412 files changed, 92941 insertions, 2158 deletions

diff --git a/usr/src/uts/common/Makefile.files b/usr/src/uts/common/Makefile.files
index 8477310f0a..055a038c24 100644
--- a/usr/src/uts/common/Makefile.files
+++ b/usr/src/uts/common/Makefile.files
@@ -24,6 +24,7 @@
 # Copyright (c) 2013 by Delphix. All rights reserved.
 # Copyright (c) 2013 by Saso Kiselkov. All rights reserved.
 # Copyright 2014 Nexenta Systems, Inc.  All rights reserved.
+# Copyright 2015, Joyent, Inc.
 #
 
 #
@@ -280,6 +281,7 @@ GENUNIX_OBJS +=	\
 		rctl.o		\
 		rctlsys.o	\
 		readlink.o	\
+		refhash.o	\
 		refstr.o	\
 		rename.o	\
 		resolvepath.o	\
@@ -427,6 +429,8 @@ PROFILE_OBJS += profile.o
 
 SYSTRACE_OBJS += systrace.o
 
+LX_SYSTRACE_OBJS += lx_systrace.o
+
 LOCKSTAT_OBJS += lockstat.o
 
 FASTTRAP_OBJS += fasttrap.o fasttrap_isa.o
@@ -491,6 +495,12 @@ PTSL_OBJS +=	tty_pts.o
 
 PTM_OBJS +=	ptm.o
 
+LX_PTM_OBJS +=	lx_ptm.o
+
+LX_AUDIO_OBJS += lx_audio.o
+
+LX_NETLINK_OBJS +=	lx_netlink.o
+
 MII_OBJS +=	mii.o mii_cicada.o mii_natsemi.o mii_intel.o mii_qualsemi.o \
 		mii_marvell.o mii_realtek.o mii_other.o
 
@@ -544,6 +554,7 @@ IP_SCTP_OBJS =	sctp.o sctp_opt_data.o sctp_output.o \
 		sctp_addr.o tn_ipopt.o tnet.o ip_netinfo.o \
 		sctp_misc.o
 IP_ILB_OBJS =	ilb.o ilb_nat.o ilb_conn.o ilb_alg_hash.o ilb_alg_rr.o
+IP_COMM_OBJS =	inet_hash.o
 
 IP_OBJS +=	igmp.o ipmp.o ip.o ip6.o ip6_asp.o ip6_if.o ip6_ire.o \
 		ip6_rts.o ip_if.o ip_ire.o ip_listutils.o ip_mroute.o \
@@ -559,7 +570,8 @@ IP_OBJS +=	igmp.o ipmp.o ip.o ip6.o ip6_asp.o ip6_if.o ip6_ire.o \
 		$(IP_TCP_OBJS) \
 		$(IP_UDP_OBJS) \
 		$(IP_SCTP_OBJS) \
-		$(IP_ILB_OBJS)
+		$(IP_ILB_OBJS) \
+		$(IP_COMM_OBJS)
 
 IP6_OBJS +=	ip6ddi.o
 
@@ -577,6 +589,8 @@ IPSECESP_OBJS += ipsecespddi.o ipsecesp.o
 
 IPSECAH_OBJS +=	ipsecahddi.o ipsecah.o sadb.o
 
+DATAFILT_OBJS += datafilt.o
+
 SPPP_OBJS +=	sppp.o sppp_dlpi.o sppp_mod.o s_common.o
 
 SPPPTUN_OBJS +=	sppptun.o sppptun_mod.o
@@ -674,6 +688,15 @@ NET80211_OBJS += net80211.o net80211_proto.o net80211_input.o \
 
 VNIC_OBJS +=	vnic_ctl.o vnic_dev.o
 
+OVERLAY_OBJS +=	overlay.o overlay_fm.o overlay_mux.o overlay_plugin.o \
+		overlay_prop.o overlay_target.o
+
+OVERLAY_VXLAN_OBJS +=	overlay_vxlan.o
+
+VND_OBJS +=	vnd.o frameio.o
+
+GSQUEUE_OBJS +=	gsqueue.o
+
 SIMNET_OBJS +=	simnet.o
 
 IB_OBJS +=	ibnex.o ibnex_ioctl.o ibnex_hca.o
@@ -986,8 +1009,12 @@ DEVPOLL_OBJS += devpoll.o
 
 DEVPOOL_OBJS += devpool.o
 
+EVENTFD_OBJS +=	eventfd.o
+
 I8042_OBJS +=	i8042.o
 
+INOTIFY_OBJS +=	inotify.o
+
 KB8042_OBJS +=	\
 		at_keyprocess.o	\
 		kb8042.o	\
@@ -1025,6 +1052,8 @@ GEN_DRV_OBJS += gen_drv.o
 
 TCLIENT_OBJS +=	tclient.o
 
+TIMERFD_OBJS +=	timerfd.o
+
 TPHCI_OBJS += tphci.o
 
 TVHCI_OBJS += tvhci.o
@@ -1060,6 +1089,8 @@ QLGE_OBJS += qlge.o qlge_dbg.o qlge_flash.o qlge_fm.o qlge_gld.o qlge_mpi.o
 
 ZCONS_OBJS += zcons.o
 
+ZFD_OBJS += zfd.o
+
 NV_SATA_OBJS += nv_sata.o
 
 SI3124_OBJS += si3124.o
@@ -1122,8 +1153,7 @@ DEVFS_OBJS +=	devfs_subr.o	devfs_vfsops.o	devfs_vnops.o
 DEV_OBJS  +=	sdev_subr.o	sdev_vfsops.o	sdev_vnops.o	\
 		sdev_ptsops.o	sdev_zvolops.o	sdev_comm.o	\
 		sdev_profile.o	sdev_ncache.o	sdev_netops.o	\
-		sdev_ipnetops.o	\
-		sdev_vtops.o
+		sdev_ipnetops.o	sdev_vtops.o	sdev_plugin.o
 
 CTFS_OBJS +=	ctfs_all.o ctfs_cdir.o ctfs_ctl.o ctfs_event.o \
 		ctfs_latest.o ctfs_root.o ctfs_sym.o ctfs_tdir.o ctfs_tmpl.o
@@ -1140,8 +1170,13 @@ PIPE_OBJS +=	pipe.o
 HSFS_OBJS +=	hsfs_node.o	hsfs_subr.o	hsfs_vfsops.o	hsfs_vnops.o \
 		hsfs_susp.o	hsfs_rrip.o	hsfs_susp_subr.o
 
+HYPRLOFS_OBJS += hyprlofs_dir.o hyprlofs_subr.o \
+		hyprlofs_vnops.o hyprlofs_vfsops.o
+
 LOFS_OBJS +=	lofs_subr.o	lofs_vfsops.o	lofs_vnops.o
 
+LXPROC_OBJS +=	lxpr_subr.o	lxpr_vfsops.o	lxpr_vnops.o
+
 NAMEFS_OBJS +=	namevfs.o	namevno.o
 
 NFS_OBJS +=	nfs_client.o	nfs_common.o	nfs_dump.o \
@@ -1249,8 +1284,8 @@ SMBSRV_OBJS +=	$(SMBSRV_SHARED_OBJS)			\
 PCFS_OBJS +=	pc_alloc.o	pc_dir.o	pc_node.o	pc_subr.o \
 		pc_vfsops.o	pc_vnops.o
 
-PROC_OBJS +=	prcontrol.o	prioctl.o	prsubr.o	prusrio.o \
-		prvfsops.o	prvnops.o
+PROC_OBJS +=	prargv.o	prcontrol.o	prioctl.o	prsubr.o \
+		prusrio.o	prvfsops.o	prvnops.o
 
 MNTFS_OBJS +=	mntvfsops.o	mntvnops.o
 
@@ -1294,6 +1329,7 @@ SMBFS_OBJS +=	smbfs_vfsops.o	smbfs_vnops.o	smbfs_node.o	\
 		smbfs_rwlock.o	smbfs_xattr.o	\
 		$(SMBFS_COMMON_OBJS)
 
+BOOTFS_OBJS +=	bootfs_construct.o bootfs_vfsops.o bootfs_vnops.o
 
 #
 #			LVM modules
@@ -1388,6 +1424,7 @@ ZFS_COMMON_OBJS +=		\
 	zfs_fuid.o		\
 	zfs_sa.o		\
 	zfs_znode.o		\
+	zfs_zone.o		\
 	zil.o			\
 	zio.o			\
 	zio_checksum.o		\
@@ -1834,7 +1871,7 @@ ZYD_OBJS += zyd.o zyd_usb.o zyd_hw.o zyd_fw.o
 
 MXFE_OBJS += mxfe.o
 
-MPTSAS_OBJS += mptsas.o mptsas_hash.o mptsas_impl.o mptsas_init.o \
+MPTSAS_OBJS += mptsas.o mptsas_impl.o mptsas_init.o \
 		mptsas_raid.o mptsas_smhba.o
 
 SFE_OBJS += sfe.o sfe_util.o
@@ -2027,6 +2064,11 @@ MEGA_SAS_OBJS = megaraid_sas.o
 MR_SAS_OBJS = ld_pd_map.o mr_sas.o mr_sas_tbolt.o mr_sas_list.o
 
 #
+#	DR_SAS module
+#
+DR_SAS_OBJS = dr_sas.o
+
+#
 #	CPQARY3 module
 #
 CPQARY3_OBJS =	cpqary3.o cpqary3_noe.o cpqary3_talk2ctlr.o	\
@@ -2065,6 +2107,20 @@ NULLDRIVER_OBJS =	nulldriver.o
 TPM_OBJS =	tpm.o tpm_hcall.o
 
 #
+# USB Fast ethernet drivers
+#
+USBGEM_OBJS = usbgem.o
+AXF_OBJS = axf_usbgem.o
+UDMF_OBJS = udmf_usbgem.o
+URF_OBJS = urf_usbgem.o
+UPF_OBJS = upf_usbgem.o
+
+#
+#	NFP objects
+#
+NFP_OBJS = hostif.o osif.o drvlist.o i21555.o i21285.o i21555d.o
+
+#
 #	BNXE objects
 #
 BNXE_OBJS +=	bnxe_cfg.o		\
diff --git a/usr/src/uts/common/Makefile.rules b/usr/src/uts/common/Makefile.rules
index c5e728fdff..e452aeb830 100644
--- a/usr/src/uts/common/Makefile.rules
+++ b/usr/src/uts/common/Makefile.rules
@@ -23,6 +23,7 @@
 # Copyright (c) 1991, 2010, Oracle and/or its affiliates. All rights reserved.
 # Copyright 2013 Garrett D'Amore <garrett@damore.org>
 # Copyright 2014 Nexenta Systems, Inc.  All rights reserved.
+# Copyright 2015 Joyent, Inc.
 #
 
 #
@@ -95,6 +96,10 @@ $(OBJS_DIR)/%.o:		$(COMMONBASE)/avl/%.c
 	$(COMPILE.c) -o $@ $<
 	$(CTFCONVERT_O)
 
+$(OBJS_DIR)/%.o:		$(COMMONBASE)/inet/%.c
+	$(COMPILE.c) -o $@ $<
+	$(CTFCONVERT_O)
+
 $(OBJS_DIR)/%.o:		$(COMMONBASE)/ucode/%.c
 	$(COMPILE.c) -o $@ $<
 	$(CTFCONVERT_O)
@@ -103,6 +108,10 @@ $(OBJS_DIR)/%.o:		$(UTSBASE)/common/brand/sn1/%.c
 	$(COMPILE.c) -o $@ $<
 	$(CTFCONVERT_O)
 
+$(OBJS_DIR)/%.o:		$(UTSBASE)/common/brand/sngl/%.c
+	$(COMPILE.c) -o $@ $<
+	$(CTFCONVERT_O)
+
 $(OBJS_DIR)/%.o:		$(UTSBASE)/common/brand/solaris10/%.c
 	$(COMPILE.c) -o $@ $<
 	$(CTFCONVERT_O)
@@ -207,6 +216,10 @@ $(OBJS_DIR)/%.o:		$(UTSBASE)/common/fs/autofs/%.c
 	$(COMPILE.c) -o $@ $<
 	$(CTFCONVERT_O)
 
+$(OBJS_DIR)/%.o:		$(UTSBASE)/common/fs/bootfs/%.c
+	$(COMPILE.c) -o $@ $<
+	$(CTFCONVERT_O)
+
 $(OBJS_DIR)/%.o:		$(UTSBASE)/common/fs/cachefs/%.c
 	$(COMPILE.c) -o $@ $<
 	$(CTFCONVERT_O)
@@ -243,10 +256,18 @@ $(OBJS_DIR)/%.o:		$(UTSBASE)/common/fs/hsfs/%.c
 	$(COMPILE.c) -o $@ $<
 	$(CTFCONVERT_O)
 
+$(OBJS_DIR)/%.o:		$(UTSBASE)/common/fs/hyprlofs/%.c
+	$(COMPILE.c) -o $@ $<
+	$(CTFCONVERT_O)
+
 $(OBJS_DIR)/%.o:		$(UTSBASE)/common/fs/lofs/%.c
 	$(COMPILE.c) -o $@ $<
 	$(CTFCONVERT_O)
 
+$(OBJS_DIR)/%.o:		$(UTSBASE)/common/fs/lxproc/%.c
+	$(COMPILE.c) -o $@ $<
+	$(CTFCONVERT_O)
+
 $(OBJS_DIR)/%.o:		$(UTSBASE)/common/fs/mntfs/%.c
 	$(COMPILE.c) -o $@ $<
 	$(CTFCONVERT_O)
@@ -758,6 +779,10 @@ $(OBJS_DIR)/%.o:		$(UTSBASE)/common/io/drm/%.c
 	$(COMPILE.c) -o $@ $<
 	$(CTFCONVERT_O)
 
+$(OBJS_DIR)/%.o:		$(UTSBASE)/common/io/dr_sas/%.c
+	$(COMPILE.c) -o $@ $<
+	$(CTFCONVERT_O)
+
 $(OBJS_DIR)/%.o:		$(UTSBASE)/common/io/efe/%.c
 	$(COMPILE.c) -o $@ $<
 	$(CTFCONVERT_O)
@@ -986,6 +1011,10 @@ $(OBJS_DIR)/%.o:		$(UTSBASE)/common/io/net80211/%.c
 	$(COMPILE.c) -o $@ $<
 	$(CTFCONVERT_O)
 
+$(OBJS_DIR)/%.o:		$(UTSBASE)/common/io/nfp/%.c
+	$(COMPILE.c) -o $@ $<
+	$(CTFCONVERT_O)
+
 $(OBJS_DIR)/%.o:		$(UTSBASE)/common/io/nge/%.c
 	$(COMPILE.c) -o $@ $<
 	$(CTFCONVERT_O)
@@ -1001,6 +1030,14 @@ $(OBJS_DIR)/%.o:		$(UTSBASE)/common/io/nxge/npi/%.c
 $(OBJS_DIR)/%.o:		$(UTSBASE)/common/io/nxge/%.s
 	$(COMPILE.s) -o $@ $<
 
+$(OBJS_DIR)/%.o:		$(UTSBASE)/common/io/overlay/%.c
+	$(COMPILE.c) -o $@ $<
+	$(CTFCONVERT_O)
+
+$(OBJS_DIR)/%.o:		$(UTSBASE)/common/io/overlay/plugins/%.c
+	$(COMPILE.c) -o $@ $<
+	$(CTFCONVERT_O)
+
 $(OBJS_DIR)/%.o:		$(UTSBASE)/common/io/pci-ide/%.c
 	$(COMPILE.c) -o $@ $<
 	$(CTFCONVERT_O)
@@ -1137,6 +1174,10 @@ $(OBJS_DIR)/%.o:		$(UTSBASE)/common/io/sdcard/targets/sdcard/%.c
 	$(COMPILE.c) -o $@ $<
 	$(CTFCONVERT_O)
 
+$(OBJS_DIR)/%.o:		$(UTSBASE)/common/io/gsqueue/%.c
+	$(COMPILE.c) -o $@ $<
+	$(CTFCONVERT_O)
+
 $(OBJS_DIR)/%.o:		$(UTSBASE)/common/io/sfe/%.c
 	$(COMPILE.c) -o $@ $<
 	$(CTFCONVERT_O)
@@ -1149,6 +1190,10 @@ $(OBJS_DIR)/%.o:		$(UTSBASE)/common/io/softmac/%.c
 	$(COMPILE.c) -o $@ $<
 	$(CTFCONVERT_O)
 
+$(OBJS_DIR)/%.o:		$(UTSBASE)/common/io/vnd/%.c
+	$(COMPILE.c) -o $@ $<
+	$(CTFCONVERT_O)
+
 $(OBJS_DIR)/%.o:		$(UTSBASE)/common/io/uath/%.c
 	$(COMPILE.c) -o $@ $<
 	$(CTFCONVERT_O)
@@ -1273,6 +1318,30 @@ $(OBJS_DIR)/%.o:		$(UTSBASE)/common/io/usb/usba10/%.c
 	$(COMPILE.c) -o $@ $<
 	$(CTFCONVERT_O)
 
+$(OBJS_DIR)/%.o:		$(UTSBASE)/common/io/usbgem/%.c
+	$(COMPILE.c) -o $@ $<
+	$(CTFCONVERT_O)
+
+$(OBJS_DIR)/%.o:		$(UTSBASE)/common/io/axf/%.c
+	$(COMPILE.c) -o $@ $<
+	$(CTFCONVERT_O)
+
+$(OBJS_DIR)/%.o:		$(UTSBASE)/common/io/udf/%.c
+	$(COMPILE.c) -o $@ $<
+	$(CTFCONVERT_O)
+
+$(OBJS_DIR)/%.o:		$(UTSBASE)/common/io/udmf/%.c
+	$(COMPILE.c) -o $@ $<
+	$(CTFCONVERT_O)
+
+$(OBJS_DIR)/%.o:		$(UTSBASE)/common/io/upf/%.c
+	$(COMPILE.c) -o $@ $<
+	$(CTFCONVERT_O)
+
+$(OBJS_DIR)/%.o:		$(UTSBASE)/common/io/urf/%.c
+	$(COMPILE.c) -o $@ $<
+	$(CTFCONVERT_O)
+
 $(OBJS_DIR)/%.o:		$(UTSBASE)/common/io/vuidmice/%.c
 	$(COMPILE.c) -o $@ $<
 	$(CTFCONVERT_O)
@@ -1445,6 +1514,10 @@ $(OBJS_DIR)/%.o:		$(UTSBASE)/common/io/vioblk/%.c
 	$(COMPILE.c) -o $@ $<
 	$(CTFCONVERT_O)
 
+$(OBJS_DIR)/%.o:		$(COMMONBASE)/idspace/%.c
+	$(COMPILE.c) -o $@ $<
+	$(CTFCONVERT_O)
+
 #
 # krtld must refer to its own bzero/bcopy until the kernel is fully linked
 #
@@ -1513,6 +1586,10 @@ $(OBJS_DIR)/%.o:		$(UTSBASE)/common/pcmcia/pcs/%.c
 	$(COMPILE.c) -o $@ $<
 	$(CTFCONVERT_O)
 
+$(OBJS_DIR)/%.o:		$(UTSBASE)/common/refhash/%.c
+	$(COMPILE.c) -o $@ $<
+	$(CTFCONVERT_O)
+
 $(OBJS_DIR)/%.o:		$(UTSBASE)/common/rpc/%.c
 	$(COMPILE.c) -o $@ $<
 	$(CTFCONVERT_O)
@@ -1628,12 +1705,18 @@ $(LINTS_DIR)/%.ln:		$(COMMONBASE)/acl/%.c
 $(LINTS_DIR)/%.ln:		$(COMMONBASE)/avl/%.c
 	@($(LHEAD) $(LINT.c) $< $(LTAIL))
 
+$(LINTS_DIR)/%.ln:		$(COMMONBASE)/inet/%.c
+	@($(LHEAD) $(LINT.c) $< $(LTAIL))
+
 $(LINTS_DIR)/%.ln:		$(COMMONBASE)/ucode/%.c
 	@($(LHEAD) $(LINT.c) $< $(LTAIL))
 
 $(LINTS_DIR)/%.ln:		$(UTSBASE)/common/brand/sn1/%.c
 	@($(LHEAD) $(LINT.c) $< $(LTAIL))
 
+$(LINTS_DIR)/%.ln:		$(UTSBASE)/common/brand/sngl/%.c
+	@($(LHEAD) $(LINT.c) $< $(LTAIL))
+
 $(LINTS_DIR)/%.ln:		$(UTSBASE)/common/brand/solaris10/%.c
 	@($(LHEAD) $(LINT.c) $< $(LTAIL))
 
@@ -1736,6 +1819,9 @@ $(LINTS_DIR)/%.ln:		$(UTSBASE)/common/fs/%.c
 $(LINTS_DIR)/%.ln:		$(UTSBASE)/common/fs/autofs/%.c
 	@($(LHEAD) $(LINT.c) $< $(LTAIL))
 
+$(LINTS_DIR)/%.ln:		$(UTSBASE)/common/fs/bootfs/%.c
+	@($(LHEAD) $(LINT.c) $< $(LTAIL))
+
 $(LINTS_DIR)/%.ln:		$(UTSBASE)/common/fs/cachefs/%.c
 	@($(LHEAD) $(LINT.c) $< $(LTAIL))
 
@@ -1763,9 +1849,15 @@ $(LINTS_DIR)/%.ln:		$(UTSBASE)/common/fs/fifofs/%.c
 $(LINTS_DIR)/%.ln:		$(UTSBASE)/common/fs/hsfs/%.c
 	@($(LHEAD) $(LINT.c) $< $(LTAIL))
 
+$(LINTS_DIR)/%.ln:		$(UTSBASE)/common/fs/hyprlofs/%.c
+	@($(LHEAD) $(LINT.c) $< $(LTAIL))
+
 $(LINTS_DIR)/%.ln:		$(UTSBASE)/common/fs/lofs/%.c
 	@($(LHEAD) $(LINT.c) $< $(LTAIL))
 
+$(LINTS_DIR)/%.ln:		$(UTSBASE)/common/fs/lxproc/%.c
+	@($(LHEAD) $(LINT.c) $< $(LTAIL))
+
 $(LINTS_DIR)/%.ln:		$(UTSBASE)/common/fs/mntfs/%.c
 	@($(LHEAD) $(LINT.c) $< $(LTAIL))
 
@@ -2111,6 +2203,9 @@ $(LINTS_DIR)/%.ln:		$(UTSBASE)/common/io/dmfe/%.c
 $(LINTS_DIR)/%.ln:		$(UTSBASE)/common/io/drm/%.c
 	@($(LHEAD) $(LINT.c) $< $(LTAIL))
 
+$(LINTS_DIR)/%.ln:		$(UTSBASE)/common/io/dr_sas/%.c
+	@($(LHEAD) $(LINT.c) $< $(LTAIL))
+
 $(LINTS_DIR)/%.ln:		$(UTSBASE)/common/io/efe/%.c
 	@($(LHEAD) $(LINT.c) $< $(LTAIL))
 
@@ -2282,6 +2377,9 @@ $(LINTS_DIR)/%.ln:		$(UTSBASE)/common/io/mwl/mwl_fw/%.c
 $(LINTS_DIR)/%.ln:		$(UTSBASE)/common/io/net80211/%.c
 	@($(LHEAD) $(LINT.c) $< $(LTAIL))
 
+$(LINTS_DIR)/%.ln:		$(UTSBASE)/common/io/nfp/%.c
+	@($(LHEAD) $(LINT.c) $< $(LTAIL))
+
 $(LINTS_DIR)/%.ln:              $(UTSBASE)/common/io/nge/%.c
 	@($(LHEAD) $(LINT.c) $< $(LTAIL))
 
@@ -2294,6 +2392,12 @@ $(LINTS_DIR)/%.ln:		$(UTSBASE)/common/io/nxge/%.s
 $(LINTS_DIR)/%.ln:		$(UTSBASE)/common/io/nxge/npi/%.c
 	@($(LHEAD) $(LINT.c) $< $(LTAIL))
 
+$(LINTS_DIR)/%.ln:		$(UTSBASE)/common/io/overlay/%.c
+	@($(LHEAD) $(LINT.c) $< $(LTAIL))
+
+$(LINTS_DIR)/%.ln:		$(UTSBASE)/common/io/overlay/plugins/%.c
+	@($(LHEAD) $(LINT.c) $< $(LTAIL))
+
 $(LINTS_DIR)/%.ln:		$(UTSBASE)/common/io/pci-ide/%.c
 	@($(LHEAD) $(LINT.c) $< $(LTAIL))
 
@@ -2501,6 +2605,21 @@ $(LINTS_DIR)/%.ln:		$(UTSBASE)/common/io/usb/usba/%.c
 $(LINTS_DIR)/%.ln:		$(UTSBASE)/common/io/usb/usba10/%.c
 	@($(LHEAD) $(LINT.c) $< $(LTAIL))
 
+$(LINTS_DIR)/%.ln:		$(UTSBASE)/common/io/usbgem/%.c
+	@($(LHEAD) $(LINT.c) $< $(LTAIL))
+
+$(LINTS_DIR)/%.ln:		$(UTSBASE)/common/io/axf/%.c
+	@($(LHEAD) $(LINT.c) $< $(LTAIL))
+
+$(LINTS_DIR)/%.ln:		$(UTSBASE)/common/io/udmf/%.c
+	@($(LHEAD) $(LINT.c) $< $(LTAIL))
+
+$(LINTS_DIR)/%.ln:		$(UTSBASE)/common/io/upf/%.c
+	@($(LHEAD) $(LINT.c) $< $(LTAIL))
+
+$(LINTS_DIR)/%.ln:		$(UTSBASE)/common/io/urf/%.c
+	@($(LHEAD) $(LINT.c) $< $(LTAIL))
+
 $(LINTS_DIR)/%.ln:		$(UTSBASE)/common/io/vuidmice/%.c
 	@($(LHEAD) $(LINT.c) $< $(LTAIL))
 
@@ -2648,6 +2767,9 @@ $(LINTS_DIR)/%.ln:		$(UTSBASE)/common/pcmcia/nexus/%.c
 $(LINTS_DIR)/%.ln:		$(UTSBASE)/common/pcmcia/pcs/%.c
 	@($(LHEAD) $(LINT.c) $< $(LTAIL))
 
+$(LINTS_DIR)/%.ln:		$(UTSBASE)/common/refhash/%.c
+	@($(LHEAD) $(LINT.c) $< $(LTAIL))
+
 $(LINTS_DIR)/%.ln:		$(UTSBASE)/common/rpc/%.c
 	@($(LHEAD) $(LINT.c) $< $(LTAIL))
 
@@ -2726,3 +2848,6 @@ $(LINTS_DIR)/%.ln:		$(UTSBASE)/common/io/skd/%.c
 
 $(LINTS_DIR)/%.ln:		$(COMMONBASE)/fsreparse/%.c
 	@($(LHEAD) $(LINT.c) $< $(LTAIL))
+
+$(LINTS_DIR)/%.ln:		$(COMMONBASE)/idspace/%.c
+	@($(LHEAD) $(LINT.c) $< $(LTAIL))
diff --git a/usr/src/uts/common/brand/lx/autofs/lx_autofs.c b/usr/src/uts/common/brand/lx/autofs/lx_autofs.c
new file mode 100644
index 0000000000..d2bb03c118
--- /dev/null
+++ b/usr/src/uts/common/brand/lx/autofs/lx_autofs.c
@@ -0,0 +1,1569 @@
+/*
+ * CDDL HEADER START
+ *
+ * The contents of this file are subject to the terms of the
+ * Common Development and Distribution License (the "License").
+ * You may not use this file except in compliance with the License.
+ *
+ * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
+ * or http://www.opensolaris.org/os/licensing.
+ * See the License for the specific language governing permissions
+ * and limitations under the License.
+ *
+ * When distributing Covered Code, include this CDDL HEADER in each
+ * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
+ * If applicable, add the following below this CDDL HEADER, with the
+ * fields enclosed by brackets "[]" replaced with your own identifying
+ * information: Portions Copyright [yyyy] [name of copyright owner]
+ *
+ * CDDL HEADER END
+ */
+
+/*
+ * Copyright 2007 Sun Microsystems, Inc.  All rights reserved.
+ * Use is subject to license terms.
+ */
+
+#pragma ident	"%Z%%M%	%I%	%E% SMI"
+
+#include <fs/fs_subr.h>
+#include <sys/atomic.h>
+#include <sys/cmn_err.h>
+#include <sys/dirent.h>
+#include <sys/fs/fifonode.h>
+#include <sys/modctl.h>
+#include <sys/mount.h>
+#include <sys/policy.h>
+#include <sys/sunddi.h>
+
+#include <sys/sysmacros.h>
+#include <sys/vfs.h>
+#include <sys/vfs_opreg.h>
+
+#include <sys/lx_autofs_impl.h>
+
+/*
+ * External functions
+ */
+extern uintptr_t		space_fetch(char *key);
+extern int			space_store(char *key, uintptr_t ptr);
+
+/*
+ * Globals
+ */
+static vfsops_t			*lx_autofs_vfsops;
+static vnodeops_t		*lx_autofs_vn_ops = NULL;
+static int			lx_autofs_fstype;
+static major_t			lx_autofs_major;
+static minor_t			lx_autofs_minor = 0;
+
+/*
+ * Support functions
+ */
+static void
+i_strfree(char *str)
+{
+	kmem_free(str, strlen(str) + 1);
+}
+
+static char *
+i_strdup(char *str)
+{
+	int	n = strlen(str);
+	char	*ptr = kmem_alloc(n + 1, KM_SLEEP);
+	bcopy(str, ptr, n + 1);
+	return (ptr);
+}
+
+static int
+i_str_to_int(char *str, int *val)
+{
+	long	res;
+
+	if (str == NULL)
+		return (-1);
+
+	if ((ddi_strtol(str, NULL, 10, &res) != 0) ||
+	    (res < INT_MIN) || (res > INT_MAX))
+		return (-1);
+
+	*val = res;
+	return (0);
+}
+
+static void
+i_stack_init(list_t *lp)
+{
+	list_create(lp,
+	    sizeof (stack_elem_t), offsetof(stack_elem_t, se_list));
+}
+
+static void
+i_stack_fini(list_t *lp)
+{
+	ASSERT(list_head(lp) == NULL);
+	list_destroy(lp);
+}
+
+static void
+i_stack_push(list_t *lp, caddr_t ptr1, caddr_t ptr2, caddr_t ptr3)
+{
+	stack_elem_t	*se;
+
+	se = kmem_alloc(sizeof (*se), KM_SLEEP);
+	se->se_ptr1 = ptr1;
+	se->se_ptr2 = ptr2;
+	se->se_ptr3 = ptr3;
+	list_insert_head(lp, se);
+}
+
+static int
+i_stack_pop(list_t *lp, caddr_t *ptr1, caddr_t *ptr2, caddr_t *ptr3)
+{
+	stack_elem_t	*se;
+
+	if ((se = list_head(lp)) == NULL)
+		return (-1);
+	list_remove(lp, se);
+	if (ptr1 != NULL)
+		*ptr1 = se->se_ptr1;
+	if (ptr2 != NULL)
+		*ptr2 = se->se_ptr2;
+	if (ptr3 != NULL)
+		*ptr3 = se->se_ptr3;
+	kmem_free(se, sizeof (*se));
+	return (0);
+}
+
+static vnode_t *
+fifo_peer_vp(vnode_t *vp)
+{
+	fifonode_t *fnp = VTOF(vp);
+	fifonode_t *fn_dest = fnp->fn_dest;
+	return (FTOV(fn_dest));
+}
+
+static vnode_t *
+i_vn_alloc(vfs_t *vfsp, vnode_t *uvp)
+{
+	lx_autofs_vfs_t	*data = vfsp->vfs_data;
+	vnode_t		*vp, *vp_old;
+
+	/* Allocate a new vnode structure in case we need it. */
+	vp = vn_alloc(KM_SLEEP);
+	vn_setops(vp, lx_autofs_vn_ops);
+	VN_SET_VFS_TYPE_DEV(vp, vfsp, uvp->v_type, uvp->v_rdev);
+	vp->v_data = uvp;
+	ASSERT(vp->v_count == 1);
+
+	/*
+	 * Take a hold on the vfs structure.  This is how unmount will
+	 * determine if there are any active vnodes in the file system.
+	 */
+	VFS_HOLD(vfsp);
+
+	/*
+	 * Check if we already have a vnode allocated for this underlying
+	 * vnode_t.
+	 */
+	mutex_enter(&data->lav_lock);
+	if (mod_hash_find(data->lav_vn_hash,
+	    (mod_hash_key_t)uvp, (mod_hash_val_t *)&vp_old) != 0) {
+
+		/*
+		 * Didn't find an existing node.
+		 * Add this node to the hash and return.
+		 */
+		VERIFY(mod_hash_insert(data->lav_vn_hash,
+		    (mod_hash_key_t)uvp,
+		    (mod_hash_val_t)vp) == 0);
+		mutex_exit(&data->lav_lock);
+		return (vp);
+	}
+
+	/* Get a hold on the existing vnode and free up the one we allocated. */
+	VN_HOLD(vp_old);
+	mutex_exit(&data->lav_lock);
+
+	/* Free up the new vnode we allocated. */
+	VN_RELE(uvp);
+	VFS_RELE(vfsp);
+	vn_invalid(vp);
+	vn_free(vp);
+
+	return (vp_old);
+}
+
+static void
+i_vn_free(vnode_t *vp)
+{
+	vfs_t		*vfsp = vp->v_vfsp;
+	lx_autofs_vfs_t	*data = vfsp->vfs_data;
+	vnode_t		*uvp = vp->v_data;
+	vnode_t	*vp_tmp;
+
+	ASSERT(MUTEX_HELD((&data->lav_lock)));
+	ASSERT(MUTEX_HELD((&vp->v_lock)));
+
+	ASSERT(vp->v_count == 0);
+
+	/* We're about to free this vnode so take it out of the hash. */
+	(void) mod_hash_remove(data->lav_vn_hash,
+	    (mod_hash_key_t)uvp, (mod_hash_val_t)&vp_tmp);
+
+	/*
+	 * No one else can lookup this vnode any more so there's no need
+	 * to hold locks.
+	 */
+	mutex_exit(&data->lav_lock);
+	mutex_exit(&vp->v_lock);
+
+	/* Release the underlying vnode. */
+	VN_RELE(uvp);
+	VFS_RELE(vfsp);
+	vn_invalid(vp);
+	vn_free(vp);
+}
+
+static lx_autofs_lookup_req_t *
+i_lalr_alloc(lx_autofs_vfs_t *data, int *dup_request, char *nm)
+{
+	lx_autofs_lookup_req_t	*lalr, *lalr_dup;
+
+	/* Pre-allocate a new automounter request before grabbing locks. */
+	lalr = kmem_zalloc(sizeof (*lalr), KM_SLEEP);
+	mutex_init(&lalr->lalr_lock, NULL, MUTEX_DEFAULT, NULL);
+	cv_init(&lalr->lalr_cv, NULL, CV_DEFAULT, NULL);
+	lalr->lalr_ref = 1;
+	lalr->lalr_pkt.lap_protover = LX_AUTOFS_PROTO_VERSION;
+
+	/* Assign a unique id for this request. */
+	lalr->lalr_pkt.lap_id = id_alloc(data->lav_ids);
+
+	/*
+	 * The token expected by the linux automount is the name of
+	 * the directory entry to look up.  (And not the entire
+	 * path that is being accessed.)
+	 */
+	lalr->lalr_pkt.lap_name_len = strlen(nm);
+	if (lalr->lalr_pkt.lap_name_len >
+	    (sizeof (lalr->lalr_pkt.lap_name) - 1)) {
+		zcmn_err(getzoneid(), CE_NOTE,
+		    "invalid autofs lookup: \"%s\"", nm);
+		id_free(data->lav_ids, lalr->lalr_pkt.lap_id);
+		kmem_free(lalr, sizeof (*lalr));
+		return (NULL);
+	}
+	(void) strlcpy(lalr->lalr_pkt.lap_name, nm,
+	    sizeof (lalr->lalr_pkt.lap_name));
+
+	/* Check for an outstanding request for this path. */
+	mutex_enter(&data->lav_lock);
+	if (mod_hash_find(data->lav_path_hash,
+	    (mod_hash_key_t)nm, (mod_hash_val_t *)&lalr_dup) == 0) {
+		/*
+		 * There's already an outstanding request for this
+		 * path so we don't need a new one.
+		 */
+		id_free(data->lav_ids, lalr->lalr_pkt.lap_id);
+		kmem_free(lalr, sizeof (*lalr));
+		lalr = lalr_dup;
+
+		/* Bump the ref count on the old request. */
+		atomic_add_int(&lalr->lalr_ref, 1);
+
+		*dup_request = 1;
+	} else {
+		/* Add it to the hashes. */
+		VERIFY(mod_hash_insert(data->lav_id_hash,
+		    (mod_hash_key_t)(uintptr_t)lalr->lalr_pkt.lap_id,
+		    (mod_hash_val_t)lalr) == 0);
+		VERIFY(mod_hash_insert(data->lav_path_hash,
+		    (mod_hash_key_t)i_strdup(nm),
+		    (mod_hash_val_t)lalr) == 0);
+
+		*dup_request = 0;
+	}
+	mutex_exit(&data->lav_lock);
+
+	return (lalr);
+}
+
+static lx_autofs_lookup_req_t *
+i_lalr_find(lx_autofs_vfs_t *data, int id)
+{
+	lx_autofs_lookup_req_t	*lalr;
+
+	/* Check for an outstanding request for this id. */
+	mutex_enter(&data->lav_lock);
+	if (mod_hash_find(data->lav_id_hash, (mod_hash_key_t)(uintptr_t)id,
+	    (mod_hash_val_t *)&lalr) != 0) {
+		mutex_exit(&data->lav_lock);
+		return (NULL);
+	}
+	atomic_add_int(&lalr->lalr_ref, 1);
+	mutex_exit(&data->lav_lock);
+	return (lalr);
+}
+
+static void
+i_lalr_complete(lx_autofs_vfs_t *data, lx_autofs_lookup_req_t *lalr)
+{
+	lx_autofs_lookup_req_t	*lalr_tmp;
+
+	/* Remove this request from the hashes so no one can look it up. */
+	mutex_enter(&data->lav_lock);
+	(void) mod_hash_remove(data->lav_id_hash,
+		    (mod_hash_key_t)(uintptr_t)lalr->lalr_pkt.lap_id,
+		    (mod_hash_val_t)&lalr_tmp);
+	(void) mod_hash_remove(data->lav_path_hash,
+		    (mod_hash_key_t)lalr->lalr_pkt.lap_name,
+		    (mod_hash_val_t)&lalr_tmp);
+	mutex_exit(&data->lav_lock);
+
+	/* Mark this requst as complete and wakeup anyone waiting on it. */
+	mutex_enter(&lalr->lalr_lock);
+	lalr->lalr_complete = 1;
+	cv_broadcast(&lalr->lalr_cv);
+	mutex_exit(&lalr->lalr_lock);
+}
+
+static void
+i_lalr_release(lx_autofs_vfs_t *data, lx_autofs_lookup_req_t *lalr)
+{
+	ASSERT(!MUTEX_HELD(&lalr->lalr_lock));
+	if (atomic_add_int_nv(&lalr->lalr_ref, -1) > 0)
+		return;
+	ASSERT(lalr->lalr_ref == 0);
+	id_free(data->lav_ids, lalr->lalr_pkt.lap_id);
+	kmem_free(lalr, sizeof (*lalr));
+}
+
+static void
+i_lalr_abort(lx_autofs_vfs_t *data, lx_autofs_lookup_req_t *lalr)
+{
+	lx_autofs_lookup_req_t	*lalr_tmp;
+
+	/*
+	 * This is a little tricky.  We're aborting the wait for this
+	 * request.  So if anyone else is waiting for this request we
+	 * can't free it, but if no one else is waiting for the request
+	 * we should free it.
+	 */
+	mutex_enter(&data->lav_lock);
+	if (atomic_add_int_nv(&lalr->lalr_ref, -1) > 0) {
+		mutex_exit(&data->lav_lock);
+		return;
+	}
+	ASSERT(lalr->lalr_ref == 0);
+
+	/* Remove this request from the hashes so no one can look it up. */
+	(void) mod_hash_remove(data->lav_id_hash,
+		    (mod_hash_key_t)(uintptr_t)lalr->lalr_pkt.lap_id,
+		    (mod_hash_val_t)&lalr_tmp);
+	(void) mod_hash_remove(data->lav_path_hash,
+		    (mod_hash_key_t)lalr->lalr_pkt.lap_name,
+		    (mod_hash_val_t)&lalr_tmp);
+	mutex_exit(&data->lav_lock);
+
+	/* It's ok to free this now because the ref count was zero. */
+	id_free(data->lav_ids, lalr->lalr_pkt.lap_id);
+	kmem_free(lalr, sizeof (*lalr));
+}
+
+static int
+i_fifo_lookup(pid_t pgrp, int fd, file_t **fpp_wr, file_t **fpp_rd)
+{
+	proc_t		*prp;
+	uf_info_t	*fip;
+	uf_entry_t	*ufp_wr, *ufp_rd = NULL;
+	file_t		*fp_wr, *fp_rd = NULL;
+	vnode_t		*vp_wr, *vp_rd;
+	int		i;
+
+	/*
+	 * sprlock() is zone aware, so assuming this mount call was
+	 * initiated by a process in a zone, if it tries to specify
+	 * a pgrp outside of it's zone this call will fail.
+	 *
+	 * Also, we want to grab hold of the main automounter process
+	 * and its going to be the group leader for pgrp, so its
+	 * pid will be equal to pgrp.
+	 */
+	prp = sprlock(pgrp);
+	if (prp == NULL)
+		return (-1);
+	mutex_exit(&prp->p_lock);
+
+	/* Now we want to access the processes open file descriptors. */
+	fip = P_FINFO(prp);
+	mutex_enter(&fip->fi_lock);
+
+	/* Sanity check fifo write fd. */
+	if (fd >= fip->fi_nfiles) {
+		mutex_exit(&fip->fi_lock);
+		mutex_enter(&prp->p_lock);
+		sprunlock(prp);
+		return (-1);
+	}
+
+	/* Get a pointer to the write fifo. */
+	UF_ENTER(ufp_wr, fip, fd);
+	if (((fp_wr = ufp_wr->uf_file) == NULL) ||
+	    ((vp_wr = fp_wr->f_vnode) == NULL) || (vp_wr->v_type != VFIFO)) {
+		/* Invalid fifo fd. */
+		UF_EXIT(ufp_wr);
+		mutex_exit(&fip->fi_lock);
+		mutex_enter(&prp->p_lock);
+		sprunlock(prp);
+		return (-1);
+	}
+
+	/*
+	 * Now we need to find the read end of the fifo (for reasons
+	 * explained below.)  We assume that the read end of the fifo
+	 * is in the same process as the write end.
+	 */
+	vp_rd = fifo_peer_vp(fp_wr->f_vnode);
+	for (i = 0; i < fip->fi_nfiles; i++) {
+		UF_ENTER(ufp_rd, fip, i);
+		if (((fp_rd = ufp_rd->uf_file) != NULL) &&
+		    (fp_rd->f_vnode == vp_rd))
+			break;
+		UF_EXIT(ufp_rd);
+	}
+	if (i == fip->fi_nfiles) {
+		/* Didn't find it. */
+		UF_EXIT(ufp_wr);
+		mutex_exit(&fip->fi_lock);
+		mutex_enter(&prp->p_lock);
+		sprunlock(prp);
+		return (-1);
+	}
+
+	/*
+	 * We need to drop fi_lock before we can try to acquire f_tlock
+	 * the good news is that the file pointers are protected because
+	 * we're still holding uf_lock.
+	 */
+	mutex_exit(&fip->fi_lock);
+
+	/*
+	 * Here we bump the open counts on the fifos.  The reason
+	 * that we do this is because when we go to write to the
+	 * fifo we want to ensure that they are actually open (and
+	 * not in the process of being closed) without having to
+	 * stop the automounter.  (If the write end of the fifo
+	 * were closed and we tried to write to it we would panic.
+	 * If the read end of the fifo was closed and we tried to
+	 * write to the other end, the process that invoked the
+	 * lookup operation would get an unexpected SIGPIPE.)
+	 */
+	mutex_enter(&fp_wr->f_tlock);
+	fp_wr->f_count++;
+	ASSERT(fp_wr->f_count >= 2);
+	mutex_exit(&fp_wr->f_tlock);
+
+	mutex_enter(&fp_rd->f_tlock);
+	fp_rd->f_count++;
+	ASSERT(fp_rd->f_count >= 2);
+	mutex_exit(&fp_rd->f_tlock);
+
+	/* Release all our locks. */
+	UF_EXIT(ufp_wr);
+	UF_EXIT(ufp_rd);
+	mutex_enter(&prp->p_lock);
+	sprunlock(prp);
+
+	/* Return the file pointers. */
+	*fpp_rd = fp_rd;
+	*fpp_wr = fp_wr;
+	return (0);
+}
+
+static uint_t
+/*ARGSUSED*/
+i_fifo_close_cb(mod_hash_key_t key, mod_hash_val_t *val, void *arg)
+{
+	int	*id = (int *)arg;
+	/* Return the key and terminate the walk. */
+	*id = (uintptr_t)key;
+	return (MH_WALK_TERMINATE);
+}
+
+static void
+i_fifo_close(lx_autofs_vfs_t *data)
+{
+	/*
+	 * Close the fifo to prevent any future requests from
+	 * getting sent to the automounter.
+	 */
+	mutex_enter(&data->lav_lock);
+	if (data->lav_fifo_wr != NULL) {
+		(void) closef(data->lav_fifo_wr);
+		data->lav_fifo_wr = NULL;
+	}
+	if (data->lav_fifo_rd != NULL) {
+		(void) closef(data->lav_fifo_rd);
+		data->lav_fifo_rd = NULL;
+	}
+	mutex_exit(&data->lav_lock);
+
+	/*
+	 * Wakeup any threads currently waiting for the automounter
+	 * note that it's possible for multiple threads to have entered
+	 * this function and to be doing the work below simultaneously.
+	 */
+	for (;;) {
+		lx_autofs_lookup_req_t	*lalr;
+		int			id;
+
+		/* Lookup the first entry in the hash. */
+		id = -1;
+		mod_hash_walk(data->lav_id_hash,
+		    i_fifo_close_cb, &id);
+		if (id == -1) {
+			/* No more id's in the hash. */
+			break;
+		}
+		if ((lalr = i_lalr_find(data, id)) == NULL) {
+			/* Someone else beat us to it. */
+			continue;
+		}
+
+		/* Mark the request as compleate and release it. */
+		i_lalr_complete(data, lalr);
+		i_lalr_release(data, lalr);
+	}
+}
+
+static int
+i_fifo_verify_rd(lx_autofs_vfs_t *data)
+{
+	proc_t		*prp;
+	uf_info_t	*fip;
+	uf_entry_t	*ufp_rd = NULL;
+	file_t		*fp_rd =  NULL;
+	vnode_t		*vp_rd;
+	int		i;
+
+	ASSERT(MUTEX_HELD((&data->lav_lock)));
+
+	/* Check if we've already been shut down. */
+	if (data->lav_fifo_wr == NULL) {
+		ASSERT(data->lav_fifo_rd == NULL);
+		return (-1);
+	}
+	vp_rd = fifo_peer_vp(data->lav_fifo_wr->f_vnode);
+
+	/*
+	 * sprlock() is zone aware, so assuming this mount call was
+	 * initiated by a process in a zone, if it tries to specify
+	 * a pgrp outside of it's zone this call will fail.
+	 *
+	 * Also, we want to grab hold of the main automounter process
+	 * and its going to be the group leader for pgrp, so its
+	 * pid will be equal to pgrp.
+	 */
+	prp = sprlock(data->lav_pgrp);
+	if (prp == NULL)
+		return (-1);
+	mutex_exit(&prp->p_lock);
+
+	/* Now we want to access the processes open file descriptors. */
+	fip = P_FINFO(prp);
+	mutex_enter(&fip->fi_lock);
+
+	/*
+	 * Now we need to find the read end of the fifo (for reasons
+	 * explained below.)  We assume that the read end of the fifo
+	 * is in the same process as the write end.
+	 */
+	for (i = 0; i < fip->fi_nfiles; i++) {
+		UF_ENTER(ufp_rd, fip, i);
+		if (((fp_rd = ufp_rd->uf_file) != NULL) &&
+		    (fp_rd->f_vnode == vp_rd))
+			break;
+		UF_EXIT(ufp_rd);
+	}
+	if (i == fip->fi_nfiles) {
+		/* Didn't find it. */
+		mutex_exit(&fip->fi_lock);
+		mutex_enter(&prp->p_lock);
+		sprunlock(prp);
+		return (-1);
+	}
+
+	/*
+	 * Seems the automounter still has the read end of the fifo
+	 * open, we're done here.  Release all our locks and exit.
+	 */
+	mutex_exit(&fip->fi_lock);
+	UF_EXIT(ufp_rd);
+	mutex_enter(&prp->p_lock);
+	sprunlock(prp);
+
+	return (0);
+}
+
+static int
+i_fifo_write(lx_autofs_vfs_t *data, lx_autofs_pkt_t *lap)
+{
+	struct uio	uio;
+	struct iovec	iov;
+	file_t		*fp_wr, *fp_rd;
+	int		error;
+
+	/*
+	 * The catch here is we need to make sure _we_ don't close
+	 * the the fifo while writing to it.  (Another thread could come
+	 * along and realize the automounter process is gone and close
+	 * the fifo.  To do this we bump the open count before we
+	 * write to the fifo.
+	 */
+	mutex_enter(&data->lav_lock);
+	if (data->lav_fifo_wr == NULL) {
+		ASSERT(data->lav_fifo_rd == NULL);
+		mutex_exit(&data->lav_lock);
+		return (ENOENT);
+	}
+	fp_wr = data->lav_fifo_wr;
+	fp_rd = data->lav_fifo_rd;
+
+	/* Bump the open count on the write fifo. */
+	mutex_enter(&fp_wr->f_tlock);
+	fp_wr->f_count++;
+	mutex_exit(&fp_wr->f_tlock);
+
+	/* Bump the open count on the read fifo. */
+	mutex_enter(&fp_rd->f_tlock);
+	fp_rd->f_count++;
+	mutex_exit(&fp_rd->f_tlock);
+
+	mutex_exit(&data->lav_lock);
+
+	iov.iov_base = (caddr_t)lap;
+	iov.iov_len = sizeof (*lap);
+	uio.uio_iov = &iov;
+	uio.uio_iovcnt = 1;
+	uio.uio_loffset = 0;
+	uio.uio_segflg = (short)UIO_SYSSPACE;
+	uio.uio_resid = sizeof (*lap);
+	uio.uio_llimit = 0;
+	uio.uio_fmode = FWRITE | FNDELAY | FNONBLOCK;
+
+	error = VOP_WRITE(fp_wr->f_vnode, &uio, 0, kcred, NULL);
+	(void) closef(fp_wr);
+	(void) closef(fp_rd);
+
+	/*
+	 * After every write we verify that the automounter still has
+	 * these files open.
+	 */
+	mutex_enter(&data->lav_lock);
+	if (i_fifo_verify_rd(data) != 0) {
+		/*
+		 * Something happened to the automounter.
+		 * Close down the communication pipe we setup.
+		 */
+		mutex_exit(&data->lav_lock);
+		i_fifo_close(data);
+		if (error != 0)
+			return (error);
+		return (ENOENT);
+	}
+	mutex_exit(&data->lav_lock);
+
+	return (error);
+}
+
+static int
+i_bs_readdir(vnode_t *dvp, list_t *dir_stack, list_t *file_stack)
+{
+	struct iovec	iov;
+	struct uio	uio;
+	dirent64_t	*dp, *dbuf;
+	vnode_t		*vp;
+	size_t		dlen, dbuflen;
+	int		eof, error, ndirents = 64;
+	char		*nm;
+
+	dlen = ndirents * (sizeof (*dbuf));
+	dbuf = kmem_alloc(dlen, KM_SLEEP);
+
+	uio.uio_iov = &iov;
+	uio.uio_iovcnt = 1;
+	uio.uio_segflg = UIO_SYSSPACE;
+	uio.uio_fmode = 0;
+	uio.uio_extflg = UIO_COPY_CACHED;
+	uio.uio_loffset = 0;
+	uio.uio_llimit = MAXOFFSET_T;
+
+	eof = 0;
+	error = 0;
+	while (!error && !eof) {
+		uio.uio_resid = dlen;
+		iov.iov_base = (char *)dbuf;
+		iov.iov_len = dlen;
+
+		(void) VOP_RWLOCK(dvp, V_WRITELOCK_FALSE, NULL);
+		if (VOP_READDIR(dvp, &uio, kcred, &eof, NULL, 0) != 0) {
+			VOP_RWUNLOCK(dvp, V_WRITELOCK_FALSE, NULL);
+			kmem_free(dbuf, dlen);
+			return (-1);
+		}
+		VOP_RWUNLOCK(dvp, V_WRITELOCK_FALSE, NULL);
+
+		if ((dbuflen = dlen - uio.uio_resid) == 0) {
+			/* We're done. */
+			break;
+		}
+
+		for (dp = dbuf; ((intptr_t)dp < (intptr_t)dbuf + dbuflen);
+			dp = (dirent64_t *)((intptr_t)dp + dp->d_reclen)) {
+
+			nm = dp->d_name;
+
+			if (strcmp(nm, ".") == 0 || strcmp(nm, "..") == 0)
+				continue;
+
+			if (VOP_LOOKUP(dvp, nm, &vp, NULL, 0, NULL, kcred,
+			    NULL, NULL, NULL) != 0) {
+				kmem_free(dbuf, dlen);
+				return (-1);
+			}
+			if (vp->v_type == VDIR) {
+				if (dir_stack != NULL) {
+					i_stack_push(dir_stack, (caddr_t)dvp,
+					    (caddr_t)vp, i_strdup(nm));
+				} else {
+					VN_RELE(vp);
+				}
+			} else {
+				if (file_stack != NULL) {
+					i_stack_push(file_stack, (caddr_t)dvp,
+					    (caddr_t)vp, i_strdup(nm));
+				} else {
+					VN_RELE(vp);
+				}
+			}
+		}
+	}
+	kmem_free(dbuf, dlen);
+	return (0);
+}
+
+static void
+i_bs_destroy(vnode_t *dvp, char *path)
+{
+	list_t	search_stack;
+	list_t	dir_stack;
+	list_t	file_stack;
+	vnode_t	*pdvp, *vp;
+	char	*dpath, *fpath;
+	int	ret;
+
+	if (VOP_LOOKUP(dvp, path, &vp, NULL, 0, NULL, kcred,
+	    NULL, NULL, NULL) != 0) {
+		/* A directory entry with this name doesn't actually exist. */
+		return;
+	}
+
+	if ((vp->v_type & VDIR) == 0) {
+		/* Easy, the directory entry is a file so delete it. */
+		VN_RELE(vp);
+		(void) VOP_REMOVE(dvp, path, kcred, NULL, 0);
+		return;
+	}
+
+	/*
+	 * The directory entry is a subdirectory, now we have a bit more
+	 * work to do.  (We'll have to recurse into the sub directory.)
+	 * It would have been much easier to do this recursively but kernel
+	 * stacks are notoriously small.
+	 */
+	i_stack_init(&search_stack);
+	i_stack_init(&dir_stack);
+	i_stack_init(&file_stack);
+
+	/* Save our newfound subdirectory into a list. */
+	i_stack_push(&search_stack, (caddr_t)dvp, (caddr_t)vp, i_strdup(path));
+
+	/* Do a recursive depth first search into the subdirectories. */
+	while (i_stack_pop(&search_stack,
+	    (caddr_t *)&pdvp, (caddr_t *)&dvp, &dpath) == 0) {
+
+		/* Get a list of the subdirectories in this directory. */
+		if (i_bs_readdir(dvp, &search_stack, NULL) != 0)
+			goto exit;
+
+		/* Save the current directory a separate stack. */
+		i_stack_push(&dir_stack, (caddr_t)pdvp, (caddr_t)dvp, dpath);
+	}
+
+	/*
+	 * Now dir_stack contains a list of directories, the deepest paths
+	 * are at the top of the list.  So let's go through and process them.
+	 */
+	while (i_stack_pop(&dir_stack,
+	    (caddr_t *)&pdvp, (caddr_t *)&dvp, &dpath) == 0) {
+
+		/* Get a list of the files in this directory. */
+		if (i_bs_readdir(dvp, NULL, &file_stack) != 0) {
+			VN_RELE(dvp);
+			i_strfree(dpath);
+			goto exit;
+		}
+
+		/* Delete all the files in this directory. */
+		while (i_stack_pop(&file_stack,
+		    NULL, (caddr_t *)&vp, &fpath) == 0) {
+			VN_RELE(vp)
+			ret = VOP_REMOVE(dvp, fpath, kcred, NULL, 0);
+			i_strfree(fpath);
+			if (ret != 0) {
+				i_strfree(dpath);
+				goto exit;
+			}
+		}
+
+		/* Delete this directory. */
+		VN_RELE(dvp);
+		ret = VOP_RMDIR(pdvp, dpath, pdvp, kcred, NULL, 0);
+		i_strfree(dpath);
+		if (ret != 0)
+			goto exit;
+	}
+
+exit:
+	while (
+	    (i_stack_pop(&search_stack, NULL, (caddr_t *)&vp, &path) == 0) ||
+	    (i_stack_pop(&dir_stack, NULL, (caddr_t *)&vp, &path) == 0) ||
+	    (i_stack_pop(&file_stack, NULL, (caddr_t *)&vp, &path) == 0)) {
+		VN_RELE(vp);
+		i_strfree(path);
+	}
+	i_stack_fini(&search_stack);
+	i_stack_fini(&dir_stack);
+	i_stack_fini(&file_stack);
+}
+
+static vnode_t *
+i_bs_create(vnode_t *dvp, char *bs_name)
+{
+	vnode_t	*vp;
+	vattr_t	vattr;
+
+	/*
+	 * After looking at the mkdir syscall path it seems we don't need
+	 * to initialize all of the vattr_t structure.
+	 */
+	bzero(&vattr, sizeof (vattr));
+	vattr.va_type = VDIR;
+	vattr.va_mode = 0755; /* u+rwx,og=rx */
+	vattr.va_mask = AT_TYPE|AT_MODE;
+
+	if (VOP_MKDIR(dvp, bs_name, &vattr, &vp, kcred, NULL, 0, NULL) != 0)
+		return (NULL);
+	return (vp);
+}
+
+static int
+i_automounter_call(vnode_t *dvp, char *nm)
+{
+	lx_autofs_lookup_req_t	*lalr;
+	lx_autofs_vfs_t		*data;
+	int			error, dup_request;
+
+	/* Get a pointer to the vfs mount data. */
+	data = dvp->v_vfsp->vfs_data;
+
+	/* The automounter only support queries in the root directory. */
+	if (dvp != data->lav_root)
+		return (ENOENT);
+
+	/*
+	 * Check if the current process is in the automounters process
+	 * group.  (If it is, the current process is either the autmounter
+	 * itself or one of it's forked child processes.)  If so, don't
+	 * redirect this lookup back into the automounter because we'll
+	 * hang.
+	 */
+	mutex_enter(&pidlock);
+	if (data->lav_pgrp == curproc->p_pgrp) {
+		mutex_exit(&pidlock);
+		return (ENOENT);
+	}
+	mutex_exit(&pidlock);
+
+	/* Verify that the automount process pipe still exists. */
+	mutex_enter(&data->lav_lock);
+	if (data->lav_fifo_wr == NULL) {
+		ASSERT(data->lav_fifo_rd == NULL);
+		mutex_exit(&data->lav_lock);
+		return (ENOENT);
+	}
+	mutex_exit(&data->lav_lock);
+
+	/* Allocate an automounter request structure. */
+	if ((lalr = i_lalr_alloc(data, &dup_request, nm)) == NULL)
+		return (ENOENT);
+
+	/*
+	 * If we were the first one to allocate this request then we
+	 * need to send it to the automounter.
+	 */
+	if ((!dup_request) &&
+	    ((error = i_fifo_write(data, &lalr->lalr_pkt)) != 0)) {
+		/*
+		 * Unable to send the request to the automounter.
+		 * Unblock any other threads waiting on the request
+		 * and release the request.
+		 */
+		i_lalr_complete(data, lalr);
+		i_lalr_release(data, lalr);
+		return (error);
+	}
+
+	/* Wait for someone to signal us that this request has compleated. */
+	mutex_enter(&lalr->lalr_lock);
+	while (!lalr->lalr_complete) {
+		if (cv_wait_sig(&lalr->lalr_cv, &lalr->lalr_lock) == 0) {
+			/* We got a signal, abort this lookup. */
+			mutex_exit(&lalr->lalr_lock);
+			i_lalr_abort(data, lalr);
+			return (EINTR);
+		}
+	}
+	mutex_exit(&lalr->lalr_lock);
+	i_lalr_release(data, lalr);
+
+	return (0);
+}
+
+static int
+i_automounter_ioctl(vnode_t *vp, int cmd, intptr_t arg)
+{
+	lx_autofs_vfs_t *data = (lx_autofs_vfs_t *)vp->v_vfsp->vfs_data;
+
+	/*
+	 * Be strict.
+	 * We only accept ioctls from the automounter process group.
+	 */
+	mutex_enter(&pidlock);
+	if (data->lav_pgrp != curproc->p_pgrp) {
+		mutex_exit(&pidlock);
+		return (ENOENT);
+	}
+	mutex_exit(&pidlock);
+
+	if ((cmd == LX_AUTOFS_IOC_READY) || (cmd == LX_AUTOFS_IOC_FAIL)) {
+		lx_autofs_lookup_req_t	*lalr;
+		int			id = arg;
+
+		/*
+		 * We don't actually care if the request failed or succeeded.
+		 * We do the same thing either way.
+		 */
+		if ((lalr = i_lalr_find(data, id)) == NULL)
+			return (ENXIO);
+
+		/* Mark the request as compleate and release it. */
+		i_lalr_complete(data, lalr);
+		i_lalr_release(data, lalr);
+		return (0);
+	}
+	if (cmd == LX_AUTOFS_IOC_CATATONIC) {
+		/* The automounter is shutting down. */
+		i_fifo_close(data);
+		return (0);
+	}
+	return (ENOTSUP);
+}
+
+static int
+i_parse_mntopt(vfs_t *vfsp, lx_autofs_vfs_t *data)
+{
+	char		*fd_str, *pgrp_str, *minproto_str, *maxproto_str;
+	int		fd, pgrp, minproto, maxproto;
+	file_t		*fp_wr, *fp_rd;
+
+	/* Require all options to be present. */
+	if ((vfs_optionisset(vfsp, LX_MNTOPT_FD, &fd_str) != 1) ||
+	    (vfs_optionisset(vfsp, LX_MNTOPT_PGRP, &pgrp_str) != 1) ||
+	    (vfs_optionisset(vfsp, LX_MNTOPT_MINPROTO, &minproto_str) != 1) ||
+	    (vfs_optionisset(vfsp, LX_MNTOPT_MAXPROTO, &maxproto_str) != 1))
+		return (EINVAL);
+
+	/* Get the values for each parameter. */
+	if ((i_str_to_int(fd_str, &fd) != 0) ||
+	    (i_str_to_int(pgrp_str, &pgrp) != 0) ||
+	    (i_str_to_int(minproto_str, &minproto) != 0) ||
+	    (i_str_to_int(maxproto_str, &maxproto) != 0))
+		return (EINVAL);
+
+	/*
+	 * We support v2 of the linux kernel automounter protocol.
+	 * Make sure the mount request we got indicates support
+	 * for this version of the protocol.
+	 */
+	if ((minproto > 2) || (maxproto < 2))
+		return (EINVAL);
+
+	/*
+	 * Now we need to lookup the fifos we'll be using
+	 * to talk to the userland automounter process.
+	 */
+	if (i_fifo_lookup(pgrp, fd, &fp_wr, &fp_rd) != 0)
+		return (EINVAL);
+
+	/* Save the mount options and fifo pointers. */
+	data->lav_fd = fd;
+	data->lav_pgrp = pgrp;
+	data->lav_fifo_rd = fp_rd;
+	data->lav_fifo_wr = fp_wr;
+	return (0);
+}
+
+/*
+ * VFS entry points
+ */
+static int
+lx_autofs_mount(vfs_t *vfsp, vnode_t *mvp, struct mounta *uap, cred_t *cr)
+{
+	lx_autofs_vfs_t	*data;
+	dev_t		dev;
+	char		name[40];
+	int		error;
+
+	if (secpolicy_fs_mount(cr, mvp, vfsp) != 0)
+		return (EPERM);
+
+	if (mvp->v_type != VDIR)
+		return (ENOTDIR);
+
+	if ((uap->flags & MS_OVERLAY) == 0 &&
+	    (mvp->v_count > 1 || (mvp->v_flag & VROOT)))
+		return (EBUSY);
+
+	/* We don't support mountes in the global zone. */
+	if (getzoneid() == GLOBAL_ZONEID)
+		return (EPERM);
+
+	/* We don't support mounting on top of ourselves. */
+	if (vn_matchops(mvp, lx_autofs_vn_ops))
+		return (EPERM);
+
+	/* Allocate a vfs struct. */
+	data = kmem_zalloc(sizeof (lx_autofs_vfs_t), KM_SLEEP);
+
+	/* Parse mount options. */
+	if ((error = i_parse_mntopt(vfsp, data)) != 0) {
+		kmem_free(data, sizeof (lx_autofs_vfs_t));
+		return (error);
+	}
+
+	/* Initialize the backing store. */
+	i_bs_destroy(mvp, LX_AUTOFS_BS_DIR);
+	if ((data->lav_bs_vp = i_bs_create(mvp, LX_AUTOFS_BS_DIR)) == NULL) {
+		kmem_free(data, sizeof (lx_autofs_vfs_t));
+		return (EBUSY);
+	}
+	data->lav_bs_name = LX_AUTOFS_BS_DIR;
+
+	/* We have to hold the underlying vnode we're mounted on. */
+	data->lav_mvp = mvp;
+	VN_HOLD(mvp);
+
+	/* Initialize vfs fields */
+	vfsp->vfs_bsize = DEV_BSIZE;
+	vfsp->vfs_fstype = lx_autofs_fstype;
+	vfsp->vfs_data = data;
+
+	/* Invent a dev_t (sigh) */
+	do {
+		dev = makedevice(lx_autofs_major,
+		    atomic_add_32_nv(&lx_autofs_minor, 1) & L_MAXMIN32);
+	} while (vfs_devismounted(dev));
+	vfsp->vfs_dev = dev;
+	vfs_make_fsid(&vfsp->vfs_fsid, dev, lx_autofs_fstype);
+
+	/* Create an id space arena for automounter requests. */
+	(void) snprintf(name, sizeof (name), "lx_autofs_id_%d",
+	    getminor(vfsp->vfs_dev));
+	data->lav_ids = id_space_create(name, 1, INT_MAX);
+
+	/* Create hashes to keep track of automounter requests. */
+	mutex_init(&data->lav_lock, NULL, MUTEX_DEFAULT, NULL);
+	(void) snprintf(name, sizeof (name), "lx_autofs_path_hash_%d",
+	    getminor(vfsp->vfs_dev));
+	data->lav_path_hash = mod_hash_create_strhash(name,
+	    LX_AUTOFS_VFS_PATH_HASH_SIZE, mod_hash_null_valdtor);
+	(void) snprintf(name, sizeof (name), "lx_autofs_id_hash_%d",
+	    getminor(vfsp->vfs_dev));
+	data->lav_id_hash = mod_hash_create_idhash(name,
+	    LX_AUTOFS_VFS_ID_HASH_SIZE, mod_hash_null_valdtor);
+
+	/* Create a hash to keep track of vnodes. */
+	(void) snprintf(name, sizeof (name), "lx_autofs_vn_hash_%d",
+	    getminor(vfsp->vfs_dev));
+	data->lav_vn_hash = mod_hash_create_ptrhash(name,
+	    LX_AUTOFS_VFS_VN_HASH_SIZE, mod_hash_null_valdtor,
+	    sizeof (vnode_t));
+
+	/* Create root vnode */
+	data->lav_root = i_vn_alloc(vfsp, data->lav_bs_vp);
+	data->lav_root->v_flag |=
+	    VROOT | VNOCACHE | VNOMAP | VNOSWAP | VNOMOUNT;
+
+	return (0);
+}
+
+static int
+lx_autofs_unmount(vfs_t *vfsp, int flag, struct cred *cr)
+{
+	lx_autofs_vfs_t *data;
+
+	if (secpolicy_fs_unmount(cr, vfsp) != 0)
+		return (EPERM);
+
+	/* We do not currently support forced unmounts. */
+	if (flag & MS_FORCE)
+		return (ENOTSUP);
+
+	/*
+	 * We should never have a reference count of less than 2: one for the
+	 * caller, one for the root vnode.
+	 */
+	ASSERT(vfsp->vfs_count >= 2);
+
+	/* If there are any outstanding vnodes, we can't unmount. */
+	if (vfsp->vfs_count > 2)
+		return (EBUSY);
+
+	/* Check for any remaining holds on the root vnode. */
+	data = vfsp->vfs_data;
+	ASSERT(data->lav_root->v_vfsp == vfsp);
+	if (data->lav_root->v_count > 1)
+		return (EBUSY);
+
+	/* Close the fifo to the automount process. */
+	if (data->lav_fifo_wr != NULL)
+		(void) closef(data->lav_fifo_wr);
+	if (data->lav_fifo_rd != NULL)
+		(void) closef(data->lav_fifo_rd);
+
+	/*
+	 * We have to release our hold on our root vnode before we can
+	 * delete the backing store.  (Since the root vnode is linked
+	 * to the backing store.)
+	 */
+	VN_RELE(data->lav_root);
+
+	/* Cleanup the backing store. */
+	i_bs_destroy(data->lav_mvp, data->lav_bs_name);
+	VN_RELE(data->lav_mvp);
+
+	/* Cleanup out remaining data structures. */
+	mod_hash_destroy_strhash(data->lav_path_hash);
+	mod_hash_destroy_idhash(data->lav_id_hash);
+	mod_hash_destroy_ptrhash(data->lav_vn_hash);
+	id_space_destroy(data->lav_ids);
+	kmem_free(data, sizeof (lx_autofs_vfs_t));
+
+	return (0);
+}
+
+static int
+lx_autofs_root(vfs_t *vfsp, vnode_t **vpp)
+{
+	lx_autofs_vfs_t	*data = vfsp->vfs_data;
+
+	*vpp = data->lav_root;
+	VN_HOLD(*vpp);
+
+	return (0);
+}
+
+static int
+lx_autofs_statvfs(vfs_t *vfsp, statvfs64_t *sp)
+{
+	lx_autofs_vfs_t	*data = vfsp->vfs_data;
+	vnode_t		*urvp = data->lav_root->v_data;
+	dev32_t		d32;
+	int		error;
+
+	if ((error = VFS_STATVFS(urvp->v_vfsp, sp)) != 0)
+		return (error);
+
+	/* Update some of values before returning. */
+	(void) cmpldev(&d32, vfsp->vfs_dev);
+	sp->f_fsid = d32;
+	(void) strlcpy(sp->f_basetype, vfssw[vfsp->vfs_fstype].vsw_name,
+	    sizeof (sp->f_basetype));
+	sp->f_flag = vf_to_stf(vfsp->vfs_flag);
+	bzero(sp->f_fstr, sizeof (sp->f_fstr));
+	return (0);
+}
+
+static const fs_operation_def_t lx_autofs_vfstops[] = {
+	{ VFSNAME_MOUNT,	{ .vfs_mount = lx_autofs_mount } },
+	{ VFSNAME_UNMOUNT,	{ .vfs_unmount = lx_autofs_unmount } },
+	{ VFSNAME_ROOT,		{ .vfs_root = lx_autofs_root } },
+	{ VFSNAME_STATVFS,	{ .vfs_statvfs = lx_autofs_statvfs } },
+	{ NULL, NULL }
+};
+
+/*
+ * VOP entry points - simple passthrough
+ *
+ * For most VOP entry points we can simply pass the request on to
+ * the underlying filesystem we're mounted on.
+ */
+static int
+lx_autofs_close(vnode_t *vp, int flag, int count, offset_t offset, cred_t *cr,
+    caller_context_t *ctp)
+{
+	vnode_t *uvp = vp->v_data;
+	return (VOP_CLOSE(uvp, flag, count, offset, cr, ctp));
+}
+
+static int
+lx_autofs_readdir(vnode_t *vp, uio_t *uiop, cred_t *cr, int *eofp,
+    caller_context_t *ctp, int flags)
+{
+	vnode_t *uvp = vp->v_data;
+	return (VOP_READDIR(uvp, uiop, cr, eofp, ctp, flags));
+}
+
+static int
+lx_autofs_access(vnode_t *vp, int mode, int flags, cred_t *cr,
+    caller_context_t *ctp)
+{
+	vnode_t *uvp = vp->v_data;
+	return (VOP_ACCESS(uvp, mode, flags, cr, ctp));
+}
+
+static int
+lx_autofs_rwlock(struct vnode *vp, int write_lock, caller_context_t *ctp)
+{
+	vnode_t *uvp = vp->v_data;
+	return (VOP_RWLOCK(uvp, write_lock, ctp));
+}
+
+static void
+lx_autofs_rwunlock(struct vnode *vp, int write_lock, caller_context_t *ctp)
+{
+	vnode_t *uvp = vp->v_data;
+	VOP_RWUNLOCK(uvp, write_lock, ctp);
+}
+
+/*ARGSUSED*/
+static int
+lx_autofs_rmdir(vnode_t *dvp, char *nm, vnode_t *cdir, cred_t *cr,
+    caller_context_t *ctp, int flags)
+{
+	vnode_t *udvp = dvp->v_data;
+
+	/*
+	 * cdir is the calling processes current directory.
+	 * If cdir is lx_autofs vnode then get its real underlying
+	 * vnode ptr.  (It seems like the only thing cdir is
+	 * ever used for is to make sure the user doesn't delete
+	 * their current directory.)
+	 */
+	if (vn_matchops(cdir, lx_autofs_vn_ops)) {
+		vnode_t *ucdir = cdir->v_data;
+		return (VOP_RMDIR(udvp, nm, ucdir, cr, ctp, flags));
+	}
+
+	return (VOP_RMDIR(udvp, nm, cdir, cr, ctp, flags));
+}
+
+/*
+ * VOP entry points - special passthrough
+ *
+ * For some VOP entry points we will first pass the request on to
+ * the underlying filesystem we're mounted on.  If there's an error
+ * then we immediately return the error, but if the request succeeds
+ * we have to do some extra work before returning.
+ */
+static int
+lx_autofs_open(vnode_t **vpp, int flag, cred_t *cr, caller_context_t *ctp)
+{
+	vnode_t		*ovp = *vpp;
+	vnode_t		*uvp = ovp->v_data;
+	int		error;
+
+	if ((error = VOP_OPEN(&uvp, flag, cr, ctp)) != 0)
+		return (error);
+
+	/* Check for clone opens. */
+	if (uvp == ovp->v_data)
+		return (0);
+
+	/* Deal with clone opens by returning a new vnode. */
+	*vpp = i_vn_alloc(ovp->v_vfsp, uvp);
+	VN_RELE(ovp);
+	return (0);
+}
+
+static int
+lx_autofs_getattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr,
+    caller_context_t *ctp)
+{
+	vnode_t		*uvp = vp->v_data;
+	int		error;
+
+	if ((error = VOP_GETATTR(uvp, vap, flags, cr, ctp)) != 0)
+		return (error);
+
+	/* Update the attributes with our filesystem id. */
+	vap->va_fsid = vp->v_vfsp->vfs_dev;
+	return (0);
+}
+
+static int
+lx_autofs_mkdir(vnode_t *dvp, char *nm, struct vattr *vap, vnode_t **vpp,
+    cred_t *cr, caller_context_t *ctp, int flags, vsecattr_t *vsecp)
+{
+	vnode_t		*udvp = dvp->v_data;
+	vnode_t		*uvp = NULL;
+	int		error;
+
+	if ((error = VOP_MKDIR(udvp, nm, vap, &uvp, cr,
+	    ctp, flags, vsecp)) != 0)
+		return (error);
+
+	/* Update the attributes with our filesystem id. */
+	vap->va_fsid = dvp->v_vfsp->vfs_dev;
+
+	/* Allocate a new vnode. */
+	*vpp = i_vn_alloc(dvp->v_vfsp, uvp);
+	return (0);
+}
+
+/*
+ * VOP entry points - custom
+ */
+/*ARGSUSED*/
+static void
+lx_autofs_inactive(struct vnode *vp, struct cred *cr, caller_context_t *ctp)
+{
+	lx_autofs_vfs_t	*data = vp->v_vfsp->vfs_data;
+
+	/*
+	 * We need to hold the vfs lock because if we're going to free
+	 * this vnode we have to prevent anyone from looking it up
+	 * in the vnode hash.
+	 */
+	mutex_enter(&data->lav_lock);
+	mutex_enter(&vp->v_lock);
+
+	if (vp->v_count < 1) {
+		panic("lx_autofs_inactive: bad v_count");
+		/*NOTREACHED*/
+	}
+
+	/* Drop the temporary hold by vn_rele now. */
+	if (--vp->v_count > 0) {
+		mutex_exit(&vp->v_lock);
+		mutex_exit(&data->lav_lock);
+		return;
+	}
+
+	/*
+	 * No one should have been blocked on this lock because we're
+	 * about to free this vnode.
+	 */
+	i_vn_free(vp);
+}
+
+static int
+lx_autofs_lookup(vnode_t *dvp, char *nm, vnode_t **vpp, struct pathname *pnp,
+    int flags, vnode_t *rdir, cred_t *cr, caller_context_t *ctp,
+    int *direntflags, pathname_t *realpnp)
+{
+	vnode_t			*udvp = dvp->v_data;
+	vnode_t			*uvp = NULL;
+	int			error;
+
+	/* First try to lookup if this path component already exitst. */
+	if ((error = VOP_LOOKUP(udvp, nm, &uvp, pnp, flags, rdir, cr, ctp,
+	    direntflags, realpnp)) == 0) {
+		*vpp = i_vn_alloc(dvp->v_vfsp, uvp);
+		return (0);
+	}
+
+	/* Only query the automounter if the path does not exist. */
+	if (error != ENOENT)
+		return (error);
+
+	/* Refer the lookup to the automounter. */
+	if ((error = i_automounter_call(dvp, nm)) != 0)
+		return (error);
+
+	/* Retry the lookup operation. */
+	if ((error = VOP_LOOKUP(udvp, nm, &uvp, pnp, flags, rdir, cr, ctp,
+	    direntflags, realpnp)) == 0) {
+		*vpp = i_vn_alloc(dvp->v_vfsp, uvp);
+		return (0);
+	}
+	return (error);
+}
+
+/*ARGSUSED*/
+static int
+lx_autofs_ioctl(vnode_t *vp, int cmd, intptr_t arg, int mode, cred_t *cr,
+    int *rvalp, caller_context_t *ctp)
+{
+	vnode_t			*uvp = vp->v_data;
+
+	/* Intercept certain ioctls. */
+	switch ((uint_t)cmd) {
+	case LX_AUTOFS_IOC_READY:
+	case LX_AUTOFS_IOC_FAIL:
+	case LX_AUTOFS_IOC_CATATONIC:
+	case LX_AUTOFS_IOC_EXPIRE:
+	case LX_AUTOFS_IOC_PROTOVER:
+	case LX_AUTOFS_IOC_SETTIMEOUT:
+		return (i_automounter_ioctl(vp, cmd, arg));
+	}
+
+	/* Pass any remaining ioctl on. */
+	return (VOP_IOCTL(uvp, cmd, arg, mode, cr, rvalp, ctp));
+}
+
+/*
+ * VOP entry points definitions
+ */
+static const fs_operation_def_t lx_autofs_tops_root[] = {
+	{ VOPNAME_OPEN,		{ .vop_open = lx_autofs_open } },
+	{ VOPNAME_CLOSE,	{ .vop_close = lx_autofs_close } },
+	{ VOPNAME_IOCTL,	{ .vop_ioctl = lx_autofs_ioctl } },
+	{ VOPNAME_RWLOCK,	{ .vop_rwlock = lx_autofs_rwlock } },
+	{ VOPNAME_RWUNLOCK,	{ .vop_rwunlock = lx_autofs_rwunlock } },
+	{ VOPNAME_GETATTR,	{ .vop_getattr = lx_autofs_getattr } },
+	{ VOPNAME_ACCESS,	{ .vop_access = lx_autofs_access } },
+	{ VOPNAME_READDIR,	{ .vop_readdir = lx_autofs_readdir } },
+	{ VOPNAME_LOOKUP,	{ .vop_lookup = lx_autofs_lookup } },
+	{ VOPNAME_INACTIVE,	{ .vop_inactive = lx_autofs_inactive } },
+	{ VOPNAME_MKDIR,	{ .vop_mkdir = lx_autofs_mkdir } },
+	{ VOPNAME_RMDIR,	{ .vop_rmdir = lx_autofs_rmdir } },
+	{ NULL }
+};
+
+/*
+ * lx_autofs_init() gets invoked via the mod_install() call in
+ * this modules _init() routine.  Therefor, the code that cleans
+ * up the structures we allocate below is actually found in
+ * our _fini() routine.
+ */
+/* ARGSUSED */
+static int
+lx_autofs_init(int fstype, char *name)
+{
+	int		error;
+
+	if ((lx_autofs_major =
+	    (major_t)space_fetch(LX_AUTOFS_SPACE_KEY_UDEV)) == 0) {
+
+		if ((lx_autofs_major = getudev()) == (major_t)-1) {
+			cmn_err(CE_WARN, "lx_autofs_init: "
+			    "can't get unique device number");
+			return (EAGAIN);
+		}
+
+		if (space_store(LX_AUTOFS_SPACE_KEY_UDEV,
+		    (uintptr_t)lx_autofs_major) != 0) {
+			cmn_err(CE_WARN, "lx_autofs_init: "
+			    "can't save unique device number");
+			return (EAGAIN);
+		}
+	}
+
+	lx_autofs_fstype = fstype;
+	if ((error = vfs_setfsops(
+	    fstype, lx_autofs_vfstops, &lx_autofs_vfsops)) != 0) {
+		cmn_err(CE_WARN, "lx_autofs_init: bad vfs ops template");
+		return (error);
+	}
+
+	if ((error = vn_make_ops("lx_autofs vnode ops",
+	    lx_autofs_tops_root, &lx_autofs_vn_ops)) != 0) {
+		VERIFY(vfs_freevfsops_by_type(fstype) == 0);
+		lx_autofs_vn_ops = NULL;
+		return (error);
+	}
+
+	return (0);
+}
+
+
+/*
+ * Module linkage
+ */
+static mntopt_t lx_autofs_mntopt[] = {
+	{ LX_MNTOPT_FD,		NULL,	0,	MO_HASVALUE },
+	{ LX_MNTOPT_PGRP,	NULL,	0,	MO_HASVALUE },
+	{ LX_MNTOPT_MINPROTO,	NULL,	0,	MO_HASVALUE },
+	{ LX_MNTOPT_MAXPROTO,	NULL,	0,	MO_HASVALUE }
+};
+
+static mntopts_t lx_autofs_mntopts = {
+	sizeof (lx_autofs_mntopt) / sizeof (mntopt_t),
+	lx_autofs_mntopt
+};
+
+static vfsdef_t vfw = {
+	VFSDEF_VERSION,
+	LX_AUTOFS_NAME,
+	lx_autofs_init,
+	VSW_HASPROTO | VSW_VOLATILEDEV,
+	&lx_autofs_mntopts
+};
+
+extern struct mod_ops mod_fsops;
+
+static struct modlfs modlfs = {
+	&mod_fsops, "linux autofs filesystem", &vfw
+};
+
+static struct modlinkage modlinkage = {
+	MODREV_1, (void *)&modlfs, NULL
+};
+
+int
+_init(void)
+{
+	return (mod_install(&modlinkage));
+}
+
+int
+_info(struct modinfo *modinfop)
+{
+	return (mod_info(&modlinkage, modinfop));
+}
+
+int
+_fini(void)
+{
+	int		error;
+
+	if ((error = mod_remove(&modlinkage)) != 0)
+		return (error);
+
+	if (lx_autofs_vn_ops != NULL) {
+		vn_freevnodeops(lx_autofs_vn_ops);
+		lx_autofs_vn_ops = NULL;
+	}
+
+	/*
+	 * In our init routine, if we get an error after calling
+	 * vfs_setfsops() we cleanup by calling vfs_freevfsops_by_type().
+	 * But we don't need to call vfs_freevfsops_by_type() here
+	 * because the fs framework did this for us as part of the
+	 * mod_remove() call above.
+	 */
+	return (0);
+}
diff --git a/usr/src/uts/common/brand/lx/dtrace/lx_systrace.c b/usr/src/uts/common/brand/lx/dtrace/lx_systrace.c
new file mode 100644
index 0000000000..510626d220
--- /dev/null
+++ b/usr/src/uts/common/brand/lx/dtrace/lx_systrace.c
@@ -0,0 +1,497 @@
+/*
+ * CDDL HEADER START
+ *
+ * The contents of this file are subject to the terms of the
+ * Common Development and Distribution License (the "License").
+ * You may not use this file except in compliance with the License.
+ *
+ * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
+ * or http://www.opensolaris.org/os/licensing.
+ * See the License for the specific language governing permissions
+ * and limitations under the License.
+ *
+ * When distributing Covered Code, include this CDDL HEADER in each
+ * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
+ * If applicable, add the following below this CDDL HEADER, with the
+ * fields enclosed by brackets "[]" replaced with your own identifying
+ * information: Portions Copyright [yyyy] [name of copyright owner]
+ *
+ * CDDL HEADER END
+ */
+
+/*
+ * Copyright 2009 Sun Microsystems, Inc.  All rights reserved.
+ * Use is subject to license terms.
+ * Copyright 2015 Joyent, Inc.
+ */
+
+
+#include <sys/modctl.h>
+#include <sys/ddi.h>
+#include <sys/sunddi.h>
+#include <sys/stat.h>
+#include <sys/conf.h>
+#include <sys/frame.h>
+#include <sys/dtrace.h>
+#include <sys/dtrace_impl.h>
+
+#include <sys/lx_brand.h>
+#include <sys/lx_impl.h>
+
+/*
+ * We store the syscall number in the low 16 bits (which limits us to 64k
+ * syscalls). The next bit indicates entry/return probe and the next bit
+ * indicates 64bit/32bit syscall.
+ */
+#define	SCALL_MASK	0xffff
+#define	ENTRY_FLAG	0x10000
+#define	SYSC_64_BIT	0x100000
+
+#define	LX_SYSTRACE_IS64BIT(x)	((int)(x) & SYSC_64_BIT)
+#define	LX_SYSTRACE_ISENTRY(x)	((int)(x) & ENTRY_FLAG)
+#define	LX_SYSTRACE_SYSNUM(x)	((int)(x) & SCALL_MASK)
+
+#define	LX_SYSTRACE32_ENTRY(id)	(ENTRY_FLAG | (id))
+#define	LX_SYSTRACE32_RETURN(id) (id)
+
+#define	LX_SYSTRACE64_ENTRY(id)	(SYSC_64_BIT | ENTRY_FLAG | (id))
+#define	LX_SYSTRACE64_RETURN(id) (SYSC_64_BIT | id)
+
+#define	LX_SYSTRACE_ENTRY_AFRAMES	2
+#define	LX_SYSTRACE_RETURN_AFRAMES	4
+
+typedef struct lx_systrace_sysent {
+	const char *lss_name;
+	dtrace_id_t lss_entry;
+	dtrace_id_t lss_return;
+} lx_systrace_sysent_t;
+
+static dev_info_t *lx_systrace_devi;
+static dtrace_provider_id_t lx_systrace_id;
+static kmutex_t lx_systrace_lock;
+static uint_t lx_systrace_nenabled;
+
+static int lx_systrace_nsysent32;
+static lx_systrace_sysent_t *lx_systrace_sysent32;
+
+#if defined(_LP64)
+static int lx_systrace_nsysent64;
+static lx_systrace_sysent_t *lx_systrace_sysent64;
+#endif
+
+/*ARGSUSED*/
+static void
+lx_systrace_entry(ulong_t sysnum, ulong_t arg0, ulong_t arg1, ulong_t arg2,
+    ulong_t arg3, ulong_t arg4, ulong_t arg5)
+{
+	dtrace_id_t id;
+
+#if defined(_LP64)
+	if ((ttoproc(curthread))->p_model == DATAMODEL_NATIVE) {
+		if (sysnum >= lx_systrace_nsysent64)
+			return;
+		id = lx_systrace_sysent64[sysnum].lss_entry;
+	} else
+#endif
+	{
+		if (sysnum >= lx_systrace_nsysent32)
+			return;
+		id = lx_systrace_sysent32[sysnum].lss_entry;
+	}
+
+	if (id == DTRACE_IDNONE)
+		return;
+	dtrace_probe(id, arg0, arg1, arg2, arg3, arg4);
+}
+
+/*ARGSUSED*/
+static void
+lx_systrace_return(ulong_t sysnum, ulong_t arg0, ulong_t arg1, ulong_t arg2,
+    ulong_t arg3, ulong_t arg4, ulong_t arg5)
+{
+	dtrace_id_t id;
+
+#if defined(_LP64)
+	if ((ttoproc(curthread))->p_model == DATAMODEL_NATIVE) {
+		if (sysnum >= lx_systrace_nsysent64)
+			return;
+		id = lx_systrace_sysent64[sysnum].lss_return;
+	} else
+#endif
+	{
+		if (sysnum >= lx_systrace_nsysent32)
+			return;
+		id = lx_systrace_sysent32[sysnum].lss_return;
+	}
+
+	if (id == DTRACE_IDNONE)
+		return;
+	dtrace_probe(id, arg0, arg1, arg2, arg3, arg4);
+}
+
+/*ARGSUSED*/
+static void
+lx_systrace_provide(void *arg, const dtrace_probedesc_t *desc)
+{
+	int i;
+
+	if (desc != NULL)
+		return;
+
+	for (i = 0; i < lx_systrace_nsysent32; i++) {
+		if (dtrace_probe_lookup(lx_systrace_id, "sys32",
+		    lx_systrace_sysent32[i].lss_name, "entry") != 0)
+			continue;
+
+		(void) dtrace_probe_create(lx_systrace_id, "sys32",
+		    lx_systrace_sysent32[i].lss_name, "entry",
+		    LX_SYSTRACE_ENTRY_AFRAMES,
+		    (void *)((uintptr_t)LX_SYSTRACE32_ENTRY(i)));
+
+		(void) dtrace_probe_create(lx_systrace_id, "sys32",
+		    lx_systrace_sysent32[i].lss_name, "return",
+		    LX_SYSTRACE_RETURN_AFRAMES,
+		    (void *)((uintptr_t)LX_SYSTRACE32_RETURN(i)));
+
+		lx_systrace_sysent32[i].lss_entry = DTRACE_IDNONE;
+		lx_systrace_sysent32[i].lss_return = DTRACE_IDNONE;
+	}
+
+#if defined(_LP64)
+	for (i = 0; i < lx_systrace_nsysent64; i++) {
+		if (dtrace_probe_lookup(lx_systrace_id, "sys64",
+		    lx_systrace_sysent64[i].lss_name, "entry") != 0)
+			continue;
+
+		(void) dtrace_probe_create(lx_systrace_id, "sys64",
+		    lx_systrace_sysent64[i].lss_name, "entry",
+		    LX_SYSTRACE_ENTRY_AFRAMES,
+		    (void *)((uintptr_t)LX_SYSTRACE64_ENTRY(i)));
+
+		(void) dtrace_probe_create(lx_systrace_id, "sys64",
+		    lx_systrace_sysent64[i].lss_name, "return",
+		    LX_SYSTRACE_RETURN_AFRAMES,
+		    (void *)((uintptr_t)LX_SYSTRACE64_RETURN(i)));
+
+		lx_systrace_sysent64[i].lss_entry = DTRACE_IDNONE;
+		lx_systrace_sysent64[i].lss_return = DTRACE_IDNONE;
+	}
+#endif
+}
+
+/*ARGSUSED*/
+static int
+lx_systrace_enable(void *arg, dtrace_id_t id, void *parg)
+{
+	int sysnum = LX_SYSTRACE_SYSNUM((uintptr_t)parg);
+
+	mutex_enter(&lx_systrace_lock);
+	if (lx_systrace_nenabled++ == 0)
+		lx_brand_systrace_enable();
+	mutex_exit(&lx_systrace_lock);
+
+	if (LX_SYSTRACE_IS64BIT((uintptr_t)parg)) {
+#if defined(_LP64)
+		ASSERT(sysnum < lx_systrace_nsysent64);
+
+		if (LX_SYSTRACE_ISENTRY((uintptr_t)parg)) {
+			lx_systrace_sysent64[sysnum].lss_entry = id;
+		} else {
+			lx_systrace_sysent64[sysnum].lss_return = id;
+		}
+#endif
+	} else {
+		ASSERT(sysnum < lx_systrace_nsysent32);
+
+		if (LX_SYSTRACE_ISENTRY((uintptr_t)parg)) {
+			lx_systrace_sysent32[sysnum].lss_entry = id;
+		} else {
+			lx_systrace_sysent32[sysnum].lss_return = id;
+		}
+	}
+	return (0);
+}
+
+/*ARGSUSED*/
+static void
+lx_systrace_disable(void *arg, dtrace_id_t id, void *parg)
+{
+	int sysnum = LX_SYSTRACE_SYSNUM((uintptr_t)parg);
+
+	if (LX_SYSTRACE_IS64BIT((uintptr_t)parg)) {
+#if defined(_LP64)
+		ASSERT(sysnum < lx_systrace_nsysent64);
+
+		if (LX_SYSTRACE_ISENTRY((uintptr_t)parg)) {
+			lx_systrace_sysent64[sysnum].lss_entry = DTRACE_IDNONE;
+		} else {
+			lx_systrace_sysent64[sysnum].lss_return = DTRACE_IDNONE;
+		}
+#endif
+	} else {
+		ASSERT(sysnum < lx_systrace_nsysent32);
+
+		if (LX_SYSTRACE_ISENTRY((uintptr_t)parg)) {
+			lx_systrace_sysent32[sysnum].lss_entry = DTRACE_IDNONE;
+		} else {
+			lx_systrace_sysent32[sysnum].lss_return = DTRACE_IDNONE;
+		}
+	}
+
+	mutex_enter(&lx_systrace_lock);
+	if (--lx_systrace_nenabled == 0)
+		lx_brand_systrace_disable();
+	mutex_exit(&lx_systrace_lock);
+}
+
+/*ARGSUSED*/
+static void
+lx_systrace_destroy(void *arg, dtrace_id_t id, void *parg)
+{
+}
+
+/*ARGSUSED*/
+static uint64_t
+lx_systrace_getarg(void *arg, dtrace_id_t id, void *parg, int argno,
+    int aframes)
+{
+	struct frame *fp = (struct frame *)dtrace_getfp();
+	uintptr_t *stack;
+	uint64_t val = 0;
+	int i;
+
+	if (argno >= 6)
+		return (0);
+
+	/*
+	 * Walk the four frames down the stack to the entry or return callback.
+	 * Our callback calls dtrace_probe() which calls dtrace_dif_variable()
+	 * which invokes this function to get the extended arguments. We get
+	 * the frame pointer in via call to dtrace_getfp() above which makes for
+	 * four frames.
+	 */
+	for (i = 0; i < 4; i++) {
+		fp = (struct frame *)fp->fr_savfp;
+	}
+
+	stack = (uintptr_t *)&fp[1];
+
+	/*
+	 * Skip the first argument to the callback -- the system call number.
+	 */
+	argno++;
+
+#ifdef __amd64
+	/*
+	 * On amd64, the first 6 arguments are passed in registers while
+	 * subsequent arguments are on the stack.
+	 */
+	argno -= 6;
+#endif
+
+	DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT);
+	val = stack[argno];
+	DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT);
+
+	return (val);
+}
+
+
+static const dtrace_pattr_t lx_systrace_attr = {
+{ DTRACE_STABILITY_EVOLVING, DTRACE_STABILITY_EVOLVING, DTRACE_CLASS_COMMON },
+{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_UNKNOWN },
+{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_ISA },
+{ DTRACE_STABILITY_EVOLVING, DTRACE_STABILITY_EVOLVING, DTRACE_CLASS_COMMON },
+{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_ISA },
+};
+
+static dtrace_pops_t lx_systrace_pops = {
+	lx_systrace_provide,
+	NULL,
+	lx_systrace_enable,
+	lx_systrace_disable,
+	NULL,
+	NULL,
+	NULL,
+	lx_systrace_getarg,
+	NULL,
+	lx_systrace_destroy
+};
+
+static int
+lx_systrace_attach(dev_info_t *devi, ddi_attach_cmd_t cmd)
+{
+	int i;
+
+	switch (cmd) {
+	case DDI_ATTACH:
+		break;
+	case DDI_RESUME:
+		return (DDI_SUCCESS);
+	default:
+		return (DDI_FAILURE);
+	}
+
+	if (ddi_create_minor_node(devi, "lx_systrace", S_IFCHR,
+	    0, DDI_PSEUDO, NULL) == DDI_FAILURE ||
+	    dtrace_register("lx-syscall", &lx_systrace_attr,
+	    DTRACE_PRIV_USER, 0, &lx_systrace_pops, NULL,
+	    &lx_systrace_id) != 0) {
+		ddi_remove_minor_node(devi, NULL);
+		return (DDI_FAILURE);
+	}
+
+	ddi_report_dev(devi);
+	lx_systrace_devi = devi;
+
+	/*
+	 * Initialize the 32-bit table.
+	 */
+	VERIFY(lx_nsysent32 > 0);
+	lx_systrace_nsysent32 = lx_nsysent32;
+	lx_systrace_sysent32 = kmem_zalloc(lx_systrace_nsysent32 *
+	    sizeof (lx_systrace_sysent_t), KM_SLEEP);
+
+	for (i = 0; i < lx_systrace_nsysent32; i++) {
+		lx_systrace_sysent32[i].lss_name = lx_sysent32[i].sy_name;
+		lx_systrace_sysent32[i].lss_entry = DTRACE_IDNONE;
+		lx_systrace_sysent32[i].lss_return = DTRACE_IDNONE;
+	}
+
+#if defined(_LP64)
+	/*
+	 * Initialize the 64-bit table.
+	 */
+	VERIFY(lx_nsysent64 > 0);
+	lx_systrace_nsysent64 = lx_nsysent64;
+	lx_systrace_sysent64 = kmem_zalloc(lx_systrace_nsysent64 *
+	    sizeof (lx_systrace_sysent_t), KM_SLEEP);
+
+	for (i = 0; i < lx_systrace_nsysent64; i++) {
+		lx_systrace_sysent64[i].lss_name = lx_sysent64[i].sy_name;
+		lx_systrace_sysent64[i].lss_entry = DTRACE_IDNONE;
+		lx_systrace_sysent64[i].lss_return = DTRACE_IDNONE;
+	}
+#endif
+
+	/*
+	 * Install probe triggers.
+	 */
+	lx_systrace_entry_ptr = lx_systrace_entry;
+	lx_systrace_return_ptr = lx_systrace_return;
+
+	return (DDI_SUCCESS);
+}
+
+/*ARGSUSED*/
+static int
+lx_systrace_detach(dev_info_t *devi, ddi_detach_cmd_t cmd)
+{
+	switch (cmd) {
+	case DDI_DETACH:
+		break;
+	case DDI_SUSPEND:
+		return (DDI_SUCCESS);
+	default:
+		return (DDI_FAILURE);
+	}
+
+	if (dtrace_unregister(lx_systrace_id) != 0)
+		return (DDI_FAILURE);
+
+	/*
+	 * Free tables.
+	 */
+	kmem_free(lx_systrace_sysent32, lx_systrace_nsysent32 *
+	    sizeof (lx_systrace_sysent_t));
+	lx_systrace_sysent32 = NULL;
+	lx_systrace_nsysent32 = 0;
+
+#if defined(_LP64)
+	kmem_free(lx_systrace_sysent64, lx_systrace_nsysent64 *
+	    sizeof (lx_systrace_sysent_t));
+	lx_systrace_sysent64 = NULL;
+	lx_systrace_nsysent64 = 0;
+#endif
+
+	/*
+	 * Reset probe triggers.
+	 */
+	lx_systrace_entry_ptr = NULL;
+	lx_systrace_return_ptr = NULL;
+
+	return (DDI_SUCCESS);
+}
+
+/*ARGSUSED*/
+static int
+lx_systrace_open(dev_t *devp, int flag, int otyp, cred_t *cred_p)
+{
+	return (0);
+}
+
+static struct cb_ops lx_systrace_cb_ops = {
+	lx_systrace_open,	/* open */
+	nodev,			/* close */
+	nulldev,		/* strategy */
+	nulldev,		/* print */
+	nodev,			/* dump */
+	nodev,			/* read */
+	nodev,			/* write */
+	nodev,			/* ioctl */
+	nodev,			/* devmap */
+	nodev,			/* mmap */
+	nodev,			/* segmap */
+	nochpoll,		/* poll */
+	ddi_prop_op,		/* cb_prop_op */
+	0,			/* streamtab */
+	D_NEW | D_MP		/* Driver compatibility flag */
+};
+
+static struct dev_ops lx_systrace_ops = {
+	DEVO_REV,		/* devo_rev */
+	0,			/* refcnt */
+	ddi_getinfo_1to1,	/* get_dev_info */
+	nulldev,		/* identify */
+	nulldev,		/* probe */
+	lx_systrace_attach,	/* attach */
+	lx_systrace_detach,	/* detach */
+	nodev,			/* reset */
+	&lx_systrace_cb_ops,	/* driver operations */
+	NULL,			/* bus operations */
+	nodev,			/* dev power */
+	ddi_quiesce_not_needed,		/* quiesce */
+};
+
+/*
+ * Module linkage information for the kernel.
+ */
+static struct modldrv modldrv = {
+	&mod_driverops,		/* module type (this is a pseudo driver) */
+	"Linux Brand System Call Tracing", /* name of module */
+	&lx_systrace_ops	/* driver ops */
+};
+
+static struct modlinkage modlinkage = {
+	MODREV_1,
+	(void *)&modldrv,
+	NULL
+};
+
+int
+_init(void)
+{
+	return (mod_install(&modlinkage));
+}
+
+int
+_info(struct modinfo *modinfop)
+{
+	return (mod_info(&modlinkage, modinfop));
+}
+
+int
+_fini(void)
+{
+	return (mod_remove(&modlinkage));
+}
diff --git a/usr/src/uts/common/brand/lx/dtrace/lx_systrace.conf b/usr/src/uts/common/brand/lx/dtrace/lx_systrace.conf
new file mode 100644
index 0000000000..e4499c8a5b
--- /dev/null
+++ b/usr/src/uts/common/brand/lx/dtrace/lx_systrace.conf
@@ -0,0 +1,27 @@
+#
+# CDDL HEADER START
+#
+# The contents of this file are subject to the terms of the
+# Common Development and Distribution License (the "License").
+# You may not use this file except in compliance with the License.
+#
+# You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
+# or http://www.opensolaris.org/os/licensing.
+# See the License for the specific language governing permissions
+# and limitations under the License.
+#
+# When distributing Covered Code, include this CDDL HEADER in each
+# file and include the License file at usr/src/OPENSOLARIS.LICENSE.
+# If applicable, add the following below this CDDL HEADER, with the
+# fields enclosed by brackets "[]" replaced with your own identifying
+# information: Portions Copyright [yyyy] [name of copyright owner]
+#
+# CDDL HEADER END
+#
+#
+# Copyright 2006 Sun Microsystems, Inc.  All rights reserved.
+# Use is subject to license terms.
+#
+#ident	"%Z%%M%	%I%	%E% SMI"
+
+name="lx_systrace" parent="pseudo" instance=0;
diff --git a/usr/src/uts/common/brand/lx/io/lx_audio.c b/usr/src/uts/common/brand/lx/io/lx_audio.c
new file mode 100644
index 0000000000..e8c6234d92
--- /dev/null
+++ b/usr/src/uts/common/brand/lx/io/lx_audio.c
@@ -0,0 +1,1996 @@
+/*
+ * CDDL HEADER START
+ *
+ * The contents of this file are subject to the terms of the
+ * Common Development and Distribution License (the "License").
+ * You may not use this file except in compliance with the License.
+ *
+ * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
+ * or http://www.opensolaris.org/os/licensing.
+ * See the License for the specific language governing permissions
+ * and limitations under the License.
+ *
+ * When distributing Covered Code, include this CDDL HEADER in each
+ * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
+ * If applicable, add the following below this CDDL HEADER, with the
+ * fields enclosed by brackets "[]" replaced with your own identifying
+ * information: Portions Copyright [yyyy] [name of copyright owner]
+ *
+ * CDDL HEADER END
+ */
+/*
+ * Copyright 2009 Sun Microsystems, Inc.  All rights reserved.
+ * Use is subject to license terms.
+ */
+
+
+#include <sys/audio.h>
+#include <sys/conf.h>
+#include <sys/debug.h>
+#include <sys/disp.h>
+#include <sys/ddi.h>
+#include <sys/file.h>
+#include <sys/id_space.h>
+#include <sys/kmem.h>
+#include <sys/lx_audio.h>
+#include <sys/mixer.h>
+#include <sys/modhash.h>
+#include <sys/stat.h>
+#include <sys/sunddi.h>
+#include <sys/sunldi.h>
+#include <sys/sysmacros.h>
+#include <sys/stropts.h>
+#include <sys/types.h>
+#include <sys/zone.h>
+
+/* Properties used by the lx_audio driver */
+#define	LXA_PROP_INPUTDEV		"inputdev"
+#define	LXA_PROP_OUTPUTDEV		"outputdev"
+
+/* default device paths used by this driver */
+#define	LXA_DEV_DEFAULT			"/dev/audio"
+#define	LXA_DEV_CUSTOM_DIR		"/dev/sound/"
+
+/* maximum possible number of concurrent opens of this driver */
+#define	LX_AUDIO_MAX_OPENS		1024
+
+/*
+ * these are default fragment size and fragment count values.
+ * these values were chosen to make quake work well on my
+ * laptop: 2Ghz Pentium M + NVIDIA GeForce Go 6400.
+ *
+ * for reference:
+ * - 1 sec of stereo output at 44Khz is about 171 Kb of data
+ * - 1 sec of mono output at 8Khz is about 8Kb of data
+ */
+#define	LXA_OSS_FRAG_SIZE		(1024)	/* 1/8 sec at 8Khz mono */
+#define	LXA_OSS_FRAG_CNT		(1024 * 2)
+
+/* maximum ammount of fragment memory we'll allow a process to mmap */
+#define	LXA_OSS_FRAG_MEM		(1024 * 1024 * 2) /* 2Mb */
+
+/* forward declarations */
+typedef struct lxa_state lxa_state_t;
+typedef struct lxa_zstate lxa_zstate_t;
+
+/*
+ * Structure and enum declarations
+ */
+typedef enum {
+	LXA_TYPE_INVALID	= 0,
+	LXA_TYPE_AUDIO		= 1,	/* audio device */
+	LXA_TYPE_AUDIOCTL	= 2	/* audio control/mixer device */
+} lxa_dev_type_t;
+
+struct lxa_zstate {
+	char			*lxa_zs_zonename;
+
+	/*
+	 * we could store the input/output audio device setting here,
+	 * but instead we're keeing them as device node properties
+	 * so that a user can easily see the audio configuration for
+	 * a zone via prtconf.
+	 */
+
+	/*
+	 * OSS doesn't support multiple opens of the audio device.
+	 * (multiple opens of the mixer device are supported.)
+	 * so here we'll keep a pointer to any open input/output
+	 * streams.  (OSS does support two opens if one is for input
+	 * and the other is for output.)
+	 */
+	lxa_state_t		*lxa_zs_istate;
+	lxa_state_t		*lxa_zs_ostate;
+
+	/*
+	 * we need to cache channel gain and balance.  channel gain and
+	 * balance map to PCM volume in OSS, which are supposedly a property
+	 * of the underlying hardware.  but in solaris, channels are
+	 * implemented in software and only exist when an audio device
+	 * is actually open.  (each open returns a unique channel.)  OSS
+	 * apps will expect consistent PCM volume set/get operations to
+	 * work even if no audio device is open.  hence, if no underlying
+	 * device is open we need to cache the gain and balance setting.
+	 */
+	lxa_mixer_levels_t	lxa_zs_pcm_levels;
+};
+
+struct lxa_state {
+	lxa_zstate_t	*lxas_zs;	/* zone state pointer */
+
+	dev_t		lxas_dev_old;	/* dev_t used to open the device */
+	dev_t		lxas_dev_new;	/* new dev_t assigned to an open */
+	int		lxas_flags;	/* original flags passed to open */
+	lxa_dev_type_t	lxas_type;	/* type of device that was opened */
+
+	int		lxas_devs_same;	/* input and output device the same? */
+
+	/* input device variables */
+	ldi_handle_t	lxas_idev_lh;		/* ldi handle for access */
+	int		lxas_idev_flags;	/* flags used for open */
+
+	/* output device variables */
+	ldi_handle_t	lxas_odev_lh;		/* ldi handle for access */
+	int		lxas_odev_flags;	/* flags used for open */
+
+	/*
+	 * since we support multiplexing of devices we need to remember
+	 * certain parameters about the devices
+	 */
+	uint_t		lxas_hw_features;
+	uint_t		lxas_sw_features;
+
+	uint_t		lxas_frag_size;
+	uint_t		lxas_frag_cnt;
+
+	/*
+	 * members needed to support mmap device access.  note that to
+	 * simplifly things we only support one mmap access per open.
+	 */
+	ddi_umem_cookie_t	lxas_umem_cookie;
+	char			*lxas_umem_ptr;
+	size_t			lxas_umem_len;
+	kthread_t		*lxas_mmap_thread;
+	int			lxas_mmap_thread_running;
+	int			lxas_mmap_thread_exit;
+	int			lxas_mmap_thread_frag;
+};
+
+/*
+ * Global variables
+ */
+dev_info_t	*lxa_dip = NULL;
+kmutex_t	lxa_lock;
+id_space_t	*lxa_minor_id = NULL;
+mod_hash_t	*lxa_state_hash = NULL;
+mod_hash_t	*lxa_zstate_hash = NULL;
+size_t		lxa_state_hash_size = 15;
+size_t		lxa_zstate_hash_size = 15;
+size_t		lxa_registered_zones = 0;
+
+/*
+ * function declarations
+ */
+static void lxa_mmap_output_disable(lxa_state_t *);
+
+/*
+ * functions
+ */
+static void
+lxa_state_close(lxa_state_t *lxa_state)
+{
+	lxa_zstate_t		*lxa_zs = lxa_state->lxas_zs;
+	minor_t			minor = getminor(lxa_state->lxas_dev_new);
+
+	/* disable any mmap output that might still be going on */
+	lxa_mmap_output_disable(lxa_state);
+
+	/*
+	 * if this was the active input/output device, unlink it from
+	 * the global zone state so that other opens of the audio device
+	 * can now succeed.
+	 */
+	mutex_enter(&lxa_lock);
+	if (lxa_zs->lxa_zs_istate == lxa_state)
+		lxa_zs->lxa_zs_istate = NULL;
+	if (lxa_zs->lxa_zs_ostate == lxa_state) {
+		lxa_zs->lxa_zs_ostate = NULL;
+	}
+	mutex_exit(&lxa_lock);
+
+	/* remove this state structure from the hash (if it's there) */
+	(void) mod_hash_remove(lxa_state_hash,
+	    (mod_hash_key_t)(uintptr_t)minor, (mod_hash_val_t *)&lxa_state);
+
+	/* close any audio device that we have open */
+	if (lxa_state->lxas_idev_lh != NULL)
+		(void) ldi_close(lxa_state->lxas_idev_lh,
+		    lxa_state->lxas_idev_flags, kcred);
+	if (lxa_state->lxas_odev_lh != NULL)
+		(void) ldi_close(lxa_state->lxas_odev_lh,
+		    lxa_state->lxas_odev_flags, kcred);
+
+	/* free up any memory allocated by mmaps */
+	if (lxa_state->lxas_umem_cookie != NULL)
+		ddi_umem_free(lxa_state->lxas_umem_cookie);
+
+	/* release the id associated with this state structure */
+	id_free(lxa_minor_id, minor);
+
+	kmem_free(lxa_state, sizeof (*lxa_state));
+}
+
+static char *
+getzonename(void)
+{
+	return (curproc->p_zone->zone_name);
+}
+
+static char *
+lxa_devprop_name(char *zname, char *pname)
+{
+	char	*zpname;
+	int	n;
+
+	ASSERT((pname != NULL) && (zname != NULL));
+
+	/* prepend the zone name to the property name */
+	n = snprintf(NULL, 0, "%s_%s", zname, pname) + 1;
+	zpname = kmem_alloc(n, KM_SLEEP);
+	(void) snprintf(zpname, n, "%s_%s", zname, pname);
+
+	return (zpname);
+}
+
+static int
+lxa_devprop_verify(char *pval)
+{
+	int	n;
+
+	ASSERT(pval != NULL);
+
+	if (strcmp(pval, "default") == 0)
+		return (0);
+
+	/* make sure the value is an integer */
+	for (n = 0; pval[n] != '\0'; n++) {
+		if ((pval[n] < '0') && (pval[n] > '9')) {
+			return (-1);
+		}
+	}
+
+	return (0);
+}
+
+static char *
+lxa_devprop_lookup(char *zname, char *pname, lxa_dev_type_t lxa_type)
+{
+	char		*zprop_name, *pval;
+	char		*dev_path;
+	int		n, rv;
+
+	ASSERT((pname != NULL) && (zname != NULL));
+	ASSERT((lxa_type == LXA_TYPE_AUDIO) || (lxa_type == LXA_TYPE_AUDIOCTL));
+
+	zprop_name = lxa_devprop_name(zname, pname);
+
+	/* attempt to lookup the property */
+	rv = ddi_prop_lookup_string(DDI_DEV_T_ANY, lxa_dip,
+	    DDI_PROP_DONTPASS | DDI_PROP_NOTPROM, zprop_name, &pval);
+	strfree(zprop_name);
+
+	if (rv != DDI_PROP_SUCCESS)
+		return (NULL);
+
+	if (lxa_devprop_verify(pval) != 0) {
+		ddi_prop_free(pval);
+		return (NULL);
+	}
+
+	if (strcmp(pval, "none") == 0) {
+		/* there is no audio device specified */
+		return (NULL);
+	} else if (strcmp(pval, "default") == 0) {
+		/* use the default audio device on the system */
+		dev_path = strdup(LXA_DEV_DEFAULT);
+	} else {
+		/* a custom audio device was specified, generate a path */
+		n = snprintf(NULL, 0, "%s%s", LXA_DEV_CUSTOM_DIR, pval) + 1;
+		dev_path = kmem_alloc(n, KM_SLEEP);
+		(void) snprintf(dev_path, n, "%s%s", LXA_DEV_CUSTOM_DIR, pval);
+	}
+	ddi_prop_free(pval);
+
+	/*
+	 * if this is an audio control device so we need to append
+	 * "ctl" to the path
+	 */
+	if (lxa_type == LXA_TYPE_AUDIOCTL) {
+		char	*tmp;
+		n = snprintf(NULL, 0, "%s%s", dev_path, "ctl") + 1;
+		tmp = kmem_alloc(n, KM_SLEEP);
+		(void) snprintf(tmp, n, "%s%s", dev_path, "ctl");
+		strfree(dev_path);
+		dev_path = tmp;
+	}
+
+	return (dev_path);
+}
+
+static int
+lxa_dev_getfeatures(lxa_state_t *lxa_state)
+{
+	audio_info_t	ai_idev, ai_odev;
+	int		n, rv;
+
+	/* set a default fragment size */
+	lxa_state->lxas_frag_size = LXA_OSS_FRAG_SIZE;
+	lxa_state->lxas_frag_cnt = LXA_OSS_FRAG_CNT;
+
+	/* get info for the currently open audio devices */
+	if ((lxa_state->lxas_idev_lh != NULL) &&
+	    ((rv = ldi_ioctl(lxa_state->lxas_idev_lh,
+	    AUDIO_GETINFO, (intptr_t)&ai_idev, FKIOCTL, kcred, &n)) != 0))
+		return (rv);
+	if ((lxa_state->lxas_odev_lh != NULL) &&
+	    ((rv = ldi_ioctl(lxa_state->lxas_odev_lh,
+	    AUDIO_GETINFO, (intptr_t)&ai_odev, FKIOCTL, kcred, &n)) != 0))
+		return (rv);
+
+	/* if we're only open for reading or writing then it's easy */
+	if (lxa_state->lxas_idev_lh == NULL) {
+		lxa_state->lxas_sw_features = ai_odev.sw_features;
+		lxa_state->lxas_hw_features = ai_odev.hw_features;
+		return (0);
+	} else if (lxa_state->lxas_odev_lh == NULL) {
+		lxa_state->lxas_sw_features = ai_idev.sw_features;
+		lxa_state->lxas_hw_features = ai_idev.hw_features;
+		return (0);
+	}
+
+	/*
+	 * well if we're open for reading and writing but the underlying
+	 * device is the same then it's also pretty easy
+	 */
+	if (lxa_state->lxas_devs_same) {
+		if ((ai_odev.sw_features != ai_idev.sw_features) ||
+		    (ai_odev.hw_features != ai_idev.hw_features)) {
+			zcmn_err(getzoneid(), CE_WARN, "lx_audio error: "
+			    "audio device reported inconsistent features");
+			return (EIO);
+		}
+		lxa_state->lxas_sw_features = ai_odev.sw_features;
+		lxa_state->lxas_hw_features = ai_odev.hw_features;
+		return (0);
+	}
+
+	/*
+	 * figure out which software features we're going to support.
+	 * we will report a feature as supported if both the input
+	 * and output device support it.
+	 */
+	lxa_state->lxas_sw_features = 0;
+	n = ai_idev.sw_features & ai_odev.sw_features;
+	if (n & AUDIO_SWFEATURE_MIXER)
+		lxa_state->lxas_sw_features |= AUDIO_SWFEATURE_MIXER;
+
+	/*
+	 * figure out which hardware features we're going to support.
+	 * for a first pass we will report a feature as supported if
+	 * both the input and output device support it.
+	 */
+	lxa_state->lxas_hw_features = 0;
+	n = ai_idev.hw_features & ai_odev.hw_features;
+	if (n & AUDIO_HWFEATURE_MSCODEC)
+		lxa_state->lxas_hw_features |= AUDIO_HWFEATURE_MSCODEC;
+
+	/*
+	 * if we made it here then we have different audio input and output
+	 * devices.  this will allow us to report support for additional
+	 * hardware features that may not supported by just the input or
+	 * output device alone.
+	 */
+
+	/* always report tha we support both playback and recording */
+	lxa_state->lxas_hw_features =
+	    AUDIO_HWFEATURE_PLAY | AUDIO_HWFEATURE_RECORD;
+
+	/* always report full duplex support */
+	lxa_state->lxas_hw_features = AUDIO_HWFEATURE_DUPLEX;
+
+	/* never report that we have input to output loopback support */
+	ASSERT((lxa_state->lxas_hw_features & AUDIO_HWFEATURE_IN2OUT) == 0);
+	return (0);
+}
+
+static int
+lxa_dev_open(lxa_state_t *lxa_state)
+{
+	char		*idev, *odev;
+	int		flags, rv;
+	ldi_handle_t	lh;
+	ldi_ident_t	li = NULL;
+
+	ASSERT((lxa_state->lxas_type == LXA_TYPE_AUDIO) ||
+	    (lxa_state->lxas_type == LXA_TYPE_AUDIOCTL));
+
+	/*
+	 * check if we have configuration properties for this zone.
+	 * if we don't then audio isn't supported in this zone.
+	 */
+	idev = lxa_devprop_lookup(getzonename(), LXA_PROP_INPUTDEV,
+	    lxa_state->lxas_type);
+	odev = lxa_devprop_lookup(getzonename(), LXA_PROP_OUTPUTDEV,
+	    lxa_state->lxas_type);
+
+	/* make sure there is at least one device to read from or write to */
+	if ((idev == NULL) && (odev == NULL))
+		return (ENODEV);
+
+	/* see if the input and output devices are actually the same device */
+	if (((idev != NULL) && (odev != NULL)) &&
+	    (strcmp(idev, odev) == 0))
+		lxa_state->lxas_devs_same = 1;
+
+	/* we don't respect FEXCL */
+	flags = lxa_state->lxas_flags & ~FEXCL;
+	if (lxa_state->lxas_type == LXA_TYPE_AUDIO) {
+		/*
+		 * if we're opening audio devices then we need to muck
+		 * with the FREAD/FWRITE flags.
+		 *
+		 * certain audio device may only support input or output
+		 * (but not both.)  so if we're multiplexing input/output
+		 * to different devices we need to make sure we don't try
+		 * and open the output device for reading and the input
+		 * device for writing.
+		 *
+		 * if we're using the same device for input/output we still
+		 * need to do this because some audio devices won't let
+		 * themselves be opened multiple times for read access.
+		 */
+		lxa_state->lxas_idev_flags = flags & ~FWRITE;
+		lxa_state->lxas_odev_flags = flags & ~FREAD;
+
+		/* make sure we have devices to read from and write to */
+		if (((flags & FREAD) && (idev == NULL)) ||
+		    ((flags & FWRITE) && (odev == NULL))) {
+			rv = ENODEV;
+			goto out;
+		}
+	} else {
+		lxa_state->lxas_idev_flags = lxa_state->lxas_odev_flags = flags;
+	}
+
+	/* get an ident to open the devices */
+	if (ldi_ident_from_dev(lxa_state->lxas_dev_new, &li) != 0) {
+		rv = ENODEV;
+		goto out;
+	}
+
+	/* open the input device */
+	lxa_state->lxas_idev_lh = NULL;
+	if (((lxa_state->lxas_type == LXA_TYPE_AUDIOCTL) ||
+	    (lxa_state->lxas_idev_flags & FREAD)) &&
+	    (idev != NULL)) {
+		rv = ldi_open_by_name(idev, lxa_state->lxas_idev_flags,
+		    kcred, &lh, li);
+		if (rv != 0) {
+			zcmn_err(getzoneid(), CE_WARN, "lxa_open_dev: "
+			    "unable to open audio device: %s", idev);
+			zcmn_err(getzoneid(), CE_WARN, "lxa_open_dev: "
+			    "possible zone audio configuration error");
+			goto out;
+		}
+		lxa_state->lxas_idev_lh = lh;
+	}
+
+	/* open the output device */
+	lxa_state->lxas_odev_lh = NULL;
+	if (((lxa_state->lxas_type == LXA_TYPE_AUDIOCTL) ||
+	    (lxa_state->lxas_odev_flags & FWRITE)) &&
+	    (odev != NULL)) {
+		rv = ldi_open_by_name(odev, lxa_state->lxas_odev_flags,
+		    kcred, &lh, li);
+		if (rv != 0) {
+			/*
+			 * If this open failed and we previously opened an
+			 * input device, it is the responsibility of the
+			 * caller to close that device after we return
+			 * failure here.
+			 */
+			zcmn_err(getzoneid(), CE_WARN, "lxa_open_dev: "
+			    "unable to open audio device: %s", odev);
+			zcmn_err(getzoneid(), CE_WARN, "lxa_open_dev: "
+			    "possible zone audio configuration error");
+			goto out;
+		}
+		lxa_state->lxas_odev_lh = lh;
+	}
+
+	/* free up stuff */
+out:
+	if (li != NULL)
+		ldi_ident_release(li);
+	if (idev != NULL)
+		strfree(idev);
+	if (odev != NULL)
+		strfree(odev);
+
+	return (rv);
+}
+
+void
+lxa_mmap_thread_exit(lxa_state_t *lxa_state)
+{
+	mutex_enter(&lxa_lock);
+	lxa_state->lxas_mmap_thread = NULL;
+	lxa_state->lxas_mmap_thread_frag = 0;
+	lxa_state->lxas_mmap_thread_running = 0;
+	lxa_state->lxas_mmap_thread_exit = 0;
+	mutex_exit(&lxa_lock);
+	thread_exit();
+	/*NOTREACHED*/
+}
+
+void
+lxa_mmap_thread(lxa_state_t *lxa_state)
+{
+	struct uio	uio, uio_null;
+	iovec_t		iovec, iovec_null;
+	uint_t		bytes_per_sec, usec_per_frag, ticks_per_frag;
+	int		rv, junk, eof, retry;
+	audio_info_t	ai;
+
+	/* we better be setup for writing to the output device */
+	ASSERT((lxa_state->lxas_flags & FWRITE) != 0);
+	ASSERT(lxa_state->lxas_odev_lh != NULL);
+
+	/* setup a uio to output one fragment */
+	uio.uio_iov = &iovec;
+	uio.uio_iovcnt = 1;
+	uio.uio_offset = 0;
+	uio.uio_segflg = UIO_SYSSPACE;
+	uio.uio_fmode = 0;
+	uio.uio_extflg = 0;
+	uio.uio_llimit = MAXOFFSET_T;
+
+	/* setup a uio to output a eof (a fragment with a length of 0) */
+	uio_null.uio_iov = &iovec_null;
+	uio_null.uio_iov->iov_len = 0;
+	uio_null.uio_iov->iov_base = NULL;
+	uio_null.uio_iovcnt = 1;
+	uio_null.uio_offset = 0;
+	uio_null.uio_segflg = UIO_SYSSPACE;
+	uio_null.uio_fmode = 0;
+	uio_null.uio_extflg = 0;
+	uio_null.uio_llimit = MAXOFFSET_T;
+	uio_null.uio_resid = 0;
+
+lxa_mmap_thread_top:
+	ASSERT(!MUTEX_HELD(&lxa_lock));
+
+	/* first drain any pending audio output */
+	if ((rv = ldi_ioctl(lxa_state->lxas_odev_lh,
+	    AUDIO_DRAIN, NULL, FKIOCTL, kcred, &junk)) != 0) {
+		cmn_err(CE_WARN, "lxa_mmap_thread: "
+		    "AUDIO_DRAIN failed, aborting audio output");
+		lxa_mmap_thread_exit(lxa_state);
+		/*NOTREACHED*/
+	}
+
+	/*
+	 * we depend on the ai.play.eof value to keep track of
+	 * audio output progress so reset it here.
+	 */
+	AUDIO_INITINFO(&ai);
+	ai.play.eof = 0;
+	if ((rv = ldi_ioctl(lxa_state->lxas_odev_lh,
+	    AUDIO_SETINFO, (intptr_t)&ai, FKIOCTL, kcred, &junk)) != 0) {
+		cmn_err(CE_WARN, "lxa_mmap_thread: "
+		    "AUDIO_SETINFO failed, aborting audio output");
+		lxa_mmap_thread_exit(lxa_state);
+		/*NOTREACHED*/
+	}
+
+	/*
+	 * we're going to need to know the sampling rate and number
+	 * of output channels to estimate how long we can sleep between
+	 * requests.
+	 */
+	if ((rv = ldi_ioctl(lxa_state->lxas_odev_lh, AUDIO_GETINFO,
+	    (intptr_t)&ai, FKIOCTL, kcred, &junk)) != 0) {
+		cmn_err(CE_WARN, "lxa_mmap_thread: "
+		    "AUDIO_GETINFO failed, aborting audio output");
+		lxa_mmap_thread_exit(lxa_state);
+		/*NOTREACHED*/
+	}
+
+	/* estimate how many ticks it takes to output a fragment of data */
+	bytes_per_sec = (ai.play.sample_rate * ai.play.channels *
+	    ai.play.precision) / 8;
+	usec_per_frag = MICROSEC * lxa_state->lxas_frag_size / bytes_per_sec;
+	ticks_per_frag = drv_usectohz(usec_per_frag);
+
+	/* queue up three fragments of of data into the output stream */
+	eof = 3;
+
+	/* sanity check the eof value */
+	ASSERT(ai.play.eof == 0);
+	ai.play.eof = 0;
+
+	/* we always start audio output at fragment 0 */
+	mutex_enter(&lxa_lock);
+	lxa_state->lxas_mmap_thread_frag = 0;
+
+	/*
+	 * we shouldn't have allowed the mapping if it isn't a multiple
+	 * of the fragment size
+	 */
+	ASSERT((lxa_state->lxas_umem_len % lxa_state->lxas_frag_size) == 0);
+
+	while (!lxa_state->lxas_mmap_thread_exit) {
+		size_t start, end;
+
+		/*
+		 * calculate the start and ending offsets of the next
+		 * fragment to output
+		 */
+		start = lxa_state->lxas_mmap_thread_frag *
+		    lxa_state->lxas_frag_size;
+		end = start + lxa_state->lxas_frag_size;
+
+		ASSERT(start < lxa_state->lxas_umem_len);
+		ASSERT(end <= lxa_state->lxas_umem_len);
+
+		/* setup the uio to output one fragment of audio */
+		uio.uio_resid = end - start;
+		uio.uio_iov->iov_len = end - start;
+		uio.uio_iov->iov_base = &lxa_state->lxas_umem_ptr[start];
+
+		/* increment the current fragment index */
+		lxa_state->lxas_mmap_thread_frag =
+		    (lxa_state->lxas_mmap_thread_frag + 1) %
+		    (lxa_state->lxas_umem_len / lxa_state->lxas_frag_size);
+
+		/* drop the audio lock before actually outputting data */
+		mutex_exit(&lxa_lock);
+
+		/*
+		 * write the fragment of audio data to the device stream
+		 * then write a eof to the stream to tell the device to
+		 * increment ai.play.eof when it's done processing the
+		 * fragment we just wrote
+		 */
+		if ((rv = ldi_write(lxa_state->lxas_odev_lh,
+		    &uio, kcred)) != 0) {
+			cmn_err(CE_WARN, "lxa_mmap_thread: "
+			    "ldi_write() failed (%d), "
+			    "resetting audio output", rv);
+			goto lxa_mmap_thread_top;
+		}
+		if ((rv = ldi_write(lxa_state->lxas_odev_lh,
+		    &uio_null, kcred)) != 0) {
+			cmn_err(CE_WARN, "lxa_mmap_thread: "
+			    "ldi_write(eof) failed (%d), "
+			    "resetting audio output", rv);
+			goto lxa_mmap_thread_top;
+		}
+
+		/*
+		 * we want to avoid buffer underrun so ensure that
+		 * there is always at least one fragment of data in the
+		 * output stream.
+		 */
+		mutex_enter(&lxa_lock);
+		if (--eof > 0) {
+			continue;
+		}
+
+		/*
+		 * now we wait until the audio device has finished outputting
+		 * at least one fragment of data.
+		 */
+		retry = 0;
+		while (!lxa_state->lxas_mmap_thread_exit && (eof == 0)) {
+			uint_t ai_eof_old = ai.play.eof;
+
+			mutex_exit(&lxa_lock);
+
+			/*
+			 * delay for the number of ticks it takes
+			 * to output one fragment of data
+			 */
+			if (ticks_per_frag > 0)
+				delay(ticks_per_frag);
+
+			/* check if we've managed to output any fragments */
+			if ((rv = ldi_ioctl(lxa_state->lxas_odev_lh,
+			    AUDIO_GETINFO, (intptr_t)&ai,
+			    FKIOCTL, kcred, &junk)) != 0) {
+				cmn_err(CE_WARN, "lxa_mmap_thread: "
+				    "AUDIO_GETINFO failed (%d), "
+				    "resetting audio output", rv);
+				/* re-start mmap audio output */
+				goto lxa_mmap_thread_top;
+			}
+
+			if (ai_eof_old == ai.play.eof) {
+				/* institute a random retry limit */
+				if (retry++ < 100) {
+					mutex_enter(&lxa_lock);
+					continue;
+				}
+				cmn_err(CE_WARN, "lxa_mmap_thread: "
+				    "output stalled, "
+				    "resetting audio output");
+				/* re-start mmap audio output */
+				goto lxa_mmap_thread_top;
+			}
+
+			if (ai.play.eof > ai_eof_old) {
+				eof = ai.play.eof - ai_eof_old;
+			} else {
+				/* eof counter wrapped around */
+				ASSERT(ai_eof_old < ai.play.eof);
+				eof = ai.play.eof + (ai_eof_old - UINTMAX_MAX);
+			}
+			/* we're done with this loop so re-aquire the lock */
+			ASSERT(eof != 0);
+			mutex_enter(&lxa_lock);
+		}
+	}
+	mutex_exit(&lxa_lock);
+	lxa_mmap_thread_exit(lxa_state);
+	/*NOTREACHED*/
+}
+
+static void
+lxa_mmap_output_disable(lxa_state_t *lxa_state)
+{
+	kt_did_t tid;
+
+	mutex_enter(&lxa_lock);
+
+	/* if the output thread isn't running there's nothing to do */
+	if (lxa_state->lxas_mmap_thread_running == 0) {
+		mutex_exit(&lxa_lock);
+		return;
+	}
+
+	/* tell the pcm mmap output thread to exit */
+	lxa_state->lxas_mmap_thread_exit = 1;
+
+	/* wait for the mmap output thread to exit */
+	tid = lxa_state->lxas_mmap_thread->t_did;
+	mutex_exit(&lxa_lock);
+	thread_join(tid);
+}
+
+static void
+lxa_mmap_output_enable(lxa_state_t *lxa_state)
+{
+	mutex_enter(&lxa_lock);
+
+	/* if the output thread is already running there's nothing to do */
+	if (lxa_state->lxas_mmap_thread_running != 0) {
+		mutex_exit(&lxa_lock);
+		return;
+	}
+
+	/* setup output state */
+	lxa_state->lxas_mmap_thread_running = 1;
+	lxa_state->lxas_mmap_thread_exit = 0;
+	lxa_state->lxas_mmap_thread_frag = 0;
+
+	/* kick off a thread to do the mmap pcm output */
+	lxa_state->lxas_mmap_thread = thread_create(NULL, 0,
+	    (void (*)())lxa_mmap_thread, lxa_state,
+	    0, &p0, TS_RUN, minclsyspri);
+	ASSERT(lxa_state->lxas_mmap_thread != NULL);
+
+	mutex_exit(&lxa_lock);
+}
+
+static int
+lxa_ioc_mmap_output(lxa_state_t *lxa_state, intptr_t arg, int mode)
+{
+	uint_t	trigger;
+
+	/* we only support output via mmap */
+	if ((lxa_state->lxas_flags & FWRITE) == 0)
+		return (EINVAL);
+
+	/* if the user hasn't mmap the device then there's nothing to do */
+	if (lxa_state->lxas_umem_cookie == NULL)
+		return (EINVAL);
+
+	/* copy in the request */
+	if (ddi_copyin((void *)arg, &trigger, sizeof (trigger), mode) != 0)
+		return (EFAULT);
+
+	/* a zero value disables output */
+	if (trigger == 0) {
+		lxa_mmap_output_disable(lxa_state);
+		return (0);
+	}
+
+	/* a non-zero value enables output */
+	lxa_mmap_output_enable(lxa_state);
+	return (0);
+}
+
+static int
+lxa_ioc_mmap_ptr(lxa_state_t *lxa_state, intptr_t arg, int mode)
+{
+	int	ptr;
+
+	/* we only support output via mmap */
+	if ((lxa_state->lxas_flags & FWRITE) == 0)
+		return (EINVAL);
+
+	/* if the user hasn't mmap the device then there's nothing to do */
+	if (lxa_state->lxas_umem_cookie == NULL)
+		return (EINVAL);
+
+	/* if the output thread isn't running then there's nothing to do */
+	if (lxa_state->lxas_mmap_thread_running == 0)
+		return (EINVAL);
+
+	mutex_enter(&lxa_lock);
+	ptr = lxa_state->lxas_mmap_thread_frag * lxa_state->lxas_frag_size;
+	mutex_exit(&lxa_lock);
+
+	if (ddi_copyout(&ptr, (void *)arg, sizeof (ptr), mode) != 0)
+		return (EFAULT);
+
+	return (0);
+}
+
+static int
+lxa_ioc_get_frag_info(lxa_state_t *lxa_state, intptr_t arg, int mode)
+{
+	lxa_frag_info_t	fi;
+
+	fi.lxa_fi_size = lxa_state->lxas_frag_size;
+	fi.lxa_fi_cnt = lxa_state->lxas_frag_cnt;
+
+	if (ddi_copyout(&fi, (void *)arg, sizeof (fi), mode) != 0)
+		return (EFAULT);
+
+	return (0);
+}
+
+static int
+lxa_ioc_set_frag_info(lxa_state_t *lxa_state, intptr_t arg, int mode)
+{
+	lxa_frag_info_t	fi;
+
+	/* if the device is mmaped we can't change the fragment settings */
+	if (lxa_state->lxas_umem_cookie != NULL)
+		return (EINVAL);
+
+	/* copy in the request */
+	if (ddi_copyin((void *)arg, &fi, sizeof (fi), mode) != 0)
+		return (EFAULT);
+
+	/* do basic bounds checking */
+	if ((fi.lxa_fi_cnt == 0) || (fi.lxa_fi_size < 16))
+		return (EINVAL);
+
+	/* don't accept size values less than 16 */
+
+	lxa_state->lxas_frag_size = fi.lxa_fi_size;
+	lxa_state->lxas_frag_cnt = fi.lxa_fi_cnt;
+
+	return (0);
+}
+
+static int
+lxa_audio_drain(lxa_state_t *lxa_state)
+{
+	int	junk;
+
+	/* only applies to output buffers */
+	if (lxa_state->lxas_odev_lh == NULL)
+		return (EINVAL);
+
+	/* can't fail so ignore the return value */
+	(void) ldi_ioctl(lxa_state->lxas_odev_lh, AUDIO_DRAIN, NULL,
+	    FKIOCTL, kcred, &junk);
+	return (0);
+}
+
+/*
+ * lxa_audio_info_merge() usage notes:
+ *
+ * - it's important to make sure NOT to get the ai_idev and ai_odev
+ *   parameters mixed up when calling lxa_audio_info_merge().
+ *
+ * - it's important for the caller to make sure that AUDIO_GETINFO
+ *   was called for the input device BEFORE the output device.  (see
+ *   the comments for merging the monitor_gain setting to see why.)
+ */
+static void
+lxa_audio_info_merge(lxa_state_t *lxa_state,
+    audio_info_t *ai_idev, audio_info_t *ai_odev, audio_info_t *ai_merged)
+{
+	/* if we're not setup for output return the intput device info */
+	if (lxa_state->lxas_odev_lh == NULL) {
+		*ai_merged = *ai_idev;
+		return;
+	}
+
+	/* if we're not setup for input return the output device info */
+	if (lxa_state->lxas_idev_lh == NULL) {
+		*ai_merged = *ai_odev;
+		return;
+	}
+
+	/* get record values from the input device */
+	ai_merged->record = ai_idev->record;
+
+	/* get play values from the output device */
+	ai_merged->play = ai_odev->play;
+
+	/* muting status only matters for the output device */
+	ai_merged->output_muted = ai_odev->output_muted;
+
+	/* we don't support device reference counts, always return 1 */
+	ai_merged->ref_cnt = 1;
+
+	/*
+	 * for supported hw/sw features report the combined feature
+	 * set we calcuated out earlier.
+	 */
+	ai_merged->hw_features = lxa_state->lxas_hw_features;
+	ai_merged->sw_features = lxa_state->lxas_sw_features;
+
+	if (!lxa_state->lxas_devs_same) {
+		/*
+		 * if the input and output devices are different
+		 * physical devices then we don't support input to
+		 * output loopback so we always report the input
+		 * to output loopback gain to be zero.
+		 */
+		ai_merged->monitor_gain = 0;
+	} else {
+		/*
+		 * the intput and output devices are actually the
+		 * same physical device.  hence it probably supports
+		 * intput to output loopback.  regardless we should
+		 * pass back the intput to output gain reported by
+		 * the device.  when we pick a value to passback we
+		 * use the output device value since that was
+		 * the most recently queried.  (we base this
+		 * decision on the assumption that io gain is
+		 * actually hardware setting in the device and
+		 * hence if it is changed on one open instance of
+		 * the device the change will be visable to all
+		 * other instances of the device.)
+		 */
+		ai_merged->monitor_gain = ai_odev->monitor_gain;
+	}
+
+	/*
+	 * for currently enabled software features always return the
+	 * merger of the two.  (of course the enabled software features
+	 * for the input and output devices should alway be the same,
+	 * so if it isn't complain.)
+	 */
+	if (ai_idev->sw_features_enabled != ai_odev->sw_features_enabled)
+		zcmn_err(getzoneid(), CE_WARN, "lx_audio: "
+		    "unexpected sofware feature state");
+	ai_merged->sw_features_enabled =
+	    ai_idev->sw_features_enabled & ai_odev->sw_features_enabled;
+}
+
+static int
+lxa_audio_setinfo(lxa_state_t *lxa_state, int cmd, intptr_t arg,
+    int mode)
+{
+	audio_info_t	ai, ai_null, ai_idev, ai_odev;
+	int		rv, junk;
+
+	/* copy in the request */
+	if (ddi_copyin((void *)arg, &ai, sizeof (ai), mode) != 0)
+		return (EFAULT);
+
+	/*
+	 * if the caller is attempting to enable a software feature that
+	 * we didn't report as supported the return an error
+	 */
+	if ((ai.sw_features_enabled != -1) &&
+	    (ai.sw_features_enabled & ~lxa_state->lxas_sw_features))
+		return (EINVAL);
+
+	/*
+	 * if a process has mmaped this device then we don't allow
+	 * changes to the play.eof field (since mmap output depends
+	 * on this field.
+	 */
+	if ((lxa_state->lxas_umem_cookie != NULL) &&
+	    (ai.play.eof != -1))
+		return (EIO);
+
+	/* initialize the new requests */
+	AUDIO_INITINFO(&ai_null);
+	ai_idev = ai_odev = ai;
+
+	/* remove audio input settings from the output device request */
+	ai_odev.record = ai_null.record;
+
+	/* remove audio output settings from the input device request */
+	ai_idev.play = ai_null.play;
+	ai_idev.output_muted = ai_null.output_muted;
+
+	/* apply settings to the intput device */
+	if ((lxa_state->lxas_idev_lh != NULL) &&
+	    ((rv = ldi_ioctl(lxa_state->lxas_idev_lh, cmd,
+	    (intptr_t)&ai_idev, FKIOCTL, kcred, &junk)) != 0))
+		return (rv);
+
+	/* apply settings to the output device */
+	if ((lxa_state->lxas_odev_lh != NULL) &&
+	    ((rv = ldi_ioctl(lxa_state->lxas_odev_lh, cmd,
+	    (intptr_t)&ai_odev, FKIOCTL, kcred, &junk)) != 0))
+		return (rv);
+
+	/*
+	 * a AUDIO_SETINFO call performs an implicit AUDIO_GETINFO to
+	 * return values (see the coments in audioio.h.) so we need
+	 * to combine the values returned from the input and output
+	 * device back into the users buffer.
+	 */
+	lxa_audio_info_merge(lxa_state, &ai_idev, &ai_odev, &ai);
+
+	/* copyout the results */
+	if (ddi_copyout(&ai, (void *)arg, sizeof (ai), mode) != 0) {
+		return (EFAULT);
+	}
+
+	return (0);
+}
+
+static int
+lxa_audio_getinfo(lxa_state_t *lxa_state, intptr_t arg, int mode)
+{
+	audio_info_t	ai, ai_idev, ai_odev;
+	int		rv, junk;
+
+	/* get the settings from the input device */
+	if ((lxa_state->lxas_idev_lh != NULL) &&
+	    ((rv = ldi_ioctl(lxa_state->lxas_idev_lh, AUDIO_GETINFO,
+	    (intptr_t)&ai_idev, FKIOCTL, kcred, &junk)) != 0))
+		return (rv);
+
+	/* get the settings from the output device */
+	if ((lxa_state->lxas_odev_lh != NULL) &&
+	    ((rv = ldi_ioctl(lxa_state->lxas_odev_lh, AUDIO_GETINFO,
+	    (intptr_t)&ai_odev, FKIOCTL, kcred, &junk)) != 0))
+		return (rv);
+
+	/*
+	 * we need to combine the values returned from the input
+	 * and output device back into a single user buffer.
+	 */
+	lxa_audio_info_merge(lxa_state, &ai_idev, &ai_odev, &ai);
+
+	/* copyout the results */
+	if (ddi_copyout(&ai, (void *)arg, sizeof (ai), mode) != 0)
+		return (EFAULT);
+
+	return (0);
+}
+
+static int
+lxa_mixer_ai_from_lh(ldi_handle_t lh, audio_info_t *ai)
+{
+	int		rv, junk;
+
+	ASSERT((lh != NULL) && (ai != NULL));
+
+	/* get the device state and channel state */
+	rv = ldi_ioctl(lh, AUDIO_GETINFO, (intptr_t)ai, FKIOCTL, kcred, &junk);
+
+	return (rv);
+}
+
+static int
+lxa_mixer_get_ai(lxa_state_t *lxa_state, audio_info_t *ai)
+{
+	audio_info_t	ai_idev, ai_odev;
+	int		rv;
+
+	/* if there is no input device, query the output device */
+	if (lxa_state->lxas_idev_lh == NULL)
+		return (lxa_mixer_ai_from_lh(lxa_state->lxas_odev_lh, ai));
+
+	/* if there is no ouput device, query the intput device */
+	if (lxa_state->lxas_odev_lh == NULL)
+		return (lxa_mixer_ai_from_lh(lxa_state->lxas_idev_lh, ai));
+
+	/*
+	 * now get the audio_info and channel information for the
+	 * underlying output device.
+	 */
+	if ((rv = lxa_mixer_ai_from_lh(lxa_state->lxas_idev_lh,
+	    &ai_idev)) != 0)
+		return (rv);
+	if ((rv = lxa_mixer_ai_from_lh(lxa_state->lxas_odev_lh,
+	    &ai_odev)) != 0)
+		return (rv);
+
+	/* now merge the audio_info structures */
+	lxa_audio_info_merge(lxa_state, &ai_idev, &ai_odev, ai);
+	return (0);
+}
+
+static int
+lxa_mixer_get_common(lxa_state_t *lxa_state, int cmd, intptr_t arg, int mode)
+{
+	lxa_mixer_levels_t	lxa_ml;
+	audio_info_t		ai;
+	int			rv;
+
+	ASSERT(lxa_state->lxas_type == LXA_TYPE_AUDIOCTL);
+
+	if ((rv = lxa_mixer_get_ai(lxa_state, &ai)) != 0)
+		return (rv);
+
+	switch (cmd) {
+	case LXA_IOC_MIXER_GET_VOL:
+		lxa_ml.lxa_ml_gain = ai.play.gain;
+		lxa_ml.lxa_ml_balance = ai.play.balance;
+		break;
+	case LXA_IOC_MIXER_GET_MIC:
+		lxa_ml.lxa_ml_gain = ai.record.gain;
+		lxa_ml.lxa_ml_balance = ai.record.balance;
+		break;
+	}
+
+	if (ddi_copyout(&lxa_ml, (void *)arg, sizeof (lxa_ml), mode) != 0)
+		return (EFAULT);
+	return (0);
+}
+
+static int
+lxa_mixer_set_common(lxa_state_t *lxa_state, int cmd, intptr_t arg, int mode)
+{
+	lxa_mixer_levels_t	lxa_ml;
+	audio_info_t		ai;
+
+	ASSERT(lxa_state->lxas_type == LXA_TYPE_AUDIOCTL);
+
+	/* get the new mixer settings */
+	if (ddi_copyin((void *)arg, &lxa_ml, sizeof (lxa_ml), mode) != 0)
+		return (EFAULT);
+
+	/* sanity check the mixer settings */
+	if (!LXA_MIXER_LEVELS_OK(&lxa_ml))
+		return (EINVAL);
+
+	/* initialize an audio_info struct with the new settings */
+	AUDIO_INITINFO(&ai);
+	switch (cmd) {
+	case LXA_IOC_MIXER_SET_VOL:
+		ai.play.gain = lxa_ml.lxa_ml_gain;
+		ai.play.balance = lxa_ml.lxa_ml_balance;
+		break;
+	case LXA_IOC_MIXER_SET_MIC:
+		ai.record.gain = lxa_ml.lxa_ml_gain;
+		ai.record.balance = lxa_ml.lxa_ml_balance;
+		break;
+	}
+
+	return (lxa_audio_setinfo(lxa_state, AUDIO_SETINFO, (intptr_t)&ai,
+	    FKIOCTL));
+}
+
+static int
+lxa_mixer_get_pcm(lxa_state_t *lxa_state, intptr_t arg, int mode)
+{
+	ASSERT(lxa_state->lxas_type == LXA_TYPE_AUDIOCTL);
+
+	/* simply return the cached pcm mixer settings */
+	mutex_enter(&lxa_lock);
+	if (ddi_copyout(&lxa_state->lxas_zs->lxa_zs_pcm_levels, (void *)arg,
+	    sizeof (lxa_state->lxas_zs->lxa_zs_pcm_levels), mode) != 0) {
+		mutex_exit(&lxa_lock);
+		return (EFAULT);
+	}
+	mutex_exit(&lxa_lock);
+	return (0);
+}
+
+static int
+lxa_mixer_set_pcm(lxa_state_t *lxa_state, intptr_t arg, int mode)
+{
+	lxa_mixer_levels_t	lxa_ml;
+	int			rv;
+
+	ASSERT(lxa_state->lxas_type == LXA_TYPE_AUDIOCTL);
+
+	/* get the new mixer settings */
+	if (ddi_copyin((void *)arg, &lxa_ml, sizeof (lxa_ml), mode) != 0)
+		return (EFAULT);
+
+	/* sanity check the mixer settings */
+	if (!LXA_MIXER_LEVELS_OK(&lxa_ml))
+		return (EINVAL);
+
+	mutex_enter(&lxa_lock);
+
+	/* if there is an active output channel, update it */
+	if (lxa_state->lxas_zs->lxa_zs_ostate != NULL) {
+		audio_info_t	ai;
+
+		/* initialize an audio_info struct with the new settings */
+		AUDIO_INITINFO(&ai);
+		ai.play.gain = lxa_ml.lxa_ml_gain;
+		ai.play.balance = lxa_ml.lxa_ml_balance;
+
+		if ((rv = lxa_audio_setinfo(lxa_state->lxas_zs->lxa_zs_ostate,
+		    AUDIO_SETINFO, (intptr_t)&ai, FKIOCTL)) != 0) {
+			mutex_exit(&lxa_lock);
+			return (rv);
+		}
+	}
+
+	/* update the cached mixer settings */
+	lxa_state->lxas_zs->lxa_zs_pcm_levels = lxa_ml;
+
+	mutex_exit(&lxa_lock);
+	return (0);
+}
+
+static int
+lxa_zone_reg(intptr_t arg, int mode)
+{
+	lxa_zone_reg_t	lxa_zr;
+	lxa_zstate_t	*lxa_zs = NULL;
+	char		*idev_name = NULL, *odev_name = NULL, *pval = NULL;
+	int		i, junk;
+
+	if (ddi_copyin((void *)arg, &lxa_zr, sizeof (lxa_zr), mode) != 0)
+		return (EFAULT);
+
+	/* make sure that zone_name is a valid string */
+	for (i = 0; i < sizeof (lxa_zr.lxa_zr_zone_name); i++)
+		if (lxa_zr.lxa_zr_zone_name[i] == '\0')
+			break;
+	if (i == sizeof (lxa_zr.lxa_zr_zone_name))
+		return (EINVAL);
+
+	/* make sure that inputdev is a valid string */
+	for (i = 0; i < sizeof (lxa_zr.lxa_zr_inputdev); i++)
+		if (lxa_zr.lxa_zr_inputdev[i] == '\0')
+			break;
+	if (i == sizeof (lxa_zr.lxa_zr_inputdev))
+		return (EINVAL);
+
+	/* make sure it's a valid inputdev property value */
+	if (lxa_devprop_verify(lxa_zr.lxa_zr_inputdev) != 0)
+		return (EINVAL);
+
+	/* make sure that outputdev is a valid string */
+	for (i = 0; i < sizeof (lxa_zr.lxa_zr_outputdev); i++)
+		if (lxa_zr.lxa_zr_outputdev[i] == '\0')
+			break;
+	if (i == sizeof (lxa_zr.lxa_zr_outputdev))
+		return (EINVAL);
+
+	/* make sure it's a valid outputdev property value */
+	if (lxa_devprop_verify(lxa_zr.lxa_zr_outputdev) != 0)
+		return (EINVAL);
+
+	/* get the property names */
+	idev_name = lxa_devprop_name(lxa_zr.lxa_zr_zone_name,
+	    LXA_PROP_INPUTDEV);
+	odev_name = lxa_devprop_name(lxa_zr.lxa_zr_zone_name,
+	    LXA_PROP_OUTPUTDEV);
+
+	/*
+	 * allocate and initialize a zone state structure
+	 * since the audio device can't possibly be opened yet
+	 * (since we're setting it up now and the zone isn't booted
+	 * yet) assign some some resonable default pcm channel settings.
+	 * also, default to one mixer channel.
+	 */
+	lxa_zs = kmem_zalloc(sizeof (*lxa_zs), KM_SLEEP);
+	lxa_zs->lxa_zs_zonename = strdup(lxa_zr.lxa_zr_zone_name);
+	lxa_zs->lxa_zs_pcm_levels.lxa_ml_gain = AUDIO_MID_GAIN;
+	lxa_zs->lxa_zs_pcm_levels.lxa_ml_balance = AUDIO_MID_BALANCE;
+
+	mutex_enter(&lxa_lock);
+
+	/*
+	 * make sure this zone isn't already registered
+	 * a zone is registered with properties for that zone exist
+	 * or there is a zone state structure for that zone
+	 */
+	if (ddi_prop_lookup_string(DDI_DEV_T_ANY, lxa_dip,
+	    DDI_PROP_DONTPASS | DDI_PROP_NOTPROM,
+	    idev_name, &pval) == DDI_PROP_SUCCESS) {
+		goto err_unlock;
+	}
+	if (ddi_prop_lookup_string(DDI_DEV_T_ANY, lxa_dip,
+	    DDI_PROP_DONTPASS | DDI_PROP_NOTPROM,
+	    odev_name, &pval) == DDI_PROP_SUCCESS) {
+		goto err_unlock;
+	}
+	if (mod_hash_find(lxa_zstate_hash,
+	    (mod_hash_key_t)lxa_zs->lxa_zs_zonename,
+	    (mod_hash_val_t *)&junk) == 0)
+		goto err_unlock;
+
+	/*
+	 * create the new properties and insert the zone state structure
+	 * into the global hash
+	 */
+	if (ddi_prop_update_string(DDI_DEV_T_NONE, lxa_dip,
+	    idev_name, lxa_zr.lxa_zr_inputdev) != DDI_PROP_SUCCESS)
+		goto err_prop_remove;
+	if (ddi_prop_update_string(DDI_DEV_T_NONE, lxa_dip,
+	    odev_name, lxa_zr.lxa_zr_outputdev) != DDI_PROP_SUCCESS)
+		goto err_prop_remove;
+	if (mod_hash_insert(lxa_zstate_hash,
+	    (mod_hash_key_t)lxa_zs->lxa_zs_zonename,
+	    (mod_hash_val_t)lxa_zs) != 0)
+		goto err_prop_remove;
+
+	/* success! */
+	lxa_registered_zones++;
+	mutex_exit(&lxa_lock);
+
+	/* cleanup */
+	strfree(idev_name);
+	strfree(odev_name);
+	return (0);
+
+err_prop_remove:
+	(void) ddi_prop_remove(DDI_DEV_T_NONE, lxa_dip, idev_name);
+	(void) ddi_prop_remove(DDI_DEV_T_NONE, lxa_dip, odev_name);
+
+err_unlock:
+	mutex_exit(&lxa_lock);
+
+	if (lxa_zs != NULL) {
+		strfree(lxa_zs->lxa_zs_zonename);
+		kmem_free(lxa_zs, sizeof (*lxa_zs));
+	}
+	if (pval != NULL)
+		ddi_prop_free(pval);
+	if (idev_name != NULL)
+		strfree(idev_name);
+	if (odev_name != NULL)
+		strfree(odev_name);
+	return (EIO);
+}
+
+static int
+lxa_zone_unreg(intptr_t arg, int mode)
+{
+	lxa_zone_reg_t	lxa_zr;
+	lxa_zstate_t	*lxa_zs = NULL;
+	char		*idev_name = NULL, *odev_name = NULL, *pval = NULL;
+	int		rv, i;
+
+	if (ddi_copyin((void *)arg, &lxa_zr, sizeof (lxa_zr), mode) != 0)
+		return (EFAULT);
+
+	/* make sure that zone_name is a valid string */
+	for (i = 0; i < sizeof (lxa_zr.lxa_zr_zone_name); i++)
+		if (lxa_zr.lxa_zr_zone_name[i] == '\0')
+			break;
+	if (i == sizeof (lxa_zr.lxa_zr_zone_name))
+		return (EINVAL);
+
+	/* get the property names */
+	idev_name = lxa_devprop_name(lxa_zr.lxa_zr_zone_name,
+	    LXA_PROP_INPUTDEV);
+	odev_name = lxa_devprop_name(lxa_zr.lxa_zr_zone_name,
+	    LXA_PROP_OUTPUTDEV);
+
+	mutex_enter(&lxa_lock);
+
+	if (lxa_registered_zones <= 0) {
+		rv = ENOENT;
+		goto err_unlock;
+	}
+
+	/* make sure this zone is actually registered */
+	if (ddi_prop_lookup_string(DDI_DEV_T_ANY, lxa_dip,
+	    DDI_PROP_DONTPASS | DDI_PROP_NOTPROM,
+	    idev_name, &pval) != DDI_PROP_SUCCESS) {
+		rv = ENOENT;
+		goto err_unlock;
+	}
+	ddi_prop_free(pval);
+	pval = NULL;
+	if (ddi_prop_lookup_string(DDI_DEV_T_ANY, lxa_dip,
+	    DDI_PROP_DONTPASS | DDI_PROP_NOTPROM,
+	    odev_name, &pval) != DDI_PROP_SUCCESS) {
+		rv = ENOENT;
+		goto err_unlock;
+	}
+	ddi_prop_free(pval);
+	pval = NULL;
+	if (mod_hash_find(lxa_zstate_hash,
+	    (mod_hash_key_t)lxa_zr.lxa_zr_zone_name,
+	    (mod_hash_val_t *)&lxa_zs) != 0) {
+		rv = ENOENT;
+		goto err_unlock;
+	}
+	ASSERT(strcmp(lxa_zr.lxa_zr_zone_name, lxa_zs->lxa_zs_zonename) == 0);
+
+	/*
+	 * if the audio device is currently in use then refuse to
+	 * unregister the zone
+	 */
+	if ((lxa_zs->lxa_zs_ostate != NULL) ||
+	    (lxa_zs->lxa_zs_ostate != NULL)) {
+		rv = EBUSY;
+		goto err_unlock;
+	}
+
+	/* success! cleanup zone config state */
+	(void) ddi_prop_remove(DDI_DEV_T_NONE, lxa_dip, idev_name);
+	(void) ddi_prop_remove(DDI_DEV_T_NONE, lxa_dip, odev_name);
+
+	/*
+	 * note, the action of removing the zone state structure from the
+	 * hash will automatically free lxa_zs->lxa_zs_zonename.
+	 *
+	 * the reason for this is that we used lxa_zs->lxa_zs_zonename
+	 * as the hash key and by default mod_hash_create_strhash() uses
+	 * mod_hash_strkey_dtor() as a the hash key destructor.  (which
+	 * free's the key for us.
+	 */
+	(void) mod_hash_remove(lxa_zstate_hash,
+	    (mod_hash_key_t)lxa_zr.lxa_zr_zone_name,
+	    (mod_hash_val_t *)&lxa_zs);
+	lxa_registered_zones--;
+	mutex_exit(&lxa_lock);
+
+	/* cleanup */
+	kmem_free(lxa_zs, sizeof (*lxa_zs));
+	strfree(idev_name);
+	strfree(odev_name);
+	return (0);
+
+err_unlock:
+	mutex_exit(&lxa_lock);
+
+	if (pval != NULL)
+		ddi_prop_free(pval);
+	if (idev_name != NULL)
+		strfree(idev_name);
+	if (odev_name != NULL)
+		strfree(odev_name);
+	return (rv);
+}
+
+static int
+lxa_ioctl_devctl(int cmd, intptr_t arg, int mode)
+{
+	/* devctl ioctls are only allowed from the global zone */
+	ASSERT(getzoneid() == 0);
+	if (getzoneid() != 0)
+		return (EINVAL);
+
+	switch (cmd) {
+	case LXA_IOC_ZONE_REG:
+		return (lxa_zone_reg(arg, mode));
+	case LXA_IOC_ZONE_UNREG:
+		return (lxa_zone_unreg(arg, mode));
+	}
+
+	return (EINVAL);
+}
+
+static int
+/*ARGSUSED*/
+lxa_open(dev_t *devp, int flags, int otyp, cred_t *credp)
+{
+	lxa_dev_type_t	open_type = LXA_TYPE_INVALID;
+	lxa_zstate_t	*lxa_zs;
+	lxa_state_t	*lxa_state;
+	minor_t		minor;
+	int		rv;
+
+	if (getminor(*devp) == LXA_MINORNUM_DEVCTL) {
+		/*
+		 * this is a devctl node, it exists to administer this
+		 * pseudo driver so it doesn't actually need access to
+		 * any underlying audio devices.  hence there is nothing
+		 * really to do here.  course, this driver should
+		 * only be administered from the global zone.
+		 */
+		ASSERT(getzoneid() == 0);
+		if (getzoneid() != 0)
+			return (EINVAL);
+		return (0);
+	}
+
+	/* lookup the zone state structure */
+	if (mod_hash_find(lxa_zstate_hash, (mod_hash_key_t)getzonename(),
+	    (mod_hash_val_t *)&lxa_zs) != 0) {
+		return (EIO);
+	}
+
+	/* determine what type of device was opened */
+	switch (getminor(*devp)) {
+	case LXA_MINORNUM_DSP:
+		open_type = LXA_TYPE_AUDIO;
+		break;
+	case LXA_MINORNUM_MIXER:
+		open_type = LXA_TYPE_AUDIOCTL;
+		break;
+	default:
+		return (EINVAL);
+	}
+	ASSERT(open_type != LXA_TYPE_INVALID);
+
+	/* all other opens are clone opens so get a new minor node */
+	minor = id_alloc(lxa_minor_id);
+
+	/* allocate and initialize the new lxa_state structure */
+	lxa_state = kmem_zalloc(sizeof (*lxa_state), KM_SLEEP);
+	lxa_state->lxas_zs = lxa_zs;
+	lxa_state->lxas_dev_old = *devp;
+	lxa_state->lxas_dev_new = makedevice(getmajor(*devp), minor);
+	lxa_state->lxas_flags = flags;
+	lxa_state->lxas_type = open_type;
+
+	/* initialize the input and output device */
+	if (((rv = lxa_dev_open(lxa_state)) != 0) ||
+	    ((rv = lxa_dev_getfeatures(lxa_state)) != 0)) {
+		lxa_state_close(lxa_state);
+		return (rv);
+	}
+
+	/*
+	 * save this audio statue structure into a hash indexed
+	 * by it's minor device number.  (this will provide a convient
+	 * way to lookup the state structure on future operations.)
+	 */
+	if (mod_hash_insert(lxa_state_hash, (mod_hash_key_t)(uintptr_t)minor,
+	    (mod_hash_val_t)lxa_state) != 0) {
+		lxa_state_close(lxa_state);
+		return (EIO);
+	}
+
+	mutex_enter(&lxa_lock);
+
+	/* apply the currently cached zone PCM mixer levels */
+	if ((lxa_state->lxas_type == LXA_TYPE_AUDIO) &&
+	    (lxa_state->lxas_odev_lh != NULL)) {
+		audio_info_t ai;
+
+		AUDIO_INITINFO(&ai);
+		ai.play.gain = lxa_zs->lxa_zs_pcm_levels.lxa_ml_gain;
+		ai.play.balance = lxa_zs->lxa_zs_pcm_levels.lxa_ml_balance;
+
+		if ((rv = lxa_audio_setinfo(lxa_state,
+		    AUDIO_SETINFO, (intptr_t)&ai, FKIOCTL)) != 0) {
+			mutex_exit(&lxa_lock);
+			lxa_state_close(lxa_state);
+			return (rv);
+		}
+	}
+
+	/*
+	 * we only allow one active open of the input or output device.
+	 * check here for duplicate opens
+	 */
+	if (lxa_state->lxas_type == LXA_TYPE_AUDIO) {
+		if ((lxa_state->lxas_idev_lh != NULL) &&
+		    (lxa_zs->lxa_zs_istate != NULL)) {
+			mutex_exit(&lxa_lock);
+			lxa_state_close(lxa_state);
+			return (EBUSY);
+		}
+		if ((lxa_state->lxas_odev_lh != NULL) &&
+		    (lxa_zs->lxa_zs_ostate != NULL)) {
+			mutex_exit(&lxa_lock);
+			lxa_state_close(lxa_state);
+			return (EBUSY);
+		}
+
+		/* not a duplicate open, update the global zone state */
+		if (lxa_state->lxas_idev_lh != NULL)
+			lxa_zs->lxa_zs_istate = lxa_state;
+		if (lxa_state->lxas_odev_lh != NULL)
+			lxa_zs->lxa_zs_ostate = lxa_state;
+	}
+	mutex_exit(&lxa_lock);
+
+	/* make sure to return our newly allocated dev_t */
+	*devp = lxa_state->lxas_dev_new;
+	return (0);
+}
+
+static int
+/*ARGSUSED*/
+lxa_close(dev_t dev, int flags, int otyp, cred_t *credp)
+{
+	lxa_state_t	*lxa_state;
+	minor_t		minor = getminor(dev);
+
+	/* handle devctl minor nodes (these nodes don't have a handle */
+	if (getminor(dev) == LXA_MINORNUM_DEVCTL)
+		return (0);
+
+	/* get the handle for this device */
+	if (mod_hash_find(lxa_state_hash, (mod_hash_key_t)(uintptr_t)minor,
+	    (mod_hash_val_t *)&lxa_state) != 0)
+		return (EINVAL);
+
+	lxa_state_close(lxa_state);
+	return (0);
+}
+
+static int
+/*ARGSUSED*/
+lxa_read(dev_t dev, struct uio *uiop, cred_t *credp)
+{
+	lxa_state_t	*lxa_state;
+	minor_t		minor = getminor(dev);
+	int		rv;
+
+	/* get the handle for this device */
+	if (mod_hash_find(lxa_state_hash, (mod_hash_key_t)(uintptr_t)minor,
+	    (mod_hash_val_t *)&lxa_state) != 0)
+		return (EINVAL);
+
+	/*
+	 * if a process has mmaped this device then we don't allow
+	 * any more reads or writes to the device
+	 */
+	if (lxa_state->lxas_umem_cookie != NULL)
+		return (EIO);
+
+	/* we can't do a read if there is no input device */
+	if (lxa_state->lxas_idev_lh == NULL)
+		return (EBADF);
+
+	/* pass the request on */
+	while (uiop->uio_resid != 0) {
+		rv = ldi_read(lxa_state->lxas_idev_lh, uiop, kcred);
+		if ((rv != 0) || (uiop->uio_fmode & (FNONBLOCK|FNDELAY))) {
+			break;
+		}
+	}
+	return (rv);
+}
+
+static int
+/*ARGSUSED*/
+lxa_write(dev_t dev, struct uio *uiop, cred_t *credp)
+{
+	lxa_state_t	*lxa_state;
+	minor_t		minor = getminor(dev);
+	int		rv;
+
+	/* get the handle for this device */
+	if (mod_hash_find(lxa_state_hash, (mod_hash_key_t)(uintptr_t)minor,
+	    (mod_hash_val_t *)&lxa_state) != 0)
+		return (EINVAL);
+
+	/*
+	 * if a process has mmaped this device then we don't allow
+	 * any more reads or writes to the device
+	 */
+	if (lxa_state->lxas_umem_cookie != NULL)
+		return (EIO);
+
+	/* we can't do a write if there is no output device */
+	if (lxa_state->lxas_odev_lh == NULL)
+		return (EBADF);
+
+	/* pass the request on */
+	while (uiop->uio_resid != 0) {
+		rv = ldi_write(lxa_state->lxas_odev_lh, uiop, kcred);
+		if ((rv != 0) || (uiop->uio_fmode & (FNONBLOCK|FNDELAY))) {
+			break;
+		}
+	}
+	return (rv);
+}
+
+static int
+/*ARGSUSED*/
+lxa_ioctl(dev_t dev, int cmd, intptr_t arg, int mode, cred_t *credp,
+    int *rvalp)
+{
+	lxa_state_t	*lxa_state;
+	minor_t		minor = getminor(dev);
+
+	/* handle devctl minor nodes (these nodes don't have a handle */
+	if (getminor(dev) == LXA_MINORNUM_DEVCTL)
+		return (lxa_ioctl_devctl(cmd, arg, mode));
+
+	/* get the handle for this device */
+	if (mod_hash_find(lxa_state_hash, (mod_hash_key_t)(uintptr_t)minor,
+	    (mod_hash_val_t *)&lxa_state) != 0)
+		return (EINVAL);
+
+	ASSERT((lxa_state->lxas_type == LXA_TYPE_AUDIO) ||
+	    (lxa_state->lxas_type == LXA_TYPE_AUDIOCTL));
+
+	switch (cmd) {
+	case LXA_IOC_GETMINORNUM:
+		{
+			int minornum = getminor(lxa_state->lxas_dev_old);
+			if (ddi_copyout(&minornum, (void *)arg,
+			    sizeof (minornum), mode) != 0)
+				return (EFAULT);
+		}
+		return (0);
+	}
+
+	if (lxa_state->lxas_type == LXA_TYPE_AUDIO) {
+		/* deal with native ioctl */
+		switch (cmd) {
+		case LXA_IOC_MMAP_OUTPUT:
+			return (lxa_ioc_mmap_output(lxa_state, arg, mode));
+		case LXA_IOC_MMAP_PTR:
+			return (lxa_ioc_mmap_ptr(lxa_state, arg, mode));
+		case LXA_IOC_GET_FRAG_INFO:
+			return (lxa_ioc_get_frag_info(lxa_state, arg, mode));
+		case LXA_IOC_SET_FRAG_INFO:
+			return (lxa_ioc_set_frag_info(lxa_state, arg, mode));
+		}
+
+		/* deal with layered ioctls */
+		switch (cmd) {
+		case AUDIO_DRAIN:
+			return (lxa_audio_drain(lxa_state));
+		case AUDIO_SETINFO:
+			return (lxa_audio_setinfo(lxa_state,
+			    AUDIO_SETINFO, arg, mode));
+		case AUDIO_GETINFO:
+			return (lxa_audio_getinfo(lxa_state, arg, mode));
+		}
+	}
+
+	if (lxa_state->lxas_type == LXA_TYPE_AUDIOCTL) {
+		/* deal with native ioctl */
+		switch (cmd) {
+		case LXA_IOC_MIXER_GET_VOL:
+			return (lxa_mixer_get_common(lxa_state,
+			    cmd, arg, mode));
+		case LXA_IOC_MIXER_SET_VOL:
+			return (lxa_mixer_set_common(lxa_state,
+			    cmd, arg, mode));
+		case LXA_IOC_MIXER_GET_MIC:
+			return (lxa_mixer_get_common(lxa_state,
+			    cmd, arg, mode));
+		case LXA_IOC_MIXER_SET_MIC:
+			return (lxa_mixer_set_common(lxa_state,
+			    cmd, arg, mode));
+		case LXA_IOC_MIXER_GET_PCM:
+			return (lxa_mixer_get_pcm(lxa_state, arg, mode));
+		case LXA_IOC_MIXER_SET_PCM:
+			return (lxa_mixer_set_pcm(lxa_state, arg, mode));
+		}
+
+	}
+
+	return (EINVAL);
+}
+
+static int
+/*ARGSUSED*/
+lxa_devmap(dev_t dev, devmap_cookie_t dhp,
+    offset_t off, size_t len, size_t *maplen, uint_t model)
+{
+	lxa_state_t		*lxa_state;
+	minor_t			minor = getminor(dev);
+	ddi_umem_cookie_t	umem_cookie;
+	void			*umem_ptr;
+	int			rv;
+
+	/* get the handle for this device */
+	if (mod_hash_find(lxa_state_hash, (mod_hash_key_t)(uintptr_t)minor,
+	    (mod_hash_val_t *)&lxa_state) != 0)
+		return (EINVAL);
+
+	/* we only support mmaping of audio devices */
+	if (lxa_state->lxas_type != LXA_TYPE_AUDIO)
+		return (EINVAL);
+
+	/* we only support output via mmap */
+	if ((lxa_state->lxas_flags & FWRITE) == 0)
+		return (EINVAL);
+
+	/* sanity check the amount of memory the user is allocating */
+	if ((len == 0) ||
+	    (len > LXA_OSS_FRAG_MEM) ||
+	    ((len % lxa_state->lxas_frag_size) != 0))
+		return (EINVAL);
+
+	/* allocate and clear memory to mmap */
+	umem_ptr = ddi_umem_alloc(len, DDI_UMEM_NOSLEEP, &umem_cookie);
+	if (umem_ptr == NULL)
+		return (ENOMEM);
+	bzero(umem_ptr, len);
+
+	/* setup the memory mappings */
+	rv = devmap_umem_setup(dhp, lxa_dip, NULL, umem_cookie, 0, len,
+	    PROT_USER | PROT_READ | PROT_WRITE, 0, NULL);
+	if (rv != 0) {
+		ddi_umem_free(umem_cookie);
+		return (EIO);
+	}
+
+	mutex_enter(&lxa_lock);
+
+	/* we only support one mmap per open */
+	if (lxa_state->lxas_umem_cookie != NULL) {
+		ASSERT(lxa_state->lxas_umem_ptr != NULL);
+		mutex_exit(&lxa_lock);
+		ddi_umem_free(umem_cookie);
+		return (EBUSY);
+	}
+	ASSERT(lxa_state->lxas_umem_ptr == NULL);
+
+	*maplen = len;
+	lxa_state->lxas_umem_len = len;
+	lxa_state->lxas_umem_ptr = umem_ptr;
+	lxa_state->lxas_umem_cookie = umem_cookie;
+	mutex_exit(&lxa_lock);
+	return (0);
+}
+
+static int
+/*ARGSUSED*/
+lxa_attach(dev_info_t *dip, ddi_attach_cmd_t cmd)
+{
+	int	instance = ddi_get_instance(dip);
+
+	if (cmd != DDI_ATTACH)
+		return (DDI_FAILURE);
+
+	ASSERT(instance == 0);
+	if (instance != 0)
+		return (DDI_FAILURE);
+
+	lxa_dip = dip;
+	mutex_init(&lxa_lock, NULL, MUTEX_DEFAULT, NULL);
+
+	/* create our minor nodes */
+	if (ddi_create_minor_node(dip, LXA_MINORNAME_DEVCTL, S_IFCHR,
+	    LXA_MINORNUM_DEVCTL, DDI_PSEUDO, 0) != DDI_SUCCESS)
+		return (DDI_FAILURE);
+
+	if (ddi_create_minor_node(dip, LXA_MINORNAME_DSP, S_IFCHR,
+	    LXA_MINORNUM_DSP, DDI_PSEUDO, 0) != DDI_SUCCESS)
+		return (DDI_FAILURE);
+
+	if (ddi_create_minor_node(dip, LXA_MINORNAME_MIXER, S_IFCHR,
+	    LXA_MINORNUM_MIXER, DDI_PSEUDO, 0) != DDI_SUCCESS)
+		return (DDI_FAILURE);
+
+	/* allocate our data structures */
+	lxa_minor_id = id_space_create("lxa_minor_id",
+	    LXA_MINORNUM_COUNT, LX_AUDIO_MAX_OPENS);
+	lxa_state_hash = mod_hash_create_idhash("lxa_state_hash",
+	    lxa_state_hash_size, mod_hash_null_valdtor);
+	lxa_zstate_hash = mod_hash_create_strhash("lxa_zstate_hash",
+	    lxa_zstate_hash_size, mod_hash_null_valdtor);
+
+	return (DDI_SUCCESS);
+}
+
+static int
+/*ARGSUSED*/
+lxa_detach(dev_info_t *dip, ddi_detach_cmd_t cmd)
+{
+	if (cmd != DDI_DETACH)
+		return (DDI_FAILURE);
+
+	ASSERT(!MUTEX_HELD(&lxa_lock));
+	if (lxa_registered_zones > 0)
+		return (DDI_FAILURE);
+
+	mod_hash_destroy_idhash(lxa_state_hash);
+	mod_hash_destroy_idhash(lxa_zstate_hash);
+	id_space_destroy(lxa_minor_id);
+	lxa_state_hash = NULL;
+	lxa_dip = NULL;
+
+	return (DDI_SUCCESS);
+}
+
+static int
+/*ARGSUSED*/
+lxa_getinfo(dev_info_t *dip, ddi_info_cmd_t infocmd, void *arg, void **resultp)
+{
+	switch (infocmd) {
+	case DDI_INFO_DEVT2DEVINFO:
+		*resultp = lxa_dip;
+		return (DDI_SUCCESS);
+
+	case DDI_INFO_DEVT2INSTANCE:
+		*resultp = (void *)0;
+		return (DDI_SUCCESS);
+	}
+	return (DDI_FAILURE);
+}
+
+/*
+ * Driver flags
+ */
+static struct cb_ops lxa_cb_ops = {
+	lxa_open,		/* open */
+	lxa_close,		/* close */
+	nodev,			/* strategy */
+	nodev,			/* print */
+	nodev,			/* dump */
+	lxa_read,		/* read */
+	lxa_write,		/* write */
+	lxa_ioctl,		/* ioctl */
+	lxa_devmap,		/* devmap */
+	nodev,			/* mmap */
+	ddi_devmap_segmap,	/* segmap */
+	nochpoll,		/* chpoll */
+	ddi_prop_op,		/* prop_op */
+	NULL,			/* cb_str */
+	D_NEW | D_MP | D_DEVMAP,
+	CB_REV,
+	NULL,
+	NULL
+};
+
+static struct dev_ops lxa_ops = {
+	DEVO_REV,
+	0,
+	lxa_getinfo,
+	nulldev,
+	nulldev,
+	lxa_attach,
+	lxa_detach,
+	nodev,
+	&lxa_cb_ops,
+	NULL,
+	NULL,
+	ddi_quiesce_not_needed,		/* quiesce */
+};
+
+/*
+ * Module linkage information for the kernel.
+ */
+static struct modldrv modldrv = {
+	&mod_driverops,		/* type of module */
+	"linux audio driver",	/* description of module */
+	&lxa_ops		/* driver ops */
+};
+
+static struct modlinkage modlinkage = {
+	MODREV_1,
+	&modldrv,
+	NULL
+};
+
+/*
+ * standard module entry points
+ */
+int
+_init(void)
+{
+	return (mod_install(&modlinkage));
+}
+
+int
+_fini(void)
+{
+	return (mod_remove(&modlinkage));
+}
+
+int
+_info(struct modinfo *modinfop)
+{
+	return (mod_info(&modlinkage, modinfop));
+}
diff --git a/usr/src/uts/common/brand/lx/io/lx_audio.conf b/usr/src/uts/common/brand/lx/io/lx_audio.conf
new file mode 100644
index 0000000000..2eeb5eb7ee
--- /dev/null
+++ b/usr/src/uts/common/brand/lx/io/lx_audio.conf
@@ -0,0 +1,27 @@
+#
+# CDDL HEADER START
+#
+# The contents of this file are subject to the terms of the
+# Common Development and Distribution License (the "License").
+# You may not use this file except in compliance with the License.
+#
+# You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
+# or http://www.opensolaris.org/os/licensing.
+# See the License for the specific language governing permissions
+# and limitations under the License.
+#
+# When distributing Covered Code, include this CDDL HEADER in each
+# file and include the License file at usr/src/OPENSOLARIS.LICENSE.
+# If applicable, add the following below this CDDL HEADER, with the
+# fields enclosed by brackets "[]" replaced with your own identifying
+# information: Portions Copyright [yyyy] [name of copyright owner]
+#
+# CDDL HEADER END
+#
+#
+# Copyright 2006 Sun Microsystems, Inc.  All rights reserved.
+# Use is subject to license terms.
+#
+#ident	"%Z%%M%	%I%	%E% SMI"
+
+name="lx_audio" parent="pseudo" instance=0;
diff --git a/usr/src/uts/common/brand/lx/io/lx_netlink.c b/usr/src/uts/common/brand/lx/io/lx_netlink.c
new file mode 100644
index 0000000000..35ee5b97ed
--- /dev/null
+++ b/usr/src/uts/common/brand/lx/io/lx_netlink.c
@@ -0,0 +1,1273 @@
+/*
+ * This file and its contents are supplied under the terms of the
+ * Common Development and Distribution License ("CDDL"), version 1.0.
+ * You may only use this file in accordance with the terms of version
+ * 1.0 of the CDDL.
+ *
+ * A full copy of the text of the CDDL should have accompanied this
+ * source.  A copy of the CDDL is also available via the Internet at
+ * http://www.illumos.org/license/CDDL.
+ */
+
+/*
+ * Copyright 2015 Joyent, Inc.
+ */
+
+/*
+ * Compatibility for the Linux netlink(7) kernel/user transport, as well as
+ * for in-kernel netlink(7) providers like rtnetlink(7).  See RFC 3549 for
+ * details of the protocol, and the Linux man pages for details of the Linux
+ * implementation that we're mimicking.
+ */
+
+#include <sys/strsubr.h>
+#include <sys/socket.h>
+#include <sys/socketvar.h>
+#include <sys/sunddi.h>
+#include <sys/sunldi.h>
+#include <sys/strsun.h>
+#include <sys/tihdr.h>
+#include <sys/sockio.h>
+#include <sys/brand.h>
+#include <inet/ip.h>
+#include <inet/ip_impl.h>
+#include <sys/lx_misc.h>
+#include <sys/ethernet.h>
+#include <sys/dlpi.h>
+#include <sys/priv_const.h>
+#include <sys/priv_impl.h>
+#include <sys/priv.h>
+#include <sys/policy.h>
+
+/*
+ * Flags in netlink header
+ */
+#define	LX_NETLINK_NLM_F_REQUEST		1
+#define	LX_NETLINK_NLM_F_MULTI			2
+#define	LX_NETLINK_NLM_F_ACK			4
+#define	LX_NETLINK_NLM_F_ECHO			8
+#define	LX_NETLINK_NLM_F_DUMP_INTR		16
+#define	LX_NETLINK_NLM_F_ROOT			0x100
+#define	LX_NETLINK_NLM_F_MATCH			0x200
+#define	LX_NETLINK_NLM_F_ATOMIC			0x400
+
+/*
+ * Generic message type constants
+ */
+#define	LX_NETLINK_NLMSG_NONE			0
+#define	LX_NETLINK_NLMSG_NOOP			1
+#define	LX_NETLINK_NLMSG_ERROR			2
+#define	LX_NETLINK_NLMSG_DONE			3
+#define	LX_NETLINK_NLMSG_OVERRUN		4
+
+/*
+ * Protocol constants.
+ */
+#define	LX_NETLINK_ROUTE			0
+#define	LX_NETLINK_UNUSED			1
+#define	LX_NETLINK_USERSOCK			2
+#define	LX_NETLINK_FIREWALL			3
+#define	LX_NETLINK_SOCK_DIAG			4
+#define	LX_NETLINK_NFLOG			5
+#define	LX_NETLINK_XFRM				6
+#define	LX_NETLINK_SELINUX			7
+#define	LX_NETLINK_ISCSI			8
+#define	LX_NETLINK_AUDIT			9
+#define	LX_NETLINK_FIB_LOOKUP			10
+#define	LX_NETLINK_CONNECTOR			11
+#define	LX_NETLINK_NETFILTER			12
+#define	LX_NETLINK_IP6_FW			13
+#define	LX_NETLINK_DNRTMSG			14
+#define	LX_NETLINK_KOBJECT_UEVENT		15
+#define	LX_NETLINK_GENERIC			16
+#define	LX_NETLINK_SCSITRANSPORT		18
+#define	LX_NETLINK_ECRYPTFS			19
+#define	LX_NETLINK_RDMA				20
+#define	LX_NETLINK_CRYPTO			21
+
+/*
+ * rtnetlink(7) attribute-related constants
+ */
+#define	LX_NETLINK_NLA_ALIGNTO			4
+
+#define	LX_NETLINK_RTM_NEWLINK			16
+#define	LX_NETLINK_RTM_DELLINK			17
+#define	LX_NETLINK_RTM_GETLINK			18
+#define	LX_NETLINK_RTM_SETLINK			19
+#define	LX_NETLINK_RTM_NEWADDR			20
+#define	LX_NETLINK_RTM_DELADDR			21
+#define	LX_NETLINK_RTM_GETADDR			22
+#define	LX_NETLINK_RTM_NEWROUTE			24
+#define	LX_NETLINK_RTM_DELROUTE			25
+#define	LX_NETLINK_RTM_GETROUTE			26
+#define	LX_NETLINK_RTM_NEWNEIGH			28
+#define	LX_NETLINK_RTM_DELNEIGH			29
+#define	LX_NETLINK_RTM_GETNEIGH			30
+#define	LX_NETLINK_RTM_NEWRULE			32
+#define	LX_NETLINK_RTM_DELRULE			33
+#define	LX_NETLINK_RTM_GETRULE			34
+#define	LX_NETLINK_RTM_NEWQDISC			36
+#define	LX_NETLINK_RTM_DELQDISC			37
+#define	LX_NETLINK_RTM_GETQDISC			38
+#define	LX_NETLINK_RTM_NEWTCLASS		40
+#define	LX_NETLINK_RTM_DELTCLASS		41
+#define	LX_NETLINK_RTM_GETTCLASS		42
+#define	LX_NETLINK_RTM_NEWTFILTER		44
+#define	LX_NETLINK_RTM_DELTFILTER		45
+#define	LX_NETLINK_RTM_GETTFILTER		46
+#define	LX_NETLINK_RTM_NEWACTION		48
+#define	LX_NETLINK_RTM_DELACTION		49
+#define	LX_NETLINK_RTM_GETACTION		50
+#define	LX_NETLINK_RTM_NEWPREFIX		52
+#define	LX_NETLINK_RTM_GETMULTICAST		58
+#define	LX_NETLINK_RTM_GETANYCAST		62
+#define	LX_NETLINK_RTM_NEWNEIGHTBL		64
+#define	LX_NETLINK_RTM_GETNEIGHTBL		66
+#define	LX_NETLINK_RTM_SETNEIGHTBL		67
+#define	LX_NETLINK_RTM_NEWNDUSEROPT		68
+#define	LX_NETLINK_RTM_NEWADDRLABEL		72
+#define	LX_NETLINK_RTM_DELADDRLABEL		73
+#define	LX_NETLINK_RTM_GETADDRLABEL		74
+#define	LX_NETLINK_RTM_GETDCB			78
+#define	LX_NETLINK_RTM_SETDCB			79
+#define	LX_NETLINK_RTM_NEWNETCONF		80
+#define	LX_NETLINK_RTM_GETNETCONF		82
+#define	LX_NETLINK_RTM_NEWMDB			84
+#define	LX_NETLINK_RTM_DELMDB			85
+#define	LX_NETLINK_RTM_GETMDB			86
+#define	LX_NETLINK_RTM_MAX			87
+
+/*
+ * rtnetlink(7) attribute constants
+ */
+#define	LX_NETLINK_RTA_UNSPEC			1
+#define	LX_NETLINK_RTA_DST			2
+#define	LX_NETLINK_RTA_SRC			3
+#define	LX_NETLINK_RTA_IIF			4
+#define	LX_NETLINK_RTA_OIF			5
+#define	LX_NETLINK_RTA_GATEWAY			6
+#define	LX_NETLINK_RTA_PRIORITY			7
+#define	LX_NETLINK_RTA_PREFSRC			8
+#define	LX_NETLINK_RTA_METRICS			9
+#define	LX_NETLINK_RTA_MULTIPATH		10
+#define	LX_NETLINK_RTA_PROTOINFO		11
+#define	LX_NETLINK_RTA_FLOW			12
+#define	LX_NETLINK_RTA_CACHEINFO		13
+#define	LX_NETLINK_RTA_SESSION			14
+#define	LX_NETLINK_RTA_MP_ALGO			15
+#define	LX_NETLINK_RTA_TABLE			16
+#define	LX_NETLINK_RTA_MARK			17
+#define	LX_NETLINK_RTA_MFC_STATS		18
+
+/*
+ * rtnetlink(7) NEWLINK/DELLINK/GETLINK constants
+ */
+#define	LX_NETLINK_IFLA_UNSPEC			0
+#define	LX_NETLINK_IFLA_ADDRESS			1
+#define	LX_NETLINK_IFLA_BROADCAST		2
+#define	LX_NETLINK_IFLA_IFNAME			3
+#define	LX_NETLINK_IFLA_MTU			4
+#define	LX_NETLINK_IFLA_LINK			5
+#define	LX_NETLINK_IFLA_QDISC			6
+#define	LX_NETLINK_IFLA_STATS			7
+#define	LX_NETLINK_IFLA_COST			8
+#define	LX_NETLINK_IFLA_PRIORITY		9
+#define	LX_NETLINK_IFLA_MASTER			10
+#define	LX_NETLINK_IFLA_WIRELESS		11
+#define	LX_NETLINK_IFLA_PROTINFO		12
+#define	LX_NETLINK_IFLA_TXQLEN			13
+#define	LX_NETLINK_IFLA_MAP			14
+#define	LX_NETLINK_IFLA_WEIGHT			15
+#define	LX_NETLINK_IFLA_OPERSTATE		16
+#define	LX_NETLINK_IFLA_LINKMODE		17
+#define	LX_NETLINK_IFLA_LINKINFO		18
+#define	LX_NETLINK_IFLA_NET_NS_PID		19
+#define	LX_NETLINK_IFLA_IFALIAS			20
+#define	LX_NETLINK_IFLA_NUM_VF			21
+#define	LX_NETLINK_IFLA_VFINFO_LIST		22
+#define	LX_NETLINK_IFLA_STATS64			23
+#define	LX_NETLINK_IFLA_VF_PORTS		24
+#define	LX_NETLINK_IFLA_PORT_SELF		25
+#define	LX_NETLINK_IFLA_AF_SPEC			26
+#define	LX_NETLINK_IFLA_GROUP			27
+#define	LX_NETLINK_IFLA_NET_NS_FD		28
+#define	LX_NETLINK_IFLA_EXT_MASK		29
+#define	LX_NETLINK_IFLA_PROMISCUITY		30
+#define	LX_NETLINK_IFLA_NUM_TX_QUEUES		31
+#define	LX_NETLINK_IFLA_NUM_RX_QUEUES		32
+#define	LX_NETLINK_IFLA_CARRIER			33
+#define	LX_NETLINK_IFLA_PHYS_PORT_ID		34
+#define	LX_NETLINK_IFLA_CARRIER_CHANGES		35
+#define	LX_NETLINK_IFLA_MAX			36
+
+/*
+ * rtnetlink(7) NEWADDR/DELADDR/GETADDR constants
+ */
+#define	LX_NETLINK_IFA_UNSPEC			0
+#define	LX_NETLINK_IFA_ADDRESS			1
+#define	LX_NETLINK_IFA_LOCAL			2
+#define	LX_NETLINK_IFA_LABEL			3
+#define	LX_NETLINK_IFA_BROADCAST		4
+#define	LX_NETLINK_IFA_ANYCAST			5
+#define	LX_NETLINK_IFA_CACHEINFO		6
+#define	LX_NETLINK_IFA_MULTICAST		7
+#define	LX_NETLINK_IFA_FLAGS			8
+#define	LX_NETLINK_IFA_MAX			9
+
+#define	LX_NETLINK_IFA_F_SECONDARY		0x01
+#define	LX_NETLINK_IFA_F_TEMPORARY		LX_NETLINK_IFA_F_SECONDARY
+#define	LX_NETLINK_IFA_F_NODAD			0x02
+#define	LX_NETLINK_IFA_F_OPTIMISTIC		0x04
+#define	LX_NETLINK_IFA_F_DADFAILED		0x08
+#define	LX_NETLINK_IFA_F_HOMEADDRESS		0x10
+#define	LX_NETLINK_IFA_F_DEPRECATED		0x20
+#define	LX_NETLINK_IFA_F_TENTATIVE		0x40
+#define	LX_NETLINK_IFA_F_PERMANENT		0x80
+#define	LX_NETLINK_IFA_F_MANAGETEMPADDR		0x100
+#define	LX_NETLINK_IFA_F_NOPREFIXROUTE		0x200
+
+/*
+ * Linux address families.
+ */
+#define	LX_AF_INET		2
+#define	LX_AF_INET6		10
+#define	LX_AF_NETLINK		16
+
+/*
+ * Linux interface flags.
+ */
+#define	LX_IFF_UP		(1<<0)
+#define	LX_IFF_BROADCAST	(1<<1)
+#define	LX_IFF_DEBUG		(1<<2)
+#define	LX_IFF_LOOPBACK		(1<<3)
+#define	LX_IFF_POINTOPOINT	(1<<4)
+#define	LX_IFF_NOTRAILERS	(1<<5)
+#define	LX_IFF_RUNNING		(1<<6)
+#define	LX_IFF_NOARP		(1<<7)
+#define	LX_IFF_PROMISC		(1<<8)
+#define	LX_IFF_ALLMULTI		(1<<9)
+#define	LX_IFF_MASTER		(1<<10)
+#define	LX_IFF_SLAVE		(1<<11)
+#define	LX_IFF_MULTICAST	(1<<12)
+#define	LX_IFF_PORTSEL		(1<<13)
+#define	LX_IFF_AUTOMEDIA	(1<<14)
+#define	LX_IFF_DYNAMIC		(1<<15)
+#define	LX_IFF_LOWER_UP		(1<<16)
+#define	LX_IFF_DORMANT		(1<<17)
+#define	LX_IFF_ECHO		(1<<18)
+
+
+/*
+ * Netlink sockopts
+ */
+#define	SOL_LX_NETLINK	270
+
+#define	LX_NETLINK_SO_ADD_MEMBERSHIP	1
+#define	LX_NETLINK_SO_DROP_MEMBERSHIP	2
+#define	LX_NETLINK_SO_PKTINFO		3
+#define	LX_NETLINK_SO_BROADCAST_ERROR	4
+#define	LX_NETLINK_SO_NO_ENOBUFS	5
+#define	LX_NETLINK_SO_RX_RING		6
+#define	LX_NETLINK_SO_TX_RING		7
+
+typedef struct lx_netlink_hdr {
+	uint32_t lxnh_len;			/* length of message */
+	uint16_t lxnh_type;			/* type of message */
+	uint16_t lxnh_flags;			/* flags */
+	uint32_t lxnh_seq;			/* sequence number */
+	uint32_t lxnh_pid;			/* sending pid */
+} lx_netlink_hdr_t;
+
+typedef struct lx_netlink_err {
+	lx_netlink_hdr_t	lxne_hdr;	/* header */
+	int32_t			lxne_errno;	/* errno */
+	lx_netlink_hdr_t	lxne_failed;	/* header of err */
+} lx_netlink_err_t;
+
+typedef struct lx_netlink_attr {
+	uint16_t	lxna_len;		/* length of attribute */
+	uint16_t	lxna_type;		/* type of attribute */
+} lx_netlink_attr_t;
+
+typedef struct lx_netlink_ifinfomsg {
+	uint8_t		lxnl_ifi_family;	/* family: AF_UNSPEC */
+	uint8_t		lxnl_ifi__pad;
+	uint16_t	lxnl_ifi_type;		/* device type */
+	uint32_t	lxnl_ifi_index;		/* interface index */
+	uint32_t	lxnl_ifi_flags;		/* device flags */
+	uint32_t 	lxnl_ifi_change;	/* unused; must be -1 */
+} lx_netlink_ifinfomsg_t;
+
+typedef struct lx_netlink_ifaddrmsg {
+	uint8_t		lxnl_ifa_family;	/* address type */
+	uint8_t		lxnl_ifa_prefixlen;	/* prefix length of address */
+	uint8_t		lxnl_ifa_flags;		/* address flags */
+	uint8_t		lxnl_ifa_scope;		/* address scope */
+	uint8_t		lxnl_ifa_index;		/* interface index */
+} lx_netlink_ifaddrmsg_t;
+
+typedef struct lx_netlink_sockaddr {
+	sa_family_t	lxnl_family;		/* AF_LX_NETLINK */
+	uint16_t	lxnl_pad;		/* padding */
+	uint32_t	lxnl_port;		/* port id */
+	uint32_t	lxnl_groups;		/* multicast groups mask */
+} lx_netlink_sockaddr_t;
+
+typedef struct lx_netlink_sock {
+	struct lx_netlink_sock *lxns_next;	/* list of lx_netlink sockets */
+	sock_upcalls_t *lxns_upcalls;		/* pointer to socket upcalls */
+	sock_upper_handle_t lxns_uphandle;	/* socket upcall handle */
+	ldi_handle_t lxns_iphandle;		/* handle to /dev/ip */
+	ldi_handle_t lxns_ip6handle;		/* handle to /dev/ip6 */
+	ldi_handle_t lxns_current;		/* current ip handle */
+	int lxns_proto;				/* protocol */
+	uint32_t lxns_port;			/* port identifier */
+	uint32_t lxns_groups;			/* group subscriptions */
+	uint32_t lxns_bufsize;			/* buffer size */
+} lx_netlink_sock_t;
+
+typedef struct lx_netlink_reply {
+	lx_netlink_hdr_t lxnr_hdr;		/* header that we're reply to */
+	lx_netlink_sock_t *lxnr_sock;		/* socket */
+	uint32_t lxnr_seq;			/* sequence number */
+	uint16_t lxnr_type;			/* type of reply */
+	mblk_t *lxnr_mp;			/* current mblk */
+	mblk_t *lxnr_err;			/* error mblk */
+	mblk_t *lxnr_mp1;			/* T_UNITDATA_IND mblk */
+	int lxnr_errno;				/* errno, if any */
+} lx_netlink_reply_t;
+
+static lx_netlink_sock_t *lx_netlink_head;	/* head of lx_netlink sockets */
+static kmutex_t lx_netlink_lock;		/* lock to protect state */
+static ldi_ident_t lx_netlink_ldi;		/* LDI handle */
+static int lx_netlink_bufsize = 4096;		/* default buffer size */
+static int lx_netlink_flowctrld;		/* # of times flow controlled */
+
+/*ARGSUSED*/
+static void
+lx_netlink_activate(sock_lower_handle_t handle,
+    sock_upper_handle_t sock_handle, sock_upcalls_t *sock_upcalls,
+    int flags, cred_t *cr)
+{
+	lx_netlink_sock_t *lxsock = (lx_netlink_sock_t *)handle;
+	struct sock_proto_props sopp;
+
+	sopp.sopp_flags = SOCKOPT_WROFF | SOCKOPT_RCVHIWAT |
+	    SOCKOPT_RCVLOWAT | SOCKOPT_MAXADDRLEN | SOCKOPT_MAXPSZ |
+	    SOCKOPT_MAXBLK | SOCKOPT_MINPSZ;
+	sopp.sopp_wroff = 0;
+	sopp.sopp_rxhiwat = SOCKET_RECVHIWATER;
+	sopp.sopp_rxlowat = SOCKET_RECVLOWATER;
+	sopp.sopp_maxaddrlen = sizeof (struct sockaddr_dl);
+	sopp.sopp_maxpsz = INFPSZ;
+	sopp.sopp_maxblk = INFPSZ;
+	sopp.sopp_minpsz = 0;
+
+	lxsock->lxns_upcalls = sock_upcalls;
+	lxsock->lxns_uphandle = sock_handle;
+
+	sock_upcalls->su_set_proto_props(sock_handle, &sopp);
+}
+
+/*ARGSUSED*/
+static int
+lx_netlink_setsockopt(sock_lower_handle_t handle, int level,
+    int option_name, const void *optval, socklen_t optlen, struct cred *cr)
+{
+	if (level == SOL_SOCKET) {
+		return (0);
+	} else if (level != SOL_LX_NETLINK) {
+		return (EOPNOTSUPP);
+	}
+
+	switch (option_name) {
+	case LX_NETLINK_SO_ADD_MEMBERSHIP:
+	case LX_NETLINK_SO_DROP_MEMBERSHIP:
+	case LX_NETLINK_SO_PKTINFO:
+	case LX_NETLINK_SO_BROADCAST_ERROR:
+	case LX_NETLINK_SO_NO_ENOBUFS:
+	case LX_NETLINK_SO_RX_RING:
+	case LX_NETLINK_SO_TX_RING:
+		/* Blatant lie */
+		return (0);
+	default:
+		return (EINVAL);
+	}
+}
+
+/*ARGSUSED*/
+static int
+lx_netlink_bind(sock_lower_handle_t handle, struct sockaddr *name,
+    socklen_t namelen, struct cred *cr)
+{
+	lx_netlink_sock_t *lxsock = (lx_netlink_sock_t *)handle;
+	lx_netlink_sockaddr_t *lxsa = (lx_netlink_sockaddr_t *)name;
+
+	if (namelen != sizeof (lx_netlink_sockaddr_t) ||
+	    lxsa->lxnl_family != AF_LX_NETLINK) {
+		return (EINVAL);
+	}
+
+
+	if (lxsa->lxnl_groups != 0) {
+		/*
+		 * On linux, CAP_NET_ADMIN is needed to bind to netlink groups.
+		 * This roughly maps to PRIV_SYS_IP_CONFIG.
+		 */
+		if (secpolicy_ip_config(cr, B_FALSE) != 0) {
+			return (EACCES);
+		}
+
+		/* Lie about group subscription for now */
+		lxsock->lxns_groups = lxsa->lxnl_groups;
+	}
+
+	/*
+	 * Linux netlink uses nl_port to identify distinct netlink sockets.
+	 * Binding to an address of nl_port=0 triggers the kernel to
+	 * automatically assign a free nl_port identifier.  Originally,
+	 * consumers of lx_netlink were required to bind with that automatic
+	 * address.  We now support non-zero values for nl_port although strict
+	 * checking to identify conflicts is not performed.  Use of the
+	 * id_space facility could be a convenient solution, if a need arose.
+	 */
+	if (lxsa->lxnl_port == 0) {
+		/*
+		 * Because we are not doing conflict detection, there is no
+		 * need to expend effort selecting a unique port for automatic
+		 * addressing during bind.
+		 */
+		lxsock->lxns_port = curproc->p_pid;
+	} else {
+		lxsock->lxns_port = lxsa->lxnl_port;
+	}
+
+	return (0);
+}
+
+/*ARGSUSED*/
+static int
+lx_netlink_getsockname(sock_lower_handle_t handle, struct sockaddr *sa,
+    socklen_t *len, struct cred *cr)
+{
+	lx_netlink_sock_t *lxsock = (lx_netlink_sock_t *)handle;
+	lx_netlink_sockaddr_t *lxsa = (lx_netlink_sockaddr_t *)sa;
+
+	if (*len < sizeof (lx_netlink_sockaddr_t))
+		return (EINVAL);
+
+	lxsa->lxnl_family = AF_LX_NETLINK;
+	lxsa->lxnl_pad = 0;
+	lxsa->lxnl_port = lxsock->lxns_port;
+	lxsa->lxnl_groups = lxsock->lxns_groups;
+
+	*len = sizeof (lx_netlink_sockaddr_t);
+
+	return (0);
+}
+
+static mblk_t *
+lx_netlink_alloc_mp1()
+{
+	return (allocb(sizeof (struct T_unitdata_ind) +
+	    sizeof (lx_netlink_sockaddr_t), 0));
+}
+
+static lx_netlink_reply_t *
+lx_netlink_reply(lx_netlink_sock_t *lxsock,
+    lx_netlink_hdr_t *hdr, uint16_t type)
+{
+	lx_netlink_reply_t *reply;
+	mblk_t *err, *mp1;
+
+	/*
+	 * We always allocate an error block to assure that even if subsequent
+	 * allocations fail, we can return an error.
+	 */
+	if ((err = allocb(sizeof (lx_netlink_err_t), 0)) == NULL)
+		return (NULL);
+
+	if ((mp1 = lx_netlink_alloc_mp1()) == NULL) {
+		freeb(err);
+		return (NULL);
+	}
+
+	reply = kmem_zalloc(sizeof (lx_netlink_reply_t), KM_SLEEP);
+	reply->lxnr_err = err;
+	reply->lxnr_sock = lxsock;
+	reply->lxnr_hdr = *hdr;
+	reply->lxnr_type = type;
+	reply->lxnr_mp1 = mp1;
+
+	return (reply);
+}
+
+static void
+lx_netlink_reply_add(lx_netlink_reply_t *reply, void *payload, uint32_t size)
+{
+	lx_netlink_hdr_t *hdr;
+	lx_netlink_sock_t *lxsock = reply->lxnr_sock;
+	uint32_t align, alignto = LX_NETLINK_NLA_ALIGNTO;
+	mblk_t *mp = reply->lxnr_mp;
+
+	if (reply->lxnr_errno)
+		return;
+
+	align = (alignto - (size & (alignto - 1))) & (alignto - 1);
+
+	hdr = (lx_netlink_hdr_t *)mp->b_rptr;
+
+	if (hdr->lxnh_len + size + align > lxsock->lxns_bufsize) {
+		reply->lxnr_errno = E2BIG;
+		return;
+	}
+
+	bcopy(payload, mp->b_wptr, size);
+
+	hdr->lxnh_len += size + align;
+	mp->b_wptr += size + align;
+}
+
+static void
+lx_netlink_reply_msg(lx_netlink_reply_t *reply, void *payload, uint32_t size)
+{
+	lx_netlink_hdr_t *hdr;
+	lx_netlink_sock_t *lxsock = reply->lxnr_sock;
+	mblk_t *mp;
+
+	if (reply->lxnr_errno)
+		return;
+
+	VERIFY(reply->lxnr_mp == NULL);
+
+	if ((reply->lxnr_mp = mp = allocb(lxsock->lxns_bufsize, 0)) == NULL) {
+		reply->lxnr_errno = ENOMEM;
+		return;
+	}
+
+	bzero(mp->b_rptr, lxsock->lxns_bufsize);
+	hdr = (lx_netlink_hdr_t *)mp->b_rptr;
+	hdr->lxnh_flags = LX_NETLINK_NLM_F_MULTI;
+	hdr->lxnh_len = sizeof (lx_netlink_hdr_t);
+	hdr->lxnh_seq = reply->lxnr_hdr.lxnh_seq;
+	hdr->lxnh_pid = lxsock->lxns_port;
+
+	mp->b_wptr += sizeof (lx_netlink_hdr_t);
+
+	if (payload == NULL) {
+		/*
+		 * A NULL payload denotes a "done" message.
+		 */
+		hdr->lxnh_type = LX_NETLINK_NLMSG_DONE;
+	} else {
+		hdr->lxnh_type = reply->lxnr_type;
+		lx_netlink_reply_add(reply, payload, size);
+	}
+}
+
+static void
+lx_netlink_reply_attr(lx_netlink_reply_t *reply, uint16_t type,
+    void *payload, uint32_t size)
+{
+	lx_netlink_attr_t attr;
+
+	attr.lxna_len = size + sizeof (lx_netlink_attr_t);
+	attr.lxna_type = type;
+
+	lx_netlink_reply_add(reply, &attr, sizeof (attr));
+	lx_netlink_reply_add(reply, payload, size);
+}
+
+static void
+lx_netlink_reply_attr_string(lx_netlink_reply_t *reply,
+    uint16_t type, const char *str)
+{
+	lx_netlink_reply_attr(reply, type, (void *)str, strlen(str) + 1);
+}
+
+static void
+lx_netlink_reply_attr_int32(lx_netlink_reply_t *reply,
+    uint16_t type, int32_t val)
+{
+	int32_t v = val;
+
+	lx_netlink_reply_attr(reply, type, &v, sizeof (int32_t));
+}
+
+static int
+lx_netlink_reply_ioctl(lx_netlink_reply_t *reply, int cmd, void *arg)
+{
+	int rval;
+
+	if (reply->lxnr_errno != 0)
+		return (reply->lxnr_errno);
+
+	if ((rval = ldi_ioctl(reply->lxnr_sock->lxns_current,
+	    cmd, (intptr_t)arg, FKIOCTL, kcred, NULL)) != 0) {
+		reply->lxnr_errno = rval;
+	}
+
+	return (rval);
+}
+
+static void
+lx_netlink_reply_sendup(lx_netlink_reply_t *reply, mblk_t *mp, mblk_t *mp1)
+{
+	lx_netlink_sock_t *lxsock = reply->lxnr_sock;
+	lx_netlink_sockaddr_t *lxsa;
+	struct T_unitdata_ind *tunit;
+	int error;
+
+	/*
+	 * To prevent the stream head from coalescing messages and to indicate
+	 * their origin, we send them as T_UNITDATA_IND messages, not as raw
+	 * M_DATA.
+	 */
+	mp1->b_cont = mp;
+
+	mp = mp1;
+	mp->b_datap->db_type = M_PROTO;
+
+	tunit = (struct T_unitdata_ind *)mp->b_rptr;
+	tunit->PRIM_type = T_UNITDATA_IND;
+	tunit->SRC_length = sizeof (lx_netlink_sockaddr_t);
+	tunit->SRC_offset = sizeof (*tunit);
+	tunit->OPT_length = 0;
+
+	lxsa = (lx_netlink_sockaddr_t *)(mp->b_rptr + sizeof (*tunit));
+	lxsa->lxnl_family = AF_LX_NETLINK;
+	lxsa->lxnl_port = 0;
+	lxsa->lxnl_groups = 0;
+	lxsa->lxnl_pad = 0;
+
+	mp->b_wptr += sizeof (*tunit) + sizeof (*lxsa);
+
+	lxsock->lxns_upcalls->su_recv(lxsock->lxns_uphandle, mp,
+	    msgdsize(mp), 0, &error, NULL);
+
+	if (error != 0)
+		lx_netlink_flowctrld++;
+}
+
+static void
+lx_netlink_reply_send(lx_netlink_reply_t *reply)
+{
+	mblk_t *mp1;
+
+	if (reply->lxnr_errno)
+		return;
+
+	if ((mp1 = lx_netlink_alloc_mp1()) == NULL) {
+		reply->lxnr_errno = ENOMEM;
+		return;
+	}
+
+	lx_netlink_reply_sendup(reply, reply->lxnr_mp, mp1);
+	reply->lxnr_mp = NULL;
+}
+
+static void
+lx_netlink_reply_done(lx_netlink_reply_t *reply)
+{
+	lx_netlink_sock_t *lxsock = reply->lxnr_sock;
+	mblk_t *mp;
+
+	/*
+	 * Denote that we're done via a message with a NULL payload.
+	 */
+	lx_netlink_reply_msg(reply, NULL, 0);
+
+	if (reply->lxnr_errno) {
+		/*
+		 * If anything failed, we'll send up an error message.
+		 */
+		lx_netlink_hdr_t *hdr;
+		lx_netlink_err_t *err;
+
+		if (reply->lxnr_mp != NULL) {
+			freeb(reply->lxnr_mp);
+			reply->lxnr_mp = NULL;
+		}
+
+		mp = reply->lxnr_err;
+		err = (lx_netlink_err_t *)mp->b_rptr;
+		hdr = &err->lxne_hdr;
+
+		hdr->lxnh_type = LX_NETLINK_NLMSG_ERROR;
+		hdr->lxnh_seq = reply->lxnr_hdr.lxnh_seq;
+		hdr->lxnh_len = sizeof (lx_netlink_err_t);
+		hdr->lxnh_seq = reply->lxnr_hdr.lxnh_seq;
+		hdr->lxnh_pid = lxsock->lxns_port;
+		err->lxne_failed = reply->lxnr_hdr;
+		err->lxne_errno = reply->lxnr_errno;
+		mp->b_wptr += sizeof (lx_netlink_err_t);
+		reply->lxnr_err = NULL;
+	} else {
+		mp = reply->lxnr_mp;
+		VERIFY(mp != NULL);
+		reply->lxnr_mp = NULL;
+	}
+
+	lx_netlink_reply_sendup(reply, mp, reply->lxnr_mp1);
+
+	if (reply->lxnr_mp != NULL)
+		freeb(reply->lxnr_mp);
+
+	if (reply->lxnr_err != NULL)
+		freeb(reply->lxnr_err);
+
+	kmem_free(reply, sizeof (lx_netlink_reply_t));
+}
+
+static int
+lx_netlink_reply_error(lx_netlink_sock_t *lxsock,
+    lx_netlink_hdr_t *hdr, int errno)
+{
+	/*
+	 * The type of the message doesn't matter, as we're going to explicitly
+	 * set lxnr_errno and therefore send only an error message.
+	 */
+	lx_netlink_reply_t *reply = lx_netlink_reply(lxsock, hdr, 0);
+
+	if (reply == NULL)
+		return (ENOMEM);
+
+	reply->lxnr_errno = errno;
+	lx_netlink_reply_done(reply);
+
+	return (0);
+}
+
+static void
+lx_netlink_getlink_lifreq(lx_netlink_reply_t *reply, struct lifreq *lifr)
+{
+	lx_netlink_ifinfomsg_t ifi;
+	int i;
+	char if_name[IFNAMSIZ];
+	struct sockaddr_dl *sdl;
+	struct sockaddr hwaddr;
+	int hwaddr_size;
+	boolean_t is_loopback;
+
+	struct {
+		int native;
+		int lx;
+	} flags[] = {
+		{ IFF_UP, LX_IFF_UP },
+		{ IFF_BROADCAST, LX_IFF_BROADCAST },
+		{ IFF_DEBUG, LX_IFF_DEBUG },
+		{ IFF_LOOPBACK, LX_IFF_LOOPBACK },
+		{ IFF_POINTOPOINT, LX_IFF_POINTOPOINT },
+		{ IFF_NOTRAILERS, LX_IFF_NOTRAILERS },
+		{ IFF_RUNNING, LX_IFF_RUNNING },
+		{ IFF_NOARP, LX_IFF_NOARP },
+		{ IFF_PROMISC, LX_IFF_PROMISC },
+		{ IFF_ALLMULTI, LX_IFF_ALLMULTI },
+		{ IFF_MULTICAST, LX_IFF_MULTICAST },
+		{ 0 }
+	};
+
+	/*
+	 * Most of the lx_netlink module is architected to emit information in
+	 * an illumos-native manner.  Socket syscalls such as getsockname will
+	 * not translate fields to values Linux programs would expect since
+	 * that conversion is performed by the generic socket emulation.
+	 *
+	 * This is _not_ true of the actual protocol output from lx_netlink.
+	 * Since translating it at the socket layer would be onerous, all
+	 * output (including constants and names) is pre-translated to values
+	 * valid for Linux.
+	 */
+
+	bzero(&ifi, sizeof (ifi));
+	ifi.lxnl_ifi_family = AF_UNSPEC;
+	ifi.lxnl_ifi_change = (uint32_t)-1;
+
+	/* Convert the name to be Linux-friendly */
+	(void) strlcpy(if_name, lifr->lifr_name, IFNAMSIZ);
+	lx_ifname_convert(if_name, LX_IF_FROMNATIVE);
+	is_loopback = (strncmp(if_name, "lo", 2) == 0);
+
+	if (lx_netlink_reply_ioctl(reply, SIOCGLIFINDEX, lifr) != 0)
+		return;
+
+	ifi.lxnl_ifi_index = lifr->lifr_index;
+
+	if (lx_netlink_reply_ioctl(reply, SIOCGLIFFLAGS, lifr) != 0)
+		return;
+
+	for (i = 0; flags[i].native; i++) {
+		if (lifr->lifr_flags & flags[i].native)
+			ifi.lxnl_ifi_flags |= flags[i].lx;
+	}
+
+	/*
+	 * Query the datalink address.
+	 * The interface type will be included in the outgoing infomsg while
+	 * the address itself will be output separately.
+	 */
+	sdl = (struct sockaddr_dl *)&lifr->lifr_addr;
+	bzero(sdl, sizeof (*sdl));
+	if (!is_loopback) {
+		lx_netlink_reply_ioctl(reply, SIOCGLIFHWADDR, lifr);
+	} else {
+		/* Simulate an empty hwaddr for loopback */
+		sdl->sdl_type = DL_LOOP;
+		sdl->sdl_alen = ETHERADDRL;
+	}
+	lx_stol_hwaddr(sdl, &hwaddr, &hwaddr_size);
+
+	ifi.lxnl_ifi_type = hwaddr.sa_family;
+	lx_netlink_reply_msg(reply, &ifi, sizeof (lx_netlink_ifinfomsg_t));
+
+	lx_netlink_reply_attr_string(reply, LX_NETLINK_IFLA_IFNAME, if_name);
+
+	if (lx_netlink_reply_ioctl(reply, SIOCGLIFMTU, lifr) != 0)
+		return;
+
+	lx_netlink_reply_attr_int32(reply, LX_NETLINK_IFLA_MTU, lifr->lifr_mtu);
+
+	if (hwaddr_size != 0) {
+		lx_netlink_reply_attr(reply, LX_NETLINK_IFLA_ADDRESS,
+		    hwaddr.sa_data, hwaddr_size);
+	}
+
+	/* Emulate a txqlen of 1. (0 for loopbacks) */
+	lx_netlink_reply_attr_int32(reply, LX_NETLINK_IFLA_TXQLEN,
+	    (is_loopback) ? 0 : 1);
+
+	lx_netlink_reply_send(reply);
+}
+
+static void
+lx_netlink_reply_eachfamily(lx_netlink_reply_t *reply,
+    void (*func)(lx_netlink_reply_t *, struct lifreq *), boolean_t distinct)
+{
+	lx_netlink_sock_t *sock = reply->lxnr_sock;
+	int nlifr, i;
+
+	struct {
+		int family;
+		ldi_handle_t handle;
+		struct lifconf lifc;
+		struct lifnum lifn;
+	} families[] = {
+		{ AF_INET, sock->lxns_iphandle },
+		{ AF_INET6, sock->lxns_ip6handle },
+		{ AF_UNSPEC }
+	}, *family, *check;
+
+	for (family = families; family->family != AF_UNSPEC; family++) {
+		struct lifconf *lifc = &family->lifc;
+		struct lifnum *lifn = &family->lifn;
+
+		lifn->lifn_family = family->family;
+		sock->lxns_current = family->handle;
+
+		if (lx_netlink_reply_ioctl(reply, SIOCGLIFNUM, lifn) != 0)
+			break;
+
+		lifc->lifc_family = lifn->lifn_family;
+		lifc->lifc_flags = 0;
+		lifc->lifc_len = lifn->lifn_count * sizeof (struct lifreq);
+		if (lifn->lifn_count == 0) {
+			lifc->lifc_buf = NULL;
+			continue;
+		}
+		lifc->lifc_buf = kmem_alloc(lifc->lifc_len, KM_SLEEP);
+
+		if (lx_netlink_reply_ioctl(reply, SIOCGLIFCONF, lifc) != 0)
+			break;
+
+		nlifr = lifc->lifc_len / sizeof (lifc->lifc_req[0]);
+
+		for (i = 0; i < nlifr; i++) {
+			if (!distinct) {
+				func(reply, &lifc->lifc_req[i]);
+				continue;
+			}
+
+			/*
+			 * If we have been asked to provide each interface
+			 * exactly once, we need to (annoyingly) check this
+			 * name against others that we've already processed for
+			 * other families.  Yes, this is quadratic time -- but
+			 * the number of interfaces per family is expected to
+			 * be very small.
+			 */
+			for (check = families; check != family; check++) {
+				struct lifconf *clifc = &check->lifc;
+				int cnlifr = clifc->lifc_len /
+				    sizeof (clifc->lifc_req[0]), j;
+				char *nm = lifc->lifc_req[i].lifr_name, *cnm;
+
+				for (j = 0; j < cnlifr; j++) {
+					cnm = clifc->lifc_req[j].lifr_name;
+
+					if (strcmp(nm, cnm) == 0)
+						break;
+				}
+
+				if (j != cnlifr)
+					break;
+			}
+
+			if (check != family)
+				continue;
+
+			func(reply, &lifc->lifc_req[i]);
+		}
+	}
+
+	for (family = families; family->family != AF_UNSPEC; family++) {
+		struct lifconf *lifc = &family->lifc;
+
+		if (lifc->lifc_buf != NULL)
+			kmem_free(lifc->lifc_buf, lifc->lifc_len);
+	}
+}
+
+/*ARGSUSED*/
+static int
+lx_netlink_getlink(lx_netlink_sock_t *lxsock, lx_netlink_hdr_t *hdr, mblk_t *mp)
+{
+	lx_netlink_reply_t *reply;
+
+	reply = lx_netlink_reply(lxsock, hdr, LX_NETLINK_RTM_NEWLINK);
+
+	if (reply == NULL)
+		return (ENOMEM);
+
+	lx_netlink_reply_eachfamily(reply, lx_netlink_getlink_lifreq, B_TRUE);
+	lx_netlink_reply_done(reply);
+
+	return (0);
+}
+
+static void
+lx_netlink_getaddr_lifreq(lx_netlink_reply_t *reply, struct lifreq *lifr)
+{
+	lx_netlink_ifaddrmsg_t ifa;
+
+	bzero(&ifa, sizeof (ifa));
+
+	if (lx_netlink_reply_ioctl(reply, SIOCGLIFINDEX, lifr) != 0)
+		return;
+
+	ifa.lxnl_ifa_index = lifr->lifr_index;
+
+	if (lx_netlink_reply_ioctl(reply, SIOCGLIFFLAGS, lifr) != 0)
+		return;
+
+	if (!(lifr->lifr_flags & IFF_NOLOCAL)) {
+		if (lx_netlink_reply_ioctl(reply, SIOCGLIFSUBNET, lifr) != 0)
+			return;
+
+		ifa.lxnl_ifa_prefixlen = lifr->lifr_addrlen;
+
+		if (lx_netlink_reply_ioctl(reply, SIOCGLIFADDR, lifr) != 0)
+			return;
+
+		if (lifr->lifr_addr.ss_family == AF_INET) {
+			struct sockaddr_in *sin;
+
+			ifa.lxnl_ifa_family = LX_AF_INET;
+
+			lx_netlink_reply_msg(reply, &ifa,
+			    sizeof (lx_netlink_ifaddrmsg_t));
+
+			sin = (struct sockaddr_in *)&lifr->lifr_addr;
+
+			lx_netlink_reply_attr_int32(reply,
+			    LX_NETLINK_IFA_ADDRESS, sin->sin_addr.s_addr);
+		} else {
+			struct sockaddr_in6 *sin;
+
+			ifa.lxnl_ifa_family = LX_AF_INET6;
+
+			lx_netlink_reply_msg(reply, &ifa,
+			    sizeof (lx_netlink_ifaddrmsg_t));
+
+			sin = (struct sockaddr_in6 *)&lifr->lifr_addr;
+
+			lx_netlink_reply_attr(reply, LX_NETLINK_IFA_ADDRESS,
+			    &sin->sin6_addr, sizeof (sin->sin6_addr));
+		}
+	}
+
+	lx_netlink_reply_send(reply);
+}
+
+/*ARGSUSED*/
+static int
+lx_netlink_getaddr(lx_netlink_sock_t *lxsock, lx_netlink_hdr_t *hdr, mblk_t *mp)
+{
+	lx_netlink_reply_t *reply;
+
+	reply = lx_netlink_reply(lxsock, hdr, LX_NETLINK_RTM_NEWADDR);
+
+	if (reply == NULL)
+		return (ENOMEM);
+
+	lx_netlink_reply_eachfamily(reply, lx_netlink_getaddr_lifreq, B_FALSE);
+	lx_netlink_reply_done(reply);
+
+	return (0);
+}
+
+/*ARGSUSED*/
+static int
+lx_netlink_audit(lx_netlink_sock_t *lxsock, lx_netlink_hdr_t *hdr, mblk_t *mp)
+{
+	/*
+	 * For all auditing messages, we return ECONNREFUSED, which seems to
+	 * keep user-level auditing happy.  (Or at least, non-suicidal.)
+	 */
+	return (ECONNREFUSED);
+}
+
+/*ARGSUSED*/
+static int
+lx_netlink_kobject_uevent(lx_netlink_sock_t *lxsock,
+    lx_netlink_hdr_t *hdr, mblk_t *mp)
+{
+	/*
+	 * For udev, we just silently accept all writes and never actually
+	 * reply with anything -- which appears to be sufficient for things
+	 * to work.
+	 */
+	return (0);
+}
+
+/*ARGSUSED*/
+static int
+lx_netlink_send(sock_lower_handle_t handle, mblk_t *mp,
+    struct nmsghdr *msg, cred_t *cr)
+{
+	lx_netlink_sock_t *lxsock = (lx_netlink_sock_t *)handle;
+	lx_netlink_hdr_t *hdr = (lx_netlink_hdr_t *)mp->b_rptr;
+	int i, rval;
+
+	static struct {
+		int proto;
+		uint16_t type;
+		int (*func)(lx_netlink_sock_t *, lx_netlink_hdr_t *, mblk_t *);
+	} handlers[] = {
+		{ LX_NETLINK_ROUTE,
+		    LX_NETLINK_RTM_GETLINK, lx_netlink_getlink },
+		{ LX_NETLINK_ROUTE,
+		    LX_NETLINK_RTM_GETADDR, lx_netlink_getaddr },
+		{ LX_NETLINK_AUDIT,
+		    LX_NETLINK_NLMSG_NONE, lx_netlink_audit },
+		{ LX_NETLINK_KOBJECT_UEVENT,
+		    LX_NETLINK_NLMSG_NONE, lx_netlink_kobject_uevent },
+		{ LX_NETLINK_NLMSG_NOOP, LX_NETLINK_NLMSG_NONE, NULL }
+	};
+
+	if (DB_TYPE(mp) != M_DATA || MBLKL(mp) < sizeof (lx_netlink_hdr_t)) {
+		freemsg(mp);
+		return (EPROTO);
+	}
+
+	for (i = 0; handlers[i].func != NULL; i++) {
+		if (lxsock->lxns_proto != handlers[i].proto)
+			continue;
+
+		if (handlers[i].type != LX_NETLINK_NLMSG_NONE &&
+		    hdr->lxnh_type != handlers[i].type)
+			continue;
+
+		rval = handlers[i].func(lxsock, hdr, mp);
+		freemsg(mp);
+
+		return (rval);
+	}
+
+	/*
+	 * An unrecognized message.  We will bounce up an EOPNOTSUPP reply.
+	 */
+	rval = lx_netlink_reply_error(lxsock, hdr, EOPNOTSUPP);
+	freemsg(mp);
+
+	return (rval);
+}
+
+/*ARGSUSED*/
+static int
+lx_netlink_close(sock_lower_handle_t handle, int flags, cred_t *cr)
+{
+	lx_netlink_sock_t *lxsock = (lx_netlink_sock_t *)handle, *sock, **prev;
+
+	mutex_enter(&lx_netlink_lock);
+
+	prev = &lx_netlink_head;
+
+	for (sock = *prev; sock != lxsock; sock = sock->lxns_next)
+		prev = &sock->lxns_next;
+
+	*prev = sock->lxns_next;
+
+	mutex_exit(&lx_netlink_lock);
+
+	(void) ldi_close(lxsock->lxns_iphandle, FREAD, kcred);
+	(void) ldi_close(lxsock->lxns_ip6handle, FREAD, kcred);
+	kmem_free(lxsock, sizeof (lx_netlink_sock_t));
+
+	return (0);
+}
+
+static sock_downcalls_t sock_lx_netlink_downcalls = {
+	lx_netlink_activate,		/* sd_activate */
+	sock_accept_notsupp,		/* sd_accept */
+	lx_netlink_bind,		/* sd_bind */
+	sock_listen_notsupp,		/* sd_listen */
+	sock_connect_notsupp,		/* sd_connect */
+	sock_getpeername_notsupp,	/* sd_getpeername */
+	lx_netlink_getsockname,		/* sd_getsockname */
+	sock_getsockopt_notsupp,	/* sd_getsockopt */
+	lx_netlink_setsockopt,		/* sd_setsockopt */
+	lx_netlink_send,		/* sd_send */
+	NULL,				/* sd_send_uio */
+	NULL,				/* sd_recv_uio */
+	NULL,				/* sd_poll */
+	sock_shutdown_notsupp,		/* sd_shutdown */
+	sock_clr_flowctrl_notsupp,	/* sd_setflowctrl */
+	sock_ioctl_notsupp,		/* sd_ioctl */
+	lx_netlink_close		/* sd_close */
+};
+
+/*ARGSUSED*/
+static sock_lower_handle_t
+lx_netlink_create(int family, int type, int proto,
+    sock_downcalls_t **sock_downcalls, uint_t *smodep, int *errorp,
+    int flags, cred_t *credp)
+{
+	lx_netlink_sock_t *lxsock;
+	ldi_handle_t handle, handle6;
+	cred_t *kcred = zone_kcred();
+	int err;
+
+	if (family != AF_LX_NETLINK ||
+	    (type != SOCK_DGRAM && type != SOCK_RAW)) {
+		*errorp = EPROTONOSUPPORT;
+		return (NULL);
+	}
+
+	switch (proto) {
+	case LX_NETLINK_ROUTE:
+	case LX_NETLINK_AUDIT:
+	case LX_NETLINK_KOBJECT_UEVENT:
+		break;
+
+	default:
+		*errorp = EPROTONOSUPPORT;
+		return (NULL);
+	}
+
+	if ((err = ldi_open_by_name(DEV_IP, FREAD, kcred,
+	    &handle, lx_netlink_ldi)) != 0) {
+		*errorp = err;
+		return (NULL);
+	}
+
+	if ((err = ldi_open_by_name(DEV_IP6, FREAD, kcred,
+	    &handle6, lx_netlink_ldi)) != 0) {
+		(void) ldi_close(handle, FREAD, kcred);
+		*errorp = err;
+		return (NULL);
+	}
+
+	*sock_downcalls = &sock_lx_netlink_downcalls;
+	*smodep = SM_ATOMIC;
+
+	lxsock = kmem_zalloc(sizeof (lx_netlink_sock_t), KM_SLEEP);
+	lxsock->lxns_iphandle = handle;
+	lxsock->lxns_ip6handle = handle6;
+	lxsock->lxns_bufsize = lx_netlink_bufsize;
+	lxsock->lxns_proto = proto;
+
+	mutex_enter(&lx_netlink_lock);
+
+	lxsock->lxns_next = lx_netlink_head;
+	lx_netlink_head = lxsock;
+
+	mutex_exit(&lx_netlink_lock);
+
+	return ((sock_lower_handle_t)lxsock);
+}
+
+static void
+lx_netlink_init(void)
+{
+	major_t major = mod_name_to_major("ip");
+	int err;
+
+	VERIFY(major != DDI_MAJOR_T_NONE);
+
+	err = ldi_ident_from_major(major, &lx_netlink_ldi);
+	VERIFY(err == 0);
+}
+
+static void
+lx_netlink_fini(void)
+{
+	ldi_ident_release(lx_netlink_ldi);
+}
+
+static smod_reg_t sinfo = {
+	SOCKMOD_VERSION,
+	"lx_netlink",
+	SOCK_UC_VERSION,
+	SOCK_DC_VERSION,
+	lx_netlink_create,
+	NULL
+};
+
+/* modldrv structure */
+static struct modlsockmod sockmod = {
+	&mod_sockmodops, "AF_LX_NETLINK socket module", &sinfo
+};
+
+/* modlinkage structure */
+static struct modlinkage ml = {
+	MODREV_1,
+	&sockmod,
+	NULL
+};
+
+int
+_init(void)
+{
+	int err;
+
+	lx_netlink_init();
+
+	if ((err = mod_install(&ml)) != 0)
+		lx_netlink_fini();
+
+	return (err);
+}
+
+int
+_info(struct modinfo *modinfop)
+{
+	return (mod_info(&ml, modinfop));
+}
+
+int
+_fini(void)
+{
+	int err = 0;
+
+	mutex_enter(&lx_netlink_lock);
+
+	if (lx_netlink_head != NULL)
+		err = EBUSY;
+
+	mutex_exit(&lx_netlink_lock);
+
+	if (err == 0 && (err = mod_remove(&ml)) == 0)
+		lx_netlink_fini();
+
+	return (err);
+}
diff --git a/usr/src/uts/common/brand/lx/io/lx_ptm.c b/usr/src/uts/common/brand/lx/io/lx_ptm.c
new file mode 100644
index 0000000000..db13b00ea4
--- /dev/null
+++ b/usr/src/uts/common/brand/lx/io/lx_ptm.c
@@ -0,0 +1,1165 @@
+/*
+ * CDDL HEADER START
+ *
+ * The contents of this file are subject to the terms of the
+ * Common Development and Distribution License (the "License").
+ * You may not use this file except in compliance with the License.
+ *
+ * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
+ * or http://www.opensolaris.org/os/licensing.
+ * See the License for the specific language governing permissions
+ * and limitations under the License.
+ *
+ * When distributing Covered Code, include this CDDL HEADER in each
+ * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
+ * If applicable, add the following below this CDDL HEADER, with the
+ * fields enclosed by brackets "[]" replaced with your own identifying
+ * information: Portions Copyright [yyyy] [name of copyright owner]
+ *
+ * CDDL HEADER END
+ */
+/*
+ * Copyright 2008 Sun Microsystems, Inc.  All rights reserved.
+ * Use is subject to license terms.
+ */
+/*
+ * Copyright 2014 Joyent, Inc.  All rights reserved.
+ */
+
+
+/*
+ * This driver attempts to emulate some of the the behaviors of
+ * Linux terminal devices (/dev/ptmx and /dev/pts/[0-9][0-9]*) on Solaris
+ *
+ * It does this by layering over the /dev/ptmx device and intercepting
+ * opens to it.
+ *
+ * This driver makes the following assumptions about the way the ptm/pts
+ * drivers on Solaris work:
+ *
+ *    - all opens of the /dev/ptmx device node return a unique dev_t.
+ *
+ *    - the dev_t minor node value for each open ptm instance corrospondes
+ *      to it's associated slave terminal device number.  ie. the path to
+ *      the slave terminal device associated with an open ptm instance
+ *      who's dev_t minor node vaue is 5, is /dev/pts/5.
+ *
+ *    - the ptm driver always allocates the lowest numbered slave terminal
+ *      device possible.
+ */
+
+#include <sys/conf.h>
+#include <sys/ddi.h>
+#include <sys/devops.h>
+#include <sys/file.h>
+#include <sys/filio.h>
+#include <sys/kstr.h>
+#include <sys/lx_ptm.h>
+#include <sys/modctl.h>
+#include <sys/pathname.h>
+#include <sys/ptms.h>
+#include <sys/ptyvar.h>
+#include <sys/stat.h>
+#include <sys/stropts.h>
+#include <sys/sunddi.h>
+#include <sys/sunldi.h>
+#include <sys/sysmacros.h>
+#include <sys/types.h>
+
+#define	LP_PTM_PATH		"/dev/ptmx"
+#define	LP_PTS_PATH		"/dev/pts/"
+#define	LP_PTS_DRV_NAME		"pts"
+#define	LP_PTS_USEC_DELAY	(5 * 1000)	/* 5 ms */
+#define	LP_PTS_USEC_DELAY_MAX	(5 * MILLISEC)	/* 5 ms */
+
+/*
+ * this driver is layered on top of the ptm driver.  we'd like to
+ * make this drivers minor name space a mirror of the ptm drivers
+ * namespace, but we can't actually do this.  the reason is that the
+ * ptm driver is opened via the clone driver.  there for no minor nodes
+ * of the ptm driver are actually accessible via the filesystem.
+ * since we're not a streams device we can't be opened by the clone
+ * driver.  there for we need to have at least minor node accessible
+ * via the filesystem so that consumers can open it.  we use the device
+ * node with a minor number of 0 for this purpose.  what this means is
+ * that minor node 0 can't be used to map ptm minor node 0.  since this
+ * minor node is now reserved we need to shift our ptm minor node
+ * mappings by one.  ie. a ptm minor node with a value of 0 will
+ * corrospond to our minor node with a value of 1.  these mappings are
+ * managed with the following macros.
+ */
+#define	DEVT_TO_INDEX(x)	LX_PTM_DEV_TO_PTS(x)
+#define	INDEX_TO_MINOR(x)	((x) + 1)
+
+/*
+ * grow our layered handle array by the same size increment that the ptm
+ * driver uses to grow the pty device space - PTY_MAXDELTA
+ */
+#define	LP_PTY_INC	128
+
+/*
+ * lx_ptm_ops contains state information about outstanding operations on the
+ * underlying master terminal device.  Currently we only track information
+ * for read operations.
+ *
+ * Note that this data has not been rolled directly into the lx_ptm_handle
+ * structure because we can't put mutex's of condition variables into
+ * lx_ptm_handle structure.  The reason is that the array of lx_ptm_handle
+ * structures linked to from the global lx_ptm state can be resized
+ * dynamically, and when it's resized, the new array is at a different
+ * memory location and the old array memory is discarded.  Mutexs and cvs
+ * are accessed based off their address, so if this array was re-sized while
+ * there were outstanding operations on any mutexs or cvs in the array
+ * then the system would tip over.  In the future the lx_ptm_handle structure
+ * array should probably be replaced with either an array of pointers to
+ * lx_ptm_handle structures or some other kind of data structure containing
+ * pointers to lx_ptm_handle structures.  Then the lx_ptm_ops structure
+ * could be folded directly into the lx_ptm_handle structures.  (This will
+ * also require the definition of a new locking mechanism to protect the
+ * contents of lx_ptm_handle structures.)
+ */
+typedef struct lx_ptm_ops {
+	int			lpo_rops;
+	kcondvar_t		lpo_rops_cv;
+	kmutex_t		lpo_rops_lock;
+} lx_ptm_ops_t;
+
+/*
+ * Every open of the master terminal device in a zone results in a new
+ * lx_ptm_handle handle allocation.  These handles are stored in an array
+ * hanging off the lx_ptm_state structure.
+ */
+typedef struct lx_ptm_handle {
+	/* Device handle to the underlying real /dev/ptmx master terminal. */
+	ldi_handle_t		lph_handle;
+
+	/* Flag to indicate if TIOCPKT mode has been enabled. */
+	int			lph_pktio;
+
+	/* Number of times the slave device has been opened/closed. */
+	int			lph_eofed;
+
+	/* Callback handler in the ptm driver to check if slave is open. */
+	ptmptsopencb_t		lph_ppocb;
+
+	/* Pointer to state for operations on underlying device. */
+	lx_ptm_ops_t		*lph_lpo;
+} lx_ptm_handle_t;
+
+/*
+ * Global state for the lx_ptm driver.
+ */
+typedef struct lx_ptm_state {
+	/* lx_ptm device devinfo pointer */
+	dev_info_t		*lps_dip;
+
+	/* LDI ident used to open underlying real /dev/ptmx master terminals. */
+	ldi_ident_t		lps_li;
+
+	/* pts drivers major number */
+	major_t			lps_pts_major;
+
+	/* rw lock used to manage access and growth of lps_lh_array */
+	krwlock_t		lps_lh_rwlock;
+
+	/* number of elements in lps_lh_array */
+	uint_t			lps_lh_count;
+
+	/* Array of handles to underlying real /dev/ptmx master terminals. */
+	lx_ptm_handle_t		*lps_lh_array;
+} lx_ptm_state_t;
+
+/* Pointer to the lx_ptm global state structure. */
+static lx_ptm_state_t	lps;
+
+/*
+ * List of modules to be autopushed onto slave terminal devices when they
+ * are opened in an lx branded zone.
+ */
+static char *lx_pts_mods[] = {
+	"ptem",
+	"ldterm",
+	"ttcompat",
+	NULL
+};
+
+static void
+lx_ptm_lh_grow(uint_t index)
+{
+	uint_t			new_lh_count, old_lh_count;
+	lx_ptm_handle_t		*new_lh_array, *old_lh_array;
+
+	/*
+	 * allocate a new array.  we drop the rw lock on the array so that
+	 * readers can still access devices in case our memory allocation
+	 * blocks.
+	 */
+	new_lh_count = MAX(lps.lps_lh_count + LP_PTY_INC, index + 1);
+	new_lh_array =
+	    kmem_zalloc(sizeof (lx_ptm_handle_t) * new_lh_count, KM_SLEEP);
+
+	/*
+	 * double check that we still actually need to increase the size
+	 * of the array
+	 */
+	rw_enter(&lps.lps_lh_rwlock, RW_WRITER);
+	if (index < lps.lps_lh_count) {
+		/* someone beat us to it so there's nothing more to do */
+		rw_exit(&lps.lps_lh_rwlock);
+		kmem_free(new_lh_array,
+		    sizeof (lx_ptm_handle_t) * new_lh_count);
+		return;
+	}
+
+	/* copy the existing data into the new array */
+	ASSERT((lps.lps_lh_count != 0) || (lps.lps_lh_array == NULL));
+	ASSERT((lps.lps_lh_count == 0) || (lps.lps_lh_array != NULL));
+	if (lps.lps_lh_count != 0) {
+		bcopy(lps.lps_lh_array, new_lh_array,
+		    sizeof (lx_ptm_handle_t) * lps.lps_lh_count);
+	}
+
+	/* save info on the old array */
+	old_lh_array = lps.lps_lh_array;
+	old_lh_count = lps.lps_lh_count;
+
+	/* install the new array */
+	lps.lps_lh_array = new_lh_array;
+	lps.lps_lh_count = new_lh_count;
+
+	rw_exit(&lps.lps_lh_rwlock);
+
+	/* free the old array */
+	if (old_lh_array != NULL) {
+		kmem_free(old_lh_array,
+		    sizeof (lx_ptm_handle_t) * old_lh_count);
+	}
+}
+
+static void
+lx_ptm_lh_insert(uint_t index, ldi_handle_t lh)
+{
+	lx_ptm_ops_t *lpo;
+
+	ASSERT(lh != NULL);
+
+	/* Allocate and initialize the ops structure */
+	lpo = kmem_zalloc(sizeof (lx_ptm_ops_t), KM_SLEEP);
+	mutex_init(&lpo->lpo_rops_lock, NULL, MUTEX_DEFAULT, NULL);
+	cv_init(&lpo->lpo_rops_cv, NULL, CV_DEFAULT, NULL);
+
+	rw_enter(&lps.lps_lh_rwlock, RW_WRITER);
+
+	/* check if we need to grow the size of the layered handle array */
+	if (index >= lps.lps_lh_count) {
+		rw_exit(&lps.lps_lh_rwlock);
+		lx_ptm_lh_grow(index);
+		rw_enter(&lps.lps_lh_rwlock, RW_WRITER);
+	}
+
+	ASSERT(index < lps.lps_lh_count);
+	ASSERT(lps.lps_lh_array[index].lph_handle == NULL);
+	ASSERT(lps.lps_lh_array[index].lph_pktio == 0);
+	ASSERT(lps.lps_lh_array[index].lph_eofed == 0);
+	ASSERT(lps.lps_lh_array[index].lph_lpo == NULL);
+
+	/* insert the new handle and return */
+	lps.lps_lh_array[index].lph_handle = lh;
+	lps.lps_lh_array[index].lph_pktio = 0;
+	lps.lps_lh_array[index].lph_eofed = 0;
+	lps.lps_lh_array[index].lph_lpo = lpo;
+
+	rw_exit(&lps.lps_lh_rwlock);
+}
+
+static ldi_handle_t
+lx_ptm_lh_remove(uint_t index)
+{
+	ldi_handle_t	lh;
+
+	rw_enter(&lps.lps_lh_rwlock, RW_WRITER);
+
+	ASSERT(index < lps.lps_lh_count);
+	ASSERT(lps.lps_lh_array[index].lph_handle != NULL);
+	ASSERT(lps.lps_lh_array[index].lph_lpo->lpo_rops == 0);
+	ASSERT(!MUTEX_HELD(&lps.lps_lh_array[index].lph_lpo->lpo_rops_lock));
+
+	/* free the write handle */
+	kmem_free(lps.lps_lh_array[index].lph_lpo, sizeof (lx_ptm_ops_t));
+	lps.lps_lh_array[index].lph_lpo = NULL;
+
+	/* remove the handle and return it */
+	lh = lps.lps_lh_array[index].lph_handle;
+	lps.lps_lh_array[index].lph_handle = NULL;
+	lps.lps_lh_array[index].lph_pktio = 0;
+	lps.lps_lh_array[index].lph_eofed = 0;
+	rw_exit(&lps.lps_lh_rwlock);
+	return (lh);
+}
+
+static void
+lx_ptm_lh_get_ppocb(uint_t index, ptmptsopencb_t *ppocb)
+{
+	rw_enter(&lps.lps_lh_rwlock, RW_WRITER);
+
+	ASSERT(index < lps.lps_lh_count);
+	ASSERT(lps.lps_lh_array[index].lph_handle != NULL);
+
+	*ppocb = lps.lps_lh_array[index].lph_ppocb;
+	rw_exit(&lps.lps_lh_rwlock);
+}
+
+static void
+lx_ptm_lh_set_ppocb(uint_t index, ptmptsopencb_t *ppocb)
+{
+	rw_enter(&lps.lps_lh_rwlock, RW_WRITER);
+
+	ASSERT(index < lps.lps_lh_count);
+	ASSERT(lps.lps_lh_array[index].lph_handle != NULL);
+
+	lps.lps_lh_array[index].lph_ppocb = *ppocb;
+	rw_exit(&lps.lps_lh_rwlock);
+}
+
+static ldi_handle_t
+lx_ptm_lh_lookup(uint_t index)
+{
+	ldi_handle_t	lh;
+
+	rw_enter(&lps.lps_lh_rwlock, RW_READER);
+
+	ASSERT(index < lps.lps_lh_count);
+	ASSERT(lps.lps_lh_array[index].lph_handle != NULL);
+
+	/* return the handle */
+	lh = lps.lps_lh_array[index].lph_handle;
+	rw_exit(&lps.lps_lh_rwlock);
+	return (lh);
+}
+
+static lx_ptm_ops_t *
+lx_ptm_lpo_lookup(uint_t index)
+{
+	lx_ptm_ops_t	*lpo;
+
+	rw_enter(&lps.lps_lh_rwlock, RW_READER);
+
+	ASSERT(index < lps.lps_lh_count);
+	ASSERT(lps.lps_lh_array[index].lph_lpo != NULL);
+
+	/* return the handle */
+	lpo = lps.lps_lh_array[index].lph_lpo;
+	rw_exit(&lps.lps_lh_rwlock);
+	return (lpo);
+}
+
+static int
+lx_ptm_lh_pktio_get(uint_t index)
+{
+	int		pktio;
+
+	rw_enter(&lps.lps_lh_rwlock, RW_READER);
+
+	ASSERT(index < lps.lps_lh_count);
+	ASSERT(lps.lps_lh_array[index].lph_handle != NULL);
+
+	/* return the pktio state */
+	pktio = lps.lps_lh_array[index].lph_pktio;
+	rw_exit(&lps.lps_lh_rwlock);
+	return (pktio);
+}
+
+static void
+lx_ptm_lh_pktio_set(uint_t index, int pktio)
+{
+	rw_enter(&lps.lps_lh_rwlock, RW_WRITER);
+
+	ASSERT(index < lps.lps_lh_count);
+	ASSERT(lps.lps_lh_array[index].lph_handle != NULL);
+
+	/* set the pktio state */
+	lps.lps_lh_array[index].lph_pktio = pktio;
+	rw_exit(&lps.lps_lh_rwlock);
+}
+
+static int
+lx_ptm_lh_eofed_get(uint_t index)
+{
+	int		eofed;
+
+	rw_enter(&lps.lps_lh_rwlock, RW_READER);
+
+	ASSERT(index < lps.lps_lh_count);
+	ASSERT(lps.lps_lh_array[index].lph_handle != NULL);
+
+	/* return the eofed state */
+	eofed = lps.lps_lh_array[index].lph_eofed;
+	rw_exit(&lps.lps_lh_rwlock);
+	return (eofed);
+}
+
+static void
+lx_ptm_lh_eofed_set(uint_t index)
+{
+	rw_enter(&lps.lps_lh_rwlock, RW_WRITER);
+
+	ASSERT(index < lps.lps_lh_count);
+	ASSERT(lps.lps_lh_array[index].lph_handle != NULL);
+
+	/* set the eofed state */
+	lps.lps_lh_array[index].lph_eofed++;
+	rw_exit(&lps.lps_lh_rwlock);
+}
+
+static int
+lx_ptm_read_start(dev_t dev)
+{
+	lx_ptm_ops_t	*lpo = lx_ptm_lpo_lookup(DEVT_TO_INDEX(dev));
+
+	mutex_enter(&lpo->lpo_rops_lock);
+	ASSERT(lpo->lpo_rops >= 0);
+
+	/* Wait for other read operations to finish */
+	while (lpo->lpo_rops != 0) {
+		if (cv_wait_sig(&lpo->lpo_rops_cv, &lpo->lpo_rops_lock) == 0) {
+			mutex_exit(&lpo->lpo_rops_lock);
+			return (-1);
+		}
+	}
+
+	/* Start a read operation */
+	VERIFY(++lpo->lpo_rops == 1);
+	mutex_exit(&lpo->lpo_rops_lock);
+	return (0);
+}
+
+static void
+lx_ptm_read_end(dev_t dev)
+{
+	lx_ptm_ops_t	*lpo = lx_ptm_lpo_lookup(DEVT_TO_INDEX(dev));
+
+	mutex_enter(&lpo->lpo_rops_lock);
+	ASSERT(lpo->lpo_rops >= 0);
+
+	/* End a read operation */
+	VERIFY(--lpo->lpo_rops == 0);
+	cv_signal(&lpo->lpo_rops_cv);
+
+	mutex_exit(&lpo->lpo_rops_lock);
+}
+
+static int
+lx_ptm_pts_isopen(dev_t dev)
+{
+	ptmptsopencb_t	ppocb;
+
+	lx_ptm_lh_get_ppocb(DEVT_TO_INDEX(dev), &ppocb);
+	return (ppocb.ppocb_func(ppocb.ppocb_arg));
+}
+
+static void
+lx_ptm_eof_read(ldi_handle_t lh)
+{
+	struct uio	uio;
+	iovec_t		iov;
+	char		junk[1];
+
+	/*
+	 * We can remove any EOF message from the head of the stream by
+	 * doing a zero byte read from the stream.
+	 */
+	iov.iov_len = 0;
+	iov.iov_base = junk;
+	uio.uio_iovcnt = 1;
+	uio.uio_iov = &iov;
+	uio.uio_resid = iov.iov_len;
+	uio.uio_offset = 0;
+	uio.uio_segflg = UIO_SYSSPACE;
+	uio.uio_fmode = 0;
+	uio.uio_extflg = 0;
+	uio.uio_llimit = MAXOFFSET_T;
+	(void) ldi_read(lh, &uio, kcred);
+}
+
+static int
+lx_ptm_eof_drop_1(dev_t dev, int *rvalp)
+{
+	ldi_handle_t	lh = lx_ptm_lh_lookup(DEVT_TO_INDEX(dev));
+	int		err, msg_size, msg_count;
+
+	*rvalp = 0;
+
+	/*
+	 * Check if there is an EOF message (represented by a zero length
+	 * data message) at the head of the stream.  Note that the
+	 * I_NREAD ioctl is a streams framework ioctl so it will succeed
+	 * even if there have been previous write errors on this stream.
+	 */
+	if ((err = ldi_ioctl(lh, I_NREAD, (intptr_t)&msg_size,
+	    FKIOCTL, kcred, &msg_count)) != 0)
+		return (err);
+
+	if ((msg_count == 0) || (msg_size != 0)) {
+		/* No EOF message found */
+		return (0);
+	}
+
+	/* Record the fact that the slave device has been closed. */
+	lx_ptm_lh_eofed_set(DEVT_TO_INDEX(dev));
+
+	/* drop the EOF */
+	lx_ptm_eof_read(lh);
+	*rvalp = 1;
+	return (0);
+}
+
+static int
+lx_ptm_eof_drop(dev_t dev, int *rvalp)
+{
+	int rval, err;
+
+	if (rvalp != NULL)
+		*rvalp = 0;
+	for (;;) {
+		if ((err = lx_ptm_eof_drop_1(dev, &rval)) != 0)
+			return (err);
+		if (rval == 0)
+			return (0);
+		if (rvalp != NULL)
+			*rvalp = 1;
+	}
+}
+
+static int
+lx_ptm_data_check(dev_t dev, int ignore_eof, int *rvalp)
+{
+	ldi_handle_t	lh = lx_ptm_lh_lookup(DEVT_TO_INDEX(dev));
+	int		err;
+
+	*rvalp = 0;
+	if (ignore_eof) {
+		int	size, rval;
+
+		if ((err = ldi_ioctl(lh, FIONREAD, (intptr_t)&size,
+		    FKIOCTL, kcred, &rval)) != 0)
+			return (err);
+		if (size != 0)
+			*rvalp = 1;
+	} else {
+		int	msg_size, msg_count;
+
+		if ((err = ldi_ioctl(lh, I_NREAD, (intptr_t)&msg_size,
+		    FKIOCTL, kcred, &msg_count)) != 0)
+			return (err);
+		if (msg_count != 0)
+			*rvalp = 1;
+	}
+	return (0);
+}
+
+static int
+lx_ptm_attach(dev_info_t *dip, ddi_attach_cmd_t cmd)
+{
+	int err;
+
+	if (cmd != DDI_ATTACH)
+		return (DDI_FAILURE);
+
+	if (ddi_create_minor_node(dip, LX_PTM_MINOR_NODE, S_IFCHR,
+	    ddi_get_instance(dip), DDI_PSEUDO, 0) != DDI_SUCCESS)
+		return (DDI_FAILURE);
+
+	err = ldi_ident_from_dip(dip, &lps.lps_li);
+	if (err != 0) {
+		ddi_remove_minor_node(dip, ddi_get_name(dip));
+		return (DDI_FAILURE);
+	}
+
+	lps.lps_dip = dip;
+	lps.lps_pts_major = ddi_name_to_major(LP_PTS_DRV_NAME);
+
+	rw_init(&lps.lps_lh_rwlock, NULL, RW_DRIVER, NULL);
+	lps.lps_lh_count = 0;
+	lps.lps_lh_array = NULL;
+
+	return (DDI_SUCCESS);
+}
+
+/*ARGSUSED*/
+static int
+lx_ptm_detach(dev_info_t *dip, ddi_detach_cmd_t cmd)
+{
+	if (cmd != DDI_DETACH)
+		return (DDI_FAILURE);
+
+	ldi_ident_release(lps.lps_li);
+	lps.lps_dip = NULL;
+
+	ASSERT((lps.lps_lh_count != 0) || (lps.lps_lh_array == NULL));
+	ASSERT((lps.lps_lh_count == 0) || (lps.lps_lh_array != NULL));
+	if (lps.lps_lh_array != NULL) {
+		kmem_free(lps.lps_lh_array,
+		    sizeof (lx_ptm_handle_t) * lps.lps_lh_count);
+		lps.lps_lh_array = NULL;
+		lps.lps_lh_count = 0;
+	}
+
+	return (DDI_SUCCESS);
+}
+
+/*ARGSUSED*/
+static int
+lx_ptm_open(dev_t *devp, int flag, int otyp, cred_t *credp)
+{
+	struct strioctl	iocb;
+	ptmptsopencb_t	ppocb = { NULL, NULL };
+	ldi_handle_t	lh;
+	major_t		maj, our_major = getmajor(*devp);
+	minor_t		min, lastmin;
+	uint_t		index, anchor = 1;
+	dev_t		ptm_dev;
+	int		err, rval = 0;
+
+	/*
+	 * Don't support the FNDELAY flag and FNONBLOCK until we either
+	 * find a Linux app that opens /dev/ptmx with the O_NDELAY
+	 * or O_NONBLOCK flags explicitly, or until we create test cases
+	 * to determine how reads of master terminal devices opened with
+	 * these flags behave in different situations on Linux.  Supporting
+	 * these flags will involve enhancing our read implementation
+	 * and changing the way it deals with EOF notifications.
+	 */
+	if (flag & (FNDELAY | FNONBLOCK))
+		return (ENOTSUP);
+
+	/*
+	 * we're layered on top of the ptm driver so open that driver
+	 * first.  (note that we're opening /dev/ptmx in the global
+	 * zone, not ourselves in the Linux zone.)
+	 */
+	err = ldi_open_by_name(LP_PTM_PATH, flag, credp, &lh, lps.lps_li);
+	if (err != 0)
+		return (err);
+
+	/* get the devt returned by the ptmx open */
+	err = ldi_get_dev(lh, &ptm_dev);
+	if (err != 0) {
+		(void) ldi_close(lh, flag, credp);
+		return (err);
+	}
+
+	/*
+	 * we're a cloning driver so here's well change the devt that we
+	 * return.  the ptmx is also a cloning driver so we'll just use
+	 * it's minor number as our minor number (it already manages it's
+	 * minor name space so no reason to duplicate the effort.)
+	 */
+	index = getminor(ptm_dev);
+	*devp = makedevice(our_major, INDEX_TO_MINOR(index));
+
+	/* Get a callback function to query if the pts device is open. */
+	iocb.ic_cmd = PTMPTSOPENCB;
+	iocb.ic_timout = 0;
+	iocb.ic_len = sizeof (ppocb);
+	iocb.ic_dp = (char *)&ppocb;
+
+	err = ldi_ioctl(lh, I_STR, (intptr_t)&iocb, FKIOCTL, kcred, &rval);
+	if ((err != 0) || (rval != 0)) {
+		(void) ldi_close(lh, flag, credp);
+		return (EIO); /* XXX return something else here? */
+	}
+	ASSERT(ppocb.ppocb_func != NULL);
+
+	/*
+	 * now setup autopush for the terminal slave device.  this is
+	 * necessary so that when a Linux program opens the device we
+	 * can push required strmod modules onto the stream.  in Solaris
+	 * this is normally done by the application that actually
+	 * allocates the terminal.
+	 */
+	maj = lps.lps_pts_major;
+	min = index;
+	lastmin = 0;
+	err = kstr_autopush(SET_AUTOPUSH, &maj, &min, &lastmin,
+	    &anchor, lx_pts_mods);
+	if (err != 0) {
+		(void) ldi_close(lh, flag, credp);
+		return (EIO); /* XXX return something else here? */
+	}
+
+	/* save off this layered handle for future accesses */
+	lx_ptm_lh_insert(index, lh);
+	lx_ptm_lh_set_ppocb(index, &ppocb);
+	return (0);
+}
+
+/*ARGSUSED*/
+static int
+lx_ptm_close(dev_t dev, int flag, int otyp, cred_t *credp)
+{
+	ldi_handle_t	lh;
+	major_t		maj;
+	minor_t		min, lastmin;
+	uint_t		index;
+	int		err;
+	int		i;
+
+	index = DEVT_TO_INDEX(dev);
+
+	/*
+	 * we must cleanup all the state associated with this major/minor
+	 * terminal pair before actually closing the ptm master device.
+	 * this is required because once the close of the ptm device is
+	 * complete major/minor terminal pair is immediatly available for
+	 * re-use in any zone.
+	 */
+
+	/* free up our saved reference for this layered handle */
+	lh = lx_ptm_lh_remove(index);
+
+	/* unconfigure autopush for the associated terminal slave device */
+	maj = lps.lps_pts_major;
+	min = index;
+	lastmin = 0;
+	for (i = 0; i < 5; i++) {
+		/*
+		 * we loop here because we don't want to release this ptm
+		 * node if autopush can't be disabled on the associated
+		 * slave device because then bad things could happen if
+		 * another brand were to get this terminal allocated
+		 * to them. If we keep failing we eventually drive on so that
+		 * things don't hang.
+		 */
+		err = kstr_autopush(CLR_AUTOPUSH, &maj, &min, &lastmin,
+		    0, NULL);
+		if (err == 0)
+			break;
+
+		cmn_err(CE_WARN, "lx zoneid %d: error %d on kstr_autopush",
+		    getzoneid(), err);
+
+		/* wait one second and try again */
+		delay(drv_usectohz(1000000));
+	}
+
+	err = ldi_close(lh, flag, credp);
+
+	/*
+	 * note that we don't have to bother with changing the permissions
+	 * on the associated slave device here.  the reason is that no one
+	 * can actually open the device untill it's associated master
+	 * device is re-opened, which will result in the permissions on
+	 * it being reset.
+	 */
+	return (err);
+}
+
+static int
+lx_ptm_read_loop(dev_t dev, struct uio *uiop, cred_t *credp, int *loop)
+{
+	ldi_handle_t	lh = lx_ptm_lh_lookup(DEVT_TO_INDEX(dev));
+	int		err, rval;
+	struct uio	uio = *uiop;
+
+	*loop = 0;
+
+	/*
+	 * Here's another way that Linux master terminals behave differently
+	 * from Solaris master terminals.  If you do a read on a Linux
+	 * master terminal (that was opened witout NDELAY and NONBLOCK)
+	 * who's corrosponding slave terminal is currently closed and
+	 * has been opened and closed at least once, Linux return -1 and
+	 * set errno to EIO where as Solaris blocks.
+	 */
+	if (lx_ptm_lh_eofed_get(DEVT_TO_INDEX(dev))) {
+		/* Slave has been opened and closed at least once. */
+		if (lx_ptm_pts_isopen(dev) == 0) {
+			/*
+			 * Slave is closed.  Make sure that data is avaliable
+			 * before attempting a read.
+			 */
+			if ((err = lx_ptm_data_check(dev, 0, &rval)) != 0)
+				return (err);
+
+			/* If there is no data available then return. */
+			if (rval == 0)
+				return (EIO);
+		}
+	}
+
+	/* Actually do the read operation. */
+	if ((err = ldi_read(lh, uiop, credp)) != 0)
+		return (err);
+
+	/* If read returned actual data then return. */
+	if (uio.uio_resid != uiop->uio_resid)
+		return (0);
+
+	/*
+	 * This was a zero byte read (ie, an EOF).  This indicates
+	 * that the slave terinal device has been closed.  Record
+	 * the fact that the slave device has been closed and retry
+	 * the read operation.
+	 */
+	lx_ptm_lh_eofed_set(DEVT_TO_INDEX(dev));
+	*loop = 1;
+	return (0);
+}
+
+static int
+lx_ptm_read(dev_t dev, struct uio *uiop, cred_t *credp)
+{
+	int		pktio = lx_ptm_lh_pktio_get(DEVT_TO_INDEX(dev));
+	int		err, loop;
+	struct uio	uio;
+	struct iovec	iovp;
+
+	ASSERT(uiop->uio_iovcnt > 0);
+
+	/*
+	 * If packet mode has been enabled (via TIOCPKT) we need to pad
+	 * all read requests with a leading byte that indicates any
+	 * relevant control status information.
+	 */
+	if (pktio != 0) {
+		/*
+		 * We'd like to write the control information into
+		 * the current buffer but we can't yet.  We don't
+		 * want to modify userspace memory here only to have
+		 * the read operation fail later.  So instead
+		 * what we'll do here is read one character from the
+		 * beginning of the memory pointed to by the uio
+		 * structure.  This will advance the output pointer
+		 * by one.  Then when the read completes successfully
+		 * we can update the byte that we passed over.  Before
+		 * we do the read make a copy of the current uiop and
+		 * iovec structs so we can write to them later.
+		 */
+		uio = *uiop;
+		iovp = *uiop->uio_iov;
+		uio.uio_iov = &iovp;
+
+		if (uwritec(uiop) == -1)
+			return (EFAULT);
+	}
+
+	do {
+		/*
+		 * Before we actually attempt a read operation we need
+		 * to make sure there's some buffer space to actually
+		 * read in some data.  We do this because if we're in
+		 * pktio mode and the caller only requested one byte,
+		 * then we've already used up that one byte and we
+		 * don't want to pass this read request.  Doing a 0
+		 * byte read (unless there is a problem with the stream
+		 * head) always returns succcess.  Normally when a streams
+		 * read returns 0 bytes we interpret that as an EOF on
+		 * the stream (ie, the slave side has been opened and
+		 * closed) and we ignore it and re-try the read operation.
+		 * So if we pass on a 0 byte read here lx_ptm_read_loop()
+		 * will tell us to loop around and we'll end up in an
+		 * infinite loop.
+		 */
+		if (uiop->uio_resid == 0)
+			break;
+
+		/*
+		 * Serialize all reads.  We need to do this so that we can
+		 * properly emulate the behavior of master terminals on Linux.
+		 * In reality this serializaion should not pose any kind of
+		 * performance problem since it would be very strange to have
+		 * multiple threads trying to read from the same master
+		 * terminal device concurrently.
+		 */
+		if (lx_ptm_read_start(dev) != 0)
+			return (EINTR);
+
+		err = lx_ptm_read_loop(dev, uiop, credp, &loop);
+		lx_ptm_read_end(dev);
+		if (err != 0)
+			return (err);
+	} while (loop != 0);
+
+	if (pktio != 0) {
+		uint8_t		pktio_data = TIOCPKT_DATA;
+
+		/*
+		 * Note that the control status information we
+		 * pass back is faked up in the sense that we
+		 * don't actually report any events, we always
+		 * report a status of 0.
+		 */
+		if (uiomove(&pktio_data, 1, UIO_READ, &uio) != 0)
+			return (EFAULT);
+	}
+
+	return (0);
+}
+
+static int
+lx_ptm_write(dev_t dev, struct uio *uiop, cred_t *credp)
+{
+	ldi_handle_t		lh = lx_ptm_lh_lookup(DEVT_TO_INDEX(dev));
+	int		err;
+
+	err = ldi_write(lh, uiop, credp);
+
+	return (err);
+}
+
+static int
+lx_ptm_ioctl(dev_t dev, int cmd, intptr_t arg, int mode, cred_t *credp,
+    int *rvalp)
+{
+	ldi_handle_t	lh = lx_ptm_lh_lookup(DEVT_TO_INDEX(dev));
+	int		err;
+
+	/*
+	 * here we need to make sure that we never allow the
+	 * I_SETSIG and I_ESETSIG ioctls to pass through.  we
+	 * do this because we can't support them.
+	 *
+	 * the native Solaris ptm device supports these ioctls because
+	 * they are streams framework ioctls and all streams devices
+	 * support them by default.  these ioctls cause the current
+	 * process to be registered with a stream and receive signals
+	 * when certain stream events occur.
+	 *
+	 * a problem arises with cleanup of these registrations
+	 * for layered drivers.
+	 *
+	 * normally the streams framework is notified whenever a
+	 * process closes any reference to a stream and it goes ahead
+	 * and cleans up these registrations.  but actual device drivers
+	 * are not notified when a process performs a close operation
+	 * unless the process is closing the last opened reference to
+	 * the device on the entire system.
+	 *
+	 * so while we could pass these ioctls on and allow processes
+	 * to register for signal delivery, we would never receive
+	 * any notification when those processes exit (or close a
+	 * stream) and we wouldn't be able to unregister them.
+	 *
+	 * luckily these operations are streams specific and Linux
+	 * doesn't support streams devices.  so it doesn't actually
+	 * seem like we need to support these ioctls.  if it turns
+	 * out that we do need to support them for some reason in
+	 * the future, the current driver model will have to be
+	 * enhanced to better support streams device layering.
+	 */
+	if ((cmd == I_SETSIG) || (cmd == I_ESETSIG))
+		return (EINVAL);
+
+	/*
+	 * here we fake up support for TIOCPKT.  Linux applications expect
+	 * /etc/ptmx to support this ioctl, but on Solaris it doesn't.
+	 * (it is supported on older bsd style ptys.)  so we'll fake
+	 * up support for it here.
+	 *
+	 * the reason that this ioctl is emulated here instead of in
+	 * userland is that this ioctl affects the results returned
+	 * from read() operations.  if this ioctl was emulated in
+	 * userland the brand library would need to intercept all
+	 * read operations and check to see if pktio was enabled
+	 * for the fd being read from.  since this ioctl only needs
+	 * to be supported on the ptmx device it makes more sense
+	 * to support it here where we can easily update the results
+	 * returned for read() operations performed on ourselves.
+	 */
+	if (cmd == TIOCPKT) {
+		int	pktio;
+
+		if (ddi_copyin((void *)arg, &pktio, sizeof (pktio),
+		    mode) != DDI_SUCCESS)
+			return (EFAULT);
+
+		if (pktio == 0)
+			lx_ptm_lh_pktio_set(DEVT_TO_INDEX(dev), 0);
+		else
+			lx_ptm_lh_pktio_set(DEVT_TO_INDEX(dev), 1);
+
+		return (0);
+	}
+
+	err = ldi_ioctl(lh, cmd, arg, mode, credp, rvalp);
+
+	return (err);
+}
+
+static int
+lx_ptm_poll_loop(dev_t dev, short events, int anyyet, short *reventsp,
+    struct pollhead **phpp, int *loop)
+{
+	ldi_handle_t	lh = lx_ptm_lh_lookup(DEVT_TO_INDEX(dev));
+	short		reventsp2;
+	int		err, rval;
+
+	*loop = 0;
+
+	/*
+	 * If the slave device has been opened and closed at least
+	 * once and the slave device is currently closed, then poll
+	 * always needs to returns immediatly.
+	 */
+	if ((lx_ptm_lh_eofed_get(DEVT_TO_INDEX(dev)) != 0) &&
+	    (lx_ptm_pts_isopen(dev) == 0)) {
+		/* In this case always return POLLHUP */
+		*reventsp = POLLHUP;
+
+		/*
+		 * Check if there really is data on the stream.
+		 * If so set the correct return flags.
+		 */
+		if ((err = lx_ptm_data_check(dev, 1, &rval)) != 0) {
+			/* Something went wrong. */
+			return (err);
+		}
+		if (rval != 0)
+			*reventsp |= (events & (POLLIN | POLLRDNORM));
+
+		/*
+		 * Is the user checking for writability?  Note that for ptm
+		 * devices Linux seems to ignore the POLLWRBAND write flag.
+		 */
+		if ((events & POLLWRNORM) == 0)
+			return (0);
+
+		/*
+		 * To check if the stream is writable we have to actually
+		 * call poll, but make sure to set anyyet to 1 to prevent
+		 * the streams framework from setting up callbacks.
+		 */
+		if ((err = ldi_poll(lh, POLLWRNORM, 1, &reventsp2, NULL)) != 0)
+			return (err);
+
+		*reventsp |= (reventsp2 & POLLWRNORM);
+	} else {
+		int lockstate;
+
+		/* The slave device is open, do the poll */
+		if ((err = ldi_poll(lh, events, anyyet, reventsp, phpp)) != 0)
+			return (err);
+
+		/*
+		 * Drop any leading EOFs on the stream.
+		 *
+		 * Note that we have to use pollunlock() here to avoid
+		 * recursive mutex enters in the poll framework.  The
+		 * reason is that if there is an EOF message on the stream
+		 * then the act of reading from the queue to remove the
+		 * message can cause the ptm drivers event service
+		 * routine to be invoked, and if there is no open
+		 * slave device then the ptm driver may generate
+		 * error messages and put them on the stream.  This
+		 * in turn will generate a poll event and the poll
+		 * framework will try to invoke any poll callbacks
+		 * associated with the stream.  In the process of
+		 * doing that the poll framework will try to aquire
+		 * locks that we are already holding.  So we need to
+		 * drop those locks here before we do our read.
+		 */
+		lockstate = pollunlock();
+		err = lx_ptm_eof_drop(dev, &rval);
+		pollrelock(lockstate);
+		if (err)
+			return (err);
+
+		/* If no EOF was dropped then return */
+		if (rval == 0)
+			return (0);
+
+		/*
+		 * An EOF was removed from the stream.  Retry the entire
+		 * poll operation from the top because polls on the ptm
+		 * device should behave differently now.
+		 */
+		*loop = 1;
+	}
+	return (0);
+}
+
+static int
+lx_ptm_poll(dev_t dev, short events, int anyyet, short *reventsp,
+    struct pollhead **phpp)
+{
+	int loop, err;
+
+	do {
+		/* Serialize ourself wrt read operations. */
+		if (lx_ptm_read_start(dev) != 0)
+			return (EINTR);
+
+		err = lx_ptm_poll_loop(dev,
+		    events, anyyet, reventsp, phpp, &loop);
+		lx_ptm_read_end(dev);
+		if (err != 0)
+			return (err);
+	} while (loop != 0);
+	return (0);
+}
+
+static struct cb_ops lx_ptm_cb_ops = {
+	lx_ptm_open,		/* open */
+	lx_ptm_close,		/* close */
+	nodev,			/* strategy */
+	nodev,			/* print */
+	nodev,			/* dump */
+	lx_ptm_read,		/* read */
+	lx_ptm_write,		/* write */
+	lx_ptm_ioctl,		/* ioctl */
+	nodev,			/* devmap */
+	nodev,			/* mmap */
+	nodev,			/* segmap */
+	lx_ptm_poll,		/* chpoll */
+	ddi_prop_op,		/* prop_op */
+	NULL,			/* cb_str */
+	D_NEW | D_MP,
+	CB_REV,
+	NULL,
+	NULL
+};
+
+static struct dev_ops lx_ptm_ops = {
+	DEVO_REV,
+	0,
+	ddi_getinfo_1to1,
+	nulldev,
+	nulldev,
+	lx_ptm_attach,
+	lx_ptm_detach,
+	nodev,
+	&lx_ptm_cb_ops,
+	NULL,
+	NULL,
+	ddi_quiesce_not_needed,		/* quiesce */
+};
+
+static struct modldrv modldrv = {
+	&mod_driverops,			/* type of module */
+	"Linux master terminal driver",	/* description of module */
+	&lx_ptm_ops			/* driver ops */
+};
+
+static struct modlinkage modlinkage = {
+	MODREV_1,
+	&modldrv,
+	NULL
+};
+
+int
+_init(void)
+{
+	return (mod_install(&modlinkage));
+}
+
+int
+_info(struct modinfo *modinfop)
+{
+	return (mod_info(&modlinkage, modinfop));
+}
+
+int
+_fini(void)
+{
+	return (mod_remove(&modlinkage));
+}
diff --git a/usr/src/uts/common/brand/lx/io/lx_ptm.conf b/usr/src/uts/common/brand/lx/io/lx_ptm.conf
new file mode 100644
index 0000000000..481b4e3c74
--- /dev/null
+++ b/usr/src/uts/common/brand/lx/io/lx_ptm.conf
@@ -0,0 +1,27 @@
+#
+# CDDL HEADER START
+#
+# The contents of this file are subject to the terms of the
+# Common Development and Distribution License (the "License").
+# You may not use this file except in compliance with the License.
+#
+# You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
+# or http://www.opensolaris.org/os/licensing.
+# See the License for the specific language governing permissions
+# and limitations under the License.
+#
+# When distributing Covered Code, include this CDDL HEADER in each
+# file and include the License file at usr/src/OPENSOLARIS.LICENSE.
+# If applicable, add the following below this CDDL HEADER, with the
+# fields enclosed by brackets "[]" replaced with your own identifying
+# information: Portions Copyright [yyyy] [name of copyright owner]
+#
+# CDDL HEADER END
+#
+#
+# Copyright 2006 Sun Microsystems, Inc.  All rights reserved.
+# Use is subject to license terms.
+#
+#ident	"%Z%%M%	%I%	%E% SMI"
+
+name="lx_ptm" parent="pseudo" instance=0;
diff --git a/usr/src/uts/common/brand/lx/os/lx_brand.c b/usr/src/uts/common/brand/lx/os/lx_brand.c
new file mode 100644
index 0000000000..c22d782b68
--- /dev/null
+++ b/usr/src/uts/common/brand/lx/os/lx_brand.c
@@ -0,0 +1,1873 @@
+/*
+ * CDDL HEADER START
+ *
+ * The contents of this file are subject to the terms of the
+ * Common Development and Distribution License (the "License").
+ * You may not use this file except in compliance with the License.
+ *
+ * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
+ * or http://www.opensolaris.org/os/licensing.
+ * See the License for the specific language governing permissions
+ * and limitations under the License.
+ *
+ * When distributing Covered Code, include this CDDL HEADER in each
+ * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
+ * If applicable, add the following below this CDDL HEADER, with the
+ * fields enclosed by brackets "[]" replaced with your own identifying
+ * information: Portions Copyright [yyyy] [name of copyright owner]
+ *
+ * CDDL HEADER END
+ */
+
+/*
+ * Copyright 2010 Sun Microsystems, Inc.  All rights reserved.
+ * Use is subject to license terms.
+ */
+
+/*
+ * Copyright 2015, Joyent, Inc. All rights reserved.
+ */
+
+/*
+ * The LX Brand: emulation of a Linux operating environment within a zone.
+ *
+ * OVERVIEW
+ *
+ * The LX brand enables a full Linux userland -- including a C library,
+ * init(1) framework, and some set of applications -- to run unmodified
+ * within an illumos zone.  Unlike illumos, where applications are expected
+ * to link against and consume functions exported from libraries, the
+ * supported Linux binary compatibility boundary is the system call
+ * interface.  By accurately emulating the behaviour of Linux system calls,
+ * Linux software can be executed in this environment as if it were running
+ * on a native Linux system.
+ *
+ * EMULATING LINUX SYSTEM CALLS
+ *
+ * Linux system calls are made in 32-bit processes via the "int 0x80"
+ * instruction; in 64-bit processes the "syscall" instruction is used, as it
+ * is with native illumos processes.  In both cases, arguments to system
+ * calls are generally passed in registers and the usermode stack is not
+ * interpreted or modified by the Linux kernel.
+ *
+ * When the emulated Linux process makes a system call, it traps into the
+ * illumos kernel.  The in-kernel brand module contains various emulation
+ * routines, and can fully service some emulated system calls; e.g. read(2)
+ * and write(2).  Other system calls require assistance from the illumos
+ * libc, bouncing back out to the brand library ("lx_brand.so.1") for
+ * emulation.
+ *
+ * The brand mechanism allows for the provision of an alternative trap
+ * handler for the various system call mechanisms.  Traditionally this was
+ * used to immediately revector execution to the usermode emulation library,
+ * which was responsible for handling all system calls.  In the interests of
+ * more accurate emulation and increased performance, much of the regular
+ * illumos system call path is now invoked.  Only the argument processing and
+ * handler dispatch are replaced by the brand, via the per-LWP
+ * "lwp_brand_syscall" interposition function pointer.
+ *
+ * THE NATIVE AND BRAND STACKS
+ *
+ * Some runtime environments (e.g. the Go language) allocate very small
+ * thread stacks, preferring to grow or split the stack as necessary.  The
+ * Linux kernel generally does not use the usermode stack when servicing
+ * system calls, so this is not a problem.  In order for our emulation to
+ * have the same zero stack impact, we must execute usermode emulation
+ * routines on an _alternate_ stack.  This is similar, in principle, to the
+ * use of sigaltstack(3C) to run signal handlers off the main thread stack.
+ *
+ * To this end, the brand library allocates and installs an alternate stack
+ * (called the "native" stack) for each LWP.  The in-kernel brand code uses
+ * this stack for usermode emulation calls and interposed signal delivery,
+ * while the emulated Linux process sees only the data on the main thread
+ * stack, known as the "brand" stack.  The stack mode is tracked in the
+ * per-LWP brand-private data, using the LX_STACK_MODE_* enum.
+ *
+ * The stack mode doubles as a system call "mode bit".  When in the
+ * LX_STACK_MODE_BRAND mode, system calls are processed as emulated Linux
+ * system calls.  In other modes, system calls are assumed to be native
+ * illumos system calls as made during brand library initialisation and
+ * usermode emulation.
+ *
+ * USERMODE EMULATION
+ *
+ * When a Linux system call cannot be emulated within the kernel, we preserve
+ * the register state of the Linux process and revector the LWP to the brand
+ * library usermode emulation handler: the "lx_emulate()" function in
+ * "lx_brand.so.1".  This revectoring is modelled on the delivery of signals,
+ * and is performed in "lx_emulate_user()".
+ *
+ * First, the emulated process state is written out to the usermode stack of
+ * the process as a "ucontext_t" object.  Arguments to the emulation routine
+ * are passed on the stack or in registers, depending on the ABI.  When the
+ * usermode emulation is complete, the result is passed back to the kernel
+ * (via the "B_EMULATION_DONE" brandsys subcommand) with the saved context
+ * for restoration.
+ *
+ * SIGNAL DELIVERY, SETCONTEXT AND GETCONTEXT
+ *
+ * When servicing emulated system calls in the usermode brand library, or
+ * during signal delivery, various state is preserved by the kernel so that
+ * the running LWP may be revectored to a handling routine.  The context
+ * allows the kernel to restart the program at the point of interruption,
+ * either at the return of the signal handler, via setcontext(3C); or after
+ * the usermode emulation request has been serviced, via B_EMULATION_DONE.
+ *
+ * In illumos native processes, the saved context (a "ucontext_t" object)
+ * includes the state of registers and the current signal mask at the point
+ * of interruption.  The context also includes a link to the most recently
+ * saved context, forming a chain to be unwound as requests complete.  The LX
+ * brand requires additional book-keeping to describe the machine state: in
+ * particular, the current stack mode and the occupied extent of the native
+ * stack.
+ *
+ * The brand code is able to interpose on the context save and restore
+ * operations in the kernel -- see "lx_savecontext()" and
+ * "lx_restorecontext()" -- to enable getcontext(3C) and setcontext(3C) to
+ * function correctly in the face of a dual stack LWP.  The brand also
+ * interposes on the signal delivery mechanism -- see "lx_sendsig()" and
+ * "lx_sendsig_stack()" -- to allow all signals to be delivered to the brand
+ * library interposer on the native stack, regardless of the interrupted
+ * execution mode.  Linux sigaltstack(2) emulation is performed entirely by
+ * the usermode brand library during signal handler interposition.
+ */
+
+#include <sys/types.h>
+#include <sys/kmem.h>
+#include <sys/errno.h>
+#include <sys/thread.h>
+#include <sys/systm.h>
+#include <sys/syscall.h>
+#include <sys/proc.h>
+#include <sys/modctl.h>
+#include <sys/cmn_err.h>
+#include <sys/model.h>
+#include <sys/exec.h>
+#include <sys/lx_impl.h>
+#include <sys/machbrand.h>
+#include <sys/lx_syscalls.h>
+#include <sys/lx_misc.h>
+#include <sys/lx_futex.h>
+#include <sys/lx_brand.h>
+#include <sys/param.h>
+#include <sys/termios.h>
+#include <sys/sunddi.h>
+#include <sys/ddi.h>
+#include <sys/vnode.h>
+#include <sys/pathname.h>
+#include <sys/auxv.h>
+#include <sys/priv.h>
+#include <sys/regset.h>
+#include <sys/privregs.h>
+#include <sys/archsystm.h>
+#include <sys/zone.h>
+#include <sys/brand.h>
+#include <sys/sdt.h>
+#include <sys/x86_archext.h>
+#include <sys/controlregs.h>
+#include <sys/core.h>
+#include <sys/stack.h>
+#include <sys/stat.h>
+#include <sys/socket.h>
+#include <lx_signum.h>
+
+int	lx_debug = 0;
+
+void	lx_init_brand_data(zone_t *);
+void	lx_free_brand_data(zone_t *);
+void	lx_setbrand(proc_t *);
+int	lx_getattr(zone_t *, int, void *, size_t *);
+int	lx_setattr(zone_t *, int, void *, size_t);
+int	lx_brandsys(int, int64_t *, uintptr_t, uintptr_t, uintptr_t,
+		uintptr_t, uintptr_t);
+void	lx_set_kern_version(zone_t *, char *);
+void	lx_copy_procdata(proc_t *, proc_t *);
+
+extern int getsetcontext(int, void *);
+extern int waitsys(idtype_t, id_t, siginfo_t *, int);
+#if defined(_SYSCALL32_IMPL)
+extern int getsetcontext32(int, void *);
+extern int waitsys32(idtype_t, id_t, siginfo_t *, int);
+#endif
+
+extern void lx_proc_exit(proc_t *);
+extern int lx_sched_affinity(int, uintptr_t, int, uintptr_t, int64_t *);
+
+extern void lx_ioctl_init();
+extern void lx_ioctl_fini();
+
+lx_systrace_f *lx_systrace_entry_ptr;
+lx_systrace_f *lx_systrace_return_ptr;
+
+static int lx_systrace_enabled;
+
+/*
+ * While this is effectively mmu.hole_start - PAGESIZE, we don't particularly
+ * want an MMU dependency here (and should there be a microprocessor without
+ * a hole, we don't want to start allocating from the top of the VA range).
+ */
+#define	LX_MAXSTACK64	0x7ffffff00000
+
+uint64_t lx_maxstack64 = LX_MAXSTACK64;
+
+static int lx_elfexec(struct vnode *vp, struct execa *uap, struct uarg *args,
+    struct intpdata *idata, int level, long *execsz, int setid,
+    caddr_t exec_file, struct cred *cred, int *brand_action);
+
+static boolean_t lx_native_exec(uint8_t, const char **);
+static uint32_t lx_map32limit(proc_t *);
+
+static void lx_savecontext(ucontext_t *);
+static void lx_restorecontext(ucontext_t *);
+static caddr_t lx_sendsig_stack(int);
+static void lx_sendsig(int);
+#if defined(_SYSCALL32_IMPL)
+static void lx_savecontext32(ucontext32_t *);
+#endif
+static int lx_setid_clear(vattr_t *, cred_t *);
+#if defined(_LP64)
+static int lx_pagefault(proc_t *, klwp_t *, caddr_t, enum fault_type,
+    enum seg_rw);
+#endif
+
+
+/* lx brand */
+struct brand_ops lx_brops = {
+	lx_init_brand_data,		/* b_init_brand_data */
+	lx_free_brand_data,		/* b_free_brand_data */
+	lx_brandsys,			/* b_brandsys */
+	lx_setbrand,			/* b_setbrand */
+	lx_getattr,			/* b_getattr */
+	lx_setattr,			/* b_setattr */
+	lx_copy_procdata,		/* b_copy_procdata */
+	lx_proc_exit,			/* b_proc_exit */
+	lx_exec,			/* b_exec */
+	lx_setrval,			/* b_lwp_setrval */
+	lx_lwpdata_alloc,		/* b_lwpdata_alloc */
+	lx_lwpdata_free,		/* b_lwpdata_free */
+	lx_initlwp,			/* b_initlwp */
+	lx_forklwp,			/* b_forklwp */
+	lx_freelwp,			/* b_freelwp */
+	lx_exitlwp,			/* b_lwpexit */
+	lx_elfexec,			/* b_elfexec */
+	NULL,				/* b_sigset_native_to_brand */
+	NULL,				/* b_sigset_brand_to_native */
+	NULL,				/* b_psig_to_proc */
+	NSIG,				/* b_nsig */
+	lx_exit_with_sig,		/* b_exit_with_sig */
+	lx_wait_filter,			/* b_wait_filter */
+	lx_native_exec,			/* b_native_exec */
+	NULL,				/* b_ptrace_exectrap */
+	lx_map32limit,			/* b_map32limit */
+	lx_stop_notify,			/* b_stop_notify */
+	lx_waitid_helper,		/* b_waitid_helper */
+	lx_sigcld_repost,		/* b_sigcld_repost */
+	lx_ptrace_issig_stop,		/* b_issig_stop */
+	lx_ptrace_sig_ignorable,	/* b_sig_ignorable */
+	lx_savecontext,			/* b_savecontext */
+#if defined(_SYSCALL32_IMPL)
+	lx_savecontext32,		/* b_savecontext32 */
+#endif
+	lx_restorecontext,		/* b_restorecontext */
+	lx_sendsig_stack,		/* b_sendsig_stack */
+	lx_sendsig,			/* b_sendsig */
+	lx_setid_clear,			/* b_setid_clear */
+#if defined(_LP64)
+	lx_pagefault			/* b_pagefault */
+#else
+	NULL
+#endif
+};
+
+struct brand_mach_ops lx_mops = {
+	NULL,
+	NULL,
+	NULL,
+	NULL,
+	NULL,
+	lx_fixsegreg,
+	lx_fsbase
+};
+
+struct brand lx_brand = {
+	BRAND_VER_1,
+	"lx",
+	&lx_brops,
+	&lx_mops,
+	sizeof (struct lx_proc_data)
+};
+
+static struct modlbrand modlbrand = {
+	&mod_brandops, "lx brand", &lx_brand
+};
+
+static struct modlinkage modlinkage = {
+	MODREV_1, (void *)&modlbrand, NULL
+};
+
+void
+lx_proc_exit(proc_t *p)
+{
+	VERIFY(p->p_brand == &lx_brand);
+	VERIFY(p->p_brand_data != NULL);
+}
+
+void
+lx_setbrand(proc_t *p)
+{
+	/* Send SIGCHLD to parent by default when child exits */
+	ptolxproc(p)->l_signal = stol_signo[SIGCHLD];
+}
+
+/* ARGSUSED */
+int
+lx_setattr(zone_t *zone, int attr, void *buf, size_t bufsize)
+{
+	char vers[LX_VERS_MAX];
+
+	if (attr == LX_KERN_VERSION_NUM) {
+		if (bufsize > (LX_VERS_MAX - 1))
+			return (ERANGE);
+		bzero(vers, LX_VERS_MAX);
+		if (copyin(buf, &vers, bufsize) != 0)
+			return (EFAULT);
+		lx_set_kern_version(zone, vers);
+		return (0);
+	}
+	return (EINVAL);
+}
+
+/* ARGSUSED */
+int
+lx_getattr(zone_t *zone, int attr, void *buf, size_t *bufsize)
+{
+	if (attr == LX_KERN_VERSION_NUM) {
+		if (*bufsize < LX_VERS_MAX)
+			return (ERANGE);
+		if (copyout(lx_get_zone_kern_version(curzone), buf,
+		    LX_VERS_MAX) != 0)
+			return (EFAULT);
+		*bufsize = LX_VERS_MAX;
+		return (0);
+	}
+	return (-EINVAL);
+}
+
+uint32_t
+lx_map32limit(proc_t *p)
+{
+	/*
+	 * To be bug-for-bug compatible with Linux, we have MAP_32BIT only
+	 * allow mappings in the first 31 bits.  This was a nuance in the
+	 * original Linux implementation circa 2002, and applications have
+	 * come to depend on its behavior.
+	 *
+	 * This is only relevant for 64-bit processes.
+	 */
+	if (p->p_model == DATAMODEL_LP64)
+		return (1 << 31);
+
+	return ((uint32_t)USERLIMIT32);
+}
+
+void
+lx_brand_systrace_enable(void)
+{
+	VERIFY(!lx_systrace_enabled);
+
+	lx_systrace_enabled = 1;
+}
+
+void
+lx_brand_systrace_disable(void)
+{
+	VERIFY(lx_systrace_enabled);
+
+	lx_systrace_enabled = 0;
+}
+
+void
+lx_lwp_set_native_stack_current(lx_lwp_data_t *lwpd, uintptr_t new_sp)
+{
+	VERIFY(lwpd->br_ntv_stack != 0);
+
+	/*
+	 * The "brand-lx-set-ntv-stack-current" probe has arguments:
+	 *   arg0: stack pointer before change
+	 *   arg1: stack pointer after change
+	 *   arg2: current stack base
+	 */
+	DTRACE_PROBE3(brand__lx__set__ntv__stack__current,
+	    uintptr_t, lwpd->br_ntv_stack_current,
+	    uintptr_t, new_sp,
+	    uintptr_t, lwpd->br_ntv_stack);
+
+	lwpd->br_ntv_stack_current = new_sp;
+}
+
+#if defined(_LP64)
+static int
+lx_pagefault(proc_t *p, klwp_t *lwp, caddr_t addr, enum fault_type type,
+    enum seg_rw rw)
+{
+	int syscall_num;
+
+	/*
+	 * We only want to handle a very specific set of circumstances.
+	 * Namely: this is a 64-bit LX-branded process attempting to execute an
+	 * address in a page for which it does not have a valid mapping.  If
+	 * this is not the case, we bail out as fast as possible.
+	 */
+	VERIFY(PROC_IS_BRANDED(p));
+	if (type != F_INVAL || rw != S_EXEC || lwp_getdatamodel(lwp) !=
+	    DATAMODEL_NATIVE) {
+		return (-1);
+	}
+
+	if (!lx_vsyscall_iscall(lwp, (uintptr_t)addr, &syscall_num)) {
+		return (-1);
+	}
+
+	/*
+	 * This is a valid vsyscall address.  We service the system call and
+	 * return 0 to signal that the pagefault has been handled completely.
+	 */
+	lx_vsyscall_enter(p, lwp, syscall_num);
+	return (0);
+}
+#endif
+
+/*
+ * This hook runs prior to sendsig() processing and allows us to nominate
+ * an alternative stack pointer for delivery of the signal handling frame.
+ * Critically, this routine should _not_ modify any LWP state as the
+ * savecontext() does not run until after this hook.
+ */
+static caddr_t
+lx_sendsig_stack(int sig)
+{
+	klwp_t *lwp = ttolwp(curthread);
+	lx_lwp_data_t *lwpd = lwptolxlwp(lwp);
+
+	/*
+	 * We want to take signal delivery on the native stack, but only if
+	 * one has been allocated and installed for this LWP.
+	 */
+	if (lwpd->br_stack_mode == LX_STACK_MODE_BRAND) {
+		/*
+		 * The program is not running on the native stack.  Return
+		 * the native stack pointer from our brand-private data so
+		 * that we may switch to it for signal handling.
+		 */
+		return ((caddr_t)lwpd->br_ntv_stack_current);
+	} else {
+		struct regs *rp = lwptoregs(lwp);
+
+		/*
+		 * Either the program is already running on the native stack,
+		 * or one has not yet been allocated for this LWP.  Use the
+		 * current stack pointer value.
+		 */
+		return ((caddr_t)rp->r_sp);
+	}
+}
+
+/*
+ * This hook runs after sendsig() processing and allows us to update the
+ * per-LWP mode flags for system calls and stacks.  The pre-signal
+ * context has already been saved and delivered to the user at this point.
+ */
+static void
+lx_sendsig(int sig)
+{
+	klwp_t *lwp = ttolwp(curthread);
+	lx_lwp_data_t *lwpd = lwptolxlwp(lwp);
+	struct regs *rp = lwptoregs(lwp);
+
+	switch (lwpd->br_stack_mode) {
+	case LX_STACK_MODE_BRAND:
+	case LX_STACK_MODE_NATIVE:
+		/*
+		 * In lx_sendsig_stack(), we nominated a stack pointer from the
+		 * native stack.  Update the stack mode, and the current in-use
+		 * extent of the native stack, accordingly:
+		 */
+		lwpd->br_stack_mode = LX_STACK_MODE_NATIVE;
+		lx_lwp_set_native_stack_current(lwpd, rp->r_sp);
+
+		/*
+		 * Fix up segment registers, etc.
+		 */
+		lx_switch_to_native(lwp);
+		break;
+
+	default:
+		/*
+		 * Otherwise, the brand library has not yet installed the
+		 * alternate stack for this LWP.  Signals will be handled on
+		 * the regular stack thread.
+		 */
+		return;
+	}
+}
+
+/*
+ * This hook runs prior to the context restoration, allowing us to take action
+ * or modify the context before it is loaded.
+ */
+static void
+lx_restorecontext(ucontext_t *ucp)
+{
+	klwp_t *lwp = ttolwp(curthread);
+	lx_lwp_data_t *lwpd = lwptolxlwp(lwp);
+	uintptr_t flags = (uintptr_t)ucp->uc_brand_data[0];
+	caddr_t sp = ucp->uc_brand_data[1];
+
+	/*
+	 * We have a saved native stack pointer value that we must restore
+	 * into the per-LWP data.
+	 */
+	if (flags & LX_UC_RESTORE_NATIVE_SP) {
+		lx_lwp_set_native_stack_current(lwpd, (uintptr_t)sp);
+	}
+
+	/*
+	 * We do not wish to restore the value of uc_link in this context,
+	 * so replace it with the value currently in the LWP.
+	 */
+	if (flags & LX_UC_IGNORE_LINK) {
+		ucp->uc_link = (ucontext_t *)lwp->lwp_oldcontext;
+	}
+
+	/*
+	 * Restore the stack mode:
+	 */
+	if (flags & LX_UC_STACK_NATIVE) {
+		lwpd->br_stack_mode = LX_STACK_MODE_NATIVE;
+	} else if (flags & LX_UC_STACK_BRAND) {
+		lwpd->br_stack_mode = LX_STACK_MODE_BRAND;
+	}
+
+#if defined(__amd64)
+	/*
+	 * Override the fs/gsbase in the context with the value provided
+	 * through the Linux arch_prctl(2) system call.
+	 */
+	if (flags & LX_UC_STACK_BRAND) {
+		if (lwpd->br_lx_fsbase != 0) {
+			ucp->uc_mcontext.gregs[REG_FSBASE] = lwpd->br_lx_fsbase;
+		}
+		if (lwpd->br_lx_gsbase != 0) {
+			ucp->uc_mcontext.gregs[REG_GSBASE] = lwpd->br_lx_gsbase;
+		}
+	}
+#endif
+}
+
+static void
+lx_savecontext(ucontext_t *ucp)
+{
+	klwp_t *lwp = ttolwp(curthread);
+	lx_lwp_data_t *lwpd = lwptolxlwp(lwp);
+	uintptr_t flags = 0;
+
+	/*
+	 * The ucontext_t affords us three private pointer-sized members in
+	 * "uc_brand_data".  We pack a variety of flags into the first element,
+	 * and an optional stack pointer in the second element.  The flags
+	 * determine which stack pointer (native or brand), if any, is stored
+	 * in the second element.  The third element may contain the system
+	 * call number; this is analogous to the "orig_[er]ax" member of a
+	 * Linux "user_regs_struct".
+	 */
+
+	if (lwpd->br_stack_mode != LX_STACK_MODE_INIT &&
+	    lwpd->br_stack_mode != LX_STACK_MODE_PREINIT) {
+		/*
+		 * Record the value of the native stack pointer to restore
+		 * when returning to this branded context:
+		 */
+		flags |= LX_UC_RESTORE_NATIVE_SP;
+		ucp->uc_brand_data[1] = (void *)lwpd->br_ntv_stack_current;
+	}
+
+	/*
+	 * Save the stack mode:
+	 */
+	if (lwpd->br_stack_mode == LX_STACK_MODE_NATIVE) {
+		flags |= LX_UC_STACK_NATIVE;
+	} else if (lwpd->br_stack_mode == LX_STACK_MODE_BRAND) {
+		flags |= LX_UC_STACK_BRAND;
+	}
+
+	/*
+	 * If we might need to restart this system call, save that information
+	 * in the context:
+	 */
+	if (lwpd->br_stack_mode == LX_STACK_MODE_BRAND) {
+		ucp->uc_brand_data[2] =
+		    (void *)(uintptr_t)lwpd->br_syscall_num;
+		if (lwpd->br_syscall_restart) {
+			flags |= LX_UC_RESTART_SYSCALL;
+		}
+	} else {
+		ucp->uc_brand_data[2] = NULL;
+	}
+
+	ucp->uc_brand_data[0] = (void *)flags;
+}
+
+#if defined(_SYSCALL32_IMPL)
+static void
+lx_savecontext32(ucontext32_t *ucp)
+{
+	klwp_t *lwp = ttolwp(curthread);
+	lx_lwp_data_t *lwpd = lwptolxlwp(lwp);
+	unsigned int flags = 0;
+
+	/*
+	 * The ucontext_t affords us three private pointer-sized members in
+	 * "uc_brand_data".  We pack a variety of flags into the first element,
+	 * and an optional stack pointer in the second element.  The flags
+	 * determine which stack pointer (native or brand), if any, is stored
+	 * in the second element.  The third element may contain the system
+	 * call number; this is analogous to the "orig_[er]ax" member of a
+	 * Linux "user_regs_struct".
+	 */
+
+	if (lwpd->br_stack_mode != LX_STACK_MODE_INIT &&
+	    lwpd->br_stack_mode != LX_STACK_MODE_PREINIT) {
+		/*
+		 * Record the value of the native stack pointer to restore
+		 * when returning to this branded context:
+		 */
+		flags |= LX_UC_RESTORE_NATIVE_SP;
+		ucp->uc_brand_data[1] = (caddr32_t)lwpd->br_ntv_stack_current;
+	}
+
+	/*
+	 * Save the stack mode:
+	 */
+	if (lwpd->br_stack_mode == LX_STACK_MODE_NATIVE) {
+		flags |= LX_UC_STACK_NATIVE;
+	} else if (lwpd->br_stack_mode == LX_STACK_MODE_BRAND) {
+		flags |= LX_UC_STACK_BRAND;
+	}
+
+	/*
+	 * If we might need to restart this system call, save that information
+	 * in the context:
+	 */
+	if (lwpd->br_stack_mode == LX_STACK_MODE_BRAND) {
+		ucp->uc_brand_data[2] = (caddr32_t)lwpd->br_syscall_num;
+		if (lwpd->br_syscall_restart) {
+			flags |= LX_UC_RESTART_SYSCALL;
+		}
+	} else {
+		ucp->uc_brand_data[2] = NULL;
+	}
+
+	ucp->uc_brand_data[0] = flags;
+}
+#endif
+
+void
+lx_init_brand_data(zone_t *zone)
+{
+	lx_zone_data_t *data;
+	ASSERT(zone->zone_brand == &lx_brand);
+	ASSERT(zone->zone_brand_data == NULL);
+	data = (lx_zone_data_t *)kmem_zalloc(sizeof (lx_zone_data_t), KM_SLEEP);
+	/*
+	 * Set the default lxzd_kernel_version to 2.4.
+	 * This can be changed by a call to setattr() during zone boot.
+	 */
+	(void) strlcpy(data->lxzd_kernel_version, "2.4.21", LX_VERS_MAX);
+
+	/*
+	 * Linux is not at all picky about address family when it comes to
+	 * supporting interface-related ioctls.  To mimic this behavior, we'll
+	 * attempt those ioctls against a ksocket configured for that purpose.
+	 */
+	(void) ksocket_socket(&data->lxzd_ioctl_sock, AF_INET, SOCK_DGRAM, 0,
+	    0, zone->zone_kcred);
+
+	zone->zone_brand_data = data;
+
+	/*
+	 * In Linux, if the init(1) process terminates the system panics.
+	 * The zone must reboot to simulate this behaviour.
+	 */
+	zone->zone_reboot_on_init_exit = B_TRUE;
+}
+
+void
+lx_free_brand_data(zone_t *zone)
+{
+	lx_zone_data_t *data = ztolxzd(zone);
+	ASSERT(data != NULL);
+	if (data->lxzd_ioctl_sock != NULL) {
+		/*
+		 * Since zone_kcred has been cleaned up already, close the
+		 * socket using the global kcred.
+		 */
+		ksocket_close(data->lxzd_ioctl_sock, kcred);
+		data->lxzd_ioctl_sock = NULL;
+	}
+	zone->zone_brand_data = NULL;
+	kmem_free(data, sizeof (*data));
+}
+
+void
+lx_unsupported(char *dmsg)
+{
+	lx_proc_data_t *pd = ttolxproc(curthread);
+
+	DTRACE_PROBE1(brand__lx__unsupported, char *, dmsg);
+
+	if (pd != NULL && (pd->l_flags & LX_PROC_STRICT_MODE) != 0) {
+		/*
+		 * If this process was run with strict mode enabled
+		 * (via LX_STRICT in the environment), we mark this
+		 * LWP as having triggered an unsupported behaviour.
+		 * This flag will be checked at an appropriate point
+		 * by lx_check_strict_failure().
+		 */
+		lx_lwp_data_t *lwpd = ttolxlwp(curthread);
+
+		lwpd->br_strict_failure = B_TRUE;
+	}
+}
+
+void
+lx_check_strict_failure(lx_lwp_data_t *lwpd)
+{
+	proc_t *p;
+
+	if (!lwpd->br_strict_failure) {
+		return;
+	}
+
+	lwpd->br_strict_failure = B_FALSE;
+
+	/*
+	 * If this process is operating in strict mode (via LX_STRICT in
+	 * the environment), and has triggered a call to
+	 * lx_unsupported(), we drop SIGSYS on it as we return.
+	 */
+	p = curproc;
+	mutex_enter(&p->p_lock);
+	sigtoproc(p, curthread, SIGSYS);
+	mutex_exit(&p->p_lock);
+}
+
+void
+lx_trace_sysenter(int syscall_num, uintptr_t *args)
+{
+	if (lx_systrace_enabled) {
+		VERIFY(lx_systrace_entry_ptr != NULL);
+
+		(*lx_systrace_entry_ptr)(syscall_num, args[0], args[1],
+		    args[2], args[3], args[4], args[5]);
+	}
+}
+
+void
+lx_trace_sysreturn(int syscall_num, long ret)
+{
+	if (lx_systrace_enabled) {
+		VERIFY(lx_systrace_return_ptr != NULL);
+
+		(*lx_systrace_return_ptr)(syscall_num, ret, ret, 0, 0, 0, 0);
+	}
+}
+
+/*
+ * Get the addresses of the user-space system call handler and attach it to
+ * the proc structure. Returning 0 indicates success; the value returned
+ * by the system call is the value stored in rval. Returning a non-zero
+ * value indicates a failure; the value returned is used to set errno, -1
+ * is returned from the syscall and the contents of rval are ignored. To
+ * set errno and have the syscall return a value other than -1 we can
+ * manually set errno and rval and return 0.
+ */
+int
+lx_brandsys(int cmd, int64_t *rval, uintptr_t arg1, uintptr_t arg2,
+    uintptr_t arg3, uintptr_t arg4, uintptr_t arg5)
+{
+	kthread_t *t = curthread;
+	klwp_t *lwp = ttolwp(t);
+	proc_t *p = ttoproc(t);
+	lx_proc_data_t *pd;
+	struct termios *termios;
+	uint_t termios_len;
+	int error;
+	int code;
+	int sig;
+	lx_brand_registration_t reg;
+	lx_lwp_data_t *lwpd = lwptolxlwp(lwp);
+
+	/*
+	 * There is one operation that is suppored for non-branded
+	 * process.  B_EXEC_BRAND.  This is the equilivant of an
+	 * exec call, but the new process that is created will be
+	 * a branded process.
+	 */
+	if (cmd == B_EXEC_BRAND) {
+		VERIFY(p->p_zone != NULL);
+		VERIFY(p->p_zone->zone_brand == &lx_brand);
+		return (exec_common(
+		    (char *)arg1, (const char **)arg2, (const char **)arg3,
+		    EBA_BRAND));
+	}
+
+	/* For all other operations this must be a branded process. */
+	if (p->p_brand == NULL)
+		return (ENOSYS);
+
+	VERIFY(p->p_brand == &lx_brand);
+	VERIFY(p->p_brand_data != NULL);
+
+	switch (cmd) {
+	case B_REGISTER:
+		if (lwpd->br_stack_mode != LX_STACK_MODE_PREINIT) {
+			lx_print("stack mode was not PREINIT during "
+			    "REGISTER\n");
+			return (EINVAL);
+		}
+
+		if (p->p_model == DATAMODEL_NATIVE) {
+			if (copyin((void *)arg1, &reg, sizeof (reg)) != 0) {
+				lx_print("Failed to copyin brand registration "
+				    "at 0x%p\n", (void *)arg1);
+				return (EFAULT);
+			}
+		}
+#ifdef _LP64
+		else {
+			/* 32-bit userland on 64-bit kernel */
+			lx_brand_registration32_t reg32;
+
+			if (copyin((void *)arg1, &reg32, sizeof (reg32)) != 0) {
+				lx_print("Failed to copyin brand registration "
+				    "at 0x%p\n", (void *)arg1);
+				return (EFAULT);
+			}
+
+			reg.lxbr_version = (uint_t)reg32.lxbr_version;
+			reg.lxbr_handler =
+			    (void *)(uintptr_t)reg32.lxbr_handler;
+			reg.lxbr_flags = reg32.lxbr_flags;
+		}
+#endif
+
+		if (reg.lxbr_version != LX_VERSION_1) {
+			lx_print("Invalid brand library version (%u)\n",
+			    reg.lxbr_version);
+			return (EINVAL);
+		}
+
+		if ((reg.lxbr_flags & ~LX_PROC_ALL) != 0) {
+			lx_print("Invalid brand flags (%u)\n",
+			    reg.lxbr_flags);
+			return (EINVAL);
+		}
+
+		lx_print("Assigning brand 0x%p and handler 0x%p to proc 0x%p\n",
+		    (void *)&lx_brand, (void *)reg.lxbr_handler, (void *)p);
+		pd = p->p_brand_data;
+		pd->l_handler = (uintptr_t)reg.lxbr_handler;
+		pd->l_flags = reg.lxbr_flags;
+
+		return (0);
+
+	case B_TTYMODES:
+		/* This is necessary for emulating TCGETS ioctls. */
+		if (ddi_prop_lookup_byte_array(DDI_DEV_T_ANY, ddi_root_node(),
+		    DDI_PROP_NOTPROM, "ttymodes", (uchar_t **)&termios,
+		    &termios_len) != DDI_SUCCESS)
+			return (EIO);
+
+		ASSERT(termios_len == sizeof (*termios));
+
+		if (copyout(&termios, (void *)arg1, sizeof (termios)) != 0) {
+			ddi_prop_free(termios);
+			return (EFAULT);
+		}
+
+		ddi_prop_free(termios);
+		return (0);
+
+	case B_ELFDATA:
+		pd = curproc->p_brand_data;
+		if (get_udatamodel() == DATAMODEL_NATIVE) {
+			if (copyout(&pd->l_elf_data, (void *)arg1,
+			    sizeof (lx_elf_data_t)) != 0) {
+				return (EFAULT);
+			}
+		}
+#if defined(_LP64)
+		else {
+			/* 32-bit userland on 64-bit kernel */
+			lx_elf_data32_t led32;
+
+			led32.ed_phdr = (int)pd->l_elf_data.ed_phdr;
+			led32.ed_phent = (int)pd->l_elf_data.ed_phent;
+			led32.ed_phnum = (int)pd->l_elf_data.ed_phnum;
+			led32.ed_entry = (int)pd->l_elf_data.ed_entry;
+			led32.ed_base = (int)pd->l_elf_data.ed_base;
+			led32.ed_ldentry = (int)pd->l_elf_data.ed_ldentry;
+
+			if (copyout(&led32, (void *)arg1,
+			    sizeof (led32)) != 0) {
+				return (EFAULT);
+			}
+		}
+#endif
+		return (0);
+
+	case B_EXEC_NATIVE:
+		return (exec_common((char *)arg1, (const char **)arg2,
+		    (const char **)arg3, EBA_NATIVE));
+
+	/*
+	 * The B_TRUSS_POINT subcommand is used so that we can make a no-op
+	 * syscall for debugging purposes (dtracing) from within the user-level
+	 * emulation.
+	 */
+	case B_TRUSS_POINT:
+		return (0);
+
+	case B_LPID_TO_SPAIR: {
+		/*
+		 * Given a Linux pid as arg1, return the Solaris pid in arg2 and
+		 * the Solaris LWP in arg3.  We also translate pid 1 (which is
+		 * hardcoded in many applications) to the zone's init process.
+		 */
+		pid_t s_pid;
+		id_t s_tid;
+
+		if ((pid_t)arg1 == 1) {
+			s_pid = p->p_zone->zone_proc_initpid;
+			/* handle the dead/missing init(1M) case */
+			if (s_pid == -1)
+				s_pid = 1;
+			s_tid = 1;
+		} else if (lx_lpid_to_spair((pid_t)arg1, &s_pid, &s_tid) < 0) {
+			return (ESRCH);
+		}
+
+		if (copyout(&s_pid, (void *)arg2, sizeof (s_pid)) != 0 ||
+		    copyout(&s_tid, (void *)arg3, sizeof (s_tid)) != 0) {
+			return (EFAULT);
+		}
+
+		return (0);
+	}
+
+	case B_SET_AFFINITY_MASK:
+	case B_GET_AFFINITY_MASK:
+		/*
+		 * Retrieve or store the CPU affinity mask for the
+		 * requested linux pid.
+		 *
+		 * arg1 is a linux PID (0 means curthread).
+		 * arg2 is the size of the given mask.
+		 * arg3 is the address of the affinity mask.
+		 */
+		return (lx_sched_affinity(cmd, arg1, arg2, arg3, rval));
+
+	case B_PTRACE_STOP_FOR_OPT:
+		return (lx_ptrace_stop_for_option((int)arg1, arg2 == 0 ?
+		    B_FALSE : B_TRUE, (ulong_t)arg3, arg4));
+
+	case B_PTRACE_CLONE_BEGIN:
+		return (lx_ptrace_set_clone_inherit((int)arg1, arg2 == 0 ?
+		    B_FALSE : B_TRUE));
+
+	case B_PTRACE_KERNEL:
+		return (lx_ptrace_kernel((int)arg1, (pid_t)arg2, arg3, arg4));
+
+	case B_HELPER_WAITID: {
+		idtype_t idtype = (idtype_t)arg1;
+		id_t id = (id_t)arg2;
+		siginfo_t *infop = (siginfo_t *)arg3;
+		int options = (int)arg4;
+
+		lwpd = ttolxlwp(curthread);
+
+		/*
+		 * Our brand-specific waitid helper only understands a subset of
+		 * the possible idtypes.  Ensure we keep to that subset here:
+		 */
+		if (idtype != P_ALL && idtype != P_PID && idtype != P_PGID) {
+			return (EINVAL);
+		}
+
+		/*
+		 * Enable the return of emulated ptrace(2) stop conditions
+		 * through lx_waitid_helper, and stash the Linux-specific
+		 * extra waitid() flags.
+		 */
+		lwpd->br_waitid_emulate = B_TRUE;
+		lwpd->br_waitid_flags = (int)arg5;
+
+#if defined(_SYSCALL32_IMPL)
+		if (get_udatamodel() != DATAMODEL_NATIVE) {
+			return (waitsys32(idtype, id, infop, options));
+		} else
+#endif
+		{
+			return (waitsys(idtype, id, infop, options));
+		}
+
+		lwpd->br_waitid_emulate = B_FALSE;
+		lwpd->br_waitid_flags = 0;
+
+		return (0);
+	}
+
+	case B_UNSUPPORTED: {
+		char dmsg[256];
+
+		if (copyin((void *)arg1, &dmsg, sizeof (dmsg)) != 0) {
+			lx_print("Failed to copyin unsupported msg "
+			    "at 0x%p\n", (void *)arg1);
+			return (EFAULT);
+		}
+		dmsg[255] = '\0';
+		lx_unsupported(dmsg);
+
+		lx_check_strict_failure(lwpd);
+
+		return (0);
+	}
+
+	case B_STORE_ARGS: {
+		/*
+		 * B_STORE_ARGS subcommand
+		 * arg1 = address of struct to be copied in
+		 * arg2 = size of the struct being copied in
+		 * arg3-arg6 ignored
+		 * rval = the amount of data copied.
+		 */
+		void *buf;
+
+		/* only have upper limit because arg2 is unsigned */
+		if (arg2 > LX_BR_ARGS_SIZE_MAX) {
+			return (EINVAL);
+		}
+
+		buf = kmem_alloc(arg2, KM_SLEEP);
+		if (copyin((void *)arg1, buf, arg2) != 0) {
+			lx_print("Failed to copyin scall arg at 0x%p\n",
+			    (void *) arg1);
+			kmem_free(buf, arg2);
+			/*
+			 * Purposely not setting br_scall_args to NULL
+			 * to preserve data for debugging.
+			 */
+			return (EFAULT);
+		}
+
+		if (lwpd->br_scall_args != NULL) {
+			ASSERT(lwpd->br_args_size > 0);
+			kmem_free(lwpd->br_scall_args,
+			    lwpd->br_args_size);
+		}
+
+		lwpd->br_scall_args = buf;
+		lwpd->br_args_size = arg2;
+		*rval = arg2;
+		return (0);
+	}
+
+	case B_HELPER_CLONE:
+		return (lx_helper_clone(rval, arg1, (void *)arg2, (void *)arg3,
+		    (void *)arg4));
+
+	case B_HELPER_SETGROUPS:
+		return (lx_helper_setgroups(arg1, (gid_t *)arg2));
+
+	case B_HELPER_SIGQUEUE:
+		return (lx_helper_rt_sigqueueinfo(arg1, arg2,
+		    (siginfo_t *)arg3));
+
+	case B_HELPER_TGSIGQUEUE:
+		return (lx_helper_rt_tgsigqueueinfo(arg1, arg2, arg3,
+		    (siginfo_t *)arg4));
+
+	case B_SET_THUNK_PID:
+		lwpd->br_lx_thunk_pid = arg1;
+		return (0);
+
+	case B_GETPID:
+		/*
+		 * The usermode clone(2) code needs to be able to call
+		 * lx_getpid() from native code:
+		 */
+		*rval = lx_getpid();
+		return (0);
+
+	case B_SET_NATIVE_STACK:
+		/*
+		 * B_SET_NATIVE_STACK subcommand
+		 * arg1 = the base of the stack to use for emulation
+		 */
+		if (lwpd->br_stack_mode != LX_STACK_MODE_PREINIT) {
+			lx_print("B_SET_NATIVE_STACK when stack was already "
+			    "set to %p\n", (void *)arg1);
+			return (EEXIST);
+		}
+
+		/*
+		 * We move from the PREINIT state, where we have no brand
+		 * emulation stack, to the INIT state.  Here, we are still
+		 * running on what will become the BRAND stack, but are running
+		 * emulation (i.e. native) code.  Once the initialisation
+		 * process for this thread has finished, we will jump to
+		 * brand-specific code, while moving to the BRAND mode.
+		 *
+		 * When a new LWP is created, lx_initlwp() will clear the
+		 * stack data.  If that LWP is actually being duplicated
+		 * into a child process by fork(2), lx_forklwp() will copy
+		 * it so that the cloned thread will keep using the same
+		 * alternate stack.
+		 */
+		lwpd->br_ntv_stack = arg1;
+		lwpd->br_stack_mode = LX_STACK_MODE_INIT;
+		lx_lwp_set_native_stack_current(lwpd, arg1);
+
+		return (0);
+
+	case B_GET_CURRENT_CONTEXT:
+		/*
+		 * B_GET_CURRENT_CONTEXT subcommand:
+		 * arg1 = address for pointer to current ucontext_t
+		 */
+
+#if defined(_SYSCALL32_IMPL)
+		if (get_udatamodel() != DATAMODEL_NATIVE) {
+			caddr32_t addr = (caddr32_t)lwp->lwp_oldcontext;
+
+			error = copyout(&addr, (void *)arg1, sizeof (addr));
+		} else
+#endif
+		{
+			error = copyout(&lwp->lwp_oldcontext, (void *)arg1,
+			    sizeof (lwp->lwp_oldcontext));
+		}
+
+		return (error != 0 ? EFAULT : 0);
+
+	case B_JUMP_TO_LINUX:
+		/*
+		 * B_JUMP_TO_LINUX subcommand:
+		 * arg1 = ucontext_t pointer for jump state
+		 */
+
+		if (arg1 == NULL)
+			return (EINVAL);
+
+		switch (lwpd->br_stack_mode) {
+		case LX_STACK_MODE_NATIVE: {
+			struct regs *rp = lwptoregs(lwp);
+
+			/*
+			 * We are on the NATIVE stack, so we must preserve
+			 * the extent of that stack.  The pointer will be
+			 * reset by a future setcontext().
+			 */
+			lx_lwp_set_native_stack_current(lwpd,
+			    (uintptr_t)rp->r_sp);
+			break;
+		}
+
+		case LX_STACK_MODE_INIT:
+			/*
+			 * The LWP is transitioning to Linux code for the first
+			 * time.
+			 */
+			break;
+
+		case LX_STACK_MODE_PREINIT:
+			/*
+			 * This LWP has not installed an alternate stack for
+			 * usermode emulation handling.
+			 */
+			return (ENOENT);
+
+		case LX_STACK_MODE_BRAND:
+			/*
+			 * The LWP should not be on the BRAND stack.
+			 */
+			exit(CLD_KILLED, SIGSYS);
+			return (0);
+		}
+
+		/*
+		 * Transfer control to Linux:
+		 */
+		return (lx_runexe(lwp, (void *)arg1));
+
+	case B_EMULATION_DONE:
+		/*
+		 * B_EMULATION_DONE subcommand:
+		 * arg1 = ucontext_t * to restore
+		 * arg2 = system call number
+		 * arg3 = return code
+		 * arg4 = if operation failed, the errno value
+		 */
+
+		/*
+		 * The first part of this operation is a setcontext() to
+		 * restore the register state to the copy we preserved
+		 * before vectoring to the usermode emulation routine.
+		 * If that fails, we return (hopefully) to the emulation
+		 * routine and it will handle the error.
+		 */
+#if (_SYSCALL32_IMPL)
+		if (get_udatamodel() != DATAMODEL_NATIVE) {
+			error = getsetcontext32(SETCONTEXT, (void *)arg1);
+		} else
+#endif
+		{
+			error = getsetcontext(SETCONTEXT, (void *)arg1);
+		}
+
+		if (error != 0) {
+			return (error);
+		}
+
+		/*
+		 * The saved Linux context has been restored.  We handle the
+		 * return value or errno with code common to the in-kernel
+		 * system call emulation.
+		 */
+		if ((error = (int)arg4) != 0) {
+			/*
+			 * lx_syscall_return() looks at the errno in the LWP,
+			 * so set it here:
+			 */
+			set_errno(error);
+		}
+		lx_syscall_return(ttolwp(curthread), (int)arg2, (long)arg3);
+
+		return (0);
+
+	case B_EXIT_AS_SIG:
+		code = CLD_KILLED;
+		sig = (int)arg1;
+		proc_is_exiting(p);
+		if (exitlwps(1) != 0) {
+			mutex_enter(&p->p_lock);
+			lwp_exit();
+		}
+		ttolwp(curthread)->lwp_cursig = sig;
+		if (sig == SIGSEGV) {
+			if (core(sig, 0) == 0)
+				code = CLD_DUMPED;
+		}
+		exit(code, sig);
+		/* NOTREACHED */
+		break;
+	}
+
+	return (EINVAL);
+}
+
+char *
+lx_get_zone_kern_version(zone_t *zone)
+{
+	return (((lx_zone_data_t *)zone->zone_brand_data)->lxzd_kernel_version);
+}
+
+void
+lx_set_kern_version(zone_t *zone, char *vers)
+{
+	lx_zone_data_t *lxzd = (lx_zone_data_t *)zone->zone_brand_data;
+
+	(void) strlcpy(lxzd->lxzd_kernel_version, vers, LX_VERS_MAX);
+}
+
+/*
+ * Linux unconditionally removes the setuid and setgid bits when changing
+ * file ownership.  This brand hook overrides the illumos native behaviour,
+ * which is based on the PRIV_FILE_SETID privilege.
+ */
+static int
+lx_setid_clear(vattr_t *vap, cred_t *cr)
+{
+	if (S_ISDIR(vap->va_mode)) {
+		return (0);
+	}
+
+	if (vap->va_mode & S_ISUID) {
+		vap->va_mask |= AT_MODE;
+		vap->va_mode &= ~S_ISUID;
+	}
+	if ((vap->va_mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
+		vap->va_mask |= AT_MODE;
+		vap->va_mode &= ~S_ISGID;
+	}
+
+	return (0);
+}
+
+/*
+ * Copy the per-process brand data from a parent proc to a child.
+ */
+void
+lx_copy_procdata(proc_t *child, proc_t *parent)
+{
+	lx_proc_data_t *cpd = child->p_brand_data;
+	lx_proc_data_t *ppd = parent->p_brand_data;
+
+	VERIFY(parent->p_brand == &lx_brand);
+	VERIFY(child->p_brand == &lx_brand);
+	VERIFY(ppd != NULL);
+	VERIFY(cpd != NULL);
+
+	*cpd = *ppd;
+
+	cpd->l_fake_limits[LX_RLFAKE_LOCKS].rlim_cur = LX_RLIM64_INFINITY;
+	cpd->l_fake_limits[LX_RLFAKE_LOCKS].rlim_max = LX_RLIM64_INFINITY;
+
+	cpd->l_fake_limits[LX_RLFAKE_NICE].rlim_cur = 20;
+	cpd->l_fake_limits[LX_RLFAKE_NICE].rlim_max = 20;
+
+	cpd->l_fake_limits[LX_RLFAKE_RTPRIO].rlim_cur = LX_RLIM64_INFINITY;
+	cpd->l_fake_limits[LX_RLFAKE_RTPRIO].rlim_max = LX_RLIM64_INFINITY;
+
+	cpd->l_fake_limits[LX_RLFAKE_RTTIME].rlim_cur = LX_RLIM64_INFINITY;
+	cpd->l_fake_limits[LX_RLFAKE_RTTIME].rlim_max = LX_RLIM64_INFINITY;
+}
+
+#if defined(_LP64)
+static void
+Ehdr32to64(Elf32_Ehdr *src, Ehdr *dst)
+{
+	bcopy(src->e_ident, dst->e_ident, sizeof (src->e_ident));
+	dst->e_type =		src->e_type;
+	dst->e_machine =	src->e_machine;
+	dst->e_version =	src->e_version;
+	dst->e_entry =		src->e_entry;
+	dst->e_phoff =		src->e_phoff;
+	dst->e_shoff =		src->e_shoff;
+	dst->e_flags =		src->e_flags;
+	dst->e_ehsize =		src->e_ehsize;
+	dst->e_phentsize =	src->e_phentsize;
+	dst->e_phnum =		src->e_phnum;
+	dst->e_shentsize =	src->e_shentsize;
+	dst->e_shnum =		src->e_shnum;
+	dst->e_shstrndx =	src->e_shstrndx;
+}
+#endif /* _LP64 */
+
+static void
+restoreexecenv(struct execenv *ep, stack_t *sp)
+{
+	klwp_t *lwp = ttolwp(curthread);
+
+	setexecenv(ep);
+	lwp->lwp_sigaltstack.ss_sp = sp->ss_sp;
+	lwp->lwp_sigaltstack.ss_size = sp->ss_size;
+	lwp->lwp_sigaltstack.ss_flags = sp->ss_flags;
+}
+
+extern int elfexec(vnode_t *, execa_t *, uarg_t *, intpdata_t *, int,
+    long *, int, caddr_t, cred_t *, int *);
+
+extern int elf32exec(struct vnode *, execa_t *, uarg_t *, intpdata_t *, int,
+    long *, int, caddr_t, cred_t *, int *);
+
+/*
+ * Exec routine called by elfexec() to load either 32-bit or 64-bit Linux
+ * binaries.
+ */
+static int
+lx_elfexec(struct vnode *vp, struct execa *uap, struct uarg *args,
+    struct intpdata *idata, int level, long *execsz, int setid,
+    caddr_t exec_file, struct cred *cred, int *brand_action)
+{
+	int		error;
+	vnode_t		*nvp;
+	Ehdr		ehdr;
+	Addr		uphdr_vaddr;
+	intptr_t	voffset;
+	char		*interp = NULL;
+	uintptr_t	ldaddr = NULL;
+	int		i;
+	proc_t		*p = ttoproc(curthread);
+	klwp_t		*lwp = ttolwp(curthread);
+	struct execenv	env;
+	struct execenv	origenv;
+	stack_t		orig_sigaltstack;
+	struct user	*up = PTOU(ttoproc(curthread));
+	lx_elf_data_t	*edp;
+	char		*lib_path = NULL;
+
+	ASSERT(ttoproc(curthread)->p_brand == &lx_brand);
+	ASSERT(ttoproc(curthread)->p_brand_data != NULL);
+
+	edp = &ttolxproc(curthread)->l_elf_data;
+
+	if (args->to_model == DATAMODEL_NATIVE) {
+		lib_path = LX_LIB_PATH;
+	}
+#if defined(_LP64)
+	else {
+		lib_path = LX_LIB_PATH32;
+	}
+#endif
+
+	/*
+	 * Set the brandname and library name for the new process so that
+	 * elfexec() puts them onto the stack.
+	 */
+	args->brandname = LX_BRANDNAME;
+	args->emulator = lib_path;
+
+#if defined(_LP64)
+	/*
+	 * To conform with the way Linux lays out the address space, we clamp
+	 * the stack to be the top of the lower region of the x86-64 canonical
+	 * form address space -- which has the side-effect of laying out the
+	 * entire address space in that lower region.  Note that this only
+	 * matters on 64-bit processes (this value will always be greater than
+	 * the size of a 32-bit address space) and doesn't actually affect
+	 * USERLIMIT:  if a Linux-branded processes wishes to map something
+	 * into the top half of the address space, it can do so -- but with
+	 * the user stack starting at the top of the bottom region, those high
+	 * virtual addresses won't be used unless explicitly directed.
+	 */
+	args->maxstack = lx_maxstack64;
+#endif
+
+	/*
+	 * We will first exec the brand library, then map in the linux
+	 * executable and the linux linker.
+	 */
+	if ((error = lookupname(lib_path, UIO_SYSSPACE, FOLLOW, NULLVPP,
+	    &nvp))) {
+		uprintf("%s: not found.", lib_path);
+		return (error);
+	}
+
+	/*
+	 * We will eventually set the p_exec member to be the vnode for the new
+	 * executable when we call setexecenv(). However, if we get an error
+	 * before that call we need to restore the execenv to its original
+	 * values so that when we return to the caller fop_close() works
+	 * properly while cleaning up from the failed exec().  Restoring the
+	 * original value will also properly decrement the 2nd VN_RELE that we
+	 * took on the brand library.
+	 */
+	origenv.ex_bssbase = p->p_bssbase;
+	origenv.ex_brkbase = p->p_brkbase;
+	origenv.ex_brksize = p->p_brksize;
+	origenv.ex_vp = p->p_exec;
+	orig_sigaltstack.ss_sp = lwp->lwp_sigaltstack.ss_sp;
+	orig_sigaltstack.ss_size = lwp->lwp_sigaltstack.ss_size;
+	orig_sigaltstack.ss_flags = lwp->lwp_sigaltstack.ss_flags;
+
+	if (args->to_model == DATAMODEL_NATIVE) {
+		error = elfexec(nvp, uap, args, idata, level + 1, execsz,
+		    setid, exec_file, cred, brand_action);
+	}
+#if defined(_LP64)
+	else {
+		error = elf32exec(nvp, uap, args, idata, level + 1, execsz,
+		    setid, exec_file, cred, brand_action);
+	}
+#endif
+	VN_RELE(nvp);
+	if (error != 0) {
+		restoreexecenv(&origenv, &orig_sigaltstack);
+		return (error);
+	}
+
+	/*
+	 * exec-ed in the brand library above.
+	 * The u_auxv vectors are now setup by elfexec to point to the
+	 * brand emulation library and its linker.
+	 */
+
+	bzero(&env, sizeof (env));
+
+	/*
+	 * map in the the Linux executable
+	 */
+	if (args->to_model == DATAMODEL_NATIVE) {
+		error = mapexec_brand(vp, args, &ehdr, &uphdr_vaddr,
+		    &voffset, exec_file, &interp, &env.ex_bssbase,
+		    &env.ex_brkbase, &env.ex_brksize, NULL, NULL);
+	}
+#if defined(_LP64)
+	else {
+		Elf32_Ehdr	ehdr32;
+		Elf32_Addr	uphdr_vaddr32;
+
+		error = mapexec32_brand(vp, args, &ehdr32, &uphdr_vaddr32,
+		    &voffset, exec_file, &interp, &env.ex_bssbase,
+		    &env.ex_brkbase, &env.ex_brksize, NULL, NULL);
+
+		Ehdr32to64(&ehdr32, &ehdr);
+
+		if (uphdr_vaddr32 == (Elf32_Addr)-1)
+			uphdr_vaddr = (Addr)-1;
+		else
+			uphdr_vaddr = uphdr_vaddr32;
+	}
+#endif
+	if (error != 0) {
+		restoreexecenv(&origenv, &orig_sigaltstack);
+
+		if (interp != NULL)
+			kmem_free(interp, MAXPATHLEN);
+
+		return (error);
+	}
+
+	/*
+	 * Save off the important properties of the lx executable. The brand
+	 * library will ask us for this data later, when it is ready to set
+	 * things up for the lx executable.
+	 */
+	edp->ed_phdr = (uphdr_vaddr == -1) ? voffset + ehdr.e_phoff :
+	    voffset + uphdr_vaddr;
+	edp->ed_entry = voffset + ehdr.e_entry;
+	edp->ed_phent = ehdr.e_phentsize;
+	edp->ed_phnum = ehdr.e_phnum;
+
+	if (interp != NULL) {
+		if (ehdr.e_type == ET_DYN) {
+			/*
+			 * This is a shared object executable, so we need to
+			 * pick a reasonable place to put the heap. Just don't
+			 * use the first page.
+			 */
+			env.ex_brkbase = (caddr_t)PAGESIZE;
+			env.ex_bssbase = (caddr_t)PAGESIZE;
+		}
+
+		/*
+		 * If the program needs an interpreter (most do), map it in and
+		 * store relevant information about it in the aux vector, where
+		 * the brand library can find it.
+		 */
+		if ((error = lookupname(interp, UIO_SYSSPACE, FOLLOW,
+		    NULLVPP, &nvp))) {
+			uprintf("%s: not found.", interp);
+			restoreexecenv(&origenv, &orig_sigaltstack);
+			kmem_free(interp, MAXPATHLEN);
+			return (error);
+		}
+
+		kmem_free(interp, MAXPATHLEN);
+		interp = NULL;
+
+		/*
+		 * map in the Linux linker
+		 */
+		if (args->to_model == DATAMODEL_NATIVE) {
+			error = mapexec_brand(nvp, args, &ehdr,
+			    &uphdr_vaddr, &voffset, exec_file, NULL, NULL,
+			    NULL, NULL, NULL, &ldaddr);
+		}
+#if defined(_LP64)
+		else {
+			Elf32_Ehdr	ehdr32;
+			Elf32_Addr	uphdr_vaddr32;
+
+			error = mapexec32_brand(nvp, args, &ehdr32,
+			    &uphdr_vaddr32, &voffset, exec_file, NULL, NULL,
+			    NULL, NULL, NULL, &ldaddr);
+
+			Ehdr32to64(&ehdr32, &ehdr);
+
+			if (uphdr_vaddr32 == (Elf32_Addr)-1)
+				uphdr_vaddr = (Addr)-1;
+			else
+				uphdr_vaddr = uphdr_vaddr32;
+		}
+#endif
+
+		VN_RELE(nvp);
+		if (error != 0) {
+			restoreexecenv(&origenv, &orig_sigaltstack);
+			return (error);
+		}
+
+		/*
+		 * Now that we know the base address of the brand's linker,
+		 * we also save this for later use by the brand library.
+		 */
+		edp->ed_base = voffset;
+		edp->ed_ldentry = voffset + ehdr.e_entry;
+	} else {
+		/*
+		 * This program has no interpreter. The lx brand library will
+		 * jump to the address in the AT_SUN_BRAND_LDENTRY aux vector,
+		 * so in this case, put the entry point of the main executable
+		 * there.
+		 */
+		if (ehdr.e_type == ET_EXEC) {
+			/*
+			 * An executable with no interpreter, this must be a
+			 * statically linked executable, which means we loaded
+			 * it at the address specified in the elf header, in
+			 * which case the e_entry field of the elf header is an
+			 * absolute address.
+			 */
+			edp->ed_ldentry = ehdr.e_entry;
+			edp->ed_entry = ehdr.e_entry;
+		} else {
+			/*
+			 * A shared object with no interpreter, we use the
+			 * calculated address from above.
+			 */
+			edp->ed_ldentry = edp->ed_entry;
+
+			/*
+			 * In all situations except an ET_DYN elf object with no
+			 * interpreter, we want to leave the brk and base
+			 * values set by mapexec_brand alone. Normally when
+			 * running ET_DYN objects on Solaris (most likely
+			 * /lib/ld.so.1) the kernel sets brk and base to 0 since
+			 * it doesn't know where to put the heap, and later the
+			 * linker will call brk() to initialize the heap in:
+			 *	usr/src/cmd/sgs/rtld/common/setup.c:setup()
+			 * after it has determined where to put it.  (This
+			 * decision is made after the linker loads and inspects
+			 * elf properties of the target executable being run.)
+			 *
+			 * So for ET_DYN Linux executables, we also don't know
+			 * where the heap should go, so we'll set the brk and
+			 * base to 0.  But in this case the Solaris linker will
+			 * not initialize the heap, so when the Linux linker
+			 * starts running there is no heap allocated.  This
+			 * seems to be ok on Linux 2.4 based systems because the
+			 * Linux linker/libc fall back to using mmap() to
+			 * allocate memory. But on 2.6 systems, running
+			 * applications by specifying them as command line
+			 * arguments to the linker results in segfaults for an
+			 * as yet undetermined reason (which seems to indicatej
+			 * that a more permanent fix for heap initalization in
+			 * these cases may be necessary).
+			 */
+			if (ehdr.e_type == ET_DYN) {
+				env.ex_bssbase = (caddr_t)0;
+				env.ex_brkbase = (caddr_t)0;
+				env.ex_brksize = 0;
+			}
+		}
+
+	}
+
+	env.ex_vp = vp;
+	setexecenv(&env);
+
+	/*
+	 * We try to keep /proc's view of the aux vector consistent with
+	 * what's on the process stack.
+	 */
+	if (args->to_model == DATAMODEL_NATIVE) {
+		auxv_t phdr_auxv[4] = {
+		    { AT_SUN_BRAND_LX_PHDR, 0 },
+		    { AT_SUN_BRAND_LX_INTERP, 0 },
+		    { AT_SUN_BRAND_LX_SYSINFO_EHDR, 0 },
+		    { AT_SUN_BRAND_AUX4, 0 }
+		};
+		phdr_auxv[0].a_un.a_val = edp->ed_phdr;
+		phdr_auxv[1].a_un.a_val = ldaddr;
+		phdr_auxv[2].a_un.a_val = 1;	/* set in lx_init */
+		phdr_auxv[3].a_type = AT_CLKTCK;
+		phdr_auxv[3].a_un.a_val = hz;
+
+		if (copyout(&phdr_auxv, args->auxp_brand,
+		    sizeof (phdr_auxv)) == -1)
+			return (EFAULT);
+	}
+#if defined(_LP64)
+	else {
+		auxv32_t phdr_auxv32[3] = {
+		    { AT_SUN_BRAND_LX_PHDR, 0 },
+		    { AT_SUN_BRAND_LX_INTERP, 0 },
+		    { AT_SUN_BRAND_AUX3, 0 }
+		};
+		phdr_auxv32[0].a_un.a_val = edp->ed_phdr;
+		phdr_auxv32[1].a_un.a_val = ldaddr;
+		phdr_auxv32[2].a_type = AT_CLKTCK;
+		phdr_auxv32[2].a_un.a_val = hz;
+
+		if (copyout(&phdr_auxv32, args->auxp_brand,
+		    sizeof (phdr_auxv32)) == -1)
+			return (EFAULT);
+	}
+#endif
+
+	/*
+	 * /proc uses the AT_ENTRY aux vector entry to deduce
+	 * the location of the executable in the address space. The user
+	 * structure contains a copy of the aux vector that needs to have those
+	 * entries patched with the values of the real lx executable (they
+	 * currently contain the values from the lx brand library that was
+	 * elfexec'd, above).
+	 *
+	 * For live processes, AT_BASE is used to locate the linker segment,
+	 * which /proc and friends will later use to find Solaris symbols
+	 * (such as rtld_db_preinit). However, for core files, /proc uses
+	 * AT_ENTRY to find the right segment to label as the executable.
+	 * So we set AT_ENTRY to be the entry point of the linux executable,
+	 * but leave AT_BASE to be the address of the Solaris linker.
+	 */
+	for (i = 0; i < __KERN_NAUXV_IMPL; i++) {
+		switch (up->u_auxv[i].a_type) {
+		case AT_ENTRY:
+			up->u_auxv[i].a_un.a_val = edp->ed_entry;
+			break;
+
+		case AT_SUN_BRAND_LX_PHDR:
+			up->u_auxv[i].a_un.a_val = edp->ed_phdr;
+			break;
+
+		case AT_SUN_BRAND_LX_INTERP:
+			up->u_auxv[i].a_un.a_val = ldaddr;
+			break;
+
+		default:
+			break;
+		}
+	}
+
+	return (0);
+}
+
+boolean_t
+lx_native_exec(uint8_t osabi, const char **interp)
+{
+	if (osabi != ELFOSABI_SOLARIS)
+		return (B_FALSE);
+
+	/*
+	 * If the process root matches the zone root, prepend /native to the
+	 * interpreter path for native executables.  Absolute precision from
+	 * VN_CMP is not necessary since any change of process root is likely
+	 * to make native binaries inaccessible via /native.
+	 *
+	 * Processes which chroot directly into /native will be able to
+	 * function as expected with no need for the prefix.
+	 */
+	if (VN_CMP(curproc->p_user.u_rdir, curproc->p_zone->zone_rootvp)) {
+		*interp = "/native";
+	}
+
+	return (B_TRUE);
+}
+
+static void
+lx_syscall_init(void)
+{
+	int i;
+
+	/*
+	 * Count up the 32-bit Linux system calls.  Note that lx_sysent32
+	 * has (LX_NSYSCALLS + 1) entries.
+	 */
+	for (i = 0; i <= LX_NSYSCALLS && lx_sysent32[i].sy_name != NULL; i++)
+		continue;
+	lx_nsysent32 = i;
+
+#if defined(_LP64)
+	/*
+	 * Count up the 64-bit Linux system calls.  Note that lx_sysent64
+	 * has (LX_NSYSCALLS + 1) entries.
+	 */
+	for (i = 0; i <= LX_NSYSCALLS && lx_sysent64[i].sy_name != NULL; i++)
+		continue;
+	lx_nsysent64 = i;
+#endif
+}
+
+int
+_init(void)
+{
+	int err = 0;
+
+	lx_syscall_init();
+
+	/* pid/tid conversion hash tables */
+	lx_pid_init();
+
+	/* for lx_ioctl() */
+	lx_ioctl_init();
+
+	/* for lx_futex() */
+	lx_futex_init();
+
+	lx_ptrace_init();
+
+	err = mod_install(&modlinkage);
+	if (err != 0) {
+		cmn_err(CE_WARN, "Couldn't install lx brand module");
+
+		/*
+		 * This looks drastic, but it should never happen.  These
+		 * two data structures should be completely free-able until
+		 * they are used by Linux processes.  Since the brand
+		 * wasn't loaded there should be no Linux processes, and
+		 * thus no way for these data structures to be modified.
+		 */
+		lx_pid_fini();
+		lx_ioctl_fini();
+		if (lx_futex_fini())
+			panic("lx brand module cannot be loaded or unloaded.");
+	}
+	return (err);
+}
+
+int
+_info(struct modinfo *modinfop)
+{
+	return (mod_info(&modlinkage, modinfop));
+}
+
+int
+_fini(void)
+{
+	int err;
+	int futex_done = 0;
+
+	/*
+	 * If there are any zones using this brand, we can't allow it to be
+	 * unloaded.
+	 */
+	if (brand_zone_count(&lx_brand))
+		return (EBUSY);
+
+	lx_ptrace_fini();
+	lx_pid_fini();
+	lx_ioctl_fini();
+
+	if ((err = lx_futex_fini()) != 0)
+		goto done;
+	futex_done = 1;
+
+	err = mod_remove(&modlinkage);
+
+done:
+	if (err) {
+		/*
+		 * If we can't unload the module, then we have to get it
+		 * back into a sane state.
+		 */
+		lx_pid_init();
+
+		if (futex_done)
+			lx_futex_init();
+
+	}
+
+	return (err);
+}
diff --git a/usr/src/uts/common/brand/lx/os/lx_misc.c b/usr/src/uts/common/brand/lx/os/lx_misc.c
new file mode 100644
index 0000000000..4b7310f1d5
--- /dev/null
+++ b/usr/src/uts/common/brand/lx/os/lx_misc.c
@@ -0,0 +1,747 @@
+/*
+ * CDDL HEADER START
+ *
+ * The contents of this file are subject to the terms of the
+ * Common Development and Distribution License (the "License").
+ * You may not use this file except in compliance with the License.
+ *
+ * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
+ * or http://www.opensolaris.org/os/licensing.
+ * See the License for the specific language governing permissions
+ * and limitations under the License.
+ *
+ * When distributing Covered Code, include this CDDL HEADER in each
+ * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
+ * If applicable, add the following below this CDDL HEADER, with the
+ * fields enclosed by brackets "[]" replaced with your own identifying
+ * information: Portions Copyright [yyyy] [name of copyright owner]
+ *
+ * CDDL HEADER END
+ */
+/*
+ * Copyright 2010 Sun Microsystems, Inc.  All rights reserved.
+ * Use is subject to license terms.
+ */
+
+/*
+ * Copyright (c) 2015, Joyent, Inc. All rights reserved.
+ */
+
+#include <sys/errno.h>
+#include <sys/systm.h>
+#include <sys/archsystm.h>
+#include <sys/privregs.h>
+#include <sys/exec.h>
+#include <sys/lwp.h>
+#include <sys/sem.h>
+#include <sys/brand.h>
+#include <sys/lx_brand.h>
+#include <sys/lx_misc.h>
+#include <sys/lx_futex.h>
+#include <sys/cmn_err.h>
+#include <sys/siginfo.h>
+#include <sys/contract/process_impl.h>
+#include <sys/x86_archext.h>
+#include <sys/sdt.h>
+#include <lx_signum.h>
+#include <lx_syscall.h>
+#include <sys/proc.h>
+#include <net/if.h>
+#include <sys/sunddi.h>
+#include <sys/dlpi.h>
+#include <sys/sysmacros.h>
+
+/* Linux specific functions and definitions */
+static void lx_save(klwp_t *);
+static void lx_restore(klwp_t *);
+
+/*
+ * Set the return code for the forked child, always zero
+ */
+/*ARGSUSED*/
+void
+lx_setrval(klwp_t *lwp, int v1, int v2)
+{
+	lwptoregs(lwp)->r_r0 = 0;
+}
+
+/*
+ * Reset process state on exec(2)
+ */
+void
+lx_exec()
+{
+	klwp_t *lwp = ttolwp(curthread);
+	struct lx_lwp_data *lwpd = lwptolxlwp(lwp);
+	proc_t *p = ttoproc(curthread);
+	lx_proc_data_t *pd = ptolxproc(p);
+	struct regs *rp = lwptoregs(lwp);
+
+	/*
+	 * Any l_handler handlers set as a result of B_REGISTER are now
+	 * invalid; clear them.
+	 */
+	pd->l_handler = NULL;
+
+	/*
+	 * If this was a multi-threaded Linux process and this lwp wasn't the
+	 * main lwp, then we need to make its Illumos and Linux PIDs match.
+	 */
+	if (curthread->t_tid != 1) {
+		lx_pid_reassign(curthread);
+	}
+
+	/*
+	 * Inform ptrace(2) that we are processing an execve(2) call so that if
+	 * we are traced we can post either the PTRACE_EVENT_EXEC event or the
+	 * legacy SIGTRAP.
+	 */
+	(void) lx_ptrace_stop_for_option(LX_PTRACE_O_TRACEEXEC, B_FALSE, 0, 0);
+
+	/* clear the fs/gsbase values until the app. can reinitialize them */
+	lwpd->br_lx_fsbase = NULL;
+	lwpd->br_ntv_fsbase = NULL;
+	lwpd->br_lx_gsbase = NULL;
+	lwpd->br_ntv_gsbase = NULL;
+
+	/*
+	 * Clear the native stack flags.  This will be reinitialised by
+	 * lx_init() in the new process image.
+	 */
+	lwpd->br_stack_mode = LX_STACK_MODE_PREINIT;
+	lwpd->br_ntv_stack = 0;
+	lwpd->br_ntv_stack_current = 0;
+
+	installctx(lwptot(lwp), lwp, lx_save, lx_restore, NULL, NULL, lx_save,
+	    NULL);
+
+	/*
+	 * clear out the tls array
+	 */
+	bzero(lwpd->br_tls, sizeof (lwpd->br_tls));
+
+	/*
+	 * reset the tls entries in the gdt
+	 */
+	kpreempt_disable();
+	lx_restore(lwp);
+	kpreempt_enable();
+
+	/*
+	 * The exec syscall doesn't return (so we don't call lx_syscall_return)
+	 * but for our ptrace emulation we need to do this so that a tracer
+	 * does not get out of sync. We know that by the time this lx_exec
+	 * function is called that the exec has succeeded.
+	 */
+	rp->r_r0 = 0;
+	lx_ptrace_stop(LX_PR_SYSEXIT);
+}
+
+void
+lx_exitlwp(klwp_t *lwp)
+{
+	struct lx_lwp_data *lwpd = lwptolxlwp(lwp);
+	proc_t *p = lwptoproc(lwp);
+	kthread_t *t;
+	sigqueue_t *sqp = NULL;
+	pid_t ppid;
+	id_t ptid;
+
+	VERIFY(MUTEX_NOT_HELD(&p->p_lock));
+
+	if (lwpd == NULL)
+		return;		/* second time thru' */
+
+	mutex_enter(&p->p_lock);
+	lx_ptrace_exit(p, lwp);
+	mutex_exit(&p->p_lock);
+
+	if (lwpd->br_robust_list != NULL) {
+		lx_futex_robust_exit((uintptr_t)lwpd->br_robust_list,
+		    lwpd->br_pid);
+	}
+
+	if (lwpd->br_clear_ctidp != NULL) {
+		(void) suword32(lwpd->br_clear_ctidp, 0);
+		(void) lx_futex((uintptr_t)lwpd->br_clear_ctidp, FUTEX_WAKE, 1,
+		    NULL, NULL, 0);
+	}
+
+	if (lwpd->br_signal != 0) {
+		/*
+		 * The first thread in a process doesn't cause a signal to
+		 * be sent when it exits.  It was created by a fork(), not
+		 * a clone(), so the parent should get signalled when the
+		 * process exits.
+		 */
+		if (lwpd->br_ptid == -1)
+			goto free;
+
+		sqp = kmem_zalloc(sizeof (sigqueue_t), KM_SLEEP);
+		/*
+		 * If br_ppid is 0, it means this is a CLONE_PARENT thread,
+		 * so the signal goes to the parent process - not to a
+		 * specific thread in this process.
+		 */
+		p = lwptoproc(lwp);
+		if (lwpd->br_ppid == 0) {
+			mutex_enter(&p->p_lock);
+			ppid = p->p_ppid;
+			t = NULL;
+		} else {
+			/*
+			 * If we have been reparented to init or if our
+			 * parent thread is gone, then nobody gets
+			 * signaled.
+			 */
+			if ((lx_lwp_ppid(lwp, &ppid, &ptid) == 1) ||
+			    (ptid == -1))
+				goto free;
+
+			mutex_enter(&pidlock);
+			if ((p = prfind(ppid)) == NULL || p->p_stat == SIDL) {
+				mutex_exit(&pidlock);
+				goto free;
+			}
+			mutex_enter(&p->p_lock);
+			mutex_exit(&pidlock);
+
+			if ((t = idtot(p, ptid)) == NULL) {
+				mutex_exit(&p->p_lock);
+				goto free;
+			}
+		}
+
+		sqp->sq_info.si_signo = lwpd->br_signal;
+		sqp->sq_info.si_code = lwpd->br_exitwhy;
+		sqp->sq_info.si_status = lwpd->br_exitwhat;
+		sqp->sq_info.si_pid = lwpd->br_pid;
+		sqp->sq_info.si_uid = crgetruid(CRED());
+		sigaddqa(p, t, sqp);
+		mutex_exit(&p->p_lock);
+		sqp = NULL;
+	}
+
+free:
+	if (lwpd->br_scall_args != NULL) {
+		ASSERT(lwpd->br_args_size > 0);
+		kmem_free(lwpd->br_scall_args, lwpd->br_args_size);
+	}
+	if (sqp)
+		kmem_free(sqp, sizeof (sigqueue_t));
+}
+
+void
+lx_freelwp(klwp_t *lwp)
+{
+	struct lx_lwp_data *lwpd = lwptolxlwp(lwp);
+	VERIFY(lwpd != NULL);
+
+	/*
+	 * Remove our system call interposer.
+	 */
+	lwp->lwp_brand_syscall = NULL;
+
+	(void) removectx(lwptot(lwp), lwp, lx_save, lx_restore, NULL, NULL,
+	    lx_save, NULL);
+	if (lwpd->br_pid != 0) {
+		lx_pid_rele(lwptoproc(lwp)->p_pid, lwptot(lwp)->t_tid);
+	}
+
+	/*
+	 * Ensure that lx_ptrace_exit() has been called to detach
+	 * ptrace(2) tracers and tracees.
+	 */
+	VERIFY(lwpd->br_ptrace_tracer == NULL);
+	VERIFY(lwpd->br_ptrace_accord == NULL);
+
+	lwp->lwp_brand = NULL;
+	kmem_free(lwpd, sizeof (struct lx_lwp_data));
+}
+
+void *
+lx_lwpdata_alloc(proc_t *p)
+{
+	lx_lwp_data_t *lwpd;
+	struct lx_pid *lpidp;
+	pid_t newpid = 0;
+	struct pid *pidp = NULL;
+
+	VERIFY(MUTEX_NOT_HELD(&p->p_lock));
+
+	/*
+	 * LWPs beyond the first will require a pid to be allocated to emulate
+	 * Linux's goofy thread model.  While this  allocation may be
+	 * unnecessary when a single-lwp process undergoes branding, it cannot
+	 * be performed during b_initlwp due to p_lock being held.
+	 */
+	if (p->p_lwpcnt > 0) {
+		if ((newpid = pid_allocate(p, 0, 0)) < 0) {
+			return (NULL);
+		}
+		pidp = pid_find(newpid);
+	}
+
+	lwpd = kmem_zalloc(sizeof (struct lx_lwp_data), KM_SLEEP);
+	lpidp = kmem_zalloc(sizeof (struct lx_pid), KM_SLEEP);
+
+	lpidp->l_pid = newpid;
+	lpidp->l_pidp = pidp;
+	lwpd->br_lpid = lpidp;
+	return (lwpd);
+}
+
+/*
+ * Free lwp brand data if an error occurred during lwp_create.
+ * Otherwise, lx_freelwp will be used to free the resources after they're
+ * associated with the lwp via lx_initlwp.
+ */
+void
+lx_lwpdata_free(void *lwpbd)
+{
+	lx_lwp_data_t *lwpd = (lx_lwp_data_t *)lwpbd;
+	VERIFY(lwpd != NULL);
+	VERIFY(lwpd->br_lpid != NULL);
+
+	if (lwpd->br_lpid->l_pidp != NULL) {
+		(void) pid_rele(lwpd->br_lpid->l_pidp);
+	}
+	kmem_free(lwpd->br_lpid, sizeof (*lwpd->br_lpid));
+	kmem_free(lwpd, sizeof (*lwpd));
+}
+
+void
+lx_initlwp(klwp_t *lwp, void *lwpbd)
+{
+	lx_lwp_data_t *lwpd = (lx_lwp_data_t *)lwpbd;
+	lx_lwp_data_t *plwpd = ttolxlwp(curthread);
+	kthread_t *tp = lwptot(lwp);
+	proc_t *p = lwptoproc(lwp);
+
+	VERIFY(MUTEX_HELD(&p->p_lock));
+	VERIFY(lwp->lwp_brand == NULL);
+
+	lwpd->br_exitwhy = CLD_EXITED;
+	lwpd->br_lwp = lwp;
+	lwpd->br_clear_ctidp = NULL;
+	lwpd->br_set_ctidp = NULL;
+	lwpd->br_signal = 0;
+	lwpd->br_stack_mode = LX_STACK_MODE_PREINIT;
+	/*
+	 * lwpd->br_affinitymask was zeroed by kmem_zalloc()
+	 * as was lwpd->br_scall_args and lwpd->br_args_size.
+	 */
+
+	/*
+	 * The first thread in a process has ppid set to the parent
+	 * process's pid, and ptid set to -1.  Subsequent threads in the
+	 * process have their ppid set to the pid of the thread that
+	 * created them, and their ptid to that thread's tid.
+	 */
+	if (tp->t_next == tp) {
+		lwpd->br_ppid = tp->t_procp->p_ppid;
+		lwpd->br_ptid = -1;
+	} else if (plwpd != NULL) {
+		bcopy(plwpd->br_tls, lwpd->br_tls, sizeof (lwpd->br_tls));
+		lwpd->br_ppid = plwpd->br_pid;
+		lwpd->br_ptid = curthread->t_tid;
+		/* The child inherits the fs/gsbase values from the parent */
+		lwpd->br_lx_fsbase = plwpd->br_lx_fsbase;
+		lwpd->br_ntv_fsbase = plwpd->br_ntv_fsbase;
+		lwpd->br_lx_gsbase = plwpd->br_lx_gsbase;
+		lwpd->br_ntv_gsbase = plwpd->br_ntv_gsbase;
+	} else {
+		/*
+		 * Oddball case: the parent thread isn't a Linux process.
+		 */
+		lwpd->br_ppid = 0;
+		lwpd->br_ptid = -1;
+	}
+	lwp->lwp_brand = lwpd;
+
+	/*
+	 * When during lx_lwpdata_alloc, we must decide whether or not to
+	 * allocate a new pid to associate with the lwp. Since p_lock is not
+	 * held at that point, the only time we can guarantee a new pid isn't
+	 * needed is when p_lwpcnt == 0.  This is because other lwps won't be
+	 * present to race with us with regards to pid allocation.
+	 *
+	 * This means that in all other cases (where p_lwpcnt > 0), we expect
+	 * that lx_lwpdata_alloc will allocate a pid for us to use here, even
+	 * if it is uneeded.  If this process is undergoing an exec, for
+	 * example, the single existing lwp will not need a new pid when it is
+	 * rebranded.  In that case, lx_pid_assign will free the uneeded pid.
+	 */
+	VERIFY(lwpd->br_lpid->l_pidp != NULL || p->p_lwpcnt == 0);
+
+	lx_pid_assign(tp, lwpd->br_lpid);
+	lwpd->br_tgid = lwpd->br_pid;
+	/*
+	 * Having performed the lx pid assignement, the lpid reference is no
+	 * longer needed.  The underlying data will be freed during lx_freelwp.
+	 */
+	lwpd->br_lpid = NULL;
+
+	installctx(lwptot(lwp), lwp, lx_save, lx_restore, NULL, NULL,
+	    lx_save, NULL);
+
+	/*
+	 * Install branded system call hook for this LWP:
+	 */
+	lwp->lwp_brand_syscall = lx_syscall_enter;
+
+	/*
+	 * If the parent LWP has a ptrace(2) tracer, the new LWP may
+	 * need to inherit that same tracer.
+	 */
+	if (plwpd != NULL) {
+		lx_ptrace_inherit_tracer(plwpd, lwpd);
+	}
+}
+
+/*
+ * There is no need to have any locking for either the source or
+ * destination struct lx_lwp_data structs.  This is always run in the
+ * thread context of the source thread, and the destination thread is
+ * always newly created and not referred to from anywhere else.
+ */
+void
+lx_forklwp(klwp_t *srclwp, klwp_t *dstlwp)
+{
+	struct lx_lwp_data *src = srclwp->lwp_brand;
+	struct lx_lwp_data *dst = dstlwp->lwp_brand;
+
+	dst->br_ppid = src->br_pid;
+	dst->br_ptid = lwptot(srclwp)->t_tid;
+	bcopy(src->br_tls, dst->br_tls, sizeof (dst->br_tls));
+
+	switch (src->br_stack_mode) {
+	case LX_STACK_MODE_BRAND:
+	case LX_STACK_MODE_NATIVE:
+		/*
+		 * The parent LWP has an alternate stack installed.
+		 * The child LWP should have the same stack base and extent.
+		 */
+		dst->br_stack_mode = src->br_stack_mode;
+		dst->br_ntv_stack = src->br_ntv_stack;
+		dst->br_ntv_stack_current = src->br_ntv_stack_current;
+		break;
+
+	default:
+		/*
+		 * Otherwise, clear the stack data for this LWP.
+		 */
+		dst->br_stack_mode = LX_STACK_MODE_PREINIT;
+		dst->br_ntv_stack = 0;
+		dst->br_ntv_stack_current = 0;
+	}
+
+	/*
+	 * copy only these flags
+	 */
+	dst->br_lwp_flags = src->br_lwp_flags & BR_CPU_BOUND;
+	dst->br_scall_args = NULL;
+}
+
+/*
+ * When switching a Linux process off the CPU, clear its GDT entries.
+ */
+/* ARGSUSED */
+static void
+lx_save(klwp_t *t)
+{
+	int i;
+
+#if defined(__amd64)
+	reset_sregs();
+#endif
+	for (i = 0; i < LX_TLSNUM; i++)
+		gdt_update_usegd(GDT_TLSMIN + i, &null_udesc);
+}
+
+/*
+ * When switching a Linux process on the CPU, set its GDT entries.
+ *
+ * For 64-bit code we don't have to worry about explicitly setting the
+ * %fsbase via wrmsr(MSR_AMD_FSBASE) here. Instead, that should happen
+ * automatically in update_sregs if we are executing in user-land. If this
+ * is the case then pcb_rupdate should be set.
+ */
+static void
+lx_restore(klwp_t *t)
+{
+	struct lx_lwp_data *lwpd = lwptolxlwp(t);
+	user_desc_t *tls;
+	int i;
+
+	ASSERT(lwpd);
+
+	tls = lwpd->br_tls;
+	for (i = 0; i < LX_TLSNUM; i++)
+		gdt_update_usegd(GDT_TLSMIN + i, &tls[i]);
+}
+
+void
+lx_set_gdt(int entry, user_desc_t *descrp)
+{
+
+	gdt_update_usegd(entry, descrp);
+}
+
+void
+lx_clear_gdt(int entry)
+{
+	gdt_update_usegd(entry, &null_udesc);
+}
+
+longlong_t
+lx_nosys()
+{
+	return (set_errno(ENOSYS));
+}
+
+/*
+ * Brand-specific routine to check if given non-Solaris standard segment
+ * register values should be modified to other values.
+ */
+/*ARGSUSED*/
+greg_t
+lx_fixsegreg(greg_t sr, model_t datamodel)
+{
+	uint16_t idx = SELTOIDX(sr);
+
+	ASSERT(sr == (sr & 0xffff));
+
+	/*
+	 * If the segment selector is a valid TLS selector, just return it.
+	 */
+	if (!SELISLDT(sr) && idx >= GDT_TLSMIN && idx <= GDT_TLSMAX)
+		return (sr | SEL_UPL);
+
+	/*
+	 * Force the SR into the LDT in ring 3 for 32-bit processes.
+	 *
+	 * 64-bit processes get the null GDT selector since they are not
+	 * allowed to have a private LDT.
+	 */
+#if defined(__amd64)
+	return (datamodel == DATAMODEL_ILP32 ? (sr | SEL_TI_LDT | SEL_UPL) : 0);
+#elif defined(__i386)
+	datamodel = datamodel;  /* datamodel currently unused for 32-bit */
+	return (sr | SEL_TI_LDT | SEL_UPL);
+#endif	/* __amd64 */
+}
+
+/*
+ * Brand-specific function to convert the fsbase as pulled from the register
+ * into a native fsbase suitable for locating the ulwp_t from the kernel.
+ */
+uintptr_t
+lx_fsbase(klwp_t *lwp, uintptr_t fsbase)
+{
+	lx_lwp_data_t *lwpd = lwp->lwp_brand;
+
+	if (lwpd->br_stack_mode != LX_STACK_MODE_BRAND ||
+	    lwpd->br_ntv_fsbase == NULL) {
+		return (fsbase);
+	}
+
+	return (lwpd->br_ntv_fsbase);
+}
+
+/*
+ * These two functions simulate winfo and post_sigcld for the lx brand. The
+ * difference is delivering a designated signal as opposed to always SIGCLD.
+ */
+static void
+lx_winfo(proc_t *pp, k_siginfo_t *ip, struct lx_proc_data *dat)
+{
+	ASSERT(MUTEX_HELD(&pidlock));
+	bzero(ip, sizeof (k_siginfo_t));
+	ip->si_signo = ltos_signo[dat->l_signal];
+	ip->si_code = pp->p_wcode;
+	ip->si_pid = pp->p_pid;
+	ip->si_ctid = PRCTID(pp);
+	ip->si_zoneid = pp->p_zone->zone_id;
+	ip->si_status = pp->p_wdata;
+	ip->si_stime = pp->p_stime;
+	ip->si_utime = pp->p_utime;
+}
+
+static void
+lx_post_exit_sig(proc_t *cp, sigqueue_t *sqp, struct lx_proc_data *dat)
+{
+	proc_t *pp = cp->p_parent;
+
+	ASSERT(MUTEX_HELD(&pidlock));
+	mutex_enter(&pp->p_lock);
+	/*
+	 * Since Linux doesn't queue SIGCHLD, or any other non RT
+	 * signals, we just blindly deliver whatever signal we can.
+	 */
+	ASSERT(sqp != NULL);
+	lx_winfo(cp, &sqp->sq_info, dat);
+	sigaddqa(pp, NULL, sqp);
+	sqp = NULL;
+	mutex_exit(&pp->p_lock);
+}
+
+
+/*
+ * Brand specific code for exiting and sending a signal to the parent, as
+ * opposed to sigcld().
+ */
+void
+lx_exit_with_sig(proc_t *cp, sigqueue_t *sqp)
+{
+	proc_t *pp = cp->p_parent;
+	lx_proc_data_t *lx_brand_data = ptolxproc(cp);
+	ASSERT(MUTEX_HELD(&pidlock));
+
+	switch (cp->p_wcode) {
+	case CLD_EXITED:
+	case CLD_DUMPED:
+	case CLD_KILLED:
+			ASSERT(cp->p_stat == SZOMB);
+			/*
+			 * The broadcast on p_srwchan_cv is a kludge to
+			 * wakeup a possible thread in uadmin(A_SHUTDOWN).
+			 */
+			cv_broadcast(&cp->p_srwchan_cv);
+
+			/*
+			 * Add to newstate list of the parent
+			 */
+			add_ns(pp, cp);
+
+			cv_broadcast(&pp->p_cv);
+			if ((pp->p_flag & SNOWAIT) ||
+			    PTOU(pp)->u_signal[SIGCLD - 1] == SIG_IGN) {
+				if (!(cp->p_pidflag & CLDWAITPID))
+					freeproc(cp);
+			} else if (!(cp->p_pidflag & CLDNOSIGCHLD) &&
+			    lx_brand_data->l_signal != 0) {
+				lx_post_exit_sig(cp, sqp, lx_brand_data);
+				sqp = NULL;
+			}
+			break;
+
+	case CLD_STOPPED:
+	case CLD_CONTINUED:
+	case CLD_TRAPPED:
+			panic("Should not be called in this case");
+	}
+
+	if (sqp)
+		siginfofree(sqp);
+}
+
+/*
+ * Filters based on arguments that have been passed in by a separate syscall
+ * using the B_STORE_ARGS mechanism. if the __WALL flag is set, no filter is
+ * applied, otherwise we look at the difference between a clone and non-clone
+ * process.
+ * The definition of a clone process in Linux is a thread that does not deliver
+ * SIGCHLD to its parent. The option __WCLONE indicates to wait only on clone
+ * processes. Without that option, a process should only wait on normal
+ * children. The following table shows the cases.
+ *
+ *                   default    __WCLONE
+ *   no SIGCHLD      -           X
+ *   SIGCHLD         X           -
+ *
+ * This is an XOR of __WCLONE being set, and SIGCHLD being the signal sent on
+ * process exit.
+ *
+ * More information on wait in lx brands can be found at
+ * usr/src/lib/brand/lx/lx_brand/common/wait.c.
+ */
+boolean_t
+lx_wait_filter(proc_t *pp, proc_t *cp)
+{
+	lx_lwp_data_t *lwpd = ttolxlwp(curthread);
+	int flags = lwpd->br_waitid_flags;
+	boolean_t ret;
+
+	if (!lwpd->br_waitid_emulate) {
+		return (B_TRUE);
+	}
+
+	mutex_enter(&cp->p_lock);
+	if (flags & LX_WALL) {
+		ret = B_TRUE;
+	} else {
+		lx_proc_data_t *pd = ptolxproc(cp);
+		boolean_t is_sigchld = B_TRUE;
+		boolean_t match_wclone = B_FALSE;
+
+		/*
+		 * When calling clone, an alternate signal can be chosen to
+		 * deliver to the parent when the child exits.
+		 */
+		if (pd != NULL && pd->l_signal != stol_signo[SIGCHLD]) {
+			is_sigchld = B_FALSE;
+		}
+		if ((flags & LX_WCLONE) != 0) {
+			match_wclone = B_TRUE;
+		}
+
+		ret = (match_wclone ^ is_sigchld) ? B_TRUE : B_FALSE;
+	}
+	mutex_exit(&cp->p_lock);
+
+	return (ret);
+}
+
+void
+lx_ifname_convert(char *ifname, lx_if_action_t act)
+{
+	if (act == LX_IF_TONATIVE) {
+		if (strncmp(ifname, "lo", IFNAMSIZ) == 0)
+			(void) strlcpy(ifname, "lo0", IFNAMSIZ);
+	} else {
+		if (strncmp(ifname, "lo0", IFNAMSIZ) == 0)
+			(void) strlcpy(ifname, "lo", IFNAMSIZ);
+	}
+}
+
+void
+lx_ifflags_convert(uint64_t *flags, lx_if_action_t act)
+{
+	uint64_t buf;
+
+	buf = *flags & (IFF_UP | IFF_BROADCAST | IFF_DEBUG |
+	    IFF_LOOPBACK | IFF_POINTOPOINT | IFF_NOTRAILERS |
+	    IFF_RUNNING | IFF_NOARP | IFF_PROMISC | IFF_ALLMULTI);
+
+	/* Linux has different shift for multicast flag */
+	if (act == LX_IF_TONATIVE) {
+		if (*flags & 0x1000)
+			buf |= IFF_MULTICAST;
+	} else {
+		if (*flags & IFF_MULTICAST)
+			buf |= 0x1000;
+	}
+	*flags = buf;
+}
+
+
+void
+lx_stol_hwaddr(const struct sockaddr_dl *src, struct sockaddr *dst, int *size)
+{
+	int copy_size = MIN(src->sdl_alen, sizeof (dst->sa_data));
+
+	switch (src->sdl_type) {
+	case DL_ETHER:
+		dst->sa_family = LX_ARPHRD_ETHER;
+		break;
+	case DL_LOOP:
+		dst->sa_family = LX_ARPHRD_LOOPBACK;
+		break;
+	default:
+		dst->sa_family = LX_ARPHRD_VOID;
+	}
+
+	bcopy(LLADDR(src), dst->sa_data, copy_size);
+	*size = copy_size;
+}
diff --git a/usr/src/uts/common/brand/lx/os/lx_pid.c b/usr/src/uts/common/brand/lx/os/lx_pid.c
new file mode 100644
index 0000000000..40179bbdaf
--- /dev/null
+++ b/usr/src/uts/common/brand/lx/os/lx_pid.c
@@ -0,0 +1,395 @@
+/*
+ * CDDL HEADER START
+ *
+ * The contents of this file are subject to the terms of the
+ * Common Development and Distribution License (the "License").
+ * You may not use this file except in compliance with the License.
+ *
+ * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
+ * or http://www.opensolaris.org/os/licensing.
+ * See the License for the specific language governing permissions
+ * and limitations under the License.
+ *
+ * When distributing Covered Code, include this CDDL HEADER in each
+ * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
+ * If applicable, add the following below this CDDL HEADER, with the
+ * fields enclosed by brackets "[]" replaced with your own identifying
+ * information: Portions Copyright [yyyy] [name of copyright owner]
+ *
+ * CDDL HEADER END
+ */
+
+/*
+ * Copyright 2009 Sun Microsystems, Inc.  All rights reserved.
+ * Use is subject to license terms.
+ * Copyright 2015, Joyent, Inc.
+ */
+
+#include <sys/types.h>
+#include <sys/param.h>
+#include <sys/sysmacros.h>
+#include <sys/bitmap.h>
+#include <sys/var.h>
+#include <sys/thread.h>
+#include <sys/proc.h>
+#include <sys/brand.h>
+#include <sys/zone.h>
+#include <sys/lx_brand.h>
+
+#define	LINUX_PROC_FACTOR	8	/* factor down the hash table by this */
+static int hash_len = 4;		/* desired average hash chain length */
+static int hash_size;			/* no of buckets in the hash table */
+
+static struct lx_pid **stol_pid_hash;
+static struct lx_pid **ltos_pid_hash;
+
+#define	LTOS_HASH(pid)		((pid) & (hash_size - 1))
+#define	STOL_HASH(pid, tid)	(((pid) + (tid)) & (hash_size - 1))
+
+static kmutex_t hash_lock;
+
+static void
+lx_pid_insert_hash(struct lx_pid *lpidp)
+{
+	int shash = STOL_HASH(lpidp->s_pid, lpidp->s_tid);
+	int lhash = LTOS_HASH(lpidp->l_pid);
+
+	ASSERT(MUTEX_HELD(&hash_lock));
+
+	lpidp->stol_next = stol_pid_hash[shash];
+	stol_pid_hash[shash] = lpidp;
+
+	lpidp->ltos_next = ltos_pid_hash[lhash];
+	ltos_pid_hash[lhash] = lpidp;
+}
+
+static struct lx_pid *
+lx_pid_remove_hash(pid_t pid, id_t tid)
+{
+	struct lx_pid **hpp;
+	struct lx_pid *lpidp = NULL;
+
+	ASSERT(MUTEX_HELD(&hash_lock));
+
+	hpp = &stol_pid_hash[STOL_HASH(pid, tid)];
+	while (*hpp) {
+		if ((*hpp)->s_pid == pid && (*hpp)->s_tid == tid) {
+			lpidp = *hpp;
+			*hpp = (*hpp)->stol_next;
+			break;
+		}
+		hpp = &(*hpp)->stol_next;
+	}
+
+	/*
+	 * when called during error recovery the pid may already
+	 * be released
+	 */
+	if (lpidp == NULL)
+		return (NULL);
+
+	hpp = &ltos_pid_hash[LTOS_HASH(lpidp->l_pid)];
+	while (*hpp) {
+		if (*hpp == lpidp) {
+			*hpp = lpidp->ltos_next;
+			break;
+		}
+		hpp = &(*hpp)->ltos_next;
+	}
+
+	return (lpidp);
+}
+
+/*
+ * given a solaris pid/tid pair, create a linux pid
+ */
+void
+lx_pid_assign(kthread_t *t, struct lx_pid *lpidp)
+{
+	proc_t *p = ttoproc(t);
+	lx_lwp_data_t *lwpd = ttolxlwp(t);
+	pid_t s_pid = p->p_pid;
+	id_t s_tid = t->t_tid;
+
+	/*
+	 * When lx_initlwp is called from lx_setbrand, p_lwpcnt will already be
+	 * equal to 1. Since lx_initlwp is being called against an lwp that
+	 * already exists, an additional pid allocation is not necessary.
+	 *
+	 * We check for this by testing br_ppid == 0.
+	 */
+	if (p->p_lwpcnt > 0 && lwpd->br_ppid != 0) {
+		/*
+		 * Assign allocated pid to any thread other than the first.
+		 * The l_pid and l_pidp fields should be populated.
+		 */
+		VERIFY(lpidp->l_pidp != NULL);
+		VERIFY(lpidp->l_pid != 0);
+	} else {
+		/*
+		 * There are cases where a pid is speculatively allocated but
+		 * is not needed.  We are obligated to free it here.
+		 */
+		if (lpidp->l_pidp != NULL) {
+			(void) pid_rele(lpidp->l_pidp);
+		}
+		lpidp->l_pidp = NULL;
+		lpidp->l_pid = s_pid;
+	}
+
+	lpidp->s_pid = s_pid;
+	lpidp->s_tid = s_tid;
+	lpidp->l_start = t->t_start;
+
+	/*
+	 * now put the pid into the linux-solaris and solaris-linux
+	 * conversion hash tables
+	 */
+	mutex_enter(&hash_lock);
+	lx_pid_insert_hash(lpidp);
+	mutex_exit(&hash_lock);
+
+	lwpd->br_pid = lpidp->l_pid;
+}
+
+/*
+ * If we are exec()ing the process, this thread's tid is about to be reset
+ * to 1.  Make sure the Linux PID bookkeeping reflects that change.
+ */
+void
+lx_pid_reassign(kthread_t *t)
+{
+	proc_t *p = ttoproc(t);
+	struct pid *old_pidp;
+	struct lx_pid *lpidp;
+
+	ASSERT(p->p_lwpcnt == 1);
+
+	mutex_enter(&hash_lock);
+
+	/*
+	 * Clean up all the traces of this thread's 'fake' Linux PID.
+	 */
+	lpidp = lx_pid_remove_hash(p->p_pid, t->t_tid);
+	ASSERT(lpidp != NULL);
+	old_pidp = lpidp->l_pidp;
+	lpidp->l_pidp = NULL;
+
+	/*
+	 * Now register this thread as (pid, 1).
+	 */
+	lpidp->l_pid = p->p_pid;
+	lpidp->s_pid = p->p_pid;
+	lpidp->s_tid = 1;
+	lx_pid_insert_hash(lpidp);
+
+	mutex_exit(&hash_lock);
+
+	if (old_pidp)
+		(void) pid_rele(old_pidp);
+}
+
+/*
+ * release a solaris pid/tid pair
+ */
+void
+lx_pid_rele(pid_t pid, id_t tid)
+{
+	struct lx_pid *lpidp;
+
+	mutex_enter(&hash_lock);
+	lpidp = lx_pid_remove_hash(pid, tid);
+	mutex_exit(&hash_lock);
+
+	if (lpidp) {
+		if (lpidp->l_pidp)
+			(void) pid_rele(lpidp->l_pidp);
+
+		kmem_free(lpidp, sizeof (*lpidp));
+	}
+}
+
+/*
+ * given a linux pid, return the solaris pid/tid pair
+ */
+int
+lx_lpid_to_spair(pid_t l_pid, pid_t *s_pid, id_t *s_tid)
+{
+	struct lx_pid *hp;
+
+	if (l_pid == 1) {
+		pid_t initpid;
+
+		/*
+		 * We are trying to look up the Linux init process for the
+		 * current zone, which we pretend has pid 1.
+		 */
+		if ((initpid = curzone->zone_proc_initpid) == -1) {
+			/*
+			 * We could not find the init process for this zone.
+			 */
+			return (-1);
+		}
+
+		if (s_pid != NULL)
+			*s_pid = initpid;
+		if (s_tid != NULL)
+			*s_tid = 1;
+
+		return (0);
+	}
+
+	mutex_enter(&hash_lock);
+	for (hp = ltos_pid_hash[LTOS_HASH(l_pid)]; hp; hp = hp->ltos_next) {
+		if (l_pid == hp->l_pid) {
+			if (s_pid)
+				*s_pid = hp->s_pid;
+			if (s_tid)
+				*s_tid = hp->s_tid;
+			break;
+		}
+	}
+	mutex_exit(&hash_lock);
+	if (hp != NULL)
+		return (0);
+
+	/*
+	 * We didn't find this pid in our translation table.
+	 * But this still could be the pid of a native process
+	 * running in the current zone so check for that here.
+	 *
+	 * Note that prfind() only searches for processes in the current zone.
+	 */
+	mutex_enter(&pidlock);
+	if (prfind(l_pid) != NULL) {
+		mutex_exit(&pidlock);
+		if (s_pid)
+			*s_pid = l_pid;
+		if (s_tid)
+			*s_tid = 0;
+		return (0);
+	}
+	mutex_exit(&pidlock);
+
+	return (-1);
+}
+
+/*
+ * Given an lwp, return the Linux pid of its parent.  If the caller
+ * wants them, we return the Solaris (pid, tid) as well.
+ */
+pid_t
+lx_lwp_ppid(klwp_t *lwp, pid_t *ppidp, id_t *ptidp)
+{
+	lx_lwp_data_t *lwpd = lwptolxlwp(lwp);
+	proc_t *p = lwptoproc(lwp);
+	struct lx_pid *hp;
+	pid_t zoneinit = curproc->p_zone->zone_proc_initpid;
+	pid_t lppid, ppid;
+
+	/*
+	 * Be sure not to return a parent pid that should be invisible
+	 * within this zone.
+	 */
+	ppid = ((p->p_flag & SZONETOP)
+	    ? curproc->p_zone->zone_zsched->p_pid : p->p_ppid);
+
+	/*
+	 * If the parent process's pid is the zone's init process, force it
+	 * to the Linux init pid value of 1.
+	 */
+	if (ppid == zoneinit)
+		ppid = 1;
+
+	/*
+	 * There are two cases in which the Linux definition of a 'parent'
+	 * matches that of Solaris:
+	 *
+	 * - if our tgid is the same as our PID, then we are either the
+	 *   first thread in the process or a CLONE_THREAD thread.
+	 *
+	 * - if the brand lwp value for ppid is 0, then we are either the
+	 *   child of a differently-branded process or a CLONE_PARENT thread.
+	 */
+	if (p->p_pid == lwpd->br_tgid || lwpd->br_ppid == 0) {
+		if (ppidp != NULL)
+			*ppidp = ppid;
+		if (ptidp != NULL)
+			*ptidp = -1;
+		return (ppid);
+	}
+
+	/*
+	 * Set the default Linux parent pid to be the pid of the zone's init
+	 * process; this will get converted back to the Linux default of 1
+	 * later.
+	 */
+	lppid = zoneinit;
+
+	/*
+	 * If the process's parent isn't init, try and look up the Linux "pid"
+	 * corresponding to the process's parent.
+	 */
+	if (ppid != 1) {
+		/*
+		 * In all other cases, we are looking for the parent of this
+		 * specific thread, which in Linux refers to the thread that
+		 * clone()d it.   We stashed that thread's PID away when this
+		 * thread was created.
+		 */
+		mutex_enter(&hash_lock);
+		for (hp = ltos_pid_hash[LTOS_HASH(lwpd->br_ppid)]; hp;
+		    hp = hp->ltos_next) {
+			if (lwpd->br_ppid == hp->l_pid) {
+				/*
+				 * We found the PID we were looking for, but
+				 * since we cached its value in this LWP's brand
+				 * structure, it has exited and been reused by
+				 * another process.
+				 */
+				if (hp->l_start > lwptot(lwp)->t_start)
+					break;
+
+				lppid = lwpd->br_ppid;
+				if (ppidp != NULL)
+					*ppidp = hp->s_pid;
+				if (ptidp != NULL)
+					*ptidp = hp->s_tid;
+
+				break;
+			}
+		}
+		mutex_exit(&hash_lock);
+	}
+
+	if (lppid == zoneinit) {
+		lppid = 1;
+
+		if (ppidp != NULL)
+			*ppidp = lppid;
+		if (ptidp != NULL)
+			*ptidp = -1;
+	}
+
+	return (lppid);
+}
+
+void
+lx_pid_init(void)
+{
+	hash_size = 1 << highbit(v.v_proc / (hash_len * LINUX_PROC_FACTOR));
+
+	stol_pid_hash = kmem_zalloc(sizeof (struct lx_pid *) * hash_size,
+	    KM_SLEEP);
+	ltos_pid_hash = kmem_zalloc(sizeof (struct lx_pid *) * hash_size,
+	    KM_SLEEP);
+
+	mutex_init(&hash_lock, NULL, MUTEX_DEFAULT, NULL);
+}
+
+void
+lx_pid_fini(void)
+{
+	kmem_free(stol_pid_hash, sizeof (struct lx_pid *) * hash_size);
+	kmem_free(ltos_pid_hash, sizeof (struct lx_pid *) * hash_size);
+}
diff --git a/usr/src/uts/common/brand/lx/os/lx_ptrace.c b/usr/src/uts/common/brand/lx/os/lx_ptrace.c
new file mode 100644
index 0000000000..1659da56ff
--- /dev/null
+++ b/usr/src/uts/common/brand/lx/os/lx_ptrace.c
@@ -0,0 +1,2377 @@
+/*
+ * This file and its contents are supplied under the terms of the
+ * Common Development and Distribution License ("CDDL"), version 1.0.
+ * You may only use this file in accordance with the terms of version
+ * 1.0 of the CDDL.
+ *
+ * A full copy of the text of the CDDL should have accompanied this
+ * source.  A copy of the CDDL is also available via the Internet at
+ * http://www.illumos.org/license/CDDL.
+ */
+
+/*
+ * Copyright 2015 Joyent, Inc.
+ */
+
+/*
+ * Emulation of the Linux ptrace(2) interface.
+ *
+ * OVERVIEW
+ *
+ * The Linux process model is somewhat different from the illumos native
+ * model.  One critical difference is that each Linux thread has a unique
+ * identifier in the pid namespace.  The lx brand assigns a pid to each LWP
+ * within the emulated process, giving the pid of the process itself to the
+ * first LWP.
+ *
+ * The Linux ptrace(2) interface allows for any LWP in a branded process to
+ * exert control over any other LWP within the same zone.  Control is exerted
+ * by the use of the ptrace(2) system call itself, which accepts a number of
+ * request codes.  Feedback on traced events is primarily received by the
+ * tracer through SIGCLD and the emulated waitpid(2) and waitid(2) system
+ * calls.  Many of the possible ptrace(2) requests will only succeed if the
+ * target LWP is in a "ptrace-stop" condition.
+ *
+ * HISTORY
+ *
+ * The brand support for ptrace(2) was originally built on top of the rich
+ * support for debugging and tracing provided through the illumos /proc
+ * interfaces, mounted at /native/proc within the zone.  The native legacy
+ * ptrace(3C) functionality was used as a starting point, but was generally
+ * insufficient for complete and precise emulation.  The extant legacy
+ * interface, and indeed our native SIGCLD and waitid(2) facilities, are
+ * focused on _process_ level concerns -- the Linux interface has been
+ * extended to be aware of LWPs as well.
+ *
+ * In order to allow us to focus on providing more complete and accurate
+ * emulation without extensive and undesirable changes to the native
+ * facilities, this second generation ptrace(2) emulation is mostly separate
+ * from any other tracing or debugging framework in the system.
+ *
+ * ATTACHING TRACERS TO TRACEES
+ *
+ * There are several ways that a child LWP may becomed traced by a tracer.
+ * To determine which attach method caused a tracee to become attached, one
+ * may inspect the "br_ptrace_attach" member of the LWP-specific brand data
+ * with the debugger.
+ *
+ * The first attach methods to consider are the attaching ptrace(2) requests:
+ *
+ *   PTRACE_TRACEME
+ *
+ *   If an LWP makes a PTRACE_TRACEME call, it will be attached as a tracee
+ *   to its parent LWP (br_ppid).  Using PTRACE_TRACEME does _not_ cause the
+ *   tracee to be held in a stop condition.  It is common practice for
+ *   consumers to raise(SIGSTOP) immediately afterward.
+ *
+ *   PTRACE_ATTACH
+ *
+ *   An LWP may attempt to trace any other LWP in this, or another, process.
+ *   We currently allow any attach where the process containing the tracer
+ *   LWP has permission to write to /proc for the process containing the
+ *   intended tracer.  This action also sends a SIGSTOP to the newly attached
+ *   tracee.
+ *
+ * The second class of attach methods are the clone(2)/fork(2) inheritance
+ * options that may be set on a tracee with PTRACE_SETOPTIONS:
+ *
+ *   PTRACE_O_TRACEFORK, PTRACE_O_TRACEVFORK and PTRACE_O_TRACECLONE
+ *
+ *   If these options have been set on a tracee, then a fork(2), vfork(2) or
+ *   clone(2) respectively will cause the newly created LWP to be traced by
+ *   the same tracer.  The same set of ptrace(2) options will also be set on
+ *   the new child.
+ *
+ * The third class of attach method is the PTRACE_CLONE flag to clone(2).
+ * This flag induces the same inheritance as PTRACE_O_TRACECLONE, but is
+ * passed by the tracee as an argument to clone(2).
+ *
+ * DETACHING TRACEES
+ *
+ * Tracees can be detached by the tracer with the PTRACE_DETACH request.
+ * This request is only valid when the tracee is in a ptrace(2) stop
+ * condition, and is itself a restarting action.
+ *
+ * If the tracer exits without detaching all of its tracees, then all of the
+ * tracees are automatically detached and restarted.  If a tracee was in
+ * "signal-delivery-stop" at the time the tracer exited, the signal will be
+ * released to the child unless it is a SIGSTOP.  We drop this instance of
+ * SIGSTOP in order to prevent the child from becoming stopped by job
+ * control.
+ *
+ * ACCORD ALLOCATION AND MANAGEMENT
+ *
+ * The "lx_ptrace_accord_t" object tracks the agreement between a tracer LWP
+ * and zero or more tracee LWPs.  It is explicitly illegal for a tracee to
+ * trace its tracer, and we block this in PTRACE_ATTACH/PTRACE_TRACEME.
+ *
+ * An LWP starts out without an accord.  If a child of that LWP calls
+ * ptrace(2) with the PTRACE_TRACEME subcommand, or if the LWP itself uses
+ * PTRACE_ATTACH, an accord will be allocated and stored on that LWP.  The
+ * accord structure is not released from that LWP until it arrives in
+ * lx_exitlwp(), as called by lwp_exit().  A new accord will not be
+ * allocated, even if one does not exist, once an LWP arrives in lx_exitlwp()
+ * and sets the LX_PTRACE_EXITING flag.  An LWP will have at most one accord
+ * structure throughout its entire lifecycle; once it has one, it has the
+ * same one until death.
+ *
+ * The accord is reference counted (lxpa_refcnt), starting at a count of one
+ * at creation to represent the link from the tracer LWP to its accord.  The
+ * accord is not freed until the reference count falls to zero.
+ *
+ * To make mutual exclusion between a detaching tracer and various notifying
+ * tracees simpler, the tracer will hold "pidlock" while it clears the
+ * accord members that point back to the tracer LWP and CV.
+ *
+ * SIGNALS AND JOB CONTROL
+ *
+ * Various actions, either directly ptrace(2) related or commonly associated
+ * with tracing, cause process- or thread-directed SIGSTOP signals to be sent
+ * to tracees.  These signals, and indeed any signal other than SIGKILL, can
+ * be suppressed by the tracer when using a restarting request (including
+ * PTRACE_DETACH) on a child.  The signal may also be substituted for a
+ * different signal.
+ *
+ * If a SIGSTOP (or other stopping signal) is not suppressed by the tracer,
+ * it will induce the regular illumos native job control stop of the entire
+ * traced process.  This is at least passingly similar to the Linux "group
+ * stop" ptrace(2) condition.
+ *
+ * SYSTEM CALL TRACING
+ *
+ * The ptrace(2) interface enables the tracer to hold the tracee on entry and
+ * exit from system calls.  When a stopped tracee is restarted through the
+ * PTRACE_SYSCALL request, the LX_PTRACE_SYSCALL flag is set until the next
+ * system call boundary.  Whether this is a "syscall-entry-stop" or
+ * "syscall-exit-stop", the tracee is held and the tracer is notified via
+ * SIGCLD/waitpid(2) in the usual way.  The flag LX_PTRACE_SYSCALL flag is
+ * cleared after each stop; for ongoing system call tracing the tracee must
+ * be continuously restarted with PTRACE_SYSCALL.
+ *
+ * EVENT STOPS
+ *
+ * Various events (particularly FORK, VFORK, CLONE, EXEC and EXIT) are
+ * enabled by the tracer through PTRACE_SETOPTIONS.  Once enabled, the tracee
+ * will be stopped at the nominated points of interest and the tracer
+ * notified.  The tracer may request additional information about the event,
+ * such as the pid of new LWPs and processes, via PTRACE_GETEVENTMSG.
+ *
+ * LOCK ORDERING RULES
+ *
+ * It is not safe, in general, to hold p_lock for two different processes at
+ * the same time.  This constraint is the primary reason for the existence
+ * (and complexity) of the ptrace(2) accord mechanism.
+ *
+ * In order to facilitate looking up accords by the "pid" of a tracer LWP,
+ * p_lock for the tracer process may be held while entering the accord mutex
+ * (lxpa_lock).  This mutex protects the accord flags and reference count.
+ * The reference count is manipulated through lx_ptrace_accord_hold() and
+ * lx_ptrace_accord_rele().
+ *
+ * DO NOT interact with the accord mutex (lxpa_lock) directly.  The
+ * lx_ptrace_accord_enter() and lx_ptrace_accord_exit() functions do various
+ * book-keeping and lock ordering enforcement and MUST be used.
+ *
+ * It is NOT legal to take ANY p_lock while holding the accord mutex
+ * (lxpa_lock).  If the lxpa_tracees_lock is to be held concurrently with
+ * lxpa_lock, lxpa_lock MUST be taken first and dropped before taking p_lock
+ * of any processes from the tracee list.
+ *
+ * It is NOT legal to take a tracee p_lock and then attempt to enter the
+ * accord mutex (or tracee list mutex) of its tracer.  When running as the
+ * tracee LWP, the tracee's hold will prevent the accord from being freed.
+ * Use of the LX_PTRACE_STOPPING or LX_PTRACE_CLONING flag in the
+ * LWP-specific brand data prevents an exiting tracer from altering the
+ * tracee until the tracee has come to an orderly stop, without requiring the
+ * tracee to hold its own p_lock the entire time it is stopping.
+ *
+ * It is not safe, in general, to enter "pidlock" while holding the p_lock of
+ * any process.  It is similarly illegal to hold any accord locks (lxpa_lock
+ * or lxpa_sublock) while attempting to enter "pidlock".  As "pidlock" is a
+ * global mutex, it should be held for the shortest possible time.
+ */
+
+#include <sys/types.h>
+#include <sys/kmem.h>
+#include <sys/ksynch.h>
+#include <sys/sysmacros.h>
+#include <sys/procfs.h>
+#include <sys/cmn_err.h>
+#include <sys/ddi.h>
+#include <sys/sunddi.h>
+#include <sys/wait.h>
+#include <sys/prsystm.h>
+#include <sys/note.h>
+
+#include <sys/brand.h>
+#include <sys/lx_brand.h>
+#include <sys/lx_impl.h>
+#include <sys/lx_misc.h>
+#include <lx_syscall.h>
+#include <lx_signum.h>
+
+
+typedef enum lx_ptrace_cont_flags_t {
+	LX_PTC_NONE = 0x00,
+	LX_PTC_SYSCALL = 0x01,
+	LX_PTC_SINGLESTEP = 0x02
+} lx_ptrace_cont_flags_t;
+
+/*
+ * Macros for checking the state of an LWP via "br_ptrace_flags":
+ */
+#define	LX_PTRACE_BUSY \
+	(LX_PTRACE_EXITING | LX_PTRACE_STOPPING | LX_PTRACE_CLONING)
+
+#define	VISIBLE(a)	(((a)->br_ptrace_flags & LX_PTRACE_EXITING) == 0)
+#define	TRACEE_BUSY(a)	(((a)->br_ptrace_flags & LX_PTRACE_BUSY) != 0)
+
+#define	ACCORD_HELD(a)	MUTEX_HELD(&(a)->lxpa_lock)
+
+static kcondvar_t lx_ptrace_busy_cv;
+static kmem_cache_t *lx_ptrace_accord_cache;
+
+/*
+ * Enter the accord mutex.
+ */
+static void
+lx_ptrace_accord_enter(lx_ptrace_accord_t *accord)
+{
+	VERIFY(MUTEX_NOT_HELD(&accord->lxpa_tracees_lock));
+
+	mutex_enter(&accord->lxpa_lock);
+}
+
+/*
+ * Exit the accord mutex.  If the reference count has dropped to zero,
+ * free the accord.
+ */
+static void
+lx_ptrace_accord_exit(lx_ptrace_accord_t *accord)
+{
+	VERIFY(ACCORD_HELD(accord));
+
+	if (accord->lxpa_refcnt > 0) {
+		mutex_exit(&accord->lxpa_lock);
+		return;
+	}
+
+	/*
+	 * When the reference count drops to zero we must free the accord.
+	 */
+	VERIFY(accord->lxpa_tracer == NULL);
+	VERIFY(MUTEX_NOT_HELD(&accord->lxpa_tracees_lock));
+	VERIFY(list_is_empty(&accord->lxpa_tracees));
+	VERIFY(accord->lxpa_flags & LX_ACC_TOMBSTONE);
+
+	mutex_destroy(&accord->lxpa_lock);
+	mutex_destroy(&accord->lxpa_tracees_lock);
+
+	kmem_cache_free(lx_ptrace_accord_cache, accord);
+}
+
+/*
+ * Drop our reference to this accord.  If this drops the reference count
+ * to zero, the next lx_ptrace_accord_exit() will free the accord.
+ */
+static void
+lx_ptrace_accord_rele(lx_ptrace_accord_t *accord)
+{
+	VERIFY(ACCORD_HELD(accord));
+
+	VERIFY(accord->lxpa_refcnt > 0);
+	accord->lxpa_refcnt--;
+}
+
+/*
+ * Place an additional hold on an accord.
+ */
+static void
+lx_ptrace_accord_hold(lx_ptrace_accord_t *accord)
+{
+	VERIFY(ACCORD_HELD(accord));
+
+	accord->lxpa_refcnt++;
+}
+
+/*
+ * Fetch the accord for this LWP.  If one has not yet been created, and the
+ * process is not exiting, allocate it now.  Must be called with p_lock held
+ * for the process containing the target LWP.
+ *
+ * If successful, we return holding the accord lock (lxpa_lock).
+ */
+static int
+lx_ptrace_accord_get_locked(klwp_t *lwp, lx_ptrace_accord_t **accordp,
+    boolean_t allocate_one)
+{
+	lx_ptrace_accord_t *lxpa;
+	lx_lwp_data_t *lwpd = lwptolxlwp(lwp);
+	proc_t *p = lwptoproc(lwp);
+
+	VERIFY(MUTEX_HELD(&p->p_lock));
+
+	/*
+	 * If this LWP does not have an accord, we wish to allocate
+	 * and install one.
+	 */
+	if ((lxpa = lwpd->br_ptrace_accord) == NULL) {
+		if (!allocate_one || !VISIBLE(lwpd)) {
+			/*
+			 * Either we do not wish to allocate an accord, or this
+			 * LWP has already begun exiting from a ptrace
+			 * perspective.
+			 */
+			*accordp = NULL;
+			return (ESRCH);
+		}
+
+		lxpa = kmem_cache_alloc(lx_ptrace_accord_cache, KM_SLEEP);
+		bzero(lxpa, sizeof (*lxpa));
+
+		/*
+		 * The initial reference count is 1 because we are referencing
+		 * it in from the soon-to-be tracer LWP.
+		 */
+		lxpa->lxpa_refcnt = 1;
+		mutex_init(&lxpa->lxpa_lock, NULL, MUTEX_DEFAULT, NULL);
+		mutex_init(&lxpa->lxpa_tracees_lock, NULL, MUTEX_DEFAULT, NULL);
+		list_create(&lxpa->lxpa_tracees, sizeof (lx_lwp_data_t),
+		    offsetof(lx_lwp_data_t, br_ptrace_linkage));
+		lxpa->lxpa_cvp = &p->p_cv;
+
+		lxpa->lxpa_tracer = lwpd;
+		lwpd->br_ptrace_accord = lxpa;
+	}
+
+	/*
+	 * Lock the accord before returning it to the caller.
+	 */
+	lx_ptrace_accord_enter(lxpa);
+
+	/*
+	 * There should be at least one active reference to this accord,
+	 * otherwise it should have been freed.
+	 */
+	VERIFY(lxpa->lxpa_refcnt > 0);
+
+	*accordp = lxpa;
+	return (0);
+}
+
+/*
+ * Accords belong to the tracer LWP.  Get the accord for this tracer or return
+ * an error if it was not possible.  To prevent deadlocks, the caller MUST NOT
+ * hold p_lock on its own or any other process.
+ *
+ * If successful, we return holding the accord lock (lxpa_lock).
+ */
+static int
+lx_ptrace_accord_get_by_pid(pid_t lxpid, lx_ptrace_accord_t **accordp)
+{
+	int ret = ESRCH;
+	pid_t apid;
+	id_t atid;
+	proc_t *aproc;
+	kthread_t *athr;
+	klwp_t *alwp;
+	lx_lwp_data_t *alwpd;
+
+	VERIFY(MUTEX_NOT_HELD(&curproc->p_lock));
+
+	/*
+	 * Locate the process containing the tracer LWP based on its Linux pid
+	 * and lock it.
+	 */
+	if (lx_lpid_to_spair(lxpid, &apid, &atid) != 0 ||
+	    (aproc = sprlock(apid)) == NULL) {
+		return (ESRCH);
+	}
+
+	/*
+	 * Locate the tracer LWP itself and ensure that it is visible to
+	 * ptrace(2).
+	 */
+	if ((athr = idtot(aproc, atid)) == NULL ||
+	    (alwp = ttolwp(athr)) == NULL ||
+	    (alwpd = lwptolxlwp(alwp)) == NULL ||
+	    !VISIBLE(alwpd)) {
+		sprunlock(aproc);
+		return (ESRCH);
+	}
+
+	/*
+	 * We should not fetch our own accord this way.
+	 */
+	if (athr == curthread) {
+		sprunlock(aproc);
+		return (EPERM);
+	}
+
+	/*
+	 * Fetch (or allocate) the accord owned by this tracer LWP:
+	 */
+	ret = lx_ptrace_accord_get_locked(alwp, accordp, B_TRUE);
+
+	/*
+	 * Unlock the process and return.
+	 */
+	sprunlock(aproc);
+	return (ret);
+}
+
+/*
+ * Get (or allocate) the ptrace(2) accord for the current LWP, acting as a
+ * tracer.  The caller MUST NOT currently hold p_lock on the process containing
+ * this LWP.
+ *
+ * If successful, we return holding the accord lock (lxpa_lock).
+ */
+static int
+lx_ptrace_accord_get(lx_ptrace_accord_t **accordp, boolean_t allocate_one)
+{
+	klwp_t *lwp = ttolwp(curthread);
+	proc_t *p = lwptoproc(lwp);
+	int ret;
+
+	VERIFY(MUTEX_NOT_HELD(&p->p_lock));
+
+	/*
+	 * Lock the tracer (this LWP).
+	 */
+	mutex_enter(&p->p_lock);
+
+	/*
+	 * Fetch (or allocate) the accord for this LWP:
+	 */
+	ret = lx_ptrace_accord_get_locked(lwp, accordp, allocate_one);
+
+	mutex_exit(&p->p_lock);
+
+	return (ret);
+}
+
+/*
+ * Restart an LWP if it is in "ptrace-stop".  This function may induce sleep,
+ * so the caller MUST NOT hold any mutexes other than p_lock for the process
+ * containing the LWP.
+ */
+static void
+lx_ptrace_restart_lwp(klwp_t *lwp)
+{
+	kthread_t *rt = lwptot(lwp);
+	proc_t *rproc = lwptoproc(lwp);
+	lx_lwp_data_t *rlwpd = lwptolxlwp(lwp);
+
+	VERIFY(rt != curthread);
+	VERIFY(MUTEX_HELD(&rproc->p_lock));
+
+	/*
+	 * Exclude potential meddling from procfs.
+	 */
+	prbarrier(rproc);
+
+	/*
+	 * Check that the LWP is still in "ptrace-stop" and, if so, restart it.
+	 */
+	thread_lock(rt);
+	if (BSTOPPED(rt) && rt->t_whystop == PR_BRAND) {
+		rt->t_schedflag |= TS_BSTART;
+		setrun_locked(rt);
+
+		/*
+		 * Clear stop reason.
+		 */
+		rlwpd->br_ptrace_whystop = 0;
+		rlwpd->br_ptrace_whatstop = 0;
+		rlwpd->br_ptrace_flags &= ~(LX_PTRACE_CLDPEND |
+		    LX_PTRACE_WAITPEND);
+	}
+	thread_unlock(rt);
+}
+
+static void
+lx_winfo(lx_lwp_data_t *remote, k_siginfo_t *ip, boolean_t waitflag,
+    pid_t *event_ppid, pid_t *event_pid)
+{
+	int signo;
+
+	/*
+	 * Populate our k_siginfo_t with data about this "ptrace-stop"
+	 * condition:
+	 */
+	bzero(ip, sizeof (*ip));
+	ip->si_signo = SIGCLD;
+	ip->si_pid = remote->br_pid;
+	ip->si_code = CLD_TRAPPED;
+
+	switch (remote->br_ptrace_whatstop) {
+	case LX_PR_SYSENTRY:
+	case LX_PR_SYSEXIT:
+		ip->si_status = SIGTRAP;
+		if (remote->br_ptrace_options & LX_PTRACE_O_TRACESYSGOOD) {
+			ip->si_status |= 0x80;
+		}
+		break;
+
+	case LX_PR_SIGNALLED:
+		signo = remote->br_ptrace_stopsig;
+		if (signo < 1 || signo >= LX_NSIG) {
+			/*
+			 * If this signal number is not valid, pretend it
+			 * was a SIGTRAP.
+			 */
+			ip->si_status = SIGTRAP;
+		} else {
+			ip->si_status = ltos_signo[signo];
+		}
+		break;
+
+	case LX_PR_EVENT:
+		ip->si_status = SIGTRAP | remote->br_ptrace_event;
+		/*
+		 * Record the Linux pid of both this LWP and the create
+		 * event we are dispatching.  We will use this information
+		 * to unblock any subsequent ptrace(2) events that depend
+		 * on this one.
+		 */
+		if (event_ppid != NULL)
+			*event_ppid = remote->br_pid;
+		if (event_pid != NULL)
+			*event_pid = (pid_t)remote->br_ptrace_eventmsg;
+		break;
+
+	default:
+		cmn_err(CE_PANIC, "unxpected stop subreason: %d",
+		    remote->br_ptrace_whatstop);
+	}
+
+	/*
+	 * If WNOWAIT was specified, do not mark the event as posted
+	 * so that it may be re-fetched on another call to waitid().
+	 */
+	if (waitflag) {
+		remote->br_ptrace_flags &= ~(LX_PTRACE_CLDPEND |
+		    LX_PTRACE_WAITPEND);
+	}
+}
+
+/*
+ * Receive notification from stop() of a PR_BRAND stop.
+ */
+void
+lx_stop_notify(proc_t *p, klwp_t *lwp, ushort_t why, ushort_t what)
+{
+	lx_lwp_data_t *lwpd = lwptolxlwp(lwp);
+	lx_ptrace_accord_t *accord;
+	klwp_t *plwp = NULL;
+	proc_t *pp = NULL;
+	lx_lwp_data_t *parent;
+	boolean_t cldpend = B_TRUE;
+	boolean_t cldpost = B_FALSE;
+	sigqueue_t *sqp = NULL;
+
+	/*
+	 * We currently only care about LX-specific stop reasons.
+	 */
+	if (why != PR_BRAND)
+		return;
+
+	switch (what) {
+	case LX_PR_SYSENTRY:
+	case LX_PR_SYSEXIT:
+	case LX_PR_SIGNALLED:
+	case LX_PR_EVENT:
+		break;
+	default:
+		cmn_err(CE_PANIC, "unexpected subreason for PR_BRAND"
+		    " stop: %d", (int)what);
+	}
+
+	/*
+	 * We should be holding the lock on our containing process.  The
+	 * STOPPING flag should have been set by lx_ptrace_stop() for all
+	 * PR_BRAND stops.
+	 */
+	VERIFY(MUTEX_HELD(&p->p_lock));
+	VERIFY(lwpd->br_ptrace_flags & LX_PTRACE_STOPPING);
+	VERIFY((accord = lwpd->br_ptrace_tracer) != NULL);
+
+	/*
+	 * We must drop our process lock to take "pidlock".  The
+	 * LX_PTRACE_STOPPING flag protects us from an exiting tracer.
+	 */
+	mutex_exit(&p->p_lock);
+
+	/*
+	 * Allocate before we enter any mutexes.
+	 */
+	sqp = kmem_zalloc(sizeof (*sqp), KM_SLEEP);
+
+	/*
+	 * We take pidlock now, which excludes all callers of waitid() and
+	 * prevents a detaching tracer from clearing critical accord members.
+	 */
+	mutex_enter(&pidlock);
+	mutex_enter(&p->p_lock);
+
+	/*
+	 * Get the ptrace(2) "parent" process, to which we may send
+	 * a SIGCLD signal later.
+	 */
+	if ((parent = accord->lxpa_tracer) != NULL &&
+	    (plwp = parent->br_lwp) != NULL) {
+		pp = lwptoproc(plwp);
+	}
+
+	/*
+	 * Our tracer should not have been modified in our absence; the
+	 * LX_PTRACE_STOPPING flag prevents it.
+	 */
+	VERIFY(lwpd->br_ptrace_tracer == accord);
+
+	/*
+	 * Stash data for this stop condition in the LWP data while we hold
+	 * both pidlock and our p_lock.
+	 */
+	lwpd->br_ptrace_whystop = why;
+	lwpd->br_ptrace_whatstop = what;
+	lwpd->br_ptrace_flags |= LX_PTRACE_WAITPEND;
+
+	/*
+	 * If this event does not depend on an event from the parent LWP,
+	 * populate the siginfo_t for the event pending on this tracee LWP.
+	 */
+	if (!(lwpd->br_ptrace_flags & LX_PTRACE_PARENT_WAIT) && pp != NULL) {
+		cldpost = B_TRUE;
+		lx_winfo(lwpd, &sqp->sq_info, B_FALSE, NULL, NULL);
+	}
+
+	/*
+	 * Drop our p_lock so that we may lock the tracer.
+	 */
+	mutex_exit(&p->p_lock);
+	if (cldpost && pp != NULL) {
+		/*
+		 * Post the SIGCLD to the tracer.
+		 */
+		mutex_enter(&pp->p_lock);
+		if (!sigismember(&pp->p_sig, SIGCLD)) {
+			sigaddqa(pp, plwp->lwp_thread, sqp);
+			cldpend = B_FALSE;
+			sqp = NULL;
+		}
+		mutex_exit(&pp->p_lock);
+	}
+
+	/*
+	 * We re-take our process lock now.  The lock will be held until
+	 * the thread is actually marked stopped, so we will not race with
+	 * lx_ptrace_lock_if_stopped() or lx_waitid_helper().
+	 */
+	mutex_enter(&p->p_lock);
+
+	/*
+	 * We clear the STOPPING flag; stop() continues to hold our p_lock
+	 * until our thread stop state is visible.
+	 */
+	lwpd->br_ptrace_flags &= ~LX_PTRACE_STOPPING;
+	lwpd->br_ptrace_flags |= LX_PTRACE_STOPPED;
+	if (cldpend) {
+		/*
+		 * We sent the SIGCLD for this new wait condition already.
+		 */
+		lwpd->br_ptrace_flags |= LX_PTRACE_CLDPEND;
+	}
+
+	/*
+	 * If lx_ptrace_exit_tracer() is trying to detach our tracer, it will
+	 * be sleeping on this CV until LX_PTRACE_STOPPING is clear.  Wake it
+	 * now.
+	 */
+	cv_broadcast(&lx_ptrace_busy_cv);
+
+	/*
+	 * While still holding pidlock, we attempt to wake our tracer from a
+	 * potential waitid() slumber.
+	 */
+	if (accord->lxpa_cvp != NULL) {
+		cv_broadcast(accord->lxpa_cvp);
+	}
+
+	/*
+	 * We release pidlock and return as we were called: with our p_lock
+	 * held.
+	 */
+	mutex_exit(&pidlock);
+
+	if (sqp != NULL) {
+		kmem_free(sqp, sizeof (*sqp));
+	}
+}
+
+/*
+ * For any restarting action (e.g. PTRACE_CONT, PTRACE_SYSCALL or
+ * PTRACE_DETACH) to be allowed, the tracee LWP must be in "ptrace-stop".  This
+ * check must ONLY be run on tracees of the current LWP.  If the check is
+ * successful, we return with the tracee p_lock held.
+ */
+static int
+lx_ptrace_lock_if_stopped(lx_ptrace_accord_t *accord, lx_lwp_data_t *remote)
+{
+	klwp_t *rlwp = remote->br_lwp;
+	proc_t *rproc = lwptoproc(rlwp);
+	kthread_t *rt = lwptot(rlwp);
+
+	/*
+	 * We must never check that we, ourselves, are stopped.  We must also
+	 * have the accord tracee list locked while we lock our tracees.
+	 */
+	VERIFY(curthread != rt);
+	VERIFY(MUTEX_HELD(&accord->lxpa_tracees_lock));
+	VERIFY(accord->lxpa_tracer == ttolxlwp(curthread));
+
+	/*
+	 * Lock the process containing the tracee LWP.
+	 */
+	mutex_enter(&rproc->p_lock);
+	if (!VISIBLE(remote)) {
+		/*
+		 * The tracee LWP is currently detaching itself as it exits.
+		 * It is no longer visible to ptrace(2).
+		 */
+		mutex_exit(&rproc->p_lock);
+		return (ESRCH);
+	}
+
+	/*
+	 * We must only check whether tracees of the current LWP are stopped.
+	 * We check this condition after confirming visibility as an exiting
+	 * tracee may no longer be completely consistent.
+	 */
+	VERIFY(remote->br_ptrace_tracer == accord);
+
+	if (!(remote->br_ptrace_flags & LX_PTRACE_STOPPED)) {
+		/*
+		 * The tracee is not in "ptrace-stop", so we release the
+		 * process.
+		 */
+		mutex_exit(&rproc->p_lock);
+		return (ESRCH);
+	}
+
+	/*
+	 * The tracee is stopped.  We return holding its process lock so that
+	 * the caller may manipulate it.
+	 */
+	return (0);
+}
+
+static int
+lx_ptrace_setoptions(lx_lwp_data_t *remote, uintptr_t options)
+{
+	/*
+	 * Check for valid options.
+	 */
+	if ((options & ~LX_PTRACE_O_ALL) != 0) {
+		return (EINVAL);
+	}
+
+	/*
+	 * Set ptrace options on the target LWP.
+	 */
+	remote->br_ptrace_options = (lx_ptrace_options_t)options;
+
+	return (0);
+}
+
+static int
+lx_ptrace_geteventmsg(lx_lwp_data_t *remote, void *umsgp)
+{
+	int error;
+
+#if defined(_SYSCALL32_IMPL)
+	if (get_udatamodel() != DATAMODEL_NATIVE) {
+		uint32_t tmp = remote->br_ptrace_eventmsg;
+
+		error = copyout(&tmp, umsgp, sizeof (uint32_t));
+	} else
+#endif
+	{
+		error = copyout(&remote->br_ptrace_eventmsg, umsgp,
+		    sizeof (ulong_t));
+	}
+
+	return (error);
+}
+
+static int
+lx_ptrace_getregs(lx_lwp_data_t *remote, void *uregsp)
+{
+	if (remote->br_stack_mode == LX_STACK_MODE_BRAND) {
+		/*
+		 * The LWP was stopped with the brand stack and register
+		 * state loaded, e.g. during a system call emulated within
+		 * the kernel.  Return the LWP register state.
+		 */
+		return (lx_regs_to_userregs(remote, uregsp));
+	} else if (remote->br_stack_mode == LX_STACK_MODE_PREINIT &&
+	    (remote->br_ptrace_whatstop == LX_PR_SIGNALLED ||
+	    remote->br_ptrace_whatstop == LX_PR_SYSEXIT)) {
+		/*
+		 * The LWP was stopped by tracing on exec.
+		 */
+		return (lx_regs_to_userregs(remote, uregsp));
+	} else if (remote->br_stack_mode == LX_STACK_MODE_NATIVE &&
+	    remote->br_ptrace_whystop == PR_BRAND &&
+	    remote->br_ptrace_whatstop == LX_PR_EVENT) {
+		/*
+		 * Called while we're ptrace event stopped by lx_exec.
+		 */
+		return (lx_regs_to_userregs(remote, uregsp));
+	} else if (remote->br_ptrace_stopucp != NULL) {
+		/*
+		 * The LWP was stopped in the usermode emulation library
+		 * but a ucontext_t for the preserved brand stack and
+		 * register state was provided.  Return the register state
+		 * from that ucontext_t.
+		 */
+		return (lx_uc_to_userregs(remote,
+		    (void *)remote->br_ptrace_stopucp, uregsp));
+	} else {
+		/*
+		 * The register state is not currently available.
+		 */
+		return (EIO);
+	}
+}
+
+static int
+lx_ptrace_setregs(lx_lwp_data_t *remote, void *uregsp)
+{
+	if (remote->br_stack_mode == LX_STACK_MODE_BRAND) {
+		/*
+		 * The LWP was stopped with the brand stack and register
+		 * state loaded, e.g. during a system call emulated within
+		 * the kernel.  Write to the LWP register state.
+		 */
+		return (lx_userregs_to_regs(remote, uregsp));
+	} else if (remote->br_ptrace_stopucp != NULL) {
+		/*
+		 * The LWP was stopped in the usermode emulation library
+		 * but a ucontext_t for the preserved brand stack and
+		 * register state was provided.  Write to the register state
+		 * in that ucontext_t.
+		 */
+		return (lx_userregs_to_uc(remote,
+		    (void *)remote->br_ptrace_stopucp, uregsp));
+	} else {
+		/*
+		 * The register state is not currently available.
+		 */
+		return (EIO);
+	}
+}
+
+/*
+ * Implements the PTRACE_CONT subcommand of the Linux ptrace(2) interface.
+ */
+static int
+lx_ptrace_cont(lx_lwp_data_t *remote, lx_ptrace_cont_flags_t flags, int signo)
+{
+	klwp_t *lwp = remote->br_lwp;
+
+	if (flags & LX_PTC_SINGLESTEP) {
+		/*
+		 * We do not currently support single-stepping.
+		 */
+		lx_unsupported("PTRACE_SINGLESTEP not currently implemented");
+		return (EINVAL);
+	}
+
+	/*
+	 * The tracer may choose to suppress the delivery of a signal, or
+	 * select an alternative signal for delivery.  If this is an
+	 * appropriate ptrace(2) "signal-delivery-stop", br_ptrace_stopsig
+	 * will be used as the new signal number.
+	 *
+	 * As with so many other aspects of the Linux ptrace(2) interface, this
+	 * may fail silently if the state machine is not aligned correctly.
+	 */
+	remote->br_ptrace_stopsig = signo;
+
+	/*
+	 * Handle the syscall-stop flag if this is a PTRACE_SYSCALL restart:
+	 */
+	if (flags & LX_PTC_SYSCALL) {
+		remote->br_ptrace_flags |= LX_PTRACE_SYSCALL;
+	} else {
+		remote->br_ptrace_flags &= ~LX_PTRACE_SYSCALL;
+	}
+
+	lx_ptrace_restart_lwp(lwp);
+
+	return (0);
+}
+
+/*
+ * Implements the PTRACE_DETACH subcommand of the Linux ptrace(2) interface.
+ *
+ * The LWP identified by the Linux pid "lx_pid" will, if it as a tracee of the
+ * current LWP, be detached and set runnable.  If the specified LWP is not
+ * currently in the "ptrace-stop" state, the routine will return ESRCH as if
+ * the LWP did not exist at all.
+ *
+ * The caller must not hold p_lock on any process.
+ */
+static int
+lx_ptrace_detach(lx_ptrace_accord_t *accord, lx_lwp_data_t *remote, int signo,
+    boolean_t *release_hold)
+{
+	klwp_t *rlwp;
+
+	rlwp = remote->br_lwp;
+
+	/*
+	 * The tracee LWP was in "ptrace-stop" and we now hold its p_lock.
+	 * Detach the LWP from the accord and set it running.
+	 */
+	VERIFY(!TRACEE_BUSY(remote));
+	remote->br_ptrace_flags &= ~(LX_PTRACE_SYSCALL | LX_PTRACE_INHERIT);
+	VERIFY(list_link_active(&remote->br_ptrace_linkage));
+	list_remove(&accord->lxpa_tracees, remote);
+
+	remote->br_ptrace_attach = LX_PTA_NONE;
+	remote->br_ptrace_tracer = NULL;
+	remote->br_ptrace_flags = 0;
+	*release_hold = B_TRUE;
+
+	/*
+	 * The tracer may, as described in lx_ptrace_cont(), choose to suppress
+	 * or modify the delivered signal.
+	 */
+	remote->br_ptrace_stopsig = signo;
+
+	lx_ptrace_restart_lwp(rlwp);
+
+	return (0);
+}
+
+/*
+ * This routine implements the PTRACE_ATTACH operation of the Linux ptrace(2)
+ * interface.
+ *
+ * This LWP is requesting to be attached as a tracer to another LWP -- the
+ * tracee.  If a ptrace accord to track the list of tracees has not yet been
+ * allocated, one will be allocated and attached to this LWP now.
+ *
+ * The "br_ptrace_tracer" on the tracee LWP is set to this accord, and the
+ * tracee LWP is then added to the "lxpa_tracees" list in the accord.  We drop
+ * locks between these two phases; the only consumer of trace events from this
+ * accord is this LWP, which obviously cannot be running waitpid(2) at the same
+ * time as this call to ptrace(2).
+ */
+static int
+lx_ptrace_attach(pid_t lx_pid)
+{
+	int error = ESRCH;
+	/*
+	 * Our (Tracer) LWP:
+	 */
+	lx_ptrace_accord_t *accord;
+	lx_lwp_data_t *lwpd = ttolxlwp(curthread);
+	/*
+	 * Remote (Tracee) LWP:
+	 */
+	pid_t rpid;
+	id_t rtid;
+	proc_t *rproc;
+	kthread_t *rthr;
+	klwp_t *rlwp;
+	lx_lwp_data_t *rlwpd;
+
+	if (lwpd->br_pid == lx_pid) {
+		/*
+		 * We cannot trace ourselves.
+		 */
+		return (EPERM);
+	}
+
+	/*
+	 * Ensure that we have an accord and obtain a lock on it.  This
+	 * routine should not fail because the LWP cannot make ptrace(2) system
+	 * calls after it has begun exiting.
+	 */
+	VERIFY0(lwpd->br_ptrace_flags & LX_PTRACE_EXITING);
+	VERIFY(lx_ptrace_accord_get(&accord, B_TRUE) == 0);
+
+	/*
+	 * Place speculative hold in case the attach is successful.
+	 */
+	lx_ptrace_accord_hold(accord);
+	lx_ptrace_accord_exit(accord);
+
+	/*
+	 * Locate the process containing the tracee LWP based on its Linux pid
+	 * and lock it.
+	 */
+	if (lx_lpid_to_spair(lx_pid, &rpid, &rtid) != 0 ||
+	    (rproc = sprlock(rpid)) == NULL) {
+		/*
+		 * We could not find the target process.
+		 */
+		goto errout;
+	}
+
+	/*
+	 * Locate the tracee LWP.
+	 */
+	if ((rthr = idtot(rproc, rtid)) == NULL ||
+	    (rlwp = ttolwp(rthr)) == NULL ||
+	    (rlwpd = lwptolxlwp(rlwp)) == NULL ||
+	    !VISIBLE(rlwpd)) {
+		/*
+		 * The LWP could not be found, was not branded, or is not
+		 * visible to ptrace(2) at this time.
+		 */
+		goto unlock_errout;
+	}
+
+	/*
+	 * We now hold the lock on the tracee.  Attempt to install ourselves
+	 * as the tracer.
+	 */
+	if (curproc != rproc && priv_proc_cred_perm(curproc->p_cred, rproc,
+	    NULL, VWRITE) != 0) {
+		/*
+		 * This process does not have permission to trace the remote
+		 * process.
+		 */
+		error = EPERM;
+	} else if (rlwpd->br_ptrace_tracer != NULL) {
+		/*
+		 * This LWP is already being traced.
+		 */
+		VERIFY(list_link_active(&rlwpd->br_ptrace_linkage));
+		VERIFY(rlwpd->br_ptrace_attach != LX_PTA_NONE);
+		error = EPERM;
+	} else {
+		lx_proc_data_t *rprocd;
+
+		/*
+		 * Bond the tracee to the accord.
+		 */
+		VERIFY0(rlwpd->br_ptrace_flags & LX_PTRACE_EXITING);
+		VERIFY(rlwpd->br_ptrace_attach == LX_PTA_NONE);
+		rlwpd->br_ptrace_attach = LX_PTA_ATTACH;
+		rlwpd->br_ptrace_tracer = accord;
+
+		/*
+		 * We had no tracer, and are thus not in the tracees list.
+		 * It is safe to take the tracee list lock while we insert
+		 * ourselves.
+		 */
+		mutex_enter(&accord->lxpa_tracees_lock);
+		VERIFY(!list_link_active(&rlwpd->br_ptrace_linkage));
+		list_insert_tail(&accord->lxpa_tracees, rlwpd);
+		mutex_exit(&accord->lxpa_tracees_lock);
+
+		/*
+		 * Send a thread-directed SIGSTOP.
+		 */
+		sigtoproc(rproc, rthr, SIGSTOP);
+
+		/*
+		 * Set the in-kernel process-wide ptrace(2) enable flag.
+		 */
+		rprocd = ttolxproc(rthr);
+		rprocd->l_ptrace = 1;
+
+		error = 0;
+	}
+
+unlock_errout:
+	/*
+	 * Unlock the process containing the tracee LWP and the accord.
+	 */
+	sprunlock(rproc);
+
+errout:
+	if (error != 0) {
+		/*
+		 * The attach was not successful.  Remove our speculative
+		 * hold.
+		 */
+		lx_ptrace_accord_enter(accord);
+		lx_ptrace_accord_rele(accord);
+		lx_ptrace_accord_exit(accord);
+	}
+
+	return (error);
+}
+
+int
+lx_ptrace_set_clone_inherit(int option, boolean_t inherit_flag)
+{
+	klwp_t *lwp = ttolwp(curthread);
+	proc_t *p = lwptoproc(lwp);
+	lx_lwp_data_t *lwpd = lwptolxlwp(lwp);
+
+	mutex_enter(&p->p_lock);
+
+	switch (option) {
+	case LX_PTRACE_O_TRACEFORK:
+	case LX_PTRACE_O_TRACEVFORK:
+	case LX_PTRACE_O_TRACECLONE:
+		lwpd->br_ptrace_clone_option = option;
+		break;
+
+	default:
+		return (EINVAL);
+	}
+
+	if (inherit_flag) {
+		lwpd->br_ptrace_flags |= LX_PTRACE_INHERIT;
+	} else {
+		lwpd->br_ptrace_flags &= ~LX_PTRACE_INHERIT;
+	}
+
+	mutex_exit(&p->p_lock);
+	return (0);
+}
+
+/*
+ * If the parent LWP is being traced, we want to attach ourselves to the
+ * same accord.
+ */
+void
+lx_ptrace_inherit_tracer(lx_lwp_data_t *src, lx_lwp_data_t *dst)
+{
+	proc_t *srcp = lwptoproc(src->br_lwp);
+	proc_t *dstp = lwptoproc(dst->br_lwp);
+	lx_ptrace_accord_t *accord;
+	boolean_t unlock = B_FALSE;
+
+	if (srcp == dstp) {
+		/*
+		 * This is syslwp_create(), so the process p_lock is already
+		 * held.
+		 */
+		VERIFY(MUTEX_HELD(&srcp->p_lock));
+	} else {
+		unlock = B_TRUE;
+		mutex_enter(&srcp->p_lock);
+	}
+
+	if ((accord = src->br_ptrace_tracer) == NULL) {
+		/*
+		 * The source LWP does not have a tracer to inherit.
+		 */
+		goto out;
+	}
+
+	/*
+	 * There are two conditions to check when determining if the new
+	 * child should inherit the same tracer (and tracing options) as its
+	 * parent.  Either condition is sufficient to trigger inheritance.
+	 */
+	dst->br_ptrace_attach = LX_PTA_NONE;
+	if ((src->br_ptrace_options & src->br_ptrace_clone_option) != 0) {
+		/*
+		 * Condition 1:
+		 * The clone(2), fork(2) and vfork(2) emulated system calls
+		 * populate "br_ptrace_clone_option" with the specific
+		 * ptrace(2) SETOPTIONS option that applies to this
+		 * operation.  If the relevant option has been enabled by the
+		 * tracer then we inherit.
+		 */
+		dst->br_ptrace_attach |= LX_PTA_INHERIT_OPTIONS;
+
+	} else if ((src->br_ptrace_flags & LX_PTRACE_INHERIT) != 0) {
+		/*
+		 * Condition 2:
+		 * If the caller opted in to inheritance with the
+		 * PTRACE_CLONE flag to clone(2), the LX_PTRACE_INHERIT flag
+		 * will be set and we inherit.
+		 */
+		dst->br_ptrace_attach |= LX_PTA_INHERIT_CLONE;
+	}
+
+	/*
+	 * These values only apply for the duration of a single clone(2), et
+	 * al, system call.
+	 */
+	src->br_ptrace_flags &= ~LX_PTRACE_INHERIT;
+	src->br_ptrace_clone_option = 0;
+
+	if (dst->br_ptrace_attach == LX_PTA_NONE) {
+		/*
+		 * No condition triggered inheritance.
+		 */
+		goto out;
+	}
+
+	/*
+	 * Set the LX_PTRACE_CLONING flag to prevent us from being detached
+	 * while our p_lock is dropped.
+	 */
+	src->br_ptrace_flags |= LX_PTRACE_CLONING;
+	mutex_exit(&srcp->p_lock);
+
+	/*
+	 * Hold the accord for the new LWP.
+	 */
+	lx_ptrace_accord_enter(accord);
+	lx_ptrace_accord_hold(accord);
+	lx_ptrace_accord_exit(accord);
+
+	/*
+	 * Install the tracer and copy the current PTRACE_SETOPTIONS options.
+	 */
+	dst->br_ptrace_tracer = accord;
+	dst->br_ptrace_options = src->br_ptrace_options;
+
+	/*
+	 * This flag prevents waitid() from seeing events for the new child
+	 * until the parent is able to post the relevant ptrace event to
+	 * the tracer.
+	 */
+	dst->br_ptrace_flags |= LX_PTRACE_PARENT_WAIT;
+
+	mutex_enter(&accord->lxpa_tracees_lock);
+	VERIFY(list_link_active(&src->br_ptrace_linkage));
+	VERIFY(!list_link_active(&dst->br_ptrace_linkage));
+	list_insert_tail(&accord->lxpa_tracees, dst);
+	mutex_exit(&accord->lxpa_tracees_lock);
+
+	/*
+	 * Relock our process and clear our busy flag.
+	 */
+	mutex_enter(&srcp->p_lock);
+	src->br_ptrace_flags &= ~LX_PTRACE_CLONING;
+
+	/*
+	 * If lx_ptrace_exit_tracer() is trying to detach our tracer, it will
+	 * be sleeping on this CV until LX_PTRACE_CLONING is clear.  Wake it
+	 * now.
+	 */
+	cv_broadcast(&lx_ptrace_busy_cv);
+
+out:
+	if (unlock) {
+		mutex_exit(&srcp->p_lock);
+	}
+}
+
+static int
+lx_ptrace_traceme(void)
+{
+	int error;
+	boolean_t did_attach = B_FALSE;
+	/*
+	 * Our (Tracee) LWP:
+	 */
+	klwp_t *lwp = ttolwp(curthread);
+	proc_t *p = lwptoproc(lwp);
+	lx_lwp_data_t *lwpd = lwptolxlwp(lwp);
+	/*
+	 * Remote (Tracer) LWP:
+	 */
+	lx_ptrace_accord_t *accord;
+
+	/*
+	 * We are intending to be the tracee.  Fetch (or allocate) the accord
+	 * for our parent LWP.
+	 */
+	if ((error = lx_ptrace_accord_get_by_pid(lx_lwp_ppid(lwp, NULL,
+	    NULL), &accord)) != 0) {
+		/*
+		 * Could not determine the Linux pid of the parent LWP, or
+		 * could not get the accord for that LWP.
+		 */
+		return (error);
+	}
+
+	/*
+	 * We now hold the accord lock.
+	 */
+	if (accord->lxpa_flags & LX_ACC_TOMBSTONE) {
+		/*
+		 * The accord is marked for death; give up now.
+		 */
+		lx_ptrace_accord_exit(accord);
+		return (ESRCH);
+	}
+
+	/*
+	 * Bump the reference count so that the accord is not freed.  We need
+	 * to drop the accord lock before we take our own p_lock.
+	 */
+	lx_ptrace_accord_hold(accord);
+	lx_ptrace_accord_exit(accord);
+
+	/*
+	 * We now lock _our_ process and determine if we can install our parent
+	 * as our tracer.
+	 */
+	mutex_enter(&p->p_lock);
+	if (lwpd->br_ptrace_tracer != NULL) {
+		/*
+		 * This LWP is already being traced.
+		 */
+		VERIFY(lwpd->br_ptrace_attach != LX_PTA_NONE);
+		error = EPERM;
+	} else {
+		/*
+		 * Bond ourselves to the accord.  We already bumped the accord
+		 * reference count.
+		 */
+		VERIFY(lwpd->br_ptrace_attach == LX_PTA_NONE);
+		lwpd->br_ptrace_attach = LX_PTA_TRACEME;
+		lwpd->br_ptrace_tracer = accord;
+		did_attach = B_TRUE;
+		error = 0;
+	}
+	mutex_exit(&p->p_lock);
+
+	/*
+	 * Lock the accord tracee list and add this LWP.  Once we are in the
+	 * tracee list, it is the responsibility of the tracer to detach us.
+	 */
+	if (error == 0) {
+		lx_ptrace_accord_enter(accord);
+		mutex_enter(&accord->lxpa_tracees_lock);
+
+		if (!(accord->lxpa_flags & LX_ACC_TOMBSTONE)) {
+			lx_proc_data_t *procd = ttolxproc(curthread);
+
+			/*
+			 * Put ourselves in the tracee list for this accord.
+			 */
+			VERIFY(!list_link_active(&lwpd->br_ptrace_linkage));
+			list_insert_tail(&accord->lxpa_tracees, lwpd);
+			mutex_exit(&accord->lxpa_tracees_lock);
+			lx_ptrace_accord_exit(accord);
+
+			/*
+			 * Set the in-kernel process-wide ptrace(2) enable
+			 * flag.
+			 */
+			procd->l_ptrace = 1;
+
+			return (0);
+		}
+		mutex_exit(&accord->lxpa_tracees_lock);
+
+		/*
+		 * The accord has been marked for death.  We must
+		 * untrace ourselves.
+		 */
+		error = ESRCH;
+		lx_ptrace_accord_exit(accord);
+	}
+
+	/*
+	 * Our optimism was unjustified: We were unable to attach.  We need to
+	 * lock the process containing this LWP again in order to remove the
+	 * tracer.
+	 */
+	VERIFY(error != 0);
+	mutex_enter(&p->p_lock);
+	if (did_attach) {
+		/*
+		 * Verify that things were as we left them:
+		 */
+		VERIFY(!list_link_active(&lwpd->br_ptrace_linkage));
+		VERIFY(lwpd->br_ptrace_tracer == accord);
+
+		lwpd->br_ptrace_attach = LX_PTA_NONE;
+		lwpd->br_ptrace_tracer = NULL;
+	}
+	mutex_exit(&p->p_lock);
+
+	/*
+	 * Remove our speculative hold on the accord, possibly causing it to be
+	 * freed in the process.
+	 */
+	lx_ptrace_accord_enter(accord);
+	lx_ptrace_accord_rele(accord);
+	lx_ptrace_accord_exit(accord);
+
+	return (error);
+}
+
+static boolean_t
+lx_ptrace_stop_common(proc_t *p, lx_lwp_data_t *lwpd, ushort_t what)
+{
+	boolean_t reset_nostop = B_FALSE;
+
+	VERIFY(MUTEX_HELD(&p->p_lock));
+
+	/*
+	 * Mark this LWP as stopping and call stop() to enter "ptrace-stop".
+	 */
+	VERIFY0(lwpd->br_ptrace_flags & LX_PTRACE_STOPPING);
+	lwpd->br_ptrace_flags |= LX_PTRACE_STOPPING;
+
+	if (lwpd->br_lwp->lwp_nostop == 1 &&
+	    lwpd->br_ptrace_event == LX_PTRACE_EVENT_EXEC) {
+		/* We need to clear this to get the signal delivered. */
+		lwpd->br_lwp->lwp_nostop = 0;
+		reset_nostop = B_TRUE;
+	}
+
+	stop(PR_BRAND, what);
+
+	if (reset_nostop) {
+		VERIFY(lwpd->br_lwp->lwp_nostop == 0);
+		lwpd->br_lwp->lwp_nostop = 1;
+	}
+
+	/*
+	 * We are back from "ptrace-stop" with our process lock held.
+	 */
+	lwpd->br_ptrace_flags &= ~(LX_PTRACE_STOPPING | LX_PTRACE_STOPPED |
+	    LX_PTRACE_CLDPEND);
+	lwpd->br_ptrace_stopucp = NULL;
+	cv_broadcast(&lx_ptrace_busy_cv);
+	mutex_exit(&p->p_lock);
+
+	return (B_TRUE);
+}
+
+int
+lx_ptrace_stop_for_option(int option, boolean_t child, ulong_t msg,
+    uintptr_t ucp)
+{
+	kthread_t *t = curthread;
+	klwp_t *lwp = ttolwp(t);
+	proc_t *p = lwptoproc(lwp);
+	lx_lwp_data_t *lwpd = lwptolxlwp(lwp);
+
+	mutex_enter(&p->p_lock);
+	if (lwpd->br_ptrace_tracer == NULL) {
+		mutex_exit(&p->p_lock);
+		return (ESRCH);
+	}
+
+	if (!child) {
+		/*
+		 * Only the first event posted by a new process is to be held
+		 * until the matching parent event is dispatched, and only if
+		 * it is a "child" event.  This is not a child event, so we
+		 * clear the wait flag.
+		 */
+		lwpd->br_ptrace_flags &= ~LX_PTRACE_PARENT_WAIT;
+	}
+
+	if (!(lwpd->br_ptrace_options & option)) {
+		if (option == LX_PTRACE_O_TRACEEXEC) {
+			/*
+			 * Without PTRACE_O_TRACEEXEC, the Linux kernel will
+			 * send SIGTRAP to the process.
+			 */
+			sigtoproc(p, t, SIGTRAP);
+			mutex_exit(&p->p_lock);
+			return (0);
+		}
+
+		/*
+		 * The flag for this trace event is not enabled, so we will not
+		 * stop.
+		 */
+		mutex_exit(&p->p_lock);
+		return (ESRCH);
+	}
+
+	if (child) {
+		switch (option) {
+		case LX_PTRACE_O_TRACECLONE:
+		case LX_PTRACE_O_TRACEFORK:
+		case LX_PTRACE_O_TRACEVFORK:
+			/*
+			 * Send the child LWP a directed SIGSTOP.
+			 */
+			sigtoproc(p, t, SIGSTOP);
+			mutex_exit(&p->p_lock);
+			return (0);
+		default:
+			goto nostop;
+		}
+	}
+
+	lwpd->br_ptrace_eventmsg = msg;
+
+	switch (option) {
+	case LX_PTRACE_O_TRACECLONE:
+		lwpd->br_ptrace_event = LX_PTRACE_EVENT_CLONE;
+		break;
+	case LX_PTRACE_O_TRACEEXEC:
+		lwpd->br_ptrace_event = LX_PTRACE_EVENT_EXEC;
+		lwpd->br_ptrace_eventmsg = 0;
+		break;
+	case LX_PTRACE_O_TRACEEXIT:
+		lwpd->br_ptrace_event = LX_PTRACE_EVENT_EXIT;
+		break;
+	case LX_PTRACE_O_TRACEFORK:
+		lwpd->br_ptrace_event = LX_PTRACE_EVENT_FORK;
+		break;
+	case LX_PTRACE_O_TRACEVFORK:
+		lwpd->br_ptrace_event = LX_PTRACE_EVENT_VFORK;
+		break;
+	case LX_PTRACE_O_TRACEVFORKDONE:
+		lwpd->br_ptrace_event = LX_PTRACE_EVENT_VFORK_DONE;
+		lwpd->br_ptrace_eventmsg = 0;
+		break;
+	default:
+		goto nostop;
+	}
+
+	/*
+	 * Userland may have passed in a ucontext_t pointer for
+	 * PTRACE_GETREGS/PTRACE_SETREGS usage while stopped.
+	 */
+	lwpd->br_ptrace_stopucp = ucp;
+
+	/*
+	 * p_lock for the process containing the tracee will be dropped by
+	 * lx_ptrace_stop_common().
+	 */
+	return (lx_ptrace_stop_common(p, lwpd, LX_PR_EVENT) ? 0 : ESRCH);
+
+nostop:
+	lwpd->br_ptrace_event = 0;
+	lwpd->br_ptrace_eventmsg = 0;
+	mutex_exit(&p->p_lock);
+	return (ESRCH);
+}
+
+boolean_t
+lx_ptrace_stop(ushort_t what)
+{
+	klwp_t *lwp = ttolwp(curthread);
+	proc_t *p = lwptoproc(lwp);
+	lx_lwp_data_t *lwpd = lwptolxlwp(lwp);
+
+	VERIFY(what == LX_PR_SYSENTRY || what == LX_PR_SYSEXIT ||
+	    what == LX_PR_SIGNALLED);
+
+	/*
+	 * If we do not have an accord, bail out early.
+	 */
+	if (lwpd->br_ptrace_tracer == NULL)
+		return (B_FALSE);
+
+	/*
+	 * Lock this process and re-check the condition.
+	 */
+	mutex_enter(&p->p_lock);
+	if (lwpd->br_ptrace_tracer == NULL) {
+		VERIFY0(lwpd->br_ptrace_flags & LX_PTRACE_SYSCALL);
+		mutex_exit(&p->p_lock);
+		return (B_FALSE);
+	}
+
+	if (what == LX_PR_SYSENTRY || what == LX_PR_SYSEXIT) {
+		/*
+		 * This is a syscall-entry-stop or syscall-exit-stop point.
+		 */
+		if (!(lwpd->br_ptrace_flags & LX_PTRACE_SYSCALL)) {
+			/*
+			 * A system call stop has not been requested.
+			 */
+			mutex_exit(&p->p_lock);
+			return (B_FALSE);
+		}
+
+		/*
+		 * The PTRACE_SYSCALL restart command applies only to the next
+		 * system call entry or exit.  The tracer must restart us with
+		 * PTRACE_SYSCALL while we are in ptrace-stop for us to fire
+		 * again at the next system call boundary.
+		 */
+		lwpd->br_ptrace_flags &= ~LX_PTRACE_SYSCALL;
+	}
+
+	/*
+	 * p_lock for the process containing the tracee will be dropped by
+	 * lx_ptrace_stop_common().
+	 */
+	return (lx_ptrace_stop_common(p, lwpd, what));
+}
+
+int
+lx_ptrace_issig_stop(proc_t *p, klwp_t *lwp)
+{
+	lx_lwp_data_t *lwpd = lwptolxlwp(lwp);
+	int lx_sig;
+
+	VERIFY(MUTEX_HELD(&p->p_lock));
+
+	/*
+	 * If we do not have an accord, bail out now.  Additionally, if there
+	 * is no valid signal then we have no reason to stop.
+	 */
+	if (lwpd->br_ptrace_tracer == NULL || lwp->lwp_cursig == SIGKILL ||
+	    (lwp->lwp_cursig == 0 || lwp->lwp_cursig > NSIG) ||
+	    (lx_sig = stol_signo[lwp->lwp_cursig]) < 1) {
+		return (0);
+	}
+
+	/*
+	 * We stash the signal on the LWP where our waitid_helper will find it
+	 * and enter the ptrace "signal-delivery-stop" condition.
+	 */
+	lwpd->br_ptrace_stopsig = lx_sig;
+	(void) lx_ptrace_stop_common(p, lwpd, LX_PR_SIGNALLED);
+	mutex_enter(&p->p_lock);
+
+	/*
+	 * When we return, the signal may have been altered or suppressed.
+	 */
+	if (lwpd->br_ptrace_stopsig != lx_sig) {
+		int native_sig;
+		lx_sig = lwpd->br_ptrace_stopsig;
+
+		if (lx_sig >= LX_NSIG) {
+			lx_sig = 0;
+		}
+
+		/*
+		 * Translate signal from Linux signal number back to
+		 * an illumos native signal.
+		 */
+		if (lx_sig >= LX_NSIG || lx_sig < 0 || (native_sig =
+		    ltos_signo[lx_sig]) < 1) {
+			/*
+			 * The signal is not deliverable.
+			 */
+			lwp->lwp_cursig = 0;
+			lwp->lwp_extsig = 0;
+			if (lwp->lwp_curinfo) {
+				siginfofree(lwp->lwp_curinfo);
+				lwp->lwp_curinfo = NULL;
+			}
+		} else {
+			/*
+			 * Alter the currently dispatching signal.
+			 */
+			if (native_sig == SIGKILL) {
+				/*
+				 * We mark ourselves the victim and request
+				 * a restart of signal processing.
+				 */
+				p->p_flag |= SKILLED;
+				p->p_flag &= ~SEXTKILLED;
+				return (-1);
+			}
+			lwp->lwp_cursig = native_sig;
+			lwp->lwp_extsig = 0;
+			if (lwp->lwp_curinfo != NULL) {
+				lwp->lwp_curinfo->sq_info.si_signo = native_sig;
+			}
+		}
+	}
+
+	lwpd->br_ptrace_stopsig = 0;
+	return (0);
+}
+
+boolean_t
+lx_ptrace_sig_ignorable(proc_t *p, int sig)
+{
+	lx_proc_data_t *lxpd = ptolxproc(p);
+
+	if (lxpd->l_ptrace != 0 && lx_stol_signo(sig, 0) != 0) {
+		/*
+		 * In order to preserve proper ptrace behavior when it comes to
+		 * signal handling, it is unacceptable to ignore any signals.
+		 * Doing so would bypass the logic in lx_ptrace_issig_stop.
+		 */
+		return (B_FALSE);
+	}
+	return (B_TRUE);
+}
+
+static void
+lx_ptrace_exit_tracer(proc_t *p, lx_lwp_data_t *lwpd,
+    lx_ptrace_accord_t *accord)
+{
+	VERIFY(MUTEX_NOT_HELD(&p->p_lock));
+
+	lx_ptrace_accord_enter(accord);
+	/*
+	 * Mark this accord for death.  This means no new tracees can be
+	 * attached to this accord.
+	 */
+	VERIFY0(accord->lxpa_flags & LX_ACC_TOMBSTONE);
+	accord->lxpa_flags |= LX_ACC_TOMBSTONE;
+	lx_ptrace_accord_exit(accord);
+
+	/*
+	 * Walk the list of tracees, detaching them and setting them runnable
+	 * if they are stopped.
+	 */
+	for (;;) {
+		klwp_t *rlwp;
+		proc_t *rproc;
+		lx_lwp_data_t *remote;
+		kmutex_t *rmp;
+
+		mutex_enter(&accord->lxpa_tracees_lock);
+		if (list_is_empty(&accord->lxpa_tracees)) {
+			mutex_exit(&accord->lxpa_tracees_lock);
+			break;
+		}
+
+		/*
+		 * Fetch the first tracee LWP in the list and lock the process
+		 * which contains it.
+		 */
+		remote = list_head(&accord->lxpa_tracees);
+		rlwp = remote->br_lwp;
+		rproc = lwptoproc(rlwp);
+		/*
+		 * The p_lock mutex persists beyond the life of the process
+		 * itself.  We save the address, here, to prevent the need to
+		 * dereference the proc_t after awaking from sleep.
+		 */
+		rmp = &rproc->p_lock;
+		mutex_enter(rmp);
+
+		if (TRACEE_BUSY(remote)) {
+			/*
+			 * This LWP is currently detaching itself on exit, or
+			 * mid-way through stop().  We must wait for this
+			 * action to be completed.  While we wait on the CV, we
+			 * must drop the accord tracee list lock.
+			 */
+			mutex_exit(&accord->lxpa_tracees_lock);
+			cv_wait(&lx_ptrace_busy_cv, rmp);
+
+			/*
+			 * While we were waiting, some state may have changed.
+			 * Restart the walk to be sure we don't miss anything.
+			 */
+			mutex_exit(rmp);
+			continue;
+		}
+
+		/*
+		 * We now hold p_lock on the process.  Remove the tracee from
+		 * the list.
+		 */
+		VERIFY(list_link_active(&remote->br_ptrace_linkage));
+		list_remove(&accord->lxpa_tracees, remote);
+
+		/*
+		 * Unlink the accord and clear our trace flags.
+		 */
+		remote->br_ptrace_attach = LX_PTA_NONE;
+		remote->br_ptrace_tracer = NULL;
+		remote->br_ptrace_flags = 0;
+
+		/*
+		 * Let go of the list lock before we restart the LWP.  We must
+		 * not hold any locks other than the process p_lock when
+		 * we call lx_ptrace_restart_lwp() as it will thread_lock
+		 * the tracee.
+		 */
+		mutex_exit(&accord->lxpa_tracees_lock);
+
+		/*
+		 * Ensure that the LWP is not stopped on our account.
+		 */
+		lx_ptrace_restart_lwp(rlwp);
+
+		/*
+		 * Unlock the former tracee.
+		 */
+		mutex_exit(rmp);
+
+		/*
+		 * Drop the hold this tracee had on the accord.
+		 */
+		lx_ptrace_accord_enter(accord);
+		lx_ptrace_accord_rele(accord);
+		lx_ptrace_accord_exit(accord);
+	}
+
+	mutex_enter(&p->p_lock);
+	lwpd->br_ptrace_accord = NULL;
+	mutex_exit(&p->p_lock);
+
+	/*
+	 * Clean up and release our hold on the accord If we completely
+	 * detached all tracee LWPs, this will free the accord.  Otherwise, it
+	 * will be freed when they complete their cleanup.
+	 *
+	 * We hold "pidlock" while clearing these members for easy exclusion of
+	 * waitid(), etc.
+	 */
+	mutex_enter(&pidlock);
+	lx_ptrace_accord_enter(accord);
+	accord->lxpa_cvp = NULL;
+	accord->lxpa_tracer = NULL;
+	mutex_exit(&pidlock);
+	lx_ptrace_accord_rele(accord);
+	lx_ptrace_accord_exit(accord);
+}
+
+static void
+lx_ptrace_exit_tracee(proc_t *p, lx_lwp_data_t *lwpd,
+    lx_ptrace_accord_t *accord)
+{
+	VERIFY(MUTEX_NOT_HELD(&p->p_lock));
+
+	/*
+	 * We are the tracee LWP.  Lock the accord tracee list and then our
+	 * containing process.
+	 */
+	mutex_enter(&accord->lxpa_tracees_lock);
+	mutex_enter(&p->p_lock);
+
+	/*
+	 * Remove our reference to the accord.  We will release our hold
+	 * later.
+	 */
+	VERIFY(lwpd->br_ptrace_tracer == accord);
+	lwpd->br_ptrace_attach = LX_PTA_NONE;
+	lwpd->br_ptrace_tracer = NULL;
+
+	/*
+	 * Remove this LWP from the accord tracee list:
+	 */
+	VERIFY(list_link_active(&lwpd->br_ptrace_linkage));
+	list_remove(&accord->lxpa_tracees, lwpd);
+
+	/*
+	 * Wake up any tracers waiting for us to detach from the accord.
+	 */
+	cv_broadcast(&lx_ptrace_busy_cv);
+	mutex_exit(&p->p_lock);
+	mutex_exit(&accord->lxpa_tracees_lock);
+
+	/*
+	 * Grab "pidlock" and wake the tracer if it is blocked in waitid().
+	 */
+	mutex_enter(&pidlock);
+	if (accord->lxpa_cvp != NULL) {
+		cv_broadcast(accord->lxpa_cvp);
+	}
+	mutex_exit(&pidlock);
+
+	/*
+	 * Release our hold on the accord.
+	 */
+	lx_ptrace_accord_enter(accord);
+	lx_ptrace_accord_rele(accord);
+	lx_ptrace_accord_exit(accord);
+}
+
+/*
+ * This routine is called from lx_exitlwp() when an LWP is ready to exit.  If
+ * this LWP is being traced, it will be detached from the tracer's accord.  The
+ * routine will also detach any LWPs being traced by this LWP.
+ */
+void
+lx_ptrace_exit(proc_t *p, klwp_t *lwp)
+{
+	lx_lwp_data_t *lwpd = lwptolxlwp(lwp);
+	lx_ptrace_accord_t *accord;
+
+	VERIFY(MUTEX_HELD(&p->p_lock));
+
+	/*
+	 * Mark our LWP as exiting from a ptrace perspective.  This will
+	 * prevent a new accord from being allocated if one does not exist
+	 * already, and will make us invisible to PTRACE_ATTACH/PTRACE_TRACEME.
+	 */
+	VERIFY0(lwpd->br_ptrace_flags & LX_PTRACE_EXITING);
+	lwpd->br_ptrace_flags |= LX_PTRACE_EXITING;
+
+	if ((accord = lwpd->br_ptrace_tracer) != NULL) {
+		/*
+		 * We are traced by another LWP and must detach ourselves.
+		 */
+		mutex_exit(&p->p_lock);
+		lx_ptrace_exit_tracee(p, lwpd, accord);
+		mutex_enter(&p->p_lock);
+	}
+
+	if ((accord = lwpd->br_ptrace_accord) != NULL) {
+		/*
+		 * We have been tracing other LWPs, and must detach from
+		 * them and clean up our accord.
+		 */
+		mutex_exit(&p->p_lock);
+		lx_ptrace_exit_tracer(p, lwpd, accord);
+		mutex_enter(&p->p_lock);
+	}
+}
+
+/*
+ * Called when a SIGCLD signal is dispatched so that we may enqueue another.
+ * Return 0 if we enqueued a signal, or -1 if not.
+ */
+int
+lx_sigcld_repost(proc_t *pp, sigqueue_t *sqp)
+{
+	klwp_t *lwp = ttolwp(curthread);
+	lx_lwp_data_t *lwpd = lwptolxlwp(lwp);
+	lx_ptrace_accord_t *accord;
+	lx_lwp_data_t *remote;
+	klwp_t *rlwp;
+	proc_t *rproc;
+	boolean_t found = B_FALSE;
+
+	VERIFY(MUTEX_HELD(&pidlock));
+	VERIFY(MUTEX_NOT_HELD(&pp->p_lock));
+	VERIFY(lwptoproc(lwp) == pp);
+
+	mutex_enter(&pp->p_lock);
+	if ((accord = lwpd->br_ptrace_accord) == NULL) {
+		/*
+		 * This LWP is not a tracer LWP, so there will be no
+		 * SIGCLD.
+		 */
+		mutex_exit(&pp->p_lock);
+		return (-1);
+	}
+	mutex_exit(&pp->p_lock);
+
+	mutex_enter(&accord->lxpa_tracees_lock);
+	for (remote = list_head(&accord->lxpa_tracees); remote != NULL;
+	    remote = list_next(&accord->lxpa_tracees, remote)) {
+		rlwp = remote->br_lwp;
+		rproc = lwptoproc(rlwp);
+
+		/*
+		 * Check if this LWP is in "ptrace-stop".  If in the correct
+		 * stop condition, lock the process containing the tracee LWP.
+		 */
+		if (lx_ptrace_lock_if_stopped(accord, remote) != 0) {
+			continue;
+		}
+
+		if (remote->br_ptrace_flags & LX_PTRACE_PARENT_WAIT) {
+			/*
+			 * This event depends on waitid() clearing out the
+			 * event of another LWP.  Skip it for now.
+			 */
+			mutex_exit(&rproc->p_lock);
+			continue;
+		}
+
+		if (!(remote->br_ptrace_flags & LX_PTRACE_CLDPEND)) {
+			/*
+			 * No SIGCLD is required for this LWP.
+			 */
+			mutex_exit(&rproc->p_lock);
+			continue;
+		}
+
+		if (!(remote->br_ptrace_flags & LX_PTRACE_WAITPEND) ||
+		    remote->br_ptrace_whystop == 0 ||
+		    remote->br_ptrace_whatstop == 0) {
+			/*
+			 * No (new) stop reason to post for this LWP.
+			 */
+			mutex_exit(&rproc->p_lock);
+			continue;
+		}
+
+		/*
+		 * We found a process of interest.  Leave the process
+		 * containing the tracee LWP locked and break out of the loop.
+		 */
+		found = B_TRUE;
+		break;
+	}
+	mutex_exit(&accord->lxpa_tracees_lock);
+
+	if (!found) {
+		return (-1);
+	}
+
+	/*
+	 * Generate siginfo for this tracee LWP.
+	 */
+	lx_winfo(remote, &sqp->sq_info, B_FALSE, NULL, NULL);
+	remote->br_ptrace_flags &= ~LX_PTRACE_CLDPEND;
+	mutex_exit(&rproc->p_lock);
+
+	mutex_enter(&pp->p_lock);
+	if (sigismember(&pp->p_sig, SIGCLD)) {
+		mutex_exit(&pp->p_lock);
+
+		mutex_enter(&rproc->p_lock);
+		remote->br_ptrace_flags |= LX_PTRACE_CLDPEND;
+		mutex_exit(&rproc->p_lock);
+
+		return (-1);
+	}
+	sigaddqa(pp, curthread, sqp);
+	mutex_exit(&pp->p_lock);
+
+	return (0);
+}
+
+/*
+ * Consume the next available ptrace(2) event queued against the accord for
+ * this LWP.  The event will be emitted as if through waitid(), and converted
+ * by lx_waitpid() and friends before the return to usermode.
+ */
+int
+lx_waitid_helper(idtype_t idtype, id_t id, k_siginfo_t *ip, int options,
+    boolean_t *brand_wants_wait, int *rval)
+{
+	lx_ptrace_accord_t *accord;
+	klwp_t *lwp = ttolwp(curthread);
+	proc_t *p = lwptoproc(lwp);
+	lx_lwp_data_t *local = lwptolxlwp(lwp);
+	lx_lwp_data_t *remote;
+	boolean_t found = B_FALSE;
+	klwp_t *rlwp = NULL;
+	proc_t *rproc = NULL;
+	pid_t event_pid = 0, event_ppid = 0;
+	boolean_t waitflag = !(options & WNOWAIT);
+
+	VERIFY(MUTEX_HELD(&pidlock));
+	VERIFY(MUTEX_NOT_HELD(&p->p_lock));
+
+	/*
+	 * By default, we do not expect waitid() to block on our account.
+	 */
+	*brand_wants_wait = B_FALSE;
+
+	if (!local->br_waitid_emulate) {
+		/*
+		 * This waitid() call is not expecting emulated results.
+		 */
+		return (-1);
+	}
+
+	switch (idtype) {
+	case P_ALL:
+	case P_PID:
+	case P_PGID:
+		break;
+	default:
+		/*
+		 * This idtype has no power here.
+		 */
+		return (-1);
+	}
+
+	if (lx_ptrace_accord_get(&accord, B_FALSE) != 0) {
+		/*
+		 * This LWP does not have an accord; it cannot be tracing.
+		 */
+		return (-1);
+	}
+
+	/*
+	 * We do not need an additional hold on the accord as it belongs to
+	 * the running, tracer, LWP.
+	 */
+	lx_ptrace_accord_exit(accord);
+
+	mutex_enter(&accord->lxpa_tracees_lock);
+	if (list_is_empty(&accord->lxpa_tracees)) {
+		/*
+		 * Though it has an accord, there are currently no tracees in
+		 * the list for this LWP.
+		 */
+		mutex_exit(&accord->lxpa_tracees_lock);
+		return (-1);
+	}
+
+	/*
+	 * Walk the list of tracees and determine if any of them have events to
+	 * report.
+	 */
+	for (remote = list_head(&accord->lxpa_tracees); remote != NULL;
+	    remote = list_next(&accord->lxpa_tracees, remote)) {
+		rlwp = remote->br_lwp;
+		rproc = lwptoproc(rlwp);
+
+		/*
+		 * If the __WALL option was passed, we unconditionally consider
+		 * every possible child.
+		 */
+		if (!(local->br_waitid_flags & LX_WALL)) {
+			/*
+			 * Otherwise, we check to see if this LWP matches an
+			 * id we are waiting for.
+			 */
+			switch (idtype) {
+			case P_ALL:
+				break;
+			case P_PID:
+				if (remote->br_pid != id)
+					continue;
+				break;
+			case P_PGID:
+				if (rproc->p_pgrp != id)
+					continue;
+				break;
+			default:
+				cmn_err(CE_PANIC, "unexpected idtype: %d",
+				    idtype);
+			}
+		}
+
+		/*
+		 * Check if this LWP is in "ptrace-stop".  If in the correct
+		 * stop condition, lock the process containing the tracee LWP.
+		 */
+		if (lx_ptrace_lock_if_stopped(accord, remote) != 0) {
+			continue;
+		}
+
+		if (remote->br_ptrace_flags & LX_PTRACE_PARENT_WAIT) {
+			/*
+			 * This event depends on waitid() clearing out the
+			 * event of another LWP.  Skip it for now.
+			 */
+			mutex_exit(&rproc->p_lock);
+			continue;
+		}
+
+		if (!(remote->br_ptrace_flags & LX_PTRACE_WAITPEND) ||
+		    remote->br_ptrace_whystop == 0 ||
+		    remote->br_ptrace_whatstop == 0) {
+			/*
+			 * No (new) stop reason to post for this LWP.
+			 */
+			mutex_exit(&rproc->p_lock);
+			continue;
+		}
+
+		/*
+		 * We found a process of interest.  Leave the process
+		 * containing the tracee LWP locked and break out of the loop.
+		 */
+		found = B_TRUE;
+		break;
+	}
+	mutex_exit(&accord->lxpa_tracees_lock);
+
+	if (!found) {
+		/*
+		 * There were no events of interest, but we have tracees.
+		 * Signal to waitid() that it should block if the provided
+		 * flags allow for it.
+		 */
+		*brand_wants_wait = B_TRUE;
+		return (-1);
+	}
+
+	/*
+	 * Populate the signal information.
+	 */
+	lx_winfo(remote, ip, waitflag, &event_ppid, &event_pid);
+
+	/*
+	 * Unlock the tracee.
+	 */
+	mutex_exit(&rproc->p_lock);
+
+	if (event_pid != 0 && event_ppid != 0) {
+		/*
+		 * We need to do another pass around the tracee list and
+		 * unblock any events that have a "happens after" relationship
+		 * with this event.
+		 */
+		mutex_enter(&accord->lxpa_tracees_lock);
+		for (remote = list_head(&accord->lxpa_tracees); remote != NULL;
+		    remote = list_next(&accord->lxpa_tracees, remote)) {
+			rlwp = remote->br_lwp;
+			rproc = lwptoproc(rlwp);
+
+			mutex_enter(&rproc->p_lock);
+
+			if (remote->br_pid != event_pid ||
+			    remote->br_ppid != event_ppid) {
+				mutex_exit(&rproc->p_lock);
+				continue;
+			}
+
+			remote->br_ptrace_flags &= ~LX_PTRACE_PARENT_WAIT;
+
+			mutex_exit(&rproc->p_lock);
+		}
+		mutex_exit(&accord->lxpa_tracees_lock);
+	}
+
+	/*
+	 * If we are consuming this wait state, we remove the SIGCLD from
+	 * the queue and post another.
+	 */
+	if (waitflag) {
+		mutex_exit(&pidlock);
+		sigcld_delete(ip);
+		sigcld_repost();
+		mutex_enter(&pidlock);
+	}
+
+	*rval = 0;
+	return (0);
+}
+
+/*
+ * Some PTRACE_* requests are handled in-kernel by this function.  It is called
+ * through brandsys() via the B_PTRACE_KERNEL subcommand.
+ */
+int
+lx_ptrace_kernel(int ptrace_op, pid_t lxpid, uintptr_t addr, uintptr_t data)
+{
+	lx_lwp_data_t *local = ttolxlwp(curthread);
+	lx_ptrace_accord_t *accord;
+	lx_lwp_data_t *remote;
+	klwp_t *rlwp;
+	proc_t *rproc;
+	int error;
+	boolean_t found = B_FALSE;
+	boolean_t release_hold = B_FALSE;
+
+	_NOTE(ARGUNUSED(addr));
+
+	/*
+	 * These actions do not require the target LWP to be traced or stopped.
+	 */
+	switch (ptrace_op) {
+	case LX_PTRACE_TRACEME:
+		return (lx_ptrace_traceme());
+
+	case LX_PTRACE_ATTACH:
+		return (lx_ptrace_attach(lxpid));
+	}
+
+	/*
+	 * Ensure that we have an accord and obtain a lock on it.  This routine
+	 * should not fail because the LWP cannot make ptrace(2) system calls
+	 * after it has begun exiting.
+	 */
+	VERIFY0(local->br_ptrace_flags & LX_PTRACE_EXITING);
+	VERIFY(lx_ptrace_accord_get(&accord, B_TRUE) == 0);
+
+	/*
+	 * The accord belongs to this (the tracer) LWP, and we have a hold on
+	 * it.  We drop the lock so that we can take other locks.
+	 */
+	lx_ptrace_accord_exit(accord);
+
+	/*
+	 * Does the tracee list contain the pid in question?
+	 */
+	mutex_enter(&accord->lxpa_tracees_lock);
+	for (remote = list_head(&accord->lxpa_tracees); remote != NULL;
+	    remote = list_next(&accord->lxpa_tracees, remote)) {
+		if (remote->br_pid == lxpid) {
+			found = B_TRUE;
+			break;
+		}
+	}
+	if (!found) {
+		/*
+		 * The requested pid does not appear in the tracee list.
+		 */
+		mutex_exit(&accord->lxpa_tracees_lock);
+		return (ESRCH);
+	}
+
+	/*
+	 * Attempt to lock the target LWP.
+	 */
+	if ((error = lx_ptrace_lock_if_stopped(accord, remote)) != 0) {
+		/*
+		 * The LWP was not in "ptrace-stop".
+		 */
+		mutex_exit(&accord->lxpa_tracees_lock);
+		return (error);
+	}
+
+	/*
+	 * The target LWP is in "ptrace-stop".  We have the containing process
+	 * locked.
+	 */
+	rlwp = remote->br_lwp;
+	rproc = lwptoproc(rlwp);
+
+	/*
+	 * Process the ptrace(2) request:
+	 */
+	switch (ptrace_op) {
+	case LX_PTRACE_DETACH:
+		error = lx_ptrace_detach(accord, remote, (int)data,
+		    &release_hold);
+		break;
+
+	case LX_PTRACE_CONT:
+		error = lx_ptrace_cont(remote, LX_PTC_NONE, (int)data);
+		break;
+
+	case LX_PTRACE_SYSCALL:
+		error = lx_ptrace_cont(remote, LX_PTC_SYSCALL, (int)data);
+		break;
+
+	case LX_PTRACE_SINGLESTEP:
+		error = lx_ptrace_cont(remote, LX_PTC_SINGLESTEP, (int)data);
+		break;
+
+	case LX_PTRACE_SETOPTIONS:
+		error = lx_ptrace_setoptions(remote, data);
+		break;
+
+	case LX_PTRACE_GETEVENTMSG:
+		error = lx_ptrace_geteventmsg(remote, (void *)data);
+		break;
+
+	case LX_PTRACE_GETREGS:
+		error = lx_ptrace_getregs(remote, (void *)data);
+		break;
+
+	case LX_PTRACE_SETREGS:
+		error = lx_ptrace_setregs(remote, (void *)data);
+		break;
+
+	default:
+		error = EINVAL;
+	}
+
+	/*
+	 * Drop the lock on both the tracee process and the tracee list.
+	 */
+	mutex_exit(&rproc->p_lock);
+	mutex_exit(&accord->lxpa_tracees_lock);
+
+	if (release_hold) {
+		/*
+		 * Release a hold from the accord.
+		 */
+		lx_ptrace_accord_enter(accord);
+		lx_ptrace_accord_rele(accord);
+		lx_ptrace_accord_exit(accord);
+	}
+
+	return (error);
+}
+
+void
+lx_ptrace_init(void)
+{
+	cv_init(&lx_ptrace_busy_cv, NULL, CV_DEFAULT, NULL);
+
+	lx_ptrace_accord_cache = kmem_cache_create("lx_ptrace_accord",
+	    sizeof (lx_ptrace_accord_t), 0, NULL, NULL, NULL, NULL, NULL, 0);
+}
+
+void
+lx_ptrace_fini(void)
+{
+	cv_destroy(&lx_ptrace_busy_cv);
+
+	kmem_cache_destroy(lx_ptrace_accord_cache);
+}
diff --git a/usr/src/uts/common/brand/lx/os/lx_syscall.c b/usr/src/uts/common/brand/lx/os/lx_syscall.c
new file mode 100644
index 0000000000..cce902a943
--- /dev/null
+++ b/usr/src/uts/common/brand/lx/os/lx_syscall.c
@@ -0,0 +1,1220 @@
+/*
+ * CDDL HEADER START
+ *
+ * The contents of this file are subject to the terms of the
+ * Common Development and Distribution License (the "License").
+ * You may not use this file except in compliance with the License.
+ *
+ * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
+ * or http://www.opensolaris.org/os/licensing.
+ * See the License for the specific language governing permissions
+ * and limitations under the License.
+ *
+ * When distributing Covered Code, include this CDDL HEADER in each
+ * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
+ * If applicable, add the following below this CDDL HEADER, with the
+ * fields enclosed by brackets "[]" replaced with your own identifying
+ * information: Portions Copyright [yyyy] [name of copyright owner]
+ *
+ * CDDL HEADER END
+ */
+
+/*
+ * Copyright 2008 Sun Microsystems, Inc.  All rights reserved.
+ * Use is subject to license terms.
+ * Copyright 2015 Joyent, Inc.
+ */
+
+#include <sys/kmem.h>
+#include <sys/errno.h>
+#include <sys/thread.h>
+#include <sys/systm.h>
+#include <sys/syscall.h>
+#include <sys/proc.h>
+#include <sys/modctl.h>
+#include <sys/cmn_err.h>
+#include <sys/model.h>
+#include <sys/privregs.h>
+#include <sys/brand.h>
+#include <sys/machbrand.h>
+#include <sys/sdt.h>
+#include <sys/lx_syscalls.h>
+#include <sys/lx_brand.h>
+#include <sys/lx_impl.h>
+#include <sys/lx_misc.h>
+#include <lx_errno.h>
+
+
+/*
+ * Flags for sysent entries:
+ */
+#define	LX_SYS_NOSYS_REASON	0x07
+#define	LX_SYS_EBPARG6		0x08
+
+/*
+ * Flags that denote the specific reason we do not have a particular system
+ * call.  These reasons are only valid if the function is NULL.
+ */
+#define	NOSYS_USERMODE		0
+#define	NOSYS_NULL		1
+#define	NOSYS_NONE		2
+#define	NOSYS_NO_EQUIV		3
+#define	NOSYS_KERNEL		4
+#define	NOSYS_UNDOC		5
+#define	NOSYS_OBSOLETE		6
+#define	NOSYS_MAX		NOSYS_OBSOLETE
+
+#if NOSYS_MAX > LX_SYS_NOSYS_REASON
+#error NOSYS reason codes must fit in LX_SYS_NOSYS_REASON
+#endif
+
+/*
+ * Strings describing the reason we do not emulate a particular system call
+ * in the kernel.
+ */
+static char *nosys_reasons[] = {
+	NULL, /* NOSYS_USERMODE means this call is emulated in usermode */
+	"Not done yet",
+	"No such Linux system call",
+	"No equivalent illumos functionality",
+	"Reads/modifies Linux kernel state",
+	"Undocumented and/or rarely used system call",
+	"Unsupported, obsolete system call"
+};
+
+
+#if defined(_LP64)
+/*
+ * System call handler table and entry count for Linux x86_64 (amd64):
+ */
+lx_sysent_t lx_sysent64[LX_NSYSCALLS + 1];
+int lx_nsysent64;
+#endif
+/*
+ * System call handler table and entry count for Linux x86 (i386):
+ */
+lx_sysent_t lx_sysent32[LX_NSYSCALLS + 1];
+int lx_nsysent32;
+
+#if defined(_LP64)
+struct lx_vsyscall
+{
+	uintptr_t lv_addr;
+	uintptr_t lv_scnum;
+} lx_vsyscalls[] = {
+	{ LX_VSYS_gettimeofday, LX_SYS_gettimeofday },
+	{ LX_VSYS_time, LX_SYS_time },
+	{ LX_VSYS_getcpu, LX_SYS_getcpu },
+	{ NULL, NULL }
+};
+#endif
+
+#if defined(__amd64)
+static int
+lx_emulate_args(klwp_t *lwp, const lx_sysent_t *s, uintptr_t *args)
+{
+	struct regs *rp = lwptoregs(lwp);
+
+	if (get_udatamodel() == DATAMODEL_NATIVE) {
+		/*
+		 * Note: Syscall argument passing is different from function
+		 * call argument passing on amd64.  For function calls, the
+		 * fourth arg is passed via %rcx, but for system calls the 4th
+		 * arg is passed via %r10.  This is because in amd64, the
+		 * syscall instruction puts the lower 32 bits of %rflags in
+		 * %r11 and puts the %rip value to %rcx.
+		 *
+		 * Appendix A of the amd64 ABI (Linux conventions) states that
+		 * syscalls are limited to 6 args and no arg is passed on the
+		 * stack.
+		 */
+		args[0] = rp->r_rdi;
+		args[1] = rp->r_rsi;
+		args[2] = rp->r_rdx;
+		args[3] = rp->r_r10;
+		args[4] = rp->r_r8;
+		args[5] = rp->r_r9;
+	} else {
+		/*
+		 * If the system call takes 6 args, then libc has stashed them
+		 * in memory at the address contained in %ebx. Except for some
+		 * syscalls which store the 6th argument in %ebp.
+		 */
+		if (s->sy_narg == 6 && !(s->sy_flags & LX_SYS_EBPARG6)) {
+			uint32_t args32[6];
+
+			if (copyin((void *)rp->r_rbx, &args32,
+			    sizeof (args32)) != 0) {
+				/*
+				 * Clear the argument vector so that the
+				 * trace probe does not expose kernel
+				 * memory.
+				 */
+				bzero(args, 6 * sizeof (uintptr_t));
+				return (set_errno(EFAULT));
+			}
+
+			args[0] = args32[0];
+			args[1] = args32[1];
+			args[2] = args32[2];
+			args[3] = args32[3];
+			args[4] = args32[4];
+			args[5] = args32[5];
+		} else {
+			args[0] = rp->r_rbx;
+			args[1] = rp->r_rcx;
+			args[2] = rp->r_rdx;
+			args[3] = rp->r_rsi;
+			args[4] = rp->r_rdi;
+			args[5] = rp->r_rbp;
+		}
+	}
+
+	return (0);
+}
+
+#else	/* !__amd64 */
+
+static int
+lx_emulate_args(klwp_t *lwp, const lx_sysent_t *s, uintptr_t *args)
+{
+	struct regs *rp = lwptoregs(lwp);
+
+	/*
+	 * If the system call takes 6 args, then libc has stashed them
+	 * in memory at the address contained in %ebx. Except for some
+	 * syscalls which store the 6th argument in %ebp.
+	 */
+	if (s->sy_narg == 6 && !(s->sy_flags & LX_SYS_EBPARG6)) {
+		if (copyin((void *)rp->r_ebx, args, 6 * sizeof (uintptr_t)) !=
+		    0) {
+			/*
+			 * Clear the argument vector so that the trace probe
+			 * does not expose kernel memory.
+			 */
+			bzero(args, 6 * sizeof (uintptr_t));
+			return (set_errno(EFAULT));
+		}
+	} else {
+		args[0] = rp->r_ebx;
+		args[1] = rp->r_ecx;
+		args[2] = rp->r_edx;
+		args[3] = rp->r_esi;
+		args[4] = rp->r_edi;
+		args[5] = rp->r_ebp;
+	}
+
+	return (0);
+}
+#endif
+
+int
+lx_syscall_return(klwp_t *lwp, int syscall_num, long ret)
+{
+	lx_lwp_data_t *lwpd = lwptolxlwp(lwp);
+	struct regs *rp = lwptoregs(lwp);
+	int error = lwp->lwp_errno;
+
+	if (error != EINTR) {
+		/*
+		 * If this system call was not interrupted, clear the system
+		 * call restart flag before lx_setcontext() can pass it to
+		 * usermode.
+		 */
+		lwpd->br_syscall_restart = B_FALSE;
+	}
+
+	if (error != 0) {
+		/*
+		 * Convert from illumos to Linux errno:
+		 */
+		ret = -lx_errno(error, EINVAL);
+	}
+
+	/*
+	 * 32-bit Linux system calls return via %eax; 64-bit calls return via
+	 * %rax.
+	 */
+	rp->r_r0 = ret;
+
+	/*
+	 * Hold for the ptrace(2) "syscall-exit-stop" condition if required by
+	 * PTRACE_SYSCALL.  Note that the register state may be modified by
+	 * tracer.
+	 */
+	lx_ptrace_stop(LX_PR_SYSEXIT);
+
+	/*
+	 * Fire the DTrace "lx-syscall:::return" probe:
+	 */
+	lx_trace_sysreturn(syscall_num, ret);
+
+	/*
+	 * Clear errno for next time.  We do not clear "br_syscall_restart" or
+	 * "br_syscall_num" as they are potentially used by "lx_savecontext()"
+	 * in the signal delivery path.
+	 */
+	lwp->lwp_errno = 0;
+
+	lx_check_strict_failure(lwpd);
+
+	/*
+	 * We want complete control of the registers on return from this
+	 * emulated Linux system call:
+	 */
+	lwp->lwp_eosys = JUSTRETURN;
+	curthread->t_post_sys = 1;
+	aston(curthread);
+
+	return (0);
+}
+
+static void
+lx_syscall_unsup_msg(lx_sysent_t *s, int syscall_num, int unsup_reason)
+{
+	char buf[100];
+
+	if (s == NULL) {
+		(void) snprintf(buf, sizeof (buf), "NOSYS (%d): out of bounds",
+		    syscall_num);
+	} else {
+		VERIFY(unsup_reason < (sizeof (nosys_reasons) /
+		    sizeof (*nosys_reasons)));
+
+		if (s->sy_name == NULL) {
+			(void) snprintf(buf, sizeof (buf), "NOSYS (%d): %s",
+			    syscall_num, nosys_reasons[unsup_reason]);
+		} else {
+			(void) snprintf(buf, sizeof (buf), "NOSYS (%s): %s",
+			    s->sy_name, nosys_reasons[unsup_reason]);
+		}
+	}
+
+	lx_unsupported(buf);
+}
+
+/*
+ * This function is used to override the processing of arguments and
+ * invocation of a handler for emulated system calls, installed on each
+ * branded LWP as "lwp_brand_syscall".  If this system call should use the
+ * native path, we return 1.  If we handled this system call (and have made
+ * arrangements with respect to post-return usermode register state) we
+ * return 0.
+ */
+int
+lx_syscall_enter(void)
+{
+	klwp_t *lwp = ttolwp(curthread);
+	lx_lwp_data_t *lwpd = lwptolxlwp(lwp);
+	struct regs *rp = lwptoregs(lwp);
+	int syscall_num;
+	int error;
+	long ret = 0;
+	lx_sysent_t *s;
+	uintptr_t args[6];
+	unsigned int unsup_reason;
+
+	/*
+	 * If we got here, we should have an LWP-specific brand data
+	 * structure.
+	 */
+	VERIFY(lwpd != NULL);
+
+	if (lwpd->br_stack_mode != LX_STACK_MODE_BRAND) {
+		/*
+		 * The lwp is not in in BRAND execution mode, so we return
+		 * to the regular native system call path.
+		 */
+		DTRACE_PROBE(brand__lx__syscall__hook__skip);
+		return (1);
+	}
+
+	/*
+	 * Clear the restartable system call flag.  This flag will be set
+	 * on in the system call handler if the call is a candidate for
+	 * a restart.  It will be saved by lx_setcontext() in the event
+	 * that we take a signal, and used in the signal handling path
+	 * to restart the system call iff SA_RESTART was set for this
+	 * signal.  Save the system call number so that we can store it
+	 * in the saved context if required.
+	 */
+	lwpd->br_syscall_restart = B_FALSE;
+	lwpd->br_syscall_num = (int)rp->r_r0;
+
+	/*
+	 * Hold for the ptrace(2) "syscall-entry-stop" condition if traced by
+	 * PTRACE_SYSCALL.  The system call number and arguments may be
+	 * modified by the tracer.
+	 */
+	lx_ptrace_stop(LX_PR_SYSENTRY);
+
+	/*
+	 * Check that the system call number is within the bounds we expect.
+	 */
+	syscall_num = lwpd->br_syscall_num;
+	if (syscall_num < 0 || syscall_num > LX_MAX_SYSCALL(lwp)) {
+		lx_syscall_unsup_msg(NULL, syscall_num, 0);
+
+		set_errno(ENOTSUP);
+		lx_syscall_return(lwp, syscall_num, -1);
+		return (0);
+	}
+
+#if defined(_LP64)
+	if (lwp_getdatamodel(lwp) == DATAMODEL_NATIVE) {
+		s = &lx_sysent64[syscall_num];
+	} else
+#endif
+	{
+		s = &lx_sysent32[syscall_num];
+	}
+
+	/*
+	 * Process the arguments for this system call and fire the DTrace
+	 * "lx-syscall:::entry" probe:
+	 */
+	error = lx_emulate_args(lwp, s, args);
+	lx_trace_sysenter(syscall_num, args);
+	if (error != 0) {
+		/*
+		 * Could not read and process the arguments.  Return the error
+		 * to the process.
+		 */
+		set_errno(error);
+		lx_syscall_return(lwp, syscall_num, -1);
+		return (0);
+	}
+
+	if (s->sy_callc != NULL) {
+		/*
+		 * Call the in-kernel handler for this Linux system call:
+		 */
+		ret = s->sy_callc(args[0], args[1], args[2], args[3], args[4],
+		    args[5]);
+		lx_syscall_return(lwp, syscall_num, ret);
+		return (0);
+	}
+
+	/*
+	 * There is no in-kernel handler.
+	 */
+	switch (unsup_reason = (s->sy_flags & LX_SYS_NOSYS_REASON)) {
+	case NOSYS_USERMODE:
+		/*
+		 * Pass to the usermode emulation routine.
+		 */
+#if defined(_LP64)
+		if (get_udatamodel() != DATAMODEL_NATIVE) {
+			lx_emulate_user32(lwp, syscall_num, args);
+		} else
+#endif
+		{
+			lx_emulate_user(lwp, syscall_num, args);
+		}
+		return (0);
+
+	default:
+		/*
+		 * We are not emulating this system call at all.
+		 */
+		lx_syscall_unsup_msg(s, syscall_num, unsup_reason);
+
+		set_errno(ENOTSUP);
+		lx_syscall_return(lwp, syscall_num, -1);
+		return (0);
+	}
+}
+
+#if defined(_LP64)
+/*
+ * Emulate vsyscall support.
+ *
+ * Linux magically maps a single page into the address space of each process,
+ * allowing them to make 'vsyscalls'.  Originally designed to counteract the
+ * perceived overhead of regular system calls, vsyscalls were implemented as
+ * code residing in userspace which could be called directly.  The userspace
+ * implementations of these vsyscalls which have now been replaced by
+ * instructions which vector into the normal syscall path.
+ *
+ * Implementing vsyscalls on Illumos is complicated by the fact that the
+ * required static address region resides inside the kernel address space.
+ * Rather than mapping a user-accessible page into the KAS, a different
+ * approach is taken.  The vsyscall gate is emulated by interposing on
+ * pagefaults in trap().  An attempt to execute a known vsyscall address will
+ * result in emulating the appropriate system call rather than inducing a
+ * SIGSEGV.
+ */
+void
+lx_vsyscall_enter(proc_t *p, klwp_t *lwp, int scnum)
+{
+	struct regs *rp = lwptoregs(lwp);
+	uintptr_t raddr;
+
+	/*
+	 * Fetch the return address from the process stack.
+	 */
+	VERIFY(MUTEX_NOT_HELD(&p->p_lock));
+	if (copyin((void *)rp->r_rsp, &raddr, sizeof (raddr)) != 0) {
+#if DEBUG
+		printf("lx_vsyscall_call: bad brand stack at vsyscall "
+		    "cmd=%s, pid=%d, sp=0x%p\n", PTOU(p)->u_comm,
+		    p->p_pid, (void *)rp->r_rsp);
+#endif
+
+		/*
+		 * The process jumped to the vsyscall address without a
+		 * correctly configured stack.  Terminate the process.
+		 */
+		exit(CLD_KILLED, SIGSEGV);
+		return;
+	}
+
+	DTRACE_PROBE1(brand__lx__vsyscall, int, scnum);
+
+	/* Simulate vectoring into the syscall */
+	rp->r_rax = scnum;
+	rp->r_rip = raddr;
+	rp->r_rsp += sizeof (uintptr_t);
+
+	lx_syscall_enter();
+}
+
+boolean_t
+lx_vsyscall_iscall(klwp_t *lwp, uintptr_t addr, int *scnum)
+{
+	lx_lwp_data_t *lwpd = lwptolxlwp(lwp);
+	int i;
+
+	if (lwpd->br_stack_mode != LX_STACK_MODE_BRAND) {
+		/*
+		 * We only handle vsyscalls when running Linux code.
+		 */
+		return (B_FALSE);
+	}
+
+	if (addr < LX_VSYSCALL_ADDR ||
+	    addr >= (LX_VSYSCALL_ADDR + LX_VSYSCALL_SIZE)) {
+		/*
+		 * Ignore faults outside the vsyscall page.
+		 */
+		return (B_FALSE);
+	}
+
+	for (i = 0; lx_vsyscalls[i].lv_addr != NULL; i++) {
+		if (addr == lx_vsyscalls[i].lv_addr) {
+			/*
+			 * This is a valid vsyscall address.
+			 */
+			*scnum = lx_vsyscalls[i].lv_scnum;
+			return (B_TRUE);
+		}
+	}
+
+	lx_unsupported("bad vsyscall access");
+	return (B_FALSE);
+}
+#endif
+
+/*
+ * Linux defines system call numbers for 32-bit x86 in the file:
+ *   arch/x86/syscalls/syscall_32.tbl
+ */
+lx_sysent_t lx_sysent32[] = {
+	{"nosys",	NULL,			NOSYS_NONE,	0}, /*  0 */
+	{"exit",	NULL,			0,		1}, /*  1 */
+	{"fork",	NULL,			0,		0}, /*  2 */
+	{"read",	lx_read,		0,		3}, /*  3 */
+	{"write",	lx_write,		0,		3}, /*  4 */
+	{"open",	lx_open,		0,		3}, /*  5 */
+	{"close",	NULL,			0,		1}, /*  6 */
+	{"waitpid",	lx_waitpid,		0,		3}, /*  7 */
+	{"creat",	NULL,			0,		2}, /*  8 */
+	{"link",	NULL,			0,		2}, /*  9 */
+	{"unlink",	NULL,			0,		1}, /* 10 */
+	{"execve",	NULL,			0,		3}, /* 11 */
+	{"chdir",	NULL,			0,		1}, /* 12 */
+	{"time",	lx_time,		0,		1}, /* 13 */
+	{"mknod",	NULL,			0,		3}, /* 14 */
+	{"chmod",	lx_chmod,		0,		2}, /* 15 */
+	{"lchown16",	lx_lchown16,		0,		3}, /* 16 */
+	{"break",	NULL,			NOSYS_OBSOLETE,	0}, /* 17 */
+	{"stat",	NULL,			NOSYS_OBSOLETE,	0}, /* 18 */
+	{"lseek",	NULL,			0,		3}, /* 19 */
+	{"getpid",	lx_getpid,		0,		0}, /* 20 */
+	{"mount",	NULL,			0,		5}, /* 21 */
+	{"umount",	NULL,			0,		1}, /* 22 */
+	{"setuid16",	NULL,			0,		1}, /* 23 */
+	{"getuid16",	NULL,			0,		0}, /* 24 */
+	{"stime",	NULL,			0,		1}, /* 25 */
+	{"ptrace",	NULL,			0,		4}, /* 26 */
+	{"alarm",	NULL,			0,		1}, /* 27 */
+	{"fstat",	NULL,			NOSYS_OBSOLETE,	0}, /* 28 */
+	{"pause",	NULL,			0,		0}, /* 29 */
+	{"utime",	NULL,			0,		2}, /* 30 */
+	{"stty",	NULL,			NOSYS_OBSOLETE,	0}, /* 31 */
+	{"gtty",	NULL,			NOSYS_OBSOLETE,	0}, /* 32 */
+	{"access",	NULL,			0,		2}, /* 33 */
+	{"nice",	NULL,			0,		1}, /* 34 */
+	{"ftime",	NULL,			NOSYS_OBSOLETE,	0}, /* 35 */
+	{"sync",	NULL,			0, 		0}, /* 36 */
+	{"kill",	lx_kill,		0,		2}, /* 37 */
+	{"rename",	NULL,			0,		2}, /* 38 */
+	{"mkdir",	lx_mkdir,		0,		2}, /* 39 */
+	{"rmdir",	NULL,			0,		1}, /* 40 */
+	{"dup",		NULL,			0,		1}, /* 41 */
+	{"pipe",	lx_pipe,		0,		1}, /* 42 */
+	{"times",	NULL,			0,		1}, /* 43 */
+	{"prof",	NULL,			NOSYS_OBSOLETE,	0}, /* 44 */
+	{"brk",		lx_brk,			0,		1}, /* 45 */
+	{"setgid16",	NULL,			0,		1}, /* 46 */
+	{"getgid16",	NULL,			0,		0}, /* 47 */
+	{"signal",	NULL,			0,		2}, /* 48 */
+	{"geteuid16",	NULL,			0,		0}, /* 49 */
+	{"getegid16",	NULL,			0,		0}, /* 50 */
+	{"acct",	NULL,			NOSYS_NO_EQUIV,	0}, /* 51 */
+	{"umount2",	NULL,			0,		2}, /* 52 */
+	{"lock",	NULL,			NOSYS_OBSOLETE,	0}, /* 53 */
+	{"ioctl",	lx_ioctl,		0,		3}, /* 54 */
+	{"fcntl",	lx_fcntl,		0,		3}, /* 55 */
+	{"mpx",		NULL,			NOSYS_OBSOLETE,	0}, /* 56 */
+	{"setpgid",	NULL,			0,		2}, /* 57 */
+	{"ulimit",	NULL,			NOSYS_OBSOLETE,	0}, /* 58 */
+	{"olduname",	NULL,			NOSYS_OBSOLETE,	0}, /* 59 */
+	{"umask",	NULL,			0,		1}, /* 60 */
+	{"chroot",	NULL,			0,		1}, /* 61 */
+	{"ustat",	NULL,			NOSYS_OBSOLETE,	2}, /* 62 */
+	{"dup2",	NULL,			0,		2}, /* 63 */
+	{"getppid",	lx_getppid,		0,		0}, /* 64 */
+	{"getpgrp",	NULL,			0,		0}, /* 65 */
+	{"setsid",	NULL,			0,		0}, /* 66 */
+	{"sigaction",	NULL,			0,		3}, /* 67 */
+	{"sgetmask",	NULL,			NOSYS_OBSOLETE,	0}, /* 68 */
+	{"ssetmask",	NULL,			NOSYS_OBSOLETE,	0}, /* 69 */
+	{"setreuid16",	NULL,			0,		2}, /* 70 */
+	{"setregid16",	NULL,			0,		2}, /* 71 */
+	{"sigsuspend",	NULL,			0,		1}, /* 72 */
+	{"sigpending",	NULL,			0,		1}, /* 73 */
+	{"sethostname",	NULL,			0,		2}, /* 74 */
+	{"setrlimit",	lx_setrlimit,		0,		2}, /* 75 */
+	{"getrlimit",	lx_oldgetrlimit,	0,		2}, /* 76 */
+	{"getrusage",	NULL,			0,		2}, /* 77 */
+	{"gettimeofday", lx_gettimeofday,	0,		2}, /* 78 */
+	{"settimeofday", NULL, 			0,		2}, /* 79 */
+	{"getgroups16",	NULL,			0,		2}, /* 80 */
+	{"setgroups16",	NULL,			0,		2}, /* 81 */
+	{"select",	NULL,			NOSYS_OBSOLETE,	0}, /* 82 */
+	{"symlink",	NULL,			0,		2}, /* 83 */
+	{"oldlstat",	NULL,			NOSYS_OBSOLETE,	0}, /* 84 */
+	{"readlink",	NULL,			0,		3}, /* 85 */
+	{"uselib",	NULL,			NOSYS_KERNEL,	0}, /* 86 */
+	{"swapon",	NULL,			NOSYS_KERNEL,	0}, /* 87 */
+	{"reboot",	NULL,			0,		4}, /* 88 */
+	{"readdir",	NULL,			0,		3}, /* 89 */
+	{"mmap",	NULL,			0,		6}, /* 90 */
+	{"munmap",	NULL,			0,		2}, /* 91 */
+	{"truncate",	NULL,			0,		2}, /* 92 */
+	{"ftruncate",	NULL,			0,		2}, /* 93 */
+	{"fchmod",	lx_fchmod,		0,		2}, /* 94 */
+	{"fchown16",	lx_fchown16,		0,		3}, /* 95 */
+	{"getpriority",	NULL,			0,		2}, /* 96 */
+	{"setpriority",	NULL,			0,		3}, /* 97 */
+	{"profil",	NULL,			NOSYS_NO_EQUIV,	0}, /* 98 */
+	{"statfs",	NULL,			0,		2}, /* 99 */
+	{"fstatfs",	NULL,			0,		2}, /* 100 */
+	{"ioperm",	NULL,			NOSYS_NO_EQUIV,	0}, /* 101 */
+	{"socketcall",	NULL,			0,		2}, /* 102 */
+	{"syslog",	NULL,			0,		3}, /* 103 */
+	{"setitimer",	NULL,			0,		3}, /* 104 */
+	{"getitimer",	NULL,			0,		2}, /* 105 */
+	{"stat",	NULL,			0,		2}, /* 106 */
+	{"lstat",	NULL,			0,		2}, /* 107 */
+	{"fstat",	NULL,			0,		2}, /* 108 */
+	{"uname",	NULL,			NOSYS_OBSOLETE,	0}, /* 109 */
+	{"oldiopl",	NULL,			NOSYS_NO_EQUIV,	0}, /* 110 */
+	{"vhangup",	NULL,			0,		0}, /* 111 */
+	{"idle",	NULL,			NOSYS_NO_EQUIV,	0}, /* 112 */
+	{"vm86old",	NULL,			NOSYS_OBSOLETE,	0}, /* 113 */
+	{"wait4",	lx_wait4,		0,		4}, /* 114 */
+	{"swapoff",	NULL,			NOSYS_KERNEL,	0}, /* 115 */
+	{"sysinfo",	lx_sysinfo32,		0,		1}, /* 116 */
+	{"ipc",		NULL,			0,		5}, /* 117 */
+	{"fsync",	NULL,			0,		1}, /* 118 */
+	{"sigreturn",	NULL,			0,		1}, /* 119 */
+	{"clone",	NULL,			0,		5}, /* 120 */
+	{"setdomainname", NULL,			0,		2}, /* 121 */
+	{"uname",	NULL,			0,		1}, /* 122 */
+	{"modify_ldt",	lx_modify_ldt,		0,		3}, /* 123 */
+	{"adjtimex",	NULL,			0,		1}, /* 124 */
+	{"mprotect",	NULL,			0,		3}, /* 125 */
+	{"sigprocmask",	NULL,			0,		3}, /* 126 */
+	{"create_module", NULL,			NOSYS_KERNEL,	0}, /* 127 */
+	{"init_module",	NULL,			NOSYS_KERNEL,	0}, /* 128 */
+	{"delete_module", NULL,			NOSYS_KERNEL,	0}, /* 129 */
+	{"get_kernel_syms", NULL,		NOSYS_KERNEL,	0}, /* 130 */
+	{"quotactl",	NULL,			NOSYS_KERNEL,	0}, /* 131 */
+	{"getpgid",	NULL,			0,		1}, /* 132 */
+	{"fchdir",	NULL,			0,		1}, /* 133 */
+	{"bdflush",	NULL,			NOSYS_KERNEL,	0}, /* 134 */
+	{"sysfs",	NULL,			0,		3}, /* 135 */
+	{"personality",	NULL,			0,		1}, /* 136 */
+	{"afs_syscall",	NULL,			NOSYS_KERNEL,	0}, /* 137 */
+	{"setfsuid16",	NULL,			0,		1}, /* 138 */
+	{"setfsgid16",	NULL,			0,		1}, /* 139 */
+	{"llseek",	NULL,			0,		5}, /* 140 */
+	{"getdents",	lx_getdents_32,		0,		3}, /* 141 */
+	{"select",	NULL,			0,		5}, /* 142 */
+	{"flock",	NULL,			0,		2}, /* 143 */
+	{"msync",	NULL,			0,		3}, /* 144 */
+	{"readv",	NULL,			0,		3}, /* 145 */
+	{"writev",	NULL,			0,		3}, /* 146 */
+	{"getsid",	NULL,			0,		1}, /* 147 */
+	{"fdatasync",	NULL,			0,		1}, /* 148 */
+	{"sysctl",	NULL,			0,		1}, /* 149 */
+	{"mlock",	NULL,			0,		2}, /* 150 */
+	{"munlock",	NULL,			0,		2}, /* 151 */
+	{"mlockall",	NULL,			0,		1}, /* 152 */
+	{"munlockall",	NULL,			0,		0}, /* 153 */
+	{"sched_setparam", NULL,		0,		2}, /* 154 */
+	{"sched_getparam", NULL,		0,		2}, /* 155 */
+	{"sched_setscheduler", NULL,		0,		3}, /* 156 */
+	{"sched_getscheduler", NULL,		0,		1}, /* 157 */
+	{"sched_yield",	lx_sched_yield,		0,		0}, /* 158 */
+	{"sched_get_priority_max", NULL, 	0,		1}, /* 159 */
+	{"sched_get_priority_min", NULL, 	0,		1}, /* 160 */
+	{"sched_rr_get_interval", NULL, 	0,		2}, /* 161 */
+	{"nanosleep",	NULL,			0,		2}, /* 162 */
+	{"mremap",	NULL,			0,		5}, /* 163 */
+	{"setresuid16",	lx_setresuid16,		0,		3}, /* 164 */
+	{"getresuid16",	NULL,			0,		3}, /* 165 */
+	{"vm86",	NULL,			NOSYS_NO_EQUIV,	0}, /* 166 */
+	{"query_module", NULL,			0,		5}, /* 167 */
+	{"poll",	NULL,			0,		3}, /* 168 */
+	{"nfsservctl",	NULL,			NOSYS_KERNEL,	0}, /* 169 */
+	{"setresgid16",	lx_setresgid16,		0,		3}, /* 170 */
+	{"getresgid16",	NULL,			0,		3}, /* 171 */
+	{"prctl",	NULL,			0,		5}, /* 172 */
+	{"rt_sigreturn", NULL,			0,		0}, /* 173 */
+	{"rt_sigaction", NULL,			0,		4}, /* 174 */
+	{"rt_sigprocmask", NULL,		0,		4}, /* 175 */
+	{"rt_sigpending", NULL,			0,		2}, /* 176 */
+	{"rt_sigtimedwait", NULL,		0,		4}, /* 177 */
+	{"rt_sigqueueinfo", NULL,		0,		3}, /* 178 */
+	{"rt_sigsuspend", NULL,			0,		2}, /* 179 */
+	{"pread64",	NULL,			0,		5}, /* 180 */
+	{"pwrite64",	NULL,			0,		5}, /* 181 */
+	{"chown16",	lx_chown16,		0,		3}, /* 182 */
+	{"getcwd",	NULL,			0,		2}, /* 183 */
+	{"capget",	NULL,			0,		2}, /* 184 */
+	{"capset",	NULL,			0,		2}, /* 185 */
+	{"sigaltstack",	NULL,			0,		2}, /* 186 */
+	{"sendfile",	NULL,			0,		4}, /* 187 */
+	{"getpmsg",	NULL,			NOSYS_OBSOLETE,	0}, /* 188 */
+	{"putpmsg",	NULL,			NOSYS_OBSOLETE,	0}, /* 189 */
+	{"vfork",	NULL,			0,		0}, /* 190 */
+	{"getrlimit",	lx_getrlimit,		0,		2}, /* 191 */
+	{"mmap2",	NULL,			LX_SYS_EBPARG6,	6}, /* 192 */
+	{"truncate64",	NULL,			0,		3}, /* 193 */
+	{"ftruncate64",	NULL,			0,		3}, /* 194 */
+	{"stat64",	NULL,			0,		2}, /* 195 */
+	{"lstat64",	NULL,			0,		2}, /* 196 */
+	{"fstat64",	NULL,			0,		2}, /* 197 */
+	{"lchown",	lx_lchown,		0,		3}, /* 198 */
+	{"getuid",	NULL,			0,		0}, /* 199 */
+	{"getgid",	NULL,			0,		0}, /* 200 */
+	{"geteuid",	NULL,			0,		0}, /* 201 */
+	{"getegid",	NULL,			0,		0}, /* 202 */
+	{"setreuid",	NULL,			0,		0}, /* 203 */
+	{"setregid",	NULL,			0,		0}, /* 204 */
+	{"getgroups",	NULL,			0,		2}, /* 205 */
+	{"setgroups",	NULL,			0,		2}, /* 206 */
+	{"fchown",	lx_fchown,		0,		3}, /* 207 */
+	{"setresuid",	lx_setresuid,		0,		3}, /* 208 */
+	{"getresuid",	NULL,			0,		3}, /* 209 */
+	{"setresgid",	lx_setresgid,		0,		3}, /* 210 */
+	{"getresgid",	NULL,			0,		3}, /* 211 */
+	{"chown",	lx_chown,		0,		3}, /* 212 */
+	{"setuid",	NULL,			0,		1}, /* 213 */
+	{"setgid",	NULL,			0,		1}, /* 214 */
+	{"setfsuid",	NULL,			0,		1}, /* 215 */
+	{"setfsgid",	NULL,			0,		1}, /* 216 */
+	{"pivot_root",	NULL,			NOSYS_KERNEL,	0}, /* 217 */
+	{"mincore",	NULL,			0,		3}, /* 218 */
+	{"madvise",	NULL,			0,		3}, /* 219 */
+	{"getdents64",	lx_getdents64,		0,		3}, /* 220 */
+	{"fcntl64",	lx_fcntl64,		0,		3}, /* 221 */
+	{"tux",		NULL,			NOSYS_NO_EQUIV,	0}, /* 222 */
+	{"security",	NULL,			NOSYS_NO_EQUIV,	0}, /* 223 */
+	{"gettid",	lx_gettid,		0,		0}, /* 224 */
+	{"readahead",	NULL,			NOSYS_NO_EQUIV,	0}, /* 225 */
+	{"setxattr",	lx_xattr,		0,		5}, /* 226 */
+	{"lsetxattr",	lx_xattr,		0,		5}, /* 227 */
+	{"fsetxattr",	lx_xattr,		0,		5}, /* 228 */
+	{"getxattr",	lx_xattr,		0,		4}, /* 229 */
+	{"lgetxattr",	lx_xattr,		0,		4}, /* 230 */
+	{"fgetxattr",	lx_xattr,		0,		4}, /* 231 */
+	{"listxattr",	lx_xattr,		0,		3}, /* 232 */
+	{"llistxattr",	lx_xattr,		0,		3}, /* 233 */
+	{"flistxattr",	lx_xattr,		0,		3}, /* 234 */
+	{"removexattr",	lx_xattr,		0,		2}, /* 235 */
+	{"lremovexattr", lx_xattr,		0,		2}, /* 236 */
+	{"fremovexattr", lx_xattr,		0,		2}, /* 237 */
+	{"tkill",	lx_tkill,		0,		2}, /* 238 */
+	{"sendfile64",	NULL,			0,		4}, /* 239 */
+	{"futex",	lx_futex,		LX_SYS_EBPARG6,	6}, /* 240 */
+	{"sched_setaffinity", NULL, 		0,		3}, /* 241 */
+	{"sched_getaffinity", NULL, 		0,		3}, /* 242 */
+	{"set_thread_area", lx_set_thread_area,	0,		1}, /* 243 */
+	{"get_thread_area", lx_get_thread_area,	0,		1}, /* 244 */
+	{"io_setup",	NULL,			0,		2}, /* 245 */
+	{"io_destroy",	NULL,			0,		1}, /* 246 */
+	{"io_getevents", NULL,			0,		5}, /* 247 */
+	{"io_submit",	NULL,			0,		3}, /* 248 */
+	{"io_cancel",	NULL,			0,		3}, /* 249 */
+	{"fadvise64",	NULL,			0,		4}, /* 250 */
+	{"nosys",	NULL,			0,		0}, /* 251 */
+	{"group_exit",	NULL,			0,		1}, /* 252 */
+	{"lookup_dcookie", NULL,		NOSYS_NO_EQUIV,	0}, /* 253 */
+	{"epoll_create", NULL,			0,		1}, /* 254 */
+	{"epoll_ctl",	NULL,			0,		4}, /* 255 */
+	{"epoll_wait",	NULL,			0,		4}, /* 256 */
+	{"remap_file_pages", NULL,		NOSYS_NO_EQUIV,	0}, /* 257 */
+	{"set_tid_address", lx_set_tid_address,	0,		1}, /* 258 */
+	{"timer_create", NULL,			0,		3}, /* 259 */
+	{"timer_settime", NULL,			0,		4}, /* 260 */
+	{"timer_gettime", NULL,			0,		2}, /* 261 */
+	{"timer_getoverrun", NULL,		0,		1}, /* 262 */
+	{"timer_delete", NULL,			0,		1}, /* 263 */
+	{"clock_settime", lx_clock_settime,	0,		2}, /* 264 */
+	{"clock_gettime", lx_clock_gettime,	0,		2}, /* 265 */
+	{"clock_getres", lx_clock_getres,	0,		2}, /* 266 */
+	{"clock_nanosleep", NULL,		0,		4}, /* 267 */
+	{"statfs64",	NULL,			0,		2}, /* 268 */
+	{"fstatfs64",	NULL,			0,		2}, /* 269 */
+	{"tgkill",	lx_tgkill,		0,		3}, /* 270 */
+
+/*
+ * The following system calls only exist in kernel 2.6 and greater:
+ */
+	{"utimes",	NULL,			0,		2}, /* 271 */
+	{"fadvise64_64", NULL, 			0,		4}, /* 272 */
+	{"vserver",	NULL,			NOSYS_NULL,	0}, /* 273 */
+	{"mbind",	NULL,			NOSYS_NULL,	0}, /* 274 */
+	{"get_mempolicy", NULL,			NOSYS_NULL,	0}, /* 275 */
+	{"set_mempolicy", NULL,			NOSYS_NULL,	0}, /* 276 */
+	{"mq_open",	NULL,			NOSYS_NULL,	0}, /* 277 */
+	{"mq_unlink",	NULL,			NOSYS_NULL,	0}, /* 278 */
+	{"mq_timedsend", NULL,			NOSYS_NULL,	0}, /* 279 */
+	{"mq_timedreceive", NULL,		NOSYS_NULL,	0}, /* 280 */
+	{"mq_notify",	NULL,			NOSYS_NULL,	0}, /* 281 */
+	{"mq_getsetattr", NULL,			NOSYS_NULL,	0}, /* 282 */
+	{"kexec_load",	NULL,			NOSYS_NULL,	0}, /* 283 */
+	{"waitid",	lx_waitid,		0,		4}, /* 284 */
+	{"sys_setaltroot", NULL,		NOSYS_NULL,	0}, /* 285 */
+	{"add_key",	NULL,			NOSYS_NULL,	0}, /* 286 */
+	{"request_key",	NULL,			NOSYS_NULL,	0}, /* 287 */
+	{"keyctl",	NULL,			NOSYS_NULL,	0}, /* 288 */
+	{"ioprio_set",	NULL,			NOSYS_NULL,	0}, /* 289 */
+	{"ioprio_get",	NULL,			NOSYS_NULL,	0}, /* 290 */
+	{"inotify_init", NULL,			0,		0}, /* 291 */
+	{"inotify_add_watch", NULL,		0,		3}, /* 292 */
+	{"inotify_rm_watch", NULL,		0,		2}, /* 293 */
+	{"migrate_pages", NULL,			NOSYS_NULL,	0}, /* 294 */
+	{"openat",	lx_openat,		0,		4}, /* 295 */
+	{"mkdirat",	lx_mkdirat,		0,		3}, /* 296 */
+	{"mknodat",	NULL,			0,		4}, /* 297 */
+	{"fchownat",	lx_fchownat,		0,		5}, /* 298 */
+	{"futimesat",	NULL,			0,		3}, /* 299 */
+	{"fstatat64",	NULL,			0,		4}, /* 300 */
+	{"unlinkat",	NULL,			0,		3}, /* 301 */
+	{"renameat",	NULL,			0,		4}, /* 302 */
+	{"linkat",	NULL,			0,		5}, /* 303 */
+	{"symlinkat",	NULL,			0,		3}, /* 304 */
+	{"readlinkat",	NULL,			0,		4}, /* 305 */
+	{"fchmodat",	lx_fchmodat,		0,		3}, /* 306 */
+	{"faccessat",	NULL,			0,		4}, /* 307 */
+	{"pselect6",	NULL,			LX_SYS_EBPARG6,	6}, /* 308 */
+	{"ppoll",	NULL,			0,		5}, /* 309 */
+	{"unshare",	NULL,			NOSYS_NULL,	0}, /* 310 */
+	{"set_robust_list", lx_set_robust_list,	0,		2}, /* 311 */
+	{"get_robust_list", lx_get_robust_list,	0,		3}, /* 312 */
+	{"splice",	NULL,			NOSYS_NULL,	0}, /* 313 */
+	{"sync_file_range", NULL,		NOSYS_NULL,	0}, /* 314 */
+	{"tee",		NULL,			NOSYS_NULL,	0}, /* 315 */
+	{"vmsplice",	NULL,			NOSYS_NULL,	0}, /* 316 */
+	{"move_pages",	NULL,			NOSYS_NULL,	0}, /* 317 */
+	{"getcpu",	lx_getcpu,		0,		3}, /* 318 */
+	{"epoll_pwait",	NULL,			0,		5}, /* 319 */
+	{"utimensat",	NULL,			0,		4}, /* 320 */
+	{"signalfd",	NULL,			NOSYS_NULL,	0}, /* 321 */
+	{"timerfd_create", NULL,		0,		2}, /* 322 */
+	{"eventfd",	NULL,			0,		1}, /* 323 */
+	{"fallocate",	NULL,			NOSYS_NULL,	0}, /* 324 */
+	{"timerfd_settime", NULL,		0,		4}, /* 325 */
+	{"timerfd_gettime", NULL,		0,		2}, /* 326 */
+	{"signalfd4",	NULL,			NOSYS_NULL,	0}, /* 327 */
+	{"eventfd2",	NULL,			0,		2}, /* 328 */
+	{"epoll_create1", NULL,			0,		1}, /* 329 */
+	{"dup3",	NULL,			0,		3}, /* 330 */
+	{"pipe2",	lx_pipe2,		0,		2}, /* 331 */
+	{"inotify_init1", NULL,			0,		1}, /* 332 */
+	{"preadv",	NULL,			0,		4}, /* 333 */
+	{"pwritev",	NULL,			0,		4}, /* 334 */
+	{"rt_tgsigqueueinfo", NULL,		0,		4}, /* 335 */
+	{"perf_event_open", NULL,		NOSYS_NULL,	0}, /* 336 */
+	{"recvmmsg",	NULL,			NOSYS_NULL,	0}, /* 337 */
+	{"fanotify_init", NULL,			NOSYS_NULL,	0}, /* 338 */
+	{"fanotify_mark", NULL,			NOSYS_NULL,	0}, /* 339 */
+	{"prlimit64",	lx_prlimit64,		0,		4}, /* 340 */
+	{"name_to_handle_at", NULL,		NOSYS_NULL,	0}, /* 341 */
+	{"open_by_handle_at", NULL,		NOSYS_NULL,	0}, /* 342 */
+	{"clock_adjtime", NULL,			NOSYS_NULL,	0}, /* 343 */
+	{"syncfs",	NULL,			NOSYS_NULL,	0}, /* 344 */
+	{"sendmmsg",	NULL,			NOSYS_NULL,	0}, /* 345 */
+	{"setns",	NULL,			NOSYS_NULL,	0}, /* 346 */
+	{"process_vm_readv", NULL,		NOSYS_NULL,	0}, /* 347 */
+	{"process_vm_writev", NULL,		NOSYS_NULL,	0}, /* 348 */
+	{"kcmp",	NULL,			NOSYS_NULL,	0}, /* 349 */
+	{"finit_module", NULL,			NOSYS_NULL,	0}, /* 350 */
+	{"sched_setattr", NULL,			NOSYS_NULL,	0}, /* 351 */
+	{"sched_getattr", NULL,			NOSYS_NULL,	0}, /* 352 */
+	{"renameat2",	NULL,			NOSYS_NULL,	0}, /* 353 */
+	{"seccomp",	NULL,			NOSYS_NULL,	0}, /* 354 */
+	{"getrandom",	lx_getrandom,		0,		3}, /* 355 */
+	{"memfd_create", NULL,			NOSYS_NULL,	0}, /* 356 */
+	{"bpf",		NULL,			NOSYS_NULL,	0}, /* 357 */
+	{"execveat",	NULL,			NOSYS_NULL,	0}, /* 358 */
+};
+
+#if defined(_LP64)
+/*
+ * Linux defines system call numbers for 64-bit x86 in the file:
+ *   arch/x86/syscalls/syscall_64.tbl
+ */
+lx_sysent_t lx_sysent64[] = {
+	{"read",	lx_read,		0,		3}, /* 0 */
+	{"write",	lx_write,		0,		3}, /* 1 */
+	{"open",	lx_open,		0,		3}, /* 2 */
+	{"close",	NULL,			0,		1}, /* 3 */
+	{"stat",	NULL,			0,		2}, /* 4 */
+	{"fstat",	NULL,			0,		2}, /* 5 */
+	{"lstat",	NULL,			0,		2}, /* 6 */
+	{"poll",	NULL,			0,		3}, /* 7 */
+	{"lseek",	NULL,			0,		3}, /* 8 */
+	{"mmap",	NULL,			0,		6}, /* 9 */
+	{"mprotect",	NULL,			0,		3}, /* 10 */
+	{"munmap",	NULL,			0,		2}, /* 11 */
+	{"brk",		lx_brk,			0,		1}, /* 12 */
+	{"rt_sigaction", NULL,			0,		4}, /* 13 */
+	{"rt_sigprocmask", NULL,		0,		4}, /* 14 */
+	{"rt_sigreturn", NULL,			0,		0}, /* 15 */
+	{"ioctl",	lx_ioctl,		0,		3}, /* 16 */
+	{"pread64",	NULL,			0,		4}, /* 17 */
+	{"pwrite64",	NULL,			0,		4}, /* 18 */
+	{"readv",	NULL,			0,		3}, /* 19 */
+	{"writev",	NULL,			0,		3}, /* 20 */
+	{"access",	NULL,			0,		2}, /* 21 */
+	{"pipe",	lx_pipe,		0,		1}, /* 22 */
+	{"select",	NULL,			0,		5}, /* 23 */
+	{"sched_yield",	lx_sched_yield,		0,		0}, /* 24 */
+	{"mremap",	NULL,			0,		5}, /* 25 */
+	{"msync",	NULL,			0,		3}, /* 26 */
+	{"mincore",	NULL,			0,		3}, /* 27 */
+	{"madvise",	NULL,			0,		3}, /* 28 */
+	{"shmget",	NULL,			0,		3}, /* 29 */
+	{"shmat",	NULL,			0,		4}, /* 30 */
+	{"shmctl",	NULL,			0,		3}, /* 31 */
+	{"dup",		NULL,			0,		1}, /* 32 */
+	{"dup2",	NULL,			0,		2}, /* 33 */
+	{"pause",	NULL,			0,		0}, /* 34 */
+	{"nanosleep",	NULL,			0,		2}, /* 35 */
+	{"getitimer",	NULL,			0,		2}, /* 36 */
+	{"alarm",	NULL,			0,		1}, /* 37 */
+	{"setitimer",	NULL,			0,		3}, /* 38 */
+	{"getpid",	lx_getpid,		0,		0}, /* 39 */
+	{"sendfile",	NULL,			0,		4}, /* 40 */
+	{"socket",	NULL,			0,		3}, /* 41 */
+	{"connect",	NULL,			0,		3}, /* 42 */
+	{"accept",	NULL,			0,		3}, /* 43 */
+	{"sendto",	NULL,			0,		6}, /* 44 */
+	{"recvfrom",	NULL,			0,		6}, /* 45 */
+	{"sendmsg",	NULL,			0,		3}, /* 46 */
+	{"recvmsg",	NULL,			0,		3}, /* 47 */
+	{"shutdown",	NULL,			0,		2}, /* 48 */
+	{"bind",	NULL,			0,		3}, /* 49 */
+	{"listen",	NULL,			0,		2}, /* 50 */
+	{"getsockname",	NULL,			0,		3}, /* 51 */
+	{"getpeername",	NULL,			0,		3}, /* 52 */
+	{"socketpair",	NULL,			0,		4}, /* 53 */
+	{"setsockopt",	NULL,			0,		5}, /* 54 */
+	{"getsockopt",	NULL,			0,		5}, /* 55 */
+	{"clone",	NULL,			0,		5}, /* 56 */
+	{"fork",	NULL,			0,		0}, /* 57 */
+	{"vfork",	NULL,			0,		0}, /* 58 */
+	{"execve",	NULL,			0,		3}, /* 59 */
+	{"exit",	NULL,			0,		1}, /* 60 */
+	{"wait4",	lx_wait4,		0,		4}, /* 61 */
+	{"kill",	lx_kill,		0,		2}, /* 62 */
+	{"uname",	NULL,			0,		1}, /* 63 */
+	{"semget",	NULL,			0,		3}, /* 64 */
+	{"semop",	NULL,			0,		3}, /* 65 */
+	{"semctl",	NULL,			0,		4}, /* 66 */
+	{"shmdt",	NULL,			0,		1}, /* 67 */
+	{"msgget",	NULL,			0,		2}, /* 68 */
+	{"msgsnd",	NULL,			0,		4}, /* 69 */
+	{"msgrcv",	NULL,			0,		5}, /* 70 */
+	{"msgctl",	NULL,			0,		3}, /* 71 */
+	{"fcntl",	lx_fcntl64,		0,		3}, /* 72 */
+	{"flock",	NULL,			0,		2}, /* 73 */
+	{"fsync",	NULL,			0,		1}, /* 74 */
+	{"fdatasync",	NULL,			0,		1}, /* 75 */
+	{"truncate",	NULL,			0,		2}, /* 76 */
+	{"ftruncate",	NULL,			0,		2}, /* 77 */
+	{"getdents",	lx_getdents_64,		0,		3}, /* 78 */
+	{"getcwd",	NULL,			0,		2}, /* 79 */
+	{"chdir",	NULL,			0,		1}, /* 80 */
+	{"fchdir",	NULL,			0,		1}, /* 81 */
+	{"rename",	NULL,			0,		2}, /* 82 */
+	{"mkdir",	lx_mkdir,		0,		2}, /* 83 */
+	{"rmdir",	NULL,			0,		1}, /* 84 */
+	{"creat",	NULL,			0,		2}, /* 85 */
+	{"link",	NULL,			0,		2}, /* 86 */
+	{"unlink",	NULL,			0,		1}, /* 87 */
+	{"symlink",	NULL,			0,		2}, /* 88 */
+	{"readlink",	NULL,			0,		3}, /* 89 */
+	{"chmod",	lx_chmod,		0,		2}, /* 90 */
+	{"fchmod",	lx_fchmod,		0,		2}, /* 91 */
+	{"chown",	lx_chown,		0,		3}, /* 92 */
+	{"fchown",	lx_fchown,		0,		3}, /* 93 */
+	{"lchown",	lx_lchown,		0,		3}, /* 94 */
+	{"umask",	NULL,			0,		1}, /* 95 */
+	{"gettimeofday", lx_gettimeofday,	0,		2}, /* 96 */
+	{"getrlimit",	lx_getrlimit,		0,		2}, /* 97 */
+	{"getrusage",	NULL,			0,		2}, /* 98 */
+	{"sysinfo",	lx_sysinfo64,		0,		1}, /* 99 */
+	{"times",	NULL,			0,		1}, /* 100 */
+	{"ptrace",	NULL,			0,		4}, /* 101 */
+	{"getuid",	NULL,			0,		0}, /* 102 */
+	{"syslog",	NULL,			0,		3}, /* 103 */
+	{"getgid",	NULL,			0,		0}, /* 104 */
+	{"setuid",	NULL,			0,		1}, /* 105 */
+	{"setgid",	NULL,			0,		1}, /* 106 */
+	{"geteuid",	NULL,			0,		0}, /* 107 */
+	{"getegid",	NULL,			0,		0}, /* 108 */
+	{"setpgid",	NULL,			0,		2}, /* 109 */
+	{"getppid",	lx_getppid,		0,		0}, /* 110 */
+	{"getpgrp",	NULL,			0,		0}, /* 111 */
+	{"setsid",	NULL,			0,		0}, /* 112 */
+	{"setreuid",	NULL,			0,		0}, /* 113 */
+	{"setregid",	NULL,			0,		0}, /* 114 */
+	{"getgroups",	NULL,			0,		2}, /* 115 */
+	{"setgroups",	NULL,			0,		2}, /* 116 */
+	{"setresuid",	lx_setresuid,		0,		3}, /* 117 */
+	{"getresuid",	NULL,			0,		3}, /* 118 */
+	{"setresgid",	lx_setresgid,		0,		3}, /* 119 */
+	{"getresgid",	NULL,			0,		3}, /* 120 */
+	{"getpgid",	NULL,			0,		1}, /* 121 */
+	{"setfsuid",	NULL,			0,		1}, /* 122 */
+	{"setfsgid",	NULL,			0,		1}, /* 123 */
+	{"getsid",	NULL,			0,		1}, /* 124 */
+	{"capget",	NULL,			0,		2}, /* 125 */
+	{"capset",	NULL,			0,		2}, /* 126 */
+	{"rt_sigpending", NULL,			0,		2}, /* 127 */
+	{"rt_sigtimedwait", NULL,		0,		4}, /* 128 */
+	{"rt_sigqueueinfo", NULL,		0,		3}, /* 129 */
+	{"rt_sigsuspend", NULL,			0,		2}, /* 130 */
+	{"sigaltstack",	NULL,			0,		2}, /* 131 */
+	{"utime",	NULL,			0,		2}, /* 132 */
+	{"mknod",	NULL,			0,		3}, /* 133 */
+	{"uselib",	NULL,			NOSYS_KERNEL,	0}, /* 134 */
+	{"personality",	NULL,			0,		1}, /* 135 */
+	{"ustat",	NULL,			NOSYS_OBSOLETE,	2}, /* 136 */
+	{"statfs",	NULL,			0,		2}, /* 137 */
+	{"fstatfs",	NULL,			0,		2}, /* 138 */
+	{"sysfs",	NULL,			0,		3}, /* 139 */
+	{"getpriority",	NULL,			0,		2}, /* 140 */
+	{"setpriority",	NULL,			0,		3}, /* 141 */
+	{"sched_setparam", NULL,		0,		2}, /* 142 */
+	{"sched_getparam", NULL,		0,		2}, /* 143 */
+	{"sched_setscheduler", NULL,		0,		3}, /* 144 */
+	{"sched_getscheduler", NULL,		0,		1}, /* 145 */
+	{"sched_get_priority_max", NULL,	0,		1}, /* 146 */
+	{"sched_get_priority_min", NULL,	0,		1}, /* 147 */
+	{"sched_rr_get_interval", NULL,		0,		2}, /* 148 */
+	{"mlock",	NULL,			0,		2}, /* 149 */
+	{"munlock",	NULL,			0,		2}, /* 150 */
+	{"mlockall",	NULL,			0,		1}, /* 151 */
+	{"munlockall",	NULL,			0,		0}, /* 152 */
+	{"vhangup",	NULL,			0,		0}, /* 153 */
+	{"modify_ldt",	lx_modify_ldt,		0,		3}, /* 154 */
+	{"pivot_root",	NULL,			NOSYS_KERNEL,	0}, /* 155 */
+	{"sysctl",	NULL,			0,		1}, /* 156 */
+	{"prctl",	NULL,			0,		5}, /* 157 */
+	{"arch_prctl",	lx_arch_prctl,		0,		2}, /* 158 */
+	{"adjtimex",	NULL,			0,		1}, /* 159 */
+	{"setrlimit",	lx_setrlimit,		0,		2}, /* 160 */
+	{"chroot",	NULL,			0,		1}, /* 161 */
+	{"sync",	NULL,			0,		0}, /* 162 */
+	{"acct",	NULL,			NOSYS_NO_EQUIV,	0}, /* 163 */
+	{"settimeofday", NULL,			0,		2}, /* 164 */
+	{"mount",	NULL,			0,		5}, /* 165 */
+	{"umount2",	NULL,			0,		2}, /* 166 */
+	{"swapon",	NULL,			NOSYS_KERNEL,	0}, /* 167 */
+	{"swapoff",	NULL,			NOSYS_KERNEL,	0}, /* 168 */
+	{"reboot",	NULL,			0,		4}, /* 169 */
+	{"sethostname",	NULL,			0,		2}, /* 170 */
+	{"setdomainname", NULL,			0,		2}, /* 171 */
+	{"iopl",	NULL,			NOSYS_NO_EQUIV,	0}, /* 172 */
+	{"ioperm",	NULL,			NOSYS_NO_EQUIV,	0}, /* 173 */
+	{"create_module", NULL,			NOSYS_KERNEL,	0}, /* 174 */
+	{"init_module",	NULL,			NOSYS_KERNEL,	0}, /* 175 */
+	{"delete_module", NULL,			NOSYS_KERNEL,	0}, /* 176 */
+	{"get_kernel_syms", NULL,		NOSYS_KERNEL,	0}, /* 177 */
+	{"query_module", NULL,			0,		5}, /* 178 */
+	{"quotactl",	NULL,			NOSYS_KERNEL,	0}, /* 179 */
+	{"nfsservctl",	NULL,			NOSYS_KERNEL,	0}, /* 180 */
+	{"getpmsg",	NULL,			NOSYS_OBSOLETE,	0}, /* 181 */
+	{"putpmsg",	NULL,			NOSYS_OBSOLETE,	0}, /* 182 */
+	{"afs_syscall",	NULL,			NOSYS_KERNEL,	0}, /* 183 */
+	{"tux",		NULL,			NOSYS_NO_EQUIV,	0}, /* 184 */
+	{"security",	NULL,			NOSYS_NO_EQUIV,	0}, /* 185 */
+	{"gettid",	lx_gettid,		0,		0}, /* 186 */
+	{"readahead",	NULL,			NOSYS_NO_EQUIV,	0}, /* 187 */
+	{"setxattr",	lx_xattr,		0,		5}, /* 188 */
+	{"lsetxattr",	lx_xattr,		0,		5}, /* 189 */
+	{"fsetxattr",	lx_xattr,		0,		5}, /* 190 */
+	{"getxattr",	lx_xattr,		0,		4}, /* 191 */
+	{"lgetxattr",	lx_xattr,		0,		4}, /* 192 */
+	{"fgetxattr",	lx_xattr,		0,		4}, /* 193 */
+	{"listxattr",	lx_xattr,		0,		3}, /* 194 */
+	{"llistxattr",	lx_xattr,		0,		3}, /* 195 */
+	{"flistxattr",	lx_xattr,		0,		3}, /* 196 */
+	{"removexattr",	lx_xattr,		0,		2}, /* 197 */
+	{"lremovexattr", lx_xattr,		0,		2}, /* 198 */
+	{"fremovexattr", lx_xattr,		0,		2}, /* 199 */
+	{"tkill",	lx_tkill,		0,		2}, /* 200 */
+	{"time",	lx_time,		0,		1}, /* 201 */
+	{"futex",	lx_futex,		0,		6}, /* 202 */
+	{"sched_setaffinity", NULL,		0,		3}, /* 203 */
+	{"sched_getaffinity", NULL,		0,		3}, /* 204 */
+	{"set_thread_area", lx_set_thread_area, 0,		1}, /* 205 */
+	{"io_setup",	NULL,			0,		2}, /* 206 */
+	{"io_destroy",	NULL,			0,		1}, /* 207 */
+	{"io_getevents", NULL,			0,		5}, /* 208 */
+	{"io_submit",	NULL,			0,		3}, /* 209 */
+	{"io_cancel",	NULL,			0,		3}, /* 210 */
+	{"get_thread_area", lx_get_thread_area,	0,		1}, /* 211 */
+	{"lookup_dcookie", NULL,		NOSYS_NO_EQUIV,	0}, /* 212 */
+	{"epoll_create", NULL,			0,		1}, /* 213 */
+	{"epoll_ctl_old", NULL,			NOSYS_NULL,	0}, /* 214 */
+	{"epoll_wait_old", NULL,		NOSYS_NULL,	0}, /* 215 */
+	{"remap_file_pages", NULL,		NOSYS_NO_EQUIV,	0}, /* 216 */
+	{"getdents64",	lx_getdents64,		0,		3}, /* 217 */
+	{"set_tid_address", lx_set_tid_address, 0,		1}, /* 218 */
+	{"restart_syscall", NULL,		NOSYS_NULL,	0}, /* 219 */
+	{"semtimedop",	NULL,			0,		4}, /* 220 */
+	{"fadvise64",	NULL,			0,		4}, /* 221 */
+	{"timer_create", NULL,			0,		3}, /* 222 */
+	{"timer_settime", NULL,			0,		4}, /* 223 */
+	{"timer_gettime", NULL,			0,		2}, /* 224 */
+	{"timer_getoverrun", NULL,		0,		1}, /* 225 */
+	{"timer_delete", NULL,			0,		1}, /* 226 */
+	{"clock_settime", lx_clock_settime,	0,		2}, /* 227 */
+	{"clock_gettime", lx_clock_gettime,	0,		2}, /* 228 */
+	{"clock_getres", lx_clock_getres,	0,		2}, /* 229 */
+	{"clock_nanosleep", NULL,		0,		4}, /* 230 */
+	{"exit_group",	NULL,			0,		1}, /* 231 */
+	{"epoll_wait",	NULL,			0,		4}, /* 232 */
+	{"epoll_ctl",	NULL,			0,		4}, /* 233 */
+	{"tgkill",	lx_tgkill,		0,		3}, /* 234 */
+	{"utimes",	NULL,			0,		2}, /* 235 */
+	{"vserver",	NULL,			NOSYS_NULL,	0}, /* 236 */
+	{"mbind",	NULL,			NOSYS_NULL,	0}, /* 237 */
+	{"set_mempolicy", NULL,			NOSYS_NULL,	0}, /* 238 */
+	{"get_mempolicy", NULL,			NOSYS_NULL,	0}, /* 239 */
+	{"mq_open",	NULL,			NOSYS_NULL,	0}, /* 240 */
+	{"mq_unlink",	NULL,			NOSYS_NULL,	0}, /* 241 */
+	{"mq_timedsend", NULL,			NOSYS_NULL,	0}, /* 242 */
+	{"mq_timedreceive", NULL,		NOSYS_NULL,	0}, /* 243 */
+	{"mq_notify",	NULL,			NOSYS_NULL,	0}, /* 244 */
+	{"mq_getsetattr", NULL,			NOSYS_NULL,	0}, /* 245 */
+	{"kexec_load",	NULL,			NOSYS_NULL,	0}, /* 246 */
+	{"waitid",	lx_waitid,		0,		4}, /* 247 */
+	{"add_key",	NULL,			NOSYS_NULL,	0}, /* 248 */
+	{"request_key",	NULL,			NOSYS_NULL,	0}, /* 249 */
+	{"keyctl",	NULL,			NOSYS_NULL,	0}, /* 250 */
+	{"ioprio_set",	NULL,			NOSYS_NULL,	0}, /* 251 */
+	{"ioprio_get",	NULL,			NOSYS_NULL,	0}, /* 252 */
+	{"inotify_init", NULL,			0,		0}, /* 253 */
+	{"inotify_add_watch", NULL,		0,		3}, /* 254 */
+	{"inotify_rm_watch", NULL,		0,		2}, /* 255 */
+	{"migrate_pages", NULL,			NOSYS_NULL,	0}, /* 256 */
+	{"openat",	lx_openat,		0,		4}, /* 257 */
+	{"mkdirat",	lx_mkdirat,		0,		3}, /* 258 */
+	{"mknodat",	NULL,			0,		4}, /* 259 */
+	{"fchownat",	lx_fchownat,		0,		5}, /* 260 */
+	{"futimesat",	NULL,			0,		3}, /* 261 */
+	{"fstatat64",	NULL,			0,		4}, /* 262 */
+	{"unlinkat",	NULL,			0,		3}, /* 263 */
+	{"renameat",	NULL,			0,		4}, /* 264 */
+	{"linkat",	NULL,			0,		5}, /* 265 */
+	{"symlinkat",	NULL,			0,		3}, /* 266 */
+	{"readlinkat",	NULL,			0,		4}, /* 267 */
+	{"fchmodat",	lx_fchmodat,		0,		3}, /* 268 */
+	{"faccessat",	NULL,			0,		4}, /* 269 */
+	{"pselect6",	NULL,			0,		6}, /* 270 */
+	{"ppoll",	NULL,			0,		5}, /* 271 */
+	{"unshare",	NULL,			NOSYS_NULL,	0}, /* 272 */
+	{"set_robust_list", lx_set_robust_list,	0,		2}, /* 273 */
+	{"get_robust_list", lx_get_robust_list,	0,		3}, /* 274 */
+	{"splice",	NULL,			NOSYS_NULL,	0}, /* 275 */
+	{"tee",		NULL,			NOSYS_NULL,	0}, /* 276 */
+	{"sync_file_range", NULL,		NOSYS_NULL,	0}, /* 277 */
+	{"vmsplice",	NULL,			NOSYS_NULL,	0}, /* 278 */
+	{"move_pages",	NULL,			NOSYS_NULL,	0}, /* 279 */
+	{"utimensat",	NULL,			0,		4}, /* 280 */
+	{"epoll_pwait",	NULL,			0,		5}, /* 281 */
+	{"signalfd",	NULL,			NOSYS_NULL,	0}, /* 282 */
+	{"timerfd_create", NULL,		0,		2}, /* 283 */
+	{"eventfd",	NULL,			0,		1}, /* 284 */
+	{"fallocate",	NULL,			NOSYS_NULL,	0}, /* 285 */
+	{"timerfd_settime", NULL,		0,		4}, /* 286 */
+	{"timerfd_gettime", NULL,		0,		2}, /* 287 */
+	{"accept4",	NULL,			0,		4}, /* 288 */
+	{"signalfd4",	NULL,			NOSYS_NULL,	0}, /* 289 */
+	{"eventfd2",	NULL,			0,		2}, /* 290 */
+	{"epoll_create1", NULL,			0,		1}, /* 291 */
+	{"dup3",	NULL,			0,		3}, /* 292 */
+	{"pipe2",	lx_pipe2,		0,		2}, /* 293 */
+	{"inotify_init1", NULL,			0,		1}, /* 294 */
+	{"preadv",	NULL,			0,		4}, /* 295 */
+	{"pwritev",	NULL,			0,		4}, /* 296 */
+	{"rt_tgsigqueueinfo", NULL, 		0,		4}, /* 297 */
+	{"perf_event_open", NULL,		NOSYS_NULL,	0}, /* 298 */
+	{"recvmmsg",	NULL,			NOSYS_NULL,	0}, /* 299 */
+	{"fanotify_init", NULL,			NOSYS_NULL,	0}, /* 300 */
+	{"fanotify_mark", NULL,			NOSYS_NULL,	0}, /* 301 */
+	{"prlimit64",	lx_prlimit64,		0,		4}, /* 302 */
+	{"name_to_handle_at", NULL,		NOSYS_NULL,	0}, /* 303 */
+	{"open_by_handle_at", NULL,		NOSYS_NULL,	0}, /* 304 */
+	{"clock_adjtime", NULL,			NOSYS_NULL,	0}, /* 305 */
+	{"syncfs",	NULL,			NOSYS_NULL,	0}, /* 306 */
+	{"sendmmsg",	NULL,			NOSYS_NULL,	0}, /* 307 */
+	{"setns",	NULL,			NOSYS_NULL,	0}, /* 309 */
+	{"getcpu",	lx_getcpu,		0,		3}, /* 309 */
+	{"process_vm_readv", NULL,		NOSYS_NULL,	0}, /* 310 */
+	{"process_vm_writev", NULL,		NOSYS_NULL,	0}, /* 311 */
+	{"kcmp",	NULL,			NOSYS_NULL,	0}, /* 312 */
+	{"finit_module", NULL,			NOSYS_NULL,	0}, /* 313 */
+	{"sched_setattr", NULL,			NOSYS_NULL,	0}, /* 314 */
+	{"sched_getattr", NULL,			NOSYS_NULL,	0}, /* 315 */
+	{"renameat2", NULL,			NOSYS_NULL,	0}, /* 316 */
+	{"seccomp",	NULL,			NOSYS_NULL,	0}, /* 317 */
+	{"getrandom",	lx_getrandom,		0,		3}, /* 318 */
+	{"memfd_create", NULL,			NOSYS_NULL,	0}, /* 319 */
+	{"kexec_file_load", NULL,		NOSYS_NULL,	0}, /* 320 */
+	{"bpf",		NULL,			NOSYS_NULL,	0}, /* 321 */
+	{"execveat",	NULL,			NOSYS_NULL,	0}, /* 322 */
+
+	/* XXX TBD gap then x32 syscalls from 512 - 544 */
+};
+#endif
diff --git a/usr/src/uts/common/brand/lx/procfs/lx_proc.h b/usr/src/uts/common/brand/lx/procfs/lx_proc.h
new file mode 100644
index 0000000000..b383a4aef7
--- /dev/null
+++ b/usr/src/uts/common/brand/lx/procfs/lx_proc.h
@@ -0,0 +1,289 @@
+/*
+ * CDDL HEADER START
+ *
+ * The contents of this file are subject to the terms of the
+ * Common Development and Distribution License (the "License").
+ * You may not use this file except in compliance with the License.
+ *
+ * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
+ * or http://www.opensolaris.org/os/licensing.
+ * See the License for the specific language governing permissions
+ * and limitations under the License.
+ *
+ * When distributing Covered Code, include this CDDL HEADER in each
+ * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
+ * If applicable, add the following below this CDDL HEADER, with the
+ * fields enclosed by brackets "[]" replaced with your own identifying
+ * information: Portions Copyright [yyyy] [name of copyright owner]
+ *
+ * CDDL HEADER END
+ */
+/*
+ * Copyright 2009 Sun Microsystems, Inc.  All rights reserved.
+ * Use is subject to license terms.
+ * Copyright 2015 Joyent, Inc.
+ */
+
+#ifdef _LXPROC_NATIVE_H
+#error Attempted to include branded lx_proc.h after native lxproc.h
+#endif
+
+#ifndef	_LXPROC_H
+#define	_LXPROC_H
+#define	_LXPROC_BRANDED_H
+
+#ifdef	__cplusplus
+extern "C" {
+#endif
+
+/*
+ * lxproc.h: declarations, data structures and macros for lxprocfs
+ */
+
+
+#include <sys/types.h>
+#include <sys/param.h>
+#include <sys/policy.h>
+#include <sys/debug.h>
+#include <sys/dirent.h>
+#include <sys/errno.h>
+#include <sys/file.h>
+#include <sys/kmem.h>
+#include <sys/pathname.h>
+#include <sys/proc.h>
+#include <sys/systm.h>
+#include <sys/var.h>
+#include <sys/user.h>
+#include <sys/t_lock.h>
+#include <sys/sysmacros.h>
+#include <sys/cred.h>
+#include <sys/priv.h>
+#include <sys/vnode.h>
+#include <sys/vfs.h>
+#include <sys/statvfs.h>
+#include <sys/cmn_err.h>
+#include <sys/zone.h>
+#include <sys/uio.h>
+#include <sys/utsname.h>
+#include <sys/dnlc.h>
+#include <sys/atomic.h>
+#include <sys/sunddi.h>
+#include <sys/sunldi.h>
+#include <vm/as.h>
+#include <vm/anon.h>
+
+/*
+ * Convert a vnode into an lxpr_mnt_t
+ */
+#define	VTOLXPM(vp)	((lxpr_mnt_t *)(vp)->v_vfsp->vfs_data)
+
+/*
+ * convert a vnode into an lxpr_node
+ */
+#define	VTOLXP(vp)	((lxpr_node_t *)(vp)->v_data)
+
+/*
+ * convert a lxprnode into a vnode
+ */
+#define	LXPTOV(lxpnp)	((lxpnp)->lxpr_vnode)
+
+/*
+ * convert a lxpr_node into zone for fs
+ */
+#define	LXPTOZ(lxpnp) \
+	(((lxpr_mnt_t *)(lxpnp)->lxpr_vnode->v_vfsp->vfs_data)->lxprm_zone)
+
+#define	LXPNSIZ		256	/* max size of lx /proc file name entries */
+
+/*
+ * Pretend that a directory entry takes 16 bytes
+ */
+#define	LXPR_SDSIZE	16
+
+/*
+ * Node/file types for lx /proc files
+ * (directories and files contained therein).
+ */
+typedef enum lxpr_nodetype {
+	LXPR_PROCDIR,		/* /proc		*/
+	LXPR_PIDDIR,		/* /proc/<pid>		*/
+	LXPR_PID_CMDLINE,	/* /proc/<pid>/cmdline	*/
+	LXPR_PID_CPU,		/* /proc/<pid>/cpu	*/
+	LXPR_PID_CURDIR,	/* /proc/<pid>/cwd	*/
+	LXPR_PID_ENV,		/* /proc/<pid>/environ	*/
+	LXPR_PID_EXE,		/* /proc/<pid>/exe	*/
+	LXPR_PID_LIMITS,	/* /proc/<pid>/limits	*/
+	LXPR_PID_MAPS,		/* /proc/<pid>/maps	*/
+	LXPR_PID_MEM,		/* /proc/<pid>/mem	*/
+	LXPR_PID_MOUNTINFO,	/* /proc/<pid>/mountinfo */
+	LXPR_PID_ROOTDIR,	/* /proc/<pid>/root	*/
+	LXPR_PID_STAT,		/* /proc/<pid>/stat	*/
+	LXPR_PID_STATM,		/* /proc/<pid>/statm	*/
+	LXPR_PID_STATUS,	/* /proc/<pid>/status	*/
+	LXPR_PID_TASKDIR,	/* /proc/<pid>/task	*/
+	LXPR_PID_TASK_IDDIR,	/* /proc/<pid>/task/<tid>		*/
+	LXPR_PID_FDDIR,		/* /proc/<pid>/fd	*/
+	LXPR_PID_FD_FD,		/* /proc/<pid>/fd/nn	*/
+	LXPR_PID_TID_CMDLINE,	/* /proc/<pid>/task/<tid>/cmdline	*/
+	LXPR_PID_TID_CPU,	/* /proc/<pid>/task/<tid>/cpu		*/
+	LXPR_PID_TID_CURDIR,	/* /proc/<pid>/task/<tid>/cwd		*/
+	LXPR_PID_TID_ENV,	/* /proc/<pid>/task/<tid>/environ	*/
+	LXPR_PID_TID_EXE,	/* /proc/<pid>/task/<tid>/exe		*/
+	LXPR_PID_TID_LIMITS,	/* /proc/<pid>/task/<tid>/limits	*/
+	LXPR_PID_TID_MAPS,	/* /proc/<pid>/task/<tid>/maps		*/
+	LXPR_PID_TID_MEM,	/* /proc/<pid>/task/<tid>/mem		*/
+	LXPR_PID_TID_MOUNTINFO,	/* /proc/<pid>/task/<tid>/mountinfo	*/
+	LXPR_PID_TID_ROOTDIR,	/* /proc/<pid>/task/<tid>/root		*/
+	LXPR_PID_TID_STAT,	/* /proc/<pid>/task/<tid>/stat		*/
+	LXPR_PID_TID_STATM,	/* /proc/<pid>/task/<tid>/statm		*/
+	LXPR_PID_TID_STATUS,	/* /proc/<pid>/task/<tid>/status	*/
+	LXPR_PID_TID_FDDIR,	/* /proc/<pid>/task/<tid>/fd		*/
+	LXPR_PID_TID_FD_FD,	/* /proc/<pid>/task/<tid>/fd/nn		*/
+	LXPR_CMDLINE,		/* /proc/cmdline	*/
+	LXPR_CPUINFO,		/* /proc/cpuinfo	*/
+	LXPR_DEVICES,		/* /proc/devices	*/
+	LXPR_DISKSTATS,		/* /proc/diskstats	*/
+	LXPR_DMA,		/* /proc/dma		*/
+	LXPR_FILESYSTEMS,	/* /proc/filesystems	*/
+	LXPR_INTERRUPTS,	/* /proc/interrupts	*/
+	LXPR_IOPORTS,		/* /proc/ioports	*/
+	LXPR_KCORE,		/* /proc/kcore		*/
+	LXPR_KMSG,		/* /proc/kmsg		*/
+	LXPR_LOADAVG,		/* /proc/loadavg	*/
+	LXPR_MEMINFO,		/* /proc/meminfo	*/
+	LXPR_MODULES,		/* /proc/modules	*/
+	LXPR_MOUNTS,		/* /proc/mounts		*/
+	LXPR_NETDIR,		/* /proc/net		*/
+	LXPR_NET_ARP,		/* /proc/net/arp	*/
+	LXPR_NET_DEV,		/* /proc/net/dev	*/
+	LXPR_NET_DEV_MCAST,	/* /proc/net/dev_mcast	*/
+	LXPR_NET_IF_INET6,	/* /proc/net/if_inet6	*/
+	LXPR_NET_IGMP,		/* /proc/net/igmp	*/
+	LXPR_NET_IP_MR_CACHE,	/* /proc/net/ip_mr_cache */
+	LXPR_NET_IP_MR_VIF,	/* /proc/net/ip_mr_vif	*/
+	LXPR_NET_IPV6_ROUTE,	/* /proc/net/ipv6_route	*/
+	LXPR_NET_MCFILTER,	/* /proc/net/mcfilter	*/
+	LXPR_NET_NETSTAT,	/* /proc/net/netstat	*/
+	LXPR_NET_RAW,		/* /proc/net/raw	*/
+	LXPR_NET_ROUTE,		/* /proc/net/route	*/
+	LXPR_NET_RPC,		/* /proc/net/rpc	*/
+	LXPR_NET_RT_CACHE,	/* /proc/net/rt_cache	*/
+	LXPR_NET_SOCKSTAT,	/* /proc/net/sockstat	*/
+	LXPR_NET_SNMP,		/* /proc/net/snmp	*/
+	LXPR_NET_STAT,		/* /proc/net/stat	*/
+	LXPR_NET_TCP,		/* /proc/net/tcp	*/
+	LXPR_NET_TCP6,		/* /proc/net/tcp6	*/
+	LXPR_NET_UDP,		/* /proc/net/udp	*/
+	LXPR_NET_UDP6,		/* /proc/net/udp6	*/
+	LXPR_NET_UNIX,		/* /proc/net/unix	*/
+	LXPR_PARTITIONS,	/* /proc/partitions	*/
+	LXPR_SELF,		/* /proc/self		*/
+	LXPR_STAT,		/* /proc/stat		*/
+	LXPR_SWAPS,		/* /proc/swaps		*/
+	LXPR_SYSDIR,		/* /proc/sys/		*/
+	LXPR_SYS_FSDIR,		/* /proc/sys/fs/	*/
+	LXPR_SYS_FS_INOTIFYDIR,	/* /proc/sys/fs/inotify	*/
+	LXPR_SYS_FS_INOTIFY_MAX_QUEUED_EVENTS,	/* inotify/max_queued_events */
+	LXPR_SYS_FS_INOTIFY_MAX_USER_INSTANCES,	/* inotify/max_user_instances */
+	LXPR_SYS_FS_INOTIFY_MAX_USER_WATCHES,	/* inotify/max_user_watches */
+	LXPR_SYS_KERNELDIR,	/* /proc/sys/kernel/	*/
+	LXPR_SYS_KERNEL_HOSTNAME,	/* /proc/sys/kernel/hostname */
+	LXPR_SYS_KERNEL_MSGMNI,	/* /proc/sys/kernel/msgmni */
+	LXPR_SYS_KERNEL_NGROUPS_MAX,	/* /proc/sys/kernel/ngroups_max */
+	LXPR_SYS_KERNEL_PID_MAX,	/* /proc/sys/kernel/pid_max */
+	LXPR_SYS_KERNEL_SHMMAX,	/* /proc/sys/kernel/shmmax */
+	LXPR_SYS_KERNEL_THREADS_MAX,	/* /proc/sys/kernel/threads-max */
+	LXPR_UPTIME,		/* /proc/uptime		*/
+	LXPR_VERSION,		/* /proc/version	*/
+	LXPR_NFILES		/* number of lx /proc file types */
+} lxpr_nodetype_t;
+
+
+/*
+ * Number of fds allowed for in the inode number calculation
+ * per process (if a process has more fds then inode numbers
+ * may be duplicated)
+ */
+#define	LXPR_FD_PERPROC 2000
+
+/*
+ * Linux sector size for /proc/diskstats
+ */
+#define	LXPR_SECTOR_SIZE	512
+
+/*
+ * external dirent characteristics
+ */
+typedef struct {
+	lxpr_nodetype_t	d_type;
+	char		*d_name;
+} lxpr_dirent_t;
+
+/*
+ * This is the lxprocfs private data object
+ * which is attached to v_data in the vnode structure
+ */
+typedef struct lxpr_node {
+	lxpr_nodetype_t	lxpr_type;	/* type of this node 		*/
+	vnode_t		*lxpr_vnode;	/* vnode for the node		*/
+	vnode_t		*lxpr_parent;	/* parent directory		*/
+	vnode_t		*lxpr_realvp;	/* real vnode, file in dirs	*/
+	timestruc_t	lxpr_time;	/* creation etc time for file	*/
+	mode_t		lxpr_mode;	/* file mode bits		*/
+	uid_t		lxpr_uid;	/* file owner			*/
+	gid_t		lxpr_gid;	/* file group owner		*/
+	pid_t		lxpr_pid;	/* pid of proc referred to	*/
+	uint_t		lxpr_desc;	/* addl. descriptor (fd or tid)	*/
+	ino_t		lxpr_ino;	/* node id 			*/
+} lxpr_node_t;
+
+struct zone;    /* forward declaration */
+
+/*
+ * This is the lxprocfs private data object
+ * which is attached to vfs_data in the vfs structure
+ */
+typedef struct lxpr_mnt {
+	lxpr_node_t	*lxprm_node;	/* node at root of proc mount */
+	struct zone	*lxprm_zone;	/* zone for this mount */
+	ldi_ident_t	lxprm_li;	/* ident for ldi */
+} lxpr_mnt_t;
+
+extern vnodeops_t	*lxpr_vnodeops;
+extern int		nproc_highbit;	/* highbit(v.v_nproc)		*/
+
+typedef struct mounta	mounta_t;
+
+extern void lxpr_initnodecache();
+extern void lxpr_fininodecache();
+extern void lxpr_initrootnode(lxpr_node_t **, vfs_t *);
+extern ino_t lxpr_inode(lxpr_nodetype_t, pid_t, int);
+extern ino_t lxpr_parentinode(lxpr_node_t *);
+extern lxpr_node_t *lxpr_getnode(vnode_t *, lxpr_nodetype_t, proc_t *, int);
+extern void lxpr_freenode(lxpr_node_t *);
+
+typedef struct lxpr_uiobuf lxpr_uiobuf_t;
+extern lxpr_uiobuf_t *lxpr_uiobuf_new(uio_t *);
+extern void lxpr_uiobuf_free(lxpr_uiobuf_t *);
+extern int lxpr_uiobuf_flush(lxpr_uiobuf_t *);
+extern void lxpr_uiobuf_seek(lxpr_uiobuf_t *, offset_t);
+extern boolean_t lxpr_uiobuf_nonblock(lxpr_uiobuf_t *);
+extern void lxpr_uiobuf_write(lxpr_uiobuf_t *, const char *, size_t);
+extern void lxpr_uiobuf_printf(lxpr_uiobuf_t *, const char *, ...);
+extern void lxpr_uiobuf_seterr(lxpr_uiobuf_t *, int);
+
+proc_t *lxpr_lock(pid_t);
+void lxpr_unlock(proc_t *);
+
+#ifdef	__cplusplus
+}
+#endif
+
+#ifndef islower
+#define	islower(x)	(((unsigned)(x) >= 'a') && ((unsigned)(x) <= 'z'))
+#endif
+#ifndef toupper
+#define	toupper(x)	(islower(x) ? (x) - 'a' + 'A' : (x))
+#endif
+
+#endif /* _LXPROC_H */
diff --git a/usr/src/uts/common/brand/lx/procfs/lx_prsubr.c b/usr/src/uts/common/brand/lx/procfs/lx_prsubr.c
new file mode 100644
index 0000000000..6dad5ce7d2
--- /dev/null
+++ b/usr/src/uts/common/brand/lx/procfs/lx_prsubr.c
@@ -0,0 +1,570 @@
+/*
+ * CDDL HEADER START
+ *
+ * The contents of this file are subject to the terms of the
+ * Common Development and Distribution License (the "License").
+ * You may not use this file except in compliance with the License.
+ *
+ * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
+ * or http://www.opensolaris.org/os/licensing.
+ * See the License for the specific language governing permissions
+ * and limitations under the License.
+ *
+ * When distributing Covered Code, include this CDDL HEADER in each
+ * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
+ * If applicable, add the following below this CDDL HEADER, with the
+ * fields enclosed by brackets "[]" replaced with your own identifying
+ * information: Portions Copyright [yyyy] [name of copyright owner]
+ *
+ * CDDL HEADER END
+ */
+/*
+ * Copyright 2008 Sun Microsystems, Inc.  All rights reserved.
+ * Use is subject to license terms.
+ * Copyright 2015 Joyent, Inc.  All rights reserved.
+ */
+
+/*
+ * lxprsubr.c: Various functions for the /lxproc vnodeops.
+ */
+
+#include <sys/varargs.h>
+
+#include <sys/cpuvar.h>
+#include <sys/mman.h>
+#include <sys/vmsystm.h>
+#include <sys/prsystm.h>
+
+#include "lx_proc.h"
+
+#define	LXPRCACHE_NAME "lxbpr_cache"
+
+static int lxpr_node_constructor(void *, void *, int);
+static void lxpr_node_destructor(void *, void *);
+
+static kmem_cache_t *lxpr_node_cache;
+
+struct lxpr_uiobuf {
+	uio_t *uiop;
+	char *buffer;
+	uint32_t buffsize;
+	char *pos;
+	size_t beg;
+	int error;
+};
+
+int lx_pr_bufsize = 4000;
+
+struct lxpr_uiobuf *
+lxpr_uiobuf_new(uio_t *uiop)
+{
+	/* Allocate memory for both lxpr_uiobuf and output buffer */
+	int bufsize = lx_pr_bufsize;
+	struct lxpr_uiobuf *uiobuf =
+	    kmem_alloc(sizeof (struct lxpr_uiobuf) + bufsize, KM_SLEEP);
+
+	uiobuf->uiop = uiop;
+	uiobuf->buffer = (char *)&uiobuf[1];
+	uiobuf->buffsize = bufsize;
+	uiobuf->pos = uiobuf->buffer;
+	uiobuf->beg = 0;
+	uiobuf->error = 0;
+
+	return (uiobuf);
+}
+
+void
+lxpr_uiobuf_free(struct lxpr_uiobuf *uiobuf)
+{
+	ASSERT(uiobuf != NULL);
+	ASSERT(uiobuf->pos == uiobuf->buffer);
+
+	kmem_free(uiobuf, sizeof (struct lxpr_uiobuf) + uiobuf->buffsize);
+}
+
+void
+lxpr_uiobuf_seek(struct lxpr_uiobuf *uiobuf, offset_t offset)
+{
+	uiobuf->uiop->uio_offset = (off_t)offset;
+}
+
+boolean_t
+lxpr_uiobuf_nonblock(struct lxpr_uiobuf *uiobuf)
+{
+	if ((uiobuf->uiop->uio_fmode & FNONBLOCK) != 0)
+		return (B_TRUE);
+	return (B_FALSE);
+}
+
+void
+lxpr_uiobuf_seterr(struct lxpr_uiobuf *uiobuf, int err)
+{
+	ASSERT(uiobuf->error == 0);
+
+	uiobuf->error = err;
+}
+
+int
+lxpr_uiobuf_flush(struct lxpr_uiobuf *uiobuf)
+{
+	off_t off = uiobuf->uiop->uio_offset;
+	caddr_t uaddr = uiobuf->buffer;
+	size_t beg = uiobuf->beg;
+	size_t size = (uintptr_t)uiobuf->pos - (uintptr_t)uaddr;
+
+	if (uiobuf->error == 0 && uiobuf->uiop->uio_resid != 0) {
+		ASSERT(off >= beg);
+
+		if (beg + size > off && off >= 0)
+			uiobuf->error =
+			    uiomove(uaddr + (off - beg), size - (off - beg),
+			    UIO_READ, uiobuf->uiop);
+
+		uiobuf->beg += size;
+	}
+
+	uiobuf->pos = uaddr;
+
+	return (uiobuf->error);
+}
+
+void
+lxpr_uiobuf_write(struct lxpr_uiobuf *uiobuf, const char *buf, size_t size)
+{
+	/* While we can still carry on */
+	while (uiobuf->error == 0 && uiobuf->uiop->uio_resid != 0) {
+		uintptr_t remain = (uintptr_t)uiobuf->buffsize -
+		    ((uintptr_t)uiobuf->pos - (uintptr_t)uiobuf->buffer);
+
+		/* Enough space in buffer? */
+		if (remain >= size) {
+			bcopy(buf, uiobuf->pos, size);
+			uiobuf->pos += size;
+			return;
+		}
+
+		/* Not enough space, so copy all we can and try again */
+		bcopy(buf, uiobuf->pos, remain);
+		uiobuf->pos += remain;
+		(void) lxpr_uiobuf_flush(uiobuf);
+		buf += remain;
+		size -= remain;
+	}
+}
+
+#define	TYPBUFFSIZE 256
+
+void
+lxpr_uiobuf_printf(struct lxpr_uiobuf *uiobuf, const char *fmt, ...)
+{
+	va_list args;
+	char buff[TYPBUFFSIZE];
+	int len;
+	char *buffer;
+
+	/* Can we still do any output */
+	if (uiobuf->error != 0 || uiobuf->uiop->uio_resid == 0)
+		return;
+
+	va_start(args, fmt);
+
+	/* Try using stack allocated buffer */
+	len = vsnprintf(buff, TYPBUFFSIZE, fmt, args);
+	if (len < TYPBUFFSIZE) {
+		va_end(args);
+		lxpr_uiobuf_write(uiobuf, buff, len);
+		return;
+	}
+
+	/* Not enough space in pre-allocated buffer */
+	buffer = kmem_alloc(len + 1, KM_SLEEP);
+
+	/*
+	 * We know we allocated the correct amount of space
+	 * so no check on the return value
+	 */
+	(void) vsnprintf(buffer, len+1, fmt, args);
+	lxpr_uiobuf_write(uiobuf, buffer, len);
+	va_end(args);
+	kmem_free(buffer, len+1);
+}
+
+/*
+ * lxpr_lock():
+ *
+ * Lookup process from pid and return with p_plock and P_PR_LOCK held.
+ */
+proc_t *
+lxpr_lock(pid_t pid)
+{
+	proc_t *p;
+	kmutex_t *mp;
+	pid_t find_pid;
+
+	ASSERT(!MUTEX_HELD(&pidlock));
+
+	for (;;) {
+		mutex_enter(&pidlock);
+
+		/*
+		 * If the pid is 1, we really want the zone's init process;
+		 * if 0 we want zsched.
+		 */
+		if (pid == 1) {
+			find_pid = curproc->p_zone->zone_proc_initpid;
+		} else if (pid == 0) {
+			find_pid = curproc->p_zone->zone_zsched->p_pid;
+		} else {
+			find_pid = pid;
+		}
+		p = prfind(find_pid);
+
+		if (p == NULL || p->p_stat == SIDL) {
+			mutex_exit(&pidlock);
+			return (NULL);
+		}
+
+		/*
+		 * p_lock is persistent, but p itself is not -- it could
+		 * vanish during cv_wait().  Load p->p_lock now so we can
+		 * drop it after cv_wait() without referencing p.
+		 */
+		mp = &p->p_lock;
+		mutex_enter(mp);
+
+		mutex_exit(&pidlock);
+
+		if (p->p_flag & SEXITING) {
+			/*
+			 * This process is exiting -- let it go.
+			 */
+			mutex_exit(mp);
+			return (NULL);
+		}
+
+		if (!(p->p_proc_flag & P_PR_LOCK))
+			break;
+
+		cv_wait(&pr_pid_cv[p->p_slot], mp);
+		mutex_exit(mp);
+	}
+
+	p->p_proc_flag |= P_PR_LOCK;
+	THREAD_KPRI_REQUEST();
+	return (p);
+}
+
+/*
+ * lxpr_unlock()
+ *
+ * Unlock locked process
+ */
+void
+lxpr_unlock(proc_t *p)
+{
+	ASSERT(p->p_proc_flag & P_PR_LOCK);
+	ASSERT(MUTEX_HELD(&p->p_lock));
+	ASSERT(!MUTEX_HELD(&pidlock));
+
+	cv_signal(&pr_pid_cv[p->p_slot]);
+	p->p_proc_flag &= ~P_PR_LOCK;
+	mutex_exit(&p->p_lock);
+	THREAD_KPRI_RELEASE();
+}
+
+void
+lxpr_initnodecache()
+{
+	lxpr_node_cache = kmem_cache_create(LXPRCACHE_NAME,
+	    sizeof (lxpr_node_t), 0,
+	    lxpr_node_constructor, lxpr_node_destructor, NULL, NULL, NULL, 0);
+}
+
+void
+lxpr_fininodecache()
+{
+	kmem_cache_destroy(lxpr_node_cache);
+}
+
+/* ARGSUSED */
+static int
+lxpr_node_constructor(void *buf, void *un, int kmflags)
+{
+	lxpr_node_t	*lxpnp = buf;
+	vnode_t		*vp;
+
+	vp = lxpnp->lxpr_vnode = vn_alloc(kmflags);
+	if (vp == NULL)
+		return (-1);
+
+	(void) vn_setops(vp, lxpr_vnodeops);
+	vp->v_data = lxpnp;
+
+	return (0);
+}
+
+/* ARGSUSED */
+static void
+lxpr_node_destructor(void *buf, void *un)
+{
+	lxpr_node_t	*lxpnp = buf;
+
+	vn_free(LXPTOV(lxpnp));
+}
+
+/*
+ * Calculate an inode number
+ *
+ * This takes various bits of info and munges them
+ * to give the inode number for an lxproc node
+ */
+ino_t
+lxpr_inode(lxpr_nodetype_t type, pid_t pid, int desc)
+{
+	if (pid == 1) {
+		pid = curproc->p_zone->zone_proc_initpid;
+	} else if (pid == 0) {
+		pid = curproc->p_zone->zone_zsched->p_pid;
+	}
+
+	switch (type) {
+	case LXPR_PIDDIR:
+		return (maxpid + pid + 1);
+	case LXPR_PID_TASK_IDDIR:
+		return (maxpid + (desc * 10));
+	case LXPR_PROCDIR:
+		return (maxpid + 2);
+	case LXPR_PID_FD_FD:
+		return (maxpid + 2 +
+		    (pid * (LXPR_FD_PERPROC + LXPR_NFILES)) +
+		    LXPR_NFILES + desc);
+	default:
+		return (maxpid + 2 +
+		    (pid * (LXPR_FD_PERPROC + LXPR_NFILES)) +
+		    type);
+	}
+}
+
+/*
+ * Return inode number of parent (directory)
+ */
+ino_t
+lxpr_parentinode(lxpr_node_t *lxpnp)
+{
+	/*
+	 * If the input node is the root then the parent inode
+	 * is the mounted on inode so just return our inode number
+	 */
+	if (lxpnp->lxpr_type != LXPR_PROCDIR)
+		return (VTOLXP(lxpnp->lxpr_parent)->lxpr_ino);
+	else
+		return (lxpnp->lxpr_ino);
+}
+
+/*
+ * Allocate a new lxproc node
+ *
+ * This also allocates the vnode associated with it
+ */
+lxpr_node_t *
+lxpr_getnode(vnode_t *dp, lxpr_nodetype_t type, proc_t *p, int desc)
+{
+	lxpr_node_t *lxpnp;
+	vnode_t *vp;
+	user_t *up;
+	timestruc_t now;
+
+	/*
+	 * Allocate a new node. It is deallocated in vop_innactive
+	 */
+	lxpnp = kmem_cache_alloc(lxpr_node_cache, KM_SLEEP);
+
+	/*
+	 * Set defaults (may be overridden below)
+	 */
+	gethrestime(&now);
+	lxpnp->lxpr_type = type;
+	lxpnp->lxpr_realvp = NULL;
+	lxpnp->lxpr_parent = dp;
+	lxpnp->lxpr_desc = desc;
+	VN_HOLD(dp);
+	if (p != NULL) {
+		if (p->p_pid == curproc->p_zone->zone_proc_initpid) {
+			lxpnp->lxpr_pid = 1;
+		} else if (p->p_pid == curproc->p_zone->zone_zsched->p_pid) {
+			lxpnp->lxpr_pid = 0;
+		} else {
+			lxpnp->lxpr_pid = p->p_pid;
+		}
+
+		lxpnp->lxpr_time = PTOU(p)->u_start;
+		lxpnp->lxpr_uid = crgetruid(p->p_cred);
+		lxpnp->lxpr_gid = crgetrgid(p->p_cred);
+		lxpnp->lxpr_ino = lxpr_inode(type, p->p_pid, desc);
+	} else {
+		/* Pretend files without a proc belong to sched */
+		lxpnp->lxpr_pid = 0;
+		lxpnp->lxpr_time = now;
+		lxpnp->lxpr_uid = lxpnp->lxpr_gid = 0;
+		lxpnp->lxpr_ino = lxpr_inode(type, 0, 0);
+	}
+
+	/* initialize the vnode data */
+	vp = lxpnp->lxpr_vnode;
+	vn_reinit(vp);
+	vp->v_flag = VNOCACHE|VNOMAP|VNOSWAP|VNOMOUNT;
+	vp->v_vfsp = dp->v_vfsp;
+
+	/*
+	 * Do node specific stuff
+	 */
+	switch (type) {
+	case LXPR_PROCDIR:
+		vp->v_flag |= VROOT;
+		vp->v_type = VDIR;
+		lxpnp->lxpr_mode = 0555;	/* read-search by everyone */
+		break;
+
+	case LXPR_PID_CURDIR:
+		ASSERT(p != NULL);
+
+		/*
+		 * Zombie check.  p_stat is officially protected by pidlock,
+		 * but we can't grab pidlock here because we already hold
+		 * p_lock.  Luckily if we look at the process exit code
+		 * we see that p_stat only transisions from SRUN to SZOMB
+		 * while p_lock is held.  Aside from this, the only other
+		 * p_stat transition that we need to be aware about is
+		 * SIDL to SRUN, but that's not a problem since lxpr_lock()
+		 * ignores nodes in the SIDL state so we'll never get a node
+		 * that isn't already in the SRUN state.
+		 */
+		if (p->p_stat == SZOMB) {
+			lxpnp->lxpr_realvp = NULL;
+		} else {
+			up = PTOU(p);
+			lxpnp->lxpr_realvp = up->u_cdir;
+			ASSERT(lxpnp->lxpr_realvp != NULL);
+			VN_HOLD(lxpnp->lxpr_realvp);
+		}
+		vp->v_type = VLNK;
+		lxpnp->lxpr_mode = 0777;	/* anyone does anything ! */
+		break;
+
+	case LXPR_PID_ROOTDIR:
+		ASSERT(p != NULL);
+		/* Zombie check.  see locking comment above */
+		if (p->p_stat == SZOMB) {
+			lxpnp->lxpr_realvp = NULL;
+		} else {
+			up = PTOU(p);
+			lxpnp->lxpr_realvp =
+			    up->u_rdir != NULL ? up->u_rdir : rootdir;
+			ASSERT(lxpnp->lxpr_realvp != NULL);
+			VN_HOLD(lxpnp->lxpr_realvp);
+		}
+		vp->v_type = VLNK;
+		lxpnp->lxpr_mode = 0777;	/* anyone does anything ! */
+		break;
+
+	case LXPR_PID_EXE:
+		ASSERT(p != NULL);
+		lxpnp->lxpr_realvp = p->p_exec;
+		if (lxpnp->lxpr_realvp != NULL) {
+			VN_HOLD(lxpnp->lxpr_realvp);
+		}
+		vp->v_type = VLNK;
+		lxpnp->lxpr_mode = 0777;
+		break;
+
+	case LXPR_SELF:
+		vp->v_type = VLNK;
+		lxpnp->lxpr_mode = 0777;	/* anyone does anything ! */
+		break;
+
+	case LXPR_PID_TASKDIR:
+		ASSERT(p != NULL);
+		vp->v_type = VDIR;
+		lxpnp->lxpr_mode = 0555;	/* read-search by everyone */
+		break;
+
+	case LXPR_PID_TASK_IDDIR:
+		ASSERT(p != NULL);
+		vp->v_type = VDIR;
+		lxpnp->lxpr_mode = 0555;	/* read-search by everyone */
+		break;
+
+	case LXPR_PID_FD_FD:
+		ASSERT(p != NULL);
+		/* lxpr_realvp is set after we return */
+		lxpnp->lxpr_mode = 0700;	/* read-write-exe owner only */
+		vp->v_type = VLNK;
+		break;
+
+	case LXPR_PID_FDDIR:
+	case LXPR_PID_TID_FDDIR:
+		ASSERT(p != NULL);
+		vp->v_type = VDIR;
+		lxpnp->lxpr_mode = 0500;	/* read-search by owner only */
+		break;
+
+	case LXPR_PIDDIR:
+		ASSERT(p != NULL);
+		vp->v_type = VDIR;
+		lxpnp->lxpr_mode = 0511;
+		break;
+
+	case LXPR_SYSDIR:
+	case LXPR_SYS_FSDIR:
+	case LXPR_SYS_FS_INOTIFYDIR:
+	case LXPR_SYS_KERNELDIR:
+	case LXPR_NETDIR:
+		vp->v_type = VDIR;
+		lxpnp->lxpr_mode = 0555;	/* read-search by all */
+		break;
+
+	case LXPR_PID_ENV:
+	case LXPR_PID_MEM:
+		ASSERT(p != NULL);
+		/*FALLTHRU*/
+	case LXPR_KCORE:
+		vp->v_type = VREG;
+		lxpnp->lxpr_mode = 0400;	/* read-only by owner only */
+		break;
+
+	default:
+		vp->v_type = VREG;
+		lxpnp->lxpr_mode = 0444;	/* read-only by all */
+		break;
+	}
+
+	return (lxpnp);
+}
+
+
+/*
+ * Free the storage obtained from lxpr_getnode().
+ */
+void
+lxpr_freenode(lxpr_node_t *lxpnp)
+{
+	ASSERT(lxpnp != NULL);
+	ASSERT(LXPTOV(lxpnp) != NULL);
+
+	/*
+	 * delete any association with realvp
+	 */
+	if (lxpnp->lxpr_realvp != NULL)
+		VN_RELE(lxpnp->lxpr_realvp);
+
+	/*
+	 * delete any association with parent vp
+	 */
+	if (lxpnp->lxpr_parent != NULL)
+		VN_RELE(lxpnp->lxpr_parent);
+
+	/*
+	 * Release the lxprnode.
+	 */
+	kmem_cache_free(lxpr_node_cache, lxpnp);
+}
diff --git a/usr/src/uts/common/brand/lx/procfs/lx_prvfsops.c b/usr/src/uts/common/brand/lx/procfs/lx_prvfsops.c
new file mode 100644
index 0000000000..e9dda7864a
--- /dev/null
+++ b/usr/src/uts/common/brand/lx/procfs/lx_prvfsops.c
@@ -0,0 +1,374 @@
+/*
+ * CDDL HEADER START
+ *
+ * The contents of this file are subject to the terms of the
+ * Common Development and Distribution License (the "License").
+ * You may not use this file except in compliance with the License.
+ *
+ * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
+ * or http://www.opensolaris.org/os/licensing.
+ * See the License for the specific language governing permissions
+ * and limitations under the License.
+ *
+ * When distributing Covered Code, include this CDDL HEADER in each
+ * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
+ * If applicable, add the following below this CDDL HEADER, with the
+ * fields enclosed by brackets "[]" replaced with your own identifying
+ * information: Portions Copyright [yyyy] [name of copyright owner]
+ *
+ * CDDL HEADER END
+ */
+/*
+ * Copyright 2007 Sun Microsystems, Inc.  All rights reserved.
+ * Use is subject to license terms.
+ */
+
+#pragma ident	"%Z%%M%	%I%	%E% SMI"
+
+/*
+ * lxprvfsops.c: vfs operations for /lxprocfs.
+ */
+
+#include <sys/types.h>
+#include <sys/param.h>
+#include <sys/cmn_err.h>
+#include <sys/cred.h>
+#include <sys/debug.h>
+#include <sys/errno.h>
+#include <sys/proc.h>
+#include <sys/stat.h>
+#include <sys/statvfs.h>
+#include <sys/sysmacros.h>
+#include <sys/systm.h>
+#include <sys/var.h>
+#include <sys/vfs.h>
+#include <sys/vfs_opreg.h>
+#include <sys/vnode.h>
+#include <sys/mode.h>
+#include <sys/signal.h>
+#include <sys/user.h>
+#include <sys/mount.h>
+#include <sys/bitmap.h>
+#include <sys/kmem.h>
+#include <sys/policy.h>
+#include <sys/modctl.h>
+#include <sys/sunddi.h>
+#include <sys/sunldi.h>
+#include <sys/lx_impl.h>
+
+#include "lx_proc.h"
+
+/* Module level parameters */
+static int	lxprocfstype;
+static dev_t	lxprocdev;
+static kmutex_t	lxpr_mount_lock;
+
+int nproc_highbit;	/* highbit(v.v_nproc) */
+
+static int lxpr_mount(vfs_t *, vnode_t *, mounta_t *, cred_t *);
+static int lxpr_unmount(vfs_t *, int, cred_t *);
+static int lxpr_root(vfs_t *, vnode_t **);
+static int lxpr_statvfs(vfs_t *, statvfs64_t *);
+static int lxpr_init(int, char *);
+
+static vfsdef_t vfw = {
+	VFSDEF_VERSION,
+	"lx_proc",
+	lxpr_init,
+	VSW_ZMOUNT,
+	NULL
+};
+
+/*
+ * Module linkage information for the kernel.
+ */
+extern struct mod_ops mod_fsops;
+
+static struct modlfs modlfs = {
+	&mod_fsops, "lx brand procfs", &vfw
+};
+
+static struct modlinkage modlinkage = {
+	MODREV_1, (void *)&modlfs, NULL
+};
+
+int
+_init(void)
+{
+	return (mod_install(&modlinkage));
+}
+
+int
+_info(struct modinfo *modinfop)
+{
+	return (mod_info(&modlinkage, modinfop));
+}
+
+int
+_fini(void)
+{
+	int retval;
+
+	/*
+	 * attempt to unload the module
+	 */
+	if ((retval = mod_remove(&modlinkage)) != 0)
+		goto done;
+
+	/*
+	 * destroy lxpr_node cache
+	 */
+	lxpr_fininodecache();
+
+	/*
+	 * clean out the vfsops and vnodeops
+	 */
+	(void) vfs_freevfsops_by_type(lxprocfstype);
+	vn_freevnodeops(lxpr_vnodeops);
+
+	mutex_destroy(&lxpr_mount_lock);
+done:
+	return (retval);
+}
+
+static int
+lxpr_init(int fstype, char *name)
+{
+	static const fs_operation_def_t lxpr_vfsops_template[] = {
+		VFSNAME_MOUNT,		{ .vfs_mount = lxpr_mount },
+		VFSNAME_UNMOUNT,	{ .vfs_unmount = lxpr_unmount },
+		VFSNAME_ROOT,		{ .vfs_root = lxpr_root },
+		VFSNAME_STATVFS,	{ .vfs_statvfs = lxpr_statvfs },
+		NULL,			NULL
+	};
+	extern const fs_operation_def_t lxpr_vnodeops_template[];
+	int error;
+	major_t dev;
+
+	nproc_highbit = highbit(v.v_proc);
+	lxprocfstype = fstype;
+	ASSERT(lxprocfstype != 0);
+
+	mutex_init(&lxpr_mount_lock, NULL, MUTEX_DEFAULT, NULL);
+
+	/*
+	 * Associate VFS ops vector with this fstype.
+	 */
+	error = vfs_setfsops(fstype, lxpr_vfsops_template, NULL);
+	if (error != 0) {
+		cmn_err(CE_WARN, "lxpr_init: bad vfs ops template");
+		return (error);
+	}
+
+	/*
+	 * Set up vnode ops vector too.
+	 */
+	error = vn_make_ops(name, lxpr_vnodeops_template, &lxpr_vnodeops);
+	if (error != 0) {
+		(void) vfs_freevfsops_by_type(fstype);
+		cmn_err(CE_WARN, "lxpr_init: bad vnode ops template");
+		return (error);
+	}
+
+	/*
+	 * Assign a unique "device" number (used by stat(2)).
+	 */
+	if ((dev = getudev()) == (major_t)-1) {
+		cmn_err(CE_WARN, "lxpr_init: can't get unique device number");
+		dev = 0;
+	}
+
+	/*
+	 * Make the pseudo device
+	 */
+	lxprocdev = makedevice(dev, 0);
+
+	/*
+	 * Initialise cache for lxpr_nodes
+	 */
+	lxpr_initnodecache();
+
+	return (0);
+}
+
+static int
+lxpr_mount(vfs_t *vfsp, vnode_t *mvp, mounta_t *uap, cred_t *cr)
+{
+	lxpr_mnt_t *lxpr_mnt;
+	zone_t *zone = curproc->p_zone;
+	ldi_ident_t li;
+	int err;
+
+	/*
+	 * must be root to mount
+	 */
+	if (secpolicy_fs_mount(cr, mvp, vfsp) != 0)
+		return (EPERM);
+
+	/*
+	 * mount point must be a directory
+	 */
+	if (mvp->v_type != VDIR)
+		return (ENOTDIR);
+
+	if (zone == global_zone) {
+		zone_t *mntzone;
+
+		mntzone = zone_find_by_path(refstr_value(vfsp->vfs_mntpt));
+		zone_rele(mntzone);
+		if (zone != mntzone)
+			return (EBUSY);
+	}
+
+	/*
+	 * Having the resource be anything but "lxproc" doesn't make sense
+	 */
+	vfs_setresource(vfsp, "lxproc", 0);
+
+	lxpr_mnt = kmem_alloc(sizeof (*lxpr_mnt), KM_SLEEP);
+
+	if ((err = ldi_ident_from_mod(&modlinkage, &li)) != 0) {
+		kmem_free(lxpr_mnt, sizeof (*lxpr_mnt));
+		return (err);
+	}
+
+	lxpr_mnt->lxprm_li = li;
+
+	mutex_enter(&lxpr_mount_lock);
+
+	/*
+	 * Ensure we don't allow overlaying mounts
+	 */
+	mutex_enter(&mvp->v_lock);
+	if ((uap->flags & MS_OVERLAY) == 0 &&
+	    (mvp->v_count > 1 || (mvp->v_flag & VROOT))) {
+		mutex_exit(&mvp->v_lock);
+		mutex_exit(&lxpr_mount_lock);
+		kmem_free(lxpr_mnt, sizeof ((*lxpr_mnt)));
+		return (EBUSY);
+	}
+	mutex_exit(&mvp->v_lock);
+
+	/*
+	 * allocate the first vnode
+	 */
+	zone_hold(lxpr_mnt->lxprm_zone = zone);
+
+	/* Arbitrarily set the parent vnode to the mounted over directory */
+	lxpr_mnt->lxprm_node = lxpr_getnode(mvp, LXPR_PROCDIR, NULL, 0);
+
+	/* Correctly set the fs for the root node */
+	lxpr_mnt->lxprm_node->lxpr_vnode->v_vfsp = vfsp;
+
+	vfs_make_fsid(&vfsp->vfs_fsid, lxprocdev, lxprocfstype);
+	vfsp->vfs_bsize = DEV_BSIZE;
+	vfsp->vfs_fstype = lxprocfstype;
+	vfsp->vfs_data = (caddr_t)lxpr_mnt;
+	vfsp->vfs_dev = lxprocdev;
+
+	mutex_exit(&lxpr_mount_lock);
+
+	return (0);
+}
+
+static int
+lxpr_unmount(vfs_t *vfsp, int flag, cred_t *cr)
+{
+	lxpr_mnt_t *lxpr_mnt = (lxpr_mnt_t *)vfsp->vfs_data;
+	vnode_t *vp;
+	int count;
+
+	ASSERT(lxpr_mnt != NULL);
+	vp = LXPTOV(lxpr_mnt->lxprm_node);
+
+	mutex_enter(&lxpr_mount_lock);
+
+	/*
+	 * must be root to unmount
+	 */
+	if (secpolicy_fs_unmount(cr, vfsp) != 0) {
+		mutex_exit(&lxpr_mount_lock);
+		return (EPERM);
+	}
+
+	/*
+	 * forced unmount is not supported by this file system
+	 */
+	if (flag & MS_FORCE) {
+		mutex_exit(&lxpr_mount_lock);
+		return (ENOTSUP);
+	}
+
+	/*
+	 * Ensure that no vnodes are in use on this mount point.
+	 */
+	mutex_enter(&vp->v_lock);
+	count = vp->v_count;
+	mutex_exit(&vp->v_lock);
+	if (count > 1) {
+		mutex_exit(&lxpr_mount_lock);
+		return (EBUSY);
+	}
+
+
+	/*
+	 * purge the dnlc cache for vnode entries
+	 * associated with this file system
+	 */
+	count = dnlc_purge_vfsp(vfsp, 0);
+
+	/*
+	 * free up the lxprnode
+	 */
+	lxpr_freenode(lxpr_mnt->lxprm_node);
+	zone_rele(lxpr_mnt->lxprm_zone);
+	kmem_free(lxpr_mnt, sizeof (*lxpr_mnt));
+
+	mutex_exit(&lxpr_mount_lock);
+
+	return (0);
+}
+
+static int
+lxpr_root(vfs_t *vfsp, vnode_t **vpp)
+{
+	lxpr_node_t *lxpnp = ((lxpr_mnt_t *)vfsp->vfs_data)->lxprm_node;
+	vnode_t *vp = LXPTOV(lxpnp);
+
+	VN_HOLD(vp);
+	*vpp = vp;
+	return (0);
+}
+
+static int
+lxpr_statvfs(vfs_t *vfsp, statvfs64_t *sp)
+{
+	int n;
+	dev32_t d32;
+	extern uint_t nproc;
+
+	n = v.v_proc - nproc;
+
+	bzero((caddr_t)sp, sizeof (*sp));
+	sp->f_bsize	= DEV_BSIZE;
+	sp->f_frsize	= DEV_BSIZE;
+	sp->f_blocks	= (fsblkcnt64_t)0;
+	sp->f_bfree	= (fsblkcnt64_t)0;
+	sp->f_bavail	= (fsblkcnt64_t)0;
+	sp->f_files	= (fsfilcnt64_t)v.v_proc + 2;
+	sp->f_ffree	= (fsfilcnt64_t)n;
+	sp->f_favail	= (fsfilcnt64_t)n;
+	(void) cmpldev(&d32, vfsp->vfs_dev);
+	sp->f_fsid	= d32;
+	/* It is guaranteed that vsw_name will fit in f_basetype */
+	(void) strcpy(sp->f_basetype, vfssw[lxprocfstype].vsw_name);
+	sp->f_flag = vf_to_stf(vfsp->vfs_flag);
+	sp->f_namemax = 64;		/* quite arbitrary */
+	bzero(sp->f_fstr, sizeof (sp->f_fstr));
+
+	/* We know f_fstr is 32 chars */
+	(void) strcpy(sp->f_fstr, "/proc");
+	(void) strcpy(&sp->f_fstr[6], "/proc");
+
+	return (0);
+}
diff --git a/usr/src/uts/common/brand/lx/procfs/lx_prvnops.c b/usr/src/uts/common/brand/lx/procfs/lx_prvnops.c
new file mode 100644
index 0000000000..9b814f7ab4
--- /dev/null
+++ b/usr/src/uts/common/brand/lx/procfs/lx_prvnops.c
@@ -0,0 +1,5296 @@
+/*
+ * CDDL HEADER START
+ *
+ * The contents of this file are subject to the terms of the
+ * Common Development and Distribution License (the "License").
+ * You may not use this file except in compliance with the License.
+ *
+ * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
+ * or http://www.opensolaris.org/os/licensing.
+ * See the License for the specific language governing permissions
+ * and limitations under the License.
+ *
+ * When distributing Covered Code, include this CDDL HEADER in each
+ * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
+ * If applicable, add the following below this CDDL HEADER, with the
+ * fields enclosed by brackets "[]" replaced with your own identifying
+ * information: Portions Copyright [yyyy] [name of copyright owner]
+ *
+ * CDDL HEADER END
+ */
+/*
+ * Copyright 2010 Sun Microsystems, Inc.  All rights reserved.
+ * Use is subject to license terms.
+ * Copyright 2015 Joyent, Inc.  All rights reserved.
+ */
+
+/*
+ * lx_proc -- a Linux-compatible /proc for the LX brand
+ *
+ * We have -- confusingly -- two implementations of Linux /proc.  One is to
+ * support native (but Linux-borne) programs that wish to view the native
+ * system through the Linux /proc model; the other -- this one -- is to
+ * support Linux binaries via the LX brand.  These two implementations differ
+ * greatly in their aspirations (and their willingness to bend the truth
+ * of the system to accommodate those aspirations); they should not be unified.
+ */
+
+#include <sys/cpupart.h>
+#include <sys/cpuvar.h>
+#include <sys/session.h>
+#include <sys/vmparam.h>
+#include <sys/mman.h>
+#include <vm/rm.h>
+#include <vm/seg_vn.h>
+#include <sys/sdt.h>
+#include <lx_signum.h>
+#include <sys/strlog.h>
+#include <sys/stropts.h>
+#include <sys/cmn_err.h>
+#include <sys/lx_brand.h>
+#include <sys/x86_archext.h>
+#include <sys/archsystm.h>
+#include <sys/fp.h>
+#include <sys/pool_pset.h>
+#include <sys/pset.h>
+#include <sys/zone.h>
+#include <sys/pghw.h>
+#include <sys/vfs_opreg.h>
+#include <sys/param.h>
+#include <sys/utsname.h>
+#include <sys/rctl.h>
+#include <sys/kstat.h>
+#include <sys/lx_misc.h>
+#include <sys/brand.h>
+#include <sys/cred_impl.h>
+#include <sys/tihdr.h>
+#include <inet/ip.h>
+#include <inet/ip_ire.h>
+#include <inet/ip6.h>
+#include <inet/ip_if.h>
+#include <inet/tcp.h>
+#include <inet/udp_impl.h>
+#include <inet/ipclassifier.h>
+#include <sys/socketvar.h>
+#include <fs/sockfs/socktpi.h>
+
+/* Dependent on procfs */
+extern kthread_t *prchoose(proc_t *);
+extern int prreadargv(proc_t *, char *, size_t, size_t *);
+extern int prreadenvv(proc_t *, char *, size_t, size_t *);
+
+#include "lx_proc.h"
+
+extern pgcnt_t swapfs_minfree;
+extern time_t boot_time;
+
+/*
+ * Pointer to the vnode ops vector for this fs.
+ * This is instantiated in lxprinit() in lxpr_vfsops.c
+ */
+vnodeops_t *lxpr_vnodeops;
+
+static int lxpr_open(vnode_t **, int, cred_t *, caller_context_t *);
+static int lxpr_close(vnode_t *, int, int, offset_t, cred_t *,
+    caller_context_t *);
+static int lxpr_read(vnode_t *, uio_t *, int, cred_t *, caller_context_t *);
+static int lxpr_getattr(vnode_t *, vattr_t *, int, cred_t *,
+    caller_context_t *);
+static int lxpr_access(vnode_t *, int, int, cred_t *, caller_context_t *);
+static int lxpr_lookup(vnode_t *, char *, vnode_t **,
+    pathname_t *, int, vnode_t *, cred_t *, caller_context_t *, int *,
+    pathname_t *);
+static int lxpr_readdir(vnode_t *, uio_t *, cred_t *, int *,
+    caller_context_t *, int);
+static int lxpr_readlink(vnode_t *, uio_t *, cred_t *, caller_context_t *);
+static int lxpr_readlink_pid_fd(lxpr_node_t *lxpnp, char *bp, size_t len);
+static int lxpr_cmp(vnode_t *, vnode_t *, caller_context_t *);
+static int lxpr_realvp(vnode_t *, vnode_t **, caller_context_t *);
+static int lxpr_sync(void);
+static void lxpr_inactive(vnode_t *, cred_t *, caller_context_t *);
+
+static vnode_t *lxpr_lookup_procdir(vnode_t *, char *);
+static vnode_t *lxpr_lookup_piddir(vnode_t *, char *);
+static vnode_t *lxpr_lookup_not_a_dir(vnode_t *, char *);
+static vnode_t *lxpr_lookup_fddir(vnode_t *, char *);
+static vnode_t *lxpr_lookup_netdir(vnode_t *, char *);
+static vnode_t *lxpr_lookup_sysdir(vnode_t *, char *);
+static vnode_t *lxpr_lookup_sys_fsdir(vnode_t *, char *);
+static vnode_t *lxpr_lookup_sys_fs_inotifydir(vnode_t *, char *);
+static vnode_t *lxpr_lookup_sys_kerneldir(vnode_t *, char *);
+static vnode_t *lxpr_lookup_taskdir(vnode_t *, char *);
+static vnode_t *lxpr_lookup_task_tid_dir(vnode_t *, char *);
+
+static int lxpr_readdir_procdir(lxpr_node_t *, uio_t *, int *);
+static int lxpr_readdir_piddir(lxpr_node_t *, uio_t *, int *);
+static int lxpr_readdir_not_a_dir(lxpr_node_t *, uio_t *, int *);
+static int lxpr_readdir_fddir(lxpr_node_t *, uio_t *, int *);
+static int lxpr_readdir_netdir(lxpr_node_t *, uio_t *, int *);
+static int lxpr_readdir_sysdir(lxpr_node_t *, uio_t *, int *);
+static int lxpr_readdir_sys_fsdir(lxpr_node_t *, uio_t *, int *);
+static int lxpr_readdir_sys_fs_inotifydir(lxpr_node_t *, uio_t *, int *);
+static int lxpr_readdir_sys_kerneldir(lxpr_node_t *, uio_t *, int *);
+static int lxpr_readdir_taskdir(lxpr_node_t *, uio_t *, int *);
+static int lxpr_readdir_task_tid_dir(lxpr_node_t *, uio_t *, int *);
+
+static void lxpr_read_invalid(lxpr_node_t *, lxpr_uiobuf_t *);
+static void lxpr_read_empty(lxpr_node_t *, lxpr_uiobuf_t *);
+static void lxpr_read_cpuinfo(lxpr_node_t *, lxpr_uiobuf_t *);
+static void lxpr_read_diskstats(lxpr_node_t *, lxpr_uiobuf_t *);
+static void lxpr_read_isdir(lxpr_node_t *, lxpr_uiobuf_t *);
+static void lxpr_read_fd(lxpr_node_t *, lxpr_uiobuf_t *);
+static void lxpr_read_kmsg(lxpr_node_t *, lxpr_uiobuf_t *, ldi_handle_t);
+static void lxpr_read_loadavg(lxpr_node_t *, lxpr_uiobuf_t *);
+static void lxpr_read_meminfo(lxpr_node_t *, lxpr_uiobuf_t *);
+static void lxpr_read_mounts(lxpr_node_t *, lxpr_uiobuf_t *);
+static void lxpr_read_partitions(lxpr_node_t *, lxpr_uiobuf_t *);
+static void lxpr_read_stat(lxpr_node_t *, lxpr_uiobuf_t *);
+static void lxpr_read_swaps(lxpr_node_t *, lxpr_uiobuf_t *);
+static void lxpr_read_uptime(lxpr_node_t *, lxpr_uiobuf_t *);
+static void lxpr_read_version(lxpr_node_t *, lxpr_uiobuf_t *);
+
+static void lxpr_read_pid_cmdline(lxpr_node_t *, lxpr_uiobuf_t *);
+static void lxpr_read_pid_env(lxpr_node_t *, lxpr_uiobuf_t *);
+static void lxpr_read_pid_limits(lxpr_node_t *, lxpr_uiobuf_t *);
+static void lxpr_read_pid_maps(lxpr_node_t *, lxpr_uiobuf_t *);
+static void lxpr_read_pid_mountinfo(lxpr_node_t *, lxpr_uiobuf_t *);
+static void lxpr_read_pid_stat(lxpr_node_t *, lxpr_uiobuf_t *);
+static void lxpr_read_pid_statm(lxpr_node_t *, lxpr_uiobuf_t *);
+static void lxpr_read_pid_status(lxpr_node_t *, lxpr_uiobuf_t *);
+
+static void lxpr_read_pid_tid_stat(lxpr_node_t *, lxpr_uiobuf_t *);
+static void lxpr_read_pid_tid_status(lxpr_node_t *, lxpr_uiobuf_t *);
+
+static void lxpr_read_net_arp(lxpr_node_t *, lxpr_uiobuf_t *);
+static void lxpr_read_net_dev(lxpr_node_t *, lxpr_uiobuf_t *);
+static void lxpr_read_net_dev_mcast(lxpr_node_t *, lxpr_uiobuf_t *);
+static void lxpr_read_net_if_inet6(lxpr_node_t *, lxpr_uiobuf_t *);
+static void lxpr_read_net_igmp(lxpr_node_t *, lxpr_uiobuf_t *);
+static void lxpr_read_net_ip_mr_cache(lxpr_node_t *, lxpr_uiobuf_t *);
+static void lxpr_read_net_ip_mr_vif(lxpr_node_t *, lxpr_uiobuf_t *);
+static void lxpr_read_net_ipv6_route(lxpr_node_t *, lxpr_uiobuf_t *);
+static void lxpr_read_net_mcfilter(lxpr_node_t *, lxpr_uiobuf_t *);
+static void lxpr_read_net_netstat(lxpr_node_t *, lxpr_uiobuf_t *);
+static void lxpr_read_net_raw(lxpr_node_t *, lxpr_uiobuf_t *);
+static void lxpr_read_net_route(lxpr_node_t *, lxpr_uiobuf_t *);
+static void lxpr_read_net_rpc(lxpr_node_t *, lxpr_uiobuf_t *);
+static void lxpr_read_net_rt_cache(lxpr_node_t *, lxpr_uiobuf_t *);
+static void lxpr_read_net_sockstat(lxpr_node_t *, lxpr_uiobuf_t *);
+static void lxpr_read_net_snmp(lxpr_node_t *, lxpr_uiobuf_t *);
+static void lxpr_read_net_stat(lxpr_node_t *, lxpr_uiobuf_t *);
+static void lxpr_read_net_tcp(lxpr_node_t *, lxpr_uiobuf_t *);
+static void lxpr_read_net_tcp6(lxpr_node_t *, lxpr_uiobuf_t *);
+static void lxpr_read_net_udp(lxpr_node_t *, lxpr_uiobuf_t *);
+static void lxpr_read_net_udp6(lxpr_node_t *, lxpr_uiobuf_t *);
+static void lxpr_read_net_unix(lxpr_node_t *, lxpr_uiobuf_t *);
+static void lxpr_read_sys_fs_inotify_max_queued_events(lxpr_node_t *,
+    lxpr_uiobuf_t *);
+static void lxpr_read_sys_fs_inotify_max_user_instances(lxpr_node_t *,
+    lxpr_uiobuf_t *);
+static void lxpr_read_sys_fs_inotify_max_user_watches(lxpr_node_t *,
+    lxpr_uiobuf_t *);
+static void lxpr_read_sys_kernel_hostname(lxpr_node_t *, lxpr_uiobuf_t *);
+static void lxpr_read_sys_kernel_msgmni(lxpr_node_t *, lxpr_uiobuf_t *);
+static void lxpr_read_sys_kernel_ngroups_max(lxpr_node_t *, lxpr_uiobuf_t *);
+static void lxpr_read_sys_kernel_pid_max(lxpr_node_t *, lxpr_uiobuf_t *);
+static void lxpr_read_sys_kernel_shmmax(lxpr_node_t *, lxpr_uiobuf_t *);
+static void lxpr_read_sys_kernel_threads_max(lxpr_node_t *, lxpr_uiobuf_t *);
+
+/*
+ * Simple conversion
+ */
+#define	btok(x)	((x) >> 10)			/* bytes to kbytes */
+#define	ptok(x)	((x) << (PAGESHIFT - 10))	/* pages to kbytes */
+
+#define	ttolxlwp(t)	((struct lx_lwp_data *)ttolwpbrand(t))
+
+extern rctl_hndl_t rc_zone_msgmni;
+extern rctl_hndl_t rc_zone_shmmax;
+#define	FOURGB	4294967295
+
+/*
+ * The maximum length of the concatenation of argument vector strings we
+ * will return to the user via the branded procfs. Likewise for the env vector.
+ */
+int lxpr_maxargvlen = 4096;
+int lxpr_maxenvvlen = 4096;
+
+/*
+ * The lx /proc vnode operations vector
+ */
+const fs_operation_def_t lxpr_vnodeops_template[] = {
+	VOPNAME_OPEN,		{ .vop_open = lxpr_open },
+	VOPNAME_CLOSE,		{ .vop_close = lxpr_close },
+	VOPNAME_READ,		{ .vop_read = lxpr_read },
+	VOPNAME_GETATTR,	{ .vop_getattr = lxpr_getattr },
+	VOPNAME_ACCESS,		{ .vop_access = lxpr_access },
+	VOPNAME_LOOKUP,		{ .vop_lookup = lxpr_lookup },
+	VOPNAME_READDIR,	{ .vop_readdir = lxpr_readdir },
+	VOPNAME_READLINK,	{ .vop_readlink = lxpr_readlink },
+	VOPNAME_FSYNC,		{ .error = lxpr_sync },
+	VOPNAME_SEEK,		{ .error = lxpr_sync },
+	VOPNAME_INACTIVE,	{ .vop_inactive = lxpr_inactive },
+	VOPNAME_CMP,		{ .vop_cmp = lxpr_cmp },
+	VOPNAME_REALVP,		{ .vop_realvp = lxpr_realvp },
+	NULL,			NULL
+};
+
+
+/*
+ * file contents of an lx /proc directory.
+ */
+static lxpr_dirent_t lx_procdir[] = {
+	{ LXPR_CMDLINE,		"cmdline" },
+	{ LXPR_CPUINFO,		"cpuinfo" },
+	{ LXPR_DEVICES,		"devices" },
+	{ LXPR_DISKSTATS,	"diskstats" },
+	{ LXPR_DMA,		"dma" },
+	{ LXPR_FILESYSTEMS,	"filesystems" },
+	{ LXPR_INTERRUPTS,	"interrupts" },
+	{ LXPR_IOPORTS,		"ioports" },
+	{ LXPR_KCORE,		"kcore" },
+	{ LXPR_KMSG,		"kmsg" },
+	{ LXPR_LOADAVG,		"loadavg" },
+	{ LXPR_MEMINFO,		"meminfo" },
+	{ LXPR_MODULES,		"modules" },
+	{ LXPR_MOUNTS,		"mounts" },
+	{ LXPR_NETDIR,		"net" },
+	{ LXPR_PARTITIONS,	"partitions" },
+	{ LXPR_SELF,		"self" },
+	{ LXPR_STAT,		"stat" },
+	{ LXPR_SWAPS,		"swaps" },
+	{ LXPR_SYSDIR,		"sys" },
+	{ LXPR_UPTIME,		"uptime" },
+	{ LXPR_VERSION,		"version" }
+};
+
+#define	PROCDIRFILES	(sizeof (lx_procdir) / sizeof (lx_procdir[0]))
+
+/*
+ * Contents of an lx /proc/<pid> directory.
+ */
+static lxpr_dirent_t piddir[] = {
+	{ LXPR_PID_CMDLINE,	"cmdline" },
+	{ LXPR_PID_CPU,		"cpu" },
+	{ LXPR_PID_CURDIR,	"cwd" },
+	{ LXPR_PID_ENV,		"environ" },
+	{ LXPR_PID_EXE,		"exe" },
+	{ LXPR_PID_LIMITS,	"limits" },
+	{ LXPR_PID_MAPS,	"maps" },
+	{ LXPR_PID_MEM,		"mem" },
+	{ LXPR_PID_MOUNTINFO,	"mountinfo" },
+	{ LXPR_PID_ROOTDIR,	"root" },
+	{ LXPR_PID_STAT,	"stat" },
+	{ LXPR_PID_STATM,	"statm" },
+	{ LXPR_PID_STATUS,	"status" },
+	{ LXPR_PID_TASKDIR,	"task" },
+	{ LXPR_PID_FDDIR,	"fd" }
+};
+
+#define	PIDDIRFILES	(sizeof (piddir) / sizeof (piddir[0]))
+
+/*
+ * Contents of an lx /proc/<pid>/task/<tid> directory.
+ */
+static lxpr_dirent_t tiddir[] = {
+	{ LXPR_PID_CMDLINE,	"cmdline" },
+	{ LXPR_PID_CPU,		"cpu" },
+	{ LXPR_PID_CURDIR,	"cwd" },
+	{ LXPR_PID_ENV,		"environ" },
+	{ LXPR_PID_EXE,		"exe" },
+	{ LXPR_PID_LIMITS,	"limits" },
+	{ LXPR_PID_MAPS,	"maps" },
+	{ LXPR_PID_MEM,		"mem" },
+	{ LXPR_PID_MOUNTINFO,	"mountinfo" },
+	{ LXPR_PID_ROOTDIR,	"root" },
+	{ LXPR_PID_TID_STAT,	"stat" },
+	{ LXPR_PID_STATM,	"statm" },
+	{ LXPR_PID_TID_STATUS,	"status" },
+	{ LXPR_PID_FDDIR,	"fd" }
+};
+
+#define	TIDDIRFILES	(sizeof (tiddir) / sizeof (tiddir[0]))
+
+#define	LX_RLIM_INFINITY	0xFFFFFFFFFFFFFFFF
+
+#define	RCTL_INFINITE(x) \
+	((x->rcv_flagaction & RCTL_LOCAL_MAXIMAL) && \
+	(x->rcv_flagaction & RCTL_GLOBAL_INFINITE))
+
+typedef struct lxpr_rlimtab {
+	char	*rlim_name;	/* limit name */
+	char	*rlim_unit;	/* limit unit */
+	char	*rlim_rctl;	/* rctl source */
+} lxpr_rlimtab_t;
+
+static lxpr_rlimtab_t lxpr_rlimtab[] = {
+	{ "Max cpu time",	"seconds",	"process.max-cpu-time" },
+	{ "Max file size",	"bytes",	"process.max-file-size" },
+	{ "Max data size",	"bytes",	"process.max-data-size" },
+	{ "Max stack size",	"bytes",	"process.max-stack-size" },
+	{ "Max core file size",	"bytes",	"process.max-core-size" },
+	{ "Max resident set",	"bytes",	"zone.max-physical-memory" },
+	{ "Max processes",	"processes",	"zone.max-lwps" },
+	{ "Max open files",	"files",	"process.max-file-descriptor" },
+	{ "Max locked memory",	"bytes",	"zone.max-locked-memory" },
+	{ "Max address space",	"bytes",	"process.max-address-space" },
+	{ "Max file locks",	"locks",	NULL },
+	{ "Max pending signals",	"signals",
+		"process.max-sigqueue-size" },
+	{ "Max msgqueue size",	"bytes",	"process.max-msg-messages" },
+	{ NULL, NULL, NULL }
+};
+
+
+/*
+ * contents of lx /proc/net directory
+ */
+static lxpr_dirent_t netdir[] = {
+	{ LXPR_NET_ARP,		"arp" },
+	{ LXPR_NET_DEV,		"dev" },
+	{ LXPR_NET_DEV_MCAST,	"dev_mcast" },
+	{ LXPR_NET_IF_INET6,	"if_inet6" },
+	{ LXPR_NET_IGMP,	"igmp" },
+	{ LXPR_NET_IP_MR_CACHE,	"ip_mr_cache" },
+	{ LXPR_NET_IP_MR_VIF,	"ip_mr_vif" },
+	{ LXPR_NET_IPV6_ROUTE,	"ipv6_route" },
+	{ LXPR_NET_MCFILTER,	"mcfilter" },
+	{ LXPR_NET_NETSTAT,	"netstat" },
+	{ LXPR_NET_RAW,		"raw" },
+	{ LXPR_NET_ROUTE,	"route" },
+	{ LXPR_NET_RPC,		"rpc" },
+	{ LXPR_NET_RT_CACHE,	"rt_cache" },
+	{ LXPR_NET_SOCKSTAT,	"sockstat" },
+	{ LXPR_NET_SNMP,	"snmp" },
+	{ LXPR_NET_STAT,	"stat" },
+	{ LXPR_NET_TCP,		"tcp" },
+	{ LXPR_NET_TCP6,	"tcp6" },
+	{ LXPR_NET_UDP,		"udp" },
+	{ LXPR_NET_UDP6,	"udp6" },
+	{ LXPR_NET_UNIX,	"unix" }
+};
+
+#define	NETDIRFILES	(sizeof (netdir) / sizeof (netdir[0]))
+
+/*
+ * contents of /proc/sys directory
+ */
+static lxpr_dirent_t sysdir[] = {
+	{ LXPR_SYS_FSDIR,	"fs" },
+	{ LXPR_SYS_KERNELDIR,	"kernel" },
+};
+
+#define	SYSDIRFILES	(sizeof (sysdir) / sizeof (sysdir[0]))
+
+/*
+ * contents of /proc/sys/fs directory
+ */
+static lxpr_dirent_t sys_fsdir[] = {
+	{ LXPR_SYS_FS_INOTIFYDIR,	"inotify" },
+};
+
+#define	SYS_FSDIRFILES (sizeof (sys_fsdir) / sizeof (sys_fsdir[0]))
+
+/*
+ * contents of /proc/sys/fs/inotify directory
+ */
+static lxpr_dirent_t sys_fs_inotifydir[] = {
+	{ LXPR_SYS_FS_INOTIFY_MAX_QUEUED_EVENTS,	"max_queued_events" },
+	{ LXPR_SYS_FS_INOTIFY_MAX_USER_INSTANCES,	"max_user_instances" },
+	{ LXPR_SYS_FS_INOTIFY_MAX_USER_WATCHES,		"max_user_watches" },
+};
+
+#define	SYS_FS_INOTIFYDIRFILES \
+	(sizeof (sys_fs_inotifydir) / sizeof (sys_fs_inotifydir[0]))
+
+/*
+ * contents of /proc/sys/kernel directory
+ */
+static lxpr_dirent_t sys_kerneldir[] = {
+	{ LXPR_SYS_KERNEL_HOSTNAME,	"hostname" },
+	{ LXPR_SYS_KERNEL_MSGMNI,	"msgmni" },
+	{ LXPR_SYS_KERNEL_NGROUPS_MAX,	"ngroups_max" },
+	{ LXPR_SYS_KERNEL_PID_MAX,	"pid_max" },
+	{ LXPR_SYS_KERNEL_SHMMAX,	"shmmax" },
+	{ LXPR_SYS_KERNEL_THREADS_MAX,	"threads-max" },
+};
+
+#define	SYS_KERNELDIRFILES (sizeof (sys_kerneldir) / sizeof (sys_kerneldir[0]))
+
+/*
+ * lxpr_open(): Vnode operation for VOP_OPEN()
+ */
+static int
+lxpr_open(vnode_t **vpp, int flag, cred_t *cr, caller_context_t *ct)
+{
+	vnode_t		*vp = *vpp;
+	lxpr_node_t	*lxpnp = VTOLXP(vp);
+	lxpr_nodetype_t	type = lxpnp->lxpr_type;
+	vnode_t		*rvp;
+	int		error = 0;
+
+	/*
+	 * We only allow reading in this file systrem
+	 */
+	if (flag & FWRITE)
+		return (EROFS);
+
+	/*
+	 * If we are opening an underlying file only allow regular files,
+	 * fifos or sockets; reject the open for anything else.
+	 * Just do it if we are opening the current or root directory.
+	 */
+	if (lxpnp->lxpr_realvp != NULL) {
+		rvp = lxpnp->lxpr_realvp;
+
+		if (type == LXPR_PID_FD_FD && rvp->v_type != VREG &&
+		    rvp->v_type != VFIFO && rvp->v_type != VSOCK) {
+			error = EACCES;
+		} else {
+			if (type == LXPR_PID_FD_FD && rvp->v_type == VFIFO) {
+				/*
+				 * This flag lets the fifo open know that
+				 * we're using proc/fd to open a fd which we
+				 * already have open. Otherwise, the fifo might
+				 * reject an open if the other end has closed.
+				 */
+				flag |= FKLYR;
+			}
+			/*
+			 * Need to hold rvp since VOP_OPEN() may release it.
+			 */
+			VN_HOLD(rvp);
+			error = VOP_OPEN(&rvp, flag, cr, ct);
+			if (error) {
+				VN_RELE(rvp);
+			} else {
+				*vpp = rvp;
+				VN_RELE(vp);
+			}
+		}
+	}
+
+	return (error);
+}
+
+
+/*
+ * lxpr_close(): Vnode operation for VOP_CLOSE()
+ */
+/* ARGSUSED */
+static int
+lxpr_close(vnode_t *vp, int flag, int count, offset_t offset, cred_t *cr,
+    caller_context_t *ct)
+{
+	lxpr_node_t	*lxpr = VTOLXP(vp);
+	lxpr_nodetype_t	type = lxpr->lxpr_type;
+
+	/*
+	 * we should never get here because the close is done on the realvp
+	 * for these nodes
+	 */
+	ASSERT(type != LXPR_PID_FD_FD &&
+	    type != LXPR_PID_CURDIR &&
+	    type != LXPR_PID_ROOTDIR &&
+	    type != LXPR_PID_EXE);
+
+	return (0);
+}
+
+static void (*lxpr_read_function[LXPR_NFILES])() = {
+	lxpr_read_isdir,		/* /proc		*/
+	lxpr_read_isdir,		/* /proc/<pid>		*/
+	lxpr_read_pid_cmdline,		/* /proc/<pid>/cmdline	*/
+	lxpr_read_empty,		/* /proc/<pid>/cpu	*/
+	lxpr_read_invalid,		/* /proc/<pid>/cwd	*/
+	lxpr_read_pid_env,		/* /proc/<pid>/environ	*/
+	lxpr_read_invalid,		/* /proc/<pid>/exe	*/
+	lxpr_read_pid_limits,		/* /proc/<pid>/limits	*/
+	lxpr_read_pid_maps,		/* /proc/<pid>/maps	*/
+	lxpr_read_empty,		/* /proc/<pid>/mem	*/
+	lxpr_read_pid_mountinfo,	/* /proc/<pid>/mountinfo */
+	lxpr_read_invalid,		/* /proc/<pid>/root	*/
+	lxpr_read_pid_stat,		/* /proc/<pid>/stat	*/
+	lxpr_read_pid_statm,		/* /proc/<pid>/statm	*/
+	lxpr_read_pid_status,		/* /proc/<pid>/status	*/
+	lxpr_read_isdir,		/* /proc/<pid>/task	*/
+	lxpr_read_isdir,		/* /proc/<pid>/task/nn	*/
+	lxpr_read_isdir,		/* /proc/<pid>/fd	*/
+	lxpr_read_fd,			/* /proc/<pid>/fd/nn	*/
+	lxpr_read_pid_cmdline,		/* /proc/<pid>/task/<tid>/cmdline */
+	lxpr_read_empty,		/* /proc/<pid>/task/<tid>/cpu	*/
+	lxpr_read_invalid,		/* /proc/<pid>/task/<tid>/cwd	*/
+	lxpr_read_pid_env,		/* /proc/<pid>/task/<tid>/environ */
+	lxpr_read_invalid,		/* /proc/<pid>/task/<tid>/exe	*/
+	lxpr_read_pid_limits,		/* /proc/<pid>/task/<tid>/limits */
+	lxpr_read_pid_maps,		/* /proc/<pid>/task/<tid>/maps	*/
+	lxpr_read_empty,		/* /proc/<pid>/task/<tid>/mem	*/
+	lxpr_read_pid_mountinfo,	/* /proc/<pid>/task/<tid>/mountinfo */
+	lxpr_read_invalid,		/* /proc/<pid>/task/<tid>/root	*/
+	lxpr_read_pid_tid_stat,		/* /proc/<pid>/task/<tid>/stat	*/
+	lxpr_read_pid_statm,		/* /proc/<pid>/task/<tid>/statm	*/
+	lxpr_read_pid_tid_status,	/* /proc/<pid>/task/<tid>/status */
+	lxpr_read_isdir,		/* /proc/<pid>/task/<tid>/fd	*/
+	lxpr_read_fd,			/* /proc/<pid>/task/<tid>/fd/nn	*/
+	lxpr_read_empty,		/* /proc/cmdline	*/
+	lxpr_read_cpuinfo,		/* /proc/cpuinfo	*/
+	lxpr_read_empty,		/* /proc/devices	*/
+	lxpr_read_diskstats,		/* /proc/diskstats	*/
+	lxpr_read_empty,		/* /proc/dma		*/
+	lxpr_read_empty,		/* /proc/filesystems	*/
+	lxpr_read_empty,		/* /proc/interrupts	*/
+	lxpr_read_empty,		/* /proc/ioports	*/
+	lxpr_read_empty,		/* /proc/kcore		*/
+	lxpr_read_invalid,		/* /proc/kmsg -- see lxpr_read() */
+	lxpr_read_loadavg,		/* /proc/loadavg	*/
+	lxpr_read_meminfo,		/* /proc/meminfo	*/
+	lxpr_read_empty,		/* /proc/modules	*/
+	lxpr_read_mounts,		/* /proc/mounts		*/
+	lxpr_read_isdir,		/* /proc/net		*/
+	lxpr_read_net_arp,		/* /proc/net/arp	*/
+	lxpr_read_net_dev,		/* /proc/net/dev	*/
+	lxpr_read_net_dev_mcast,	/* /proc/net/dev_mcast	*/
+	lxpr_read_net_if_inet6,		/* /proc/net/if_inet6	*/
+	lxpr_read_net_igmp,		/* /proc/net/igmp	*/
+	lxpr_read_net_ip_mr_cache,	/* /proc/net/ip_mr_cache */
+	lxpr_read_net_ip_mr_vif,	/* /proc/net/ip_mr_vif	*/
+	lxpr_read_net_ipv6_route,	/* /proc/net/ipv6_route	*/
+	lxpr_read_net_mcfilter,		/* /proc/net/mcfilter	*/
+	lxpr_read_net_netstat,		/* /proc/net/netstat	*/
+	lxpr_read_net_raw,		/* /proc/net/raw	*/
+	lxpr_read_net_route,		/* /proc/net/route	*/
+	lxpr_read_net_rpc,		/* /proc/net/rpc	*/
+	lxpr_read_net_rt_cache,		/* /proc/net/rt_cache	*/
+	lxpr_read_net_sockstat,		/* /proc/net/sockstat	*/
+	lxpr_read_net_snmp,		/* /proc/net/snmp	*/
+	lxpr_read_net_stat,		/* /proc/net/stat	*/
+	lxpr_read_net_tcp,		/* /proc/net/tcp	*/
+	lxpr_read_net_tcp6,		/* /proc/net/tcp6	*/
+	lxpr_read_net_udp,		/* /proc/net/udp	*/
+	lxpr_read_net_udp6,		/* /proc/net/udp6	*/
+	lxpr_read_net_unix,		/* /proc/net/unix	*/
+	lxpr_read_partitions,		/* /proc/partitions	*/
+	lxpr_read_invalid,		/* /proc/self		*/
+	lxpr_read_stat,			/* /proc/stat		*/
+	lxpr_read_swaps,		/* /proc/swaps		*/
+	lxpr_read_invalid,		/* /proc/sys		*/
+	lxpr_read_invalid,		/* /proc/sys/fs		*/
+	lxpr_read_invalid,		/* /proc/sys/fs/inotify	*/
+	lxpr_read_sys_fs_inotify_max_queued_events, /* max_queued_events */
+	lxpr_read_sys_fs_inotify_max_user_instances, /* max_user_instances */
+	lxpr_read_sys_fs_inotify_max_user_watches, /* max_user_watches */
+	lxpr_read_invalid,		/* /proc/sys/kernel	*/
+	lxpr_read_sys_kernel_hostname,	/* /proc/sys/kernel/hostname */
+	lxpr_read_sys_kernel_msgmni,	/* /proc/sys/kernel/msgmni */
+	lxpr_read_sys_kernel_ngroups_max, /* /proc/sys/kernel/ngroups_max */
+	lxpr_read_sys_kernel_pid_max,	/* /proc/sys/kernel/pid_max */
+	lxpr_read_sys_kernel_shmmax,	/* /proc/sys/kernel/shmmax */
+	lxpr_read_sys_kernel_threads_max, /* /proc/sys/kernel/threads-max */
+	lxpr_read_uptime,		/* /proc/uptime		*/
+	lxpr_read_version,		/* /proc/version	*/
+};
+
+/*
+ * Array of lookup functions, indexed by lx /proc file type.
+ */
+static vnode_t *(*lxpr_lookup_function[LXPR_NFILES])() = {
+	lxpr_lookup_procdir,		/* /proc		*/
+	lxpr_lookup_piddir,		/* /proc/<pid>		*/
+	lxpr_lookup_not_a_dir,		/* /proc/<pid>/cmdline	*/
+	lxpr_lookup_not_a_dir,		/* /proc/<pid>/cpu	*/
+	lxpr_lookup_not_a_dir,		/* /proc/<pid>/cwd	*/
+	lxpr_lookup_not_a_dir,		/* /proc/<pid>/environ	*/
+	lxpr_lookup_not_a_dir,		/* /proc/<pid>/exe	*/
+	lxpr_lookup_not_a_dir,		/* /proc/<pid>/limits	*/
+	lxpr_lookup_not_a_dir,		/* /proc/<pid>/maps	*/
+	lxpr_lookup_not_a_dir,		/* /proc/<pid>/mem	*/
+	lxpr_lookup_not_a_dir,		/* /proc/<pid>/mountinfo */
+	lxpr_lookup_not_a_dir,		/* /proc/<pid>/root	*/
+	lxpr_lookup_not_a_dir,		/* /proc/<pid>/stat	*/
+	lxpr_lookup_not_a_dir,		/* /proc/<pid>/statm	*/
+	lxpr_lookup_not_a_dir,		/* /proc/<pid>/status	*/
+	lxpr_lookup_taskdir,		/* /proc/<pid>/task	*/
+	lxpr_lookup_task_tid_dir,	/* /proc/<pid>/task/nn	*/
+	lxpr_lookup_fddir,		/* /proc/<pid>/fd	*/
+	lxpr_lookup_not_a_dir,		/* /proc/<pid>/fd/nn	*/
+	lxpr_lookup_not_a_dir,		/* /proc/<pid>/task/<tid>/cmdline */
+	lxpr_lookup_not_a_dir,		/* /proc/<pid>/task/<tid>/cpu	*/
+	lxpr_lookup_not_a_dir,		/* /proc/<pid>/task/<tid>/cwd	*/
+	lxpr_lookup_not_a_dir,		/* /proc/<pid>/task/<tid>/environ */
+	lxpr_lookup_not_a_dir,		/* /proc/<pid>/task/<tid>/exe	*/
+	lxpr_lookup_not_a_dir,		/* /proc/<pid>/task/<tid>/limits */
+	lxpr_lookup_not_a_dir,		/* /proc/<pid>/task/<tid>/maps	*/
+	lxpr_lookup_not_a_dir,		/* /proc/<pid>/task/<tid>/mem	*/
+	lxpr_lookup_not_a_dir,		/* /proc/<pid>/task/<tid>/mountinfo */
+	lxpr_lookup_not_a_dir,		/* /proc/<pid>/task/<tid>/root	*/
+	lxpr_lookup_not_a_dir,		/* /proc/<pid>/task/<tid>/stat	*/
+	lxpr_lookup_not_a_dir,		/* /proc/<pid>/task/<tid>/statm	*/
+	lxpr_lookup_not_a_dir,		/* /proc/<pid>/task/<tid>/status */
+	lxpr_lookup_fddir,		/* /proc/<pid>/task/<tid>/fd	*/
+	lxpr_lookup_not_a_dir,		/* /proc/<pid>/task/<tid>/fd/nn	*/
+	lxpr_lookup_not_a_dir,		/* /proc/cmdline	*/
+	lxpr_lookup_not_a_dir,		/* /proc/cpuinfo	*/
+	lxpr_lookup_not_a_dir,		/* /proc/devices	*/
+	lxpr_lookup_not_a_dir,		/* /proc/diskstats	*/
+	lxpr_lookup_not_a_dir,		/* /proc/dma		*/
+	lxpr_lookup_not_a_dir,		/* /proc/filesystems	*/
+	lxpr_lookup_not_a_dir,		/* /proc/interrupts	*/
+	lxpr_lookup_not_a_dir,		/* /proc/ioports	*/
+	lxpr_lookup_not_a_dir,		/* /proc/kcore		*/
+	lxpr_lookup_not_a_dir,		/* /proc/kmsg		*/
+	lxpr_lookup_not_a_dir,		/* /proc/loadavg	*/
+	lxpr_lookup_not_a_dir,		/* /proc/meminfo	*/
+	lxpr_lookup_not_a_dir,		/* /proc/modules	*/
+	lxpr_lookup_not_a_dir,		/* /proc/mounts		*/
+	lxpr_lookup_netdir,		/* /proc/net		*/
+	lxpr_lookup_not_a_dir,		/* /proc/net/arp	*/
+	lxpr_lookup_not_a_dir,		/* /proc/net/dev	*/
+	lxpr_lookup_not_a_dir,		/* /proc/net/dev_mcast	*/
+	lxpr_lookup_not_a_dir,		/* /proc/net/if_inet6	*/
+	lxpr_lookup_not_a_dir,		/* /proc/net/igmp	*/
+	lxpr_lookup_not_a_dir,		/* /proc/net/ip_mr_cache */
+	lxpr_lookup_not_a_dir,		/* /proc/net/ip_mr_vif	*/
+	lxpr_lookup_not_a_dir,		/* /proc/net/ipv6_route	*/
+	lxpr_lookup_not_a_dir,		/* /proc/net/mcfilter	*/
+	lxpr_lookup_not_a_dir,		/* /proc/net/netstat	*/
+	lxpr_lookup_not_a_dir,		/* /proc/net/raw	*/
+	lxpr_lookup_not_a_dir,		/* /proc/net/route	*/
+	lxpr_lookup_not_a_dir,		/* /proc/net/rpc	*/
+	lxpr_lookup_not_a_dir,		/* /proc/net/rt_cache	*/
+	lxpr_lookup_not_a_dir,		/* /proc/net/sockstat	*/
+	lxpr_lookup_not_a_dir,		/* /proc/net/snmp	*/
+	lxpr_lookup_not_a_dir,		/* /proc/net/stat	*/
+	lxpr_lookup_not_a_dir,		/* /proc/net/tcp	*/
+	lxpr_lookup_not_a_dir,		/* /proc/net/tcp6	*/
+	lxpr_lookup_not_a_dir,		/* /proc/net/udp	*/
+	lxpr_lookup_not_a_dir,		/* /proc/net/udp6	*/
+	lxpr_lookup_not_a_dir,		/* /proc/net/unix	*/
+	lxpr_lookup_not_a_dir,		/* /proc/partitions	*/
+	lxpr_lookup_not_a_dir,		/* /proc/self		*/
+	lxpr_lookup_not_a_dir,		/* /proc/stat		*/
+	lxpr_lookup_not_a_dir,		/* /proc/swaps		*/
+	lxpr_lookup_sysdir,		/* /proc/sys		*/
+	lxpr_lookup_sys_fsdir,		/* /proc/sys/fs		*/
+	lxpr_lookup_sys_fs_inotifydir,	/* /proc/sys/fs/inotify	*/
+	lxpr_lookup_not_a_dir,		/* .../inotify/max_queued_events */
+	lxpr_lookup_not_a_dir,		/* .../inotify/max_user_instances */
+	lxpr_lookup_not_a_dir,		/* .../inotify/max_user_watches */
+	lxpr_lookup_sys_kerneldir,	/* /proc/sys/kernel	*/
+	lxpr_lookup_not_a_dir,		/* /proc/sys/kernel/hostname */
+	lxpr_lookup_not_a_dir,		/* /proc/sys/kernel/msgmni */
+	lxpr_lookup_not_a_dir,		/* /proc/sys/kernel/ngroups_max */
+	lxpr_lookup_not_a_dir,		/* /proc/sys/kernel/pid_max */
+	lxpr_lookup_not_a_dir,		/* /proc/sys/kernel/shmmax */
+	lxpr_lookup_not_a_dir,		/* /proc/sys/kernel/threads-max */
+	lxpr_lookup_not_a_dir,		/* /proc/uptime		*/
+	lxpr_lookup_not_a_dir,		/* /proc/version	*/
+};
+
+/*
+ * Array of readdir functions, indexed by /proc file type.
+ */
+static int (*lxpr_readdir_function[LXPR_NFILES])() = {
+	lxpr_readdir_procdir,		/* /proc		*/
+	lxpr_readdir_piddir,		/* /proc/<pid>		*/
+	lxpr_readdir_not_a_dir,		/* /proc/<pid>/cmdline	*/
+	lxpr_readdir_not_a_dir,		/* /proc/<pid>/cpu	*/
+	lxpr_readdir_not_a_dir,		/* /proc/<pid>/cwd	*/
+	lxpr_readdir_not_a_dir,		/* /proc/<pid>/environ	*/
+	lxpr_readdir_not_a_dir,		/* /proc/<pid>/exe	*/
+	lxpr_readdir_not_a_dir,		/* /proc/<pid>/limits	*/
+	lxpr_readdir_not_a_dir,		/* /proc/<pid>/maps	*/
+	lxpr_readdir_not_a_dir,		/* /proc/<pid>/mem	*/
+	lxpr_readdir_not_a_dir,		/* /proc/<pid>/mountinfo */
+	lxpr_readdir_not_a_dir,		/* /proc/<pid>/root	*/
+	lxpr_readdir_not_a_dir,		/* /proc/<pid>/stat	*/
+	lxpr_readdir_not_a_dir,		/* /proc/<pid>/statm	*/
+	lxpr_readdir_not_a_dir,		/* /proc/<pid>/status	*/
+	lxpr_readdir_taskdir,		/* /proc/<pid>/task	*/
+	lxpr_readdir_task_tid_dir,	/* /proc/<pid>/task/nn	*/
+	lxpr_readdir_fddir,		/* /proc/<pid>/fd	*/
+	lxpr_readdir_not_a_dir,		/* /proc/<pid>/fd/nn	*/
+	lxpr_readdir_not_a_dir,		/* /proc/<pid>/task/<tid>/cmdline */
+	lxpr_readdir_not_a_dir,		/* /proc/<pid>/task/<tid>/cpu	*/
+	lxpr_readdir_not_a_dir,		/* /proc/<pid>/task/<tid>/cwd	*/
+	lxpr_readdir_not_a_dir,		/* /proc/<pid>/task/<tid>/environ */
+	lxpr_readdir_not_a_dir,		/* /proc/<pid>/task/<tid>/exe	*/
+	lxpr_readdir_not_a_dir,		/* /proc/<pid>/task/<tid>/limits */
+	lxpr_readdir_not_a_dir,		/* /proc/<pid>/task/<tid>/maps	*/
+	lxpr_readdir_not_a_dir,		/* /proc/<pid>/task/<tid>/mem	*/
+	lxpr_readdir_not_a_dir,		/* /proc/<pid>/task/<tid>/mountinfo */
+	lxpr_readdir_not_a_dir,		/* /proc/<pid>/task/<tid>/root	*/
+	lxpr_readdir_not_a_dir,		/* /proc/<pid>/task/<tid>/stat	*/
+	lxpr_readdir_not_a_dir,		/* /proc/<pid>/task/<tid>/statm	*/
+	lxpr_readdir_not_a_dir,		/* /proc/<pid>/task/<tid>/status */
+	lxpr_readdir_fddir,		/* /proc/<pid>/task/<tid>/fd	*/
+	lxpr_readdir_not_a_dir,		/* /proc/<pid>/task/<tid>/fd/nn	*/
+	lxpr_readdir_not_a_dir,		/* /proc/cmdline	*/
+	lxpr_readdir_not_a_dir,		/* /proc/cpuinfo	*/
+	lxpr_readdir_not_a_dir,		/* /proc/devices	*/
+	lxpr_readdir_not_a_dir,		/* /proc/diskstats	*/
+	lxpr_readdir_not_a_dir,		/* /proc/dma		*/
+	lxpr_readdir_not_a_dir,		/* /proc/filesystems	*/
+	lxpr_readdir_not_a_dir,		/* /proc/interrupts	*/
+	lxpr_readdir_not_a_dir,		/* /proc/ioports	*/
+	lxpr_readdir_not_a_dir,		/* /proc/kcore		*/
+	lxpr_readdir_not_a_dir,		/* /proc/kmsg		*/
+	lxpr_readdir_not_a_dir,		/* /proc/loadavg	*/
+	lxpr_readdir_not_a_dir,		/* /proc/meminfo	*/
+	lxpr_readdir_not_a_dir,		/* /proc/modules	*/
+	lxpr_readdir_not_a_dir,		/* /proc/mounts		*/
+	lxpr_readdir_netdir,		/* /proc/net		*/
+	lxpr_readdir_not_a_dir,		/* /proc/net/arp	*/
+	lxpr_readdir_not_a_dir,		/* /proc/net/dev	*/
+	lxpr_readdir_not_a_dir,		/* /proc/net/dev_mcast	*/
+	lxpr_readdir_not_a_dir,		/* /proc/net/if_inet6	*/
+	lxpr_readdir_not_a_dir,		/* /proc/net/igmp	*/
+	lxpr_readdir_not_a_dir,		/* /proc/net/ip_mr_cache */
+	lxpr_readdir_not_a_dir,		/* /proc/net/ip_mr_vif	*/
+	lxpr_readdir_not_a_dir,		/* /proc/net/ipv6_route	*/
+	lxpr_readdir_not_a_dir,		/* /proc/net/mcfilter	*/
+	lxpr_readdir_not_a_dir,		/* /proc/net/netstat	*/
+	lxpr_readdir_not_a_dir,		/* /proc/net/raw	*/
+	lxpr_readdir_not_a_dir,		/* /proc/net/route	*/
+	lxpr_readdir_not_a_dir,		/* /proc/net/rpc	*/
+	lxpr_readdir_not_a_dir,		/* /proc/net/rt_cache	*/
+	lxpr_readdir_not_a_dir,		/* /proc/net/sockstat	*/
+	lxpr_readdir_not_a_dir,		/* /proc/net/snmp	*/
+	lxpr_readdir_not_a_dir,		/* /proc/net/stat	*/
+	lxpr_readdir_not_a_dir,		/* /proc/net/tcp	*/
+	lxpr_readdir_not_a_dir,		/* /proc/net/tcp6	*/
+	lxpr_readdir_not_a_dir,		/* /proc/net/udp	*/
+	lxpr_readdir_not_a_dir,		/* /proc/net/udp6	*/
+	lxpr_readdir_not_a_dir,		/* /proc/net/unix	*/
+	lxpr_readdir_not_a_dir,		/* /proc/partitions	*/
+	lxpr_readdir_not_a_dir,		/* /proc/self		*/
+	lxpr_readdir_not_a_dir,		/* /proc/stat		*/
+	lxpr_readdir_not_a_dir,		/* /proc/swaps		*/
+	lxpr_readdir_sysdir,		/* /proc/sys		*/
+	lxpr_readdir_sys_fsdir,		/* /proc/sys/fs		*/
+	lxpr_readdir_sys_fs_inotifydir,	/* /proc/sys/fs/inotify	*/
+	lxpr_readdir_not_a_dir,		/* .../inotify/max_queued_events */
+	lxpr_readdir_not_a_dir,		/* .../inotify/max_user_instances */
+	lxpr_readdir_not_a_dir,		/* .../inotify/max_user_watches	*/
+	lxpr_readdir_sys_kerneldir,	/* /proc/sys/kernel	*/
+	lxpr_readdir_not_a_dir,		/* /proc/sys/kernel/hostname */
+	lxpr_readdir_not_a_dir,		/* /proc/sys/kernel/msgmni */
+	lxpr_readdir_not_a_dir,		/* /proc/sys/kernel/ngroups_max */
+	lxpr_readdir_not_a_dir,		/* /proc/sys/kernel/pid_max */
+	lxpr_readdir_not_a_dir,		/* /proc/sys/kernel/shmmax */
+	lxpr_readdir_not_a_dir,		/* /proc/sys/kernel/threads-max */
+	lxpr_readdir_not_a_dir,		/* /proc/uptime		*/
+	lxpr_readdir_not_a_dir,		/* /proc/version	*/
+};
+
+
+/*
+ * lxpr_read(): Vnode operation for VOP_READ()
+ *
+ * As the format of all the files that can be read in the lx procfs is human
+ * readable and not binary structures there do not have to be different
+ * read variants depending on whether the reading process model is 32 or 64 bits
+ * (at least in general, and certainly the difference is unlikely to be enough
+ * to justify have different routines for 32 and 64 bit reads
+ */
+/* ARGSUSED */
+static int
+lxpr_read(vnode_t *vp, uio_t *uiop, int ioflag, cred_t *cr,
+    caller_context_t *ct)
+{
+	lxpr_node_t *lxpnp = VTOLXP(vp);
+	lxpr_nodetype_t type = lxpnp->lxpr_type;
+	lxpr_uiobuf_t *uiobuf = lxpr_uiobuf_new(uiop);
+	int error;
+
+	ASSERT(type < LXPR_NFILES);
+
+	if (type == LXPR_KMSG) {
+		ldi_ident_t	li = VTOLXPM(vp)->lxprm_li;
+		ldi_handle_t	ldih;
+		struct strioctl	str;
+		int		rv;
+
+		/*
+		 * Open the zone's console device using the layered driver
+		 * interface.
+		 */
+		if ((error =
+		    ldi_open_by_name("/dev/log", FREAD, cr, &ldih, li)) != 0)
+			return (error);
+
+		/*
+		 * Send an ioctl to the underlying console device, letting it
+		 * know we're interested in getting console messages.
+		 */
+		str.ic_cmd = I_CONSLOG;
+		str.ic_timout = 0;
+		str.ic_len = 0;
+		str.ic_dp = NULL;
+		if ((error = ldi_ioctl(ldih, I_STR,
+		    (intptr_t)&str, FKIOCTL, cr, &rv)) != 0)
+			return (error);
+
+		lxpr_read_kmsg(lxpnp, uiobuf, ldih);
+
+		if ((error = ldi_close(ldih, FREAD, cr)) != 0)
+			return (error);
+	} else {
+		lxpr_read_function[type](lxpnp, uiobuf);
+	}
+
+	error = lxpr_uiobuf_flush(uiobuf);
+	lxpr_uiobuf_free(uiobuf);
+
+	return (error);
+}
+
+/*
+ * lxpr_read_invalid(), lxpr_read_isdir(), lxpr_read_empty()
+ *
+ * Various special case reads:
+ * - trying to read a directory
+ * - invalid file (used to mean a file that should be implemented,
+ *   but isn't yet)
+ * - empty file
+ * - wait to be able to read a file that will never have anything to read
+ */
+/* ARGSUSED */
+static void
+lxpr_read_isdir(lxpr_node_t *lxpnp, lxpr_uiobuf_t *uiobuf)
+{
+	lxpr_uiobuf_seterr(uiobuf, EISDIR);
+}
+
+/* ARGSUSED */
+static void
+lxpr_read_invalid(lxpr_node_t *lxpnp, lxpr_uiobuf_t *uiobuf)
+{
+	lxpr_uiobuf_seterr(uiobuf, EINVAL);
+}
+
+/* ARGSUSED */
+static void
+lxpr_read_empty(lxpr_node_t *lxpnp, lxpr_uiobuf_t *uiobuf)
+{
+}
+
+/*
+ * lxpr_read_pid_cmdline(): read argument vector from process
+ */
+static void
+lxpr_read_pid_cmdline(lxpr_node_t *lxpnp, lxpr_uiobuf_t *uiobuf)
+{
+	proc_t *p;
+	char *buf;
+	size_t asz = lxpr_maxargvlen, sz;
+
+	ASSERT(lxpnp->lxpr_type == LXPR_PID_CMDLINE ||
+	    lxpnp->lxpr_type == LXPR_PID_TID_CMDLINE);
+
+	buf = kmem_alloc(asz, KM_SLEEP);
+
+	p = lxpr_lock(lxpnp->lxpr_pid);
+	if (p == NULL) {
+		lxpr_uiobuf_seterr(uiobuf, EINVAL);
+		kmem_free(buf, asz);
+		return;
+	}
+
+	if (prreadargv(p, buf, asz, &sz) != 0) {
+		lxpr_uiobuf_seterr(uiobuf, EINVAL);
+	} else {
+		lxpr_uiobuf_write(uiobuf, buf, sz);
+	}
+
+	lxpr_unlock(p);
+	kmem_free(buf, asz);
+}
+
+/*
+ * lxpr_read_pid_env(): read env vector from process
+ */
+static void
+lxpr_read_pid_env(lxpr_node_t *lxpnp, lxpr_uiobuf_t *uiobuf)
+{
+	proc_t *p;
+	char *buf;
+	size_t asz = lxpr_maxenvvlen, sz;
+
+	ASSERT(lxpnp->lxpr_type == LXPR_PID_ENV);
+
+	buf = kmem_alloc(asz, KM_SLEEP);
+
+	p = lxpr_lock(lxpnp->lxpr_pid);
+	if (p == NULL) {
+		lxpr_uiobuf_seterr(uiobuf, EINVAL);
+		kmem_free(buf, asz);
+		return;
+	}
+
+	if (prreadenvv(p, buf, asz, &sz) != 0) {
+		lxpr_uiobuf_seterr(uiobuf, EINVAL);
+	} else {
+		lxpr_uiobuf_write(uiobuf, buf, sz);
+	}
+
+	lxpr_unlock(p);
+	kmem_free(buf, asz);
+}
+
+/*
+ * lxpr_read_pid_limits(): ulimit file
+ */
+static void
+lxpr_read_pid_limits(lxpr_node_t *lxpnp, lxpr_uiobuf_t *uiobuf)
+{
+	proc_t *p;
+	rctl_qty_t cur, max;
+	rctl_val_t *oval, *nval;
+	rctl_hndl_t hndl;
+	char *kname;
+	int i;
+
+	ASSERT(lxpnp->lxpr_type == LXPR_PID_LIMITS ||
+	    lxpnp->lxpr_type == LXPR_PID_TID_LIMITS);
+
+	nval = kmem_alloc(sizeof (rctl_val_t), KM_SLEEP);
+
+	p = lxpr_lock(lxpnp->lxpr_pid);
+	if (p == NULL) {
+		kmem_free(nval, sizeof (rctl_val_t));
+		lxpr_uiobuf_seterr(uiobuf, EINVAL);
+		return;
+	}
+
+	lxpr_uiobuf_printf(uiobuf, "%-25s %-20s %-20s %-10s\n",
+	    "Limit", "Soft Limit", "Hard Limit", "Units");
+	for (i = 0; lxpr_rlimtab[i].rlim_name != NULL; i++) {
+		kname = lxpr_rlimtab[i].rlim_rctl;
+		/* default to unlimited for resources without an analog */
+		cur = RLIM_INFINITY;
+		max = RLIM_INFINITY;
+		if (kname != NULL) {
+			hndl = rctl_hndl_lookup(kname);
+			oval = NULL;
+			while ((hndl != -1) &&
+			    rctl_local_get(hndl, oval, nval, p) == 0) {
+				oval = nval;
+				switch (nval->rcv_privilege) {
+				case RCPRIV_BASIC:
+					if (!RCTL_INFINITE(nval))
+						cur = nval->rcv_value;
+					break;
+				case RCPRIV_PRIVILEGED:
+					if (!RCTL_INFINITE(nval))
+						max = nval->rcv_value;
+					break;
+				}
+			}
+		}
+
+		lxpr_uiobuf_printf(uiobuf, "%-25s", lxpr_rlimtab[i].rlim_name);
+		if (cur == RLIM_INFINITY || cur == LX_RLIM_INFINITY) {
+			lxpr_uiobuf_printf(uiobuf, " %-20s", "unlimited");
+		} else {
+			lxpr_uiobuf_printf(uiobuf, " %-20lu", cur);
+		}
+		if (max == RLIM_INFINITY || max == LX_RLIM_INFINITY) {
+			lxpr_uiobuf_printf(uiobuf, " %-20s", "unlimited");
+		} else {
+			lxpr_uiobuf_printf(uiobuf, " %-20lu", max);
+		}
+		lxpr_uiobuf_printf(uiobuf, " %-10s\n",
+		    lxpr_rlimtab[i].rlim_unit);
+	}
+
+	lxpr_unlock(p);
+	kmem_free(nval, sizeof (rctl_val_t));
+}
+
+/*
+ * lxpr_read_pid_maps(): memory map file
+ */
+static void
+lxpr_read_pid_maps(lxpr_node_t *lxpnp, lxpr_uiobuf_t *uiobuf)
+{
+	proc_t *p;
+	struct as *as;
+	struct seg *seg;
+	char *buf;
+	int buflen = MAXPATHLEN;
+	struct print_data {
+		uintptr_t saddr;
+		uintptr_t eaddr;
+		int type;
+		char prot[5];
+		uintptr_t offset;
+		vnode_t *vp;
+		struct print_data *next;
+	} *print_head = NULL;
+	struct print_data **print_tail = &print_head;
+	struct print_data *pbuf;
+
+	ASSERT(lxpnp->lxpr_type == LXPR_PID_MAPS ||
+	    lxpnp->lxpr_type == LXPR_PID_TID_MAPS);
+
+	p = lxpr_lock(lxpnp->lxpr_pid);
+	if (p == NULL) {
+		lxpr_uiobuf_seterr(uiobuf, EINVAL);
+		return;
+	}
+
+	as = p->p_as;
+
+	if (as == &kas) {
+		lxpr_unlock(p);
+		return;
+	}
+
+	mutex_exit(&p->p_lock);
+
+	/* Iterate over all segments in the address space */
+	AS_LOCK_ENTER(as, &as->a_lock, RW_READER);
+	for (seg = AS_SEGFIRST(as); seg != NULL; seg = AS_SEGNEXT(as, seg)) {
+		vnode_t *vp;
+		uint_t protbits;
+
+		pbuf = kmem_alloc(sizeof (*pbuf), KM_SLEEP);
+
+		pbuf->saddr = (uintptr_t)seg->s_base;
+		pbuf->eaddr = pbuf->saddr + seg->s_size;
+		pbuf->type = SEGOP_GETTYPE(seg, seg->s_base);
+
+		/*
+		 * Cheat and only use the protection bits of the first page
+		 * in the segment
+		 */
+		(void) strncpy(pbuf->prot, "----", sizeof (pbuf->prot));
+		(void) SEGOP_GETPROT(seg, seg->s_base, 0, &protbits);
+
+		if (protbits & PROT_READ)	   pbuf->prot[0] = 'r';
+		if (protbits & PROT_WRITE)	   pbuf->prot[1] = 'w';
+		if (protbits & PROT_EXEC)	   pbuf->prot[2] = 'x';
+		if (pbuf->type & MAP_SHARED)	   pbuf->prot[3] = 's';
+		else if (pbuf->type & MAP_PRIVATE) pbuf->prot[3] = 'p';
+
+		if (seg->s_ops == &segvn_ops &&
+		    SEGOP_GETVP(seg, seg->s_base, &vp) == 0 &&
+		    vp != NULL && vp->v_type == VREG) {
+			VN_HOLD(vp);
+			pbuf->vp = vp;
+		} else {
+			pbuf->vp = NULL;
+		}
+
+		pbuf->offset = SEGOP_GETOFFSET(seg, (caddr_t)pbuf->saddr);
+
+		pbuf->next = NULL;
+		*print_tail = pbuf;
+		print_tail = &pbuf->next;
+	}
+	AS_LOCK_EXIT(as, &as->a_lock);
+	mutex_enter(&p->p_lock);
+	lxpr_unlock(p);
+
+	buf = kmem_alloc(buflen, KM_SLEEP);
+
+	/* print the data we've extracted */
+	pbuf = print_head;
+	while (pbuf != NULL) {
+		struct print_data *pbuf_next;
+		vattr_t vattr;
+
+		int maj = 0;
+		int min = 0;
+		ino_t inode = 0;
+
+		*buf = '\0';
+		if (pbuf->vp != NULL) {
+			vattr.va_mask = AT_FSID | AT_NODEID;
+			if (VOP_GETATTR(pbuf->vp, &vattr, 0, CRED(),
+			    NULL) == 0) {
+				maj = getmajor(vattr.va_fsid);
+				min = getminor(vattr.va_fsid);
+				inode = vattr.va_nodeid;
+			}
+			(void) vnodetopath(NULL, pbuf->vp, buf, buflen, CRED());
+			VN_RELE(pbuf->vp);
+		}
+
+		if (p->p_model == DATAMODEL_LP64) {
+			lxpr_uiobuf_printf(uiobuf,
+			    "%08llx-%08llx %s %08llx %02x:%02x %llu%s%s\n",
+			    pbuf->saddr, pbuf->eaddr, pbuf->prot, pbuf->offset,
+			    maj, min, inode, *buf != '\0' ? " " : "", buf);
+		} else {
+			lxpr_uiobuf_printf(uiobuf,
+			    "%08x-%08x %s %08x %02x:%02x %llu%s%s\n",
+			    (uint32_t)pbuf->saddr, (uint32_t)pbuf->eaddr,
+			    pbuf->prot, (uint32_t)pbuf->offset, maj, min,
+			    inode, *buf != '\0' ? " " : "", buf);
+		}
+
+		pbuf_next = pbuf->next;
+		kmem_free(pbuf, sizeof (*pbuf));
+		pbuf = pbuf_next;
+	}
+
+	kmem_free(buf, buflen);
+}
+
+/*
+ * lxpr_read_pid_mountinfo(): information about process mount points. e.g.:
+ *    14 19 0:13 / /sys rw,nosuid,nodev,noexec,relatime - sysfs sysfs rw
+ * mntid parid devnums root mntpnt mntopts - fstype mntsrc superopts
+ *
+ * We have to make up several of these fields.
+ */
+static void
+lxpr_read_pid_mountinfo(lxpr_node_t *lxpnp, lxpr_uiobuf_t *uiobuf)
+{
+	struct vfs *vfsp;
+	struct vfs *vfslist;
+	zone_t *zone = LXPTOZ(lxpnp);
+	struct print_data {
+		refstr_t *vfs_mntpt;
+		refstr_t *vfs_resource;
+		uint_t vfs_flag;
+		int vfs_fstype;
+		dev_t vfs_dev;
+		struct print_data *next;
+	} *print_head = NULL;
+	struct print_data **print_tail = &print_head;
+	struct print_data *printp;
+	int root_id = 15;	/* use a made-up value */
+	int mnt_id;
+
+	ASSERT(lxpnp->lxpr_type == LXPR_PID_MOUNTINFO ||
+	    lxpnp->lxpr_type == LXPR_PID_TID_MOUNTINFO);
+
+	vfs_list_read_lock();
+
+	/* root is the top-level, it does not appear in this output */
+	if (zone == global_zone) {
+		vfsp = vfslist = rootvfs;
+	} else {
+		vfsp = vfslist = zone->zone_vfslist;
+		/*
+		 * If the zone has a root entry, it will be the first in
+		 * the list.  If it doesn't, we conjure one up.
+		 */
+		if (vfslist == NULL || strcmp(refstr_value(vfsp->vfs_mntpt),
+		    zone->zone_rootpath) != 0) {
+			struct vfs *tvfsp;
+			/*
+			 * The root of the zone is not a mount point.  The vfs
+			 * we want to report is that of the zone's root vnode.
+			 */
+			tvfsp = zone->zone_rootvp->v_vfsp;
+
+			lxpr_uiobuf_printf(uiobuf,
+			    "%d 1 %d:%d / / %s - %s / %s\n",
+			    root_id,
+			    major(tvfsp->vfs_dev), minor(vfsp->vfs_dev),
+			    tvfsp->vfs_flag & VFS_RDONLY ? "ro" : "rw",
+			    vfssw[tvfsp->vfs_fstype].vsw_name,
+			    tvfsp->vfs_flag & VFS_RDONLY ? "ro" : "rw");
+
+		}
+		if (vfslist == NULL) {
+			vfs_list_unlock();
+			return;
+		}
+	}
+
+	/*
+	 * Later on we have to do a lookupname, which can end up causing
+	 * another vfs_list_read_lock() to be called. Which can lead to a
+	 * deadlock. To avoid this, we extract the data we need into a local
+	 * list, then we can run this list without holding vfs_list_read_lock()
+	 * We keep the list in the same order as the vfs_list
+	 */
+	do {
+		/* Skip mounts we shouldn't show */
+		if (vfsp->vfs_flag & VFS_NOMNTTAB) {
+			goto nextfs;
+		}
+
+		printp = kmem_alloc(sizeof (*printp), KM_SLEEP);
+		refstr_hold(vfsp->vfs_mntpt);
+		printp->vfs_mntpt = vfsp->vfs_mntpt;
+		refstr_hold(vfsp->vfs_resource);
+		printp->vfs_resource = vfsp->vfs_resource;
+		printp->vfs_flag = vfsp->vfs_flag;
+		printp->vfs_fstype = vfsp->vfs_fstype;
+		printp->vfs_dev = vfsp->vfs_dev;
+		printp->next = NULL;
+
+		*print_tail = printp;
+		print_tail = &printp->next;
+
+nextfs:
+		vfsp = (zone == global_zone) ?
+		    vfsp->vfs_next : vfsp->vfs_zone_next;
+
+	} while (vfsp != vfslist);
+
+	vfs_list_unlock();
+
+	mnt_id = root_id + 1;
+
+	/*
+	 * now we can run through what we've extracted without holding
+	 * vfs_list_read_lock()
+	 */
+	printp = print_head;
+	while (printp != NULL) {
+		struct print_data *printp_next;
+		const char *resource;
+		char *mntpt;
+		struct vnode *vp;
+		int error;
+
+		mntpt = (char *)refstr_value(printp->vfs_mntpt);
+		resource = refstr_value(printp->vfs_resource);
+
+		if (mntpt != NULL && mntpt[0] != '\0')
+			mntpt = ZONE_PATH_TRANSLATE(mntpt, zone);
+		else
+			mntpt = "-";
+
+		error = lookupname(mntpt, UIO_SYSSPACE, FOLLOW, NULLVPP, &vp);
+
+		if (error != 0)
+			goto nextp;
+
+		if (!(vp->v_flag & VROOT)) {
+			VN_RELE(vp);
+			goto nextp;
+		}
+		VN_RELE(vp);
+
+		if (resource != NULL && resource[0] != '\0') {
+			if (resource[0] == '/') {
+				resource = ZONE_PATH_VISIBLE(resource, zone) ?
+				    ZONE_PATH_TRANSLATE(resource, zone) : mntpt;
+			}
+		} else {
+			resource = "none";
+		}
+
+		/*
+		 * XXX parent ID is not tracked correctly here. Currently we
+		 * always assume the parent ID is the root ID.
+		 */
+		lxpr_uiobuf_printf(uiobuf,
+		    "%d %d %d:%d / %s %s - %s %s %s\n",
+		    mnt_id, root_id,
+		    major(printp->vfs_dev), minor(printp->vfs_dev),
+		    mntpt,
+		    printp->vfs_flag & VFS_RDONLY ? "ro" : "rw",
+		    vfssw[printp->vfs_fstype].vsw_name,
+		    resource,
+		    printp->vfs_flag & VFS_RDONLY ? "ro" : "rw");
+
+nextp:
+		printp_next = printp->next;
+		refstr_rele(printp->vfs_mntpt);
+		refstr_rele(printp->vfs_resource);
+		kmem_free(printp, sizeof (*printp));
+		printp = printp_next;
+
+		mnt_id++;
+	}
+}
+
+/*
+ * lxpr_read_pid_statm(): memory status file
+ */
+static void
+lxpr_read_pid_statm(lxpr_node_t *lxpnp, lxpr_uiobuf_t *uiobuf)
+{
+	proc_t *p;
+	struct as *as;
+	size_t vsize;
+	size_t rss;
+
+	ASSERT(lxpnp->lxpr_type == LXPR_PID_STATM ||
+	    lxpnp->lxpr_type == LXPR_PID_TID_STATM);
+
+	p = lxpr_lock(lxpnp->lxpr_pid);
+	if (p == NULL) {
+		lxpr_uiobuf_seterr(uiobuf, EINVAL);
+		return;
+	}
+
+	as = p->p_as;
+
+	mutex_exit(&p->p_lock);
+
+	AS_LOCK_ENTER(as, &as->a_lock, RW_READER);
+	vsize = btopr(as->a_resvsize);
+	rss = rm_asrss(as);
+	AS_LOCK_EXIT(as, &as->a_lock);
+
+	mutex_enter(&p->p_lock);
+	lxpr_unlock(p);
+
+	lxpr_uiobuf_printf(uiobuf,
+	    "%lu %lu %lu %lu %lu %lu %lu\n",
+	    vsize, rss, 0l, rss, 0l, 0l, 0l);
+}
+
+/*
+ * Look for either the main thread (lookup_id is 0) or the specified thread.
+ * If we're looking for the main thread but the proc does not have one, we
+ * fallback to using prchoose to get any thread available.
+ */
+static kthread_t *
+lxpr_get_thread(proc_t *p, uint_t lookup_id)
+{
+	kthread_t *t;
+	uint_t emul_tid;
+	lx_lwp_data_t *lwpd;
+	pid_t pid = p->p_pid;
+	pid_t init_pid = curproc->p_zone->zone_proc_initpid;
+
+	/* get specified thread  */
+	if ((t = p->p_tlist) == NULL)
+		return (NULL);
+
+	do {
+		if (lookup_id == 0 && t->t_tid == 1) {
+			thread_lock(t);
+			return (t);
+		}
+
+		lwpd = ttolxlwp(t);
+		if (lwpd == NULL) {
+			emul_tid = t->t_tid;
+		} else {
+			if (pid == init_pid && lookup_id == 1) {
+				emul_tid = t->t_tid;
+			} else {
+				emul_tid = lwpd->br_pid;
+			}
+		}
+		if (emul_tid == lookup_id) {
+			thread_lock(t);
+			return (t);
+		}
+	} while ((t = t->t_forw) != p->p_tlist);
+
+	if (lookup_id == 0)
+		return (prchoose(p));
+	return (NULL);
+}
+
+/*
+ * Lookup the real pid for procs 0 or 1.
+ */
+static pid_t
+get_real_pid(pid_t p)
+{
+	pid_t find_pid;
+
+	if (p == 1) {
+		find_pid = curproc->p_zone->zone_proc_initpid;
+	} else if (p == 0) {
+		find_pid = curproc->p_zone->zone_zsched->p_pid;
+	} else {
+		find_pid = p;
+	}
+
+	return (find_pid);
+}
+
+/*
+ * pid/tid common code to read status file
+ */
+static void
+lxpr_read_status_common(lxpr_node_t *lxpnp, lxpr_uiobuf_t *uiobuf,
+    uint_t lookup_id)
+{
+	proc_t *p;
+	kthread_t *t;
+	user_t *up;
+	cred_t *cr;
+	const gid_t *groups;
+	int    ngroups;
+	struct as *as;
+	char *status;
+	pid_t pid, ppid;
+	k_sigset_t current, ignore, handle;
+	int    i, lx_sig;
+	pid_t real_pid;
+
+	real_pid = get_real_pid(lxpnp->lxpr_pid);
+	p = lxpr_lock(real_pid);
+	if (p == NULL) {
+		lxpr_uiobuf_seterr(uiobuf, EINVAL);
+		return;
+	}
+
+	pid = p->p_pid;
+
+	/*
+	 * Convert pid to the Linux default of 1 if we're the zone's init
+	 * process or if we're the zone's zsched the pid is 0.
+	 */
+	if (pid == curproc->p_zone->zone_proc_initpid) {
+		pid = 1;
+		ppid = 0;	/* parent pid for init is 0 */
+	} else if (pid == curproc->p_zone->zone_zsched->p_pid) {
+		pid = 0;	/* zsched is pid 0 */
+		ppid = 0;	/* parent pid for zsched is itself */
+	} else {
+		/*
+		 * Make sure not to reference parent PIDs that reside outside
+		 * the zone
+		 */
+		ppid = ((p->p_flag & SZONETOP)
+		    ? curproc->p_zone->zone_zsched->p_pid : p->p_ppid);
+
+		/*
+		 * Convert ppid to the Linux default of 1 if our parent is the
+		 * zone's init process
+		 */
+		if (ppid == curproc->p_zone->zone_proc_initpid)
+			ppid = 1;
+	}
+
+	t = lxpr_get_thread(p, lookup_id);
+	if (t != NULL) {
+		switch (t->t_state) {
+		case TS_SLEEP:
+			status = "S (sleeping)";
+			break;
+		case TS_RUN:
+		case TS_ONPROC:
+			status = "R (running)";
+			break;
+		case TS_ZOMB:
+			status = "Z (zombie)";
+			break;
+		case TS_STOPPED:
+			status = "T (stopped)";
+			break;
+		default:
+			status = "! (unknown)";
+			break;
+		}
+		thread_unlock(t);
+	} else {
+		if (lookup_id != 0) {
+			/* we can't find this specific thread */
+			lxpr_uiobuf_seterr(uiobuf, EINVAL);
+			lxpr_unlock(p);
+			return;
+		}
+
+		/*
+		 * there is a hole in the exit code, where a proc can have
+		 * no threads but it is yet to be flagged SZOMB. We will
+		 * assume we are about to become a zombie
+		 */
+		status = "Z (zombie)";
+	}
+
+	up = PTOU(p);
+	mutex_enter(&p->p_crlock);
+	crhold(cr = p->p_cred);
+	mutex_exit(&p->p_crlock);
+
+	lxpr_uiobuf_printf(uiobuf,
+	    "Name:\t%s\n"
+	    "State:\t%s\n"
+	    "Tgid:\t%d\n"
+	    "Pid:\t%d\n"
+	    "PPid:\t%d\n"
+	    "TracerPid:\t%d\n"
+	    "Uid:\t%u\t%u\t%u\t%u\n"
+	    "Gid:\t%u\t%u\t%u\t%u\n"
+	    "FDSize:\t%d\n"
+	    "Groups:\t",
+	    up->u_comm,
+	    status,
+	    pid, /* thread group id - same as pid */
+	    (lookup_id == 0) ? pid : lxpnp->lxpr_desc,
+	    ppid,
+	    0,
+	    crgetruid(cr), crgetuid(cr), crgetsuid(cr), crgetuid(cr),
+	    crgetrgid(cr), crgetgid(cr), crgetsgid(cr), crgetgid(cr),
+	    p->p_fno_ctl);
+
+
+	ngroups = crgetngroups(cr);
+	groups  = crgetgroups(cr);
+	for (i = 0; i < ngroups; i++) {
+		lxpr_uiobuf_printf(uiobuf,
+		    "%u ",
+		    groups[i]);
+	}
+	crfree(cr);
+
+	as = p->p_as;
+	if ((p->p_stat != SZOMB) && !(p->p_flag & SSYS) && (as != &kas)) {
+		size_t vsize, nlocked, rss;
+
+		mutex_exit(&p->p_lock);
+		AS_LOCK_ENTER(as, &as->a_lock, RW_READER);
+		vsize = as->a_resvsize;
+		rss = rm_asrss(as);
+		AS_LOCK_EXIT(as, &as->a_lock);
+		mutex_enter(&p->p_lock);
+		nlocked = p->p_locked_mem;
+
+		lxpr_uiobuf_printf(uiobuf,
+		    "\n"
+		    "VmSize:\t%8lu kB\n"
+		    "VmLck:\t%8lu kB\n"
+		    "VmRSS:\t%8lu kB\n"
+		    "VmData:\t%8lu kB\n"
+		    "VmStk:\t%8lu kB\n"
+		    "VmExe:\t%8lu kB\n"
+		    "VmLib:\t%8lu kB",
+		    btok(vsize),
+		    btok(nlocked),
+		    ptok(rss),
+		    0l,
+		    btok(p->p_stksize),
+		    ptok(rss),
+		    0l);
+	}
+
+	lxpr_uiobuf_printf(uiobuf, "\nThreads:\t%u", p->p_lwpcnt);
+
+	sigemptyset(&current);
+	sigemptyset(&ignore);
+	sigemptyset(&handle);
+
+	for (i = 1; i < NSIG; i++) {
+		lx_sig = stol_signo[i];
+
+		if ((lx_sig > 0) && (lx_sig <= LX_NSIG)) {
+			if (sigismember(&p->p_sig, i))
+				sigaddset(&current, lx_sig);
+
+			if (up->u_signal[i - 1] == SIG_IGN)
+				sigaddset(&ignore, lx_sig);
+			else if (up->u_signal[i - 1] != SIG_DFL)
+				sigaddset(&handle, lx_sig);
+		}
+	}
+
+	lxpr_uiobuf_printf(uiobuf,
+	    "\n"
+	    "SigPnd:\t%08x%08x\n"
+	    "SigBlk:\t%08x%08x\n"
+	    "SigIgn:\t%08x%08x\n"
+	    "SigCgt:\t%08x%08x\n"
+	    "CapInh:\t%016x\n"
+	    "CapPrm:\t%016x\n"
+	    "CapEff:\t%016x\n",
+	    current.__sigbits[1], current.__sigbits[0],
+	    0, 0, /* signals blocked on per thread basis */
+	    ignore.__sigbits[1], ignore.__sigbits[0],
+	    handle.__sigbits[1], handle.__sigbits[0],
+	    /* Can't do anything with linux capabilities */
+	    0,
+	    0,
+	    0);
+
+	lxpr_unlock(p);
+}
+
+/*
+ * lxpr_read_pid_status(): status file
+ */
+static void
+lxpr_read_pid_status(lxpr_node_t *lxpnp, lxpr_uiobuf_t *uiobuf)
+{
+	ASSERT(lxpnp->lxpr_type == LXPR_PID_STATUS);
+
+	lxpr_read_status_common(lxpnp, uiobuf, 0);
+}
+
+/*
+ * lxpr_read_pid_tid_status(): status file
+ */
+static void
+lxpr_read_pid_tid_status(lxpr_node_t *lxpnp, lxpr_uiobuf_t *uiobuf)
+{
+	ASSERT(lxpnp->lxpr_type == LXPR_PID_TID_STATUS);
+	lxpr_read_status_common(lxpnp, uiobuf, lxpnp->lxpr_desc);
+}
+
+/*
+ * pid/tid common code to read stat file
+ */
+static void
+lxpr_read_stat_common(lxpr_node_t *lxpnp, lxpr_uiobuf_t *uiobuf,
+    uint_t lookup_id)
+{
+	proc_t *p;
+	kthread_t *t;
+	struct as *as;
+	char stat;
+	pid_t pid, ppid, pgpid, spid;
+	gid_t psgid;
+	dev_t psdev;
+	size_t rss, vsize;
+	int nice, pri;
+	caddr_t wchan;
+	processorid_t cpu;
+	pid_t real_pid;
+
+	real_pid = get_real_pid(lxpnp->lxpr_pid);
+	p = lxpr_lock(real_pid);
+	if (p == NULL) {
+		lxpr_uiobuf_seterr(uiobuf, EINVAL);
+		return;
+	}
+
+	pid = p->p_pid;
+
+	/*
+	 * Set Linux defaults if we're the zone's init process
+	 */
+	if (pid == curproc->p_zone->zone_proc_initpid) {
+		pid = 1;		/* PID for init */
+		ppid = 0;		/* parent PID for init is 0 */
+		pgpid = 0;		/* process group for init is 0 */
+		psgid = (gid_t)-1;	/* credential GID for init is -1 */
+		spid = 0;		/* session id for init is 0 */
+		psdev = 0;		/* session device for init is 0 */
+	} else if (pid == curproc->p_zone->zone_zsched->p_pid) {
+		pid = 0;		/* PID for zsched */
+		ppid = 0;		/* parent PID for zsched is 0 */
+		pgpid = 0;		/* process group for zsched is 0 */
+		psgid = (gid_t)-1;	/* credential GID for zsched is -1 */
+		spid = 0;		/* session id for zsched is 0 */
+		psdev = 0;		/* session device for zsched is 0 */
+	} else {
+		/*
+		 * Make sure not to reference parent PIDs that reside outside
+		 * the zone
+		 */
+		ppid = ((p->p_flag & SZONETOP) ?
+		    curproc->p_zone->zone_zsched->p_pid : p->p_ppid);
+
+		/*
+		 * Convert ppid to the Linux default of 1 if our parent is the
+		 * zone's init process
+		 */
+		if (ppid == curproc->p_zone->zone_proc_initpid)
+			ppid = 1;
+
+		pgpid = p->p_pgrp;
+
+		mutex_enter(&p->p_splock);
+		mutex_enter(&p->p_sessp->s_lock);
+		spid = p->p_sessp->s_sid;
+		psdev = p->p_sessp->s_dev;
+		if (p->p_sessp->s_cred)
+			psgid = crgetgid(p->p_sessp->s_cred);
+		else
+			psgid = crgetgid(p->p_cred);
+
+		mutex_exit(&p->p_sessp->s_lock);
+		mutex_exit(&p->p_splock);
+	}
+
+	t = lxpr_get_thread(p, lookup_id);
+	if (t != NULL) {
+		switch (t->t_state) {
+		case TS_SLEEP:
+			stat = 'S'; break;
+		case TS_RUN:
+		case TS_ONPROC:
+			stat = 'R'; break;
+		case TS_ZOMB:
+			stat = 'Z'; break;
+		case TS_STOPPED:
+			stat = 'T'; break;
+		default:
+			stat = '!'; break;
+		}
+
+		if (CL_DONICE(t, NULL, 0, &nice) != 0)
+			nice = 0;
+
+		pri = t->t_pri;
+		wchan = t->t_wchan;
+		cpu = t->t_cpu->cpu_id;
+		thread_unlock(t);
+	} else {
+		if (lookup_id != 0) {
+			/* we can't find this specific thread */
+			lxpr_uiobuf_seterr(uiobuf, EINVAL);
+			lxpr_unlock(p);
+			return;
+		}
+
+		/* Only zombies have no threads */
+		stat = 'Z';
+		nice = 0;
+		pri = 0;
+		wchan = 0;
+		cpu = 0;
+	}
+	as = p->p_as;
+	mutex_exit(&p->p_lock);
+	AS_LOCK_ENTER(as, &as->a_lock, RW_READER);
+	vsize = as->a_resvsize;
+	rss = rm_asrss(as);
+	AS_LOCK_EXIT(as, &as->a_lock);
+	mutex_enter(&p->p_lock);
+
+	lxpr_uiobuf_printf(uiobuf,
+	    "%d (%s) %c %d %d %d %d %d "
+	    "%lu %lu %lu %lu %lu "
+	    "%lu %lu %ld %ld "
+	    "%d %d %d "
+	    "%lu "
+	    "%lu "
+	    "%lu %ld %llu "
+	    "%lu %lu %u "
+	    "%lu %lu "
+	    "%lu %lu %lu %lu "
+	    "%lu "
+	    "%lu %lu "
+	    "%d "
+	    "%d"
+	    "\n",
+	    (lookup_id == 0) ? pid : lxpnp->lxpr_desc,
+	    PTOU(p)->u_comm, stat, ppid, pgpid, spid, psdev, psgid,
+	    0l, 0l, 0l, 0l, 0l, /* flags, minflt, cminflt, majflt, cmajflt */
+	    p->p_utime, p->p_stime, p->p_cutime, p->p_cstime,
+	    pri, nice, p->p_lwpcnt,
+	    0l, /* itrealvalue (time before next SIGALRM) */
+	    PTOU(p)->u_ticks,
+	    vsize, rss, p->p_vmem_ctl,
+	    0l, 0l, USRSTACK, /* startcode, endcode, startstack */
+	    0l, 0l, /* kstkesp, kstkeip */
+	    0l, 0l, 0l, 0l, /* signal, blocked, sigignore, sigcatch */
+	    wchan,
+	    0l, 0l, /* nswap, cnswap */
+	    0, /* exit_signal */
+	    cpu);
+
+	lxpr_unlock(p);
+}
+
+/*
+ * lxpr_read_pid_stat(): pid stat file
+ */
+static void
+lxpr_read_pid_stat(lxpr_node_t *lxpnp, lxpr_uiobuf_t *uiobuf)
+{
+	ASSERT(lxpnp->lxpr_type == LXPR_PID_STAT);
+
+	lxpr_read_stat_common(lxpnp, uiobuf, 0);
+}
+
+/*
+ * lxpr_read_pid_tid_stat(): pid stat file
+ */
+static void
+lxpr_read_pid_tid_stat(lxpr_node_t *lxpnp, lxpr_uiobuf_t *uiobuf)
+{
+	ASSERT(lxpnp->lxpr_type == LXPR_PID_TID_STAT);
+	lxpr_read_stat_common(lxpnp, uiobuf, lxpnp->lxpr_desc);
+}
+
+/* ARGSUSED */
+static void
+lxpr_read_net_arp(lxpr_node_t *lxpnp, lxpr_uiobuf_t *uiobuf)
+{
+}
+
+struct lxpr_ifstat {
+	uint64_t rx_bytes;
+	uint64_t rx_packets;
+	uint64_t rx_errors;
+	uint64_t rx_drop;
+	uint64_t tx_bytes;
+	uint64_t tx_packets;
+	uint64_t tx_errors;
+	uint64_t tx_drop;
+	uint64_t collisions;
+	uint64_t rx_multicast;
+};
+
+static void *
+lxpr_kstat_read(kstat_t *kn, boolean_t byname, size_t *size, int *num)
+{
+	kstat_t *kp;
+	int i, nrec = 0;
+	size_t bufsize;
+	void *buf = NULL;
+
+	if (byname == B_TRUE) {
+		kp = kstat_hold_byname(kn->ks_module, kn->ks_instance,
+		    kn->ks_name, getzoneid());
+	} else {
+		kp = kstat_hold_bykid(kn->ks_kid, getzoneid());
+	}
+	if (kp == NULL) {
+		return (NULL);
+	}
+	if (kp->ks_flags & KSTAT_FLAG_INVALID) {
+		kstat_rele(kp);
+		return (NULL);
+	}
+
+	bufsize = kp->ks_data_size + 1;
+	kstat_rele(kp);
+
+	/*
+	 * The kstat in question is released so that kmem_alloc(KM_SLEEP) is
+	 * performed without it held.  After the alloc, the kstat is reacquired
+	 * and its size is checked again. If the buffer is no longer large
+	 * enough, the alloc and check are repeated up to three times.
+	 */
+	for (i = 0; i < 2; i++) {
+		buf = kmem_alloc(bufsize, KM_SLEEP);
+
+		/* Check if bufsize still appropriate */
+		if (byname == B_TRUE) {
+			kp = kstat_hold_byname(kn->ks_module, kn->ks_instance,
+			    kn->ks_name, getzoneid());
+		} else {
+			kp = kstat_hold_bykid(kn->ks_kid, getzoneid());
+		}
+		if (kp == NULL || kp->ks_flags & KSTAT_FLAG_INVALID) {
+			if (kp != NULL) {
+				kstat_rele(kp);
+			}
+			kmem_free(buf, bufsize);
+			return (NULL);
+		}
+		KSTAT_ENTER(kp);
+		(void) KSTAT_UPDATE(kp, KSTAT_READ);
+		if (bufsize < kp->ks_data_size) {
+			kmem_free(buf, bufsize);
+			buf = NULL;
+			bufsize = kp->ks_data_size + 1;
+			KSTAT_EXIT(kp);
+			kstat_rele(kp);
+			continue;
+		} else {
+			if (KSTAT_SNAPSHOT(kp, buf, KSTAT_READ) != 0) {
+				kmem_free(buf, bufsize);
+				buf = NULL;
+			}
+			nrec = kp->ks_ndata;
+			KSTAT_EXIT(kp);
+			kstat_rele(kp);
+			break;
+		}
+	}
+
+	if (buf != NULL) {
+		*size = bufsize;
+		*num = nrec;
+	}
+	return (buf);
+}
+
+static int
+lxpr_kstat_ifstat(kstat_t *kn, struct lxpr_ifstat *ifs)
+{
+	kstat_named_t *kp;
+	int i, num;
+	size_t size;
+
+	/*
+	 * Search by name instead of by kid since there's a small window to
+	 * race against kstats being added/removed.
+	 */
+	bzero(ifs, sizeof (*ifs));
+	kp = (kstat_named_t *)lxpr_kstat_read(kn, B_TRUE, &size, &num);
+	if (kp == NULL)
+		return (-1);
+	for (i = 0; i < num; i++) {
+		if (strncmp(kp[i].name, "rbytes64", KSTAT_STRLEN) == 0)
+			ifs->rx_bytes = kp[i].value.ui64;
+		else if (strncmp(kp[i].name, "ipackets64", KSTAT_STRLEN) == 0)
+			ifs->rx_packets = kp[i].value.ui64;
+		else if (strncmp(kp[i].name, "ierrors", KSTAT_STRLEN) == 0)
+			ifs->rx_errors = kp[i].value.ui32;
+		else if (strncmp(kp[i].name, "norcvbuf", KSTAT_STRLEN) == 0)
+			ifs->rx_drop = kp[i].value.ui32;
+		else if (strncmp(kp[i].name, "multircv", KSTAT_STRLEN) == 0)
+			ifs->rx_multicast = kp[i].value.ui32;
+		else if (strncmp(kp[i].name, "obytes64", KSTAT_STRLEN) == 0)
+			ifs->tx_bytes = kp[i].value.ui64;
+		else if (strncmp(kp[i].name, "opackets64", KSTAT_STRLEN) == 0)
+			ifs->tx_packets = kp[i].value.ui64;
+		else if (strncmp(kp[i].name, "oerrors", KSTAT_STRLEN) == 0)
+			ifs->tx_errors = kp[i].value.ui32;
+		else if (strncmp(kp[i].name, "noxmtbuf", KSTAT_STRLEN) == 0)
+			ifs->tx_drop = kp[i].value.ui32;
+		else if (strncmp(kp[i].name, "collisions", KSTAT_STRLEN) == 0)
+			ifs->collisions = kp[i].value.ui32;
+	}
+	kmem_free(kp, size);
+	return (0);
+}
+
+/* ARGSUSED */
+static void
+lxpr_read_net_dev(lxpr_node_t *lxpnp, lxpr_uiobuf_t *uiobuf)
+{
+	kstat_t *ksr;
+	kstat_t ks0;
+	int i, nidx;
+	size_t sidx;
+	struct lxpr_ifstat ifs;
+
+	lxpr_uiobuf_printf(uiobuf, "Inter-|   Receive                   "
+	    "                             |  Transmit\n");
+	lxpr_uiobuf_printf(uiobuf, " face |bytes    packets errs drop fifo"
+	    " frame compressed multicast|bytes    packets errs drop fifo"
+	    " colls carrier compressed\n");
+
+	ks0.ks_kid = 0;
+	ksr = (kstat_t *)lxpr_kstat_read(&ks0, B_FALSE, &sidx, &nidx);
+	if (ksr == NULL)
+		return;
+
+	for (i = 1; i < nidx; i++) {
+		if (strncmp(ksr[i].ks_module, "link", KSTAT_STRLEN) == 0 ||
+		    strncmp(ksr[i].ks_module, "lo", KSTAT_STRLEN) == 0) {
+			if (lxpr_kstat_ifstat(&ksr[i], &ifs) != 0)
+				continue;
+
+			/* Overwriting the name is ok in the local snapshot */
+			lx_ifname_convert(ksr[i].ks_name, LX_IF_FROMNATIVE);
+			lxpr_uiobuf_printf(uiobuf, "%6s: %7llu %7llu %4lu "
+			    "%4lu %4u %5u %10u %9lu %8llu %7llu %4lu %4lu %4u "
+			    "%5lu %7u %10u\n",
+			    ksr[i].ks_name,
+			    ifs.rx_bytes, ifs.rx_packets,
+			    ifs.rx_errors, ifs.rx_drop,
+			    0, 0, 0, ifs.rx_multicast,
+			    ifs.tx_bytes, ifs.tx_packets,
+			    ifs.tx_errors, ifs.tx_drop,
+			    0, ifs.collisions, 0, 0);
+		}
+	}
+
+	kmem_free(ksr, sidx);
+}
+
+/* ARGSUSED */
+static void
+lxpr_read_net_dev_mcast(lxpr_node_t *lxpnp, lxpr_uiobuf_t *uiobuf)
+{
+}
+
+static void
+lxpr_inet6_out(const in6_addr_t *addr, char buf[33])
+{
+	const uint8_t *ip = addr->s6_addr;
+	char digits[] = "0123456789abcdef";
+	int i;
+	for (i = 0; i < 16; i++) {
+		buf[2 * i] = digits[ip[i] >> 4];
+		buf[2 * i + 1] = digits[ip[i] & 0xf];
+	}
+	buf[32] = '\0';
+}
+
+/* ARGSUSED */
+static void
+lxpr_read_net_if_inet6(lxpr_node_t *lxpnp, lxpr_uiobuf_t *uiobuf)
+{
+	netstack_t *ns;
+	ip_stack_t *ipst;
+	ill_t *ill;
+	ipif_t *ipif;
+	ill_walk_context_t	ctx;
+	char ifname[LIFNAMSIZ], ip6out[33];
+
+	ns = netstack_get_current();
+	if (ns == NULL)
+		return;
+	ipst = ns->netstack_ip;
+
+	rw_enter(&ipst->ips_ill_g_lock, RW_READER);
+	ill = ILL_START_WALK_V6(&ctx, ipst);
+
+	for (; ill != NULL; ill = ill_next(&ctx, ill)) {
+		for (ipif = ill->ill_ipif; ipif != NULL;
+		    ipif = ipif->ipif_next) {
+			uint_t index = ill->ill_phyint->phyint_ifindex;
+			int plen = ip_mask_to_plen_v6(&ipif->ipif_v6net_mask);
+			in6addr_scope_t scope = ip_addr_scope_v6(
+			    &ipif->ipif_v6lcl_addr);
+			/* Always report PERMANENT flag */
+			int flag = 0x80;
+
+			ipif_get_name(ipif, ifname, sizeof (ifname));
+			lx_ifname_convert(ifname, LX_IF_FROMNATIVE);
+			lxpr_inet6_out(&ipif->ipif_v6lcl_addr, ip6out);
+			/* Scope output is shifted on Linux */
+			scope = scope << 4;
+
+			lxpr_uiobuf_printf(uiobuf, "%32s %02x %02x %02x %02x"
+			    " %8s\n", ip6out, index, plen, scope, flag, ifname);
+		}
+	}
+	rw_exit(&ipst->ips_ill_g_lock);
+	netstack_rele(ns);
+}
+
+/* ARGSUSED */
+static void
+lxpr_read_net_igmp(lxpr_node_t *lxpnp, lxpr_uiobuf_t *uiobuf)
+{
+}
+
+/* ARGSUSED */
+static void
+lxpr_read_net_ip_mr_cache(lxpr_node_t *lxpnp, lxpr_uiobuf_t *uiobuf)
+{
+}
+
+/* ARGSUSED */
+static void
+lxpr_read_net_ip_mr_vif(lxpr_node_t *lxpnp, lxpr_uiobuf_t *uiobuf)
+{
+}
+
+static void
+lxpr_format_route_ipv6(ire_t *ire, lxpr_uiobuf_t *uiobuf)
+{
+	uint32_t flags;
+	char name[IFNAMSIZ];
+	char ipv6addr[33];
+
+	lxpr_inet6_out(&ire->ire_addr_v6, ipv6addr);
+	lxpr_uiobuf_printf(uiobuf, "%s %02x ", ipv6addr,
+	    ip_mask_to_plen_v6(&ire->ire_mask_v6));
+
+	/* punt on this for now */
+	lxpr_uiobuf_printf(uiobuf, "%s %02x ",
+	    "00000000000000000000000000000000", 0);
+
+	lxpr_inet6_out(&ire->ire_gateway_addr_v6, ipv6addr);
+	lxpr_uiobuf_printf(uiobuf, "%s", ipv6addr);
+
+	flags = ire->ire_flags &
+	    (RTF_UP|RTF_GATEWAY|RTF_HOST|RTF_DYNAMIC|RTF_MODIFIED);
+	/* Linux's RTF_LOCAL equivalent */
+	if (ire->ire_metrics.iulp_local)
+		flags |= 0x80000000;
+
+	if (ire->ire_ill != NULL) {
+		ill_get_name(ire->ire_ill, name, sizeof (name));
+		lx_ifname_convert(name, LX_IF_FROMNATIVE);
+	} else {
+		name[0] = '\0';
+	}
+
+	lxpr_uiobuf_printf(uiobuf, " %08x %08x %08x %08x %8s\n",
+	    0, /* metric */
+	    ire->ire_refcnt,
+	    0,
+	    flags,
+	    name);
+}
+
+/* ARGSUSED */
+static void
+lxpr_read_net_ipv6_route(lxpr_node_t *lxpnp, lxpr_uiobuf_t *uiobuf)
+{
+	netstack_t *ns;
+	ip_stack_t *ipst;
+
+	ns = netstack_get_current();
+	if (ns == NULL)
+		return;
+	ipst = ns->netstack_ip;
+
+	/*
+	 * LX branded zones are expected to have exclusive IP stack, hence
+	 * using ALL_ZONES as the zoneid filter.
+	 */
+	ire_walk_v6(&lxpr_format_route_ipv6, uiobuf, ALL_ZONES, ipst);
+
+	netstack_rele(ns);
+}
+
+/* ARGSUSED */
+static void
+lxpr_read_net_mcfilter(lxpr_node_t *lxpnp, lxpr_uiobuf_t *uiobuf)
+{
+}
+
+/* ARGSUSED */
+static void
+lxpr_read_net_netstat(lxpr_node_t *lxpnp, lxpr_uiobuf_t *uiobuf)
+{
+}
+
+/* ARGSUSED */
+static void
+lxpr_read_net_raw(lxpr_node_t *lxpnp, lxpr_uiobuf_t *uiobuf)
+{
+}
+
+#define	LXPR_SKIP_ROUTE(type)	\
+	(((IRE_IF_CLONE | IRE_BROADCAST | IRE_MULTICAST | \
+	IRE_NOROUTE | IRE_LOOPBACK | IRE_LOCAL) & type) != 0)
+
+static void
+lxpr_format_route_ipv4(ire_t *ire, lxpr_uiobuf_t *uiobuf)
+{
+	uint32_t flags;
+	char name[IFNAMSIZ];
+	ill_t *ill;
+	ire_t *nire;
+	ipif_t *ipif;
+	ipaddr_t gateway;
+
+	if (LXPR_SKIP_ROUTE(ire->ire_type) || ire->ire_testhidden != 0)
+		return;
+
+	/* These route flags have direct Linux equivalents */
+	flags = ire->ire_flags &
+	    (RTF_UP|RTF_GATEWAY|RTF_HOST|RTF_DYNAMIC|RTF_MODIFIED);
+
+	/*
+	 * Search for a suitable IRE for naming purposes.
+	 * On Linux, the default route is typically associated with the
+	 * interface used to access gateway.  The default IRE on Illumos
+	 * typically lacks an ill reference but its parent might have one.
+	 */
+	nire = ire;
+	do {
+		ill = nire->ire_ill;
+		nire = nire->ire_dep_parent;
+	} while (ill == NULL && nire != NULL);
+	if (ill != NULL) {
+		ill_get_name(ill, name, sizeof (name));
+		lx_ifname_convert(name, LX_IF_FROMNATIVE);
+	} else {
+		name[0] = '*';
+		name[1] = '\0';
+	}
+
+	/*
+	 * Linux suppresses the gateway address for directly connected
+	 * interface networks.  To emulate this behavior, we walk all addresses
+	 * of a given route interface.  If one matches the gateway, it is
+	 * displayed as NULL.
+	 */
+	gateway = ire->ire_gateway_addr;
+	if ((ill = ire->ire_ill) != NULL) {
+		for (ipif = ill->ill_ipif; ipif != NULL;
+		    ipif = ipif->ipif_next) {
+			if (ipif->ipif_lcl_addr == gateway) {
+				gateway = 0;
+				break;
+			}
+		}
+	}
+
+	lxpr_uiobuf_printf(uiobuf, "%s\t%08X\t%08X\t%04X\t%d\t%u\t"
+	    "%d\t%08X\t%d\t%u\t%u\n",
+	    name,
+	    ire->ire_addr,
+	    gateway,
+	    flags, 0, 0,
+	    0, /* priority */
+	    ire->ire_mask,
+	    0, 0, /* mss, window */
+	    ire->ire_metrics.iulp_rtt);
+}
+
+/* ARGSUSED */
+static void
+lxpr_read_net_route(lxpr_node_t *lxpnp, lxpr_uiobuf_t *uiobuf)
+{
+	netstack_t *ns;
+	ip_stack_t *ipst;
+
+	lxpr_uiobuf_printf(uiobuf, "Iface\tDestination\tGateway \tFlags\t"
+	    "RefCnt\tUse\tMetric\tMask\t\tMTU\tWindow\tIRTT\n");
+
+	ns = netstack_get_current();
+	if (ns == NULL)
+		return;
+	ipst = ns->netstack_ip;
+
+	/*
+	 * LX branded zones are expected to have exclusive IP stack, hence
+	 * using ALL_ZONES as the zoneid filter.
+	 */
+	ire_walk_v4(&lxpr_format_route_ipv4, uiobuf, ALL_ZONES, ipst);
+
+	netstack_rele(ns);
+}
+
+/* ARGSUSED */
+static void
+lxpr_read_net_rpc(lxpr_node_t *lxpnp, lxpr_uiobuf_t *uiobuf)
+{
+}
+
+/* ARGSUSED */
+static void
+lxpr_read_net_rt_cache(lxpr_node_t *lxpnp, lxpr_uiobuf_t *uiobuf)
+{
+}
+
+/* ARGSUSED */
+static void
+lxpr_read_net_sockstat(lxpr_node_t *lxpnp, lxpr_uiobuf_t *uiobuf)
+{
+}
+
+typedef struct lxpr_snmp_table {
+	const char *lst_proto;
+	const char *lst_fields[];
+} lxpr_snmp_table_t;
+
+static lxpr_snmp_table_t lxpr_snmp_ip = { "ip",
+	{
+	"forwarding", "defaultTTL", "inReceives", "inHdrErrors",
+	"inAddrErrors", "forwDatagrams", "inUnknownProtos", "inDiscards",
+	"inDelivers", "outRequests", "outDiscards", "outNoRoutes",
+	"reasmTimeout", "reasmReqds", "reasmOKs", "reasmFails", "fragOKs",
+	"fragFails", "fragCreates",
+	NULL
+	}
+};
+static lxpr_snmp_table_t lxpr_snmp_icmp = { "icmp",
+	{
+	"inMsgs", "inErrors", "inCsumErrors", "inDestUnreachs", "inTimeExcds",
+	"inParmProbs", "inSrcQuenchs", "inRedirects", "inEchos", "inEchoReps",
+	"inTimestamps", "inTimestampReps", "inAddrMasks", "inAddrMaskReps",
+	"outMsgs", "outErrors", "outDestUnreachs", "outTimeExcds",
+	"outParmProbs", "outSrcQuenchs", "outRedirects", "outEchos",
+	"outEchoReps", "outTimestamps", "outTimestampReps", "outAddrMasks",
+	"outAddrMaskReps",
+	NULL
+	}
+};
+static lxpr_snmp_table_t lxpr_snmp_tcp = { "tcp",
+	{
+	"rtoAlgorithm", "rtoMin", "rtoMax", "maxConn", "activeOpens",
+	"passiveOpens", "attemptFails", "estabResets", "currEstab", "inSegs",
+	"outSegs", "retransSegs", "inErrs", "outRsts", "inCsumErrors",
+	NULL
+	}
+};
+static lxpr_snmp_table_t lxpr_snmp_udp = { "udp",
+	{
+	"inDatagrams", "noPorts", "inErrors", "outDatagrams", "rcvbufErrors",
+	"sndbufErrors", "inCsumErrors",
+	NULL
+	}
+};
+
+static lxpr_snmp_table_t *lxpr_net_snmptab[] = {
+	&lxpr_snmp_ip,
+	&lxpr_snmp_icmp,
+	&lxpr_snmp_tcp,
+	&lxpr_snmp_udp,
+	NULL
+};
+
+static void
+lxpr_kstat_print_tab(lxpr_uiobuf_t *uiobuf, lxpr_snmp_table_t *table,
+    kstat_t *kn)
+{
+	kstat_named_t *klist;
+	char upname[KSTAT_STRLEN], upfield[KSTAT_STRLEN];
+	int i, j, num;
+	size_t size;
+
+	klist = (kstat_named_t *)lxpr_kstat_read(kn, B_TRUE, &size, &num);
+	if (klist == NULL)
+		return;
+
+	/* Print the header line, fields capitalized */
+	(void) strncpy(upname, table->lst_proto, KSTAT_STRLEN);
+	upname[0] = toupper(upname[0]);
+	lxpr_uiobuf_printf(uiobuf, "%s:", upname);
+	for (i = 0; table->lst_fields[i] != NULL; i++) {
+		(void) strncpy(upfield, table->lst_fields[i], KSTAT_STRLEN);
+		upfield[0] = toupper(upfield[0]);
+		lxpr_uiobuf_printf(uiobuf, " %s", upfield);
+	}
+	lxpr_uiobuf_printf(uiobuf, "\n%s:", upname);
+
+	/* Then loop back through to print the value line. */
+	for (i = 0; table->lst_fields[i] != NULL; i++) {
+		kstat_named_t *kpoint = NULL;
+		for (j = 0; j < num; j++) {
+			if (strncmp(klist[j].name, table->lst_fields[i],
+			    KSTAT_STRLEN) == 0) {
+				kpoint = &klist[j];
+				break;
+			}
+		}
+		if (kpoint == NULL) {
+			/* Output 0 for unknown fields */
+			lxpr_uiobuf_printf(uiobuf, " 0");
+		} else {
+			switch (kpoint->data_type) {
+			case KSTAT_DATA_INT32:
+				lxpr_uiobuf_printf(uiobuf, " %d",
+				    kpoint->value.i32);
+				break;
+			case KSTAT_DATA_UINT32:
+				lxpr_uiobuf_printf(uiobuf, " %u",
+				    kpoint->value.ui32);
+				break;
+			case KSTAT_DATA_INT64:
+				lxpr_uiobuf_printf(uiobuf, " %ld",
+				    kpoint->value.l);
+				break;
+			case KSTAT_DATA_UINT64:
+				lxpr_uiobuf_printf(uiobuf, " %lu",
+				    kpoint->value.ul);
+				break;
+			}
+		}
+	}
+	lxpr_uiobuf_printf(uiobuf, "\n");
+	kmem_free(klist, size);
+}
+
+/* ARGSUSED */
+static void
+lxpr_read_net_snmp(lxpr_node_t *lxpnp, lxpr_uiobuf_t *uiobuf)
+{
+	kstat_t *ksr;
+	kstat_t ks0;
+	lxpr_snmp_table_t **table = lxpr_net_snmptab;
+	int i, t, nidx;
+	size_t sidx;
+
+	ks0.ks_kid = 0;
+	ksr = (kstat_t *)lxpr_kstat_read(&ks0, B_FALSE, &sidx, &nidx);
+	if (ksr == NULL)
+		return;
+
+	for (t = 0; table[t] != NULL; t++) {
+		for (i = 0; i < nidx; i++) {
+			if (strncmp(ksr[i].ks_class, "mib2", KSTAT_STRLEN) != 0)
+				continue;
+			if (strncmp(ksr[i].ks_name, table[t]->lst_proto,
+			    KSTAT_STRLEN) == 0) {
+				lxpr_kstat_print_tab(uiobuf, table[t], &ksr[i]);
+				break;
+			}
+		}
+	}
+	kmem_free(ksr, sidx);
+}
+
+/* ARGSUSED */
+static void
+lxpr_read_net_stat(lxpr_node_t *lxpnp, lxpr_uiobuf_t *uiobuf)
+{
+}
+
+static int
+lxpr_convert_tcp_state(int st)
+{
+	/*
+	 * Derived from the enum located in the Linux kernel sources:
+	 * include/net/tcp_states.h
+	 */
+	switch (st) {
+	case TCPS_ESTABLISHED:
+		return (1);
+	case TCPS_SYN_SENT:
+		return (2);
+	case TCPS_SYN_RCVD:
+		return (3);
+	case TCPS_FIN_WAIT_1:
+		return (4);
+	case TCPS_FIN_WAIT_2:
+		return (5);
+	case TCPS_TIME_WAIT:
+		return (6);
+	case TCPS_CLOSED:
+		return (7);
+	case TCPS_CLOSE_WAIT:
+		return (8);
+	case TCPS_LAST_ACK:
+		return (9);
+	case TCPS_LISTEN:
+		return (10);
+	case TCPS_CLOSING:
+		return (11);
+	default:
+		/* No translation for TCPS_IDLE, TCPS_BOUND or anything else */
+		return (0);
+	}
+}
+
+static void
+lxpr_format_tcp(lxpr_uiobuf_t *uiobuf, ushort_t ipver)
+{
+	int i, sl = 0;
+	connf_t *connfp;
+	conn_t *connp;
+	netstack_t *ns;
+	ip_stack_t *ipst;
+
+	ASSERT(ipver == IPV4_VERSION || ipver == IPV6_VERSION);
+	if (ipver == IPV4_VERSION) {
+		lxpr_uiobuf_printf(uiobuf, "  sl  local_address rem_address   "
+		    "st tx_queue rx_queue tr tm->when retrnsmt   uid  timeout "
+		    "inode\n");
+	} else {
+		lxpr_uiobuf_printf(uiobuf, "  sl  "
+		    "local_address                         "
+		    "remote_address                        "
+		    "st tx_queue rx_queue tr tm->when retrnsmt   "
+		    "uid  timeout inode\n");
+	}
+	/*
+	 * Due to differences between the Linux and illumos TCP
+	 * implementations, some data will be omitted from the output here.
+	 *
+	 * Valid fields:
+	 *  - local_address
+	 *  - remote_address
+	 *  - st
+	 *  - tx_queue
+	 *  - rx_queue
+	 *  - uid
+	 *  - inode
+	 *
+	 * Omitted/invalid fields
+	 *  - tr
+	 *  - tm->when
+	 *  - retrnsmt
+	 *  - timeout
+	 */
+
+	ns = netstack_get_current();
+	if (ns == NULL)
+		return;
+	ipst = ns->netstack_ip;
+
+	for (i = 0; i < CONN_G_HASH_SIZE; i++) {
+		connfp = &ipst->ips_ipcl_globalhash_fanout[i];
+		connp = NULL;
+		while ((connp =
+		    ipcl_get_next_conn(connfp, connp, IPCL_TCPCONN)) != NULL) {
+			tcp_t *tcp;
+			vattr_t attr;
+			sonode_t *so = (sonode_t *)connp->conn_upper_handle;
+			vnode_t *vp = (so != NULL) ? so->so_vnode : NULL;
+			if (connp->conn_ipversion != ipver)
+				continue;
+			tcp = connp->conn_tcp;
+			if (ipver == IPV4_VERSION) {
+				lxpr_uiobuf_printf(uiobuf,
+				    "%4d: %08X:%04X %08X:%04X ",
+				    ++sl,
+				    connp->conn_laddr_v4,
+				    ntohs(connp->conn_lport),
+				    connp->conn_faddr_v4,
+				    ntohs(connp->conn_fport));
+			} else {
+				lxpr_uiobuf_printf(uiobuf, "%4d: "
+				    "%08X%08X%08X%08X:%04X "
+				    "%08X%08X%08X%08X:%04X ",
+				    ++sl,
+				    connp->conn_laddr_v6.s6_addr32[0],
+				    connp->conn_laddr_v6.s6_addr32[1],
+				    connp->conn_laddr_v6.s6_addr32[2],
+				    connp->conn_laddr_v6.s6_addr32[3],
+				    ntohs(connp->conn_lport),
+				    connp->conn_faddr_v6.s6_addr32[0],
+				    connp->conn_faddr_v6.s6_addr32[1],
+				    connp->conn_faddr_v6.s6_addr32[2],
+				    connp->conn_faddr_v6.s6_addr32[3],
+				    ntohs(connp->conn_fport));
+			}
+
+			/* fetch the simulated inode for the socket */
+			if (vp == NULL ||
+			    VOP_GETATTR(vp, &attr, 0, CRED(), NULL) != 0)
+				attr.va_nodeid = 0;
+
+			lxpr_uiobuf_printf(uiobuf,
+			    "%02X %08X:%08X %02X:%08X %08X "
+			    "%5u %8d %lu %d %p %u %u %u %u %d\n",
+			    lxpr_convert_tcp_state(tcp->tcp_state),
+			    tcp->tcp_rcv_cnt, tcp->tcp_unsent, /* rx/tx queue */
+			    0, 0, /* tr, when */
+			    0, /* per-connection rexmits aren't tracked today */
+			    connp->conn_cred->cr_uid,
+			    0, /* timeout */
+			    /* inode + more */
+			    (ino_t)attr.va_nodeid, 0, NULL, 0, 0, 0, 0, 0);
+		}
+	}
+	netstack_rele(ns);
+}
+
+/* ARGSUSED */
+static void
+lxpr_read_net_tcp(lxpr_node_t *lxpnp, lxpr_uiobuf_t *uiobuf)
+{
+	lxpr_format_tcp(uiobuf, IPV4_VERSION);
+}
+
+/* ARGSUSED */
+static void
+lxpr_read_net_tcp6(lxpr_node_t *lxpnp, lxpr_uiobuf_t *uiobuf)
+{
+	lxpr_format_tcp(uiobuf, IPV6_VERSION);
+}
+
+static void
+lxpr_format_udp(lxpr_uiobuf_t *uiobuf, ushort_t ipver)
+{
+	int i, sl = 0;
+	connf_t *connfp;
+	conn_t *connp;
+	netstack_t *ns;
+	ip_stack_t *ipst;
+
+	ASSERT(ipver == IPV4_VERSION || ipver == IPV6_VERSION);
+	if (ipver == IPV4_VERSION) {
+		lxpr_uiobuf_printf(uiobuf, "  sl  local_address rem_address"
+		    "   st tx_queue rx_queue tr tm->when retrnsmt   uid"
+		    "  timeout inode ref pointer drops\n");
+	} else {
+		lxpr_uiobuf_printf(uiobuf, "  sl  "
+		    "local_address                         "
+		    "remote_address                        "
+		    "st tx_queue rx_queue tr tm->when retrnsmt   "
+		    "uid  timeout inode ref pointer drops\n");
+	}
+	/*
+	 * Due to differences between the Linux and illumos UDP
+	 * implementations, some data will be omitted from the output here.
+	 *
+	 * Valid fields:
+	 *  - local_address
+	 *  - remote_address
+	 *  - st: limited
+	 *  - uid
+	 *
+	 * Omitted/invalid fields
+	 *  - tx_queue
+	 *  - rx_queue
+	 *  - tr
+	 *  - tm->when
+	 *  - retrnsmt
+	 *  - timeout
+	 *  - inode
+	 */
+
+	ns = netstack_get_current();
+	if (ns == NULL)
+		return;
+	ipst = ns->netstack_ip;
+
+	for (i = 0; i < CONN_G_HASH_SIZE; i++) {
+		connfp = &ipst->ips_ipcl_globalhash_fanout[i];
+		connp = NULL;
+		while ((connp =
+		    ipcl_get_next_conn(connfp, connp, IPCL_UDPCONN)) != NULL) {
+			udp_t *udp;
+			int state = 0;
+			vattr_t attr;
+			sonode_t *so = (sonode_t *)connp->conn_upper_handle;
+			vnode_t *vp = (so != NULL) ? so->so_vnode : NULL;
+			if (connp->conn_ipversion != ipver)
+				continue;
+			udp = connp->conn_udp;
+			if (ipver == IPV4_VERSION) {
+				lxpr_uiobuf_printf(uiobuf,
+				    "%4d: %08X:%04X %08X:%04X ",
+				    ++sl,
+				    connp->conn_laddr_v4,
+				    ntohs(connp->conn_lport),
+				    connp->conn_faddr_v4,
+				    ntohs(connp->conn_fport));
+			} else {
+				lxpr_uiobuf_printf(uiobuf, "%4d: "
+				    "%08X%08X%08X%08X:%04X "
+				    "%08X%08X%08X%08X:%04X ",
+				    ++sl,
+				    connp->conn_laddr_v6.s6_addr32[0],
+				    connp->conn_laddr_v6.s6_addr32[1],
+				    connp->conn_laddr_v6.s6_addr32[2],
+				    connp->conn_laddr_v6.s6_addr32[3],
+				    ntohs(connp->conn_lport),
+				    connp->conn_faddr_v6.s6_addr32[0],
+				    connp->conn_faddr_v6.s6_addr32[1],
+				    connp->conn_faddr_v6.s6_addr32[2],
+				    connp->conn_faddr_v6.s6_addr32[3],
+				    ntohs(connp->conn_fport));
+			}
+
+			switch (udp->udp_state) {
+			case TS_UNBND:
+			case TS_IDLE:
+				state = 7;
+				break;
+			case TS_DATA_XFER:
+				state = 1;
+				break;
+			}
+
+			/* fetch the simulated inode for the socket */
+			if (vp == NULL ||
+			    VOP_GETATTR(vp, &attr, 0, CRED(), NULL) != 0)
+				attr.va_nodeid = 0;
+
+			lxpr_uiobuf_printf(uiobuf,
+			    "%02X %08X:%08X %02X:%08X %08X "
+			    "%5u %8d %lu %d %p %d\n",
+			    state,
+			    0, 0, /* rx/tx queue */
+			    0, 0, /* tr, when */
+			    0, /* retrans */
+			    connp->conn_cred->cr_uid,
+			    0, /* timeout */
+			    /* inode, ref, pointer, drops */
+			    (ino_t)attr.va_nodeid, 0, NULL, 0);
+		}
+	}
+	netstack_rele(ns);
+}
+
+/* ARGSUSED */
+static void
+lxpr_read_net_udp(lxpr_node_t *lxpnp, lxpr_uiobuf_t *uiobuf)
+{
+	lxpr_format_udp(uiobuf, IPV4_VERSION);
+}
+
+/* ARGSUSED */
+static void
+lxpr_read_net_udp6(lxpr_node_t *lxpnp, lxpr_uiobuf_t *uiobuf)
+{
+	lxpr_format_udp(uiobuf, IPV6_VERSION);
+}
+
+/* ARGSUSED */
+static void
+lxpr_read_net_unix(lxpr_node_t *lxpnp, lxpr_uiobuf_t *uiobuf)
+{
+	sonode_t *so;
+	zoneid_t zoneid = getzoneid();
+
+	lxpr_uiobuf_printf(uiobuf, "Num       RefCount Protocol Flags    Type "
+	    "St Inode Path\n");
+
+	mutex_enter(&socklist.sl_lock);
+	for (so = socklist.sl_list; so != NULL;
+	    so = _SOTOTPI(so)->sti_next_so) {
+		vnode_t *vp = so->so_vnode;
+		vattr_t attr;
+		sotpi_info_t *sti;
+		const char *name = NULL;
+		int status = 0;
+		int type = 0;
+		int flags = 0;
+
+		/* Only process active sonodes in this zone */
+		if (so->so_count == 0 || so->so_zoneid != zoneid)
+			continue;
+
+		/*
+		 * Grab the inode, if possible.
+		 * This must be done before entering so_lock.
+		 */
+		if (vp == NULL ||
+		    VOP_GETATTR(vp, &attr, 0, CRED(), NULL) != 0)
+			attr.va_nodeid = 0;
+
+		mutex_enter(&so->so_lock);
+		sti = _SOTOTPI(so);
+
+		if (sti->sti_laddr_sa != NULL &&
+		    sti->sti_laddr_len > 0) {
+			name = sti->sti_laddr_sa->sa_data;
+		} else if (sti->sti_faddr_sa != NULL &&
+		    sti->sti_faddr_len > 0) {
+			name = sti->sti_faddr_sa->sa_data;
+		}
+
+		/*
+		 * Derived from enum values in Linux kernel source:
+		 * include/uapi/linux/net.h
+		 */
+		if ((so->so_state & SS_ISDISCONNECTING) != 0) {
+			status = 4;
+		} else if ((so->so_state & SS_ISCONNECTING) != 0) {
+			status = 2;
+		} else if ((so->so_state & SS_ISCONNECTED) != 0) {
+			status = 3;
+		} else {
+			status = 1;
+			/* Add ACC flag for stream-type server sockets */
+			if (so->so_type != SOCK_DGRAM &&
+			    sti->sti_laddr_sa != NULL)
+				flags |= 0x10000;
+		}
+
+		/* Convert to Linux type */
+		switch (so->so_type) {
+		case SOCK_DGRAM:
+			type = 2;
+			break;
+		case SOCK_SEQPACKET:
+			type = 5;
+			break;
+		default:
+			type = 1;
+		}
+
+		lxpr_uiobuf_printf(uiobuf, "%p: %08X %08X %08X %04X %02X %5llu",
+		    so,
+		    so->so_count,
+		    0, /* proto, always 0 */
+		    flags,
+		    type,
+		    status,
+		    (ino_t)attr.va_nodeid);
+
+		/*
+		 * Due to shortcomings in the abstract socket emulation, they
+		 * cannot be properly represented here (as @<path>).
+		 *
+		 * This will be the case until they are better implemented.
+		 */
+		if (name != NULL)
+			lxpr_uiobuf_printf(uiobuf, " %s\n", name);
+		else
+			lxpr_uiobuf_printf(uiobuf, "\n");
+		mutex_exit(&so->so_lock);
+	}
+	mutex_exit(&socklist.sl_lock);
+}
+
+/*
+ * lxpr_read_kmsg(): read the contents of the kernel message queue. We
+ * translate this into the reception of console messages for this zone; each
+ * read copies out a single zone console message, or blocks until the next one
+ * is produced, unless we're open non-blocking, in which case we return after
+ * 1ms.
+ */
+
+#define	LX_KMSG_PRI	"<0>"
+
+static void
+lxpr_read_kmsg(lxpr_node_t *lxpnp, struct lxpr_uiobuf *uiobuf, ldi_handle_t lh)
+{
+	mblk_t		*mp;
+	timestruc_t	to;
+	timestruc_t	*tp = NULL;
+
+	ASSERT(lxpnp->lxpr_type == LXPR_KMSG);
+
+	if (lxpr_uiobuf_nonblock(uiobuf)) {
+		to.tv_sec = 0;
+		to.tv_nsec = 1000000; /* 1msec */
+		tp = &to;
+	}
+
+	if (ldi_getmsg(lh, &mp, tp) == 0) {
+		/*
+		 * lx procfs doesn't like successive reads to the same file
+		 * descriptor unless we do an explicit rewind each time.
+		 */
+		lxpr_uiobuf_seek(uiobuf, 0);
+
+		lxpr_uiobuf_printf(uiobuf, "%s%s", LX_KMSG_PRI,
+		    mp->b_cont->b_rptr);
+
+		freemsg(mp);
+	}
+}
+
+/*
+ * lxpr_read_loadavg(): read the contents of the "loadavg" file.  We do just
+ * enough for uptime and other simple lxproc readers to work
+ */
+extern int nthread;
+
+static void
+lxpr_read_loadavg(lxpr_node_t *lxpnp, lxpr_uiobuf_t *uiobuf)
+{
+	ulong_t avenrun1;
+	ulong_t avenrun5;
+	ulong_t avenrun15;
+	ulong_t avenrun1_cs;
+	ulong_t avenrun5_cs;
+	ulong_t avenrun15_cs;
+	int loadavg[3];
+	int *loadbuf;
+	cpupart_t *cp;
+	zone_t *zone = LXPTOZ(lxpnp);
+
+	uint_t nrunnable = 0;
+	rctl_qty_t nlwps;
+
+	ASSERT(lxpnp->lxpr_type == LXPR_LOADAVG);
+
+	mutex_enter(&cpu_lock);
+
+	/*
+	 * Need to add up values over all CPU partitions. If pools are active,
+	 * only report the values of the zone's partition, which by definition
+	 * includes the current CPU.
+	 */
+	if (pool_pset_enabled()) {
+		psetid_t psetid = zone_pset_get(curproc->p_zone);
+
+		ASSERT(curproc->p_zone != &zone0);
+		cp = CPU->cpu_part;
+
+		nrunnable = cp->cp_nrunning + cp->cp_nrunnable;
+		(void) cpupart_get_loadavg(psetid, &loadavg[0], 3);
+		loadbuf = &loadavg[0];
+	} else {
+		cp = cp_list_head;
+		do {
+			nrunnable += cp->cp_nrunning + cp->cp_nrunnable;
+		} while ((cp = cp->cp_next) != cp_list_head);
+
+		loadbuf = zone == global_zone ?
+		    &avenrun[0] : zone->zone_avenrun;
+	}
+
+	/*
+	 * If we're in the non-global zone, we'll report the total number of
+	 * LWPs in the zone for the "nproc" parameter of /proc/loadavg,
+	 * otherwise will just use nthread (which will include kernel threads,
+	 * but should be good enough for lxproc).
+	 */
+	nlwps = zone == global_zone ? nthread : zone->zone_nlwps;
+
+	mutex_exit(&cpu_lock);
+
+	avenrun1 = loadbuf[0] >> FSHIFT;
+	avenrun1_cs = ((loadbuf[0] & (FSCALE-1)) * 100) >> FSHIFT;
+	avenrun5 = loadbuf[1] >> FSHIFT;
+	avenrun5_cs = ((loadbuf[1] & (FSCALE-1)) * 100) >> FSHIFT;
+	avenrun15 = loadbuf[2] >> FSHIFT;
+	avenrun15_cs = ((loadbuf[2] & (FSCALE-1)) * 100) >> FSHIFT;
+
+	lxpr_uiobuf_printf(uiobuf,
+	    "%ld.%02d %ld.%02d %ld.%02d %d/%d %d\n",
+	    avenrun1, avenrun1_cs,
+	    avenrun5, avenrun5_cs,
+	    avenrun15, avenrun15_cs,
+	    nrunnable, nlwps, 0);
+}
+
+/*
+ * lxpr_read_meminfo(): read the contents of the "meminfo" file.
+ */
+static void
+lxpr_read_meminfo(lxpr_node_t *lxpnp, lxpr_uiobuf_t *uiobuf)
+{
+	zone_t *zone = LXPTOZ(lxpnp);
+	int global = zone == global_zone;
+	long total_mem, free_mem, total_swap, used_swap;
+
+	ASSERT(lxpnp->lxpr_type == LXPR_MEMINFO);
+
+	if (global || zone->zone_phys_mem_ctl == UINT64_MAX) {
+		total_mem = physmem * PAGESIZE;
+		free_mem = freemem * PAGESIZE;
+	} else {
+		total_mem = zone->zone_phys_mem_ctl;
+		free_mem = zone->zone_phys_mem_ctl - zone->zone_phys_mem;
+	}
+
+	if (global || zone->zone_max_swap_ctl == UINT64_MAX) {
+		total_swap = k_anoninfo.ani_max * PAGESIZE;
+		used_swap = k_anoninfo.ani_phys_resv * PAGESIZE;
+	} else {
+		mutex_enter(&zone->zone_mem_lock);
+		total_swap = zone->zone_max_swap_ctl;
+		used_swap = zone->zone_max_swap;
+		mutex_exit(&zone->zone_mem_lock);
+	}
+
+	lxpr_uiobuf_printf(uiobuf,
+	    "MemTotal:  %8lu kB\n"
+	    "MemFree:   %8lu kB\n"
+	    "MemShared: %8u kB\n"
+	    "Buffers:   %8u kB\n"
+	    "Cached:    %8u kB\n"
+	    "SwapCached:%8u kB\n"
+	    "Active:    %8u kB\n"
+	    "Inactive:  %8u kB\n"
+	    "HighTotal: %8u kB\n"
+	    "HighFree:  %8u kB\n"
+	    "LowTotal:  %8u kB\n"
+	    "LowFree:   %8u kB\n"
+	    "SwapTotal: %8lu kB\n"
+	    "SwapFree:  %8lu kB\n",
+	    btok(total_mem),				/* MemTotal */
+	    btok(free_mem),				/* MemFree */
+	    0,						/* MemShared */
+	    0,						/* Buffers */
+	    0,						/* Cached */
+	    0,						/* SwapCached */
+	    0,						/* Active */
+	    0,						/* Inactive */
+	    0,						/* HighTotal */
+	    0,						/* HighFree */
+	    btok(total_mem),				/* LowTotal */
+	    btok(free_mem),				/* LowFree */
+	    btok(total_swap),				/* SwapTotal */
+	    btok(total_swap - used_swap));		/* SwapFree */
+}
+
+/*
+ * lxpr_read_mounts():
+ */
+/* ARGSUSED */
+static void
+lxpr_read_mounts(lxpr_node_t *lxpnp, lxpr_uiobuf_t *uiobuf)
+{
+	struct vfs *vfsp;
+	struct vfs *vfslist;
+	zone_t *zone = LXPTOZ(lxpnp);
+	struct print_data {
+		refstr_t *vfs_mntpt;
+		refstr_t *vfs_resource;
+		uint_t vfs_flag;
+		int vfs_fstype;
+		struct print_data *next;
+	} *print_head = NULL;
+	struct print_data **print_tail = &print_head;
+	struct print_data *printp;
+
+	vfs_list_read_lock();
+
+	if (zone == global_zone) {
+		vfsp = vfslist = rootvfs;
+	} else {
+		vfsp = vfslist = zone->zone_vfslist;
+		/*
+		 * If the zone has a root entry, it will be the first in
+		 * the list.  If it doesn't, we conjure one up.
+		 */
+		if (vfslist == NULL || strcmp(refstr_value(vfsp->vfs_mntpt),
+		    zone->zone_rootpath) != 0) {
+			struct vfs *tvfsp;
+			/*
+			 * The root of the zone is not a mount point.  The vfs
+			 * we want to report is that of the zone's root vnode.
+			 */
+			tvfsp = zone->zone_rootvp->v_vfsp;
+
+			lxpr_uiobuf_printf(uiobuf,
+			    "/ / %s %s 0 0\n",
+			    vfssw[tvfsp->vfs_fstype].vsw_name,
+			    tvfsp->vfs_flag & VFS_RDONLY ? "ro" : "rw");
+
+		}
+		if (vfslist == NULL) {
+			vfs_list_unlock();
+			return;
+		}
+	}
+
+	/*
+	 * Later on we have to do a lookupname, which can end up causing
+	 * another vfs_list_read_lock() to be called. Which can lead to a
+	 * deadlock. To avoid this, we extract the data we need into a local
+	 * list, then we can run this list without holding vfs_list_read_lock()
+	 * We keep the list in the same order as the vfs_list
+	 */
+	do {
+		/* Skip mounts we shouldn't show */
+		if (vfsp->vfs_flag & VFS_NOMNTTAB) {
+			goto nextfs;
+		}
+
+		printp = kmem_alloc(sizeof (*printp), KM_SLEEP);
+		refstr_hold(vfsp->vfs_mntpt);
+		printp->vfs_mntpt = vfsp->vfs_mntpt;
+		refstr_hold(vfsp->vfs_resource);
+		printp->vfs_resource = vfsp->vfs_resource;
+		printp->vfs_flag = vfsp->vfs_flag;
+		printp->vfs_fstype = vfsp->vfs_fstype;
+		printp->next = NULL;
+
+		*print_tail = printp;
+		print_tail = &printp->next;
+
+nextfs:
+		vfsp = (zone == global_zone) ?
+		    vfsp->vfs_next : vfsp->vfs_zone_next;
+
+	} while (vfsp != vfslist);
+
+	vfs_list_unlock();
+
+	/*
+	 * now we can run through what we've extracted without holding
+	 * vfs_list_read_lock()
+	 */
+	printp = print_head;
+	while (printp != NULL) {
+		struct print_data *printp_next;
+		const char *resource;
+		char *mntpt;
+		struct vnode *vp;
+		int error;
+
+		mntpt = (char *)refstr_value(printp->vfs_mntpt);
+		resource = refstr_value(printp->vfs_resource);
+
+		if (mntpt != NULL && mntpt[0] != '\0')
+			mntpt = ZONE_PATH_TRANSLATE(mntpt, zone);
+		else
+			mntpt = "-";
+
+		error = lookupname(mntpt, UIO_SYSSPACE, FOLLOW, NULLVPP, &vp);
+
+		if (error != 0)
+			goto nextp;
+
+		if (!(vp->v_flag & VROOT)) {
+			VN_RELE(vp);
+			goto nextp;
+		}
+		VN_RELE(vp);
+
+		if (resource != NULL && resource[0] != '\0') {
+			if (resource[0] == '/') {
+				resource = ZONE_PATH_VISIBLE(resource, zone) ?
+				    ZONE_PATH_TRANSLATE(resource, zone) :
+				    mntpt;
+			}
+		} else {
+			resource = "-";
+		}
+
+		lxpr_uiobuf_printf(uiobuf,
+		    "%s %s %s %s 0 0\n",
+		    resource, mntpt, vfssw[printp->vfs_fstype].vsw_name,
+		    printp->vfs_flag & VFS_RDONLY ? "ro" : "rw");
+
+nextp:
+		printp_next = printp->next;
+		refstr_rele(printp->vfs_mntpt);
+		refstr_rele(printp->vfs_resource);
+		kmem_free(printp, sizeof (*printp));
+		printp = printp_next;
+
+	}
+}
+
+/*
+ * lxpr_read_partitions():
+ *
+ * Over the years, /proc/partitions has been made considerably smaller -- to
+ * the point that it really is only major number, minor number, number of
+ * blocks (which we report as 0), and partition name.  We support this only
+ * because some things want to see it to make sense of /proc/diskstats.
+ */
+/* ARGSUSED */
+static void
+lxpr_read_partitions(lxpr_node_t *lxpnp, lxpr_uiobuf_t *uiobuf)
+{
+
+	kstat_t *ksr;
+	kstat_t ks0;
+	int nidx, num, i;
+	size_t sidx, size;
+
+	ASSERT(lxpnp->lxpr_type == LXPR_PARTITIONS);
+
+	ks0.ks_kid = 0;
+	ksr = (kstat_t *)lxpr_kstat_read(&ks0, B_FALSE, &sidx, &nidx);
+
+	if (ksr == NULL)
+		return;
+
+	lxpr_uiobuf_printf(uiobuf, "major minor  #blocks  name\n\n");
+
+	for (i = 1; i < nidx; i++) {
+		kstat_t *ksp = &ksr[i];
+		kstat_io_t *kip;
+
+		if (ksp->ks_type != KSTAT_TYPE_IO ||
+		    strcmp(ksp->ks_class, "disk") != 0)
+			continue;
+
+		if ((kip = (kstat_io_t *)lxpr_kstat_read(ksp, B_TRUE,
+		    &size, &num)) == NULL)
+			continue;
+
+		if (size < sizeof (kstat_io_t)) {
+			kmem_free(kip, size);
+			continue;
+		}
+
+		lxpr_uiobuf_printf(uiobuf, "%4d %7d %10d %s\n",
+		    mod_name_to_major(ksp->ks_module),
+		    ksp->ks_instance, 0, ksp->ks_name);
+
+		kmem_free(kip, size);
+	}
+
+	kmem_free(ksr, sidx);
+}
+
+/*
+ * lxpr_read_diskstats():
+ *
+ * See the block comment above the per-device output-generating line for the
+ * details of the format.
+ */
+/* ARGSUSED */
+static void
+lxpr_read_diskstats(lxpr_node_t *lxpnp, lxpr_uiobuf_t *uiobuf)
+{
+	kstat_t *ksr;
+	kstat_t ks0;
+	int nidx, num, i;
+	size_t sidx, size;
+
+	ASSERT(lxpnp->lxpr_type == LXPR_DISKSTATS);
+
+	ks0.ks_kid = 0;
+	ksr = (kstat_t *)lxpr_kstat_read(&ks0, B_FALSE, &sidx, &nidx);
+
+	if (ksr == NULL)
+		return;
+
+	for (i = 1; i < nidx; i++) {
+		kstat_t *ksp = &ksr[i];
+		kstat_io_t *kip;
+
+		if (ksp->ks_type != KSTAT_TYPE_IO ||
+		    strcmp(ksp->ks_class, "disk") != 0)
+			continue;
+
+		if ((kip = (kstat_io_t *)lxpr_kstat_read(ksp, B_TRUE,
+		    &size, &num)) == NULL)
+			continue;
+
+		if (size < sizeof (kstat_io_t)) {
+			kmem_free(kip, size);
+			continue;
+		}
+
+		/*
+		 * /proc/diskstats is defined to have one line of output for
+		 * each block device, with each line containing the following
+		 * 14 fields:
+		 *
+		 *	1 - major number
+		 *	2 - minor mumber
+		 *	3 - device name
+		 *	4 - reads completed successfully
+		 * 	5 - reads merged
+		 *	6 - sectors read
+		 *	7 - time spent reading (ms)
+		 *	8 - writes completed
+		 *	9 - writes merged
+		 *	10 - sectors written
+		 *	11 - time spent writing (ms)
+		 *	12 - I/Os currently in progress
+		 *	13 - time spent doing I/Os (ms)
+		 *	14 - weighted time spent doing I/Os (ms)
+		 *
+		 * One small hiccup:  we don't actually keep track of time
+		 * spent reading vs. time spent writing -- we keep track of
+		 * time waiting vs. time actually performing I/O.  While we
+		 * could divide the total time by the I/O mix (making the
+		 * obviously wrong assumption that I/O operations all take the
+		 * same amount of time), this has the undesirable side-effect
+		 * of moving backwards.  Instead, we report the total time
+		 * (read + write) for all three stats (read, write, total).
+		 * This is also a lie of sorts, but it should be more
+		 * immediately clear to the user that reads and writes are
+		 * each being double-counted as the other.
+		 */
+		lxpr_uiobuf_printf(uiobuf, "%4d %7d %s "
+		    "%llu %llu %llu %llu "
+		    "%llu %llu %llu %llu "
+		    "%llu %llu %llu\n",
+		    mod_name_to_major(ksp->ks_module),
+		    ksp->ks_instance, ksp->ks_name,
+		    (uint64_t)kip->reads, 0LL,
+		    kip->nread / (uint64_t)LXPR_SECTOR_SIZE,
+		    (kip->rtime + kip->wtime) / (uint64_t)(NANOSEC / MILLISEC),
+		    (uint64_t)kip->writes, 0LL,
+		    kip->nwritten / (uint64_t)LXPR_SECTOR_SIZE,
+		    (kip->rtime + kip->wtime) / (uint64_t)(NANOSEC / MILLISEC),
+		    (uint64_t)(kip->rcnt + kip->wcnt),
+		    (kip->rtime + kip->wtime) / (uint64_t)(NANOSEC / MILLISEC),
+		    (kip->rlentime + kip->wlentime) /
+		    (uint64_t)(NANOSEC / MILLISEC));
+
+		kmem_free(kip, size);
+	}
+
+	kmem_free(ksr, sidx);
+}
+
+/*
+ * lxpr_read_version(): read the contents of the "version" file.
+ */
+/* ARGSUSED */
+static void
+lxpr_read_version(lxpr_node_t *lxpnp, lxpr_uiobuf_t *uiobuf)
+{
+	char *vers;
+
+	vers = lx_get_zone_kern_version(LXPTOZ(lxpnp));
+
+	lxpr_uiobuf_printf(uiobuf,
+	    "%s version %s (%s version %d.%d.%d) "
+	    "#%s SMP %s\n",
+	    LX_UNAME_SYSNAME, vers,
+#if defined(__GNUC__)
+	    "gcc",
+	    __GNUC__,
+	    __GNUC_MINOR__,
+	    __GNUC_PATCHLEVEL__,
+#else
+	    "Sun C",
+	    __SUNPRO_C / 0x100,
+	    (__SUNPRO_C & 0xff) / 0x10,
+	    __SUNPRO_C & 0xf,
+#endif
+	    LX_UNAME_VERSION,
+	    "00:00:00 00/00/00");
+}
+
+/*
+ * lxpr_read_stat(): read the contents of the "stat" file.
+ *
+ */
+/* ARGSUSED */
+static void
+lxpr_read_stat(lxpr_node_t *lxpnp, lxpr_uiobuf_t *uiobuf)
+{
+	cpu_t *cp, *cpstart;
+	int pools_enabled;
+	ulong_t idle_cum = 0;
+	ulong_t sys_cum  = 0;
+	ulong_t user_cum = 0;
+	ulong_t irq_cum = 0;
+	ulong_t cpu_nrunnable_cum = 0;
+	ulong_t w_io_cum = 0;
+
+	ulong_t pgpgin_cum    = 0;
+	ulong_t pgpgout_cum   = 0;
+	ulong_t pgswapout_cum = 0;
+	ulong_t pgswapin_cum  = 0;
+	ulong_t intr_cum = 0;
+	ulong_t pswitch_cum = 0;
+	ulong_t forks_cum = 0;
+	hrtime_t msnsecs[NCMSTATES];
+	char *lx_kern_version = lx_get_zone_kern_version(LXPTOZ(lxpnp));
+	/* temporary variable since scalehrtime modifies data in place */
+	hrtime_t tmptime;
+
+	ASSERT(lxpnp->lxpr_type == LXPR_STAT);
+
+	mutex_enter(&cpu_lock);
+	pools_enabled = pool_pset_enabled();
+
+	/* Calculate cumulative stats */
+	cp = cpstart = CPU->cpu_part->cp_cpulist;
+	do {
+		int i;
+
+		/*
+		 * Don't count CPUs that aren't even in the system
+		 * or aren't up yet.
+		 */
+		if ((cp->cpu_flags & CPU_EXISTS) == 0) {
+			continue;
+		}
+
+		get_cpu_mstate(cp, msnsecs);
+
+		idle_cum += NSEC_TO_TICK(msnsecs[CMS_IDLE]);
+		sys_cum  += NSEC_TO_TICK(msnsecs[CMS_SYSTEM]);
+		user_cum += NSEC_TO_TICK(msnsecs[CMS_USER]);
+
+		pgpgin_cum += CPU_STATS(cp, vm.pgpgin);
+		pgpgout_cum += CPU_STATS(cp, vm.pgpgout);
+		pgswapin_cum += CPU_STATS(cp, vm.pgswapin);
+		pgswapout_cum += CPU_STATS(cp, vm.pgswapout);
+
+		if (strncmp(lx_kern_version, "2.4", 3) != 0) {
+			cpu_nrunnable_cum += cp->cpu_disp->disp_nrunnable;
+			w_io_cum += CPU_STATS(cp, sys.iowait);
+			for (i = 0; i < NCMSTATES; i++) {
+				tmptime = cp->cpu_intracct[i];
+				scalehrtime(&tmptime);
+				irq_cum += NSEC_TO_TICK(tmptime);
+			}
+		}
+
+		for (i = 0; i < PIL_MAX; i++)
+			intr_cum += CPU_STATS(cp, sys.intr[i]);
+
+		pswitch_cum += CPU_STATS(cp, sys.pswitch);
+		forks_cum += CPU_STATS(cp, sys.sysfork);
+		forks_cum += CPU_STATS(cp, sys.sysvfork);
+
+		if (pools_enabled)
+			cp = cp->cpu_next_part;
+		else
+			cp = cp->cpu_next;
+	} while (cp != cpstart);
+
+	if (strncmp(lx_kern_version, "2.4", 3) != 0) {
+		lxpr_uiobuf_printf(uiobuf,
+		    "cpu %lu %lu %lu %lu %lu %lu %lu\n",
+		    user_cum, 0L, sys_cum, idle_cum, 0L, irq_cum, 0L);
+	} else {
+		lxpr_uiobuf_printf(uiobuf,
+		    "cpu %lu %lu %lu %lu\n",
+		    user_cum, 0L, sys_cum, idle_cum);
+	}
+
+	/* Do per processor stats */
+	do {
+		int i;
+
+		ulong_t idle_ticks;
+		ulong_t sys_ticks;
+		ulong_t user_ticks;
+		ulong_t irq_ticks = 0;
+
+		/*
+		 * Don't count CPUs that aren't even in the system
+		 * or aren't up yet.
+		 */
+		if ((cp->cpu_flags & CPU_EXISTS) == 0) {
+			continue;
+		}
+
+		get_cpu_mstate(cp, msnsecs);
+
+		idle_ticks = NSEC_TO_TICK(msnsecs[CMS_IDLE]);
+		sys_ticks  = NSEC_TO_TICK(msnsecs[CMS_SYSTEM]);
+		user_ticks = NSEC_TO_TICK(msnsecs[CMS_USER]);
+
+		for (i = 0; i < NCMSTATES; i++) {
+			tmptime = cp->cpu_intracct[i];
+			scalehrtime(&tmptime);
+			irq_ticks += NSEC_TO_TICK(tmptime);
+		}
+
+		if (strncmp(lx_kern_version, "2.4", 3) != 0) {
+			lxpr_uiobuf_printf(uiobuf,
+			    "cpu%d %lu %lu %lu %lu %lu %lu %lu\n",
+			    cp->cpu_id, user_ticks, 0L, sys_ticks, idle_ticks,
+			    0L, irq_ticks, 0L);
+		} else {
+			lxpr_uiobuf_printf(uiobuf,
+			    "cpu%d %lu %lu %lu %lu\n",
+			    cp->cpu_id,
+			    user_ticks, 0L, sys_ticks, idle_ticks);
+		}
+
+		if (pools_enabled)
+			cp = cp->cpu_next_part;
+		else
+			cp = cp->cpu_next;
+	} while (cp != cpstart);
+
+	mutex_exit(&cpu_lock);
+
+	if (strncmp(lx_kern_version, "2.4", 3) != 0) {
+		lxpr_uiobuf_printf(uiobuf,
+		    "page %lu %lu\n"
+		    "swap %lu %lu\n"
+		    "intr %lu\n"
+		    "ctxt %lu\n"
+		    "btime %lu\n"
+		    "processes %lu\n"
+		    "procs_running %lu\n"
+		    "procs_blocked %lu\n",
+		    pgpgin_cum, pgpgout_cum,
+		    pgswapin_cum, pgswapout_cum,
+		    intr_cum,
+		    pswitch_cum,
+		    boot_time,
+		    forks_cum,
+		    cpu_nrunnable_cum,
+		    w_io_cum);
+	} else {
+		lxpr_uiobuf_printf(uiobuf,
+		    "page %lu %lu\n"
+		    "swap %lu %lu\n"
+		    "intr %lu\n"
+		    "ctxt %lu\n"
+		    "btime %lu\n"
+		    "processes %lu\n",
+		    pgpgin_cum, pgpgout_cum,
+		    pgswapin_cum, pgswapout_cum,
+		    intr_cum,
+		    pswitch_cum,
+		    boot_time,
+		    forks_cum);
+	}
+}
+
+/*
+ * lxpr_read_swaps():
+ *
+ * We don't support swap files or partitions, so just provide a dummy file with
+ * the necessary header.
+ */
+/* ARGSUSED */
+static void
+lxpr_read_swaps(lxpr_node_t *lxpnp, lxpr_uiobuf_t *uiobuf)
+{
+	lxpr_uiobuf_printf(uiobuf,
+	    "Filename                                "
+	    "Type            Size    Used    Priority\n");
+}
+
+/*
+ * inotify tunables exported via /proc.
+ */
+extern int inotify_maxevents;
+extern int inotify_maxinstances;
+extern int inotify_maxwatches;
+
+static void
+lxpr_read_sys_fs_inotify_max_queued_events(lxpr_node_t *lxpnp,
+    lxpr_uiobuf_t *uiobuf)
+{
+	ASSERT(lxpnp->lxpr_type == LXPR_SYS_FS_INOTIFY_MAX_QUEUED_EVENTS);
+	lxpr_uiobuf_printf(uiobuf, "%d\n", inotify_maxevents);
+}
+
+static void
+lxpr_read_sys_fs_inotify_max_user_instances(lxpr_node_t *lxpnp,
+    lxpr_uiobuf_t *uiobuf)
+{
+	ASSERT(lxpnp->lxpr_type == LXPR_SYS_FS_INOTIFY_MAX_USER_INSTANCES);
+	lxpr_uiobuf_printf(uiobuf, "%d\n", inotify_maxinstances);
+}
+
+static void
+lxpr_read_sys_fs_inotify_max_user_watches(lxpr_node_t *lxpnp,
+    lxpr_uiobuf_t *uiobuf)
+{
+	ASSERT(lxpnp->lxpr_type == LXPR_SYS_FS_INOTIFY_MAX_USER_WATCHES);
+	lxpr_uiobuf_printf(uiobuf, "%d\n", inotify_maxwatches);
+}
+
+static void
+lxpr_read_sys_kernel_hostname(lxpr_node_t *lxpnp, lxpr_uiobuf_t *uiobuf)
+{
+	ASSERT(lxpnp->lxpr_type == LXPR_SYS_KERNEL_HOSTNAME);
+	lxpr_uiobuf_printf(uiobuf, "%s\n", uts_nodename());
+}
+
+static void
+lxpr_read_sys_kernel_msgmni(lxpr_node_t *lxpnp, lxpr_uiobuf_t *uiobuf)
+{
+	rctl_qty_t val;
+
+	ASSERT(lxpnp->lxpr_type == LXPR_SYS_KERNEL_MSGMNI);
+
+	mutex_enter(&curproc->p_lock);
+	val = rctl_enforced_value(rc_zone_msgmni,
+	    curproc->p_zone->zone_rctls, curproc);
+	mutex_exit(&curproc->p_lock);
+
+	lxpr_uiobuf_printf(uiobuf, "%u\n", (uint_t)val);
+}
+
+static void
+lxpr_read_sys_kernel_ngroups_max(lxpr_node_t *lxpnp, lxpr_uiobuf_t *uiobuf)
+{
+	ASSERT(lxpnp->lxpr_type == LXPR_SYS_KERNEL_NGROUPS_MAX);
+	lxpr_uiobuf_printf(uiobuf, "%d\n", ngroups_max);
+}
+
+static void
+lxpr_read_sys_kernel_pid_max(lxpr_node_t *lxpnp, lxpr_uiobuf_t *uiobuf)
+{
+	ASSERT(lxpnp->lxpr_type == LXPR_SYS_KERNEL_PID_MAX);
+	lxpr_uiobuf_printf(uiobuf, "%d\n", maxpid);
+}
+
+static void
+lxpr_read_sys_kernel_shmmax(lxpr_node_t *lxpnp, lxpr_uiobuf_t *uiobuf)
+{
+	rctl_qty_t val;
+
+	ASSERT(lxpnp->lxpr_type == LXPR_SYS_KERNEL_SHMMAX);
+
+	mutex_enter(&curproc->p_lock);
+	val = rctl_enforced_value(rc_zone_shmmax,
+	    curproc->p_zone->zone_rctls, curproc);
+	mutex_exit(&curproc->p_lock);
+
+	if (val > FOURGB)
+		val = FOURGB;
+
+	lxpr_uiobuf_printf(uiobuf, "%u\n", (uint_t)val);
+}
+
+static void
+lxpr_read_sys_kernel_threads_max(lxpr_node_t *lxpnp, lxpr_uiobuf_t *uiobuf)
+{
+	ASSERT(lxpnp->lxpr_type == LXPR_SYS_KERNEL_THREADS_MAX);
+	lxpr_uiobuf_printf(uiobuf, "%d\n", curproc->p_zone->zone_nlwps_ctl);
+}
+
+/*
+ * lxpr_read_uptime(): read the contents of the "uptime" file.
+ *
+ * format is: "%.2lf, %.2lf",uptime_secs, idle_secs
+ * Use fixed point arithmetic to get 2 decimal places
+ */
+/* ARGSUSED */
+static void
+lxpr_read_uptime(lxpr_node_t *lxpnp, lxpr_uiobuf_t *uiobuf)
+{
+	cpu_t *cp, *cpstart;
+	int pools_enabled;
+	ulong_t idle_cum = 0;
+	ulong_t cpu_count = 0;
+	ulong_t idle_s;
+	ulong_t idle_cs;
+	ulong_t up_s;
+	ulong_t up_cs;
+	hrtime_t birthtime;
+	hrtime_t centi_sec = 10000000;  /* 10^7 */
+
+	ASSERT(lxpnp->lxpr_type == LXPR_UPTIME);
+
+	/* Calculate cumulative stats */
+	mutex_enter(&cpu_lock);
+	pools_enabled = pool_pset_enabled();
+
+	cp = cpstart = CPU;
+	do {
+		/*
+		 * Don't count CPUs that aren't even in the system
+		 * or aren't up yet.
+		 */
+		if ((cp->cpu_flags & CPU_EXISTS) == 0) {
+			continue;
+		}
+
+		idle_cum += CPU_STATS(cp, sys.cpu_ticks_idle);
+		idle_cum += CPU_STATS(cp, sys.cpu_ticks_wait);
+		cpu_count += 1;
+
+		if (pools_enabled)
+			cp = cp->cpu_next_part;
+		else
+			cp = cp->cpu_next;
+	} while (cp != cpstart);
+	mutex_exit(&cpu_lock);
+
+	/* Getting the Zone zsched process startup time */
+	birthtime = LXPTOZ(lxpnp)->zone_zsched->p_mstart;
+	up_cs = (gethrtime() - birthtime) / centi_sec;
+	up_s = up_cs / 100;
+	up_cs %= 100;
+
+	ASSERT(cpu_count > 0);
+	idle_cum /= cpu_count;
+	idle_s = idle_cum / hz;
+	idle_cs = idle_cum % hz;
+	idle_cs *= 100;
+	idle_cs /= hz;
+
+	lxpr_uiobuf_printf(uiobuf,
+	    "%ld.%02d %ld.%02d\n", up_s, up_cs, idle_s, idle_cs);
+}
+
+static const char *amd_x_edx[] = {
+	NULL,	NULL,	NULL,	NULL,
+	NULL,	NULL,	NULL,	NULL,
+	NULL,	NULL,	NULL,	"syscall",
+	NULL,	NULL,	NULL,	NULL,
+	NULL,	NULL,	NULL,	"mp",
+	"nx",	NULL,	"mmxext", NULL,
+	NULL,	NULL,	NULL,	NULL,
+	NULL,	"lm",	"3dnowext", "3dnow"
+};
+
+static const char *amd_x_ecx[] = {
+	"lahf_lm", NULL, "svm", NULL,
+	"altmovcr8"
+};
+
+static const char *tm_x_edx[] = {
+	"recovery", "longrun", NULL, "lrti"
+};
+
+/*
+ * Intel calls no-execute "xd" in its docs, but Linux still reports it as "nx."
+ */
+static const char *intc_x_edx[] = {
+	NULL,	NULL,	NULL,	NULL,
+	NULL,	NULL,	NULL,	NULL,
+	NULL,	NULL,	NULL,	"syscall",
+	NULL,	NULL,	NULL,	NULL,
+	NULL,	NULL,	NULL,	NULL,
+	"nx",	NULL,	NULL,   NULL,
+	NULL,	NULL,	NULL,	NULL,
+	NULL,	"lm",   NULL,   NULL
+};
+
+static const char *intc_edx[] = {
+	"fpu",	"vme",	"de",	"pse",
+	"tsc",	"msr",	"pae",	"mce",
+	"cx8",	"apic",	 NULL,	"sep",
+	"mtrr",	"pge",	"mca",	"cmov",
+	"pat",	"pse36", "pn",	"clflush",
+	NULL,	"dts",	"acpi",	"mmx",
+	"fxsr",	"sse",	"sse2",	"ss",
+	"ht",	"tm",	"ia64",	"pbe"
+};
+
+/*
+ * "sse3" on linux is called "pni" (Prescott New Instructions).
+ */
+static const char *intc_ecx[] = {
+	"pni",	NULL,	NULL, "monitor",
+	"ds_cpl", NULL,	NULL, "est",
+	"tm2",	NULL,	"cid", NULL,
+	NULL,	"cx16",	"xtpr"
+};
+
+static void
+lxpr_read_cpuinfo(lxpr_node_t *lxpnp, lxpr_uiobuf_t *uiobuf)
+{
+	int i;
+	uint32_t bits;
+	cpu_t *cp, *cpstart;
+	int pools_enabled;
+	const char **fp;
+	char brandstr[CPU_IDSTRLEN];
+	struct cpuid_regs cpr;
+	int maxeax;
+	int std_ecx, std_edx, ext_ecx, ext_edx;
+
+	ASSERT(lxpnp->lxpr_type == LXPR_CPUINFO);
+
+	mutex_enter(&cpu_lock);
+	pools_enabled = pool_pset_enabled();
+
+	cp = cpstart = CPU;
+	do {
+		/*
+		 * This returns the maximum eax value for standard cpuid
+		 * functions in eax.
+		 */
+		cpr.cp_eax = 0;
+		(void) cpuid_insn(cp, &cpr);
+		maxeax = cpr.cp_eax;
+
+		/*
+		 * Get standard x86 feature flags.
+		 */
+		cpr.cp_eax = 1;
+		(void) cpuid_insn(cp, &cpr);
+		std_ecx = cpr.cp_ecx;
+		std_edx = cpr.cp_edx;
+
+		/*
+		 * Now get extended feature flags.
+		 */
+		cpr.cp_eax = 0x80000001;
+		(void) cpuid_insn(cp, &cpr);
+		ext_ecx = cpr.cp_ecx;
+		ext_edx = cpr.cp_edx;
+
+		(void) cpuid_getbrandstr(cp, brandstr, CPU_IDSTRLEN);
+
+		lxpr_uiobuf_printf(uiobuf,
+		    "processor\t: %d\n"
+		    "vendor_id\t: %s\n"
+		    "cpu family\t: %d\n"
+		    "model\t\t: %d\n"
+		    "model name\t: %s\n"
+		    "stepping\t: %d\n"
+		    "cpu MHz\t\t: %u.%03u\n",
+		    cp->cpu_id, cpuid_getvendorstr(cp), cpuid_getfamily(cp),
+		    cpuid_getmodel(cp), brandstr, cpuid_getstep(cp),
+		    (uint32_t)(cpu_freq_hz / 1000000),
+		    ((uint32_t)(cpu_freq_hz / 1000)) % 1000);
+
+		lxpr_uiobuf_printf(uiobuf, "cache size\t: %u KB\n",
+		    getl2cacheinfo(cp, NULL, NULL, NULL) / 1024);
+
+		if (is_x86_feature(x86_featureset, X86FSET_HTT)) {
+			/*
+			 * 'siblings' is used for HT-style threads
+			 */
+			lxpr_uiobuf_printf(uiobuf,
+			    "physical id\t: %lu\n"
+			    "siblings\t: %u\n",
+			    pg_plat_hw_instance_id(cp, PGHW_CHIP),
+			    cpuid_get_ncpu_per_chip(cp));
+		}
+
+		/*
+		 * Since we're relatively picky about running on older hardware,
+		 * we can be somewhat cavalier about the answers to these ones.
+		 *
+		 * In fact, given the hardware we support, we just say:
+		 *
+		 *	fdiv_bug	: no	(if we're on a 64-bit kernel)
+		 *	hlt_bug		: no
+		 *	f00f_bug	: no
+		 *	coma_bug	: no
+		 *	wp		: yes	(write protect in supervsr mode)
+		 */
+		lxpr_uiobuf_printf(uiobuf,
+		    "fdiv_bug\t: %s\n"
+		    "hlt_bug \t: no\n"
+		    "f00f_bug\t: no\n"
+		    "coma_bug\t: no\n"
+		    "fpu\t\t: %s\n"
+		    "fpu_exception\t: %s\n"
+		    "cpuid level\t: %d\n"
+		    "flags\t\t:",
+#if defined(__i386)
+		    fpu_pentium_fdivbug ? "yes" : "no",
+#else
+		    "no",
+#endif /* __i386 */
+		    fpu_exists ? "yes" : "no", fpu_exists ? "yes" : "no",
+		    maxeax);
+
+		for (bits = std_edx, fp = intc_edx, i = 0;
+		    i < sizeof (intc_edx) / sizeof (intc_edx[0]); fp++, i++)
+			if ((bits & (1 << i)) != 0 && *fp)
+				lxpr_uiobuf_printf(uiobuf, " %s", *fp);
+
+		/*
+		 * name additional features where appropriate
+		 */
+		switch (x86_vendor) {
+		case X86_VENDOR_Intel:
+			for (bits = ext_edx, fp = intc_x_edx, i = 0;
+			    i < sizeof (intc_x_edx) / sizeof (intc_x_edx[0]);
+			    fp++, i++)
+				if ((bits & (1 << i)) != 0 && *fp)
+					lxpr_uiobuf_printf(uiobuf, " %s", *fp);
+			break;
+
+		case X86_VENDOR_AMD:
+			for (bits = ext_edx, fp = amd_x_edx, i = 0;
+			    i < sizeof (amd_x_edx) / sizeof (amd_x_edx[0]);
+			    fp++, i++)
+				if ((bits & (1 << i)) != 0 && *fp)
+					lxpr_uiobuf_printf(uiobuf, " %s", *fp);
+
+			for (bits = ext_ecx, fp = amd_x_ecx, i = 0;
+			    i < sizeof (amd_x_ecx) / sizeof (amd_x_ecx[0]);
+			    fp++, i++)
+				if ((bits & (1 << i)) != 0 && *fp)
+					lxpr_uiobuf_printf(uiobuf, " %s", *fp);
+			break;
+
+		case X86_VENDOR_TM:
+			for (bits = ext_edx, fp = tm_x_edx, i = 0;
+			    i < sizeof (tm_x_edx) / sizeof (tm_x_edx[0]);
+			    fp++, i++)
+				if ((bits & (1 << i)) != 0 && *fp)
+					lxpr_uiobuf_printf(uiobuf, " %s", *fp);
+			break;
+		default:
+			break;
+		}
+
+		for (bits = std_ecx, fp = intc_ecx, i = 0;
+		    i < sizeof (intc_ecx) / sizeof (intc_ecx[0]); fp++, i++)
+			if ((bits & (1 << i)) != 0 && *fp)
+				lxpr_uiobuf_printf(uiobuf, " %s", *fp);
+
+		lxpr_uiobuf_printf(uiobuf, "\n\n");
+
+		if (pools_enabled)
+			cp = cp->cpu_next_part;
+		else
+			cp = cp->cpu_next;
+	} while (cp != cpstart);
+
+	mutex_exit(&cpu_lock);
+}
+
+/* ARGSUSED */
+static void
+lxpr_read_fd(lxpr_node_t *lxpnp, lxpr_uiobuf_t *uiobuf)
+{
+	ASSERT(lxpnp->lxpr_type == LXPR_PID_FD_FD);
+	lxpr_uiobuf_seterr(uiobuf, EFAULT);
+}
+
+/*
+ * lxpr_getattr(): Vnode operation for VOP_GETATTR()
+ */
+static int
+lxpr_getattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr,
+    caller_context_t *ct)
+{
+	register lxpr_node_t *lxpnp = VTOLXP(vp);
+	lxpr_nodetype_t type = lxpnp->lxpr_type;
+	extern uint_t nproc;
+	int error;
+
+	/*
+	 * Return attributes of underlying vnode if ATTR_REAL
+	 *
+	 * but keep fd files with the symlink permissions
+	 */
+	if (lxpnp->lxpr_realvp != NULL && (flags & ATTR_REAL)) {
+		vnode_t *rvp = lxpnp->lxpr_realvp;
+
+		/*
+		 * withold attribute information to owner or root
+		 */
+		if ((error = VOP_ACCESS(rvp, 0, 0, cr, ct)) != 0) {
+			return (error);
+		}
+
+		/*
+		 * now its attributes
+		 */
+		if ((error = VOP_GETATTR(rvp, vap, flags, cr, ct)) != 0) {
+			return (error);
+		}
+
+		/*
+		 * if it's a file in lx /proc/pid/fd/xx then set its
+		 * mode and keep it looking like a symlink, fifo or socket
+		 */
+		if (type == LXPR_PID_FD_FD) {
+			vap->va_mode = lxpnp->lxpr_mode;
+			vap->va_type = lxpnp->lxpr_realvp->v_type;
+			vap->va_size = 0;
+			vap->va_nlink = 1;
+		}
+		return (0);
+	}
+
+	/* Default attributes, that may be overridden below */
+	bzero(vap, sizeof (*vap));
+	vap->va_atime = vap->va_mtime = vap->va_ctime = lxpnp->lxpr_time;
+	vap->va_nlink = 1;
+	vap->va_type = vp->v_type;
+	vap->va_mode = lxpnp->lxpr_mode;
+	vap->va_fsid = vp->v_vfsp->vfs_dev;
+	vap->va_blksize = DEV_BSIZE;
+	vap->va_uid = lxpnp->lxpr_uid;
+	vap->va_gid = lxpnp->lxpr_gid;
+	vap->va_nodeid = lxpnp->lxpr_ino;
+
+	switch (type) {
+	case LXPR_PROCDIR:
+		vap->va_nlink = nproc + 2 + PROCDIRFILES;
+		vap->va_size = (nproc + 2 + PROCDIRFILES) * LXPR_SDSIZE;
+		break;
+	case LXPR_PIDDIR:
+		vap->va_nlink = PIDDIRFILES;
+		vap->va_size = PIDDIRFILES * LXPR_SDSIZE;
+		break;
+	case LXPR_PID_TASK_IDDIR:
+		vap->va_nlink = TIDDIRFILES;
+		vap->va_size = TIDDIRFILES * LXPR_SDSIZE;
+		break;
+	case LXPR_SELF:
+		vap->va_uid = crgetruid(curproc->p_cred);
+		vap->va_gid = crgetrgid(curproc->p_cred);
+		break;
+	case LXPR_PID_FD_FD:
+	case LXPR_PID_TID_FD_FD:
+		/*
+		 * Restore VLNK type for lstat-type activity.
+		 * See lxpr_readlink for more details.
+		 */
+		if ((flags & FOLLOW) == 0)
+			vap->va_type = VLNK;
+	default:
+		break;
+	}
+
+	vap->va_nblocks = (fsblkcnt64_t)btod(vap->va_size);
+	return (0);
+}
+
+/*
+ * lxpr_access(): Vnode operation for VOP_ACCESS()
+ */
+static int
+lxpr_access(vnode_t *vp, int mode, int flags, cred_t *cr, caller_context_t *ct)
+{
+	lxpr_node_t *lxpnp = VTOLXP(vp);
+	int shift = 0;
+	proc_t *tp;
+
+	/* lx /proc is a read only file system */
+	if (mode & VWRITE)
+		return (EROFS);
+
+	/*
+	 * If this is a restricted file, check access permissions.
+	 */
+	switch (lxpnp->lxpr_type) {
+	case LXPR_PIDDIR:
+		return (0);
+	case LXPR_PID_CURDIR:
+	case LXPR_PID_ENV:
+	case LXPR_PID_EXE:
+	case LXPR_PID_LIMITS:
+	case LXPR_PID_MAPS:
+	case LXPR_PID_MEM:
+	case LXPR_PID_ROOTDIR:
+	case LXPR_PID_FDDIR:
+	case LXPR_PID_FD_FD:
+	case LXPR_PID_TID_FDDIR:
+	case LXPR_PID_TID_FD_FD:
+		if ((tp = lxpr_lock(lxpnp->lxpr_pid)) == NULL)
+			return (ENOENT);
+		if (tp != curproc && secpolicy_proc_access(cr) != 0 &&
+		    priv_proc_cred_perm(cr, tp, NULL, mode) != 0) {
+			lxpr_unlock(tp);
+			return (EACCES);
+		}
+		lxpr_unlock(tp);
+	default:
+		break;
+	}
+
+	if (lxpnp->lxpr_realvp != NULL) {
+		/*
+		 * For these we use the underlying vnode's accessibility.
+		 */
+		return (VOP_ACCESS(lxpnp->lxpr_realvp, mode, flags, cr, ct));
+	}
+
+	/* If user is root allow access regardless of permission bits */
+	if (secpolicy_proc_access(cr) == 0)
+		return (0);
+
+	/*
+	 * Access check is based on only one of owner, group, public.  If not
+	 * owner, then check group.  If not a member of the group, then check
+	 * public access.
+	 */
+	if (crgetuid(cr) != lxpnp->lxpr_uid) {
+		shift += 3;
+		if (!groupmember((uid_t)lxpnp->lxpr_gid, cr))
+			shift += 3;
+	}
+
+	mode &= ~(lxpnp->lxpr_mode << shift);
+
+	if (mode == 0)
+		return (0);
+
+	return (EACCES);
+}
+
+/* ARGSUSED */
+static vnode_t *
+lxpr_lookup_not_a_dir(vnode_t *dp, char *comp)
+{
+	return (NULL);
+}
+
+/*
+ * lxpr_lookup(): Vnode operation for VOP_LOOKUP()
+ */
+/* ARGSUSED */
+static int
+lxpr_lookup(vnode_t *dp, char *comp, vnode_t **vpp, pathname_t *pathp,
+	int flags, vnode_t *rdir, cred_t *cr, caller_context_t *ct,
+	int *direntflags, pathname_t *realpnp)
+{
+	lxpr_node_t *lxpnp = VTOLXP(dp);
+	lxpr_nodetype_t type = lxpnp->lxpr_type;
+	int error;
+
+	ASSERT(dp->v_type == VDIR);
+	ASSERT(type < LXPR_NFILES);
+
+	/*
+	 * we should never get here because the lookup
+	 * is done on the realvp for these nodes
+	 */
+	ASSERT(type != LXPR_PID_FD_FD &&
+	    type != LXPR_PID_CURDIR &&
+	    type != LXPR_PID_ROOTDIR);
+
+	/*
+	 * restrict lookup permission to owner or root
+	 */
+	if ((error = lxpr_access(dp, VEXEC, 0, cr, ct)) != 0) {
+		return (error);
+	}
+
+	/*
+	 * Just return the parent vnode if that's where we are trying to go.
+	 */
+	if (strcmp(comp, "..") == 0) {
+		VN_HOLD(lxpnp->lxpr_parent);
+		*vpp = lxpnp->lxpr_parent;
+		return (0);
+	}
+
+	/*
+	 * Special handling for directory searches.  Note: null component name
+	 * denotes that the current directory is being searched.
+	 */
+	if ((dp->v_type == VDIR) && (*comp == '\0' || strcmp(comp, ".") == 0)) {
+		VN_HOLD(dp);
+		*vpp = dp;
+		return (0);
+	}
+
+	*vpp = (lxpr_lookup_function[type](dp, comp));
+	return ((*vpp == NULL) ? ENOENT : 0);
+}
+
+/*
+ * Do a sequential search on the given directory table
+ */
+static vnode_t *
+lxpr_lookup_common(vnode_t *dp, char *comp, proc_t *p,
+    lxpr_dirent_t *dirtab, int dirtablen)
+{
+	lxpr_node_t *lxpnp;
+	int count;
+
+	for (count = 0; count < dirtablen; count++) {
+		if (strcmp(dirtab[count].d_name, comp) == 0) {
+			lxpnp = lxpr_getnode(dp, dirtab[count].d_type, p, 0);
+			dp = LXPTOV(lxpnp);
+			ASSERT(dp != NULL);
+			return (dp);
+		}
+	}
+	return (NULL);
+}
+
+static vnode_t *
+lxpr_lookup_piddir(vnode_t *dp, char *comp)
+{
+	proc_t *p;
+
+	ASSERT(VTOLXP(dp)->lxpr_type == LXPR_PIDDIR);
+
+	p = lxpr_lock(VTOLXP(dp)->lxpr_pid);
+	if (p == NULL)
+		return (NULL);
+
+	dp = lxpr_lookup_common(dp, comp, p, piddir, PIDDIRFILES);
+
+	lxpr_unlock(p);
+
+	return (dp);
+}
+
+/*
+ * Lookup one of the process's task ID's.
+ */
+static vnode_t *
+lxpr_lookup_taskdir(vnode_t *dp, char *comp)
+{
+	lxpr_node_t *dlxpnp = VTOLXP(dp);
+	lxpr_node_t *lxpnp;
+	proc_t *p;
+	pid_t real_pid;
+	uint_t tid;
+	int c;
+	kthread_t *t;
+
+	ASSERT(dlxpnp->lxpr_type == LXPR_PID_TASKDIR);
+
+	/*
+	 * convert the string rendition of the filename to a thread ID
+	 */
+	tid = 0;
+	while ((c = *comp++) != '\0') {
+		int otid;
+		if (c < '0' || c > '9')
+			return (NULL);
+
+		otid = tid;
+		tid = 10 * tid + c - '0';
+		/* integer overflow */
+		if (tid / 10 != otid)
+			return (NULL);
+	}
+
+	/*
+	 * get the proc to work with and lock it
+	 */
+	real_pid = get_real_pid(dlxpnp->lxpr_pid);
+	p = lxpr_lock(real_pid);
+	if ((p == NULL))
+		return (NULL);
+
+	/*
+	 * If the process is a zombie or system process
+	 * it can't have any threads.
+	 */
+	if ((p->p_stat == SZOMB) || (p->p_flag & SSYS) || (p->p_as == &kas)) {
+		lxpr_unlock(p);
+		return (NULL);
+	}
+
+	t = lxpr_get_thread(p, tid);
+	if (t == NULL) {
+		lxpr_unlock(p);
+		return (NULL);
+	}
+	thread_unlock(t);
+
+	/*
+	 * Allocate and fill in a new lx /proc taskid node.
+	 * Instead of the last arg being a fd, it is a tid.
+	 */
+	lxpnp = lxpr_getnode(dp, LXPR_PID_TASK_IDDIR, p, tid);
+	dp = LXPTOV(lxpnp);
+	ASSERT(dp != NULL);
+	lxpr_unlock(p);
+	return (dp);
+}
+
+/*
+ * Lookup one of the process's task ID's.
+ */
+static vnode_t *
+lxpr_lookup_task_tid_dir(vnode_t *dp, char *comp)
+{
+	lxpr_node_t *dlxpnp = VTOLXP(dp);
+	lxpr_node_t *lxpnp;
+	proc_t *p;
+	pid_t real_pid;
+	kthread_t *t;
+	int i;
+
+	ASSERT(dlxpnp->lxpr_type == LXPR_PID_TASK_IDDIR);
+
+	/*
+	 * get the proc to work with and lock it
+	 */
+	real_pid = get_real_pid(dlxpnp->lxpr_pid);
+	p = lxpr_lock(real_pid);
+	if ((p == NULL))
+		return (NULL);
+
+	/*
+	 * If the process is a zombie or system process
+	 * it can't have any threads.
+	 */
+	if ((p->p_stat == SZOMB) || (p->p_flag & SSYS) || (p->p_as == &kas)) {
+		lxpr_unlock(p);
+		return (NULL);
+	}
+
+	/* need to confirm tid is still there */
+	t = lxpr_get_thread(p, dlxpnp->lxpr_desc);
+	if (t == NULL) {
+		lxpr_unlock(p);
+		return (NULL);
+	}
+	thread_unlock(t);
+
+	/*
+	 * allocate and fill in the new lx /proc taskid dir node
+	 */
+	for (i = 0; i < TIDDIRFILES; i++) {
+		if (strcmp(tiddir[i].d_name, comp) == 0) {
+			lxpnp = lxpr_getnode(dp, tiddir[i].d_type, p,
+			    dlxpnp->lxpr_desc);
+			dp = LXPTOV(lxpnp);
+			ASSERT(dp != NULL);
+			lxpr_unlock(p);
+			return (dp);
+		}
+	}
+
+	lxpr_unlock(p);
+	return (NULL);
+}
+
+/*
+ * Lookup one of the process's open files.
+ */
+static vnode_t *
+lxpr_lookup_fddir(vnode_t *dp, char *comp)
+{
+	lxpr_node_t *dlxpnp = VTOLXP(dp);
+	lxpr_node_t *lxpnp;
+	vnode_t *vp = NULL;
+	proc_t *p;
+	file_t *fp;
+	uint_t fd;
+	int c;
+	uf_entry_t *ufp;
+	uf_info_t *fip;
+
+	ASSERT(dlxpnp->lxpr_type == LXPR_PID_FDDIR ||
+	    dlxpnp->lxpr_type == LXPR_PID_TID_FDDIR);
+
+	/*
+	 * convert the string rendition of the filename
+	 * to a file descriptor
+	 */
+	fd = 0;
+	while ((c = *comp++) != '\0') {
+		int ofd;
+		if (c < '0' || c > '9')
+			return (NULL);
+
+		ofd = fd;
+		fd = 10*fd + c - '0';
+		/* integer overflow */
+		if (fd / 10 != ofd)
+			return (NULL);
+	}
+
+	/*
+	 * get the proc to work with and lock it
+	 */
+	p = lxpr_lock(dlxpnp->lxpr_pid);
+	if ((p == NULL))
+		return (NULL);
+
+	/*
+	 * If the process is a zombie or system process
+	 * it can't have any open files.
+	 */
+	if ((p->p_stat == SZOMB) || (p->p_flag & SSYS) || (p->p_as == &kas)) {
+		lxpr_unlock(p);
+		return (NULL);
+	}
+
+	/*
+	 * get us a fresh node/vnode
+	 */
+	lxpnp = lxpr_getnode(dp, LXPR_PID_FD_FD, p, fd);
+
+	/*
+	 * Drop p_lock, but keep the process P_PR_LOCK'd to prevent it from
+	 * going away while we dereference into fi_list.
+	 */
+	mutex_exit(&p->p_lock);
+
+	/*
+	 * get open file info
+	 */
+	fip = (&(p)->p_user.u_finfo);
+	mutex_enter(&fip->fi_lock);
+
+	if (fd < fip->fi_nfiles) {
+		UF_ENTER(ufp, fip, fd);
+		/*
+		 * ensure the fd is still kosher.
+		 * it may have gone between the readdir and
+		 * the lookup
+		 */
+		if (fip->fi_list[fd].uf_file == NULL) {
+			mutex_exit(&fip->fi_lock);
+			UF_EXIT(ufp);
+			mutex_enter(&p->p_lock);
+			lxpr_unlock(p);
+			lxpr_freenode(lxpnp);
+			return (NULL);
+		}
+
+		if ((fp = ufp->uf_file) != NULL)
+			vp = fp->f_vnode;
+		UF_EXIT(ufp);
+	}
+	mutex_exit(&fip->fi_lock);
+
+	if (vp == NULL) {
+		mutex_enter(&p->p_lock);
+		lxpr_unlock(p);
+		lxpr_freenode(lxpnp);
+		return (NULL);
+	} else {
+		/*
+		 * Fill in the lxpr_node so future references will be able to
+		 * find the underlying vnode. The vnode is held on the realvp.
+		 */
+		lxpnp->lxpr_realvp = vp;
+		VN_HOLD(lxpnp->lxpr_realvp);
+		/*
+		 * For certain entries (sockets, pipes, etc), Linux expects a
+		 * bogus-named symlink.  If that's the case, report the type as
+		 * VNON to bypass link-following elsewhere in the vfs system.
+		 *
+		 * See lxpr_readlink for more details.
+		 */
+		if (lxpr_readlink_pid_fd(lxpnp, NULL, 0) == 0)
+			LXPTOV(lxpnp)->v_type = VNON;
+	}
+
+	mutex_enter(&p->p_lock);
+	lxpr_unlock(p);
+	dp = LXPTOV(lxpnp);
+	ASSERT(dp != NULL);
+
+	return (dp);
+}
+
+static vnode_t *
+lxpr_lookup_netdir(vnode_t *dp, char *comp)
+{
+	ASSERT(VTOLXP(dp)->lxpr_type == LXPR_NETDIR);
+
+	dp = lxpr_lookup_common(dp, comp, NULL, netdir, NETDIRFILES);
+
+	return (dp);
+}
+
+static vnode_t *
+lxpr_lookup_procdir(vnode_t *dp, char *comp)
+{
+	ASSERT(VTOLXP(dp)->lxpr_type == LXPR_PROCDIR);
+
+	/*
+	 * We know all the names of files & dirs in our file system structure
+	 * except those that are pid names.  These change as pids are created/
+	 * deleted etc., so we just look for a number as the first char to see
+	 * if we are we doing pid lookups.
+	 *
+	 * Don't need to check for "self" as it is implemented as a symlink
+	 */
+	if (*comp >= '0' && *comp <= '9') {
+		pid_t pid = 0;
+		lxpr_node_t *lxpnp = NULL;
+		proc_t *p;
+		int c;
+
+		while ((c = *comp++) != '\0')
+			pid = 10 * pid + c - '0';
+
+		/*
+		 * Can't continue if the process is still loading or it doesn't
+		 * really exist yet (or maybe it just died!)
+		 */
+		p = lxpr_lock(pid);
+		if (p == NULL)
+			return (NULL);
+
+		if (secpolicy_basic_procinfo(CRED(), p, curproc) != 0) {
+			lxpr_unlock(p);
+			return (NULL);
+		}
+
+		/*
+		 * allocate and fill in a new lx /proc node
+		 */
+		lxpnp = lxpr_getnode(dp, LXPR_PIDDIR, p, 0);
+
+		lxpr_unlock(p);
+
+		dp = LXPTOV(lxpnp);
+		ASSERT(dp != NULL);
+
+		return (dp);
+	}
+
+	/* Lookup fixed names */
+	return (lxpr_lookup_common(dp, comp, NULL, lx_procdir, PROCDIRFILES));
+}
+
+static vnode_t *
+lxpr_lookup_sysdir(vnode_t *dp, char *comp)
+{
+	ASSERT(VTOLXP(dp)->lxpr_type == LXPR_SYSDIR);
+	return (lxpr_lookup_common(dp, comp, NULL, sysdir, SYSDIRFILES));
+}
+
+static vnode_t *
+lxpr_lookup_sys_kerneldir(vnode_t *dp, char *comp)
+{
+	ASSERT(VTOLXP(dp)->lxpr_type == LXPR_SYS_KERNELDIR);
+	return (lxpr_lookup_common(dp, comp, NULL, sys_kerneldir,
+	    SYS_KERNELDIRFILES));
+}
+
+static vnode_t *
+lxpr_lookup_sys_fsdir(vnode_t *dp, char *comp)
+{
+	ASSERT(VTOLXP(dp)->lxpr_type == LXPR_SYS_FSDIR);
+	return (lxpr_lookup_common(dp, comp, NULL, sys_fsdir,
+	    SYS_FSDIRFILES));
+}
+
+static vnode_t *
+lxpr_lookup_sys_fs_inotifydir(vnode_t *dp, char *comp)
+{
+	ASSERT(VTOLXP(dp)->lxpr_type == LXPR_SYS_FS_INOTIFYDIR);
+	return (lxpr_lookup_common(dp, comp, NULL, sys_fs_inotifydir,
+	    SYS_FS_INOTIFYDIRFILES));
+}
+
+/*
+ * lxpr_readdir(): Vnode operation for VOP_READDIR()
+ */
+/* ARGSUSED */
+static int
+lxpr_readdir(vnode_t *dp, uio_t *uiop, cred_t *cr, int *eofp,
+	caller_context_t *ct, int flags)
+{
+	lxpr_node_t *lxpnp = VTOLXP(dp);
+	lxpr_nodetype_t type = lxpnp->lxpr_type;
+	ssize_t uresid;
+	off_t uoffset;
+	int error;
+
+	ASSERT(dp->v_type == VDIR);
+	ASSERT(type < LXPR_NFILES);
+
+	/*
+	 * we should never get here because the readdir
+	 * is done on the realvp for these nodes
+	 */
+	ASSERT(type != LXPR_PID_FD_FD &&
+	    type != LXPR_PID_CURDIR &&
+	    type != LXPR_PID_ROOTDIR);
+
+	/*
+	 * restrict readdir permission to owner or root
+	 */
+	if ((error = lxpr_access(dp, VREAD, 0, cr, ct)) != 0)
+		return (error);
+
+	uoffset = uiop->uio_offset;
+	uresid = uiop->uio_resid;
+
+	/* can't do negative reads */
+	if (uoffset < 0 || uresid <= 0)
+		return (EINVAL);
+
+	/* can't read directory entries that don't exist! */
+	if (uoffset % LXPR_SDSIZE)
+		return (ENOENT);
+
+	return (lxpr_readdir_function[lxpnp->lxpr_type](lxpnp, uiop, eofp));
+}
+
+/* ARGSUSED */
+static int
+lxpr_readdir_not_a_dir(lxpr_node_t *lxpnp, uio_t *uiop, int *eofp)
+{
+	return (ENOTDIR);
+}
+
+/*
+ * This has the common logic for returning directory entries
+ */
+static int
+lxpr_readdir_common(lxpr_node_t *lxpnp, uio_t *uiop, int *eofp,
+    lxpr_dirent_t *dirtab, int dirtablen)
+{
+	/* bp holds one dirent64 structure */
+	longlong_t bp[DIRENT64_RECLEN(LXPNSIZ) / sizeof (longlong_t)];
+	dirent64_t *dirent = (dirent64_t *)bp;
+	ssize_t oresid;	/* save a copy for testing later */
+	ssize_t uresid;
+
+	oresid = uiop->uio_resid;
+
+	/* clear out the dirent buffer */
+	bzero(bp, sizeof (bp));
+
+	/*
+	 * Satisfy user request
+	 */
+	while ((uresid = uiop->uio_resid) > 0) {
+		int dirindex;
+		off_t uoffset;
+		int reclen;
+		int error;
+
+		uoffset = uiop->uio_offset;
+		dirindex  = (uoffset / LXPR_SDSIZE) - 2;
+
+		if (uoffset == 0) {
+
+			dirent->d_ino = lxpnp->lxpr_ino;
+			dirent->d_name[0] = '.';
+			dirent->d_name[1] = '\0';
+			reclen = DIRENT64_RECLEN(1);
+
+		} else if (uoffset == LXPR_SDSIZE) {
+
+			dirent->d_ino = lxpr_parentinode(lxpnp);
+			dirent->d_name[0] = '.';
+			dirent->d_name[1] = '.';
+			dirent->d_name[2] = '\0';
+			reclen = DIRENT64_RECLEN(2);
+
+		} else if (dirindex >= 0 && dirindex < dirtablen) {
+			int slen = strlen(dirtab[dirindex].d_name);
+
+			dirent->d_ino = lxpr_inode(dirtab[dirindex].d_type,
+			    lxpnp->lxpr_pid, 0);
+
+			VERIFY(slen < LXPNSIZ);
+			(void) strcpy(dirent->d_name, dirtab[dirindex].d_name);
+			reclen = DIRENT64_RECLEN(slen);
+
+		} else {
+			/* Run out of table entries */
+			if (eofp) {
+				*eofp = 1;
+			}
+			return (0);
+		}
+
+		dirent->d_off = (off64_t)(uoffset + LXPR_SDSIZE);
+		dirent->d_reclen = (ushort_t)reclen;
+
+		/*
+		 * if the size of the data to transfer is greater
+		 * that that requested then we can't do it this transfer.
+		 */
+		if (reclen > uresid) {
+			/*
+			 * Error if no entries have been returned yet.
+			 */
+			if (uresid == oresid) {
+				return (EINVAL);
+			}
+			break;
+		}
+
+		/*
+		 * uiomove() updates both uiop->uio_resid and uiop->uio_offset
+		 * by the same amount.  But we want uiop->uio_offset to change
+		 * in increments of LXPR_SDSIZE, which is different from the
+		 * number of bytes being returned to the user.  So we set
+		 * uiop->uio_offset separately, ignoring what uiomove() does.
+		 */
+		if ((error = uiomove((caddr_t)dirent, reclen, UIO_READ,
+		    uiop)) != 0)
+			return (error);
+
+		uiop->uio_offset = uoffset + LXPR_SDSIZE;
+	}
+
+	/* Have run out of space, but could have just done last table entry */
+	if (eofp) {
+		*eofp =
+		    (uiop->uio_offset >= ((dirtablen+2) * LXPR_SDSIZE)) ? 1 : 0;
+	}
+	return (0);
+}
+
+
+static int
+lxpr_readdir_procdir(lxpr_node_t *lxpnp, uio_t *uiop, int *eofp)
+{
+	/* bp holds one dirent64 structure */
+	longlong_t bp[DIRENT64_RECLEN(LXPNSIZ) / sizeof (longlong_t)];
+	dirent64_t *dirent = (dirent64_t *)bp;
+	ssize_t oresid;	/* save a copy for testing later */
+	ssize_t uresid;
+	off_t uoffset;
+	zoneid_t zoneid;
+	pid_t pid;
+	int error;
+	int ceof;
+
+	ASSERT(lxpnp->lxpr_type == LXPR_PROCDIR);
+
+	oresid = uiop->uio_resid;
+	zoneid = LXPTOZ(lxpnp)->zone_id;
+
+	/*
+	 * We return directory entries in the order: "." and ".." then the
+	 * unique lxproc files, then the directories corresponding to the
+	 * running processes.  We have defined this as the ordering because
+	 * it allows us to more easily keep track of where we are betwen calls
+	 * to getdents().  If the number of processes changes between calls
+	 * then we can't lose track of where we are in the lxproc files.
+	 */
+
+	/* Do the fixed entries */
+	error = lxpr_readdir_common(lxpnp, uiop, &ceof, lx_procdir,
+	    PROCDIRFILES);
+
+	/* Finished if we got an error or if we couldn't do all the table */
+	if (error != 0 || ceof == 0)
+		return (error);
+
+	/* clear out the dirent buffer */
+	bzero(bp, sizeof (bp));
+
+	/* Do the process entries */
+	while ((uresid = uiop->uio_resid) > 0) {
+		proc_t *p;
+		int len;
+		int reclen;
+		int i;
+
+		uoffset = uiop->uio_offset;
+
+		/*
+		 * Stop when entire proc table has been examined.
+		 */
+		i = (uoffset / LXPR_SDSIZE) - 2 - PROCDIRFILES;
+		if (i < 0 || i >= v.v_proc) {
+			/* Run out of table entries */
+			if (eofp) {
+				*eofp = 1;
+			}
+			return (0);
+		}
+		mutex_enter(&pidlock);
+
+		/*
+		 * Skip indices for which there is no pid_entry, PIDs for
+		 * which there is no corresponding process, a PID of 0,
+		 * and anything the security policy doesn't allow
+		 * us to look at.
+		 */
+		if ((p = pid_entry(i)) == NULL || p->p_stat == SIDL ||
+		    p->p_pid == 0 ||
+		    secpolicy_basic_procinfo(CRED(), p, curproc) != 0) {
+			mutex_exit(&pidlock);
+			goto next;
+		}
+		mutex_exit(&pidlock);
+
+		/*
+		 * Convert pid to the Linux default of 1 if we're the zone's
+		 * init process, or 0 if zsched, otherwise use the value from
+		 * the proc structure
+		 */
+		if (p->p_pid == curproc->p_zone->zone_proc_initpid) {
+			pid = 1;
+		} else if (p->p_pid == curproc->p_zone->zone_zsched->p_pid) {
+			pid = 0;
+		} else {
+			pid = p->p_pid;
+		}
+
+		/*
+		 * If this /proc was mounted in the global zone, view
+		 * all procs; otherwise, only view zone member procs.
+		 */
+		if (zoneid != GLOBAL_ZONEID && p->p_zone->zone_id != zoneid) {
+			goto next;
+		}
+
+		ASSERT(p->p_stat != 0);
+
+		dirent->d_ino = lxpr_inode(LXPR_PIDDIR, pid, 0);
+		len = snprintf(dirent->d_name, LXPNSIZ, "%d", pid);
+		ASSERT(len < LXPNSIZ);
+		reclen = DIRENT64_RECLEN(len);
+
+		dirent->d_off = (off64_t)(uoffset + LXPR_SDSIZE);
+		dirent->d_reclen = (ushort_t)reclen;
+
+		/*
+		 * if the size of the data to transfer is greater
+		 * that that requested then we can't do it this transfer.
+		 */
+		if (reclen > uresid) {
+			/*
+			 * Error if no entries have been returned yet.
+			 */
+			if (uresid == oresid)
+				return (EINVAL);
+			break;
+		}
+
+		/*
+		 * uiomove() updates both uiop->uio_resid and uiop->uio_offset
+		 * by the same amount.  But we want uiop->uio_offset to change
+		 * in increments of LXPR_SDSIZE, which is different from the
+		 * number of bytes being returned to the user.  So we set
+		 * uiop->uio_offset separately, in the increment of this for
+		 * the loop, ignoring what uiomove() does.
+		 */
+		if ((error = uiomove((caddr_t)dirent, reclen, UIO_READ,
+		    uiop)) != 0)
+			return (error);
+next:
+		uiop->uio_offset = uoffset + LXPR_SDSIZE;
+	}
+
+	if (eofp != NULL) {
+		*eofp = (uiop->uio_offset >=
+		    ((v.v_proc + PROCDIRFILES + 2) * LXPR_SDSIZE)) ? 1 : 0;
+	}
+
+	return (0);
+}
+
+static int
+lxpr_readdir_piddir(lxpr_node_t *lxpnp, uio_t *uiop, int *eofp)
+{
+	proc_t *p;
+	pid_t find_pid;
+
+	ASSERT(lxpnp->lxpr_type == LXPR_PIDDIR);
+
+	/* can't read its contents if it died */
+	mutex_enter(&pidlock);
+
+	if (lxpnp->lxpr_pid == 1) {
+		find_pid = curproc->p_zone->zone_proc_initpid;
+	} else if (lxpnp->lxpr_pid == 0) {
+		find_pid = curproc->p_zone->zone_zsched->p_pid;
+	} else {
+		find_pid = lxpnp->lxpr_pid;
+	}
+	p = prfind(find_pid);
+
+	if (p == NULL || p->p_stat == SIDL) {
+		mutex_exit(&pidlock);
+		return (ENOENT);
+	}
+	mutex_exit(&pidlock);
+
+	return (lxpr_readdir_common(lxpnp, uiop, eofp, piddir, PIDDIRFILES));
+}
+
+static int
+lxpr_readdir_netdir(lxpr_node_t *lxpnp, uio_t *uiop, int *eofp)
+{
+	ASSERT(lxpnp->lxpr_type == LXPR_NETDIR);
+	return (lxpr_readdir_common(lxpnp, uiop, eofp, netdir, NETDIRFILES));
+}
+
+static int
+lxpr_readdir_taskdir(lxpr_node_t *lxpnp, uio_t *uiop, int *eofp)
+{
+	/* bp holds one dirent64 structure */
+	longlong_t bp[DIRENT64_RECLEN(LXPNSIZ) / sizeof (longlong_t)];
+	dirent64_t *dirent = (dirent64_t *)bp;
+	ssize_t oresid;	/* save a copy for testing later */
+	ssize_t uresid;
+	off_t uoffset;
+	int error;
+	int ceof;
+	proc_t *p;
+	int tiddirsize = -1;
+	int tasknum;
+	pid_t real_pid;
+	kthread_t *t;
+
+	ASSERT(lxpnp->lxpr_type == LXPR_PID_TASKDIR);
+
+	oresid = uiop->uio_resid;
+
+	real_pid = get_real_pid(lxpnp->lxpr_pid);
+	p = lxpr_lock(real_pid);
+
+	/* can't read its contents if it died */
+	if (p == NULL) {
+		return (ENOENT);
+	}
+	if (p->p_stat == SIDL) {
+		lxpr_unlock(p);
+		return (ENOENT);
+	}
+
+	if ((p->p_stat == SZOMB) || (p->p_flag & SSYS) || (p->p_as == &kas))
+		tiddirsize = 0;
+
+	/*
+	 * Drop p_lock, but keep the process P_PR_LOCK'd to prevent it from
+	 * going away while we iterate over its threads.
+	 */
+	mutex_exit(&p->p_lock);
+
+	if (tiddirsize == -1)
+		tiddirsize = p->p_lwpcnt;
+
+	/* Do the fixed entries (in this case just "." & "..") */
+	error = lxpr_readdir_common(lxpnp, uiop, &ceof, 0, 0);
+
+	/* Finished if we got an error or if we couldn't do all the table */
+	if (error != 0 || ceof == 0)
+		goto out;
+
+	if ((t = p->p_tlist) == NULL) {
+		if (eofp != NULL)
+			*eofp = 1;
+		goto out;
+	}
+
+	/* clear out the dirent buffer */
+	bzero(bp, sizeof (bp));
+
+	/*
+	 * Loop until user's request is satisfied or until all thread's have
+	 * been returned.
+	 */
+	for (tasknum = 0; (uresid = uiop->uio_resid) > 0; tasknum++) {
+		int i;
+		int reclen;
+		int len;
+		uint_t emul_tid;
+		lx_lwp_data_t *lwpd;
+
+		uoffset = uiop->uio_offset;
+
+		/*
+		 * Stop at the end of the thread list
+		 */
+		i = (uoffset / LXPR_SDSIZE) - 2;
+		if (i < 0 || i >= tiddirsize) {
+			if (eofp) {
+				*eofp = 1;
+			}
+			goto out;
+		}
+
+		if (i != tasknum)
+			goto next;
+
+		lwpd = ttolxlwp(t);
+		if (lwpd == NULL) {
+			emul_tid = t->t_tid;
+		} else {
+			emul_tid = lwpd->br_pid;
+			/*
+			 * Convert pid to Linux default of 1 if we're the
+			 * zone's init.
+			 */
+			if (emul_tid == curproc->p_zone->zone_proc_initpid)
+				emul_tid = 1;
+		}
+
+		dirent->d_ino = lxpr_inode(LXPR_PID_TASK_IDDIR, lxpnp->lxpr_pid,
+		    emul_tid);
+		len = snprintf(dirent->d_name, LXPNSIZ, "%d", emul_tid);
+		ASSERT(len < LXPNSIZ);
+		reclen = DIRENT64_RECLEN(len);
+
+		dirent->d_off = (off64_t)(uoffset + LXPR_SDSIZE);
+		dirent->d_reclen = (ushort_t)reclen;
+
+		if (reclen > uresid) {
+			/*
+			 * Error if no entries have been returned yet.
+			 */
+			if (uresid == oresid)
+				error = EINVAL;
+			goto out;
+		}
+
+		/*
+		 * uiomove() updates both uiop->uio_resid and uiop->uio_offset
+		 * by the same amount.  But we want uiop->uio_offset to change
+		 * in increments of LXPR_SDSIZE, which is different from the
+		 * number of bytes being returned to the user.  So we set
+		 * uiop->uio_offset separately, in the increment of this for
+		 * the loop, ignoring what uiomove() does.
+		 */
+		if ((error = uiomove((caddr_t)dirent, reclen, UIO_READ,
+		    uiop)) != 0)
+			goto out;
+
+next:
+		uiop->uio_offset = uoffset + LXPR_SDSIZE;
+
+		if ((t = t->t_forw) == p->p_tlist) {
+			if (eofp != NULL)
+				*eofp = 1;
+			goto out;
+		}
+	}
+
+	if (eofp != NULL)
+		*eofp = 0;
+
+out:
+	mutex_enter(&p->p_lock);
+	lxpr_unlock(p);
+	return (error);
+}
+
+static int
+lxpr_readdir_task_tid_dir(lxpr_node_t *lxpnp, uio_t *uiop, int *eofp)
+{
+	proc_t *p;
+	pid_t real_pid;
+	kthread_t *t;
+
+	ASSERT(lxpnp->lxpr_type == LXPR_PID_TASK_IDDIR);
+
+	mutex_enter(&pidlock);
+
+	real_pid = get_real_pid(lxpnp->lxpr_pid);
+	p = prfind(real_pid);
+
+	/* can't read its contents if it died */
+	if (p == NULL || p->p_stat == SIDL) {
+		mutex_exit(&pidlock);
+		return (ENOENT);
+	}
+
+	mutex_exit(&pidlock);
+
+	/* need to confirm tid is still there */
+	t = lxpr_get_thread(p, lxpnp->lxpr_desc);
+	if (t == NULL) {
+		/* we can't find this specific thread */
+		return (NULL);
+	}
+	thread_unlock(t);
+
+	return (lxpr_readdir_common(lxpnp, uiop, eofp, tiddir, TIDDIRFILES));
+}
+
+static int
+lxpr_readdir_fddir(lxpr_node_t *lxpnp, uio_t *uiop, int *eofp)
+{
+	/* bp holds one dirent64 structure */
+	longlong_t bp[DIRENT64_RECLEN(LXPNSIZ) / sizeof (longlong_t)];
+	dirent64_t *dirent = (dirent64_t *)bp;
+	ssize_t oresid;	/* save a copy for testing later */
+	ssize_t uresid;
+	off_t uoffset;
+	int error;
+	int ceof;
+	proc_t *p;
+	int fddirsize = -1;
+	uf_info_t *fip;
+
+	ASSERT(lxpnp->lxpr_type == LXPR_PID_FDDIR ||
+	    lxpnp->lxpr_type == LXPR_PID_TID_FDDIR);
+
+	oresid = uiop->uio_resid;
+
+	/* can't read its contents if it died */
+	p = lxpr_lock(lxpnp->lxpr_pid);
+	if (p == NULL)
+		return (ENOENT);
+
+	if ((p->p_stat == SZOMB) || (p->p_flag & SSYS) || (p->p_as == &kas))
+		fddirsize = 0;
+
+	/*
+	 * Drop p_lock, but keep the process P_PR_LOCK'd to prevent it from
+	 * going away while we iterate over its fi_list.
+	 */
+	mutex_exit(&p->p_lock);
+
+	/* Get open file info */
+	fip = (&(p)->p_user.u_finfo);
+	mutex_enter(&fip->fi_lock);
+
+	if (fddirsize == -1)
+		fddirsize = fip->fi_nfiles;
+
+	/* Do the fixed entries (in this case just "." & "..") */
+	error = lxpr_readdir_common(lxpnp, uiop, &ceof, 0, 0);
+
+	/* Finished if we got an error or if we couldn't do all the table */
+	if (error != 0 || ceof == 0)
+		goto out;
+
+	/* clear out the dirent buffer */
+	bzero(bp, sizeof (bp));
+
+	/*
+	 * Loop until user's request is satisfied or until
+	 * all file descriptors have been examined.
+	 */
+	for (; (uresid = uiop->uio_resid) > 0;
+	    uiop->uio_offset = uoffset + LXPR_SDSIZE) {
+		int reclen;
+		int fd;
+		int len;
+
+		uoffset = uiop->uio_offset;
+
+		/*
+		 * Stop at the end of the fd list
+		 */
+		fd = (uoffset / LXPR_SDSIZE) - 2;
+		if (fd < 0 || fd >= fddirsize) {
+			if (eofp) {
+				*eofp = 1;
+			}
+			goto out;
+		}
+
+		if (fip->fi_list[fd].uf_file == NULL)
+			continue;
+
+		dirent->d_ino = lxpr_inode(LXPR_PID_FD_FD, lxpnp->lxpr_pid, fd);
+		len = snprintf(dirent->d_name, LXPNSIZ, "%d", fd);
+		ASSERT(len < LXPNSIZ);
+		reclen = DIRENT64_RECLEN(len);
+
+		dirent->d_off = (off64_t)(uoffset + LXPR_SDSIZE);
+		dirent->d_reclen = (ushort_t)reclen;
+
+		if (reclen > uresid) {
+			/*
+			 * Error if no entries have been returned yet.
+			 */
+			if (uresid == oresid)
+				error = EINVAL;
+			goto out;
+		}
+
+		if ((error = uiomove((caddr_t)dirent, reclen, UIO_READ,
+		    uiop)) != 0)
+			goto out;
+	}
+
+	if (eofp != NULL) {
+		*eofp =
+		    (uiop->uio_offset >= ((fddirsize+2) * LXPR_SDSIZE)) ? 1 : 0;
+	}
+
+out:
+	mutex_exit(&fip->fi_lock);
+	mutex_enter(&p->p_lock);
+	lxpr_unlock(p);
+	return (error);
+}
+
+static int
+lxpr_readdir_sysdir(lxpr_node_t *lxpnp, uio_t *uiop, int *eofp)
+{
+	ASSERT(lxpnp->lxpr_type == LXPR_SYSDIR);
+	return (lxpr_readdir_common(lxpnp, uiop, eofp, sysdir, SYSDIRFILES));
+}
+
+static int
+lxpr_readdir_sys_fsdir(lxpr_node_t *lxpnp, uio_t *uiop, int *eofp)
+{
+	ASSERT(lxpnp->lxpr_type == LXPR_SYS_FSDIR);
+	return (lxpr_readdir_common(lxpnp, uiop, eofp, sys_fsdir,
+	    SYS_FSDIRFILES));
+}
+
+static int
+lxpr_readdir_sys_fs_inotifydir(lxpr_node_t *lxpnp, uio_t *uiop, int *eofp)
+{
+	ASSERT(lxpnp->lxpr_type == LXPR_SYS_FS_INOTIFYDIR);
+	return (lxpr_readdir_common(lxpnp, uiop, eofp, sys_fs_inotifydir,
+	    SYS_FS_INOTIFYDIRFILES));
+}
+
+static int
+lxpr_readdir_sys_kerneldir(lxpr_node_t *lxpnp, uio_t *uiop, int *eofp)
+{
+	ASSERT(lxpnp->lxpr_type == LXPR_SYS_KERNELDIR);
+	return (lxpr_readdir_common(lxpnp, uiop, eofp, sys_kerneldir,
+	    SYS_KERNELDIRFILES));
+}
+
+/*
+ * lxpr_readlink(): Vnode operation for VOP_READLINK()
+ */
+/* ARGSUSED */
+static int
+lxpr_readlink(vnode_t *vp, uio_t *uiop, cred_t *cr, caller_context_t *ct)
+{
+	char bp[MAXPATHLEN + 1];
+	size_t buflen = sizeof (bp);
+	lxpr_node_t *lxpnp = VTOLXP(vp);
+	vnode_t *rvp = lxpnp->lxpr_realvp;
+	pid_t pid;
+	int error = 0;
+
+	/*
+	 * Linux does something very "clever" for /proc/<pid>/fd/<num> entries.
+	 * Open FDs are represented as symlinks, the link contents
+	 * corresponding to the open resource.  For plain files or devices,
+	 * this isn't absurd since one can dereference the symlink to query
+	 * the underlying resource.  For sockets or pipes, it becomes ugly in a
+	 * hurry.  To maintain this human-readable output, those FD symlinks
+	 * point to bogus targets such as "socket:[<inodenum>]".  This requires
+	 * circumventing vfs since the stat/lstat behavior on those FD entries
+	 * will be unusual. (A stat must retrieve information about the open
+	 * socket or pipe.  It cannot fail because the link contents point to
+	 * an absent file.)
+	 *
+	 * To accomplish this, lxpr_getnode returns an vnode typed VNON for FD
+	 * entries.  This bypasses code paths which would normally
+	 * short-circuit on symlinks and allows us to emulate the vfs behavior
+	 * expected by /proc consumers.
+	 */
+	if (vp->v_type != VLNK && lxpnp->lxpr_type != LXPR_PID_FD_FD)
+		return (EINVAL);
+
+	/* Try to produce a symlink name for anything that has a realvp */
+	if (rvp != NULL) {
+		if ((error = lxpr_access(vp, VREAD, 0, CRED(), ct)) != 0)
+			return (error);
+		if ((error = vnodetopath(NULL, rvp, bp, buflen, CRED())) != 0) {
+			/*
+			 * Special handling possible for /proc/<pid>/fd/<num>
+			 * Generate <type>:[<inode>] links, if allowed.
+			 */
+			if (lxpnp->lxpr_type != LXPR_PID_FD_FD ||
+			    lxpr_readlink_pid_fd(lxpnp, bp, buflen) != 0) {
+				return (error);
+			}
+		}
+	} else {
+		switch (lxpnp->lxpr_type) {
+		case LXPR_SELF:
+			/*
+			 * Convert pid to the Linux default of 1 if we're the
+			 * zone's init process or 0 if zsched.
+			 */
+			if (curproc->p_pid ==
+			    curproc->p_zone->zone_proc_initpid) {
+				pid = 1;
+			} else if (curproc->p_pid ==
+			    curproc->p_zone->zone_zsched->p_pid) {
+				pid = 0;
+			} else {
+				pid = curproc->p_pid;
+			}
+
+			/*
+			 * Don't need to check result as every possible int
+			 * will fit within MAXPATHLEN bytes.
+			 */
+			(void) snprintf(bp, buflen, "%d", pid);
+			break;
+		case LXPR_PID_CURDIR:
+		case LXPR_PID_ROOTDIR:
+		case LXPR_PID_EXE:
+			return (EACCES);
+		default:
+			/*
+			 * Need to return error so that nothing thinks
+			 * that the symlink is empty and hence "."
+			 */
+			return (EINVAL);
+		}
+	}
+
+	/* copy the link data to user space */
+	return (uiomove(bp, strlen(bp), UIO_READ, uiop));
+}
+
+/*
+ * Attempt to create Linux-proc-style fake symlinks contents for supported
+ * /proc/<pid>/fd/<#> entries.
+ */
+static int
+lxpr_readlink_pid_fd(lxpr_node_t *lxpnp, char *bp, size_t len)
+{
+	const char *format;
+	vnode_t *rvp = lxpnp->lxpr_realvp;
+	vattr_t attr;
+
+	switch (rvp->v_type) {
+	case VSOCK:
+		format = "socket:[%lu]";
+		break;
+	case VFIFO:
+		format = "pipe:[%lu]";
+		break;
+	default:
+		return (-1);
+	}
+
+	/* Fetch the inode of the underlying vnode */
+	if (VOP_GETATTR(rvp, &attr, 0, CRED(), NULL) != 0)
+		return (-1);
+
+	if (bp != NULL)
+		(void) snprintf(bp, len, format, (ino_t)attr.va_nodeid);
+	return (0);
+}
+
+/*
+ * lxpr_inactive(): Vnode operation for VOP_INACTIVE()
+ * Vnode is no longer referenced, deallocate the file
+ * and all its resources.
+ */
+/* ARGSUSED */
+static void
+lxpr_inactive(vnode_t *vp, cred_t *cr, caller_context_t *ct)
+{
+	lxpr_freenode(VTOLXP(vp));
+}
+
+/*
+ * lxpr_sync(): Vnode operation for VOP_SYNC()
+ */
+static int
+lxpr_sync()
+{
+	/*
+	 * Nothing to sync but this function must never fail
+	 */
+	return (0);
+}
+
+/*
+ * lxpr_cmp(): Vnode operation for VOP_CMP()
+ */
+static int
+lxpr_cmp(vnode_t *vp1, vnode_t *vp2, caller_context_t *ct)
+{
+	vnode_t *rvp;
+
+	while (vn_matchops(vp1, lxpr_vnodeops) &&
+	    (rvp = VTOLXP(vp1)->lxpr_realvp) != NULL) {
+		vp1 = rvp;
+	}
+
+	while (vn_matchops(vp2, lxpr_vnodeops) &&
+	    (rvp = VTOLXP(vp2)->lxpr_realvp) != NULL) {
+		vp2 = rvp;
+	}
+
+	if (vn_matchops(vp1, lxpr_vnodeops) || vn_matchops(vp2, lxpr_vnodeops))
+		return (vp1 == vp2);
+	return (VOP_CMP(vp1, vp2, ct));
+}
+
+/*
+ * lxpr_realvp(): Vnode operation for VOP_REALVP()
+ */
+static int
+lxpr_realvp(vnode_t *vp, vnode_t **vpp, caller_context_t *ct)
+{
+	vnode_t *rvp;
+
+	if ((rvp = VTOLXP(vp)->lxpr_realvp) != NULL) {
+		vp = rvp;
+		if (VOP_REALVP(vp, &rvp, ct) == 0)
+			vp = rvp;
+	}
+
+	*vpp = vp;
+	return (0);
+}
diff --git a/usr/src/uts/common/brand/lx/sys/lx_audio.h b/usr/src/uts/common/brand/lx/sys/lx_audio.h
new file mode 100644
index 0000000000..cbb3431c4b
--- /dev/null
+++ b/usr/src/uts/common/brand/lx/sys/lx_audio.h
@@ -0,0 +1,130 @@
+/*
+ * CDDL HEADER START
+ *
+ * The contents of this file are subject to the terms of the
+ * Common Development and Distribution License (the "License").
+ * You may not use this file except in compliance with the License.
+ *
+ * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
+ * or http://www.opensolaris.org/os/licensing.
+ * See the License for the specific language governing permissions
+ * and limitations under the License.
+ *
+ * When distributing Covered Code, include this CDDL HEADER in each
+ * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
+ * If applicable, add the following below this CDDL HEADER, with the
+ * fields enclosed by brackets "[]" replaced with your own identifying
+ * information: Portions Copyright [yyyy] [name of copyright owner]
+ *
+ * CDDL HEADER END
+ */
+/*
+ * Copyright 2006 Sun Microsystems, Inc.  All rights reserved.
+ * Use is subject to license terms.
+ */
+
+#ifndef	_LX_AUDIO_H
+#define	_LX_AUDIO_H
+
+#pragma ident	"%Z%%M%	%I%	%E% SMI"
+
+#include <sys/zone.h>
+
+#ifdef	__cplusplus
+extern "C" {
+#endif
+
+/*
+ * name for this driver
+ */
+#define	LX_AUDIO_DRV			"lx_audio"
+
+/*
+ * names for the minor nodes this driver exports
+ */
+#define	LXA_MINORNAME_DEVCTL		"lx_devctl"
+#define	LXA_MINORNAME_DSP		"lx_dsp"
+#define	LXA_MINORNAME_MIXER		"lx_mixer"
+
+/*
+ * minor numbers for the minor nodes this driver exporrts
+ */
+#define	LXA_MINORNUM_DEVCTL		0
+#define	LXA_MINORNUM_DSP		1
+#define	LXA_MINORNUM_MIXER		2
+#define	LXA_MINORNUM_COUNT		3
+
+/*
+ * driver ioctls
+ *
+ * note that we're layering on top of solaris audio devices so we want
+ * to make sure that our ioctls namespace doesn't conflict with theirs.
+ * looking in sys/audioio.h and sys/mixer.h we see that they seem to
+ * use an _IO key of 'A' and 'M', so we'll choose an _IO key of 'a.'
+ */
+
+/*
+ * administrative ioctls.
+ * these ioctls are only supported on the DEVCTL minor node
+ */
+#define	LXA_IOC_ZONE_REG		(_IOR('a', 0, lxa_zone_reg_t))
+#define	LXA_IOC_ZONE_UNREG		(_IOR('a', 1, lxa_zone_reg_t))
+
+
+/*
+ * audio and mixer device ioctls
+ * these ioctls are supported on DSP and MIXER minor nodes.
+ */
+#define	LXA_IOC_GETMINORNUM		(_IOR('a', 20, int))
+
+/*
+ * audio device ioctls.
+ * these ioctls are supports on DSP minor nodes.
+ */
+#define	LXA_IOC_MMAP_OUTPUT		(_IOR('a', 41, int))
+#define	LXA_IOC_MMAP_PTR		(_IOR('a', 42, int))
+#define	LXA_IOC_GET_FRAG_INFO		(_IOR('a', 43, lxa_frag_info_t))
+#define	LXA_IOC_SET_FRAG_INFO		(_IOR('a', 44, lxa_frag_info_t))
+
+/*
+ * mixer device ioctls.
+ * these ioctls are supports on MIXER minor nodes.
+ */
+#define	LXA_IOC_MIXER_GET_VOL		(_IOR('a', 60, lxa_mixer_levels_t))
+#define	LXA_IOC_MIXER_SET_VOL		(_IOR('a', 61, lxa_mixer_levels_t))
+#define	LXA_IOC_MIXER_GET_MIC		(_IOR('a', 62, lxa_mixer_levels_t))
+#define	LXA_IOC_MIXER_SET_MIC		(_IOR('a', 63, lxa_mixer_levels_t))
+#define	LXA_IOC_MIXER_GET_PCM		(_IOR('a', 64, lxa_mixer_levels_t))
+#define	LXA_IOC_MIXER_SET_PCM		(_IOR('a', 65, lxa_mixer_levels_t))
+
+/* command structure for LXA_IOC_ZONE_REG */
+#define	LXA_INTSTRLEN 32
+typedef struct lxa_zone_reg {
+	char	lxa_zr_zone_name[ZONENAME_MAX];
+	char	lxa_zr_inputdev[LXA_INTSTRLEN];
+	char	lxa_zr_outputdev[LXA_INTSTRLEN];
+} lxa_zone_reg_t;
+
+/* command structure for LXA_IOC_GET_FRAG_INFO and LXA_IOC_SET_FRAG_INFO */
+typedef struct lxa_frag_info {
+	int	lxa_fi_size;
+	int	lxa_fi_cnt;
+} lxa_frag_info_t;
+
+/* command structure for LXA_IOC_MIXER_GET_* and LXA_IOC_MIXER_SET_* */
+typedef struct lxa_mixer_levels {
+	int	lxa_ml_gain;
+	int	lxa_ml_balance;
+} lxa_mixer_levels_t;
+
+/* verify that a solaris mixer level structure has valid values */
+#define	LXA_MIXER_LEVELS_OK(x) (((x)->lxa_ml_gain >= AUDIO_MIN_GAIN) && \
+				((x)->lxa_ml_gain <= AUDIO_MAX_GAIN) && \
+				((x)->lxa_ml_balance >= AUDIO_LEFT_BALANCE) && \
+				((x)->lxa_ml_balance <= AUDIO_RIGHT_BALANCE))
+
+#ifdef	__cplusplus
+}
+#endif
+
+#endif	/* _LX_AUDIO_H */
diff --git a/usr/src/uts/common/brand/lx/sys/lx_autofs.h b/usr/src/uts/common/brand/lx/sys/lx_autofs.h
new file mode 100644
index 0000000000..a6ad86f594
--- /dev/null
+++ b/usr/src/uts/common/brand/lx/sys/lx_autofs.h
@@ -0,0 +1,337 @@
+/*
+ * CDDL HEADER START
+ *
+ * The contents of this file are subject to the terms of the
+ * Common Development and Distribution License (the "License").
+ * You may not use this file except in compliance with the License.
+ *
+ * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
+ * or http://www.opensolaris.org/os/licensing.
+ * See the License for the specific language governing permissions
+ * and limitations under the License.
+ *
+ * When distributing Covered Code, include this CDDL HEADER in each
+ * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
+ * If applicable, add the following below this CDDL HEADER, with the
+ * fields enclosed by brackets "[]" replaced with your own identifying
+ * information: Portions Copyright [yyyy] [name of copyright owner]
+ *
+ * CDDL HEADER END
+ */
+
+/*
+ * Copyright 2006 Sun Microsystems, Inc.  All rights reserved.
+ * Use is subject to license terms.
+ */
+/*
+ * Copyright 2015 Joyent, Inc.
+ */
+
+#ifndef	_LX_AUTOFS_H
+#define	_LX_AUTOFS_H
+
+/*
+ * The lx_autofs filesystem exists to emulate the Linux autofs filesystem
+ * and provide support for the Linux "automount" automounter.
+ *
+ *
+ *
+ * +++ Linux automounter background.
+ *
+ * Linux has two automounters: "amd" and "automount"
+ *
+ * 1) "amd" is a userland NFS server.  It basically mounts an NFS filesystem
+ * at an automount point, and it acts as the NFS server for the mount.  When
+ * an access is done to that NFS filesystem, the access is redirected by the
+ * kernel to the "amd" process via rpc.  "amd" then looks up any information
+ * required to resolve the requests, mounts real NFS filesystems if
+ * necessary, and returns.  "amd" has it's own strange configuration
+ * mechanism that doesn't seem to be very compatabile with Solaris's network
+ * based automounter map support.
+ *
+ * 2) "automount" is the other Linux automounter.  It utilizes a kernel
+ * filesystem (autofs) to provide it's functionality.  Basically, it mounts
+ * the autofs filesystem at any automounter controlled mount point.  This
+ * filesystem then intercepts and redirects lookup operations (and only
+ * lookup ops) to the userland automounter process via a pipe.  (The
+ * pipe to the automounter is establised via mount options when the autofs
+ * filesystem is mounted.)  When the automounter recieves a request via this
+ * pipe, it does lookups to whatever backing store it's configured to use,
+ * does mkdir operations on the autofs filesystem, mounts remote NFS
+ * filesystems on any leaf directories it just created, and signals the
+ * autofs filesystem via an ioctl to let it know that the lookup can
+ * continue.
+ *
+ *
+ *
+ * +++ Linux autofs (and automount daemon) notes
+ *
+ * Since we're mimicking the behavior of the Linux autofs filesystem it's
+ * important to document some of it's observed behavior here since there's
+ * no doubt that in the future this behavior will change.  These comments
+ * apply to the behavior of the automounter as observed on a system
+ * running Linux v2.4.21 (autofs is bundled with the Linux kernel).
+ *
+ * A) Autofs allows root owned, non-automounter processes to create
+ * directories in the autofs filesystem.  The autofs filesystem treats the
+ * automounter's process group as special, but it doesn't prevent root
+ * processes outside of the automounter's process group from creating new
+ * directories in the autofs filesystem.
+ *
+ * B) Autofs doesn't allow creation of any non-directory entries in the
+ * autofs filesystem.  No entity can create files (e.g. /bin/touch or
+ * VOP_CREATE/VOP_SYMLINK/etc.)  The only entries that can exist within
+ * the autofs filesystem are directories.
+ *
+ * C) Autofs only intercepts vop lookup operations.  Notably, it does _not_
+ * intercept and re-direct vop readdir operations.  This means that the
+ * observed behavior of the Linux automounter can be considerably different
+ * from that of the Solaris automounter.  Specifically, on Solaris if autofs
+ * mount point is mounted _without_ the -nobrowse option then if a user does
+ * an ls operation (which translates into a vop readdir operation) then the
+ * automounter will intercept that operation and list all the possible
+ * directories and mount points without actually mounting any filesystems.
+ * Essentially, all automounter managed mount points on Linux will behave
+ * like "-nobrowse" mount points on Solaris.  Here's an example to
+ * illustrate this.  If /ws was mounted on Solaris without the -nobrowse
+ * option and an auto_ws yp map was setup as the backing store for this
+ * mount point, then an "ls /ws" would list all the keys in the map as
+ * valid directories, but an "ls /ws" on Linux would list an emptry
+ * directory.
+ *
+ * D) NFS mounts are performed by the automount process.  When the automount
+ * process gets a redirected lookup request, it determines _all_ the
+ * possible remote mount points for that request, creates directory paths
+ * via mkdir, and mounts the remote filesystems on the newly created paths.
+ * So for example, if a machine called mcescher exported /var/crash and
+ * /var/core, an "ls /net/mcescher" would result in the following actions
+ * being done by the automounter:
+ * 	mkdir /net/mcescher
+ * 	mkdir /net/mcescher/var
+ * 	mkdir /net/mcescher/var/crash
+ * 	mkdir /net/mcescher/var/core
+ * 	mount mcescher:/var/crash /var/crash
+ * 	mount mcescher:/var/crash /var/core
+ * once the automounter compleated the work above it would signal the autofs
+ * filesystem (via an ioctl) that the lookup could continue.
+ *
+ * E.1) Autofs only redirects vop lookup operations for path entries that
+ * don't already exist in the autofs filesystem.  So for the example above,
+ * an initial (after the start of the automounter) "ls /net/mcescher" would
+ * result in a request to the automounter.  A subsequest "ls /net/mcescher"
+ * would not result in a request to the automounter.  Even if
+ * /net/mcescher/var/crash and /net/mcescher/var/core were manually unmounted
+ * after the initial "ls /net/mcescher", a subsequest "ls /net/mcescher"
+ * would not result in a new request to the automounter.
+ *
+ * E.2) Autofs lookup requests that are sent to the automounter only include
+ * the root directory path component.  So for example, after starting up
+ * the automounter if a user were to do a "ls /net/mcescher/var/crash", the
+ * lookup request actually sent to the automounter would just be for
+ * "mcescher".  (The same request as if the user had done "ls /net/mcescher".)
+ *
+ * E.3) The two statements above aren't entirely entirely true.  The Linux
+ * autofs filesystem will also redirect lookup operations for leaf
+ * directories that don't have a filesystem mounted on them.  Using the
+ * example above, if a user did a "ls /net/mcescher", then manually
+ * unmounted /net/mcescher/var/crash, and then did an "ls
+ * /net/mcescher/var/crash", this would result in a request for
+ * "mcescher/var/crash" being sent to the automounter.  The strange thing
+ * (a Linux bug perhaps) is that the automounter won't do anything with this
+ * request and the lookup will fail.
+ *
+ * F) The autofs filesystem communication protocol (what ioctls it supports
+ * and what data it passes to the automount process) are versioned.  The
+ * source for the userland automount daemon (i looked at version v3.1.7)
+ * seemed to support two versions of the Linux kernel autofs implementation.
+ * Both versions supported communiciation with a pipe and the format of the
+ * structure passed via this pipe was the same.  The difference between the
+ * two versions was in the functionality supported.  (The v3 version has
+ * additional ioctls to support automount timeouts.)
+ *
+ *
+ *
+ * +++ lx_autofs notes
+ *
+ * 1) In general, the lx_autofs filesystem tries to mimic the behavior of the
+ * Linux autofs filesystem with the following exceptions:
+ *
+ * 	1.1) We don't bother to implement the E.3 functionality listed above
+ * 	since it doesn't appear to be of any use.
+ *
+ * 	1.2) We only implement v2 of the automounter protocol since
+ * 	implementing v3 would take a _lot_ more work.  If this proves to be a
+ * 	problem we can re-visit this decision later.  (More details about v3
+ * 	support are included in comments below.)
+ *
+ * 2) In general, the approach taken for lx_autofs is to keep it as simple
+ * as possible and to minimize it's memory usage.  To do this all information
+ * about the contents of the lx_autofs filesystem are mirrored in the
+ * underlying filesystem that lx_autofs is mounted on and most vop operations
+ * are simply passed onto this underlying filesystem.  This means we don't
+ * have to implement most the complex operations that a full filesystem
+ * normally has to implement.  It also means that most of our filesystem state
+ * (wrt the contents of the filesystem) doesn't actually have to be stored
+ * in memory, we can simply go to the underlying filesystem to get it when
+ * it's requested.  For the purposes of discussion, we'll call the underlying
+ * filesystem the "backing store."
+ *
+ * The backing store is actually directory called ".lx_afs" which is created in
+ * the directory where the lx_autofs filesystem is mounted.  When the lx_autofs
+ * filesystem is unmounted this backing store directory is deleted.  If this
+ * directory exists at mount time (perhaps the system crashed while a previous
+ * lx_autofs instance was mounted at the same location) it will be deleted.
+ * There are a few implications of using a backing store worth mentioning.
+ *
+ * 	2.1) lx_autofs can't be mounted on a read only filesystem.  If this
+ * 	proves to be a problem we can probably move the location of the
+ * 	backing store.
+ *
+ * 	2.2) If the backing store filesystem runs out of space then the
+ * 	automounter process won't be able to create more directories and mount
+ * 	new filesystems.  Of course, strange failures usually happen when
+ * 	filesystems run out of space.
+ *
+ * 3) Why aren't we using gfs?  gfs has two different usage models.
+ *
+ * 	3.1) I'm my own filesystem but i'm using gfs to help with managing
+ * 	readdir operations.
+ *
+ * 	3.2) I'm a gfs filesystem and gfs is managing all my vnodes
+ *
+ * We're not using the 3.1 interfaces because we don't implement readdir
+ * ourselves.  We pass all readdir operations onto the backing store
+ * filesystem and utilize its readdir implementation.
+ *
+ * We're not using the 3.2 interfaces because they are really designed for
+ * in memory filesystems where all of the filesystem state is stored in
+ * memory.  They don't lend themselves to filesystems where part of the
+ * state is in memory and part of the state is on disk.
+ *
+ * For more information on gfs take a look at the block comments in the
+ * top of gfs.c
+ */
+
+#ifdef	__cplusplus
+extern "C" {
+#endif
+
+/*
+ * Note that the name of the actual Solaris filesystem is lx_afs and not
+ * lx_autofs.  This is becase filesystem names are stupidly limited to 8
+ * characters.
+ */
+#define	LX_AUTOFS_NAME			"lx_afs"
+
+/*
+ * Mount options supported.
+ */
+#define	LX_MNTOPT_FD			"fd"
+#define	LX_MNTOPT_PGRP			"pgrp"
+#define	LX_MNTOPT_MINPROTO		"minproto"
+#define	LX_MNTOPT_MAXPROTO		"maxproto"
+#define	LX_MNTOPT_MODE			"mode"
+#define	LX_MNTOPT_SIZE			"size"
+
+/* Version of the Linux kernel automount protocol we support. */
+#define	LX_AUTOFS_PROTO_VERSION		2
+
+/*
+ * Command structure sent to automount process from lx_autofs via a pipe.
+ * This structure is the same for v2 and v3 of the automount protocol
+ * (the communication pipe is established at mount time).
+ */
+typedef struct lx_autofs_pkt {
+	int	lap_protover;	/* protocol version number */
+	int	lap_constant;	/* always set to 0 */
+	int	lap_id;		/* every pkt must have a unique id */
+	int	lap_name_len;	/* don't include newline or NULL */
+	char	lap_name[256];	/* path component to lookup */
+} lx_autofs_pkt_t;
+
+/*
+ * Ioctls supprted (v2 protocol).
+ */
+#define	LX_AUTOFS_IOC_READY		0x00009360 /* arg: int */
+#define	LX_AUTOFS_IOC_FAIL		0x00009361 /* arg: int */
+#define	LX_AUTOFS_IOC_CATATONIC		0x00009362 /* arg: <none> */
+
+/*
+ * Ioctls not supported (v3 protocol).
+ *
+ * Initially we're only going to support v2 of the Linux kernel automount
+ * protocol.  This means that we don't support the following ioctls.
+ *
+ * 1) The protocol version ioctl (by not supporting it the automounter
+ * will assume version 2).
+ *
+ * 2) Automounter timeout ioctls.  For v3 and later the automounter can
+ * be started with a timeout option.  It will notify the filesystem of
+ * this timeout and, if any automounter filesystem root directory entry
+ * is not in use, it will notify the automounter via the LX_AUTOFS_IOC_EXPIRE
+ * ioctl.  For example, if the timeout is 60 seconds, the Linux
+ * automounter will use the LX_AUTOFS_IOC_EXPIRE ioctl to query for
+ * timeouts more often than that.  (v3.1.7 of the automount daemon would
+ * perform this ioctl every <timeout>/4 seconds.)  Then, if the autofs
+ * filesystem will
+ * report top level directories that aren't in use to the automounter
+ * via this ioctl.  If /net was managed by the automounter and
+ * there were the following mount points:
+ *	/net/jurassic/var/crash
+ *	/net/mcescher/var/crash
+ * and no one was looking at any crash dumps on mcescher but someone
+ * was analyzing a crash dump on jurassic, then after <timeout> seconds
+ * had passed the autofs filesystem would let the automounter know that
+ * "mcescher" could be unmounted.  (Note the granularity of notification
+ * is directories in the root of the autofs filesystem.)  Here's two
+ * ideas for how this functionality could be implemented on Solaris:
+ *
+ * 2.1) The easy incomplete way.  Don't do any in-use detection.  Simply
+ * tell the automounter it can try to unmount the filesystem every time
+ * the specified timeout passes.  If the filesystem is in use then the
+ * unmount will fail.  This would break down for remote hosts with multiple
+ * mounts.  For example, if the automounter had mounted the following
+ * filesystems:
+ *	/net/jurassic/var/crash
+ *	/net/jurassic/var/core
+ * and the user was looking at a core file, and the timeout expired, the
+ * automounter would recieve notification to unmount "jurassic".  Then
+ * it would unmount crash (which would succeed) and then to try unmount
+ * core (which would fail).  After that (since the automounter only
+ * performs mounts for failed lookups in the root autofs directory)
+ * future access to /net/jurassic/var/crash would result to access
+ * to an empty autofs directory.  We might be able to work around
+ * this by caching which root autofs directories we've timed out,
+ * then any access to paths that contain those directories could be
+ * stalled and we could resend another request to the automounter.
+ * This could work if the automounter ignores mount failures.
+ *
+ * 2.2) The hard correct way.  The real difficulty here is detecting
+ * files in use on other filesystems (say NFS) that have been mounted
+ * on top of autofs.  (Detecting in use autofs vnodes should be easy.)
+ * to do this we would probably have to create a new brand op to intercept
+ * mount/umount filesystem operations.  Then using this entry point we
+ * could detect mounts of other filesystems on top of lx_autofs.  When
+ * a successful mount finishes we would use the FEM (file event
+ * monitoring) framework to push a module onto that filesystem and
+ * intercept VOP operations that allocate/free vnodes in that filesystem.
+ * (We would also then have to track mount operations on top of that
+ * filesystem, etc.)  this would allow us to properly detect any
+ * usage of subdirectories of an autofs directory.
+ */
+#define	LX_AUTOFS_IOC_PROTOVER		0x80049363 /* arg: int */
+#define	LX_AUTOFS_IOC_EXPIRE		0x81109365 /* arg: lx_autofs_expire * */
+#define	LX_AUTOFS_IOC_SETTIMEOUT	0xc0049364 /* arg: ulong_t */
+
+typedef struct lx_autofs_expire {
+	int	lap_protover;	/* protol version number */
+	int	lap_constant;	/* always set to 1 */
+	int	lap_name_len;	/* don't include newline or NULL */
+	char	lap_name[256];	/* path component that has timed out */
+} lx_autofs_expire_t;
+
+#ifdef	__cplusplus
+}
+#endif
+
+#endif	/* _LX_AUTOFS_H */
diff --git a/usr/src/uts/common/brand/lx/sys/lx_autofs_impl.h b/usr/src/uts/common/brand/lx/sys/lx_autofs_impl.h
new file mode 100644
index 0000000000..9c5517b8d5
--- /dev/null
+++ b/usr/src/uts/common/brand/lx/sys/lx_autofs_impl.h
@@ -0,0 +1,121 @@
+/*
+ * CDDL HEADER START
+ *
+ * The contents of this file are subject to the terms of the
+ * Common Development and Distribution License (the "License").
+ * You may not use this file except in compliance with the License.
+ *
+ * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
+ * or http://www.opensolaris.org/os/licensing.
+ * See the License for the specific language governing permissions
+ * and limitations under the License.
+ *
+ * When distributing Covered Code, include this CDDL HEADER in each
+ * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
+ * If applicable, add the following below this CDDL HEADER, with the
+ * fields enclosed by brackets "[]" replaced with your own identifying
+ * information: Portions Copyright [yyyy] [name of copyright owner]
+ *
+ * CDDL HEADER END
+ */
+
+/*
+ * Copyright 2006 Sun Microsystems, Inc.  All rights reserved.
+ * Use is subject to license terms.
+ */
+
+#ifndef	_LX_AUTOFS_IMPL_H
+#define	_LX_AUTOFS_IMPL_H
+
+#pragma ident	"%Z%%M%	%I%	%E% SMI"
+
+#ifdef	__cplusplus
+extern "C" {
+#endif
+
+#include <sys/file.h>
+#include <sys/id_space.h>
+#include <sys/modhash.h>
+#include <sys/vnode.h>
+
+#include <sys/lx_autofs.h>
+
+/*
+ * Space key.
+ * Used to persist data across lx_autofs filesystem module unloads.
+ */
+#define	LX_AUTOFS_SPACE_KEY_UDEV	LX_AUTOFS_NAME "_udev"
+
+/*
+ * Name of the backing store directory.
+ */
+#define	LX_AUTOFS_BS_DIR		"." LX_AUTOFS_NAME
+
+#define	LX_AUTOFS_VFS_ID_HASH_SIZE	15
+#define	LX_AUTOFS_VFS_PATH_HASH_SIZE	15
+#define	LX_AUTOFS_VFS_VN_HASH_SIZE	15
+
+/*
+ * VFS data object.
+ */
+typedef struct lx_autofs_vfs {
+	/* Info about the underlying filesystem and backing store. */
+	vnode_t		*lav_mvp;
+	char		*lav_bs_name;
+	vnode_t		*lav_bs_vp;
+
+	/* Info about the automounter process managing this filesystem. */
+	int		lav_fd;
+	pid_t		lav_pgrp;
+	file_t		*lav_fifo_wr;
+	file_t		*lav_fifo_rd;
+
+	/* Each automount requests needs a unique id. */
+	id_space_t	*lav_ids;
+
+	/* All remaining structure members are protected by lav_lock. */
+	kmutex_t	lav_lock;
+
+	/* Hashes to keep track of outstanding automounter requests. */
+	mod_hash_t	*lav_path_hash;
+	mod_hash_t	*lav_id_hash;
+
+	/* We need to keep track of all our vnodes. */
+	vnode_t		*lav_root;
+	mod_hash_t	*lav_vn_hash;
+} lx_autofs_vfs_t;
+
+/*
+ * Structure to keep track of requests sent to the automounter.
+ */
+typedef struct lx_autofs_lookup_req {
+	/* Packet that gets sent to the automounter. */
+	lx_autofs_pkt_t	lalr_pkt;
+
+	/* Reference count.  Always updated atomically. */
+	uint_t		lalr_ref;
+
+	/*
+	 * Fields to keep track and sync threads waiting on a lookup.
+	 * Fields are protected by lalr_lock.
+	 */
+	kmutex_t	lalr_lock;
+	kcondvar_t	lalr_cv;
+	int		lalr_complete;
+} lx_autofs_lookup_req_t;
+
+/*
+ * Generic stack structure.
+ */
+typedef struct stack_elem {
+	list_node_t	se_list;
+	caddr_t		se_ptr1;
+	caddr_t		se_ptr2;
+	caddr_t		se_ptr3;
+} stack_elem_t;
+
+#ifdef	__cplusplus
+}
+#endif
+
+#endif	/* _LX_AUTOFS_IMPL_H */
diff --git a/usr/src/uts/common/brand/lx/sys/lx_brand.h b/usr/src/uts/common/brand/lx/sys/lx_brand.h
new file mode 100644
index 0000000000..882b25743a
--- /dev/null
+++ b/usr/src/uts/common/brand/lx/sys/lx_brand.h
@@ -0,0 +1,698 @@
+/*
+ * CDDL HEADER START
+ *
+ * The contents of this file are subject to the terms of the
+ * Common Development and Distribution License (the "License").
+ * You may not use this file except in compliance with the License.
+ *
+ * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
+ * or http://www.opensolaris.org/os/licensing.
+ * See the License for the specific language governing permissions
+ * and limitations under the License.
+ *
+ * When distributing Covered Code, include this CDDL HEADER in each
+ * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
+ * If applicable, add the following below this CDDL HEADER, with the
+ * fields enclosed by brackets "[]" replaced with your own identifying
+ * information: Portions Copyright [yyyy] [name of copyright owner]
+ *
+ * CDDL HEADER END
+ */
+/*
+ * Copyright 2010 Sun Microsystems, Inc.  All rights reserved.
+ * Use is subject to license terms.
+ */
+
+/*
+ * Copyright 2015 Joyent, Inc.
+ */
+
+#ifndef _LX_BRAND_H
+#define	_LX_BRAND_H
+
+#ifndef _ASM
+#include <sys/types.h>
+#include <sys/cpuvar.h>
+#include <sys/zone.h>
+#include <sys/ksocket.h>
+#endif
+
+#ifdef	__cplusplus
+extern "C" {
+#endif
+
+#define	LX_BRANDNAME	"lx"
+
+/*
+ * Brand uname info
+ */
+#define	LX_UNAME_SYSNAME	"Linux"
+#define	LX_UNAME_RELEASE_2_6	"2.6.18"
+#define	LX_UNAME_RELEASE_2_4	"2.4.21"
+#define	LX_UNAME_VERSION	"BrandZ virtual linux"
+#define	LX_UNAME_MACHINE32	"i686"
+#define	LX_UNAME_MACHINE64	"x86_64"
+
+#define	LX_LIB_PATH32	"/native/usr/lib/lx_brand.so.1"
+#define	LX_LIB_PATH64	"/native/usr/lib/amd64/lx_brand.so.1"
+
+#if defined(_LP64)
+#define	LX_LIB_PATH		LX_LIB_PATH64
+#define	LX_UNAME_MACHINE	LX_UNAME_MACHINE64
+#else
+#define	LX_LIB_PATH		LX_LIB_PATH32
+#define	LX_UNAME_MACHINE	LX_UNAME_MACHINE32
+#endif
+
+/*
+ * This must be large enough for both the 32-bit table and 64-bit table.
+ */
+#define	LX_NSYSCALLS		358
+
+/*
+ * brand(2) subcommands
+ *
+ * Everything >= 128 is a brand-specific subcommand.
+ * > 192 is reserved for in-kernel emulated system calls.
+ */
+#define	B_LPID_TO_SPAIR		128
+#define	B_GET_CURRENT_CONTEXT	129
+#define	B_EMULATION_DONE	130
+#define	B_PTRACE_KERNEL		131
+#define	B_SET_AFFINITY_MASK	132
+#define	B_GET_AFFINITY_MASK	133
+#define	B_PTRACE_CLONE_BEGIN	134
+#define	B_PTRACE_STOP_FOR_OPT	135
+#define	B_UNSUPPORTED		136
+#define	B_STORE_ARGS		137
+#define	B_GETPID		138
+#define	B_JUMP_TO_LINUX		139
+#define	B_SET_THUNK_PID		140
+#define	B_EXIT_AS_SIG		141
+#define	B_HELPER_WAITID		142
+#define	B_HELPER_CLONE		143
+#define	B_HELPER_SETGROUPS	144
+#define	B_HELPER_SIGQUEUE	145
+#define	B_HELPER_TGSIGQUEUE	146
+#define	B_SET_NATIVE_STACK	147
+
+#ifndef _ASM
+/*
+ * Support for Linux PTRACE_SETOPTIONS handling.
+ */
+typedef enum lx_ptrace_options {
+	LX_PTRACE_O_TRACESYSGOOD =	0x0001,
+	LX_PTRACE_O_TRACEFORK =		0x0002,
+	LX_PTRACE_O_TRACEVFORK =	0x0004,
+	LX_PTRACE_O_TRACECLONE =	0x0008,
+	LX_PTRACE_O_TRACEEXEC =		0x0010,
+	LX_PTRACE_O_TRACEVFORKDONE =	0x0020,
+	LX_PTRACE_O_TRACEEXIT =		0x0040,
+	LX_PTRACE_O_TRACESECCOMP =	0x0080
+} lx_ptrace_options_t;
+
+#define	LX_PTRACE_O_ALL							\
+	(LX_PTRACE_O_TRACESYSGOOD | LX_PTRACE_O_TRACEFORK | 		\
+	LX_PTRACE_O_TRACEVFORK | LX_PTRACE_O_TRACECLONE | 		\
+	LX_PTRACE_O_TRACEEXEC | LX_PTRACE_O_TRACEVFORKDONE |		\
+	LX_PTRACE_O_TRACEEXIT | LX_PTRACE_O_TRACESECCOMP)
+#endif /* !_ASM */
+
+/* siginfo si_status for traced events */
+#define	LX_PTRACE_EVENT_FORK		0x100
+#define	LX_PTRACE_EVENT_VFORK		0x200
+#define	LX_PTRACE_EVENT_CLONE		0x300
+#define	LX_PTRACE_EVENT_EXEC		0x400
+#define	LX_PTRACE_EVENT_VFORK_DONE	0x500
+#define	LX_PTRACE_EVENT_EXIT		0x600
+#define	LX_PTRACE_EVENT_SECCOMP		0x700
+
+/*
+ * Brand-private values for the "pr_what" member of lwpstatus, for use with the
+ * PR_BRAND stop reason.  These reasons are validated in lx_stop_notify();
+ * update it if you add new reasons here.
+ */
+#define	LX_PR_SYSENTRY		1
+#define	LX_PR_SYSEXIT		2
+#define	LX_PR_SIGNALLED		3
+#define	LX_PR_EVENT		4
+
+
+#define	LX_VERSION_1		1
+#define	LX_VERSION		LX_VERSION_1
+
+#define	LX_KERN_VERSION_NUM	ZONE_ATTR_BRAND_ATTRS
+
+/*
+ * Aux vector containing phdr of Linux executable and ehdr of interpreter
+ * (if any), both of which are used by lx_librtld_db to ascertain r_debug.
+ * We repurpose the 3rd brand-specific aux vector slot for the Linux
+ * AT_SYSINFO_EHDR entry (we modify the a_type in the brand library).
+ */
+#define	AT_SUN_BRAND_LX_PHDR	AT_SUN_BRAND_AUX1
+#define	AT_SUN_BRAND_LX_INTERP	AT_SUN_BRAND_AUX2
+#define	AT_SUN_BRAND_LX_SYSINFO_EHDR	AT_SUN_BRAND_AUX3
+
+/* Aux vector containing hz value */
+#define	AT_CLKTCK	17
+/* Aux vector containing vDSO addr */
+#define	AT_SYSINFO_EHDR	33
+
+/*
+ * Usermode emulation routines are run on an alternate stack allocated by
+ * the brand library.  Every LWP in a process will incur this overhead beyond
+ * the regular thread stack:
+ */
+#define	LX_NATIVE_STACK_PAGE_COUNT	64
+
+/*
+ * When returning in a new child process created with vfork(2) (or CLONE_VFORK)
+ * we discard some of the native stack to prevent corruption of the parent
+ * emulation state.
+ */
+#define	LX_NATIVE_STACK_VFORK_GAP	0x3000
+
+#ifndef	_ASM
+
+extern struct brand lx_brand;
+
+typedef struct lx_brand_registration {
+	uint_t lxbr_version;		/* version number */
+	void *lxbr_handler;		/* base address of handler */
+	uint32_t lxbr_flags;		/* LX_PROC_* registration flags */
+} lx_brand_registration_t;
+
+typedef struct lx_brand_registration32 {
+	uint_t lxbr_version;		/* version number */
+	uint32_t lxbr_handler;		/* base address of handler */
+	uint32_t lxbr_flags;		/* LX_PROC_* registration flags */
+} lx_brand_registration32_t;
+
+#ifdef __amd64
+
+typedef struct lx_regs {
+	long lxr_fs;
+	long lxr_rdi;
+	long lxr_rsi;
+	long lxr_rbp;
+	long lxr_rsp;
+	long lxr_rbx;
+	long lxr_rdx;
+	long lxr_rcx;
+	long lxr_rax;
+	long lxr_r8;
+	long lxr_r9;
+	long lxr_r10;
+	long lxr_r11;
+	long lxr_r12;
+	long lxr_r13;
+	long lxr_r14;
+	long lxr_r15;
+	long lxr_rip;
+
+	long lxr_orig_rax;
+} lx_regs_t;
+
+typedef struct lx_regs32 {
+	uint32_t lxr_gs;
+	uint32_t lxr_edi;
+	uint32_t lxr_esi;
+	uint32_t lxr_ebp;
+	uint32_t lxr_esp;
+	uint32_t lxr_ebx;
+	uint32_t lxr_edx;
+	uint32_t lxr_ecx;
+	uint32_t lxr_eax;
+	uint32_t lxr_eip;
+
+	uint32_t lxr_orig_eax;
+} lx_regs32_t;
+
+#else /* ! __amd64 */
+
+typedef struct lx_regs {
+	long lxr_gs;
+	long lxr_edi;
+	long lxr_esi;
+	long lxr_ebp;
+	long lxr_esp;
+	long lxr_ebx;
+	long lxr_edx;
+	long lxr_ecx;
+	long lxr_eax;
+	long lxr_eip;
+
+	long lxr_orig_eax;
+} lx_regs_t;
+
+#endif /* __amd64 */
+
+#ifdef __amd64
+/*
+ * The 64-bit native "user_regs_struct" Linux structure.
+ */
+typedef struct lx_user_regs {
+	long lxur_r15;
+	long lxur_r14;
+	long lxur_r13;
+	long lxur_r12;
+	long lxur_rbp;
+	long lxur_rbx;
+	long lxur_r11;
+	long lxur_r10;
+	long lxur_r9;
+	long lxur_r8;
+	long lxur_rax;
+	long lxur_rcx;
+	long lxur_rdx;
+	long lxur_rsi;
+	long lxur_rdi;
+	long lxur_orig_rax;
+	long lxur_rip;
+	long lxur_xcs;
+	long lxur_rflags;
+	long lxur_rsp;
+	long lxur_xss;
+	long lxur_xfs_base;
+	long lxur_xgs_base;
+	long lxur_xds;
+	long lxur_xes;
+	long lxur_xfs;
+	long lxur_xgs;
+} lx_user_regs_t;
+
+#if defined(_KERNEL) && defined(_SYSCALL32_IMPL)
+/*
+ * 64-bit kernel view of the 32-bit "user_regs_struct" Linux structure.
+ */
+typedef struct lx_user_regs32 {
+	int32_t lxur_ebx;
+	int32_t lxur_ecx;
+	int32_t lxur_edx;
+	int32_t lxur_esi;
+	int32_t lxur_edi;
+	int32_t lxur_ebp;
+	int32_t lxur_eax;
+	int32_t lxur_xds;
+	int32_t lxur_xes;
+	int32_t lxur_xfs;
+	int32_t lxur_xgs;
+	int32_t lxur_orig_eax;
+	int32_t lxur_eip;
+	int32_t lxur_xcs;
+	int32_t lxur_eflags;
+	int32_t lxur_esp;
+	int32_t lxur_xss;
+} lx_user_regs32_t;
+#endif /* defined(_KERNEL) && defined(_SYSCALL32_IMPL) */
+
+#else /* !__amd64 */
+/*
+ * The 32-bit native "user_regs_struct" Linux structure.
+ */
+typedef struct lx_user_regs {
+	long lxur_ebx;
+	long lxur_ecx;
+	long lxur_edx;
+	long lxur_esi;
+	long lxur_edi;
+	long lxur_ebp;
+	long lxur_eax;
+	long lxur_xds;
+	long lxur_xes;
+	long lxur_xfs;
+	long lxur_xgs;
+	long lxur_orig_eax;
+	long lxur_eip;
+	long lxur_xcs;
+	long lxur_eflags;
+	long lxur_esp;
+	long lxur_xss;
+} lx_user_regs_t;
+#endif /* __amd64 */
+
+#endif /* _ASM */
+
+/*
+ * GDT usage
+ */
+#define	GDT_TLSMIN	(GDT_BRANDMIN)
+#define	GDT_TLSMAX	(GDT_TLSMIN + 2)
+#define	LX_TLSNUM	(GDT_TLSMAX - GDT_TLSMIN)
+
+#ifndef _ASM
+
+/*
+ * Stores information needed by the lx linker to launch the main
+ * lx executable.
+ */
+typedef struct lx_elf_data64 {
+	uintptr_t	ed_phdr;
+	uintptr_t	ed_phent;
+	uintptr_t	ed_phnum;
+	uintptr_t	ed_entry;
+	uintptr_t	ed_base;
+	uintptr_t	ed_ldentry;
+} lx_elf_data64_t;
+
+typedef struct lx_elf_data32 {
+	uint32_t	ed_phdr;
+	uint32_t	ed_phent;
+	uint32_t	ed_phnum;
+	uint32_t	ed_entry;
+	uint32_t	ed_base;
+	uint32_t	ed_ldentry;
+} lx_elf_data32_t;
+
+#if defined(_LP64)
+typedef lx_elf_data64_t lx_elf_data_t;
+#else
+typedef lx_elf_data32_t lx_elf_data_t;
+#endif
+
+typedef enum lx_proc_flags {
+	LX_PROC_INSTALL_MODE = 0x01,
+	LX_PROC_STRICT_MODE = 0x02
+} lx_proc_flags_t;
+
+#define	LX_PROC_ALL	(LX_PROC_INSTALL_MODE | LX_PROC_STRICT_MODE)
+
+#ifdef	_KERNEL
+
+#define	LX_RLFAKE_LOCKS		0
+#define	LX_RLFAKE_NICE		1
+#define	LX_RLFAKE_RTPRIO	2
+#define	LX_RLFAKE_RTTIME	3
+
+#define	LX_RLFAKE_NLIMITS	4
+
+#define	LX_RLIM64_INFINITY	(~0ULL)
+
+typedef struct {
+	uint64_t	rlim_cur;
+	uint64_t	rlim_max;
+} lx_rlimit64_t;
+
+typedef struct lx_proc_data {
+	uintptr_t l_handler;	/* address of user-space handler */
+	pid_t l_ppid;		/* pid of originating parent proc */
+	uint64_t l_ptrace;	/* process being observed with ptrace */
+	lx_elf_data_t l_elf_data; /* ELF data for linux executable */
+	int l_signal;		/* signal to deliver to parent when this */
+				/* thread group dies */
+	lx_proc_flags_t l_flags;
+
+	lx_rlimit64_t l_fake_limits[LX_RLFAKE_NLIMITS];
+} lx_proc_data_t;
+
+#endif	/* _KERNEL */
+
+/*
+ * A data type big enough to bitmap all Linux possible cpus.
+ * The bitmap size is defined as 1024 cpus in the Linux 2.4 and 2.6 man pages
+ * for sched_getaffinity() and sched_getaffinity().
+ */
+#define	LX_NCPU		(1024)
+#define	LX_AFF_ULONGS	(LX_NCPU / (8 * sizeof (ulong_t)))
+typedef ulong_t lx_affmask_t[LX_AFF_ULONGS];
+
+/* Max. length of kernel version string */
+#define	LX_VERS_MAX	16
+
+/*
+ * Flag values for uc_brand_data[0] in the ucontext_t:
+ */
+#define	LX_UC_STACK_NATIVE	0x00001
+#define	LX_UC_STACK_BRAND	0x00002
+#define	LX_UC_RESTORE_NATIVE_SP	0x00010
+#define	LX_UC_FRAME_IS_SYSCALL	0x00100
+#define	LX_UC_RESTART_SYSCALL	0x01000
+#define	LX_UC_IGNORE_LINK	0x10000
+
+#ifdef	_KERNEL
+
+typedef struct lx_lwp_data lx_lwp_data_t;
+
+/*
+ * Flag values for "lxpa_flags" on a ptrace(2) accord.
+ */
+typedef enum lx_accord_flags {
+	LX_ACC_TOMBSTONE = 0x01
+} lx_accord_flags_t;
+
+/*
+ * Flags values for "br_ptrace_flags" in the LWP-specific data.
+ */
+typedef enum lx_ptrace_state {
+	LX_PTRACE_SYSCALL = 0x01,
+	LX_PTRACE_EXITING = 0x02,
+	LX_PTRACE_STOPPING = 0x04,
+	LX_PTRACE_INHERIT = 0x08,
+	LX_PTRACE_STOPPED = 0x10,
+	LX_PTRACE_PARENT_WAIT = 0x20,
+	LX_PTRACE_CLDPEND = 0x40,
+	LX_PTRACE_CLONING = 0x80,
+	LX_PTRACE_WAITPEND = 0x100
+} lx_ptrace_state_t;
+
+/*
+ * A ptrace(2) accord represents the relationship between a tracer LWP and the
+ * set of LWPs that it is tracing: the tracees.  This data structure belongs
+ * primarily to the tracer, but is reference counted so that it may be freed by
+ * whoever references it last.
+ */
+typedef struct lx_ptrace_accord {
+	kmutex_t		lxpa_lock;
+	uint_t			lxpa_refcnt;
+	lx_accord_flags_t	lxpa_flags;
+
+	/*
+	 * The tracer must hold "pidlock" while clearing these fields for
+	 * exclusion of waitid(), etc.
+	 */
+	lx_lwp_data_t		*lxpa_tracer;
+	kcondvar_t		*lxpa_cvp;
+
+	/*
+	 * The "lxpa_tracees_lock" mutex protects the tracee list.
+	 */
+	kmutex_t		lxpa_tracees_lock;
+	list_t			lxpa_tracees;
+} lx_ptrace_accord_t;
+
+/*
+ * These values are stored in the per-LWP data for a tracee when it is attached
+ * to a tracer.  They record the method that was used to attach.
+ */
+typedef enum lx_ptrace_attach {
+	LX_PTA_NONE = 0x00,		/* not attached */
+	LX_PTA_ATTACH = 0x01,		/* due to tracer using PTRACE_ATTACH */
+	LX_PTA_TRACEME = 0x02,		/* due to child using PTRACE_TRACEME */
+	LX_PTA_INHERIT_CLONE = 0x04,	/* due to PTRACE_CLONE clone(2) flag */
+	LX_PTA_INHERIT_OPTIONS = 0x08	/* due to PTRACE_SETOPTIONS options */
+} lx_ptrace_attach_t;
+
+typedef enum lx_stack_mode {
+	LX_STACK_MODE_PREINIT = 0,
+	LX_STACK_MODE_INIT,
+	LX_STACK_MODE_NATIVE,
+	LX_STACK_MODE_BRAND
+} lx_stack_mode_t;
+
+struct lx_pid {
+	pid_t	s_pid;			/* the SunOS pid and ... */
+	id_t	s_tid;			/* ... tid pair */
+	pid_t	l_pid;			/* the corresponding linux pid */
+	time_t	l_start;		/* birthday of this pid */
+	struct pid *l_pidp;
+	struct lx_pid *stol_next;	/* link in stol hash table */
+	struct lx_pid *ltos_next;	/* link in ltos hash table */
+};
+
+/*
+ * lx-specific data in the klwp_t
+ */
+struct lx_lwp_data {
+	uint_t	br_lwp_flags;		/* misc. flags */
+	klwp_t	*br_lwp;		/* back pointer to container lwp */
+	int	br_signal;		/* signal to send to parent when */
+					/* clone()'ed child terminates */
+	int	br_exitwhy;		/* reason for thread (process) exit */
+	int	br_exitwhat;		/* exit code / killing signal */
+	lx_affmask_t br_affinitymask;	/* bitmask of CPU sched affinities */
+	struct user_desc br_tls[LX_TLSNUM];
+			/* descriptors used by libc for TLS */
+	ulong_t	br_lx_fsbase;		/* lx fsbase for 64-bit thread ptr */
+	ulong_t	br_ntv_fsbase;		/* native fsbase 64-bit thread ptr */
+	ulong_t	br_lx_gsbase;		/* lx user-land gsbase */
+	ulong_t	br_ntv_gsbase;		/* native user-land gsbase */
+	pid_t	br_pid;			/* converted pid for this thread */
+	pid_t	br_tgid;		/* thread group ID for this thread */
+	pid_t	br_ppid;		/* parent pid for this thread */
+	id_t	br_ptid;		/* parent tid for this thread */
+	void	*br_clear_ctidp;	/* clone thread id ptr */
+	void	*br_set_ctidp;		/* clone thread id ptr */
+	void	*br_robust_list;	/* robust lock list, if any */
+
+	/*
+	 * The following struct is used by some system calls to pass extra
+	 * flags into the kernel without impinging on the namespace for
+	 * illumos.
+	 */
+	void	*br_scall_args;
+	int	br_args_size; /* size in bytes of br_scall_args */
+
+	boolean_t br_waitid_emulate;
+	int br_waitid_flags;
+
+	lx_ptrace_state_t br_ptrace_flags; /* ptrace state for this LWP */
+	lx_ptrace_options_t br_ptrace_options; /* PTRACE_SETOPTIONS options */
+	lx_ptrace_options_t br_ptrace_clone_option; /* current clone(2) type */
+
+	lx_ptrace_attach_t br_ptrace_attach; /* how did we get attached */
+	lx_ptrace_accord_t *br_ptrace_accord; /* accord for this tracer LWP */
+	lx_ptrace_accord_t *br_ptrace_tracer; /* accord tracing this LWP */
+	list_node_t br_ptrace_linkage;	/* linkage for lxpa_tracees list */
+
+	ushort_t br_ptrace_whystop; 	/* stop reason, 0 for no stop */
+	ushort_t br_ptrace_whatstop;	/* stop sub-reason */
+
+	int32_t br_ptrace_stopsig;	/* stop signal, 0 for no signal */
+	uintptr_t br_ptrace_stopucp;	/* usermode ucontext_t pointer */
+
+	uint_t	br_ptrace_event;
+	ulong_t	br_ptrace_eventmsg;
+
+	int	br_syscall_num;		/* current system call number */
+	boolean_t br_syscall_restart;	/* should restart on EINTR */
+
+	/*
+	 * Store the LX_STACK_MODE for this LWP, and the current extent of the
+	 * native (emulation) stack.  This is similar, in principle, to the
+	 * sigaltstack mechanism for signal handling.  We also use this mode
+	 * flag to determine how to process system calls from this LWP.
+	 */
+	lx_stack_mode_t	br_stack_mode;
+	uintptr_t br_ntv_stack;
+	uintptr_t br_ntv_stack_current;
+
+	/*
+	 * If this pid is set, we return it with getpid().  This allows the
+	 * thunking server to interpose on the pid returned to the Linux
+	 * syslog software.
+	 */
+	pid_t	br_lx_thunk_pid;
+
+	/*
+	 * If strict mode is enabled (via LX_STRICT in the environment), any
+	 * call to lx_unsupported() will set this boolean to B_TRUE.  This will
+	 * cause us to drop SIGSYS on the LWP as it attempts to return to
+	 * usermode.
+	 */
+	boolean_t br_strict_failure;
+
+	/*
+	 * Hold a pre-allocated lx_pid structure to be used during lx_initlwp.
+	 */
+	struct lx_pid *br_lpid;
+};
+
+/*
+ * Upper limit on br_args_size, low because this value can persist until
+ * overridden with another value, and the size is given from userland.
+ */
+#define	LX_BR_ARGS_SIZE_MAX	(1024)
+
+/* brand specific data */
+typedef struct lx_zone_data {
+	char lxzd_kernel_version[LX_VERS_MAX];
+	ksocket_t lxzd_ioctl_sock;
+} lx_zone_data_t;
+
+#define	BR_CPU_BOUND	0x0001
+
+#define	ttolxlwp(t)	((struct lx_lwp_data *)ttolwpbrand(t))
+#define	lwptolxlwp(l)	((struct lx_lwp_data *)lwptolwpbrand(l))
+#define	ttolxproc(t)	\
+	(((t)->t_procp->p_brand == &lx_brand) ? \
+	(struct lx_proc_data *)(t)->t_procp->p_brand_data : NULL)
+#define	ptolxproc(p)	\
+	(((p)->p_brand == &lx_brand) ? \
+	(struct lx_proc_data *)(p)->p_brand_data : NULL)
+#define	ztolxzd(z)		\
+	(((z)->zone_brand == &lx_brand) ?  \
+	(lx_zone_data_t *)(z)->zone_brand_data : NULL)
+
+/* Macro for converting to system call arguments. */
+#define	LX_ARGS(scall) ((struct lx_##scall##_args *)\
+	(ttolxlwp(curthread)->br_scall_args))
+
+/*
+ * Determine the upper bound on the system call number:
+ */
+#if defined(_LP64)
+#define	LX_MAX_SYSCALL(lwp)						\
+	((lwp_getdatamodel(lwp) == DATAMODEL_NATIVE) ?			\
+	    lx_nsysent64 : lx_nsysent32)
+#else
+#define	LX_MAX_SYSCALL(lwp)	lx_nsysent32
+#endif
+
+extern char *lx_get_zone_kern_version(zone_t *);
+
+extern void lx_lwp_set_native_stack_current(lx_lwp_data_t *, uintptr_t);
+extern void lx_divert(klwp_t *, uintptr_t);
+extern int lx_runexe(klwp_t *, void *);
+extern void lx_switch_to_native(klwp_t *);
+extern int lx_regs_to_userregs(lx_lwp_data_t *, void *);
+extern int lx_uc_to_userregs(lx_lwp_data_t *, void *, void *);
+extern int lx_userregs_to_regs(lx_lwp_data_t *lwpd, void *);
+extern int lx_userregs_to_uc(lx_lwp_data_t *lwpd, void *, void *);
+
+extern int lx_syscall_enter(void);
+extern int lx_syscall_return(klwp_t *, int, long);
+
+extern void lx_trace_sysenter(int, uintptr_t *);
+extern void lx_trace_sysreturn(int, long);
+
+extern void lx_emulate_user(klwp_t *, int, uintptr_t *);
+#if defined(_SYSCALL32_IMPL)
+extern void lx_emulate_user32(klwp_t *, int, uintptr_t *);
+#endif
+
+extern int lx_debug;
+#define	lx_print	if (lx_debug) printf
+
+extern void lx_pid_assign(kthread_t *, struct lx_pid *);
+extern void lx_pid_reassign(kthread_t *);
+extern void lx_pid_rele(pid_t, id_t);
+extern pid_t lx_lpid_to_spair(pid_t, pid_t *, id_t *);
+extern pid_t lx_lwp_ppid(klwp_t *, pid_t *, id_t *);
+extern void lx_pid_init(void);
+extern void lx_pid_fini(void);
+
+/*
+ * In-Kernel Linux System Call Description.
+ */
+typedef struct lx_sysent {
+	char	*sy_name;
+	long	(*sy_callc)();
+	char	sy_flags;
+	char	sy_narg;
+} lx_sysent_t;
+
+#if defined(_LP64)
+extern lx_sysent_t lx_sysent64[LX_NSYSCALLS + 1];
+extern int lx_nsysent64;
+#endif
+extern lx_sysent_t lx_sysent32[LX_NSYSCALLS + 1];
+extern int lx_nsysent32;
+
+#endif	/* _KERNEL */
+#endif /* _ASM */
+
+#ifdef	__cplusplus
+}
+#endif
+
+#endif	/* _LX_BRAND_H */
diff --git a/usr/src/uts/common/brand/lx/sys/lx_fcntl.h b/usr/src/uts/common/brand/lx/sys/lx_fcntl.h
new file mode 100644
index 0000000000..759831ea38
--- /dev/null
+++ b/usr/src/uts/common/brand/lx/sys/lx_fcntl.h
@@ -0,0 +1,153 @@
+/*
+ * CDDL HEADER START
+ *
+ * The contents of this file are subject to the terms of the
+ * Common Development and Distribution License (the "License").
+ * You may not use this file except in compliance with the License.
+ *
+ * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
+ * or http://www.opensolaris.org/os/licensing.
+ * See the License for the specific language governing permissions
+ * and limitations under the License.
+ *
+ * When distributing Covered Code, include this CDDL HEADER in each
+ * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
+ * If applicable, add the following below this CDDL HEADER, with the
+ * fields enclosed by brackets "[]" replaced with your own identifying
+ * information: Portions Copyright [yyyy] [name of copyright owner]
+ *
+ * CDDL HEADER END
+ */
+/*
+ * Copyright 2009 Sun Microsystems, Inc.  All rights reserved.
+ * Use is subject to license terms.
+ * Copyright 2015 Joyent, Inc.
+ */
+
+#ifndef _SYS_LX_FCNTL_H
+#define	_SYS_LX_FCNTL_H
+
+#ifdef	__cplusplus
+extern "C" {
+#endif
+
+/*
+ * Lx open/fcntl flags
+ */
+#define	LX_O_RDONLY		00
+#define	LX_O_WRONLY		01
+#define	LX_O_RDWR		02
+#define	LX_O_CREAT		0100
+#define	LX_O_EXCL		0200
+#define	LX_O_NOCTTY		0400
+#define	LX_O_TRUNC		01000
+#define	LX_O_APPEND		02000
+#define	LX_O_NONBLOCK		04000
+#define	LX_O_NDELAY		LX_O_NONBLOCK
+#define	LX_O_SYNC		010000
+#define	LX_O_FSYNC		LX_O_SYNC
+#define	LX_O_ASYNC		020000
+#define	LX_O_DIRECT		040000
+#define	LX_O_LARGEFILE		0100000
+#define	LX_O_DIRECTORY		0200000
+#define	LX_O_NOFOLLOW		0400000
+/* lx flag for pipe2 */
+#define	LX_O_CLOEXEC		02000000
+
+#define	LX_F_DUPFD		0
+#define	LX_F_GETFD		1
+#define	LX_F_SETFD		2
+#define	LX_F_GETFL		3
+#define	LX_F_SETFL		4
+#define	LX_F_GETLK		5
+#define	LX_F_SETLK		6
+#define	LX_F_SETLKW		7
+#define	LX_F_SETOWN		8
+#define	LX_F_GETOWN		9
+#define	LX_F_SETSIG		10
+#define	LX_F_GETSIG		11
+
+#define	LX_F_GETLK64		12
+#define	LX_F_SETLK64		13
+#define	LX_F_SETLKW64		14
+
+#define	LX_F_SETLEASE		1024
+#define	LX_F_GETLEASE		1025
+#define	LX_F_NOTIFY		1026
+#define	LX_F_CANCELLK		1029
+#define	LX_F_DUPFD_CLOEXEC	1030
+#define	LX_F_SETPIPE_SZ		1031
+#define	LX_F_GETPIPE_SZ		1032
+
+#define	LX_F_RDLCK		0
+#define	LX_F_WRLCK		1
+#define	LX_F_UNLCK		2
+
+/*
+ * Lx flock codes.
+ */
+#define	LX_NAME_MAX		255
+#define	LX_LOCK_SH		1	/* shared */
+#define	LX_LOCK_EX		2	/* exclusive */
+#define	LX_LOCK_NB		4	/* non-blocking */
+#define	LX_LOCK_UN		8	/* unlock */
+
+/*
+ * On Linux the constants AT_REMOVEDIR and AT_EACCESS have the same value.
+ * AT_REMOVEDIR is used only by unlinkat and AT_EACCESS is used only by
+ * faccessat.
+ */
+#define	LX_AT_FDCWD		(-100)
+#define	LX_AT_SYMLINK_NOFOLLOW	0x100
+#define	LX_AT_REMOVEDIR		0x200
+#define	LX_AT_EACCESS		0x200
+#define	LX_AT_SYMLINK_FOLLOW	0x400
+#define	LX_AT_NO_AUTOMOUNT	0x800
+#define	LX_AT_EMPTY_PATH	0x1000
+
+typedef struct lx_flock {
+	short		l_type;
+	short		l_whence;
+	long		l_start;
+	long		l_len;
+	int		l_pid;
+} lx_flock_t;
+
+typedef struct lx_flock64 {
+	short		l_type;
+	short		l_whence;
+	long long	l_start;
+	long long	l_len;
+	int		l_pid;
+} lx_flock64_t;
+
+#if defined(_KERNEL)
+
+/*
+ * 64-bit kernel view of 32-bit usermode structs.
+ */
+#pragma pack(4)
+typedef struct lx_flock32 {
+	int16_t		l_type;
+	int16_t		l_whence;
+	int32_t		l_start;
+	int32_t		l_len;
+	int32_t		l_pid;
+} lx_flock32_t;
+
+typedef struct lx_flock64_32 {
+	int16_t		l_type;
+	int16_t		l_whence;
+	int64_t		l_start;
+	int64_t		l_len;
+	int32_t		l_pid;
+} lx_flock64_32_t;
+#pragma pack()
+
+#endif /* _KERNEL && _SYSCALL32_IMPL */
+
+#ifdef	__cplusplus
+}
+#endif
+
+#endif	/* _SYS_LX_FCNTL_H */
diff --git a/usr/src/uts/common/brand/lx/sys/lx_futex.h b/usr/src/uts/common/brand/lx/sys/lx_futex.h
new file mode 100644
index 0000000000..a400b3bd83
--- /dev/null
+++ b/usr/src/uts/common/brand/lx/sys/lx_futex.h
@@ -0,0 +1,121 @@
+/*
+ * CDDL HEADER START
+ *
+ * The contents of this file are subject to the terms of the
+ * Common Development and Distribution License (the "License").
+ * You may not use this file except in compliance with the License.
+ *
+ * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
+ * or http://www.opensolaris.org/os/licensing.
+ * See the License for the specific language governing permissions
+ * and limitations under the License.
+ *
+ * When distributing Covered Code, include this CDDL HEADER in each
+ * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
+ * If applicable, add the following below this CDDL HEADER, with the
+ * fields enclosed by brackets "[]" replaced with your own identifying
+ * information: Portions Copyright [yyyy] [name of copyright owner]
+ *
+ * CDDL HEADER END
+ */
+/*
+ * Copyright 2007 Sun Microsystems, Inc.  All rights reserved.
+ * Use is subject to license terms.
+ */
+
+/*
+ * Copyright (c) 2014, Joyent, Inc.  All rights reserved.
+ */
+
+#ifndef _SYS_LX_FUTEX_H
+#define	_SYS_LX_FUTEX_H
+
+#ifdef	__cplusplus
+extern "C" {
+#endif
+
+#define	FUTEX_WAIT		0
+#define	FUTEX_WAKE		1
+#define	FUTEX_FD		2
+#define	FUTEX_REQUEUE		3
+#define	FUTEX_CMP_REQUEUE	4
+#define	FUTEX_WAKE_OP		5
+#define	FUTEX_LOCK_PI		6
+#define	FUTEX_UNLOCK_PI		7
+#define	FUTEX_TRYLOCK_PI	8
+#define	FUTEX_WAIT_BITSET	9
+#define	FUTEX_WAKE_BITSET	10
+#define	FUTEX_WAIT_REQUEUE_PI	11
+#define	FUTEX_CMP_REQUEUE_PI	12
+#define	FUTEX_MAX_CMD		FUTEX_CMP_REQUEUE_PI
+
+/*
+ * Flags that can be OR'd into a futex operation.
+ */
+#define	FUTEX_CMD_MASK		0x007f
+#define	FUTEX_PRIVATE_FLAG	0x0080
+#define	FUTEX_CLOCK_REALTIME	0x0100
+
+#define	FUTEX_BITSET_MATCH_ANY	0xffffffff
+/*
+ * FUTEX_WAKE_OP operations
+ */
+#define	FUTEX_OP_SET		0	/* *(int *)UADDR2 = OPARG; */
+#define	FUTEX_OP_ADD		1	/* *(int *)UADDR2 += OPARG; */
+#define	FUTEX_OP_OR		2	/* *(int *)UADDR2 |= OPARG; */
+#define	FUTEX_OP_ANDN		3	/* *(int *)UADDR2 &= ~OPARG; */
+#define	FUTEX_OP_XOR		4	/* *(int *)UADDR2 ^= OPARG; */
+
+/*
+ * FUTEX_WAKE_OP comparison operations
+ */
+#define	FUTEX_OP_CMP_EQ		0	/* if (oldval == CMPARG) wake */
+#define	FUTEX_OP_CMP_NE		1	/* if (oldval != CMPARG) wake */
+#define	FUTEX_OP_CMP_LT		2	/* if (oldval < CMPARG) wake */
+#define	FUTEX_OP_CMP_LE		3	/* if (oldval <= CMPARG) wake */
+#define	FUTEX_OP_CMP_GT		4	/* if (oldval > CMPARG) wake */
+#define	FUTEX_OP_CMP_GE		5	/* if (oldval >= CMPARG) wake */
+
+/*
+ * The encoding of the FUTEX_WAKE_OP operation in 32 bits:
+ *
+ *	+--+-- - --+-- - --+-- - --+-- - --+
+ *	|S |OP     |CMP    |OPARG  |CMPARG |
+ *	+--+-- - --+-- - --+-- - --+-- - --+
+ *	|31|30 - 28|27 - 24|23 - 12|11 -  0|
+ *
+ * The S bit denotes that the OPARG should be (1 << OPARG) instead of OPARG.
+ * (Yes, this whole thing is entirely absurd -- see the block comment in
+ * lx_futex.c for an explanation of this nonsense.)  Macros to extract the
+ * various components from the operation, given the above encoding:
+ */
+#define	FUTEX_OP_OP(x)		(((x) >> 28) & 7)
+#define	FUTEX_OP_CMP(x)		(((x) >> 24) & 15)
+#define	FUTEX_OP_OPARG(x)	(((x) >> 31) ? (1 << (((x) << 8) >> 20)) : \
+				((((x) << 8) >> 20)))
+#define	FUTEX_OP_CMPARG(x)	(((x) << 20) >> 20)
+
+#ifdef _KERNEL
+
+#define	FUTEX_WAITERS			0x80000000
+#define	FUTEX_OWNER_DIED		0x40000000
+#define	FUTEX_TID_MASK			0x3fffffff
+
+#define	FUTEX_ROBUST_LOCK_PI		1
+#define	FUTEX_ROBUST_LIST_LIMIT		2048
+
+extern long lx_futex(uintptr_t addr, int cmd, int val, uintptr_t lx_timeout,
+    uintptr_t addr2, int val2);
+extern void lx_futex_init(void);
+extern int lx_futex_fini(void);
+extern long lx_set_robust_list(void *listp, size_t len);
+extern long lx_get_robust_list(pid_t pid, void **listp, size_t *lenp);
+extern void lx_futex_robust_exit(uintptr_t addr, uint32_t tid);
+
+#endif /* _KERNEL */
+
+#ifdef	__cplusplus
+}
+#endif
+
+#endif	/* _SYS_LX_FUTEX_H */
diff --git a/usr/src/uts/common/brand/lx/sys/lx_impl.h b/usr/src/uts/common/brand/lx/sys/lx_impl.h
new file mode 100644
index 0000000000..03b9d43038
--- /dev/null
+++ b/usr/src/uts/common/brand/lx/sys/lx_impl.h
@@ -0,0 +1,52 @@
+/*
+ * CDDL HEADER START
+ *
+ * The contents of this file are subject to the terms of the
+ * Common Development and Distribution License (the "License").
+ * You may not use this file except in compliance with the License.
+ *
+ * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
+ * or http://www.opensolaris.org/os/licensing.
+ * See the License for the specific language governing permissions
+ * and limitations under the License.
+ *
+ * When distributing Covered Code, include this CDDL HEADER in each
+ * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
+ * If applicable, add the following below this CDDL HEADER, with the
+ * fields enclosed by brackets "[]" replaced with your own identifying
+ * information: Portions Copyright [yyyy] [name of copyright owner]
+ *
+ * CDDL HEADER END
+ */
+/*
+ * Copyright 2006 Sun Microsystems, Inc.  All rights reserved.
+ * Use is subject to license terms.
+ * Copyright 2014 Joyent, Inc.  All rights reserved.
+ */
+
+#ifndef	_LX_IMPL_H
+#define	_LX_IMPL_H
+
+#include <sys/types.h>
+
+#ifdef	__cplusplus
+extern "C" {
+#endif
+
+typedef void (lx_systrace_f)(ulong_t, ulong_t, ulong_t, ulong_t, ulong_t,
+    ulong_t, ulong_t);
+
+
+extern lx_systrace_f *lx_systrace_entry_ptr;
+extern lx_systrace_f *lx_systrace_return_ptr;
+
+extern void lx_brand_systrace_enable(void);
+extern void lx_brand_systrace_disable(void);
+
+extern void lx_unsupported(char *);
+
+#ifdef	__cplusplus
+}
+#endif
+
+#endif	/* _LX_IMPL_H */
diff --git a/usr/src/uts/common/brand/lx/sys/lx_ldt.h b/usr/src/uts/common/brand/lx/sys/lx_ldt.h
new file mode 100644
index 0000000000..825933e86c
--- /dev/null
+++ b/usr/src/uts/common/brand/lx/sys/lx_ldt.h
@@ -0,0 +1,91 @@
+/*
+ * CDDL HEADER START
+ *
+ * The contents of this file are subject to the terms of the
+ * Common Development and Distribution License (the "License").
+ * You may not use this file except in compliance with the License.
+ *
+ * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
+ * or http://www.opensolaris.org/os/licensing.
+ * See the License for the specific language governing permissions
+ * and limitations under the License.
+ *
+ * When distributing Covered Code, include this CDDL HEADER in each
+ * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
+ * If applicable, add the following below this CDDL HEADER, with the
+ * fields enclosed by brackets "[]" replaced with your own identifying
+ * information: Portions Copyright [yyyy] [name of copyright owner]
+ *
+ * CDDL HEADER END
+ */
+/*
+ * Copyright 2006 Sun Microsystems, Inc.  All rights reserved.
+ * Use is subject to license terms.
+ */
+/*
+ * Copyright 2014 Joyent, Inc.  All rights reserved.
+ */
+
+#ifndef _SYS_LINUX_LDT_H
+#define	_SYS_LINUX_LDT_H
+
+#include <sys/segments.h>
+
+#ifdef	__cplusplus
+extern "C" {
+#endif
+
+struct ldt_info {
+	uint_t	entry_number;
+	uint_t	base_addr;
+	uint_t	limit;
+	uint_t	seg_32bit:1,
+		contents:2,
+		read_exec_only:1,
+		limit_in_pages:1,
+		seg_not_present:1,
+		useable:1;
+};
+
+#define	LDT_INFO_EMPTY(info)						\
+	((info)->base_addr == 0 && (info)->limit == 0 &&		\
+	(info)->contents == 0 && (info)->read_exec_only == 1 &&		\
+	(info)->seg_32bit == 0 && (info)->limit_in_pages == 0 &&	\
+	(info)->seg_not_present == 1 && (info)->useable == 0)
+
+#if defined(__amd64)
+#define	SETMODE(desc)	(desc)->usd_long = SDP_SHORT;
+#else
+#define	SETMODE(desc)
+#endif
+
+#define	LDT_INFO_TO_DESC(info, desc)	{				\
+	USEGD_SETBASE(desc, (info)->base_addr);				\
+	USEGD_SETLIMIT(desc, (info)->limit);				\
+	(desc)->usd_type = ((info)->contents << 2) | 			\
+	    ((info)->read_exec_only ^ 1) << 1 | 0x10;			\
+	(desc)->usd_dpl = SEL_UPL;					\
+	(desc)->usd_p = (info)->seg_not_present ^ 1;			\
+	(desc)->usd_def32 = (info)->seg_32bit;				\
+	(desc)->usd_gran = (info)->limit_in_pages;			\
+	(desc)->usd_avl = (info)->useable;				\
+	SETMODE(desc);							\
+}
+
+#define	DESC_TO_LDT_INFO(desc, info)	{				\
+	bzero((info), sizeof (*(info)));				\
+	(info)->base_addr = USEGD_GETBASE(desc);			\
+	(info)->limit = USEGD_GETLIMIT(desc);				\
+	(info)->seg_not_present = (desc)->usd_p ^ 1;			\
+	(info)->contents = ((desc)->usd_type >> 2) & 3;			\
+	(info)->read_exec_only = (((desc)->usd_type >> 1) & 1) ^ 1;	\
+	(info)->seg_32bit = (desc)->usd_def32;				\
+	(info)->limit_in_pages = (desc)->usd_gran;			\
+	(info)->useable = (desc)->usd_avl;				\
+}
+
+#ifdef	__cplusplus
+}
+#endif
+
+#endif	/* _SYS_LINUX_LDT_H */
diff --git a/usr/src/uts/common/brand/lx/sys/lx_misc.h b/usr/src/uts/common/brand/lx/sys/lx_misc.h
new file mode 100644
index 0000000000..a3df150eef
--- /dev/null
+++ b/usr/src/uts/common/brand/lx/sys/lx_misc.h
@@ -0,0 +1,92 @@
+/*
+ * This file and its contents are supplied under the terms of the
+ * Common Development and Distribution License ("CDDL"), version 1.0.
+ * You may only use this file in accordance with the terms of version
+ * 1.0 of the CDDL.
+ *
+ * A full copy of the text of the CDDL should have accompanied this
+ * source.  A copy of the CDDL is also available via the Internet at
+ * http://www.illumos.org/license/CDDL.
+ */
+
+/*
+ * Copyright 2015 Joyent, Inc.  All rights reserved.
+ */
+
+#ifndef _SYS__LX_MISC_H
+#define	_SYS__LX_MISC_H
+
+#ifdef	__cplusplus
+extern "C" {
+#endif
+
+#include <sys/lx_brand.h>
+
+#ifdef _KERNEL
+
+extern void lx_setrval(klwp_t *, int, int);
+extern void lx_exec();
+extern void lx_exitlwp(klwp_t *);
+extern void lx_freelwp(klwp_t *);
+extern void *lx_lwpdata_alloc(proc_t *);
+extern void lx_lwpdata_free(void *);
+extern void lx_initlwp(klwp_t *, void *);
+extern void lx_forklwp(klwp_t *, klwp_t *);
+
+extern void lx_set_gdt(int, user_desc_t *);
+extern void lx_clear_gdt(int);
+
+extern longlong_t lx_nosys();
+
+extern greg_t lx_fixsegreg(greg_t, model_t);
+extern uintptr_t lx_fsbase(klwp_t *, uintptr_t);
+extern void lx_exit_with_sig(proc_t *, sigqueue_t *);
+extern boolean_t lx_wait_filter(proc_t *, proc_t *);
+
+typedef enum lx_if_action {
+	LX_IF_FROMNATIVE,
+	LX_IF_TONATIVE
+} lx_if_action_t;
+
+/* Linux ARP protocol hardware identifiers */
+#define	LX_ARPHRD_ETHER		1	/* Ethernet */
+#define	LX_ARPHRD_LOOPBACK	772	/* Loopback */
+#define	LX_ARPHRD_VOID		0xffff	/* Unknown */
+
+extern void lx_ifname_convert(char *, lx_if_action_t);
+extern void lx_ifflags_convert(uint64_t *, lx_if_action_t);
+extern void lx_stol_hwaddr(const struct sockaddr_dl *, struct sockaddr *,
+    int *);
+
+extern boolean_t lx_ptrace_stop(ushort_t);
+extern void lx_stop_notify(proc_t *, klwp_t *, ushort_t, ushort_t);
+extern void lx_ptrace_init(void);
+extern void lx_ptrace_fini(void);
+extern int lx_ptrace_kernel(int, pid_t, uintptr_t, uintptr_t);
+extern int lx_waitid_helper(idtype_t, id_t, k_siginfo_t *, int, boolean_t *,
+    int *);
+extern void lx_ptrace_exit(proc_t *, klwp_t *);
+extern void lx_ptrace_inherit_tracer(lx_lwp_data_t *, lx_lwp_data_t *);
+extern int lx_ptrace_stop_for_option(int, boolean_t, ulong_t, uintptr_t);
+extern int lx_ptrace_set_clone_inherit(int, boolean_t);
+extern int lx_sigcld_repost(proc_t *, sigqueue_t *);
+extern int lx_ptrace_issig_stop(proc_t *, klwp_t *);
+extern boolean_t lx_ptrace_sig_ignorable(proc_t *, int);
+
+extern int lx_helper_clone(int64_t *, int, void *, void *, void *);
+extern int lx_helper_setgroups(int, gid_t *);
+extern int lx_helper_rt_sigqueueinfo(pid_t, int, siginfo_t *);
+extern int lx_helper_rt_tgsigqueueinfo(pid_t, pid_t, int, siginfo_t *);
+
+extern boolean_t lx_vsyscall_iscall(klwp_t *, uintptr_t, int *);
+extern void lx_vsyscall_enter(proc_t *, klwp_t *, int);
+
+extern void lx_check_strict_failure(lx_lwp_data_t *);
+
+#endif
+
+#ifdef	__cplusplus
+}
+#endif
+
+#endif	/* _SYS__LX_MISC_H */
diff --git a/usr/src/uts/common/brand/lx/sys/lx_ptm.h b/usr/src/uts/common/brand/lx/sys/lx_ptm.h
new file mode 100644
index 0000000000..74bbc939a3
--- /dev/null
+++ b/usr/src/uts/common/brand/lx/sys/lx_ptm.h
@@ -0,0 +1,44 @@
+/*
+ * CDDL HEADER START
+ *
+ * The contents of this file are subject to the terms of the
+ * Common Development and Distribution License (the "License").
+ * You may not use this file except in compliance with the License.
+ *
+ * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
+ * or http://www.opensolaris.org/os/licensing.
+ * See the License for the specific language governing permissions
+ * and limitations under the License.
+ *
+ * When distributing Covered Code, include this CDDL HEADER in each
+ * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
+ * If applicable, add the following below this CDDL HEADER, with the
+ * fields enclosed by brackets "[]" replaced with your own identifying
+ * information: Portions Copyright [yyyy] [name of copyright owner]
+ *
+ * CDDL HEADER END
+ */
+/*
+ * Copyright 2006 Sun Microsystems, Inc.  All rights reserved.
+ * Use is subject to license terms.
+ */
+
+#ifndef	_SYS_PTM_LINUX_H
+#define	_SYS_PTM_LINUX_H
+
+#pragma ident	"%Z%%M%	%I%	%E% SMI"
+
+#ifdef	__cplusplus
+extern "C" {
+#endif
+
+#define	LX_PTM_DRV		"lx_ptm"
+#define	LX_PTM_MINOR_NODE	"lx_ptmajor"
+
+#define	LX_PTM_DEV_TO_PTS(dev)	(getminor(dev) - 1)
+
+#ifdef	__cplusplus
+}
+#endif
+
+#endif	/* _SYS_PTM_LINUX_H */
diff --git a/usr/src/uts/common/brand/lx/sys/lx_sched.h b/usr/src/uts/common/brand/lx/sys/lx_sched.h
new file mode 100644
index 0000000000..b0ae748f3c
--- /dev/null
+++ b/usr/src/uts/common/brand/lx/sys/lx_sched.h
@@ -0,0 +1,60 @@
+/*
+ * CDDL HEADER START
+ *
+ * The contents of this file are subject to the terms of the
+ * Common Development and Distribution License (the "License").
+ * You may not use this file except in compliance with the License.
+ *
+ * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
+ * or http://www.opensolaris.org/os/licensing.
+ * See the License for the specific language governing permissions
+ * and limitations under the License.
+ *
+ * When distributing Covered Code, include this CDDL HEADER in each
+ * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
+ * If applicable, add the following below this CDDL HEADER, with the
+ * fields enclosed by brackets "[]" replaced with your own identifying
+ * information: Portions Copyright [yyyy] [name of copyright owner]
+ *
+ * CDDL HEADER END
+ */
+/*
+ * Copyright 2006 Sun Microsystems, Inc.  All rights reserved.
+ * Use is subject to license terms.
+ */
+
+#ifndef _SYS_LINUX_SCHED_H
+#define	_SYS_LINUX_SCHED_H
+
+#pragma ident	"%Z%%M%	%I%	%E% SMI"
+
+#include <sys/procset.h>
+#include <sys/priocntl.h>
+
+#ifdef	__cplusplus
+extern "C" {
+#endif
+
+/*
+ * Linux scheduler policies.
+ */
+#define	LX_SCHED_OTHER		0
+#define	LX_SCHED_FIFO		1
+#define	LX_SCHED_RR		2
+
+#define	LX_PRI_MAX		99
+
+typedef	int l_pid_t;
+
+struct lx_sched_param {
+	int	lx_sched_prio;
+};
+
+extern int sched_setprocset(procset_t *, l_pid_t);
+extern long do_priocntlsys(int, procset_t *, void *);
+
+#ifdef	__cplusplus
+}
+#endif
+
+#endif	/* _SYS_LINUX_SCHED_H */
diff --git a/usr/src/uts/common/brand/lx/sys/lx_siginfo.h b/usr/src/uts/common/brand/lx/sys/lx_siginfo.h
new file mode 100644
index 0000000000..9f606b614f
--- /dev/null
+++ b/usr/src/uts/common/brand/lx/sys/lx_siginfo.h
@@ -0,0 +1,190 @@
+/*
+ * CDDL HEADER START
+ *
+ * The contents of this file are subject to the terms of the
+ * Common Development and Distribution License (the "License").
+ * You may not use this file except in compliance with the License.
+ *
+ * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
+ * or http://www.opensolaris.org/os/licensing.
+ * See the License for the specific language governing permissions
+ * and limitations under the License.
+ *
+ * When distributing Covered Code, include this CDDL HEADER in each
+ * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
+ * If applicable, add the following below this CDDL HEADER, with the
+ * fields enclosed by brackets "[]" replaced with your own identifying
+ * information: Portions Copyright [yyyy] [name of copyright owner]
+ *
+ * CDDL HEADER END
+ */
+/*
+ * Copyright 2007 Sun Microsystems, Inc.  All rights reserved.
+ * Use is subject to license terms.
+ * Copyright 2015 Joyent, Inc.
+ */
+
+#ifndef _LX_SIGINFO_H
+#define	_LX_SIGINFO_H
+
+#include <sys/lx_types.h>
+
+#ifdef	__cplusplus
+extern "C" {
+#endif
+
+/*
+ * lx_siginfo_t lsi_code values
+ *
+ *	LX_SI_ASYNCNL:	Sent by asynch name lookup completion
+ *	LX_SI_DETHREAD:	Sent by execve() killing subsidiary threads
+ *	LX_SI_SIGIO:	Sent by queued SIGIO
+ *	LX_SI_ASYNCIO:	Sent by asynchronous I/O completion
+ *	LX_SI_MESGQ:	Sent by real time message queue state change
+ *	LX_SI_TIMER:	Sent by timer expiration
+ *	LX_SI_QUEUE:	Sent by sigqueue
+ *	LX_SI_USER:	Sent by kill, sigsend, raise, etc.
+ *	LX_SI_KERNEL:	Sent by kernel
+ *	LX_SI_CODE_NOT_EXIST: Error code. When translating from Linux to
+ *	    illumos errors, if there is no translation available, this value
+ *	    should be used. This value should have no meaning as an si_code in
+ *	    illumos or Linux.
+ *
+ * At present, LX_SI_ASYNCNL, LX_SI_DETHREAD, and LX_SI_SIGIO are unused by
+ * BrandZ.
+ */
+#define	LX_SI_CODE_NOT_EXIST	(-61)
+#define	LX_SI_ASYNCNL		(-60)
+#define	LX_SI_DETHREAD		(-7)
+#define	LX_SI_TKILL		(-6)
+#define	LX_SI_SIGIO		(-5)
+#define	LX_SI_ASYNCIO		(-4)
+#define	LX_SI_MESGQ		(-3)
+#define	LX_SI_TIMER		(-2)
+#define	LX_SI_QUEUE		(-1)
+#define	LX_SI_USER		(0)
+#define	LX_SI_KERNEL		(0x80)
+
+#define	LX_SI_MAX_SIZE		128
+#define	LX_SI_PAD_SIZE_32	((LX_SI_MAX_SIZE / sizeof (int)) - 3)
+#define	LX_SI_PAD_SIZE_64	((LX_SI_MAX_SIZE / sizeof (int)) - 4)
+
+#if defined(_LP64)
+/*
+ * Because of the odd number (3) of ints before the union, we need to account
+ * for the smaller padding needed on x64 due to the union being offset to an 8
+ * byte boundary.
+ */
+#define	LX_SI_PAD_SIZE		LX_SI_PAD_SIZE_64
+#else
+#define	LX_SI_PAD_SIZE		LX_SI_PAD_SIZE_32
+#endif
+
+typedef struct lx_siginfo {
+	int lsi_signo;
+	int lsi_errno;
+	int lsi_code;
+	union {
+		int _pad[LX_SI_PAD_SIZE];
+
+		struct {
+			pid_t _pid;
+			lx_uid16_t _uid;
+		} _kill;
+
+		struct {
+			uint_t _timer1;
+			uint_t _timer2;
+		} _timer;
+
+		struct {
+			pid_t _pid;
+			lx_uid16_t _uid;
+			union sigval _sigval;
+		} _rt;
+
+		struct {
+			pid_t _pid;
+			lx_uid16_t _uid;
+			int _status;
+			clock_t _utime;
+			clock_t _stime;
+		} _sigchld;
+
+		struct {
+			void *_addr;
+		} _sigfault;
+
+		struct {
+			int _band;
+			int _fd;
+		} _sigpoll;
+	} _sifields;
+} lx_siginfo_t;
+
+#if defined(_KERNEL) && defined(_SYSCALL32_IMPL)
+/*
+ * 64-bit kernel view of the 32-bit "lx_siginfo_t" object.
+ */
+#pragma pack(4)
+typedef struct lx_siginfo32 {
+	int lsi_signo;
+	int lsi_errno;
+	int lsi_code;
+	union {
+		int _pad[LX_SI_PAD_SIZE_32];
+
+		struct {
+			pid32_t _pid;
+			lx_uid16_t _uid;
+		} _kill;
+
+		struct {
+			uint_t _timer1;
+			uint_t _timer2;
+		} _timer;
+
+		struct {
+			pid32_t _pid;
+			lx_uid16_t _uid;
+			union sigval32 _sigval;
+		} _rt;
+
+		struct {
+			pid32_t _pid;
+			lx_uid16_t _uid;
+			int _status;
+			clock32_t _utime;
+			clock32_t _stime;
+		} _sigchld;
+
+		struct {
+			caddr32_t _addr;
+		} _sigfault;
+
+		struct {
+			int _band;
+			int _fd;
+		} _sigpoll;
+	} _sifields;
+} lx_siginfo32_t;
+#pragma pack()
+#endif /* defined(_KERNEL) && defined(_SYSCALL32_IMPL) */
+
+#define	lsi_pid		_sifields._kill._pid
+#define	lsi_uid		_sifields._kill._uid
+#define	lsi_status	_sifields._sigchld._status
+#define	lsi_utime	_sifields._sigchld._utime
+#define	lsi_stime	_sifields._sigchld._stime
+#define	lsi_value	_sifields._rt._sigval
+#define	lsi_int		_sifields._rt._sigval.sivalx_int
+#define	lsi_ptr		_sifields._rt._sigval.sivalx_ptr
+#define	lsi_addr	_sifields._sigfault._addr
+#define	lsi_band	_sifields._sigpoll._band
+#define	lsi_fd		_sifields._sigpoll._fd
+
+#ifdef	__cplusplus
+}
+#endif
+
+#endif	/* _LX_SIGINFO_H */
diff --git a/usr/src/uts/common/brand/lx/sys/lx_syscalls.h b/usr/src/uts/common/brand/lx/sys/lx_syscalls.h
new file mode 100644
index 0000000000..0f2977c47c
--- /dev/null
+++ b/usr/src/uts/common/brand/lx/sys/lx_syscalls.h
@@ -0,0 +1,128 @@
+/*
+ * CDDL HEADER START
+ *
+ * The contents of this file are subject to the terms of the
+ * Common Development and Distribution License (the "License").
+ * You may not use this file except in compliance with the License.
+ *
+ * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
+ * or http://www.opensolaris.org/os/licensing.
+ * See the License for the specific language governing permissions
+ * and limitations under the License.
+ *
+ * When distributing Covered Code, include this CDDL HEADER in each
+ * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
+ * If applicable, add the following below this CDDL HEADER, with the
+ * fields enclosed by brackets "[]" replaced with your own identifying
+ * information: Portions Copyright [yyyy] [name of copyright owner]
+ *
+ * CDDL HEADER END
+ */
+
+/*
+ * Copyright 2006 Sun Microsystems, Inc.  All rights reserved.
+ * Use is subject to license terms.
+ * Copyright 2015 Joyent, Inc.
+ */
+
+#ifndef _SYS_LINUX_SYSCALLS_H
+#define	_SYS_LINUX_SYSCALLS_H
+
+#ifdef	__cplusplus
+extern "C" {
+#endif
+
+#ifdef	_KERNEL
+
+extern long lx_arch_prctl();
+extern long lx_brk();
+extern long lx_chmod();
+extern long lx_chown();
+extern long lx_chown16();
+extern long lx_clock_getres();
+extern long lx_clock_gettime();
+extern long lx_clock_settime();
+extern long lx_fchmod();
+extern long lx_fchmodat();
+extern long lx_fchown();
+extern long lx_fchown16();
+extern long lx_fchownat();
+extern long lx_fcntl();
+extern long lx_fcntl64();
+extern long lx_futex();
+extern long lx_get_robust_list();
+extern long lx_get_thread_area();
+extern long lx_getcpu();
+extern long lx_getdents_32();
+extern long lx_getdents_64();
+extern long lx_getdents64();
+extern long lx_getpid();
+extern long lx_getppid();
+extern long lx_getrandom();
+extern long lx_getrlimit();
+extern long lx_gettid();
+extern long lx_gettimeofday();
+extern long lx_ioctl();
+extern long lx_kill();
+extern long lx_lchown();
+extern long lx_lchown16();
+extern long lx_mkdir();
+extern long lx_mkdirat();
+extern long lx_modify_ldt();
+extern long lx_oldgetrlimit();
+extern long lx_open();
+extern long lx_openat();
+extern long lx_pipe();
+extern long lx_pipe2();
+extern long lx_prlimit64();
+extern long lx_read();
+extern long lx_sched_getparam();
+extern long lx_sched_getscheduler();
+extern long lx_sched_rr_get_interval();
+extern long lx_sched_setparam();
+extern long lx_sched_setscheduler();
+extern long lx_sched_yield();
+extern long lx_set_robust_list();
+extern long lx_set_thread_area();
+extern long lx_set_tid_address();
+extern long lx_setresgid();
+extern long lx_setresgid16();
+extern long lx_setresuid();
+extern long lx_setresuid16();
+extern long lx_setrlimit();
+extern long lx_sysinfo32();
+extern long lx_sysinfo64();
+extern long lx_tgkill();
+extern long lx_time();
+extern long lx_tkill();
+extern long lx_wait4();
+extern long lx_waitid();
+extern long lx_waitpid();
+extern long lx_write();
+extern long lx_xattr();
+
+#if defined(_LP64)
+/*
+ * Linux vsyscall addresses:
+ */
+#define	LX_VSYS_gettimeofday	(uintptr_t)0xffffffffff600000
+#define	LX_VSYS_time		(uintptr_t)0xffffffffff600400
+#define	LX_VSYS_getcpu		(uintptr_t)0xffffffffff600800
+
+#define	LX_VSYSCALL_ADDR		(uintptr_t)0xffffffffff600000
+#define	LX_VSYSCALL_SIZE		(uintptr_t)0x1000
+/*
+ * System call numbers for vsyscall revectoring:
+ */
+#define	LX_SYS_gettimeofday	96
+#define	LX_SYS_time		201
+#define	LX_SYS_getcpu		309
+#endif
+
+#endif	/* _KERNEL */
+
+#ifdef	__cplusplus
+}
+#endif
+
+#endif	/* _SYS_LINUX_SYSCALLS_H */
diff --git a/usr/src/uts/common/brand/lx/sys/lx_types.h b/usr/src/uts/common/brand/lx/sys/lx_types.h
new file mode 100644
index 0000000000..922c412020
--- /dev/null
+++ b/usr/src/uts/common/brand/lx/sys/lx_types.h
@@ -0,0 +1,115 @@
+/*
+ * CDDL HEADER START
+ *
+ * The contents of this file are subject to the terms of the
+ * Common Development and Distribution License (the "License").
+ * You may not use this file except in compliance with the License.
+ *
+ * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
+ * or http://www.opensolaris.org/os/licensing.
+ * See the License for the specific language governing permissions
+ * and limitations under the License.
+ *
+ * When distributing Covered Code, include this CDDL HEADER in each
+ * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
+ * If applicable, add the following below this CDDL HEADER, with the
+ * fields enclosed by brackets "[]" replaced with your own identifying
+ * information: Portions Copyright [yyyy] [name of copyright owner]
+ *
+ * CDDL HEADER END
+ */
+/*
+ * Copyright 2006 Sun Microsystems, Inc.  All rights reserved.
+ * Use is subject to license terms.
+ * Copyright 2015 Joyent, Inc.  All rights reserved.
+ */
+
+#ifndef _SYS_LX_TYPES_H
+#define	_SYS_LX_TYPES_H
+
+#ifdef	__cplusplus
+extern "C" {
+#endif
+
+#ifndef	_KERNEL
+
+#define	SHRT_MIN	(-32768)	/* min value of a "short int" */
+#define	SHRT_MAX	32767		/* max value of a "short int" */
+#define	USHRT_MAX	65535		/* max of "unsigned short int" */
+#define	INT_MIN		(-2147483647-1) /* min value of an "int" */
+#define	INT_MAX		2147483647	/* max value of an "int" */
+#define	UINT_MAX	4294967295U	/* max value of an "unsigned int" */
+
+#if defined(_LP64)
+#define	LONG_MAX	9223372036854775807L
+#define	ULONG_MAX	18446744073709551615UL
+#else
+#define	LONG_MAX	2147483647L	/* max value of a 32-bit "long int" */
+#define	ULONG_MAX	4294967295UL	/* max value of a 32-bit "ulong int" */
+#endif
+
+#endif /* !_KERNEL */
+
+#define	LX_SYS_UTS_LN	65
+
+struct lx_utsname {
+	char	sysname[LX_SYS_UTS_LN];
+	char	nodename[LX_SYS_UTS_LN];
+	char	release[LX_SYS_UTS_LN];
+	char	version[LX_SYS_UTS_LN];
+	char	machine[LX_SYS_UTS_LN];
+	char	domainname[LX_SYS_UTS_LN];
+};
+
+typedef	uint64_t	lx_dev_t;
+typedef	uint16_t	lx_dev16_t;
+typedef	uint32_t	lx_ino_t;
+typedef	uint64_t	lx_ino64_t;
+typedef	uint32_t	lx_uid_t;
+typedef	uint16_t	lx_uid16_t;
+typedef	uint32_t	lx_gid_t;
+typedef	uint16_t	lx_gid16_t;
+typedef	uint32_t	lx_off_t;
+typedef	uint64_t	lx_off64_t;
+typedef	uint32_t	lx_blksize_t;
+typedef	uint32_t	lx_blkcnt_t;
+typedef	uint64_t	lx_blkcnt64_t;
+typedef	uint32_t	lx_mode_t;
+typedef	uint16_t	lx_mode16_t;
+
+#define	LX_UID16_TO_UID32(uid16)	\
+	(((uid16) == (lx_uid16_t)-1) ? ((lx_uid_t)-1) : (lx_uid_t)(uid16))
+
+#define	LX_GID16_TO_GID32(gid16)     \
+	(((gid16) == (lx_gid16_t)-1) ? ((lx_gid_t)-1) : (lx_gid_t)(gid16))
+
+/* Overflow values default to NFS nobody. */
+
+#define	UID16_OVERFLOW	((lx_uid16_t)65534)
+#define	GID16_OVERFLOW	((lx_gid16_t)65534)
+
+/*
+ * All IDs with high word non-zero are converted to default overflow values to
+ * avoid inadvertent truncation to zero (root) (!).
+ */
+#define	LX_UID32_TO_UID16(uid32)	\
+	((((uid32) & 0xffff0000) == 0)  ? ((lx_uid16_t)(uid32)) : \
+	    (((uid32) == ((lx_uid_t)-1)) ? ((lx_uid16_t)-1) : UID16_OVERFLOW))
+
+#define	LX_GID32_TO_GID16(gid32)	\
+	((((gid32) & 0xffff0000) == 0)  ? ((lx_gid16_t)(gid32)) : \
+	    (((gid32) == ((lx_gid_t)-1)) ? ((lx_gid16_t)-1) : GID16_OVERFLOW))
+
+struct lx_timespec {
+	time_t	ts_sec;
+	long	ts_nsec;
+};
+
+#define	LX_32TO64(lo, hi)	\
+	((uint64_t)((uint64_t)(lo) | ((uint64_t)(hi) << 32)))
+
+#ifdef	__cplusplus
+}
+#endif
+
+#endif	/* _SYS_LX_TYPES_H */
diff --git a/usr/src/uts/common/brand/lx/syscall/lx_brk.c b/usr/src/uts/common/brand/lx/syscall/lx_brk.c
new file mode 100644
index 0000000000..19a7577ac0
--- /dev/null
+++ b/usr/src/uts/common/brand/lx/syscall/lx_brk.c
@@ -0,0 +1,57 @@
+/*
+ * CDDL HEADER START
+ *
+ * The contents of this file are subject to the terms of the
+ * Common Development and Distribution License (the "License").
+ * You may not use this file except in compliance with the License.
+ *
+ * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
+ * or http://www.opensolaris.org/os/licensing.
+ * See the License for the specific language governing permissions
+ * and limitations under the License.
+ *
+ * When distributing Covered Code, include this CDDL HEADER in each
+ * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
+ * If applicable, add the following below this CDDL HEADER, with the
+ * fields enclosed by brackets "[]" replaced with your own identifying
+ * information: Portions Copyright [yyyy] [name of copyright owner]
+ *
+ * CDDL HEADER END
+ */
+/*
+ * Copyright 2006 Sun Microsystems, Inc.  All rights reserved.
+ * Use is subject to license terms.
+ */
+
+#pragma ident	"%Z%%M%	%I%	%E% SMI"
+
+#include <sys/types.h>
+#include <sys/systm.h>
+#include <sys/errno.h>
+
+/*
+ * The brk() system call needs to be in-kernel because Linux expects a call to
+ * brk(0) to return the current breakpoint.  In Solaris, the process breakpoint
+ * is setup and managed by libc.  Due to the way we link our libraries and the
+ * need for Linux to manage its own breakpoint, this has to remain in the
+ * kernel.
+ */
+extern int brk(caddr_t);
+
+long
+lx_brk(caddr_t nva)
+{
+	proc_t *p = curproc;
+	klwp_t *lwp = ttolwp(curthread);
+
+	if (nva != 0) {
+		(void) brk(nva);
+
+		/*
+		 * Despite claims to the contrary in the manpage, when Linux
+		 * brk() fails, errno is left unchanged.
+		 */
+		lwp->lwp_errno = 0;
+	}
+	return ((long)(p->p_brkbase + p->p_brksize));
+}
diff --git a/usr/src/uts/common/brand/lx/syscall/lx_chmod.c b/usr/src/uts/common/brand/lx/syscall/lx_chmod.c
new file mode 100644
index 0000000000..66a4e35a63
--- /dev/null
+++ b/usr/src/uts/common/brand/lx/syscall/lx_chmod.c
@@ -0,0 +1,70 @@
+/*
+ * This file and its contents are supplied under the terms of the
+ * Common Development and Distribution License ("CDDL"), version 1.0.
+ * You may only use this file in accordance with the terms of version
+ * 1.0 of the CDDL.
+ *
+ * A full copy of the text of the CDDL should have accompanied this
+ * source.  A copy of the CDDL is also available via the Internet at
+ * http://www.illumos.org/license/CDDL.
+ */
+
+/*
+ * Copyright 2015 Joyent, Inc.
+ */
+
+#include <sys/fcntl.h>
+#include <sys/thread.h>
+#include <sys/klwp.h>
+#include <sys/lx_brand.h>
+#include <sys/lx_fcntl.h>
+
+/*
+ * From "uts/common/syscall/chmod.c":
+ */
+extern int fchmodat(int, char *, int, int);
+
+static long
+lx_fchmodat_wrapper(int fd, char *path, int mode, int flag)
+{
+	long rval;
+
+	if (fd == LX_AT_FDCWD) {
+		fd = AT_FDCWD;
+	}
+
+	if ((rval = fchmodat(fd, path, mode, flag)) != 0) {
+		lx_proc_data_t *pd = ttolxproc(curthread);
+		klwp_t *lwp = ttolwp(curthread);
+
+		/*
+		 * If the process is in "install mode", return success
+		 * if the operation failed due to an absent file.
+		 */
+		if ((pd->l_flags & LX_PROC_INSTALL_MODE) &&
+		    lwp->lwp_errno == ENOENT) {
+			lwp->lwp_errno = 0;
+			return (0);
+		}
+	}
+
+	return (rval);
+}
+
+long
+lx_fchmodat(int fd, char *path, int mode)
+{
+	return (lx_fchmodat_wrapper(fd, path, mode, 0));
+}
+
+long
+lx_fchmod(int fd, int mode)
+{
+	return (lx_fchmodat_wrapper(fd, NULL, mode, 0));
+}
+
+long
+lx_chmod(char *path, int mode)
+{
+	return (lx_fchmodat_wrapper(AT_FDCWD, path, mode, 0));
+}
diff --git a/usr/src/uts/common/brand/lx/syscall/lx_chown.c b/usr/src/uts/common/brand/lx/syscall/lx_chown.c
new file mode 100644
index 0000000000..c0326ee619
--- /dev/null
+++ b/usr/src/uts/common/brand/lx/syscall/lx_chown.c
@@ -0,0 +1,137 @@
+/*
+ * This file and its contents are supplied under the terms of the
+ * Common Development and Distribution License ("CDDL"), version 1.0.
+ * You may only use this file in accordance with the terms of version
+ * 1.0 of the CDDL.
+ *
+ * A full copy of the text of the CDDL should have accompanied this
+ * source.  A copy of the CDDL is also available via the Internet at
+ * http://www.illumos.org/license/CDDL.
+ */
+
+/*
+ * Copyright 2015 Joyent, Inc.
+ */
+
+#include <sys/systm.h>
+#include <sys/fcntl.h>
+#include <sys/lx_brand.h>
+#include <sys/lx_fcntl.h>
+#include <sys/lx_types.h>
+
+/*
+ * From "uts/common/syscall/chown.c":
+ */
+extern int fchownat(int, char *, uid_t, gid_t, int);
+
+long
+lx_fchownat_wrapper(int fd, char *path, uid_t uid, gid_t gid, int native_flag)
+{
+	long rval;
+
+	if (fd == LX_AT_FDCWD) {
+		fd = AT_FDCWD;
+	}
+
+	if ((rval = fchownat(fd, path, uid, gid, native_flag)) != 0) {
+		lx_proc_data_t *pd = ttolxproc(curthread);
+		klwp_t *lwp = ttolwp(curthread);
+
+		/*
+		 * If the process is in "install mode", return success
+		 * if the operation failed due to an absent file.
+		 */
+		if ((pd->l_flags & LX_PROC_INSTALL_MODE) &&
+		    lwp->lwp_errno == ENOENT) {
+			lwp->lwp_errno = 0;
+			return (0);
+		}
+	}
+
+	return (rval);
+}
+
+long
+lx_fchownat(int fd, char *path, uid_t uid, gid_t gid, int flag)
+{
+	char c;
+	int native_flag = 0;
+
+	if (copyin(path, &c, sizeof (c)) != 0) {
+		return (set_errno(EFAULT));
+	}
+
+	if (flag & LX_AT_EMPTY_PATH) {
+		/*
+		 * According to fchownat(2), when AT_EMPTY_PATH is set: "if
+		 * path is an empty string, operate on the file referred to by
+		 * fd".  We pass NULL in place of the empty string, which
+		 * causes fchownat() to operate on the fd we passed without an
+		 * additional lookup.
+		 */
+		if (c == '\0') {
+			path = NULL;
+		}
+
+		flag &= ~LX_AT_EMPTY_PATH;
+	} else {
+		/*
+		 * Otherwise, a file with no filename obviously cannot be
+		 * present in the directory.
+		 */
+		if (c == '\0') {
+			return (set_errno(ENOENT));
+		}
+	}
+
+	if (flag & LX_AT_SYMLINK_NOFOLLOW) {
+		flag &= ~LX_AT_SYMLINK_NOFOLLOW;
+		native_flag |= AT_SYMLINK_NOFOLLOW;
+	}
+
+	if (flag != 0) {
+		return (set_errno(EINVAL));
+	}
+
+	return (lx_fchownat_wrapper(fd, path, uid, gid, native_flag));
+}
+
+long
+lx_fchown(int fd, uid_t uid, gid_t gid)
+{
+	return (lx_fchownat_wrapper(fd, NULL, uid, gid, 0));
+}
+
+long
+lx_lchown(char *path, uid_t uid, gid_t gid)
+{
+	return (lx_fchownat_wrapper(AT_FDCWD, path, uid, gid,
+	    AT_SYMLINK_NOFOLLOW));
+}
+
+long
+lx_chown(char *path, uid_t uid, gid_t gid)
+{
+	return (lx_fchownat_wrapper(AT_FDCWD, path, uid, gid, 0));
+}
+
+long
+lx_fchown16(int fd, lx_uid16_t uid, lx_gid16_t gid)
+{
+	return (lx_fchownat_wrapper(fd, NULL, LX_UID16_TO_UID32(uid),
+	    LX_GID16_TO_GID32(gid), 0));
+}
+
+long
+lx_lchown16(char *path, uid_t uid, gid_t gid)
+{
+	return (lx_fchownat_wrapper(AT_FDCWD, path, LX_UID16_TO_UID32(uid),
+	    LX_GID16_TO_GID32(gid), AT_SYMLINK_NOFOLLOW));
+}
+
+long
+lx_chown16(char *path, lx_uid16_t uid, lx_gid16_t gid)
+{
+	return (lx_fchownat_wrapper(AT_FDCWD, path, LX_UID16_TO_UID32(uid),
+	    LX_GID16_TO_GID32(gid), 0));
+}
diff --git a/usr/src/uts/common/brand/lx/syscall/lx_clone.c b/usr/src/uts/common/brand/lx/syscall/lx_clone.c
new file mode 100644
index 0000000000..50cdeaeab9
--- /dev/null
+++ b/usr/src/uts/common/brand/lx/syscall/lx_clone.c
@@ -0,0 +1,143 @@
+/*
+ * CDDL HEADER START
+ *
+ * The contents of this file are subject to the terms of the
+ * Common Development and Distribution License (the "License").
+ * You may not use this file except in compliance with the License.
+ *
+ * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
+ * or http://www.opensolaris.org/os/licensing.
+ * See the License for the specific language governing permissions
+ * and limitations under the License.
+ *
+ * When distributing Covered Code, include this CDDL HEADER in each
+ * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
+ * If applicable, add the following below this CDDL HEADER, with the
+ * fields enclosed by brackets "[]" replaced with your own identifying
+ * information: Portions Copyright [yyyy] [name of copyright owner]
+ *
+ * CDDL HEADER END
+ */
+/*
+ * Copyright 2006 Sun Microsystems, Inc.  All rights reserved.
+ * Use is subject to license terms.
+ * Copyright 2015 Joyent, Inc.
+ */
+
+#include <sys/types.h>
+#include <sys/systm.h>
+#include <sys/errno.h>
+#include <sys/brand.h>
+#include <sys/lx_brand.h>
+#include <sys/lx_ldt.h>
+#include <sys/lx_misc.h>
+#include <lx_signum.h>
+#include <lx_syscall.h>
+#include <sys/x86_archext.h>
+#include <sys/controlregs.h>
+
+/*
+ * Our lwp has already been created at this point, so this routine is
+ * responsible for setting up all the state needed to track this as a
+ * linux cloned thread.
+ */
+/* ARGSUSED */
+int
+lx_helper_clone(int64_t *rval, int flags, void *ptidp, void *tls, void *ctidp)
+{
+	struct lx_lwp_data *lwpd = ttolxlwp(curthread);
+	struct lx_proc_data *lproc = ttolxproc(curthread);
+	struct ldt_info info;
+	struct user_desc descr;
+	int tls_index;
+	int entry = -1;
+	int signo;
+
+	signo = flags & LX_CSIGNAL;
+	if (signo < 0 || signo > LX_NSIG)
+		return (set_errno(EINVAL));
+
+	if (!(flags & LX_CLONE_THREAD)) {
+		lproc->l_signal = signo;
+	} else {
+		if (flags & LX_CLONE_SETTLS) {
+			if (get_udatamodel() == DATAMODEL_ILP32) {
+				if (copyin((caddr_t)tls, &info, sizeof (info)))
+					return (set_errno(EFAULT));
+
+				if (LDT_INFO_EMPTY(&info))
+					return (set_errno(EINVAL));
+
+				entry = info.entry_number;
+				if (entry < GDT_TLSMIN || entry > GDT_TLSMAX)
+					return (set_errno(EINVAL));
+
+				tls_index = entry - GDT_TLSMIN;
+
+				/*
+				 * Convert the user-space structure into a real
+				 * x86 descriptor and copy it into this LWP's
+				 * TLS array.  We also load it into the GDT.
+				 */
+				LDT_INFO_TO_DESC(&info, &descr);
+				bcopy(&descr, &lwpd->br_tls[tls_index],
+				    sizeof (descr));
+				lx_set_gdt(entry, &lwpd->br_tls[tls_index]);
+			} else {
+				/*
+				 * Set the Linux %fsbase for this LWP.  We will
+				 * restore it the next time we return to Linux
+				 * via setcontext()/lx_restorecontext().
+				 */
+				lwpd->br_lx_fsbase = (uintptr_t)tls;
+			}
+		}
+
+		lwpd->br_clear_ctidp =
+		    (flags & LX_CLONE_CHILD_CLEARTID) ?  ctidp : NULL;
+
+		if (signo && ! (flags & LX_CLONE_DETACH))
+			lwpd->br_signal = signo;
+		else
+			lwpd->br_signal = 0;
+
+		if (flags & LX_CLONE_THREAD)
+			lwpd->br_tgid = curthread->t_procp->p_pid;
+
+		if (flags & LX_CLONE_PARENT)
+			lwpd->br_ppid = 0;
+
+		if ((flags & LX_CLONE_CHILD_SETTID) && (ctidp != NULL) &&
+		    (suword32(ctidp, lwpd->br_pid) != 0)) {
+			if (entry >= 0)
+				lx_clear_gdt(entry);
+			return (set_errno(EFAULT));
+		}
+		if ((flags & LX_CLONE_PARENT_SETTID) && (ptidp != NULL) &&
+		    (suword32(ptidp, lwpd->br_pid) != 0)) {
+			if (entry >= 0)
+				lx_clear_gdt(entry);
+			return (set_errno(EFAULT));
+		}
+	}
+
+	*rval = lwpd->br_pid;
+	return (0);
+}
+
+long
+lx_set_tid_address(int *tidp)
+{
+	struct lx_lwp_data *lwpd = ttolxlwp(curthread);
+	long rv;
+
+	lwpd->br_clear_ctidp = tidp;
+
+	if (curproc->p_pid == curproc->p_zone->zone_proc_initpid) {
+		rv = 1;
+	} else {
+		rv = lwpd->br_pid;
+	}
+
+	return (rv);
+}
diff --git a/usr/src/uts/common/brand/lx/syscall/lx_cpu.c b/usr/src/uts/common/brand/lx/syscall/lx_cpu.c
new file mode 100644
index 0000000000..ec8b7576d8
--- /dev/null
+++ b/usr/src/uts/common/brand/lx/syscall/lx_cpu.c
@@ -0,0 +1,35 @@
+/*
+ * This file and its contents are supplied under the terms of the
+ * Common Development and Distribution License ("CDDL"), version 1.0.
+ * You may only use this file in accordance with the terms of version
+ * 1.0 of the CDDL.
+ *
+ * A full copy of the text of the CDDL should have accompanied this
+ * source.  A copy of the CDDL is also available via the Internet at
+ * http://www.illumos.org/license/CDDL.
+ */
+
+/*
+ * Copyright 2015 Joyent, Inc.
+ */
+
+#include <sys/systm.h>
+#include <sys/thread.h>
+#include <sys/cpuvar.h>
+#include <sys/cmn_err.h>
+#include <sys/lx_impl.h>
+
+/*
+ * We support neither the second argument (NUMA node), nor the third (obsolete
+ * pre-2.6.24 caching functionality which was ultimately broken).
+ */
+long
+lx_getcpu(unsigned int *cpu, uintptr_t p2, uintptr_t p3)
+{
+	unsigned int curcpu = curthread->t_cpu->cpu_id;
+
+	if (copyout(&curcpu, cpu, sizeof (curcpu)) != 0)
+		return (set_errno(EFAULT));
+
+	return (0);
+}
diff --git a/usr/src/uts/common/brand/lx/syscall/lx_fcntl.c b/usr/src/uts/common/brand/lx/syscall/lx_fcntl.c
new file mode 100644
index 0000000000..d3615c704f
--- /dev/null
+++ b/usr/src/uts/common/brand/lx/syscall/lx_fcntl.c
@@ -0,0 +1,520 @@
+/*
+ * This file and its contents are supplied under the terms of the
+ * Common Development and Distribution License ("CDDL"), version 1.0.
+ * You may only use this file in accordance with the terms of version
+ * 1.0 of the CDDL.
+ *
+ * A full copy of the text of the CDDL should have accompanied this
+ * source.  A copy of the CDDL is also available via the Internet at
+ * http://www.illumos.org/license/CDDL.
+ */
+
+/*
+ * Copyright 2015 Joyent, Inc.
+ */
+
+#include <sys/systm.h>
+#include <sys/zone.h>
+#include <sys/types.h>
+#include <sys/filio.h>
+#include <sys/fcntl.h>
+#include <sys/stat.h>
+#include <sys/cmn_err.h>
+#include <sys/lx_impl.h>
+#include <sys/lx_brand.h>
+#include <sys/lx_fcntl.h>
+#include <sys/lx_misc.h>
+
+extern int fcntl(int, int, intptr_t);
+extern int flock_check(vnode_t *, flock64_t *, offset_t, offset_t);
+
+#define	LTOS_FLOCK(l, s)						\
+{									\
+	s->l_type = ltos_type(l->l_type);				\
+	s->l_whence = l->l_whence;					\
+	s->l_start = l->l_start;					\
+	s->l_len = l->l_len;						\
+	s->l_sysid = 0;			/* not defined in linux */	\
+	s->l_pid = (pid_t)l->l_pid;					\
+}
+
+#define	STOL_FLOCK(s, l)						\
+{									\
+	l->l_type = stol_type(s->l_type);				\
+	l->l_whence = s->l_whence;					\
+	l->l_start = s->l_start;					\
+	l->l_len = s->l_len;						\
+	l->l_pid = (int)s->l_pid;					\
+}
+
+static short
+ltos_type(short l_type)
+{
+	switch (l_type) {
+	case LX_F_RDLCK:
+		return (F_RDLCK);
+	case LX_F_WRLCK:
+		return (F_WRLCK);
+	case LX_F_UNLCK:
+		return (F_UNLCK);
+	default:
+		return (-1);
+	}
+}
+
+static short
+stol_type(short l_type)
+{
+	switch (l_type) {
+	case F_RDLCK:
+		return (LX_F_RDLCK);
+	case F_WRLCK:
+		return (LX_F_WRLCK);
+	case F_UNLCK:
+		return (LX_F_UNLCK);
+	default:
+		/* can't ever happen */
+		return (0);
+	}
+}
+
+static void
+ltos_flock(struct lx_flock *l, struct flock64 *s)
+{
+	LTOS_FLOCK(l, s)
+}
+
+static void
+stol_flock(struct flock64 *s, struct lx_flock *l)
+{
+	STOL_FLOCK(s, l)
+}
+
+static void
+ltos_flock64(struct lx_flock64_32 *l, struct flock64 *s)
+{
+	LTOS_FLOCK(l, s)
+}
+
+static void
+stol_flock64(struct flock64 *s, struct lx_flock64_32 *l)
+{
+	STOL_FLOCK(s, l)
+}
+
+static int
+lx_fcntl_getfl(int fd)
+{
+	int retval;
+	int rc;
+
+	retval = fcntl(fd, F_GETFL, 0);
+	if (ttolwp(curthread)->lwp_errno != 0)
+		return (ttolwp(curthread)->lwp_errno);
+
+	if ((retval & O_ACCMODE) == O_RDONLY)
+		rc = LX_O_RDONLY;
+	else if ((retval & O_ACCMODE) == O_WRONLY)
+		rc = LX_O_WRONLY;
+	else
+		rc = LX_O_RDWR;
+	/* O_NDELAY != O_NONBLOCK, so we need to check for both */
+	if (retval & O_NDELAY)
+		rc |= LX_O_NDELAY;
+	if (retval & O_NONBLOCK)
+		rc |= LX_O_NONBLOCK;
+	if (retval & O_APPEND)
+		rc |= LX_O_APPEND;
+	if (retval & O_SYNC)
+		rc |= LX_O_SYNC;
+	if (retval & O_LARGEFILE)
+		rc |= LX_O_LARGEFILE;
+	if (retval & FASYNC)
+		rc |= LX_O_ASYNC;
+
+	return (rc);
+}
+
+static int
+lx_fcntl_setfl(int fd, ulong_t arg)
+{
+	int new_arg;
+
+	new_arg = 0;
+	/* LX_O_NDELAY == LX_O_NONBLOCK, so we only check for one */
+	if (arg & LX_O_NDELAY)
+		new_arg |= O_NONBLOCK;
+	if (arg & LX_O_APPEND)
+		new_arg |= O_APPEND;
+	if (arg & LX_O_SYNC)
+		new_arg |= O_SYNC;
+	if (arg & LX_O_LARGEFILE)
+		new_arg |= O_LARGEFILE;
+	if (arg & LX_O_ASYNC)
+		new_arg |= FASYNC;
+
+	return (fcntl(fd, F_SETFL, new_arg));
+}
+
+static int
+lx_fcntl_common(int fd, int cmd, ulong_t arg)
+{
+	int		rc = 0;
+	pid_t		pid;
+	int		error;
+	int		rv;
+	int32_t		flag;
+	file_t		*fp;
+
+	/*
+	 * We depend on the call to fcntl to set the errno if necessary.
+	 */
+	ttolwp(curthread)->lwp_errno = 0;
+
+	switch (cmd) {
+	case LX_F_SETSIG:
+	case LX_F_GETSIG:
+	case LX_F_SETLEASE:
+	case LX_F_GETLEASE:
+	case LX_F_NOTIFY:
+	case LX_F_CANCELLK:
+	case LX_F_SETPIPE_SZ:
+	case LX_F_GETPIPE_SZ:
+		{
+			char buf[80];
+
+			(void) snprintf(buf, sizeof (buf),
+			    "unsupported fcntl command: %d", cmd);
+			lx_unsupported(buf);
+		}
+		return (set_errno(ENOTSUP));
+
+	case LX_F_DUPFD:
+		rc = fcntl(fd, F_DUPFD, arg);
+		break;
+
+	case LX_F_DUPFD_CLOEXEC:
+		rc = fcntl(fd, F_DUPFD_CLOEXEC, arg);
+		break;
+
+	case LX_F_GETFD:
+		rc = fcntl(fd, F_GETFD, 0);
+		break;
+
+	case LX_F_SETFD:
+		rc = fcntl(fd, F_SETFD, arg);
+		break;
+
+	case LX_F_GETFL:
+		rc = lx_fcntl_getfl(fd);
+		break;
+
+	case LX_F_SETFL:
+		rc = lx_fcntl_setfl(fd, arg);
+		break;
+
+	case LX_F_SETOWN:
+		pid = (pid_t)arg;
+		if (pid == 1) {
+			/* Setown for the init process uses the real pid. */
+			pid = curzone->zone_proc_initpid;
+		}
+
+		if ((fp = getf(fd)) == NULL)
+			return (set_errno(EBADF));
+
+		rv = 0;
+
+		flag = fp->f_flag | get_udatamodel() | FKIOCTL;
+		error = VOP_IOCTL(fp->f_vnode, FIOSETOWN, (intptr_t)&pid,
+		    flag, CRED(), &rv, NULL);
+		releasef(fd);
+		if (error != 0)
+			return (set_errno(error));
+
+		rc = 0;
+		break;
+
+	case LX_F_GETOWN:
+		if ((fp = getf(fd)) == NULL)
+			return (set_errno(EBADF));
+
+		rv = 0;
+
+		flag = fp->f_flag | get_udatamodel() | FKIOCTL;
+		error = VOP_IOCTL(fp->f_vnode, FIOGETOWN, (intptr_t)&pid,
+		    flag, CRED(), &rv, NULL);
+		releasef(fd);
+		if (error != 0)
+			return (set_errno(error));
+
+		if (pid == curzone->zone_proc_initpid) {
+			/* Getown for the init process returns 1. */
+			pid = 1;
+		}
+
+		rc = pid;
+		break;
+
+	default:
+		return (set_errno(EINVAL));
+	}
+
+	return (rc);
+}
+
+static int
+lx_fcntl_lock_cmd_to_s(int lx_cmd)
+{
+	switch (lx_cmd) {
+	case LX_F_GETLK:
+		return (F_GETLK);
+	case LX_F_SETLK:
+		return (F_SETLK);
+	case LX_F_SETLKW:
+		return (F_SETLKW);
+	case LX_F_GETLK64:
+		return (F_GETLK64);
+	case LX_F_SETLK64:
+		return (F_SETLK64);
+	case LX_F_SETLKW64:
+		return (F_SETLKW64);
+	default:
+		VERIFY(0);
+		/*NOTREACHED*/
+		return (0);
+	}
+}
+
+/*
+ * This is a pain but we can't re-use the fcntl code for locking since it does
+ * its own copyin/copyout for the flock struct. Since we have to convert the
+ * struct we have to do our own copyin/out. Thus we replicate the fcntl code for
+ * these 3 cmds. Luckily it's not much.
+ */
+static int
+lx_fcntl_lock(int fd, int lx_cmd, void *arg)
+{
+	int cmd;
+	int error = 0;
+	file_t *fp;
+	vnode_t *vp;
+	int flag;
+	offset_t maxoffset;
+	u_offset_t offset;
+	model_t datamodel;
+	lx_flock_t lxflk;
+	lx_flock64_32_t lxflk64;
+	struct flock64 bf;
+
+	if ((fp = getf(fd)) == NULL)
+		return (set_errno(EBADF));
+
+	maxoffset = MAXOFF_T;
+	datamodel = DATAMODEL_NATIVE;
+#if defined(_SYSCALL32_IMPL)
+	if ((datamodel = get_udatamodel()) == DATAMODEL_ILP32)
+		maxoffset = MAXOFF32_T;
+#endif
+	vp = fp->f_vnode;
+	flag = fp->f_flag;
+	offset = fp->f_offset;
+
+	cmd = lx_fcntl_lock_cmd_to_s(lx_cmd);
+
+	switch (cmd) {
+	case F_GETLK:
+	case F_SETLK:
+	case F_SETLKW:
+		if (datamodel == DATAMODEL_NATIVE) {
+			if (copyin(arg, &lxflk, sizeof (lx_flock_t)) != 0) {
+				error = EFAULT;
+				break;
+			}
+		}
+#if defined(_SYSCALL32_IMPL)
+		else {
+			lx_flock32_t lxflk32;
+
+			if (copyin(arg, &lxflk32, sizeof (lxflk32)) != 0) {
+				error = EFAULT;
+				break;
+			}
+
+			lxflk.l_type = lxflk32.l_type;
+			lxflk.l_whence = lxflk32.l_whence;
+			lxflk.l_start = (off64_t)lxflk32.l_start;
+			lxflk.l_len = (off64_t)lxflk32.l_len;
+			lxflk.l_pid = lxflk32.l_pid;
+		}
+#endif /* _SYSCALL32_IMPL */
+
+		ltos_flock(&lxflk, &bf);
+
+		if ((error = flock_check(vp, &bf, offset, maxoffset)) != 0)
+			break;
+
+		if ((error = VOP_FRLOCK(vp, cmd, &bf, flag, offset, NULL,
+		    fp->f_cred, NULL)) != 0)
+			break;
+
+		if (cmd != F_GETLK)
+			break;
+
+		/*
+		 * The command is GETLK, return result.
+		 */
+		stol_flock(&bf, &lxflk);
+
+		/*
+		 * If no lock is found, only the type field is changed.
+		 */
+		if (lxflk.l_type == LX_F_UNLCK) {
+			/* l_type always first entry, always a short */
+			if (copyout(&lxflk.l_type, &((lx_flock_t *)arg)->l_type,
+			    sizeof (lxflk.l_type)))
+				error = EFAULT;
+			break;
+		}
+
+		if (bf.l_start > maxoffset || bf.l_len > maxoffset) {
+			error = EOVERFLOW;
+			break;
+		}
+
+		if (datamodel == DATAMODEL_NATIVE) {
+			if (copyout(&lxflk, arg, sizeof (lxflk)) != 0) {
+				error = EFAULT;
+				break;
+			}
+		}
+#if defined(_SYSCALL32_IMPL)
+		else {
+			lx_flock32_t lxflk32;
+
+			if (bf.l_start > MAXOFF32_T || bf.l_len > MAXOFF32_T) {
+				error = EOVERFLOW;
+				break;
+			}
+
+			lxflk32.l_type = lxflk.l_type;
+			lxflk32.l_whence = lxflk.l_whence;
+			lxflk32.l_start = lxflk.l_start;
+			lxflk32.l_len = lxflk.l_len;
+			lxflk32.l_pid = lxflk.l_pid;
+
+			if (copyout(&lxflk32, arg, sizeof (lxflk32)) != 0) {
+				error = EFAULT;
+				break;
+			}
+		}
+#endif /* _SYSCALL32_IMPL */
+		break;
+
+	case F_GETLK64:
+	case F_SETLK64:
+	case F_SETLKW64:
+		/*
+		 * Large File support is only used for ILP32 apps.
+		 */
+		if (datamodel != DATAMODEL_ILP32) {
+			error = EINVAL;
+			break;
+		}
+
+		if (cmd == F_GETLK64)
+			cmd = F_GETLK;
+		else if (cmd == F_SETLK64)
+			cmd = F_SETLK;
+		else if (cmd == F_SETLKW64)
+			cmd = F_SETLKW;
+
+		if (copyin(arg, &lxflk64, sizeof (lxflk64)) != 0) {
+			error = EFAULT;
+			break;
+		}
+
+		ltos_flock64(&lxflk64, &bf);
+
+		if ((error = flock_check(vp, &bf, offset, MAXOFFSET_T)) != 0)
+			break;
+
+		if ((error = VOP_FRLOCK(vp, cmd, &bf, flag, offset, NULL,
+		    fp->f_cred, NULL)) != 0)
+			break;
+
+		if (cmd != F_GETLK)
+			break;
+
+		/*
+		 * The command is GETLK, return result.
+		 */
+		stol_flock64(&bf, &lxflk64);
+
+		/*
+		 * If no lock is found, only the type field is changed.
+		 */
+		if (lxflk64.l_type == LX_F_UNLCK) {
+			/* l_type always first entry, always a short */
+			if (copyout(&lxflk64.l_type,
+			    &((lx_flock64_t *)arg)->l_type,
+			    sizeof (lxflk64.l_type)))
+				error = EFAULT;
+			break;
+		}
+
+		if (bf.l_start > maxoffset || bf.l_len > maxoffset) {
+			error = EOVERFLOW;
+			break;
+		}
+
+		if (copyout(&lxflk64, arg, sizeof (lxflk64)) != 0) {
+			error = EFAULT;
+			break;
+		}
+		break;
+	}
+
+	releasef(fd);
+	if (error)
+		return (set_errno(error));
+
+	return (0);
+}
+
+long
+lx_fcntl(int fd, int cmd, intptr_t arg)
+{
+	switch (cmd) {
+	case LX_F_GETLK64:
+	case LX_F_SETLK64:
+	case LX_F_SETLKW64:
+		/* The 64-bit fcntl commands must go through fcntl64(). */
+		return (set_errno(EINVAL));
+
+	case LX_F_GETLK:
+	case LX_F_SETLK:
+	case LX_F_SETLKW:
+		return (lx_fcntl_lock(fd, cmd, (void *)arg));
+
+	default:
+		return (lx_fcntl_common(fd, cmd, arg));
+	}
+}
+
+long
+lx_fcntl64(int fd, int cmd, intptr_t arg)
+{
+	switch (cmd) {
+	case LX_F_GETLK:
+	case LX_F_SETLK:
+	case LX_F_SETLKW:
+	case LX_F_GETLK64:
+	case LX_F_SETLKW64:
+	case LX_F_SETLK64:
+		return (lx_fcntl_lock(fd, cmd, (void *)arg));
+
+	default:
+		return (lx_fcntl_common(fd, cmd, (ulong_t)arg));
+	}
+}
diff --git a/usr/src/uts/common/brand/lx/syscall/lx_futex.c b/usr/src/uts/common/brand/lx/syscall/lx_futex.c
new file mode 100644
index 0000000000..f6a8f33f46
--- /dev/null
+++ b/usr/src/uts/common/brand/lx/syscall/lx_futex.c
@@ -0,0 +1,1077 @@
+/*
+ * CDDL HEADER START
+ *
+ * The contents of this file are subject to the terms of the
+ * Common Development and Distribution License (the "License").
+ * You may not use this file except in compliance with the License.
+ *
+ * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
+ * or http://www.opensolaris.org/os/licensing.
+ * See the License for the specific language governing permissions
+ * and limitations under the License.
+ *
+ * When distributing Covered Code, include this CDDL HEADER in each
+ * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
+ * If applicable, add the following below this CDDL HEADER, with the
+ * fields enclosed by brackets "[]" replaced with your own identifying
+ * information: Portions Copyright [yyyy] [name of copyright owner]
+ *
+ * CDDL HEADER END
+ */
+/*
+ * Copyright 2006 Sun Microsystems, Inc.  All rights reserved.
+ * Use is subject to license terms.
+ */
+
+/*
+ * Copyright 2015 Joyent, Inc.
+ */
+
+#include <sys/types.h>
+#include <sys/systm.h>
+#include <sys/errno.h>
+#include <sys/debug.h>
+#include <vm/as.h>
+#include <vm/seg.h>
+#include <vm/seg_vn.h>
+#include <vm/page.h>
+#include <sys/priv.h>
+#include <sys/mman.h>
+#include <sys/timer.h>
+#include <sys/condvar.h>
+#include <sys/inttypes.h>
+#include <sys/cmn_err.h>
+#include <sys/brand.h>
+#include <sys/lx_brand.h>
+#include <sys/lx_futex.h>
+#include <sys/lx_impl.h>
+
+/*
+ * Futexes are a Linux-specific implementation of inter-process mutexes.
+ * They are designed to use shared memory for simple, uncontested
+ * operations, and rely on the kernel to resolve any contention issues.
+ *
+ * Most of the information in this section comes from the paper "Futexes
+ * Are Tricky", by Ulrich Drepper.  This paper is currently available at:
+ * http://people.redhat.com/~drepper/futex.pdf.
+ *
+ * A futex itself a 4-byte integer, which must be 4-byte aligned.  The
+ * value of this integer is expected to be modified using user-level atomic
+ * operations.  The futex(4) design itself does not impose any semantic
+ * constraints on the value stored in the futex; it is up to the
+ * application to define its own protocol.
+ *
+ * When the application decides that kernel intervention is required, it
+ * will use the futex(2) system call.  There are 5 different operations
+ * that can be performed on a futex, using this system call.  Since this
+ * interface has evolved over time, there are several different prototypes
+ * available to the user.  Fortunately, there is only a single kernel-level
+ * interface:
+ *
+ * long sys_futex(void *futex1, int cmd, int val1,
+ * 	struct timespec	*timeout, void *futex2, int val2)
+ *
+ * The kernel-level operations that may be performed on a futex are:
+ *
+ * FUTEX_WAIT
+ *
+ *	Atomically verify that futex1 contains the value val1.  If it
+ *	doesn't, return EWOULDBLOCK.  If it does contain the expected
+ *	value, the thread will sleep until somebody performs a FUTEX_WAKE
+ *	on the futex.  The caller may also specify a timeout, indicating
+ *	the maximum time the thread should sleep.  If the timer expires,
+ *	the call returns ETIMEDOUT.  If the thread is awoken with a signal,
+ *	the call returns EINTR.  Otherwise, the call returns 0.
+ *
+ * FUTEX_WAKE
+ *
+ *	Wake up val1 processes that are waiting on futex1.  The call
+ *	returns the number of blocked threads that were woken up.
+ *
+ * FUTEX_WAIT_BITSET/FUTEX_WAKE_BITSET
+ *
+ *	Similar to FUTEX_WAIT/FUTEX_WAKE, but each takes an additional argument
+ *	denoting a bit vector, with wakers will only waking waiters that match
+ *	in one or more bits.  These semantics are dubious enough, but the
+ *	interface has an inconsistency that is glaring even by the
+ *	embarrassingly low standards that Linux sets for itself:  the timeout
+ *	argument to FUTEX_WAIT_BITSET is absolute, not relative as it is for
+ *	FUTEX_WAIT.  And as if that weren't enough unnecessary complexity,
+ *	the caller may specify this absolute timeout to be against either
+ *	CLOCK_MONOTONIC or CLOCK_REALTIME -- but only for FUTEX_WAIT_BITSET,
+ *	of course!
+ *
+ * FUTEX_WAKE_OP
+ *
+ *	The implementation of a conditional variable in terms of futexes
+ *	actually uses two futexes:  one to assure sequential access and one to
+ *	represent the condition variable.  This implementation gives rise to a
+ *	particular performance problem whereby a thread is awoken on the futex
+ *	that represents the condition variable only to have to (potentially)
+ *	immediately wait on the futex that protects the condition variable.
+ *	(Do not confuse the futex that serves to protect the condition variable
+ *	with the pthread_mutex_t associated with pthread_cond_t -- which
+ *	represents a third futex.)  To (over)solve this problem, FUTEX_WAKE_OP
+ *	was invented, which performs an atomic compare-and-exchange on a
+ *	second address in a specified fashion (that is, with a specified
+ *	operation).  Here are the possible operations (OPARG is defined
+ *	to be 12 bit value embedded in the operation):
+ *
+ *	- FUTEX_OP_SET: Sets the value at the second address to OPARG
+ *	- FUTEX_OP_ADD: Adds the value to OPARG
+ *	- FUTEX_OP_OR: OR's the value with OPARG
+ *	- FUTEX_OP_ANDN: Performs a negated AND of the value with OPARG
+ *	- FUTEX_OP_XOR: XOR's the value with OPARG
+ *
+ *	After this compare-and-exchange on the second address, a FUTEX_WAKE is
+ *	performed on the first address and -- if the compare-and-exchange
+ *	matches a specified result based on a specified comparison operation --
+ *	a FUTEX_WAKE is performed on the second address.  Here are the possible
+ *	comparison operations:
+ *
+ *	- FUTEX_OP_CMP_EQ: If old value is CMPARG, wake
+ *	- FUTEX_OP_CMP_NE: If old value is not equal to CMPARG, wake
+ *	- FUTEX_OP_CMP_LT: If old value is less than CMPARG, wake
+ *	- FUTEX_OP_CMP_LE: If old value is less than or equal to CMPARG, wake
+ *	- FUTEX_OP_CMP_GT: If old value is greater than CMPARG, wake
+ *	- FUTEX_OP_CMP_GE: If old value is greater than or equal to CMPARG, wake
+ *
+ *	As a practical matter, the only way that this is used (or, some might
+ *	argue, is usable) is by the implementation of pthread_cond_signal(),
+ *	which uses FUTEX_WAKE_OP to -- in a single system call -- unlock the
+ *	futex that protects the condition variable and wake the futex that
+ *	represents the condition variable.  The second wake-up is conditional
+ *	because the futex that protects the condition variable (rather than the
+ *	one that represents it) may or may not have waiters.  Given that this
+ *	is the use case, FUTEX_WAKE_OP is falsely generic: despite allowing for
+ *	five different kinds of operations and six different kinds of
+ *	comparision operations, in practice only one is used.  (Namely, setting
+ *	to 0 and waking if the old value is greater than 1 -- which denotes
+ *	that waiters are present and the wakeup should be performed.) Moreover,
+ *	because FUTEX_WAKE_OP does not (and cannot) optimize anything in the
+ *	case that the pthread_mutex_t associated with the pthread_cond_t is
+ *	held at the time of a pthread_cond_signal(), this entire mechanism is
+ *	essentially for naught in this case.  As one can imagine (and can
+ *	verify on just about any source base that uses pthread_cond_signal()),
+ *	it is overwhelmingly the common case that the lock associated with the
+ *	pthread_cond_t is held at the time of pthread_cond_signal(), assuring
+ *	that the problem that all of this complexity was designed to solve
+ *	isn't, in fact, solved because the signalled thread simply wakes up
+ *	only to block again on the held mutex.  Cue a slow clap!
+ *
+ * FUTEX_CMP_REQUEUE
+ *
+ *	If the value stored in futex1 matches that passed in in val2, wake
+ *	up val1 processes that are waiting on futex1.  Otherwise, return
+ *	EAGAIN.
+ *
+ *	If there are more than val1 threads waiting on the futex, remove
+ *	the remaining threads from this futex, and requeue them on futex2.
+ *	The caller can limit the number of threads being requeued by
+ *	encoding an integral numerical value in the position usually used
+ *	for the timeout pointer.
+ *
+ *	The call returns the number of blocked threads that were woken up
+ *	or requeued.
+ *
+ * FUTEX_REQUEUE
+ *
+ *	 Identical to FUTEX_CMP_REQUEUE except that it does not use val2.
+ *	 This command has been declared broken and obsolete, but we still
+ *	 need to support it.
+ *
+ * FUTEX_FD
+ *
+ *	Return a file descriptor, which can be used to refer to the futex.
+ *	This operation was broken by design, and was blessedly removed in
+ *	Linux 2.6.26 ("because it was inherently racy"); it should go without
+ *	saying that we don't support this operation.
+ */
+
+/*
+ * This structure is used to track all the threads currently waiting on a
+ * futex.  There is one fwaiter_t for each blocked thread.  We store all
+ * fwaiter_t's in a hash structure, indexed by the memid_t of the integer
+ * containing the futex's value.
+ *
+ * At the moment, all fwaiter_t's for a single futex are simply dumped into
+ * the hash bucket.  If futex contention ever becomes a hot path, we can
+ * chain a single futex's waiters together.
+ */
+typedef struct fwaiter {
+	memid_t		fw_memid;	/* memid of the user-space futex */
+	kcondvar_t	fw_cv;		/* cond var */
+	struct fwaiter	*fw_next;	/* hash queue */
+	struct fwaiter	*fw_prev;	/* hash queue */
+	uint32_t	fw_bits;	/* bits waiting on */
+	volatile int	fw_woken;
+} fwaiter_t;
+
+/*
+ * The structure of the robust_list, as set with the set_robust_list() system
+ * call.  See lx_futex_robust_exit(), below, for details.
+ */
+typedef struct futex_robust_list {
+	uintptr_t	frl_head;	/* list of robust locks held */
+	uint64_t	frl_offset;	/* offset of lock word within a lock */
+	uintptr_t	frl_pending;	/* pending operation */
+} futex_robust_list_t;
+
+#if defined(_SYSCALL32_IMPL)
+
+#pragma pack(4)
+typedef struct futex_robust_list32 {
+	uint32_t	frl_head;	/* list of robust locks held */
+	uint32_t	frl_offset;	/* offset of lock word within a lock */
+	uint32_t	frl_pending;	/* pending operation */
+} futex_robust_list32_t;
+#pragma pack()
+
+#endif
+
+#define	MEMID_COPY(s, d) \
+	{ (d)->val[0] = (s)->val[0]; (d)->val[1] = (s)->val[1]; }
+#define	MEMID_EQUAL(s, d) \
+	((d)->val[0] == (s)->val[0] && (d)->val[1] == (s)->val[1])
+
+/* Borrowed from the page freelist hash code.  */
+#define	HASH_SHIFT_SZ	7
+#define	HASH_SIZE	(1 << HASH_SHIFT_SZ)
+#define	HASH_FUNC(id)						\
+	((((uintptr_t)((id)->val[1]) >> PAGESHIFT) +			\
+	((uintptr_t)((id)->val[1]) >> (PAGESHIFT + HASH_SHIFT_SZ)) +	\
+	((uintptr_t)((id)->val[0]) >> 3) +				\
+	((uintptr_t)((id)->val[0]) >> (3 + HASH_SHIFT_SZ)) +		\
+	((uintptr_t)((id)->val[0]) >> (3 + 2 * HASH_SHIFT_SZ))) &	\
+	(HASH_SIZE - 1))
+
+static fwaiter_t *futex_hash[HASH_SIZE];
+static kmutex_t futex_hash_lock[HASH_SIZE];
+
+static void
+futex_hashin(fwaiter_t *fwp)
+{
+	int index;
+
+	index = HASH_FUNC(&fwp->fw_memid);
+	ASSERT(MUTEX_HELD(&futex_hash_lock[index]));
+
+	fwp->fw_prev = NULL;
+	fwp->fw_next = futex_hash[index];
+	if (fwp->fw_next)
+		fwp->fw_next->fw_prev = fwp;
+	futex_hash[index] = fwp;
+}
+
+static void
+futex_hashout(fwaiter_t *fwp)
+{
+	int index;
+
+	index = HASH_FUNC(&fwp->fw_memid);
+	ASSERT(MUTEX_HELD(&futex_hash_lock[index]));
+
+	if (fwp->fw_prev)
+		fwp->fw_prev->fw_next = fwp->fw_next;
+	if (fwp->fw_next)
+		fwp->fw_next->fw_prev = fwp->fw_prev;
+	if (futex_hash[index] == fwp)
+		futex_hash[index] = fwp->fw_next;
+
+	fwp->fw_prev = NULL;
+	fwp->fw_next = NULL;
+}
+
+/*
+ * Go to sleep until somebody does a WAKE operation on this futex, we get a
+ * signal, or the timeout expires.
+ */
+static int
+futex_wait(memid_t *memid, caddr_t addr,
+    int val, timespec_t *timeout, uint32_t bits)
+{
+	int err, ret;
+	int32_t curval;
+	fwaiter_t fw;
+	int index;
+
+	fw.fw_woken = 0;
+	fw.fw_bits = bits;
+
+	MEMID_COPY(memid, &fw.fw_memid);
+	cv_init(&fw.fw_cv, NULL, CV_DEFAULT, NULL);
+
+	index = HASH_FUNC(&fw.fw_memid);
+	mutex_enter(&futex_hash_lock[index]);
+
+	if (fuword32(addr, (uint32_t *)&curval)) {
+		err = set_errno(EFAULT);
+		goto out;
+	}
+	if (curval != val) {
+		err = set_errno(EWOULDBLOCK);
+		goto out;
+	}
+
+	futex_hashin(&fw);
+
+	err = 0;
+	while ((fw.fw_woken == 0) && (err == 0)) {
+		ret = cv_waituntil_sig(&fw.fw_cv, &futex_hash_lock[index],
+		    timeout, timechanged);
+		if (ret < 0) {
+			err = set_errno(ETIMEDOUT);
+		} else if (ret == 0) {
+			/*
+			 * According to signal(7), a futex(2) call with the
+			 * FUTEX_WAIT operation is restartable.
+			 */
+			ttolxlwp(curthread)->br_syscall_restart = B_TRUE;
+			err = set_errno(EINTR);
+		}
+	}
+
+	/*
+	 * The futex is normally hashed out in wakeup.  If we timed out or
+	 * got a signal, we need to hash it out here instead.
+	 */
+	if (fw.fw_woken == 0)
+		futex_hashout(&fw);
+
+out:
+	mutex_exit(&futex_hash_lock[index]);
+
+	return (err);
+}
+
+/*
+ * Wake up to wake_threads threads that are blocked on the futex at memid.
+ */
+static int
+futex_wake(memid_t *memid, int wake_threads, uint32_t mask)
+{
+	fwaiter_t *fwp, *next;
+	int index;
+	int ret = 0;
+
+	index = HASH_FUNC(memid);
+
+	mutex_enter(&futex_hash_lock[index]);
+
+	for (fwp = futex_hash[index]; fwp && ret < wake_threads; fwp = next) {
+		next = fwp->fw_next;
+		if (MEMID_EQUAL(&fwp->fw_memid, memid) &&
+		    (fwp->fw_bits & mask)) {
+			futex_hashout(fwp);
+			fwp->fw_woken = 1;
+			cv_signal(&fwp->fw_cv);
+			ret++;
+		}
+	}
+
+	mutex_exit(&futex_hash_lock[index]);
+
+	return (ret);
+}
+
+static int
+futex_wake_op_execute(int32_t *addr, int32_t val3)
+{
+	int32_t op = FUTEX_OP_OP(val3);
+	int32_t cmp = FUTEX_OP_CMP(val3);
+	int32_t cmparg = FUTEX_OP_CMPARG(val3);
+	int32_t oparg, oldval, newval;
+	label_t ljb;
+	int rval;
+
+	if ((uintptr_t)addr >= KERNELBASE)
+		return (set_errno(EFAULT));
+
+	if (on_fault(&ljb))
+		return (set_errno(EFAULT));
+
+	oparg = FUTEX_OP_OPARG(val3);
+
+	do {
+		oldval = *addr;
+		newval = oparg;
+
+		switch (op) {
+		case FUTEX_OP_SET:
+			break;
+
+		case FUTEX_OP_ADD:
+			newval += oparg;
+			break;
+
+		case FUTEX_OP_OR:
+			newval |= oparg;
+			break;
+
+		case FUTEX_OP_ANDN:
+			newval &= ~oparg;
+			break;
+
+		case FUTEX_OP_XOR:
+			newval ^= oparg;
+			break;
+
+		default:
+			no_fault();
+			return (set_errno(EINVAL));
+		}
+	} while (atomic_cas_32((uint32_t *)addr, oldval, newval) != oldval);
+
+	no_fault();
+
+	switch (cmp) {
+	case FUTEX_OP_CMP_EQ:
+		rval = (oldval == cmparg);
+		break;
+
+	case FUTEX_OP_CMP_NE:
+		rval = (oldval != cmparg);
+		break;
+
+	case FUTEX_OP_CMP_LT:
+		rval = (oldval < cmparg);
+		break;
+
+	case FUTEX_OP_CMP_LE:
+		rval = (oldval <= cmparg);
+		break;
+
+	case FUTEX_OP_CMP_GT:
+		rval = (oldval > cmparg);
+		break;
+
+	case FUTEX_OP_CMP_GE:
+		rval = (oldval >= cmparg);
+		break;
+
+	default:
+		return (set_errno(EINVAL));
+	}
+
+	return (rval);
+}
+
+static int
+futex_wake_op(memid_t *memid, caddr_t addr2, memid_t *memid2,
+    int wake_threads, int wake_threads2, int val3)
+{
+	kmutex_t *l1, *l2;
+	int ret = 0, ret2 = 0, wake;
+	fwaiter_t *fwp, *next;
+	int index1, index2;
+
+	index1 = HASH_FUNC(memid);
+	index2 = HASH_FUNC(memid2);
+
+	if (index1 == index2) {
+		l1 = &futex_hash_lock[index1];
+		l2 = NULL;
+	} else if (index1 < index2) {
+		l1 = &futex_hash_lock[index1];
+		l2 = &futex_hash_lock[index2];
+	} else {
+		l1 = &futex_hash_lock[index2];
+		l2 = &futex_hash_lock[index1];
+	}
+
+	mutex_enter(l1);
+	if (l2 != NULL)
+		mutex_enter(l2);
+
+	/* LINTED: alignment */
+	if ((wake = futex_wake_op_execute((int32_t *)addr2, val3)) < 0)
+		goto out;
+
+	for (fwp = futex_hash[index1]; fwp; fwp = next) {
+		next = fwp->fw_next;
+		if (!MEMID_EQUAL(&fwp->fw_memid, memid))
+			continue;
+
+		futex_hashout(fwp);
+		fwp->fw_woken = 1;
+		cv_signal(&fwp->fw_cv);
+		if (++ret >= wake_threads) {
+			break;
+		}
+	}
+
+	if (!wake)
+		goto out;
+
+	for (fwp = futex_hash[index2]; fwp; fwp = next) {
+		next = fwp->fw_next;
+		if (!MEMID_EQUAL(&fwp->fw_memid, memid2))
+			continue;
+
+		futex_hashout(fwp);
+		fwp->fw_woken = 1;
+		cv_signal(&fwp->fw_cv);
+		if (++ret2 >= wake_threads2) {
+			break;
+		}
+	}
+
+	ret += ret2;
+out:
+	if (l2 != NULL)
+		mutex_exit(l2);
+	mutex_exit(l1);
+
+	return (ret);
+}
+
+/*
+ * Wake up to wake_threads waiting on the futex at memid.  If there are
+ * more than that many threads waiting, requeue the remaining threads on
+ * the futex at requeue_memid.
+ */
+static int
+futex_requeue(memid_t *memid, memid_t *requeue_memid, int wake_threads,
+	ulong_t requeue_threads, caddr_t addr, int *cmpval)
+{
+	fwaiter_t *fwp, *next;
+	int index1, index2;
+	int ret = 0;
+	int32_t curval;
+	kmutex_t *l1, *l2;
+
+	/*
+	 * To ensure that we don't miss a wakeup if the value of cmpval
+	 * changes, we need to grab locks on both the original and new hash
+	 * buckets.  To avoid deadlock, we always grab the lower-indexed
+	 * lock first.
+	 */
+	index1 = HASH_FUNC(memid);
+	index2 = HASH_FUNC(requeue_memid);
+
+	if (index1 == index2) {
+		l1 = &futex_hash_lock[index1];
+		l2 = NULL;
+	} else if (index1 < index2) {
+		l1 = &futex_hash_lock[index1];
+		l2 = &futex_hash_lock[index2];
+	} else {
+		l1 = &futex_hash_lock[index2];
+		l2 = &futex_hash_lock[index1];
+	}
+
+	mutex_enter(l1);
+	if (l2 != NULL)
+		mutex_enter(l2);
+
+	if (cmpval != NULL) {
+		if (fuword32(addr, (uint32_t *)&curval)) {
+			ret = -EFAULT;
+			goto out;
+		}
+		if (curval != *cmpval) {
+			ret = -EAGAIN;
+			goto out;
+		}
+	}
+
+	for (fwp = futex_hash[index1]; fwp; fwp = next) {
+		next = fwp->fw_next;
+		if (!MEMID_EQUAL(&fwp->fw_memid, memid))
+			continue;
+
+		futex_hashout(fwp);
+		if (ret++ < wake_threads) {
+			fwp->fw_woken = 1;
+			cv_signal(&fwp->fw_cv);
+		} else {
+			MEMID_COPY(requeue_memid, &fwp->fw_memid);
+			futex_hashin(fwp);
+
+			if ((ret - wake_threads) >= requeue_threads)
+				break;
+		}
+	}
+
+out:
+	if (l2 != NULL)
+		mutex_exit(l2);
+	mutex_exit(l1);
+
+	if (ret < 0)
+		return (set_errno(-ret));
+	return (ret);
+}
+
+/*
+ * Copy in the relative timeout provided by the application and convert it
+ * to an absolute timeout.  Sadly, this is complicated by the different
+ * timeout of semantics of FUTEX_WAIT vs. FUTEX_WAIT_BITSET.  (Yes, you read
+ * that correctly; FUTEX_WAIT and FUTEX_WAIT_BITSET have different timeout
+ * semantics; see the block comment at the top of the file for commentary
+ * on this inanity.)
+ */
+static int
+get_timeout(void *lx_timeout, timestruc_t *timeout, int cmd, int clock)
+{
+	timestruc_t now;
+
+	if (get_udatamodel() == DATAMODEL_NATIVE) {
+		if (copyin(lx_timeout, timeout, sizeof (timestruc_t)))
+			return (EFAULT);
+	}
+#ifdef _SYSCALL32_IMPL
+	else {
+		timestruc32_t timeout32;
+		if (copyin(lx_timeout, &timeout32, sizeof (timestruc32_t)))
+			return (EFAULT);
+		timeout->tv_sec = (time_t)timeout32.tv_sec;
+		timeout->tv_nsec = timeout32.tv_nsec;
+	}
+#endif
+	if (itimerspecfix(timeout))
+		return (EINVAL);
+
+	if (cmd == FUTEX_WAIT) {
+		/*
+		 * We've been given a relative time; add it to the current
+		 * time to derive an absolute time.
+		 */
+		gethrestime(&now);
+		timespecadd(timeout, &now);
+	} else {
+		/*
+		 * This is a FUTEX_WAIT_BITSET operation, which (1) specifies
+		 * the timeout as an absolute rather than a relative timeout
+		 * and (2) allows for different clock types to be specified.
+		 * If the clock is CLOCK_REALTIME, we actually have nothing
+		 * to do -- but if this is CLOCK_MONOTONIC, we need to convert
+		 * our absolute time back into a relative time and then add
+		 * it to our current hrestime to get an absolute CLOCK_REALTIME
+		 * timeout.
+		 */
+		if (clock == CLOCK_MONOTONIC) {
+			/*
+			 * Get our current time, and subtract it from our
+			 * timeout to get the relative value.
+			 */
+			hrt2ts(gethrtime(), &now);
+			timespecsub(timeout, &now);
+
+			/*
+			 * If our timeout is in the past, set it to be 0.
+			 */
+			if (timeout->tv_sec < 0) {
+				timeout->tv_sec = 0;
+				timeout->tv_nsec = 0;
+			}
+
+			/*
+			 * Add the relative time back into the current time.
+			 */
+			gethrestime(&now);
+			timespecadd(timeout, &now);
+		}
+	}
+
+	return (0);
+}
+
+long
+lx_futex(uintptr_t addr, int op, int val, uintptr_t lx_timeout,
+	uintptr_t addr2, int val3)
+{
+	struct as *as = curproc->p_as;
+	memid_t memid, memid2;
+	timestruc_t timeout;
+	timestruc_t *tptr = NULL;
+	int val2 = NULL;
+	int rval = 0;
+	int cmd = op & FUTEX_CMD_MASK;
+	int private = op & FUTEX_PRIVATE_FLAG;
+	char dmsg[32];
+
+	/* must be aligned on int boundary */
+	if (addr & 0x3)
+		return (set_errno(EINVAL));
+
+	/* Sanity check the futex command */
+	if (cmd < 0 || cmd > FUTEX_MAX_CMD)
+		return (set_errno(EINVAL));
+
+	if (cmd == FUTEX_FD) {
+		/*
+		 * FUTEX_FD was sentenced to death for grievous crimes of
+		 * semantics against humanity; it has been ripped out of Linux
+		 * and will never be supported by us.
+		 */
+		(void) snprintf(dmsg, sizeof (dmsg), "futex 0x%x", cmd);
+		lx_unsupported(dmsg);
+		return (set_errno(ENOSYS));
+	}
+
+	switch (cmd) {
+	case FUTEX_LOCK_PI:
+	case FUTEX_UNLOCK_PI:
+	case FUTEX_TRYLOCK_PI:
+	case FUTEX_WAIT_REQUEUE_PI:
+	case FUTEX_CMP_REQUEUE_PI:
+		/*
+		 * These are operations that we don't currently support, but
+		 * may well need to in the future.  For now, callers need to
+		 * deal with these being missing -- but if and as that changes,
+		 * they may well need to be implemented.
+		 */
+		(void) snprintf(dmsg, sizeof (dmsg), "futex 0x%x", cmd);
+		lx_unsupported(dmsg);
+		return (set_errno(ENOSYS));
+	}
+
+	if ((op & FUTEX_CLOCK_REALTIME) && cmd != FUTEX_WAIT_BITSET) {
+		/*
+		 * Linux only allows FUTEX_CLOCK_REALTIME to be set on the
+		 * FUTEX_WAIT_BITSET and FUTEX_WAIT_REQUEUE_PI commands.
+		 */
+		return (set_errno(ENOSYS));
+	}
+
+	/* Copy in the timeout structure from userspace. */
+	if ((cmd == FUTEX_WAIT || cmd == FUTEX_WAIT_BITSET) &&
+	    lx_timeout != NULL) {
+		rval = get_timeout((timespec_t *)lx_timeout, &timeout, cmd,
+		    op & FUTEX_CLOCK_REALTIME ? CLOCK_REALTIME :
+		    CLOCK_MONOTONIC);
+
+		if (rval != 0)
+			return (set_errno(rval));
+		tptr = &timeout;
+	}
+
+	switch (cmd) {
+	case FUTEX_REQUEUE:
+	case FUTEX_CMP_REQUEUE:
+	case FUTEX_WAKE_OP:
+		/*
+		 * lx_timeout is nominally a pointer to a userspace address.
+		 * For several commands, however, it actually contains
+		 * an additional integer parameter.  This is horrible, and
+		 * the people who did this to us should be sorry.
+		 */
+		val2 = (int)lx_timeout;
+	}
+
+	/*
+	 * Translate the process-specific, user-space futex virtual
+	 * address(es) to a universal memid.  If the private bit is set, we
+	 * can just use our as plus the virtual address, saving quite a bit
+	 * of effort.
+	 */
+	if (private) {
+		memid.val[0] = (uintptr_t)as;
+		memid.val[1] = (uintptr_t)addr;
+	} else {
+		rval = as_getmemid(as, (void *)addr, &memid);
+		if (rval != 0)
+			return (set_errno(rval));
+	}
+
+	if (cmd == FUTEX_REQUEUE || cmd == FUTEX_CMP_REQUEUE ||
+	    cmd == FUTEX_WAKE_OP) {
+		if (addr2 & 0x3)
+			return (set_errno(EINVAL));
+
+		if (private) {
+			memid2.val[0] = (uintptr_t)as;
+			memid2.val[1] = (uintptr_t)addr2;
+		} else {
+			rval = as_getmemid(as, (void *)addr2, &memid2);
+			if (rval)
+				return (set_errno(rval));
+		}
+	}
+
+	switch (cmd) {
+	case FUTEX_WAIT:
+		rval = futex_wait(&memid, (void *)addr, val,
+		    tptr, FUTEX_BITSET_MATCH_ANY);
+		break;
+
+	case FUTEX_WAIT_BITSET:
+		rval = futex_wait(&memid, (void *)addr, val, tptr, val3);
+		break;
+
+	case FUTEX_WAKE:
+		rval = futex_wake(&memid, val, FUTEX_BITSET_MATCH_ANY);
+		break;
+
+	case FUTEX_WAKE_BITSET:
+		rval = futex_wake(&memid, val, val3);
+		break;
+
+	case FUTEX_WAKE_OP:
+		rval = futex_wake_op(&memid, (void *)addr2, &memid2,
+		    val, val2, val3);
+		break;
+
+	case FUTEX_CMP_REQUEUE:
+	case FUTEX_REQUEUE:
+		rval = futex_requeue(&memid, &memid2, val,
+		    val2, (void *)addr2, &val3);
+
+		break;
+	}
+
+	return (rval);
+}
+
+/*
+ * Does the dirty work of actually dropping a held robust lock in the event
+ * of the untimely death of the owner; see lx_futex_robust_exit(), below.
+ */
+static void
+lx_futex_robust_drop(uintptr_t addr, uint32_t tid)
+{
+	memid_t memid;
+	uint32_t oldval, newval;
+
+	VERIFY(addr + sizeof (uint32_t) < KERNELBASE);
+
+	do {
+		fuword32_noerr((void *)addr, &oldval);
+
+		if ((oldval & FUTEX_TID_MASK) != tid)
+			return;
+
+		newval = (oldval & FUTEX_WAITERS) | FUTEX_OWNER_DIED;
+	} while (atomic_cas_32((uint32_t *)addr, oldval, newval) != oldval);
+
+	/*
+	 * We have now denoted that this lock's owner is dead; we need to
+	 * wake any waiters.
+	 */
+	if (as_getmemid(curproc->p_as, (void *)addr, &memid) != 0)
+		return;
+
+	(void) futex_wake(&memid, 1, FUTEX_BITSET_MATCH_ANY);
+}
+
+/*
+ * Called when a thread is exiting.  The role of the kernel is very clearly
+ * spelled out in the Linux design document entitled robust-futex-ABI.txt:
+ * we must (carefully!) iterate over the list of held locks pointed to by
+ * the robust list head; for each lock, we'll check to see if the calling
+ * (exiting) thread is the owner, and if so, denote that the lock is dead
+ * and wake any waiters.  (The "pending" field of the head points to a lock
+ * that is in transition; it should be dropped if held.)  If there are any
+ * errors through here at all (including memory operations), we abort the
+ * entire operation.
+ */
+void
+lx_futex_robust_exit(uintptr_t addr, uint32_t tid)
+{
+	futex_robust_list_t list;
+	uintptr_t entry, next;
+	model_t model = get_udatamodel();
+	int length = 0;
+	label_t ljb;
+
+	if (on_fault(&ljb))
+		return;
+
+	if (addr + sizeof (futex_robust_list_t) >= KERNELBASE)
+		goto out;
+
+	if (model == DATAMODEL_NATIVE) {
+		copyin_noerr((void *)addr, &list, sizeof (list));
+	}
+#if defined(_SYSCALL32_IMPL)
+	else {
+		futex_robust_list32_t list32;
+
+		copyin_noerr((void *)addr, &list32, sizeof (list32));
+		list.frl_head = list32.frl_head;
+		list.frl_offset = list32.frl_offset;
+		list.frl_pending = list32.frl_pending;
+	}
+#endif
+
+	/*
+	 * Strip off the PI bit, if any.
+	 */
+	entry = list.frl_head & ~FUTEX_ROBUST_LOCK_PI;
+
+	while (entry != addr && length++ < FUTEX_ROBUST_LIST_LIMIT) {
+		if (entry + list.frl_offset + sizeof (uint32_t) >= KERNELBASE)
+			goto out;
+
+		if (model == DATAMODEL_NATIVE) {
+			fulword_noerr((void *)entry, &next);
+		}
+#if defined(_SYSCALL32_IMPL)
+		else {
+			uint32_t next32;
+			fuword32_noerr((void *)entry, &next32);
+			next = next32;
+		}
+#endif
+
+		/*
+		 * Drop the robust mutex -- but only if our pending lock didn't
+		 * somehow sneak on there.
+		 */
+		if (entry != list.frl_pending)
+			lx_futex_robust_drop(entry + list.frl_offset, tid);
+
+		entry = next & ~FUTEX_LOCK_PI;
+	}
+
+	/*
+	 * Finally, drop the pending lock if there is one.
+	 */
+	if (list.frl_pending != NULL && list.frl_pending +
+	    list.frl_offset + sizeof (uint32_t) < KERNELBASE)
+		lx_futex_robust_drop(list.frl_pending + list.frl_offset, tid);
+
+out:
+	no_fault();
+}
+
+long
+lx_set_robust_list(void *listp, size_t len)
+{
+	proc_t *p = curproc;
+	klwp_t *lwp = ttolwp(curthread);
+	struct lx_lwp_data *lwpd = lwptolxlwp(lwp);
+
+	if (get_udatamodel() == DATAMODEL_NATIVE) {
+		if (len != sizeof (futex_robust_list_t))
+			return (set_errno(EINVAL));
+	}
+#if defined(_SYSCALL32_IMPL)
+	else {
+		if (len != sizeof (futex_robust_list32_t))
+			return (set_errno(EINVAL));
+	}
+#endif
+
+	/*
+	 * To assure that we are serialized with respect to any racing call
+	 * to lx_get_robust_list(), we lock ourselves to set the value.  (Note
+	 * that sprunlock() drops p_lock.)
+	 */
+	mutex_enter(&p->p_lock);
+	sprlock_proc(p);
+	lwpd->br_robust_list = listp;
+	sprunlock(p);
+
+	return (0);
+}
+
+long
+lx_get_robust_list(pid_t pid, void **listp, size_t *lenp)
+{
+	model_t model = get_udatamodel();
+	pid_t rpid;
+	id_t rtid;
+	proc_t *rproc;
+	klwp_t *rlwp;
+	lx_lwp_data_t *rlwpd;
+	kthread_t *rthr;
+	void *list;
+	int err = 0;
+
+	if (pid == 0) {
+		/*
+		 * A pid of 0 denotes the current thread; we lock the current
+		 * process even though it isn't strictly necessary (we can't
+		 * race with set_robust_list() because a thread may only set
+		 * its robust list on itself).
+		 */
+		rproc = curproc;
+		rlwpd = lwptolxlwp(ttolwp(curthread));
+		mutex_enter(&curproc->p_lock);
+		sprlock_proc(rproc);
+	} else {
+		if (lx_lpid_to_spair(pid, &rpid, &rtid) != 0 ||
+		    (rproc = sprlock(rpid)) == NULL) {
+			/*
+			 * We couldn't find the specified process.
+			 */
+			return (set_errno(ESRCH));
+		}
+
+		if (rproc->p_model != model ||
+		    (rthr = idtot(rproc, rtid)) == NULL ||
+		    (rlwp = ttolwp(rthr)) == NULL ||
+		    (rlwpd = lwptolxlwp(rlwp)) == NULL) {
+			/*
+			 * The target process does not match our data model, or
+			 * we couldn't find the LWP, or the target process is
+			 * not branded.
+			 */
+			err = ESRCH;
+			goto out;
+		}
+	}
+
+	if (curproc != rproc &&
+	    priv_proc_cred_perm(curproc->p_cred, rproc, NULL, VREAD) != 0) {
+		/*
+		 * We don't have the permission to examine the target.
+		 */
+		err = EPERM;
+		goto out;
+	}
+
+	list = rlwpd->br_robust_list;
+
+out:
+	sprunlock(rproc);
+
+	if (err != 0)
+		return (set_errno(err));
+
+	if (model == DATAMODEL_NATIVE) {
+		if (sulword(listp, (uintptr_t)list) != 0)
+			return (set_errno(EFAULT));
+
+		if (sulword(lenp, sizeof (futex_robust_list_t)) != 0)
+			return (set_errno(EFAULT));
+	}
+#if defined(_SYSCALL32_IMPL)
+	else {
+		if (suword32(listp, (uint32_t)(uintptr_t)list) != 0)
+			return (set_errno(EFAULT));
+
+		if (suword32(lenp, sizeof (futex_robust_list32_t)) != 0)
+			return (set_errno(EFAULT));
+	}
+#endif
+
+	return (0);
+}
+
+void
+lx_futex_init(void)
+{
+	int i;
+
+	for (i = 0; i < HASH_SIZE; i++)
+		mutex_init(&futex_hash_lock[i], NULL, MUTEX_DEFAULT, NULL);
+	bzero(futex_hash, sizeof (futex_hash));
+}
+
+int
+lx_futex_fini(void)
+{
+	int i, err;
+
+	err = 0;
+	for (i = 0; (err == 0) && (i < HASH_SIZE); i++) {
+		mutex_enter(&futex_hash_lock[i]);
+		if (futex_hash[i] != NULL)
+			err = EBUSY;
+		mutex_exit(&futex_hash_lock[i]);
+	}
+	return (err);
+}
diff --git a/usr/src/uts/common/brand/lx/syscall/lx_getdents.c b/usr/src/uts/common/brand/lx/syscall/lx_getdents.c
new file mode 100644
index 0000000000..a41ab19d65
--- /dev/null
+++ b/usr/src/uts/common/brand/lx/syscall/lx_getdents.c
@@ -0,0 +1,349 @@
+/*
+ * CDDL HEADER START
+ *
+ * The contents of this file are subject to the terms of the
+ * Common Development and Distribution License (the "License").
+ * You may not use this file except in compliance with the License.
+ *
+ * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
+ * or http://www.opensolaris.org/os/licensing.
+ * See the License for the specific language governing permissions
+ * and limitations under the License.
+ *
+ * When distributing Covered Code, include this CDDL HEADER in each
+ * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
+ * If applicable, add the following below this CDDL HEADER, with the
+ * fields enclosed by brackets "[]" replaced with your own identifying
+ * information: Portions Copyright [yyyy] [name of copyright owner]
+ *
+ * CDDL HEADER END
+ */
+
+/*
+ * Copyright 2015 Joyent, Inc.
+ */
+
+#include <sys/systm.h>
+#include <sys/filio.h>
+#include <sys/types.h>
+#include <sys/stat.h>
+#include <sys/inttypes.h>
+#include <sys/vnode.h>
+#include <sys/dirent.h>
+#include <sys/errno.h>
+#include <sys/file.h>
+#include <sys/sysmacros.h>
+#include <sys/kmem.h>
+#include <sys/sunddi.h>
+
+#include <sys/lx_types.h>
+#include <sys/lx_misc.h>
+
+#define	LX_NAMEMAX	256
+
+#define	LX_GETDENTS_MAX_BUFSZ	65536
+
+/*
+ * Because the Linux dirent has an extra field (d_type), it's possible that
+ * each entry will be 8 bytes larger due to padding.  To prevent overrun during
+ * translation, the illumos-native buffer is sized pessimistically.
+ */
+#define	LTOS_GETDENTS_BUFSZ(bufsz, datasz)	\
+	(((bufsz) / (datasz + 8)) * sizeof (struct dirent))
+
+/*
+ * Record must be long enough to house d_name string, null terminator and
+ * d_type field.  It's then padded to nearest 8-byte boundary
+ */
+#define	LX_RECLEN(l, t)	\
+	((offsetof(t, d_name) + 2 + (l) + 7) & ~7)
+
+/*
+ * Bytes after d_name string until d_reclen should be zeroed.
+ * Includes zero-terminating d_name
+ */
+#define	LX_ZEROLEN(l, t)	\
+	(LX_RECLEN(l, t) -	\
+	((offsetof(t, d_name) + (l))))
+
+/* The output format of getdents differs if the caller is 32 or 64 bit. */
+struct lx_dirent_32 {
+	uint32_t	d_ino;
+	int32_t		d_off;
+	ushort_t	d_reclen;
+	char		d_name[1];
+	uchar_t		d_type;
+};
+
+struct lx_dirent_64 {
+	uint64_t	d_ino;
+	int64_t		d_off;
+	ushort_t	d_reclen;
+	char		d_name[1];
+	uchar_t		d_type;
+};
+
+static long
+lx_getdents_common(int fd, caddr_t uptr, size_t count,
+    unsigned int lx_size, int (*outcb)(caddr_t, caddr_t, int))
+{
+	vnode_t *vp;
+	file_t *fp;
+	struct uio auio;
+	struct iovec aiov;
+	int error;
+	int sbufsz, lbufsz, bufsz;
+	void *lbuf, *sbuf;
+	size_t outb = 0;
+
+	if (count < lx_size) {
+		return (set_errno(EINVAL));
+	}
+	if ((fp = getf(fd)) == NULL) {
+		return (set_errno(EBADF));
+	}
+	vp = fp->f_vnode;
+	if (vp->v_type != VDIR) {
+		releasef(fd);
+		return (set_errno(ENOTDIR));
+	}
+	if (!(fp->f_flag & FREAD)) {
+		releasef(fd);
+		return (set_errno(EBADF));
+	}
+
+	if (count > LX_GETDENTS_MAX_BUFSZ) {
+		/*
+		 * If the target buffer passed to us is huge, keep the
+		 * translation buffers moderate in size.  Iteration will be
+		 * used to fill the request.
+		 */
+		lbufsz = LX_GETDENTS_MAX_BUFSZ;
+		sbufsz = LTOS_GETDENTS_BUFSZ(LX_GETDENTS_MAX_BUFSZ, lx_size);
+	} else if (count < (lx_size + MAXPATHLEN)) {
+		/*
+		 * If the target buffer is tiny, allocate a Linux-format buffer
+		 * big enough to hold at least one max-length row while keeping
+		 * the illumos-format buffer pesimistic in size.
+		 *
+		 * Assuming the buffer is truely tiny, it's likely that the
+		 * result will not fit and an EINVAL will be tossed.
+		 */
+		lbufsz = (lx_size + MAXPATHLEN);
+		sbufsz = MAX((LTOS_GETDENTS_BUFSZ(count, lx_size)),
+		    sizeof (struct dirent));
+	} else {
+		lbufsz = count;
+		sbufsz = LTOS_GETDENTS_BUFSZ(count, lx_size);
+	}
+	bufsz = sbufsz;
+	lbuf = kmem_alloc(lbufsz, KM_SLEEP);
+	sbuf = kmem_alloc(sbufsz, KM_SLEEP);
+
+	aiov.iov_base = sbuf;
+	aiov.iov_len = sbufsz;
+	auio.uio_iov = &aiov;
+	auio.uio_iovcnt = 1;
+	auio.uio_loffset = fp->f_offset;
+	auio.uio_segflg = UIO_SYSSPACE;
+	auio.uio_resid = sbufsz;
+	auio.uio_fmode = 0;
+	auio.uio_extflg = UIO_COPY_CACHED;
+
+	/*
+	 * Since we use a conservative buffer allocation for the differing
+	 * struct sizing and Linux places fewer limits on getdents buffers in
+	 * general, there's a chance we'll undershoot on the record count.
+	 * When this happens, we can simply repeat the READDIR operation until
+	 * the available records are exhausted or we've filled the user buffer.
+	 */
+	while (1) {
+		int at_eof, res;
+		(void) VOP_RWLOCK(vp, V_WRITELOCK_FALSE, NULL);
+		error = VOP_READDIR(vp, &auio, fp->f_cred, &at_eof, NULL, 0);
+		VOP_RWUNLOCK(vp, V_WRITELOCK_FALSE, NULL);
+		if (error != 0 || auio.uio_resid == sbufsz) {
+			break;
+		}
+		res = outcb(sbuf, lbuf, bufsz - auio.uio_resid);
+		VERIFY(res <= lbufsz);
+		if (res == 0) {
+			/* no records to copyout from this batch */
+			break;
+		} else if (res > count) {
+			/*
+			 * For very small buffer sizes, it's possible that a
+			 * single record is too large due to a long filename.
+			 */
+			error = EINVAL;
+			break;
+		}
+
+		VERIFY(outb + res <= count);
+		if (copyout(lbuf, (void *)(uptr + outb), res) != 0) {
+			error = EFAULT;
+			break;
+		}
+		outb += res;
+
+		if (at_eof != 0 || (count - outb) < (lx_size + MAXPATHLEN)) {
+			/*
+			 * If there are no records left or the remaining buffer
+			 * space is not large enough to hold a max-length
+			 * filename, do not continue iteration.
+			 */
+			break;
+		}
+
+		/*
+		 * We undershot the request buffer.
+		 * Reset for another READDIR, taking care not to overshoot.
+		 */
+		bufsz = MIN(sbufsz, LTOS_GETDENTS_BUFSZ(count - outb, lx_size));
+		auio.uio_resid = bufsz;
+		aiov.iov_len = bufsz;
+		aiov.iov_base = sbuf;
+	}
+
+	kmem_free(lbuf, lbufsz);
+	kmem_free(sbuf, sbufsz);
+
+	if (error) {
+		releasef(fd);
+		return (set_errno(error));
+	}
+
+	fp->f_offset = auio.uio_loffset;
+	releasef(fd);
+	return (outb);
+}
+
+
+static int
+lx_getdents_format32(caddr_t sbuf, caddr_t lbuf, int len)
+{
+	struct dirent *sd;
+	struct lx_dirent_32 *ld;
+	int namelen;
+	int size = 0;
+
+	while (len > 0) {
+		sd = (struct dirent *)sbuf;
+		ld = (struct lx_dirent_32 *)lbuf;
+		namelen = MIN(strlen(sd->d_name), LX_NAMEMAX - 1);
+
+		ld->d_ino = sd->d_ino;
+		ld->d_off = sd->d_off;
+		(void) strncpy(ld->d_name, sd->d_name, namelen);
+		ld->d_name[namelen] = 0;
+		ld->d_reclen = (ushort_t)LX_RECLEN(namelen,
+		    struct lx_dirent_32);
+		/* Zero out any alignment padding and d_type */
+		bzero(ld->d_name + namelen,
+		    LX_ZEROLEN(namelen, struct lx_dirent_32));
+
+		len -= sd->d_reclen;
+		size += ld->d_reclen;
+		sbuf += sd->d_reclen;
+		lbuf += ld->d_reclen;
+	}
+	return (size);
+}
+
+static int
+lx_getdents_format64(caddr_t sbuf, caddr_t lbuf, int len)
+{
+	struct dirent *sd;
+	struct lx_dirent_64 *ld;
+	int namelen;
+	int size = 0;
+
+	while (len > 0) {
+		sd = (struct dirent *)sbuf;
+		ld = (struct lx_dirent_64 *)lbuf;
+		namelen = MIN(strlen(sd->d_name), LX_NAMEMAX - 1);
+
+		ld->d_ino = sd->d_ino;
+		ld->d_off = sd->d_off;
+		(void) strncpy(ld->d_name, sd->d_name, namelen);
+		ld->d_name[namelen] = 0;
+		ld->d_reclen = (ushort_t)LX_RECLEN(namelen,
+		    struct lx_dirent_64);
+		/* Zero out any alignment padding and d_type */
+		bzero(ld->d_name + namelen,
+		    LX_ZEROLEN(namelen, struct lx_dirent_64));
+
+		len -= sd->d_reclen;
+		size += ld->d_reclen;
+		sbuf += sd->d_reclen;
+		lbuf += ld->d_reclen;
+	}
+	return (size);
+}
+
+long
+lx_getdents_32(int fd, caddr_t buf, size_t count)
+{
+	return (lx_getdents_common(fd, buf, count,
+	    sizeof (struct lx_dirent_32), lx_getdents_format32));
+}
+
+long
+lx_getdents_64(int fd, caddr_t buf, size_t count)
+{
+	return (lx_getdents_common(fd, buf, count,
+	    sizeof (struct lx_dirent_64), lx_getdents_format64));
+}
+
+struct lx_dirent64 {
+	uint64_t	d_ino;
+	int64_t		d_off;
+	ushort_t	d_reclen;
+	uchar_t		d_type;
+	char		d_name[1];
+};
+
+#define	LX_RECLEN64(namelen)	\
+	((offsetof(struct lx_dirent64, d_name) + 1 + (namelen) + 7) & ~7)
+
+#define	LX_ZEROLEN64(namelen)	\
+	(LX_RECLEN64(namelen) -	\
+	((offsetof(struct lx_dirent64, d_name) + (namelen))))
+
+static int
+lx_getdents64_format(caddr_t sbuf, caddr_t lbuf, int len)
+{
+	struct dirent *sd;
+	struct lx_dirent64 *ld;
+	int namelen;
+	int size = 0;
+
+	while (len > 0) {
+		sd = (struct dirent *)sbuf;
+		ld = (struct lx_dirent64 *)lbuf;
+		namelen = MIN(strlen(sd->d_name), LX_NAMEMAX - 1);
+
+		ld->d_ino = sd->d_ino;
+		ld->d_off = sd->d_off;
+		ld->d_type = 0;
+		(void) strncpy(ld->d_name, sd->d_name, namelen);
+		ld->d_name[namelen] = 0;
+		ld->d_reclen = (ushort_t)LX_RECLEN64(namelen);
+		/* Zero out any alignment padding */
+		bzero(ld->d_name + namelen, LX_ZEROLEN64(namelen));
+
+		len -= sd->d_reclen;
+		size += ld->d_reclen;
+		sbuf += sd->d_reclen;
+		lbuf += ld->d_reclen;
+	}
+	return (size);
+}
+
+
+long
+lx_getdents64(int fd, caddr_t buf, size_t count)
+{
+	return (lx_getdents_common(fd, buf, count,
+	    sizeof (struct lx_dirent64), lx_getdents64_format));
+}
diff --git a/usr/src/uts/common/brand/lx/syscall/lx_getpid.c b/usr/src/uts/common/brand/lx/syscall/lx_getpid.c
new file mode 100644
index 0000000000..c2506f52c5
--- /dev/null
+++ b/usr/src/uts/common/brand/lx/syscall/lx_getpid.c
@@ -0,0 +1,79 @@
+/*
+ * CDDL HEADER START
+ *
+ * The contents of this file are subject to the terms of the
+ * Common Development and Distribution License (the "License").
+ * You may not use this file except in compliance with the License.
+ *
+ * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
+ * or http://www.opensolaris.org/os/licensing.
+ * See the License for the specific language governing permissions
+ * and limitations under the License.
+ *
+ * When distributing Covered Code, include this CDDL HEADER in each
+ * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
+ * If applicable, add the following below this CDDL HEADER, with the
+ * fields enclosed by brackets "[]" replaced with your own identifying
+ * information: Portions Copyright [yyyy] [name of copyright owner]
+ *
+ * CDDL HEADER END
+ */
+/*
+ * Copyright 2006 Sun Microsystems, Inc.  All rights reserved.
+ * Use is subject to license terms.
+ */
+/*
+ * Copyright 2015 Joyent, Inc.
+ */
+
+#include <sys/zone.h>
+#include <sys/types.h>
+#include <sys/systm.h>
+#include <sys/thread.h>
+#include <sys/cpuvar.h>
+#include <sys/brand.h>
+#include <sys/lx_brand.h>
+
+/*
+ * return the pid
+ */
+long
+lx_getpid(void)
+{
+	lx_lwp_data_t *lwpd = ttolxlwp(curthread);
+	long rv;
+
+	if (curproc->p_pid == curproc->p_zone->zone_proc_initpid) {
+		rv = 1;
+	} else {
+		VERIFY(lwpd != NULL);
+
+		if (lwpd->br_lx_thunk_pid != 0) {
+			rv = lwpd->br_lx_thunk_pid;
+		} else {
+			rv = lwpd->br_tgid;
+		}
+	}
+
+	return (rv);
+}
+
+/*
+ * return the parent pid
+ */
+long
+lx_getppid(void)
+{
+	return (lx_lwp_ppid(ttolwp(curthread), NULL, NULL));
+}
+
+/*
+ * return the thread id
+ */
+long
+lx_gettid(void)
+{
+	lx_lwp_data_t *lwpd = ttolxlwp(curthread);
+
+	return (lwpd->br_pid);
+}
diff --git a/usr/src/uts/common/brand/lx/syscall/lx_getrandom.c b/usr/src/uts/common/brand/lx/syscall/lx_getrandom.c
new file mode 100644
index 0000000000..acc4073483
--- /dev/null
+++ b/usr/src/uts/common/brand/lx/syscall/lx_getrandom.c
@@ -0,0 +1,33 @@
+/*
+ * This file and its contents are supplied under the terms of the
+ * Common Development and Distribution License ("CDDL"), version 1.0.
+ * You may only use this file in accordance with the terms of version
+ * 1.0 of the CDDL.
+ *
+ * A full copy of the text of the CDDL should have accompanied this
+ * source.  A copy of the CDDL is also available via the Internet at
+ * http://www.illumos.org/license/CDDL.
+ */
+
+/*
+ * Copyright 2015 Joyent, Inc.
+ */
+
+#include <sys/brand.h>
+#include <sys/lx_brand.h>
+
+/*
+ * From "uts/common/syscall/getrandom.c":
+ */
+extern int getrandom(void *, size_t, int);
+
+long
+lx_getrandom(void *bufp, size_t buflen, int flags)
+{
+	/*
+	 * According to signal(7), calls to getrandom(2) are restartable.
+	 */
+	ttolxlwp(curthread)->br_syscall_restart = B_TRUE;
+
+	return (getrandom(bufp, buflen, flags));
+}
diff --git a/usr/src/uts/common/brand/lx/syscall/lx_id.c b/usr/src/uts/common/brand/lx/syscall/lx_id.c
new file mode 100644
index 0000000000..baa41f52fa
--- /dev/null
+++ b/usr/src/uts/common/brand/lx/syscall/lx_id.c
@@ -0,0 +1,296 @@
+/*
+ * CDDL HEADER START
+ *
+ * The contents of this file are subject to the terms of the
+ * Common Development and Distribution License (the "License").
+ * You may not use this file except in compliance with the License.
+ *
+ * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
+ * or http://www.opensolaris.org/os/licensing.
+ * See the License for the specific language governing permissions
+ * and limitations under the License.
+ *
+ * When distributing Covered Code, include this CDDL HEADER in each
+ * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
+ * If applicable, add the following below this CDDL HEADER, with the
+ * fields enclosed by brackets "[]" replaced with your own identifying
+ * information: Portions Copyright [yyyy] [name of copyright owner]
+ *
+ * CDDL HEADER END
+ */
+/*
+ * Copyright 2006 Sun Microsystems, Inc.  All rights reserved.
+ * Use is subject to license terms.
+ */
+/*
+ * Copyright 2015 Joyent, Inc.
+ */
+
+#include <sys/types.h>
+#include <sys/systm.h>
+#include <sys/errno.h>
+#include <sys/zone.h>
+#include <sys/cred_impl.h>
+#include <sys/policy.h>
+
+typedef ushort_t	l_uid16_t;
+typedef ushort_t	l_gid16_t;
+typedef uint_t		l_uid_t;
+typedef uint_t		l_gid_t;
+
+#define	LINUX_UID16_TO_UID32(uid16)	\
+	(((uid16) == (l_uid16_t)-1) ? ((l_uid_t)-1) : (l_uid_t)(uid16))
+
+#define	LINUX_GID16_TO_GID32(gid16)     \
+	(((gid16) == (l_gid16_t)-1) ? ((l_gid_t)-1) : (l_gid_t)(gid16))
+
+#define	LX_NGROUPS_MAX	32
+extern int setgroups(int, gid_t *);
+
+/*
+ * This function is based on setreuid in common/syscall/uid.c and exists
+ * because illumos does not have a way to explicitly set the saved uid (suid)
+ * from any other system call.
+ */
+long
+lx_setresuid(l_uid_t ruid, l_uid_t euid, l_uid_t suid)
+{
+	proc_t	*p;
+	int	error = 0;
+	int	do_nocd = 0;
+	int	uidchge = 0;
+	uid_t	oldruid = ruid;
+	cred_t	*cr, *newcr;
+	zoneid_t zoneid = getzoneid();
+
+	if ((ruid != -1 && (ruid > MAXUID)) ||
+	    (euid != -1 && (euid > MAXUID)) ||
+	    (suid != -1 && (suid > MAXUID))) {
+		error = EINVAL;
+		goto done;
+	}
+
+	/*
+	 * Need to pre-allocate the new cred structure before grabbing
+	 * the p_crlock mutex.
+	 */
+	newcr = cralloc();
+
+	p = ttoproc(curthread);
+
+retry:
+	mutex_enter(&p->p_crlock);
+	cr = p->p_cred;
+
+	if (ruid != -1 &&
+	    ruid != cr->cr_ruid && ruid != cr->cr_uid &&
+	    ruid != cr->cr_suid && secpolicy_allow_setid(cr, ruid, B_FALSE)) {
+		error = EPERM;
+	} else if (euid != -1 &&
+	    euid != cr->cr_ruid && euid != cr->cr_uid &&
+	    euid != cr->cr_suid && secpolicy_allow_setid(cr, euid, B_FALSE)) {
+		error = EPERM;
+	} else if (suid != -1 &&
+	    suid != cr->cr_ruid && suid != cr->cr_uid &&
+	    suid != cr->cr_suid && secpolicy_allow_setid(cr, suid, B_FALSE)) {
+		error = EPERM;
+	} else {
+		if (!uidchge && ruid != -1 && cr->cr_ruid != ruid) {
+			/*
+			 * The ruid of the process is going to change. In order
+			 * to avoid a race condition involving the
+			 * process count associated with the newly given ruid,
+			 * we increment the count before assigning the
+			 * credential to the process.
+			 * To do that, we'll have to take pidlock, so we first
+			 * release p_crlock.
+			 */
+			mutex_exit(&p->p_crlock);
+			uidchge = 1;
+			mutex_enter(&pidlock);
+			upcount_inc(ruid, zoneid);
+			mutex_exit(&pidlock);
+			/*
+			 * As we released p_crlock we can't rely on the cr
+			 * we read. So retry the whole thing.
+			 */
+			goto retry;
+		}
+		crhold(cr);
+		crcopy_to(cr, newcr);
+		p->p_cred = newcr;
+
+		if (euid != -1)
+			newcr->cr_uid = euid;
+		if (suid != -1)
+			newcr->cr_suid = suid;
+		if (ruid != -1) {
+			oldruid = newcr->cr_ruid;
+			newcr->cr_ruid = ruid;
+			ASSERT(ruid != oldruid ? uidchge : 1);
+		}
+
+		/*
+		 * A process that gives up its privilege
+		 * must be marked to produce no core dump.
+		 */
+		if ((cr->cr_uid != newcr->cr_uid ||
+		    cr->cr_ruid != newcr->cr_ruid ||
+		    cr->cr_suid != newcr->cr_suid))
+			do_nocd = 1;
+
+		crfree(cr);
+	}
+	mutex_exit(&p->p_crlock);
+
+	/*
+	 * We decrement the number of processes associated with the oldruid
+	 * to match the increment above, even if the ruid of the process
+	 * did not change or an error occurred (oldruid == uid).
+	 */
+	if (uidchge) {
+		ASSERT(oldruid != -1 && ruid != -1);
+		mutex_enter(&pidlock);
+		upcount_dec(oldruid, zoneid);
+		mutex_exit(&pidlock);
+	}
+
+	if (error == 0) {
+		if (do_nocd) {
+			mutex_enter(&p->p_lock);
+			p->p_flag |= SNOCD;
+			mutex_exit(&p->p_lock);
+		}
+		crset(p, newcr);	/* broadcast to process threads */
+		goto done;
+	}
+	crfree(newcr);
+done:
+	if (error)
+		return (set_errno(error));
+	else
+		return (0);
+}
+
+long
+lx_setresuid16(l_uid16_t ruid16, l_uid16_t euid16, l_uid16_t suid16)
+{
+	long	rval;
+
+	rval = lx_setresuid(
+	    LINUX_UID16_TO_UID32(ruid16),
+	    LINUX_UID16_TO_UID32(euid16),
+	    LINUX_UID16_TO_UID32(suid16));
+
+	return (rval);
+}
+
+/*
+ * This function is based on setregid in common/syscall/gid.c
+ */
+long
+lx_setresgid(l_gid_t rgid, l_gid_t egid, l_gid_t sgid)
+{
+	proc_t	*p;
+	int	error = 0;
+	int	do_nocd = 0;
+	cred_t	*cr, *newcr;
+
+	if ((rgid != -1 && (rgid > MAXUID)) ||
+	    (egid != -1 && (egid > MAXUID)) ||
+	    (sgid != -1 && (sgid > MAXUID))) {
+		error = EINVAL;
+		goto done;
+	}
+
+	/*
+	 * Need to pre-allocate the new cred structure before grabbing
+	 * the p_crlock mutex.
+	 */
+	newcr = cralloc();
+
+	p = ttoproc(curthread);
+	mutex_enter(&p->p_crlock);
+	cr = p->p_cred;
+
+	if (rgid != -1 &&
+	    rgid != cr->cr_rgid && rgid != cr->cr_gid &&
+	    rgid != cr->cr_sgid && secpolicy_allow_setid(cr, -1, B_FALSE)) {
+		error = EPERM;
+	} else if (egid != -1 &&
+	    egid != cr->cr_rgid && egid != cr->cr_gid &&
+	    egid != cr->cr_sgid && secpolicy_allow_setid(cr, -1, B_FALSE)) {
+		error = EPERM;
+	} else if (sgid != -1 &&
+	    sgid != cr->cr_rgid && sgid != cr->cr_gid &&
+	    sgid != cr->cr_sgid && secpolicy_allow_setid(cr, -1, B_FALSE)) {
+		error = EPERM;
+	} else {
+		crhold(cr);
+		crcopy_to(cr, newcr);
+		p->p_cred = newcr;
+
+		if (egid != -1)
+			newcr->cr_gid = egid;
+		if (sgid != -1)
+			newcr->cr_sgid = sgid;
+		if (rgid != -1)
+			newcr->cr_rgid = rgid;
+
+		/*
+		 * A process that gives up its privilege
+		 * must be marked to produce no core dump.
+		 */
+		if ((cr->cr_gid != newcr->cr_gid ||
+		    cr->cr_rgid != newcr->cr_rgid ||
+		    cr->cr_sgid != newcr->cr_sgid))
+			do_nocd = 1;
+
+		crfree(cr);
+	}
+	mutex_exit(&p->p_crlock);
+
+	if (error == 0) {
+		if (do_nocd) {
+			mutex_enter(&p->p_lock);
+			p->p_flag |= SNOCD;
+			mutex_exit(&p->p_lock);
+		}
+		crset(p, newcr);	/* broadcast to process threads */
+		goto done;
+	}
+	crfree(newcr);
+done:
+	if (error)
+		return (set_errno(error));
+	else
+		return (0);
+}
+
+long
+lx_setresgid16(l_gid16_t rgid16, l_gid16_t egid16, l_gid16_t sgid16)
+{
+	long	rval;
+
+	rval = lx_setresgid(
+	    LINUX_GID16_TO_GID32(rgid16),
+	    LINUX_GID16_TO_GID32(egid16),
+	    LINUX_GID16_TO_GID32(sgid16));
+
+	return (rval);
+}
+
+/*
+ * Linux defines NGROUPS_MAX to be 32, but on illumos it is only 16. We employ
+ * the terrible hack below so that tests may proceed, if only on DEBUG kernels.
+ */
+long
+lx_helper_setgroups(int ngroups, gid_t *grouplist)
+{
+#ifdef DEBUG
+	if (ngroups > ngroups_max && ngroups <= LX_NGROUPS_MAX)
+		ngroups = ngroups_max;
+#endif /* DEBUG */
+
+	return (setgroups(ngroups, grouplist));
+}
diff --git a/usr/src/uts/common/brand/lx/syscall/lx_ioctl.c b/usr/src/uts/common/brand/lx/syscall/lx_ioctl.c
new file mode 100644
index 0000000000..b9c24198f9
--- /dev/null
+++ b/usr/src/uts/common/brand/lx/syscall/lx_ioctl.c
@@ -0,0 +1,1220 @@
+/*
+ * This file and its contents are supplied under the terms of the
+ * Common Development and Distribution License ("CDDL"), version 1.0.
+ * You may only use this file in accordance with the terms of version
+ * 1.0 of the CDDL.
+ *
+ * A full copy of the text of the CDDL should have accompanied this
+ * source.  A copy of the CDDL is also available via the Internet at
+ * http://www.illumos.org/license/CDDL.
+ */
+
+/*
+ * Copyright 2015 Joyent, Inc.  All rights reserved.
+ */
+
+#include <sys/errno.h>
+#include <sys/systm.h>
+#include <sys/file.h>
+#include <sys/filio.h>
+#include <sys/vnode.h>
+#include <sys/fcntl.h>
+#include <sys/termio.h>
+#include <sys/termios.h>
+#include <sys/ptyvar.h>
+#include <net/if.h>
+#include <net/if_dl.h>
+#include <sys/sockio.h>
+#include <sys/stropts.h>
+#include <sys/ptms.h>
+#include <sys/cred.h>
+#include <sys/cred_impl.h>
+#include <sys/sysmacros.h>
+#include <sys/lx_misc.h>
+#include <sys/lx_ptm.h>
+#include <sys/sunddi.h>
+#include <sys/thread.h>
+#include <sys/proc.h>
+#include <sys/session.h>
+#include <sys/kmem.h>
+#include <sys/socket.h>
+#include <sys/socketvar.h>
+#include <net/if_arp.h>
+#include <sys/ioccom.h>
+#include <sys/dtrace.h>
+#include <sys/ethernet.h>
+#include <sys/dlpi.h>
+
+/*
+ * Supported ioctls
+ */
+#define	LX_TCGETS		0x5401
+#define	LX_TCSETS		0x5402
+#define	LX_TCSETSW		0x5403
+#define	LX_TCSETSF		0x5404
+#define	LX_TCGETA		0x5405
+#define	LX_TCSETA		0x5406
+#define	LX_TCSETAW		0x5407
+#define	LX_TCSETAF		0x5408
+#define	LX_TCSBRK		0x5409
+#define	LX_TCXONC		0x540a
+#define	LX_TCFLSH		0x540b
+#define	LX_TIOCEXCL		0x540c
+#define	LX_TIOCNXCL		0x540d
+#define	LX_TIOCSCTTY		0x540e
+#define	LX_TIOCGPGRP		0x540f
+#define	LX_TIOCSPGRP		0x5410
+#define	LX_TIOCOUTQ		0x5411
+#define	LX_TIOCSTI		0x5412
+#define	LX_TIOCGWINSZ		0x5413
+#define	LX_TIOCSWINSZ		0x5414
+#define	LX_TIOCMGET		0x5415
+#define	LX_TIOCMBIS		0x5416
+#define	LX_TIOCMBIC		0x5417
+#define	LX_TIOCMSET		0x5418
+#define	LX_TIOCGSOFTCAR		0x5419
+#define	LX_TIOCSSOFTCAR		0x541a
+#define	LX_FIONREAD		0x541b
+#define	LX_TIOCPKT		0x5420
+#define	LX_FIONBIO		0x5421
+#define	LX_TIOCNOTTY		0x5422
+#define	LX_TIOCSETD		0x5423
+#define	LX_TIOCGETD		0x5424
+#define	LX_TCSBRKP		0x5425
+#define	LX_TIOCGSID		0x5429
+#define	LX_TIOCGPTN		0x80045430
+#define	LX_TIOCSPTLCK		0x40045431
+#define	LX_FIONCLEX		0x5450
+#define	LX_FIOCLEX		0x5451
+#define	LX_FIOASYNC		0x5452
+#define	LX_FIOSETOWN		0x8901
+#define	LX_SIOCSPGRP		0x8902
+#define	LX_FIOGETOWN		0x8903
+#define	LX_SIOCGPGRP		0x8904
+#define	LX_SIOCATMARK		0x8905
+#define	LX_SIOCGSTAMP		0x8906
+#define	LX_SIOCGIFCONF		0x8912
+#define	LX_SIOCGIFFLAGS		0x8913
+#define	LX_SIOCSIFFLAGS		0x8914
+#define	LX_SIOCGIFADDR		0x8915
+#define	LX_SIOCSIFADDR		0x8916
+#define	LX_SIOCGIFDSTADDR	0x8917
+#define	LX_SIOCSIFDSTADDR	0x8918
+#define	LX_SIOCGIFBRDADDR	0x8919
+#define	LX_SIOCSIFBRDADDR	0x891a
+#define	LX_SIOCGIFNETMASK	0x891b
+#define	LX_SIOCSIFNETMASK	0x891c
+#define	LX_SIOCGIFMETRIC	0x891d
+#define	LX_SIOCSIFMETRIC	0x891e
+#define	LX_SIOCGIFMEM		0x891f
+#define	LX_SIOCSIFMEM		0x8920
+#define	LX_SIOCGIFMTU		0x8921
+#define	LX_SIOCSIFMTU		0x8922
+#define	LX_SIOCSIFHWADDR	0x8924
+#define	LX_SIOCGIFHWADDR	0x8927
+#define	LX_SIOCGIFINDEX		0x8933
+#define	LX_SIOCGIFTXQLEN	0x8942
+
+#define	FLUSER(fp)	fp->f_flag | get_udatamodel()
+#define	FLFAKE(fp)	fp->f_flag | FKIOCTL
+
+/*
+ * LX_NCC must be different from LX_NCCS since while the termio and termios
+ * structures may look similar they are fundamentally different sizes and
+ * have different members.
+ */
+#define	LX_NCC	8
+#define	LX_NCCS	19
+
+struct lx_termio {
+	unsigned short c_iflag;		/* input mode flags */
+	unsigned short c_oflag;		/* output mode flags */
+	unsigned short c_cflag;		/* control mode flags */
+	unsigned short c_lflag;		/* local mode flags */
+	unsigned char c_line;		/* line discipline */
+	unsigned char c_cc[LX_NCC];	/* control characters */
+};
+
+struct lx_termios {
+	uint32_t c_iflag;		/* input mode flags */
+	uint32_t c_oflag;		/* output mode flags */
+	uint32_t c_cflag;		/* control mode flags */
+	uint32_t c_lflag;		/* local mode flags */
+	unsigned char c_line;		/* line discipline */
+	unsigned char c_cc[LX_NCCS];	/* control characters */
+};
+
+/*
+ * c_cc characters which are valid for lx_termio and lx_termios
+ */
+#define	LX_VINTR	0
+#define	LX_VQUIT	1
+#define	LX_VERASE	2
+#define	LX_VKILL	3
+#define	LX_VEOF		4
+#define	LX_VTIME	5
+#define	LX_VMIN		6
+#define	LX_VSWTC	7
+
+/*
+ * c_cc characters which are valid for lx_termios
+ */
+#define	LX_VSTART	8
+#define	LX_VSTOP	9
+#define	LX_VSUSP	10
+#define	LX_VEOL		11
+#define	LX_VREPRINT	12
+#define	LX_VDISCARD	13
+#define	LX_VWERASE	14
+#define	LX_VLNEXT	15
+#define	LX_VEOL2	16
+
+/* VSD key for lx_cc information */
+static uint_t lx_ioctl_vsd = 0;
+
+extern int lx_lpid_to_spair(pid_t l_pid, pid_t *s_pid, id_t *s_tid);
+
+/* Terminal helpers */
+
+static void
+l2s_termios(struct lx_termios *l_tios, struct termios *s_tios)
+{
+	ASSERT((l_tios != NULL) && (s_tios != NULL));
+
+	bzero(s_tios, sizeof (*s_tios));
+
+	s_tios->c_iflag = l_tios->c_iflag;
+	s_tios->c_oflag = l_tios->c_oflag;
+	s_tios->c_cflag = l_tios->c_cflag;
+	s_tios->c_lflag = l_tios->c_lflag;
+
+	if (s_tios->c_lflag & ICANON) {
+		s_tios->c_cc[VEOF] = l_tios->c_cc[LX_VEOF];
+		s_tios->c_cc[VEOL] = l_tios->c_cc[LX_VEOL];
+	} else {
+		s_tios->c_cc[VMIN] = l_tios->c_cc[LX_VMIN];
+		s_tios->c_cc[VTIME] = l_tios->c_cc[LX_VTIME];
+	}
+
+	s_tios->c_cc[VEOL2] = l_tios->c_cc[LX_VEOL2];
+	s_tios->c_cc[VERASE] = l_tios->c_cc[LX_VERASE];
+	s_tios->c_cc[VKILL] = l_tios->c_cc[LX_VKILL];
+	s_tios->c_cc[VREPRINT] = l_tios->c_cc[LX_VREPRINT];
+	s_tios->c_cc[VLNEXT] = l_tios->c_cc[LX_VLNEXT];
+	s_tios->c_cc[VWERASE] = l_tios->c_cc[LX_VWERASE];
+	s_tios->c_cc[VINTR] = l_tios->c_cc[LX_VINTR];
+	s_tios->c_cc[VQUIT] = l_tios->c_cc[LX_VQUIT];
+	s_tios->c_cc[VSWTCH] = l_tios->c_cc[LX_VSWTC];
+	s_tios->c_cc[VSTART] = l_tios->c_cc[LX_VSTART];
+	s_tios->c_cc[VSTOP] = l_tios->c_cc[LX_VSTOP];
+	s_tios->c_cc[VSUSP] = l_tios->c_cc[LX_VSUSP];
+	s_tios->c_cc[VDISCARD] = l_tios->c_cc[LX_VDISCARD];
+}
+
+static void
+l2s_termio(struct lx_termio *l_tio, struct termio *s_tio)
+{
+	ASSERT((l_tio != NULL) && (s_tio != NULL));
+
+	bzero(s_tio, sizeof (*s_tio));
+
+	s_tio->c_iflag = l_tio->c_iflag;
+	s_tio->c_oflag = l_tio->c_oflag;
+	s_tio->c_cflag = l_tio->c_cflag;
+	s_tio->c_lflag = l_tio->c_lflag;
+
+	if (s_tio->c_lflag & ICANON) {
+		s_tio->c_cc[VEOF] = l_tio->c_cc[LX_VEOF];
+	} else {
+		s_tio->c_cc[VMIN] = l_tio->c_cc[LX_VMIN];
+		s_tio->c_cc[VTIME] = l_tio->c_cc[LX_VTIME];
+	}
+
+	s_tio->c_cc[VINTR] = l_tio->c_cc[LX_VINTR];
+	s_tio->c_cc[VQUIT] = l_tio->c_cc[LX_VQUIT];
+	s_tio->c_cc[VERASE] = l_tio->c_cc[LX_VERASE];
+	s_tio->c_cc[VKILL] = l_tio->c_cc[LX_VKILL];
+	s_tio->c_cc[VSWTCH] = l_tio->c_cc[LX_VSWTC];
+}
+
+static void
+termios2lx_cc(struct lx_termios *l_tios, struct lx_cc *lio)
+{
+	ASSERT((l_tios != NULL) && (lio != NULL));
+
+	bzero(lio, sizeof (*lio));
+
+	lio->veof = l_tios->c_cc[LX_VEOF];
+	lio->veol = l_tios->c_cc[LX_VEOL];
+	lio->vmin = l_tios->c_cc[LX_VMIN];
+	lio->vtime = l_tios->c_cc[LX_VTIME];
+}
+
+static void
+termio2lx_cc(struct lx_termio *l_tio, struct lx_cc *lio)
+{
+	ASSERT((l_tio != NULL) && (lio != NULL));
+
+	bzero(lio, sizeof (*lio));
+
+	lio->veof = l_tio->c_cc[LX_VEOF];
+	lio->veol = 0;
+	lio->vmin = l_tio->c_cc[LX_VMIN];
+	lio->vtime = l_tio->c_cc[LX_VTIME];
+}
+
+static void
+s2l_termios(struct termios *s_tios, struct lx_termios *l_tios)
+{
+	ASSERT((s_tios != NULL) && (l_tios != NULL));
+
+	bzero(l_tios, sizeof (*l_tios));
+
+	l_tios->c_iflag = s_tios->c_iflag;
+	l_tios->c_oflag = s_tios->c_oflag;
+	l_tios->c_cflag = s_tios->c_cflag;
+	l_tios->c_lflag = s_tios->c_lflag;
+
+	if (s_tios->c_lflag & ICANON) {
+		l_tios->c_cc[LX_VEOF] = s_tios->c_cc[VEOF];
+		l_tios->c_cc[LX_VEOL] = s_tios->c_cc[VEOL];
+	} else {
+		l_tios->c_cc[LX_VMIN] = s_tios->c_cc[VMIN];
+		l_tios->c_cc[LX_VTIME] = s_tios->c_cc[VTIME];
+	}
+
+	l_tios->c_cc[LX_VEOL2] = s_tios->c_cc[VEOL2];
+	l_tios->c_cc[LX_VERASE] = s_tios->c_cc[VERASE];
+	l_tios->c_cc[LX_VKILL] = s_tios->c_cc[VKILL];
+	l_tios->c_cc[LX_VREPRINT] = s_tios->c_cc[VREPRINT];
+	l_tios->c_cc[LX_VLNEXT] = s_tios->c_cc[VLNEXT];
+	l_tios->c_cc[LX_VWERASE] = s_tios->c_cc[VWERASE];
+	l_tios->c_cc[LX_VINTR] = s_tios->c_cc[VINTR];
+	l_tios->c_cc[LX_VQUIT] = s_tios->c_cc[VQUIT];
+	l_tios->c_cc[LX_VSWTC] = s_tios->c_cc[VSWTCH];
+	l_tios->c_cc[LX_VSTART] = s_tios->c_cc[VSTART];
+	l_tios->c_cc[LX_VSTOP] = s_tios->c_cc[VSTOP];
+	l_tios->c_cc[LX_VSUSP] = s_tios->c_cc[VSUSP];
+	l_tios->c_cc[LX_VDISCARD] = s_tios->c_cc[VDISCARD];
+}
+
+static void
+s2l_termio(struct termio *s_tio, struct lx_termio *l_tio)
+{
+	ASSERT((s_tio != NULL) && (l_tio != NULL));
+
+	bzero(l_tio, sizeof (*l_tio));
+
+	l_tio->c_iflag = s_tio->c_iflag;
+	l_tio->c_oflag = s_tio->c_oflag;
+	l_tio->c_cflag = s_tio->c_cflag;
+	l_tio->c_lflag = s_tio->c_lflag;
+
+	if (s_tio->c_lflag & ICANON) {
+		l_tio->c_cc[LX_VEOF] = s_tio->c_cc[VEOF];
+	} else {
+		l_tio->c_cc[LX_VMIN] = s_tio->c_cc[VMIN];
+		l_tio->c_cc[LX_VTIME] = s_tio->c_cc[VTIME];
+	}
+
+	l_tio->c_cc[LX_VINTR] = s_tio->c_cc[VINTR];
+	l_tio->c_cc[LX_VQUIT] = s_tio->c_cc[VQUIT];
+	l_tio->c_cc[LX_VERASE] = s_tio->c_cc[VERASE];
+	l_tio->c_cc[LX_VKILL] = s_tio->c_cc[VKILL];
+	l_tio->c_cc[LX_VSWTC] = s_tio->c_cc[VSWTCH];
+}
+
+static void
+set_lx_cc(vnode_t *vp, struct lx_cc *lio)
+{
+	struct lx_cc *cur;
+	/*
+	 * Linux expects that the termio/termios control characters are
+	 * preserved more strictly than illumos supports.  In order to preserve
+	 * the illusion that the characters are maintained, they are stored as
+	 * vnode-specific data.
+	 */
+	mutex_enter(&vp->v_vsd_lock);
+	cur = (struct lx_cc *)vsd_get(vp, lx_ioctl_vsd);
+	if (cur == NULL) {
+		cur = kmem_alloc(sizeof (struct lx_cc), KM_SLEEP);
+		bcopy(lio, cur, sizeof (struct lx_cc));
+		(void) vsd_set(vp, lx_ioctl_vsd, cur);
+	} else {
+		bcopy(lio, cur, sizeof (struct lx_cc));
+	}
+	mutex_exit(&vp->v_vsd_lock);
+}
+
+static int
+get_lx_cc(vnode_t *vp, struct lx_cc *lio)
+{
+	struct lx_cc *cur;
+	int rv = 1;
+	mutex_enter(&vp->v_vsd_lock);
+	cur = (struct lx_cc *)vsd_get(vp, lx_ioctl_vsd);
+	if (cur != NULL) {
+		bcopy(cur, lio, sizeof (*lio));
+		rv = 0;
+	}
+	mutex_exit(&vp->v_vsd_lock);
+	return (rv);
+}
+
+/* Socket helpers */
+
+typedef struct lx_ifreq32 {
+	char	ifr_name[IFNAMSIZ];
+	union {
+		struct	sockaddr ifru_addr;
+	};
+} lx_ifreq32_t;
+
+typedef struct lx_ifreq64 {
+	char	ifr_name[IFNAMSIZ];
+	union {
+		struct	sockaddr ifru_addr;
+		/* pad this out to the Linux size */
+		uint64_t	ifmap[3];
+	};
+} lx_ifreq64_t;
+
+typedef struct lx_ifconf32 {
+	int32_t	if_len;
+	caddr32_t if_buf;
+} lx_ifconf32_t;
+
+typedef struct lx_ifconf64 {
+	int32_t	if_len;
+	caddr_t if_buf;
+} lx_ifconf64_t;
+
+
+/* Generic translators */
+
+static int
+ict_pass(file_t *fp, int cmd, intptr_t arg, int lxcmd)
+{
+	int error = 0;
+	int rv;
+
+	error = VOP_IOCTL(fp->f_vnode, cmd, arg, FLUSER(fp), fp->f_cred, &rv,
+	    NULL);
+	return ((error != 0) ? set_errno(error) : 0);
+}
+
+static int
+ict_fionbio(file_t *fp, int cmd, intptr_t arg, int lxcmd)
+{
+	vnode_t *vp;
+	int32_t iflag, flags;
+	int error;
+
+	if (copyin((caddr_t)arg, &iflag, sizeof (iflag)))
+		return (set_errno(EFAULT));
+
+	mutex_enter(&fp->f_tlock);
+	vp = fp->f_vnode;
+	flags = fp->f_flag;
+	/* Linux sets NONBLOCK instead of FIONBIO */
+	if (iflag)
+		flags |= FNONBLOCK;
+	else
+		flags &= ~FNONBLOCK;
+	/* push the flag down */
+	error = VOP_SETFL(vp, fp->f_flag, flags, fp->f_cred, NULL);
+	fp->f_flag = flags;
+	mutex_exit(&fp->f_tlock);
+	return ((error != 0) ? set_errno(error) : 0);
+}
+
+static int
+ict_fionread(file_t *fp, int cmd, intptr_t arg, int lxcmd)
+{
+	vnode_t *vp;
+	struct vattr vattr;
+	int error = 0;
+	int rv;
+	/*
+	 * offset is int32_t because that is what FIONREAD is defined in terms
+	 * of.  We cap at INT_MAX as in other cases for this ioctl.
+	 */
+	int32_t offset;
+
+	vp = fp->f_vnode;
+
+	if (vp->v_type == VREG || vp->v_type == VDIR) {
+		vattr.va_mask = AT_SIZE;
+		error = VOP_GETATTR(vp, &vattr, 0, fp->f_cred, NULL);
+		if (error != 0)
+			return (set_errno(error));
+		offset = MIN(vattr.va_size - fp->f_offset, INT_MAX);
+		if (copyout(&offset, (caddr_t)arg, sizeof (offset)))
+			return (set_errno(EFAULT));
+	} else {
+		error = VOP_IOCTL(vp, FIONREAD, arg, FLUSER(fp), fp->f_cred,
+		    &rv, NULL);
+		if (error)
+			return (set_errno(error));
+	}
+	return (0);
+}
+
+/* Terminal-related translators */
+
+static int
+ict_tcsets(file_t *fp, int cmd, intptr_t arg, int lxcmd)
+{
+	struct lx_termios	l_tios;
+	struct termios		s_tios;
+	struct lx_cc		lio;
+	int			error, rv;
+
+	ASSERT(cmd == TCSETS || cmd == TCSETSW || cmd == TCSETSF);
+
+	if (copyin((struct lx_termios *)arg, &l_tios, sizeof (l_tios)) != 0)
+		return (set_errno(EFAULT));
+	termios2lx_cc(&l_tios, &lio);
+	l2s_termios(&l_tios, &s_tios);
+
+	error = VOP_IOCTL(fp->f_vnode, cmd, (intptr_t)&s_tios,
+	    FLFAKE(fp), fp->f_cred, &rv, NULL);
+	if (error)
+		return (set_errno(error));
+	/* preserve lx_cc */
+	set_lx_cc(fp->f_vnode, &lio);
+
+	return (0);
+}
+
+static int
+ict_tcseta(file_t *fp, int cmd, intptr_t arg, int lxcmd)
+{
+	struct lx_termio	l_tio;
+	struct termio		s_tio;
+	struct lx_cc		lio;
+	int			error, rv;
+
+	ASSERT(cmd == TCSETA || cmd == TCSETAW || cmd == TCSETAF);
+
+	if (copyin((struct lx_termio *)arg, &l_tio, sizeof (l_tio)) != 0)
+		return (set_errno(EFAULT));
+	l2s_termio(&l_tio, &s_tio);
+	termio2lx_cc(&l_tio, &lio);
+
+	error = VOP_IOCTL(fp->f_vnode, cmd, (intptr_t)&s_tio,
+	    FLFAKE(fp), fp->f_cred, &rv, NULL);
+	if (error)
+		return (set_errno(error));
+	/* preserve lx_cc */
+	set_lx_cc(fp->f_vnode, &lio);
+
+	return (0);
+}
+
+static int
+ict_tcgets_ptm(file_t *fp, int cmd, intptr_t arg, int lxcmd)
+{
+	struct lx_termios	l_tios;
+	struct termios		s_tios, *s_tiosd;
+	uint_t			s_tiosl;
+
+	/* get termios defaults */
+	if (ddi_prop_lookup_byte_array(DDI_DEV_T_ANY, ddi_root_node(),
+	    DDI_PROP_NOTPROM, "ttymodes", (uchar_t **)&s_tiosd,
+	    &s_tiosl) != DDI_SUCCESS)
+		return (EIO);
+	ASSERT(s_tiosl == sizeof (*s_tiosd));
+	bcopy(s_tiosd, &s_tios, sizeof (s_tios));
+	ddi_prop_free(s_tiosd);
+
+	/* Now munge the data to how Linux wants it. */
+	s2l_termios(&s_tios, &l_tios);
+	if (copyout(&l_tios, (struct lx_termios *)arg, sizeof (l_tios)) != 0)
+		return (set_errno(EFAULT));
+
+	return (0);
+}
+
+static int
+ict_tcgets_native(file_t *fp, int cmd, intptr_t arg, int lxcmd)
+{
+	struct lx_termios	l_tios;
+	struct termios		s_tios;
+	struct lx_cc		lio;
+	int			error, rv;
+
+	error = VOP_IOCTL(fp->f_vnode, cmd, (intptr_t)&s_tios,
+	    FLFAKE(fp), fp->f_cred, &rv, NULL);
+	if (error)
+		return (set_errno(error));
+
+	/* Now munge the data to how Linux wants it. */
+	s2l_termios(&s_tios, &l_tios);
+
+	/* return preserved lx_cc */
+	if (get_lx_cc(fp->f_vnode, &lio) == 0) {
+		l_tios.c_cc[LX_VEOF] = lio.veof;
+		l_tios.c_cc[LX_VEOL] = lio.veol;
+		l_tios.c_cc[LX_VMIN] = lio.vmin;
+		l_tios.c_cc[LX_VTIME] = lio.vtime;
+	}
+
+	if (copyout(&l_tios, (struct lx_termios *)arg, sizeof (l_tios)) != 0)
+		return (set_errno(EFAULT));
+
+	return (0);
+}
+
+static int
+ict_tcgets(file_t *fp, int cmd, intptr_t arg, int lxcmd)
+{
+	if (getmajor(fp->f_vnode->v_rdev) == ddi_name_to_major(LX_PTM_DRV))
+		return (ict_tcgets_ptm(fp, cmd, arg, lxcmd));
+	else
+		return (ict_tcgets_native(fp, cmd, arg, lxcmd));
+}
+
+static int
+ict_tcgeta(file_t *fp, int cmd, intptr_t arg, int lxcmd)
+{
+	struct lx_termio	l_tio;
+	struct termio		s_tio;
+	struct lx_cc		lio;
+	int			error, rv;
+
+	error = VOP_IOCTL(fp->f_vnode, cmd, (intptr_t)&s_tio,
+	    FLFAKE(fp), fp->f_cred, &rv, NULL);
+	if (error)
+		return (set_errno(error));
+
+	s2l_termio(&s_tio, &l_tio);
+	/* return preserved lx_cc */
+	if (get_lx_cc(fp->f_vnode, &lio) == 0) {
+		l_tio.c_cc[LX_VEOF] = lio.veof;
+		l_tio.c_cc[LX_VMIN] = lio.vmin;
+		l_tio.c_cc[LX_VTIME] = lio.vtime;
+	}
+
+	if (copyout(&l_tio, (struct lx_termios *)arg, sizeof (l_tio)) != 0)
+		return (set_errno(EFAULT));
+
+	return (0);
+}
+
+static int
+ict_tiocspgrp(file_t *fp, int cmd, intptr_t arg, int lxcmd)
+{
+	pid_t lpid, spid, tid;
+	int error, rv;
+
+	/* Converting to the illumos pid is necessary */
+	if (copyin((pid_t *)arg, &lpid, sizeof (lpid)) < 0)
+		return (set_errno(EFAULT));
+	if (lx_lpid_to_spair(lpid, &spid, &tid) < 0)
+		return (set_errno(EPERM));
+
+	error = VOP_IOCTL(fp->f_vnode, cmd, (intptr_t)&spid,
+	    fp->f_flag |FKIOCTL, fp->f_cred, &rv, NULL);
+	return ((error != 0) ? set_errno(error) : 0);
+}
+
+static int
+ict_tcsbrkp(file_t *fp, int cmd, intptr_t arg, int lxcmd)
+{
+	int rv, error;
+	/* use null duration to emulate TCSBRKP */
+	int dur = 0;
+	error = VOP_IOCTL(fp->f_vnode, TCSBRK, (intptr_t)&dur,
+	    FLFAKE(fp), fp->f_cred, &rv, NULL);
+	return ((error != 0) ? set_errno(error) : 0);
+}
+
+static int
+ict_tiocgpgrp(file_t *fp, int cmd, intptr_t arg, int lxcmd)
+{
+	pid_t	spgrp;
+	int	error, rv;
+
+	error = VOP_IOCTL(fp->f_vnode, cmd, (intptr_t)&spgrp, FLFAKE(fp),
+	    fp->f_cred, &rv, NULL);
+	if (error == 0) {
+		if (spgrp == curproc->p_zone->zone_proc_initpid) {
+			spgrp = 1;
+		}
+		if (copyout(&spgrp, (caddr_t)arg, sizeof (spgrp))) {
+			return (set_errno(EFAULT));
+		}
+	}
+	return ((error != 0) ? set_errno(error) : 0);
+}
+
+static int
+ict_sptlock(file_t *fp, int cmd, intptr_t arg, int lxcmd)
+{
+	struct strioctl istr;
+	int error, rv;
+
+	istr.ic_cmd = UNLKPT;
+	istr.ic_len = 0;
+	istr.ic_timout = 0;
+	istr.ic_dp = NULL;
+	error = VOP_IOCTL(fp->f_vnode, I_STR, (intptr_t)&istr,
+	    fp->f_flag |FKIOCTL, fp->f_cred, &rv, NULL);
+	/*
+	 * The success/fail return values are different between Linux
+	 * and illumos.   Linux expects 0 or -1.  Illumos can return
+	 * positive number on success.
+	 */
+	return ((error != 0) ? set_errno(error) : 0);
+}
+
+static int
+ict_gptn(file_t *fp, int cmd, intptr_t arg, int lxcmd)
+{
+	struct strioctl istr;
+	cred_t *cr;
+	pt_own_t pto;
+	int error, rv;
+	int ptyno;
+
+	/* This operation is only valid for the lx_ptm device. */
+	if (getmajor(fp->f_vnode->v_rdev) != ddi_name_to_major(LX_PTM_DRV))
+		return (set_errno(ENOTTY));
+
+	cr = CRED();
+	pto.pto_ruid = cr->cr_uid;
+	pto.pto_rgid = cr->cr_gid;
+
+	istr.ic_cmd = OWNERPT;
+	istr.ic_len = sizeof (pto);
+	istr.ic_timout = 0;
+	istr.ic_dp = (char *)&pto;
+	error = VOP_IOCTL(fp->f_vnode, I_STR, (intptr_t)&istr,
+	    FLFAKE(fp), fp->f_cred, &rv, NULL);
+
+	if (error)
+		return (set_errno((error == ENOTTY) ? error: EACCES));
+
+	ptyno = getminor(fp->f_vnode->v_rdev) - 1;
+	if (copyout(&ptyno, (caddr_t)arg, sizeof (ptyno)))
+		return (set_errno(EFAULT));
+
+	return (0);
+}
+
+static int
+ict_tiocgwinsz(file_t *fp, int cmd, intptr_t arg, int lxcmd)
+{
+	int error, rv;
+	struct winsize winsize;
+
+	error = VOP_IOCTL(fp->f_vnode, cmd, arg, FLUSER(fp), fp->f_cred, &rv,
+	    NULL);
+
+	if (error == EINVAL) {
+		bzero(&winsize, sizeof (winsize));
+		if (copyout(&winsize, (caddr_t)arg, sizeof (winsize)))
+			return (set_errno(EFAULT));
+	} else if (error != 0) {
+		return (set_errno(error));
+	}
+
+	return (0);
+}
+
+static int
+ict_tiocsctty(file_t *fp, int cmd, intptr_t arg, int lxcmd)
+{
+	pid_t	ttysid, mysid;
+	int	error, rv;
+	proc_t *p = curproc;
+
+	/* getsid */
+	mutex_enter(&p->p_splock);
+	mysid = p->p_sessp->s_sid;
+	mutex_exit(&p->p_splock);
+
+	/*
+	 * Report success if we already control the tty.
+	 * If no one controls it, TIOCSCTTY will change that later.
+	 */
+	error = VOP_IOCTL(fp->f_vnode, TIOCGSID, (intptr_t)&ttysid,
+	    FLFAKE(fp), fp->f_cred, &rv, NULL);
+	if (error == 0 && ttysid == mysid)
+		return (0);
+
+	/*
+	 * Need to make sure we're a session leader, otherwise the
+	 * TIOCSCTTY ioctl will fail.
+	 */
+	mutex_enter(&pidlock);
+	if (p->p_sessp->s_sidp != p->p_pidp && !pgmembers(p->p_pid)) {
+		mutex_exit(&pidlock);
+		sess_create();
+	} else {
+		mutex_exit(&pidlock);
+	}
+
+	error = VOP_IOCTL(fp->f_vnode, cmd, 0, FLUSER(fp),
+	    fp->f_cred, &rv, NULL);
+	return ((error != 0) ? set_errno(error) : 0);
+}
+
+/* Socket-related translators */
+
+static int
+ict_siocatmark(file_t *fp, int cmd, intptr_t arg, int lxcmd)
+{
+	vnode_t *vp = fp->f_vnode;
+	int error, rv;
+	/*
+	 * Linux expects a SIOCATMARK of a UDP socket to return ENOTTY, while
+	 * Illumos allows it. Linux prior to 2.6.39 returned EINVAL for this.
+	 */
+	if (vp->v_type != VSOCK || VTOSO(vp)->so_type != SOCK_STREAM)
+		return (set_errno(ENOTTY));
+
+	error = VOP_IOCTL(fp->f_vnode, cmd, arg, FLUSER(fp), fp->f_cred, &rv,
+	    NULL);
+	if (error)
+		return (set_errno(error));
+
+	return (0);
+}
+
+static int
+ict_if_ioctl(vnode_t *vn, int cmd, intptr_t arg, int flags, cred_t *cred)
+{
+	int error, rv;
+	lx_zone_data_t *lxzd = ztolxzd(curproc->p_zone);
+	ksocket_t ks;
+
+	ASSERT(lxzd != NULL);
+	ks = lxzd->lxzd_ioctl_sock;
+
+	/*
+	 * For ioctls of this type, Illumos is strict about address family
+	 * whereas Linux is lenient.  This strictness can be avoided by using
+	 * an internal AF_INET ksocket.
+	 */
+	if (ks != NULL) {
+		error = ksocket_ioctl(ks, cmd, arg, &rv, cred);
+	} else {
+		error = VOP_IOCTL(vn, cmd, arg, flags, cred, &rv, NULL);
+	}
+
+	return (error);
+}
+
+static int
+ict_sioghwaddr(file_t *fp, struct lifreq *lreq)
+{
+	struct sockaddr_dl *sdl = (struct sockaddr_dl *)&lreq->lifr_addr;
+	struct sockaddr hwaddr;
+	int error, size;
+
+	error = ict_if_ioctl(fp->f_vnode, SIOCGLIFHWADDR, (intptr_t)lreq,
+	    FLFAKE(fp), fp->f_cred);
+
+	if (error == EADDRNOTAVAIL &&
+	    strncmp(lreq->lifr_name, "lo", 2) == 0) {
+		/* Emulate success on suspected loopbacks */
+		sdl->sdl_type = DL_LOOP;
+		sdl->sdl_alen = ETHERADDRL;
+		bzero(LLADDR(sdl), sdl->sdl_alen);
+		error = 0;
+	}
+
+	if (error == 0) {
+		bzero(&hwaddr, sizeof (hwaddr));
+		lx_stol_hwaddr(sdl, &hwaddr, &size);
+		bcopy(&hwaddr, &lreq->lifr_addr,
+		    size + sizeof (sdl->sdl_family));
+	}
+
+	return (error);
+}
+
+static int
+ict_siolifreq(file_t *fp, int cmd, intptr_t arg, int lxcmd)
+{
+	struct ifreq	req;
+	struct lifreq	lreq;
+	int		error, len;
+
+	/* Convert from Linux ifreq to illumos lifreq */
+	if (curproc->p_model == DATAMODEL_LP64)
+		len = sizeof (lx_ifreq64_t);
+	else
+		len = sizeof (lx_ifreq32_t);
+	if (copyin((struct ifreq *)arg, &req, len) != 0)
+		return (set_errno(EFAULT));
+	bzero(&lreq, sizeof (lreq));
+	strncpy(lreq.lifr_name, req.ifr_name, IFNAMSIZ);
+	bcopy(&req.ifr_ifru, &lreq.lifr_lifru, len - IFNAMSIZ);
+	lx_ifname_convert(lreq.lifr_name, LX_IF_TONATIVE);
+
+	switch (cmd) {
+	case SIOCGIFADDR:
+	case SIOCSIFADDR:
+	case SIOCGIFDSTADDR:
+	case SIOCSIFDSTADDR:
+	case SIOCGIFBRDADDR:
+	case SIOCSIFBRDADDR:
+	case SIOCGIFNETMASK:
+	case SIOCSIFNETMASK:
+	case SIOCGIFMETRIC:
+	case SIOCSIFMETRIC:
+	case SIOCGIFMTU:
+	case SIOCSIFMTU:
+		/*
+		 * Convert cmd from SIO*IF* to SIO*LIF*.
+		 * This is needed since Linux allows ifreq operations on ipv6
+		 * sockets where illumos does not.
+		 */
+		cmd = ((cmd & IOC_INOUT) |
+		    _IOW('i', ((cmd & 0xff) + 100), struct lifreq));
+		error = ict_if_ioctl(fp->f_vnode, cmd, (intptr_t)&lreq,
+		    FLFAKE(fp), fp->f_cred);
+		break;
+	case SIOCGIFINDEX:
+		cmd = SIOCGLIFINDEX;
+		error = ict_if_ioctl(fp->f_vnode, cmd, (intptr_t)&lreq,
+		    FLFAKE(fp), fp->f_cred);
+		break;
+	case SIOCGIFFLAGS:
+		cmd = SIOCGLIFFLAGS;
+		error = ict_if_ioctl(fp->f_vnode, cmd, (intptr_t)&lreq,
+		    FLFAKE(fp), fp->f_cred);
+		if (error == 0)
+			lx_ifflags_convert(&lreq.lifr_flags, LX_IF_FROMNATIVE);
+		break;
+	case SIOCSIFFLAGS:
+		cmd = SIOCSLIFFLAGS;
+		lx_ifflags_convert(&lreq.lifr_flags, LX_IF_TONATIVE);
+		error = ict_if_ioctl(fp->f_vnode, cmd, (intptr_t)&lreq,
+		    FLFAKE(fp), fp->f_cred);
+		break;
+	case SIOCGIFHWADDR:
+		error = ict_sioghwaddr(fp, &lreq);
+		break;
+	case LX_SIOCGIFTXQLEN:
+		/*
+		 * Illumos lacks the notion of txqlen.  Confirm the provided
+		 * interface is valid with SIOCGLIFINDEX and return a fake
+		 * txqlen of 1.  Loopback devices will report txqlen of 0.
+		 */
+		if (strncmp(lreq.lifr_name, "lo", 2) == 0) {
+			lreq.lifr_index = 0;
+			error = 0;
+			break;
+		}
+		cmd = SIOCGLIFINDEX;
+		error = ict_if_ioctl(fp->f_vnode, cmd, (intptr_t)&lreq,
+		    FLFAKE(fp), fp->f_cred);
+		if (error == 0) {
+			/* lifr_index aliases to the qlen field */
+			lreq.lifr_index = 1;
+		}
+		break;
+	case LX_SIOCSIFHWADDR:
+		/*
+		 * We're not going to support SIOCSIFHWADDR, but we need to be
+		 * able to check the result of the copyin first to see if the
+		 * command should have returned EFAULT.
+		 */
+	default:
+		error = EINVAL;
+	}
+
+	if (error != 0)
+		return (set_errno(error));
+
+	/* Convert back to a Linux ifreq */
+	lx_ifname_convert(lreq.lifr_name, LX_IF_FROMNATIVE);
+	bzero(&req, sizeof (req));
+	strncpy(req.ifr_name, lreq.lifr_name, IFNAMSIZ);
+	bcopy(&lreq.lifr_lifru, &req.ifr_ifru, len - IFNAMSIZ);
+
+	if (copyout(&req, (struct lifreq *)arg, len) != 0)
+		return (set_errno(EFAULT));
+
+	return (0);
+}
+
+static int
+ict_siocgifconf32(file_t *fp, int cmd, intptr_t arg, int lxcmd)
+{
+	lx_ifconf32_t	conf;
+	lx_ifreq32_t	*oreq;
+	struct ifconf	sconf;
+	int		ifcount, error, i, buf_len;
+
+	if (copyin((lx_ifconf32_t *)arg, &conf, sizeof (conf)) != 0)
+		return (set_errno(EFAULT));
+
+	/* They want to know how many interfaces there are. */
+	if (conf.if_len <= 0 || conf.if_buf == NULL) {
+		error = ict_if_ioctl(fp->f_vnode, SIOCGIFNUM,
+		    (intptr_t)&ifcount, FLFAKE(fp), fp->f_cred);
+		if (error != 0)
+			return (set_errno(error));
+
+		conf.if_len = ifcount * sizeof (lx_ifreq32_t);
+
+		if (copyout(&conf, (lx_ifconf32_t *)arg, sizeof (conf)) != 0)
+			return (set_errno(EFAULT));
+		return (0);
+	} else {
+		ifcount = conf.if_len / sizeof (lx_ifreq32_t);
+	}
+
+	/* Get interface configuration list. */
+	sconf.ifc_len = ifcount * sizeof (struct ifreq);
+	sconf.ifc_req = (struct ifreq *)kmem_alloc(sconf.ifc_len, KM_SLEEP);
+
+	error = ict_if_ioctl(fp->f_vnode, cmd, (intptr_t)&sconf, FLFAKE(fp),
+	    fp->f_cred);
+	if (error != 0) {
+		kmem_free(sconf.ifc_req, ifcount * sizeof (struct ifreq));
+		return (set_errno(error));
+	}
+
+	/* Convert data to Linux format & rename interfaces */
+	buf_len = ifcount * sizeof (lx_ifreq32_t);
+	oreq = (lx_ifreq32_t *)kmem_alloc(buf_len, KM_SLEEP);
+	for (i = 0; i < sconf.ifc_len / sizeof (struct ifreq); i++) {
+		bcopy(&sconf.ifc_req[i], oreq + i, sizeof (lx_ifreq32_t));
+		lx_ifname_convert(oreq[i].ifr_name, LX_IF_FROMNATIVE);
+	}
+	conf.if_len = i * sizeof (*oreq);
+	kmem_free(sconf.ifc_req, ifcount * sizeof (struct ifreq));
+
+	error = 0;
+	if (copyout(oreq, (caddr_t)(uintptr_t)conf.if_buf, conf.if_len) != 0 ||
+	    copyout(&conf, (lx_ifconf32_t *)arg, sizeof (conf)) != 0)
+		error = set_errno(EFAULT);
+
+	kmem_free(oreq, buf_len);
+	return (error);
+}
+
+static int
+ict_siocgifconf64(file_t *fp, int cmd, intptr_t arg, int lxcmd)
+{
+	lx_ifconf64_t	conf;
+	lx_ifreq64_t	*oreq;
+	struct ifconf	sconf;
+	int		ifcount, error, i, buf_len;
+
+	if (copyin((lx_ifconf64_t *)arg, &conf, sizeof (conf)) != 0)
+		return (set_errno(EFAULT));
+
+	/* They want to know how many interfaces there are. */
+	if (conf.if_len <= 0 || conf.if_buf == NULL) {
+		error = ict_if_ioctl(fp->f_vnode, SIOCGIFNUM,
+		    (intptr_t)&ifcount, FLFAKE(fp), fp->f_cred);
+		if (error != 0)
+			return (set_errno(error));
+
+		conf.if_len = ifcount * sizeof (lx_ifreq64_t);
+
+		if (copyout(&conf, (lx_ifconf64_t *)arg, sizeof (conf)) != 0)
+			return (set_errno(EFAULT));
+		return (0);
+	} else {
+		ifcount = conf.if_len / sizeof (lx_ifreq64_t);
+	}
+
+	/* Get interface configuration list. */
+	sconf.ifc_len = ifcount * sizeof (struct ifreq);
+	sconf.ifc_req = (struct ifreq *)kmem_alloc(sconf.ifc_len, KM_SLEEP);
+
+	error = ict_if_ioctl(fp->f_vnode, cmd, (intptr_t)&sconf, FLFAKE(fp),
+	    fp->f_cred);
+	if (error != 0) {
+		kmem_free(sconf.ifc_req, ifcount * sizeof (struct ifreq));
+		return (set_errno(error));
+	}
+
+	/* Convert data to Linux format & rename interfaces */
+	buf_len = ifcount * sizeof (lx_ifreq64_t);
+	oreq = (lx_ifreq64_t *)kmem_alloc(buf_len, KM_SLEEP);
+	for (i = 0; i < sconf.ifc_len / sizeof (struct ifreq); i++) {
+		bcopy(&sconf.ifc_req[i], oreq + i, sizeof (lx_ifreq64_t));
+		lx_ifname_convert(oreq[i].ifr_name, LX_IF_FROMNATIVE);
+	}
+	conf.if_len = i * sizeof (*oreq);
+	kmem_free(sconf.ifc_req, ifcount * sizeof (struct ifreq));
+
+	error = 0;
+	if (copyout(oreq, (caddr_t)(uintptr_t)conf.if_buf, conf.if_len) != 0 ||
+	    copyout(&conf, (lx_ifconf64_t *)arg, sizeof (conf)) != 0)
+		error = set_errno(EFAULT);
+
+	kmem_free(oreq, buf_len);
+	return (error);
+}
+
+static int
+ict_siocgifconf(file_t *fp, int cmd, intptr_t arg, int lxcmd)
+{
+	if (curproc->p_model == DATAMODEL_LP64)
+		return (ict_siocgifconf64(fp, cmd, arg, lxcmd));
+	else
+		return (ict_siocgifconf32(fp, cmd, arg, lxcmd));
+}
+
+/* Structure used to define an ioctl translator. */
+typedef struct ioc_cmd_translator {
+	int	ict_lxcmd;
+	int	ict_cmd;
+	int	(*ict_func)(file_t *fp, int cmd, intptr_t arg, int lxcmd);
+} ioc_cmd_translator_t;
+
+#define	IOC_CMD_TRANSLATOR_PASS(ioc_cmd_sym)				\
+	{ (int)LX_##ioc_cmd_sym, (int)ioc_cmd_sym, ict_pass },
+
+#define	IOC_CMD_TRANSLATOR_FILTER(ioc_cmd_sym, ioct_handler)		\
+	{ (int)LX_##ioc_cmd_sym, (int)ioc_cmd_sym, ioct_handler },
+
+#define	IOC_CMD_TRANSLATOR_CUSTOM(ioc_cmd_sym, ioct_handler)		\
+	{ (int)ioc_cmd_sym, (int)ioc_cmd_sym, ioct_handler },
+
+#define	IOC_CMD_TRANSLATOR_PTHRU(ioc_cmd_sym)				\
+	{ (int)ioc_cmd_sym, (int)ioc_cmd_sym, ict_pass },
+
+#define	IOC_CMD_TRANSLATOR_END						\
+	{0, 0, NULL}
+
+static ioc_cmd_translator_t ioc_translators[] = {
+	IOC_CMD_TRANSLATOR_FILTER(FIONBIO,	ict_fionbio)
+	IOC_CMD_TRANSLATOR_FILTER(FIONREAD,	ict_fionread)
+	IOC_CMD_TRANSLATOR_PASS(FIOGETOWN)
+	IOC_CMD_TRANSLATOR_PASS(FIOASYNC)
+
+	/* streams related */
+	IOC_CMD_TRANSLATOR_PASS(TCXONC)
+	IOC_CMD_TRANSLATOR_PASS(TCFLSH)
+	IOC_CMD_TRANSLATOR_PASS(TIOCEXCL)
+	IOC_CMD_TRANSLATOR_PASS(TIOCNXCL)
+	IOC_CMD_TRANSLATOR_PASS(TIOCSTI)
+	IOC_CMD_TRANSLATOR_PASS(TIOCSWINSZ)
+	IOC_CMD_TRANSLATOR_PASS(TIOCMBIS)
+	IOC_CMD_TRANSLATOR_PASS(TIOCMBIC)
+	IOC_CMD_TRANSLATOR_PASS(TIOCMSET)
+	IOC_CMD_TRANSLATOR_PASS(TIOCSETD)
+	IOC_CMD_TRANSLATOR_PASS(TCSBRK)
+
+	/* terminal related */
+	IOC_CMD_TRANSLATOR_PASS(TIOCGETD)
+	IOC_CMD_TRANSLATOR_PASS(TIOCGSID)
+	IOC_CMD_TRANSLATOR_PASS(TIOCNOTTY)
+	IOC_CMD_TRANSLATOR_PASS(TIOCPKT)
+
+	IOC_CMD_TRANSLATOR_FILTER(TCSETS,		ict_tcsets)
+	IOC_CMD_TRANSLATOR_FILTER(TCSETSW,		ict_tcsets)
+	IOC_CMD_TRANSLATOR_FILTER(TCSETSF,		ict_tcsets)
+	IOC_CMD_TRANSLATOR_FILTER(TCSETA,		ict_tcseta)
+	IOC_CMD_TRANSLATOR_FILTER(TCSETAW,		ict_tcseta)
+	IOC_CMD_TRANSLATOR_FILTER(TCSETAF,		ict_tcseta)
+	IOC_CMD_TRANSLATOR_FILTER(TCGETS,		ict_tcgets)
+	IOC_CMD_TRANSLATOR_FILTER(TCGETA,		ict_tcgeta)
+	IOC_CMD_TRANSLATOR_FILTER(TIOCGWINSZ,		ict_tiocgwinsz)
+	IOC_CMD_TRANSLATOR_CUSTOM(LX_TCSBRKP,		ict_tcsbrkp)
+	IOC_CMD_TRANSLATOR_FILTER(TIOCSPGRP,		ict_tiocspgrp)
+	IOC_CMD_TRANSLATOR_FILTER(TIOCGPGRP,		ict_tiocgpgrp)
+	IOC_CMD_TRANSLATOR_CUSTOM(LX_TIOCSPTLCK,	ict_sptlock)
+	IOC_CMD_TRANSLATOR_CUSTOM(LX_TIOCGPTN,		ict_gptn)
+	IOC_CMD_TRANSLATOR_FILTER(TIOCSCTTY,		ict_tiocsctty)
+
+	/* socket related */
+	IOC_CMD_TRANSLATOR_PASS(SIOCSPGRP)
+	IOC_CMD_TRANSLATOR_PASS(SIOCGPGRP)
+	IOC_CMD_TRANSLATOR_PASS(SIOCGSTAMP)
+	IOC_CMD_TRANSLATOR_FILTER(SIOCATMARK,		ict_siocatmark)
+	IOC_CMD_TRANSLATOR_FILTER(SIOCGIFFLAGS,		ict_siolifreq)
+	IOC_CMD_TRANSLATOR_FILTER(SIOCSIFFLAGS,		ict_siolifreq)
+	IOC_CMD_TRANSLATOR_FILTER(SIOCGIFADDR,		ict_siolifreq)
+	IOC_CMD_TRANSLATOR_FILTER(SIOCSIFADDR,		ict_siolifreq)
+	IOC_CMD_TRANSLATOR_FILTER(SIOCGIFDSTADDR,	ict_siolifreq)
+	IOC_CMD_TRANSLATOR_FILTER(SIOCSIFDSTADDR,	ict_siolifreq)
+	IOC_CMD_TRANSLATOR_FILTER(SIOCGIFBRDADDR,	ict_siolifreq)
+	IOC_CMD_TRANSLATOR_FILTER(SIOCSIFBRDADDR,	ict_siolifreq)
+	IOC_CMD_TRANSLATOR_FILTER(SIOCGIFNETMASK,	ict_siolifreq)
+	IOC_CMD_TRANSLATOR_FILTER(SIOCSIFNETMASK,	ict_siolifreq)
+	IOC_CMD_TRANSLATOR_FILTER(SIOCGIFMETRIC,	ict_siolifreq)
+	IOC_CMD_TRANSLATOR_FILTER(SIOCSIFMETRIC,	ict_siolifreq)
+	IOC_CMD_TRANSLATOR_FILTER(SIOCGIFMTU,		ict_siolifreq)
+	IOC_CMD_TRANSLATOR_FILTER(SIOCSIFMTU,		ict_siolifreq)
+	IOC_CMD_TRANSLATOR_FILTER(SIOCGIFHWADDR,	ict_siolifreq)
+	IOC_CMD_TRANSLATOR_CUSTOM(LX_SIOCSIFHWADDR,	ict_siolifreq)
+	IOC_CMD_TRANSLATOR_FILTER(SIOCGIFINDEX,		ict_siolifreq)
+	IOC_CMD_TRANSLATOR_CUSTOM(LX_SIOCGIFTXQLEN,	ict_siolifreq)
+	IOC_CMD_TRANSLATOR_FILTER(SIOCGIFCONF,		ict_siocgifconf)
+
+	/* dtrace related */
+	IOC_CMD_TRANSLATOR_PTHRU(DTRACEHIOC_ADD)
+	IOC_CMD_TRANSLATOR_PTHRU(DTRACEHIOC_REMOVE)
+	IOC_CMD_TRANSLATOR_PTHRU(DTRACEHIOC_ADDDOF)
+
+	IOC_CMD_TRANSLATOR_END
+};
+
+static void
+lx_ioctl_vsd_free(void *data)
+{
+	kmem_free(data, sizeof (struct lx_cc));
+}
+
+void
+lx_ioctl_init()
+{
+	vsd_create(&lx_ioctl_vsd, lx_ioctl_vsd_free);
+}
+
+void
+lx_ioctl_fini()
+{
+	vsd_destroy(&lx_ioctl_vsd);
+}
+
+long
+lx_ioctl(int fdes, int cmd, intptr_t arg)
+{
+	file_t *fp;
+	int res = 0;
+	ioc_cmd_translator_t *ict;
+
+	if (cmd == LX_FIOCLEX || cmd == LX_FIONCLEX) {
+		res = f_setfd_error(fdes, (cmd == LX_FIOCLEX) ? FD_CLOEXEC : 0);
+		return ((res != 0) ? set_errno(res) : 0);
+	}
+
+	if ((fp = getf(fdes)) == NULL)
+		return (set_errno(EBADF));
+
+	/*
+	 * Today, none of the ioctls supported by the emulation possess
+	 * overlapping cmd values.  Because of that, no type interrogation of
+	 * the fd is done before executing specific ioctl emulation.  It's
+	 * assumed that the vnode-specific logic called by the emulation
+	 * function will reject ioctl commands not supported by the fd.
+	 */
+	ict = ioc_translators;
+	while (ict->ict_func != NULL) {
+		if (ict->ict_lxcmd == cmd)
+			break;
+		ict++;
+	}
+
+	if (ict->ict_func != NULL) {
+		res = ict->ict_func(fp, ict->ict_cmd, arg, ict->ict_lxcmd);
+	} else {
+		res = set_errno(EINVAL);
+	}
+
+	releasef(fdes);
+	return (res);
+}
diff --git a/usr/src/uts/common/brand/lx/syscall/lx_kill.c b/usr/src/uts/common/brand/lx/syscall/lx_kill.c
new file mode 100644
index 0000000000..908d83b6be
--- /dev/null
+++ b/usr/src/uts/common/brand/lx/syscall/lx_kill.c
@@ -0,0 +1,399 @@
+/*
+ * CDDL HEADER START
+ *
+ * The contents of this file are subject to the terms of the
+ * Common Development and Distribution License (the "License").
+ * You may not use this file except in compliance with the License.
+ *
+ * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
+ * or http://www.opensolaris.org/os/licensing.
+ * See the License for the specific language governing permissions
+ * and limitations under the License.
+ *
+ * When distributing Covered Code, include this CDDL HEADER in each
+ * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
+ * If applicable, add the following below this CDDL HEADER, with the
+ * fields enclosed by brackets "[]" replaced with your own identifying
+ * information: Portions Copyright [yyyy] [name of copyright owner]
+ *
+ * CDDL HEADER END
+ */
+/*
+ * Copyright 2006 Sun Microsystems, Inc.  All rights reserved.
+ * Use is subject to license terms.
+ * Copyright 2015 Joyent, Inc.
+ */
+
+
+
+
+#include <sys/types.h>
+#include <sys/systm.h>
+#include <sys/errno.h>
+#include <sys/proc.h>
+#include <sys/zone.h>
+#include <sys/thread.h>
+#include <sys/signal.h>
+#include <sys/brand.h>
+#include <sys/lx_brand.h>
+#include <lx_signum.h>
+#include <sys/contract/process_impl.h>
+
+extern int kill(pid_t, int);
+
+/*
+ * Check if it is legal to send this signal to the init process.  Linux
+ * kill(2) semantics dictate that no _unhandled_ signal may be sent to pid
+ * 1.
+ */
+static int
+init_sig_check(int sig, pid_t pid)
+{
+	proc_t *p;
+	int rv = 0;
+
+	mutex_enter(&pidlock);
+
+	if (((p = prfind(pid)) == NULL) || (p->p_stat == SIDL))
+		rv = ESRCH;
+	else if (sig && (sigismember(&cantmask, sig) ||
+	    (PTOU(p)->u_signal[sig-1] == SIG_DFL) ||
+	    (PTOU(p)->u_signal[sig-1] == SIG_IGN)))
+		rv = EPERM;
+
+	mutex_exit(&pidlock);
+
+	return (rv);
+}
+
+static long
+lx_thrkill(pid_t tgid, pid_t pid, int lx_sig, boolean_t tgkill)
+{
+	kthread_t *t;
+	proc_t *pp, *cp = curproc;
+	pid_t initpid;
+	sigqueue_t *sqp;
+	int tid = 1;	/* default tid */
+	int sig, rv;
+
+	/*
+	 * Unlike kill(2), Linux tkill(2) doesn't allow signals to
+	 * be sent to process IDs <= 0 as it doesn't overlay any special
+	 * semantics on the pid.
+	 */
+	if ((pid <= 0) || ((lx_sig < 0) || (lx_sig > LX_NSIG)) ||
+	    ((sig = ltos_signo[lx_sig]) < 0))
+		return (set_errno(EINVAL));
+
+	/*
+	 * If the Linux pid is 1, translate the pid to the actual init
+	 * pid for the zone.  Note that Linux dictates that no unhandled
+	 * signals may be sent to init, so check for that, too.
+	 *
+	 * Otherwise, extract the tid and real pid from the Linux pid.
+	 */
+	initpid = cp->p_zone->zone_proc_initpid;
+	if (pid == 1)
+		pid = initpid;
+	if ((pid == initpid) && ((rv = init_sig_check(sig, pid)) != 0))
+		return (set_errno(rv));
+	else if (lx_lpid_to_spair(pid, &pid, &tid) < 0)
+		return (set_errno(ESRCH));
+
+	if (tgkill && tgid != pid)
+		return (set_errno(ESRCH));
+
+	sqp = kmem_zalloc(sizeof (sigqueue_t), KM_SLEEP);
+
+	/*
+	 * Find the process for the passed pid...
+	 */
+	mutex_enter(&pidlock);
+	if (((pp = prfind(pid)) == NULL) || (pp->p_stat == SIDL)) {
+		mutex_exit(&pidlock);
+		rv = set_errno(ESRCH);
+		goto free_and_exit;
+	}
+	mutex_enter(&pp->p_lock);
+	mutex_exit(&pidlock);
+
+	/*
+	 * Deny permission to send the signal if either of the following
+	 * is true:
+	 *
+	 *	+ The signal is SIGCONT and the target pid is not in the same
+	 *	  session as the sender
+	 *
+	 *	+ prochasprocperm() shows the user lacks sufficient permission
+	 *	  to send the signal to the target pid
+	 */
+	if (((sig == SIGCONT) && (pp->p_sessp != cp->p_sessp)) ||
+	    (!prochasprocperm(pp, cp, CRED()))) {
+		mutex_exit(&pp->p_lock);
+		rv = set_errno(EPERM);
+		goto free_and_exit;
+	}
+
+	/* check for the tid */
+	if ((t = idtot(pp, tid)) == NULL) {
+		mutex_exit(&pp->p_lock);
+		rv = set_errno(ESRCH);
+		goto free_and_exit;
+	}
+
+	/* a signal of 0 means just check for the existence of the thread */
+	if (lx_sig == 0) {
+		mutex_exit(&pp->p_lock);
+		rv = 0;
+		goto free_and_exit;
+	}
+
+	sqp->sq_info.si_signo = sig;
+	sqp->sq_info.si_code = SI_LWP;
+	sqp->sq_info.si_pid = cp->p_pid;
+	sqp->sq_info.si_zoneid = getzoneid();
+	sqp->sq_info.si_uid = crgetruid(CRED());
+	sigaddqa(pp, t, sqp);
+
+	mutex_exit(&pp->p_lock);
+
+	return (0);
+
+free_and_exit:
+	kmem_free(sqp, sizeof (sigqueue_t));
+	return (rv);
+}
+
+long
+lx_tgkill(pid_t tgid, pid_t pid, int lx_sig)
+{
+	return (lx_thrkill(tgid, pid, lx_sig, B_TRUE));
+}
+
+long
+lx_tkill(pid_t pid, int lx_sig)
+{
+	return (lx_thrkill(0, pid, lx_sig, B_FALSE));
+}
+
+long
+lx_kill(pid_t lx_pid, int lx_sig)
+{
+	pid_t s_pid, initpid;
+	sigsend_t v;
+	zone_t *zone = curproc->p_zone;
+	struct proc *p;
+	int err, sig, nfound;
+
+	if ((lx_sig < 0) || (lx_sig > LX_NSIG) ||
+	    ((sig = ltos_signo[lx_sig]) < 0))
+		return (set_errno(EINVAL));
+
+	/*
+	 * Since some linux apps rely on init(1M) having PID 1, we
+	 * transparently translate 1 to the real init(1M)'s pid.  We then
+	 * check to be sure that it is legal for this process to send this
+	 * signal to init(1M).
+	 */
+	initpid = zone->zone_proc_initpid;
+	if (lx_pid == 1 || lx_pid == -1) {
+		s_pid = initpid;
+	} else if (lx_pid == 0) {
+		s_pid = 0;
+	} else if (lx_pid > 0) {
+		if (lx_lpid_to_spair(lx_pid, &s_pid, NULL) != 0) {
+			/*
+			 * If we didn't find this pid that means it doesn't
+			 * exist in this zone.
+			 */
+			return (set_errno(ESRCH));
+		}
+	} else {
+		ASSERT(lx_pid < 0);
+		if (lx_lpid_to_spair(-lx_pid, &s_pid, NULL) != 0) {
+			/*
+			 * If we didn't find this pid it means that the
+			 * process group leader doesn't exist in this zone.
+			 * In this case assuming that the Linux pid is
+			 * the same as the Solaris pid will get us the
+			 * correct behavior.
+			 */
+			s_pid = -lx_pid;
+		}
+	}
+
+	if ((s_pid == initpid) && ((err = init_sig_check(sig, s_pid)) != 0))
+		return (set_errno(err));
+
+	/*
+	 * For individual processes, kill() semantics are the same between
+	 * Solaris and Linux.
+	 */
+	if (lx_pid >= 0)
+		return (kill(s_pid, sig));
+
+	/*
+	 * In Solaris, sending a signal to -pid means "send a signal to
+	 * everyone in process group pid."  In Linux it means "send a
+	 * signal to everyone in the group other than init."  Sending a
+	 * signal to -1 means "send a signal to every process except init
+	 * and myself."
+	 */
+
+	bzero(&v, sizeof (v));
+	v.sig = sig;
+	v.checkperm = 1;
+	v.sicode = SI_USER;
+	err = 0;
+
+	mutex_enter(&pidlock);
+
+	p = (lx_pid == -1) ? practive : pgfind(s_pid);
+	nfound = 0;
+	while (err == 0 && p != NULL) {
+		if ((p->p_zone == zone) && (p->p_stat != SIDL) &&
+		    (p->p_pid != initpid) && (lx_pid < -1 || p != curproc)) {
+			nfound++;
+			err = sigsendproc(p, &v);
+		}
+
+		p = (lx_pid == -1) ? p->p_next : p->p_pglink;
+	}
+	mutex_exit(&pidlock);
+	if (nfound == 0)
+		err = ESRCH;
+	else if (err == 0 && v.perm == 0)
+		err = EPERM;
+	return (err ? set_errno(err) : 0);
+}
+
+/*
+ * This handles the unusual case where the user sends a non-queueable signal
+ * through rt_sigqueueinfo. Signals sent with codes that indicate they are
+ * queuable are sent through the sigqueue syscall via the user level function
+ * lx_rt_sigqueueinfo().
+ */
+int
+lx_helper_rt_sigqueueinfo(pid_t tgid, int sig, siginfo_t *uinfo)
+{
+	proc_t *target_proc;
+	pid_t s_pid;
+	zone_t *zone = curproc->p_zone;
+	sigsend_t send;
+	int err;
+	siginfo_t kinfo;
+
+	if (copyin(uinfo, &kinfo, sizeof (siginfo_t)) != 0)
+		return (set_errno(EFAULT));
+	/* Unlike in lx_kill, this process id must be exact, no negatives. */
+	if (tgid == 0)
+		return (set_errno(ESRCH));
+	if (tgid < 0)
+		return (set_errno(EINVAL));
+	/*
+	 * Translate init directly, otherwise use the convenient utility
+	 * function to translate. Since we're sending to the whole group, we
+	 * only need the solaris pid, and not the lwp id.
+	 */
+	if (tgid == 1) {
+		s_pid = zone->zone_proc_initpid;
+	} else {
+		if (lx_lpid_to_spair(tgid, &s_pid, NULL) != 0) {
+			/*
+			 * If we didn't find this pid that means it doesn't
+			 * exist in this zone.
+			 */
+			return (set_errno(ESRCH));
+		}
+	}
+	/*
+	 * We shouldn't have queuable signals here, those are sent elsewhere by
+	 * the usermode handler for this emulated call.
+	 */
+	if (!SI_CANQUEUE(kinfo.si_code)) {
+		return (set_errno(EINVAL));
+	}
+	/* Since our signal shouldn't queue, we just call sigsendproc(). */
+	bzero(&send, sizeof (send));
+	send.sig = sig;
+	send.checkperm = 1;
+	send.sicode = kinfo.si_code;
+	send.value = kinfo.si_value;
+
+	mutex_enter(&pidlock);
+	target_proc = prfind(s_pid);
+	err = 0;
+	if (target_proc != NULL) {
+		err = sigsendproc(target_proc, &send);
+		if (err == 0 && send.perm == 0)
+			err = EPERM;
+	} else {
+		err = ESRCH;
+	}
+	mutex_exit(&pidlock);
+
+	return (err ? set_errno(err) : 0);
+}
+
+/*
+ * Unlike the above function, this handles all system calls to rt_tgsigqueue
+ * regardless of si_code.
+ */
+int
+lx_helper_rt_tgsigqueueinfo(pid_t tgid, pid_t tid, int sig, siginfo_t *uinfo)
+{
+	id_t s_tid;
+	pid_t s_pid;
+	proc_t *target_proc;
+	sigqueue_t *sqp;
+	kthread_t *t;
+	siginfo_t kinfo;
+
+	if (copyin(uinfo, &kinfo, sizeof (siginfo_t)) != 0)
+		return (set_errno(EFAULT));
+	if (lx_lpid_to_spair(tid, &s_pid, &s_tid) != 0)
+		return (set_errno(ESRCH));
+	/*
+	 * For group leaders, solaris pid == linux pid, so the solaris leader
+	 * pid should be the same as the tgid but since the tgid comes in via
+	 * the syscall we need to check for an invalid value.
+	 */
+	if (s_pid != tgid)
+		return (set_errno(EINVAL));
+
+	mutex_enter(&pidlock);
+	target_proc = prfind(s_pid);
+	if (target_proc != NULL)
+		mutex_enter(&target_proc->p_lock);
+	mutex_exit(&pidlock);
+
+	if (target_proc == NULL) {
+		return (set_errno(ESRCH));
+	}
+	if (sig < 0 || sig >= NSIG)
+		return (set_errno(EINVAL));
+
+	/*
+	 * Some code adapted from lwp_kill, duplicated here because we do some
+	 * customization to the sq_info field of sqp.
+	 */
+	if ((t = idtot(target_proc, s_tid)) == NULL) {
+		mutex_exit(&target_proc->p_lock);
+		return (set_errno(ESRCH));
+	}
+	/* Just checking for existence of the process, not sending a signal. */
+	if (sig == 0) {
+		mutex_exit(&target_proc->p_lock);
+		return (0);
+	}
+	sqp = kmem_zalloc(sizeof (sigqueue_t), KM_SLEEP);
+	sqp->sq_info.si_signo = sig;
+	sqp->sq_info.si_code = kinfo.si_code;
+	sqp->sq_info.si_pid = target_proc->p_pid;
+	sqp->sq_info.si_ctid = PRCTID(target_proc);
+	sqp->sq_info.si_zoneid = getzoneid();
+	sqp->sq_info.si_uid = crgetruid(CRED());
+	sigaddqa(target_proc, t, sqp);
+	mutex_exit(&target_proc->p_lock);
+	return (0);
+}
diff --git a/usr/src/uts/common/brand/lx/syscall/lx_mkdir.c b/usr/src/uts/common/brand/lx/syscall/lx_mkdir.c
new file mode 100644
index 0000000000..2f29f56d5f
--- /dev/null
+++ b/usr/src/uts/common/brand/lx/syscall/lx_mkdir.c
@@ -0,0 +1,38 @@
+/*
+ * This file and its contents are supplied under the terms of the
+ * Common Development and Distribution License ("CDDL"), version 1.0.
+ * You may only use this file in accordance with the terms of version
+ * 1.0 of the CDDL.
+ *
+ * A full copy of the text of the CDDL should have accompanied this
+ * source.  A copy of the CDDL is also available via the Internet at
+ * http://www.illumos.org/license/CDDL.
+ */
+
+/*
+ * Copyright 2015 Joyent, Inc.
+ */
+
+#include <sys/fcntl.h>
+#include <sys/lx_fcntl.h>
+
+/*
+ * From "uts/common/syscall/mkdir.c":
+ */
+extern int mkdirat(int, char *, int);
+
+long
+lx_mkdirat(int fd, char *dname, int dmode)
+{
+	if (fd == LX_AT_FDCWD) {
+		fd = AT_FDCWD;
+	}
+
+	return (mkdirat(fd, dname, dmode));
+}
+
+long
+lx_mkdir(char *dname, int dmode)
+{
+	return (mkdirat(AT_FDCWD, dname, dmode));
+}
diff --git a/usr/src/uts/common/brand/lx/syscall/lx_modify_ldt.c b/usr/src/uts/common/brand/lx/syscall/lx_modify_ldt.c
new file mode 100644
index 0000000000..aa6e12a7d8
--- /dev/null
+++ b/usr/src/uts/common/brand/lx/syscall/lx_modify_ldt.c
@@ -0,0 +1,121 @@
+/*
+ * CDDL HEADER START
+ *
+ * The contents of this file are subject to the terms of the
+ * Common Development and Distribution License (the "License").
+ * You may not use this file except in compliance with the License.
+ *
+ * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
+ * or http://www.opensolaris.org/os/licensing.
+ * See the License for the specific language governing permissions
+ * and limitations under the License.
+ *
+ * When distributing Covered Code, include this CDDL HEADER in each
+ * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
+ * If applicable, add the following below this CDDL HEADER, with the
+ * fields enclosed by brackets "[]" replaced with your own identifying
+ * information: Portions Copyright [yyyy] [name of copyright owner]
+ *
+ * CDDL HEADER END
+ */
+/*
+ * Copyright 2006 Sun Microsystems, Inc.  All rights reserved.
+ * Use is subject to license terms.
+ */
+
+
+#pragma ident	"%Z%%M%	%I%	%E% SMI"
+
+#include <sys/types.h>
+#include <sys/systm.h>
+#include <sys/errno.h>
+#include <sys/segments.h>
+#include <sys/archsystm.h>
+#include <sys/proc.h>
+#include <sys/sysi86.h>
+#include <sys/cmn_err.h>
+#include <sys/lx_ldt.h>
+
+/*
+ * Read the ldt_info structure in from the Linux app, convert it to an ssd
+ * structure, and then call setdscr() to do all the heavy lifting.
+ */
+static int
+write_ldt(void *data, ulong_t count)
+{
+	user_desc_t usd;
+	struct ssd ssd;
+	struct ldt_info ldt_inf;
+	proc_t *pp = curthread->t_procp;
+	int err;
+
+	if (count != sizeof (ldt_inf))
+		return (set_errno(EINVAL));
+
+	if (copyin(data, &ldt_inf, sizeof (ldt_inf)))
+		return (set_errno(EFAULT));
+
+	if (ldt_inf.entry_number >= MAXNLDT)
+		return (set_errno(EINVAL));
+
+	LDT_INFO_TO_DESC(&ldt_inf, &usd);
+	usd_to_ssd(&usd, &ssd, SEL_LDT(ldt_inf.entry_number));
+
+	/*
+	 * Get everyone into a safe state before changing the LDT.
+	 */
+	if (!holdlwps(SHOLDFORK1))
+		return (set_errno(EINTR));
+
+	err = setdscr(&ssd);
+
+	/*
+	 * Release the hounds!
+	 */
+	mutex_enter(&pp->p_lock);
+	continuelwps(pp);
+	mutex_exit(&pp->p_lock);
+
+	return (err ? set_errno(err) : 0);
+}
+
+static int
+read_ldt(void *uptr, ulong_t count)
+{
+	proc_t *pp = curproc;
+	int bytes;
+
+	if (pp->p_ldt == NULL)
+		return (0);
+
+	bytes = (pp->p_ldtlimit + 1) * sizeof (user_desc_t);
+	if (bytes > count)
+		bytes = count;
+
+	if (copyout(pp->p_ldt, uptr, bytes))
+		return (set_errno(EFAULT));
+
+	return (bytes);
+}
+
+long
+lx_modify_ldt(int op, void *data, ulong_t count)
+{
+	int rval;
+
+	switch (op) {
+	case 0:
+		rval = read_ldt(data, count);
+		break;
+
+	case 1:
+		rval = write_ldt(data, count);
+		break;
+
+	default:
+		rval = set_errno(ENOSYS);
+		break;
+	}
+
+	return (rval);
+}
diff --git a/usr/src/uts/common/brand/lx/syscall/lx_open.c b/usr/src/uts/common/brand/lx/syscall/lx_open.c
new file mode 100644
index 0000000000..3e13cef54f
--- /dev/null
+++ b/usr/src/uts/common/brand/lx/syscall/lx_open.c
@@ -0,0 +1,221 @@
+/*
+ * CDDL HEADER START
+ *
+ * The contents of this file are subject to the terms of the
+ * Common Development and Distribution License (the "License").
+ * You may not use this file except in compliance with the License.
+ *
+ * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
+ * or http://www.opensolaris.org/os/licensing.
+ * See the License for the specific language governing permissions
+ * and limitations under the License.
+ *
+ * When distributing Covered Code, include this CDDL HEADER in each
+ * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
+ * If applicable, add the following below this CDDL HEADER, with the
+ * fields enclosed by brackets "[]" replaced with your own identifying
+ * information: Portions Copyright [yyyy] [name of copyright owner]
+ *
+ * CDDL HEADER END
+ */
+
+/*
+ * Copyright 2009 Sun Microsystems, Inc.  All rights reserved.
+ * Use is subject to license terms.
+ * Copyright 2015 Joyent, Inc.  All rights reserved.
+ */
+
+#include <sys/systm.h>
+#include <sys/fcntl.h>
+#include <sys/filio.h>
+#include <sys/types.h>
+#include <sys/stat.h>
+#include <sys/inttypes.h>
+
+#include <sys/lx_types.h>
+#include <sys/lx_fcntl.h>
+#include <sys/lx_misc.h>
+
+extern int fcntl(int, int, intptr_t);
+extern int openat(int, char *, int, int);
+extern int open(char *, int, int);
+extern int close(int);
+extern int ioctl(int, int, intptr_t);
+extern int lookupnameat(char *, enum uio_seg, int, vnode_t **, vnode_t **,
+    vnode_t *);
+
+static int
+ltos_open_flags(uintptr_t p2)
+{
+	int flags;
+
+	if ((p2 & O_ACCMODE) == LX_O_RDONLY)
+		flags = O_RDONLY;
+	else if ((p2 & O_ACCMODE) == LX_O_WRONLY)
+		flags = O_WRONLY;
+	else
+		flags = O_RDWR;
+
+	if (p2 & LX_O_CREAT) {
+		flags |= O_CREAT;
+	}
+
+	if (p2 & LX_O_EXCL)
+		flags |= O_EXCL;
+	if (p2 & LX_O_NOCTTY)
+		flags |= O_NOCTTY;
+	if (p2 & LX_O_TRUNC)
+		flags |= O_TRUNC;
+	if (p2 & LX_O_APPEND)
+		flags |= O_APPEND;
+	if (p2 & LX_O_NONBLOCK)
+		flags |= O_NONBLOCK;
+	if (p2 & LX_O_SYNC)
+		flags |= O_SYNC;
+	if (p2 & LX_O_LARGEFILE)
+		flags |= O_LARGEFILE;
+	if (p2 & LX_O_NOFOLLOW)
+		flags |= O_NOFOLLOW;
+	if (p2 & LX_O_CLOEXEC)
+		flags |= O_CLOEXEC;
+
+	/*
+	 * Linux uses the LX_O_DIRECT flag to do raw, synchronous I/O to the
+	 * device backing the fd in question.  Illumos doesn't have similar
+	 * functionality, but we can attempt to simulate it using the flags
+	 * (O_RSYNC|O_SYNC) and directio(3C).
+	 *
+	 * The LX_O_DIRECT flag also requires that the transfer size and
+	 * alignment of I/O buffers be a multiple of the logical block size for
+	 * the underlying file system, but frankly there isn't an easy way to
+	 * support that functionality without doing something like adding an
+	 * fcntl(2) flag to denote LX_O_DIRECT mode.
+	 *
+	 * Since LX_O_DIRECT is merely a performance advisory, we'll just
+	 * emulate what we can and trust that the only applications expecting
+	 * an error when performing I/O from a misaligned buffer or when
+	 * passing a transfer size is not a multiple of the underlying file
+	 * system block size will be test suites.
+	 */
+	if (p2 & LX_O_DIRECT)
+		flags |= (O_RSYNC|O_SYNC);
+
+	return (flags);
+}
+
+static int
+lx_open_postprocess(int fd, int fmode)
+{
+	if (fmode & LX_O_DIRECT) {
+		(void) ioctl(fd, _FIODIRECTIO, DIRECTIO_ON);
+	}
+
+	/*
+	 * Set the ASYNC flag if passsed.
+	 */
+	if (fmode & LX_O_ASYNC) {
+		int res;
+
+		if ((res = fcntl(fd, F_SETFL, FASYNC)) != 0) {
+			(void) close(fd);
+			return (res);
+		}
+	}
+
+	return (fd);
+}
+
+long
+lx_openat(int atfd, char *path, int fmode, int cmode)
+{
+	int flags, fd;
+	mode_t mode = 0;
+
+	if (atfd == LX_AT_FDCWD)
+		atfd = AT_FDCWD;
+
+	flags = ltos_open_flags(fmode);
+
+	/*
+	 * We use the FSEARCH flag to make sure this is a directory. We have to
+	 * explicitly add 1 to emulate the FREAD/FWRITE mapping of the OPENMODE
+	 * macro since it won't get set via OPENMODE when FSEARCH is used.
+	 */
+	if (fmode & LX_O_DIRECTORY) {
+		flags |= FSEARCH;
+		flags++;
+	}
+
+	if (flags & O_CREAT)
+		mode = (mode_t)cmode;
+
+	ttolwp(curthread)->lwp_errno = 0;
+	fd = openat(atfd, path, flags, mode);
+	if (ttolwp(curthread)->lwp_errno != 0) {
+		if ((fmode & LX_O_DIRECTORY) &&
+		    ttolwp(curthread)->lwp_errno != ENOTDIR) {
+			/*
+			 * We got an error trying to open a file as a directory.
+			 * We need to determine if we should return the original
+			 * error or ENOTDIR.
+			 */
+			vnode_t *startvp;
+			vnode_t *vp;
+			int oerror, error = 0;
+
+			oerror = ttolwp(curthread)->lwp_errno;
+
+			if (atfd == AT_FDCWD) {
+				/* regular open */
+				startvp = NULL;
+			} else {
+				char startchar;
+
+				if (copyin(path, &startchar, sizeof (char)))
+					return (set_errno(oerror));
+
+				/* if startchar is / then startfd is ignored */
+				if (startchar == '/') {
+					startvp = NULL;
+				} else {
+					file_t *startfp;
+
+					if ((startfp = getf(atfd)) == NULL)
+						return (set_errno(oerror));
+					startvp = startfp->f_vnode;
+					VN_HOLD(startvp);
+					releasef(atfd);
+				}
+			}
+
+			if (lookupnameat(path, UIO_USERSPACE,
+			    (fmode & LX_O_NOFOLLOW) ?  NO_FOLLOW : FOLLOW,
+			    NULLVPP, &vp, startvp) != 0) {
+				if (startvp != NULL)
+					VN_RELE(startvp);
+				return (set_errno(oerror));
+			}
+
+			if (startvp != NULL)
+				VN_RELE(startvp);
+
+			if (vp->v_type != VDIR)
+				error = ENOTDIR;
+
+			VN_RELE(vp);
+			if (error != 0)
+				return (set_errno(ENOTDIR));
+
+			set_errno(oerror);
+		}
+		return (ttolwp(curthread)->lwp_errno);
+	}
+
+	return (lx_open_postprocess(fd, fmode));
+}
+
+long
+lx_open(char *path, int fmode, int cmode)
+{
+	return (lx_openat(LX_AT_FDCWD, path, fmode, cmode));
+}
diff --git a/usr/src/uts/common/brand/lx/syscall/lx_pipe.c b/usr/src/uts/common/brand/lx/syscall/lx_pipe.c
new file mode 100644
index 0000000000..fe354a8d54
--- /dev/null
+++ b/usr/src/uts/common/brand/lx/syscall/lx_pipe.c
@@ -0,0 +1,202 @@
+/*
+ * CDDL HEADER START
+ *
+ * The contents of this file are subject to the terms of the
+ * Common Development and Distribution License (the "License").
+ * You may not use this file except in compliance with the License.
+ *
+ * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
+ * or http://www.opensolaris.org/os/licensing.
+ * See the License for the specific language governing permissions
+ * and limitations under the License.
+ *
+ * When distributing Covered Code, include this CDDL HEADER in each
+ * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
+ * If applicable, add the following below this CDDL HEADER, with the
+ * fields enclosed by brackets "[]" replaced with your own identifying
+ * information: Portions Copyright [yyyy] [name of copyright owner]
+ *
+ * CDDL HEADER END
+ */
+/*
+ * Copyright (c) 1984, 1986, 1987, 1988, 1989 AT&T. All Rights Reserved.
+ *
+ * Copyright 2007 Sun Microsystems, Inc.  All rights reserved.
+ * Use is subject to license terms.
+ * Copyright 2013 OmniTI Computer Consulting, Inc. All rights reserved.
+ * Copyright 2015 Joyent, Inc.
+ */
+
+#include <sys/zone.h>
+#include <sys/types.h>
+#include <sys/systm.h>
+#include <sys/thread.h>
+#include <sys/cpuvar.h>
+#include <sys/cred.h>
+#include <sys/user.h>
+#include <sys/vnode.h>
+#include <sys/file.h>
+#include <sys/stream.h>
+#include <sys/strsubr.h>
+#include <sys/errno.h>
+#include <sys/debug.h>
+#include <sys/fs/fifonode.h>
+#include <sys/fcntl.h>
+#include <sys/brand.h>
+#include <sys/lx_brand.h>
+
+#define	LX_O_NONBLOCK	04000
+#define	LX_O_CLOEXEC	02000000
+
+/*
+ * Based on native pipe(2) system call, except that the pipe is half-duplex.
+ */
+static int
+lx_hd_pipe(intptr_t arg, int flags)
+{
+	vnode_t *vp1, *vp2;
+	struct file *fp1, *fp2;
+	int error = 0;
+	int flag1, flag2, iflags;
+	int fd1, fd2;
+
+	/*
+	 * Validate allowed flags.
+	 */
+	if ((flags & ~(FCLOEXEC|FNONBLOCK)) != 0) {
+		return (set_errno(EINVAL));
+	}
+	/*
+	 * Allocate and initialize two vnodes.
+	 */
+	makepipe(&vp1, &vp2);
+
+	/*
+	 * Allocate and initialize two file table entries and two
+	 * file pointers. The first file pointer is open for read and the
+	 * second is open for write.
+	 */
+	if ((error = falloc(vp1, FREAD, &fp1, &fd1)) != 0) {
+		VN_RELE(vp1);
+		VN_RELE(vp2);
+		return (set_errno(error));
+	}
+
+	if ((error = falloc(vp2, FWRITE, &fp2, &fd2)) != 0)
+		goto out2;
+
+	/*
+	 * Create two stream heads and attach to each vnode.
+	 */
+	if ((error = fifo_stropen(&vp1, FREAD, fp1->f_cred, 0, 0)) != 0)
+		goto out;
+
+	if ((error = fifo_stropen(&vp2, FWRITE, fp2->f_cred, 0, 0)) != 0) {
+		(void) VOP_CLOSE(vp1, FREAD, 1, (offset_t)0,
+		    fp1->f_cred, NULL);
+		goto out;
+	}
+
+	strmate(vp1, vp2);
+
+	VTOF(vp1)->fn_ino = VTOF(vp2)->fn_ino = fifogetid();
+
+	/*
+	 * Set the O_NONBLOCK flag if requested.
+	 */
+	if (flags & FNONBLOCK) {
+		flag1 = fp1->f_flag;
+		flag2 = fp2->f_flag;
+		iflags = flags & FNONBLOCK;
+
+		if ((error = VOP_SETFL(vp1, flag1, iflags, fp1->f_cred,
+		    NULL)) != 0) {
+			goto out_vop_close;
+		}
+		fp1->f_flag |= iflags;
+
+		if ((error = VOP_SETFL(vp2, flag2, iflags, fp2->f_cred,
+		    NULL)) != 0) {
+			goto out_vop_close;
+		}
+		fp2->f_flag |= iflags;
+	}
+
+	/*
+	 * Return the file descriptors to the user. They now
+	 * point to two different vnodes which have different
+	 * stream heads.
+	 */
+	if (copyout(&fd1, &((int *)arg)[0], sizeof (int)) ||
+	    copyout(&fd2, &((int *)arg)[1], sizeof (int))) {
+		error = EFAULT;
+		goto out_vop_close;
+	}
+
+	/*
+	 * Now fill in the entries that falloc reserved
+	 */
+	mutex_exit(&fp1->f_tlock);
+	mutex_exit(&fp2->f_tlock);
+	setf(fd1, fp1);
+	setf(fd2, fp2);
+
+	/*
+	 * Optionally set the FCLOEXEC flag
+	 */
+	if ((flags & FCLOEXEC) != 0) {
+		f_setfd(fd1, FD_CLOEXEC);
+		f_setfd(fd2, FD_CLOEXEC);
+	}
+
+	return (0);
+out_vop_close:
+	(void) VOP_CLOSE(vp1, FREAD, 1, (offset_t)0, fp1->f_cred, NULL);
+	(void) VOP_CLOSE(vp2, FWRITE, 1, (offset_t)0, fp2->f_cred, NULL);
+out:
+	setf(fd2, NULL);
+	unfalloc(fp2);
+out2:
+	setf(fd1, NULL);
+	unfalloc(fp1);
+	VN_RELE(vp1);
+	VN_RELE(vp2);
+	return (set_errno(error));
+}
+
+/*
+ * pipe(2) system call.
+ */
+long
+lx_pipe(intptr_t arg)
+{
+	return (lx_hd_pipe(arg, 0));
+}
+
+/*
+ * pipe2(2) system call.
+ */
+long
+lx_pipe2(intptr_t arg, int lxflags)
+{
+	int flags = 0;
+
+	/*
+	 * Validate allowed flags.
+	 */
+	if ((lxflags & ~(LX_O_NONBLOCK | LX_O_CLOEXEC)) != 0) {
+		return (set_errno(EINVAL));
+	}
+
+	/*
+	 * Convert from Linux flags to illumos flags.
+	 */
+	if (lxflags & LX_O_NONBLOCK) {
+		flags |= FNONBLOCK;
+	}
+	if (lxflags & LX_O_CLOEXEC) {
+		flags |= FCLOEXEC;
+	}
+
+	return (lx_hd_pipe(arg, flags));
+}
diff --git a/usr/src/uts/common/brand/lx/syscall/lx_rlimit.c b/usr/src/uts/common/brand/lx/syscall/lx_rlimit.c
new file mode 100644
index 0000000000..6581ead25b
--- /dev/null
+++ b/usr/src/uts/common/brand/lx/syscall/lx_rlimit.c
@@ -0,0 +1,575 @@
+/*
+ * This file and its contents are supplied under the terms of the
+ * Common Development and Distribution License ("CDDL"), version 1.0.
+ * You may only use this file in accordance with the terms of version
+ * 1.0 of the CDDL.
+ *
+ * A full copy of the text of the CDDL should have accompanied this
+ * source.  A copy of the CDDL is also available via the Internet at
+ * http://www.illumos.org/license/CDDL.
+ */
+
+/*
+ * Copyright 2015 Joyent, Inc.
+ */
+
+#include <sys/systm.h>
+#include <sys/thread.h>
+#include <sys/zone.h>
+#include <sys/cpuvar.h>
+#include <sys/cmn_err.h>
+#include <sys/lx_impl.h>
+#include <sys/lx_brand.h>
+
+#define	LX_RLIMIT_CPU		0
+#define	LX_RLIMIT_FSIZE		1
+#define	LX_RLIMIT_DATA		2
+#define	LX_RLIMIT_STACK		3
+#define	LX_RLIMIT_CORE		4
+#define	LX_RLIMIT_RSS		5
+#define	LX_RLIMIT_NPROC		6
+#define	LX_RLIMIT_NOFILE	7
+#define	LX_RLIMIT_MEMLOCK	8
+#define	LX_RLIMIT_AS		9
+#define	LX_RLIMIT_LOCKS		10	/* NA limit on locks, early 2.4 only */
+#define	LX_RLIMIT_SIGPENDING	11
+#define	LX_RLIMIT_MSGQUEUE	12
+#define	LX_RLIMIT_NICE		13	/* NA ceiling for nice */
+#define	LX_RLIMIT_RTPRIO	14	/* NA ceiling on the RT priority */
+#define	LX_RLIMIT_RTTIME	15	/* NA cpu limit for RT proc. */
+
+#define	LX_RLIMIT_NLIMITS	16
+
+#define	RCTL_INFINITE(x) \
+	((x->rcv_flagaction & RCTL_LOCAL_MAXIMAL) && \
+	(x->rcv_flagaction & RCTL_GLOBAL_INFINITE))
+
+typedef struct {
+	ulong_t	rlim_cur;
+	ulong_t	rlim_max;
+} lx_rlimit_t;
+
+typedef struct {
+	uint32_t	rlim_cur;
+	uint32_t	rlim_max;
+} lx_rlimit32_t;
+
+/*
+ * Linux supports many of the same resources that we do, but on Illumos these
+ * are rctls. Instead of using rlimit, we use rctls for all of the limits.
+ * This table is used to translate Linux rlimit keys into the Illumos legacy
+ * rlimit. We then primarily use the rctl/rlimit compatability code to
+ * manage these.
+ */
+static int l_to_r[LX_RLIMIT_NLIMITS] = {
+	RLIMIT_CPU,		/* 0 CPU */
+	RLIMIT_FSIZE,		/* 1 FSIZE */
+	RLIMIT_DATA,		/* 2 DATA */
+	RLIMIT_STACK,		/* 3 STACK */
+	RLIMIT_CORE,		/* 4 CORE */
+	-1,			/* 5 RSS */
+	-1,			/* 6 NPROC */
+	RLIMIT_NOFILE,		/* 7 NOFILE */
+	-1,			/* 8 MEMLOCK */
+	RLIMIT_AS,		/* 9 AS */
+	-1,			/* 10 LOCKS */
+	-1, 			/* 11 SIGPENDING */
+	-1, 			/* 12 MSGQUEUE */
+	-1,			/* 13 NICE */
+	-1,			/* 14 RTPRIO */
+	-1			/* 15 RTTIME */
+};
+
+/*
+ * Magic value Linux uses to indicate infinity
+ */
+#define	LX_RLIM_INFINITY_N	ULONG_MAX
+
+static void
+lx_get_rctl(char *nm, struct rlimit64 *rlp64)
+{
+	rctl_hndl_t hndl;
+	rctl_val_t *oval, *nval;
+
+	rlp64->rlim_cur = RLIM_INFINITY;
+	rlp64->rlim_max = RLIM_INFINITY;
+
+	nval = kmem_alloc(sizeof (rctl_val_t), KM_SLEEP);
+	mutex_enter(&curproc->p_lock);
+
+	hndl = rctl_hndl_lookup(nm);
+	oval = NULL;
+	while ((hndl != -1) && rctl_local_get(hndl, oval, nval, curproc) == 0) {
+		oval = nval;
+		switch (nval->rcv_privilege) {
+		case RCPRIV_BASIC:
+			if (!RCTL_INFINITE(nval))
+				rlp64->rlim_cur = nval->rcv_value;
+			break;
+		case RCPRIV_PRIVILEGED:
+			if (!RCTL_INFINITE(nval))
+				rlp64->rlim_max = nval->rcv_value;
+			break;
+		}
+	}
+
+	mutex_exit(&curproc->p_lock);
+	kmem_free(nval, sizeof (rctl_val_t));
+
+	if (rlp64->rlim_cur == RLIM_INFINITY &&
+	    rlp64->rlim_max != RLIM_INFINITY)
+		rlp64->rlim_cur = rlp64->rlim_max;
+}
+
+static int
+lx_getrlimit_common(int lx_resource, uint64_t *rlim_curp, uint64_t *rlim_maxp)
+{
+	lx_proc_data_t *pd = ptolxproc(curproc);
+	int resource;
+	int64_t cur = -1;
+	boolean_t cur_inf = B_FALSE;
+	int64_t max = -1;
+	boolean_t max_inf = B_FALSE;
+	struct rlimit64 rlim64;
+
+	if (lx_resource < 0 || lx_resource >= LX_RLIMIT_NLIMITS)
+		return (EINVAL);
+
+	switch (lx_resource) {
+	case LX_RLIMIT_LOCKS:
+		rlim64.rlim_cur = pd->l_fake_limits[LX_RLFAKE_LOCKS].rlim_cur;
+		rlim64.rlim_max = pd->l_fake_limits[LX_RLFAKE_LOCKS].rlim_max;
+		break;
+
+	case LX_RLIMIT_NICE:
+		rlim64.rlim_cur = pd->l_fake_limits[LX_RLFAKE_NICE].rlim_cur;
+		rlim64.rlim_max = pd->l_fake_limits[LX_RLFAKE_NICE].rlim_max;
+		break;
+
+	case LX_RLIMIT_RTPRIO:
+		rlim64.rlim_cur = pd->l_fake_limits[LX_RLFAKE_RTPRIO].rlim_cur;
+		rlim64.rlim_max = pd->l_fake_limits[LX_RLFAKE_RTPRIO].rlim_max;
+		break;
+
+	case LX_RLIMIT_RTTIME:
+		rlim64.rlim_cur = pd->l_fake_limits[LX_RLFAKE_RTTIME].rlim_cur;
+		rlim64.rlim_max = pd->l_fake_limits[LX_RLFAKE_RTTIME].rlim_max;
+		break;
+
+	case LX_RLIMIT_RSS:
+		/* zone.max-physical-memory */
+		rlim64.rlim_cur = rlim64.rlim_max = curzone->zone_phys_mem_ctl;
+		break;
+
+	case LX_RLIMIT_NPROC:
+		/*  zone.max-lwps */
+		rlim64.rlim_cur = rlim64.rlim_max = curzone->zone_nlwps_ctl;
+		break;
+
+	case LX_RLIMIT_MEMLOCK:
+		/* zone.max-locked-memory */
+		rlim64.rlim_cur = rlim64.rlim_max =
+		    curzone->zone_locked_mem_ctl;
+		break;
+
+	case LX_RLIMIT_SIGPENDING:
+		lx_get_rctl("process.max-sigqueue-size", &rlim64);
+		break;
+
+	case LX_RLIMIT_MSGQUEUE:
+		lx_get_rctl("process.max-msg-messages", &rlim64);
+		break;
+
+	default:
+		resource = l_to_r[lx_resource];
+
+		mutex_enter(&curproc->p_lock);
+		(void) rctl_rlimit_get(rctlproc_legacy[resource], curproc,
+		    &rlim64);
+		mutex_exit(&curproc->p_lock);
+		break;
+	}
+
+
+	if (rlim64.rlim_cur == RLIM64_INFINITY) {
+		cur = LX_RLIM_INFINITY_N;
+	} else {
+		cur = rlim64.rlim_cur;
+	}
+	if (rlim64.rlim_max == RLIM64_INFINITY) {
+		max = LX_RLIM_INFINITY_N;
+	} else {
+		max = rlim64.rlim_max;
+	}
+
+	if (lx_resource == LX_RLIMIT_STACK && cur > INT_MAX) {
+		/*
+		 * Stunningly, Linux has somehow managed to confuse the concept
+		 * of a "limit" with that of a "default" -- and the value of
+		 * RLIMIT_STACK is used by NPTL as the _default_ stack size if
+		 * it isn't specified. (!!)  Even for a system that prides
+		 * itself on slapdash castles of junk, this is an amazingly
+		 * willful act of incompetence -- and one that is gleefully
+		 * confessed in the pthread_create() man page: "if the
+		 * RLIMIT_STACK soft resource limit at the time the program
+		 * started has any value other than 'unlimited', then it
+		 * determines the default stack size of new threads."  A
+		 * typical stack limit for us is 32TB; if it needs to be said,
+		 * setting the default stack size to be 32TB doesn't work so
+		 * well!  Of course, glibc dropping a deuce in its pants
+		 * becomes our problem -- so to prevent smelly accidents we
+		 * tell Linux that any stack limit over the old (32-bit) values
+		 * for infinity are just infinitely large.
+		 */
+		cur_inf = B_TRUE;
+		max_inf = B_TRUE;
+	}
+
+	if (cur_inf) {
+		*rlim_curp = LX_RLIM64_INFINITY;
+	} else {
+		*rlim_curp = cur;
+	}
+
+	if (max_inf) {
+		*rlim_maxp = LX_RLIM64_INFINITY;
+	} else {
+		*rlim_maxp = max;
+	}
+
+	return (0);
+}
+
+/*
+ * This is the 'new' getrlimit, variously called getrlimit or ugetrlimit
+ * in Linux headers and code.  The only difference between this and the old
+ * getrlimit (variously called getrlimit or old_getrlimit) is the value of
+ * RLIM_INFINITY, which is smaller for the older version.  Modern code will
+ * use this version by default.
+ */
+long
+lx_getrlimit(int resource, lx_rlimit_t *rlp)
+{
+	int rv;
+	lx_rlimit_t rl;
+	uint64_t rlim_cur, rlim_max;
+
+	rv = lx_getrlimit_common(resource, &rlim_cur, &rlim_max);
+	if (rv != 0)
+		return (set_errno(rv));
+
+	if (get_udatamodel() == DATAMODEL_NATIVE) {
+		if (rlim_cur == LX_RLIM64_INFINITY)
+			rl.rlim_cur = LX_RLIM_INFINITY_N;
+		else if (rlim_cur > LX_RLIM_INFINITY_N)
+			rl.rlim_cur = LX_RLIM_INFINITY_N;
+		else
+			rl.rlim_cur = (ulong_t)rlim_cur;
+
+		if (rlim_max == LX_RLIM64_INFINITY)
+			rl.rlim_max = LX_RLIM_INFINITY_N;
+		else if (rlim_max > LX_RLIM_INFINITY_N)
+			rl.rlim_max = LX_RLIM_INFINITY_N;
+		else
+			rl.rlim_max = (ulong_t)rlim_max;
+
+		if (copyout(&rl, rlp, sizeof (rl)) != 0)
+			return (set_errno(EFAULT));
+	}
+#ifdef _SYSCALL32_IMPL
+	else {
+		lx_rlimit32_t rl32;
+
+		if (rlim_cur > UINT_MAX)
+			rl.rlim_cur = UINT_MAX;
+		else
+			rl.rlim_cur = (ulong_t)rlim_cur;
+
+		if (rlim_max > UINT_MAX)
+			rl.rlim_max = UINT_MAX;
+		else
+			rl.rlim_max = (ulong_t)rlim_max;
+
+		rl32.rlim_cur = rl.rlim_cur;
+		rl32.rlim_max = rl.rlim_max;
+
+		if (copyout(&rl32, rlp, sizeof (rl32)) != 0)
+			return (set_errno(EFAULT));
+	}
+#endif
+
+	return (0);
+}
+
+/*
+ * This is the 'old' getrlimit, variously called getrlimit or old_getrlimit
+ * in Linux headers and code.  The only difference between this and the new
+ * getrlimit (variously called getrlimit or ugetrlimit) is the value of
+ * RLIM_INFINITY, which is smaller for the older version.
+ *
+ * This is only used for 32-bit code.
+ */
+long
+lx_oldgetrlimit(int resource, lx_rlimit_t *rlp)
+{
+	int rv;
+	lx_rlimit32_t rl32;
+	uint64_t rlim_cur, rlim_max;
+
+	rv = lx_getrlimit_common(resource, &rlim_cur, &rlim_max);
+	if (rv != 0)
+		return (set_errno(rv));
+
+	if (rlim_cur > INT_MAX)
+		rl32.rlim_cur = INT_MAX;
+	else
+		rl32.rlim_cur = (ulong_t)rlim_cur;
+
+	if (rlim_max > INT_MAX)
+		rl32.rlim_max = INT_MAX;
+	else
+		rl32.rlim_max = (ulong_t)rlim_cur;
+
+	if (copyout(&rl32, rlp, sizeof (rl32)) != 0)
+		return (set_errno(EFAULT));
+
+	return (0);
+}
+
+static int
+lx_set_rctl(char *nm, struct rlimit64 *rlp64)
+{
+	int err;
+	rctl_hndl_t hndl;
+	rctl_alloc_gp_t *gp;
+
+	gp = rctl_rlimit_set_prealloc(1);
+
+	mutex_enter(&curproc->p_lock);
+
+	hndl = rctl_hndl_lookup(nm);
+
+	/*
+	 * We're not supposed to do this but since we want all our rctls to
+	 * behave like rlimits, we take advantage of this function to set up
+	 * this way.
+	 */
+	err = rctl_rlimit_set(hndl, curproc, rlp64, gp, RCTL_LOCAL_DENY, 0,
+	    CRED());
+
+	mutex_exit(&curproc->p_lock);
+
+	rctl_prealloc_destroy(gp);
+
+	return (err);
+}
+
+static int
+lx_setrlimit_common(int lx_resource, uint64_t rlim_cur, uint64_t rlim_max)
+{
+	lx_proc_data_t *pd = ptolxproc(curproc);
+	int err;
+	int resource;
+	rctl_alloc_gp_t *gp;
+	struct rlimit64 rl64;
+
+	if (lx_resource < 0 || lx_resource >= LX_RLIMIT_NLIMITS)
+		return (EINVAL);
+
+	switch (lx_resource) {
+	case LX_RLIMIT_LOCKS:
+		pd->l_fake_limits[LX_RLFAKE_LOCKS].rlim_cur = rlim_cur;
+		pd->l_fake_limits[LX_RLFAKE_LOCKS].rlim_max = rlim_max;
+		break;
+
+	case LX_RLIMIT_NICE:
+		pd->l_fake_limits[LX_RLFAKE_NICE].rlim_cur = rlim_cur;
+		pd->l_fake_limits[LX_RLFAKE_NICE].rlim_max = rlim_max;
+		break;
+
+	case LX_RLIMIT_RTPRIO:
+		pd->l_fake_limits[LX_RLFAKE_RTPRIO].rlim_cur = rlim_cur;
+		pd->l_fake_limits[LX_RLFAKE_RTPRIO].rlim_max = rlim_max;
+		break;
+
+	case LX_RLIMIT_RTTIME:
+		pd->l_fake_limits[LX_RLFAKE_RTTIME].rlim_cur = rlim_cur;
+		pd->l_fake_limits[LX_RLFAKE_RTTIME].rlim_max = rlim_max;
+		break;
+
+	case LX_RLIMIT_RSS:
+		/*
+		 * zone.max-physical-memory
+		 * Since we're emulating the value via a zone rctl, we can't
+		 * set that from within the zone. Lie and say we set the value.
+		 */
+		break;
+
+	case LX_RLIMIT_NPROC:
+		/*
+		 * zone.max-lwps
+		 * Since we're emulating the value via a zone rctl, we can't
+		 * set that from within the zone. Lie and say we set the value.
+		 */
+		break;
+
+	case LX_RLIMIT_MEMLOCK:
+		/*
+		 * zone.max-locked-memory
+		 * Since we're emulating the value via a zone rctl, we can't
+		 * set that from within the zone. Lie and say we set the value.
+		 */
+		break;
+
+	case LX_RLIMIT_SIGPENDING:
+		/*
+		 * On Ubuntu at least, the login and sshd processes expect to
+		 * set this limit to 16k and login will fail if this fails. On
+		 * Illumos we have a system limit of 8k and normally the
+		 * privileged limit is 512. We simply pretend this works to
+		 * allow login to work.
+		 */
+		if (rlim_max > 8192)
+			return (0);
+
+		rl64.rlim_cur = rlim_cur;
+		rl64.rlim_max = rlim_max;
+		if ((err = lx_set_rctl("process.max-sigqueue-size", &rl64))
+		    != 0)
+			return (set_errno(err));
+		break;
+
+	case LX_RLIMIT_MSGQUEUE:
+		rl64.rlim_cur = rlim_cur;
+		rl64.rlim_max = rlim_max;
+		if ((err = lx_set_rctl("process.max-msg-messages", &rl64)) != 0)
+			return (set_errno(err));
+		break;
+
+	default:
+		resource = l_to_r[lx_resource];
+
+		/*
+		 * Linux limits the max number of open files to 1m and there is
+		 * a test for this.
+		 */
+		if (lx_resource == LX_RLIMIT_NOFILE && rlim_max > (1024 * 1024))
+			return (EPERM);
+
+		rl64.rlim_cur = rlim_cur;
+		rl64.rlim_max = rlim_max;
+		gp = rctl_rlimit_set_prealloc(1);
+
+		mutex_enter(&curproc->p_lock);
+		err = rctl_rlimit_set(rctlproc_legacy[resource], curproc,
+		    &rl64, gp, rctlproc_flags[resource],
+		    rctlproc_signals[resource], CRED());
+		mutex_exit(&curproc->p_lock);
+
+		rctl_prealloc_destroy(gp);
+		if (err != 0)
+			return (set_errno(err));
+		break;
+	}
+
+	return (0);
+}
+
+long
+lx_setrlimit(int resource, lx_rlimit_t *rlp)
+{
+	int rv;
+	lx_rlimit_t rl;
+	uint64_t rlim_cur, rlim_max;
+
+	if (get_udatamodel() == DATAMODEL_NATIVE) {
+		if (copyin(rlp, &rl, sizeof (rl)) != 0)
+			return (set_errno(EFAULT));
+	}
+#ifdef _SYSCALL32_IMPL
+	else {
+		lx_rlimit32_t rl32;
+
+		if (copyin(rlp, &rl32, sizeof (rl32)) != 0)
+			return (set_errno(EFAULT));
+
+		rl.rlim_cur = rl32.rlim_cur;
+		rl.rlim_max = rl32.rlim_max;
+	}
+#endif
+
+	if ((rl.rlim_max != LX_RLIM_INFINITY_N &&
+	    rl.rlim_cur == LX_RLIM_INFINITY_N) ||
+	    rl.rlim_cur > rl.rlim_max)
+		return (set_errno(EINVAL));
+
+	if (rl.rlim_cur == LX_RLIM_INFINITY_N)
+		rlim_cur = LX_RLIM64_INFINITY;
+	else
+		rlim_cur = rl.rlim_cur;
+
+	if (rl.rlim_max == LX_RLIM_INFINITY_N)
+		rlim_max = LX_RLIM64_INFINITY;
+	else
+		rlim_max = rl.rlim_max;
+
+	rv = lx_setrlimit_common(resource, rlim_cur, rlim_max);
+	if (rv != 0)
+		return (set_errno(rv));
+	return (0);
+}
+
+/*
+ * From the man page:
+ * The Linux-specific prlimit() system call combines and extends the
+ * functionality of setrlimit() and getrlimit(). It can be used to both set
+ * and get the resource limits of an arbitrary process.
+ *
+ * If pid is 0, then the call applies to the calling process.
+ */
+long
+lx_prlimit64(pid_t pid, int resource, lx_rlimit64_t *nrlp, lx_rlimit64_t *orlp)
+{
+	int rv;
+	lx_rlimit64_t nrl, orl;
+
+	if (pid != 0) {
+		/* XXX TBD if needed */
+		char buf[80];
+
+		(void) snprintf(buf, sizeof (buf),
+		    "setting prlimit %d for another process\n", resource);
+		lx_unsupported(buf);
+		return (ENOTSUP);
+	}
+
+	if (orlp != NULL) {
+		/* we first get the current limits */
+		rv = lx_getrlimit_common(resource, &orl.rlim_cur,
+		    &orl.rlim_max);
+		if (rv != 0)
+			return (set_errno(rv));
+	}
+
+	if (nrlp != NULL) {
+		if (copyin(nrlp, &nrl, sizeof (nrl)) != 0)
+			return (set_errno(EFAULT));
+
+		if ((nrl.rlim_max != LX_RLIM64_INFINITY &&
+		    nrl.rlim_cur == LX_RLIM64_INFINITY) ||
+		    nrl.rlim_cur > nrl.rlim_max)
+			return (set_errno(EINVAL));
+
+		rv = lx_setrlimit_common(resource, nrl.rlim_cur, nrl.rlim_max);
+		if (rv != 0)
+			return (set_errno(rv));
+	}
+
+	if (orlp != NULL) {
+		/* now return the original limits, if necessary */
+		if (copyout(&orl, orlp, sizeof (orl)) != 0)
+			return (set_errno(EFAULT));
+	}
+
+	return (0);
+}
diff --git a/usr/src/uts/common/brand/lx/syscall/lx_rw.c b/usr/src/uts/common/brand/lx/syscall/lx_rw.c
new file mode 100644
index 0000000000..b21a81da48
--- /dev/null
+++ b/usr/src/uts/common/brand/lx/syscall/lx_rw.c
@@ -0,0 +1,60 @@
+/*
+ * This file and its contents are supplied under the terms of the
+ * Common Development and Distribution License ("CDDL"), version 1.0.
+ * You may only use this file in accordance with the terms of version
+ * 1.0 of the CDDL.
+ *
+ * A full copy of the text of the CDDL should have accompanied this
+ * source.  A copy of the CDDL is also available via the Internet at
+ * http://www.illumos.org/license/CDDL.
+ */
+
+/*
+ * Copyright 2015 Joyent, Inc.
+ */
+
+#include <sys/errno.h>
+#include <sys/systm.h>
+#include <sys/file.h>
+#include <sys/vnode.h>
+#include <sys/brand.h>
+#include <sys/lx_brand.h>
+
+/* uts/common/syscall/rw.c */
+extern ssize_t read(int fdes, void *cbuf, size_t count);
+extern ssize_t write(int fdes, void *cbuf, size_t count);
+
+long
+lx_read(int fd, void *buf, size_t nbyte)
+{
+	file_t *fp;
+	vtype_t t;
+
+	if ((fp = getf(fd)) == NULL)
+		return (set_errno(EBADF));
+	t = fp->f_vnode->v_type;
+	releasef(fd);
+
+	if (t == VDIR)
+		return (set_errno(EISDIR));
+
+	/*
+	 * If read(2) returns EINTR, we want to signal that restarting the
+	 * system call is acceptable:
+	 */
+	ttolxlwp(curthread)->br_syscall_restart = B_TRUE;
+
+	return (read(fd, buf, nbyte));
+}
+
+long
+lx_write(int fd, void *buf, size_t nbyte)
+{
+	/*
+	 * If write(2) returns EINTR, we want to signal that restarting the
+	 * system call is acceptable:
+	 */
+	ttolxlwp(curthread)->br_syscall_restart = B_TRUE;
+
+	return (write(fd, buf, nbyte));
+}
diff --git a/usr/src/uts/common/brand/lx/syscall/lx_sched.c b/usr/src/uts/common/brand/lx/syscall/lx_sched.c
new file mode 100644
index 0000000000..0def559e29
--- /dev/null
+++ b/usr/src/uts/common/brand/lx/syscall/lx_sched.c
@@ -0,0 +1,524 @@
+/*
+ * CDDL HEADER START
+ *
+ * The contents of this file are subject to the terms of the
+ * Common Development and Distribution License (the "License").
+ * You may not use this file except in compliance with the License.
+ *
+ * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
+ * or http://www.opensolaris.org/os/licensing.
+ * See the License for the specific language governing permissions
+ * and limitations under the License.
+ *
+ * When distributing Covered Code, include this CDDL HEADER in each
+ * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
+ * If applicable, add the following below this CDDL HEADER, with the
+ * fields enclosed by brackets "[]" replaced with your own identifying
+ * information: Portions Copyright [yyyy] [name of copyright owner]
+ *
+ * CDDL HEADER END
+ */
+/*
+ * Copyright 2006 Sun Microsystems, Inc.  All rights reserved.
+ * Use is subject to license terms.
+ */
+/*
+ * Copyright 2015 Joyent, Inc.
+ */
+
+#include <sys/types.h>
+#include <sys/systm.h>
+#include <sys/errno.h>
+#include <sys/proc.h>
+#include <sys/cpu.h>
+#include <sys/rtpriocntl.h>
+#include <sys/tspriocntl.h>
+#include <sys/processor.h>
+#include <sys/brand.h>
+#include <sys/lx_sched.h>
+#include <sys/lx_brand.h>
+
+extern int yield();
+extern long priocntl_common(int, procset_t *, int, caddr_t, caddr_t, uio_seg_t);
+
+long
+lx_sched_yield(void)
+{
+	yield();
+
+	return (0);
+}
+
+int
+lx_sched_affinity(int cmd, uintptr_t pid, int len, uintptr_t maskp,
+    int64_t *rval)
+{
+	pid_t		s_pid;
+	id_t		s_tid;
+	kthread_t	*t = curthread;
+	lx_lwp_data_t	*lx_lwp;
+
+	if (cmd != B_GET_AFFINITY_MASK && cmd != B_SET_AFFINITY_MASK)
+		return (set_errno(EINVAL));
+
+	/*
+	 * The caller wants to know how large the mask should be.
+	 */
+	if (cmd == B_GET_AFFINITY_MASK && len == 0) {
+		*rval = sizeof (lx_affmask_t);
+		return (0);
+	}
+
+	/*
+	 * Otherwise, ensure they have a large enough mask.
+	 */
+	if (cmd == B_GET_AFFINITY_MASK && len < sizeof (lx_affmask_t)) {
+		*rval = -1;
+		return (set_errno(EINVAL));
+	}
+
+	if (pid == 0) {
+		s_pid = curproc->p_pid;
+		s_tid = curthread->t_tid;
+	} else if (lx_lpid_to_spair((pid_t)pid, &s_pid, &s_tid) == -1) {
+		return (set_errno(ESRCH));
+	}
+
+	/*
+	 * For now, we only support manipulating threads in the
+	 * same process.
+	 */
+	if (curproc->p_pid != s_pid)
+		return (set_errno(EPERM));
+
+	/*
+	 * We must hold the process lock so that the thread list
+	 * doesn't change while we're looking at it. We'll hold
+	 * the lock until we no longer reference the
+	 * corresponding lwp.
+	 */
+
+	mutex_enter(&curproc->p_lock);
+
+	do {
+		if (t->t_tid == s_tid)
+			break;
+		t = t->t_forw;
+	} while (t != curthread);
+
+	/*
+	 * If the given PID is in the current thread's process,
+	 * then we _must_ find it in the process's thread list.
+	 */
+	ASSERT(t->t_tid == s_tid);
+
+	lx_lwp = t->t_lwp->lwp_brand;
+
+	if (cmd == B_SET_AFFINITY_MASK) {
+		if (copyin_nowatch((void *)maskp, &lx_lwp->br_affinitymask,
+		    sizeof (lx_affmask_t)) != 0) {
+			mutex_exit(&curproc->p_lock);
+			return (set_errno(EFAULT));
+		}
+
+		*rval = 0;
+	} else {
+		if (copyout_nowatch(&lx_lwp->br_affinitymask, (void *)maskp,
+		    sizeof (lx_affmask_t)) != 0) {
+			mutex_exit(&curproc->p_lock);
+			return (set_errno(EFAULT));
+		}
+
+		*rval = sizeof (lx_affmask_t);
+	}
+
+	mutex_exit(&curproc->p_lock);
+	return (0);
+}
+
+long
+lx_sched_setscheduler(l_pid_t pid, int policy, struct lx_sched_param *param)
+{
+	klwp_t *lwp = ttolwp(curthread);
+	procset_t procset;
+	procset_t procset_cid;
+	pcparms_t pcparm;
+	pcinfo_t pcinfo;
+	struct lx_sched_param sched_param;
+	tsparms_t *tsp;
+	int prio, maxupri;
+	int rv;
+
+	if (pid < 0)
+		return (set_errno(ESRCH));
+
+	if ((rv = sched_setprocset(&procset, pid)))
+		return (rv);
+
+	if (copyin(param, &sched_param, sizeof (sched_param)))
+		return (set_errno(EFAULT));
+
+	prio = sched_param.lx_sched_prio;
+
+	if (policy < 0) {
+		/*
+		 * get the class id
+		 */
+		pcparm.pc_cid = PC_CLNULL;
+		(void) do_priocntlsys(PC_GETPARMS, &procset, &pcparm);
+		if (lwp->lwp_errno)
+			return (lwp->lwp_errno);
+
+		/*
+		 * get the current policy
+		 */
+		bzero(&pcinfo, sizeof (pcinfo));
+		pcinfo.pc_cid = pcparm.pc_cid;
+		(void) do_priocntlsys(PC_GETCLINFO, &procset, &pcinfo);
+		if (lwp->lwp_errno)
+			return (lwp->lwp_errno);
+
+		if (strcmp(pcinfo.pc_clname, "TS") == 0) {
+			policy = LX_SCHED_OTHER;
+		} else if (strcmp(pcinfo.pc_clname, "RT") == 0) {
+			policy = ((rtparms_t *)pcparm.pc_clparms)->rt_tqnsecs ==
+			    RT_TQINF ? LX_SCHED_FIFO : LX_SCHED_RR;
+		} else {
+			return (set_errno(EINVAL));
+		}
+	}
+
+	bzero(&pcinfo, sizeof (pcinfo));
+	bzero(&pcparm, sizeof (pcparm));
+	setprocset(&procset_cid, POP_AND, P_PID, 0, P_ALL, 0);
+	switch (policy) {
+	case LX_SCHED_FIFO:
+	case LX_SCHED_RR:
+		(void) strcpy(pcinfo.pc_clname, "RT");
+		(void) do_priocntlsys(PC_GETCID, &procset_cid, &pcinfo);
+		if (lwp->lwp_errno)
+			return (lwp->lwp_errno);
+
+		if (prio < 0 ||
+		    prio > ((rtinfo_t *)pcinfo.pc_clinfo)->rt_maxpri)
+			return (set_errno(EINVAL));
+		pcparm.pc_cid = pcinfo.pc_cid;
+		((rtparms_t *)pcparm.pc_clparms)->rt_pri = prio;
+		((rtparms_t *)pcparm.pc_clparms)->rt_tqnsecs =
+		    policy == LX_SCHED_RR ? RT_TQDEF : RT_TQINF;
+		break;
+
+	case LX_SCHED_OTHER:
+		(void) strcpy(pcinfo.pc_clname, "TS");
+		(void) do_priocntlsys(PC_GETCID, &procset_cid, &pcinfo);
+		if (lwp->lwp_errno)
+			return (lwp->lwp_errno);
+
+		maxupri = ((tsinfo_t *)pcinfo.pc_clinfo)->ts_maxupri;
+		if (prio > maxupri || prio < -maxupri)
+			return (set_errno(EINVAL));
+
+		pcparm.pc_cid = pcinfo.pc_cid;
+		tsp = (tsparms_t *)pcparm.pc_clparms;
+		tsp->ts_upri = prio;
+		tsp->ts_uprilim = TS_NOCHANGE;
+		break;
+
+	default:
+		return (set_errno(EINVAL));
+	}
+
+	/*
+	 * finally set scheduling policy and parameters
+	 */
+	(void) do_priocntlsys(PC_SETPARMS, &procset, &pcparm);
+
+	return (0);
+}
+
+long
+lx_sched_getscheduler(l_pid_t pid)
+{
+	klwp_t *lwp = ttolwp(curthread);
+	procset_t procset;
+	pcparms_t pcparm;
+	pcinfo_t pcinfo;
+	int policy;
+	int rv;
+
+	if (pid < 0)
+		return (set_errno(ESRCH));
+
+	if ((rv = sched_setprocset(&procset, pid)))
+		return (rv);
+
+	/*
+	 * get the class id
+	 */
+	pcparm.pc_cid = PC_CLNULL;
+	(void) do_priocntlsys(PC_GETPARMS, &procset, &pcparm);
+	if (lwp->lwp_errno)
+		return (lwp->lwp_errno);
+
+	/*
+	 * get the class info and identify the equivalent linux policy
+	 */
+	bzero(&pcinfo, sizeof (pcinfo));
+	pcinfo.pc_cid = pcparm.pc_cid;
+	(void) do_priocntlsys(PC_GETCLINFO, &procset, &pcinfo);
+	if (lwp->lwp_errno)
+		return (lwp->lwp_errno);
+
+	if (strcmp(pcinfo.pc_clname, "TS") == 0)
+		policy = LX_SCHED_OTHER;
+	else if (strcmp(pcinfo.pc_clname, "RT") == 0)
+		policy = ((rtparms_t *)pcparm.pc_clparms)->rt_tqnsecs ==
+		    RT_TQINF ? LX_SCHED_FIFO : LX_SCHED_RR;
+	else
+		policy = set_errno(EINVAL);
+
+	return (policy);
+}
+
+long
+lx_sched_setparam(l_pid_t pid, struct lx_sched_param *param)
+{
+	klwp_t *lwp = ttolwp(curthread);
+	procset_t procset;
+	procset_t procset_cid;
+	pcparms_t pcparm;
+	pcinfo_t pcinfo;
+	struct lx_sched_param sched_param;
+	tsparms_t *tsp;
+	int policy;
+	int prio, maxupri;
+	int rv;
+
+	if (pid < 0)
+		return (set_errno(ESRCH));
+
+	if ((rv = sched_setprocset(&procset, pid)))
+		return (rv);
+
+	if (copyin(param, &sched_param, sizeof (sched_param)))
+		return (set_errno(EFAULT));
+
+	prio = sched_param.lx_sched_prio;
+
+	/*
+	 * get the class id
+	 */
+	pcparm.pc_cid = PC_CLNULL;
+	(void) do_priocntlsys(PC_GETPARMS, &procset, &pcparm);
+	if (lwp->lwp_errno)
+		return (lwp->lwp_errno);
+
+	/*
+	 * get the current policy
+	 */
+	bzero(&pcinfo, sizeof (pcinfo));
+	pcinfo.pc_cid = pcparm.pc_cid;
+	(void) do_priocntlsys(PC_GETCLINFO, &procset, &pcinfo);
+	if (lwp->lwp_errno)
+		return (lwp->lwp_errno);
+
+	if (strcmp(pcinfo.pc_clname, "TS") == 0)
+		policy = LX_SCHED_OTHER;
+	else if (strcmp(pcinfo.pc_clname, "RT") == 0)
+		policy = ((rtparms_t *)pcparm.pc_clparms)->rt_tqnsecs ==
+		    RT_TQINF ? LX_SCHED_FIFO : LX_SCHED_RR;
+	else
+		return (set_errno(EINVAL));
+
+	bzero(&pcinfo, sizeof (pcinfo));
+	bzero(&pcparm, sizeof (pcparm));
+	setprocset(&procset_cid, POP_AND, P_PID, 0, P_ALL, 0);
+	switch (policy) {
+	case LX_SCHED_FIFO:
+	case LX_SCHED_RR:
+		(void) strcpy(pcinfo.pc_clname, "RT");
+		(void) do_priocntlsys(PC_GETCID, &procset_cid, &pcinfo);
+		if (lwp->lwp_errno)
+			return (lwp->lwp_errno);
+
+		if (prio < 0 ||
+		    prio > ((rtinfo_t *)pcinfo.pc_clinfo)->rt_maxpri)
+			return (set_errno(EINVAL));
+		pcparm.pc_cid = pcinfo.pc_cid;
+		((rtparms_t *)pcparm.pc_clparms)->rt_pri = prio;
+		((rtparms_t *)pcparm.pc_clparms)->rt_tqnsecs =
+		    policy == LX_SCHED_RR ? RT_TQDEF : RT_TQINF;
+		break;
+
+	case LX_SCHED_OTHER:
+		(void) strcpy(pcinfo.pc_clname, "TS");
+		(void) do_priocntlsys(PC_GETCID, &procset_cid, &pcinfo);
+		if (lwp->lwp_errno)
+			return (lwp->lwp_errno);
+
+		maxupri = ((tsinfo_t *)pcinfo.pc_clinfo)->ts_maxupri;
+		if (prio > maxupri || prio < -maxupri)
+			return (set_errno(EINVAL));
+
+		pcparm.pc_cid = pcinfo.pc_cid;
+		tsp = (tsparms_t *)pcparm.pc_clparms;
+		tsp->ts_upri = prio;
+		tsp->ts_uprilim = TS_NOCHANGE;
+		break;
+
+	default:
+		return (set_errno(EINVAL));
+	}
+
+	/*
+	 * finally set scheduling policy and parameters
+	 */
+	(void) do_priocntlsys(PC_SETPARMS, &procset, &pcparm);
+
+	return (0);
+}
+
+long
+lx_sched_getparam(l_pid_t pid, struct lx_sched_param *param)
+{
+	klwp_t *lwp = ttolwp(curthread);
+	struct lx_sched_param local_param;
+	procset_t procset;
+	pcparms_t pcparm;
+	pcinfo_t pcinfo;
+	tsinfo_t *tsi;
+	int prio, scale;
+	int rv;
+
+	if (pid < 0)
+		return (set_errno(ESRCH));
+
+	if ((rv = sched_setprocset(&procset, pid)))
+		return (rv);
+
+	/*
+	 * get the class id
+	 */
+	pcparm.pc_cid = PC_CLNULL;
+	(void) do_priocntlsys(PC_GETPARMS, &procset, &pcparm);
+	if (lwp->lwp_errno)
+		return (lwp->lwp_errno);
+
+	/*
+	 * get the class info and identify the equivalent linux policy
+	 */
+	bzero(&pcinfo, sizeof (pcinfo));
+	pcinfo.pc_cid = pcparm.pc_cid;
+	(void) do_priocntlsys(PC_GETCLINFO, &procset, &pcinfo);
+	if (lwp->lwp_errno)
+		return (lwp->lwp_errno);
+
+	bzero(&local_param, sizeof (local_param));
+	if (strcmp(pcinfo.pc_clname, "TS") == 0) {
+		/*
+		 * I don't know if we need to do this, coz it can't be
+		 * changed from zero anyway.....
+		 */
+		tsi = (tsinfo_t *)pcinfo.pc_clinfo;
+		prio = ((tsparms_t *)pcparm.pc_clparms)->ts_upri;
+		scale = tsi->ts_maxupri;
+		if (scale == 0)
+			local_param.lx_sched_prio = 0;
+		else
+			local_param.lx_sched_prio = -(prio * 20) / scale;
+	} else if (strcmp(pcinfo.pc_clname, "RT") == 0) {
+		local_param.lx_sched_prio =
+		    ((rtparms_t *)pcparm.pc_clparms)->rt_pri;
+	} else {
+		rv = set_errno(EINVAL);
+	}
+
+	if (rv == 0)
+		if (copyout(&local_param, param, sizeof (local_param)))
+			return (set_errno(EFAULT));
+
+	return (rv);
+}
+
+long
+lx_sched_rr_get_interval(l_pid_t pid, struct timespec *ival)
+{
+	klwp_t *lwp = ttolwp(curthread);
+	struct timespec interval;
+	procset_t procset;
+	pcparms_t pcparm;
+	pcinfo_t pcinfo;
+	int rv;
+
+	if (pid < 0)
+		return (set_errno(ESRCH));
+
+	if ((rv = sched_setprocset(&procset, pid)))
+		return (rv);
+
+	/*
+	 * get the class id
+	 */
+	pcparm.pc_cid = PC_CLNULL;
+	(void) do_priocntlsys(PC_GETPARMS, &procset, &pcparm);
+	if (lwp->lwp_errno)
+		return (lwp->lwp_errno);
+
+	/*
+	 * get the class info and identify the equivalent linux policy
+	 */
+	setprocset(&procset, POP_AND, P_PID, 0, P_ALL, 0);
+	bzero(&pcinfo, sizeof (pcinfo));
+	(void) strcpy(pcinfo.pc_clname, "RT");
+	(void) do_priocntlsys(PC_GETCID, &procset, &pcinfo);
+	if (lwp->lwp_errno)
+		return (lwp->lwp_errno);
+
+	if (pcparm.pc_cid == pcinfo.pc_cid &&
+	    ((rtparms_t *)pcparm.pc_clparms)->rt_tqnsecs != RT_TQINF) {
+		interval.tv_sec = ((rtparms_t *)pcparm.pc_clparms)->rt_tqsecs;
+		interval.tv_nsec = ((rtparms_t *)pcparm.pc_clparms)->rt_tqnsecs;
+
+		if (copyout(&interval, ival, sizeof (interval)))
+			return (set_errno(EFAULT));
+
+		return (0);
+	}
+
+	return (set_errno(EINVAL));
+}
+
+int
+sched_setprocset(procset_t *procset, l_pid_t pid)
+{
+	id_t lid, rid;
+	idtype_t lidtype, ridtype;
+
+	/*
+	 * define the target lwp
+	 */
+	if (pid == 0) {
+		ridtype = P_ALL;
+		lidtype = P_PID;
+		rid = 0;
+		lid = P_MYID;
+	} else {
+		if (lx_lpid_to_spair(pid, &pid, &lid) < 0)
+			return (set_errno(ESRCH));
+		if (pid != curproc->p_pid)
+			return (set_errno(ESRCH));
+		rid = 0;
+		ridtype = P_ALL;
+		lidtype = P_LWPID;
+	}
+	setprocset(procset, POP_AND, lidtype, lid, ridtype, rid);
+
+	return (0);
+}
+
+long
+do_priocntlsys(int cmd, procset_t *procset, void *arg)
+{
+	return (priocntl_common(PC_VERSION, procset, cmd, (caddr_t)arg, 0,
+	    UIO_SYSSPACE));
+}
diff --git a/usr/src/uts/common/brand/lx/syscall/lx_sysinfo.c b/usr/src/uts/common/brand/lx/syscall/lx_sysinfo.c
new file mode 100644
index 0000000000..449d5882d4
--- /dev/null
+++ b/usr/src/uts/common/brand/lx/syscall/lx_sysinfo.c
@@ -0,0 +1,218 @@
+/*
+ * CDDL HEADER START
+ *
+ * The contents of this file are subject to the terms of the
+ * Common Development and Distribution License (the "License").
+ * You may not use this file except in compliance with the License.
+ *
+ * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
+ * or http://www.opensolaris.org/os/licensing.
+ * See the License for the specific language governing permissions
+ * and limitations under the License.
+ *
+ * When distributing Covered Code, include this CDDL HEADER in each
+ * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
+ * If applicable, add the following below this CDDL HEADER, with the
+ * fields enclosed by brackets "[]" replaced with your own identifying
+ * information: Portions Copyright [yyyy] [name of copyright owner]
+ *
+ * CDDL HEADER END
+ */
+/*
+ * Copyright 2006 Sun Microsystems, Inc.  All rights reserved.
+ * Use is subject to license terms.
+ * Copyright 2015 Joyent, Inc.
+ */
+
+#include <vm/anon.h>
+#include <sys/systm.h>
+#include <sys/sysmacros.h>
+#include <sys/zone.h>
+#include <sys/time.h>
+
+typedef struct lx_sysinfo {
+	int64_t si_uptime;	/* Seconds since boot */
+	uint64_t si_loads[3];	/* 1, 5, and 15 minute avg runq length */
+	uint64_t si_totalram;	/* Total memory size */
+	uint64_t si_freeram;	/* Available memory */
+	uint64_t si_sharedram;	/* Shared memory */
+	uint64_t si_bufferram;	/* Buffer memory */
+	uint64_t si_totalswap;	/* Total swap space */
+	uint64_t si_freeswap;	/* Avail swap space */
+	uint16_t si_procs;	/* Process count */
+	uint16_t si_pad;	/* Padding */
+	uint64_t si_totalhigh;	/* High memory size */
+	uint64_t si_freehigh;	/* Avail high memory */
+	uint32_t si_mem_unit;	/* Unit size of memory fields */
+} lx_sysinfo_t;
+
+#if defined(_SYSCALL32_IMPL)
+/*
+ * 64-bit kernel view of the 32-bit usermode struct.
+ */
+#pragma pack(4)
+typedef struct lx_sysinfo32 {
+	int32_t si_uptime;	/* Seconds since boot */
+	uint32_t si_loads[3];	/* 1, 5, and 15 minute avg runq length */
+	uint32_t si_totalram;	/* Total memory size */
+	uint32_t si_freeram;	/* Available memory */
+	uint32_t si_sharedram;	/* Shared memory */
+	uint32_t si_bufferram;	/* Buffer memory */
+	uint32_t si_totalswap;	/* Total swap space */
+	uint32_t si_freeswap;	/* Avail swap space */
+	uint16_t si_procs;	/* Process count */
+	uint16_t si_pad;	/* Padding */
+	uint32_t si_totalhigh;	/* High memory size */
+	uint32_t si_freehigh;	/* Avail high memory */
+	uint32_t si_mem_unit;	/* Unit size of memory fields */
+	char __si_pad[8];
+} lx_sysinfo32_t;
+#pragma pack()
+#endif
+
+extern pgcnt_t swapfs_minfree;
+
+static void
+lx_sysinfo_common(lx_sysinfo_t *si)
+{
+	zone_t *zone = curthread->t_procp->p_zone;
+	uint64_t zphysmem, zfreemem, ztotswap, zfreeswap;
+
+	si->si_uptime = gethrestime_sec() - zone->zone_boot_time;
+
+	si->si_loads[0] = zone->zone_hp_avenrun[0];
+	si->si_loads[1] = zone->zone_hp_avenrun[1];
+	si->si_loads[2] = zone->zone_hp_avenrun[2];
+
+	/*
+	 * In linux each thread looks like a process, so we conflate the
+	 * two in this stat as well.
+	 */
+	si->si_procs = (int32_t)zone->zone_nlwps;
+
+	/*
+	 * If memory or swap limits are set on the zone, use those, otherwise
+	 * use the system values. physmem and freemem are in pages, but the
+	 * zone values are in bytes. Likewise, ani_max and ani_free are in
+	 * pages.
+	 */
+	if (zone->zone_phys_mem_ctl == UINT64_MAX) {
+		zphysmem = physmem;
+		zfreemem = freemem;
+	} else {
+		zphysmem = btop(zone->zone_phys_mem_ctl);
+		zfreemem = btop(zone->zone_phys_mem_ctl - zone->zone_phys_mem);
+	}
+
+	if (zone->zone_max_swap_ctl == UINT64_MAX) {
+		ztotswap = k_anoninfo.ani_max;
+		zfreeswap = k_anoninfo.ani_free;
+	} else {
+		/*
+		 * See the comment in swapctl for a description of how free is
+		 * calculated within a zone.
+		 */
+		rctl_qty_t used;
+		spgcnt_t avail;
+		uint64_t max;
+
+		avail = MAX((spgcnt_t)(availrmem - swapfs_minfree), 0);
+		max = k_anoninfo.ani_max + k_anoninfo.ani_mem_resv + avail;
+
+		mutex_enter(&zone->zone_mem_lock);
+		ztotswap = btop(zone->zone_max_swap_ctl);
+		used = btop(zone->zone_max_swap);
+		mutex_exit(&zone->zone_mem_lock);
+
+		zfreeswap = MIN(ztotswap, max) - used;
+	}
+
+	/*
+	 * If the maximum memory stat is less than 1^20 pages (i.e. 4GB),
+	 * then we report the result in bytes.  Otherwise we use pages.
+	 * Once we start supporting >1TB systems/zones, we'll need a third
+	 * option.
+	 */
+	if (MAX(zphysmem, ztotswap) < 1024 * 1024) {
+		si->si_totalram = ptob(zphysmem);
+		si->si_freeram = ptob(zfreemem);
+		si->si_totalswap = ptob(ztotswap);
+		si->si_freeswap = ptob(zfreeswap);
+		si->si_mem_unit = 1;
+	} else {
+		si->si_totalram = zphysmem;
+		si->si_freeram = zfreemem;
+		si->si_totalswap = ztotswap;
+		si->si_freeswap = zfreeswap;
+		si->si_mem_unit = PAGESIZE;
+	}
+	si->si_bufferram = 0;
+	si->si_sharedram = 0;
+
+	/*
+	 * These two stats refer to high physical memory.  If an
+	 * application running in a Linux zone cares about this, then
+	 * either it or we are broken.
+	 */
+	si->si_totalhigh = 0;
+	si->si_freehigh = 0;
+}
+
+long
+lx_sysinfo64(caddr_t sip)
+{
+	lx_sysinfo_t si;
+
+	bzero(&si, sizeof (si));
+	lx_sysinfo_common(&si);
+
+	if (copyout(&si, sip, sizeof (si)) != 0) {
+		return (set_errno(EFAULT));
+	}
+
+	return (0);
+}
+
+#if defined(_SYSCALL32_IMPL)
+long
+lx_sysinfo32(caddr_t sip)
+{
+	lx_sysinfo_t si;
+	lx_sysinfo32_t si32;
+	int i;
+
+	lx_sysinfo_common(&si);
+
+	/*
+	 * Convert the lx_sysinfo_t into the legacy 32-bit view:
+	 */
+	bzero(&si32, sizeof (si32));
+	si32.si_uptime = si.si_uptime;
+
+	for (i = 0; i < 3; i++) {
+		if ((si.si_loads[i]) > 0x7fffffff)
+			si32.si_loads[i] = 0x7fffffff;
+		else
+			si32.si_loads[i] = si.si_loads[i];
+	}
+
+	si32.si_procs = si.si_procs;
+	si32.si_totalram = si.si_totalram;
+	si32.si_freeram = si.si_freeram;
+	si32.si_totalswap = si.si_totalswap;
+	si32.si_freeswap = si.si_freeswap;
+	si32.si_mem_unit = si.si_mem_unit;
+
+	si32.si_bufferram = si.si_bufferram;
+	si32.si_sharedram = si.si_sharedram;
+
+	si32.si_totalhigh = si.si_totalhigh;
+	si32.si_freehigh = si.si_freehigh;
+
+	if (copyout(&si32, sip, sizeof (si32)) != 0) {
+		return (set_errno(EFAULT));
+	}
+
+	return (0);
+}
+#endif
diff --git a/usr/src/uts/common/brand/lx/syscall/lx_thread_area.c b/usr/src/uts/common/brand/lx/syscall/lx_thread_area.c
new file mode 100644
index 0000000000..48d91b09cc
--- /dev/null
+++ b/usr/src/uts/common/brand/lx/syscall/lx_thread_area.c
@@ -0,0 +1,196 @@
+/*
+ * CDDL HEADER START
+ *
+ * The contents of this file are subject to the terms of the
+ * Common Development and Distribution License (the "License").
+ * You may not use this file except in compliance with the License.
+ *
+ * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
+ * or http://www.opensolaris.org/os/licensing.
+ * See the License for the specific language governing permissions
+ * and limitations under the License.
+ *
+ * When distributing Covered Code, include this CDDL HEADER in each
+ * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
+ * If applicable, add the following below this CDDL HEADER, with the
+ * fields enclosed by brackets "[]" replaced with your own identifying
+ * information: Portions Copyright [yyyy] [name of copyright owner]
+ *
+ * CDDL HEADER END
+ */
+/*
+ * Copyright 2006 Sun Microsystems, Inc.  All rights reserved.
+ * Use is subject to license terms.
+ * Copyright 2015 Joyent, Inc.
+ */
+
+#include <sys/types.h>
+#include <sys/systm.h>
+#include <sys/errno.h>
+#include <sys/cpuvar.h>
+#include <sys/archsystm.h>
+#include <sys/proc.h>
+#include <sys/brand.h>
+#include <sys/lx_brand.h>
+#include <sys/lx_ldt.h>
+#include <sys/lx_misc.h>
+#include <sys/x86_archext.h>
+#include <sys/controlregs.h>
+#include <lx_syscall.h>
+
+long
+lx_arch_prctl(int code, ulong_t addr)
+{
+#if defined(__amd64)
+	klwp_t *lwp = ttolwp(curthread);
+	lx_lwp_data_t *llwp = lwptolxlwp(lwp);
+	pcb_t *pcb = &lwp->lwp_pcb;
+
+	switch (code) {
+	case LX_ARCH_GET_FS:
+		if (copyout(&llwp->br_lx_fsbase, (void *)addr,
+		    sizeof (llwp->br_lx_fsbase)) != 0) {
+			return (set_errno(EFAULT));
+		}
+		break;
+
+	case LX_ARCH_SET_FS:
+		llwp->br_lx_fsbase = addr;
+
+		kpreempt_disable();
+		if (pcb->pcb_fsbase != llwp->br_lx_fsbase) {
+			pcb->pcb_fsbase = llwp->br_lx_fsbase;
+
+			/*
+			 * Ensure we go out via update_sregs.
+			 */
+			pcb->pcb_rupdate = 1;
+		}
+		kpreempt_enable();
+		break;
+
+	case LX_ARCH_GET_GS:
+		if (copyout(&llwp->br_lx_gsbase, (void *)addr,
+		    sizeof (llwp->br_lx_gsbase)) != 0) {
+			return (set_errno(EFAULT));
+		}
+		break;
+
+	case LX_ARCH_SET_GS:
+		llwp->br_lx_gsbase = addr;
+
+		kpreempt_disable();
+		if (pcb->pcb_gsbase != llwp->br_lx_gsbase) {
+			pcb->pcb_gsbase = llwp->br_lx_gsbase;
+
+			/*
+			 * Ensure we go out via update_sregs.
+			 */
+			pcb->pcb_rupdate = 1;
+		}
+		kpreempt_enable();
+		break;
+
+	default:
+		return (set_errno(EINVAL));
+	}
+#endif
+
+	return (0);
+}
+
+long
+lx_get_thread_area(struct ldt_info *inf)
+{
+	struct lx_lwp_data *jlwp = ttolxlwp(curthread);
+	struct ldt_info ldt_inf;
+	user_desc_t *dscrp;
+	int entry;
+
+	if (fuword32(&inf->entry_number, (uint32_t *)&entry))
+		return (set_errno(EFAULT));
+
+	if (entry < GDT_TLSMIN || entry > GDT_TLSMAX)
+		return (set_errno(EINVAL));
+
+	dscrp = jlwp->br_tls + entry - GDT_TLSMIN;
+
+	/*
+	 * convert the solaris ldt to the linux format expected by the
+	 * caller
+	 */
+	DESC_TO_LDT_INFO(dscrp, &ldt_inf);
+	ldt_inf.entry_number = entry;
+
+	if (copyout(&ldt_inf, inf, sizeof (struct ldt_info)))
+		return (set_errno(EFAULT));
+
+	return (0);
+}
+
+long
+lx_set_thread_area(struct ldt_info *inf)
+{
+	struct lx_lwp_data *jlwp = ttolxlwp(curthread);
+	struct ldt_info ldt_inf;
+	user_desc_t *dscrp;
+	int entry;
+	int i;
+
+	/* Check that casts for accessing the words in user_desc are valid */
+	ASSERT(sizeof (user_desc_t) == 8);
+
+	if (copyin(inf, &ldt_inf, sizeof (ldt_inf)))
+		return (set_errno(EFAULT));
+
+	entry = ldt_inf.entry_number;
+	if (entry == -1) {
+		/*
+		 * Find an empty entry in the tls for this thread.
+		 * The casts assume each user_desc_t entry is 8 bytes.
+		 */
+		for (i = 0, dscrp = jlwp->br_tls; i < LX_TLSNUM; i++, dscrp++) {
+			if (((uint_t *)dscrp)[0] == 0 &&
+			    ((uint_t *)dscrp)[1] == 0)
+				break;
+		}
+
+		if (i < LX_TLSNUM) {
+			/*
+			 * found one
+			 */
+			entry = i + GDT_TLSMIN;
+			if (suword32(&inf->entry_number, entry))
+				return (set_errno(EFAULT));
+		} else {
+			return (set_errno(ESRCH));
+		}
+	}
+
+	if (entry < GDT_TLSMIN || entry > GDT_TLSMAX)
+		return (set_errno(EINVAL));
+
+	/*
+	 * convert the linux ldt info to standard intel descriptor
+	 */
+	dscrp = jlwp->br_tls + entry - GDT_TLSMIN;
+
+	if (LDT_INFO_EMPTY(&ldt_inf)) {
+		((uint_t *)dscrp)[0] = 0;
+		((uint_t *)dscrp)[1] = 0;
+	} else {
+		LDT_INFO_TO_DESC(&ldt_inf, dscrp);
+	}
+
+	/*
+	 * update the gdt with the new descriptor
+	 */
+	kpreempt_disable();
+
+	for (i = 0, dscrp = jlwp->br_tls; i < LX_TLSNUM; i++, dscrp++)
+		lx_set_gdt(GDT_TLSMIN + i, dscrp);
+
+	kpreempt_enable();
+
+	return (0);
+}
diff --git a/usr/src/uts/common/brand/lx/syscall/lx_timer.c b/usr/src/uts/common/brand/lx/syscall/lx_timer.c
new file mode 100644
index 0000000000..aa22228808
--- /dev/null
+++ b/usr/src/uts/common/brand/lx/syscall/lx_timer.c
@@ -0,0 +1,367 @@
+/*
+ * This file and its contents are supplied under the terms of the
+ * Common Development and Distribution License ("CDDL"), version 1.0.
+ * You may only use this file in accordance with the terms of version
+ * 1.0 of the CDDL.
+ *
+ * A full copy of the text of the CDDL should have accompanied this
+ * source.  A copy of the CDDL is also available via the Internet at
+ * http://www.illumos.org/license/CDDL.
+ */
+
+/*
+ * Copyright 2015 Joyent, Inc.
+ */
+
+/*
+ * The illumos kernel provides two clock backends: CLOCK_REALTIME, the
+ * adjustable system wall clock; and CLOCK_HIGHRES, the monotonically
+ * increasing time source that is not subject to drift or adjustment.  By
+ * contrast, the Linux kernel is furnished with an overabundance of narrowly
+ * differentiated clock types.
+ *
+ * Fortunately, most of the commonly used Linux clock types are either similar
+ * enough to the native clock backends that they can be directly mapped, or
+ * represent queries to the per-process and per-LWP microstate counters.
+ */
+
+#include <sys/time.h>
+#include <sys/systm.h>
+#include <sys/cmn_err.h>
+#include <sys/lx_impl.h>
+
+/*
+ * From "uts/common/os/timer.c":
+ */
+extern int clock_settime(clockid_t, timespec_t *);
+extern int clock_gettime(clockid_t, timespec_t *);
+extern int clock_getres(clockid_t, timespec_t *);
+
+
+static int lx_emul_clock_getres(clockid_t, timespec_t *);
+static int lx_emul_clock_gettime(clockid_t, timespec_t *);
+static int lx_emul_clock_settime(clockid_t, timespec_t *);
+
+typedef struct lx_clock_backend {
+	clockid_t lclk_ntv_id;
+	int (*lclk_clock_getres)(clockid_t, timespec_t *);
+	int (*lclk_clock_gettime)(clockid_t, timespec_t *);
+	int (*lclk_clock_settime)(clockid_t, timespec_t *);
+} lx_clock_backend_t;
+
+/*
+ * NOTE: The Linux man pages state this structure is obsolete and is
+ * unsupported, so it is declared here for sizing purposes only.
+ */
+struct lx_timezone {
+	int tz_minuteswest;	/* minutes W of Greenwich */
+	int tz_dsttime;		/* type of dst correction */
+};
+
+/*
+ * Use the native clock_* system call implementation, but with a translated
+ * clock identifier:
+ */
+#define	NATIVE(ntv_id)							\
+	{ ntv_id, clock_getres, clock_gettime, clock_settime }
+
+/*
+ * This backend is not supported, so we provide an emulation handler:
+ */
+#define	EMUL(ntv_id)							\
+	{ ntv_id, lx_emul_clock_getres, lx_emul_clock_gettime,		\
+	    lx_emul_clock_settime }
+
+static lx_clock_backend_t lx_clock_backends[] = {
+	NATIVE(CLOCK_REALTIME),		/* LX_CLOCK_REALTIME */
+	NATIVE(CLOCK_HIGHRES),		/* LX_CLOCK_MONOTONIC */
+	EMUL(CLOCK_PROCESS_CPUTIME_ID),	/* LX_CLOCK_PROCESS_CPUTIME_ID */
+	EMUL(CLOCK_THREAD_CPUTIME_ID),	/* LX_CLOCK_THREAD_CPUTIME_ID */
+	NATIVE(CLOCK_HIGHRES),		/* LX_CLOCK_MONOTONIC_RAW */
+	NATIVE(CLOCK_REALTIME),		/* LX_CLOCK_REALTIME_COARSE */
+	NATIVE(CLOCK_HIGHRES)		/* LX_CLOCK_MONOTONIC_COARSE */
+};
+
+#define	LX_CLOCK_MAX \
+	(sizeof (lx_clock_backends) / sizeof (lx_clock_backends[0]))
+#define	LX_CLOCK_BACKEND(clk) \
+	((clk) < LX_CLOCK_MAX && (clk) >= 0 ? &lx_clock_backends[(clk)] : NULL)
+
+static int
+lx_emul_clock_settime(clockid_t clock, timespec_t *tp)
+{
+	return (set_errno(EINVAL));
+}
+
+static int
+lx_emul_clock_gettime(clockid_t clock, timespec_t *tp)
+{
+	timespec_t t;
+
+	switch (clock) {
+	case CLOCK_PROCESS_CPUTIME_ID: {
+		proc_t *p = ttoproc(curthread);
+		hrtime_t snsecs, unsecs;
+
+		/*
+		 * Based on getrusage() in "rusagesys.c":
+		 */
+		mutex_enter(&p->p_lock);
+		unsecs = mstate_aggr_state(p, LMS_USER);
+		snsecs = mstate_aggr_state(p, LMS_SYSTEM);
+		mutex_exit(&p->p_lock);
+
+		hrt2ts(unsecs + snsecs, &t);
+		break;
+	}
+
+	case CLOCK_THREAD_CPUTIME_ID: {
+		klwp_t *lwp = ttolwp(curthread);
+		struct mstate *ms = &lwp->lwp_mstate;
+		hrtime_t snsecs, unsecs;
+
+		/*
+		 * Based on getrusage_lwp() in "rusagesys.c":
+		 */
+		unsecs = ms->ms_acct[LMS_USER];
+		snsecs = ms->ms_acct[LMS_SYSTEM] + ms->ms_acct[LMS_TRAP];
+
+		scalehrtime(&unsecs);
+		scalehrtime(&snsecs);
+
+		hrt2ts(unsecs + snsecs, &t);
+		break;
+	}
+
+	default:
+		return (set_errno(EINVAL));
+	}
+
+#if defined(_SYSCALL32_IMPL)
+	if (get_udatamodel() != DATAMODEL_NATIVE) {
+		timespec32_t t32;
+
+		if (TIMESPEC_OVERFLOW(&t)) {
+			return (set_errno(EOVERFLOW));
+		}
+		TIMESPEC_TO_TIMESPEC32(&t32, &t);
+
+		if (copyout(&t32, tp, sizeof (t32)) != 0) {
+			return (set_errno(EFAULT));
+		}
+
+		return (0);
+	}
+#endif
+
+	if (copyout(&t, tp, sizeof (t)) != 0) {
+		return (set_errno(EFAULT));
+	}
+
+	return (0);
+}
+
+static int
+lx_emul_clock_getres(clockid_t clock, timespec_t *tp)
+{
+	timespec_t t;
+
+	if (tp == NULL) {
+		return (0);
+	}
+
+	switch (clock) {
+	case CLOCK_PROCESS_CPUTIME_ID:
+	case CLOCK_THREAD_CPUTIME_ID:
+		/*
+		 * These clock backends return microstate accounting values for
+		 * the LWP or the entire process.  The Linux kernel claims they
+		 * have nanosecond resolution; so will we.
+		 */
+		t.tv_sec = 0;
+		t.tv_nsec = 1;
+		break;
+
+	default:
+		return (set_errno(EINVAL));
+	}
+
+#if defined(_SYSCALL32_IMPL)
+	if (get_udatamodel() != DATAMODEL_NATIVE) {
+		timespec32_t t32;
+
+		if (TIMESPEC_OVERFLOW(&t)) {
+			return (set_errno(EOVERFLOW));
+		}
+		TIMESPEC_TO_TIMESPEC32(&t32, &t);
+
+		if (copyout(&t32, tp, sizeof (t32)) != 0) {
+			return (set_errno(EFAULT));
+		}
+
+		return (0);
+	}
+#endif
+
+	if (copyout(&t, tp, sizeof (t)) != 0) {
+		return (set_errno(EFAULT));
+	}
+
+	return (0);
+}
+
+static void
+lx_clock_unsupported(int clock)
+{
+	char buf[100];
+
+	(void) snprintf(buf, sizeof (buf), "unsupported clock: %d", clock);
+	lx_unsupported(buf);
+}
+
+long
+lx_clock_settime(int clock, timespec_t *tp)
+{
+	lx_clock_backend_t *backend;
+
+	if ((backend = LX_CLOCK_BACKEND(clock)) == NULL) {
+		lx_clock_unsupported(clock);
+		return (set_errno(EINVAL));
+	}
+
+	return (backend->lclk_clock_settime(backend->lclk_ntv_id, tp));
+}
+
+long
+lx_clock_gettime(int clock, timespec_t *tp)
+{
+	lx_clock_backend_t *backend;
+
+	if ((backend = LX_CLOCK_BACKEND(clock)) == NULL) {
+		lx_clock_unsupported(clock);
+		return (set_errno(EINVAL));
+	}
+
+	return (backend->lclk_clock_gettime(backend->lclk_ntv_id, tp));
+}
+
+long
+lx_clock_getres(int clock, timespec_t *tp)
+{
+	lx_clock_backend_t *backend;
+
+	if (tp == NULL) {
+		return (0);
+	}
+
+	if ((backend = LX_CLOCK_BACKEND(clock)) == NULL) {
+		lx_clock_unsupported(clock);
+		return (set_errno(EINVAL));
+	}
+
+	return (backend->lclk_clock_getres(backend->lclk_ntv_id, tp));
+}
+
+
+long
+lx_gettimeofday(struct timeval *tvp, struct lx_timezone *tzp)
+{
+	struct lx_timezone tz;
+
+	bzero(&tz, sizeof (tz));
+
+	/*
+	 * We want to be similar to libc which just does a fasttrap to
+	 * gethrestime and simply converts that result. We follow how uniqtime
+	 * does the conversion but we can't use that code since it does some
+	 * extra work which can cause the result to bounce around based on which
+	 * CPU we run on.
+	 */
+	if (tvp != NULL) {
+		struct timeval tv;
+		timestruc_t ts;
+		int usec, nsec;
+
+		gethrestime(&ts);
+		nsec = ts.tv_nsec;
+		usec = nsec + (nsec >> 2);
+		usec = nsec + (usec >> 1);
+		usec = nsec + (usec >> 2);
+		usec = nsec + (usec >> 4);
+		usec = nsec - (usec >> 3);
+		usec = nsec + (usec >> 2);
+		usec = nsec + (usec >> 3);
+		usec = nsec + (usec >> 4);
+		usec = nsec + (usec >> 1);
+		usec = nsec + (usec >> 6);
+		usec = usec >> 10;
+
+		tv.tv_sec = ts.tv_sec;
+		tv.tv_usec = usec;
+
+		if (get_udatamodel() == DATAMODEL_NATIVE) {
+			if (copyout(&tv, tvp, sizeof (tv)) != 0)
+				return (set_errno(EFAULT));
+		}
+#ifdef _SYSCALL32_IMPL
+		else {
+			struct timeval32 tv32;
+
+			if (TIMEVAL_OVERFLOW(&tv))
+				return (set_errno(EOVERFLOW));
+			TIMEVAL_TO_TIMEVAL32(&tv32, &tv);
+
+			if (copyout(&tv32, tvp, sizeof (tv32)))
+				return (set_errno(EFAULT));
+		}
+#endif
+	}
+
+	/*
+	 * The Linux man page states use of the second parameter is obsolete,
+	 * but gettimeofday(2) should still return EFAULT if it is set
+	 * to a bad non-NULL pointer (sigh...)
+	 */
+	if (tzp != NULL && copyout(&tz, tzp, sizeof (tz)) != 0)
+		return (set_errno(EFAULT));
+
+	return (0);
+}
+
+/*
+ * On Linux a bad buffer will set errno to EFAULT, and on Illumos the failure
+ * mode is documented as "undefined."
+ */
+long
+lx_time(time_t *tp)
+{
+	timestruc_t ts;
+	struct timeval tv;
+
+	gethrestime(&ts);
+	tv.tv_sec = ts.tv_sec;
+	tv.tv_usec = 0;
+
+	if (get_udatamodel() == DATAMODEL_NATIVE) {
+		if (tp != NULL &&
+		    copyout(&tv.tv_sec, tp, sizeof (tv.tv_sec)) != 0)
+			return (set_errno(EFAULT));
+
+		return (tv.tv_sec);
+	}
+#ifdef _SYSCALL32_IMPL
+	else {
+		struct timeval32 tv32;
+
+		if (TIMEVAL_OVERFLOW(&tv))
+			return (set_errno(EOVERFLOW));
+		TIMEVAL_TO_TIMEVAL32(&tv32, &tv);
+
+		if (tp != NULL &&
+		    copyout(&tv32.tv_sec, tp, sizeof (tv32.tv_sec)))
+			return (set_errno(EFAULT));
+
+		return (tv32.tv_sec);
+	}
+#endif
+}
diff --git a/usr/src/uts/common/brand/lx/syscall/lx_wait.c b/usr/src/uts/common/brand/lx/syscall/lx_wait.c
new file mode 100644
index 0000000000..e6c5607e43
--- /dev/null
+++ b/usr/src/uts/common/brand/lx/syscall/lx_wait.c
@@ -0,0 +1,473 @@
+/*
+ * CDDL HEADER START
+ *
+ * The contents of this file are subject to the terms of the
+ * Common Development and Distribution License (the "License").
+ * You may not use this file except in compliance with the License.
+ *
+ * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
+ * or http://www.opensolaris.org/os/licensing.
+ * See the License for the specific language governing permissions
+ * and limitations under the License.
+ *
+ * When distributing Covered Code, include this CDDL HEADER in each
+ * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
+ * If applicable, add the following below this CDDL HEADER, with the
+ * fields enclosed by brackets "[]" replaced with your own identifying
+ * information: Portions Copyright [yyyy] [name of copyright owner]
+ *
+ * CDDL HEADER END
+ */
+
+/*
+ * Copyright 2007 Sun Microsystems, Inc.  All rights reserved.
+ * Use is subject to license terms.
+ * Copyright 2015 Joyent, Inc.
+ */
+
+/*
+ * wait() family of functions.
+ *
+ * The first minor difference between the Linux and Solaris family of wait()
+ * calls is that the values for WNOHANG and WUNTRACED are different. Thankfully,
+ * the exit status values are identical between the two implementations.
+ *
+ * Things get very different and very complicated when we introduce the Linux
+ * threading model.  Under linux, both threads and child processes are
+ * represented as processes.  However, the behavior of wait() with respect to
+ * each child varies according to the flags given to clone()
+ *
+ *	SIGCHLD 	The SIGCHLD signal should be sent on termination
+ *	CLONE_THREAD	The child shares the same thread group as the parent
+ *	CLONE_DETACHED	The parent receives no notification when the child exits
+ *
+ * The following flags control the Linux behavior w.r.t. the above attributes:
+ *
+ * 	__WALL		Wait on all children, regardless of type
+ * 	__WCLONE	Wait only on non-SIGCHLD children
+ * 	__WNOTHREAD	Don't wait on children of other threads in this group
+ *
+ * The following chart shows whether wait() returns when the child exits:
+ *
+ *                           default    __WCLONE    __WALL
+ *           no SIGCHLD		-	    X	      X
+ *              SIGCHLD		X	    -	      X
+ *
+ * The following chart shows whether wait() returns when the grandchild exits:
+ *
+ *                           default   __WNOTHREAD
+ * 	no CLONE_THREAD		-	    -
+ *         CLONE_THREAD		X	    -
+ *
+ * The CLONE_DETACHED flag is universal - when the child exits, no state is
+ * stored and wait() has no effect.
+ *
+ * XXX Support the above combination of options, or some reasonable subset that
+ *     covers at least fork() and pthread_create().
+ */
+
+#include <sys/wait.h>
+#include <sys/brand.h>
+#include <sys/lx_brand.h>
+#include <sys/lx_types.h>
+#include <sys/lx_siginfo.h>
+#include <lx_signum.h>
+#include <lx_errno.h>
+#include <lx_syscall.h>
+
+/*
+ * From "uts/common/os/exit.c" and "uts/common/syscall/rusagesys.c":
+ */
+extern int waitid(idtype_t, id_t, k_siginfo_t *, int);
+extern int rusagesys(int, void *, void *, void *, void *);
+
+/*
+ * Convert between Linux options and Solaris options, returning -1 if any
+ * invalid flags are found.
+ */
+#define	LX_WNOHANG	0x00000001
+#define	LX_WUNTRACED	0x00000002
+#define	LX_WSTOPPED	LX_WUNTRACED
+#define	LX_WEXITED	0x00000004
+#define	LX_WCONTINUED	0x00000008
+#define	LX_WNOWAIT	0x01000000
+
+#define	LX_WNOTHREAD	0x20000000
+#define	LX_WALL		0x40000000
+#define	LX_WCLONE	0x80000000
+
+#define	LX_P_ALL	0x0
+#define	LX_P_PID	0x1
+#define	LX_P_GID	0x2
+
+/*
+ * Split the passed waitpid/waitid options into two separate variables:
+ * those for the native illumos waitid(2), and the extra Linux-specific
+ * options we will handle in our brand-specific code.
+ */
+static int
+ltos_options(uintptr_t options, int *native_options, int *extra_options)
+{
+	int newoptions = 0;
+
+	if (((options) & ~(LX_WNOHANG | LX_WUNTRACED | LX_WEXITED |
+	    LX_WCONTINUED | LX_WNOWAIT | LX_WNOTHREAD | LX_WALL |
+	    LX_WCLONE)) != 0) {
+		return (-1);
+	}
+
+	*extra_options = options & (LX_WNOTHREAD | LX_WALL | LX_WCLONE);
+
+	if (options & LX_WNOHANG)
+		newoptions |= WNOHANG;
+	if (options & LX_WUNTRACED)
+		newoptions |= WUNTRACED;
+	if (options & LX_WEXITED)
+		newoptions |= WEXITED;
+	if (options & LX_WCONTINUED)
+		newoptions |= WCONTINUED;
+	if (options & LX_WNOWAIT)
+		newoptions |= WNOWAIT;
+
+	/*
+	 * The trapped option is implicit on Linux.
+	 */
+	newoptions |= WTRAPPED;
+
+	*native_options = newoptions;
+	return (0);
+}
+
+static int
+lx_wstat(int code, int status)
+{
+	int stat = 0;
+
+	switch (code) {
+	case CLD_EXITED:
+		stat = status << 8;
+		break;
+	case CLD_DUMPED:
+		stat = lx_stol_signo(status, SIGKILL) | WCOREFLG;
+		break;
+	case CLD_KILLED:
+		stat = lx_stol_signo(status, SIGKILL);
+		break;
+	case CLD_TRAPPED:
+	case CLD_STOPPED:
+		stat = (lx_stol_status(status, SIGKILL) << 8) | WSTOPFLG;
+		break;
+	case CLD_CONTINUED:
+		stat = WCONTFLG;
+		break;
+	}
+
+	return (stat);
+}
+
+static int
+lx_call_waitid(idtype_t idtype, id_t id, k_siginfo_t *sip, int native_options,
+    int extra_options)
+{
+	lx_lwp_data_t *lwpd = ttolxlwp(curthread);
+	int error;
+
+	/*
+	 * Our brand-specific waitid helper only understands a subset of
+	 * the possible idtypes.  Ensure we keep to that subset here:
+	 */
+	if (idtype != P_ALL && idtype != P_PID && idtype != P_PGID) {
+		return (EINVAL);
+	}
+
+	/*
+	 * Enable the return of emulated ptrace(2) stop conditions
+	 * through lx_waitid_helper, and stash the Linux-specific
+	 * extra waitid() flags.
+	 */
+	lwpd->br_waitid_emulate = B_TRUE;
+	lwpd->br_waitid_flags = extra_options;
+
+	if ((error = waitid(idtype, id, sip, native_options)) == EINTR) {
+		/*
+		 * According to signal(7), the wait4(2), waitid(2), and
+		 * waitpid(2) system calls are restartable.
+		 */
+		ttolxlwp(curthread)->br_syscall_restart = B_TRUE;
+	}
+
+	lwpd->br_waitid_emulate = B_FALSE;
+	lwpd->br_waitid_flags = 0;
+
+	return (error);
+}
+
+long
+lx_wait4(uintptr_t p1, uintptr_t p2, uintptr_t p3, uintptr_t p4)
+{
+	k_siginfo_t info = { 0 };
+	idtype_t idtype;
+	id_t id;
+	int status = 0;
+	pid_t pid = (pid_t)p1;
+	int error;
+	int native_options, extra_options;
+	int *statusp = (int *)p2;
+	void *rup = (void *)p4;
+
+	if (ltos_options(p3, &native_options, &extra_options) == -1) {
+		return (set_errno(EINVAL));
+	}
+
+	if (pid > maxpid) {
+		return (set_errno(ECHILD));
+	}
+
+	/*
+	 * While not listed as a valid return code, Linux's wait4(2) does,
+	 * in fact, get an EFAULT if either the status pointer or rusage
+	 * pointer is invalid. Since a failed waitpid should leave child
+	 * process in a state where a future wait4(2) will succeed, we
+	 * check them by copying out the values their buffers originally
+	 * contained.  (We need to do this as a failed system call should
+	 * never affect the contents of a passed buffer.)
+	 *
+	 * This will fail if the buffers in question are write-only.
+	 */
+	if (statusp != NULL) {
+		if (copyin(statusp, &status, sizeof (status)) != 0 ||
+		    copyout(&status, statusp, sizeof (status)) != 0) {
+			return (set_errno(EFAULT));
+		}
+	}
+
+	/*
+	 * Do the same check for the "struct rusage" pointer, which differs
+	 * in size for 32- and 64-bit processes.
+	 */
+	if (rup != NULL) {
+		struct rusage ru;
+		void *krup = &ru;
+		size_t rusz = sizeof (ru);
+#if defined(_SYSCALL32_IMPL)
+		struct rusage32 ru32;
+
+		if (get_udatamodel() != DATAMODEL_NATIVE) {
+			krup = &ru32;
+			rusz = sizeof (ru32);
+		}
+#endif
+
+		if (copyin(rup, krup, rusz) != 0 ||
+		    copyout(krup, rup, rusz) != 0) {
+			return (set_errno(EFAULT));
+		}
+	}
+
+	if (pid < -1) {
+		idtype = P_PGID;
+		id = -pid;
+	} else if (pid == -1) {
+		idtype = P_ALL;
+		id = 0;
+	} else if (pid == 0) {
+		idtype = P_PGID;
+		mutex_enter(&pidlock);
+		id = curproc->p_pgrp;
+		mutex_exit(&pidlock);
+	} else {
+		idtype = P_PID;
+		id = pid;
+	}
+
+	native_options |= (WEXITED | WTRAPPED);
+
+	if ((error = lx_call_waitid(idtype, id, &info, native_options,
+	    extra_options)) != 0) {
+		return (set_errno(error));
+	}
+
+	/*
+	 * If the WNOHANG flag was specified and no child was found return 0.
+	 */
+	if ((native_options & WNOHANG) && info.si_pid == 0) {
+		return (0);
+	}
+
+	status = lx_wstat(info.si_code, info.si_status);
+
+	/*
+	 * Unfortunately if this attempt to copy out either the status or the
+	 * rusage fails, the process will be in an inconsistent state as
+	 * subsequent calls to wait for the same child will fail where they
+	 * should succeed on a Linux system. This, however, is rather
+	 * unlikely since we tested the validity of both above.
+	 */
+	if (statusp != NULL) {
+		if (copyout(&status, statusp, sizeof (status)) != 0) {
+			return (set_errno(EFAULT));
+		}
+	}
+
+	if (rup != NULL) {
+		if ((error = rusagesys(_RUSAGESYS_GETRUSAGE_CHLD, rup, NULL,
+		    NULL, NULL)) != 0) {
+			return (set_errno(error));
+		}
+	}
+
+	return (info.si_pid);
+}
+
+long
+lx_waitpid(uintptr_t p1, uintptr_t p2, uintptr_t p3)
+{
+	return (lx_wait4(p1, p2, p3, NULL));
+}
+
+static int
+stol_ksiginfo(k_siginfo_t *sip, uintptr_t lxsip)
+{
+	lx_siginfo_t lsi;
+
+	bzero(&lsi, sizeof (lsi));
+	lsi.lsi_signo = lx_stol_signo(sip->si_signo, SIGCLD);
+	lsi.lsi_code = lx_stol_sigcode(sip->si_code);
+	lsi.lsi_errno = lx_errno(sip->si_errno, EINVAL);
+
+	switch (lsi.lsi_signo) {
+	case LX_SIGPOLL:
+		lsi.lsi_band = sip->si_band;
+		lsi.lsi_fd = sip->si_fd;
+		break;
+
+	case LX_SIGCHLD:
+		lsi.lsi_pid = sip->si_pid;
+		if (sip->si_code <= 0 || sip->si_code == CLD_EXITED) {
+			lsi.lsi_status = sip->si_status;
+		} else {
+			lsi.lsi_status = lx_stol_status(sip->si_status,
+			    SIGKILL);
+		}
+		lsi.lsi_utime = sip->si_utime;
+		lsi.lsi_stime = sip->si_stime;
+		break;
+
+	case LX_SIGILL:
+	case LX_SIGBUS:
+	case LX_SIGFPE:
+	case LX_SIGSEGV:
+		lsi.lsi_addr = sip->si_addr;
+		break;
+
+	default:
+		lsi.lsi_pid = sip->si_pid;
+		lsi.lsi_uid = LX_UID32_TO_UID16(sip->si_uid);
+	}
+
+	if (copyout(&lsi, (void *)lxsip, sizeof (lsi)) != 0) {
+		return (set_errno(EFAULT));
+	}
+
+	return (0);
+}
+
+#if defined(_SYSCALL32_IMPL)
+static int
+stol_ksiginfo32(k_siginfo_t *sip, uintptr_t lxsip)
+{
+	lx_siginfo32_t lsi;
+
+	bzero(&lsi, sizeof (lsi));
+	lsi.lsi_signo = lx_stol_signo(sip->si_signo, SIGCLD);
+	lsi.lsi_code = lx_stol_sigcode(sip->si_code);
+	lsi.lsi_errno = lx_errno(sip->si_errno, EINVAL);
+
+	switch (lsi.lsi_signo) {
+	case LX_SIGPOLL:
+		lsi.lsi_band = sip->si_band;
+		lsi.lsi_fd = sip->si_fd;
+		break;
+
+	case LX_SIGCHLD:
+		lsi.lsi_pid = sip->si_pid;
+		if (sip->si_code <= 0 || sip->si_code == CLD_EXITED) {
+			lsi.lsi_status = sip->si_status;
+		} else {
+			lsi.lsi_status = lx_stol_status(sip->si_status,
+			    SIGKILL);
+		}
+		lsi.lsi_utime = sip->si_utime;
+		lsi.lsi_stime = sip->si_stime;
+		break;
+
+	case LX_SIGILL:
+	case LX_SIGBUS:
+	case LX_SIGFPE:
+	case LX_SIGSEGV:
+		lsi.lsi_addr = (caddr32_t)(uintptr_t)sip->si_addr;
+		break;
+
+	default:
+		lsi.lsi_pid = sip->si_pid;
+		lsi.lsi_uid = LX_UID32_TO_UID16(sip->si_uid);
+	}
+
+	if (copyout(&lsi, (void *)lxsip, sizeof (lsi)) != 0) {
+		return (set_errno(EFAULT));
+	}
+
+	return (0);
+}
+#endif
+
+long
+lx_waitid(uintptr_t idtype, uintptr_t id, uintptr_t infop, uintptr_t opt)
+{
+	int error;
+	int native_options, extra_options;
+	k_siginfo_t info = { 0 };
+
+	if (ltos_options(opt, &native_options, &extra_options) == -1) {
+		return (set_errno(EINVAL));
+	}
+
+	if (((opt) & (LX_WEXITED | LX_WSTOPPED | LX_WCONTINUED)) == 0) {
+		return (set_errno(EINVAL));
+	}
+
+	switch (idtype) {
+	case LX_P_ALL:
+		idtype = P_ALL;
+		break;
+	case LX_P_PID:
+		idtype = P_PID;
+		break;
+	case LX_P_GID:
+		idtype = P_PGID;
+		break;
+	default:
+		return (set_errno(EINVAL));
+	}
+
+	if ((error = lx_call_waitid(idtype, id, &info, native_options,
+	    extra_options)) != 0) {
+		return (set_errno(error));
+	}
+
+	/*
+	 * If the WNOHANG flag was specified and no child was found return 0.
+	 */
+	if ((native_options & WNOHANG) && info.si_pid == 0) {
+		return (0);
+	}
+
+#if defined(_SYSCALL32_IMPL)
+	if (get_udatamodel() != DATAMODEL_NATIVE) {
+		return (stol_ksiginfo32(&info, infop));
+	} else
+#endif
+	{
+		return (stol_ksiginfo(&info, infop));
+	}
+}
diff --git a/usr/src/uts/common/brand/lx/syscall/lx_xattr.c b/usr/src/uts/common/brand/lx/syscall/lx_xattr.c
new file mode 100644
index 0000000000..ea23c3e4b8
--- /dev/null
+++ b/usr/src/uts/common/brand/lx/syscall/lx_xattr.c
@@ -0,0 +1,31 @@
+/*
+ * This file and its contents are supplied under the terms of the
+ * Common Development and Distribution License ("CDDL"), version 1.0.
+ * You may only use this file in accordance with the terms of version
+ * 1.0 of the CDDL.
+ *
+ * A full copy of the text of the CDDL should have accompanied this
+ * source.  A copy of the CDDL is also available via the Internet at
+ * http://www.illumos.org/license/CDDL.
+ */
+
+/*
+ * Copyright 2015 Joyent, Inc.
+ */
+
+#include <sys/errno.h>
+#include <sys/systm.h>
+#include <sys/file.h>
+
+/*
+ * *xattr() family of functions.
+ *
+ * These are currently unimplemented.  We return EOPNOTSUPP for now, rather
+ * than using NOSYS_NO_EQUIV to avoid unwanted stderr output from ls(1).
+ */
+
+long
+lx_xattr(void)
+{
+	return (set_errno(EOPNOTSUPP));
+}
diff --git a/usr/src/uts/common/brand/lx/sysfs/lx_sysfs.h b/usr/src/uts/common/brand/lx/sysfs/lx_sysfs.h
new file mode 100644
index 0000000000..a30fdd9c06
--- /dev/null
+++ b/usr/src/uts/common/brand/lx/sysfs/lx_sysfs.h
@@ -0,0 +1,192 @@
+/*
+ * This file and its contents are supplied under the terms of the
+ * Common Development and Distribution License ("CDDL"), version 1.0.
+ * You may only use this file in accordance with the terms of version
+ * 1.0 of the CDDL.
+ *
+ * A full copy of the text of the CDDL should have accompanied this
+ * source.  A copy of the CDDL is also available via the Internet at
+ * http://www.illumos.org/license/CDDL.
+ */
+
+/*
+ * Copyright 2015 Joyent, Inc.
+ */
+
+#ifndef	_LXSYSFS_H
+#define	_LXSYSFS_H
+
+#ifdef	__cplusplus
+extern "C" {
+#endif
+
+/*
+ * lx_sysfs.h: declarations, data structures and macros for lx_sysfs
+ */
+
+
+#include <sys/types.h>
+#include <sys/param.h>
+#include <sys/policy.h>
+#include <sys/debug.h>
+#include <sys/dirent.h>
+#include <sys/errno.h>
+#include <sys/file.h>
+#include <sys/kmem.h>
+#include <sys/pathname.h>
+#include <sys/systm.h>
+#include <sys/var.h>
+#include <sys/user.h>
+#include <sys/t_lock.h>
+#include <sys/sysmacros.h>
+#include <sys/cred.h>
+#include <sys/priv.h>
+#include <sys/vnode.h>
+#include <sys/vfs.h>
+#include <sys/statvfs.h>
+#include <sys/cmn_err.h>
+#include <sys/zone.h>
+#include <sys/uio.h>
+#include <sys/utsname.h>
+#include <sys/dnlc.h>
+#include <sys/atomic.h>
+#include <sys/sunddi.h>
+#include <sys/sunldi.h>
+#include <vm/as.h>
+#include <vm/anon.h>
+#include <sys/netstack.h>
+#include <inet/ip.h>
+#include <inet/ip_if.h>
+
+/*
+ * Convert a vnode into an lxsys_mnt_t
+ */
+#define	VTOLXSM(vp)	((lxsys_mnt_t *)(vp)->v_vfsp->vfs_data)
+
+/*
+ * convert a vnode into an lxsys_node
+ */
+#define	VTOLXS(vp)	((lxsys_node_t *)(vp)->v_data)
+
+/*
+ * convert a lxsys_node into a vnode
+ */
+#define	LXSTOV(lxsnp)	((lxsnp)->lxsys_vnode)
+
+/*
+ * convert a lxsys_node into zone for fs
+ */
+#define	LXSTOZ(lxsnp) \
+	(((lxsys_mnt_t *)(lxsnp)->lxsys_vnode->v_vfsp->vfs_data)->lxsysm_zone)
+
+#define	LXSNSIZ		256	/* max size of lx /sys file name entries */
+
+/*
+ * Pretend that a directory entry takes 16 bytes
+ */
+#define	LXSYS_SDSIZE	16
+
+/* Root sysfs lxsys_instance */
+#define	LXSYS_INST_ROOT	0
+
+/*
+ * Node/file types for lx /sys files
+ * (directories and files contained therein).
+ */
+typedef enum lxsys_nodetype {
+	LXSYS_NONE,		/* None-type to keep inodes non-zero	*/
+	LXSYS_STATIC,		/* Statically defined entries		*/
+	LXSYS_CLASS_NET,	/* /sys/class/net/<iface>		*/
+	LXSYS_DEVICES_NET,	/* /sys/devices/virtual/net/<iface>	*/
+	LXSYS_MAXTYPE,		/* type limit				*/
+} lxsys_nodetype_t;
+
+/*
+ * external dirent characteristics
+ */
+typedef struct {
+	unsigned int	d_idnum;
+	char		*d_name;
+} lxsys_dirent_t;
+
+typedef struct {
+	unsigned int	dl_instance;
+	lxsys_dirent_t	*dl_list;
+	int		dl_length;
+} lxsys_dirlookup_t;
+
+/*
+ * This is the lx sysfs private data object
+ * which is attached to v_data in the vnode structure
+ */
+struct lxsys_node;
+typedef struct lxsys_node lxsys_node_t;
+struct lxsys_node {
+	lxsys_nodetype_t	lxsys_type;	/* type ID of node 	*/
+	unsigned int		lxsys_instance;	/* instance ID node	*/
+	unsigned int		lxsys_endpoint;	/* endpoint ID node	*/
+	vnode_t			*lxsys_vnode;	/* vnode for the node	*/
+	vnode_t			*lxsys_parent;	/* parent directory	*/
+	vnode_t			*lxsys_realvp;	/* real vnode		*/
+	lxsys_node_t		*lxsys_next;	/* next list entry	*/
+	timestruc_t		lxsys_time;	/* creation time	*/
+	mode_t			lxsys_mode;	/* file mode bits	*/
+	uid_t			lxsys_uid;	/* file owner		*/
+	gid_t			lxsys_gid;	/* file group owner	*/
+	ino_t			lxsys_ino;	/* node id		*/
+};
+
+/*
+ * This is the lxsysfs private data object
+ * which is attached to vfs_data in the vfs structure
+ */
+typedef struct lxsys_mnt {
+	kmutex_t	lxsysm_lock;	/* protects fields		*/
+	lxsys_node_t	*lxsysm_node;	/* node at root of sys mount	*/
+	zone_t		*lxsysm_zone;	/* zone for this mount		*/
+} lxsys_mnt_t;
+
+extern vnodeops_t	*lxsys_vnodeops;
+
+typedef struct mounta	mounta_t;
+
+extern void lxsys_initnodecache();
+extern void lxsys_fininodecache();
+extern ino_t lxsys_inode(lxsys_nodetype_t, unsigned int, unsigned int);
+extern ino_t lxsys_parentinode(lxsys_node_t *);
+extern lxsys_node_t *lxsys_getnode(vnode_t *, lxsys_nodetype_t, unsigned int,
+    unsigned int);
+extern lxsys_node_t *lxsys_getnode_static(vnode_t *, unsigned int);
+extern void lxsys_freenode(lxsys_node_t *);
+
+extern netstack_t *lxsys_netstack(lxsys_node_t *);
+extern ill_t *lxsys_find_ill(ip_stack_t *, uint_t);
+
+typedef struct lxpr_uiobuf {
+	uio_t *uiop;
+	char *buffer;
+	uint32_t bufsize;
+	char *pos;
+	size_t beg;
+	int error;
+} lxsys_uiobuf_t;
+
+extern lxsys_uiobuf_t *lxsys_uiobuf_new(uio_t *);
+extern void lxsys_uiobuf_free(lxsys_uiobuf_t *);
+extern void lxsys_uiobuf_seterr(lxsys_uiobuf_t *, int);
+extern int lxsys_uiobuf_flush(lxsys_uiobuf_t *);
+extern void lxsys_uiobuf_write(lxsys_uiobuf_t *, const char *, size_t);
+extern void lxsys_uiobuf_printf(lxsys_uiobuf_t *uiobuf, const char *fmt, ...);
+
+#ifdef	__cplusplus
+}
+#endif
+
+#ifndef islower
+#define	islower(x)	(((unsigned)(x) >= 'a') && ((unsigned)(x) <= 'z'))
+#endif
+#ifndef toupper
+#define	toupper(x)	(islower(x) ? (x) - 'a' + 'A' : (x))
+#endif
+
+#endif /* _LXSYSFS_H */
diff --git a/usr/src/uts/common/brand/lx/sysfs/lx_syssubr.c b/usr/src/uts/common/brand/lx/sysfs/lx_syssubr.c
new file mode 100644
index 0000000000..36da7fe546
--- /dev/null
+++ b/usr/src/uts/common/brand/lx/sysfs/lx_syssubr.c
@@ -0,0 +1,461 @@
+/*
+ * This file and its contents are supplied under the terms of the
+ * Common Development and Distribution License ("CDDL"), version 1.0.
+ * You may only use this file in accordance with the terms of version
+ * 1.0 of the CDDL.
+ *
+ * A full copy of the text of the CDDL should have accompanied this
+ * source.  A copy of the CDDL is also available via the Internet at
+ * http://www.illumos.org/license/CDDL.
+ */
+
+/*
+ * Copyright 2015 Joyent, Inc.
+ */
+
+/*
+ * lx_syssubr.c: Various functions for the /sys vnodeops.
+ */
+
+#include <sys/varargs.h>
+
+#include <sys/cpuvar.h>
+#include <sys/mman.h>
+#include <sys/vmsystm.h>
+#include <sys/prsystm.h>
+
+#include "lx_sysfs.h"
+
+#define	LXSYSCACHE_NAME "lxsys_cache"
+
+static int lxsys_node_constructor(void *, void *, int);
+static void lxsys_node_destructor(void *, void *);
+
+static kmem_cache_t *lxsys_node_cache;
+
+void
+lxsys_initnodecache()
+{
+	lxsys_node_cache = kmem_cache_create(LXSYSCACHE_NAME,
+	    sizeof (lxsys_node_t), 0,
+	    lxsys_node_constructor, lxsys_node_destructor, NULL, NULL, NULL, 0);
+}
+
+void
+lxsys_fininodecache()
+{
+	kmem_cache_destroy(lxsys_node_cache);
+}
+
+/* ARGSUSED */
+static int
+lxsys_node_constructor(void *buf, void *un, int kmflags)
+{
+	lxsys_node_t	*lxsnp = buf;
+	vnode_t		*vp;
+
+	vp = lxsnp->lxsys_vnode = vn_alloc(kmflags);
+	if (vp == NULL)
+		return (-1);
+
+	(void) vn_setops(vp, lxsys_vnodeops);
+	vp->v_data = lxsnp;
+
+	return (0);
+}
+
+/* ARGSUSED */
+static void
+lxsys_node_destructor(void *buf, void *un)
+{
+	lxsys_node_t	*lxsnp = buf;
+
+	vn_free(LXSTOV(lxsnp));
+}
+
+/*
+ * Calculate an inode number
+ *
+ * This takes various bits of info and munges them
+ * to give the inode number for an lxsys node
+ */
+ino_t
+lxsys_inode(lxsys_nodetype_t type, unsigned int instance,
+    unsigned int endpoint)
+{
+	/*
+	 * Sysfs Inode format:
+	 * 0000AABBBBCC
+	 *
+	 * AA - TYPE
+	 * BBBB - INSTANCE
+	 * CC - ENDPOINT
+	 */
+	ASSERT(instance <= 0xffff);
+	ASSERT(endpoint <= 0xff);
+
+	return ((ino_t)(type << 24)|(instance << 8)|endpoint);
+}
+
+/*
+ * Return inode number of parent (directory)
+ */
+ino_t
+lxsys_parentinode(lxsys_node_t *lxsnp)
+{
+	/*
+	 * If the input node is the root then the parent inode
+	 * is the mounted on inode so just return our inode number
+	 */
+	if (lxsnp->lxsys_type == LXSYS_STATIC &&
+	    lxsnp->lxsys_instance == LXSYS_INST_ROOT) {
+		return (lxsnp->lxsys_ino);
+	} else {
+		return (VTOLXS(lxsnp->lxsys_parent)->lxsys_ino);
+	}
+}
+
+/*
+ * Allocate a new lxsys node
+ *
+ * This also allocates the vnode associated with it
+ */
+lxsys_node_t *
+lxsys_getnode(vnode_t *dp, lxsys_nodetype_t type, unsigned int instance,
+    unsigned int endpoint)
+{
+	lxsys_node_t *lxsnp;
+	vnode_t *vp;
+	timestruc_t now;
+
+	/*
+	 * Allocate a new node. It is deallocated in vop_innactive
+	 */
+	lxsnp = kmem_cache_alloc(lxsys_node_cache, KM_SLEEP);
+
+	/*
+	 * Set defaults (may be overridden below)
+	 */
+	gethrestime(&now);
+	lxsnp->lxsys_type = type;
+	lxsnp->lxsys_instance = instance;
+	lxsnp->lxsys_endpoint = endpoint;
+	lxsnp->lxsys_next = NULL;
+	lxsnp->lxsys_realvp = NULL;
+	lxsnp->lxsys_parent = dp;
+	VN_HOLD(dp);
+
+	lxsnp->lxsys_time = now;
+	lxsnp->lxsys_uid = lxsnp->lxsys_gid = 0;
+	lxsnp->lxsys_ino = lxsys_inode(type, instance, endpoint);
+
+	/* initialize the vnode data */
+	vp = lxsnp->lxsys_vnode;
+	vn_reinit(vp);
+	vp->v_flag = VNOCACHE|VNOMAP|VNOSWAP|VNOMOUNT;
+	vp->v_vfsp = dp->v_vfsp;
+
+	/*
+	 * Default to a directory with open permissions.
+	 * Specific components will override this
+	 */
+	if (type == LXSYS_STATIC && instance == LXSYS_INST_ROOT) {
+		vp->v_flag |= VROOT;
+	}
+	vp->v_type = VDIR;
+	lxsnp->lxsys_mode = 0555;
+
+	return (lxsnp);
+}
+
+lxsys_node_t *
+lxsys_getnode_static(vnode_t *dp, unsigned int instance)
+{
+	lxsys_mnt_t *lxsm = VTOLXSM(dp);
+	lxsys_node_t *lnp;
+
+	mutex_enter(&lxsm->lxsysm_lock);
+	lnp = lxsm->lxsysm_node;
+	while (1) {
+		if (lnp->lxsys_instance == instance) {
+			VERIFY(lnp->lxsys_parent == dp);
+
+			VN_HOLD(lnp->lxsys_vnode);
+			mutex_exit(&lxsm->lxsysm_lock);
+			return (lnp);
+		} else if (lnp->lxsys_next == NULL) {
+			break;
+		}
+		lnp = lnp->lxsys_next;
+	}
+
+	/*
+	 * No persistent node found.
+	 * Create one and add it to the end of the list.
+	 */
+	lnp->lxsys_next = lxsys_getnode(dp, LXSYS_STATIC, instance, 0);
+	lnp = lnp->lxsys_next;
+	/* Allow mounts on static entries */
+	LXSTOV(lnp)->v_flag &= (~VNOMOUNT);
+
+	mutex_exit(&lxsm->lxsysm_lock);
+	return (lnp);
+}
+
+/* Clean up persistence for static lxsys_node */
+int
+lxsys_freenode_static(lxsys_node_t *lnp)
+{
+	lxsys_node_t *plnp;
+	vnode_t *vp = LXSTOV(lnp);
+	lxsys_mnt_t *lxsm = VTOLXSM(vp);
+
+	if (lnp->lxsys_instance == LXSYS_INST_ROOT) {
+		/*
+		 * The root vnode does not need special cleanup since it
+		 * anchors the list and is freed by lxsys_unmount.
+		 */
+		return (0);
+	}
+
+	mutex_enter(&lxsm->lxsysm_lock);
+
+	/*
+	 * It is possible that a different process acquired a fresh reference
+	 * to this vnode via lookup while we were waiting on the lxsysm_lock.
+	 * To avoid freeing the vnode out from under them, we will double-check
+	 * v_count and bail from the fop_inactive if it was grabbed.
+	 */
+	mutex_enter(&vp->v_lock);
+	if (vp->v_count != 1) {
+		VERIFY(vp->v_count > 0);
+
+		mutex_exit(&vp->v_lock);
+		mutex_exit(&lxsm->lxsysm_lock);
+		return (-1);
+	}
+	mutex_exit(&vp->v_lock);
+
+	/* search for the record pointing to lnp */
+	plnp = lxsm->lxsysm_node;
+	while (plnp != NULL && plnp->lxsys_next != lnp) {
+		plnp = plnp->lxsys_next;
+	}
+	/* entry should always be found */
+	VERIFY(plnp != NULL);
+	plnp->lxsys_next = lnp->lxsys_next;
+
+	mutex_exit(&lxsm->lxsysm_lock);
+	return (0);
+}
+
+/*
+ * Free the storage obtained from lxsys_getnode().
+ */
+void
+lxsys_freenode(lxsys_node_t *lxsnp)
+{
+	vnode_t *vp = LXSTOV(lxsnp);
+
+	VERIFY(vp != NULL);
+
+	if (lxsnp->lxsys_type == LXSYS_STATIC) {
+		if (lxsys_freenode_static(lxsnp) != 0) {
+			return;
+		}
+	}
+
+	/*
+	 * delete any association with realvp
+	 */
+	if (lxsnp->lxsys_realvp != NULL)
+		VN_RELE(lxsnp->lxsys_realvp);
+
+	/*
+	 * delete any association with parent vp
+	 */
+	if (lxsnp->lxsys_parent != NULL)
+		VN_RELE(lxsnp->lxsys_parent);
+
+	/*
+	 * Release the lxsysnode.
+	 */
+	kmem_cache_free(lxsys_node_cache, lxsnp);
+}
+
+/*
+ * Get the netstack associated with this lxsys mount
+ */
+netstack_t *
+lxsys_netstack(lxsys_node_t *lnp)
+{
+	zone_t *zone = VTOLXSM(LXSTOV(lnp))->lxsysm_zone;
+	netstack_t *ns = zone->zone_netstack;
+
+	VERIFY(ns != NULL);
+
+	if (ns->netstack_flags & (NSF_UNINIT|NSF_CLOSING)) {
+		ns = NULL;
+	} else {
+		netstack_hold(ns);
+	}
+
+	return (ns);
+}
+
+ill_t *
+lxsys_find_ill(ip_stack_t *ipst, uint_t ifindex)
+{
+	ill_t *ill;
+	phyint_t *phyi;
+
+	rw_enter(&ipst->ips_ill_g_lock, RW_READER);
+	phyi = avl_find(&ipst->ips_phyint_g_list->phyint_list_avl_by_index,
+	    (void *) &ifindex, NULL);
+	if (phyi != NULL) {
+		/*
+		 * Since interface information presented via /sys is not
+		 * specific to IPv4 or IPv6, an ill reference from either
+		 * protocol will be adequate.  Check both, starting with IPv4
+		 * for a valid reference to use.
+		 */
+		for (ill = phyi->phyint_illv4; ill != phyi->phyint_illv6;
+		    ill = phyi->phyint_illv6) {
+			if (ill != NULL) {
+				mutex_enter(&ill->ill_lock);
+				if (!ILL_IS_CONDEMNED(ill)) {
+					ill_refhold_locked(ill);
+					mutex_exit(&ill->ill_lock);
+					rw_exit(&ipst->ips_ill_g_lock);
+					return (ill);
+				}
+				mutex_exit(&ill->ill_lock);
+			}
+		}
+	}
+	rw_exit(&ipst->ips_ill_g_lock);
+	return (NULL);
+}
+
+
+#define	LXSYSUIOBUFSZ	4096
+
+lxsys_uiobuf_t *
+lxsys_uiobuf_new(uio_t *uiop)
+{
+	/* Allocate memory for both lxsys_uiobuf and output buffer */
+	int bufsize = LXSYSUIOBUFSZ;
+	lxsys_uiobuf_t *uiobuf =
+	    kmem_alloc(sizeof (lxsys_uiobuf_t) + bufsize, KM_SLEEP);
+
+	uiobuf->uiop = uiop;
+	uiobuf->buffer = (char *)&uiobuf[1];
+	uiobuf->bufsize = bufsize;
+	uiobuf->pos = uiobuf->buffer;
+	uiobuf->beg = 0;
+	uiobuf->error = 0;
+
+	return (uiobuf);
+}
+
+void
+lxsys_uiobuf_free(lxsys_uiobuf_t *uiobuf)
+{
+	ASSERT(uiobuf != NULL);
+	ASSERT(uiobuf->pos == uiobuf->buffer);
+
+	kmem_free(uiobuf, sizeof (lxsys_uiobuf_t) + uiobuf->bufsize);
+}
+
+void
+lxsys_uiobuf_seterr(lxsys_uiobuf_t *uiobuf, int err)
+{
+	ASSERT(uiobuf->error == 0);
+
+	uiobuf->error = err;
+}
+
+int
+lxsys_uiobuf_flush(lxsys_uiobuf_t *uiobuf)
+{
+	off_t off = uiobuf->uiop->uio_offset;
+	caddr_t uaddr = uiobuf->buffer;
+	size_t beg = uiobuf->beg;
+	size_t size = (uintptr_t)uiobuf->pos - (uintptr_t)uaddr;
+
+	if (uiobuf->error == 0 && uiobuf->uiop->uio_resid != 0) {
+		ASSERT(off >= beg);
+
+		if (beg + size > off && off >= 0)
+			uiobuf->error =
+			    uiomove(uaddr + (off - beg), size - (off - beg),
+			    UIO_READ, uiobuf->uiop);
+
+		uiobuf->beg += size;
+	}
+
+	uiobuf->pos = uaddr;
+
+	return (uiobuf->error);
+}
+
+void
+lxsys_uiobuf_write(lxsys_uiobuf_t *uiobuf, const char *buf, size_t size)
+{
+	/* While we can still carry on */
+	while (uiobuf->error == 0 && uiobuf->uiop->uio_resid != 0) {
+		uintptr_t remain = (uintptr_t)uiobuf->bufsize -
+		    ((uintptr_t)uiobuf->pos - (uintptr_t)uiobuf->buffer);
+
+		/* Enough space in buffer? */
+		if (remain >= size) {
+			bcopy(buf, uiobuf->pos, size);
+			uiobuf->pos += size;
+			return;
+		}
+
+		/* Not enough space, so copy all we can and try again */
+		bcopy(buf, uiobuf->pos, remain);
+		uiobuf->pos += remain;
+		(void) lxsys_uiobuf_flush(uiobuf);
+		buf += remain;
+		size -= remain;
+	}
+}
+
+#define	TYPBUFFSIZE 256
+
+void
+lxsys_uiobuf_printf(lxsys_uiobuf_t *uiobuf, const char *fmt, ...)
+{
+	va_list args;
+	char buff[TYPBUFFSIZE];
+	int len;
+	char *buffer;
+
+	/* Can we still do any output */
+	if (uiobuf->error != 0 || uiobuf->uiop->uio_resid == 0)
+		return;
+
+	va_start(args, fmt);
+
+	/* Try using stack allocated buffer */
+	len = vsnprintf(buff, TYPBUFFSIZE, fmt, args);
+	if (len < TYPBUFFSIZE) {
+		va_end(args);
+		lxsys_uiobuf_write(uiobuf, buff, len);
+		return;
+	}
+
+	/* Not enough space in pre-allocated buffer */
+	buffer = kmem_alloc(len + 1, KM_SLEEP);
+
+	/*
+	 * We know we allocated the correct amount of space
+	 * so no check on the return value
+	 */
+	(void) vsnprintf(buffer, len+1, fmt, args);
+	lxsys_uiobuf_write(uiobuf, buffer, len);
+	va_end(args);
+	kmem_free(buffer, len+1);
+}
diff --git a/usr/src/uts/common/brand/lx/sysfs/lx_sysvfsops.c b/usr/src/uts/common/brand/lx/sysfs/lx_sysvfsops.c
new file mode 100644
index 0000000000..c33e61345e
--- /dev/null
+++ b/usr/src/uts/common/brand/lx/sysfs/lx_sysvfsops.c
@@ -0,0 +1,348 @@
+/*
+ * This file and its contents are supplied under the terms of the
+ * Common Development and Distribution License ("CDDL"), version 1.0.
+ * You may only use this file in accordance with the terms of version
+ * 1.0 of the CDDL.
+ *
+ * A full copy of the text of the CDDL should have accompanied this
+ * source.  A copy of the CDDL is also available via the Internet at
+ * http://www.illumos.org/license/CDDL.
+ */
+
+/*
+ * Copyright 2015 Joyent, Inc.
+ */
+
+/*
+ * lxsysvfsops.c: vfs operations for lx sysfs.
+ */
+
+#include <sys/types.h>
+#include <sys/param.h>
+#include <sys/cmn_err.h>
+#include <sys/cred.h>
+#include <sys/debug.h>
+#include <sys/errno.h>
+#include <sys/proc.h>
+#include <sys/stat.h>
+#include <sys/statvfs.h>
+#include <sys/sysmacros.h>
+#include <sys/systm.h>
+#include <sys/var.h>
+#include <sys/vfs.h>
+#include <sys/vfs_opreg.h>
+#include <sys/vnode.h>
+#include <sys/mode.h>
+#include <sys/signal.h>
+#include <sys/user.h>
+#include <sys/mount.h>
+#include <sys/bitmap.h>
+#include <sys/kmem.h>
+#include <sys/policy.h>
+#include <sys/modctl.h>
+#include <sys/sunddi.h>
+#include <sys/sunldi.h>
+#include <sys/lx_impl.h>
+
+#include "lx_sysfs.h"
+
+/* Module level parameters */
+static int	lxsysfstype;
+static dev_t	lxsysdev;
+static kmutex_t	lxsys_mount_lock;
+
+static int lxsys_mount(vfs_t *, vnode_t *, mounta_t *, cred_t *);
+static int lxsys_unmount(vfs_t *, int, cred_t *);
+static int lxsys_root(vfs_t *, vnode_t **);
+static int lxsys_statvfs(vfs_t *, statvfs64_t *);
+static int lxsys_init(int, char *);
+
+static vfsdef_t vfw = {
+	VFSDEF_VERSION,
+	"lx_sysfs",
+	lxsys_init,
+	VSW_ZMOUNT,
+	NULL
+};
+
+/*
+ * Module linkage information for the kernel.
+ */
+extern struct mod_ops mod_fsops;
+
+static struct modlfs modlfs = {
+	&mod_fsops, "lx brand sysfs", &vfw
+};
+
+static struct modlinkage modlinkage = {
+	MODREV_1, (void *)&modlfs, NULL
+};
+
+int
+_init(void)
+{
+	return (mod_install(&modlinkage));
+}
+
+int
+_info(struct modinfo *modinfop)
+{
+	return (mod_info(&modlinkage, modinfop));
+}
+
+int
+_fini(void)
+{
+	int retval;
+
+	/*
+	 * attempt to unload the module
+	 */
+	if ((retval = mod_remove(&modlinkage)) != 0)
+		goto done;
+
+	/*
+	 * destroy lxsys_node cache
+	 */
+	lxsys_fininodecache();
+
+	/*
+	 * clean out the vfsops and vnodeops
+	 */
+	(void) vfs_freevfsops_by_type(lxsysfstype);
+	vn_freevnodeops(lxsys_vnodeops);
+
+	mutex_destroy(&lxsys_mount_lock);
+done:
+	return (retval);
+}
+
+static int
+lxsys_init(int fstype, char *name)
+{
+	static const fs_operation_def_t lxsys_vfsops_template[] = {
+		VFSNAME_MOUNT,		{ .vfs_mount = lxsys_mount },
+		VFSNAME_UNMOUNT,	{ .vfs_unmount = lxsys_unmount },
+		VFSNAME_ROOT,		{ .vfs_root = lxsys_root },
+		VFSNAME_STATVFS,	{ .vfs_statvfs = lxsys_statvfs },
+		NULL,			NULL
+	};
+	extern const fs_operation_def_t lxsys_vnodeops_template[];
+	int error;
+	major_t dev;
+
+	lxsysfstype = fstype;
+	ASSERT(lxsysfstype != 0);
+
+	mutex_init(&lxsys_mount_lock, NULL, MUTEX_DEFAULT, NULL);
+
+	/*
+	 * Associate VFS ops vector with this fstype.
+	 */
+	error = vfs_setfsops(fstype, lxsys_vfsops_template, NULL);
+	if (error != 0) {
+		cmn_err(CE_WARN, "lxsys_init: bad vfs ops template");
+		return (error);
+	}
+
+	/*
+	 * Set up vnode ops vector too.
+	 */
+	error = vn_make_ops(name, lxsys_vnodeops_template, &lxsys_vnodeops);
+	if (error != 0) {
+		(void) vfs_freevfsops_by_type(fstype);
+		cmn_err(CE_WARN, "lxsys_init: bad vnode ops template");
+		return (error);
+	}
+
+	/*
+	 * Assign a unique "device" number (used by stat(2)).
+	 */
+	if ((dev = getudev()) == (major_t)-1) {
+		cmn_err(CE_WARN, "lxsys_init: can't get unique device number");
+		dev = 0;
+	}
+
+	/*
+	 * Make the pseudo device
+	 */
+	lxsysdev = makedevice(dev, 0);
+
+	/*
+	 * Initialise cache for lxsys_nodes
+	 */
+	lxsys_initnodecache();
+
+	return (0);
+}
+
+static int
+lxsys_mount(vfs_t *vfsp, vnode_t *mvp, mounta_t *uap, cred_t *cr)
+{
+	lxsys_mnt_t *lxsys_mnt;
+	zone_t *zone = curproc->p_zone;
+
+	/*
+	 * must be root to mount
+	 */
+	if (secpolicy_fs_mount(cr, mvp, vfsp) != 0)
+		return (EPERM);
+
+	/*
+	 * mount point must be a directory
+	 */
+	if (mvp->v_type != VDIR)
+		return (ENOTDIR);
+
+	if (zone == global_zone) {
+		zone_t *mntzone;
+
+		mntzone = zone_find_by_path(refstr_value(vfsp->vfs_mntpt));
+		zone_rele(mntzone);
+		if (zone != mntzone)
+			return (EBUSY);
+	}
+
+	/*
+	 * Having the resource be anything but "lxsys" doesn't make sense
+	 */
+	vfs_setresource(vfsp, "lxsys", 0);
+
+	lxsys_mnt = kmem_alloc(sizeof (*lxsys_mnt), KM_SLEEP);
+
+	mutex_enter(&lxsys_mount_lock);
+
+	/*
+	 * Ensure we don't allow overlaying mounts
+	 */
+	mutex_enter(&mvp->v_lock);
+	if ((uap->flags & MS_OVERLAY) == 0 &&
+	    (mvp->v_count > 1 || (mvp->v_flag & VROOT))) {
+		mutex_exit(&mvp->v_lock);
+		mutex_exit(&lxsys_mount_lock);
+		kmem_free(lxsys_mnt, sizeof ((*lxsys_mnt)));
+		return (EBUSY);
+	}
+	mutex_exit(&mvp->v_lock);
+
+
+	mutex_init(&lxsys_mnt->lxsysm_lock, NULL, MUTEX_DEFAULT, NULL);
+	zone_hold(lxsys_mnt->lxsysm_zone = zone);
+
+	/* Arbitrarily set the parent vnode to the mounted over directory */
+	lxsys_mnt->lxsysm_node = lxsys_getnode(mvp, LXSYS_STATIC,
+	    LXSYS_INST_ROOT, 0);
+	lxsys_mnt->lxsysm_node->lxsys_next = NULL;
+
+	/* Correctly set the fs for the root node */
+	lxsys_mnt->lxsysm_node->lxsys_vnode->v_vfsp = vfsp;
+
+	vfs_make_fsid(&vfsp->vfs_fsid, lxsysdev, lxsysfstype);
+	vfsp->vfs_bsize = DEV_BSIZE;
+	vfsp->vfs_fstype = lxsysfstype;
+	vfsp->vfs_data = (caddr_t)lxsys_mnt;
+	vfsp->vfs_dev = lxsysdev;
+
+	mutex_exit(&lxsys_mount_lock);
+
+	return (0);
+}
+
+static int
+lxsys_unmount(vfs_t *vfsp, int flag, cred_t *cr)
+{
+	lxsys_mnt_t *lxsys_mnt = (lxsys_mnt_t *)vfsp->vfs_data;
+	lxsys_node_t *lnp;
+	vnode_t *vp;
+	int count;
+
+	VERIFY(lxsys_mnt != NULL);
+
+	mutex_enter(&lxsys_mount_lock);
+
+	/* must be root to unmount */
+	if (secpolicy_fs_unmount(cr, vfsp) != 0) {
+		mutex_exit(&lxsys_mount_lock);
+		return (EPERM);
+	}
+
+	/* forced unmount is not supported by this fs */
+	if (flag & MS_FORCE) {
+		mutex_exit(&lxsys_mount_lock);
+		return (ENOTSUP);
+	}
+
+	/* Ensure that no vnodes are in use on this mount point. */
+	lnp = lxsys_mnt->lxsysm_node;
+	vp = LXSTOV(lnp);
+	mutex_enter(&vp->v_lock);
+	count = vp->v_count;
+	mutex_exit(&vp->v_lock);
+	if (count > 1) {
+		mutex_exit(&lxsys_mount_lock);
+		return (EBUSY);
+	}
+
+	/*
+	 * If there are no references to the root vnode the list of persistent
+	 * static vnodes should be empty
+	 */
+	VERIFY(lnp->lxsys_next == NULL);
+
+	(void) dnlc_purge_vfsp(vfsp, 0);
+
+	lxsys_mnt->lxsysm_node = NULL;
+	lxsys_freenode(lnp);
+	zone_rele(lxsys_mnt->lxsysm_zone);
+	vfsp->vfs_data = NULL;
+	kmem_free(lxsys_mnt, sizeof (*lxsys_mnt));
+
+	mutex_exit(&lxsys_mount_lock);
+
+	return (0);
+}
+
+static int
+lxsys_root(vfs_t *vfsp, vnode_t **vpp)
+{
+	lxsys_mnt_t *lxsm = (lxsys_mnt_t *)vfsp->vfs_data;
+	vnode_t *vp;
+
+	VERIFY(lxsm != NULL);
+	VERIFY(lxsm->lxsysm_node != NULL);
+
+	vp = LXSTOV(lxsm->lxsysm_node);
+	VN_HOLD(vp);
+	*vpp = vp;
+
+	return (0);
+}
+
+static int
+lxsys_statvfs(vfs_t *vfsp, statvfs64_t *sp)
+{
+	dev32_t d32;
+
+	bzero((caddr_t)sp, sizeof (*sp));
+	sp->f_bsize	= DEV_BSIZE;
+	sp->f_frsize	= DEV_BSIZE;
+	sp->f_blocks	= (fsblkcnt64_t)0;
+	sp->f_bfree	= (fsblkcnt64_t)0;
+	sp->f_bavail	= (fsblkcnt64_t)0;
+	sp->f_files	= (fsfilcnt64_t)3;
+	sp->f_ffree	= (fsfilcnt64_t)0;	/* none */
+	sp->f_favail	= (fsfilcnt64_t)0;	/* none */
+	(void) cmpldev(&d32, vfsp->vfs_dev);
+	sp->f_fsid	= d32;
+	/* It is guaranteed that vsw_name will fit in f_basetype */
+	(void) strcpy(sp->f_basetype, vfssw[lxsysfstype].vsw_name);
+	sp->f_flag = vf_to_stf(vfsp->vfs_flag);
+	sp->f_namemax = 64;		/* quite arbitrary */
+	bzero(sp->f_fstr, sizeof (sp->f_fstr));
+
+	/* We know f_fstr is 32 chars */
+	(void) strcpy(sp->f_fstr, "/sys");
+	(void) strcpy(&sp->f_fstr[6], "/sys");
+
+	return (0);
+}
diff --git a/usr/src/uts/common/brand/lx/sysfs/lx_sysvnops.c b/usr/src/uts/common/brand/lx/sysfs/lx_sysvnops.c
new file mode 100644
index 0000000000..0383904f83
--- /dev/null
+++ b/usr/src/uts/common/brand/lx/sysfs/lx_sysvnops.c
@@ -0,0 +1,1182 @@
+/*
+ * This file and its contents are supplied under the terms of the
+ * Common Development and Distribution License ("CDDL"), version 1.0.
+ * You may only use this file in accordance with the terms of version
+ * 1.0 of the CDDL.
+ *
+ * A full copy of the text of the CDDL should have accompanied this
+ * source.  A copy of the CDDL is also available via the Internet at
+ * http://www.illumos.org/license/CDDL.
+ */
+
+/*
+ * Copyright 2015 Joyent, Inc.
+ */
+
+/*
+ * lx_sysfs -- a Linux-compatible /sys for the LX brand
+ */
+
+#include <vm/seg_vn.h>
+#include <sys/sdt.h>
+#include <sys/strlog.h>
+#include <sys/stropts.h>
+#include <sys/cmn_err.h>
+#include <sys/lx_brand.h>
+#include <sys/x86_archext.h>
+#include <sys/archsystm.h>
+#include <sys/fp.h>
+#include <sys/pool_pset.h>
+#include <sys/pset.h>
+#include <sys/zone.h>
+#include <sys/pghw.h>
+#include <sys/vfs_opreg.h>
+#include <sys/param.h>
+#include <sys/utsname.h>
+#include <sys/lx_misc.h>
+#include <sys/brand.h>
+#include <sys/cred_impl.h>
+#include <sys/tihdr.h>
+#include <sys/sunddi.h>
+#include <sys/vnode.h>
+#include <sys/netstack.h>
+#include <sys/ethernet.h>
+#include <inet/ip_arp.h>
+
+#include "lx_sysfs.h"
+
+/*
+ * Pointer to the vnode ops vector for this fs.
+ * This is instantiated in lxsys_init() in lx_sysvfsops.c
+ */
+vnodeops_t *lxsys_vnodeops;
+
+static int lxsys_open(vnode_t **, int, cred_t *, caller_context_t *);
+static int lxsys_close(vnode_t *, int, int, offset_t, cred_t *,
+    caller_context_t *);
+static int lxsys_read(vnode_t *, uio_t *, int, cred_t *, caller_context_t *);
+static int lxsys_getattr(vnode_t *, vattr_t *, int, cred_t *,
+    caller_context_t *);
+static int lxsys_access(vnode_t *, int, int, cred_t *, caller_context_t *);
+static int lxsys_lookup(vnode_t *, char *, vnode_t **,
+    pathname_t *, int, vnode_t *, cred_t *, caller_context_t *, int *,
+    pathname_t *);
+static int lxsys_readdir(vnode_t *, uio_t *, cred_t *, int *,
+    caller_context_t *, int);
+static int lxsys_readlink(vnode_t *, uio_t *, cred_t *, caller_context_t *);
+static int lxsys_cmp(vnode_t *, vnode_t *, caller_context_t *);
+static int lxsys_realvp(vnode_t *, vnode_t **, caller_context_t *);
+static int lxsys_sync(void);
+static void lxsys_inactive(vnode_t *, cred_t *, caller_context_t *);
+
+static vnode_t *lxsys_lookup_static(lxsys_node_t *, char *);
+static vnode_t *lxsys_lookup_class_netdir(lxsys_node_t *, char *);
+static vnode_t *lxsys_lookup_devices_virtual_netdir(lxsys_node_t *, char *);
+
+static int lxsys_read_devices_virtual_net(lxsys_node_t *, lxsys_uiobuf_t *);
+
+static int lxsys_readdir_static(lxsys_node_t *, uio_t *, int *);
+static int lxsys_readdir_class_netdir(lxsys_node_t *, uio_t *, int *);
+static int lxsys_readdir_devices_virtual_netdir(lxsys_node_t *, uio_t *, int *);
+
+static int lxsys_readlink_class_net(lxsys_node_t *lnp, char *buf, size_t len);
+
+/*
+ * The lx /sys vnode operations vector
+ */
+const fs_operation_def_t lxsys_vnodeops_template[] = {
+	VOPNAME_OPEN,		{ .vop_open = lxsys_open },
+	VOPNAME_CLOSE,		{ .vop_close = lxsys_close },
+	VOPNAME_READ,		{ .vop_read = lxsys_read },
+	VOPNAME_GETATTR,	{ .vop_getattr = lxsys_getattr },
+	VOPNAME_ACCESS,		{ .vop_access = lxsys_access },
+	VOPNAME_LOOKUP,		{ .vop_lookup = lxsys_lookup },
+	VOPNAME_READDIR,	{ .vop_readdir = lxsys_readdir },
+	VOPNAME_READLINK,	{ .vop_readlink = lxsys_readlink },
+	VOPNAME_FSYNC,		{ .error = lxsys_sync },
+	VOPNAME_SEEK,		{ .error = lxsys_sync },
+	VOPNAME_INACTIVE,	{ .vop_inactive = lxsys_inactive },
+	VOPNAME_CMP,		{ .vop_cmp = lxsys_cmp },
+	VOPNAME_REALVP,		{ .vop_realvp = lxsys_realvp },
+	NULL,			NULL
+};
+
+/*
+ * Sysfs Inode format:
+ * 0000AABBBBCC
+ *
+ * AA - TYPE
+ * BBBB - INSTANCE
+ * CC - ENDPOINT
+ *
+ * Where TYPE is one of:
+ * 1 - SYS_STATIC
+ * 2 - SYS_CLASS_NET
+ * 3 - SYS_DEVICES_NET
+ *
+ * Static entries will have assigned INSTANCE identifiers:
+ * - 0: /sys
+ * - 1: /sys/class
+ * - 2: /sys/devices
+ * - 3: /sys/fs
+ * - 4: /sys/class/net
+ * - 5: /sys/devices/virtual
+ * - 7: /sys/devices/system
+ * - 8: /sys/fs/cgroup
+ * - 9: /sys/devices/virtual/net
+ *
+ * Dynamic /sys/class/net/<interface> symlinks will use an INSTANCE derived
+ * from the corresonding ifindex.
+ *
+ * Dynamic /sys/devices/virtual/net/<interface>/<entries> directories will use
+ * an INSTANCE derived from the ifindex and statically assigned ENDPOINT IDs
+ * for the contained entries.
+ */
+
+#define	LXSYS_INST_CLASSDIR			1
+#define	LXSYS_INST_DEVICESDIR			2
+#define	LXSYS_INST_FSDIR			3
+#define	LXSYS_INST_CLASS_NETDIR			4
+#define	LXSYS_INST_DEVICES_VIRTUALDIR		5
+#define	LXSYS_INST_DEVICES_SYSTEMDIR		6
+#define	LXSYS_INST_FS_CGROUPDIR			7
+#define	LXSYS_INST_DEVICES_VIRTUAL_NETDIR	8
+
+/*
+ * file contents of an lx /sys directory.
+ */
+static lxsys_dirent_t dirlist_root[] = {
+	{ LXSYS_INST_CLASSDIR,		"class" },
+	{ LXSYS_INST_DEVICESDIR,	"devices" },
+	{ LXSYS_INST_FSDIR,		"fs" }
+};
+static lxsys_dirent_t dirlist_empty[] = {};
+static lxsys_dirent_t dirlist_class[] = {
+	{ LXSYS_INST_CLASS_NETDIR,	"net" }
+};
+static lxsys_dirent_t dirlist_fs[] = {
+	{ LXSYS_INST_FS_CGROUPDIR,	"cgroup" }
+};
+static lxsys_dirent_t dirlist_devices[] = {
+	{ LXSYS_INST_DEVICES_SYSTEMDIR,		"system" },
+	{ LXSYS_INST_DEVICES_VIRTUALDIR,	"virtual" }
+};
+static lxsys_dirent_t dirlist_devices_virtual[] = {
+	{ LXSYS_INST_DEVICES_VIRTUAL_NETDIR,	"net" }
+};
+
+
+#define	LXSYS_ENDP_NET_ADDRESS	1
+#define	LXSYS_ENDP_NET_ADDRLEN	2
+#define	LXSYS_ENDP_NET_FLAGS	3
+#define	LXSYS_ENDP_NET_IFINDEX	4
+#define	LXSYS_ENDP_NET_MTU	5
+#define	LXSYS_ENDP_NET_TXQLEN	6
+#define	LXSYS_ENDP_NET_TYPE	7
+
+static lxsys_dirent_t dirlist_devices_virtual_net[] = {
+	{ LXSYS_ENDP_NET_ADDRESS,	"address" },
+	{ LXSYS_ENDP_NET_ADDRLEN,	"addr_len" },
+	{ LXSYS_ENDP_NET_FLAGS,		"flags" },
+	{ LXSYS_ENDP_NET_IFINDEX,	"ifindex" },
+	{ LXSYS_ENDP_NET_MTU,		"mtu" },
+	{ LXSYS_ENDP_NET_TXQLEN,	"tx_queue_len" },
+	{ LXSYS_ENDP_NET_TYPE,		"type" }
+};
+
+#define	SYSDIRLISTSZ(l)	(sizeof (l) / sizeof ((l)[0]))
+
+#define	SYSDLENT(i, l)	{ i, l, SYSDIRLISTSZ(l) }
+static lxsys_dirlookup_t lxsys_dirlookup[] = {
+	SYSDLENT(LXSYS_INST_ROOT, dirlist_root),
+	SYSDLENT(LXSYS_INST_CLASSDIR, dirlist_class),
+	SYSDLENT(LXSYS_INST_FSDIR, dirlist_fs),
+	SYSDLENT(LXSYS_INST_FS_CGROUPDIR, dirlist_empty),
+	SYSDLENT(LXSYS_INST_DEVICESDIR, dirlist_devices),
+	SYSDLENT(LXSYS_INST_DEVICES_VIRTUALDIR, dirlist_devices_virtual)
+};
+
+
+/*
+ * Array of lookup functions, indexed by lx /sys file type.
+ */
+static vnode_t *(*lxsys_lookup_function[LXSYS_MAXTYPE])() = {
+	NULL,					/* LXSYS_NONE		*/
+	lxsys_lookup_static,			/* LXSYS_STATIC		*/
+	lxsys_lookup_class_netdir,		/* LXSYS_CLASS_NET	*/
+	lxsys_lookup_devices_virtual_netdir,	/* LXSYS_DEVICES_NET	*/
+};
+
+/*
+ * Array of readdir functions, indexed by /sys file type.
+ */
+static int (*lxsys_readdir_function[LXSYS_MAXTYPE])() = {
+	NULL,					/* LXSYS_NONE		*/
+	lxsys_readdir_static,			/* LXSYS_STATIC		*/
+	lxsys_readdir_class_netdir,		/* LXSYS_CLASS_NET	*/
+	lxsys_readdir_devices_virtual_netdir,	/* LXSYS_DEVICES_NET	*/
+};
+
+/*
+ * Array of read functions, indexed by /sys file type.
+ */
+static int (*lxsys_read_function[LXSYS_MAXTYPE])() = {
+	NULL,					/* LXSYS_NONE		*/
+	NULL,					/* LXSYS_STATIC		*/
+	NULL,					/* LXSYS_CLASS_NET	*/
+	lxsys_read_devices_virtual_net,		/* LXSYS_DEVICES_NET	*/
+};
+
+/*
+ * Array of readlink functions, indexed by /sys file type.
+ */
+static int (*lxsys_readlink_function[LXSYS_MAXTYPE])() = {
+	NULL,					/* LXSYS_NONE		*/
+	NULL,					/* LXSYS_STATIC		*/
+	lxsys_readlink_class_net,		/* LXSYS_CLASS_NET	*/
+	NULL,					/* LXSYS_DEVICES_NET	*/
+};
+
+
+
+/*
+ * lxsys_open(): Vnode operation for VOP_OPEN()
+ */
+static int
+lxsys_open(vnode_t **vpp, int flag, cred_t *cr, caller_context_t *ct)
+{
+	vnode_t		*vp = *vpp;
+	lxsys_node_t	*lxsnp = VTOLXS(vp);
+	vnode_t		*rvp;
+	int		error = 0;
+
+	/*
+	 * We only allow reading in this file system
+	 */
+	if (flag & FWRITE)
+		return (EROFS);
+
+	/*
+	 * If we are opening an underlying file only allow regular files,
+	 * reject the open for anything else.
+	 * Just do it if we are opening the current or root directory.
+	 */
+	if (lxsnp->lxsys_realvp != NULL) {
+		rvp = lxsnp->lxsys_realvp;
+
+		/*
+		 * Need to hold rvp since VOP_OPEN() may release it.
+		 */
+		VN_HOLD(rvp);
+		error = VOP_OPEN(&rvp, flag, cr, ct);
+		if (error) {
+			VN_RELE(rvp);
+		} else {
+			*vpp = rvp;
+			VN_RELE(vp);
+		}
+	}
+
+	return (error);
+}
+
+
+/*
+ * lxsys_close(): Vnode operation for VOP_CLOSE()
+ */
+/* ARGSUSED */
+static int
+lxsys_close(vnode_t *vp, int flag, int count, offset_t offset, cred_t *cr,
+    caller_context_t *ct)
+{
+	return (0);
+}
+
+
+/*
+ * lxsys_read(): Vnode operation for VOP_READ()
+ * All we currently have in this fs are directories.
+ */
+/* ARGSUSED */
+static int
+lxsys_read(vnode_t *vp, uio_t *uiop, int ioflag, cred_t *cr,
+    caller_context_t *ct)
+{
+	lxsys_node_t *lnp = VTOLXS(vp);
+	lxsys_nodetype_t type = lnp->lxsys_type;
+	int (*rlfunc)();
+	int error;
+	lxsys_uiobuf_t *luio;
+
+	VERIFY(type > LXSYS_NONE && type < LXSYS_MAXTYPE);
+
+	if (vp->v_type == VDIR) {
+		return (EISDIR);
+	}
+
+	rlfunc = lxsys_read_function[type];
+	if (rlfunc != NULL) {
+		luio = lxsys_uiobuf_new(uiop);
+		if ((error = rlfunc(lnp, luio)) == 0) {
+			error = lxsys_uiobuf_flush(luio);
+		}
+		lxsys_uiobuf_free(luio);
+	} else {
+		error = EIO;
+	}
+
+	return (error);
+}
+
+/*
+ * lxsys_getattr(): Vnode operation for VOP_GETATTR()
+ */
+static int
+lxsys_getattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr,
+    caller_context_t *ct)
+{
+	register lxsys_node_t *lxsnp = VTOLXS(vp);
+	int error;
+
+	/*
+	 * Return attributes of underlying vnode if ATTR_REAL
+	 *
+	 * but keep fd files with the symlink permissions
+	 */
+	if (lxsnp->lxsys_realvp != NULL && (flags & ATTR_REAL)) {
+		vnode_t *rvp = lxsnp->lxsys_realvp;
+
+		/*
+		 * limit attribute information to owner or root
+		 */
+		if ((error = VOP_ACCESS(rvp, 0, 0, cr, ct)) != 0) {
+			return (error);
+		}
+
+		/*
+		 * now its attributes
+		 */
+		if ((error = VOP_GETATTR(rvp, vap, flags, cr, ct)) != 0) {
+			return (error);
+		}
+
+		return (0);
+	}
+
+	/* Default attributes, that may be overridden below */
+	bzero(vap, sizeof (*vap));
+	vap->va_atime = vap->va_mtime = vap->va_ctime = lxsnp->lxsys_time;
+	vap->va_nlink = 1;
+	vap->va_type = vp->v_type;
+	vap->va_mode = lxsnp->lxsys_mode;
+	vap->va_fsid = vp->v_vfsp->vfs_dev;
+	vap->va_blksize = DEV_BSIZE;
+	vap->va_uid = lxsnp->lxsys_uid;
+	vap->va_gid = lxsnp->lxsys_gid;
+	vap->va_nodeid = lxsnp->lxsys_ino;
+
+	vap->va_nblocks = (fsblkcnt64_t)btod(vap->va_size);
+	return (0);
+}
+
+/*
+ * lxsys_access(): Vnode operation for VOP_ACCESS()
+ */
+static int
+lxsys_access(vnode_t *vp, int mode, int flags, cred_t *cr, caller_context_t *ct)
+{
+	lxsys_node_t *lxsnp = VTOLXS(vp);
+	int shift = 0;
+
+	/*
+	 * Although our lx sysfs is basically a read only file system, Linux
+	 * expects it to be writable so we can't just error if (mode & VWRITE).
+	 */
+
+	if (lxsnp->lxsys_realvp != NULL) {
+		/*
+		 * For these we use the underlying vnode's accessibility.
+		 */
+		return (VOP_ACCESS(lxsnp->lxsys_realvp, mode, flags, cr, ct));
+	}
+
+	/* If user is root allow access regardless of permission bits */
+	if (secpolicy_proc_access(cr) == 0)
+		return (0);
+
+	/*
+	 * Access check is based on only one of owner, group, public.  If not
+	 * owner, then check group.  If not a member of the group, then check
+	 * public access.
+	 */
+	if (crgetuid(cr) != lxsnp->lxsys_uid) {
+		shift += 3;
+		if (!groupmember((uid_t)lxsnp->lxsys_gid, cr))
+			shift += 3;
+	}
+
+	mode &= ~(lxsnp->lxsys_mode << shift);
+
+	if (mode == 0)
+		return (0);
+
+	return (EACCES);
+}
+
+/*
+ * lxsys_lookup(): Vnode operation for VOP_LOOKUP()
+ */
+/* ARGSUSED */
+static int
+lxsys_lookup(vnode_t *dp, char *comp, vnode_t **vpp, pathname_t *pathp,
+    int flags, vnode_t *rdir, cred_t *cr, caller_context_t *ct,
+    int *direntflags, pathname_t *realpnp)
+{
+	lxsys_node_t *lxsnp = VTOLXS(dp);
+	lxsys_nodetype_t type = lxsnp->lxsys_type;
+	int error;
+
+	VERIFY(dp->v_type == VDIR);
+	VERIFY(type > LXSYS_NONE && type < LXSYS_MAXTYPE);
+
+	/*
+	 * restrict lookup permission to owner or root
+	 */
+	if ((error = lxsys_access(dp, VEXEC, 0, cr, ct)) != 0) {
+		return (error);
+	}
+
+	/*
+	 * Just return the parent vnode if that's where we are trying to go.
+	 */
+	if (strcmp(comp, "..") == 0) {
+		VN_HOLD(lxsnp->lxsys_parent);
+		*vpp = lxsnp->lxsys_parent;
+		return (0);
+	}
+
+	/*
+	 * Special handling for directory searches.  Note: null component name
+	 * denotes that the current directory is being searched.
+	 */
+	if ((dp->v_type == VDIR) && (*comp == '\0' || strcmp(comp, ".") == 0)) {
+		VN_HOLD(dp);
+		*vpp = dp;
+		return (0);
+	}
+
+	*vpp = (lxsys_lookup_function[type](lxsnp, comp));
+	return ((*vpp == NULL) ? ENOENT : 0);
+}
+
+static vnode_t *
+lxsys_lookup_static(lxsys_node_t *ldp, char *comp)
+{
+	lxsys_dirent_t *dirent = NULL;
+	int i, len = 0;
+
+	for (i = 0; i < SYSDIRLISTSZ(lxsys_dirlookup); i++) {
+		if (ldp->lxsys_instance == lxsys_dirlookup[i].dl_instance) {
+			dirent = lxsys_dirlookup[i].dl_list;
+			len = lxsys_dirlookup[i].dl_length;
+			break;
+		}
+	}
+	if (dirent == NULL) {
+		return (NULL);
+	}
+
+	for (i = 0; i < len; i++) {
+		if (strncmp(comp, dirent[i].d_name, MAXPATHLEN) == 0) {
+			lxsys_nodetype_t node_type = ldp->lxsys_type;
+			unsigned int node_instance = 0;
+			lxsys_node_t *lnp;
+
+			switch (dirent[i].d_idnum) {
+			case LXSYS_INST_CLASS_NETDIR:
+				node_type = LXSYS_CLASS_NET;
+				break;
+			case LXSYS_INST_DEVICES_VIRTUAL_NETDIR:
+				node_type = LXSYS_DEVICES_NET;
+				break;
+			default:
+				/* Another static node */
+				node_instance = dirent[i].d_idnum;
+			}
+			if (node_type == LXSYS_STATIC) {
+				lnp = lxsys_getnode_static(ldp->lxsys_vnode,
+				    node_instance);
+			} else {
+				lnp = lxsys_getnode(ldp->lxsys_vnode,
+				    node_type, node_instance, 0);
+			}
+			return (lnp->lxsys_vnode);
+		}
+	}
+	return (NULL);
+}
+
+static vnode_t *
+lxsys_lookup_class_netdir(lxsys_node_t *ldp, char *comp)
+{
+	vnode_t *result = NULL;
+	lxsys_node_t *lnp;
+	netstack_t *ns;
+	ip_stack_t *ipst;
+	avl_tree_t *phytree;
+	phyint_t *phyi;
+	char ifname[LIFNAMSIZ];
+
+	if (ldp->lxsys_type != LXSYS_CLASS_NET ||
+	    ldp->lxsys_instance != 0) {
+		/* Lookups only allowed at directory level */
+		return (NULL);
+	}
+
+	(void) strncpy(ifname, comp, LIFNAMSIZ);
+	lx_ifname_convert(ifname, LX_IF_TONATIVE);
+
+	if ((ns = lxsys_netstack(ldp)) == NULL) {
+		return (NULL);
+	}
+	ipst = ns->netstack_ip;
+	rw_enter(&ipst->ips_ill_g_lock, RW_READER);
+
+	phytree = &ipst->ips_phyint_g_list->phyint_list_avl_by_name;
+	phyi = avl_find(phytree, ifname, NULL);
+	if (phyi != NULL) {
+		lnp = lxsys_getnode(ldp->lxsys_vnode, ldp->lxsys_type,
+		    phyi->phyint_ifindex, 0);
+		result = lnp->lxsys_vnode;
+		result->v_type = VLNK;
+	}
+
+	rw_exit(&ipst->ips_ill_g_lock);
+	netstack_rele(ns);
+
+	return (result);
+}
+
+static vnode_t *
+lxsys_lookup_devices_virtual_netdir(lxsys_node_t *ldp, char *comp)
+{
+	lxsys_node_t *lnp;
+
+	if (ldp->lxsys_instance == 0) {
+		/* top-level interface listing */
+		vnode_t *result = NULL;
+		netstack_t *ns;
+		ip_stack_t *ipst;
+		avl_tree_t *phytree;
+		phyint_t *phyi;
+		char ifname[LIFNAMSIZ];
+
+		(void) strncpy(ifname, comp, LIFNAMSIZ);
+		lx_ifname_convert(ifname, LX_IF_TONATIVE);
+
+		if ((ns = lxsys_netstack(ldp)) == NULL) {
+			return (NULL);
+		}
+		ipst = ns->netstack_ip;
+		rw_enter(&ipst->ips_ill_g_lock, RW_READER);
+
+		phytree = &ipst->ips_phyint_g_list->phyint_list_avl_by_name;
+		phyi = avl_find(phytree, ifname, NULL);
+		if (phyi != NULL) {
+			lnp = lxsys_getnode(ldp->lxsys_vnode, ldp->lxsys_type,
+			    phyi->phyint_ifindex, 0);
+			result = lnp->lxsys_vnode;
+		}
+
+		rw_exit(&ipst->ips_ill_g_lock);
+		netstack_rele(ns);
+
+		return (result);
+	} else if (ldp->lxsys_endpoint == 0) {
+		/* interface-level sub-item listing */
+		int i, size;
+		lxsys_dirent_t *dirent;
+
+		size = SYSDIRLISTSZ(dirlist_devices_virtual_net);
+		for (i = 0; i < size; i++) {
+			dirent = &dirlist_devices_virtual_net[i];
+			if (strncmp(comp, dirent->d_name, LXSNSIZ) == 0) {
+				lnp = lxsys_getnode(ldp->lxsys_vnode,
+				    ldp->lxsys_type, ldp->lxsys_instance,
+				    dirent->d_idnum);
+				lnp->lxsys_vnode->v_type = VREG;
+				lnp->lxsys_mode = 0444;
+				return (lnp->lxsys_vnode);
+			}
+		}
+	}
+
+	return (NULL);
+}
+
+
+static int
+lxsys_read_devices_virtual_net(lxsys_node_t *lnp, lxsys_uiobuf_t *luio)
+{
+	netstack_t *ns;
+	ill_t *ill;
+	uint_t ifindex = lnp->lxsys_instance;
+	uint8_t *addr;
+	uint64_t flags;
+	int error = 0;
+
+	if (ifindex == 0 || lnp->lxsys_endpoint == 0) {
+		return (EISDIR);
+	}
+
+	if ((ns = lxsys_netstack(lnp)) == NULL) {
+		return (EIO);
+	}
+
+	ill = lxsys_find_ill(ns->netstack_ip, ifindex);
+	if (ill == NULL) {
+		netstack_rele(ns);
+		return (EIO);
+	}
+
+	switch (lnp->lxsys_endpoint) {
+	case LXSYS_ENDP_NET_ADDRESS:
+		if (ill->ill_phys_addr_length != ETHERADDRL) {
+			lxsys_uiobuf_printf(luio, "00:00:00:00:00:00\n");
+			break;
+		}
+		addr = ill->ill_phys_addr;
+		lxsys_uiobuf_printf(luio,
+		    "%02hhx:%02hhx:%02hhx:%02hhx:%02hhx:%02hhx\n",
+		    addr[0], addr[1], addr[2], addr[3], addr[4], addr[5]);
+		break;
+	case LXSYS_ENDP_NET_ADDRLEN:
+		lxsys_uiobuf_printf(luio, "%u\n",
+		    IS_LOOPBACK(ill) ? ETHERADDRL : ill->ill_phys_addr_length);
+		break;
+	case LXSYS_ENDP_NET_FLAGS:
+		flags = (ill->ill_flags | ill->ill_ipif->ipif_flags |
+		    ill->ill_phyint->phyint_flags) & 0xffff;
+		lx_ifflags_convert(&flags, LX_IF_FROMNATIVE);
+		lxsys_uiobuf_printf(luio, "0x%x\n", flags);
+		break;
+	case LXSYS_ENDP_NET_IFINDEX:
+		lxsys_uiobuf_printf(luio, "%u\n", ifindex);
+		break;
+	case LXSYS_ENDP_NET_MTU:
+		lxsys_uiobuf_printf(luio, "%u\n", ill->ill_mtu);
+		break;
+	case LXSYS_ENDP_NET_TXQLEN:
+		/* perpetuate the txqlen lie */
+		if (IS_LOOPBACK(ill)) {
+			lxsys_uiobuf_printf(luio, "0\n");
+		} else {
+			lxsys_uiobuf_printf(luio, "1\n");
+		}
+		break;
+	case LXSYS_ENDP_NET_TYPE:
+		lxsys_uiobuf_printf(luio, "%u\n",
+		    IS_LOOPBACK(ill) ? LX_ARPHRD_LOOPBACK :
+		    arp_hw_type(ill->ill_mactype));
+		break;
+	default:
+		error = EIO;
+	}
+
+	ill_refrele(ill);
+	netstack_rele(ns);
+	return (error);
+}
+
+/*
+ * lxsys_readdir(): Vnode operation for VOP_READDIR()
+ */
+/* ARGSUSED */
+static int
+lxsys_readdir(vnode_t *dp, uio_t *uiop, cred_t *cr, int *eofp,
+    caller_context_t *ct, int flags)
+{
+	lxsys_node_t *lxsnp = VTOLXS(dp);
+	lxsys_nodetype_t type = lxsnp->lxsys_type;
+	ssize_t uresid;
+	off_t uoffset;
+	int error, leof;
+
+	ASSERT(dp->v_type == VDIR);
+	VERIFY(type > LXSYS_NONE && type < LXSYS_MAXTYPE);
+
+	/*
+	 * restrict readdir permission to owner or root
+	 */
+	if ((error = lxsys_access(dp, VREAD, 0, cr, ct)) != 0)
+		return (error);
+
+	uoffset = uiop->uio_offset;
+	uresid = uiop->uio_resid;
+
+	/* can't do negative reads */
+	if (uoffset < 0 || uresid <= 0)
+		return (EINVAL);
+
+	/* can't read directory entries that don't exist! */
+	if (uoffset % LXSYS_SDSIZE)
+		return (ENOENT);
+
+	/* Free lower functions from having to check eofp == NULL */
+	if (eofp == NULL) {
+		eofp = &leof;
+	}
+
+	return (lxsys_readdir_function[lxsnp->lxsys_type](lxsnp, uiop, eofp));
+}
+
+static int
+lxsys_dirent_out(dirent64_t *d, ushort_t n, struct uio *uio)
+{
+	int error;
+	off_t offset = uio->uio_offset;
+
+	/*
+	 * uiomove() updates both uiop->uio_resid and uiop->uio_offset by the
+	 * same amount.  But we want uiop->uio_offset to change in increments
+	 * of LXSYS_SDSIZE, which is different from the number of bytes being
+	 * returned to the user.  To accomplish this, we set uiop->uio_offset
+	 * separately on success, overriding what uiomove() does.
+	 */
+	d->d_off = (off64_t)(offset + LXSYS_SDSIZE);
+	d->d_reclen = n;
+	if ((error = uiomove(d, n, UIO_READ, uio)) != 0) {
+		return (error);
+	}
+	uio->uio_offset = offset + LXSYS_SDSIZE;
+	return (0);
+}
+
+/*
+ * This has the common logic for returning directory entries
+ */
+static int
+lxsys_readdir_common(lxsys_node_t *lxsnp, uio_t *uiop, int *eofp,
+    lxsys_dirent_t *dirtab, int dirtablen)
+{
+	/* bp holds one dirent64 structure */
+	longlong_t bp[DIRENT64_RECLEN(LXSNSIZ) / sizeof (longlong_t)];
+	dirent64_t *dirent = (dirent64_t *)bp;
+	ssize_t oresid;	/* save a copy for testing later */
+	ssize_t uresid;
+
+	oresid = uiop->uio_resid;
+
+	/* clear out the dirent buffer */
+	bzero(bp, sizeof (bp));
+
+	/* Satisfy user request */
+	while ((uresid = uiop->uio_resid) > 0) {
+		int dirindex;
+		off_t uoffset;
+		int reclen;
+		int error;
+
+		uoffset = uiop->uio_offset;
+		dirindex  = (uoffset / LXSYS_SDSIZE) - 2;
+
+		if (uoffset == 0) {
+
+			dirent->d_ino = lxsnp->lxsys_ino;
+			dirent->d_name[0] = '.';
+			dirent->d_name[1] = '\0';
+			reclen = DIRENT64_RECLEN(1);
+
+		} else if (uoffset == LXSYS_SDSIZE) {
+
+			dirent->d_ino = lxsys_parentinode(lxsnp);
+			dirent->d_name[0] = '.';
+			dirent->d_name[1] = '.';
+			dirent->d_name[2] = '\0';
+			reclen = DIRENT64_RECLEN(2);
+
+		} else if (dirindex >= 0 && dirindex < dirtablen) {
+
+			int slen = strlen(dirtab[dirindex].d_name);
+
+			dirent->d_ino = lxsys_inode(LXSYS_STATIC,
+			    dirtab[dirindex].d_idnum, 0);
+			(void) strcpy(dirent->d_name, dirtab[dirindex].d_name);
+			reclen = DIRENT64_RECLEN(slen);
+
+		} else {
+			/* Run out of table entries */
+			*eofp = 1;
+			return (0);
+		}
+
+		/*
+		 * If the size of the data to transfer is greater than the
+		 * user-provided buffer, we cannot continue.
+		 */
+		if (reclen > uresid) {
+			/* Error if no entries have been returned yet. */
+			if (uresid == oresid) {
+				return (EINVAL);
+			}
+			break;
+		}
+
+		if ((error = lxsys_dirent_out(dirent, reclen, uiop)) != 0) {
+			return (error);
+		}
+	}
+
+	/* Have run out of space, but could have just done last table entry */
+	*eofp = (uiop->uio_offset >= ((dirtablen+2) * LXSYS_SDSIZE)) ?  1 : 0;
+	return (0);
+}
+
+static int
+lxsys_readdir_subdir(lxsys_node_t *lxsnp, uio_t *uiop, int *eofp,
+    lxsys_dirent_t *dirtab, int dirtablen)
+{
+	/* bp holds one dirent64 structure */
+	longlong_t bp[DIRENT64_RECLEN(LXSNSIZ) / sizeof (longlong_t)];
+	dirent64_t *dirent = (dirent64_t *)bp;
+	ssize_t oresid;	/* save a copy for testing later */
+	ssize_t uresid;
+
+	oresid = uiop->uio_resid;
+
+	/* clear out the dirent buffer */
+	bzero(bp, sizeof (bp));
+
+	/* Satisfy user request */
+	while ((uresid = uiop->uio_resid) > 0) {
+		int dirindex;
+		off_t uoffset;
+		int reclen;
+		int error;
+
+		uoffset = uiop->uio_offset;
+		dirindex  = (uoffset / LXSYS_SDSIZE) - 2;
+
+		if (uoffset == 0) {
+
+			dirent->d_ino = lxsnp->lxsys_ino;
+			dirent->d_name[0] = '.';
+			dirent->d_name[1] = '\0';
+			reclen = DIRENT64_RECLEN(1);
+
+		} else if (uoffset == LXSYS_SDSIZE) {
+
+			dirent->d_ino = lxsys_parentinode(lxsnp);
+			dirent->d_name[0] = '.';
+			dirent->d_name[1] = '.';
+			dirent->d_name[2] = '\0';
+			reclen = DIRENT64_RECLEN(2);
+
+		} else if (dirindex >= 0 && dirindex < dirtablen) {
+
+			int slen = strlen(dirtab[dirindex].d_name);
+
+			dirent->d_ino = lxsys_inode(lxsnp->lxsys_type,
+			    lxsnp->lxsys_instance, dirtab[dirindex].d_idnum);
+			(void) strcpy(dirent->d_name, dirtab[dirindex].d_name);
+			reclen = DIRENT64_RECLEN(slen);
+
+		} else {
+			/* Run out of table entries */
+			*eofp = 1;
+			return (0);
+		}
+
+		/*
+		 * If the size of the data to transfer is greater than the
+		 * user-provided buffer, we cannot continue.
+		 */
+		if (reclen > uresid) {
+			/* Error if no entries have been returned yet. */
+			if (uresid == oresid) {
+				return (EINVAL);
+			}
+			break;
+		}
+
+		if ((error = lxsys_dirent_out(dirent, reclen, uiop)) != 0) {
+			return (error);
+		}
+	}
+
+	/* Have run out of space, but could have just done last table entry */
+	*eofp = (uiop->uio_offset >= ((dirtablen+2) * LXSYS_SDSIZE)) ?  1 : 0;
+	return (0);
+}
+
+static int
+lxsys_readdir_ifaces(lxsys_node_t *ldp, struct uio *uiop, int *eofp,
+    lxsys_nodetype_t type)
+{
+	longlong_t bp[DIRENT64_RECLEN(LXSNSIZ) / sizeof (longlong_t)];
+	dirent64_t *dirent = (dirent64_t *)bp;
+	ssize_t oresid, uresid;
+	netstack_t *ns;
+	ip_stack_t *ipst;
+	avl_tree_t *phytree;
+	phyint_t *phyi;
+	int error, i;
+
+
+	/* Emit "." and ".." entries */
+	oresid = uiop->uio_resid;
+	error = lxsys_readdir_common(ldp, uiop, eofp, NULL, 0);
+	if (error != 0 || *eofp == 0) {
+		return (error);
+	}
+
+	if ((ns = lxsys_netstack(ldp)) == NULL) {
+		*eofp = 1;
+		return (0);
+	}
+	ipst = ns->netstack_ip;
+
+	rw_enter(&ipst->ips_ill_g_lock, RW_READER);
+	phytree = &ipst->ips_phyint_g_list->phyint_list_avl_by_index;
+	phyi = avl_first(phytree);
+	if (phyi == NULL) {
+		*eofp = 1;
+	}
+	bzero(bp, sizeof (bp));
+
+	/*
+	 * Skip records we have already passed with the offset.
+	 * This accounts for the two "." and ".." records already seen.
+	 */
+	for (i = (uiop->uio_offset/LXSYS_SDSIZE) - 2; i > 0; i--) {
+		if ((phyi = avl_walk(phytree, phyi, AVL_AFTER)) == NULL) {
+			*eofp = 1;
+			break;
+		}
+	}
+
+	while ((uresid = uiop->uio_resid) > 0 && phyi != NULL) {
+		uint_t ifindex;
+		int reclen;
+
+		ifindex = phyi->phyint_ifindex;
+		(void) strncpy(dirent->d_name, phyi->phyint_name, LIFNAMSIZ);
+		lx_ifname_convert(dirent->d_name, LX_IF_FROMNATIVE);
+		dirent->d_ino = lxsys_inode(type, ifindex, 0);
+		reclen = DIRENT64_RECLEN(strlen(dirent->d_name));
+
+		if (reclen > uresid) {
+			if (uresid == oresid) {
+				/* Not enough space for one record */
+				error = EINVAL;
+			}
+			break;
+		}
+		if ((error = lxsys_dirent_out(dirent, reclen, uiop)) != 0) {
+			break;
+		}
+
+		if ((phyi = avl_walk(phytree, phyi, AVL_AFTER)) == NULL) {
+			*eofp = 1;
+			break;
+		}
+	}
+
+	rw_exit(&ipst->ips_ill_g_lock);
+	netstack_rele(ns);
+	return (error);
+}
+
+
+static int
+lxsys_readdir_static(lxsys_node_t *lnp, uio_t *uiop, int *eofp)
+{
+	lxsys_dirent_t *dirent = NULL;
+	int i, len = 0;
+
+	for (i = 0; i < SYSDIRLISTSZ(lxsys_dirlookup); i++) {
+		if (lnp->lxsys_instance == lxsys_dirlookup[i].dl_instance) {
+			dirent = lxsys_dirlookup[i].dl_list;
+			len = lxsys_dirlookup[i].dl_length;
+			break;
+		}
+	}
+
+	if (dirent == NULL) {
+		return (ENOTDIR);
+	}
+
+	return (lxsys_readdir_common(lnp, uiop, eofp, dirent, len));
+}
+
+static int
+lxsys_readdir_class_netdir(lxsys_node_t *lnp, uio_t *uiop, int *eofp)
+{
+	if (lnp->lxsys_type != LXSYS_CLASS_NET ||
+	    lnp->lxsys_instance != 0) {
+		/*
+		 * Since /sys/class/net contains only symlinks, readdir
+		 * operations should not be performed anywhere except the top
+		 * level (instance == 0).
+		 */
+		return (ENOTDIR);
+	}
+
+	return (lxsys_readdir_ifaces(lnp, uiop, eofp, LXSYS_CLASS_NET));
+}
+
+static int
+lxsys_readdir_devices_virtual_netdir(lxsys_node_t *lnp, uio_t *uiop, int *eofp)
+{
+	int error;
+
+	if (lnp->lxsys_instance == 0) {
+		/* top-level interface listing */
+		error = lxsys_readdir_ifaces(lnp, uiop, eofp,
+		    LXSYS_DEVICES_NET);
+	} else if (lnp->lxsys_endpoint == 0) {
+		/* interface-level sub-item listing */
+		error = lxsys_readdir_subdir(lnp, uiop, eofp,
+		    dirlist_devices_virtual_net,
+		    SYSDIRLISTSZ(dirlist_devices_virtual_net));
+	} else {
+		/* there shouldn't be subdirs below this */
+		error = ENOTDIR;
+	}
+
+	return (error);
+}
+
+/*
+ * lxsys_readlink(): Vnode operation for VOP_READLINK()
+ */
+/* ARGSUSED */
+static int
+lxsys_readlink(vnode_t *vp, uio_t *uiop, cred_t *cr, caller_context_t *ct)
+{
+	char buf[MAXPATHLEN + 1];
+	lxsys_node_t *lnp = VTOLXS(vp);
+	lxsys_nodetype_t type = lnp->lxsys_type;
+	int (*rlfunc)();
+	int error;
+
+	VERIFY(type > LXSYS_NONE && type < LXSYS_MAXTYPE);
+
+	if (vp->v_type != VLNK) {
+		return (EINVAL);
+	}
+
+	rlfunc = lxsys_readlink_function[lnp->lxsys_type];
+	if (rlfunc != NULL) {
+		if ((error = rlfunc(lnp, buf, sizeof (buf))) == 0) {
+			error = uiomove(buf, strlen(buf), UIO_READ, uiop);
+		}
+	} else {
+		error = EINVAL;
+	}
+
+	return (error);
+}
+
+
+static int
+lxsys_readlink_class_net(lxsys_node_t *lnp, char *buf, size_t len)
+{
+	netstack_t *ns;
+	ip_stack_t *ipst;
+	avl_tree_t *phytree;
+	phyint_t *phyi;
+	uint_t ifindex;
+	char ifname[LIFNAMSIZ];
+	int error = EINVAL;
+
+	if ((ifindex = lnp->lxsys_instance) == 0) {
+		return (error);
+	}
+
+	if ((ns = lxsys_netstack(lnp)) == NULL) {
+		return (error);
+	}
+	ipst = ns->netstack_ip;
+	rw_enter(&ipst->ips_ill_g_lock, RW_READER);
+
+	phytree = &ipst->ips_phyint_g_list->phyint_list_avl_by_index;
+	phyi = avl_find(phytree, &ifindex, NULL);
+	if (phyi != NULL) {
+		(void) strncpy(ifname, phyi->phyint_name, LIFNAMSIZ);
+		lx_ifname_convert(ifname, LX_IF_FROMNATIVE);
+		(void) snprintf(buf, len, "/sys/devices/virtual/net/%s",
+		    ifname);
+		error = 0;
+	}
+
+	rw_exit(&ipst->ips_ill_g_lock);
+	netstack_rele(ns);
+	return (error);
+}
+
+/*
+ * lxsys_inactive(): Vnode operation for VOP_INACTIVE()
+ * Vnode is no longer referenced, deallocate the file
+ * and all its resources.
+ */
+/* ARGSUSED */
+static void
+lxsys_inactive(vnode_t *vp, cred_t *cr, caller_context_t *ct)
+{
+	lxsys_freenode(VTOLXS(vp));
+}
+
+/*
+ * lxsys_sync(): Vnode operation for VOP_SYNC()
+ */
+static int
+lxsys_sync()
+{
+	/*
+	 * Nothing to sync but this function must never fail
+	 */
+	return (0);
+}
+
+/*
+ * lxsys_cmp(): Vnode operation for VOP_CMP()
+ */
+static int
+lxsys_cmp(vnode_t *vp1, vnode_t *vp2, caller_context_t *ct)
+{
+	vnode_t *rvp;
+
+	while (vn_matchops(vp1, lxsys_vnodeops) &&
+	    (rvp = VTOLXS(vp1)->lxsys_realvp) != NULL) {
+		vp1 = rvp;
+	}
+
+	while (vn_matchops(vp2, lxsys_vnodeops) &&
+	    (rvp = VTOLXS(vp2)->lxsys_realvp) != NULL) {
+		vp2 = rvp;
+	}
+
+	if (vn_matchops(vp1, lxsys_vnodeops) ||
+	    vn_matchops(vp2, lxsys_vnodeops))
+		return (vp1 == vp2);
+	return (VOP_CMP(vp1, vp2, ct));
+}
+
+/*
+ * lxsys_realvp(): Vnode operation for VOP_REALVP()
+ */
+static int
+lxsys_realvp(vnode_t *vp, vnode_t **vpp, caller_context_t *ct)
+{
+	vnode_t *rvp;
+
+	if ((rvp = VTOLXS(vp)->lxsys_realvp) != NULL) {
+		vp = rvp;
+		if (VOP_REALVP(vp, &rvp, ct) == 0)
+			vp = rvp;
+	}
+
+	*vpp = vp;
+	return (0);
+}
diff --git a/usr/src/uts/common/brand/sn1/sn1_brand.c b/usr/src/uts/common/brand/sn1/sn1_brand.c
index d61928d578..7f99dc5f33 100644
--- a/usr/src/uts/common/brand/sn1/sn1_brand.c
+++ b/usr/src/uts/common/brand/sn1/sn1_brand.c
@@ -21,6 +21,7 @@
 
 /*
  * Copyright (c) 2006, 2010, Oracle and/or its affiliates. All rights reserved.
+ * Copyright 2015 Joyent, Inc. All rights reserved.
  */
 
 #include <sys/errno.h>
@@ -48,37 +49,59 @@ void	sn1_setbrand(proc_t *);
 int	sn1_getattr(zone_t *, int, void *, size_t *);
 int	sn1_setattr(zone_t *, int, void *, size_t);
 int	sn1_brandsys(int, int64_t *, uintptr_t, uintptr_t, uintptr_t,
-		uintptr_t, uintptr_t, uintptr_t);
+		uintptr_t, uintptr_t);
 void	sn1_copy_procdata(proc_t *, proc_t *);
-void	sn1_proc_exit(struct proc *, klwp_t *);
+void	sn1_proc_exit(struct proc *);
 void	sn1_exec();
-int	sn1_initlwp(klwp_t *);
+void	sn1_initlwp(klwp_t *, void *);
 void	sn1_forklwp(klwp_t *, klwp_t *);
 void	sn1_freelwp(klwp_t *);
 void	sn1_lwpexit(klwp_t *);
 int	sn1_elfexec(vnode_t *, execa_t *, uarg_t *, intpdata_t *, int,
-	long *, int, caddr_t, cred_t *, int);
+	long *, int, caddr_t, cred_t *, int *);
 
 /* sn1 brand */
 struct brand_ops sn1_brops = {
-	sn1_init_brand_data,
-	sn1_free_brand_data,
-	sn1_brandsys,
-	sn1_setbrand,
-	sn1_getattr,
-	sn1_setattr,
-	sn1_copy_procdata,
-	sn1_proc_exit,
-	sn1_exec,
-	lwp_setrval,
-	sn1_initlwp,
-	sn1_forklwp,
-	sn1_freelwp,
-	sn1_lwpexit,
-	sn1_elfexec,
-	NULL,
-	NULL,
-	NSIG,
+	sn1_init_brand_data,		/* b_init_brand_data */
+	sn1_free_brand_data,		/* b_free_brand_data */
+	sn1_brandsys,			/* b_brandsys */
+	sn1_setbrand,			/* b_setbrand */
+	sn1_getattr,			/* b_getattr */
+	sn1_setattr,			/* b_setattr */
+	sn1_copy_procdata,		/* b_copy_procdata */
+	sn1_proc_exit,			/* b_proc_exit */
+	sn1_exec,			/* b_exec */
+	lwp_setrval,			/* b_lwp_setrval */
+	NULL,				/* b_lwpdata_alloc */
+	NULL,				/* b_lwpdata_free */
+	sn1_initlwp,			/* b_initlwp */
+	sn1_forklwp,			/* b_forklwp */
+	sn1_freelwp,			/* b_freelwp */
+	sn1_lwpexit,			/* b_lwpexit */
+	sn1_elfexec,			/* b_elfexec */
+	NULL,				/* b_sigset_native_to_brand */
+	NULL,				/* b_sigset_brand_to_native */
+	NULL,				/* b_psig_to_proc */
+	NSIG,				/* b_nsig */
+	NULL,				/* b_exit_with_sig */
+	NULL,				/* b_wait_filter */
+	NULL,				/* b_native_exec */
+	NULL,				/* b_ptrace_exectrap */
+	NULL,				/* b_map32limit */
+	NULL,				/* b_stop_notify */
+	NULL,				/* b_waitid_helper */
+	NULL,				/* b_sigcld_repost */
+	NULL,				/* b_issig_stop */
+	NULL,				/* b_sig_ignorable */
+	NULL,				/* b_savecontext */
+#if defined(_SYSCALL32_IMPL)
+	NULL,				/* b_savecontext32 */
+#endif
+	NULL,				/* b_restorecontext */
+	NULL,				/* b_sendsig_stack */
+	NULL,				/* b_sendsig */
+	NULL,				/* b_setid_clear */
+	NULL				/* b_pagefault */
 };
 
 #ifdef	sparc
@@ -94,9 +117,12 @@ struct brand_mach_ops sn1_mops = {
 
 struct brand_mach_ops sn1_mops = {
 	sn1_brand_sysenter_callback,
+	NULL,
 	sn1_brand_int91_callback,
 	sn1_brand_syscall_callback,
-	sn1_brand_syscall32_callback
+	sn1_brand_syscall32_callback,
+	NULL,
+	NULL
 };
 
 #else	/* ! __amd64 */
@@ -104,7 +130,10 @@ struct brand_mach_ops sn1_mops = {
 struct brand_mach_ops sn1_mops = {
 	sn1_brand_sysenter_callback,
 	NULL,
+	NULL,
 	sn1_brand_syscall_callback,
+	NULL,
+	NULL,
 	NULL
 };
 #endif	/* __amd64 */
@@ -115,7 +144,8 @@ struct brand	sn1_brand = {
 	BRAND_VER_1,
 	"sn1",
 	&sn1_brops,
-	&sn1_mops
+	&sn1_mops,
+	sizeof (brand_proc_data_t),
 };
 
 static struct modlbrand modlbrand = {
@@ -151,7 +181,7 @@ sn1_setattr(zone_t *zone, int attr, void *buf, size_t bufsize)
 /*ARGSUSED*/
 int
 sn1_brandsys(int cmd, int64_t *rval, uintptr_t arg1, uintptr_t arg2,
-    uintptr_t arg3, uintptr_t arg4, uintptr_t arg5, uintptr_t arg6)
+    uintptr_t arg3, uintptr_t arg4, uintptr_t arg5)
 {
 	int	res;
 
@@ -171,9 +201,9 @@ sn1_copy_procdata(proc_t *child, proc_t *parent)
 }
 
 void
-sn1_proc_exit(struct proc *p, klwp_t *l)
+sn1_proc_exit(struct proc *p)
 {
-	brand_solaris_proc_exit(p, l, &sn1_brand);
+	brand_solaris_proc_exit(p, &sn1_brand);
 }
 
 void
@@ -182,10 +212,10 @@ sn1_exec()
 	brand_solaris_exec(&sn1_brand);
 }
 
-int
-sn1_initlwp(klwp_t *l)
+void
+sn1_initlwp(klwp_t *l, void *bd)
 {
-	return (brand_solaris_initlwp(l, &sn1_brand));
+	brand_solaris_initlwp(l, &sn1_brand);
 }
 
 void
@@ -221,11 +251,11 @@ sn1_init_brand_data(zone_t *zone)
 int
 sn1_elfexec(vnode_t *vp, execa_t *uap, uarg_t *args, intpdata_t *idatap,
 	int level, long *execsz, int setid, caddr_t exec_file, cred_t *cred,
-	int brand_action)
+	int *brand_action)
 {
 	return (brand_solaris_elfexec(vp, uap, args, idatap, level, execsz,
 	    setid, exec_file, cred, brand_action, &sn1_brand, SN1_BRANDNAME,
-	    SN1_LIB, SN1_LIB32, SN1_LINKER, SN1_LINKER32));
+	    SN1_LIB, SN1_LIB32));
 }
 
 int
diff --git a/usr/src/uts/common/brand/sn1/sn1_brand.h b/usr/src/uts/common/brand/sn1/sn1_brand.h
index b487745e21..fef9dc128b 100644
--- a/usr/src/uts/common/brand/sn1/sn1_brand.h
+++ b/usr/src/uts/common/brand/sn1/sn1_brand.h
@@ -20,6 +20,7 @@
  */
 /*
  * Copyright (c) 2006, 2010, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2014, Joyent, Inc. All rights reserved.
  */
 
 #ifndef _SN1_BRAND_H
@@ -37,20 +38,14 @@ extern "C" {
 #define	SN1_VERSION		SN1_VERSION_1
 
 #define	SN1_LIB_NAME		"sn1_brand.so.1"
-#define	SN1_LINKER_NAME		"ld.so.1"
 
 #define	SN1_LIB32		BRAND_NATIVE_DIR "usr/lib/" SN1_LIB_NAME
-#define	SN1_LINKER32		"/lib/" SN1_LINKER_NAME
-
 #define	SN1_LIB64		BRAND_NATIVE_DIR "usr/lib/64/" SN1_LIB_NAME
-#define	SN1_LINKER64		"/lib/64/" SN1_LINKER_NAME
 
 #if defined(_LP64)
 #define	SN1_LIB		SN1_LIB64
-#define	SN1_LINKER	SN1_LINKER64
 #else /* !_LP64 */
 #define	SN1_LIB		SN1_LIB32
-#define	SN1_LINKER	SN1_LINKER32
 #endif /* !_LP64 */
 
 #if defined(_KERNEL)
diff --git a/usr/src/uts/common/brand/sngl/sngl_brand.c b/usr/src/uts/common/brand/sngl/sngl_brand.c
new file mode 100644
index 0000000000..d5e1808d89
--- /dev/null
+++ b/usr/src/uts/common/brand/sngl/sngl_brand.c
@@ -0,0 +1,276 @@
+/*
+ * CDDL HEADER START
+ *
+ * The contents of this file are subject to the terms of the
+ * Common Development and Distribution License (the "License").
+ * You may not use this file except in compliance with the License.
+ *
+ * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
+ * or http://www.opensolaris.org/os/licensing.
+ * See the License for the specific language governing permissions
+ * and limitations under the License.
+ *
+ * When distributing Covered Code, include this CDDL HEADER in each
+ * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
+ * If applicable, add the following below this CDDL HEADER, with the
+ * fields enclosed by brackets "[]" replaced with your own identifying
+ * information: Portions Copyright [yyyy] [name of copyright owner]
+ *
+ * CDDL HEADER END
+ */
+
+/*
+ * Copyright 2015, Joyent, Inc. All rights reserved.
+ */
+
+#include <sys/errno.h>
+#include <sys/exec.h>
+#include <sys/file.h>
+#include <sys/kmem.h>
+#include <sys/modctl.h>
+#include <sys/model.h>
+#include <sys/proc.h>
+#include <sys/syscall.h>
+#include <sys/systm.h>
+#include <sys/thread.h>
+#include <sys/cmn_err.h>
+#include <sys/archsystm.h>
+#include <sys/pathname.h>
+#include <sys/sunddi.h>
+
+#include <sys/machbrand.h>
+#include <sys/brand.h>
+#include "sngl_brand.h"
+
+char *sngl_emulation_table = NULL;
+
+void	sngl_init_brand_data(zone_t *);
+void	sngl_free_brand_data(zone_t *);
+void	sngl_setbrand(proc_t *);
+int	sngl_getattr(zone_t *, int, void *, size_t *);
+int	sngl_setattr(zone_t *, int, void *, size_t);
+int	sngl_brandsys(int, int64_t *, uintptr_t, uintptr_t, uintptr_t,
+	uintptr_t, uintptr_t);
+void	sngl_copy_procdata(proc_t *, proc_t *);
+void	sngl_proc_exit(struct proc *);
+void	sngl_exec();
+void	sngl_initlwp(klwp_t *, void *);
+void	sngl_forklwp(klwp_t *, klwp_t *);
+void	sngl_freelwp(klwp_t *);
+void	sngl_lwpexit(klwp_t *);
+int	sngl_elfexec(vnode_t *, execa_t *, uarg_t *, intpdata_t *, int,
+	long *, int, caddr_t, cred_t *, int *);
+
+/* SNGL brand */
+struct brand_ops sngl_brops = {
+	sngl_init_brand_data,		/* b_init_brand_data */
+	sngl_free_brand_data,		/* b_free_brand_data */
+	sngl_brandsys,			/* b_brandsys */
+	sngl_setbrand,			/* b_setbrand */
+	sngl_getattr,			/* b_getattr */
+	sngl_setattr,			/* b_setattr */
+	sngl_copy_procdata,		/* b_copy_procdata */
+	sngl_proc_exit,			/* b_proc_exit */
+	sngl_exec,			/* b_exec */
+	lwp_setrval,			/* b_lwp_setrval */
+	NULL,				/* b_lwpdata_alloc */
+	NULL,				/* b_lwpdata_free */
+	sngl_initlwp,			/* b_initlwp */
+	sngl_forklwp,			/* b_forklwp */
+	sngl_freelwp,			/* b_freelwp */
+	sngl_lwpexit,			/* b_lwpexit */
+	sngl_elfexec,			/* b_elfexec */
+	NULL,				/* b_sigset_native_to_brand */
+	NULL,				/* b_sigset_brand_to_native */
+	NULL,				/* b_psig_to_proc */
+	NSIG,				/* b_nsig */
+	NULL,				/* b_exit_with_sig */
+	NULL,				/* b_wait_filter */
+	NULL,				/* b_native_exec */
+	NULL,				/* b_ptrace_exectrap */
+	NULL,				/* b_map32limit */
+	NULL,				/* b_stop_notify */
+	NULL,				/* b_waitid_helper */
+	NULL,				/* b_sigcld_repost */
+	NULL,				/* b_issig_stop */
+	NULL,				/* b_sig_ignorable */
+	NULL,				/* b_savecontext */
+#if defined(_SYSCALL32_IMPL)
+	NULL,				/* b_savecontext32 */
+#endif
+	NULL,				/* b_restorecontext */
+	NULL,				/* b_sendsig_stack */
+	NULL,				/* b_sendsig */
+	NULL,				/* b_setid_clear */
+	NULL				/* b_pagefault */
+};
+
+#ifdef	__amd64
+
+struct brand_mach_ops sngl_mops = {
+	sngl_brand_sysenter_callback,
+	sngl_brand_int91_callback,
+	sngl_brand_syscall_callback,
+	sngl_brand_syscall32_callback,
+	NULL
+};
+
+#else	/* ! __amd64 */
+
+struct brand_mach_ops sngl_mops = {
+	sngl_brand_sysenter_callback,
+	NULL,
+	sngl_brand_syscall_callback,
+	NULL,
+	NULL
+};
+#endif	/* __amd64 */
+
+struct brand	sngl_brand = {
+	BRAND_VER_1,
+	"sngl",
+	&sngl_brops,
+	&sngl_mops,
+	sizeof (brand_proc_data_t),
+};
+
+static struct modlbrand modlbrand = {
+	&mod_brandops,		/* type of module */
+	"SNGL Brand",		/* description of module */
+	&sngl_brand		/* driver ops */
+};
+
+static struct modlinkage modlinkage = {
+	MODREV_1, (void *)&modlbrand, NULL
+};
+
+void
+sngl_setbrand(proc_t *p)
+{
+	brand_solaris_setbrand(p, &sngl_brand);
+}
+
+/*ARGSUSED*/
+int
+sngl_getattr(zone_t *zone, int attr, void *buf, size_t *bufsize)
+{
+	return (EINVAL);
+}
+
+/*ARGSUSED*/
+int
+sngl_setattr(zone_t *zone, int attr, void *buf, size_t bufsize)
+{
+	return (EINVAL);
+}
+
+/*ARGSUSED*/
+int
+sngl_brandsys(int cmd, int64_t *rval, uintptr_t arg1, uintptr_t arg2,
+    uintptr_t arg3, uintptr_t arg4, uintptr_t arg5)
+{
+	int	res;
+
+	*rval = 0;
+	res = brand_solaris_cmd(cmd, arg1, arg2, arg3, &sngl_brand,
+	    SNGL_VERSION);
+	if (res >= 0)
+		return (res);
+
+	return (EINVAL);
+}
+
+void
+sngl_copy_procdata(proc_t *child, proc_t *parent)
+{
+	brand_solaris_copy_procdata(child, parent, &sngl_brand);
+}
+
+void
+sngl_proc_exit(struct proc *p)
+{
+	brand_solaris_proc_exit(p, &sngl_brand);
+}
+
+void
+sngl_exec()
+{
+	brand_solaris_exec(&sngl_brand);
+}
+
+void
+sngl_initlwp(klwp_t *l, void *bd)
+{
+	brand_solaris_initlwp(l, &sngl_brand);
+}
+
+void
+sngl_forklwp(klwp_t *p, klwp_t *c)
+{
+	brand_solaris_forklwp(p, c, &sngl_brand);
+}
+
+void
+sngl_freelwp(klwp_t *l)
+{
+	brand_solaris_freelwp(l, &sngl_brand);
+}
+
+void
+sngl_lwpexit(klwp_t *l)
+{
+	brand_solaris_lwpexit(l, &sngl_brand);
+}
+
+void
+sngl_free_brand_data(zone_t *zone)
+{
+}
+
+void
+sngl_init_brand_data(zone_t *zone)
+{
+}
+
+int
+sngl_elfexec(vnode_t *vp, execa_t *uap, uarg_t *args, intpdata_t *idatap,
+	int level, long *execsz, int setid, caddr_t exec_file, cred_t *cred,
+	int *brand_action)
+{
+	return (brand_solaris_elfexec(vp, uap, args, idatap, level, execsz,
+	    setid, exec_file, cred, brand_action, &sngl_brand, SNGL_BRANDNAME,
+	    SNGL_LIB, SNGL_LIB32));
+}
+
+int
+_init(void)
+{
+	int err;
+
+	/*
+	 * Set up the table to interpose on open.
+	 */
+	sngl_emulation_table = kmem_zalloc(NSYSCALL, KM_SLEEP);
+	sngl_emulation_table[SYS_open] = 1;			/*   5 */
+
+	err = mod_install(&modlinkage);
+	if (err) {
+		cmn_err(CE_WARN, "Couldn't install brand module");
+		kmem_free(sngl_emulation_table, NSYSCALL);
+	}
+
+	return (err);
+}
+
+int
+_info(struct modinfo *modinfop)
+{
+	return (mod_info(&modlinkage, modinfop));
+}
+
+int
+_fini(void)
+{
+	return (brand_solaris_fini(&sngl_emulation_table, &modlinkage,
+	    &sngl_brand));
+}
diff --git a/usr/src/uts/common/brand/sngl/sngl_brand.h b/usr/src/uts/common/brand/sngl/sngl_brand.h
new file mode 100644
index 0000000000..7c6cbd5412
--- /dev/null
+++ b/usr/src/uts/common/brand/sngl/sngl_brand.h
@@ -0,0 +1,65 @@
+/*
+ * CDDL HEADER START
+ *
+ * The contents of this file are subject to the terms of the
+ * Common Development and Distribution License (the "License").
+ * You may not use this file except in compliance with the License.
+ *
+ * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
+ * or http://www.opensolaris.org/os/licensing.
+ * See the License for the specific language governing permissions
+ * and limitations under the License.
+ *
+ * When distributing Covered Code, include this CDDL HEADER in each
+ * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
+ * If applicable, add the following below this CDDL HEADER, with the
+ * fields enclosed by brackets "[]" replaced with your own identifying
+ * information: Portions Copyright [yyyy] [name of copyright owner]
+ *
+ * CDDL HEADER END
+ */
+
+/*
+ * Copyright 2014, Joyent, Inc. All rights reserved.
+ */
+
+#ifndef _SNGL_BRAND_H
+#define	_SNGL_BRAND_H
+
+#ifdef	__cplusplus
+extern "C" {
+#endif
+
+#include <sys/brand.h>
+
+#define	SNGL_BRANDNAME		"sngl"
+
+#define	SNGL_VERSION		1
+
+#define	SNGL_LIB_NAME		"sngl_brand.so.1"
+#define	SNGL_LIB32		"/system/usr/lib/" SNGL_LIB_NAME
+#define	SNGL_LIB64		"/system/usr/lib/64/" SNGL_LIB_NAME
+
+#if defined(_LP64)
+#define	SNGL_LIB	SNGL_LIB64
+#else /* !_LP64 */
+#define	SNGL_LIB	SNGL_LIB32
+#endif /* !_LP64 */
+
+#if defined(_KERNEL)
+
+void sngl_brand_syscall_callback(void);
+void sngl_brand_sysenter_callback(void);
+
+#if defined(__amd64)
+void sngl_brand_syscall32_callback(void);
+void sngl_brand_int91_callback(void);
+#endif /* __amd64 */
+
+#endif /* _KERNEL */
+
+#ifdef	__cplusplus
+}
+#endif
+
+#endif	/* _SNGL_BRAND_H */
diff --git a/usr/src/uts/common/brand/solaris10/s10_brand.c b/usr/src/uts/common/brand/solaris10/s10_brand.c
index f24b864eef..f70f54a788 100644
--- a/usr/src/uts/common/brand/solaris10/s10_brand.c
+++ b/usr/src/uts/common/brand/solaris10/s10_brand.c
@@ -22,6 +22,7 @@
 /*
  * Copyright (c) 2013, OmniTI Computer Consulting, Inc. All rights reserved.
  * Copyright (c) 2009, 2010, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2015, Joyent, Inc. All rights reserved.
  */
 
 #include <sys/errno.h>
@@ -51,39 +52,61 @@ void	s10_setbrand(proc_t *);
 int	s10_getattr(zone_t *, int, void *, size_t *);
 int	s10_setattr(zone_t *, int, void *, size_t);
 int	s10_brandsys(int, int64_t *, uintptr_t, uintptr_t, uintptr_t,
-		uintptr_t, uintptr_t, uintptr_t);
+		uintptr_t, uintptr_t);
 void	s10_copy_procdata(proc_t *, proc_t *);
-void	s10_proc_exit(struct proc *, klwp_t *);
+void	s10_proc_exit(struct proc *);
 void	s10_exec();
-int	s10_initlwp(klwp_t *);
+void	s10_initlwp(klwp_t *, void *);
 void	s10_forklwp(klwp_t *, klwp_t *);
 void	s10_freelwp(klwp_t *);
 void	s10_lwpexit(klwp_t *);
 int	s10_elfexec(vnode_t *, execa_t *, uarg_t *, intpdata_t *, int,
-	long *, int, caddr_t, cred_t *, int);
+	long *, int, caddr_t, cred_t *, int *);
 void	s10_sigset_native_to_s10(sigset_t *);
 void	s10_sigset_s10_to_native(sigset_t *);
 
 /* s10 brand */
 struct brand_ops s10_brops = {
-	s10_init_brand_data,
-	s10_free_brand_data,
-	s10_brandsys,
-	s10_setbrand,
-	s10_getattr,
-	s10_setattr,
-	s10_copy_procdata,
-	s10_proc_exit,
-	s10_exec,
-	lwp_setrval,
-	s10_initlwp,
-	s10_forklwp,
-	s10_freelwp,
-	s10_lwpexit,
-	s10_elfexec,
-	s10_sigset_native_to_s10,
-	s10_sigset_s10_to_native,
-	S10_NSIG,
+	s10_init_brand_data,		/* b_init_brand_data */
+	s10_free_brand_data,		/* b_free_brand_data */
+	s10_brandsys,			/* b_brandsys */
+	s10_setbrand,			/* b_setbrand */
+	s10_getattr,			/* b_getattr */
+	s10_setattr,			/* b_setattr */
+	s10_copy_procdata,		/* b_copy_procdata */
+	s10_proc_exit,			/* b_proc_exit */
+	s10_exec,			/* b_exec */
+	lwp_setrval,			/* b_lwp_setrval */
+	NULL,				/* b_lwpdata_alloc */
+	NULL,				/* b_lwpdata_free */
+	s10_initlwp,			/* b_initlwp */
+	s10_forklwp,			/* b_forklwp */
+	s10_freelwp,			/* b_freelwp */
+	s10_lwpexit,			/* b_lwpexit */
+	s10_elfexec,			/* b_elfexec */
+	s10_sigset_native_to_s10,	/* b_sigset_native_to_brand */
+	s10_sigset_s10_to_native,	/* b_sigset_brand_to_native */
+	NULL,				/* b_psig_to_proc */
+	S10_NSIG,			/* b_nsig */
+	NULL,				/* b_exit_with_sig */
+	NULL,				/* b_wait_filter */
+	NULL,				/* b_native_exec */
+	NULL,				/* b_ptrace_exectrap */
+	NULL,				/* b_map32limit */
+	NULL,				/* b_stop_notify */
+	NULL,				/* b_waitid_helper */
+	NULL,				/* b_sigcld_repost */
+	NULL,				/* b_issig_stop */
+	NULL,				/* b_sig_ignorable */
+	NULL,				/* b_savecontext */
+#if defined(_SYSCALL32_IMPL)
+	NULL,				/* b_savecontext32 */
+#endif
+	NULL,				/* b_restorecontext */
+	NULL,				/* b_sendsig_stack */
+	NULL,				/* b_sendsig */
+	NULL,				/* b_setid_clear */
+	NULL				/* b_pagefault */
 };
 
 #ifdef	sparc
@@ -99,9 +122,12 @@ struct brand_mach_ops s10_mops = {
 
 struct brand_mach_ops s10_mops = {
 	s10_brand_sysenter_callback,
+	NULL,
 	s10_brand_int91_callback,
 	s10_brand_syscall_callback,
-	s10_brand_syscall32_callback
+	s10_brand_syscall32_callback,
+	NULL,
+	NULL
 };
 
 #else	/* ! __amd64 */
@@ -109,7 +135,10 @@ struct brand_mach_ops s10_mops = {
 struct brand_mach_ops s10_mops = {
 	s10_brand_sysenter_callback,
 	NULL,
+	NULL,
 	s10_brand_syscall_callback,
+	NULL,
+	NULL,
 	NULL
 };
 #endif	/* __amd64 */
@@ -120,7 +149,8 @@ struct brand	s10_brand = {
 	BRAND_VER_1,
 	"solaris10",
 	&s10_brops,
-	&s10_mops
+	&s10_mops,
+	sizeof (brand_proc_data_t),
 };
 
 static struct modlbrand modlbrand = {
@@ -252,7 +282,7 @@ s10_native(void *cmd, void *args)
 /*ARGSUSED*/
 int
 s10_brandsys(int cmd, int64_t *rval, uintptr_t arg1, uintptr_t arg2,
-    uintptr_t arg3, uintptr_t arg4, uintptr_t arg5, uintptr_t arg6)
+    uintptr_t arg3, uintptr_t arg4, uintptr_t arg5)
 {
 	proc_t	*p = curproc;
 	int	res;
@@ -326,9 +356,9 @@ s10_copy_procdata(proc_t *child, proc_t *parent)
 }
 
 void
-s10_proc_exit(struct proc *p, klwp_t *l)
+s10_proc_exit(struct proc *p)
 {
-	brand_solaris_proc_exit(p, l, &s10_brand);
+	brand_solaris_proc_exit(p, &s10_brand);
 }
 
 void
@@ -337,10 +367,10 @@ s10_exec()
 	brand_solaris_exec(&s10_brand);
 }
 
-int
-s10_initlwp(klwp_t *l)
+void
+s10_initlwp(klwp_t *l, void *bd)
 {
-	return (brand_solaris_initlwp(l, &s10_brand));
+	brand_solaris_initlwp(l, &s10_brand);
 }
 
 void
@@ -390,11 +420,11 @@ s10_init_brand_data(zone_t *zone)
 int
 s10_elfexec(vnode_t *vp, execa_t *uap, uarg_t *args, intpdata_t *idatap,
 	int level, long *execsz, int setid, caddr_t exec_file, cred_t *cred,
-	int brand_action)
+	int *brand_action)
 {
 	return (brand_solaris_elfexec(vp, uap, args, idatap, level, execsz,
 	    setid, exec_file, cred, brand_action, &s10_brand, S10_BRANDNAME,
-	    S10_LIB, S10_LIB32, S10_LINKER, S10_LINKER32));
+	    S10_LIB, S10_LIB32));
 }
 
 void
diff --git a/usr/src/uts/common/brand/solaris10/s10_brand.h b/usr/src/uts/common/brand/solaris10/s10_brand.h
index 11f9853f48..ffef485e12 100644
--- a/usr/src/uts/common/brand/solaris10/s10_brand.h
+++ b/usr/src/uts/common/brand/solaris10/s10_brand.h
@@ -22,6 +22,7 @@
 /*
  * Copyright (c) 2013, OmniTI Computer Consulting, Inc. All rights reserved.
  * Copyright (c) 2009, 2010, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2014, Joyent, Inc. All rights reserved.
  */
 
 #ifndef _S10_BRAND_H
@@ -42,17 +43,12 @@ extern "C" {
 #define	S10_LINKER_NAME		"ld.so.1"
 
 #define	S10_LIB32		BRAND_NATIVE_DIR "usr/lib/" S10_LIB_NAME
-#define	S10_LINKER32		"/lib/" S10_LINKER_NAME
-
 #define	S10_LIB64		BRAND_NATIVE_DIR "usr/lib/64/" S10_LIB_NAME
-#define	S10_LINKER64		"/lib/64/" S10_LINKER_NAME
 
 #if defined(_LP64)
 #define	S10_LIB		S10_LIB64
-#define	S10_LINKER	S10_LINKER64
 #else /* !_LP64 */
 #define	S10_LIB		S10_LIB32
-#define	S10_LINKER	S10_LINKER32
 #endif /* !_LP64 */
 
 /*
diff --git a/usr/src/uts/common/conf/param.c b/usr/src/uts/common/conf/param.c
index 6b3ba51f31..a71be771fd 100644
--- a/usr/src/uts/common/conf/param.c
+++ b/usr/src/uts/common/conf/param.c
@@ -22,6 +22,7 @@
 /*
  * Copyright 2014 Nexenta Systems, Inc.  All rights reserved.
  * Copyright (c) 1983, 2010, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2012, Joyent, Inc. All rights reserved.
  * Copyright 2012 Milan Jurik. All rights reserved.
  */
 
@@ -559,8 +560,8 @@ char *isa_list = architecture;
 static pgcnt_t original_physmem = 0;
 
 #define	MIN_DEFAULT_MAXUSERS	8u
-#define	MAX_DEFAULT_MAXUSERS	2048u
-#define	MAX_MAXUSERS		4096u
+#define	MAX_DEFAULT_MAXUSERS	10000u
+#define	MAX_MAXUSERS		20000u
 
 void
 param_preset(void)
@@ -572,7 +573,7 @@ void
 param_calc(int platform_max_nprocs)
 {
 	/*
-	 * Default to about one "user" per megabyte, taking into
+	 * Default to about one "user" per 8MB, taking into
 	 * account both physical and virtual constraints.
 	 * Note: 2^20 is a meg; shifting right by (20 - PAGESHIFT)
 	 * converts pages to megs without integer overflow.
@@ -586,8 +587,9 @@ param_calc(int platform_max_nprocs)
 	if (maxusers == 0) {
 		pgcnt_t physmegs = physmem >> (20 - PAGESHIFT);
 		pgcnt_t virtmegs = vmem_size(heap_arena, VMEM_FREE) >> 20;
-		maxusers = MIN(MAX(MIN(physmegs, virtmegs),
-		    MIN_DEFAULT_MAXUSERS), MAX_DEFAULT_MAXUSERS);
+		maxusers = MIN(physmegs, virtmegs) >> 3; /* divide by 8 */
+		maxusers = MAX(maxusers, MIN_DEFAULT_MAXUSERS);
+		maxusers = MIN(maxusers, MAX_DEFAULT_MAXUSERS);
 	}
 	if (maxusers > MAX_MAXUSERS) {
 		maxusers = MAX_MAXUSERS;
diff --git a/usr/src/uts/common/crypto/api/kcf_random.c b/usr/src/uts/common/crypto/api/kcf_random.c
index bc72fa984a..75072fb686 100644
--- a/usr/src/uts/common/crypto/api/kcf_random.c
+++ b/usr/src/uts/common/crypto/api/kcf_random.c
@@ -70,6 +70,7 @@
 #include <sys/cpuvar.h>
 #include <sys/taskq.h>
 #include <rng/fips_random.h>
+#include <sys/strlog.h>
 
 #define	RNDPOOLSIZE		1024	/* Pool size in bytes */
 #define	MINEXTRACTBYTES		20
@@ -933,7 +934,8 @@ rnd_handler(void *arg)
 	int len = 0;
 
 	if (!rng_prov_found && rng_ok_to_log) {
-		cmn_err(CE_WARN, "No randomness provider enabled for "
+		(void) strlog(0, 0, 0, SL_NOTE,
+		    "No randomness provider enabled for "
 		    "/dev/random. Use cryptoadm(1M) to enable a provider.");
 		rng_ok_to_log = B_FALSE;
 	}
diff --git a/usr/src/uts/common/crypto/core/kcf_sched.c b/usr/src/uts/common/crypto/core/kcf_sched.c
index f461fe048c..8b2760b237 100644
--- a/usr/src/uts/common/crypto/core/kcf_sched.c
+++ b/usr/src/uts/common/crypto/core/kcf_sched.c
@@ -1027,9 +1027,9 @@ kcfpool_svc(void *arg)
 			case 0:
 			case -1:
 				/*
-				 * Woke up with no work to do. Check
-				 * if this thread should exit. We keep
-				 * at least kcf_minthreads.
+				 * Woke up with no work to do. Check if we
+				 * should lwp_exit() (which won't return). We
+				 * keep at least kcf_minthreads.
 				 */
 				if (kcfpool->kp_threads > kcf_minthreads) {
 					KCF_ATOMIC_DECR(kcfpool->kp_threads);
diff --git a/usr/src/uts/common/ctf/ctf_mod.c b/usr/src/uts/common/ctf/ctf_mod.c
index b34cf400cd..421b922c96 100644
--- a/usr/src/uts/common/ctf/ctf_mod.c
+++ b/usr/src/uts/common/ctf/ctf_mod.c
@@ -24,8 +24,6 @@
  * Use is subject to license terms.
  */
 
-#pragma ident	"%Z%%M%	%I%	%E% SMI"
-
 #include <sys/sysmacros.h>
 #include <sys/modctl.h>
 #include <sys/debug.h>
@@ -117,6 +115,15 @@ ctf_version(int version)
 
 /*ARGSUSED*/
 ctf_file_t *
+ctf_fdcreate_int(int fd, int *errp, ctf_sect_t *ctfp)
+{
+	if (errp != NULL)
+		*errp = ENOTSUP;
+	return (NULL);
+}
+
+/*ARGSUSED*/
+ctf_file_t *
 ctf_modopen(struct module *mp, int *error)
 {
 	ctf_sect_t ctfsect, symsect, strsect;
diff --git a/usr/src/uts/common/disp/cmt.c b/usr/src/uts/common/disp/cmt.c
index 1c5e1f79a9..3ecbf39393 100644
--- a/usr/src/uts/common/disp/cmt.c
+++ b/usr/src/uts/common/disp/cmt.c
@@ -201,13 +201,15 @@ pg_cmt_cpu_startup(cpu_t *cp)
 
 /*
  * Return non-zero if thread can migrate between "from" and "to"
- * without a performance penalty
+ * without a performance penalty.  This is true only if we share a core on
+ * virtually any CPU; sharing the last-level cache is insufficient to make
+ * migration possible without penalty.
  */
 int
 pg_cmt_can_migrate(cpu_t *from, cpu_t *to)
 {
-	if (from->cpu_physid->cpu_cacheid ==
-	    to->cpu_physid->cpu_cacheid)
+	if (from->cpu_physid->cpu_coreid ==
+	    to->cpu_physid->cpu_coreid)
 		return (1);
 	return (0);
 }
diff --git a/usr/src/uts/common/disp/cpucaps.c b/usr/src/uts/common/disp/cpucaps.c
index 46f53faab6..2a4365ff73 100644
--- a/usr/src/uts/common/disp/cpucaps.c
+++ b/usr/src/uts/common/disp/cpucaps.c
@@ -22,6 +22,7 @@
 /*
  * Copyright 2009 Sun Microsystems, Inc.  All rights reserved.
  * Use is subject to license terms.
+ * Copyright 2013 Joyent, Inc.  All rights reserved.
  */
 
 #include <sys/disp.h>
@@ -74,6 +75,32 @@
  * Putting threads on wait queues in random places while running in the
  * kernel might lead to all kinds of locking problems.
  *
+ * Bursting
+ * ========
+ *
+ * CPU bursting occurs when the CPU usage is over the baseline but under the
+ * cap.  The baseline CPU (zone.cpu-baseline) is set in a multi-tenant
+ * environment so that we know how much CPU is allocated for a tenant under
+ * normal utilization.  We can then track how much time a zone is spending
+ * over the "normal" CPU utilization expected for that zone using the
+ * "above_base_sec" kstat. This kstat is cumulative.
+ *
+ * If the zone has a burst limit (zone.cpu-burst-time) then the zone can
+ * burst for that period of time (in seconds) before the effective cap is
+ * lowered to the baseline.  Once the effective cap is lowered, the zone
+ * will run at the baseline for the burst limit before the effective cap is
+ * raised again to the full value.  This will allow the zone to burst again.
+ * We can watch this behavior using the kstats.  The "effective" kstat shows
+ * which cap is being used, the baseline value or the burst value.  The
+ * "burst_limit_sec" shows the value of the zone.cpu-burst-time rctl and the
+ * "bursting_sec" kstat shows how many seconds the zone has currently been
+ * bursting.  When the CPU load is continuously greater than the baseline,
+ * bursting_sec will increase, up to the burst_limit_sec value, then the
+ * effective kstat will drop to the baseline and the bursting_sec value will
+ * decrease until it hits 0, at which time the effective kstat will return to
+ * the full burst value and the bursting_sec value will begin to increase
+ * again.
+ *
  * Accounting
  * ==========
  *
@@ -203,18 +230,28 @@ static void caps_update();
  */
 struct cap_kstat {
 	kstat_named_t	cap_value;
+	kstat_named_t	cap_baseline;
+	kstat_named_t	cap_effective;
+	kstat_named_t	cap_burst_limit;
+	kstat_named_t	cap_bursting;
 	kstat_named_t	cap_usage;
 	kstat_named_t	cap_nwait;
 	kstat_named_t	cap_below;
 	kstat_named_t	cap_above;
+	kstat_named_t	cap_above_base;
 	kstat_named_t	cap_maxusage;
 	kstat_named_t	cap_zonename;
 } cap_kstat = {
 	{ "value",	KSTAT_DATA_UINT64 },
+	{ "baseline",	KSTAT_DATA_UINT64 },
+	{ "effective",	KSTAT_DATA_UINT64 },
+	{ "burst_limit_sec", KSTAT_DATA_UINT64 },
+	{ "bursting_sec", KSTAT_DATA_UINT64 },
 	{ "usage",	KSTAT_DATA_UINT64 },
 	{ "nwait",	KSTAT_DATA_UINT64 },
 	{ "below_sec",	KSTAT_DATA_UINT64 },
 	{ "above_sec",	KSTAT_DATA_UINT64 },
+	{ "above_base_sec", KSTAT_DATA_UINT64 },
 	{ "maxusage",	KSTAT_DATA_UINT64 },
 	{ "zonename",	KSTAT_DATA_STRING },
 };
@@ -311,7 +348,7 @@ cap_enable(list_t *l, cpucap_t *cap, hrtime_t value)
 	cap->cap_below = cap->cap_above = 0;
 	cap->cap_maxusage = 0;
 	cap->cap_usage = 0;
-	cap->cap_value = value;
+	cap->cap_value = cap->cap_chk_value = value;
 	waitq_unblock(&cap->cap_waitq);
 	if (CPUCAPS_OFF()) {
 		cpucaps_enabled = B_TRUE;
@@ -340,19 +377,21 @@ cap_disable(list_t *l, cpucap_t *cap)
 	ASSERT(CAP_ENABLED(cap));
 
 	waitq_block(&cap->cap_waitq);
+
+	/* do this first to avoid race with cap_kstat_update */
+	if (cap->cap_kstat != NULL) {
+		kstat_delete(cap->cap_kstat);
+		cap->cap_kstat = NULL;
+	}
+
 	list_remove(l, cap);
 	if (list_is_empty(&capped_projects) && list_is_empty(&capped_zones)) {
 		cpucaps_enabled = B_FALSE;
 		cpucaps_clock_callout = NULL;
 	}
-	cap->cap_value = 0;
+	cap->cap_value = cap->cap_chk_value = 0;
 	cap->cap_project = NULL;
 	cap->cap_zone = NULL;
-	if (cap->cap_kstat != NULL) {
-		kstat_delete(cap->cap_kstat);
-		cap->cap_kstat = NULL;
-	}
-
 }
 
 /*
@@ -487,6 +526,8 @@ cap_walk(list_t *l, void (*cb)(cpucap_t *, int64_t))
  * The waitq_isempty check is performed without the waitq lock. If a new thread
  * is placed on the waitq right after the check, it will be picked up during the
  * next invocation of cap_poke_waitq().
+ *
+ * Called once per tick for zones.
  */
 /* ARGSUSED */
 static void
@@ -494,15 +535,92 @@ cap_poke_waitq(cpucap_t *cap, int64_t gen)
 {
 	ASSERT(MUTEX_HELD(&caps_lock));
 
-	if (cap->cap_usage >= cap->cap_value) {
+	if (cap->cap_base != 0) {
+		/*
+		 * Because of the way usage is calculated and decayed, its
+		 * possible for the zone to be slightly over its cap, but we
+		 * don't want to count that after we have reduced the effective
+		 * cap to the baseline.  That way the zone will be able to
+		 * burst again after the burst_limit has expired.
+		 */
+		if (cap->cap_usage > cap->cap_base &&
+		    cap->cap_chk_value == cap->cap_value) {
+			cap->cap_above_base++;
+
+			/*
+			 * If bursting is limited and we've been bursting
+			 * longer than we're supposed to, then set the
+			 * effective cap to the baseline.
+			 */
+			if (cap->cap_burst_limit != 0) {
+				cap->cap_bursting++;
+				if (cap->cap_bursting >= cap->cap_burst_limit)
+					cap->cap_chk_value = cap->cap_base;
+			}
+		} else if (cap->cap_bursting > 0) {
+			/*
+			 * We're not bursting now, but we were, decay the
+			 * bursting timer.
+			 */
+			cap->cap_bursting--;
+			/*
+			 * Reset the effective cap once we decay to 0 so we
+			 * can burst again.
+			 */
+			if (cap->cap_bursting == 0 &&
+			    cap->cap_chk_value != cap->cap_value)
+				cap->cap_chk_value = cap->cap_value;
+		}
+	}
+
+	if (cap->cap_usage >= cap->cap_chk_value) {
 		cap->cap_above++;
 	} else {
 		waitq_t *wq = &cap->cap_waitq;
 
 		cap->cap_below++;
 
-		if (!waitq_isempty(wq))
-			waitq_runone(wq);
+		if (!waitq_isempty(wq)) {
+			int i, ndequeue, p;
+
+			/*
+			 * Since this function is only called once per tick,
+			 * we can hit a situation where we have artificially
+			 * limited the project/zone below its cap.  This would
+			 * happen if we have multiple threads queued up but
+			 * only dequeued one thread/tick. To avoid this we
+			 * dequeue multiple threads, calculated based on the
+			 * usage percentage of the cap. It is possible that we
+			 * could dequeue too many threads and some of them
+			 * might be put back on the wait queue quickly, but
+			 * since we know that threads are on the wait queue
+			 * because we're capping, we know that there is unused
+			 * CPU cycles anyway, so this extra work would not
+			 * hurt. Also, the ndequeue number is only an upper
+			 * bound and we might dequeue less, depending on how
+			 * many threads are actually in the wait queue. The
+			 * ndequeue values are empirically derived and could be
+			 * adjusted or calculated in another way if necessary.
+			 */
+			p = (int)((100 * cap->cap_usage) / cap->cap_chk_value);
+			if (p >= 98)
+				ndequeue = 10;
+			else if (p >= 95)
+				ndequeue = 20;
+			else if (p >= 90)
+				ndequeue = 40;
+			else if (p >= 85)
+				ndequeue = 80;
+			else
+				ndequeue = 160;
+
+			for (i = 0; i < ndequeue; i++) {
+				waitq_runone(wq);
+				if (waitq_isempty(wq))
+					break;
+			}
+			DTRACE_PROBE2(cpucaps__pokeq, int, p, int, i);
+		}
 	}
 }
 
@@ -629,14 +747,14 @@ cap_project_zone_modify_walker(kproject_t *kpj, void *arg)
 		 * Remove all projects in this zone without caps
 		 * from the capped_projects list.
 		 */
-		if (project_cap->cap_value == MAX_USAGE) {
+		if (project_cap->cap_chk_value == MAX_USAGE) {
 			cap_project_disable(kpj);
 		}
 	} else if (CAP_DISABLED(project_cap)) {
 		/*
 		 * Add the project to capped_projects list.
 		 */
-		ASSERT(project_cap->cap_value == 0);
+		ASSERT(project_cap->cap_chk_value == 0);
 		cap_project_enable(kpj, MAX_USAGE);
 	}
 	mutex_exit(&caps_lock);
@@ -746,7 +864,7 @@ cpucaps_zone_set(zone_t *zone, rctl_qty_t cap_val)
 		/*
 		 * No state transitions, just change the value
 		 */
-		cap->cap_value = value;
+		cap->cap_value = cap->cap_chk_value = value;
 	}
 
 	ASSERT(MUTEX_HELD(&caps_lock));
@@ -757,6 +875,108 @@ cpucaps_zone_set(zone_t *zone, rctl_qty_t cap_val)
 }
 
 /*
+ * Set zone's base cpu value to base_val
+ */
+int
+cpucaps_zone_set_base(zone_t *zone, rctl_qty_t base_val)
+{
+	cpucap_t *cap = NULL;
+	hrtime_t value;
+
+	ASSERT(base_val <= MAXCAP);
+	if (base_val > MAXCAP)
+		base_val = MAXCAP;
+
+	if (CPUCAPS_OFF() || !ZONE_IS_CAPPED(zone))
+		return (0);
+
+	if (zone->zone_cpucap == NULL)
+		cap = cap_alloc();
+
+	mutex_enter(&caps_lock);
+
+	if (cpucaps_busy) {
+		mutex_exit(&caps_lock);
+		return (EBUSY);
+	}
+
+	/*
+	 * Double-check whether zone->zone_cpucap is NULL, now with caps_lock
+	 * held. If it is still NULL, assign a newly allocated cpucap to it.
+	 */
+	if (zone->zone_cpucap == NULL) {
+		zone->zone_cpucap = cap;
+	} else if (cap != NULL) {
+		cap_free(cap);
+	}
+
+	cap = zone->zone_cpucap;
+
+	value = base_val * cap_tick_cost;
+	if (value < 0 || value > cap->cap_value)
+		value = 0;
+
+	cap->cap_base = value;
+
+	mutex_exit(&caps_lock);
+
+	return (0);
+}
+
+/*
+ * Set zone's maximum burst time in seconds.  A burst time of 0 means that
+ * the zone can run over its baseline indefinitely.
+ */
+int
+cpucaps_zone_set_burst_time(zone_t *zone, rctl_qty_t base_val)
+{
+	cpucap_t *cap = NULL;
+	hrtime_t value;
+
+	ASSERT(base_val <= INT_MAX);
+	/* Treat the default as 0 - no limit */
+	if (base_val == INT_MAX)
+		base_val = 0;
+	if (base_val > INT_MAX)
+		base_val = INT_MAX;
+
+	if (CPUCAPS_OFF() || !ZONE_IS_CAPPED(zone))
+		return (0);
+
+	if (zone->zone_cpucap == NULL)
+		cap = cap_alloc();
+
+	mutex_enter(&caps_lock);
+
+	if (cpucaps_busy) {
+		mutex_exit(&caps_lock);
+		return (EBUSY);
+	}
+
+	/*
+	 * Double-check whether zone->zone_cpucap is NULL, now with caps_lock
+	 * held. If it is still NULL, assign a newly allocated cpucap to it.
+	 */
+	if (zone->zone_cpucap == NULL) {
+		zone->zone_cpucap = cap;
+	} else if (cap != NULL) {
+		cap_free(cap);
+	}
+
+	cap = zone->zone_cpucap;
+
+	value = SEC_TO_TICK(base_val);
+	if (value < 0)
+		value = 0;
+
+	cap->cap_burst_limit = value;
+
+	mutex_exit(&caps_lock);
+
+	return (0);
+}
+
+/*
  * The project is going away so disable its cap.
  */
 void
@@ -902,7 +1122,7 @@ cpucaps_project_set(kproject_t *kpj, rctl_qty_t cap_val)
 		if (CAP_DISABLED(cap))
 			cap_project_enable(kpj, value);
 		else
-			cap->cap_value = value;
+			cap->cap_value = cap->cap_chk_value = value;
 	} else if (CAP_ENABLED(cap)) {
 		/*
 		 * User requested to drop a cap on the project. If it is part of
@@ -910,7 +1130,7 @@ cpucaps_project_set(kproject_t *kpj, rctl_qty_t cap_val)
 		 * otherwise disable the cap.
 		 */
 		if (ZONE_IS_CAPPED(kpj->kpj_zone)) {
-			cap->cap_value = MAX_USAGE;
+			cap->cap_value = cap->cap_chk_value = MAX_USAGE;
 		} else {
 			cap_project_disable(kpj);
 		}
@@ -948,6 +1168,26 @@ cpucaps_zone_get(zone_t *zone)
 }
 
 /*
+ * Get current zone baseline.
+ */
+rctl_qty_t
+cpucaps_zone_get_base(zone_t *zone)
+{
+	return (zone->zone_cpucap != NULL ?
+	    (rctl_qty_t)(zone->zone_cpucap->cap_base / cap_tick_cost) : 0);
+}
+
+/*
+ * Get current zone maximum burst time.
+ */
+rctl_qty_t
+cpucaps_zone_get_burst_time(zone_t *zone)
+{
+	return (zone->zone_cpucap != NULL ?
+	    (rctl_qty_t)(TICK_TO_SEC(zone->zone_cpucap->cap_burst_limit)) : 0);
+}
+
+/*
  * Charge project of thread t the time thread t spent on CPU since previously
  * adjusted.
  *
@@ -1045,7 +1285,7 @@ cpucaps_charge(kthread_id_t t, caps_sc_t *csc, cpucaps_charge_t charge_type)
 
 	project_cap = kpj->kpj_cpucap;
 
-	if (project_cap->cap_usage >= project_cap->cap_value) {
+	if (project_cap->cap_usage >= project_cap->cap_chk_value) {
 		t->t_schedflag |= TS_PROJWAITQ;
 		rc = B_TRUE;
 	} else if (t->t_schedflag & TS_PROJWAITQ) {
@@ -1059,7 +1299,7 @@ cpucaps_charge(kthread_id_t t, caps_sc_t *csc, cpucaps_charge_t charge_type)
 	} else {
 		cpucap_t *zone_cap = zone->zone_cpucap;
 
-		if (zone_cap->cap_usage >= zone_cap->cap_value) {
+		if (zone_cap->cap_usage >= zone_cap->cap_chk_value) {
 			t->t_schedflag |= TS_ZONEWAITQ;
 			rc = B_TRUE;
 		} else if (t->t_schedflag & TS_ZONEWAITQ) {
@@ -1119,6 +1359,7 @@ cpucaps_enforce(kthread_t *t)
 
 /*
  * Convert internal cap statistics into values exported by cap kstat.
+ * Note that the kstat is held throughout this function but caps_lock is not.
  */
 static int
 cap_kstat_update(kstat_t *ksp, int rw)
@@ -1133,6 +1374,12 @@ cap_kstat_update(kstat_t *ksp, int rw)
 
 	capsp->cap_value.value.ui64 =
 	    ROUND_SCALE(cap->cap_value, cap_tick_cost);
+	capsp->cap_baseline.value.ui64 =
+	    ROUND_SCALE(cap->cap_base, cap_tick_cost);
+	capsp->cap_effective.value.ui64 =
+	    ROUND_SCALE(cap->cap_chk_value, cap_tick_cost);
+	capsp->cap_burst_limit.value.ui64 =
+	    ROUND_SCALE(cap->cap_burst_limit, tick_sec);
 	capsp->cap_usage.value.ui64 =
 	    ROUND_SCALE(cap->cap_usage, cap_tick_cost);
 	capsp->cap_maxusage.value.ui64 =
@@ -1140,6 +1387,10 @@ cap_kstat_update(kstat_t *ksp, int rw)
 	capsp->cap_nwait.value.ui64 = cap->cap_waitq.wq_count;
 	capsp->cap_below.value.ui64 = ROUND_SCALE(cap->cap_below, tick_sec);
 	capsp->cap_above.value.ui64 = ROUND_SCALE(cap->cap_above, tick_sec);
+	capsp->cap_above_base.value.ui64 =
+	    ROUND_SCALE(cap->cap_above_base, tick_sec);
+	capsp->cap_bursting.value.ui64 =
+	    ROUND_SCALE(cap->cap_bursting, tick_sec);
 	kstat_named_setstr(&capsp->cap_zonename, zonename);
 
 	return (0);
diff --git a/usr/src/uts/common/disp/disp.c b/usr/src/uts/common/disp/disp.c
index 0c2c0b4993..5f9c2c68a2 100644
--- a/usr/src/uts/common/disp/disp.c
+++ b/usr/src/uts/common/disp/disp.c
@@ -23,6 +23,10 @@
  * Use is subject to license terms.
  */
 
+/*
+ * Copyright (c) 2012, Joyent, Inc. All rights reserved.
+ */
+
 /*	Copyright (c) 1984, 1986, 1987, 1988, 1989 AT&T	*/
 /*	  All Rights Reserved  	*/
 
@@ -105,7 +109,7 @@ static void	cpu_resched(cpu_t *cp, pri_t tpri);
 /*
  * If this is set, only interrupt threads will cause kernel preemptions.
  * This is done by changing the value of kpreemptpri.  kpreemptpri
- * will either be the max sysclass pri + 1 or the min interrupt pri.
+ * will either be the max sysclass pri or the min interrupt pri.
  */
 int	only_intr_kpreempt;
 
@@ -252,7 +256,23 @@ dispinit(void)
 				maxglobpri = cl_maxglobpri;
 		}
 	}
-	kpreemptpri = (pri_t)v.v_maxsyspri + 1;
+
+	/*
+	 * Historically, kpreemptpri was set to v_maxsyspri + 1 -- which is
+	 * to say, maxclsyspri + 1.  However, over time, the system has used
+	 * more and more asynchronous kernel threads, with an increasing number
+	 * of these doing work on direct behalf of higher-level software (e.g.,
+	 * network processing).  This has led to potential priority inversions:
+	 * threads doing low-priority lengthy kernel work can effectively
+	 * delay kernel-level processing of higher-priority data. To minimize
+	 * such inversions, we set kpreemptpri to be v_maxsyspri; anything in
+	 * the kernel that runs at maxclsyspri will therefore induce kernel
+	 * preemption, and this priority should be used if/when an asynchronous
+	 * thread (or, as is often the case, task queue) is performing a task
+	 * on behalf of higher-level software (or any task that is otherwise
+	 * latency-sensitve).
+	 */
+	kpreemptpri = (pri_t)v.v_maxsyspri;
 	if (kpqpri == KPQPRI)
 		kpqpri = kpreemptpri;
 
@@ -2258,7 +2278,7 @@ disp_getbest(disp_t *dp)
 		 * placed earlier.
 		 */
 		if (tcp == NULL ||
-		    pri >= minclsyspri ||
+		    (pri >= minclsyspri && tp->t_procp == &p0) ||
 		    tp->t_cpu != tcp)
 			break;
 
diff --git a/usr/src/uts/common/disp/thread.c b/usr/src/uts/common/disp/thread.c
index f2685af534..ae6c5eef16 100644
--- a/usr/src/uts/common/disp/thread.c
+++ b/usr/src/uts/common/disp/thread.c
@@ -21,7 +21,7 @@
 
 /*
  * Copyright (c) 1991, 2010, Oracle and/or its affiliates. All rights reserved.
- * Copyright (c) 2013, Joyent, Inc.  All rights reserved.
+ * Copyright (c) 2015, Joyent, Inc.  All rights reserved.
  */
 
 #include <sys/types.h>
@@ -75,6 +75,10 @@
 #include <sys/cpucaps.h>
 #include <sys/kiconv.h>
 
+#ifndef	STACK_GROWTH_DOWN
+#error Stacks do not grow downward; 3b2 zombie attack detected!
+#endif
+
 struct kmem_cache *thread_cache;	/* cache of free threads */
 struct kmem_cache *lwp_cache;		/* cache of free lwps */
 struct kmem_cache *turnstile_cache;	/* cache of free turnstiles */
@@ -372,7 +376,7 @@ thread_create(
 		if (stksize <= sizeof (kthread_t) + PTR24_ALIGN)
 			cmn_err(CE_PANIC, "thread_create: proposed stack size"
 			    " too small to hold thread.");
-#ifdef STACK_GROWTH_DOWN
+
 		stksize -= SA(sizeof (kthread_t) + PTR24_ALIGN - 1);
 		stksize &= -PTR24_ALIGN;	/* make thread aligned */
 		t = (kthread_t *)(stk + stksize);
@@ -381,13 +385,6 @@ thread_create(
 			audit_thread_create(t);
 		t->t_stk = stk + stksize;
 		t->t_stkbase = stk;
-#else	/* stack grows to larger addresses */
-		stksize -= SA(sizeof (kthread_t));
-		t = (kthread_t *)(stk);
-		bzero(t, sizeof (kthread_t));
-		t->t_stk = stk + sizeof (kthread_t);
-		t->t_stkbase = stk + stksize + sizeof (kthread_t);
-#endif	/* STACK_GROWTH_DOWN */
 		t->t_flag |= T_TALLOCSTK;
 		t->t_swap = stk;
 	} else {
@@ -400,13 +397,8 @@ thread_create(
 		 * Initialize t_stk to the kernel stack pointer to use
 		 * upon entry to the kernel
 		 */
-#ifdef STACK_GROWTH_DOWN
 		t->t_stk = stk + stksize;
 		t->t_stkbase = stk;
-#else
-		t->t_stk = stk;			/* 3b2-like */
-		t->t_stkbase = stk + stksize;
-#endif /* STACK_GROWTH_DOWN */
 	}
 
 	if (kmem_stackinfo != 0) {
@@ -589,6 +581,9 @@ thread_exit(void)
 	if ((t->t_proc_flag & TP_ZTHREAD) != 0)
 		cmn_err(CE_PANIC, "thread_exit: zthread_exit() not called");
 
+	if ((t->t_flag & T_SPLITSTK) != 0)
+		cmn_err(CE_PANIC, "thread_exit: called when stack is split");
+
 	tsd_exit();		/* Clean up this thread's TSD */
 
 	kcpc_passivate();	/* clean up performance counter state */
@@ -1050,6 +1045,8 @@ installctx(
 	ctx->free_op = free;
 	ctx->arg = arg;
 	ctx->next = t->t_ctx;
+	ctx->save_ts = 0;
+	ctx->restore_ts = 0;
 	t->t_ctx = ctx;
 }
 
@@ -1124,9 +1121,12 @@ savectx(kthread_t *t)
 	struct ctxop *ctx;
 
 	ASSERT(t == curthread);
-	for (ctx = t->t_ctx; ctx != 0; ctx = ctx->next)
-		if (ctx->save_op != NULL)
+	for (ctx = t->t_ctx; ctx != 0; ctx = ctx->next) {
+		if (ctx->save_op != NULL) {
+			ctx->save_ts = gethrtime_unscaled();
 			(ctx->save_op)(ctx->arg);
+		}
+	}
 }
 
 void
@@ -1135,9 +1135,12 @@ restorectx(kthread_t *t)
 	struct ctxop *ctx;
 
 	ASSERT(t == curthread);
-	for (ctx = t->t_ctx; ctx != 0; ctx = ctx->next)
-		if (ctx->restore_op != NULL)
+	for (ctx = t->t_ctx; ctx != 0; ctx = ctx->next) {
+		if (ctx->restore_op != NULL) {
+			ctx->restore_ts = gethrtime_unscaled();
 			(ctx->restore_op)(ctx->arg);
+		}
+	}
 }
 
 void
@@ -1883,6 +1886,103 @@ thread_change_pri(kthread_t *t, pri_t disp_pri, int front)
 	return (on_rq);
 }
 
+
+/*
+ * There are occasions in the kernel when we need much more stack than we
+ * allocate by default, but we do not wish to have that work done
+ * asynchronously by another thread.  To accommodate these scenarios, we allow
+ * for a split stack (also known as a "segmented stack") whereby a new stack
+ * is dynamically allocated and the current thread jumps onto it for purposes
+ * of executing the specified function.  After the specified function returns,
+ * the stack is deallocated and control is returned to the caller.  This
+ * functionality is implemented by thread_splitstack(), below; there are a few
+ * constraints on its use:
+ *
+ * - The caller must be in a context where it is safe to block for memory.
+ * - The caller cannot be in a t_onfault context
+ * - The called function must not call thread_exit() while on the split stack
+ *
+ * The code will explicitly panic if these constraints are violated.  Notably,
+ * however, thread_splitstack() _can_ be called on a split stack -- there
+ * is no limit to the level that split stacks can nest.
+ *
+ * When the stack is split, it is constructed such that stack backtraces
+ * from kernel debuggers continue to function -- though note that DTrace's
+ * stack() action and stackdepth function will only show the stack up to and
+ * including thread_splitstack_run(); DTrace explicitly bounds itself to
+ * pointers that exist within the current declared stack as a safety
+ * mechanism.
+ */
+void
+thread_splitstack(void (*func)(void *), void *arg, size_t stksize)
+{
+	kthread_t *t = curthread;
+	caddr_t ostk, ostkbase, stk;
+	ushort_t otflag;
+
+	if (t->t_onfault != NULL)
+		panic("thread_splitstack: called with non-NULL t_onfault");
+
+	ostk = t->t_stk;
+	ostkbase = t->t_stkbase;
+	otflag = t->t_flag;
+
+	stksize = roundup(stksize, PAGESIZE);
+
+	if (stksize < default_stksize)
+		stksize = default_stksize;
+
+	if (stksize == default_stksize) {
+		stk = (caddr_t)segkp_cache_get(segkp_thread);
+	} else {
+		stksize = roundup(stksize, PAGESIZE);
+		stk = (caddr_t)segkp_get(segkp, stksize,
+		    (KPD_HASREDZONE | KPD_NO_ANON | KPD_LOCKED));
+	}
+
+	/*
+	 * We're going to lock ourselves before we set T_SPLITSTK to assure
+	 * that we're not swapped out in the meantime.  (Note that we don't
+	 * bother to set t_swap, as we're not going to be swapped out.)
+	 */
+	thread_lock(t);
+
+	if (!(otflag & T_SPLITSTK))
+		t->t_flag |= T_SPLITSTK;
+
+	t->t_stk = stk + stksize;
+	t->t_stkbase = stk;
+
+	thread_unlock(t);
+
+	/*
+	 * Now actually run on the new (split) stack...
+	 */
+	thread_splitstack_run(t->t_stk, func, arg);
+
+	/*
+	 * We're back onto our own stack; lock ourselves and restore our
+	 * pre-split state.
+	 */
+	thread_lock(t);
+
+	t->t_stk = ostk;
+	t->t_stkbase = ostkbase;
+
+	if (!(otflag & T_SPLITSTK))
+		t->t_flag &= ~T_SPLITSTK;
+
+	thread_unlock(t);
+
+	/*
+	 * Now that we are entirely back on our own stack, call back into
+	 * the platform layer to perform any platform-specific cleanup.
+	 */
+	thread_splitstack_cleanup();
+
+	segkp_release(segkp, stk);
+}
+
 /*
  * Tunable kmem_stackinfo is set, fill the kernel thread stack with a
  * specific pattern.
diff --git a/usr/src/uts/common/disp/thread_intr.c b/usr/src/uts/common/disp/thread_intr.c
index 67ccc6922f..c840bdf31a 100644
--- a/usr/src/uts/common/disp/thread_intr.c
+++ b/usr/src/uts/common/disp/thread_intr.c
@@ -23,19 +23,10 @@
  * Copyright 2006 Sun Microsystems, Inc.  All rights reserved.
  * Use is subject to license terms.
  */
-
 /*
- * FILE NOTICE BEGIN
- *
- * This file should not be modified.  If you wish to modify it or have it
- * modified, please contact Sun Microsystems at <LFI149367@-sun-.-com->
- * (without anti-spam dashes)
- *
- * FILE NOTICE END
+ * Copyright 2015, Joyent, Inc.
  */
 
-#pragma ident	"%Z%%M%	%I%	%E% SMI"
-
 #include <sys/cpuvar.h>
 #include <sys/stack.h>
 #include <vm/seg_kp.h>
@@ -44,6 +35,17 @@
 #include <sys/sysmacros.h>
 
 /*
+ * Use a slightly larger thread stack size for interrupt threads rather than the
+ * default. This is useful for cases where the networking stack may do an rx and
+ * a tx in the context of a single interrupt and when combined with various
+ * promisc hooks that need memory, can cause us to get dangerously close to the
+ * edge of the traditional stack sizes. This is only a few pages more than a
+ * traditional stack and given that we don't have that many interrupt threads,
+ * the memory costs end up being more than worthwhile.
+ */
+#define	LL_INTR_STKSZ	(32 * 1024)
+
+/*
  * Create and initialize an interrupt thread.
  */
 static void
@@ -51,7 +53,7 @@ thread_create_intr(cpu_t *cp)
 {
 	kthread_t *tp;
 
-	tp = thread_create(NULL, 0,
+	tp = thread_create(NULL, LL_INTR_STKSZ,
 	    (void (*)())thread_create_intr, NULL, 0, &p0, TS_ONPROC, 0);
 
 	/*
@@ -97,9 +99,12 @@ thread_create_intr(cpu_t *cp)
 }
 
 /*
- * Allocate a given number of interrupt threads for a given CPU.
- * These threads will get freed by cpu_destroy_bound_threads()
- * when CPU gets unconfigured.
+ * Allocate a given number of interrupt threads for a given CPU.  These threads
+ * will get freed by cpu_destroy_bound_threads() when CPU gets unconfigured.
+ *
+ * Note, high level interrupts are always serviced using cpu_intr_stack and are
+ * not allowed to block. Low level interrupts or soft-interrupts use the
+ * kthread_t's that we create through the calls to thread_create_intr().
  */
 void
 cpu_intr_alloc(cpu_t *cp, int n)
@@ -110,6 +115,6 @@ cpu_intr_alloc(cpu_t *cp, int n)
 		thread_create_intr(cp);
 
 	cp->cpu_intr_stack = (caddr_t)segkp_get(segkp, INTR_STACK_SIZE,
-		KPD_HASREDZONE | KPD_NO_ANON | KPD_LOCKED) +
-		INTR_STACK_SIZE - SA(MINFRAME);
+	    KPD_HASREDZONE | KPD_NO_ANON | KPD_LOCKED) +
+	    INTR_STACK_SIZE - SA(MINFRAME);
 }
diff --git a/usr/src/uts/common/dtrace/dtrace.c b/usr/src/uts/common/dtrace/dtrace.c
index aa4b1d67e0..1a7876ca57 100644
--- a/usr/src/uts/common/dtrace/dtrace.c
+++ b/usr/src/uts/common/dtrace/dtrace.c
@@ -785,8 +785,8 @@ dtrace_canload(uint64_t addr, size_t sz, dtrace_mstate_t *mstate,
 			if (DTRACE_INRANGE(addr, sz, &vp->v_path, psz))
 				return (1);
 
-			if (vp->v_path != NULL && DTRACE_INRANGE(addr, sz,
-			    vp->v_path, strlen(vp->v_path) + 1)) {
+			if (DTRACE_INRANGE(addr, sz, vp->v_path,
+			    strlen(vp->v_path) + 1)) {
 				return (1);
 			}
 
@@ -845,7 +845,7 @@ static int
 dtrace_vcanload(void *src, dtrace_diftype_t *type, dtrace_mstate_t *mstate,
     dtrace_vstate_t *vstate)
 {
-	size_t sz;
+	size_t sz, strsize;
 	ASSERT(type->dtdt_flags & DIF_TF_BYREF);
 
 	/*
@@ -855,11 +855,24 @@ dtrace_vcanload(void *src, dtrace_diftype_t *type, dtrace_mstate_t *mstate,
 	if ((mstate->dtms_access & DTRACE_ACCESS_KERNEL) != 0)
 		return (1);
 
-	if (type->dtdt_kind == DIF_TYPE_STRING)
-		sz = dtrace_strlen(src,
-		    vstate->dtvs_state->dts_options[DTRACEOPT_STRSIZE]) + 1;
-	else
+	if (type->dtdt_kind == DIF_TYPE_STRING) {
+		dtrace_state_t *state = vstate->dtvs_state;
+
+		if (state != NULL) {
+			strsize = state->dts_options[DTRACEOPT_STRSIZE];
+		} else {
+			/*
+			 * In helper context, we have a NULL state; fall back
+			 * to using the system-wide default for the string size
+			 * in this case.
+			 */
+			strsize = dtrace_strsize_default;
+		}
+
+		sz = dtrace_strlen(src, strsize) + 1;
+	} else {
 		sz = type->dtdt_size;
+	}
 
 	return (dtrace_canload((uintptr_t)src, sz, mstate, vstate));
 }
@@ -7506,7 +7519,7 @@ dtrace_cred2priv(cred_t *cr, uint32_t *privp, uid_t *uidp, zoneid_t *zoneidp)
 		priv = DTRACE_PRIV_ALL;
 	} else {
 		*uidp = crgetuid(cr);
-		*zoneidp = crgetzoneid(cr);
+		*zoneidp = crgetzonedid(cr);
 
 		priv = 0;
 		if (PRIV_POLICY_ONLY(cr, PRIV_DTRACE_KERNEL, B_FALSE))
@@ -8002,7 +8015,7 @@ dtrace_register(const char *name, const dtrace_pattr_t *pap, uint32_t priv,
 	provider->dtpv_priv.dtpp_flags = priv;
 	if (cr != NULL) {
 		provider->dtpv_priv.dtpp_uid = crgetuid(cr);
-		provider->dtpv_priv.dtpp_zoneid = crgetzoneid(cr);
+		provider->dtpv_priv.dtpp_zoneid = crgetzonedid(cr);
 	}
 	provider->dtpv_pops = *pops;
 
@@ -8613,6 +8626,7 @@ dtrace_probe_enable(const dtrace_probedesc_t *desc, dtrace_enabling_t *enab)
 	uint32_t priv;
 	uid_t uid;
 	zoneid_t zoneid;
+	dtrace_state_t *state = enab->dten_vstate->dtvs_state;
 
 	ASSERT(MUTEX_HELD(&dtrace_lock));
 	dtrace_ecb_create_cache = NULL;
@@ -8627,8 +8641,22 @@ dtrace_probe_enable(const dtrace_probedesc_t *desc, dtrace_enabling_t *enab)
 	}
 
 	dtrace_probekey(desc, &pkey);
-	dtrace_cred2priv(enab->dten_vstate->dtvs_state->dts_cred.dcr_cred,
-	    &priv, &uid, &zoneid);
+	dtrace_cred2priv(state->dts_cred.dcr_cred, &priv, &uid, &zoneid);
+
+	if ((priv & DTRACE_PRIV_ZONEOWNER) &&
+	    state->dts_options[DTRACEOPT_ZONE] != DTRACEOPT_UNSET) {
+		/*
+		 * If we have the privilege of instrumenting all zones but we
+		 * have been told to instrument but one, we will spoof this up
+		 * depriving ourselves of DTRACE_PRIV_ZONEOWNER for purposes
+		 * of dtrace_match().  (Note that DTRACEOPT_ZONE is not for
+		 * security but rather for performance: it allows the global
+		 * zone to instrument USDT probes in a local zone without
+		 * requiring all zones to be instrumented.)
+		 */
+		priv &= ~DTRACE_PRIV_ZONEOWNER;
+		zoneid = state->dts_options[DTRACEOPT_ZONE];
+	}
 
 	return (dtrace_match(&pkey, priv, uid, zoneid, dtrace_ecb_create_enable,
 	    enab));
diff --git a/usr/src/uts/common/dtrace/sdt_subr.c b/usr/src/uts/common/dtrace/sdt_subr.c
index 157acc25fc..3d350ff278 100644
--- a/usr/src/uts/common/dtrace/sdt_subr.c
+++ b/usr/src/uts/common/dtrace/sdt_subr.c
@@ -97,6 +97,10 @@ static dtrace_pattr_t iscsi_attr = {
 { DTRACE_STABILITY_EVOLVING, DTRACE_STABILITY_EVOLVING, DTRACE_CLASS_ISA },
 };
 
+/*
+ * When adding a new provider you must add it before sdt as sdt is a catch all
+ * for remaining probes.
+ */
 sdt_provider_t sdt_providers[] = {
 	{ "vtrace", "__vtrace_", &vtrace_attr },
 	{ "sysinfo", "__cpu_sysinfo_", &info_attr, DTRACE_PRIV_USER },
@@ -117,6 +121,7 @@ sdt_provider_t sdt_providers[] = {
 	{ "fc", "__fc_", &fc_attr },
 	{ "srp", "__srp_", &fc_attr },
 	{ "sysevent", "__sysevent_", &stab_attr },
+	{ "vnd", "__vnd_", &stab_attr },
 	{ "sdt", NULL, &sdt_attr },
 	{ NULL }
 };
@@ -1151,6 +1156,34 @@ sdt_argdesc_t sdt_args[] = {
 	{ "fc", "abts-receive", 2, 2, "fct_i_remote_port_t *",
 	    "fc_port_info_t *" },
 
+	{ "vnd", "flow-blocked", 0, 0, "vnd_str_t *", "ifinfo_t *" },
+	{ "vnd", "flow-blocked", 1, 1, "uint64_t", "uint64_t" },
+	{ "vnd", "flow-blocked", 2, 2, "uintptr_t", "uintptr_t" },
+	{ "vnd", "flow-resumed", 0, 0, "vnd_str_t *", "ifinfo_t *" },
+	{ "vnd", "flow-resumed", 1, 1, "uint64_t", "uint64_t" },
+	{ "vnd", "flow-resumed", 2, 2, "uintptr_t", "uintptr_t" },
+	{ "vnd", "drop-in", 0, 0, "mblk_t *", "pktinfo_t *" },
+	{ "vnd", "drop-in", 1, 1, "vnd_str_t *", "ifinfo_t *" },
+	{ "vnd", "drop-in", 2, 2, "mblk_t *", "etherinfo_t *" },
+	{ "vnd", "drop-in", 3, 3, "const char *", "const char *" },
+	{ "vnd", "drop-out", 0, 0, "mblk_t *", "pktinfo_t *" },
+	{ "vnd", "drop-out", 1, 1, "vnd_str_t *", "ifinfo_t *" },
+	{ "vnd", "drop-out", 2, 2, "mblk_t *", "etherinfo_t *" },
+	{ "vnd", "drop-out", 3, 3, "const char *", "const char *" },
+	{ "vnd", "drop-ctl", 0, 0, "mblk_t *", "pktinfo_t *" },
+	{ "vnd", "drop-ctl", 1, 1, "vnd_str_t *", "ifinfo_t *" },
+	{ "vnd", "drop-ctl", 2, 2, "mblk_t *", "etherinfo_t *" },
+	{ "vnd", "drop-ctl", 3, 3, "const char *", "const char *" },
+	{ "vnd", "send", 0, 0, "mblk_t *", "pktinfo_t *" },
+	{ "vnd", "send", 1, 1, "void *", "csinfo_t *" },
+	{ "vnd", "send", 2, 2, "void *", "ipinfo_t *" },
+	{ "vnd", "send", 3, 3, "vnd_str_t *", "ifinfo_t *" },
+	{ "vnd", "send", 4, 4, "mblk_t *", "etherinfo_t *" },
+	{ "vnd", "recv", 0, 0, "mblk_t *", "pktinfo_t *" },
+	{ "vnd", "recv", 1, 1, "void *", "csinfo_t *" },
+	{ "vnd", "recv", 2, 2, "void *", "ipinfo_t *" },
+	{ "vnd", "recv", 3, 3, "vnd_str_t *", "ifinfo_t *" },
+	{ "vnd", "recv", 4, 4, "mblk_t *", "etherinfo_t *" },
 
 	{ NULL }
 };
diff --git a/usr/src/uts/common/exec/aout/aout.c b/usr/src/uts/common/exec/aout/aout.c
index fc45bd9544..5dbb2ed28c 100644
--- a/usr/src/uts/common/exec/aout/aout.c
+++ b/usr/src/uts/common/exec/aout/aout.c
@@ -22,6 +22,7 @@
  * Copyright 2008 Sun Microsystems, Inc.  All rights reserved.
  * Use is subject to license terms.
  * Copyright (c) 2011 Bayard G. Bell. All rights reserved.
+ * Copyright 2015, Joyent, Inc.
  */
 
 #include <sys/types.h>
@@ -54,7 +55,7 @@
 
 static int aoutexec(vnode_t *vp, execa_t *uap, uarg_t *args,
     intpdata_t *idatap, int level, long *execsz, int setid,
-    caddr_t exec_file, cred_t *cred, int brand_action);
+    caddr_t exec_file, cred_t *cred, int *brand_action);
 static int get_aout_head(struct vnode **vpp, struct exdata *edp, long *execsz,
     int *isdyn);
 static int aoutcore(vnode_t *vp, proc_t *pp, cred_t *credp,
@@ -130,7 +131,7 @@ _info(struct modinfo *modinfop)
 static int
 aoutexec(vnode_t *vp, struct execa *uap, struct uarg *args,
     struct intpdata *idatap, int level, long *execsz, int setid,
-    caddr_t exec_file, cred_t *cred, int brand_action)
+    caddr_t exec_file, cred_t *cred, int *brand_action)
 {
 	auxv32_t auxflags_auxv32;
 	int error;
diff --git a/usr/src/uts/common/exec/elf/elf.c b/usr/src/uts/common/exec/elf/elf.c
index 0eeb4e798c..936a3f1da1 100644
--- a/usr/src/uts/common/exec/elf/elf.c
+++ b/usr/src/uts/common/exec/elf/elf.c
@@ -26,7 +26,7 @@
 /*	Copyright (c) 1984, 1986, 1987, 1988, 1989 AT&T	*/
 /*	  All Rights Reserved  	*/
 /*
- * Copyright (c) 2013, Joyent, Inc.  All rights reserved.
+ * Copyright (c) 2015, Joyent, Inc. All rights reserved.
  */
 
 #include <sys/types.h>
@@ -163,12 +163,16 @@ dtrace_safe_phdr(Phdr *phdrp, struct uarg *args, uintptr_t base)
 }
 
 /*
- * Map in the executable pointed to by vp. Returns 0 on success.
+ * Map in the executable pointed to by vp. Returns 0 on success.  Note that
+ * this function currently has the maximum number of arguments allowed by
+ * modstubs on x86 (MAXNARG)!  Do _not_ add to this function signature without
+ * adding to MAXNARG.  (Better yet, do not add to this monster of a function
+ * signature!)
  */
 int
 mapexec_brand(vnode_t *vp, uarg_t *args, Ehdr *ehdr, Addr *uphdr_vaddr,
-    intptr_t *voffset, caddr_t exec_file, int *interp, caddr_t *bssbase,
-    caddr_t *brkbase, size_t *brksize, uintptr_t *lddatap)
+    intptr_t *voffset, caddr_t exec_file, char **interpp, caddr_t *bssbase,
+    caddr_t *brkbase, size_t *brksize, uintptr_t *lddatap, uintptr_t *minaddrp)
 {
 	size_t		len;
 	struct vattr	vat;
@@ -180,6 +184,7 @@ mapexec_brand(vnode_t *vp, uarg_t *args, Ehdr *ehdr, Addr *uphdr_vaddr,
 	Phdr		*junk = NULL;
 	Phdr		*dynphdr = NULL;
 	Phdr		*dtrphdr = NULL;
+	char		*interp = NULL;
 	uintptr_t	lddata;
 	long		execsz;
 	intptr_t	minaddr;
@@ -187,6 +192,9 @@ mapexec_brand(vnode_t *vp, uarg_t *args, Ehdr *ehdr, Addr *uphdr_vaddr,
 	if (lddatap != NULL)
 		*lddatap = NULL;
 
+	if (minaddrp != NULL)
+		*minaddrp = NULL;
+
 	if (error = execpermissions(vp, &vat, args)) {
 		uprintf("%s: Cannot execute %s\n", exec_file, args->pathname);
 		return (error);
@@ -212,25 +220,65 @@ mapexec_brand(vnode_t *vp, uarg_t *args, Ehdr *ehdr, Addr *uphdr_vaddr,
 	    &junk, &dtrphdr, NULL, bssbase, brkbase, voffset, &minaddr,
 	    len, &execsz, brksize)) {
 		uprintf("%s: Cannot map %s\n", exec_file, args->pathname);
+		if (uphdr != NULL && uphdr->p_flags == 0)
+			kmem_free(uphdr, sizeof (Phdr));
 		kmem_free(phdrbase, phdrsize);
 		return (error);
 	}
 
+	if (minaddrp != NULL)
+		*minaddrp = minaddr;
+
 	/*
-	 * Inform our caller if the executable needs an interpreter.
+	 * If the executable requires an interpreter, determine its name.
 	 */
-	*interp = (dynphdr == NULL) ? 0 : 1;
+	if (dynphdr != NULL) {
+		ssize_t	resid;
+
+		if (dynphdr->p_filesz > MAXPATHLEN || dynphdr->p_filesz == 0) {
+			uprintf("%s: Invalid interpreter\n", exec_file);
+			kmem_free(phdrbase, phdrsize);
+			return (ENOEXEC);
+		}
+
+		interp = kmem_alloc(MAXPATHLEN, KM_SLEEP);
+
+		if ((error = vn_rdwr(UIO_READ, vp, interp, dynphdr->p_filesz,
+		    (offset_t)dynphdr->p_offset, UIO_SYSSPACE, 0,
+		    (rlim64_t)0, CRED(), &resid)) != 0 || resid != 0 ||
+		    interp[dynphdr->p_filesz - 1] != '\0') {
+			uprintf("%s: Cannot obtain interpreter pathname\n",
+			    exec_file);
+			kmem_free(interp, MAXPATHLEN);
+			kmem_free(phdrbase, phdrsize);
+			return (error != 0 ? error : ENOEXEC);
+		}
+	}
 
 	/*
 	 * If this is a statically linked executable, voffset should indicate
 	 * the address of the executable itself (it normally holds the address
 	 * of the interpreter).
 	 */
-	if (ehdr->e_type == ET_EXEC && *interp == 0)
+	if (ehdr->e_type == ET_EXEC && interp == NULL)
 		*voffset = minaddr;
 
+	/*
+	 * If the caller has asked for the interpreter name, return it (it's
+	 * up to the caller to free it); if the caller hasn't asked for it,
+	 * free it ourselves.
+	 */
+	if (interpp != NULL) {
+		*interpp = interp;
+	} else if (interp != NULL) {
+		kmem_free(interp, MAXPATHLEN);
+	}
+
 	if (uphdr != NULL) {
 		*uphdr_vaddr = uphdr->p_vaddr;
+
+		if (uphdr->p_flags == 0)
+			kmem_free(uphdr, sizeof (Phdr));
 	} else {
 		*uphdr_vaddr = (Addr)-1;
 	}
@@ -243,13 +291,13 @@ mapexec_brand(vnode_t *vp, uarg_t *args, Ehdr *ehdr, Addr *uphdr_vaddr,
 int
 elfexec(vnode_t *vp, execa_t *uap, uarg_t *args, intpdata_t *idatap,
     int level, long *execsz, int setid, caddr_t exec_file, cred_t *cred,
-    int brand_action)
+    int *brand_action)
 {
 	caddr_t		phdrbase = NULL;
 	caddr_t 	bssbase = 0;
 	caddr_t 	brkbase = 0;
 	size_t		brksize = 0;
-	ssize_t		dlnsize;
+	ssize_t		dlnsize, nsize = 0;
 	aux_entry_t	*aux;
 	int		error;
 	ssize_t		resid;
@@ -273,6 +321,7 @@ elfexec(vnode_t *vp, execa_t *uap, uarg_t *args, intpdata_t *idatap,
 	int		hasauxv = 0;
 	int		hasdy = 0;
 	int		branded = 0;
+	int		dynuphdr = 0;
 
 	struct proc *p = ttoproc(curthread);
 	struct user *up = PTOU(p);
@@ -339,14 +388,40 @@ elfexec(vnode_t *vp, execa_t *uap, uarg_t *args, intpdata_t *idatap,
 #endif	/* _LP64 */
 
 	/*
-	 * We delay invoking the brand callback until we've figured out
-	 * what kind of elf binary we're trying to run, 32-bit or 64-bit.
-	 * We do this because now the brand library can just check
-	 * args->to_model to see if the target is 32-bit or 64-bit without
-	 * having do duplicate all the code above.
+	 * We delay invoking the brand callback until we've figured out what
+	 * kind of elf binary we're trying to run, 32-bit or 64-bit.  We do this
+	 * because now the brand library can just check args->to_model to see if
+	 * the target is 32-bit or 64-bit without having do duplicate all the
+	 * code above.
+	 *
+	 * We also give the brand a chance to indicate that based on the ELF
+	 * OSABI of the target binary it should become unbranded and optionally
+	 * indicate that it should be treated as existing in a specific prefix.
+	 *
+	 * Note that if a brand opts to go down this route it does not actually
+	 * end up being debranded. In other words, future programs that exec
+	 * will still be considered for branding unless this escape hatch is
+	 * used. Consider the case of lx brand for example. If a user runs
+	 * /native/usr/sbin/dtrace -c /bin/ls, the isaexec and normal executable
+	 * of DTrace that's in /native will take this escape hatch and be run
+	 * and interpreted using the normal system call table; however, the
+	 * execution of a non-illumos binary in the form of /bin/ls will still
+	 * be branded and be subject to all of the normal actions of the brand.
 	 */
+	if ((level < 2) && (*brand_action != EBA_NATIVE) &&
+	    (PROC_IS_BRANDED(p)) && (BROP(p)->b_native_exec != NULL)) {
+		if (BROP(p)->b_native_exec(ehdrp->e_ident[EI_OSABI],
+		    &args->brand_nroot) == B_TRUE) {
+			ASSERT(ehdrp->e_ident[EI_OSABI]);
+			*brand_action = EBA_NATIVE;
+			/* Add one for the trailing '/' in the path */
+			if (args->brand_nroot != NULL)
+				nsize = strlen(args->brand_nroot) + 1;
+		}
+	}
+
 	if ((level < 2) &&
-	    (brand_action != EBA_NATIVE) && (PROC_IS_BRANDED(p))) {
+	    (*brand_action != EBA_NATIVE) && (PROC_IS_BRANDED(p))) {
 		error = BROP(p)->b_elfexec(vp, uap, args,
 		    idatap, level + 1, execsz, setid, exec_file, cred,
 		    brand_action);
@@ -417,14 +492,15 @@ elfexec(vnode_t *vp, execa_t *uap, uarg_t *args, intpdata_t *idatap,
 		 *	AT_BASE
 		 *	AT_FLAGS
 		 *	AT_PAGESZ
+		 *	AT_RANDOM	(added in stk_copyout)
 		 *	AT_SUN_AUXFLAGS
 		 *	AT_SUN_HWCAP
 		 *	AT_SUN_HWCAP2
-		 *	AT_SUN_PLATFORM (added in stk_copyout)
-		 *	AT_SUN_EXECNAME (added in stk_copyout)
+		 *	AT_SUN_PLATFORM	(added in stk_copyout)
+		 *	AT_SUN_EXECNAME	(added in stk_copyout)
 		 *	AT_NULL
 		 *
-		 * total == 9
+		 * total == 10
 		 */
 		if (hasdy && hasu) {
 			/*
@@ -439,7 +515,7 @@ elfexec(vnode_t *vp, execa_t *uap, uarg_t *args, intpdata_t *idatap,
 			 *
 			 * total = 5
 			 */
-			args->auxsize = (9 + 5) * sizeof (aux_entry_t);
+			args->auxsize = (10 + 5) * sizeof (aux_entry_t);
 		} else if (hasdy) {
 			/*
 			 * Has PT_INTERP but no PT_PHDR
@@ -449,9 +525,9 @@ elfexec(vnode_t *vp, execa_t *uap, uarg_t *args, intpdata_t *idatap,
 			 *
 			 * total = 2
 			 */
-			args->auxsize = (9 + 2) * sizeof (aux_entry_t);
+			args->auxsize = (10 + 2) * sizeof (aux_entry_t);
 		} else {
-			args->auxsize = 9 * sizeof (aux_entry_t);
+			args->auxsize = 10 * sizeof (aux_entry_t);
 		}
 	} else {
 		args->auxsize = 0;
@@ -464,13 +540,21 @@ elfexec(vnode_t *vp, execa_t *uap, uarg_t *args, intpdata_t *idatap,
 	if (args->emulator != NULL)
 		args->auxsize += sizeof (aux_entry_t);
 
-	if ((brand_action != EBA_NATIVE) && (PROC_IS_BRANDED(p))) {
+	/*
+	 * If this is a native binary that's been given a modified interpreter
+	 * root, inform it that the native system exists at that root.
+	 */
+	if (args->brand_nroot != NULL) {
+		args->auxsize += sizeof (aux_entry_t);
+	}
+
+	if ((*brand_action != EBA_NATIVE) && (PROC_IS_BRANDED(p))) {
 		branded = 1;
 		/*
-		 * We will be adding 4 entries to the aux vectors.  One for
-		 * the the brandname and 3 for the brand specific aux vectors.
+		 * We will be adding 5 entries to the aux vectors.  One for
+		 * the the brandname and 4 for the brand specific aux vectors.
 		 */
-		args->auxsize += 4 * sizeof (aux_entry_t);
+		args->auxsize += 5 * sizeof (aux_entry_t);
 	}
 
 	/* Hardware/Software capabilities */
@@ -501,7 +585,8 @@ elfexec(vnode_t *vp, execa_t *uap, uarg_t *args, intpdata_t *idatap,
 	aux = bigwad->elfargs;
 	/*
 	 * Move args to the user's stack.
-	 * This can fill in the AT_SUN_PLATFORM and AT_SUN_EXECNAME aux entries.
+	 * This can fill in the AT_SUN_PLATFORM, AT_SUN_EXECNAME and AT_RANDOM
+	 * aux entries.
 	 */
 	if ((error = exec_args(uap, args, idatap, (void **)&aux)) != 0) {
 		if (error == -1) {
@@ -528,6 +613,14 @@ elfexec(vnode_t *vp, execa_t *uap, uarg_t *args, intpdata_t *idatap,
 	    len, execsz, &brksize)) != 0)
 		goto bad;
 
+	if (uphdr != NULL) {
+		/*
+		 * Our uphdr has been dynamically allocated if (and only if)
+		 * its program header flags are clear.
+		 */
+		dynuphdr = (uphdr->p_flags == 0);
+	}
+
 	if (uphdr != NULL && dyphdr == NULL)
 		goto bad;
 
@@ -542,17 +635,22 @@ elfexec(vnode_t *vp, execa_t *uap, uarg_t *args, intpdata_t *idatap,
 		char		*p;
 		struct vnode	*nvp;
 
-		dlnsize = dyphdr->p_filesz;
+		dlnsize = dyphdr->p_filesz + nsize;
 
 		if (dlnsize > MAXPATHLEN || dlnsize <= 0)
 			goto bad;
 
+		if (nsize != 0) {
+			bcopy(args->brand_nroot, dlnp, nsize - 1);
+			dlnp[nsize - 1] = '/';
+		}
+
 		/*
 		 * Read in "interpreter" pathname.
 		 */
-		if ((error = vn_rdwr(UIO_READ, vp, dlnp, dyphdr->p_filesz,
-		    (offset_t)dyphdr->p_offset, UIO_SYSSPACE, 0, (rlim64_t)0,
-		    CRED(), &resid)) != 0) {
+		if ((error = vn_rdwr(UIO_READ, vp, dlnp + nsize,
+		    dyphdr->p_filesz, (offset_t)dyphdr->p_offset, UIO_SYSSPACE,
+		    0, (rlim64_t)0, CRED(), &resid)) != 0) {
 			uprintf("%s: Cannot obtain interpreter pathname\n",
 			    exec_file);
 			goto bad;
@@ -688,9 +786,10 @@ elfexec(vnode_t *vp, execa_t *uap, uarg_t *args, intpdata_t *idatap,
 
 		dtrphdr = NULL;
 
-		error = mapelfexec(nvp, ehdrp, nphdrs, phdrbase, &junk, &junk,
+		error = mapelfexec(nvp, ehdrp, nphdrs, phdrbase, NULL, &junk,
 		    &junk, &dtrphdr, NULL, NULL, NULL, &voffset, NULL, len,
 		    execsz, NULL);
+
 		if (error || junk != NULL) {
 			VN_RELE(nvp);
 			uprintf("%s: Cannot map %s\n", exec_file, dlnp);
@@ -718,8 +817,8 @@ elfexec(vnode_t *vp, execa_t *uap, uarg_t *args, intpdata_t *idatap,
 	if (hasauxv) {
 		int auxf = AF_SUN_HWCAPVERIFY;
 		/*
-		 * Note: AT_SUN_PLATFORM and AT_SUN_EXECNAME were filled in via
-		 * exec_args()
+		 * Note: AT_SUN_PLATFORM, AT_SUN_EXECNAME and AT_RANDOM were
+		 * filled in via exec_args()
 		 */
 		ADDAUX(aux, AT_BASE, voffset)
 		ADDAUX(aux, AT_FLAGS, at_flags)
@@ -747,7 +846,7 @@ elfexec(vnode_t *vp, execa_t *uap, uarg_t *args, intpdata_t *idatap,
 		 * malicious user within the zone from crafting a wrapper to
 		 * run native suid commands with unsecure libraries interposed.
 		 */
-		if ((brand_action == EBA_NATIVE) && (PROC_IS_BRANDED(p) &&
+		if ((*brand_action == EBA_NATIVE) && (PROC_IS_BRANDED(p) &&
 		    (setid &= ~EXECSETID_SETID) != 0))
 			auxf &= ~AF_SUN_SETUGID;
 
@@ -789,6 +888,7 @@ elfexec(vnode_t *vp, execa_t *uap, uarg_t *args, intpdata_t *idatap,
 			ADDAUX(aux, AT_SUN_BRAND_AUX1, 0)
 			ADDAUX(aux, AT_SUN_BRAND_AUX2, 0)
 			ADDAUX(aux, AT_SUN_BRAND_AUX3, 0)
+			ADDAUX(aux, AT_SUN_BRAND_AUX4, 0)
 		}
 
 		ADDAUX(aux, AT_NULL, 0)
@@ -896,6 +996,8 @@ bad:
 	if (error == 0)
 		error = ENOEXEC;
 out:
+	if (dynuphdr)
+		kmem_free(uphdr, sizeof (Phdr));
 	if (phdrbase != NULL)
 		kmem_free(phdrbase, phdrsize);
 	if (cap != NULL)
@@ -1182,7 +1284,7 @@ mapelfexec(
 	size_t *brksize)
 {
 	Phdr *phdr;
-	int i, prot, error;
+	int i, prot, error, lastprot = 0;
 	caddr_t addr = NULL;
 	size_t zfodsz;
 	int ptload = 0;
@@ -1190,6 +1292,7 @@ mapelfexec(
 	off_t offset;
 	int hsize = ehdr->e_phentsize;
 	caddr_t mintmp = (caddr_t)-1;
+	uintptr_t lastaddr = NULL;
 	extern int use_brk_lpg;
 
 	if (ehdr->e_type == ET_DYN) {
@@ -1220,13 +1323,11 @@ mapelfexec(
 	} else {
 		*voffset = 0;
 	}
+
 	phdr = (Phdr *)phdrbase;
 	for (i = nphdrs; i > 0; i--) {
 		switch (phdr->p_type) {
 		case PT_LOAD:
-			if ((*dyphdr != NULL) && (*uphdr == NULL))
-				return (0);
-
 			ptload = 1;
 			prot = PROT_USER;
 			if (phdr->p_flags & PF_R)
@@ -1238,6 +1339,34 @@ mapelfexec(
 
 			addr = (caddr_t)((uintptr_t)phdr->p_vaddr + *voffset);
 
+			if ((*dyphdr != NULL) && uphdr != NULL &&
+			    (*uphdr == NULL)) {
+				/*
+				 * The PT_PHDR program header is, strictly
+				 * speaking, optional.  If we find that this
+				 * is missing, we will determine the location
+				 * of the program headers based on the address
+				 * of the lowest PT_LOAD segment (namely, this
+				 * one):  we subtract the p_offset to get to
+				 * the ELF header and then add back the program
+				 * header offset to get to the program headers.
+				 * We then cons up a Phdr that corresponds to
+				 * the (missing) PT_PHDR, setting the flags
+				 * to 0 to denote that this is artificial and
+				 * should (must) be freed by the caller.
+				 */
+				Phdr *cons;
+
+				cons = kmem_zalloc(sizeof (Phdr), KM_SLEEP);
+
+				cons->p_flags = 0;
+				cons->p_type = PT_PHDR;
+				cons->p_vaddr = ((uintptr_t)addr -
+				    phdr->p_offset) + ehdr->e_phoff;
+
+				*uphdr = cons;
+			}
+
 			/*
 			 * Keep track of the segment with the lowest starting
 			 * address.
@@ -1245,6 +1374,41 @@ mapelfexec(
 			if (addr < mintmp)
 				mintmp = addr;
 
+			/*
+			 * Segments need not correspond to page boundaries:
+			 * they are permitted to share a page.  If two PT_LOAD
+			 * segments share the same page, and the permissions
+			 * of the segments differ, the behavior is historically
+			 * that the permissions of the latter segment are used
+			 * for the page that the two segments share.  This is
+			 * also historically a non-issue:  binaries generated
+			 * by most anything will make sure that two PT_LOAD
+			 * segments with differing permissions don't actually
+			 * share any pages.  However, there exist some crazy
+			 * things out there (including at least an obscure
+			 * Portuguese teaching language called G-Portugol) that
+			 * actually do the wrong thing and expect it to work:
+			 * they have a segment with execute permission share
+			 * a page with a subsequent segment that does not
+			 * have execute permissions and expect the resulting
+			 * shared page to in fact be executable.  To accommodate
+			 * such broken link editors, we take advantage of a
+			 * latitude explicitly granted to the loader:  it is
+			 * permitted to make _any_ PT_LOAD segment executable
+			 * (provided that it is readable or writable).  If we
+			 * see that we're sharing a page and that the previous
+			 * page was executable, we will add execute permissions
+			 * to our segment.
+			 */
+			if (btop(lastaddr) == btop((uintptr_t)addr) &&
+			    (phdr->p_flags & (PF_R | PF_W)) &&
+			    (lastprot & PROT_EXEC)) {
+				prot |= PROT_EXEC;
+			}
+
+			lastaddr = (uintptr_t)addr + phdr->p_filesz;
+			lastprot = prot;
+
 			zfodsz = (size_t)phdr->p_memsz - phdr->p_filesz;
 
 			offset = phdr->p_offset;
@@ -1319,9 +1483,12 @@ mapelfexec(
 			break;
 
 		case PT_PHDR:
-			if (ptload)
+			if (ptload || phdr->p_flags == 0)
 				goto bad;
-			*uphdr = phdr;
+
+			if (uphdr != NULL)
+				*uphdr = phdr;
+
 			break;
 
 		case PT_NULL:
@@ -2170,7 +2337,7 @@ static struct modlexec modlexec = {
 extern int elf32exec(vnode_t *vp, execa_t *uap, uarg_t *args,
 			intpdata_t *idatap, int level, long *execsz,
 			int setid, caddr_t exec_file, cred_t *cred,
-			int brand_action);
+			int *brand_action);
 extern int elf32core(vnode_t *vp, proc_t *p, cred_t *credp,
 			rlim64_t rlimit, int sig, core_content_t content);
 
diff --git a/usr/src/uts/common/exec/intp/intp.c b/usr/src/uts/common/exec/intp/intp.c
index 998c0f1f8e..316f4c07e0 100644
--- a/usr/src/uts/common/exec/intp/intp.c
+++ b/usr/src/uts/common/exec/intp/intp.c
@@ -22,6 +22,7 @@
  * Copyright 2010 Sun Microsystems, Inc.  All rights reserved.
  * Use is subject to license terms.
  * Copyright 2012 Milan Jurik. All rights reserved.
+ * Copyright 2015, Joyent, Inc.
  */
 
 /*	Copyright (c) 1988 AT&T	*/
@@ -53,7 +54,7 @@
 #include <sys/modctl.h>
 
 extern int intpexec(struct vnode *, struct execa *, struct uarg *,
-    struct intpdata *, int, long *, int, caddr_t, struct cred *, int);
+    struct intpdata *, int, long *, int, caddr_t, struct cred *, int *);
 
 static struct execsw esw = {
 	intpmagicstr,
@@ -169,7 +170,7 @@ intpexec(
 	int setid,
 	caddr_t exec_file,
 	struct cred *cred,
-	int brand_action)
+	int *brand_action)
 {
 	_NOTE(ARGUNUSED(brand_action))
 	vnode_t *nvp;
@@ -180,6 +181,7 @@ intpexec(
 	char *opath;
 	char devfd[19]; /* 32-bit int fits in 10 digits + 8 for "/dev/fd/" */
 	int fd = -1;
+	int custom_action = EBA_NONE;
 
 	if (level) {		/* Can't recurse */
 		error = ENOEXEC;
@@ -228,7 +230,7 @@ intpexec(
 	}
 
 	error = gexec(&nvp, uap, args, &idata, ++level, execsz, exec_file, cred,
-	    EBA_NONE);
+	    &custom_action);
 
 	if (!error) {
 		/*
diff --git a/usr/src/uts/common/exec/java/java.c b/usr/src/uts/common/exec/java/java.c
index 7cf9126e8d..4ce223a682 100644
--- a/usr/src/uts/common/exec/java/java.c
+++ b/usr/src/uts/common/exec/java/java.c
@@ -21,6 +21,7 @@
 /*
  * Copyright 2010 Sun Microsystems, Inc.  All rights reserved.
  * Use is subject to license terms.
+ * Copyright 2015, Joyent, Inc.
  */
 
 /*
@@ -85,7 +86,7 @@ char *jexec_arg = "-jar";
 static int
 javaexec(vnode_t *vp, struct execa *uap, struct uarg *args,
     struct intpdata *idatap, int level, long *execsz, int setid,
-    caddr_t execfile, cred_t *cred, int brand_action)
+    caddr_t execfile, cred_t *cred, int *brand_action)
 {
 	struct intpdata idata;
 	int error;
diff --git a/usr/src/uts/common/exec/shbin/shbin.c b/usr/src/uts/common/exec/shbin/shbin.c
index a6cf129d9d..a431264103 100644
--- a/usr/src/uts/common/exec/shbin/shbin.c
+++ b/usr/src/uts/common/exec/shbin/shbin.c
@@ -22,6 +22,7 @@
 /*
  * Copyright 2010 Sun Microsystems, Inc.  All rights reserved.
  * Use is subject to license terms.
+ * Copyright 2015, Joyent, Inc.
  */
 
 #include <sys/types.h>
@@ -58,7 +59,7 @@ shbinexec(
 	int setid,
 	caddr_t exec_file,
 	struct cred *cred,
-	int brand_action);
+	int *brand_action);
 
 #define	SHBIN_CNTL(x)	((x)&037)
 #define	SHBINMAGIC_LEN	4
@@ -162,7 +163,7 @@ shbinexec(
 	int setid,
 	caddr_t exec_file,
 	struct cred *cred,
-	int brand_action)
+	int *brand_action)
 {
 	_NOTE(ARGUNUSED(brand_action))
 	vnode_t *nvp;
diff --git a/usr/src/uts/common/fs/bootfs/bootfs_construct.c b/usr/src/uts/common/fs/bootfs/bootfs_construct.c
new file mode 100644
index 0000000000..b909b5d121
--- /dev/null
+++ b/usr/src/uts/common/fs/bootfs/bootfs_construct.c
@@ -0,0 +1,368 @@
+/*
+ * This file and its contents are supplied under the terms of the
+ * Common Development and Distribution License ("CDDL"), version 1.0.
+ * You may only use this file in accordance with the terms of version
+ * 1.0 of the CDDL.
+ *
+ * A full copy of the text of the CDDL should have accompanied this
+ * source.  A copy of the CDDL is also available via the Internet at
+ * http://www.illumos.org/license/CDDL.
+ */
+
+/*
+ * Copyright (c) 2015 Joyent, Inc.  All rights reserved.
+ */
+
+/*
+ * This file takes care of reading the boot time modules and constructing them
+ * into the appropriate series of vnodes.
+ */
+
+#include <sys/conf.h>
+#include <sys/ddi.h>
+#include <sys/sunddi.h>
+#include <sys/vfs.h>
+#include <sys/sysmacros.h>
+#include <sys/stat.h>
+
+#include <sys/fs/bootfs_impl.h>
+
+kmem_cache_t *bootfs_node_cache;
+
+static const vattr_t bootfs_vattr_dir = {
+	AT_ALL,					/* va_mask */
+	VDIR,					/* va_type */
+	S_IFDIR | 0555,				/* va_mode */
+	0,					/* va_uid */
+	0,					/* va_gid */
+	0,					/* va_fsid */
+	0,					/* va_nodeid */
+	1,					/* va_nlink */
+	0,					/* va_size */
+	0,					/* va_atime */
+	0,					/* va_mtime */
+	0,					/* va_ctime */
+	0,					/* va_rdev */
+	0,					/* va_blksize */
+	0,					/* va_nblocks */
+	0					/* va_seq */
+};
+
+static const vattr_t bootfs_vattr_reg = {
+	AT_ALL,					/* va_mask */
+	VREG,					/* va_type */
+	S_IFREG | 0555,				/* va_mode */
+	0,					/* va_uid */
+	0,					/* va_gid */
+	0,					/* va_fsid */
+	0,					/* va_nodeid */
+	1,					/* va_nlink */
+	0,					/* va_size */
+	0,					/* va_atime */
+	0,					/* va_mtime */
+	0,					/* va_ctime */
+	0,					/* va_rdev */
+	0,					/* va_blksize */
+	0,					/* va_nblocks */
+	0					/* va_seq */
+};
+
+/*ARGSUSED*/
+int
+bootfs_node_constructor(void *buf, void *arg, int kmflags)
+{
+	bootfs_node_t *bnp = buf;
+
+	bnp->bvn_vnp = vn_alloc(kmflags);
+	if (bnp->bvn_vnp == NULL)
+		return (-1);
+
+	return (0);
+}
+
+/*ARGSUSED*/
+void
+bootfs_node_destructor(void *buf, void *arg)
+{
+	bootfs_node_t *bnp = buf;
+
+	vn_free(bnp->bvn_vnp);
+}
+
+static int
+bootfs_comparator(const void *a, const void *b)
+{
+	const bootfs_node_t *lfs, *rfs;
+	int ret;
+
+	lfs = a;
+	rfs = b;
+
+	ret = strcmp(lfs->bvn_name, rfs->bvn_name);
+	if (ret > 0)
+		ret = 1;
+	if (ret < 0)
+		ret = -1;
+	return (ret);
+}
+
+static void
+bootfs_node_init(bootfs_t *bfs, bootfs_node_t *bnp, const struct vattr *vap,
+    const char *name, size_t namelen)
+{
+	timestruc_t now;
+
+	vn_reinit(bnp->bvn_vnp);
+
+	bnp->bvn_vnp->v_flag |= VNOSWAP;
+	bnp->bvn_vnp->v_type = vap->va_type;
+	bnp->bvn_vnp->v_vfsp = bfs->bfs_vfsp;
+	bnp->bvn_vnp->v_rdev = 0;
+	bnp->bvn_vnp->v_data = (caddr_t)bnp;
+	vn_setops(bnp->bvn_vnp, bootfs_vnodeops);
+
+	bnp->bvn_name = kmem_alloc(namelen + 1, KM_SLEEP);
+	bcopy(name, bnp->bvn_name, namelen);
+	bnp->bvn_name[namelen] = '\0';
+	if (vap->va_type == VDIR) {
+		avl_create(&bnp->bvn_dir, bootfs_comparator,
+		    sizeof (bootfs_node_t),
+		    offsetof(bootfs_node_t, bvn_link));
+	}
+	bzero(&bnp->bvn_link, sizeof (avl_node_t));
+	bcopy(vap, &bnp->bvn_attr, sizeof (vattr_t));
+
+	gethrestime(&now);
+	bnp->bvn_attr.va_atime = now;
+	bnp->bvn_attr.va_ctime = now;
+	bnp->bvn_attr.va_mtime = now;
+	bnp->bvn_attr.va_fsid = makedevice(bootfs_major, bfs->bfs_minor);
+	bnp->bvn_attr.va_nodeid = bfs->bfs_ninode;
+	bnp->bvn_attr.va_blksize = PAGESIZE;
+	bfs->bfs_ninode++;
+	list_insert_tail(&bfs->bfs_nodes, bnp);
+}
+
+static void
+bootfs_mkroot(bootfs_t *bfs)
+{
+	bootfs_node_t *bnp;
+
+	bnp = kmem_cache_alloc(bootfs_node_cache, KM_SLEEP);
+	bootfs_node_init(bfs, bnp, &bootfs_vattr_dir, "/", 1);
+	bnp->bvn_vnp->v_flag |= VROOT;
+	bnp->bvn_parent = bnp;
+	bfs->bfs_rootvn = bnp;
+	bfs->bfs_stat.bfss_ndirs.value.ui32++;
+	vn_exists(bnp->bvn_vnp);
+}
+
+static int
+bootfs_mknode(bootfs_t *bfs, bootfs_node_t *parent, bootfs_node_t **outp,
+    const char *name, size_t namelen, const vattr_t *vap, uintptr_t addr,
+    uint64_t size)
+{
+	bootfs_node_t *bnp;
+	bootfs_node_t sn;
+	avl_index_t where;
+	char *buf;
+
+	ASSERT(parent->bvn_attr.va_type == VDIR);
+	buf = kmem_alloc(namelen + 1, KM_SLEEP);
+	bcopy(name, buf, namelen);
+	buf[namelen] = '\0';
+	sn.bvn_name = buf;
+	if ((bnp = avl_find(&parent->bvn_dir, &sn, &where)) != NULL) {
+		kmem_free(buf, namelen + 1);
+		/* Directories can collide, files cannot */
+		if (vap->va_type == VDIR) {
+			*outp = bnp;
+			return (0);
+		}
+		return (EEXIST);
+	}
+	kmem_free(buf, namelen + 1);
+
+	bnp = kmem_cache_alloc(bootfs_node_cache, KM_SLEEP);
+	bootfs_node_init(bfs, bnp, vap, name, namelen);
+	bnp->bvn_parent = parent;
+	avl_add(&parent->bvn_dir, bnp);
+	*outp = bnp;
+
+	if (vap->va_type == VDIR) {
+		parent->bvn_attr.va_size++;
+		parent->bvn_attr.va_nlink++;
+		bfs->bfs_stat.bfss_ndirs.value.ui32++;
+	} else {
+		bnp->bvn_addr = addr;
+		bnp->bvn_size = size;
+		bfs->bfs_stat.bfss_nfiles.value.ui32++;
+		bfs->bfs_stat.bfss_nbytes.value.ui64 += size;
+		bnp->bvn_attr.va_nblocks = P2ROUNDUP(size, 512) >> 9;
+		bnp->bvn_attr.va_size = size;
+	}
+
+	vn_exists(bnp->bvn_vnp);
+
+	return (0);
+}
+
+/*
+ * Given the address, size, and path a boot-time module would like, go through
+ * and create all of the directory entries that are required and then the file
+ * itself. If someone has passed in a module that has the same name as another
+ * one, we honor the first one.
+ */
+static int
+bootfs_construct_entry(bootfs_t *bfs, uintptr_t addr, uint64_t size,
+    const char *mname)
+{
+	char *sp;
+	size_t nlen;
+	int ret;
+	bootfs_node_t *nbnp;
+
+	const char *p = mname;
+	bootfs_node_t *bnp = bfs->bfs_rootvn;
+
+	if (*p == '\0')
+		return (EINVAL);
+
+	for (;;) {
+		/* First eliminate all leading / characters. */
+		while (*p == '/')
+			p++;
+
+		/* A name with all slashes or ending in a / */
+		if (*p == '\0')
+			return (EINVAL);
+
+		sp = strchr(p, '/');
+		if (sp == NULL)
+			break;
+		nlen = (ptrdiff_t)sp - (ptrdiff_t)p;
+		if (strncmp(p, ".", nlen) == 0) {
+			p = sp + 1;
+			continue;
+		}
+
+		if (strncmp(p, "..", nlen) == 0) {
+			bnp = bnp->bvn_parent;
+			p = sp + 1;
+			continue;
+		}
+
+		VERIFY(bootfs_mknode(bfs, bnp, &nbnp, p, nlen,
+		    &bootfs_vattr_dir, addr, size) == 0);
+		p = sp + 1;
+		bnp = nbnp;
+	}
+
+	nlen = strlen(p);
+	ret = bootfs_mknode(bfs, bnp, &nbnp, p, nlen, &bootfs_vattr_reg,
+	    addr, size);
+	if (ret != 0)
+		return (ret);
+
+	return (0);
+}
+
+/*
+ * We're going to go through every boot time module and construct the
+ * appropriate vnodes for them now. Because there are very few of these that
+ * exist, generally on the order of a handful, we're going to create them all
+ * when the file system is initialized and then tear them all down when the
+ * module gets unloaded.
+ *
+ * The information about the modules is contained in properties on the root of
+ * the devinfo tree. Specifically there are three properties per module:
+ *
+ *   - module-size-%d	int64_t size, in bytes, of the boot time module.
+ *   - module-addr-%d	The address of the boot time module
+ *   - module-name-%d	The string name of the boot time module
+ *
+ * Note that the module-size and module-addr fields are always 64-bit values
+ * regardless of being on a 32-bit or 64-bit kernel. module-name is a string
+ * property.
+ *
+ * There is no property that indicates the total number of such modules. Modules
+ * start at 0 and work their way up incrementally. The first time we can't find
+ * a module or a property, then we stop.
+ */
+void
+bootfs_construct(bootfs_t *bfs)
+{
+	uint_t id = 0, ndata;
+	char paddr[64], psize[64], pname[64], *mname;
+	dev_info_t *root;
+	uchar_t *datap;
+	uint64_t size = 0, addr = 0;
+	int ret;
+
+	bootfs_mkroot(bfs);
+	root = ddi_root_node();
+
+	for (;;) {
+		if (id == UINT32_MAX)
+			break;
+
+		if (snprintf(paddr, sizeof (paddr), "module-addr-%d", id) >
+		    sizeof (paddr))
+			break;
+
+		if (snprintf(psize, sizeof (paddr), "module-size-%d", id) >
+		    sizeof (paddr))
+			break;
+
+		if (snprintf(pname, sizeof (paddr), "module-name-%d", id) >
+		    sizeof (paddr))
+			break;
+
+		if (ddi_prop_lookup_byte_array(DDI_DEV_T_ANY, root,
+		    DDI_PROP_DONTPASS, paddr, &datap, &ndata) !=
+		    DDI_PROP_SUCCESS)
+			break;
+
+		if (ndata == 8)
+			bcopy(datap, &addr, sizeof (uint64_t));
+		ddi_prop_free(datap);
+		if (ndata != 8)
+			break;
+
+		if (ddi_prop_lookup_byte_array(DDI_DEV_T_ANY, root,
+		    DDI_PROP_DONTPASS, psize, &datap, &ndata) !=
+		    DDI_PROP_SUCCESS)
+			break;
+		if (ndata == 8)
+			bcopy(datap, &size, sizeof (uint64_t));
+		ddi_prop_free(datap);
+		if (ndata != 8)
+			break;
+
+		if (ddi_prop_lookup_string(DDI_DEV_T_ANY, root,
+		    DDI_PROP_DONTPASS, pname, &mname) != DDI_PROP_SUCCESS)
+			break;
+
+		ret = bootfs_construct_entry(bfs, addr, size, mname);
+		if (ret == EINVAL)
+			bfs->bfs_stat.bfss_ndiscards.value.ui32++;
+		if (ret == EEXIST)
+			bfs->bfs_stat.bfss_ndups.value.ui32++;
+		ddi_prop_free(mname);
+
+		id++;
+	}
+}
+
+void
+bootfs_destruct(bootfs_t *bfs)
+{
+	bootfs_node_t *bnp;
+
+	while ((bnp = list_remove_head(&bfs->bfs_nodes)) != NULL) {
+		ASSERT(bnp->bvn_vnp->v_count == 1);
+		VN_RELE(bnp->bvn_vnp);
+		kmem_free(bnp->bvn_name, strlen(bnp->bvn_name) + 1);
+		kmem_cache_free(bootfs_node_cache, bnp);
+	}
+}
diff --git a/usr/src/uts/common/fs/bootfs/bootfs_vfsops.c b/usr/src/uts/common/fs/bootfs/bootfs_vfsops.c
new file mode 100644
index 0000000000..e642e86169
--- /dev/null
+++ b/usr/src/uts/common/fs/bootfs/bootfs_vfsops.c
@@ -0,0 +1,321 @@
+/*
+ * This file and its contents are supplied under the terms of the
+ * Common Development and Distribution License ("CDDL"), version 1.0.
+ * You may only use this file in accordance with the terms of version
+ * 1.0 of the CDDL.
+ *
+ * A full copy of the text of the CDDL should have accompanied this
+ * source.  A copy of the CDDL is also available via the Internet at
+ * http://www.illumos.org/license/CDDL.
+ */
+
+/*
+ * Copyright (c) 2015 Joyent, Inc.
+ */
+
+#include <sys/errno.h>
+#include <sys/modctl.h>
+#include <sys/types.h>
+#include <sys/mkdev.h>
+#include <sys/ddi.h>
+#include <sys/sunddi.h>
+#include <sys/vfs.h>
+#include <sys/vfs_opreg.h>
+#include <sys/systm.h>
+#include <sys/id_space.h>
+#include <sys/cmn_err.h>
+#include <sys/ksynch.h>
+#include <sys/policy.h>
+#include <sys/mount.h>
+#include <sys/sysmacros.h>
+
+#include <sys/fs/bootfs_impl.h>
+
+/*
+ * While booting, additional types of modules and files can be passed in to the
+ * loader. These include the familiar boot archive, as well as, a module hash
+ * and additional modules that are interpreted as files. As part of the handoff
+ * in early boot, information about these modules are saved as properties on the
+ * root of the devinfo tree, similar to other boot-time properties.
+ *
+ * This file system provides a read-only view of those additional files. Due to
+ * its limited scope, it has a slightly simpler construction than several other
+ * file systems. When mounted, it looks for the corresponding properties and
+ * creates bootfs_node_t's and vnodes for all of the corresponding files and
+ * directories that exist along the way. At this time, there are currently a
+ * rather small number of files passed in this way.
+ *
+ * This does lead to one behavior that folks used to other file systems might
+ * find peculiar. Because we are not always actively creating and destroying the
+ * required vnodes on demand, the count on the root vnode will not be going up
+ * accordingly with the existence of other vnodes. This means that a bootfs file
+ * system that is not in use will have all of its vnodes exist with a v_count of
+ * one.
+ */
+
+major_t bootfs_major;
+static int bootfs_fstype;
+static id_space_t *bootfs_idspace;
+static uint64_t bootfs_nactive;
+static kmutex_t bootfs_lock;
+
+static const char *bootfs_name = "bootfs";
+
+static int
+bootfs_mount(vfs_t *vfsp, vnode_t *mvp, struct mounta *uap, cred_t *cr)
+{
+	int ret;
+	bootfs_t *bfs;
+	struct pathname dpn;
+	dev_t fsdev;
+
+	if ((ret = secpolicy_fs_mount(cr, mvp, vfsp)) != 0)
+		return (ret);
+
+	if (mvp->v_type != VDIR)
+		return (ENOTDIR);
+
+	if (uap->flags & MS_REMOUNT)
+		return (EBUSY);
+
+	mutex_enter(&mvp->v_lock);
+	if ((uap->flags & MS_OVERLAY) == 0 &&
+	    (mvp->v_count != 1 || (mvp->v_flag & VROOT))) {
+		mutex_exit(&mvp->v_lock);
+		return (EBUSY);
+	}
+	mutex_exit(&mvp->v_lock);
+
+	/*
+	 * We indicate that the backing store is bootfs. We don't want to use
+	 * swap, because folks might think that this is putting all the data
+	 * into memory ala tmpfs. Rather these modules are always in memory and
+	 * there's nothing to be done about that.
+	 */
+	vfs_setresource(vfsp, bootfs_name, 0);
+	bfs = kmem_zalloc(sizeof (bootfs_t), KM_NOSLEEP | KM_NORMALPRI);
+	if (bfs == NULL)
+		return (ENOMEM);
+
+	ret = pn_get(uap->dir,
+	    (uap->flags & MS_SYSSPACE) ? UIO_SYSSPACE : UIO_USERSPACE, &dpn);
+	if (ret != 0) {
+		kmem_free(bfs, sizeof (bfs));
+		return (ret);
+	}
+
+	bfs->bfs_minor = id_alloc(bootfs_idspace);
+	bfs->bfs_kstat = kstat_create_zone("bootfs", bfs->bfs_minor, "bootfs",
+	    "fs", KSTAT_TYPE_NAMED,
+	    sizeof (bootfs_stat_t) / sizeof (kstat_named_t),
+	    KSTAT_FLAG_VIRTUAL, GLOBAL_ZONEID);
+	if (bfs->bfs_kstat == NULL) {
+		id_free(bootfs_idspace, bfs->bfs_minor);
+		pn_free(&dpn);
+		kmem_free(bfs, sizeof (bfs));
+		return (ENOMEM);
+	}
+	bfs->bfs_kstat->ks_data = &bfs->bfs_stat;
+
+	fsdev = makedevice(bootfs_major, bfs->bfs_minor);
+	bfs->bfs_vfsp = vfsp;
+
+	vfsp->vfs_data = (caddr_t)bfs;
+	vfsp->vfs_fstype = bootfs_fstype;
+	vfsp->vfs_dev = fsdev;
+	vfsp->vfs_bsize = PAGESIZE;
+	vfsp->vfs_flag |= VFS_RDONLY | VFS_NOSETUID | VFS_NOTRUNC |
+	    VFS_UNLINKABLE;
+	vfs_make_fsid(&vfsp->vfs_fsid, fsdev, bootfs_fstype);
+	bfs->bfs_mntpath = kmem_alloc(dpn.pn_pathlen + 1, KM_SLEEP);
+	bcopy(dpn.pn_path, bfs->bfs_mntpath, dpn.pn_pathlen);
+	bfs->bfs_mntpath[dpn.pn_pathlen] = '\0';
+	pn_free(&dpn);
+	list_create(&bfs->bfs_nodes, sizeof (bootfs_node_t),
+	    offsetof(bootfs_node_t, bvn_alink));
+
+	kstat_named_init(&bfs->bfs_stat.bfss_nfiles, "nfiles",
+	    KSTAT_DATA_UINT32);
+	kstat_named_init(&bfs->bfs_stat.bfss_ndirs, "ndirs",
+	    KSTAT_DATA_UINT32);
+	kstat_named_init(&bfs->bfs_stat.bfss_nbytes, "nbytes",
+	    KSTAT_DATA_UINT64);
+	kstat_named_init(&bfs->bfs_stat.bfss_ndups, "ndup",
+	    KSTAT_DATA_UINT32);
+	kstat_named_init(&bfs->bfs_stat.bfss_ndiscards, "ndiscard",
+	    KSTAT_DATA_UINT32);
+
+	bootfs_construct(bfs);
+
+	kstat_install(bfs->bfs_kstat);
+
+	return (0);
+}
+
+static int
+bootfs_unmount(vfs_t *vfsp, int flag, cred_t *cr)
+{
+	int ret;
+	bootfs_t *bfs = vfsp->vfs_data;
+	bootfs_node_t *bnp;
+
+	if ((ret = secpolicy_fs_unmount(cr, vfsp)) != 0)
+		return (ret);
+
+	if (flag & MS_FORCE)
+		return (ENOTSUP);
+
+	for (bnp = list_head(&bfs->bfs_nodes); bnp != NULL;
+	    bnp = list_next(&bfs->bfs_nodes, bnp)) {
+		mutex_enter(&bnp->bvn_vnp->v_lock);
+		if (bnp->bvn_vnp->v_count > 1) {
+			mutex_exit(&bnp->bvn_vnp->v_lock);
+			return (EBUSY);
+		}
+		mutex_exit(&bnp->bvn_vnp->v_lock);
+	}
+
+	kstat_delete(bfs->bfs_kstat);
+	bootfs_destruct(bfs);
+	list_destroy(&bfs->bfs_nodes);
+	kmem_free(bfs->bfs_mntpath, strlen(bfs->bfs_mntpath) + 1);
+	id_free(bootfs_idspace, bfs->bfs_minor);
+	kmem_free(bfs, sizeof (bootfs_t));
+	return (0);
+}
+
+static int
+bootfs_root(vfs_t *vfsp, vnode_t **vpp)
+{
+	bootfs_t *bfs;
+
+	bfs = (bootfs_t *)vfsp->vfs_data;
+	*vpp = bfs->bfs_rootvn->bvn_vnp;
+	VN_HOLD(*vpp)
+
+	return (0);
+}
+
+static int
+bootfs_statvfs(vfs_t *vfsp, struct statvfs64 *sbp)
+{
+	const bootfs_t *bfs = (bootfs_t *)vfsp;
+	dev32_t d32;
+
+	sbp->f_bsize = PAGESIZE;
+	sbp->f_frsize = PAGESIZE;
+
+	sbp->f_blocks = bfs->bfs_stat.bfss_nbytes.value.ui64 >> PAGESHIFT;
+	sbp->f_bfree = 0;
+	sbp->f_bavail = 0;
+
+	sbp->f_files = bfs->bfs_stat.bfss_nfiles.value.ui32 +
+	    bfs->bfs_stat.bfss_ndirs.value.ui32;
+	sbp->f_ffree = 0;
+	sbp->f_favail = 0;
+
+	(void) cmpldev(&d32, vfsp->vfs_dev);
+	sbp->f_fsid = d32;
+	(void) strlcpy(sbp->f_basetype, bootfs_name, FSTYPSZ);
+	bzero(sbp->f_fstr, sizeof (sbp->f_fstr));
+
+	return (0);
+}
+
+static const fs_operation_def_t bootfs_vfsops_tmpl[] = {
+	VFSNAME_MOUNT,		{ .vfs_mount = bootfs_mount },
+	VFSNAME_UNMOUNT,	{ .vfs_unmount = bootfs_unmount },
+	VFSNAME_ROOT,		{ .vfs_root = bootfs_root },
+	VFSNAME_STATVFS,	{ .vfs_statvfs = bootfs_statvfs },
+	NULL,			NULL
+};
+
+static int
+bootfs_init(int fstype, char *name)
+{
+	int ret;
+
+	bootfs_fstype = fstype;
+	ASSERT(bootfs_fstype != 0);
+
+	ret = vfs_setfsops(fstype, bootfs_vfsops_tmpl, NULL);
+	if (ret != 0)
+		return (ret);
+
+	ret = vn_make_ops(name, bootfs_vnodeops_template, &bootfs_vnodeops);
+	if (ret != 0) {
+		(void) vfs_freevfsops_by_type(bootfs_fstype);
+		return (ret);
+	}
+
+	bootfs_major = getudev();
+	if (bootfs_major == (major_t)-1) {
+		cmn_err(CE_WARN, "bootfs_init: Can't get unique device number");
+		bootfs_major = 0;
+	}
+
+	bootfs_nactive = 0;
+	return (0);
+}
+
+static mntopts_t bootfs_mntopts = {
+	0, NULL
+};
+
+static vfsdef_t bootfs_vfsdef = {
+	VFSDEF_VERSION,
+	"bootfs",
+	bootfs_init,
+	VSW_HASPROTO|VSW_STATS,
+	&bootfs_mntopts
+};
+
+static struct modlfs bootfs_modlfs = {
+	&mod_fsops, "boot-time modules file system", &bootfs_vfsdef
+};
+
+static struct modlinkage bootfs_modlinkage = {
+	MODREV_1, &bootfs_modlfs, NULL
+};
+
+int
+_init(void)
+{
+	bootfs_node_cache = kmem_cache_create("bootfs_node_cache",
+	    sizeof (bootfs_node_t), 0, bootfs_node_constructor,
+	    bootfs_node_destructor, NULL, NULL, NULL, 0);
+	bootfs_idspace = id_space_create("bootfs_minors", 1, INT32_MAX);
+	mutex_init(&bootfs_lock, NULL, MUTEX_DEFAULT, NULL);
+
+	return (mod_install(&bootfs_modlinkage));
+}
+
+int
+_info(struct modinfo *modinfop)
+{
+	return (mod_info(&bootfs_modlinkage, modinfop));
+}
+
+int
+_fini(void)
+{
+	int err;
+
+	mutex_enter(&bootfs_lock);
+	if (bootfs_nactive > 0) {
+		mutex_exit(&bootfs_lock);
+		return (EBUSY);
+	}
+	mutex_exit(&bootfs_lock);
+
+	err = mod_remove(&bootfs_modlinkage);
+	if (err != 0)
+		return (err);
+
+	(void) vfs_freevfsops_by_type(bootfs_fstype);
+	vn_freevnodeops(bootfs_vnodeops);
+	id_space_destroy(bootfs_idspace);
+	mutex_destroy(&bootfs_lock);
+	kmem_cache_destroy(bootfs_node_cache);
+	return (err);
+}
diff --git a/usr/src/uts/common/fs/bootfs/bootfs_vnops.c b/usr/src/uts/common/fs/bootfs/bootfs_vnops.c
new file mode 100644
index 0000000000..f63d0a4f24
--- /dev/null
+++ b/usr/src/uts/common/fs/bootfs/bootfs_vnops.c
@@ -0,0 +1,544 @@
+/*
+ * This file and its contents are supplied under the terms of the
+ * Common Development and Distribution License ("CDDL"), version 1.0.
+ * You may only use this file in accordance with the terms of version
+ * 1.0 of the CDDL.
+ *
+ * A full copy of the text of the CDDL should have accompanied this
+ * source.  A copy of the CDDL is also available via the Internet at
+ * http://www.illumos.org/license/CDDL.
+ */
+
+/*
+ * Copyright (c) 2014 Joyent, Inc.  All rights reserved.
+ */
+
+/*
+ * bootfs vnode operations
+ */
+
+#include <sys/types.h>
+#include <sys/uio.h>
+#include <sys/sunddi.h>
+#include <sys/errno.h>
+#include <sys/vfs_opreg.h>
+#include <sys/vnode.h>
+#include <sys/mman.h>
+#include <fs/fs_subr.h>
+#include <sys/policy.h>
+#include <sys/sysmacros.h>
+#include <sys/dirent.h>
+#include <sys/uio.h>
+#include <vm/pvn.h>
+#include <vm/hat.h>
+#include <vm/seg_map.h>
+#include <vm/seg_vn.h>
+#include <sys/vmsystm.h>
+
+#include <sys/fs/bootfs_impl.h>
+
+struct vnodeops *bootfs_vnodeops;
+
+/*ARGSUSED*/
+static int
+bootfs_open(vnode_t **vpp, int flag, cred_t *cr, caller_context_t *ct)
+{
+	return (0);
+}
+
+/*ARGSUSED*/
+static int
+bootfs_close(vnode_t *vp, int flag, int count, offset_t offset, cred_t *cr,
+    caller_context_t *ct)
+{
+	return (0);
+}
+
+/*ARGSUSED*/
+static int
+bootfs_read(vnode_t *vp, struct uio *uiop, int ioflag, cred_t *cr,
+    caller_context_t *ct)
+{
+	int err;
+	ssize_t sres = uiop->uio_resid;
+	bootfs_node_t *bnp = vp->v_data;
+
+	if (vp->v_type == VDIR)
+		return (EISDIR);
+
+	if (vp->v_type != VREG)
+		return (EINVAL);
+
+	if (uiop->uio_loffset < 0)
+		return (EINVAL);
+
+	if (uiop->uio_loffset >= bnp->bvn_size)
+		return (0);
+
+	err = 0;
+	while (uiop->uio_resid != 0) {
+		caddr_t base;
+		long offset, frem;
+		ulong_t poff, segoff;
+		size_t bytes;
+		int relerr;
+
+		offset = uiop->uio_loffset;
+		poff = offset & PAGEOFFSET;
+		bytes = MIN(PAGESIZE - poff, uiop->uio_resid);
+
+		frem = bnp->bvn_size - offset;
+		if (frem <= 0) {
+			err = 0;
+			break;
+		}
+
+		/* Don't read past EOF */
+		bytes = MIN(bytes, frem);
+
+		/*
+		 * Segmaps are likely larger than our page size, so make sure we
+		 * have the proper offfset into the resulting segmap data.
+		 */
+		segoff = (offset & PAGEMASK) & MAXBOFFSET;
+
+		base = segmap_getmapflt(segkmap, vp, offset & MAXBMASK, bytes,
+		    1, S_READ);
+
+		err = uiomove(base + segoff + poff, bytes, UIO_READ, uiop);
+		relerr = segmap_release(segkmap, base, 0);
+
+		if (err == 0)
+			err = relerr;
+
+		if (err != 0)
+			break;
+	}
+
+	/* Even if we had an error in a partial read, return success */
+	if (uiop->uio_resid > sres)
+		err = 0;
+
+	gethrestime(&bnp->bvn_attr.va_atime);
+
+	return (err);
+}
+
+/*ARGSUSED*/
+static int
+bootfs_ioctl(vnode_t *vp, int cmd, intptr_t data, int flag,
+    cred_t *cr, int *rvalp, caller_context_t *ct)
+{
+	return (ENOTTY);
+}
+
+/*ARGSUSED*/
+static int
+bootfs_getattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr,
+    caller_context_t *ct)
+{
+	uint32_t mask;
+	bootfs_node_t *bpn = (bootfs_node_t *)vp->v_data;
+
+	mask = vap->va_mask;
+	bcopy(&bpn->bvn_attr, vap, sizeof (vattr_t));
+	vap->va_mask = mask;
+	return (0);
+}
+
+/*ARGSUSED*/
+static int
+bootfs_access(vnode_t *vp, int mode, int flags, cred_t *cr,
+    caller_context_t *ct)
+{
+	int shift = 0;
+	bootfs_node_t *bpn = (bootfs_node_t *)vp->v_data;
+
+	if (crgetuid(cr) != bpn->bvn_attr.va_uid) {
+		shift += 3;
+		if (groupmember(bpn->bvn_attr.va_gid, cr) == 0)
+			shift += 3;
+	}
+
+	return (secpolicy_vnode_access2(cr, vp, bpn->bvn_attr.va_uid,
+	    bpn->bvn_attr.va_mode << shift, mode));
+}
+
+/*ARGSUSED*/
+static int
+bootfs_lookup(vnode_t *dvp, char *nm, vnode_t **vpp, struct pathname *pnp,
+    int flags, vnode_t *rdir, cred_t *cr, caller_context_t *ct,
+    int *direntflags, pathname_t *realpnp)
+{
+	avl_index_t where;
+	bootfs_node_t sn, *bnp;
+	bootfs_node_t *bpp = (bootfs_node_t *)dvp->v_data;
+
+	if (flags & LOOKUP_XATTR)
+		return (EINVAL);
+
+	if (bpp->bvn_attr.va_type != VDIR)
+		return (ENOTDIR);
+
+	if (*nm == '\0' || strcmp(nm, ".") == 0) {
+		VN_HOLD(dvp);
+		*vpp = dvp;
+		return (0);
+	}
+
+	if (strcmp(nm, "..") == 0) {
+		VN_HOLD(bpp->bvn_parent->bvn_vnp);
+		*vpp = bpp->bvn_parent->bvn_vnp;
+		return (0);
+	}
+
+	sn.bvn_name = nm;
+	bnp = avl_find(&bpp->bvn_dir, &sn, &where);
+	if (bnp == NULL)
+		return (ENOENT);
+
+	VN_HOLD(bnp->bvn_vnp);
+	*vpp = bnp->bvn_vnp;
+	return (0);
+}
+
+/*ARGSUSED*/
+static int
+bootfs_readdir(vnode_t *vp, struct uio *uiop, cred_t *cr, int *eofp,
+    caller_context_t *ct, int flags)
+{
+	bootfs_node_t *bnp = (bootfs_node_t *)vp->v_data;
+	dirent64_t *dp;
+	void *buf;
+	ulong_t bsize, brem;
+	offset_t coff, roff;
+	int dlen, ret;
+	bootfs_node_t *dnp;
+	boolean_t first = B_TRUE;
+
+	if (uiop->uio_loffset >= MAXOFF_T) {
+		if (eofp != NULL)
+			*eofp = 1;
+		return (0);
+	}
+
+	if (uiop->uio_iovcnt != 1)
+		return (EINVAL);
+
+	if (!(uiop->uio_iov->iov_len > 0))
+		return (EINVAL);
+
+	if (vp->v_type != VDIR)
+		return (ENOTDIR);
+
+	roff = uiop->uio_loffset;
+	coff = 0;
+	brem = bsize = uiop->uio_iov->iov_len;
+	buf = kmem_alloc(bsize, KM_SLEEP);
+	dp = buf;
+
+	/*
+	 * Recall that offsets here are done based on the name of the dirent
+	 * excluding the null terminator. Therefore `.` is always at 0, `..` is
+	 * always at 1, and then the first real dirent is at 3. This offset is
+	 * what's actually stored when we update the offset in the structure.
+	 */
+	if (roff == 0) {
+		dlen = DIRENT64_RECLEN(1);
+		if (first == B_TRUE) {
+			if (dlen > brem) {
+				kmem_free(buf, bsize);
+				return (EINVAL);
+			}
+			first = B_FALSE;
+		}
+		dp->d_ino = (ino64_t)bnp->bvn_attr.va_nodeid;
+		dp->d_off = 0;
+		dp->d_reclen = (ushort_t)dlen;
+		(void) strncpy(dp->d_name, ".", DIRENT64_NAMELEN(dlen));
+		dp = (struct dirent64 *)((uintptr_t)dp + dp->d_reclen);
+		brem -= dlen;
+	}
+
+	if (roff <= 1) {
+		dlen = DIRENT64_RECLEN(2);
+		if (first == B_TRUE) {
+			if (dlen > brem) {
+				kmem_free(buf, bsize);
+				return (EINVAL);
+			}
+			first = B_FALSE;
+		}
+		dp->d_ino = (ino64_t)bnp->bvn_parent->bvn_attr.va_nodeid;
+		dp->d_off = 1;
+		dp->d_reclen = (ushort_t)dlen;
+		(void) strncpy(dp->d_name, "..", DIRENT64_NAMELEN(dlen));
+		dp = (struct dirent64 *)((uintptr_t)dp + dp->d_reclen);
+		brem -= dlen;
+	}
+
+	coff = 3;
+	for (dnp = avl_first(&bnp->bvn_dir); dnp != NULL;
+	    dnp = AVL_NEXT(&bnp->bvn_dir, dnp)) {
+		size_t nlen = strlen(dnp->bvn_name);
+
+		if (roff > coff) {
+			coff += nlen;
+			continue;
+		}
+
+		dlen = DIRENT64_RECLEN(nlen);
+		if (dlen > brem) {
+			if (first == B_TRUE) {
+				kmem_free(buf, bsize);
+				return (EINVAL);
+			}
+			break;
+		}
+		first = B_FALSE;
+
+		dp->d_ino = (ino64_t)dnp->bvn_attr.va_nodeid;
+		dp->d_off = coff;
+		dp->d_reclen = (ushort_t)dlen;
+		(void) strncpy(dp->d_name, dnp->bvn_name,
+		    DIRENT64_NAMELEN(dlen));
+		dp = (struct dirent64 *)((uintptr_t)dp + dp->d_reclen);
+		brem -= dlen;
+		coff += nlen;
+	}
+
+	ret = uiomove(buf, (bsize - brem), UIO_READ, uiop);
+
+	if (ret == 0) {
+		if (dnp == NULL) {
+			coff++;
+			if (eofp != NULL)
+				*eofp = 1;
+		} else if (eofp != NULL) {
+			*eofp = 0;
+		}
+		uiop->uio_loffset = coff;
+	}
+	gethrestime(&bnp->bvn_attr.va_atime);
+	kmem_free(buf, bsize);
+	return (ret);
+}
+
+/*ARGSUSED*/
+static void
+bootfs_inactive(vnode_t *vp, cred_t *cr, caller_context_t *ct)
+{
+}
+
+/*ARGSUSED*/
+static int
+bootfs_rwlock(vnode_t *vp, int write_lock, caller_context_t *ct)
+{
+	if (write_lock != 0)
+		return (EINVAL);
+	return (0);
+}
+
+/*ARGSUSED*/
+static void
+bootfs_rwunlock(vnode_t *vp, int write_lock, caller_context_t *ct)
+{
+}
+
+/*ARGSUSED*/
+static int
+bootfs_seek(vnode_t *vp, offset_t ooff, offset_t *noffp,
+    caller_context_t *ct)
+{
+	bootfs_node_t *bnp = (bootfs_node_t *)vp->v_data;
+	if (vp->v_type == VDIR)
+		return (0);
+	return ((*noffp < 0 || *noffp > bnp->bvn_size ? EINVAL : 0));
+}
+
+/*
+ * We need to fill in a single page of a vnode's memory based on the actual data
+ * from the kernel. We'll use this node's sliding window into physical memory
+ * and update one page at a time.
+ */
+/*ARGSUSED*/
+static int
+bootfs_getapage(vnode_t *vp, u_offset_t off, size_t len, uint_t *protp,
+    page_t *pl[], size_t plsz, struct seg *seg, caddr_t addr, enum seg_rw rw,
+    cred_t *cr)
+{
+	bootfs_node_t *bnp = vp->v_data;
+	page_t *pp, *fpp;
+	pfn_t pfn;
+
+	for (;;) {
+		/* Easy case where the page exists */
+		pp = page_lookup(vp, off, rw == S_CREATE ? SE_EXCL : SE_SHARED);
+		if (pp != NULL) {
+			if (pl != NULL) {
+				pl[0] = pp;
+				pl[1] = NULL;
+			} else {
+				page_unlock(pp);
+			}
+			return (0);
+		}
+
+		pp = page_create_va(vp, off, PAGESIZE, PG_EXCL | PG_WAIT, seg,
+		    addr);
+
+		/*
+		 * If we didn't get the page, that means someone else beat us to
+		 * creating this so we need to try again.
+		 */
+		if (pp != NULL)
+			break;
+	}
+
+	pfn = btop((bnp->bvn_addr + off) & PAGEMASK);
+	fpp = page_numtopp_nolock(pfn);
+
+	if (ppcopy(fpp, pp) == 0) {
+		pvn_read_done(pp, B_ERROR);
+		return (EIO);
+	}
+
+	if (pl != NULL) {
+		pvn_plist_init(pp, pl, plsz, off, PAGESIZE, rw);
+	} else {
+		pvn_io_done(pp);
+	}
+
+	return (0);
+}
+
+/*ARGSUSED*/
+static int
+bootfs_getpage(vnode_t *vp, offset_t off, size_t len, uint_t *protp,
+    page_t *pl[], size_t plsz, struct seg *seg, caddr_t addr, enum seg_rw rw,
+    cred_t *cr, caller_context_t *ct)
+{
+	int err;
+	bootfs_node_t *bnp = vp->v_data;
+
+	if (off + len > bnp->bvn_size + PAGEOFFSET)
+		return (EFAULT);
+
+	if (len <= PAGESIZE)
+		err = bootfs_getapage(vp, (u_offset_t)off, len, protp, pl,
+		    plsz, seg, addr, rw, cr);
+	else
+		err = pvn_getpages(bootfs_getapage, vp, (u_offset_t)off, len,
+		    protp, pl, plsz, seg, addr, rw, cr);
+
+	return (err);
+}
+
+/*ARGSUSED*/
+static int
+bootfs_map(vnode_t *vp, offset_t off, struct as *as, caddr_t *addrp,
+    size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr,
+    caller_context_t *ct)
+{
+	int ret;
+	segvn_crargs_t vn_a;
+
+#ifdef	_ILP32
+	if (len > MAXOFF_T)
+		return (ENOMEM);
+#endif
+
+	if (vp->v_flag & VNOMAP)
+		return (ENOSYS);
+
+	if (off < 0 || off > MAXOFFSET_T - off)
+		return (ENXIO);
+
+	if (vp->v_type != VREG)
+		return (ENODEV);
+
+	if (prot & PROT_WRITE)
+		return (ENOTSUP);
+
+	as_rangelock(as);
+	ret = choose_addr(as, addrp, len, off, ADDR_VACALIGN, flags);
+	if (ret != 0) {
+		as_rangeunlock(as);
+		return (ret);
+	}
+
+	vn_a.vp = vp;
+	vn_a.offset = (u_offset_t)off;
+	vn_a.type = flags & MAP_TYPE;
+	vn_a.prot = prot;
+	vn_a.maxprot = maxprot;
+	vn_a.cred = cr;
+	vn_a.amp = NULL;
+	vn_a.flags = flags & ~MAP_TYPE;
+	vn_a.szc = 0;
+	vn_a.lgrp_mem_policy_flags = 0;
+
+	ret = as_map(as, *addrp, len, segvn_create, &vn_a);
+
+	as_rangeunlock(as);
+	return (ret);
+
+}
+
+/*ARGSUSED*/
+static int
+bootfs_addmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr,
+    size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr,
+    caller_context_t *ct)
+{
+	return (0);
+}
+
+/*ARGSUSED*/
+static int
+bootfs_delmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr,
+    size_t len, uint_t prot, uint_t maxprot, uint_t flags, cred_t *cr,
+    caller_context_t *ct)
+{
+	return (0);
+}
+
+static int
+bootfs_pathconf(vnode_t *vp, int cmd, ulong_t *valp, cred_t *cr,
+    caller_context_t *ct)
+{
+	int ret;
+
+	switch (cmd) {
+	case _PC_TIMESTAMP_RESOLUTION:
+		*valp = 1L;
+		ret = 0;
+		break;
+	default:
+		ret = fs_pathconf(vp, cmd, valp, cr, ct);
+	}
+
+	return (ret);
+}
+
+const fs_operation_def_t bootfs_vnodeops_template[] = {
+	VOPNAME_OPEN,		{ .vop_open = bootfs_open },
+	VOPNAME_CLOSE,		{ .vop_close = bootfs_close },
+	VOPNAME_READ,		{ .vop_read = bootfs_read },
+	VOPNAME_IOCTL,		{ .vop_ioctl = bootfs_ioctl },
+	VOPNAME_GETATTR,	{ .vop_getattr = bootfs_getattr },
+	VOPNAME_ACCESS,		{ .vop_access = bootfs_access },
+	VOPNAME_LOOKUP,		{ .vop_lookup = bootfs_lookup },
+	VOPNAME_READDIR,	{ .vop_readdir = bootfs_readdir },
+	VOPNAME_INACTIVE,	{ .vop_inactive = bootfs_inactive },
+	VOPNAME_RWLOCK,		{ .vop_rwlock = bootfs_rwlock },
+	VOPNAME_RWUNLOCK,	{ .vop_rwunlock = bootfs_rwunlock },
+	VOPNAME_SEEK,		{ .vop_seek = bootfs_seek },
+	VOPNAME_GETPAGE,	{ .vop_getpage = bootfs_getpage },
+	VOPNAME_MAP,		{ .vop_map = bootfs_map },
+	VOPNAME_ADDMAP,		{ .vop_addmap = bootfs_addmap },
+	VOPNAME_DELMAP,		{ .vop_delmap = bootfs_delmap },
+	VOPNAME_PATHCONF,	{ .vop_pathconf = bootfs_pathconf },
+	VOPNAME_VNEVENT,	{ .vop_vnevent = fs_vnevent_nosupport },
+	NULL,			NULL
+};
diff --git a/usr/src/uts/common/fs/dev/sdev_netops.c b/usr/src/uts/common/fs/dev/sdev_netops.c
index 4eaf38f484..41441ec52d 100644
--- a/usr/src/uts/common/fs/dev/sdev_netops.c
+++ b/usr/src/uts/common/fs/dev/sdev_netops.c
@@ -21,6 +21,7 @@
 /*
  * Copyright 2009 Sun Microsystems, Inc.  All rights reserved.
  * Use is subject to license terms.
+ * Copyright (c) 2014, Joyent, Inc.  All rights reserved.
  */
 
 /*
@@ -41,8 +42,102 @@
 #include <sys/zone.h>
 #include <sys/dls.h>
 
+static const char *devnet_zpath = "/dev/net/zone/";
 struct vnodeops		*devnet_vnodeops;
 
+static zoneid_t
+devnet_nodetozone(sdev_node_t *dv)
+{
+	char *zname = NULL, *dup;
+	zone_t *zone;
+	int duplen;
+	zoneid_t zid;
+
+	/*
+	 * If in a non-global zone, always return it's zid no matter what the
+	 * node is.
+	 */
+	zid = getzoneid();
+	if (zid != GLOBAL_ZONEID)
+		return (zid);
+
+	/*
+	 * If it doesn't have /dev/net/zone/ then it can't be a specific zone
+	 * we're targetting.
+	 */
+	if (strncmp(devnet_zpath, dv->sdev_path, strlen(devnet_zpath)) != 0)
+		return (GLOBAL_ZONEID);
+
+	if (dv->sdev_vnode->v_type == VDIR) {
+		zone = zone_find_by_name(dv->sdev_name);
+	} else {
+		/* Non directories have the form /dev/net/zone/%z/%s */
+		dup = strdup(dv->sdev_path);
+		duplen = strlen(dup);
+		zname = strrchr(dup, '/');
+		*zname = '\0';
+		zname--;
+		zname = strrchr(dup, '/');
+		zname++;
+		zone = zone_find_by_name(zname);
+		kmem_free(dup, duplen + 1);
+	}
+	if (zone == NULL)
+		return (GLOBAL_ZONEID);
+	zid = zone->zone_id;
+	zone_rele(zone);
+	return (zid);
+}
+
+static int
+devnet_mkdir(struct sdev_node *ddv, char *name)
+{
+	sdev_node_t *dv;
+	struct vattr va;
+	int ret;
+
+	ASSERT(RW_WRITE_HELD(&ddv->sdev_contents));
+	dv = sdev_cache_lookup(ddv, name);
+	if (dv != NULL) {
+		SDEV_SIMPLE_RELE(dv);
+		return (EEXIST);
+	}
+
+	va = *sdev_getdefault_attr(VDIR);
+	gethrestime(&va.va_atime);
+	va.va_mtime = va.va_atime;
+	va.va_ctime = va.va_atime;
+
+	ret = sdev_mknode(ddv, name, &dv, &va, NULL, NULL, kcred, SDEV_READY);
+	if (ret != 0)
+		return (ret);
+	SDEV_SIMPLE_RELE(dv);
+	return (0);
+}
+
+/*
+ * We basically need to walk down the directory path to determine what we should
+ * do. At the top level of /dev/net, only the directory /dev/net/zone is valid,
+ * and it is always valid. Following on that, /dev/net/zone/%zonename is valid
+ * if and only if we can look up that zone name. If it's not, or it's some other
+ * name, then it's SDEV_VTOR_INVALID.
+ */
+static int
+devnet_dirvalidate(struct sdev_node *dv)
+{
+	zone_t *zonep;
+	char *path = "/dev/net/zone";
+
+	if (strcmp(path, dv->sdev_path) == 0)
+		return (SDEV_VTOR_VALID);
+
+	zonep = zone_find_by_name(dv->sdev_name);
+	if (zonep == NULL)
+		return (SDEV_VTOR_INVALID);
+	zone_rele(zonep);
+	return (SDEV_VTOR_VALID);
+}
+
 /*
  * Check if a net sdev_node is still valid - i.e. it represents a current
  * network link.
@@ -60,11 +155,20 @@ devnet_validate(struct sdev_node *dv)
 
 	ASSERT(dv->sdev_state == SDEV_READY);
 
-	if (dls_mgmt_get_linkid(dv->sdev_name, &linkid) != 0)
+	if (dv->sdev_vnode->v_type == VDIR)
+		return (devnet_dirvalidate(dv));
+
+	if (strncmp(devnet_zpath, dv->sdev_path, strlen(devnet_zpath)) == 0) {
+		ASSERT(SDEV_IS_GLOBAL(dv));
+		zoneid = devnet_nodetozone(dv);
+	} else {
+		zoneid = getzoneid();
+	}
+
+	if (dls_mgmt_get_linkid_in_zone(dv->sdev_name, &linkid, zoneid) != 0)
 		return (SDEV_VTOR_INVALID);
-	if (SDEV_IS_GLOBAL(dv))
+	if (zoneid == GLOBAL_ZONEID)
 		return (SDEV_VTOR_VALID);
-	zoneid = getzoneid();
 	return (zone_check_datalink(&zoneid, linkid) == 0 ?
 	    SDEV_VTOR_VALID : SDEV_VTOR_INVALID);
 }
@@ -74,13 +178,14 @@ devnet_validate(struct sdev_node *dv)
  * a net entry when the node is not found in the cache.
  */
 static int
-devnet_create_rvp(const char *nm, struct vattr *vap, dls_dl_handle_t *ddhp)
+devnet_create_rvp(const char *nm, struct vattr *vap, dls_dl_handle_t *ddhp,
+    zoneid_t zid)
 {
 	timestruc_t now;
 	dev_t dev;
 	int error;
 
-	if ((error = dls_devnet_open(nm, ddhp, &dev)) != 0) {
+	if ((error = dls_devnet_open_in_zone(nm, ddhp, &dev, zid)) != 0) {
 		sdcmn_err12(("devnet_create_rvp: not a valid vanity name "
 		    "network node: %s\n", nm));
 		return (error);
@@ -116,6 +221,7 @@ devnet_lookup(struct vnode *dvp, char *nm, struct vnode **vpp,
 	struct sdev_node *ddv = VTOSDEV(dvp);
 	struct sdev_node *dv = NULL;
 	dls_dl_handle_t ddh = NULL;
+	zone_t *zone;
 	struct vattr vattr;
 	int nmlen;
 	int error = ENOENT;
@@ -123,6 +229,9 @@ devnet_lookup(struct vnode *dvp, char *nm, struct vnode **vpp,
 	if (SDEVTOV(ddv)->v_type != VDIR)
 		return (ENOTDIR);
 
+	if (!SDEV_IS_GLOBAL(ddv) && crgetzoneid(cred) == GLOBAL_ZONEID)
+		return (EPERM);
+
 	/*
 	 * Empty name or ., return node itself.
 	 */
@@ -145,6 +254,12 @@ devnet_lookup(struct vnode *dvp, char *nm, struct vnode **vpp,
 	rw_enter(&ddv->sdev_contents, RW_WRITER);
 
 	/*
+	 * ZOMBIED parent does not allow new node creation, bail out early.
+	 */
+	if (ddv->sdev_state == SDEV_ZOMBIE)
+		goto failed;
+
+	/*
 	 * directory cache lookup:
 	 */
 	if ((dv = sdev_cache_lookup(ddv, nm)) != NULL) {
@@ -153,13 +268,42 @@ devnet_lookup(struct vnode *dvp, char *nm, struct vnode **vpp,
 			goto found;
 	}
 
+	if (SDEV_IS_GLOBAL(ddv)) {
+		/*
+		 * Check for /dev/net/zone
+		 */
+		if (strcmp("zone", nm) == 0 && strcmp("/dev/net",
+		    ddv->sdev_path) == 0) {
+			(void) devnet_mkdir(ddv, nm);
+			dv = sdev_cache_lookup(ddv, nm);
+			ASSERT(dv != NULL);
+			goto found;
+		}
+
+		/*
+		 * Check for /dev/net/zone/%z. We can't use devnet_zpath due to
+		 * its trailing slash.
+		 */
+		if (strcmp("/dev/net/zone", ddv->sdev_path) == 0) {
+			zone = zone_find_by_name(nm);
+			if (zone == NULL)
+				goto failed;
+			(void) devnet_mkdir(ddv, nm);
+			zone_rele(zone);
+			dv = sdev_cache_lookup(ddv, nm);
+			ASSERT(dv != NULL);
+			goto found;
+		}
+	} else if (strcmp("/dev/net", ddv->sdev_path) != 0) {
+		goto failed;
+	}
+
 	/*
-	 * ZOMBIED parent does not allow new node creation, bail out early.
+	 * We didn't find what we were looking for. What that is depends a lot
+	 * on what directory we're in.
 	 */
-	if (ddv->sdev_state == SDEV_ZOMBIE)
-		goto failed;
 
-	error = devnet_create_rvp(nm, &vattr, &ddh);
+	error = devnet_create_rvp(nm, &vattr, &ddh, devnet_nodetozone(ddv));
 	if (error != 0)
 		goto failed;
 
@@ -219,7 +363,7 @@ devnet_filldir_datalink(datalink_id_t linkid, void *arg)
 	if ((dv = sdev_cache_lookup(ddv, (char *)link)) != NULL)
 		goto found;
 
-	if (devnet_create_rvp(link, &vattr, &ddh) != 0)
+	if (devnet_create_rvp(link, &vattr, &ddh, devnet_nodetozone(arg)) != 0)
 		return (0);
 
 	ASSERT(ddh != NULL);
@@ -244,16 +388,77 @@ found:
 	return (0);
 }
 
+/*
+ * Fill in all the entries for the current zone.
+ */
 static void
-devnet_filldir(struct sdev_node *ddv)
+devnet_fillzone(struct sdev_node *ddv, zoneid_t zid)
 {
-	sdev_node_t	*dv, *next;
 	datalink_id_t	linkid;
 
+	ASSERT(RW_WRITE_HELD(&ddv->sdev_contents));
+	if (zid == GLOBAL_ZONEID) {
+		ASSERT(SDEV_IS_GLOBAL(ddv));
+		linkid = DATALINK_INVALID_LINKID;
+		do {
+			linkid = dls_mgmt_get_next(linkid, DATALINK_CLASS_ALL,
+			    DATALINK_ANY_MEDIATYPE, DLMGMT_ACTIVE);
+			if (linkid != DATALINK_INVALID_LINKID)
+				(void) devnet_filldir_datalink(linkid, ddv);
+		} while (linkid != DATALINK_INVALID_LINKID);
+	} else {
+		(void) zone_datalink_walk(zid,  devnet_filldir_datalink, ddv);
+	}
+}
+
+/*
+ * Callback for zone_walk when filling up /dev/net/zone/...
+ */
+static int
+devnet_fillzdir_cb(zone_t *zonep, void *arg)
+{
+	sdev_node_t *ddv = arg;
+
+	ASSERT(RW_WRITE_HELD(&ddv->sdev_contents));
+	(void) devnet_mkdir(ddv, zonep->zone_name);
+	return (0);
+}
+
+/*
+ * Fill in a directory that isn't the top level /dev/net.
+ */
+static void
+devnet_fillzdir(struct sdev_node *ddv)
+{
+	zone_t *zonep;
+	char *path = "/dev/net/zone";
+
+	if (strcmp(path, ddv->sdev_path) == 0) {
+		(void) zone_walk(devnet_fillzdir_cb, ddv);
+		return;
+	}
+
+	zonep = zone_find_by_name(ddv->sdev_name);
+	if (zonep == NULL)
+		return;
+	devnet_fillzone(ddv, zonep->zone_id);
+	zone_rele(zonep);
+}
+
+static void
+devnet_filldir(struct sdev_node *ddv)
+{
+	int ret;
+	sdev_node_t *dv, *next;
+
 	ASSERT(RW_READ_HELD(&ddv->sdev_contents));
 	if (rw_tryupgrade(&ddv->sdev_contents) == NULL) {
 		rw_exit(&ddv->sdev_contents);
 		rw_enter(&ddv->sdev_contents, RW_WRITER);
+		if (ddv->sdev_state == SDEV_ZOMBIE) {
+			rw_exit(&ddv->sdev_contents);
+			return;
+		}
 	}
 
 	for (dv = SDEV_FIRST_ENTRY(ddv); dv; dv = next) {
@@ -276,31 +481,38 @@ devnet_filldir(struct sdev_node *ddv)
 
 		if (SDEVTOV(dv)->v_count > 0)
 			continue;
+
 		SDEV_HOLD(dv);
+
+		/*
+		 * Clean out everything underneath before we remove ourselves.
+		 */
+		if (SDEVTOV(ddv)->v_type == VDIR) {
+			ret = sdev_cleandir(dv, NULL, 0);
+			ASSERT(ret == 0);
+		}
 		/* remove the cache node */
 		(void) sdev_cache_update(ddv, &dv, dv->sdev_name,
 		    SDEV_CACHE_DELETE);
 		SDEV_RELE(dv);
 	}
 
+	if (strcmp(ddv->sdev_path, "/dev/net") != 0) {
+		devnet_fillzdir(ddv);
+		goto done;
+	}
+
 	if (((ddv->sdev_flags & SDEV_BUILD) == 0) && !dls_devnet_rebuild())
 		goto done;
 
 	if (SDEV_IS_GLOBAL(ddv)) {
-		linkid = DATALINK_INVALID_LINKID;
-		do {
-			linkid = dls_mgmt_get_next(linkid, DATALINK_CLASS_ALL,
-			    DATALINK_ANY_MEDIATYPE, DLMGMT_ACTIVE);
-			if (linkid != DATALINK_INVALID_LINKID)
-				(void) devnet_filldir_datalink(linkid, ddv);
-		} while (linkid != DATALINK_INVALID_LINKID);
+		devnet_fillzone(ddv, GLOBAL_ZONEID);
+		(void) devnet_mkdir(ddv, "zone");
 	} else {
-		(void) zone_datalink_walk(getzoneid(),
-		    devnet_filldir_datalink, ddv);
+		devnet_fillzone(ddv, getzoneid());
 	}
 
 	ddv->sdev_flags &= ~SDEV_BUILD;
-
 done:
 	rw_downgrade(&ddv->sdev_contents);
 }
@@ -319,6 +531,9 @@ devnet_readdir(struct vnode *dvp, struct uio *uiop, struct cred *cred,
 
 	ASSERT(sdvp);
 
+	if (crgetzoneid(cred) == GLOBAL_ZONEID && !SDEV_IS_GLOBAL(sdvp))
+		return (EPERM);
+
 	if (uiop->uio_offset == 0)
 		devnet_filldir(sdvp);
 
diff --git a/usr/src/uts/common/fs/dev/sdev_plugin.c b/usr/src/uts/common/fs/dev/sdev_plugin.c
new file mode 100644
index 0000000000..885191175f
--- /dev/null
+++ b/usr/src/uts/common/fs/dev/sdev_plugin.c
@@ -0,0 +1,913 @@
+/*
+ * This file and its contents are supplied under the terms of the
+ * Common Development and Distribution License ("CDDL"), version 1.0.
+ * You may only use this file in accordance with the terms of version
+ * 1.0 of the CDDL.
+ *
+ * A full copy of the text of the CDDL should have accompanied this
+ * source.  A copy of the CDDL is also available via the Internet at
+ * http://www.illumos.org/license/CDDL.
+ */
+
+/*
+ * Copyright (c) 2014, Joyent, Inc. All rights reserved.
+ */
+
+/*
+ * Dynamic directory plugin interface for sdev.
+ *
+ * The sdev plugin interfaces provides a means for a dynamic directory based on
+ * in-kernel state to be simply created. Traditionally, dynamic directories were
+ * built into sdev itself. While these legacy plugins are useful, it makes more
+ * sense for these pieces of functionality to live with the individual drivers.
+ *
+ * The plugin interface requires folks to implement three interfaces and
+ * provides a series of callbacks that can be made in the context of those
+ * interfaces to interrogate the sdev_node_t without having to leak
+ * implementation details of the sdev_node_t. These interfaces are:
+ *
+ *   o spo_validate
+ *
+ *   Given a particular node, answer the question as to whether or not this
+ *   entry is still valid. Here, plugins should use the name and the dev_t
+ *   associated with the node to verify that it matches something that still
+ *   exists.
+ *
+ *   o spo_filldir
+ *
+ *   Fill all the entries inside of a directory. Note that some of these entries
+ *   may already exist.
+ *
+ *   o spo_inactive
+ *
+ *   The given node is no longer being used. This allows the consumer to
+ *   potentially tear down anything that was being held open related to this.
+ *   Note that this only fires when the given sdev_node_t becomes a zombie.
+ *
+ * During these callbacks a consumer is not allowed to register or unregister a
+ * plugin, especially their own. They may call the sdev_ctx style functions. All
+ * callbacks fire in a context where blocking is allowed (eg. the spl is below
+ * LOCK_LEVEL).
+ *
+ * When a plugin is added, we create its directory in the global zone. By doing
+ * that, we ensure that something isn't already there and that nothing else can
+ * come along and try and create something without our knowledge. We only have
+ * to create it in the GZ and not for all other instances of sdev because an
+ * instance of sdev that isn't at /dev does not have dynamic directories, and
+ * second, any instance of sdev present in a non-global zone cannot create
+ * anything, therefore we know that by it not being in the global zone's
+ * instance of sdev that we're good to go.
+ *
+ * Lock Ordering
+ * -------------
+ *
+ * The global sdev_plugin_lock must be held before any of the individual
+ * sdev_plugin_t`sp_lock. Further, once any plugin related lock has been held,
+ * it is not legal to take any holds on any sdev_node_t or to grab the
+ * sdev_node_t`contents_lock in any way.
+ */
+
+#include <sys/types.h>
+#include <sys/stat.h>
+#include <sys/fs/sdev_impl.h>
+#include <sys/fs/sdev_plugin.h>
+#include <fs/fs_subr.h>
+#include <sys/ddi.h>
+#include <sys/sunddi.h>
+#include <sys/ksynch.h>
+#include <sys/sysmacros.h>
+#include <sys/list.h>
+#include <sys/ctype.h>
+
+kmutex_t sdev_plugin_lock;
+list_t sdev_plugin_list;
+kmem_cache_t *sdev_plugin_cache;
+struct vnodeops *sdev_plugin_vnops;
+
+#define	SDEV_PLUGIN_NAMELEN	64
+
+typedef struct sdev_plugin {
+	list_node_t sp_link;
+	char sp_name[SDEV_PLUGIN_NAMELEN];	/* E */
+	int sp_nflags;				/* E */
+	struct vnodeops *sp_vnops;		/* E */
+	sdev_plugin_ops_t *sp_pops;		/* E */
+	boolean_t sp_islegacy;			/* E */
+	int (*sp_lvtor)(sdev_node_t *);		/* E */
+	kmutex_t sp_lock;			/* Protects everything below */
+	kcondvar_t sp_nodecv;
+	size_t sp_nnodes;
+} sdev_plugin_t;
+
+/* ARGSUSED */
+static int
+sdev_plugin_cache_constructor(void *buf, void *arg, int tags)
+{
+	sdev_plugin_t *spp = buf;
+	mutex_init(&spp->sp_lock, NULL, MUTEX_DRIVER, 0);
+	cv_init(&spp->sp_nodecv, NULL, CV_DRIVER, NULL);
+	return (0);
+}
+
+/* ARGSUSED */
+static void
+sdev_plugin_cache_destructor(void *buf, void *arg)
+{
+	sdev_plugin_t *spp = buf;
+	cv_destroy(&spp->sp_nodecv);
+	mutex_destroy(&spp->sp_lock);
+}
+
+enum vtype
+sdev_ctx_vtype(sdev_ctx_t ctx)
+{
+	sdev_node_t *sdp = (sdev_node_t *)ctx;
+
+	ASSERT(RW_LOCK_HELD(&sdp->sdev_contents));
+	return (sdp->sdev_vnode->v_type);
+}
+
+const char *
+sdev_ctx_path(sdev_ctx_t ctx)
+{
+	sdev_node_t *sdp = (sdev_node_t *)ctx;
+
+	ASSERT(RW_LOCK_HELD(&sdp->sdev_contents));
+	return (sdp->sdev_path);
+}
+
+const char *
+sdev_ctx_name(sdev_ctx_t ctx)
+{
+	sdev_node_t *sdp = (sdev_node_t *)ctx;
+
+	ASSERT(RW_LOCK_HELD(&sdp->sdev_contents));
+	return (sdp->sdev_name);
+}
+
+/*
+ * Currently we only support psasing through a single flag -- SDEV_IS_GLOBAL.
+ */
+sdev_ctx_flags_t
+sdev_ctx_flags(sdev_ctx_t ctx)
+{
+	sdev_node_t *sdp = (sdev_node_t *)ctx;
+
+	ASSERT(RW_LOCK_HELD(&sdp->sdev_contents));
+	return (sdp->sdev_flags & SDEV_GLOBAL);
+}
+
+/*
+ * Return some amount of private data specific to the vtype. In the case of a
+ * character or block device this is the device number.
+ */
+const void *
+sdev_ctx_vtype_data(sdev_ctx_t ctx)
+{
+	sdev_node_t *sdp = (sdev_node_t *)ctx;
+	void *ret;
+
+	ASSERT(RW_LOCK_HELD(&sdp->sdev_contents));
+	switch (sdp->sdev_vnode->v_type) {
+	case VCHR:
+	case VBLK:
+		ret = (void *)(uintptr_t)(sdp->sdev_vnode->v_rdev);
+		break;
+	default:
+		ret = NULL;
+		break;
+	}
+
+	return (ret);
+}
+
+/*
+ * Use the same rules as zones for a name. isalphanum + '-', '_', and '.'.
+ */
+static int
+sdev_plugin_name_isvalid(const char *c, int buflen)
+{
+	int i;
+
+	for (i = 0; i < buflen; i++, c++) {
+		if (*c == '\0')
+			return (1);
+
+		if (!isalnum(*c) && *c != '-' && *c != '_' && *c != '.')
+			return (0);
+	}
+	/* Never found a null terminator */
+	return (0);
+}
+
+static int
+sdev_plugin_mknode(sdev_plugin_t *spp, sdev_node_t *sdvp, char *name,
+    vattr_t *vap)
+{
+	int ret;
+	sdev_node_t *svp;
+
+	ASSERT(RW_WRITE_HELD(&sdvp->sdev_contents));
+	ASSERT(spp != NULL);
+	svp = sdev_cache_lookup(sdvp, name);
+	if (svp != NULL) {
+		SDEV_SIMPLE_RELE(svp);
+		return (EEXIST);
+	}
+
+	ret = sdev_mknode(sdvp, name, &svp, vap, NULL, NULL, kcred,
+	    SDEV_READY);
+	if (ret != 0)
+		return (ret);
+	SDEV_SIMPLE_RELE(svp);
+
+	return (0);
+}
+
+/*
+ * Plugin node creation callbacks
+ */
+int
+sdev_plugin_mkdir(sdev_ctx_t ctx, char *name)
+{
+	sdev_node_t *sdvp;
+	timestruc_t now;
+	struct vattr vap;
+
+	if (sdev_plugin_name_isvalid(name, SDEV_PLUGIN_NAMELEN) == 0)
+		return (EINVAL);
+
+	sdvp = (sdev_node_t *)ctx;
+	ASSERT(sdvp->sdev_private != NULL);
+	ASSERT(RW_WRITE_HELD(&sdvp->sdev_contents));
+
+	vap = *sdev_getdefault_attr(VDIR);
+	gethrestime(&now);
+	vap.va_atime = now;
+	vap.va_mtime = now;
+	vap.va_ctime = now;
+
+	return (sdev_plugin_mknode(sdvp->sdev_private, sdvp, name, &vap));
+}
+
+int
+sdev_plugin_mknod(sdev_ctx_t ctx, char *name, mode_t mode, dev_t dev)
+{
+	sdev_node_t *sdvp;
+	timestruc_t now;
+	struct vattr vap;
+
+	if (sdev_plugin_name_isvalid(name, SDEV_PLUGIN_NAMELEN) == 0)
+		return (EINVAL);
+
+	sdvp = (sdev_node_t *)ctx;
+	ASSERT(RW_WRITE_HELD(&sdvp->sdev_contents));
+	if (mode != S_IFCHR && mode != S_IFBLK)
+		return (EINVAL);
+
+	ASSERT(sdvp->sdev_private != NULL);
+
+	vap = *sdev_getdefault_attr(mode == S_IFCHR ? VCHR : VBLK);
+	gethrestime(&now);
+	vap.va_atime = now;
+	vap.va_mtime = now;
+	vap.va_ctime = now;
+	vap.va_rdev = dev;
+	vap.va_mode = mode | 0666;
+
+	/* Despite the similar name, this is in fact a different function */
+	return (sdev_plugin_mknode(sdvp->sdev_private, sdvp, name, &vap));
+
+}
+
+static int
+sdev_plugin_validate(sdev_node_t *sdp)
+{
+	int ret;
+	sdev_plugin_t *spp;
+
+	ASSERT(sdp->sdev_private != NULL);
+	spp = sdp->sdev_private;
+	ASSERT(spp->sp_islegacy == B_FALSE);
+	ASSERT(spp->sp_pops != NULL);
+	rw_enter(&sdp->sdev_contents, RW_READER);
+	ret = spp->sp_pops->spo_validate((uintptr_t)sdp);
+	rw_exit(&sdp->sdev_contents);
+	return (ret);
+}
+
+static void
+sdev_plugin_validate_dir(sdev_node_t *sdvp)
+{
+	int ret;
+	sdev_node_t *svp, *next;
+
+	ASSERT(RW_WRITE_HELD(&sdvp->sdev_contents));
+
+	for (svp = SDEV_FIRST_ENTRY(sdvp); svp != NULL; svp = next) {
+
+		next = SDEV_NEXT_ENTRY(sdvp, svp);
+		ASSERT(svp->sdev_state != SDEV_ZOMBIE);
+		/* skip nodes that aren't ready */
+		if (svp->sdev_state == SDEV_INIT)
+			continue;
+
+		switch (sdev_plugin_validate(svp)) {
+		case SDEV_VTOR_VALID:
+		case SDEV_VTOR_SKIP:
+			continue;
+		case SDEV_VTOR_INVALID:
+		case SDEV_VTOR_STALE:
+			break;
+		}
+
+		SDEV_HOLD(svp);
+
+		/*
+		 * Clean out everything underneath this node before we
+		 * remove it.
+		 */
+		if (svp->sdev_vnode->v_type == VDIR) {
+			ret = sdev_cleandir(svp, NULL, 0);
+			ASSERT(ret == 0);
+		}
+		/* remove the cache node */
+		(void) sdev_cache_update(sdvp, &svp, svp->sdev_name,
+		    SDEV_CACHE_DELETE);
+		SDEV_RELE(svp);
+	}
+}
+
+/* ARGSUSED */
+static int
+sdev_plugin_vop_readdir(struct vnode *dvp, struct uio *uiop, struct cred *cred,
+    int *eofp, caller_context_t *ct_unused, int flags_unused)
+{
+	int ret;
+	sdev_node_t *sdvp = VTOSDEV(dvp);
+	sdev_plugin_t *spp;
+
+	ASSERT(RW_READ_HELD(&sdvp->sdev_contents));
+
+	/* Sanity check we're not a zombie before we do anyting else */
+	if (sdvp->sdev_state == SDEV_ZOMBIE)
+		return (ENOENT);
+
+	spp = sdvp->sdev_private;
+	ASSERT(spp != NULL);
+	ASSERT(spp->sp_islegacy == B_FALSE);
+	ASSERT(spp->sp_pops != NULL);
+
+	if (crgetzoneid(cred) == GLOBAL_ZONEID && !SDEV_IS_GLOBAL(sdvp))
+		return (EPERM);
+
+	if (uiop->uio_offset == 0) {
+		/*
+		 * We upgrade to a write lock and grab the plugin's lock along
+		 * the way. We're almost certainly going to get creation
+		 * callbacks, so this is the only safe way to go.
+		 */
+		if (rw_tryupgrade(&sdvp->sdev_contents) == 0) {
+			rw_exit(&sdvp->sdev_contents);
+			rw_enter(&sdvp->sdev_contents, RW_WRITER);
+			if (sdvp->sdev_state == SDEV_ZOMBIE) {
+				rw_downgrade(&sdvp->sdev_contents);
+				return (ENOENT);
+			}
+		}
+
+		sdev_plugin_validate_dir(sdvp);
+		ret = spp->sp_pops->spo_filldir((uintptr_t)sdvp);
+		rw_downgrade(&sdvp->sdev_contents);
+		if (ret != 0)
+			return (ret);
+	}
+
+	return (devname_readdir_func(dvp, uiop, cred, eofp, 0));
+}
+
+/*
+ * If we don't have a callback function that returns a failure, then sdev will
+ * try to create a node for us which violates all of our basic assertions. To
+ * work around that we create our own callback for devname_lookup_func which
+ * always returns ENOENT as at this point either it was created with the filldir
+ * callback or it was not.
+ */
+/*ARGSUSED*/
+static int
+sdev_plugin_vop_lookup_cb(sdev_node_t *ddv, char *nm, void **arg, cred_t *cred,
+    void *unused, char *unused2)
+{
+	return (ENOENT);
+}
+
+/* ARGSUSED */
+static int
+sdev_plugin_vop_lookup(struct vnode *dvp, char *nm, struct vnode **vpp,
+    struct pathname *pnp, int flags, struct vnode *rdir, struct cred *cred,
+    caller_context_t *ct, int *direntflags, pathname_t *realpnp)
+{
+	int ret;
+	sdev_node_t *sdvp;
+	sdev_plugin_t *spp;
+
+	/* execute access is required to search the directory */
+	if ((ret = VOP_ACCESS(dvp, VEXEC, 0, cred, ct)) != 0)
+		return (ret);
+
+	sdvp = VTOSDEV(dvp);
+	spp = sdvp->sdev_private;
+	ASSERT(spp != NULL);
+	ASSERT(spp->sp_islegacy == B_FALSE);
+	ASSERT(spp->sp_pops != NULL);
+
+	if (crgetzoneid(cred) == GLOBAL_ZONEID && !SDEV_IS_GLOBAL(sdvp))
+		return (EPERM);
+
+	/*
+	 * Go straight for the write lock.
+	 */
+	rw_enter(&sdvp->sdev_contents, RW_WRITER);
+	if (sdvp->sdev_state == SDEV_ZOMBIE) {
+		rw_exit(&sdvp->sdev_contents);
+		return (ENOENT);
+	}
+	sdev_plugin_validate_dir(sdvp);
+	ret = spp->sp_pops->spo_filldir((uintptr_t)sdvp);
+	rw_exit(&sdvp->sdev_contents);
+	if (ret != 0)
+		return (ret);
+
+	return (devname_lookup_func(sdvp, nm, vpp, cred,
+	    sdev_plugin_vop_lookup_cb, SDEV_VATTR));
+}
+
+/*
+ * sdev is not a good citizen. We get inactive callbacks whenever a vnode goes
+ * to zero, but isn't necessairily a zombie yet. As such, to make things easier
+ * for users, we only fire the inactive callback when the node becomes a zombie
+ * and thus will be torn down here.
+ */
+static void
+sdev_plugin_vop_inactive_cb(struct vnode *dvp)
+{
+	sdev_node_t *sdp = VTOSDEV(dvp);
+	sdev_plugin_t *spp = sdp->sdev_private;
+
+	rw_enter(&sdp->sdev_contents, RW_READER);
+	if (sdp->sdev_state != SDEV_ZOMBIE) {
+		rw_exit(&sdp->sdev_contents);
+		return;
+	}
+	spp->sp_pops->spo_inactive((uintptr_t)sdp);
+	mutex_enter(&spp->sp_lock);
+	VERIFY(spp->sp_nnodes > 0);
+	spp->sp_nnodes--;
+	cv_signal(&spp->sp_nodecv);
+	mutex_exit(&spp->sp_lock);
+	rw_exit(&sdp->sdev_contents);
+}
+
+/*ARGSUSED*/
+static void
+sdev_plugin_vop_inactive(struct vnode *dvp, struct cred *cred,
+    caller_context_t *ct)
+{
+	sdev_node_t *sdp = VTOSDEV(dvp);
+	sdev_plugin_t *spp = sdp->sdev_private;
+	ASSERT(sdp->sdev_private != NULL);
+	ASSERT(spp->sp_islegacy == B_FALSE);
+	devname_inactive_func(dvp, cred, sdev_plugin_vop_inactive_cb);
+}
+
+const fs_operation_def_t sdev_plugin_vnodeops_tbl[] = {
+	VOPNAME_READDIR,	{ .vop_readdir = sdev_plugin_vop_readdir },
+	VOPNAME_LOOKUP,		{ .vop_lookup = sdev_plugin_vop_lookup },
+	VOPNAME_INACTIVE,	{ .vop_inactive = sdev_plugin_vop_inactive },
+	VOPNAME_CREATE,		{ .error = fs_nosys },
+	VOPNAME_REMOVE,		{ .error = fs_nosys },
+	VOPNAME_MKDIR,		{ .error = fs_nosys },
+	VOPNAME_RMDIR,		{ .error = fs_nosys },
+	VOPNAME_SYMLINK,	{ .error = fs_nosys },
+	VOPNAME_SETSECATTR,	{ .error = fs_nosys },
+	NULL,			NULL
+};
+
+/*
+ * construct a new template with overrides from vtab
+ */
+static fs_operation_def_t *
+sdev_merge_vtab(const fs_operation_def_t tab[])
+{
+	fs_operation_def_t *new;
+	const fs_operation_def_t *tab_entry;
+
+	/* make a copy of standard vnode ops table */
+	new = kmem_alloc(sdev_vnodeops_tbl_size, KM_SLEEP);
+	bcopy((void *)sdev_vnodeops_tbl, new, sdev_vnodeops_tbl_size);
+
+	/* replace the overrides from tab */
+	for (tab_entry = tab; tab_entry->name != NULL; tab_entry++) {
+		fs_operation_def_t *std_entry = new;
+		while (std_entry->name) {
+			if (strcmp(tab_entry->name, std_entry->name) == 0) {
+				std_entry->func = tab_entry->func;
+				break;
+			}
+			std_entry++;
+		}
+	}
+
+	return (new);
+}
+
+/* free memory allocated by sdev_merge_vtab */
+static void
+sdev_free_vtab(fs_operation_def_t *new)
+{
+	kmem_free(new, sdev_vnodeops_tbl_size);
+}
+
+/*
+ * Register a new plugin.
+ */
+sdev_plugin_hdl_t
+sdev_plugin_register(const char *name, sdev_plugin_ops_t *ops, int *errp)
+{
+	int ret, err;
+	sdev_plugin_t *spp, *iter;
+	vnode_t *vp, *nvp;
+	sdev_node_t *sdp, *slp;
+	timestruc_t now;
+	struct vattr vap;
+
+	/*
+	 * Some consumers don't care about why they failed. To keep the code
+	 * simple, we'll just pretend they gave us something.
+	 */
+	if (errp == NULL)
+		errp = &err;
+
+	if (sdev_plugin_name_isvalid(name, SDEV_PLUGIN_NAMELEN) == 0) {
+		*errp = EINVAL;
+		return (NULL);
+	}
+
+	if (ops->spo_version != 1) {
+		*errp = EINVAL;
+		return (NULL);
+	}
+
+	if (ops->spo_validate == NULL || ops->spo_filldir == NULL ||
+	    ops->spo_inactive == NULL) {
+		*errp = EINVAL;
+		return (NULL);
+	}
+
+	if ((ops->spo_flags & ~SDEV_PLUGIN_FLAGS_MASK) != 0) {
+		*errp = EINVAL;
+		return (NULL);
+	}
+
+	spp = kmem_cache_alloc(sdev_plugin_cache, KM_SLEEP);
+	(void) strlcpy(spp->sp_name, name, SDEV_PLUGIN_NAMELEN);
+
+	spp->sp_pops = ops;
+	spp->sp_nflags = SDEV_DYNAMIC | SDEV_VTOR;
+	if (ops->spo_flags & SDEV_PLUGIN_NO_NCACHE)
+		spp->sp_nflags |= SDEV_NO_NCACHE;
+	if (ops->spo_flags & SDEV_PLUGIN_SUBDIR)
+		spp->sp_nflags |= SDEV_SUBDIR;
+	spp->sp_vnops = sdev_plugin_vnops;
+	spp->sp_islegacy = B_FALSE;
+	spp->sp_lvtor = NULL;
+	spp->sp_nnodes = 0;
+
+	/*
+	 * Make sure it's unique, nothing exists with this name already, and add
+	 * it to the list. We also need to go through and grab the sdev
+	 * root node as we cannot grab any sdev node locks once we've grabbed
+	 * the sdev_plugin_lock. We effectively assert that if a directory is
+	 * not present in the GZ's /dev, then it doesn't exist in any of the
+	 * local zones.
+	 */
+	ret = vn_openat("/dev", UIO_SYSSPACE, FREAD, 0, &vp, 0, 0, rootdir, -1);
+	if (ret != 0) {
+		*errp = ret;
+		kmem_cache_free(sdev_plugin_cache, spp);
+		return (NULL);
+	}
+	/* Make sure we have the real vnode */
+	if (VOP_REALVP(vp, &nvp, NULL) == 0) {
+		VN_HOLD(nvp);
+		VN_RELE(vp);
+		vp = nvp;
+		nvp = NULL;
+	}
+	VERIFY(vp->v_op == sdev_vnodeops);
+	sdp = VTOSDEV(vp);
+	rw_enter(&sdp->sdev_contents, RW_WRITER);
+	slp = sdev_cache_lookup(sdp, spp->sp_name);
+	if (slp != NULL) {
+		SDEV_RELE(slp);
+		rw_exit(&sdp->sdev_contents);
+		VN_RELE(vp);
+		*errp = EEXIST;
+		kmem_cache_free(sdev_plugin_cache, spp);
+		return (NULL);
+	}
+
+	mutex_enter(&sdev_plugin_lock);
+	for (iter = list_head(&sdev_plugin_list); iter != NULL;
+	    iter = list_next(&sdev_plugin_list, iter)) {
+		if (strcmp(spp->sp_name, iter->sp_name) == 0) {
+			mutex_exit(&sdev_plugin_lock);
+			rw_exit(&sdp->sdev_contents);
+			VN_RELE(vp);
+			*errp = EEXIST;
+			kmem_cache_free(sdev_plugin_cache, spp);
+			return (NULL);
+		}
+	}
+
+	list_insert_tail(&sdev_plugin_list, spp);
+	mutex_exit(&sdev_plugin_lock);
+
+	/*
+	 * Now go ahead and create the top level directory for the global zone.
+	 */
+	vap = *sdev_getdefault_attr(VDIR);
+	gethrestime(&now);
+	vap.va_atime = now;
+	vap.va_mtime = now;
+	vap.va_ctime = now;
+
+	(void) sdev_plugin_mknode(spp, sdp, spp->sp_name, &vap);
+
+	rw_exit(&sdp->sdev_contents);
+	VN_RELE(vp);
+
+	return ((sdev_plugin_hdl_t)spp);
+}
+
+static void
+sdev_plugin_unregister_cb(sdev_node_t *rdp, void *arg)
+{
+	sdev_plugin_t *spp = arg;
+	sdev_node_t *sdp;
+
+	rw_enter(&rdp->sdev_contents, RW_WRITER);
+	sdp = sdev_cache_lookup(rdp, spp->sp_name);
+	/* If it doesn't exist, we're done here */
+	if (sdp == NULL) {
+		rw_exit(&rdp->sdev_contents);
+		return;
+	}
+
+	/*
+	 * We first delete the directory before recursively marking everything
+	 * else stale. This ordering should ensure that we don't accidentally
+	 * miss anything.
+	 */
+	sdev_cache_update(rdp, &sdp, spp->sp_name, SDEV_CACHE_DELETE);
+	sdev_stale(sdp);
+	SDEV_RELE(sdp);
+	rw_exit(&rdp->sdev_contents);
+}
+
+/*
+ * Remove a plugin. This will block until everything has become a zombie, thus
+ * guaranteeing the caller that nothing will call into them again once this call
+ * returns. While the call is ongoing, it could be called into. Note that while
+ * this is ongoing, it will block other mounts.
+ */
+int
+sdev_plugin_unregister(sdev_plugin_hdl_t hdl)
+{
+	sdev_plugin_t *spp = (sdev_plugin_t *)hdl;
+	if (spp->sp_islegacy)
+		return (EINVAL);
+
+	mutex_enter(&sdev_plugin_lock);
+	list_remove(&sdev_plugin_list, spp);
+	mutex_exit(&sdev_plugin_lock);
+
+	sdev_mnt_walk(sdev_plugin_unregister_cb, spp);
+	mutex_enter(&spp->sp_lock);
+	while (spp->sp_nnodes > 0)
+		cv_wait(&spp->sp_nodecv, &spp->sp_lock);
+	mutex_exit(&spp->sp_lock);
+	kmem_cache_free(sdev_plugin_cache, spp);
+	return (0);
+}
+
+/*
+ * Register an old sdev style plugin to deal with what used to be in the vtab.
+ */
+static int
+sdev_plugin_register_legacy(struct sdev_vop_table *vtp)
+{
+	sdev_plugin_t *spp;
+
+	spp = kmem_cache_alloc(sdev_plugin_cache, KM_SLEEP);
+	(void) strlcpy(spp->sp_name, vtp->vt_name, SDEV_PLUGIN_NAMELEN);
+	spp->sp_islegacy = B_TRUE;
+	spp->sp_pops = NULL;
+	spp->sp_nflags = vtp->vt_flags;
+	spp->sp_lvtor = vtp->vt_vtor;
+	spp->sp_nnodes = 0;
+
+	if (vtp->vt_service != NULL) {
+		fs_operation_def_t *templ;
+		templ = sdev_merge_vtab(vtp->vt_service);
+		if (vn_make_ops(vtp->vt_name,
+		    (const fs_operation_def_t *)templ,
+		    &spp->sp_vnops) != 0) {
+			cmn_err(CE_WARN, "%s: malformed vnode ops\n",
+			    vtp->vt_name);
+			sdev_free_vtab(templ);
+			kmem_cache_free(sdev_plugin_cache, spp);
+			return (1);
+		}
+
+		if (vtp->vt_global_vops) {
+			*(vtp->vt_global_vops) = spp->sp_vnops;
+		}
+
+		sdev_free_vtab(templ);
+	} else {
+		spp->sp_vnops = sdev_vnodeops;
+	}
+
+	/*
+	 * No need to check for EEXIST here. These are loaded as a part of the
+	 * sdev's initialization function. Further, we don't have to create them
+	 * as that's taken care of in sdev's mount for the GZ.
+	 */
+	mutex_enter(&sdev_plugin_lock);
+	list_insert_tail(&sdev_plugin_list, spp);
+	mutex_exit(&sdev_plugin_lock);
+
+	return (0);
+}
+
+/*
+ * We need to match off of the sdev_path, not the sdev_name. We are only allowed
+ * to exist directly under /dev.
+ */
+static sdev_plugin_t *
+sdev_match(sdev_node_t *dv)
+{
+	int vlen;
+	const char *path;
+	sdev_plugin_t *spp;
+
+	if (strlen(dv->sdev_path) <= 5)
+		return (NULL);
+
+	if (strncmp(dv->sdev_path, "/dev/", 5) != 0)
+		return (NULL);
+	path = dv->sdev_path + 5;
+
+	mutex_enter(&sdev_plugin_lock);
+
+	for (spp = list_head(&sdev_plugin_list); spp != NULL;
+	    spp = list_next(&sdev_plugin_list, spp)) {
+		if (strcmp(spp->sp_name, path) == 0) {
+			mutex_exit(&sdev_plugin_lock);
+			return (spp);
+		}
+
+		if (spp->sp_nflags & SDEV_SUBDIR) {
+			vlen = strlen(spp->sp_name);
+			if ((strncmp(spp->sp_name, path,
+			    vlen - 1) == 0) && path[vlen] == '/') {
+				mutex_exit(&sdev_plugin_lock);
+				return (spp);
+			}
+
+		}
+	}
+
+	mutex_exit(&sdev_plugin_lock);
+	return (NULL);
+}
+
+void
+sdev_set_no_negcache(sdev_node_t *dv)
+{
+	char *path;
+	sdev_plugin_t *spp;
+
+	ASSERT(dv->sdev_path);
+	path = dv->sdev_path + strlen("/dev/");
+
+	mutex_enter(&sdev_plugin_lock);
+	for (spp = list_head(&sdev_plugin_list); spp != NULL;
+	    spp = list_next(&sdev_plugin_list, spp)) {
+		if (strcmp(spp->sp_name, path) == 0) {
+			if (spp->sp_nflags & SDEV_NO_NCACHE)
+				dv->sdev_flags |= SDEV_NO_NCACHE;
+			break;
+		}
+	}
+	mutex_exit(&sdev_plugin_lock);
+}
+
+struct vnodeops *
+sdev_get_vop(sdev_node_t *dv)
+{
+	char *path;
+	sdev_plugin_t *spp;
+
+	path = dv->sdev_path;
+	ASSERT(path);
+
+	/* gets the relative path to /dev/ */
+	path += 5;
+
+	if ((spp = sdev_match(dv)) != NULL) {
+		dv->sdev_flags |= spp->sp_nflags;
+		if (SDEV_IS_PERSIST(dv->sdev_dotdot) &&
+		    (SDEV_IS_PERSIST(dv) || !SDEV_IS_DYNAMIC(dv)))
+			dv->sdev_flags |= SDEV_PERSIST;
+		return (spp->sp_vnops);
+	}
+
+	/* child inherits the persistence of the parent */
+	if (SDEV_IS_PERSIST(dv->sdev_dotdot))
+		dv->sdev_flags |= SDEV_PERSIST;
+	return (sdev_vnodeops);
+}
+
+void *
+sdev_get_vtor(sdev_node_t *dv)
+{
+	sdev_plugin_t *spp;
+
+	if (dv->sdev_private == NULL) {
+		spp = sdev_match(dv);
+		if (spp == NULL)
+			return (NULL);
+	} else {
+		spp = dv->sdev_private;
+	}
+
+	if (spp->sp_islegacy)
+		return ((void *)spp->sp_lvtor);
+	else
+		return ((void *)sdev_plugin_validate);
+}
+
+void
+sdev_plugin_nodeready(sdev_node_t *sdp)
+{
+	sdev_plugin_t *spp;
+
+	ASSERT(RW_WRITE_HELD(&sdp->sdev_contents));
+	ASSERT(sdp->sdev_private == NULL);
+
+	spp = sdev_match(sdp);
+	if (spp == NULL)
+		return;
+	if (spp->sp_islegacy)
+		return;
+	sdp->sdev_private = spp;
+	mutex_enter(&spp->sp_lock);
+	spp->sp_nnodes++;
+	mutex_exit(&spp->sp_lock);
+}
+
+int
+sdev_plugin_init(void)
+{
+	sdev_vop_table_t *vtp;
+	fs_operation_def_t *templ;
+
+	sdev_plugin_cache = kmem_cache_create("sdev_plugin",
+	    sizeof (sdev_plugin_t), 0, sdev_plugin_cache_constructor,
+	    sdev_plugin_cache_destructor, NULL, NULL, NULL, 0);
+	if (sdev_plugin_cache == NULL)
+		return (1);
+	mutex_init(&sdev_plugin_lock, NULL, MUTEX_DRIVER, NULL);
+	list_create(&sdev_plugin_list, sizeof (sdev_plugin_t),
+	    offsetof(sdev_plugin_t, sp_link));
+
+	/*
+	 * Register all of the legacy vnops
+	 */
+	for (vtp = &vtab[0]; vtp->vt_name != NULL; vtp++)
+		if (sdev_plugin_register_legacy(vtp) != 0)
+			return (1);
+
+	templ = sdev_merge_vtab(sdev_plugin_vnodeops_tbl);
+	if (vn_make_ops("sdev_plugin",
+	    (const fs_operation_def_t *)templ,
+	    &sdev_plugin_vnops) != 0) {
+		sdev_free_vtab(templ);
+		return (1);
+	}
+
+	sdev_free_vtab(templ);
+	return (0);
+}
diff --git a/usr/src/uts/common/fs/dev/sdev_subr.c b/usr/src/uts/common/fs/dev/sdev_subr.c
index b4b27e6285..16439a66a6 100644
--- a/usr/src/uts/common/fs/dev/sdev_subr.c
+++ b/usr/src/uts/common/fs/dev/sdev_subr.c
@@ -20,7 +20,7 @@
  */
 /*
  * Copyright (c) 2006, 2010, Oracle and/or its affiliates. All rights reserved.
- * Copyright (c) 2013, Joyent, Inc. All rights reserved.
+ * Copyright (c) 2014, Joyent, Inc. All rights reserved.
  */
 
 /*
@@ -150,12 +150,6 @@ vattr_t sdev_vattr_chr = {
 kmem_cache_t	*sdev_node_cache;	/* sdev_node cache */
 int		devtype;		/* fstype */
 
-/* static */
-static struct vnodeops *sdev_get_vop(struct sdev_node *);
-static void sdev_set_no_negcache(struct sdev_node *);
-static fs_operation_def_t *sdev_merge_vtab(const fs_operation_def_t []);
-static void sdev_free_vtab(fs_operation_def_t *);
-
 static void
 sdev_prof_free(struct sdev_node *dv)
 {
@@ -318,6 +312,7 @@ sdev_nodeinit(struct sdev_node *ddv, char *nm, struct sdev_node **newdv,
 	(void) snprintf(dv->sdev_path, len, "%s/%s", ddv->sdev_path, nm);
 	/* overwritten for VLNK nodes */
 	dv->sdev_symlink = NULL;
+	list_link_init(&dv->sdev_plist);
 
 	vp = SDEVTOV(dv);
 	vn_reinit(vp);
@@ -406,6 +401,7 @@ sdev_nodeready(struct sdev_node *dv, struct vattr *vap, struct vnode *avp,
 	} else {
 		dv->sdev_nlink = 1;
 	}
+	sdev_plugin_nodeready(dv);
 
 	if (!(SDEV_IS_GLOBAL(dv))) {
 		dv->sdev_origin = (struct sdev_node *)args;
@@ -502,37 +498,22 @@ sdev_mkroot(struct vfs *vfsp, dev_t devdev, struct vnode *mvp,
 	return (dv);
 }
 
-/* directory dependent vop table */
-struct sdev_vop_table {
-	char *vt_name;				/* subdirectory name */
-	const fs_operation_def_t *vt_service;	/* vnodeops table */
-	struct vnodeops *vt_vops;		/* constructed vop */
-	struct vnodeops **vt_global_vops;	/* global container for vop */
-	int (*vt_vtor)(struct sdev_node *);	/* validate sdev_node */
-	int vt_flags;
-};
-
-/*
- * A nice improvement would be to provide a plug-in mechanism
- * for this table instead of a const table.
- */
-static struct sdev_vop_table vtab[] =
-{
-	{ "pts", devpts_vnodeops_tbl, NULL, &devpts_vnodeops, devpts_validate,
+struct sdev_vop_table vtab[] = {
+	{ "pts", devpts_vnodeops_tbl, &devpts_vnodeops, devpts_validate,
 	SDEV_DYNAMIC | SDEV_VTOR },
 
-	{ "vt", devvt_vnodeops_tbl, NULL, &devvt_vnodeops, devvt_validate,
+	{ "vt", devvt_vnodeops_tbl, &devvt_vnodeops, devvt_validate,
 	SDEV_DYNAMIC | SDEV_VTOR },
 
-	{ "zvol", devzvol_vnodeops_tbl, NULL, &devzvol_vnodeops,
+	{ "zvol", devzvol_vnodeops_tbl, &devzvol_vnodeops,
 	devzvol_validate, SDEV_ZONED | SDEV_DYNAMIC | SDEV_VTOR | SDEV_SUBDIR },
 
-	{ "zcons", NULL, NULL, NULL, NULL, SDEV_NO_NCACHE },
+	{ "zcons", NULL, NULL, NULL, SDEV_NO_NCACHE },
 
-	{ "net", devnet_vnodeops_tbl, NULL, &devnet_vnodeops, devnet_validate,
-	SDEV_DYNAMIC | SDEV_VTOR },
+	{ "net", devnet_vnodeops_tbl, &devnet_vnodeops, devnet_validate,
+	SDEV_DYNAMIC | SDEV_VTOR | SDEV_SUBDIR },
 
-	{ "ipnet", devipnet_vnodeops_tbl, NULL, &devipnet_vnodeops,
+	{ "ipnet", devipnet_vnodeops_tbl, &devipnet_vnodeops,
 	devipnet_validate, SDEV_DYNAMIC | SDEV_VTOR | SDEV_NO_NCACHE },
 
 	/*
@@ -547,132 +528,14 @@ static struct sdev_vop_table vtab[] =
 	 * preventing a mkdir.
 	 */
 
-	{ "lofi", NULL, NULL, NULL, NULL,
+	{ "lofi", NULL, NULL, NULL,
 	    SDEV_ZONED | SDEV_DYNAMIC | SDEV_PERSIST },
-	{ "rlofi", NULL, NULL, NULL, NULL,
+	{ "rlofi", NULL, NULL, NULL,
 	    SDEV_ZONED | SDEV_DYNAMIC | SDEV_PERSIST },
 
-	{ NULL, NULL, NULL, NULL, NULL, 0}
+	{ NULL, NULL, NULL, NULL, 0}
 };
 
-/*
- * We need to match off of the sdev_path, not the sdev_name. We are only allowed
- * to exist directly under /dev.
- */
-struct sdev_vop_table *
-sdev_match(struct sdev_node *dv)
-{
-	int vlen;
-	int i;
-	const char *path;
-
-	if (strlen(dv->sdev_path) <= 5)
-		return (NULL);
-
-	if (strncmp(dv->sdev_path, "/dev/", 5) != 0)
-		return (NULL);
-	path = dv->sdev_path + 5;
-
-	for (i = 0; vtab[i].vt_name; i++) {
-		if (strcmp(vtab[i].vt_name, path) == 0)
-			return (&vtab[i]);
-		if (vtab[i].vt_flags & SDEV_SUBDIR) {
-			vlen = strlen(vtab[i].vt_name);
-			if ((strncmp(vtab[i].vt_name, path,
-			    vlen - 1) == 0) && path[vlen] == '/')
-				return (&vtab[i]);
-		}
-
-	}
-	return (NULL);
-}
-
-/*
- *  sets a directory's vnodeops if the directory is in the vtab;
- */
-static struct vnodeops *
-sdev_get_vop(struct sdev_node *dv)
-{
-	struct sdev_vop_table *vtp;
-	char *path;
-
-	path = dv->sdev_path;
-	ASSERT(path);
-
-	/* gets the relative path to /dev/ */
-	path += 5;
-
-	/* gets the vtab entry it matches */
-	if ((vtp = sdev_match(dv)) != NULL) {
-		dv->sdev_flags |= vtp->vt_flags;
-		if (SDEV_IS_PERSIST(dv->sdev_dotdot) &&
-		    (SDEV_IS_PERSIST(dv) || !SDEV_IS_DYNAMIC(dv)))
-			dv->sdev_flags |= SDEV_PERSIST;
-
-		if (vtp->vt_vops) {
-			if (vtp->vt_global_vops)
-				*(vtp->vt_global_vops) = vtp->vt_vops;
-
-			return (vtp->vt_vops);
-		}
-
-		if (vtp->vt_service) {
-			fs_operation_def_t *templ;
-			templ = sdev_merge_vtab(vtp->vt_service);
-			if (vn_make_ops(vtp->vt_name,
-			    (const fs_operation_def_t *)templ,
-			    &vtp->vt_vops) != 0) {
-				cmn_err(CE_PANIC, "%s: malformed vnode ops\n",
-				    vtp->vt_name);
-				/*NOTREACHED*/
-			}
-			if (vtp->vt_global_vops) {
-				*(vtp->vt_global_vops) = vtp->vt_vops;
-			}
-			sdev_free_vtab(templ);
-
-			return (vtp->vt_vops);
-		}
-
-		return (sdev_vnodeops);
-	}
-
-	/* child inherits the persistence of the parent */
-	if (SDEV_IS_PERSIST(dv->sdev_dotdot))
-		dv->sdev_flags |= SDEV_PERSIST;
-
-	return (sdev_vnodeops);
-}
-
-static void
-sdev_set_no_negcache(struct sdev_node *dv)
-{
-	int i;
-	char *path;
-
-	ASSERT(dv->sdev_path);
-	path = dv->sdev_path + strlen("/dev/");
-
-	for (i = 0; vtab[i].vt_name; i++) {
-		if (strcmp(vtab[i].vt_name, path) == 0) {
-			if (vtab[i].vt_flags & SDEV_NO_NCACHE)
-				dv->sdev_flags |= SDEV_NO_NCACHE;
-			break;
-		}
-	}
-}
-
-void *
-sdev_get_vtor(struct sdev_node *dv)
-{
-	struct sdev_vop_table *vtp;
-
-	vtp = sdev_match(dv);
-	if (vtp)
-		return ((void *)vtp->vt_vtor);
-	else
-		return (NULL);
-}
 
 /*
  * Build the base root inode
@@ -952,8 +815,11 @@ sdev_nodedestroy(struct sdev_node *dv, uint_t flags)
 		dv->sdev_path = NULL;
 	}
 
-	if (!SDEV_IS_GLOBAL(dv))
+	if (!SDEV_IS_GLOBAL(dv)) {
 		sdev_prof_free(dv);
+		if (dv->sdev_vnode->v_type != VLNK && dv->sdev_origin != NULL)
+			SDEV_RELE(dv->sdev_origin);
+	}
 
 	if (SDEVTOV(dv)->v_type == VDIR) {
 		ASSERT(SDEV_FIRST_ENTRY(dv) == NULL);
@@ -967,6 +833,7 @@ sdev_nodedestroy(struct sdev_node *dv, uint_t flags)
 	(void) memset((void *)&dv->sdev_instance_data, 0,
 	    sizeof (dv->sdev_instance_data));
 	vn_invalid(SDEVTOV(dv));
+	dv->sdev_private = NULL;
 	kmem_cache_free(sdev_node_cache, dv);
 }
 
@@ -2948,46 +2815,6 @@ sdev_modctl_devexists(const char *path)
 	return (error);
 }
 
-extern int sdev_vnodeops_tbl_size;
-
-/*
- * construct a new template with overrides from vtab
- */
-static fs_operation_def_t *
-sdev_merge_vtab(const fs_operation_def_t tab[])
-{
-	fs_operation_def_t *new;
-	const fs_operation_def_t *tab_entry;
-
-	/* make a copy of standard vnode ops table */
-	new = kmem_alloc(sdev_vnodeops_tbl_size, KM_SLEEP);
-	bcopy((void *)sdev_vnodeops_tbl, new, sdev_vnodeops_tbl_size);
-
-	/* replace the overrides from tab */
-	for (tab_entry = tab; tab_entry->name != NULL; tab_entry++) {
-		fs_operation_def_t *std_entry = new;
-		while (std_entry->name) {
-			if (strcmp(tab_entry->name, std_entry->name) == 0) {
-				std_entry->func = tab_entry->func;
-				break;
-			}
-			std_entry++;
-		}
-		if (std_entry->name == NULL)
-			cmn_err(CE_NOTE, "sdev_merge_vtab: entry %s unused.",
-			    tab_entry->name);
-	}
-
-	return (new);
-}
-
-/* free memory allocated by sdev_merge_vtab */
-static void
-sdev_free_vtab(fs_operation_def_t *new)
-{
-	kmem_free(new, sdev_vnodeops_tbl_size);
-}
-
 /*
  * a generic setattr() function
  *
diff --git a/usr/src/uts/common/fs/dev/sdev_vfsops.c b/usr/src/uts/common/fs/dev/sdev_vfsops.c
index ea9cb6374a..6f32f47635 100644
--- a/usr/src/uts/common/fs/dev/sdev_vfsops.c
+++ b/usr/src/uts/common/fs/dev/sdev_vfsops.c
@@ -169,7 +169,13 @@ devinit(int fstype, char *name)
 
 	if ((devmajor = getudev()) == (major_t)-1) {
 		cmn_err(CE_WARN, "%s: can't get unique dev", sdev_vfssw.name);
-		return (1);
+		return (ENXIO);
+	}
+
+	if (sdev_plugin_init() != 0) {
+		cmn_err(CE_WARN, "%s: failed to set init plugin subsystem",
+		    sdev_vfssw.name);
+		return (EIO);
 	}
 
 	/* initialize negative cache */
@@ -332,6 +338,7 @@ sdev_mount(struct vfs *vfsp, struct vnode *mvp, struct mounta *uap,
 		ASSERT(sdev_origins);
 		dv->sdev_flags &= ~SDEV_GLOBAL;
 		dv->sdev_origin = sdev_origins->sdev_root;
+		SDEV_HOLD(dv->sdev_origin);
 	} else {
 		sdev_ncache_setup();
 		rw_enter(&dv->sdev_contents, RW_WRITER);
@@ -504,3 +511,17 @@ sdev_mntinfo_rele(struct sdev_data *mntinfo)
 	SDEVTOV(mntinfo->sdev_root)->v_count--;
 	mutex_exit(&sdev_lock);
 }
+
+void
+sdev_mnt_walk(void (*func)(struct sdev_node *, void *), void *arg)
+{
+	struct sdev_data *mntinfo;
+
+	mutex_enter(&sdev_lock);
+	mntinfo = sdev_mntinfo;
+	while (mntinfo != NULL) {
+		func(mntinfo->sdev_root, arg);
+		mntinfo = mntinfo->sdev_next;
+	}
+	mutex_exit(&sdev_lock);
+}
diff --git a/usr/src/uts/common/fs/dev/sdev_zvolops.c b/usr/src/uts/common/fs/dev/sdev_zvolops.c
index e4c3acf787..495cf450de 100644
--- a/usr/src/uts/common/fs/dev/sdev_zvolops.c
+++ b/usr/src/uts/common/fs/dev/sdev_zvolops.c
@@ -21,7 +21,7 @@
 /*
  * Copyright 2009 Sun Microsystems, Inc.  All rights reserved.
  * Use is subject to license terms.
- * Copyright 2013 Joyent, Inc.  All rights reserved.
+ * Copyright 2015 Joyent, Inc.  All rights reserved.
  */
 
 /* vnode ops for the /dev/zvol directory */
@@ -370,8 +370,10 @@ devzvol_create_pool_dirs(struct vnode *dvp)
 		ASSERT(dvp->v_count > 0);
 		rc = VOP_LOOKUP(dvp, nvpair_name(elem), &vp, NULL, 0,
 		    NULL, kcred, NULL, 0, NULL);
-		/* should either work, or not be visible from a zone */
-		ASSERT(rc == 0 || rc == ENOENT);
+		/*
+		 * should either work or we should get an error if this should
+		 * not be visible from the zone, or disallowed in the zone
+		 */
 		if (rc == 0)
 			VN_RELE(vp);
 		pools++;
diff --git a/usr/src/uts/common/fs/dnlc.c b/usr/src/uts/common/fs/dnlc.c
index 25327d2852..c949117da6 100644
--- a/usr/src/uts/common/fs/dnlc.c
+++ b/usr/src/uts/common/fs/dnlc.c
@@ -921,50 +921,6 @@ dnlc_fs_purge1(vnodeops_t *vop)
 }
 
 /*
- * Perform a reverse lookup in the DNLC.  This will find the first occurrence of
- * the vnode.  If successful, it will return the vnode of the parent, and the
- * name of the entry in the given buffer.  If it cannot be found, or the buffer
- * is too small, then it will return NULL.  Note that this is a highly
- * inefficient function, since the DNLC is constructed solely for forward
- * lookups.
- */
-vnode_t *
-dnlc_reverse_lookup(vnode_t *vp, char *buf, size_t buflen)
-{
-	nc_hash_t *nch;
-	ncache_t *ncp;
-	vnode_t *pvp;
-
-	if (!doingcache)
-		return (NULL);
-
-	for (nch = nc_hash; nch < &nc_hash[nc_hashsz]; nch++) {
-		mutex_enter(&nch->hash_lock);
-		ncp = nch->hash_next;
-		while (ncp != (ncache_t *)nch) {
-			/*
-			 * We ignore '..' entries since it can create
-			 * confusion and infinite loops.
-			 */
-			if (ncp->vp == vp && !(ncp->namlen == 2 &&
-			    0 == bcmp(ncp->name, "..", 2)) &&
-			    ncp->namlen < buflen) {
-				bcopy(ncp->name, buf, ncp->namlen);
-				buf[ncp->namlen] = '\0';
-				pvp = ncp->dp;
-				/* VN_HOLD 2 of 2 in this file */
-				VN_HOLD_CALLER(pvp);
-				mutex_exit(&nch->hash_lock);
-				return (pvp);
-			}
-			ncp = ncp->hash_next;
-		}
-		mutex_exit(&nch->hash_lock);
-	}
-
-	return (NULL);
-}
-/*
  * Utility routine to search for a cache entry. Return the
  * ncache entry if found, NULL otherwise.
  */
diff --git a/usr/src/uts/common/fs/fem.c b/usr/src/uts/common/fs/fem.c
index b4e28cc860..5f524def30 100644
--- a/usr/src/uts/common/fs/fem.c
+++ b/usr/src/uts/common/fs/fem.c
@@ -23,6 +23,10 @@
  * Use is subject to license terms.
  */
 
+/*
+ * Copyright (c) 2015, Joyent, Inc.  All rights reserved.
+ */
+
 #include <sys/types.h>
 #include <sys/atomic.h>
 #include <sys/kmem.h>
@@ -33,11 +37,12 @@
 #include <sys/systm.h>
 #include <sys/cmn_err.h>
 #include <sys/debug.h>
-
 #include <sys/fem.h>
 #include <sys/vfs.h>
 #include <sys/vnode.h>
 #include <sys/vfs_opreg.h>
+#include <sys/stack.h>
+#include <sys/archsystm.h>
 
 #define	NNODES_DEFAULT	8	/* Default number of nodes in a fem_list */
 /*
@@ -291,6 +296,536 @@ _op_find(femarg_t *ap, void **fp, int offs0, int offs1)
 }
 #endif
 
+/*
+ * File event monitoring handoffs
+ *
+ * File event monitoring relies on being able to inject stack frames between
+ * vnode consumers and the underlying file systems.  This becomes problematic
+ * when there exist many monitors, as kernel stack depth is finite.  The model
+ * very much encodes this injected frame:  the flow of control deliberately
+ * lies with the monitor, not with the monitoring system.  While we could
+ * conceivably address this by allowing each subsystem to install at most
+ * one monitor per vnode (and impose on subsystems that they handle any
+ * of their own consumer multiplexing internally), this in fact exports a
+ * substantial amount of run-time complexity to deal with an uncommon case
+ * (and, it must be said, assumes a small number of consuming subsystems).
+ * To allow our abstraction to remain clean, we instead check our remaining
+ * stack in every vnext_*() call; if the amount of stack remaining is lower
+ * than a threshold (fem_stack_needed), we call thread_splitstack() to carry
+ * on the execution of the monitors and the underlying vnode operation on a
+ * split stack.  Because we can only pass a single argument to our split stack
+ * function, we must marshal our arguments, the mechanics of which are somewhat
+ * ornate in terms of the code: to marshal in a type-safe manner, we define a
+ * baton that is a union of payload structures for each kind of operation,
+ * loading the per-operation payload explicitly and calling into common handoff
+ * code that itself calls thread_splitstack().  The function passed to
+ * thread_splitstack() is a per-entry point function that continues monitor
+ * processing given the specified (marshalled) arguments.  While this method
+ * is a little verbose to implement, it has the advantage of being relatively
+ * robust (that is, broadly type-safe) while imposing minimal burden on each
+ * vnext_*() entry point.
+ *
+ * In terms of the implementation:
+ *
+ * - The FEM_BATON_n macros define the per-entry point baton structures
+ * - The fem_baton_payload_t contains the union of these structures
+ * - The FEM_VNEXTn_DECL macros declare the post-handoff entry point
+ * - The FEM_VNEXTn macros constitute the per-handoff entry point
+ *
+ * Note that we don't use variadic macros -- we define a variant of these
+ * macros for each of our relevant argument counts.  This may seem overly
+ * explicit, but it is deliberate:  the object here is to minimize the
+ * future maintenance burden by minimizing the likelihood of introduced
+ * error --  not to minimize the number of characters in this source file.
+ */
+
+#ifndef STACK_GROWTH_DOWN
+#error Downward stack growth assumed.
+#endif
+
+int fem_stack_toodeep;
+uintptr_t fem_stack_needed = 8 * 1024;
+size_t fem_handoff_stacksize = 128 * 1024;
+
+#define	FEM_TOODEEP() (STACK_BIAS + (uintptr_t)getfp() - \
+	(uintptr_t)curthread->t_stkbase < fem_stack_needed)
+
+#define	FEM_BATON_1(what, t0, l0)					\
+	struct {							\
+		void *fb_##what##_arg0;					\
+		caller_context_t *fb_##what##_ct;			\
+		t0 fb_##what##_##l0;					\
+	} fb_##what
+
+#define	FEM_BATON_2(what, t0, l0, t1, l1)				\
+	struct {							\
+		void *fb_##what##_arg0;					\
+		caller_context_t *fb_##what##_ct;			\
+		t0 fb_##what##_##l0;					\
+		t1 fb_##what##_##l1;					\
+	} fb_##what
+
+#define	FEM_BATON_3(what, t0, l0, t1, l1, t2, l2)			\
+	struct {							\
+		void *fb_##what##_arg0;					\
+		caller_context_t *fb_##what##_ct;			\
+		t0 fb_##what##_##l0;					\
+		t1 fb_##what##_##l1;					\
+		t2 fb_##what##_##l2;					\
+	} fb_##what
+
+#define	FEM_BATON_4(what, t0, l0, t1, l1, t2, l2, t3, l3)		\
+	struct {							\
+		void *fb_##what##_arg0;					\
+		caller_context_t *fb_##what##_ct;			\
+		t0 fb_##what##_##l0;					\
+		t1 fb_##what##_##l1;					\
+		t2 fb_##what##_##l2;					\
+		t3 fb_##what##_##l3;					\
+	} fb_##what
+
+#define	FEM_BATON_5(what, t0, l0, t1, l1, t2, l2, t3, l3, t4, l4)	\
+	struct {							\
+		void *fb_##what##_arg0;					\
+		caller_context_t *fb_##what##_ct;			\
+		t0 fb_##what##_##l0;					\
+		t1 fb_##what##_##l1;					\
+		t2 fb_##what##_##l2;					\
+		t3 fb_##what##_##l3;					\
+		t4 fb_##what##_##l4;					\
+	} fb_##what
+
+#define	FEM_BATON_6(what, t0, l0, t1, l1, t2, l2, t3, l3, t4, l4, t5, l5) \
+	struct {							\
+		void *fb_##what##_arg0;					\
+		caller_context_t *fb_##what##_ct;			\
+		t0 fb_##what##_##l0;					\
+		t1 fb_##what##_##l1;					\
+		t2 fb_##what##_##l2;					\
+		t3 fb_##what##_##l3;					\
+		t4 fb_##what##_##l4;					\
+		t5 fb_##what##_##l5;					\
+	} fb_##what
+
+#define	FEM_BATON_8(what, t0, l0, t1, l1, t2, l2, t3, l3, t4, l4, t5, l5, \
+    t6, l6, t7, l7) \
+	struct {							\
+		void *fb_##what##_arg0;					\
+		caller_context_t *fb_##what##_ct;			\
+		t0 fb_##what##_##l0;					\
+		t1 fb_##what##_##l1;					\
+		t2 fb_##what##_##l2;					\
+		t3 fb_##what##_##l3;					\
+		t4 fb_##what##_##l4;					\
+		t5 fb_##what##_##l5;					\
+		t6 fb_##what##_##l6;					\
+		t7 fb_##what##_##l7;					\
+	} fb_##what
+
+#define	FEM_BATON_9(what, t0, l0, t1, l1, t2, l2, t3, l3, t4, l4, t5, l5, \
+    t6, l6, t7, l7, t8, l8) \
+	struct {							\
+		void *fb_##what##_arg0;					\
+		caller_context_t *fb_##what##_ct;			\
+		t0 fb_##what##_##l0;					\
+		t1 fb_##what##_##l1;					\
+		t2 fb_##what##_##l2;					\
+		t3 fb_##what##_##l3;					\
+		t4 fb_##what##_##l4;					\
+		t5 fb_##what##_##l5;					\
+		t6 fb_##what##_##l6;					\
+		t7 fb_##what##_##l7;					\
+		t8 fb_##what##_##l8;					\
+	} fb_##what
+
+typedef union {
+	FEM_BATON_2(open, int, mode, cred_t *, cr);
+	FEM_BATON_4(close, int, flag, int, count,
+	    offset_t, offset, cred_t *, cr);
+	FEM_BATON_3(read, uio_t *, uiop, int, ioflag, cred_t *, cr);
+	FEM_BATON_3(write, uio_t *, uiop, int, ioflag, cred_t *, cr);
+	FEM_BATON_5(ioctl, int, cmd, intptr_t, arg,
+	    int, flag, cred_t *, cr, int *, rvalp);
+	FEM_BATON_3(setfl, int, oflags, int, nflags, cred_t *, cr);
+	FEM_BATON_3(getattr, vattr_t *, vap, int, flags, cred_t *, cr);
+	FEM_BATON_3(setattr, vattr_t *, vap, int, flags, cred_t *, cr);
+	FEM_BATON_3(access, int, mode, int, flags, cred_t *, cr);
+	FEM_BATON_8(lookup, char *, nm, vnode_t **, vpp,
+	    pathname_t *, pnp, int, flags, vnode_t *, rdir,
+	    cred_t *, cr, int *, direntflags, pathname_t *, realpnp);
+	FEM_BATON_8(create, char *, name, vattr_t *, vap,
+	    vcexcl_t, excl, int, mode, vnode_t **, vpp,
+	    cred_t *, cr, int, flag, vsecattr_t *, vsecp);
+	FEM_BATON_3(remove, char *, nm, cred_t *, cr, int, flags);
+	FEM_BATON_4(link, vnode_t *, svp, char *, tnm,
+	    cred_t *, cr, int, flags);
+	FEM_BATON_5(rename, char *, snm, vnode_t *, tdvp,
+	    char *, tnm, cred_t *, cr, int, flags);
+	FEM_BATON_6(mkdir, char *, dirname, vattr_t *, vap,
+	    vnode_t **, vpp, cred_t *, cr, int, flags,
+	    vsecattr_t *, vsecp);
+	FEM_BATON_4(rmdir, char *, nm, vnode_t *, cdir,
+	    cred_t *, cr, int, flags);
+	FEM_BATON_4(readdir, uio_t *, uiop, cred_t *, cr,
+	    int *, eofp, int, flags);
+	FEM_BATON_5(symlink, char *, linkname, vattr_t *, vap,
+	    char *, target, cred_t *, cr, int, flags);
+	FEM_BATON_2(readlink, uio_t *, uiop, cred_t *, cr);
+	FEM_BATON_2(fsync, int, syncflag, cred_t *, cr);
+	FEM_BATON_1(inactive, cred_t *, cr);
+	FEM_BATON_1(fid, fid_t *, fidp);
+	FEM_BATON_1(rwlock, int, write_lock);
+	FEM_BATON_1(rwunlock, int, write_lock);
+	FEM_BATON_2(seek, offset_t, ooff, offset_t *, noffp);
+	FEM_BATON_1(cmp, vnode_t *, vp2);
+	FEM_BATON_6(frlock, int, cmd, struct flock64 *, bfp,
+	    int, flag, offset_t, offset, struct flk_callback *, flk_cbp,
+	    cred_t *, cr);
+	FEM_BATON_5(space, int, cmd, struct flock64 *, bfp,
+	    int, flag, offset_t, offset, cred_t *, cr);
+	FEM_BATON_1(realvp, vnode_t **, vpp);
+	FEM_BATON_9(getpage, offset_t, off, size_t, len,
+	    uint_t *, protp, struct page **, plarr, size_t, plsz,
+	    struct seg *, seg, caddr_t, addr, enum seg_rw, rw,
+	    cred_t *, cr);
+	FEM_BATON_4(putpage, offset_t, off, size_t, len,
+	    int, flags, cred_t *, cr);
+	FEM_BATON_8(map, offset_t, off, struct as *, as,
+	    caddr_t *, addrp, size_t, len, uchar_t, prot,
+	    uchar_t, maxprot, uint_t, flags, cred_t *, cr);
+	FEM_BATON_8(addmap, offset_t, off, struct as *, as,
+	    caddr_t, addr, size_t, len, uchar_t, prot,
+	    uchar_t, maxprot, uint_t, flags, cred_t *, cr);
+	FEM_BATON_8(delmap, offset_t, off, struct as *, as,
+	    caddr_t, addr, size_t, len, uint_t, prot,
+	    uint_t, maxprot, uint_t, flags, cred_t *, cr);
+	FEM_BATON_4(poll, short, events, int, anyyet,
+	    short *, reventsp, struct pollhead **, phpp);
+	FEM_BATON_3(dump, caddr_t, addr, offset_t, lbdn, offset_t, dblks);
+	FEM_BATON_3(pathconf, int, cmd, ulong_t *, valp, cred_t *, cr);
+	FEM_BATON_5(pageio, struct page *, pp, u_offset_t, io_off,
+	    size_t, io_len, int, flags, cred_t *, cr);
+	FEM_BATON_2(dumpctl, int, action, offset_t *, blkp);
+	FEM_BATON_4(dispose, struct page *, pp, int, flag,
+	    int, dn, cred_t *, cr);
+	FEM_BATON_3(setsecattr, vsecattr_t *, vsap, int, flag, cred_t *, cr);
+	FEM_BATON_3(getsecattr, vsecattr_t *, vsap, int, flag, cred_t *, cr);
+	FEM_BATON_4(shrlock, int, cmd, struct shrlock *, shr,
+	    int, flag, cred_t *, cr);
+	FEM_BATON_3(vnevent, vnevent_t, vnevent, vnode_t *, dvp, char *, cname);
+	FEM_BATON_3(reqzcbuf, enum uio_rw, ioflag,
+	    xuio_t *, xuiop, cred_t *, cr);
+	FEM_BATON_2(retzcbuf, xuio_t *, xuiop, cred_t *, cr);
+} fem_baton_payload_t;
+
+typedef struct {
+	fem_baton_payload_t fb_payload;
+	int (*fb_func)();
+	void (*fb_handoff)();
+	int fb_rval;
+} fem_baton_t;
+
+static int
+fem_handoff(fem_baton_t *bp)
+{
+	fem_stack_toodeep++;
+	thread_splitstack(bp->fb_handoff, bp, fem_handoff_stacksize);
+
+	return (bp->fb_rval);
+}
+
+#define	FEM_VNEXT3_DECL(what, a0, a1, a2)				\
+void									\
+fem_handoff_##what(fem_baton_t *bp)					\
+{									\
+	bp->fb_rval = bp->fb_func(					\
+	    bp->fb_payload.fb_##what.fb_##what##_##a0,			\
+	    bp->fb_payload.fb_##what.fb_##what##_##a1,			\
+	    bp->fb_payload.fb_##what.fb_##what##_##a2);			\
+}
+
+#define	FEM_VNEXT4_DECL(what, a0, a1, a2, a3)				\
+void									\
+fem_handoff_##what(fem_baton_t *bp)					\
+{									\
+	bp->fb_rval = bp->fb_func(					\
+	    bp->fb_payload.fb_##what.fb_##what##_##a0,			\
+	    bp->fb_payload.fb_##what.fb_##what##_##a1,			\
+	    bp->fb_payload.fb_##what.fb_##what##_##a2,			\
+	    bp->fb_payload.fb_##what.fb_##what##_##a3);			\
+}
+
+#define	FEM_VNEXT5_DECL(what, a0, a1, a2, a3, a4)			\
+void									\
+fem_handoff_##what(fem_baton_t *bp)					\
+{									\
+	bp->fb_rval = bp->fb_func(					\
+	    bp->fb_payload.fb_##what.fb_##what##_##a0,			\
+	    bp->fb_payload.fb_##what.fb_##what##_##a1,			\
+	    bp->fb_payload.fb_##what.fb_##what##_##a2,			\
+	    bp->fb_payload.fb_##what.fb_##what##_##a3,			\
+	    bp->fb_payload.fb_##what.fb_##what##_##a4);			\
+}
+
+#define	FEM_VNEXT6_DECL(what, a0, a1, a2, a3, a4, a5)			\
+void									\
+fem_handoff_##what(fem_baton_t *bp)					\
+{									\
+	bp->fb_rval = bp->fb_func(					\
+	    bp->fb_payload.fb_##what.fb_##what##_##a0,			\
+	    bp->fb_payload.fb_##what.fb_##what##_##a1,			\
+	    bp->fb_payload.fb_##what.fb_##what##_##a2,			\
+	    bp->fb_payload.fb_##what.fb_##what##_##a3,			\
+	    bp->fb_payload.fb_##what.fb_##what##_##a4,			\
+	    bp->fb_payload.fb_##what.fb_##what##_##a5);			\
+}
+
+#define	FEM_VNEXT7_DECL(what, a0, a1, a2, a3, a4, a5, a6)		\
+void									\
+fem_handoff_##what(fem_baton_t *bp)					\
+{									\
+	bp->fb_rval = bp->fb_func(					\
+	    bp->fb_payload.fb_##what.fb_##what##_##a0,			\
+	    bp->fb_payload.fb_##what.fb_##what##_##a1,			\
+	    bp->fb_payload.fb_##what.fb_##what##_##a2,			\
+	    bp->fb_payload.fb_##what.fb_##what##_##a3,			\
+	    bp->fb_payload.fb_##what.fb_##what##_##a4,			\
+	    bp->fb_payload.fb_##what.fb_##what##_##a5,			\
+	    bp->fb_payload.fb_##what.fb_##what##_##a6);			\
+}
+
+#define	FEM_VNEXT8_DECL(what, a0, a1, a2, a3, a4, a5, a6, a7)		\
+void									\
+fem_handoff_##what(fem_baton_t *bp)					\
+{									\
+	bp->fb_rval = bp->fb_func(					\
+	    bp->fb_payload.fb_##what.fb_##what##_##a0,			\
+	    bp->fb_payload.fb_##what.fb_##what##_##a1,			\
+	    bp->fb_payload.fb_##what.fb_##what##_##a2,			\
+	    bp->fb_payload.fb_##what.fb_##what##_##a3,			\
+	    bp->fb_payload.fb_##what.fb_##what##_##a4,			\
+	    bp->fb_payload.fb_##what.fb_##what##_##a5,			\
+	    bp->fb_payload.fb_##what.fb_##what##_##a6,			\
+	    bp->fb_payload.fb_##what.fb_##what##_##a7);			\
+}
+
+#define	FEM_VNEXT10_DECL(what, a0, a1, a2, a3, a4, a5, a6, a7, a8, a9)	\
+void									\
+fem_handoff_##what(fem_baton_t *bp)					\
+{									\
+	bp->fb_rval = bp->fb_func(					\
+	    bp->fb_payload.fb_##what.fb_##what##_##a0,			\
+	    bp->fb_payload.fb_##what.fb_##what##_##a1,			\
+	    bp->fb_payload.fb_##what.fb_##what##_##a2,			\
+	    bp->fb_payload.fb_##what.fb_##what##_##a3,			\
+	    bp->fb_payload.fb_##what.fb_##what##_##a4,			\
+	    bp->fb_payload.fb_##what.fb_##what##_##a5,			\
+	    bp->fb_payload.fb_##what.fb_##what##_##a6,			\
+	    bp->fb_payload.fb_##what.fb_##what##_##a7,			\
+	    bp->fb_payload.fb_##what.fb_##what##_##a8,			\
+	    bp->fb_payload.fb_##what.fb_##what##_##a9);			\
+}
+
+#define	FEM_VNEXT11_DECL(what, a0, a1, a2, a3, a4, a5, a6, a7, a8, a9, a10) \
+void									\
+fem_handoff_##what(fem_baton_t *bp)					\
+{									\
+	bp->fb_rval = bp->fb_func(					\
+	    bp->fb_payload.fb_##what.fb_##what##_##a0,			\
+	    bp->fb_payload.fb_##what.fb_##what##_##a1,			\
+	    bp->fb_payload.fb_##what.fb_##what##_##a2,			\
+	    bp->fb_payload.fb_##what.fb_##what##_##a3,			\
+	    bp->fb_payload.fb_##what.fb_##what##_##a4,			\
+	    bp->fb_payload.fb_##what.fb_##what##_##a5,			\
+	    bp->fb_payload.fb_##what.fb_##what##_##a6,			\
+	    bp->fb_payload.fb_##what.fb_##what##_##a7,			\
+	    bp->fb_payload.fb_##what.fb_##what##_##a8,			\
+	    bp->fb_payload.fb_##what.fb_##what##_##a9,			\
+	    bp->fb_payload.fb_##what.fb_##what##_##a10);		\
+}
+
+#define	FEM_VNEXT3(what, func, a0, a1, a2)				\
+	if (FEM_TOODEEP()) {						\
+		fem_baton_t *baton;					\
+		int rval;						\
+									\
+		baton = kmem_alloc(sizeof (fem_baton_t), KM_SLEEP);	\
+		baton->fb_payload.fb_##what.fb_##what##_##a0 = a0;	\
+		baton->fb_payload.fb_##what.fb_##what##_##a1 = a1;	\
+		baton->fb_payload.fb_##what.fb_##what##_##a2 = a2;	\
+		baton->fb_handoff = fem_handoff_##what;			\
+		baton->fb_func = func;					\
+									\
+		rval = fem_handoff(baton);				\
+		kmem_free(baton, sizeof (fem_baton_t));			\
+									\
+		return (rval);						\
+	}								\
+	return (func(a0, a1, a2))
+
+#define	FEM_VNEXT4(what, func, a0, a1, a2, a3)				\
+	if (FEM_TOODEEP()) {						\
+		fem_baton_t *baton;					\
+		int rval;						\
+									\
+		baton = kmem_alloc(sizeof (fem_baton_t), KM_SLEEP);	\
+		baton->fb_payload.fb_##what.fb_##what##_##a0 = a0;	\
+		baton->fb_payload.fb_##what.fb_##what##_##a1 = a1;	\
+		baton->fb_payload.fb_##what.fb_##what##_##a2 = a2;	\
+		baton->fb_payload.fb_##what.fb_##what##_##a3 = a3;	\
+		baton->fb_handoff = fem_handoff_##what;			\
+		baton->fb_func = func;					\
+									\
+		rval = fem_handoff(baton);				\
+		kmem_free(baton, sizeof (fem_baton_t));			\
+									\
+		return (rval);						\
+	}								\
+	return (func(a0, a1, a2, a3))
+
+#define	FEM_VNEXT5(what, func, a0, a1, a2, a3, a4)			\
+	if (FEM_TOODEEP()) {						\
+		fem_baton_t *baton;					\
+		int rval;						\
+									\
+		baton = kmem_alloc(sizeof (fem_baton_t), KM_SLEEP);	\
+		baton->fb_payload.fb_##what.fb_##what##_##a0 = a0;	\
+		baton->fb_payload.fb_##what.fb_##what##_##a1 = a1;	\
+		baton->fb_payload.fb_##what.fb_##what##_##a2 = a2;	\
+		baton->fb_payload.fb_##what.fb_##what##_##a3 = a3;	\
+		baton->fb_payload.fb_##what.fb_##what##_##a4 = a4;	\
+		baton->fb_handoff = fem_handoff_##what;			\
+		baton->fb_func = func;					\
+									\
+		rval = fem_handoff(baton);				\
+		kmem_free(baton, sizeof (fem_baton_t));			\
+									\
+		return (rval);						\
+	}								\
+	return (func(a0, a1, a2, a3, a4))
+
+#define	FEM_VNEXT6(what, func, a0, a1, a2, a3, a4, a5)			\
+	if (FEM_TOODEEP()) {						\
+		fem_baton_t *baton;					\
+		int rval;						\
+									\
+		baton = kmem_alloc(sizeof (fem_baton_t), KM_SLEEP);	\
+		baton->fb_payload.fb_##what.fb_##what##_##a0 = a0;	\
+		baton->fb_payload.fb_##what.fb_##what##_##a1 = a1;	\
+		baton->fb_payload.fb_##what.fb_##what##_##a2 = a2;	\
+		baton->fb_payload.fb_##what.fb_##what##_##a3 = a3;	\
+		baton->fb_payload.fb_##what.fb_##what##_##a4 = a4;	\
+		baton->fb_payload.fb_##what.fb_##what##_##a5 = a5;	\
+		baton->fb_handoff = fem_handoff_##what;			\
+		baton->fb_func = func;					\
+									\
+		rval = fem_handoff(baton);				\
+		kmem_free(baton, sizeof (fem_baton_t));			\
+									\
+		return (rval);						\
+	}								\
+	return (func(a0, a1, a2, a3, a4, a5))
+
+#define	FEM_VNEXT7(what, func, a0, a1, a2, a3, a4, a5, a6)		\
+	if (FEM_TOODEEP()) {						\
+		fem_baton_t *baton;					\
+		int rval;						\
+									\
+		baton = kmem_alloc(sizeof (fem_baton_t), KM_SLEEP);	\
+		baton->fb_payload.fb_##what.fb_##what##_##a0 = a0;	\
+		baton->fb_payload.fb_##what.fb_##what##_##a1 = a1;	\
+		baton->fb_payload.fb_##what.fb_##what##_##a2 = a2;	\
+		baton->fb_payload.fb_##what.fb_##what##_##a3 = a3;	\
+		baton->fb_payload.fb_##what.fb_##what##_##a4 = a4;	\
+		baton->fb_payload.fb_##what.fb_##what##_##a5 = a5;	\
+		baton->fb_payload.fb_##what.fb_##what##_##a6 = a6;	\
+		baton->fb_handoff = fem_handoff_##what;			\
+		baton->fb_func = func;					\
+									\
+		rval = fem_handoff(baton);				\
+		kmem_free(baton, sizeof (fem_baton_t));			\
+									\
+		return (rval);						\
+	}								\
+	return (func(a0, a1, a2, a3, a4, a5, a6))
+
+#define	FEM_VNEXT8(what, func, a0, a1, a2, a3, a4, a5, a6, a7)		\
+	if (FEM_TOODEEP()) {						\
+		fem_baton_t *baton;					\
+		int rval;						\
+									\
+		baton = kmem_alloc(sizeof (fem_baton_t), KM_SLEEP);	\
+		baton->fb_payload.fb_##what.fb_##what##_##a0 = a0;	\
+		baton->fb_payload.fb_##what.fb_##what##_##a1 = a1;	\
+		baton->fb_payload.fb_##what.fb_##what##_##a2 = a2;	\
+		baton->fb_payload.fb_##what.fb_##what##_##a3 = a3;	\
+		baton->fb_payload.fb_##what.fb_##what##_##a4 = a4;	\
+		baton->fb_payload.fb_##what.fb_##what##_##a5 = a5;	\
+		baton->fb_payload.fb_##what.fb_##what##_##a6 = a6;	\
+		baton->fb_payload.fb_##what.fb_##what##_##a7 = a7;	\
+		baton->fb_handoff = fem_handoff_##what;			\
+		baton->fb_func = func;					\
+									\
+		rval = fem_handoff(baton);				\
+		kmem_free(baton, sizeof (fem_baton_t));			\
+									\
+		return (rval);						\
+	}								\
+	return (func(a0, a1, a2, a3, a4, a5, a6, a7))
+
+#define	FEM_VNEXT10(what, func, a0, a1, a2, a3, a4, a5, a6, a7, a8, a9)	\
+	if (FEM_TOODEEP()) {						\
+		fem_baton_t *baton;					\
+		int rval;						\
+									\
+		baton = kmem_alloc(sizeof (fem_baton_t), KM_SLEEP);	\
+		baton->fb_payload.fb_##what.fb_##what##_##a0 = a0;	\
+		baton->fb_payload.fb_##what.fb_##what##_##a1 = a1;	\
+		baton->fb_payload.fb_##what.fb_##what##_##a2 = a2;	\
+		baton->fb_payload.fb_##what.fb_##what##_##a3 = a3;	\
+		baton->fb_payload.fb_##what.fb_##what##_##a4 = a4;	\
+		baton->fb_payload.fb_##what.fb_##what##_##a5 = a5;	\
+		baton->fb_payload.fb_##what.fb_##what##_##a6 = a6;	\
+		baton->fb_payload.fb_##what.fb_##what##_##a7 = a7;	\
+		baton->fb_payload.fb_##what.fb_##what##_##a8 = a8;	\
+		baton->fb_payload.fb_##what.fb_##what##_##a9 = a9;	\
+		baton->fb_handoff = fem_handoff_##what;			\
+		baton->fb_func = func;					\
+									\
+		rval = fem_handoff(baton);				\
+		kmem_free(baton, sizeof (fem_baton_t));			\
+									\
+		return (rval);						\
+	}								\
+	return (func(a0, a1, a2, a3, a4, a5, a6, a7, a8, a9))
+
+#define	FEM_VNEXT11(what, func, a0, a1, a2, a3, a4, a5, a6, a7, a8, a9, a10) \
+	if (FEM_TOODEEP()) {						\
+		fem_baton_t *baton;					\
+		int rval;						\
+									\
+		baton = kmem_alloc(sizeof (fem_baton_t), KM_SLEEP);	\
+		baton->fb_payload.fb_##what.fb_##what##_##a0 = a0;	\
+		baton->fb_payload.fb_##what.fb_##what##_##a1 = a1;	\
+		baton->fb_payload.fb_##what.fb_##what##_##a2 = a2;	\
+		baton->fb_payload.fb_##what.fb_##what##_##a3 = a3;	\
+		baton->fb_payload.fb_##what.fb_##what##_##a4 = a4;	\
+		baton->fb_payload.fb_##what.fb_##what##_##a5 = a5;	\
+		baton->fb_payload.fb_##what.fb_##what##_##a6 = a6;	\
+		baton->fb_payload.fb_##what.fb_##what##_##a7 = a7;	\
+		baton->fb_payload.fb_##what.fb_##what##_##a8 = a8;	\
+		baton->fb_payload.fb_##what.fb_##what##_##a9 = a9;	\
+		baton->fb_payload.fb_##what.fb_##what##_##a10 = a10;	\
+		baton->fb_handoff = fem_handoff_##what;			\
+		baton->fb_func = func;					\
+									\
+		rval = fem_handoff(baton);				\
+		kmem_free(baton, sizeof (fem_baton_t));			\
+									\
+		return (rval);						\
+	}								\
+	return (func(a0, a1, a2, a3, a4, a5, a6, a7, a8, a9, a10))
+
 static fem_t *
 fem_alloc()
 {
@@ -2036,10 +2571,60 @@ static struct fs_operation_def fshead_vfs_spec[]  = {
  * 5.  Return by invoking the base operation with the base object.
  *
  * for each classification, there needs to be at least one "next" operation
- * for each "head"operation.
- *
+ * for each "head" operation.  Note that we also use the FEM_VNEXTn_DECL macros
+ * to define the function to run when the stack is split; see the discussion
+ * on "File event monitoring handoffs", above.
  */
 
+FEM_VNEXT4_DECL(open, arg0, mode, cr, ct)
+FEM_VNEXT6_DECL(close, arg0, flag, count, offset, cr, ct)
+FEM_VNEXT5_DECL(read, arg0, uiop, ioflag, cr, ct)
+FEM_VNEXT5_DECL(write, arg0, uiop, ioflag, cr, ct)
+FEM_VNEXT7_DECL(ioctl, arg0, cmd, arg, flag, cr, rvalp, ct)
+FEM_VNEXT5_DECL(setfl, arg0, oflags, nflags, cr, ct)
+FEM_VNEXT5_DECL(getattr, arg0, vap, flags, cr, ct)
+FEM_VNEXT5_DECL(setattr, arg0, vap, flags, cr, ct)
+FEM_VNEXT5_DECL(access, arg0, mode, flags, cr, ct)
+FEM_VNEXT10_DECL(lookup, arg0, nm, vpp, pnp, flags, rdir,
+    cr, ct, direntflags, realpnp)
+FEM_VNEXT10_DECL(create, arg0, name, vap, excl, mode, vpp, cr, flag, ct, vsecp)
+FEM_VNEXT5_DECL(remove, arg0, nm, cr, ct, flags)
+FEM_VNEXT6_DECL(link, arg0, svp, tnm, cr, ct, flags)
+FEM_VNEXT7_DECL(rename, arg0, snm, tdvp, tnm, cr, ct, flags)
+FEM_VNEXT8_DECL(mkdir, arg0, dirname, vap, vpp, cr, ct, flags, vsecp)
+FEM_VNEXT6_DECL(rmdir, arg0, nm, cdir, cr, ct, flags)
+FEM_VNEXT6_DECL(readdir, arg0, uiop, cr, eofp, ct, flags)
+FEM_VNEXT7_DECL(symlink, arg0, linkname, vap, target, cr, ct, flags)
+FEM_VNEXT4_DECL(readlink, arg0, uiop, cr, ct)
+FEM_VNEXT4_DECL(fsync, arg0, syncflag, cr, ct)
+FEM_VNEXT3_DECL(fid, arg0, fidp, ct)
+FEM_VNEXT3_DECL(rwlock, arg0, write_lock, ct)
+FEM_VNEXT4_DECL(seek, arg0, ooff, noffp, ct)
+FEM_VNEXT3_DECL(cmp, arg0, vp2, ct)
+FEM_VNEXT8_DECL(frlock, arg0, cmd, bfp, flag, offset, flk_cbp, cr, ct)
+FEM_VNEXT7_DECL(space, arg0, cmd, bfp, flag, offset, cr, ct)
+FEM_VNEXT3_DECL(realvp, arg0, vpp, ct)
+FEM_VNEXT11_DECL(getpage, arg0, off, len, protp, plarr, plsz,
+    seg, addr, rw, cr, ct)
+FEM_VNEXT6_DECL(putpage, arg0, off, len, flags, cr, ct)
+FEM_VNEXT10_DECL(map, arg0, off, as, addrp, len, prot, maxprot,
+    flags, cr, ct)
+FEM_VNEXT10_DECL(addmap, arg0, off, as, addr, len, prot, maxprot,
+    flags, cr, ct)
+FEM_VNEXT10_DECL(delmap, arg0, off, as, addr, len, prot, maxprot,
+    flags, cr, ct)
+FEM_VNEXT6_DECL(poll, arg0, events, anyyet, reventsp, phpp, ct)
+FEM_VNEXT5_DECL(dump, arg0, addr, lbdn, dblks, ct)
+FEM_VNEXT5_DECL(pathconf, arg0, cmd, valp, cr, ct)
+FEM_VNEXT7_DECL(pageio, arg0, pp, io_off, io_len, flags, cr, ct)
+FEM_VNEXT4_DECL(dumpctl, arg0, action, blkp, ct)
+FEM_VNEXT5_DECL(setsecattr, arg0, vsap, flag, cr, ct)
+FEM_VNEXT5_DECL(getsecattr, arg0, vsap, flag, cr, ct)
+FEM_VNEXT6_DECL(shrlock, arg0, cmd, shr, flag, cr, ct)
+FEM_VNEXT5_DECL(vnevent, arg0, vnevent, dvp, cname, ct)
+FEM_VNEXT5_DECL(reqzcbuf, arg0, ioflag, xuiop, cr, ct)
+FEM_VNEXT4_DECL(retzcbuf, arg0, xuiop, cr, ct)
+
 int
 vnext_open(femarg_t *vf, int mode, cred_t *cr, caller_context_t *ct)
 {
@@ -2051,7 +2636,7 @@ vnext_open(femarg_t *vf, int mode, cred_t *cr, caller_context_t *ct)
 	vsop_find(vf, &func, int, &arg0, vop_open, femop_open);
 	ASSERT(func != NULL);
 	ASSERT(arg0 != NULL);
-	return ((*func)(arg0, mode, cr, ct));
+	FEM_VNEXT4(open, func, arg0, mode, cr, ct);
 }
 
 int
@@ -2066,7 +2651,7 @@ vnext_close(femarg_t *vf, int flag, int count, offset_t offset, cred_t *cr,
 	vsop_find(vf, &func, int, &arg0, vop_close, femop_close);
 	ASSERT(func != NULL);
 	ASSERT(arg0 != NULL);
-	return ((*func)(arg0, flag, count, offset, cr, ct));
+	FEM_VNEXT6(close, func, arg0, flag, count, offset, cr, ct);
 }
 
 int
@@ -2081,7 +2666,7 @@ vnext_read(femarg_t *vf, uio_t *uiop, int ioflag, cred_t *cr,
 	vsop_find(vf, &func, int, &arg0, vop_read, femop_read);
 	ASSERT(func != NULL);
 	ASSERT(arg0 != NULL);
-	return ((*func)(arg0, uiop, ioflag, cr, ct));
+	FEM_VNEXT5(read, func, arg0, uiop, ioflag, cr, ct);
 }
 
 int
@@ -2096,7 +2681,7 @@ vnext_write(femarg_t *vf, uio_t *uiop, int ioflag, cred_t *cr,
 	vsop_find(vf, &func, int, &arg0, vop_write, femop_write);
 	ASSERT(func != NULL);
 	ASSERT(arg0 != NULL);
-	return ((*func)(arg0, uiop, ioflag, cr, ct));
+	FEM_VNEXT5(write, func, arg0, uiop, ioflag, cr, ct);
 }
 
 int
@@ -2111,7 +2696,7 @@ vnext_ioctl(femarg_t *vf, int cmd, intptr_t arg, int flag, cred_t *cr,
 	vsop_find(vf, &func, int, &arg0, vop_ioctl, femop_ioctl);
 	ASSERT(func != NULL);
 	ASSERT(arg0 != NULL);
-	return ((*func)(arg0, cmd, arg, flag, cr, rvalp, ct));
+	FEM_VNEXT7(ioctl, func, arg0, cmd, arg, flag, cr, rvalp, ct);
 }
 
 int
@@ -2126,7 +2711,7 @@ vnext_setfl(femarg_t *vf, int oflags, int nflags, cred_t *cr,
 	vsop_find(vf, &func, int, &arg0, vop_setfl, femop_setfl);
 	ASSERT(func != NULL);
 	ASSERT(arg0 != NULL);
-	return ((*func)(arg0, oflags, nflags, cr, ct));
+	FEM_VNEXT5(setfl, func, arg0, oflags, nflags, cr, ct);
 }
 
 int
@@ -2141,7 +2726,7 @@ vnext_getattr(femarg_t *vf, vattr_t *vap, int flags, cred_t *cr,
 	vsop_find(vf, &func, int, &arg0, vop_getattr, femop_getattr);
 	ASSERT(func != NULL);
 	ASSERT(arg0 != NULL);
-	return ((*func)(arg0, vap, flags, cr, ct));
+	FEM_VNEXT5(getattr, func, arg0, vap, flags, cr, ct);
 }
 
 int
@@ -2156,7 +2741,7 @@ vnext_setattr(femarg_t *vf, vattr_t *vap, int flags, cred_t *cr,
 	vsop_find(vf, &func, int, &arg0, vop_setattr, femop_setattr);
 	ASSERT(func != NULL);
 	ASSERT(arg0 != NULL);
-	return ((*func)(arg0, vap, flags, cr, ct));
+	FEM_VNEXT5(setattr, func, arg0, vap, flags, cr, ct);
 }
 
 int
@@ -2171,7 +2756,7 @@ vnext_access(femarg_t *vf, int mode, int flags, cred_t *cr,
 	vsop_find(vf, &func, int, &arg0, vop_access, femop_access);
 	ASSERT(func != NULL);
 	ASSERT(arg0 != NULL);
-	return ((*func)(arg0, mode, flags, cr, ct));
+	FEM_VNEXT5(access, func, arg0, mode, flags, cr, ct);
 }
 
 int
@@ -2187,8 +2772,8 @@ vnext_lookup(femarg_t *vf, char *nm, vnode_t **vpp, pathname_t *pnp,
 	vsop_find(vf, &func, int, &arg0, vop_lookup, femop_lookup);
 	ASSERT(func != NULL);
 	ASSERT(arg0 != NULL);
-	return ((*func)(arg0, nm, vpp, pnp, flags, rdir, cr, ct,
-	    direntflags, realpnp));
+	FEM_VNEXT10(lookup, func, arg0, nm, vpp, pnp, flags, rdir, cr, ct,
+	    direntflags, realpnp);
 }
 
 int
@@ -2204,7 +2789,8 @@ vnext_create(femarg_t *vf, char *name, vattr_t *vap, vcexcl_t excl,
 	vsop_find(vf, &func, int, &arg0, vop_create, femop_create);
 	ASSERT(func != NULL);
 	ASSERT(arg0 != NULL);
-	return ((*func)(arg0, name, vap, excl, mode, vpp, cr, flag, ct, vsecp));
+	FEM_VNEXT10(create, func, arg0, name, vap, excl,
+	    mode, vpp, cr, flag, ct, vsecp);
 }
 
 int
@@ -2219,7 +2805,7 @@ vnext_remove(femarg_t *vf, char *nm, cred_t *cr, caller_context_t *ct,
 	vsop_find(vf, &func, int, &arg0, vop_remove, femop_remove);
 	ASSERT(func != NULL);
 	ASSERT(arg0 != NULL);
-	return ((*func)(arg0, nm, cr, ct, flags));
+	FEM_VNEXT5(remove, func, arg0, nm, cr, ct, flags);
 }
 
 int
@@ -2234,7 +2820,7 @@ vnext_link(femarg_t *vf, vnode_t *svp, char *tnm, cred_t *cr,
 	vsop_find(vf, &func, int, &arg0, vop_link, femop_link);
 	ASSERT(func != NULL);
 	ASSERT(arg0 != NULL);
-	return ((*func)(arg0, svp, tnm, cr, ct, flags));
+	FEM_VNEXT6(link, func, arg0, svp, tnm, cr, ct, flags);
 }
 
 int
@@ -2249,7 +2835,7 @@ vnext_rename(femarg_t *vf, char *snm, vnode_t *tdvp, char *tnm, cred_t *cr,
 	vsop_find(vf, &func, int, &arg0, vop_rename, femop_rename);
 	ASSERT(func != NULL);
 	ASSERT(arg0 != NULL);
-	return ((*func)(arg0, snm, tdvp, tnm, cr, ct, flags));
+	FEM_VNEXT7(rename, func, arg0, snm, tdvp, tnm, cr, ct, flags);
 }
 
 int
@@ -2264,7 +2850,7 @@ vnext_mkdir(femarg_t *vf, char *dirname, vattr_t *vap, vnode_t **vpp,
 	vsop_find(vf, &func, int, &arg0, vop_mkdir, femop_mkdir);
 	ASSERT(func != NULL);
 	ASSERT(arg0 != NULL);
-	return ((*func)(arg0, dirname, vap, vpp, cr, ct, flags, vsecp));
+	FEM_VNEXT8(mkdir, func, arg0, dirname, vap, vpp, cr, ct, flags, vsecp);
 }
 
 int
@@ -2279,7 +2865,7 @@ vnext_rmdir(femarg_t *vf, char *nm, vnode_t *cdir, cred_t *cr,
 	vsop_find(vf, &func, int, &arg0, vop_rmdir, femop_rmdir);
 	ASSERT(func != NULL);
 	ASSERT(arg0 != NULL);
-	return ((*func)(arg0, nm, cdir, cr, ct, flags));
+	FEM_VNEXT6(rmdir, func, arg0, nm, cdir, cr, ct, flags);
 }
 
 int
@@ -2294,7 +2880,7 @@ vnext_readdir(femarg_t *vf, uio_t *uiop, cred_t *cr, int *eofp,
 	vsop_find(vf, &func, int, &arg0, vop_readdir, femop_readdir);
 	ASSERT(func != NULL);
 	ASSERT(arg0 != NULL);
-	return ((*func)(arg0, uiop, cr, eofp, ct, flags));
+	FEM_VNEXT6(readdir, func, arg0, uiop, cr, eofp, ct, flags);
 }
 
 int
@@ -2309,7 +2895,7 @@ vnext_symlink(femarg_t *vf, char *linkname, vattr_t *vap, char *target,
 	vsop_find(vf, &func, int, &arg0, vop_symlink, femop_symlink);
 	ASSERT(func != NULL);
 	ASSERT(arg0 != NULL);
-	return ((*func)(arg0, linkname, vap, target, cr, ct, flags));
+	FEM_VNEXT7(symlink, func, arg0, linkname, vap, target, cr, ct, flags);
 }
 
 int
@@ -2323,7 +2909,7 @@ vnext_readlink(femarg_t *vf, uio_t *uiop, cred_t *cr, caller_context_t *ct)
 	vsop_find(vf, &func, int, &arg0, vop_readlink, femop_readlink);
 	ASSERT(func != NULL);
 	ASSERT(arg0 != NULL);
-	return ((*func)(arg0, uiop, cr, ct));
+	FEM_VNEXT4(readlink, func, arg0, uiop, cr, ct);
 }
 
 int
@@ -2337,7 +2923,7 @@ vnext_fsync(femarg_t *vf, int syncflag, cred_t *cr, caller_context_t *ct)
 	vsop_find(vf, &func, int, &arg0, vop_fsync, femop_fsync);
 	ASSERT(func != NULL);
 	ASSERT(arg0 != NULL);
-	return ((*func)(arg0, syncflag, cr, ct));
+	FEM_VNEXT4(fsync, func, arg0, syncflag, cr, ct);
 }
 
 void
@@ -2365,7 +2951,7 @@ vnext_fid(femarg_t *vf, fid_t *fidp, caller_context_t *ct)
 	vsop_find(vf, &func, int, &arg0, vop_fid, femop_fid);
 	ASSERT(func != NULL);
 	ASSERT(arg0 != NULL);
-	return ((*func)(arg0, fidp, ct));
+	FEM_VNEXT3(fid, func, arg0, fidp, ct);
 }
 
 int
@@ -2379,7 +2965,7 @@ vnext_rwlock(femarg_t *vf, int write_lock, caller_context_t *ct)
 	vsop_find(vf, &func, int, &arg0, vop_rwlock, femop_rwlock);
 	ASSERT(func != NULL);
 	ASSERT(arg0 != NULL);
-	return ((*func)(arg0, write_lock, ct));
+	FEM_VNEXT3(rwlock, func, arg0, write_lock, ct);
 }
 
 void
@@ -2407,7 +2993,7 @@ vnext_seek(femarg_t *vf, offset_t ooff, offset_t *noffp, caller_context_t *ct)
 	vsop_find(vf, &func, int, &arg0, vop_seek, femop_seek);
 	ASSERT(func != NULL);
 	ASSERT(arg0 != NULL);
-	return ((*func)(arg0, ooff, noffp, ct));
+	FEM_VNEXT4(seek, func, arg0, ooff, noffp, ct);
 }
 
 int
@@ -2421,7 +3007,7 @@ vnext_cmp(femarg_t *vf, vnode_t *vp2, caller_context_t *ct)
 	vsop_find(vf, &func, int, &arg0, vop_cmp, femop_cmp);
 	ASSERT(func != NULL);
 	ASSERT(arg0 != NULL);
-	return ((*func)(arg0, vp2, ct));
+	FEM_VNEXT3(cmp, func, arg0, vp2, ct);
 }
 
 int
@@ -2437,7 +3023,7 @@ vnext_frlock(femarg_t *vf, int cmd, struct flock64 *bfp, int flag,
 	vsop_find(vf, &func, int, &arg0, vop_frlock, femop_frlock);
 	ASSERT(func != NULL);
 	ASSERT(arg0 != NULL);
-	return ((*func)(arg0, cmd, bfp, flag, offset, flk_cbp, cr, ct));
+	FEM_VNEXT8(frlock, func, arg0, cmd, bfp, flag, offset, flk_cbp, cr, ct);
 }
 
 int
@@ -2452,7 +3038,7 @@ vnext_space(femarg_t *vf, int cmd, struct flock64 *bfp, int flag,
 	vsop_find(vf, &func, int, &arg0, vop_space, femop_space);
 	ASSERT(func != NULL);
 	ASSERT(arg0 != NULL);
-	return ((*func)(arg0, cmd, bfp, flag, offset, cr, ct));
+	FEM_VNEXT7(space, func, arg0, cmd, bfp, flag, offset, cr, ct);
 }
 
 int
@@ -2466,7 +3052,7 @@ vnext_realvp(femarg_t *vf, vnode_t **vpp, caller_context_t *ct)
 	vsop_find(vf, &func, int, &arg0, vop_realvp, femop_realvp);
 	ASSERT(func != NULL);
 	ASSERT(arg0 != NULL);
-	return ((*func)(arg0, vpp, ct));
+	FEM_VNEXT3(realvp, func, arg0, vpp, ct);
 }
 
 int
@@ -2482,8 +3068,8 @@ vnext_getpage(femarg_t *vf, offset_t off, size_t len, uint_t *protp,
 	vsop_find(vf, &func, int, &arg0, vop_getpage, femop_getpage);
 	ASSERT(func != NULL);
 	ASSERT(arg0 != NULL);
-	return ((*func)(arg0, off, len, protp, plarr, plsz, seg, addr, rw,
-	    cr, ct));
+	FEM_VNEXT11(getpage, func, arg0, off, len, protp,
+	    plarr, plsz, seg, addr, rw, cr, ct);
 }
 
 int
@@ -2498,7 +3084,7 @@ vnext_putpage(femarg_t *vf, offset_t off, size_t len, int flags,
 	vsop_find(vf, &func, int, &arg0, vop_putpage, femop_putpage);
 	ASSERT(func != NULL);
 	ASSERT(arg0 != NULL);
-	return ((*func)(arg0, off, len, flags, cr, ct));
+	FEM_VNEXT6(putpage, func, arg0, off, len, flags, cr, ct);
 }
 
 int
@@ -2514,8 +3100,8 @@ vnext_map(femarg_t *vf, offset_t off, struct as *as, caddr_t *addrp,
 	vsop_find(vf, &func, int, &arg0, vop_map, femop_map);
 	ASSERT(func != NULL);
 	ASSERT(arg0 != NULL);
-	return ((*func)(arg0, off, as, addrp, len, prot, maxprot, flags,
-	    cr, ct));
+	FEM_VNEXT10(map, func, arg0, off, as, addrp, len, prot, maxprot, flags,
+	    cr, ct);
 }
 
 int
@@ -2531,8 +3117,8 @@ vnext_addmap(femarg_t *vf, offset_t off, struct as *as, caddr_t addr,
 	vsop_find(vf, &func, int, &arg0, vop_addmap, femop_addmap);
 	ASSERT(func != NULL);
 	ASSERT(arg0 != NULL);
-	return ((*func)(arg0, off, as, addr, len, prot, maxprot, flags,
-	    cr, ct));
+	FEM_VNEXT10(addmap, func, arg0, off, as, addr, len, prot, maxprot,
+	    flags, cr, ct);
 }
 
 int
@@ -2548,8 +3134,8 @@ vnext_delmap(femarg_t *vf, offset_t off, struct as *as, caddr_t addr,
 	vsop_find(vf, &func, int, &arg0, vop_delmap, femop_delmap);
 	ASSERT(func != NULL);
 	ASSERT(arg0 != NULL);
-	return ((*func)(arg0, off, as, addr, len, prot, maxprot, flags,
-	    cr, ct));
+	FEM_VNEXT10(delmap, func, arg0, off, as, addr, len, prot, maxprot,
+	    flags, cr, ct);
 }
 
 int
@@ -2564,7 +3150,7 @@ vnext_poll(femarg_t *vf, short events, int anyyet, short *reventsp,
 	vsop_find(vf, &func, int, &arg0, vop_poll, femop_poll);
 	ASSERT(func != NULL);
 	ASSERT(arg0 != NULL);
-	return ((*func)(arg0, events, anyyet, reventsp, phpp, ct));
+	FEM_VNEXT6(poll, func, arg0, events, anyyet, reventsp, phpp, ct);
 }
 
 int
@@ -2579,7 +3165,7 @@ vnext_dump(femarg_t *vf, caddr_t addr, offset_t lbdn, offset_t dblks,
 	vsop_find(vf, &func, int, &arg0, vop_dump, femop_dump);
 	ASSERT(func != NULL);
 	ASSERT(arg0 != NULL);
-	return ((*func)(arg0, addr, lbdn, dblks, ct));
+	FEM_VNEXT5(dump, func, arg0, addr, lbdn, dblks, ct);
 }
 
 int
@@ -2594,7 +3180,7 @@ vnext_pathconf(femarg_t *vf, int cmd, ulong_t *valp, cred_t *cr,
 	vsop_find(vf, &func, int, &arg0, vop_pathconf, femop_pathconf);
 	ASSERT(func != NULL);
 	ASSERT(arg0 != NULL);
-	return ((*func)(arg0, cmd, valp, cr, ct));
+	FEM_VNEXT5(pathconf, func, arg0, cmd, valp, cr, ct);
 }
 
 int
@@ -2609,7 +3195,7 @@ vnext_pageio(femarg_t *vf, struct page *pp, u_offset_t io_off,
 	vsop_find(vf, &func, int, &arg0, vop_pageio, femop_pageio);
 	ASSERT(func != NULL);
 	ASSERT(arg0 != NULL);
-	return ((*func)(arg0, pp, io_off, io_len, flags, cr, ct));
+	FEM_VNEXT7(pageio, func, arg0, pp, io_off, io_len, flags, cr, ct);
 }
 
 int
@@ -2623,7 +3209,7 @@ vnext_dumpctl(femarg_t *vf, int action, offset_t *blkp, caller_context_t *ct)
 	vsop_find(vf, &func, int, &arg0, vop_dumpctl, femop_dumpctl);
 	ASSERT(func != NULL);
 	ASSERT(arg0 != NULL);
-	return ((*func)(arg0, action, blkp, ct));
+	FEM_VNEXT4(dumpctl, func, arg0, action, blkp, ct);
 }
 
 void
@@ -2653,7 +3239,7 @@ vnext_setsecattr(femarg_t *vf, vsecattr_t *vsap, int flag, cred_t *cr,
 	vsop_find(vf, &func, int, &arg0, vop_setsecattr, femop_setsecattr);
 	ASSERT(func != NULL);
 	ASSERT(arg0 != NULL);
-	return ((*func)(arg0, vsap, flag, cr, ct));
+	FEM_VNEXT5(setsecattr, func, arg0, vsap, flag, cr, ct);
 }
 
 int
@@ -2668,7 +3254,7 @@ vnext_getsecattr(femarg_t *vf, vsecattr_t *vsap, int flag, cred_t *cr,
 	vsop_find(vf, &func, int, &arg0, vop_getsecattr, femop_getsecattr);
 	ASSERT(func != NULL);
 	ASSERT(arg0 != NULL);
-	return ((*func)(arg0, vsap, flag, cr, ct));
+	FEM_VNEXT5(getsecattr, func, arg0, vsap, flag, cr, ct);
 }
 
 int
@@ -2683,7 +3269,7 @@ vnext_shrlock(femarg_t *vf, int cmd, struct shrlock *shr, int flag,
 	vsop_find(vf, &func, int, &arg0, vop_shrlock, femop_shrlock);
 	ASSERT(func != NULL);
 	ASSERT(arg0 != NULL);
-	return ((*func)(arg0, cmd, shr, flag, cr, ct));
+	FEM_VNEXT6(shrlock, func, arg0, cmd, shr, flag, cr, ct);
 }
 
 int
@@ -2698,7 +3284,7 @@ vnext_vnevent(femarg_t *vf, vnevent_t vnevent, vnode_t *dvp, char *cname,
 	vsop_find(vf, &func, int, &arg0, vop_vnevent, femop_vnevent);
 	ASSERT(func != NULL);
 	ASSERT(arg0 != NULL);
-	return ((*func)(arg0, vnevent, dvp, cname, ct));
+	FEM_VNEXT5(vnevent, func, arg0, vnevent, dvp, cname, ct);
 }
 
 int
@@ -2713,7 +3299,7 @@ vnext_reqzcbuf(femarg_t *vf, enum uio_rw ioflag, xuio_t *xuiop, cred_t *cr,
 	vsop_find(vf, &func, int, &arg0, vop_reqzcbuf, femop_reqzcbuf);
 	ASSERT(func != NULL);
 	ASSERT(arg0 != NULL);
-	return ((*func)(arg0, ioflag, xuiop, cr, ct));
+	FEM_VNEXT5(reqzcbuf, func, arg0, ioflag, xuiop, cr, ct);
 }
 
 int
@@ -2727,7 +3313,7 @@ vnext_retzcbuf(femarg_t *vf, xuio_t *xuiop, cred_t *cr, caller_context_t *ct)
 	vsop_find(vf, &func, int, &arg0, vop_retzcbuf, femop_retzcbuf);
 	ASSERT(func != NULL);
 	ASSERT(arg0 != NULL);
-	return ((*func)(arg0, xuiop, cr, ct));
+	FEM_VNEXT4(retzcbuf, func, arg0, xuiop, cr, ct);
 }
 
 int
diff --git a/usr/src/uts/common/fs/fifofs/fifosubr.c b/usr/src/uts/common/fs/fifofs/fifosubr.c
index 6e56000ffe..56204c6741 100644
--- a/usr/src/uts/common/fs/fifofs/fifosubr.c
+++ b/usr/src/uts/common/fs/fifofs/fifosubr.c
@@ -614,9 +614,12 @@ fifo_stropen(vnode_t **vpp, int flag, cred_t *crp, int dotwist, int lockheld)
 	/*
 	 * The other end of the pipe is almost closed so
 	 * reject any other open on this end of the pipe
-	 * This only happens with a pipe mounted under namefs
+	 * This normally only happens with a pipe mounted under namefs, but
+	 * we can also see an open via proc/fd, which should still succeed.
+	 * To indicate the proc/fd case the FKLYR flag is passed.
 	 */
-	if ((fnp->fn_flag & (FIFOCLOSE|ISPIPE)) == (FIFOCLOSE|ISPIPE)) {
+	if ((fnp->fn_flag & (FIFOCLOSE|ISPIPE)) == (FIFOCLOSE|ISPIPE) &&
+	    (flag & FKLYR) == 0) {
 		fifo_cleanup(oldvp, flag);
 		cv_broadcast(&fnp->fn_wait_cv);
 		if (!lockheld)
diff --git a/usr/src/uts/common/fs/fifofs/fifovnops.c b/usr/src/uts/common/fs/fifofs/fifovnops.c
index ac89e430c7..fee2924093 100644
--- a/usr/src/uts/common/fs/fifofs/fifovnops.c
+++ b/usr/src/uts/common/fs/fifofs/fifovnops.c
@@ -27,7 +27,9 @@
  * Use is subject to license terms.
  */
 
-#pragma ident	"%Z%%M%	%I%	%E% SMI"
+/*
+ * Copyright (c) 2014, Joyent, Inc.  All rights reserved.
+ */
 
 /*
  * FIFOFS file system vnode operations.  This file system
@@ -1832,17 +1834,16 @@ fifo_poll(vnode_t *vp, short events, int anyyet, short *reventsp,
 	}
 
 	/*
-	 * if we happened to get something, return
+	 * if we happened to get something and we're not edge-triggered, return
 	 */
-
-	if ((*reventsp = (short)retevents) != 0) {
+	if ((*reventsp = (short)retevents) != 0 && !(events & POLLET)) {
 		mutex_exit(&fnp->fn_lock->flk_lock);
 		return (0);
 	}
 
 	/*
-	 * If poll() has not found any events yet, set up event cell
-	 * to wake up the poll if a requested event occurs on this
+	 * If poll() has not found any events yet or we're edge-triggered, set
+	 * up event cell to wake up the poll if a requested event occurs on this
 	 * pipe/fifo.
 	 */
 	if (!anyyet) {
diff --git a/usr/src/uts/common/fs/fs_subr.c b/usr/src/uts/common/fs/fs_subr.c
index 744d269716..ec5b145a86 100644
--- a/usr/src/uts/common/fs/fs_subr.c
+++ b/usr/src/uts/common/fs/fs_subr.c
@@ -25,6 +25,7 @@
 /*
  * Copyright (c) 1989, 2010, Oracle and/or its affiliates. All rights reserved.
  * Copyright 2011 Nexenta Systems, Inc.  All rights reserved.
+ * Copyright 2015 Joyent, Inc.
  */
 
 /*
@@ -246,6 +247,7 @@ fs_frlock(register vnode_t *vp, int cmd, struct flock64 *bfp, int flag,
 	int frcmd;
 	int nlmid;
 	int error = 0;
+	boolean_t skip_lock = B_FALSE;
 	flk_callback_t serialize_callback;
 	int serialize = 0;
 	v_mode_t mode;
@@ -265,6 +267,17 @@ fs_frlock(register vnode_t *vp, int cmd, struct flock64 *bfp, int flag,
 		}
 		break;
 
+	case F_OFD_GETLK:
+		/*
+		 * TBD we do not support remote OFD locks at this time.
+		 */
+		if (flag & (F_REMOTELOCK | F_PXFSLOCK)) {
+			error = EINVAL;
+			goto done;
+		}
+		skip_lock = B_TRUE;
+		break;
+
 	case F_SETLK_NBMAND:
 		/*
 		 * Are NBMAND locks allowed on this file?
@@ -326,6 +339,20 @@ fs_frlock(register vnode_t *vp, int cmd, struct flock64 *bfp, int flag,
 		}
 		break;
 
+	case F_OFD_SETLK:
+	case F_OFD_SETLKW:
+	case F_FLOCK:
+	case F_FLOCKW:
+		/*
+		 * TBD we do not support remote OFD locks at this time.
+		 */
+		if (flag & (F_REMOTELOCK | F_PXFSLOCK)) {
+			error = EINVAL;
+			goto done;
+		}
+		skip_lock = B_TRUE;
+		break;
+
 	case F_HASREMOTELOCKS:
 		nlmid = GETNLMID(bfp->l_sysid);
 		if (nlmid != 0) {	/* booted as a cluster */
@@ -354,7 +381,8 @@ fs_frlock(register vnode_t *vp, int cmd, struct flock64 *bfp, int flag,
 		flk_cbp = &serialize_callback;
 	}
 
-	error = reclock(vp, bfp, frcmd, flag, offset, flk_cbp);
+	if (!skip_lock)
+		error = reclock(vp, bfp, frcmd, flag, offset, flk_cbp);
 
 done:
 	if (serialize)
diff --git a/usr/src/uts/common/fs/hyprlofs/hyprlofs_dir.c b/usr/src/uts/common/fs/hyprlofs/hyprlofs_dir.c
new file mode 100644
index 0000000000..05ee2c6e09
--- /dev/null
+++ b/usr/src/uts/common/fs/hyprlofs/hyprlofs_dir.c
@@ -0,0 +1,640 @@
+/*
+ * CDDL HEADER START
+ *
+ * The contents of this file are subject to the terms of the
+ * Common Development and Distribution License (the "License").
+ * You may not use this file except in compliance with the License.
+ *
+ * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
+ * or http://www.opensolaris.org/os/licensing.
+ * See the License for the specific language governing permissions
+ * and limitations under the License.
+ *
+ * When distributing Covered Code, include this CDDL HEADER in each
+ * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
+ * If applicable, add the following below this CDDL HEADER, with the
+ * fields enclosed by brackets "[]" replaced with your own identifying
+ * information: Portions Copyright [yyyy] [name of copyright owner]
+ *
+ * CDDL HEADER END
+ */
+/*
+ * Copyright 2012, Joyent, Inc.  All rights reserved.
+ */
+
+#include <sys/types.h>
+#include <sys/param.h>
+#include <sys/sysmacros.h>
+#include <sys/systm.h>
+#include <sys/time.h>
+#include <sys/vfs.h>
+#include <sys/vnode.h>
+#include <sys/errno.h>
+#include <sys/cmn_err.h>
+#include <sys/cred.h>
+#include <sys/stat.h>
+#include <sys/policy.h>
+#include <sys/fs/hyprlofs_info.h>
+
+static int hldir_make_hlnode(hlnode_t *, hlfsmount_t *, vattr_t *, enum de_op,
+		vnode_t *, hlnode_t **, cred_t *);
+static int hldiraddentry(hlnode_t *, hlnode_t *, char *);
+
+
+#define	HL_HASH_SIZE	8192		/* must be power of 2 */
+#define	HL_MUTEX_SIZE	64
+
+static hldirent_t	*hl_hashtable[HL_HASH_SIZE];
+static kmutex_t		 hl_hashmutex[HL_MUTEX_SIZE];
+
+#define	HL_HASH_INDEX(a)	((a) & (HL_HASH_SIZE-1))
+#define	HL_MUTEX_INDEX(a)	((a) & (HL_MUTEX_SIZE-1))
+
+#define	HYPRLOFS_HASH(tp, name, hash)				\
+	{							\
+		char Xc, *Xcp;					\
+		hash = (uint_t)(uintptr_t)(tp) >> 8;		\
+		for (Xcp = (name); (Xc = *Xcp) != 0; Xcp++)	\
+			hash = (hash << 4) + hash + (uint_t)Xc;	\
+	}
+
+void
+hyprlofs_hash_init(void)
+{
+	int	ix;
+
+	for (ix = 0; ix < HL_MUTEX_SIZE; ix++)
+		mutex_init(&hl_hashmutex[ix], NULL, MUTEX_DEFAULT, NULL);
+}
+
+static void
+hyprlofs_hash_in(hldirent_t *h)
+{
+	uint_t		hash;
+	hldirent_t	**prevpp;
+	kmutex_t	*hmtx;
+
+	HYPRLOFS_HASH(h->hld_parent, h->hld_name, hash);
+	h->hld_hash = hash;
+	prevpp = &hl_hashtable[HL_HASH_INDEX(hash)];
+	hmtx = &hl_hashmutex[HL_MUTEX_INDEX(hash)];
+	mutex_enter(hmtx);
+	h->hld_link = *prevpp;
+	*prevpp = h;
+	mutex_exit(hmtx);
+}
+
+/* Remove hldirent *h from the hash list. */
+static void
+hyprlofs_hash_out(hldirent_t *h)
+{
+	uint_t		hash;
+	hldirent_t	**prevpp;
+	kmutex_t	*hmtx;
+
+	hash = h->hld_hash;
+	prevpp = &hl_hashtable[HL_HASH_INDEX(hash)];
+	hmtx = &hl_hashmutex[HL_MUTEX_INDEX(hash)];
+	mutex_enter(hmtx);
+	while (*prevpp != h)
+		prevpp = &(*prevpp)->hld_link;
+	*prevpp = h->hld_link;
+	mutex_exit(hmtx);
+}
+
+static hldirent_t *
+hyprlofs_hash_lookup(char *name, hlnode_t *parent, uint_t hold,
+    hlnode_t **found)
+{
+	hldirent_t	*l;
+	uint_t		hash;
+	kmutex_t	*hmtx;
+	hlnode_t	*hnp;
+
+	HYPRLOFS_HASH(parent, name, hash);
+	hmtx = &hl_hashmutex[HL_MUTEX_INDEX(hash)];
+	mutex_enter(hmtx);
+	l = hl_hashtable[HL_HASH_INDEX(hash)];
+	while (l) {
+		if (l->hld_hash == hash && l->hld_parent == parent &&
+		    strcmp(l->hld_name, name) == 0) {
+			/*
+			 * Ensure that the hlnode that we put a hold on is the
+			 * same one that we pass back. Thus the temp. var
+			 * hnp is necessary.
+			 */
+			hnp = l->hld_hlnode;
+			if (hold) {
+				ASSERT(hnp);
+				hlnode_hold(hnp);
+			}
+			if (found)
+				*found = hnp;
+			mutex_exit(hmtx);
+			return (l);
+		} else {
+			l = l->hld_link;
+		}
+	}
+	mutex_exit(hmtx);
+	return (NULL);
+}
+
+/*
+ * Search directory 'parent' for entry 'name'.
+ *
+ * The calling thread can't hold the write version of the rwlock for the
+ * directory being searched
+ *
+ * On success *foundtp points to the found hlnode with its vnode held.
+ */
+int
+hyprlofs_dirlookup(hlnode_t *parent, char *name, hlnode_t **foundtp, cred_t *cr)
+{
+	int error;
+
+	*foundtp = NULL;
+	if (parent->hln_type != VDIR)
+		return (ENOTDIR);
+
+	if ((error = hyprlofs_taccess(parent, VEXEC, cr)))
+		return (error);
+
+	if (*name == '\0') {
+		hlnode_hold(parent);
+		*foundtp = parent;
+		return (0);
+	}
+
+	/*
+	 * Search the directory for the matching name. We need the lock
+	 * protecting the hln_dir list so that it doesn't change out from
+	 * underneath us. hyprlofs_hash_lookup() will pass back the hlnode
+	 * with a hold on it.
+	 */
+	if (hyprlofs_hash_lookup(name, parent, 1, foundtp) != NULL) {
+		ASSERT(*foundtp);
+		return (0);
+	}
+
+	return (ENOENT);
+}
+
+/*
+ * Enter a directory entry (either a file or subdir, depending on op) for
+ * 'name' and 'hp' into directory 'dir'
+ */
+int
+hyprlofs_direnter(
+	hlfsmount_t	*hm,
+	hlnode_t	*dir,		/* target directory to make entry in */
+	char		*name,		/* name of entry */
+	enum de_op	op,		/* entry operation */
+	vnode_t		*realvp,	/* real vnode */
+	vattr_t		*va,
+	hlnode_t	**hpp,		/* return hlnode */
+	cred_t		*cr)
+{
+	hldirent_t *hdp;
+	hlnode_t *found = NULL;
+	hlnode_t *hp;
+	int error = 0;
+	char *s;
+
+	/* hln_rwlock is held to serialize direnter and dirdeletes */
+	ASSERT(RW_WRITE_HELD(&dir->hln_rwlock));
+	ASSERT(dir->hln_type == VDIR);
+
+	/* Don't allow '/' characters in pathname component */
+	for (s = name; *s; s++)
+		if (*s == '/')
+			return (EACCES);
+
+	if (name[0] == '\0')
+		panic("hyprlofs_direnter: NULL name");
+
+	/*
+	 * This might be a "dangling detached directory". It could have been
+	 * removed, but a reference to it kept in u_cwd. Don't bother searching
+	 * it, and with any luck the user will get tired of dealing with us and
+	 * cd to some absolute pathway. This is in ufs, too.
+	 */
+	if (dir->hln_nlink == 0) {
+		return (ENOENT);
+	}
+
+	/* Search for the entry.  Return "found" if it exists. */
+	hdp = hyprlofs_hash_lookup(name, dir, 1, &found);
+
+	if (hdp) {
+		ASSERT(found);
+		switch (op) {
+		case DE_CREATE:
+		case DE_MKDIR:
+			if (hpp) {
+				*hpp = found;
+				error = EEXIST;
+			} else {
+				hlnode_rele(found);
+			}
+			break;
+		}
+	} else {
+
+		/*
+		 * The entry does not exist. Check write perms in dir to see if
+		 * entry can be created.
+		 */
+		if ((error = hyprlofs_taccess(dir, VWRITE, cr)))
+			return (error);
+
+		/* Make new hlnode and directory entry as required. */
+		if ((error = hldir_make_hlnode(dir, hm, va, op, realvp, &hp,
+		    cr)))
+			return (error);
+
+		if ((error = hldiraddentry(dir, hp, name))) {
+			/* Unmake the inode we just made. */
+			rw_enter(&hp->hln_rwlock, RW_WRITER);
+			if ((hp->hln_type) == VDIR) {
+				ASSERT(hdp == NULL);
+				/* cleanup allocs made by hyprlofs_dirinit() */
+				hyprlofs_dirtrunc(hp);
+			}
+			mutex_enter(&hp->hln_tlock);
+			hp->hln_nlink = 0;
+			mutex_exit(&hp->hln_tlock);
+			gethrestime(&hp->hln_ctime);
+			rw_exit(&hp->hln_rwlock);
+			hlnode_rele(hp);
+			hp = NULL;
+		} else if (hpp) {
+			*hpp = hp;
+		} else {
+			hlnode_rele(hp);
+		}
+	}
+
+	return (error);
+}
+
+/*
+ * Delete entry hp of name "nm" from dir. Free dir entry space and decrement
+ * link count on hlnode(s).
+ */
+int
+hyprlofs_dirdelete(hlnode_t *dir, hlnode_t *hp, char *nm, enum dr_op op,
+    cred_t *cr)
+{
+	hldirent_t *hpdp;
+	int error;
+	size_t namelen;
+	hlnode_t *hnp;
+	timestruc_t now;
+
+	ASSERT(RW_WRITE_HELD(&dir->hln_rwlock));
+	ASSERT(RW_WRITE_HELD(&hp->hln_rwlock));
+	ASSERT(dir->hln_type == VDIR);
+
+	if (nm[0] == '\0')
+		panic("hyprlofs_dirdelete: NULL name for %p", (void *)hp);
+
+	/* return error if removing . or .. */
+	if (nm[0] == '.') {
+		if (nm[1] == '\0')
+			return (EINVAL);
+		if (nm[1] == '.' && nm[2] == '\0')
+			return (EEXIST); /* thus in ufs */
+	}
+
+	if ((error = hyprlofs_taccess(dir, VEXEC|VWRITE, cr)) != 0)
+		return (error);
+
+	if (dir->hln_dir == NULL)
+		return (ENOENT);
+
+	hpdp = hyprlofs_hash_lookup(nm, dir, 0, &hnp);
+	if (hpdp == NULL) {
+		/*
+		 * If it is gone, some other thread got here first!
+		 * Return error ENOENT.
+		 */
+		return (ENOENT);
+	}
+
+	/*
+	 * If the hlnode in the hldirent changed (shouldn't happen since we
+	 * don't support rename) then original is gone, so return that status
+	 * (same as UFS).
+	 */
+	if (hp != hnp)
+		return (ENOENT);
+
+	hyprlofs_hash_out(hpdp);
+
+	/* Take hpdp out of the directory list. */
+	ASSERT(hpdp->hld_next != hpdp);
+	ASSERT(hpdp->hld_prev != hpdp);
+	if (hpdp->hld_prev) {
+		hpdp->hld_prev->hld_next = hpdp->hld_next;
+	}
+	if (hpdp->hld_next) {
+		hpdp->hld_next->hld_prev = hpdp->hld_prev;
+	}
+
+	/*
+	 * If the roving slot pointer happens to match hpdp, point it at the
+	 * previous dirent.
+	 */
+	if (dir->hln_dir->hld_prev == hpdp) {
+		dir->hln_dir->hld_prev = hpdp->hld_prev;
+	}
+	ASSERT(hpdp->hld_next != hpdp);
+	ASSERT(hpdp->hld_prev != hpdp);
+
+	/* hpdp points to the correct directory entry */
+	namelen = strlen(hpdp->hld_name) + 1;
+
+	kmem_free(hpdp, sizeof (hldirent_t) + namelen);
+	dir->hln_size -= (sizeof (hldirent_t) + namelen);
+	dir->hln_dirents--;
+
+	gethrestime(&now);
+	dir->hln_mtime = now;
+	dir->hln_ctime = now;
+	hp->hln_ctime = now;
+
+	ASSERT(hp->hln_nlink > 0);
+	DECR_COUNT(&hp->hln_nlink, &hp->hln_tlock);
+	if (op == DR_RMDIR && hp->hln_type == VDIR) {
+		hyprlofs_dirtrunc(hp);
+		ASSERT(hp->hln_nlink == 0);
+	}
+	return (0);
+}
+
+/*
+ * hyprlofs_dirinit initializes a dir with '.' and '..' entries without
+ * checking perms and locking
+ */
+void
+hyprlofs_dirinit(
+	hlnode_t *parent,	/* parent of directory to initialize */
+	hlnode_t *dir)		/* the new directory */
+{
+	hldirent_t *dot, *dotdot;
+	timestruc_t now;
+
+	ASSERT(RW_WRITE_HELD(&parent->hln_rwlock));
+	ASSERT(dir->hln_type == VDIR);
+
+	dot = kmem_zalloc(sizeof (hldirent_t) + 2, KM_SLEEP);
+	dotdot = kmem_zalloc(sizeof (hldirent_t) + 3, KM_SLEEP);
+
+	/* Initialize the entries */
+	dot->hld_hlnode = dir;
+	dot->hld_offset = 0;
+	dot->hld_name = (char *)dot + sizeof (hldirent_t);
+	dot->hld_name[0] = '.';
+	dot->hld_parent = dir;
+	hyprlofs_hash_in(dot);
+
+	dotdot->hld_hlnode = parent;
+	dotdot->hld_offset = 1;
+	dotdot->hld_name = (char *)dotdot + sizeof (hldirent_t);
+	dotdot->hld_name[0] = '.';
+	dotdot->hld_name[1] = '.';
+	dotdot->hld_parent = dir;
+	hyprlofs_hash_in(dotdot);
+
+	/* Initialize directory entry list. */
+	dot->hld_next = dotdot;
+	dot->hld_prev = dotdot;
+	dotdot->hld_next = NULL;
+	dotdot->hld_prev = dot;
+
+	gethrestime(&now);
+	dir->hln_mtime = now;
+	dir->hln_ctime = now;
+
+	/*
+	 * Since hyprlofs_dirinit is called with both dir and parent being the
+	 * same for the root vnode, we need to increment this before we set
+	 * hln_nlink = 2 below.
+	 */
+	INCR_COUNT(&parent->hln_nlink, &parent->hln_tlock);
+	parent->hln_ctime = now;
+
+	dir->hln_dir = dot;
+	dir->hln_size = 2 * sizeof (hldirent_t) + 5; /* dot and dotdot */
+	dir->hln_dirents = 2;
+	dir->hln_nlink = 2;
+}
+
+
+/*
+ * hyprlofs_dirtrunc removes all dir entries under this dir.
+ */
+void
+hyprlofs_dirtrunc(hlnode_t *dir)
+{
+	hldirent_t *hdp;
+	hlnode_t *tp;
+	size_t namelen;
+	timestruc_t now;
+
+	ASSERT(RW_WRITE_HELD(&dir->hln_rwlock));
+	ASSERT(dir->hln_type == VDIR);
+
+	if (dir->hln_looped)
+		return;
+
+	for (hdp = dir->hln_dir; hdp; hdp = dir->hln_dir) {
+		ASSERT(hdp->hld_next != hdp);
+		ASSERT(hdp->hld_prev != hdp);
+		ASSERT(hdp->hld_hlnode);
+
+		dir->hln_dir = hdp->hld_next;
+		namelen = strlen(hdp->hld_name) + 1;
+
+		/*
+		 * Adjust the link counts to account for this dir entry removal.
+		 */
+		tp = hdp->hld_hlnode;
+
+		ASSERT(tp->hln_nlink > 0);
+		DECR_COUNT(&tp->hln_nlink, &tp->hln_tlock);
+
+		hyprlofs_hash_out(hdp);
+
+		kmem_free(hdp, sizeof (hldirent_t) + namelen);
+		dir->hln_size -= (sizeof (hldirent_t) + namelen);
+		dir->hln_dirents--;
+	}
+
+	gethrestime(&now);
+	dir->hln_mtime = now;
+	dir->hln_ctime = now;
+
+	ASSERT(dir->hln_dir == NULL);
+	ASSERT(dir->hln_size == 0);
+	ASSERT(dir->hln_dirents == 0);
+}
+
+static int
+hldiraddentry(
+    hlnode_t	*dir,	/* target directory to make entry in */
+    hlnode_t	*hp,	/* new hlnode */
+    char	*name)
+{
+	hldirent_t	*hdp, *hpdp;
+	size_t		namelen, alloc_size;
+	timestruc_t	now;
+
+	/*
+	 * Make sure the parent dir wasn't removed from underneath the caller.
+	 */
+	if (dir->hln_dir == NULL)
+		return (ENOENT);
+
+	/* Check that everything is on the same FS. */
+	if (hp->hln_vnode->v_vfsp != dir->hln_vnode->v_vfsp)
+		return (EXDEV);
+
+	/* Alloc and init dir entry */
+	namelen = strlen(name) + 1;
+	alloc_size = namelen + sizeof (hldirent_t);
+	hdp = kmem_zalloc(alloc_size, KM_NORMALPRI | KM_NOSLEEP);
+	if (hdp == NULL)
+		return (ENOSPC);
+
+	dir->hln_size += alloc_size;
+	dir->hln_dirents++;
+	hdp->hld_hlnode = hp;
+	hdp->hld_parent = dir;
+
+	/* The dir entry and its name were allocated sequentially. */
+	hdp->hld_name = (char *)hdp + sizeof (hldirent_t);
+	(void) strcpy(hdp->hld_name, name);
+
+	hyprlofs_hash_in(hdp);
+
+	/*
+	 * Some utilities expect the size of a directory to remain fairly
+	 * static.  For example, a routine which unlinks files between calls to
+	 * readdir(); the size of the dir changes from underneath it and so the
+	 * real dir offset in bytes is invalid.  To circumvent this problem, we
+	 * initialize a dir entry with a phony offset, and use this offset to
+	 * determine end of file in hyprlofs_readdir.
+	 */
+	hpdp = dir->hln_dir->hld_prev;
+	/*
+	 * Install at first empty "slot" in directory list.
+	 */
+	while (hpdp->hld_next != NULL && (hpdp->hld_next->hld_offset -
+	    hpdp->hld_offset) <= 1) {
+		ASSERT(hpdp->hld_next != hpdp);
+		ASSERT(hpdp->hld_prev != hpdp);
+		ASSERT(hpdp->hld_next->hld_offset > hpdp->hld_offset);
+		hpdp = hpdp->hld_next;
+	}
+	hdp->hld_offset = hpdp->hld_offset + 1;
+
+	/*
+	 * If we're at the end of the dirent list and the offset (which is
+	 * necessarily the largest offset in this dir) is more than twice the
+	 * number of dirents, that means the dir is 50% holes.  At this point
+	 * we reset the slot pointer back to the beginning of the dir so we
+	 * start using the holes. The idea is that if there are N dirents,
+	 * there must also be N holes, so we can satisfy the next N creates by
+	 * walking at most 2N entries; thus the average cost of a create is
+	 * constant. Note that we use the first dirent's hld_prev as the roving
+	 * slot pointer. This saves a word in every dirent.
+	 */
+	if (hpdp->hld_next == NULL && hpdp->hld_offset > 2 * dir->hln_dirents)
+		dir->hln_dir->hld_prev = dir->hln_dir->hld_next;
+	else
+		dir->hln_dir->hld_prev = hdp;
+
+	ASSERT(hpdp->hld_next != hpdp);
+	ASSERT(hpdp->hld_prev != hpdp);
+
+	hdp->hld_next = hpdp->hld_next;
+	if (hdp->hld_next) {
+		hdp->hld_next->hld_prev = hdp;
+	}
+	hdp->hld_prev = hpdp;
+	hpdp->hld_next = hdp;
+
+	ASSERT(hdp->hld_next != hdp);
+	ASSERT(hdp->hld_prev != hdp);
+	ASSERT(hpdp->hld_next != hpdp);
+	ASSERT(hpdp->hld_prev != hpdp);
+
+	gethrestime(&now);
+	dir->hln_mtime = now;
+	dir->hln_ctime = now;
+
+	return (0);
+}
+
+static int
+hldir_make_hlnode(hlnode_t *dir, hlfsmount_t *hm, vattr_t *va, enum de_op op,
+    vnode_t *realvp, hlnode_t **newnode, cred_t *cr)
+{
+	hlnode_t	*hp;
+	enum vtype	type;
+
+	ASSERT(va != NULL);
+	ASSERT(op == DE_CREATE || op == DE_MKDIR);
+	if (((va->va_mask & AT_ATIME) && TIMESPEC_OVERFLOW(&va->va_atime)) ||
+	    ((va->va_mask & AT_MTIME) && TIMESPEC_OVERFLOW(&va->va_mtime)))
+		return (EOVERFLOW);
+	type = va->va_type;
+	hp = kmem_zalloc(sizeof (hlnode_t), KM_SLEEP);
+	hyprlofs_node_init(hm, hp, va, cr);
+
+	hp->hln_vnode->v_rdev = hp->hln_rdev = NODEV;
+	hp->hln_vnode->v_type = type;
+	hp->hln_uid = crgetuid(cr);
+
+	/*
+	 * To determine the gid of the created file:
+	 *   If the directory's set-gid bit is set, set the gid to the gid
+	 *   of the parent dir, otherwise, use the process's gid.
+	 */
+	if (dir->hln_mode & VSGID)
+		hp->hln_gid = dir->hln_gid;
+	else
+		hp->hln_gid = crgetgid(cr);
+
+	/*
+	 * If we're creating a dir and the parent dir has the set-GID bit set,
+	 * set it on the new dir. Otherwise, if the user is neither privileged
+	 * nor a member of the file's new group, clear the file's set-GID bit.
+	 */
+	if (dir->hln_mode & VSGID && type == VDIR)
+		hp->hln_mode |= VSGID;
+	else {
+		if ((hp->hln_mode & VSGID) &&
+		    secpolicy_vnode_setids_setgids(cr, hp->hln_gid) != 0)
+			hp->hln_mode &= ~VSGID;
+	}
+
+	if (va->va_mask & AT_ATIME)
+		hp->hln_atime = va->va_atime;
+	if (va->va_mask & AT_MTIME)
+		hp->hln_mtime = va->va_mtime;
+
+	if (op == DE_MKDIR) {
+		hyprlofs_dirinit(dir, hp);
+		hp->hln_looped = 0;
+	} else {
+		hp->hln_realvp = realvp;
+		hp->hln_size = va->va_size;
+		hp->hln_looped = 1;
+	}
+
+	*newnode = hp;
+	return (0);
+}
diff --git a/usr/src/uts/common/fs/hyprlofs/hyprlofs_subr.c b/usr/src/uts/common/fs/hyprlofs/hyprlofs_subr.c
new file mode 100644
index 0000000000..1d857309f3
--- /dev/null
+++ b/usr/src/uts/common/fs/hyprlofs/hyprlofs_subr.c
@@ -0,0 +1,127 @@
+/*
+ * CDDL HEADER START
+ *
+ * The contents of this file are subject to the terms of the
+ * Common Development and Distribution License (the "License").
+ * You may not use this file except in compliance with the License.
+ *
+ * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
+ * or http://www.opensolaris.org/os/licensing.
+ * See the License for the specific language governing permissions
+ * and limitations under the License.
+ *
+ * When distributing Covered Code, include this CDDL HEADER in each
+ * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
+ * If applicable, add the following below this CDDL HEADER, with the
+ * fields enclosed by brackets "[]" replaced with your own identifying
+ * information: Portions Copyright [yyyy] [name of copyright owner]
+ *
+ * CDDL HEADER END
+ */
+/*
+ * Copyright (c) 2012, Joyent, Inc. All rights reserved.
+ */
+
+#include <sys/types.h>
+#include <sys/errno.h>
+#include <sys/param.h>
+#include <sys/t_lock.h>
+#include <sys/systm.h>
+#include <sys/sysmacros.h>
+#include <sys/debug.h>
+#include <sys/time.h>
+#include <sys/cmn_err.h>
+#include <sys/vnode.h>
+#include <sys/stat.h>
+#include <sys/mode.h>
+#include <sys/vfs.h>
+#include <sys/cred.h>
+#include <sys/kmem.h>
+#include <sys/atomic.h>
+#include <sys/policy.h>
+#include <sys/fs/hyprlofs_info.h>
+
+#define	MODESHIFT	3
+
+/* Initialize a hlnode and add it to file list under mount point. */
+void
+hyprlofs_node_init(hlfsmount_t *hm, hlnode_t *h, vattr_t *vap, cred_t *cr)
+{
+	vnode_t *vp;
+	timestruc_t now;
+
+	ASSERT(vap != NULL);
+
+	rw_init(&h->hln_rwlock, NULL, RW_DEFAULT, NULL);
+	mutex_init(&h->hln_tlock, NULL, MUTEX_DEFAULT, NULL);
+	h->hln_mode = MAKEIMODE(vap->va_type, vap->va_mode);
+	h->hln_mask = 0;
+	h->hln_type = vap->va_type;
+	h->hln_nodeid = (ino64_t)(uint32_t)((uintptr_t)h >> 3);
+	h->hln_nlink = 1;
+	h->hln_size = 0;
+
+	if (cr == NULL) {
+		h->hln_uid = vap->va_uid;
+		h->hln_gid = vap->va_gid;
+	} else {
+		h->hln_uid = crgetuid(cr);
+		h->hln_gid = crgetgid(cr);
+	}
+
+	h->hln_fsid = hm->hlm_dev;
+	h->hln_rdev = vap->va_rdev;
+	h->hln_blksize = PAGESIZE;
+	h->hln_nblocks = 0;
+	gethrestime(&now);
+	h->hln_atime = now;
+	h->hln_mtime = now;
+	h->hln_ctime = now;
+	h->hln_seq = 0;
+	h->hln_dir = NULL;
+
+	h->hln_vnode = vn_alloc(KM_SLEEP);
+	vp = HLNTOV(h);
+	vn_setops(vp, hyprlofs_vnodeops);
+	vp->v_vfsp = hm->hlm_vfsp;
+	vp->v_type = vap->va_type;
+	vp->v_rdev = vap->va_rdev;
+	vp->v_data = (caddr_t)h;
+	mutex_enter(&hm->hlm_contents);
+	/*
+	 * Increment the pseudo generation number for this hlnode. Since
+	 * hlnodes are allocated and freed, there really is no particular
+	 * generation number for a new hlnode.  Just fake it by using a
+	 * counter in each file system.
+	 */
+	h->hln_gen = hm->hlm_gen++;
+
+	/*
+	 * Add new hlnode to end of linked list of hlnodes for this hyprlofs
+	 * Root dir is handled specially in hyprlofs_mount.
+	 */
+	if (hm->hlm_rootnode != (hlnode_t *)NULL) {
+		h->hln_forw = NULL;
+		h->hln_back = hm->hlm_rootnode->hln_back;
+		h->hln_back->hln_forw = hm->hlm_rootnode->hln_back = h;
+	}
+	mutex_exit(&hm->hlm_contents);
+	vn_exists(vp);
+}
+
+int
+hyprlofs_taccess(void *vtp, int mode, cred_t *cr)
+{
+	hlnode_t *hp = vtp;
+	int shift = 0;
+
+	/* Check access based on owner, group and public perms in hlnode. */
+	if (crgetuid(cr) != hp->hln_uid) {
+		shift += MODESHIFT;
+		if (groupmember(hp->hln_gid, cr) == 0)
+			shift += MODESHIFT;
+	}
+
+	return (secpolicy_vnode_access2(cr, HLNTOV(hp), hp->hln_uid,
+	    hp->hln_mode << shift, mode));
+}
diff --git a/usr/src/uts/common/fs/hyprlofs/hyprlofs_vfsops.c b/usr/src/uts/common/fs/hyprlofs/hyprlofs_vfsops.c
new file mode 100644
index 0000000000..c582a8cac2
--- /dev/null
+++ b/usr/src/uts/common/fs/hyprlofs/hyprlofs_vfsops.c
@@ -0,0 +1,614 @@
+/*
+ * CDDL HEADER START
+ *
+ * The contents of this file are subject to the terms of the
+ * Common Development and Distribution License (the "License").
+ * You may not use this file except in compliance with the License.
+ *
+ * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
+ * or http://www.opensolaris.org/os/licensing.
+ * See the License for the specific language governing permissions
+ * and limitations under the License.
+ *
+ * When distributing Covered Code, include this CDDL HEADER in each
+ * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
+ * If applicable, add the following below this CDDL HEADER, with the
+ * fields enclosed by brackets "[]" replaced with your own identifying
+ * information: Portions Copyright [yyyy] [name of copyright owner]
+ *
+ * CDDL HEADER END
+ */
+/*
+ * Copyright (c) 2012, Joyent, Inc. All rights reserved.
+ */
+
+/*
+ * Hyperlofs is a hybrid file system combining features of the tmpfs(7FS) and
+ * lofs(7FS) file systems.  It is modeled on code from both of these file
+ * systems.
+ *
+ * The purpose is to create a high performance name space for files on which
+ * applications will compute.  Given a large number of data files with various
+ * owners, we want to construct a view onto those files such that only a subset
+ * is visible to the applications and such that the view can be changed very
+ * quickly as compute progresses.  Entries in the name space are not mounts and
+ * thus do not appear in the mnttab.  Entries in the name space are allowed to
+ * refer to files on different backing file systems.  Intermediate directories
+ * in the name space exist only in-memory, ala tmpfs.  There are no leaf nodes
+ * in the name space except for entries that refer to backing files ala lofs.
+ *
+ * The name space is managed via ioctls issued on the mounted file system and
+ * is mostly read-only for the compute applications.  That is, applications
+ * cannot create new files in the name space. If a file is unlinked by an
+ * application, that only removes the file from the name space, the backing
+ * file remains in place.  It is possible for applications to write-through to
+ * the backing files if the file system is mounted read-write.
+ *
+ * The name space is managed via the HYPRLOFS_ADD_ENTRIES, HYPRLOFS_RM_ENTRIES,
+ * and HYPRLOFS_RM_ALL ioctls on the top-level mount.
+ *
+ * The HYPRLOFS_ADD_ENTRIES ioctl specifies path(s) to the backing file(s) and
+ * the name(s) for the file(s) in the name space.  The name(s) may be path(s)
+ * which will be relative to the root of the mount and thus cannot begin with
+ * a /. If the name is a path, it does not have to correspond to any backing
+ * path. The intermediate directories will only exist in the name space. The
+ * entry(ies) will be added to the name space.
+ *
+ * The HYPRLOFS_RM_ENTRIES ioctl specifies the name(s) of the file(s) in the
+ * name space which should be removed.  The name(s) may be path(s) which will
+ * be relative to the root of the mount and thus cannot begin with a /.  The
+ * named entry(ies) will be removed.
+ *
+ * The HYPRLOFS_RM_ALL ioctl will remove all mappings from the name space.
+ */
+
+#include <sys/types.h>
+#include <sys/param.h>
+#include <sys/sysmacros.h>
+#include <sys/kmem.h>
+#include <sys/time.h>
+#include <sys/pathname.h>
+#include <sys/vfs.h>
+#include <sys/vfs_opreg.h>
+#include <sys/vnode.h>
+#include <sys/stat.h>
+#include <sys/uio.h>
+#include <sys/stat.h>
+#include <sys/errno.h>
+#include <sys/cmn_err.h>
+#include <sys/cred.h>
+#include <sys/statvfs.h>
+#include <sys/mount.h>
+#include <sys/debug.h>
+#include <sys/systm.h>
+#include <sys/mntent.h>
+#include <fs/fs_subr.h>
+#include <vm/page.h>
+#include <vm/anon.h>
+#include <sys/model.h>
+#include <sys/policy.h>
+
+#include <sys/fs/swapnode.h>
+#include <sys/fs/hyprlofs_info.h>
+
+static int hyprlofsfstype;
+
+/*
+ * hyprlofs vfs operations.
+ */
+static int hyprlofsinit(int, char *);
+static int hyprlofs_mount(vfs_t *, vnode_t *, struct mounta *, cred_t *);
+static int hyprlofs_unmount(vfs_t *, int, cred_t *);
+static int hyprlofs_root(vfs_t *, vnode_t **);
+static int hyprlofs_statvfs(vfs_t *, struct statvfs64 *);
+static int hyprlofs_vget(vfs_t *, vnode_t **, struct fid *);
+
+/*
+ * Loadable module wrapper
+ */
+#include <sys/modctl.h>
+
+static mntopts_t hyprlofs_mntopts;
+
+static vfsdef_t vfw = {
+	VFSDEF_VERSION,
+	"hyprlofs",
+	hyprlofsinit,
+	VSW_HASPROTO|VSW_CANREMOUNT|VSW_STATS|VSW_ZMOUNT,
+	&hyprlofs_mntopts
+};
+
+static mntopts_t hyprlofs_mntopts = {
+	0, NULL
+};
+
+/*
+ * Module linkage information
+ */
+static struct modlfs modlfs = {
+	&mod_fsops, "filesystem for hyprlofs", &vfw
+};
+
+static struct modlinkage modlinkage = {
+	MODREV_1, &modlfs, NULL
+};
+
+int
+_init()
+{
+	return (mod_install(&modlinkage));
+}
+
+int
+_fini()
+{
+	int error;
+
+	error = mod_remove(&modlinkage);
+	if (error)
+		return (error);
+	/*
+	 * Tear down the operations vectors
+	 */
+	(void) vfs_freevfsops_by_type(hyprlofsfstype);
+	vn_freevnodeops(hyprlofs_vnodeops);
+	return (0);
+}
+
+int
+_info(struct modinfo *modinfop)
+{
+	return (mod_info(&modlinkage, modinfop));
+}
+
+/*
+ * The following are patchable variables limiting the amount of system
+ * resources hyprlofs can use.
+ *
+ * hyprlofs_maxkmem limits the amount of kernel kmem_alloc memory hyprlofs can
+ * use for it's data structures (e.g. hlnodes, directory entries). It is set
+ * as a percentage of physical memory which is determined when hyprlofs is
+ * first used in the system.
+ *
+ * hyprlofs_minfree is the minimum amount of swap space that hyprlofs leaves for
+ * the rest of the system. If the amount of free swap space in the system
+ * (i.e. anoninfo.ani_free) drops below hyprlofs_minfree, hyprlofs anon
+ * allocations will fail.
+ */
+size_t hyprlofs_maxkmem = 0;
+size_t hyprlofs_minfree = 0;
+size_t hyprlofs_kmemspace;	/* bytes of kernel heap used by all hyprlofs */
+
+static major_t hyprlofs_major;
+static minor_t hyprlofs_minor;
+static kmutex_t	hyprlofs_minor_lock;
+
+/*
+ * initialize global hyprlofs locks and hashes when loading hyprlofs module
+ */
+static int
+hyprlofsinit(int fstype, char *name)
+{
+	static const fs_operation_def_t hl_vfsops_template[] = {
+		VFSNAME_MOUNT,		{ .vfs_mount = hyprlofs_mount },
+		VFSNAME_UNMOUNT,	{ .vfs_unmount = hyprlofs_unmount },
+		VFSNAME_ROOT,		{ .vfs_root = hyprlofs_root },
+		VFSNAME_STATVFS,	{ .vfs_statvfs = hyprlofs_statvfs },
+		VFSNAME_VGET,		{ .vfs_vget = hyprlofs_vget },
+		NULL,			NULL
+	};
+	int error;
+	extern  void    hyprlofs_hash_init();
+
+	hyprlofs_hash_init();
+	hyprlofsfstype = fstype;
+	ASSERT(hyprlofsfstype != 0);
+
+	error = vfs_setfsops(fstype, hl_vfsops_template, NULL);
+	if (error != 0) {
+		cmn_err(CE_WARN, "hyprlofsinit: bad vfs ops template");
+		return (error);
+	}
+
+	error = vn_make_ops(name, hyprlofs_vnodeops_template,
+	    &hyprlofs_vnodeops);
+	if (error != 0) {
+		(void) vfs_freevfsops_by_type(fstype);
+		cmn_err(CE_WARN, "hyprlofsinit: bad vnode ops template");
+		return (error);
+	}
+
+	/*
+	 * hyprlofs_minfree is an absolute limit of swap space which still
+	 * allows other processes to execute.  Set it if its not patched.
+	 */
+	if (hyprlofs_minfree == 0)
+		hyprlofs_minfree = btopr(HYPRLOFSMINFREE);
+
+	if ((hyprlofs_major = getudev()) == (major_t)-1) {
+		cmn_err(CE_WARN,
+		    "hyprlofsinit: Can't get unique device number.");
+		hyprlofs_major = 0;
+	}
+	mutex_init(&hyprlofs_minor_lock, NULL, MUTEX_DEFAULT, NULL);
+	return (0);
+}
+
+static int
+hyprlofs_mount(vfs_t *vfsp, vnode_t *mvp, struct mounta *uap, cred_t *cr)
+{
+	hlfsmount_t *hm = NULL;
+	hlnode_t *hp;
+	struct pathname dpn;
+	int error;
+	vattr_t rattr;
+	int got_attrs;
+
+	if ((error = secpolicy_fs_mount(cr, mvp, vfsp)) != 0)
+		return (error);
+	if (secpolicy_hyprlofs_control(cr) != 0)
+		return (EPERM);
+
+	if (mvp->v_type != VDIR)
+		return (ENOTDIR);
+
+	if (uap->flags & MS_REMOUNT)
+		return (EBUSY);
+
+	mutex_enter(&mvp->v_lock);
+	if ((uap->flags & MS_OVERLAY) == 0 &&
+	    (mvp->v_count != 1 || (mvp->v_flag & VROOT))) {
+		mutex_exit(&mvp->v_lock);
+		return (EBUSY);
+	}
+	mutex_exit(&mvp->v_lock);
+
+	/* Having the resource be anything but "swap" doesn't make sense. */
+	vfs_setresource(vfsp, "swap", 0);
+
+	if ((error = pn_get(uap->dir,
+	    (uap->flags & MS_SYSSPACE) ? UIO_SYSSPACE : UIO_USERSPACE,
+	    &dpn)) != 0)
+		goto out;
+
+	if ((hm = kmem_zalloc(sizeof (hlfsmount_t),
+	    KM_NORMALPRI | KM_NOSLEEP)) == NULL) {
+		pn_free(&dpn);
+		error = ENOMEM;
+		goto out;
+	}
+
+	/* Get an available minor device number for this mount */
+	mutex_enter(&hyprlofs_minor_lock);
+	do {
+		hyprlofs_minor = (hyprlofs_minor + 1) & L_MAXMIN32;
+		hm->hlm_dev = makedevice(hyprlofs_major, hyprlofs_minor);
+	} while (vfs_devismounted(hm->hlm_dev));
+	mutex_exit(&hyprlofs_minor_lock);
+
+	/*
+	 * Set but don't bother entering the mutex since hlfsmount is not on
+	 * the mount list yet.
+	 */
+	mutex_init(&hm->hlm_contents, NULL, MUTEX_DEFAULT, NULL);
+
+	hm->hlm_vfsp = vfsp;
+
+	vfsp->vfs_data = (caddr_t)hm;
+	vfsp->vfs_fstype = hyprlofsfstype;
+	vfsp->vfs_dev = hm->hlm_dev;
+	vfsp->vfs_bsize = PAGESIZE;
+	vfsp->vfs_flag |= VFS_NOTRUNC;
+	vfs_make_fsid(&vfsp->vfs_fsid, hm->hlm_dev, hyprlofsfstype);
+	hm->hlm_mntpath = kmem_zalloc(dpn.pn_pathlen + 1, KM_SLEEP);
+	(void) strcpy(hm->hlm_mntpath, dpn.pn_path);
+
+	/* allocate and initialize root hlnode structure */
+	bzero(&rattr, sizeof (vattr_t));
+	rattr.va_mode = (mode_t)(S_IFDIR | 0777);
+	rattr.va_type = VDIR;
+	rattr.va_rdev = 0;
+	hp = kmem_zalloc(sizeof (hlnode_t), KM_SLEEP);
+	hyprlofs_node_init(hm, hp, &rattr, cr);
+
+	/* Get the mode, uid, and gid from the underlying mount point. */
+	rattr.va_mask = AT_MODE|AT_UID|AT_GID;
+	got_attrs = VOP_GETATTR(mvp, &rattr, 0, cr, NULL);
+
+	rw_enter(&hp->hln_rwlock, RW_WRITER);
+	HLNTOV(hp)->v_flag |= VROOT;
+
+	/*
+	 * If the getattr succeeded, use its results, otherwise allow the
+	 * previously set defaults to prevail.
+	 */
+	if (got_attrs == 0) {
+		hp->hln_mode = rattr.va_mode;
+		hp->hln_uid = rattr.va_uid;
+		hp->hln_gid = rattr.va_gid;
+	}
+
+	/*
+	 * Initialize linked list of hlnodes so that the back pointer of the
+	 * root hlnode always points to the last one on the list and the
+	 * forward pointer of the last node is null
+	 */
+	hp->hln_back = hp;
+	hp->hln_forw = NULL;
+	hp->hln_nlink = 0;
+	hm->hlm_rootnode = hp;
+
+	hyprlofs_dirinit(hp, hp);
+
+	rw_exit(&hp->hln_rwlock);
+
+	pn_free(&dpn);
+	error = 0;
+
+out:
+	return (error);
+}
+
+static int
+hyprlofs_unmount(vfs_t *vfsp, int flag, cred_t *cr)
+{
+	hlfsmount_t *hm = (hlfsmount_t *)VFSTOHLM(vfsp);
+	hlnode_t *hnp, *cancel;
+	vnode_t	*vp;
+	int error;
+
+	if ((error = secpolicy_fs_unmount(cr, vfsp)) != 0)
+		return (error);
+	if (secpolicy_hyprlofs_control(cr) != 0)
+		return (EPERM);
+
+	/*
+	 * forced unmount is not supported by this file system
+	 * and thus, ENOTSUP, is being returned.
+	 */
+	if (flag & MS_FORCE)
+		return (ENOTSUP);
+
+	mutex_enter(&hm->hlm_contents);
+
+	/*
+	 * If there are no open files, only the root node should have a ref cnt.
+	 * With hlm_contents held, nothing can be added or removed. There may
+	 * be some dirty pages.  To prevent fsflush from disrupting the unmount,
+	 * put a hold on each node while scanning. If we find a previously
+	 * referenced node, undo the holds we have placed and fail EBUSY.
+	 */
+	hnp = hm->hlm_rootnode;
+	if (HLNTOV(hnp)->v_count > 1) {
+		mutex_exit(&hm->hlm_contents);
+		return (EBUSY);
+	}
+
+	for (hnp = hnp->hln_forw; hnp; hnp = hnp->hln_forw) {
+		if ((vp = HLNTOV(hnp))->v_count > 0) {
+			cancel = hm->hlm_rootnode->hln_forw;
+			while (cancel != hnp) {
+				vp = HLNTOV(cancel);
+				ASSERT(vp->v_count > 0);
+				VN_RELE(vp);
+				cancel = cancel->hln_forw;
+			}
+			mutex_exit(&hm->hlm_contents);
+			return (EBUSY);
+		}
+		VN_HOLD(vp);
+	}
+
+	/* We can drop the mutex now because no one can find this mount */
+	mutex_exit(&hm->hlm_contents);
+
+	/*
+	 * Free all alloc'd memory associated with this FS. To do this, we go
+	 * through the file list twice, once to remove all the dir entries, and
+	 * then to remove all the files.
+	 */
+
+	/* Remove all directory entries */
+	for (hnp = hm->hlm_rootnode; hnp; hnp = hnp->hln_forw) {
+		rw_enter(&hnp->hln_rwlock, RW_WRITER);
+		if (hnp->hln_type == VDIR)
+			hyprlofs_dirtrunc(hnp);
+		rw_exit(&hnp->hln_rwlock);
+	}
+
+	ASSERT(hm->hlm_rootnode);
+
+	/*
+	 * All links are gone, v_count is keeping nodes in place. VN_RELE
+	 * should make the node disappear, unless somebody is holding pages
+	 * against it.  Wait and retry until it disappears.
+	 *
+	 * We re-acquire the lock to prevent others who have a HOLD on a hlnode
+	 * from blowing it away (in hyprlofs_inactive) while we're trying to
+	 * get to it here. Once we have a HOLD on it we know it'll stick around.
+	 */
+	mutex_enter(&hm->hlm_contents);
+
+	/* Remove all the files (except the rootnode) backwards. */
+	while ((hnp = hm->hlm_rootnode->hln_back) != hm->hlm_rootnode) {
+		mutex_exit(&hm->hlm_contents);
+		/* Note we handled the link count in pass 2 above. */
+		vp = HLNTOV(hnp);
+		VN_RELE(vp);
+		mutex_enter(&hm->hlm_contents);
+		/*
+		 * It's still there after the RELE. Someone else like pageout
+		 * has a hold on it so wait a bit and then try again.
+		 */
+		if (hnp == hm->hlm_rootnode->hln_back) {
+			VN_HOLD(vp);
+			mutex_exit(&hm->hlm_contents);
+			delay(hz / 4);
+			mutex_enter(&hm->hlm_contents);
+		}
+	}
+	mutex_exit(&hm->hlm_contents);
+
+	VN_RELE(HLNTOV(hm->hlm_rootnode));
+
+	ASSERT(hm->hlm_mntpath);
+
+	kmem_free(hm->hlm_mntpath, strlen(hm->hlm_mntpath) + 1);
+
+	mutex_destroy(&hm->hlm_contents);
+	kmem_free(hm, sizeof (hlfsmount_t));
+
+	return (0);
+}
+
+/* Return root hlnode for given vnode */
+static int
+hyprlofs_root(vfs_t *vfsp, vnode_t **vpp)
+{
+	hlfsmount_t *hm = (hlfsmount_t *)VFSTOHLM(vfsp);
+	hlnode_t *hp = hm->hlm_rootnode;
+	vnode_t *vp;
+
+	ASSERT(hp);
+
+	vp = HLNTOV(hp);
+	VN_HOLD(vp);
+	*vpp = vp;
+	return (0);
+}
+
+static int
+hyprlofs_statvfs(vfs_t *vfsp, struct statvfs64 *sbp)
+{
+	hlfsmount_t *hm = (hlfsmount_t *)VFSTOHLM(vfsp);
+	ulong_t	blocks;
+	dev32_t d32;
+	zoneid_t eff_zid;
+	struct zone *zp;
+
+	/*
+	 * The FS may have been mounted by the GZ on behalf of the NGZ.  In
+	 * that case, the hlfsmount zone_id will be the global zone.  We want
+	 * to show the swap cap inside the zone in this case, even though the
+	 * FS was mounted by the GZ.
+	 */
+	if (curproc->p_zone->zone_id != GLOBAL_ZONEUNIQID)
+		zp = curproc->p_zone;
+	else
+		zp = hm->hlm_vfsp->vfs_zone;
+
+	if (zp == NULL)
+		eff_zid = GLOBAL_ZONEUNIQID;
+	else
+		eff_zid = zp->zone_id;
+
+	sbp->f_bsize = PAGESIZE;
+	sbp->f_frsize = PAGESIZE;
+
+	/*
+	 * Find the amount of available physical and memory swap
+	 */
+	mutex_enter(&anoninfo_lock);
+	ASSERT(k_anoninfo.ani_max >= k_anoninfo.ani_phys_resv);
+	blocks = (ulong_t)CURRENT_TOTAL_AVAILABLE_SWAP;
+	mutex_exit(&anoninfo_lock);
+
+	if (blocks > hyprlofs_minfree)
+		sbp->f_bfree = blocks - hyprlofs_minfree;
+	else
+		sbp->f_bfree = 0;
+
+	sbp->f_bavail = sbp->f_bfree;
+
+	/*
+	 * Total number of blocks is what's available plus what's been used
+	 */
+	sbp->f_blocks = (fsblkcnt64_t)(sbp->f_bfree);
+
+	if (eff_zid != GLOBAL_ZONEUNIQID &&
+	    zp->zone_max_swap_ctl != UINT64_MAX) {
+		/*
+		 * If the fs is used by a NGZ with a swap cap, then report the
+		 * capped size.
+		 */
+		rctl_qty_t cap, used;
+		pgcnt_t pgcap, pgused;
+
+		mutex_enter(&zp->zone_mem_lock);
+		cap = zp->zone_max_swap_ctl;
+		used = zp->zone_max_swap;
+		mutex_exit(&zp->zone_mem_lock);
+
+		pgcap = btop(cap);
+		pgused = btop(used);
+
+		sbp->f_bfree = MIN(pgcap - pgused, sbp->f_bfree);
+		sbp->f_bavail = sbp->f_bfree;
+		sbp->f_blocks = MIN(pgcap, sbp->f_blocks);
+	}
+
+	/*
+	 * This is fairly inaccurate since it doesn't take into account the
+	 * names stored in the directory entries.
+	 */
+	sbp->f_ffree = sbp->f_files = ptob(availrmem) /
+	    (sizeof (hlnode_t) + sizeof (hldirent_t));
+
+	sbp->f_favail = (fsfilcnt64_t)(sbp->f_ffree);
+	(void) cmpldev(&d32, vfsp->vfs_dev);
+	sbp->f_fsid = d32;
+	(void) strcpy(sbp->f_basetype, vfssw[hyprlofsfstype].vsw_name);
+	(void) strncpy(sbp->f_fstr, hm->hlm_mntpath, sizeof (sbp->f_fstr));
+	/*
+	 * ensure null termination
+	 */
+	sbp->f_fstr[sizeof (sbp->f_fstr) - 1] = '\0';
+	sbp->f_flag = vf_to_stf(vfsp->vfs_flag);
+	sbp->f_namemax = MAXNAMELEN - 1;
+	return (0);
+}
+
+static int
+hyprlofs_vget(vfs_t *vfsp, vnode_t **vpp, struct fid *fidp)
+{
+	hlfid_t *hfid;
+	hlfsmount_t *hm = (hlfsmount_t *)VFSTOHLM(vfsp);
+	hlnode_t *hp = NULL;
+
+	hfid = (hlfid_t *)fidp;
+	*vpp = NULL;
+
+	mutex_enter(&hm->hlm_contents);
+	for (hp = hm->hlm_rootnode; hp; hp = hp->hln_forw) {
+		mutex_enter(&hp->hln_tlock);
+		if (hp->hln_nodeid == hfid->hlfid_ino) {
+			/*
+			 * If the gen numbers don't match we know the file
+			 * won't be found since only one hlnode can have this
+			 * number at a time.
+			 */
+			if (hp->hln_gen != hfid->hlfid_gen ||
+			    hp->hln_nlink == 0) {
+				mutex_exit(&hp->hln_tlock);
+				mutex_exit(&hm->hlm_contents);
+				return (0);
+			}
+			*vpp = (vnode_t *)HLNTOV(hp);
+
+			VN_HOLD(*vpp);
+
+			if ((hp->hln_mode & S_ISVTX) &&
+			    !(hp->hln_mode & (S_IXUSR | S_IFDIR))) {
+				mutex_enter(&(*vpp)->v_lock);
+				(*vpp)->v_flag |= VISSWAP;
+				mutex_exit(&(*vpp)->v_lock);
+			}
+			mutex_exit(&hp->hln_tlock);
+			mutex_exit(&hm->hlm_contents);
+			return (0);
+		}
+		mutex_exit(&hp->hln_tlock);
+	}
+	mutex_exit(&hm->hlm_contents);
+	return (0);
+}
diff --git a/usr/src/uts/common/fs/hyprlofs/hyprlofs_vnops.c b/usr/src/uts/common/fs/hyprlofs/hyprlofs_vnops.c
new file mode 100644
index 0000000000..a2064dfa1f
--- /dev/null
+++ b/usr/src/uts/common/fs/hyprlofs/hyprlofs_vnops.c
@@ -0,0 +1,1441 @@
+/*
+ * CDDL HEADER START
+ *
+ * The contents of this file are subject to the terms of the
+ * Common Development and Distribution License (the "License").
+ * You may not use this file except in compliance with the License.
+ *
+ * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
+ * or http://www.opensolaris.org/os/licensing.
+ * See the License for the specific language governing permissions
+ * and limitations under the License.
+ *
+ * When distributing Covered Code, include this CDDL HEADER in each
+ * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
+ * If applicable, add the following below this CDDL HEADER, with the
+ * fields enclosed by brackets "[]" replaced with your own identifying
+ * information: Portions Copyright [yyyy] [name of copyright owner]
+ *
+ * CDDL HEADER END
+ */
+
+/*
+ * Copyright 2015 Joyent, Inc.  All rights reserved.
+ */
+
+#include <sys/types.h>
+#include <sys/param.h>
+#include <sys/t_lock.h>
+#include <sys/systm.h>
+#include <sys/sysmacros.h>
+#include <sys/user.h>
+#include <sys/time.h>
+#include <sys/vfs.h>
+#include <sys/vfs_opreg.h>
+#include <sys/vnode.h>
+#include <sys/file.h>
+#include <sys/fcntl.h>
+#include <sys/flock.h>
+#include <sys/kmem.h>
+#include <sys/errno.h>
+#include <sys/stat.h>
+#include <sys/cred.h>
+#include <sys/dirent.h>
+#include <sys/pathname.h>
+#include <sys/fs/hyprlofs.h>
+#include <sys/fs/hyprlofs_info.h>
+#include <sys/mman.h>
+#include <vm/pvn.h>
+#include <sys/cmn_err.h>
+#include <sys/buf.h>
+#include <sys/policy.h>
+#include <fs/fs_subr.h>
+#include <sys/ddi.h>
+#include <sys/sunddi.h>
+
+static int hyprlofs_add_entry(vnode_t *, char *, char *, cred_t *,
+		caller_context_t *);
+static int hyprlofs_rm_entry(vnode_t *, char *, cred_t *, caller_context_t *,
+		int);
+static int hyprlofs_rm_all(vnode_t *, cred_t *, caller_context_t *, int);
+static int hyprlofs_remove(vnode_t *, char *, cred_t *, caller_context_t *,
+		int);
+static int hyprlofs_get_all(vnode_t *, intptr_t, cred_t *, caller_context_t *,
+		int);
+
+/*
+ * This is a somewhat arbitrary upper limit on the number of entries we can
+ * pass in on a single add/rm ioctl call.  This is only used to validate that
+ * the input list looks sane.
+ */
+#define	MAX_IOCTL_PARAMS	100000
+
+static int
+hyprlofs_open(vnode_t **vpp, int flag, cred_t *cr, caller_context_t *ct)
+{
+	vnode_t *rvp;
+	int error;
+
+	rvp = REALVP(*vpp);
+
+	if (VTOHLN(*vpp)->hln_looped == 0)
+		return (0);
+
+	/*
+	 * looped back, pass through to real vnode. Need to hold new reference
+	 * to vp since VOP_OPEN() may decide to release it.
+	 */
+	VN_HOLD(rvp);
+	error = VOP_OPEN(&rvp, flag, cr, ct);
+	ASSERT(rvp->v_count > 1);
+	VN_RELE(rvp);
+
+	return (error);
+}
+
+static int
+hyprlofs_close(vnode_t *vp, int flag, int count, offset_t offset, cred_t *cr,
+    caller_context_t *ct)
+{
+	if (VTOHLN(vp)->hln_looped == 0) {
+		cleanlocks(vp, ttoproc(curthread)->p_pid, 0);
+		cleanshares(vp, ttoproc(curthread)->p_pid);
+		return (0);
+	}
+
+	return (VOP_CLOSE(REALVP(vp), flag, count, offset, cr, ct));
+}
+
+static int
+hyprlofs_read(vnode_t *vp, struct uio *uiop, int ioflag, cred_t *cr,
+    caller_context_t *ct)
+{
+	if (vp->v_type == VDIR)
+		return (EISDIR);
+	return (VOP_READ(REALVP(vp), uiop, ioflag, cr, ct));
+}
+
+static int
+hyprlofs_write(vnode_t *vp, struct uio *uiop, int ioflag, cred_t *cr,
+    caller_context_t *ct)
+{
+	/* We don't support writing to non-regular files */
+	if (vp->v_type != VREG)
+		return (EINVAL);
+
+	if (vn_is_readonly(vp))
+		return (EROFS);
+
+	return (VOP_WRITE(REALVP(vp), uiop, ioflag, cr, ct));
+}
+
+/* ARGSUSED */
+static int
+hyprlofs_ioctl(vnode_t *vp, int cmd, intptr_t data, int flag,
+    cred_t *cr, int *rvalp, caller_context_t *ct)
+{
+	int len, cnt, error;
+	int i;
+	model_t model;
+	char path[MAXPATHLEN];
+	char nm[MAXPATHLEN];
+
+	/* We only support the hyprlofs ioctls on the root vnode */
+	if (!(vp->v_flag & VROOT))
+		return (ENOTTY);
+
+	/*
+	 * Check if managing hyprlofs is allowed.
+	 */
+	if (secpolicy_hyprlofs_control(cr) != 0)
+		return (EPERM);
+
+	if (cmd == HYPRLOFS_ADD_ENTRIES || cmd == HYPRLOFS_RM_ENTRIES) {
+		model = get_udatamodel();
+
+		if (model == DATAMODEL_NATIVE) {
+			hyprlofs_entries_t ebuf;
+			hyprlofs_entry_t *e;
+
+			if (copyin((void *)data, &ebuf, sizeof (ebuf)))
+				return (EFAULT);
+			cnt = ebuf.hle_len;
+			if (cnt > MAX_IOCTL_PARAMS)
+				return (EINVAL);
+			len = sizeof (hyprlofs_entry_t) * cnt;
+
+			e = kmem_alloc(len, KM_SLEEP);
+			if (copyin((void *)(ebuf.hle_entries), e, len)) {
+				kmem_free(e, len);
+				return (EFAULT);
+			}
+
+			for (i = 0; i < cnt; i++) {
+				if (e[i].hle_nlen == 0 ||
+				    e[i].hle_nlen > MAXPATHLEN)
+					return (EINVAL);
+
+				if (copyin(e[i].hle_name, nm, e[i].hle_nlen)
+				    != 0) {
+					kmem_free(e, len);
+					return (EFAULT);
+				}
+				nm[e[i].hle_nlen] = '\0';
+
+				if (cmd == HYPRLOFS_ADD_ENTRIES) {
+					if (e[i].hle_plen == 0 ||
+					    e[i].hle_plen > MAXPATHLEN)
+						return (EINVAL);
+
+					if (copyin(e[i].hle_path, path,
+					    e[i].hle_plen) != 0) {
+						kmem_free(e, len);
+						return (EFAULT);
+					}
+					path[e[i].hle_plen] = '\0';
+
+					if ((error = hyprlofs_add_entry(vp,
+					    path, nm, cr, ct)) != 0) {
+						kmem_free(e, len);
+						return (error);
+					}
+				} else {
+					if ((error = hyprlofs_rm_entry(vp, nm,
+					    cr, ct, flag)) != 0) {
+						kmem_free(e, len);
+						return (error);
+					}
+				}
+			}
+
+			kmem_free(e, len);
+			return (0);
+
+		} else {
+			hyprlofs_entries32_t ebuf32;
+			hyprlofs_entry32_t *e32;
+
+			if (copyin((void *)data, &ebuf32, sizeof (ebuf32)))
+				return (EFAULT);
+
+			cnt = ebuf32.hle_len;
+			if (cnt > MAX_IOCTL_PARAMS)
+				return (EINVAL);
+			len = sizeof (hyprlofs_entry32_t) * cnt;
+
+			e32 = kmem_alloc(len, KM_SLEEP);
+			if (copyin((void *)(unsigned long)(ebuf32.hle_entries),
+			    e32, len)) {
+				kmem_free(e32, len);
+				return (EFAULT);
+			}
+
+			for (i = 0; i < cnt; i++) {
+				if (e32[i].hle_nlen == 0 ||
+				    e32[i].hle_nlen > MAXPATHLEN)
+					return (EINVAL);
+
+				if (copyin((void *)(unsigned long)
+				    e32[i].hle_name, nm,
+				    e32[i].hle_nlen) != 0) {
+					kmem_free(e32, len);
+					return (EFAULT);
+				}
+				nm[e32[i].hle_nlen] = '\0';
+
+				if (cmd == HYPRLOFS_ADD_ENTRIES) {
+					if (e32[i].hle_plen == 0 ||
+					    e32[i].hle_plen > MAXPATHLEN)
+						return (EINVAL);
+
+					if (copyin((void *)(unsigned long)
+					    e32[i].hle_path, path,
+					    e32[i].hle_plen) != 0) {
+						kmem_free(e32, len);
+						return (EFAULT);
+					}
+					path[e32[i].hle_plen] = '\0';
+
+					if ((error = hyprlofs_add_entry(vp,
+					    path, nm, cr, ct)) != 0) {
+						kmem_free(e32, len);
+						return (error);
+					}
+				} else {
+					if ((error = hyprlofs_rm_entry(vp, nm,
+					    cr, ct, flag)) != 0) {
+						kmem_free(e32, len);
+						return (error);
+					}
+				}
+			}
+
+			kmem_free(e32, len);
+			return (0);
+		}
+	}
+
+	if (cmd == HYPRLOFS_RM_ALL) {
+		return (hyprlofs_rm_all(vp, cr, ct, flag));
+	}
+
+	if (cmd == HYPRLOFS_GET_ENTRIES) {
+		return (hyprlofs_get_all(vp, data, cr, ct, flag));
+	}
+
+	return (ENOTTY);
+}
+
+static int
+hyprlofs_getattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr,
+    caller_context_t *ct)
+{
+	hlnode_t *tp = (hlnode_t *)VTOHLN(vp);
+	vattr_t tmp_va;
+
+	if (tp->hln_looped == 1) {
+		int error;
+
+		if ((error = VOP_GETATTR(REALVP(vp), &tmp_va, flags, cr,
+		    ct)) != 0)
+			return (error);
+	}
+
+	mutex_enter(&tp->hln_tlock);
+	vap->va_type = vp->v_type;
+	vap->va_mode = tp->hln_mode & MODEMASK;
+	vap->va_uid = tp->hln_uid;
+	vap->va_gid = tp->hln_gid;
+	vap->va_fsid = tp->hln_fsid;
+	vap->va_nodeid = (ino64_t)tp->hln_nodeid;
+	vap->va_nlink = tp->hln_nlink;
+	vap->va_size = (u_offset_t)tp->hln_size;
+	vap->va_atime = tp->hln_atime;
+	vap->va_mtime = tp->hln_mtime;
+	vap->va_ctime = tp->hln_ctime;
+	vap->va_blksize = PAGESIZE;
+	vap->va_rdev = tp->hln_rdev;
+	vap->va_seq = tp->hln_seq;
+
+	if (tp->hln_looped == 1) {
+		vap->va_nblocks = tmp_va.va_nblocks;
+	} else {
+		vap->va_nblocks =
+		    (fsblkcnt64_t)btodb(ptob(btopr(vap->va_size)));
+	}
+	mutex_exit(&tp->hln_tlock);
+	return (0);
+}
+
+/*ARGSUSED4*/
+static int
+hyprlofs_setattr(vnode_t *vp, vattr_t *vap, int flags,
+    cred_t *cr, caller_context_t *ct)
+{
+	hlnode_t *tp = (hlnode_t *)VTOHLN(vp);
+	int error = 0;
+	vattr_t *get;
+	long mask;
+
+	/*
+	 * Cannot set these attributes
+	 */
+	if ((vap->va_mask & AT_NOSET) || (vap->va_mask & AT_XVATTR))
+		return (EINVAL);
+
+	mutex_enter(&tp->hln_tlock);
+
+	get = &tp->hln_attr;
+	/*
+	 * Change file access modes. Must be owner or have sufficient
+	 * privileges.
+	 */
+	error = secpolicy_vnode_setattr(cr, vp, vap, get, flags,
+	    hyprlofs_taccess, tp);
+
+	if (error)
+		goto out;
+
+	mask = vap->va_mask;
+
+	if (mask & AT_MODE) {
+		get->va_mode &= S_IFMT;
+		get->va_mode |= vap->va_mode & ~S_IFMT;
+	}
+
+	if (mask & AT_UID)
+		get->va_uid = vap->va_uid;
+	if (mask & AT_GID)
+		get->va_gid = vap->va_gid;
+	if (mask & AT_ATIME)
+		get->va_atime = vap->va_atime;
+	if (mask & AT_MTIME)
+		get->va_mtime = vap->va_mtime;
+
+	if (mask & (AT_UID | AT_GID | AT_MODE | AT_MTIME))
+		gethrestime(&tp->hln_ctime);
+
+out:
+	mutex_exit(&tp->hln_tlock);
+	return (error);
+}
+
+static int
+hyprlofs_access(vnode_t *vp, int mode, int flags, cred_t *cr,
+    caller_context_t *ct)
+{
+	hlnode_t *tp = (hlnode_t *)VTOHLN(vp);
+	int error;
+
+	if (mode & VWRITE) {
+		if (vp->v_type == VREG && vn_is_readonly(vp))
+			return (EROFS);
+	}
+	if (VTOHLN(vp)->hln_looped == 1)
+		return (VOP_ACCESS(REALVP(vp), mode, flags, cr, ct));
+
+	mutex_enter(&tp->hln_tlock);
+	error = hyprlofs_taccess(tp, mode, cr);
+	mutex_exit(&tp->hln_tlock);
+	return (error);
+}
+
+/* ARGSUSED3 */
+static int
+hyprlofs_lookup(vnode_t *dvp, char *nm, vnode_t **vpp, struct pathname *pnp,
+    int flags, vnode_t *rdir, cred_t *cr, caller_context_t *ct,
+    int *direntflags, pathname_t *realpnp)
+{
+	hlnode_t *tp = (hlnode_t *)VTOHLN(dvp);
+	hlnode_t *ntp = NULL;
+	int error;
+
+	if (VTOHLN(dvp)->hln_looped == 1)
+		return (VOP_LOOKUP(REALVP(dvp), nm, vpp, pnp, flags, rdir,
+		    cr, ct, direntflags, realpnp));
+
+	if (flags & LOOKUP_XATTR)
+		return (EINVAL);
+
+	/* Null component name is a synonym for directory being searched. */
+	if (*nm == '\0') {
+		VN_HOLD(dvp);
+		*vpp = dvp;
+		return (0);
+	}
+	ASSERT(tp);
+
+	if ((error = hyprlofs_dirlookup(tp, nm, &ntp, cr)) == 0) {
+		ASSERT(ntp);
+		*vpp = HLNTOV(ntp);
+	}
+	return (error);
+}
+
+/*
+ * Create the loopback from the hyprlofs vnode to the real vnode.
+ */
+static int
+hyprlofs_loopback(vnode_t *dvp, vnode_t *rvp, char *nm, vattr_t *vap,
+    int mode, cred_t *cr, caller_context_t *ct)
+{
+	hlnode_t *parent;
+	hlfsmount_t *tm;
+	int error;
+	hlnode_t *oldtp;
+	vnode_t *vp;
+
+	parent = (hlnode_t *)VTOHLN(dvp);
+	tm = (hlfsmount_t *)VTOHLM(dvp);
+	error = 0;
+	oldtp = NULL;
+
+	if (vap->va_type == VREG && (vap->va_mode & VSVTX)) {
+		/* we don't support the sticky bit */
+		vap->va_mode &= ~VSVTX;
+	} else if (vap->va_type == VNON) {
+		return (EINVAL);
+	}
+
+	/* Null component name is a synonym for directory being searched. */
+	if (*nm == '\0') {
+		VN_HOLD(dvp);
+		oldtp = parent;
+	} else {
+		error = hyprlofs_dirlookup(parent, nm, &oldtp, cr);
+	}
+
+	if (error == 0) {	/* name found */
+		ASSERT(oldtp);
+
+		rw_enter(&oldtp->hln_rwlock, RW_WRITER);
+
+		/*
+		 * if create/read-only an existing directory, allow it
+		 */
+		if ((oldtp->hln_type == VDIR) && (mode & VWRITE))
+			error = EISDIR;
+		else {
+			error = hyprlofs_taccess(oldtp, mode, cr);
+		}
+
+		if (error) {
+			rw_exit(&oldtp->hln_rwlock);
+			hlnode_rele(oldtp);
+			return (error);
+		}
+
+		vp = HLNTOV(oldtp);
+		rw_exit(&oldtp->hln_rwlock);
+
+		if (vp->v_type == VREG) {
+			hlnode_rele(oldtp);
+			return (EEXIST);
+		}
+
+		vnevent_create(vp, ct);
+		return (0);
+	}
+
+	if (error != ENOENT)
+		return (error);
+
+	rw_enter(&parent->hln_rwlock, RW_WRITER);
+	error = hyprlofs_direnter(tm, parent, nm, DE_CREATE, rvp, vap, NULL,
+	    cr);
+	rw_exit(&parent->hln_rwlock);
+
+	return (error);
+}
+
+/*
+ * Create an in-memory directory based on the add-entry ioctl name.
+ * If the dir exists, return EEXIST but still also return node in vpp.
+ */
+static int
+hyprlofs_mkdir(vnode_t *dvp, char *nm, vattr_t *va, vnode_t **vpp, cred_t *cr)
+{
+	hlnode_t *parent = (hlnode_t *)VTOHLN(dvp);
+	hlnode_t *self = NULL;
+	hlfsmount_t *tm = (hlfsmount_t *)VTOHLM(dvp);
+	int error;
+
+	/*
+	 * Might be dangling directory.  Catch it here, because a ENOENT return
+	 * from hyprlofs_dirlookup() is a valid return.
+	 */
+	if (parent->hln_nlink == 0)
+		return (ENOENT);
+
+	error = hyprlofs_dirlookup(parent, nm, &self, cr);
+	if (error == 0) {
+		ASSERT(self);
+		hlnode_rele(self);
+		/* We can't loop in under a looped in directory */
+		if (self->hln_looped)
+			return (EACCES);
+		*vpp = HLNTOV(self);
+		return (EEXIST);
+	}
+	if (error != ENOENT)
+		return (error);
+
+	rw_enter(&parent->hln_rwlock, RW_WRITER);
+	error = hyprlofs_direnter(tm, parent, nm, DE_MKDIR, (vnode_t *)NULL,
+	    va, &self, cr);
+	rw_exit(&parent->hln_rwlock);
+
+	if (error == 0 || error == EEXIST) {
+		hlnode_rele(self);
+		*vpp = HLNTOV(self);
+	}
+
+	return (error);
+}
+
+/*
+ * Loop in a file or directory into the namespace.
+ */
+static int
+hyprlofs_add_entry(vnode_t *vp, char *fspath, char *fsname,
+    cred_t *cr, caller_context_t *ct)
+{
+	int error;
+	char *p, *pnm;
+	vnode_t *realvp, *dvp;
+	vattr_t va;
+
+	/*
+	 * Get vnode for the real file/dir. We'll have a hold on realvp which
+	 * we won't vn_rele until hyprlofs_inactive.
+	 */
+	if ((error = lookupname(fspath, UIO_SYSSPACE, FOLLOW, NULLVPP,
+	    &realvp)) != 0)
+		return (error);
+
+	/* no devices allowed */
+	if (IS_DEVVP(realvp)) {
+		VN_RELE(realvp);
+		return (ENODEV);
+	}
+
+	/*
+	 * realvp may be an AUTOFS node, in which case we perform a VOP_ACCESS
+	 * to trigger the mount of the intended filesystem. This causes a
+	 * loopback mount of the intended filesystem instead of the AUTOFS
+	 * filesystem.
+	 */
+	if ((error = VOP_ACCESS(realvp, 0, 0, cr, NULL)) != 0) {
+		VN_RELE(realvp);
+		return (error);
+	}
+
+	/*
+	 * We're interested in the top most filesystem. This is specially
+	 * important when fspath is a trigger AUTOFS node, since we're really
+	 * interested in mounting the filesystem AUTOFS mounted as result of
+	 * the VOP_ACCESS() call not the AUTOFS node itself.
+	 */
+	if (vn_mountedvfs(realvp) != NULL) {
+		if ((error = traverse(&realvp)) != 0) {
+			VN_RELE(realvp);
+			return (error);
+		}
+	}
+
+	va.va_type = VNON;
+	/*
+	 * If the target name is a path, make sure we have all of the
+	 * intermediate directories, creating them if necessary.
+	 */
+	dvp = vp;
+	pnm = p = fsname;
+
+	/* path cannot be absolute */
+	if (*p == '/') {
+		VN_RELE(realvp);
+		return (EINVAL);
+	}
+
+	for (p = strchr(pnm, '/'); p != NULL; p = strchr(pnm, '/')) {
+		if (va.va_type == VNON)
+			/* use the top-level dir as the template va for mkdir */
+			if ((error = VOP_GETATTR(vp, &va, 0, cr, NULL)) != 0) {
+				VN_RELE(realvp);
+				return (error);
+			}
+
+		*p = '\0';
+
+		/* Path component cannot be empty or relative */
+		if (pnm[0] == '\0' ||
+		    (pnm[0] == '.' && pnm[1] == '.' && pnm[2] == '\0')) {
+			VN_RELE(realvp);
+			return (EINVAL);
+		}
+
+		if ((error = hyprlofs_mkdir(dvp, pnm, &va, &dvp, cr)) != 0 &&
+		    error != EEXIST) {
+			VN_RELE(realvp);
+			return (error);
+		}
+
+		*p = '/';
+		pnm = p + 1;
+	}
+
+	/* The file name is required */
+	if (pnm[0] == '\0') {
+		VN_RELE(realvp);
+		return (EINVAL);
+	}
+
+	/* Now use the real file's va as the template va */
+	if ((error = VOP_GETATTR(realvp, &va, 0, cr, NULL)) != 0) {
+		VN_RELE(realvp);
+		return (error);
+	}
+
+	/* Make the vnode */
+	error = hyprlofs_loopback(dvp, realvp, pnm, &va, va.va_mode, cr, ct);
+	if (error != 0)
+		VN_RELE(realvp);
+	return (error);
+}
+
+/*
+ * Remove a looped in file from the namespace.
+ */
+static int
+hyprlofs_rm_entry(vnode_t *dvp, char *fsname, cred_t *cr, caller_context_t *ct,
+    int flags)
+{
+	int error;
+	char *p, *pnm;
+	hlnode_t *parent;
+	hlnode_t *fndtp;
+
+	pnm = p = fsname;
+
+	/* path cannot be absolute */
+	if (*p == '/')
+		return (EINVAL);
+
+	/*
+	 * If the target name is a path, get the containing dir and simple
+	 * file name.
+	 */
+	parent = (hlnode_t *)VTOHLN(dvp);
+	for (p = strchr(pnm, '/'); p != NULL; p = strchr(pnm, '/')) {
+		*p = '\0';
+
+		/* Path component cannot be empty or relative */
+		if (pnm[0] == '\0' ||
+		    (pnm[0] == '.' && pnm[1] == '.' && pnm[2] == '\0'))
+			return (EINVAL);
+
+		if ((error = hyprlofs_dirlookup(parent, pnm, &fndtp, cr)) != 0)
+			return (error);
+
+		dvp = HLNTOV(fndtp);
+		parent = fndtp;
+		pnm = p + 1;
+	}
+
+	/* The file name is required */
+	if (pnm[0] == '\0')
+		return (EINVAL);
+
+	/* Remove the entry from the parent dir */
+	return (hyprlofs_remove(dvp, pnm, cr, ct, flags));
+}
+
+/*
+ * Remove all looped in files from the namespace.
+ */
+static int
+hyprlofs_rm_all(vnode_t *dvp, cred_t *cr, caller_context_t *ct,
+    int flags)
+{
+	int error = 0;
+	hlnode_t *hp = (hlnode_t *)VTOHLN(dvp);
+	hldirent_t *hdp;
+
+	hlnode_hold(hp);
+
+	/*
+	 * There's a window here where someone could have removed
+	 * all the entries in the directory after we put a hold on the
+	 * vnode but before we grabbed the rwlock.  Just return.
+	 */
+	if (hp->hln_dir == NULL) {
+		if (hp->hln_nlink) {
+			panic("empty directory 0x%p", (void *)hp);
+			/*NOTREACHED*/
+		}
+		goto done;
+	}
+
+	hdp = hp->hln_dir;
+	while (hdp) {
+		hlnode_t *fndhp;
+
+		if (strcmp(hdp->hld_name, ".") == 0 ||
+		    strcmp(hdp->hld_name, "..") == 0) {
+			hdp = hdp->hld_next;
+			continue;
+		}
+
+		/* This holds the fndhp vnode */
+		error = hyprlofs_dirlookup(hp, hdp->hld_name, &fndhp, cr);
+		if (error != 0)
+			goto done;
+		hlnode_rele(fndhp);
+
+		if (fndhp->hln_looped == 0) {
+			/* recursively remove contents of this subdir */
+			if (fndhp->hln_type == VDIR) {
+				vnode_t *tvp = HLNTOV(fndhp);
+
+				error = hyprlofs_rm_all(tvp, cr, ct, flags);
+				if (error != 0)
+					goto done;
+			}
+		}
+
+		/* remove the entry */
+		error = hyprlofs_remove(dvp, hdp->hld_name, cr, ct, flags);
+		if (error != 0)
+			goto done;
+
+		hdp = hp->hln_dir;
+	}
+
+done:
+	hlnode_rele(hp);
+	return (error);
+}
+
+/*
+ * Get a list of all looped in files in the namespace.
+ */
+static int
+hyprlofs_get_all_entries(vnode_t *dvp, hyprlofs_curr_entry_t *hcp,
+    char *prefix, int *pcnt, int n_max,
+    cred_t *cr, caller_context_t *ct, int flags)
+{
+	int error = 0;
+	int too_big = 0;
+	int cnt;
+	int len;
+	hlnode_t *hp = (hlnode_t *)VTOHLN(dvp);
+	hldirent_t *hdp;
+	char *path;
+
+	cnt = *pcnt;
+	path = kmem_alloc(MAXPATHLEN, KM_SLEEP);
+
+	hlnode_hold(hp);
+
+	/*
+	 * There's a window here where someone could have removed
+	 * all the entries in the directory after we put a hold on the
+	 * vnode but before we grabbed the rwlock.  Just return.
+	 */
+	if (hp->hln_dir == NULL) {
+		if (hp->hln_nlink) {
+			panic("empty directory 0x%p", (void *)hp);
+			/*NOTREACHED*/
+		}
+		goto done;
+	}
+
+	hdp = hp->hln_dir;
+	while (hdp) {
+		hlnode_t *fndhp;
+		vnode_t *tvp;
+
+		if (strcmp(hdp->hld_name, ".") == 0 ||
+		    strcmp(hdp->hld_name, "..") == 0) {
+			hdp = hdp->hld_next;
+			continue;
+		}
+
+		/* This holds the fndhp vnode */
+		error = hyprlofs_dirlookup(hp, hdp->hld_name, &fndhp, cr);
+		if (error != 0)
+			goto done;
+		hlnode_rele(fndhp);
+
+		if (fndhp->hln_looped == 0) {
+			/* recursively get contents of this subdir */
+			VERIFY(fndhp->hln_type == VDIR);
+			tvp = HLNTOV(fndhp);
+
+			if (*prefix == '\0')
+				(void) strlcpy(path, hdp->hld_name, MAXPATHLEN);
+			else
+				(void) snprintf(path, MAXPATHLEN, "%s/%s",
+				    prefix, hdp->hld_name);
+
+			error = hyprlofs_get_all_entries(tvp, hcp, path,
+			    &cnt, n_max, cr, ct, flags);
+
+			if (error == E2BIG) {
+				too_big = 1;
+				error = 0;
+			}
+			if (error != 0)
+				goto done;
+		} else {
+			if (cnt < n_max) {
+				char *p;
+
+				if (*prefix == '\0')
+					(void) strlcpy(path, hdp->hld_name,
+					    MAXPATHLEN);
+				else
+					(void) snprintf(path, MAXPATHLEN,
+					    "%s/%s", prefix, hdp->hld_name);
+
+				len = strlen(path);
+				ASSERT(len <= MAXPATHLEN);
+				if (copyout(path, (void *)(hcp[cnt].hce_name),
+				    len)) {
+					error = EFAULT;
+					goto done;
+				}
+
+				tvp = REALVP(HLNTOV(fndhp));
+				if (tvp->v_path == vn_vpath_empty) {
+					p = "<unknown>";
+				} else {
+					p = tvp->v_path;
+				}
+				len = strlen(p);
+				ASSERT(len <= MAXPATHLEN);
+				if (copyout(p, (void *)(hcp[cnt].hce_path),
+				    len)) {
+					error = EFAULT;
+					goto done;
+				}
+			}
+
+			cnt++;
+			if (cnt > n_max)
+				too_big = 1;
+		}
+
+		hdp = hdp->hld_next;
+	}
+
+done:
+	hlnode_rele(hp);
+	kmem_free(path, MAXPATHLEN);
+
+	*pcnt = cnt;
+	if (error == 0 && too_big == 1)
+		error = E2BIG;
+
+	return (error);
+}
+
+/*
+ * Return a list of all looped in files in the namespace.
+ */
+static int
+hyprlofs_get_all(vnode_t *dvp, intptr_t data, cred_t *cr, caller_context_t *ct,
+    int flags)
+{
+	int limit, cnt, error;
+	model_t model;
+	hyprlofs_curr_entry_t *e;
+
+	model = get_udatamodel();
+
+	if (model == DATAMODEL_NATIVE) {
+		hyprlofs_curr_entries_t ebuf;
+
+		if (copyin((void *)data, &ebuf, sizeof (ebuf)))
+			return (EFAULT);
+		limit = ebuf.hce_cnt;
+		e = ebuf.hce_entries;
+		if (limit > MAX_IOCTL_PARAMS)
+			return (EINVAL);
+
+	} else {
+		hyprlofs_curr_entries32_t ebuf32;
+
+		if (copyin((void *)data, &ebuf32, sizeof (ebuf32)))
+			return (EFAULT);
+
+		limit = ebuf32.hce_cnt;
+		e = (hyprlofs_curr_entry_t *)(unsigned long)
+		    (ebuf32.hce_entries);
+		if (limit > MAX_IOCTL_PARAMS)
+			return (EINVAL);
+	}
+
+	cnt = 0;
+	error = hyprlofs_get_all_entries(dvp, e, "", &cnt, limit, cr, ct,
+	    flags);
+
+	if (error == 0 || error == E2BIG) {
+		if (model == DATAMODEL_NATIVE) {
+			hyprlofs_curr_entries_t ebuf;
+
+			ebuf.hce_cnt = cnt;
+			if (copyout(&ebuf, (void *)data, sizeof (ebuf)))
+				return (EFAULT);
+
+		} else {
+			hyprlofs_curr_entries32_t ebuf32;
+
+			ebuf32.hce_cnt = cnt;
+			if (copyout(&ebuf32, (void *)data, sizeof (ebuf32)))
+				return (EFAULT);
+		}
+	}
+
+	return (error);
+}
+
+/* ARGSUSED3 */
+static int
+hyprlofs_remove(vnode_t *dvp, char *nm, cred_t *cr, caller_context_t *ct,
+    int flags)
+{
+	hlnode_t *parent = (hlnode_t *)VTOHLN(dvp);
+	int error;
+	hlnode_t *hp = NULL;
+
+	/* This holds the hp vnode */
+	error = hyprlofs_dirlookup(parent, nm, &hp, cr);
+	if (error)
+		return (error);
+
+	ASSERT(hp);
+	rw_enter(&parent->hln_rwlock, RW_WRITER);
+	rw_enter(&hp->hln_rwlock, RW_WRITER);
+
+	error = hyprlofs_dirdelete(parent, hp, nm, DR_REMOVE, cr);
+
+	rw_exit(&hp->hln_rwlock);
+	rw_exit(&parent->hln_rwlock);
+	vnevent_remove(HLNTOV(hp), dvp, nm, ct);
+
+	/*
+	 * We've now dropped the dir link so by rele-ing our vnode we should
+	 * clean up in hyprlofs_inactive.
+	 */
+	hlnode_rele(hp);
+
+	return (error);
+}
+
+/* ARGSUSED4 */
+static int
+hyprlofs_rmdir(vnode_t *dvp, char *nm, vnode_t *cdir, cred_t *cr,
+    caller_context_t *ct, int flags)
+{
+	hlnode_t *parent = (hlnode_t *)VTOHLN(dvp);
+	hlnode_t *self = NULL;
+	vnode_t *vp;
+	int error = 0;
+
+	/* Return error if removing . or .. */
+	if (strcmp(nm, ".") == 0)
+		return (EINVAL);
+	if (strcmp(nm, "..") == 0)
+		return (EEXIST); /* Should be ENOTEMPTY */
+	error = hyprlofs_dirlookup(parent, nm, &self, cr);
+	if (error)
+		return (error);
+
+	rw_enter(&parent->hln_rwlock, RW_WRITER);
+	rw_enter(&self->hln_rwlock, RW_WRITER);
+
+	vp = HLNTOV(self);
+	if (vp == dvp || vp == cdir) {
+		error = EINVAL;
+		goto done1;
+	}
+	if (self->hln_type != VDIR) {
+		error = ENOTDIR;
+		goto done1;
+	}
+
+	/*
+	 * When a dir is looped in, we only remove the in-memory dir, not the
+	 * backing dir.
+	 */
+	if (self->hln_looped == 0) {
+		mutex_enter(&self->hln_tlock);
+		if (self->hln_nlink > 2) {
+			mutex_exit(&self->hln_tlock);
+			error = EEXIST;
+			goto done1;
+		}
+		mutex_exit(&self->hln_tlock);
+
+		if (vn_vfswlock(vp)) {
+			error = EBUSY;
+			goto done1;
+		}
+		if (vn_mountedvfs(vp) != NULL) {
+			error = EBUSY;
+			goto done;
+		}
+
+		/*
+		 * Check for an empty directory, i.e. only includes entries for
+		 * "." and ".."
+		 */
+		if (self->hln_dirents > 2) {
+			error = EEXIST;		/* SIGH should be ENOTEMPTY */
+			/*
+			 * Update atime because checking hln_dirents is
+			 * equivalent to reading the directory
+			 */
+			gethrestime(&self->hln_atime);
+			goto done;
+		}
+
+		error = hyprlofs_dirdelete(parent, self, nm, DR_RMDIR, cr);
+	} else {
+		error = hyprlofs_dirdelete(parent, self, nm, DR_REMOVE, cr);
+	}
+
+done:
+	if (self->hln_looped == 0)
+		vn_vfsunlock(vp);
+done1:
+	rw_exit(&self->hln_rwlock);
+	rw_exit(&parent->hln_rwlock);
+	vnevent_rmdir(HLNTOV(self), dvp, nm, ct);
+
+	/*
+	 * We've now dropped the dir link so by rele-ing our vnode we should
+	 * clean up in hyprlofs_inactive.
+	 */
+	hlnode_rele(self);
+
+	return (error);
+}
+
+static int
+hyprlofs_readdir(vnode_t *vp, struct uio *uiop, cred_t *cr, int *eofp,
+    caller_context_t *ct, int flags)
+{
+	hlnode_t *hp = (hlnode_t *)VTOHLN(vp);
+	hldirent_t *hdp;
+	int error = 0;
+	size_t namelen;
+	struct dirent64 *dp;
+	ulong_t offset;
+	ulong_t total_bytes_wanted;
+	long outcount = 0;
+	long bufsize;
+	int reclen;
+	caddr_t outbuf;
+
+	if (VTOHLN(vp)->hln_looped == 1)
+		return (VOP_READDIR(REALVP(vp), uiop, cr, eofp, ct, flags));
+
+	if (uiop->uio_loffset >= MAXOFF_T) {
+		if (eofp)
+			*eofp = 1;
+		return (0);
+	}
+	/* assuming syscall has already called hln_rwlock */
+	ASSERT(RW_READ_HELD(&hp->hln_rwlock));
+
+	if (uiop->uio_iovcnt != 1)
+		return (EINVAL);
+
+	if (vp->v_type != VDIR)
+		return (ENOTDIR);
+
+	/*
+	 * There's a window here where someone could have removed
+	 * all the entries in the directory after we put a hold on the
+	 * vnode but before we grabbed the rwlock.  Just return.
+	 */
+	if (hp->hln_dir == NULL) {
+		if (hp->hln_nlink) {
+			panic("empty directory 0x%p", (void *)hp);
+			/*NOTREACHED*/
+		}
+		return (0);
+	}
+
+	/* Get space for multiple dir entries */
+	total_bytes_wanted = uiop->uio_iov->iov_len;
+	bufsize = total_bytes_wanted + sizeof (struct dirent64);
+	outbuf = kmem_alloc(bufsize, KM_SLEEP);
+
+	dp = (struct dirent64 *)((uintptr_t)outbuf);
+
+	offset = 0;
+	hdp = hp->hln_dir;
+	while (hdp) {
+		namelen = strlen(hdp->hld_name);	/* no +1 needed */
+		offset = hdp->hld_offset;
+		if (offset >= uiop->uio_offset) {
+			reclen = (int)DIRENT64_RECLEN(namelen);
+			if (outcount + reclen > total_bytes_wanted) {
+				if (!outcount)
+					/* Buffer too small for any entries. */
+					error = EINVAL;
+				break;
+			}
+			ASSERT(hdp->hld_hlnode != NULL);
+
+			/* zero out uninitialized bytes */
+			(void) strncpy(dp->d_name, hdp->hld_name,
+			    DIRENT64_NAMELEN(reclen));
+			dp->d_reclen = (ushort_t)reclen;
+			dp->d_ino = (ino64_t)hdp->hld_hlnode->hln_nodeid;
+			dp->d_off = (offset_t)hdp->hld_offset + 1;
+			dp = (struct dirent64 *)
+			    ((uintptr_t)dp + dp->d_reclen);
+			outcount += reclen;
+			ASSERT(outcount <= bufsize);
+		}
+		hdp = hdp->hld_next;
+	}
+
+	if (!error)
+		error = uiomove(outbuf, outcount, UIO_READ, uiop);
+
+	if (!error) {
+		/*
+		 * If we reached the end of the list our offset should now be
+		 * just past the end.
+		 */
+		if (!hdp) {
+			offset += 1;
+			if (eofp)
+				*eofp = 1;
+		} else if (eofp)
+			*eofp = 0;
+		uiop->uio_offset = offset;
+	}
+	gethrestime(&hp->hln_atime);
+	kmem_free(outbuf, bufsize);
+	return (error);
+}
+
+static int
+hyprlofs_fsync(vnode_t *vp, int syncflag, cred_t *cr, caller_context_t *ct)
+{
+	if (VTOHLN(vp)->hln_looped == 1)
+		return (VOP_FSYNC(REALVP(vp), syncflag, cr, ct));
+	return (0);
+}
+
+/* ARGSUSED */
+static void
+hyprlofs_inactive(vnode_t *vp, cred_t *cr, caller_context_t *ct)
+{
+	hlnode_t *hp = (hlnode_t *)VTOHLN(vp);
+	hlfsmount_t *hm = (hlfsmount_t *)VFSTOHLM(vp->v_vfsp);
+
+	rw_enter(&hp->hln_rwlock, RW_WRITER);
+
+	mutex_enter(&hp->hln_tlock);
+	mutex_enter(&vp->v_lock);
+	ASSERT(vp->v_count >= 1);
+
+	/*
+	 * If we don't have the last hold or the link count is non-zero,
+	 * there's nothing to do except drop our hold.
+	 */
+	if (vp->v_count > 1 || hp->hln_nlink != 0) {
+		vp->v_count--;
+		mutex_exit(&vp->v_lock);
+		mutex_exit(&hp->hln_tlock);
+		rw_exit(&hp->hln_rwlock);
+		return;
+	}
+
+	mutex_exit(&vp->v_lock);
+	mutex_exit(&hp->hln_tlock);
+
+	/* release hold on the real vnode now */
+	if (hp->hln_looped == 1 && hp->hln_realvp != NULL)
+		VN_RELE(hp->hln_realvp);
+
+	/* Here's our chance to send invalid event while we're between locks */
+	vn_invalid(HLNTOV(hp));
+
+	mutex_enter(&hm->hlm_contents);
+	if (hp->hln_forw == NULL)
+		hm->hlm_rootnode->hln_back = hp->hln_back;
+	else
+		hp->hln_forw->hln_back = hp->hln_back;
+	hp->hln_back->hln_forw = hp->hln_forw;
+	mutex_exit(&hm->hlm_contents);
+	rw_exit(&hp->hln_rwlock);
+	rw_destroy(&hp->hln_rwlock);
+	mutex_destroy(&hp->hln_tlock);
+	vn_free(HLNTOV(hp));
+	kmem_free(hp, sizeof (hlnode_t));
+}
+
+static int
+hyprlofs_fid(vnode_t *vp, struct fid *fidp, caller_context_t *ct)
+{
+	hlnode_t *hp = (hlnode_t *)VTOHLN(vp);
+	hlfid_t *hfid;
+
+	if (VTOHLN(vp)->hln_looped == 1)
+		return (VOP_FID(REALVP(vp), fidp, ct));
+
+	if (fidp->fid_len < (sizeof (hlfid_t) - sizeof (ushort_t))) {
+		fidp->fid_len = sizeof (hlfid_t) - sizeof (ushort_t);
+		return (ENOSPC);
+	}
+
+	hfid = (hlfid_t *)fidp;
+	bzero(hfid, sizeof (hlfid_t));
+	hfid->hlfid_len = (int)sizeof (hlfid_t) - sizeof (ushort_t);
+
+	hfid->hlfid_ino = hp->hln_nodeid;
+	hfid->hlfid_gen = hp->hln_gen;
+
+	return (0);
+}
+
+static int
+hyprlofs_getpage(vnode_t *vp, offset_t off, size_t len, uint_t *protp,
+    page_t *pl[], size_t plsz, struct seg *seg, caddr_t addr, enum seg_rw rw,
+    cred_t *cr, caller_context_t *ct)
+{
+	/* return EACCES to be consistent with mmap */
+	if (VTOHLN(vp)->hln_looped != 1)
+		return (EACCES);
+	return (VOP_GETPAGE(REALVP(vp), off, len, protp, pl, plsz, seg, addr,
+	    rw, cr, ct));
+}
+
+int
+hyprlofs_putpage(vnode_t *vp, offset_t off, size_t len, int flags,
+    cred_t *cr, caller_context_t *ct)
+{
+	/* return EACCES to be consistent with mmap */
+	if (VTOHLN(vp)->hln_looped != 1)
+		return (EACCES);
+	return (VOP_PUTPAGE(REALVP(vp), off, len, flags, cr, ct));
+}
+
+static int
+hyprlofs_map(vnode_t *vp, offset_t off, struct as *as, caddr_t *addrp,
+    size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr,
+    caller_context_t *ct)
+{
+	/* return EACCES to be consistent with mmap */
+	if (VTOHLN(vp)->hln_looped != 1)
+		return (EACCES);
+	return (VOP_MAP(REALVP(vp), off, as, addrp, len, prot, maxprot, flags,
+	    cr, ct));
+}
+
+static int
+hyprlofs_addmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr,
+    size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr,
+    caller_context_t *ct)
+{
+	/* return EACCES to be consistent with mmap */
+	if (VTOHLN(vp)->hln_looped != 1)
+		return (EACCES);
+	return (VOP_ADDMAP(REALVP(vp), off, as, addr, len, prot, maxprot,
+	    flags, cr, ct));
+}
+
+static int
+hyprlofs_delmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr,
+    size_t len, uint_t prot, uint_t maxprot, uint_t flags, cred_t *cr,
+    caller_context_t *ct)
+{
+	/* return EACCES to be consistent with mmap */
+	if (VTOHLN(vp)->hln_looped != 1)
+		return (EACCES);
+	return (VOP_DELMAP(REALVP(vp), off, as, addr, len, prot, maxprot,
+	    flags, cr, ct));
+}
+
+static int
+hyprlofs_space(vnode_t *vp, int cmd, struct flock64 *bfp, int flag,
+    offset_t offset, cred_t *cr, caller_context_t *ct)
+{
+	/* return EACCES to be consistent with mmap */
+	if (VTOHLN(vp)->hln_looped != 1)
+		return (EACCES);
+	return (VOP_SPACE(REALVP(vp), cmd, bfp, flag, offset, cr, ct));
+}
+
+static int
+hyprlofs_seek(vnode_t *vp, offset_t ooff, offset_t *noffp,
+    caller_context_t *ct)
+{
+	if (VTOHLN(vp)->hln_looped == 0)
+		return ((*noffp < 0 || *noffp > MAXOFFSET_T) ? EINVAL : 0);
+
+	return (VOP_SEEK(REALVP(vp), ooff, noffp, ct));
+}
+
+static int
+hyprlofs_rwlock(vnode_t *vp, int write_lock, caller_context_t *ct)
+{
+	hlnode_t *hp = VTOHLN(vp);
+
+	if (hp->hln_looped == 1)
+		return (VOP_RWLOCK(REALVP(vp), write_lock, ct));
+
+	if (write_lock) {
+		rw_enter(&hp->hln_rwlock, RW_WRITER);
+	} else {
+		rw_enter(&hp->hln_rwlock, RW_READER);
+	}
+	return (write_lock);
+}
+
+static void
+hyprlofs_rwunlock(vnode_t *vp, int write_lock, caller_context_t *ct)
+{
+	hlnode_t *hp = VTOHLN(vp);
+
+	if (hp->hln_looped == 1) {
+		VOP_RWUNLOCK(REALVP(vp), write_lock, ct);
+		return;
+	}
+
+	rw_exit(&hp->hln_rwlock);
+}
+
+static int
+hyprlofs_pathconf(vnode_t *vp, int cmd, ulong_t *valp, cred_t *cr,
+    caller_context_t *ct)
+{
+	int error;
+
+	if (VTOHLN(vp)->hln_looped == 1)
+		return (VOP_PATHCONF(REALVP(vp), cmd, valp, cr, ct));
+
+	switch (cmd) {
+	case _PC_XATTR_ENABLED:
+	case _PC_XATTR_EXISTS:
+	case _PC_SATTR_ENABLED:
+	case _PC_SATTR_EXISTS:
+		error = EINVAL;
+		break;
+	case _PC_TIMESTAMP_RESOLUTION:
+		/* nanosecond timestamp resolution */
+		*valp = 1L;
+		error = 0;
+		break;
+	default:
+		error = fs_pathconf(vp, cmd, valp, cr, ct);
+	}
+	return (error);
+}
+
+
+struct vnodeops *hyprlofs_vnodeops;
+
+const fs_operation_def_t hyprlofs_vnodeops_template[] = {
+	VOPNAME_OPEN,		{ .vop_open = hyprlofs_open },
+	VOPNAME_CLOSE,		{ .vop_close = hyprlofs_close },
+	VOPNAME_READ,		{ .vop_read = hyprlofs_read },
+	VOPNAME_WRITE,		{ .vop_write = hyprlofs_write },
+	VOPNAME_IOCTL,		{ .vop_ioctl = hyprlofs_ioctl },
+	VOPNAME_GETATTR,	{ .vop_getattr = hyprlofs_getattr },
+	VOPNAME_SETATTR,	{ .vop_setattr = hyprlofs_setattr },
+	VOPNAME_ACCESS,		{ .vop_access = hyprlofs_access },
+	VOPNAME_LOOKUP,		{ .vop_lookup = hyprlofs_lookup },
+	VOPNAME_CREATE,		{ .error = fs_error },
+	VOPNAME_REMOVE,		{ .vop_remove = hyprlofs_remove },
+	VOPNAME_LINK,		{ .error = fs_error },
+	VOPNAME_RENAME,		{ .error = fs_error },
+	VOPNAME_MKDIR,		{ .error = fs_error },
+	VOPNAME_RMDIR,		{ .vop_rmdir = hyprlofs_rmdir },
+	VOPNAME_READDIR,	{ .vop_readdir = hyprlofs_readdir },
+	VOPNAME_SYMLINK,	{ .error = fs_error },
+	VOPNAME_READLINK,	{ .error = fs_error },
+	VOPNAME_FSYNC,		{ .vop_fsync = hyprlofs_fsync },
+	VOPNAME_INACTIVE,	{ .vop_inactive = hyprlofs_inactive },
+	VOPNAME_FID,		{ .vop_fid = hyprlofs_fid },
+	VOPNAME_RWLOCK,		{ .vop_rwlock = hyprlofs_rwlock },
+	VOPNAME_RWUNLOCK,	{ .vop_rwunlock = hyprlofs_rwunlock },
+	VOPNAME_SEEK,		{ .vop_seek = hyprlofs_seek },
+	VOPNAME_SPACE,		{ .vop_space = hyprlofs_space },
+	VOPNAME_GETPAGE,	{ .vop_getpage = hyprlofs_getpage },
+	VOPNAME_PUTPAGE,	{ .vop_putpage = hyprlofs_putpage },
+	VOPNAME_MAP,		{ .vop_map = hyprlofs_map },
+	VOPNAME_ADDMAP,		{ .vop_addmap = hyprlofs_addmap },
+	VOPNAME_DELMAP,		{ .vop_delmap = hyprlofs_delmap },
+	VOPNAME_PATHCONF,	{ .vop_pathconf = hyprlofs_pathconf },
+	VOPNAME_VNEVENT,	{ .vop_vnevent = fs_vnevent_support },
+	NULL,			NULL
+};
diff --git a/usr/src/uts/common/fs/lookup.c b/usr/src/uts/common/fs/lookup.c
index 55ffb94805..59ec5d1829 100644
--- a/usr/src/uts/common/fs/lookup.c
+++ b/usr/src/uts/common/fs/lookup.c
@@ -21,6 +21,7 @@
 
 /*
  * Copyright 2015 Nexenta Systems, Inc.  All rights reserved.
+ * Copyright (c) 2015, Joyent, Inc. All rights reserved.
  * Copyright (c) 1988, 2010, Oracle and/or its affiliates. All rights reserved.
  */
 
@@ -57,6 +58,7 @@
 #include <sys/zone.h>
 #include <sys/dnlc.h>
 #include <sys/fs/snode.h>
+#include <sys/brand.h>
 
 /* Controls whether paths are stored with vnodes. */
 int vfs_vnode_path = 1;
@@ -977,6 +979,96 @@ localpath(char *path, struct vnode *vrootp, cred_t *cr)
 }
 
 /*
+ * Clean a stale v_path from a vnode.  This is only performed if the v_path has
+ * not been altered since it was found to be stale
+ */
+static void
+vnode_clear_vpath(vnode_t *vp, char *vpath_old)
+{
+	mutex_enter(&vp->v_lock);
+	if (vp->v_path != vn_vpath_empty && vp->v_path == vpath_old) {
+		vp->v_path = vn_vpath_empty;
+		mutex_exit(&vp->v_lock);
+		kmem_free(vpath_old, strlen(vpath_old) + 1);
+	} else {
+		mutex_exit(&vp->v_lock);
+	}
+}
+
+/*
+ * Validate that a pathname refers to a given vnode.
+ */
+static int
+vnode_valid_pn(vnode_t *vp, vnode_t *vrootp, pathname_t *pn, pathname_t *rpn,
+    int flags, cred_t *cr)
+{
+	vnode_t *compvp;
+	/*
+	 * If we are in a zone or a chroot environment, then we have to
+	 * take additional steps, since the path to the root might not
+	 * be readable with the current credentials, even though the
+	 * process can legitmately access the file.  In this case, we
+	 * do the following:
+	 *
+	 * lookuppnvp() with all privileges to get the resolved path.
+	 * call localpath() to get the local portion of the path, and
+	 * continue as normal.
+	 *
+	 * If the the conversion to a local path fails, then we continue
+	 * as normal.  This is a heuristic to make process object file
+	 * paths available from within a zone.  Because lofs doesn't
+	 * support page operations, the vnode stored in the seg_t is
+	 * actually the underlying real vnode, not the lofs node itself.
+	 * Most of the time, the lofs path is the same as the underlying
+	 * vnode (for example, /usr/lib/libc.so.1).
+	 */
+	if (vrootp != rootdir) {
+		char *local = NULL;
+
+		VN_HOLD(rootdir);
+		if (lookuppnvp(pn, rpn, FOLLOW, NULL, &compvp, rootdir,
+		    rootdir, kcred) == 0) {
+			local = localpath(rpn->pn_path, vrootp, kcred);
+			VN_RELE(compvp);
+		}
+
+		/*
+		 * The original pn was changed through lookuppnvp().
+		 * Set it to local for next validation attempt.
+		 */
+		if (local) {
+			(void) pn_set(pn, local);
+		} else {
+			return (1);
+		}
+	}
+
+	/*
+	 * We should have a local path at this point, so start the search from
+	 * the root of the current process.
+	 */
+	VN_HOLD(vrootp);
+	if (vrootp != rootdir)
+		VN_HOLD(vrootp);
+	if (lookuppnvp(pn, rpn, FOLLOW | flags, NULL, &compvp, vrootp, vrootp,
+	    cr) == 0) {
+		/*
+		 * Check to see if the returned vnode is the same as the one we
+		 * expect.
+		 */
+		if (vn_compare(vp, compvp) ||
+		    vnode_match(vp, compvp, cr)) {
+			VN_RELE(compvp);
+			return (0);
+		} else {
+			VN_RELE(compvp);
+		}
+	}
+
+	return (1);
+}
+
+/*
  * Given a directory, return the full, resolved path.  This looks up "..",
  * searches for the given vnode in the parent, appends the component, etc.  It
  * is used to implement vnodetopath() and getcwd() when the cached path fails.
@@ -995,6 +1087,8 @@ dirtopath(vnode_t *vrootp, vnode_t *vp, char *buf, size_t buflen, int flags,
 	char		*bufloc;
 	size_t		dlen = DIRENT64_RECLEN(MAXPATHLEN);
 	refstr_t	*mntpt;
+	char *vpath_cached;
+	boolean_t vpath_stale;
 
 	/* Operation only allowed on directories */
 	ASSERT(vp->v_type == VDIR);
@@ -1088,40 +1182,28 @@ dirtopath(vnode_t *vrootp, vnode_t *vp, char *buf, size_t buflen, int flags,
 		 * Shortcut: see if this vnode has correct v_path. If so,
 		 * we have the work done.
 		 */
+		vpath_cached = NULL;
+		vpath_stale = B_FALSE;
 		mutex_enter(&vp->v_lock);
-		if (vp->v_path != NULL) {
-
-			if ((err = pn_set(&pn, vp->v_path)) == 0) {
-				mutex_exit(&vp->v_lock);
-				rpn.pn_path = rpn.pn_buf;
-
-				/*
-				 * Ensure the v_path pointing to correct vnode
-				 */
-				VN_HOLD(vrootp);
-				if (vrootp != rootdir)
-					VN_HOLD(vrootp);
-				if (lookuppnvp(&pn, &rpn, flags, NULL,
-				    &cmpvp, vrootp, vrootp, cr) == 0) {
-
-					if (VN_CMP(vp, cmpvp)) {
-						VN_RELE(cmpvp);
+		if (vp->v_path != vn_vpath_empty &&
+		    pn_set(&pn, vp->v_path) == 0) {
+			vpath_cached = vp->v_path;
+			mutex_exit(&vp->v_lock);
+			rpn.pn_path = rpn.pn_buf;
 
-						complen = strlen(rpn.pn_path);
-						bufloc -= complen;
-						if (bufloc < buf) {
-							err = ERANGE;
-							goto out;
-						}
-						bcopy(rpn.pn_path, bufloc,
-						    complen);
-						break;
-					} else {
-						VN_RELE(cmpvp);
-					}
+			/* Ensure the v_path pointing to correct vnode */
+			if (vnode_valid_pn(vp, vrootp, &pn, &rpn, flags,
+			    cr) == 0) {
+				complen = strlen(rpn.pn_path);
+				bufloc -= complen;
+				if (bufloc < buf) {
+					err = ERANGE;
+					goto out;
 				}
+				bcopy(rpn.pn_path, bufloc, complen);
+				break;
 			} else {
-				mutex_exit(&vp->v_lock);
+				vpath_stale = B_TRUE;
 			}
 		} else {
 			mutex_exit(&vp->v_lock);
@@ -1166,38 +1248,6 @@ dirtopath(vnode_t *vrootp, vnode_t *vp, char *buf, size_t buflen, int flags,
 		}
 
 		/*
-		 * Try to obtain the path component from dnlc cache
-		 * before searching through the directory.
-		 */
-		if ((cmpvp = dnlc_reverse_lookup(vp, dbuf, dlen)) != NULL) {
-			/*
-			 * If we got parent vnode as a result,
-			 * then the answered path is correct.
-			 */
-			if (VN_CMP(cmpvp, pvp)) {
-				VN_RELE(cmpvp);
-				complen = strlen(dbuf);
-				bufloc -= complen;
-				if (bufloc <= buf) {
-					err = ENAMETOOLONG;
-					goto out;
-				}
-				bcopy(dbuf, bufloc, complen);
-
-				/* Prepend a slash to the current path */
-				*--bufloc = '/';
-
-				/* And continue with the next component */
-				VN_RELE(vp);
-				vp = pvp;
-				pvp = NULL;
-				continue;
-			} else {
-				VN_RELE(cmpvp);
-			}
-		}
-
-		/*
 		 * Search the parent directory for the entry corresponding to
 		 * this vnode.
 		 */
@@ -1215,6 +1265,11 @@ dirtopath(vnode_t *vrootp, vnode_t *vp, char *buf, size_t buflen, int flags,
 		/* Prepend a slash to the current path.  */
 		*--bufloc = '/';
 
+		/* Clear vp->v_path if it was found to be stale. */
+		if (vpath_stale == B_TRUE) {
+			vnode_clear_vpath(vp, vpath_cached);
+		}
+
 		/* And continue with the next component */
 		VN_RELE(vp);
 		vp = pvp;
@@ -1306,144 +1361,49 @@ vnodetopath_common(vnode_t *vrootp, vnode_t *vp, char *buf, size_t buflen,
 			VN_RELE(vp);
 	}
 
-	pn_alloc(&pn);
 
 	/*
-	 * Check to see if we have a cached path in the vnode.
+	 * Check to see if we have a valid cached path in the vnode.
 	 */
+	pn_alloc(&pn);
 	mutex_enter(&vp->v_lock);
-	if (vp->v_path != NULL) {
+	if (vp->v_path != vn_vpath_empty) {
 		(void) pn_set(&pn, vp->v_path);
 		mutex_exit(&vp->v_lock);
 
-		pn_alloc(&rpn);
-
 		/* We should only cache absolute paths */
 		ASSERT(pn.pn_buf[0] == '/');
 
-		/*
-		 * If we are in a zone or a chroot environment, then we have to
-		 * take additional steps, since the path to the root might not
-		 * be readable with the current credentials, even though the
-		 * process can legitmately access the file.  In this case, we
-		 * do the following:
-		 *
-		 * lookuppnvp() with all privileges to get the resolved path.
-		 * call localpath() to get the local portion of the path, and
-		 * continue as normal.
-		 *
-		 * If the the conversion to a local path fails, then we continue
-		 * as normal.  This is a heuristic to make process object file
-		 * paths available from within a zone.  Because lofs doesn't
-		 * support page operations, the vnode stored in the seg_t is
-		 * actually the underlying real vnode, not the lofs node itself.
-		 * Most of the time, the lofs path is the same as the underlying
-		 * vnode (for example, /usr/lib/libc.so.1).
-		 */
-		if (vrootp != rootdir) {
-			char *local = NULL;
-			VN_HOLD(rootdir);
-			if (lookuppnvp(&pn, &rpn, FOLLOW,
-			    NULL, &compvp, rootdir, rootdir, kcred) == 0) {
-				local = localpath(rpn.pn_path, vrootp,
-				    kcred);
-				VN_RELE(compvp);
-			}
-
-			/*
-			 * The original pn was changed through lookuppnvp().
-			 * Set it to local for next validation attempt.
-			 */
-			if (local) {
-				(void) pn_set(&pn, local);
-			} else {
-				goto notcached;
+		pn_alloc(&rpn);
+		if (vnode_valid_pn(vp, vrootp, &pn, &rpn, flags, cr) == 0) {
+			/* Return the result, if we're able. */
+			if (buflen > rpn.pn_pathlen) {
+				bcopy(rpn.pn_path, buf, rpn.pn_pathlen + 1);
+				pn_free(&pn);
+				pn_free(&rpn);
+				VN_RELE(vrootp);
+				if (doclose) {
+					(void) VOP_CLOSE(vp, FREAD, 1, 0, cr,
+					    NULL);
+					VN_RELE(vp);
+				}
+				return (0);
 			}
 		}
-
 		/*
-		 * We should have a local path at this point, so start the
-		 * search from the root of the current process.
+		 * A stale v_path will be purged by the later dirtopath lookup.
 		 */
-		VN_HOLD(vrootp);
-		if (vrootp != rootdir)
-			VN_HOLD(vrootp);
-		ret = lookuppnvp(&pn, &rpn, FOLLOW | flags, NULL,
-		    &compvp, vrootp, vrootp, cr);
-		if (ret == 0) {
-			/*
-			 * Check to see if the returned vnode is the same as
-			 * the one we expect.  If not, give up.
-			 */
-			if (!vn_compare(vp, compvp) &&
-			    !vnode_match(vp, compvp, cr)) {
-				VN_RELE(compvp);
-				goto notcached;
-			}
-
-			VN_RELE(compvp);
-
-			/*
-			 * Return the result.
-			 */
-			if (buflen <= rpn.pn_pathlen)
-				goto notcached;
-
-			bcopy(rpn.pn_path, buf, rpn.pn_pathlen + 1);
-			pn_free(&pn);
-			pn_free(&rpn);
-			VN_RELE(vrootp);
-			if (doclose) {
-				(void) VOP_CLOSE(vp, FREAD, 1, 0, cr, NULL);
-				VN_RELE(vp);
-			}
-			return (0);
-		}
-
-notcached:
 		pn_free(&rpn);
 	} else {
 		mutex_exit(&vp->v_lock);
 	}
-
 	pn_free(&pn);
 
 	if (vp->v_type != VDIR) {
-		/*
-		 * If we don't have a directory, try to find it in the dnlc via
-		 * reverse lookup.  Once this is found, we can use the regular
-		 * directory search to find the full path.
-		 */
-		if ((pvp = dnlc_reverse_lookup(vp, path, MAXNAMELEN)) != NULL) {
-			/*
-			 * Check if we have read privilege so, that
-			 * we can lookup the path in the directory
-			 */
-			ret = 0;
-			if ((flags & LOOKUP_CHECKREAD)) {
-				ret = VOP_ACCESS(pvp, VREAD, 0, cr, NULL);
-			}
-			if (ret == 0) {
-				ret = dirtopath(vrootp, pvp, buf, buflen,
-				    flags, cr);
-			}
-			if (ret == 0) {
-				len = strlen(buf);
-				if (len + strlen(path) + 1 >= buflen) {
-					ret = ENAMETOOLONG;
-				} else {
-					if (buf[len - 1] != '/')
-						buf[len++] = '/';
-					bcopy(path, buf + len,
-					    strlen(path) + 1);
-				}
-			}
-
-			VN_RELE(pvp);
-		} else
-			ret = ENOENT;
-	} else
+		ret = ENOENT;
+	} else {
 		ret = dirtopath(vrootp, vp, buf, buflen, flags, cr);
+	}
 
 	VN_RELE(vrootp);
 	if (doclose) {
diff --git a/usr/src/uts/common/fs/lxproc/lxpr_subr.c b/usr/src/uts/common/fs/lxproc/lxpr_subr.c
new file mode 100644
index 0000000000..3c1405d4af
--- /dev/null
+++ b/usr/src/uts/common/fs/lxproc/lxpr_subr.c
@@ -0,0 +1,524 @@
+/*
+ * CDDL HEADER START
+ *
+ * The contents of this file are subject to the terms of the
+ * Common Development and Distribution License (the "License").
+ * You may not use this file except in compliance with the License.
+ *
+ * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
+ * or http://www.opensolaris.org/os/licensing.
+ * See the License for the specific language governing permissions
+ * and limitations under the License.
+ *
+ * When distributing Covered Code, include this CDDL HEADER in each
+ * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
+ * If applicable, add the following below this CDDL HEADER, with the
+ * fields enclosed by brackets "[]" replaced with your own identifying
+ * information: Portions Copyright [yyyy] [name of copyright owner]
+ *
+ * CDDL HEADER END
+ */
+/*
+ * Copyright 2008 Sun Microsystems, Inc.  All rights reserved.
+ * Use is subject to license terms.
+ */
+
+/*
+ * Copyright (c) 2012, Joyent, Inc. All rights reserved.
+ */
+
+#include <sys/varargs.h>
+#include <sys/cpuvar.h>
+#include <sys/mman.h>
+#include <sys/vmsystm.h>
+#include <sys/prsystm.h>
+
+#include "lxproc.h"
+
+#define	LXPRCACHE_NAME "lxpr_cache"
+
+static int lxpr_node_constructor(void *, void *, int);
+static void lxpr_node_destructor(void *, void *);
+
+static kmem_cache_t *lxpr_node_cache;
+
+struct lxpr_uiobuf {
+	uio_t *uiop;
+	char *buffer;
+	uint32_t buffsize;
+	char *pos;
+	size_t beg;
+	int error;
+};
+
+int lxpr_bufsize = 4000;
+
+struct lxpr_uiobuf *
+lxpr_uiobuf_new(uio_t *uiop)
+{
+	/* Allocate memory for both lxpr_uiobuf and output buffer */
+	int bufsize = lxpr_bufsize;
+	struct lxpr_uiobuf *uiobuf =
+	    kmem_alloc(sizeof (struct lxpr_uiobuf) + bufsize, KM_SLEEP);
+
+	uiobuf->uiop = uiop;
+	uiobuf->buffer = (char *)&uiobuf[1];
+	uiobuf->buffsize = bufsize;
+	uiobuf->pos = uiobuf->buffer;
+	uiobuf->beg = 0;
+	uiobuf->error = 0;
+
+	return (uiobuf);
+}
+
+void
+lxpr_uiobuf_free(struct lxpr_uiobuf *uiobuf)
+{
+	ASSERT(uiobuf != NULL);
+	ASSERT(uiobuf->pos == uiobuf->buffer);
+
+	kmem_free(uiobuf, sizeof (struct lxpr_uiobuf) + uiobuf->buffsize);
+}
+
+void
+lxpr_uiobuf_seek(struct lxpr_uiobuf *uiobuf, offset_t offset)
+{
+	uiobuf->uiop->uio_offset = (off_t)offset;
+}
+
+void
+lxpr_uiobuf_seterr(struct lxpr_uiobuf *uiobuf, int err)
+{
+	ASSERT(uiobuf->error == 0);
+
+	uiobuf->error = err;
+}
+
+int
+lxpr_uiobuf_flush(struct lxpr_uiobuf *uiobuf)
+{
+	off_t off = uiobuf->uiop->uio_offset;
+	caddr_t uaddr = uiobuf->buffer;
+	size_t beg = uiobuf->beg;
+	size_t size = (uintptr_t)uiobuf->pos - (uintptr_t)uaddr;
+
+	if (uiobuf->error == 0 && uiobuf->uiop->uio_resid != 0) {
+		ASSERT(off >= beg);
+
+		if (beg + size > off && off >= 0)
+			uiobuf->error =
+			    uiomove(uaddr + (off - beg), size - (off - beg),
+			    UIO_READ, uiobuf->uiop);
+
+		uiobuf->beg += size;
+	}
+
+	uiobuf->pos = uaddr;
+
+	return (uiobuf->error);
+}
+
+void
+lxpr_uiobuf_write(struct lxpr_uiobuf *uiobuf, const char *buf, size_t size)
+{
+	/* While we can still carry on */
+	while (uiobuf->error == 0 && uiobuf->uiop->uio_resid != 0) {
+		uintptr_t remain = (uintptr_t)uiobuf->buffsize -
+		    ((uintptr_t)uiobuf->pos - (uintptr_t)uiobuf->buffer);
+
+		/* Enough space in buffer? */
+		if (remain >= size) {
+			bcopy(buf, uiobuf->pos, size);
+			uiobuf->pos += size;
+			return;
+		}
+
+		/* Not enough space, so copy all we can and try again */
+		bcopy(buf, uiobuf->pos, remain);
+		uiobuf->pos += remain;
+		(void) lxpr_uiobuf_flush(uiobuf);
+		buf += remain;
+		size -= remain;
+	}
+}
+
+#define	TYPBUFFSIZE 256
+
+void
+lxpr_uiobuf_printf(struct lxpr_uiobuf *uiobuf, const char *fmt, ...)
+{
+	va_list args;
+	char buff[TYPBUFFSIZE];
+	int len;
+	char *buffer;
+
+	/* Can we still do any output */
+	if (uiobuf->error != 0 || uiobuf->uiop->uio_resid == 0)
+		return;
+
+	va_start(args, fmt);
+
+	/* Try using stack allocated buffer */
+	len = vsnprintf(buff, TYPBUFFSIZE, fmt, args);
+	if (len < TYPBUFFSIZE) {
+		va_end(args);
+		lxpr_uiobuf_write(uiobuf, buff, len);
+		return;
+	}
+
+	/* Not enough space in pre-allocated buffer */
+	buffer = kmem_alloc(len + 1, KM_SLEEP);
+
+	/*
+	 * We know we allocated the correct amount of space
+	 * so no check on the return value
+	 */
+	(void) vsnprintf(buffer, len+1, fmt, args);
+	lxpr_uiobuf_write(uiobuf, buffer, len);
+	va_end(args);
+	kmem_free(buffer, len+1);
+}
+
+/*
+ * lxpr_lock():
+ *
+ * Lookup process from pid and return with p_plock and P_PR_LOCK held.
+ */
+proc_t *
+lxpr_lock(pid_t pid)
+{
+	proc_t *p;
+	kmutex_t *mp;
+
+	ASSERT(!MUTEX_HELD(&pidlock));
+
+	for (;;) {
+		mutex_enter(&pidlock);
+
+		/*
+		 * If the pid is 1, we really want the zone's init process
+		 */
+		p = prfind((pid == 1) ?
+		    curproc->p_zone->zone_proc_initpid : pid);
+
+		if (p == NULL || p->p_stat == SIDL) {
+			mutex_exit(&pidlock);
+			return (NULL);
+		}
+
+		/*
+		 * p_lock is persistent, but p itself is not -- it could
+		 * vanish during cv_wait().  Load p->p_lock now so we can
+		 * drop it after cv_wait() without referencing p.
+		 */
+		mp = &p->p_lock;
+		mutex_enter(mp);
+
+		mutex_exit(&pidlock);
+
+		if (p->p_flag & SEXITING) {
+			/*
+			 * This process is exiting -- let it go.
+			 */
+			mutex_exit(mp);
+			return (NULL);
+		}
+
+		if (!(p->p_proc_flag & P_PR_LOCK))
+			break;
+
+		cv_wait(&pr_pid_cv[p->p_slot], mp);
+		mutex_exit(mp);
+	}
+
+	p->p_proc_flag |= P_PR_LOCK;
+	THREAD_KPRI_REQUEST();
+	return (p);
+}
+
+/*
+ * lxpr_unlock()
+ *
+ * Unlock locked process
+ */
+void
+lxpr_unlock(proc_t *p)
+{
+	ASSERT(p->p_proc_flag & P_PR_LOCK);
+	ASSERT(MUTEX_HELD(&p->p_lock));
+	ASSERT(!MUTEX_HELD(&pidlock));
+
+	cv_signal(&pr_pid_cv[p->p_slot]);
+	p->p_proc_flag &= ~P_PR_LOCK;
+	mutex_exit(&p->p_lock);
+	THREAD_KPRI_RELEASE();
+}
+
+void
+lxpr_initnodecache()
+{
+	lxpr_node_cache = kmem_cache_create(LXPRCACHE_NAME,
+	    sizeof (lxpr_node_t), 0,
+	    lxpr_node_constructor, lxpr_node_destructor, NULL, NULL, NULL, 0);
+}
+
+void
+lxpr_fininodecache()
+{
+	kmem_cache_destroy(lxpr_node_cache);
+}
+
+/* ARGSUSED */
+static int
+lxpr_node_constructor(void *buf, void *un, int kmflags)
+{
+	lxpr_node_t	*lxpnp = buf;
+	vnode_t		*vp;
+
+	vp = lxpnp->lxpr_vnode = vn_alloc(kmflags);
+	if (vp == NULL)
+		return (-1);
+
+	(void) vn_setops(vp, lxpr_vnodeops);
+	vp->v_data = lxpnp;
+
+	return (0);
+}
+
+/* ARGSUSED */
+static void
+lxpr_node_destructor(void *buf, void *un)
+{
+	lxpr_node_t	*lxpnp = buf;
+
+	vn_free(LXPTOV(lxpnp));
+}
+
+/*
+ * Calculate an inode number
+ *
+ * This takes various bits of info and munges them
+ * to give the inode number for an lxproc node
+ */
+ino_t
+lxpr_inode(lxpr_nodetype_t type, pid_t pid, int fd)
+{
+	if (pid == 1)
+		pid = curproc->p_zone->zone_proc_initpid;
+
+	switch (type) {
+	case LXPR_PIDDIR:
+		return (pid + 1);
+	case LXPR_PROCDIR:
+		return (maxpid + 2);
+	case LXPR_PID_FD_FD:
+		return (maxpid + 2 +
+		    (pid * (LXPR_FD_PERPROC + LXPR_NFILES)) +
+		    LXPR_NFILES + fd);
+	default:
+		return (maxpid + 2 +
+		    (pid * (LXPR_FD_PERPROC + LXPR_NFILES)) +
+		    type);
+	}
+}
+
+/*
+ * Return inode number of parent (directory)
+ */
+ino_t
+lxpr_parentinode(lxpr_node_t *lxpnp)
+{
+	/*
+	 * If the input node is the root then the parent inode
+	 * is the mounted on inode so just return our inode number
+	 */
+	if (lxpnp->lxpr_type != LXPR_PROCDIR)
+		return (VTOLXP(lxpnp->lxpr_parent)->lxpr_ino);
+	else
+		return (lxpnp->lxpr_ino);
+}
+
+/*
+ * Allocate a new lxproc node
+ *
+ * This also allocates the vnode associated with it
+ */
+lxpr_node_t *
+lxpr_getnode(vnode_t *dp, lxpr_nodetype_t type, proc_t *p, int fd)
+{
+	lxpr_node_t *lxpnp;
+	vnode_t *vp;
+	user_t *up;
+	timestruc_t now;
+
+	/*
+	 * Allocate a new node. It is deallocated in vop_innactive
+	 */
+	lxpnp = kmem_cache_alloc(lxpr_node_cache, KM_SLEEP);
+
+	/*
+	 * Set defaults (may be overridden below)
+	 */
+	gethrestime(&now);
+	lxpnp->lxpr_type = type;
+	lxpnp->lxpr_realvp = NULL;
+	lxpnp->lxpr_parent = dp;
+	VN_HOLD(dp);
+	if (p != NULL) {
+		lxpnp->lxpr_pid = ((p->p_pid ==
+		    curproc->p_zone->zone_proc_initpid) ? 1 : p->p_pid);
+
+		lxpnp->lxpr_time = PTOU(p)->u_start;
+		lxpnp->lxpr_uid = crgetruid(p->p_cred);
+		lxpnp->lxpr_gid = crgetrgid(p->p_cred);
+		lxpnp->lxpr_ino = lxpr_inode(type, p->p_pid, fd);
+	} else {
+		/* Pretend files without a proc belong to sched */
+		lxpnp->lxpr_pid = 0;
+		lxpnp->lxpr_time = now;
+		lxpnp->lxpr_uid = lxpnp->lxpr_gid = 0;
+		lxpnp->lxpr_ino = lxpr_inode(type, 0, 0);
+	}
+
+	/* initialize the vnode data */
+	vp = lxpnp->lxpr_vnode;
+	vn_reinit(vp);
+	vp->v_flag = VNOCACHE|VNOMAP|VNOSWAP|VNOMOUNT;
+	vp->v_vfsp = dp->v_vfsp;
+
+	/*
+	 * Do node specific stuff
+	 */
+	switch (type) {
+	case LXPR_PROCDIR:
+		vp->v_flag |= VROOT;
+		vp->v_type = VDIR;
+		lxpnp->lxpr_mode = 0555;	/* read-search by everyone */
+		break;
+
+	case LXPR_PID_CURDIR:
+		ASSERT(p != NULL);
+
+		/*
+		 * Zombie check.  p_stat is officially protected by pidlock,
+		 * but we can't grab pidlock here because we already hold
+		 * p_lock.  Luckily if we look at the process exit code
+		 * we see that p_stat only transisions from SRUN to SZOMB
+		 * while p_lock is held.  Aside from this, the only other
+		 * p_stat transition that we need to be aware about is
+		 * SIDL to SRUN, but that's not a problem since lxpr_lock()
+		 * ignores nodes in the SIDL state so we'll never get a node
+		 * that isn't already in the SRUN state.
+		 */
+		if (p->p_stat == SZOMB) {
+			lxpnp->lxpr_realvp = NULL;
+		} else {
+			up = PTOU(p);
+			lxpnp->lxpr_realvp = up->u_cdir;
+			ASSERT(lxpnp->lxpr_realvp != NULL);
+			VN_HOLD(lxpnp->lxpr_realvp);
+		}
+		vp->v_type = VLNK;
+		lxpnp->lxpr_mode = 0777;	/* anyone does anything ! */
+		break;
+
+	case LXPR_PID_ROOTDIR:
+		ASSERT(p != NULL);
+		/* Zombie check.  see locking comment above */
+		if (p->p_stat == SZOMB) {
+			lxpnp->lxpr_realvp = NULL;
+		} else {
+			up = PTOU(p);
+			lxpnp->lxpr_realvp =
+			    up->u_rdir != NULL ? up->u_rdir : rootdir;
+			ASSERT(lxpnp->lxpr_realvp != NULL);
+			VN_HOLD(lxpnp->lxpr_realvp);
+		}
+		vp->v_type = VLNK;
+		lxpnp->lxpr_mode = 0777;	/* anyone does anything ! */
+		break;
+
+	case LXPR_PID_EXE:
+		ASSERT(p != NULL);
+		lxpnp->lxpr_realvp = p->p_exec;
+		if (lxpnp->lxpr_realvp != NULL) {
+			VN_HOLD(lxpnp->lxpr_realvp);
+		}
+		vp->v_type = VLNK;
+		lxpnp->lxpr_mode = 0777;
+		break;
+
+	case LXPR_SELF:
+		vp->v_type = VLNK;
+		lxpnp->lxpr_mode = 0777;	/* anyone does anything ! */
+		break;
+
+	case LXPR_PID_FD_FD:
+		ASSERT(p != NULL);
+		/* lxpr_realvp is set after we return */
+		vp->v_type = VLNK;
+		lxpnp->lxpr_mode = 0700;	/* read-write-exe owner only */
+		break;
+
+	case LXPR_PID_FDDIR:
+		ASSERT(p != NULL);
+		vp->v_type = VDIR;
+		lxpnp->lxpr_mode = 0500;	/* read-search by owner only */
+		break;
+
+	case LXPR_PIDDIR:
+		ASSERT(p != NULL);
+		vp->v_type = VDIR;
+		lxpnp->lxpr_mode = 0511;
+		break;
+
+	case LXPR_NETDIR:
+		vp->v_type = VDIR;
+		lxpnp->lxpr_mode = 0555;	/* read-search by all */
+		break;
+
+	case LXPR_PID_ENV:
+	case LXPR_PID_MEM:
+		ASSERT(p != NULL);
+		/*FALLTHRU*/
+	case LXPR_KCORE:
+		vp->v_type = VREG;
+		lxpnp->lxpr_mode = 0400;	/* read-only by owner only */
+		break;
+
+	default:
+		vp->v_type = VREG;
+		lxpnp->lxpr_mode = 0444;	/* read-only by all */
+		break;
+	}
+
+	return (lxpnp);
+}
+
+
+/*
+ * Free the storage obtained from lxpr_getnode().
+ */
+void
+lxpr_freenode(lxpr_node_t *lxpnp)
+{
+	ASSERT(lxpnp != NULL);
+	ASSERT(LXPTOV(lxpnp) != NULL);
+
+	/*
+	 * delete any association with realvp
+	 */
+	if (lxpnp->lxpr_realvp != NULL)
+		VN_RELE(lxpnp->lxpr_realvp);
+
+	/*
+	 * delete any association with parent vp
+	 */
+	if (lxpnp->lxpr_parent != NULL)
+		VN_RELE(lxpnp->lxpr_parent);
+
+	/*
+	 * Release the lxprnode.
+	 */
+	kmem_cache_free(lxpr_node_cache, lxpnp);
+}
diff --git a/usr/src/uts/common/fs/lxproc/lxpr_vfsops.c b/usr/src/uts/common/fs/lxproc/lxpr_vfsops.c
new file mode 100644
index 0000000000..1bb7bd3823
--- /dev/null
+++ b/usr/src/uts/common/fs/lxproc/lxpr_vfsops.c
@@ -0,0 +1,367 @@
+/*
+ * CDDL HEADER START
+ *
+ * The contents of this file are subject to the terms of the
+ * Common Development and Distribution License (the "License").
+ * You may not use this file except in compliance with the License.
+ *
+ * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
+ * or http://www.opensolaris.org/os/licensing.
+ * See the License for the specific language governing permissions
+ * and limitations under the License.
+ *
+ * When distributing Covered Code, include this CDDL HEADER in each
+ * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
+ * If applicable, add the following below this CDDL HEADER, with the
+ * fields enclosed by brackets "[]" replaced with your own identifying
+ * information: Portions Copyright [yyyy] [name of copyright owner]
+ *
+ * CDDL HEADER END
+ */
+/*
+ * Copyright 2007 Sun Microsystems, Inc.  All rights reserved.
+ * Use is subject to license terms.
+ */
+
+/*
+ * Copyright (c) 2012, Joyent, Inc. All rights reserved.
+ */
+
+#include <sys/types.h>
+#include <sys/param.h>
+#include <sys/cmn_err.h>
+#include <sys/cred.h>
+#include <sys/debug.h>
+#include <sys/errno.h>
+#include <sys/proc.h>
+#include <sys/stat.h>
+#include <sys/statvfs.h>
+#include <sys/sysmacros.h>
+#include <sys/systm.h>
+#include <sys/var.h>
+#include <sys/vfs.h>
+#include <sys/vfs_opreg.h>
+#include <sys/vnode.h>
+#include <sys/mode.h>
+#include <sys/signal.h>
+#include <sys/user.h>
+#include <sys/mount.h>
+#include <sys/bitmap.h>
+#include <sys/kmem.h>
+#include <sys/policy.h>
+#include <sys/modctl.h>
+#include <sys/sunddi.h>
+#include <sys/sunldi.h>
+
+#include "lxproc.h"
+
+/* Module level parameters */
+static int	lxprocfstype;
+static dev_t	lxprocdev;
+static kmutex_t	lxpr_mount_lock;
+
+int nproc_highbit;	/* highbit(v.v_nproc) */
+
+static int lxpr_mount(vfs_t *, vnode_t *, mounta_t *, cred_t *);
+static int lxpr_unmount(vfs_t *, int, cred_t *);
+static int lxpr_root(vfs_t *, vnode_t **);
+static int lxpr_statvfs(vfs_t *, statvfs64_t *);
+static int lxpr_init(int, char *);
+
+static vfsdef_t vfw = {
+	VFSDEF_VERSION,
+	"lxproc",
+	lxpr_init,
+	VSW_ZMOUNT,
+	NULL
+};
+
+/*
+ * Module linkage information for the kernel.
+ */
+extern struct mod_ops mod_fsops;
+
+static struct modlfs modlfs = {
+	&mod_fsops, "generic linux procfs", &vfw
+};
+
+static struct modlinkage modlinkage = {
+	MODREV_1, (void *)&modlfs, NULL
+};
+
+int
+_init(void)
+{
+	return (mod_install(&modlinkage));
+}
+
+int
+_info(struct modinfo *modinfop)
+{
+	return (mod_info(&modlinkage, modinfop));
+}
+
+int
+_fini(void)
+{
+	int retval;
+
+	/*
+	 * attempt to unload the module
+	 */
+	if ((retval = mod_remove(&modlinkage)) != 0)
+		goto done;
+
+	/*
+	 * destroy lxpr_node cache
+	 */
+	lxpr_fininodecache();
+
+	/*
+	 * clean out the vfsops and vnodeops
+	 */
+	(void) vfs_freevfsops_by_type(lxprocfstype);
+	vn_freevnodeops(lxpr_vnodeops);
+
+	mutex_destroy(&lxpr_mount_lock);
+done:
+	return (retval);
+}
+
+static int
+lxpr_init(int fstype, char *name)
+{
+	static const fs_operation_def_t lxpr_vfsops_template[] = {
+		VFSNAME_MOUNT,		{ .vfs_mount = lxpr_mount },
+		VFSNAME_UNMOUNT,	{ .vfs_unmount = lxpr_unmount },
+		VFSNAME_ROOT,		{ .vfs_root = lxpr_root },
+		VFSNAME_STATVFS,	{ .vfs_statvfs = lxpr_statvfs },
+		NULL,			NULL
+	};
+	extern const fs_operation_def_t lxpr_vnodeops_template[];
+	int error;
+	major_t dev;
+
+	nproc_highbit = highbit(v.v_proc);
+	lxprocfstype = fstype;
+	ASSERT(lxprocfstype != 0);
+
+	mutex_init(&lxpr_mount_lock, NULL, MUTEX_DEFAULT, NULL);
+
+	/*
+	 * Associate VFS ops vector with this fstype.
+	 */
+	error = vfs_setfsops(fstype, lxpr_vfsops_template, NULL);
+	if (error != 0) {
+		cmn_err(CE_WARN, "lxpr_init: bad vfs ops template");
+		return (error);
+	}
+
+	/*
+	 * Set up vnode ops vector too.
+	 */
+	error = vn_make_ops(name, lxpr_vnodeops_template, &lxpr_vnodeops);
+	if (error != 0) {
+		(void) vfs_freevfsops_by_type(fstype);
+		cmn_err(CE_WARN, "lxpr_init: bad vnode ops template");
+		return (error);
+	}
+
+	/*
+	 * Assign a unique "device" number (used by stat(2)).
+	 */
+	if ((dev = getudev()) == (major_t)-1) {
+		cmn_err(CE_WARN, "lxpr_init: can't get unique device number");
+		dev = 0;
+	}
+
+	/*
+	 * Make the pseudo device
+	 */
+	lxprocdev = makedevice(dev, 0);
+
+	/*
+	 * Initialize cache for lxpr_nodes
+	 */
+	lxpr_initnodecache();
+
+	return (0);
+}
+
+static int
+lxpr_mount(vfs_t *vfsp, vnode_t *mvp, mounta_t *uap, cred_t *cr)
+{
+	lxpr_mnt_t *lxpr_mnt;
+	zone_t *zone = curproc->p_zone;
+	ldi_ident_t li;
+	int err;
+
+	/*
+	 * must be root to mount
+	 */
+	if (secpolicy_fs_mount(cr, mvp, vfsp) != 0)
+		return (EPERM);
+
+	/*
+	 * mount point must be a directory
+	 */
+	if (mvp->v_type != VDIR)
+		return (ENOTDIR);
+
+	if (zone == global_zone) {
+		zone_t *mntzone;
+
+		mntzone = zone_find_by_path(refstr_value(vfsp->vfs_mntpt));
+		zone_rele(mntzone);
+		if (zone != mntzone)
+			return (EBUSY);
+	}
+
+	/*
+	 * Having the resource be anything but "lxproc" doesn't make sense
+	 */
+	vfs_setresource(vfsp, "lxproc", 0);
+
+	lxpr_mnt = kmem_alloc(sizeof (*lxpr_mnt), KM_SLEEP);
+
+	if ((err = ldi_ident_from_mod(&modlinkage, &li)) != 0) {
+		kmem_free(lxpr_mnt, sizeof (*lxpr_mnt));
+		return (err);
+	}
+
+	lxpr_mnt->lxprm_li = li;
+
+	mutex_enter(&lxpr_mount_lock);
+
+	/*
+	 * Ensure we don't allow overlaying mounts
+	 */
+	mutex_enter(&mvp->v_lock);
+	if ((uap->flags & MS_OVERLAY) == 0 &&
+	    (mvp->v_count > 1 || (mvp->v_flag & VROOT))) {
+		mutex_exit(&mvp->v_lock);
+		mutex_exit(&lxpr_mount_lock);
+		kmem_free(lxpr_mnt, sizeof ((*lxpr_mnt)));
+		return (EBUSY);
+	}
+	mutex_exit(&mvp->v_lock);
+
+	/*
+	 * allocate the first vnode
+	 */
+	zone_hold(lxpr_mnt->lxprm_zone = zone);
+
+	/* Arbitrarily set the parent vnode to the mounted over directory */
+	lxpr_mnt->lxprm_node = lxpr_getnode(mvp, LXPR_PROCDIR, NULL, 0);
+
+	/* Correctly set the fs for the root node */
+	lxpr_mnt->lxprm_node->lxpr_vnode->v_vfsp = vfsp;
+
+	vfs_make_fsid(&vfsp->vfs_fsid, lxprocdev, lxprocfstype);
+	vfsp->vfs_bsize = DEV_BSIZE;
+	vfsp->vfs_fstype = lxprocfstype;
+	vfsp->vfs_data = (caddr_t)lxpr_mnt;
+	vfsp->vfs_dev = lxprocdev;
+
+	mutex_exit(&lxpr_mount_lock);
+
+	return (0);
+}
+
+static int
+lxpr_unmount(vfs_t *vfsp, int flag, cred_t *cr)
+{
+	lxpr_mnt_t *lxpr_mnt = (lxpr_mnt_t *)vfsp->vfs_data;
+	vnode_t *vp;
+	int count;
+
+	ASSERT(lxpr_mnt != NULL);
+	vp = LXPTOV(lxpr_mnt->lxprm_node);
+
+	mutex_enter(&lxpr_mount_lock);
+
+	/*
+	 * must be root to unmount
+	 */
+	if (secpolicy_fs_unmount(cr, vfsp) != 0) {
+		mutex_exit(&lxpr_mount_lock);
+		return (EPERM);
+	}
+
+	/*
+	 * forced unmount is not supported by this file system
+	 */
+	if (flag & MS_FORCE) {
+		mutex_exit(&lxpr_mount_lock);
+		return (ENOTSUP);
+	}
+
+	/*
+	 * Ensure that no vnodes are in use on this mount point.
+	 */
+	mutex_enter(&vp->v_lock);
+	count = vp->v_count;
+	mutex_exit(&vp->v_lock);
+	if (count > 1) {
+		mutex_exit(&lxpr_mount_lock);
+		return (EBUSY);
+	}
+
+	/*
+	 * purge the dnlc cache for vnode entries
+	 * associated with this file system
+	 */
+	count = dnlc_purge_vfsp(vfsp, 0);
+
+	/*
+	 * free up the lxprnode
+	 */
+	lxpr_freenode(lxpr_mnt->lxprm_node);
+	zone_rele(lxpr_mnt->lxprm_zone);
+	kmem_free(lxpr_mnt, sizeof (*lxpr_mnt));
+
+	mutex_exit(&lxpr_mount_lock);
+
+	return (0);
+}
+
+static int
+lxpr_root(vfs_t *vfsp, vnode_t **vpp)
+{
+	lxpr_node_t *lxpnp = ((lxpr_mnt_t *)vfsp->vfs_data)->lxprm_node;
+	vnode_t *vp = LXPTOV(lxpnp);
+
+	VN_HOLD(vp);
+	*vpp = vp;
+	return (0);
+}
+
+static int
+lxpr_statvfs(vfs_t *vfsp, statvfs64_t *sp)
+{
+	int n;
+	dev32_t d32;
+	extern uint_t nproc;
+
+	n = v.v_proc - nproc;
+
+	bzero((caddr_t)sp, sizeof (*sp));
+	sp->f_bsize	= DEV_BSIZE;
+	sp->f_frsize	= DEV_BSIZE;
+	sp->f_blocks	= (fsblkcnt64_t)0;
+	sp->f_bfree	= (fsblkcnt64_t)0;
+	sp->f_bavail	= (fsblkcnt64_t)0;
+	sp->f_files	= (fsfilcnt64_t)v.v_proc + 2;
+	sp->f_ffree	= (fsfilcnt64_t)n;
+	sp->f_favail	= (fsfilcnt64_t)n;
+	(void) cmpldev(&d32, vfsp->vfs_dev);
+	sp->f_fsid	= d32;
+	/* It is guaranteed that vsw_name will fit in f_basetype */
+	(void) strcpy(sp->f_basetype, vfssw[lxprocfstype].vsw_name);
+	sp->f_flag = vf_to_stf(vfsp->vfs_flag);
+	sp->f_namemax = 64;		/* quite arbitrary */
+
+	(void) strcpy(sp->f_fstr, "lxproc");
+
+	return (0);
+}
diff --git a/usr/src/uts/common/fs/lxproc/lxpr_vnops.c b/usr/src/uts/common/fs/lxproc/lxpr_vnops.c
new file mode 100644
index 0000000000..48f4efc1bf
--- /dev/null
+++ b/usr/src/uts/common/fs/lxproc/lxpr_vnops.c
@@ -0,0 +1,3099 @@
+/*
+ * CDDL HEADER START
+ *
+ * The contents of this file are subject to the terms of the
+ * Common Development and Distribution License (the "License").
+ * You may not use this file except in compliance with the License.
+ *
+ * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
+ * or http://www.opensolaris.org/os/licensing.
+ * See the License for the specific language governing permissions
+ * and limitations under the License.
+ *
+ * When distributing Covered Code, include this CDDL HEADER in each
+ * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
+ * If applicable, add the following below this CDDL HEADER, with the
+ * fields enclosed by brackets "[]" replaced with your own identifying
+ * information: Portions Copyright [yyyy] [name of copyright owner]
+ *
+ * CDDL HEADER END
+ */
+/*
+ * Copyright 2010 Sun Microsystems, Inc.  All rights reserved.
+ * Use is subject to license terms.
+ */
+
+/*
+ * Copyright 2015 Joyent, Inc. All rights reserved.
+ */
+
+/*
+ * lxproc -- a loosely Linux-compatible /proc
+ *
+ * We have -- confusingly -- two implementations of Linux /proc.  One is to
+ * support the LX brand with a Linux /proc entirely compatible with the Linux
+ * world view; the other -- this one -- is to support native (but Linux-borne)
+ * programs that wish to view the native system via the Linux /proc model.  So
+ * the aspiration here is to provide something that sufficiently approximates
+ * the Linux /proc implementation for purposes of offering some compatibility
+ * for simple Linux /proc readers (e.g., ps/top/htop).  However, it is not
+ * intended to exactly mimic Linux semantics; when choosing between offering
+ * compatibility and telling the truth, we emphatically pick the truth.  A
+ * particular glaring example of this is the Linux notion of "tasks" (that is,
+ * threads), which -- due to historical misadventures on Linux -- allocate their
+ * identifiers from the process identifier space.  (That is, each thread has in
+ * effect a pid.)  Some Linux /proc readers have come to depend on this
+ * attribute, and become confused when threads appear with proper identifiers,
+ * so we simply opt for the pre-2.6 behavior, and do not present the tasks
+ * directory at all.  Similarly, when choosing between offering compatibility
+ * and remaining consistent with our broader security model, we (obviously)
+ * choose security over compatibility.  In short, this is meant to be a best
+ * effort -- no more -- and as such, it should not be unified with the much
+ * more complete Linux /proc implementation found in the LX brand.
+ */
+
+#include <sys/cpupart.h>
+#include <sys/cpuvar.h>
+#include <sys/session.h>
+#include <sys/vmparam.h>
+#include <sys/mman.h>
+#include <vm/rm.h>
+#include <vm/seg_vn.h>
+#include <sys/sdt.h>
+#include <sys/strlog.h>
+#include <sys/stropts.h>
+#include <sys/cmn_err.h>
+#include <sys/x86_archext.h>
+#include <sys/archsystm.h>
+#include <sys/fp.h>
+#include <sys/pool_pset.h>
+#include <sys/pset.h>
+#include <sys/zone.h>
+#include <sys/pghw.h>
+#include <sys/vfs_opreg.h>
+
+/* Dependent on procfs */
+extern kthread_t *prchoose(proc_t *);
+
+#include "lxproc.h"
+
+extern pgcnt_t swapfs_minfree;
+extern time_t boot_time;
+
+/*
+ * Pointer to the vnode ops vector for this fs.
+ * This is instantiated in lxprinit() in lxpr_vfsops.c
+ */
+vnodeops_t *lxpr_vnodeops;
+
+static int lxpr_open(vnode_t **, int, cred_t *, caller_context_t *);
+static int lxpr_close(vnode_t *, int, int, offset_t, cred_t *,
+    caller_context_t *);
+static int lxpr_read(vnode_t *, uio_t *, int, cred_t *, caller_context_t *);
+static int lxpr_getattr(vnode_t *, vattr_t *, int, cred_t *,
+    caller_context_t *);
+static int lxpr_access(vnode_t *, int, int, cred_t *, caller_context_t *);
+static int lxpr_lookup(vnode_t *, char *, vnode_t **,
+    pathname_t *, int, vnode_t *, cred_t *, caller_context_t *, int *,
+    pathname_t *);
+static int lxpr_readdir(vnode_t *, uio_t *, cred_t *, int *,
+    caller_context_t *, int);
+static int lxpr_readlink(vnode_t *, uio_t *, cred_t *, caller_context_t *);
+static int lxpr_cmp(vnode_t *, vnode_t *, caller_context_t *);
+static int lxpr_realvp(vnode_t *, vnode_t **, caller_context_t *);
+static int lxpr_sync(void);
+static void lxpr_inactive(vnode_t *, cred_t *, caller_context_t *);
+
+static vnode_t *lxpr_lookup_procdir(vnode_t *, char *);
+static vnode_t *lxpr_lookup_piddir(vnode_t *, char *);
+static vnode_t *lxpr_lookup_not_a_dir(vnode_t *, char *);
+static vnode_t *lxpr_lookup_fddir(vnode_t *, char *);
+static vnode_t *lxpr_lookup_netdir(vnode_t *, char *);
+
+static int lxpr_readdir_procdir(lxpr_node_t *, uio_t *, int *);
+static int lxpr_readdir_piddir(lxpr_node_t *, uio_t *, int *);
+static int lxpr_readdir_not_a_dir(lxpr_node_t *, uio_t *, int *);
+static int lxpr_readdir_fddir(lxpr_node_t *, uio_t *, int *);
+static int lxpr_readdir_netdir(lxpr_node_t *, uio_t *, int *);
+
+static void lxpr_read_invalid(lxpr_node_t *, lxpr_uiobuf_t *);
+static void lxpr_read_empty(lxpr_node_t *, lxpr_uiobuf_t *);
+static void lxpr_read_cpuinfo(lxpr_node_t *, lxpr_uiobuf_t *);
+static void lxpr_read_isdir(lxpr_node_t *, lxpr_uiobuf_t *);
+static void lxpr_read_fd(lxpr_node_t *, lxpr_uiobuf_t *);
+static void lxpr_read_kmsg(lxpr_node_t *, lxpr_uiobuf_t *, ldi_handle_t);
+static void lxpr_read_loadavg(lxpr_node_t *, lxpr_uiobuf_t *);
+static void lxpr_read_meminfo(lxpr_node_t *, lxpr_uiobuf_t *);
+static void lxpr_read_mounts(lxpr_node_t *, lxpr_uiobuf_t *);
+static void lxpr_read_partitions(lxpr_node_t *, lxpr_uiobuf_t *);
+static void lxpr_read_stat(lxpr_node_t *, lxpr_uiobuf_t *);
+static void lxpr_read_uptime(lxpr_node_t *, lxpr_uiobuf_t *);
+static void lxpr_read_version(lxpr_node_t *, lxpr_uiobuf_t *);
+
+static void lxpr_read_pid_cmdline(lxpr_node_t *, lxpr_uiobuf_t *);
+static void lxpr_read_pid_maps(lxpr_node_t *, lxpr_uiobuf_t *);
+static void lxpr_read_pid_stat(lxpr_node_t *, lxpr_uiobuf_t *);
+static void lxpr_read_pid_statm(lxpr_node_t *, lxpr_uiobuf_t *);
+static void lxpr_read_pid_status(lxpr_node_t *, lxpr_uiobuf_t *);
+
+static void lxpr_read_net_arp(lxpr_node_t *, lxpr_uiobuf_t *);
+static void lxpr_read_net_dev(lxpr_node_t *, lxpr_uiobuf_t *);
+static void lxpr_read_net_dev_mcast(lxpr_node_t *, lxpr_uiobuf_t *);
+static void lxpr_read_net_igmp(lxpr_node_t *, lxpr_uiobuf_t *);
+static void lxpr_read_net_ip_mr_cache(lxpr_node_t *, lxpr_uiobuf_t *);
+static void lxpr_read_net_ip_mr_vif(lxpr_node_t *, lxpr_uiobuf_t *);
+static void lxpr_read_net_mcfilter(lxpr_node_t *, lxpr_uiobuf_t *);
+static void lxpr_read_net_netstat(lxpr_node_t *, lxpr_uiobuf_t *);
+static void lxpr_read_net_raw(lxpr_node_t *, lxpr_uiobuf_t *);
+static void lxpr_read_net_route(lxpr_node_t *, lxpr_uiobuf_t *);
+static void lxpr_read_net_rpc(lxpr_node_t *, lxpr_uiobuf_t *);
+static void lxpr_read_net_rt_cache(lxpr_node_t *, lxpr_uiobuf_t *);
+static void lxpr_read_net_sockstat(lxpr_node_t *, lxpr_uiobuf_t *);
+static void lxpr_read_net_snmp(lxpr_node_t *, lxpr_uiobuf_t *);
+static void lxpr_read_net_stat(lxpr_node_t *, lxpr_uiobuf_t *);
+static void lxpr_read_net_tcp(lxpr_node_t *, lxpr_uiobuf_t *);
+static void lxpr_read_net_udp(lxpr_node_t *, lxpr_uiobuf_t *);
+static void lxpr_read_net_unix(lxpr_node_t *, lxpr_uiobuf_t *);
+
+/*
+ * Simple conversion
+ */
+#define	btok(x)	((x) >> 10)			/* bytes to kbytes */
+#define	ptok(x)	((x) << (PAGESHIFT - 10))	/* pages to kbytes */
+
+/*
+ * The lxproc vnode operations vector
+ */
+const fs_operation_def_t lxpr_vnodeops_template[] = {
+	VOPNAME_OPEN,		{ .vop_open = lxpr_open },
+	VOPNAME_CLOSE,		{ .vop_close = lxpr_close },
+	VOPNAME_READ,		{ .vop_read = lxpr_read },
+	VOPNAME_GETATTR,	{ .vop_getattr = lxpr_getattr },
+	VOPNAME_ACCESS,		{ .vop_access = lxpr_access },
+	VOPNAME_LOOKUP,		{ .vop_lookup = lxpr_lookup },
+	VOPNAME_READDIR,	{ .vop_readdir = lxpr_readdir },
+	VOPNAME_READLINK,	{ .vop_readlink = lxpr_readlink },
+	VOPNAME_FSYNC,		{ .error = lxpr_sync },
+	VOPNAME_SEEK,		{ .error = lxpr_sync },
+	VOPNAME_INACTIVE,	{ .vop_inactive = lxpr_inactive },
+	VOPNAME_CMP,		{ .vop_cmp = lxpr_cmp },
+	VOPNAME_REALVP,		{ .vop_realvp = lxpr_realvp },
+	NULL,			NULL
+};
+
+/*
+ * file contents of an lxproc directory.
+ */
+static lxpr_dirent_t lxpr_dir[] = {
+	{ LXPR_CMDLINE,		"cmdline" },
+	{ LXPR_CPUINFO,		"cpuinfo" },
+	{ LXPR_DEVICES,		"devices" },
+	{ LXPR_DMA,		"dma" },
+	{ LXPR_FILESYSTEMS,	"filesystems" },
+	{ LXPR_INTERRUPTS,	"interrupts" },
+	{ LXPR_IOPORTS,		"ioports" },
+	{ LXPR_KCORE,		"kcore" },
+	{ LXPR_KMSG,		"kmsg" },
+	{ LXPR_LOADAVG,		"loadavg" },
+	{ LXPR_MEMINFO,		"meminfo" },
+	{ LXPR_MOUNTS,		"mounts" },
+	{ LXPR_NETDIR,		"net" },
+	{ LXPR_PARTITIONS,	"partitions" },
+	{ LXPR_SELF,		"self" },
+	{ LXPR_STAT,		"stat" },
+	{ LXPR_UPTIME,		"uptime" },
+	{ LXPR_VERSION,		"version" }
+};
+
+#define	PROCDIRFILES	(sizeof (lxpr_dir) / sizeof (lxpr_dir[0]))
+
+/*
+ * Contents of an /lxproc/<pid> directory.
+ */
+static lxpr_dirent_t piddir[] = {
+	{ LXPR_PID_CMDLINE,	"cmdline" },
+	{ LXPR_PID_CPU,		"cpu" },
+	{ LXPR_PID_CURDIR,	"cwd" },
+	{ LXPR_PID_ENV,		"environ" },
+	{ LXPR_PID_EXE,		"exe" },
+	{ LXPR_PID_MAPS,	"maps" },
+	{ LXPR_PID_MEM,		"mem" },
+	{ LXPR_PID_ROOTDIR,	"root" },
+	{ LXPR_PID_STAT,	"stat" },
+	{ LXPR_PID_STATM,	"statm" },
+	{ LXPR_PID_STATUS,	"status" },
+	{ LXPR_PID_FDDIR,	"fd" }
+};
+
+#define	PIDDIRFILES	(sizeof (piddir) / sizeof (piddir[0]))
+
+/*
+ * contents of /lxproc/net directory
+ */
+static lxpr_dirent_t netdir[] = {
+	{ LXPR_NET_ARP,		"arp" },
+	{ LXPR_NET_DEV,		"dev" },
+	{ LXPR_NET_DEV_MCAST,	"dev_mcast" },
+	{ LXPR_NET_IGMP,	"igmp" },
+	{ LXPR_NET_IP_MR_CACHE,	"ip_mr_cache" },
+	{ LXPR_NET_IP_MR_VIF,	"ip_mr_vif" },
+	{ LXPR_NET_MCFILTER,	"mcfilter" },
+	{ LXPR_NET_NETSTAT,	"netstat" },
+	{ LXPR_NET_RAW,		"raw" },
+	{ LXPR_NET_ROUTE,	"route" },
+	{ LXPR_NET_RPC,		"rpc" },
+	{ LXPR_NET_RT_CACHE,	"rt_cache" },
+	{ LXPR_NET_SOCKSTAT,	"sockstat" },
+	{ LXPR_NET_SNMP,	"snmp" },
+	{ LXPR_NET_STAT,	"stat" },
+	{ LXPR_NET_TCP,		"tcp" },
+	{ LXPR_NET_UDP,		"udp" },
+	{ LXPR_NET_UNIX,	"unix" }
+};
+
+#define	NETDIRFILES	(sizeof (netdir) / sizeof (netdir[0]))
+
+/*
+ * These are the major signal number differences between Linux and native:
+ *
+ * 	====================================
+ * 	| Number | Linux      | Native     |
+ * 	| ====== | =========  | ========== |
+ *	|    7   | SIGBUS     | SIGEMT     |
+ *	|   10   | SIGUSR1    | SIGBUS     |
+ *	|   12   | SIGUSR2    | SIGSYS     |
+ *	|   16   | SIGSTKFLT  | SIGUSR1    |
+ *	|   17   | SIGCHLD    | SIGUSR2    |
+ * 	|   18   | SIGCONT    | SIGCHLD    |
+ *	|   19   | SIGSTOP    | SIGPWR     |
+ * 	|   20   | SIGTSTP    | SIGWINCH   |
+ * 	|   21   | SIGTTIN    | SIGURG     |
+ * 	|   22   | SIGTTOU    | SIGPOLL    |
+ *	|   23   | SIGURG     | SIGSTOP    |
+ * 	|   24   | SIGXCPU    | SIGTSTP    |
+ *	|   25   | SIGXFSZ    | SIGCONT    |
+ *	|   26   | SIGVTALARM | SIGTTIN    |
+ *	|   27   | SIGPROF    | SIGTTOU    |
+ *	|   28   | SIGWINCH   | SIGVTALARM |
+ *	|   29   | SIGPOLL    | SIGPROF    |
+ *	|   30   | SIGPWR     | SIGXCPU    |
+ *	|   31   | SIGSYS     | SIGXFSZ    |
+ * 	====================================
+ *
+ * Not every Linux signal maps to a native signal, nor does every native
+ * signal map to a Linux counterpart. However, when signals do map, the
+ * mapping is unique.
+ */
+static int
+lxpr_sigmap[NSIG] = {
+	0,
+	LX_SIGHUP,
+	LX_SIGINT,
+	LX_SIGQUIT,
+	LX_SIGILL,
+	LX_SIGTRAP,
+	LX_SIGABRT,
+	LX_SIGSTKFLT,
+	LX_SIGFPE,
+	LX_SIGKILL,
+	LX_SIGBUS,
+	LX_SIGSEGV,
+	LX_SIGSYS,
+	LX_SIGPIPE,
+	LX_SIGALRM,
+	LX_SIGTERM,
+	LX_SIGUSR1,
+	LX_SIGUSR2,
+	LX_SIGCHLD,
+	LX_SIGPWR,
+	LX_SIGWINCH,
+	LX_SIGURG,
+	LX_SIGPOLL,
+	LX_SIGSTOP,
+	LX_SIGTSTP,
+	LX_SIGCONT,
+	LX_SIGTTIN,
+	LX_SIGTTOU,
+	LX_SIGVTALRM,
+	LX_SIGPROF,
+	LX_SIGXCPU,
+	LX_SIGXFSZ,
+	-1,			/* 32:  illumos SIGWAITING */
+	-1,			/* 33:  illumos SIGLWP */
+	-1,			/* 34:  illumos SIGFREEZE */
+	-1,			/* 35:  illumos SIGTHAW */
+	-1,			/* 36:  illumos SIGCANCEL */
+	-1,			/* 37:  illumos SIGLOST */
+	-1,			/* 38:  illumos SIGXRES */
+	-1,			/* 39:  illumos SIGJVM1 */
+	-1,			/* 40:  illumos SIGJVM2 */
+	-1,			/* 41:  illumos SIGINFO */
+	LX_SIGRTMIN,		/* 42:  illumos _SIGRTMIN */
+	LX_SIGRTMIN + 1,
+	LX_SIGRTMIN + 2,
+	LX_SIGRTMIN + 3,
+	LX_SIGRTMIN + 4,
+	LX_SIGRTMIN + 5,
+	LX_SIGRTMIN + 6,
+	LX_SIGRTMIN + 7,
+	LX_SIGRTMIN + 8,
+	LX_SIGRTMIN + 9,
+	LX_SIGRTMIN + 10,
+	LX_SIGRTMIN + 11,
+	LX_SIGRTMIN + 12,
+	LX_SIGRTMIN + 13,
+	LX_SIGRTMIN + 14,
+	LX_SIGRTMIN + 15,
+	LX_SIGRTMIN + 16,
+	LX_SIGRTMIN + 17,
+	LX_SIGRTMIN + 18,
+	LX_SIGRTMIN + 19,
+	LX_SIGRTMIN + 20,
+	LX_SIGRTMIN + 21,
+	LX_SIGRTMIN + 22,
+	LX_SIGRTMIN + 23,
+	LX_SIGRTMIN + 24,
+	LX_SIGRTMIN + 25,
+	LX_SIGRTMIN + 26,
+	LX_SIGRTMIN + 27,
+	LX_SIGRTMIN + 28,
+	LX_SIGRTMIN + 29,
+	LX_SIGRTMIN + 30,
+	LX_SIGRTMAX
+};
+
+/*
+ * lxpr_open(): Vnode operation for VOP_OPEN()
+ */
+static int
+lxpr_open(vnode_t **vpp, int flag, cred_t *cr, caller_context_t *ct)
+{
+	vnode_t		*vp = *vpp;
+	lxpr_node_t	*lxpnp = VTOLXP(vp);
+	lxpr_nodetype_t	type = lxpnp->lxpr_type;
+	vnode_t		*rvp;
+	int		error = 0;
+
+	/*
+	 * We only allow reading in this file systrem
+	 */
+	if (flag & FWRITE)
+		return (EROFS);
+
+	/*
+	 * If we are opening an underlying file only allow regular files
+	 * reject the open for anything but a regular file.
+	 * Just do it if we are opening the current or root directory.
+	 */
+	if (lxpnp->lxpr_realvp != NULL) {
+		rvp = lxpnp->lxpr_realvp;
+
+		if (type == LXPR_PID_FD_FD && rvp->v_type != VREG)
+			error = EACCES;
+		else {
+			/*
+			 * Need to hold rvp since VOP_OPEN() may release it.
+			 */
+			VN_HOLD(rvp);
+			error = VOP_OPEN(&rvp, flag, cr, ct);
+			if (error) {
+				VN_RELE(rvp);
+			} else {
+				*vpp = rvp;
+				VN_RELE(vp);
+			}
+		}
+	}
+
+	return (error);
+}
+
+
+/*
+ * lxpr_close(): Vnode operation for VOP_CLOSE()
+ */
+/* ARGSUSED */
+static int
+lxpr_close(vnode_t *vp, int flag, int count, offset_t offset, cred_t *cr,
+    caller_context_t *ct)
+{
+	lxpr_node_t	*lxpr = VTOLXP(vp);
+	lxpr_nodetype_t	type = lxpr->lxpr_type;
+
+	/*
+	 * we should never get here because the close is done on the realvp
+	 * for these nodes
+	 */
+	ASSERT(type != LXPR_PID_FD_FD &&
+	    type != LXPR_PID_CURDIR &&
+	    type != LXPR_PID_ROOTDIR &&
+	    type != LXPR_PID_EXE);
+
+	return (0);
+}
+
+static void (*lxpr_read_function[LXPR_NFILES])() = {
+	lxpr_read_isdir,		/* /proc		*/
+	lxpr_read_isdir,		/* /proc/<pid>		*/
+	lxpr_read_pid_cmdline,		/* /proc/<pid>/cmdline	*/
+	lxpr_read_empty,		/* /proc/<pid>/cpu	*/
+	lxpr_read_invalid,		/* /proc/<pid>/cwd	*/
+	lxpr_read_empty,		/* /proc/<pid>/environ	*/
+	lxpr_read_invalid,		/* /proc/<pid>/exe	*/
+	lxpr_read_pid_maps,		/* /proc/<pid>/maps	*/
+	lxpr_read_empty,		/* /proc/<pid>/mem	*/
+	lxpr_read_invalid,		/* /proc/<pid>/root	*/
+	lxpr_read_pid_stat,		/* /proc/<pid>/stat	*/
+	lxpr_read_pid_statm,		/* /proc/<pid>/statm	*/
+	lxpr_read_pid_status,		/* /proc/<pid>/status	*/
+	lxpr_read_isdir,		/* /proc/<pid>/fd	*/
+	lxpr_read_fd,			/* /proc/<pid>/fd/nn	*/
+	lxpr_read_empty,		/* /proc/cmdline	*/
+	lxpr_read_cpuinfo,		/* /proc/cpuinfo	*/
+	lxpr_read_empty,		/* /proc/devices	*/
+	lxpr_read_empty,		/* /proc/dma		*/
+	lxpr_read_empty,		/* /proc/filesystems	*/
+	lxpr_read_empty,		/* /proc/interrupts	*/
+	lxpr_read_empty,		/* /proc/ioports	*/
+	lxpr_read_empty,		/* /proc/kcore		*/
+	lxpr_read_invalid,		/* /proc/kmsg -- see lxpr_read() */
+	lxpr_read_loadavg,		/* /proc/loadavg	*/
+	lxpr_read_meminfo,		/* /proc/meminfo	*/
+	lxpr_read_mounts,		/* /proc/mounts		*/
+	lxpr_read_isdir,		/* /proc/net		*/
+	lxpr_read_net_arp,		/* /proc/net/arp	*/
+	lxpr_read_net_dev,		/* /proc/net/dev	*/
+	lxpr_read_net_dev_mcast,	/* /proc/net/dev_mcast	*/
+	lxpr_read_net_igmp,		/* /proc/net/igmp	*/
+	lxpr_read_net_ip_mr_cache,	/* /proc/net/ip_mr_cache */
+	lxpr_read_net_ip_mr_vif,	/* /proc/net/ip_mr_vif	*/
+	lxpr_read_net_mcfilter,		/* /proc/net/mcfilter	*/
+	lxpr_read_net_netstat,		/* /proc/net/netstat	*/
+	lxpr_read_net_raw,		/* /proc/net/raw	*/
+	lxpr_read_net_route,		/* /proc/net/route	*/
+	lxpr_read_net_rpc,		/* /proc/net/rpc	*/
+	lxpr_read_net_rt_cache,		/* /proc/net/rt_cache	*/
+	lxpr_read_net_sockstat,		/* /proc/net/sockstat	*/
+	lxpr_read_net_snmp,		/* /proc/net/snmp	*/
+	lxpr_read_net_stat,		/* /proc/net/stat	*/
+	lxpr_read_net_tcp,		/* /proc/net/tcp	*/
+	lxpr_read_net_udp,		/* /proc/net/udp	*/
+	lxpr_read_net_unix,		/* /proc/net/unix	*/
+	lxpr_read_partitions,		/* /proc/partitions	*/
+	lxpr_read_invalid,		/* /proc/self		*/
+	lxpr_read_stat,			/* /proc/stat		*/
+	lxpr_read_uptime,		/* /proc/uptime		*/
+	lxpr_read_version,		/* /proc/version	*/
+};
+
+/*
+ * Array of lookup functions, indexed by /lxproc file type.
+ */
+static vnode_t *(*lxpr_lookup_function[LXPR_NFILES])() = {
+	lxpr_lookup_procdir,		/* /proc		*/
+	lxpr_lookup_piddir,		/* /proc/<pid>		*/
+	lxpr_lookup_not_a_dir,		/* /proc/<pid>/cmdline	*/
+	lxpr_lookup_not_a_dir,		/* /proc/<pid>/cpu	*/
+	lxpr_lookup_not_a_dir,		/* /proc/<pid>/cwd	*/
+	lxpr_lookup_not_a_dir,		/* /proc/<pid>/environ	*/
+	lxpr_lookup_not_a_dir,		/* /proc/<pid>/exe	*/
+	lxpr_lookup_not_a_dir,		/* /proc/<pid>/maps	*/
+	lxpr_lookup_not_a_dir,		/* /proc/<pid>/mem	*/
+	lxpr_lookup_not_a_dir,		/* /proc/<pid>/root	*/
+	lxpr_lookup_not_a_dir,		/* /proc/<pid>/stat	*/
+	lxpr_lookup_not_a_dir,		/* /proc/<pid>/statm	*/
+	lxpr_lookup_not_a_dir,		/* /proc/<pid>/status	*/
+	lxpr_lookup_fddir,		/* /proc/<pid>/fd	*/
+	lxpr_lookup_not_a_dir,		/* /proc/<pid>/fd/nn	*/
+	lxpr_lookup_not_a_dir,		/* /proc/cmdline	*/
+	lxpr_lookup_not_a_dir,		/* /proc/cpuinfo	*/
+	lxpr_lookup_not_a_dir,		/* /proc/devices	*/
+	lxpr_lookup_not_a_dir,		/* /proc/dma		*/
+	lxpr_lookup_not_a_dir,		/* /proc/filesystems	*/
+	lxpr_lookup_not_a_dir,		/* /proc/interrupts	*/
+	lxpr_lookup_not_a_dir,		/* /proc/ioports	*/
+	lxpr_lookup_not_a_dir,		/* /proc/kcore		*/
+	lxpr_lookup_not_a_dir,		/* /proc/kmsg		*/
+	lxpr_lookup_not_a_dir,		/* /proc/loadavg	*/
+	lxpr_lookup_not_a_dir,		/* /proc/meminfo	*/
+	lxpr_lookup_not_a_dir,		/* /proc/mounts		*/
+	lxpr_lookup_netdir,		/* /proc/net		*/
+	lxpr_lookup_not_a_dir,		/* /proc/net/arp	*/
+	lxpr_lookup_not_a_dir,		/* /proc/net/dev	*/
+	lxpr_lookup_not_a_dir,		/* /proc/net/dev_mcast	*/
+	lxpr_lookup_not_a_dir,		/* /proc/net/igmp	*/
+	lxpr_lookup_not_a_dir,		/* /proc/net/ip_mr_cache */
+	lxpr_lookup_not_a_dir,		/* /proc/net/ip_mr_vif	*/
+	lxpr_lookup_not_a_dir,		/* /proc/net/mcfilter	*/
+	lxpr_lookup_not_a_dir,		/* /proc/net/netstat	*/
+	lxpr_lookup_not_a_dir,		/* /proc/net/raw	*/
+	lxpr_lookup_not_a_dir,		/* /proc/net/route	*/
+	lxpr_lookup_not_a_dir,		/* /proc/net/rpc	*/
+	lxpr_lookup_not_a_dir,		/* /proc/net/rt_cache	*/
+	lxpr_lookup_not_a_dir,		/* /proc/net/sockstat	*/
+	lxpr_lookup_not_a_dir,		/* /proc/net/snmp	*/
+	lxpr_lookup_not_a_dir,		/* /proc/net/stat	*/
+	lxpr_lookup_not_a_dir,		/* /proc/net/tcp	*/
+	lxpr_lookup_not_a_dir,		/* /proc/net/udp	*/
+	lxpr_lookup_not_a_dir,		/* /proc/net/unix	*/
+	lxpr_lookup_not_a_dir,		/* /proc/partitions	*/
+	lxpr_lookup_not_a_dir,		/* /proc/self		*/
+	lxpr_lookup_not_a_dir,		/* /proc/stat		*/
+	lxpr_lookup_not_a_dir,		/* /proc/uptime		*/
+	lxpr_lookup_not_a_dir,		/* /proc/version	*/
+};
+
+/*
+ * Array of readdir functions, indexed by /proc file type.
+ */
+static int (*lxpr_readdir_function[LXPR_NFILES])() = {
+	lxpr_readdir_procdir,		/* /proc		*/
+	lxpr_readdir_piddir,		/* /proc/<pid>		*/
+	lxpr_readdir_not_a_dir,		/* /proc/<pid>/cmdline	*/
+	lxpr_readdir_not_a_dir,		/* /proc/<pid>/cpu	*/
+	lxpr_readdir_not_a_dir,		/* /proc/<pid>/cwd	*/
+	lxpr_readdir_not_a_dir,		/* /proc/<pid>/environ	*/
+	lxpr_readdir_not_a_dir,		/* /proc/<pid>/exe	*/
+	lxpr_readdir_not_a_dir,		/* /proc/<pid>/maps	*/
+	lxpr_readdir_not_a_dir,		/* /proc/<pid>/mem	*/
+	lxpr_readdir_not_a_dir,		/* /proc/<pid>/root	*/
+	lxpr_readdir_not_a_dir,		/* /proc/<pid>/stat	*/
+	lxpr_readdir_not_a_dir,		/* /proc/<pid>/statm	*/
+	lxpr_readdir_not_a_dir,		/* /proc/<pid>/status	*/
+	lxpr_readdir_fddir,		/* /proc/<pid>/fd	*/
+	lxpr_readdir_not_a_dir,		/* /proc/<pid>/fd/nn	*/
+	lxpr_readdir_not_a_dir,		/* /proc/cmdline	*/
+	lxpr_readdir_not_a_dir,		/* /proc/cpuinfo	*/
+	lxpr_readdir_not_a_dir,		/* /proc/devices	*/
+	lxpr_readdir_not_a_dir,		/* /proc/dma		*/
+	lxpr_readdir_not_a_dir,		/* /proc/filesystems	*/
+	lxpr_readdir_not_a_dir,		/* /proc/interrupts	*/
+	lxpr_readdir_not_a_dir,		/* /proc/ioports	*/
+	lxpr_readdir_not_a_dir,		/* /proc/kcore		*/
+	lxpr_readdir_not_a_dir,		/* /proc/kmsg		*/
+	lxpr_readdir_not_a_dir,		/* /proc/loadavg	*/
+	lxpr_readdir_not_a_dir,		/* /proc/meminfo	*/
+	lxpr_readdir_not_a_dir,		/* /proc/mounts		*/
+	lxpr_readdir_netdir,		/* /proc/net		*/
+	lxpr_readdir_not_a_dir,		/* /proc/net/arp	*/
+	lxpr_readdir_not_a_dir,		/* /proc/net/dev	*/
+	lxpr_readdir_not_a_dir,		/* /proc/net/dev_mcast	*/
+	lxpr_readdir_not_a_dir,		/* /proc/net/igmp	*/
+	lxpr_readdir_not_a_dir,		/* /proc/net/ip_mr_cache */
+	lxpr_readdir_not_a_dir,		/* /proc/net/ip_mr_vif	*/
+	lxpr_readdir_not_a_dir,		/* /proc/net/mcfilter	*/
+	lxpr_readdir_not_a_dir,		/* /proc/net/netstat	*/
+	lxpr_readdir_not_a_dir,		/* /proc/net/raw	*/
+	lxpr_readdir_not_a_dir,		/* /proc/net/route	*/
+	lxpr_readdir_not_a_dir,		/* /proc/net/rpc	*/
+	lxpr_readdir_not_a_dir,		/* /proc/net/rt_cache	*/
+	lxpr_readdir_not_a_dir,		/* /proc/net/sockstat	*/
+	lxpr_readdir_not_a_dir,		/* /proc/net/snmp	*/
+	lxpr_readdir_not_a_dir,		/* /proc/net/stat	*/
+	lxpr_readdir_not_a_dir,		/* /proc/net/tcp	*/
+	lxpr_readdir_not_a_dir,		/* /proc/net/udp	*/
+	lxpr_readdir_not_a_dir,		/* /proc/net/unix	*/
+	lxpr_readdir_not_a_dir,		/* /proc/partitions	*/
+	lxpr_readdir_not_a_dir,		/* /proc/self		*/
+	lxpr_readdir_not_a_dir,		/* /proc/stat		*/
+	lxpr_readdir_not_a_dir,		/* /proc/uptime		*/
+	lxpr_readdir_not_a_dir,		/* /proc/version	*/
+};
+
+
+/*
+ * lxpr_read(): Vnode operation for VOP_READ()
+ *
+ * As the format of all the files that can be read in lxproc is human readable
+ * and not binary structures there do not have to be different read variants
+ * depending on whether the reading process model is 32- or 64-bit.
+ */
+/* ARGSUSED */
+static int
+lxpr_read(vnode_t *vp, uio_t *uiop, int ioflag, cred_t *cr,
+    caller_context_t *ct)
+{
+	lxpr_node_t *lxpnp = VTOLXP(vp);
+	lxpr_nodetype_t type = lxpnp->lxpr_type;
+	lxpr_uiobuf_t *uiobuf = lxpr_uiobuf_new(uiop);
+	int error;
+
+	ASSERT(type < LXPR_NFILES);
+
+	if (type == LXPR_KMSG) {
+		ldi_ident_t	li = VTOLXPM(vp)->lxprm_li;
+		ldi_handle_t	ldih;
+		struct strioctl	str;
+		int		rv;
+
+		/*
+		 * Open the zone's console device using the layered driver
+		 * interface.
+		 */
+		if ((error =
+		    ldi_open_by_name("/dev/log", FREAD, cr, &ldih, li)) != 0)
+			return (error);
+
+		/*
+		 * Send an ioctl to the underlying console device, letting it
+		 * know we're interested in getting console messages.
+		 */
+		str.ic_cmd = I_CONSLOG;
+		str.ic_timout = 0;
+		str.ic_len = 0;
+		str.ic_dp = NULL;
+		if ((error = ldi_ioctl(ldih, I_STR,
+		    (intptr_t)&str, FKIOCTL, cr, &rv)) != 0)
+			return (error);
+
+		lxpr_read_kmsg(lxpnp, uiobuf, ldih);
+
+		if ((error = ldi_close(ldih, FREAD, cr)) != 0)
+			return (error);
+	} else {
+		lxpr_read_function[type](lxpnp, uiobuf);
+	}
+
+	error = lxpr_uiobuf_flush(uiobuf);
+	lxpr_uiobuf_free(uiobuf);
+
+	return (error);
+}
+
+/*
+ * lxpr_read_invalid(), lxpr_read_isdir(), lxpr_read_empty()
+ *
+ * Various special case reads:
+ * - trying to read a directory
+ * - invalid file (used to mean a file that should be implemented,
+ *   but isn't yet)
+ * - empty file
+ * - wait to be able to read a file that will never have anything to read
+ */
+/* ARGSUSED */
+static void
+lxpr_read_isdir(lxpr_node_t *lxpnp, lxpr_uiobuf_t *uiobuf)
+{
+	lxpr_uiobuf_seterr(uiobuf, EISDIR);
+}
+
+/* ARGSUSED */
+static void
+lxpr_read_invalid(lxpr_node_t *lxpnp, lxpr_uiobuf_t *uiobuf)
+{
+	lxpr_uiobuf_seterr(uiobuf, EINVAL);
+}
+
+/* ARGSUSED */
+static void
+lxpr_read_empty(lxpr_node_t *lxpnp, lxpr_uiobuf_t *uiobuf)
+{
+}
+
+/*
+ * lxpr_read_pid_cmdline():
+ *
+ * This is not precisely compatible with Linux: the Linux cmdline returns argv
+ * with the correct separation using \0 between the arguments, but we cannot do
+ * that without copying the real argv from the correct process context.  This
+ * is too difficult to attempt so we pretend that the entire cmdline is just
+ * argv[0]. This is good enough for ps and htop to display correctly, but might
+ * cause some other things not to work correctly.
+ */
+static void
+lxpr_read_pid_cmdline(lxpr_node_t *lxpnp, lxpr_uiobuf_t *uiobuf)
+{
+	proc_t *p;
+	char *buf;
+
+	ASSERT(lxpnp->lxpr_type == LXPR_PID_CMDLINE);
+
+	p = lxpr_lock(lxpnp->lxpr_pid);
+	if (p == NULL) {
+		lxpr_uiobuf_seterr(uiobuf, EINVAL);
+		return;
+	}
+
+	buf = PTOU(p)->u_argv != 0 ? PTOU(p)->u_psargs : PTOU(p)->u_comm;
+
+	lxpr_uiobuf_write(uiobuf, buf, strlen(buf) + 1);
+	lxpr_unlock(p);
+}
+
+/*
+ * lxpr_read_pid_maps(): memory map file
+ */
+static void
+lxpr_read_pid_maps(lxpr_node_t *lxpnp, lxpr_uiobuf_t *uiobuf)
+{
+	proc_t *p;
+	struct as *as;
+	struct seg *seg;
+	char *buf;
+	int buflen = MAXPATHLEN;
+	struct print_data {
+		caddr_t saddr;
+		caddr_t eaddr;
+		int type;
+		char prot[5];
+		uint32_t offset;
+		vnode_t *vp;
+		struct print_data *next;
+	} *print_head = NULL;
+	struct print_data **print_tail = &print_head;
+	struct print_data *pbuf;
+
+	ASSERT(lxpnp->lxpr_type == LXPR_PID_MAPS);
+
+	p = lxpr_lock(lxpnp->lxpr_pid);
+	if (p == NULL) {
+		lxpr_uiobuf_seterr(uiobuf, EINVAL);
+		return;
+	}
+
+	as = p->p_as;
+
+	if (as == &kas) {
+		lxpr_unlock(p);
+		return;
+	}
+
+	mutex_exit(&p->p_lock);
+
+	/* Iterate over all segments in the address space */
+	AS_LOCK_ENTER(as, &as->a_lock, RW_READER);
+	for (seg = AS_SEGFIRST(as); seg != NULL; seg = AS_SEGNEXT(as, seg)) {
+		vnode_t *vp;
+		uint_t protbits;
+
+		pbuf = kmem_alloc(sizeof (*pbuf), KM_SLEEP);
+
+		pbuf->saddr = seg->s_base;
+		pbuf->eaddr = seg->s_base+seg->s_size;
+		pbuf->type = SEGOP_GETTYPE(seg, seg->s_base);
+
+		/*
+		 * Cheat and only use the protection bits of the first page
+		 * in the segment
+		 */
+		(void) strncpy(pbuf->prot, "----", sizeof (pbuf->prot));
+		(void) SEGOP_GETPROT(seg, seg->s_base, 0, &protbits);
+
+		if (protbits & PROT_READ)	   pbuf->prot[0] = 'r';
+		if (protbits & PROT_WRITE)	   pbuf->prot[1] = 'w';
+		if (protbits & PROT_EXEC)	   pbuf->prot[2] = 'x';
+		if (pbuf->type & MAP_SHARED)	   pbuf->prot[3] = 's';
+		else if (pbuf->type & MAP_PRIVATE) pbuf->prot[3] = 'p';
+
+		if (seg->s_ops == &segvn_ops &&
+		    SEGOP_GETVP(seg, seg->s_base, &vp) == 0 &&
+		    vp != NULL && vp->v_type == VREG) {
+			VN_HOLD(vp);
+			pbuf->vp = vp;
+		} else {
+			pbuf->vp = NULL;
+		}
+
+		pbuf->offset = (uint32_t)SEGOP_GETOFFSET(seg, pbuf->saddr);
+
+		pbuf->next = NULL;
+		*print_tail = pbuf;
+		print_tail = &pbuf->next;
+	}
+	AS_LOCK_EXIT(as, &as->a_lock);
+	mutex_enter(&p->p_lock);
+	lxpr_unlock(p);
+
+	buf = kmem_alloc(buflen, KM_SLEEP);
+
+	/* print the data we've extracted */
+	pbuf = print_head;
+	while (pbuf != NULL) {
+		struct print_data *pbuf_next;
+		vattr_t vattr;
+
+		int maj = 0;
+		int min = 0;
+		u_longlong_t inode = 0;
+
+		*buf = '\0';
+		if (pbuf->vp != NULL) {
+			vattr.va_mask = AT_FSID | AT_NODEID;
+			if (VOP_GETATTR(pbuf->vp, &vattr, 0, CRED(),
+			    NULL) == 0) {
+				maj = getmajor(vattr.va_fsid);
+				min = getminor(vattr.va_fsid);
+				inode = vattr.va_nodeid;
+			}
+			(void) vnodetopath(NULL, pbuf->vp, buf, buflen, CRED());
+			VN_RELE(pbuf->vp);
+		}
+
+		if (*buf != '\0') {
+			lxpr_uiobuf_printf(uiobuf,
+			    "%08x-%08x %s %08x %02d:%03d %lld %s\n",
+			    pbuf->saddr, pbuf->eaddr, pbuf->prot, pbuf->offset,
+			    maj, min, inode, buf);
+		} else {
+			lxpr_uiobuf_printf(uiobuf,
+			    "%08x-%08x %s %08x %02d:%03d %lld\n",
+			    pbuf->saddr, pbuf->eaddr, pbuf->prot, pbuf->offset,
+			    maj, min, inode);
+		}
+
+		pbuf_next = pbuf->next;
+		kmem_free(pbuf, sizeof (*pbuf));
+		pbuf = pbuf_next;
+	}
+
+	kmem_free(buf, buflen);
+}
+
+/*
+ * lxpr_read_pid_statm(): memory status file
+ */
+static void
+lxpr_read_pid_statm(lxpr_node_t *lxpnp, lxpr_uiobuf_t *uiobuf)
+{
+	proc_t *p;
+	struct as *as;
+	size_t vsize;
+	size_t rss;
+
+	ASSERT(lxpnp->lxpr_type == LXPR_PID_STATM);
+
+	p = lxpr_lock(lxpnp->lxpr_pid);
+	if (p == NULL) {
+		lxpr_uiobuf_seterr(uiobuf, EINVAL);
+		return;
+	}
+
+	as = p->p_as;
+
+	mutex_exit(&p->p_lock);
+
+	AS_LOCK_ENTER(as, &as->a_lock, RW_READER);
+	vsize = btopr(as->a_resvsize);
+	rss = rm_asrss(as);
+	AS_LOCK_EXIT(as, &as->a_lock);
+
+	mutex_enter(&p->p_lock);
+	lxpr_unlock(p);
+
+	lxpr_uiobuf_printf(uiobuf,
+	    "%lu %lu %lu %lu %lu %lu %lu\n",
+	    vsize, rss, 0l, rss, 0l, 0l, 0l);
+}
+
+/*
+ * lxpr_read_pid_status(): status file
+ */
+static void
+lxpr_read_pid_status(lxpr_node_t *lxpnp, lxpr_uiobuf_t *uiobuf)
+{
+	proc_t *p;
+	kthread_t *t;
+	user_t *up;
+	cred_t *cr;
+	const gid_t *groups;
+	int    ngroups;
+	struct as *as;
+	char *status;
+	pid_t pid, ppid;
+	size_t vsize;
+	size_t rss;
+	k_sigset_t current, ignore, handle;
+	int    i, lx_sig;
+
+	ASSERT(lxpnp->lxpr_type == LXPR_PID_STATUS);
+
+	p = lxpr_lock(lxpnp->lxpr_pid);
+	if (p == NULL) {
+		lxpr_uiobuf_seterr(uiobuf, EINVAL);
+		return;
+	}
+
+	pid = p->p_pid;
+
+	/*
+	 * Convert pid to the Linux default of 1 if we're the zone's init
+	 * process
+	 */
+	if (pid == curproc->p_zone->zone_proc_initpid) {
+		pid = 1;
+		ppid = 0;	/* parent pid for init is 0 */
+	} else {
+		/*
+		 * Make sure not to reference parent PIDs that reside outside
+		 * the zone
+		 */
+		ppid = ((p->p_flag & SZONETOP)
+		    ? curproc->p_zone->zone_zsched->p_pid : p->p_ppid);
+
+		/*
+		 * Convert ppid to the Linux default of 1 if our parent is the
+		 * zone's init process
+		 */
+		if (ppid == curproc->p_zone->zone_proc_initpid)
+			ppid = 1;
+	}
+
+	t = prchoose(p);
+	if (t != NULL) {
+		switch (t->t_state) {
+		case TS_SLEEP:
+			status = "S (sleeping)";
+			break;
+		case TS_RUN:
+		case TS_ONPROC:
+			status = "R (running)";
+			break;
+		case TS_ZOMB:
+			status = "Z (zombie)";
+			break;
+		case TS_STOPPED:
+			status = "T (stopped)";
+			break;
+		default:
+			status = "! (unknown)";
+			break;
+		}
+		thread_unlock(t);
+	} else {
+		/*
+		 * there is a hole in the exit code, where a proc can have
+		 * no threads but it is yet to be flagged SZOMB. We will
+		 * assume we are about to become a zombie
+		 */
+		status = "Z (zombie)";
+	}
+
+	up = PTOU(p);
+	mutex_enter(&p->p_crlock);
+	crhold(cr = p->p_cred);
+	mutex_exit(&p->p_crlock);
+
+	lxpr_uiobuf_printf(uiobuf,
+	    "Name:\t%s\n"
+	    "State:\t%s\n"
+	    "Tgid:\t%d\n"
+	    "Pid:\t%d\n"
+	    "PPid:\t%d\n"
+	    "TracerPid:\t%d\n"
+	    "Uid:\t%u\t%u\t%u\t%u\n"
+	    "Gid:\t%u\t%u\t%u\t%u\n"
+	    "FDSize:\t%d\n"
+	    "Groups:\t",
+	    up->u_comm,
+	    status,
+	    pid, /* thread group id - same as pid */
+	    pid,
+	    ppid,
+	    0,
+	    crgetruid(cr), crgetuid(cr), crgetsuid(cr), crgetuid(cr),
+	    crgetrgid(cr), crgetgid(cr), crgetsgid(cr), crgetgid(cr),
+	    p->p_fno_ctl);
+
+	ngroups = crgetngroups(cr);
+	groups  = crgetgroups(cr);
+	for (i = 0; i < ngroups; i++) {
+		lxpr_uiobuf_printf(uiobuf,
+		    "%u ",
+		    groups[i]);
+	}
+	crfree(cr);
+
+	as = p->p_as;
+	if ((p->p_stat != SZOMB) && !(p->p_flag & SSYS) && (as != &kas)) {
+		mutex_exit(&p->p_lock);
+		AS_LOCK_ENTER(as, &as->a_lock, RW_READER);
+		vsize = as->a_resvsize;
+		rss = rm_asrss(as);
+		AS_LOCK_EXIT(as, &as->a_lock);
+		mutex_enter(&p->p_lock);
+
+		lxpr_uiobuf_printf(uiobuf,
+		    "\n"
+		    "VmSize:\t%8lu kB\n"
+		    "VmLck:\t%8lu kB\n"
+		    "VmRSS:\t%8lu kB\n"
+		    "VmData:\t%8lu kB\n"
+		    "VmStk:\t%8lu kB\n"
+		    "VmExe:\t%8lu kB\n"
+		    "VmLib:\t%8lu kB",
+		    btok(vsize),
+		    0l,
+		    ptok(rss),
+		    0l,
+		    btok(p->p_stksize),
+		    ptok(rss),
+		    0l);
+	}
+
+	sigemptyset(&current);
+	sigemptyset(&ignore);
+	sigemptyset(&handle);
+
+	for (i = 1; i < NSIG; i++) {
+		lx_sig = lxpr_sigmap[i];
+
+		if ((lx_sig > 0) && (lx_sig <= LX_NSIG)) {
+			if (sigismember(&p->p_sig, i))
+				sigaddset(&current, lx_sig);
+
+			if (up->u_signal[i - 1] == SIG_IGN)
+				sigaddset(&ignore, lx_sig);
+			else if (up->u_signal[i - 1] != SIG_DFL)
+				sigaddset(&handle, lx_sig);
+		}
+	}
+
+	lxpr_uiobuf_printf(uiobuf,
+	    "\n"
+	    "SigPnd:\t%08x%08x\n"
+	    "SigBlk:\t%08x%08x\n"
+	    "SigIgn:\t%08x%08x\n"
+	    "SigCgt:\t%08x%08x\n"
+	    "CapInh:\t%016x\n"
+	    "CapPrm:\t%016x\n"
+	    "CapEff:\t%016x\n",
+	    current.__sigbits[1], current.__sigbits[0],
+	    0, 0, /* signals blocked on per thread basis */
+	    ignore.__sigbits[1], ignore.__sigbits[0],
+	    handle.__sigbits[1], handle.__sigbits[0],
+	    /* Can't do anything with linux capabilities */
+	    0,
+	    0,
+	    0);
+
+	lxpr_unlock(p);
+}
+
+
+/*
+ * lxpr_read_pid_stat(): pid stat file
+ */
+static void
+lxpr_read_pid_stat(lxpr_node_t *lxpnp, lxpr_uiobuf_t *uiobuf)
+{
+	proc_t *p;
+	kthread_t *t;
+	struct as *as;
+	char stat;
+	pid_t pid, ppid, pgpid, spid;
+	gid_t psgid;
+	dev_t psdev;
+	size_t rss, vsize;
+	int nice, pri;
+	caddr_t wchan;
+	processorid_t cpu;
+
+	ASSERT(lxpnp->lxpr_type == LXPR_PID_STAT);
+
+	p = lxpr_lock(lxpnp->lxpr_pid);
+	if (p == NULL) {
+		lxpr_uiobuf_seterr(uiobuf, EINVAL);
+		return;
+	}
+
+	pid = p->p_pid;
+
+	/*
+	 * Set Linux defaults if we're the zone's init process
+	 */
+	if (pid == curproc->p_zone->zone_proc_initpid) {
+		pid = 1;		/* PID for init */
+		ppid = 0;		/* parent PID for init is 0 */
+		pgpid = 0;		/* process group for init is 0 */
+		psgid = (gid_t)-1;	/* credential GID for init is -1 */
+		spid = 0;		/* session id for init is 0 */
+		psdev = 0;		/* session device for init is 0 */
+	} else {
+		/*
+		 * Make sure not to reference parent PIDs that reside outside
+		 * the zone
+		 */
+		ppid = ((p->p_flag & SZONETOP) ?
+		    curproc->p_zone->zone_zsched->p_pid : p->p_ppid);
+
+		/*
+		 * Convert ppid to the Linux default of 1 if our parent is the
+		 * zone's init process
+		 */
+		if (ppid == curproc->p_zone->zone_proc_initpid)
+			ppid = 1;
+
+		pgpid = p->p_pgrp;
+
+		mutex_enter(&p->p_splock);
+		mutex_enter(&p->p_sessp->s_lock);
+		spid = p->p_sessp->s_sid;
+		psdev = p->p_sessp->s_dev;
+		if (p->p_sessp->s_cred)
+			psgid = crgetgid(p->p_sessp->s_cred);
+		else
+			psgid = crgetgid(p->p_cred);
+
+		mutex_exit(&p->p_sessp->s_lock);
+		mutex_exit(&p->p_splock);
+	}
+
+	t = prchoose(p);
+	if (t != NULL) {
+		switch (t->t_state) {
+		case TS_SLEEP:
+			stat = 'S'; break;
+		case TS_RUN:
+		case TS_ONPROC:
+			stat = 'R'; break;
+		case TS_ZOMB:
+			stat = 'Z'; break;
+		case TS_STOPPED:
+			stat = 'T'; break;
+		default:
+			stat = '!'; break;
+		}
+
+		if (CL_DONICE(t, NULL, 0, &nice) != 0)
+			nice = 0;
+
+		pri = t->t_pri;
+		wchan = t->t_wchan;
+		cpu = t->t_cpu->cpu_id;
+		thread_unlock(t);
+	} else {
+		/* Only zombies have no threads */
+		stat = 'Z';
+		nice = 0;
+		pri = 0;
+		wchan = 0;
+		cpu = 0;
+	}
+	as = p->p_as;
+	mutex_exit(&p->p_lock);
+	AS_LOCK_ENTER(as, &as->a_lock, RW_READER);
+	vsize = as->a_resvsize;
+	rss = rm_asrss(as);
+	AS_LOCK_EXIT(as, &as->a_lock);
+	mutex_enter(&p->p_lock);
+
+	lxpr_uiobuf_printf(uiobuf,
+	    "%d (%s) %c %d %d %d %d %d "
+	    "%lu %lu %lu %lu %lu "
+	    "%lu %lu %ld %ld "
+	    "%d %d %d "
+	    "%lu "
+	    "%lu "
+	    "%lu %ld %llu "
+	    "%lu %lu %u "
+	    "%lu %lu "
+	    "%lu %lu %lu %lu "
+	    "%lu "
+	    "%lu %lu "
+	    "%d "
+	    "%d"
+	    "\n",
+	    pid, PTOU(p)->u_comm, stat, ppid, pgpid, spid, psdev, psgid,
+	    0l, 0l, 0l, 0l, 0l, /* flags, minflt, cminflt, majflt, cmajflt */
+	    p->p_utime, p->p_stime, p->p_cutime, p->p_cstime,
+	    pri, nice, p->p_lwpcnt,
+	    0l, /* itrealvalue (time before next SIGALRM) */
+	    PTOU(p)->u_ticks,
+	    vsize, rss, p->p_vmem_ctl,
+	    0l, 0l, USRSTACK, /* startcode, endcode, startstack */
+	    0l, 0l, /* kstkesp, kstkeip */
+	    0l, 0l, 0l, 0l, /* signal, blocked, sigignore, sigcatch */
+	    wchan,
+	    0l, 0l, /* nswap, cnswap */
+	    0, /* exit_signal */
+	    cpu);
+
+	lxpr_unlock(p);
+}
+
+/* ARGSUSED */
+static void
+lxpr_read_net_arp(lxpr_node_t *lxpnp, lxpr_uiobuf_t *uiobuf)
+{
+}
+
+/* ARGSUSED */
+static void
+lxpr_read_net_dev(lxpr_node_t *lxpnp, lxpr_uiobuf_t *uiobuf)
+{
+	lxpr_uiobuf_printf(uiobuf, "Inter-|   Receive                   "
+	    "                             |  Transmit\n");
+	lxpr_uiobuf_printf(uiobuf, " face |bytes    packets errs drop fifo"
+	    " frame compressed multicast|bytes    packets errs drop fifo"
+	    " colls carrier compressed\n");
+
+	/*
+	 * Data about each interface should go here, but that shouldn't be added
+	 * unless there is an lxproc reader that actually makes use of it (and
+	 * doesn't need anything else that we refuse to provide)...
+	 */
+}
+
+/* ARGSUSED */
+static void
+lxpr_read_net_dev_mcast(lxpr_node_t *lxpnp, lxpr_uiobuf_t *uiobuf)
+{
+}
+
+/* ARGSUSED */
+static void
+lxpr_read_net_igmp(lxpr_node_t *lxpnp, lxpr_uiobuf_t *uiobuf)
+{
+}
+
+/* ARGSUSED */
+static void
+lxpr_read_net_ip_mr_cache(lxpr_node_t *lxpnp, lxpr_uiobuf_t *uiobuf)
+{
+}
+
+/* ARGSUSED */
+static void
+lxpr_read_net_ip_mr_vif(lxpr_node_t *lxpnp, lxpr_uiobuf_t *uiobuf)
+{
+}
+
+/* ARGSUSED */
+static void
+lxpr_read_net_mcfilter(lxpr_node_t *lxpnp, lxpr_uiobuf_t *uiobuf)
+{
+}
+
+/* ARGSUSED */
+static void
+lxpr_read_net_netstat(lxpr_node_t *lxpnp, lxpr_uiobuf_t *uiobuf)
+{
+}
+
+/* ARGSUSED */
+static void
+lxpr_read_net_raw(lxpr_node_t *lxpnp, lxpr_uiobuf_t *uiobuf)
+{
+}
+
+/* ARGSUSED */
+static void
+lxpr_read_net_route(lxpr_node_t *lxpnp, lxpr_uiobuf_t *uiobuf)
+{
+}
+
+/* ARGSUSED */
+static void
+lxpr_read_net_rpc(lxpr_node_t *lxpnp, lxpr_uiobuf_t *uiobuf)
+{
+}
+
+/* ARGSUSED */
+static void
+lxpr_read_net_rt_cache(lxpr_node_t *lxpnp, lxpr_uiobuf_t *uiobuf)
+{
+}
+
+/* ARGSUSED */
+static void
+lxpr_read_net_sockstat(lxpr_node_t *lxpnp, lxpr_uiobuf_t *uiobuf)
+{
+}
+
+/* ARGSUSED */
+static void
+lxpr_read_net_snmp(lxpr_node_t *lxpnp, lxpr_uiobuf_t *uiobuf)
+{
+}
+
+/* ARGSUSED */
+static void
+lxpr_read_net_stat(lxpr_node_t *lxpnp, lxpr_uiobuf_t *uiobuf)
+{
+}
+
+/* ARGSUSED */
+static void
+lxpr_read_net_tcp(lxpr_node_t *lxpnp, lxpr_uiobuf_t *uiobuf)
+{
+}
+
+/* ARGSUSED */
+static void
+lxpr_read_net_udp(lxpr_node_t *lxpnp, lxpr_uiobuf_t *uiobuf)
+{
+}
+
+/* ARGSUSED */
+static void
+lxpr_read_net_unix(lxpr_node_t *lxpnp, lxpr_uiobuf_t *uiobuf)
+{
+}
+
+/*
+ * lxpr_read_kmsg(): read the contents of the kernel message queue. We
+ * translate this into the reception of console messages for this zone; each
+ * read copies out a single zone console message, or blocks until the next one
+ * is produced.
+ */
+
+#define	LX_KMSG_PRI	"<0>"
+
+static void
+lxpr_read_kmsg(lxpr_node_t *lxpnp, struct lxpr_uiobuf *uiobuf, ldi_handle_t lh)
+{
+	mblk_t		*mp;
+
+	ASSERT(lxpnp->lxpr_type == LXPR_KMSG);
+
+	if (ldi_getmsg(lh, &mp, NULL) == 0) {
+		/*
+		 * lxproc doesn't like successive reads to the same file
+		 * descriptor unless we do an explicit rewind each time.
+		 */
+		lxpr_uiobuf_seek(uiobuf, 0);
+
+		lxpr_uiobuf_printf(uiobuf, "%s%s", LX_KMSG_PRI,
+		    mp->b_cont->b_rptr);
+
+		freemsg(mp);
+	}
+}
+
+/*
+ * lxpr_read_loadavg(): read the contents of the "loadavg" file.  We do just
+ * enough for uptime and other simple lxproc readers to work
+ */
+extern int nthread;
+
+static void
+lxpr_read_loadavg(lxpr_node_t *lxpnp, lxpr_uiobuf_t *uiobuf)
+{
+	ulong_t avenrun1;
+	ulong_t avenrun5;
+	ulong_t avenrun15;
+	ulong_t avenrun1_cs;
+	ulong_t avenrun5_cs;
+	ulong_t avenrun15_cs;
+	int loadavg[3];
+	int *loadbuf;
+	cpupart_t *cp;
+	zone_t *zone = LXPTOZ(lxpnp);
+
+	uint_t nrunnable = 0;
+	rctl_qty_t nlwps;
+
+	ASSERT(lxpnp->lxpr_type == LXPR_LOADAVG);
+
+	mutex_enter(&cpu_lock);
+
+	/*
+	 * Need to add up values over all CPU partitions. If pools are active,
+	 * only report the values of the zone's partition, which by definition
+	 * includes the current CPU.
+	 */
+	if (pool_pset_enabled()) {
+		psetid_t psetid = zone_pset_get(curproc->p_zone);
+
+		ASSERT(curproc->p_zone != &zone0);
+		cp = CPU->cpu_part;
+
+		nrunnable = cp->cp_nrunning + cp->cp_nrunnable;
+		(void) cpupart_get_loadavg(psetid, &loadavg[0], 3);
+		loadbuf = &loadavg[0];
+	} else {
+		cp = cp_list_head;
+		do {
+			nrunnable += cp->cp_nrunning + cp->cp_nrunnable;
+		} while ((cp = cp->cp_next) != cp_list_head);
+
+		loadbuf = zone == global_zone ?
+		    &avenrun[0] : zone->zone_avenrun;
+	}
+
+	/*
+	 * If we're in the non-global zone, we'll report the total number of
+	 * LWPs in the zone for the "nproc" parameter of /proc/loadavg,
+	 * otherwise will just use nthread (which will include kernel threads,
+	 * but should be good enough for lxproc).
+	 */
+	nlwps = zone == global_zone ? nthread : zone->zone_nlwps;
+
+	mutex_exit(&cpu_lock);
+
+	avenrun1 = loadbuf[0] >> FSHIFT;
+	avenrun1_cs = ((loadbuf[0] & (FSCALE-1)) * 100) >> FSHIFT;
+	avenrun5 = loadbuf[1] >> FSHIFT;
+	avenrun5_cs = ((loadbuf[1] & (FSCALE-1)) * 100) >> FSHIFT;
+	avenrun15 = loadbuf[2] >> FSHIFT;
+	avenrun15_cs = ((loadbuf[2] & (FSCALE-1)) * 100) >> FSHIFT;
+
+	lxpr_uiobuf_printf(uiobuf,
+	    "%ld.%02d %ld.%02d %ld.%02d %d/%d %d\n",
+	    avenrun1, avenrun1_cs,
+	    avenrun5, avenrun5_cs,
+	    avenrun15, avenrun15_cs,
+	    nrunnable, nlwps, 0);
+}
+
+/*
+ * lxpr_read_meminfo(): read the contents of the "meminfo" file.
+ */
+static void
+lxpr_read_meminfo(lxpr_node_t *lxpnp, lxpr_uiobuf_t *uiobuf)
+{
+	zone_t *zone = LXPTOZ(lxpnp);
+	int global = zone == global_zone;
+	long total_mem, free_mem, total_swap, used_swap;
+
+	ASSERT(lxpnp->lxpr_type == LXPR_MEMINFO);
+
+	if (global || zone->zone_phys_mem_ctl == UINT64_MAX) {
+		total_mem = physmem * PAGESIZE;
+		free_mem = freemem * PAGESIZE;
+	} else {
+		total_mem = zone->zone_phys_mem_ctl;
+		free_mem = zone->zone_phys_mem_ctl - zone->zone_phys_mem;
+	}
+
+	if (global || zone->zone_max_swap_ctl == UINT64_MAX) {
+		total_swap = k_anoninfo.ani_max * PAGESIZE;
+		used_swap = k_anoninfo.ani_phys_resv * PAGESIZE;
+	} else {
+		mutex_enter(&zone->zone_mem_lock);
+		total_swap = zone->zone_max_swap_ctl;
+		used_swap = zone->zone_max_swap;
+		mutex_exit(&zone->zone_mem_lock);
+	}
+
+	lxpr_uiobuf_printf(uiobuf,
+	    "        total:     used:    free:  shared: buffers:  cached:\n"
+	    "Mem:  %8lu %8lu %8lu %8u %8u %8u\n"
+	    "Swap: %8lu %8lu %8lu\n"
+	    "MemTotal:  %8lu kB\n"
+	    "MemFree:   %8lu kB\n"
+	    "MemShared: %8u kB\n"
+	    "Buffers:   %8u kB\n"
+	    "Cached:    %8u kB\n"
+	    "SwapCached:%8u kB\n"
+	    "Active:    %8u kB\n"
+	    "Inactive:  %8u kB\n"
+	    "HighTotal: %8u kB\n"
+	    "HighFree:  %8u kB\n"
+	    "LowTotal:  %8u kB\n"
+	    "LowFree:   %8u kB\n"
+	    "SwapTotal: %8lu kB\n"
+	    "SwapFree:  %8lu kB\n",
+	    total_mem, total_mem - free_mem, free_mem, 0, 0, 0,
+	    total_swap, used_swap, total_swap - used_swap,
+	    btok(total_mem),				/* MemTotal */
+	    btok(free_mem),				/* MemFree */
+	    0,						/* MemShared */
+	    0,						/* Buffers */
+	    0,						/* Cached */
+	    0,						/* SwapCached */
+	    0,						/* Active */
+	    0,						/* Inactive */
+	    0,						/* HighTotal */
+	    0,						/* HighFree */
+	    btok(total_mem),				/* LowTotal */
+	    btok(free_mem),				/* LowFree */
+	    btok(total_swap),				/* SwapTotal */
+	    btok(total_swap - used_swap));		/* SwapFree */
+}
+
+/*
+ * lxpr_read_mounts():
+ */
+/* ARGSUSED */
+static void
+lxpr_read_mounts(lxpr_node_t *lxpnp, lxpr_uiobuf_t *uiobuf)
+{
+	struct vfs *vfsp;
+	struct vfs *vfslist;
+	zone_t *zone = LXPTOZ(lxpnp);
+	struct print_data {
+		refstr_t *vfs_mntpt;
+		refstr_t *vfs_resource;
+		uint_t vfs_flag;
+		int vfs_fstype;
+		struct print_data *next;
+	} *print_head = NULL;
+	struct print_data **print_tail = &print_head;
+	struct print_data *printp;
+
+	vfs_list_read_lock();
+
+	if (zone == global_zone) {
+		vfsp = vfslist = rootvfs;
+	} else {
+		vfsp = vfslist = zone->zone_vfslist;
+		/*
+		 * If the zone has a root entry, it will be the first in
+		 * the list.  If it doesn't, we conjure one up.
+		 */
+		if (vfslist == NULL || strcmp(refstr_value(vfsp->vfs_mntpt),
+		    zone->zone_rootpath) != 0) {
+			struct vfs *tvfsp;
+			/*
+			 * The root of the zone is not a mount point.  The vfs
+			 * we want to report is that of the zone's root vnode.
+			 */
+			tvfsp = zone->zone_rootvp->v_vfsp;
+
+			lxpr_uiobuf_printf(uiobuf,
+			    "/ / %s %s 0 0\n",
+			    vfssw[tvfsp->vfs_fstype].vsw_name,
+			    tvfsp->vfs_flag & VFS_RDONLY ? "ro" : "rw");
+
+		}
+		if (vfslist == NULL) {
+			vfs_list_unlock();
+			return;
+		}
+	}
+
+	/*
+	 * Later on we have to do a lookupname, which can end up causing
+	 * another vfs_list_read_lock() to be called. Which can lead to a
+	 * deadlock. To avoid this, we extract the data we need into a local
+	 * list, then we can run this list without holding vfs_list_read_lock()
+	 * We keep the list in the same order as the vfs_list
+	 */
+	do {
+		/* Skip mounts we shouldn't show */
+		if (vfsp->vfs_flag & VFS_NOMNTTAB) {
+			goto nextfs;
+		}
+
+		printp = kmem_alloc(sizeof (*printp), KM_SLEEP);
+		refstr_hold(vfsp->vfs_mntpt);
+		printp->vfs_mntpt = vfsp->vfs_mntpt;
+		refstr_hold(vfsp->vfs_resource);
+		printp->vfs_resource = vfsp->vfs_resource;
+		printp->vfs_flag = vfsp->vfs_flag;
+		printp->vfs_fstype = vfsp->vfs_fstype;
+		printp->next = NULL;
+
+		*print_tail = printp;
+		print_tail = &printp->next;
+
+nextfs:
+		vfsp = (zone == global_zone) ?
+		    vfsp->vfs_next : vfsp->vfs_zone_next;
+
+	} while (vfsp != vfslist);
+
+	vfs_list_unlock();
+
+	/*
+	 * now we can run through what we've extracted without holding
+	 * vfs_list_read_lock()
+	 */
+	printp = print_head;
+	while (printp != NULL) {
+		struct print_data *printp_next;
+		const char *resource;
+		char *mntpt;
+		struct vnode *vp;
+		int error;
+
+		mntpt = (char *)refstr_value(printp->vfs_mntpt);
+		resource = refstr_value(printp->vfs_resource);
+
+		if (mntpt != NULL && mntpt[0] != '\0')
+			mntpt = ZONE_PATH_TRANSLATE(mntpt, zone);
+		else
+			mntpt = "-";
+
+		error = lookupname(mntpt, UIO_SYSSPACE, FOLLOW, NULLVPP, &vp);
+
+		if (error != 0)
+			goto nextp;
+
+		if (!(vp->v_flag & VROOT)) {
+			VN_RELE(vp);
+			goto nextp;
+		}
+		VN_RELE(vp);
+
+		if (resource != NULL && resource[0] != '\0') {
+			if (resource[0] == '/') {
+				resource = ZONE_PATH_VISIBLE(resource, zone) ?
+				    ZONE_PATH_TRANSLATE(resource, zone) :
+				    mntpt;
+			}
+		} else {
+			resource = "-";
+		}
+
+		lxpr_uiobuf_printf(uiobuf,
+		    "%s %s %s %s 0 0\n",
+		    resource, mntpt, vfssw[printp->vfs_fstype].vsw_name,
+		    printp->vfs_flag & VFS_RDONLY ? "ro" : "rw");
+
+nextp:
+		printp_next = printp->next;
+		refstr_rele(printp->vfs_mntpt);
+		refstr_rele(printp->vfs_resource);
+		kmem_free(printp, sizeof (*printp));
+		printp = printp_next;
+
+	}
+}
+
+/*
+ * lxpr_read_partitions():
+ *
+ * We don't support partitions in a local zone because it requires access to
+ * physical devices.  But we need to fake up enough of the file to show that we
+ * have no partitions.
+ */
+/* ARGSUSED */
+static void
+lxpr_read_partitions(lxpr_node_t *lxpnp, lxpr_uiobuf_t *uiobuf)
+{
+	lxpr_uiobuf_printf(uiobuf,
+	    "major minor  #blocks  name     rio rmerge rsect ruse "
+	    "wio wmerge wsect wuse running use aveq\n\n");
+}
+
+/*
+ * lxpr_read_version(): read the contents of the "version" file.  Note that
+ * we don't lie here -- we don't pretend that we're Linux.  If lxproc is to
+ * be used in a Linux-branded zone, there will need to be a mount option to
+ * indicate that Linux should be more fully mimicked.
+ */
+/* ARGSUSED */
+static void
+lxpr_read_version(lxpr_node_t *lxpnp, lxpr_uiobuf_t *uiobuf)
+{
+	lxpr_uiobuf_printf(uiobuf,
+	    "%s version %s (%s version %d.%d.%d) "
+	    "#%s SMP %s\n",
+	    utsname.sysname, utsname.release,
+#if defined(__GNUC__)
+	    "gcc",
+	    __GNUC__,
+	    __GNUC_MINOR__,
+	    __GNUC_PATCHLEVEL__,
+#else
+	    "Sun C",
+	    __SUNPRO_C / 0x100,
+	    (__SUNPRO_C & 0xff) / 0x10,
+	    __SUNPRO_C & 0xf,
+#endif
+	    utsname.version,
+	    "00:00:00 00/00/00");
+}
+
+/*
+ * lxpr_read_stat(): read the contents of the "stat" file.
+ *
+ */
+/* ARGSUSED */
+static void
+lxpr_read_stat(lxpr_node_t *lxpnp, lxpr_uiobuf_t *uiobuf)
+{
+	cpu_t *cp, *cpstart;
+	int pools_enabled;
+	ulong_t idle_cum = 0;
+	ulong_t sys_cum  = 0;
+	ulong_t user_cum = 0;
+	ulong_t irq_cum = 0;
+	ulong_t cpu_nrunnable_cum = 0;
+	ulong_t w_io_cum = 0;
+
+	ulong_t pgpgin_cum    = 0;
+	ulong_t pgpgout_cum   = 0;
+	ulong_t pgswapout_cum = 0;
+	ulong_t pgswapin_cum  = 0;
+	ulong_t intr_cum = 0;
+	ulong_t pswitch_cum = 0;
+	ulong_t forks_cum = 0;
+	hrtime_t msnsecs[NCMSTATES];
+
+	/* temporary variable since scalehrtime modifies data in place */
+	hrtime_t tmptime;
+
+	ASSERT(lxpnp->lxpr_type == LXPR_STAT);
+
+	mutex_enter(&cpu_lock);
+	pools_enabled = pool_pset_enabled();
+
+	/* Calculate cumulative stats */
+	cp = cpstart = CPU->cpu_part->cp_cpulist;
+	do {
+		int i;
+
+		/*
+		 * Don't count CPUs that aren't even in the system
+		 * or aren't up yet.
+		 */
+		if ((cp->cpu_flags & CPU_EXISTS) == 0) {
+			continue;
+		}
+
+		get_cpu_mstate(cp, msnsecs);
+
+		idle_cum += NSEC_TO_TICK(msnsecs[CMS_IDLE]);
+		sys_cum  += NSEC_TO_TICK(msnsecs[CMS_SYSTEM]);
+		user_cum += NSEC_TO_TICK(msnsecs[CMS_USER]);
+
+		pgpgin_cum += CPU_STATS(cp, vm.pgpgin);
+		pgpgout_cum += CPU_STATS(cp, vm.pgpgout);
+		pgswapin_cum += CPU_STATS(cp, vm.pgswapin);
+		pgswapout_cum += CPU_STATS(cp, vm.pgswapout);
+
+		cpu_nrunnable_cum += cp->cpu_disp->disp_nrunnable;
+		w_io_cum += CPU_STATS(cp, sys.iowait);
+		for (i = 0; i < NCMSTATES; i++) {
+			tmptime = cp->cpu_intracct[i];
+			scalehrtime(&tmptime);
+			irq_cum += NSEC_TO_TICK(tmptime);
+		}
+
+		for (i = 0; i < PIL_MAX; i++)
+			intr_cum += CPU_STATS(cp, sys.intr[i]);
+
+		pswitch_cum += CPU_STATS(cp, sys.pswitch);
+		forks_cum += CPU_STATS(cp, sys.sysfork);
+		forks_cum += CPU_STATS(cp, sys.sysvfork);
+
+		if (pools_enabled)
+			cp = cp->cpu_next_part;
+		else
+			cp = cp->cpu_next;
+	} while (cp != cpstart);
+
+	lxpr_uiobuf_printf(uiobuf, "cpu %lu %lu %lu %lu %lu %lu %lu\n",
+	    user_cum, 0L, sys_cum, idle_cum, 0L, irq_cum, 0L);
+
+	/* Do per processor stats */
+	do {
+		int i;
+
+		ulong_t idle_ticks;
+		ulong_t sys_ticks;
+		ulong_t user_ticks;
+		ulong_t irq_ticks = 0;
+
+		/*
+		 * Don't count CPUs that aren't even in the system
+		 * or aren't up yet.
+		 */
+		if ((cp->cpu_flags & CPU_EXISTS) == 0) {
+			continue;
+		}
+
+		get_cpu_mstate(cp, msnsecs);
+
+		idle_ticks = NSEC_TO_TICK(msnsecs[CMS_IDLE]);
+		sys_ticks  = NSEC_TO_TICK(msnsecs[CMS_SYSTEM]);
+		user_ticks = NSEC_TO_TICK(msnsecs[CMS_USER]);
+
+		for (i = 0; i < NCMSTATES; i++) {
+			tmptime = cp->cpu_intracct[i];
+			scalehrtime(&tmptime);
+			irq_ticks += NSEC_TO_TICK(tmptime);
+		}
+
+		lxpr_uiobuf_printf(uiobuf,
+		    "cpu%d %lu %lu %lu %lu %lu %lu %lu\n",
+		    cp->cpu_id, user_ticks, 0L, sys_ticks, idle_ticks,
+		    0L, irq_ticks, 0L);
+
+		if (pools_enabled)
+			cp = cp->cpu_next_part;
+		else
+			cp = cp->cpu_next;
+	} while (cp != cpstart);
+
+	mutex_exit(&cpu_lock);
+
+	lxpr_uiobuf_printf(uiobuf,
+	    "page %lu %lu\n"
+	    "swap %lu %lu\n"
+	    "intr %lu\n"
+	    "ctxt %lu\n"
+	    "btime %lu\n"
+	    "processes %lu\n"
+	    "procs_running %lu\n"
+	    "procs_blocked %lu\n",
+	    pgpgin_cum, pgpgout_cum,
+	    pgswapin_cum, pgswapout_cum,
+	    intr_cum,
+	    pswitch_cum,
+	    boot_time,
+	    forks_cum,
+	    cpu_nrunnable_cum,
+	    w_io_cum);
+}
+
+/*
+ * lxpr_read_uptime(): read the contents of the "uptime" file.
+ *
+ * format is: "%.2lf, %.2lf",uptime_secs, idle_secs
+ * Use fixed point arithmetic to get 2 decimal places
+ */
+/* ARGSUSED */
+static void
+lxpr_read_uptime(lxpr_node_t *lxpnp, lxpr_uiobuf_t *uiobuf)
+{
+	cpu_t *cp, *cpstart;
+	int pools_enabled;
+	ulong_t idle_cum = 0;
+	ulong_t cpu_count = 0;
+	ulong_t idle_s;
+	ulong_t idle_cs;
+	ulong_t up_s;
+	ulong_t up_cs;
+	hrtime_t birthtime;
+	hrtime_t centi_sec = 10000000;  /* 10^7 */
+
+	ASSERT(lxpnp->lxpr_type == LXPR_UPTIME);
+
+	/* Calculate cumulative stats */
+	mutex_enter(&cpu_lock);
+	pools_enabled = pool_pset_enabled();
+
+	cp = cpstart = CPU;
+	do {
+		/*
+		 * Don't count CPUs that aren't even in the system
+		 * or aren't up yet.
+		 */
+		if ((cp->cpu_flags & CPU_EXISTS) == 0) {
+			continue;
+		}
+
+		idle_cum += CPU_STATS(cp, sys.cpu_ticks_idle);
+		idle_cum += CPU_STATS(cp, sys.cpu_ticks_wait);
+		cpu_count += 1;
+
+		if (pools_enabled)
+			cp = cp->cpu_next_part;
+		else
+			cp = cp->cpu_next;
+	} while (cp != cpstart);
+	mutex_exit(&cpu_lock);
+
+	/* Getting the Zone zsched process startup time */
+	birthtime = LXPTOZ(lxpnp)->zone_zsched->p_mstart;
+	up_cs = (gethrtime() - birthtime) / centi_sec;
+	up_s = up_cs / 100;
+	up_cs %= 100;
+
+	ASSERT(cpu_count > 0);
+	idle_cum /= cpu_count;
+	idle_s = idle_cum / hz;
+	idle_cs = idle_cum % hz;
+	idle_cs *= 100;
+	idle_cs /= hz;
+
+	lxpr_uiobuf_printf(uiobuf,
+	    "%ld.%02d %ld.%02d\n", up_s, up_cs, idle_s, idle_cs);
+}
+
+static const char *amd_x_edx[] = {
+	NULL,	NULL,	NULL,	NULL,
+	NULL,	NULL,	NULL,	NULL,
+	NULL,	NULL,	NULL,	"syscall",
+	NULL,	NULL,	NULL,	NULL,
+	NULL,	NULL,	NULL,	"mp",
+	"nx",	NULL,	"mmxext", NULL,
+	NULL,	NULL,	NULL,	NULL,
+	NULL,	"lm",	"3dnowext", "3dnow"
+};
+
+static const char *amd_x_ecx[] = {
+	"lahf_lm", NULL, "svm", NULL,
+	"altmovcr8"
+};
+
+static const char *tm_x_edx[] = {
+	"recovery", "longrun", NULL, "lrti"
+};
+
+/*
+ * Intel calls no-execute "xd" in its docs, but Linux still reports it as "nx."
+ */
+static const char *intc_x_edx[] = {
+	NULL,	NULL,	NULL,	NULL,
+	NULL,	NULL,	NULL,	NULL,
+	NULL,	NULL,	NULL,	"syscall",
+	NULL,	NULL,	NULL,	NULL,
+	NULL,	NULL,	NULL,	NULL,
+	"nx",	NULL,	NULL,   NULL,
+	NULL,	NULL,	NULL,	NULL,
+	NULL,	"lm",   NULL,   NULL
+};
+
+static const char *intc_edx[] = {
+	"fpu",	"vme",	"de",	"pse",
+	"tsc",	"msr",	"pae",	"mce",
+	"cx8",	"apic",	 NULL,	"sep",
+	"mtrr",	"pge",	"mca",	"cmov",
+	"pat",	"pse36", "pn",	"clflush",
+	NULL,	"dts",	"acpi",	"mmx",
+	"fxsr",	"sse",	"sse2",	"ss",
+	"ht",	"tm",	"ia64",	"pbe"
+};
+
+/*
+ * "sse3" on linux is called "pni" (Prescott New Instructions).
+ */
+static const char *intc_ecx[] = {
+	"pni",	NULL,	NULL, "monitor",
+	"ds_cpl", NULL,	NULL, "est",
+	"tm2",	NULL,	"cid", NULL,
+	NULL,	"cx16",	"xtpr"
+};
+
+static void
+lxpr_read_cpuinfo(lxpr_node_t *lxpnp, lxpr_uiobuf_t *uiobuf)
+{
+	int i;
+	uint32_t bits;
+	cpu_t *cp, *cpstart;
+	int pools_enabled;
+	const char **fp;
+	char brandstr[CPU_IDSTRLEN];
+	struct cpuid_regs cpr;
+	int maxeax;
+	int std_ecx, std_edx, ext_ecx, ext_edx;
+
+	ASSERT(lxpnp->lxpr_type == LXPR_CPUINFO);
+
+	mutex_enter(&cpu_lock);
+	pools_enabled = pool_pset_enabled();
+
+	cp = cpstart = CPU;
+	do {
+		/*
+		 * This returns the maximum eax value for standard cpuid
+		 * functions in eax.
+		 */
+		cpr.cp_eax = 0;
+		(void) cpuid_insn(cp, &cpr);
+		maxeax = cpr.cp_eax;
+
+		/*
+		 * Get standard x86 feature flags.
+		 */
+		cpr.cp_eax = 1;
+		(void) cpuid_insn(cp, &cpr);
+		std_ecx = cpr.cp_ecx;
+		std_edx = cpr.cp_edx;
+
+		/*
+		 * Now get extended feature flags.
+		 */
+		cpr.cp_eax = 0x80000001;
+		(void) cpuid_insn(cp, &cpr);
+		ext_ecx = cpr.cp_ecx;
+		ext_edx = cpr.cp_edx;
+
+		(void) cpuid_getbrandstr(cp, brandstr, CPU_IDSTRLEN);
+
+		lxpr_uiobuf_printf(uiobuf,
+		    "processor\t: %d\n"
+		    "vendor_id\t: %s\n"
+		    "cpu family\t: %d\n"
+		    "model\t\t: %d\n"
+		    "model name\t: %s\n"
+		    "stepping\t: %d\n"
+		    "cpu MHz\t\t: %u.%03u\n",
+		    cp->cpu_id, cpuid_getvendorstr(cp), cpuid_getfamily(cp),
+		    cpuid_getmodel(cp), brandstr, cpuid_getstep(cp),
+		    (uint32_t)(cpu_freq_hz / 1000000),
+		    ((uint32_t)(cpu_freq_hz / 1000)) % 1000);
+
+		lxpr_uiobuf_printf(uiobuf, "cache size\t: %u KB\n",
+		    getl2cacheinfo(cp, NULL, NULL, NULL) / 1024);
+
+		if (is_x86_feature(x86_featureset, X86FSET_HTT)) {
+			/*
+			 * 'siblings' is used for HT-style threads
+			 */
+			lxpr_uiobuf_printf(uiobuf,
+			    "physical id\t: %lu\n"
+			    "siblings\t: %u\n",
+			    pg_plat_hw_instance_id(cp, PGHW_CHIP),
+			    cpuid_get_ncpu_per_chip(cp));
+		}
+
+		/*
+		 * Since we're relatively picky about running on older hardware,
+		 * we can be somewhat cavalier about the answers to these ones.
+		 *
+		 * In fact, given the hardware we support, we just say:
+		 *
+		 *	fdiv_bug	: no	(if we're on a 64-bit kernel)
+		 *	hlt_bug		: no
+		 *	f00f_bug	: no
+		 *	coma_bug	: no
+		 *	wp		: yes	(write protect in supervsr mode)
+		 */
+		lxpr_uiobuf_printf(uiobuf,
+		    "fdiv_bug\t: %s\n"
+		    "hlt_bug \t: no\n"
+		    "f00f_bug\t: no\n"
+		    "coma_bug\t: no\n"
+		    "fpu\t\t: %s\n"
+		    "fpu_exception\t: %s\n"
+		    "cpuid level\t: %d\n"
+		    "flags\t\t:",
+#if defined(__i386)
+		    fpu_pentium_fdivbug ? "yes" : "no",
+#else
+		    "no",
+#endif /* __i386 */
+		    fpu_exists ? "yes" : "no", fpu_exists ? "yes" : "no",
+		    maxeax);
+
+		for (bits = std_edx, fp = intc_edx, i = 0;
+		    i < sizeof (intc_edx) / sizeof (intc_edx[0]); fp++, i++)
+			if ((bits & (1 << i)) != 0 && *fp)
+				lxpr_uiobuf_printf(uiobuf, " %s", *fp);
+
+		/*
+		 * name additional features where appropriate
+		 */
+		switch (x86_vendor) {
+		case X86_VENDOR_Intel:
+			for (bits = ext_edx, fp = intc_x_edx, i = 0;
+			    i < sizeof (intc_x_edx) / sizeof (intc_x_edx[0]);
+			    fp++, i++)
+				if ((bits & (1 << i)) != 0 && *fp)
+					lxpr_uiobuf_printf(uiobuf, " %s", *fp);
+			break;
+
+		case X86_VENDOR_AMD:
+			for (bits = ext_edx, fp = amd_x_edx, i = 0;
+			    i < sizeof (amd_x_edx) / sizeof (amd_x_edx[0]);
+			    fp++, i++)
+				if ((bits & (1 << i)) != 0 && *fp)
+					lxpr_uiobuf_printf(uiobuf, " %s", *fp);
+
+			for (bits = ext_ecx, fp = amd_x_ecx, i = 0;
+			    i < sizeof (amd_x_ecx) / sizeof (amd_x_ecx[0]);
+			    fp++, i++)
+				if ((bits & (1 << i)) != 0 && *fp)
+					lxpr_uiobuf_printf(uiobuf, " %s", *fp);
+			break;
+
+		case X86_VENDOR_TM:
+			for (bits = ext_edx, fp = tm_x_edx, i = 0;
+			    i < sizeof (tm_x_edx) / sizeof (tm_x_edx[0]);
+			    fp++, i++)
+				if ((bits & (1 << i)) != 0 && *fp)
+					lxpr_uiobuf_printf(uiobuf, " %s", *fp);
+			break;
+		default:
+			break;
+		}
+
+		for (bits = std_ecx, fp = intc_ecx, i = 0;
+		    i < sizeof (intc_ecx) / sizeof (intc_ecx[0]); fp++, i++)
+			if ((bits & (1 << i)) != 0 && *fp)
+				lxpr_uiobuf_printf(uiobuf, " %s", *fp);
+
+		lxpr_uiobuf_printf(uiobuf, "\n\n");
+
+		if (pools_enabled)
+			cp = cp->cpu_next_part;
+		else
+			cp = cp->cpu_next;
+	} while (cp != cpstart);
+
+	mutex_exit(&cpu_lock);
+}
+
+/* ARGSUSED */
+static void
+lxpr_read_fd(lxpr_node_t *lxpnp, lxpr_uiobuf_t *uiobuf)
+{
+	ASSERT(lxpnp->lxpr_type == LXPR_PID_FD_FD);
+	lxpr_uiobuf_seterr(uiobuf, EFAULT);
+}
+
+/*
+ * lxpr_getattr(): Vnode operation for VOP_GETATTR()
+ */
+static int
+lxpr_getattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr,
+    caller_context_t *ct)
+{
+	register lxpr_node_t *lxpnp = VTOLXP(vp);
+	lxpr_nodetype_t type = lxpnp->lxpr_type;
+	extern uint_t nproc;
+	int error;
+
+	/*
+	 * Return attributes of underlying vnode if ATTR_REAL
+	 *
+	 * but keep fd files with the symlink permissions
+	 */
+	if (lxpnp->lxpr_realvp != NULL && (flags & ATTR_REAL)) {
+		vnode_t *rvp = lxpnp->lxpr_realvp;
+
+		/*
+		 * withold attribute information to owner or root
+		 */
+		if ((error = VOP_ACCESS(rvp, 0, 0, cr, ct)) != 0) {
+			return (error);
+		}
+
+		/*
+		 * now its attributes
+		 */
+		if ((error = VOP_GETATTR(rvp, vap, flags, cr, ct)) != 0) {
+			return (error);
+		}
+
+		/*
+		 * if it's a file in lx /proc/pid/fd/xx then set its
+		 * mode and keep it looking like a symlink
+		 */
+		if (type == LXPR_PID_FD_FD) {
+			vap->va_mode = lxpnp->lxpr_mode;
+			vap->va_type = vp->v_type;
+			vap->va_size = 0;
+			vap->va_nlink = 1;
+		}
+		return (0);
+	}
+
+	/* Default attributes, that may be overridden below */
+	bzero(vap, sizeof (*vap));
+	vap->va_atime = vap->va_mtime = vap->va_ctime = lxpnp->lxpr_time;
+	vap->va_nlink = 1;
+	vap->va_type = vp->v_type;
+	vap->va_mode = lxpnp->lxpr_mode;
+	vap->va_fsid = vp->v_vfsp->vfs_dev;
+	vap->va_blksize = DEV_BSIZE;
+	vap->va_uid = lxpnp->lxpr_uid;
+	vap->va_gid = lxpnp->lxpr_gid;
+	vap->va_nodeid = lxpnp->lxpr_ino;
+
+	switch (type) {
+	case LXPR_PROCDIR:
+		vap->va_nlink = nproc + 2 + PROCDIRFILES;
+		vap->va_size = (nproc + 2 + PROCDIRFILES) * LXPR_SDSIZE;
+		break;
+	case LXPR_PIDDIR:
+		vap->va_nlink = PIDDIRFILES;
+		vap->va_size = PIDDIRFILES * LXPR_SDSIZE;
+		break;
+	case LXPR_SELF:
+		vap->va_uid = crgetruid(curproc->p_cred);
+		vap->va_gid = crgetrgid(curproc->p_cred);
+		break;
+	default:
+		break;
+	}
+
+	vap->va_nblocks = (fsblkcnt64_t)btod(vap->va_size);
+	return (0);
+}
+
+/*
+ * lxpr_access(): Vnode operation for VOP_ACCESS()
+ */
+static int
+lxpr_access(vnode_t *vp, int mode, int flags, cred_t *cr, caller_context_t *ct)
+{
+	lxpr_node_t *lxpnp = VTOLXP(vp);
+	int shift = 0;
+	proc_t *tp;
+
+	/* lx /proc is a read only file system */
+	if (mode & VWRITE)
+		return (EROFS);
+
+	/*
+	 * If this is a restricted file, check access permissions.
+	 */
+	switch (lxpnp->lxpr_type) {
+	case LXPR_PIDDIR:
+		return (0);
+	case LXPR_PID_CURDIR:
+	case LXPR_PID_ENV:
+	case LXPR_PID_EXE:
+	case LXPR_PID_MAPS:
+	case LXPR_PID_MEM:
+	case LXPR_PID_ROOTDIR:
+	case LXPR_PID_FDDIR:
+	case LXPR_PID_FD_FD:
+		if ((tp = lxpr_lock(lxpnp->lxpr_pid)) == NULL)
+			return (ENOENT);
+		if (tp != curproc && secpolicy_proc_access(cr) != 0 &&
+		    priv_proc_cred_perm(cr, tp, NULL, mode) != 0) {
+			lxpr_unlock(tp);
+			return (EACCES);
+		}
+		lxpr_unlock(tp);
+	default:
+		break;
+	}
+
+	if (lxpnp->lxpr_realvp != NULL) {
+		/*
+		 * For these we use the underlying vnode's accessibility.
+		 */
+		return (VOP_ACCESS(lxpnp->lxpr_realvp, mode, flags, cr, ct));
+	}
+
+	/* If user is root allow access regardless of permission bits */
+	if (secpolicy_proc_access(cr) == 0)
+		return (0);
+
+	/*
+	 * Access check is based on only one of owner, group, public.  If not
+	 * owner, then check group.  If not a member of the group, then check
+	 * public access.
+	 */
+	if (crgetuid(cr) != lxpnp->lxpr_uid) {
+		shift += 3;
+		if (!groupmember((uid_t)lxpnp->lxpr_gid, cr))
+			shift += 3;
+	}
+
+	mode &= ~(lxpnp->lxpr_mode << shift);
+
+	if (mode == 0)
+		return (0);
+
+	return (EACCES);
+}
+
+/* ARGSUSED */
+static vnode_t *
+lxpr_lookup_not_a_dir(vnode_t *dp, char *comp)
+{
+	return (NULL);
+}
+
+/*
+ * lxpr_lookup(): Vnode operation for VOP_LOOKUP()
+ */
+/* ARGSUSED */
+static int
+lxpr_lookup(vnode_t *dp, char *comp, vnode_t **vpp, pathname_t *pathp,
+	int flags, vnode_t *rdir, cred_t *cr, caller_context_t *ct,
+	int *direntflags, pathname_t *realpnp)
+{
+	lxpr_node_t *lxpnp = VTOLXP(dp);
+	lxpr_nodetype_t type = lxpnp->lxpr_type;
+	int error;
+
+	ASSERT(dp->v_type == VDIR);
+	ASSERT(type < LXPR_NFILES);
+
+	/*
+	 * we should never get here because the lookup
+	 * is done on the realvp for these nodes
+	 */
+	ASSERT(type != LXPR_PID_FD_FD &&
+	    type != LXPR_PID_CURDIR &&
+	    type != LXPR_PID_ROOTDIR);
+
+	/*
+	 * restrict lookup permission to owner or root
+	 */
+	if ((error = lxpr_access(dp, VEXEC, 0, cr, ct)) != 0) {
+		return (error);
+	}
+
+	/*
+	 * Just return the parent vnode if that's where we are trying to go.
+	 */
+	if (strcmp(comp, "..") == 0) {
+		VN_HOLD(lxpnp->lxpr_parent);
+		*vpp = lxpnp->lxpr_parent;
+		return (0);
+	}
+
+	/*
+	 * Special handling for directory searches.  Note: null component name
+	 * denotes that the current directory is being searched.
+	 */
+	if ((dp->v_type == VDIR) && (*comp == '\0' || strcmp(comp, ".") == 0)) {
+		VN_HOLD(dp);
+		*vpp = dp;
+		return (0);
+	}
+
+	*vpp = (lxpr_lookup_function[type](dp, comp));
+	return ((*vpp == NULL) ? ENOENT : 0);
+}
+
+/*
+ * Do a sequential search on the given directory table
+ */
+static vnode_t *
+lxpr_lookup_common(vnode_t *dp, char *comp, proc_t *p,
+    lxpr_dirent_t *dirtab, int dirtablen)
+{
+	lxpr_node_t *lxpnp;
+	int count;
+
+	for (count = 0; count < dirtablen; count++) {
+		if (strcmp(dirtab[count].d_name, comp) == 0) {
+			lxpnp = lxpr_getnode(dp, dirtab[count].d_type, p, 0);
+			dp = LXPTOV(lxpnp);
+			ASSERT(dp != NULL);
+			return (dp);
+		}
+	}
+	return (NULL);
+}
+
+static vnode_t *
+lxpr_lookup_piddir(vnode_t *dp, char *comp)
+{
+	proc_t *p;
+
+	ASSERT(VTOLXP(dp)->lxpr_type == LXPR_PIDDIR);
+
+	p = lxpr_lock(VTOLXP(dp)->lxpr_pid);
+	if (p == NULL)
+		return (NULL);
+
+	dp = lxpr_lookup_common(dp, comp, p, piddir, PIDDIRFILES);
+
+	lxpr_unlock(p);
+
+	return (dp);
+}
+
+/*
+ * Lookup one of the process's open files.
+ */
+static vnode_t *
+lxpr_lookup_fddir(vnode_t *dp, char *comp)
+{
+	lxpr_node_t *dlxpnp = VTOLXP(dp);
+	lxpr_node_t *lxpnp;
+	vnode_t *vp = NULL;
+	proc_t *p;
+	file_t *fp;
+	uint_t fd;
+	int c;
+	uf_entry_t *ufp;
+	uf_info_t *fip;
+
+	ASSERT(dlxpnp->lxpr_type == LXPR_PID_FDDIR);
+
+	/*
+	 * convert the string rendition of the filename
+	 * to a file descriptor
+	 */
+	fd = 0;
+	while ((c = *comp++) != '\0') {
+		int ofd;
+		if (c < '0' || c > '9')
+			return (NULL);
+
+		ofd = fd;
+		fd = 10*fd + c - '0';
+		/* integer overflow */
+		if (fd / 10 != ofd)
+			return (NULL);
+	}
+
+	/*
+	 * get the proc to work with and lock it
+	 */
+	p = lxpr_lock(dlxpnp->lxpr_pid);
+	if ((p == NULL))
+		return (NULL);
+
+	/*
+	 * If the process is a zombie or system process
+	 * it can't have any open files.
+	 */
+	if ((p->p_stat == SZOMB) || (p->p_flag & SSYS) || (p->p_as == &kas)) {
+		lxpr_unlock(p);
+		return (NULL);
+	}
+
+	/*
+	 * get us a fresh node/vnode
+	 */
+	lxpnp = lxpr_getnode(dp, LXPR_PID_FD_FD, p, fd);
+
+	/*
+	 * Drop p_lock, but keep the process P_PR_LOCK'd to prevent it from
+	 * going away while we dereference into fi_list.
+	 */
+	mutex_exit(&p->p_lock);
+
+	/*
+	 * get open file info
+	 */
+	fip = (&(p)->p_user.u_finfo);
+	mutex_enter(&fip->fi_lock);
+
+	if (fd < fip->fi_nfiles) {
+		UF_ENTER(ufp, fip, fd);
+		/*
+		 * ensure the fd is still kosher.
+		 * it may have gone between the readdir and
+		 * the lookup
+		 */
+		if (fip->fi_list[fd].uf_file == NULL) {
+			mutex_exit(&fip->fi_lock);
+			UF_EXIT(ufp);
+			mutex_enter(&p->p_lock);
+			lxpr_unlock(p);
+			lxpr_freenode(lxpnp);
+			return (NULL);
+		}
+
+		if ((fp = ufp->uf_file) != NULL)
+			vp = fp->f_vnode;
+		UF_EXIT(ufp);
+	}
+	mutex_exit(&fip->fi_lock);
+
+	if (vp == NULL) {
+		mutex_enter(&p->p_lock);
+		lxpr_unlock(p);
+		lxpr_freenode(lxpnp);
+		return (NULL);
+	} else {
+		/*
+		 * Fill in the lxpr_node so future references will be able to
+		 * find the underlying vnode. The vnode is held on the realvp.
+		 */
+		lxpnp->lxpr_realvp = vp;
+		VN_HOLD(lxpnp->lxpr_realvp);
+	}
+
+	mutex_enter(&p->p_lock);
+	lxpr_unlock(p);
+	dp = LXPTOV(lxpnp);
+	ASSERT(dp != NULL);
+
+	return (dp);
+}
+
+static vnode_t *
+lxpr_lookup_netdir(vnode_t *dp, char *comp)
+{
+	ASSERT(VTOLXP(dp)->lxpr_type == LXPR_NETDIR);
+
+	dp = lxpr_lookup_common(dp, comp, NULL, netdir, NETDIRFILES);
+
+	return (dp);
+}
+
+static vnode_t *
+lxpr_lookup_procdir(vnode_t *dp, char *comp)
+{
+	ASSERT(VTOLXP(dp)->lxpr_type == LXPR_PROCDIR);
+
+	/*
+	 * We know all the names of files & dirs in our file system structure
+	 * except those that are pid names.  These change as pids are created/
+	 * deleted etc., so we just look for a number as the first char to see
+	 * if we are we doing pid lookups.
+	 *
+	 * Don't need to check for "self" as it is implemented as a symlink
+	 */
+	if (*comp >= '0' && *comp <= '9') {
+		pid_t pid = 0;
+		lxpr_node_t *lxpnp = NULL;
+		proc_t *p;
+		int c;
+
+		while ((c = *comp++) != '\0')
+			pid = 10 * pid + c - '0';
+
+		/*
+		 * Can't continue if the process is still loading or it doesn't
+		 * really exist yet (or maybe it just died!)
+		 */
+		p = lxpr_lock(pid);
+		if (p == NULL)
+			return (NULL);
+
+		if (secpolicy_basic_procinfo(CRED(), p, curproc) != 0) {
+			lxpr_unlock(p);
+			return (NULL);
+		}
+
+		/*
+		 * allocate and fill in a new lxpr node
+		 */
+		lxpnp = lxpr_getnode(dp, LXPR_PIDDIR, p, 0);
+
+		lxpr_unlock(p);
+
+		dp = LXPTOV(lxpnp);
+		ASSERT(dp != NULL);
+
+		return (dp);
+	}
+
+	/* Lookup fixed names */
+	return (lxpr_lookup_common(dp, comp, NULL, lxpr_dir, PROCDIRFILES));
+}
+
+/*
+ * lxpr_readdir(): Vnode operation for VOP_READDIR()
+ */
+/* ARGSUSED */
+static int
+lxpr_readdir(vnode_t *dp, uio_t *uiop, cred_t *cr, int *eofp,
+	caller_context_t *ct, int flags)
+{
+	lxpr_node_t *lxpnp = VTOLXP(dp);
+	lxpr_nodetype_t type = lxpnp->lxpr_type;
+	ssize_t uresid;
+	off_t uoffset;
+	int error;
+
+	ASSERT(dp->v_type == VDIR);
+	ASSERT(type < LXPR_NFILES);
+
+	/*
+	 * we should never get here because the readdir
+	 * is done on the realvp for these nodes
+	 */
+	ASSERT(type != LXPR_PID_FD_FD &&
+	    type != LXPR_PID_CURDIR &&
+	    type != LXPR_PID_ROOTDIR);
+
+	/*
+	 * restrict readdir permission to owner or root
+	 */
+	if ((error = lxpr_access(dp, VREAD, 0, cr, ct)) != 0)
+		return (error);
+
+	uoffset = uiop->uio_offset;
+	uresid = uiop->uio_resid;
+
+	/* can't do negative reads */
+	if (uoffset < 0 || uresid <= 0)
+		return (EINVAL);
+
+	/* can't read directory entries that don't exist! */
+	if (uoffset % LXPR_SDSIZE)
+		return (ENOENT);
+
+	return (lxpr_readdir_function[lxpnp->lxpr_type](lxpnp, uiop, eofp));
+}
+
+/* ARGSUSED */
+static int
+lxpr_readdir_not_a_dir(lxpr_node_t *lxpnp, uio_t *uiop, int *eofp)
+{
+	return (ENOTDIR);
+}
+
+/*
+ * This has the common logic for returning directory entries
+ */
+static int
+lxpr_readdir_common(lxpr_node_t *lxpnp, uio_t *uiop, int *eofp,
+    lxpr_dirent_t *dirtab, int dirtablen)
+{
+	/* bp holds one dirent64 structure */
+	longlong_t bp[DIRENT64_RECLEN(LXPNSIZ) / sizeof (longlong_t)];
+	dirent64_t *dirent = (dirent64_t *)bp;
+	ssize_t oresid;	/* save a copy for testing later */
+	ssize_t uresid;
+
+	oresid = uiop->uio_resid;
+
+	/* clear out the dirent buffer */
+	bzero(bp, sizeof (bp));
+
+	/*
+	 * Satisfy user request
+	 */
+	while ((uresid = uiop->uio_resid) > 0) {
+		int dirindex;
+		off_t uoffset;
+		int reclen;
+		int error;
+
+		uoffset = uiop->uio_offset;
+		dirindex  = (uoffset / LXPR_SDSIZE) - 2;
+
+		if (uoffset == 0) {
+
+			dirent->d_ino = lxpnp->lxpr_ino;
+			dirent->d_name[0] = '.';
+			dirent->d_name[1] = '\0';
+			reclen = DIRENT64_RECLEN(1);
+
+		} else if (uoffset == LXPR_SDSIZE) {
+
+			dirent->d_ino = lxpr_parentinode(lxpnp);
+			dirent->d_name[0] = '.';
+			dirent->d_name[1] = '.';
+			dirent->d_name[2] = '\0';
+			reclen = DIRENT64_RECLEN(2);
+
+		} else if (dirindex >= 0 && dirindex < dirtablen) {
+			int slen = strlen(dirtab[dirindex].d_name);
+
+			dirent->d_ino = lxpr_inode(dirtab[dirindex].d_type,
+			    lxpnp->lxpr_pid, 0);
+
+			VERIFY(slen < LXPNSIZ);
+			(void) strcpy(dirent->d_name, dirtab[dirindex].d_name);
+			reclen = DIRENT64_RECLEN(slen);
+
+		} else {
+			/* Run out of table entries */
+			if (eofp) {
+				*eofp = 1;
+			}
+			return (0);
+		}
+
+		dirent->d_off = (off64_t)(uoffset + LXPR_SDSIZE);
+		dirent->d_reclen = (ushort_t)reclen;
+
+		/*
+		 * if the size of the data to transfer is greater
+		 * that that requested then we can't do it this transfer.
+		 */
+		if (reclen > uresid) {
+			/*
+			 * Error if no entries have been returned yet.
+			 */
+			if (uresid == oresid) {
+				return (EINVAL);
+			}
+			break;
+		}
+
+		/*
+		 * uiomove() updates both uiop->uio_resid and uiop->uio_offset
+		 * by the same amount.  But we want uiop->uio_offset to change
+		 * in increments of LXPR_SDSIZE, which is different from the
+		 * number of bytes being returned to the user.  So we set
+		 * uiop->uio_offset separately, ignoring what uiomove() does.
+		 */
+		if ((error = uiomove((caddr_t)dirent, reclen, UIO_READ,
+		    uiop)) != 0)
+			return (error);
+
+		uiop->uio_offset = uoffset + LXPR_SDSIZE;
+	}
+
+	/* Have run out of space, but could have just done last table entry */
+	if (eofp) {
+		*eofp =
+		    (uiop->uio_offset >= ((dirtablen+2) * LXPR_SDSIZE)) ? 1 : 0;
+	}
+	return (0);
+}
+
+
+static int
+lxpr_readdir_procdir(lxpr_node_t *lxpnp, uio_t *uiop, int *eofp)
+{
+	/* bp holds one dirent64 structure */
+	longlong_t bp[DIRENT64_RECLEN(LXPNSIZ) / sizeof (longlong_t)];
+	dirent64_t *dirent = (dirent64_t *)bp;
+	ssize_t oresid;	/* save a copy for testing later */
+	ssize_t uresid;
+	off_t uoffset;
+	zoneid_t zoneid;
+	pid_t pid;
+	int error;
+	int ceof;
+
+	ASSERT(lxpnp->lxpr_type == LXPR_PROCDIR);
+
+	oresid = uiop->uio_resid;
+	zoneid = LXPTOZ(lxpnp)->zone_id;
+
+	/*
+	 * We return directory entries in the order: "." and ".." then the
+	 * unique lxproc files, then the directories corresponding to the
+	 * running processes.  We have defined this as the ordering because
+	 * it allows us to more easily keep track of where we are betwen calls
+	 * to getdents().  If the number of processes changes between calls
+	 * then we can't lose track of where we are in the lxproc files.
+	 */
+
+	/* Do the fixed entries */
+	error = lxpr_readdir_common(lxpnp, uiop, &ceof, lxpr_dir,
+	    PROCDIRFILES);
+
+	/* Finished if we got an error or if we couldn't do all the table */
+	if (error != 0 || ceof == 0)
+		return (error);
+
+	/* clear out the dirent buffer */
+	bzero(bp, sizeof (bp));
+
+	/* Do the process entries */
+	while ((uresid = uiop->uio_resid) > 0) {
+		proc_t *p;
+		int len;
+		int reclen;
+		int i;
+
+		uoffset = uiop->uio_offset;
+
+		/*
+		 * Stop when entire proc table has been examined.
+		 */
+		i = (uoffset / LXPR_SDSIZE) - 2 - PROCDIRFILES;
+		if (i < 0 || i >= v.v_proc) {
+			/* Run out of table entries */
+			if (eofp) {
+				*eofp = 1;
+			}
+			return (0);
+		}
+		mutex_enter(&pidlock);
+
+		/*
+		 * Skip indices for which there is no pid_entry, PIDs for
+		 * which there is no corresponding process, a PID of 0,
+		 * and anything the security policy doesn't allow
+		 * us to look at.
+		 */
+		if ((p = pid_entry(i)) == NULL || p->p_stat == SIDL ||
+		    p->p_pid == 0 ||
+		    secpolicy_basic_procinfo(CRED(), p, curproc) != 0) {
+			mutex_exit(&pidlock);
+			goto next;
+		}
+		mutex_exit(&pidlock);
+
+		/*
+		 * Convert pid to the Linux default of 1 if we're the zone's
+		 * init process, otherwise use the value from the proc
+		 * structure
+		 */
+		pid = ((p->p_pid != curproc->p_zone->zone_proc_initpid) ?
+		    p->p_pid : 1);
+
+		/*
+		 * If this /proc was mounted in the global zone, view
+		 * all procs; otherwise, only view zone member procs.
+		 */
+		if (zoneid != GLOBAL_ZONEID && p->p_zone->zone_id != zoneid) {
+			goto next;
+		}
+
+		ASSERT(p->p_stat != 0);
+
+		dirent->d_ino = lxpr_inode(LXPR_PIDDIR, pid, 0);
+		len = snprintf(dirent->d_name, LXPNSIZ, "%d", pid);
+		ASSERT(len < LXPNSIZ);
+		reclen = DIRENT64_RECLEN(len);
+
+		dirent->d_off = (off64_t)(uoffset + LXPR_SDSIZE);
+		dirent->d_reclen = (ushort_t)reclen;
+
+		/*
+		 * if the size of the data to transfer is greater
+		 * that that requested then we can't do it this transfer.
+		 */
+		if (reclen > uresid) {
+			/*
+			 * Error if no entries have been returned yet.
+			 */
+			if (uresid == oresid)
+				return (EINVAL);
+			break;
+		}
+
+		/*
+		 * uiomove() updates both uiop->uio_resid and uiop->uio_offset
+		 * by the same amount.  But we want uiop->uio_offset to change
+		 * in increments of LXPR_SDSIZE, which is different from the
+		 * number of bytes being returned to the user.  So we set
+		 * uiop->uio_offset separately, in the increment of this for
+		 * the loop, ignoring what uiomove() does.
+		 */
+		if ((error = uiomove((caddr_t)dirent, reclen, UIO_READ,
+		    uiop)) != 0)
+			return (error);
+next:
+		uiop->uio_offset = uoffset + LXPR_SDSIZE;
+	}
+
+	if (eofp != NULL) {
+		*eofp = (uiop->uio_offset >=
+		    ((v.v_proc + PROCDIRFILES + 2) * LXPR_SDSIZE)) ? 1 : 0;
+	}
+
+	return (0);
+}
+
+static int
+lxpr_readdir_piddir(lxpr_node_t *lxpnp, uio_t *uiop, int *eofp)
+{
+	proc_t *p;
+
+	ASSERT(lxpnp->lxpr_type == LXPR_PIDDIR);
+
+	/* can't read its contents if it died */
+	mutex_enter(&pidlock);
+
+	p = prfind((lxpnp->lxpr_pid == 1) ?
+	    curproc->p_zone->zone_proc_initpid : lxpnp->lxpr_pid);
+
+	if (p == NULL || p->p_stat == SIDL) {
+		mutex_exit(&pidlock);
+		return (ENOENT);
+	}
+	mutex_exit(&pidlock);
+
+	return (lxpr_readdir_common(lxpnp, uiop, eofp, piddir, PIDDIRFILES));
+}
+
+static int
+lxpr_readdir_netdir(lxpr_node_t *lxpnp, uio_t *uiop, int *eofp)
+{
+	ASSERT(lxpnp->lxpr_type == LXPR_NETDIR);
+	return (lxpr_readdir_common(lxpnp, uiop, eofp, netdir, NETDIRFILES));
+}
+
+static int
+lxpr_readdir_fddir(lxpr_node_t *lxpnp, uio_t *uiop, int *eofp)
+{
+	/* bp holds one dirent64 structure */
+	longlong_t bp[DIRENT64_RECLEN(LXPNSIZ) / sizeof (longlong_t)];
+	dirent64_t *dirent = (dirent64_t *)bp;
+	ssize_t oresid;	/* save a copy for testing later */
+	ssize_t uresid;
+	off_t uoffset;
+	int error;
+	int ceof;
+	proc_t *p;
+	int fddirsize = -1;
+	uf_info_t *fip;
+
+	ASSERT(lxpnp->lxpr_type == LXPR_PID_FDDIR);
+
+	oresid = uiop->uio_resid;
+
+	/* can't read its contents if it died */
+	p = lxpr_lock(lxpnp->lxpr_pid);
+	if (p == NULL)
+		return (ENOENT);
+
+	if ((p->p_stat == SZOMB) || (p->p_flag & SSYS) || (p->p_as == &kas))
+		fddirsize = 0;
+
+	/*
+	 * Drop p_lock, but keep the process P_PR_LOCK'd to prevent it from
+	 * going away while we iterate over its fi_list.
+	 */
+	mutex_exit(&p->p_lock);
+
+	/* Get open file info */
+	fip = (&(p)->p_user.u_finfo);
+	mutex_enter(&fip->fi_lock);
+
+	if (fddirsize == -1)
+		fddirsize = fip->fi_nfiles;
+
+	/* Do the fixed entries (in this case just "." & "..") */
+	error = lxpr_readdir_common(lxpnp, uiop, &ceof, 0, 0);
+
+	/* Finished if we got an error or if we couldn't do all the table */
+	if (error != 0 || ceof == 0)
+		goto out;
+
+	/* clear out the dirent buffer */
+	bzero(bp, sizeof (bp));
+
+	/*
+	 * Loop until user's request is satisfied or until
+	 * all file descriptors have been examined.
+	 */
+	for (; (uresid = uiop->uio_resid) > 0;
+	    uiop->uio_offset = uoffset + LXPR_SDSIZE) {
+		int reclen;
+		int fd;
+		int len;
+
+		uoffset = uiop->uio_offset;
+
+		/*
+		 * Stop at the end of the fd list
+		 */
+		fd = (uoffset / LXPR_SDSIZE) - 2;
+		if (fd < 0 || fd >= fddirsize) {
+			if (eofp) {
+				*eofp = 1;
+			}
+			goto out;
+		}
+
+		if (fip->fi_list[fd].uf_file == NULL)
+			continue;
+
+		dirent->d_ino = lxpr_inode(LXPR_PID_FD_FD, lxpnp->lxpr_pid, fd);
+		len = snprintf(dirent->d_name, LXPNSIZ, "%d", fd);
+		ASSERT(len < LXPNSIZ);
+		reclen = DIRENT64_RECLEN(len);
+
+		dirent->d_off = (off64_t)(uoffset + LXPR_SDSIZE);
+		dirent->d_reclen = (ushort_t)reclen;
+
+		if (reclen > uresid) {
+			/*
+			 * Error if no entries have been returned yet.
+			 */
+			if (uresid == oresid)
+				error = EINVAL;
+			goto out;
+		}
+
+		if ((error = uiomove((caddr_t)dirent, reclen, UIO_READ,
+		    uiop)) != 0)
+			goto out;
+	}
+
+	if (eofp != NULL) {
+		*eofp =
+		    (uiop->uio_offset >= ((fddirsize+2) * LXPR_SDSIZE)) ? 1 : 0;
+	}
+
+out:
+	mutex_exit(&fip->fi_lock);
+	mutex_enter(&p->p_lock);
+	lxpr_unlock(p);
+	return (error);
+}
+
+
+/*
+ * lxpr_readlink(): Vnode operation for VOP_READLINK()
+ */
+/* ARGSUSED */
+static int
+lxpr_readlink(vnode_t *vp, uio_t *uiop, cred_t *cr, caller_context_t *ct)
+{
+	char bp[MAXPATHLEN + 1];
+	size_t buflen = sizeof (bp);
+	lxpr_node_t *lxpnp = VTOLXP(vp);
+	vnode_t *rvp = lxpnp->lxpr_realvp;
+	pid_t pid;
+	int error = 0;
+
+	/* must be a symbolic link file */
+	if (vp->v_type != VLNK)
+		return (EINVAL);
+
+	/* Try to produce a symlink name for anything that has a realvp */
+	if (rvp != NULL) {
+		if ((error = lxpr_access(vp, VREAD, 0, CRED(), ct)) != 0)
+			return (error);
+		if ((error = vnodetopath(NULL, rvp, bp, buflen, CRED())) != 0)
+			return (error);
+	} else {
+		switch (lxpnp->lxpr_type) {
+		case LXPR_SELF:
+			/*
+			 * Convert pid to the Linux default of 1 if we're the
+			 * zone's init process
+			 */
+			pid = ((curproc->p_pid !=
+			    curproc->p_zone->zone_proc_initpid)
+			    ? curproc->p_pid : 1);
+
+			/*
+			 * Don't need to check result as every possible int
+			 * will fit within MAXPATHLEN bytes.
+			 */
+			(void) snprintf(bp, buflen, "%d", pid);
+			break;
+		case LXPR_PID_CURDIR:
+		case LXPR_PID_ROOTDIR:
+		case LXPR_PID_EXE:
+			return (EACCES);
+		default:
+			/*
+			 * Need to return error so that nothing thinks
+			 * that the symlink is empty and hence "."
+			 */
+			return (EINVAL);
+		}
+	}
+
+	/* copy the link data to user space */
+	return (uiomove(bp, strlen(bp), UIO_READ, uiop));
+}
+
+/*
+ * lxpr_inactive(): Vnode operation for VOP_INACTIVE()
+ * Vnode is no longer referenced, deallocate the file
+ * and all its resources.
+ */
+/* ARGSUSED */
+static void
+lxpr_inactive(vnode_t *vp, cred_t *cr, caller_context_t *ct)
+{
+	lxpr_freenode(VTOLXP(vp));
+}
+
+/*
+ * lxpr_sync(): Vnode operation for VOP_SYNC()
+ */
+static int
+lxpr_sync()
+{
+	/*
+	 * Nothing to sync but this function must never fail
+	 */
+	return (0);
+}
+
+/*
+ * lxpr_cmp(): Vnode operation for VOP_CMP()
+ */
+static int
+lxpr_cmp(vnode_t *vp1, vnode_t *vp2, caller_context_t *ct)
+{
+	vnode_t *rvp;
+
+	while (vn_matchops(vp1, lxpr_vnodeops) &&
+	    (rvp = VTOLXP(vp1)->lxpr_realvp) != NULL) {
+		vp1 = rvp;
+	}
+
+	while (vn_matchops(vp2, lxpr_vnodeops) &&
+	    (rvp = VTOLXP(vp2)->lxpr_realvp) != NULL) {
+		vp2 = rvp;
+	}
+
+	if (vn_matchops(vp1, lxpr_vnodeops) || vn_matchops(vp2, lxpr_vnodeops))
+		return (vp1 == vp2);
+
+	return (VOP_CMP(vp1, vp2, ct));
+}
+
+/*
+ * lxpr_realvp(): Vnode operation for VOP_REALVP()
+ */
+static int
+lxpr_realvp(vnode_t *vp, vnode_t **vpp, caller_context_t *ct)
+{
+	vnode_t *rvp;
+
+	if ((rvp = VTOLXP(vp)->lxpr_realvp) != NULL) {
+		vp = rvp;
+		if (VOP_REALVP(vp, &rvp, ct) == 0)
+			vp = rvp;
+	}
+
+	*vpp = vp;
+	return (0);
+}
diff --git a/usr/src/uts/common/fs/lxproc/lxproc.h b/usr/src/uts/common/fs/lxproc/lxproc.h
new file mode 100644
index 0000000000..e3718372e0
--- /dev/null
+++ b/usr/src/uts/common/fs/lxproc/lxproc.h
@@ -0,0 +1,277 @@
+/*
+ * CDDL HEADER START
+ *
+ * The contents of this file are subject to the terms of the
+ * Common Development and Distribution License (the "License").
+ * You may not use this file except in compliance with the License.
+ *
+ * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
+ * or http://www.opensolaris.org/os/licensing.
+ * See the License for the specific language governing permissions
+ * and limitations under the License.
+ *
+ * When distributing Covered Code, include this CDDL HEADER in each
+ * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
+ * If applicable, add the following below this CDDL HEADER, with the
+ * fields enclosed by brackets "[]" replaced with your own identifying
+ * information: Portions Copyright [yyyy] [name of copyright owner]
+ *
+ * CDDL HEADER END
+ */
+/*
+ * Copyright 2009 Sun Microsystems, Inc.  All rights reserved.
+ * Use is subject to license terms.
+ */
+
+/*
+ * Copyright 2015 Joyent, Inc. All rights reserved.
+ */
+
+#ifdef _LXPROC_BRANDED_H
+#error Attempted to include native lxproc.h after branded lx_proc.h
+#endif
+
+#ifndef	_LXPROC_H
+#define	_LXPROC_H
+#define	_LXPROC_NATIVE_H
+
+#ifdef	__cplusplus
+extern "C" {
+#endif
+
+/*
+ * lxproc.h: declarations, data structures and macros for lxprocfs
+ */
+#include <sys/types.h>
+#include <sys/param.h>
+#include <sys/policy.h>
+#include <sys/debug.h>
+#include <sys/dirent.h>
+#include <sys/errno.h>
+#include <sys/file.h>
+#include <sys/kmem.h>
+#include <sys/pathname.h>
+#include <sys/proc.h>
+#include <sys/systm.h>
+#include <sys/var.h>
+#include <sys/user.h>
+#include <sys/t_lock.h>
+#include <sys/sysmacros.h>
+#include <sys/cred.h>
+#include <sys/priv.h>
+#include <sys/vnode.h>
+#include <sys/vfs.h>
+#include <sys/statvfs.h>
+#include <sys/cmn_err.h>
+#include <sys/zone.h>
+#include <sys/uio.h>
+#include <sys/utsname.h>
+#include <sys/dnlc.h>
+#include <sys/atomic.h>
+#include <sys/sunddi.h>
+#include <sys/sunldi.h>
+#include <vm/as.h>
+#include <vm/anon.h>
+
+#define	LX_SIGHUP	1
+#define	LX_SIGINT	2
+#define	LX_SIGQUIT	3
+#define	LX_SIGILL	4
+#define	LX_SIGTRAP	5
+#define	LX_SIGABRT	6
+#define	LX_SIGIOT	6
+#define	LX_SIGBUS	7
+#define	LX_SIGFPE	8
+#define	LX_SIGKILL	9
+#define	LX_SIGUSR1	10
+#define	LX_SIGSEGV	11
+#define	LX_SIGUSR2	12
+#define	LX_SIGPIPE	13
+#define	LX_SIGALRM	14
+#define	LX_SIGTERM	15
+#define	LX_SIGSTKFLT	16
+#define	LX_SIGCHLD	17
+#define	LX_SIGCONT	18
+#define	LX_SIGSTOP	19
+#define	LX_SIGTSTP	20
+#define	LX_SIGTTIN	21
+#define	LX_SIGTTOU	22
+#define	LX_SIGURG	23
+#define	LX_SIGXCPU	24
+#define	LX_SIGXFSZ	25
+#define	LX_SIGVTALRM	26
+#define	LX_SIGPROF	27
+#define	LX_SIGWINCH	28
+#define	LX_SIGIO	29
+#define	LX_SIGPOLL	LX_SIGIO
+#define	LX_SIGPWR	30
+#define	LX_SIGSYS	31
+#define	LX_SIGUNUSED	31
+
+#define	LX_NSIG		64	/* Linux _NSIG */
+
+#define	LX_SIGRTMIN	32
+#define	LX_SIGRTMAX	LX_NSIG
+
+/*
+ * Convert a vnode into an lxpr_mnt_t
+ */
+#define	VTOLXPM(vp)	((lxpr_mnt_t *)(vp)->v_vfsp->vfs_data)
+
+/*
+ * convert a vnode into an lxpr_node
+ */
+#define	VTOLXP(vp)	((lxpr_node_t *)(vp)->v_data)
+
+/*
+ * convert a lxprnode into a vnode
+ */
+#define	LXPTOV(lxpnp)	((lxpnp)->lxpr_vnode)
+
+/*
+ * convert a lxpr_node into zone for fs
+ */
+#define	LXPTOZ(lxpnp) \
+	(((lxpr_mnt_t *)(lxpnp)->lxpr_vnode->v_vfsp->vfs_data)->lxprm_zone)
+
+#define	LXPNSIZ		256	/* max size of lx /proc file name entries */
+
+/*
+ * Pretend that a directory entry takes 16 bytes
+ */
+#define	LXPR_SDSIZE	16
+
+/*
+ * Node/file types for lx /proc files
+ * (directories and files contained therein).
+ */
+typedef enum lxpr_nodetype {
+	LXPR_PROCDIR,		/* /proc		*/
+	LXPR_PIDDIR,		/* /proc/<pid>		*/
+	LXPR_PID_CMDLINE,	/* /proc/<pid>/cmdline	*/
+	LXPR_PID_CPU,		/* /proc/<pid>/cpu	*/
+	LXPR_PID_CURDIR,	/* /proc/<pid>/cwd	*/
+	LXPR_PID_ENV,		/* /proc/<pid>/environ	*/
+	LXPR_PID_EXE,		/* /proc/<pid>/exe	*/
+	LXPR_PID_MAPS,		/* /proc/<pid>/maps	*/
+	LXPR_PID_MEM,		/* /proc/<pid>/mem	*/
+	LXPR_PID_ROOTDIR,	/* /proc/<pid>/root	*/
+	LXPR_PID_STAT,		/* /proc/<pid>/stat	*/
+	LXPR_PID_STATM,		/* /proc/<pid>/statm	*/
+	LXPR_PID_STATUS,	/* /proc/<pid>/status	*/
+	LXPR_PID_FDDIR,		/* /proc/<pid>/fd	*/
+	LXPR_PID_FD_FD,		/* /proc/<pid>/fd/nn	*/
+	LXPR_CMDLINE,		/* /proc/cmdline	*/
+	LXPR_CPUINFO,		/* /proc/cpuinfo	*/
+	LXPR_DEVICES,		/* /proc/devices	*/
+	LXPR_DMA,		/* /proc/dma		*/
+	LXPR_FILESYSTEMS,	/* /proc/filesystems	*/
+	LXPR_INTERRUPTS,	/* /proc/interrupts	*/
+	LXPR_IOPORTS,		/* /proc/ioports	*/
+	LXPR_KCORE,		/* /proc/kcore		*/
+	LXPR_KMSG,		/* /proc/kmsg		*/
+	LXPR_LOADAVG,		/* /proc/loadavg	*/
+	LXPR_MEMINFO,		/* /proc/meminfo	*/
+	LXPR_MOUNTS,		/* /proc/mounts		*/
+	LXPR_NETDIR,		/* /proc/net		*/
+	LXPR_NET_ARP,		/* /proc/net/arp	*/
+	LXPR_NET_DEV,		/* /proc/net/dev	*/
+	LXPR_NET_DEV_MCAST,	/* /proc/net/dev_mcast	*/
+	LXPR_NET_IGMP,		/* /proc/net/igmp	*/
+	LXPR_NET_IP_MR_CACHE,	/* /proc/net/ip_mr_cache */
+	LXPR_NET_IP_MR_VIF,	/* /proc/net/ip_mr_vif	*/
+	LXPR_NET_MCFILTER,	/* /proc/net/mcfilter	*/
+	LXPR_NET_NETSTAT,	/* /proc/net/netstat	*/
+	LXPR_NET_RAW,		/* /proc/net/raw	*/
+	LXPR_NET_ROUTE,		/* /proc/net/route	*/
+	LXPR_NET_RPC,		/* /proc/net/rpc	*/
+	LXPR_NET_RT_CACHE,	/* /proc/net/rt_cache	*/
+	LXPR_NET_SOCKSTAT,	/* /proc/net/sockstat	*/
+	LXPR_NET_SNMP,		/* /proc/net/snmp	*/
+	LXPR_NET_STAT,		/* /proc/net/stat	*/
+	LXPR_NET_TCP,		/* /proc/net/tcp	*/
+	LXPR_NET_UDP,		/* /proc/net/udp	*/
+	LXPR_NET_UNIX,		/* /proc/net/unix	*/
+	LXPR_PARTITIONS,	/* /proc/partitions	*/
+	LXPR_SELF,		/* /proc/self		*/
+	LXPR_STAT,		/* /proc/stat		*/
+	LXPR_UPTIME,		/* /proc/uptime		*/
+	LXPR_VERSION,		/* /proc/version	*/
+	LXPR_NFILES		/* number of lx /proc file types */
+} lxpr_nodetype_t;
+
+/*
+ * Number of fds allowed for in the inode number calculation
+ * per process (if a process has more fds then inode numbers
+ * may be duplicated)
+ */
+#define	LXPR_FD_PERPROC 2000
+
+/*
+ * external dirent characteristics
+ */
+#define	LXPRMAXNAMELEN	14
+typedef struct {
+	lxpr_nodetype_t	d_type;
+	char		d_name[LXPRMAXNAMELEN];
+} lxpr_dirent_t;
+
+/*
+ * This is the lxprocfs private data object
+ * which is attached to v_data in the vnode structure
+ */
+typedef struct lxpr_node {
+	lxpr_nodetype_t	lxpr_type;	/* type of this node 		*/
+	vnode_t		*lxpr_vnode;	/* vnode for the node		*/
+	vnode_t		*lxpr_parent;	/* parent directory		*/
+	vnode_t		*lxpr_realvp;	/* real vnode, file in dirs	*/
+	timestruc_t	lxpr_time;	/* creation etc time for file	*/
+	mode_t		lxpr_mode;	/* file mode bits		*/
+	uid_t		lxpr_uid;	/* file owner			*/
+	gid_t		lxpr_gid;	/* file group owner		*/
+	pid_t		lxpr_pid;	/* pid of proc referred to	*/
+	ino_t		lxpr_ino;	/* node id 			*/
+} lxpr_node_t;
+
+struct zone;    /* forward declaration */
+
+/*
+ * This is the lxprocfs private data object
+ * which is attached to vfs_data in the vfs structure
+ */
+typedef struct lxpr_mnt {
+	lxpr_node_t	*lxprm_node;	/* node at root of proc mount */
+	struct zone	*lxprm_zone;	/* zone for this mount */
+	ldi_ident_t	lxprm_li;	/* ident for ldi */
+} lxpr_mnt_t;
+
+extern vnodeops_t	*lxpr_vnodeops;
+extern int		nproc_highbit;	/* highbit(v.v_nproc)		*/
+
+typedef struct mounta	mounta_t;
+
+extern void lxpr_initnodecache();
+extern void lxpr_fininodecache();
+extern void lxpr_initrootnode(lxpr_node_t **, vfs_t *);
+extern ino_t lxpr_inode(lxpr_nodetype_t, pid_t, int);
+extern ino_t lxpr_parentinode(lxpr_node_t *);
+extern lxpr_node_t *lxpr_getnode(vnode_t *, lxpr_nodetype_t, proc_t *, int);
+extern void lxpr_freenode(lxpr_node_t *);
+
+typedef struct lxpr_uiobuf lxpr_uiobuf_t;
+extern lxpr_uiobuf_t *lxpr_uiobuf_new(uio_t *);
+extern void lxpr_uiobuf_free(lxpr_uiobuf_t *);
+extern int lxpr_uiobuf_flush(lxpr_uiobuf_t *);
+extern void lxpr_uiobuf_seek(lxpr_uiobuf_t *, offset_t);
+extern void lxpr_uiobuf_write(lxpr_uiobuf_t *, const char *, size_t);
+extern void lxpr_uiobuf_printf(lxpr_uiobuf_t *, const char *, ...);
+extern void lxpr_uiobuf_seterr(lxpr_uiobuf_t *, int);
+
+proc_t *lxpr_lock(pid_t);
+void lxpr_unlock(proc_t *);
+
+#ifdef	__cplusplus
+}
+#endif
+
+#endif /* _LXPROC_H */
diff --git a/usr/src/uts/common/fs/nfs/nfs3_vfsops.c b/usr/src/uts/common/fs/nfs/nfs3_vfsops.c
index 207a708771..2176dcb9de 100644
--- a/usr/src/uts/common/fs/nfs/nfs3_vfsops.c
+++ b/usr/src/uts/common/fs/nfs/nfs3_vfsops.c
@@ -20,6 +20,7 @@
  */
 /*
  * Copyright (c) 1986, 2010, Oracle and/or its affiliates. All rights reserved.
+ * Copyright 2013, Joyent, Inc. All rights reserved.
  */
 
 /*
diff --git a/usr/src/uts/common/fs/nfs/nfs3_vnops.c b/usr/src/uts/common/fs/nfs/nfs3_vnops.c
index b7354c168a..cd1e08d5d7 100644
--- a/usr/src/uts/common/fs/nfs/nfs3_vnops.c
+++ b/usr/src/uts/common/fs/nfs/nfs3_vnops.c
@@ -29,7 +29,7 @@
  */
 
 /*
- * Copyright (c) 2013, Joyent, Inc. All rights reserved.
+ * Copyright (c) 2014, Joyent, Inc. All rights reserved.
  * Copyright 2015 Nexenta Systems, Inc.  All rights reserved.
  */
 
@@ -3353,10 +3353,9 @@ nfs3rename(vnode_t *odvp, char *onm, vnode_t *ndvp, char *nnm, cred_t *cr,
 		if (nvp)
 			vnevent_rename_dest(nvp, ndvp, nnm, ct);
 
-		if (odvp != ndvp)
-			vnevent_rename_dest_dir(ndvp, ct);
 		ASSERT(ovp != NULL);
 		vnevent_rename_src(ovp, odvp, onm, ct);
+		vnevent_rename_dest_dir(ndvp, ovp, nnm, ct);
 	}
 
 	if (nvp) {
diff --git a/usr/src/uts/common/fs/nfs/nfs4_srv_ns.c b/usr/src/uts/common/fs/nfs/nfs4_srv_ns.c
index ce0c9485a6..3ee41939ac 100644
--- a/usr/src/uts/common/fs/nfs/nfs4_srv_ns.c
+++ b/usr/src/uts/common/fs/nfs/nfs4_srv_ns.c
@@ -22,6 +22,7 @@
 /*
  * Copyright 2014 Nexenta Systems, Inc.  All rights reserved.
  * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2015, Joyent, Inc.
  */
 
 #include <sys/systm.h>
@@ -178,12 +179,12 @@ pseudo_exportfs(vnode_t *vp, fid_t *fid, struct exp_visible *vis_head,
 	kex = &exi->exi_export;
 	kex->ex_flags = EX_PSEUDO;
 
-	vpathlen = vp->v_path ? strlen(vp->v_path) : 0;
+	vpathlen = strlen(vp->v_path);
 	kex->ex_pathlen = vpathlen + strlen(PSEUDOFS_SUFFIX);
 	kex->ex_path = kmem_alloc(kex->ex_pathlen + 1, KM_SLEEP);
 
 	if (vpathlen)
-		(void) strcpy(kex->ex_path, vp->v_path);
+		(void) strncpy(kex->ex_path, vp->v_path, vpathlen);
 	(void) strcpy(kex->ex_path + vpathlen, PSEUDOFS_SUFFIX);
 
 	/* Transfer the secinfo data from exdata to this new pseudo node */
diff --git a/usr/src/uts/common/fs/nfs/nfs4_vfsops.c b/usr/src/uts/common/fs/nfs/nfs4_vfsops.c
index 151cb62403..55f6c95289 100644
--- a/usr/src/uts/common/fs/nfs/nfs4_vfsops.c
+++ b/usr/src/uts/common/fs/nfs/nfs4_vfsops.c
@@ -22,6 +22,7 @@
 /*
  * Copyright 2015 Nexenta Systems, Inc.  All rights reserved.
  * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
+ * Copyright 2013, Joyent, Inc. All rights reserved.
  */
 
 /*
diff --git a/usr/src/uts/common/fs/nfs/nfs4_vnops.c b/usr/src/uts/common/fs/nfs/nfs4_vnops.c
index d6bf384a8b..2b3fdfdd55 100644
--- a/usr/src/uts/common/fs/nfs/nfs4_vnops.c
+++ b/usr/src/uts/common/fs/nfs/nfs4_vnops.c
@@ -34,7 +34,7 @@
  */
 
 /*
- * Copyright (c) 2013, Joyent, Inc. All rights reserved.
+ * Copyright (c) 2014, Joyent, Inc. All rights reserved.
  */
 
 #include <sys/param.h>
@@ -8061,8 +8061,9 @@ link_call:
 	 * vnode if it already existed.
 	 */
 	if (error == 0) {
-		vnode_t *tvp;
+		vnode_t *tvp, *tovp;
 		rnode4_t *trp;
+
 		/*
 		 * Notify the vnode. Each links is represented by
 		 * a different vnode, in nfsv4.
@@ -8075,23 +8076,20 @@ link_call:
 			vnevent_rename_dest(tvp, ndvp, nnm, ct);
 		}
 
-		/*
-		 * if the source and destination directory are not the
-		 * same notify the destination directory.
-		 */
-		if (VTOR4(odvp) != VTOR4(ndvp)) {
-			trp = VTOR4(ndvp);
-			tvp = ndvp;
-			if (IS_SHADOW(ndvp, trp))
-				tvp = RTOV4(trp);
-			vnevent_rename_dest_dir(tvp, ct);
-		}
-
 		trp = VTOR4(ovp);
-		tvp = ovp;
+		tovp = ovp;
 		if (IS_SHADOW(ovp, trp))
+			tovp = RTOV4(trp);
+
+		vnevent_rename_src(tovp, odvp, onm, ct);
+
+		trp = VTOR4(ndvp);
+		tvp = ndvp;
+
+		if (IS_SHADOW(ndvp, trp))
 			tvp = RTOV4(trp);
-		vnevent_rename_src(tvp, odvp, onm, ct);
+
+		vnevent_rename_dest_dir(tvp, tovp, nnm, ct);
 	}
 
 	if (nvp) {
diff --git a/usr/src/uts/common/fs/nfs/nfs_auth.c b/usr/src/uts/common/fs/nfs/nfs_auth.c
index 3410340581..a8bcfdf438 100644
--- a/usr/src/uts/common/fs/nfs/nfs_auth.c
+++ b/usr/src/uts/common/fs/nfs/nfs_auth.c
@@ -22,6 +22,7 @@
 /*
  * Copyright 2015 Nexenta Systems, Inc.  All rights reserved.
  * Copyright (c) 1995, 2010, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2015 Joyent, Inc.  All rights reserved.
  */
 
 #include <sys/param.h>
@@ -559,11 +560,16 @@ retry:
 			*access = res.ares.auth_perm;
 			*srv_uid = res.ares.auth_srv_uid;
 			*srv_gid = res.ares.auth_srv_gid;
-			*srv_gids_cnt = res.ares.auth_srv_gids.len;
-			*srv_gids = kmem_alloc(*srv_gids_cnt * sizeof (gid_t),
-			    KM_SLEEP);
-			bcopy(res.ares.auth_srv_gids.val, *srv_gids,
-			    *srv_gids_cnt * sizeof (gid_t));
+
+			if ((*srv_gids_cnt = res.ares.auth_srv_gids.len) != 0) {
+				*srv_gids = kmem_alloc(*srv_gids_cnt *
+				    sizeof (gid_t), KM_SLEEP);
+				bcopy(res.ares.auth_srv_gids.val, *srv_gids,
+				    *srv_gids_cnt * sizeof (gid_t));
+			} else {
+				*srv_gids = NULL;
+			}
+
 			break;
 
 		case NFSAUTH_DR_EFAIL:
@@ -1114,9 +1120,13 @@ wait:
 		if (gid != NULL)
 			*gid = p->auth_srv_gid;
 		if (ngids != NULL && gids != NULL) {
-			*ngids = p->auth_srv_ngids;
-			*gids = kmem_alloc(*ngids * sizeof (gid_t), KM_SLEEP);
-			bcopy(p->auth_srv_gids, *gids, *ngids * sizeof (gid_t));
+			if ((*ngids = p->auth_srv_ngids) != 0) {
+				size_t sz = *ngids * sizeof (gid_t);
+				*gids = kmem_alloc(sz, KM_SLEEP);
+				bcopy(p->auth_srv_gids, *gids, sz);
+			} else {
+				*gids = NULL;
+			}
 		}
 
 		access = p->auth_access;
diff --git a/usr/src/uts/common/fs/nfs/nfs_server.c b/usr/src/uts/common/fs/nfs/nfs_server.c
index 7e94c62734..dcc64def59 100644
--- a/usr/src/uts/common/fs/nfs/nfs_server.c
+++ b/usr/src/uts/common/fs/nfs/nfs_server.c
@@ -23,6 +23,7 @@
  * Copyright (c) 2011 Bayard G. Bell. All rights reserved.
  * Copyright (c) 2013 by Delphix. All rights reserved.
  * Copyright 2014 Nexenta Systems, Inc.  All rights reserved.
+ * Copyright (c) 2015, Joyent, Inc.
  */
 
 /*
@@ -2570,6 +2571,9 @@ nfs_srvinit(void)
 {
 	int error;
 
+	if (getzoneid() != GLOBAL_ZONEID)
+		return (EACCES);
+
 	error = nfs_exportinit();
 	if (error != 0)
 		return (error);
@@ -3208,7 +3212,7 @@ nfs_getflabel(vnode_t *vp, struct exportinfo *exi)
 	char *path;
 
 	mutex_enter(&vp->v_lock);
-	if (vp->v_path != NULL) {
+	if (vp->v_path != vn_vpath_empty) {
 		zone = zone_find_by_any_path(vp->v_path, B_FALSE);
 		mutex_exit(&vp->v_lock);
 	} else {
diff --git a/usr/src/uts/common/fs/nfs/nfs_vfsops.c b/usr/src/uts/common/fs/nfs/nfs_vfsops.c
index 57b21778b4..ffd5380a86 100644
--- a/usr/src/uts/common/fs/nfs/nfs_vfsops.c
+++ b/usr/src/uts/common/fs/nfs/nfs_vfsops.c
@@ -20,6 +20,7 @@
  */
 /*
  * Copyright (c) 1986, 2010, Oracle and/or its affiliates. All rights reserved.
+ * Copyright 2013, Joyent, Inc. All rights reserved.
  *
  *	Copyright (c) 1983,1984,1985,1986,1987,1988,1989  AT&T.
  *	All rights reserved.
diff --git a/usr/src/uts/common/fs/nfs/nfs_vnops.c b/usr/src/uts/common/fs/nfs/nfs_vnops.c
index 1a1082bcb8..8c321c8aa3 100644
--- a/usr/src/uts/common/fs/nfs/nfs_vnops.c
+++ b/usr/src/uts/common/fs/nfs/nfs_vnops.c
@@ -26,7 +26,7 @@
  */
 
 /*
- * Copyright (c) 2013, Joyent, Inc. All rights reserved.
+ * Copyright (c) 2014, Joyent, Inc. All rights reserved.
  * Copyright 2015 Nexenta Systems, Inc.  All rights reserved.
  */
 
@@ -2688,11 +2688,9 @@ nfsrename(vnode_t *odvp, char *onm, vnode_t *ndvp, char *nnm, cred_t *cr,
 		if (nvp)
 			vnevent_rename_dest(nvp, ndvp, nnm, ct);
 
-		if (odvp != ndvp)
-			vnevent_rename_dest_dir(ndvp, ct);
-
 		ASSERT(ovp != NULL);
 		vnevent_rename_src(ovp, odvp, onm, ct);
+		vnevent_rename_dest_dir(ndvp, ovp, nnm, ct);
 	}
 
 	if (nvp) {
diff --git a/usr/src/uts/common/fs/pcfs/pc_dir.c b/usr/src/uts/common/fs/pcfs/pc_dir.c
index a9ee604b7c..21a0b1a4bd 100644
--- a/usr/src/uts/common/fs/pcfs/pc_dir.c
+++ b/usr/src/uts/common/fs/pcfs/pc_dir.c
@@ -24,6 +24,10 @@
  * Use is subject to license terms.
  */
 
+/*
+ * Copyright (c) 2014, Joyent, Inc. All rights reserved.
+ */
+
 #include <sys/param.h>
 #include <sys/errno.h>
 #include <sys/systm.h>
@@ -822,10 +826,10 @@ top:
 			return (error);
 		}
 	}
-out:
-	vnevent_rename_src(PCTOV(pcp), PCTOV(dp), snm, ctp);
-	if (dp != tdp) {
-		vnevent_rename_dest_dir(PCTOV(tdp), ctp);
+
+	if (error == 0) {
+		vnevent_rename_src(PCTOV(pcp), PCTOV(dp), snm, ctp);
+		vnevent_rename_dest_dir(PCTOV(tdp), PCTOV(pcp), tnm, ctp);
 	}
 
 	VN_RELE(PCTOV(pcp));
diff --git a/usr/src/uts/common/fs/portfs/port.c b/usr/src/uts/common/fs/portfs/port.c
index 14be8cbbae..11b7386269 100644
--- a/usr/src/uts/common/fs/portfs/port.c
+++ b/usr/src/uts/common/fs/portfs/port.c
@@ -24,7 +24,9 @@
  * Use is subject to license terms.
  */
 
-#pragma ident	"%Z%%M%	%I%	%E% SMI"
+/*
+ * Copyright (c) 2015 Joyent, Inc.  All rights reserved.
+ */
 
 #include <sys/types.h>
 #include <sys/systm.h>
@@ -1381,12 +1383,18 @@ portnowait:
 
 	if (model == DATAMODEL_NATIVE) {
 		eventsz = sizeof (port_event_t);
-		kevp = kmem_alloc(eventsz * nmax, KM_NOSLEEP);
-		if (kevp == NULL) {
-			if (nmax > pp->port_max_list)
-				nmax = pp->port_max_list;
-			kevp = kmem_alloc(eventsz * nmax, KM_SLEEP);
+
+		if (nmax == 0) {
+			kevp = NULL;
+		} else {
+			kevp = kmem_alloc(eventsz * nmax, KM_NOSLEEP);
+			if (kevp == NULL) {
+				if (nmax > pp->port_max_list)
+					nmax = pp->port_max_list;
+				kevp = kmem_alloc(eventsz * nmax, KM_SLEEP);
+			}
 		}
+
 		results = kevp;
 		lev = NULL;	/* start with first event in the queue */
 		for (nevents = 0; nevents < nmax; ) {
@@ -1423,12 +1431,18 @@ portnowait:
 		port_event32_t	*kevp32;
 
 		eventsz = sizeof (port_event32_t);
-		kevp32 = kmem_alloc(eventsz * nmax, KM_NOSLEEP);
-		if (kevp32 == NULL) {
-			if (nmax > pp->port_max_list)
-				nmax = pp->port_max_list;
-			kevp32 = kmem_alloc(eventsz * nmax, KM_SLEEP);
+
+		if (nmax == 0) {
+			kevp32 = NULL;
+		} else {
+			kevp32 = kmem_alloc(eventsz * nmax, KM_NOSLEEP);
+			if (kevp32 == NULL) {
+				if (nmax > pp->port_max_list)
+					nmax = pp->port_max_list;
+				kevp32 = kmem_alloc(eventsz * nmax, KM_SLEEP);
+			}
 		}
+
 		results = kevp32;
 		lev = NULL;	/* start with first event in the queue */
 		for (nevents = 0; nevents < nmax; ) {
diff --git a/usr/src/uts/common/fs/portfs/port_vnops.c b/usr/src/uts/common/fs/portfs/port_vnops.c
index b2f5088e06..ab95c0a1f8 100644
--- a/usr/src/uts/common/fs/portfs/port_vnops.c
+++ b/usr/src/uts/common/fs/portfs/port_vnops.c
@@ -24,6 +24,10 @@
  * Use is subject to license terms.
  */
 
+/*
+ * Copyright (c) 2014, Joyent, Inc.  All rights reserved.
+ */
+
 #include <sys/types.h>
 #include <sys/vnode.h>
 #include <sys/vfs_opreg.h>
@@ -294,14 +298,10 @@ port_poll(vnode_t *vp, short events, int anyyet, short *reventsp,
 		levents |= POLLOUT;
 	levents &= events;
 	*reventsp = levents;
-	if (levents == 0) {
-		if (!anyyet) {
-			*phpp = &pp->port_pollhd;
-			portq->portq_flags |=
-			    events & POLLIN ? PORTQ_POLLIN : 0;
-			portq->portq_flags |=
-			    events & POLLOUT ? PORTQ_POLLOUT : 0;
-		}
+	if ((levents == 0 && !anyyet) || (events & POLLET)) {
+		*phpp = &pp->port_pollhd;
+		portq->portq_flags |= events & POLLIN ? PORTQ_POLLIN : 0;
+		portq->portq_flags |= events & POLLOUT ? PORTQ_POLLOUT : 0;
 	}
 	mutex_exit(&portq->portq_mutex);
 	return (0);
diff --git a/usr/src/uts/common/fs/proc/prargv.c b/usr/src/uts/common/fs/proc/prargv.c
new file mode 100644
index 0000000000..a4ad82d661
--- /dev/null
+++ b/usr/src/uts/common/fs/proc/prargv.c
@@ -0,0 +1,441 @@
+/*
+ * This file and its contents are supplied under the terms of the
+ * Common Development and Distribution License ("CDDL"), version 1.0.
+ * You may only use this file in accordance with the terms of version
+ * 1.0 of the CDDL.
+ *
+ * A full copy of the text of the CDDL should have accompanied this
+ * source.  A copy of the CDDL is also available via the Internet at
+ * http://www.illumos.org/license/CDDL.
+ */
+
+/*
+ * Copyright 2015, Joyent, Inc.
+ */
+
+#include <sys/types.h>
+#include <sys/sunddi.h>
+#include <sys/proc.h>
+#include <sys/procfs.h>
+#include <sys/sysmacros.h>
+#include <vm/as.h>
+
+/*
+ * Safely read a contiguous region of memory from 'addr' in the address space
+ * of a particular process into the supplied kernel buffer (*buf, sz).
+ * Partially mapped regions will result in a partial read terminating at the
+ * first hole in the address space.  The number of bytes actually read is
+ * returned to the caller via 'rdsz'.
+ */
+static int
+prreadbuf(proc_t *p, uintptr_t ustart, uint8_t *buf, size_t sz, size_t *rdsz)
+{
+	int error = 0;
+	size_t rem = sz;
+	off_t pos = 0;
+
+	if (rdsz != NULL)
+		*rdsz = 0;
+
+	while (rem != 0) {
+		uintptr_t addr = ustart + pos;
+		size_t len = MIN(rem, PAGESIZE - (addr & PAGEOFFSET));
+
+		if ((error = uread(p, buf + pos, len, addr)) != 0) {
+			if (error == ENXIO) {
+				/*
+				 * ENXIO from uread() indicates that the page
+				 * does not exist.  This will simply be a
+				 * partial read.
+				 */
+				error = 0;
+			}
+			break;
+		}
+
+		rem -= len;
+		pos += len;
+	}
+
+	if (rdsz != NULL)
+		*rdsz = pos;
+
+	return (error);
+}
+
+/*
+ * Attempt to read the argument vector (argv) from this process.  The caller
+ * must hold the p_lock mutex, and have marked the process P_PR_LOCK (e.g. via
+ * prlock or lx_prlock).
+ *
+ * The caller must provide a buffer (buf, buflen).  We will concatenate each
+ * argument string (including the NUL terminator) into this buffer.  The number
+ * of characters written to this buffer (including the final NUL terminator)
+ * will be stored in 'slen'.
+ */
+int
+prreadargv(proc_t *p, char *buf, size_t bufsz, size_t *slen)
+{
+	int error;
+	user_t *up;
+	struct as *as;
+	size_t pos = 0;
+	caddr_t *argv = NULL;
+	size_t argvsz = 0;
+	int i;
+
+	VERIFY(MUTEX_HELD(&p->p_lock));
+	VERIFY(p->p_proc_flag & P_PR_LOCK);
+
+	up = PTOU(p);
+	as = p->p_as;
+
+	if ((p->p_flag & SSYS) || as == &kas || up->u_argv == NULL) {
+		/*
+		 * Return the regular psargs string to the caller.
+		 */
+		bcopy(up->u_psargs, buf, MIN(bufsz, sizeof (up->u_psargs)));
+		buf[bufsz - 1] = '\0';
+		*slen = strlen(buf) + 1;
+
+		return (0);
+	}
+
+	/*
+	 * Allocate space to store argv array.
+	 */
+	argvsz = up->u_argc * (p->p_model == DATAMODEL_ILP32 ?
+	    sizeof (caddr32_t) : sizeof (caddr_t));
+	argv = kmem_alloc(argvsz, KM_SLEEP);
+
+	/*
+	 * Extract the argv array from the target process.  Drop p_lock
+	 * while we do I/O to avoid deadlock with the clock thread.
+	 */
+	mutex_exit(&p->p_lock);
+	if ((error = prreadbuf(p, up->u_argv, (uint8_t *)argv, argvsz,
+	    NULL)) != 0) {
+		kmem_free(argv, argvsz);
+		mutex_enter(&p->p_lock);
+		VERIFY(p->p_proc_flag & P_PR_LOCK);
+		return (-1);
+	}
+
+	/*
+	 * Read each argument string from the pointers in the argv array.
+	 */
+	pos = 0;
+	for (i = 0; i < up->u_argc; i++) {
+		size_t rdsz, trysz;
+		uintptr_t arg;
+		off_t j;
+		boolean_t found_nul;
+		boolean_t do_retry = B_TRUE;
+
+#ifdef	_SYSCALL32_IMPL
+		if (p->p_model == DATAMODEL_ILP32) {
+			arg = (uintptr_t)((caddr32_t *)argv)[i];
+		} else {
+			arg = (uintptr_t)argv[i];
+		}
+#else
+		arg = (uintptr_t)argv[i];
+#endif
+
+		/*
+		 * Stop trying to read arguments if we reach a NULL
+		 * pointer in the vector.
+		 */
+		if (arg == NULL)
+			break;
+
+		/*
+		 * Stop reading if we have read the maximum length
+		 * we can return to the user.
+		 */
+		if (pos >= bufsz)
+			break;
+
+		/*
+		 * Initially we try a short read, on the assumption that
+		 * most individual argument strings are less than 80
+		 * characters long.
+		 */
+		if ((trysz = MIN(80, bufsz - pos - 1)) < 80) {
+			/*
+			 * We don't have room in the target buffer for even
+			 * an entire short read, so there is no need to retry
+			 * with a longer read.
+			 */
+			do_retry = B_FALSE;
+		}
+
+retry:
+		/*
+		 * Read string data for this argument.  Leave room
+		 * in the buffer for a final NUL terminator.
+		 */
+		if ((error = prreadbuf(p, arg, (uint8_t *)&buf[pos], trysz,
+		    &rdsz)) != 0) {
+			/*
+			 * There was a problem reading this string
+			 * from the process.  Give up.
+			 */
+			break;
+		}
+
+		/*
+		 * Find the NUL terminator.
+		 */
+		found_nul = B_FALSE;
+		for (j = 0; j < rdsz; j++) {
+			if (buf[pos + j] == '\0') {
+				found_nul = B_TRUE;
+				break;
+			}
+		}
+
+		if (!found_nul && do_retry) {
+			/*
+			 * We did not find a NUL terminator, but this
+			 * was a first pass short read.  Try once more
+			 * with feeling.
+			 */
+			trysz = bufsz - pos - 1;
+			do_retry = B_FALSE;
+			goto retry;
+		}
+
+		/*
+		 * Commit the string we read to the buffer.
+		 */
+		pos += j + 1;
+		if (!found_nul && pos < bufsz) {
+			/*
+			 * A NUL terminator was not found; add one.
+			 */
+			buf[pos++] = '\0';
+		}
+	}
+
+	/*
+	 * Ensure the entire string is NUL-terminated.
+	 */
+	buf[bufsz - 1] = '\0';
+
+	mutex_enter(&p->p_lock);
+	VERIFY(p->p_proc_flag & P_PR_LOCK);
+	kmem_free(argv, argvsz);
+
+	/*
+	 * If the operation was a success, return the copied string length
+	 * to the caller.
+	 */
+	*slen = (error == 0) ? pos : 0;
+
+	return (error);
+}
+
+/*
+ * Similar to prreadargv except reads the env vector. This is slightly more
+ * complex because there is no count for the env vector that corresponds to
+ * u_argc.
+ */
+int
+prreadenvv(proc_t *p, char *buf, size_t bufsz, size_t *slen)
+{
+	int error;
+	user_t *up;
+	struct as *as;
+	size_t pos = 0;
+	caddr_t *envp = NULL;
+	uintptr_t tmpp = NULL;
+	size_t envpsz = 0, rdsz = 0;
+	int i;
+	int cnt, bound;
+
+	VERIFY(MUTEX_HELD(&p->p_lock));
+	VERIFY(p->p_proc_flag & P_PR_LOCK);
+
+	up = PTOU(p);
+	as = p->p_as;
+
+	if ((p->p_flag & SSYS) || as == &kas || up->u_envp == NULL) {
+		/*
+		 * Return empty string.
+		 */
+		buf[0] = '\0';
+		*slen = 1;
+
+		return (0);
+	}
+
+	/*
+	 * Drop p_lock while we do I/O to avoid deadlock with the clock thread.
+	 */
+	mutex_exit(&p->p_lock);
+
+	/*
+	 * We first have to count how many env entries we have. This is
+	 * somewhat painful. We extract the env entries from the target process
+	 * one entry at a time. Stop trying to read env entries if we reach a
+	 * NULL pointer in the vector or hit our upper bound (which we take
+	 * as the bufsz/4) to ensure we don't run off.
+	 */
+	rdsz = (p->p_model == DATAMODEL_ILP32 ?
+	    sizeof (caddr32_t) : sizeof (caddr_t));
+	bound = (int)(bufsz / 4);
+	for (cnt = 0, tmpp = up->u_envp; cnt < bound; cnt++, tmpp += rdsz) {
+		caddr_t tmp = NULL;
+
+		if ((error = prreadbuf(p, tmpp, (uint8_t *)&tmp, rdsz,
+		    NULL)) != 0) {
+			mutex_enter(&p->p_lock);
+			VERIFY(p->p_proc_flag & P_PR_LOCK);
+			return (-1);
+		}
+
+		if (tmp == NULL)
+			break;
+	}
+	if (cnt == 0) {
+		/* Return empty string. */
+		buf[0] = '\0';
+		*slen = 1;
+		mutex_enter(&p->p_lock);
+		VERIFY(p->p_proc_flag & P_PR_LOCK);
+		return (0);
+	}
+
+	/*
+	 * Allocate space to store env array.
+	 */
+	envpsz = cnt * (p->p_model == DATAMODEL_ILP32 ?
+	    sizeof (caddr32_t) : sizeof (caddr_t));
+	envp = kmem_alloc(envpsz, KM_SLEEP);
+
+	/*
+	 * Extract the env array from the target process.
+	 */
+	if ((error = prreadbuf(p, up->u_envp, (uint8_t *)envp, envpsz,
+	    NULL)) != 0) {
+		kmem_free(envp, envpsz);
+		mutex_enter(&p->p_lock);
+		VERIFY(p->p_proc_flag & P_PR_LOCK);
+		return (-1);
+	}
+
+	/*
+	 * Read each env string from the pointers in the env array.
+	 */
+	pos = 0;
+	for (i = 0; i < cnt; i++) {
+		size_t rdsz, trysz;
+		uintptr_t ev;
+		off_t j;
+		boolean_t found_nul;
+		boolean_t do_retry = B_TRUE;
+
+#ifdef	_SYSCALL32_IMPL
+		if (p->p_model == DATAMODEL_ILP32) {
+			ev = (uintptr_t)((caddr32_t *)envp)[i];
+		} else {
+			ev = (uintptr_t)envp[i];
+		}
+#else
+		ev = (uintptr_t)envp[i];
+#endif
+
+		/*
+		 * Stop trying to read env entries if we reach a NULL
+		 * pointer in the vector.
+		 */
+		if (ev == NULL)
+			break;
+
+		/*
+		 * Stop reading if we have read the maximum length
+		 * we can return to the user.
+		 */
+		if (pos >= bufsz)
+			break;
+
+		/*
+		 * Initially we try a short read, on the assumption that
+		 * most individual env strings are less than 80
+		 * characters long.
+		 */
+		if ((trysz = MIN(80, bufsz - pos - 1)) < 80) {
+			/*
+			 * We don't have room in the target buffer for even
+			 * an entire short read, so there is no need to retry
+			 * with a longer read.
+			 */
+			do_retry = B_FALSE;
+		}
+
+retry:
+		/*
+		 * Read string data for this env var.  Leave room
+		 * in the buffer for a final NUL terminator.
+		 */
+		if ((error = prreadbuf(p, ev, (uint8_t *)&buf[pos], trysz,
+		    &rdsz)) != 0) {
+			/*
+			 * There was a problem reading this string
+			 * from the process.  Give up.
+			 */
+			break;
+		}
+
+		/*
+		 * Find the NUL terminator.
+		 */
+		found_nul = B_FALSE;
+		for (j = 0; j < rdsz; j++) {
+			if (buf[pos + j] == '\0') {
+				found_nul = B_TRUE;
+				break;
+			}
+		}
+
+		if (!found_nul && do_retry) {
+			/*
+			 * We did not find a NUL terminator, but this
+			 * was a first pass short read.  Try once more
+			 * with feeling.
+			 */
+			trysz = bufsz - pos - 1;
+			do_retry = B_FALSE;
+			goto retry;
+		}
+
+		/*
+		 * Commit the string we read to the buffer.
+		 */
+		pos += j + 1;
+		if (!found_nul && pos < bufsz) {
+			/*
+			 * A NUL terminator was not found; add one.
+			 */
+			buf[pos++] = '\0';
+		}
+	}
+
+	/*
+	 * Ensure the entire string is NUL-terminated.
+	 */
+	buf[bufsz - 1] = '\0';
+
+	mutex_enter(&p->p_lock);
+	VERIFY(p->p_proc_flag & P_PR_LOCK);
+	kmem_free(envp, envpsz);
+
+	/*
+	 * If the operation was a success, return the copied string length
+	 * to the caller.
+	 */
+	*slen = (error == 0) ? pos : 0;
+
+	return (error);
+}
diff --git a/usr/src/uts/common/fs/proc/prcontrol.c b/usr/src/uts/common/fs/proc/prcontrol.c
index a73a64a4a4..7e99d23b97 100644
--- a/usr/src/uts/common/fs/proc/prcontrol.c
+++ b/usr/src/uts/common/fs/proc/prcontrol.c
@@ -25,7 +25,7 @@
  */
 
 /*
- * Copyright (c) 2013, Joyent, Inc.  All rights reserved.
+ * Copyright 2015, Joyent, Inc.
  */
 
 #include <sys/types.h>
@@ -1481,7 +1481,7 @@ pr_setsig(prnode_t *pnp, siginfo_t *sip)
 		} else if (t->t_state == TS_STOPPED && sig == SIGKILL) {
 			/* If SIGKILL, set stopped lwp running */
 			p->p_stopsig = 0;
-			t->t_schedflag |= TS_XSTART | TS_PSTART;
+			t->t_schedflag |= TS_XSTART | TS_PSTART | TS_BSTART;
 			t->t_dtrace_stop = 0;
 			setrun_locked(t);
 		}
@@ -2276,9 +2276,17 @@ pr_szoneid(proc_t *p, zoneid_t zoneid, cred_t *cr)
 		return (EPERM);
 	if (zoneid != GLOBAL_ZONEID && zoneid != p->p_zone->zone_id)
 		return (EINVAL);
-	if ((zptr = zone_find_by_id(zoneid)) == NULL)
-		return (EINVAL);
+	/*
+	 * We cannot hold p_lock when we call zone_find_by_id since that can
+	 * lead to a deadlock. zone_find_by_id() takes zonehash_lock.
+	 * zone_enter() can hold the zonehash_lock and needs p_lock when it
+	 * calls task_join.
+	 */
 	mutex_exit(&p->p_lock);
+	if ((zptr = zone_find_by_id(zoneid)) == NULL) {
+		mutex_enter(&p->p_lock);
+		return (EINVAL);
+	}
 	mutex_enter(&p->p_crlock);
 	oldcred = p->p_cred;
 	crhold(oldcred);
diff --git a/usr/src/uts/common/fs/proc/prdata.h b/usr/src/uts/common/fs/proc/prdata.h
index 8ea516bf82..72f26b3c05 100644
--- a/usr/src/uts/common/fs/proc/prdata.h
+++ b/usr/src/uts/common/fs/proc/prdata.h
@@ -27,7 +27,7 @@
 /*	  All Rights Reserved  	*/
 
 /*
- * Copyright (c) 2013, Joyent, Inc. All rights reserved.
+ * Copyright 2015, Joyent, Inc.
  */
 
 #ifndef _SYS_PROC_PRDATA_H
@@ -123,6 +123,7 @@ typedef enum prnodetype {
 #if defined(__i386) || defined(__amd64)
 	PR_LDT,			/* /proc/<pid>/ldt			*/
 #endif
+	PR_ARGV,		/* /proc/<pid>/argv			*/
 	PR_USAGE,		/* /proc/<pid>/usage			*/
 	PR_LUSAGE,		/* /proc/<pid>/lusage			*/
 	PR_PAGEDATA,		/* /proc/<pid>/pagedata			*/
@@ -347,6 +348,8 @@ extern	int	pr_unset(proc_t *, long);
 extern	void	pr_sethold(prnode_t *, sigset_t *);
 extern	void	pr_setfault(proc_t *, fltset_t *);
 extern	int	prusrio(proc_t *, enum uio_rw, struct uio *, int);
+extern	int	prreadargv(proc_t *, char *, size_t, size_t *);
+extern	int	prreadenvv(proc_t *, char *, size_t, size_t *);
 extern	int	prwritectl(vnode_t *, struct uio *, cred_t *);
 extern	int	prlock(prnode_t *, int);
 extern	void	prunmark(proc_t *);
diff --git a/usr/src/uts/common/fs/proc/prsubr.c b/usr/src/uts/common/fs/proc/prsubr.c
index 7801fd0ac8..284bf8cb88 100644
--- a/usr/src/uts/common/fs/proc/prsubr.c
+++ b/usr/src/uts/common/fs/proc/prsubr.c
@@ -21,7 +21,7 @@
 
 /*
  * Copyright (c) 1989, 2010, Oracle and/or its affiliates. All rights reserved.
- * Copyright (c) 2013, Joyent, Inc. All rights reserved.
+ * Copyright 2015, Joyent, Inc.
  */
 
 /*	Copyright (c) 1984, 1986, 1987, 1988, 1989 AT&T	*/
@@ -201,6 +201,7 @@ prchoose(proc_t *p)
 			case PR_SYSEXIT:
 			case PR_SIGNALLED:
 			case PR_FAULTED:
+			case PR_BRAND:
 				/*
 				 * Make an lwp calling exit() be the
 				 * last lwp seen in the process.
diff --git a/usr/src/uts/common/fs/proc/prvnops.c b/usr/src/uts/common/fs/proc/prvnops.c
index 411c9b8b0b..276f54ae3a 100644
--- a/usr/src/uts/common/fs/proc/prvnops.c
+++ b/usr/src/uts/common/fs/proc/prvnops.c
@@ -21,7 +21,7 @@
 
 /*
  * Copyright (c) 1989, 2010, Oracle and/or its affiliates. All rights reserved.
- * Copyright (c) 2013, Joyent, Inc. All rights reserved.
+ * Copyright 2015, Joyent, Inc.
  */
 
 /*	Copyright (c) 1984,	 1986, 1987, 1988, 1989 AT&T	*/
@@ -96,6 +96,11 @@ struct prdirect {
 #define	PRSDSIZE	(sizeof (struct prdirect))
 
 /*
+ * Maximum length of the /proc/$$/argv file:
+ */
+int prmaxargvlen = 4096;
+
+/*
  * Directory characteristics.
  */
 typedef struct prdirent {
@@ -166,6 +171,8 @@ static prdirent_t piddir[] = {
 	{ PR_LDT,	27 * sizeof (prdirent_t), sizeof (prdirent_t),
 		"ldt" },
 #endif
+	{ PR_ARGV,	28 * sizeof (prdirent_t), sizeof (prdirent_t),
+		"argv" },
 };
 
 #define	NPIDDIRFILES	(sizeof (piddir) / sizeof (piddir[0]) - 2)
@@ -582,6 +589,7 @@ static int pr_read_inval(), pr_read_as(), pr_read_status(),
 #if defined(__x86)
 	pr_read_ldt(),
 #endif
+	pr_read_argv(),
 	pr_read_usage(), pr_read_lusage(), pr_read_pagedata(),
 	pr_read_watch(), pr_read_lwpstatus(), pr_read_lwpsinfo(),
 	pr_read_lwpusage(), pr_read_xregs(), pr_read_priv(),
@@ -610,6 +618,7 @@ static int (*pr_read_function[PR_NFILES])() = {
 #if defined(__x86)
 	pr_read_ldt,		/* /proc/<pid>/ldt			*/
 #endif
+	pr_read_argv,		/* /proc/<pid>/argv			*/
 	pr_read_usage,		/* /proc/<pid>/usage			*/
 	pr_read_lusage,		/* /proc/<pid>/lusage			*/
 	pr_read_pagedata,	/* /proc/<pid>/pagedata			*/
@@ -672,6 +681,41 @@ pr_uioread(void *base, long count, uio_t *uiop)
 }
 
 static int
+pr_read_argv(prnode_t *pnp, uio_t *uiop)
+{
+	char *args;
+	int error;
+	size_t asz = prmaxargvlen, sz;
+
+	/*
+	 * Allocate a scratch buffer for collection of the process arguments.
+	 */
+	args = kmem_alloc(asz, KM_SLEEP);
+
+	ASSERT(pnp->pr_type == PR_ARGV);
+
+	if ((error = prlock(pnp, ZNO)) != 0) {
+		kmem_free(args, asz);
+		return (error);
+	}
+
+	if ((error = prreadargv(pnp->pr_common->prc_proc, args, asz,
+	    &sz)) != 0) {
+		prunlock(pnp);
+		kmem_free(args, asz);
+		return (error);
+	}
+
+	prunlock(pnp);
+
+	error = pr_uioread(args, sz, uiop);
+
+	kmem_free(args, asz);
+
+	return (error);
+}
+
+static int
 pr_read_as(prnode_t *pnp, uio_t *uiop)
 {
 	int error;
@@ -1767,6 +1811,7 @@ static int (*pr_read_function_32[PR_NFILES])() = {
 #if defined(__x86)
 	pr_read_ldt,		/* /proc/<pid>/ldt			*/
 #endif
+	pr_read_argv,		/* /proc/<pid>/argv			*/
 	pr_read_usage_32,	/* /proc/<pid>/usage			*/
 	pr_read_lusage_32,	/* /proc/<pid>/lusage			*/
 	pr_read_pagedata_32,	/* /proc/<pid>/pagedata			*/
@@ -2686,6 +2731,103 @@ prread(vnode_t *vp, uio_t *uiop, int ioflag, cred_t *cr, caller_context_t *ct)
 #endif
 }
 
+/*
+ * We make pr_write_psinfo_fname() somewhat simpler by asserting at compile
+ * time that PRFNSZ has the same definition as MAXCOMLEN.
+ */
+#if PRFNSZ != MAXCOMLEN
+#error PRFNSZ/MAXCOMLEN mismatch
+#endif
+
+static int
+pr_write_psinfo_fname(prnode_t *pnp, uio_t *uiop)
+{
+	char fname[PRFNSZ];
+	int offset = offsetof(psinfo_t, pr_fname), error;
+
+#ifdef _SYSCALL32_IMPL
+	if (curproc->p_model != DATAMODEL_LP64)
+		offset = offsetof(psinfo32_t, pr_fname);
+#endif
+
+	/*
+	 * If this isn't a write to pr_fname (or if the size doesn't match
+	 * PRFNSZ) return.
+	 */
+	if (uiop->uio_offset != offset || uiop->uio_resid != PRFNSZ)
+		return (0);
+
+	if ((error = uiomove(fname, PRFNSZ, UIO_WRITE, uiop)) != 0)
+		return (error);
+
+	fname[PRFNSZ - 1] = '\0';
+
+	if ((error = prlock(pnp, ZNO)) != 0)
+		return (error);
+
+	bcopy(fname, pnp->pr_common->prc_proc->p_user.u_comm, PRFNSZ);
+
+	prunlock(pnp);
+
+	return (0);
+}
+
+/*
+ * We make pr_write_psinfo_psargs() somewhat simpler by asserting at compile
+ * time that PRARGSZ has the same definition as PSARGSZ.
+ */
+#if PRARGSZ != PSARGSZ
+#error PRARGSZ/PSARGSZ mismatch
+#endif
+
+static int
+pr_write_psinfo_psargs(prnode_t *pnp, uio_t *uiop)
+{
+	char psargs[PRARGSZ];
+	int offset = offsetof(psinfo_t, pr_psargs), error;
+
+#ifdef _SYSCALL32_IMPL
+	if (curproc->p_model != DATAMODEL_LP64)
+		offset = offsetof(psinfo32_t, pr_psargs);
+#endif
+
+	/*
+	 * If this isn't a write to pr_psargs (or if the size doesn't match
+	 * PRARGSZ) return.
+	 */
+	if (uiop->uio_offset != offset || uiop->uio_resid != PRARGSZ)
+		return (0);
+
+	if ((error = uiomove(psargs, PRARGSZ, UIO_WRITE, uiop)) != 0)
+		return (error);
+
+	psargs[PRARGSZ - 1] = '\0';
+
+	if ((error = prlock(pnp, ZNO)) != 0)
+		return (error);
+
+	bcopy(psargs, pnp->pr_common->prc_proc->p_user.u_psargs, PRARGSZ);
+
+	prunlock(pnp);
+
+	return (0);
+}
+
+int
+pr_write_psinfo(prnode_t *pnp, uio_t *uiop)
+{
+	int error;
+
+	if ((error = pr_write_psinfo_fname(pnp, uiop)) != 0)
+		return (error);
+
+	if ((error = pr_write_psinfo_psargs(pnp, uiop)) != 0)
+		return (error);
+
+	return (0);
+}
+
+
 /* ARGSUSED */
 static int
 prwrite(vnode_t *vp, uio_t *uiop, int ioflag, cred_t *cr, caller_context_t *ct)
@@ -2764,6 +2906,9 @@ prwrite(vnode_t *vp, uio_t *uiop, int ioflag, cred_t *cr, caller_context_t *ct)
 			uiop->uio_resid = resid;
 		return (error);
 
+	case PR_PSINFO:
+		return (pr_write_psinfo(pnp, uiop));
+
 	default:
 		return ((vp->v_type == VDIR)? EISDIR : EBADF);
 	}
@@ -3047,6 +3192,13 @@ prgetattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr,
 	case PR_AUXV:
 		vap->va_size = __KERN_NAUXV_IMPL * PR_OBJSIZE(auxv32_t, auxv_t);
 		break;
+	case PR_ARGV:
+		if ((p->p_flag & SSYS) || p->p_as == &kas) {
+			vap->va_size = PSARGSZ;
+		} else {
+			vap->va_size = prmaxargvlen;
+		}
+		break;
 #if defined(__x86)
 	case PR_LDT:
 		mutex_exit(&p->p_lock);
@@ -3222,6 +3374,7 @@ praccess(vnode_t *vp, int mode, int flags, cred_t *cr, caller_context_t *ct)
 	case PR_USAGE:
 	case PR_LUSAGE:
 	case PR_LWPUSAGE:
+	case PR_ARGV:
 		p = pr_p_lock(pnp);
 		mutex_exit(&pr_pidlock);
 		if (p == NULL)
@@ -3307,6 +3460,7 @@ static vnode_t *(*pr_lookup_function[PR_NFILES])() = {
 #if defined(__x86)
 	pr_lookup_notdir,	/* /proc/<pid>/ldt			*/
 #endif
+	pr_lookup_notdir,	/* /proc/<pid>/argv			*/
 	pr_lookup_notdir,	/* /proc/<pid>/usage			*/
 	pr_lookup_notdir,	/* /proc/<pid>/lusage			*/
 	pr_lookup_notdir,	/* /proc/<pid>/pagedata			*/
@@ -4546,11 +4700,15 @@ prgetnode(vnode_t *dp, prnodetype_t type)
 		break;
 
 	case PR_PSINFO:
+		pnp->pr_mode = 0644;	/* readable by all + owner can write */
+		break;
+
 	case PR_LPSINFO:
 	case PR_LWPSINFO:
 	case PR_USAGE:
 	case PR_LUSAGE:
 	case PR_LWPUSAGE:
+	case PR_ARGV:
 		pnp->pr_mode = 0444;	/* read-only by all */
 		break;
 
@@ -4656,6 +4814,7 @@ static int (*pr_readdir_function[PR_NFILES])() = {
 #if defined(__x86)
 	pr_readdir_notdir,	/* /proc/<pid>/ldt			*/
 #endif
+	pr_readdir_notdir,	/* /proc/<pid>/argv			*/
 	pr_readdir_notdir,	/* /proc/<pid>/usage			*/
 	pr_readdir_notdir,	/* /proc/<pid>/lusage			*/
 	pr_readdir_notdir,	/* /proc/<pid>/pagedata			*/
@@ -4805,6 +4964,7 @@ pr_readdir_piddir(prnode_t *pnp, uio_t *uiop, int *eofp)
 			case PR_PROCDIR:
 			case PR_PSINFO:
 			case PR_USAGE:
+			case PR_ARGV:
 				break;
 			default:
 				continue;
@@ -6010,7 +6170,7 @@ prpoll(vnode_t *vp, short events, int anyyet, short *reventsp,
 	}
 
 	*reventsp = revents;
-	if (!anyyet && revents == 0) {
+	if ((!anyyet && revents == 0) || (events & POLLET)) {
 		/*
 		 * Arrange to wake up the polling lwp when
 		 * the target process/lwp stops or terminates
diff --git a/usr/src/uts/common/fs/sockfs/sockcommon.c b/usr/src/uts/common/fs/sockfs/sockcommon.c
index 703e26ea61..682f1d867b 100644
--- a/usr/src/uts/common/fs/sockfs/sockcommon.c
+++ b/usr/src/uts/common/fs/sockfs/sockcommon.c
@@ -21,6 +21,7 @@
 
 /*
  * Copyright (c) 2008, 2010, Oracle and/or its affiliates. All rights reserved.
+ * Copyright 2015, Joyent, Inc.
  */
 
 #include <sys/types.h>
@@ -501,6 +502,9 @@ sonode_constructor(void *buf, void *cdrarg, int kmflags)
 	cv_init(&so->so_copy_cv, NULL, CV_DEFAULT, NULL);
 	cv_init(&so->so_closing_cv, NULL, CV_DEFAULT, NULL);
 
+	so->so_krecv_cb = NULL;
+	so->so_krecv_arg = NULL;
+
 	return (0);
 }
 
@@ -654,6 +658,10 @@ sonode_fini(struct sonode *so)
 	if (so->so_filter_top != NULL)
 		sof_sonode_cleanup(so);
 
+	/* Clean up any remnants of krecv callbacks */
+	so->so_krecv_cb = NULL;
+	so->so_krecv_arg = NULL;
+
 	ASSERT(list_is_empty(&so->so_acceptq_list));
 	ASSERT(list_is_empty(&so->so_acceptq_defer));
 	ASSERT(!list_link_active(&so->so_acceptq_node));
diff --git a/usr/src/uts/common/fs/sockfs/sockcommon_sops.c b/usr/src/uts/common/fs/sockfs/sockcommon_sops.c
index 0be628f329..40c368736d 100644
--- a/usr/src/uts/common/fs/sockfs/sockcommon_sops.c
+++ b/usr/src/uts/common/fs/sockfs/sockcommon_sops.c
@@ -24,7 +24,7 @@
  */
 
 /*
- * Copyright (c) 2014, Joyent, Inc.  All rights reserved.
+ * Copyright (c) 2015, Joyent, Inc.  All rights reserved.
  */
 
 #include <sys/types.h>
@@ -128,7 +128,7 @@ so_bind(struct sonode *so, struct sockaddr *name, socklen_t namelen,
 {
 	int error;
 
-	SO_BLOCK_FALLBACK(so, SOP_BIND(so, name, namelen, flags, cr));
+	SO_BLOCK_FALLBACK_SAFE(so, SOP_BIND(so, name, namelen, flags, cr));
 
 	ASSERT(flags == _SOBIND_XPG4_2 || flags == _SOBIND_SOCKBSD);
 
@@ -586,11 +586,6 @@ so_sendmblk(struct sonode *so, struct nmsghdr *msg, int fflag,
 
 	SO_BLOCK_FALLBACK(so, SOP_SENDMBLK(so, msg, fflag, cr, mpp));
 
-	if ((so->so_mode & SM_SENDFILESUPP) == 0) {
-		SO_UNBLOCK_FALLBACK(so);
-		return (EOPNOTSUPP);
-	}
-
 	error = so_sendmblk_impl(so, msg, fflag, cr, mpp, so->so_filter_top,
 	    B_FALSE);
 
@@ -653,7 +648,7 @@ so_getsockname(struct sonode *so, struct sockaddr *addr,
 {
 	int error;
 
-	SO_BLOCK_FALLBACK(so, SOP_GETSOCKNAME(so, addr, addrlen, cr));
+	SO_BLOCK_FALLBACK_SAFE(so, SOP_GETSOCKNAME(so, addr, addrlen, cr));
 
 	if (so->so_filter_active == 0 ||
 	    (error = sof_filter_getsockname(so, addr, addrlen, cr)) < 0)
@@ -702,7 +697,7 @@ so_getsockopt(struct sonode *so, int level, int option_name,
 	if (level == SOL_FILTER)
 		return (sof_getsockopt(so, option_name, optval, optlenp, cr));
 
-	SO_BLOCK_FALLBACK(so,
+	SO_BLOCK_FALLBACK_SAFE(so,
 	    SOP_GETSOCKOPT(so, level, option_name, optval, optlenp, flags, cr));
 
 	if ((so->so_filter_active == 0 ||
@@ -791,7 +786,7 @@ so_setsockopt(struct sonode *so, int level, int option_name,
 	if (level == SOL_FILTER)
 		return (sof_setsockopt(so, option_name, optval, optlen, cr));
 
-	SO_BLOCK_FALLBACK(so,
+	SO_BLOCK_FALLBACK_SAFE(so,
 	    SOP_SETSOCKOPT(so, level, option_name, optval, optlen, cr));
 
 	/* X/Open requires this check */
@@ -953,6 +948,13 @@ so_poll(struct sonode *so, short events, int anyyet, short *reventsp,
 	if (!list_is_empty(&so->so_acceptq_list))
 		*reventsp |= (POLLIN|POLLRDNORM) & events;
 
+	/*
+	 * If we're looking for POLLRDHUP, indicate it if we have sent the
+	 * last rx signal for the socket.
+	 */
+	if ((events & POLLRDHUP) && (state & SS_SENTLASTREADSIG))
+		*reventsp |= POLLRDHUP;
+
 	/* Data */
 	/* so_downcalls is null for sctp */
 	if (so->so_downcalls != NULL && so->so_downcalls->sd_poll != NULL) {
@@ -988,14 +990,20 @@ so_poll(struct sonode *so, short events, int anyyet, short *reventsp,
 			*reventsp |= POLLHUP;
 	}
 
-	if (!*reventsp && !anyyet) {
+	if ((!*reventsp && !anyyet) || (events & POLLET)) {
 		/* Check for read events again, but this time under lock */
 		if (events & (POLLIN|POLLRDNORM)) {
 			mutex_enter(&so->so_lock);
 			if (SO_HAVE_DATA(so) ||
 			    !list_is_empty(&so->so_acceptq_list)) {
+				if (events & POLLET) {
+					so->so_pollev |= SO_POLLEV_IN;
+					*phpp = &so->so_poll_list;
+				}
+
 				mutex_exit(&so->so_lock);
 				*reventsp |= (POLLIN|POLLRDNORM) & events;
+
 				return (0);
 			} else {
 				so->so_pollev |= SO_POLLEV_IN;
@@ -1316,6 +1324,26 @@ so_queue_msg_impl(struct sonode *so, mblk_t *mp,
 		}
 	}
 
+	mutex_enter(&so->so_lock);
+	if (so->so_krecv_cb != NULL) {
+		boolean_t cont;
+		so_krecv_f func = so->so_krecv_cb;
+		void *arg = so->so_krecv_arg;
+
+		mutex_exit(&so->so_lock);
+		cont = func(so, mp, msg_size, flags & MSG_OOB, arg);
+		mutex_enter(&so->so_lock);
+		if (cont == B_TRUE) {
+			space_left = so->so_rcvbuf;
+		} else {
+			so->so_rcv_queued = so->so_rcvlowat;
+			*errorp = ENOSPC;
+			space_left = -1;
+		}
+		goto done_unlock;
+	}
+	mutex_exit(&so->so_lock);
+
 	if (flags & MSG_OOB) {
 		so_queue_oob(so, mp, msg_size);
 		mutex_enter(&so->so_lock);
@@ -1594,6 +1622,13 @@ so_recvmsg(struct sonode *so, struct nmsghdr *msg, struct uio *uiop,
 		return (ENOTCONN);
 	}
 
+	mutex_enter(&so->so_lock);
+	if (so->so_krecv_cb != NULL) {
+		mutex_exit(&so->so_lock);
+		return (EOPNOTSUPP);
+	}
+	mutex_exit(&so->so_lock);
+
 	if (msg->msg_flags & MSG_PEEK)
 		msg->msg_flags &= ~MSG_WAITALL;
 
diff --git a/usr/src/uts/common/fs/sockfs/sockcommon_subr.c b/usr/src/uts/common/fs/sockfs/sockcommon_subr.c
index 957c8f93b4..9604ea5dba 100644
--- a/usr/src/uts/common/fs/sockfs/sockcommon_subr.c
+++ b/usr/src/uts/common/fs/sockfs/sockcommon_subr.c
@@ -24,6 +24,7 @@
  */
 /*
  * Copyright 2014, OmniTI Computer Consulting, Inc. All rights reserved.
+ * Copyright 2015 Joyent, Inc.
  */
 
 #include <sys/types.h>
@@ -2276,9 +2277,9 @@ so_tpi_fallback(struct sonode *so, struct cred *cr)
 	fbfunc = sp->sp_smod_info->smod_proto_fallback_func;
 
 	/*
-	 * Cannot fallback if the socket has active filters
+	 * Cannot fallback if the socket has active filters or a krecv callback.
 	 */
-	if (so->so_filter_active > 0)
+	if (so->so_filter_active > 0 || so->so_krecv_cb != NULL)
 		return (EINVAL);
 
 	switch (so->so_family) {
@@ -2456,3 +2457,50 @@ out:
 
 	return (error);
 }
+
+int
+so_krecv_set(sonode_t *so, so_krecv_f cb, void *arg)
+{
+	int ret;
+
+	if (cb == NULL && arg != NULL)
+		return (EINVAL);
+
+	SO_BLOCK_FALLBACK(so, so_krecv_set(so, cb, arg));
+
+	mutex_enter(&so->so_lock);
+	if (so->so_state & SS_FALLBACK_COMP) {
+		mutex_exit(&so->so_lock);
+		SO_UNBLOCK_FALLBACK(so);
+		return (ENOTSUP);
+	}
+
+	ret = so_lock_read(so, 0);
+	VERIFY(ret == 0);
+	/*
+	 * Other consumers may actually care about getting extant data delivered
+	 * to them, when they come along, they should figure out the best API
+	 * for that.
+	 */
+	so_rcv_flush(so);
+
+	so->so_krecv_cb = cb;
+	so->so_krecv_arg = arg;
+
+	so_unlock_read(so);
+	mutex_exit(&so->so_lock);
+	SO_UNBLOCK_FALLBACK(so);
+
+	return (0);
+}
+
+void
+so_krecv_unblock(sonode_t *so)
+{
+	mutex_enter(&so->so_lock);
+	VERIFY(so->so_krecv_cb != NULL);
+
+	so->so_rcv_queued = 0;
+	(void) so_check_flow_control(so);
+	mutex_exit(&so->so_lock);
+}
diff --git a/usr/src/uts/common/fs/sockfs/sockfilter.c b/usr/src/uts/common/fs/sockfs/sockfilter.c
index 971523945e..7dca6ae6fc 100644
--- a/usr/src/uts/common/fs/sockfs/sockfilter.c
+++ b/usr/src/uts/common/fs/sockfs/sockfilter.c
@@ -20,6 +20,7 @@
  */
 /*
  * Copyright (c) 2010, Oracle and/or its affiliates. All rights reserved.
+ * Copyright 2015 Joyent, Inc.
  */
 
 #include <sys/systm.h>
@@ -246,6 +247,18 @@ sof_setsockopt_impl(struct sonode *so, int option_name,
 
 		/* Module loaded OK, so there must be an ops vector */
 		ASSERT(ent->sofe_mod != NULL);
+
+		/*
+		 * Check again to confirm ATTACH is ok. See if the the module
+		 * is not SOF_ATT_SAFE after an unsafe operation has taken
+		 * place.
+		 */
+		if ((ent->sofe_mod->sofm_flags & SOF_ATT_SAFE) == 0 &&
+		    so->so_state & SS_FILOP_UNSF) {
+			sof_instance_destroy(inst);
+			return (EINVAL);
+		}
+
 		inst->sofi_ops = &ent->sofe_mod->sofm_ops;
 
 		SOF_STAT_ADD(inst, tot_active_attach, 1);
@@ -1444,7 +1457,13 @@ sof_filter_ioctl(struct sonode *so, int cmd, intptr_t arg, int mode,
  * sof_register(version, name, ops, flags)
  *
  * Register a socket filter identified by name `name' and which should use
- * the ops vector `ops' for event notification. `flags' should be set to 0.
+ * the ops vector `ops' for event notification. `flags' should be set to 0
+ * by default for "unsafe" modules or SOF_ATT_SAFE for "safe" modules. An
+ * unsafe filter is one that cannot be attached after any socket operation has
+ * occured. This is the legacy default. A "safe" filter can be attached even
+ * after some basic initial socket operations have taken place. This set is
+ * currently bind, getsockname, getsockopt and setsockopt. The order in which
+ * a "safe" filter can be attached is more relaxed, and thus more flexible.
  * On success 0 is returned, otherwise an errno is returned.
  */
 int
@@ -1452,14 +1471,13 @@ sof_register(int version, const char *name, const sof_ops_t *ops, int flags)
 {
 	sof_module_t *mod;
 
-	_NOTE(ARGUNUSED(flags));
-
 	if (version != SOF_VERSION)
 		return (EINVAL);
 
 	mod = kmem_zalloc(sizeof (sof_module_t), KM_SLEEP);
 	mod->sofm_name = kmem_alloc(strlen(name) + 1, KM_SLEEP);
 	(void) strcpy(mod->sofm_name, name);
+	mod->sofm_flags = flags;
 	mod->sofm_ops = *ops;
 
 	mutex_enter(&sof_module_lock);
diff --git a/usr/src/uts/common/fs/sockfs/sockfilter_impl.h b/usr/src/uts/common/fs/sockfs/sockfilter_impl.h
index 7f7aece1f1..cf2ad8b20d 100644
--- a/usr/src/uts/common/fs/sockfs/sockfilter_impl.h
+++ b/usr/src/uts/common/fs/sockfs/sockfilter_impl.h
@@ -20,6 +20,7 @@
  */
 /*
  * Copyright (c) 2010, Oracle and/or its affiliates. All rights reserved.
+ * Copyright 2015 Joyent, Inc.
  */
 
 #ifndef	_SOCKFS_SOCKFILTER_H
@@ -51,6 +52,7 @@ typedef struct sof_kstat	sof_kstat_t;
 
 struct sof_module {
 	char		*sofm_name;
+	int		sofm_flags;
 	sof_ops_t	sofm_ops;
 	uint_t		sofm_refcnt;
 	list_node_t	sofm_node;
diff --git a/usr/src/uts/common/fs/sockfs/socknotify.c b/usr/src/uts/common/fs/sockfs/socknotify.c
index 3d5ba2a7e8..3f858afecc 100644
--- a/usr/src/uts/common/fs/sockfs/socknotify.c
+++ b/usr/src/uts/common/fs/sockfs/socknotify.c
@@ -377,7 +377,7 @@ i_so_notify_last_rx(struct sonode *so, int *pollev, int *sigev)
 		so->so_state |= SS_SENTLASTREADSIG;
 		so->so_pollev &= ~SO_POLLEV_IN;
 
-		*pollev |= POLLIN|POLLRDNORM;
+		*pollev |= POLLIN|POLLRDNORM|POLLRDHUP;
 		*sigev |= SOCKETSIG_READ;
 
 		return (1);
diff --git a/usr/src/uts/common/fs/sockfs/socksubr.c b/usr/src/uts/common/fs/sockfs/socksubr.c
index 06d76044e5..095d84fc42 100644
--- a/usr/src/uts/common/fs/sockfs/socksubr.c
+++ b/usr/src/uts/common/fs/sockfs/socksubr.c
@@ -21,6 +21,7 @@
 
 /*
  * Copyright (c) 1995, 2010, Oracle and/or its affiliates. All rights reserved.
+ * Copyright 2015, Joyent, Inc. All rights reserved.
  */
 
 #include <sys/types.h>
@@ -1879,7 +1880,7 @@ ssize_t
 soreadfile(file_t *fp, uchar_t *buf, u_offset_t fileoff, int *err, size_t size)
 {
 	struct uio auio;
-	struct iovec aiov[MSG_MAXIOVLEN];
+	struct iovec aiov[1];
 	register vnode_t *vp;
 	int ioflag, rwflag;
 	ssize_t cnt;
diff --git a/usr/src/uts/common/fs/sockfs/socksyscalls.c b/usr/src/uts/common/fs/sockfs/socksyscalls.c
index a86cda937c..067eaedd7c 100644
--- a/usr/src/uts/common/fs/sockfs/socksyscalls.c
+++ b/usr/src/uts/common/fs/sockfs/socksyscalls.c
@@ -21,10 +21,10 @@
 
 /*
  * Copyright (c) 1995, 2010, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2013, OmniTI Computer Consulting, Inc. All rights reserved.
+ * Copyright 2015, Joyent, Inc.  All rights reserved.
  */
 
-/* Copyright (c) 2013, OmniTI Computer Consulting, Inc. All rights reserved. */
-
 #include <sys/types.h>
 #include <sys/t_lock.h>
 #include <sys/param.h>
@@ -51,6 +51,7 @@
 #include <sys/cmn_err.h>
 #include <sys/vmsystm.h>
 #include <sys/policy.h>
+#include <sys/limits.h>
 
 #include <sys/socket.h>
 #include <sys/socketvar.h>
@@ -83,12 +84,6 @@ extern void	nl7c_init(void);
 extern int	sockfs_defer_nl7c_init;
 
 /*
- * Note: DEF_IOV_MAX is defined and used as it is in "fs/vncalls.c"
- *	 as there isn't a formal definition of IOV_MAX ???
- */
-#define	MSG_MAXIOVLEN	16
-
-/*
  * Kernel component of socket creation.
  *
  * The socket library determines which version number to use.
@@ -1023,9 +1018,10 @@ recvmsg(int sock, struct nmsghdr *msg, int flags)
 	STRUCT_HANDLE(nmsghdr, umsgptr);
 	struct nmsghdr lmsg;
 	struct uio auio;
-	struct iovec aiov[MSG_MAXIOVLEN];
+	struct iovec buf[IOV_MAX_STACK], *aiov = buf;
+	ssize_t iovsize = 0;
 	int iovcnt;
-	ssize_t len;
+	ssize_t len, rval;
 	int i;
 	int *flagsp;
 	model_t	model;
@@ -1068,22 +1064,37 @@ recvmsg(int sock, struct nmsghdr *msg, int flags)
 
 	iovcnt = lmsg.msg_iovlen;
 
-	if (iovcnt <= 0 || iovcnt > MSG_MAXIOVLEN) {
+	if (iovcnt <= 0 || iovcnt > IOV_MAX) {
 		return (set_errno(EMSGSIZE));
 	}
 
+	if (iovcnt > IOV_MAX_STACK) {
+		iovsize = iovcnt * sizeof (struct iovec);
+		aiov = kmem_alloc(iovsize, KM_SLEEP);
+	}
+
 #ifdef _SYSCALL32_IMPL
 	/*
 	 * 32-bit callers need to have their iovec expanded, while ensuring
 	 * that they can't move more than 2Gbytes of data in a single call.
 	 */
 	if (model == DATAMODEL_ILP32) {
-		struct iovec32 aiov32[MSG_MAXIOVLEN];
+		struct iovec32 buf32[IOV_MAX_STACK], *aiov32 = buf32;
+		ssize_t iov32size;
 		ssize32_t count32;
 
-		if (copyin((struct iovec32 *)lmsg.msg_iov, aiov32,
-		    iovcnt * sizeof (struct iovec32)))
+		iov32size = iovcnt * sizeof (struct iovec32);
+		if (iovsize != 0)
+			aiov32 = kmem_alloc(iov32size, KM_SLEEP);
+
+		if (copyin((struct iovec32 *)lmsg.msg_iov, aiov32, iov32size)) {
+			if (iovsize != 0) {
+				kmem_free(aiov32, iov32size);
+				kmem_free(aiov, iovsize);
+			}
+
 			return (set_errno(EFAULT));
+		}
 
 		count32 = 0;
 		for (i = 0; i < iovcnt; i++) {
@@ -1091,15 +1102,28 @@ recvmsg(int sock, struct nmsghdr *msg, int flags)
 
 			iovlen32 = aiov32[i].iov_len;
 			count32 += iovlen32;
-			if (iovlen32 < 0 || count32 < 0)
+			if (iovlen32 < 0 || count32 < 0) {
+				if (iovsize != 0) {
+					kmem_free(aiov32, iov32size);
+					kmem_free(aiov, iovsize);
+				}
+
 				return (set_errno(EINVAL));
+			}
+
 			aiov[i].iov_len = iovlen32;
 			aiov[i].iov_base =
 			    (caddr_t)(uintptr_t)aiov32[i].iov_base;
 		}
+
+		if (iovsize != 0)
+			kmem_free(aiov32, iov32size);
 	} else
 #endif /* _SYSCALL32_IMPL */
 	if (copyin(lmsg.msg_iov, aiov, iovcnt * sizeof (struct iovec))) {
+		if (iovsize != 0)
+			kmem_free(aiov, iovsize);
+
 		return (set_errno(EFAULT));
 	}
 	len = 0;
@@ -1107,6 +1131,9 @@ recvmsg(int sock, struct nmsghdr *msg, int flags)
 		ssize_t iovlen = aiov[i].iov_len;
 		len += iovlen;
 		if (iovlen < 0 || len < 0) {
+			if (iovsize != 0)
+				kmem_free(aiov, iovsize);
+
 			return (set_errno(EINVAL));
 		}
 	}
@@ -1121,12 +1148,20 @@ recvmsg(int sock, struct nmsghdr *msg, int flags)
 	    (do_useracc == 0 ||
 	    useracc(lmsg.msg_control, lmsg.msg_controllen,
 	    B_WRITE) != 0)) {
+		if (iovsize != 0)
+			kmem_free(aiov, iovsize);
+
 		return (set_errno(EFAULT));
 	}
 
-	return (recvit(sock, &lmsg, &auio, flags,
+	rval = recvit(sock, &lmsg, &auio, flags,
 	    STRUCT_FADDR(umsgptr, msg_namelen),
-	    STRUCT_FADDR(umsgptr, msg_controllen), flagsp));
+	    STRUCT_FADDR(umsgptr, msg_controllen), flagsp);
+
+	if (iovsize != 0)
+		kmem_free(aiov, iovsize);
+
+	return (rval);
 }
 
 /*
@@ -1264,9 +1299,10 @@ sendmsg(int sock, struct nmsghdr *msg, int flags)
 	struct nmsghdr lmsg;
 	STRUCT_DECL(nmsghdr, u_lmsg);
 	struct uio auio;
-	struct iovec aiov[MSG_MAXIOVLEN];
+	struct iovec buf[IOV_MAX_STACK], *aiov = buf;
+	ssize_t iovsize = 0;
 	int iovcnt;
-	ssize_t len;
+	ssize_t len, rval;
 	int i;
 	model_t	model;
 
@@ -1309,7 +1345,7 @@ sendmsg(int sock, struct nmsghdr *msg, int flags)
 
 	iovcnt = lmsg.msg_iovlen;
 
-	if (iovcnt <= 0 || iovcnt > MSG_MAXIOVLEN) {
+	if (iovcnt <= 0 || iovcnt > IOV_MAX) {
 		/*
 		 * Unless this is XPG 4.2 we allow iovcnt == 0 to
 		 * be compatible with SunOS 4.X and 4.4BSD.
@@ -1318,19 +1354,34 @@ sendmsg(int sock, struct nmsghdr *msg, int flags)
 			return (set_errno(EMSGSIZE));
 	}
 
+	if (iovcnt > IOV_MAX_STACK) {
+		iovsize = iovcnt * sizeof (struct iovec);
+		aiov = kmem_alloc(iovsize, KM_SLEEP);
+	}
+
 #ifdef _SYSCALL32_IMPL
 	/*
 	 * 32-bit callers need to have their iovec expanded, while ensuring
 	 * that they can't move more than 2Gbytes of data in a single call.
 	 */
 	if (model == DATAMODEL_ILP32) {
-		struct iovec32 aiov32[MSG_MAXIOVLEN];
+		struct iovec32 buf32[IOV_MAX_STACK], *aiov32 = buf32;
+		ssize_t iov32size;
 		ssize32_t count32;
 
+		iov32size = iovcnt * sizeof (struct iovec32);
+		if (iovsize != 0)
+			aiov32 = kmem_alloc(iov32size, KM_SLEEP);
+
 		if (iovcnt != 0 &&
-		    copyin((struct iovec32 *)lmsg.msg_iov, aiov32,
-		    iovcnt * sizeof (struct iovec32)))
+		    copyin((struct iovec32 *)lmsg.msg_iov, aiov32, iov32size)) {
+			if (iovsize != 0) {
+				kmem_free(aiov32, iov32size);
+				kmem_free(aiov, iovsize);
+			}
+
 			return (set_errno(EFAULT));
+		}
 
 		count32 = 0;
 		for (i = 0; i < iovcnt; i++) {
@@ -1338,17 +1389,30 @@ sendmsg(int sock, struct nmsghdr *msg, int flags)
 
 			iovlen32 = aiov32[i].iov_len;
 			count32 += iovlen32;
-			if (iovlen32 < 0 || count32 < 0)
+			if (iovlen32 < 0 || count32 < 0) {
+				if (iovsize != 0) {
+					kmem_free(aiov32, iov32size);
+					kmem_free(aiov, iovsize);
+				}
+
 				return (set_errno(EINVAL));
+			}
+
 			aiov[i].iov_len = iovlen32;
 			aiov[i].iov_base =
 			    (caddr_t)(uintptr_t)aiov32[i].iov_base;
 		}
+
+		if (iovsize != 0)
+			kmem_free(aiov32, iov32size);
 	} else
 #endif /* _SYSCALL32_IMPL */
 	if (iovcnt != 0 &&
 	    copyin(lmsg.msg_iov, aiov,
 	    (unsigned)iovcnt * sizeof (struct iovec))) {
+		if (iovsize != 0)
+			kmem_free(aiov, iovsize);
+
 		return (set_errno(EFAULT));
 	}
 	len = 0;
@@ -1356,6 +1420,9 @@ sendmsg(int sock, struct nmsghdr *msg, int flags)
 		ssize_t iovlen = aiov[i].iov_len;
 		len += iovlen;
 		if (iovlen < 0 || len < 0) {
+			if (iovsize != 0)
+				kmem_free(aiov, iovsize);
+
 			return (set_errno(EINVAL));
 		}
 	}
@@ -1366,7 +1433,12 @@ sendmsg(int sock, struct nmsghdr *msg, int flags)
 	auio.uio_segflg = UIO_USERSPACE;
 	auio.uio_limit = 0;
 
-	return (sendit(sock, &lmsg, &auio, flags));
+	rval = sendit(sock, &lmsg, &auio, flags);
+
+	if (iovsize != 0)
+		kmem_free(aiov, iovsize);
+
+	return (rval);
 }
 
 ssize_t
diff --git a/usr/src/uts/common/fs/sockfs/socktpi.c b/usr/src/uts/common/fs/sockfs/socktpi.c
index d33f53f7a6..485e73eb02 100644
--- a/usr/src/uts/common/fs/sockfs/socktpi.c
+++ b/usr/src/uts/common/fs/sockfs/socktpi.c
@@ -6272,6 +6272,13 @@ sotpi_poll(
 	if (sti->sti_conn_ind_head != NULL)
 		*reventsp |= (POLLIN|POLLRDNORM) & events;
 
+	if (so->so_state & SS_CANTRCVMORE) {
+		*reventsp |= POLLRDHUP & events;
+
+		if (so->so_state & SS_CANTSENDMORE)
+			*reventsp |= POLLHUP;
+	}
+
 	if (so->so_state & SS_OOBPEND)
 		*reventsp |= POLLRDBAND & events;
 
diff --git a/usr/src/uts/common/fs/swapfs/swap_subr.c b/usr/src/uts/common/fs/swapfs/swap_subr.c
index 74c4302da9..a4d983665b 100644
--- a/usr/src/uts/common/fs/swapfs/swap_subr.c
+++ b/usr/src/uts/common/fs/swapfs/swap_subr.c
@@ -110,9 +110,11 @@ swapfs_recalc(pgcnt_t pgs)
 		 * memory that can be used as swap space should do so by
 		 * setting swapfs_desfree at boot time, not swapfs_minfree.
 		 * However, swapfs_minfree is tunable by install as a
-		 * workaround for bugid 1147463.
+		 * workaround for bugid 1147463. Note swapfs_minfree is set
+		 * to 1/8th of memory, but clamped at the limit of 256 MB.
 		 */
-		new_swapfs_minfree = MAX(btopr(2 * 1024 * 1024), pgs >> 3);
+		new_swapfs_minfree = MIN(MAX(btopr(2 * 1024 * 1024), pgs >> 3),
+		    btopr(256 * 1024 * 1024));
 	}
 
 	/*
diff --git a/usr/src/uts/common/fs/tmpfs/tmp_dir.c b/usr/src/uts/common/fs/tmpfs/tmp_dir.c
index f6621c8097..387cc6ae54 100644
--- a/usr/src/uts/common/fs/tmpfs/tmp_dir.c
+++ b/usr/src/uts/common/fs/tmpfs/tmp_dir.c
@@ -516,7 +516,7 @@ tdirdelete(
 	 */
 	namelen = strlen(tpdp->td_name) + 1;
 
-	tmp_memfree(tpdp, sizeof (struct tdirent) + namelen);
+	kmem_free(tpdp, sizeof (struct tdirent) + namelen);
 	dir->tn_size -= (sizeof (struct tdirent) + namelen);
 	dir->tn_dirents--;
 
@@ -549,8 +549,8 @@ tdirinit(
 	ASSERT(RW_WRITE_HELD(&parent->tn_rwlock));
 	ASSERT(dir->tn_type == VDIR);
 
-	dot = tmp_memalloc(sizeof (struct tdirent) + 2, TMP_MUSTHAVE);
-	dotdot = tmp_memalloc(sizeof (struct tdirent) + 3, TMP_MUSTHAVE);
+	dot = kmem_zalloc(sizeof (struct tdirent) + 2, KM_SLEEP);
+	dotdot = kmem_zalloc(sizeof (struct tdirent) + 3, KM_SLEEP);
 
 	/*
 	 * Initialize the entries
@@ -650,7 +650,7 @@ tdirtrunc(struct tmpnode *dir)
 
 		tmpfs_hash_out(tdp);
 
-		tmp_memfree(tdp, sizeof (struct tdirent) + namelen);
+		kmem_free(tdp, sizeof (struct tdirent) + namelen);
 		dir->tn_size -= (sizeof (struct tdirent) + namelen);
 		dir->tn_dirents--;
 	}
@@ -925,7 +925,7 @@ tdiraddentry(
 	 */
 	namelen = strlen(name) + 1;
 	alloc_size = namelen + sizeof (struct tdirent);
-	tdp = tmp_memalloc(alloc_size, 0);
+	tdp = kmem_zalloc(alloc_size, KM_NOSLEEP | KM_NORMALPRI);
 	if (tdp == NULL)
 		return (ENOSPC);
 
@@ -1025,7 +1025,7 @@ tdirmaketnode(
 	    ((va->va_mask & AT_MTIME) && TIMESPEC_OVERFLOW(&va->va_mtime)))
 		return (EOVERFLOW);
 	type = va->va_type;
-	tp = tmp_memalloc(sizeof (struct tmpnode), TMP_MUSTHAVE);
+	tp = kmem_zalloc(sizeof (struct tmpnode), KM_SLEEP);
 	tmpnode_init(tm, tp, va, cred);
 
 	/* setup normal file/dir's extended attribute directory */
diff --git a/usr/src/uts/common/fs/tmpfs/tmp_subr.c b/usr/src/uts/common/fs/tmpfs/tmp_subr.c
index 2e59d28d80..e6e2b392fe 100644
--- a/usr/src/uts/common/fs/tmpfs/tmp_subr.c
+++ b/usr/src/uts/common/fs/tmpfs/tmp_subr.c
@@ -20,6 +20,7 @@
  */
 /*
  * Copyright (c) 1990, 2010, Oracle and/or its affiliates. All rights reserved.
+ * Copyright 2015 Joyent, Inc.
  */
 
 #include <sys/types.h>
@@ -40,9 +41,19 @@
 #include <sys/policy.h>
 #include <sys/fs/tmp.h>
 #include <sys/fs/tmpnode.h>
+#include <sys/ddi.h>
+#include <sys/sunddi.h>
+
+#define	KILOBYTE	1024
+#define	MEGABYTE	(1024 * KILOBYTE)
+#define	GIGABYTE	(1024 * MEGABYTE)
 
 #define	MODESHIFT	3
 
+#define	VALIDMODEBITS	07777
+
+extern pgcnt_t swapfs_minfree;
+
 int
 tmp_taccess(void *vtp, int mode, struct cred *cred)
 {
@@ -71,7 +82,6 @@ tmp_taccess(void *vtp, int mode, struct cred *cred)
  * a plain file and you have write access to that file.
  * Function returns 0 if remove access is granted.
  */
-
 int
 tmp_sticky_remove_access(struct tmpnode *dir, struct tmpnode *entry,
 	struct cred *cr)
@@ -89,111 +99,122 @@ tmp_sticky_remove_access(struct tmpnode *dir, struct tmpnode *entry,
 }
 
 /*
- * Allocate zeroed memory if tmpfs_maxkmem has not been exceeded
- * or the 'musthave' flag is set.  'musthave' allocations should
- * always be subordinate to normal allocations so that tmpfs_maxkmem
- * can't be exceeded by more than a few KB.  Example: when creating
- * a new directory, the tmpnode is a normal allocation; if that
- * succeeds, the dirents for "." and ".." are 'musthave' allocations.
- */
-void *
-tmp_memalloc(size_t size, int musthave)
-{
-	static time_t last_warning;
-	time_t now;
-
-	if (atomic_add_long_nv(&tmp_kmemspace, size) < tmpfs_maxkmem ||
-	    musthave)
-		return (kmem_zalloc(size, KM_SLEEP));
-
-	atomic_add_long(&tmp_kmemspace, -size);
-	now = gethrestime_sec();
-	if (last_warning != now) {
-		last_warning = now;
-		cmn_err(CE_WARN, "tmp_memalloc: tmpfs over memory limit");
-	}
-	return (NULL);
-}
-
-void
-tmp_memfree(void *cp, size_t size)
-{
-	kmem_free(cp, size);
-	atomic_add_long(&tmp_kmemspace, -size);
-}
-
-/*
  * Convert a string containing a number (number of bytes) to a pgcnt_t,
  * containing the corresponding number of pages. On 32-bit kernels, the
  * maximum value encoded in 'str' is PAGESIZE * ULONG_MAX, while the value
  * returned in 'maxpg' is at most ULONG_MAX.
  *
- * If the number is followed by a "k" or "K", the value is converted from
- * kilobytes to bytes.  If it is followed by an "m" or "M" it is converted
- * from megabytes to bytes.  If it is not followed by a character it is
- * assumed to be in bytes. Multiple letter options are allowed, so for instance
- * '2mk' is interpreted as 2gb.
+ * The number may be followed by a magnitude suffix: "k" or "K" for kilobytes;
+ * "m" or "M" for megabytes; "g" or "G" for gigabytes.  This interface allows
+ * for an arguably esoteric interpretation of multiple suffix characters:
+ * namely, they cascade.  For example, the caller may specify "2mk", which is
+ * interpreted as 2 gigabytes.  It would seem, at this late stage, that the
+ * horse has left not only the barn but indeed the country, and possibly the
+ * entire planetary system. Alternatively, the number may be followed by a
+ * single '%' sign, indicating the size is a percentage of either the zone's
+ * swap limit or the system's overall swap size.
  *
  * Parse and overflow errors are detected and a non-zero number returned on
  * error.
  */
-
 int
 tmp_convnum(char *str, pgcnt_t *maxpg)
 {
-	uint64_t num = 0, oldnum;
+	u_longlong_t num = 0;
 #ifdef _LP64
-	uint64_t max_bytes = ULONG_MAX;
+	u_longlong_t max_bytes = ULONG_MAX;
 #else
-	uint64_t max_bytes = PAGESIZE * (uint64_t)ULONG_MAX;
+	u_longlong_t max_bytes = PAGESIZE * (uint64_t)ULONG_MAX;
 #endif
 	char *c;
-
-	if (str == NULL)
+	const struct convchar {
+		char *cc_char;
+		uint64_t cc_factor;
+	} convchars[] = {
+		{ "kK", KILOBYTE },
+		{ "mM", MEGABYTE },
+		{ "gG", GIGABYTE },
+		{ NULL, 0 }
+	};
+
+	if (str == NULL) {
 		return (EINVAL);
+	}
 	c = str;
 
 	/*
-	 * Convert str to number
+	 * Convert the initial numeric portion of the input string.
 	 */
-	while ((*c >= '0') && (*c <= '9')) {
-		oldnum = num;
-		num = num * 10 + (*c++ - '0');
-		if (oldnum > num) /* overflow */
+	if (ddi_strtoull(str, &c, 10, &num) != 0) {
+		return (EINVAL);
+	}
+
+	/*
+	 * Handle a size in percent. Anything other than a single percent
+	 * modifier is invalid. We use either the zone's swap limit or the
+	 * system's total available swap size as the initial value. Perform the
+	 * intermediate calculation in pages to avoid overflow.
+	 */
+	if (*c == '\%') {
+		u_longlong_t cap;
+
+		if (*(c + 1) != '\0')
+			return (EINVAL);
+
+		if (num > 100)
 			return (EINVAL);
+
+		cap = (u_longlong_t)curproc->p_zone->zone_max_swap_ctl;
+		if (cap == UINT64_MAX) {
+			/*
+			 * Use the amount of available physical and memory swap
+			 */
+			mutex_enter(&anoninfo_lock);
+			cap = TOTAL_AVAILABLE_SWAP;
+			mutex_exit(&anoninfo_lock);
+		} else {
+			cap = btop(cap);
+		}
+
+		num = ptob(cap * num / 100);
+		goto done;
 	}
 
 	/*
-	 * Terminate on null
+	 * Apply the (potentially cascading) magnitude suffixes until an
+	 * invalid character is found, or the string comes to an end.
 	 */
-	while (*c != '\0') {
-		switch (*c++) {
+	for (; *c != '\0'; c++) {
+		int i;
+
+		for (i = 0; convchars[i].cc_char != NULL; i++) {
+			/*
+			 * Check if this character matches this multiplier
+			 * class:
+			 */
+			if (strchr(convchars[i].cc_char, *c) != NULL) {
+				/*
+				 * Check for overflow:
+				 */
+				if (num > max_bytes / convchars[i].cc_factor) {
+					return (EINVAL);
+				}
+
+				num *= convchars[i].cc_factor;
+				goto valid_char;
+			}
+		}
 
 		/*
-		 * convert from kilobytes
+		 * This was not a valid multiplier suffix character.
 		 */
-		case 'k':
-		case 'K':
-			if (num > max_bytes / 1024) /* will overflow */
-				return (EINVAL);
-			num *= 1024;
-			break;
+		return (EINVAL);
 
-		/*
-		 * convert from megabytes
-		 */
-		case 'm':
-		case 'M':
-			if (num > max_bytes / (1024 * 1024)) /* will overflow */
-				return (EINVAL);
-			num *= 1024 * 1024;
-			break;
-
-		default:
-			return (EINVAL);
-		}
+valid_char:
+		continue;
 	}
 
+done:
 	/*
 	 * Since btopr() rounds up to page granularity, this round-up can
 	 * cause an overflow only if 'num' is between (max_bytes - PAGESIZE)
@@ -204,3 +225,29 @@ tmp_convnum(char *str, pgcnt_t *maxpg)
 		return (EINVAL);
 	return (0);
 }
+
+/*
+ * Parse an octal mode string for use as the permissions set for the root
+ * of the tmpfs mount.
+ */
+int
+tmp_convmode(char *str, mode_t *mode)
+{
+	ulong_t num;
+	char *c;
+
+	if (str == NULL) {
+		return (EINVAL);
+	}
+
+	if (ddi_strtoul(str, &c, 8, &num) != 0) {
+		return (EINVAL);
+	}
+
+	if ((num & ~VALIDMODEBITS) != 0) {
+		return (EINVAL);
+	}
+
+	*mode = VALIDMODEBITS & num;
+	return (0);
+}
diff --git a/usr/src/uts/common/fs/tmpfs/tmp_vfsops.c b/usr/src/uts/common/fs/tmpfs/tmp_vfsops.c
index f8a36a528f..3c088c442c 100644
--- a/usr/src/uts/common/fs/tmpfs/tmp_vfsops.c
+++ b/usr/src/uts/common/fs/tmpfs/tmp_vfsops.c
@@ -20,6 +20,7 @@
  */
 /*
  * Copyright (c) 1990, 2010, Oracle and/or its affiliates. All rights reserved.
+ * Copyright 2015 Joyent, Inc.
  */
 
 #include <sys/types.h>
@@ -55,6 +56,15 @@
 static int tmpfsfstype;
 
 /*
+ * tmpfs_mountcount is used to prevent module unloads while there is still
+ * state from a former mount hanging around. With forced umount support, the
+ * filesystem module must not be allowed to go away before the last
+ * VFS_FREEVFS() call has been made. Since this is just an atomic counter,
+ * there's no need for locking.
+ */
+static uint32_t	tmpfs_mountcount;
+
+/*
  * tmpfs vfs operations.
  */
 static int tmpfsinit(int, char *);
@@ -64,6 +74,7 @@ static int tmp_unmount(struct vfs *, int, struct cred *);
 static int tmp_root(struct vfs *, struct vnode **);
 static int tmp_statvfs(struct vfs *, struct statvfs64 *);
 static int tmp_vget(struct vfs *, struct vnode **, struct fid *);
+static void tmp_freevfs(vfs_t *vfsp);
 
 /*
  * Loadable module wrapper
@@ -76,7 +87,7 @@ static vfsdef_t vfw = {
 	VFSDEF_VERSION,
 	"tmpfs",
 	tmpfsinit,
-	VSW_HASPROTO|VSW_STATS|VSW_ZMOUNT,
+	VSW_HASPROTO|VSW_CANREMOUNT|VSW_STATS|VSW_ZMOUNT,
 	&tmpfs_proto_opttbl
 };
 
@@ -90,7 +101,8 @@ static mntopt_t tmpfs_options[] = {
 	/* Option name		Cancel Opt	Arg	Flags		Data */
 	{ MNTOPT_XATTR,		xattr_cancel,	NULL,	MO_DEFAULT,	NULL},
 	{ MNTOPT_NOXATTR,	noxattr_cancel,	NULL,	NULL,		NULL},
-	{ "size",		NULL,		"0",	MO_HASVALUE,	NULL}
+	{ "size",		NULL,		"0",	MO_HASVALUE,	NULL},
+	{ "mode",		NULL,		NULL,	MO_HASVALUE,	NULL}
 };
 
 
@@ -121,6 +133,14 @@ _fini()
 {
 	int error;
 
+	/*
+	 * If a forceably unmounted instance is still hanging around, we cannot
+	 * allow the module to be unloaded because that would cause panics once
+	 * the VFS framework decides it's time to call into VFS_FREEVFS().
+	 */
+	if (tmpfs_mountcount)
+		return (EBUSY);
+
 	error = mod_remove(&modlinkage);
 	if (error)
 		return (error);
@@ -139,14 +159,6 @@ _info(struct modinfo *modinfop)
 }
 
 /*
- * The following are patchable variables limiting the amount of system
- * resources tmpfs can use.
- *
- * tmpfs_maxkmem limits the amount of kernel kmem_alloc memory
- * tmpfs can use for it's data structures (e.g. tmpnodes, directory entries)
- * It is not determined by setting a hard limit but rather as a percentage of
- * physical memory which is determined when tmpfs is first used in the system.
- *
  * tmpfs_minfree is the minimum amount of swap space that tmpfs leaves for
  * the rest of the system.  In other words, if the amount of free swap space
  * in the system (i.e. anoninfo.ani_free) drops below tmpfs_minfree, tmpfs
@@ -155,9 +167,7 @@ _info(struct modinfo *modinfop)
  * There is also a per mount limit on the amount of swap space
  * (tmount.tm_anonmax) settable via a mount option.
  */
-size_t tmpfs_maxkmem = 0;
 size_t tmpfs_minfree = 0;
-size_t tmp_kmemspace;		/* bytes of kernel heap used by all tmpfs */
 
 static major_t tmpfs_major;
 static minor_t tmpfs_minor;
@@ -176,6 +186,7 @@ tmpfsinit(int fstype, char *name)
 		VFSNAME_ROOT,		{ .vfs_root = tmp_root },
 		VFSNAME_STATVFS,	{ .vfs_statvfs = tmp_statvfs },
 		VFSNAME_VGET,		{ .vfs_vget = tmp_vget },
+		VFSNAME_FREEVFS,	{ .vfs_freevfs = tmp_freevfs },
 		NULL,			NULL
 	};
 	int error;
@@ -210,27 +221,17 @@ tmpfsinit(int fstype, char *name)
 		tmpfs_minfree = btopr(TMPMINFREE);
 	}
 
-	/*
-	 * The maximum amount of space tmpfs can allocate is
-	 * TMPMAXPROCKMEM percent of kernel memory
-	 */
-	if (tmpfs_maxkmem == 0)
-		tmpfs_maxkmem = MAX(PAGESIZE, kmem_maxavail() / TMPMAXFRACKMEM);
-
 	if ((tmpfs_major = getudev()) == (major_t)-1) {
 		cmn_err(CE_WARN, "tmpfsinit: Can't get unique device number.");
 		tmpfs_major = 0;
 	}
 	mutex_init(&tmpfs_minor_lock, NULL, MUTEX_DEFAULT, NULL);
+	tmpfs_mountcount = 0;
 	return (0);
 }
 
 static int
-tmp_mount(
-	struct vfs *vfsp,
-	struct vnode *mvp,
-	struct mounta *uap,
-	struct cred *cr)
+tmp_mount(vfs_t *vfsp, vnode_t *mvp, struct mounta *uap, cred_t *cr)
 {
 	struct tmount *tm = NULL;
 	struct tmpnode *tp;
@@ -239,8 +240,9 @@ tmp_mount(
 	pgcnt_t anonmax;
 	struct vattr rattr;
 	int got_attrs;
-
-	char *sizestr;
+	boolean_t mode_arg = B_FALSE;
+	mode_t root_mode = 0777;
+	char *argstr;
 
 	if ((error = secpolicy_fs_mount(cr, mvp, vfsp)) != 0)
 		return (error);
@@ -249,7 +251,7 @@ tmp_mount(
 		return (ENOTDIR);
 
 	mutex_enter(&mvp->v_lock);
-	if ((uap->flags & MS_OVERLAY) == 0 &&
+	if ((uap->flags & MS_REMOUNT) == 0 && (uap->flags & MS_OVERLAY) == 0 &&
 	    (mvp->v_count != 1 || (mvp->v_flag & VROOT))) {
 		mutex_exit(&mvp->v_lock);
 		return (EBUSY);
@@ -275,18 +277,45 @@ tmp_mount(
 	 * tm_anonmax is set according to the mount arguments
 	 * if any.  Otherwise, it is set to a maximum value.
 	 */
-	if (vfs_optionisset(vfsp, "size", &sizestr)) {
-		if ((error = tmp_convnum(sizestr, &anonmax)) != 0)
+	if (vfs_optionisset(vfsp, "size", &argstr)) {
+		if ((error = tmp_convnum(argstr, &anonmax)) != 0)
 			goto out;
 	} else {
 		anonmax = ULONG_MAX;
 	}
 
+	/*
+	 * The "mode" mount argument allows the operator to override the
+	 * permissions of the root of the tmpfs mount.
+	 */
+	if (vfs_optionisset(vfsp, "mode", &argstr)) {
+		if ((error = tmp_convmode(argstr, &root_mode)) != 0) {
+			goto out;
+		}
+		mode_arg = B_TRUE;
+	}
+
 	if (error = pn_get(uap->dir,
 	    (uap->flags & MS_SYSSPACE) ? UIO_SYSSPACE : UIO_USERSPACE, &dpn))
 		goto out;
 
-	if ((tm = tmp_memalloc(sizeof (struct tmount), 0)) == NULL) {
+	if (uap->flags & MS_REMOUNT) {
+		tm = (struct tmount *)VFSTOTM(vfsp);
+
+		/*
+		 * If we change the size so its less than what is currently
+		 * being used, we allow that. The file system will simply be
+		 * full until enough files have been removed to get below the
+		 * new max.
+		 */
+		mutex_enter(&tm->tm_contents);
+		tm->tm_anonmax = anonmax;
+		mutex_exit(&tm->tm_contents);
+		goto out;
+	}
+
+	if ((tm = kmem_zalloc(sizeof (struct tmount),
+	    KM_NOSLEEP | KM_NORMALPRI)) == NULL) {
 		pn_free(&dpn);
 		error = ENOMEM;
 		goto out;
@@ -318,17 +347,17 @@ tmp_mount(
 	vfsp->vfs_bsize = PAGESIZE;
 	vfsp->vfs_flag |= VFS_NOTRUNC;
 	vfs_make_fsid(&vfsp->vfs_fsid, tm->tm_dev, tmpfsfstype);
-	tm->tm_mntpath = tmp_memalloc(dpn.pn_pathlen + 1, TMP_MUSTHAVE);
+	tm->tm_mntpath = kmem_zalloc(dpn.pn_pathlen + 1, KM_SLEEP);
 	(void) strcpy(tm->tm_mntpath, dpn.pn_path);
 
 	/*
 	 * allocate and initialize root tmpnode structure
 	 */
 	bzero(&rattr, sizeof (struct vattr));
-	rattr.va_mode = (mode_t)(S_IFDIR | 0777);	/* XXX modes */
+	rattr.va_mode = (mode_t)(S_IFDIR | root_mode);
 	rattr.va_type = VDIR;
 	rattr.va_rdev = 0;
-	tp = tmp_memalloc(sizeof (struct tmpnode), TMP_MUSTHAVE);
+	tp = kmem_zalloc(sizeof (struct tmpnode), KM_SLEEP);
 	tmpnode_init(tm, tp, &rattr, cr);
 
 	/*
@@ -345,7 +374,14 @@ tmp_mount(
 	 * the previously set hardwired defaults to prevail.
 	 */
 	if (got_attrs == 0) {
-		tp->tn_mode = rattr.va_mode;
+		if (!mode_arg) {
+			/*
+			 * Only use the underlying mount point for the
+			 * mode if the "mode" mount argument was not
+			 * provided.
+			 */
+			tp->tn_mode = rattr.va_mode;
+		}
 		tp->tn_uid = rattr.va_uid;
 		tp->tn_gid = rattr.va_gid;
 	}
@@ -366,6 +402,7 @@ tmp_mount(
 
 	pn_free(&dpn);
 	error = 0;
+	atomic_inc_32(&tmpfs_mountcount);
 
 out:
 	if (error == 0)
@@ -381,36 +418,107 @@ tmp_unmount(struct vfs *vfsp, int flag, struct cred *cr)
 	struct tmpnode *tnp, *cancel;
 	struct vnode	*vp;
 	int error;
+	uint_t cnt;
+	int i;
 
 	if ((error = secpolicy_fs_unmount(cr, vfsp)) != 0)
 		return (error);
 
-	/*
-	 * forced unmount is not supported by this file system
-	 * and thus, ENOTSUP, is being returned.
-	 */
-	if (flag & MS_FORCE)
-		return (ENOTSUP);
-
 	mutex_enter(&tm->tm_contents);
 
 	/*
-	 * If there are no open files, only the root node should have
-	 * a reference count.
+	 * In the normal unmount case (non-forced unmount), if there are no
+	 * open files, only the root node should have a reference count.
+	 *
 	 * With tm_contents held, nothing can be added or removed.
 	 * There may be some dirty pages.  To prevent fsflush from
 	 * disrupting the unmount, put a hold on each node while scanning.
 	 * If we find a previously referenced node, undo the holds we have
 	 * placed and fail EBUSY.
+	 *
+	 * However, in the case of a forced umount, things are a bit different.
+	 * An additional VFS_HOLD is added for each outstanding VN_HOLD to
+	 * ensure that the file system is not cleaned up (tmp_freevfs) until
+	 * the last vfs hold is dropped. This happens in tmp_inactive as the
+	 * vnodes are released. Also, we can't add an additional VN_HOLD in
+	 * this case since that would prevent tmp_inactive from ever being
+	 * called. Finally, we do need to drop the zone ref now (zone_rele_ref)
+	 * so that the zone is not blocked waiting for the final file system
+	 * cleanup.
 	 */
 	tnp = tm->tm_rootnode;
-	if (TNTOV(tnp)->v_count > 1) {
+
+	vp = TNTOV(tnp);
+	mutex_enter(&vp->v_lock);
+	cnt = vp->v_count;
+	if (flag & MS_FORCE) {
+		vfsp->vfs_flag |= VFS_UNMOUNTED;
+		/* Extra hold which we rele below when we drop the zone ref */
+		VFS_HOLD(vfsp);
+
+		for (i = 1; i < cnt; i++)
+			VFS_HOLD(vfsp);
+
+		/* drop the mutex now because no one can find this mount */
+		mutex_exit(&tm->tm_contents);
+	} else if (cnt > 1) {
+		mutex_exit(&vp->v_lock);
 		mutex_exit(&tm->tm_contents);
 		return (EBUSY);
 	}
+	mutex_exit(&vp->v_lock);
 
+	/*
+	 * Check for open files. An open file causes everything to unwind
+	 * unless this is a forced umount.
+	 */
 	for (tnp = tnp->tn_forw; tnp; tnp = tnp->tn_forw) {
-		if ((vp = TNTOV(tnp))->v_count > 0) {
+		vp = TNTOV(tnp);
+		mutex_enter(&vp->v_lock);
+		cnt = vp->v_count;
+		if (flag & MS_FORCE) {
+			for (i = 0; i < cnt; i++)
+				VFS_HOLD(vfsp);
+
+			/*
+			 * In the case of a forced umount don't add an
+			 * additional VN_HOLD on the already held vnodes, like
+			 * we do in the non-forced unmount case. If the
+			 * cnt > 0, then the vnode already has at least one
+			 * hold and we need tmp_inactive to get called when the
+			 * last pre-existing hold on the node is released so
+			 * that we can VFS_RELE the VFS holds we just added.
+			 */
+			if (cnt == 0) {
+				/* directly add VN_HOLD since have the lock */
+				vp->v_count++;
+			}
+
+			mutex_exit(&vp->v_lock);
+
+			/*
+			 * If the tmpnode has any pages associated with it
+			 * (i.e. if it's a normal file with non-zero size), the
+			 * tmpnode could still be discovered by pageout or
+			 * fsflush via the page vnode pointers. To prevent this
+			 * from interfering with the tmp_freevfs, truncate the
+			 * tmpnode now.
+			 */
+			if (tnp->tn_size != 0 && tnp->tn_type == VREG) {
+				rw_enter(&tnp->tn_rwlock, RW_WRITER);
+				rw_enter(&tnp->tn_contents, RW_WRITER);
+
+				(void) tmpnode_trunc(tm, tnp, 0);
+
+				rw_exit(&tnp->tn_contents);
+				rw_exit(&tnp->tn_rwlock);
+
+				ASSERT(tnp->tn_size == 0);
+				ASSERT(tnp->tn_nblocks == 0);
+			}
+		} else if (cnt > 0) {
+			/* An open file; unwind the holds we've been adding. */
+			mutex_exit(&vp->v_lock);
 			cancel = tm->tm_rootnode->tn_forw;
 			while (cancel != tnp) {
 				vp = TNTOV(cancel);
@@ -420,14 +528,50 @@ tmp_unmount(struct vfs *vfsp, int flag, struct cred *cr)
 			}
 			mutex_exit(&tm->tm_contents);
 			return (EBUSY);
+		} else {
+			/* directly add a VN_HOLD since we have the lock */
+			vp->v_count++;
+			mutex_exit(&vp->v_lock);
 		}
-		VN_HOLD(vp);
 	}
 
-	/*
-	 * We can drop the mutex now because no one can find this mount
-	 */
-	mutex_exit(&tm->tm_contents);
+	if (flag & MS_FORCE) {
+		/*
+		 * Drop the zone ref now since we don't know how long it will
+		 * be until the final vfs_rele is called by tmp_inactive.
+		 */
+		if (vfsp->vfs_zone) {
+			zone_rele_ref(&vfsp->vfs_implp->vi_zone_ref,
+			    ZONE_REF_VFS);
+			vfsp->vfs_zone = 0;
+		}
+		/* We can now drop the extra hold we added above. */
+		VFS_RELE(vfsp);
+	} else {
+		/*
+		 * For the non-forced case, we can drop the mutex now because
+		 * no one can find this mount anymore
+		 */
+		vfsp->vfs_flag |= VFS_UNMOUNTED;
+		mutex_exit(&tm->tm_contents);
+	}
+
+	return (0);
+}
+
+/*
+ * Implementation of VFS_FREEVFS() to support forced umounts. This is called by
+ * the vfs framework after umount and the last VFS_RELE, to trigger the release
+ * of any resources still associated with the given vfs_t. We only add
+ * additional VFS_HOLDs during the forced umount case, so this is normally
+ * called immediately after tmp_umount.
+ */
+void
+tmp_freevfs(vfs_t *vfsp)
+{
+	struct tmount *tm = (struct tmount *)VFSTOTM(vfsp);
+	struct tmpnode *tnp;
+	struct vnode	*vp;
 
 	/*
 	 * Free all kmemalloc'd and anonalloc'd memory associated with
@@ -437,6 +581,16 @@ tmp_unmount(struct vfs *vfsp, int flag, struct cred *cr)
 	 * tmpnode_free which assumes that the directory entry has been
 	 * removed before the file.
 	 */
+
+	/*
+	 * Now that we are tearing ourselves down we need to remove the
+	 * UNMOUNTED flag. If we don't, we'll later hit a VN_RELE when we remove
+	 * files from the system causing us to have a negative value. Doing this
+	 * seems a bit better than trying to set a flag on the tmount that says
+	 * we're tearing down.
+	 */
+	vfsp->vfs_flag &= ~VFS_UNMOUNTED;
+
 	/*
 	 * Remove all directory entries
 	 */
@@ -503,15 +657,16 @@ tmp_unmount(struct vfs *vfsp, int flag, struct cred *cr)
 
 	ASSERT(tm->tm_mntpath);
 
-	tmp_memfree(tm->tm_mntpath, strlen(tm->tm_mntpath) + 1);
+	kmem_free(tm->tm_mntpath, strlen(tm->tm_mntpath) + 1);
 
 	ASSERT(tm->tm_anonmem == 0);
 
 	mutex_destroy(&tm->tm_contents);
 	mutex_destroy(&tm->tm_renamelck);
-	tmp_memfree(tm, sizeof (struct tmount));
+	kmem_free(tm, sizeof (struct tmount));
 
-	return (0);
+	/* Allow _fini() to succeed now */
+	atomic_dec_32(&tmpfs_mountcount);
 }
 
 /*
@@ -614,13 +769,7 @@ tmp_statvfs(struct vfs *vfsp, struct statvfs64 *sbp)
 	 * available to tmpfs.  This is fairly inaccurate since it doesn't
 	 * take into account the names stored in the directory entries.
 	 */
-	if (tmpfs_maxkmem > tmp_kmemspace)
-		sbp->f_ffree = (tmpfs_maxkmem - tmp_kmemspace) /
-		    (sizeof (struct tmpnode) + sizeof (struct tdirent));
-	else
-		sbp->f_ffree = 0;
-
-	sbp->f_files = tmpfs_maxkmem /
+	sbp->f_ffree = sbp->f_files = ptob(availrmem) /
 	    (sizeof (struct tmpnode) + sizeof (struct tdirent));
 	sbp->f_favail = (fsfilcnt64_t)(sbp->f_ffree);
 	(void) cmpldev(&d32, vfsp->vfs_dev);
diff --git a/usr/src/uts/common/fs/tmpfs/tmp_vnops.c b/usr/src/uts/common/fs/tmpfs/tmp_vnops.c
index 61607a6dcc..464d638db0 100644
--- a/usr/src/uts/common/fs/tmpfs/tmp_vnops.c
+++ b/usr/src/uts/common/fs/tmpfs/tmp_vnops.c
@@ -25,7 +25,7 @@
  */
 
 /*
- * Copyright (c) 2012, Joyent, Inc. All rights reserved.
+ * Copyright (c) 2014, Joyent, Inc. All rights reserved.
  * Copyright 2015 Nexenta Systems, Inc.  All rights reserved.
  */
 
@@ -584,6 +584,10 @@ tmp_read(struct vnode *vp, struct uio *uiop, int ioflag, cred_t *cred,
 	struct tmount *tm = (struct tmount *)VTOTM(vp);
 	int error;
 
+	/* If the filesystem was umounted by force, return immediately. */
+	if (vp->v_vfsp->vfs_flag & VFS_UNMOUNTED)
+		return (EIO);
+
 	/*
 	 * We don't currently support reading non-regular files
 	 */
@@ -613,6 +617,10 @@ tmp_write(struct vnode *vp, struct uio *uiop, int ioflag, struct cred *cred,
 	struct tmount *tm = (struct tmount *)VTOTM(vp);
 	int error;
 
+	/* If the filesystem was umounted by force, return immediately. */
+	if (vp->v_vfsp->vfs_flag & VFS_UNMOUNTED)
+		return (EIO);
+
 	/*
 	 * We don't currently support writing to non-regular files
 	 */
@@ -833,6 +841,9 @@ tmp_lookup(
 	struct tmpnode *ntp = NULL;
 	int error;
 
+	/* If the filesystem was umounted by force, return immediately. */
+	if (dvp->v_vfsp->vfs_flag & VFS_UNMOUNTED)
+		return (EIO);
 
 	/* allow cd into @ dir */
 	if (flags & LOOKUP_XATTR) {
@@ -871,8 +882,7 @@ tmp_lookup(
 				return (error);
 			}
 
-			xdp = tmp_memalloc(sizeof (struct tmpnode),
-			    TMP_MUSTHAVE);
+			xdp = kmem_zalloc(sizeof (struct tmpnode), KM_SLEEP);
 			tm = VTOTM(dvp);
 			tmpnode_init(tm, xdp, &tp->tn_attr, NULL);
 			/*
@@ -1186,7 +1196,7 @@ tmp_rename(
 	struct tmpnode *toparent;
 	struct tmpnode *fromtp = NULL;	/* source tmpnode */
 	struct tmount *tm = (struct tmount *)VTOTM(odvp);
-	int error;
+	int error, err;
 	int samedir = 0;	/* set if odvp == ndvp */
 	struct vnode *realvp;
 
@@ -1262,15 +1272,6 @@ tmp_rename(
 			error = 0;
 		goto done;
 	}
-	vnevent_rename_src(TNTOV(fromtp), odvp, onm, ct);
-
-	/*
-	 * Notify the target directory if not same as
-	 * source directory.
-	 */
-	if (ndvp != odvp) {
-		vnevent_rename_dest_dir(ndvp, ct);
-	}
 
 	/*
 	 * Unlink from source.
@@ -1278,7 +1279,7 @@ tmp_rename(
 	rw_enter(&fromparent->tn_rwlock, RW_WRITER);
 	rw_enter(&fromtp->tn_rwlock, RW_WRITER);
 
-	error = tdirdelete(fromparent, fromtp, onm, DR_RENAME, cred);
+	error = err = tdirdelete(fromparent, fromtp, onm, DR_RENAME, cred);
 
 	/*
 	 * The following handles the case where our source tmpnode was
@@ -1293,6 +1294,12 @@ tmp_rename(
 
 	rw_exit(&fromtp->tn_rwlock);
 	rw_exit(&fromparent->tn_rwlock);
+
+	if (err == 0) {
+		vnevent_rename_src(TNTOV(fromtp), odvp, onm, ct);
+		vnevent_rename_dest_dir(ndvp, TNTOV(fromtp), nnm, ct);
+	}
+
 done:
 	tmpnode_rele(fromtp);
 	mutex_exit(&tm->tm_renamelck);
@@ -1459,6 +1466,10 @@ tmp_readdir(
 	int reclen;
 	caddr_t outbuf;
 
+	/* If the filesystem was umounted by force, return immediately. */
+	if (vp->v_vfsp->vfs_flag & VFS_UNMOUNTED)
+		return (EIO);
+
 	if (uiop->uio_loffset >= MAXOFF_T) {
 		if (eofp)
 			*eofp = 1;
@@ -1597,7 +1608,7 @@ tmp_symlink(
 		return (error);
 	}
 	len = strlen(tnm) + 1;
-	cp = tmp_memalloc(len, 0);
+	cp = kmem_alloc(len, KM_NOSLEEP | KM_NORMALPRI);
 	if (cp == NULL) {
 		tmpnode_rele(self);
 		return (ENOSPC);
@@ -1662,10 +1673,27 @@ top:
 	 * there's little to do -- just drop our hold.
 	 */
 	if (vp->v_count > 1 || tp->tn_nlink != 0) {
-		vp->v_count--;
+		if (vp->v_vfsp->vfs_flag & VFS_UNMOUNTED) {
+			/*
+			 * Since the file system was forcibly unmounted, we can
+			 * have a case (v_count == 1, tn_nlink != 0) where this
+			 * file was open so we didn't add an extra hold on the
+			 * file in tmp_unmount. We are counting on the
+			 * interaction of the hold made in tmp_unmount and
+			 * rele-ed in tmp_vfsfree so we need to be sure we
+			 * don't decrement in this case.
+			 */
+			if (vp->v_count > 1)
+				vp->v_count--;
+		} else {
+			vp->v_count--;
+		}
 		mutex_exit(&vp->v_lock);
 		mutex_exit(&tp->tn_tlock);
 		rw_exit(&tp->tn_rwlock);
+		/* If the filesystem was umounted by force, rele the vfs ref */
+		if (tm->tm_vfsp->vfs_flag & VFS_UNMOUNTED)
+			VFS_RELE(tm->tm_vfsp);
 		return;
 	}
 
@@ -1690,7 +1718,7 @@ top:
 			goto top;
 		}
 		if (tp->tn_type == VLNK)
-			tmp_memfree(tp->tn_symlink, tp->tn_size + 1);
+			kmem_free(tp->tn_symlink, tp->tn_size + 1);
 	}
 
 	/*
@@ -1724,7 +1752,11 @@ top:
 	rw_destroy(&tp->tn_rwlock);
 	mutex_destroy(&tp->tn_tlock);
 	vn_free(TNTOV(tp));
-	tmp_memfree(tp, sizeof (struct tmpnode));
+	kmem_free(tp, sizeof (struct tmpnode));
+
+	/* If the filesystem was umounted by force, rele the vfs ref */
+	if (tm->tm_vfsp->vfs_flag & VFS_UNMOUNTED)
+		VFS_RELE(tm->tm_vfsp);
 }
 
 /* ARGSUSED2 */
@@ -1846,6 +1878,10 @@ tmp_getapage(
 	struct vnode *pvp;
 	u_offset_t poff;
 
+	/* If the filesystem was umounted by force, return immediately. */
+	if (vp->v_vfsp->vfs_flag & VFS_UNMOUNTED)
+		return (EIO);
+
 	if (protp != NULL)
 		*protp = PROT_ALL;
 again:
@@ -2067,6 +2103,10 @@ tmp_putapage(
 	u_offset_t offset;
 	u_offset_t tmpoff;
 
+	/* If the filesystem was umounted by force, return immediately. */
+	if (vp->v_vfsp->vfs_flag & VFS_UNMOUNTED)
+		return (EIO);
+
 	ASSERT(PAGE_LOCKED(pp));
 
 	/* Kluster in tmp_klustsize chunks */
diff --git a/usr/src/uts/common/fs/udfs/udf_dir.c b/usr/src/uts/common/fs/udfs/udf_dir.c
index c1e2c74a87..def046a0bf 100644
--- a/usr/src/uts/common/fs/udfs/udf_dir.c
+++ b/usr/src/uts/common/fs/udfs/udf_dir.c
@@ -20,6 +20,7 @@
  */
 /*
  * Copyright (c) 1998, 2010, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2014, Joyent, Inc. All rights reserved.
  */
 
 #include <sys/types.h>
@@ -562,9 +563,8 @@ out:
 				    namep, ctp);
 			}
 
-			if (sdp != tdp) {
-				vnevent_rename_dest_dir(ITOV(tdp), ctp);
-			}
+			vnevent_rename_dest_dir(ITOV(tdp), ITOV(tip),
+			    namep, ctp);
 		}
 
 		/*
diff --git a/usr/src/uts/common/fs/udfs/udf_vnops.c b/usr/src/uts/common/fs/udfs/udf_vnops.c
index 307d3987ed..2d8de23399 100644
--- a/usr/src/uts/common/fs/udfs/udf_vnops.c
+++ b/usr/src/uts/common/fs/udfs/udf_vnops.c
@@ -911,7 +911,7 @@ udf_rename(
 	caller_context_t *ct,
 	int flags)
 {
-	int32_t error = 0;
+	int32_t error = 0, err;
 	struct udf_vfs *udf_vfsp;
 	struct ud_inode *sip;		/* source inode */
 	struct ud_inode *sdp, *tdp;	/* source and target parent inode */
@@ -995,7 +995,6 @@ udf_rename(
 		rw_exit(&tdp->i_rwlock);
 		goto errout;
 	}
-	vnevent_rename_src(ITOV(sip), sdvp, snm, ct);
 	rw_exit(&tdp->i_rwlock);
 
 	rw_enter(&sdp->i_rwlock, RW_WRITER);
@@ -1006,11 +1005,15 @@ udf_rename(
 	 * If the entry has changed just forget about it.  Release
 	 * the source inode.
 	 */
-	if ((error = ud_dirremove(sdp, snm, sip, (struct vnode *)0,
+	if ((error = err = ud_dirremove(sdp, snm, sip, (struct vnode *)0,
 	    DR_RENAME, cr, ct)) == ENOENT) {
 		error = 0;
 	}
 	rw_exit(&sdp->i_rwlock);
+
+	if (err == 0)
+		vnevent_rename_src(ITOV(sip), sdvp, snm, ct);
+
 errout:
 	ITIMES(sdp);
 	ITIMES(tdp);
diff --git a/usr/src/uts/common/fs/ufs/ufs_vnops.c b/usr/src/uts/common/fs/ufs/ufs_vnops.c
index fcffd952ed..c77872b11d 100644
--- a/usr/src/uts/common/fs/ufs/ufs_vnops.c
+++ b/usr/src/uts/common/fs/ufs/ufs_vnops.c
@@ -21,7 +21,7 @@
 
 /*
  * Copyright (c) 1984, 2010, Oracle and/or its affiliates. All rights reserved.
- * Copyright (c) 2013, Joyent, Inc. All rights reserved.
+ * Copyright (c) 2014, Joyent, Inc. All rights reserved.
  */
 
 /*	Copyright (c) 1983, 1984, 1985, 1986, 1987, 1988, 1989 AT&T	*/
@@ -3367,11 +3367,10 @@ ufs_rename(
 	struct inode *ip = NULL;	/* check inode */
 	struct inode *sdp;		/* old (source) parent inode */
 	struct inode *tdp;		/* new (target) parent inode */
-	struct vnode *svp = NULL;	/* source vnode */
 	struct vnode *tvp = NULL;	/* target vnode, if it exists */
 	struct vnode *realvp;
 	struct ufsvfs *ufsvfsp;
-	struct ulockfs *ulp;
+	struct ulockfs *ulp = NULL;
 	struct ufs_slot slot;
 	timestruc_t now;
 	int error;
@@ -3380,7 +3379,7 @@ ufs_rename(
 	krwlock_t *first_lock;
 	krwlock_t *second_lock;
 	krwlock_t *reverse_lock;
-	int serr, terr;
+	int terr;
 
 	sdp = VTOI(sdvp);
 	slot.fbp = NULL;
@@ -3389,34 +3388,13 @@ ufs_rename(
 	if (VOP_REALVP(tdvp, &realvp, ct) == 0)
 		tdvp = realvp;
 
+	/* Must do this before taking locks in case of DNLC miss */
 	terr = ufs_eventlookup(tdvp, tnm, cr, &tvp);
-	serr = ufs_eventlookup(sdvp, snm, cr, &svp);
-
-	if ((serr == 0) && ((terr == 0) || (terr == ENOENT))) {
-		if (tvp != NULL)
-			vnevent_rename_dest(tvp, tdvp, tnm, ct);
-
-		/*
-		 * Notify the target directory of the rename event
-		 * if source and target directories are not the same.
-		 */
-		if (sdvp != tdvp)
-			vnevent_rename_dest_dir(tdvp, ct);
-
-		if (svp != NULL)
-			vnevent_rename_src(svp, sdvp, snm, ct);
-	}
-
-	if (tvp != NULL)
-		VN_RELE(tvp);
-
-	if (svp != NULL)
-		VN_RELE(svp);
 
 retry_rename:
 	error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_RENAME_MASK);
 	if (error)
-		goto out;
+		goto unlock;
 
 	if (ulp)
 		TRANS_BEGIN_CSYNC(ufsvfsp, issync, TOP_RENAME,
@@ -3712,6 +3690,9 @@ retry_firstlock:
 		goto errout;
 	}
 
+	if (terr == 0 && tvp != NULL)
+		vnevent_rename_dest(tvp, tdvp, tnm, ct);
+
 	/*
 	 * Unlink the source.
 	 * Remove the source entry.  ufs_dirremove() checks that the entry
@@ -3723,6 +3704,9 @@ retry_firstlock:
 	    DR_RENAME, cr)) == ENOENT)
 		error = 0;
 
+	vnevent_rename_src(ITOV(sip), sdvp, snm, ct);
+	vnevent_rename_dest_dir(tdvp, ITOV(sip), tnm, ct);
+
 errout:
 	if (slot.fbp)
 		fbrelse(slot.fbp, S_OTHER);
@@ -3732,15 +3716,17 @@ errout:
 		rw_exit(&sdp->i_rwlock);
 	}
 
-	VN_RELE(ITOV(sip));
-
 unlock:
+	if (tvp != NULL)
+		VN_RELE(tvp);
+	if (sip != NULL)
+		VN_RELE(ITOV(sip));
+
 	if (ulp) {
 		TRANS_END_CSYNC(ufsvfsp, error, issync, TOP_RENAME, trans_size);
 		ufs_lockfs_end(ulp);
 	}
 
-out:
 	return (error);
 }
 
diff --git a/usr/src/uts/common/fs/vfs.c b/usr/src/uts/common/fs/vfs.c
index 6e93d056df..b680b1168c 100644
--- a/usr/src/uts/common/fs/vfs.c
+++ b/usr/src/uts/common/fs/vfs.c
@@ -20,7 +20,7 @@
  */
 /*
  * Copyright (c) 1988, 2010, Oracle and/or its affiliates. All rights reserved.
- * Copyright (c) 2012, Joyent, Inc. All rights reserved.
+ * Copyright (c) 2015, Joyent, Inc.
  */
 
 /*	Copyright (c) 1983, 1984, 1985, 1986, 1987, 1988, 1989 AT&T	*/
@@ -235,7 +235,8 @@ fsop_root(vfs_t *vfsp, vnode_t **vpp)
 	 * Make sure this root has a path.  With lofs, it is possible to have
 	 * a NULL mountpoint.
 	 */
-	if (ret == 0 && vfsp->vfs_mntpt != NULL && (*vpp)->v_path == NULL) {
+	if (ret == 0 && vfsp->vfs_mntpt != NULL &&
+	    (*vpp)->v_path == vn_vpath_empty) {
 		mntpt = vfs_getmntpoint(vfsp);
 		vn_setpath_str(*vpp, refstr_value(mntpt),
 		    strlen(refstr_value(mntpt)));
@@ -905,6 +906,7 @@ vfs_mountroot(void)
 	vfs_mountfs("mntfs", "/etc/mnttab", "/etc/mnttab");
 	vfs_mountfs("tmpfs", "/etc/svc/volatile", "/etc/svc/volatile");
 	vfs_mountfs("objfs", "objfs", OBJFS_ROOT);
+	vfs_mountfs("bootfs", "bootfs", "/system/boot");
 
 	if (getzoneid() == GLOBAL_ZONEID) {
 		vfs_mountfs("sharefs", "sharefs", "/etc/dfs/sharetab");
@@ -3899,6 +3901,8 @@ vfs_to_modname(const char *vfstype)
 		vfstype = "fdfs";
 	} else if (strncmp(vfstype, "nfs", 3) == 0) {
 		vfstype = "nfs";
+	} else if (strcmp(vfstype, "lxproc") == 0) {
+		vfstype = "lxprocfs";
 	}
 
 	return (vfstype);
diff --git a/usr/src/uts/common/fs/vnode.c b/usr/src/uts/common/fs/vnode.c
index 4abb040de0..9fb1ea702a 100644
--- a/usr/src/uts/common/fs/vnode.c
+++ b/usr/src/uts/common/fs/vnode.c
@@ -21,7 +21,7 @@
 
 /*
  * Copyright (c) 1988, 2010, Oracle and/or its affiliates. All rights reserved.
- * Copyright (c) 2013, Joyent, Inc. All rights reserved.
+ * Copyright (c) 2015, Joyent, Inc.
  */
 
 /*	Copyright (c) 1983, 1984, 1985, 1986, 1987, 1988, 1989 AT&T	*/
@@ -66,6 +66,7 @@
 #include <fs/fs_subr.h>
 #include <sys/taskq.h>
 #include <fs/fs_reparse.h>
+#include <sys/time.h>
 
 /* Determine if this vnode is a file that is read-only */
 #define	ISROFILE(vp)	\
@@ -102,6 +103,9 @@ kmutex_t	vskstat_tree_lock;
 /* Global variable which enables/disables the vopstats collection */
 int vopstats_enabled = 1;
 
+/* Global used for empty/invalid v_path */
+char *vn_vpath_empty = "";
+
 /*
  * forward declarations for internal vnode specific data (vsd)
  */
@@ -200,6 +204,11 @@ static void 		(**vsd_destructor)(void *);
 		cr = crgetmapped(cr);					\
 	}
 
+#define	VOP_LATENCY_10MS	10000000
+#define	VOP_LATENCY_100MS	100000000
+#define	VOP_LATENCY_1S		1000000000
+#define	VOP_LATENCY_10S		10000000000
+
 /*
  * Convert stat(2) formats to vnode types and vice versa.  (Knows about
  * numerical order of S_IFMT and vnode types.)
@@ -2284,7 +2293,7 @@ vn_cache_constructor(void *buf, void *cdrarg, int kmflags)
 	cv_init(&vp->v_cv, NULL, CV_DEFAULT, NULL);
 	rw_init(&vp->v_nbllock, NULL, RW_DEFAULT, NULL);
 	vp->v_femhead = NULL;	/* Must be done before vn_reinit() */
-	vp->v_path = NULL;
+	vp->v_path = vn_vpath_empty;
 	vp->v_mpssdata = NULL;
 	vp->v_vsd = NULL;
 	vp->v_fopdata = NULL;
@@ -2331,6 +2340,7 @@ void
 vn_recycle(vnode_t *vp)
 {
 	ASSERT(vp->v_pages == NULL);
+	VERIFY(vp->v_path != NULL);
 
 	/*
 	 * XXX - This really belongs in vn_reinit(), but we have some issues
@@ -2353,9 +2363,9 @@ vn_recycle(vnode_t *vp)
 		kmem_free(vp->v_femhead, sizeof (*(vp->v_femhead)));
 		vp->v_femhead = NULL;
 	}
-	if (vp->v_path) {
+	if (vp->v_path != vn_vpath_empty) {
 		kmem_free(vp->v_path, strlen(vp->v_path) + 1);
-		vp->v_path = NULL;
+		vp->v_path = vn_vpath_empty;
 	}
 
 	if (vp->v_fopdata != NULL) {
@@ -2427,9 +2437,10 @@ vn_free(vnode_t *vp)
 	 */
 	ASSERT((vp->v_count == 0) || (vp->v_count == 1));
 	ASSERT(vp->v_count_dnlc == 0);
-	if (vp->v_path != NULL) {
+	VERIFY(vp->v_path != NULL);
+	if (vp->v_path != vn_vpath_empty) {
 		kmem_free(vp->v_path, strlen(vp->v_path) + 1);
-		vp->v_path = NULL;
+		vp->v_path = vn_vpath_empty;
 	}
 
 	/* If FEM was in use, make sure everything gets cleaned up */
@@ -2516,6 +2527,7 @@ vnevent_rename_src(vnode_t *vp, vnode_t *dvp, char *name, caller_context_t *ct)
 	if (vp == NULL || vp->v_femhead == NULL) {
 		return;
 	}
+	(void) VOP_VNEVENT(dvp, VE_RENAME_SRC_DIR, vp, name, ct);
 	(void) VOP_VNEVENT(vp, VE_RENAME_SRC, dvp, name, ct);
 }
 
@@ -2530,12 +2542,13 @@ vnevent_rename_dest(vnode_t *vp, vnode_t *dvp, char *name,
 }
 
 void
-vnevent_rename_dest_dir(vnode_t *vp, caller_context_t *ct)
+vnevent_rename_dest_dir(vnode_t *vp, vnode_t *nvp, char *name,
+    caller_context_t *ct)
 {
 	if (vp == NULL || vp->v_femhead == NULL) {
 		return;
 	}
-	(void) VOP_VNEVENT(vp, VE_RENAME_DEST_DIR, NULL, NULL, ct);
+	(void) VOP_VNEVENT(vp, VE_RENAME_DEST_DIR, nvp, name, ct);
 }
 
 void
@@ -2951,7 +2964,7 @@ vn_setpath(vnode_t *rootvp, struct vnode *startvp, struct vnode *vp,
 	 * the potential for deadlock.
 	 */
 	mutex_enter(&base->v_lock);
-	if (base->v_path == NULL) {
+	if (base->v_path == vn_vpath_empty) {
 		mutex_exit(&base->v_lock);
 		return;
 	}
@@ -2978,7 +2991,8 @@ vn_setpath(vnode_t *rootvp, struct vnode *startvp, struct vnode *vp,
 	rpath = kmem_alloc(rpathalloc, KM_SLEEP);
 
 	mutex_enter(&base->v_lock);
-	if (base->v_path == NULL || strlen(base->v_path) != rpathlen) {
+	if (base->v_path == vn_vpath_empty ||
+	    strlen(base->v_path) != rpathlen) {
 		mutex_exit(&base->v_lock);
 		kmem_free(rpath, rpathalloc);
 		return;
@@ -2992,7 +3006,7 @@ vn_setpath(vnode_t *rootvp, struct vnode *startvp, struct vnode *vp,
 	rpath[rpathlen + plen] = '\0';
 
 	mutex_enter(&vp->v_lock);
-	if (vp->v_path != NULL) {
+	if (vp->v_path != vn_vpath_empty) {
 		mutex_exit(&vp->v_lock);
 		kmem_free(rpath, rpathalloc);
 	} else {
@@ -3012,7 +3026,7 @@ vn_setpath_str(struct vnode *vp, const char *str, size_t len)
 	char *buf = kmem_alloc(len + 1, KM_SLEEP);
 
 	mutex_enter(&vp->v_lock);
-	if (vp->v_path != NULL) {
+	if (vp->v_path != vn_vpath_empty) {
 		mutex_exit(&vp->v_lock);
 		kmem_free(buf, len + 1);
 		return;
@@ -3036,10 +3050,10 @@ vn_renamepath(vnode_t *dvp, vnode_t *vp, const char *nm, size_t len)
 
 	mutex_enter(&vp->v_lock);
 	tmp = vp->v_path;
-	vp->v_path = NULL;
+	vp->v_path = vn_vpath_empty;
 	mutex_exit(&vp->v_lock);
 	vn_setpath(rootdir, dvp, vp, nm, len);
-	if (tmp != NULL)
+	if (tmp != vn_vpath_empty)
 		kmem_free(tmp, strlen(tmp) + 1);
 }
 
@@ -3054,7 +3068,7 @@ vn_copypath(struct vnode *src, struct vnode *dst)
 	int alloc;
 
 	mutex_enter(&src->v_lock);
-	if (src->v_path == NULL) {
+	if (src->v_path == vn_vpath_empty) {
 		mutex_exit(&src->v_lock);
 		return;
 	}
@@ -3064,7 +3078,7 @@ vn_copypath(struct vnode *src, struct vnode *dst)
 	mutex_exit(&src->v_lock);
 	buf = kmem_alloc(alloc, KM_SLEEP);
 	mutex_enter(&src->v_lock);
-	if (src->v_path == NULL || strlen(src->v_path) + 1 != alloc) {
+	if (src->v_path == vn_vpath_empty || strlen(src->v_path) + 1 != alloc) {
 		mutex_exit(&src->v_lock);
 		kmem_free(buf, alloc);
 		return;
@@ -3073,7 +3087,7 @@ vn_copypath(struct vnode *src, struct vnode *dst)
 	mutex_exit(&src->v_lock);
 
 	mutex_enter(&dst->v_lock);
-	if (dst->v_path != NULL) {
+	if (dst->v_path != vn_vpath_empty) {
 		mutex_exit(&dst->v_lock);
 		kmem_free(buf, alloc);
 		return;
@@ -3231,14 +3245,57 @@ fop_read(
 	cred_t *cr,
 	caller_context_t *ct)
 {
-	int	err;
 	ssize_t	resid_start = uiop->uio_resid;
+	zone_t	*zonep = curzone;
+	zone_vfs_kstat_t *zvp = zonep->zone_vfs_stats;
+
+	hrtime_t start, lat;
+	ssize_t len;
+	int err;
+
+	if (vp->v_type == VREG || vp->v_type == VDIR || vp->v_type == VBLK) {
+		start = gethrtime();
+
+		mutex_enter(&zonep->zone_vfs_lock);
+		kstat_runq_enter(&zonep->zone_vfs_rwstats);
+		mutex_exit(&zonep->zone_vfs_lock);
+	}
 
 	VOPXID_MAP_CR(vp, cr);
 
 	err = (*(vp)->v_op->vop_read)(vp, uiop, ioflag, cr, ct);
-	VOPSTATS_UPDATE_IO(vp, read,
-	    read_bytes, (resid_start - uiop->uio_resid));
+	len = resid_start - uiop->uio_resid;
+
+	VOPSTATS_UPDATE_IO(vp, read, read_bytes, len);
+
+	if (vp->v_type == VREG || vp->v_type == VDIR || vp->v_type == VBLK) {
+		mutex_enter(&zonep->zone_vfs_lock);
+		zonep->zone_vfs_rwstats.reads++;
+		zonep->zone_vfs_rwstats.nread += len;
+		kstat_runq_exit(&zonep->zone_vfs_rwstats);
+		mutex_exit(&zonep->zone_vfs_lock);
+
+		lat = gethrtime() - start;
+
+		if (lat >= VOP_LATENCY_10MS) {
+			if (lat < VOP_LATENCY_100MS)
+				atomic_inc_64(&zvp->zv_10ms_ops.value.ui64);
+			else if (lat < VOP_LATENCY_1S) {
+				atomic_inc_64(&zvp->zv_10ms_ops.value.ui64);
+				atomic_inc_64(&zvp->zv_100ms_ops.value.ui64);
+			} else if (lat < VOP_LATENCY_10S) {
+				atomic_inc_64(&zvp->zv_10ms_ops.value.ui64);
+				atomic_inc_64(&zvp->zv_100ms_ops.value.ui64);
+				atomic_inc_64(&zvp->zv_1s_ops.value.ui64);
+			} else {
+				atomic_inc_64(&zvp->zv_10ms_ops.value.ui64);
+				atomic_inc_64(&zvp->zv_100ms_ops.value.ui64);
+				atomic_inc_64(&zvp->zv_1s_ops.value.ui64);
+				atomic_inc_64(&zvp->zv_10s_ops.value.ui64);
+			}
+		}
+	}
+
 	return (err);
 }
 
@@ -3250,14 +3307,62 @@ fop_write(
 	cred_t *cr,
 	caller_context_t *ct)
 {
-	int	err;
 	ssize_t	resid_start = uiop->uio_resid;
+	zone_t	*zonep = curzone;
+	zone_vfs_kstat_t *zvp = zonep->zone_vfs_stats;
+
+	hrtime_t start, lat;
+	ssize_t len;
+	int	err;
+
+	/*
+	 * For the purposes of VFS kstat consumers, the "waitq" calculation is
+	 * repurposed as the active queue for VFS write operations.  There's no
+	 * actual wait queue for VFS operations.
+	 */
+	if (vp->v_type == VREG || vp->v_type == VDIR || vp->v_type == VBLK) {
+		start = gethrtime();
+
+		mutex_enter(&zonep->zone_vfs_lock);
+		kstat_waitq_enter(&zonep->zone_vfs_rwstats);
+		mutex_exit(&zonep->zone_vfs_lock);
+	}
 
 	VOPXID_MAP_CR(vp, cr);
 
 	err = (*(vp)->v_op->vop_write)(vp, uiop, ioflag, cr, ct);
-	VOPSTATS_UPDATE_IO(vp, write,
-	    write_bytes, (resid_start - uiop->uio_resid));
+	len = resid_start - uiop->uio_resid;
+
+	VOPSTATS_UPDATE_IO(vp, write, write_bytes, len);
+
+	if (vp->v_type == VREG || vp->v_type == VDIR || vp->v_type == VBLK) {
+		mutex_enter(&zonep->zone_vfs_lock);
+		zonep->zone_vfs_rwstats.writes++;
+		zonep->zone_vfs_rwstats.nwritten += len;
+		kstat_waitq_exit(&zonep->zone_vfs_rwstats);
+		mutex_exit(&zonep->zone_vfs_lock);
+
+		lat = gethrtime() - start;
+
+		if (lat >= VOP_LATENCY_10MS) {
+			if (lat < VOP_LATENCY_100MS)
+				atomic_inc_64(&zvp->zv_10ms_ops.value.ui64);
+			else if (lat < VOP_LATENCY_1S) {
+				atomic_inc_64(&zvp->zv_10ms_ops.value.ui64);
+				atomic_inc_64(&zvp->zv_100ms_ops.value.ui64);
+			} else if (lat < VOP_LATENCY_10S) {
+				atomic_inc_64(&zvp->zv_10ms_ops.value.ui64);
+				atomic_inc_64(&zvp->zv_100ms_ops.value.ui64);
+				atomic_inc_64(&zvp->zv_1s_ops.value.ui64);
+			} else {
+				atomic_inc_64(&zvp->zv_10ms_ops.value.ui64);
+				atomic_inc_64(&zvp->zv_100ms_ops.value.ui64);
+				atomic_inc_64(&zvp->zv_1s_ops.value.ui64);
+				atomic_inc_64(&zvp->zv_10s_ops.value.ui64);
+			}
+		}
+	}
+
 	return (err);
 }
 
@@ -3421,7 +3526,7 @@ fop_lookup(
 	}
 	if (ret == 0 && *vpp) {
 		VOPSTATS_UPDATE(*vpp, lookup);
-		if ((*vpp)->v_path == NULL) {
+		if ((*vpp)->v_path == vn_vpath_empty) {
 			vn_setpath(rootdir, dvp, *vpp, nm, strlen(nm));
 		}
 	}
@@ -3463,7 +3568,7 @@ fop_create(
 	    (dvp, name, vap, excl, mode, vpp, cr, flags, ct, vsecp);
 	if (ret == 0 && *vpp) {
 		VOPSTATS_UPDATE(*vpp, create);
-		if ((*vpp)->v_path == NULL) {
+		if ((*vpp)->v_path == vn_vpath_empty) {
 			vn_setpath(rootdir, dvp, *vpp, name, strlen(name));
 		}
 	}
@@ -3585,7 +3690,7 @@ fop_mkdir(
 	    (dvp, dirname, vap, vpp, cr, ct, flags, vsecp);
 	if (ret == 0 && *vpp) {
 		VOPSTATS_UPDATE(*vpp, mkdir);
-		if ((*vpp)->v_path == NULL) {
+		if ((*vpp)->v_path == vn_vpath_empty) {
 			vn_setpath(rootdir, dvp, *vpp, dirname,
 			    strlen(dirname));
 		}
diff --git a/usr/src/uts/common/fs/zfs/arc.c b/usr/src/uts/common/fs/zfs/arc.c
index 09d2e1dd8f..d2b584cf02 100644
--- a/usr/src/uts/common/fs/zfs/arc.c
+++ b/usr/src/uts/common/fs/zfs/arc.c
@@ -129,6 +129,7 @@
 #include <sys/vdev.h>
 #include <sys/vdev_impl.h>
 #include <sys/dsl_pool.h>
+#include <sys/zfs_zone.h>
 #include <sys/multilist.h>
 #ifdef _KERNEL
 #include <sys/vmsystm.h>
@@ -4277,6 +4278,14 @@ top:
 		rzio = zio_read(pio, spa, bp, buf->b_data, size,
 		    arc_read_done, buf, priority, zio_flags, zb);
 
+		/*
+		 * At this point, this read I/O has already missed in the ARC
+		 * and will be going through to the disk.  The I/O throttle
+		 * should delay this I/O if this zone is using more than its I/O
+		 * priority allows.
+		 */
+		zfs_zone_io_throttle(ZFS_ZONE_IOP_READ);
+
 		if (*arc_flags & ARC_FLAG_WAIT)
 			return (zio_wait(rzio));
 
diff --git a/usr/src/uts/common/fs/zfs/dbuf.c b/usr/src/uts/common/fs/zfs/dbuf.c
index 465dfd08b2..4b644f7479 100644
--- a/usr/src/uts/common/fs/zfs/dbuf.c
+++ b/usr/src/uts/common/fs/zfs/dbuf.c
@@ -617,8 +617,19 @@ dbuf_read_impl(dmu_buf_impl_t *db, zio_t *zio, uint32_t *flags)
 		arc_space_consume(DN_MAX_BONUSLEN, ARC_SPACE_OTHER);
 		if (bonuslen < DN_MAX_BONUSLEN)
 			bzero(db->db.db_data, DN_MAX_BONUSLEN);
-		if (bonuslen)
-			bcopy(DN_BONUS(dn->dn_phys), db->db.db_data, bonuslen);
+
+		if (bonuslen) {
+			/*
+			 * Absent byzantine on-disk corruption, we fully expect
+			 * our bonuslen to be no more than DN_MAX_BONUSLEN --
+			 * but we nonetheless explicitly clamp it on the bcopy()
+			 * to prevent any on-disk corruption from becoming
+			 * rampant in-kernel corruption.
+			 */
+			bcopy(DN_BONUS(dn->dn_phys), db->db.db_data,
+			    MIN(bonuslen, DN_MAX_BONUSLEN));
+		}
+
 		DB_DNODE_EXIT(db);
 		db->db_state = DB_CACHED;
 		mutex_exit(&db->db_mtx);
diff --git a/usr/src/uts/common/fs/zfs/dmu.c b/usr/src/uts/common/fs/zfs/dmu.c
index 43bcfee91c..bfe5cda5e4 100644
--- a/usr/src/uts/common/fs/zfs/dmu.c
+++ b/usr/src/uts/common/fs/zfs/dmu.c
@@ -1799,7 +1799,6 @@ dmu_write_policy(objset_t *os, dnode_t *dn, int level, int wp, zio_prop_t *zp)
 			if (!zio_checksum_table[checksum].ci_dedup)
 				dedup_verify = B_TRUE;
 		}
-
 		/*
 		 * Enable nopwrite if we have a cryptographically secure
 		 * checksum that has no known collisions (i.e. SHA-256)
diff --git a/usr/src/uts/common/fs/zfs/dmu_tx.c b/usr/src/uts/common/fs/zfs/dmu_tx.c
index 56ce2f0c27..c2b1dca1c0 100644
--- a/usr/src/uts/common/fs/zfs/dmu_tx.c
+++ b/usr/src/uts/common/fs/zfs/dmu_tx.c
@@ -38,11 +38,11 @@
 #include <sys/sa_impl.h>
 #include <sys/zfs_context.h>
 #include <sys/varargs.h>
+#include <sys/zfs_zone.h>
 
 typedef void (*dmu_tx_hold_func_t)(dmu_tx_t *tx, struct dnode *dn,
     uint64_t arg1, uint64_t arg2);
 
-
 dmu_tx_t *
 dmu_tx_create_dd(dsl_dir_t *dd)
 {
@@ -223,6 +223,8 @@ dmu_tx_count_write(dmu_tx_hold_t *txh, uint64_t off, uint64_t len)
 	if (len == 0)
 		return;
 
+	zfs_zone_io_throttle(ZFS_ZONE_IOP_LOGICAL_WRITE);
+
 	min_bs = SPA_MINBLOCKSHIFT;
 	max_bs = highbit64(txh->txh_tx->tx_objset->os_recordsize) - 1;
 	min_ibs = DN_MIN_INDBLKSHIFT;
diff --git a/usr/src/uts/common/fs/zfs/dsl_dir.c b/usr/src/uts/common/fs/zfs/dsl_dir.c
index 6a20ab3ca2..1222b8933a 100644
--- a/usr/src/uts/common/fs/zfs/dsl_dir.c
+++ b/usr/src/uts/common/fs/zfs/dsl_dir.c
@@ -42,6 +42,7 @@
 #include <sys/zio.h>
 #include <sys/arc.h>
 #include <sys/sunddi.h>
+#include <sys/zfs_zone.h>
 #include <sys/zfeature.h>
 #include <sys/policy.h>
 #include <sys/zfs_znode.h>
@@ -1266,7 +1267,7 @@ dsl_dir_tempreserve_space(dsl_dir_t *dd, uint64_t lsize, uint64_t asize,
 			 * locks are held.
 			 */
 			txg_delay(dd->dd_pool, tx->tx_txg,
-			    MSEC2NSEC(10), MSEC2NSEC(10));
+			    zfs_zone_txg_delay(), MSEC2NSEC(10));
 			err = SET_ERROR(ERESTART);
 		}
 	}
diff --git a/usr/src/uts/common/fs/zfs/dsl_pool.c b/usr/src/uts/common/fs/zfs/dsl_pool.c
index 8c5d820ede..8ad7c789d3 100644
--- a/usr/src/uts/common/fs/zfs/dsl_pool.c
+++ b/usr/src/uts/common/fs/zfs/dsl_pool.c
@@ -42,6 +42,7 @@
 #include <sys/zfs_znode.h>
 #include <sys/spa_impl.h>
 #include <sys/dsl_deadlist.h>
+#include <sys/zfs_zone.h>
 #include <sys/bptree.h>
 #include <sys/zfeature.h>
 #include <sys/zil_impl.h>
diff --git a/usr/src/uts/common/fs/zfs/metaslab.c b/usr/src/uts/common/fs/zfs/metaslab.c
index 9ba9fd3841..02cccae29d 100644
--- a/usr/src/uts/common/fs/zfs/metaslab.c
+++ b/usr/src/uts/common/fs/zfs/metaslab.c
@@ -63,6 +63,11 @@ uint64_t metaslab_gang_bang = SPA_MAXBLOCKSIZE + 1;	/* force gang blocks */
 int zfs_condense_pct = 200;
 
 /*
+ * Never condense any space map.  This is for debugging/recovery only.
+ */
+int zfs_condense_never = 0;
+
+/*
  * Condensing a metaslab is not guaranteed to actually reduce the amount of
  * space used on disk. In particular, a space map uses data in increments of
  * MAX(1 << ashift, space_map_blksize), so a metaslab might use the
@@ -1656,6 +1661,9 @@ metaslab_should_condense(metaslab_t *msp)
 	ASSERT(MUTEX_HELD(&msp->ms_lock));
 	ASSERT(msp->ms_loaded);
 
+	if (zfs_condense_never != 0)
+		return (B_FALSE);
+
 	/*
 	 * Use the ms_size_tree range tree, which is ordered by size, to
 	 * obtain the largest segment in the free tree. We always condense
diff --git a/usr/src/uts/common/fs/zfs/sa.c b/usr/src/uts/common/fs/zfs/sa.c
index 0b5b37f5fb..5a2602271b 100644
--- a/usr/src/uts/common/fs/zfs/sa.c
+++ b/usr/src/uts/common/fs/zfs/sa.c
@@ -24,6 +24,7 @@
  * Portions Copyright 2011 iXsystems, Inc
  * Copyright (c) 2013 by Delphix. All rights reserved.
  * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
+ * Copyright (c) 2015 Joyent, Inc.  All rights reserved.
  */
 
 #include <sys/zfs_context.h>
@@ -406,15 +407,18 @@ sa_add_layout_entry(objset_t *os, sa_attr_type_t *attrs, int attr_count,
 {
 	sa_os_t *sa = os->os_sa;
 	sa_lot_t *tb, *findtb;
-	int i;
+	int i, size;
 	avl_index_t loc;
 
 	ASSERT(MUTEX_HELD(&sa->sa_lock));
 	tb = kmem_zalloc(sizeof (sa_lot_t), KM_SLEEP);
 	tb->lot_attr_count = attr_count;
-	tb->lot_attrs = kmem_alloc(sizeof (sa_attr_type_t) * attr_count,
-	    KM_SLEEP);
-	bcopy(attrs, tb->lot_attrs, sizeof (sa_attr_type_t) * attr_count);
+
+	if ((size = sizeof (sa_attr_type_t) * attr_count) != 0) {
+		tb->lot_attrs = kmem_alloc(size, KM_SLEEP);
+		bcopy(attrs, tb->lot_attrs, size);
+	}
+
 	tb->lot_num = lot_num;
 	tb->lot_hash = hash;
 	tb->lot_instance = 0;
diff --git a/usr/src/uts/common/fs/zfs/sys/vdev_impl.h b/usr/src/uts/common/fs/zfs/sys/vdev_impl.h
index 17a18a3199..bde9b0a7ab 100644
--- a/usr/src/uts/common/fs/zfs/sys/vdev_impl.h
+++ b/usr/src/uts/common/fs/zfs/sys/vdev_impl.h
@@ -116,6 +116,7 @@ struct vdev_queue {
 	avl_tree_t	vq_read_offset_tree;
 	avl_tree_t	vq_write_offset_tree;
 	uint64_t	vq_last_offset;
+	zoneid_t	vq_last_zone_id;
 	hrtime_t	vq_io_complete_ts; /* time last i/o completed */
 	kmutex_t	vq_lock;
 };
diff --git a/usr/src/uts/common/fs/zfs/sys/zfs_zone.h b/usr/src/uts/common/fs/zfs/sys/zfs_zone.h
new file mode 100644
index 0000000000..f1431b3f55
--- /dev/null
+++ b/usr/src/uts/common/fs/zfs/sys/zfs_zone.h
@@ -0,0 +1,63 @@
+/*
+ * CDDL HEADER START
+ *
+ * The contents of this file are subject to the terms of the
+ * Common Development and Distribution License (the "License").
+ * You may not use this file except in compliance with the License.
+ *
+ * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
+ * or http://www.opensolaris.org/os/licensing.
+ * See the License for the specific language governing permissions
+ * and limitations under the License.
+ *
+ * When distributing Covered Code, include this CDDL HEADER in each
+ * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
+ * If applicable, add the following below this CDDL HEADER, with the
+ * fields enclosed by brackets "[]" replaced with your own identifying
+ * information: Portions Copyright [yyyy] [name of copyright owner]
+ *
+ * CDDL HEADER END
+ */
+/*
+ * Copyright 2015, Joyent, Inc. All rights reserved.
+ */
+
+#ifndef	_SYS_FS_ZFS_ZONE_H
+#define	_SYS_FS_ZFS_ZONE_H
+
+#ifdef _KERNEL
+#include <sys/isa_defs.h>
+#include <sys/types32.h>
+#include <sys/vdev_impl.h>
+#include <sys/zio.h>
+#endif
+
+#ifdef	__cplusplus
+extern "C" {
+#endif
+
+typedef enum {
+	ZFS_ZONE_IOP_READ = 0,
+	ZFS_ZONE_IOP_WRITE,
+	ZFS_ZONE_IOP_LOGICAL_WRITE,
+} zfs_zone_iop_type_t;
+
+extern void zfs_zone_io_throttle(zfs_zone_iop_type_t);
+
+extern void zfs_zone_zio_init(zio_t *);
+extern void zfs_zone_zio_start(zio_t *);
+extern void zfs_zone_zio_done(zio_t *);
+extern void zfs_zone_zio_dequeue(zio_t *);
+extern void zfs_zone_zio_enqueue(zio_t *);
+extern void zfs_zone_report_txg_sync(void *);
+extern hrtime_t zfs_zone_txg_delay();
+#ifdef _KERNEL
+extern zio_t *zfs_zone_schedule(vdev_queue_t *, zio_priority_t, avl_index_t,
+    avl_tree_t *);
+#endif
+
+#ifdef	__cplusplus
+}
+#endif
+
+#endif	/* _SYS_FS_ZFS_ZONE_H */
diff --git a/usr/src/uts/common/fs/zfs/sys/zio.h b/usr/src/uts/common/fs/zfs/sys/zio.h
index 198dc92387..78c9355438 100644
--- a/usr/src/uts/common/fs/zfs/sys/zio.h
+++ b/usr/src/uts/common/fs/zfs/sys/zio.h
@@ -421,7 +421,8 @@ struct zio {
 	const zio_vsd_ops_t *io_vsd_ops;
 
 	uint64_t	io_offset;
-	hrtime_t	io_timestamp;
+	hrtime_t	io_timestamp;	/* time I/O entered zio pipeline */
+	hrtime_t	io_dispatched;	/* time I/O was dispatched to disk */
 	avl_node_t	io_queue_node;
 	avl_node_t	io_offset_node;
 
@@ -450,6 +451,7 @@ struct zio {
 	zio_cksum_report_t *io_cksum_report;
 	uint64_t	io_ena;
 
+	zoneid_t	io_zoneid;	/* zone which originated this I/O */
 	/* Taskq dispatching state */
 	taskq_ent_t	io_tqent;
 };
diff --git a/usr/src/uts/common/fs/zfs/txg.c b/usr/src/uts/common/fs/zfs/txg.c
index 191259e75b..915c9bb4b2 100644
--- a/usr/src/uts/common/fs/zfs/txg.c
+++ b/usr/src/uts/common/fs/zfs/txg.c
@@ -31,6 +31,7 @@
 #include <sys/dsl_pool.h>
 #include <sys/dsl_scan.h>
 #include <sys/callb.h>
+#include <sys/zfs_zone.h>
 
 /*
  * ZFS Transaction Groups
@@ -506,6 +507,8 @@ txg_sync_thread(dsl_pool_t *dp)
 		    txg, tx->tx_quiesce_txg_waiting, tx->tx_sync_txg_waiting);
 		mutex_exit(&tx->tx_sync_lock);
 
+		zfs_zone_report_txg_sync(dp);
+
 		start = ddi_get_lbolt();
 		spa_sync(spa, txg);
 		delta = ddi_get_lbolt() - start;
diff --git a/usr/src/uts/common/fs/zfs/vdev_disk.c b/usr/src/uts/common/fs/zfs/vdev_disk.c
index ed4a8b773b..e4cc42452a 100644
--- a/usr/src/uts/common/fs/zfs/vdev_disk.c
+++ b/usr/src/uts/common/fs/zfs/vdev_disk.c
@@ -21,11 +21,13 @@
 /*
  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
  * Copyright (c) 2012, 2014 by Delphix. All rights reserved.
+ * Copyright (c) 2013, Joyent, Inc. All rights reserved.
  * Copyright 2013 Nexenta Systems, Inc.  All rights reserved.
  * Copyright (c) 2013 Joyent, Inc.  All rights reserved.
  */
 
 #include <sys/zfs_context.h>
+#include <sys/zfs_zone.h>
 #include <sys/spa_impl.h>
 #include <sys/refcount.h>
 #include <sys/vdev_disk.h>
@@ -44,6 +46,11 @@ extern ldi_ident_t zfs_li;
 
 static void vdev_disk_close(vdev_t *);
 
+typedef struct vdev_disk_buf {
+	buf_t	vdb_buf;
+	zio_t	*vdb_io;
+} vdev_disk_buf_t;
+
 typedef struct vdev_disk_ldi_cb {
 	list_node_t		lcb_next;
 	ldi_callback_id_t	lcb_id;
@@ -127,6 +134,8 @@ vdev_disk_off_finalize(ldi_handle_t lh, ldi_ev_cookie_t ecookie,
     int ldi_result, void *arg, void *ev_data)
 {
 	vdev_t *vd = (vdev_t *)arg;
+	vdev_disk_t *dvd = vd->vdev_tsd;
+	vdev_disk_ldi_cb_t *lcb;
 
 	/*
 	 * Ignore events other than offline.
@@ -586,6 +595,7 @@ static void
 vdev_disk_close(vdev_t *vd)
 {
 	vdev_disk_t *dvd = vd->vdev_tsd;
+	vdev_disk_ldi_cb_t *lcb;
 
 	if (vd->vdev_reopening || dvd == NULL)
 		return;
@@ -812,6 +822,8 @@ vdev_disk_io_start(zio_t *zio)
 	bp->b_bufsize = zio->io_size;
 	bp->b_iodone = (int (*)())vdev_disk_io_intr;
 
+	zfs_zone_zio_start(zio);
+
 	/* ldi_strategy() will return non-zero only on programming errors */
 	VERIFY(ldi_strategy(dvd->vd_lh, bp) == 0);
 }
@@ -821,6 +833,8 @@ vdev_disk_io_done(zio_t *zio)
 {
 	vdev_t *vd = zio->io_vd;
 
+	zfs_zone_zio_done(zio);
+
 	/*
 	 * If the device returned EIO, then attempt a DKIOCSTATE ioctl to see if
 	 * the device has been removed.  If this is the case, then we trigger an
diff --git a/usr/src/uts/common/fs/zfs/vdev_queue.c b/usr/src/uts/common/fs/zfs/vdev_queue.c
index 79d6e13b3b..4c02013405 100644
--- a/usr/src/uts/common/fs/zfs/vdev_queue.c
+++ b/usr/src/uts/common/fs/zfs/vdev_queue.c
@@ -21,6 +21,7 @@
 /*
  * Copyright 2009 Sun Microsystems, Inc.  All rights reserved.
  * Use is subject to license terms.
+ * Copyright 2013, Joyent, Inc. All rights reserved.
  */
 
 /*
@@ -33,6 +34,7 @@
 #include <sys/zio.h>
 #include <sys/avl.h>
 #include <sys/dsl_pool.h>
+#include <sys/zfs_zone.h>
 
 /*
  * ZFS I/O Scheduler
@@ -141,7 +143,7 @@ uint32_t zfs_vdev_sync_write_min_active = 10;
 uint32_t zfs_vdev_sync_write_max_active = 10;
 uint32_t zfs_vdev_async_read_min_active = 1;
 uint32_t zfs_vdev_async_read_max_active = 3;
-uint32_t zfs_vdev_async_write_min_active = 1;
+uint32_t zfs_vdev_async_write_min_active = 3;
 uint32_t zfs_vdev_async_write_max_active = 10;
 uint32_t zfs_vdev_scrub_min_active = 1;
 uint32_t zfs_vdev_scrub_max_active = 2;
@@ -237,6 +239,8 @@ vdev_queue_init(vdev_t *vd)
 	    vdev_queue_offset_compare, sizeof (zio_t),
 	    offsetof(struct zio, io_offset_node));
 
+	vq->vq_last_zone_id = 0;
+
 	for (zio_priority_t p = 0; p < ZIO_PRIORITY_NUM_QUEUEABLE; p++) {
 		int (*compfn) (const void *, const void *);
 
@@ -274,6 +278,7 @@ vdev_queue_io_add(vdev_queue_t *vq, zio_t *zio)
 {
 	spa_t *spa = zio->io_spa;
 	ASSERT3U(zio->io_priority, <, ZIO_PRIORITY_NUM_QUEUEABLE);
+	zfs_zone_zio_enqueue(zio);
 	avl_add(vdev_queue_class_tree(vq, zio->io_priority), zio);
 	avl_add(vdev_queue_type_tree(vq, zio->io_type), zio);
 
@@ -289,6 +294,7 @@ vdev_queue_io_remove(vdev_queue_t *vq, zio_t *zio)
 {
 	spa_t *spa = zio->io_spa;
 	ASSERT3U(zio->io_priority, <, ZIO_PRIORITY_NUM_QUEUEABLE);
+	zfs_zone_zio_dequeue(zio);
 	avl_remove(vdev_queue_class_tree(vq, zio->io_priority), zio);
 	avl_remove(vdev_queue_type_tree(vq, zio->io_type), zio);
 
@@ -648,7 +654,11 @@ again:
 	search.io_timestamp = 0;
 	search.io_offset = vq->vq_last_offset + 1;
 	VERIFY3P(avl_find(tree, &search, &idx), ==, NULL);
+#ifdef _KERNEL
+	zio = zfs_zone_schedule(vq, p, idx, tree);
+#else
 	zio = avl_nearest(tree, idx, AVL_AFTER);
+#endif
 	if (zio == NULL)
 		zio = avl_first(tree);
 	ASSERT3U(zio->io_priority, ==, p);
diff --git a/usr/src/uts/common/fs/zfs/zfs_dir.c b/usr/src/uts/common/fs/zfs/zfs_dir.c
index bd7424b55b..2adb297937 100644
--- a/usr/src/uts/common/fs/zfs/zfs_dir.c
+++ b/usr/src/uts/common/fs/zfs/zfs_dir.c
@@ -21,6 +21,7 @@
 /*
  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
  * Copyright (c) 2013, 2014 by Delphix. All rights reserved.
+ * Copyright (c) 2015, Joyent, Inc.
  */
 
 #include <sys/types.h>
@@ -847,9 +848,9 @@ zfs_link_destroy(zfs_dirlock_t *dl, znode_t *zp, dmu_tx_t *tx, int flag,
 		if (zp->z_links <= zp_is_dir) {
 			zfs_panic_recover("zfs: link count on %s is %u, "
 			    "should be at least %u",
-			    zp->z_vnode->v_path ? zp->z_vnode->v_path :
-			    "<unknown>", (int)zp->z_links,
-			    zp_is_dir + 1);
+			    zp->z_vnode->v_path != vn_vpath_empty ?
+			    zp->z_vnode->v_path : "<unknown>",
+			    (int)zp->z_links, zp_is_dir + 1);
 			zp->z_links = zp_is_dir + 1;
 		}
 		if (--zp->z_links == zp_is_dir) {
diff --git a/usr/src/uts/common/fs/zfs/zfs_ioctl.c b/usr/src/uts/common/fs/zfs/zfs_ioctl.c
index 692b49611d..47b0b6f650 100644
--- a/usr/src/uts/common/fs/zfs/zfs_ioctl.c
+++ b/usr/src/uts/common/fs/zfs/zfs_ioctl.c
@@ -22,8 +22,8 @@
 /*
  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
  * Portions Copyright 2011 Martin Matuska
- * Copyright 2015 Nexenta Systems, Inc.  All rights reserved.
- * Copyright (c) 2014, Joyent, Inc. All rights reserved.
+ * Copyright 2011 Nexenta Systems, Inc.  All rights reserved.
+ * Copyright (c) 2015, Joyent, Inc. All rights reserved.
  * Copyright (c) 2011, 2014 by Delphix. All rights reserved.
  * Copyright (c) 2013 by Saso Kiselkov. All rights reserved.
  * Copyright (c) 2013 Steven Hartland. All rights reserved.
@@ -611,9 +611,10 @@ zfs_secpolicy_setprop(const char *dsname, zfs_prop_t prop, nvpair_t *propval,
 	 * Check permissions for special properties.
 	 */
 	switch (prop) {
+	case ZFS_PROP_DEDUP:
 	case ZFS_PROP_ZONED:
 		/*
-		 * Disallow setting of 'zoned' from within a local zone.
+		 * Disallow setting these properties from within a local zone.
 		 */
 		if (!INGLOBALZONE(curproc))
 			return (SET_ERROR(EPERM));
@@ -943,6 +944,9 @@ zfs_secpolicy_recv(zfs_cmd_t *zc, nvlist_t *innvl, cred_t *cr)
 {
 	int error;
 
+	if (secpolicy_fs_import(cr) != 0)
+		return (set_errno(EPERM));
+
 	if ((error = zfs_secpolicy_write_perms(zc->zc_name,
 	    ZFS_DELEG_PERM_RECEIVE, cr)) != 0)
 		return (error);
@@ -2034,7 +2038,8 @@ zfs_ioc_vdev_setfru(zfs_cmd_t *zc)
 }
 
 static int
-zfs_ioc_objset_stats_impl(zfs_cmd_t *zc, objset_t *os)
+zfs_ioc_objset_stats_impl(zfs_cmd_t *zc, objset_t *os,
+    boolean_t cachedpropsonly)
 {
 	int error = 0;
 	nvlist_t *nv;
@@ -2052,7 +2057,8 @@ zfs_ioc_objset_stats_impl(zfs_cmd_t *zc, objset_t *os)
 		 * XXX reading with out owning
 		 */
 		if (!zc->zc_objset_stats.dds_inconsistent &&
-		    dmu_objset_type(os) == DMU_OST_ZVOL) {
+		    dmu_objset_type(os) == DMU_OST_ZVOL &&
+		    !cachedpropsonly) {
 			error = zvol_get_stats(os, nv);
 			if (error == EIO)
 				return (error);
@@ -2079,11 +2085,24 @@ static int
 zfs_ioc_objset_stats(zfs_cmd_t *zc)
 {
 	objset_t *os;
+	nvlist_t *nvl = NULL;
+	boolean_t cachedpropsonly = B_FALSE;
 	int error;
 
+	if (zc->zc_nvlist_src != NULL &&
+	    (error = get_nvlist(zc->zc_nvlist_src, zc->zc_nvlist_src_size,
+	    zc->zc_iflags, &nvl) != 0))
+		return (error);
+
+	if (nvl != NULL) {
+		(void) nvlist_lookup_boolean_value(nvl, "cachedpropsonly",
+		    &cachedpropsonly);
+		nvlist_free(nvl);
+	}
+
 	error = dmu_objset_hold(zc->zc_name, FTAG, &os);
 	if (error == 0) {
-		error = zfs_ioc_objset_stats_impl(zc, os);
+		error = zfs_ioc_objset_stats_impl(zc, os, cachedpropsonly);
 		dmu_objset_rele(os, FTAG);
 	}
 
@@ -2277,8 +2296,21 @@ static int
 zfs_ioc_snapshot_list_next(zfs_cmd_t *zc)
 {
 	objset_t *os;
+	nvlist_t *nvl = NULL;
+	boolean_t cachedpropsonly = B_FALSE;
 	int error;
 
+	if (zc->zc_nvlist_src != NULL &&
+	    (error = get_nvlist(zc->zc_nvlist_src, zc->zc_nvlist_src_size,
+	    zc->zc_iflags, &nvl) != 0))
+		return (error);
+
+	if (nvl != NULL) {
+		(void) nvlist_lookup_boolean_value(nvl, "cachedpropsonly",
+		    &cachedpropsonly);
+		nvlist_free(nvl);
+	}
+
 	error = dmu_objset_hold(zc->zc_name, FTAG, &os);
 	if (error != 0) {
 		return (error == ENOENT ? ESRCH : error);
@@ -2307,8 +2339,10 @@ zfs_ioc_snapshot_list_next(zfs_cmd_t *zc)
 			objset_t *ossnap;
 
 			error = dmu_objset_from_ds(ds, &ossnap);
-			if (error == 0)
-				error = zfs_ioc_objset_stats_impl(zc, ossnap);
+			if (error == 0) {
+				error = zfs_ioc_objset_stats_impl(zc,
+				    ossnap, cachedpropsonly);
+			}
 			dsl_dataset_rele(ds, FTAG);
 		}
 	} else if (error == ENOENT) {
@@ -3018,6 +3052,7 @@ zfs_fill_zplprops_impl(objset_t *os, uint64_t zplver,
 	uint64_t sense = ZFS_PROP_UNDEFINED;
 	uint64_t norm = ZFS_PROP_UNDEFINED;
 	uint64_t u8 = ZFS_PROP_UNDEFINED;
+	int error;
 
 	ASSERT(zplprops != NULL);
 
@@ -3061,8 +3096,9 @@ zfs_fill_zplprops_impl(objset_t *os, uint64_t zplver,
 	VERIFY(nvlist_add_uint64(zplprops,
 	    zfs_prop_to_name(ZFS_PROP_VERSION), zplver) == 0);
 
-	if (norm == ZFS_PROP_UNDEFINED)
-		VERIFY(zfs_get_zplprop(os, ZFS_PROP_NORMALIZE, &norm) == 0);
+	if (norm == ZFS_PROP_UNDEFINED &&
+	    (error = zfs_get_zplprop(os, ZFS_PROP_NORMALIZE, &norm)) != 0)
+		return (error);
 	VERIFY(nvlist_add_uint64(zplprops,
 	    zfs_prop_to_name(ZFS_PROP_NORMALIZE), norm) == 0);
 
@@ -3071,13 +3107,15 @@ zfs_fill_zplprops_impl(objset_t *os, uint64_t zplver,
 	 */
 	if (norm)
 		u8 = 1;
-	if (u8 == ZFS_PROP_UNDEFINED)
-		VERIFY(zfs_get_zplprop(os, ZFS_PROP_UTF8ONLY, &u8) == 0);
+	if (u8 == ZFS_PROP_UNDEFINED &&
+	    (error = zfs_get_zplprop(os, ZFS_PROP_UTF8ONLY, &u8)) != 0)
+		return (error);
 	VERIFY(nvlist_add_uint64(zplprops,
 	    zfs_prop_to_name(ZFS_PROP_UTF8ONLY), u8) == 0);
 
-	if (sense == ZFS_PROP_UNDEFINED)
-		VERIFY(zfs_get_zplprop(os, ZFS_PROP_CASE, &sense) == 0);
+	if (sense == ZFS_PROP_UNDEFINED &&
+	    (error = zfs_get_zplprop(os, ZFS_PROP_CASE, &sense)) != 0)
+		return (error);
 	VERIFY(nvlist_add_uint64(zplprops,
 	    zfs_prop_to_name(ZFS_PROP_CASE), sense) == 0);
 
diff --git a/usr/src/uts/common/fs/zfs/zfs_vfsops.c b/usr/src/uts/common/fs/zfs/zfs_vfsops.c
index 8cf83b399a..f9b7986c56 100644
--- a/usr/src/uts/common/fs/zfs/zfs_vfsops.c
+++ b/usr/src/uts/common/fs/zfs/zfs_vfsops.c
@@ -21,6 +21,7 @@
 /*
  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
  * Copyright (c) 2013 by Delphix. All rights reserved.
+ * Copyright (c) 2012, Joyent, Inc. All rights reserved.
  */
 
 /* Portions Copyright 2010 Robert Milkowski */
@@ -1950,6 +1951,17 @@ zfs_umount(vfs_t *vfsp, int fflag, cred_t *cr)
 	if (zfsvfs->z_ctldir != NULL)
 		zfsctl_destroy(zfsvfs);
 
+	/*
+	 * If we're doing a forced unmount on a dataset which still has
+	 * references and is in a zone, then we need to cleanup the zone
+	 * reference at this point or else the zone will never be able to
+	 * shutdown.
+	 */
+	if ((fflag & MS_FORCE) && vfsp->vfs_count > 1 && vfsp->vfs_zone) {
+		zone_rele_ref(&vfsp->vfs_implp->vi_zone_ref, ZONE_REF_VFS);
+		vfsp->vfs_zone = NULL;
+	}
+
 	return (0);
 }
 
diff --git a/usr/src/uts/common/fs/zfs/zfs_vnops.c b/usr/src/uts/common/fs/zfs/zfs_vnops.c
index bf75097d2a..ddec59ffc9 100644
--- a/usr/src/uts/common/fs/zfs/zfs_vnops.c
+++ b/usr/src/uts/common/fs/zfs/zfs_vnops.c
@@ -22,11 +22,16 @@
  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
  * Copyright (c) 2012, 2014 by Delphix. All rights reserved.
  * Copyright 2014 Nexenta Systems, Inc.  All rights reserved.
+ * Copyright 2015 Joyent, Inc.  All rights reserved.
  */
 
 /* Portions Copyright 2007 Jeremy Teo */
 /* Portions Copyright 2010 Robert Milkowski */
 
+/*
+ * Copyright (c) 2013, Joyent, Inc. All rights reserved.
+ */
+
 #include <sys/types.h>
 #include <sys/param.h>
 #include <sys/time.h>
@@ -662,6 +667,7 @@ zfs_write(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct)
 	rl_t		*rl;
 	int		max_blksz = zfsvfs->z_max_blksz;
 	int		error = 0;
+	int		prev_error;
 	arc_buf_t	*abuf;
 	iovec_t		*aiov = NULL;
 	xuio_t		*xuio = NULL;
@@ -683,6 +689,17 @@ zfs_write(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct)
 	if (limit == RLIM64_INFINITY || limit > MAXOFFSET_T)
 		limit = MAXOFFSET_T;
 
+	/*
+	 * Pre-fault the pages to ensure slow (eg NFS) pages
+	 * don't hold up txg.
+	 * Skip this if uio contains loaned arc_buf.
+	 */
+	if ((uio->uio_extflg == UIO_XUIO) &&
+	    (((xuio_t *)uio)->xu_type == UIOTYPE_ZEROCOPY))
+		xuio = (xuio_t *)uio;
+	else
+		uio_prefaultpages(n, uio);
+
 	ZFS_ENTER(zfsvfs);
 	ZFS_VERIFY_ZP(zp);
 
@@ -735,17 +752,6 @@ zfs_write(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct)
 	}
 
 	/*
-	 * Pre-fault the pages to ensure slow (eg NFS) pages
-	 * don't hold up txg.
-	 * Skip this if uio contains loaned arc_buf.
-	 */
-	if ((uio->uio_extflg == UIO_XUIO) &&
-	    (((xuio_t *)uio)->xu_type == UIOTYPE_ZEROCOPY))
-		xuio = (xuio_t *)uio;
-	else
-		uio_prefaultpages(MIN(n, max_blksz), uio);
-
-	/*
 	 * If in append mode, set the io offset pointer to eof.
 	 */
 	if (ioflag & FAPPEND) {
@@ -966,7 +972,6 @@ zfs_write(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct)
 		while ((end_size = zp->z_size) < uio->uio_loffset) {
 			(void) atomic_cas_64(&zp->z_size, end_size,
 			    uio->uio_loffset);
-			ASSERT(error == 0);
 		}
 		/*
 		 * If we are replaying and eof is non zero then force
@@ -976,18 +981,20 @@ zfs_write(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct)
 		if (zfsvfs->z_replay && zfsvfs->z_replay_eof != 0)
 			zp->z_size = zfsvfs->z_replay_eof;
 
+		/*
+		 * Keep track of a possible pre-existing error from a partial
+		 * write via dmu_write_uio_dbuf above.
+		 */
+		prev_error = error;
 		error = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
 
 		zfs_log_write(zilog, tx, TX_WRITE, zp, woff, tx_bytes, ioflag);
 		dmu_tx_commit(tx);
 
-		if (error != 0)
+		if (prev_error != 0 || error != 0)
 			break;
 		ASSERT(tx_bytes == nbytes);
 		n -= nbytes;
-
-		if (!xuio && n > 0)
-			uio_prefaultpages(MIN(n, max_blksz), uio);
 	}
 
 	zfs_range_unlock(rl);
@@ -3659,18 +3666,6 @@ top:
 		}
 	}
 
-	vnevent_rename_src(ZTOV(szp), sdvp, snm, ct);
-	if (tzp)
-		vnevent_rename_dest(ZTOV(tzp), tdvp, tnm, ct);
-
-	/*
-	 * notify the target directory if it is not the same
-	 * as source directory.
-	 */
-	if (tdvp != sdvp) {
-		vnevent_rename_dest_dir(tdvp, ct);
-	}
-
 	tx = dmu_tx_create(zfsvfs->z_os);
 	dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE);
 	dmu_tx_hold_sa(tx, sdzp->z_sa_hdl, B_FALSE);
@@ -3711,8 +3706,12 @@ top:
 		return (error);
 	}
 
-	if (tzp)	/* Attempt to remove the existing target */
+	if (tzp) {
+		/* Attempt to remove the existing target */
 		error = zfs_link_destroy(tdl, tzp, tx, zflg, NULL);
+		if (error == 0)
+			vnevent_rename_dest(ZTOV(tzp), tdvp, tnm, ct);
+	}
 
 	if (error == 0) {
 		error = zfs_link_create(tdl, szp, tx, ZRENAMING);
@@ -3754,6 +3753,11 @@ top:
 	}
 
 	dmu_tx_commit(tx);
+
+	if (error == 0) {
+		vnevent_rename_src(ZTOV(szp), sdvp, snm, ct);
+		vnevent_rename_dest_dir(tdvp, ZTOV(szp), tnm, ct);
+	}
 out:
 	if (zl != NULL)
 		zfs_rename_unlock(&zl);
@@ -4244,6 +4248,8 @@ zfs_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp,
 		    &zp->z_pflags, 8);
 		zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime,
 		    B_TRUE);
+		err = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
+
 		zfs_log_write(zfsvfs->z_log, tx, TX_WRITE, zp, off, len, 0);
 	}
 	dmu_tx_commit(tx);
@@ -4779,10 +4785,6 @@ zfs_delmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr,
 	ASSERT3U(VTOZ(vp)->z_mapcnt, >=, pages);
 	atomic_add_64(&VTOZ(vp)->z_mapcnt, -pages);
 
-	if ((flags & MAP_SHARED) && (prot & PROT_WRITE) &&
-	    vn_has_cached_data(vp))
-		(void) VOP_PUTPAGE(vp, off, len, B_ASYNC, cr, ct);
-
 	return (0);
 }
 
diff --git a/usr/src/uts/common/fs/zfs/zfs_zone.c b/usr/src/uts/common/fs/zfs/zfs_zone.c
new file mode 100644
index 0000000000..4861c64f8e
--- /dev/null
+++ b/usr/src/uts/common/fs/zfs/zfs_zone.c
@@ -0,0 +1,1336 @@
+/*
+ * This file and its contents are supplied under the terms of the
+ * Common Development and Distribution License ("CDDL"), version 1.0.
+ * You may only use this file in accordance with the terms of version
+ * 1.0 of the CDDL.
+ *
+ * A full copy of the text of the CDDL should have accompanied this
+ * source.  A copy of the CDDL is also available via the Internet at
+ * http://www.illumos.org/license/CDDL.
+ */
+
+/*
+ * Copyright 2015, Joyent, Inc. All rights reserved.
+ */
+
+/*
+ * The ZFS/Zone I/O throttle and scheduler attempts to ensure fair access to
+ * ZFS I/O resources for each zone.
+ *
+ * I/O contention can be major pain point on a multi-tenant system. A single
+ * zone can issue a stream of I/O operations, usually synchronous writes, which
+ * disrupt I/O performance for all other zones. This problem is further
+ * exacerbated by ZFS, which buffers all asynchronous writes in a single TXG,
+ * a set of blocks which are atomically synced to disk. The process of
+ * syncing a TXG can occupy all of a device's I/O bandwidth, thereby starving
+ * out any pending read operations.
+ *
+ * There are two facets to this capability; the throttle and the scheduler.
+ *
+ * Throttle
+ *
+ * The requirements on the throttle are:
+ *
+ *     1) Ensure consistent and predictable I/O latency across all zones.
+ *     2) Sequential and random workloads have very different characteristics,
+ *        so it is a non-starter to track IOPS or throughput.
+ *     3) A zone should be able to use the full disk bandwidth if no other zone
+ *        is actively using the disk.
+ *
+ * The throttle has two components: one to track and account for each zone's
+ * I/O requests, and another to throttle each zone's operations when it
+ * exceeds its fair share of disk I/O. When the throttle detects that a zone is
+ * consuming more than is appropriate, each read or write system call is
+ * delayed by up to 100 microseconds, which we've found is sufficient to allow
+ * other zones to interleave I/O requests during those delays.
+ *
+ * Note: The throttle will delay each logical I/O (as opposed to the physical
+ * I/O which will likely be issued asynchronously), so it may be easier to
+ * think of the I/O throttle delaying each read/write syscall instead of the
+ * actual I/O operation. For each zone, the throttle tracks an ongoing average
+ * of read and write operations performed to determine the overall I/O
+ * utilization for each zone.
+ *
+ * The throttle calculates a I/O utilization metric for each zone using the
+ * following formula:
+ *
+ *     (# of read syscalls) x (Average read latency) +
+ *     (# of write syscalls) x (Average write latency)
+ *
+ * Once each zone has its utilization metric, the I/O throttle will compare I/O
+ * utilization across all zones, and if a zone has a higher-than-average I/O
+ * utilization, system calls from that zone are throttled. That is, if one
+ * zone has a much higher utilization, that zone's delay is increased by 5
+ * microseconds, up to a maximum of 100 microseconds. Conversely, if a zone is
+ * already throttled and has a lower utilization than average, its delay will
+ * be lowered by 5 microseconds.
+ *
+ * The throttle calculation is driven by IO activity, but since IO does not
+ * happen at fixed intervals, timestamps are used to track when the last update
+ * was made and to drive recalculation.
+ *
+ * The throttle recalculates each zone's I/O usage and throttle delay (if any)
+ * on the zfs_zone_adjust_time interval. Overall I/O latency is maintained as
+ * a decayed average which is updated on the zfs_zone_sys_avg_cycle interval.
+ *
+ * Scheduler
+ *
+ * The I/O scheduler manages the vdev queues – the queues of pending I/Os to
+ * issue to the disks. It only makes scheduling decisions for the two
+ * synchronous I/O queues (read & write).
+ *
+ * The scheduler maintains how many I/Os in the queue are from each zone, and
+ * if one zone has a disproportionately large number of I/Os in the queue, the
+ * scheduler will allow certain I/Os from the underutilized zones to be "bumped"
+ * and pulled from the middle of the queue. This bump allows zones with a small
+ * number of I/Os (so small they may not even be taken into account by the
+ * throttle) to complete quickly instead of waiting behind dozens of I/Os from
+ * other zones.
+ */
+
+#include <sys/spa.h>
+#include <sys/vdev_impl.h>
+#include <sys/zfs_zone.h>
+
+#ifndef _KERNEL
+
+/*
+ * Stubs for when compiling for user-land.
+ */
+
+void
+zfs_zone_io_throttle(zfs_zone_iop_type_t type)
+{
+}
+
+void
+zfs_zone_zio_init(zio_t *zp)
+{
+}
+
+void
+zfs_zone_zio_start(zio_t *zp)
+{
+}
+
+void
+zfs_zone_zio_done(zio_t *zp)
+{
+}
+
+void
+zfs_zone_zio_dequeue(zio_t *zp)
+{
+}
+
+void
+zfs_zone_zio_enqueue(zio_t *zp)
+{
+}
+
+/*ARGSUSED*/
+void
+zfs_zone_report_txg_sync(void *dp)
+{
+}
+
+hrtime_t
+zfs_zone_txg_delay()
+{
+	return (MSEC2NSEC(10));
+}
+
+#else
+
+/*
+ * The real code.
+ */
+
+#include <sys/systm.h>
+#include <sys/thread.h>
+#include <sys/proc.h>
+#include <sys/types.h>
+#include <sys/param.h>
+#include <sys/time.h>
+#include <sys/atomic.h>
+#include <sys/zio.h>
+#include <sys/zone.h>
+#include <sys/avl.h>
+#include <sys/sdt.h>
+#include <sys/ddi.h>
+
+/*
+ * The zone throttle delays read and write operations from certain zones based
+ * on each zone's IO utilitzation.  Once a cycle (defined by zfs_zone_cycle_time
+ * below), the delays for each zone are recalculated based on the utilization
+ * over the previous window.
+ */
+boolean_t	zfs_zone_delay_enable = B_TRUE;	/* enable IO throttle */
+uint16_t	zfs_zone_delay_step = 5;	/* usec amnt to change delay */
+uint16_t	zfs_zone_delay_ceiling = 100;	/* usec delay max */
+
+boolean_t	zfs_zone_priority_enable = B_TRUE;  /* enable IO priority */
+
+/*
+ * For certain workloads, one zone may be issuing primarily sequential I/O and
+ * another primarily random I/O.  The sequential I/O will complete much more
+ * quickly than the random I/O, driving the average system latency for those
+ * operations way down.  As a result, the random I/O may be throttled back, even
+ * though the sequential I/O should be throttled to allow the random I/O more
+ * access to the disk.
+ *
+ * This tunable limits the discrepancy between the read and write system
+ * latency.  If one becomes excessively high, this tunable prevents the I/O
+ * throttler from exacerbating the imbalance.
+ */
+uint_t		zfs_zone_rw_lat_limit = 10;
+
+/*
+ * The I/O throttle will only start delaying zones when it detects disk
+ * utilization has reached a certain level.  This tunable controls the
+ * threshold at which the throttle will start delaying zones.  When the number
+ * of vdevs is small, the calculation should correspond closely with the %b
+ * column from iostat -- but as the number of vdevs becomes large, it will
+ * correlate less and less to any single device (therefore making it a poor
+ * approximation for the actual I/O utilization on such systems).  We
+ * therefore use our derived utilization conservatively:  we know that low
+ * derived utilization does indeed correlate to low I/O use -- but that a high
+ * rate of derived utilization does not necesarily alone denote saturation;
+ * where we see a high rate of utilization, we also look for laggard I/Os to
+ * attempt to detect saturation.
+ */
+uint_t		zfs_zone_util_threshold = 80;
+uint_t		zfs_zone_underutil_threshold = 60;
+
+/*
+ * There are three important tunables here:  zfs_zone_laggard_threshold denotes
+ * the threshold at which an I/O is considered to be of notably high latency;
+ * zfs_zone_laggard_recent denotes the number of microseconds before the
+ * current time after which the last laggard is considered to be sufficiently
+ * recent to merit increasing the throttle; zfs_zone_laggard_ancient denotes
+ * the microseconds before the current time before which the last laggard is
+ * considered to be sufficiently old to merit decreasing the throttle.  The
+ * most important tunable of these three is the zfs_zone_laggard_threshold: in
+ * modeling data from a large public cloud, this tunable was found to have a
+ * much greater effect on the throttle than the two time-based thresholds.
+ * This must be set high enough to not result in spurious throttling, but not
+ * so high as to allow pathological I/O to persist in the system.
+ */
+uint_t		zfs_zone_laggard_threshold = 50000;	/* 50 ms */
+uint_t		zfs_zone_laggard_recent = 1000000;	/* 1000 ms */
+uint_t		zfs_zone_laggard_ancient = 5000000;	/* 5000 ms */
+
+/*
+ * Throughout this subsystem, our timestamps are in microseconds.  Our system
+ * average cycle is one second or 1 million microseconds.  Our zone counter
+ * update cycle is two seconds or 2 million microseconds.  We use a longer
+ * duration for that cycle because some ops can see a little over two seconds of
+ * latency when they are being starved by another zone.
+ */
+uint_t 		zfs_zone_sys_avg_cycle = 1000000;	/* 1 s */
+uint_t 		zfs_zone_cycle_time = 2000000;		/* 2 s */
+
+/*
+ * How often the I/O throttle will reevaluate each zone's utilization, in
+ * microseconds. Default is 1/4 sec.
+ */
+uint_t 		zfs_zone_adjust_time = 250000;		/* 250 ms */
+
+typedef struct {
+	hrtime_t	cycle_start;
+	int		cycle_cnt;
+	hrtime_t	cycle_lat;
+	hrtime_t	sys_avg_lat;
+} sys_lat_cycle_t;
+
+typedef struct {
+	hrtime_t zi_now;
+	uint_t zi_avgrlat;
+	uint_t zi_avgwlat;
+	uint64_t zi_totpri;
+	uint64_t zi_totutil;
+	int zi_active;
+	uint_t zi_diskutil;
+	boolean_t zi_underutil;
+	boolean_t zi_overutil;
+} zoneio_stats_t;
+
+static sys_lat_cycle_t	rd_lat;
+static sys_lat_cycle_t	wr_lat;
+
+/*
+ * Some basic disk stats to determine disk utilization. The utilization info
+ * for all disks on the system is aggregated into these values.
+ *
+ * Overall disk utilization for the current cycle is calculated as:
+ *
+ * ((zfs_disk_rtime - zfs_disk_last_rtime) * 100)
+ * ----------------------------------------------
+ *    ((now - zfs_zone_last_checked) * 1000);
+ */
+kmutex_t	zfs_disk_lock;		/* protects the following: */
+uint_t		zfs_disk_rcnt;		/* Number of outstanding IOs */
+hrtime_t	zfs_disk_rtime = 0; /* cummulative sum of time performing IO */
+hrtime_t	zfs_disk_rlastupdate = 0; /* time last IO dispatched */
+
+hrtime_t	zfs_disk_last_rtime = 0; /* prev. cycle's zfs_disk_rtime val */
+/* time that we last updated per-zone throttle info */
+hrtime_t	zfs_zone_last_checked = 0;
+hrtime_t	zfs_disk_last_laggard = 0;
+
+/*
+ * Data used to keep track of how often txg sync is running.
+ */
+extern int	zfs_txg_timeout;
+static uint_t	txg_last_check;
+static uint_t	txg_cnt;
+static uint_t	txg_sync_rate;
+
+boolean_t	zfs_zone_schedule_enable = B_TRUE;	/* enable IO sched. */
+/*
+ * Threshold for when zio scheduling should kick in.
+ *
+ * This threshold is based on the zfs_vdev_sync_read_max_active value for the
+ * number of I/Os that can be pending on a device.  If there are more than the
+ * max_active ops already queued up, beyond those already issued to the vdev,
+ * then use zone-based scheduling to get the next synchronous zio.
+ */
+uint32_t	zfs_zone_schedule_thresh = 10;
+
+/*
+ * On each pass of the scheduler we increment the zone's weight (up to this
+ * maximum). The weight is used by the scheduler to prevent starvation so
+ * that zones which haven't been able to do any IO over many iterations
+ * will max out thier weight to this value.
+ */
+#define	SCHED_WEIGHT_MAX	20
+
+/*
+ * Tunables for delay throttling when TXG sync is occurring.
+ *
+ * If the zone is performing a write and we're doing above normal TXG syncing,
+ * then throttle for longer than normal. The zone's wait time is multiplied
+ * by the scale (zfs_zone_txg_throttle_scale).
+ */
+int		zfs_zone_txg_throttle_scale = 2;
+hrtime_t	zfs_zone_txg_delay_nsec = MSEC2NSEC(20);
+
+typedef struct {
+	int		zq_qdepth;
+	zio_priority_t	zq_queue;
+	int		zq_priority;
+	int		zq_wt;
+	zoneid_t	zq_zoneid;
+} zone_q_bump_t;
+
+/*
+ * This uses gethrtime() but returns a value in usecs.
+ */
+#define	GET_USEC_TIME		(gethrtime() / 1000)
+#define	NANO_TO_MICRO(x)	(x / (NANOSEC / MICROSEC))
+
+/*
+ * Keep track of the zone's ZFS IOPs.
+ *
+ * See the comment on the zfs_zone_io_throttle function for which/how IOPs are
+ * accounted for.
+ *
+ * If the number of ops is >1 then we can just use that value.  However,
+ * if the number of ops is <2 then we might have a zone which is trying to do
+ * IO but is not able to get any ops through the system.  We don't want to lose
+ * track of this zone so we factor in its decayed count into the current count.
+ *
+ * Each cycle (zfs_zone_sys_avg_cycle) we want to update the decayed count.
+ * However, since this calculation is driven by IO activity and since IO does
+ * not happen at fixed intervals, we use a timestamp to see when the last update
+ * was made.  If it was more than one cycle ago, then we need to decay the
+ * historical count by the proper number of additional cycles in which no IO was
+ * performed.
+ *
+ * Return a time delta indicating how far into the current cycle we are or 0
+ * if the last IO was more than a cycle ago.
+ */
+static hrtime_t
+compute_historical_zone_cnt(hrtime_t unow, sys_zio_cntr_t *cp)
+{
+	hrtime_t delta;
+	int	gen_cnt;
+
+	/*
+	 * Check if its time to recompute a new zone count.
+	 * If we're still collecting data for the current cycle, return false.
+	 */
+	delta = unow - cp->cycle_start;
+	if (delta < zfs_zone_cycle_time)
+		return (delta);
+
+	/* A previous cycle is past, compute the new zone count. */
+
+	/*
+	 * Figure out how many generations we have to decay the historical
+	 * count, since multiple cycles may have elapsed since our last IO.
+	 * We depend on int rounding here.
+	 */
+	gen_cnt = (int)(delta / zfs_zone_cycle_time);
+
+	/* If more than 5 cycles since last the IO, reset count. */
+	if (gen_cnt > 5) {
+		cp->zone_avg_cnt = 0;
+	} else {
+		/* Update the count. */
+		int	i;
+
+		/*
+		 * If the zone did more than 1 IO, just use its current count
+		 * as the historical value, otherwise decay the historical
+		 * count and factor that into the new historical count.  We
+		 * pick a threshold > 1 so that we don't lose track of IO due
+		 * to int rounding.
+		 */
+		if (cp->cycle_cnt > 1)
+			cp->zone_avg_cnt = cp->cycle_cnt;
+		else
+			cp->zone_avg_cnt = cp->cycle_cnt +
+			    (cp->zone_avg_cnt / 2);
+
+		/*
+		 * If more than one generation has elapsed since the last
+		 * update, decay the values further.
+		 */
+		for (i = 1; i < gen_cnt; i++)
+			cp->zone_avg_cnt = cp->zone_avg_cnt / 2;
+	}
+
+	/* A new cycle begins. */
+	cp->cycle_start = unow;
+	cp->cycle_cnt = 0;
+
+	return (0);
+}
+
+/*
+ * Add IO op data to the zone.
+ */
+static void
+add_zone_iop(zone_t *zonep, hrtime_t unow, zfs_zone_iop_type_t op)
+{
+	switch (op) {
+	case ZFS_ZONE_IOP_READ:
+		(void) compute_historical_zone_cnt(unow, &zonep->zone_rd_ops);
+		zonep->zone_rd_ops.cycle_cnt++;
+		break;
+	case ZFS_ZONE_IOP_WRITE:
+		(void) compute_historical_zone_cnt(unow, &zonep->zone_wr_ops);
+		zonep->zone_wr_ops.cycle_cnt++;
+		break;
+	case ZFS_ZONE_IOP_LOGICAL_WRITE:
+		(void) compute_historical_zone_cnt(unow, &zonep->zone_lwr_ops);
+		zonep->zone_lwr_ops.cycle_cnt++;
+		break;
+	}
+}
+
+/*
+ * Use a decaying average to keep track of the overall system latency.
+ *
+ * We want to have the recent activity heavily weighted, but if the
+ * activity decreases or stops, then the average should quickly decay
+ * down to the new value.
+ *
+ * Each cycle (zfs_zone_sys_avg_cycle) we want to update the decayed average.
+ * However, since this calculation is driven by IO activity and since IO does
+ * not happen at fixed intervals, we use a timestamp to see when the last
+ * update was made. If it was more than one cycle ago, then we need to decay
+ * the average by the proper number of additional cycles in which no IO was
+ * performed.
+ *
+ * Return true if we actually computed a new system average.
+ * If we're still within an active cycle there is nothing to do, return false.
+ */
+static boolean_t
+compute_new_sys_avg(hrtime_t unow, sys_lat_cycle_t *cp)
+{
+	hrtime_t delta;
+	int	gen_cnt;
+
+	/*
+	 * Check if its time to recompute a new average.
+	 * If we're still collecting data for the current cycle, return false.
+	 */
+	delta = unow - cp->cycle_start;
+	if (delta < zfs_zone_sys_avg_cycle)
+		return (B_FALSE);
+
+	/* A previous cycle is past, compute a new system average. */
+
+	/*
+	 * Figure out how many generations we have to decay, since multiple
+	 * cycles may have elapsed since our last IO.
+	 * We count on int rounding here.
+	 */
+	gen_cnt = (int)(delta / zfs_zone_sys_avg_cycle);
+
+	/* If more than 5 cycles since last the IO, reset average. */
+	if (gen_cnt > 5) {
+		cp->sys_avg_lat = 0;
+	} else {
+		/* Update the average. */
+		int	i;
+
+		cp->sys_avg_lat =
+		    (cp->sys_avg_lat + cp->cycle_lat) / (1 + cp->cycle_cnt);
+
+		/*
+		 * If more than one generation has elapsed since the last
+		 * update, decay the values further.
+		 */
+		for (i = 1; i < gen_cnt; i++)
+			cp->sys_avg_lat = cp->sys_avg_lat / 2;
+	}
+
+	/* A new cycle begins. */
+	cp->cycle_start = unow;
+	cp->cycle_cnt = 0;
+	cp->cycle_lat = 0;
+
+	return (B_TRUE);
+}
+
+static void
+add_sys_iop(hrtime_t unow, int op, int lat)
+{
+	switch (op) {
+	case ZFS_ZONE_IOP_READ:
+		(void) compute_new_sys_avg(unow, &rd_lat);
+		rd_lat.cycle_cnt++;
+		rd_lat.cycle_lat += lat;
+		break;
+	case ZFS_ZONE_IOP_WRITE:
+		(void) compute_new_sys_avg(unow, &wr_lat);
+		wr_lat.cycle_cnt++;
+		wr_lat.cycle_lat += lat;
+		break;
+	}
+}
+
+/*
+ * Get the zone IO counts.
+ */
+static uint_t
+calc_zone_cnt(hrtime_t unow, sys_zio_cntr_t *cp)
+{
+	hrtime_t delta;
+	uint_t cnt;
+
+	if ((delta = compute_historical_zone_cnt(unow, cp)) == 0) {
+		/*
+		 * No activity in the current cycle, we already have the
+		 * historical data so we'll use that.
+		 */
+		cnt = cp->zone_avg_cnt;
+	} else {
+		/*
+		 * If we're less than half way through the cycle then use
+		 * the current count plus half the historical count, otherwise
+		 * just use the current count.
+		 */
+		if (delta < (zfs_zone_cycle_time / 2))
+			cnt = cp->cycle_cnt + (cp->zone_avg_cnt / 2);
+		else
+			cnt = cp->cycle_cnt;
+	}
+
+	return (cnt);
+}
+
+/*
+ * Get the average read/write latency in usecs for the system.
+ */
+static uint_t
+calc_avg_lat(hrtime_t unow, sys_lat_cycle_t *cp)
+{
+	if (compute_new_sys_avg(unow, cp)) {
+		/*
+		 * No activity in the current cycle, we already have the
+		 * historical data so we'll use that.
+		 */
+		return (cp->sys_avg_lat);
+	} else {
+		/*
+		 * We're within a cycle; weight the current activity higher
+		 * compared to the historical data and use that.
+		 */
+		DTRACE_PROBE3(zfs__zone__calc__wt__avg,
+		    uintptr_t, cp->sys_avg_lat,
+		    uintptr_t, cp->cycle_lat,
+		    uintptr_t, cp->cycle_cnt);
+
+		return ((cp->sys_avg_lat + (cp->cycle_lat * 8)) /
+		    (1 + (cp->cycle_cnt * 8)));
+	}
+}
+
+/*
+ * Account for the current IOP on the zone and for the system as a whole.
+ * The latency parameter is in usecs.
+ */
+static void
+add_iop(zone_t *zonep, hrtime_t unow, zfs_zone_iop_type_t op, hrtime_t lat)
+{
+	/* Add op to zone */
+	add_zone_iop(zonep, unow, op);
+
+	/* Track system latency */
+	if (op != ZFS_ZONE_IOP_LOGICAL_WRITE)
+		add_sys_iop(unow, op, lat);
+}
+
+/*
+ * Calculate and return the total number of read ops, write ops and logical
+ * write ops for the given zone.  If the zone has issued operations of any type
+ * return a non-zero value, otherwise return 0.
+ */
+static int
+get_zone_io_cnt(hrtime_t unow, zone_t *zonep, uint_t *rops, uint_t *wops,
+    uint_t *lwops)
+{
+	*rops = calc_zone_cnt(unow, &zonep->zone_rd_ops);
+	*wops = calc_zone_cnt(unow, &zonep->zone_wr_ops);
+	*lwops = calc_zone_cnt(unow, &zonep->zone_lwr_ops);
+
+	DTRACE_PROBE4(zfs__zone__io__cnt, uintptr_t, zonep->zone_id,
+	    uintptr_t, *rops, uintptr_t, *wops, uintptr_t, *lwops);
+
+	return (*rops | *wops | *lwops);
+}
+
+/*
+ * Get the average read/write latency in usecs for the system.
+ */
+static void
+get_sys_avg_lat(hrtime_t unow, uint_t *rlat, uint_t *wlat)
+{
+	*rlat = calc_avg_lat(unow, &rd_lat);
+	*wlat = calc_avg_lat(unow, &wr_lat);
+
+	/*
+	 * In an attempt to improve the accuracy of the throttling algorithm,
+	 * assume that IO operations can't have zero latency.  Instead, assume
+	 * a reasonable lower bound for each operation type. If the actual
+	 * observed latencies are non-zero, use those latency values instead.
+	 */
+	if (*rlat == 0)
+		*rlat = 1000;
+	if (*wlat == 0)
+		*wlat = 1000;
+
+	DTRACE_PROBE2(zfs__zone__sys__avg__lat, uintptr_t, *rlat,
+	    uintptr_t, *wlat);
+}
+
+/*
+ * Find disk utilization for each zone and average utilization for all active
+ * zones.
+ */
+static int
+zfs_zone_wait_adjust_calculate_cb(zone_t *zonep, void *arg)
+{
+	zoneio_stats_t *sp = arg;
+	uint_t rops, wops, lwops;
+
+	if (zonep->zone_id == GLOBAL_ZONEID ||
+	    get_zone_io_cnt(sp->zi_now, zonep, &rops, &wops, &lwops) == 0) {
+		zonep->zone_io_util = 0;
+		return (0);
+	}
+
+	zonep->zone_io_util = (rops * sp->zi_avgrlat) +
+	    (wops * sp->zi_avgwlat) + (lwops * sp->zi_avgwlat);
+	sp->zi_totutil += zonep->zone_io_util;
+
+	if (zonep->zone_io_util > 0) {
+		sp->zi_active++;
+		sp->zi_totpri += zonep->zone_zfs_io_pri;
+	}
+
+	/*
+	 * sdt:::zfs-zone-utilization
+	 *
+	 *	arg0: zone ID
+	 *	arg1: read operations observed during time window
+	 *	arg2: physical write operations observed during time window
+	 *	arg3: logical write ops observed during time window
+	 *	arg4: calculated utilization given read and write ops
+	 *	arg5: I/O priority assigned to this zone
+	 */
+	DTRACE_PROBE6(zfs__zone__utilization, uint_t, zonep->zone_id,
+	    uint_t, rops, uint_t, wops, uint_t, lwops,
+	    uint_t, zonep->zone_io_util, uint_t, zonep->zone_zfs_io_pri);
+
+	return (0);
+}
+
+static void
+zfs_zone_delay_inc(zone_t *zonep)
+{
+	if (zonep->zone_io_delay < zfs_zone_delay_ceiling)
+		zonep->zone_io_delay += zfs_zone_delay_step;
+}
+
+static void
+zfs_zone_delay_dec(zone_t *zonep)
+{
+	if (zonep->zone_io_delay > 0)
+		zonep->zone_io_delay -= zfs_zone_delay_step;
+}
+
+/*
+ * For all zones "far enough" away from the average utilization, increase that
+ * zones delay.  Otherwise, reduce its delay.
+ */
+static int
+zfs_zone_wait_adjust_delay_cb(zone_t *zonep, void *arg)
+{
+	zoneio_stats_t *sp = arg;
+	uint16_t delay = zonep->zone_io_delay;
+	uint_t fairutil = 0;
+
+	zonep->zone_io_util_above_avg = B_FALSE;
+
+	/*
+	 * Given the calculated total utilitzation for all zones, calculate the
+	 * fair share of I/O for this zone.
+	 */
+	if (zfs_zone_priority_enable && sp->zi_totpri > 0) {
+		fairutil = (sp->zi_totutil * zonep->zone_zfs_io_pri) /
+		    sp->zi_totpri;
+	} else if (sp->zi_active > 0) {
+		fairutil = sp->zi_totutil / sp->zi_active;
+	}
+
+	/*
+	 * Adjust each IO's delay.  If the overall delay becomes too high, avoid
+	 * increasing beyond the ceiling value.
+	 */
+	if (zonep->zone_io_util > fairutil && sp->zi_overutil) {
+		zonep->zone_io_util_above_avg = B_TRUE;
+
+		if (sp->zi_active > 1)
+			zfs_zone_delay_inc(zonep);
+	} else if (zonep->zone_io_util < fairutil || sp->zi_underutil ||
+	    sp->zi_active <= 1) {
+		zfs_zone_delay_dec(zonep);
+	}
+
+	/*
+	 * sdt:::zfs-zone-throttle
+	 *
+	 *	arg0: zone ID
+	 *	arg1: old delay for this zone
+	 *	arg2: new delay for this zone
+	 *	arg3: calculated fair I/O utilization
+	 *	arg4: actual I/O utilization
+	 */
+	DTRACE_PROBE5(zfs__zone__throttle, uintptr_t, zonep->zone_id,
+	    uintptr_t, delay, uintptr_t, zonep->zone_io_delay,
+	    uintptr_t, fairutil, uintptr_t, zonep->zone_io_util);
+
+	return (0);
+}
+
+/*
+ * Examine the utilization between different zones, and adjust the delay for
+ * each zone appropriately.
+ */
+static void
+zfs_zone_wait_adjust(hrtime_t unow, hrtime_t last_checked)
+{
+	zoneio_stats_t stats;
+	hrtime_t laggard_udelta = 0;
+
+	(void) bzero(&stats, sizeof (stats));
+
+	stats.zi_now = unow;
+	get_sys_avg_lat(unow, &stats.zi_avgrlat, &stats.zi_avgwlat);
+
+	if (stats.zi_avgrlat > stats.zi_avgwlat * zfs_zone_rw_lat_limit)
+		stats.zi_avgrlat = stats.zi_avgwlat * zfs_zone_rw_lat_limit;
+	else if (stats.zi_avgrlat * zfs_zone_rw_lat_limit < stats.zi_avgwlat)
+		stats.zi_avgwlat = stats.zi_avgrlat * zfs_zone_rw_lat_limit;
+
+	if (zone_walk(zfs_zone_wait_adjust_calculate_cb, &stats) != 0)
+		return;
+
+	/*
+	 * Calculate disk utilization for the most recent period.
+	 */
+	if (zfs_disk_last_rtime == 0 || unow - last_checked <= 0) {
+		stats.zi_diskutil = 0;
+	} else {
+		stats.zi_diskutil =
+		    ((zfs_disk_rtime - zfs_disk_last_rtime) * 100) /
+		    ((unow - last_checked) * 1000);
+	}
+	zfs_disk_last_rtime = zfs_disk_rtime;
+
+	if (unow > zfs_disk_last_laggard)
+		laggard_udelta = unow - zfs_disk_last_laggard;
+
+	/*
+	 * To minimize porpoising, we have three separate states for our
+	 * assessment of I/O performance:  overutilized, underutilized, and
+	 * neither overutilized nor underutilized.  We will increment the
+	 * throttle if a zone is using more than its fair share _and_ I/O
+	 * is overutilized; we will decrement the throttle if a zone is using
+	 * less than its fair share _or_ I/O is underutilized.
+	 */
+	stats.zi_underutil = stats.zi_diskutil < zfs_zone_underutil_threshold ||
+	    laggard_udelta > zfs_zone_laggard_ancient;
+
+	stats.zi_overutil = stats.zi_diskutil > zfs_zone_util_threshold &&
+	    laggard_udelta < zfs_zone_laggard_recent;
+
+	/*
+	 * sdt:::zfs-zone-stats
+	 *
+	 * Statistics observed over the last period:
+	 *
+	 *	arg0: average system read latency
+	 *	arg1: average system write latency
+	 *	arg2: number of active zones
+	 *	arg3: total I/O 'utilization' for all zones
+	 *	arg4: total I/O priority of all active zones
+	 *	arg5: calculated disk utilization
+	 */
+	DTRACE_PROBE6(zfs__zone__stats, uintptr_t, stats.zi_avgrlat,
+	    uintptr_t, stats.zi_avgwlat, uintptr_t, stats.zi_active,
+	    uintptr_t, stats.zi_totutil, uintptr_t, stats.zi_totpri,
+	    uintptr_t, stats.zi_diskutil);
+
+	(void) zone_walk(zfs_zone_wait_adjust_delay_cb, &stats);
+}
+
+/*
+ * Callback used to calculate a zone's IO schedule priority.
+ *
+ * We scan the zones looking for ones with ops in the queue.  Out of those,
+ * we pick the one that calculates to the highest schedule priority.
+ */
+static int
+get_sched_pri_cb(zone_t *zonep, void *arg)
+{
+	int pri;
+	uint_t cnt;
+	zone_q_bump_t *qbp = arg;
+	zio_priority_t p = qbp->zq_queue;
+
+	cnt = zonep->zone_zfs_queued[p];
+	if (cnt == 0) {
+		zonep->zone_zfs_weight = 0;
+		return (0);
+	}
+
+	/*
+	 * On each pass, increment the zone's weight.  We use this as input
+	 * to the calculation to prevent starvation.  The value is reset
+	 * each time we issue an IO for this zone so zones which haven't
+	 * done any IO over several iterations will see their weight max
+	 * out.
+	 */
+	if (zonep->zone_zfs_weight < SCHED_WEIGHT_MAX)
+		zonep->zone_zfs_weight++;
+
+	/*
+	 * This zone's IO priority is the inverse of the number of IOs
+	 * the zone has enqueued * zone's configured priority * weight.
+	 * The queue depth has already been scaled by 10 to avoid problems
+	 * with int rounding.
+	 *
+	 * This means that zones with fewer IOs in the queue will get
+	 * preference unless other zone's assigned priority pulls them
+	 * ahead.  The weight is factored in to help ensure that zones
+	 * which haven't done IO in a while aren't getting starved.
+	 */
+	pri = (qbp->zq_qdepth / cnt) *
+	    zonep->zone_zfs_io_pri * zonep->zone_zfs_weight;
+
+	/*
+	 * If this zone has a higher priority than what we found so far,
+	 * it becomes the new leading contender.
+	 */
+	if (pri > qbp->zq_priority) {
+		qbp->zq_zoneid = zonep->zone_id;
+		qbp->zq_priority = pri;
+		qbp->zq_wt = zonep->zone_zfs_weight;
+	}
+	return (0);
+}
+
+/*
+ * See if we need to bump a zone's zio to the head of the queue. This is only
+ * done on the two synchronous I/O queues (see the block comment on the
+ * zfs_zone_schedule function). We get the correct vdev_queue_class_t and
+ * queue depth from our caller.
+ *
+ * For single-threaded synchronous processes a zone cannot get more than
+ * 1 op into the queue at a time unless the zone is running multiple processes
+ * in parallel.  This can cause an imbalance in performance if there are zones
+ * with many parallel processes (and ops in the queue) vs. other zones which
+ * are doing simple single-threaded processes, such as interactive tasks in the
+ * shell.  These zones can get backed up behind a deep queue and their IO
+ * performance will appear to be very poor as a result.  This can make the
+ * zone work badly for interactive behavior.
+ *
+ * The scheduling algorithm kicks in once we start to get a deeper queue.
+ * Once that occurs, we look at all of the zones to see which one calculates
+ * to the highest priority.  We bump that zone's first zio to the head of the
+ * queue.
+ *
+ * We use a counter on the zone so that we can quickly find how many ops each
+ * zone has in the queue without having to search the entire queue itself.
+ * This scales better since the number of zones is expected to be on the
+ * order of 10-100 whereas the queue depth can be in the range of 50-2000.
+ * In addition, since the zio's in the queue only have the zoneid, we would
+ * have to look up the zone for each zio enqueued and that means the overhead
+ * for scanning the queue each time would be much higher.
+ *
+ * In all cases, we fall back to simply pulling the next op off the queue
+ * if something should go wrong.
+ */
+static zio_t *
+get_next_zio(vdev_queue_class_t *vqc, int qdepth, zio_priority_t p,
+    avl_tree_t *tree)
+{
+	zone_q_bump_t qbump;
+	zio_t *zp = NULL, *zphead;
+	int cnt = 0;
+
+	/* To avoid problems with int rounding, scale the queue depth by 10 */
+	qbump.zq_qdepth = qdepth * 10;
+	qbump.zq_priority = 0;
+	qbump.zq_zoneid = 0;
+	qbump.zq_queue = p;
+	(void) zone_walk(get_sched_pri_cb, &qbump);
+
+	zphead = avl_first(tree);
+
+	/* Check if the scheduler didn't pick a zone for some reason!? */
+	if (qbump.zq_zoneid != 0) {
+		for (zp = avl_first(tree); zp != NULL;
+		    zp = avl_walk(tree, zp, AVL_AFTER)) {
+			if (zp->io_zoneid == qbump.zq_zoneid)
+				break;
+			cnt++;
+		}
+	}
+
+	if (zp == NULL) {
+		zp = zphead;
+	} else if (zp != zphead) {
+		/*
+		 * Only fire the probe if we actually picked a different zio
+		 * than the one already at the head of the queue.
+		 */
+		DTRACE_PROBE4(zfs__zone__sched__bump, uint_t, zp->io_zoneid,
+		    uint_t, cnt, int, qbump.zq_priority, int, qbump.zq_wt);
+	}
+
+	return (zp);
+}
+
+/*
+ * Add our zone ID to the zio so we can keep track of which zones are doing
+ * what, even when the current thread processing the zio is not associated
+ * with the zone (e.g. the kernel taskq which pushes out TX groups).
+ */
+void
+zfs_zone_zio_init(zio_t *zp)
+{
+	zone_t	*zonep = curzone;
+
+	zp->io_zoneid = zonep->zone_id;
+}
+
+/*
+ * Track and throttle IO operations per zone. Called from:
+ *   - dmu_tx_count_write for (logical) write ops (both dataset and zvol writes
+ *     go through this path)
+ *   - arc_read for read ops that miss the ARC (both dataset and zvol)
+ * For each operation, increment that zone's counter based on the type of
+ * operation, then delay the operation, if necessary.
+ *
+ * There are three basic ways that we can see write ops:
+ * 1) An application does write syscalls.  Those ops go into a TXG which
+ *    we'll count here.  Sometime later a kernel taskq thread (we'll see the
+ *    vdev IO as zone 0) will perform some number of physical writes to commit
+ *    the TXG to disk.  Those writes are not associated with the zone which
+ *    made the write syscalls and the number of operations is not correlated
+ *    between the taskq and the zone. We only see logical writes in this
+ *    function, we see the physcial writes in the zfs_zone_zio_start and
+ *    zfs_zone_zio_done functions.
+ * 2) An application opens a file with O_SYNC.  Each write will result in
+ *    an operation which we'll see here plus a low-level vdev write from
+ *    that zone.
+ * 3) An application does write syscalls followed by an fsync().  We'll
+ *    count the writes going into a TXG here.  We'll also see some number
+ *    (usually much smaller, maybe only 1) of low-level vdev writes from this
+ *    zone when the fsync is performed, plus some other low-level vdev writes
+ *    from the taskq in zone 0 (are these metadata writes?).
+ *
+ * 4) In addition to the above, there are misc. system-level writes, such as
+ *    writing out dirty pages to swap, or sync(2) calls, which will be handled
+ *    by the global zone and which we count but don't generally worry about.
+ *
+ * Because of the above, we can see writes twice; first because this function
+ * is always called by a zone thread for logical writes, but then we also will
+ * count the physical writes that are performed at a low level via
+ * zfs_zone_zio_start. Without this, it can look like a non-global zone never
+ * writes (case 1). Depending on when the TXG is synced, the counts may be in
+ * the same sample bucket or in a different one.
+ *
+ * Tracking read operations is simpler due to their synchronous semantics.  The
+ * zfs_read function -- called as a result of a read(2) syscall -- will always
+ * retrieve the data to be read through arc_read and we only come into this
+ * function when we have an arc miss.
+ */
+void
+zfs_zone_io_throttle(zfs_zone_iop_type_t type)
+{
+	zone_t *zonep = curzone;
+	hrtime_t unow, last_checked;
+	uint16_t wait;
+
+	unow = GET_USEC_TIME;
+
+	/*
+	 * Only bump the counter for logical writes here.  The counters for
+	 * tracking physical IO operations are handled in zfs_zone_zio_done.
+	 */
+	if (type == ZFS_ZONE_IOP_LOGICAL_WRITE) {
+		mutex_enter(&zonep->zone_stg_io_lock);
+		add_iop(zonep, unow, type, 0);
+		mutex_exit(&zonep->zone_stg_io_lock);
+	}
+
+	if (!zfs_zone_delay_enable)
+		return;
+
+	/*
+	 * If the zone's I/O priority is set to zero, don't throttle that zone's
+	 * operations at all.
+	 */
+	if (zonep->zone_zfs_io_pri == 0)
+		return;
+
+	/*
+	 * XXX There's a potential race here in that more than one thread may
+	 * update the zone delays concurrently.  The worst outcome is corruption
+	 * of our data to track each zone's IO, so the algorithm may make
+	 * incorrect throttling decisions until the data is refreshed.
+	 */
+	last_checked = zfs_zone_last_checked;
+	if ((unow - last_checked) > zfs_zone_adjust_time) {
+		zfs_zone_last_checked = unow;
+		zfs_zone_wait_adjust(unow, last_checked);
+	}
+
+	if ((wait = zonep->zone_io_delay) > 0) {
+		/*
+		 * If this is a write and we're doing above normal TXG
+		 * syncing, then throttle for longer than normal.
+		 */
+		if (type == ZFS_ZONE_IOP_LOGICAL_WRITE &&
+		    (txg_cnt > 1 || txg_sync_rate > 1))
+			wait *= zfs_zone_txg_throttle_scale;
+
+		/*
+		 * sdt:::zfs-zone-wait
+		 *
+		 *	arg0: zone ID
+		 *	arg1: type of IO operation
+		 *	arg2: time to delay (in us)
+		 */
+		DTRACE_PROBE3(zfs__zone__wait, uintptr_t, zonep->zone_id,
+		    uintptr_t, type, uintptr_t, wait);
+
+		drv_usecwait(wait);
+
+		if (zonep->zone_vfs_stats != NULL) {
+			atomic_inc_64(&zonep->zone_vfs_stats->
+			    zv_delay_cnt.value.ui64);
+			atomic_add_64(&zonep->zone_vfs_stats->
+			    zv_delay_time.value.ui64, wait);
+		}
+	}
+}
+
+/*
+ * XXX Ignore the pool pointer parameter for now.
+ *
+ * Keep track to see if the TXG sync rate is running above the expected rate.
+ * If so, this implies that we are filling TXG's at a high rate due to a heavy
+ * write workload.  We use this as input into the zone throttle.
+ *
+ * This function is called every 5 seconds (zfs_txg_timeout) under a normal
+ * write load.  In this case, the sync rate is going to be 1.  When there
+ * is a heavy write load, TXG's fill up fast and the sync thread will write
+ * the TXG more frequently (perhaps once a second).  In this case the rate
+ * will be > 1.  The sync rate is a lagging indicator since it can be up
+ * to 5 seconds old.  We use the txg_cnt to keep track of the rate in the
+ * current 5 second interval and txg_sync_rate to keep track of the previous
+ * 5 second interval.  In that way we don't have a period (1 or more seconds)
+ * where the txg_cnt == 0 and we cut back on throttling even though the rate
+ * is still high.
+ */
+/*ARGSUSED*/
+void
+zfs_zone_report_txg_sync(void *dp)
+{
+	uint_t now;
+
+	txg_cnt++;
+	now = (uint_t)(gethrtime() / NANOSEC);
+	if ((now - txg_last_check) >= zfs_txg_timeout) {
+		txg_sync_rate = txg_cnt / 2;
+		txg_cnt = 0;
+		txg_last_check = now;
+	}
+}
+
+hrtime_t
+zfs_zone_txg_delay()
+{
+	if (curzone->zone_io_util_above_avg)
+		return (zfs_zone_txg_delay_nsec);
+
+	return (MSEC2NSEC(10));
+}
+
+/*
+ * Called from vdev_disk_io_start when an IO hits the end of the zio pipeline
+ * and is issued.
+ * Keep track of start time for latency calculation in zfs_zone_zio_done.
+ */
+void
+zfs_zone_zio_start(zio_t *zp)
+{
+	zone_t	*zonep;
+
+	/*
+	 * I/Os of type ZIO_TYPE_IOCTL are used to flush the disk cache, not for
+	 * an actual I/O operation.  Ignore those operations as they relate to
+	 * throttling and scheduling.
+	 */
+	if (zp->io_type == ZIO_TYPE_IOCTL)
+		return;
+
+	if ((zonep = zone_find_by_id(zp->io_zoneid)) == NULL)
+		return;
+
+	mutex_enter(&zonep->zone_zfs_lock);
+	if (zp->io_type == ZIO_TYPE_READ)
+		kstat_runq_enter(&zonep->zone_zfs_rwstats);
+	zonep->zone_zfs_weight = 0;
+	mutex_exit(&zonep->zone_zfs_lock);
+
+	mutex_enter(&zfs_disk_lock);
+	zp->io_dispatched = gethrtime();
+
+	if (zfs_disk_rcnt++ != 0)
+		zfs_disk_rtime += (zp->io_dispatched - zfs_disk_rlastupdate);
+	zfs_disk_rlastupdate = zp->io_dispatched;
+	mutex_exit(&zfs_disk_lock);
+
+	zone_rele(zonep);
+}
+
+/*
+ * Called from vdev_disk_io_done when an IO completes.
+ * Increment our counter for zone ops.
+ * Calculate the IO latency avg. for this zone.
+ */
+void
+zfs_zone_zio_done(zio_t *zp)
+{
+	zone_t	*zonep;
+	hrtime_t now, unow, udelta;
+
+	if (zp->io_type == ZIO_TYPE_IOCTL)
+		return;
+
+	if (zp->io_dispatched == 0)
+		return;
+
+	if ((zonep = zone_find_by_id(zp->io_zoneid)) == NULL)
+		return;
+
+	now = gethrtime();
+	unow = NANO_TO_MICRO(now);
+	udelta = unow - NANO_TO_MICRO(zp->io_dispatched);
+
+	mutex_enter(&zonep->zone_zfs_lock);
+
+	/*
+	 * To calculate the wsvc_t average, keep a cumulative sum of all the
+	 * wait time before each I/O was dispatched.  Since most writes are
+	 * asynchronous, only track the wait time for read I/Os.
+	 */
+	if (zp->io_type == ZIO_TYPE_READ) {
+		zonep->zone_zfs_rwstats.reads++;
+		zonep->zone_zfs_rwstats.nread += zp->io_size;
+
+		zonep->zone_zfs_stats->zz_waittime.value.ui64 +=
+		    zp->io_dispatched - zp->io_timestamp;
+
+		kstat_runq_exit(&zonep->zone_zfs_rwstats);
+	} else {
+		zonep->zone_zfs_rwstats.writes++;
+		zonep->zone_zfs_rwstats.nwritten += zp->io_size;
+	}
+
+	mutex_exit(&zonep->zone_zfs_lock);
+
+	mutex_enter(&zfs_disk_lock);
+	zfs_disk_rcnt--;
+	zfs_disk_rtime += (now - zfs_disk_rlastupdate);
+	zfs_disk_rlastupdate = now;
+
+	if (udelta > zfs_zone_laggard_threshold)
+		zfs_disk_last_laggard = unow;
+
+	mutex_exit(&zfs_disk_lock);
+
+	if (zfs_zone_delay_enable) {
+		mutex_enter(&zonep->zone_stg_io_lock);
+		add_iop(zonep, unow, zp->io_type == ZIO_TYPE_READ ?
+		    ZFS_ZONE_IOP_READ : ZFS_ZONE_IOP_WRITE, udelta);
+		mutex_exit(&zonep->zone_stg_io_lock);
+	}
+
+	zone_rele(zonep);
+
+	/*
+	 * sdt:::zfs-zone-latency
+	 *
+	 *	arg0: zone ID
+	 *	arg1: type of I/O operation
+	 *	arg2: I/O latency (in us)
+	 */
+	DTRACE_PROBE3(zfs__zone__latency, uintptr_t, zp->io_zoneid,
+	    uintptr_t, zp->io_type, uintptr_t, udelta);
+}
+
+void
+zfs_zone_zio_dequeue(zio_t *zp)
+{
+	zio_priority_t p;
+	zone_t	*zonep;
+
+	p = zp->io_priority;
+	if (p != ZIO_PRIORITY_SYNC_READ && p != ZIO_PRIORITY_SYNC_WRITE)
+		return;
+
+	/* We depend on p being defined as either 0 or 1 */
+	ASSERT(p < 2);
+
+	if ((zonep = zone_find_by_id(zp->io_zoneid)) == NULL)
+		return;
+
+	mutex_enter(&zonep->zone_stg_io_lock);
+	ASSERT(zonep->zone_zfs_queued[p] > 0);
+	if (zonep->zone_zfs_queued[p] == 0)
+		cmn_err(CE_WARN, "zfs_zone_zio_dequeue: count==0");
+	else
+		zonep->zone_zfs_queued[p]--;
+	mutex_exit(&zonep->zone_stg_io_lock);
+	zone_rele(zonep);
+}
+
+void
+zfs_zone_zio_enqueue(zio_t *zp)
+{
+	zio_priority_t p;
+	zone_t	*zonep;
+
+	p = zp->io_priority;
+	if (p != ZIO_PRIORITY_SYNC_READ && p != ZIO_PRIORITY_SYNC_WRITE)
+		return;
+
+	/* We depend on p being defined as either 0 or 1 */
+	ASSERT(p < 2);
+
+	if ((zonep = zone_find_by_id(zp->io_zoneid)) == NULL)
+		return;
+
+	mutex_enter(&zonep->zone_stg_io_lock);
+	zonep->zone_zfs_queued[p]++;
+	mutex_exit(&zonep->zone_stg_io_lock);
+	zone_rele(zonep);
+}
+
+/*
+ * Called from vdev_queue_io_to_issue. That function is where zio's are listed
+ * in FIFO order on one of the sync queues, then pulled off (by
+ * vdev_queue_io_remove) and issued.  We potentially do zone-based scheduling
+ * here to find a zone's zio deeper in the sync queue and issue that instead
+ * of simply doing FIFO.
+ *
+ * We only do zone-based zio scheduling for the two synchronous I/O queues
+ * (read & write). These queues are normally serviced in FIFO order but we
+ * may decide to move a zone's zio to the head of the line. A typical I/O
+ * load will be mostly synchronous reads and some asynchronous writes (which
+ * are scheduled differently due to transaction groups). There will also be
+ * some synchronous writes for those apps which want to ensure their data is on
+ * disk. We want to make sure that a zone with a single-threaded app (e.g. the
+ * shell) that is doing synchronous I/O (typically reads) isn't penalized by
+ * other zones which are doing lots of synchronous I/O because they have many
+ * running threads.
+ *
+ * The vq->vq_lock mutex is held when we're executing this function so we
+ * can safely access the "last zone" variable on the queue.
+ */
+zio_t *
+zfs_zone_schedule(vdev_queue_t *vq, zio_priority_t p, avl_index_t idx,
+    avl_tree_t *tree)
+{
+	vdev_queue_class_t *vqc = &vq->vq_class[p];
+	uint_t cnt;
+	zoneid_t last_zone;
+	zio_t *zio;
+
+	ASSERT(MUTEX_HELD(&vq->vq_lock));
+
+	/* Don't change the order on the LBA ordered queues. */
+	if (p != ZIO_PRIORITY_SYNC_READ && p != ZIO_PRIORITY_SYNC_WRITE)
+		return (avl_nearest(tree, idx, AVL_AFTER));
+
+	/* We depend on p being defined as either 0 or 1 */
+	ASSERT(p < 2);
+
+	cnt = avl_numnodes(tree);
+	last_zone = vq->vq_last_zone_id;
+
+	/*
+	 * If there are only a few zios in the queue then just issue the head.
+	 * If there are more than a few zios already queued up, then use
+	 * scheduling to get the next zio.
+	 */
+	if (!zfs_zone_schedule_enable || cnt < zfs_zone_schedule_thresh)
+		zio = avl_nearest(tree, idx, AVL_AFTER);
+	else
+		zio = get_next_zio(vqc, cnt, p, tree);
+
+	vq->vq_last_zone_id = zio->io_zoneid;
+
+	/*
+	 * Probe with 4 args; the number of IOs in the queue, the zone that
+	 * was last scheduled off this queue, the zone that was associated
+	 * with the next IO that is scheduled, and which queue (priority).
+	 */
+	DTRACE_PROBE4(zfs__zone__sched, uint_t, cnt, uint_t, last_zone,
+	    uint_t, zio->io_zoneid, uint_t, p);
+
+	return (zio);
+}
+
+#endif
diff --git a/usr/src/uts/common/fs/zfs/zil.c b/usr/src/uts/common/fs/zfs/zil.c
index 38b57e0123..f9f93999db 100644
--- a/usr/src/uts/common/fs/zfs/zil.c
+++ b/usr/src/uts/common/fs/zfs/zil.c
@@ -21,6 +21,7 @@
 /*
  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
  * Copyright (c) 2011, 2014 by Delphix. All rights reserved.
+ * Copyright (c) 2014, Joyent, Inc. All rights reserved.
  */
 
 /* Portions Copyright 2010 Robert Milkowski */
@@ -1801,9 +1802,18 @@ zil_close(zilog_t *zilog)
 	if (lwb != NULL)
 		txg = lwb->lwb_max_txg;
 	mutex_exit(&zilog->zl_lock);
-	if (txg)
+
+	if (zilog_is_dirty(zilog)) {
+		/*
+		 * If we're dirty, always wait for the current transaction --
+		 * our lwb_max_txg may be in the past.
+		 */
+		txg_wait_synced(zilog->zl_dmu_pool, 0);
+	} else if (txg) {
 		txg_wait_synced(zilog->zl_dmu_pool, txg);
-	ASSERT(!zilog_is_dirty(zilog));
+	}
+
+	VERIFY(!zilog_is_dirty(zilog));
 
 	taskq_destroy(zilog->zl_clean_taskq);
 	zilog->zl_clean_taskq = NULL;
diff --git a/usr/src/uts/common/fs/zfs/zio.c b/usr/src/uts/common/fs/zfs/zio.c
index aa0f2945dd..2fa3213be0 100644
--- a/usr/src/uts/common/fs/zfs/zio.c
+++ b/usr/src/uts/common/fs/zfs/zio.c
@@ -22,6 +22,7 @@
  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
  * Copyright (c) 2011, 2015 by Delphix. All rights reserved.
  * Copyright (c) 2011 Nexenta Systems, Inc. All rights reserved.
+ * Copyright (c) 2013 by Saso Kiselkov. All rights reserved.
  */
 
 #include <sys/sysmacros.h>
@@ -39,6 +40,7 @@
 #include <sys/ddt.h>
 #include <sys/blkptr.h>
 #include <sys/zfeature.h>
+#include <sys/zfs_zone.h>
 
 /*
  * ==========================================================================
@@ -557,11 +559,14 @@ zio_create(zio_t *pio, spa_t *spa, uint64_t txg, const blkptr_t *bp,
 		zio->io_bookmark = *zb;
 
 	if (pio != NULL) {
+		zio->io_zoneid = pio->io_zoneid;
 		if (zio->io_logical == NULL)
 			zio->io_logical = pio->io_logical;
 		if (zio->io_child_type == ZIO_CHILD_GANG)
 			zio->io_gang_leader = pio->io_gang_leader;
 		zio_add_child(pio, zio);
+	} else {
+		zfs_zone_zio_init(zio);
 	}
 
 	return (zio);
diff --git a/usr/src/uts/common/fs/zfs/zvol.c b/usr/src/uts/common/fs/zfs/zvol.c
index 80888103fe..f681b1dc65 100644
--- a/usr/src/uts/common/fs/zfs/zvol.c
+++ b/usr/src/uts/common/fs/zfs/zvol.c
@@ -25,7 +25,7 @@
  *
  * Copyright 2011 Nexenta Systems, Inc.  All rights reserved.
  * Copyright (c) 2012, 2014 by Delphix. All rights reserved.
- * Copyright (c) 2013, Joyent, Inc. All rights reserved.
+ * Copyright (c) 2014, Joyent, Inc. All rights reserved.
  */
 
 /*
@@ -83,6 +83,7 @@
 #include <sys/zvol.h>
 #include <sys/dumphdr.h>
 #include <sys/zil_impl.h>
+#include <sys/sdt.h>
 #include <sys/dbuf.h>
 #include <sys/dmu_tx.h>
 #include <sys/zfeature.h>
@@ -137,6 +138,11 @@ typedef struct zvol_state {
 #define	ZVOL_EXCL	0x4
 #define	ZVOL_WCE	0x8
 
+#define	VOP_LATENCY_10MS	10000000
+#define	VOP_LATENCY_100MS	100000000
+#define	VOP_LATENCY_1S		1000000000
+#define	VOP_LATENCY_10S		10000000000
+
 /*
  * zvol maximum transfer in one DMU tx.
  */
@@ -1378,6 +1384,9 @@ zvol_read(dev_t dev, uio_t *uio, cred_t *cr)
 	uint64_t volsize;
 	rl_t *rl;
 	int error = 0;
+	zone_t *zonep = curzone;
+	uint64_t tot_bytes;
+	hrtime_t start, lat;
 
 	zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
 	if (zv == NULL)
@@ -1394,6 +1403,14 @@ zvol_read(dev_t dev, uio_t *uio, cred_t *cr)
 		return (error);
 	}
 
+	DTRACE_PROBE3(zvol__uio__start, dev_t, dev, uio_t *, uio, int, 0);
+
+	mutex_enter(&zonep->zone_vfs_lock);
+	kstat_runq_enter(&zonep->zone_vfs_rwstats);
+	mutex_exit(&zonep->zone_vfs_lock);
+	start = gethrtime();
+	tot_bytes = 0;
+
 	rl = zfs_range_lock(&zv->zv_znode, uio->uio_loffset, uio->uio_resid,
 	    RL_READER);
 	while (uio->uio_resid > 0 && uio->uio_loffset < volsize) {
@@ -1403,6 +1420,7 @@ zvol_read(dev_t dev, uio_t *uio, cred_t *cr)
 		if (bytes > volsize - uio->uio_loffset)
 			bytes = volsize - uio->uio_loffset;
 
+		tot_bytes += bytes;
 		error =  dmu_read_uio(zv->zv_objset, ZVOL_OBJ, uio, bytes);
 		if (error) {
 			/* convert checksum errors into IO errors */
@@ -1412,6 +1430,39 @@ zvol_read(dev_t dev, uio_t *uio, cred_t *cr)
 		}
 	}
 	zfs_range_unlock(rl);
+
+	mutex_enter(&zonep->zone_vfs_lock);
+	zonep->zone_vfs_rwstats.reads++;
+	zonep->zone_vfs_rwstats.nread += tot_bytes;
+	kstat_runq_exit(&zonep->zone_vfs_rwstats);
+	mutex_exit(&zonep->zone_vfs_lock);
+
+	lat = gethrtime() - start;
+
+	if (lat >= VOP_LATENCY_10MS) {
+		zone_vfs_kstat_t *zvp;
+
+		zvp = zonep->zone_vfs_stats;
+		if (lat < VOP_LATENCY_100MS) {
+			atomic_inc_64(&zvp->zv_10ms_ops.value.ui64);
+		} else if (lat < VOP_LATENCY_1S) {
+			atomic_inc_64(&zvp->zv_10ms_ops.value.ui64);
+			atomic_inc_64(&zvp->zv_100ms_ops.value.ui64);
+		} else if (lat < VOP_LATENCY_10S) {
+			atomic_inc_64(&zvp->zv_10ms_ops.value.ui64);
+			atomic_inc_64(&zvp->zv_100ms_ops.value.ui64);
+			atomic_inc_64(&zvp->zv_1s_ops.value.ui64);
+		} else {
+			atomic_inc_64(&zvp->zv_10ms_ops.value.ui64);
+			atomic_inc_64(&zvp->zv_100ms_ops.value.ui64);
+			atomic_inc_64(&zvp->zv_1s_ops.value.ui64);
+			atomic_inc_64(&zvp->zv_10s_ops.value.ui64);
+		}
+	}
+
+	DTRACE_PROBE4(zvol__uio__done, dev_t, dev, uio_t *, uio, int, 0, int,
+	    error);
+
 	return (error);
 }
 
@@ -1425,6 +1476,9 @@ zvol_write(dev_t dev, uio_t *uio, cred_t *cr)
 	rl_t *rl;
 	int error = 0;
 	boolean_t sync;
+	zone_t *zonep = curzone;
+	uint64_t tot_bytes;
+	hrtime_t start, lat;
 
 	zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
 	if (zv == NULL)
@@ -1441,6 +1495,19 @@ zvol_write(dev_t dev, uio_t *uio, cred_t *cr)
 		return (error);
 	}
 
+	DTRACE_PROBE3(zvol__uio__start, dev_t, dev, uio_t *, uio, int, 1);
+
+	/*
+	 * For the purposes of VFS kstat consumers, the "waitq" calculation is
+	 * repurposed as the active queue for zvol write operations. There's no
+	 * actual wait queue for zvol operations.
+	 */
+	mutex_enter(&zonep->zone_vfs_lock);
+	kstat_waitq_enter(&zonep->zone_vfs_rwstats);
+	mutex_exit(&zonep->zone_vfs_lock);
+	start = gethrtime();
+	tot_bytes = 0;
+
 	sync = !(zv->zv_flags & ZVOL_WCE) ||
 	    (zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS);
 
@@ -1454,6 +1521,7 @@ zvol_write(dev_t dev, uio_t *uio, cred_t *cr)
 		if (bytes > volsize - off)	/* don't write past the end */
 			bytes = volsize - off;
 
+		tot_bytes += bytes;
 		dmu_tx_hold_write(tx, ZVOL_OBJ, off, bytes);
 		error = dmu_tx_assign(tx, TXG_WAIT);
 		if (error) {
@@ -1471,6 +1539,39 @@ zvol_write(dev_t dev, uio_t *uio, cred_t *cr)
 	zfs_range_unlock(rl);
 	if (sync)
 		zil_commit(zv->zv_zilog, ZVOL_OBJ);
+
+	DTRACE_PROBE4(zvol__uio__done, dev_t, dev, uio_t *, uio, int, 1, int,
+	    error);
+
+	mutex_enter(&zonep->zone_vfs_lock);
+	zonep->zone_vfs_rwstats.writes++;
+	zonep->zone_vfs_rwstats.nwritten += tot_bytes;
+	kstat_waitq_exit(&zonep->zone_vfs_rwstats);
+	mutex_exit(&zonep->zone_vfs_lock);
+
+	lat = gethrtime() - start;
+
+	if (lat >= VOP_LATENCY_10MS) {
+		zone_vfs_kstat_t *zvp;
+
+		zvp = zonep->zone_vfs_stats;
+		if (lat < VOP_LATENCY_100MS) {
+			atomic_inc_64(&zvp->zv_10ms_ops.value.ui64);
+		} else if (lat < VOP_LATENCY_1S) {
+			atomic_inc_64(&zvp->zv_10ms_ops.value.ui64);
+			atomic_inc_64(&zvp->zv_100ms_ops.value.ui64);
+		} else if (lat < VOP_LATENCY_10S) {
+			atomic_inc_64(&zvp->zv_10ms_ops.value.ui64);
+			atomic_inc_64(&zvp->zv_100ms_ops.value.ui64);
+			atomic_inc_64(&zvp->zv_1s_ops.value.ui64);
+		} else {
+			atomic_inc_64(&zvp->zv_10ms_ops.value.ui64);
+			atomic_inc_64(&zvp->zv_100ms_ops.value.ui64);
+			atomic_inc_64(&zvp->zv_1s_ops.value.ui64);
+			atomic_inc_64(&zvp->zv_10s_ops.value.ui64);
+		}
+	}
+
 	return (error);
 }
 
diff --git a/usr/src/uts/common/inet/inet_hash.h b/usr/src/uts/common/inet/inet_hash.h
new file mode 100644
index 0000000000..a790a797d1
--- /dev/null
+++ b/usr/src/uts/common/inet/inet_hash.h
@@ -0,0 +1,37 @@
+/*
+ * This file and its contents are supplied under the terms of the
+ * Common Development and Distribution License ("CDDL"), version 1.0.
+ * You may only use this file in accordance with the terms of version
+ * 1.0 of the CDDL.
+ *
+ * A full copy of the text of the CDDL should have accompanied this
+ * source.  A copy of the CDDL is also available via the Internet at
+ * http://www.illumos.org/license/CDDL.
+ */
+
+/*
+ * Copyright 2015 Joyent, Inc.
+ */
+
+#ifndef _INET_INET_HASH_H
+#define	_INET_INET_HASH_H
+
+/*
+ * Common packet hashing routines shared across MAC, UDP, and others.
+ */
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+#define	INET_PKT_HASH_L2	0x01
+#define	INET_PKT_HASH_L3	0x02
+#define	INET_PKT_HASH_L4	0x04
+
+extern uint64_t inet_pkt_hash(uint_t, mblk_t *, uint8_t);
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif /* _INET_INET_HASH_H */
diff --git a/usr/src/uts/common/inet/ip/ip_attr.c b/usr/src/uts/common/inet/ip/ip_attr.c
index 85ee142dfc..c350d67c2d 100644
--- a/usr/src/uts/common/inet/ip/ip_attr.c
+++ b/usr/src/uts/common/inet/ip/ip_attr.c
@@ -909,6 +909,11 @@ ixa_safe_copy(ip_xmit_attr_t *src, ip_xmit_attr_t *ixa)
 	 */
 	if (ixa->ixa_free_flags & IXA_FREE_CRED)
 		crhold(ixa->ixa_cred);
+
+	/*
+	 * There is no cleanup in progress on this new copy.
+	 */
+	ixa->ixa_tcpcleanup = IXATC_IDLE;
 }
 
 /*
diff --git a/usr/src/uts/common/inet/ip/ip_if.c b/usr/src/uts/common/inet/ip/ip_if.c
index 94b2396147..16a9b7e349 100644
--- a/usr/src/uts/common/inet/ip/ip_if.c
+++ b/usr/src/uts/common/inet/ip/ip_if.c
@@ -25,6 +25,9 @@
  * Copyright 2013 Joyent, Inc.
  * Copyright (c) 2014, OmniTI Computer Consulting, Inc. All rights reserved.
  */
+/*
+ * Copyright (c) 2013, Joyent, Inc. All rights reserved.
+ */
 
 /*
  * This file contains the interface control functions for IP.
diff --git a/usr/src/uts/common/inet/ip/ip_squeue.c b/usr/src/uts/common/inet/ip/ip_squeue.c
index 33a2fa5935..dedb4dadcc 100644
--- a/usr/src/uts/common/inet/ip/ip_squeue.c
+++ b/usr/src/uts/common/inet/ip/ip_squeue.c
@@ -163,7 +163,7 @@ ip_squeue_create(pri_t pri)
 {
 	squeue_t *sqp;
 
-	sqp = squeue_create(ip_squeue_worker_wait, pri);
+	sqp = squeue_create(ip_squeue_worker_wait, pri, B_TRUE);
 	ASSERT(sqp != NULL);
 	if (ip_squeue_create_callback != NULL)
 		ip_squeue_create_callback(sqp);
diff --git a/usr/src/uts/common/inet/ip/ip_tunables.c b/usr/src/uts/common/inet/ip/ip_tunables.c
index 58e3b59ff3..dda05a316d 100644
--- a/usr/src/uts/common/inet/ip/ip_tunables.c
+++ b/usr/src/uts/common/inet/ip/ip_tunables.c
@@ -20,6 +20,7 @@
  */
 /*
  * Copyright (c) 1991, 2010, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2012, Joyent, Inc. All rights reserved.
  * Copyright (c) 2013 by Delphix. All rights reserved.
  * Copyright (c) 2012, Joyent, Inc. All rights reserved.
  */
diff --git a/usr/src/uts/common/inet/ip/ipsecesp.c b/usr/src/uts/common/inet/ip/ipsecesp.c
index c325e8dc26..2ca770ebe9 100644
--- a/usr/src/uts/common/inet/ip/ipsecesp.c
+++ b/usr/src/uts/common/inet/ip/ipsecesp.c
@@ -234,8 +234,7 @@ esp_kstat_init(ipsecesp_stack_t *espstack, netstackid_t stackid)
 {
 	espstack->esp_ksp = kstat_create_netstack("ipsecesp", 0, "esp_stat",
 	    "net", KSTAT_TYPE_NAMED,
-	    sizeof (esp_kstats_t) / sizeof (kstat_named_t),
-	    KSTAT_FLAG_PERSISTENT, stackid);
+	    sizeof (esp_kstats_t) / sizeof (kstat_named_t), 0, stackid);
 
 	if (espstack->esp_ksp == NULL || espstack->esp_ksp->ks_data == NULL)
 		return (B_FALSE);
diff --git a/usr/src/uts/common/inet/ipf/ip_fil_solaris.c b/usr/src/uts/common/inet/ipf/ip_fil_solaris.c
index f958ca2261..227d2075f8 100644
--- a/usr/src/uts/common/inet/ipf/ip_fil_solaris.c
+++ b/usr/src/uts/common/inet/ipf/ip_fil_solaris.c
@@ -83,6 +83,14 @@ static	int	ipf_hook6_loop_out __P((hook_event_token_t, hook_data_t,
 static	int	ipf_hook6_loop_in __P((hook_event_token_t, hook_data_t,
     void *));
 static	int     ipf_hook6 __P((hook_data_t, int, int, void *));
+static	int	ipf_hookvndl3v4_in __P((hook_event_token_t, hook_data_t,
+    void *));
+static	int	ipf_hookvndl3v6_in __P((hook_event_token_t, hook_data_t,
+    void *));
+static	int	ipf_hookvndl3v4_out __P((hook_event_token_t, hook_data_t,
+    void *));
+static	int	ipf_hookvndl3v6_out __P((hook_event_token_t, hook_data_t,
+    void *));
 extern	int	ipf_geniter __P((ipftoken_t *, ipfgeniter_t *, ipf_stack_t *));
 extern	int	ipf_frruleiter __P((void *, int, void *, ipf_stack_t *));
 
@@ -152,6 +160,16 @@ char *hook6_loop_in_gz = 	"ipfilter_hook6_loop_in_gz";
 char *hook6_loop_out = 		"ipfilter_hook6_loop_out";
 char *hook6_loop_out_gz = 	"ipfilter_hook6_loop_out_gz";
 
+/* vnd IPv4/v6 hook names */
+char *hook4_vnd_in =		"ipfilter_hookvndl3v4_in";
+char *hook4_vnd_in_gz =		"ipfilter_hookvndl3v4_in_gz";
+char *hook6_vnd_in =		"ipfilter_hookvndl3v6_in";
+char *hook6_vnd_in_gz =		"ipfilter_hookvndl3v6_in_gz";
+char *hook4_vnd_out =		"ipfilter_hookvndl3v4_out";
+char *hook4_vnd_out_gz =	"ipfilter_hookvndl3v4_out_gz";
+char *hook6_vnd_out =		"ipfilter_hookvndl3v6_out";
+char *hook6_vnd_out_gz =	"ipfilter_hookvndl3v6_out_gz";
+
 /* ------------------------------------------------------------------------ */
 /* Function:    ipldetach                                                   */
 /* Returns:     int - 0 == success, else error.                             */
@@ -248,6 +266,31 @@ ipf_stack_t *ifs;
 		ifs->ifs_ipf_ipv4 = NULL;
 	}
 
+	/*
+	 * Remove VND hooks
+	 */
+	if (ifs->ifs_ipf_vndl3v4 != NULL) {
+		UNDO_HOOK(ifs_ipf_vndl3v4, ifs_hookvndl3v4_physical_in,
+		    NH_PHYSICAL_IN, ifs_ipfhookvndl3v4_in);
+		UNDO_HOOK(ifs_ipf_vndl3v4, ifs_hookvndl3v4_physical_out,
+		    NH_PHYSICAL_OUT, ifs_ipfhookvndl3v4_out);
+
+		if (net_protocol_release(ifs->ifs_ipf_vndl3v4) != 0)
+			goto detach_failed;
+		ifs->ifs_ipf_vndl3v4 = NULL;
+	}
+
+	if (ifs->ifs_ipf_vndl3v6 != NULL) {
+		UNDO_HOOK(ifs_ipf_vndl3v6, ifs_hookvndl3v6_physical_in,
+		    NH_PHYSICAL_IN, ifs_ipfhookvndl3v6_in);
+		UNDO_HOOK(ifs_ipf_vndl3v6, ifs_hookvndl3v6_physical_out,
+		    NH_PHYSICAL_OUT, ifs_ipfhookvndl3v6_out);
+
+		if (net_protocol_release(ifs->ifs_ipf_vndl3v6) != 0)
+			goto detach_failed;
+		ifs->ifs_ipf_vndl3v6 = NULL;
+	}
+
 #undef UNDO_HOOK
 
 #ifdef	IPFDEBUG
@@ -445,6 +488,48 @@ ipf_stack_t *ifs;
 	}
 
 	/*
+	 * Add VND INET hooks
+	 */
+	ifs->ifs_ipf_vndl3v4 = net_protocol_lookup(id, NHF_VND_INET);
+	if (ifs->ifs_ipf_vndl3v4 == NULL)
+		goto hookup_failed;
+
+	HOOK_INIT_GZ_BEFORE(ifs->ifs_ipfhookvndl3v4_in, ipf_hookvndl3v4_in,
+	    hook4_vnd_in, hook4_vnd_in_gz, ifs);
+	HOOK_INIT_GZ_AFTER(ifs->ifs_ipfhookvndl3v4_out, ipf_hookvndl3v4_out,
+	    hook4_vnd_out, hook4_vnd_out_gz, ifs);
+	ifs->ifs_hookvndl3v4_physical_in = (net_hook_register(ifs->ifs_ipf_vndl3v4,
+	    NH_PHYSICAL_IN, ifs->ifs_ipfhookvndl3v4_in) == 0);
+	if (!ifs->ifs_hookvndl3v4_physical_in)
+		goto hookup_failed;
+
+	ifs->ifs_hookvndl3v4_physical_out = (net_hook_register(ifs->ifs_ipf_vndl3v4,
+	    NH_PHYSICAL_OUT, ifs->ifs_ipfhookvndl3v4_out) == 0);
+	if (!ifs->ifs_hookvndl3v4_physical_out)
+		goto hookup_failed;
+
+
+	/*
+	 * VND INET6 hooks
+	 */
+	ifs->ifs_ipf_vndl3v6 = net_protocol_lookup(id, NHF_VND_INET6);
+	if (ifs->ifs_ipf_vndl3v6 == NULL)
+		goto hookup_failed;
+
+	HOOK_INIT_GZ_BEFORE(ifs->ifs_ipfhookvndl3v6_in, ipf_hookvndl3v6_in,
+	    hook6_vnd_in, hook6_vnd_in_gz, ifs);
+	HOOK_INIT_GZ_AFTER(ifs->ifs_ipfhookvndl3v6_out, ipf_hookvndl3v6_out,
+	    hook6_vnd_out, hook6_vnd_out_gz, ifs);
+	ifs->ifs_hookvndl3v6_physical_in = (net_hook_register(ifs->ifs_ipf_vndl3v6,
+	    NH_PHYSICAL_IN, ifs->ifs_ipfhookvndl3v6_in) == 0);
+	if (!ifs->ifs_hookvndl3v6_physical_in)
+		goto hookup_failed;
+
+	ifs->ifs_hookvndl3v6_physical_out = (net_hook_register(ifs->ifs_ipf_vndl3v6,
+	    NH_PHYSICAL_OUT, ifs->ifs_ipfhookvndl3v6_out) == 0);
+	if (!ifs->ifs_hookvndl3v6_physical_out)
+		goto hookup_failed;
+	/*
 	 * Reacquire ipf_global, now it is safe.
 	 */
 	WRITE_ENTER(&ifs->ifs_ipf_global);
@@ -1011,7 +1096,6 @@ cred_t *cp;
 		return ENXIO;
 	unit = isp->ipfs_minor;
 
-
         /*
 	 * ipf_find_stack returns with a read lock on ifs_ipf_global
 	 */
@@ -2045,6 +2129,42 @@ int ipf_hook6_loop_out(hook_event_token_t token, hook_data_t info, void *arg)
 }
 
 /* ------------------------------------------------------------------------ */
+/* Function:    ipf_hookvndl3_in					    */
+/* Returns:     int - 0 == packet ok, else problem, free packet if not done */
+/* Parameters:  event(I)     - pointer to event                             */
+/*              info(I)      - pointer to hook information for firewalling  */
+/*                                                                          */
+/* The vnd hooks are private hooks to ON. They represents a layer 2         */
+/* datapath generally used to implement virtual machines. The driver sends  */
+/* along L3 packets of either type IP or IPv6. The ethertype to distinguish */
+/* them is in the upper 16 bits while the remaining bits are the            */
+/* traditional packet hook flags.                                           */
+/*                                                                          */
+/* They end up calling the appropriate traditional ip hooks.                */
+/* ------------------------------------------------------------------------ */
+/*ARGSUSED*/
+int ipf_hookvndl3v4_in(hook_event_token_t token, hook_data_t info, void *arg)
+{
+	return ipf_hook4_in(token, info, arg);
+}
+
+int ipf_hookvndl3v6_in(hook_event_token_t token, hook_data_t info, void *arg)
+{
+	return ipf_hook6_in(token, info, arg);
+}
+
+/*ARGSUSED*/
+int ipf_hookvndl3v4_out(hook_event_token_t token, hook_data_t info, void *arg)
+{
+	return ipf_hook4_out(token, info, arg);
+}
+
+int ipf_hookvndl3v6_out(hook_event_token_t token, hook_data_t info, void *arg)
+{
+	return ipf_hook6_out(token, info, arg);
+}
+
+/* ------------------------------------------------------------------------ */
 /* Function:    ipf_hook4_loop_in                                           */
 /* Returns:     int - 0 == packet ok, else problem, free packet if not done */
 /* Parameters:  event(I)     - pointer to event                             */
diff --git a/usr/src/uts/common/inet/ipf/ipf.conf b/usr/src/uts/common/inet/ipf/ipf.conf
index 6b36f9fdbf..f49e024a72 100644
--- a/usr/src/uts/common/inet/ipf/ipf.conf
+++ b/usr/src/uts/common/inet/ipf/ipf.conf
@@ -1,3 +1,8 @@
 #
 #
 name="ipf" parent="pseudo" instance=0;
+
+# Increase the state table limits. fr_statemax should be ~70% of fr_statesize,
+# and both should be prime numbers
+fr_statesize=151007;
+fr_statemax=113279;
diff --git a/usr/src/uts/common/inet/ipf/netinet/ipf_stack.h b/usr/src/uts/common/inet/ipf/netinet/ipf_stack.h
index a239f1c1ca..9aa2478c6a 100644
--- a/usr/src/uts/common/inet/ipf/netinet/ipf_stack.h
+++ b/usr/src/uts/common/inet/ipf/netinet/ipf_stack.h
@@ -125,6 +125,10 @@ struct ipf_stack {
 	hook_t		*ifs_ipfhook6_loop_in;
 	hook_t		*ifs_ipfhook6_loop_out;
 	hook_t		*ifs_ipfhook6_nicevents;
+	hook_t		*ifs_ipfhookvndl3v4_in;
+	hook_t		*ifs_ipfhookvndl3v6_in;
+	hook_t		*ifs_ipfhookvndl3v4_out;
+	hook_t		*ifs_ipfhookvndl3v6_out;
 
 	/* flags to indicate whether hooks are registered. */
 	boolean_t	ifs_hook4_physical_in;
@@ -137,10 +141,16 @@ struct ipf_stack {
 	boolean_t	ifs_hook6_nic_events;
 	boolean_t	ifs_hook6_loopback_in;
 	boolean_t	ifs_hook6_loopback_out;
+	boolean_t	ifs_hookvndl3v4_physical_in;
+	boolean_t	ifs_hookvndl3v6_physical_in;
+	boolean_t	ifs_hookvndl3v4_physical_out;
+	boolean_t	ifs_hookvndl3v6_physical_out;
 
 	int		ifs_ipf_loopback;
 	net_handle_t	ifs_ipf_ipv4;
 	net_handle_t	ifs_ipf_ipv6;
+	net_handle_t	ifs_ipf_vndl3v4;
+	net_handle_t	ifs_ipf_vndl3v6;
 
 	/* ip_auth.c */
 	int			ifs_fr_authsize;
diff --git a/usr/src/uts/common/inet/ipf/solaris.c b/usr/src/uts/common/inet/ipf/solaris.c
index c541f4dddc..5d56debc31 100644
--- a/usr/src/uts/common/inet/ipf/solaris.c
+++ b/usr/src/uts/common/inet/ipf/solaris.c
@@ -625,7 +625,6 @@ ipf_stack_shutdown(const netid_t id, void *arg)
 /*
  * Destroy things for ipf for one stack.
  */
-/* ARGSUSED */
 static void
 ipf_stack_destroy_one(const netid_t id, ipf_stack_t *ifs)
 {
diff --git a/usr/src/uts/common/inet/sockmods/datafilt.c b/usr/src/uts/common/inet/sockmods/datafilt.c
new file mode 100644
index 0000000000..6e1171de46
--- /dev/null
+++ b/usr/src/uts/common/inet/sockmods/datafilt.c
@@ -0,0 +1,116 @@
+/*
+ * This file and its contents are supplied under the terms of the
+ * Common Development and Distribution License ("CDDL"), version 1.0.
+ * You may only use this file in accordance with the terms of version
+ * 1.0 of the CDDL.
+ *
+ * A full copy of the text of the CDDL should have accompanied this
+ * source.  A copy of the CDDL is also available via the Internet at
+ * http://www.illumos.org/license/CDDL.
+ */
+
+/*
+ * Copyright (c) 2012, OmniTI Computer Consulting, Inc. All rights reserved.
+ */
+
+/*
+ * This file implements a socketfilter used to deter TCP connections.
+ * To defer a connection means to delay the return of accept(3SOCKET)
+ * until at least one byte is ready to be read(2). This filter may be
+ * applied automatically or programmatically through the use of
+ * soconfig(1M) and setsockopt(3SOCKET).
+ */
+
+#include <sys/kmem.h>
+#include <sys/systm.h>
+#include <sys/stropts.h>
+#include <sys/strsun.h>
+#include <sys/socketvar.h>
+#include <sys/sockfilter.h>
+#include <sys/note.h>
+#include <sys/taskq.h>
+
+#define	DATAFILT_MODULE "datafilt"
+
+static struct modlmisc dataf_modlmisc = {
+	&mod_miscops,
+	"Kernel data-ready socket filter"
+};
+
+static struct modlinkage dataf_modlinkage = {
+	MODREV_1,
+	&dataf_modlmisc,
+	NULL
+};
+
+static sof_rval_t
+dataf_attach_passive_cb(sof_handle_t handle, sof_handle_t ph,
+    void *parg, struct sockaddr *laddr, socklen_t laddrlen,
+    struct sockaddr *faddr, socklen_t faddrlen, void **cookiep)
+{
+	_NOTE(ARGUNUSED(handle, ph, parg, laddr, laddrlen, faddr, faddrlen,
+	cookiep));
+	return (SOF_RVAL_DEFER);
+}
+
+static void
+dataf_detach_cb(sof_handle_t handle, void *cookie, cred_t *cr)
+{
+	_NOTE(ARGUNUSED(handle, cookie, cr));
+}
+
+static mblk_t *
+dataf_data_in_cb(sof_handle_t handle, void *cookie, mblk_t *mp, int flags,
+    size_t *lenp)
+{
+	_NOTE(ARGUNUSED(cookie, flags, lenp));
+
+	if (mp != NULL && MBLKL(mp) > 0) {
+		sof_newconn_ready(handle);
+		sof_bypass(handle);
+	}
+
+	return (mp);
+}
+
+static sof_ops_t dataf_ops = {
+	.sofop_attach_passive = dataf_attach_passive_cb,
+	.sofop_detach = dataf_detach_cb,
+	.sofop_data_in = dataf_data_in_cb
+};
+
+int
+_init(void)
+{
+	int err;
+
+	/*
+	 * This module is safe to attach even after some preliminary socket
+	 * setup calls have taken place. See the comment for SOF_ATT_SAFE.
+	 */
+	err = sof_register(SOF_VERSION, DATAFILT_MODULE, &dataf_ops,
+	    SOF_ATT_SAFE);
+	if (err != 0)
+		return (err);
+	if ((err = mod_install(&dataf_modlinkage)) != 0)
+		(void) sof_unregister(DATAFILT_MODULE);
+
+	return (err);
+}
+
+int
+_fini(void)
+{
+	int err;
+
+	if ((err = sof_unregister(DATAFILT_MODULE)) != 0)
+		return (err);
+
+	return (mod_remove(&dataf_modlinkage));
+}
+
+int
+_info(struct modinfo *modinfop)
+{
+	return (mod_info(&dataf_modlinkage, modinfop));
+}
diff --git a/usr/src/uts/common/inet/sockmods/socksctp.c b/usr/src/uts/common/inet/sockmods/socksctp.c
index bcd8e8d8fa..8b07c9b36e 100644
--- a/usr/src/uts/common/inet/sockmods/socksctp.c
+++ b/usr/src/uts/common/inet/sockmods/socksctp.c
@@ -21,6 +21,7 @@
 
 /*
  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2015 Joyent, Inc.  All rights reserved.
  */
 
 #include <sys/types.h>
@@ -403,7 +404,6 @@ sosctp_connect(struct sonode *so, struct sockaddr *name,
 	}
 
 	if (name == NULL || namelen == 0) {
-		mutex_exit(&so->so_lock);
 		error = EINVAL;
 		eprintsoline(so, error);
 		goto done;
diff --git a/usr/src/uts/common/inet/squeue.c b/usr/src/uts/common/inet/squeue.c
index 2e08dc359b..1009f0700f 100644
--- a/usr/src/uts/common/inet/squeue.c
+++ b/usr/src/uts/common/inet/squeue.c
@@ -23,7 +23,7 @@
  */
 
 /*
- * Copyright 2012 Joyent, Inc.  All rights reserved.
+ * Copyright (c) 2014 Joyent, Inc.  All rights reserved.
  */
 
 /*
@@ -61,6 +61,10 @@
  * connection are processed on that squeue. The connection ("conn") to
  * squeue mapping is stored in "conn_t" member "conn_sqp".
  *
+ * If the squeue is not related to TCP/IP, then the value of sqp->sq_isip is
+ * false and it will not have an associated conn_t, which means many aspects of
+ * the system, such as polling and swtiching squeues will not be used.
+ *
  * Since the processing of the connection cuts across multiple layers
  * but still allows packets for different connnection to be processed on
  * other CPU/squeues, squeues are also termed as "Vertical Perimeter" or
@@ -244,7 +248,7 @@ squeue_init(void)
 
 /* ARGSUSED */
 squeue_t *
-squeue_create(clock_t wait, pri_t pri)
+squeue_create(clock_t wait, pri_t pri, boolean_t isip)
 {
 	squeue_t *sqp = kmem_cache_alloc(squeue_cache, KM_SLEEP);
 
@@ -260,11 +264,36 @@ squeue_create(clock_t wait, pri_t pri)
 
 	sqp->sq_enter = squeue_enter;
 	sqp->sq_drain = squeue_drain;
+	sqp->sq_isip = isip;
 
 	return (sqp);
 }
 
 /*
+ * We need to kill the threads and then clean up. We should VERIFY that
+ * polling is disabled so we don't have to worry about disassociating from
+ * MAC/IP/etc.
+ */
+void
+squeue_destroy(squeue_t *sqp)
+{
+	kt_did_t worker, poll;
+	mutex_enter(&sqp->sq_lock);
+	VERIFY(!(sqp->sq_state & (SQS_POLL_THR_QUIESCED |
+	    SQS_POLL_QUIESCE_DONE | SQS_PAUSE | SQS_EXIT)));
+	worker = sqp->sq_worker->t_did;
+	poll = sqp->sq_poll_thr->t_did;
+	sqp->sq_state |= SQS_EXIT;
+	cv_signal(&sqp->sq_poll_cv);
+	cv_signal(&sqp->sq_worker_cv);
+	mutex_exit(&sqp->sq_lock);
+
+	thread_join(poll);
+	thread_join(worker);
+	kmem_cache_free(squeue_cache, sqp);
+}
+
+/*
  * Bind squeue worker thread to the specified CPU, given by CPU id.
  * If the CPU id  value is -1, bind the worker thread to the value
  * specified in sq_bind field. If a thread is already bound to a
@@ -475,18 +504,21 @@ squeue_enter(squeue_t *sqp, mblk_t *mp, mblk_t *tail, uint32_t cnt,
 			 * Handle squeue switching. More details in the
 			 * block comment at the top of the file
 			 */
-			if (connp->conn_sqp == sqp) {
+			if (sqp->sq_isip == B_FALSE || connp->conn_sqp == sqp) {
 				SQUEUE_DBG_SET(sqp, mp, proc, connp,
 				    tag);
-				connp->conn_on_sqp = B_TRUE;
+				if (sqp->sq_isip == B_TRUE)
+					connp->conn_on_sqp = B_TRUE;
 				DTRACE_PROBE3(squeue__proc__start, squeue_t *,
 				    sqp, mblk_t *, mp, conn_t *, connp);
 				(*proc)(connp, mp, sqp, ira);
 				DTRACE_PROBE2(squeue__proc__end, squeue_t *,
 				    sqp, conn_t *, connp);
-				connp->conn_on_sqp = B_FALSE;
+				if (sqp->sq_isip == B_TRUE) {
+					connp->conn_on_sqp = B_FALSE;
+					CONN_DEC_REF(connp);
+				}
 				SQUEUE_DBG_CLEAR(sqp);
-				CONN_DEC_REF(connp);
 			} else {
 				SQUEUE_ENTER_ONE(connp->conn_sqp, mp, proc,
 				    connp, ira, SQ_FILL, SQTAG_SQUEUE_CHANGE);
@@ -513,7 +545,7 @@ squeue_enter(squeue_t *sqp, mblk_t *mp, mblk_t *tail, uint32_t cnt,
 				return;
 			}
 		} else {
-			if (ira != NULL) {
+			if (sqp->sq_isip == B_TRUE && ira != NULL) {
 				mblk_t	*attrmp;
 
 				ASSERT(cnt == 1);
@@ -587,7 +619,8 @@ squeue_enter(squeue_t *sqp, mblk_t *mp, mblk_t *tail, uint32_t cnt,
 		if (!(sqp->sq_state & SQS_REENTER) &&
 		    (process_flag != SQ_FILL) && (sqp->sq_first == NULL) &&
 		    (sqp->sq_run == curthread) && (cnt == 1) &&
-		    (connp->conn_on_sqp == B_FALSE)) {
+		    (sqp->sq_isip == B_FALSE ||
+		    connp->conn_on_sqp == B_FALSE)) {
 			sqp->sq_state |= SQS_REENTER;
 			mutex_exit(&sqp->sq_lock);
 
@@ -602,15 +635,21 @@ squeue_enter(squeue_t *sqp, mblk_t *mp, mblk_t *tail, uint32_t cnt,
 			 * Handle squeue switching. More details in the
 			 * block comment at the top of the file
 			 */
-			if (connp->conn_sqp == sqp) {
-				connp->conn_on_sqp = B_TRUE;
+			if (sqp->sq_isip == B_FALSE || connp->conn_sqp == sqp) {
+				SQUEUE_DBG_SET(sqp, mp, proc, connp,
+				    tag);
+				if (sqp->sq_isip == B_TRUE)
+					connp->conn_on_sqp = B_TRUE;
 				DTRACE_PROBE3(squeue__proc__start, squeue_t *,
 				    sqp, mblk_t *, mp, conn_t *, connp);
 				(*proc)(connp, mp, sqp, ira);
 				DTRACE_PROBE2(squeue__proc__end, squeue_t *,
 				    sqp, conn_t *, connp);
-				connp->conn_on_sqp = B_FALSE;
-				CONN_DEC_REF(connp);
+				if (sqp->sq_isip == B_TRUE) {
+					connp->conn_on_sqp = B_FALSE;
+					CONN_DEC_REF(connp);
+				}
+				SQUEUE_DBG_CLEAR(sqp);
 			} else {
 				SQUEUE_ENTER_ONE(connp->conn_sqp, mp, proc,
 				    connp, ira, SQ_FILL, SQTAG_SQUEUE_CHANGE);
@@ -631,7 +670,7 @@ squeue_enter(squeue_t *sqp, mblk_t *mp, mblk_t *tail, uint32_t cnt,
 #ifdef DEBUG
 		mp->b_tag = tag;
 #endif
-		if (ira != NULL) {
+		if (sqp->sq_isip && ira != NULL) {
 			mblk_t	*attrmp;
 
 			ASSERT(cnt == 1);
@@ -779,7 +818,7 @@ again:
 		mp->b_prev = NULL;
 
 		/* Is there an ip_recv_attr_t to handle? */
-		if (ip_recv_attr_is_mblk(mp)) {
+		if (sqp->sq_isip == B_TRUE && ip_recv_attr_is_mblk(mp)) {
 			mblk_t	*attrmp = mp;
 
 			ASSERT(attrmp->b_cont != NULL);
@@ -804,20 +843,25 @@ again:
 
 
 		/*
-		 * Handle squeue switching. More details in the
-		 * block comment at the top of the file
+		 * Handle squeue switching. More details in the block comment at
+		 * the top of the file. non-IP squeues cannot switch, as there
+		 * is no conn_t.
 		 */
-		if (connp->conn_sqp == sqp) {
+		if (sqp->sq_isip == B_FALSE || connp->conn_sqp == sqp) {
 			SQUEUE_DBG_SET(sqp, mp, proc, connp,
 			    mp->b_tag);
-			connp->conn_on_sqp = B_TRUE;
+			if (sqp->sq_isip == B_TRUE)
+				connp->conn_on_sqp = B_TRUE;
 			DTRACE_PROBE3(squeue__proc__start, squeue_t *,
 			    sqp, mblk_t *, mp, conn_t *, connp);
 			(*proc)(connp, mp, sqp, ira);
 			DTRACE_PROBE2(squeue__proc__end, squeue_t *,
 			    sqp, conn_t *, connp);
-			connp->conn_on_sqp = B_FALSE;
-			CONN_DEC_REF(connp);
+			if (sqp->sq_isip == B_TRUE) {
+				connp->conn_on_sqp = B_FALSE;
+				CONN_DEC_REF(connp);
+			}
+			SQUEUE_DBG_CLEAR(sqp);
 		} else {
 			SQUEUE_ENTER_ONE(connp->conn_sqp, mp, proc, connp, ira,
 			    SQ_FILL, SQTAG_SQUEUE_CHANGE);
@@ -1051,6 +1095,11 @@ squeue_polling_thread(squeue_t *sqp)
 		cv_wait(async, lock);
 		CALLB_CPR_SAFE_END(&cprinfo, lock);
 
+		if (sqp->sq_state & SQS_EXIT) {
+			mutex_exit(lock);
+			thread_exit();
+		}
+
 		ctl_state = sqp->sq_state & (SQS_POLL_THR_CONTROL |
 		    SQS_POLL_THR_QUIESCED);
 		if (ctl_state != 0) {
@@ -1076,6 +1125,9 @@ squeue_polling_thread(squeue_t *sqp)
 		    (SQS_PROC|SQS_POLLING|SQS_GET_PKTS)) ==
 		    (SQS_PROC|SQS_POLLING|SQS_GET_PKTS));
 
+		/* Only IP related squeues should reach this point */
+		VERIFY(sqp->sq_isip == B_TRUE);
+
 poll_again:
 		sq_rx_ring = sqp->sq_rx_ring;
 		sq_get_pkts = sq_rx_ring->rr_rx;
@@ -1205,6 +1257,7 @@ squeue_worker_thr_control(squeue_t *sqp)
 	ill_rx_ring_t	*rx_ring;
 
 	ASSERT(MUTEX_HELD(&sqp->sq_lock));
+	VERIFY(sqp->sq_isip == B_TRUE);
 
 	if (sqp->sq_state & SQS_POLL_RESTART) {
 		/* Restart implies a previous quiesce. */
@@ -1316,6 +1369,11 @@ squeue_worker(squeue_t *sqp)
 
 	for (;;) {
 		for (;;) {
+			if (sqp->sq_state & SQS_EXIT) {
+				mutex_exit(lock);
+				thread_exit();
+			}
+
 			/*
 			 * If the poll thread has handed control to us
 			 * we need to break out of the wait.
@@ -1412,6 +1470,7 @@ squeue_synch_enter(conn_t *connp, mblk_t *use_mp)
 
 again:
 	sqp = connp->conn_sqp;
+	VERIFY(sqp->sq_isip == B_TRUE);
 
 	mutex_enter(&sqp->sq_lock);
 	if (sqp->sq_first == NULL && !(sqp->sq_state & SQS_PROC)) {
@@ -1487,6 +1546,7 @@ void
 squeue_synch_exit(conn_t *connp)
 {
 	squeue_t *sqp = connp->conn_sqp;
+	VERIFY(sqp->sq_isip == B_TRUE);
 
 	mutex_enter(&sqp->sq_lock);
 	if (sqp->sq_run == curthread) {
diff --git a/usr/src/uts/common/inet/tcp_impl.h b/usr/src/uts/common/inet/tcp_impl.h
index 1b20e40aca..3488630f2c 100644
--- a/usr/src/uts/common/inet/tcp_impl.h
+++ b/usr/src/uts/common/inet/tcp_impl.h
@@ -20,7 +20,7 @@
  */
 /*
  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
- * Copyright (c) 2011, Joyent Inc. All rights reserved.
+ * Copyright (c) 2013, Joyent Inc. All rights reserved.
  * Copyright (c) 2013, OmniTI Computer Consulting, Inc. All rights reserved.
  * Copyright (c) 2013, 2014 by Delphix. All rights reserved.
  */
@@ -61,9 +61,9 @@ extern sock_downcalls_t sock_tcp_downcalls;
  * by setting it to 0.
  */
 #define	TCP_XMIT_LOWATER	4096
-#define	TCP_XMIT_HIWATER	49152
+#define	TCP_XMIT_HIWATER	128000
 #define	TCP_RECV_LOWATER	2048
-#define	TCP_RECV_HIWATER	128000
+#define	TCP_RECV_HIWATER	1048576
 
 /*
  * Bind hash list size and has function.  It has to be a power of 2 for
diff --git a/usr/src/uts/common/inet/udp/udp.c b/usr/src/uts/common/inet/udp/udp.c
index 5a15aea4de..eec3beddd2 100644
--- a/usr/src/uts/common/inet/udp/udp.c
+++ b/usr/src/uts/common/inet/udp/udp.c
@@ -22,6 +22,7 @@
  * Copyright (c) 1991, 2010, Oracle and/or its affiliates. All rights reserved.
  * Copyright 2013 Nexenta Systems, Inc.  All rights reserved.
  * Copyright 2014, OmniTI Computer Consulting, Inc. All rights reserved.
+ * Copyright 2015, Joyent, Inc.
  */
 /* Copyright (c) 1990 Mentat Inc. */
 
@@ -76,7 +77,8 @@
 #include <inet/ipclassifier.h>
 #include <sys/squeue_impl.h>
 #include <inet/ipnet.h>
-#include <sys/ethernet.h>
+#include <sys/vxlan.h>
+#include <inet/inet_hash.h>
 
 #include <sys/tsol/label.h>
 #include <sys/tsol/tnet.h>
@@ -346,6 +348,89 @@ void (*cl_inet_unbind)(netstackid_t stack_id, uint8_t protocol,
 typedef union T_primitives *t_primp_t;
 
 /*
+ * Various protocols that encapsulate UDP have no real use for the source port.
+ * Instead, they want to vary the source port to provide better equal-cost
+ * multipathing and other systems that use fanout. Consider something like
+ * VXLAN. If you're actually sending multiple different streams to a single
+ * host, if you don't vary the source port, then the tuple of ( SRC IP, DST IP,
+ * SRC Port, DST Port) will always be the same.
+ *
+ * Here, we return a port to hash this to, if we know how to hash it. If for
+ * some reason we can't perform an L4 hash, then we just return the default
+ * value, usually the default port. After we determine the hash we transform it
+ * so that it's in the range of [ min, max ].
+ *
+ * We'd like to avoid a pull up for the sake of performing the hash. If the
+ * first mblk_t doesn't have the full protocol header, then we just send it to
+ * the default. If for some reason we have an encapsulated packet that has its
+ * protocol header in different parts of an mblk_t, then we'll go with the
+ * default port. This means that that if a driver isn't consistent about how it
+ * generates the frames for a given flow, it will not always be consistently
+ * hashed. That should be an uncommon event.
+ */
+uint16_t
+udp_srcport_hash(mblk_t *mp, int type, uint16_t min, uint16_t max,
+    uint16_t def)
+{
+	size_t szused = 0;
+	struct ether_header *ether;
+	struct ether_vlan_header *vether;
+	ip6_t *ip6h;
+	ipha_t *ipha;
+	uint16_t sap;
+	uint64_t hash;
+	uint32_t mod;
+
+	ASSERT(min <= max);
+
+	if (type != UDP_HASH_VXLAN)
+		return (def);
+
+	if (!IS_P2ALIGNED(mp->b_rptr, sizeof (uint16_t)))
+		return (def);
+
+	/*
+	 * The following logic is VXLAN specific to get at the header, if we
+	 * have formats, eg. GENEVE, then we should ignore this.
+	 *
+	 * The kernel overlay device often puts a first mblk_t for the data
+	 * which is just the encap. If so, then we're going to use that and try
+	 * to avoid a pull up.
+	 */
+	if (MBLKL(mp) == VXLAN_HDR_LEN) {
+		if (mp->b_cont == NULL)
+			return (def);
+		mp = mp->b_cont;
+		ether = (struct ether_header *)mp->b_rptr;
+	} else if (MBLKL(mp) < VXLAN_HDR_LEN) {
+		return (def);
+	} else {
+		szused = VXLAN_HDR_LEN;
+		ether = (struct ether_header *)((uintptr_t)mp->b_rptr + szused);
+	}
+
+	/* Can we hold a MAC header? */
+	if (MBLKL(mp) + szused < sizeof (struct ether_header))
+		return (def);
+
+	/*
+	 * We need to lie about the starting offset into the message block for
+	 * convenience. Undo it at the end. We know that inet_pkt_hash() won't
+	 * modify the mblk_t.
+	 */
+	mp->b_rptr += szused;
+	hash = inet_pkt_hash(DL_ETHER, mp, INET_PKT_HASH_L2 |
+	    INET_PKT_HASH_L3 | INET_PKT_HASH_L4);
+	mp->b_rptr -= szused;
+
+	if (hash == 0)
+		return (def);
+
+	mod = max - min + 1;
+	return ((hash % mod) + min);
+}
+
+/*
  * Return the next anonymous port in the privileged port range for
  * bind checking.
  *
@@ -1583,6 +1668,11 @@ udp_opt_get(conn_t *connp, t_scalar_t level, t_scalar_t name,
 			*i1 = udp->udp_rcvhdr ? 1 : 0;
 			mutex_exit(&connp->conn_lock);
 			return (sizeof (int));
+		case UDP_SRCPORT_HASH:
+			mutex_enter(&connp->conn_lock);
+			*i1 = udp->udp_vxlanhash;
+			mutex_exit(&connp->conn_lock);
+			return (sizeof (int));
 		}
 	}
 	mutex_enter(&connp->conn_lock);
@@ -1718,6 +1808,26 @@ udp_do_opt_set(conn_opt_arg_t *coa, int level, int name,
 			udp->udp_rcvhdr = onoff;
 			mutex_exit(&connp->conn_lock);
 			return (0);
+		case UDP_SRCPORT_HASH:
+			/*
+			 * This should have already been verified, but double
+			 * check.
+			 */
+			if ((error = secpolicy_ip_config(cr, B_FALSE)) != 0) {
+				return (error);
+			}
+
+			/* First see if the val is something we understand */
+			if (*i1 != UDP_HASH_DISABLE && *i1 != UDP_HASH_VXLAN)
+				return (EINVAL);
+
+			if (!checkonly) {
+				mutex_enter(&connp->conn_lock);
+				udp->udp_vxlanhash = *i1;
+				mutex_exit(&connp->conn_lock);
+			}
+			/* Fully handled this option. */
+			return (0);
 		}
 		break;
 	}
@@ -2001,13 +2111,25 @@ udp_prepend_hdr(conn_t *connp, ip_xmit_attr_t *ixa, const ip_pkt_t *ipp,
 	uint32_t	cksum;
 	udp_t		*udp = connp->conn_udp;
 	boolean_t	insert_spi = udp->udp_nat_t_endpoint;
+	boolean_t	hash_srcport = udp->udp_vxlanhash;
 	uint_t		ulp_hdr_len;
+	uint16_t	srcport;
 
 	data_len = msgdsize(data_mp);
 	ulp_hdr_len = UDPH_SIZE;
 	if (insert_spi)
 		ulp_hdr_len += sizeof (uint32_t);
 
+	/*
+	 * If we have source port hashing going on, determine the hash before
+	 * we modify the mblk_t.
+	 */
+	if (hash_srcport == B_TRUE) {
+		srcport = udp_srcport_hash(mp, UDP_HASH_VXLAN,
+		    IPPORT_DYNAMIC_MIN, IPPORT_DYNAMIC_MAX,
+		    ntohs(connp->conn_lport));
+	}
+
 	mp = conn_prepend_hdr(ixa, ipp, v6src, v6dst, IPPROTO_UDP, flowinfo,
 	    ulp_hdr_len, data_mp, data_len, us->us_wroff_extra, &cksum, errorp);
 	if (mp == NULL) {
@@ -2019,7 +2141,11 @@ udp_prepend_hdr(conn_t *connp, ip_xmit_attr_t *ixa, const ip_pkt_t *ipp,
 	ixa->ixa_pktlen = data_len + ixa->ixa_ip_hdr_length;
 
 	udpha = (udpha_t *)(mp->b_rptr + ixa->ixa_ip_hdr_length);
-	udpha->uha_src_port = connp->conn_lport;
+	if (hash_srcport == B_TRUE) {
+		udpha->uha_src_port = htons(srcport);
+	} else {
+		udpha->uha_src_port = connp->conn_lport;
+	}
 	udpha->uha_dst_port = dstport;
 	udpha->uha_checksum = 0;
 	udpha->uha_length = htons(data_len);
@@ -3194,6 +3320,7 @@ udp_prepend_header_template(conn_t *connp, ip_xmit_attr_t *ixa, mblk_t *mp,
 	udp_t		*udp = connp->conn_udp;
 	udp_stack_t	*us = udp->udp_us;
 	boolean_t	insert_spi = udp->udp_nat_t_endpoint;
+	boolean_t	hash_srcport = udp->udp_vxlanhash;
 	uint_t		pktlen;
 	uint_t		alloclen;
 	uint_t		copylen;
@@ -3202,10 +3329,21 @@ udp_prepend_header_template(conn_t *connp, ip_xmit_attr_t *ixa, mblk_t *mp,
 	udpha_t		*udpha;
 	uint32_t	cksum;
 	ip_pkt_t	*ipp;
+	uint16_t	srcport;
 
 	ASSERT(MUTEX_HELD(&connp->conn_lock));
 
 	/*
+	 * If we have source port hashing going on, determine the hash before
+	 * we modify the mblk_t.
+	 */
+	if (hash_srcport == B_TRUE) {
+		srcport = udp_srcport_hash(mp, UDP_HASH_VXLAN,
+		    IPPORT_DYNAMIC_MIN, IPPORT_DYNAMIC_MAX,
+		    ntohs(connp->conn_lport));
+	}
+
+	/*
 	 * Copy the header template and leave space for an SPI
 	 */
 	copylen = connp->conn_ht_iphc_len;
@@ -3303,6 +3441,9 @@ udp_prepend_header_template(conn_t *connp, ip_xmit_attr_t *ixa, mblk_t *mp,
 		*((uint32_t *)(udpha + 1)) = 0;
 
 	udpha->uha_dst_port = dstport;
+	if (hash_srcport == B_TRUE)
+		udpha->uha_src_port = htons(srcport);
+
 	return (mp);
 }
 
diff --git a/usr/src/uts/common/inet/udp/udp_opt_data.c b/usr/src/uts/common/inet/udp/udp_opt_data.c
index c279bb4a21..fec3dec4c9 100644
--- a/usr/src/uts/common/inet/udp/udp_opt_data.c
+++ b/usr/src/uts/common/inet/udp/udp_opt_data.c
@@ -21,6 +21,7 @@
 /*
  * Copyright 2009 Sun Microsystems, Inc.  All rights reserved.
  * Use is subject to license terms.
+ * Copyright 2015, Joyent, Inc.
  */
 
 #include <sys/types.h>
@@ -292,6 +293,7 @@ opdes_t	udp_opt_arr[] = {
 	},
 { UDP_NAT_T_ENDPOINT, IPPROTO_UDP, OA_RW, OA_RW, OP_PRIVPORT, 0, sizeof (int),
 	0 },
+{ UDP_SRCPORT_HASH, IPPROTO_UDP, OA_R, OA_RW, OP_CONFIG, 0, sizeof (int), 0 }
 };
 
 /*
diff --git a/usr/src/uts/common/inet/udp_impl.h b/usr/src/uts/common/inet/udp_impl.h
index 6a31ce5c22..3b07abc309 100644
--- a/usr/src/uts/common/inet/udp_impl.h
+++ b/usr/src/uts/common/inet/udp_impl.h
@@ -20,6 +20,7 @@
  */
 /*
  * Copyright (c) 2000, 2010, Oracle and/or its affiliates. All rights reserved.
+ * Copyright 2015, Joyent, Inc.
  */
 
 #ifndef	_UDP_IMPL_H
@@ -178,8 +179,11 @@ typedef	struct udp_s {
 		udp_issocket : 1,	/* socket mode; sockfs is on top */
 		udp_nat_t_endpoint : 1,	/* UDP_NAT_T_ENDPOINT option */
 		udp_rcvhdr : 1,		/* UDP_RCVHDR option */
+		udp_vxlanhash: 1,	/* UDP_SRCPORT_HASH option */
+					/* Because there's only VXLAN, cheat */
+					/* and only use a single bit */
 
-		udp_pad_to_bit_31 : 29;
+		udp_pad_to_bit_31 : 28;
 
 	/* Following 2 fields protected by the uf_lock */
 	struct udp_s	*udp_bind_hash; /* Bind hash chain */
diff --git a/usr/src/uts/common/io/1394/targets/av1394/av1394_async.c b/usr/src/uts/common/io/1394/targets/av1394/av1394_async.c
index 94323582d6..c2c16e848b 100644
--- a/usr/src/uts/common/io/1394/targets/av1394/av1394_async.c
+++ b/usr/src/uts/common/io/1394/targets/av1394/av1394_async.c
@@ -24,7 +24,9 @@
  * Use is subject to license terms.
  */
 
-#pragma ident	"%Z%%M%	%I%	%E% SMI"
+/*
+ * Copyright (c) 2014, Joyent, Inc. All rights reserved.
+ */
 
 /*
  * av1394 asynchronous module
@@ -359,9 +361,10 @@ av1394_async_poll(av1394_inst_t *avp, short events, int anyyet, short *reventsp,
 	AV1394_TNF_ENTER(av1394_async_poll);
 
 	if (events & POLLIN) {
-		if (av1394_peekq(rq)) {
+		if (av1394_peekq(rq))
 			*reventsp |= POLLIN;
-		} else if (!anyyet) {
+
+		if ((!*reventsp && !anyyet) || (events & POLLET)) {
 			mutex_enter(&ap->a_mutex);
 			ap->a_pollevents |= POLLIN;
 			*phpp = &ap->a_pollhead;
diff --git a/usr/src/uts/common/io/aggr/aggr_port.c b/usr/src/uts/common/io/aggr/aggr_port.c
index 00545d2c03..a39110255a 100644
--- a/usr/src/uts/common/io/aggr/aggr_port.c
+++ b/usr/src/uts/common/io/aggr/aggr_port.c
@@ -21,6 +21,7 @@
 /*
  * Copyright 2010 Sun Microsystems, Inc.  All rights reserved.
  * Use is subject to license terms.
+ * Copyright 2012 OmniTI Computer Consulting, Inc  All rights reserved.
  */
 
 /*
@@ -528,8 +529,13 @@ aggr_port_promisc(aggr_port_t *port, boolean_t on)
 
 	if (on) {
 		mac_rx_clear(port->lp_mch);
+		/* We use the promisc callback because without hardware
+		 * rings, we deliver through flows that will cause duplicate
+		 * delivery of packets when we've flipped into this mode
+		 * to compensate for the lack of hardware MAC matching
+		 */
 		rc = mac_promisc_add(port->lp_mch, MAC_CLIENT_PROMISC_ALL,
-		    aggr_recv_cb, port, &port->lp_mphp,
+		    aggr_recv_promisc_cb, port, &port->lp_mphp,
 		    MAC_PROMISC_FLAGS_NO_TX_LOOP);
 		if (rc != 0) {
 			mac_rx_set(port->lp_mch, aggr_recv_cb, port);
diff --git a/usr/src/uts/common/io/aggr/aggr_recv.c b/usr/src/uts/common/io/aggr/aggr_recv.c
index 2bdb7872e3..0dfe234b70 100644
--- a/usr/src/uts/common/io/aggr/aggr_recv.c
+++ b/usr/src/uts/common/io/aggr/aggr_recv.c
@@ -21,6 +21,7 @@
 /*
  * Copyright 2008 Sun Microsystems, Inc.  All rights reserved.
  * Use is subject to license terms.
+ * Copyright 2012 OmniTI Computer Consulting, Inc  All rights reserved.
  */
 
 /*
@@ -68,16 +69,27 @@ aggr_recv_lacp(aggr_port_t *port, mac_resource_handle_t mrh, mblk_t *mp)
 
 /*
  * Callback function invoked by MAC service module when packets are
- * made available by a MAC port.
+ * made available by a MAC port, both in promisc_on mode and not.
  */
 /* ARGSUSED */
-void
-aggr_recv_cb(void *arg, mac_resource_handle_t mrh, mblk_t *mp,
-    boolean_t loopback)
+static void
+aggr_recv_path_cb(void *arg, mac_resource_handle_t mrh, mblk_t *mp,
+    boolean_t loopback, boolean_t promisc_path)
 {
 	aggr_port_t *port = (aggr_port_t *)arg;
 	aggr_grp_t *grp = port->lp_grp;
 
+	/* In the case where lp_promisc_on has been turned on to
+	 * compensate for insufficient hardware MAC matching and
+	 * hardware rings are not in use we will fall back to
+	 * using flows for delivery which can result in duplicates
+	 * pushed up the stack. Only respect the chosen path.
+	 */
+	if (port->lp_promisc_on != promisc_path) {
+		freemsgchain(mp);
+		return;
+	}
+
 	if (grp->lg_lacp_mode == AGGR_LACP_OFF) {
 		aggr_mac_rx(grp->lg_mh, mrh, mp);
 	} else {
@@ -161,3 +173,19 @@ aggr_recv_cb(void *arg, mac_resource_handle_t mrh, mblk_t *mp,
 		}
 	}
 }
+
+/* ARGSUSED */
+void
+aggr_recv_cb(void *arg, mac_resource_handle_t mrh, mblk_t *mp,
+    boolean_t loopback)
+{
+	aggr_recv_path_cb(arg, mrh, mp, loopback, B_FALSE);
+}
+
+/* ARGSUSED */
+void
+aggr_recv_promisc_cb(void *arg, mac_resource_handle_t mrh, mblk_t *mp,
+    boolean_t loopback)
+{
+	aggr_recv_path_cb(arg, mrh, mp, loopback, B_TRUE);
+}
diff --git a/usr/src/uts/common/io/axf/ax88172reg.h b/usr/src/uts/common/io/axf/ax88172reg.h
new file mode 100644
index 0000000000..8ca6ebc187
--- /dev/null
+++ b/usr/src/uts/common/io/axf/ax88172reg.h
@@ -0,0 +1,163 @@
+/*
+ * @(#)ax88172reg.h	1.1 09/06/15
+ * Macro definitions for ASIX AX88172 USB to fast ethernet controler
+ * based on ASIX AX88172/88772 data sheet
+ * This file is public domain. Coded by M.Murayama (KHF04453@nifty.com)
+ */
+
+#ifndef __AX88172_H__
+#define __AX88172_H__
+
+/*
+ * Vendor command definitions
+ */
+#define	VCMD_READ_SRAM			0x02
+#define	VCMD_WRITE_RXSRAM		0x03
+#define	VCMD_WRITE_TXSRAM		0x04
+#define	VCMD_SOFTWARE_MII_OP		0x06
+#define	VCMD_READ_MII_REG		0x07
+#define	VCMD_WRITE_MII_REG		0x08
+#define	VCMD_READ_MII_OPMODE		0x09
+#define	VCMD_HARDWARE_MII_OP		0x0a
+#define	VCMD_READ_SROM			0x0b
+#define	VCMD_WRITE_SROM			0x0c
+#define	VCMD_WRITE_SROM_ENABLE		0x0d
+#define	VCMD_WRITE_SROM_DISABLE		0x0e
+#define	VCMD_READ_RXCTRL		0x0f
+#define	VCMD_WRITE_RXCTRL		0x10
+#define	VCMD_READ_IPGS			0x11
+#define	VCMD_WRITE_IPG			0x12
+#define	VCMD_WRITE_IPG1			0x13
+#define	VCMD_WRITE_IPG2			0x14
+#define	VCMD_READ_MCAST_FILTER		0x15
+#define	VCMD_WRITE_MCAST_FILTER		0x16
+#define	VCMD_READ_NODE_ID		0x17
+#define	VCMD_READ_PHY_IDS		0x19
+#define	VCMD_READ_MEDIUM_STATUS		0x1a
+#define	VCMD_WRITE_MEDIUM_STATUS	0x1b
+#define	VCMD_SET_MONITOR_MODE		0x1c
+#define	VCMD_GET_MONITOR_MODE		0x1d
+#define	VCMD_READ_GPIO			0x1e
+#define	VCMD_WRITE_GPIO			0x1f
+
+/* ax88772 only,  currently not supported */
+#define	VCMD_WRITE_IPGS_88772		0x12
+#define	VCMD_READ_NODE_ID_88772		0x13
+#define	VCMD_WRITE_NODE_ID_88772	0x14
+#define	VCMD_WRITE_TEST_REG_88772	0x17
+#define	VCMD_SOFTWARE_RESET_88772	0x20
+#define	VCMD_READ_PHY_SELECT_88772	0x21
+#define	VCMD_WRITE_PHY_SELECT_88772	0x22
+
+
+/*
+ * Register definitions
+ */
+
+/* Rx control register */
+#define	RCR_SO		0x80	/* Start Operation */
+#define	RCR_AP_88772	0x20	/* accept physical address from mcast filter */
+#define	RCR_AM		0x10	/* accept multicast address */
+#define	RCR_AB		0x08	/* accept broadcast address */
+#define	RCR_SEP		0x04	/* save error packet */
+#define	RCR_AMALL	0x02	/* accept all multicast address */
+#define	RCR_PRO		0x01	/* promiscious, all frames received */
+
+#define	RCR_MFB	0x0300
+#define		RCR_MFB_SHIFT	8
+#define		RCR_MFB_2K	(0U << RCR_MFB_SHIFT)
+#define		RCR_MFB_4K	(1U << RCR_MFB_SHIFT)
+#define		RCR_MFB_8K	(2U << RCR_MFB_SHIFT)
+#define		RCR_MFB_16K	(3U << RCR_MFB_SHIFT)
+
+#define	RCR_BITS	\
+	"\020"	\
+	"\010SO"	\
+	"\006AP"	\
+	"\005AM"	\
+	"\004AB"	\
+	"\003SEP"	\
+	"\002AMALL"	\
+	"\001PRO"
+
+/* Medium status register */
+#define	MSR_SM		0x1000	/* super mac support */
+#define	MSR_SBP		0x0800	/* stop backpressure */
+#define	MSR_PS		0x0200	/* port speed in mii mode */
+#define	MSR_RE		0x0100	/* rx enable */
+#define	MSR_PF		0x0080	/* check only length/type for pause frame */
+#define	MSR_JFE		0x0040	/* jumbo frame enable */
+#define	MSR_TFC		0x0020	/* tx flow control enable */
+#define	MSR_RFC		0x0010	/* rx flow control enable (178) */
+#define	MSR_FCEN	0x0010	/* flow control enable (172/772) */
+#define	MSR_ENCK	0x0008	/* Enable GTX_CLK and TXC clock output (178) */
+#define	MSR_TXABT	0x0004	/* Tx abort allow, always set */
+#define	MSR_FDPX	0x0002	/* full duplex */
+#define	MSR_GM		0x0001	/* Gigabit mode (178) */
+
+#define	MSR_BITS	\
+	"\020"	\
+	"\015SM"	\
+	"\014SBP"	\
+	"\012PS"	\
+	"\011RE"	\
+	"\005FCEN"	\
+	"\004ENCK"	\
+	"\003TXABT"	\
+	"\002FDPX"	\
+	"\001GM"
+
+/* monitor mode register */
+#define	MMR_RWMP	0x04	/* remote wakeup by magic pkt */
+#define	MMR_RWLU	0x02	/* remote wakeup by linkup */
+#define	MMR_MOM		0x01	/* monitor mode 1:en, 0:dis */
+
+#define	MMR_BITS	\
+	"\020"	\
+	"\003RWMP"	\
+	"\002RWLU"	\
+	"\001MOM"
+
+/* GPIO register */
+#define	GPIO_RSE	0x80	/* reload serial eeprom (88772)*/
+#define	GPIO_DATA2	0x20
+#define	GPIO_EN2	0x10
+#define	GPIO_DATA1	0x08
+#define	GPIO_EN1	0x04
+#define	GPIO_DATA0	0x02
+#define	GPIO_EN0	0x01
+
+#define	GPIO_BITS	\
+	"\020"		\
+	"\010RSE"	\
+	"\006DATA2"	\
+	"\005EN2"	\
+	"\004DATA1"	\
+	"\003EN1"	\
+	"\002DATA0"	\
+	"\001EN0"
+
+/* Software reset register */
+#define	SWRST_IPPD	0x40	/* internal phy power down control */
+#define	SWRST_IPRL	0x20	/* internal phy reset control */
+#define	SWRST_BZ	0x10	/* force Bulk In to return zero-length pkt */
+#define	SWRST_PRL	0x08	/* external phy reset pin level */
+#define	SWRST_PRTE	0x04	/* external phy tri-state enable */
+#define	SWRST_RT	0x02	/* clear frame length error for Bulk-Out */
+#define	SWRST_RR	0x01	/* clear frame length error for Bulk-In */
+
+#define	SWRST_BITS	\
+	"\020"		\
+	"\007IPPD"	\
+	"\006IPRL"	\
+	"\005BZ"	\
+	"\004PRL"	\
+	"\003PRTE"	\
+	"\002RT"	\
+	"\001RR"
+
+/* Software PHY Select Status register */
+#define	SPSS_ASEL	0x02	/* 1:auto select 0:manual select */
+#define	SPSS_PSEL	0x01	/* 1:intenal phy, 0:external (when ASEL=0) */
+
+#endif /* __AX88172_H__ */
diff --git a/usr/src/uts/common/io/axf/axf_usbgem.c b/usr/src/uts/common/io/axf/axf_usbgem.c
new file mode 100644
index 0000000000..28963f6849
--- /dev/null
+++ b/usr/src/uts/common/io/axf/axf_usbgem.c
@@ -0,0 +1,1539 @@
+/*
+ * axf_usbgem.c : ASIX AX88172/772 USB to Fast Ethernet Driver for Solaris
+ *
+ * Copyright (c) 2004-2012 Masayuki Murayama.  All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions are met:
+ *
+ * 1. Redistributions of source code must retain the above copyright notice,
+ *    this list of conditions and the following disclaimer.
+ *
+ * 2. Redistributions in binary form must reproduce the above copyright notice,
+ *    this list of conditions and the following disclaimer in the documentation
+ *    and/or other materials provided with the distribution.
+ *
+ * 3. Neither the name of the author nor the names of its contributors may be
+ *    used to endorse or promote products derived from this software without
+ *    specific prior written permission.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
+ * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
+ * COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
+ * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
+ * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS
+ * OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
+ * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+ * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
+ * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH
+ * DAMAGE.
+ */
+
+#pragma ident "@(#)axf_usbgem.c	1.3 12/02/09"
+
+/*
+ *  Changelog:
+ */
+
+/*
+ * TODO
+ * handle RXMODE_ENABLE in set_rx_filter()
+ */
+/* ======================================================= */
+
+/*
+ * Solaris system header files and macros
+ */
+
+/* minimum kernel headers for drivers */
+#include <sys/types.h>
+#include <sys/conf.h>
+#include <sys/debug.h>
+#include <sys/kmem.h>
+#include <sys/modctl.h>
+#include <sys/errno.h>
+#include <sys/ddi.h>
+#include <sys/sunddi.h>
+#include <sys/byteorder.h>
+
+/* ethernet stuff */
+#include <sys/ethernet.h>
+
+/* interface card depend stuff */
+#include <sys/stropts.h>
+#include <sys/stream.h>
+#include <sys/strlog.h>
+#include <sys/usb/usba.h>
+#include "usbgem.h"
+
+/* hardware stuff */
+#include "usbgem_mii.h"
+#include "ax88172reg.h"
+
+char	ident[] = "ax88x72 usbnic driver v" VERSION;
+
+/*
+ * Useful macros
+ */
+#define	CHECK_AND_JUMP(err, label)	if (err != USB_SUCCESS) goto label
+#define	LE16P(p)	((((uint8_t *)(p))[1] << 8) | ((uint8_t *)(p))[0])
+
+#define	AX88172(dp)	\
+	(((struct axf_dev *)(dp)->private)->chip->type == CHIP_TYPE_AX88172)
+
+#define	AX88772(dp)	\
+	(((struct axf_dev *)(dp)->private)->chip->type == CHIP_TYPE_AX88772)
+
+/*
+ * Debugging
+ */
+#ifdef DEBUG_LEVEL
+static int axf_debug = DEBUG_LEVEL;
+#define	DPRINTF(n, args)	if (axf_debug > (n)) cmn_err args
+#else
+#define	DPRINTF(n, args)
+#endif
+
+/*
+ * Our configration for ax88172
+ */
+/* timeouts */
+#define	ONESEC		(drv_usectohz(1*1000000))
+
+/*
+ * RX/TX buffer size
+ */
+
+/*
+ * Local device definitions
+ */
+struct chip_info {
+	uint16_t	vid;	/* usb vendor id */
+	uint16_t	pid;	/* usb product id */
+	int		type;
+	uint8_t		gpio_reset[2];
+	uint8_t		gpio_speed[2];
+	uint8_t		gpio_duplex[2];
+	char		*name;
+#define	CHIP_TYPE_AX88172	0
+#define	CHIP_TYPE_AX88772	1
+#define	CHIP_TYPE_AX88178	2
+};
+
+#define	GPIO_DEFAULT	{0x00, 0x15}, {0, 0}, {0, 0}
+struct chip_info chiptbl_88x7x[] = {
+/* AX88172 */
+{
+	/* Planex UE2-100TX, Hawking UF200, TrendNet TU2-ET100 */
+	0x07b8, 0x420a, CHIP_TYPE_AX88172,
+
+	/*
+	 * the default setting covers below:
+	 * gpio bit2 has to be 0 and gpio bit0 has to be 1
+	 */
+	{0, 0},
+	{GPIO_EN1, GPIO_DATA1 | GPIO_EN1},
+	{0, 0},
+	"Planex UE2-100TX",	/* tested */
+},
+{
+	0x2001, 0x1a00, CHIP_TYPE_AX88172,
+	{0x9f, 0x9e}, {0, 0}, {0, 0},
+	"D-Link dube100",	/* XXX */
+},
+{
+	0x077b, 0x2226, CHIP_TYPE_AX88172,
+	GPIO_DEFAULT,
+	"Linksys USB200M",
+},
+{
+	0x0846, 0x1040, CHIP_TYPE_AX88172,
+	GPIO_DEFAULT,
+	"Netgear FA120",
+},
+{
+	0x0b95, 0x1720, CHIP_TYPE_AX88172,
+	GPIO_DEFAULT,
+	"Intellinet, ST Lab USB Ethernet",
+},
+{
+	0x08dd, 0x90ff, CHIP_TYPE_AX88172,
+	GPIO_DEFAULT,
+	"Billionton Systems, USB2AR",
+},
+{
+	0x0557, 0x2009, CHIP_TYPE_AX88172,
+	GPIO_DEFAULT,
+	"ATEN UC210T",
+},
+{
+	0x0411, 0x003d, CHIP_TYPE_AX88172,
+	GPIO_DEFAULT,
+	"Buffalo LUA-U2-KTX",
+},
+{
+	0x6189, 0x182d, CHIP_TYPE_AX88172,
+	GPIO_DEFAULT,
+	"Sitecom LN-029 USB 2.0 10/100 Ethernet adapter",
+},
+{
+	0x07aa, 0x0017, CHIP_TYPE_AX88172,
+	GPIO_DEFAULT,
+	"corega FEther USB2-TX",
+},
+{
+	0x1189, 0x0893, CHIP_TYPE_AX88172,
+	GPIO_DEFAULT,
+	"Surecom EP-1427X-2",
+},
+{
+	0x1631, 0x6200, CHIP_TYPE_AX88172,
+	GPIO_DEFAULT,
+	"goodway corp usb gwusb2e",
+},
+/* AX88772 and AX88178 */
+{
+	0x13b1, 0x0018, CHIP_TYPE_AX88772,
+	{0, 0}, {0, 0}, {0, 0},
+	"Linksys USB200M rev.2",
+},
+{
+	0x1557, 0x7720, CHIP_TYPE_AX88772,
+	{0, 0}, {0, 0}, {0, 0},
+	"0Q0 cable ethernet",
+},
+{
+	0x07d1, 0x3c05, CHIP_TYPE_AX88772,
+	{0, 0}, {0, 0}, {0, 0},
+	"DLink DUB E100 ver B1",
+},
+{
+	0x2001, 0x3c05, CHIP_TYPE_AX88772,
+	{0, 0}, {0, 0}, {0, 0},
+	"DLink DUB E100 ver B1(2)",
+},
+{
+	0x05ac, 0x1402, CHIP_TYPE_AX88772,
+	{0, 0}, {0, 0}, {0, 0},
+	"Apple Ethernet USB Adapter",
+},
+{
+	0x1737, 0x0039, CHIP_TYPE_AX88178,
+	{0, 0}, {0, 0}, {0, 0},
+	"Linksys USB1000",
+},
+{
+	0x0411, 0x006e, CHIP_TYPE_AX88178,
+	{0, 0}, {0, 0}, {0, 0},
+	"Buffalo LUA-U2-KGT/LUA-U2-GT",
+},
+{
+	0x04bb, 0x0930, CHIP_TYPE_AX88178,
+	{0, 0}, {0, 0}, {0, 0},
+	"I/O DATA ETG-US2",
+},
+{
+	0x050d, 0x5055, CHIP_TYPE_AX88178,
+	{0, 0}, {0, 0}, {0, 0},
+	"Belkin F5D5055",
+},
+{
+	/* generic ax88772 must be the last entry */
+	/* planex UE-200TX-G */
+	0x0b95, 0x7720, CHIP_TYPE_AX88772,
+	{0, 0}, {0, 0}, {0, 0},
+	"ASIX AX88772/AX88178",	/* tested */
+},
+};
+
+#define	CHIPTABLESIZE	(sizeof (chiptbl_88x7x) / sizeof (struct chip_info))
+
+struct axf_dev {
+	/*
+	 * Misc HW information
+	 */
+	struct chip_info	*chip;
+	uint8_t			ipg[3];
+	uint8_t			gpio;
+	uint16_t		rcr;
+	uint16_t		msr;
+	uint8_t			last_link_state;
+	boolean_t		phy_has_reset;
+};
+
+/*
+ * private functions
+ */
+
+/* mii operations */
+static uint16_t axf_mii_read(struct usbgem_dev *, uint_t, int *errp);
+static void axf_mii_write(struct usbgem_dev *, uint_t, uint16_t, int *errp);
+
+/* nic operations */
+static int axf_reset_chip(struct usbgem_dev *);
+static int axf_init_chip(struct usbgem_dev *);
+static int axf_start_chip(struct usbgem_dev *);
+static int axf_stop_chip(struct usbgem_dev *);
+static int axf_set_media(struct usbgem_dev *);
+static int axf_set_rx_filter(struct usbgem_dev *);
+static int axf_get_stats(struct usbgem_dev *);
+static void  axf_interrupt(struct usbgem_dev *, mblk_t *);
+
+/* packet operations */
+static mblk_t *axf_tx_make_packet(struct usbgem_dev *, mblk_t *);
+static mblk_t *axf_rx_make_packet(struct usbgem_dev *, mblk_t *);
+
+/* =============================================================== */
+/*
+ * I/O functions
+ */
+/* =============================================================== */
+#define	OUT(dp, req, val, ix, len, buf, errp, label)	\
+	if ((*(errp) = usbgem_ctrl_out((dp), 	\
+	/* bmRequestType */ USB_DEV_REQ_HOST_TO_DEV	\
+		    | USB_DEV_REQ_TYPE_VENDOR | USB_DEV_REQ_RCPT_DEV,	\
+	/* bRequest */ (req),	\
+	/* wValue */   (val),	\
+	/* wIndex */   (ix),	\
+	/* wLength */  (len),	\
+	/* value */    (buf),	\
+	/* size */     (len))) != USB_SUCCESS) goto label
+
+#define	IN(dp, req, val, ix, len, buf, errp, label)	\
+	if ((*(errp) = usbgem_ctrl_in((dp), 	\
+	/* bmRequestType */ USB_DEV_REQ_DEV_TO_HOST	\
+		    | USB_DEV_REQ_TYPE_VENDOR | USB_DEV_REQ_RCPT_DEV,	\
+	/* bRequest */ (req),	\
+	/* wValue */   (val),	\
+	/* wIndex */   (ix),	\
+	/* wLength */  (len),	\
+	/* valuep */   (buf),	\
+	/* size */     (len))) != USB_SUCCESS) goto label
+
+/* =============================================================== */
+/*
+ * Hardware manupilation
+ */
+/* =============================================================== */
+static int
+axf_reset_phy(struct usbgem_dev *dp)
+{
+	uint8_t	phys[2];
+	uint8_t	val8;
+	int	err;
+	struct axf_dev	*lp = dp->private;
+
+	DPRINTF(2, (CE_CONT, "!%s: %s: called", dp->name, __func__));
+
+	if (AX88172(dp)) {
+		delay(drv_usectohz(5000));
+		IN(dp, VCMD_READ_GPIO, 0, 0, 1, &val8, &err, usberr);
+
+		DPRINTF(0, (CE_CONT, "!%s: %s: gpio 0x%b",
+		    dp->name, __func__, val8, GPIO_BITS));
+
+		/* reset MII PHY */
+		val8 = lp->chip->gpio_reset[1]
+		    | lp->chip->gpio_speed[dp->speed]
+		    | lp->chip->gpio_duplex[dp->full_duplex];
+
+		OUT(dp, VCMD_WRITE_GPIO,
+		    val8, 0, 0, NULL, &err, usberr);
+		delay(drv_usectohz(5000));
+
+		val8 = lp->chip->gpio_reset[0]
+		    | lp->chip->gpio_speed[dp->speed]
+		    | lp->chip->gpio_duplex[dp->full_duplex];
+
+		OUT(dp, VCMD_WRITE_GPIO,
+		    val8, 0, 0, NULL, &err, usberr);
+		delay(drv_usectohz(5000));
+	} else {
+		lp->gpio = GPIO_RSE | GPIO_DATA2 | GPIO_EN2;
+		OUT(dp, VCMD_WRITE_GPIO, lp->gpio, 0,
+		    0, NULL, &err, usberr);
+		drv_usecwait(1000);
+
+		OUT(dp, VCMD_WRITE_PHY_SELECT_88772,
+		    dp->mii_phy_addr == 16 ? 1 : 0, 0, 0, NULL, &err, usberr);
+
+		OUT(dp, VCMD_SOFTWARE_RESET_88772,
+		    SWRST_IPPD | SWRST_PRL, 0, 0, NULL, &err, usberr);
+		delay(drv_usectohz(150*1000));
+		OUT(dp, VCMD_SOFTWARE_RESET_88772,
+		    0, 0, 0, NULL, &err, usberr);
+
+		OUT(dp, VCMD_SOFTWARE_RESET_88772,
+		    dp->mii_phy_addr == 16 ? SWRST_IPRL : SWRST_PRTE,
+		    0, 0, NULL, &err, usberr);
+		delay(drv_usectohz(150*1000));
+	}
+
+
+	return (USB_SUCCESS);
+
+usberr:
+	return (USB_FAILURE);
+}
+
+static int
+axf_reset_chip(struct usbgem_dev *dp)
+{
+	int	err = USB_SUCCESS;
+
+	if (AX88172(dp)) {
+		/* there are no ways to reset nic */
+		return (USB_SUCCESS);
+	}
+#ifdef NEVER
+	OUT(dp, VCMD_SOFTWARE_RESET_88772,
+	    SWRST_RR | SWRST_RT, 0, 0, NULL, &err, usberr);
+	OUT(dp, VCMD_SOFTWARE_RESET_88772,
+	    0, 0, 0, NULL, &err, usberr);
+usberr:
+#endif
+	return (err);
+}
+
+/*
+ * Setup ax88172
+ */
+static int
+axf_init_chip(struct usbgem_dev *dp)
+{
+	int		i;
+	uint32_t	val;
+	int		err = USB_SUCCESS;
+	uint16_t	reg;
+	uint8_t		buf[2];
+	uint16_t	tmp16;
+	struct axf_dev	*lp = dp->private;
+
+	DPRINTF(2, (CE_CONT, "!%s: %s: called", dp->name, __func__));
+
+	/* rx conrol register: read default value */
+	if (!AX88172(dp)) {
+		/* clear rx control */
+		OUT(dp, VCMD_WRITE_RXCTRL, 0, 0, 0, NULL, &err, usberr);
+	}
+
+	IN(dp, VCMD_READ_RXCTRL, 0, 0, 2, buf, &err, usberr);
+	lp->rcr = LE16P(buf);
+	DPRINTF(0, (CE_CONT, "!%s: %s: rcr(default):%b",
+	    dp->name, __func__, lp->rcr, RCR_BITS));
+
+	lp->rcr &= ~RCR_SO;
+
+	/* Media status register */
+	if (AX88172(dp)) {
+#ifdef notdef
+		lp->msr = MSR_TXABT;
+#else
+		lp->msr = 0;
+#endif
+	} else {
+		lp->msr = MSR_RE | MSR_TXABT;
+	}
+	DPRINTF(0, (CE_CONT, "!%s: %s: msr:%b",
+	    dp->name, __func__, lp->msr, MSR_BITS));
+	err = axf_set_media(dp);
+	CHECK_AND_JUMP(err, usberr);
+
+	/* write IPG0-2 registers */
+	if (AX88172(dp)) {
+		OUT(dp, VCMD_WRITE_IPG, lp->ipg[0], 0, 0, NULL, &err, usberr);
+		OUT(dp, VCMD_WRITE_IPG1, lp->ipg[1], 0, 0, NULL, &err, usberr);
+		OUT(dp, VCMD_WRITE_IPG2, lp->ipg[2], 0, 0, NULL, &err, usberr);
+	} else {
+		/* EMPTY */
+	}
+#ifdef ENABLE_RX_IN_INIT_CHIP
+	/* enable Rx */
+	lp->rcr |= RCR_SO;
+	OUT(dp, VCMD_WRITE_RXCTRL, lp->rcr, 0, 0, NULL, &err, usberr);
+#endif
+usberr:
+	DPRINTF(2, (CE_CONT, "!%s: %s: end (%s)",
+	    dp->name, __func__,
+	    err, err == USB_SUCCESS ? "success" : "error"));
+	return (err);
+}
+
+static int
+axf_start_chip(struct usbgem_dev *dp)
+{
+	int	err = USB_SUCCESS;
+	struct axf_dev	*lp = dp->private;
+#ifndef ENABLE_RX_IN_INIT_CHIP
+	/* enable Rx */
+	lp->rcr |= RCR_SO;
+	OUT(dp, VCMD_WRITE_RXCTRL, lp->rcr, 0, 0, NULL, &err, usberr);
+
+usberr:
+	DPRINTF(2, (CE_CONT, "!%s: %s: end (%s)",
+	    dp->name, __func__,
+	    err, err == USB_SUCCESS ? "success" : "error"));
+#endif
+	return (err);
+}
+
+static int
+axf_stop_chip(struct usbgem_dev *dp)
+{
+	int	err = USB_SUCCESS;
+	struct axf_dev	*lp = dp->private;
+
+	/* Disable Rx */
+	lp->rcr &= ~RCR_SO;
+	OUT(dp, VCMD_WRITE_RXCTRL, lp->rcr, 0, 0, NULL, &err, usberr);
+
+	/*
+	 * Restore factory mac address
+	 * if we have changed current mac address
+	 */
+	if (!AX88172(dp) &&
+	    bcmp(dp->dev_addr.ether_addr_octet,
+	    dp->cur_addr.ether_addr_octet,
+	    ETHERADDRL) != 0) {
+		OUT(dp, VCMD_WRITE_NODE_ID_88772, 0, 0,
+		    ETHERADDRL, dp->cur_addr.ether_addr_octet, &err, usberr);
+	}
+usberr:
+	return (axf_reset_chip(dp));
+}
+
+static int
+axf_get_stats(struct usbgem_dev *dp)
+{
+	/* EMPTY */
+	return (USB_SUCCESS);
+}
+
+static uint_t
+axf_mcast_hash(struct usbgem_dev *dp, const uint8_t *addr)
+{
+	return (usbgem_ether_crc_be(addr) >> (32 - 6));
+}
+
+static int
+axf_set_rx_filter(struct usbgem_dev *dp)
+{
+	int		i;
+	uint8_t		mode;
+	uint8_t		mhash[8];
+	uint8_t		buf[2];
+	uint_t		h;
+	int		err = USB_SUCCESS;
+	struct axf_dev	*lp = dp->private;
+
+	DPRINTF(2, (CE_CONT, "!%s: %s: called, rxmode:%x",
+	    dp->name, __func__, dp->rxmode));
+
+	if (lp->rcr & RCR_SO) {
+		/* set promiscuous mode  before changing it. */
+		OUT(dp, VCMD_WRITE_RXCTRL,
+		    lp->rcr | RCR_PRO, 0, 0, NULL, &err, usberr);
+	}
+
+	lp->rcr &= ~(RCR_AP_88772 | RCR_AM | RCR_SEP | RCR_AMALL | RCR_PRO);
+	mode = RCR_AB;	/* accept broadcast packets */
+
+	bzero(mhash, sizeof (mhash));
+
+	if (dp->rxmode & RXMODE_PROMISC) {
+		/* promiscious mode implies all multicast and all physical */
+		mode |= RCR_PRO;
+	} else if ((dp->rxmode & RXMODE_ALLMULTI) || dp->mc_count > 32) {
+		/* accept all multicast packets */
+		mode |= RCR_AMALL;
+	} else if (dp->mc_count > 0) {
+		/*
+		 * make hash table to select interresting
+		 * multicast address only.
+		 */
+		mode |= RCR_AM;
+		for (i = 0; i < dp->mc_count; i++) {
+			h = dp->mc_list[i].hash;
+			mhash[h / 8] |= 1 << (h % 8);
+		}
+	}
+	if (AX88172(dp)) {
+		if (bcmp(dp->dev_addr.ether_addr_octet,
+		    dp->cur_addr.ether_addr_octet, ETHERADDRL) != 0) {
+			/*
+			 * we use promiscious mode instead of changing the
+			 * mac address in ax88172
+			 */
+			mode |= RCR_PRO;
+		}
+	} else {
+		OUT(dp, VCMD_WRITE_NODE_ID_88772, 0, 0,
+		    ETHERADDRL, dp->cur_addr.ether_addr_octet, &err, usberr);
+	}
+	lp->rcr |= mode;
+
+	/* set multicast hash table */
+	if (mode & RCR_AM) {
+		/* need to set up multicast hash table */
+		OUT(dp, VCMD_WRITE_MCAST_FILTER, 0, 0,
+		    sizeof (mhash), mhash, &err, usberr);
+	}
+
+	/* update rcr */
+	OUT(dp, VCMD_WRITE_RXCTRL, lp->rcr, 0,
+	    0, NULL, &err, usberr);
+
+#if DEBUG_LEVEL > 1
+	/* verify rxctrl reg */
+	IN(dp, VCMD_READ_RXCTRL, 0, 0, 2, buf, &err, usberr);
+	cmn_err(CE_CONT, "!%s: %s: rcr:%b returned",
+	    dp->name, __func__, LE16P(buf), RCR_BITS);
+#endif
+usberr:
+	DPRINTF(2, (CE_CONT, "!%s: %s: end (%s)",
+	    dp->name, __func__,
+	    err, err == USB_SUCCESS ? "success" : "error"));
+	return (err);
+}
+
+static int
+axf_set_media(struct usbgem_dev *dp)
+{
+	uint8_t	val8;
+	uint8_t	gpio;
+	uint8_t	gpio_old;
+	int	err = USB_SUCCESS;
+	uint16_t	msr;
+	struct axf_dev	*lp = dp->private;
+
+	IN(dp, VCMD_READ_GPIO, 0, 0, 1, &gpio, &err, usberr);
+
+	DPRINTF(0, (CE_CONT, "!%s: %s: called, gpio:%b",
+	    dp->name, __func__, gpio, GPIO_BITS));
+
+	msr = lp->msr;
+	gpio_old = gpio;
+	gpio = lp->chip->gpio_reset[0];
+
+	/* setup speed */
+	if (AX88172(dp)) {
+		/* EMPTY */
+	} else {
+		msr &= ~(MSR_PS | MSR_GM | MSR_ENCK);
+
+		switch (dp->speed) {
+		case USBGEM_SPD_1000:
+			msr |= MSR_GM | MSR_ENCK;
+			break;
+
+		case USBGEM_SPD_100:
+			msr |= MSR_PS;
+			break;
+
+		case USBGEM_SPD_10:
+			break;
+		}
+	}
+	gpio |= lp->chip->gpio_speed[dp->speed == USBGEM_SPD_100 ? 1 : 0];
+
+	/* select duplex */
+	msr &= ~MSR_FDPX;
+	if (dp->full_duplex) {
+		msr |= MSR_FDPX;
+
+		/* select flow control */
+		if (AX88172(dp)) {
+			msr &= ~MSR_FCEN;
+			switch (dp->flow_control) {
+			case FLOW_CONTROL_TX_PAUSE:
+			case FLOW_CONTROL_SYMMETRIC:
+			case FLOW_CONTROL_RX_PAUSE:
+				msr |= MSR_FCEN;
+				break;
+			}
+		} else {
+			msr &= ~(MSR_RFC | MSR_TFC);
+			switch (dp->flow_control) {
+			case FLOW_CONTROL_TX_PAUSE:
+				msr |= MSR_TFC;
+				break;
+
+			case FLOW_CONTROL_SYMMETRIC:
+				msr |= MSR_TFC | MSR_RFC;
+				break;
+
+			case FLOW_CONTROL_RX_PAUSE:
+				msr |= MSR_RFC;
+				break;
+			}
+		}
+	}
+	gpio |= lp->chip->gpio_duplex[dp->full_duplex ? 1 : 0];
+
+	/* update medium status register */
+	lp->msr = msr;
+	OUT(dp, VCMD_WRITE_MEDIUM_STATUS, lp->msr, 0,
+	    0, NULL, &err, usberr);
+
+	if (gpio != gpio_old) {
+		/* LED control required for some products */
+		OUT(dp, VCMD_WRITE_GPIO,
+		    gpio, 0, 0, NULL, &err, usberr);
+	}
+
+usberr:
+	DPRINTF(2, (CE_CONT, "!%s: %s: end (%s)",
+	    dp->name, __func__,
+	    err, err == USB_SUCCESS ? "success" : "error"));
+	return (err);
+}
+
+#define	FILL_PKT_HEADER(bp, len)	{	\
+	(bp)[0] = (uint8_t)(len);	\
+	(bp)[1] = (uint8_t)((len) >> 8);	\
+	(bp)[2] = (uint8_t)(~(len));	\
+	(bp)[3] = (uint8_t)((~(len)) >> 8);	\
+}
+
+#define	PKT_HEADER_SIZE	4
+
+/*
+ * send/receive packet check
+ */
+static mblk_t *
+axf_tx_make_packet(struct usbgem_dev *dp, mblk_t *mp)
+{
+	int		n;
+	size_t		len;
+	size_t		pkt_size;
+	mblk_t		*new;
+	mblk_t		*tp;
+	uint8_t		*bp;
+	uint8_t		*last_pos;
+	uint_t		align_mask;
+	size_t		header_size;
+	int		pad_size;
+
+	len = msgdsize(mp);
+
+	if (AX88172(dp)) {
+#ifdef notdef
+		align_mask = 63;
+#else
+		align_mask = 511;
+#endif
+		header_size = 0;
+
+		if (len >= ETHERMIN && mp->b_cont == NULL &&
+		    (len & align_mask) != 0) {
+			/* use the mp "as is" */
+			return (mp);
+		}
+	} else {
+		align_mask = 511;
+		header_size = PKT_HEADER_SIZE;
+	}
+
+	/*
+	 * re-allocate the mp
+	 */
+	/* minimum ethernet packet size of ETHERMIN */
+	pkt_size = max(len, ETHERMIN);
+
+	if (((pkt_size + header_size) & align_mask) == 0) {
+		/* padding is required in usb communication */
+		pad_size = PKT_HEADER_SIZE;
+	} else {
+		pad_size = 0;
+	}
+
+	if ((new = allocb(header_size + pkt_size + pad_size, 0)) == NULL) {
+		return (NULL);
+	}
+
+	bp = new->b_rptr;
+	if (header_size) {
+		uint16_t	tmp;
+
+		/* add a header */
+		tmp = (uint16_t)pkt_size;
+		FILL_PKT_HEADER(bp, tmp);
+		bp += header_size;
+	}
+
+	/* copy contents of the buffer */
+	for (tp = mp; tp; tp = tp->b_cont) {
+		n = tp->b_wptr - tp->b_rptr;
+		bcopy(tp->b_rptr, bp, n);
+		bp += n;
+	}
+
+	/* add pads for ethernet packets */
+	last_pos = new->b_rptr + header_size + pkt_size;
+	while (bp < last_pos) {
+		*bp++ = 0;
+	}
+
+	/* add a zero-length pad segment for usb communications */
+	if (pad_size) {
+		/* add a dummy header for zero-length packet */
+		FILL_PKT_HEADER(bp, 0);
+		bp += pad_size;
+	}
+
+	/* close the payload of the packet */
+	new->b_wptr = bp;
+
+	return (new);
+}
+
+static void
+axf_dump_packet(struct usbgem_dev *dp, uint8_t *bp, int n)
+{
+	int	i;
+
+	for (i = 0; i < n; i += 8, bp += 8) {
+		cmn_err(CE_CONT, "%02x %02x %02x %02x %02x %02x %02x %02x",
+		    bp[0], bp[1], bp[2], bp[3], bp[4], bp[5], bp[6], bp[7]);
+	}
+}
+
+static mblk_t *
+axf_rx_make_packet(struct usbgem_dev *dp, mblk_t *mp)
+{
+	mblk_t	*tp;
+	int	rest;
+
+	if (AX88172(dp)) {
+		return (mp);
+	}
+
+	tp = mp;
+	rest = tp->b_wptr - tp->b_rptr;
+
+	if (rest <= PKT_HEADER_SIZE) {
+		/*
+		 * the usb bulk-in frame doesn't include any valid
+		 * ethernet packets.
+		 */
+		return (NULL);
+	}
+
+	for (; ; ) {
+		uint16_t	len;
+		uint16_t	cksum;
+
+		/* analyse the header of the received usb frame */
+		len = LE16P(tp->b_rptr + 0);
+		cksum = LE16P(tp->b_rptr + 2);
+
+		/* test if the header is valid */
+		if (len + cksum != 0xffff) {
+			/* discard whole the packet */
+			cmn_err(CE_WARN,
+			    "!%s: %s: corrupted header:%04x %04x",
+			    dp->name, __func__, len, cksum);
+			return (NULL);
+		}
+#if DEBUG_LEVEL > 0
+		if (len < ETHERMIN || len > ETHERMAX) {
+			cmn_err(CE_NOTE,
+			    "!%s: %s: incorrect pktsize:%d",
+			    dp->name, __func__, len);
+		}
+#endif
+		/* extract a ethernet packet from the bulk-in frame */
+		tp->b_rptr += PKT_HEADER_SIZE;
+		tp->b_wptr = tp->b_rptr + len;
+
+		if (len & 1) {
+			/*
+			 * skip a tailing pad byte if the packet
+			 * length is odd
+			 */
+			len++;
+		}
+		rest -= len + PKT_HEADER_SIZE;
+
+		if (rest <= PKT_HEADER_SIZE) {
+			/* no more vaild ethernet packets */
+			break;
+		}
+
+#if DEBUG_LEVEL > 10
+		axf_dump_packet(dp, tp->b_wptr, 18);
+#endif
+		/* allocate a mblk_t header for the next ethernet packet */
+		tp->b_next = dupb(mp);
+		tp->b_next->b_rptr = tp->b_rptr + len;
+		tp = tp->b_next;
+	}
+
+	return (mp);
+}
+
+/*
+ * MII Interfaces
+ */
+static uint16_t
+axf_mii_read(struct usbgem_dev *dp, uint_t index, int *errp)
+{
+	uint8_t		buf[2];
+	uint16_t	val;
+
+	DPRINTF(4, (CE_CONT, "!%s: %s: called, ix:%d",
+	    dp->name, __func__, index));
+
+	/* switch to software MII operation mode */
+	OUT(dp, VCMD_SOFTWARE_MII_OP, 0, 0, 0, NULL, errp, usberr);
+
+	/* Read MII register */
+	IN(dp, VCMD_READ_MII_REG, dp->mii_phy_addr, index,
+	    2, buf, errp, usberr);
+
+	/* switch to hardware MII operation mode */
+	OUT(dp, VCMD_HARDWARE_MII_OP, 0, 0, 0, NULL, errp, usberr);
+
+	return (LE16P(buf));
+
+usberr:
+	cmn_err(CE_CONT,
+	    "!%s: %s: usberr(%d) detected", dp->name, __func__, *errp);
+	return (0);
+}
+
+static void
+axf_mii_write(struct usbgem_dev *dp, uint_t index, uint16_t val, int *errp)
+{
+	uint8_t		buf[2];
+
+	DPRINTF(4, (CE_CONT, "!%s: %s called, reg:%x val:%x",
+	    dp->name, __func__, index, val));
+
+	/* switch software MII operation mode */
+	OUT(dp, VCMD_SOFTWARE_MII_OP, 0, 0, 0, NULL, errp, usberr);
+
+	/* Write to the specified MII register */
+	buf[0] = (uint8_t)val;
+	buf[1] = (uint8_t)(val >> 8);
+	OUT(dp, VCMD_WRITE_MII_REG, dp->mii_phy_addr, index,
+	    2, buf, errp, usberr);
+
+	/* switch to hardware MII operation mode */
+	OUT(dp, VCMD_HARDWARE_MII_OP, 0, 0, 0, NULL, errp, usberr);
+
+usberr:
+	;
+}
+
+static void
+axf_interrupt(struct usbgem_dev *dp, mblk_t *mp)
+{
+	uint8_t	*bp;
+	struct axf_dev	*lp = dp->private;
+
+	bp = mp->b_rptr;
+
+	DPRINTF(2, (CE_CONT,
+	    "!%s: %s: size:%d, %02x %02x %02x %02x %02x %02x %02x %02x",
+	    dp->name, __func__, mp->b_wptr - mp->b_rptr,
+	    bp[0], bp[1], bp[2], bp[3], bp[4], bp[5], bp[6], bp[7]));
+
+	if (lp->last_link_state ^ bp[2]) {
+		usbgem_mii_update_link(dp);
+	}
+
+	lp->last_link_state = bp[2];
+}
+
+/* ======================================================== */
+/*
+ * OS depend (device driver DKI) routine
+ */
+/* ======================================================== */
+#ifdef DEBUG_LEVEL
+static void
+axf_eeprom_dump(struct usbgem_dev *dp, int size)
+{
+	int	i;
+	int	err;
+	uint8_t	w0[2], w1[2], w2[2], w3[2];
+
+	cmn_err(CE_CONT, "!%s: eeprom dump:", dp->name);
+
+	err = USB_SUCCESS;
+
+	for (i = 0; i < size; i += 4) {
+		IN(dp, VCMD_READ_SROM, i + 0, 0, 2, w0, &err, usberr);
+		IN(dp, VCMD_READ_SROM, i + 1, 0, 2, w1, &err, usberr);
+		IN(dp, VCMD_READ_SROM, i + 2, 0, 2, w2, &err, usberr);
+		IN(dp, VCMD_READ_SROM, i + 3, 0, 2, w3, &err, usberr);
+		cmn_err(CE_CONT, "!0x%02x: 0x%04x 0x%04x 0x%04x 0x%04x",
+		    i,
+		    (w0[1] << 8) | w0[0],
+		    (w1[1] << 8) | w1[0],
+		    (w2[1] << 8) | w2[0],
+		    (w3[1] << 8) | w3[0]);
+	}
+usberr:
+	;
+}
+#endif
+
+static int
+axf_attach_chip(struct usbgem_dev *dp)
+{
+	uint8_t	phys[2];
+	int	err;
+	uint_t	vcmd;
+	int	ret;
+#ifdef CONFIG_FULLSIZE_VLAN
+	uint8_t	maxpktsize[2];
+	uint16_t	vlan_pktsize;
+#endif
+#ifdef DEBUG_LEVEL
+	uint8_t	val8;
+#endif
+	struct axf_dev	*lp = dp->private;
+
+	DPRINTF(0, (CE_CONT, "!%s: %s enter", dp->name, __func__));
+
+	ret = USB_SUCCESS;
+	/*
+	 * mac address in EEPROM has loaded to ID registers.
+	 */
+	vcmd = AX88172(dp) ? VCMD_READ_NODE_ID : VCMD_READ_NODE_ID_88772;
+	IN(dp, vcmd, 0, 0,
+	    ETHERADDRL, dp->dev_addr.ether_addr_octet, &err, usberr);
+
+	/*
+	 * setup IPG values
+	 */
+	lp->ipg[0] = 0x15;
+	lp->ipg[1] = 0x0c;
+	lp->ipg[2] = 0x12;
+
+	/*
+	 * We cannot scan phy because the nic returns undefined
+	 * value, i.e. remained garbage, when MII phy is not at the
+	 * specified index.
+	 */
+#ifdef DEBUG_LEVELx
+	if (lp->chip->vid == 0x07b8 && lp->chip->pid == 0x420a) {
+		/*
+		 * restore the original phy address of brain
+		 * damaged Planex UE2-100TX
+		 */
+		OUT(dp, VCMD_WRITE_SROM_ENABLE, 0, 0, 0, NULL, &err, usberr);
+		OUT(dp, VCMD_WRITE_SROM, 0x11, 0xe004, 0, NULL, &err, usberr);
+		OUT(dp, VCMD_WRITE_SROM_DISABLE, 0, 0, 0, NULL, &err, usberr);
+	}
+#endif
+	if (AX88172(dp)) {
+		IN(dp, VCMD_READ_PHY_IDS, 0, 0, 2, &phys, &err, usberr);
+		dp->mii_phy_addr = phys[1];
+		DPRINTF(0, (CE_CONT, "!%s: %s: phys_addr:%d %d",
+		    dp->name, __func__, phys[0], phys[1]));
+	} else {
+		/* use built-in phy */
+		dp->mii_phy_addr = 0x10;
+	}
+
+	dp->misc_flag |= USBGEM_VLAN;
+#ifdef CONFIG_FULLSIZE_VLAN
+	if (AX88172(dp) || AX88772(dp)) {
+		/* check max packet size in srom */
+		IN(dp, VCMD_READ_SROM, 0x10, 0, 2, maxpktsize, &err, usberr);
+		vlan_pktsize = ETHERMAX + ETHERFCSL + 4 /* VTAG_SIZE */;
+
+		if (LE16P(maxpktsize) < vlan_pktsize) {
+			cmn_err(CE_CONT,
+			    "!%s: %s: max packet size in srom is too small, "
+			    "changing %d -> %d, do power cycle for the device",
+			    dp->name, __func__,
+			    LE16P(maxpktsize), vlan_pktsize);
+			OUT(dp, VCMD_WRITE_SROM_ENABLE,
+			    0, 0, 0, NULL, &err, usberr);
+			OUT(dp, VCMD_WRITE_SROM, 0x10,
+			    vlan_pktsize, 0, NULL, &err, usberr);
+			OUT(dp, VCMD_WRITE_SROM_DISABLE,
+			    0, 0, 0, NULL, &err, usberr);
+
+			/* need to power off the device */
+			ret = USB_FAILURE;
+		}
+	}
+#endif
+#ifdef DEBUG_LEVEL
+	IN(dp, VCMD_READ_GPIO, 0, 0, 1, &val8, &err, usberr);
+	cmn_err(CE_CONT,
+	    "!%s: %s: ipg 0x%02x 0x%02x 0x%02x, gpio 0x%b",
+	    dp->name, __func__, lp->ipg[0], lp->ipg[1], lp->ipg[2],
+	    val8, GPIO_BITS);
+#endif
+	/* fix rx buffer size */
+	if (!AX88172(dp)) {
+		dp->rx_buf_len = 2048;
+	}
+
+#if DEBUG_LEVEL > 0
+	axf_eeprom_dump(dp, 0x20);
+#endif
+	return (ret);
+
+usberr:
+	cmn_err(CE_WARN, "%s: %s: usb error detected (%d)",
+	    dp->name, __func__, err);
+	return (USB_FAILURE);
+}
+
+static boolean_t
+axf_scan_phy(struct usbgem_dev *dp)
+{
+	int	i;
+	int	err;
+	uint16_t	val;
+	int	phy_addr_saved;
+	struct axf_dev	*lp = dp->private;
+
+	DPRINTF(0, (CE_CONT, "!%s: %s: called", dp->name, __func__));
+
+	phy_addr_saved = dp->mii_phy_addr;
+
+	/* special probe routine for unreliable MII addr */
+#define	PROBE_PAT	\
+	(MII_ABILITY_100BASE_TX_FD |	\
+	MII_ABILITY_100BASE_TX |	\
+	MII_ABILITY_10BASE_T_FD |	\
+	MII_ABILITY_10BASE_T)
+
+	for (i = 0; i < 32; i++) {
+		dp->mii_phy_addr = i;
+		axf_mii_write(dp, MII_AN_ADVERT, 0, &err);
+		if (err != USBGEM_SUCCESS) {
+			break;
+		}
+		val = axf_mii_read(dp, MII_AN_ADVERT, &err);
+		if (err != USBGEM_SUCCESS) {
+			break;
+		}
+		if (val != 0) {
+			DPRINTF(0, (CE_CONT, "!%s: %s: index:%d,  val %b != 0",
+			    dp->name, __func__, i, val, MII_ABILITY_BITS));
+			continue;
+		}
+
+		axf_mii_write(dp, MII_AN_ADVERT, PROBE_PAT, &err);
+		if (err != USBGEM_SUCCESS) {
+			break;
+		}
+		val = axf_mii_read(dp, MII_AN_ADVERT, &err);
+		if (err != USBGEM_SUCCESS) {
+			break;
+		}
+		if ((val & MII_ABILITY_TECH) != PROBE_PAT) {
+			DPRINTF(0, (CE_CONT, "!%s: %s: "
+			    "index:%d,  pat:%x != val:%b",
+			    dp->name, __func__, i,
+			    PROBE_PAT, val, MII_ABILITY_BITS));
+			continue;
+		}
+
+		/* found */
+		dp->mii_phy_addr = phy_addr_saved;
+		return (i);
+	}
+#undef PROBE_PAT
+	if (i == 32) {
+		cmn_err(CE_CONT, "!%s: %s: no mii phy found",
+		    dp->name, __func__);
+	} else {
+		cmn_err(CE_CONT, "!%s: %s: i/o error while scanning phy",
+		    dp->name, __func__);
+	}
+	dp->mii_phy_addr = phy_addr_saved;
+	return (-1);
+}
+
+static int
+axf_mii_probe(struct usbgem_dev *dp)
+{
+	int	my_guess;
+	int	err;
+	uint8_t	old_11th[2];
+	uint8_t	new_11th[2];
+	struct axf_dev	*lp = dp->private;
+
+	DPRINTF(0, (CE_CONT, "!%s: %s: called", dp->name, __func__));
+	(void) axf_reset_phy(dp);
+	lp->phy_has_reset = B_TRUE;
+
+	if (AX88172(dp)) {
+		my_guess = axf_scan_phy(dp);
+		if (my_guess >= 0 && my_guess < 32 &&
+		    my_guess != dp->mii_phy_addr) {
+			/*
+			 * phy addr in srom is wrong, need to fix it
+			 */
+			IN(dp, VCMD_READ_SROM,
+			    0x11, 0, 2, old_11th, &err, usberr);
+
+			new_11th[0] = my_guess;
+			new_11th[1] = old_11th[1];
+
+			OUT(dp, VCMD_WRITE_SROM_ENABLE,
+			    0, 0, 0, NULL, &err, usberr);
+			OUT(dp, VCMD_WRITE_SROM,
+			    0x11, LE16P(new_11th), 0, NULL, &err, usberr);
+			OUT(dp, VCMD_WRITE_SROM_DISABLE,
+			    0, 0, 0, NULL, &err, usberr);
+#if 1
+			/* XXX - read back, but it doesn't work, why? */
+			delay(drv_usectohz(1000*1000));
+			IN(dp, VCMD_READ_SROM,
+			    0x11, 0, 2, new_11th, &err, usberr);
+#endif
+			cmn_err(CE_NOTE, "!%s: %s: phy addr in srom fixed: "
+			    "%04x -> %04x",
+			    dp->name, __func__,
+			    LE16P(old_11th), LE16P(new_11th));
+			return (USBGEM_FAILURE);
+usberr:
+			cmn_err(CE_NOTE,
+			    "!%s: %s:  failed to patch phy addr, "
+			    "current: %04x",
+			    dp->name, __func__, LE16P(old_11th));
+			return (USBGEM_FAILURE);
+		}
+	}
+	return (usbgem_mii_probe_default(dp));
+}
+
+static int
+axf_mii_init(struct usbgem_dev *dp)
+{
+	struct axf_dev	*lp = dp->private;
+
+	DPRINTF(2, (CE_CONT, "!%s: %s: called", dp->name, __func__));
+
+	if (!lp->phy_has_reset) {
+		(void) axf_reset_phy(dp);
+	}
+
+	/* prepare to reset phy on the next reconnect or resume */
+	lp->phy_has_reset = B_FALSE;
+
+	return (USB_SUCCESS);
+}
+
+static int
+axfattach(dev_info_t *dip, ddi_attach_cmd_t cmd)
+{
+	int			i;
+	ddi_iblock_cookie_t	c;
+	int			ret;
+	int			revid;
+	int			unit;
+	int			vid;
+	int			pid;
+	struct chip_info	*p;
+	int			len;
+	const char		*drv_name;
+	struct usbgem_dev	*dp;
+	void			*base;
+	struct usbgem_conf	*ugcp;
+	struct axf_dev		*lp;
+
+	unit = ddi_get_instance(dip);
+	drv_name = ddi_driver_name(dip);
+
+	DPRINTF(3, (CE_CONT, "!%s%d: %s: called, cmd:%d",
+	    drv_name, unit, __func__, cmd));
+
+	if (cmd == DDI_ATTACH) {
+		/*
+		 * Check if the chip is supported.
+		 */
+		vid = ddi_prop_get_int(DDI_DEV_T_ANY, dip, DDI_PROP_DONTPASS,
+		    "usb-vendor-id", -1);
+		pid = ddi_prop_get_int(DDI_DEV_T_ANY, dip, DDI_PROP_DONTPASS,
+		    "usb-product-id", -1);
+		revid = ddi_prop_get_int(DDI_DEV_T_ANY, dip, DDI_PROP_DONTPASS,
+		    "usb-revision-id", -1);
+
+		for (i = 0, p = chiptbl_88x7x; i < CHIPTABLESIZE; i++, p++) {
+			if (p->vid == vid && p->pid == pid) {
+				/* found */
+				cmn_err(CE_CONT, "!%s%d: %s "
+				    "(vid: 0x%04x, did: 0x%04x, revid: 0x%02x)",
+				    drv_name, unit, p->name, vid, pid, revid);
+				goto chip_found;
+			}
+		}
+
+		/* Not found */
+		cmn_err(CE_WARN, "!%s: %s: wrong usb venid/prodid (0x%x, 0x%x)",
+		    drv_name, __func__, vid, pid);
+
+		/* assume 88772 */
+		p = &chiptbl_88x7x[CHIPTABLESIZE - 1];
+chip_found:
+		/*
+		 * construct usbgem configration
+		 */
+		ugcp = kmem_zalloc(sizeof (*ugcp), KM_SLEEP);
+
+		/* name */
+		/*
+		 * softmac requires that ppa is the instance number
+		 * of the device, otherwise it hangs in seaching the device.
+		 */
+		sprintf(ugcp->usbgc_name, "%s%d", drv_name, unit);
+		ugcp->usbgc_ppa = unit;
+
+		ugcp->usbgc_ifnum = 0;
+		ugcp->usbgc_alt = 0;
+
+		ugcp->usbgc_tx_list_max = 64;
+
+		ugcp->usbgc_rx_header_len = 0;
+		ugcp->usbgc_rx_list_max = 64;
+
+		/* time out parameters */
+		ugcp->usbgc_tx_timeout = USBGEM_TX_TIMEOUT;
+		ugcp->usbgc_tx_timeout_interval = USBGEM_TX_TIMEOUT_INTERVAL;
+
+		/* flow control */
+		/*
+		 * XXX - flow control caused link down frequently under
+		 * heavy traffic
+		 */
+		ugcp->usbgc_flow_control = FLOW_CONTROL_RX_PAUSE;
+
+		/* MII timeout parameters */
+		ugcp->usbgc_mii_link_watch_interval = ONESEC;
+		ugcp->usbgc_mii_an_watch_interval = ONESEC/5;
+		ugcp->usbgc_mii_reset_timeout = MII_RESET_TIMEOUT; /* 1 sec */
+		ugcp->usbgc_mii_an_timeout = MII_AN_TIMEOUT;	/* 5 sec */
+		ugcp->usbgc_mii_an_wait = 0;
+		ugcp->usbgc_mii_linkdown_timeout = MII_LINKDOWN_TIMEOUT;
+
+		ugcp->usbgc_mii_an_delay = ONESEC/10;
+		ugcp->usbgc_mii_linkdown_action = MII_ACTION_RSA;
+		ugcp->usbgc_mii_linkdown_timeout_action = MII_ACTION_RESET;
+		ugcp->usbgc_mii_dont_reset = B_FALSE;
+		ugcp->usbgc_mii_hw_link_detection = B_TRUE;
+		ugcp->usbgc_mii_stop_mac_on_linkdown = B_FALSE;
+
+		/* I/O methods */
+
+		/* mac operation */
+		ugcp->usbgc_attach_chip = &axf_attach_chip;
+		ugcp->usbgc_reset_chip = &axf_reset_chip;
+		ugcp->usbgc_init_chip = &axf_init_chip;
+		ugcp->usbgc_start_chip = &axf_start_chip;
+		ugcp->usbgc_stop_chip = &axf_stop_chip;
+		ugcp->usbgc_multicast_hash = &axf_mcast_hash;
+
+		ugcp->usbgc_set_rx_filter = &axf_set_rx_filter;
+		ugcp->usbgc_set_media = &axf_set_media;
+		ugcp->usbgc_get_stats = &axf_get_stats;
+		ugcp->usbgc_interrupt = &axf_interrupt;
+
+		/* packet operation */
+		ugcp->usbgc_tx_make_packet = &axf_tx_make_packet;
+		ugcp->usbgc_rx_make_packet = &axf_rx_make_packet;
+
+		/* mii operations */
+		ugcp->usbgc_mii_probe = &axf_mii_probe;
+		ugcp->usbgc_mii_init = &axf_mii_init;
+		ugcp->usbgc_mii_config = &usbgem_mii_config_default;
+		ugcp->usbgc_mii_read = &axf_mii_read;
+		ugcp->usbgc_mii_write = &axf_mii_write;
+
+		/* mtu */
+		ugcp->usbgc_min_mtu = ETHERMTU;
+		ugcp->usbgc_max_mtu = ETHERMTU;
+		ugcp->usbgc_default_mtu = ETHERMTU;
+
+		lp = kmem_zalloc(sizeof (struct axf_dev), KM_SLEEP);
+		lp->chip = p;
+		lp->last_link_state = 0;
+		lp->phy_has_reset = B_FALSE;
+
+		dp = usbgem_do_attach(dip, ugcp, lp, sizeof (struct axf_dev));
+
+		kmem_free(ugcp, sizeof (*ugcp));
+
+		if (dp != NULL) {
+			return (DDI_SUCCESS);
+		}
+
+err_free_mem:
+		kmem_free(lp, sizeof (struct axf_dev));
+err_close_pipe:
+err:
+		return (DDI_FAILURE);
+	}
+
+	if (cmd == DDI_RESUME) {
+		return (usbgem_resume(dip));
+	}
+
+	return (DDI_FAILURE);
+}
+
+static int
+axfdetach(dev_info_t *dip, ddi_detach_cmd_t cmd)
+{
+	int	ret;
+
+	if (cmd == DDI_DETACH) {
+		ret = usbgem_do_detach(dip);
+		if (ret != DDI_SUCCESS) {
+			return (DDI_FAILURE);
+		}
+		return (DDI_SUCCESS);
+	}
+	if (cmd == DDI_SUSPEND) {
+		return (usbgem_suspend(dip));
+	}
+	return (DDI_FAILURE);
+}
+
+/* ======================================================== */
+/*
+ * OS depend (loadable streams driver) routine
+ */
+/* ======================================================== */
+#ifdef USBGEM_CONFIG_GLDv3
+USBGEM_STREAM_OPS(axf_ops, axfattach, axfdetach);
+#else
+static	struct module_info axfminfo = {
+	0,			/* mi_idnum */
+	"axf",			/* mi_idname */
+	0,			/* mi_minpsz */
+	ETHERMTU,		/* mi_maxpsz */
+	ETHERMTU*128,		/* mi_hiwat */
+	1,			/* mi_lowat */
+};
+
+static	struct qinit axfrinit = {
+	(int (*)()) NULL,	/* qi_putp */
+	usbgem_rsrv,		/* qi_srvp */
+	usbgem_open,		/* qi_qopen */
+	usbgem_close,		/* qi_qclose */
+	(int (*)()) NULL,	/* qi_qadmin */
+	&axfminfo,		/* qi_minfo */
+	NULL			/* qi_mstat */
+};
+
+static	struct qinit axfwinit = {
+	usbgem_wput,		/* qi_putp */
+	usbgem_wsrv,		/* qi_srvp */
+	(int (*)()) NULL,	/* qi_qopen */
+	(int (*)()) NULL,	/* qi_qclose */
+	(int (*)()) NULL,	/* qi_qadmin */
+	&axfminfo,		/* qi_minfo */
+	NULL			/* qi_mstat */
+};
+
+static struct streamtab	axf_info = {
+	&axfrinit,	/* st_rdinit */
+	&axfwinit,	/* st_wrinit */
+	NULL,		/* st_muxrinit */
+	NULL		/* st_muxwrinit */
+};
+
+static	struct cb_ops cb_axf_ops = {
+	nulldev,	/* cb_open */
+	nulldev,	/* cb_close */
+	nodev,		/* cb_strategy */
+	nodev,		/* cb_print */
+	nodev,		/* cb_dump */
+	nodev,		/* cb_read */
+	nodev,		/* cb_write */
+	nodev,		/* cb_ioctl */
+	nodev,		/* cb_devmap */
+	nodev,		/* cb_mmap */
+	nodev,		/* cb_segmap */
+	nochpoll,	/* cb_chpoll */
+	ddi_prop_op,	/* cb_prop_op */
+	&axf_info,	/* cb_stream */
+	D_NEW|D_MP	/* cb_flag */
+};
+
+static	struct dev_ops axf_ops = {
+	DEVO_REV,	/* devo_rev */
+	0,		/* devo_refcnt */
+	usbgem_getinfo,	/* devo_getinfo */
+	nulldev,	/* devo_identify */
+	nulldev,	/* devo_probe */
+	axfattach,	/* devo_attach */
+	axfdetach,	/* devo_detach */
+	nodev,		/* devo_reset */
+	&cb_axf_ops,	/* devo_cb_ops */
+	NULL,		/* devo_bus_ops */
+	usbgem_power,	/* devo_power */
+#if DEVO_REV >= 4
+	usbgem_quiesce,	/* devo_quiesce */
+#endif
+};
+#endif
+
+static struct modldrv modldrv = {
+	&mod_driverops,	/* Type of module.  This one is a driver */
+	ident,
+	&axf_ops,	/* driver ops */
+};
+
+static struct modlinkage modlinkage = {
+	MODREV_1, &modldrv, NULL
+};
+
+/* ======================================================== */
+/*
+ * _init : done
+ */
+/* ======================================================== */
+int
+_init(void)
+{
+	int 	status;
+
+	DPRINTF(2, (CE_CONT, "!axf: _init: called"));
+
+	status = usbgem_mod_init(&axf_ops, "axf");
+	if (status != DDI_SUCCESS) {
+		return (status);
+	}
+	status = mod_install(&modlinkage);
+	if (status != DDI_SUCCESS) {
+		usbgem_mod_fini(&axf_ops);
+	}
+	return (status);
+}
+
+/*
+ * _fini : done
+ */
+int
+_fini(void)
+{
+	int	status;
+
+	DPRINTF(2, (CE_CONT, "!axf: _fini: called"));
+	status = mod_remove(&modlinkage);
+	if (status == DDI_SUCCESS) {
+		usbgem_mod_fini(&axf_ops);
+	}
+	return (status);
+}
+
+int
+_info(struct modinfo *modinfop)
+{
+	return (mod_info(&modlinkage, modinfop));
+}
diff --git a/usr/src/uts/common/io/cons.c b/usr/src/uts/common/io/cons.c
index 6ef1b0b9f7..495ae93cf9 100644
--- a/usr/src/uts/common/io/cons.c
+++ b/usr/src/uts/common/io/cons.c
@@ -21,6 +21,7 @@
 
 /*
  * Copyright (c) 1982, 2010, Oracle and/or its affiliates. All rights reserved.
+ * Copyright 2015, Joyent, Inc. All rights reserved.
  */
 
 /*
@@ -53,6 +54,7 @@
 #include <sys/vnode.h>
 #include <sys/uio.h>
 #include <sys/stat.h>
+#include <sys/limits.h>
 
 #include <sys/console.h>
 #include <sys/consdev.h>
@@ -414,14 +416,24 @@ cnwrite(dev_t dev, struct uio *uio, struct cred *cred)
 	 */
 	if (vsconsvp != NULL && vsconsvp->v_stream != NULL) {
 		struiod_t uiod;
+		struct iovec buf[IOV_MAX_STACK];
+		int iovlen = 0;
+
+		if (uio->uio_iovcnt > IOV_MAX_STACK) {
+			iovlen = uio->uio_iovcnt * sizeof (iovec_t);
+			uiod.d_iov = kmem_alloc(iovlen, KM_SLEEP);
+		} else {
+			uiod.d_iov = buf;
+		}
 
 		/*
 		 * strwrite modifies uio so need to make copy.
 		 */
-		(void) uiodup(uio, &uiod.d_uio, uiod.d_iov,
-		    sizeof (uiod.d_iov) / sizeof (*uiod.d_iov));
+		(void) uiodup(uio, &uiod.d_uio, uiod.d_iov, uio->uio_iovcnt);
 
 		(void) strwrite(vsconsvp, &uiod.d_uio, cred);
+		if (iovlen != 0)
+			kmem_free(uiod.d_iov, iovlen);
 	}
 
 	if (rconsvp->v_stream != NULL)
diff --git a/usr/src/uts/common/io/devpoll.c b/usr/src/uts/common/io/devpoll.c
index a3fcbbba03..764636e218 100644
--- a/usr/src/uts/common/io/devpoll.c
+++ b/usr/src/uts/common/io/devpoll.c
@@ -25,6 +25,7 @@
 
 /*
  * Copyright (c) 2012 by Delphix. All rights reserved.
+ * Copyright (c) 2015, Joyent, Inc. All rights reserved.
  */
 
 #include <sys/types.h>
@@ -45,6 +46,8 @@
 #include <sys/devpoll.h>
 #include <sys/rctl.h>
 #include <sys/resource.h>
+#include <sys/schedctl.h>
+#include <sys/epoll.h>
 
 #define	RESERVED	1
 
@@ -237,7 +240,8 @@ dpinfo(dev_info_t *dip, ddi_info_cmd_t infocmd, void *arg, void **result)
  *	 stale entries!
  */
 static int
-dp_pcache_poll(pollfd_t *pfdp, pollcache_t *pcp, nfds_t nfds, int *fdcntp)
+dp_pcache_poll(dp_entry_t *dpep, void *dpbuf,
+    pollcache_t *pcp, nfds_t nfds, int *fdcntp)
 {
 	int		start, ostart, end;
 	int		fdcnt, fd;
@@ -247,7 +251,10 @@ dp_pcache_poll(pollfd_t *pfdp, pollcache_t *pcp, nfds_t nfds, int *fdcntp)
 	boolean_t	no_wrap;
 	pollhead_t	*php;
 	polldat_t	*pdp;
+	pollfd_t	*pfdp;
+	epoll_event_t	*epoll;
 	int		error = 0;
+	short		mask = POLLRDHUP | POLLWRBAND;
 
 	ASSERT(MUTEX_HELD(&pcp->pc_lock));
 	if (pcp->pc_bitmap == NULL) {
@@ -257,6 +264,14 @@ dp_pcache_poll(pollfd_t *pfdp, pollcache_t *pcp, nfds_t nfds, int *fdcntp)
 		 */
 		return (error);
 	}
+
+	if (dpep->dpe_flag & DP_ISEPOLLCOMPAT) {
+		pfdp = NULL;
+		epoll = (epoll_event_t *)dpbuf;
+	} else {
+		pfdp = (pollfd_t *)dpbuf;
+		epoll = NULL;
+	}
 retry:
 	start = ostart = pcp->pc_mapstart;
 	end = pcp->pc_mapend;
@@ -316,11 +331,32 @@ repoll:
 				 * polling a closed fd. Hope this will remind
 				 * user to do a POLLREMOVE.
 				 */
-				pfdp[fdcnt].fd = fd;
-				pfdp[fdcnt].revents = POLLNVAL;
-				fdcnt++;
+				if (pfdp != NULL) {
+					pfdp[fdcnt].fd = fd;
+					pfdp[fdcnt].revents = POLLNVAL;
+					fdcnt++;
+					continue;
+				}
+
+				/*
+				 * In the epoll compatibility case, we actually
+				 * perform the implicit removal to remain
+				 * closer to the epoll semantics.
+				 */
+				ASSERT(epoll != NULL);
+
+				pdp->pd_fp = NULL;
+				pdp->pd_events = 0;
+
+				if (php != NULL) {
+					pollhead_delete(php, pdp);
+					pdp->pd_php = NULL;
+				}
+
+				BT_CLEAR(pcp->pc_bitmap, fd);
 				continue;
 			}
+
 			if (fp != pdp->pd_fp) {
 				/*
 				 * user is polling on a cached fd which was
@@ -376,9 +412,69 @@ repoll:
 			}
 
 			if (revent != 0) {
-				pfdp[fdcnt].fd = fd;
-				pfdp[fdcnt].events = pdp->pd_events;
-				pfdp[fdcnt].revents = revent;
+				if (pfdp != NULL) {
+					pfdp[fdcnt].fd = fd;
+					pfdp[fdcnt].events = pdp->pd_events;
+					pfdp[fdcnt].revents = revent;
+				} else {
+					epoll_event_t *ep = &epoll[fdcnt];
+
+					ASSERT(epoll != NULL);
+					ep->data.u64 = pdp->pd_epolldata;
+
+					/*
+					 * If any of the event bits are set for
+					 * which poll and epoll representations
+					 * differ, swizzle in the native epoll
+					 * values.
+					 */
+					if (revent & mask) {
+						ep->events = (revent & ~mask) |
+						    ((revent & POLLRDHUP) ?
+						    EPOLLRDHUP : 0) |
+						    ((revent & POLLWRBAND) ?
+						    EPOLLWRBAND : 0);
+					} else {
+						ep->events = revent;
+					}
+
+					/*
+					 * We define POLLWRNORM to be POLLOUT,
+					 * but epoll has separate definitions
+					 * for them; if POLLOUT is set and the
+					 * user has asked for EPOLLWRNORM, set
+					 * that as well.
+					 */
+					if ((revent & POLLOUT) &&
+					    (pdp->pd_events & EPOLLWRNORM)) {
+						ep->events |= EPOLLWRNORM;
+					}
+				}
+
+				/*
+				 * If POLLET is set, clear the bit in the
+				 * bitmap -- which effectively latches the
+				 * edge on a pollwakeup() from the driver.
+				 */
+				if (pdp->pd_events & POLLET)
+					BT_CLEAR(pcp->pc_bitmap, fd);
+
+				/*
+				 * If POLLONESHOT is set, perform the implicit
+				 * POLLREMOVE.
+				 */
+				if (pdp->pd_events & POLLONESHOT) {
+					pdp->pd_fp = NULL;
+					pdp->pd_events = 0;
+
+					if (php != NULL) {
+						pollhead_delete(php, pdp);
+						pdp->pd_php = NULL;
+					}
+
+					BT_CLEAR(pcp->pc_bitmap, fd);
+				}
+
 				fdcnt++;
 			} else if (php != NULL) {
 				/*
@@ -392,7 +488,7 @@ repoll:
 				 * in bitmap.
 				 */
 				if ((pdp->pd_php != NULL) &&
-				    ((pcp->pc_flag & T_POLLWAKE) == 0)) {
+				    ((pcp->pc_flag & PC_POLLWAKE) == 0)) {
 					BT_CLEAR(pcp->pc_bitmap, fd);
 				}
 				if (pdp->pd_php == NULL) {
@@ -473,11 +569,15 @@ dpopen(dev_t *devp, int flag, int otyp, cred_t *credp)
 	/*
 	 * allocate a pollcache skeleton here. Delay allocating bitmap
 	 * structures until dpwrite() time, since we don't know the
-	 * optimal size yet.
+	 * optimal size yet.  We also delay setting the pid until either
+	 * dpwrite() or attempt to poll on the instance, allowing parents
+	 * to create instances of /dev/poll for their children.  (In the
+	 * epoll compatibility case, this check isn't performed to maintain
+	 * semantic compatibility.)
 	 */
 	pcp = pcache_alloc();
 	dpep->dpe_pcache = pcp;
-	pcp->pc_pid = curproc->p_pid;
+	pcp->pc_pid = -1;
 	*devp = makedevice(getmajor(*devp), minordev);  /* clone the driver */
 	mutex_enter(&devpoll_lock);
 	ASSERT(minordev < dptblsize);
@@ -499,7 +599,9 @@ dpwrite(dev_t dev, struct uio *uiop, cred_t *credp)
 	dp_entry_t	*dpep;
 	pollcache_t	*pcp;
 	pollfd_t	*pollfdp, *pfdp;
-	int		error;
+	dvpoll_epollfd_t *epfdp;
+	uintptr_t	limit;
+	int		error, size;
 	ssize_t		uiosize;
 	nfds_t		pollfdnum;
 	struct pollhead	*php = NULL;
@@ -515,11 +617,23 @@ dpwrite(dev_t dev, struct uio *uiop, cred_t *credp)
 	ASSERT(dpep != NULL);
 	mutex_exit(&devpoll_lock);
 	pcp = dpep->dpe_pcache;
-	if (curproc->p_pid != pcp->pc_pid) {
-		return (EACCES);
+
+	if (!(dpep->dpe_flag & DP_ISEPOLLCOMPAT) &&
+	    curproc->p_pid != pcp->pc_pid) {
+		if (pcp->pc_pid != -1)
+			return (EACCES);
+
+		pcp->pc_pid = curproc->p_pid;
 	}
+
+	if (dpep->dpe_flag & DP_ISEPOLLCOMPAT) {
+		size = sizeof (dvpoll_epollfd_t);
+	} else {
+		size = sizeof (pollfd_t);
+	}
+
 	uiosize = uiop->uio_resid;
-	pollfdnum = uiosize / sizeof (pollfd_t);
+	pollfdnum = uiosize / size;
 	mutex_enter(&curproc->p_lock);
 	if (pollfdnum > (uint_t)rctl_enforced_value(
 	    rctlproc_legacy[RLIMIT_NOFILE], curproc->p_rctls, curproc)) {
@@ -534,6 +648,7 @@ dpwrite(dev_t dev, struct uio *uiop, cred_t *credp)
 	 * each polled fd to the cached set.
 	 */
 	pollfdp = kmem_alloc(uiosize, KM_SLEEP);
+	limit = (uintptr_t)pollfdp + (pollfdnum * size);
 
 	/*
 	 * Although /dev/poll uses the write(2) interface to cache fds, it's
@@ -555,9 +670,27 @@ dpwrite(dev_t dev, struct uio *uiop, cred_t *credp)
 	mutex_enter(&dpep->dpe_lock);
 	dpep->dpe_writerwait++;
 	while (dpep->dpe_refcnt != 0) {
+		/*
+		 * We need to do a bit of a dance here:  we need to drop
+		 * our dpe_lock and grab the pc_lock to broadcast the pc_cv to
+		 * kick any DP_POLL/DP_PPOLL sleepers.
+		 */
+		mutex_exit(&dpep->dpe_lock);
+		mutex_enter(&pcp->pc_lock);
+		pcp->pc_flag |= PC_WRITEWANTED;
+		cv_broadcast(&pcp->pc_cv);
+		mutex_exit(&pcp->pc_lock);
+		mutex_enter(&dpep->dpe_lock);
+
+		if (dpep->dpe_refcnt == 0)
+			break;
+
 		if (!cv_wait_sig_swap(&dpep->dpe_cv, &dpep->dpe_lock)) {
 			dpep->dpe_writerwait--;
 			mutex_exit(&dpep->dpe_lock);
+			mutex_enter(&pcp->pc_lock);
+			pcp->pc_flag &= ~PC_WRITEWANTED;
+			mutex_exit(&pcp->pc_lock);
 			kmem_free(pollfdp, uiosize);
 			return (set_errno(EINTR));
 		}
@@ -565,24 +698,103 @@ dpwrite(dev_t dev, struct uio *uiop, cred_t *credp)
 	dpep->dpe_writerwait--;
 	dpep->dpe_flag |= DP_WRITER_PRESENT;
 	dpep->dpe_refcnt++;
+
 	mutex_exit(&dpep->dpe_lock);
 
 	mutex_enter(&pcp->pc_lock);
+	pcp->pc_flag &= ~PC_WRITEWANTED;
+
 	if (pcp->pc_bitmap == NULL) {
 		pcache_create(pcp, pollfdnum);
 	}
-	for (pfdp = pollfdp; pfdp < pollfdp + pollfdnum; pfdp++) {
+	for (pfdp = pollfdp; (uintptr_t)pfdp < limit;
+	    pfdp = (pollfd_t *)((uintptr_t)pfdp + size)) {
 		fd = pfdp->fd;
-		if ((uint_t)fd >= P_FINFO(curproc)->fi_nfiles)
+		if ((uint_t)fd >= P_FINFO(curproc)->fi_nfiles) {
+			/*
+			 * epoll semantics demand that we return EBADF if our
+			 * specified fd is invalid.
+			 */
+			if (dpep->dpe_flag & DP_ISEPOLLCOMPAT) {
+				error = EBADF;
+				break;
+			}
+
 			continue;
+		}
+
 		pdp = pcache_lookup_fd(pcp, fd);
 		if (pfdp->events != POLLREMOVE) {
+
+			fp = NULL;
+
 			if (pdp == NULL) {
+				/*
+				 * If we're in epoll compatibility mode, check
+				 * that the fd is valid before allocating
+				 * anything for it; epoll semantics demand that
+				 * we return EBADF if our specified fd is
+				 * invalid.
+				 */
+				if (dpep->dpe_flag & DP_ISEPOLLCOMPAT) {
+					if ((fp = getf(fd)) == NULL) {
+						error = EBADF;
+						break;
+					}
+
+					/*
+					 * To (greatly) reduce EEXIST false
+					 * positives, we denote that this
+					 * fp has been epoll()'d; see below.
+					 */
+					fp->f_flag2 |= FEPOLLED;
+				}
+
 				pdp = pcache_alloc_fd(0);
 				pdp->pd_fd = fd;
 				pdp->pd_pcache = pcp;
 				pcache_insert_fd(pcp, pdp, pollfdnum);
+			} else {
+				/*
+				 * epoll semantics demand that we error out if
+				 * a file descriptor is added twice, which we
+				 * check (imperfectly) by checking if we both
+				 * have the file descriptor cached and the
+				 * file pointer that correponds to the file
+				 * descriptor matches our cached value.  (If
+				 * there is a pointer mismatch, the file
+				 * descriptor was closed without being removed.
+				 * The converse is clearly not true, however,
+				 * so to narrow the window by which a spurious
+				 * EEXIST may be returned, we also check if
+				 * this fp has been added to an epoll control
+				 * descriptor in the past; if it hasn't, we
+				 * know that this is due to fp reuse -- it's
+				 * not a true EEXIST case.  (By performing this
+				 * additional check, we limit the window of
+				 * spurious EEXIST to situations where a single
+				 * file descriptor is being used across two or
+				 * more epoll control descriptors -- and even
+				 * then, the file descriptor must be closed and
+				 * reused in a relatively tight time span.)
+				 */
+				if ((dpep->dpe_flag & DP_ISEPOLLCOMPAT) &&
+				    pdp->pd_fp != NULL &&
+				    (fp = getf(fd)) != NULL &&
+				    fp == pdp->pd_fp &&
+				    (fp->f_flag2 & FEPOLLED)) {
+					error = EEXIST;
+					releasef(fd);
+					break;
+				}
 			}
+
+			if (dpep->dpe_flag & DP_ISEPOLLCOMPAT) {
+				/* LINTED pointer alignment */
+				epfdp = (dvpoll_epollfd_t *)pfdp;
+				pdp->pd_epolldata = epfdp->dpep_data;
+			}
+
 			ASSERT(pdp->pd_fd == fd);
 			ASSERT(pdp->pd_pcache == pcp);
 			if (fd >= pcp->pc_mapsize) {
@@ -593,7 +805,7 @@ dpwrite(dev_t dev, struct uio *uiop, cred_t *credp)
 			if (fd > pcp->pc_mapend) {
 				pcp->pc_mapend = fd;
 			}
-			if ((fp = getf(fd)) == NULL) {
+			if (fp == NULL && (fp = getf(fd)) == NULL) {
 				/*
 				 * The fd is not valid. Since we can't pass
 				 * this error back in the write() call, set
@@ -609,7 +821,7 @@ dpwrite(dev_t dev, struct uio *uiop, cred_t *credp)
 			 * same poll events.
 			 */
 			if ((pdp->pd_events == pfdp->events) &&
-			    (pdp->pd_fp != NULL)) {
+			    (pdp->pd_fp == fp)) {
 				/*
 				 * the events are already cached
 				 */
@@ -665,7 +877,17 @@ dpwrite(dev_t dev, struct uio *uiop, cred_t *credp)
 			}
 			releasef(fd);
 		} else {
-			if (pdp == NULL) {
+			if (pdp == NULL || pdp->pd_fp == NULL) {
+				if (dpep->dpe_flag & DP_ISEPOLLCOMPAT) {
+					/*
+					 * As with the add case (above), epoll
+					 * semantics demand that we error out
+					 * in this case.
+					 */
+					error = ENOENT;
+					break;
+				}
+
 				continue;
 			}
 			ASSERT(pdp->pd_fd == fd);
@@ -690,6 +912,17 @@ dpwrite(dev_t dev, struct uio *uiop, cred_t *credp)
 	return (error);
 }
 
+#define	DP_SIGMASK_RESTORE(ksetp) {					\
+	if (ksetp != NULL) {						\
+		mutex_enter(&p->p_lock);				\
+		if (lwp->lwp_cursig == 0) {				\
+			t->t_hold = lwp->lwp_sigoldmask;		\
+			t->t_flag &= ~T_TOMASK;				\
+		}							\
+		mutex_exit(&p->p_lock);					\
+	}								\
+}
+
 /*ARGSUSED*/
 static int
 dpioctl(dev_t dev, int cmd, intptr_t arg, int mode, cred_t *credp, int *rvalp)
@@ -701,7 +934,7 @@ dpioctl(dev_t dev, int cmd, intptr_t arg, int mode, cred_t *credp, int *rvalp)
 	int		error = 0;
 	STRUCT_DECL(dvpoll, dvpoll);
 
-	if (cmd == DP_POLL) {
+	if (cmd == DP_POLL || cmd == DP_PPOLL) {
 		/* do this now, before we sleep on DP_WRITER_PRESENT */
 		now = gethrtime();
 	}
@@ -713,10 +946,39 @@ dpioctl(dev_t dev, int cmd, intptr_t arg, int mode, cred_t *credp, int *rvalp)
 	mutex_exit(&devpoll_lock);
 	ASSERT(dpep != NULL);
 	pcp = dpep->dpe_pcache;
-	if (curproc->p_pid != pcp->pc_pid)
-		return (EACCES);
 
 	mutex_enter(&dpep->dpe_lock);
+
+	if (cmd == DP_EPOLLCOMPAT) {
+		if (dpep->dpe_refcnt != 0) {
+			/*
+			 * We can't turn on epoll compatibility while there
+			 * are outstanding operations.
+			 */
+			mutex_exit(&dpep->dpe_lock);
+			return (EBUSY);
+		}
+
+		/*
+		 * epoll compatibility is a one-way street: there's no way
+		 * to turn it off for a particular open.
+		 */
+		dpep->dpe_flag |= DP_ISEPOLLCOMPAT;
+		mutex_exit(&dpep->dpe_lock);
+
+		return (0);
+	}
+
+	if (!(dpep->dpe_flag & DP_ISEPOLLCOMPAT) &&
+	    curproc->p_pid != pcp->pc_pid) {
+		if (pcp->pc_pid != -1) {
+			mutex_exit(&dpep->dpe_lock);
+			return (EACCES);
+		}
+
+		pcp->pc_pid = curproc->p_pid;
+	}
+
 	while ((dpep->dpe_flag & DP_WRITER_PRESENT) ||
 	    (dpep->dpe_writerwait != 0)) {
 		if (!cv_wait_sig_swap(&dpep->dpe_cv, &dpep->dpe_lock)) {
@@ -729,15 +991,36 @@ dpioctl(dev_t dev, int cmd, intptr_t arg, int mode, cred_t *credp, int *rvalp)
 
 	switch (cmd) {
 	case	DP_POLL:
+	case	DP_PPOLL:
 	{
 		pollstate_t	*ps;
 		nfds_t		nfds;
 		int		fdcnt = 0;
+		size_t		size, fdsize, dpsize;
 		hrtime_t	deadline = 0;
+		k_sigset_t	*ksetp = NULL;
+		k_sigset_t	kset;
+		sigset_t	set;
+		kthread_t	*t = curthread;
+		klwp_t		*lwp = ttolwp(t);
+		struct proc	*p = ttoproc(curthread);
 
 		STRUCT_INIT(dvpoll, mode);
-		error = copyin((caddr_t)arg, STRUCT_BUF(dvpoll),
-		    STRUCT_SIZE(dvpoll));
+
+		/*
+		 * The dp_setp member is only required/consumed for DP_PPOLL,
+		 * which otherwise uses the same structure as DP_POLL.
+		 */
+		if (cmd == DP_POLL) {
+			dpsize = (uintptr_t)STRUCT_FADDR(dvpoll, dp_setp) -
+			    (uintptr_t)STRUCT_FADDR(dvpoll, dp_fds);
+		} else {
+			ASSERT(cmd == DP_PPOLL);
+			dpsize = STRUCT_SIZE(dvpoll);
+		}
+
+		error = copyin((caddr_t)arg, STRUCT_BUF(dvpoll), dpsize);
+
 		if (error) {
 			DP_REFRELE(dpep);
 			return (EFAULT);
@@ -755,6 +1038,52 @@ dpioctl(dev_t dev, int cmd, intptr_t arg, int mode, cred_t *credp, int *rvalp)
 			deadline += now;
 		}
 
+		if (cmd == DP_PPOLL) {
+			void *setp = STRUCT_FGETP(dvpoll, dp_setp);
+
+			if (setp != NULL) {
+				if (copyin(setp, &set, sizeof (set))) {
+					DP_REFRELE(dpep);
+					return (EFAULT);
+				}
+
+				sigutok(&set, &kset);
+				ksetp = &kset;
+
+				mutex_enter(&p->p_lock);
+				schedctl_finish_sigblock(t);
+				lwp->lwp_sigoldmask = t->t_hold;
+				t->t_hold = *ksetp;
+				t->t_flag |= T_TOMASK;
+
+				/*
+				 * Like ppoll() with a non-NULL sigset, we'll
+				 * call cv_reltimedwait_sig() just to check for
+				 * signals.  This call will return immediately
+				 * with either 0 (signalled) or -1 (no signal).
+				 * There are some conditions whereby we can
+				 * get 0 from cv_reltimedwait_sig() without
+				 * a true signal (e.g., a directed stop), so
+				 * we restore our signal mask in the unlikely
+				 * event that lwp_cursig is 0.
+				 */
+				if (!cv_reltimedwait_sig(&t->t_delay_cv,
+				    &p->p_lock, 0, TR_CLOCK_TICK)) {
+					if (lwp->lwp_cursig == 0) {
+						t->t_hold = lwp->lwp_sigoldmask;
+						t->t_flag &= ~T_TOMASK;
+					}
+
+					mutex_exit(&p->p_lock);
+
+					DP_REFRELE(dpep);
+					return (EINTR);
+				}
+
+				mutex_exit(&p->p_lock);
+			}
+		}
+
 		if ((nfds = STRUCT_FGET(dvpoll, dp_nfds)) == 0) {
 			/*
 			 * We are just using DP_POLL to sleep, so
@@ -762,17 +1091,29 @@ dpioctl(dev_t dev, int cmd, intptr_t arg, int mode, cred_t *credp, int *rvalp)
 			 * Do not check for signals if we have a zero timeout.
 			 */
 			DP_REFRELE(dpep);
-			if (deadline == 0)
+			if (deadline == 0) {
+				DP_SIGMASK_RESTORE(ksetp);
 				return (0);
+			}
+
 			mutex_enter(&curthread->t_delay_lock);
 			while ((error =
 			    cv_timedwait_sig_hrtime(&curthread->t_delay_cv,
 			    &curthread->t_delay_lock, deadline)) > 0)
 				continue;
 			mutex_exit(&curthread->t_delay_lock);
+
+			DP_SIGMASK_RESTORE(ksetp);
+
 			return (error == 0 ? EINTR : 0);
 		}
 
+		if (dpep->dpe_flag & DP_ISEPOLLCOMPAT) {
+			size = nfds * (fdsize = sizeof (epoll_event_t));
+		} else {
+			size = nfds * (fdsize = sizeof (pollfd_t));
+		}
+
 		/*
 		 * XXX It would be nice not to have to alloc each time, but it
 		 * requires another per thread structure hook. This can be
@@ -782,37 +1123,45 @@ dpioctl(dev_t dev, int cmd, intptr_t arg, int mode, cred_t *credp, int *rvalp)
 			curthread->t_pollstate = pollstate_create();
 			ps = curthread->t_pollstate;
 		}
-		if (ps->ps_dpbufsize < nfds) {
-			struct proc *p = ttoproc(curthread);
+
+		if (ps->ps_dpbufsize < size) {
 			/*
-			 * The maximum size should be no large than
-			 * current maximum open file count.
+			 * If nfds is larger than twice the current maximum
+			 * open file count, we'll silently clamp it.  This
+			 * only limits our exposure to allocating an
+			 * inordinate amount of kernel memory; it doesn't
+			 * otherwise affect the semantics.  (We have this
+			 * check at twice the maximum instead of merely the
+			 * maximum because some applications pass an nfds that
+			 * is only slightly larger than their limit.)
 			 */
 			mutex_enter(&p->p_lock);
-			if (nfds > p->p_fno_ctl) {
-				mutex_exit(&p->p_lock);
-				DP_REFRELE(dpep);
-				return (EINVAL);
+			if ((nfds >> 1) > p->p_fno_ctl) {
+				nfds = p->p_fno_ctl;
+				size = nfds * fdsize;
 			}
 			mutex_exit(&p->p_lock);
-			kmem_free(ps->ps_dpbuf, sizeof (pollfd_t) *
-			    ps->ps_dpbufsize);
-			ps->ps_dpbuf = kmem_zalloc(sizeof (pollfd_t) *
-			    nfds, KM_SLEEP);
-			ps->ps_dpbufsize = nfds;
+
+			if (ps->ps_dpbufsize < size) {
+				kmem_free(ps->ps_dpbuf, ps->ps_dpbufsize);
+				ps->ps_dpbuf = kmem_zalloc(size, KM_SLEEP);
+				ps->ps_dpbufsize = size;
+			}
 		}
 
 		mutex_enter(&pcp->pc_lock);
 		for (;;) {
-			pcp->pc_flag = 0;
-			error = dp_pcache_poll(ps->ps_dpbuf, pcp, nfds, &fdcnt);
+			pcp->pc_flag &= ~PC_POLLWAKE;
+
+			error = dp_pcache_poll(dpep, ps->ps_dpbuf,
+			    pcp, nfds, &fdcnt);
 			if (fdcnt > 0 || error != 0)
 				break;
 
 			/*
 			 * A pollwake has happened since we polled cache.
 			 */
-			if (pcp->pc_flag & T_POLLWAKE)
+			if (pcp->pc_flag & PC_POLLWAKE)
 				continue;
 
 			/*
@@ -822,8 +1171,40 @@ dpioctl(dev_t dev, int cmd, intptr_t arg, int mode, cred_t *credp, int *rvalp)
 				/* immediate timeout; do not check signals */
 				break;
 			}
-			error = cv_timedwait_sig_hrtime(&pcp->pc_cv,
-			    &pcp->pc_lock, deadline);
+
+			if (!(pcp->pc_flag & PC_WRITEWANTED)) {
+				error = cv_timedwait_sig_hrtime(&pcp->pc_cv,
+				    &pcp->pc_lock, deadline);
+			} else {
+				error = 1;
+			}
+
+			if (error > 0 && (pcp->pc_flag & PC_WRITEWANTED)) {
+				/*
+				 * We've been kicked off of our cv because a
+				 * writer wants in.  We're going to drop our
+				 * reference count and then wait until the
+				 * writer is gone -- at which point we'll
+				 * reacquire the pc_lock and call into
+				 * dp_pcache_poll() to get the updated state.
+				 */
+				mutex_exit(&pcp->pc_lock);
+
+				mutex_enter(&dpep->dpe_lock);
+				dpep->dpe_refcnt--;
+				cv_broadcast(&dpep->dpe_cv);
+
+				while ((dpep->dpe_flag & DP_WRITER_PRESENT) ||
+				    (dpep->dpe_writerwait != 0)) {
+					error = cv_wait_sig_swap(&dpep->dpe_cv,
+					    &dpep->dpe_lock);
+				}
+
+				dpep->dpe_refcnt++;
+				mutex_exit(&dpep->dpe_lock);
+				mutex_enter(&pcp->pc_lock);
+			}
+
 			/*
 			 * If we were awakened by a signal or timeout
 			 * then break the loop, else poll again.
@@ -837,9 +1218,11 @@ dpioctl(dev_t dev, int cmd, intptr_t arg, int mode, cred_t *credp, int *rvalp)
 		}
 		mutex_exit(&pcp->pc_lock);
 
+		DP_SIGMASK_RESTORE(ksetp);
+
 		if (error == 0 && fdcnt > 0) {
-			if (copyout(ps->ps_dpbuf, STRUCT_FGETP(dvpoll,
-			    dp_fds), sizeof (pollfd_t) * fdcnt)) {
+			if (copyout(ps->ps_dpbuf,
+			    STRUCT_FGETP(dvpoll, dp_fds), fdcnt * fdsize)) {
 				DP_REFRELE(dpep);
 				return (EFAULT);
 			}
@@ -901,10 +1284,25 @@ static int
 dppoll(dev_t dev, short events, int anyyet, short *reventsp,
     struct pollhead **phpp)
 {
+	minor_t		minor;
+	dp_entry_t	*dpep;
+
+	minor = getminor(dev);
+
+	mutex_enter(&devpoll_lock);
+	dpep = devpolltbl[minor];
+	ASSERT(dpep != NULL);
+	mutex_exit(&devpoll_lock);
+
 	/*
 	 * Polling on a /dev/poll fd is not fully supported yet.
 	 */
-	*reventsp = POLLERR;
+	if (dpep->dpe_flag & DP_ISEPOLLCOMPAT) {
+		/* no error in epoll compat. mode */
+		*reventsp = 0;
+	} else {
+		*reventsp = POLLERR;
+	}
 	return (0);
 }
 
diff --git a/usr/src/uts/common/io/dld/dld_drv.c b/usr/src/uts/common/io/dld/dld_drv.c
index 40cbe86170..62bc4a8ecf 100644
--- a/usr/src/uts/common/io/dld/dld_drv.c
+++ b/usr/src/uts/common/io/dld/dld_drv.c
@@ -20,6 +20,7 @@
  */
 /*
  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
+ * Copyright 2015, Joyent Inc.
  */
 
 /*
@@ -701,7 +702,8 @@ drv_ioc_prop_common(dld_ioc_macprop_t *prop, intptr_t arg, boolean_t set,
 				err = EACCES;
 				goto done;
 			}
-			err = dls_devnet_setzid(dlh, dzp->diz_zid);
+			err = dls_devnet_setzid(dlh, dzp->diz_zid,
+			    dzp->diz_transient);
 		} else {
 			kprop->pr_perm_flags = MAC_PROP_PERM_RW;
 			(*(zoneid_t *)kprop->pr_val) = dls_devnet_getzid(dlh);
@@ -865,7 +867,7 @@ drv_ioc_rename(void *karg, intptr_t arg, int mode, cred_t *cred, int *rvalp)
 		return (err);
 
 	if ((err = dls_devnet_rename(dir->dir_linkid1, dir->dir_linkid2,
-	    dir->dir_link)) != 0)
+	    dir->dir_link, dir->dir_zoneinit)) != 0)
 		return (err);
 
 	if (dir->dir_linkid2 == DATALINK_INVALID_LINKID)
@@ -1376,7 +1378,8 @@ static dld_ioc_modentry_t dld_ioc_modtable[] = {
 	{SIMNET_IOC,	"simnet", 0, NULL, 0},
 	{BRIDGE_IOC,	"bridge", 0, NULL, 0},
 	{IPTUN_IOC,	"iptun", 0, NULL, 0},
-	{IBPART_IOC,	"ibp", -1, NULL, 0}
+	{IBPART_IOC,	"ibp", -1, NULL, 0},
+	{OVERLAY_IOC,	"overlay", 0, NULL, 0}
 };
 #define	DLDIOC_CNT	\
 	(sizeof (dld_ioc_modtable) / sizeof (dld_ioc_modentry_t))
diff --git a/usr/src/uts/common/io/dld/dld_proto.c b/usr/src/uts/common/io/dld/dld_proto.c
index a438e43d91..79c3d8260a 100644
--- a/usr/src/uts/common/io/dld/dld_proto.c
+++ b/usr/src/uts/common/io/dld/dld_proto.c
@@ -41,7 +41,7 @@ static proto_reqfunc_t proto_info_req, proto_attach_req, proto_detach_req,
     proto_bind_req, proto_unbind_req, proto_promiscon_req, proto_promiscoff_req,
     proto_enabmulti_req, proto_disabmulti_req, proto_physaddr_req,
     proto_setphysaddr_req, proto_udqos_req, proto_req, proto_capability_req,
-    proto_notify_req, proto_passive_req;
+    proto_notify_req, proto_passive_req, proto_exclusive_req;
 
 static void proto_capability_advertise(dld_str_t *, mblk_t *);
 static int dld_capab_poll_disable(dld_str_t *, dld_capab_poll_t *);
@@ -121,6 +121,9 @@ dld_proto(dld_str_t *dsp, mblk_t *mp)
 	case DL_PASSIVE_REQ:
 		proto_passive_req(dsp, mp);
 		break;
+	case DL_EXCLUSIVE_REQ:
+		proto_exclusive_req(dsp, mp);
+		break;
 	default:
 		proto_req(dsp, mp);
 		break;
@@ -605,6 +608,10 @@ proto_promiscon_req(dld_str_t *dsp, mblk_t *mp)
 		new_flags |= DLS_PROMISC_PHYS;
 		break;
 
+	case DL_PROMISC_RX_ONLY:
+		new_flags |= DLS_PROMISC_RX_ONLY;
+		break;
+
 	default:
 		dl_err = DL_NOTSUPPORTED;
 		goto failed2;
@@ -692,12 +699,24 @@ proto_promiscoff_req(dld_str_t *dsp, mblk_t *mp)
 		new_flags &= ~DLS_PROMISC_PHYS;
 		break;
 
+	case DL_PROMISC_RX_ONLY:
+		if (!(dsp->ds_promisc & DLS_PROMISC_RX_ONLY)) {
+			dl_err = DL_NOTENAB;
+			goto failed;
+		}
+		new_flags &= ~DLS_PROMISC_RX_ONLY;
+		break;
+
 	default:
 		dl_err = DL_NOTSUPPORTED;
 		mac_perim_exit(mph);
 		goto failed;
 	}
 
+	/* DLS_PROMISC_RX_ONLY can't be a solo flag */
+	if (new_flags == DLS_PROMISC_RX_ONLY)
+		new_flags = 0;
+
 	/*
 	 * Adjust channel promiscuity.
 	 */
@@ -1295,7 +1314,8 @@ proto_passive_req(dld_str_t *dsp, mblk_t *mp)
 	 * If we've already become active by issuing an active primitive,
 	 * then it's too late to try to become passive.
 	 */
-	if (dsp->ds_passivestate == DLD_ACTIVE) {
+	if (dsp->ds_passivestate == DLD_ACTIVE ||
+	    dsp->ds_passivestate == DLD_EXCLUSIVE) {
 		dl_err = DL_OUTSTATE;
 		goto failed;
 	}
@@ -1354,7 +1374,12 @@ dld_capab_direct(dld_str_t *dsp, void *data, uint_t flags)
 		dls_rx_set(dsp, (dls_rx_t)direct->di_rx_cf,
 		    direct->di_rx_ch);
 
-		direct->di_tx_df = (uintptr_t)str_mdata_fastpath_put;
+		if (direct->di_flags & DI_DIRECT_RAW) {
+			direct->di_tx_df =
+			    (uintptr_t)str_mdata_raw_fastpath_put;
+		} else {
+			direct->di_tx_df = (uintptr_t)str_mdata_fastpath_put;
+		}
 		direct->di_tx_dh = dsp;
 		direct->di_tx_cb_df = (uintptr_t)mac_client_tx_notify;
 		direct->di_tx_cb_dh = dsp->ds_mch;
@@ -1516,8 +1541,9 @@ dld_capab(dld_str_t *dsp, uint_t type, void *data, uint_t flags)
 	 * completes. So we limit the check to DLD_ENABLE case.
 	 */
 	if ((flags == DLD_ENABLE && type != DLD_CAPAB_PERIM) &&
-	    (dsp->ds_sap != ETHERTYPE_IP ||
-	    !check_mod_above(dsp->ds_rq, "ip"))) {
+	    ((dsp->ds_sap != ETHERTYPE_IP ||
+	    !check_mod_above(dsp->ds_rq, "ip")) &&
+	    !check_mod_above(dsp->ds_rq, "vnd"))) {
 		return (ENOTSUP);
 	}
 
@@ -1599,9 +1625,15 @@ proto_capability_advertise(dld_str_t *dsp, mblk_t *mp)
 	}
 
 	/*
-	 * Direct capability negotiation interface between IP and DLD
+	 * Direct capability negotiation interface between IP/VND and DLD. Note
+	 * that for vnd we only allow the case where the media type is the
+	 * native media type so we know that there are no transformations that
+	 * would have to happen to the mac header that it receives.
 	 */
-	if (dsp->ds_sap == ETHERTYPE_IP && check_mod_above(dsp->ds_rq, "ip")) {
+	if ((dsp->ds_sap == ETHERTYPE_IP &&
+	    check_mod_above(dsp->ds_rq, "ip")) ||
+	    (check_mod_above(dsp->ds_rq, "vnd") &&
+	    dsp->ds_mip->mi_media == dsp->ds_mip->mi_nativemedia)) {
 		dld_capable = B_TRUE;
 		subsize += sizeof (dl_capability_sub_t) +
 		    sizeof (dl_capab_dld_t);
@@ -1720,3 +1752,36 @@ dld_capabilities_disable(dld_str_t *dsp)
 	if (dsp->ds_polling)
 		(void) dld_capab_poll_disable(dsp, NULL);
 }
+
+static void
+proto_exclusive_req(dld_str_t *dsp, mblk_t *mp)
+{
+	int ret = 0;
+	t_uscalar_t dl_err;
+	mac_perim_handle_t mph;
+
+	if (dsp->ds_passivestate != DLD_UNINITIALIZED) {
+		dl_err = DL_OUTSTATE;
+		goto failed;
+	}
+
+	if (MBLKL(mp) < DL_EXCLUSIVE_REQ_SIZE) {
+		dl_err = DL_BADPRIM;
+		goto failed;
+	}
+
+	mac_perim_enter_by_mh(dsp->ds_mh, &mph);
+	ret = dls_exclusive_set(dsp, B_TRUE);
+	mac_perim_exit(mph);
+
+	if (ret != 0) {
+		dl_err = DL_SYSERR;
+		goto failed;
+	}
+
+	dsp->ds_passivestate = DLD_EXCLUSIVE;
+	dlokack(dsp->ds_wq, mp, DL_EXCLUSIVE_REQ);
+	return;
+failed:
+	dlerrorack(dsp->ds_wq, mp, DL_EXCLUSIVE_REQ, dl_err, (t_uscalar_t)ret);
+}
diff --git a/usr/src/uts/common/io/dld/dld_str.c b/usr/src/uts/common/io/dld/dld_str.c
index 6f0d0b9a6c..f5308e70ff 100644
--- a/usr/src/uts/common/io/dld/dld_str.c
+++ b/usr/src/uts/common/io/dld/dld_str.c
@@ -854,6 +854,77 @@ i_dld_ether_header_update_tag(mblk_t *mp, uint_t pri, uint16_t vid,
 	return (mp);
 }
 
+static boolean_t
+i_dld_raw_ether_check(dld_str_t *dsp, mac_header_info_t *mhip, mblk_t **mpp)
+{
+	mblk_t *mp = *mpp;
+	mblk_t *newmp;
+	uint_t pri, vid, dvid;
+
+	dvid = mac_client_vid(dsp->ds_mch);
+
+	/*
+	 * Discard the packet if this is a VLAN stream but the VID in
+	 * the packet is not correct.
+	 */
+	vid = VLAN_ID(mhip->mhi_tci);
+	if ((dvid != VLAN_ID_NONE) && (vid != VLAN_ID_NONE))
+		return (B_FALSE);
+
+	/*
+	 * Discard the packet if this packet is a tagged packet
+	 * but both pri and VID are 0.
+	 */
+	pri = VLAN_PRI(mhip->mhi_tci);
+	if (mhip->mhi_istagged && !mhip->mhi_ispvid && pri == 0 &&
+	    vid == VLAN_ID_NONE)
+		return (B_FALSE);
+
+	/*
+	 * Update the priority bits to the per-stream priority if
+	 * priority is not set in the packet. Update the VID for
+	 * packets on a VLAN stream.
+	 */
+	pri = (pri == 0) ? dsp->ds_pri : 0;
+	if ((pri != 0) || (dvid != VLAN_ID_NONE)) {
+		if ((newmp = i_dld_ether_header_update_tag(mp, pri,
+		    dvid, dsp->ds_dlp->dl_tagmode)) == NULL) {
+			return (B_FALSE);
+		}
+		*mpp = newmp;
+	}
+
+	return (B_TRUE);
+}
+
+mac_tx_cookie_t
+str_mdata_raw_fastpath_put(dld_str_t *dsp, mblk_t *mp, uintptr_t f_hint,
+    uint16_t flag)
+{
+	boolean_t is_ethernet = (dsp->ds_mip->mi_media == DL_ETHER);
+	mac_header_info_t mhi;
+	mac_tx_cookie_t cookie;
+
+	if (mac_vlan_header_info(dsp->ds_mh, mp, &mhi) != 0)
+		goto discard;
+
+	if (is_ethernet) {
+		if (i_dld_raw_ether_check(dsp, &mhi, &mp) == B_FALSE)
+			goto discard;
+	}
+
+	if ((cookie = DLD_TX(dsp, mp, f_hint, flag)) != NULL) {
+		DLD_SETQFULL(dsp);
+	}
+	return (cookie);
+discard:
+	/* TODO: bump kstat? */
+	freemsg(mp);
+	return (NULL);
+}
+
+
+
 /*
  * M_DATA put (IP fast-path mode)
  */
@@ -902,7 +973,6 @@ str_mdata_raw_put(dld_str_t *dsp, mblk_t *mp)
 	mblk_t *bp, *newmp;
 	size_t size;
 	mac_header_info_t mhi;
-	uint_t pri, vid, dvid;
 	uint_t max_sdu;
 
 	/*
@@ -948,38 +1018,8 @@ str_mdata_raw_put(dld_str_t *dsp, mblk_t *mp)
 		goto discard;
 
 	if (is_ethernet) {
-		dvid = mac_client_vid(dsp->ds_mch);
-
-		/*
-		 * Discard the packet if this is a VLAN stream but the VID in
-		 * the packet is not correct.
-		 */
-		vid = VLAN_ID(mhi.mhi_tci);
-		if ((dvid != VLAN_ID_NONE) && (vid != VLAN_ID_NONE))
-			goto discard;
-
-		/*
-		 * Discard the packet if this packet is a tagged packet
-		 * but both pri and VID are 0.
-		 */
-		pri = VLAN_PRI(mhi.mhi_tci);
-		if (mhi.mhi_istagged && !mhi.mhi_ispvid && pri == 0 &&
-		    vid == VLAN_ID_NONE)
+		if (i_dld_raw_ether_check(dsp, &mhi, &mp) == B_FALSE)
 			goto discard;
-
-		/*
-		 * Update the priority bits to the per-stream priority if
-		 * priority is not set in the packet. Update the VID for
-		 * packets on a VLAN stream.
-		 */
-		pri = (pri == 0) ? dsp->ds_pri : 0;
-		if ((pri != 0) || (dvid != VLAN_ID_NONE)) {
-			if ((newmp = i_dld_ether_header_update_tag(mp, pri,
-			    dvid, dsp->ds_dlp->dl_tagmode)) == NULL) {
-				goto discard;
-			}
-			mp = newmp;
-		}
 	}
 
 	if (DLD_TX(dsp, mp, 0, 0) != NULL) {
diff --git a/usr/src/uts/common/io/dls/dls.c b/usr/src/uts/common/io/dls/dls.c
index 92993ada58..9fa649943c 100644
--- a/usr/src/uts/common/io/dls/dls.c
+++ b/usr/src/uts/common/io/dls/dls.c
@@ -25,7 +25,7 @@
  */
 
 /*
- * Copyright (c) 2013 Joyent, Inc.  All rights reserved.
+ * Copyright (c) 2014 Joyent, Inc.  All rights reserved.
  */
 
 /*
@@ -248,19 +248,44 @@ dls_promisc(dld_str_t *dsp, uint32_t new_flags)
 {
 	int err = 0;
 	uint32_t old_flags = dsp->ds_promisc;
+	uint32_t new_type = new_flags & ~DLS_PROMISC_RX_ONLY;
 	mac_client_promisc_type_t mptype = MAC_CLIENT_PROMISC_ALL;
+	uint16_t mac_flags = 0;
 
 	ASSERT(MAC_PERIM_HELD(dsp->ds_mh));
 	ASSERT(!(new_flags & ~(DLS_PROMISC_SAP | DLS_PROMISC_MULTI |
-	    DLS_PROMISC_PHYS)));
+	    DLS_PROMISC_PHYS | DLS_PROMISC_RX_ONLY)));
+
+	/*
+	 * Asking us just to turn on DLS_PROMISC_RX_ONLY is not valid.
+	 */
+	if (new_flags == DLS_PROMISC_RX_ONLY)
+		return (EINVAL);
 
 	/*
 	 * If the user has only requested DLS_PROMISC_MULTI then we need to make
 	 * sure that they don't see all packets.
 	 */
-	if (new_flags == DLS_PROMISC_MULTI)
+	if (new_type == DLS_PROMISC_MULTI)
 		mptype = MAC_CLIENT_PROMISC_MULTI;
 
+	/*
+	 * Look at new flags and figure out the correct mac promisc flags.
+	 * If we've only requested DLS_PROMISC_SAP and not _MULTI or _PHYS,
+	 * don't turn on physical promisc mode.
+	 */
+	if (new_flags & DLS_PROMISC_RX_ONLY)
+		mac_flags |= MAC_PROMISC_FLAGS_NO_TX_LOOP;
+	if (new_type == DLS_PROMISC_SAP)
+		mac_flags |= MAC_PROMISC_FLAGS_NO_PHYS;
+
+	/*
+	 * There are three cases we care about here with respect to MAC. Going
+	 * from nothing to something, something to nothing, something to
+	 * something where we need to change how we're getting stuff from mac.
+	 * In the last case, as long as they're not equal, we need to assume
+	 * something has changed and do something about it.
+	 */
 	if (dsp->ds_promisc == 0 && new_flags != 0) {
 		/*
 		 * If only DLS_PROMISC_SAP, we don't turn on the
@@ -268,9 +293,7 @@ dls_promisc(dld_str_t *dsp, uint32_t new_flags)
 		 */
 		dsp->ds_promisc = new_flags;
 		err = mac_promisc_add(dsp->ds_mch, mptype,
-		    dls_rx_promisc, dsp, &dsp->ds_mph,
-		    (new_flags != DLS_PROMISC_SAP) ? 0 :
-		    MAC_PROMISC_FLAGS_NO_PHYS);
+		    dls_rx_promisc, dsp, &dsp->ds_mph, mac_flags);
 		if (err != 0) {
 			dsp->ds_promisc = old_flags;
 			return (err);
@@ -296,19 +319,13 @@ dls_promisc(dld_str_t *dsp, uint32_t new_flags)
 			    MAC_CLIENT_PROMISC_ALL, dls_rx_vlan_promisc, dsp,
 			    &dsp->ds_vlan_mph, MAC_PROMISC_FLAGS_NO_PHYS);
 		}
-	} else if (dsp->ds_promisc == DLS_PROMISC_SAP && new_flags != 0 &&
-	    new_flags != dsp->ds_promisc) {
-		/*
-		 * If the old flag is PROMISC_SAP, but the current flag has
-		 * changed to some new non-zero value, we need to turn the
-		 * physical promiscuous mode.
-		 */
+	} else if (new_flags != 0 && new_flags != old_flags) {
 		ASSERT(dsp->ds_mph != NULL);
 		mac_promisc_remove(dsp->ds_mph);
 		/* Honors both after-remove and before-add semantics! */
 		dsp->ds_promisc = new_flags;
 		err = mac_promisc_add(dsp->ds_mch, mptype,
-		    dls_rx_promisc, dsp, &dsp->ds_mph, 0);
+		    dls_rx_promisc, dsp, &dsp->ds_mph, mac_flags);
 		if (err != 0)
 			dsp->ds_promisc = old_flags;
 	} else {
@@ -629,6 +646,22 @@ boolean_t
 dls_accept_promisc(dld_str_t *dsp, mac_header_info_t *mhip, dls_rx_t *ds_rx,
     void **ds_rx_arg, boolean_t loopback)
 {
+	if (dsp->ds_promisc == 0) {
+		/*
+		 * If there are active walkers of the mi_promisc_list when
+		 * promiscuousness is disabled, ds_promisc will be cleared,
+		 * but the DLS will remain on the mi_promisc_list until the
+		 * walk is completed.  If we do not recognize this case here,
+		 * we won't properly execute the ds_promisc case in the common
+		 * accept routine -- and we will potentially accept a packet
+		 * that has originated with this DLS (which in turn can
+		 * induce recursion and death by stack overflow).  If
+		 * ds_promisc is zero, we know that we are in this window --
+		 * and we refuse to accept the packet.
+		 */
+		return (B_FALSE);
+	}
+
 	return (dls_accept_common(dsp, mhip, ds_rx, ds_rx_arg, B_TRUE,
 	    loopback));
 }
@@ -659,7 +692,10 @@ dls_mac_active_set(dls_link_t *dlp)
 		 * Set the function to start receiving packets.
 		 */
 		mac_rx_set(dlp->dl_mch, i_dls_link_rx, dlp);
+	} else if (dlp->dl_exclusive == B_TRUE) {
+		return (EBUSY);
 	}
+
 	dlp->dl_nactive++;
 	return (0);
 }
@@ -685,7 +721,11 @@ dls_active_set(dld_str_t *dsp)
 	if (dsp->ds_passivestate == DLD_PASSIVE)
 		return (0);
 
-	/* If we're already active, then there's nothing more to do. */
+	if (dsp->ds_dlp->dl_exclusive == B_TRUE &&
+	    dsp->ds_passivestate != DLD_EXCLUSIVE)
+		return (EBUSY);
+
+	/* If we're already active, we need to check the link's exclusivity */
 	if ((dsp->ds_nactive == 0) &&
 	    ((err = dls_mac_active_set(dsp->ds_dlp)) != 0)) {
 		/* except for ENXIO all other errors are mapped to EBUSY */
@@ -694,7 +734,8 @@ dls_active_set(dld_str_t *dsp)
 		return (err);
 	}
 
-	dsp->ds_passivestate = DLD_ACTIVE;
+	dsp->ds_passivestate = dsp->ds_dlp->dl_exclusive == B_TRUE ?
+	    DLD_EXCLUSIVE : DLD_ACTIVE;
 	dsp->ds_nactive++;
 	return (0);
 }
@@ -725,7 +766,32 @@ dls_active_clear(dld_str_t *dsp, boolean_t all)
 	if (dsp->ds_nactive != 0)
 		return;
 
-	ASSERT(dsp->ds_passivestate == DLD_ACTIVE);
+	ASSERT(dsp->ds_passivestate == DLD_ACTIVE ||
+	    dsp->ds_passivestate == DLD_EXCLUSIVE);
 	dls_mac_active_clear(dsp->ds_dlp);
+	/*
+	 * We verify below to ensure that no other part of DLS has mucked with
+	 * our exclusive state.
+	 */
+	if (dsp->ds_passivestate == DLD_EXCLUSIVE)
+		VERIFY(dls_exclusive_set(dsp, B_FALSE) == 0);
 	dsp->ds_passivestate = DLD_UNINITIALIZED;
 }
+
+int
+dls_exclusive_set(dld_str_t *dsp, boolean_t enable)
+{
+	ASSERT(MAC_PERIM_HELD(dsp->ds_mh));
+
+	if (enable == B_FALSE) {
+		dsp->ds_dlp->dl_exclusive = B_FALSE;
+		return (0);
+	}
+
+	if (dsp->ds_dlp->dl_nactive != 0)
+		return (EBUSY);
+
+	dsp->ds_dlp->dl_exclusive = B_TRUE;
+
+	return (0);
+}
diff --git a/usr/src/uts/common/io/dls/dls_link.c b/usr/src/uts/common/io/dls/dls_link.c
index 6b92a81e77..c5363e8194 100644
--- a/usr/src/uts/common/io/dls/dls_link.c
+++ b/usr/src/uts/common/io/dls/dls_link.c
@@ -602,6 +602,7 @@ i_dls_link_destroy(dls_link_t *dlp)
 	dlp->dl_mip = NULL;
 	dlp->dl_unknowns = 0;
 	dlp->dl_nonip_cnt = 0;
+	dlp->dl_exclusive = B_FALSE;
 	kmem_cache_free(i_dls_link_cachep, dlp);
 }
 
diff --git a/usr/src/uts/common/io/dls/dls_mgmt.c b/usr/src/uts/common/io/dls/dls_mgmt.c
index 049c4bd757..6111d62475 100644
--- a/usr/src/uts/common/io/dls/dls_mgmt.c
+++ b/usr/src/uts/common/io/dls/dls_mgmt.c
@@ -21,6 +21,7 @@
 /*
  * Copyright 2009 Sun Microsystems, Inc.  All rights reserved.
  * Use is subject to license terms.
+ * Copyright (c) 2013 Joyent, Inc.  All rights reserved.
  */
 
 /*
@@ -105,12 +106,13 @@ typedef struct dls_devnet_s {
 	zoneid_t	dd_zid;		/* current zone */
 	boolean_t	dd_prop_loaded;
 	taskqid_t	dd_prop_taskid;
+	boolean_t	dd_transient;	/* link goes away when zone does */
 } dls_devnet_t;
 
 static int i_dls_devnet_create_iptun(const char *, const char *,
     datalink_id_t *);
 static int i_dls_devnet_destroy_iptun(datalink_id_t);
-static int i_dls_devnet_setzid(dls_devnet_t *, zoneid_t, boolean_t);
+static int i_dls_devnet_setzid(dls_devnet_t *, zoneid_t, boolean_t, boolean_t);
 static int dls_devnet_unset(const char *, datalink_id_t *, boolean_t);
 
 /*ARGSUSED*/
@@ -145,7 +147,12 @@ dls_zone_remove(datalink_id_t linkid, void *arg)
 	dls_devnet_t *ddp;
 
 	if (dls_devnet_hold_tmp(linkid, &ddp) == 0) {
-		(void) dls_devnet_setzid(ddp, GLOBAL_ZONEID);
+		/*
+		 * Don't bother moving transient links back to the global zone
+		 * since we will simply delete them in dls_devnet_unset.
+		 */
+		if (!ddp->dd_transient)
+			(void) dls_devnet_setzid(ddp, GLOBAL_ZONEID, B_FALSE);
 		dls_devnet_rele_tmp(ddp);
 	}
 	return (0);
@@ -526,6 +533,27 @@ dls_mgmt_get_linkid(const char *link, datalink_id_t *linkid)
 
 	getlinkid.ld_cmd = DLMGMT_CMD_GETLINKID;
 	(void) strlcpy(getlinkid.ld_link, link, MAXLINKNAMELEN);
+	getlinkid.ld_zoneid = getzoneid();
+
+	if ((err = i_dls_mgmt_upcall(&getlinkid, sizeof (getlinkid), &retval,
+	    sizeof (retval))) == 0) {
+		*linkid = retval.lr_linkid;
+	}
+	return (err);
+}
+
+int
+dls_mgmt_get_linkid_in_zone(const char *link, datalink_id_t *linkid,
+    zoneid_t zid)
+{
+	dlmgmt_door_getlinkid_t		getlinkid;
+	dlmgmt_getlinkid_retval_t	retval;
+	int				err;
+
+	ASSERT(getzoneid() == GLOBAL_ZONEID || zid == getzoneid());
+	getlinkid.ld_cmd = DLMGMT_CMD_GETLINKID;
+	(void) strlcpy(getlinkid.ld_link, link, MAXLINKNAMELEN);
+	getlinkid.ld_zoneid = zid;
 
 	if ((err = i_dls_mgmt_upcall(&getlinkid, sizeof (getlinkid), &retval,
 	    sizeof (retval))) == 0) {
@@ -534,6 +562,7 @@ dls_mgmt_get_linkid(const char *link, datalink_id_t *linkid)
 	return (err);
 }
 
+
 datalink_id_t
 dls_mgmt_get_next(datalink_id_t linkid, datalink_class_t class,
     datalink_media_t dmedia, uint32_t flags)
@@ -740,12 +769,23 @@ dls_devnet_stat_update(kstat_t *ksp, int rw)
  * Create the "link" kstats.
  */
 static void
-dls_devnet_stat_create(dls_devnet_t *ddp, zoneid_t zoneid)
+dls_devnet_stat_create(dls_devnet_t *ddp, zoneid_t zoneid, zoneid_t newzoneid)
 {
 	kstat_t	*ksp;
+	char	*nm;
+	char	kname[MAXLINKNAMELEN];
+
+	if (zoneid != newzoneid) {
+		ASSERT(zoneid == GLOBAL_ZONEID);
+		(void) snprintf(kname, sizeof (kname), "z%d_%s", newzoneid,
+		    ddp->dd_linkname);
+		nm = kname;
+	} else {
+		nm = ddp->dd_linkname;
+	}
 
-	if (dls_stat_create("link", 0, ddp->dd_linkname, zoneid,
-	    dls_devnet_stat_update, ddp, &ksp) == 0) {
+	if (dls_stat_create("link", 0, nm, zoneid,
+	    dls_devnet_stat_update, ddp, &ksp, newzoneid) == 0) {
 		ASSERT(ksp != NULL);
 		if (zoneid == ddp->dd_owner_zid) {
 			ASSERT(ddp->dd_ksp == NULL);
@@ -765,12 +805,12 @@ dls_devnet_stat_destroy(dls_devnet_t *ddp, zoneid_t zoneid)
 {
 	if (zoneid == ddp->dd_owner_zid) {
 		if (ddp->dd_ksp != NULL) {
-			kstat_delete(ddp->dd_ksp);
+			dls_stat_delete(ddp->dd_ksp);
 			ddp->dd_ksp = NULL;
 		}
 	} else {
 		if (ddp->dd_zone_ksp != NULL) {
-			kstat_delete(ddp->dd_zone_ksp);
+			dls_stat_delete(ddp->dd_zone_ksp);
 			ddp->dd_zone_ksp = NULL;
 		}
 	}
@@ -781,15 +821,25 @@ dls_devnet_stat_destroy(dls_devnet_t *ddp, zoneid_t zoneid)
  * and create the new set using the new name.
  */
 static void
-dls_devnet_stat_rename(dls_devnet_t *ddp)
+dls_devnet_stat_rename(dls_devnet_t *ddp, boolean_t zoneinit)
 {
 	if (ddp->dd_ksp != NULL) {
-		kstat_delete(ddp->dd_ksp);
+		dls_stat_delete(ddp->dd_ksp);
 		ddp->dd_ksp = NULL;
 	}
-	/* We can't rename a link while it's assigned to a non-global zone. */
+	if (zoneinit && ddp->dd_zone_ksp != NULL) {
+		dls_stat_delete(ddp->dd_zone_ksp);
+		ddp->dd_zone_ksp = NULL;
+	}
+	/*
+	 * We can't rename a link while it's assigned to a non-global zone
+	 * unless we're first initializing the zone while readying it.
+	 */
 	ASSERT(ddp->dd_zone_ksp == NULL);
-	dls_devnet_stat_create(ddp, ddp->dd_owner_zid);
+	dls_devnet_stat_create(ddp, ddp->dd_owner_zid,
+	    (zoneinit ? ddp->dd_zid : ddp->dd_owner_zid));
+	if (zoneinit)
+		dls_devnet_stat_create(ddp, ddp->dd_zid, ddp->dd_zid);
 }
 
 /*
@@ -878,7 +928,8 @@ done:
 	rw_exit(&i_dls_devnet_lock);
 	if (err == 0) {
 		if (zoneid != GLOBAL_ZONEID &&
-		    (err = i_dls_devnet_setzid(ddp, zoneid, B_FALSE)) != 0)
+		    (err = i_dls_devnet_setzid(ddp, zoneid, B_FALSE,
+		    B_FALSE)) != 0)
 			(void) dls_devnet_unset(macname, &linkid, B_TRUE);
 		/*
 		 * The kstat subsystem holds its own locks (rather perimeter)
@@ -887,7 +938,7 @@ done:
 		 * lock hierarchy is kstat locks -> i_dls_devnet_lock.
 		 */
 		if (stat_create)
-			dls_devnet_stat_create(ddp, zoneid);
+			dls_devnet_stat_create(ddp, zoneid, zoneid);
 		if (ddpp != NULL)
 			*ddpp = ddp;
 	}
@@ -924,17 +975,78 @@ dls_devnet_unset(const char *macname, datalink_id_t *id, boolean_t wait)
 	ASSERT(ddp->dd_ref != 0);
 	if ((ddp->dd_ref != 1) || (!wait &&
 	    (ddp->dd_tref != 0 || ddp->dd_prop_taskid != NULL))) {
-		mutex_exit(&ddp->dd_mutex);
-		rw_exit(&i_dls_devnet_lock);
-		return (EBUSY);
+		int zstatus = 0;
+
+		/*
+		 * There are a couple of alternatives that might be going on
+		 * here; a) the zone is shutting down and it has a transient
+		 * link assigned, in which case we want to clean it up instead
+		 * of moving it back to the global zone, or b) its possible
+		 * that we're trying to clean up an orphaned vnic that was
+		 * delegated to a zone and which wasn't cleaned up properly
+		 * when the zone went away.  Check for either of these cases
+		 * before we simply return EBUSY.
+		 *
+		 * zstatus indicates which situation we are dealing with:
+		 *	 0 - means return EBUSY
+		 *	 1 - means case (a), cleanup transient link
+		 *	-1 - means case (b), orphained VNIC
+		 */
+		if (ddp->dd_ref > 1 && ddp->dd_zid != GLOBAL_ZONEID) {
+			zone_t	*zp;
+
+			if ((zp = zone_find_by_id(ddp->dd_zid)) == NULL) {
+				zstatus = -1;
+			} else {
+				if (ddp->dd_transient) {
+					zone_status_t s = zone_status_get(zp);
+
+					if (s >= ZONE_IS_SHUTTING_DOWN)
+						zstatus = 1;
+				}
+				zone_rele(zp);
+			}
+		}
+
+		if (zstatus == 0) {
+			mutex_exit(&ddp->dd_mutex);
+			rw_exit(&i_dls_devnet_lock);
+			return (EBUSY);
+		}
+
+		/*
+		 * We want to delete the link, reset ref to 1;
+		 */
+		if (zstatus == -1)
+			/* Log a warning, but continue in this case */
+			cmn_err(CE_WARN, "clear orphaned datalink: %s\n",
+			    ddp->dd_linkname);
+		ddp->dd_ref = 1;
 	}
 
 	ddp->dd_flags |= DD_CONDEMNED;
 	ddp->dd_ref--;
 	*id = ddp->dd_linkid;
 
-	if (ddp->dd_zid != GLOBAL_ZONEID)
-		(void) i_dls_devnet_setzid(ddp, GLOBAL_ZONEID, B_FALSE);
+	if (ddp->dd_zid != GLOBAL_ZONEID) {
+		/*
+		 * We need to release the dd_mutex before we try and destroy the
+		 * stat. When we destroy it, we'll need to grab the lock for the
+		 * kstat but if there's a concurrent reader of the kstat, we'll
+		 * be blocked on it. This will lead to deadlock because these
+		 * kstats employ a ks_update function (dls_devnet_stat_update)
+		 * which needs the dd_mutex that we currently hold.
+		 *
+		 * Because we've already flagged the dls_devnet_t as
+		 * DD_CONDEMNED and we still have a write lock on
+		 * i_dls_devnet_lock, we should be able to release the dd_mutex.
+		 */
+		mutex_exit(&ddp->dd_mutex);
+		dls_devnet_stat_destroy(ddp, ddp->dd_zid);
+		mutex_enter(&ddp->dd_mutex);
+		(void) i_dls_devnet_setzid(ddp, GLOBAL_ZONEID, B_FALSE,
+		    B_FALSE);
+	}
 
 	/*
 	 * Remove this dls_devnet_t from the hash table.
@@ -960,8 +1072,15 @@ dls_devnet_unset(const char *macname, datalink_id_t *id, boolean_t wait)
 		ASSERT(ddp->dd_tref == 0 && ddp->dd_prop_taskid == NULL);
 	}
 
-	if (ddp->dd_linkid != DATALINK_INVALID_LINKID)
+	if (ddp->dd_linkid != DATALINK_INVALID_LINKID) {
+		/*
+		 * See the earlier call in this function for an explanation.
+		 */
+		mutex_exit(&ddp->dd_mutex);
 		dls_devnet_stat_destroy(ddp, ddp->dd_owner_zid);
+		mutex_enter(&ddp->dd_mutex);
+	}
+
 
 	ddp->dd_prop_loaded = B_FALSE;
 	ddp->dd_linkid = DATALINK_INVALID_LINKID;
@@ -1111,7 +1230,7 @@ dls_devnet_rele(dls_devnet_t *ddp)
 }
 
 static int
-dls_devnet_hold_by_name(const char *link, dls_devnet_t **ddpp)
+dls_devnet_hold_by_name(const char *link, dls_devnet_t **ddpp, zoneid_t zid)
 {
 	char			drv[MAXLINKNAMELEN];
 	uint_t			ppa;
@@ -1121,7 +1240,7 @@ dls_devnet_hold_by_name(const char *link, dls_devnet_t **ddpp)
 	dls_dev_handle_t	ddh;
 	int			err;
 
-	if ((err = dls_mgmt_get_linkid(link, &linkid)) == 0)
+	if ((err = dls_mgmt_get_linkid_in_zone(link, &linkid, zid)) == 0)
 		return (dls_devnet_hold(linkid, ddpp));
 
 	/*
@@ -1261,9 +1380,15 @@ dls_devnet_phydev(datalink_id_t vlanid, dev_t *devp)
  *
  *    This case does not change the <link name, linkid> mapping, so the link's
  *    kstats need to be updated with using name associated the given id2.
+ *
+ * The zonename parameter is used to allow us to create a VNIC in the global
+ * zone which is assigned to a non-global zone.  Since there is a race condition
+ * in the create process if two VNICs have the same name, we need to rename it
+ * after it has been assigned to the zone.
  */
 int
-dls_devnet_rename(datalink_id_t id1, datalink_id_t id2, const char *link)
+dls_devnet_rename(datalink_id_t id1, datalink_id_t id2, const char *link,
+    boolean_t zoneinit)
 {
 	dls_dev_handle_t	ddh = NULL;
 	int			err = 0;
@@ -1313,13 +1438,16 @@ dls_devnet_rename(datalink_id_t id1, datalink_id_t id2, const char *link)
 	 * is currently accessing the link kstats, or if the link is on-loan
 	 * to a non-global zone. Then set the DD_KSTAT_CHANGING flag to
 	 * prevent any access to the kstats while we delete and recreate
-	 * kstats below.
+	 * kstats below.  However, we skip this check if we're renaming the
+	 * vnic as part of bringing it up for a zone.
 	 */
 	mutex_enter(&ddp->dd_mutex);
-	if (ddp->dd_ref > 1) {
-		mutex_exit(&ddp->dd_mutex);
-		err = EBUSY;
-		goto done;
+	if (!zoneinit) {
+		if (ddp->dd_ref > 1) {
+			mutex_exit(&ddp->dd_mutex);
+			err = EBUSY;
+			goto done;
+		}
 	}
 
 	ddp->dd_flags |= DD_KSTAT_CHANGING;
@@ -1333,7 +1461,15 @@ dls_devnet_rename(datalink_id_t id1, datalink_id_t id2, const char *link)
 		/* rename mac client name and its flow if exists */
 		if ((err = mac_open(ddp->dd_mac, &mh)) != 0)
 			goto done;
-		(void) mac_rename_primary(mh, link);
+		if (zoneinit) {
+			char tname[MAXLINKNAMELEN];
+
+			(void) snprintf(tname, sizeof (tname), "z%d_%s",
+			    ddp->dd_zid, link);
+			(void) mac_rename_primary(mh, tname);
+		} else {
+			(void) mac_rename_primary(mh, link);
+		}
 		mac_close(mh);
 		goto done;
 	}
@@ -1406,7 +1542,7 @@ done:
 	 */
 	rw_exit(&i_dls_devnet_lock);
 	if (err == 0)
-		dls_devnet_stat_rename(ddp);
+		dls_devnet_stat_rename(ddp, zoneinit);
 
 	if (clear_dd_flag) {
 		mutex_enter(&ddp->dd_mutex);
@@ -1421,7 +1557,8 @@ done:
 }
 
 static int
-i_dls_devnet_setzid(dls_devnet_t *ddp, zoneid_t new_zoneid, boolean_t setprop)
+i_dls_devnet_setzid(dls_devnet_t *ddp, zoneid_t new_zoneid, boolean_t setprop,
+    boolean_t transient)
 {
 	int			err;
 	mac_perim_handle_t	mph;
@@ -1454,6 +1591,7 @@ i_dls_devnet_setzid(dls_devnet_t *ddp, zoneid_t new_zoneid, boolean_t setprop)
 	}
 	if ((err = dls_link_setzid(ddp->dd_mac, new_zoneid)) == 0) {
 		ddp->dd_zid = new_zoneid;
+		ddp->dd_transient = transient;
 		devnet_need_rebuild = B_TRUE;
 	}
 
@@ -1468,7 +1606,7 @@ done:
 }
 
 int
-dls_devnet_setzid(dls_dl_handle_t ddh, zoneid_t new_zid)
+dls_devnet_setzid(dls_dl_handle_t ddh, zoneid_t new_zid, boolean_t transient)
 {
 	dls_devnet_t	*ddp;
 	int		err;
@@ -1490,7 +1628,7 @@ dls_devnet_setzid(dls_dl_handle_t ddh, zoneid_t new_zid)
 		refheld = B_TRUE;
 	}
 
-	if ((err = i_dls_devnet_setzid(ddh, new_zid, B_TRUE)) != 0) {
+	if ((err = i_dls_devnet_setzid(ddh, new_zid, B_TRUE, transient)) != 0) {
 		if (refheld)
 			dls_devnet_rele(ddp);
 		return (err);
@@ -1507,7 +1645,7 @@ dls_devnet_setzid(dls_dl_handle_t ddh, zoneid_t new_zid)
 	if (old_zid != GLOBAL_ZONEID)
 		dls_devnet_stat_destroy(ddh, old_zid);
 	if (new_zid != GLOBAL_ZONEID)
-		dls_devnet_stat_create(ddh, new_zid);
+		dls_devnet_stat_create(ddh, new_zid, new_zid);
 
 	return (0);
 }
@@ -1545,15 +1683,19 @@ dls_devnet_islinkvisible(datalink_id_t linkid, zoneid_t zoneid)
  * Access a vanity naming node.
  */
 int
-dls_devnet_open(const char *link, dls_dl_handle_t *dhp, dev_t *devp)
+dls_devnet_open_in_zone(const char *link, dls_dl_handle_t *dhp, dev_t *devp,
+    zoneid_t zid)
 {
 	dls_devnet_t	*ddp;
 	dls_link_t	*dlp;
-	zoneid_t	zid = getzoneid();
+	zoneid_t	czid = getzoneid();
 	int		err;
 	mac_perim_handle_t	mph;
 
-	if ((err = dls_devnet_hold_by_name(link, &ddp)) != 0)
+	if (czid != GLOBAL_ZONEID && czid != zid)
+		return (ENOENT);
+
+	if ((err = dls_devnet_hold_by_name(link, &ddp, zid)) != 0)
 		return (err);
 
 	dls_devnet_prop_task_wait(ddp);
@@ -1586,6 +1728,12 @@ dls_devnet_open(const char *link, dls_dl_handle_t *dhp, dev_t *devp)
 	return (0);
 }
 
+int
+dls_devnet_open(const char *link, dls_dl_handle_t *dhp, dev_t *devp)
+{
+	return (dls_devnet_open_in_zone(link, dhp, devp, getzoneid()));
+}
+
 /*
  * Close access to a vanity naming node.
  */
diff --git a/usr/src/uts/common/io/dls/dls_stat.c b/usr/src/uts/common/io/dls/dls_stat.c
index 51e4be7260..82dceff278 100644
--- a/usr/src/uts/common/io/dls/dls_stat.c
+++ b/usr/src/uts/common/io/dls/dls_stat.c
@@ -21,6 +21,7 @@
 /*
  * Copyright 2009 Sun Microsystems, Inc.  All rights reserved.
  * Use is subject to license terms.
+ * Copyright 2011 Joyent, Inc.  All rights reserved.
  */
 
 /*
@@ -30,30 +31,33 @@
 #include <sys/dld_impl.h>
 #include <sys/mac_ether.h>
 
-static mac_stat_info_t	i_dls_si[] = {
-	{ MAC_STAT_IFSPEED, "ifspeed", KSTAT_DATA_UINT64, 0 },
-	{ MAC_STAT_MULTIRCV, "multircv", KSTAT_DATA_UINT32, 0 },
-	{ MAC_STAT_BRDCSTRCV, "brdcstrcv", KSTAT_DATA_UINT32, 0 },
-	{ MAC_STAT_MULTIXMT, "multixmt", KSTAT_DATA_UINT32, 0 },
-	{ MAC_STAT_BRDCSTXMT, "brdcstxmt", KSTAT_DATA_UINT32, 0 },
-	{ MAC_STAT_NORCVBUF, "norcvbuf", KSTAT_DATA_UINT32, 0 },
-	{ MAC_STAT_IERRORS, "ierrors", KSTAT_DATA_UINT32, 0 },
-	{ MAC_STAT_NOXMTBUF, "noxmtbuf", KSTAT_DATA_UINT32, 0 },
-	{ MAC_STAT_OERRORS, "oerrors", KSTAT_DATA_UINT32, 0 },
-	{ MAC_STAT_COLLISIONS, "collisions", KSTAT_DATA_UINT32, 0 },
-	{ MAC_STAT_RBYTES, "rbytes", KSTAT_DATA_UINT32, 0 },
-	{ MAC_STAT_IPACKETS, "ipackets", KSTAT_DATA_UINT32, 0 },
-	{ MAC_STAT_OBYTES, "obytes", KSTAT_DATA_UINT32, 0 },
-	{ MAC_STAT_OPACKETS, "opackets", KSTAT_DATA_UINT32, 0 },
-	{ MAC_STAT_RBYTES, "rbytes64", KSTAT_DATA_UINT64, 0 },
-	{ MAC_STAT_IPACKETS, "ipackets64", KSTAT_DATA_UINT64, 0 },
-	{ MAC_STAT_OBYTES, "obytes64", KSTAT_DATA_UINT64, 0 },
-	{ MAC_STAT_OPACKETS, "opackets64", KSTAT_DATA_UINT64, 0 },
-	{ MAC_STAT_LINK_STATE, "link_state", KSTAT_DATA_UINT32,
-	    (uint64_t)LINK_STATE_UNKNOWN}
-};
-
-#define	STAT_INFO_COUNT	(sizeof (i_dls_si) / sizeof (i_dls_si[0]))
+/*
+ * structure for link kstats
+ */
+typedef struct {
+	kstat_named_t	dk_ifspeed;
+	kstat_named_t	dk_multircv;
+	kstat_named_t	dk_brdcstrcv;
+	kstat_named_t	dk_multixmt;
+	kstat_named_t	dk_brdcstxmt;
+	kstat_named_t	dk_norcvbuf;
+	kstat_named_t	dk_ierrors;
+	kstat_named_t	dk_noxmtbuf;
+	kstat_named_t	dk_oerrors;
+	kstat_named_t	dk_collisions;
+	kstat_named_t	dk_rbytes;
+	kstat_named_t	dk_ipackets;
+	kstat_named_t	dk_obytes;
+	kstat_named_t	dk_opackets;
+	kstat_named_t	dk_rbytes64;
+	kstat_named_t	dk_ipackets64;
+	kstat_named_t	dk_obytes64;
+	kstat_named_t	dk_opackets64;
+	kstat_named_t	dk_link_state;
+	kstat_named_t	dk_link_duplex;
+	kstat_named_t	dk_unknowns;
+	kstat_named_t	dk_zonename;
+} dls_kstat_t;
 
 /*
  * Exported functions.
@@ -61,42 +65,54 @@ static mac_stat_info_t	i_dls_si[] = {
 int
 dls_stat_update(kstat_t *ksp, dls_link_t *dlp, int rw)
 {
-	kstat_named_t	*knp;
-	uint_t		i;
-	uint64_t	val;
+	dls_kstat_t *dkp = ksp->ks_data;
 
 	if (rw != KSTAT_READ)
 		return (EACCES);
 
-	knp = (kstat_named_t *)ksp->ks_data;
-	for (i = 0; i < STAT_INFO_COUNT; i++) {
-		val = mac_stat_get(dlp->dl_mh, i_dls_si[i].msi_stat);
-
-		switch (i_dls_si[i].msi_type) {
-		case KSTAT_DATA_UINT64:
-			knp->value.ui64 = val;
-			break;
-		case KSTAT_DATA_UINT32:
-			knp->value.ui32 = (uint32_t)val;
-			break;
-		default:
-			ASSERT(B_FALSE);
-		}
-
-		knp++;
-	}
+	dkp->dk_ifspeed.value.ui64 = mac_stat_get(dlp->dl_mh, MAC_STAT_IFSPEED);
+	dkp->dk_multircv.value.ui32 = mac_stat_get(dlp->dl_mh,
+	    MAC_STAT_MULTIRCV);
+	dkp->dk_brdcstrcv.value.ui32 = mac_stat_get(dlp->dl_mh,
+	    MAC_STAT_BRDCSTRCV);
+	dkp->dk_multixmt.value.ui32 = mac_stat_get(dlp->dl_mh,
+	    MAC_STAT_MULTIXMT);
+	dkp->dk_brdcstxmt.value.ui32 = mac_stat_get(dlp->dl_mh,
+	    MAC_STAT_BRDCSTXMT);
+	dkp->dk_norcvbuf.value.ui32 = mac_stat_get(dlp->dl_mh,
+	    MAC_STAT_NORCVBUF);
+	dkp->dk_ierrors.value.ui32 = mac_stat_get(dlp->dl_mh, MAC_STAT_IERRORS);
+	dkp->dk_noxmtbuf.value.ui32 = mac_stat_get(dlp->dl_mh,
+	    MAC_STAT_NOXMTBUF);
+	dkp->dk_oerrors.value.ui32 = mac_stat_get(dlp->dl_mh, MAC_STAT_OERRORS);
+	dkp->dk_collisions.value.ui32 = mac_stat_get(dlp->dl_mh,
+	    MAC_STAT_COLLISIONS);
+	dkp->dk_rbytes.value.ui32 = mac_stat_get(dlp->dl_mh, MAC_STAT_RBYTES);
+	dkp->dk_ipackets.value.ui32 = mac_stat_get(dlp->dl_mh,
+	    MAC_STAT_IPACKETS);
+	dkp->dk_obytes.value.ui32 = mac_stat_get(dlp->dl_mh, MAC_STAT_OBYTES);
+	dkp->dk_opackets.value.ui32 = mac_stat_get(dlp->dl_mh,
+	    MAC_STAT_OPACKETS);
+	dkp->dk_rbytes64.value.ui64 = mac_stat_get(dlp->dl_mh, MAC_STAT_RBYTES);
+	dkp->dk_ipackets64.value.ui64 = mac_stat_get(dlp->dl_mh,
+	    MAC_STAT_IPACKETS);
+	dkp->dk_obytes64.value.ui64 = mac_stat_get(dlp->dl_mh, MAC_STAT_OBYTES);
+	dkp->dk_opackets64.value.ui64 = mac_stat_get(dlp->dl_mh,
+	    MAC_STAT_OPACKETS);
+	dkp->dk_link_state.value.ui32 = mac_stat_get(dlp->dl_mh,
+	    MAC_STAT_LINK_STATE);
 
 	/*
 	 * Ethernet specific kstat "link_duplex"
 	 */
 	if (dlp->dl_mip->mi_nativemedia != DL_ETHER) {
-		knp->value.ui32 = LINK_DUPLEX_UNKNOWN;
+		dkp->dk_link_duplex.value.ui32 = LINK_DUPLEX_UNKNOWN;
 	} else {
-		val = mac_stat_get(dlp->dl_mh, ETHER_STAT_LINK_DUPLEX);
-		knp->value.ui32 = (uint32_t)val;
+		dkp->dk_link_duplex.value.ui32 =
+		    (uint32_t)mac_stat_get(dlp->dl_mh, ETHER_STAT_LINK_DUPLEX);
 	}
-	knp++;
-	knp->value.ui32 = dlp->dl_unknowns;
+
+	dkp->dk_unknowns.value.ui32 = dlp->dl_unknowns;
 
 	return (0);
 }
@@ -104,30 +120,66 @@ dls_stat_update(kstat_t *ksp, dls_link_t *dlp, int rw)
 int
 dls_stat_create(const char *module, int instance, const char *name,
     zoneid_t zoneid, int (*update)(struct kstat *, int), void *private,
-    kstat_t **kspp)
+    kstat_t **kspp, zoneid_t newzoneid)
 {
 	kstat_t		*ksp;
-	kstat_named_t	*knp;
-	uint_t		i;
+	zone_t		*zone;
+	dls_kstat_t	*dkp;
 
 	if ((ksp = kstat_create_zone(module, instance, name, "net",
-	    KSTAT_TYPE_NAMED, STAT_INFO_COUNT + 2, 0, zoneid)) == NULL) {
+	    KSTAT_TYPE_NAMED, sizeof (dls_kstat_t) / sizeof (kstat_named_t),
+	    KSTAT_FLAG_VIRTUAL, zoneid)) == NULL) {
 		return (EINVAL);
 	}
 
 	ksp->ks_update = update;
 	ksp->ks_private = private;
+	dkp = ksp->ks_data = kmem_zalloc(sizeof (dls_kstat_t), KM_SLEEP);
+	if ((zone = zone_find_by_id(newzoneid)) != NULL) {
+		ksp->ks_data_size += strlen(zone->zone_name) + 1;
+	}
 
-	knp = (kstat_named_t *)ksp->ks_data;
-	for (i = 0; i < STAT_INFO_COUNT; i++) {
-		kstat_named_init(knp, i_dls_si[i].msi_name,
-		    i_dls_si[i].msi_type);
-		knp++;
+	kstat_named_init(&dkp->dk_ifspeed, "ifspeed", KSTAT_DATA_UINT64);
+	kstat_named_init(&dkp->dk_multircv, "multircv", KSTAT_DATA_UINT32);
+	kstat_named_init(&dkp->dk_brdcstrcv, "brdcstrcv", KSTAT_DATA_UINT32);
+	kstat_named_init(&dkp->dk_multixmt, "multixmt", KSTAT_DATA_UINT32);
+	kstat_named_init(&dkp->dk_brdcstxmt, "brdcstxmt", KSTAT_DATA_UINT32);
+	kstat_named_init(&dkp->dk_norcvbuf, "norcvbuf", KSTAT_DATA_UINT32);
+	kstat_named_init(&dkp->dk_ierrors, "ierrors", KSTAT_DATA_UINT32);
+	kstat_named_init(&dkp->dk_noxmtbuf, "noxmtbuf", KSTAT_DATA_UINT32);
+	kstat_named_init(&dkp->dk_oerrors, "oerrors", KSTAT_DATA_UINT32);
+	kstat_named_init(&dkp->dk_collisions, "collisions", KSTAT_DATA_UINT32);
+	kstat_named_init(&dkp->dk_rbytes, "rbytes", KSTAT_DATA_UINT32);
+	kstat_named_init(&dkp->dk_ipackets, "ipackets", KSTAT_DATA_UINT32);
+	kstat_named_init(&dkp->dk_obytes, "obytes", KSTAT_DATA_UINT32);
+	kstat_named_init(&dkp->dk_opackets, "opackets", KSTAT_DATA_UINT32);
+	kstat_named_init(&dkp->dk_rbytes64, "rbytes64", KSTAT_DATA_UINT64);
+	kstat_named_init(&dkp->dk_ipackets64, "ipackets64", KSTAT_DATA_UINT64);
+	kstat_named_init(&dkp->dk_obytes64, "obytes64", KSTAT_DATA_UINT64);
+	kstat_named_init(&dkp->dk_opackets64, "opackets64", KSTAT_DATA_UINT64);
+	kstat_named_init(&dkp->dk_link_state, "link_state", KSTAT_DATA_UINT32);
+	kstat_named_init(&dkp->dk_link_duplex, "link_duplex",
+		    KSTAT_DATA_UINT32);
+	kstat_named_init(&dkp->dk_unknowns, "unknowns", KSTAT_DATA_UINT32);
+	kstat_named_init(&dkp->dk_zonename, "zonename", KSTAT_DATA_STRING);
+
+	if (zone != NULL) {
+		kstat_named_setstr(&dkp->dk_zonename, zone->zone_name);
+		zone_rele(zone);
 	}
 
-	kstat_named_init(knp++, "link_duplex", KSTAT_DATA_UINT32);
-	kstat_named_init(knp, "unknowns", KSTAT_DATA_UINT32);
 	kstat_install(ksp);
 	*kspp = ksp;
 	return (0);
 }
+
+void
+dls_stat_delete(kstat_t *ksp)
+{
+	void *data;
+	if (ksp != NULL) {
+		data = ksp->ks_data;
+		kstat_delete(ksp);
+		kmem_free(data, sizeof (dls_kstat_t));
+	}
+}
diff --git a/usr/src/uts/common/io/dr_sas/THIRDPARTYLICENSE b/usr/src/uts/common/io/dr_sas/THIRDPARTYLICENSE
new file mode 100644
index 0000000000..00aefb6f51
--- /dev/null
+++ b/usr/src/uts/common/io/dr_sas/THIRDPARTYLICENSE
@@ -0,0 +1,32 @@
+/*
+ * MegaRAID device driver for SAS2.0 controllers
+ * Copyright (c) 2009, LSI Logic Corporation.
+ * All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions are met:
+ *
+ * 1. Redistributions of source code must retain the above copyright notice,
+ *    this list of conditions and the following disclaimer.
+ *
+ * 2. Redistributions in binary form must reproduce the above copyright notice,
+ *    this list of conditions and the following disclaimer in the documentation
+ *    and/or other materials provided with the distribution.
+ *
+ * 3. Neither the name of the author nor the names of its contributors may be
+ *    used to endorse or promote products derived from this software without
+ *    specific prior written permission.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
+ * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
+ * COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
+ * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
+ * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS
+ * OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
+ * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+ * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
+ * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH
+ * DAMAGE.
+ */
diff --git a/usr/src/uts/common/io/dr_sas/THIRDPARTYLICENSE.descrip b/usr/src/uts/common/io/dr_sas/THIRDPARTYLICENSE.descrip
new file mode 100644
index 0000000000..ac6d2d1b15
--- /dev/null
+++ b/usr/src/uts/common/io/dr_sas/THIRDPARTYLICENSE.descrip
@@ -0,0 +1 @@
+DR_SAS DRIVER
diff --git a/usr/src/uts/common/io/dr_sas/dr_sas.c b/usr/src/uts/common/io/dr_sas/dr_sas.c
new file mode 100644
index 0000000000..5b1dc82938
--- /dev/null
+++ b/usr/src/uts/common/io/dr_sas/dr_sas.c
@@ -0,0 +1,5506 @@
+/*
+ * dr_sas.c: source for dr_sas driver
+ *
+ * MegaRAID device driver for SAS2.0 controllers
+ * Copyright (c) 2008-2009, LSI Logic Corporation.
+ * All rights reserved.
+ *
+ * Version:
+ * Author:
+ *		Arun Chandrashekhar
+ *		Manju R
+ *        	Rajesh Prabhakaran
+ *        	Seokmann Ju
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions are met:
+ *
+ * 1. Redistributions of source code must retain the above copyright notice,
+ *    this list of conditions and the following disclaimer.
+ *
+ * 2. Redistributions in binary form must reproduce the above copyright notice,
+ *    this list of conditions and the following disclaimer in the documentation
+ *    and/or other materials provided with the distribution.
+ *
+ * 3. Neither the name of the author nor the names of its contributors may be
+ *    used to endorse or promote products derived from this software without
+ *    specific prior written permission.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
+ * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
+ * COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
+ * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
+ * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS
+ * OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
+ * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+ * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
+ * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH
+ * DAMAGE.
+ */
+
+/*
+ * Copyright 2009 Sun Microsystems, Inc.  All rights reserved.
+ * Use is subject to license terms.
+ */
+
+#include <sys/types.h>
+#include <sys/param.h>
+#include <sys/file.h>
+#include <sys/errno.h>
+#include <sys/open.h>
+#include <sys/cred.h>
+#include <sys/modctl.h>
+#include <sys/conf.h>
+#include <sys/devops.h>
+#include <sys/cmn_err.h>
+#include <sys/kmem.h>
+#include <sys/stat.h>
+#include <sys/mkdev.h>
+#include <sys/pci.h>
+#include <sys/scsi/scsi.h>
+#include <sys/ddi.h>
+#include <sys/sunddi.h>
+#include <sys/atomic.h>
+#include <sys/signal.h>
+#include <sys/fs/dv_node.h>	/* devfs_clean */
+
+#include "dr_sas.h"
+
+/*
+ * FMA header files
+ */
+#include <sys/ddifm.h>
+#include <sys/fm/protocol.h>
+#include <sys/fm/util.h>
+#include <sys/fm/io/ddi.h>
+
+/*
+ * Local static data
+ */
+static void	*drsas_state = NULL;
+static int 	debug_level_g = CL_NONE;
+
+#pragma weak scsi_hba_open
+#pragma weak scsi_hba_close
+#pragma weak scsi_hba_ioctl
+
+static ddi_dma_attr_t drsas_generic_dma_attr = {
+	DMA_ATTR_V0,		/* dma_attr_version */
+	0,			/* low DMA address range */
+	0xFFFFFFFFU,		/* high DMA address range */
+	0xFFFFFFFFU,		/* DMA counter register  */
+	8,			/* DMA address alignment */
+	0x07,			/* DMA burstsizes  */
+	1,			/* min DMA size */
+	0xFFFFFFFFU,		/* max DMA size */
+	0xFFFFFFFFU,		/* segment boundary */
+	DRSAS_MAX_SGE_CNT,	/* dma_attr_sglen */
+	512,			/* granularity of device */
+	0			/* bus specific DMA flags */
+};
+
+int32_t drsas_max_cap_maxxfer = 0x1000000;
+
+/*
+ * cb_ops contains base level routines
+ */
+static struct cb_ops drsas_cb_ops = {
+	drsas_open,		/* open */
+	drsas_close,		/* close */
+	nodev,			/* strategy */
+	nodev,			/* print */
+	nodev,			/* dump */
+	nodev,			/* read */
+	nodev,			/* write */
+	drsas_ioctl,		/* ioctl */
+	nodev,			/* devmap */
+	nodev,			/* mmap */
+	nodev,			/* segmap */
+	nochpoll,		/* poll */
+	nodev,			/* cb_prop_op */
+	0,			/* streamtab  */
+	D_NEW | D_HOTPLUG,	/* cb_flag */
+	CB_REV,			/* cb_rev */
+	nodev,			/* cb_aread */
+	nodev			/* cb_awrite */
+};
+
+/*
+ * dev_ops contains configuration routines
+ */
+static struct dev_ops drsas_ops = {
+	DEVO_REV,		/* rev, */
+	0,			/* refcnt */
+	drsas_getinfo,		/* getinfo */
+	nulldev,		/* identify */
+	nulldev,		/* probe */
+	drsas_attach,		/* attach */
+	drsas_detach,		/* detach */
+	drsas_reset,		/* reset */
+	&drsas_cb_ops,		/* char/block ops */
+	NULL,			/* bus ops */
+	NULL,			/* power */
+	ddi_quiesce_not_supported,		/* quiesce */
+};
+
+char _depends_on[] = "misc/scsi";
+
+static struct modldrv modldrv = {
+	&mod_driverops,		/* module type - driver */
+	DRSAS_VERSION,
+	&drsas_ops,		/* driver ops */
+};
+
+static struct modlinkage modlinkage = {
+	MODREV_1,	/* ml_rev - must be MODREV_1 */
+	&modldrv,	/* ml_linkage */
+	NULL		/* end of driver linkage */
+};
+
+static struct ddi_device_acc_attr endian_attr = {
+	DDI_DEVICE_ATTR_V0,
+	DDI_STRUCTURE_LE_ACC,
+	DDI_STRICTORDER_ACC
+};
+
+
+/*
+ * ************************************************************************** *
+ *                                                                            *
+ *         common entry points - for loadable kernel modules                  *
+ *                                                                            *
+ * ************************************************************************** *
+ */
+
+int
+_init(void)
+{
+	int ret;
+
+	con_log(CL_ANN1, (CE_NOTE, "chkpnt:%s:%d", __func__, __LINE__));
+
+	ret = ddi_soft_state_init(&drsas_state,
+	    sizeof (struct drsas_instance), 0);
+
+	if (ret != DDI_SUCCESS) {
+		con_log(CL_ANN, (CE_WARN, "dr_sas: could not init state"));
+		return (ret);
+	}
+
+	if ((ret = scsi_hba_init(&modlinkage)) != DDI_SUCCESS) {
+		con_log(CL_ANN, (CE_WARN, "dr_sas: could not init scsi hba"));
+		ddi_soft_state_fini(&drsas_state);
+		return (ret);
+	}
+
+	ret = mod_install(&modlinkage);
+
+	if (ret != DDI_SUCCESS) {
+		con_log(CL_ANN, (CE_WARN, "dr_sas: mod_install failed"));
+		scsi_hba_fini(&modlinkage);
+		ddi_soft_state_fini(&drsas_state);
+	}
+
+	return (ret);
+}
+
+int
+_info(struct modinfo *modinfop)
+{
+	con_log(CL_ANN1, (CE_NOTE, "chkpnt:%s:%d", __func__, __LINE__));
+
+	return (mod_info(&modlinkage, modinfop));
+}
+
+int
+_fini(void)
+{
+	int ret;
+
+	con_log(CL_ANN1, (CE_NOTE, "chkpnt:%s:%d", __func__, __LINE__));
+
+	if ((ret = mod_remove(&modlinkage)) != DDI_SUCCESS)
+		return (ret);
+
+	scsi_hba_fini(&modlinkage);
+
+	ddi_soft_state_fini(&drsas_state);
+
+	return (ret);
+}
+
+
+/*
+ * ************************************************************************** *
+ *                                                                            *
+ *               common entry points - for autoconfiguration                  *
+ *                                                                            *
+ * ************************************************************************** *
+ */
+
+static int
+drsas_attach(dev_info_t *dip, ddi_attach_cmd_t cmd)
+{
+	int		instance_no;
+	int		nregs;
+	uint8_t		added_isr_f = 0;
+	uint8_t		added_soft_isr_f = 0;
+	uint8_t		create_devctl_node_f = 0;
+	uint8_t		create_scsi_node_f = 0;
+	uint8_t		create_ioc_node_f = 0;
+	uint8_t		tran_alloc_f = 0;
+	uint8_t 	irq;
+	uint16_t	vendor_id;
+	uint16_t	device_id;
+	uint16_t	subsysvid;
+	uint16_t	subsysid;
+	uint16_t	command;
+	off_t		reglength = 0;
+	int		intr_types = 0;
+	char		*data;
+	int		msi_enable = 0;
+
+	scsi_hba_tran_t		*tran;
+	ddi_dma_attr_t  tran_dma_attr;
+	struct drsas_instance	*instance;
+
+	con_log(CL_ANN1, (CE_NOTE, "chkpnt:%s:%d", __func__, __LINE__));
+
+	/* CONSTCOND */
+	ASSERT(NO_COMPETING_THREADS);
+
+	instance_no = ddi_get_instance(dip);
+
+	/*
+	 * check to see whether this device is in a DMA-capable slot.
+	 */
+	if (ddi_slaveonly(dip) == DDI_SUCCESS) {
+		con_log(CL_ANN, (CE_WARN,
+		    "dr_sas%d: Device in slave-only slot, unused",
+		    instance_no));
+		return (DDI_FAILURE);
+	}
+
+	switch (cmd) {
+		case DDI_ATTACH:
+			con_log(CL_DLEVEL1, (CE_NOTE, "dr_sas: DDI_ATTACH"));
+			/* allocate the soft state for the instance */
+			if (ddi_soft_state_zalloc(drsas_state, instance_no)
+			    != DDI_SUCCESS) {
+				con_log(CL_ANN, (CE_WARN,
+				    "dr_sas%d: Failed to allocate soft state",
+				    instance_no));
+
+				return (DDI_FAILURE);
+			}
+
+			instance = (struct drsas_instance *)ddi_get_soft_state
+			    (drsas_state, instance_no);
+
+			if (instance == NULL) {
+				con_log(CL_ANN, (CE_WARN,
+				    "dr_sas%d: Bad soft state", instance_no));
+
+				ddi_soft_state_free(drsas_state, instance_no);
+
+				return (DDI_FAILURE);
+			}
+
+			bzero((caddr_t)instance,
+			    sizeof (struct drsas_instance));
+
+			instance->func_ptr = kmem_zalloc(
+			    sizeof (struct drsas_func_ptr), KM_SLEEP);
+			ASSERT(instance->func_ptr);
+
+			/* Setup the PCI configuration space handles */
+			if (pci_config_setup(dip, &instance->pci_handle) !=
+			    DDI_SUCCESS) {
+				con_log(CL_ANN, (CE_WARN,
+				    "dr_sas%d: pci config setup failed ",
+				    instance_no));
+
+				kmem_free(instance->func_ptr,
+				    sizeof (struct drsas_func_ptr));
+				ddi_soft_state_free(drsas_state, instance_no);
+
+				return (DDI_FAILURE);
+			}
+
+			if (ddi_dev_nregs(dip, &nregs) != DDI_SUCCESS) {
+				con_log(CL_ANN, (CE_WARN,
+				    "dr_sas: failed to get registers."));
+
+				pci_config_teardown(&instance->pci_handle);
+				kmem_free(instance->func_ptr,
+				    sizeof (struct drsas_func_ptr));
+				ddi_soft_state_free(drsas_state, instance_no);
+
+				return (DDI_FAILURE);
+			}
+
+			vendor_id = pci_config_get16(instance->pci_handle,
+			    PCI_CONF_VENID);
+			device_id = pci_config_get16(instance->pci_handle,
+			    PCI_CONF_DEVID);
+
+			subsysvid = pci_config_get16(instance->pci_handle,
+			    PCI_CONF_SUBVENID);
+			subsysid = pci_config_get16(instance->pci_handle,
+			    PCI_CONF_SUBSYSID);
+
+			pci_config_put16(instance->pci_handle, PCI_CONF_COMM,
+			    (pci_config_get16(instance->pci_handle,
+			    PCI_CONF_COMM) | PCI_COMM_ME));
+			irq = pci_config_get8(instance->pci_handle,
+			    PCI_CONF_ILINE);
+
+			con_log(CL_DLEVEL1, (CE_CONT, "dr_sas%d: "
+			    "0x%x:0x%x 0x%x:0x%x, irq:%d drv-ver:%s",
+			    instance_no, vendor_id, device_id, subsysvid,
+			    subsysid, irq, DRSAS_VERSION));
+
+			/* enable bus-mastering */
+			command = pci_config_get16(instance->pci_handle,
+			    PCI_CONF_COMM);
+
+			if (!(command & PCI_COMM_ME)) {
+				command |= PCI_COMM_ME;
+
+				pci_config_put16(instance->pci_handle,
+				    PCI_CONF_COMM, command);
+
+				con_log(CL_ANN, (CE_CONT, "dr_sas%d: "
+				    "enable bus-mastering", instance_no));
+			} else {
+				con_log(CL_DLEVEL1, (CE_CONT, "dr_sas%d: "
+				"bus-mastering already set", instance_no));
+			}
+
+			/* initialize function pointers */
+			if ((device_id == PCI_DEVICE_ID_LSI_2108VDE) ||
+			    (device_id == PCI_DEVICE_ID_LSI_2108V)) {
+				con_log(CL_DLEVEL1, (CE_CONT, "dr_sas%d: "
+				    "2108V/DE detected", instance_no));
+				instance->func_ptr->read_fw_status_reg =
+				    read_fw_status_reg_ppc;
+				instance->func_ptr->issue_cmd = issue_cmd_ppc;
+				instance->func_ptr->issue_cmd_in_sync_mode =
+				    issue_cmd_in_sync_mode_ppc;
+				instance->func_ptr->issue_cmd_in_poll_mode =
+				    issue_cmd_in_poll_mode_ppc;
+				instance->func_ptr->enable_intr =
+				    enable_intr_ppc;
+				instance->func_ptr->disable_intr =
+				    disable_intr_ppc;
+				instance->func_ptr->intr_ack = intr_ack_ppc;
+			} else {
+				con_log(CL_ANN, (CE_WARN,
+				    "dr_sas: Invalid device detected"));
+
+				pci_config_teardown(&instance->pci_handle);
+				kmem_free(instance->func_ptr,
+				    sizeof (struct drsas_func_ptr));
+				ddi_soft_state_free(drsas_state, instance_no);
+
+				return (DDI_FAILURE);
+			}
+
+			instance->baseaddress = pci_config_get32(
+			    instance->pci_handle, PCI_CONF_BASE0);
+			instance->baseaddress &= 0x0fffc;
+
+			instance->dip		= dip;
+			instance->vendor_id	= vendor_id;
+			instance->device_id	= device_id;
+			instance->subsysvid	= subsysvid;
+			instance->subsysid	= subsysid;
+			instance->instance	= instance_no;
+
+			/* Initialize FMA */
+			instance->fm_capabilities = ddi_prop_get_int(
+			    DDI_DEV_T_ANY, instance->dip, DDI_PROP_DONTPASS,
+			    "fm-capable", DDI_FM_EREPORT_CAPABLE |
+			    DDI_FM_ACCCHK_CAPABLE | DDI_FM_DMACHK_CAPABLE
+			    | DDI_FM_ERRCB_CAPABLE);
+
+			drsas_fm_init(instance);
+
+			/* Initialize Interrupts */
+			if ((ddi_dev_regsize(instance->dip,
+			    REGISTER_SET_IO_2108, &reglength) != DDI_SUCCESS) ||
+			    reglength < MINIMUM_MFI_MEM_SZ) {
+				return (DDI_FAILURE);
+			}
+			if (reglength > DEFAULT_MFI_MEM_SZ) {
+				reglength = DEFAULT_MFI_MEM_SZ;
+				con_log(CL_DLEVEL1, (CE_NOTE,
+				    "dr_sas: register length to map is "
+				    "0x%lx bytes", reglength));
+			}
+			if (ddi_regs_map_setup(instance->dip,
+			    REGISTER_SET_IO_2108, &instance->regmap, 0,
+			    reglength, &endian_attr, &instance->regmap_handle)
+			    != DDI_SUCCESS) {
+				con_log(CL_ANN, (CE_NOTE,
+				    "dr_sas: couldn't map control registers"));
+				goto fail_attach;
+			}
+
+			/*
+			 * Disable Interrupt Now.
+			 * Setup Software interrupt
+			 */
+			instance->func_ptr->disable_intr(instance);
+
+			msi_enable = 0;
+			if (ddi_prop_lookup_string(DDI_DEV_T_ANY, dip, 0,
+			    "drsas-enable-msi", &data) == DDI_SUCCESS) {
+				if (strncmp(data, "yes", 3) == 0) {
+					msi_enable = 1;
+					con_log(CL_ANN, (CE_WARN,
+					    "msi_enable = %d ENABLED",
+					    msi_enable));
+				}
+				ddi_prop_free(data);
+			}
+
+			con_log(CL_DLEVEL1, (CE_WARN, "msi_enable = %d",
+			    msi_enable));
+
+			/* Check for all supported interrupt types */
+			if (ddi_intr_get_supported_types(
+			    dip, &intr_types) != DDI_SUCCESS) {
+				con_log(CL_ANN, (CE_WARN,
+				    "ddi_intr_get_supported_types() failed"));
+				goto fail_attach;
+			}
+
+			con_log(CL_DLEVEL1, (CE_NOTE,
+			    "ddi_intr_get_supported_types() ret: 0x%x",
+			    intr_types));
+
+			/* Initialize and Setup Interrupt handler */
+			if (msi_enable && (intr_types & DDI_INTR_TYPE_MSIX)) {
+				if (drsas_add_intrs(instance,
+				    DDI_INTR_TYPE_MSIX) != DDI_SUCCESS) {
+					con_log(CL_ANN, (CE_WARN,
+					    "MSIX interrupt query failed"));
+					goto fail_attach;
+				}
+				instance->intr_type = DDI_INTR_TYPE_MSIX;
+			} else if (msi_enable && (intr_types &
+			    DDI_INTR_TYPE_MSI)) {
+				if (drsas_add_intrs(instance,
+				    DDI_INTR_TYPE_MSI) != DDI_SUCCESS) {
+					con_log(CL_ANN, (CE_WARN,
+					    "MSI interrupt query failed"));
+					goto fail_attach;
+				}
+				instance->intr_type = DDI_INTR_TYPE_MSI;
+			} else if (intr_types & DDI_INTR_TYPE_FIXED) {
+				msi_enable = 0;
+				if (drsas_add_intrs(instance,
+				    DDI_INTR_TYPE_FIXED) != DDI_SUCCESS) {
+					con_log(CL_ANN, (CE_WARN,
+					    "FIXED interrupt query failed"));
+					goto fail_attach;
+				}
+				instance->intr_type = DDI_INTR_TYPE_FIXED;
+			} else {
+				con_log(CL_ANN, (CE_WARN, "Device cannot "
+				    "suppport either FIXED or MSI/X "
+				    "interrupts"));
+				goto fail_attach;
+			}
+
+			added_isr_f = 1;
+
+			/* setup the mfi based low level driver */
+			if (init_mfi(instance) != DDI_SUCCESS) {
+				con_log(CL_ANN, (CE_WARN, "dr_sas: "
+				"could not initialize the low level driver"));
+
+				goto fail_attach;
+			}
+
+			/* Initialize all Mutex */
+			INIT_LIST_HEAD(&instance->completed_pool_list);
+			mutex_init(&instance->completed_pool_mtx,
+			    "completed_pool_mtx", MUTEX_DRIVER,
+			    DDI_INTR_PRI(instance->intr_pri));
+
+			mutex_init(&instance->int_cmd_mtx, "int_cmd_mtx",
+			    MUTEX_DRIVER, DDI_INTR_PRI(instance->intr_pri));
+			cv_init(&instance->int_cmd_cv, NULL, CV_DRIVER, NULL);
+
+			mutex_init(&instance->cmd_pool_mtx, "cmd_pool_mtx",
+			    MUTEX_DRIVER, DDI_INTR_PRI(instance->intr_pri));
+
+			/* Register our soft-isr for highlevel interrupts. */
+			instance->isr_level = instance->intr_pri;
+			if (instance->isr_level == HIGH_LEVEL_INTR) {
+				if (ddi_add_softintr(dip, DDI_SOFTINT_HIGH,
+				    &instance->soft_intr_id, NULL, NULL,
+				    drsas_softintr, (caddr_t)instance) !=
+				    DDI_SUCCESS) {
+					con_log(CL_ANN, (CE_WARN,
+					    " Software ISR did not register"));
+
+					goto fail_attach;
+				}
+
+				added_soft_isr_f = 1;
+			}
+
+			/* Allocate a transport structure */
+			tran = scsi_hba_tran_alloc(dip, SCSI_HBA_CANSLEEP);
+
+			if (tran == NULL) {
+				con_log(CL_ANN, (CE_WARN,
+				    "scsi_hba_tran_alloc failed"));
+				goto fail_attach;
+			}
+
+			tran_alloc_f = 1;
+
+			instance->tran = tran;
+
+			tran->tran_hba_private	= instance;
+			tran->tran_tgt_init	= drsas_tran_tgt_init;
+			tran->tran_tgt_probe	= scsi_hba_probe;
+			tran->tran_tgt_free	= drsas_tran_tgt_free;
+			tran->tran_init_pkt	= drsas_tran_init_pkt;
+			tran->tran_start	= drsas_tran_start;
+			tran->tran_abort	= drsas_tran_abort;
+			tran->tran_reset	= drsas_tran_reset;
+			tran->tran_getcap	= drsas_tran_getcap;
+			tran->tran_setcap	= drsas_tran_setcap;
+			tran->tran_destroy_pkt	= drsas_tran_destroy_pkt;
+			tran->tran_dmafree	= drsas_tran_dmafree;
+			tran->tran_sync_pkt	= drsas_tran_sync_pkt;
+			tran->tran_bus_config	= drsas_tran_bus_config;
+
+			tran_dma_attr = drsas_generic_dma_attr;
+			tran_dma_attr.dma_attr_sgllen = instance->max_num_sge;
+
+			/* Attach this instance of the hba */
+			if (scsi_hba_attach_setup(dip, &tran_dma_attr, tran, 0)
+			    != DDI_SUCCESS) {
+				con_log(CL_ANN, (CE_WARN,
+				    "scsi_hba_attach failed"));
+
+				goto fail_attach;
+			}
+
+			/* create devctl node for cfgadm command */
+			if (ddi_create_minor_node(dip, "devctl",
+			    S_IFCHR, INST2DEVCTL(instance_no),
+			    DDI_NT_SCSI_NEXUS, 0) == DDI_FAILURE) {
+				con_log(CL_ANN, (CE_WARN,
+				    "dr_sas: failed to create devctl node."));
+
+				goto fail_attach;
+			}
+
+			create_devctl_node_f = 1;
+
+			/* create scsi node for cfgadm command */
+			if (ddi_create_minor_node(dip, "scsi", S_IFCHR,
+			    INST2SCSI(instance_no),
+			    DDI_NT_SCSI_ATTACHMENT_POINT, 0) ==
+			    DDI_FAILURE) {
+				con_log(CL_ANN, (CE_WARN,
+				    "dr_sas: failed to create scsi node."));
+
+				goto fail_attach;
+			}
+
+			create_scsi_node_f = 1;
+
+			(void) sprintf(instance->iocnode, "%d:lsirdctl",
+			    instance_no);
+
+			/*
+			 * Create a node for applications
+			 * for issuing ioctl to the driver.
+			 */
+			if (ddi_create_minor_node(dip, instance->iocnode,
+			    S_IFCHR, INST2LSIRDCTL(instance_no),
+			    DDI_PSEUDO, 0) == DDI_FAILURE) {
+				con_log(CL_ANN, (CE_WARN,
+				    "dr_sas: failed to create ioctl node."));
+
+				goto fail_attach;
+			}
+
+			create_ioc_node_f = 1;
+
+			/* Create a taskq to handle dr events */
+			if ((instance->taskq = ddi_taskq_create(dip,
+			    "drsas_dr_taskq", 1,
+			    TASKQ_DEFAULTPRI, 0)) == NULL) {
+				con_log(CL_ANN, (CE_WARN,
+				    "dr_sas: failed to create taskq "));
+				instance->taskq = NULL;
+				goto fail_attach;
+			}
+
+			/* enable interrupt */
+			instance->func_ptr->enable_intr(instance);
+
+			/* initiate AEN */
+			if (start_mfi_aen(instance)) {
+				con_log(CL_ANN, (CE_WARN,
+				    "dr_sas: failed to initiate AEN."));
+				goto fail_initiate_aen;
+			}
+
+			con_log(CL_DLEVEL1, (CE_NOTE,
+			    "AEN started for instance %d.", instance_no));
+
+			/* Finally! We are on the air.  */
+			ddi_report_dev(dip);
+
+			if (drsas_check_acc_handle(instance->regmap_handle) !=
+			    DDI_SUCCESS) {
+				goto fail_attach;
+			}
+			if (drsas_check_acc_handle(instance->pci_handle) !=
+			    DDI_SUCCESS) {
+				goto fail_attach;
+			}
+			instance->dr_ld_list =
+			    kmem_zalloc(MRDRV_MAX_LD * sizeof (struct drsas_ld),
+			    KM_SLEEP);
+			break;
+		case DDI_PM_RESUME:
+			con_log(CL_ANN, (CE_NOTE,
+			    "dr_sas: DDI_PM_RESUME"));
+			break;
+		case DDI_RESUME:
+			con_log(CL_ANN, (CE_NOTE,
+			    "dr_sas: DDI_RESUME"));
+			break;
+		default:
+			con_log(CL_ANN, (CE_WARN,
+			    "dr_sas: invalid attach cmd=%x", cmd));
+			return (DDI_FAILURE);
+	}
+
+	return (DDI_SUCCESS);
+
+fail_initiate_aen:
+fail_attach:
+	if (create_devctl_node_f) {
+		ddi_remove_minor_node(dip, "devctl");
+	}
+
+	if (create_scsi_node_f) {
+		ddi_remove_minor_node(dip, "scsi");
+	}
+
+	if (create_ioc_node_f) {
+		ddi_remove_minor_node(dip, instance->iocnode);
+	}
+
+	if (tran_alloc_f) {
+		scsi_hba_tran_free(tran);
+	}
+
+
+	if (added_soft_isr_f) {
+		ddi_remove_softintr(instance->soft_intr_id);
+	}
+
+	if (added_isr_f) {
+		drsas_rem_intrs(instance);
+	}
+
+	if (instance && instance->taskq) {
+		ddi_taskq_destroy(instance->taskq);
+	}
+
+	drsas_fm_ereport(instance, DDI_FM_DEVICE_NO_RESPONSE);
+	ddi_fm_service_impact(instance->dip, DDI_SERVICE_LOST);
+
+	drsas_fm_fini(instance);
+
+	pci_config_teardown(&instance->pci_handle);
+
+	ddi_soft_state_free(drsas_state, instance_no);
+
+	con_log(CL_ANN, (CE_NOTE,
+	    "dr_sas: return failure from drsas_attach"));
+
+	return (DDI_FAILURE);
+}
+
+/*ARGSUSED*/
+static int
+drsas_getinfo(dev_info_t *dip, ddi_info_cmd_t cmd,  void *arg, void **resultp)
+{
+	int	rval;
+	int	drsas_minor = getminor((dev_t)arg);
+
+	struct drsas_instance	*instance;
+
+	con_log(CL_ANN1, (CE_NOTE, "chkpnt:%s:%d", __func__, __LINE__));
+
+	switch (cmd) {
+		case DDI_INFO_DEVT2DEVINFO:
+			instance = (struct drsas_instance *)
+			    ddi_get_soft_state(drsas_state,
+			    MINOR2INST(drsas_minor));
+
+			if (instance == NULL) {
+				*resultp = NULL;
+				rval = DDI_FAILURE;
+			} else {
+				*resultp = instance->dip;
+				rval = DDI_SUCCESS;
+			}
+			break;
+		case DDI_INFO_DEVT2INSTANCE:
+			*resultp = (void *)instance;
+			rval = DDI_SUCCESS;
+			break;
+		default:
+			*resultp = NULL;
+			rval = DDI_FAILURE;
+	}
+
+	return (rval);
+}
+
+static int
+drsas_detach(dev_info_t *dip, ddi_detach_cmd_t cmd)
+{
+	int	instance_no;
+
+	struct drsas_instance	*instance;
+
+	con_log(CL_ANN1, (CE_NOTE, "chkpnt:%s:%d", __func__, __LINE__));
+
+	/* CONSTCOND */
+	ASSERT(NO_COMPETING_THREADS);
+
+	instance_no = ddi_get_instance(dip);
+
+	instance = (struct drsas_instance *)ddi_get_soft_state(drsas_state,
+	    instance_no);
+
+	if (!instance) {
+		con_log(CL_ANN, (CE_WARN,
+		    "dr_sas:%d could not get instance in detach",
+		    instance_no));
+
+		return (DDI_FAILURE);
+	}
+
+	con_log(CL_ANN, (CE_NOTE,
+	    "dr_sas%d: detaching device 0x%4x:0x%4x:0x%4x:0x%4x",
+	    instance_no, instance->vendor_id, instance->device_id,
+	    instance->subsysvid, instance->subsysid));
+
+	switch (cmd) {
+	case DDI_DETACH:
+		con_log(CL_ANN, (CE_NOTE,
+		    "drsas_detach: DDI_DETACH"));
+
+		if (scsi_hba_detach(dip) != DDI_SUCCESS) {
+			con_log(CL_ANN, (CE_WARN,
+			    "dr_sas:%d failed to detach",
+			    instance_no));
+
+			return (DDI_FAILURE);
+		}
+
+		scsi_hba_tran_free(instance->tran);
+
+		flush_cache(instance);
+
+		if (abort_aen_cmd(instance, instance->aen_cmd)) {
+			con_log(CL_ANN, (CE_WARN, "drsas_detach: "
+			    "failed to abort prevous AEN command"));
+
+			return (DDI_FAILURE);
+		}
+
+		instance->func_ptr->disable_intr(instance);
+
+		if (instance->isr_level == HIGH_LEVEL_INTR) {
+			ddi_remove_softintr(instance->soft_intr_id);
+		}
+
+		drsas_rem_intrs(instance);
+
+		if (instance->taskq) {
+			ddi_taskq_destroy(instance->taskq);
+		}
+		kmem_free(instance->dr_ld_list, MRDRV_MAX_LD
+		    * sizeof (struct drsas_ld));
+		free_space_for_mfi(instance);
+
+		drsas_fm_fini(instance);
+
+		pci_config_teardown(&instance->pci_handle);
+
+		kmem_free(instance->func_ptr,
+		    sizeof (struct drsas_func_ptr));
+
+		ddi_soft_state_free(drsas_state, instance_no);
+		break;
+	case DDI_PM_SUSPEND:
+		con_log(CL_ANN, (CE_NOTE,
+		    "drsas_detach: DDI_PM_SUSPEND"));
+
+		break;
+	case DDI_SUSPEND:
+		con_log(CL_ANN, (CE_NOTE,
+		    "drsas_detach: DDI_SUSPEND"));
+
+		break;
+	default:
+		con_log(CL_ANN, (CE_WARN,
+		    "invalid detach command:0x%x", cmd));
+		return (DDI_FAILURE);
+	}
+
+	return (DDI_SUCCESS);
+}
+
+/*
+ * ************************************************************************** *
+ *                                                                            *
+ *             common entry points - for character driver types               *
+ *                                                                            *
+ * ************************************************************************** *
+ */
+static  int
+drsas_open(dev_t *dev, int openflags, int otyp, cred_t *credp)
+{
+	int	rval = 0;
+
+	con_log(CL_ANN1, (CE_NOTE, "chkpnt:%s:%d", __func__, __LINE__));
+
+	/* Check root permissions */
+	if (drv_priv(credp) != 0) {
+		con_log(CL_ANN, (CE_WARN,
+		    "dr_sas: Non-root ioctl access denied!"));
+		return (EPERM);
+	}
+
+	/* Verify we are being opened as a character device */
+	if (otyp != OTYP_CHR) {
+		con_log(CL_ANN, (CE_WARN,
+		    "dr_sas: ioctl node must be a char node"));
+		return (EINVAL);
+	}
+
+	if (ddi_get_soft_state(drsas_state, MINOR2INST(getminor(*dev)))
+	    == NULL) {
+		return (ENXIO);
+	}
+
+	if (scsi_hba_open) {
+		rval = scsi_hba_open(dev, openflags, otyp, credp);
+	}
+
+	return (rval);
+}
+
+static  int
+drsas_close(dev_t dev, int openflags, int otyp, cred_t *credp)
+{
+	int	rval = 0;
+
+	con_log(CL_ANN1, (CE_NOTE, "chkpnt:%s:%d", __func__, __LINE__));
+
+	/* no need for locks! */
+
+	if (scsi_hba_close) {
+		rval = scsi_hba_close(dev, openflags, otyp, credp);
+	}
+
+	return (rval);
+}
+
+static int
+drsas_ioctl(dev_t dev, int cmd, intptr_t arg, int mode, cred_t *credp,
+    int *rvalp)
+{
+	int	rval = 0;
+
+	struct drsas_instance	*instance;
+	struct drsas_ioctl	*ioctl;
+	struct drsas_aen	aen;
+	int i;
+	con_log(CL_ANN1, (CE_NOTE, "chkpnt:%s:%d", __func__, __LINE__));
+
+	instance = ddi_get_soft_state(drsas_state, MINOR2INST(getminor(dev)));
+
+	if (instance == NULL) {
+		/* invalid minor number */
+		con_log(CL_ANN, (CE_WARN, "dr_sas: adapter not found."));
+		return (ENXIO);
+	}
+
+	ioctl = (struct drsas_ioctl *)kmem_zalloc(sizeof (struct drsas_ioctl),
+	    KM_SLEEP);
+	ASSERT(ioctl);
+
+	switch ((uint_t)cmd) {
+		case DRSAS_IOCTL_FIRMWARE:
+			for (i = 0; i < sizeof (struct drsas_ioctl); i++) {
+				if (ddi_copyin((uint8_t *)arg+i,
+				    (uint8_t *)ioctl+i, 1, mode)) {
+					con_log(CL_ANN, (CE_WARN, "drsas_ioctl "
+					    "ERROR IOCTL copyin"));
+					kmem_free(ioctl,
+					    sizeof (struct drsas_ioctl));
+					return (EFAULT);
+				}
+			}
+			if (ioctl->control_code == DRSAS_DRIVER_IOCTL_COMMON) {
+				rval = handle_drv_ioctl(instance, ioctl, mode);
+			} else {
+				rval = handle_mfi_ioctl(instance, ioctl, mode);
+			}
+			for (i = 0; i < sizeof (struct drsas_ioctl) - 1; i++) {
+				if (ddi_copyout((uint8_t *)ioctl+i,
+				    (uint8_t *)arg+i, 1, mode)) {
+					con_log(CL_ANN, (CE_WARN,
+					    "drsas_ioctl: ddi_copyout "
+					    "failed"));
+					rval = 1;
+					break;
+				}
+			}
+
+			break;
+		case DRSAS_IOCTL_AEN:
+			for (i = 0; i < sizeof (struct drsas_aen); i++) {
+				if (ddi_copyin((uint8_t *)arg+i,
+				    (uint8_t *)&aen+i, 1, mode)) {
+					con_log(CL_ANN, (CE_WARN,
+					    "drsas_ioctl: "
+					    "ERROR AEN copyin"));
+					kmem_free(ioctl,
+					    sizeof (struct drsas_ioctl));
+					return (EFAULT);
+				}
+			}
+
+			rval = handle_mfi_aen(instance, &aen);
+			for (i = 0; i < sizeof (struct drsas_aen); i++) {
+				if (ddi_copyout((uint8_t *)&aen + i,
+				    (uint8_t *)arg + i, 1, mode)) {
+					con_log(CL_ANN, (CE_WARN,
+					    "drsas_ioctl: "
+					    "ddi_copyout failed"));
+					rval = 1;
+					break;
+				}
+			}
+
+			break;
+		default:
+			rval = scsi_hba_ioctl(dev, cmd, arg,
+			    mode, credp, rvalp);
+
+			con_log(CL_DLEVEL1, (CE_NOTE, "drsas_ioctl: "
+			    "scsi_hba_ioctl called, ret = %x.", rval));
+	}
+
+	kmem_free(ioctl, sizeof (struct drsas_ioctl));
+	return (rval);
+}
+
+/*
+ * ************************************************************************** *
+ *                                                                            *
+ *               common entry points - for block driver types                 *
+ *                                                                            *
+ * ************************************************************************** *
+ */
+/*ARGSUSED*/
+static int
+drsas_reset(dev_info_t *dip, ddi_reset_cmd_t cmd)
+{
+	int	instance_no;
+
+	struct drsas_instance	*instance;
+
+	instance_no = ddi_get_instance(dip);
+	instance = (struct drsas_instance *)ddi_get_soft_state
+	    (drsas_state, instance_no);
+
+	con_log(CL_ANN1, (CE_NOTE, "chkpnt:%s:%d", __func__, __LINE__));
+
+	if (!instance) {
+		con_log(CL_ANN, (CE_WARN, "dr_sas:%d could not get adapter "
+		    "in reset", instance_no));
+		return (DDI_FAILURE);
+	}
+
+	instance->func_ptr->disable_intr(instance);
+
+	con_log(CL_ANN1, (CE_NOTE, "flushing cache for instance %d",
+	    instance_no));
+
+	flush_cache(instance);
+
+	return (DDI_SUCCESS);
+}
+
+
+/*
+ * ************************************************************************** *
+ *                                                                            *
+ *                          entry points (SCSI HBA)                           *
+ *                                                                            *
+ * ************************************************************************** *
+ */
+/*ARGSUSED*/
+static int
+drsas_tran_tgt_init(dev_info_t *hba_dip, dev_info_t *tgt_dip,
+		scsi_hba_tran_t *tran, struct scsi_device *sd)
+{
+	struct drsas_instance *instance;
+	uint16_t tgt = sd->sd_address.a_target;
+	uint8_t lun = sd->sd_address.a_lun;
+
+	con_log(CL_ANN1, (CE_NOTE, "drsas_tgt_init target %d lun %d",
+	    tgt, lun));
+
+	instance = ADDR2MR(&sd->sd_address);
+
+	if (ndi_dev_is_persistent_node(tgt_dip) == 0) {
+		(void) ndi_merge_node(tgt_dip, drsas_name_node);
+		ddi_set_name_addr(tgt_dip, NULL);
+
+		con_log(CL_ANN1, (CE_NOTE, "drsas_tgt_init in "
+		    "ndi_dev_is_persistent_node DDI_FAILURE t = %d l = %d",
+		    tgt, lun));
+		return (DDI_FAILURE);
+	}
+
+	con_log(CL_ANN1, (CE_NOTE, "drsas_tgt_init dev_dip %p tgt_dip %p",
+	    (void *)instance->dr_ld_list[tgt].dip, (void *)tgt_dip));
+
+	if (tgt < MRDRV_MAX_LD && lun == 0) {
+		if (instance->dr_ld_list[tgt].dip == NULL &&
+		    strcmp(ddi_driver_name(sd->sd_dev), "sd") == 0) {
+			instance->dr_ld_list[tgt].dip = tgt_dip;
+			instance->dr_ld_list[tgt].lun_type = DRSAS_LD_LUN;
+		}
+	}
+	return (DDI_SUCCESS);
+}
+
+/*ARGSUSED*/
+static void
+drsas_tran_tgt_free(dev_info_t *hba_dip, dev_info_t *tgt_dip,
+    scsi_hba_tran_t *hba_tran, struct scsi_device *sd)
+{
+	struct drsas_instance *instance;
+	int tgt = sd->sd_address.a_target;
+	int lun = sd->sd_address.a_lun;
+
+	instance = ADDR2MR(&sd->sd_address);
+
+	con_log(CL_ANN1, (CE_NOTE, "tgt_free t = %d l = %d", tgt, lun));
+
+	if (tgt < MRDRV_MAX_LD && lun == 0) {
+		if (instance->dr_ld_list[tgt].dip == tgt_dip) {
+			instance->dr_ld_list[tgt].dip = NULL;
+		}
+	}
+}
+
+static dev_info_t *
+drsas_find_child(struct drsas_instance *instance, uint16_t tgt, uint8_t lun)
+{
+	dev_info_t *child = NULL;
+	char addr[SCSI_MAXNAMELEN];
+	char tmp[MAXNAMELEN];
+
+	(void) sprintf(addr, "%x,%x", tgt, lun);
+	for (child = ddi_get_child(instance->dip); child;
+	    child = ddi_get_next_sibling(child)) {
+
+		if (drsas_name_node(child, tmp, MAXNAMELEN) !=
+		    DDI_SUCCESS) {
+			continue;
+		}
+
+		if (strcmp(addr, tmp) == 0) {
+			break;
+		}
+	}
+	con_log(CL_ANN1, (CE_NOTE, "drsas_find_child: return child = %p",
+	    (void *)child));
+	return (child);
+}
+
+static int
+drsas_name_node(dev_info_t *dip, char *name, int len)
+{
+	int tgt, lun;
+
+	tgt = ddi_prop_get_int(DDI_DEV_T_ANY, dip,
+	    DDI_PROP_DONTPASS, "target", -1);
+	con_log(CL_ANN1, (CE_NOTE,
+	    "drsas_name_node: dip %p tgt %d", (void *)dip, tgt));
+	if (tgt == -1) {
+		return (DDI_FAILURE);
+	}
+	lun = ddi_prop_get_int(DDI_DEV_T_ANY, dip, DDI_PROP_DONTPASS,
+	    "lun", -1);
+	con_log(CL_ANN1,
+	    (CE_NOTE, "drsas_name_node: tgt %d lun %d", tgt, lun));
+	if (lun == -1) {
+		return (DDI_FAILURE);
+	}
+	(void) snprintf(name, len, "%x,%x", tgt, lun);
+	return (DDI_SUCCESS);
+}
+
+static struct scsi_pkt *
+drsas_tran_init_pkt(struct scsi_address *ap, register struct scsi_pkt *pkt,
+	struct buf *bp, int cmdlen, int statuslen, int tgtlen,
+	int flags, int (*callback)(), caddr_t arg)
+{
+	struct scsa_cmd	*acmd;
+	struct drsas_instance	*instance;
+	struct scsi_pkt	*new_pkt;
+
+	con_log(CL_ANN1, (CE_NOTE, "chkpnt:%s:%d", __func__, __LINE__));
+
+	instance = ADDR2MR(ap);
+
+	/* step #1 : pkt allocation */
+	if (pkt == NULL) {
+		pkt = scsi_hba_pkt_alloc(instance->dip, ap, cmdlen, statuslen,
+		    tgtlen, sizeof (struct scsa_cmd), callback, arg);
+		if (pkt == NULL) {
+			return (NULL);
+		}
+
+		acmd = PKT2CMD(pkt);
+
+		/*
+		 * Initialize the new pkt - we redundantly initialize
+		 * all the fields for illustrative purposes.
+		 */
+		acmd->cmd_pkt		= pkt;
+		acmd->cmd_flags		= 0;
+		acmd->cmd_scblen	= statuslen;
+		acmd->cmd_cdblen	= cmdlen;
+		acmd->cmd_dmahandle	= NULL;
+		acmd->cmd_ncookies	= 0;
+		acmd->cmd_cookie	= 0;
+		acmd->cmd_cookiecnt	= 0;
+		acmd->cmd_nwin		= 0;
+
+		pkt->pkt_address	= *ap;
+		pkt->pkt_comp		= (void (*)())NULL;
+		pkt->pkt_flags		= 0;
+		pkt->pkt_time		= 0;
+		pkt->pkt_resid		= 0;
+		pkt->pkt_state		= 0;
+		pkt->pkt_statistics	= 0;
+		pkt->pkt_reason		= 0;
+		new_pkt			= pkt;
+	} else {
+		acmd = PKT2CMD(pkt);
+		new_pkt = NULL;
+	}
+
+	/* step #2 : dma allocation/move */
+	if (bp && bp->b_bcount != 0) {
+		if (acmd->cmd_dmahandle == NULL) {
+			if (drsas_dma_alloc(instance, pkt, bp, flags,
+			    callback) == DDI_FAILURE) {
+				if (new_pkt) {
+					scsi_hba_pkt_free(ap, new_pkt);
+				}
+				return ((struct scsi_pkt *)NULL);
+			}
+		} else {
+			if (drsas_dma_move(instance, pkt, bp) == DDI_FAILURE) {
+				return ((struct scsi_pkt *)NULL);
+			}
+		}
+	}
+
+	return (pkt);
+}
+
+static int
+drsas_tran_start(struct scsi_address *ap, register struct scsi_pkt *pkt)
+{
+	uchar_t 	cmd_done = 0;
+
+	struct drsas_instance	*instance = ADDR2MR(ap);
+	struct drsas_cmd	*cmd;
+
+	con_log(CL_ANN1, (CE_NOTE, "chkpnt:%s:%d:SCSI CDB[0]=0x%x",
+	    __func__, __LINE__, pkt->pkt_cdbp[0]));
+
+	pkt->pkt_reason	= CMD_CMPLT;
+	*pkt->pkt_scbp = STATUS_GOOD; /* clear arq scsi_status */
+
+	cmd = build_cmd(instance, ap, pkt, &cmd_done);
+
+	/*
+	 * Check if the command is already completed by the drsas_build_cmd()
+	 * routine. In which case the busy_flag would be clear and scb will be
+	 * NULL and appropriate reason provided in pkt_reason field
+	 */
+	if (cmd_done) {
+		pkt->pkt_reason = CMD_CMPLT;
+		pkt->pkt_scbp[0] = STATUS_GOOD;
+		pkt->pkt_state |= STATE_GOT_BUS | STATE_GOT_TARGET
+		    | STATE_SENT_CMD;
+		if (((pkt->pkt_flags & FLAG_NOINTR) == 0) && pkt->pkt_comp) {
+			(*pkt->pkt_comp)(pkt);
+		}
+
+		return (TRAN_ACCEPT);
+	}
+
+	if (cmd == NULL) {
+		return (TRAN_BUSY);
+	}
+
+	if ((pkt->pkt_flags & FLAG_NOINTR) == 0) {
+		if (instance->fw_outstanding > instance->max_fw_cmds) {
+			con_log(CL_ANN, (CE_CONT, "dr_sas:Firmware busy"));
+			return_mfi_pkt(instance, cmd);
+			return (TRAN_BUSY);
+		}
+
+		/* Synchronize the Cmd frame for the controller */
+		(void) ddi_dma_sync(cmd->frame_dma_obj.dma_handle, 0, 0,
+		    DDI_DMA_SYNC_FORDEV);
+
+		instance->func_ptr->issue_cmd(cmd, instance);
+
+	} else {
+		struct drsas_header *hdr = &cmd->frame->hdr;
+
+		cmd->sync_cmd = DRSAS_TRUE;
+
+		instance->func_ptr-> issue_cmd_in_poll_mode(instance, cmd);
+
+		pkt->pkt_reason		= CMD_CMPLT;
+		pkt->pkt_statistics	= 0;
+		pkt->pkt_state |= STATE_XFERRED_DATA | STATE_GOT_STATUS;
+
+		switch (ddi_get8(cmd->frame_dma_obj.acc_handle,
+		    &hdr->cmd_status)) {
+		case MFI_STAT_OK:
+			pkt->pkt_scbp[0] = STATUS_GOOD;
+			break;
+
+		case MFI_STAT_SCSI_DONE_WITH_ERROR:
+
+			pkt->pkt_reason	= CMD_CMPLT;
+			pkt->pkt_statistics = 0;
+
+			((struct scsi_status *)pkt->pkt_scbp)->sts_chk = 1;
+			break;
+
+		case MFI_STAT_DEVICE_NOT_FOUND:
+			pkt->pkt_reason		= CMD_DEV_GONE;
+			pkt->pkt_statistics	= STAT_DISCON;
+			break;
+
+		default:
+			((struct scsi_status *)pkt->pkt_scbp)->sts_busy = 1;
+		}
+
+		return_mfi_pkt(instance, cmd);
+		(void) drsas_common_check(instance, cmd);
+
+		if (pkt->pkt_comp) {
+			(*pkt->pkt_comp)(pkt);
+		}
+
+	}
+
+	return (TRAN_ACCEPT);
+}
+
+/*ARGSUSED*/
+static int
+drsas_tran_abort(struct scsi_address *ap, struct scsi_pkt *pkt)
+{
+	con_log(CL_ANN1, (CE_NOTE, "chkpnt:%s:%d", __func__, __LINE__));
+
+	/* abort command not supported by H/W */
+
+	return (DDI_FAILURE);
+}
+
+/*ARGSUSED*/
+static int
+drsas_tran_reset(struct scsi_address *ap, int level)
+{
+	con_log(CL_ANN1, (CE_NOTE, "chkpnt:%s:%d", __func__, __LINE__));
+
+	/* reset command not supported by H/W */
+
+	return (DDI_FAILURE);
+
+}
+
+/*ARGSUSED*/
+static int
+drsas_tran_getcap(struct scsi_address *ap, char *cap, int whom)
+{
+	int	rval = 0;
+
+	struct drsas_instance	*instance = ADDR2MR(ap);
+
+	con_log(CL_ANN1, (CE_NOTE, "chkpnt:%s:%d", __func__, __LINE__));
+
+	/* we do allow inquiring about capabilities for other targets */
+	if (cap == NULL) {
+		return (-1);
+	}
+
+	switch (scsi_hba_lookup_capstr(cap)) {
+	case SCSI_CAP_DMA_MAX:
+		/* Limit to 16MB max transfer */
+		rval = drsas_max_cap_maxxfer;
+		break;
+	case SCSI_CAP_MSG_OUT:
+		rval = 1;
+		break;
+	case SCSI_CAP_DISCONNECT:
+		rval = 0;
+		break;
+	case SCSI_CAP_SYNCHRONOUS:
+		rval = 0;
+		break;
+	case SCSI_CAP_WIDE_XFER:
+		rval = 1;
+		break;
+	case SCSI_CAP_TAGGED_QING:
+		rval = 1;
+		break;
+	case SCSI_CAP_UNTAGGED_QING:
+		rval = 1;
+		break;
+	case SCSI_CAP_PARITY:
+		rval = 1;
+		break;
+	case SCSI_CAP_INITIATOR_ID:
+		rval = instance->init_id;
+		break;
+	case SCSI_CAP_ARQ:
+		rval = 1;
+		break;
+	case SCSI_CAP_LINKED_CMDS:
+		rval = 0;
+		break;
+	case SCSI_CAP_RESET_NOTIFICATION:
+		rval = 1;
+		break;
+	case SCSI_CAP_GEOMETRY:
+		rval = -1;
+
+		break;
+	default:
+		con_log(CL_DLEVEL2, (CE_NOTE, "Default cap coming 0x%x",
+		    scsi_hba_lookup_capstr(cap)));
+		rval = -1;
+		break;
+	}
+
+	return (rval);
+}
+
+/*ARGSUSED*/
+static int
+drsas_tran_setcap(struct scsi_address *ap, char *cap, int value, int whom)
+{
+	int		rval = 1;
+
+	con_log(CL_ANN1, (CE_NOTE, "chkpnt:%s:%d", __func__, __LINE__));
+
+	/* We don't allow setting capabilities for other targets */
+	if (cap == NULL || whom == 0) {
+		return (-1);
+	}
+
+	switch (scsi_hba_lookup_capstr(cap)) {
+		case SCSI_CAP_DMA_MAX:
+		case SCSI_CAP_MSG_OUT:
+		case SCSI_CAP_PARITY:
+		case SCSI_CAP_LINKED_CMDS:
+		case SCSI_CAP_RESET_NOTIFICATION:
+		case SCSI_CAP_DISCONNECT:
+		case SCSI_CAP_SYNCHRONOUS:
+		case SCSI_CAP_UNTAGGED_QING:
+		case SCSI_CAP_WIDE_XFER:
+		case SCSI_CAP_INITIATOR_ID:
+		case SCSI_CAP_ARQ:
+			/*
+			 * None of these are settable via
+			 * the capability interface.
+			 */
+			break;
+		case SCSI_CAP_TAGGED_QING:
+			rval = 1;
+			break;
+		case SCSI_CAP_SECTOR_SIZE:
+			rval = 1;
+			break;
+
+		case SCSI_CAP_TOTAL_SECTORS:
+			rval = 1;
+			break;
+		default:
+			rval = -1;
+			break;
+	}
+
+	return (rval);
+}
+
+static void
+drsas_tran_destroy_pkt(struct scsi_address *ap, struct scsi_pkt *pkt)
+{
+	struct scsa_cmd *acmd = PKT2CMD(pkt);
+
+	con_log(CL_ANN1, (CE_NOTE, "chkpnt:%s:%d", __func__, __LINE__));
+
+	if (acmd->cmd_flags & CFLAG_DMAVALID) {
+		acmd->cmd_flags &= ~CFLAG_DMAVALID;
+
+		(void) ddi_dma_unbind_handle(acmd->cmd_dmahandle);
+
+		ddi_dma_free_handle(&acmd->cmd_dmahandle);
+
+		acmd->cmd_dmahandle = NULL;
+	}
+
+	/* free the pkt */
+	scsi_hba_pkt_free(ap, pkt);
+}
+
+/*ARGSUSED*/
+static void
+drsas_tran_dmafree(struct scsi_address *ap, struct scsi_pkt *pkt)
+{
+	register struct scsa_cmd *acmd = PKT2CMD(pkt);
+
+	con_log(CL_ANN1, (CE_NOTE, "chkpnt:%s:%d", __func__, __LINE__));
+
+	if (acmd->cmd_flags & CFLAG_DMAVALID) {
+		acmd->cmd_flags &= ~CFLAG_DMAVALID;
+
+		(void) ddi_dma_unbind_handle(acmd->cmd_dmahandle);
+
+		ddi_dma_free_handle(&acmd->cmd_dmahandle);
+
+		acmd->cmd_dmahandle = NULL;
+	}
+}
+
+/*ARGSUSED*/
+static void
+drsas_tran_sync_pkt(struct scsi_address *ap, struct scsi_pkt *pkt)
+{
+	register struct scsa_cmd	*acmd = PKT2CMD(pkt);
+
+	con_log(CL_ANN1, (CE_NOTE, "chkpnt:%s:%d", __func__, __LINE__));
+
+	if (acmd->cmd_flags & CFLAG_DMAVALID) {
+		(void) ddi_dma_sync(acmd->cmd_dmahandle, acmd->cmd_dma_offset,
+		    acmd->cmd_dma_len, (acmd->cmd_flags & CFLAG_DMASEND) ?
+		    DDI_DMA_SYNC_FORDEV : DDI_DMA_SYNC_FORCPU);
+	}
+}
+
+/*
+ * drsas_isr(caddr_t)
+ *
+ * The Interrupt Service Routine
+ *
+ * Collect status for all completed commands and do callback
+ *
+ */
+static uint_t
+drsas_isr(struct drsas_instance *instance)
+{
+	int		need_softintr;
+	uint32_t	producer;
+	uint32_t	consumer;
+	uint32_t	context;
+
+	struct drsas_cmd	*cmd;
+
+	con_log(CL_ANN1, (CE_NOTE, "chkpnt:%s:%d", __func__, __LINE__));
+
+	ASSERT(instance);
+	if ((instance->intr_type == DDI_INTR_TYPE_FIXED) &&
+	    !instance->func_ptr->intr_ack(instance)) {
+		return (DDI_INTR_UNCLAIMED);
+	}
+
+	(void) ddi_dma_sync(instance->mfi_internal_dma_obj.dma_handle,
+	    0, 0, DDI_DMA_SYNC_FORCPU);
+
+	if (drsas_check_dma_handle(instance->mfi_internal_dma_obj.dma_handle)
+	    != DDI_SUCCESS) {
+		drsas_fm_ereport(instance, DDI_FM_DEVICE_NO_RESPONSE);
+		ddi_fm_service_impact(instance->dip, DDI_SERVICE_LOST);
+		return (DDI_INTR_UNCLAIMED);
+	}
+
+	producer = ddi_get32(instance->mfi_internal_dma_obj.acc_handle,
+	    instance->producer);
+	consumer = ddi_get32(instance->mfi_internal_dma_obj.acc_handle,
+	    instance->consumer);
+
+	con_log(CL_ANN1, (CE_CONT, " producer %x consumer %x ",
+	    producer, consumer));
+	if (producer == consumer) {
+		con_log(CL_ANN1, (CE_WARN, "producer = consumer case"));
+		return (DDI_INTR_UNCLAIMED);
+	}
+	mutex_enter(&instance->completed_pool_mtx);
+
+	while (consumer != producer) {
+		context = ddi_get32(instance->mfi_internal_dma_obj.acc_handle,
+		    &instance->reply_queue[consumer]);
+		cmd = instance->cmd_list[context];
+		mlist_add_tail(&cmd->list, &instance->completed_pool_list);
+
+		consumer++;
+		if (consumer == (instance->max_fw_cmds + 1)) {
+			consumer = 0;
+		}
+	}
+
+	mutex_exit(&instance->completed_pool_mtx);
+
+	ddi_put32(instance->mfi_internal_dma_obj.acc_handle,
+	    instance->consumer, consumer);
+	(void) ddi_dma_sync(instance->mfi_internal_dma_obj.dma_handle,
+	    0, 0, DDI_DMA_SYNC_FORDEV);
+
+	if (instance->softint_running) {
+		need_softintr = 0;
+	} else {
+		need_softintr = 1;
+	}
+
+	if (instance->isr_level == HIGH_LEVEL_INTR) {
+		if (need_softintr) {
+			ddi_trigger_softintr(instance->soft_intr_id);
+		}
+	} else {
+		/*
+		 * Not a high-level interrupt, therefore call the soft level
+		 * interrupt explicitly
+		 */
+		(void) drsas_softintr(instance);
+	}
+
+	return (DDI_INTR_CLAIMED);
+}
+
+
+/*
+ * ************************************************************************** *
+ *                                                                            *
+ *                                  libraries                                 *
+ *                                                                            *
+ * ************************************************************************** *
+ */
+/*
+ * get_mfi_pkt : Get a command from the free pool
+ * After successful allocation, the caller of this routine
+ * must clear the frame buffer (memset to zero) before
+ * using the packet further.
+ *
+ * ***** Note *****
+ * After clearing the frame buffer the context id of the
+ * frame buffer SHOULD be restored back.
+ */
+static struct drsas_cmd *
+get_mfi_pkt(struct drsas_instance *instance)
+{
+	mlist_t 		*head = &instance->cmd_pool_list;
+	struct drsas_cmd	*cmd = NULL;
+
+	mutex_enter(&instance->cmd_pool_mtx);
+	ASSERT(mutex_owned(&instance->cmd_pool_mtx));
+
+	if (!mlist_empty(head)) {
+		cmd = mlist_entry(head->next, struct drsas_cmd, list);
+		mlist_del_init(head->next);
+	}
+	if (cmd != NULL)
+		cmd->pkt = NULL;
+	mutex_exit(&instance->cmd_pool_mtx);
+
+	return (cmd);
+}
+
+/*
+ * return_mfi_pkt : Return a cmd to free command pool
+ */
+static void
+return_mfi_pkt(struct drsas_instance *instance, struct drsas_cmd *cmd)
+{
+	mutex_enter(&instance->cmd_pool_mtx);
+	ASSERT(mutex_owned(&instance->cmd_pool_mtx));
+
+	mlist_add(&cmd->list, &instance->cmd_pool_list);
+
+	mutex_exit(&instance->cmd_pool_mtx);
+}
+
+/*
+ * destroy_mfi_frame_pool
+ */
+static void
+destroy_mfi_frame_pool(struct drsas_instance *instance)
+{
+	int		i;
+	uint32_t	max_cmd = instance->max_fw_cmds;
+
+	struct drsas_cmd	*cmd;
+
+	/* return all frames to pool */
+	for (i = 0; i < max_cmd+1; i++) {
+
+		cmd = instance->cmd_list[i];
+
+		if (cmd->frame_dma_obj_status == DMA_OBJ_ALLOCATED)
+			(void) drsas_free_dma_obj(instance, cmd->frame_dma_obj);
+
+		cmd->frame_dma_obj_status  = DMA_OBJ_FREED;
+	}
+
+}
+
+/*
+ * create_mfi_frame_pool
+ */
+static int
+create_mfi_frame_pool(struct drsas_instance *instance)
+{
+	int		i = 0;
+	int		cookie_cnt;
+	uint16_t	max_cmd;
+	uint16_t	sge_sz;
+	uint32_t	sgl_sz;
+	uint32_t	tot_frame_size;
+
+	struct drsas_cmd	*cmd;
+
+	max_cmd = instance->max_fw_cmds;
+
+	sge_sz	= sizeof (struct drsas_sge64);
+
+	/* calculated the number of 64byte frames required for SGL */
+	sgl_sz		= sge_sz * instance->max_num_sge;
+	tot_frame_size	= sgl_sz + MRMFI_FRAME_SIZE + SENSE_LENGTH;
+
+	con_log(CL_DLEVEL3, (CE_NOTE, "create_mfi_frame_pool: "
+	    "sgl_sz %x tot_frame_size %x", sgl_sz, tot_frame_size));
+
+	while (i < max_cmd+1) {
+		cmd = instance->cmd_list[i];
+
+		cmd->frame_dma_obj.size	= tot_frame_size;
+		cmd->frame_dma_obj.dma_attr = drsas_generic_dma_attr;
+		cmd->frame_dma_obj.dma_attr.dma_attr_addr_hi = 0xFFFFFFFFU;
+		cmd->frame_dma_obj.dma_attr.dma_attr_count_max = 0xFFFFFFFFU;
+		cmd->frame_dma_obj.dma_attr.dma_attr_sgllen = 1;
+		cmd->frame_dma_obj.dma_attr.dma_attr_align = 64;
+
+
+		cookie_cnt = drsas_alloc_dma_obj(instance, &cmd->frame_dma_obj,
+		    (uchar_t)DDI_STRUCTURE_LE_ACC);
+
+		if (cookie_cnt == -1 || cookie_cnt > 1) {
+			con_log(CL_ANN, (CE_WARN,
+			    "create_mfi_frame_pool: could not alloc."));
+			return (DDI_FAILURE);
+		}
+
+		bzero(cmd->frame_dma_obj.buffer, tot_frame_size);
+
+		cmd->frame_dma_obj_status = DMA_OBJ_ALLOCATED;
+		cmd->frame = (union drsas_frame *)cmd->frame_dma_obj.buffer;
+		cmd->frame_phys_addr =
+		    cmd->frame_dma_obj.dma_cookie[0].dmac_address;
+
+		cmd->sense = (uint8_t *)(((unsigned long)
+		    cmd->frame_dma_obj.buffer) +
+		    tot_frame_size - SENSE_LENGTH);
+		cmd->sense_phys_addr =
+		    cmd->frame_dma_obj.dma_cookie[0].dmac_address +
+		    tot_frame_size - SENSE_LENGTH;
+
+		if (!cmd->frame || !cmd->sense) {
+			con_log(CL_ANN, (CE_NOTE,
+			    "dr_sas: pci_pool_alloc failed"));
+
+			return (ENOMEM);
+		}
+
+		ddi_put32(cmd->frame_dma_obj.acc_handle,
+		    &cmd->frame->io.context, cmd->index);
+		i++;
+
+		con_log(CL_DLEVEL3, (CE_NOTE, "[%x]-%x",
+		    cmd->index, cmd->frame_phys_addr));
+	}
+
+	return (DDI_SUCCESS);
+}
+
+/*
+ * free_additional_dma_buffer
+ */
+static void
+free_additional_dma_buffer(struct drsas_instance *instance)
+{
+	if (instance->mfi_internal_dma_obj.status == DMA_OBJ_ALLOCATED) {
+		(void) drsas_free_dma_obj(instance,
+		    instance->mfi_internal_dma_obj);
+		instance->mfi_internal_dma_obj.status = DMA_OBJ_FREED;
+	}
+
+	if (instance->mfi_evt_detail_obj.status == DMA_OBJ_ALLOCATED) {
+		(void) drsas_free_dma_obj(instance,
+		    instance->mfi_evt_detail_obj);
+		instance->mfi_evt_detail_obj.status = DMA_OBJ_FREED;
+	}
+}
+
+/*
+ * alloc_additional_dma_buffer
+ */
+static int
+alloc_additional_dma_buffer(struct drsas_instance *instance)
+{
+	uint32_t	reply_q_sz;
+	uint32_t	internal_buf_size = PAGESIZE*2;
+
+	/* max cmds plus 1 + producer & consumer */
+	reply_q_sz = sizeof (uint32_t) * (instance->max_fw_cmds + 1 + 2);
+
+	instance->mfi_internal_dma_obj.size = internal_buf_size;
+	instance->mfi_internal_dma_obj.dma_attr	= drsas_generic_dma_attr;
+	instance->mfi_internal_dma_obj.dma_attr.dma_attr_addr_hi = 0xFFFFFFFFU;
+	instance->mfi_internal_dma_obj.dma_attr.dma_attr_count_max =
+	    0xFFFFFFFFU;
+	instance->mfi_internal_dma_obj.dma_attr.dma_attr_sgllen	= 1;
+
+	if (drsas_alloc_dma_obj(instance, &instance->mfi_internal_dma_obj,
+	    (uchar_t)DDI_STRUCTURE_LE_ACC) != 1) {
+		con_log(CL_ANN, (CE_WARN,
+		    "dr_sas: could not alloc reply queue"));
+		return (DDI_FAILURE);
+	}
+
+	bzero(instance->mfi_internal_dma_obj.buffer, internal_buf_size);
+
+	instance->mfi_internal_dma_obj.status |= DMA_OBJ_ALLOCATED;
+
+	instance->producer = (uint32_t *)((unsigned long)
+	    instance->mfi_internal_dma_obj.buffer);
+	instance->consumer = (uint32_t *)((unsigned long)
+	    instance->mfi_internal_dma_obj.buffer + 4);
+	instance->reply_queue = (uint32_t *)((unsigned long)
+	    instance->mfi_internal_dma_obj.buffer + 8);
+	instance->internal_buf = (caddr_t)(((unsigned long)
+	    instance->mfi_internal_dma_obj.buffer) + reply_q_sz + 8);
+	instance->internal_buf_dmac_add =
+	    instance->mfi_internal_dma_obj.dma_cookie[0].dmac_address +
+	    (reply_q_sz + 8);
+	instance->internal_buf_size = internal_buf_size -
+	    (reply_q_sz + 8);
+
+	/* allocate evt_detail */
+	instance->mfi_evt_detail_obj.size = sizeof (struct drsas_evt_detail);
+	instance->mfi_evt_detail_obj.dma_attr = drsas_generic_dma_attr;
+	instance->mfi_evt_detail_obj.dma_attr.dma_attr_addr_hi = 0xFFFFFFFFU;
+	instance->mfi_evt_detail_obj.dma_attr.dma_attr_count_max = 0xFFFFFFFFU;
+	instance->mfi_evt_detail_obj.dma_attr.dma_attr_sgllen = 1;
+	instance->mfi_evt_detail_obj.dma_attr.dma_attr_align = 1;
+
+	if (drsas_alloc_dma_obj(instance, &instance->mfi_evt_detail_obj,
+	    (uchar_t)DDI_STRUCTURE_LE_ACC) != 1) {
+		con_log(CL_ANN, (CE_WARN, "alloc_additional_dma_buffer: "
+		    "could not allocate data transfer buffer."));
+		return (DDI_FAILURE);
+	}
+
+	bzero(instance->mfi_evt_detail_obj.buffer,
+	    sizeof (struct drsas_evt_detail));
+
+	instance->mfi_evt_detail_obj.status |= DMA_OBJ_ALLOCATED;
+
+	return (DDI_SUCCESS);
+}
+
+/*
+ * free_space_for_mfi
+ */
+static void
+free_space_for_mfi(struct drsas_instance *instance)
+{
+	int		i;
+	uint32_t	max_cmd = instance->max_fw_cmds;
+
+	/* already freed */
+	if (instance->cmd_list == NULL) {
+		return;
+	}
+
+	free_additional_dma_buffer(instance);
+
+	/* first free the MFI frame pool */
+	destroy_mfi_frame_pool(instance);
+
+	/* free all the commands in the cmd_list */
+	for (i = 0; i < instance->max_fw_cmds+1; i++) {
+		kmem_free(instance->cmd_list[i],
+		    sizeof (struct drsas_cmd));
+
+		instance->cmd_list[i] = NULL;
+	}
+
+	/* free the cmd_list buffer itself */
+	kmem_free(instance->cmd_list,
+	    sizeof (struct drsas_cmd *) * (max_cmd+1));
+
+	instance->cmd_list = NULL;
+
+	INIT_LIST_HEAD(&instance->cmd_pool_list);
+}
+
+/*
+ * alloc_space_for_mfi
+ */
+static int
+alloc_space_for_mfi(struct drsas_instance *instance)
+{
+	int		i;
+	uint32_t	max_cmd;
+	size_t		sz;
+
+	struct drsas_cmd	*cmd;
+
+	max_cmd = instance->max_fw_cmds;
+
+	/* reserve 1 more slot for flush_cache */
+	sz = sizeof (struct drsas_cmd *) * (max_cmd+1);
+
+	/*
+	 * instance->cmd_list is an array of struct drsas_cmd pointers.
+	 * Allocate the dynamic array first and then allocate individual
+	 * commands.
+	 */
+	instance->cmd_list = kmem_zalloc(sz, KM_SLEEP);
+	ASSERT(instance->cmd_list);
+
+	for (i = 0; i < max_cmd+1; i++) {
+		instance->cmd_list[i] = kmem_zalloc(sizeof (struct drsas_cmd),
+		    KM_SLEEP);
+		ASSERT(instance->cmd_list[i]);
+	}
+
+	INIT_LIST_HEAD(&instance->cmd_pool_list);
+
+	/* add all the commands to command pool (instance->cmd_pool) */
+	for (i = 0; i < max_cmd; i++) {
+		cmd		= instance->cmd_list[i];
+		cmd->index	= i;
+
+		mlist_add_tail(&cmd->list, &instance->cmd_pool_list);
+	}
+
+	/* single slot for flush_cache won't be added in command pool */
+	cmd		= instance->cmd_list[max_cmd];
+	cmd->index	= i;
+
+	/* create a frame pool and assign one frame to each cmd */
+	if (create_mfi_frame_pool(instance)) {
+		con_log(CL_ANN, (CE_NOTE, "error creating frame DMA pool"));
+		return (DDI_FAILURE);
+	}
+
+	/* create a frame pool and assign one frame to each cmd */
+	if (alloc_additional_dma_buffer(instance)) {
+		con_log(CL_ANN, (CE_NOTE, "error creating frame DMA pool"));
+		return (DDI_FAILURE);
+	}
+
+	return (DDI_SUCCESS);
+}
+
+/*
+ * get_ctrl_info
+ */
+static int
+get_ctrl_info(struct drsas_instance *instance,
+    struct drsas_ctrl_info *ctrl_info)
+{
+	int	ret = 0;
+
+	struct drsas_cmd		*cmd;
+	struct drsas_dcmd_frame	*dcmd;
+	struct drsas_ctrl_info	*ci;
+
+	cmd = get_mfi_pkt(instance);
+
+	if (!cmd) {
+		con_log(CL_ANN, (CE_WARN,
+		    "Failed to get a cmd for ctrl info"));
+		return (DDI_FAILURE);
+	}
+	/* Clear the frame buffer and assign back the context id */
+	(void) memset((char *)&cmd->frame[0], 0, sizeof (union drsas_frame));
+	ddi_put32(cmd->frame_dma_obj.acc_handle, &cmd->frame->hdr.context,
+	    cmd->index);
+
+	dcmd = &cmd->frame->dcmd;
+
+	ci = (struct drsas_ctrl_info *)instance->internal_buf;
+
+	if (!ci) {
+		con_log(CL_ANN, (CE_WARN,
+		    "Failed to alloc mem for ctrl info"));
+		return_mfi_pkt(instance, cmd);
+		return (DDI_FAILURE);
+	}
+
+	(void) memset(ci, 0, sizeof (struct drsas_ctrl_info));
+
+	/* for( i = 0; i < DCMD_MBOX_SZ; i++ ) dcmd->mbox.b[i] = 0; */
+	(void) memset(dcmd->mbox.b, 0, DCMD_MBOX_SZ);
+
+	ddi_put8(cmd->frame_dma_obj.acc_handle, &dcmd->cmd, MFI_CMD_OP_DCMD);
+	ddi_put8(cmd->frame_dma_obj.acc_handle, &dcmd->cmd_status,
+	    MFI_CMD_STATUS_POLL_MODE);
+	ddi_put8(cmd->frame_dma_obj.acc_handle, &dcmd->sge_count, 1);
+	ddi_put16(cmd->frame_dma_obj.acc_handle, &dcmd->flags,
+	    MFI_FRAME_DIR_READ);
+	ddi_put16(cmd->frame_dma_obj.acc_handle, &dcmd->timeout, 0);
+	ddi_put32(cmd->frame_dma_obj.acc_handle, &dcmd->data_xfer_len,
+	    sizeof (struct drsas_ctrl_info));
+	ddi_put32(cmd->frame_dma_obj.acc_handle, &dcmd->opcode,
+	    DR_DCMD_CTRL_GET_INFO);
+	ddi_put32(cmd->frame_dma_obj.acc_handle, &dcmd->sgl.sge32[0].phys_addr,
+	    instance->internal_buf_dmac_add);
+	ddi_put32(cmd->frame_dma_obj.acc_handle, &dcmd->sgl.sge32[0].length,
+	    sizeof (struct drsas_ctrl_info));
+
+	cmd->frame_count = 1;
+
+	if (!instance->func_ptr->issue_cmd_in_poll_mode(instance, cmd)) {
+		ret = 0;
+		ddi_rep_get8(cmd->frame_dma_obj.acc_handle,
+		    (uint8_t *)ctrl_info, (uint8_t *)ci,
+		    sizeof (struct drsas_ctrl_info), DDI_DEV_AUTOINCR);
+	} else {
+		con_log(CL_ANN, (CE_WARN, "get_ctrl_info: Ctrl info failed"));
+		ret = -1;
+	}
+
+	return_mfi_pkt(instance, cmd);
+	if (drsas_common_check(instance, cmd) != DDI_SUCCESS) {
+		ret = -1;
+	}
+
+	return (ret);
+}
+
+/*
+ * abort_aen_cmd
+ */
+static int
+abort_aen_cmd(struct drsas_instance *instance,
+    struct drsas_cmd *cmd_to_abort)
+{
+	int	ret = 0;
+
+	struct drsas_cmd		*cmd;
+	struct drsas_abort_frame	*abort_fr;
+
+	cmd = get_mfi_pkt(instance);
+
+	if (!cmd) {
+		con_log(CL_ANN, (CE_WARN,
+		    "Failed to get a cmd for ctrl info"));
+		return (DDI_FAILURE);
+	}
+	/* Clear the frame buffer and assign back the context id */
+	(void) memset((char *)&cmd->frame[0], 0, sizeof (union drsas_frame));
+	ddi_put32(cmd->frame_dma_obj.acc_handle, &cmd->frame->hdr.context,
+	    cmd->index);
+
+	abort_fr = &cmd->frame->abort;
+
+	/* prepare and issue the abort frame */
+	ddi_put8(cmd->frame_dma_obj.acc_handle,
+	    &abort_fr->cmd, MFI_CMD_OP_ABORT);
+	ddi_put8(cmd->frame_dma_obj.acc_handle, &abort_fr->cmd_status,
+	    MFI_CMD_STATUS_SYNC_MODE);
+	ddi_put16(cmd->frame_dma_obj.acc_handle, &abort_fr->flags, 0);
+	ddi_put32(cmd->frame_dma_obj.acc_handle, &abort_fr->abort_context,
+	    cmd_to_abort->index);
+	ddi_put32(cmd->frame_dma_obj.acc_handle,
+	    &abort_fr->abort_mfi_phys_addr_lo, cmd_to_abort->frame_phys_addr);
+	ddi_put32(cmd->frame_dma_obj.acc_handle,
+	    &abort_fr->abort_mfi_phys_addr_hi, 0);
+
+	instance->aen_cmd->abort_aen = 1;
+
+	cmd->sync_cmd = DRSAS_TRUE;
+	cmd->frame_count = 1;
+
+	if (instance->func_ptr->issue_cmd_in_sync_mode(instance, cmd)) {
+		con_log(CL_ANN, (CE_WARN,
+		    "abort_aen_cmd: issue_cmd_in_sync_mode failed"));
+		ret = -1;
+	} else {
+		ret = 0;
+	}
+
+	instance->aen_cmd->abort_aen = 1;
+	instance->aen_cmd = 0;
+
+	return_mfi_pkt(instance, cmd);
+	(void) drsas_common_check(instance, cmd);
+
+	return (ret);
+}
+
+/*
+ * init_mfi
+ */
+static int
+init_mfi(struct drsas_instance *instance)
+{
+	struct drsas_cmd		*cmd;
+	struct drsas_ctrl_info		ctrl_info;
+	struct drsas_init_frame		*init_frame;
+	struct drsas_init_queue_info	*initq_info;
+
+	/* we expect the FW state to be READY */
+	if (mfi_state_transition_to_ready(instance)) {
+		con_log(CL_ANN, (CE_WARN, "dr_sas: F/W is not ready"));
+		goto fail_ready_state;
+	}
+
+	/* get various operational parameters from status register */
+	instance->max_num_sge =
+	    (instance->func_ptr->read_fw_status_reg(instance) &
+	    0xFF0000) >> 0x10;
+	/*
+	 * Reduce the max supported cmds by 1. This is to ensure that the
+	 * reply_q_sz (1 more than the max cmd that driver may send)
+	 * does not exceed max cmds that the FW can support
+	 */
+	instance->max_fw_cmds =
+	    instance->func_ptr->read_fw_status_reg(instance) & 0xFFFF;
+	instance->max_fw_cmds = instance->max_fw_cmds - 1;
+
+	instance->max_num_sge =
+	    (instance->max_num_sge > DRSAS_MAX_SGE_CNT) ?
+	    DRSAS_MAX_SGE_CNT : instance->max_num_sge;
+
+	/* create a pool of commands */
+	if (alloc_space_for_mfi(instance) != DDI_SUCCESS)
+		goto fail_alloc_fw_space;
+
+	/*
+	 * Prepare a init frame. Note the init frame points to queue info
+	 * structure. Each frame has SGL allocated after first 64 bytes. For
+	 * this frame - since we don't need any SGL - we use SGL's space as
+	 * queue info structure
+	 */
+	cmd = get_mfi_pkt(instance);
+	/* Clear the frame buffer and assign back the context id */
+	(void) memset((char *)&cmd->frame[0], 0, sizeof (union drsas_frame));
+	ddi_put32(cmd->frame_dma_obj.acc_handle, &cmd->frame->hdr.context,
+	    cmd->index);
+
+	init_frame = (struct drsas_init_frame *)cmd->frame;
+	initq_info = (struct drsas_init_queue_info *)
+	    ((unsigned long)init_frame + 64);
+
+	(void) memset(init_frame, 0, MRMFI_FRAME_SIZE);
+	(void) memset(initq_info, 0, sizeof (struct drsas_init_queue_info));
+
+	ddi_put32(cmd->frame_dma_obj.acc_handle, &initq_info->init_flags, 0);
+
+	ddi_put32(cmd->frame_dma_obj.acc_handle,
+	    &initq_info->reply_queue_entries, instance->max_fw_cmds + 1);
+
+	ddi_put32(cmd->frame_dma_obj.acc_handle,
+	    &initq_info->producer_index_phys_addr_hi, 0);
+	ddi_put32(cmd->frame_dma_obj.acc_handle,
+	    &initq_info->producer_index_phys_addr_lo,
+	    instance->mfi_internal_dma_obj.dma_cookie[0].dmac_address);
+
+	ddi_put32(cmd->frame_dma_obj.acc_handle,
+	    &initq_info->consumer_index_phys_addr_hi, 0);
+	ddi_put32(cmd->frame_dma_obj.acc_handle,
+	    &initq_info->consumer_index_phys_addr_lo,
+	    instance->mfi_internal_dma_obj.dma_cookie[0].dmac_address + 4);
+
+	ddi_put32(cmd->frame_dma_obj.acc_handle,
+	    &initq_info->reply_queue_start_phys_addr_hi, 0);
+	ddi_put32(cmd->frame_dma_obj.acc_handle,
+	    &initq_info->reply_queue_start_phys_addr_lo,
+	    instance->mfi_internal_dma_obj.dma_cookie[0].dmac_address + 8);
+
+	ddi_put8(cmd->frame_dma_obj.acc_handle,
+	    &init_frame->cmd, MFI_CMD_OP_INIT);
+	ddi_put8(cmd->frame_dma_obj.acc_handle, &init_frame->cmd_status,
+	    MFI_CMD_STATUS_POLL_MODE);
+	ddi_put16(cmd->frame_dma_obj.acc_handle, &init_frame->flags, 0);
+	ddi_put32(cmd->frame_dma_obj.acc_handle,
+	    &init_frame->queue_info_new_phys_addr_lo,
+	    cmd->frame_phys_addr + 64);
+	ddi_put32(cmd->frame_dma_obj.acc_handle,
+	    &init_frame->queue_info_new_phys_addr_hi, 0);
+
+	ddi_put32(cmd->frame_dma_obj.acc_handle, &init_frame->data_xfer_len,
+	    sizeof (struct drsas_init_queue_info));
+
+	cmd->frame_count = 1;
+
+	/* issue the init frame in polled mode */
+	if (instance->func_ptr->issue_cmd_in_poll_mode(instance, cmd)) {
+		con_log(CL_ANN, (CE_WARN, "failed to init firmware"));
+		goto fail_fw_init;
+	}
+
+	return_mfi_pkt(instance, cmd);
+	if (drsas_common_check(instance, cmd) != DDI_SUCCESS) {
+		goto fail_fw_init;
+	}
+
+	/* gather misc FW related information */
+	if (!get_ctrl_info(instance, &ctrl_info)) {
+		instance->max_sectors_per_req = ctrl_info.max_request_size;
+		con_log(CL_ANN1, (CE_NOTE, "product name %s ld present %d",
+		    ctrl_info.product_name, ctrl_info.ld_present_count));
+	} else {
+		instance->max_sectors_per_req = instance->max_num_sge *
+		    PAGESIZE / 512;
+	}
+
+	if (drsas_check_acc_handle(instance->regmap_handle) != DDI_SUCCESS) {
+		goto fail_fw_init;
+	}
+
+	return (DDI_SUCCESS);
+
+fail_fw_init:
+fail_alloc_fw_space:
+
+	free_space_for_mfi(instance);
+
+fail_ready_state:
+	ddi_regs_map_free(&instance->regmap_handle);
+
+fail_mfi_reg_setup:
+	return (DDI_FAILURE);
+}
+
+/*
+ * mfi_state_transition_to_ready	: Move the FW to READY state
+ *
+ * @reg_set			: MFI register set
+ */
+static int
+mfi_state_transition_to_ready(struct drsas_instance *instance)
+{
+	int		i;
+	uint8_t		max_wait;
+	uint32_t	fw_ctrl;
+	uint32_t	fw_state;
+	uint32_t	cur_state;
+
+	fw_state =
+	    instance->func_ptr->read_fw_status_reg(instance) & MFI_STATE_MASK;
+	con_log(CL_ANN1, (CE_NOTE,
+	    "mfi_state_transition_to_ready:FW state = 0x%x", fw_state));
+
+	while (fw_state != MFI_STATE_READY) {
+		con_log(CL_ANN, (CE_NOTE,
+		    "mfi_state_transition_to_ready:FW state%x", fw_state));
+
+		switch (fw_state) {
+		case MFI_STATE_FAULT:
+			con_log(CL_ANN, (CE_NOTE,
+			    "dr_sas: FW in FAULT state!!"));
+
+			return (ENODEV);
+		case MFI_STATE_WAIT_HANDSHAKE:
+			/* set the CLR bit in IMR0 */
+			con_log(CL_ANN, (CE_NOTE,
+			    "dr_sas: FW waiting for HANDSHAKE"));
+			/*
+			 * PCI_Hot Plug: MFI F/W requires
+			 * (MFI_INIT_CLEAR_HANDSHAKE|MFI_INIT_HOTPLUG)
+			 * to be set
+			 */
+			/* WR_IB_MSG_0(MFI_INIT_CLEAR_HANDSHAKE, instance); */
+			WR_IB_DOORBELL(MFI_INIT_CLEAR_HANDSHAKE |
+			    MFI_INIT_HOTPLUG, instance);
+
+			max_wait	= 2;
+			cur_state	= MFI_STATE_WAIT_HANDSHAKE;
+			break;
+		case MFI_STATE_BOOT_MESSAGE_PENDING:
+			/* set the CLR bit in IMR0 */
+			con_log(CL_ANN, (CE_NOTE,
+			    "dr_sas: FW state boot message pending"));
+			/*
+			 * PCI_Hot Plug: MFI F/W requires
+			 * (MFI_INIT_CLEAR_HANDSHAKE|MFI_INIT_HOTPLUG)
+			 * to be set
+			 */
+			WR_IB_DOORBELL(MFI_INIT_HOTPLUG, instance);
+
+			max_wait	= 10;
+			cur_state	= MFI_STATE_BOOT_MESSAGE_PENDING;
+			break;
+		case MFI_STATE_OPERATIONAL:
+			/* bring it to READY state; assuming max wait 2 secs */
+			instance->func_ptr->disable_intr(instance);
+			con_log(CL_ANN1, (CE_NOTE,
+			    "dr_sas: FW in OPERATIONAL state"));
+			/*
+			 * PCI_Hot Plug: MFI F/W requires
+			 * (MFI_INIT_READY | MFI_INIT_MFIMODE | MFI_INIT_ABORT)
+			 * to be set
+			 */
+			/* WR_IB_DOORBELL(MFI_INIT_READY, instance); */
+			WR_IB_DOORBELL(MFI_RESET_FLAGS, instance);
+
+			max_wait	= 10;
+			cur_state	= MFI_STATE_OPERATIONAL;
+			break;
+		case MFI_STATE_UNDEFINED:
+			/* this state should not last for more than 2 seconds */
+			con_log(CL_ANN, (CE_NOTE, "FW state undefined"));
+
+			max_wait	= 2;
+			cur_state	= MFI_STATE_UNDEFINED;
+			break;
+		case MFI_STATE_BB_INIT:
+			max_wait	= 2;
+			cur_state	= MFI_STATE_BB_INIT;
+			break;
+		case MFI_STATE_FW_INIT:
+			max_wait	= 2;
+			cur_state	= MFI_STATE_FW_INIT;
+			break;
+		case MFI_STATE_DEVICE_SCAN:
+			max_wait	= 10;
+			cur_state	= MFI_STATE_DEVICE_SCAN;
+			break;
+		default:
+			con_log(CL_ANN, (CE_NOTE,
+			    "dr_sas: Unknown state 0x%x", fw_state));
+			return (ENODEV);
+		}
+
+		/* the cur_state should not last for more than max_wait secs */
+		for (i = 0; i < (max_wait * MILLISEC); i++) {
+			/* fw_state = RD_OB_MSG_0(instance) & MFI_STATE_MASK; */
+			fw_state =
+			    instance->func_ptr->read_fw_status_reg(instance) &
+			    MFI_STATE_MASK;
+
+			if (fw_state == cur_state) {
+				delay(1 * drv_usectohz(MILLISEC));
+			} else {
+				break;
+			}
+		}
+
+		/* return error if fw_state hasn't changed after max_wait */
+		if (fw_state == cur_state) {
+			con_log(CL_ANN, (CE_NOTE,
+			    "FW state hasn't changed in %d secs", max_wait));
+			return (ENODEV);
+		}
+	};
+
+	fw_ctrl = RD_IB_DOORBELL(instance);
+
+	con_log(CL_ANN1, (CE_NOTE,
+	    "mfi_state_transition_to_ready:FW ctrl = 0x%x", fw_ctrl));
+
+	/*
+	 * Write 0xF to the doorbell register to do the following.
+	 * - Abort all outstanding commands (bit 0).
+	 * - Transition from OPERATIONAL to READY state (bit 1).
+	 * - Discard (possible) low MFA posted in 64-bit mode (bit-2).
+	 * - Set to release FW to continue running (i.e. BIOS handshake
+	 *   (bit 3).
+	 */
+	WR_IB_DOORBELL(0xF, instance);
+
+	if (drsas_check_acc_handle(instance->regmap_handle) != DDI_SUCCESS) {
+		return (ENODEV);
+	}
+	return (DDI_SUCCESS);
+}
+
+/*
+ * get_seq_num
+ */
+static int
+get_seq_num(struct drsas_instance *instance,
+    struct drsas_evt_log_info *eli)
+{
+	int	ret = DDI_SUCCESS;
+
+	dma_obj_t			dcmd_dma_obj;
+	struct drsas_cmd		*cmd;
+	struct drsas_dcmd_frame		*dcmd;
+
+	cmd = get_mfi_pkt(instance);
+
+	if (!cmd) {
+		cmn_err(CE_WARN, "dr_sas: failed to get a cmd");
+		return (ENOMEM);
+	}
+	/* Clear the frame buffer and assign back the context id */
+	(void) memset((char *)&cmd->frame[0], 0, sizeof (union drsas_frame));
+	ddi_put32(cmd->frame_dma_obj.acc_handle, &cmd->frame->hdr.context,
+	    cmd->index);
+
+	dcmd	= &cmd->frame->dcmd;
+
+	/* allocate the data transfer buffer */
+	dcmd_dma_obj.size = sizeof (struct drsas_evt_log_info);
+	dcmd_dma_obj.dma_attr = drsas_generic_dma_attr;
+	dcmd_dma_obj.dma_attr.dma_attr_addr_hi = 0xFFFFFFFFU;
+	dcmd_dma_obj.dma_attr.dma_attr_count_max = 0xFFFFFFFFU;
+	dcmd_dma_obj.dma_attr.dma_attr_sgllen = 1;
+	dcmd_dma_obj.dma_attr.dma_attr_align = 1;
+
+	if (drsas_alloc_dma_obj(instance, &dcmd_dma_obj,
+	    (uchar_t)DDI_STRUCTURE_LE_ACC) != 1) {
+		con_log(CL_ANN, (CE_WARN,
+		    "get_seq_num: could not allocate data transfer buffer."));
+		return (DDI_FAILURE);
+	}
+
+	(void) memset(dcmd_dma_obj.buffer, 0,
+	    sizeof (struct drsas_evt_log_info));
+
+	(void) memset(dcmd->mbox.b, 0, DCMD_MBOX_SZ);
+
+	ddi_put8(cmd->frame_dma_obj.acc_handle, &dcmd->cmd, MFI_CMD_OP_DCMD);
+	ddi_put8(cmd->frame_dma_obj.acc_handle, &dcmd->cmd_status, 0);
+	ddi_put8(cmd->frame_dma_obj.acc_handle, &dcmd->sge_count, 1);
+	ddi_put16(cmd->frame_dma_obj.acc_handle, &dcmd->flags,
+	    MFI_FRAME_DIR_READ);
+	ddi_put16(cmd->frame_dma_obj.acc_handle, &dcmd->timeout, 0);
+	ddi_put32(cmd->frame_dma_obj.acc_handle, &dcmd->data_xfer_len,
+	    sizeof (struct drsas_evt_log_info));
+	ddi_put32(cmd->frame_dma_obj.acc_handle, &dcmd->opcode,
+	    DR_DCMD_CTRL_EVENT_GET_INFO);
+	ddi_put32(cmd->frame_dma_obj.acc_handle, &dcmd->sgl.sge32[0].length,
+	    sizeof (struct drsas_evt_log_info));
+	ddi_put32(cmd->frame_dma_obj.acc_handle, &dcmd->sgl.sge32[0].phys_addr,
+	    dcmd_dma_obj.dma_cookie[0].dmac_address);
+
+	cmd->sync_cmd = DRSAS_TRUE;
+	cmd->frame_count = 1;
+
+	if (instance->func_ptr->issue_cmd_in_sync_mode(instance, cmd)) {
+		cmn_err(CE_WARN, "get_seq_num: "
+		    "failed to issue DRSAS_DCMD_CTRL_EVENT_GET_INFO");
+		ret = DDI_FAILURE;
+	} else {
+		/* copy the data back into callers buffer */
+		ddi_rep_get8(cmd->frame_dma_obj.acc_handle, (uint8_t *)eli,
+		    (uint8_t *)dcmd_dma_obj.buffer,
+		    sizeof (struct drsas_evt_log_info), DDI_DEV_AUTOINCR);
+		ret = DDI_SUCCESS;
+	}
+
+	if (drsas_free_dma_obj(instance, dcmd_dma_obj) != DDI_SUCCESS)
+		ret = DDI_FAILURE;
+
+	return_mfi_pkt(instance, cmd);
+	if (drsas_common_check(instance, cmd) != DDI_SUCCESS) {
+		ret = DDI_FAILURE;
+	}
+	return (ret);
+}
+
+/*
+ * start_mfi_aen
+ */
+static int
+start_mfi_aen(struct drsas_instance *instance)
+{
+	int	ret = 0;
+
+	struct drsas_evt_log_info	eli;
+	union drsas_evt_class_locale	class_locale;
+
+	/* get the latest sequence number from FW */
+	(void) memset(&eli, 0, sizeof (struct drsas_evt_log_info));
+
+	if (get_seq_num(instance, &eli)) {
+		cmn_err(CE_WARN, "start_mfi_aen: failed to get seq num");
+		return (-1);
+	}
+
+	/* register AEN with FW for latest sequence number plus 1 */
+	class_locale.members.reserved	= 0;
+	class_locale.members.locale	= DR_EVT_LOCALE_ALL;
+	class_locale.members.class	= DR_EVT_CLASS_INFO;
+	ret = register_mfi_aen(instance, eli.newest_seq_num + 1,
+	    class_locale.word);
+
+	if (ret) {
+		cmn_err(CE_WARN, "start_mfi_aen: aen registration failed");
+		return (-1);
+	}
+
+	return (ret);
+}
+
+/*
+ * flush_cache
+ */
+static void
+flush_cache(struct drsas_instance *instance)
+{
+	struct drsas_cmd		*cmd = NULL;
+	struct drsas_dcmd_frame		*dcmd;
+	uint32_t	max_cmd = instance->max_fw_cmds;
+
+	cmd = instance->cmd_list[max_cmd];
+
+	if (cmd == NULL)
+		return;
+
+	dcmd = &cmd->frame->dcmd;
+
+	(void) memset(dcmd->mbox.b, 0, DCMD_MBOX_SZ);
+
+	ddi_put8(cmd->frame_dma_obj.acc_handle, &dcmd->cmd, MFI_CMD_OP_DCMD);
+	ddi_put8(cmd->frame_dma_obj.acc_handle, &dcmd->cmd_status, 0x0);
+	ddi_put8(cmd->frame_dma_obj.acc_handle, &dcmd->sge_count, 0);
+	ddi_put16(cmd->frame_dma_obj.acc_handle, &dcmd->flags,
+	    MFI_FRAME_DIR_NONE);
+	ddi_put16(cmd->frame_dma_obj.acc_handle, &dcmd->timeout, 0);
+	ddi_put32(cmd->frame_dma_obj.acc_handle, &dcmd->data_xfer_len, 0);
+	ddi_put32(cmd->frame_dma_obj.acc_handle, &dcmd->opcode,
+	    DR_DCMD_CTRL_CACHE_FLUSH);
+	ddi_put8(cmd->frame_dma_obj.acc_handle, &dcmd->mbox.b[0],
+	    DR_FLUSH_CTRL_CACHE | DR_FLUSH_DISK_CACHE);
+
+	cmd->frame_count = 1;
+
+	if (instance->func_ptr->issue_cmd_in_poll_mode(instance, cmd)) {
+		con_log(CL_ANN1, (CE_WARN,
+	    "flush_cache: failed to issue MFI_DCMD_CTRL_CACHE_FLUSH"));
+	}
+	con_log(CL_DLEVEL1, (CE_NOTE, "done"));
+}
+
+/*
+ * service_mfi_aen-	Completes an AEN command
+ * @instance:			Adapter soft state
+ * @cmd:			Command to be completed
+ *
+ */
+static void
+service_mfi_aen(struct drsas_instance *instance, struct drsas_cmd *cmd)
+{
+	uint32_t	seq_num;
+	struct drsas_evt_detail *evt_detail =
+	    (struct drsas_evt_detail *)instance->mfi_evt_detail_obj.buffer;
+	int		rval = 0;
+	int		tgt = 0;
+	ddi_acc_handle_t		acc_handle;
+
+	acc_handle = cmd->frame_dma_obj.acc_handle;
+
+	cmd->cmd_status = ddi_get8(acc_handle, &cmd->frame->io.cmd_status);
+
+	if (cmd->cmd_status == ENODATA) {
+		cmd->cmd_status = 0;
+	}
+
+	/*
+	 * log the MFI AEN event to the sysevent queue so that
+	 * application will get noticed
+	 */
+	if (ddi_log_sysevent(instance->dip, DDI_VENDOR_LSI, "LSIMEGA", "SAS",
+	    NULL, NULL, DDI_NOSLEEP) != DDI_SUCCESS) {
+		int	instance_no = ddi_get_instance(instance->dip);
+		con_log(CL_ANN, (CE_WARN,
+		    "dr_sas%d: Failed to log AEN event", instance_no));
+	}
+	/*
+	 * Check for any ld devices that has changed state. i.e. online
+	 * or offline.
+	 */
+	con_log(CL_ANN1, (CE_NOTE,
+	    "AEN: code = %x class = %x locale = %x args = %x",
+	    ddi_get32(acc_handle, &evt_detail->code),
+	    evt_detail->cl.members.class,
+	    ddi_get16(acc_handle, &evt_detail->cl.members.locale),
+	    ddi_get8(acc_handle, &evt_detail->arg_type)));
+
+	switch (ddi_get32(acc_handle, &evt_detail->code)) {
+	case DR_EVT_CFG_CLEARED: {
+		for (tgt = 0; tgt < MRDRV_MAX_LD; tgt++) {
+			if (instance->dr_ld_list[tgt].dip != NULL) {
+				rval = drsas_service_evt(instance, tgt, 0,
+				    DRSAS_EVT_UNCONFIG_TGT, NULL);
+				con_log(CL_ANN1, (CE_WARN,
+				    "dr_sas: CFG CLEARED AEN rval = %d "
+				    "tgt id = %d", rval, tgt));
+			}
+		}
+		break;
+	}
+
+	case DR_EVT_LD_DELETED: {
+		rval = drsas_service_evt(instance,
+		    ddi_get16(acc_handle, &evt_detail->args.ld.target_id), 0,
+		    DRSAS_EVT_UNCONFIG_TGT, NULL);
+		con_log(CL_ANN1, (CE_WARN, "dr_sas: LD DELETED AEN rval = %d "
+		    "tgt id = %d index = %d", rval,
+		    ddi_get16(acc_handle, &evt_detail->args.ld.target_id),
+		    ddi_get8(acc_handle, &evt_detail->args.ld.ld_index)));
+		break;
+	} /* End of DR_EVT_LD_DELETED */
+
+	case DR_EVT_LD_CREATED: {
+		rval = drsas_service_evt(instance,
+		    ddi_get16(acc_handle, &evt_detail->args.ld.target_id), 0,
+		    DRSAS_EVT_CONFIG_TGT, NULL);
+		con_log(CL_ANN1, (CE_WARN, "dr_sas: LD CREATED AEN rval = %d "
+		    "tgt id = %d index = %d", rval,
+		    ddi_get16(acc_handle, &evt_detail->args.ld.target_id),
+		    ddi_get8(acc_handle, &evt_detail->args.ld.ld_index)));
+		break;
+	} /* End of DR_EVT_LD_CREATED */
+	} /* End of Main Switch */
+
+	/* get copy of seq_num and class/locale for re-registration */
+	seq_num = ddi_get32(acc_handle, &evt_detail->seq_num);
+	seq_num++;
+	(void) memset(instance->mfi_evt_detail_obj.buffer, 0,
+	    sizeof (struct drsas_evt_detail));
+
+	ddi_put8(acc_handle, &cmd->frame->dcmd.cmd_status, 0x0);
+	ddi_put32(acc_handle, &cmd->frame->dcmd.mbox.w[0], seq_num);
+
+	instance->aen_seq_num = seq_num;
+
+	cmd->frame_count = 1;
+
+	/* Issue the aen registration frame */
+	instance->func_ptr->issue_cmd(cmd, instance);
+}
+
+/*
+ * complete_cmd_in_sync_mode -	Completes an internal command
+ * @instance:			Adapter soft state
+ * @cmd:			Command to be completed
+ *
+ * The issue_cmd_in_sync_mode() function waits for a command to complete
+ * after it issues a command. This function wakes up that waiting routine by
+ * calling wake_up() on the wait queue.
+ */
+static void
+complete_cmd_in_sync_mode(struct drsas_instance *instance,
+    struct drsas_cmd *cmd)
+{
+	cmd->cmd_status = ddi_get8(cmd->frame_dma_obj.acc_handle,
+	    &cmd->frame->io.cmd_status);
+
+	cmd->sync_cmd = DRSAS_FALSE;
+
+	if (cmd->cmd_status == ENODATA) {
+		cmd->cmd_status = 0;
+	}
+
+	cv_broadcast(&instance->int_cmd_cv);
+}
+
+/*
+ * drsas_softintr - The Software ISR
+ * @param arg	: HBA soft state
+ *
+ * called from high-level interrupt if hi-level interrupt are not there,
+ * otherwise triggered as a soft interrupt
+ */
+static uint_t
+drsas_softintr(struct drsas_instance *instance)
+{
+	struct scsi_pkt		*pkt;
+	struct scsa_cmd		*acmd;
+	struct drsas_cmd	*cmd;
+	struct mlist_head	*pos, *next;
+	mlist_t			process_list;
+	struct drsas_header	*hdr;
+	struct scsi_arq_status	*arqstat;
+
+	con_log(CL_ANN1, (CE_CONT, "drsas_softintr called"));
+
+	ASSERT(instance);
+	mutex_enter(&instance->completed_pool_mtx);
+
+	if (mlist_empty(&instance->completed_pool_list)) {
+		mutex_exit(&instance->completed_pool_mtx);
+		return (DDI_INTR_UNCLAIMED);
+	}
+
+	instance->softint_running = 1;
+
+	INIT_LIST_HEAD(&process_list);
+	mlist_splice(&instance->completed_pool_list, &process_list);
+	INIT_LIST_HEAD(&instance->completed_pool_list);
+
+	mutex_exit(&instance->completed_pool_mtx);
+
+	/* perform all callbacks first, before releasing the SCBs */
+	mlist_for_each_safe(pos, next, &process_list) {
+		cmd = mlist_entry(pos, struct drsas_cmd, list);
+
+		/* syncronize the Cmd frame for the controller */
+		(void) ddi_dma_sync(cmd->frame_dma_obj.dma_handle,
+		    0, 0, DDI_DMA_SYNC_FORCPU);
+
+		if (drsas_check_dma_handle(cmd->frame_dma_obj.dma_handle) !=
+		    DDI_SUCCESS) {
+			drsas_fm_ereport(instance, DDI_FM_DEVICE_NO_RESPONSE);
+			ddi_fm_service_impact(instance->dip, DDI_SERVICE_LOST);
+			return (DDI_INTR_UNCLAIMED);
+		}
+
+		hdr = &cmd->frame->hdr;
+
+		/* remove the internal command from the process list */
+		mlist_del_init(&cmd->list);
+
+		switch (ddi_get8(cmd->frame_dma_obj.acc_handle, &hdr->cmd)) {
+		case MFI_CMD_OP_PD_SCSI:
+		case MFI_CMD_OP_LD_SCSI:
+		case MFI_CMD_OP_LD_READ:
+		case MFI_CMD_OP_LD_WRITE:
+			/*
+			 * MFI_CMD_OP_PD_SCSI and MFI_CMD_OP_LD_SCSI
+			 * could have been issued either through an
+			 * IO path or an IOCTL path. If it was via IOCTL,
+			 * we will send it to internal completion.
+			 */
+			if (cmd->sync_cmd == DRSAS_TRUE) {
+				complete_cmd_in_sync_mode(instance, cmd);
+				break;
+			}
+
+			/* regular commands */
+			acmd =	cmd->cmd;
+			pkt =	CMD2PKT(acmd);
+
+			if (acmd->cmd_flags & CFLAG_DMAVALID) {
+				if (acmd->cmd_flags & CFLAG_CONSISTENT) {
+					(void) ddi_dma_sync(acmd->cmd_dmahandle,
+					    acmd->cmd_dma_offset,
+					    acmd->cmd_dma_len,
+					    DDI_DMA_SYNC_FORCPU);
+				}
+			}
+
+			pkt->pkt_reason		= CMD_CMPLT;
+			pkt->pkt_statistics	= 0;
+			pkt->pkt_state = STATE_GOT_BUS
+			    | STATE_GOT_TARGET | STATE_SENT_CMD
+			    | STATE_XFERRED_DATA | STATE_GOT_STATUS;
+
+			con_log(CL_ANN1, (CE_CONT,
+			    "CDB[0] = %x completed for %s: size %lx context %x",
+			    pkt->pkt_cdbp[0], ((acmd->islogical) ? "LD" : "PD"),
+			    acmd->cmd_dmacount, hdr->context));
+
+			if (pkt->pkt_cdbp[0] == SCMD_INQUIRY) {
+				struct scsi_inquiry	*inq;
+
+				if (acmd->cmd_dmacount != 0) {
+					bp_mapin(acmd->cmd_buf);
+					inq = (struct scsi_inquiry *)
+					    acmd->cmd_buf->b_un.b_addr;
+
+					/* don't expose physical drives to OS */
+					if (acmd->islogical &&
+					    (hdr->cmd_status == MFI_STAT_OK)) {
+						display_scsi_inquiry(
+						    (caddr_t)inq);
+					} else if ((hdr->cmd_status ==
+					    MFI_STAT_OK) && inq->inq_dtype ==
+					    DTYPE_DIRECT) {
+
+						display_scsi_inquiry(
+						    (caddr_t)inq);
+
+						/* for physical disk */
+						hdr->cmd_status =
+						    MFI_STAT_DEVICE_NOT_FOUND;
+					}
+				}
+			}
+
+			switch (hdr->cmd_status) {
+			case MFI_STAT_OK:
+				pkt->pkt_scbp[0] = STATUS_GOOD;
+				break;
+			case MFI_STAT_LD_CC_IN_PROGRESS:
+			case MFI_STAT_LD_RECON_IN_PROGRESS:
+				pkt->pkt_scbp[0] = STATUS_GOOD;
+				break;
+			case MFI_STAT_LD_INIT_IN_PROGRESS:
+				con_log(CL_ANN,
+				    (CE_WARN, "Initialization in Progress"));
+				pkt->pkt_reason	= CMD_TRAN_ERR;
+
+				break;
+			case MFI_STAT_SCSI_DONE_WITH_ERROR:
+				con_log(CL_ANN1, (CE_CONT, "scsi_done error"));
+
+				pkt->pkt_reason	= CMD_CMPLT;
+				((struct scsi_status *)
+				    pkt->pkt_scbp)->sts_chk = 1;
+
+				if (pkt->pkt_cdbp[0] == SCMD_TEST_UNIT_READY) {
+
+					con_log(CL_ANN,
+					    (CE_WARN, "TEST_UNIT_READY fail"));
+
+				} else {
+					pkt->pkt_state |= STATE_ARQ_DONE;
+					arqstat = (void *)(pkt->pkt_scbp);
+					arqstat->sts_rqpkt_reason = CMD_CMPLT;
+					arqstat->sts_rqpkt_resid = 0;
+					arqstat->sts_rqpkt_state |=
+					    STATE_GOT_BUS | STATE_GOT_TARGET
+					    | STATE_SENT_CMD
+					    | STATE_XFERRED_DATA;
+					*(uint8_t *)&arqstat->sts_rqpkt_status =
+					    STATUS_GOOD;
+					ddi_rep_get8(
+					    cmd->frame_dma_obj.acc_handle,
+					    (uint8_t *)
+					    &(arqstat->sts_sensedata),
+					    cmd->sense,
+					    acmd->cmd_scblen -
+					    offsetof(struct scsi_arq_status,
+					    sts_sensedata), DDI_DEV_AUTOINCR);
+				}
+				break;
+			case MFI_STAT_LD_OFFLINE:
+			case MFI_STAT_DEVICE_NOT_FOUND:
+				con_log(CL_ANN1, (CE_CONT,
+				    "device not found error"));
+				pkt->pkt_reason	= CMD_DEV_GONE;
+				pkt->pkt_statistics  = STAT_DISCON;
+				break;
+			case MFI_STAT_LD_LBA_OUT_OF_RANGE:
+				pkt->pkt_state |= STATE_ARQ_DONE;
+				pkt->pkt_reason	= CMD_CMPLT;
+				((struct scsi_status *)
+				    pkt->pkt_scbp)->sts_chk = 1;
+
+				arqstat = (void *)(pkt->pkt_scbp);
+				arqstat->sts_rqpkt_reason = CMD_CMPLT;
+				arqstat->sts_rqpkt_resid = 0;
+				arqstat->sts_rqpkt_state |= STATE_GOT_BUS
+				    | STATE_GOT_TARGET | STATE_SENT_CMD
+				    | STATE_XFERRED_DATA;
+				*(uint8_t *)&arqstat->sts_rqpkt_status =
+				    STATUS_GOOD;
+
+				arqstat->sts_sensedata.es_valid = 1;
+				arqstat->sts_sensedata.es_key =
+				    KEY_ILLEGAL_REQUEST;
+				arqstat->sts_sensedata.es_class =
+				    CLASS_EXTENDED_SENSE;
+
+				/*
+				 * LOGICAL BLOCK ADDRESS OUT OF RANGE:
+				 * ASC: 0x21h; ASCQ: 0x00h;
+				 */
+				arqstat->sts_sensedata.es_add_code = 0x21;
+				arqstat->sts_sensedata.es_qual_code = 0x00;
+
+				break;
+
+			default:
+				con_log(CL_ANN, (CE_CONT, "Unknown status!"));
+				pkt->pkt_reason	= CMD_TRAN_ERR;
+
+				break;
+			}
+
+			atomic_add_16(&instance->fw_outstanding, (-1));
+
+			return_mfi_pkt(instance, cmd);
+
+			(void) drsas_common_check(instance, cmd);
+
+			if (acmd->cmd_dmahandle) {
+				if (drsas_check_dma_handle(
+				    acmd->cmd_dmahandle) != DDI_SUCCESS) {
+					ddi_fm_service_impact(instance->dip,
+					    DDI_SERVICE_UNAFFECTED);
+					pkt->pkt_reason = CMD_TRAN_ERR;
+					pkt->pkt_statistics = 0;
+				}
+			}
+
+			/* Call the callback routine */
+			if (((pkt->pkt_flags & FLAG_NOINTR) == 0) &&
+			    pkt->pkt_comp) {
+				(*pkt->pkt_comp)(pkt);
+			}
+
+			break;
+		case MFI_CMD_OP_SMP:
+		case MFI_CMD_OP_STP:
+			complete_cmd_in_sync_mode(instance, cmd);
+			break;
+		case MFI_CMD_OP_DCMD:
+			/* see if got an event notification */
+			if (ddi_get32(cmd->frame_dma_obj.acc_handle,
+			    &cmd->frame->dcmd.opcode) ==
+			    DR_DCMD_CTRL_EVENT_WAIT) {
+				if ((instance->aen_cmd == cmd) &&
+				    (instance->aen_cmd->abort_aen)) {
+					con_log(CL_ANN, (CE_WARN,
+					    "drsas_softintr: "
+					    "aborted_aen returned"));
+				} else {
+					atomic_add_16(&instance->fw_outstanding,
+					    (-1));
+					service_mfi_aen(instance, cmd);
+				}
+			} else {
+				complete_cmd_in_sync_mode(instance, cmd);
+			}
+
+			break;
+		case MFI_CMD_OP_ABORT:
+			con_log(CL_ANN, (CE_WARN, "MFI_CMD_OP_ABORT complete"));
+			/*
+			 * MFI_CMD_OP_ABORT successfully completed
+			 * in the synchronous mode
+			 */
+			complete_cmd_in_sync_mode(instance, cmd);
+			break;
+		default:
+			drsas_fm_ereport(instance, DDI_FM_DEVICE_NO_RESPONSE);
+			ddi_fm_service_impact(instance->dip, DDI_SERVICE_LOST);
+
+			if (cmd->pkt != NULL) {
+				pkt = cmd->pkt;
+				if (((pkt->pkt_flags & FLAG_NOINTR) == 0) &&
+				    pkt->pkt_comp) {
+					(*pkt->pkt_comp)(pkt);
+				}
+			}
+			con_log(CL_ANN, (CE_WARN, "Cmd type unknown !"));
+			break;
+		}
+	}
+
+	instance->softint_running = 0;
+
+	return (DDI_INTR_CLAIMED);
+}
+
+/*
+ * drsas_alloc_dma_obj
+ *
+ * Allocate the memory and other resources for an dma object.
+ */
+static int
+drsas_alloc_dma_obj(struct drsas_instance *instance, dma_obj_t *obj,
+    uchar_t endian_flags)
+{
+	int	i;
+	size_t	alen = 0;
+	uint_t	cookie_cnt;
+	struct ddi_device_acc_attr tmp_endian_attr;
+
+	tmp_endian_attr = endian_attr;
+	tmp_endian_attr.devacc_attr_endian_flags = endian_flags;
+
+	i = ddi_dma_alloc_handle(instance->dip, &obj->dma_attr,
+	    DDI_DMA_SLEEP, NULL, &obj->dma_handle);
+	if (i != DDI_SUCCESS) {
+
+		switch (i) {
+			case DDI_DMA_BADATTR :
+				con_log(CL_ANN, (CE_WARN,
+				"Failed ddi_dma_alloc_handle- Bad attribute"));
+				break;
+			case DDI_DMA_NORESOURCES :
+				con_log(CL_ANN, (CE_WARN,
+				"Failed ddi_dma_alloc_handle- No Resources"));
+				break;
+			default :
+				con_log(CL_ANN, (CE_WARN,
+				"Failed ddi_dma_alloc_handle: "
+				"unknown status %d", i));
+				break;
+		}
+
+		return (-1);
+	}
+
+	if ((ddi_dma_mem_alloc(obj->dma_handle, obj->size, &tmp_endian_attr,
+	    DDI_DMA_RDWR | DDI_DMA_STREAMING, DDI_DMA_SLEEP, NULL,
+	    &obj->buffer, &alen, &obj->acc_handle) != DDI_SUCCESS) ||
+	    alen < obj->size) {
+
+		ddi_dma_free_handle(&obj->dma_handle);
+
+		con_log(CL_ANN, (CE_WARN, "Failed : ddi_dma_mem_alloc"));
+
+		return (-1);
+	}
+
+	if (ddi_dma_addr_bind_handle(obj->dma_handle, NULL, obj->buffer,
+	    obj->size, DDI_DMA_RDWR | DDI_DMA_STREAMING, DDI_DMA_SLEEP,
+	    NULL, &obj->dma_cookie[0], &cookie_cnt) != DDI_SUCCESS) {
+
+		ddi_dma_mem_free(&obj->acc_handle);
+		ddi_dma_free_handle(&obj->dma_handle);
+
+		con_log(CL_ANN, (CE_WARN, "Failed : ddi_dma_addr_bind_handle"));
+
+		return (-1);
+	}
+
+	if (drsas_check_dma_handle(obj->dma_handle) != DDI_SUCCESS) {
+		ddi_fm_service_impact(instance->dip, DDI_SERVICE_LOST);
+		return (-1);
+	}
+
+	if (drsas_check_acc_handle(obj->acc_handle) != DDI_SUCCESS) {
+		ddi_fm_service_impact(instance->dip, DDI_SERVICE_LOST);
+		return (-1);
+	}
+
+	return (cookie_cnt);
+}
+
+/*
+ * drsas_free_dma_obj(struct drsas_instance *, dma_obj_t)
+ *
+ * De-allocate the memory and other resources for an dma object, which must
+ * have been alloated by a previous call to drsas_alloc_dma_obj()
+ */
+static int
+drsas_free_dma_obj(struct drsas_instance *instance, dma_obj_t obj)
+{
+
+	if (drsas_check_dma_handle(obj.dma_handle) != DDI_SUCCESS) {
+		ddi_fm_service_impact(instance->dip, DDI_SERVICE_UNAFFECTED);
+		return (DDI_FAILURE);
+	}
+
+	if (drsas_check_acc_handle(obj.acc_handle) != DDI_SUCCESS) {
+		ddi_fm_service_impact(instance->dip, DDI_SERVICE_UNAFFECTED);
+		return (DDI_FAILURE);
+	}
+
+	(void) ddi_dma_unbind_handle(obj.dma_handle);
+	ddi_dma_mem_free(&obj.acc_handle);
+	ddi_dma_free_handle(&obj.dma_handle);
+
+	return (DDI_SUCCESS);
+}
+
+/*
+ * drsas_dma_alloc(instance_t *, struct scsi_pkt *, struct buf *,
+ * int, int (*)())
+ *
+ * Allocate dma resources for a new scsi command
+ */
+static int
+drsas_dma_alloc(struct drsas_instance *instance, struct scsi_pkt *pkt,
+    struct buf *bp, int flags, int (*callback)())
+{
+	int	dma_flags;
+	int	(*cb)(caddr_t);
+	int	i;
+
+	ddi_dma_attr_t	tmp_dma_attr = drsas_generic_dma_attr;
+	struct scsa_cmd	*acmd = PKT2CMD(pkt);
+
+	acmd->cmd_buf = bp;
+
+	if (bp->b_flags & B_READ) {
+		acmd->cmd_flags &= ~CFLAG_DMASEND;
+		dma_flags = DDI_DMA_READ;
+	} else {
+		acmd->cmd_flags |= CFLAG_DMASEND;
+		dma_flags = DDI_DMA_WRITE;
+	}
+
+	if (flags & PKT_CONSISTENT) {
+		acmd->cmd_flags |= CFLAG_CONSISTENT;
+		dma_flags |= DDI_DMA_CONSISTENT;
+	}
+
+	if (flags & PKT_DMA_PARTIAL) {
+		dma_flags |= DDI_DMA_PARTIAL;
+	}
+
+	dma_flags |= DDI_DMA_REDZONE;
+
+	cb = (callback == NULL_FUNC) ? DDI_DMA_DONTWAIT : DDI_DMA_SLEEP;
+
+	tmp_dma_attr.dma_attr_sgllen = instance->max_num_sge;
+	tmp_dma_attr.dma_attr_addr_hi = 0xffffffffffffffffull;
+
+	if ((i = ddi_dma_alloc_handle(instance->dip, &tmp_dma_attr,
+	    cb, 0, &acmd->cmd_dmahandle)) != DDI_SUCCESS) {
+		switch (i) {
+		case DDI_DMA_BADATTR:
+			bioerror(bp, EFAULT);
+			return (DDI_FAILURE);
+
+		case DDI_DMA_NORESOURCES:
+			bioerror(bp, 0);
+			return (DDI_FAILURE);
+
+		default:
+			con_log(CL_ANN, (CE_PANIC, "ddi_dma_alloc_handle: "
+			    "impossible result (0x%x)", i));
+			bioerror(bp, EFAULT);
+			return (DDI_FAILURE);
+		}
+	}
+
+	i = ddi_dma_buf_bind_handle(acmd->cmd_dmahandle, bp, dma_flags,
+	    cb, 0, &acmd->cmd_dmacookies[0], &acmd->cmd_ncookies);
+
+	switch (i) {
+	case DDI_DMA_PARTIAL_MAP:
+		if ((dma_flags & DDI_DMA_PARTIAL) == 0) {
+			con_log(CL_ANN, (CE_PANIC, "ddi_dma_buf_bind_handle: "
+			    "DDI_DMA_PARTIAL_MAP impossible"));
+			goto no_dma_cookies;
+		}
+
+		if (ddi_dma_numwin(acmd->cmd_dmahandle, &acmd->cmd_nwin) ==
+		    DDI_FAILURE) {
+			con_log(CL_ANN, (CE_PANIC, "ddi_dma_numwin failed"));
+			goto no_dma_cookies;
+		}
+
+		if (ddi_dma_getwin(acmd->cmd_dmahandle, acmd->cmd_curwin,
+		    &acmd->cmd_dma_offset, &acmd->cmd_dma_len,
+		    &acmd->cmd_dmacookies[0], &acmd->cmd_ncookies) ==
+		    DDI_FAILURE) {
+
+			con_log(CL_ANN, (CE_PANIC, "ddi_dma_getwin failed"));
+			goto no_dma_cookies;
+		}
+
+		goto get_dma_cookies;
+	case DDI_DMA_MAPPED:
+		acmd->cmd_nwin = 1;
+		acmd->cmd_dma_len = 0;
+		acmd->cmd_dma_offset = 0;
+
+get_dma_cookies:
+		i = 0;
+		acmd->cmd_dmacount = 0;
+		for (;;) {
+			acmd->cmd_dmacount +=
+			    acmd->cmd_dmacookies[i++].dmac_size;
+
+			if (i == instance->max_num_sge ||
+			    i == acmd->cmd_ncookies)
+				break;
+
+			ddi_dma_nextcookie(acmd->cmd_dmahandle,
+			    &acmd->cmd_dmacookies[i]);
+		}
+
+		acmd->cmd_cookie = i;
+		acmd->cmd_cookiecnt = i;
+
+		acmd->cmd_flags |= CFLAG_DMAVALID;
+
+		if (bp->b_bcount >= acmd->cmd_dmacount) {
+			pkt->pkt_resid = bp->b_bcount - acmd->cmd_dmacount;
+		} else {
+			pkt->pkt_resid = 0;
+		}
+
+		return (DDI_SUCCESS);
+	case DDI_DMA_NORESOURCES:
+		bioerror(bp, 0);
+		break;
+	case DDI_DMA_NOMAPPING:
+		bioerror(bp, EFAULT);
+		break;
+	case DDI_DMA_TOOBIG:
+		bioerror(bp, EINVAL);
+		break;
+	case DDI_DMA_INUSE:
+		con_log(CL_ANN, (CE_PANIC, "ddi_dma_buf_bind_handle:"
+		    " DDI_DMA_INUSE impossible"));
+		break;
+	default:
+		con_log(CL_ANN, (CE_PANIC, "ddi_dma_buf_bind_handle: "
+		    "impossible result (0x%x)", i));
+		break;
+	}
+
+no_dma_cookies:
+	ddi_dma_free_handle(&acmd->cmd_dmahandle);
+	acmd->cmd_dmahandle = NULL;
+	acmd->cmd_flags &= ~CFLAG_DMAVALID;
+	return (DDI_FAILURE);
+}
+
+/*
+ * drsas_dma_move(struct drsas_instance *, struct scsi_pkt *, struct buf *)
+ *
+ * move dma resources to next dma window
+ *
+ */
+static int
+drsas_dma_move(struct drsas_instance *instance, struct scsi_pkt *pkt,
+    struct buf *bp)
+{
+	int	i = 0;
+
+	struct scsa_cmd	*acmd = PKT2CMD(pkt);
+
+	/*
+	 * If there are no more cookies remaining in this window,
+	 * must move to the next window first.
+	 */
+	if (acmd->cmd_cookie == acmd->cmd_ncookies) {
+		if (acmd->cmd_curwin == acmd->cmd_nwin && acmd->cmd_nwin == 1) {
+			return (DDI_SUCCESS);
+		}
+
+		/* at last window, cannot move */
+		if (++acmd->cmd_curwin >= acmd->cmd_nwin) {
+			return (DDI_FAILURE);
+		}
+
+		if (ddi_dma_getwin(acmd->cmd_dmahandle, acmd->cmd_curwin,
+		    &acmd->cmd_dma_offset, &acmd->cmd_dma_len,
+		    &acmd->cmd_dmacookies[0], &acmd->cmd_ncookies) ==
+		    DDI_FAILURE) {
+			return (DDI_FAILURE);
+		}
+
+		acmd->cmd_cookie = 0;
+	} else {
+		/* still more cookies in this window - get the next one */
+		ddi_dma_nextcookie(acmd->cmd_dmahandle,
+		    &acmd->cmd_dmacookies[0]);
+	}
+
+	/* get remaining cookies in this window, up to our maximum */
+	for (;;) {
+		acmd->cmd_dmacount += acmd->cmd_dmacookies[i++].dmac_size;
+		acmd->cmd_cookie++;
+
+		if (i == instance->max_num_sge ||
+		    acmd->cmd_cookie == acmd->cmd_ncookies) {
+			break;
+		}
+
+		ddi_dma_nextcookie(acmd->cmd_dmahandle,
+		    &acmd->cmd_dmacookies[i]);
+	}
+
+	acmd->cmd_cookiecnt = i;
+
+	if (bp->b_bcount >= acmd->cmd_dmacount) {
+		pkt->pkt_resid = bp->b_bcount - acmd->cmd_dmacount;
+	} else {
+		pkt->pkt_resid = 0;
+	}
+
+	return (DDI_SUCCESS);
+}
+
+/*
+ * build_cmd
+ */
+static struct drsas_cmd *
+build_cmd(struct drsas_instance *instance, struct scsi_address *ap,
+    struct scsi_pkt *pkt, uchar_t *cmd_done)
+{
+	uint16_t	flags = 0;
+	uint32_t	i;
+	uint32_t 	context;
+	uint32_t	sge_bytes;
+	ddi_acc_handle_t acc_handle;
+	struct drsas_cmd		*cmd;
+	struct drsas_sge64		*mfi_sgl;
+	struct scsa_cmd			*acmd = PKT2CMD(pkt);
+	struct drsas_pthru_frame 	*pthru;
+	struct drsas_io_frame		*ldio;
+
+	/* find out if this is logical or physical drive command.  */
+	acmd->islogical = MRDRV_IS_LOGICAL(ap);
+	acmd->device_id = MAP_DEVICE_ID(instance, ap);
+	*cmd_done = 0;
+
+	/* get the command packet */
+	if (!(cmd = get_mfi_pkt(instance))) {
+		return (NULL);
+	}
+
+	acc_handle = cmd->frame_dma_obj.acc_handle;
+
+	/* Clear the frame buffer and assign back the context id */
+	(void) memset((char *)&cmd->frame[0], 0, sizeof (union drsas_frame));
+	ddi_put32(acc_handle, &cmd->frame->hdr.context, cmd->index);
+
+	cmd->pkt = pkt;
+	cmd->cmd = acmd;
+
+	/* lets get the command directions */
+	if (acmd->cmd_flags & CFLAG_DMASEND) {
+		flags = MFI_FRAME_DIR_WRITE;
+
+		if (acmd->cmd_flags & CFLAG_CONSISTENT) {
+			(void) ddi_dma_sync(acmd->cmd_dmahandle,
+			    acmd->cmd_dma_offset, acmd->cmd_dma_len,
+			    DDI_DMA_SYNC_FORDEV);
+		}
+	} else if (acmd->cmd_flags & ~CFLAG_DMASEND) {
+		flags = MFI_FRAME_DIR_READ;
+
+		if (acmd->cmd_flags & CFLAG_CONSISTENT) {
+			(void) ddi_dma_sync(acmd->cmd_dmahandle,
+			    acmd->cmd_dma_offset, acmd->cmd_dma_len,
+			    DDI_DMA_SYNC_FORCPU);
+		}
+	} else {
+		flags = MFI_FRAME_DIR_NONE;
+	}
+
+	flags |= MFI_FRAME_SGL64;
+
+	switch (pkt->pkt_cdbp[0]) {
+
+	/*
+	 * case SCMD_SYNCHRONIZE_CACHE:
+	 * 	flush_cache(instance);
+	 *	return_mfi_pkt(instance, cmd);
+	 *	*cmd_done = 1;
+	 *
+	 *	return (NULL);
+	 */
+
+	case SCMD_READ:
+	case SCMD_WRITE:
+	case SCMD_READ_G1:
+	case SCMD_WRITE_G1:
+		if (acmd->islogical) {
+			ldio = (struct drsas_io_frame *)cmd->frame;
+
+			/*
+			 * preare the Logical IO frame:
+			 * 2nd bit is zero for all read cmds
+			 */
+			ddi_put8(acc_handle, &ldio->cmd,
+			    (pkt->pkt_cdbp[0] & 0x02) ? MFI_CMD_OP_LD_WRITE
+			    : MFI_CMD_OP_LD_READ);
+			ddi_put8(acc_handle, &ldio->cmd_status, 0x0);
+			ddi_put8(acc_handle, &ldio->scsi_status, 0x0);
+			ddi_put8(acc_handle, &ldio->target_id, acmd->device_id);
+			ddi_put16(acc_handle, &ldio->timeout, 0);
+			ddi_put8(acc_handle, &ldio->reserved_0, 0);
+			ddi_put16(acc_handle, &ldio->pad_0, 0);
+			ddi_put16(acc_handle, &ldio->flags, flags);
+
+			/* Initialize sense Information */
+			bzero(cmd->sense, SENSE_LENGTH);
+			ddi_put8(acc_handle, &ldio->sense_len, SENSE_LENGTH);
+			ddi_put32(acc_handle, &ldio->sense_buf_phys_addr_hi, 0);
+			ddi_put32(acc_handle, &ldio->sense_buf_phys_addr_lo,
+			    cmd->sense_phys_addr);
+			ddi_put32(acc_handle, &ldio->start_lba_hi, 0);
+			ddi_put8(acc_handle, &ldio->access_byte,
+			    (acmd->cmd_cdblen != 6) ? pkt->pkt_cdbp[1] : 0);
+			ddi_put8(acc_handle, &ldio->sge_count,
+			    acmd->cmd_cookiecnt);
+			mfi_sgl = (struct drsas_sge64	*)&ldio->sgl;
+
+			context = ddi_get32(acc_handle, &ldio->context);
+
+			if (acmd->cmd_cdblen == CDB_GROUP0) {
+				ddi_put32(acc_handle, &ldio->lba_count, (
+				    (uint16_t)(pkt->pkt_cdbp[4])));
+
+				ddi_put32(acc_handle, &ldio->start_lba_lo, (
+				    ((uint32_t)(pkt->pkt_cdbp[3])) |
+				    ((uint32_t)(pkt->pkt_cdbp[2]) << 8) |
+				    ((uint32_t)((pkt->pkt_cdbp[1]) & 0x1F)
+				    << 16)));
+			} else if (acmd->cmd_cdblen == CDB_GROUP1) {
+				ddi_put32(acc_handle, &ldio->lba_count, (
+				    ((uint16_t)(pkt->pkt_cdbp[8])) |
+				    ((uint16_t)(pkt->pkt_cdbp[7]) << 8)));
+
+				ddi_put32(acc_handle, &ldio->start_lba_lo, (
+				    ((uint32_t)(pkt->pkt_cdbp[5])) |
+				    ((uint32_t)(pkt->pkt_cdbp[4]) << 8) |
+				    ((uint32_t)(pkt->pkt_cdbp[3]) << 16) |
+				    ((uint32_t)(pkt->pkt_cdbp[2]) << 24)));
+			} else if (acmd->cmd_cdblen == CDB_GROUP2) {
+				ddi_put32(acc_handle, &ldio->lba_count, (
+				    ((uint16_t)(pkt->pkt_cdbp[9])) |
+				    ((uint16_t)(pkt->pkt_cdbp[8]) << 8) |
+				    ((uint16_t)(pkt->pkt_cdbp[7]) << 16) |
+				    ((uint16_t)(pkt->pkt_cdbp[6]) << 24)));
+
+				ddi_put32(acc_handle, &ldio->start_lba_lo, (
+				    ((uint32_t)(pkt->pkt_cdbp[5])) |
+				    ((uint32_t)(pkt->pkt_cdbp[4]) << 8) |
+				    ((uint32_t)(pkt->pkt_cdbp[3]) << 16) |
+				    ((uint32_t)(pkt->pkt_cdbp[2]) << 24)));
+			} else if (acmd->cmd_cdblen == CDB_GROUP3) {
+				ddi_put32(acc_handle, &ldio->lba_count, (
+				    ((uint16_t)(pkt->pkt_cdbp[13])) |
+				    ((uint16_t)(pkt->pkt_cdbp[12]) << 8) |
+				    ((uint16_t)(pkt->pkt_cdbp[11]) << 16) |
+				    ((uint16_t)(pkt->pkt_cdbp[10]) << 24)));
+
+				ddi_put32(acc_handle, &ldio->start_lba_lo, (
+				    ((uint32_t)(pkt->pkt_cdbp[9])) |
+				    ((uint32_t)(pkt->pkt_cdbp[8]) << 8) |
+				    ((uint32_t)(pkt->pkt_cdbp[7]) << 16) |
+				    ((uint32_t)(pkt->pkt_cdbp[6]) << 24)));
+
+				ddi_put32(acc_handle, &ldio->start_lba_lo, (
+				    ((uint32_t)(pkt->pkt_cdbp[5])) |
+				    ((uint32_t)(pkt->pkt_cdbp[4]) << 8) |
+				    ((uint32_t)(pkt->pkt_cdbp[3]) << 16) |
+				    ((uint32_t)(pkt->pkt_cdbp[2]) << 24)));
+			}
+
+			break;
+		}
+		/* fall through For all non-rd/wr cmds */
+	default:
+
+		switch (pkt->pkt_cdbp[0]) {
+		case SCMD_MODE_SENSE:
+		case SCMD_MODE_SENSE_G1: {
+			union scsi_cdb	*cdbp;
+			uint16_t	page_code;
+
+			cdbp = (void *)pkt->pkt_cdbp;
+			page_code = (uint16_t)cdbp->cdb_un.sg.scsi[0];
+			switch (page_code) {
+			case 0x3:
+			case 0x4:
+				(void) drsas_mode_sense_build(pkt);
+				return_mfi_pkt(instance, cmd);
+				*cmd_done = 1;
+				return (NULL);
+			}
+			break;
+		}
+		default:
+			break;
+		}
+
+		pthru	= (struct drsas_pthru_frame *)cmd->frame;
+
+		/* prepare the DCDB frame */
+		ddi_put8(acc_handle, &pthru->cmd, (acmd->islogical) ?
+		    MFI_CMD_OP_LD_SCSI : MFI_CMD_OP_PD_SCSI);
+		ddi_put8(acc_handle, &pthru->cmd_status, 0x0);
+		ddi_put8(acc_handle, &pthru->scsi_status, 0x0);
+		ddi_put8(acc_handle, &pthru->target_id, acmd->device_id);
+		ddi_put8(acc_handle, &pthru->lun, 0);
+		ddi_put8(acc_handle, &pthru->cdb_len, acmd->cmd_cdblen);
+		ddi_put16(acc_handle, &pthru->timeout, 0);
+		ddi_put16(acc_handle, &pthru->flags, flags);
+		ddi_put32(acc_handle, &pthru->data_xfer_len,
+		    acmd->cmd_dmacount);
+		ddi_put8(acc_handle, &pthru->sge_count, acmd->cmd_cookiecnt);
+		mfi_sgl			= (struct drsas_sge64 *)&pthru->sgl;
+
+		bzero(cmd->sense, SENSE_LENGTH);
+		ddi_put8(acc_handle, &pthru->sense_len, SENSE_LENGTH);
+		ddi_put32(acc_handle, &pthru->sense_buf_phys_addr_hi, 0);
+		ddi_put32(acc_handle, &pthru->sense_buf_phys_addr_lo,
+		    cmd->sense_phys_addr);
+
+		context = ddi_get32(acc_handle, &pthru->context);
+		ddi_rep_put8(acc_handle, (uint8_t *)pkt->pkt_cdbp,
+		    (uint8_t *)pthru->cdb, acmd->cmd_cdblen, DDI_DEV_AUTOINCR);
+
+		break;
+	}
+#ifdef lint
+	context = context;
+#endif
+	/* prepare the scatter-gather list for the firmware */
+	for (i = 0; i < acmd->cmd_cookiecnt; i++, mfi_sgl++) {
+		ddi_put64(acc_handle, &mfi_sgl->phys_addr,
+		    acmd->cmd_dmacookies[i].dmac_laddress);
+		ddi_put32(acc_handle, &mfi_sgl->length,
+		    acmd->cmd_dmacookies[i].dmac_size);
+	}
+
+	sge_bytes = sizeof (struct drsas_sge64)*acmd->cmd_cookiecnt;
+
+	cmd->frame_count = (sge_bytes / MRMFI_FRAME_SIZE) +
+	    ((sge_bytes % MRMFI_FRAME_SIZE) ? 1 : 0) + 1;
+
+	if (cmd->frame_count >= 8) {
+		cmd->frame_count = 8;
+	}
+
+	return (cmd);
+}
+
+/*
+ * issue_mfi_pthru
+ */
+static int
+issue_mfi_pthru(struct drsas_instance *instance, struct drsas_ioctl *ioctl,
+    struct drsas_cmd *cmd, int mode)
+{
+	void		*ubuf;
+	uint32_t	kphys_addr = 0;
+	uint32_t	xferlen = 0;
+	uint_t		model;
+	ddi_acc_handle_t	acc_handle = cmd->frame_dma_obj.acc_handle;
+	dma_obj_t			pthru_dma_obj;
+	struct drsas_pthru_frame	*kpthru;
+	struct drsas_pthru_frame	*pthru;
+	int i;
+	pthru = &cmd->frame->pthru;
+	kpthru = (struct drsas_pthru_frame *)&ioctl->frame[0];
+
+	model = ddi_model_convert_from(mode & FMODELS);
+	if (model == DDI_MODEL_ILP32) {
+		con_log(CL_ANN1, (CE_NOTE, "issue_mfi_pthru: DDI_MODEL_LP32"));
+
+		xferlen	= kpthru->sgl.sge32[0].length;
+
+		ubuf	= (void *)(ulong_t)kpthru->sgl.sge32[0].phys_addr;
+	} else {
+#ifdef _ILP32
+		con_log(CL_ANN1, (CE_NOTE, "issue_mfi_pthru: DDI_MODEL_LP32"));
+		xferlen	= kpthru->sgl.sge32[0].length;
+		ubuf	= (void *)(ulong_t)kpthru->sgl.sge32[0].phys_addr;
+#else
+		con_log(CL_ANN1, (CE_NOTE, "issue_mfi_pthru: DDI_MODEL_LP64"));
+		xferlen	= kpthru->sgl.sge64[0].length;
+		ubuf	= (void *)(ulong_t)kpthru->sgl.sge64[0].phys_addr;
+#endif
+	}
+
+	if (xferlen) {
+		/* means IOCTL requires DMA */
+		/* allocate the data transfer buffer */
+		pthru_dma_obj.size = xferlen;
+		pthru_dma_obj.dma_attr = drsas_generic_dma_attr;
+		pthru_dma_obj.dma_attr.dma_attr_addr_hi = 0xFFFFFFFFU;
+		pthru_dma_obj.dma_attr.dma_attr_count_max = 0xFFFFFFFFU;
+		pthru_dma_obj.dma_attr.dma_attr_sgllen = 1;
+		pthru_dma_obj.dma_attr.dma_attr_align = 1;
+
+		/* allocate kernel buffer for DMA */
+		if (drsas_alloc_dma_obj(instance, &pthru_dma_obj,
+		    (uchar_t)DDI_STRUCTURE_LE_ACC) != 1) {
+			con_log(CL_ANN, (CE_WARN, "issue_mfi_pthru: "
+			    "could not allocate data transfer buffer."));
+			return (DDI_FAILURE);
+		}
+
+		/* If IOCTL requires DMA WRITE, do ddi_copyin IOCTL data copy */
+		if (kpthru->flags & MFI_FRAME_DIR_WRITE) {
+			for (i = 0; i < xferlen; i++) {
+				if (ddi_copyin((uint8_t *)ubuf+i,
+				    (uint8_t *)pthru_dma_obj.buffer+i,
+				    1, mode)) {
+					con_log(CL_ANN, (CE_WARN,
+					    "issue_mfi_pthru : "
+					    "copy from user space failed"));
+					return (DDI_FAILURE);
+				}
+			}
+		}
+
+		kphys_addr = pthru_dma_obj.dma_cookie[0].dmac_address;
+	}
+
+	ddi_put8(acc_handle, &pthru->cmd, kpthru->cmd);
+	ddi_put8(acc_handle, &pthru->sense_len, kpthru->sense_len);
+	ddi_put8(acc_handle, &pthru->cmd_status, 0);
+	ddi_put8(acc_handle, &pthru->scsi_status, 0);
+	ddi_put8(acc_handle, &pthru->target_id, kpthru->target_id);
+	ddi_put8(acc_handle, &pthru->lun, kpthru->lun);
+	ddi_put8(acc_handle, &pthru->cdb_len, kpthru->cdb_len);
+	ddi_put8(acc_handle, &pthru->sge_count, kpthru->sge_count);
+	ddi_put16(acc_handle, &pthru->timeout, kpthru->timeout);
+	ddi_put32(acc_handle, &pthru->data_xfer_len, kpthru->data_xfer_len);
+
+	ddi_put32(acc_handle, &pthru->sense_buf_phys_addr_hi, 0);
+	/* pthru->sense_buf_phys_addr_lo = cmd->sense_phys_addr; */
+	ddi_put32(acc_handle, &pthru->sense_buf_phys_addr_lo, 0);
+
+	ddi_rep_put8(acc_handle, (uint8_t *)kpthru->cdb, (uint8_t *)pthru->cdb,
+	    pthru->cdb_len, DDI_DEV_AUTOINCR);
+
+	ddi_put16(acc_handle, &pthru->flags, kpthru->flags & ~MFI_FRAME_SGL64);
+	ddi_put32(acc_handle, &pthru->sgl.sge32[0].length, xferlen);
+	ddi_put32(acc_handle, &pthru->sgl.sge32[0].phys_addr, kphys_addr);
+
+	cmd->sync_cmd = DRSAS_TRUE;
+	cmd->frame_count = 1;
+
+	if (instance->func_ptr->issue_cmd_in_sync_mode(instance, cmd)) {
+		con_log(CL_ANN, (CE_WARN,
+		    "issue_mfi_pthru: fw_ioctl failed"));
+	} else {
+		if (xferlen && kpthru->flags & MFI_FRAME_DIR_READ) {
+			for (i = 0; i < xferlen; i++) {
+				if (ddi_copyout(
+				    (uint8_t *)pthru_dma_obj.buffer+i,
+				    (uint8_t *)ubuf+i, 1, mode)) {
+					con_log(CL_ANN, (CE_WARN,
+					    "issue_mfi_pthru : "
+					    "copy to user space failed"));
+					return (DDI_FAILURE);
+				}
+			}
+		}
+	}
+
+	kpthru->cmd_status = ddi_get8(acc_handle, &pthru->cmd_status);
+	kpthru->scsi_status = ddi_get8(acc_handle, &pthru->scsi_status);
+
+	con_log(CL_ANN, (CE_NOTE, "issue_mfi_pthru: cmd_status %x, "
+	    "scsi_status %x", kpthru->cmd_status, kpthru->scsi_status));
+
+	if (xferlen) {
+		/* free kernel buffer */
+		if (drsas_free_dma_obj(instance, pthru_dma_obj) != DDI_SUCCESS)
+			return (DDI_FAILURE);
+	}
+
+	return (DDI_SUCCESS);
+}
+
+/*
+ * issue_mfi_dcmd
+ */
+static int
+issue_mfi_dcmd(struct drsas_instance *instance, struct drsas_ioctl *ioctl,
+    struct drsas_cmd *cmd, int mode)
+{
+	void		*ubuf;
+	uint32_t	kphys_addr = 0;
+	uint32_t	xferlen = 0;
+	uint32_t	model;
+	dma_obj_t	dcmd_dma_obj;
+	struct drsas_dcmd_frame	*kdcmd;
+	struct drsas_dcmd_frame	*dcmd;
+	ddi_acc_handle_t	acc_handle = cmd->frame_dma_obj.acc_handle;
+	int i;
+	dcmd = &cmd->frame->dcmd;
+	kdcmd = (struct drsas_dcmd_frame *)&ioctl->frame[0];
+
+	model = ddi_model_convert_from(mode & FMODELS);
+	if (model == DDI_MODEL_ILP32) {
+		con_log(CL_ANN1, (CE_NOTE, "issue_mfi_dcmd: DDI_MODEL_ILP32"));
+
+		xferlen	= kdcmd->sgl.sge32[0].length;
+
+		ubuf	= (void *)(ulong_t)kdcmd->sgl.sge32[0].phys_addr;
+	} else {
+#ifdef _ILP32
+		con_log(CL_ANN1, (CE_NOTE, "issue_mfi_dcmd: DDI_MODEL_ILP32"));
+		xferlen	= kdcmd->sgl.sge32[0].length;
+		ubuf	= (void *)(ulong_t)kdcmd->sgl.sge32[0].phys_addr;
+#else
+		con_log(CL_ANN1, (CE_NOTE, "issue_mfi_dcmd: DDI_MODEL_LP64"));
+		xferlen	= kdcmd->sgl.sge64[0].length;
+		ubuf	= (void *)(ulong_t)kdcmd->sgl.sge64[0].phys_addr;
+#endif
+	}
+	if (xferlen) {
+		/* means IOCTL requires DMA */
+		/* allocate the data transfer buffer */
+		dcmd_dma_obj.size = xferlen;
+		dcmd_dma_obj.dma_attr = drsas_generic_dma_attr;
+		dcmd_dma_obj.dma_attr.dma_attr_addr_hi = 0xFFFFFFFFU;
+		dcmd_dma_obj.dma_attr.dma_attr_count_max = 0xFFFFFFFFU;
+		dcmd_dma_obj.dma_attr.dma_attr_sgllen = 1;
+		dcmd_dma_obj.dma_attr.dma_attr_align = 1;
+
+		/* allocate kernel buffer for DMA */
+		if (drsas_alloc_dma_obj(instance, &dcmd_dma_obj,
+		    (uchar_t)DDI_STRUCTURE_LE_ACC) != 1) {
+			con_log(CL_ANN, (CE_WARN, "issue_mfi_dcmd: "
+			    "could not allocate data transfer buffer."));
+			return (DDI_FAILURE);
+		}
+
+		/* If IOCTL requires DMA WRITE, do ddi_copyin IOCTL data copy */
+		if (kdcmd->flags & MFI_FRAME_DIR_WRITE) {
+			for (i = 0; i < xferlen; i++) {
+				if (ddi_copyin((uint8_t *)ubuf + i,
+				    (uint8_t *)dcmd_dma_obj.buffer + i,
+				    1, mode)) {
+					con_log(CL_ANN, (CE_WARN,
+					    "issue_mfi_dcmd : "
+					    "copy from user space failed"));
+					return (DDI_FAILURE);
+				}
+			}
+		}
+
+		kphys_addr = dcmd_dma_obj.dma_cookie[0].dmac_address;
+	}
+
+	ddi_put8(acc_handle, &dcmd->cmd, kdcmd->cmd);
+	ddi_put8(acc_handle, &dcmd->cmd_status, 0);
+	ddi_put8(acc_handle, &dcmd->sge_count, kdcmd->sge_count);
+	ddi_put16(acc_handle, &dcmd->timeout, kdcmd->timeout);
+	ddi_put32(acc_handle, &dcmd->data_xfer_len, kdcmd->data_xfer_len);
+	ddi_put32(acc_handle, &dcmd->opcode, kdcmd->opcode);
+
+	ddi_rep_put8(acc_handle, (uint8_t *)kdcmd->mbox.b,
+	    (uint8_t *)dcmd->mbox.b, DCMD_MBOX_SZ, DDI_DEV_AUTOINCR);
+
+	ddi_put16(acc_handle, &dcmd->flags, kdcmd->flags & ~MFI_FRAME_SGL64);
+	ddi_put32(acc_handle, &dcmd->sgl.sge32[0].length, xferlen);
+	ddi_put32(acc_handle, &dcmd->sgl.sge32[0].phys_addr, kphys_addr);
+
+	cmd->sync_cmd = DRSAS_TRUE;
+	cmd->frame_count = 1;
+
+	if (instance->func_ptr->issue_cmd_in_sync_mode(instance, cmd)) {
+		con_log(CL_ANN, (CE_WARN, "issue_mfi_dcmd: fw_ioctl failed"));
+	} else {
+		if (xferlen && (kdcmd->flags & MFI_FRAME_DIR_READ)) {
+			for (i = 0; i < xferlen; i++) {
+				if (ddi_copyout(
+				    (uint8_t *)dcmd_dma_obj.buffer + i,
+				    (uint8_t *)ubuf + i,
+				    1, mode)) {
+					con_log(CL_ANN, (CE_WARN,
+					    "issue_mfi_dcmd : "
+					    "copy to user space failed"));
+					return (DDI_FAILURE);
+				}
+			}
+		}
+	}
+
+	kdcmd->cmd_status = ddi_get8(acc_handle, &dcmd->cmd_status);
+
+	if (xferlen) {
+		/* free kernel buffer */
+		if (drsas_free_dma_obj(instance, dcmd_dma_obj) != DDI_SUCCESS)
+			return (DDI_FAILURE);
+	}
+
+	return (DDI_SUCCESS);
+}
+
+/*
+ * issue_mfi_smp
+ */
+static int
+issue_mfi_smp(struct drsas_instance *instance, struct drsas_ioctl *ioctl,
+    struct drsas_cmd *cmd, int mode)
+{
+	void		*request_ubuf;
+	void		*response_ubuf;
+	uint32_t	request_xferlen = 0;
+	uint32_t	response_xferlen = 0;
+	uint_t		model;
+	dma_obj_t			request_dma_obj;
+	dma_obj_t			response_dma_obj;
+	ddi_acc_handle_t	acc_handle = cmd->frame_dma_obj.acc_handle;
+	struct drsas_smp_frame		*ksmp;
+	struct drsas_smp_frame		*smp;
+	struct drsas_sge32		*sge32;
+#ifndef _ILP32
+	struct drsas_sge64		*sge64;
+#endif
+	int i;
+	uint64_t			tmp_sas_addr;
+
+	smp = &cmd->frame->smp;
+	ksmp = (struct drsas_smp_frame *)&ioctl->frame[0];
+
+	model = ddi_model_convert_from(mode & FMODELS);
+	if (model == DDI_MODEL_ILP32) {
+		con_log(CL_ANN1, (CE_NOTE, "issue_mfi_smp: DDI_MODEL_ILP32"));
+
+		sge32			= &ksmp->sgl[0].sge32[0];
+		response_xferlen	= sge32[0].length;
+		request_xferlen		= sge32[1].length;
+		con_log(CL_ANN, (CE_NOTE, "issue_mfi_smp: "
+		    "response_xferlen = %x, request_xferlen = %x",
+		    response_xferlen, request_xferlen));
+
+		response_ubuf	= (void *)(ulong_t)sge32[0].phys_addr;
+		request_ubuf	= (void *)(ulong_t)sge32[1].phys_addr;
+		con_log(CL_ANN1, (CE_NOTE, "issue_mfi_smp: "
+		    "response_ubuf = %p, request_ubuf = %p",
+		    response_ubuf, request_ubuf));
+	} else {
+#ifdef _ILP32
+		con_log(CL_ANN1, (CE_NOTE, "issue_mfi_smp: DDI_MODEL_ILP32"));
+
+		sge32			= &ksmp->sgl[0].sge32[0];
+		response_xferlen	= sge32[0].length;
+		request_xferlen		= sge32[1].length;
+		con_log(CL_ANN, (CE_NOTE, "issue_mfi_smp: "
+		    "response_xferlen = %x, request_xferlen = %x",
+		    response_xferlen, request_xferlen));
+
+		response_ubuf	= (void *)(ulong_t)sge32[0].phys_addr;
+		request_ubuf	= (void *)(ulong_t)sge32[1].phys_addr;
+		con_log(CL_ANN1, (CE_NOTE, "issue_mfi_smp: "
+		    "response_ubuf = %p, request_ubuf = %p",
+		    response_ubuf, request_ubuf));
+#else
+		con_log(CL_ANN1, (CE_NOTE, "issue_mfi_smp: DDI_MODEL_LP64"));
+
+		sge64			= &ksmp->sgl[0].sge64[0];
+		response_xferlen	= sge64[0].length;
+		request_xferlen		= sge64[1].length;
+
+		response_ubuf	= (void *)(ulong_t)sge64[0].phys_addr;
+		request_ubuf	= (void *)(ulong_t)sge64[1].phys_addr;
+#endif
+	}
+	if (request_xferlen) {
+		/* means IOCTL requires DMA */
+		/* allocate the data transfer buffer */
+		request_dma_obj.size = request_xferlen;
+		request_dma_obj.dma_attr = drsas_generic_dma_attr;
+		request_dma_obj.dma_attr.dma_attr_addr_hi = 0xFFFFFFFFU;
+		request_dma_obj.dma_attr.dma_attr_count_max = 0xFFFFFFFFU;
+		request_dma_obj.dma_attr.dma_attr_sgllen = 1;
+		request_dma_obj.dma_attr.dma_attr_align = 1;
+
+		/* allocate kernel buffer for DMA */
+		if (drsas_alloc_dma_obj(instance, &request_dma_obj,
+		    (uchar_t)DDI_STRUCTURE_LE_ACC) != 1) {
+			con_log(CL_ANN, (CE_WARN, "issue_mfi_smp: "
+			    "could not allocate data transfer buffer."));
+			return (DDI_FAILURE);
+		}
+
+		/* If IOCTL requires DMA WRITE, do ddi_copyin IOCTL data copy */
+		for (i = 0; i < request_xferlen; i++) {
+			if (ddi_copyin((uint8_t *)request_ubuf + i,
+			    (uint8_t *)request_dma_obj.buffer + i,
+			    1, mode)) {
+				con_log(CL_ANN, (CE_WARN, "issue_mfi_smp: "
+				    "copy from user space failed"));
+				return (DDI_FAILURE);
+			}
+		}
+	}
+
+	if (response_xferlen) {
+		/* means IOCTL requires DMA */
+		/* allocate the data transfer buffer */
+		response_dma_obj.size = response_xferlen;
+		response_dma_obj.dma_attr = drsas_generic_dma_attr;
+		response_dma_obj.dma_attr.dma_attr_addr_hi = 0xFFFFFFFFU;
+		response_dma_obj.dma_attr.dma_attr_count_max = 0xFFFFFFFFU;
+		response_dma_obj.dma_attr.dma_attr_sgllen = 1;
+		response_dma_obj.dma_attr.dma_attr_align = 1;
+
+		/* allocate kernel buffer for DMA */
+		if (drsas_alloc_dma_obj(instance, &response_dma_obj,
+		    (uchar_t)DDI_STRUCTURE_LE_ACC) != 1) {
+			con_log(CL_ANN, (CE_WARN, "issue_mfi_smp: "
+			    "could not allocate data transfer buffer."));
+			return (DDI_FAILURE);
+		}
+
+		/* If IOCTL requires DMA WRITE, do ddi_copyin IOCTL data copy */
+		for (i = 0; i < response_xferlen; i++) {
+			if (ddi_copyin((uint8_t *)response_ubuf + i,
+			    (uint8_t *)response_dma_obj.buffer + i,
+			    1, mode)) {
+				con_log(CL_ANN, (CE_WARN, "issue_mfi_smp: "
+				    "copy from user space failed"));
+				return (DDI_FAILURE);
+			}
+		}
+	}
+
+	ddi_put8(acc_handle, &smp->cmd, ksmp->cmd);
+	ddi_put8(acc_handle, &smp->cmd_status, 0);
+	ddi_put8(acc_handle, &smp->connection_status, 0);
+	ddi_put8(acc_handle, &smp->sge_count, ksmp->sge_count);
+	/* smp->context		= ksmp->context; */
+	ddi_put16(acc_handle, &smp->timeout, ksmp->timeout);
+	ddi_put32(acc_handle, &smp->data_xfer_len, ksmp->data_xfer_len);
+
+	bcopy((void *)&ksmp->sas_addr, (void *)&tmp_sas_addr,
+	    sizeof (uint64_t));
+	ddi_put64(acc_handle, &smp->sas_addr, tmp_sas_addr);
+
+	ddi_put16(acc_handle, &smp->flags, ksmp->flags & ~MFI_FRAME_SGL64);
+
+	model = ddi_model_convert_from(mode & FMODELS);
+	if (model == DDI_MODEL_ILP32) {
+		con_log(CL_ANN1, (CE_NOTE,
+		    "handle_drv_ioctl: DDI_MODEL_ILP32"));
+
+		sge32 = &smp->sgl[0].sge32[0];
+		ddi_put32(acc_handle, &sge32[0].length, response_xferlen);
+		ddi_put32(acc_handle, &sge32[0].phys_addr,
+		    response_dma_obj.dma_cookie[0].dmac_address);
+		ddi_put32(acc_handle, &sge32[1].length, request_xferlen);
+		ddi_put32(acc_handle, &sge32[1].phys_addr,
+		    request_dma_obj.dma_cookie[0].dmac_address);
+	} else {
+#ifdef _ILP32
+		con_log(CL_ANN1, (CE_NOTE,
+		    "handle_drv_ioctl: DDI_MODEL_ILP32"));
+		sge32 = &smp->sgl[0].sge32[0];
+		ddi_put32(acc_handle, &sge32[0].length, response_xferlen);
+		ddi_put32(acc_handle, &sge32[0].phys_addr,
+		    response_dma_obj.dma_cookie[0].dmac_address);
+		ddi_put32(acc_handle, &sge32[1].length, request_xferlen);
+		ddi_put32(acc_handle, &sge32[1].phys_addr,
+		    request_dma_obj.dma_cookie[0].dmac_address);
+#else
+		con_log(CL_ANN1, (CE_NOTE,
+		    "issue_mfi_smp: DDI_MODEL_LP64"));
+		sge64 = &smp->sgl[0].sge64[0];
+		ddi_put32(acc_handle, &sge64[0].length, response_xferlen);
+		ddi_put64(acc_handle, &sge64[0].phys_addr,
+		    response_dma_obj.dma_cookie[0].dmac_address);
+		ddi_put32(acc_handle, &sge64[1].length, request_xferlen);
+		ddi_put64(acc_handle, &sge64[1].phys_addr,
+		    request_dma_obj.dma_cookie[0].dmac_address);
+#endif
+	}
+	con_log(CL_ANN1, (CE_NOTE, "issue_mfi_smp : "
+	    "smp->response_xferlen = %d, smp->request_xferlen = %d "
+	    "smp->data_xfer_len = %d", ddi_get32(acc_handle, &sge32[0].length),
+	    ddi_get32(acc_handle, &sge32[1].length),
+	    ddi_get32(acc_handle, &smp->data_xfer_len)));
+
+	cmd->sync_cmd = DRSAS_TRUE;
+	cmd->frame_count = 1;
+
+	if (instance->func_ptr->issue_cmd_in_sync_mode(instance, cmd)) {
+		con_log(CL_ANN, (CE_WARN,
+		    "issue_mfi_smp: fw_ioctl failed"));
+	} else {
+		con_log(CL_ANN1, (CE_NOTE,
+		    "issue_mfi_smp: copy to user space"));
+
+		if (request_xferlen) {
+			for (i = 0; i < request_xferlen; i++) {
+				if (ddi_copyout(
+				    (uint8_t *)request_dma_obj.buffer +
+				    i, (uint8_t *)request_ubuf + i,
+				    1, mode)) {
+					con_log(CL_ANN, (CE_WARN,
+					    "issue_mfi_smp : copy to user space"
+					    " failed"));
+					return (DDI_FAILURE);
+				}
+			}
+		}
+
+		if (response_xferlen) {
+			for (i = 0; i < response_xferlen; i++) {
+				if (ddi_copyout(
+				    (uint8_t *)response_dma_obj.buffer
+				    + i, (uint8_t *)response_ubuf
+				    + i, 1, mode)) {
+					con_log(CL_ANN, (CE_WARN,
+					    "issue_mfi_smp : copy to "
+					    "user space failed"));
+					return (DDI_FAILURE);
+				}
+			}
+		}
+	}
+
+	ksmp->cmd_status = ddi_get8(acc_handle, &smp->cmd_status);
+	con_log(CL_ANN1, (CE_NOTE, "issue_mfi_smp: smp->cmd_status = %d",
+	    ddi_get8(acc_handle, &smp->cmd_status)));
+
+
+	if (request_xferlen) {
+		/* free kernel buffer */
+		if (drsas_free_dma_obj(instance, request_dma_obj) !=
+		    DDI_SUCCESS)
+			return (DDI_FAILURE);
+	}
+
+	if (response_xferlen) {
+		/* free kernel buffer */
+		if (drsas_free_dma_obj(instance, response_dma_obj) !=
+		    DDI_SUCCESS)
+			return (DDI_FAILURE);
+	}
+
+	return (DDI_SUCCESS);
+}
+
+/*
+ * issue_mfi_stp
+ */
+static int
+issue_mfi_stp(struct drsas_instance *instance, struct drsas_ioctl *ioctl,
+    struct drsas_cmd *cmd, int mode)
+{
+	void		*fis_ubuf;
+	void		*data_ubuf;
+	uint32_t	fis_xferlen = 0;
+	uint32_t	data_xferlen = 0;
+	uint_t		model;
+	dma_obj_t	fis_dma_obj;
+	dma_obj_t	data_dma_obj;
+	struct drsas_stp_frame	*kstp;
+	struct drsas_stp_frame	*stp;
+	ddi_acc_handle_t	acc_handle = cmd->frame_dma_obj.acc_handle;
+	int i;
+
+	stp = &cmd->frame->stp;
+	kstp = (struct drsas_stp_frame *)&ioctl->frame[0];
+
+	model = ddi_model_convert_from(mode & FMODELS);
+	if (model == DDI_MODEL_ILP32) {
+		con_log(CL_ANN1, (CE_NOTE, "issue_mfi_stp: DDI_MODEL_ILP32"));
+
+		fis_xferlen	= kstp->sgl.sge32[0].length;
+		data_xferlen	= kstp->sgl.sge32[1].length;
+
+		fis_ubuf	= (void *)(ulong_t)kstp->sgl.sge32[0].phys_addr;
+		data_ubuf	= (void *)(ulong_t)kstp->sgl.sge32[1].phys_addr;
+	}
+	else
+	{
+#ifdef _ILP32
+		con_log(CL_ANN1, (CE_NOTE, "issue_mfi_stp: DDI_MODEL_ILP32"));
+
+		fis_xferlen	= kstp->sgl.sge32[0].length;
+		data_xferlen	= kstp->sgl.sge32[1].length;
+
+		fis_ubuf	= (void *)(ulong_t)kstp->sgl.sge32[0].phys_addr;
+		data_ubuf	= (void *)(ulong_t)kstp->sgl.sge32[1].phys_addr;
+#else
+		con_log(CL_ANN1, (CE_NOTE, "issue_mfi_stp: DDI_MODEL_LP64"));
+
+		fis_xferlen	= kstp->sgl.sge64[0].length;
+		data_xferlen	= kstp->sgl.sge64[1].length;
+
+		fis_ubuf	= (void *)(ulong_t)kstp->sgl.sge64[0].phys_addr;
+		data_ubuf	= (void *)(ulong_t)kstp->sgl.sge64[1].phys_addr;
+#endif
+	}
+
+
+	if (fis_xferlen) {
+		con_log(CL_ANN, (CE_NOTE, "issue_mfi_stp: "
+		    "fis_ubuf = %p fis_xferlen = %x", fis_ubuf, fis_xferlen));
+
+		/* means IOCTL requires DMA */
+		/* allocate the data transfer buffer */
+		fis_dma_obj.size = fis_xferlen;
+		fis_dma_obj.dma_attr = drsas_generic_dma_attr;
+		fis_dma_obj.dma_attr.dma_attr_addr_hi = 0xFFFFFFFFU;
+		fis_dma_obj.dma_attr.dma_attr_count_max	= 0xFFFFFFFFU;
+		fis_dma_obj.dma_attr.dma_attr_sgllen = 1;
+		fis_dma_obj.dma_attr.dma_attr_align = 1;
+
+		/* allocate kernel buffer for DMA */
+		if (drsas_alloc_dma_obj(instance, &fis_dma_obj,
+		    (uchar_t)DDI_STRUCTURE_LE_ACC) != 1) {
+			con_log(CL_ANN, (CE_WARN, "issue_mfi_stp : "
+			    "could not allocate data transfer buffer."));
+			return (DDI_FAILURE);
+		}
+
+		/* If IOCTL requires DMA WRITE, do ddi_copyin IOCTL data copy */
+		for (i = 0; i < fis_xferlen; i++) {
+			if (ddi_copyin((uint8_t *)fis_ubuf + i,
+			    (uint8_t *)fis_dma_obj.buffer + i, 1, mode)) {
+				con_log(CL_ANN, (CE_WARN, "issue_mfi_stp: "
+				    "copy from user space failed"));
+				return (DDI_FAILURE);
+			}
+		}
+	}
+
+	if (data_xferlen) {
+		con_log(CL_ANN, (CE_NOTE, "issue_mfi_stp: data_ubuf = %p "
+		    "data_xferlen = %x", data_ubuf, data_xferlen));
+
+		/* means IOCTL requires DMA */
+		/* allocate the data transfer buffer */
+		data_dma_obj.size = data_xferlen;
+		data_dma_obj.dma_attr = drsas_generic_dma_attr;
+		data_dma_obj.dma_attr.dma_attr_addr_hi = 0xFFFFFFFFU;
+		data_dma_obj.dma_attr.dma_attr_count_max = 0xFFFFFFFFU;
+		data_dma_obj.dma_attr.dma_attr_sgllen = 1;
+		data_dma_obj.dma_attr.dma_attr_align = 1;
+
+/* allocate kernel buffer for DMA */
+		if (drsas_alloc_dma_obj(instance, &data_dma_obj,
+		    (uchar_t)DDI_STRUCTURE_LE_ACC) != 1) {
+			con_log(CL_ANN, (CE_WARN, "issue_mfi_stp: "
+			    "could not allocate data transfer buffer."));
+			return (DDI_FAILURE);
+		}
+
+		/* If IOCTL requires DMA WRITE, do ddi_copyin IOCTL data copy */
+		for (i = 0; i < data_xferlen; i++) {
+			if (ddi_copyin((uint8_t *)data_ubuf + i,
+			    (uint8_t *)data_dma_obj.buffer + i, 1, mode)) {
+				con_log(CL_ANN, (CE_WARN, "issue_mfi_stp: "
+				    "copy from user space failed"));
+				return (DDI_FAILURE);
+			}
+		}
+	}
+
+	ddi_put8(acc_handle, &stp->cmd, kstp->cmd);
+	ddi_put8(acc_handle, &stp->cmd_status, 0);
+	ddi_put8(acc_handle, &stp->connection_status, 0);
+	ddi_put8(acc_handle, &stp->target_id, kstp->target_id);
+	ddi_put8(acc_handle, &stp->sge_count, kstp->sge_count);
+
+	ddi_put16(acc_handle, &stp->timeout, kstp->timeout);
+	ddi_put32(acc_handle, &stp->data_xfer_len, kstp->data_xfer_len);
+
+	ddi_rep_put8(acc_handle, (uint8_t *)kstp->fis, (uint8_t *)stp->fis, 10,
+	    DDI_DEV_AUTOINCR);
+
+	ddi_put16(acc_handle, &stp->flags, kstp->flags & ~MFI_FRAME_SGL64);
+	ddi_put32(acc_handle, &stp->stp_flags, kstp->stp_flags);
+	ddi_put32(acc_handle, &stp->sgl.sge32[0].length, fis_xferlen);
+	ddi_put32(acc_handle, &stp->sgl.sge32[0].phys_addr,
+	    fis_dma_obj.dma_cookie[0].dmac_address);
+	ddi_put32(acc_handle, &stp->sgl.sge32[1].length, data_xferlen);
+	ddi_put32(acc_handle, &stp->sgl.sge32[1].phys_addr,
+	    data_dma_obj.dma_cookie[0].dmac_address);
+
+	cmd->sync_cmd = DRSAS_TRUE;
+	cmd->frame_count = 1;
+
+	if (instance->func_ptr->issue_cmd_in_sync_mode(instance, cmd)) {
+		con_log(CL_ANN, (CE_WARN, "issue_mfi_stp: fw_ioctl failed"));
+	} else {
+
+		if (fis_xferlen) {
+			for (i = 0; i < fis_xferlen; i++) {
+				if (ddi_copyout(
+				    (uint8_t *)fis_dma_obj.buffer + i,
+				    (uint8_t *)fis_ubuf + i, 1, mode)) {
+					con_log(CL_ANN, (CE_WARN,
+					    "issue_mfi_stp : copy to "
+					    "user space failed"));
+					return (DDI_FAILURE);
+				}
+			}
+		}
+	}
+	if (data_xferlen) {
+		for (i = 0; i < data_xferlen; i++) {
+			if (ddi_copyout(
+			    (uint8_t *)data_dma_obj.buffer + i,
+			    (uint8_t *)data_ubuf + i, 1, mode)) {
+				con_log(CL_ANN, (CE_WARN,
+				    "issue_mfi_stp : copy to"
+				    " user space failed"));
+				return (DDI_FAILURE);
+			}
+		}
+	}
+
+	kstp->cmd_status = ddi_get8(acc_handle, &stp->cmd_status);
+
+	if (fis_xferlen) {
+		/* free kernel buffer */
+		if (drsas_free_dma_obj(instance, fis_dma_obj) != DDI_SUCCESS)
+			return (DDI_FAILURE);
+	}
+
+	if (data_xferlen) {
+		/* free kernel buffer */
+		if (drsas_free_dma_obj(instance, data_dma_obj) != DDI_SUCCESS)
+			return (DDI_FAILURE);
+	}
+
+	return (DDI_SUCCESS);
+}
+
+/*
+ * fill_up_drv_ver
+ */
+static void
+fill_up_drv_ver(struct drsas_drv_ver *dv)
+{
+	(void) memset(dv, 0, sizeof (struct drsas_drv_ver));
+
+	(void) memcpy(dv->signature, "$LSI LOGIC$", strlen("$LSI LOGIC$"));
+	(void) memcpy(dv->os_name, "Solaris", strlen("Solaris"));
+	(void) memcpy(dv->drv_name, "dr_sas", strlen("dr_sas"));
+	(void) memcpy(dv->drv_ver, DRSAS_VERSION, strlen(DRSAS_VERSION));
+	(void) memcpy(dv->drv_rel_date, DRSAS_RELDATE,
+	    strlen(DRSAS_RELDATE));
+}
+
+/*
+ * handle_drv_ioctl
+ */
+static int
+handle_drv_ioctl(struct drsas_instance *instance, struct drsas_ioctl *ioctl,
+    int mode)
+{
+	int	i;
+	int	rval = DDI_SUCCESS;
+	int	*props = NULL;
+	void	*ubuf;
+
+	uint8_t		*pci_conf_buf;
+	uint32_t	xferlen;
+	uint32_t	num_props;
+	uint_t		model;
+	struct drsas_dcmd_frame	*kdcmd;
+	struct drsas_drv_ver	dv;
+	struct drsas_pci_information pi;
+
+	kdcmd = (struct drsas_dcmd_frame *)&ioctl->frame[0];
+
+	model = ddi_model_convert_from(mode & FMODELS);
+	if (model == DDI_MODEL_ILP32) {
+		con_log(CL_ANN1, (CE_NOTE,
+		    "handle_drv_ioctl: DDI_MODEL_ILP32"));
+
+		xferlen	= kdcmd->sgl.sge32[0].length;
+
+		ubuf = (void *)(ulong_t)kdcmd->sgl.sge32[0].phys_addr;
+	} else {
+#ifdef _ILP32
+		con_log(CL_ANN1, (CE_NOTE,
+		    "handle_drv_ioctl: DDI_MODEL_ILP32"));
+		xferlen	= kdcmd->sgl.sge32[0].length;
+		ubuf = (void *)(ulong_t)kdcmd->sgl.sge32[0].phys_addr;
+#else
+		con_log(CL_ANN1, (CE_NOTE,
+		    "handle_drv_ioctl: DDI_MODEL_LP64"));
+		xferlen	= kdcmd->sgl.sge64[0].length;
+		ubuf = (void *)(ulong_t)kdcmd->sgl.sge64[0].phys_addr;
+#endif
+	}
+	con_log(CL_ANN1, (CE_NOTE, "handle_drv_ioctl: "
+	    "dataBuf=%p size=%d bytes", ubuf, xferlen));
+
+	switch (kdcmd->opcode) {
+	case DRSAS_DRIVER_IOCTL_DRIVER_VERSION:
+		con_log(CL_ANN1, (CE_NOTE, "handle_drv_ioctl: "
+		    "DRSAS_DRIVER_IOCTL_DRIVER_VERSION"));
+
+		fill_up_drv_ver(&dv);
+		for (i = 0; i < xferlen; i++) {
+			if (ddi_copyout((uint8_t *)&dv + i, (uint8_t *)ubuf + i,
+			    1, mode)) {
+				con_log(CL_ANN, (CE_WARN, "handle_drv_ioctl: "
+				    "DRSAS_DRIVER_IOCTL_DRIVER_VERSION"
+				    " : copy to user space failed"));
+				kdcmd->cmd_status = 1;
+				rval = DDI_FAILURE;
+				break;
+			}
+		}
+		if (i == xferlen)
+			kdcmd->cmd_status = 0;
+		break;
+	case DRSAS_DRIVER_IOCTL_PCI_INFORMATION:
+		con_log(CL_ANN1, (CE_NOTE, "handle_drv_ioctl: "
+		    "DRSAS_DRIVER_IOCTL_PCI_INFORMAITON"));
+
+		if (ddi_prop_lookup_int_array(DDI_DEV_T_ANY, instance->dip,
+		    0, "reg", &props, &num_props)) {
+			con_log(CL_ANN, (CE_WARN, "handle_drv_ioctl: "
+			    "DRSAS_DRIVER_IOCTL_PCI_INFORMATION : "
+			    "ddi_prop_look_int_array failed"));
+			rval = DDI_FAILURE;
+		} else {
+
+			pi.busNumber = (props[0] >> 16) & 0xFF;
+			pi.deviceNumber = (props[0] >> 11) & 0x1f;
+			pi.functionNumber = (props[0] >> 8) & 0x7;
+			ddi_prop_free((void *)props);
+		}
+
+		pci_conf_buf = (uint8_t *)&pi.pciHeaderInfo;
+
+		for (i = 0; i < (sizeof (struct drsas_pci_information) -
+		    offsetof(struct drsas_pci_information, pciHeaderInfo));
+		    i++) {
+			pci_conf_buf[i] =
+			    pci_config_get8(instance->pci_handle, i);
+		}
+		for (i = 0; i < xferlen; i++) {
+			if (ddi_copyout((uint8_t *)&pi + i, (uint8_t *)ubuf + i,
+			    1, mode)) {
+				con_log(CL_ANN, (CE_WARN, "handle_drv_ioctl: "
+				    "DRSAS_DRIVER_IOCTL_PCI_INFORMATION"
+				    " : copy to user space failed"));
+				kdcmd->cmd_status = 1;
+				rval = DDI_FAILURE;
+				break;
+			}
+		}
+
+		if (i == xferlen)
+			kdcmd->cmd_status = 0;
+
+		break;
+	default:
+		con_log(CL_ANN, (CE_WARN, "handle_drv_ioctl: "
+		    "invalid driver specific IOCTL opcode = 0x%x",
+		    kdcmd->opcode));
+		kdcmd->cmd_status = 1;
+		rval = DDI_FAILURE;
+		break;
+	}
+
+	return (rval);
+}
+
+/*
+ * handle_mfi_ioctl
+ */
+static int
+handle_mfi_ioctl(struct drsas_instance *instance, struct drsas_ioctl *ioctl,
+    int mode)
+{
+	int	rval = DDI_SUCCESS;
+
+	struct drsas_header	*hdr;
+	struct drsas_cmd	*cmd;
+
+	cmd = get_mfi_pkt(instance);
+
+	if (!cmd) {
+		con_log(CL_ANN, (CE_WARN, "dr_sas: "
+		    "failed to get a cmd packet"));
+		return (DDI_FAILURE);
+	}
+
+	/* Clear the frame buffer and assign back the context id */
+	(void) memset((char *)&cmd->frame[0], 0, sizeof (union drsas_frame));
+	ddi_put32(cmd->frame_dma_obj.acc_handle, &cmd->frame->hdr.context,
+	    cmd->index);
+
+	hdr = (struct drsas_header *)&ioctl->frame[0];
+
+	switch (hdr->cmd) {
+	case MFI_CMD_OP_DCMD:
+		rval = issue_mfi_dcmd(instance, ioctl, cmd, mode);
+		break;
+	case MFI_CMD_OP_SMP:
+		rval = issue_mfi_smp(instance, ioctl, cmd, mode);
+		break;
+	case MFI_CMD_OP_STP:
+		rval = issue_mfi_stp(instance, ioctl, cmd, mode);
+		break;
+	case MFI_CMD_OP_LD_SCSI:
+	case MFI_CMD_OP_PD_SCSI:
+		rval = issue_mfi_pthru(instance, ioctl, cmd, mode);
+		break;
+	default:
+		con_log(CL_ANN, (CE_WARN, "handle_mfi_ioctl: "
+		    "invalid mfi ioctl hdr->cmd = %d", hdr->cmd));
+		rval = DDI_FAILURE;
+		break;
+	}
+
+
+	return_mfi_pkt(instance, cmd);
+	if (drsas_common_check(instance, cmd) != DDI_SUCCESS)
+		rval = DDI_FAILURE;
+	return (rval);
+}
+
+/*
+ * AEN
+ */
+static int
+handle_mfi_aen(struct drsas_instance *instance, struct drsas_aen *aen)
+{
+	int	rval = 0;
+
+	rval = register_mfi_aen(instance, instance->aen_seq_num,
+	    aen->class_locale_word);
+
+	aen->cmd_status = (uint8_t)rval;
+
+	return (rval);
+}
+
+static int
+register_mfi_aen(struct drsas_instance *instance, uint32_t seq_num,
+    uint32_t class_locale_word)
+{
+	int	ret_val;
+
+	struct drsas_cmd	*cmd, *aen_cmd;
+	struct drsas_dcmd_frame	*dcmd;
+	union drsas_evt_class_locale	curr_aen;
+	union drsas_evt_class_locale	prev_aen;
+
+	/*
+	 * If there an AEN pending already (aen_cmd), check if the
+	 * class_locale of that pending AEN is inclusive of the new
+	 * AEN request we currently have. If it is, then we don't have
+	 * to do anything. In other words, whichever events the current
+	 * AEN request is subscribing to, have already been subscribed
+	 * to.
+	 *
+	 * If the old_cmd is _not_ inclusive, then we have to abort
+	 * that command, form a class_locale that is superset of both
+	 * old and current and re-issue to the FW
+	 */
+
+	curr_aen.word = class_locale_word;
+	aen_cmd = instance->aen_cmd;
+	if (aen_cmd) {
+		prev_aen.word = ddi_get32(aen_cmd->frame_dma_obj.acc_handle,
+		    &aen_cmd->frame->dcmd.mbox.w[1]);
+
+		/*
+		 * A class whose enum value is smaller is inclusive of all
+		 * higher values. If a PROGRESS (= -1) was previously
+		 * registered, then a new registration requests for higher
+		 * classes need not be sent to FW. They are automatically
+		 * included.
+		 *
+		 * Locale numbers don't have such hierarchy. They are bitmap
+		 * values
+		 */
+		if ((prev_aen.members.class <= curr_aen.members.class) &&
+		    !((prev_aen.members.locale & curr_aen.members.locale) ^
+		    curr_aen.members.locale)) {
+			/*
+			 * Previously issued event registration includes
+			 * current request. Nothing to do.
+			 */
+
+			return (0);
+		} else {
+			curr_aen.members.locale |= prev_aen.members.locale;
+
+			if (prev_aen.members.class < curr_aen.members.class)
+				curr_aen.members.class = prev_aen.members.class;
+
+			ret_val = abort_aen_cmd(instance, aen_cmd);
+
+			if (ret_val) {
+				con_log(CL_ANN, (CE_WARN, "register_mfi_aen: "
+				    "failed to abort prevous AEN command"));
+
+				return (ret_val);
+			}
+		}
+	} else {
+		curr_aen.word = class_locale_word;
+	}
+
+	cmd = get_mfi_pkt(instance);
+
+	if (!cmd)
+		return (ENOMEM);
+	/* Clear the frame buffer and assign back the context id */
+	(void) memset((char *)&cmd->frame[0], 0, sizeof (union drsas_frame));
+	ddi_put32(cmd->frame_dma_obj.acc_handle, &cmd->frame->hdr.context,
+	    cmd->index);
+
+	dcmd = &cmd->frame->dcmd;
+
+	/* for(i = 0; i < DCMD_MBOX_SZ; i++) dcmd->mbox.b[i] = 0; */
+	(void) memset(dcmd->mbox.b, 0, DCMD_MBOX_SZ);
+
+	(void) memset(instance->mfi_evt_detail_obj.buffer, 0,
+	    sizeof (struct drsas_evt_detail));
+
+	/* Prepare DCMD for aen registration */
+	ddi_put8(cmd->frame_dma_obj.acc_handle, &dcmd->cmd, MFI_CMD_OP_DCMD);
+	ddi_put8(cmd->frame_dma_obj.acc_handle, &dcmd->cmd_status, 0x0);
+	ddi_put8(cmd->frame_dma_obj.acc_handle, &dcmd->sge_count, 1);
+	ddi_put16(cmd->frame_dma_obj.acc_handle, &dcmd->flags,
+	    MFI_FRAME_DIR_READ);
+	ddi_put16(cmd->frame_dma_obj.acc_handle, &dcmd->timeout, 0);
+	ddi_put32(cmd->frame_dma_obj.acc_handle, &dcmd->data_xfer_len,
+	    sizeof (struct drsas_evt_detail));
+	ddi_put32(cmd->frame_dma_obj.acc_handle, &dcmd->opcode,
+	    DR_DCMD_CTRL_EVENT_WAIT);
+	ddi_put32(cmd->frame_dma_obj.acc_handle, &dcmd->mbox.w[0], seq_num);
+	ddi_put32(cmd->frame_dma_obj.acc_handle, &dcmd->mbox.w[1],
+	    curr_aen.word);
+	ddi_put32(cmd->frame_dma_obj.acc_handle, &dcmd->sgl.sge32[0].phys_addr,
+	    instance->mfi_evt_detail_obj.dma_cookie[0].dmac_address);
+	ddi_put32(cmd->frame_dma_obj.acc_handle, &dcmd->sgl.sge32[0].length,
+	    sizeof (struct drsas_evt_detail));
+
+	instance->aen_seq_num = seq_num;
+
+
+	/*
+	 * Store reference to the cmd used to register for AEN. When an
+	 * application wants us to register for AEN, we have to abort this
+	 * cmd and re-register with a new EVENT LOCALE supplied by that app
+	 */
+	instance->aen_cmd = cmd;
+
+	cmd->frame_count = 1;
+
+	/* Issue the aen registration frame */
+	/* atomic_add_16 (&instance->fw_outstanding, 1); */
+	instance->func_ptr->issue_cmd(cmd, instance);
+
+	return (0);
+}
+
+static void
+display_scsi_inquiry(caddr_t scsi_inq)
+{
+#define	MAX_SCSI_DEVICE_CODE	14
+	int		i;
+	char		inquiry_buf[256] = {0};
+	int		len;
+	const char	*const scsi_device_types[] = {
+		"Direct-Access    ",
+		"Sequential-Access",
+		"Printer          ",
+		"Processor        ",
+		"WORM             ",
+		"CD-ROM           ",
+		"Scanner          ",
+		"Optical Device   ",
+		"Medium Changer   ",
+		"Communications   ",
+		"Unknown          ",
+		"Unknown          ",
+		"Unknown          ",
+		"Enclosure        ",
+	};
+
+	len = 0;
+
+	len += snprintf(inquiry_buf + len, 265 - len, "  Vendor: ");
+	for (i = 8; i < 16; i++) {
+		len += snprintf(inquiry_buf + len, 265 - len, "%c",
+		    scsi_inq[i]);
+	}
+
+	len += snprintf(inquiry_buf + len, 265 - len, "  Model: ");
+
+	for (i = 16; i < 32; i++) {
+		len += snprintf(inquiry_buf + len, 265 - len, "%c",
+		    scsi_inq[i]);
+	}
+
+	len += snprintf(inquiry_buf + len, 265 - len, "  Rev: ");
+
+	for (i = 32; i < 36; i++) {
+		len += snprintf(inquiry_buf + len, 265 - len, "%c",
+		    scsi_inq[i]);
+	}
+
+	len += snprintf(inquiry_buf + len, 265 - len, "\n");
+
+
+	i = scsi_inq[0] & 0x1f;
+
+
+	len += snprintf(inquiry_buf + len, 265 - len, "  Type:   %s ",
+	    i < MAX_SCSI_DEVICE_CODE ? scsi_device_types[i] :
+	    "Unknown          ");
+
+
+	len += snprintf(inquiry_buf + len, 265 - len,
+	    "                 ANSI SCSI revision: %02x", scsi_inq[2] & 0x07);
+
+	if ((scsi_inq[2] & 0x07) == 1 && (scsi_inq[3] & 0x0f) == 1) {
+		len += snprintf(inquiry_buf + len, 265 - len, " CCS\n");
+	} else {
+		len += snprintf(inquiry_buf + len, 265 - len, "\n");
+	}
+
+	con_log(CL_ANN1, (CE_CONT, inquiry_buf));
+}
+
+static int
+read_fw_status_reg_ppc(struct drsas_instance *instance)
+{
+	return ((int)RD_OB_SCRATCH_PAD_0(instance));
+}
+
+static void
+issue_cmd_ppc(struct drsas_cmd *cmd, struct drsas_instance *instance)
+{
+	atomic_add_16(&instance->fw_outstanding, 1);
+
+	/* Issue the command to the FW */
+	WR_IB_QPORT((cmd->frame_phys_addr) |
+	    (((cmd->frame_count - 1) << 1) | 1), instance);
+}
+
+/*
+ * issue_cmd_in_sync_mode
+ */
+static int
+issue_cmd_in_sync_mode_ppc(struct drsas_instance *instance,
+    struct drsas_cmd *cmd)
+{
+	int		i;
+	uint32_t	msecs = MFI_POLL_TIMEOUT_SECS * (10 * MILLISEC);
+
+	con_log(CL_ANN1, (CE_NOTE, "issue_cmd_in_sync_mode_ppc: called"));
+
+	cmd->cmd_status	= ENODATA;
+
+	WR_IB_QPORT((cmd->frame_phys_addr) |
+	    (((cmd->frame_count - 1) << 1) | 1), instance);
+
+	mutex_enter(&instance->int_cmd_mtx);
+
+	for (i = 0; i < msecs && (cmd->cmd_status == ENODATA); i++) {
+		cv_wait(&instance->int_cmd_cv, &instance->int_cmd_mtx);
+	}
+
+	mutex_exit(&instance->int_cmd_mtx);
+
+	con_log(CL_ANN1, (CE_NOTE, "issue_cmd_in_sync_mode_ppc: done"));
+
+	if (i < (msecs -1)) {
+		return (DDI_SUCCESS);
+	} else {
+		return (DDI_FAILURE);
+	}
+}
+
+/*
+ * issue_cmd_in_poll_mode
+ */
+static int
+issue_cmd_in_poll_mode_ppc(struct drsas_instance *instance,
+    struct drsas_cmd *cmd)
+{
+	int		i;
+	uint16_t	flags;
+	uint32_t	msecs = MFI_POLL_TIMEOUT_SECS * MILLISEC;
+	struct drsas_header *frame_hdr;
+
+	con_log(CL_ANN1, (CE_NOTE, "issue_cmd_in_poll_mode_ppc: called"));
+
+	frame_hdr = (struct drsas_header *)cmd->frame;
+	ddi_put8(cmd->frame_dma_obj.acc_handle, &frame_hdr->cmd_status,
+	    MFI_CMD_STATUS_POLL_MODE);
+	flags = ddi_get16(cmd->frame_dma_obj.acc_handle, &frame_hdr->flags);
+	flags 	|= MFI_FRAME_DONT_POST_IN_REPLY_QUEUE;
+
+	ddi_put16(cmd->frame_dma_obj.acc_handle, &frame_hdr->flags, flags);
+
+	/* issue the frame using inbound queue port */
+	WR_IB_QPORT((cmd->frame_phys_addr) |
+	    (((cmd->frame_count - 1) << 1) | 1), instance);
+
+	/* wait for cmd_status to change from 0xFF */
+	for (i = 0; i < msecs && (
+	    ddi_get8(cmd->frame_dma_obj.acc_handle, &frame_hdr->cmd_status)
+	    == MFI_CMD_STATUS_POLL_MODE); i++) {
+		drv_usecwait(MILLISEC); /* wait for 1000 usecs */
+	}
+
+	if (ddi_get8(cmd->frame_dma_obj.acc_handle, &frame_hdr->cmd_status)
+	    == MFI_CMD_STATUS_POLL_MODE) {
+		con_log(CL_ANN, (CE_NOTE, "issue_cmd_in_poll_mode: "
+		    "cmd polling timed out"));
+		return (DDI_FAILURE);
+	}
+
+	return (DDI_SUCCESS);
+}
+
+static void
+enable_intr_ppc(struct drsas_instance *instance)
+{
+	uint32_t	mask;
+
+	con_log(CL_ANN1, (CE_NOTE, "enable_intr_ppc: called"));
+
+	/* WR_OB_DOORBELL_CLEAR(0xFFFFFFFF, instance); */
+	WR_OB_DOORBELL_CLEAR(OB_DOORBELL_CLEAR_MASK, instance);
+
+	/* WR_OB_INTR_MASK(~0x80000000, instance); */
+	WR_OB_INTR_MASK(~(MFI_REPLY_2108_MESSAGE_INTR_MASK), instance);
+
+	/* dummy read to force PCI flush */
+	mask = RD_OB_INTR_MASK(instance);
+
+	con_log(CL_ANN1, (CE_NOTE, "enable_intr_ppc: "
+	    "outbound_intr_mask = 0x%x", mask));
+}
+
+static void
+disable_intr_ppc(struct drsas_instance *instance)
+{
+	uint32_t	mask;
+
+	con_log(CL_ANN1, (CE_NOTE, "disable_intr_ppc: called"));
+
+	con_log(CL_ANN1, (CE_NOTE, "disable_intr_ppc: before : "
+	    "outbound_intr_mask = 0x%x", RD_OB_INTR_MASK(instance)));
+
+	/* WR_OB_INTR_MASK(0xFFFFFFFF, instance); */
+	WR_OB_INTR_MASK(OB_INTR_MASK, instance);
+
+	con_log(CL_ANN1, (CE_NOTE, "disable_intr_ppc: after : "
+	    "outbound_intr_mask = 0x%x", RD_OB_INTR_MASK(instance)));
+
+	/* dummy read to force PCI flush */
+	mask = RD_OB_INTR_MASK(instance);
+#ifdef lint
+	mask = mask;
+#endif
+}
+
+static int
+intr_ack_ppc(struct drsas_instance *instance)
+{
+	uint32_t	status;
+
+	con_log(CL_ANN1, (CE_NOTE, "intr_ack_ppc: called"));
+
+	/* check if it is our interrupt */
+	status = RD_OB_INTR_STATUS(instance);
+
+	con_log(CL_ANN1, (CE_NOTE, "intr_ack_ppc: status = 0x%x", status));
+
+	if (!(status & MFI_REPLY_2108_MESSAGE_INTR)) {
+		return (DDI_INTR_UNCLAIMED);
+	}
+
+	/* clear the interrupt by writing back the same value */
+	WR_OB_DOORBELL_CLEAR(status, instance);
+
+	/* dummy READ */
+	status = RD_OB_INTR_STATUS(instance);
+
+	con_log(CL_ANN1, (CE_NOTE, "intr_ack_ppc: interrupt cleared"));
+
+	return (DDI_INTR_CLAIMED);
+}
+
+static int
+drsas_common_check(struct drsas_instance *instance,
+    struct  drsas_cmd *cmd)
+{
+	int ret = DDI_SUCCESS;
+
+	if (drsas_check_dma_handle(cmd->frame_dma_obj.dma_handle) !=
+	    DDI_SUCCESS) {
+		ddi_fm_service_impact(instance->dip, DDI_SERVICE_UNAFFECTED);
+		if (cmd->pkt != NULL) {
+			cmd->pkt->pkt_reason = CMD_TRAN_ERR;
+			cmd->pkt->pkt_statistics = 0;
+		}
+		ret = DDI_FAILURE;
+	}
+	if (drsas_check_dma_handle(instance->mfi_internal_dma_obj.dma_handle)
+	    != DDI_SUCCESS) {
+		ddi_fm_service_impact(instance->dip, DDI_SERVICE_UNAFFECTED);
+		if (cmd->pkt != NULL) {
+			cmd->pkt->pkt_reason = CMD_TRAN_ERR;
+			cmd->pkt->pkt_statistics = 0;
+		}
+		ret = DDI_FAILURE;
+	}
+	if (drsas_check_dma_handle(instance->mfi_evt_detail_obj.dma_handle) !=
+	    DDI_SUCCESS) {
+		ddi_fm_service_impact(instance->dip, DDI_SERVICE_UNAFFECTED);
+		if (cmd->pkt != NULL) {
+			cmd->pkt->pkt_reason = CMD_TRAN_ERR;
+			cmd->pkt->pkt_statistics = 0;
+		}
+		ret = DDI_FAILURE;
+	}
+	if (drsas_check_acc_handle(instance->regmap_handle) != DDI_SUCCESS) {
+		ddi_fm_service_impact(instance->dip, DDI_SERVICE_UNAFFECTED);
+
+		ddi_fm_acc_err_clear(instance->regmap_handle, DDI_FME_VER0);
+
+		if (cmd->pkt != NULL) {
+			cmd->pkt->pkt_reason = CMD_TRAN_ERR;
+			cmd->pkt->pkt_statistics = 0;
+		}
+		ret = DDI_FAILURE;
+	}
+
+	return (ret);
+}
+
+/*ARGSUSED*/
+static int
+drsas_fm_error_cb(dev_info_t *dip, ddi_fm_error_t *err, const void *impl_data)
+{
+	/*
+	 * as the driver can always deal with an error in any dma or
+	 * access handle, we can just return the fme_status value.
+	 */
+	pci_ereport_post(dip, err, NULL);
+	return (err->fme_status);
+}
+
+static void
+drsas_fm_init(struct drsas_instance *instance)
+{
+	/* Need to change iblock to priority for new MSI intr */
+	ddi_iblock_cookie_t fm_ibc;
+
+	/* Only register with IO Fault Services if we have some capability */
+	if (instance->fm_capabilities) {
+		/* Adjust access and dma attributes for FMA */
+		endian_attr.devacc_attr_access = DDI_FLAGERR_ACC;
+		drsas_generic_dma_attr.dma_attr_flags = DDI_DMA_FLAGERR;
+
+		/*
+		 * Register capabilities with IO Fault Services.
+		 * fm_capabilities will be updated to indicate
+		 * capabilities actually supported (not requested.)
+		 */
+
+		ddi_fm_init(instance->dip, &instance->fm_capabilities, &fm_ibc);
+
+		/*
+		 * Initialize pci ereport capabilities if ereport
+		 * capable (should always be.)
+		 */
+
+		if (DDI_FM_EREPORT_CAP(instance->fm_capabilities) ||
+		    DDI_FM_ERRCB_CAP(instance->fm_capabilities)) {
+			pci_ereport_setup(instance->dip);
+		}
+
+		/*
+		 * Register error callback if error callback capable.
+		 */
+		if (DDI_FM_ERRCB_CAP(instance->fm_capabilities)) {
+			ddi_fm_handler_register(instance->dip,
+			    drsas_fm_error_cb, (void*) instance);
+		}
+	} else {
+		endian_attr.devacc_attr_access = DDI_DEFAULT_ACC;
+		drsas_generic_dma_attr.dma_attr_flags = 0;
+	}
+}
+
+static void
+drsas_fm_fini(struct drsas_instance *instance)
+{
+	/* Only unregister FMA capabilities if registered */
+	if (instance->fm_capabilities) {
+		/*
+		 * Un-register error callback if error callback capable.
+		 */
+		if (DDI_FM_ERRCB_CAP(instance->fm_capabilities)) {
+			ddi_fm_handler_unregister(instance->dip);
+		}
+
+		/*
+		 * Release any resources allocated by pci_ereport_setup()
+		 */
+		if (DDI_FM_EREPORT_CAP(instance->fm_capabilities) ||
+		    DDI_FM_ERRCB_CAP(instance->fm_capabilities)) {
+			pci_ereport_teardown(instance->dip);
+		}
+
+		/* Unregister from IO Fault Services */
+		ddi_fm_fini(instance->dip);
+
+		/* Adjust access and dma attributes for FMA */
+		endian_attr.devacc_attr_access = DDI_DEFAULT_ACC;
+		drsas_generic_dma_attr.dma_attr_flags = 0;
+	}
+}
+
+int
+drsas_check_acc_handle(ddi_acc_handle_t handle)
+{
+	ddi_fm_error_t de;
+
+	if (handle == NULL) {
+		return (DDI_FAILURE);
+	}
+
+	ddi_fm_acc_err_get(handle, &de, DDI_FME_VERSION);
+
+	return (de.fme_status);
+}
+
+int
+drsas_check_dma_handle(ddi_dma_handle_t handle)
+{
+	ddi_fm_error_t de;
+
+	if (handle == NULL) {
+		return (DDI_FAILURE);
+	}
+
+	ddi_fm_dma_err_get(handle, &de, DDI_FME_VERSION);
+
+	return (de.fme_status);
+}
+
+void
+drsas_fm_ereport(struct drsas_instance *instance, char *detail)
+{
+	uint64_t ena;
+	char buf[FM_MAX_CLASS];
+
+	(void) snprintf(buf, FM_MAX_CLASS, "%s.%s", DDI_FM_DEVICE, detail);
+	ena = fm_ena_generate(0, FM_ENA_FMT1);
+	if (DDI_FM_EREPORT_CAP(instance->fm_capabilities)) {
+		ddi_fm_ereport_post(instance->dip, buf, ena, DDI_NOSLEEP,
+		    FM_VERSION, DATA_TYPE_UINT8, FM_EREPORT_VERSION, NULL);
+	}
+}
+
+static int
+drsas_add_intrs(struct drsas_instance *instance, int intr_type)
+{
+
+	dev_info_t *dip = instance->dip;
+	int	avail, actual, count;
+	int	i, flag, ret;
+
+	con_log(CL_DLEVEL1, (CE_WARN, "drsas_add_intrs: intr_type = %x",
+	    intr_type));
+
+	/* Get number of interrupts */
+	ret = ddi_intr_get_nintrs(dip, intr_type, &count);
+	if ((ret != DDI_SUCCESS) || (count == 0)) {
+		con_log(CL_ANN, (CE_WARN, "ddi_intr_get_nintrs() failed:"
+		    "ret %d count %d", ret, count));
+
+		return (DDI_FAILURE);
+	}
+
+	con_log(CL_DLEVEL1, (CE_WARN, "drsas_add_intrs: count = %d ", count));
+
+	/* Get number of available interrupts */
+	ret = ddi_intr_get_navail(dip, intr_type, &avail);
+	if ((ret != DDI_SUCCESS) || (avail == 0)) {
+		con_log(CL_ANN, (CE_WARN, "ddi_intr_get_navail() failed:"
+		    "ret %d avail %d", ret, avail));
+
+		return (DDI_FAILURE);
+	}
+	con_log(CL_DLEVEL1, (CE_WARN, "drsas_add_intrs: avail = %d ", avail));
+
+	/* Only one interrupt routine. So limit the count to 1 */
+	if (count > 1) {
+		count = 1;
+	}
+
+	/*
+	 * Allocate an array of interrupt handlers. Currently we support
+	 * only one interrupt. The framework can be extended later.
+	 */
+	instance->intr_size = count * sizeof (ddi_intr_handle_t);
+	instance->intr_htable = kmem_zalloc(instance->intr_size, KM_SLEEP);
+	ASSERT(instance->intr_htable);
+
+	flag = ((intr_type == DDI_INTR_TYPE_MSI) || (intr_type ==
+	    DDI_INTR_TYPE_MSIX)) ? DDI_INTR_ALLOC_STRICT:DDI_INTR_ALLOC_NORMAL;
+
+	/* Allocate interrupt */
+	ret = ddi_intr_alloc(dip, instance->intr_htable, intr_type, 0,
+	    count, &actual, flag);
+
+	if ((ret != DDI_SUCCESS) || (actual == 0)) {
+		con_log(CL_ANN, (CE_WARN, "drsas_add_intrs: "
+		    "avail = %d", avail));
+		kmem_free(instance->intr_htable, instance->intr_size);
+		return (DDI_FAILURE);
+	}
+	if (actual < count) {
+		con_log(CL_ANN, (CE_WARN, "drsas_add_intrs: "
+		    "Requested = %d  Received = %d", count, actual));
+	}
+	instance->intr_cnt = actual;
+
+	/*
+	 * Get the priority of the interrupt allocated.
+	 */
+	if ((ret = ddi_intr_get_pri(instance->intr_htable[0],
+	    &instance->intr_pri)) != DDI_SUCCESS) {
+		con_log(CL_ANN, (CE_WARN, "drsas_add_intrs: "
+		    "get priority call failed"));
+
+		for (i = 0; i < actual; i++) {
+			(void) ddi_intr_free(instance->intr_htable[i]);
+		}
+		kmem_free(instance->intr_htable, instance->intr_size);
+		return (DDI_FAILURE);
+	}
+
+	/*
+	 * Test for high level mutex. we don't support them.
+	 */
+	if (instance->intr_pri >= ddi_intr_get_hilevel_pri()) {
+		con_log(CL_ANN, (CE_WARN, "drsas_add_intrs: "
+		    "High level interrupts not supported."));
+
+		for (i = 0; i < actual; i++) {
+			(void) ddi_intr_free(instance->intr_htable[i]);
+		}
+		kmem_free(instance->intr_htable, instance->intr_size);
+		return (DDI_FAILURE);
+	}
+
+	con_log(CL_DLEVEL1, (CE_NOTE, "drsas_add_intrs: intr_pri = 0x%x ",
+	    instance->intr_pri));
+
+	/* Call ddi_intr_add_handler() */
+	for (i = 0; i < actual; i++) {
+		ret = ddi_intr_add_handler(instance->intr_htable[i],
+		    (ddi_intr_handler_t *)drsas_isr, (caddr_t)instance,
+		    (caddr_t)(uintptr_t)i);
+
+		if (ret != DDI_SUCCESS) {
+			con_log(CL_ANN, (CE_WARN, "drsas_add_intrs:"
+			    "failed %d", ret));
+
+			for (i = 0; i < actual; i++) {
+				(void) ddi_intr_free(instance->intr_htable[i]);
+			}
+			kmem_free(instance->intr_htable, instance->intr_size);
+			return (DDI_FAILURE);
+		}
+
+	}
+
+	con_log(CL_DLEVEL1, (CE_WARN, " ddi_intr_add_handler done"));
+
+	if ((ret = ddi_intr_get_cap(instance->intr_htable[0],
+	    &instance->intr_cap)) != DDI_SUCCESS) {
+		con_log(CL_ANN, (CE_WARN, "ddi_intr_get_cap() failed %d",
+		    ret));
+
+		/* Free already allocated intr */
+		for (i = 0; i < actual; i++) {
+			(void) ddi_intr_remove_handler(
+			    instance->intr_htable[i]);
+			(void) ddi_intr_free(instance->intr_htable[i]);
+		}
+		kmem_free(instance->intr_htable, instance->intr_size);
+		return (DDI_FAILURE);
+	}
+
+	if (instance->intr_cap &  DDI_INTR_FLAG_BLOCK) {
+		con_log(CL_ANN, (CE_WARN, "Calling ddi_intr_block _enable"));
+
+		(void) ddi_intr_block_enable(instance->intr_htable,
+		    instance->intr_cnt);
+	} else {
+		con_log(CL_ANN, (CE_NOTE, " calling ddi_intr_enable"));
+
+		for (i = 0; i < instance->intr_cnt; i++) {
+			(void) ddi_intr_enable(instance->intr_htable[i]);
+			con_log(CL_ANN, (CE_NOTE, "ddi intr enable returns "
+			    "%d", i));
+		}
+	}
+
+	return (DDI_SUCCESS);
+
+}
+
+
+static void
+drsas_rem_intrs(struct drsas_instance *instance)
+{
+	int i;
+
+	con_log(CL_ANN, (CE_NOTE, "drsas_rem_intrs called"));
+
+	/* Disable all interrupts first */
+	if (instance->intr_cap & DDI_INTR_FLAG_BLOCK) {
+		(void) ddi_intr_block_disable(instance->intr_htable,
+		    instance->intr_cnt);
+	} else {
+		for (i = 0; i < instance->intr_cnt; i++) {
+			(void) ddi_intr_disable(instance->intr_htable[i]);
+		}
+	}
+
+	/* Remove all the handlers */
+
+	for (i = 0; i < instance->intr_cnt; i++) {
+		(void) ddi_intr_remove_handler(instance->intr_htable[i]);
+		(void) ddi_intr_free(instance->intr_htable[i]);
+	}
+
+	kmem_free(instance->intr_htable, instance->intr_size);
+}
+
+static int
+drsas_tran_bus_config(dev_info_t *parent, uint_t flags,
+    ddi_bus_config_op_t op, void *arg, dev_info_t **childp)
+{
+	struct drsas_instance *instance;
+	int config;
+	int rval;
+
+	char *ptr = NULL;
+	int tgt, lun;
+
+	con_log(CL_ANN1, (CE_NOTE, "Bus config called for op = %x", op));
+
+	if ((instance = ddi_get_soft_state(drsas_state,
+	    ddi_get_instance(parent))) == NULL) {
+		return (NDI_FAILURE);
+	}
+
+	/* Hold nexus during bus_config */
+	ndi_devi_enter(parent, &config);
+	switch (op) {
+	case BUS_CONFIG_ONE: {
+
+		/* parse wwid/target name out of name given */
+		if ((ptr = strchr((char *)arg, '@')) == NULL) {
+			rval = NDI_FAILURE;
+			break;
+		}
+		ptr++;
+
+		if (drsas_parse_devname(arg, &tgt, &lun) != 0) {
+			rval = NDI_FAILURE;
+			break;
+		}
+
+		if (lun == 0) {
+			rval = drsas_config_ld(instance, tgt, lun, childp);
+		} else {
+			rval = NDI_FAILURE;
+		}
+
+		break;
+	}
+	case BUS_CONFIG_DRIVER:
+	case BUS_CONFIG_ALL: {
+
+		rval = drsas_config_all_devices(instance);
+
+		rval = NDI_SUCCESS;
+		break;
+	}
+	}
+
+	if (rval == NDI_SUCCESS) {
+		rval = ndi_busop_bus_config(parent, flags, op, arg, childp, 0);
+
+	}
+	ndi_devi_exit(parent, config);
+
+	con_log(CL_ANN1, (CE_NOTE, "drsas_tran_bus_config: rval = %x",
+	    rval));
+	return (rval);
+}
+
+static int
+drsas_config_all_devices(struct drsas_instance *instance)
+{
+	int rval, tgt;
+
+	for (tgt = 0; tgt < MRDRV_MAX_LD; tgt++) {
+		(void) drsas_config_ld(instance, tgt, 0, NULL);
+
+	}
+
+	rval = NDI_SUCCESS;
+	return (rval);
+}
+
+static int
+drsas_parse_devname(char *devnm, int *tgt, int *lun)
+{
+	char devbuf[SCSI_MAXNAMELEN];
+	char *addr;
+	char *p,  *tp, *lp;
+	long num;
+
+	/* Parse dev name and address */
+	(void) strcpy(devbuf, devnm);
+	addr = "";
+	for (p = devbuf; *p != '\0'; p++) {
+		if (*p == '@') {
+			addr = p + 1;
+			*p = '\0';
+		} else if (*p == ':') {
+			*p = '\0';
+			break;
+		}
+	}
+
+	/* Parse target and lun */
+	for (p = tp = addr, lp = NULL; *p != '\0'; p++) {
+		if (*p == ',') {
+			lp = p + 1;
+			*p = '\0';
+			break;
+		}
+	}
+	if (tgt && tp) {
+		if (ddi_strtol(tp, NULL, 0x10, &num)) {
+			return (DDI_FAILURE); /* Can declare this as constant */
+		}
+			*tgt = (int)num;
+	}
+	if (lun && lp) {
+		if (ddi_strtol(lp, NULL, 0x10, &num)) {
+			return (DDI_FAILURE);
+		}
+			*lun = (int)num;
+	}
+	return (DDI_SUCCESS);  /* Success case */
+}
+
+static int
+drsas_config_ld(struct drsas_instance *instance, uint16_t tgt,
+    uint8_t lun, dev_info_t **ldip)
+{
+	struct scsi_device *sd;
+	dev_info_t *child;
+	int rval;
+
+	con_log(CL_ANN1, (CE_NOTE, "drsas_config_ld: t = %d l = %d",
+	    tgt, lun));
+
+	if ((child = drsas_find_child(instance, tgt, lun)) != NULL) {
+		if (ldip) {
+			*ldip = child;
+		}
+		con_log(CL_ANN1, (CE_NOTE,
+		    "drsas_config_ld: Child = %p found t = %d l = %d",
+		    (void *)child, tgt, lun));
+		return (NDI_SUCCESS);
+	}
+
+	sd = kmem_zalloc(sizeof (struct scsi_device), KM_SLEEP);
+	sd->sd_address.a_hba_tran = instance->tran;
+	sd->sd_address.a_target = (uint16_t)tgt;
+	sd->sd_address.a_lun = (uint8_t)lun;
+
+	if (scsi_hba_probe(sd, NULL) == SCSIPROBE_EXISTS)
+		rval = drsas_config_scsi_device(instance, sd, ldip);
+	else
+		rval = NDI_FAILURE;
+
+	/* sd_unprobe is blank now. Free buffer manually */
+	if (sd->sd_inq) {
+		kmem_free(sd->sd_inq, SUN_INQSIZE);
+		sd->sd_inq = (struct scsi_inquiry *)NULL;
+	}
+
+	kmem_free(sd, sizeof (struct scsi_device));
+	con_log(CL_ANN1, (CE_NOTE, "drsas_config_ld: return rval = %d",
+	    rval));
+	return (rval);
+}
+
+static int
+drsas_config_scsi_device(struct drsas_instance *instance,
+    struct scsi_device *sd, dev_info_t **dipp)
+{
+	char *nodename = NULL;
+	char **compatible = NULL;
+	int ncompatible = 0;
+	char *childname;
+	dev_info_t *ldip = NULL;
+	int tgt = sd->sd_address.a_target;
+	int lun = sd->sd_address.a_lun;
+	int dtype = sd->sd_inq->inq_dtype & DTYPE_MASK;
+	int rval;
+
+	con_log(CL_ANN1, (CE_WARN, "dr_sas: scsi_device t%dL%d", tgt, lun));
+	scsi_hba_nodename_compatible_get(sd->sd_inq, NULL, dtype,
+	    NULL, &nodename, &compatible, &ncompatible);
+
+	if (nodename == NULL) {
+		con_log(CL_ANN1, (CE_WARN, "dr_sas: Found no compatible driver "
+		    "for t%dL%d", tgt, lun));
+		rval = NDI_FAILURE;
+		goto finish;
+	}
+
+	childname = (dtype == DTYPE_DIRECT) ? "sd" : nodename;
+	con_log(CL_ANN1, (CE_WARN,
+	    "dr_sas: Childname = %2s nodename = %s", childname, nodename));
+
+	/* Create a dev node */
+	rval = ndi_devi_alloc(instance->dip, childname, DEVI_SID_NODEID, &ldip);
+	con_log(CL_ANN1, (CE_WARN,
+	    "dr_sas_config_scsi_device: ndi_devi_alloc rval = %x", rval));
+	if (rval == NDI_SUCCESS) {
+		if (ndi_prop_update_int(DDI_DEV_T_NONE, ldip, "target", tgt) !=
+		    DDI_PROP_SUCCESS) {
+			con_log(CL_ANN1, (CE_WARN, "dr_sas: unable to create "
+			    "property for t%dl%d target", tgt, lun));
+			rval = NDI_FAILURE;
+			goto finish;
+		}
+		if (ndi_prop_update_int(DDI_DEV_T_NONE, ldip, "lun", lun) !=
+		    DDI_PROP_SUCCESS) {
+			con_log(CL_ANN1, (CE_WARN, "dr_sas: unable to create "
+			    "property for t%dl%d lun", tgt, lun));
+			rval = NDI_FAILURE;
+			goto finish;
+		}
+
+		if (ndi_prop_update_string_array(DDI_DEV_T_NONE, ldip,
+		    "compatible", compatible, ncompatible) !=
+		    DDI_PROP_SUCCESS) {
+			con_log(CL_ANN1, (CE_WARN, "dr_sas: unable to create "
+			    "property for t%dl%d compatible", tgt, lun));
+			rval = NDI_FAILURE;
+			goto finish;
+		}
+
+		rval = ndi_devi_online(ldip, NDI_ONLINE_ATTACH);
+		if (rval != NDI_SUCCESS) {
+			con_log(CL_ANN1, (CE_WARN, "dr_sas: unable to online "
+			    "t%dl%d", tgt, lun));
+			ndi_prop_remove_all(ldip);
+			(void) ndi_devi_free(ldip);
+		} else {
+			con_log(CL_ANN1, (CE_WARN, "dr_sas: online Done :"
+			    "0 t%dl%d", tgt, lun));
+		}
+
+	}
+finish:
+	if (dipp) {
+		*dipp = ldip;
+	}
+
+	con_log(CL_DLEVEL1, (CE_WARN,
+	    "dr_sas: config_scsi_device rval = %d t%dL%d",
+	    rval, tgt, lun));
+	scsi_hba_nodename_compatible_free(nodename, compatible);
+	return (rval);
+}
+
+/*ARGSUSED*/
+static int
+drsas_service_evt(struct drsas_instance *instance, int tgt, int lun, int event,
+    uint64_t wwn)
+{
+	struct drsas_eventinfo *mrevt = NULL;
+
+	con_log(CL_ANN1, (CE_NOTE,
+	    "drsas_service_evt called for t%dl%d event = %d",
+	    tgt, lun, event));
+
+	if ((instance->taskq == NULL) || (mrevt =
+	    kmem_zalloc(sizeof (struct drsas_eventinfo), KM_NOSLEEP)) == NULL) {
+		return (ENOMEM);
+	}
+
+	mrevt->instance = instance;
+	mrevt->tgt = tgt;
+	mrevt->lun = lun;
+	mrevt->event = event;
+
+	if ((ddi_taskq_dispatch(instance->taskq,
+	    (void (*)(void *))drsas_issue_evt_taskq, mrevt, DDI_NOSLEEP)) !=
+	    DDI_SUCCESS) {
+		con_log(CL_ANN1, (CE_NOTE,
+		    "dr_sas: Event task failed for t%dl%d event = %d",
+		    tgt, lun, event));
+		kmem_free(mrevt, sizeof (struct drsas_eventinfo));
+		return (DDI_FAILURE);
+	}
+	return (DDI_SUCCESS);
+}
+
+static void
+drsas_issue_evt_taskq(struct drsas_eventinfo *mrevt)
+{
+	struct drsas_instance *instance = mrevt->instance;
+	dev_info_t *dip, *pdip;
+	int circ1 = 0;
+	char *devname;
+
+	con_log(CL_ANN1, (CE_NOTE, "drsas_issue_evt_taskq: called for"
+	    " tgt %d lun %d event %d",
+	    mrevt->tgt, mrevt->lun, mrevt->event));
+
+	if (mrevt->tgt < MRDRV_MAX_LD && mrevt->lun == 0) {
+		dip = instance->dr_ld_list[mrevt->tgt].dip;
+	} else {
+		return;
+	}
+
+	ndi_devi_enter(instance->dip, &circ1);
+	switch (mrevt->event) {
+	case DRSAS_EVT_CONFIG_TGT:
+		if (dip == NULL) {
+
+			if (mrevt->lun == 0) {
+				(void) drsas_config_ld(instance, mrevt->tgt,
+				    0, NULL);
+			}
+			con_log(CL_ANN1, (CE_NOTE,
+			    "dr_sas: EVT_CONFIG_TGT called:"
+			    " for tgt %d lun %d event %d",
+			    mrevt->tgt, mrevt->lun, mrevt->event));
+
+		} else {
+			con_log(CL_ANN1, (CE_NOTE,
+			    "dr_sas: EVT_CONFIG_TGT dip != NULL:"
+			    " for tgt %d lun %d event %d",
+			    mrevt->tgt, mrevt->lun, mrevt->event));
+		}
+		break;
+	case DRSAS_EVT_UNCONFIG_TGT:
+		if (dip) {
+			if (i_ddi_devi_attached(dip)) {
+
+				pdip = ddi_get_parent(dip);
+
+				devname = kmem_zalloc(MAXNAMELEN + 1, KM_SLEEP);
+				(void) ddi_deviname(dip, devname);
+
+				(void) devfs_clean(pdip, devname + 1,
+				    DV_CLEAN_FORCE);
+				kmem_free(devname, MAXNAMELEN + 1);
+			}
+			(void) ndi_devi_offline(dip, NDI_DEVI_REMOVE);
+			con_log(CL_ANN1, (CE_NOTE,
+			    "dr_sas: EVT_UNCONFIG_TGT called:"
+			    " for tgt %d lun %d event %d",
+			    mrevt->tgt, mrevt->lun, mrevt->event));
+		} else {
+			con_log(CL_ANN1, (CE_NOTE,
+			    "dr_sas: EVT_UNCONFIG_TGT dip == NULL:"
+			    " for tgt %d lun %d event %d",
+			    mrevt->tgt, mrevt->lun, mrevt->event));
+		}
+		break;
+	}
+	kmem_free(mrevt, sizeof (struct drsas_eventinfo));
+	ndi_devi_exit(instance->dip, circ1);
+}
+
+static int
+drsas_mode_sense_build(struct scsi_pkt *pkt)
+{
+	union scsi_cdb		*cdbp;
+	uint16_t 		page_code;
+	struct scsa_cmd		*acmd;
+	struct buf		*bp;
+	struct mode_header	*modehdrp;
+
+	cdbp = (void *)pkt->pkt_cdbp;
+	page_code = cdbp->cdb_un.sg.scsi[0];
+	acmd = PKT2CMD(pkt);
+	bp = acmd->cmd_buf;
+	if ((!bp) && bp->b_un.b_addr && bp->b_bcount && acmd->cmd_dmacount) {
+		con_log(CL_ANN1, (CE_WARN, "Failing MODESENSE Command"));
+		/* ADD pkt statistics as Command failed. */
+		return (NULL);
+	}
+
+	bp_mapin(bp);
+	bzero(bp->b_un.b_addr, bp->b_bcount);
+
+	switch (page_code) {
+		case 0x3: {
+			struct mode_format *page3p = NULL;
+			modehdrp = (struct mode_header *)(bp->b_un.b_addr);
+			modehdrp->bdesc_length = MODE_BLK_DESC_LENGTH;
+
+			page3p = (void *)((caddr_t)modehdrp +
+			    MODE_HEADER_LENGTH + MODE_BLK_DESC_LENGTH);
+			page3p->mode_page.code = 0x3;
+			page3p->mode_page.length =
+			    (uchar_t)(sizeof (struct mode_format));
+			page3p->data_bytes_sect = 512;
+			page3p->sect_track = 63;
+			break;
+		}
+		case 0x4: {
+			struct mode_geometry *page4p = NULL;
+			modehdrp = (struct mode_header *)(bp->b_un.b_addr);
+			modehdrp->bdesc_length = MODE_BLK_DESC_LENGTH;
+
+			page4p = (void *)((caddr_t)modehdrp +
+			    MODE_HEADER_LENGTH + MODE_BLK_DESC_LENGTH);
+			page4p->mode_page.code = 0x4;
+			page4p->mode_page.length =
+			    (uchar_t)(sizeof (struct mode_geometry));
+			page4p->heads = 255;
+			page4p->rpm = 10000;
+			break;
+		}
+		default:
+			break;
+	}
+	return (NULL);
+}
diff --git a/usr/src/uts/common/io/dr_sas/dr_sas.conf b/usr/src/uts/common/io/dr_sas/dr_sas.conf
new file mode 100644
index 0000000000..3792f43ca4
--- /dev/null
+++ b/usr/src/uts/common/io/dr_sas/dr_sas.conf
@@ -0,0 +1,15 @@
+#
+# Copyright (c) 2008-2009, LSI Logic Corporation.
+# All rights reserved.
+#
+# Copyright 2009 Sun Microsystems, Inc.  All rights reserved.
+# Use is subject to license terms.
+#
+
+#
+# dr_sas.conf for sol 10 (and later) for all supported architectures
+#
+# global definitions
+
+# MSI specific flag. user can uncomment this line and set flag "yes" to enable MSI
+#drsas-enable-msi="yes";
diff --git a/usr/src/uts/common/io/dr_sas/dr_sas.h b/usr/src/uts/common/io/dr_sas/dr_sas.h
new file mode 100644
index 0000000000..8f78658edf
--- /dev/null
+++ b/usr/src/uts/common/io/dr_sas/dr_sas.h
@@ -0,0 +1,1766 @@
+/*
+ * dr_sas.h: header for dr_sas
+ *
+ * Solaris MegaRAID driver for SAS2.0 controllers
+ * Copyright (c) 2008-2009, LSI Logic Corporation.
+ * All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions are met:
+ *
+ * 1. Redistributions of source code must retain the above copyright notice,
+ *    this list of conditions and the following disclaimer.
+ *
+ * 2. Redistributions in binary form must reproduce the above copyright notice,
+ *    this list of conditions and the following disclaimer in the documentation
+ *    and/or other materials provided with the distribution.
+ *
+ * 3. Neither the name of the author nor the names of its contributors may be
+ *    used to endorse or promote products derived from this software without
+ *    specific prior written permission.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
+ * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
+ * COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
+ * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
+ * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS
+ * OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
+ * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+ * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
+ * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH
+ * DAMAGE.
+ */
+
+/*
+ * Copyright 2009 Sun Microsystems, Inc.  All rights reserved.
+ * Use is subject to license terms.
+ */
+
+#ifndef	_DR_SAS_H_
+#define	_DR_SAS_H_
+
+#ifdef	__cplusplus
+extern "C" {
+#endif
+
+#include <sys/scsi/scsi.h>
+#include "dr_sas_list.h"
+
+/*
+ * MegaRAID SAS2.0 Driver meta data
+ */
+#define	DRSAS_VERSION				"LSIv2.0"
+#define	DRSAS_RELDATE				"Jan 9, 2009"
+
+#define	DRSAS_TRUE				1
+#define	DRSAS_FALSE				0
+
+/*
+ * MegaRAID SAS2.0 device id conversion definitions.
+ */
+#define	INST2LSIRDCTL(x)		((x) << INST_MINOR_SHIFT)
+
+/*
+ * MegaRAID SAS2.0 supported controllers
+ */
+#define	PCI_DEVICE_ID_LSI_2108VDE		0x0078
+#define	PCI_DEVICE_ID_LSI_2108V			0x0079
+
+/*
+ * Register Index for 2108 Controllers.
+ */
+#define	REGISTER_SET_IO_2108			(2)
+
+#define	DRSAS_MAX_SGE_CNT			0x50
+
+#define	DRSAS_IOCTL_DRIVER			0x12341234
+#define	DRSAS_IOCTL_FIRMWARE			0x12345678
+#define	DRSAS_IOCTL_AEN				0x87654321
+
+#define	DRSAS_1_SECOND				1000000
+
+/* Dynamic Enumeration Flags */
+#define	DRSAS_PD_LUN		1
+#define	DRSAS_LD_LUN		0
+#define	DRSAS_PD_TGT_MAX	255
+#define	DRSAS_GET_PD_MAX(s)	((s)->dr_pd_max)
+#define	WWN_STRLEN		17
+
+/*
+ * =====================================
+ * MegaRAID SAS2.0 MFI firmware definitions
+ * =====================================
+ */
+/*
+ * MFI stands for  MegaRAID SAS2.0 FW Interface. This is just a moniker for
+ * protocol between the software and firmware. Commands are issued using
+ * "message frames"
+ */
+
+/*
+ * FW posts its state in upper 4 bits of outbound_msg_0 register
+ */
+#define	MFI_STATE_SHIFT 			28
+#define	MFI_STATE_MASK				((uint32_t)0xF<<MFI_STATE_SHIFT)
+#define	MFI_STATE_UNDEFINED			((uint32_t)0x0<<MFI_STATE_SHIFT)
+#define	MFI_STATE_BB_INIT			((uint32_t)0x1<<MFI_STATE_SHIFT)
+#define	MFI_STATE_FW_INIT			((uint32_t)0x4<<MFI_STATE_SHIFT)
+#define	MFI_STATE_WAIT_HANDSHAKE		((uint32_t)0x6<<MFI_STATE_SHIFT)
+#define	MFI_STATE_FW_INIT_2			((uint32_t)0x7<<MFI_STATE_SHIFT)
+#define	MFI_STATE_DEVICE_SCAN			((uint32_t)0x8<<MFI_STATE_SHIFT)
+#define	MFI_STATE_BOOT_MESSAGE_PENDING		((uint32_t)0x9<<MFI_STATE_SHIFT)
+#define	MFI_STATE_FLUSH_CACHE			((uint32_t)0xA<<MFI_STATE_SHIFT)
+#define	MFI_STATE_READY				((uint32_t)0xB<<MFI_STATE_SHIFT)
+#define	MFI_STATE_OPERATIONAL			((uint32_t)0xC<<MFI_STATE_SHIFT)
+#define	MFI_STATE_FAULT				((uint32_t)0xF<<MFI_STATE_SHIFT)
+
+#define	MRMFI_FRAME_SIZE			64
+
+/*
+ * During FW init, clear pending cmds & reset state using inbound_msg_0
+ *
+ * ABORT	: Abort all pending cmds
+ * READY	: Move from OPERATIONAL to READY state; discard queue info
+ * MFIMODE	: Discard (possible) low MFA posted in 64-bit mode (??)
+ * CLR_HANDSHAKE: FW is waiting for HANDSHAKE from BIOS or Driver
+ */
+#define	MFI_INIT_ABORT				0x00000001
+#define	MFI_INIT_READY				0x00000002
+#define	MFI_INIT_MFIMODE			0x00000004
+#define	MFI_INIT_CLEAR_HANDSHAKE		0x00000008
+#define	MFI_INIT_HOTPLUG			0x00000010
+#define	MFI_STOP_ADP				0x00000020
+#define	MFI_RESET_FLAGS		MFI_INIT_READY|MFI_INIT_MFIMODE|MFI_INIT_ABORT
+
+/*
+ * MFI frame flags
+ */
+#define	MFI_FRAME_POST_IN_REPLY_QUEUE		0x0000
+#define	MFI_FRAME_DONT_POST_IN_REPLY_QUEUE	0x0001
+#define	MFI_FRAME_SGL32				0x0000
+#define	MFI_FRAME_SGL64				0x0002
+#define	MFI_FRAME_SENSE32			0x0000
+#define	MFI_FRAME_SENSE64			0x0004
+#define	MFI_FRAME_DIR_NONE			0x0000
+#define	MFI_FRAME_DIR_WRITE			0x0008
+#define	MFI_FRAME_DIR_READ			0x0010
+#define	MFI_FRAME_DIR_BOTH			0x0018
+
+/*
+ * Definition for cmd_status
+ */
+#define	MFI_CMD_STATUS_POLL_MODE		0xFF
+#define	MFI_CMD_STATUS_SYNC_MODE		0xFF
+
+/*
+ * MFI command opcodes
+ */
+#define	MFI_CMD_OP_INIT				0x00
+#define	MFI_CMD_OP_LD_READ			0x01
+#define	MFI_CMD_OP_LD_WRITE			0x02
+#define	MFI_CMD_OP_LD_SCSI			0x03
+#define	MFI_CMD_OP_PD_SCSI			0x04
+#define	MFI_CMD_OP_DCMD				0x05
+#define	MFI_CMD_OP_ABORT			0x06
+#define	MFI_CMD_OP_SMP				0x07
+#define	MFI_CMD_OP_STP				0x08
+
+#define	DR_DCMD_CTRL_GET_INFO			0x01010000
+
+#define	DR_DCMD_CTRL_CACHE_FLUSH		0x01101000
+#define	DR_FLUSH_CTRL_CACHE			0x01
+#define	DR_FLUSH_DISK_CACHE			0x02
+
+#define	DR_DCMD_CTRL_SHUTDOWN			0x01050000
+#define	DRSAS_ENABLE_DRIVE_SPINDOWN		0x01
+
+#define	DR_DCMD_CTRL_EVENT_GET_INFO		0x01040100
+#define	DR_DCMD_CTRL_EVENT_GET			0x01040300
+#define	DR_DCMD_CTRL_EVENT_WAIT			0x01040500
+#define	DR_DCMD_LD_GET_PROPERTIES		0x03030000
+#define	DR_DCMD_PD_GET_INFO			0x02020000
+
+/*
+ * Solaris Specific MAX values
+ */
+#define	MAX_SGL					24
+/*
+ * MFI command completion codes
+ */
+enum MFI_STAT {
+	MFI_STAT_OK				= 0x00,
+	MFI_STAT_INVALID_CMD			= 0x01,
+	MFI_STAT_INVALID_DCMD			= 0x02,
+	MFI_STAT_INVALID_PARAMETER		= 0x03,
+	MFI_STAT_INVALID_SEQUENCE_NUMBER	= 0x04,
+	MFI_STAT_ABORT_NOT_POSSIBLE		= 0x05,
+	MFI_STAT_APP_HOST_CODE_NOT_FOUND	= 0x06,
+	MFI_STAT_APP_IN_USE			= 0x07,
+	MFI_STAT_APP_NOT_INITIALIZED		= 0x08,
+	MFI_STAT_ARRAY_INDEX_INVALID		= 0x09,
+	MFI_STAT_ARRAY_ROW_NOT_EMPTY		= 0x0a,
+	MFI_STAT_CONFIG_RESOURCE_CONFLICT	= 0x0b,
+	MFI_STAT_DEVICE_NOT_FOUND		= 0x0c,
+	MFI_STAT_DRIVE_TOO_SMALL		= 0x0d,
+	MFI_STAT_FLASH_ALLOC_FAIL		= 0x0e,
+	MFI_STAT_FLASH_BUSY			= 0x0f,
+	MFI_STAT_FLASH_ERROR			= 0x10,
+	MFI_STAT_FLASH_IMAGE_BAD		= 0x11,
+	MFI_STAT_FLASH_IMAGE_INCOMPLETE		= 0x12,
+	MFI_STAT_FLASH_NOT_OPEN			= 0x13,
+	MFI_STAT_FLASH_NOT_STARTED		= 0x14,
+	MFI_STAT_FLUSH_FAILED			= 0x15,
+	MFI_STAT_HOST_CODE_NOT_FOUNT		= 0x16,
+	MFI_STAT_LD_CC_IN_PROGRESS		= 0x17,
+	MFI_STAT_LD_INIT_IN_PROGRESS		= 0x18,
+	MFI_STAT_LD_LBA_OUT_OF_RANGE		= 0x19,
+	MFI_STAT_LD_MAX_CONFIGURED		= 0x1a,
+	MFI_STAT_LD_NOT_OPTIMAL			= 0x1b,
+	MFI_STAT_LD_RBLD_IN_PROGRESS		= 0x1c,
+	MFI_STAT_LD_RECON_IN_PROGRESS		= 0x1d,
+	MFI_STAT_LD_WRONG_RAID_LEVEL		= 0x1e,
+	MFI_STAT_MAX_SPARES_EXCEEDED		= 0x1f,
+	MFI_STAT_MEMORY_NOT_AVAILABLE		= 0x20,
+	MFI_STAT_MFC_HW_ERROR			= 0x21,
+	MFI_STAT_NO_HW_PRESENT			= 0x22,
+	MFI_STAT_NOT_FOUND			= 0x23,
+	MFI_STAT_NOT_IN_ENCL			= 0x24,
+	MFI_STAT_PD_CLEAR_IN_PROGRESS		= 0x25,
+	MFI_STAT_PD_TYPE_WRONG			= 0x26,
+	MFI_STAT_PR_DISABLED			= 0x27,
+	MFI_STAT_ROW_INDEX_INVALID		= 0x28,
+	MFI_STAT_SAS_CONFIG_INVALID_ACTION	= 0x29,
+	MFI_STAT_SAS_CONFIG_INVALID_DATA	= 0x2a,
+	MFI_STAT_SAS_CONFIG_INVALID_PAGE	= 0x2b,
+	MFI_STAT_SAS_CONFIG_INVALID_TYPE	= 0x2c,
+	MFI_STAT_SCSI_DONE_WITH_ERROR		= 0x2d,
+	MFI_STAT_SCSI_IO_FAILED			= 0x2e,
+	MFI_STAT_SCSI_RESERVATION_CONFLICT	= 0x2f,
+	MFI_STAT_SHUTDOWN_FAILED		= 0x30,
+	MFI_STAT_TIME_NOT_SET			= 0x31,
+	MFI_STAT_WRONG_STATE			= 0x32,
+	MFI_STAT_LD_OFFLINE			= 0x33,
+	/* UNUSED: 0x34 to 0xfe */
+	MFI_STAT_INVALID_STATUS			= 0xFF
+};
+
+enum DR_EVT_CLASS {
+	DR_EVT_CLASS_DEBUG		= -2,
+	DR_EVT_CLASS_PROGRESS		= -1,
+	DR_EVT_CLASS_INFO		=  0,
+	DR_EVT_CLASS_WARNING		=  1,
+	DR_EVT_CLASS_CRITICAL		=  2,
+	DR_EVT_CLASS_FATAL		=  3,
+	DR_EVT_CLASS_DEAD		=  4
+};
+
+enum DR_EVT_LOCALE {
+	DR_EVT_LOCALE_LD		= 0x0001,
+	DR_EVT_LOCALE_PD		= 0x0002,
+	DR_EVT_LOCALE_ENCL		= 0x0004,
+	DR_EVT_LOCALE_BBU		= 0x0008,
+	DR_EVT_LOCALE_SAS		= 0x0010,
+	DR_EVT_LOCALE_CTRL		= 0x0020,
+	DR_EVT_LOCALE_CONFIG		= 0x0040,
+	DR_EVT_LOCALE_CLUSTER		= 0x0080,
+	DR_EVT_LOCALE_ALL		= 0xffff
+};
+
+#define	DR_EVT_CFG_CLEARED		0x0004
+#define	DR_EVT_LD_CREATED		0x008a
+#define	DR_EVT_LD_DELETED		0x008b
+#define	DR_EVT_PD_REMOVED_EXT		0x00f8
+#define	DR_EVT_PD_INSERTED_EXT		0x00f7
+
+enum LD_STATE {
+	LD_OFFLINE		= 0,
+	LD_PARTIALLY_DEGRADED	= 1,
+	LD_DEGRADED		= 2,
+	LD_OPTIMAL		= 3,
+	LD_INVALID		= 0xFF
+};
+
+enum DRSAS_EVT {
+	DRSAS_EVT_CONFIG_TGT	= 0,
+	DRSAS_EVT_UNCONFIG_TGT	= 1,
+	DRSAS_EVT_UNCONFIG_SMP	= 2
+};
+
+#define	DMA_OBJ_ALLOCATED	1
+#define	DMA_OBJ_REALLOCATED	2
+#define	DMA_OBJ_FREED		3
+
+/*
+ * dma_obj_t	- Our DMA object
+ * @param buffer	: kernel virtual address
+ * @param size		: size of the data to be allocated
+ * @param acc_handle	: access handle
+ * @param dma_handle	: dma handle
+ * @param dma_cookie	: scatter-gather list
+ * @param dma_attr	: dma attributes for this buffer
+ * Our DMA object. The caller must initialize the size and dma attributes
+ * (dma_attr) fields before allocating the resources.
+ */
+typedef struct {
+	caddr_t			buffer;
+	uint32_t		size;
+	ddi_acc_handle_t	acc_handle;
+	ddi_dma_handle_t	dma_handle;
+	ddi_dma_cookie_t	dma_cookie[DRSAS_MAX_SGE_CNT];
+	ddi_dma_attr_t		dma_attr;
+	uint8_t			status;
+	uint8_t			reserved[3];
+} dma_obj_t;
+
+struct drsas_eventinfo {
+	struct drsas_instance	*instance;
+	int 			tgt;
+	int 			lun;
+	int 			event;
+};
+
+struct drsas_ld {
+	dev_info_t		*dip;
+	uint8_t 		lun_type;
+	uint8_t 		reserved[3];
+};
+
+struct drsas_pd {
+	dev_info_t		*dip;
+	uint8_t 		lun_type;
+	uint8_t 		dev_id;
+	uint8_t 		flags;
+	uint8_t 		reserved;
+};
+
+struct drsas_pd_info {
+	uint16_t	deviceId;
+	uint16_t	seqNum;
+	uint8_t		inquiryData[96];
+	uint8_t		vpdPage83[64];
+	uint8_t		notSupported;
+	uint8_t		scsiDevType;
+	uint8_t		a;
+	uint8_t		device_speed;
+	uint32_t	mediaerrcnt;
+	uint32_t	other;
+	uint32_t	pred;
+	uint32_t	lastpred;
+	uint16_t	fwState;
+	uint8_t		disabled;
+	uint8_t		linkspwwd;
+	uint32_t	ddfType;
+	struct {
+		uint8_t	count;
+		uint8_t	isPathBroken;
+		uint8_t	connectorIndex[2];
+		uint8_t	reserved[4];
+		uint64_t sasAddr[2];
+		uint8_t	reserved2[16];
+	} pathInfo;
+};
+
+typedef struct drsas_instance {
+	uint32_t	*producer;
+	uint32_t	*consumer;
+
+	uint32_t	*reply_queue;
+	dma_obj_t	mfi_internal_dma_obj;
+
+	uint8_t		init_id;
+	uint8_t		reserved[3];
+
+	uint16_t	max_num_sge;
+	uint16_t	max_fw_cmds;
+	uint32_t	max_sectors_per_req;
+
+	struct drsas_cmd **cmd_list;
+
+	mlist_t		cmd_pool_list;
+	kmutex_t	cmd_pool_mtx;
+
+	mlist_t		cmd_pend_list;
+	kmutex_t	cmd_pend_mtx;
+
+	dma_obj_t	mfi_evt_detail_obj;
+	struct drsas_cmd *aen_cmd;
+
+	uint32_t	aen_seq_num;
+	uint32_t	aen_class_locale_word;
+
+	scsi_hba_tran_t		*tran;
+
+	kcondvar_t	int_cmd_cv;
+	kmutex_t	int_cmd_mtx;
+
+	kcondvar_t	aen_cmd_cv;
+	kmutex_t	aen_cmd_mtx;
+
+	kcondvar_t	abort_cmd_cv;
+	kmutex_t	abort_cmd_mtx;
+
+	dev_info_t		*dip;
+	ddi_acc_handle_t	pci_handle;
+
+	timeout_id_t	timeout_id;
+	uint32_t	unique_id;
+	uint16_t	fw_outstanding;
+	caddr_t		regmap;
+	ddi_acc_handle_t	regmap_handle;
+	uint8_t		isr_level;
+	ddi_iblock_cookie_t	iblock_cookie;
+	ddi_iblock_cookie_t	soft_iblock_cookie;
+	ddi_softintr_t		soft_intr_id;
+	uint8_t		softint_running;
+	kmutex_t	completed_pool_mtx;
+	mlist_t		completed_pool_list;
+
+	caddr_t		internal_buf;
+	uint32_t	internal_buf_dmac_add;
+	uint32_t	internal_buf_size;
+
+	uint16_t	vendor_id;
+	uint16_t	device_id;
+	uint16_t	subsysvid;
+	uint16_t	subsysid;
+	int		instance;
+	int		baseaddress;
+	char		iocnode[16];
+
+	int		fm_capabilities;
+
+	struct drsas_func_ptr *func_ptr;
+	/* MSI interrupts specific */
+	ddi_intr_handle_t *intr_htable;
+	int		intr_type;
+	int		intr_cnt;
+	size_t		intr_size;
+	uint_t		intr_pri;
+	int		intr_cap;
+
+	ddi_taskq_t	*taskq;
+	struct drsas_ld	*dr_ld_list;
+} drsas_t;
+
+struct drsas_func_ptr {
+	int (*read_fw_status_reg)(struct drsas_instance *);
+	void (*issue_cmd)(struct drsas_cmd *, struct drsas_instance *);
+	int (*issue_cmd_in_sync_mode)(struct drsas_instance *,
+	    struct drsas_cmd *);
+	int (*issue_cmd_in_poll_mode)(struct drsas_instance *,
+	    struct drsas_cmd *);
+	void (*enable_intr)(struct drsas_instance *);
+	void (*disable_intr)(struct drsas_instance *);
+	int (*intr_ack)(struct drsas_instance *);
+};
+
+/*
+ * ### Helper routines ###
+ */
+
+/*
+ * con_log() - console log routine
+ * @param level		: indicates the severity of the message.
+ * @fparam mt		: format string
+ *
+ * con_log displays the error messages on the console based on the current
+ * debug level. Also it attaches the appropriate kernel severity level with
+ * the message.
+ *
+ *
+ * console messages debug levels
+ */
+#define	CL_NONE		0	/* No debug information */
+#define	CL_ANN		1	/* print unconditionally, announcements */
+#define	CL_ANN1		2	/* No o/p  */
+#define	CL_DLEVEL1	3	/* debug level 1, informative */
+#define	CL_DLEVEL2	4	/* debug level 2, verbose */
+#define	CL_DLEVEL3	5	/* debug level 3, very verbose */
+
+#ifdef __SUNPRO_C
+#define	__func__ ""
+#endif
+
+#define	con_log(level, fmt) { if (debug_level_g >= level) cmn_err fmt; }
+
+/*
+ * ### SCSA definitions ###
+ */
+#define	PKT2TGT(pkt)	((pkt)->pkt_address.a_target)
+#define	PKT2LUN(pkt)	((pkt)->pkt_address.a_lun)
+#define	PKT2TRAN(pkt)	((pkt)->pkt_adress.a_hba_tran)
+#define	ADDR2TRAN(ap)	((ap)->a_hba_tran)
+
+#define	TRAN2MR(tran)	(struct drsas_instance *)(tran)->tran_hba_private)
+#define	ADDR2MR(ap)	(TRAN2MR(ADDR2TRAN(ap))
+
+#define	PKT2CMD(pkt)	((struct scsa_cmd *)(pkt)->pkt_ha_private)
+#define	CMD2PKT(sp)	((sp)->cmd_pkt)
+#define	PKT2REQ(pkt)	(&(PKT2CMD(pkt)->request))
+
+#define	CMD2ADDR(cmd)	(&CMD2PKT(cmd)->pkt_address)
+#define	CMD2TRAN(cmd)	(CMD2PKT(cmd)->pkt_address.a_hba_tran)
+#define	CMD2MR(cmd)	(TRAN2MR(CMD2TRAN(cmd)))
+
+#define	CFLAG_DMAVALID		0x0001	/* requires a dma operation */
+#define	CFLAG_DMASEND		0x0002	/* Transfer from the device */
+#define	CFLAG_CONSISTENT	0x0040	/* consistent data transfer */
+
+/*
+ * ### Data structures for ioctl inteface and internal commands ###
+ */
+
+/*
+ * Data direction flags
+ */
+#define	UIOC_RD		0x00001
+#define	UIOC_WR		0x00002
+
+#define	SCP2HOST(scp)		(scp)->device->host	/* to host */
+#define	SCP2HOSTDATA(scp)	SCP2HOST(scp)->hostdata	/* to soft state */
+#define	SCP2CHANNEL(scp)	(scp)->device->channel	/* to channel */
+#define	SCP2TARGET(scp)		(scp)->device->id	/* to target */
+#define	SCP2LUN(scp)		(scp)->device->lun	/* to LUN */
+
+#define	SCSIHOST2ADAP(host)	(((caddr_t *)(host->hostdata))[0])
+#define	SCP2ADAPTER(scp)				\
+	(struct drsas_instance *)SCSIHOST2ADAP(SCP2HOST(scp))
+
+#define	MRDRV_IS_LOGICAL_SCSA(instance, acmd)		\
+	(acmd->device_id < MRDRV_MAX_LD) ? 1 : 0
+#define	MRDRV_IS_LOGICAL(ap)				\
+	((ap->a_target < MRDRV_MAX_LD) && (ap->a_lun == 0)) ? 1 : 0
+#define	MAP_DEVICE_ID(instance, ap)			\
+	(ap->a_target)
+
+#define	HIGH_LEVEL_INTR			1
+#define	NORMAL_LEVEL_INTR		0
+
+/*
+ * scsa_cmd  - Per-command mr private data
+ * @param cmd_dmahandle		:  dma handle
+ * @param cmd_dmacookies	:  current dma cookies
+ * @param cmd_pkt		:  scsi_pkt reference
+ * @param cmd_dmacount		:  dma count
+ * @param cmd_cookie		:  next cookie
+ * @param cmd_ncookies		:  cookies per window
+ * @param cmd_cookiecnt		:  cookies per sub-win
+ * @param cmd_nwin		:  number of dma windows
+ * @param cmd_curwin		:  current dma window
+ * @param cmd_dma_offset	:  current window offset
+ * @param cmd_dma_len		:  current window length
+ * @param cmd_flags		:  private flags
+ * @param cmd_cdblen		:  length of cdb
+ * @param cmd_scblen		:  length of scb
+ * @param cmd_buf		:  command buffer
+ * @param channel		:  channel for scsi sub-system
+ * @param target		:  target for scsi sub-system
+ * @param lun			:  LUN for scsi sub-system
+ *
+ * - Allocated at same time as scsi_pkt by scsi_hba_pkt_alloc(9E)
+ * - Pointed to by pkt_ha_private field in scsi_pkt
+ */
+struct scsa_cmd {
+	ddi_dma_handle_t	cmd_dmahandle;
+	ddi_dma_cookie_t	cmd_dmacookies[DRSAS_MAX_SGE_CNT];
+	struct scsi_pkt		*cmd_pkt;
+	ulong_t			cmd_dmacount;
+	uint_t			cmd_cookie;
+	uint_t			cmd_ncookies;
+	uint_t			cmd_cookiecnt;
+	uint_t			cmd_nwin;
+	uint_t			cmd_curwin;
+	off_t			cmd_dma_offset;
+	ulong_t			cmd_dma_len;
+	ulong_t			cmd_flags;
+	uint_t			cmd_cdblen;
+	uint_t			cmd_scblen;
+	struct buf		*cmd_buf;
+	ushort_t		device_id;
+	uchar_t			islogical;
+	uchar_t			lun;
+	struct drsas_device	*drsas_dev;
+};
+
+
+struct drsas_cmd {
+	union drsas_frame	*frame;
+	uint32_t		frame_phys_addr;
+	uint8_t			*sense;
+	uint32_t		sense_phys_addr;
+	dma_obj_t		frame_dma_obj;
+	uint8_t			frame_dma_obj_status;
+
+	uint32_t		index;
+	uint8_t			sync_cmd;
+	uint8_t			cmd_status;
+	uint16_t		abort_aen;
+	mlist_t			list;
+	uint32_t		frame_count;
+	struct scsa_cmd		*cmd;
+	struct scsi_pkt		*pkt;
+};
+
+#define	MAX_MGMT_ADAPTERS			1024
+#define	IOC_SIGNATURE				"MR-SAS"
+
+#define	IOC_CMD_FIRMWARE			0x0
+#define	DRSAS_DRIVER_IOCTL_COMMON		0xF0010000
+#define	DRSAS_DRIVER_IOCTL_DRIVER_VERSION	0xF0010100
+#define	DRSAS_DRIVER_IOCTL_PCI_INFORMATION	0xF0010200
+#define	DRSAS_DRIVER_IOCTL_MRRAID_STATISTICS	0xF0010300
+
+
+#define	DRSAS_MAX_SENSE_LENGTH			32
+
+struct drsas_mgmt_info {
+
+	uint16_t			count;
+	struct drsas_instance		*instance[MAX_MGMT_ADAPTERS];
+	uint16_t			map[MAX_MGMT_ADAPTERS];
+	int				max_index;
+};
+
+#pragma pack(1)
+
+/*
+ * SAS controller properties
+ */
+struct drsas_ctrl_prop {
+	uint16_t	seq_num;
+	uint16_t	pred_fail_poll_interval;
+	uint16_t	intr_throttle_count;
+	uint16_t	intr_throttle_timeouts;
+
+	uint8_t		rebuild_rate;
+	uint8_t		patrol_read_rate;
+	uint8_t		bgi_rate;
+	uint8_t		cc_rate;
+	uint8_t		recon_rate;
+
+	uint8_t		cache_flush_interval;
+
+	uint8_t		spinup_drv_count;
+	uint8_t		spinup_delay;
+
+	uint8_t		cluster_enable;
+	uint8_t		coercion_mode;
+	uint8_t		disk_write_cache_disable;
+	uint8_t		alarm_enable;
+
+	uint8_t		reserved[44];
+};
+
+/*
+ * SAS controller information
+ */
+struct drsas_ctrl_info {
+	/* PCI device information */
+	struct {
+		uint16_t	vendor_id;
+		uint16_t	device_id;
+		uint16_t	sub_vendor_id;
+		uint16_t	sub_device_id;
+		uint8_t	reserved[24];
+	} pci;
+
+	/* Host interface information */
+	struct {
+		uint8_t	PCIX		: 1;
+		uint8_t	PCIE		: 1;
+		uint8_t	iSCSI		: 1;
+		uint8_t	SAS_3G		: 1;
+		uint8_t	reserved_0	: 4;
+		uint8_t	reserved_1[6];
+		uint8_t	port_count;
+		uint64_t	port_addr[8];
+	} host_interface;
+
+	/* Device (backend) interface information */
+	struct {
+		uint8_t	SPI		: 1;
+		uint8_t	SAS_3G		: 1;
+		uint8_t	SATA_1_5G	: 1;
+		uint8_t	SATA_3G		: 1;
+		uint8_t	reserved_0	: 4;
+		uint8_t	reserved_1[6];
+		uint8_t	port_count;
+		uint64_t	port_addr[8];
+	} device_interface;
+
+	/* List of components residing in flash. All str are null terminated */
+	uint32_t	image_check_word;
+	uint32_t	image_component_count;
+
+	struct {
+		char	name[8];
+		char	version[32];
+		char	build_date[16];
+		char	built_time[16];
+	} image_component[8];
+
+	/*
+	 * List of flash components that have been flashed on the card, but
+	 * are not in use, pending reset of the adapter. This list will be
+	 * empty if a flash operation has not occurred. All stings are null
+	 * terminated
+	 */
+	uint32_t	pending_image_component_count;
+
+	struct {
+		char	name[8];
+		char	version[32];
+		char	build_date[16];
+		char	build_time[16];
+	} pending_image_component[8];
+
+	uint8_t		max_arms;
+	uint8_t		max_spans;
+	uint8_t		max_arrays;
+	uint8_t		max_lds;
+
+	char		product_name[80];
+	char		serial_no[32];
+
+	/*
+	 * Other physical/controller/operation information. Indicates the
+	 * presence of the hardware
+	 */
+	struct {
+		uint32_t	bbu		: 1;
+		uint32_t	alarm		: 1;
+		uint32_t	nvram		: 1;
+		uint32_t	uart		: 1;
+		uint32_t	reserved	: 28;
+	} hw_present;
+
+	uint32_t	current_fw_time;
+
+	/* Maximum data transfer sizes */
+	uint16_t		max_concurrent_cmds;
+	uint16_t		max_sge_count;
+	uint32_t		max_request_size;
+
+	/* Logical and physical device counts */
+	uint16_t		ld_present_count;
+	uint16_t		ld_degraded_count;
+	uint16_t		ld_offline_count;
+
+	uint16_t		pd_present_count;
+	uint16_t		pd_disk_present_count;
+	uint16_t		pd_disk_pred_failure_count;
+	uint16_t		pd_disk_failed_count;
+
+	/* Memory size information */
+	uint16_t		nvram_size;
+	uint16_t		memory_size;
+	uint16_t		flash_size;
+
+	/* Error counters */
+	uint16_t		mem_correctable_error_count;
+	uint16_t		mem_uncorrectable_error_count;
+
+	/* Cluster information */
+	uint8_t		cluster_permitted;
+	uint8_t		cluster_active;
+	uint8_t		reserved_1[2];
+
+	/* Controller capabilities structures */
+	struct {
+		uint32_t	raid_level_0	: 1;
+		uint32_t	raid_level_1	: 1;
+		uint32_t	raid_level_5	: 1;
+		uint32_t	raid_level_1E	: 1;
+		uint32_t	reserved	: 28;
+	} raid_levels;
+
+	struct {
+		uint32_t	rbld_rate		: 1;
+		uint32_t	cc_rate			: 1;
+		uint32_t	bgi_rate		: 1;
+		uint32_t	recon_rate		: 1;
+		uint32_t	patrol_rate		: 1;
+		uint32_t	alarm_control		: 1;
+		uint32_t	cluster_supported	: 1;
+		uint32_t	bbu			: 1;
+		uint32_t	spanning_allowed	: 1;
+		uint32_t	dedicated_hotspares	: 1;
+		uint32_t	revertible_hotspares	: 1;
+		uint32_t	foreign_config_import	: 1;
+		uint32_t	self_diagnostic		: 1;
+		uint32_t	reserved		: 19;
+	} adapter_operations;
+
+	struct {
+		uint32_t	read_policy	: 1;
+		uint32_t	write_policy	: 1;
+		uint32_t	io_policy	: 1;
+		uint32_t	access_policy	: 1;
+		uint32_t	reserved	: 28;
+	} ld_operations;
+
+	struct {
+		uint8_t	min;
+		uint8_t	max;
+		uint8_t	reserved[2];
+	} stripe_size_operations;
+
+	struct {
+		uint32_t	force_online	: 1;
+		uint32_t	force_offline	: 1;
+		uint32_t	force_rebuild	: 1;
+		uint32_t	reserved	: 29;
+	} pd_operations;
+
+	struct {
+		uint32_t	ctrl_supports_sas	: 1;
+		uint32_t	ctrl_supports_sata	: 1;
+		uint32_t	allow_mix_in_encl	: 1;
+		uint32_t	allow_mix_in_ld		: 1;
+		uint32_t	allow_sata_in_cluster	: 1;
+		uint32_t	reserved		: 27;
+	} pd_mix_support;
+
+	/* Include the controller properties (changeable items) */
+	uint8_t				reserved_2[12];
+	struct drsas_ctrl_prop		properties;
+
+	uint8_t				pad[0x800 - 0x640];
+};
+
+/*
+ * ==================================
+ * MegaRAID SAS2.0 driver definitions
+ * ==================================
+ */
+#define	MRDRV_MAX_NUM_CMD			1024
+
+#define	MRDRV_MAX_PD_CHANNELS			2
+#define	MRDRV_MAX_LD_CHANNELS			2
+#define	MRDRV_MAX_CHANNELS			(MRDRV_MAX_PD_CHANNELS + \
+						MRDRV_MAX_LD_CHANNELS)
+#define	MRDRV_MAX_DEV_PER_CHANNEL		128
+#define	MRDRV_DEFAULT_INIT_ID			-1
+#define	MRDRV_MAX_CMD_PER_LUN			1000
+#define	MRDRV_MAX_LUN				1
+#define	MRDRV_MAX_LD				64
+
+#define	MRDRV_RESET_WAIT_TIME			300
+#define	MRDRV_RESET_NOTICE_INTERVAL		5
+
+#define	DRSAS_IOCTL_CMD				0
+
+/*
+ * FW can accept both 32 and 64 bit SGLs. We want to allocate 32/64 bit
+ * SGLs based on the size of dma_addr_t
+ */
+#define	IS_DMA64		(sizeof (dma_addr_t) == 8)
+
+#define	IB_MSG_0_OFF			0x10	/* XScale */
+#define	OB_MSG_0_OFF			0x18	/* XScale */
+#define	IB_DOORBELL_OFF			0x20	/* XScale & ROC */
+#define	OB_INTR_STATUS_OFF		0x30	/* XScale & ROC */
+#define	OB_INTR_MASK_OFF		0x34	/* XScale & ROC */
+#define	IB_QPORT_OFF			0x40	/* XScale & ROC */
+#define	OB_DOORBELL_CLEAR_OFF		0xA0	/* ROC */
+#define	OB_SCRATCH_PAD_0_OFF		0xB0	/* ROC */
+#define	OB_INTR_MASK			0xFFFFFFFF
+#define	OB_DOORBELL_CLEAR_MASK		0xFFFFFFFF
+
+/*
+ * All MFI register set macros accept drsas_register_set*
+ */
+#define	WR_IB_MSG_0(v, instance) 	ddi_put32((instance)->regmap_handle, \
+	(uint32_t *)((uintptr_t)(instance)->regmap + IB_MSG_0_OFF), (v))
+
+#define	RD_OB_MSG_0(instance) 		ddi_get32((instance)->regmap_handle, \
+	(uint32_t *)((uintptr_t)(instance)->regmap + OB_MSG_0_OFF))
+
+#define	WR_IB_DOORBELL(v, instance)	ddi_put32((instance)->regmap_handle, \
+	(uint32_t *)((uintptr_t)(instance)->regmap + IB_DOORBELL_OFF), (v))
+
+#define	RD_IB_DOORBELL(instance)	ddi_get32((instance)->regmap_handle, \
+	(uint32_t *)((uintptr_t)(instance)->regmap + IB_DOORBELL_OFF))
+
+#define	WR_OB_INTR_STATUS(v, instance) 	ddi_put32((instance)->regmap_handle, \
+	(uint32_t *)((uintptr_t)(instance)->regmap + OB_INTR_STATUS_OFF), (v))
+
+#define	RD_OB_INTR_STATUS(instance) 	ddi_get32((instance)->regmap_handle, \
+	(uint32_t *)((uintptr_t)(instance)->regmap + OB_INTR_STATUS_OFF))
+
+#define	WR_OB_INTR_MASK(v, instance) 	ddi_put32((instance)->regmap_handle, \
+	(uint32_t *)((uintptr_t)(instance)->regmap + OB_INTR_MASK_OFF), (v))
+
+#define	RD_OB_INTR_MASK(instance) 	ddi_get32((instance)->regmap_handle, \
+	(uint32_t *)((uintptr_t)(instance)->regmap + OB_INTR_MASK_OFF))
+
+#define	WR_IB_QPORT(v, instance) 	ddi_put32((instance)->regmap_handle, \
+	(uint32_t *)((uintptr_t)(instance)->regmap + IB_QPORT_OFF), (v))
+
+#define	WR_OB_DOORBELL_CLEAR(v, instance) ddi_put32((instance)->regmap_handle, \
+	(uint32_t *)((uintptr_t)(instance)->regmap + OB_DOORBELL_CLEAR_OFF), \
+	(v))
+
+#define	RD_OB_SCRATCH_PAD_0(instance) 	ddi_get32((instance)->regmap_handle, \
+	(uint32_t *)((uintptr_t)(instance)->regmap + OB_SCRATCH_PAD_0_OFF))
+
+/*
+ * When FW is in MFI_STATE_READY or MFI_STATE_OPERATIONAL, the state data
+ * of Outbound Msg Reg 0 indicates max concurrent cmds supported, max SGEs
+ * supported per cmd and if 64-bit MFAs (M64) is enabled or disabled.
+ */
+#define	MFI_OB_INTR_STATUS_MASK		0x00000002
+
+/*
+ * This MFI_REPLY_2108_MESSAGE_INTR flag is used also
+ * in enable_intr_ppc also. Hence bit 2, i.e. 0x4 has
+ * been set in this flag along with bit 1.
+ */
+#define	MFI_REPLY_2108_MESSAGE_INTR		0x00000001
+#define	MFI_REPLY_2108_MESSAGE_INTR_MASK	0x00000005
+
+#define	MFI_POLL_TIMEOUT_SECS		60
+
+#define	MFI_ENABLE_INTR(instance)  ddi_put32((instance)->regmap_handle, \
+	(uint32_t *)((uintptr_t)(instance)->regmap + OB_INTR_MASK_OFF), 1)
+#define	MFI_DISABLE_INTR(instance)					\
+{									\
+	uint32_t disable = 1;						\
+	uint32_t mask =  ddi_get32((instance)->regmap_handle, 		\
+	    (uint32_t *)((uintptr_t)(instance)->regmap + OB_INTR_MASK_OFF));\
+	mask &= ~disable;						\
+	ddi_put32((instance)->regmap_handle, (uint32_t *)		\
+	    (uintptr_t)((instance)->regmap + OB_INTR_MASK_OFF), mask);	\
+}
+
+/* By default, the firmware programs for 8 Kbytes of memory */
+#define	DEFAULT_MFI_MEM_SZ	8192
+#define	MINIMUM_MFI_MEM_SZ	4096
+
+/* DCMD Message Frame MAILBOX0-11 */
+#define	DCMD_MBOX_SZ		12
+
+
+struct drsas_register_set {
+	uint32_t	reserved_0[4];
+
+	uint32_t	inbound_msg_0;
+	uint32_t	inbound_msg_1;
+	uint32_t	outbound_msg_0;
+	uint32_t	outbound_msg_1;
+
+	uint32_t	inbound_doorbell;
+	uint32_t	inbound_intr_status;
+	uint32_t	inbound_intr_mask;
+
+	uint32_t	outbound_doorbell;
+	uint32_t	outbound_intr_status;
+	uint32_t	outbound_intr_mask;
+
+	uint32_t	reserved_1[2];
+
+	uint32_t	inbound_queue_port;
+	uint32_t	outbound_queue_port;
+
+	uint32_t 	reserved_2[22];
+
+	uint32_t 	outbound_doorbell_clear;
+
+	uint32_t 	reserved_3[3];
+
+	uint32_t 	outbound_scratch_pad;
+
+	uint32_t 	reserved_4[3];
+
+	uint32_t 	inbound_low_queue_port;
+
+	uint32_t 	inbound_high_queue_port;
+
+	uint32_t 	reserved_5;
+	uint32_t 	index_registers[820];
+};
+
+struct drsas_sge32 {
+	uint32_t	phys_addr;
+	uint32_t	length;
+};
+
+struct drsas_sge64 {
+	uint64_t	phys_addr;
+	uint32_t	length;
+};
+
+union drsas_sgl {
+	struct drsas_sge32	sge32[1];
+	struct drsas_sge64	sge64[1];
+};
+
+struct drsas_header {
+	uint8_t		cmd;
+	uint8_t		sense_len;
+	uint8_t		cmd_status;
+	uint8_t		scsi_status;
+
+	uint8_t		target_id;
+	uint8_t		lun;
+	uint8_t		cdb_len;
+	uint8_t		sge_count;
+
+	uint32_t	context;
+	uint8_t		req_id;
+	uint8_t		msgvector;
+	uint16_t	pad_0;
+
+	uint16_t	flags;
+	uint16_t	timeout;
+	uint32_t	data_xferlen;
+};
+
+union drsas_sgl_frame {
+	struct drsas_sge32	sge32[8];
+	struct drsas_sge64	sge64[5];
+};
+
+struct drsas_init_frame {
+	uint8_t		cmd;
+	uint8_t		reserved_0;
+	uint8_t		cmd_status;
+
+	uint8_t		reserved_1;
+	uint32_t	reserved_2;
+
+	uint32_t	context;
+	uint8_t		req_id;
+	uint8_t		msgvector;
+	uint16_t	pad_0;
+
+	uint16_t	flags;
+	uint16_t	reserved_3;
+	uint32_t	data_xfer_len;
+
+	uint32_t	queue_info_new_phys_addr_lo;
+	uint32_t	queue_info_new_phys_addr_hi;
+	uint32_t	queue_info_old_phys_addr_lo;
+	uint32_t	queue_info_old_phys_addr_hi;
+
+	uint32_t	reserved_4[6];
+};
+
+struct drsas_init_queue_info {
+	uint32_t		init_flags;
+	uint32_t		reply_queue_entries;
+
+	uint32_t		reply_queue_start_phys_addr_lo;
+	uint32_t		reply_queue_start_phys_addr_hi;
+	uint32_t		producer_index_phys_addr_lo;
+	uint32_t		producer_index_phys_addr_hi;
+	uint32_t		consumer_index_phys_addr_lo;
+	uint32_t		consumer_index_phys_addr_hi;
+};
+
+struct drsas_io_frame {
+	uint8_t			cmd;
+	uint8_t			sense_len;
+	uint8_t			cmd_status;
+	uint8_t			scsi_status;
+
+	uint8_t			target_id;
+	uint8_t			access_byte;
+	uint8_t			reserved_0;
+	uint8_t			sge_count;
+
+	uint32_t		context;
+	uint8_t			req_id;
+	uint8_t			msgvector;
+	uint16_t		pad_0;
+
+	uint16_t		flags;
+	uint16_t		timeout;
+	uint32_t		lba_count;
+
+	uint32_t		sense_buf_phys_addr_lo;
+	uint32_t		sense_buf_phys_addr_hi;
+
+	uint32_t		start_lba_lo;
+	uint32_t		start_lba_hi;
+
+	union drsas_sgl		sgl;
+};
+
+struct drsas_pthru_frame {
+	uint8_t			cmd;
+	uint8_t			sense_len;
+	uint8_t			cmd_status;
+	uint8_t			scsi_status;
+
+	uint8_t			target_id;
+	uint8_t			lun;
+	uint8_t			cdb_len;
+	uint8_t			sge_count;
+
+	uint32_t		context;
+	uint8_t			req_id;
+	uint8_t			msgvector;
+	uint16_t		pad_0;
+
+	uint16_t		flags;
+	uint16_t		timeout;
+	uint32_t		data_xfer_len;
+
+	uint32_t		sense_buf_phys_addr_lo;
+	uint32_t		sense_buf_phys_addr_hi;
+
+	uint8_t			cdb[16];
+	union drsas_sgl		sgl;
+};
+
+struct drsas_dcmd_frame {
+	uint8_t			cmd;
+	uint8_t			reserved_0;
+	uint8_t			cmd_status;
+	uint8_t			reserved_1[4];
+	uint8_t			sge_count;
+
+	uint32_t		context;
+	uint8_t			req_id;
+	uint8_t			msgvector;
+	uint16_t		pad_0;
+
+	uint16_t		flags;
+	uint16_t		timeout;
+
+	uint32_t		data_xfer_len;
+	uint32_t		opcode;
+
+	union {
+		uint8_t b[DCMD_MBOX_SZ];
+		uint16_t s[6];
+		uint32_t w[3];
+	} mbox;
+
+	union drsas_sgl		sgl;
+};
+
+struct drsas_abort_frame {
+	uint8_t		cmd;
+	uint8_t		reserved_0;
+	uint8_t		cmd_status;
+
+	uint8_t		reserved_1;
+	uint32_t	reserved_2;
+
+	uint32_t	context;
+	uint8_t		req_id;
+	uint8_t		msgvector;
+	uint16_t	pad_0;
+
+	uint16_t	flags;
+	uint16_t	reserved_3;
+	uint32_t	reserved_4;
+
+	uint32_t	abort_context;
+	uint32_t	pad_1;
+
+	uint32_t	abort_mfi_phys_addr_lo;
+	uint32_t	abort_mfi_phys_addr_hi;
+
+	uint32_t	reserved_5[6];
+};
+
+struct drsas_smp_frame {
+	uint8_t		cmd;
+	uint8_t		reserved_1;
+	uint8_t		cmd_status;
+	uint8_t		connection_status;
+
+	uint8_t		reserved_2[3];
+	uint8_t		sge_count;
+
+	uint32_t	context;
+	uint8_t		req_id;
+	uint8_t		msgvector;
+	uint16_t	pad_0;
+
+	uint16_t	flags;
+	uint16_t	timeout;
+
+	uint32_t	data_xfer_len;
+
+	uint64_t	sas_addr;
+
+	union drsas_sgl	sgl[2];
+};
+
+struct drsas_stp_frame {
+	uint8_t		cmd;
+	uint8_t		reserved_1;
+	uint8_t		cmd_status;
+	uint8_t		connection_status;
+
+	uint8_t		target_id;
+	uint8_t		reserved_2[2];
+	uint8_t		sge_count;
+
+	uint32_t	context;
+	uint8_t		req_id;
+	uint8_t		msgvector;
+	uint16_t	pad_0;
+
+	uint16_t	flags;
+	uint16_t	timeout;
+
+	uint32_t	data_xfer_len;
+
+	uint16_t	fis[10];
+	uint32_t	stp_flags;
+	union drsas_sgl	sgl;
+};
+
+union drsas_frame {
+	struct drsas_header		hdr;
+	struct drsas_init_frame		init;
+	struct drsas_io_frame		io;
+	struct drsas_pthru_frame	pthru;
+	struct drsas_dcmd_frame		dcmd;
+	struct drsas_abort_frame	abort;
+	struct drsas_smp_frame		smp;
+	struct drsas_stp_frame		stp;
+
+	uint8_t			raw_bytes[64];
+};
+
+typedef struct drsas_pd_address {
+	uint16_t	device_id;
+	uint16_t	encl_id;
+
+	union {
+		struct {
+			uint8_t encl_index;
+			uint8_t slot_number;
+		} pd_address;
+		struct {
+			uint8_t	encl_position;
+			uint8_t	encl_connector_index;
+		} encl_address;
+	}address;
+
+	uint8_t	scsi_dev_type;
+
+	union {
+		uint8_t		port_bitmap;
+		uint8_t		port_numbers;
+	} connected;
+
+	uint64_t		sas_addr[2];
+} drsas_pd_address_t;
+
+union drsas_evt_class_locale {
+	struct {
+		uint16_t	locale;
+		uint8_t		reserved;
+		int8_t		class;
+	} members;
+
+	uint32_t	word;
+};
+
+struct drsas_evt_log_info {
+	uint32_t	newest_seq_num;
+	uint32_t	oldest_seq_num;
+	uint32_t	clear_seq_num;
+	uint32_t	shutdown_seq_num;
+	uint32_t	boot_seq_num;
+};
+
+struct drsas_progress {
+	uint16_t	progress;
+	uint16_t	elapsed_seconds;
+};
+
+struct drsas_evtarg_ld {
+	uint16_t	target_id;
+	uint8_t		ld_index;
+	uint8_t		reserved;
+};
+
+struct drsas_evtarg_pd {
+	uint16_t	device_id;
+	uint8_t		encl_index;
+	uint8_t		slot_number;
+};
+
+struct drsas_evt_detail {
+	uint32_t	seq_num;
+	uint32_t	time_stamp;
+	uint32_t	code;
+	union drsas_evt_class_locale	cl;
+	uint8_t		arg_type;
+	uint8_t		reserved1[15];
+
+	union {
+		struct {
+			struct drsas_evtarg_pd	pd;
+			uint8_t			cdb_length;
+			uint8_t			sense_length;
+			uint8_t			reserved[2];
+			uint8_t			cdb[16];
+			uint8_t			sense[64];
+		} cdbSense;
+
+		struct drsas_evtarg_ld		ld;
+
+		struct {
+			struct drsas_evtarg_ld	ld;
+			uint64_t		count;
+		} ld_count;
+
+		struct {
+			uint64_t		lba;
+			struct drsas_evtarg_ld	ld;
+		} ld_lba;
+
+		struct {
+			struct drsas_evtarg_ld	ld;
+			uint32_t		prevOwner;
+			uint32_t		newOwner;
+		} ld_owner;
+
+		struct {
+			uint64_t		ld_lba;
+			uint64_t		pd_lba;
+			struct drsas_evtarg_ld	ld;
+			struct drsas_evtarg_pd	pd;
+		} ld_lba_pd_lba;
+
+		struct {
+			struct drsas_evtarg_ld	ld;
+			struct drsas_progress	prog;
+		} ld_prog;
+
+		struct {
+			struct drsas_evtarg_ld	ld;
+			uint32_t		prev_state;
+			uint32_t		new_state;
+		} ld_state;
+
+		struct {
+			uint64_t		strip;
+			struct drsas_evtarg_ld	ld;
+		} ld_strip;
+
+		struct drsas_evtarg_pd		pd;
+
+		struct {
+			struct drsas_evtarg_pd	pd;
+			uint32_t		err;
+		} pd_err;
+
+		struct {
+			uint64_t		lba;
+			struct drsas_evtarg_pd	pd;
+		} pd_lba;
+
+		struct {
+			uint64_t		lba;
+			struct drsas_evtarg_pd	pd;
+			struct drsas_evtarg_ld	ld;
+		} pd_lba_ld;
+
+		struct {
+			struct drsas_evtarg_pd	pd;
+			struct drsas_progress	prog;
+		} pd_prog;
+
+		struct {
+			struct drsas_evtarg_pd	pd;
+			uint32_t		prevState;
+			uint32_t		newState;
+		} pd_state;
+
+		struct {
+			uint16_t	vendorId;
+			uint16_t	deviceId;
+			uint16_t	subVendorId;
+			uint16_t	subDeviceId;
+		} pci;
+
+		uint32_t	rate;
+		char		str[96];
+
+		struct {
+			uint32_t	rtc;
+			uint32_t	elapsedSeconds;
+		} time;
+
+		struct {
+			uint32_t	ecar;
+			uint32_t	elog;
+			char		str[64];
+		} ecc;
+
+		drsas_pd_address_t	pd_addr;
+
+		uint8_t		b[96];
+		uint16_t	s[48];
+		uint32_t	w[24];
+		uint64_t	d[12];
+	} args;
+
+	char	description[128];
+
+};
+
+/* only 63 are usable by the application */
+#define	MAX_LOGICAL_DRIVES			64
+/* only 255 physical devices may be used */
+#define	MAX_PHYSICAL_DEVICES			256
+#define	MAX_PD_PER_ENCLOSURE			64
+/* maximum disks per array */
+#define	MAX_ROW_SIZE				32
+/* maximum spans per logical drive */
+#define	MAX_SPAN_DEPTH				8
+/* maximum number of arrays a hot spare may be dedicated to */
+#define	MAX_ARRAYS_DEDICATED			16
+/* maximum number of arrays which may exist */
+#define	MAX_ARRAYS				128
+/* maximum number of foreign configs that may ha managed at once */
+#define	MAX_FOREIGN_CONFIGS			8
+/* maximum spares (global and dedicated combined) */
+#define	MAX_SPARES_FOR_THE_CONTROLLER		MAX_PHYSICAL_DEVICES
+/* maximum possible Target IDs (i.e. 0 to 63) */
+#define	MAX_TARGET_ID				63
+/* maximum number of supported enclosures */
+#define	MAX_ENCLOSURES				32
+/* maximum number of PHYs per controller */
+#define	MAX_PHYS_PER_CONTROLLER			16
+/* maximum number of LDs per array (due to DDF limitations) */
+#define	MAX_LDS_PER_ARRAY			16
+
+/*
+ * -----------------------------------------------------------------------------
+ * -----------------------------------------------------------------------------
+ *
+ * Logical Drive commands
+ *
+ * -----------------------------------------------------------------------------
+ * -----------------------------------------------------------------------------
+ */
+#define	DR_DCMD_LD	0x03000000,	/* Logical Device (LD) opcodes */
+
+/*
+ * Input:	dcmd.opcode	- DR_DCMD_LD_GET_LIST
+ *		dcmd.mbox	- reserved
+ *		dcmd.sge IN	- ptr to returned DR_LD_LIST structure
+ * Desc:	Return the logical drive list structure
+ * Status:	No error
+ */
+
+/*
+ * defines the logical drive reference structure
+ */
+typedef	union _DR_LD_REF {	/* LD reference structure */
+	struct {
+		uint8_t	targetId; /* LD target id (0 to MAX_TARGET_ID) */
+		uint8_t	reserved; /* reserved for in line with DR_PD_REF */
+		uint16_t seqNum;  /* Sequence Number */
+	} ld_ref;
+	uint32_t ref;		/* shorthand reference to full 32-bits */
+} DR_LD_REF;			/* 4 bytes */
+
+/*
+ * defines the logical drive list structure
+ */
+typedef struct _DR_LD_LIST {
+	uint32_t	ldCount;	/* number of LDs */
+	uint32_t	reserved;	/* pad to 8-byte boundary */
+	struct {
+		DR_LD_REF ref;	/* LD reference */
+		uint8_t	state;		/* current LD state (DR_LD_STATE) */
+		uint8_t	reserved[3];	/* pad to 8-byte boundary */
+		uint64_t size;		/* LD size */
+	} ldList[MAX_LOGICAL_DRIVES];
+} DR_LD_LIST;
+
+struct drsas_drv_ver {
+	uint8_t	signature[12];
+	uint8_t	os_name[16];
+	uint8_t	os_ver[12];
+	uint8_t	drv_name[20];
+	uint8_t	drv_ver[32];
+	uint8_t	drv_rel_date[20];
+};
+
+#define	PCI_TYPE0_ADDRESSES		6
+#define	PCI_TYPE1_ADDRESSES		2
+#define	PCI_TYPE2_ADDRESSES		5
+
+struct drsas_pci_common_header {
+	uint16_t	vendorID;		/* (ro) */
+	uint16_t	deviceID;		/* (ro) */
+	uint16_t	command;		/* Device control */
+	uint16_t	status;
+	uint8_t		revisionID;		/* (ro) */
+	uint8_t		progIf;			/* (ro) */
+	uint8_t		subClass;		/* (ro) */
+	uint8_t		baseClass;		/* (ro) */
+	uint8_t		cacheLineSize;		/* (ro+) */
+	uint8_t		latencyTimer;		/* (ro+) */
+	uint8_t		headerType;		/* (ro) */
+	uint8_t		bist;			/* Built in self test */
+
+	union {
+	    struct {
+		uint32_t	baseAddresses[PCI_TYPE0_ADDRESSES];
+		uint32_t	cis;
+		uint16_t	subVendorID;
+		uint16_t	subSystemID;
+		uint32_t	romBaseAddress;
+		uint8_t		capabilitiesPtr;
+		uint8_t		reserved1[3];
+		uint32_t	reserved2;
+		uint8_t		interruptLine;
+		uint8_t		interruptPin;	/* (ro) */
+		uint8_t		minimumGrant;	/* (ro) */
+		uint8_t		maximumLatency;	/* (ro) */
+	    } type_0;
+
+	    struct {
+		uint32_t	baseAddresses[PCI_TYPE1_ADDRESSES];
+		uint8_t		primaryBus;
+		uint8_t		secondaryBus;
+		uint8_t		subordinateBus;
+		uint8_t		secondaryLatency;
+		uint8_t		ioBase;
+		uint8_t		ioLimit;
+		uint16_t	secondaryStatus;
+		uint16_t	memoryBase;
+		uint16_t	memoryLimit;
+		uint16_t	prefetchBase;
+		uint16_t	prefetchLimit;
+		uint32_t	prefetchBaseUpper32;
+		uint32_t	prefetchLimitUpper32;
+		uint16_t	ioBaseUpper16;
+		uint16_t	ioLimitUpper16;
+		uint8_t		capabilitiesPtr;
+		uint8_t		reserved1[3];
+		uint32_t	romBaseAddress;
+		uint8_t		interruptLine;
+		uint8_t		interruptPin;
+		uint16_t	bridgeControl;
+	    } type_1;
+
+	    struct {
+		uint32_t	socketRegistersBaseAddress;
+		uint8_t		capabilitiesPtr;
+		uint8_t		reserved;
+		uint16_t	secondaryStatus;
+		uint8_t		primaryBus;
+		uint8_t		secondaryBus;
+		uint8_t		subordinateBus;
+		uint8_t		secondaryLatency;
+		struct {
+			uint32_t	base;
+			uint32_t	limit;
+		} range[PCI_TYPE2_ADDRESSES-1];
+		uint8_t		interruptLine;
+		uint8_t		interruptPin;
+		uint16_t	bridgeControl;
+	    } type_2;
+	} header;
+};
+
+struct drsas_pci_link_capability {
+	union {
+	    struct {
+		uint32_t linkSpeed		:4;
+		uint32_t linkWidth		:6;
+		uint32_t aspmSupport		:2;
+		uint32_t losExitLatency		:3;
+		uint32_t l1ExitLatency		:3;
+		uint32_t rsvdp			:6;
+		uint32_t portNumber		:8;
+	    } bits;
+
+	    uint32_t asUlong;
+	} cap;
+
+};
+
+struct drsas_pci_link_status_capability {
+	union {
+	    struct {
+		uint16_t linkSpeed		:4;
+		uint16_t negotiatedLinkWidth	:6;
+		uint16_t linkTrainingError	:1;
+		uint16_t linkTraning		:1;
+		uint16_t slotClockConfig	:1;
+		uint16_t rsvdZ			:3;
+	    } bits;
+
+	    uint16_t asUshort;
+	} stat_cap;
+
+	uint16_t reserved;
+
+};
+
+struct drsas_pci_capabilities {
+	struct drsas_pci_link_capability	linkCapability;
+	struct drsas_pci_link_status_capability linkStatusCapability;
+};
+
+struct drsas_pci_information
+{
+	uint32_t		busNumber;
+	uint8_t			deviceNumber;
+	uint8_t			functionNumber;
+	uint8_t			interruptVector;
+	uint8_t			reserved;
+	struct drsas_pci_common_header pciHeaderInfo;
+	struct drsas_pci_capabilities capability;
+	uint8_t			reserved2[32];
+};
+
+struct drsas_ioctl {
+	uint16_t	version;
+	uint16_t	controller_id;
+	uint8_t		signature[8];
+	uint32_t	reserved_1;
+	uint32_t	control_code;
+	uint32_t	reserved_2[2];
+	uint8_t		frame[64];
+	union drsas_sgl_frame sgl_frame;
+	uint8_t		sense_buff[DRSAS_MAX_SENSE_LENGTH];
+	uint8_t		data[1];
+};
+
+struct drsas_aen {
+	uint16_t	host_no;
+	uint16_t	cmd_status;
+	uint32_t	seq_num;
+	uint32_t	class_locale_word;
+};
+#pragma pack()
+
+#ifndef	DDI_VENDOR_LSI
+#define	DDI_VENDOR_LSI		"LSI"
+#endif /* DDI_VENDOR_LSI */
+
+static int	drsas_getinfo(dev_info_t *, ddi_info_cmd_t,  void *, void **);
+static int	drsas_attach(dev_info_t *, ddi_attach_cmd_t);
+static int	drsas_reset(dev_info_t *, ddi_reset_cmd_t);
+static int	drsas_detach(dev_info_t *, ddi_detach_cmd_t);
+static int	drsas_open(dev_t *, int, int, cred_t *);
+static int	drsas_close(dev_t, int, int, cred_t *);
+static int	drsas_ioctl(dev_t, int, intptr_t, int, cred_t *, int *);
+
+static int	drsas_tran_tgt_init(dev_info_t *, dev_info_t *,
+		    scsi_hba_tran_t *, struct scsi_device *);
+static struct scsi_pkt *drsas_tran_init_pkt(struct scsi_address *, register
+		    struct scsi_pkt *, struct buf *, int, int, int, int,
+		    int (*)(), caddr_t);
+static int	drsas_tran_start(struct scsi_address *,
+		    register struct scsi_pkt *);
+static int	drsas_tran_abort(struct scsi_address *, struct scsi_pkt *);
+static int	drsas_tran_reset(struct scsi_address *, int);
+static int	drsas_tran_getcap(struct scsi_address *, char *, int);
+static int	drsas_tran_setcap(struct scsi_address *, char *, int, int);
+static void	drsas_tran_destroy_pkt(struct scsi_address *,
+		    struct scsi_pkt *);
+static void	drsas_tran_dmafree(struct scsi_address *, struct scsi_pkt *);
+static void	drsas_tran_sync_pkt(struct scsi_address *, struct scsi_pkt *);
+static uint_t	drsas_isr();
+static uint_t	drsas_softintr();
+
+static int	init_mfi(struct drsas_instance *);
+static int	drsas_free_dma_obj(struct drsas_instance *, dma_obj_t);
+static int	drsas_alloc_dma_obj(struct drsas_instance *, dma_obj_t *,
+		    uchar_t);
+static struct drsas_cmd *get_mfi_pkt(struct drsas_instance *);
+static void	return_mfi_pkt(struct drsas_instance *,
+		    struct drsas_cmd *);
+
+static void	free_space_for_mfi(struct drsas_instance *);
+static void	free_additional_dma_buffer(struct drsas_instance *);
+static int	alloc_additional_dma_buffer(struct drsas_instance *);
+static int	read_fw_status_reg_ppc(struct drsas_instance *);
+static void	issue_cmd_ppc(struct drsas_cmd *, struct drsas_instance *);
+static int	issue_cmd_in_poll_mode_ppc(struct drsas_instance *,
+		    struct drsas_cmd *);
+static int	issue_cmd_in_sync_mode_ppc(struct drsas_instance *,
+		    struct drsas_cmd *);
+static void	enable_intr_ppc(struct drsas_instance *);
+static void	disable_intr_ppc(struct drsas_instance *);
+static int	intr_ack_ppc(struct drsas_instance *);
+static int	mfi_state_transition_to_ready(struct drsas_instance *);
+static void	destroy_mfi_frame_pool(struct drsas_instance *);
+static int	create_mfi_frame_pool(struct drsas_instance *);
+static int	drsas_dma_alloc(struct drsas_instance *, struct scsi_pkt *,
+		    struct buf *, int, int (*)());
+static int	drsas_dma_move(struct drsas_instance *,
+			struct scsi_pkt *, struct buf *);
+static void	flush_cache(struct drsas_instance *instance);
+static void	display_scsi_inquiry(caddr_t);
+static int	start_mfi_aen(struct drsas_instance *instance);
+static int	handle_drv_ioctl(struct drsas_instance *instance,
+		    struct drsas_ioctl *ioctl, int mode);
+static int	handle_mfi_ioctl(struct drsas_instance *instance,
+		    struct drsas_ioctl *ioctl, int mode);
+static int	handle_mfi_aen(struct drsas_instance *instance,
+		    struct drsas_aen *aen);
+static void	fill_up_drv_ver(struct drsas_drv_ver *dv);
+static struct drsas_cmd *build_cmd(struct drsas_instance *instance,
+		    struct scsi_address *ap, struct scsi_pkt *pkt,
+		    uchar_t *cmd_done);
+static int	register_mfi_aen(struct drsas_instance *instance,
+		    uint32_t seq_num, uint32_t class_locale_word);
+static int	issue_mfi_pthru(struct drsas_instance *instance, struct
+		    drsas_ioctl *ioctl, struct drsas_cmd *cmd, int mode);
+static int	issue_mfi_dcmd(struct drsas_instance *instance, struct
+		    drsas_ioctl *ioctl, struct drsas_cmd *cmd, int mode);
+static int	issue_mfi_smp(struct drsas_instance *instance, struct
+		    drsas_ioctl *ioctl, struct drsas_cmd *cmd, int mode);
+static int	issue_mfi_stp(struct drsas_instance *instance, struct
+		    drsas_ioctl *ioctl, struct drsas_cmd *cmd, int mode);
+static int	abort_aen_cmd(struct drsas_instance *instance,
+		    struct drsas_cmd *cmd_to_abort);
+
+static int	drsas_common_check(struct drsas_instance *instance,
+		    struct  drsas_cmd *cmd);
+static void	drsas_fm_init(struct drsas_instance *instance);
+static void	drsas_fm_fini(struct drsas_instance *instance);
+static int	drsas_fm_error_cb(dev_info_t *, ddi_fm_error_t *,
+		    const void *);
+static void	drsas_fm_ereport(struct drsas_instance *instance,
+		    char *detail);
+static int	drsas_check_dma_handle(ddi_dma_handle_t handle);
+static int	drsas_check_acc_handle(ddi_acc_handle_t handle);
+
+static void	drsas_rem_intrs(struct drsas_instance *instance);
+static int	drsas_add_intrs(struct drsas_instance *instance, int intr_type);
+
+static void	drsas_tran_tgt_free(dev_info_t *, dev_info_t *,
+		    scsi_hba_tran_t *, struct scsi_device *);
+static int	drsas_tran_bus_config(dev_info_t *, uint_t,
+		    ddi_bus_config_op_t, void *, dev_info_t **);
+static int	drsas_parse_devname(char *, int *, int *);
+static int	drsas_config_all_devices(struct drsas_instance *);
+static int 	drsas_config_scsi_device(struct drsas_instance *,
+		    struct scsi_device *, dev_info_t **);
+static int 	drsas_config_ld(struct drsas_instance *, uint16_t,
+				uint8_t, dev_info_t **);
+static dev_info_t *drsas_find_child(struct drsas_instance *, uint16_t,
+			uint8_t);
+static int	drsas_name_node(dev_info_t *, char *, int);
+static void	drsas_issue_evt_taskq(struct drsas_eventinfo *);
+static int	drsas_service_evt(struct drsas_instance *, int, int, int,
+			uint64_t);
+static int	drsas_mode_sense_build(struct scsi_pkt *);
+
+#ifdef	__cplusplus
+}
+#endif
+
+#endif /* _DR_SAS_H_ */
diff --git a/usr/src/uts/common/io/dr_sas/dr_sas_list.h b/usr/src/uts/common/io/dr_sas/dr_sas_list.h
new file mode 100644
index 0000000000..4154a77796
--- /dev/null
+++ b/usr/src/uts/common/io/dr_sas/dr_sas_list.h
@@ -0,0 +1,212 @@
+/*
+ * dr_sas_list.h: header for dr_sas
+ *
+ * Solaris MegaRAID driver for SAS2.0 controllers
+ * Copyright (c) 2008-2009, LSI Logic Corporation.
+ * All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions are met:
+ *
+ * 1. Redistributions of source code must retain the above copyright notice,
+ *    this list of conditions and the following disclaimer.
+ *
+ * 2. Redistributions in binary form must reproduce the above copyright notice,
+ *    this list of conditions and the following disclaimer in the documentation
+ *    and/or other materials provided with the distribution.
+ *
+ * 3. Neither the name of the author nor the names of its contributors may be
+ *    used to endorse or promote products derived from this software without
+ *    specific prior written permission.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
+ * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
+ * COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
+ * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
+ * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS
+ * OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
+ * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+ * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
+ * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH
+ * DAMAGE.
+ */
+
+/*
+ * Copyright 2009 Sun Microsystems, Inc.  All rights reserved.
+ * Use is subject to license terms.
+ */
+
+#ifndef	_DR_SAS_LIST_H_
+#define	_DR_SAS_LIST_H_
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+/*
+ * Simple doubly linked list implementation.
+ *
+ * Some of the internal functions ("__xxx") are useful when
+ * manipulating whole lists rather than single entries, as
+ * sometimes we already know the next/prev entries and we can
+ * generate better code by using them directly rather than
+ * using the generic single-entry routines.
+ */
+
+struct mlist_head {
+	struct mlist_head *next, *prev;
+};
+
+typedef struct mlist_head mlist_t;
+
+#define	LIST_HEAD_INIT(name) { &(name), &(name) }
+
+#define	LIST_HEAD(name) \
+	struct mlist_head name = LIST_HEAD_INIT(name)
+
+#define	INIT_LIST_HEAD(ptr) { \
+	(ptr)->next = (ptr); (ptr)->prev = (ptr); \
+}
+
+
+/*
+ * Insert a new entry between two known consecutive entries.
+ *
+ * This is only for internal list manipulation where we know
+ * the prev/next entries already!
+ */
+static void __list_add(struct mlist_head *new,
+	struct mlist_head *prev,
+	struct mlist_head *next)
+{
+	next->prev = new;
+	new->next = next;
+	new->prev = prev;
+	prev->next = new;
+}
+
+
+/*
+ * mlist_add - add a new entry
+ * @new: new entry to be added
+ * @head: list head to add it after
+ *
+ * Insert a new entry after the specified head.
+ * This is good for implementing stacks.
+ */
+static void mlist_add(struct mlist_head *new, struct mlist_head *head)
+{
+	__list_add(new, head, head->next);
+}
+
+
+/*
+ * mlist_add_tail - add a new entry
+ * @new: new entry to be added
+ * @head: list head to add it before
+ *
+ * Insert a new entry before the specified head.
+ * This is useful for implementing queues.
+ */
+static void mlist_add_tail(struct mlist_head *new, struct mlist_head *head)
+{
+	__list_add(new, head->prev, head);
+}
+
+
+
+/*
+ * Delete a list entry by making the prev/next entries
+ * point to each other.
+ *
+ * This is only for internal list manipulation where we know
+ * the prev/next entries already!
+ */
+static void __list_del(struct mlist_head *prev,
+			struct mlist_head *next)
+{
+	next->prev = prev;
+	prev->next = next;
+}
+
+
+/*
+ * mlist_del_init - deletes entry from list and reinitialize it.
+ * @entry: the element to delete from the list.
+ */
+static void mlist_del_init(struct mlist_head *entry)
+{
+	__list_del(entry->prev, entry->next);
+	INIT_LIST_HEAD(entry);
+}
+
+
+/*
+ * mlist_empty - tests whether a list is empty
+ * @head: the list to test.
+ */
+static int mlist_empty(struct mlist_head *head)
+{
+	return (head->next == head);
+}
+
+
+/*
+ * mlist_splice - join two lists
+ * @list: the new list to add.
+ * @head: the place to add it in the first list.
+ */
+static void mlist_splice(struct mlist_head *list, struct mlist_head *head)
+{
+	struct mlist_head *first = list->next;
+
+	if (first != list) {
+		struct mlist_head *last = list->prev;
+		struct mlist_head *at = head->next;
+
+		first->prev = head;
+		head->next = first;
+
+		last->next = at;
+		at->prev = last;
+	}
+}
+
+
+/*
+ * mlist_entry - get the struct for this entry
+ * @ptr:	the &struct mlist_head pointer.
+ * @type:	the type of the struct this is embedded in.
+ * @member:	the name of the list_struct within the struct.
+ */
+#define	mlist_entry(ptr, type, member) \
+	((type *)((size_t)(ptr) - offsetof(type, member)))
+
+
+/*
+ * mlist_for_each	-	iterate over a list
+ * @pos:	the &struct mlist_head to use as a loop counter.
+ * @head:	the head for your list.
+ */
+#define	mlist_for_each(pos, head) \
+	for (pos = (head)->next, prefetch(pos->next); pos != (head); \
+		pos = pos->next, prefetch(pos->next))
+
+
+/*
+ * mlist_for_each_safe - iterate over a list safe against removal of list entry
+ * @pos:	the &struct mlist_head to use as a loop counter.
+ * @n:		another &struct mlist_head to use as temporary storage
+ * @head:	the head for your list.
+ */
+#define	mlist_for_each_safe(pos, n, head) \
+	for (pos = (head)->next, n = pos->next; pos != (head); \
+		pos = n, n = pos->next)
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif /* _DR_SAS_LIST_H_ */
diff --git a/usr/src/uts/common/io/eventfd.c b/usr/src/uts/common/io/eventfd.c
new file mode 100644
index 0000000000..6683a9ca8e
--- /dev/null
+++ b/usr/src/uts/common/io/eventfd.c
@@ -0,0 +1,414 @@
+/*
+ * This file and its contents are supplied under the terms of the
+ * Common Development and Distribution License ("CDDL"), version 1.0.
+ * You may only use this file in accordance with the terms of version
+ * 1.0 of the CDDL.
+ *
+ * A full copy of the text of the CDDL should have accompanied this
+ * source.  A copy of the CDDL is also available via the Internet at
+ * http://www.illumos.org/license/CDDL.
+ */
+
+/*
+ * Copyright (c) 2015 Joyent, Inc.  All rights reserved.
+ */
+
+/*
+ * Support for the eventfd facility, a Linux-borne facility for user-generated
+ * file descriptor-based events.
+ */
+
+#include <sys/ddi.h>
+#include <sys/sunddi.h>
+#include <sys/eventfd.h>
+#include <sys/conf.h>
+#include <sys/vmem.h>
+#include <sys/sysmacros.h>
+#include <sys/filio.h>
+#include <sys/stat.h>
+#include <sys/file.h>
+
+struct eventfd_state;
+typedef struct eventfd_state eventfd_state_t;
+
+struct eventfd_state {
+	kmutex_t efd_lock;			/* lock protecting state */
+	boolean_t efd_semaphore;		/* boolean: sema. semantics */
+	kcondvar_t efd_cv;			/* condvar */
+	pollhead_t efd_pollhd;			/* poll head */
+	uint64_t efd_value;			/* value */
+	eventfd_state_t *efd_next;		/* next state on global list */
+};
+
+/*
+ * Internal global variables.
+ */
+static kmutex_t		eventfd_lock;		/* lock protecting state */
+static dev_info_t	*eventfd_devi;		/* device info */
+static vmem_t		*eventfd_minor;		/* minor number arena */
+static void		*eventfd_softstate;	/* softstate pointer */
+static eventfd_state_t	*eventfd_state;		/* global list of state */
+
+/*ARGSUSED*/
+static int
+eventfd_open(dev_t *devp, int flag, int otyp, cred_t *cred_p)
+{
+	eventfd_state_t *state;
+	major_t major = getemajor(*devp);
+	minor_t minor = getminor(*devp);
+
+	if (minor != EVENTFDMNRN_EVENTFD)
+		return (ENXIO);
+
+	mutex_enter(&eventfd_lock);
+
+	minor = (minor_t)(uintptr_t)vmem_alloc(eventfd_minor, 1,
+	    VM_BESTFIT | VM_SLEEP);
+
+	if (ddi_soft_state_zalloc(eventfd_softstate, minor) != DDI_SUCCESS) {
+		vmem_free(eventfd_minor, (void *)(uintptr_t)minor, 1);
+		mutex_exit(&eventfd_lock);
+		return (NULL);
+	}
+
+	state = ddi_get_soft_state(eventfd_softstate, minor);
+	*devp = makedevice(major, minor);
+
+	state->efd_next = eventfd_state;
+	eventfd_state = state;
+
+	mutex_exit(&eventfd_lock);
+
+	return (0);
+}
+
+/*ARGSUSED*/
+static int
+eventfd_read(dev_t dev, uio_t *uio, cred_t *cr)
+{
+	eventfd_state_t *state;
+	minor_t minor = getminor(dev);
+	uint64_t val, oval;
+	int err;
+
+	if (uio->uio_resid < sizeof (val))
+		return (EINVAL);
+
+	state = ddi_get_soft_state(eventfd_softstate, minor);
+
+	mutex_enter(&state->efd_lock);
+
+	while (state->efd_value == 0) {
+		if (uio->uio_fmode & (FNDELAY|FNONBLOCK)) {
+			mutex_exit(&state->efd_lock);
+			return (EAGAIN);
+		}
+
+		if (!cv_wait_sig_swap(&state->efd_cv, &state->efd_lock)) {
+			mutex_exit(&state->efd_lock);
+			return (EINTR);
+		}
+	}
+
+	/*
+	 * We have a non-zero value and we own the lock; our behavior now
+	 * depends on whether or not EFD_SEMAPHORE was set when the eventfd
+	 * was created.
+	 */
+	val = oval = state->efd_value;
+
+	if (state->efd_semaphore) {
+		state->efd_value--;
+		val = 1;
+	} else {
+		state->efd_value = 0;
+	}
+
+	err = uiomove(&val, sizeof (val), UIO_READ, uio);
+
+	mutex_exit(&state->efd_lock);
+
+	if (oval == EVENTFD_VALMAX) {
+		cv_broadcast(&state->efd_cv);
+		pollwakeup(&state->efd_pollhd, POLLWRNORM | POLLOUT);
+	}
+
+	return (err);
+}
+
+/*ARGSUSED*/
+static int
+eventfd_write(dev_t dev, struct uio *uio, cred_t *credp)
+{
+	eventfd_state_t *state;
+	minor_t minor = getminor(dev);
+	uint64_t val, oval;
+	int err;
+
+	if (uio->uio_resid < sizeof (val))
+		return (EINVAL);
+
+	if ((err = uiomove(&val, sizeof (val), UIO_WRITE, uio)) != 0)
+		return (err);
+
+	if (val > EVENTFD_VALMAX)
+		return (EINVAL);
+
+	state = ddi_get_soft_state(eventfd_softstate, minor);
+
+	mutex_enter(&state->efd_lock);
+
+	while (val > EVENTFD_VALMAX - state->efd_value) {
+		if (uio->uio_fmode & (FNDELAY|FNONBLOCK)) {
+			mutex_exit(&state->efd_lock);
+			return (EAGAIN);
+		}
+
+		if (!cv_wait_sig_swap(&state->efd_cv, &state->efd_lock)) {
+			mutex_exit(&state->efd_lock);
+			return (EINTR);
+		}
+	}
+
+	/*
+	 * We now know that we can add the value without overflowing.
+	 */
+	state->efd_value = (oval = state->efd_value) + val;
+
+	mutex_exit(&state->efd_lock);
+
+	if (oval == 0) {
+		cv_broadcast(&state->efd_cv);
+		pollwakeup(&state->efd_pollhd, POLLRDNORM | POLLIN);
+	}
+
+	return (0);
+}
+
+/*ARGSUSED*/
+static int
+eventfd_poll(dev_t dev, short events, int anyyet, short *reventsp,
+    struct pollhead **phpp)
+{
+	eventfd_state_t *state;
+	minor_t minor = getminor(dev);
+	short revents = 0;
+
+	state = ddi_get_soft_state(eventfd_softstate, minor);
+
+	mutex_enter(&state->efd_lock);
+
+	if (state->efd_value > 0)
+		revents |= POLLRDNORM | POLLIN;
+
+	if (state->efd_value < EVENTFD_VALMAX)
+		revents |= POLLWRNORM | POLLOUT;
+
+	if (!(*reventsp = revents & events) && !anyyet)
+		*phpp = &state->efd_pollhd;
+
+	mutex_exit(&state->efd_lock);
+
+	return (0);
+}
+
+/*ARGSUSED*/
+static int
+eventfd_ioctl(dev_t dev, int cmd, intptr_t arg, int md, cred_t *cr, int *rv)
+{
+	eventfd_state_t *state;
+	minor_t minor = getminor(dev);
+
+	state = ddi_get_soft_state(eventfd_softstate, minor);
+
+	switch (cmd) {
+	case EVENTFDIOC_SEMAPHORE: {
+		mutex_enter(&state->efd_lock);
+		state->efd_semaphore ^= 1;
+		mutex_exit(&state->efd_lock);
+
+		return (0);
+	}
+
+	default:
+		break;
+	}
+
+	return (ENOTTY);
+}
+
+/*ARGSUSED*/
+static int
+eventfd_close(dev_t dev, int flag, int otyp, cred_t *cred_p)
+{
+	eventfd_state_t *state, **sp;
+	minor_t minor = getminor(dev);
+
+	state = ddi_get_soft_state(eventfd_softstate, minor);
+
+	if (state->efd_pollhd.ph_list != NULL) {
+		pollwakeup(&state->efd_pollhd, POLLERR);
+		pollhead_clean(&state->efd_pollhd);
+	}
+
+	mutex_enter(&eventfd_lock);
+
+	/*
+	 * Remove our state from our global list.
+	 */
+	for (sp = &eventfd_state; *sp != state; sp = &((*sp)->efd_next))
+		VERIFY(*sp != NULL);
+
+	*sp = (*sp)->efd_next;
+
+	ddi_soft_state_free(eventfd_softstate, minor);
+	vmem_free(eventfd_minor, (void *)(uintptr_t)minor, 1);
+
+	mutex_exit(&eventfd_lock);
+
+	return (0);
+}
+
+/*ARGSUSED*/
+static int
+eventfd_attach(dev_info_t *devi, ddi_attach_cmd_t cmd)
+{
+	mutex_enter(&eventfd_lock);
+
+	if (ddi_soft_state_init(&eventfd_softstate,
+	    sizeof (eventfd_state_t), 0) != 0) {
+		cmn_err(CE_NOTE, "/dev/eventfd failed to create soft state");
+		mutex_exit(&eventfd_lock);
+		return (DDI_FAILURE);
+	}
+
+	if (ddi_create_minor_node(devi, "eventfd", S_IFCHR,
+	    EVENTFDMNRN_EVENTFD, DDI_PSEUDO, NULL) == DDI_FAILURE) {
+		cmn_err(CE_NOTE, "/dev/eventfd couldn't create minor node");
+		ddi_soft_state_fini(&eventfd_softstate);
+		mutex_exit(&eventfd_lock);
+		return (DDI_FAILURE);
+	}
+
+	ddi_report_dev(devi);
+	eventfd_devi = devi;
+
+	eventfd_minor = vmem_create("eventfd_minor", (void *)EVENTFDMNRN_CLONE,
+	    UINT32_MAX - EVENTFDMNRN_CLONE, 1, NULL, NULL, NULL, 0,
+	    VM_SLEEP | VMC_IDENTIFIER);
+
+	mutex_exit(&eventfd_lock);
+
+	return (DDI_SUCCESS);
+}
+
+/*ARGSUSED*/
+static int
+eventfd_detach(dev_info_t *dip, ddi_detach_cmd_t cmd)
+{
+	switch (cmd) {
+	case DDI_DETACH:
+		break;
+
+	case DDI_SUSPEND:
+		return (DDI_SUCCESS);
+
+	default:
+		return (DDI_FAILURE);
+	}
+
+	mutex_enter(&eventfd_lock);
+	vmem_destroy(eventfd_minor);
+
+	ddi_remove_minor_node(eventfd_devi, NULL);
+	eventfd_devi = NULL;
+
+	ddi_soft_state_fini(&eventfd_softstate);
+	mutex_exit(&eventfd_lock);
+
+	return (DDI_SUCCESS);
+}
+
+/*ARGSUSED*/
+static int
+eventfd_info(dev_info_t *dip, ddi_info_cmd_t infocmd, void *arg, void **result)
+{
+	int error;
+
+	switch (infocmd) {
+	case DDI_INFO_DEVT2DEVINFO:
+		*result = (void *)eventfd_devi;
+		error = DDI_SUCCESS;
+		break;
+	case DDI_INFO_DEVT2INSTANCE:
+		*result = (void *)0;
+		error = DDI_SUCCESS;
+		break;
+	default:
+		error = DDI_FAILURE;
+	}
+	return (error);
+}
+
+static struct cb_ops eventfd_cb_ops = {
+	eventfd_open,		/* open */
+	eventfd_close,		/* close */
+	nulldev,		/* strategy */
+	nulldev,		/* print */
+	nodev,			/* dump */
+	eventfd_read,		/* read */
+	eventfd_write,		/* write */
+	eventfd_ioctl,		/* ioctl */
+	nodev,			/* devmap */
+	nodev,			/* mmap */
+	nodev,			/* segmap */
+	eventfd_poll,		/* poll */
+	ddi_prop_op,		/* cb_prop_op */
+	0,			/* streamtab  */
+	D_NEW | D_MP		/* Driver compatibility flag */
+};
+
+static struct dev_ops eventfd_ops = {
+	DEVO_REV,		/* devo_rev */
+	0,			/* refcnt */
+	eventfd_info,		/* get_dev_info */
+	nulldev,		/* identify */
+	nulldev,		/* probe */
+	eventfd_attach,		/* attach */
+	eventfd_detach,		/* detach */
+	nodev,			/* reset */
+	&eventfd_cb_ops,	/* driver operations */
+	NULL,			/* bus operations */
+	nodev,			/* dev power */
+	ddi_quiesce_not_needed,	/* quiesce */
+};
+
+static struct modldrv modldrv = {
+	&mod_driverops,		/* module type (this is a pseudo driver) */
+	"eventfd support",	/* name of module */
+	&eventfd_ops,		/* driver ops */
+};
+
+static struct modlinkage modlinkage = {
+	MODREV_1,
+	(void *)&modldrv,
+	NULL
+};
+
+int
+_init(void)
+{
+	return (mod_install(&modlinkage));
+}
+
+int
+_info(struct modinfo *modinfop)
+{
+	return (mod_info(&modlinkage, modinfop));
+}
+
+int
+_fini(void)
+{
+	return (mod_remove(&modlinkage));
+}
diff --git a/usr/src/uts/common/io/eventfd.conf b/usr/src/uts/common/io/eventfd.conf
new file mode 100644
index 0000000000..f9c6dc11b2
--- /dev/null
+++ b/usr/src/uts/common/io/eventfd.conf
@@ -0,0 +1,16 @@
+#
+# This file and its contents are supplied under the terms of the
+# Common Development and Distribution License ("CDDL"), version 1.0.
+# You may only use this file in accordance with the terms of version
+# 1.0 of the CDDL.
+#
+# A full copy of the text of the CDDL should have accompanied this
+# source.  A copy of the CDDL is also available via the Internet at
+# http://www.illumos.org/license/CDDL.
+#
+
+#
+# Copyright (c) 2014 Joyent, Inc.  All rights reserved.
+#
+
+name="eventfd" parent="pseudo" instance=0;
diff --git a/usr/src/uts/common/io/fibre-channel/fca/oce/oce_rx.c b/usr/src/uts/common/io/fibre-channel/fca/oce/oce_rx.c
index c8af6ae527..2930fe578c 100644
--- a/usr/src/uts/common/io/fibre-channel/fca/oce/oce_rx.c
+++ b/usr/src/uts/common/io/fibre-channel/fca/oce/oce_rx.c
@@ -532,8 +532,7 @@ oce_drain_rq_cq(void *arg)
 			if (dev->function_mode & FLEX10_MODE) {
 				if (cqe->u0.s.vlan_tag_present &&
 				    cqe->u0.s.qnq) {
-					oce_rx_insert_tag(mp,
-					    cqe->u0.s.vlan_tag);
+					oce_rx_insert_tag(mp, cqe->u0.s.vlan_tag);
 				}
 			} else if (cqe->u0.s.vlan_tag_present) {
 				oce_rx_insert_tag(mp, cqe->u0.s.vlan_tag);
diff --git a/usr/src/uts/common/io/fibre-channel/impl/fctl.c b/usr/src/uts/common/io/fibre-channel/impl/fctl.c
index 634de6c6dd..87105e779d 100644
--- a/usr/src/uts/common/io/fibre-channel/impl/fctl.c
+++ b/usr/src/uts/common/io/fibre-channel/impl/fctl.c
@@ -24,6 +24,7 @@
  */
 /*
  * Copyright 2012 Garrett D'Amore <garrett@damore.org>.  All rights reserved.
+ * Copyright (c) 2015 Joyent, Inc.  All rights reserved.
  */
 /*
  * Fibre channel Transport Library (fctl)
@@ -5500,6 +5501,11 @@ fc_ulp_get_adapter_paths(char *pathList, int count)
 		maxPorts ++;
 	}
 
+	if (maxPorts == 0) {
+		mutex_exit(&fctl_port_lock);
+		return (0);
+	}
+
 	/* Now allocate a buffer to store all the pointers for comparisons */
 	portList = kmem_zalloc(sizeof (fc_local_port_t *) * maxPorts, KM_SLEEP);
 
diff --git a/usr/src/uts/common/io/gsqueue/gsqueue.c b/usr/src/uts/common/io/gsqueue/gsqueue.c
new file mode 100644
index 0000000000..b484b16142
--- /dev/null
+++ b/usr/src/uts/common/io/gsqueue/gsqueue.c
@@ -0,0 +1,612 @@
+/*
+ * This file and its contents are supplied under the terms of the
+ * Common Development and Distribution License ("CDDL"), version 1.0.
+ * You may only use this file in accordance with the terms of version
+ * 1.0 of the CDDL.
+ *
+ * A full copy of the text of the CDDL should have accompanied this
+ * source.  A copy of the CDDL is also available via the Internet at
+ * http://www.illumos.org/license/CDDL.
+ */
+
+/*
+ * Copyright (c) 2014 Joyent, Inc.  All rights reserved.
+ */
+
+/*
+ * Serialization queues are a technique used in illumos to provide what's
+ * commonly known as a 'vertical' perimeter. The idea (described a bit in
+ * uts/common/inet/squeue.c) is to provide a means to make sure that message
+ * blocks (mblk_t) are processed in a specific order. Subsystems like ip and vnd
+ * consume these on different policies, ip on a conn_t basis, vnd on a per
+ * device basis, and use this to ensure that only one packet is being processed
+ * at a given time.
+ *
+ * Serialization queues were originally used by ip. As part of that
+ * implementation, many of the details of ip were baked into it. That includes
+ * things like conn_t, ip receive attributes, and the notion of sets. While an
+ * individual serialization queue, or gsqueue_t, is a useful level of
+ * abstraction, it isn't the basis on which monst consumers want to manage them.
+ * Instead, we have the notion of a set of serialization queues. These sets are
+ * DR (CPU Dynamic reconfiguration) aware, and allow consumers to have a
+ * gsqueue_t per CPU to fanout on without managing them all itself. In the
+ * original implementation, this existed, but they were heavily tied into the
+ * infrastructure of IP, and its notion of polling on the underlying MAC
+ * devices.
+ *
+ * The result of that past is a new interface to serialization queues and a
+ * similar, but slightly different, abstraction to sets of these
+ * (gsqueue_set_t).  When designing this there are two different approaches that
+ * one could consider. The first is that the system has one gsqueue_set_t that
+ * the entire world shares, whether IP or some other consumer. The other is that
+ * every consumer has their own set.
+ *
+ * The trade offs between these two failure modes are the pathological failure
+ * modes. There is no guarantee that any two consumers here are equivalent. In
+ * fact, they very likely have very different latency profiles. If they are
+ * being processed in the same queue, that can lead to very odd behaviors. More
+ * generally, if we have a series of processing functions from one consumer
+ * which are generally short, and another which are generally long, that'll
+ * cause undue latency that's harder to observe. If we instead take the approach
+ * that each consumer should have its own set that it fans out over then we
+ * won't end up with the problem that a given serialization queue will have
+ * multiple latency profiles, but instead we'll see cpu contention for the bound
+ * gsqueue_t worker thread. Keep in mind though, that only the gsqueue_t worker
+ * thread is bound and it is in fact possible for it to be processed by other
+ * threads on other CPUs.
+ *
+ * We've opted to go down the second path, so each consumer has its own
+ * independent set of serialization queues that it is bound over.
+ *
+ * Structure Hierarchies
+ * ---------------------
+ *
+ * At the top level, we have a single list of gsqueue_set_t. The gsqueue_set_t
+ * encapsulates all the per-CPU gsqueue_t that exist in the form of
+ * gsqueue_cpu_t.  The gsqueue_cpu_t has been designed such that it could
+ * accommodate more than one gsqueue_t, but today there is a one to one mapping.
+ *
+ * We maintain two different lists of gsqueue_cpu_t, the active and defunct
+ * sets. The active set is maintained in the array `gs_cpus`. There are NCPU
+ * entries available in `gs_cpus` with the total number of currently active cpus
+ * described in `gs_ncpus`. The ordering of `gs_cpus` is unimportant.  When
+ * there is no longer a need for a given binding (see the following section for
+ * more explanation on when this is the case) then we move the entry to the
+ * `gs_defunct` list which is just a singly linked list of gsqueue_cpu_t.
+ *
+ * In addition, each gsqueue_set_t can have a series of callbacks registered
+ * with it. These are described in the following section. Graphically, a given
+ * gsqueue_set_t looks roughly like the following:
+ *
+ *     +---------------+
+ *     | gsqueue_set_t |
+ *     +---------------+
+ *       |    |     |
+ *       |    |     * . . . gs_cpus
+ *       |    |     |
+ *       |    |     |      +-------------------------------------------------+
+ *       |    |     +----->| gsqueue_cpu_t || gsqueue_cpu_t || gsqueue_cpu_t |...
+ *       |    |            +-------------------------------------------------+
+ *       |    |
+ *       |    * . . . gs_defunct
+ *       |    |
+ *       |    |    +---------------+   +---------------+   +---------------+
+ *       |    +--->| gsqueue_cpu_t |-->| gsqueue_cpu_t |-->| gsqueue_cpu_t |...
+ *       |         +---------------+   +---------------+   +---------------+
+ *       * . . . gs_cbs
+ *       |
+ *       |    +--------------+   +--------------+  +--------------+
+ *       +--->| gsqueue_cb_t |-->| gsqueue_cb_t |->| gsqueue_cb_t |...
+ *            +--------------+   +--------------+  +--------------+
+ *
+ * CPU DR, gsqueue_t, and gsqueue_t
+ * --------------------------------
+ *
+ * Recall, that every serialization queue (gsqueue_t or squeue_t) has a worker
+ * thread that may end up doing work. As part of supporting fanout, we have one
+ * gsqueue_t per CPU, and its worker thread is bound to that CPU. Because of
+ * this binding, we need to deal with CPU DR changes.
+ *
+ * The gsqueue driver maintains a single CPU DR callback that is used for the
+ * entire sub-system. We break down CPU DR events into three groups. Offline
+ * events, online events, and events we can ignore. When the first group occurs,
+ * we need to go through every gsqueue_t, find the gsqueue_cpu_t that
+ * corresponds to that processor id, and unbind all of its gsqueue_t's. It's
+ * rather important that we only unbind the gsqueue_t's and not actually destroy
+ * them. When this happens, they could very easily have data queued inside of
+ * them and it's unreasonable to just throw out everything in them at this
+ * point. The data remains intact and service continues uinterrupted.
+ *
+ * When we receive an online event, we do the opposite. We try to find a
+ * gsqueue_cpu_t that previously was bound to this CPU (by leaving its gqc_cpuid
+ * field intact) in the defunct list. If we find one, we remove it from the
+ * defunct list and add it to the active list as well as binding the gsqueue_t
+ * to the CPU in question. If we don't find one, then we create a new one.
+ *
+ * To deal with these kinds of situations, we allow a consumer to register
+ * callbacks for the gsqueue_t that they are interested in. These callbacks will
+ * fire whenever we are handling a topology change. The design of the callbacks
+ * is not that the user can take any administrative action during them, but
+ * rather set something for them to do asynchronously. It is illegal to make any
+ * calls into the gsqueue system while you are in a callback.
+ *
+ * Locking
+ * -------
+ *
+ * The lock ordering here is fairly straightforward. Due to our use of CPU
+ * binding and the CPU DR callbacks, we have an additional lock to consider
+ * cpu_lock. Because of that, the following are the rules for locking:
+ *
+ *
+ *   o If performing binding operations, you must grab cpu_lock. cpu_lock is
+ *     also at the top of the order.
+ *
+ *   o cpu_lock > gsqueue_lock > gsqueue_t`gs_lock > squeue_t`sq_lock
+ *     If you need to take multiple locks, you must take the greatest
+ *     (left-most) one first.
+ */
+
+#include <sys/types.h>
+#include <sys/conf.h>
+#include <sys/stat.h>
+#include <sys/kmem.h>
+#include <sys/stream.h>
+#include <sys/modctl.h>
+#include <sys/cpuvar.h>
+#include <sys/list.h>
+#include <sys/sysmacros.h>
+
+#include <sys/gsqueue.h>
+#include <sys/squeue_impl.h>
+
+typedef struct gsqueue_cb {
+	struct gsqueue_cb *gcb_next;
+	gsqueue_cb_f gcb_func;
+	void *gcb_arg;
+} gsqueue_cb_t;
+
+typedef struct gsqueue_cpu {
+	struct gsqueue_cpu *gqc_next;
+	squeue_t *gqc_head;
+	processorid_t gqc_cpuid;
+} gsqueue_cpu_t;
+
+struct gsqueue_set {
+	list_node_t gs_next;
+	uint_t gs_wwait;
+	pri_t gs_wpri;
+	kmutex_t gs_lock;
+	int gs_ncpus;
+	gsqueue_cpu_t **gs_cpus;
+	gsqueue_cpu_t *gs_defunct;
+	gsqueue_cb_t *gs_cbs;
+};
+
+static kmutex_t gsqueue_lock;
+static list_t gsqueue_list;
+static kmem_cache_t *gsqueue_cb_cache;
+static kmem_cache_t *gsqueue_cpu_cache;
+static kmem_cache_t *gsqueue_set_cache;
+
+static gsqueue_cpu_t *
+gsqueue_cpu_create(uint_t wwait, pri_t wpri, processorid_t cpuid)
+{
+	gsqueue_cpu_t *scp;
+
+	scp = kmem_cache_alloc(gsqueue_cpu_cache, KM_SLEEP);
+
+	scp->gqc_next = NULL;
+	scp->gqc_cpuid = cpuid;
+	scp->gqc_head = squeue_create(wwait, wpri, B_FALSE);
+	scp->gqc_head->sq_state = SQS_DEFAULT;
+	squeue_bind(scp->gqc_head, cpuid);
+
+	return (scp);
+}
+
+static void
+gsqueue_cpu_destroy(gsqueue_cpu_t *scp)
+{
+	squeue_destroy(scp->gqc_head);
+	kmem_cache_free(gsqueue_cpu_cache, scp);
+}
+
+gsqueue_set_t *
+gsqueue_set_create(uint_t wwait, pri_t wpri)
+{
+	int i;
+	gsqueue_set_t *gssp;
+
+	gssp = kmem_cache_alloc(gsqueue_set_cache, KM_SLEEP);
+	gssp->gs_wwait = wwait;
+	gssp->gs_wpri = wpri;
+	gssp->gs_ncpus = 0;
+
+	/*
+	 * We're grabbing CPU lock. Once we let go of it we have to ensure all
+	 * set up of the gsqueue_set_t is complete, as it'll be in there for the
+	 * various CPU DR bits.
+	 */
+	mutex_enter(&cpu_lock);
+
+	for (i = 0; i < NCPU; i++) {
+		gsqueue_cpu_t *scp;
+		cpu_t *cp = cpu_get(i);
+		if (cp != NULL && CPU_ACTIVE(cp) &&
+		    cp->cpu_flags & CPU_EXISTS) {
+			scp = gsqueue_cpu_create(wwait, wpri, cp->cpu_id);
+			gssp->gs_cpus[gssp->gs_ncpus] = scp;
+			gssp->gs_ncpus++;
+		}
+	}
+
+	/* Finally we can add it to our global list and be done */
+	mutex_enter(&gsqueue_lock);
+	list_insert_tail(&gsqueue_list, gssp);
+	mutex_exit(&gsqueue_lock);
+	mutex_exit(&cpu_lock);
+
+	return (gssp);
+}
+
+void
+gsqueue_set_destroy(gsqueue_set_t *gssp)
+{
+	int i;
+	gsqueue_cpu_t *scp;
+
+	/*
+	 * Go through and unbind all of the squeues while cpu_lock is held and
+	 * move them to the defunct list. Once that's done, we don't need to do
+	 * anything else with cpu_lock.
+	 */
+	mutex_enter(&cpu_lock);
+	mutex_enter(&gsqueue_lock);
+	list_remove(&gsqueue_list, gssp);
+	mutex_exit(&gsqueue_lock);
+
+	mutex_enter(&gssp->gs_lock);
+
+	for (i = 0; i < gssp->gs_ncpus; i++) {
+		scp = gssp->gs_cpus[i];
+		squeue_unbind(scp->gqc_head);
+		scp->gqc_next = gssp->gs_defunct;
+		gssp->gs_defunct = scp;
+		gssp->gs_cpus[i] = NULL;
+	}
+	gssp->gs_ncpus = 0;
+
+	mutex_exit(&gssp->gs_lock);
+	mutex_exit(&cpu_lock);
+
+	while (gssp->gs_defunct != NULL) {
+		gsqueue_cpu_t *scp;
+
+		scp = gssp->gs_defunct;
+		gssp->gs_defunct = scp->gqc_next;
+		gsqueue_cpu_destroy(scp);
+	}
+
+	while (gssp->gs_cbs != NULL) {
+		gsqueue_cb_t *cbp;
+
+		cbp = gssp->gs_cbs;
+		gssp->gs_cbs = cbp->gcb_next;
+		kmem_cache_free(gsqueue_cb_cache, cbp);
+	}
+
+	ASSERT(gssp->gs_ncpus == 0);
+	ASSERT(gssp->gs_defunct == NULL);
+	ASSERT(gssp->gs_cbs == NULL);
+	kmem_cache_free(gsqueue_set_cache, gssp);
+}
+
+gsqueue_t *
+gsqueue_set_get(gsqueue_set_t *gssp, uint_t index)
+{
+	squeue_t *sqp;
+	gsqueue_cpu_t *scp;
+
+	mutex_enter(&gssp->gs_lock);
+	scp = gssp->gs_cpus[index % gssp->gs_ncpus];
+	sqp = scp->gqc_head;
+	mutex_exit(&gssp->gs_lock);
+	return ((gsqueue_t *)sqp);
+}
+
+uintptr_t
+gsqueue_set_cb_add(gsqueue_set_t *gssp, gsqueue_cb_f cb, void *arg)
+{
+	gsqueue_cb_t *cbp;
+
+	cbp = kmem_cache_alloc(gsqueue_cb_cache, KM_SLEEP);
+	cbp->gcb_func = cb;
+	cbp->gcb_arg = arg;
+
+	mutex_enter(&gssp->gs_lock);
+	cbp->gcb_next = gssp->gs_cbs;
+	gssp->gs_cbs = cbp;
+	mutex_exit(&gssp->gs_lock);
+	return ((uintptr_t)cbp);
+}
+
+int
+gsqueue_set_cb_remove(gsqueue_set_t *gssp, uintptr_t id)
+{
+	gsqueue_cb_t *cbp, *prev;
+	mutex_enter(&gssp->gs_lock);
+	cbp = gssp->gs_cbs;
+	prev = NULL;
+	while (cbp != NULL) {
+		if ((uintptr_t)cbp != id) {
+			prev = cbp;
+			cbp = cbp->gcb_next;
+			continue;
+		}
+
+		if (prev == NULL) {
+			gssp->gs_cbs = cbp->gcb_next;
+		} else {
+			prev->gcb_next = cbp->gcb_next;
+		}
+
+		mutex_exit(&gssp->gs_lock);
+		kmem_cache_free(gsqueue_cb_cache, cbp);
+		return (0);
+	}
+	mutex_exit(&gssp->gs_lock);
+	return (-1);
+}
+
+void
+gsqueue_enter_one(gsqueue_t *gsp, mblk_t *mp, gsqueue_proc_f func, void *arg,
+    int flags, uint8_t tag)
+{
+	squeue_t *sqp = (squeue_t *)gsp;
+
+	ASSERT(mp->b_next == NULL);
+	ASSERT(mp->b_prev == NULL);
+	mp->b_queue = (queue_t *)func;
+	mp->b_prev = arg;
+	sqp->sq_enter(sqp, mp, mp, 1, NULL, flags, tag);
+}
+
+static void
+gsqueue_notify(gsqueue_set_t *gssp, squeue_t *sqp, boolean_t online)
+{
+	gsqueue_cb_t *cbp;
+
+	ASSERT(MUTEX_HELD(&gssp->gs_lock));
+	cbp = gssp->gs_cbs;
+	while (cbp != NULL) {
+		cbp->gcb_func(gssp, (gsqueue_t *)sqp, cbp->gcb_arg, online);
+		cbp = cbp->gcb_next;
+	}
+
+}
+
+/*
+ * When we online a processor we need to go through and either bind a defunct
+ * squeue or create a new one. We'll try to reuse a gsqueue_cpu_t from the
+ * defunct list that used to be on that processor. If no such gsqueue_cpu_t
+ * exists, then we'll create a new one. We'd rather avoid taking over an
+ * existing defunct one that used to be on another CPU, as its not unreasonable
+ * to believe that its CPU will come back. More CPUs are offlined and onlined by
+ * the administrator or by creating cpu sets than actually get offlined by FMA.
+ */
+static void
+gsqueue_handle_online(processorid_t id)
+{
+	gsqueue_set_t *gssp;
+
+	ASSERT(MUTEX_HELD(&cpu_lock));
+	mutex_enter(&gsqueue_lock);
+	for (gssp = list_head(&gsqueue_list); gssp != NULL;
+	    gssp = list_next(&gsqueue_list, gssp)) {
+		gsqueue_cpu_t *scp;
+
+		mutex_enter(&gssp->gs_lock);
+		scp = gssp->gs_defunct;
+		while (scp != NULL) {
+			if (scp->gqc_cpuid == id)
+				break;
+			scp = scp->gqc_next;
+		}
+
+		if (scp == NULL) {
+			scp = gsqueue_cpu_create(gssp->gs_wwait,
+			    gssp->gs_wpri, id);
+		} else {
+			squeue_bind(scp->gqc_head, id);
+		}
+		ASSERT(gssp->gs_ncpus < NCPU);
+		gssp->gs_cpus[gssp->gs_ncpus] = scp;
+		gssp->gs_ncpus++;
+		gsqueue_notify(gssp, scp->gqc_head, B_TRUE);
+		mutex_exit(&gssp->gs_lock);
+	}
+	mutex_exit(&gsqueue_lock);
+}
+
+static void
+gsqueue_handle_offline(processorid_t id)
+{
+	gsqueue_set_t *gssp;
+
+	ASSERT(MUTEX_HELD(&cpu_lock));
+	mutex_enter(&gsqueue_lock);
+	for (gssp = list_head(&gsqueue_list); gssp != NULL;
+	    gssp = list_next(&gsqueue_list, gssp)) {
+		int i;
+		gsqueue_cpu_t *scp = NULL;
+
+		mutex_enter(&gssp->gs_lock);
+		for (i = 0; i < gssp->gs_ncpus; i++) {
+			if (gssp->gs_cpus[i]->gqc_cpuid == id) {
+				scp = gssp->gs_cpus[i];
+				break;
+			}
+		}
+
+		if (scp != NULL) {
+			squeue_unbind(scp->gqc_head);
+			scp->gqc_next = gssp->gs_defunct;
+			gssp->gs_defunct = scp;
+			gssp->gs_cpus[i] = gssp->gs_cpus[gssp->gs_ncpus-1];
+			gssp->gs_ncpus--;
+			gsqueue_notify(gssp, scp->gqc_head, B_FALSE);
+		}
+		mutex_exit(&gssp->gs_lock);
+	}
+	mutex_exit(&gsqueue_lock);
+}
+
+/* ARGSUSED */
+static int
+gsqueue_cpu_setup(cpu_setup_t what, int id, void *unused)
+{
+	cpu_t *cp;
+
+	ASSERT(MUTEX_HELD(&cpu_lock));
+	cp = cpu_get(id);
+	switch (what) {
+	case CPU_CONFIG:
+	case CPU_ON:
+	case CPU_INIT:
+	case CPU_CPUPART_IN:
+		if (cp != NULL && CPU_ACTIVE(cp) && cp->cpu_flags & CPU_EXISTS)
+			gsqueue_handle_online(cp->cpu_id);
+		break;
+	case CPU_UNCONFIG:
+	case CPU_OFF:
+	case CPU_CPUPART_OUT:
+		gsqueue_handle_offline(cp->cpu_id);
+		break;
+	default:
+		break;
+	}
+
+	return (0);
+}
+
+
+/* ARGSUSED */
+static int
+gsqueue_set_cache_construct(void *buf, void *arg, int kmflags)
+{
+	gsqueue_set_t *gssp = buf;
+
+	gssp->gs_cpus = kmem_alloc(sizeof (gsqueue_cpu_t *) * NCPU, kmflags);
+	if (gssp->gs_cpus == NULL)
+		return (-1);
+
+	mutex_init(&gssp->gs_lock, NULL, MUTEX_DRIVER, NULL);
+	gssp->gs_ncpus = 0;
+	gssp->gs_defunct = NULL;
+	gssp->gs_cbs = NULL;
+
+	return (0);
+}
+
+static void
+gsqueue_set_cache_destruct(void *buf, void *arg)
+{
+	gsqueue_set_t *gssp = buf;
+
+	kmem_free(gssp->gs_cpus, sizeof (gsqueue_cpu_t *) * NCPU);
+	gssp->gs_cpus = NULL;
+	mutex_destroy(&gssp->gs_lock);
+}
+
+static void
+gsqueue_ddiinit(void)
+{
+	list_create(&gsqueue_list, sizeof (gsqueue_set_t),
+	    offsetof(gsqueue_set_t, gs_next));
+	mutex_init(&gsqueue_lock, NULL, MUTEX_DRIVER, NULL);
+
+	gsqueue_cb_cache = kmem_cache_create("gsqueue_cb_cache",
+	    sizeof (gsqueue_cb_t),
+	    0, NULL, NULL, NULL, NULL, NULL, 0);
+	gsqueue_cpu_cache = kmem_cache_create("gsqueue_cpu_cache",
+	    sizeof (gsqueue_cpu_t),
+	    0, NULL, NULL, NULL, NULL, NULL, 0);
+	gsqueue_set_cache = kmem_cache_create("squeue_set_cache",
+	    sizeof (gsqueue_set_t),
+	    0, gsqueue_set_cache_construct, gsqueue_set_cache_destruct,
+	    NULL, NULL, NULL, 0);
+
+
+	mutex_enter(&cpu_lock);
+	register_cpu_setup_func(gsqueue_cpu_setup, NULL);
+	mutex_exit(&cpu_lock);
+}
+
+static int
+gsqueue_ddifini(void)
+{
+	mutex_enter(&gsqueue_lock);
+	if (list_is_empty(&gsqueue_list) == 0) {
+		mutex_exit(&gsqueue_lock);
+		return (EBUSY);
+	}
+	list_destroy(&gsqueue_list);
+	mutex_exit(&gsqueue_lock);
+
+	mutex_enter(&cpu_lock);
+	register_cpu_setup_func(gsqueue_cpu_setup, NULL);
+	mutex_exit(&cpu_lock);
+
+	kmem_cache_destroy(gsqueue_set_cache);
+	kmem_cache_destroy(gsqueue_cpu_cache);
+	kmem_cache_destroy(gsqueue_cb_cache);
+
+	mutex_destroy(&gsqueue_lock);
+
+	return (0);
+}
+
+static struct modlmisc		gsqueue_modmisc = {
+	&mod_miscops,
+	"gsqueue"
+};
+
+static struct modlinkage	gsqueue_modlinkage = {
+	MODREV_1,
+	&gsqueue_modmisc,
+	NULL
+};
+
+int
+_init(void)
+{
+	int ret;
+
+	gsqueue_ddiinit();
+	if ((ret = mod_install(&gsqueue_modlinkage)) != 0) {
+		VERIFY(gsqueue_ddifini() == 0);
+		return (ret);
+	}
+
+	return (ret);
+}
+
+int
+_info(struct modinfo *modinfop)
+{
+	return (mod_info(&gsqueue_modlinkage, modinfop));
+}
+
+int
+_fini(void)
+{
+	int ret;
+
+	if ((ret = gsqueue_ddifini()) != 0)
+		return (ret);
+
+	if ((ret = mod_remove(&gsqueue_modlinkage)) != 0)
+		return (ret);
+
+	return (0);
+}
diff --git a/usr/src/uts/common/io/inotify.c b/usr/src/uts/common/io/inotify.c
new file mode 100644
index 0000000000..c5c7b379e3
--- /dev/null
+++ b/usr/src/uts/common/io/inotify.c
@@ -0,0 +1,1508 @@
+/*
+ * This file and its contents are supplied under the terms of the
+ * Common Development and Distribution License ("CDDL"), version 1.0.
+ * You may only use this file in accordance with the terms of version
+ * 1.0 of the CDDL.
+ *
+ * A full copy of the text of the CDDL should have accompanied this
+ * source.  A copy of the CDDL is also available via the Internet at
+ * http://www.illumos.org/license/CDDL.
+ */
+
+/*
+ * Copyright (c) 2015 Joyent, Inc.  All rights reserved.
+ */
+
+/*
+ * Support for the inotify facility, a Linux-borne facility for asynchronous
+ * notification of certain events on specified files or directories.  Our
+ * implementation broadly leverages the file event monitoring facility, and
+ * would actually be quite straightforward were it not for a very serious
+ * blunder in the inotify interface:  in addition to allowing for one to be
+ * notified on events on a particular file or directory, inotify also allows
+ * for one to be notified on certain events on files _within_ a watched
+ * directory -- even though those events have absolutely nothing to do with
+ * the directory itself.  This leads to all sorts of madness because file
+ * operations are (of course) not undertaken on paths but rather on open
+ * files -- and the relationships between open files and the paths that resolve
+ * to those files are neither static nor isomorphic.  We implement this
+ * concept by having _child watches_ when directories are watched with events
+ * in IN_CHILD_EVENTS.  We add child watches when a watch on a directory is
+ * first added, and we modify those child watches dynamically as files are
+ * created, deleted, moved into or moved out of the specified directory.  This
+ * mechanism works well, absent hard links.  Hard links, unfortunately, break
+ * this rather badly, and the user is warned that watches on directories that
+ * have multiple directory entries referring to the same file may behave
+ * unexpectedly.
+ */
+
+#include <sys/ddi.h>
+#include <sys/sunddi.h>
+#include <sys/inotify.h>
+#include <sys/fem.h>
+#include <sys/conf.h>
+#include <sys/stat.h>
+#include <sys/vfs_opreg.h>
+#include <sys/vmem.h>
+#include <sys/avl.h>
+#include <sys/sysmacros.h>
+#include <sys/cyclic.h>
+#include <sys/filio.h>
+
+struct inotify_state;
+struct inotify_kevent;
+
+typedef struct inotify_watch inotify_watch_t;
+typedef struct inotify_state inotify_state_t;
+typedef struct inotify_kevent inotify_kevent_t;
+
+struct inotify_watch {
+	kmutex_t inw_lock;			/* lock protecting ref count */
+	int inw_refcnt;				/* reference count */
+	uint8_t inw_zombie:1;			/* boolean: is zombie */
+	uint8_t inw_fired:1;			/* boolean: fired one-shot */
+	uint8_t inw_active:1;			/* boolean: watch is active */
+	uint8_t inw_orphaned:1;			/* boolean: orphaned */
+	kcondvar_t inw_cv;			/* condvar for zombifier */
+	uint32_t inw_mask;			/* mask of watch */
+	int32_t inw_wd;				/* watch descriptor */
+	vnode_t *inw_vp;			/* underlying vnode */
+	inotify_watch_t *inw_parent;		/* parent, if a child */
+	avl_node_t inw_byvp;			/* watches by vnode */
+	avl_node_t inw_bywd;			/* watches by descriptor */
+	avl_tree_t inw_children;		/* children, if a parent */
+	char *inw_name;				/* name, if a child */
+	list_node_t inw_orphan;			/* orphan list */
+	cred_t *inw_cred;			/* cred, if orphaned */
+	inotify_state_t *inw_state;		/* corresponding state */
+};
+
+struct inotify_kevent {
+	inotify_kevent_t *ine_next;		/* next event in queue */
+	struct inotify_event ine_event;		/* event (variable size) */
+};
+
+#define	INOTIFY_EVENT_LENGTH(ev) \
+	(sizeof (inotify_kevent_t) + (ev)->ine_event.len)
+
+struct inotify_state {
+	kmutex_t ins_lock;			/* lock protecting state */
+	avl_tree_t ins_byvp;			/* watches by vnode */
+	avl_tree_t ins_bywd;			/* watches by descriptor */
+	vmem_t *ins_wds;			/* watch identifier arena */
+	int ins_maxwatches;			/* maximum number of watches */
+	int ins_maxevents;			/* maximum number of events */
+	int ins_nevents;			/* current # of events */
+	int32_t ins_size;			/* total size of events */
+	inotify_kevent_t *ins_head;		/* head of event queue */
+	inotify_kevent_t *ins_tail;		/* tail of event queue */
+	pollhead_t ins_pollhd;			/* poll head */
+	kcondvar_t ins_cv;			/* condvar for reading */
+	list_t ins_orphans;			/* orphan list */
+	cyclic_id_t ins_cleaner;		/* cyclic for cleaning */
+	inotify_watch_t *ins_zombies;		/* zombie watch list */
+	cred_t *ins_cred;			/* creator's credentials */
+	inotify_state_t *ins_next;		/* next state on global list */
+};
+
+/*
+ * Tunables (exported read-only in lx-branded zones via /proc).
+ */
+int	inotify_maxwatches = 8192;		/* max watches per instance */
+int	inotify_maxevents = 16384;		/* max events */
+int	inotify_maxinstances = 128;		/* max instances per user */
+
+/*
+ * Internal global variables.
+ */
+static kmutex_t		inotify_lock;		/* lock protecting state */
+static dev_info_t	*inotify_devi;		/* device info */
+static fem_t		*inotify_femp;		/* FEM pointer */
+static vmem_t		*inotify_minor;		/* minor number arena */
+static void		*inotify_softstate;	/* softstate pointer */
+static inotify_state_t	*inotify_state;		/* global list if state */
+
+static void inotify_watch_event(inotify_watch_t *, uint64_t, char *);
+static void inotify_watch_insert(inotify_watch_t *, vnode_t *, char *);
+static void inotify_watch_delete(inotify_watch_t *, uint32_t);
+static void inotify_watch_remove(inotify_state_t *state,
+	inotify_watch_t *watch);
+
+static int
+inotify_fop_close(femarg_t *vf, int flag, int count, offset_t offset,
+    cred_t *cr, caller_context_t *ct)
+{
+	inotify_watch_t *watch = vf->fa_fnode->fn_available;
+	int rval;
+
+	if ((rval = vnext_close(vf, flag, count, offset, cr, ct)) == 0) {
+		inotify_watch_event(watch, flag & FWRITE ?
+		    IN_CLOSE_WRITE : IN_CLOSE_NOWRITE, NULL);
+	}
+
+	return (rval);
+}
+
+static int
+inotify_fop_create(femarg_t *vf, char *name, vattr_t *vap, vcexcl_t excl,
+    int mode, vnode_t **vpp, cred_t *cr, int flag, caller_context_t *ct,
+    vsecattr_t *vsecp)
+{
+	inotify_watch_t *watch = vf->fa_fnode->fn_available;
+	int rval;
+
+	if ((rval = vnext_create(vf, name, vap, excl, mode,
+	    vpp, cr, flag, ct, vsecp)) == 0) {
+		inotify_watch_insert(watch, *vpp, name);
+		inotify_watch_event(watch, IN_CREATE, name);
+	}
+
+	return (rval);
+}
+
+static int
+inotify_fop_link(femarg_t *vf, vnode_t *svp, char *tnm, cred_t *cr,
+    caller_context_t *ct, int flags)
+{
+	inotify_watch_t *watch = vf->fa_fnode->fn_available;
+	int rval;
+
+	if ((rval = vnext_link(vf, svp, tnm, cr, ct, flags)) == 0) {
+		inotify_watch_insert(watch, svp, tnm);
+		inotify_watch_event(watch, IN_CREATE, tnm);
+	}
+
+	return (rval);
+}
+
+static int
+inotify_fop_mkdir(femarg_t *vf, char *name, vattr_t *vap, vnode_t **vpp,
+    cred_t *cr, caller_context_t *ct, int flags, vsecattr_t *vsecp)
+{
+	inotify_watch_t *watch = vf->fa_fnode->fn_available;
+	int rval;
+
+	if ((rval = vnext_mkdir(vf, name, vap, vpp, cr,
+	    ct, flags, vsecp)) == 0) {
+		inotify_watch_insert(watch, *vpp, name);
+		inotify_watch_event(watch, IN_CREATE | IN_ISDIR, name);
+	}
+
+	return (rval);
+}
+
+static int
+inotify_fop_open(femarg_t *vf, int mode, cred_t *cr, caller_context_t *ct)
+{
+	inotify_watch_t *watch = vf->fa_fnode->fn_available;
+	int rval;
+
+	if ((rval = vnext_open(vf, mode, cr, ct)) == 0)
+		inotify_watch_event(watch, IN_OPEN, NULL);
+
+	return (rval);
+}
+
+static int
+inotify_fop_read(femarg_t *vf, struct uio *uiop, int ioflag, struct cred *cr,
+    caller_context_t *ct)
+{
+	inotify_watch_t *watch = vf->fa_fnode->fn_available;
+	int rval = vnext_read(vf, uiop, ioflag, cr, ct);
+	inotify_watch_event(watch, IN_ACCESS, NULL);
+
+	return (rval);
+}
+
+static int
+inotify_fop_readdir(femarg_t *vf, uio_t *uiop, cred_t *cr, int *eofp,
+    caller_context_t *ct, int flags)
+{
+	inotify_watch_t *watch = vf->fa_fnode->fn_available;
+	int rval = vnext_readdir(vf, uiop, cr, eofp, ct, flags);
+	inotify_watch_event(watch, IN_ACCESS | IN_ISDIR, NULL);
+
+	return (rval);
+}
+
+int
+inotify_fop_remove(femarg_t *vf, char *nm, cred_t *cr, caller_context_t *ct,
+    int flags)
+{
+	inotify_watch_t *watch = vf->fa_fnode->fn_available;
+	int rval;
+
+	if ((rval = vnext_remove(vf, nm, cr, ct, flags)) == 0)
+		inotify_watch_event(watch, IN_DELETE, nm);
+
+	return (rval);
+}
+
+int
+inotify_fop_rmdir(femarg_t *vf, char *nm, vnode_t *cdir, cred_t *cr,
+    caller_context_t *ct, int flags)
+{
+	inotify_watch_t *watch = vf->fa_fnode->fn_available;
+	int rval;
+
+	if ((rval = vnext_rmdir(vf, nm, cdir, cr, ct, flags)) == 0)
+		inotify_watch_event(watch, IN_DELETE | IN_ISDIR, nm);
+
+	return (rval);
+}
+
+static int
+inotify_fop_setattr(femarg_t *vf, vattr_t *vap, int flags, cred_t *cr,
+    caller_context_t *ct)
+{
+	inotify_watch_t *watch = vf->fa_fnode->fn_available;
+	int rval;
+
+	if ((rval = vnext_setattr(vf, vap, flags, cr, ct)) == 0)
+		inotify_watch_event(watch, IN_ATTRIB, NULL);
+
+	return (rval);
+}
+
+static int
+inotify_fop_write(femarg_t *vf, struct uio *uiop, int ioflag, struct cred *cr,
+    caller_context_t *ct)
+{
+	inotify_watch_t *watch = vf->fa_fnode->fn_available;
+	int rval = vnext_write(vf, uiop, ioflag, cr, ct);
+	inotify_watch_event(watch, IN_MODIFY, NULL);
+
+	return (rval);
+}
+
+static int
+inotify_fop_vnevent(femarg_t *vf, vnevent_t vnevent, vnode_t *dvp, char *name,
+    caller_context_t *ct)
+{
+	inotify_watch_t *watch = vf->fa_fnode->fn_available;
+
+	switch (vnevent) {
+	case VE_RENAME_SRC:
+		inotify_watch_event(watch, IN_MOVE_SELF, NULL);
+		inotify_watch_delete(watch, IN_MOVE_SELF);
+		break;
+	case VE_REMOVE:
+		/*
+		 * Linux will apparently fire an IN_ATTRIB event when the link
+		 * count changes (including when it drops to 0 on a remove).
+		 * This is merely somewhat odd; what is amazing is that this
+		 * IN_ATTRIB event is not visible on an inotify watch on the
+		 * parent directory.  (IN_ATTRIB events are normally sent to
+		 * watches on the parent directory).  While it's hard to
+		 * believe that this constitutes desired semantics, ltp
+		 * unfortunately tests this case (if implicitly); in the name
+		 * of bug-for-bug compatibility, we fire IN_ATTRIB iff we are
+		 * explicitly watching the file that has been removed.
+		 */
+		if (watch->inw_parent == NULL)
+			inotify_watch_event(watch, IN_ATTRIB, NULL);
+
+		/*FALLTHROUGH*/
+	case VE_RENAME_DEST:
+		inotify_watch_event(watch, IN_DELETE_SELF, NULL);
+		inotify_watch_delete(watch, IN_DELETE_SELF);
+		break;
+	case VE_RMDIR:
+		/*
+		 * It seems that IN_ISDIR should really be OR'd in here, but
+		 * Linux doesn't seem to do that in this case; for the sake of
+		 * bug-for-bug compatibility, we don't do it either.
+		 */
+		inotify_watch_event(watch, IN_DELETE_SELF, NULL);
+		inotify_watch_delete(watch, IN_DELETE_SELF);
+		break;
+	case VE_CREATE:
+		inotify_watch_event(watch, IN_MODIFY | IN_ATTRIB, NULL);
+		break;
+	case VE_LINK:
+		inotify_watch_event(watch, IN_ATTRIB, NULL);
+		break;
+	case VE_RENAME_SRC_DIR:
+		inotify_watch_event(watch, IN_MOVED_FROM, name);
+		break;
+	case VE_RENAME_DEST_DIR:
+		if (name == NULL)
+			name = dvp->v_path;
+
+		inotify_watch_insert(watch, dvp, name);
+		inotify_watch_event(watch, IN_MOVED_TO, name);
+		break;
+	case VE_SUPPORT:
+	case VE_MOUNTEDOVER:
+	case VE_TRUNCATE:
+		break;
+	}
+
+	return (vnext_vnevent(vf, vnevent, dvp, name, ct));
+}
+
+const fs_operation_def_t inotify_vnodesrc_template[] = {
+	VOPNAME_CLOSE,		{ .femop_close = inotify_fop_close },
+	VOPNAME_CREATE,		{ .femop_create = inotify_fop_create },
+	VOPNAME_LINK,		{ .femop_link = inotify_fop_link },
+	VOPNAME_MKDIR,		{ .femop_mkdir = inotify_fop_mkdir },
+	VOPNAME_OPEN,		{ .femop_open = inotify_fop_open },
+	VOPNAME_READ,		{ .femop_read = inotify_fop_read },
+	VOPNAME_READDIR,	{ .femop_readdir = inotify_fop_readdir },
+	VOPNAME_REMOVE,		{ .femop_remove = inotify_fop_remove },
+	VOPNAME_RMDIR,		{ .femop_rmdir = inotify_fop_rmdir },
+	VOPNAME_SETATTR,	{ .femop_setattr = inotify_fop_setattr },
+	VOPNAME_WRITE,		{ .femop_write = inotify_fop_write },
+	VOPNAME_VNEVENT,	{ .femop_vnevent = inotify_fop_vnevent },
+	NULL, NULL
+};
+
+static int
+inotify_watch_cmpwd(inotify_watch_t *lhs, inotify_watch_t *rhs)
+{
+	if (lhs->inw_wd < rhs->inw_wd)
+		return (-1);
+
+	if (lhs->inw_wd > rhs->inw_wd)
+		return (1);
+
+	return (0);
+}
+
+static int
+inotify_watch_cmpvp(inotify_watch_t *lhs, inotify_watch_t *rhs)
+{
+	uintptr_t lvp = (uintptr_t)lhs->inw_vp, rvp = (uintptr_t)rhs->inw_vp;
+
+	if (lvp < rvp)
+		return (-1);
+
+	if (lvp > rvp)
+		return (1);
+
+	return (0);
+}
+
+static void
+inotify_watch_hold(inotify_watch_t *watch)
+{
+	mutex_enter(&watch->inw_lock);
+	VERIFY(watch->inw_refcnt > 0);
+	watch->inw_refcnt++;
+	mutex_exit(&watch->inw_lock);
+}
+
+static void
+inotify_watch_release(inotify_watch_t *watch)
+{
+	mutex_enter(&watch->inw_lock);
+	VERIFY(watch->inw_refcnt > 1);
+
+	if (--watch->inw_refcnt == 1 && watch->inw_zombie) {
+		/*
+		 * We're down to our last reference; kick anyone that might be
+		 * waiting.
+		 */
+		cv_signal(&watch->inw_cv);
+	}
+
+	mutex_exit(&watch->inw_lock);
+}
+
+static void
+inotify_watch_event(inotify_watch_t *watch, uint64_t mask, char *name)
+{
+	inotify_kevent_t *event, *tail;
+	inotify_state_t *state = watch->inw_state;
+	uint32_t wd = watch->inw_wd, cookie = 0, len;
+	int align = sizeof (uintptr_t) - 1;
+	boolean_t removal = mask & IN_REMOVAL ? B_TRUE : B_FALSE;
+	inotify_watch_t *source = watch;
+
+	if (!(mask &= watch->inw_mask) || mask == IN_ISDIR)
+		return;
+
+	if (watch->inw_parent != NULL) {
+		/*
+		 * This is an event on the child; if this isn't a valid child
+		 * event, return.  Otherwise, we move our watch to be our
+		 * parent (which we know is around because we have a hold on
+		 * it) and continue.
+		 */
+		if (!(mask & IN_CHILD_EVENTS))
+			return;
+
+		name = watch->inw_name;
+		watch = watch->inw_parent;
+	}
+
+	if (!removal) {
+		mutex_enter(&state->ins_lock);
+
+		if (watch->inw_zombie ||
+		    watch->inw_fired || !watch->inw_active) {
+			mutex_exit(&state->ins_lock);
+			return;
+		}
+	} else {
+		if (!watch->inw_active)
+			return;
+
+		VERIFY(MUTEX_HELD(&state->ins_lock));
+	}
+
+	/*
+	 * If this is an operation on a directory and it's a child event
+	 * (event if it's not on a child), we specify IN_ISDIR.
+	 */
+	if (source->inw_vp->v_type == VDIR && (mask & IN_CHILD_EVENTS))
+		mask |= IN_ISDIR;
+
+	if (mask & (IN_MOVED_FROM | IN_MOVED_TO))
+		cookie = (uint32_t)curthread->t_did;
+
+	if (state->ins_nevents >= state->ins_maxevents) {
+		/*
+		 * We're at our maximum number of events -- turn our event
+		 * into an IN_Q_OVERFLOW event, which will be coalesced if
+		 * it's already the tail event.
+		 */
+		mask = IN_Q_OVERFLOW;
+		wd = (uint32_t)-1;
+		cookie = 0;
+		len = 0;
+	}
+
+	if ((tail = state->ins_tail) != NULL && tail->ine_event.wd == wd &&
+	    tail->ine_event.mask == mask && tail->ine_event.cookie == cookie &&
+	    ((tail->ine_event.len == 0 && len == 0) ||
+	    (name != NULL && tail->ine_event.len != 0 &&
+	    strcmp(tail->ine_event.name, name) == 0))) {
+		/*
+		 * This is an implicitly coalesced event; we're done.
+		 */
+		if (!removal)
+			mutex_exit(&state->ins_lock);
+		return;
+	}
+
+	if (name != NULL) {
+		if ((len = strlen(name) + 1) & align)
+			len += (align + 1) - (len & align);
+	} else {
+		len = 0;
+	}
+
+	event = kmem_zalloc(sizeof (inotify_kevent_t) + len, KM_SLEEP);
+	event->ine_event.wd = wd;
+	event->ine_event.mask = (uint32_t)mask;
+	event->ine_event.cookie = cookie;
+	event->ine_event.len = len;
+
+	if (name != NULL)
+		strcpy(event->ine_event.name, name);
+
+	if (tail != NULL) {
+		tail->ine_next = event;
+	} else {
+		VERIFY(state->ins_head == NULL);
+		state->ins_head = event;
+		cv_broadcast(&state->ins_cv);
+	}
+
+	state->ins_tail = event;
+	state->ins_nevents++;
+	state->ins_size += sizeof (event->ine_event) + len;
+
+	if (removal)
+		return;
+
+	if ((watch->inw_mask & IN_ONESHOT) && !watch->inw_fired) {
+		/*
+		 * If this is a one-shot, we need to remove the watch.  (Note
+		 * that this will recurse back into inotify_watch_event() to
+		 * fire the IN_IGNORED event -- but with "removal" set.)
+		 */
+		watch->inw_fired = 1;
+		inotify_watch_remove(state, watch);
+	}
+
+	mutex_exit(&state->ins_lock);
+	pollwakeup(&state->ins_pollhd, POLLRDNORM | POLLIN);
+}
+
+/*
+ * Destroy a watch.  By the time we're in here, the watch must have exactly
+ * one reference.
+ */
+static void
+inotify_watch_destroy(inotify_watch_t *watch)
+{
+	VERIFY(MUTEX_HELD(&watch->inw_lock));
+
+	if (watch->inw_name != NULL)
+		kmem_free(watch->inw_name, strlen(watch->inw_name) + 1);
+
+	kmem_free(watch, sizeof (inotify_watch_t));
+}
+
+/*
+ * Zombify a watch.  By the time we come in here, it must be true that the
+ * watch has already been fem_uninstall()'d -- the only reference should be
+ * in the state's data structure.  If we can get away with freeing it, we'll
+ * do that -- but if the reference count is greater than one due to an active
+ * vnode operation, we'll put this watch on the zombie list on the state
+ * structure.
+ */
+static void
+inotify_watch_zombify(inotify_watch_t *watch)
+{
+	inotify_state_t *state = watch->inw_state;
+
+	VERIFY(MUTEX_HELD(&state->ins_lock));
+	VERIFY(!watch->inw_zombie);
+
+	watch->inw_zombie = 1;
+
+	if (watch->inw_parent != NULL) {
+		inotify_watch_release(watch->inw_parent);
+	} else {
+		avl_remove(&state->ins_byvp, watch);
+		avl_remove(&state->ins_bywd, watch);
+		vmem_free(state->ins_wds, (void *)(uintptr_t)watch->inw_wd, 1);
+		watch->inw_wd = -1;
+	}
+
+	mutex_enter(&watch->inw_lock);
+
+	if (watch->inw_refcnt == 1) {
+		/*
+		 * There are no operations in flight and there is no way
+		 * for anyone to discover this watch -- we can destroy it.
+		 */
+		inotify_watch_destroy(watch);
+	} else {
+		/*
+		 * There are operations in flight; we will need to enqueue
+		 * this for later destruction.
+		 */
+		watch->inw_parent = state->ins_zombies;
+		state->ins_zombies = watch;
+		mutex_exit(&watch->inw_lock);
+	}
+}
+
+static inotify_watch_t *
+inotify_watch_add(inotify_state_t *state, inotify_watch_t *parent,
+    const char *name, vnode_t *vp, uint32_t mask)
+{
+	inotify_watch_t *watch;
+	int err;
+
+	VERIFY(MUTEX_HELD(&state->ins_lock));
+
+	watch = kmem_zalloc(sizeof (inotify_watch_t), KM_SLEEP);
+
+	watch->inw_vp = vp;
+	watch->inw_mask = mask;
+	watch->inw_state = state;
+	watch->inw_refcnt = 1;
+
+	if (parent == NULL) {
+		watch->inw_wd = (int)(uintptr_t)vmem_alloc(state->ins_wds,
+		    1, VM_BESTFIT | VM_SLEEP);
+		avl_add(&state->ins_byvp, watch);
+		avl_add(&state->ins_bywd, watch);
+
+		avl_create(&watch->inw_children,
+		    (int(*)(const void *, const void *))inotify_watch_cmpvp,
+		    sizeof (inotify_watch_t),
+		    offsetof(inotify_watch_t, inw_byvp));
+	} else {
+		VERIFY(name != NULL);
+		inotify_watch_hold(parent);
+		watch->inw_mask &= IN_CHILD_EVENTS;
+		watch->inw_parent = parent;
+		watch->inw_name = kmem_alloc(strlen(name) + 1, KM_SLEEP);
+		strcpy(watch->inw_name, name);
+
+		avl_add(&parent->inw_children, watch);
+	}
+
+	/*
+	 * Add our monitor to the vnode.  We must not have the watch lock held
+	 * when we do this, as it will immediately hold our watch.
+	 */
+	err = fem_install(vp, inotify_femp, watch, OPARGUNIQ,
+	    (void (*)(void *))inotify_watch_hold,
+	    (void (*)(void *))inotify_watch_release);
+
+	VERIFY(err == 0);
+
+	return (watch);
+}
+
+/*
+ * Remove a (non-child) watch.  This is called from either synchronous context
+ * via inotify_rm_watch() or monitor context via either a vnevent or a
+ * one-shot.
+ */
+static void
+inotify_watch_remove(inotify_state_t *state, inotify_watch_t *watch)
+{
+	inotify_watch_t *child;
+	int err;
+
+	VERIFY(MUTEX_HELD(&state->ins_lock));
+	VERIFY(watch->inw_parent == NULL);
+
+	err = fem_uninstall(watch->inw_vp, inotify_femp, watch);
+	VERIFY(err == 0);
+
+	/*
+	 * If we have children, we're going to remove them all and set them
+	 * all to be zombies.
+	 */
+	while ((child = avl_first(&watch->inw_children)) != NULL) {
+		VERIFY(child->inw_parent == watch);
+		avl_remove(&watch->inw_children, child);
+
+		err = fem_uninstall(child->inw_vp, inotify_femp, child);
+		VERIFY(err == 0);
+
+		/*
+		 * If this child watch has been orphaned, remove it from the
+		 * state's list of orphans.
+		 */
+		if (child->inw_orphaned) {
+			list_remove(&state->ins_orphans, child);
+			crfree(child->inw_cred);
+		}
+
+		VN_RELE(child->inw_vp);
+
+		/*
+		 * We're down (or should be down) to a single reference to
+		 * this child watch; it's safe to zombify it.
+		 */
+		inotify_watch_zombify(child);
+	}
+
+	inotify_watch_event(watch, IN_IGNORED | IN_REMOVAL, NULL);
+	VN_RELE(watch->inw_vp);
+
+	/*
+	 * It's now safe to zombify the watch -- we know that the only reference
+	 * can come from operations in flight.
+	 */
+	inotify_watch_zombify(watch);
+}
+
+/*
+ * Delete a watch.  Should only be called from VOP context.
+ */
+static void
+inotify_watch_delete(inotify_watch_t *watch, uint32_t event)
+{
+	inotify_state_t *state = watch->inw_state;
+	inotify_watch_t cmp = { .inw_vp = watch->inw_vp }, *parent;
+	int err;
+
+	if (event != IN_DELETE_SELF && !(watch->inw_mask & IN_CHILD_EVENTS))
+		return;
+
+	mutex_enter(&state->ins_lock);
+
+	if (watch->inw_zombie) {
+		mutex_exit(&state->ins_lock);
+		return;
+	}
+
+	if ((parent = watch->inw_parent) == NULL) {
+		if (event == IN_DELETE_SELF) {
+			/*
+			 * If we're here because we're being deleted and we
+			 * are not a child watch, we need to delete the entire
+			 * watch, children and all.
+			 */
+			inotify_watch_remove(state, watch);
+		}
+
+		mutex_exit(&state->ins_lock);
+		return;
+	} else {
+		if (event == IN_DELETE_SELF &&
+		    !(parent->inw_mask & IN_EXCL_UNLINK)) {
+			/*
+			 * This is a child watch for a file that is being
+			 * removed and IN_EXCL_UNLINK has not been specified;
+			 * indicate that it is orphaned and add it to the list
+			 * of orphans.  (This list will be checked by the
+			 * cleaning cyclic to determine when the watch has
+			 * become the only hold on the vnode, at which point
+			 * the watch can be zombified.)  Note that we check
+			 * if the watch is orphaned before we orphan it:  hard
+			 * links make it possible for VE_REMOVE to be called
+			 * multiple times on the same vnode. (!)
+			 */
+			if (!watch->inw_orphaned) {
+				watch->inw_orphaned = 1;
+				watch->inw_cred = CRED();
+				crhold(watch->inw_cred);
+				list_insert_head(&state->ins_orphans, watch);
+			}
+
+			mutex_exit(&state->ins_lock);
+			return;
+		}
+
+		if (watch->inw_orphaned) {
+			/*
+			 * If we're here, a file was orphaned and then later
+			 * moved -- which almost certainly means that hard
+			 * links are on the scene.  We choose the orphan over
+			 * the move because we don't want to spuriously
+			 * drop events if we can avoid it.
+			 */
+			crfree(watch->inw_cred);
+			list_remove(&state->ins_orphans, watch);
+		}
+	}
+
+	if (avl_find(&parent->inw_children, &cmp, NULL) == NULL) {
+		/*
+		 * This watch has already been deleted from the parent.
+		 */
+		mutex_exit(&state->ins_lock);
+		return;
+	}
+
+	avl_remove(&parent->inw_children, watch);
+	err = fem_uninstall(watch->inw_vp, inotify_femp, watch);
+	VERIFY(err == 0);
+
+	VN_RELE(watch->inw_vp);
+
+	/*
+	 * It's now safe to zombify the watch -- which won't actually delete
+	 * it as we know that the reference count is greater than 1.
+	 */
+	inotify_watch_zombify(watch);
+	mutex_exit(&state->ins_lock);
+}
+
+/*
+ * Insert a new child watch.  Should only be called from VOP context when
+ * a child is created in a watched directory.
+ */
+static void
+inotify_watch_insert(inotify_watch_t *watch, vnode_t *vp, char *name)
+{
+	inotify_state_t *state = watch->inw_state;
+	inotify_watch_t cmp = { .inw_vp = vp };
+
+	if (!(watch->inw_mask & IN_CHILD_EVENTS))
+		return;
+
+	mutex_enter(&state->ins_lock);
+
+	if (watch->inw_zombie || watch->inw_parent != NULL || vp == NULL) {
+		mutex_exit(&state->ins_lock);
+		return;
+	}
+
+	if (avl_find(&watch->inw_children, &cmp, NULL) != NULL) {
+		mutex_exit(&state->ins_lock);
+		return;
+	}
+
+	VN_HOLD(vp);
+	watch = inotify_watch_add(state, watch, name, vp, watch->inw_mask);
+	VERIFY(watch != NULL);
+
+	mutex_exit(&state->ins_lock);
+}
+
+
+static int
+inotify_add_watch(inotify_state_t *state, vnode_t *vp, uint32_t mask,
+    int32_t *wdp)
+{
+	inotify_watch_t *watch, cmp = { .inw_vp = vp };
+	uint32_t set;
+
+	set = (mask & (IN_ALL_EVENTS | IN_MODIFIERS)) | IN_UNMASKABLE;
+
+	/*
+	 * Lookup our vnode to determine if we already have a watch on it.
+	 */
+	mutex_enter(&state->ins_lock);
+
+	if ((watch = avl_find(&state->ins_byvp, &cmp, NULL)) == NULL) {
+		/*
+		 * We don't have this watch; allocate a new one, provided that
+		 * we have fewer than our limit.
+		 */
+		if (avl_numnodes(&state->ins_bywd) >= state->ins_maxwatches) {
+			mutex_exit(&state->ins_lock);
+			return (ENOSPC);
+		}
+
+		VN_HOLD(vp);
+		watch = inotify_watch_add(state, NULL, NULL, vp, set);
+		*wdp = watch->inw_wd;
+		mutex_exit(&state->ins_lock);
+
+		return (0);
+	}
+
+	VERIFY(!watch->inw_zombie);
+
+	if (!(mask & IN_MASK_ADD)) {
+		/*
+		 * Note that if we're resetting our event mask and we're
+		 * transitioning from an event mask that includes child events
+		 * to one that doesn't, there will be potentially some stale
+		 * child watches.  This is basically fine:  they won't fire,
+		 * and they will correctly be removed when the watch is
+		 * removed.
+		 */
+		watch->inw_mask = 0;
+	}
+
+	watch->inw_mask |= set;
+
+	*wdp = watch->inw_wd;
+
+	mutex_exit(&state->ins_lock);
+
+	return (0);
+}
+
+static int
+inotify_add_child(inotify_state_t *state, vnode_t *vp, char *name)
+{
+	inotify_watch_t *watch, cmp = { .inw_vp = vp };
+	vnode_t *cvp;
+	int err;
+
+	/*
+	 * Verify that the specified child doesn't have a directory component
+	 * within it.
+	 */
+	if (strchr(name, '/') != NULL)
+		return (EINVAL);
+
+	/*
+	 * Lookup the underlying file.  Note that this will succeed even if
+	 * we don't have permissions to actually read the file.
+	 */
+	if ((err = lookupnameat(name,
+	    UIO_SYSSPACE, NO_FOLLOW, NULL, &cvp, vp)) != 0) {
+		return (err);
+	}
+
+	/*
+	 * Use our vnode to find our watch, and then add our child watch to it.
+	 */
+	mutex_enter(&state->ins_lock);
+
+	if ((watch = avl_find(&state->ins_byvp, &cmp, NULL)) == NULL) {
+		/*
+		 * This is unexpected -- it means that we don't have the
+		 * watch that we thought we had.
+		 */
+		mutex_exit(&state->ins_lock);
+		VN_RELE(cvp);
+		return (ENXIO);
+	}
+
+	/*
+	 * Now lookup the child vnode in the watch; we'll only add it if it
+	 * isn't already there.
+	 */
+	cmp.inw_vp = cvp;
+
+	if (avl_find(&watch->inw_children, &cmp, NULL) != NULL) {
+		mutex_exit(&state->ins_lock);
+		VN_RELE(cvp);
+		return (0);
+	}
+
+	watch = inotify_watch_add(state, watch, name, cvp, watch->inw_mask);
+	VERIFY(watch != NULL);
+	mutex_exit(&state->ins_lock);
+
+	return (0);
+}
+
+static int
+inotify_rm_watch(inotify_state_t *state, int32_t wd)
+{
+	inotify_watch_t *watch, cmp = { .inw_wd = wd };
+
+	mutex_enter(&state->ins_lock);
+
+	if ((watch = avl_find(&state->ins_bywd, &cmp, NULL)) == NULL) {
+		mutex_exit(&state->ins_lock);
+		return (EINVAL);
+	}
+
+	inotify_watch_remove(state, watch);
+	mutex_exit(&state->ins_lock);
+
+	return (0);
+}
+
+static int
+inotify_activate(inotify_state_t *state, int32_t wd)
+{
+	inotify_watch_t *watch, cmp = { .inw_wd = wd };
+
+	mutex_enter(&state->ins_lock);
+
+	if ((watch = avl_find(&state->ins_bywd, &cmp, NULL)) == NULL) {
+		mutex_exit(&state->ins_lock);
+		return (EINVAL);
+	}
+
+	watch->inw_active = 1;
+
+	mutex_exit(&state->ins_lock);
+
+	return (0);
+}
+
+/*
+ * Called periodically as a cyclic to process the orphans and zombies.
+ */
+static void
+inotify_clean(void *arg)
+{
+	inotify_state_t *state = arg;
+	inotify_watch_t *watch, *parent, *next, **prev;
+	cred_t *savecred;
+	int err;
+
+	mutex_enter(&state->ins_lock);
+
+	for (watch = list_head(&state->ins_orphans);
+	    watch != NULL; watch = next) {
+		next = list_next(&state->ins_orphans, watch);
+
+		VERIFY(!watch->inw_zombie);
+		VERIFY((parent = watch->inw_parent) != NULL);
+
+		if (watch->inw_vp->v_count > 1)
+			continue;
+
+		avl_remove(&parent->inw_children, watch);
+		err = fem_uninstall(watch->inw_vp, inotify_femp, watch);
+		VERIFY(err == 0);
+
+		list_remove(&state->ins_orphans, watch);
+
+		/*
+		 * For purposes of releasing the vnode, we need to switch our
+		 * cred to be the cred of the orphaning thread (which we held
+		 * at the time this watch was orphaned).
+		 */
+		savecred = curthread->t_cred;
+		curthread->t_cred = watch->inw_cred;
+		VN_RELE(watch->inw_vp);
+		crfree(watch->inw_cred);
+		curthread->t_cred = savecred;
+
+		inotify_watch_zombify(watch);
+	}
+
+	prev = &state->ins_zombies;
+
+	while ((watch = *prev) != NULL) {
+		mutex_enter(&watch->inw_lock);
+
+		if (watch->inw_refcnt == 1) {
+			*prev = watch->inw_parent;
+			inotify_watch_destroy(watch);
+			continue;
+		}
+
+		prev = &watch->inw_parent;
+		mutex_exit(&watch->inw_lock);
+	}
+
+	mutex_exit(&state->ins_lock);
+}
+
+/*ARGSUSED*/
+static int
+inotify_open(dev_t *devp, int flag, int otyp, cred_t *cred_p)
+{
+	inotify_state_t *state;
+	major_t major = getemajor(*devp);
+	minor_t minor = getminor(*devp);
+	int instances = 0;
+	cyc_handler_t hdlr;
+	cyc_time_t when;
+	char c[64];
+
+	if (minor != INOTIFYMNRN_INOTIFY)
+		return (ENXIO);
+
+	mutex_enter(&inotify_lock);
+
+	for (state = inotify_state; state != NULL; state = state->ins_next) {
+		if (state->ins_cred == cred_p)
+			instances++;
+	}
+
+	if (instances >= inotify_maxinstances) {
+		mutex_exit(&inotify_lock);
+		return (EMFILE);
+	}
+
+	minor = (minor_t)(uintptr_t)vmem_alloc(inotify_minor, 1,
+	    VM_BESTFIT | VM_SLEEP);
+
+	if (ddi_soft_state_zalloc(inotify_softstate, minor) != DDI_SUCCESS) {
+		vmem_free(inotify_minor, (void *)(uintptr_t)minor, 1);
+		mutex_exit(&inotify_lock);
+		return (NULL);
+	}
+
+	state = ddi_get_soft_state(inotify_softstate, minor);
+	*devp = makedevice(major, minor);
+
+	crhold(cred_p);
+	state->ins_cred = cred_p;
+	state->ins_next = inotify_state;
+	inotify_state = state;
+
+	(void) snprintf(c, sizeof (c), "inotify_watchid_%d", minor);
+	state->ins_wds = vmem_create(c, (void *)1, UINT32_MAX, 1,
+	    NULL, NULL, NULL, 0, VM_SLEEP | VMC_IDENTIFIER);
+
+	avl_create(&state->ins_bywd,
+	    (int(*)(const void *, const void *))inotify_watch_cmpwd,
+	    sizeof (inotify_watch_t),
+	    offsetof(inotify_watch_t, inw_bywd));
+
+	avl_create(&state->ins_byvp,
+	    (int(*)(const void *, const void *))inotify_watch_cmpvp,
+	    sizeof (inotify_watch_t),
+	    offsetof(inotify_watch_t, inw_byvp));
+
+	list_create(&state->ins_orphans, sizeof (inotify_watch_t),
+	    offsetof(inotify_watch_t, inw_orphan));
+
+	state->ins_maxwatches = inotify_maxwatches;
+	state->ins_maxevents = inotify_maxevents;
+
+	mutex_exit(&inotify_lock);
+
+	mutex_enter(&cpu_lock);
+
+	hdlr.cyh_func = inotify_clean;
+	hdlr.cyh_level = CY_LOW_LEVEL;
+	hdlr.cyh_arg = state;
+
+	when.cyt_when = 0;
+	when.cyt_interval = NANOSEC;
+
+	state->ins_cleaner = cyclic_add(&hdlr, &when);
+	mutex_exit(&cpu_lock);
+
+	return (0);
+}
+
+/*ARGSUSED*/
+static int
+inotify_read(dev_t dev, uio_t *uio, cred_t *cr)
+{
+	inotify_state_t *state;
+	inotify_kevent_t *event;
+	minor_t minor = getminor(dev);
+	int err = 0, nevents = 0;
+	size_t len;
+
+	state = ddi_get_soft_state(inotify_softstate, minor);
+
+	mutex_enter(&state->ins_lock);
+
+	while (state->ins_head == NULL) {
+		if (uio->uio_fmode & (FNDELAY|FNONBLOCK)) {
+			mutex_exit(&state->ins_lock);
+			return (EAGAIN);
+		}
+
+		if (!cv_wait_sig_swap(&state->ins_cv, &state->ins_lock)) {
+			mutex_exit(&state->ins_lock);
+			return (EINTR);
+		}
+	}
+
+	/*
+	 * We have events and we have our lock; return as many as we can.
+	 */
+	while ((event = state->ins_head) != NULL) {
+		len = sizeof (event->ine_event) + event->ine_event.len;
+
+		if (uio->uio_resid < len) {
+			if (nevents == 0)
+				err = EINVAL;
+			break;
+		}
+
+		nevents++;
+
+		if ((err = uiomove(&event->ine_event, len, UIO_READ, uio)) != 0)
+			break;
+
+		VERIFY(state->ins_nevents > 0);
+		state->ins_nevents--;
+
+		VERIFY(state->ins_size > 0);
+		state->ins_size -= len;
+
+		if ((state->ins_head = event->ine_next) == NULL) {
+			VERIFY(event == state->ins_tail);
+			VERIFY(state->ins_nevents == 0);
+			state->ins_tail = NULL;
+		}
+
+		kmem_free(event, INOTIFY_EVENT_LENGTH(event));
+	}
+
+	mutex_exit(&state->ins_lock);
+
+	return (err);
+}
+
+/*ARGSUSED*/
+static int
+inotify_poll(dev_t dev, short events, int anyyet, short *reventsp,
+    struct pollhead **phpp)
+{
+	inotify_state_t *state;
+	minor_t minor = getminor(dev);
+
+	state = ddi_get_soft_state(inotify_softstate, minor);
+
+	mutex_enter(&state->ins_lock);
+
+	if (state->ins_head != NULL) {
+		*reventsp = events & (POLLRDNORM | POLLIN);
+	} else {
+		*reventsp = 0;
+
+		if (!anyyet)
+			*phpp = &state->ins_pollhd;
+	}
+
+	mutex_exit(&state->ins_lock);
+
+	return (0);
+}
+
+/*ARGSUSED*/
+static int
+inotify_ioctl(dev_t dev, int cmd, intptr_t arg, int md, cred_t *cr, int *rv)
+{
+	inotify_state_t *state;
+	minor_t minor = getminor(dev);
+	file_t *fp;
+	int rval;
+
+	state = ddi_get_soft_state(inotify_softstate, minor);
+
+	switch (cmd) {
+	case INOTIFYIOC_ADD_WATCH: {
+		inotify_addwatch_t addwatch;
+		file_t *fp;
+
+		if (copyin((void *)arg, &addwatch, sizeof (addwatch)) != 0)
+			return (EFAULT);
+
+		if ((fp = getf(addwatch.inaw_fd)) == NULL)
+			return (EBADF);
+
+		rval = inotify_add_watch(state, fp->f_vnode,
+		    addwatch.inaw_mask, rv);
+
+		releasef(addwatch.inaw_fd);
+		return (rval);
+	}
+
+	case INOTIFYIOC_ADD_CHILD: {
+		inotify_addchild_t addchild;
+		char name[MAXPATHLEN];
+
+		if (copyin((void *)arg, &addchild, sizeof (addchild)) != 0)
+			return (EFAULT);
+
+		if (copyinstr(addchild.inac_name, name, MAXPATHLEN, NULL) != 0)
+			return (EFAULT);
+
+		if ((fp = getf(addchild.inac_fd)) == NULL)
+			return (EBADF);
+
+		rval = inotify_add_child(state, fp->f_vnode, name);
+
+		releasef(addchild.inac_fd);
+		return (rval);
+	}
+
+	case INOTIFYIOC_RM_WATCH:
+		return (inotify_rm_watch(state, arg));
+
+	case INOTIFYIOC_ACTIVATE:
+		return (inotify_activate(state, arg));
+
+	case FIONREAD: {
+		int32_t size;
+
+		mutex_enter(&state->ins_lock);
+		size = state->ins_size;
+		mutex_exit(&state->ins_lock);
+
+		if (copyout(&size, (void *)arg, sizeof (size)) != 0)
+			return (EFAULT);
+
+		return (0);
+	}
+
+	default:
+		break;
+	}
+
+	return (ENOTTY);
+}
+
+/*ARGSUSED*/
+static int
+inotify_close(dev_t dev, int flag, int otyp, cred_t *cred_p)
+{
+	inotify_state_t *state, **sp;
+	inotify_watch_t *watch, *zombies;
+	inotify_kevent_t *event;
+	minor_t minor = getminor(dev);
+
+	state = ddi_get_soft_state(inotify_softstate, minor);
+
+	if (state->ins_pollhd.ph_list != NULL) {
+		pollwakeup(&state->ins_pollhd, POLLERR);
+		pollhead_clean(&state->ins_pollhd);
+	}
+
+	mutex_enter(&state->ins_lock);
+
+	/*
+	 * First, destroy all of our watches.
+	 */
+	while ((watch = avl_first(&state->ins_bywd)) != NULL)
+		inotify_watch_remove(state, watch);
+
+	/*
+	 * And now destroy our event queue.
+	 */
+	while ((event = state->ins_head) != NULL) {
+		state->ins_head = event->ine_next;
+		kmem_free(event, INOTIFY_EVENT_LENGTH(event));
+	}
+
+	zombies = state->ins_zombies;
+	state->ins_zombies = NULL;
+	mutex_exit(&state->ins_lock);
+
+	/*
+	 * Now that our state lock is dropped, we can synchronously wait on
+	 * any zombies.
+	 */
+	while ((watch = zombies) != NULL) {
+		zombies = zombies->inw_parent;
+
+		mutex_enter(&watch->inw_lock);
+
+		while (watch->inw_refcnt > 1)
+			cv_wait(&watch->inw_cv, &watch->inw_lock);
+
+		inotify_watch_destroy(watch);
+	}
+
+	mutex_enter(&cpu_lock);
+	cyclic_remove(state->ins_cleaner);
+	mutex_exit(&cpu_lock);
+
+	mutex_enter(&inotify_lock);
+
+	/*
+	 * Remove our state from our global list, and release our hold on
+	 * the cred.
+	 */
+	for (sp = &inotify_state; *sp != state; sp = &((*sp)->ins_next))
+		VERIFY(*sp != NULL);
+
+	*sp = (*sp)->ins_next;
+	crfree(state->ins_cred);
+
+	ddi_soft_state_free(inotify_softstate, minor);
+	vmem_free(inotify_minor, (void *)(uintptr_t)minor, 1);
+
+	mutex_exit(&inotify_lock);
+
+	return (0);
+}
+
+/*ARGSUSED*/
+static int
+inotify_attach(dev_info_t *devi, ddi_attach_cmd_t cmd)
+{
+	mutex_enter(&inotify_lock);
+
+	if (ddi_soft_state_init(&inotify_softstate,
+	    sizeof (inotify_state_t), 0) != 0) {
+		cmn_err(CE_NOTE, "/dev/inotify failed to create soft state");
+		mutex_exit(&inotify_lock);
+		return (DDI_FAILURE);
+	}
+
+	if (ddi_create_minor_node(devi, "inotify", S_IFCHR,
+	    INOTIFYMNRN_INOTIFY, DDI_PSEUDO, NULL) == DDI_FAILURE) {
+		cmn_err(CE_NOTE, "/dev/inotify couldn't create minor node");
+		ddi_soft_state_fini(&inotify_softstate);
+		mutex_exit(&inotify_lock);
+		return (DDI_FAILURE);
+	}
+
+	if (fem_create("inotify_fem",
+	    inotify_vnodesrc_template, &inotify_femp) != 0) {
+		cmn_err(CE_NOTE, "/dev/inotify couldn't create FEM state");
+		ddi_remove_minor_node(devi, NULL);
+		ddi_soft_state_fini(&inotify_softstate);
+		mutex_exit(&inotify_lock);
+		return (DDI_FAILURE);
+	}
+
+	ddi_report_dev(devi);
+	inotify_devi = devi;
+
+	inotify_minor = vmem_create("inotify_minor", (void *)INOTIFYMNRN_CLONE,
+	    UINT32_MAX - INOTIFYMNRN_CLONE, 1, NULL, NULL, NULL, 0,
+	    VM_SLEEP | VMC_IDENTIFIER);
+
+	mutex_exit(&inotify_lock);
+
+	return (DDI_SUCCESS);
+}
+
+/*ARGSUSED*/
+static int
+inotify_detach(dev_info_t *dip, ddi_detach_cmd_t cmd)
+{
+	switch (cmd) {
+	case DDI_DETACH:
+		break;
+
+	case DDI_SUSPEND:
+		return (DDI_SUCCESS);
+
+	default:
+		return (DDI_FAILURE);
+	}
+
+	mutex_enter(&inotify_lock);
+	fem_free(inotify_femp);
+	vmem_destroy(inotify_minor);
+
+	ddi_remove_minor_node(inotify_devi, NULL);
+	inotify_devi = NULL;
+
+	ddi_soft_state_fini(&inotify_softstate);
+	mutex_exit(&inotify_lock);
+
+	return (DDI_SUCCESS);
+}
+
+/*ARGSUSED*/
+static int
+inotify_info(dev_info_t *dip, ddi_info_cmd_t infocmd, void *arg, void **result)
+{
+	int error;
+
+	switch (infocmd) {
+	case DDI_INFO_DEVT2DEVINFO:
+		*result = (void *)inotify_devi;
+		error = DDI_SUCCESS;
+		break;
+	case DDI_INFO_DEVT2INSTANCE:
+		*result = (void *)0;
+		error = DDI_SUCCESS;
+		break;
+	default:
+		error = DDI_FAILURE;
+	}
+	return (error);
+}
+
+static struct cb_ops inotify_cb_ops = {
+	inotify_open,		/* open */
+	inotify_close,		/* close */
+	nulldev,		/* strategy */
+	nulldev,		/* print */
+	nodev,			/* dump */
+	inotify_read,		/* read */
+	nodev,			/* write */
+	inotify_ioctl,		/* ioctl */
+	nodev,			/* devmap */
+	nodev,			/* mmap */
+	nodev,			/* segmap */
+	inotify_poll,		/* poll */
+	ddi_prop_op,		/* cb_prop_op */
+	0,			/* streamtab  */
+	D_NEW | D_MP		/* Driver compatibility flag */
+};
+
+static struct dev_ops inotify_ops = {
+	DEVO_REV,		/* devo_rev */
+	0,			/* refcnt */
+	inotify_info,		/* get_dev_info */
+	nulldev,		/* identify */
+	nulldev,		/* probe */
+	inotify_attach,		/* attach */
+	inotify_detach,		/* detach */
+	nodev,			/* reset */
+	&inotify_cb_ops,	/* driver operations */
+	NULL,			/* bus operations */
+	nodev,			/* dev power */
+	ddi_quiesce_not_needed,	/* quiesce */
+};
+
+static struct modldrv modldrv = {
+	&mod_driverops,		/* module type (this is a pseudo driver) */
+	"inotify support",	/* name of module */
+	&inotify_ops,		/* driver ops */
+};
+
+static struct modlinkage modlinkage = {
+	MODREV_1,
+	(void *)&modldrv,
+	NULL
+};
+
+int
+_init(void)
+{
+	return (mod_install(&modlinkage));
+}
+
+int
+_info(struct modinfo *modinfop)
+{
+	return (mod_info(&modlinkage, modinfop));
+}
+
+int
+_fini(void)
+{
+	return (mod_remove(&modlinkage));
+}
diff --git a/usr/src/uts/common/io/inotify.conf b/usr/src/uts/common/io/inotify.conf
new file mode 100644
index 0000000000..ce9da6180f
--- /dev/null
+++ b/usr/src/uts/common/io/inotify.conf
@@ -0,0 +1,16 @@
+#
+# This file and its contents are supplied under the terms of the
+# Common Development and Distribution License ("CDDL"), version 1.0.
+# You may only use this file in accordance with the terms of version
+# 1.0 of the CDDL.
+#
+# A full copy of the text of the CDDL should have accompanied this
+# source.  A copy of the CDDL is also available via the Internet at
+# http://www.illumos.org/license/CDDL.
+#
+
+#
+# Copyright (c) 2014 Joyent, Inc.  All rights reserved.
+#
+
+name="inotify" parent="pseudo" instance=0;
diff --git a/usr/src/uts/common/io/ixgbe/ixgbe_main.c b/usr/src/uts/common/io/ixgbe/ixgbe_main.c
index 848e3470c7..117b7da16a 100644
--- a/usr/src/uts/common/io/ixgbe/ixgbe_main.c
+++ b/usr/src/uts/common/io/ixgbe/ixgbe_main.c
@@ -1792,6 +1792,7 @@ ixgbe_cbfunc(dev_info_t *dip, ddi_cb_action_t cbaction, void *cbarg,
     void *arg1, void *arg2)
 {
 	ixgbe_t *ixgbe = (ixgbe_t *)arg1;
+	int prev = ixgbe->intr_cnt;
 
 	switch (cbaction) {
 	/* IRM callback */
@@ -1805,7 +1806,8 @@ ixgbe_cbfunc(dev_info_t *dip, ddi_cb_action_t cbaction, void *cbarg,
 		if (ixgbe_intr_adjust(ixgbe, cbaction, count) !=
 		    DDI_SUCCESS) {
 			ixgbe_error(ixgbe,
-			    "IRM CB: Failed to adjust interrupts");
+			    "IRM CB: Failed to adjust interrupts [%d %d %d]",
+			    cbaction, count, prev);
 			goto cb_fail;
 		}
 		break;
diff --git a/usr/src/uts/common/io/ksocket/ksocket.c b/usr/src/uts/common/io/ksocket/ksocket.c
index 49ca6f0475..25da45be39 100644
--- a/usr/src/uts/common/io/ksocket/ksocket.c
+++ b/usr/src/uts/common/io/ksocket/ksocket.c
@@ -22,6 +22,7 @@
 /*
  * Copyright 2011 Nexenta Systems, Inc.  All rights reserved.
  * Copyright (c) 2008, 2010, Oracle and/or its affiliates. All rights reserved.
+ * Copyright 2015, Joyent, Inc.
  */
 
 #include <sys/file.h>
@@ -820,7 +821,7 @@ ksocket_spoll(ksocket_t ks, int timo, short events, short *revents,
 		if (error != 0 || *revents != 0)
 			break;
 
-		if (pcp->pc_flag & T_POLLWAKE)
+		if (pcp->pc_flag & PC_POLLWAKE)
 			continue;
 
 		if (timo == -1) {
@@ -931,3 +932,15 @@ ksocket_rele(ksocket_t ks)
 			cv_signal(&so->so_closing_cv);
 	}
 }
+
+int
+ksocket_krecv_set(ksocket_t ks, ksocket_krecv_f cb, void *arg)
+{
+	return (so_krecv_set(KSTOSO(ks), (so_krecv_f)cb, arg));
+}
+
+void
+ksocket_krecv_unblock(ksocket_t ks)
+{
+	return (so_krecv_unblock(KSTOSO(ks)));
+}
diff --git a/usr/src/uts/common/io/ksocket/ksocket_impl.h b/usr/src/uts/common/io/ksocket/ksocket_impl.h
index ac5251540f..516a68d358 100644
--- a/usr/src/uts/common/io/ksocket/ksocket_impl.h
+++ b/usr/src/uts/common/io/ksocket/ksocket_impl.h
@@ -22,11 +22,17 @@
 /*
  * Copyright 2008 Sun Microsystems, Inc.  All rights reserved.
  * Use is subject to license terms.
+ * Copyright 2015, Joyent, Inc.
  */
 
 #ifndef	_INET_KSOCKET_KSOCKET_IMPL_H
 #define	_INET_KSOCKET_KSOCKET_IMPL_H
 
+/*
+ * Note that if this relationship ever changes, the logic in ksocket_krecv_set
+ * must be updated and we must maintain local state about this on whatever the
+ * new ksocket object is.
+ */
 #define	KSTOSO(ks)	((struct sonode *)(ks))
 #define	SOTOKS(so)	((ksocket_t)(uintptr_t)(so))
 
diff --git a/usr/src/uts/common/io/mac/mac.c b/usr/src/uts/common/io/mac/mac.c
index 98ec8b4366..99bdbda0cd 100644
--- a/usr/src/uts/common/io/mac/mac.c
+++ b/usr/src/uts/common/io/mac/mac.c
@@ -3101,6 +3101,9 @@ mac_prop_check_size(mac_prop_id_t id, uint_t valsize, boolean_t is_range)
 	case MAC_PROP_WL_MLME:
 		minsize = sizeof (wl_mlme_t);
 		break;
+	case MAC_PROP_VN_PROMISC_FILTERED:
+		minsize = sizeof (boolean_t);
+		break;
 	}
 
 	return (valsize >= minsize);
diff --git a/usr/src/uts/common/io/mac/mac_client.c b/usr/src/uts/common/io/mac/mac_client.c
index f4dd8aab25..d719899901 100644
--- a/usr/src/uts/common/io/mac/mac_client.c
+++ b/usr/src/uts/common/io/mac/mac_client.c
@@ -21,7 +21,7 @@
 
 /*
  * Copyright (c) 2008, 2010, Oracle and/or its affiliates. All rights reserved.
- * Copyright (c) 2014, Joyent, Inc.  All rights reserved.
+ * Copyright 2015, Joyent, Inc.
  */
 
 /*
@@ -3256,7 +3256,8 @@ mac_promisc_add(mac_client_handle_t mch, mac_client_promisc_type_t type,
 	}
 
 	if ((mcip->mci_state_flags & MCIS_IS_VNIC) &&
-	    type == MAC_CLIENT_PROMISC_ALL) {
+	    type == MAC_CLIENT_PROMISC_ALL &&
+	    (mcip->mci_protect_flags & MPT_FLAG_PROMISC_FILTERED)) {
 		/*
 		 * The function is being invoked by the upper MAC client
 		 * of a VNIC. The VNIC should only see the traffic
@@ -4135,16 +4136,15 @@ mac_info_get(const char *name, mac_info_t *minfop)
 /*
  * To get the capabilities that MAC layer cares about, such as rings, factory
  * mac address, vnic or not, it should directly invoke this function.  If the
- * link is part of a bridge, then the only "capability" it has is the inability
- * to do zero copy.
+ * link is part of a bridge, then the link is unable to do zero copy.
  */
 boolean_t
 i_mac_capab_get(mac_handle_t mh, mac_capab_t cap, void *cap_data)
 {
 	mac_impl_t *mip = (mac_impl_t *)mh;
 
-	if (mip->mi_bridge_link != NULL)
-		return (cap == MAC_CAPAB_NO_ZCOPY);
+	if (mip->mi_bridge_link != NULL && cap == MAC_CAPAB_NO_ZCOPY)
+		return (B_TRUE);
 	else if (mip->mi_callbacks->mc_callbacks & MC_GETCAPAB)
 		return (mip->mi_getcapab(mip->mi_driver, cap, cap_data));
 	else
@@ -4323,7 +4323,13 @@ mac_addr_len(mac_handle_t mh)
 boolean_t
 mac_is_vnic(mac_handle_t mh)
 {
-	return (((mac_impl_t *)mh)->mi_state_flags & MIS_IS_VNIC);
+	return ((((mac_impl_t *)mh)->mi_state_flags & MIS_IS_VNIC) != 0);
+}
+
+boolean_t
+mac_is_overlay(mac_handle_t mh)
+{
+	return ((((mac_impl_t *)mh)->mi_state_flags & MIS_IS_OVERLAY) != 0);
 }
 
 mac_handle_t
@@ -5537,3 +5543,23 @@ mac_client_set_rings(mac_client_handle_t mch, int rxrings, int txrings)
 		mrp->mrp_ntxrings = txrings;
 	}
 }
+
+boolean_t
+mac_get_promisc_filtered(mac_client_handle_t mch)
+{
+	mac_client_impl_t	*mcip = (mac_client_impl_t *)mch;
+
+	return (mcip->mci_protect_flags & MPT_FLAG_PROMISC_FILTERED);
+}
+
+void
+mac_set_promisc_filtered(mac_client_handle_t mch, boolean_t enable)
+{
+	mac_client_impl_t	*mcip = (mac_client_impl_t *)mch;
+
+	ASSERT(MAC_PERIM_HELD((mac_handle_t)mcip->mci_mip));
+	if (enable)
+		mcip->mci_protect_flags |= MPT_FLAG_PROMISC_FILTERED;
+	else
+		mcip->mci_protect_flags &= ~MPT_FLAG_PROMISC_FILTERED;
+}
diff --git a/usr/src/uts/common/io/mac/mac_datapath_setup.c b/usr/src/uts/common/io/mac/mac_datapath_setup.c
index 5ca673ea6e..b0f184b161 100644
--- a/usr/src/uts/common/io/mac/mac_datapath_setup.c
+++ b/usr/src/uts/common/io/mac/mac_datapath_setup.c
@@ -20,6 +20,7 @@
  */
 /*
  * Copyright (c) 2008, 2010, Oracle and/or its affiliates. All rights reserved.
+ * Copyright 2015, Joyent, Inc.
  */
 
 #include <sys/types.h>
@@ -603,6 +604,7 @@ mac_srs_cpu_setup(cpu_setup_t what, int id, void *arg)
  *
  * TODO: Cleanup and tighten some of the assumptions.
  */
+boolean_t mac_check_overlay = B_TRUE;
 boolean_t mac_use_bw_heuristic = B_TRUE;
 static int
 mac_compute_soft_ring_count(flow_entry_t *flent, int rx_srs_cnt, int maxcpus)
@@ -610,6 +612,7 @@ mac_compute_soft_ring_count(flow_entry_t *flent, int rx_srs_cnt, int maxcpus)
 	uint64_t cpu_speed, bw = 0;
 	int srings = 0;
 	boolean_t bw_enabled = B_FALSE;
+	mac_client_impl_t *mcip = flent->fe_mcip;
 
 	ASSERT(!(flent->fe_type & FLOW_USER));
 	if (flent->fe_resource_props.mrp_mask & MRP_MAXBW &&
@@ -637,7 +640,16 @@ mac_compute_soft_ring_count(flow_entry_t *flent, int rx_srs_cnt, int maxcpus)
 			 */
 			if (mac_soft_ring_enable)
 				srings = srings * 2;
+		} else if (mac_check_overlay == B_TRUE &&
+		    (mcip->mci_state_flags & MCIS_IS_VNIC) != 0) {
+			/* Is this a VNIC on an overlay? */
+			mac_handle_t mh = (mac_handle_t)mcip->mci_mip;
+			if (mac_is_overlay(mh) == B_TRUE) {
+				srings = mac_rx_soft_ring_10gig_count;
+			}
 		}
+
+
 	} else {
 		/*
 		 * Soft ring computation using CPU speed and specified
diff --git a/usr/src/uts/common/io/mac/mac_protect.c b/usr/src/uts/common/io/mac/mac_protect.c
index cd7fcb9a5d..e341f3d562 100644
--- a/usr/src/uts/common/io/mac/mac_protect.c
+++ b/usr/src/uts/common/io/mac/mac_protect.c
@@ -2297,6 +2297,9 @@ mac_protect_init(mac_client_impl_t *mcip)
 	    sizeof (dhcpv6_cid_t), offsetof(dhcpv6_cid_t, dc_node));
 	avl_create(&mcip->mci_v6_dyn_ip, compare_dhcpv6_ip,
 	    sizeof (dhcpv6_addr_t), offsetof(dhcpv6_addr_t, da_node));
+
+	if (mcip->mci_state_flags & MCIS_IS_VNIC)
+		mcip->mci_protect_flags |= MPT_FLAG_PROMISC_FILTERED;
 }
 
 void
diff --git a/usr/src/uts/common/io/mac/mac_provider.c b/usr/src/uts/common/io/mac/mac_provider.c
index 57d1996d84..a17795be22 100644
--- a/usr/src/uts/common/io/mac/mac_provider.c
+++ b/usr/src/uts/common/io/mac/mac_provider.c
@@ -21,6 +21,7 @@
 
 /*
  * Copyright (c) 2008, 2010, Oracle and/or its affiliates. All rights reserved.
+ * Copyright 2015, Joyent, Inc.
  */
 
 #include <sys/types.h>
@@ -350,6 +351,9 @@ mac_register(mac_register_t *mregp, mac_handle_t *mhp)
 	if (i_mac_capab_get((mac_handle_t)mip, MAC_CAPAB_AGGR, NULL))
 		mip->mi_state_flags |= MIS_IS_AGGR;
 
+	if (i_mac_capab_get((mac_handle_t)mip, MAC_CAPAB_OVERLAY, NULL))
+		mip->mi_state_flags |= MIS_IS_OVERLAY;
+
 	mac_addr_factory_init(mip);
 
 	/*
diff --git a/usr/src/uts/common/io/mac/mac_sched.c b/usr/src/uts/common/io/mac/mac_sched.c
index eb179b07c7..6fcdb468c2 100644
--- a/usr/src/uts/common/io/mac/mac_sched.c
+++ b/usr/src/uts/common/io/mac/mac_sched.c
@@ -1370,7 +1370,7 @@ int mac_srs_worker_wakeup_ticks = 0;
  * said, the constant is left as a static variable to allow it to be
  * dynamically tuned in the field if and as needed.
  */
-static uintptr_t mac_rx_srs_stack_needed = 10240;
+static uintptr_t mac_rx_srs_stack_needed = 13312;
 static uint_t mac_rx_srs_stack_toodeep;
 
 #ifndef STACK_GROWTH_DOWN
diff --git a/usr/src/uts/common/io/mac/mac_stat.c b/usr/src/uts/common/io/mac/mac_stat.c
index 31972f94d8..c1a5c9c069 100644
--- a/usr/src/uts/common/io/mac/mac_stat.c
+++ b/usr/src/uts/common/io/mac/mac_stat.c
@@ -21,6 +21,7 @@
 /*
  * Copyright 2010 Sun Microsystems, Inc.  All rights reserved.
  * Use is subject to license terms.
+ * Copyright 2013 Joyent, Inc.  All rights reserved.
  */
 
 /*
@@ -390,8 +391,8 @@ i_mac_stat_create(void *handle, const char *modname, const char *statname,
 	kstat_t		*ksp;
 	kstat_named_t	*knp;
 
-	ksp = kstat_create(modname, 0, statname, "net",
-	    KSTAT_TYPE_NAMED, count, 0);
+	ksp = kstat_create_zone(modname, 0, statname, "net",
+	    KSTAT_TYPE_NAMED, count, 0, getzoneid());
 
 	if (ksp == NULL)
 		return (NULL);
@@ -948,9 +949,9 @@ mac_driver_stat_create(mac_impl_t *mip)
 	major_t		major = getmajor(mip->mi_phy_dev);
 
 	count = MAC_MOD_NKSTAT + MAC_NKSTAT + mip->mi_type->mt_statcount;
-	ksp = kstat_create((const char *)ddi_major_to_name(major),
+	ksp = kstat_create_zone((const char *)ddi_major_to_name(major),
 	    getminor(mip->mi_phy_dev) - 1, MAC_KSTAT_NAME,
-	    MAC_KSTAT_CLASS, KSTAT_TYPE_NAMED, count, 0);
+	    MAC_KSTAT_CLASS, KSTAT_TYPE_NAMED, count, 0, getzoneid());
 	if (ksp == NULL)
 		return;
 
diff --git a/usr/src/uts/common/io/mr_sas/mr_sas.conf b/usr/src/uts/common/io/mr_sas/mr_sas.conf
index cfda434e23..6c585c6a42 100644
--- a/usr/src/uts/common/io/mr_sas/mr_sas.conf
+++ b/usr/src/uts/common/io/mr_sas/mr_sas.conf
@@ -13,3 +13,11 @@
 # Fast-Path specific flag. Default is "yes".
 # mrsas-enable-fp="yes";
 
+flow_control="dmult" queue="qsort" tape="sctp";
+
+# MSI specific flag. To enable MSI modify the flag value to "yes"
+mrsas-enable-msi="yes";
+
+# Fast-Path specific flag. To enable Fast-Path modify the flag value to "yes"
+mrsas-enable-fp="yes";
+
diff --git a/usr/src/uts/common/io/nfp/THIRDPARTYLICENSE b/usr/src/uts/common/io/nfp/THIRDPARTYLICENSE
new file mode 100644
index 0000000000..187088ff34
--- /dev/null
+++ b/usr/src/uts/common/io/nfp/THIRDPARTYLICENSE
@@ -0,0 +1,19 @@
+Copyright (c) 2014, Thales UK Limited
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in
+all copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
+THE SOFTWARE.
diff --git a/usr/src/uts/common/io/nfp/THIRDPARTYLICENSE.descrip b/usr/src/uts/common/io/nfp/THIRDPARTYLICENSE.descrip
new file mode 100644
index 0000000000..cde8b65b37
--- /dev/null
+++ b/usr/src/uts/common/io/nfp/THIRDPARTYLICENSE.descrip
@@ -0,0 +1 @@
+NFAST CRYPTO ACCELERATOR DRIVER
diff --git a/usr/src/uts/common/io/nfp/autoversion.h b/usr/src/uts/common/io/nfp/autoversion.h
new file mode 100644
index 0000000000..b9021942b2
--- /dev/null
+++ b/usr/src/uts/common/io/nfp/autoversion.h
@@ -0,0 +1,21 @@
+/*
+
+(C) Copyright nCipher Corporation Ltd 2002-2008 All rights reserved
+
+Copyright (c) 2008-2013 Thales e-Security All rights reserved
+
+Copyright (c) 2014 Thales UK All rights reserved
+
+*/
+
+/* AUTOGENERATED - DO NOT EDIT */
+#ifndef AUTOVERSION_H
+#define AUTOVERSION_H
+
+#define VERSION_RELEASEMAJOR 2
+#define VERSION_RELEASEMINOR 26
+#define VERSION_RELEASEPATCH 40
+#define VERSION_NO "2.26.40cam999"
+#define VERSION_COMPNAME "nfdrv"
+
+#endif
diff --git a/usr/src/uts/common/io/nfp/drvlist.c b/usr/src/uts/common/io/nfp/drvlist.c
new file mode 100644
index 0000000000..a04b1fd5b0
--- /dev/null
+++ b/usr/src/uts/common/io/nfp/drvlist.c
@@ -0,0 +1,19 @@
+/*
+
+(C) Copyright nCipher Corporation Ltd 2002-2008 All rights reserved
+
+Copyright (c) 2008-2013 Thales e-Security All rights reserved
+
+Copyright (c) 2014 Thales UK All rights reserved
+
+*/
+
+#include "nfp_common.h"
+#include "nfp_cmd.h"
+
+const nfpcmd_dev *nfp_drvlist[] = {
+  &i21285_cmddev,
+  &i21555_cmddev,
+  NULL
+};
+
diff --git a/usr/src/uts/common/io/nfp/hostif.c b/usr/src/uts/common/io/nfp/hostif.c
new file mode 100644
index 0000000000..684be703ea
--- /dev/null
+++ b/usr/src/uts/common/io/nfp/hostif.c
@@ -0,0 +1,1192 @@
+/*
+
+hostif.c: nFast PCI driver for Solaris 2.5, 2.6, 2.7 and 2.8
+
+(C) Copyright nCipher Corporation Ltd 2002-2008 All rights reserved
+
+Copyright (c) 2008-2013 Thales e-Security All rights reserved
+
+Copyright (c) 2014 Thales UK All rights reserved
+
+history
+
+06/05/1998 jsh  Original solaris 2.6
+21/05/1999 jsh  added support for solaris 2.5
+10/06/1999 jsh  added support for solaris 2.7 (32 and 64 bit)
+??/??/2001 jsh  added support for solaris 2.8 (32 and 64 bit)
+16/10/2001 jsh  moved from nfast to new structure in nfdrv
+12/02/2002 jsh  added high level interrupt support
+
+*/
+
+#include <sys/types.h>
+#include <sys/param.h>
+#include <sys/errno.h>
+#include <sys/file.h>
+#include <sys/conf.h>
+#include <sys/uio.h>
+#include <sys/map.h>
+#include <sys/debug.h>
+#include <sys/modctl.h>
+#include <sys/kmem.h>
+#include <sys/cmn_err.h>
+#include <sys/open.h>
+#include <sys/stat.h>
+#include <sys/ddi.h>
+#include <sys/sunddi.h>
+#include <sys/pci.h>
+
+#include "nfp_common.h"
+#include "nfp_hostif.h"
+#include "nfp_osif.h"
+#include "nfp_cmd.h"
+
+#include "nfp.h"
+
+/* mapped memory attributes, no-swap endianess (done in higher level) */
+static struct ddi_device_acc_attr nosw_attr = {
+  DDI_DEVICE_ATTR_V0,
+  DDI_NEVERSWAP_ACC,
+  DDI_STRICTORDER_ACC
+};
+
+/* dma attributes */
+static ddi_dma_attr_t dma_attrs = {
+  DMA_ATTR_V0,            /* version number */
+  (uint64_t)0x0,          /* low address */
+  (uint64_t)0xffffffff,   /* high address */
+  (uint64_t)0xffffff,     /* DMA counter max */
+  (uint64_t)0x1,          /* alignment */
+  0x0c,                   /* burst sizes */
+  0x1,                    /* minimum transfer size */
+  (uint64_t)0x3ffffff,    /* maximum transfer size */
+  (uint64_t)0x7fff,       /* maximum segment size */
+  1,                      /* no scatter/gather lists */
+  1,                      /* granularity */
+  0                       /* DMA flags */
+};
+
+/*
+ * Debug message control
+ * Debug Levels:
+ *  0 = no messages
+ *  1 = Errors
+ *  2 = Subroutine calls & control flow
+ *  3 = I/O Data (verbose!)
+ * Can be set with adb or in the /etc/system file with
+ * "set nfp:nfp_debug=<value>"
+ */
+
+int nfp_debug= 1;
+
+static void *state_head; /* opaque handle top of state structs */
+
+static int nfp_open(dev_t *dev, int openflags, int otyp, cred_t *credp);
+static int nfp_close(dev_t dev, int openflags, int otyp, cred_t *credp);
+static int nfp_release_dev( dev_info_t *dip );
+
+static int nfp_read(dev_t dev, struct uio *uiop, cred_t *credp);
+static int nfp_write(dev_t dev, struct uio *uiop, cred_t *credp);
+static int nfp_strategy(struct buf *bp);
+
+static int nfp_ioctl(dev_t dev, int cmd, ioctlptr_t arg, int mode, cred_t *credp, int *rvalp);
+static int nfp_chpoll(dev_t dev, short events, int anyyet, short *reventsp,
+                    struct pollhead **phpp);
+
+static void nfp_wrtimeout (void *pdev);
+static void nfp_rdtimeout (void *pdev);
+
+static int nfp_getinfo(dev_info_t *dip, ddi_info_cmd_t infocmd, void *arg, void **result);
+static int nfp_attach(dev_info_t *dip, ddi_attach_cmd_t cmd);
+static int nfp_detach(dev_info_t *dip, ddi_detach_cmd_t cmd);
+
+static void nfp_read_complete_final(nfp_dev *pdev, int ok);
+static void nfp_write_complete_final(nfp_dev *pdev, int ok);
+
+/* nfp file ops --------------------------------------------------- */
+
+static struct cb_ops nfp_cb_ops = {
+  nfp_open,
+  nfp_close,
+  nodev,                /* no nfp_strategy */
+  nodev,                /* no print routine */
+  nodev,                /* no dump routine */
+  nfp_read,
+  nfp_write,
+  nfp_ioctl,
+  nodev,                /* no devmap routine */
+  nodev,                /* no mmap routine */
+  nodev,                /* no segmap routine */
+  nfp_chpoll,
+  ddi_prop_op,
+  0,            /* not a STREAMS driver, no cb_str routine */
+  D_NEW | D_MP | EXTRA_CB_FLAGS, /* must be safe for multi-thread/multi-processor */
+  CB_REV,
+  nodev,                /* aread */
+  nodev                 /* awrite */
+};
+
+static struct dev_ops nfp_ops = {
+  DEVO_REV,               /* DEVO_REV indicated by manual */
+  0,                      /* device reference count       */
+  nfp_getinfo,
+  nulldev,                /* identify */
+  nulldev,                /* probe */
+  nfp_attach,
+  nfp_detach,
+  nodev,                  /* device reset routine         */
+  &nfp_cb_ops,
+  (struct bus_ops *)0,    /* bus operations               */
+};
+
+extern struct mod_ops mod_driverops;
+static struct modldrv modldrv = {
+  &mod_driverops,
+  NFP_DRVNAME,
+  &nfp_ops,
+};
+
+static struct modlinkage modlinkage = {
+  MODREV_1,               /* MODREV_1 indicated by manual */
+  (void *)&modldrv,
+  NULL,                   /* termination of list of linkage structures */
+};
+
+/* interface resource allocation */
+
+int nfp_alloc_pci_push( nfp_dev *pdev ) {
+  /* allocate resources needed for PCI Push,
+   * if not already allocated.
+   * return True if successful
+   */
+  nfp_err ret;
+  uint_t cookie_count;
+	size_t real_length;
+
+  if(!pdev->read_buf) {
+    /* allocate read buffer */
+    pdev->read_buf = kmem_zalloc( NFP_READBUF_SIZE, KM_NOSLEEP );
+  }
+  if(!pdev->read_buf) {
+    nfp_log( NFP_DBG1, "nfp_attach: kmem_zalloc read buffer failed");
+    pdev->read_buf = NULL;
+    return 0;
+  }
+
+  if(!pdev->rd_dma_ok) {
+    /* allocate dma handle for read buffer */
+    ret = ddi_dma_alloc_handle( pdev->dip,
+                                &dma_attrs,
+                                DDI_DMA_DONTWAIT,
+                                NULL,
+                                &pdev->read_dma_handle );
+    if( ret != DDI_SUCCESS ) {
+      nfp_log( NFP_DBG1,
+               "nfp_alloc_pci_push: ddi_dma_alloc_handle failed (%d)",
+               ret );
+      return 0;
+    }
+
+    /* Allocate the memory for dma transfers */
+    ret = ddi_dma_mem_alloc(pdev->read_dma_handle, NFP_READBUF_SIZE, &nosw_attr,
+			    DDI_DMA_CONSISTENT, DDI_DMA_DONTWAIT, NULL,
+			    (caddr_t*)&pdev->read_buf, &real_length, &pdev->acchandle);
+    if (ret != DDI_SUCCESS) {
+      nfp_log( NFP_DBG1, "nfp_alloc_pci_push: ddi_dma_mem_alloc failed (%d)", ret);
+      ddi_dma_free_handle( &pdev->read_dma_handle );
+      return 0;
+    }
+
+    ret = ddi_dma_addr_bind_handle( pdev->read_dma_handle,
+                                    NULL, /* kernel address space */
+                                    (caddr_t)pdev->read_buf, real_length,
+                                    DDI_DMA_READ | DDI_DMA_CONSISTENT, /* dma flags */
+                                    DDI_DMA_DONTWAIT, NULL,
+                                    &pdev->read_dma_cookie, &cookie_count );
+    if( ret != DDI_DMA_MAPPED ) {
+      nfp_log( NFP_DBG1,
+               "nfp_alloc_pci_push: ddi_dma_addr_bind_handle failed (%d)",
+               ret);
+      ddi_dma_mem_free(&pdev->acchandle);
+      ddi_dma_free_handle( &pdev->read_dma_handle );
+      return 0;
+    }
+    if( cookie_count > 1 ) {
+      nfp_log( NFP_DBG1,
+               "nfp_alloc_pci_push: error:"
+               " ddi_dma_addr_bind_handle wants %d transfers",
+               cookie_count);
+      ddi_dma_mem_free(&pdev->acchandle);
+      (void) ddi_dma_unbind_handle( pdev->read_dma_handle );
+      ddi_dma_free_handle( &pdev->read_dma_handle );
+      return 0;
+    }
+    pdev->rd_dma_ok = 1;
+  }
+  return pdev->rd_dma_ok;
+}
+
+void nfp_free_pci_push( nfp_dev *pdev ) {
+  /* free resources allocated to PCI Push */
+  if( pdev->rd_dma_ok ) {
+    (void) ddi_dma_sync(pdev->read_dma_handle,0,0,DDI_DMA_SYNC_FORKERNEL);
+    ddi_dma_mem_free(&pdev->acchandle);
+    (void) ddi_dma_unbind_handle( pdev->read_dma_handle );
+    ddi_dma_free_handle( &pdev->read_dma_handle );
+    pdev->rd_dma_ok = 0;
+  }
+  if( pdev->read_buf ) {
+    kmem_free( pdev->read_buf, NFP_READBUF_SIZE );
+    pdev->read_buf = NULL;
+  }
+}
+
+/* include definition of nfp_set_ifvers() */
+#define nfp_ifvers NFDEV_IF_PCI_PUSH
+#include "nfp_ifvers.c"
+#undef nfp_ifvers
+
+/*--------------------*/
+/*  nfp_isr           */
+/*--------------------*/
+
+static u_int nfp_isr( char *pdev_in ) {
+  /* LINTED: alignment */
+  nfp_dev *pdev= (nfp_dev *)pdev_in;
+  nfp_err ne;
+  int handled;
+
+  nfp_log( NFP_DBG3, "nfp_isr: entered");
+
+  if( !pdev ) {
+    nfp_log( NFP_DBG1, "nfp_isr: cannot find dev");
+    return DDI_INTR_UNCLAIMED;
+  }
+
+  /* The isr needs to be mutex'ed - an SMP can call us while we're still
+   * running!
+   */
+  mutex_enter(&pdev->low_mutex);
+  ne= pdev->cmddev->isr( pdev->common.cmdctx, &handled );
+  mutex_exit(&pdev->low_mutex);
+
+  if( !ne && handled )
+    return DDI_INTR_CLAIMED;
+  if (ne)
+    nfp_log( NFP_DBG1, "nfp_isr: failed");
+  else
+    nfp_log( NFP_DBG3, "nfp_isr: unclaimed");
+  return DDI_INTR_UNCLAIMED;
+}
+
+static u_int nfp_soft_isr( char *pdev_in ) {
+  /* LINTED: alignment */
+  nfp_dev *pdev= (nfp_dev *)pdev_in;
+  int rd, wr;
+
+  nfp_log( NFP_DBG3, "nfp_soft_isr: entered");
+
+  if( !pdev ) {
+    nfp_log( NFP_DBG1, "nfp_soft_isr: cannot find dev");
+    return DDI_INTR_UNCLAIMED;
+  }
+  rd= wr= 0;
+  
+  mutex_enter(&pdev->high_mutex);
+  if(pdev->high_read) {
+    pdev->high_read= 0;
+    mutex_exit(&pdev->high_mutex);
+    rd= 1;
+  }
+  if(pdev->high_write) {
+    pdev->high_write= 0;
+    wr= 1;
+  }
+  mutex_exit(&pdev->high_mutex);
+
+  if(rd) {
+    nfp_log( NFP_DBG3, "nfp_soft_isr: read done");
+    nfp_read_complete_final(pdev, pdev->rd_ok);
+  }
+  if(wr) {
+    nfp_log( NFP_DBG3, "nfp_soft_isr: write done");
+    nfp_write_complete_final(pdev, pdev->wr_ok);
+  }
+  if( rd || wr )
+    return DDI_INTR_CLAIMED;
+
+  nfp_log( NFP_DBG2, "nfp_isr: unclaimed");
+  return DDI_INTR_UNCLAIMED;
+}
+
+
+/*-------------------------*/
+/*  nfp_read               */
+/*-------------------------*/
+
+void nfp_read_complete(nfp_dev *pdev, int ok) {
+  nfp_log( NFP_DBG2,"nfp_read_complete: entering");
+
+  if(pdev->high_intr) {
+    nfp_log(NFP_DBG2, "nfp_read_complete: high_intr");
+    mutex_enter(&pdev->high_mutex);
+    nfp_log(NFP_DBG3, "nfp_read_complete: high_mutex entered");
+    if(pdev->high_read)
+      nfp_log(NFP_DBG1, "nfp_read_complete: high_read allread set!");
+    pdev->high_read= 1;
+    pdev->rd_ok= ok;
+    nfp_log(NFP_DBG3, "nfp_read_complete: exiting high_mutex");
+    mutex_exit(&pdev->high_mutex);
+    ddi_trigger_softintr(pdev->soft_int_id);
+  } else
+    nfp_read_complete_final( pdev, ok );
+  nfp_log( NFP_DBG2,"nfp_read_complete: exiting");
+}
+
+static void nfp_read_complete_final(nfp_dev *pdev, int ok) {
+  nfp_log( NFP_DBG2,"nfp_read_complete_final: entering");
+  if(pdev->rdtimeout)
+    (void) untimeout(pdev->rdtimeout);
+  if(!pdev->rd_outstanding) {
+    nfp_log( NFP_DBG1,"nfp_read_complete_final: !pdev->rd_outstanding");
+  }
+  nfp_log( NFP_DBG2,"nfp_read_complete_final: pdev->rd_outstanding=0, ok %d", ok);
+  mutex_enter(&pdev->isr_mutex);
+  pdev->rd_outstanding= 0;
+  pdev->rd_ready= 1;
+  pdev->rd_ok= ok;
+  cv_broadcast(&pdev->rd_cv);
+  mutex_exit(&pdev->isr_mutex);
+  pollwakeup (&pdev->pollhead, POLLRDNORM);
+  nfp_log( NFP_DBG2,"nfp_read_complete_final: exiting");
+}
+
+static void nfp_rdtimeout( void *pdev_in )
+{
+  nfp_dev *pdev= (nfp_dev *)pdev_in;
+
+  nfp_log( NFP_DBG1, "nfp_rdtimeout: read timed out");
+
+  if (!pdev) {
+    nfp_log( NFP_DBG1, "nfp_rdtimeout: NULL pdev." );
+    return;
+  }
+  pdev->rdtimeout= 0;
+  nfp_read_complete_final(pdev, 0);
+}
+
+/* ARGSUSED */
+static int nfp_read(dev_t dev, struct uio *uiop, cred_t *credp) {
+  int ret;
+  nfp_log( NFP_DBG2, "nfp_read: entered" );
+  if (ddi_get_soft_state(state_head, getminor(dev)) != NULL) {
+    nfp_log( NFP_DBG1, "nfp_read: unable to get nfp_dev");
+    return (ENODEV);
+  }
+  nfp_log( NFP_DBG2, "nfp_read: about to physio." );
+  ret = physio(nfp_strategy, (struct buf *)0, dev, B_READ, minphys, uiop );
+  if(ret)
+    nfp_log( NFP_DBG1, "nfp_read: physio returned %x.", ret );
+  return ret;
+}
+
+/*-------------------------*/
+/*  nfp_write              */
+/*-------------------------*/
+
+void nfp_write_complete( nfp_dev *pdev, int ok) {
+  nfp_log( NFP_DBG2,"nfp_write_complete: entering");
+
+  if(pdev->high_intr) {
+    mutex_enter(&pdev->high_mutex);
+    if(pdev->high_write)
+      nfp_log(NFP_DBG1, "nfp_write_complete: high_write allread set!");
+    pdev->high_write= 1;
+    pdev->wr_ok= ok;
+    mutex_exit(&pdev->high_mutex);
+    ddi_trigger_softintr(pdev->soft_int_id);
+  } else
+    nfp_write_complete_final( pdev, ok );
+  nfp_log( NFP_DBG2,"nfp_write_complete: exiting");
+}
+
+static void nfp_write_complete_final( nfp_dev *pdev, int ok) {
+  struct buf *local_wr_bp;
+  nfp_log( NFP_DBG2,"nfp_write_complete_final: entering");
+  if(pdev->wrtimeout)
+    (void) untimeout(pdev->wrtimeout);
+
+  if (!pdev->wr_bp) {
+    nfp_log( NFP_DBG2, "nfp_write_complete_final: write: wr_bp == NULL." );
+    return;
+  }
+
+  bp_mapout(pdev->wr_bp);
+  pdev->wr_bp->b_resid = ok ? 0 : pdev->wr_bp->b_bcount;
+  /* Make sure we set wr_ready before calling biodone to avoid a race */
+  pdev->wr_ready = 1;
+  bioerror(pdev->wr_bp, ok ? 0 : ENXIO);
+  local_wr_bp = pdev->wr_bp;
+  pdev->wr_bp = 0;
+  biodone(local_wr_bp);
+  nfp_log( NFP_DBG2, "nfp_write_complete_final: isr_mutex extited");
+  pollwakeup (&pdev->pollhead, POLLWRNORM);
+
+  nfp_log( NFP_DBG2, "nfp_write_complete_final: leaving");
+}
+
+static void nfp_wrtimeout( void *pdev_in )
+{
+  nfp_dev *pdev= (nfp_dev *)pdev_in;
+
+  nfp_log( NFP_DBG1, "nfp_wrtimeout: write timed out");
+
+  if (!pdev) {
+    nfp_log( NFP_DBG1, "nfp_wrtimeout: NULL pdev." );
+    return;
+  }
+  pdev->wrtimeout= 0;
+  nfp_write_complete_final(pdev, 0);
+}
+
+/* ARGSUSED */
+static int nfp_write(dev_t dev, struct uio *uiop, cred_t *credp) {
+  int ret;
+  nfp_log( NFP_DBG2, "nfp_write: entered." );
+  if (ddi_get_soft_state(state_head, getminor(dev)) == NULL) {
+    nfp_log( NFP_DBG1, "nfp_chread: unable to get nfp_dev.");
+    return (ENODEV);
+  }
+  nfp_log( NFP_DBG2, "nfp_write: about to physio." );
+  ret = physio(nfp_strategy, (struct buf *)0, dev, B_WRITE, minphys, uiop );
+  if(ret)
+    nfp_log( NFP_DBG1, "nfp_write: physio returned %x.", ret );
+  return ret;
+}
+
+/*-------------------------*/
+/*  nfp_strategy           */
+/*-------------------------*/
+
+#define NFP_STRAT_ERR(thebp,err,txt) \
+      nfp_log( NFP_DBG1, "nfp_strategy: " txt ".\n"); \
+      (thebp)->b_resid = (thebp)->b_bcount; \
+      bioerror ((thebp), err); \
+      biodone ((thebp));
+
+static int nfp_strategy(struct buf *bp) {
+  register struct nfp_dev *pdev;
+  nfp_err ne;
+  
+  nfp_log( NFP_DBG2, "nfp_strategy: entered." );
+  if (!(pdev = ddi_get_soft_state(state_head, getminor(bp->b_edev)))) {
+    NFP_STRAT_ERR (bp, ENXIO, "unable to get nfp_dev");
+    return (0);
+  }
+
+  if (bp->b_flags & B_READ) {
+    int count;
+    /* read */
+    if (!pdev->rd_ready) {
+      NFP_STRAT_ERR (bp,ENXIO,"read called when not ready");
+      return (0);
+    }
+    pdev->rd_ready=0;
+    pdev->rd_pending = 0;
+    if( !pdev->rd_ok) {
+      NFP_STRAT_ERR (bp,ENXIO,"read failed");
+      return (0);
+    }
+    /* copy data from module */
+    if(pdev->ifvers >= NFDEV_IF_PCI_PUSH) {
+      nfp_log( NFP_DBG3, "nfp_strategy: copying kernel read buffer");
+      if( ddi_dma_sync(pdev->read_dma_handle,0,0,DDI_DMA_SYNC_FORKERNEL) != DDI_SUCCESS )
+      {
+        NFP_STRAT_ERR(bp,ENXIO,"ddi_dma_sync(read_dma_handle) failed");
+        return (0);
+      }
+      /* LINTED: alignment */
+      count= *(unsigned int *)(pdev->read_buf+4);
+      count= FROM_LE32_MEM(&count);
+      nfp_log( NFP_DBG3, "nfp_strategy: read count %d", count);
+      if(count<0 || count>bp->b_bcount) {
+        NFP_STRAT_ERR(bp,ENXIO,"bad read byte count from device");
+        nfp_log( NFP_DBG1, "nfp_strategy: bad read byte count (%d) from device", count);
+        return (0);
+      }
+      bp_mapin (bp);
+      bcopy( pdev->read_buf + 8, bp->b_un.b_addr, count );
+      bp_mapout (bp);
+    } else {
+      bp_mapin (bp);
+      ne=  pdev->cmddev->read_block( bp->b_un.b_addr, bp->b_bcount, pdev->common.cmdctx, &count );
+      bp_mapout (bp);
+      if( ne != NFP_SUCCESS) {
+        NFP_STRAT_ERR (bp,nfp_oserr(ne),"read_block failed");
+        return (0);
+      }
+    }
+    bioerror(bp, 0);
+    bp->b_resid = 0;
+    biodone (bp);
+  } else {
+    /* write */
+    if (!pdev->wr_ready) {
+      NFP_STRAT_ERR (bp,ENXIO,"write called when not ready");
+      return (0);
+    }
+    if (pdev->wr_bp) {
+      NFP_STRAT_ERR (bp,ENXIO,"wr_bp != NULL");
+      return (0);
+    }
+    pdev->wrtimeout= timeout(nfp_wrtimeout, (caddr_t)pdev, NFP_TIMEOUT_SEC * drv_usectohz(1000000));
+    pdev->wr_bp = bp;
+    pdev->wr_ready = 0;
+    bp_mapin (bp);
+    ne= pdev->cmddev->write_block( bp->b_un.b_addr, bp->b_bcount,  pdev->common.cmdctx);
+    if( ne != NFP_SUCCESS ) {
+      bp_mapout (bp);
+      (void) untimeout(pdev->wrtimeout);
+      pdev->wr_bp = 0;
+      pdev->wr_ready = 1;
+      NFP_STRAT_ERR (bp,nfp_oserr(ne),"write failed");
+      return (0);
+    }
+  }
+  nfp_log( NFP_DBG2, "nfp_strategy: leaving");
+
+  return (0);
+}
+
+
+/*--------------------*/
+/*  poll / select     */
+/*--------------------*/
+
+static int nfp_chpoll(dev_t dev, short events, int anyyet, short *reventsp,
+                      struct pollhead **phpp) {
+  nfp_dev *pdev;
+  short revents;
+
+  if (!(pdev = ddi_get_soft_state(state_head, getminor(dev)))) {
+    nfp_log( NFP_DBG1, "nfp_chpoll: unable to get nfp_dev");
+    *reventsp=0;
+    return (0);
+  }
+  nfp_log( NFP_DBG2, "nfp_chpoll: entered %x", events);
+
+  revents=0;
+  if (events&POLLWRNORM) {
+    if (pdev->wr_ready) {
+      nfp_log( NFP_DBG2, "nfp_chpoll: write ready");
+      revents|=POLLWRNORM;
+    }
+  }
+
+  if (events&POLLRDNORM) {
+    if (pdev->rd_ready) {
+      nfp_log( NFP_DBG2, "nfp_chpoll: read ready");
+      revents|=POLLRDNORM;
+    }
+  }
+
+  if (!revents && !anyyet) {
+    *phpp=&pdev->pollhead;
+  }
+  *reventsp=revents;
+
+  nfp_log( NFP_DBG2, "nfp_chpoll: leaving");
+  return (0);
+}
+
+
+/*--------------------*/
+/*  ioctl             */
+/*--------------------*/
+
+/* ARGSUSED */
+static int nfp_ioctl(dev_t dev, int cmd, ioctlptr_t arg, int mode, cred_t *credp, int *rvalp) {
+  register struct nfp_dev *pdev;
+
+  nfp_log( NFP_DBG2, "nfp_ioctl: entered." );
+
+  if (!(pdev = ddi_get_soft_state(state_head, getminor(dev)))) {
+    nfp_log( NFP_DBG1, "nfp_ioctl: unable to get nfp dev.");
+    return (ENXIO);
+  }
+
+  switch (cmd) {
+  case NFDEV_IOCTL_ENQUIRY:
+    {
+      long *outp;
+      int outlen;
+      nfdev_enquiry_str enq_data;
+
+      enq_data.busno = (unsigned int)-1;
+      enq_data.slotno = (unsigned char)-1;
+
+      /* get our bus and slot num */
+      if (ddi_getlongprop (DDI_DEV_T_NONE,
+                           pdev->dip, 0, "reg",
+                           (caddr_t)&outp, &outlen) != DDI_PROP_NOT_FOUND) {
+        nfp_log( NFP_DBG2, "ddi_getlongprop('reg') ok." );
+        if( outlen > 0 ) {
+          enq_data.busno = ((*outp)>>16) & 0xff;
+          enq_data.slotno = ((*outp)>>11) & 0x1f;
+          nfp_log( NFP_DBG2, "busno %d, slotno %d.",
+                   enq_data.busno, enq_data.slotno );
+        }
+      } else
+        nfp_log( NFP_DBG1, "ddi_getlongprop('reg') failed." );
+
+      if( ddi_copyout( (char *)&enq_data, (void *)arg, sizeof(enq_data), mode ) != 0 ) {
+        nfp_log( NFP_DBG1, "ddi_copyout() failed." );
+        return EFAULT;
+      }
+    }
+    break;
+
+  case NFDEV_IOCTL_ENSUREREADING:
+    {
+      unsigned int addr, len;
+      nfp_err  ret;
+      if( ddi_copyin( (void *)arg, (char *)&len, sizeof(unsigned int), mode ) != 0 ) {
+        nfp_log( NFP_DBG1, "ddi_copyin() failed." );
+        return (EFAULT);
+      }
+      /* signal a read to the module */
+      nfp_log( NFP_DBG2, "nfp_ioctl: signalling read request to module, len = %x.", len );
+      if (len>8192) {
+        nfp_log( NFP_DBG1, "nfp_ioctl: len >8192 = %x.", len );
+        return EINVAL;
+      }
+      if (pdev->rd_outstanding==1) {
+        nfp_log( NFP_DBG1, "nfp_ioctl: not about to call read with read outstanding.");
+        return EIO;
+      }
+
+      addr= 0;
+      if(pdev->ifvers >= NFDEV_IF_PCI_PUSH) {
+        if( len > NFP_READBUF_SIZE ) {
+          nfp_log( NFP_DBG1, "nfp_ioctl: len > NFP_READBUF_SIZE = %x.", len );
+          return EINVAL;
+        }
+        addr= pdev->read_dma_cookie.dmac_address;
+      }
+
+      pdev->rd_outstanding = 1;
+      nfp_log( NFP_DBG2,"nfp_ioctl: pdev->rd_outstanding=1");
+
+      /* setup timeout timer */
+      pdev->rdtimeout= timeout(nfp_rdtimeout, (caddr_t)pdev, NFP_TIMEOUT_SEC * drv_usectohz(1000000));
+
+      nfp_log( NFP_DBG2, "nfp_ioctl: read request");
+      ret = pdev->cmddev->ensure_reading(addr, len, pdev->common.cmdctx);
+      if ( ret != NFP_SUCCESS ) {
+        (void) untimeout(pdev->rdtimeout);
+        pdev->rdtimeout = 0;
+        pdev->rd_outstanding = 0;
+        nfp_log( NFP_DBG1, "nfp_ioctl : cmddev->ensure_reading failed ");
+        return nfp_oserr( ret );
+      }
+    }
+    break;
+
+  case NFDEV_IOCTL_PCI_IFVERS:
+    {
+      int vers;
+
+      nfp_log( NFP_DBG2, "nfp_ioctl: NFDEV_IOCTL_PCI_IFVERS");
+
+      if( ddi_copyin( (void *)arg, (char *)&vers, sizeof(vers), mode ) != 0 ) {
+        nfp_log( NFP_DBG1, "ddi_copyin() failed." );
+        return (EFAULT);
+      }
+
+      if( pdev->rd_outstanding ) {
+        nfp_log( NFP_DBG1, "nfp_ioctl: can't set ifvers %d as read outstanding", vers);
+        return EIO;
+      }
+
+      nfp_set_ifvers(pdev, vers);
+      if( pdev->ifvers != vers ) {
+        nfp_log( NFP_DBG1, "nfp_ioctl: can't set ifvers %d", vers);
+        return EIO;
+      }
+    }
+    break;
+
+  case NFDEV_IOCTL_STATS:
+    {
+      if( ddi_copyout( (char *)&(pdev->common.stats),
+                       (void *)arg,
+                       sizeof(nfdev_stats_str),
+                       mode ) != 0 ) {
+        nfp_log( NFP_DBG1, "ddi_copyout() failed." );
+        return EFAULT;
+      }
+    }
+    break;
+
+  default:
+    nfp_log( NFP_DBG1, "nfp_ioctl: unknown ioctl." );
+    return EINVAL;
+  }
+
+  return 0;
+}
+
+/*-------------------------*/
+/*  nfp_open               */
+/*-------------------------*/
+
+/* ARGSUSED */
+int nfp_open(dev_t *dev, int openflags, int otyp, cred_t *credp)
+{       
+  nfp_err ret;
+  register struct nfp_dev *pdev;
+ 
+  nfp_log( NFP_DBG2, "entered nfp_open." );
+      
+  pdev = (nfp_dev *)ddi_get_soft_state(state_head, getminor(*dev));
+  
+  if( !pdev ) {
+    nfp_log( NFP_DBG1, "nfp_open: unable to get nfp dev.");
+    return (ENODEV);
+  }     
+        
+  if( otyp != OTYP_CHR ) {
+    nfp_log( NFP_DBG1, "nfp_open: not opened as character device");
+    return (EINVAL);
+  } 
+    
+  mutex_enter(&pdev->busy_mutex);
+    
+  if (pdev->busy) {
+    mutex_exit(&pdev->busy_mutex);
+    nfp_log( NFP_DBG1, "nfp_open: device busy");
+    return EBUSY;
+  } 
+  pdev->busy= 1;
+  mutex_exit(&pdev->busy_mutex);
+
+  /* use oldest possible interface until told otherwise */
+  pdev->ifvers= NFDEV_IF_STANDARD;
+  nfp_log( NFP_DBG3, "nfp_open: setting ifvers %d", pdev->ifvers);
+  pdev->rd_ready= 0; /* drop any old data */
+ 
+  ret = pdev->cmddev->open(pdev->common.cmdctx);
+  if( ret != NFP_SUCCESS ) {
+    nfp_log( NFP_DBG1, "nfp_open : cmddev->open failed ");
+    return nfp_oserr( ret );
+  } 
+
+  nfp_log( NFP_DBG2, "nfp_open: done");
+
+  return 0;
+}
+
+/*--------------------*/
+/*  nfp_close         */
+/*--------------------*/
+
+/* ARGSUSED */
+static int nfp_close(dev_t dev, int openflags, int otyp, cred_t *credp) {
+  nfp_dev *pdev;
+  nfp_err ret;
+
+  nfp_log( NFP_DBG2, "nfp_close: entered");
+
+  pdev = (struct nfp_dev *)ddi_get_soft_state(state_head, getminor(dev));
+  if( !pdev ) {
+    nfp_log( NFP_DBG1, "nfp_close: cannot find dev.");
+    return ENODEV;
+  }
+
+  mutex_enter(&pdev->isr_mutex);
+  if(pdev->rd_outstanding) {
+    int lbolt, err;
+    nfp_get_lbolt(&lbolt, err);
+    if(!err)
+      (void) cv_timedwait(&pdev->rd_cv, &pdev->isr_mutex, lbolt + (NFP_TIMEOUT_SEC * drv_usectohz(1000000)) );
+  }
+  mutex_exit(&pdev->isr_mutex);
+  ret = pdev->cmddev->close(pdev->common.cmdctx);
+  if (ret != NFP_SUCCESS ) {
+    nfp_log( NFP_DBG1, " nfp_close : cmddev->close failed");
+    return nfp_oserr( ret );
+  }
+
+  mutex_enter(&pdev->busy_mutex);
+  pdev->busy= 0;
+  mutex_exit(&pdev->busy_mutex);
+
+  return 0;
+}
+
+/****************************************************************************
+
+  nfp driver config
+
+ ****************************************************************************/
+
+/*-------------------------*/
+/*  nfp_getinfo            */
+/*-------------------------*/
+
+/* ARGSUSED */
+static int nfp_getinfo(dev_info_t *dip, ddi_info_cmd_t infocmd, void *arg, void **result) {
+  int error;
+  nfp_dev *pdev;
+
+  nfp_log( NFP_DBG2, "nfp_getinfo: entered" );
+
+  pdev = (struct nfp_dev *)ddi_get_soft_state(state_head, getminor((dev_t)arg));
+  if( !pdev ) {
+    nfp_log( NFP_DBG1, "nfp_close: cannot find dev.");
+    return ENODEV;
+  }
+
+  switch (infocmd) {
+  case DDI_INFO_DEVT2DEVINFO:
+    if (pdev == NULL) {
+      *result = NULL;
+      error = DDI_FAILURE;
+    } else {
+      /*
+       * don't need to use a MUTEX even though we are
+       * accessing our instance structure; dev->dip
+       * never changes.
+       */
+      *result = pdev->dip;
+      error = DDI_SUCCESS;
+    }
+    break;
+  case DDI_INFO_DEVT2INSTANCE:
+    *result = (void *)(uintptr_t)getminor((dev_t)arg);
+    error = DDI_SUCCESS;
+    break;
+  default:
+    *result = NULL;
+    error = DDI_FAILURE;
+  }
+
+  nfp_log( NFP_DBG2, "nfp_getinfo: leaving." );
+  return (error);
+}
+
+/*-------------------------*/
+/*  nfp_release            */
+/*-------------------------*/
+
+static int nfp_release_dev( dev_info_t *dip ) {
+  nfp_dev *pdev;
+  int instance, i;
+  nfp_err ret;
+
+  nfp_log( NFP_DBG2, "nfp_release_dev: entering" );
+
+  instance = ddi_get_instance(dip);
+  pdev = (struct nfp_dev *)ddi_get_soft_state(state_head, instance);
+  if (pdev) {
+    nfp_log( NFP_DBG3, "nfp_release_dev: removing device" );
+
+    nfp_free_pci_push(pdev);
+
+    if( pdev->cmddev ) {
+      nfp_log( NFP_DBG3, "nfp_release_dev: destroying cmd dev" );
+       ret = pdev->cmddev->destroy(pdev->common.cmdctx);
+       if (ret != NFP_SUCCESS) {
+         nfp_log( NFP_DBG1, " nfp_release_dev : cmddev->destroy failed ");
+         return nfp_oserr( ret );
+       }
+    }
+
+    if(pdev->high_iblock_cookie) {
+      nfp_log( NFP_DBG3, "nfp_release_dev: removing high and soft irq" );
+      ddi_remove_softintr(pdev->soft_int_id);
+      ddi_remove_intr(pdev->dip, 0, pdev->high_iblock_cookie);
+      mutex_destroy( &pdev->busy_mutex );
+      cv_destroy( &pdev->rd_cv );
+      mutex_destroy( &pdev->isr_mutex );
+      mutex_destroy( &pdev->high_mutex );
+    } else if(pdev->iblock_cookie) {
+      nfp_log( NFP_DBG3, "nfp_release_dev: removing irq" );
+      ddi_remove_intr(pdev->dip, 0, pdev->iblock_cookie);
+      mutex_destroy( &pdev->busy_mutex );
+      cv_destroy( &pdev->rd_cv );
+      mutex_destroy( &pdev->isr_mutex );
+    }
+    if(pdev->low_iblock_cookie) {
+      ddi_remove_intr(pdev->dip, 0, pdev->low_iblock_cookie);
+      mutex_destroy( &pdev->low_mutex);
+    }
+
+    for(i=0;i<6;i++) {
+      if( pdev->common.extra[i] ) {
+        nfp_log( NFP_DBG3, "nfp_release_dev: unmapping BAR %d", i );
+        ddi_regs_map_free ((ddi_acc_handle_t *)&pdev->common.extra[i]);
+      }
+    }
+
+    ddi_remove_minor_node(dip, NULL);
+
+    if (pdev->conf_handle)
+      pci_config_teardown( &pdev->conf_handle );
+
+    ddi_soft_state_free(state_head, instance);
+  }
+  nfp_log( NFP_DBG2, "nfp_release: finished" );
+
+  return DDI_SUCCESS;
+}
+
+
+/*-------------------------*/
+/*  nfp_attach             */
+/*-------------------------*/
+
+static int nfp_attach(dev_info_t *dip, ddi_attach_cmd_t cmd) {
+  int instance;
+  nfp_dev *pdev = NULL;
+  int intres;
+  uint16_t device, vendor, sub_device, sub_vendor;
+  long *outp;
+  nfpcmd_dev const *cmddev;
+  int index, i;
+  nfp_err ret;
+
+  nfp_log( NFP_DBG2, "nfp_attach: entered." );
+
+  if (cmd != DDI_ATTACH) {
+    nfp_log( NFP_DBG1, "nfp_attach: bad command." );
+    goto bailout;
+  }
+
+  instance = ddi_get_instance(dip);
+
+  if (ddi_soft_state_zalloc(state_head, instance) != 0) {
+    nfp_log( NFP_DBG1, "nfp_attach: ddi_soft_state_zalloc() failed." );
+    goto bailout;
+  }
+
+  pdev = (struct nfp_dev *)ddi_get_soft_state(state_head, instance);
+  if( !pdev ) {
+    nfp_log( NFP_DBG1, "nfp_attach: cannot find dev.");
+    return ENODEV;
+  }
+  pdev->dip = dip;
+
+  /* map in pci config registers */
+  if (pci_config_setup(dip, &pdev->conf_handle)) {
+    nfp_log( NFP_DBG1, "nfp_attach: pci_config_setup() failed." );
+    goto bailout;
+  }
+
+  /* find out what we have got */
+  vendor= PCI_CONFIG_GET16( pdev->conf_handle, PCI_CONF_VENID );
+  device = PCI_CONFIG_GET16( pdev->conf_handle, PCI_CONF_DEVID );
+  sub_vendor = PCI_CONFIG_GET16( pdev->conf_handle, PCI_CONF_SUBVENID );
+  sub_device = PCI_CONFIG_GET16( pdev->conf_handle, PCI_CONF_SUBSYSID );
+
+  index= 0;
+  while( (cmddev = nfp_drvlist[index++]) != NULL ) {
+    if( cmddev->vendorid == vendor &&
+        cmddev->deviceid == device &&
+        cmddev->sub_vendorid == sub_vendor &&
+        cmddev->sub_deviceid == sub_device )
+      break;
+  }
+  if( !cmddev ) {
+    nfp_log( NFP_DBG1, "nfp_attach: unknonw device." );
+    goto bailout;
+  }
+
+  /* map BARs */
+  for( i=0; i<6; i++ ) {
+    if( cmddev->bar_sizes[i] ) {
+      off_t size;
+      if( ddi_dev_regsize(dip, i+1, &size) != DDI_SUCCESS) {
+        nfp_log( NFP_DBG1, "nfp_attach: ddi_dev_regsize() failed for BAR %d", i );
+        goto bailout;
+      }
+      if( size < (cmddev->bar_sizes[i] & ~NFP_MEMBAR_MASK) ) { 
+        nfp_log( NFP_DBG1, "nfp_attach: BAR %d too small %x (%x)", i, size, (cmddev->bar_sizes[i] & ~0xF) );
+        goto bailout;
+      }
+      if (ddi_regs_map_setup(dip, i+1, (caddr_t *)&pdev->common.bar[i],
+                         0, cmddev->bar_sizes[i] & ~NFP_MEMBAR_MASK, &nosw_attr, (ddi_acc_handle_t *)&pdev->common.extra[i] )) { 
+        nfp_log( NFP_DBG1, "nfp_attach: ddi_regs_map_setup() failed for BAR %d", i );
+        goto bailout;
+      }
+      nfp_log( NFP_DBG3, "nfp_attach: BAR[%d] mapped to %x (%x)", i, pdev->common.bar[i], size );
+    }
+  }
+  
+  pdev->read_buf = NULL;
+  pdev->rd_dma_ok = 0;
+
+  /* attach to minor node */
+  if (ddi_create_minor_node(dip, "nfp", S_IFCHR, instance, (char *)cmddev->name, 0) == DDI_FAILURE) {
+    ddi_remove_minor_node(dip, NULL);
+    nfp_log( NFP_DBG1, "nfp_attach: ddi_create_minor_node() failed." );
+    goto bailout;
+  }
+  
+  pdev->wr_ready = 1;
+  pdev->rd_ready = 0;
+  pdev->rd_pending = 0;
+  pdev->rd_outstanding = 0;
+  pdev->busy=0; 
+  pdev->cmddev= cmddev;
+  
+  ret = pdev->cmddev->create(&pdev->common);
+  if( ret != NFP_SUCCESS) {
+    nfp_log( NFP_DBG1, "nfp_attach: failed to create command device");
+    goto bailout;
+  }
+  pdev->common.dev= pdev;
+
+  if (ddi_intr_hilevel(dip, 0) != 0){
+    nfp_log( NFP_DBG2, "nfp_attach: high-level interrupt");
+    if( ddi_get_iblock_cookie(dip, 0, &pdev->high_iblock_cookie) ) {
+      nfp_log( NFP_DBG1, "nfp_attach: ddi_get_iblock_cookie(high) failed." );
+      goto bailout;
+    } 
+    if( ddi_get_iblock_cookie(dip, 0, &pdev->low_iblock_cookie) ) {
+      nfp_log( NFP_DBG1, "nfp_attach: ddi_get_iblock_cookie(low) failed." );
+      goto bailout;
+    }
+    mutex_init(&pdev->high_mutex, NULL, MUTEX_DRIVER,
+                (void *)pdev->high_iblock_cookie);
+    mutex_init(&pdev->low_mutex, NULL, MUTEX_DRIVER,
+                (void *)pdev->low_iblock_cookie);
+    if (ddi_add_intr(dip, 0, NULL,
+                NULL, nfp_isr,
+                (caddr_t)pdev) != DDI_SUCCESS) {
+      nfp_log( NFP_DBG1, "nfp_attach: ddi_add_intr(high) failed." );
+      goto bailout;
+    }
+    if( ddi_get_soft_iblock_cookie(dip, DDI_SOFTINT_HIGH,
+                &pdev->iblock_cookie) ) {
+      nfp_log( NFP_DBG1, "nfp_attach: ddi_get_iblock_cookie(soft) failed." );
+      goto bailout;
+    }
+    mutex_init(&pdev->isr_mutex, NULL, MUTEX_DRIVER,
+                (void *)pdev->iblock_cookie);
+    if (ddi_add_softintr(dip, DDI_SOFTINT_HIGH, &pdev->soft_int_id,
+                &pdev->iblock_cookie, NULL,
+                nfp_soft_isr, (caddr_t)pdev) != DDI_SUCCESS)
+                goto bailout;
+    pdev->high_intr= 1;
+  } else {
+    nfp_log( NFP_DBG2, "nfp_attach: low-level interrupt");
+
+    if (ddi_get_iblock_cookie (dip, 0, &pdev->iblock_cookie)) {
+      nfp_log( NFP_DBG1, "nfp_attach: ddi_get_iblock_cookie() failed." );
+      goto bailout;
+    }
+  
+    mutex_init(&pdev->isr_mutex, "nfp isr mutex", MUTEX_DRIVER, (void *)pdev->iblock_cookie);
+  
+    if (ddi_add_intr(dip, 0, NULL,
+                     (ddi_idevice_cookie_t *)NULL, nfp_isr,
+                     (caddr_t)pdev) != DDI_SUCCESS) {
+      nfp_log( NFP_DBG1, "nfp_attach: ddi_add_intr() failed." );
+      goto bailout;
+    }
+  }
+  mutex_init(&pdev->busy_mutex, "nfp busy mutex", MUTEX_DRIVER, NULL );
+  cv_init(&pdev->rd_cv, "nfp read condvar", CV_DRIVER, NULL );
+
+  /* get our bus and slot num */
+  if (ddi_getlongprop (DDI_DEV_T_NONE, 
+                       pdev->dip, 0, "reg",
+                       (caddr_t)&outp, &intres) != DDI_PROP_NOT_FOUND) {
+    nfp_log( NFP_DBG2, "nfp_attach: ddi_getlongprop('reg') ok." );
+    if( intres > 0 ) {
+      nfp_log( NFP_DBG1, "nfp_attach: found PCI nfast bus %x slot %x.",
+               ((*outp)>>16) & 0xff, ((*outp)>>11) & 0x1f );
+    }
+  }
+  
+  nfp_log( NFP_DBG2, "nfp_attach: attach succeeded." );
+  return DDI_SUCCESS;
+  
+bailout:
+  (void) nfp_release_dev( dip );
+
+  return DDI_FAILURE;
+}
+
+/*-------------------------*/
+/*  nfp_detach             */
+/*-------------------------*/
+
+/* 
+ * When our driver is unloaded, nfp_detach cleans up and frees the resources
+ * we allocated in nfp_attach.
+ */
+static int nfp_detach(dev_info_t *dip, ddi_detach_cmd_t cmd) {
+  if (cmd != DDI_DETACH)
+    return (DDI_FAILURE);
+
+  (void) nfp_release_dev(dip);
+
+  return (DDI_SUCCESS);
+}
+
+/*-------------------------*/
+/*  _init                  */
+/*-------------------------*/
+
+int _init(void) {
+  register int error;
+
+  nfp_log( NFP_DBG2, "_init: entered" );
+
+  if ((error = ddi_soft_state_init(&state_head, sizeof (struct nfp_dev), 1)) != 0) {
+    nfp_log( NFP_DBG1, "_init: soft_state_init() failed" );
+    return (error);
+  }
+  
+  if ((error = mod_install(&modlinkage)) != 0) {
+    nfp_log( NFP_DBG1, "_init: mod_install() failed" );
+    ddi_soft_state_fini(&state_head);
+  }
+  
+  nfp_log( NFP_DBG2, "_init: leaving" );
+  return (error);
+}
+
+/*-------------------------*/
+/*  _info                  */
+/*-------------------------*/
+
+int _info(struct modinfo *modinfop) { 
+  nfp_log( NFP_DBG2, "_info: entered" );
+  
+  return (mod_info(&modlinkage, modinfop));
+}
+
+/*-------------------------*/
+/*  _fini                  */
+/*-------------------------*/
+
+int _fini(void) {
+  int status;
+  
+  nfp_log( NFP_DBG2, "_fini: entered" );
+  
+  if ((status = mod_remove(&modlinkage)) != 0) {
+    nfp_log( NFP_DBG2, "_fini: mod_remove() failed." );
+    return (status);
+  }
+  
+  ddi_soft_state_fini(&state_head);
+  
+  nfp_log( NFP_DBG2, "_fini: leaving" );
+  
+  return (status);
+}
+
diff --git a/usr/src/uts/common/io/nfp/i21285.c b/usr/src/uts/common/io/nfp/i21285.c
new file mode 100644
index 0000000000..f51a09188d
--- /dev/null
+++ b/usr/src/uts/common/io/nfp/i21285.c
@@ -0,0 +1,310 @@
+/*
+
+i21285.c: nCipher PCI HSM intel/digital 21285 command driver
+
+(C) Copyright nCipher Corporation Ltd 2002-2008 All rights reserved
+
+Copyright (c) 2008-2013 Thales e-Security All rights reserved
+
+Copyright (c) 2014 Thales UK All rights reserved
+
+
+history
+
+09/10/2001 jsh  Original
+
+*/
+
+#include "nfp_common.h"
+#include "nfp_error.h"
+#include "nfp_hostif.h"
+#include "nfp_osif.h"
+#include "i21285.h"
+#include "nfp_cmd.h"
+#include "nfpci.h"
+
+/* create ------------------------------------------------------- */
+
+static nfp_err i21285_create( nfp_cdev *pdev ) {
+  unsigned int tmp32;
+
+  nfp_log( NFP_DBG2, "i21285_create: entered");
+  pdev->cmdctx= pdev;  /* set our context to just be a pointer to our nfp_cdev */
+
+  nfp_log( NFP_DBG2, "i21285_create: enable doorbell");
+  if(!pdev->bar[ IOBAR ]) {
+    nfp_log( NFP_DBG1, "i21285_create: null BAR[%d]", IOBAR );
+    return NFP_ENOMEM;
+  }
+  TO_LE32_IO( &tmp32, DOORBELL_ENABLE | POSTLIST_ENABLE);
+  nfp_outl( pdev, IOBAR, I21285_OFFSET_INTERRUPT_MASK, tmp32 );
+
+  return NFP_SUCCESS;
+}
+
+/* stop ------------------------------------------------------- */
+
+static nfp_err i21285_destroy( void * ctx ) {
+  nfp_cdev *pdev;
+  unsigned int tmp32;
+
+  nfp_log( NFP_DBG2, "i21285_destroy: entered");
+
+  pdev= (nfp_cdev *)ctx;
+  if(!pdev) {
+    nfp_log( NFP_DBG1, "i21285_destroy: NULL pdev");
+    return NFP_ENODEV;
+  }
+  if(!pdev->bar[ IOBAR ]) {
+    nfp_log( NFP_DBG1, "i21285_destroy: null BAR[%d]", IOBAR );
+    return NFP_ENOMEM;
+  }
+  TO_LE32_IO( &tmp32, DOORBELL_DISABLE | POSTLIST_DISABLE );
+  nfp_outl( pdev, IOBAR, I21285_OFFSET_INTERRUPT_MASK, tmp32 );
+
+  return NFP_SUCCESS;
+}
+
+/* open ------------------------------------------------------- */
+
+/* ARGSUSED */
+static nfp_err i21285_open( void * ctx ) {
+  nfp_log( NFP_DBG2, "i21285_open: entered");
+
+  return NFP_SUCCESS;
+}
+
+/* close ------------------------------------------------------- */
+
+/* ARGSUSED */
+static nfp_err i21285_close( void * ctx ) {
+  nfp_log( NFP_DBG2, "i21285_close: entered");
+
+  return NFP_SUCCESS;
+}
+
+/* isr ------------------------------------------------------- */
+
+static nfp_err i21285_isr( void *ctx, int *handled ) {
+  nfp_cdev *pdev;
+  unsigned int doorbell;
+  unsigned int tmp32;
+
+  nfp_log( NFP_DBG3, "i21285_isr: entered");
+
+  *handled= 0;
+  pdev= (nfp_cdev *)ctx;
+  if(!pdev) {
+    nfp_log( NFP_DBG1, "i21285_isr: NULL pdev");
+    return NFP_ENODEV;
+  }
+
+  doorbell= nfp_inl( pdev, IOBAR, I21285_OFFSET_DOORBELL);
+  doorbell= FROM_LE32_IO(&doorbell) & 0xffff;
+  while( doorbell && doorbell != 0xffff) {
+    *handled= 1;
+    /* service interrupts */
+    if( doorbell & (NFAST_INT_DEVICE_WRITE_OK | NFAST_INT_DEVICE_WRITE_FAILED)) {
+      TO_LE32_IO( &tmp32, NFAST_INT_DEVICE_WRITE_OK | NFAST_INT_DEVICE_WRITE_FAILED);
+      nfp_outl( pdev, IOBAR, I21285_OFFSET_DOORBELL, tmp32 );
+
+      nfp_log(NFP_DBG2, "i21285_isr: write done interrupt, ok = %d.", doorbell & NFAST_INT_DEVICE_WRITE_OK ? 1 : 0 );
+
+      nfp_write_complete(pdev->dev, doorbell & NFAST_INT_DEVICE_WRITE_OK ? 1 : 0 );
+    }
+
+    if( doorbell & (NFAST_INT_DEVICE_READ_OK | NFAST_INT_DEVICE_READ_FAILED)) {
+       TO_LE32_IO( &tmp32, NFAST_INT_DEVICE_READ_OK | NFAST_INT_DEVICE_READ_FAILED );
+       nfp_outl( pdev, IOBAR, I21285_OFFSET_DOORBELL, tmp32 );
+
+      nfp_log(NFP_DBG2, "i21285_isr: read ack interrupt, ok = %d.", doorbell & NFAST_INT_DEVICE_READ_OK ? 1 : 0 );
+      nfp_read_complete( pdev->dev, doorbell & NFAST_INT_DEVICE_READ_OK ? 1 : 0);
+    }
+
+    if( doorbell & ~(NFAST_INT_DEVICE_READ_OK  | NFAST_INT_DEVICE_READ_FAILED |
+                     NFAST_INT_DEVICE_WRITE_OK | NFAST_INT_DEVICE_WRITE_FAILED)) {
+      nfp_log( NFP_DBG1, "i21285_isr: unexpected interrupt %x", doorbell );
+      TO_LE32_IO( &tmp32, 0xffff & doorbell );
+      nfp_outl( pdev, IOBAR, I21285_OFFSET_DOORBELL, tmp32 );
+    }
+    doorbell= nfp_inl( pdev, IOBAR, I21285_OFFSET_DOORBELL);
+    doorbell= FROM_LE32_IO(&doorbell) & 0xffff;
+  }
+  return 0;
+}
+
+/* write ------------------------------------------------------- */
+
+static nfp_err i21285_write( const char *block, int len, void *ctx ) {
+  nfp_cdev *cdev;
+  unsigned int hdr[2];
+  nfp_err ne;
+  unsigned int tmp32;
+
+  nfp_log( NFP_DBG2, "i21285_write: entered");
+
+  cdev= (nfp_cdev *)ctx;
+  if(!cdev) {
+    nfp_log( NFP_DBG1, "i21285_write: NULL pdev");
+    return NFP_ENODEV;
+  }
+
+  nfp_log(NFP_DBG2, "i21285_write: pdev->bar[ MEMBAR ]= %x\n", cdev->bar[ MEMBAR ]);
+  nfp_log(NFP_DBG2, "i21285_write: pdev->bar[ IOBAR ]= %x\n", cdev->bar[ IOBAR ]);
+  if(!cdev->bar[ MEMBAR ]) {
+    nfp_log( NFP_DBG1, "i21285_write: null BAR[%d]", MEMBAR );
+    return NFP_ENOMEM;
+  }
+  ne= nfp_copy_from_user_to_dev( cdev, MEMBAR, NFPCI_JOBS_WR_DATA, block, len);
+  if (ne) {
+    nfp_log( NFP_DBG1, "i21285_write: nfp_copy_from_user_to_dev failed");
+    return ne;
+  }
+  TO_LE32_MEM(&hdr[0], NFPCI_JOB_CONTROL);
+  TO_LE32_MEM(&hdr[1], len);
+
+  ne= nfp_copy_to_dev( cdev, MEMBAR, NFPCI_JOBS_WR_CONTROL, (const char *)hdr, 8);
+  if (ne) {
+    nfp_log( NFP_DBG1, "i21285_write: nfp_copy_to_dev failed");
+    return ne;
+  }
+
+  ne= nfp_copy_from_dev( cdev, MEMBAR, NFPCI_JOBS_WR_LENGTH, (char *)hdr, 4);
+  if (ne) {
+    nfp_log( NFP_DBG1, "i21285_write: nfp_copy_from_dev failed");
+    return ne;
+  }
+  
+  TO_LE32_MEM( &tmp32, len );
+  if ( hdr[0] != tmp32 ) {
+    nfp_log( NFP_DBG1, "i21285_write: length not written");
+    return NFP_EIO;
+  }
+
+  TO_LE32_IO( &tmp32, NFAST_INT_HOST_WRITE_REQUEST);
+
+  nfp_outl( cdev, IOBAR, I21285_OFFSET_DOORBELL, tmp32 );
+
+  nfp_log( NFP_DBG2, "i21285_write: done");
+  return NFP_SUCCESS;
+}
+
+/* read ------------------------------------------------------- */
+
+static nfp_err i21285_read( char *block, int len, void *ctx, int *rcount) {
+  nfp_cdev *cdev;
+  nfp_err ne;
+  int count;
+
+  nfp_log( NFP_DBG2, "i21285_read: entered, len %d", len);
+  *rcount= 0;
+
+  cdev= (nfp_cdev *)ctx;
+  if(!cdev) {
+    nfp_log( NFP_DBG1, "i21285_read: NULL pdev");
+    return NFP_ENODEV;
+  }
+
+  if(!cdev->bar[ MEMBAR ]) {
+    nfp_log( NFP_DBG1, "i21285_read: null BAR[%d]", MEMBAR );
+    return NFP_ENOMEM;
+  }
+  ne= nfp_copy_from_dev( cdev, MEMBAR, NFPCI_JOBS_RD_LENGTH, (char *)&count, 4);
+  if(ne) {
+    nfp_log( NFP_DBG1, "i21285_read: nfp_copy_from_dev failed.");
+    return ne;
+  }
+  count= FROM_LE32_MEM(&count);
+  if(count<0 || count>len) {
+    nfp_log( NFP_DBG1, "i21285_read: bad byte count (%d) from device", count);
+    return NFP_EIO;
+  }
+  ne= nfp_copy_to_user_from_dev( cdev, MEMBAR, NFPCI_JOBS_RD_DATA, block, count);
+  if( ne ) {
+    nfp_log( NFP_DBG1, "i21285_read: nfp_copy_to_user_from_dev failed.");
+    return ne;
+  }
+  nfp_log( NFP_DBG2, "i21285_read: done");
+  *rcount= count;
+  return NFP_SUCCESS;
+}
+
+/* chupdate  ------------------------------------------------------- */
+
+/* ARGSUSED */
+static nfp_err i21285_chupdate( char *data, int len, void *ctx ) {
+  nfp_log( NFP_DBG1, "i21285_chupdate: NYI");
+  return NFP_SUCCESS;
+}
+
+/* ensure reading -------------------------------------------------- */
+
+static nfp_err i21285_ensure_reading( unsigned int addr, int len, void *ctx ) {
+  nfp_cdev *cdev;
+  unsigned int hdr[2];
+  unsigned int tmp32;
+  nfp_err ne;
+
+  nfp_log( NFP_DBG2, "i21285_ensure_reading: entered");
+
+  if(addr) {
+    nfp_log( NFP_DBG2, "i21285_ensure_reading: bad addr");
+    return -NFP_EINVAL;
+  }
+
+  cdev= (nfp_cdev *)ctx;
+  if(!cdev) {
+    nfp_log( NFP_DBG1, "i21285_ensure_reading: NULL pdev");
+    return NFP_ENODEV;
+  }
+
+  if(!cdev->bar[ MEMBAR ]) {
+    nfp_log( NFP_DBG1, "i21285_ensure_reading: null BAR[%d]", MEMBAR );
+    return NFP_ENXIO;
+  }
+  nfp_log( NFP_DBG3, "i21285_ensure_reading: pdev->bar[ MEMBAR ]= %x", cdev->bar[ MEMBAR ]);
+  nfp_log( NFP_DBG3, "i21285_ensure_reading: pdev->bar[ IOBAR ]= %x", cdev->bar[ IOBAR ]);
+  TO_LE32_MEM( &hdr[0], NFPCI_JOB_CONTROL);
+  TO_LE32_MEM( &hdr[1], len);
+  ne= nfp_copy_to_dev( cdev, MEMBAR, NFPCI_JOBS_RD_CONTROL, (const char *)hdr, 8);
+  if (ne) {
+    nfp_log( NFP_DBG1, "i21285_ensure_reading: nfp_copy_to_dev failed");
+    return ne;
+  }
+  ne= nfp_copy_from_dev( cdev, MEMBAR, NFPCI_JOBS_RD_LENGTH, (char *)hdr, 4);
+  if (ne) {
+    nfp_log( NFP_DBG1, "i21285_ensure_reading: nfp_copy_from_dev failed");
+    return ne;
+  }
+  TO_LE32_MEM( &tmp32, len );
+  if ( hdr[0] != tmp32 ) {
+    nfp_log( NFP_DBG1, "i21285_ensure_reading: len not written");
+    return NFP_EIO;
+  };
+  TO_LE32_IO( &tmp32, NFAST_INT_HOST_READ_REQUEST );
+  nfp_outl( cdev, IOBAR, I21285_OFFSET_DOORBELL, tmp32 );
+
+  return NFP_SUCCESS;
+}
+
+/* command device structure ------------------------------------- */
+
+
+const nfpcmd_dev i21285_cmddev = {
+  "nCipher Gen 1 PCI",
+  PCI_VENDOR_ID_DEC, PCI_DEVICE_ID_DEC_21285,
+  PCI_VENDOR_ID_NCIPHER, PCI_DEVICE_ID_NFAST_GEN1,
+  { 0, IOSIZE | PCI_BASE_ADDRESS_SPACE_IO, NFPCI_RAM_MINSIZE, 0, 0, 0 },
+  NFP_CMD_FLG_NEED_IOBUF,
+  i21285_create,
+  i21285_destroy,
+  i21285_open,
+  i21285_close,
+  i21285_isr,
+  i21285_write,
+  i21285_read,
+  i21285_chupdate,
+  i21285_ensure_reading,
+  0, /* no debug */
+};
+  
diff --git a/usr/src/uts/common/io/nfp/i21285.h b/usr/src/uts/common/io/nfp/i21285.h
new file mode 100644
index 0000000000..4ea1d853ec
--- /dev/null
+++ b/usr/src/uts/common/io/nfp/i21285.h
@@ -0,0 +1,43 @@
+/*
+
+(C) Copyright nCipher Corporation Ltd 2002-2008 All rights reserved
+
+Copyright (c) 2008-2013 Thales e-Security All rights reserved
+
+Copyright (c) 2014 Thales UK All rights reserved
+
+*/
+
+#ifndef NFP_I21285_H
+#define NFP_I21285_H
+
+#ifndef PCI_VENDOR_ID_DEC
+#define PCI_VENDOR_ID_DEC               0x1011
+#endif
+#ifndef PCI_DEVICE_ID_DEC_21285
+#define PCI_DEVICE_ID_DEC_21285         0x1065
+#endif
+#ifndef PCI_VENDOR_ID_NCIPHER
+#define PCI_VENDOR_ID_NCIPHER           0x0100
+#endif
+
+#ifndef PCI_DEVICE_ID_NFAST_GEN1
+#define PCI_DEVICE_ID_NFAST_GEN1	0x0100
+#endif
+
+#define I21285_OFFSET_DOORBELL		0x60
+#define I21285_OFFSET_INTERRUPT_MASK	0x34
+
+#define DOORBELL_ENABLE 0x0
+#define DOORBELL_DISABLE 0x4
+
+#define POSTLIST_ENABLE 0x0
+#define POSTLIST_DISABLE 0x8
+
+#define IOBAR	1
+#define MEMBAR	2
+
+#define IOSIZE	0x80
+#define MEMSIZE	0x100000
+
+#endif
diff --git a/usr/src/uts/common/io/nfp/i21555.c b/usr/src/uts/common/io/nfp/i21555.c
new file mode 100644
index 0000000000..82024dc800
--- /dev/null
+++ b/usr/src/uts/common/io/nfp/i21555.c
@@ -0,0 +1,423 @@
+/*
+
+i21555.c: nCipher PCI HSM intel 21555 command driver
+
+(C) Copyright nCipher Corporation Ltd 2002-2008 All rights reserved
+
+Copyright (c) 2008-2013 Thales e-Security All rights reserved
+
+Copyright (c) 2014 Thales UK All rights reserved
+
+history
+
+09/10/2001 jsh  Original
+
+*/
+
+#include "nfp_common.h"
+#include "nfp_error.h"
+#include "nfp_hostif.h"
+#include "nfp_osif.h"
+#include "i21555.h"
+#include "nfp_cmd.h"
+#include "nfpci.h"
+
+/* started ------------------------------------------------------
+ *
+ * Check that device is ready to talk, by checking that
+ * the i21555 has master enabled on its secondary interface
+ */
+
+static nfp_err i21555_started( nfp_cdev *pdev ) {
+  unsigned int tmp32;
+#ifdef CONFIGSPACE_DEBUG
+  unsigned int reg32[64];
+  int i;
+#endif
+  nfp_err ne;
+
+  nfp_log( NFP_DBG2, "i21555_started: entered");
+
+#ifdef CONFIGSPACE_DEBUG
+  /* Suck up all the registers */
+  for (i=0; i < 64; i++) {
+    ne = nfp_config_inl( pdev, i*4, &reg32[i] );
+  }
+
+  for (i=0; i < 16; i++) {
+    int j = i * 4;
+    nfp_log( NFP_DBG3, "i21555 config reg %2x: %08x %08x %08x %08x", j*4,
+        reg32[j], reg32[j+1], reg32[j+2], reg32[j+3]);
+  }
+#endif
+
+  ne = nfp_config_inl( pdev, I21555_CFG_SEC_CMD_STATUS, &tmp32 );
+  if (ne) {
+    /* succeed if PCI config reads are not implemented */
+    if (ne == NFP_EUNKNOWN)
+      return NFP_SUCCESS;
+    nfp_log( NFP_DBG1, "i21555_started: nfp_config_inl failed");
+    return ne;
+  }
+
+  tmp32= FROM_LE32_IO(&tmp32) & 0xffff;
+
+  if ( tmp32 & CFG_CMD_MASTER ) {
+    nfp_log( NFP_DBG3, "i21555_started: Yes %x", tmp32);
+    return NFP_SUCCESS;
+  } else {
+    nfp_log( NFP_DBG1, "i21555_started: device not started yet %x", tmp32);
+    return NFP_ESTARTING;
+  }
+}
+
+/* create ------------------------------------------------------- */
+
+static nfp_err i21555_create( nfp_cdev *pdev ) {
+  unsigned int tmp32;
+
+  nfp_log( NFP_DBG2, "i21555_create: entered");
+  pdev->cmdctx= pdev;  /* set our context to just be a pointer to our nfp_cdev */
+
+  if(!pdev->bar[ IOBAR ]) {
+    nfp_log( NFP_DBG1, "i21555_create: null BAR[%d]", IOBAR );
+    return NFP_ENOMEM;
+  }
+  nfp_log( NFP_DBG2, "i21555_create: enable doorbell");
+  TO_LE32_IO( &tmp32, I21555_DOORBELL_PRI_ENABLE );
+  nfp_outl( pdev, IOBAR, I21555_OFFSET_DOORBELL_PRI_SET_MASK, tmp32 );
+  nfp_outl( pdev, IOBAR, I21555_OFFSET_DOORBELL_PRI_CLEAR_MASK, tmp32 );
+  return NFP_SUCCESS;
+}
+
+/* stop ------------------------------------------------------- */
+
+static nfp_err i21555_destroy( void * ctx ) {
+  nfp_cdev *pdev;
+  unsigned int tmp32;
+
+  nfp_log( NFP_DBG2, "i21555_destroy: entered");
+
+  pdev= (nfp_cdev *)ctx;
+  if(!pdev) {
+    nfp_log( NFP_DBG1, "i21555_destroy: NULL pdev");
+    return NFP_ENODEV;
+  }
+  if(!pdev->bar[ IOBAR ]) {
+    nfp_log( NFP_DBG1, "i21555_destroy: null BAR[%d]", IOBAR );
+    return NFP_ENOMEM;
+  }
+  TO_LE32_IO( &tmp32, I21555_DOORBELL_PRI_DISABLE );
+  nfp_outl( pdev, IOBAR, I21555_OFFSET_DOORBELL_PRI_SET_MASK, tmp32 );
+  nfp_outl( pdev, IOBAR, I21555_OFFSET_DOORBELL_PRI_CLEAR_MASK, tmp32 );
+
+  return NFP_SUCCESS;
+}
+
+/* open ------------------------------------------------------- */
+
+/* ARGSUSED */
+static nfp_err i21555_open( void * ctx ) {
+
+  nfp_log( NFP_DBG2, "i21555_open: entered");
+
+  return NFP_SUCCESS;
+}
+
+/* close ------------------------------------------------------- */
+
+/* ARGSUSED */
+static nfp_err i21555_close( void * ctx ) {
+  nfp_log( NFP_DBG2, "i21555_close: entered");
+
+  return NFP_SUCCESS;
+}
+
+/* isr ------------------------------------------------------- */
+
+static nfp_err i21555_isr( void *ctx, int *handled ) {
+  nfp_cdev *pdev;
+  nfp_err ne;
+  unsigned short doorbell;
+  unsigned short tmp16;
+
+  nfp_log( NFP_DBG3, "i21555_isr: entered");
+
+  *handled= 0;
+  pdev= (nfp_cdev *)ctx;
+  if(!pdev) {
+    nfp_log( NFP_DBG1, "i21555_isr: NULL pdev");
+    return NFP_ENODEV;
+  }
+
+  pdev->stats.isr++;
+
+  if(!pdev->bar[ IOBAR ]) {
+    nfp_log( NFP_DBG1, "i21555_isr: null BAR[%d]", IOBAR );
+    return NFP_ENOMEM;
+  }
+
+  /* This interrupt may not be from our module, so check that it actually is
+   * us before handling it.
+   */
+  ne = i21555_started( pdev );
+  if (ne) {
+    if (ne != NFP_ESTARTING) {
+      nfp_log( NFP_DBG1, "i21555_isr: i21555_started failed");
+    }
+    return ne;
+  }
+
+  doorbell= nfp_inw( pdev, IOBAR, I21555_OFFSET_DOORBELL_PRI_SET);
+  doorbell= FROM_LE16_IO(&doorbell);
+  while( doorbell && doorbell != 0xffff) {
+    *handled= 1;
+    /* service interrupts */
+    if( doorbell & (NFAST_INT_DEVICE_WRITE_OK | NFAST_INT_DEVICE_WRITE_FAILED)) {
+      pdev->stats.isr_write++;
+      TO_LE16_IO(&tmp16,NFAST_INT_DEVICE_WRITE_OK | NFAST_INT_DEVICE_WRITE_FAILED);
+      nfp_outw( pdev, IOBAR, I21555_OFFSET_DOORBELL_PRI_CLEAR, tmp16 );
+
+      nfp_log( NFP_DBG2, "i21555_isr: write done interrupt, ok = %d.", doorbell & NFAST_INT_DEVICE_WRITE_OK ? 1 : 0 );
+
+      nfp_write_complete(pdev->dev, doorbell & NFAST_INT_DEVICE_WRITE_OK ? 1 : 0 );
+    }
+
+    if( doorbell & (NFAST_INT_DEVICE_READ_OK | NFAST_INT_DEVICE_READ_FAILED)) {
+      pdev->stats.isr_read++;
+      TO_LE16_IO(&tmp16,NFAST_INT_DEVICE_READ_OK | NFAST_INT_DEVICE_READ_FAILED);
+      nfp_outw( pdev, IOBAR, I21555_OFFSET_DOORBELL_PRI_CLEAR, tmp16 );
+
+      nfp_log( NFP_DBG2, "i21555_isr: read ack interrupt, ok = %d.", doorbell & NFAST_INT_DEVICE_READ_OK ? 1 : 0 );
+      nfp_read_complete( pdev->dev, doorbell & NFAST_INT_DEVICE_READ_OK ? 1 : 0);
+    }
+
+    if( doorbell & ~(NFAST_INT_DEVICE_READ_OK  | NFAST_INT_DEVICE_READ_FAILED |
+                     NFAST_INT_DEVICE_WRITE_OK | NFAST_INT_DEVICE_WRITE_FAILED)) {
+      TO_LE16_IO(&tmp16,doorbell);
+      nfp_outw( pdev, IOBAR, I21555_OFFSET_DOORBELL_PRI_CLEAR, tmp16 );
+      nfp_log( NFP_DBG1, "i21555_isr: unexpected interrupt %x", doorbell );
+    }
+    doorbell= nfp_inw( pdev, IOBAR, I21555_OFFSET_DOORBELL_PRI_SET);
+    doorbell= FROM_LE16_IO(&doorbell);
+  }
+  nfp_log( NFP_DBG3, "i21555_isr: exiting");
+  return 0;
+}
+
+/* write ------------------------------------------------------- */
+
+static nfp_err i21555_write( const char *block, int len, void *ctx) {
+  nfp_cdev *cdev;
+  unsigned int hdr[2];
+  nfp_err ne;
+  unsigned short tmp16;
+  unsigned int tmp32;
+
+  nfp_log( NFP_DBG2, "i21555_write: entered");
+
+  cdev= (nfp_cdev *)ctx;
+  if(!cdev) {
+    nfp_log( NFP_DBG1, "i21555_write: NULL cdev");
+    return NFP_ENODEV;
+  }
+
+  cdev->stats.write_fail++;
+
+  if(!cdev->bar[ IOBAR ]) {
+    nfp_log( NFP_DBG1, "i21555_write: null BAR[%d]", IOBAR );
+    return NFP_ENOMEM;
+  }
+
+  ne = i21555_started( cdev );
+  if (ne) {
+    if (ne != NFP_ESTARTING) {
+      nfp_log( NFP_DBG1, "i21555_write: i21555_started failed");
+    }
+    return ne;
+  }
+
+  nfp_log( NFP_DBG3, "i21555_write: cdev->bar[ MEMBAR ]= %x", cdev->bar[ MEMBAR ]);
+  nfp_log( NFP_DBG3, "i21555_write: cdev->bar[ IOBAR ]= %x", cdev->bar[ IOBAR ]);
+  nfp_log( NFP_DBG3, "i21555_write: block len %d", len ); 
+  ne= nfp_copy_from_user_to_dev( cdev, MEMBAR, NFPCI_JOBS_WR_DATA, block, len);
+  if (ne) {
+    nfp_log( NFP_DBG1, "i21555_write: nfp_copy_from_user_to_dev failed");
+    return ne;
+  }
+  TO_LE32_MEM(&hdr[0], NFPCI_JOB_CONTROL);
+  TO_LE32_MEM(&hdr[1], len);
+  ne= nfp_copy_to_dev( cdev, MEMBAR, NFPCI_JOBS_WR_CONTROL, (const char *)hdr, 8);
+  if (ne) {
+    nfp_log( NFP_DBG1, "i21555_write: nfp_copy_to_dev failed");
+    return ne;
+  }
+
+  ne= nfp_copy_from_dev( cdev, MEMBAR, NFPCI_JOBS_WR_LENGTH, (char *)hdr, 4);
+  if (ne) {
+    nfp_log( NFP_DBG1, "i21555_write: nfp_copy_from_dev failed");
+    return ne;
+  }
+
+  TO_LE32_MEM(&tmp32, len);
+  if ( hdr[0] != tmp32 ) {
+    nfp_log( NFP_DBG1, "i21555_write: length not written");
+    return NFP_EIO;
+  }
+  TO_LE16_IO(&tmp16, NFAST_INT_HOST_WRITE_REQUEST >> 16);
+  nfp_outw( cdev, IOBAR, I21555_OFFSET_DOORBELL_SEC_SET, tmp16);
+
+  cdev->stats.write_fail--;
+  cdev->stats.write_block++;
+  cdev->stats.write_byte += len;
+
+  nfp_log( NFP_DBG2, "i21555_write: done");
+  return NFP_SUCCESS;
+}
+
+/* read ------------------------------------------------------- */
+
+static nfp_err i21555_read( char *block, int len, void *ctx, int *rcount) {
+  nfp_cdev *cdev;
+  nfp_err ne;
+  int count;
+
+  nfp_log( NFP_DBG2, "i21555_read: entered");
+  *rcount= 0;
+
+  cdev= (nfp_cdev *)ctx;
+  if(!cdev) {
+    nfp_log( NFP_DBG1, "i21555_read: NULL pdev");
+    return NFP_ENODEV;
+  }
+
+  cdev->stats.read_fail++;
+
+  if(!cdev->bar[ IOBAR ]) {
+    nfp_log( NFP_DBG1, "i21555_read: null BAR[%d]", IOBAR );
+    return NFP_ENOMEM;
+  }
+
+  ne= nfp_copy_from_dev( cdev, MEMBAR, NFPCI_JOBS_RD_LENGTH, (char *)&count, 4);
+  if (ne) {
+    nfp_log( NFP_DBG1, "i21555_read: nfp_copy_from_dev failed.");
+    return ne;
+  }
+  count= FROM_LE32_MEM(&count);
+  if(count<0 || count>len) {
+    nfp_log( NFP_DBG1, "i21555_read: bad byte count (%d) from device", count);
+    return NFP_EIO;
+  }
+  ne= nfp_copy_to_user_from_dev( cdev, MEMBAR, NFPCI_JOBS_RD_DATA, block, count);
+  if (ne) {
+    nfp_log( NFP_DBG1, "i21555_read: nfp_copy_to_user failed.");
+    return ne;
+  }
+  nfp_log( NFP_DBG2, "i21555_read: done");
+  *rcount= count;
+  cdev->stats.read_fail--;
+  cdev->stats.read_block++;
+  cdev->stats.read_byte += len;
+  return NFP_SUCCESS;
+}
+
+/* chupdate  ------------------------------------------------------- */
+
+/* ARGSUSED */
+static nfp_err i21555_chupdate( char *data, int len, void *ctx ) {
+  nfp_log( NFP_DBG1, "i21555_chupdate: NYI");
+  return NFP_SUCCESS;
+}
+
+/* ensure reading -------------------------------------------------- */
+
+static nfp_err i21555_ensure_reading( unsigned int addr, int len, void *ctx ) {
+  nfp_cdev *cdev;
+  unsigned int hdr[3];
+  unsigned short tmp16;
+  unsigned int tmp32;
+  nfp_err ne;
+  int hdr_len;
+
+  nfp_log( NFP_DBG2, "i21555_ensure_reading: entered");
+
+  cdev= (nfp_cdev *)ctx;
+  if(!cdev) {
+    nfp_log( NFP_DBG1, "i21555_ensure_reading: NULL pdev");
+    return NFP_ENODEV;
+  }
+
+  cdev->stats.ensure_fail++;
+
+  if(!cdev->bar[ IOBAR ]) {
+    nfp_log( NFP_DBG1, "i21555_ensure_reading: null BAR[%d]", IOBAR );
+    return NFP_ENOMEM;
+  }
+
+  ne = i21555_started( cdev );
+  if (ne) {
+    if (ne != NFP_ESTARTING) {
+      nfp_log( NFP_DBG1, "i21555_ensure_reading: i21555_started failed");
+    }
+    return ne;
+  }
+
+  nfp_log( NFP_DBG3, "i21555_ensure_reading: pdev->bar[ MEMBAR ]= %x", cdev->bar[ MEMBAR ]);
+  nfp_log( NFP_DBG3, "i21555_ensure_reading: pdev->bar[ IOBAR ]= %x", cdev->bar[ IOBAR ]);
+  if(addr) {
+    nfp_log( NFP_DBG3, "i21555_ensure_reading: new format, addr %x", addr);
+    TO_LE32_MEM(&hdr[0], NFPCI_JOB_CONTROL_PCI_PUSH);
+    TO_LE32_MEM(&hdr[1], len);
+    TO_LE32_MEM(&hdr[2], addr);
+    hdr_len= 12;
+  } else {
+    TO_LE32_MEM(&hdr[0], NFPCI_JOB_CONTROL);
+    TO_LE32_MEM(&hdr[1], len);
+    hdr_len= 8;
+  }
+  ne= nfp_copy_to_dev( cdev, MEMBAR, NFPCI_JOBS_RD_CONTROL, (const char *)hdr, hdr_len);
+  if (ne) {
+    nfp_log( NFP_DBG1, "i21555_ensure_reading: nfp_copy_to_dev failed");
+    return ne;
+  }
+
+  ne= nfp_copy_from_dev( cdev, MEMBAR, NFPCI_JOBS_RD_LENGTH, (char *)hdr, 4);
+  if (ne) {
+    nfp_log( NFP_DBG1, "i21555_ensure_reading: nfp_copy_from_dev failed");
+    return ne;
+  }
+
+  TO_LE32_MEM(&tmp32, len);
+
+  if ( hdr[0] != tmp32 ) {
+    nfp_log( NFP_DBG1, "i21555_ensure_reading: len not written");
+    return NFP_EIO;
+  }
+  TO_LE16_IO( &tmp16, NFAST_INT_HOST_READ_REQUEST >> 16);
+  nfp_outw( cdev, IOBAR, I21555_OFFSET_DOORBELL_SEC_SET, tmp16);
+
+  cdev->stats.ensure_fail--;
+  cdev->stats.ensure++;
+
+  return NFP_SUCCESS;
+}
+
+/* command device structure ------------------------------------- */
+
+const nfpcmd_dev i21555_cmddev = {
+  "nCipher Gen 2 PCI",
+  PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_INTEL_21555,
+  PCI_VENDOR_ID_NCIPHER, PCI_SUBSYSTEM_ID_NFAST_REV1,
+  { 0, IOSIZE | PCI_BASE_ADDRESS_SPACE_IO, NFPCI_RAM_MINSIZE_JOBS, 0, 0, 0 },
+  NFP_CMD_FLG_NEED_IOBUF,
+  i21555_create,
+  i21555_destroy,
+  i21555_open,
+  i21555_close,
+  i21555_isr,
+  i21555_write,
+  i21555_read,
+  i21555_chupdate,
+  i21555_ensure_reading,
+  i21555_debug,
+};
diff --git a/usr/src/uts/common/io/nfp/i21555.h b/usr/src/uts/common/io/nfp/i21555.h
new file mode 100644
index 0000000000..d8f3965938
--- /dev/null
+++ b/usr/src/uts/common/io/nfp/i21555.h
@@ -0,0 +1,51 @@
+/*
+
+(C) Copyright nCipher Corporation Ltd 2002-2008 All rights reserved
+
+Copyright (c) 2008-2013 Thales e-Security All rights reserved
+
+Copyright (c) 2014 Thales UK All rights reserved
+
+*/
+
+#ifndef I21555_H
+#define I21555_H
+
+#ifndef PCI_VENDOR_ID_INTEL
+#define PCI_VENDOR_ID_INTEL             0x8086
+#endif
+
+#ifndef PCI_DEVICE_ID_INTEL_21555
+#define PCI_DEVICE_ID_INTEL_21555       0xb555
+#endif
+
+#ifndef PCI_VENDOR_ID_NCIPHER
+#define PCI_VENDOR_ID_NCIPHER           0x0100
+#endif
+
+#ifndef PCI_SUBSYSTEM_ID_NFAST_REV1
+#define PCI_SUBSYSTEM_ID_NFAST_REV1     0x0100
+#endif
+
+#define I21555_OFFSET_DOORBELL_PRI_SET		0x9C
+#define I21555_OFFSET_DOORBELL_SEC_SET		0x9E
+#define I21555_OFFSET_DOORBELL_PRI_CLEAR	0x98
+
+#define I21555_OFFSET_DOORBELL_PRI_SET_MASK	0xA4
+#define I21555_OFFSET_DOORBELL_PRI_CLEAR_MASK	0xA0
+
+#define I21555_DOORBELL_PRI_ENABLE 0x0000
+#define I21555_DOORBELL_PRI_DISABLE 0xFFFF
+
+#define I21555_CFG_SEC_CMD_STATUS 0x44
+
+#define CFG_CMD_MASTER 0x0004
+
+#define IOBAR   1
+#define MEMBAR  2
+
+#define IOSIZE  0x100
+
+extern nfp_err i21555_debug( int cmd, void *ctx );
+
+#endif
diff --git a/usr/src/uts/common/io/nfp/i21555d.c b/usr/src/uts/common/io/nfp/i21555d.c
new file mode 100644
index 0000000000..183ace8275
--- /dev/null
+++ b/usr/src/uts/common/io/nfp/i21555d.c
@@ -0,0 +1,28 @@
+/*
+
+i21555d.c: nCipher PCI HSM intel 21555 debug ioctl
+
+(C) Copyright nCipher Corporation Ltd 2002-2008 All rights reserved
+
+Copyright (c) 2008-2013 Thales e-Security All rights reserved
+
+Copyright (c) 2014 Thales UK All rights reserved
+
+
+history
+
+15/05/2002 jsh  Original, does nothing
+
+*/
+
+#include "nfp_common.h"
+#include "nfp_error.h"
+#include "nfp_osif.h"
+#include "i21555.h"
+
+/* ARGSUSED */
+nfp_err i21555_debug( int cmd, void *ctx) {
+  nfp_log( NFP_DBG1, "i21555_debug: entered");
+
+  return NFP_EUNKNOWN;
+}
diff --git a/usr/src/uts/common/io/nfp/nfdev-common.h b/usr/src/uts/common/io/nfp/nfdev-common.h
new file mode 100644
index 0000000000..8a97bf2c63
--- /dev/null
+++ b/usr/src/uts/common/io/nfp/nfdev-common.h
@@ -0,0 +1,141 @@
+/*
+
+(C) Copyright nCipher Corporation Ltd 2002-2008 All rights reserved
+
+Copyright (c) 2008-2013 Thales e-Security All rights reserved
+
+Copyright (c) 2014 Thales UK All rights reserved
+
+*/
+/** \file nfdev-common.h
+ *
+ * \brief nFast device driver (not generic SCSI) ioctl struct definition file
+ *  include NFDEV-$(system) for ioctl number definitions
+ *
+ *  1998.07.13	jsh	Started
+ *
+ * 
+ */
+
+#ifndef NFDEV_COMMON_H
+#define NFDEV_COMMON_H
+
+/**
+ * Result of the ENQUIRY ioctl.
+ */
+typedef struct nfdev_enquiry_str {
+  unsigned int  busno; /**< Which bus is the PCI device on. */
+  unsigned char slotno; /**< Which slot is the PCI device in. */
+  unsigned char reserved[3]; /**< for consistant struct alignment */
+} nfdev_enquiry_str;
+
+/**
+ * Result of the STATS ioctl.
+ */
+typedef struct nfdev_stats_str {
+  unsigned long  isr; /**< Count interrupts. */
+  unsigned long  isr_read; /**< Count read interrupts. */
+  unsigned long  isr_write; /**< Count write interrupts. */
+  unsigned long  write_fail; /**< Count write failures. */
+  unsigned long  write_block; /**< Count blocks written. */
+  unsigned long  write_byte; /**< Count bytes written. */
+  unsigned long  read_fail; /**< Count read failures. */
+  unsigned long  read_block; /**< Count blocks read. */
+  unsigned long  read_byte; /**< Count bytes read. */
+  unsigned long  ensure_fail; /**< Count read request failures. */
+  unsigned long  ensure; /**< Count read requests. */
+} nfdev_stats_str;
+
+/**
+ * Input to the CONTROL ioctl.
+ */
+typedef struct nfdev_control_str {
+  unsigned control; /**< Control flags. */
+} nfdev_control_str;
+
+/** Control bit indicating host supports MOI control */
+#define NFDEV_CONTROL_HOST_MOI 0x0001
+
+/** Index of control bits indicating desired mode
+ *
+ * Desired mode follows the M_ModuleMode enumeration.
+ */
+#define NFDEV_CONTROL_MODE_SHIFT 1
+
+/** Detect a backwards-compatible control value
+ *
+ * Returns true if the request control value "makes no difference", i.e.
+ * and the failure of an attempt to set it is therefore uninteresting.
+ */
+#define NFDEV_CONTROL_HARMLESS(c) ((c) <= 1)
+
+/**
+ * Result of the STATUS ioctl.
+ */
+typedef struct nfdev_status_str {
+  unsigned  status; /**< Status flags. */
+  char      error[8]; /**< Error string. */
+} nfdev_status_str;
+
+/** Monitor firmware supports MOI control and error reporting */
+#define NFDEV_STATUS_MONITOR_MOI 0x0001
+
+/** Application firmware supports MOI control and error reporting */
+#define NFDEV_STATUS_APPLICATION_MOI 0x0002
+
+/** Application firmware running and supports error reporting */
+#define NFDEV_STATUS_APPLICATION_RUNNING 0x0004
+
+/** HSM failed
+ *
+ * Consult error[] for additional information.
+ */
+#define NFDEV_STATUS_FAILED 0x0008
+
+/** Standard PCI interface. */
+#define NFDEV_IF_STANDARD	0x01
+
+/** PCI interface with results pushed from device
+ *  via DMA.
+ */
+#define NFDEV_IF_PCI_PUSH	0x02
+
+/* platform independant base ioctl numbers */
+
+/** Enquiry ioctl.
+ *  \return nfdev_enquiry_str describing the attached device. */
+#define NFDEV_IOCTL_NUM_ENQUIRY        0x01
+/** Channel Update ioctl.
+ *  \deprecated */
+#define NFDEV_IOCTL_NUM_CHUPDATE       0x02
+/** Ensure Reading ioctl.
+ *  Signal a read request to the device.
+ *  \param (unsigned int) Length of data to be read.
+ */
+#define NFDEV_IOCTL_NUM_ENSUREREADING  0x03
+/** Device Count ioctl.
+ *  Not implemented for on all platforms.
+ *  \return (int) the number of attached devices. */
+#define NFDEV_IOCTL_NUM_DEVCOUNT       0x04
+/** Internal Debug ioctl.
+ *  Not implemented in release drivers. */
+#define NFDEV_IOCTL_NUM_DEBUG          0x05
+/** PCI Interface Version ioctl.
+ *  \param (int) Maximum PCI interface version
+ *   supported by the user of the device. */
+#define NFDEV_IOCTL_NUM_PCI_IFVERS     0x06
+/** Statistics ioctl.
+ *  \return nfdev_enquiry_str describing the attached device. */
+#define NFDEV_IOCTL_NUM_STATS          0x07
+
+/** Module control ioctl
+ * \param (nfdev_control_str) Value to write to HSM control register
+ */
+#define NFDEV_IOCTL_NUM_CONTROL        0x08
+
+/** Module state ioctl
+ * \return (nfdev_status_str) Values read from HSM status/error registers
+ */
+#define NFDEV_IOCTL_NUM_STATUS         0x09
+
+#endif
diff --git a/usr/src/uts/common/io/nfp/nfdev-solaris.h b/usr/src/uts/common/io/nfp/nfdev-solaris.h
new file mode 100644
index 0000000000..923b902e46
--- /dev/null
+++ b/usr/src/uts/common/io/nfp/nfdev-solaris.h
@@ -0,0 +1,37 @@
+/*
+
+nfdev-solaris.h: nFast solaris specific device ioctl interface.
+
+(C) Copyright nCipher Corporation Ltd 1998-2008 All rights reserved
+
+Copyright (c) 2008-2013 Thales e-Security All rights reserved
+
+Copyright (c) 2014 Thales UK All rights reserved
+
+history
+
+14/07/1998 jsh  Original
+
+*/
+
+#ifndef NFDEV_SOLARIS_H
+#define NFDEV_SOLARIS_H
+
+#include "nfdev-common.h"
+
+#define NFDEV_IOCTL_TYPE ('n'<<8)
+
+#define NFDEV_IOCTL_ENQUIRY		( NFDEV_IOCTL_TYPE | \
+					  NFDEV_IOCTL_NUM_ENQUIRY )
+#define NFDEV_IOCTL_ENSUREREADING	( NFDEV_IOCTL_TYPE | \
+					  NFDEV_IOCTL_NUM_ENSUREREADING )
+#define NFDEV_IOCTL_DEVCOUNT		( NFDEV_IOCTL_TYPE | \
+					  NFDEV_IOCTL_NUM_DEVCOUNT )
+#define NFDEV_IOCTL_DEBUG		( NFDEV_IOCTL_TYPE | \
+					  NFDEV_IOCTL_NUM_DEBUG )
+#define NFDEV_IOCTL_PCI_IFVERS		( NFDEV_IOCTL_TYPE | \
+					  NFDEV_IOCTL_NUM_PCI_IFVERS )
+#define NFDEV_IOCTL_STATS		( NFDEV_IOCTL_TYPE | \
+					  NFDEV_IOCTL_NUM_STATS )
+
+#endif /* NFDEV_SOLARIS_H */
diff --git a/usr/src/uts/common/io/nfp/nfp.h b/usr/src/uts/common/io/nfp/nfp.h
new file mode 100644
index 0000000000..9704f04fbc
--- /dev/null
+++ b/usr/src/uts/common/io/nfp/nfp.h
@@ -0,0 +1,113 @@
+/*
+
+nfp.h: nFast PCI driver for Solaris 2.5, 2.6 and 2.7
+
+(C) Copyright nCipher Corporation Ltd 2001-2008 All rights reserved
+
+Copyright (c) 2008-2013 Thales e-Security All rights reserved
+
+Copyright (c) 2014 Thales UK All rights reserved
+
+history
+
+06/05/1998 jsh  Original solaris 2.6
+21/05/1999 jsh  added support for solaris 2.5
+10/06/1999 jsh  added support for solaris 2.7 (32 and 64 bit)
+16/10/2001 jsh  moved from nfast to new structure in nfdrv
+
+*/
+
+#ifndef NFP_H
+#define NFP_H
+
+#ifndef _KERNEL
+#error Hello?  this is a driver, please compile with -D_KERNEL
+#endif
+
+#if ( CH_KERNELVER < 260 )
+typedef int ioctlptr_t;
+typedef unsigned short uint16_t;
+#define DDI_GET32    ddi_getl
+#define DDI_PUT32    ddi_putl
+#define DDI_GET16    ddi_getw
+#define DDI_PUT16    ddi_putw
+#define DDI_REP_GET8 ddi_rep_getb
+#define DDI_REP_PUT8 ddi_rep_putb
+#define DDI_REP_GET32 ddi_rep_getl
+#define DDI_REP_PUT32 ddi_rep_putl
+#define PCI_CONFIG_GET16 pci_config_getw
+#else /* ( CH_KERNELVER >= 260 ) */
+typedef intptr_t ioctlptr_t;
+#define DDI_GET32    ddi_get32
+#define DDI_PUT32    ddi_put32
+#define DDI_GET16    ddi_get16
+#define DDI_PUT16    ddi_put16
+#define DDI_REP_GET8 ddi_rep_get8
+#define DDI_REP_PUT8 ddi_rep_put8
+#define DDI_REP_GET32 ddi_rep_get32
+#define DDI_REP_PUT32 ddi_rep_put32
+#define PCI_CONFIG_GET16 pci_config_get16
+#endif
+
+#if ( CH_KERNELVER < 270 )
+typedef int nfp_timeout_t;
+#define EXTRA_CB_FLAGS 0
+#define VSXPRINTF(s, n, format, ap) vsprintf (s, format, ap)
+#else /* ( CH_KERNELVER >= 270 ) */
+typedef timeout_id_t nfp_timeout_t;
+#define EXTRA_CB_FLAGS D_64BIT
+#define VSXPRINTF(s, n, format, ap) vsnprintf(s, n, format, ap)
+#endif
+
+typedef struct nfp_dev {
+  int rd_ok;
+  int wr_ok;
+
+  int ifvers;
+
+  /* for PCI push read interface */
+  unsigned char *read_buf;
+  ddi_dma_handle_t read_dma_handle;
+  ddi_dma_cookie_t read_dma_cookie;
+
+  ddi_acc_handle_t acchandle;
+
+  int rd_dma_ok;
+
+  nfp_timeout_t wrtimeout;
+  nfp_timeout_t rdtimeout;
+
+  struct buf *wr_bp;
+  int wr_ready;
+  int rd_ready;
+  int rd_pending;
+  int rd_outstanding;
+  kcondvar_t rd_cv;
+
+  struct pollhead pollhead;
+  dev_info_t *dip;
+
+  ddi_iblock_cookie_t high_iblock_cookie; /* for mutex */
+  ddi_iblock_cookie_t low_iblock_cookie; /* for mutex */
+  kmutex_t high_mutex;
+  kmutex_t low_mutex;
+  int high_intr;
+  ddi_softintr_t soft_int_id;
+  int high_read;
+  int high_write;
+
+  ddi_iblock_cookie_t iblock_cookie; /* for mutex */
+  kmutex_t isr_mutex;
+
+  kmutex_t busy_mutex;
+  int busy;
+ 
+  ddi_acc_handle_t conf_handle;
+
+  nfp_cdev common;
+  const nfpcmd_dev *cmddev;
+} nfp_dev;
+
+extern struct nfp_dev *nfp_dev_list[];
+
+#endif /* NFP_H */
diff --git a/usr/src/uts/common/io/nfp/nfp_cmd.h b/usr/src/uts/common/io/nfp/nfp_cmd.h
new file mode 100644
index 0000000000..db8af0b2f9
--- /dev/null
+++ b/usr/src/uts/common/io/nfp/nfp_cmd.h
@@ -0,0 +1,68 @@
+/*
+
+nfp_cmd.h: nCipher PCI HSM command driver decalrations
+
+(C) Copyright nCipher Corporation Ltd 2002-2008 All rights reserved
+
+Copyright (c) 2008-2013 Thales e-Security All rights reserved
+
+Copyright (c) 2014 Thales UK All rights reserved
+
+history
+
+10/10/2001 jsh  Original
+
+*/
+
+#ifndef NFPCMD_H
+#define NFPCMD_H
+
+#include "nfp_hostif.h"
+#include "nfp_error.h"
+
+/* read and write called with userspace buffer */
+
+typedef struct nfpcmd_dev {
+  const char *name;
+  unsigned short vendorid, deviceid,
+                 sub_vendorid, sub_deviceid;
+  unsigned int bar_sizes[6];    /* includes IO bit */
+  unsigned int flags;
+  nfp_err (*create)(struct nfp_cdev *pdev);
+  nfp_err (*destroy)(void * ctx);
+  nfp_err (*open)(void * ctx);
+  nfp_err (*close)(void * ctx);
+  nfp_err (*isr)(void *ctx, int *handled);
+  nfp_err (*write_block)( const char *ublock, int len, void *ctx );
+  nfp_err (*read_block)( char *ublock, int len, void *ctx, int *rcount);
+  nfp_err (*channel_update)( char *data, int len, void *ctx);
+  nfp_err (*ensure_reading)( unsigned int addr, int len, void *ctx );
+  nfp_err (*debug)( int cmd, void *ctx);
+} nfpcmd_dev;
+
+#define NFP_CMD_FLG_NEED_IOBUF	0x1
+
+/* list of all supported drivers ---------------------------------------- */
+
+extern const nfpcmd_dev *nfp_drvlist[];
+
+extern const nfpcmd_dev i21285_cmddev;
+extern const nfpcmd_dev i21555_cmddev;
+extern const nfpcmd_dev bcm5820_cmddev;
+
+#ifndef PCI_BASE_ADDRESS_SPACE_IO
+#define PCI_BASE_ADDRESS_SPACE_IO	0x1
+#endif
+
+#define NFP_MAXDEV	16
+
+
+#define NFP_MEMBAR_MASK    ~0xf
+#define NFP_IOBAR_MASK     ~0x3
+/*
+   This masks off the bottom bits of the PCI_CSR_BAR which signify that the
+   BAR is an IO BAR rather than a MEM BAR 
+*/ 
+
+#endif
+
diff --git a/usr/src/uts/common/io/nfp/nfp_common.h b/usr/src/uts/common/io/nfp/nfp_common.h
new file mode 100644
index 0000000000..d1d2100fea
--- /dev/null
+++ b/usr/src/uts/common/io/nfp/nfp_common.h
@@ -0,0 +1,68 @@
+/*
+
+(C) Copyright nCipher Corporation Ltd 2002-2008 All rights reserved
+
+Copyright (c) 2008-2013 Thales e-Security All rights reserved
+
+Copyright (c) 2014 Thales UK All rights reserved
+
+*/
+
+#ifndef NFP_COMMON_H
+#define NFP_COMMON_H
+
+#include <sys/types.h>
+#include <sys/conf.h>
+
+typedef uint32_t UINT32;
+typedef uint8_t BYTE;
+
+#define DEFINE_NFPCI_PACKED_STRUCTS
+#include "nfpci.h"
+#include "nfdev-solaris.h"
+
+typedef int oserr_t;
+
+#if CH_BIGENDIAN
+
+/* Big Endian Sparc */
+
+#define SWP32(x) \
+( (((unsigned int)(x)>>24)&0xff) | (((unsigned int)(x)>>8)&0xff00) | (((unsigned int)(x)<<8)&0xff0000) | (((unsigned int)(x)<<24)&0xff000000) ) 
+
+#define SWP16(x) ( (((x)>>8)&0xff) | (((x)<<8)&0xff00) )
+
+#define FROM_LE32_IO(x)		SWP32(*x)
+#define TO_LE32_IO(x,y)		*x=SWP32(y)
+
+#define FROM_LE32_MEM(x)	SWP32(*x)
+#define TO_LE32_MEM(x,y)	*x=SWP32(y)
+
+#define FROM_LE16_IO(x)		SWP16(*x)
+#define TO_LE16_IO(x,y)		*x=SWP16(y)
+
+#else
+
+/* Little Endian x86 */
+
+#define FROM_LE32_IO(x) (*x)
+#define TO_LE32_IO(x,y) (*x=y)
+
+#define FROM_LE32_MEM(x) (*x)
+#define TO_LE32_MEM(x,y) (*x=y)
+
+#define FROM_LE16_IO(x) (*x)
+#define TO_LE16_IO(x,y) (*x=y)
+
+#endif /* !CH_BIGENDIAN */
+
+#include <sys/types.h>
+
+#if CH_KERNELVER == 260
+#define nfp_get_lbolt( lbolt, err ) err= drv_getparm( LBOLT, lbolt )
+#else
+#define nfp_get_lbolt( lbolt, err ) { *lbolt= ddi_get_lbolt(); err= 0; }
+#endif
+
+#endif
+
diff --git a/usr/src/uts/common/io/nfp/nfp_error.h b/usr/src/uts/common/io/nfp/nfp_error.h
new file mode 100644
index 0000000000..d64cb78fd4
--- /dev/null
+++ b/usr/src/uts/common/io/nfp/nfp_error.h
@@ -0,0 +1,48 @@
+/*
+
+nfp_error.h: nCipher PCI HSM error handling
+
+(C) Copyright nCipher Corporation Ltd 2002-2008 All rights reserved
+
+Copyright (c) 2008-2013 Thales e-Security All rights reserved
+
+Copyright (c) 2014 Thales UK All rights reserved
+
+history
+
+05/12/2001 jsh  Original
+
+*/
+
+#ifndef NFP_ERROR_H
+#define NFP_ERROR_H
+
+#include "nfp_common.h"
+
+#define NFP_SUCCESS	0x0
+#define NFP_EFAULT      0x1
+#define NFP_ENOMEM	0x2
+#define NFP_EINVAL	0x3
+#define NFP_EIO		0x4
+#define NFP_ENXIO	0x5
+#define NFP_ENODEV	0x6
+#define NFP_EINTR	0x7
+#define NFP_ESTARTING	0x8
+#define NFP_EAGAIN	0x9
+#define NFP_EUNKNOWN	0x100
+
+typedef int nfp_err;
+
+extern oserr_t nfp_oserr( nfp_err nerr );
+extern nfp_err nfp_error( oserr_t oerr );
+
+#define nfr( x) \
+  return nfp_error((x))
+
+#define nfer(x, fn, msg) \
+  { oserr_t err=(x); if(err) { nfp_log( NFP_DBG1, #fn ": " msg); return nfp_error(err); } }
+
+#define er(x, fn, msg ) \
+{ nfp_err err=(x); if(err) { nfp_log( NFP_DBG1, #fn ": " msg); return err; } }
+
+#endif
diff --git a/usr/src/uts/common/io/nfp/nfp_hostif.h b/usr/src/uts/common/io/nfp/nfp_hostif.h
new file mode 100644
index 0000000000..3e7d8187e5
--- /dev/null
+++ b/usr/src/uts/common/io/nfp/nfp_hostif.h
@@ -0,0 +1,54 @@
+/*
+
+nfp_hostif.h: nCipher PCI HSM host interface declarations
+
+(C) Copyright nCipher Corporation Ltd 2002-2008 All rights reserved
+
+Copyright (c) 2008-2013 Thales e-Security All rights reserved
+
+Copyright (c) 2014 Thales UK All rights reserved
+
+history
+
+10/10/2001 jsh  Original
+
+*/
+
+#ifndef NFP_HOSTIF_H
+#define NFP_HOSTIF_H
+
+#include "nfdev-common.h"
+
+struct nfp_dev;
+
+/* common device structure */
+
+typedef struct nfp_cdev {
+  unsigned char *bar[6];
+  void *extra[6];
+
+  int busno;
+  int slotno;
+
+  void *cmdctx;
+
+  char *iobuf;
+
+  struct nfp_dev* dev;
+
+  struct nfdev_stats_str stats;
+
+} nfp_cdev;
+
+/* callbacks from command drivers -------------------------------------- */
+
+void nfp_read_complete(  struct nfp_dev *pdev, int ok);
+void nfp_write_complete( struct nfp_dev *pdev, int ok);
+
+#define NFP_READ_MAX (8 * 1024)
+#define NFP_READBUF_SIZE (NFP_READ_MAX + 8)
+#define NFP_TIMEOUT_SEC 10
+
+#define NFP_DRVNAME "nCipher nFast PCI driver"
+
+#endif
diff --git a/usr/src/uts/common/io/nfp/nfp_ifvers.c b/usr/src/uts/common/io/nfp/nfp_ifvers.c
new file mode 100644
index 0000000000..807b4f24c5
--- /dev/null
+++ b/usr/src/uts/common/io/nfp/nfp_ifvers.c
@@ -0,0 +1,51 @@
+/*
+
+(C) Copyright nCipher Corporation Ltd 2002-2008 All rights reserved
+
+Copyright (c) 2008-2013 Thales e-Security All rights reserved
+
+Copyright (c) 2014 Thales UK All rights reserved
+
+*/
+
+/*
+ * nfp_ifervs.c  - common pci interface versioning
+ *
+ * uses:
+ *
+ * int pdev->ifvers
+ *     device interface version
+ *
+ * int nfp_ifvers
+ *     interface version limit
+ * 
+ * int nfp_alloc_pci_push( nfp_dev *pdev )
+ *     allocates resources needed for PCI Push,
+ *     if not already allocated, and return True if successful
+ *
+ * void nfp_free_pci_push( nfp_dev *pdev ) {
+ *     frees any resources allocated to PCI Push
+ */
+
+void nfp_set_ifvers( nfp_dev *pdev, int vers ) {
+  if( nfp_ifvers != 0 && vers > nfp_ifvers ) {
+    nfp_log( NFP_DBG2,
+             "nfp_set_ifvers: can't set ifvers %d"
+             " as nfp_ifvers wants max ifvers %d",
+             vers, nfp_ifvers);
+    return;
+  }
+  if( vers >= NFDEV_IF_PCI_PUSH ) {
+    if(!nfp_alloc_pci_push(pdev)) {
+      nfp_log( NFP_DBG1,
+               "nfp_set_ifvers: can't set ifvers %d"
+               " as resources not available",
+               vers);
+      return;
+    }
+  } else {
+    nfp_free_pci_push(pdev);
+  }
+  pdev->ifvers= vers;
+  nfp_log( NFP_DBG3, "nfp_set_ifvers: setting ifvers %d", vers);
+}
diff --git a/usr/src/uts/common/io/nfp/nfp_osif.h b/usr/src/uts/common/io/nfp/nfp_osif.h
new file mode 100644
index 0000000000..17ffe469ce
--- /dev/null
+++ b/usr/src/uts/common/io/nfp/nfp_osif.h
@@ -0,0 +1,105 @@
+/*
+
+nfp_osif.h: nCipher PCI HSM OS interface declarations
+
+(C) Copyright nCipher Corporation Ltd 2002-2008 All rights reserved
+
+Copyright (c) 2008-2013 Thales e-Security All rights reserved
+
+Copyright (c) 2014 Thales UK All rights reserved
+
+history
+
+10/10/2001 jsh  Original
+
+*/
+
+#ifndef NFP_OSIF_H
+#define NFP_OSIF_H
+
+#include "nfp_hostif.h"
+#include "nfp_error.h"
+
+/* general typedefs ----------------------------------------------- */
+
+typedef volatile unsigned int reg32;
+typedef volatile unsigned short reg16;
+typedef volatile unsigned char reg8;
+
+/* sempaphores, mutexs and events --------------------------------- */
+
+#if 0
+extern nfp_err nfp_sema_init( nfp_sema *sema, int initial);
+extern void nfp_sema_destroy( nfp_sema *sema );
+extern void nfp_sema_post( nfp_sema *sema );
+extern void nfp_sema_wait( nfp_sema *sema );
+extern int nfp_sema_wait_sig( nfp_sema *sema );
+
+extern nfp_err nfp_mutex_init( nfp_mutex *mutex );
+extern void nfp_mutex_destroy( nfp_mutex *mutex );
+extern void nfp_mutex_enter( nfp_mutex *mutex );
+extern void nfp_mutex_exit( nfp_mutex *mutex );
+
+extern nfp_err nfp_event_init( nfp_event *event );
+extern void nfp_event_destroy( nfp_event *event );
+extern void nfp_event_set( nfp_event *event );
+extern void nfp_event_clear( nfp_event *event );
+extern void nfp_event_wait( nfp_event *event );
+extern void nfp_event_wait_sig( nfp_event *event );
+
+#endif
+
+/* timeouts ------------------------------------------------------ */
+
+extern void nfp_sleep( int ms );
+
+/* memory handling ----------------------------------------------- */
+
+#define KMALLOC_DMA	0
+#define KMALLOC_CACHED	1
+
+extern void *nfp_kmalloc( int size, int flags );
+extern void *nfp_krealloc( void *ptr, int size, int flags );
+extern void nfp_kfree( void * );
+
+/* config space access ------------------------------------------------ */
+
+/* return Little Endian 32 bit config register */
+extern nfp_err nfp_config_inl( nfp_cdev *pdev, int offset, unsigned int *res );
+
+/* io space access ------------------------------------------------ */
+
+extern unsigned int nfp_inl( nfp_cdev *pdev, int bar, int offset );
+extern unsigned short nfp_inw( nfp_cdev *pdev, int bar, int offset );
+extern void nfp_outl( nfp_cdev *pdev, int bar, int offset, unsigned int data );
+extern void nfp_outw( nfp_cdev *pdev, int bar, int offset, unsigned short data );
+
+/* user and device memory space access ---------------------------- */
+
+/* NB these 2 functions are not guarenteed to be re-entrant for a given device */
+extern nfp_err nfp_copy_from_user_to_dev( nfp_cdev *cdev, int bar, int offset, const char *ubuf, int len);
+extern nfp_err nfp_copy_to_user_from_dev( nfp_cdev *cdev, int bar, int offset, char *ubuf, int len);
+
+extern nfp_err nfp_copy_from_user( char *kbuf, const char *ubuf, int len );
+extern nfp_err nfp_copy_to_user( char *ubuf, const char *kbuf, int len );
+
+extern nfp_err nfp_copy_from_dev( nfp_cdev *cdev, int bar, int offset, char *kbuf, int len );
+extern nfp_err nfp_copy_to_dev( nfp_cdev *cdev, int bar, int offset, const char *kbuf, int len);
+
+/* debug ------------------------------------------------------------ */
+
+#define NFP_DBG1	1
+#define NFP_DBGE	NFP_DBG1
+#define NFP_DBG2	2
+#define NFP_DBG3	3
+#define NFP_DBG4	4
+
+#ifdef STRANGE_VARARGS
+extern void nfp_log();
+#else
+extern void nfp_log( int severity, const char *format, ...);
+#endif
+
+extern int nfp_debug;
+
+#endif
diff --git a/usr/src/uts/common/io/nfp/nfpci.h b/usr/src/uts/common/io/nfp/nfpci.h
new file mode 100644
index 0000000000..793f5995e6
--- /dev/null
+++ b/usr/src/uts/common/io/nfp/nfpci.h
@@ -0,0 +1,171 @@
+/*
+
+(C) Copyright nCipher Corporation Ltd 2002-2008 All rights reserved
+
+Copyright (c) 2008-2013 Thales e-Security All rights reserved
+
+Copyright (c) 2014 Thales UK All rights reserved
+
+*/
+
+/*
+*
+*  NFPCI.H	- nFast PCI interface definition file
+*
+*
+*
+*  1998.06.09	IH	Started
+*
+* The interface presented by nFast PCI devices consists of:
+*
+* A region of shared RAM used for data transfer & control information
+* A doorbell interrupt register, so both sides can give each other interrupts
+* A number of DMA channels for transferring data
+*/
+
+#ifndef NFPCI_H
+#define NFPCI_H
+
+/* Sizes of some regions */
+#define NFPCI_RAM_MINSIZE	0x00100000
+/* This is the minimum size of shared RAM. In future it may be possible to
+   negotiate larger sizes of shared RAM or auto-detect how big it is */
+#define NFPCI_RAM_MINSIZE_JOBS	0x00020000 /* standard jobs only */
+#define NFPCI_RAM_MINSIZE_KERN	0x00040000 /* standard and kernel jobs */
+
+/* Offsets within shared memory space.
+   The following main regions are:
+     jobs input area
+     jobs output area
+     kernel jobs input area
+     kernel output area
+*/
+
+#define NFPCI_OFFSET_JOBS		0x00000000
+#define NFPCI_OFFSET_JOBS_WR		0x00000000
+#define NFPCI_OFFSET_JOBS_RD		0x00010000
+#define NFPCI_OFFSET_KERN		0x00020000
+#define NFPCI_OFFSET_KERN_WR		0x00020000
+#define NFPCI_OFFSET_KERN_RD		0x00030000
+
+/* Interrupts, defined by bit position in doorbell register */
+
+/* Interrupts from device to host */
+#define NFAST_INT_DEVICE_WRITE_OK               0x00000001
+#define NFAST_INT_DEVICE_WRITE_FAILED           0x00000002
+#define NFAST_INT_DEVICE_READ_OK                0x00000004
+#define NFAST_INT_DEVICE_READ_FAILED            0x00000008
+#define NFAST_INT_DEVICE_KERN_WRITE_OK		0x00000010
+#define NFAST_INT_DEVICE_KERN_WRITE_FAILED	0x00000020
+#define NFAST_INT_DEVICE_KERN_READ_OK		0x00000040
+#define NFAST_INT_DEVICE_KERN_READ_FAILED	0x00000080
+
+/* Interrupts from host to device */
+#define NFAST_INT_HOST_WRITE_REQUEST            0x00010000
+#define NFAST_INT_HOST_READ_REQUEST             0x00020000
+#define NFAST_INT_HOST_DEBUG                    0x00040000
+#define NFAST_INT_HOST_KERN_WRITE_REQUEST	0x00080000
+#define NFAST_INT_HOST_KERN_READ_REQUEST	0x00100000
+
+/* Ordinary job submission ------------------------ */
+
+/* The NFPCI_OFFSET_JOBS_WR and NFPCI_OFFSET_JOBS_RD regions are defined
+   by the following (byte) address offsets... */
+
+#define NFPCI_OFFSET_CONTROL	0x0
+#define NFPCI_OFFSET_LENGTH	0x4
+#define NFPCI_OFFSET_DATA	0x8
+#define NFPCI_OFFSET_PUSH_ADDR	0x8
+
+#define NFPCI_JOBS_WR_CONTROL	(NFPCI_OFFSET_JOBS_WR + NFPCI_OFFSET_CONTROL)
+#define NFPCI_JOBS_WR_LENGTH	(NFPCI_OFFSET_JOBS_WR + NFPCI_OFFSET_LENGTH)
+#define NFPCI_JOBS_WR_DATA	(NFPCI_OFFSET_JOBS_WR + NFPCI_OFFSET_DATA)
+#define NFPCI_MAX_JOBS_WR_LEN		(0x0000FFF8)
+
+#define NFPCI_JOBS_RD_CONTROL	(NFPCI_OFFSET_JOBS_RD + NFPCI_OFFSET_CONTROL)
+#define NFPCI_JOBS_RD_LENGTH	(NFPCI_OFFSET_JOBS_RD + NFPCI_OFFSET_LENGTH)
+#define NFPCI_JOBS_RD_DATA	(NFPCI_OFFSET_JOBS_RD + NFPCI_OFFSET_DATA)
+/* address in PCI space of host buffer for NFPCI_JOB_CONTROL_PCI_PUSH */
+#define NFPCI_JOBS_RD_PUSH_ADDR	(NFPCI_OFFSET_JOBS_RD + NFPCI_OFFSET_PUSH_ADDR)
+#define NFPCI_MAX_JOBS_RD_LEN		(0x000FFF8)
+
+/* Kernel inferface job submission ---------------- */
+
+#define NFPCI_KERN_WR_CONTROL   (NFPCI_OFFSET_KERN_WR + NFPCI_OFFSET_CONTROL)
+#define NFPCI_KERN_WR_LENGTH    (NFPCI_OFFSET_KERN_WR + NFPCI_OFFSET_LENGTH)
+#define NFPCI_KERN_WR_DATA      (NFPCI_OFFSET_KERN_WR + NFPCI_OFFSET_DATA)
+#define NFPCI_MAX_KERN_WR_LEN      (0x0000FFF8)
+
+#define NFPCI_KERN_RD_CONTROL   (NFPCI_OFFSET_KERN_RD + NFPCI_OFFSET_CONTROL)
+#define NFPCI_KERN_RD_LENGTH    (NFPCI_OFFSET_KERN_RD + NFPCI_OFFSET_LENGTH)
+#define NFPCI_KERN_RD_DATA      (NFPCI_OFFSET_KERN_RD + NFPCI_OFFSET_DATA)
+/* address in PCI space of host buffer for NFPCI_JOB_CONTROL_PCI_PUSH */
+#define NFPCI_KERN_RD_ADDR      (NFPCI_OFFSET_KERN_RD + NFPCI_OFFSET_PUSH_ADDR)
+#define NFPCI_MAX_KERN_RD_LEN		(0x000FFF8)
+
+#ifdef DEFINE_NFPCI_PACKED_STRUCTS
+typedef struct
+{
+  UINT32	controlword;
+  UINT32	length;		/* length of data to follow */
+  union {
+    BYTE	data[1];
+    UINT32	addr;
+  } uu;
+}
+  NFPCI_JOBS_BLOCK;
+#endif
+
+
+#define NFPCI_JOB_CONTROL		0x00000001
+#define NFPCI_JOB_CONTROL_PCI_PUSH	0x00000002
+/*
+   The 'Control' word is analogous to the SCSI read/write address;
+   1 = standard push/pull IO
+   2 = push/push IO
+
+   To submit a block of job data, the host:
+   - sets the (32-bit, little-endian) word at NFPCI_JOBS_WR_CONTROL to NFPCI_JOB_CONTROL
+   - sets the word at NFPCI_JOBS_WR_LENGTH to the length of the data
+   - copies the data to NFPCI_JOBS_WR_DATA
+   - sets interrupt NFAST_INT_HOST_WRITE_REQUEST in the doorbell register
+   - awaits the NFAST_INT_DEVICE_WRITE_OK (or _FAILED) interrupts back
+
+   To read a block of jobs back, the host:
+   - sets the word at NFPCI_JOBS_RD_CONTROL to NFPCI_JOB_CONTROL
+   - sets the word at NFPCI_JOBS_RD_LENGTH to the max length for returned data
+   - sets interrupt NFAST_INT_HOST_READ_REQUEST
+   - awaits the NFAST_INT_DEVICE_READ_OK (or _FAILED) interrupt
+   - reads the data from NFPCI_JOBS_RD_DATA; the module will set the word at
+	NFPCI_JOBS_RD_LENGTH to its actual length.
+
+   Optionally the host can request the PCI read data to be pushed to host PCI mapped ram:
+   - allocates a contiguous PCI addressable buffer for a NFPCI_JOBS_BLOCK of max
+        size NFPCI_MAX_JOBS_RD_LEN (or NFPCI_MAX_KERN_RD_LEN) + 8
+   - sets the word at NFPCI_JOBS_RD_CONTROL to NFPCI_JOB_CONTROL_PCI_PUSH
+   - sets the word at NFPCI_JOBS_RD_LENGTH to the max length for returned data
+   - sets the word at NFPCI_JOBS_RD_PUSH_ADDR to be the host PCI address of
+        the buffer
+   - sets interrupt NFAST_INT_HOST_READ_REQUEST
+   - awaits the NFAST_INT_DEVICE_READ_OK (or _FAILED) interrupt
+   - reads the data from the buffer at NFPCI_OFFSET_DATA in the buffer.  The
+        module will set NFPCI_OFFSET_LENGTH to the actual length.
+*/
+
+#define NFPCI_SCRATCH_CONTROL       0
+
+#define NFPCI_SCRATCH_CONTROL_HOST_MOI   (1<<0)
+#define NFPCI_SCRATCH_CONTROL_MODE_SHIFT 1
+#define NFPCI_SCRATCH_CONTROL_MODE_MASK  (3<<NFPCI_SCRATCH_CONTROL_MODE_SHIFT)
+
+#define NFPCI_SCRATCH_STATUS        1
+
+#define NFPCI_SCRATCH_STATUS_MONITOR_MOI         (1<<0)
+#define NFPCI_SCRATCH_STATUS_APPLICATION_MOI     (1<<1)
+#define NFPCI_SCRATCH_STATUS_APPLICATION_RUNNING (1<<2)
+#define NFPCI_SCRATCH_STATUS_ERROR               (1<<3)
+
+#define NFPCI_SCRATCH_ERROR_LO      2
+#define NFPCI_SCRATCH_ERROR_HI      3
+
+#endif
diff --git a/usr/src/uts/common/io/nfp/osif.c b/usr/src/uts/common/io/nfp/osif.c
new file mode 100644
index 0000000000..fba62f9a37
--- /dev/null
+++ b/usr/src/uts/common/io/nfp/osif.c
@@ -0,0 +1,184 @@
+/*
+
+(C) Copyright nCipher Corporation Ltd 2002-2008 All rights reserved
+
+Copyright (c) 2008-2013 Thales e-Security All rights reserved
+
+Copyright (c) 2014 Thales UK All rights reserved
+
+*/
+
+#include <sys/types.h>
+#include <sys/param.h>
+#include <sys/errno.h>
+#include <sys/file.h>
+#include <sys/conf.h>
+#include <sys/uio.h>
+#include <sys/map.h>
+#include <sys/debug.h>
+#include <sys/modctl.h>
+#include <sys/kmem.h>
+#include <sys/cmn_err.h>
+#include <sys/open.h>
+#include <sys/stat.h>
+#include <sys/ddi.h>
+#include <sys/sunddi.h>
+#include <sys/pci.h>
+
+#include "nfp_common.h"
+#include "nfp_hostif.h"
+#include "nfp_error.h"
+#include "nfp_osif.h"
+#include "nfp_cmd.h"
+#include "nfp.h"
+#include "autoversion.h"
+
+/* config space access ---------------------------------- */
+
+nfp_err nfp_config_inl( nfp_cdev *pdev, int offset, unsigned int *res ) {
+  unsigned int tmp32;
+  if ( !pdev || !pdev->dev || !pdev->dev->conf_handle )
+    return NFP_ENODEV;
+
+/* pci_config_get32() does byte swapping, so put back to LE */
+  tmp32 = pci_config_get32( pdev->dev->conf_handle, offset );
+  TO_LE32_IO(res, tmp32);
+
+  return NFP_SUCCESS;
+}
+
+/* user space memory access ---------------------------------- */
+
+nfp_err nfp_copy_from_user( char *kbuf, const char *ubuf, int len) {
+  bcopy(ubuf, kbuf, len);
+  return 0;
+}
+
+nfp_err nfp_copy_to_user( char *ubuf, const char *kbuf, int len) {
+  bcopy(kbuf, ubuf, len);
+  return 0;
+}
+
+nfp_err nfp_copy_from_user_to_dev( nfp_cdev *cdev, int bar, int offset, const char *ubuf, int len) {
+  /* dirty hack on Solaris, as we are called from strategy we are, in fact, copying from kernel mem */
+  return nfp_copy_to_dev( cdev, bar, offset, ubuf, len );
+}
+
+nfp_err nfp_copy_to_user_from_dev( nfp_cdev *cdev, int bar, int offset, char *ubuf, int len) {
+  /* dirty hack on Solaris, as we are called from strategy we are, in fact, copying to kernel mem */
+  return nfp_copy_from_dev( cdev, bar, offset, ubuf, len );
+}
+
+nfp_err nfp_copy_from_dev( nfp_cdev *cdev, int bar, int offset, char *kbuf, int len) {
+  if( len & 0x3 || offset & 0x3 )
+    DDI_REP_GET8( cdev->extra[bar], (unsigned char *)kbuf, cdev->bar[bar] + offset, len, DDI_DEV_AUTOINCR);
+  else
+    /* LINTED: alignment */
+    DDI_REP_GET32( cdev->extra[bar], (unsigned int *)kbuf, (unsigned int *)(cdev->bar[bar] + offset), len / 4, DDI_DEV_AUTOINCR);
+  return NFP_SUCCESS;
+}
+
+nfp_err nfp_copy_to_dev( nfp_cdev *cdev, int bar, int offset, const char *kbuf, int len) {
+  if( len & 0x3 || offset & 0x3 )
+    DDI_REP_PUT8( cdev->extra[bar], (unsigned char *)kbuf, cdev->bar[bar] + offset, len, DDI_DEV_AUTOINCR );
+  else
+    /* LINTED: alignment */
+    DDI_REP_PUT32( cdev->extra[bar], (unsigned int *)kbuf, (unsigned int *)(cdev->bar[bar] + offset), len / 4, DDI_DEV_AUTOINCR );
+  return NFP_SUCCESS;
+}
+
+/* pci io space access --------------------------------------- */
+
+unsigned int nfp_inl( nfp_cdev *pdev, int bar, int offset ) {
+  nfp_log( NFP_DBG3, "nfp_inl: addr %x", (uintptr_t) pdev->bar[bar] + offset);
+  /* LINTED: alignment */
+  return DDI_GET32( pdev->extra[bar], (uint32_t *)(pdev->bar[bar] + offset) );
+}
+
+unsigned short nfp_inw( nfp_cdev *pdev, int bar, int offset ) {
+  nfp_log( NFP_DBG3, "nfp_inw: addr %x", (uintptr_t) pdev->bar[bar] + offset);
+  /* LINTED: alignment */
+  return DDI_GET16( pdev->extra[bar], (unsigned short *)(pdev->bar[ bar ] + offset) );
+}
+
+void nfp_outl( nfp_cdev *pdev, int bar, int offset, unsigned int data ) {
+  nfp_log( NFP_DBG3, "nfp_outl: addr %x, data %x", (uintptr_t) pdev->bar[bar] + offset, data);
+  /* LINTED: alignment */
+  DDI_PUT32( pdev->extra[bar], (uint32_t *)(pdev->bar[ bar ] + offset), data ); 
+}
+
+void nfp_outw( nfp_cdev *pdev, int bar, int offset, unsigned short data ) {
+  nfp_log( NFP_DBG3, "nfp_outl: addr %x, data %x", (uintptr_t) pdev->bar[bar] + offset, data);
+  /* LINTED: alignment */
+  DDI_PUT16( pdev->extra[bar], (unsigned short *)(pdev->bar[ bar ] + offset), data ); 
+}
+
+/* logging ---------------------------------------------------- */
+
+void nfp_log( int level, const char *fmt, ...)
+{
+  auto char buf[256];
+  va_list ap;
+
+  switch (level) {
+  case NFP_DBG4: if (nfp_debug < 4) break;
+  /*FALLTHROUGH*/
+  case NFP_DBG3: if (nfp_debug < 3) break;
+  /*FALLTHROUGH*/
+  case NFP_DBG2: if (nfp_debug < 2) break;
+  /*FALLTHROUGH*/
+  case NFP_DBG1: if (nfp_debug < 1) break;
+  /*FALLTHROUGH*/
+  default:
+    va_start(ap, fmt);
+    (void) vsnprintf(buf, 256, fmt, ap);
+    va_end(ap);
+    cmn_err(CE_CONT, "!" VERSION_COMPNAME " " VERSION_NO ": %s\n", buf);
+    break;
+  }
+}
+
+struct errstr {
+  int oserr;
+  nfp_err nferr;
+};
+
+
+static struct errstr errtab[] = {
+  { EFAULT, NFP_EFAULT },
+  { ENOMEM, NFP_ENOMEM },
+  { EINVAL, NFP_EINVAL },
+  { EIO,    NFP_EIO    },
+  { ENXIO,  NFP_ENXIO  },
+  { ENODEV, NFP_ENODEV  },
+  { EINVAL, NFP_EUNKNOWN },
+  { 0, 0 }
+};
+
+nfp_err nfp_error( int oserr )
+{
+  struct errstr *perr;
+  if(!oserr)
+    return 0;
+  perr= errtab;
+  while(perr->nferr) {
+   if(perr->oserr == oserr)
+     return perr->nferr;
+   perr++;
+  }
+  return NFP_EUNKNOWN;
+}
+
+int nfp_oserr( nfp_err nferr )
+{
+  struct errstr *perr;
+  if(nferr == NFP_SUCCESS)
+    return 0;
+  perr= errtab;
+  while(perr->nferr) {
+   if(perr->nferr == nferr)
+     return perr->oserr;
+   perr++;
+  }
+  return EIO;
+}
diff --git a/usr/src/uts/common/io/overlay/overlay.c b/usr/src/uts/common/io/overlay/overlay.c
new file mode 100644
index 0000000000..78bc418548
--- /dev/null
+++ b/usr/src/uts/common/io/overlay/overlay.c
@@ -0,0 +1,2172 @@
+/*
+ * This file and its contents are supplied under the terms of the
+ * Common Development and Distribution License ("CDDL"), version 1.0.
+ * You may only use this file in accordance with the terms of version
+ * 1.0 of the CDDL.
+ *
+ * A full copy of the text of the CDDL should have accompanied this
+ * source.  A copy of the CDDL is also available via the Internet at
+ * http://www.illumos.org/license/CDDL.
+ */
+
+/*
+ * Copyright 2015 Joyent, Inc.
+ */
+
+/*
+ * Overlay Devices
+ *
+ * Overlay devices provide a means for creating overlay networks, a means of
+ * multiplexing multiple logical, isolated, and discrete layer two and layer
+ * three networks on top of one physical network.
+ *
+ * In general, these overlay devices encapsulate the logic to answer two
+ * different questions:
+ *
+ *   1) How should I transform a packet to put it on the wire?
+ *   2) Where should I send a transformed packet?
+ *
+ * Each overlay device is presented to the user as a GLDv3 device. While the
+ * link itself cannot have an IP interface created on top of it, it allows for
+ * additional GLDv3 devices, such as a VNIC, to be created on top of it which
+ * can be plumbed up with IP interfaces.
+ *
+ *
+ * --------------------
+ * General Architecture
+ * --------------------
+ *
+ * The logical overlay device that a user sees in dladm(1M) is a combination of
+ * two different components that work together. The first component is this
+ * kernel module, which is responsible for answering question one -- how should
+ * I transform a packet to put it on the wire.
+ *
+ * The second component is what we call the virtual ARP daemon, or varpd. It is
+ * a userland component that is responsible for answering the second question --
+ * Where should I send a transformed packet. Instances of the kernel overlay
+ * GLDv3 device ask varpd the question of where should a packet go.
+ *
+ * The split was done for a few reasons. Importantly, we wanted to keep the act
+ * of generating encapsulated packets in the kernel so as to ensure that the
+ * general data path was fast and also kept simple. On the flip side, while the
+ * question of where should something go may be simple, it may often be
+ * complicated and need to interface with several different external or
+ * distributed systems. In those cases, it's simpler to allow for the full
+ * flexibility of userland to be brought to bear to solve that problem and in
+ * general, the path isn't very common.
+ *
+ * The following is what makes up the logical overlay device that a user would
+ * create with dladm(1M).
+ *
+ *       Kernel                                     Userland
+ *   . . . . . . . . . . . . . . . . . . . . .   . . . . . . . . . . . . .
+ *   . +--------+   +--------+  +--------+   .   .                       .
+ *   . | VNIC 0 |   | VNIC 1 |  | VNIC 2 |   .   .                       .
+ *   . +--------+   +--------+  +--------+   .   .                       .
+ *   .     |            |           |        .   .                       .
+ *   .     |            |           |        .   .                       .
+ *   .     +------------+-----------+        .   .                       .
+ *   .                  |              . . /dev/overlay                  .
+ *   .           +--------------+      .     .   .       +------------+  .
+ *   .           |              |      .     .   .       |            |  .
+ *   .           |    Overlay   |======*=================|   Virtual  |  .
+ *   .           | GLDv3 Device |========================| ARP Daemon |  .
+ *   .           |              |            .   .       |            |  .
+ *   .           +--------------+            .   .       +------------+  .
+ *   .                  |                    .   .              |        .
+ *   .                  |                    .   .              |        .
+ *   .           +----------------+          .   .         +--------+    .
+ *   .           |  Overlay       |          .   .         | varpd  |    .
+ *   .           |  Encapsulation |          .   .         | Lookup |    .
+ *   .           |  Plugin        |          .   .         | Plugin |    .
+ *   .           +----------------+          .   .         +--------+    .
+ *   . . . . . . . . . . . . . . . . . . . . .   . . . . . . . . . . . . .
+ *
+ *
+ * This image shows the two different components and where they live.
+ * Importantly, it also shows that both the kernel overlay device and the
+ * userland varpd both support plugins. The plugins actually implement the
+ * things that users care about and the APIs have been designed to try to
+ * minimize the amount of things that a module writer needs to worry about it.
+ *
+ * IDENTIFIERS
+ *
+ * Every overlay device is defined by a unique identifier which is the overlay
+ * identifier. Its purpose is similar to that of a VLAN identifier, it's a
+ * unique number that is used to differentiate between different entries on the
+ * wire.
+ *
+ * ENCAPSULATION
+ *
+ * An overlay encapsulation plugin is a kernel miscellaneous module whose
+ * purpose is to contain knowledge about how to transform packets to put them
+ * onto the wire and to take them off. An example of an encapsulation plugin is
+ * vxlan. It's also how support for things like nvgre or geneve would be brought
+ * into the system.
+ *
+ * Each encapsulation plugins defines a series of operation vectors and
+ * properties. For the full details on everything they should provide, please
+ * read uts/common/sys/overlay_plugin.h. The encapsulation plugin is responsible
+ * for telling the system what information is required to send a packet. For
+ * example, vxlan is defined to send everything over a UDP packet and therefore
+ * requires a port and an IP address, while nvgre on the other hand is its own
+ * IP type and therefore just requires an IP address. In addition, it also
+ * provides information about the kind of socket that should be created. This is
+ * used by the kernel multiplexor, more of that in the Kernel Components
+ * section.
+ *
+ * LOOKUPS
+ *
+ * The kernel communicates requests for lookups over the character device
+ * /dev/overlay. varpd is responsible for listening for requests on that device
+ * and answering them. The character device is specific to the target path and
+ * varpd.
+ *
+ * Much as the kernel overlay module handles the bulk of the scaffolding but
+ * leaves the important work to the encapsulation plugin, varpd provides a
+ * similar role and leaves the full brunt of lookups to a userland dynamic
+ * shared object which implements the logic of lookups.
+ *
+ * Each lookup plugin defines a series of operation vectors and properties. For
+ * the full details on everything that they should provide, please read
+ * lib/varpd/libvarpd/libvarpd_provider.h. Essentially, they are given a MAC
+ * address and asked to give an address on the physical network that it should
+ * be sent to. In addition, they handle questions related to how to handle
+ * things like broadcast and multicast traffic, etc.
+ *
+ * ----------
+ * Properties
+ * ----------
+ *
+ * A device from a dladm perspective has a unique set of properties that are
+ * combined from three different sources:
+ *
+ *   1) Generic properties that every overlay device has
+ *   2) Properties that are specific to the encapsulation plugin
+ *   3) Properties that are specific to the lookup plugin
+ *
+ * All of these are exposed in a single set of properties in dladm. Note that
+ * these are not necessarily traditional link properties. However, if something
+ * is both a traditional GLDv3 link property, say the MTU of a device, and a
+ * specific property here, than the driver ensures that all existing GLDv3
+ * specific means of manipulating it are used and wraps up its private property
+ * interfaces to ensure that works.
+ *
+ * Properties in the second and third category are prefixed with the name of
+ * their module. For example, the vxlan encapsulation module has a property
+ * called the 'listen_ip'. This property would show up in dladm as
+ * 'vxlan/listen_ip'. This allows different plugins to both use similar names
+ * for similar properties and to also have independent name spaces so that
+ * overlapping names do not conflict with anything else.
+ *
+ * While the kernel combines both sets one and two into a single coherent view,
+ * it does not do anything with respect to the properties that are owned by the
+ * lookup plugin -- those are owned wholly by varpd. Instead, libdladm is in
+ * charge of bridging these two worlds into one magical experience for the user.
+ * It carries the burden of knowing about both overlay specific and varpd
+ * specific properties. Importantly, we want to maintain this distinction. We
+ * don't want to treat the kernel as an arbitrary key/value store for varpd and
+ * we want the kernel to own its own data and not have to ask userland for
+ * information that it owns.
+ *
+ * Every property in the system has the following attributes:
+ *
+ *   o A name
+ *   o A type
+ *   o A size
+ *   o Permissions
+ *   o Default value
+ *   o Valid value ranges
+ *   o A value
+ *
+ * Everything except for the value is obtained by callers through the propinfo
+ * callbacks and a property has a maximum size of OVERLAY_PROP_SIZEMAX,
+ * currently 256 bytes.
+ *
+ * The following are the supported types of properties:
+ *
+ * 	OVERLAY_PROP_T_INT
+ *
+ * 		A signed integer, its length is 8 bytes, corresponding to a
+ * 		int64_t.
+ *
+ * 	OVERLAY_PROP_T_UINT
+ *
+ * 		An unsigned integer, its length is 8 bytes, corresponding to a
+ * 		uint64_t.
+ *
+ * 	OVERLAY_PROP_T_IP
+ *
+ * 		A struct in6_addr, it has a fixed size.
+ *
+ * 	OVERLAY_PROP_T_STRING
+ *
+ * 		A null-terminated character string encoded in either ASCII or
+ * 		UTF-8. Note that the size of the string includes the null
+ * 		terminator.
+ *
+ * The next thing that we apply to a property is its permission. The permissions
+ * are put together by the bitwise or of the following flags and values.
+ *
+ * 	OVERLAY_PROP_PERM_REQ
+ *
+ * 		This indicates a required property. A property that is required
+ * 		must be set by a consumer before the device can be created. If a
+ * 		required property has a default property, this constraint is
+ * 		loosened because the default property defines the value.
+ *
+ * 	OVERLAY_PORP_PERM_READ
+ *
+ * 		This indicates that a property can be read. All properties will
+ * 		have this value set.
+ *
+ * 	OVERLAY_PROP_PERM_WRITE
+ *
+ * 		This indicates that a property can be written to and thus
+ * 		updated by userland. Properties that are only intended to
+ * 		display information, will not have OVERLAY_PROP_PERM_WRITE set.
+ *
+ * In addition, a few additional values are defined as a convenience to
+ * consumers. The first, OVERLAY_PROP_PERM_RW, is a combination of
+ * OVERLAY_PROP_PERM_READ and OVERLAY_PERM_PROP_WRITE. The second,
+ * OVERLAY_PROP_PERM_RRW, is a combination of OVERLAY_PROP_PERM_REQ,
+ * OVERLAY_PROP_PERM_READ, and OVERLAY_PROP_PERM_WRITE. The protection mode of a
+ * property should generally be a constant across its lifetime.
+ *
+ * A property may optionally have a default value. If it does have a default
+ * value, and that property is not set to be a different value, then the default
+ * value is inherited automatically. It also means that if the default value is
+ * acceptable, there is no need to set the value for a required property. For
+ * example, the vxlan module has the vxlan/listen_port property which is
+ * required, but has a default value of 4789 (the IANA assigned port). Because
+ * of that default value, there is no need for it to be set.
+ *
+ * Finally, a property may declare a list of valid values. These valid values
+ * are used for display purposes, they are not enforced by the broader system,
+ * but merely allow a means for the information to be communicated to the user
+ * through dladm(1M). Like a default value, this is optional.
+ *
+ * The general scaffolding does not do very much with respect to the getting and
+ * setting of properties. That is really owned by the individual plugins
+ * themselves.
+ *
+ * -----------------------------
+ * Destinations and Plugin Types
+ * -----------------------------
+ *
+ * Both encapsulation and lookup plugins define the kinds of destinations that
+ * they know how to support. There are three different pieces of information
+ * that can be used to address to a destination currently, all of which is
+ * summarized in the type overlay_point_t. Any combination of these is
+ * supported.
+ *
+ * 	OVERLAY_PLUGIN_D_ETHERNET
+ *
+ * 		An Ethernet MAC address is required.
+ *
+ * 	OVERLAY_PLUGIN_D_IP
+ *
+ * 		An IP address is required. All IP addresses used by the overlay
+ * 		system are transmitted as IPv6 addresses. IPv4 addresses can be
+ * 		represented by using IPv4-mapped IPv6 addresses.
+ *
+ * 	OVERLAY_PLUGIN_D_PORT
+ *
+ * 		A TCP/UDP port is required.
+ *
+ * A kernel encapsulation plugin declares which of these that it requires, it's
+ * a static set. On the other hand, a userland lookup plugin can be built to
+ * support all of these or any combination thereof. It gets passed the required
+ * destination type, based on the kernel encapsulation method, and then it makes
+ * the determination as to whether or not it supports it. For example, the
+ * direct plugin can support either an IP or both an IP and a port, it simply
+ * doesn't display the direct/dest_port property in the cases where a port is
+ * not required to support this.
+ *
+ * The user lookup plugins have two different modes of operation which
+ * determines how they interact with the broader system and how look ups are
+ * performed. These types are:
+ *
+ * 	OVERLAY_TARGET_POINT
+ *
+ * 		A point to point plugin has a single static definition for where
+ * 		to send all traffic. Every packet in the system always gets sent
+ * 		to the exact same destination which is programmed into the
+ * 		kernel when the general device is activated.
+ *
+ * 	OVERLAY_TARGET_DYNAMIC
+ *
+ * 		A dynamic plugin does not have a single static definition.
+ * 		Instead, for each destination, the kernel makes an asynchronous
+ * 		request to varpd to determine where the packet should be routed,
+ * 		and if a specific destination is found, then that destination is
+ * 		cached in the overlay device's target cache.
+ *
+ * This distinction, while important for the general overlay device's operation,
+ * is not important to the encapsulation plugins. They don't need to know about
+ * any of these pieces. It's just a concern for varpd, the userland plugin, and
+ * the general overlay scaffolding.
+ *
+ * When an overlay device is set to OVERLAY_TARGET_POINT, then it does not
+ * maintain a target cache, and instead just keeps track of the destination and
+ * always sends encapsulated packets to that address. When the target type is of
+ * OVERLAY_TARGET_DYNAMIC, then the kernel maintains a cache of all such
+ * destinations. These destinations are kept around in an instance of a
+ * reference hash that is specific to the given overlay device. Entries in the
+ * cache can be invalidated and replaced by varpd and its lookup plugins.
+ *
+ * ----------------------------------
+ * Kernel Components and Architecture
+ * ----------------------------------
+ *
+ * There are multiple pieces inside the kernel that work together, there is the
+ * general overlay_dev_t structure, which is the logical GLDv3 device, but it
+ * itself has references to things like an instance of an encapsulation plugin,
+ * a pointer to a mux and a target cache. It can roughly be summarized in the
+ * following image:
+ *
+ *     +------------------+
+ *     | global           |
+ *     | overlay list     |
+ *     | overlay_dev_list |
+ *     +------------------+
+ *        |
+ *        |  +-----------------------+            +---------------+
+ *        +->| GLDv3 Device          |----------->| GLDv3 Device  | -> ...
+ *           | overlay_dev_t         |            | overlay_dev_t |
+ *           |                       |            +---------------+
+ *           |                       |
+ *           | mac_handle_t     -----+---> GLDv3 handle to MAC
+ *           | datalink_id_t    -----+---> Datalink ID used by DLS
+ *           | overlay_dev_flag_t ---+---> Device state
+ *           | uint_t           -----+---> Curent device MTU
+ *           | uint_t           -----+---> In-progress RX operations
+ *           | uint_t           -----+---> In-progress TX operations
+ *           | char[]           -----+---> FMA degraded message
+ *           | void *           -----+---> plugin private data
+ *           | overlay_target_t * ---+---------------------+
+ *           | overlay_plugin_t * ---+---------+           |
+ *           +-----------------------+         |           |
+ *                           ^                 |           |
+ *   +--------------------+  |                 |           |
+ *   | Kernel Socket      |  |                 |           |
+ *   | Multiplexor        |  |                 |           |
+ *   | overlay_mux_t      |  |                 |           |
+ *   |                    |  |                 |           |
+ *   | avl_tree_t        -+--+                 |           |
+ *   | uint_t            -+--> socket family   |           |
+ *   | uint_t            -+--> socket type     |           |
+ *   | uint_t            -+--> socket protocol |           |
+ *   | ksocket_t         -+--> I/O socket      |           |
+ *   | struct sockaddr * -+--> ksocket address |           |
+ *   | overlay_plugin_t --+--------+           |           |
+ *   +--------------------+        |           |           |
+ *                                 |           |           |
+ *   +-------------------------+   |           |           |
+ *   | Encap Plugin            |<--+-----------+           |
+ *   | overlay_plugin_t        |                           |
+ *   |                         |                           |
+ *   | char *               ---+--> plugin name            |
+ *   | overlay_plugin_ops_t * -+--> plugin downcalls       |
+ *   | char ** (props)      ---+--> property list          |
+ *   | uint_t               ---+--> id length              |
+ *   | overlay_plugin_flags_t -+--> plugin flags           |
+ *   | overlay_plugin_dest_t --+--> destination type       v
+ *   +-------------------------+                    +-------------------------+
+ *                                                  |   Target Cache          |
+ *                                                  |   overlay_target_t      |
+ *                                                  |                         |
+ *                                    cache mode <--+- overlay_target_mode_t  |
+ *                                     dest type <--+- overlay_plugin_dest_t  |
+ *                                   cache flags <--+- overlay_target_flag_t  |
+ *                                     varpd id  <--+- uint64_t               |
+ *                       outstanding varpd reqs. <--+- uint_t                 |
+ *                   OVERLAY_TARGET_POINT state  <--+- overlay_target_point_t |
+ *               OVERLAY_TARGET_DYNAMIC state <-+---+- overlay_target_dyn_t   |
+ *                                              |   +-------------------------+
+ *                      +-----------------------+
+ *                      |
+ *                      v
+ *   +-------------------------------+   +------------------------+
+ *   | Target Entry                  |-->| Target Entry           |--> ...
+ *   | overlay_target_entry_t        |   | overlay_target_entry_t |
+ *   |                               |   +------------------------+
+ *   |                               |
+ *   | overlay_target_entry_flags_t -+--> Entry flags
+ *   | uint8_t[ETHERADDRL]        ---+--> Target MAC address
+ *   | overlay_target_point_t     ---+--> Target underlay address
+ *   | mblk_t *                   ---+--> outstanding mblk head
+ *   | mblk_t *                   ---+--> outstanding mblk tail
+ *   | size_t                     ---+--> outstanding mblk size
+ *   +-------------------------------+
+ *
+ * The primary entries that we care about are the overlay_dev_t, which
+ * correspond to each overlay device that is created with dladm(1M). Globally,
+ * these devices are maintained in a simple list_t which is protected with a
+ * lock.  Hence, these include important information such as the mac_handle_t
+ * and a datalink_id_t which is used to interact with the broader MAC and DLS
+ * ecosystem. We also maintain additional information such as the current state,
+ * outstanding operations, the mtu, and importantly, the plugin's private data.
+ * This is the instance of an encapsulation plugin that gets created as part of
+ * creating an overlay device. Another aspect of this is that the overlay_dev_t
+ * also includes information with respect to FMA. For more information, see the
+ * FMA section.
+ *
+ * Each overlay_dev_t has a pointer to a plugin, a mux, and a target. The plugin
+ * is the encapsulation plugin. This allows the device to make downcalls into it
+ * based on doing things like getting and setting properties. Otherwise, the
+ * plugin itself is a fairly straightforward entity. They are maintained in an
+ * (not pictured above) list. The plugins themselves mostly maintain things like
+ * the static list of properties, what kind of destination they require, and the
+ * operations vector. A given module may contain more if necessary.
+ *
+ * The next piece of the puzzle is the mux, or a multiplexor. The mux itself
+ * maintains a ksocket and it is through the mux that we send and receive
+ * message blocks. The mux represents a socket type and address, as well as a
+ * plugin. Multiple overlay_dev_t devices may then share the same mux. For
+ * example, consider the case where you have different instances of vxlan all on
+ * the same underlay network. These would all logically share the same IP
+ * address and port that packets are sent and received on; however, what differs
+ * is the decapuslation ID.
+ *
+ * Each mux maintains a ksocket_t which is similar to a socket(3SOCKET). Unlike
+ * a socket, we enable a direct callback on the ksocket. This means that
+ * whenever a message block chain is received, rather than sitting there and
+ * getting a callback in a context and kicking that back out to a taskq. Instead
+ * data comes into the callback function overlay_mux_recv().
+ *
+ * The mux is given encapsulated packets (via overlay_m_tx, the GLDv3 tx
+ * function) to transmit. It receives encapsulated packets, decapsulates them to
+ * determine the overlay identifier, looks up the given device that matches that
+ * identifier, and then causes the broader MAC world to receive the packet with
+ * a call to mac_rx().
+ *
+ * Today, we don't do too much that's special with the ksocket; however, as
+ * hardware is gaining understanding for these encapuslation protocols, we'll
+ * probably want to think of better ways to get those capabilities passed down
+ * and potentially better ways to program receive filters so they get directly
+ * to us. Though, that's all fantasy future land.
+ *
+ * The next part of the puzzle is the target cache. The purpose of the target
+ * cache is to cache where we should send a packet on the underlay network,
+ * given its mac address. The target cache operates in two modes depending on
+ * whether the lookup module was declared to OVERLAY_TARGET_POINT or
+ * OVERLAY_TARGET_DYANMIC.
+ *
+ * In the case where the target cache has been programmed to be
+ * OVERLAY_TARGET_POINT, then we only maintain a single overlay_target_point_t
+ * which has the destination that we send everything, no matter the destination
+ * mac address.
+ *
+ * On the other hand, when we have an instance of OVERLAY_TARGET_DYNAMIC, things
+ * are much more interesting and as a result, more complicated. We primarily
+ * store lists of overlay_target_entry_t's which are stored in both an avl tree
+ * and a refhash_t. The primary look up path uses the refhash_t and the avl tree
+ * is only used for a few of the target ioctls used to dump data such that we
+ * can get a consistent iteration order for things like dladm show-overlay -t.
+ * The key that we use for the reference hashtable is based on the mac address
+ * in the cache and currently we just do a simple CRC32 to transform it into a
+ * hash.
+ *
+ * Each entry maintains a set of flags to indicate the current status of the
+ * request. The flags may indicate one of three states: that current cache entry
+ * is valid, that the current cache entry has been directed to drop all output,
+ * and that the current cache entry is invalid and may be being looked up. In
+ * the case where it's valid, we just take the destination address and run with
+ * it.
+ *
+ * If it's invalid and a lookup has not been made, then we start the process
+ * that prepares a query that will make its way up to varpd. The cache entry
+ * entry maintains a message block chain of outstanding message blocks and a
+ * size. These lists are populated only when we don't know the answer as to
+ * where should these be sent. The size entry is used to cap the amount of
+ * outstanding data that we don't know the answer to. If we exceed a cap on the
+ * amount of outstanding data (currently 1 Mb), then we'll drop any additional
+ * packets. Once we get an answer indicating a valid destination, we transmit
+ * any outstanding data to that place. For the full story on how we look that up
+ * will be discussed in the section on the Target Cache Lifecycle.
+ *
+ * ------------------------
+ * FMA and Degraded Devices
+ * ------------------------
+ *
+ * Every kernel overlay device keeps track of its FMA state. Today in FMA we
+ * cannot represent partitions between resources nor can we represent that a
+ * given minor node of a psuedo device has failed -- if we degrade the overlay
+ * device, then the entire dev_info_t is degraded. However, we still want to be
+ * able to indicate to administrators that things may go wrong.
+ *
+ * To this end, we've added a notion of a degraded state to every overlay
+ * device. This state is primarily dictated by userland and it can happen for
+ * various reasons. Generally, because a userland lookup plugin has been
+ * partitioned, or something has gone wrong such that there is no longer any
+ * userland lookup module for a device, then we'll mark it degraded.
+ *
+ * As long as any of our minor instances is degraded, then we'll fire off the
+ * FMA event to note that. Once the last degraded instance is no longer
+ * degraded, then we'll end up telling FMA that we're all clean.
+ *
+ * To help administrators get a better sense of which of the various minor
+ * devices is wrong, we store the odd_fmamsg[] character array. This character
+ * array can be fetched with doing a dladm show-overlay -f.
+ *
+ * When an overlay device is degraded, we update the link status of a device to
+ * ensure that dladm and other tooling that is trying to check the link status
+ * note that it is down due to this error.
+ *
+ * -----------------------
+ * Target Cache Life Cycle
+ * -----------------------
+ *
+ * This section only applies when we have a lookup plugin of
+ * OVERLAY_TARGET_DYNAMIC. None of this applies to those of type
+ * OVERLAY_TARGET_POINT.
+ *
+ * While we got into the target cache in the general architecture section, it's
+ * worth going into more details as to how this actually works and showing some
+ * examples and state machines. Recall that a target cache entry basically has
+ * the following state transition diagram:
+ *
+ * Initial state
+ *    . . .           . . . first access       . . . varpd lookup enqueued
+ *        .           .                        .
+ *        .           .                        .
+ *     +-------+      .     +----------+       .
+ *     |  No   |------*---->| Invalid  |-------*----+
+ *     | Entry |            |  Entry   |            |
+ *     +-------+            +----------+            |
+ *                 varpd      ^      ^   varpd      |
+ *                 invalidate |      |   drop       |
+ *                      . . . *      * . .          v
+ *          +-------+         |      |         +---------+
+ *          | Entry |--->-----+      +----<----| Entry   |
+ *          | Valid |<----------*---------<----| Pending |->-+     varpd
+ *          +-------+           .              +---------+   * . . drop, but
+ *                              . varpd                ^     |     other queued
+ *                              . success              |     |     entries
+ *                                                     +-----+
+ *
+ * When the table is first created, it is empty. As we attempt to lookup entries
+ * and we find there is no entry at all, we'll create a new table entry for it.
+ * At that point the entry is technically in an invalid state, that means that
+ * we have no valid data from varpd. In that case, we'll go ahead and queue the
+ * packet into the entry's pending chain, and queue a varpd lookup, setting the
+ * OVERLAY_ENTRY_F_PENDING flag in the progress.
+ *
+ * If additional mblk_t's come in for this entry, we end up appending them to
+ * the tail of the chain, if and only if, we don't exceed the threshold for the
+ * amount of space they can take up. An entry remains pending until we get a
+ * varpd reply. If varpd replies with a valid results, we move to the valid
+ * entry state, and remove the OVERLAY_ENTRY_F_PENDING flag and set it with one
+ * of OVERLAY_ENTRY_F_VALID or OVERLAY_ENTRY_F_DROP as appropriate.
+ *
+ * Once an entry is valid, it stays valid until user land tells us to invalidate
+ * it with an ioctl or replace it, OVERLAY_TARG_CACHE_REMOE and
+ * OVERLAY_TARG_CACHE_SET respectively.
+ *
+ * If the lookup fails with a call to drop the packet, then the next state is
+ * determined by the state of the queue. If the set of outstanding entries is
+ * empty, then we just transition back to the invalid state. If instead, the
+ * set of outstanding entries is not empty, then we'll queue another entry and
+ * stay in the same state, repeating this until the number of requests is
+ * drained.
+ *
+ * The following images describes the flow of a given lookup and where the
+ * overlay_target_entry_t is at any given time.
+ *
+ *     +-------------------+
+ *     | Invalid Entry     |		An entry starts off as an invalid entry
+ *     | de:ad:be:ef:00:00 |		and only exists in the target cache.
+ *     +-------------------+
+ *
+ * 	~~~~
+ *
+ *     +---------------------+
+ *     | Global list_t       |		A mblk_t comes in for an entry. We
+ *     | overlay_target_list |		append it to the overlay_target_list.
+ *     +---------------------+
+ *                   |
+ *                   v
+ *             +-------------------+      +-------------------+
+ *             | Pending Entry     |----->| Pending Entry     |--->...
+ *             | 42:5e:1a:10:d6:2d |      | de:ad:be:ef:00:00 |
+ *             +-------------------+      +-------------------+
+ *
+ * 	~~~~
+ *
+ *     +--------------------------+
+ *     | /dev/overlay minor state |	User land said that it would look up an
+ *     | overlay_target_hdl_t     |	entry for us. We remove it from the
+ *     +--------------------------+	global list and add it to the handle's
+ *                  |			outstanding list.
+ *                  |
+ *                  v
+ *            +-------------------+      +-------------------+
+ *            | Pending Entry     |----->| Pending Entry     |
+ *            | 90:b8:d0:79:02:dd |      | de:ad:be:ef:00:00 |
+ *            +-------------------+      +-------------------+
+ *
+ * 	~~~~
+ *
+ *     +-------------------+
+ *     | Valid Entry       |		varpd returned an answer with
+ *     | de:ad:be:ef:00:00 |		OVERLAY_IOC_RESPOND and the target cache
+ *     | 10.169.23.42:4789 |		entry is now populated with a
+ *     +-------------------+		destination and marked as valid
+ *
+ *
+ * The lookup mechanism is performed via a series of operations on the character
+ * psuedo-device /dev/overlay. The only thing that uses this device is the
+ * userland daemon varpd. /dev/overlay is a cloneable device, each open of it
+ * granting a new minor number which maintains its own state. We maintain this
+ * state so that way if an outstanding lookup was queued to something that
+ * crashed or closed its handle without responding, we can know about this and
+ * thus handle it appropriately.
+ *
+ * When a lookup is first created it's added to our global list of outstanding
+ * lookups. To service requests, userland is required to perform an ioctl to ask
+ * for a request. We will block it in the kernel a set amount of time waiting
+ * for a request. When we give a request to a given minor instance of the
+ * device, we remove it from the global list and append the request to the
+ * device's list of outstanding entries, for the reasons we discussed above.
+ * When a lookup comes in, we give user land a smaller amount of information
+ * specific to that packet, the overlay_targ_lookup_t. It includes a request id
+ * to identify this, and then the overlay id, the varpd id, the header and
+ * packet size, the source and destination mac address, the SAP, and any
+ * potential VLAN header.
+ *
+ * At that point, it stays in that outstanding list until one of two ioctls are
+ * returned: OVERLAY_TARG_RESPOND or OVERLAY_TARG_DROP. During this time,
+ * userland may also perform other operations. For example, it may use
+ * OVERLAY_TARG_PKT to get a copy of this packet so it can perform more in-depth
+ * analysis of what to do beyond what we gave it initially. This is useful for
+ * providing proxy arp and the like. Finally, there are two other ioctls that
+ * varpd can then do. The first is OVERLAY_TARG_INJECT which injects the
+ * non-jumbo frame packet up into that mac device and OVERLAY_TARG_RESEND which
+ * causes us to encapsulate and send out the packet they've given us.
+ *
+ *
+ * Finally, through the target cache, several ioctls are provided to allow for
+ * interrogation and management of the cache. They allow for individual entries
+ * to be retrieved, set, or have the entire table flushed. For the full set of
+ * ioctls here and what they do, take a look at uts/common/sys/overlay_target.h.
+ *
+ * ------------------
+ * Sample Packet Flow
+ * ------------------
+ *
+ * There's a lot of pieces here, hopefully an example of how this all fits
+ * together will help clarify and elucidate what's going on. We're going to
+ * first track an outgoing packet, eg. one that is sent from an IP interface on
+ * a VNIC on top of an overlay device, and then we'll look at what it means to
+ * respond to that.
+ *
+ *
+ *    +----------------+        +--------------+            +------------------+
+ *    | IP/DLS send    |------->| MAC sends it |----------->| mblk_t reaches   |
+ *    | packet to MAC  |        | to the GLDv3 |            | overlay GLDv3 tx |
+ *    +----------------+        | VNIC device  |            | overlay_m_tx()   |
+ *                              +--------------+            +------------------+
+ *                                                                   |
+ *                             . lookup              . cache         |
+ *                             . drop                . miss          v
+ *            +---------+      .       +--------+    .      +------------------+
+ *            | freemsg |<-----*-------| varpd  |<---*------| Lookup each mblk |
+ *            | mblk_t  |              | lookup |           | in the target    |
+ *            +---------+              | queued |           | cache            |
+ *                ^                    +--------+           +------------------+
+ *      on send   |                        |                         |     cache
+ *      error . . *                        *. . lookup               * . . hit
+ *                |                        |    success              v
+ *                |                        |                +------------------+
+ *    +-----------------+                  +--------------->| call plugin      |
+ *    | Send out        |                                   | ovpo_encap() to  |
+ *    | overlay_mux_t's |<----------------------------------| get encap mblk_t |
+ *    | ksocket         |                                   +------------------+
+ *    +-----------------+
+ *
+ * The receive end point looks a little different and looks more like:
+ *
+ *  +------------------+     +----------------+    +-----------+
+ *  | mblk_t comes off |---->| enter netstack |--->| delivered |---+
+ *  | the physical     |     | IP stack       |    |     to    |   * . . direct
+ *  | device           |     +----------------+    |  ksocket  |   |   callback
+ *  +------------------+                           +-----------+   |
+ *                       . overlay id                              |
+ *                       . not found                               v
+ *       +-----------+   .      +-----------------+       +--------------------+
+ *       | freemsg   |<--*------| call plugin     |<------| overlay_mux_recv() |
+ *       | mblk_t    |          | ovpo_decap() to |       +--------------------+
+ *       +-----------+          | decap mblk_t    |
+ *                              +-----------------+
+ *                                     |
+ *                                     * . . overlay id
+ *                                     v     found
+ *                                 +--------+      +----------------+
+ *                                 | adjust |----->| call mac_rx    |
+ *                                 | mblk_t |      | on original    |
+ *                                 +--------+      | decaped packet |
+ *                                                 +----------------+
+ *
+ * ------------------
+ * Netstack Awareness
+ * ------------------
+ *
+ * In the above image we note that this enters a netstack. Today the only
+ * netstack that can be is the global zone as the overlay driver itself is not
+ * exactly netstack aware. What this really means is that varpd cannot run in a
+ * non-global zone and an overlay device cannot belong to a non-global zone.
+ * Non-global zones can still have a VNIC assigned to them that's been created
+ * over the overlay device the same way they would if it had been created over
+ * an etherstub or a physical device.
+ *
+ * The majority of the work to make it netstack aware is straightforward and the
+ * biggest thing is to create a netstack module that allows us to hook into
+ * netstack (and thus zone) creation and destruction.  From there, we need to
+ * amend the target cache lookup routines that we discussed earlier to not have
+ * a global outstanding list and a global list of handles, but rather, one per
+ * netstack.
+ *
+ * For the mux, we'll need to open the ksocket in the context of the zone, we
+ * can likely do this with a properly composed credential, but we'll need to do
+ * some more work on that path. Finally, we'll want to make sure the dld ioctls
+ * are aware of the zoneid of the caller and we use that appropriately and store
+ * it in the overlay_dev_t.
+ *
+ * -----------
+ * GLDv3 Notes
+ * -----------
+ *
+ * The overlay driver implements a GLDv3 device. Parts of GLDv3 are more
+ * relevant and other parts are much less relevant for us. For example, the
+ * GLDv3 is used to toggle the device being put into and out of promiscuous
+ * mode, to program MAC addresses for unicast and multicast hardware filters.
+ * Today, an overlay device doesn't have a notion of promiscuous mode nor does
+ * it have a notion of unicast and multicast addresses programmed into the
+ * device. Instead, for the purposes of the hardware filter, we don't do
+ * anything and just always accept new addresses being added and removed.
+ *
+ * If the GLDv3 start function has not been called, then we will not use this
+ * device for I/O purposes. Any calls to transmit or receive should be dropped,
+ * though the GLDv3 guarantees us that transmit will not be called without
+ * calling start. Similarly, once stop is called, then no packets can be dealt
+ * with.
+ *
+ * Today we don't support the stat interfaces, though there's no good reason
+ * that we shouldn't assemble some of the stats based on what we have in the
+ * future.
+ *
+ * When it comes to link properties, many of the traditional link properties do
+ * not apply and many others MAC handles for us. For example, we don't need to
+ * implement anything for overlay_m_getprop() to deal with returning the MTU, as
+ * MAC never calls into us for that. As such, there isn't much of anything to
+ * support in terms of properties.
+ *
+ * Today, we don't support any notion of hardware capabilities. However, if
+ * future NIC hardware or other changes to the system cause it to make sense for
+ * us to emulate logical groups, then we should do that. However, we still do
+ * implement a capab function so that we can identify ourselves as an overlay
+ * device to the broader MAC framework. This is done mostly so that a device
+ * created on top of us can have fanout rings as we don't try to lie about a
+ * speed for our device.
+ *
+ * The other question is what should be done for a device's MTU and margin. We
+ * set our minimum supported MTU to be the minimum value that an IP network may
+ * be set to 576 -- which mimics what an etherstub does. On the flip side, we
+ * have our upper bound set to 8900. This value comes from the fact that a lot
+ * of jumbo networks use their maximum as 9000. As such, we want to reserve 100
+ * bytes, which isn't exactly the most accurate number, but it'll be good enough
+ * for now. Because of that, our default MTU off of these devices is 1400, as
+ * the default MTU for everything is usually 1500 or whatever the underlying
+ * device is at; however, this is a bit simpler than asking the netstack what
+ * are all the IP interfaces at. It also calls into question how PMTU and PMTU
+ * discovery should work here. The challenge, especially for
+ * OVERLAY_TARG_DYNAMIC is that the MTU to any of the places will vary and it's
+ * not clear that if you have a single bad entry that the overall MTU should be
+ * lowered. Instead, we should figure out a better way of determining these
+ * kinds of PMTU errors and appropriately alerting the administrator via FMA.
+ *
+ * Regarding margin, we allow a margin of up to VLAN_TAGSZ depending on whether
+ * or not the underlying encapsulation device supports VLAN tags. If it does,
+ * then we'll set the margin to allow for it, otherwise, we will not.
+ */
+
+#include <sys/conf.h>
+#include <sys/errno.h>
+#include <sys/stat.h>
+#include <sys/ddi.h>
+#include <sys/sunddi.h>
+#include <sys/modctl.h>
+#include <sys/policy.h>
+#include <sys/stream.h>
+#include <sys/strsubr.h>
+#include <sys/strsun.h>
+#include <sys/types.h>
+#include <sys/kmem.h>
+#include <sys/param.h>
+#include <sys/sysmacros.h>
+#include <sys/ddifm.h>
+
+#include <sys/dls.h>
+#include <sys/dld_ioc.h>
+#include <sys/mac_provider.h>
+#include <sys/mac_client_priv.h>
+#include <sys/mac_ether.h>
+#include <sys/vlan.h>
+
+#include <sys/overlay_impl.h>
+
+dev_info_t *overlay_dip;
+static kmutex_t overlay_dev_lock;
+static list_t overlay_dev_list;
+static uint8_t overlay_macaddr[ETHERADDRL] =
+	{ 0x00, 0x00, 0x00, 0x00, 0x00, 0x00 };
+
+typedef enum overlay_dev_prop {
+	OVERLAY_DEV_P_MTU = 0,
+	OVERLAY_DEV_P_VNETID,
+	OVERLAY_DEV_P_ENCAP,
+	OVERLAY_DEV_P_VARPDID
+} overlay_dev_prop_t;
+
+#define	OVERLAY_DEV_NPROPS	4
+static const char *overlay_dev_props[] = {
+	"mtu",
+	"vnetid",
+	"encap",
+	"varpd/id"
+};
+
+#define	OVERLAY_MTU_MIN	576
+#define	OVERLAY_MTU_DEF	1400
+#define	OVERLAY_MTU_MAX	8900
+
+overlay_dev_t *
+overlay_hold_by_dlid(datalink_id_t id)
+{
+	overlay_dev_t *o;
+
+	mutex_enter(&overlay_dev_lock);
+	for (o = list_head(&overlay_dev_list); o != NULL;
+	    o = list_next(&overlay_dev_list, o)) {
+		if (id == o->odd_linkid) {
+			mutex_enter(&o->odd_lock);
+			o->odd_ref++;
+			mutex_exit(&o->odd_lock);
+			mutex_exit(&overlay_dev_lock);
+			return (o);
+		}
+	}
+
+	mutex_exit(&overlay_dev_lock);
+	return (NULL);
+}
+
+void
+overlay_hold_rele(overlay_dev_t *odd)
+{
+	mutex_enter(&odd->odd_lock);
+	ASSERT(odd->odd_ref > 0);
+	odd->odd_ref--;
+	mutex_exit(&odd->odd_lock);
+}
+
+void
+overlay_io_start(overlay_dev_t *odd, overlay_dev_flag_t flag)
+{
+	ASSERT(flag == OVERLAY_F_IN_RX || flag == OVERLAY_F_IN_TX);
+	ASSERT(MUTEX_HELD(&odd->odd_lock));
+
+	if (flag & OVERLAY_F_IN_RX)
+		odd->odd_rxcount++;
+	if (flag & OVERLAY_F_IN_TX)
+		odd->odd_txcount++;
+	odd->odd_flags |= flag;
+}
+
+void
+overlay_io_done(overlay_dev_t *odd, overlay_dev_flag_t flag)
+{
+	boolean_t signal = B_FALSE;
+
+	ASSERT(flag == OVERLAY_F_IN_RX || flag == OVERLAY_F_IN_TX);
+	ASSERT(MUTEX_HELD(&odd->odd_lock));
+
+	if (flag & OVERLAY_F_IN_RX) {
+		ASSERT(odd->odd_rxcount > 0);
+		odd->odd_rxcount--;
+		if (odd->odd_rxcount == 0) {
+			signal = B_TRUE;
+			odd->odd_flags &= ~OVERLAY_F_IN_RX;
+		}
+	}
+	if (flag & OVERLAY_F_IN_TX) {
+		ASSERT(odd->odd_txcount > 0);
+		odd->odd_txcount--;
+		if (odd->odd_txcount == 0) {
+			signal = B_TRUE;
+			odd->odd_flags &= ~OVERLAY_F_IN_TX;
+		}
+	}
+
+	if (signal == B_TRUE)
+		cv_broadcast(&odd->odd_iowait);
+}
+
+static void
+overlay_io_wait(overlay_dev_t *odd, overlay_dev_flag_t flag)
+{
+	ASSERT((flag & ~OVERLAY_F_IOMASK) == 0);
+	ASSERT(MUTEX_HELD(&odd->odd_lock));
+
+	while (odd->odd_flags & flag) {
+		cv_wait(&odd->odd_iowait, &odd->odd_lock);
+	}
+}
+
+void
+overlay_dev_iter(overlay_dev_iter_f func, void *arg)
+{
+	overlay_dev_t *odd;
+
+	mutex_enter(&overlay_dev_lock);
+	for (odd = list_head(&overlay_dev_list); odd != NULL;
+	    odd = list_next(&overlay_dev_list, odd)) {
+		if (func(odd, arg) != 0) {
+			mutex_exit(&overlay_dev_lock);
+			return;
+		}
+	}
+	mutex_exit(&overlay_dev_lock);
+}
+
+/* ARGSUSED */
+static int
+overlay_m_stat(void *arg, uint_t stat, uint64_t *val)
+{
+	return (ENOTSUP);
+}
+
+static int
+overlay_m_start(void *arg)
+{
+	overlay_dev_t *odd = arg;
+	overlay_mux_t *mux;
+	int ret, domain, family, prot;
+	struct sockaddr_storage storage;
+	socklen_t slen;
+
+	mutex_enter(&odd->odd_lock);
+	if ((odd->odd_flags & OVERLAY_F_ACTIVATED) == 0) {
+		mutex_exit(&odd->odd_lock);
+		return (EAGAIN);
+	}
+	mutex_exit(&odd->odd_lock);
+
+	ret = odd->odd_plugin->ovp_ops->ovpo_socket(odd->odd_pvoid, &domain,
+	    &family, &prot, (struct sockaddr *)&storage, &slen);
+	if (ret != 0)
+		return (ret);
+
+	mux = overlay_mux_open(odd->odd_plugin, domain, family, prot,
+	    (struct sockaddr *)&storage, slen, &ret);
+	if (mux == NULL)
+		return (ret);
+
+	overlay_mux_add_dev(mux, odd);
+	odd->odd_mux = mux;
+	mutex_enter(&odd->odd_lock);
+	ASSERT(!(odd->odd_flags & OVERLAY_F_IN_MUX));
+	odd->odd_flags |= OVERLAY_F_IN_MUX;
+	mutex_exit(&odd->odd_lock);
+
+	return (0);
+}
+
+static void
+overlay_m_stop(void *arg)
+{
+	overlay_dev_t *odd = arg;
+
+	/*
+	 * The MAC Perimeter is held here, so we don't have to worry about
+	 * synchornizing this with respect to metadata operations.
+	 */
+	mutex_enter(&odd->odd_lock);
+	VERIFY(odd->odd_flags & OVERLAY_F_IN_MUX);
+	VERIFY(!(odd->odd_flags & OVERLAY_F_MDDROP));
+	odd->odd_flags |= OVERLAY_F_MDDROP;
+	overlay_io_wait(odd, OVERLAY_F_IOMASK);
+	mutex_exit(&odd->odd_lock);
+
+	overlay_mux_remove_dev(odd->odd_mux, odd);
+	overlay_mux_close(odd->odd_mux);
+	odd->odd_mux = NULL;
+
+	mutex_enter(&odd->odd_lock);
+	odd->odd_flags &= ~OVERLAY_F_IN_MUX;
+	odd->odd_flags &= ~OVERLAY_F_MDDROP;
+	VERIFY((odd->odd_flags & OVERLAY_F_STOPMASK) == 0);
+	mutex_exit(&odd->odd_lock);
+}
+
+/*
+ * For more info on this, see the big theory statement.
+ */
+/* ARGSUSED */
+static int
+overlay_m_promisc(void *arg, boolean_t on)
+{
+	return (0);
+}
+
+/*
+ * For more info on this, see the big theory statement.
+ */
+/* ARGSUSED */
+static int
+overlay_m_multicast(void *arg, boolean_t add, const uint8_t *addrp)
+{
+	return (0);
+}
+
+/*
+ * For more info on this, see the big theory statement.
+ */
+/* ARGSUSED */
+static int
+overlay_m_unicast(void *arg, const uint8_t *macaddr)
+{
+	return (0);
+}
+
+mblk_t *
+overlay_m_tx(void *arg, mblk_t *mp_chain)
+{
+	overlay_dev_t *odd = arg;
+	mblk_t *mp, *ep;
+	int ret;
+	ovep_encap_info_t einfo;
+	struct msghdr hdr;
+
+	mutex_enter(&odd->odd_lock);
+	if ((odd->odd_flags & OVERLAY_F_MDDROP) ||
+	    !(odd->odd_flags & OVERLAY_F_IN_MUX)) {
+		mutex_exit(&odd->odd_lock);
+		freemsgchain(mp_chain);
+		return (NULL);
+	}
+	overlay_io_start(odd, OVERLAY_F_IN_TX);
+	mutex_exit(&odd->odd_lock);
+
+	bzero(&hdr, sizeof (struct msghdr));
+
+	bzero(&einfo, sizeof (ovep_encap_info_t));
+	einfo.ovdi_id = odd->odd_vid;
+	mp = mp_chain;
+	while (mp != NULL) {
+		socklen_t slen;
+		struct sockaddr_storage storage;
+
+		mp_chain = mp->b_next;
+		mp->b_next = NULL;
+		ep = NULL;
+
+		ret = overlay_target_lookup(odd, mp,
+		    (struct sockaddr *)&storage, &slen);
+		if (ret != OVERLAY_TARGET_OK) {
+			if (ret == OVERLAY_TARGET_DROP)
+				freemsg(mp);
+			mp = mp_chain;
+			continue;
+		}
+
+		hdr.msg_name = &storage;
+		hdr.msg_namelen = slen;
+
+		ret = odd->odd_plugin->ovp_ops->ovpo_encap(odd->odd_mh, mp,
+		    &einfo, &ep);
+		if (ret != 0 || ep == NULL) {
+			freemsg(mp);
+			goto out;
+		}
+
+		ep->b_cont = mp;
+		ret = overlay_mux_tx(odd->odd_mux, &hdr, ep);
+		if (ret != 0)
+			goto out;
+
+		mp = mp_chain;
+	}
+
+out:
+	mutex_enter(&odd->odd_lock);
+	overlay_io_done(odd, OVERLAY_F_IN_TX);
+	mutex_exit(&odd->odd_lock);
+	return (mp_chain);
+}
+
+/* ARGSUSED */
+static void
+overlay_m_ioctl(void *arg, queue_t *q, mblk_t *mp)
+{
+	miocnak(q, mp, 0, ENOTSUP);
+}
+
+/* ARGSUSED */
+static boolean_t
+overlay_m_getcapab(void *arg, mac_capab_t cap, void *cap_data)
+{
+	/*
+	 * Tell MAC we're an overlay.
+	 */
+	if (cap == MAC_CAPAB_OVERLAY)
+		return (B_TRUE);
+	return (B_FALSE);
+}
+
+/* ARGSUSED */
+static int
+overlay_m_setprop(void *arg, const char *pr_name, mac_prop_id_t pr_num,
+    uint_t pr_valsize, const void *pr_val)
+{
+	uint32_t mtu, old;
+	int err;
+	overlay_dev_t *odd = arg;
+
+	if (pr_num != MAC_PROP_MTU)
+		return (ENOTSUP);
+
+	bcopy(pr_val, &mtu, sizeof (mtu));
+	if (mtu < OVERLAY_MTU_MIN || mtu > OVERLAY_MTU_MAX)
+		return (EINVAL);
+
+	mutex_enter(&odd->odd_lock);
+	old = odd->odd_mtu;
+	odd->odd_mtu = mtu;
+	err = mac_maxsdu_update(odd->odd_mh, mtu);
+	if (err != 0)
+		odd->odd_mtu = old;
+	mutex_exit(&odd->odd_lock);
+
+	return (err);
+}
+
+/* ARGSUSED */
+static int
+overlay_m_getprop(void *arg, const char *pr_name, mac_prop_id_t pr_num,
+    uint_t pr_valsize, void *pr_val)
+{
+	return (ENOTSUP);
+}
+
+/* ARGSUSED */
+static void
+overlay_m_propinfo(void *arg, const char *pr_name, mac_prop_id_t pr_num,
+    mac_prop_info_handle_t prh)
+{
+	if (pr_num != MAC_PROP_MTU)
+		return;
+
+	mac_prop_info_set_default_uint32(prh, OVERLAY_MTU_DEF);
+	mac_prop_info_set_range_uint32(prh, OVERLAY_MTU_MIN, OVERLAY_MTU_MAX);
+}
+
+static mac_callbacks_t overlay_m_callbacks = {
+	.mc_callbacks = (MC_IOCTL | MC_GETCAPAB | MC_SETPROP | MC_GETPROP |
+	    MC_PROPINFO),
+	.mc_getstat = overlay_m_stat,
+	.mc_start = overlay_m_start,
+	.mc_stop = overlay_m_stop,
+	.mc_setpromisc = overlay_m_promisc,
+	.mc_multicst = overlay_m_multicast,
+	.mc_unicst = overlay_m_unicast,
+	.mc_tx = overlay_m_tx,
+	.mc_ioctl = overlay_m_ioctl,
+	.mc_getcapab = overlay_m_getcapab,
+	.mc_getprop = overlay_m_getprop,
+	.mc_setprop = overlay_m_setprop,
+	.mc_propinfo = overlay_m_propinfo
+};
+
+static boolean_t
+overlay_valid_name(const char *name, size_t buflen)
+{
+	size_t actlen;
+	int err, i;
+
+	for (i = 0; i < buflen; i++) {
+		if (name[i] == '\0')
+			break;
+	}
+
+	if (i == 0 || i == buflen)
+		return (B_FALSE);
+	actlen = i;
+	if (strchr(name, '/') != NULL)
+		return (B_FALSE);
+	if (u8_validate((char *)name, actlen, NULL,
+	    U8_VALIDATE_ENTIRE, &err) < 0)
+		return (B_FALSE);
+	return (B_TRUE);
+}
+
+/* ARGSUSED */
+static int
+overlay_i_create(void *karg, intptr_t arg, int mode, cred_t *cred, int *rvalp)
+{
+	int err;
+	uint64_t maxid;
+	overlay_dev_t *odd, *o;
+	mac_register_t *mac;
+	overlay_ioc_create_t *oicp = karg;
+
+	if (overlay_valid_name(oicp->oic_encap, MAXLINKNAMELEN) == B_FALSE)
+		return (EINVAL);
+
+	odd = kmem_zalloc(sizeof (overlay_dev_t), KM_SLEEP);
+	odd->odd_linkid = oicp->oic_linkid;
+	odd->odd_plugin = overlay_plugin_lookup(oicp->oic_encap);
+	if (odd->odd_plugin == NULL) {
+		kmem_free(odd, sizeof (overlay_dev_t));
+		return (ENOENT);
+	}
+	err = odd->odd_plugin->ovp_ops->ovpo_init((overlay_handle_t)odd,
+	    &odd->odd_pvoid);
+	if (err != 0) {
+		odd->odd_plugin->ovp_ops->ovpo_fini(odd->odd_pvoid);
+		overlay_plugin_rele(odd->odd_plugin);
+		kmem_free(odd, sizeof (overlay_dev_t));
+		return (EINVAL);
+	}
+
+	/*
+	 * Make sure that our virtual network id is valid for the given plugin
+	 * that we're working with.
+	 */
+	ASSERT(odd->odd_plugin->ovp_id_size <= 8);
+	maxid = UINT64_MAX;
+	if (odd->odd_plugin->ovp_id_size != 8)
+		maxid = (1ULL << (odd->odd_plugin->ovp_id_size * 8)) - 1ULL;
+	if (oicp->oic_vnetid > maxid) {
+		odd->odd_plugin->ovp_ops->ovpo_fini(odd->odd_pvoid);
+		overlay_plugin_rele(odd->odd_plugin);
+		kmem_free(odd, sizeof (overlay_dev_t));
+		return (EINVAL);
+	}
+	odd->odd_vid = oicp->oic_vnetid;
+
+	mac = mac_alloc(MAC_VERSION);
+	if (mac == NULL) {
+		mutex_exit(&overlay_dev_lock);
+		odd->odd_plugin->ovp_ops->ovpo_fini(odd->odd_pvoid);
+		overlay_plugin_rele(odd->odd_plugin);
+		kmem_free(odd, sizeof (overlay_dev_t));
+		return (EINVAL);
+	}
+
+	mac->m_type_ident = MAC_PLUGIN_IDENT_ETHER;
+	mac->m_driver = odd;
+	mac->m_dip = overlay_dip;
+	mac->m_dst_addr = NULL;
+	mac->m_callbacks = &overlay_m_callbacks;
+	mac->m_pdata = NULL;
+	mac->m_pdata_size = 0;
+
+	mac->m_priv_props = NULL;
+
+	/* Let mac handle this itself. */
+	mac->m_instance = (uint_t)-1;
+
+	/*
+	 * There is no real source address that should be used here, but saying
+	 * that we're not ethernet is going to cause its own problems. At the
+	 * end of the say, this is fine.
+	 */
+	mac->m_src_addr = overlay_macaddr;
+
+	/*
+	 * Start with the default MTU as the max SDU. If the MTU is changed, the
+	 * SDU will be changed to reflect that.
+	 */
+	mac->m_min_sdu = 1;
+	mac->m_max_sdu = OVERLAY_MTU_DEF;
+	mac->m_multicast_sdu = 0;
+
+	/*
+	 * The underlying device doesn't matter, instead this comes from the
+	 * encapsulation protocol and whether or not they allow VLAN tags.
+	 */
+	if (odd->odd_plugin->ovp_flags & OVEP_F_VLAN_TAG) {
+		mac->m_margin = VLAN_TAGSZ;
+	} else {
+		mac->m_margin = 0;
+	}
+
+	/*
+	 * Today, we have no MAC virtualization, it may make sense in the future
+	 * to go ahead and emulate some subset of this, but it doesn't today.
+	 */
+	mac->m_v12n = MAC_VIRT_NONE;
+
+	mutex_enter(&overlay_dev_lock);
+	for (o = list_head(&overlay_dev_list); o != NULL;
+	    o = list_next(&overlay_dev_list, o)) {
+		if (o->odd_linkid == oicp->oic_linkid) {
+			mutex_exit(&overlay_dev_lock);
+			odd->odd_plugin->ovp_ops->ovpo_fini(odd->odd_pvoid);
+			overlay_plugin_rele(odd->odd_plugin);
+			kmem_free(odd, sizeof (overlay_dev_t));
+			return (EEXIST);
+		}
+
+		if (o->odd_vid == oicp->oic_vnetid &&
+		    o->odd_plugin == odd->odd_plugin) {
+			mutex_exit(&overlay_dev_lock);
+			odd->odd_plugin->ovp_ops->ovpo_fini(odd->odd_pvoid);
+			overlay_plugin_rele(odd->odd_plugin);
+			kmem_free(odd, sizeof (overlay_dev_t));
+			return (EEXIST);
+		}
+	}
+
+	err = mac_register(mac, &odd->odd_mh);
+	mac_free(mac);
+	if (err != 0) {
+		mutex_exit(&overlay_dev_lock);
+		odd->odd_plugin->ovp_ops->ovpo_fini(odd->odd_pvoid);
+		overlay_plugin_rele(odd->odd_plugin);
+		kmem_free(odd, sizeof (overlay_dev_t));
+		return (err);
+	}
+
+	err = dls_devnet_create(odd->odd_mh, odd->odd_linkid,
+	    crgetzoneid(cred));
+	if (err != 0) {
+		mutex_exit(&overlay_dev_lock);
+		(void) mac_unregister(odd->odd_mh);
+		odd->odd_plugin->ovp_ops->ovpo_fini(odd->odd_pvoid);
+		overlay_plugin_rele(odd->odd_plugin);
+		kmem_free(odd, sizeof (overlay_dev_t));
+		return (err);
+	}
+
+	mutex_init(&odd->odd_lock, NULL, MUTEX_DRIVER, NULL);
+	cv_init(&odd->odd_iowait, NULL, CV_DRIVER, NULL);
+	odd->odd_ref = 0;
+	odd->odd_flags = 0;
+	list_insert_tail(&overlay_dev_list, odd);
+	mutex_exit(&overlay_dev_lock);
+
+	return (0);
+}
+
+/* ARGSUSED */
+static int
+overlay_i_activate(void *karg, intptr_t arg, int mode, cred_t *cred, int *rvalp)
+{
+	int i, ret;
+	overlay_dev_t *odd;
+	mac_perim_handle_t mph;
+	overlay_ioc_activate_t *oiap = karg;
+	overlay_ioc_propinfo_t *infop;
+	overlay_ioc_prop_t *oip;
+	overlay_prop_handle_t phdl;
+
+	odd = overlay_hold_by_dlid(oiap->oia_linkid);
+	if (odd == NULL)
+		return (ENOENT);
+
+	infop = kmem_alloc(sizeof (overlay_ioc_propinfo_t), KM_SLEEP);
+	oip = kmem_alloc(sizeof (overlay_ioc_prop_t), KM_SLEEP);
+	phdl = (overlay_prop_handle_t)infop;
+
+	mac_perim_enter_by_mh(odd->odd_mh, &mph);
+	mutex_enter(&odd->odd_lock);
+	if (odd->odd_flags & OVERLAY_F_ACTIVATED) {
+		mutex_exit(&odd->odd_lock);
+		mac_perim_exit(mph);
+		overlay_hold_rele(odd);
+		kmem_free(infop, sizeof (overlay_ioc_propinfo_t));
+		kmem_free(oip, sizeof (overlay_ioc_prop_t));
+		return (EEXIST);
+	}
+	mutex_exit(&odd->odd_lock);
+
+	for (i = 0; i < odd->odd_plugin->ovp_nprops; i++) {
+		const char *pname = odd->odd_plugin->ovp_props[i];
+		bzero(infop, sizeof (overlay_ioc_propinfo_t));
+		overlay_prop_init(phdl);
+		ret = odd->odd_plugin->ovp_ops->ovpo_propinfo(pname, phdl);
+		if (ret != 0) {
+			mac_perim_exit(mph);
+			overlay_hold_rele(odd);
+			kmem_free(infop, sizeof (overlay_ioc_propinfo_t));
+			kmem_free(oip, sizeof (overlay_ioc_prop_t));
+			return (ret);
+		}
+
+		if ((infop->oipi_prot & OVERLAY_PROP_PERM_REQ) == 0)
+			continue;
+		bzero(oip, sizeof (overlay_ioc_prop_t));
+		oip->oip_size = sizeof (oip->oip_value);
+		ret = odd->odd_plugin->ovp_ops->ovpo_getprop(odd->odd_pvoid,
+		    pname, oip->oip_value, &oip->oip_size);
+		if (ret != 0) {
+			mac_perim_exit(mph);
+			overlay_hold_rele(odd);
+			kmem_free(infop, sizeof (overlay_ioc_propinfo_t));
+			kmem_free(oip, sizeof (overlay_ioc_prop_t));
+			return (ret);
+		}
+		if (oip->oip_size == 0) {
+			mac_perim_exit(mph);
+			overlay_hold_rele(odd);
+			kmem_free(infop, sizeof (overlay_ioc_propinfo_t));
+			kmem_free(oip, sizeof (overlay_ioc_prop_t));
+			return (EINVAL);
+		}
+	}
+
+	mutex_enter(&odd->odd_lock);
+	if ((odd->odd_flags & OVERLAY_F_VARPD) == 0) {
+		mutex_exit(&odd->odd_lock);
+		mac_perim_exit(mph);
+		overlay_hold_rele(odd);
+		kmem_free(infop, sizeof (overlay_ioc_propinfo_t));
+		kmem_free(oip, sizeof (overlay_ioc_prop_t));
+		return (ENXIO);
+	}
+
+	ASSERT((odd->odd_flags & OVERLAY_F_ACTIVATED) == 0);
+	odd->odd_flags |= OVERLAY_F_ACTIVATED;
+	mutex_exit(&odd->odd_lock);
+
+	mac_perim_exit(mph);
+	overlay_hold_rele(odd);
+	kmem_free(infop, sizeof (overlay_ioc_propinfo_t));
+	kmem_free(oip, sizeof (overlay_ioc_prop_t));
+
+	return (0);
+}
+
+/* ARGSUSED */
+static int
+overlay_i_delete(void *karg, intptr_t arg, int mode, cred_t *cred, int *rvalp)
+{
+	overlay_ioc_delete_t *oidp = karg;
+	overlay_dev_t *odd;
+	datalink_id_t tid;
+	int ret;
+
+	odd = overlay_hold_by_dlid(oidp->oid_linkid);
+	if (odd == NULL) {
+		return (ENOENT);
+	}
+
+	mutex_enter(&odd->odd_lock);
+	/* If we're not the only hold, we're busy */
+	if (odd->odd_ref != 1) {
+		mutex_exit(&odd->odd_lock);
+		overlay_hold_rele(odd);
+		return (EBUSY);
+	}
+
+	if (odd->odd_flags & OVERLAY_F_IN_MUX) {
+		mutex_exit(&odd->odd_lock);
+		overlay_hold_rele(odd);
+		return (EBUSY);
+	}
+
+	/*
+	 * To remove this, we need to first remove it from dls and then remove
+	 * it from mac. The act of removing it from mac will check if there are
+	 * devices on top of this, eg. vnics. If there are, then that will fail
+	 * and we'll have to go through and recreate the dls entry. Only after
+	 * mac_unregister has succeeded, then we'll go through and actually free
+	 * everything and drop the dev lock.
+	 */
+	ret = dls_devnet_destroy(odd->odd_mh, &tid, B_TRUE);
+	if (ret != 0) {
+		overlay_hold_rele(odd);
+		return (ret);
+	}
+
+	ASSERT(oidp->oid_linkid == tid);
+	ret = mac_disable(odd->odd_mh);
+	if (ret != 0) {
+		(void) dls_devnet_create(odd->odd_mh, odd->odd_linkid,
+		    crgetzoneid(cred));
+		overlay_hold_rele(odd);
+		return (ret);
+	}
+
+	overlay_target_quiesce(odd->odd_target);
+
+	mutex_enter(&overlay_dev_lock);
+	list_remove(&overlay_dev_list, odd);
+	mutex_exit(&overlay_dev_lock);
+
+	cv_destroy(&odd->odd_iowait);
+	mutex_destroy(&odd->odd_lock);
+	overlay_target_free(odd);
+	odd->odd_plugin->ovp_ops->ovpo_fini(odd->odd_pvoid);
+	overlay_plugin_rele(odd->odd_plugin);
+	kmem_free(odd, sizeof (overlay_dev_t));
+
+	return (0);
+}
+
+/* ARGSUSED */
+static int
+overlay_i_nprops(void *karg, intptr_t arg, int mode, cred_t *cred,
+    int *rvalp)
+{
+	overlay_dev_t *odd;
+	overlay_ioc_nprops_t *on = karg;
+
+	odd = overlay_hold_by_dlid(on->oipn_linkid);
+	if (odd == NULL)
+		return (ENOENT);
+	on->oipn_nprops = odd->odd_plugin->ovp_nprops + OVERLAY_DEV_NPROPS;
+	overlay_hold_rele(odd);
+
+	return (0);
+}
+
+static int
+overlay_propinfo_plugin_cb(overlay_plugin_t *opp, void *arg)
+{
+	overlay_prop_handle_t phdl = arg;
+	overlay_prop_set_range_str(phdl, opp->ovp_name);
+	return (0);
+}
+
+static int
+overlay_i_name_to_propid(overlay_dev_t *odd, const char *name, uint_t *id)
+{
+	int i;
+
+	for (i = 0; i < OVERLAY_DEV_NPROPS; i++) {
+		if (strcmp(overlay_dev_props[i], name) == 0) {
+			*id = i;
+			return (0);
+		}
+	}
+
+	for (i = 0; i < odd->odd_plugin->ovp_nprops; i++) {
+		if (strcmp(odd->odd_plugin->ovp_props[i], name) == 0) {
+			*id = i + OVERLAY_DEV_NPROPS;
+			return (0);
+		}
+	}
+
+	return (ENOENT);
+}
+
+static void
+overlay_i_propinfo_mtu(overlay_dev_t *odd, overlay_prop_handle_t phdl)
+{
+	uint32_t def;
+	mac_propval_range_t range;
+	uint_t perm;
+
+	ASSERT(MAC_PERIM_HELD(odd->odd_mh));
+
+	bzero(&range, sizeof (mac_propval_range_t));
+	range.mpr_count = 1;
+	if (mac_prop_info(odd->odd_mh, MAC_PROP_MTU, "mtu", &def,
+	    sizeof (def), &range, &perm) != 0)
+		return;
+
+	if (perm == MAC_PROP_PERM_READ)
+		overlay_prop_set_prot(phdl, OVERLAY_PROP_PERM_READ);
+	else if (perm == MAC_PROP_PERM_WRITE)
+		overlay_prop_set_prot(phdl, OVERLAY_PROP_PERM_WRITE);
+	else if (perm == MAC_PROP_PERM_RW)
+		overlay_prop_set_prot(phdl, OVERLAY_PROP_PERM_RW);
+
+	overlay_prop_set_type(phdl, OVERLAY_PROP_T_UINT);
+	overlay_prop_set_default(phdl, &def, sizeof (def));
+	overlay_prop_set_range_uint32(phdl, range.mpr_range_uint32[0].mpur_min,
+	    range.mpr_range_uint32[0].mpur_max);
+}
+
+/* ARGSUSED */
+static int
+overlay_i_propinfo(void *karg, intptr_t arg, int mode, cred_t *cred,
+    int *rvalp)
+{
+	overlay_dev_t *odd;
+	int ret;
+	mac_perim_handle_t mph;
+	uint_t propid = UINT_MAX;
+	overlay_ioc_propinfo_t *oip = karg;
+	overlay_prop_handle_t phdl = (overlay_prop_handle_t)oip;
+
+	odd = overlay_hold_by_dlid(oip->oipi_linkid);
+	if (odd == NULL)
+		return (ENOENT);
+
+	overlay_prop_init(phdl);
+	mac_perim_enter_by_mh(odd->odd_mh, &mph);
+
+	/*
+	 * If the id is -1, then the property that we're looking for is named in
+	 * oipi_name and we should fill in its id. Otherwise, we've been given
+	 * an id and we need to turn that into a name for our plugin's sake. The
+	 * id is our own fabrication for property discovery.
+	 */
+	if (oip->oipi_id == -1) {
+		/*
+		 * Determine if it's a known generic property or it belongs to a
+		 * module by checking against the list of known names.
+		 */
+		oip->oipi_name[OVERLAY_PROP_NAMELEN-1] = '\0';
+		if ((ret = overlay_i_name_to_propid(odd, oip->oipi_name,
+		    &propid)) != 0) {
+			overlay_hold_rele(odd);
+			mac_perim_exit(mph);
+			return (ret);
+		}
+		oip->oipi_id = propid;
+		if (propid >= OVERLAY_DEV_NPROPS) {
+			ret = odd->odd_plugin->ovp_ops->ovpo_propinfo(
+			    oip->oipi_name, phdl);
+			overlay_hold_rele(odd);
+			mac_perim_exit(mph);
+			return (ret);
+
+		}
+	} else if (oip->oipi_id >= OVERLAY_DEV_NPROPS) {
+		uint_t id = oip->oipi_id - OVERLAY_DEV_NPROPS;
+
+		if (id >= odd->odd_plugin->ovp_nprops) {
+			overlay_hold_rele(odd);
+			mac_perim_exit(mph);
+			return (EINVAL);
+		}
+		ret = odd->odd_plugin->ovp_ops->ovpo_propinfo(
+		    odd->odd_plugin->ovp_props[id], phdl);
+		overlay_hold_rele(odd);
+		mac_perim_exit(mph);
+		return (ret);
+	} else if (oip->oipi_id < -1) {
+		overlay_hold_rele(odd);
+		mac_perim_exit(mph);
+		return (EINVAL);
+	} else {
+		ASSERT(oip->oipi_id < OVERLAY_DEV_NPROPS);
+		ASSERT(oip->oipi_id >= 0);
+		propid = oip->oipi_id;
+		(void) strlcpy(oip->oipi_name, overlay_dev_props[propid],
+		    sizeof (oip->oipi_name));
+	}
+
+	switch (propid) {
+	case OVERLAY_DEV_P_MTU:
+		overlay_i_propinfo_mtu(odd, phdl);
+		break;
+	case OVERLAY_DEV_P_VNETID:
+		overlay_prop_set_prot(phdl, OVERLAY_PROP_PERM_RW);
+		overlay_prop_set_type(phdl, OVERLAY_PROP_T_UINT);
+		overlay_prop_set_nodefault(phdl);
+		break;
+	case OVERLAY_DEV_P_ENCAP:
+		overlay_prop_set_prot(phdl, OVERLAY_PROP_PERM_READ);
+		overlay_prop_set_type(phdl, OVERLAY_PROP_T_STRING);
+		overlay_prop_set_nodefault(phdl);
+		overlay_plugin_walk(overlay_propinfo_plugin_cb, phdl);
+		break;
+	case OVERLAY_DEV_P_VARPDID:
+		overlay_prop_set_prot(phdl, OVERLAY_PROP_PERM_READ);
+		overlay_prop_set_type(phdl, OVERLAY_PROP_T_UINT);
+		overlay_prop_set_nodefault(phdl);
+		break;
+	default:
+		overlay_hold_rele(odd);
+		mac_perim_exit(mph);
+		return (ENOENT);
+	}
+
+	overlay_hold_rele(odd);
+	mac_perim_exit(mph);
+	return (0);
+}
+
+/* ARGSUSED */
+static int
+overlay_i_getprop(void *karg, intptr_t arg, int mode, cred_t *cred,
+    int *rvalp)
+{
+	int ret;
+	overlay_dev_t *odd;
+	mac_perim_handle_t mph;
+	overlay_ioc_prop_t *oip = karg;
+	uint_t propid, mtu;
+
+	odd = overlay_hold_by_dlid(oip->oip_linkid);
+	if (odd == NULL)
+		return (ENOENT);
+
+	mac_perim_enter_by_mh(odd->odd_mh, &mph);
+	oip->oip_size = OVERLAY_PROP_SIZEMAX;
+	oip->oip_name[OVERLAY_PROP_NAMELEN-1] = '\0';
+	if (oip->oip_id == -1) {
+		int i;
+
+		for (i = 0; i < OVERLAY_DEV_NPROPS; i++) {
+			if (strcmp(overlay_dev_props[i], oip->oip_name) == 0)
+				break;
+			if (i == OVERLAY_DEV_NPROPS) {
+				ret = odd->odd_plugin->ovp_ops->ovpo_getprop(
+				    odd->odd_pvoid, oip->oip_name,
+				    oip->oip_value, &oip->oip_size);
+				overlay_hold_rele(odd);
+				mac_perim_exit(mph);
+				return (ret);
+			}
+		}
+
+		propid = i;
+	} else if (oip->oip_id >= OVERLAY_DEV_NPROPS) {
+		uint_t id = oip->oip_id - OVERLAY_DEV_NPROPS;
+
+		if (id > odd->odd_plugin->ovp_nprops) {
+			overlay_hold_rele(odd);
+			mac_perim_exit(mph);
+			return (EINVAL);
+		}
+		ret = odd->odd_plugin->ovp_ops->ovpo_getprop(odd->odd_pvoid,
+		    odd->odd_plugin->ovp_props[id], oip->oip_value,
+		    &oip->oip_size);
+		overlay_hold_rele(odd);
+		mac_perim_exit(mph);
+		return (ret);
+	} else if (oip->oip_id < -1) {
+		overlay_hold_rele(odd);
+		mac_perim_exit(mph);
+		return (EINVAL);
+	} else {
+		ASSERT(oip->oip_id < OVERLAY_DEV_NPROPS);
+		ASSERT(oip->oip_id >= 0);
+		propid = oip->oip_id;
+	}
+
+	ret = 0;
+	switch (propid) {
+	case OVERLAY_DEV_P_MTU:
+		/*
+		 * The MTU is always set and retrieved through MAC, to allow for
+		 * MAC to do whatever it wants, as really that property belongs
+		 * to MAC. This is important for things where vnics have hold on
+		 * the MTU.
+		 */
+		mac_sdu_get(odd->odd_mh, NULL, &mtu);
+		bcopy(&mtu, oip->oip_value, sizeof (uint_t));
+		oip->oip_size = sizeof (uint_t);
+		break;
+	case OVERLAY_DEV_P_VNETID:
+		/*
+		 * While it's read-only while inside of a mux, we're not in a
+		 * context that can guarantee that. Therefore we always grab the
+		 * overlay_dev_t's odd_lock.
+		 */
+		mutex_enter(&odd->odd_lock);
+		bcopy(&odd->odd_vid, oip->oip_value, sizeof (uint64_t));
+		mutex_exit(&odd->odd_lock);
+		oip->oip_size = sizeof (uint64_t);
+		break;
+	case OVERLAY_DEV_P_ENCAP:
+		oip->oip_size = strlcpy((char *)oip->oip_value,
+		    odd->odd_plugin->ovp_name, oip->oip_size);
+		break;
+	case OVERLAY_DEV_P_VARPDID:
+		mutex_enter(&odd->odd_lock);
+		if (odd->odd_flags & OVERLAY_F_VARPD) {
+			const uint64_t val = odd->odd_target->ott_id;
+			bcopy(&val, oip->oip_value, sizeof (uint64_t));
+			oip->oip_size = sizeof (uint64_t);
+		} else {
+			oip->oip_size = 0;
+		}
+		mutex_exit(&odd->odd_lock);
+		break;
+	default:
+		ret = ENOENT;
+	}
+
+	overlay_hold_rele(odd);
+	mac_perim_exit(mph);
+	return (ret);
+}
+
+static void
+overlay_setprop_vnetid(overlay_dev_t *odd, uint64_t vnetid)
+{
+	mutex_enter(&odd->odd_lock);
+
+	/* Simple case, not active */
+	if (!(odd->odd_flags & OVERLAY_F_IN_MUX)) {
+		odd->odd_vid = vnetid;
+		mutex_exit(&odd->odd_lock);
+		return;
+	}
+
+	/*
+	 * In the hard case, we need to set the drop flag, quiesce I/O and then
+	 * we can go ahead and do everything.
+	 */
+	odd->odd_flags |= OVERLAY_F_MDDROP;
+	overlay_io_wait(odd, OVERLAY_F_IOMASK);
+	mutex_exit(&odd->odd_lock);
+
+	overlay_mux_remove_dev(odd->odd_mux, odd);
+	mutex_enter(&odd->odd_lock);
+	odd->odd_vid = vnetid;
+	mutex_exit(&odd->odd_lock);
+	overlay_mux_add_dev(odd->odd_mux, odd);
+
+	mutex_enter(&odd->odd_lock);
+	ASSERT(odd->odd_flags & OVERLAY_F_IN_MUX);
+	odd->odd_flags &= ~OVERLAY_F_IN_MUX;
+	mutex_exit(&odd->odd_lock);
+}
+
+/* ARGSUSED */
+static int
+overlay_i_setprop(void *karg, intptr_t arg, int mode, cred_t *cred,
+    int *rvalp)
+{
+	int ret;
+	overlay_dev_t *odd;
+	overlay_ioc_prop_t *oip = karg;
+	uint_t propid = UINT_MAX;
+	mac_perim_handle_t mph;
+	uint64_t maxid, *vidp;
+
+	if (oip->oip_size > OVERLAY_PROP_SIZEMAX)
+		return (EINVAL);
+
+	odd = overlay_hold_by_dlid(oip->oip_linkid);
+	if (odd == NULL)
+		return (ENOENT);
+
+	oip->oip_name[OVERLAY_PROP_NAMELEN-1] = '\0';
+	mac_perim_enter_by_mh(odd->odd_mh, &mph);
+	mutex_enter(&odd->odd_lock);
+	if (odd->odd_flags & OVERLAY_F_ACTIVATED) {
+		mac_perim_exit(mph);
+		mutex_exit(&odd->odd_lock);
+		return (ENOTSUP);
+	}
+	mutex_exit(&odd->odd_lock);
+	if (oip->oip_id == -1) {
+		int i;
+
+		for (i = 0; i < OVERLAY_DEV_NPROPS; i++) {
+			if (strcmp(overlay_dev_props[i], oip->oip_name) == 0)
+				break;
+			if (i == OVERLAY_DEV_NPROPS) {
+				ret = odd->odd_plugin->ovp_ops->ovpo_setprop(
+				    odd->odd_pvoid, oip->oip_name,
+				    oip->oip_value, oip->oip_size);
+				overlay_hold_rele(odd);
+				mac_perim_exit(mph);
+				return (ret);
+			}
+		}
+
+		propid = i;
+	} else if (oip->oip_id >= OVERLAY_DEV_NPROPS) {
+		uint_t id = oip->oip_id - OVERLAY_DEV_NPROPS;
+
+		if (id > odd->odd_plugin->ovp_nprops) {
+			mac_perim_exit(mph);
+			overlay_hold_rele(odd);
+			return (EINVAL);
+		}
+		ret = odd->odd_plugin->ovp_ops->ovpo_setprop(odd->odd_pvoid,
+		    odd->odd_plugin->ovp_props[id], oip->oip_value,
+		    oip->oip_size);
+		mac_perim_exit(mph);
+		overlay_hold_rele(odd);
+		return (ret);
+	} else if (oip->oip_id < -1) {
+		mac_perim_exit(mph);
+		overlay_hold_rele(odd);
+		return (EINVAL);
+	} else {
+		ASSERT(oip->oip_id < OVERLAY_DEV_NPROPS);
+		ASSERT(oip->oip_id >= 0);
+		propid = oip->oip_id;
+	}
+
+	ret = 0;
+	switch (propid) {
+	case OVERLAY_DEV_P_MTU:
+		ret = mac_set_prop(odd->odd_mh, MAC_PROP_MTU, "mtu",
+		    oip->oip_value, oip->oip_size);
+		break;
+	case OVERLAY_DEV_P_VNETID:
+		if (oip->oip_size != sizeof (uint64_t)) {
+			ret = EINVAL;
+			break;
+		}
+		vidp = (uint64_t *)oip->oip_value;
+		ASSERT(odd->odd_plugin->ovp_id_size <= 8);
+		maxid = UINT64_MAX;
+		if (odd->odd_plugin->ovp_id_size != 8)
+			maxid = (1ULL << (odd->odd_plugin->ovp_id_size * 8)) -
+			    1ULL;
+		if (*vidp >= maxid) {
+			ret = EINVAL;
+			break;
+		}
+		overlay_setprop_vnetid(odd, *vidp);
+		break;
+	case OVERLAY_DEV_P_ENCAP:
+	case OVERLAY_DEV_P_VARPDID:
+		ret = EPERM;
+		break;
+	default:
+		ret = ENOENT;
+	}
+
+	mac_perim_exit(mph);
+	overlay_hold_rele(odd);
+	return (ret);
+}
+
+/* ARGSUSED */
+static int
+overlay_i_status(void *karg, intptr_t arg, int mode, cred_t *cred,
+    int *rvalp)
+{
+	overlay_dev_t *odd;
+	overlay_ioc_status_t *os = karg;
+
+	odd = overlay_hold_by_dlid(os->ois_linkid);
+	if (odd == NULL)
+		return (ENOENT);
+
+	mutex_enter(&odd->odd_lock);
+	if ((odd->odd_flags & OVERLAY_F_DEGRADED) != 0) {
+		os->ois_status = OVERLAY_I_DEGRADED;
+		if (odd->odd_fmamsg != NULL) {
+			(void) strlcpy(os->ois_message, odd->odd_fmamsg,
+			    OVERLAY_STATUS_BUFLEN);
+		} else {
+			os->ois_message[0] = '\0';
+		}
+
+	} else {
+		os->ois_status = OVERLAY_I_OK;
+		os->ois_message[0] = '\0';
+	}
+	mutex_exit(&odd->odd_lock);
+	overlay_hold_rele(odd);
+
+	return (0);
+}
+
+static dld_ioc_info_t overlay_ioc_list[] = {
+	{ OVERLAY_IOC_CREATE, DLDCOPYIN, sizeof (overlay_ioc_create_t),
+		overlay_i_create, secpolicy_dl_config },
+	{ OVERLAY_IOC_ACTIVATE, DLDCOPYIN, sizeof (overlay_ioc_activate_t),
+		overlay_i_activate, secpolicy_dl_config },
+	{ OVERLAY_IOC_DELETE, DLDCOPYIN, sizeof (overlay_ioc_delete_t),
+		overlay_i_delete, secpolicy_dl_config },
+	{ OVERLAY_IOC_PROPINFO, DLDCOPYIN | DLDCOPYOUT,
+		sizeof (overlay_ioc_propinfo_t), overlay_i_propinfo,
+		secpolicy_dl_config },
+	{ OVERLAY_IOC_GETPROP, DLDCOPYIN | DLDCOPYOUT,
+		sizeof (overlay_ioc_prop_t), overlay_i_getprop,
+		secpolicy_dl_config },
+	{ OVERLAY_IOC_SETPROP, DLDCOPYIN,
+		sizeof (overlay_ioc_prop_t), overlay_i_setprop,
+		secpolicy_dl_config },
+	{ OVERLAY_IOC_NPROPS, DLDCOPYIN | DLDCOPYOUT,
+		sizeof (overlay_ioc_nprops_t), overlay_i_nprops,
+		secpolicy_dl_config },
+	{ OVERLAY_IOC_STATUS, DLDCOPYIN | DLDCOPYOUT,
+		sizeof (overlay_ioc_status_t), overlay_i_status,
+		NULL }
+};
+
+static int
+overlay_attach(dev_info_t *dip, ddi_attach_cmd_t cmd)
+{
+	int fmcap = DDI_FM_EREPORT_CAPABLE;
+	if (cmd != DDI_ATTACH)
+		return (DDI_FAILURE);
+
+	if (overlay_dip != NULL || ddi_get_instance(dip) != 0)
+		return (DDI_FAILURE);
+
+	ddi_fm_init(dip, &fmcap, NULL);
+
+	if (ddi_create_minor_node(dip, OVERLAY_CTL, S_IFCHR,
+	    ddi_get_instance(dip), DDI_PSEUDO, 0) == DDI_FAILURE)
+		return (DDI_FAILURE);
+
+	if (dld_ioc_register(OVERLAY_IOC, overlay_ioc_list,
+	    DLDIOCCNT(overlay_ioc_list)) != 0) {
+		ddi_remove_minor_node(dip, OVERLAY_CTL);
+		return (DDI_FAILURE);
+	}
+
+	overlay_dip = dip;
+	return (DDI_SUCCESS);
+}
+
+/* ARGSUSED */
+static int
+overlay_getinfo(dev_info_t *dip, ddi_info_cmd_t cmd, void *arg, void **resp)
+{
+	int error;
+
+	switch (cmd) {
+	case DDI_INFO_DEVT2DEVINFO:
+		*resp = (void *)overlay_dip;
+		error = DDI_SUCCESS;
+		break;
+	case DDI_INFO_DEVT2INSTANCE:
+		*resp = (void *)0;
+		error = DDI_SUCCESS;
+		break;
+	default:
+		error = DDI_FAILURE;
+		break;
+	}
+
+	return (error);
+}
+
+static int
+overlay_detach(dev_info_t *dip, ddi_detach_cmd_t cmd)
+{
+	if (cmd != DDI_DETACH)
+		return (DDI_FAILURE);
+
+	mutex_enter(&overlay_dev_lock);
+	if (!list_is_empty(&overlay_dev_list) || overlay_target_busy()) {
+		mutex_exit(&overlay_dev_lock);
+		return (EBUSY);
+	}
+	mutex_exit(&overlay_dev_lock);
+
+
+	dld_ioc_unregister(OVERLAY_IOC);
+	ddi_remove_minor_node(dip, OVERLAY_CTL);
+	ddi_fm_fini(dip);
+	overlay_dip = NULL;
+	return (DDI_SUCCESS);
+}
+
+static struct cb_ops overlay_cbops = {
+	overlay_target_open,	/* cb_open */
+	overlay_target_close,	/* cb_close */
+	nodev,			/* cb_strategy */
+	nodev,			/* cb_print */
+	nodev,			/* cb_dump */
+	nodev,			/* cb_read */
+	nodev,			/* cb_write */
+	overlay_target_ioctl,	/* cb_ioctl */
+	nodev,			/* cb_devmap */
+	nodev,			/* cb_mmap */
+	nodev,			/* cb_segmap */
+	nochpoll,		/* cb_chpoll */
+	ddi_prop_op,		/* cb_prop_op */
+	NULL,			/* cb_stream */
+	D_MP,			/* cb_flag */
+	CB_REV,			/* cb_rev */
+	nodev,			/* cb_aread */
+	nodev,			/* cb_awrite */
+};
+
+static struct dev_ops overlay_dev_ops = {
+	DEVO_REV,		/* devo_rev */
+	0,			/* devo_refcnt */
+	overlay_getinfo,	/* devo_getinfo */
+	nulldev,		/* devo_identify */
+	nulldev,		/* devo_probe */
+	overlay_attach,		/* devo_attach */
+	overlay_detach,		/* devo_detach */
+	nulldev,		/* devo_reset */
+	&overlay_cbops,		/* devo_cb_ops */
+	NULL,			/* devo_bus_ops */
+	NULL,			/* devo_power */
+	ddi_quiesce_not_supported	/* devo_quiesce */
+};
+
+static struct modldrv overlay_modldrv = {
+	&mod_driverops,
+	"Overlay Network Driver",
+	&overlay_dev_ops
+};
+
+static struct modlinkage overlay_linkage = {
+	MODREV_1,
+	&overlay_modldrv
+};
+
+static int
+overlay_init(void)
+{
+	mutex_init(&overlay_dev_lock, NULL, MUTEX_DRIVER, NULL);
+	list_create(&overlay_dev_list, sizeof (overlay_dev_t),
+	    offsetof(overlay_dev_t, odd_link));
+	overlay_mux_init();
+	overlay_plugin_init();
+	overlay_target_init();
+
+	return (DDI_SUCCESS);
+}
+
+static void
+overlay_fini(void)
+{
+	overlay_target_fini();
+	overlay_plugin_fini();
+	overlay_mux_fini();
+	mutex_destroy(&overlay_dev_lock);
+	list_destroy(&overlay_dev_list);
+}
+
+int
+_init(void)
+{
+	int err;
+
+	if ((err = overlay_init()) != DDI_SUCCESS)
+		return (err);
+
+	mac_init_ops(NULL, "overlay");
+	err = mod_install(&overlay_linkage);
+	if (err != DDI_SUCCESS) {
+		overlay_fini();
+		return (err);
+	}
+
+	return (0);
+}
+
+int
+_info(struct modinfo *modinfop)
+{
+	return (mod_info(&overlay_linkage, modinfop));
+}
+
+int
+_fini(void)
+{
+	int err;
+
+	err = mod_remove(&overlay_linkage);
+	if (err != 0)
+		return (err);
+
+	overlay_fini();
+	return (0);
+}
diff --git a/usr/src/uts/common/io/overlay/overlay.conf b/usr/src/uts/common/io/overlay/overlay.conf
new file mode 100644
index 0000000000..4b62fafd94
--- /dev/null
+++ b/usr/src/uts/common/io/overlay/overlay.conf
@@ -0,0 +1,16 @@
+#
+# This file and its contents are supplied under the terms of the
+# Common Development and Distribution License ("CDDL"), version 1.0.
+# You may only use this file in accordance with the terms of version
+# 1.0 of the CDDL.
+#
+# A full copy of the text of the CDDL should have accompanied this
+# source.  A copy of the CDDL is also available via the Internet at
+# http://www.illumos.org/license/CDDL.
+#
+
+#
+# Copyright 2015, Joyent, Inc.
+#
+
+name="overlay" parent="pseudo" instance=0;
diff --git a/usr/src/uts/common/io/overlay/overlay.mapfile b/usr/src/uts/common/io/overlay/overlay.mapfile
new file mode 100644
index 0000000000..800d72dc2b
--- /dev/null
+++ b/usr/src/uts/common/io/overlay/overlay.mapfile
@@ -0,0 +1,46 @@
+#
+# This file and its contents are supplied under the terms of the
+# Common Development and Distribution License ("CDDL"), version 1.0.
+# You may only use this file in accordance with the terms of version
+# 1.0 of the CDDL.
+#
+# A full copy of the text of the CDDL should have accompanied this
+# source.  A copy of the CDDL is also available via the Internet at
+# http://www.illumos.org/license/CDDL.
+#
+
+#
+# Copyright 2015 Joyent, Inc.
+#
+
+#
+# MAPFILE HEADER START
+#
+# WARNING:  STOP NOW.  DO NOT MODIFY THIS FILE.
+# Object versioning must comply with the rules detailed in
+#
+#	usr/src/lib/README.mapfiles
+#
+# You should not be making modifications here until you've read the most current
+# copy of that file. If you need help, contact a gatekeeper for guidance.
+#
+# MAPFILE HEADER END
+#
+
+$mapfile_version 2
+
+SYMBOL_VERSION ILLUMOSprivate {
+    global:
+	# DDI Interfaces
+	_fini;
+	_init;
+	_info;
+
+	# Encapsualation Plugin interfaces
+	overlay_plugin_alloc;
+	overlay_plugin_free;
+	overlay_plugin_register;
+	overlay_plugin_unregister;
+    local:
+	*;
+};
diff --git a/usr/src/uts/common/io/overlay/overlay_fm.c b/usr/src/uts/common/io/overlay/overlay_fm.c
new file mode 100644
index 0000000000..f26356ceaf
--- /dev/null
+++ b/usr/src/uts/common/io/overlay/overlay_fm.c
@@ -0,0 +1,84 @@
+/*
+ * This file and its contents are supplied under the terms of the
+ * Common Development and Distribution License ("CDDL"), version 1.0.
+ * You may only use this file in accordance with the terms of version
+ * 1.0 of the CDDL.
+ *
+ * A full copy of the text of the CDDL should have accompanied this
+ * source.  A copy of the CDDL is also available via the Internet at
+ * http://www.illumos.org/license/CDDL.
+ */
+
+/*
+ * Copyright 2015 Joyent, Inc.
+ */
+
+/*
+ * Overlay device FMA operations.
+ *
+ * For more information, see the big theory statement in
+ * uts/common/io/overlay/overlay.c
+ */
+
+#include <sys/ddifm.h>
+#include <sys/overlay_impl.h>
+
+kmutex_t overlay_fm_lock;
+uint_t overlay_fm_count;
+
+void
+overlay_fm_init(void)
+{
+	overlay_fm_count = 0;
+	mutex_init(&overlay_fm_lock, NULL, MUTEX_DRIVER, NULL);
+}
+
+void
+overlay_fm_fini(void)
+{
+	VERIFY(overlay_fm_count == 0);
+	mutex_destroy(&overlay_fm_lock);
+}
+
+void
+overlay_fm_degrade(overlay_dev_t *odd, const char *msg)
+{
+	mutex_enter(&overlay_fm_lock);
+	mutex_enter(&odd->odd_lock);
+
+	if (msg != NULL)
+		(void) strlcpy(odd->odd_fmamsg, msg, OVERLAY_STATUS_BUFLEN);
+
+	if (odd->odd_flags & OVERLAY_F_DEGRADED)
+		goto out;
+
+	odd->odd_flags |= OVERLAY_F_DEGRADED;
+	overlay_fm_count++;
+	if (overlay_fm_count == 1) {
+		ddi_fm_service_impact(overlay_dip, DDI_SERVICE_DEGRADED);
+		mac_link_update(odd->odd_mh, LINK_STATE_DOWN);
+	}
+out:
+	mutex_exit(&odd->odd_lock);
+	mutex_exit(&overlay_fm_lock);
+}
+
+void
+overlay_fm_restore(overlay_dev_t *odd)
+{
+	mutex_enter(&overlay_fm_lock);
+	mutex_enter(&odd->odd_lock);
+	if (!(odd->odd_flags & OVERLAY_F_DEGRADED))
+		goto out;
+
+	odd->odd_fmamsg[0] = '\0';
+	odd->odd_flags &= ~OVERLAY_F_DEGRADED;
+	overlay_fm_count--;
+	if (overlay_fm_count == 0) {
+		ddi_fm_service_impact(overlay_dip, DDI_SERVICE_RESTORED);
+		mac_link_update(odd->odd_mh, LINK_STATE_UP);
+	}
+out:
+	mutex_exit(&odd->odd_lock);
+	mutex_exit(&overlay_fm_lock);
+}
diff --git a/usr/src/uts/common/io/overlay/overlay_mux.c b/usr/src/uts/common/io/overlay/overlay_mux.c
new file mode 100644
index 0000000000..9f70e8c83e
--- /dev/null
+++ b/usr/src/uts/common/io/overlay/overlay_mux.c
@@ -0,0 +1,354 @@
+/*
+ * This file and its contents are supplied under the terms of the
+ * Common Development and Distribution License ("CDDL"), version 1.0.
+ * You may only use this file in accordance with the terms of version
+ * 1.0 of the CDDL.
+ *
+ * A full copy of the text of the CDDL should have accompanied this
+ * source.  A copy of the CDDL is also available via the Internet at
+ * http://www.illumos.org/license/CDDL.
+ */
+
+/*
+ * Copyright 2015 Joyent, Inc.
+ */
+
+/*
+ * Overlay device ksocket multiplexer.
+ *
+ * For more information, see the big theory statement in
+ * uts/common/io/overlay/overlay.c
+ */
+
+#include <sys/types.h>
+#include <sys/socket.h>
+#include <sys/ksynch.h>
+#include <sys/ksocket.h>
+#include <sys/avl.h>
+#include <sys/list.h>
+#include <sys/sysmacros.h>
+#include <sys/strsubr.h>
+#include <sys/strsun.h>
+#include <sys/tihdr.h>
+
+#include <sys/overlay_impl.h>
+
+#include <sys/sdt.h>
+
+#define	OVERLAY_FREEMSG(mp, reason) \
+    DTRACE_PROBE2(overlay__fremsg, mblk_t *, mp, char *, reason)
+
+static list_t overlay_mux_list;
+static kmutex_t overlay_mux_lock;
+
+void
+overlay_mux_init(void)
+{
+	list_create(&overlay_mux_list, sizeof (overlay_mux_t),
+	    offsetof(overlay_mux_t, omux_lnode));
+	mutex_init(&overlay_mux_lock, NULL, MUTEX_DRIVER, NULL);
+}
+
+void
+overlay_mux_fini(void)
+{
+	mutex_destroy(&overlay_mux_lock);
+	list_destroy(&overlay_mux_list);
+}
+
+static int
+overlay_mux_comparator(const void *a, const void *b)
+{
+	const overlay_dev_t *odl, *odr;
+	odl = a;
+	odr = b;
+	if (odl->odd_vid > odr->odd_vid)
+		return (1);
+	else if (odl->odd_vid < odr->odd_vid)
+		return (-1);
+	else
+		return (0);
+}
+
+/*
+ * This is the central receive data path. We need to decode the packet, if we
+ * can, and then deliver it to the appropriate overlay.
+ */
+/* ARGSUSED */
+static boolean_t
+overlay_mux_recv(ksocket_t ks, mblk_t *mpchain, size_t msgsize, int oob,
+    void *arg)
+{
+	mblk_t *mp, *nmp, *fmp;
+	overlay_mux_t *mux = arg;
+
+	/*
+	 * We may have a received a chain of messages. Each messsage in the
+	 * chain will likely have a T_unitdata_ind attached to it as an M_PROTO.
+	 * If we aren't getting that, we should probably drop that for the
+	 * moment.
+	 */
+	for (mp = mpchain; mp != NULL; mp = nmp) {
+		struct T_unitdata_ind *tudi;
+		ovep_encap_info_t infop;
+		overlay_dev_t od, *odd;
+		int ret;
+
+		nmp = mp->b_next;
+		mp->b_next = NULL;
+
+		if (DB_TYPE(mp) != M_PROTO) {
+			OVERLAY_FREEMSG(mp, "first one isn't M_PROTO");
+			freemsg(mp);
+			continue;
+		}
+
+		if (mp->b_cont == NULL) {
+			OVERLAY_FREEMSG(mp, "missing a b_cont");
+			freemsg(mp);
+			continue;
+		}
+
+		tudi = (struct T_unitdata_ind *)mp->b_rptr;
+		if (tudi->PRIM_type != T_UNITDATA_IND) {
+			OVERLAY_FREEMSG(mp, "Not a T_unitdata_ind *");
+			freemsg(mp);
+			continue;
+		}
+
+		/*
+		 * In the future, we'll care about the source information
+		 * for purposes of telling varpd for oob invalidation. But for
+		 * now, just drop that block.
+		 */
+		fmp = mp;
+		mp = fmp->b_cont;
+		fmp->b_cont = NULL;
+		freemsg(fmp);
+
+		/*
+		 * Decap and deliver.
+		 */
+		bzero(&infop, sizeof (ovep_encap_info_t));
+		ret = mux->omux_plugin->ovp_ops->ovpo_decap(NULL, mp, &infop);
+		if (ret != 0) {
+			OVERLAY_FREEMSG(mp, "decap failed");
+			freemsg(mp);
+			continue;
+		}
+		if (MBLKL(mp) > infop.ovdi_hdr_size) {
+			mp->b_rptr += infop.ovdi_hdr_size;
+		} else {
+			while (infop.ovdi_hdr_size != 0) {
+				size_t rem, blkl;
+
+				if (mp == NULL)
+					break;
+
+				blkl = MBLKL(mp);
+				rem = MIN(infop.ovdi_hdr_size, blkl);
+				infop.ovdi_hdr_size -= rem;
+				mp->b_rptr += rem;
+				if (rem == blkl) {
+					fmp = mp;
+					mp = fmp->b_cont;
+					fmp->b_cont = NULL;
+					OVERLAY_FREEMSG(mp,
+					    "freed a fmp block");
+					freemsg(fmp);
+				}
+			}
+			if (mp == NULL) {
+				OVERLAY_FREEMSG(mp, "freed it all...");
+				continue;
+			}
+		}
+
+
+		od.odd_vid = infop.ovdi_id;
+		mutex_enter(&mux->omux_lock);
+		odd = avl_find(&mux->omux_devices, &od, NULL);
+		if (odd == NULL) {
+			mutex_exit(&mux->omux_lock);
+			OVERLAY_FREEMSG(mp, "no matching vid");
+			freemsg(mp);
+			continue;
+		}
+		mutex_enter(&odd->odd_lock);
+		if ((odd->odd_flags & OVERLAY_F_MDDROP) ||
+		    !(odd->odd_flags & OVERLAY_F_IN_MUX)) {
+			mutex_exit(&odd->odd_lock);
+			mutex_exit(&mux->omux_lock);
+			OVERLAY_FREEMSG(mp, "dev dropped");
+			freemsg(mp);
+			continue;
+		}
+		overlay_io_start(odd, OVERLAY_F_IN_RX);
+		mutex_exit(&odd->odd_lock);
+		mutex_exit(&mux->omux_lock);
+
+		mac_rx(odd->odd_mh, NULL, mp);
+
+		mutex_enter(&odd->odd_lock);
+		overlay_io_done(odd, OVERLAY_F_IN_RX);
+		mutex_exit(&odd->odd_lock);
+	}
+
+	return (B_TRUE);
+}
+
+/*
+ * Register a given device with a socket backend. If no such device socket
+ * exists, create a new one.
+ */
+overlay_mux_t *
+overlay_mux_open(overlay_plugin_t *opp, int domain, int family, int protocol,
+    struct sockaddr *addr, socklen_t len, int *errp)
+{
+	int err;
+	overlay_mux_t *mux;
+	ksocket_t ksock;
+
+	if (errp == NULL)
+		errp = &err;
+
+	mutex_enter(&overlay_mux_lock);
+	for (mux = list_head(&overlay_mux_list); mux != NULL;
+	    mux = list_next(&overlay_mux_list, mux)) {
+		if (domain == mux->omux_domain &&
+		    family == mux->omux_family &&
+		    protocol == mux->omux_protocol &&
+		    len == mux->omux_alen &&
+		    bcmp(addr, mux->omux_addr, len) == 0) {
+
+			if (opp != mux->omux_plugin) {
+				*errp = EEXIST;
+				return (NULL);
+			}
+
+			mutex_enter(&mux->omux_lock);
+			mux->omux_count++;
+			mutex_exit(&mux->omux_lock);
+			mutex_exit(&overlay_mux_lock);
+			*errp = 0;
+			return (mux);
+		}
+	}
+
+	/*
+	 * Today we aren't zone-aware and only exist in the global zone. When we
+	 * allow for things to exist in the non-global zone, we'll want to use a
+	 * credential that's actually specific to the zone.
+	 */
+	*errp = ksocket_socket(&ksock, domain, family, protocol, KSOCKET_SLEEP,
+	    kcred);
+	if (*errp != 0) {
+		mutex_exit(&overlay_mux_lock);
+		return (NULL);
+	}
+
+	*errp = ksocket_bind(ksock, addr, len, kcred);
+	if (*errp != 0) {
+		mutex_exit(&overlay_mux_lock);
+		ksocket_close(ksock, kcred);
+		return (NULL);
+	}
+
+	/*
+	 * Ask our lower layer to optionally toggle anything they need on this
+	 * socket. Because a socket is owned by a single type of plugin, we can
+	 * then ask it to perform any additional socket set up it'd like to do.
+	 */
+	if (opp->ovp_ops->ovpo_sockopt != NULL &&
+	    (*errp = opp->ovp_ops->ovpo_sockopt(ksock)) != 0) {
+		mutex_exit(&overlay_mux_lock);
+		ksocket_close(ksock, kcred);
+		return (NULL);
+	}
+
+	mux = kmem_alloc(sizeof (overlay_mux_t), KM_SLEEP);
+	list_link_init(&mux->omux_lnode);
+	mux->omux_ksock = ksock;
+	mux->omux_plugin = opp;
+	mux->omux_domain = domain;
+	mux->omux_family = family;
+	mux->omux_protocol = protocol;
+	mux->omux_addr = kmem_alloc(len, KM_SLEEP);
+	bcopy(addr, mux->omux_addr, len);
+	mux->omux_alen = len;
+	mux->omux_count = 1;
+	avl_create(&mux->omux_devices, overlay_mux_comparator,
+	    sizeof (overlay_dev_t), offsetof(overlay_dev_t, odd_muxnode));
+	mutex_init(&mux->omux_lock, NULL, MUTEX_DRIVER, NULL);
+
+
+	/* Once this is called, we need to expect to rx data */
+	*errp = ksocket_krecv_set(ksock, overlay_mux_recv, mux);
+	if (*errp != 0) {
+		ksocket_close(ksock, kcred);
+		mutex_destroy(&mux->omux_lock);
+		avl_destroy(&mux->omux_devices);
+		kmem_free(mux->omux_addr, len);
+		kmem_free(mux, sizeof (overlay_mux_t));
+		return (NULL);
+	}
+
+	list_insert_tail(&overlay_mux_list, mux);
+	mutex_exit(&overlay_mux_lock);
+
+	*errp = 0;
+	return (mux);
+}
+
+void
+overlay_mux_close(overlay_mux_t *mux)
+{
+	mutex_enter(&overlay_mux_lock);
+	mutex_enter(&mux->omux_lock);
+	mux->omux_count--;
+	if (mux->omux_count != 0) {
+		mutex_exit(&mux->omux_lock);
+		mutex_exit(&overlay_mux_lock);
+		return;
+	}
+	list_remove(&overlay_mux_list, mux);
+	mutex_exit(&mux->omux_lock);
+	mutex_exit(&overlay_mux_lock);
+
+	ksocket_close(mux->omux_ksock, kcred);
+	avl_destroy(&mux->omux_devices);
+	kmem_free(mux->omux_addr, mux->omux_alen);
+	kmem_free(mux, sizeof (overlay_mux_t));
+}
+
+void
+overlay_mux_add_dev(overlay_mux_t *mux, overlay_dev_t *odd)
+{
+	mutex_enter(&mux->omux_lock);
+	avl_add(&mux->omux_devices, odd);
+	mutex_exit(&mux->omux_lock);
+}
+
+void
+overlay_mux_remove_dev(overlay_mux_t *mux, overlay_dev_t *odd)
+{
+	mutex_enter(&mux->omux_lock);
+	avl_remove(&mux->omux_devices, odd);
+	mutex_exit(&mux->omux_lock);
+}
+
+int
+overlay_mux_tx(overlay_mux_t *mux, struct msghdr *hdr, mblk_t *mp)
+{
+	int ret;
+
+	/*
+	 * It'd be nice to be able to use MSG_MBLK_QUICKRELE, unfortunately,
+	 * that isn't actually supported by UDP at this time.
+	 */
+	ret = ksocket_sendmblk(mux->omux_ksock, hdr, 0, &mp, kcred);
+	if (ret != 0)
+		freemsg(mp);
+
+	return (ret);
+}
diff --git a/usr/src/uts/common/io/overlay/overlay_plugin.c b/usr/src/uts/common/io/overlay/overlay_plugin.c
new file mode 100644
index 0000000000..348ddb92a2
--- /dev/null
+++ b/usr/src/uts/common/io/overlay/overlay_plugin.c
@@ -0,0 +1,281 @@
+/*
+ * This file and its contents are supplied under the terms of the
+ * Common Development and Distribution License ("CDDL"), version 1.0.
+ * You may only use this file in accordance with the terms of version
+ * 1.0 of the CDDL.
+ *
+ * A full copy of the text of the CDDL should have accompanied this
+ * source.  A copy of the CDDL is also available via the Internet at
+ * http://www.illumos.org/license/CDDL.
+ */
+
+/*
+ * Copyright 2015 Joyent, Inc.
+ */
+
+/*
+ * Overlay device encapsulation plugin management
+ *
+ * For more information, see the big theory statement in
+ * uts/common/io/overlay/overlay.c
+ */
+
+#include <sys/types.h>
+#include <sys/kmem.h>
+#include <sys/ddi.h>
+#include <sys/sunddi.h>
+#include <sys/errno.h>
+#include <sys/sysmacros.h>
+#include <sys/modctl.h>
+
+#include <sys/overlay_impl.h>
+
+static kmem_cache_t *overlay_plugin_cache;
+static kmutex_t overlay_plugin_lock;
+static list_t overlay_plugin_list;
+
+#define	OVERLAY_MODDIR	"overlay"
+
+/* ARGSUSED */
+static int
+overlay_plugin_cache_constructor(void *buf, void *arg, int kmflags)
+{
+	overlay_plugin_t *opp = buf;
+
+	mutex_init(&opp->ovp_mutex, NULL, MUTEX_DRIVER, NULL);
+	list_link_init(&opp->ovp_link);
+
+	return (0);
+}
+
+/* ARGSUSED */
+static void
+overlay_plugin_cache_destructor(void *buf, void *arg)
+{
+	overlay_plugin_t *opp = buf;
+	ASSERT(list_link_active(&opp->ovp_link) == 0);
+	mutex_destroy(&opp->ovp_mutex);
+}
+
+void
+overlay_plugin_init(void)
+{
+	mutex_init(&overlay_plugin_lock, NULL, MUTEX_DRIVER, 0);
+
+	/*
+	 * In the future we may want to have a reaper to unload unused modules
+	 * to help the kernel be able to reclaim memory.
+	 */
+	overlay_plugin_cache = kmem_cache_create("overlay_plugin_cache",
+	    sizeof (overlay_plugin_t), 0, overlay_plugin_cache_constructor,
+	    overlay_plugin_cache_destructor, NULL, NULL, NULL, 0);
+	list_create(&overlay_plugin_list, sizeof (overlay_plugin_t),
+	    offsetof(overlay_plugin_t, ovp_link));
+}
+
+void
+overlay_plugin_fini(void)
+{
+	mutex_enter(&overlay_plugin_lock);
+	VERIFY(list_is_empty(&overlay_plugin_list));
+	mutex_exit(&overlay_plugin_lock);
+
+	list_destroy(&overlay_plugin_list);
+	kmem_cache_destroy(overlay_plugin_cache);
+	mutex_destroy(&overlay_plugin_lock);
+}
+
+overlay_plugin_register_t *
+overlay_plugin_alloc(uint_t version)
+{
+	overlay_plugin_register_t *ovrp;
+	/* Version 1 is the only one that exists */
+	if (version != OVEP_VERSION_ONE)
+		return (NULL);
+
+	ovrp = kmem_zalloc(sizeof (overlay_plugin_register_t), KM_SLEEP);
+	ovrp->ovep_version = version;
+	return (ovrp);
+}
+
+void
+overlay_plugin_free(overlay_plugin_register_t *ovrp)
+{
+	kmem_free(ovrp, sizeof (overlay_plugin_register_t));
+}
+
+int
+overlay_plugin_register(overlay_plugin_register_t *ovrp)
+{
+	overlay_plugin_t *opp, *ipp;
+
+	/* Sanity check parameters of the registration */
+	if (ovrp->ovep_version != OVEP_VERSION_ONE)
+		return (EINVAL);
+
+	if (ovrp->ovep_name == NULL || ovrp->ovep_ops == NULL)
+		return (EINVAL);
+
+	if ((ovrp->ovep_flags & ~(OVEP_F_VLAN_TAG)) != 0)
+		return (EINVAL);
+
+	if (ovrp->ovep_id_size < 1)
+		return (EINVAL);
+
+	/* Don't support anything that has an id size larger than 8 bytes */
+	if (ovrp->ovep_id_size > 8)
+		return (ENOTSUP);
+
+	if (ovrp->ovep_dest == OVERLAY_PLUGIN_D_INVALID)
+		return (EINVAL);
+
+	if ((ovrp->ovep_dest & ~OVERLAY_PLUGIN_D_MASK) != 0)
+		return (EINVAL);
+
+	if (ovrp->ovep_ops->ovpo_callbacks != 0)
+		return (EINVAL);
+	if (ovrp->ovep_ops->ovpo_init == NULL)
+		return (EINVAL);
+	if (ovrp->ovep_ops->ovpo_fini == NULL)
+		return (EINVAL);
+	if (ovrp->ovep_ops->ovpo_encap == NULL)
+		return (EINVAL);
+	if (ovrp->ovep_ops->ovpo_decap == NULL)
+		return (EINVAL);
+	if (ovrp->ovep_ops->ovpo_socket == NULL)
+		return (EINVAL);
+	if (ovrp->ovep_ops->ovpo_getprop == NULL)
+		return (EINVAL);
+	if (ovrp->ovep_ops->ovpo_setprop == NULL)
+		return (EINVAL);
+	if (ovrp->ovep_ops->ovpo_propinfo == NULL)
+		return (EINVAL);
+
+
+	opp = kmem_cache_alloc(overlay_plugin_cache, KM_SLEEP);
+	opp->ovp_active = 0;
+	opp->ovp_name = ovrp->ovep_name;
+	opp->ovp_ops = ovrp->ovep_ops;
+	opp->ovp_props = ovrp->ovep_props;
+	opp->ovp_id_size = ovrp->ovep_id_size;
+	opp->ovp_flags = ovrp->ovep_flags;
+	opp->ovp_dest = ovrp->ovep_dest;
+
+	opp->ovp_nprops = 0;
+	if (ovrp->ovep_props != NULL) {
+		while (ovrp->ovep_props[opp->ovp_nprops] != NULL) {
+			if (strlen(ovrp->ovep_props[opp->ovp_nprops]) >=
+			    OVERLAY_PROP_NAMELEN) {
+				mutex_exit(&overlay_plugin_lock);
+				kmem_cache_free(overlay_plugin_cache, opp);
+				return (EINVAL);
+			}
+			opp->ovp_nprops++;
+		}
+	}
+
+	mutex_enter(&overlay_plugin_lock);
+	for (ipp = list_head(&overlay_plugin_list); ipp != NULL;
+	    ipp = list_next(&overlay_plugin_list, ipp)) {
+		if (strcmp(ipp->ovp_name, opp->ovp_name) == 0) {
+			mutex_exit(&overlay_plugin_lock);
+			kmem_cache_free(overlay_plugin_cache, opp);
+			return (EEXIST);
+		}
+	}
+	list_insert_tail(&overlay_plugin_list, opp);
+	mutex_exit(&overlay_plugin_lock);
+
+	return (0);
+}
+
+int
+overlay_plugin_unregister(const char *name)
+{
+	overlay_plugin_t *opp;
+
+	mutex_enter(&overlay_plugin_lock);
+	for (opp = list_head(&overlay_plugin_list); opp != NULL;
+	    opp = list_next(&overlay_plugin_list, opp)) {
+		if (strcmp(opp->ovp_name, name) == 0)
+			break;
+	}
+
+	if (opp == NULL) {
+		mutex_exit(&overlay_plugin_lock);
+		return (ENOENT);
+	}
+
+	mutex_enter(&opp->ovp_mutex);
+	if (opp->ovp_active > 0) {
+		mutex_exit(&opp->ovp_mutex);
+		mutex_exit(&overlay_plugin_lock);
+		return (EBUSY);
+	}
+	mutex_exit(&opp->ovp_mutex);
+
+	list_remove(&overlay_plugin_list, opp);
+	mutex_exit(&overlay_plugin_lock);
+
+	kmem_cache_free(overlay_plugin_cache, opp);
+	return (0);
+}
+
+overlay_plugin_t *
+overlay_plugin_lookup(const char *name)
+{
+	overlay_plugin_t *opp;
+	boolean_t trymodload = B_FALSE;
+
+	for (;;) {
+		mutex_enter(&overlay_plugin_lock);
+		for (opp = list_head(&overlay_plugin_list); opp != NULL;
+		    opp = list_next(&overlay_plugin_list, opp)) {
+			if (strcmp(name, opp->ovp_name) == 0) {
+				mutex_enter(&opp->ovp_mutex);
+				opp->ovp_active++;
+				mutex_exit(&opp->ovp_mutex);
+				mutex_exit(&overlay_plugin_lock);
+				return (opp);
+			}
+		}
+		mutex_exit(&overlay_plugin_lock);
+
+		if (trymodload == B_TRUE)
+			return (NULL);
+
+		/*
+		 * If we didn't find it, it may still exist, but just not have
+		 * been a loaded module. In that case, we'll do one attempt to
+		 * load it.
+		 */
+		if (modload(OVERLAY_MODDIR, (char *)name) == -1)
+			return (NULL);
+		trymodload = B_TRUE;
+	}
+
+}
+
+void
+overlay_plugin_rele(overlay_plugin_t *opp)
+{
+	mutex_enter(&opp->ovp_mutex);
+	ASSERT(opp->ovp_active > 0);
+	opp->ovp_active--;
+	mutex_exit(&opp->ovp_mutex);
+}
+
+void
+overlay_plugin_walk(overlay_plugin_walk_f func, void *arg)
+{
+	overlay_plugin_t *opp;
+	mutex_enter(&overlay_plugin_lock);
+	for (opp = list_head(&overlay_plugin_list); opp != NULL;
+	    opp = list_next(&overlay_plugin_list, opp)) {
+		if (func(opp, arg) != 0) {
+			mutex_exit(&overlay_plugin_lock);
+			return;
+		}
+	}
+	mutex_exit(&overlay_plugin_lock);
+}
diff --git a/usr/src/uts/common/io/overlay/overlay_prop.c b/usr/src/uts/common/io/overlay/overlay_prop.c
new file mode 100644
index 0000000000..ba1ea2a629
--- /dev/null
+++ b/usr/src/uts/common/io/overlay/overlay_prop.c
@@ -0,0 +1,122 @@
+/*
+ * This file and its contents are supplied under the terms of the
+ * Common Development and Distribution License ("CDDL"), version 1.0.
+ * You may only use this file in accordance with the terms of version
+ * 1.0 of the CDDL.
+ *
+ * A full copy of the text of the CDDL should have accompanied this
+ * source.  A copy of the CDDL is also available via the Internet at
+ * http://www.illumos.org/license/CDDL.
+ */
+
+/*
+ * Copyright 2015, Joyent, Inc.
+ */
+
+/*
+ * Routines for manipulating property information structures.
+ *
+ * For more information, see the big theory statement in
+ * uts/common/io/overlay/overlay.c
+ */
+
+#include <sys/overlay_impl.h>
+
+void
+overlay_prop_init(overlay_prop_handle_t phdl)
+{
+	overlay_ioc_propinfo_t *infop = (overlay_ioc_propinfo_t *)phdl;
+	mac_propval_range_t *rangep = (mac_propval_range_t *)infop->oipi_poss;
+
+	infop->oipi_posssize = sizeof (mac_propval_range_t);
+	bzero(rangep, sizeof (mac_propval_range_t));
+}
+
+void
+overlay_prop_set_name(overlay_prop_handle_t phdl, const char *name)
+{
+	overlay_ioc_propinfo_t *infop = (overlay_ioc_propinfo_t *)phdl;
+	(void) strlcpy(infop->oipi_name, name, OVERLAY_PROP_NAMELEN);
+}
+
+void
+overlay_prop_set_prot(overlay_prop_handle_t phdl, overlay_prop_prot_t prot)
+{
+	overlay_ioc_propinfo_t *infop = (overlay_ioc_propinfo_t *)phdl;
+	infop->oipi_prot = prot;
+}
+
+void
+overlay_prop_set_type(overlay_prop_handle_t phdl, overlay_prop_type_t type)
+{
+	overlay_ioc_propinfo_t *infop = (overlay_ioc_propinfo_t *)phdl;
+	infop->oipi_type = type;
+}
+
+int
+overlay_prop_set_default(overlay_prop_handle_t phdl, void *def, ssize_t len)
+{
+	overlay_ioc_propinfo_t *infop = (overlay_ioc_propinfo_t *)phdl;
+
+	if (len > OVERLAY_PROP_SIZEMAX)
+		return (E2BIG);
+
+	if (len < 0)
+		return (EOVERFLOW);
+
+	bcopy(def, infop->oipi_default, len);
+	infop->oipi_defsize = (uint32_t)len;
+
+	return (0);
+}
+
+void
+overlay_prop_set_nodefault(overlay_prop_handle_t phdl)
+{
+	overlay_ioc_propinfo_t *infop = (overlay_ioc_propinfo_t *)phdl;
+	infop->oipi_default[0] = '\0';
+	infop->oipi_defsize = 0;
+}
+
+void
+overlay_prop_set_range_uint32(overlay_prop_handle_t phdl, uint32_t min,
+    uint32_t max)
+{
+	overlay_ioc_propinfo_t *infop = (overlay_ioc_propinfo_t *)phdl;
+	mac_propval_range_t *rangep = (mac_propval_range_t *)infop->oipi_poss;
+
+	if (rangep->mpr_count != 0 && rangep->mpr_type != MAC_PROPVAL_UINT32)
+		return;
+
+	if (infop->oipi_posssize + sizeof (mac_propval_uint32_range_t) >
+	    sizeof (infop->oipi_poss))
+		return;
+
+	infop->oipi_posssize += sizeof (mac_propval_uint32_range_t);
+	rangep->mpr_count++;
+	rangep->mpr_type = MAC_PROPVAL_UINT32;
+	rangep->u.mpr_uint32[rangep->mpr_count-1].mpur_min = min;
+	rangep->u.mpr_uint32[rangep->mpr_count-1].mpur_max = max;
+}
+
+void
+overlay_prop_set_range_str(overlay_prop_handle_t phdl, const char *str)
+{
+	size_t len = strlen(str) + 1; /* Account for a null terminator */
+	overlay_ioc_propinfo_t *infop = (overlay_ioc_propinfo_t *)phdl;
+	mac_propval_range_t *rangep = (mac_propval_range_t *)infop->oipi_poss;
+	mac_propval_str_range_t *pstr = &rangep->u.mpr_str;
+
+	if (rangep->mpr_count != 0 && rangep->mpr_type != MAC_PROPVAL_STR)
+		return;
+
+	if (infop->oipi_posssize + len > sizeof (infop->oipi_poss))
+		return;
+
+	rangep->mpr_count++;
+	rangep->mpr_type = MAC_PROPVAL_STR;
+	strlcpy((char *)&pstr->mpur_data[pstr->mpur_nextbyte], str,
+	    sizeof (infop->oipi_poss) - infop->oipi_posssize);
+	pstr->mpur_nextbyte += len;
+	infop->oipi_posssize += len;
+}
diff --git a/usr/src/uts/common/io/overlay/overlay_target.c b/usr/src/uts/common/io/overlay/overlay_target.c
new file mode 100644
index 0000000000..c2da20c883
--- /dev/null
+++ b/usr/src/uts/common/io/overlay/overlay_target.c
@@ -0,0 +1,1662 @@
+/*
+ * This file and its contents are supplied under the terms of the
+ * Common Development and Distribution License ("CDDL"), version 1.0.
+ * You may only use this file in accordance with the terms of version
+ * 1.0 of the CDDL.
+ *
+ * A full copy of the text of the CDDL should have accompanied this
+ * source.  A copy of the CDDL is also available via the Internet at
+ * http://www.illumos.org/license/CDDL.
+ */
+
+/*
+ * Copyright 2015 Joyent, Inc.
+ */
+
+/*
+ * Overlay device target cache management
+ *
+ * For more information, see the big theory statement in
+ * uts/common/io/overlay/overlay.c
+ */
+
+#include <sys/types.h>
+#include <sys/ethernet.h>
+#include <sys/kmem.h>
+#include <sys/policy.h>
+#include <sys/sysmacros.h>
+#include <sys/stream.h>
+#include <sys/strsun.h>
+#include <sys/strsubr.h>
+#include <sys/mac_provider.h>
+#include <sys/mac_client.h>
+#include <sys/mac_client_priv.h>
+#include <sys/vlan.h>
+#include <sys/crc32.h>
+#include <sys/cred.h>
+#include <sys/file.h>
+#include <sys/errno.h>
+#include <sys/ddi.h>
+#include <sys/sunddi.h>
+
+#include <sys/overlay_impl.h>
+#include <sys/sdt.h>
+
+/*
+ * This is total straw man, but at least it's a prime number. Here we're
+ * going to have to go through and do a lot of evaluation and understanding as
+ * to how these target caches should grow and shrink, as well as, memory
+ * pressure and evictions. This just gives us a starting point that'll be 'good
+ * enough', until it's not.
+ */
+#define	OVERLAY_HSIZE	823
+
+/*
+ * We use this data structure to keep track of what requests have been actively
+ * allocated to a given instance so we know what to put back on the pending
+ * list.
+ */
+typedef struct overlay_target_hdl {
+	minor_t oth_minor;		/* RO */
+	zoneid_t oth_zoneid;		/* RO */
+	int oth_oflags;			/* RO */
+	list_node_t oth_link;		/* overlay_target_lock */
+	kmutex_t oth_lock;
+	list_t	oth_outstanding;	/* oth_lock */
+} overlay_target_hdl_t;
+
+typedef int (*overlay_target_copyin_f)(const void *, void **, size_t *, int);
+typedef int (*overlay_target_ioctl_f)(overlay_target_hdl_t *, void *);
+typedef int (*overlay_target_copyout_f)(void *, void *, size_t, int);
+
+typedef struct overaly_target_ioctl {
+	int		oti_cmd;	/* ioctl id */
+	boolean_t	oti_write;	/* ioctl requires FWRITE */
+	boolean_t	oti_ncopyout;	/* copyout data? */
+	overlay_target_copyin_f oti_copyin;	/* copyin func */
+	overlay_target_ioctl_f oti_func; /* function to call */
+	overlay_target_copyout_f oti_copyout;	/* copyin func */
+	size_t		oti_size;	/* size of user level structure */
+} overlay_target_ioctl_t;
+
+static kmem_cache_t *overlay_target_cache;
+static kmem_cache_t *overlay_entry_cache;
+static id_space_t *overlay_thdl_idspace;
+static void *overlay_thdl_state;
+
+/*
+ * When we support overlay devices in the NGZ, then all of these need to become
+ * zone aware, by plugging into the netstack engine and becoming per-netstack
+ * data.
+ */
+static list_t overlay_thdl_list;
+static kmutex_t overlay_target_lock;
+static kcondvar_t overlay_target_condvar;
+static list_t overlay_target_list;
+static boolean_t overlay_target_excl;
+
+/*
+ * Outstanding data per hash table entry.
+ */
+static int overlay_ent_size = 128 * 1024;
+
+/* ARGSUSED */
+static int
+overlay_target_cache_constructor(void *buf, void *arg, int kmflgs)
+{
+	overlay_target_t *ott = buf;
+
+	mutex_init(&ott->ott_lock, NULL, MUTEX_DRIVER, NULL);
+	cv_init(&ott->ott_cond, NULL, CV_DRIVER, NULL);
+	return (0);
+}
+
+/* ARGSUSED */
+static void
+overlay_target_cache_destructor(void *buf, void *arg)
+{
+	overlay_target_t *ott = buf;
+
+	cv_destroy(&ott->ott_cond);
+	mutex_destroy(&ott->ott_lock);
+}
+
+/* ARGSUSED */
+static int
+overlay_entry_cache_constructor(void *buf, void *arg, int kmflgs)
+{
+	overlay_target_entry_t *ote = buf;
+
+	bzero(ote, sizeof (overlay_target_entry_t));
+	mutex_init(&ote->ote_lock, NULL, MUTEX_DRIVER, NULL);
+	return (0);
+}
+
+/* ARGSUSED */
+static void
+overlay_entry_cache_destructor(void *buf, void *arg)
+{
+	overlay_target_entry_t *ote = buf;
+
+	mutex_destroy(&ote->ote_lock);
+}
+
+static uint64_t
+overlay_mac_hash(const void *v)
+{
+	uint32_t crc;
+	CRC32(crc, v, ETHERADDRL, -1U, crc32_table);
+	return (crc);
+}
+
+static int
+overlay_mac_cmp(const void *a, const void *b)
+{
+	return (bcmp(a, b, ETHERADDRL));
+}
+
+/* ARGSUSED */
+static void
+overlay_target_entry_dtor(void *arg)
+{
+	overlay_target_entry_t *ote = arg;
+
+	ote->ote_flags = 0;
+	bzero(ote->ote_addr, ETHERADDRL);
+	ote->ote_ott = NULL;
+	ote->ote_odd = NULL;
+	freemsgchain(ote->ote_chead);
+	ote->ote_chead = ote->ote_ctail = NULL;
+	ote->ote_mbsize = 0;
+	ote->ote_vtime = 0;
+	kmem_cache_free(overlay_entry_cache, ote);
+}
+
+static int
+overlay_mac_avl(const void *a, const void *b)
+{
+	int i;
+	const overlay_target_entry_t *l, *r;
+	l = a;
+	r = b;
+
+	for (i = 0; i < ETHERADDRL; i++) {
+		if (l->ote_addr[i] > r->ote_addr[i])
+			return (1);
+		else if (l->ote_addr[i] < r->ote_addr[i])
+			return (-1);
+	}
+
+	return (0);
+}
+
+void
+overlay_target_init(void)
+{
+	int ret;
+	ret = ddi_soft_state_init(&overlay_thdl_state,
+	    sizeof (overlay_target_hdl_t), 1);
+	VERIFY(ret == 0);
+	overlay_target_cache = kmem_cache_create("overlay_target",
+	    sizeof (overlay_target_t), 0, overlay_target_cache_constructor,
+	    overlay_target_cache_destructor, NULL, NULL, NULL, 0);
+	overlay_entry_cache = kmem_cache_create("overlay_entry",
+	    sizeof (overlay_target_entry_t), 0, overlay_entry_cache_constructor,
+	    overlay_entry_cache_destructor, NULL, NULL, NULL, 0);
+	mutex_init(&overlay_target_lock, NULL, MUTEX_DRIVER, NULL);
+	cv_init(&overlay_target_condvar, NULL, CV_DRIVER, NULL);
+	list_create(&overlay_target_list, sizeof (overlay_target_entry_t),
+	    offsetof(overlay_target_entry_t, ote_qlink));
+	list_create(&overlay_thdl_list, sizeof (overlay_target_hdl_t),
+	    offsetof(overlay_target_hdl_t, oth_link));
+	overlay_thdl_idspace = id_space_create("overlay_target_minors",
+	    1, INT32_MAX);
+}
+
+void
+overlay_target_fini(void)
+{
+	id_space_destroy(overlay_thdl_idspace);
+	list_destroy(&overlay_thdl_list);
+	list_destroy(&overlay_target_list);
+	cv_destroy(&overlay_target_condvar);
+	mutex_destroy(&overlay_target_lock);
+	kmem_cache_destroy(overlay_entry_cache);
+	kmem_cache_destroy(overlay_target_cache);
+	ddi_soft_state_fini(&overlay_thdl_state);
+}
+
+void
+overlay_target_free(overlay_dev_t *odd)
+{
+	if (odd->odd_target == NULL)
+		return;
+
+	if (odd->odd_target->ott_mode == OVERLAY_TARGET_DYNAMIC) {
+		refhash_t *rp = odd->odd_target->ott_u.ott_dyn.ott_dhash;
+		avl_tree_t *ap = &odd->odd_target->ott_u.ott_dyn.ott_tree;
+		overlay_target_entry_t *ote;
+
+		/*
+		 * Our AVL tree and hashtable contain the same elements,
+		 * therefore we should just remove it from the tree, but then
+		 * delete the entries when we remove them from the hash table
+		 * (which happens through the refhash dtor).
+		 */
+		while ((ote = avl_first(ap)) != NULL)
+			avl_remove(ap, ote);
+
+		avl_destroy(ap);
+		for (ote = refhash_first(rp); ote != NULL;
+		    ote = refhash_next(rp, ote)) {
+			refhash_remove(rp, ote);
+		}
+		refhash_destroy(rp);
+	}
+
+	ASSERT(odd->odd_target->ott_ocount == 0);
+	kmem_cache_free(overlay_target_cache, odd->odd_target);
+}
+
+int
+overlay_target_busy()
+{
+	int ret;
+
+	mutex_enter(&overlay_target_lock);
+	ret = !list_is_empty(&overlay_thdl_list);
+	mutex_exit(&overlay_target_lock);
+
+	return (ret);
+}
+
+static void
+overlay_target_queue(overlay_target_entry_t *entry)
+{
+	mutex_enter(&overlay_target_lock);
+	mutex_enter(&entry->ote_ott->ott_lock);
+	if (entry->ote_ott->ott_flags & OVERLAY_T_TEARDOWN) {
+		mutex_exit(&entry->ote_ott->ott_lock);
+		mutex_exit(&overlay_target_lock);
+		return;
+	}
+	entry->ote_ott->ott_ocount++;
+	mutex_exit(&entry->ote_ott->ott_lock);
+	list_insert_tail(&overlay_target_list, entry);
+	cv_signal(&overlay_target_condvar);
+	mutex_exit(&overlay_target_lock);
+}
+
+void
+overlay_target_quiesce(overlay_target_t *ott)
+{
+	if (ott == NULL)
+		return;
+	mutex_enter(&ott->ott_lock);
+	ott->ott_flags |= OVERLAY_T_TEARDOWN;
+	while (ott->ott_ocount != 0)
+		cv_wait(&ott->ott_cond, &ott->ott_lock);
+	mutex_exit(&ott->ott_lock);
+}
+
+/*
+ * This functions assumes that the destination mode is OVERLAY_PLUGIN_D_IP |
+ * OVERLAY_PLUGIN_D_PORT. As we don't have an implementation of anything else at
+ * this time, say for NVGRE, we drop all packets that mcuh this.
+ */
+int
+overlay_target_lookup(overlay_dev_t *odd, mblk_t *mp, struct sockaddr *sock,
+    socklen_t *slenp)
+{
+	int ret;
+	struct sockaddr_in6 *v6;
+	overlay_target_t *ott;
+	mac_header_info_t mhi;
+	overlay_target_entry_t *entry;
+
+	ASSERT(odd->odd_target != NULL);
+
+	/*
+	 * At this point, the overlay device is in a mux which means that it's
+	 * been activated. At this point, parts of the target, such as the mode
+	 * and the destination are now read-only and we don't have to worry
+	 * about synchronization for them.
+	 */
+	ott = odd->odd_target;
+	if (ott->ott_dest != (OVERLAY_PLUGIN_D_IP | OVERLAY_PLUGIN_D_PORT))
+		return (OVERLAY_TARGET_DROP);
+
+	v6 = (struct sockaddr_in6 *)sock;
+	bzero(v6, sizeof (struct sockaddr_in6));
+	v6->sin6_family = AF_INET6;
+
+	if (ott->ott_mode == OVERLAY_TARGET_POINT) {
+		mutex_enter(&ott->ott_lock);
+		bcopy(&ott->ott_u.ott_point.otp_ip, &v6->sin6_addr,
+		    sizeof (struct in6_addr));
+		v6->sin6_port = htons(ott->ott_u.ott_point.otp_port);
+		mutex_exit(&ott->ott_lock);
+		*slenp = sizeof (struct sockaddr_in6);
+
+		return (OVERLAY_TARGET_OK);
+	}
+
+	ASSERT(ott->ott_mode == OVERLAY_TARGET_DYNAMIC);
+
+	/*
+	 * Note we only want the MAC address here, therefore we won't bother
+	 * using mac_vlan_header_info(). If any caller needs the vlan info at
+	 * this point, this should change to a call to mac_vlan_header_info().
+	 */
+	if (mac_header_info(odd->odd_mh, mp, &mhi) != 0)
+		return (OVERLAY_TARGET_DROP);
+	mutex_enter(&ott->ott_lock);
+	entry = refhash_lookup(ott->ott_u.ott_dyn.ott_dhash,
+	    mhi.mhi_daddr);
+	if (entry == NULL) {
+		entry = kmem_cache_alloc(overlay_entry_cache,
+		    KM_NOSLEEP | KM_NORMALPRI);
+		if (entry == NULL) {
+			mutex_exit(&ott->ott_lock);
+			return (OVERLAY_TARGET_DROP);
+		}
+		bcopy(mhi.mhi_daddr, entry->ote_addr, ETHERADDRL);
+		entry->ote_chead = entry->ote_ctail = mp;
+		entry->ote_mbsize = msgsize(mp);
+		entry->ote_flags |= OVERLAY_ENTRY_F_PENDING;
+		entry->ote_ott = ott;
+		entry->ote_odd = odd;
+		refhash_insert(ott->ott_u.ott_dyn.ott_dhash, entry);
+		avl_add(&ott->ott_u.ott_dyn.ott_tree, entry);
+		mutex_exit(&ott->ott_lock);
+		overlay_target_queue(entry);
+		return (OVERLAY_TARGET_ASYNC);
+	}
+	refhash_hold(ott->ott_u.ott_dyn.ott_dhash, entry);
+	mutex_exit(&ott->ott_lock);
+
+	mutex_enter(&entry->ote_lock);
+	if (entry->ote_flags & OVERLAY_ENTRY_F_DROP) {
+		ret = OVERLAY_TARGET_DROP;
+	} else if (entry->ote_flags & OVERLAY_ENTRY_F_VALID) {
+		bcopy(&entry->ote_dest.otp_ip, &v6->sin6_addr,
+		    sizeof (struct in6_addr));
+		v6->sin6_port = htons(entry->ote_dest.otp_port);
+		*slenp = sizeof (struct sockaddr_in6);
+		ret = OVERLAY_TARGET_OK;
+	} else {
+		size_t mlen = msgsize(mp);
+
+		if (mlen + entry->ote_mbsize > overlay_ent_size) {
+			ret = OVERLAY_TARGET_DROP;
+		} else {
+			if (entry->ote_ctail != NULL) {
+				ASSERT(entry->ote_ctail->b_next ==
+				    NULL);
+				entry->ote_ctail->b_next = mp;
+				entry->ote_ctail = mp;
+			} else {
+				entry->ote_chead = mp;
+				entry->ote_ctail = mp;
+			}
+			entry->ote_mbsize += mlen;
+			if ((entry->ote_flags &
+			    OVERLAY_ENTRY_F_PENDING) == 0) {
+				entry->ote_flags |=
+				    OVERLAY_ENTRY_F_PENDING;
+				overlay_target_queue(entry);
+			}
+			ret = OVERLAY_TARGET_ASYNC;
+		}
+	}
+	mutex_exit(&entry->ote_lock);
+
+	mutex_enter(&ott->ott_lock);
+	refhash_rele(ott->ott_u.ott_dyn.ott_dhash, entry);
+	mutex_exit(&ott->ott_lock);
+
+	return (ret);
+}
+
+/* ARGSUSED */
+static int
+overlay_target_info(overlay_target_hdl_t *thdl, void *arg)
+{
+	overlay_dev_t *odd;
+	overlay_targ_info_t *oti = arg;
+
+	odd = overlay_hold_by_dlid(oti->oti_linkid);
+	if (odd == NULL)
+		return (ENOENT);
+
+	mutex_enter(&odd->odd_lock);
+	oti->oti_flags = 0;
+	oti->oti_needs = odd->odd_plugin->ovp_dest;
+	if (odd->odd_flags & OVERLAY_F_DEGRADED)
+		oti->oti_flags |= OVERLAY_TARG_INFO_F_DEGRADED;
+	if (odd->odd_flags & OVERLAY_F_ACTIVATED)
+		oti->oti_flags |= OVERLAY_TARG_INFO_F_ACTIVE;
+	oti->oti_vnetid = odd->odd_vid;
+	mutex_exit(&odd->odd_lock);
+	overlay_hold_rele(odd);
+	return (0);
+}
+
+/* ARGSUSED */
+static int
+overlay_target_associate(overlay_target_hdl_t *thdl, void *arg)
+{
+	overlay_dev_t *odd;
+	overlay_target_t *ott;
+	overlay_targ_associate_t *ota = arg;
+
+	odd = overlay_hold_by_dlid(ota->ota_linkid);
+	if (odd == NULL)
+		return (ENOENT);
+
+	if (ota->ota_id == 0) {
+		overlay_hold_rele(odd);
+		return (EINVAL);
+	}
+
+	if (ota->ota_mode != OVERLAY_TARGET_POINT &&
+	    ota->ota_mode != OVERLAY_TARGET_DYNAMIC) {
+		overlay_hold_rele(odd);
+		return (EINVAL);
+	}
+
+	if (ota->ota_provides != odd->odd_plugin->ovp_dest) {
+		overlay_hold_rele(odd);
+		return (EINVAL);
+	}
+
+	if (ota->ota_mode == OVERLAY_TARGET_POINT) {
+		if (ota->ota_provides & OVERLAY_PLUGIN_D_IP) {
+			if (IN6_IS_ADDR_UNSPECIFIED(&ota->ota_point.otp_ip) ||
+			    IN6_IS_ADDR_V4COMPAT(&ota->ota_point.otp_ip) ||
+			    IN6_IS_ADDR_V4MAPPED_ANY(&ota->ota_point.otp_ip)) {
+				overlay_hold_rele(odd);
+				return (EINVAL);
+			}
+		}
+
+		if (ota->ota_provides & OVERLAY_PLUGIN_D_PORT) {
+			if (ota->ota_point.otp_port == 0) {
+				overlay_hold_rele(odd);
+				return (EINVAL);
+			}
+		}
+	}
+
+	ott = kmem_cache_alloc(overlay_target_cache, KM_SLEEP);
+	ott->ott_flags = 0;
+	ott->ott_ocount = 0;
+	ott->ott_mode = ota->ota_mode;
+	ott->ott_dest = ota->ota_provides;
+	ott->ott_id = ota->ota_id;
+
+	if (ott->ott_mode == OVERLAY_TARGET_POINT) {
+		bcopy(&ota->ota_point, &ott->ott_u.ott_point,
+		    sizeof (overlay_target_point_t));
+	} else {
+		ott->ott_u.ott_dyn.ott_dhash = refhash_create(OVERLAY_HSIZE,
+		    overlay_mac_hash, overlay_mac_cmp,
+		    overlay_target_entry_dtor, sizeof (overlay_target_entry_t),
+		    offsetof(overlay_target_entry_t, ote_reflink),
+		    offsetof(overlay_target_entry_t, ote_addr), KM_SLEEP);
+		avl_create(&ott->ott_u.ott_dyn.ott_tree, overlay_mac_avl,
+		    sizeof (overlay_target_entry_t),
+		    offsetof(overlay_target_entry_t, ote_avllink));
+	}
+	mutex_enter(&odd->odd_lock);
+	if (odd->odd_flags & OVERLAY_F_VARPD) {
+		mutex_exit(&odd->odd_lock);
+		kmem_cache_free(overlay_target_cache, ott);
+		overlay_hold_rele(odd);
+		return (EEXIST);
+	}
+
+	odd->odd_flags |= OVERLAY_F_VARPD;
+	odd->odd_target = ott;
+	mutex_exit(&odd->odd_lock);
+
+	/*
+	 * Now that we've successfully integrated ourselves here, we should note
+	 * that the link state is now up, and transition it away from UKNOWN.
+	 */
+	mac_link_update(odd->odd_mh, LINK_STATE_UP);
+
+	overlay_hold_rele(odd);
+
+
+	return (0);
+}
+
+
+/* ARGSUSED */
+static int
+overlay_target_degrade(overlay_target_hdl_t *thdl, void *arg)
+{
+	overlay_dev_t *odd;
+	overlay_targ_degrade_t *otd = arg;
+
+	odd = overlay_hold_by_dlid(otd->otd_linkid);
+	if (odd == NULL)
+		return (ENOENT);
+
+	overlay_fm_degrade(odd, otd->otd_buf);
+	overlay_hold_rele(odd);
+	return (0);
+}
+
+/* ARGSUSED */
+static int
+overlay_target_restore(overlay_target_hdl_t *thdl, void *arg)
+{
+	overlay_dev_t *odd;
+	overlay_targ_id_t *otid = arg;
+
+	odd = overlay_hold_by_dlid(otid->otid_linkid);
+	if (odd == NULL)
+		return (ENOENT);
+
+	overlay_fm_restore(odd);
+	overlay_hold_rele(odd);
+	return (0);
+}
+
+/* ARGSUSED */
+static int
+overlay_target_disassociate(overlay_target_hdl_t *thdl, void *arg)
+{
+	overlay_dev_t *odd;
+	overlay_targ_id_t *otid = arg;
+
+	odd = overlay_hold_by_dlid(otid->otid_linkid);
+	if (odd == NULL)
+		return (ENOENT);
+
+	mutex_enter(&odd->odd_lock);
+	odd->odd_flags &= ~OVERLAY_F_VARPD;
+	mutex_exit(&odd->odd_lock);
+
+	/*
+	 * Without a varpd instance, we should consider ourselves link down.
+	 */
+	mac_link_update(odd->odd_mh, LINK_STATE_DOWN);
+
+	overlay_hold_rele(odd);
+	return (0);
+
+}
+
+static int
+overlay_target_lookup_request(overlay_target_hdl_t *thdl, void *arg)
+{
+	overlay_targ_lookup_t *otl = arg;
+	overlay_target_entry_t *entry;
+	clock_t ret, timeout;
+	mac_header_info_t mhi;
+
+	timeout = ddi_get_lbolt() + drv_usectohz(MICROSEC);
+again:
+	mutex_enter(&overlay_target_lock);
+	while (list_is_empty(&overlay_target_list)) {
+		ret = cv_timedwait(&overlay_target_condvar,
+		    &overlay_target_lock, timeout);
+		if (ret == -1) {
+			mutex_exit(&overlay_target_lock);
+			return (ETIME);
+		}
+	}
+	entry = list_remove_head(&overlay_target_list);
+	mutex_exit(&overlay_target_lock);
+	mutex_enter(&entry->ote_lock);
+	if (entry->ote_flags & OVERLAY_ENTRY_F_VALID) {
+		ASSERT(entry->ote_chead == NULL);
+		mutex_exit(&entry->ote_lock);
+		goto again;
+	}
+	ASSERT(entry->ote_chead != NULL);
+
+	/*
+	 * If we have a bogon that doesn't have a valid mac header, drop it and
+	 * try again.
+	 */
+	if (mac_vlan_header_info(entry->ote_odd->odd_mh, entry->ote_chead,
+	    &mhi) != 0) {
+		boolean_t queue = B_FALSE;
+		mblk_t *mp = entry->ote_chead;
+		entry->ote_chead = mp->b_next;
+		mp->b_next = NULL;
+		if (entry->ote_ctail == mp)
+			entry->ote_ctail = entry->ote_chead;
+		entry->ote_mbsize -= msgsize(mp);
+		if (entry->ote_chead != NULL)
+			queue = B_TRUE;
+		mutex_exit(&entry->ote_lock);
+		if (queue == B_TRUE)
+			overlay_target_queue(entry);
+		freemsg(mp);
+		goto again;
+	}
+
+	otl->otl_dlid = entry->ote_odd->odd_linkid;
+	otl->otl_reqid = (uintptr_t)entry;
+	otl->otl_varpdid = entry->ote_ott->ott_id;
+	otl->otl_vnetid = entry->ote_odd->odd_vid;
+
+	otl->otl_hdrsize = mhi.mhi_hdrsize;
+	otl->otl_pktsize = msgsize(entry->ote_chead) - otl->otl_hdrsize;
+	bcopy(mhi.mhi_daddr, otl->otl_dstaddr, ETHERADDRL);
+	bcopy(mhi.mhi_saddr, otl->otl_srcaddr, ETHERADDRL);
+	otl->otl_dsttype = mhi.mhi_dsttype;
+	otl->otl_sap = mhi.mhi_bindsap;
+	otl->otl_vlan = VLAN_ID(mhi.mhi_tci);
+	mutex_exit(&entry->ote_lock);
+
+	mutex_enter(&thdl->oth_lock);
+	list_insert_tail(&thdl->oth_outstanding, entry);
+	mutex_exit(&thdl->oth_lock);
+
+	return (0);
+}
+
+static int
+overlay_target_lookup_respond(overlay_target_hdl_t *thdl, void *arg)
+{
+	const overlay_targ_resp_t *otr = arg;
+	overlay_target_entry_t *entry;
+	mblk_t *mp;
+
+	mutex_enter(&thdl->oth_lock);
+	for (entry = list_head(&thdl->oth_outstanding); entry != NULL;
+	    entry = list_next(&thdl->oth_outstanding, entry)) {
+		if ((uintptr_t)entry == otr->otr_reqid)
+			break;
+	}
+
+	if (entry == NULL) {
+		mutex_exit(&thdl->oth_lock);
+		return (EINVAL);
+	}
+	list_remove(&thdl->oth_outstanding, entry);
+	mutex_exit(&thdl->oth_lock);
+
+	mutex_enter(&entry->ote_lock);
+	bcopy(&otr->otr_answer, &entry->ote_dest,
+	    sizeof (overlay_target_point_t));
+	entry->ote_flags &= ~OVERLAY_ENTRY_F_PENDING;
+	entry->ote_flags |= OVERLAY_ENTRY_F_VALID;
+	mp = entry->ote_chead;
+	entry->ote_chead = NULL;
+	entry->ote_ctail = NULL;
+	entry->ote_mbsize = 0;
+	entry->ote_vtime = gethrtime();
+	mutex_exit(&entry->ote_lock);
+
+	/*
+	 * For now do an in-situ drain.
+	 */
+	mp = overlay_m_tx(entry->ote_odd, mp);
+	freemsgchain(mp);
+
+	mutex_enter(&entry->ote_ott->ott_lock);
+	entry->ote_ott->ott_ocount--;
+	cv_signal(&entry->ote_ott->ott_cond);
+	mutex_exit(&entry->ote_ott->ott_lock);
+
+	return (0);
+}
+
+static int
+overlay_target_lookup_drop(overlay_target_hdl_t *thdl, void *arg)
+{
+	const overlay_targ_resp_t *otr = arg;
+	overlay_target_entry_t *entry;
+	mblk_t *mp;
+	boolean_t queue = B_FALSE;
+
+	mutex_enter(&thdl->oth_lock);
+	for (entry = list_head(&thdl->oth_outstanding); entry != NULL;
+	    entry = list_next(&thdl->oth_outstanding, entry)) {
+		if ((uintptr_t)entry == otr->otr_reqid)
+			break;
+	}
+
+	if (entry == NULL) {
+		mutex_exit(&thdl->oth_lock);
+		return (EINVAL);
+	}
+	list_remove(&thdl->oth_outstanding, entry);
+	mutex_exit(&thdl->oth_lock);
+
+	mutex_enter(&entry->ote_lock);
+
+	/* Safeguard against a confused varpd */
+	if (entry->ote_flags & OVERLAY_ENTRY_F_VALID) {
+		entry->ote_flags &= ~OVERLAY_ENTRY_F_PENDING;
+		DTRACE_PROBE1(overlay__target__valid__drop,
+		    overlay_target_entry_t *, entry);
+		mutex_exit(&entry->ote_lock);
+		goto done;
+	}
+
+	mp = entry->ote_chead;
+	if (mp != NULL) {
+		entry->ote_chead = mp->b_next;
+		mp->b_next = NULL;
+		if (entry->ote_ctail == mp)
+			entry->ote_ctail = entry->ote_chead;
+		entry->ote_mbsize -= msgsize(mp);
+	}
+	if (entry->ote_chead != NULL) {
+		queue = B_TRUE;
+		entry->ote_flags |= OVERLAY_ENTRY_F_PENDING;
+	} else {
+		entry->ote_flags &= ~OVERLAY_ENTRY_F_PENDING;
+	}
+	mutex_exit(&entry->ote_lock);
+
+	if (queue == B_TRUE)
+		overlay_target_queue(entry);
+	freemsg(mp);
+
+done:
+	mutex_enter(&entry->ote_ott->ott_lock);
+	entry->ote_ott->ott_ocount--;
+	cv_signal(&entry->ote_ott->ott_cond);
+	mutex_exit(&entry->ote_ott->ott_lock);
+
+	return (0);
+}
+
+/* ARGSUSED */
+static int
+overlay_target_pkt_copyin(const void *ubuf, void **outp, size_t *bsize,
+    int flags)
+{
+	overlay_targ_pkt_t *pkt;
+	overlay_targ_pkt32_t *pkt32;
+
+	pkt = kmem_alloc(sizeof (overlay_targ_pkt_t), KM_SLEEP);
+	*outp = pkt;
+	*bsize = sizeof (overlay_targ_pkt_t);
+	if (ddi_model_convert_from(flags & FMODELS) == DDI_MODEL_ILP32) {
+		uintptr_t addr;
+
+		if (ddi_copyin(ubuf, pkt, sizeof (overlay_targ_pkt32_t),
+		    flags & FKIOCTL) != 0) {
+			kmem_free(pkt, *bsize);
+			return (EFAULT);
+		}
+		pkt32 = (overlay_targ_pkt32_t *)pkt;
+		addr = pkt32->otp_buf;
+		pkt->otp_buf = (void *)addr;
+	} else {
+		if (ddi_copyin(ubuf, pkt, *bsize, flags & FKIOCTL) != 0) {
+			kmem_free(pkt, *bsize);
+			return (EFAULT);
+		}
+	}
+	return (0);
+}
+
+static int
+overlay_target_pkt_copyout(void *ubuf, void *buf, size_t bufsize,
+    int flags)
+{
+	if (ddi_model_convert_from(flags & FMODELS) == DDI_MODEL_ILP32) {
+		overlay_targ_pkt_t *pkt = buf;
+		overlay_targ_pkt32_t *pkt32 = buf;
+		uintptr_t addr = (uintptr_t)pkt->otp_buf;
+		pkt32->otp_buf = (caddr32_t)addr;
+		if (ddi_copyout(buf, ubuf, sizeof (overlay_targ_pkt32_t),
+		    flags & FKIOCTL) != 0)
+			return (EFAULT);
+	} else {
+		if (ddi_copyout(buf, ubuf, bufsize, flags & FKIOCTL) != 0)
+			return (EFAULT);
+	}
+	return (0);
+}
+
+static int
+overlay_target_packet(overlay_target_hdl_t *thdl, void *arg)
+{
+	overlay_targ_pkt_t *pkt = arg;
+	overlay_target_entry_t *entry;
+	mblk_t *mp;
+	size_t mlen;
+	size_t boff;
+
+	mutex_enter(&thdl->oth_lock);
+	for (entry = list_head(&thdl->oth_outstanding); entry != NULL;
+	    entry = list_next(&thdl->oth_outstanding, entry)) {
+		if ((uintptr_t)entry == pkt->otp_reqid)
+			break;
+	}
+
+	if (entry == NULL) {
+		mutex_exit(&thdl->oth_lock);
+		return (EINVAL);
+	}
+	mutex_enter(&entry->ote_lock);
+	mutex_exit(&thdl->oth_lock);
+	mp = entry->ote_chead;
+	/* Protect against a rogue varpd */
+	if (mp == NULL) {
+		mutex_exit(&entry->ote_lock);
+		return (EINVAL);
+	}
+	mlen = MIN(msgsize(mp), pkt->otp_size);
+	pkt->otp_size = mlen;
+	boff = 0;
+	while (mlen > 0) {
+		size_t wlen = MIN(MBLKL(mp), mlen);
+		if (ddi_copyout(mp->b_rptr,
+		    (void *)((uintptr_t)pkt->otp_buf + boff),
+		    wlen, 0) != 0) {
+			mutex_exit(&entry->ote_lock);
+			return (EFAULT);
+		}
+		mlen -= wlen;
+		boff += wlen;
+		mp = mp->b_cont;
+	}
+	mutex_exit(&entry->ote_lock);
+	return (0);
+}
+
+static int
+overlay_target_inject(overlay_target_hdl_t *thdl, void *arg)
+{
+	overlay_targ_pkt_t *pkt = arg;
+	overlay_target_entry_t *entry;
+	overlay_dev_t *odd;
+	mblk_t *mp;
+
+	if (pkt->otp_size > ETHERMAX + VLAN_TAGSZ)
+		return (EINVAL);
+
+	mp = allocb(pkt->otp_size, 0);
+	if (mp == NULL)
+		return (ENOMEM);
+
+	if (ddi_copyin(pkt->otp_buf, mp->b_rptr, pkt->otp_size, 0) != 0) {
+		freeb(mp);
+		return (EFAULT);
+	}
+	mp->b_wptr += pkt->otp_size;
+
+	if (pkt->otp_linkid != UINT64_MAX) {
+		odd = overlay_hold_by_dlid(pkt->otp_linkid);
+		if (odd == NULL) {
+			freeb(mp);
+			return (ENOENT);
+		}
+	} else {
+		mutex_enter(&thdl->oth_lock);
+		for (entry = list_head(&thdl->oth_outstanding); entry != NULL;
+		    entry = list_next(&thdl->oth_outstanding, entry)) {
+			if ((uintptr_t)entry == pkt->otp_reqid)
+				break;
+		}
+
+		if (entry == NULL) {
+			mutex_exit(&thdl->oth_lock);
+			freeb(mp);
+			return (ENOENT);
+		}
+		odd = entry->ote_odd;
+		mutex_exit(&thdl->oth_lock);
+	}
+
+	mutex_enter(&odd->odd_lock);
+	overlay_io_start(odd, OVERLAY_F_IN_RX);
+	mutex_exit(&odd->odd_lock);
+
+	mac_rx(odd->odd_mh, NULL, mp);
+
+	mutex_enter(&odd->odd_lock);
+	overlay_io_done(odd, OVERLAY_F_IN_RX);
+	mutex_exit(&odd->odd_lock);
+
+	return (0);
+}
+
+static int
+overlay_target_resend(overlay_target_hdl_t *thdl, void *arg)
+{
+	overlay_targ_pkt_t *pkt = arg;
+	overlay_target_entry_t *entry;
+	overlay_dev_t *odd;
+	mblk_t *mp;
+
+	if (pkt->otp_size > ETHERMAX + VLAN_TAGSZ)
+		return (EINVAL);
+
+	mp = allocb(pkt->otp_size, 0);
+	if (mp == NULL)
+		return (ENOMEM);
+
+	if (ddi_copyin(pkt->otp_buf, mp->b_rptr, pkt->otp_size, 0) != 0) {
+		freeb(mp);
+		return (EFAULT);
+	}
+	mp->b_wptr += pkt->otp_size;
+
+	if (pkt->otp_linkid != UINT64_MAX) {
+		odd = overlay_hold_by_dlid(pkt->otp_linkid);
+		if (odd == NULL) {
+			freeb(mp);
+			return (ENOENT);
+		}
+	} else {
+		mutex_enter(&thdl->oth_lock);
+		for (entry = list_head(&thdl->oth_outstanding); entry != NULL;
+		    entry = list_next(&thdl->oth_outstanding, entry)) {
+			if ((uintptr_t)entry == pkt->otp_reqid)
+				break;
+		}
+
+		if (entry == NULL) {
+			mutex_exit(&thdl->oth_lock);
+			freeb(mp);
+			return (ENOENT);
+		}
+		odd = entry->ote_odd;
+		mutex_exit(&thdl->oth_lock);
+	}
+
+	mp = overlay_m_tx(odd, mp);
+	freemsgchain(mp);
+
+	return (0);
+}
+
+typedef struct overlay_targ_list_int {
+	boolean_t	otli_count;
+	uint32_t	otli_cur;
+	uint32_t	otli_nents;
+	uint32_t	otli_ents[];
+} overlay_targ_list_int_t;
+
+static int
+overlay_target_list_copyin(const void *ubuf, void **outp, size_t *bsize,
+    int flags)
+{
+	overlay_targ_list_t n;
+	overlay_targ_list_int_t *otl;
+
+	if (ddi_copyin(ubuf, &n, sizeof (overlay_targ_list_t),
+	    flags & FKIOCTL) != 0)
+		return (EFAULT);
+
+	/*
+	 */
+	if (n.otl_nents >= INT32_MAX / sizeof (uint32_t))
+		return (EINVAL);
+	*bsize = sizeof (overlay_targ_list_int_t) +
+	    sizeof (uint32_t) * n.otl_nents;
+	otl = kmem_zalloc(*bsize, KM_SLEEP);
+	otl->otli_cur = 0;
+	otl->otli_nents = n.otl_nents;
+	if (otl->otli_nents != 0) {
+		otl->otli_count = B_FALSE;
+		if (ddi_copyin((void *)((uintptr_t)ubuf +
+		    offsetof(overlay_targ_list_t, otl_ents)),
+		    otl->otli_ents, n.otl_nents * sizeof (uint32_t),
+		    flags & FKIOCTL) != 0) {
+			kmem_free(otl, *bsize);
+			return (EFAULT);
+		}
+	} else {
+		otl->otli_count = B_TRUE;
+	}
+
+	*outp = otl;
+	return (0);
+}
+
+static int
+overlay_target_ioctl_list_cb(overlay_dev_t *odd, void *arg)
+{
+	overlay_targ_list_int_t *otl = arg;
+
+	if (otl->otli_cur < otl->otli_nents)
+		otl->otli_ents[otl->otli_cur] = odd->odd_linkid;
+	otl->otli_cur++;
+	return (0);
+}
+
+/* ARGSUSED */
+static int
+overlay_target_ioctl_list(overlay_target_hdl_t *thdl, void *arg)
+{
+	overlay_dev_iter(overlay_target_ioctl_list_cb, arg);
+	return (0);
+}
+
+/* ARGSUSED */
+static int
+overlay_target_list_copyout(void *ubuf, void *buf, size_t bufsize, int flags)
+{
+	overlay_targ_list_int_t *otl = buf;
+
+	if (ddi_copyout(&otl->otli_cur, ubuf, sizeof (uint32_t),
+	    flags & FKIOCTL) != 0)
+		return (EFAULT);
+
+	if (otl->otli_count == B_FALSE) {
+		if (ddi_copyout(otl->otli_ents,
+		    (void *)((uintptr_t)ubuf +
+		    offsetof(overlay_targ_list_t, otl_ents)),
+		    sizeof (uint32_t) * otl->otli_nents,
+		    flags & FKIOCTL) != 0)
+			return (EFAULT);
+	}
+	return (0);
+}
+
+/* ARGSUSED */
+static int
+overlay_target_cache_get(overlay_target_hdl_t *thdl, void *arg)
+{
+	int ret = 0;
+	overlay_dev_t *odd;
+	overlay_target_t *ott;
+	overlay_targ_cache_t *otc = arg;
+
+	odd = overlay_hold_by_dlid(otc->otc_linkid);
+	if (odd == NULL)
+		return (ENOENT);
+
+	mutex_enter(&odd->odd_lock);
+	if (!(odd->odd_flags & OVERLAY_F_VARPD)) {
+		mutex_exit(&odd->odd_lock);
+		overlay_hold_rele(odd);
+		return (ENXIO);
+	}
+	ott = odd->odd_target;
+	if (ott->ott_mode != OVERLAY_TARGET_POINT &&
+	    ott->ott_mode != OVERLAY_TARGET_DYNAMIC) {
+		mutex_exit(&odd->odd_lock);
+		overlay_hold_rele(odd);
+		return (ENOTSUP);
+	}
+	mutex_enter(&ott->ott_lock);
+	mutex_exit(&odd->odd_lock);
+
+	if (ott->ott_mode == OVERLAY_TARGET_POINT) {
+		otc->otc_entry.otce_flags = 0;
+		bcopy(&ott->ott_u.ott_point, &otc->otc_entry.otce_dest,
+		    sizeof (overlay_target_point_t));
+	} else {
+		overlay_target_entry_t *ote;
+		ote = refhash_lookup(ott->ott_u.ott_dyn.ott_dhash,
+		    otc->otc_entry.otce_mac);
+		if (ote != NULL) {
+			mutex_enter(&ote->ote_lock);
+			if ((ote->ote_flags &
+			    OVERLAY_ENTRY_F_VALID_MASK) != 0) {
+				if (ote->ote_flags & OVERLAY_ENTRY_F_DROP) {
+					otc->otc_entry.otce_flags =
+					    OVERLAY_TARGET_CACHE_DROP;
+				} else {
+					otc->otc_entry.otce_flags = 0;
+					bcopy(&ote->ote_dest,
+					    &otc->otc_entry.otce_dest,
+					    sizeof (overlay_target_point_t));
+				}
+				ret = 0;
+			} else {
+				ret = ENOENT;
+			}
+			mutex_exit(&ote->ote_lock);
+		} else {
+			ret = ENOENT;
+		}
+	}
+
+	mutex_exit(&ott->ott_lock);
+	overlay_hold_rele(odd);
+
+	return (ret);
+}
+
+/* ARGSUSED */
+static int
+overlay_target_cache_set(overlay_target_hdl_t *thdl, void *arg)
+{
+	overlay_dev_t *odd;
+	overlay_target_t *ott;
+	overlay_target_entry_t *ote;
+	overlay_targ_cache_t *otc = arg;
+	mblk_t *mp = NULL;
+
+	if (otc->otc_entry.otce_flags & ~OVERLAY_TARGET_CACHE_DROP)
+		return (EINVAL);
+
+	odd = overlay_hold_by_dlid(otc->otc_linkid);
+	if (odd == NULL)
+		return (ENOENT);
+
+	mutex_enter(&odd->odd_lock);
+	if (!(odd->odd_flags & OVERLAY_F_VARPD)) {
+		mutex_exit(&odd->odd_lock);
+		overlay_hold_rele(odd);
+		return (ENXIO);
+	}
+	ott = odd->odd_target;
+	if (ott->ott_mode != OVERLAY_TARGET_DYNAMIC) {
+		mutex_exit(&odd->odd_lock);
+		overlay_hold_rele(odd);
+		return (ENOTSUP);
+	}
+	mutex_enter(&ott->ott_lock);
+	mutex_exit(&odd->odd_lock);
+
+	ote = refhash_lookup(ott->ott_u.ott_dyn.ott_dhash,
+	    otc->otc_entry.otce_mac);
+	if (ote == NULL) {
+		ote = kmem_cache_alloc(overlay_entry_cache, KM_SLEEP);
+		bcopy(otc->otc_entry.otce_mac, ote->ote_addr, ETHERADDRL);
+		ote->ote_chead = ote->ote_ctail = NULL;
+		ote->ote_mbsize = 0;
+		ote->ote_ott = ott;
+		ote->ote_odd = odd;
+		mutex_enter(&ote->ote_lock);
+		refhash_insert(ott->ott_u.ott_dyn.ott_dhash, ote);
+		avl_add(&ott->ott_u.ott_dyn.ott_tree, ote);
+	} else {
+		mutex_enter(&ote->ote_lock);
+	}
+
+	if (otc->otc_entry.otce_flags & OVERLAY_TARGET_CACHE_DROP) {
+		ote->ote_flags |= OVERLAY_ENTRY_F_DROP;
+	} else {
+		ote->ote_flags |= OVERLAY_ENTRY_F_VALID;
+		bcopy(&otc->otc_entry.otce_dest, &ote->ote_dest,
+		    sizeof (overlay_target_point_t));
+		mp = ote->ote_chead;
+		ote->ote_chead = NULL;
+		ote->ote_ctail = NULL;
+		ote->ote_mbsize = 0;
+		ote->ote_vtime = gethrtime();
+	}
+
+	mutex_exit(&ote->ote_lock);
+	mutex_exit(&ott->ott_lock);
+
+	if (mp != NULL) {
+		mp = overlay_m_tx(ote->ote_odd, mp);
+		freemsgchain(mp);
+	}
+
+	overlay_hold_rele(odd);
+
+	return (0);
+}
+
+/* ARGSUSED */
+static int
+overlay_target_cache_remove(overlay_target_hdl_t *thdl, void *arg)
+{
+	int ret = 0;
+	overlay_dev_t *odd;
+	overlay_target_t *ott;
+	overlay_target_entry_t *ote;
+	overlay_targ_cache_t *otc = arg;
+
+	odd = overlay_hold_by_dlid(otc->otc_linkid);
+	if (odd == NULL)
+		return (ENOENT);
+
+	mutex_enter(&odd->odd_lock);
+	if (!(odd->odd_flags & OVERLAY_F_VARPD)) {
+		mutex_exit(&odd->odd_lock);
+		overlay_hold_rele(odd);
+		return (ENXIO);
+	}
+	ott = odd->odd_target;
+	if (ott->ott_mode != OVERLAY_TARGET_DYNAMIC) {
+		mutex_exit(&odd->odd_lock);
+		overlay_hold_rele(odd);
+		return (ENOTSUP);
+	}
+	mutex_enter(&ott->ott_lock);
+	mutex_exit(&odd->odd_lock);
+
+	ote = refhash_lookup(ott->ott_u.ott_dyn.ott_dhash,
+	    otc->otc_entry.otce_mac);
+	if (ote != NULL) {
+		mutex_enter(&ote->ote_lock);
+		ote->ote_flags &= ~OVERLAY_ENTRY_F_VALID_MASK;
+		mutex_exit(&ote->ote_lock);
+		ret = 0;
+	} else {
+		ret = ENOENT;
+	}
+
+	mutex_exit(&ott->ott_lock);
+	overlay_hold_rele(odd);
+
+	return (ret);
+}
+
+/* ARGSUSED */
+static int
+overlay_target_cache_flush(overlay_target_hdl_t *thdl, void *arg)
+{
+	avl_tree_t *avl;
+	overlay_dev_t *odd;
+	overlay_target_t *ott;
+	overlay_target_entry_t *ote;
+	overlay_targ_cache_t *otc = arg;
+
+	odd = overlay_hold_by_dlid(otc->otc_linkid);
+	if (odd == NULL)
+		return (ENOENT);
+
+	mutex_enter(&odd->odd_lock);
+	if (!(odd->odd_flags & OVERLAY_F_VARPD)) {
+		mutex_exit(&odd->odd_lock);
+		overlay_hold_rele(odd);
+		return (ENXIO);
+	}
+	ott = odd->odd_target;
+	if (ott->ott_mode != OVERLAY_TARGET_DYNAMIC) {
+		mutex_exit(&odd->odd_lock);
+		overlay_hold_rele(odd);
+		return (ENOTSUP);
+	}
+	mutex_enter(&ott->ott_lock);
+	mutex_exit(&odd->odd_lock);
+	avl = &ott->ott_u.ott_dyn.ott_tree;
+
+	for (ote = avl_first(avl); ote != NULL; ote = AVL_NEXT(avl, ote)) {
+		mutex_enter(&ote->ote_lock);
+		ote->ote_flags &= ~OVERLAY_ENTRY_F_VALID_MASK;
+		mutex_exit(&ote->ote_lock);
+	}
+	ote = refhash_lookup(ott->ott_u.ott_dyn.ott_dhash,
+	    otc->otc_entry.otce_mac);
+
+	mutex_exit(&ott->ott_lock);
+	overlay_hold_rele(odd);
+
+	return (0);
+}
+
+static int
+overlay_target_cache_iter_copyin(const void *ubuf, void **outp, size_t *bsize,
+    int flags)
+{
+	overlay_targ_cache_iter_t base, *iter;
+
+	if (ddi_copyin(ubuf, &base, sizeof (overlay_targ_cache_iter_t),
+	    flags & FKIOCTL) != 0)
+		return (EFAULT);
+
+	if (base.otci_count > OVERLAY_TARGET_ITER_MAX)
+		return (E2BIG);
+
+	if (base.otci_count == 0)
+		return (EINVAL);
+
+	*bsize = sizeof (overlay_targ_cache_iter_t) +
+	    base.otci_count * sizeof (overlay_targ_cache_entry_t);
+	iter = kmem_alloc(*bsize, KM_SLEEP);
+	bcopy(&base, iter, sizeof (overlay_targ_cache_iter_t));
+	*outp = iter;
+
+	return (0);
+}
+
+typedef struct overlay_targ_cache_marker {
+	uint8_t		otcm_mac[ETHERADDRL];
+	uint16_t	otcm_done;
+} overlay_targ_cache_marker_t;
+
+/* ARGSUSED */
+static int
+overlay_target_cache_iter(overlay_target_hdl_t *thdl, void *arg)
+{
+	overlay_dev_t *odd;
+	overlay_target_t *ott;
+	overlay_target_entry_t lookup, *ent;
+	overlay_targ_cache_marker_t *mark;
+	avl_index_t where;
+	avl_tree_t *avl;
+	uint16_t written = 0;
+
+	overlay_targ_cache_iter_t *iter = arg;
+	mark = (void *)&iter->otci_marker;
+
+	if (mark->otcm_done != 0) {
+		iter->otci_count = 0;
+		return (0);
+	}
+
+	odd = overlay_hold_by_dlid(iter->otci_linkid);
+	if (odd == NULL)
+		return (ENOENT);
+
+	mutex_enter(&odd->odd_lock);
+	if (!(odd->odd_flags & OVERLAY_F_VARPD)) {
+		mutex_exit(&odd->odd_lock);
+		overlay_hold_rele(odd);
+		return (ENXIO);
+	}
+	ott = odd->odd_target;
+	if (ott->ott_mode != OVERLAY_TARGET_DYNAMIC &&
+	    ott->ott_mode != OVERLAY_TARGET_POINT) {
+		mutex_exit(&odd->odd_lock);
+		overlay_hold_rele(odd);
+		return (ENOTSUP);
+	}
+
+	/*
+	 * Holding this lock across the entire iteration probably isn't very
+	 * good. We should perhaps add an r/w lock for the avl tree. But we'll
+	 * wait until we now it's necessary before we do more.
+	 */
+	mutex_enter(&ott->ott_lock);
+	mutex_exit(&odd->odd_lock);
+
+	if (ott->ott_mode == OVERLAY_TARGET_POINT) {
+		overlay_targ_cache_entry_t *out = &iter->otci_ents[0];
+		bzero(out->otce_mac, ETHERADDRL);
+		out->otce_flags = 0;
+		bcopy(&ott->ott_u.ott_point, &out->otce_dest,
+		    sizeof (overlay_target_point_t));
+		written++;
+		mark->otcm_done = 1;
+	}
+
+	avl = &ott->ott_u.ott_dyn.ott_tree;
+	bcopy(mark->otcm_mac, lookup.ote_addr, ETHERADDRL);
+	ent = avl_find(avl, &lookup, &where);
+
+	/*
+	 * NULL ent means that the entry does not exist, so we want to start
+	 * with the closest node in the tree. This means that we implicitly rely
+	 * on the tree's order and the first node will be the mac 00:00:00:00:00
+	 * and the last will be ff:ff:ff:ff:ff:ff.
+	 */
+	if (ent == NULL) {
+		ent = avl_nearest(avl, where, AVL_AFTER);
+		if (ent == NULL) {
+			mark->otcm_done = 1;
+			goto done;
+		}
+	}
+
+	for (; ent != NULL && written < iter->otci_count;
+	    ent = AVL_NEXT(avl, ent)) {
+		overlay_targ_cache_entry_t *out = &iter->otci_ents[written];
+		mutex_enter(&ent->ote_lock);
+		if ((ent->ote_flags & OVERLAY_ENTRY_F_VALID_MASK) == 0) {
+			mutex_exit(&ent->ote_lock);
+			continue;
+		}
+		bcopy(ent->ote_addr, out->otce_mac, ETHERADDRL);
+		out->otce_flags = 0;
+		if (ent->ote_flags & OVERLAY_ENTRY_F_DROP)
+			out->otce_flags |= OVERLAY_TARGET_CACHE_DROP;
+		if (ent->ote_flags & OVERLAY_ENTRY_F_VALID)
+			bcopy(&ent->ote_dest, &out->otce_dest,
+			    sizeof (overlay_target_point_t));
+		written++;
+		mutex_exit(&ent->ote_lock);
+	}
+
+	if (ent != NULL) {
+		bcopy(ent->ote_addr, mark->otcm_mac, ETHERADDRL);
+	} else {
+		mark->otcm_done = 1;
+	}
+
+done:
+	iter->otci_count = written;
+	mutex_exit(&ott->ott_lock);
+	overlay_hold_rele(odd);
+
+	return (0);
+}
+
+/* ARGSUSED */
+static int
+overlay_target_cache_iter_copyout(void *ubuf, void *buf, size_t bufsize,
+    int flags)
+{
+	size_t outsize;
+	const overlay_targ_cache_iter_t *iter = buf;
+
+	outsize = sizeof (overlay_targ_cache_iter_t) +
+	    iter->otci_count * sizeof (overlay_targ_cache_entry_t);
+
+	if (ddi_copyout(buf, ubuf, outsize, flags & FKIOCTL) != 0)
+		return (EFAULT);
+
+	return (0);
+}
+
+static overlay_target_ioctl_t overlay_target_ioctab[] = {
+	{ OVERLAY_TARG_INFO, B_TRUE, B_TRUE,
+		NULL, overlay_target_info,
+		NULL, sizeof (overlay_targ_info_t)	},
+	{ OVERLAY_TARG_ASSOCIATE, B_TRUE, B_FALSE,
+		NULL, overlay_target_associate,
+		NULL, sizeof (overlay_targ_associate_t)	},
+	{ OVERLAY_TARG_DISASSOCIATE, B_TRUE, B_FALSE,
+		NULL, overlay_target_disassociate,
+		NULL, sizeof (overlay_targ_id_t)	},
+	{ OVERLAY_TARG_DEGRADE, B_TRUE, B_FALSE,
+		NULL, overlay_target_degrade,
+		NULL, sizeof (overlay_targ_degrade_t)	},
+	{ OVERLAY_TARG_RESTORE, B_TRUE, B_FALSE,
+		NULL, overlay_target_restore,
+		NULL, sizeof (overlay_targ_id_t)	},
+	{ OVERLAY_TARG_LOOKUP, B_FALSE, B_TRUE,
+		NULL, overlay_target_lookup_request,
+		NULL, sizeof (overlay_targ_lookup_t)	},
+	{ OVERLAY_TARG_RESPOND, B_TRUE, B_FALSE,
+		NULL, overlay_target_lookup_respond,
+		NULL, sizeof (overlay_targ_resp_t)	},
+	{ OVERLAY_TARG_DROP, B_TRUE, B_FALSE,
+		NULL, overlay_target_lookup_drop,
+		NULL, sizeof (overlay_targ_resp_t)	},
+	{ OVERLAY_TARG_PKT, B_TRUE, B_TRUE,
+		overlay_target_pkt_copyin,
+		overlay_target_packet,
+		overlay_target_pkt_copyout,
+		sizeof (overlay_targ_pkt_t)		},
+	{ OVERLAY_TARG_INJECT, B_TRUE, B_FALSE,
+		overlay_target_pkt_copyin,
+		overlay_target_inject,
+		NULL, sizeof (overlay_targ_pkt_t)	},
+	{ OVERLAY_TARG_RESEND, B_TRUE, B_FALSE,
+		overlay_target_pkt_copyin,
+		overlay_target_resend,
+		NULL, sizeof (overlay_targ_pkt_t)	},
+	{ OVERLAY_TARG_LIST, B_FALSE, B_TRUE,
+		overlay_target_list_copyin,
+		overlay_target_ioctl_list,
+		overlay_target_list_copyout,
+		sizeof (overlay_targ_list_t)		},
+	{ OVERLAY_TARG_CACHE_GET, B_FALSE, B_TRUE,
+		NULL, overlay_target_cache_get,
+		NULL, sizeof (overlay_targ_cache_t)	},
+	{ OVERLAY_TARG_CACHE_SET, B_TRUE, B_TRUE,
+		NULL, overlay_target_cache_set,
+		NULL, sizeof (overlay_targ_cache_t)	},
+	{ OVERLAY_TARG_CACHE_REMOVE, B_TRUE, B_TRUE,
+		NULL, overlay_target_cache_remove,
+		NULL, sizeof (overlay_targ_cache_t)	},
+	{ OVERLAY_TARG_CACHE_FLUSH, B_TRUE, B_TRUE,
+		NULL, overlay_target_cache_flush,
+		NULL, sizeof (overlay_targ_cache_t)	},
+	{ OVERLAY_TARG_CACHE_ITER, B_FALSE, B_TRUE,
+		overlay_target_cache_iter_copyin,
+		overlay_target_cache_iter,
+		overlay_target_cache_iter_copyout,
+		sizeof (overlay_targ_cache_iter_t)		},
+	{ 0 }
+};
+
+int
+overlay_target_open(dev_t *devp, int flags, int otype, cred_t *credp)
+{
+	minor_t mid;
+	overlay_target_hdl_t *thdl;
+
+	if (secpolicy_dl_config(credp) != 0)
+		return (EPERM);
+
+	if (getminor(*devp) != 0)
+		return (ENXIO);
+
+	if (otype & OTYP_BLK)
+		return (EINVAL);
+
+	if (flags & ~(FREAD | FWRITE | FEXCL))
+		return (EINVAL);
+
+	if ((flags & FWRITE) &&
+	    !(flags & FEXCL))
+		return (EINVAL);
+
+	if (!(flags & FREAD) && !(flags & FWRITE))
+		return (EINVAL);
+
+	if (crgetzoneid(credp) != GLOBAL_ZONEID)
+		return (EPERM);
+
+	mid = id_alloc(overlay_thdl_idspace);
+	if (ddi_soft_state_zalloc(overlay_thdl_state, mid) != 0) {
+		id_free(overlay_thdl_idspace, mid);
+		return (ENXIO);
+	}
+
+	thdl = ddi_get_soft_state(overlay_thdl_state, mid);
+	VERIFY(thdl != NULL);
+	thdl->oth_minor = mid;
+	thdl->oth_zoneid = crgetzoneid(credp);
+	thdl->oth_oflags = flags;
+	mutex_init(&thdl->oth_lock, NULL, MUTEX_DRIVER, NULL);
+	list_create(&thdl->oth_outstanding, sizeof (overlay_target_entry_t),
+	    offsetof(overlay_target_entry_t, ote_qlink));
+	*devp = makedevice(getmajor(*devp), mid);
+
+	mutex_enter(&overlay_target_lock);
+	if ((flags & FEXCL) && overlay_target_excl == B_TRUE) {
+		mutex_exit(&overlay_target_lock);
+		list_destroy(&thdl->oth_outstanding);
+		mutex_destroy(&thdl->oth_lock);
+		ddi_soft_state_free(overlay_thdl_state, mid);
+		id_free(overlay_thdl_idspace, mid);
+		return (EEXIST);
+	} else if ((flags & FEXCL) != 0) {
+		VERIFY(overlay_target_excl == B_FALSE);
+		overlay_target_excl = B_TRUE;
+	}
+	list_insert_tail(&overlay_thdl_list, thdl);
+	mutex_exit(&overlay_target_lock);
+
+	return (0);
+}
+
+/* ARGSUSED */
+int
+overlay_target_ioctl(dev_t dev, int cmd, intptr_t arg, int mode, cred_t *credp,
+    int *rvalp)
+{
+	overlay_target_ioctl_t *ioc;
+	overlay_target_hdl_t *thdl;
+
+	if (secpolicy_dl_config(credp) != 0)
+		return (EPERM);
+
+	if ((thdl = ddi_get_soft_state(overlay_thdl_state,
+	    getminor(dev))) == NULL)
+		return (ENXIO);
+
+	for (ioc = &overlay_target_ioctab[0]; ioc->oti_cmd != 0; ioc++) {
+		int ret;
+		caddr_t buf;
+		size_t bufsize;
+
+		if (ioc->oti_cmd != cmd)
+			continue;
+
+		if (ioc->oti_write == B_TRUE && !(mode & FWRITE))
+			return (EBADF);
+
+		if (ioc->oti_copyin == NULL) {
+			bufsize = ioc->oti_size;
+			buf = kmem_alloc(bufsize, KM_SLEEP);
+			if (ddi_copyin((void *)(uintptr_t)arg, buf, bufsize,
+			    mode & FKIOCTL) != 0) {
+				kmem_free(buf, bufsize);
+				return (EFAULT);
+			}
+		} else {
+			if ((ret = ioc->oti_copyin((void *)(uintptr_t)arg,
+			    (void **)&buf, &bufsize, mode)) != 0)
+				return (ret);
+		}
+
+		ret = ioc->oti_func(thdl, buf);
+		if (ret == 0 && ioc->oti_size != 0 &&
+		    ioc->oti_ncopyout == B_TRUE) {
+			if (ioc->oti_copyout == NULL) {
+				if (ddi_copyout(buf, (void *)(uintptr_t)arg,
+				    bufsize, mode & FKIOCTL) != 0)
+					ret = EFAULT;
+			} else {
+				ret = ioc->oti_copyout((void *)(uintptr_t)arg,
+				    buf, bufsize, mode);
+			}
+		}
+
+		kmem_free(buf, bufsize);
+		return (ret);
+	}
+
+	return (ENOTTY);
+}
+
+/* ARGSUSED */
+int
+overlay_target_close(dev_t dev, int flags, int otype, cred_t *credp)
+{
+	overlay_target_hdl_t *thdl;
+	overlay_target_entry_t *entry;
+	minor_t mid = getminor(dev);
+
+	if ((thdl = ddi_get_soft_state(overlay_thdl_state, mid)) == NULL)
+		return (ENXIO);
+
+	mutex_enter(&overlay_target_lock);
+	list_remove(&overlay_thdl_list, thdl);
+	mutex_enter(&thdl->oth_lock);
+	while ((entry = list_remove_head(&thdl->oth_outstanding)) != NULL)
+		list_insert_tail(&overlay_target_list, entry);
+	cv_signal(&overlay_target_condvar);
+	mutex_exit(&thdl->oth_lock);
+	if ((thdl->oth_oflags & FEXCL) != 0) {
+		VERIFY(overlay_target_excl == B_TRUE);
+		overlay_target_excl = B_FALSE;
+	}
+	mutex_exit(&overlay_target_lock);
+
+	list_destroy(&thdl->oth_outstanding);
+	mutex_destroy(&thdl->oth_lock);
+	mid = thdl->oth_minor;
+	ddi_soft_state_free(overlay_thdl_state, mid);
+	id_free(overlay_thdl_idspace, mid);
+
+	return (0);
+}
diff --git a/usr/src/uts/common/io/overlay/plugins/overlay_vxlan.c b/usr/src/uts/common/io/overlay/plugins/overlay_vxlan.c
new file mode 100644
index 0000000000..8b4e4ecb42
--- /dev/null
+++ b/usr/src/uts/common/io/overlay/plugins/overlay_vxlan.c
@@ -0,0 +1,372 @@
+/*
+ * This file and its contents are supplied under the terms of the
+ * Common Development and Distribution License ("CDDL"), version 1.0.
+ * You may only use this file in accordance with the terms of version
+ * 1.0 of the CDDL.
+ *
+ * A full copy of the text of the CDDL should have accompanied this
+ * source.  A copy of the CDDL is also available via the Internet at
+ * http://www.illumos.org/license/CDDL.
+ */
+
+/*
+ * Copyright 2015 Joyent, Inc.
+ */
+
+/*
+ * VXLAN encapsulation module
+ *
+ *
+ * The VXLAN header looks as follows in network byte order:
+ *
+ * |0        3| 4 |5                     31|
+ * +----------+---+------------------------+
+ * | Reserved | I | Reserved               |
+ * +---------------------------------------+
+ * | Virtual Network ID         | Reserved |
+ * +----------------------------+----------+
+ * |0                         23|24      31|
+ *
+ * All reserved values must be 0. The I bit must be 1. We call the top
+ * word the VXLAN magic field for the time being. The second word is
+ * definitely not the most friendly way to operate. Specifically, the ID
+ * is a 24-bit big endian value, but we have to make sure not to use the
+ * reserved byte.
+ *
+ * For us, VXLAN encapsulation is a fairly straightforward implementation. It
+ * only has two properties, a listen_ip and a listen_port. These determine on
+ * what address we should be listening on. While we do not have a default
+ * address to listen upon, we do have a default port, which is the IANA assigned
+ * port for VXLAN -- 4789.
+ */
+
+#include <sys/overlay_plugin.h>
+#include <sys/modctl.h>
+#include <sys/errno.h>
+#include <sys/byteorder.h>
+#include <sys/vxlan.h>
+#include <inet/ip.h>
+#include <netinet/in.h>
+#include <sys/strsun.h>
+#include <netinet/udp.h>
+
+static const char *vxlan_ident = "vxlan";
+static uint16_t vxlan_defport = IPPORT_VXLAN;
+
+/*
+ * Should we enable UDP source port hashing for fanout.
+ */
+boolean_t vxlan_fanout = B_TRUE;
+
+static const char *vxlan_props[] = {
+	"vxlan/listen_ip",
+	"vxlan/listen_port",
+	NULL
+};
+
+typedef struct vxlan {
+	kmutex_t vxl_lock;
+	overlay_handle_t vxl_oh;
+	uint16_t vxl_lport;
+	boolean_t vxl_hladdr;
+	struct in6_addr vxl_laddr;
+} vxlan_t;
+
+static int
+vxlan_o_init(overlay_handle_t oh, void **outp)
+{
+	vxlan_t *vxl;
+
+	vxl = kmem_alloc(sizeof (vxlan_t), KM_SLEEP);
+	*outp = vxl;
+	mutex_init(&vxl->vxl_lock, NULL, MUTEX_DRIVER, NULL);
+	vxl->vxl_oh = oh;
+	vxl->vxl_lport = vxlan_defport;
+	vxl->vxl_hladdr = B_FALSE;
+
+	return (0);
+}
+
+static void
+vxlan_o_fini(void *arg)
+{
+	vxlan_t *vxl = arg;
+
+	mutex_destroy(&vxl->vxl_lock);
+	kmem_free(arg, sizeof (vxlan_t));
+}
+
+static int
+vxlan_o_socket(void *arg, int *dp, int *fp, int *pp, struct sockaddr *addr,
+    socklen_t *slenp)
+{
+	vxlan_t *vxl = arg;
+	struct sockaddr_in6 *in;
+
+	in = (struct sockaddr_in6 *)addr;
+	*dp = AF_INET6;
+	*fp = SOCK_DGRAM;
+	*pp = 0;
+	bzero(in, sizeof (struct sockaddr_in6));
+	in->sin6_family = AF_INET6;
+
+	/*
+	 * We should consider a more expressive private errno set that
+	 * provider's can use.
+	 */
+	mutex_enter(&vxl->vxl_lock);
+	if (vxl->vxl_hladdr == B_FALSE) {
+		mutex_exit(&vxl->vxl_lock);
+		return (EINVAL);
+	}
+	in->sin6_port = htons(vxl->vxl_lport);
+	in->sin6_addr = vxl->vxl_laddr;
+	mutex_exit(&vxl->vxl_lock);
+	*slenp = sizeof (struct sockaddr_in6);
+
+	return (0);
+}
+
+static int
+vxlan_o_sockopt(ksocket_t ksock)
+{
+	int val, err;
+	if (vxlan_fanout == B_FALSE)
+		return (0);
+
+	val = UDP_HASH_VXLAN;
+	err = ksocket_setsockopt(ksock, IPPROTO_UDP, UDP_SRCPORT_HASH, &val,
+	    sizeof (val), kcred);
+	return (err);
+}
+
+/* ARGSUSED */
+static int
+vxlan_o_encap(void *arg, mblk_t *mp, ovep_encap_info_t *einfop,
+    mblk_t **outp)
+{
+	mblk_t *ob;
+	vxlan_hdr_t *vxh;
+
+	ASSERT(einfop->ovdi_id < (1 << 24));
+
+	/*
+	 * This allocation could get hot. We may want to have a good way to
+	 * cache and handle this allocation the same way that IP does with
+	 * keeping around a message block per entry, or basically treating this
+	 * as an immutable message block in the system. Basically freemsg() will
+	 * be a nop, but we'll do the right thing with respect to the rest of
+	 * the chain.
+	 */
+	ob = allocb(VXLAN_HDR_LEN, 0);
+	if (ob == NULL)
+		return (ENOMEM);
+
+	vxh = (vxlan_hdr_t *)ob->b_rptr;
+	vxh->vxlan_flags = ntohl(VXLAN_F_VDI);
+	vxh->vxlan_id = htonl((uint32_t)einfop->ovdi_id << VXLAN_ID_SHIFT);
+	ob->b_wptr += VXLAN_HDR_LEN;
+	*outp = ob;
+
+	return (0);
+}
+
+/* ARGSUSED */
+static int
+vxlan_o_decap(void *arg, mblk_t *mp, ovep_encap_info_t *dinfop)
+{
+	vxlan_hdr_t *vxh;
+
+	if (MBLKL(mp) < sizeof (vxlan_hdr_t))
+		return (EINVAL);
+	vxh = (vxlan_hdr_t *)mp->b_rptr;
+	if ((ntohl(vxh->vxlan_flags) & VXLAN_F_VDI) == 0)
+		return (EINVAL);
+
+	dinfop->ovdi_id = ntohl(vxh->vxlan_id) >> VXLAN_ID_SHIFT;
+	dinfop->ovdi_hdr_size = VXLAN_HDR_LEN;
+
+	return (0);
+}
+
+static int
+vxlan_o_getprop(void *arg, const char *pr_name, void *buf, uint32_t *bufsize)
+{
+	vxlan_t *vxl = arg;
+
+	/* vxlan/listen_ip */
+	if (strcmp(pr_name, vxlan_props[0]) == 0) {
+		if (*bufsize < sizeof (struct in6_addr))
+			return (EOVERFLOW);
+
+		mutex_enter(&vxl->vxl_lock);
+		if (vxl->vxl_hladdr == B_FALSE) {
+			*bufsize = 0;
+		} else {
+			bcopy(&vxl->vxl_laddr, buf, sizeof (struct in6_addr));
+			*bufsize = sizeof (struct in6_addr);
+		}
+		mutex_exit(&vxl->vxl_lock);
+		return (0);
+	}
+
+	/* vxlan/listen_port */
+	if (strcmp(pr_name, vxlan_props[1]) == 0) {
+		uint64_t val;
+		if (*bufsize < sizeof (uint64_t))
+			return (EOVERFLOW);
+
+		mutex_enter(&vxl->vxl_lock);
+		val = vxl->vxl_lport;
+		bcopy(&val, buf, sizeof (uint64_t));
+		*bufsize = sizeof (uint64_t);
+		mutex_exit(&vxl->vxl_lock);
+		return (0);
+	}
+
+	return (EINVAL);
+}
+
+static int
+vxlan_o_setprop(void *arg, const char *pr_name, const void *buf,
+    uint32_t bufsize)
+{
+	vxlan_t *vxl = arg;
+
+	/* vxlan/listen_ip */
+	if (strcmp(pr_name, vxlan_props[0]) == 0) {
+		const struct in6_addr *ipv6 = buf;
+		if (bufsize != sizeof (struct in6_addr))
+			return (EINVAL);
+
+		if (IN6_IS_ADDR_V4COMPAT(ipv6))
+			return (EINVAL);
+
+		if (IN6_IS_ADDR_MULTICAST(ipv6))
+			return (EINVAL);
+
+		if (IN6_IS_ADDR_6TO4(ipv6))
+			return (EINVAL);
+
+		if (IN6_IS_ADDR_V4MAPPED(ipv6)) {
+			ipaddr_t v4;
+			IN6_V4MAPPED_TO_IPADDR(ipv6, v4);
+			if (IN_MULTICAST(v4))
+				return (EINVAL);
+		}
+
+		mutex_enter(&vxl->vxl_lock);
+		vxl->vxl_hladdr = B_TRUE;
+		bcopy(ipv6, &vxl->vxl_laddr, sizeof (struct in6_addr));
+		mutex_exit(&vxl->vxl_lock);
+
+		return (0);
+	}
+
+	/* vxlan/listen_port */
+	if (strcmp(pr_name, vxlan_props[1]) == 0) {
+		const uint64_t *valp = buf;
+		if (bufsize != 8)
+			return (EINVAL);
+
+		if (*valp == 0 || *valp > UINT16_MAX)
+			return (EINVAL);
+
+		mutex_enter(&vxl->vxl_lock);
+		vxl->vxl_lport = *valp;
+		mutex_exit(&vxl->vxl_lock);
+		return (0);
+	}
+	return (EINVAL);
+}
+
+static int
+vxlan_o_propinfo(const char *pr_name, overlay_prop_handle_t phdl)
+{
+	/* vxlan/listen_ip */
+	if (strcmp(pr_name, vxlan_props[0]) == 0) {
+		overlay_prop_set_name(phdl, vxlan_props[0]);
+		overlay_prop_set_prot(phdl, OVERLAY_PROP_PERM_RRW);
+		overlay_prop_set_type(phdl, OVERLAY_PROP_T_IP);
+		overlay_prop_set_nodefault(phdl);
+		return (0);
+	}
+
+	if (strcmp(pr_name, vxlan_props[1]) == 0) {
+		overlay_prop_set_name(phdl, vxlan_props[1]);
+		overlay_prop_set_prot(phdl, OVERLAY_PROP_PERM_RRW);
+		overlay_prop_set_type(phdl, OVERLAY_PROP_T_UINT);
+		(void) overlay_prop_set_default(phdl, &vxlan_defport,
+		    sizeof (vxlan_defport));
+		overlay_prop_set_range_uint32(phdl, 1, UINT16_MAX);
+		return (0);
+	}
+
+	return (EINVAL);
+}
+
+static struct overlay_plugin_ops vxlan_o_ops = {
+	0,
+	vxlan_o_init,
+	vxlan_o_fini,
+	vxlan_o_encap,
+	vxlan_o_decap,
+	vxlan_o_socket,
+	vxlan_o_sockopt,
+	vxlan_o_getprop,
+	vxlan_o_setprop,
+	vxlan_o_propinfo
+};
+
+static struct modlmisc vxlan_modlmisc = {
+	&mod_miscops,
+	"VXLAN encap plugin"
+};
+
+static struct modlinkage vxlan_modlinkage = {
+	MODREV_1,
+	&vxlan_modlmisc
+};
+
+int
+_init(void)
+{
+	int err;
+	overlay_plugin_register_t *ovrp;
+
+	ovrp = overlay_plugin_alloc(OVEP_VERSION);
+	if (ovrp == NULL)
+		return (ENOTSUP);
+	ovrp->ovep_name = vxlan_ident;
+	ovrp->ovep_ops = &vxlan_o_ops;
+	ovrp->ovep_id_size = VXLAN_ID_LEN;
+	ovrp->ovep_flags = OVEP_F_VLAN_TAG;
+	ovrp->ovep_dest = OVERLAY_PLUGIN_D_IP | OVERLAY_PLUGIN_D_PORT;
+	ovrp->ovep_props = vxlan_props;
+
+	if ((err = overlay_plugin_register(ovrp)) == 0) {
+		if ((err = mod_install(&vxlan_modlinkage)) != 0) {
+			(void) overlay_plugin_unregister(vxlan_ident);
+		}
+	}
+
+	overlay_plugin_free(ovrp);
+	return (err);
+}
+
+int
+_info(struct modinfo *modinfop)
+{
+	return (mod_info(&vxlan_modlinkage, modinfop));
+}
+
+int
+_fini(void)
+{
+	int err;
+
+	if ((err = overlay_plugin_unregister(vxlan_ident)) != 0)
+		return (err);
+
+	return (mod_remove(&vxlan_modlinkage));
+}
diff --git a/usr/src/uts/common/io/pseudo.conf b/usr/src/uts/common/io/pseudo.conf
index 42248e93d6..08affec609 100644
--- a/usr/src/uts/common/io/pseudo.conf
+++ b/usr/src/uts/common/io/pseudo.conf
@@ -22,8 +22,7 @@
 #
 # Copyright 2003 Sun Microsystems, Inc.  All rights reserved.
 # Use is subject to license terms.
-#
-# ident	"%Z%%M%	%I%	%E% SMI"
+# Copyright 2014 Joyent, Inc.  All rights reserved.
 #
 # This file is private to the pseudonex driver.  It should not be edited.
 #
@@ -38,3 +37,9 @@ name="pseudo" class="root" instance=0;
 # /pseudo; it has as its children the zone console pseudo nodes.
 #
 name="zconsnex" parent="/pseudo" instance=1 valid-children="zcons";
+
+#
+# zfdnex is an alias for pseudo; this node is instantiated as a child of
+# /pseudo; it has as its children the zone fd pseudo nodes.
+#
+name="zfdnex" parent="/pseudo" instance=2 valid-children="zfd";
diff --git a/usr/src/uts/common/io/pseudonex.c b/usr/src/uts/common/io/pseudonex.c
index f83b0abf39..0ae06f88cc 100644
--- a/usr/src/uts/common/io/pseudonex.c
+++ b/usr/src/uts/common/io/pseudonex.c
@@ -83,6 +83,8 @@ static int pseudonex_detach(dev_info_t *, ddi_detach_cmd_t);
 static int pseudonex_open(dev_t *, int, int, cred_t *);
 static int pseudonex_close(dev_t, int, int, cred_t *);
 static int pseudonex_ioctl(dev_t, int, intptr_t, int, cred_t *, int *);
+static int pseudonex_fm_init(dev_info_t *, dev_info_t *, int,
+    ddi_iblock_cookie_t *);
 static int pseudonex_ctl(dev_info_t *, dev_info_t *, ddi_ctl_enum_t, void *,
     void *);
 
@@ -90,6 +92,8 @@ static void *pseudonex_state;
 
 typedef struct pseudonex_state {
 	dev_info_t *pnx_devi;
+	int pnx_fmcap;
+	ddi_iblock_cookie_t pnx_fm_ibc;
 } pseudonex_state_t;
 
 static struct bus_ops pseudonex_bus_ops = {
@@ -116,7 +120,7 @@ static struct bus_ops pseudonex_bus_ops = {
 	NULL,			/* bus_intr_ctl */
 	NULL,			/* bus_config */
 	NULL,			/* bus_unconfig */
-	NULL,			/* bus_fm_init */
+	pseudonex_fm_init,	/* bus_fm_init */
 	NULL,			/* bus_fm_fini */
 	NULL,			/* bus_fm_access_enter */
 	NULL,			/* bus_fm_access_exit */
@@ -228,6 +232,9 @@ pseudonex_attach(dev_info_t *devi, ddi_attach_cmd_t cmd)
 	pnx_state = ddi_get_soft_state(pseudonex_state, instance);
 	pnx_state->pnx_devi = devi;
 
+	pnx_state->pnx_fmcap = DDI_FM_EREPORT_CAPABLE;
+	ddi_fm_init(devi, &pnx_state->pnx_fmcap, &pnx_state->pnx_fm_ibc);
+
 	if (ddi_create_minor_node(devi, "devctl", S_IFCHR, instance,
 	    DDI_NT_NEXUS, 0) != DDI_SUCCESS) {
 		ddi_remove_minor_node(devi, NULL);
@@ -247,6 +254,10 @@ pseudonex_detach(dev_info_t *devi, ddi_detach_cmd_t cmd)
 	if (cmd == DDI_SUSPEND)
 		return (DDI_SUCCESS);
 
+	if (cmd != DDI_DETACH)
+		return (DDI_FAILURE);
+
+	ddi_fm_fini(devi);
 	ddi_remove_minor_node(devi, NULL);
 	ddi_soft_state_free(pseudonex_state, instance);
 	return (DDI_SUCCESS);
@@ -375,6 +386,19 @@ pseudonex_auto_assign(dev_info_t *child)
 }
 
 static int
+pseudonex_fm_init(dev_info_t *dip, dev_info_t *tdip, int cap,
+    ddi_iblock_cookie_t *ibc)
+{
+	pseudonex_state_t *pnx_state;
+
+	pnx_state = ddi_get_soft_state(pseudonex_state, ddi_get_instance(dip));
+	ASSERT(pnx_state != NULL);
+	ASSERT(ibc != NULL);
+	*ibc = pnx_state->pnx_fm_ibc;
+	return (pnx_state->pnx_fmcap & cap);
+}
+
+static int
 pseudonex_ctl(dev_info_t *dip, dev_info_t *rdip, ddi_ctl_enum_t ctlop,
     void *arg, void *result)
 {
diff --git a/usr/src/uts/common/io/ptm.c b/usr/src/uts/common/io/ptm.c
index 400e9ffd10..07ffddc123 100644
--- a/usr/src/uts/common/io/ptm.c
+++ b/usr/src/uts/common/io/ptm.c
@@ -447,6 +447,18 @@ ptmclose(queue_t *rqp, int flag, cred_t *credp)
 	return (0);
 }
 
+static boolean_t
+ptmptsopencb(ptmptsopencb_arg_t arg)
+{
+	struct pt_ttys	*ptmp = (struct pt_ttys *)arg;
+	boolean_t rval;
+
+	PT_ENTER_READ(ptmp);
+	rval = (ptmp->pt_nullmsg != NULL);
+	PT_EXIT_READ(ptmp);
+	return (rval);
+}
+
 /*
  * The wput procedure will only handle ioctl and flush messages.
  */
@@ -574,6 +586,41 @@ ptmwput(queue_t *qp, mblk_t *mp)
 			miocack(qp, mp, 0, 0);
 			break;
 		}
+		case PTMPTSOPENCB:
+		{
+			mblk_t		*dp;	/* ioctl reply data */
+			ptmptsopencb_t	*ppocb;
+
+			/* only allow the kernel to invoke this ioctl */
+			if (iocp->ioc_cr != kcred) {
+				miocnak(qp, mp, 0, EINVAL);
+				break;
+			}
+
+			/* we don't support transparent ioctls */
+			ASSERT(iocp->ioc_count != TRANSPARENT);
+			if (iocp->ioc_count == TRANSPARENT) {
+				miocnak(qp, mp, 0, EINVAL);
+				break;
+			}
+
+			/* allocate a response message */
+			dp = allocb(sizeof (ptmptsopencb_t), BPRI_MED);
+			if (dp == NULL) {
+				miocnak(qp, mp, 0, EAGAIN);
+				break;
+			}
+
+			/* initialize the ioctl results */
+			ppocb = (ptmptsopencb_t *)dp->b_rptr;
+			ppocb->ppocb_func = ptmptsopencb;
+			ppocb->ppocb_arg = (ptmptsopencb_arg_t)ptmp;
+
+			/* send the reply data */
+			mioc2ack(mp, dp, sizeof (ptmptsopencb_t), 0);
+			qreply(qp, mp);
+			break;
+		}
 		}
 		break;
 
diff --git a/usr/src/uts/common/io/scsi/adapters/mpt_sas/mptsas.c b/usr/src/uts/common/io/scsi/adapters/mpt_sas/mptsas.c
index f094c9a510..db9197ca36 100644
--- a/usr/src/uts/common/io/scsi/adapters/mpt_sas/mptsas.c
+++ b/usr/src/uts/common/io/scsi/adapters/mpt_sas/mptsas.c
@@ -22,7 +22,7 @@
 /*
  * Copyright (c) 2009, 2010, Oracle and/or its affiliates. All rights reserved.
  * Copyright 2014 Nexenta Systems, Inc. All rights reserved.
- * Copyright (c) 2014, Joyent, Inc. All rights reserved.
+ * Copyright 2015, Joyent, Inc.
  * Copyright 2014 OmniTI Computer Consulting, Inc. All rights reserved.
  * Copyright (c) 2014, Tegile Systems Inc. All rights reserved.
  */
@@ -72,6 +72,7 @@
 #include <sys/file.h>
 #include <sys/policy.h>
 #include <sys/model.h>
+#include <sys/refhash.h>
 #include <sys/sysevent.h>
 #include <sys/sysevent/eventdefs.h>
 #include <sys/sysevent/dr.h>
@@ -98,7 +99,6 @@
 #include <sys/scsi/adapters/mpt_sas/mptsas_var.h>
 #include <sys/scsi/adapters/mpt_sas/mptsas_ioctl.h>
 #include <sys/scsi/adapters/mpt_sas/mptsas_smhba.h>
-#include <sys/scsi/adapters/mpt_sas/mptsas_hash.h>
 #include <sys/raidioctl.h>
 
 #include <sys/fs/dv_node.h>	/* devfs_clean */
diff --git a/usr/src/uts/common/io/scsi/targets/sd.c b/usr/src/uts/common/io/scsi/targets/sd.c
index ae1e7e0fc3..dc5dc22e37 100644
--- a/usr/src/uts/common/io/scsi/targets/sd.c
+++ b/usr/src/uts/common/io/scsi/targets/sd.c
@@ -3503,9 +3503,13 @@ sd_set_mmc_caps(sd_ssc_t *ssc)
 		 * according to the successful response to the page
 		 * 0x2A mode sense request.
 		 */
-		scsi_log(SD_DEVINFO(un), sd_label, CE_WARN,
-		    "sd_set_mmc_caps: Mode Sense returned "
-		    "invalid block descriptor length\n");
+		/*
+		 * The following warning occurs due to the KVM CD-ROM
+		 * mishandling the multi-media commands.  Ignore it.
+		 * scsi_log(SD_DEVINFO(un), sd_label, CE_WARN,
+		 *     "sd_set_mmc_caps: Mode Sense returned "
+		 *     "invalid block descriptor length\n");
+		 */
 		kmem_free(buf, BUFLEN_MODE_CDROM_CAP);
 		return;
 	}
@@ -4450,18 +4454,77 @@ sd_sdconf_id_match(struct sd_lun *un, char *id, int idlen)
 {
 	struct scsi_inquiry	*sd_inq;
 	int 			rval = SD_SUCCESS;
+	char			*p;
+	int			chk_vidlen = 0, chk_pidlen = 0;
+	int			has_tail = 0;
+	static const int	VSZ = sizeof (sd_inq->inq_vid);
+	static const int	PSZ = sizeof (sd_inq->inq_pid);
 
 	ASSERT(un != NULL);
 	sd_inq = un->un_sd->sd_inq;
 	ASSERT(id != NULL);
 
 	/*
-	 * We use the inq_vid as a pointer to a buffer containing the
-	 * vid and pid and use the entire vid/pid length of the table
-	 * entry for the comparison. This works because the inq_pid
-	 * data member follows inq_vid in the scsi_inquiry structure.
+	 * We would like to use the inq_vid as a pointer to a buffer
+	 * containing the vid and pid and use the entire vid/pid length of
+	 * the table entry for the comparison.  However, this does not work
+	 * because, while the inq_pid data member follows inq_vid in the
+	 * scsi_inquiry structure, we do not control the contents of this
+	 * buffer, and some broken devices violate SPC 4.3.1 and return
+	 * fields with null bytes in them.
+	 */
+	chk_vidlen = MIN(VSZ, idlen);
+	p = id + chk_vidlen - 1;
+	while (*p == ' ' && chk_vidlen > 0) {
+		--p;
+		--chk_vidlen;
+	}
+
+	/*
+	 * If it's all spaces, check the whole thing.
 	 */
-	if (strncasecmp(sd_inq->inq_vid, id, idlen) != 0) {
+	if (chk_vidlen == 0)
+		chk_vidlen = MIN(VSZ, idlen);
+
+	if (idlen > VSZ) {
+		chk_pidlen = idlen - VSZ;
+		p = id + idlen - 1;
+		while (*p == ' ' && chk_pidlen > 0) {
+			--p;
+			--chk_pidlen;
+		}
+		if (chk_pidlen == 0)
+			chk_pidlen = MIN(PSZ, idlen - VSZ);
+	}
+
+	/*
+	 * There's one more thing we need to do here.  If the user specified
+	 * an ID with trailing spaces, we need to make sure the inquiry
+	 * vid/pid has only spaces or NULs after the check length; otherwise, it
+	 * can't match.
+	 */
+	if (idlen > chk_vidlen && chk_vidlen < VSZ) {
+		for (p = sd_inq->inq_vid + chk_vidlen;
+		    p < sd_inq->inq_vid + VSZ; ++p) {
+			if (*p != ' ' && *p != '\0') {
+				++has_tail;
+				break;
+			}
+		}
+	}
+	if (idlen > chk_pidlen + VSZ && chk_pidlen < PSZ) {
+		for (p = sd_inq->inq_pid + chk_pidlen;
+		    p < sd_inq->inq_pid + PSZ; ++p) {
+			if (*p != ' ' && *p != '\0') {
+				++has_tail;
+				break;
+			}
+		}
+	}
+
+	if (has_tail || strncasecmp(sd_inq->inq_vid, id, chk_vidlen) != 0 ||
+	    (idlen > VSZ &&
+	    strncasecmp(sd_inq->inq_pid, id + VSZ, chk_pidlen) != 0)) {
 		/*
 		 * The user id string is compared to the inquiry vid/pid
 		 * using a case insensitive comparison and ignoring
diff --git a/usr/src/uts/common/io/timerfd.c b/usr/src/uts/common/io/timerfd.c
new file mode 100644
index 0000000000..2f59c0f3e2
--- /dev/null
+++ b/usr/src/uts/common/io/timerfd.c
@@ -0,0 +1,548 @@
+/*
+ * This file and its contents are supplied under the terms of the
+ * Common Development and Distribution License ("CDDL"), version 1.0.
+ * You may only use this file in accordance with the terms of version
+ * 1.0 of the CDDL.
+ *
+ * A full copy of the text of the CDDL should have accompanied this
+ * source.  A copy of the CDDL is also available via the Internet at
+ * http://www.illumos.org/license/CDDL.
+ */
+
+/*
+ * Copyright (c) 2015 Joyent, Inc.  All rights reserved.
+ */
+
+/*
+ * Support for the timerfd facility, a Linux-borne facility that allows
+ * POSIX.1b timers to be created and manipulated via a file descriptor
+ * interface.
+ */
+
+#include <sys/ddi.h>
+#include <sys/sunddi.h>
+#include <sys/timerfd.h>
+#include <sys/conf.h>
+#include <sys/vmem.h>
+#include <sys/sysmacros.h>
+#include <sys/filio.h>
+#include <sys/stat.h>
+#include <sys/file.h>
+#include <sys/timer.h>
+
+struct timerfd_state;
+typedef struct timerfd_state timerfd_state_t;
+
+struct timerfd_state {
+	kmutex_t tfd_lock;			/* lock protecting state */
+	kcondvar_t tfd_cv;			/* condvar */
+	pollhead_t tfd_pollhd;			/* poll head */
+	uint64_t tfd_fired;			/* # of times fired */
+	itimer_t tfd_itimer;			/* underlying itimer */
+	timerfd_state_t *tfd_next;		/* next state on global list */
+};
+
+/*
+ * Internal global variables.
+ */
+static kmutex_t		timerfd_lock;		/* lock protecting state */
+static dev_info_t	*timerfd_devi;		/* device info */
+static vmem_t		*timerfd_minor;		/* minor number arena */
+static void		*timerfd_softstate;	/* softstate pointer */
+static timerfd_state_t	*timerfd_state;		/* global list of state */
+
+static itimer_t *
+timerfd_itimer_lock(timerfd_state_t *state)
+{
+	itimer_t *it = &state->tfd_itimer;
+
+	mutex_enter(&state->tfd_lock);
+
+	while (it->it_lock & ITLK_LOCKED) {
+		it->it_blockers++;
+		cv_wait(&it->it_cv, &state->tfd_lock);
+		it->it_blockers--;
+	}
+
+	it->it_lock |= ITLK_LOCKED;
+
+	mutex_exit(&state->tfd_lock);
+
+	return (it);
+}
+
+static void
+timerfd_itimer_unlock(timerfd_state_t *state, itimer_t *it)
+{
+	VERIFY(it == &state->tfd_itimer);
+	VERIFY(it->it_lock & ITLK_LOCKED);
+
+	mutex_enter(&state->tfd_lock);
+
+	it->it_lock &= ~ITLK_LOCKED;
+
+	if (it->it_blockers)
+		cv_signal(&it->it_cv);
+
+	mutex_exit(&state->tfd_lock);
+}
+
+static void
+timerfd_fire(itimer_t *it)
+{
+	timerfd_state_t *state = it->it_frontend;
+	uint64_t oval;
+
+	mutex_enter(&state->tfd_lock);
+	oval = state->tfd_fired++;
+	mutex_exit(&state->tfd_lock);
+
+	if (oval == 0) {
+		cv_broadcast(&state->tfd_cv);
+		pollwakeup(&state->tfd_pollhd, POLLRDNORM | POLLIN);
+	}
+}
+
+/*ARGSUSED*/
+static int
+timerfd_open(dev_t *devp, int flag, int otyp, cred_t *cred_p)
+{
+	timerfd_state_t *state;
+	major_t major = getemajor(*devp);
+	minor_t minor = getminor(*devp);
+
+	if (minor != TIMERFDMNRN_TIMERFD)
+		return (ENXIO);
+
+	mutex_enter(&timerfd_lock);
+
+	minor = (minor_t)(uintptr_t)vmem_alloc(timerfd_minor, 1,
+	    VM_BESTFIT | VM_SLEEP);
+
+	if (ddi_soft_state_zalloc(timerfd_softstate, minor) != DDI_SUCCESS) {
+		vmem_free(timerfd_minor, (void *)(uintptr_t)minor, 1);
+		mutex_exit(&timerfd_lock);
+		return (NULL);
+	}
+
+	state = ddi_get_soft_state(timerfd_softstate, minor);
+	*devp = makedevice(major, minor);
+
+	state->tfd_next = timerfd_state;
+	timerfd_state = state;
+
+	mutex_exit(&timerfd_lock);
+
+	return (0);
+}
+
+/*ARGSUSED*/
+static int
+timerfd_read(dev_t dev, uio_t *uio, cred_t *cr)
+{
+	timerfd_state_t *state;
+	minor_t minor = getminor(dev);
+	uint64_t val;
+	int err;
+
+	if (uio->uio_resid < sizeof (val))
+		return (EINVAL);
+
+	state = ddi_get_soft_state(timerfd_softstate, minor);
+
+	mutex_enter(&state->tfd_lock);
+
+	while (state->tfd_fired == 0) {
+		if (uio->uio_fmode & (FNDELAY|FNONBLOCK)) {
+			mutex_exit(&state->tfd_lock);
+			return (EAGAIN);
+		}
+
+		if (!cv_wait_sig_swap(&state->tfd_cv, &state->tfd_lock)) {
+			mutex_exit(&state->tfd_lock);
+			return (EINTR);
+		}
+	}
+
+	/*
+	 * Our tfd_fired is non-zero; slurp its value and then clear it.
+	 */
+	val = state->tfd_fired;
+	state->tfd_fired = 0;
+	mutex_exit(&state->tfd_lock);
+
+	err = uiomove(&val, sizeof (val), UIO_READ, uio);
+
+	return (err);
+}
+
+/*ARGSUSED*/
+static int
+timerfd_poll(dev_t dev, short events, int anyyet, short *reventsp,
+    struct pollhead **phpp)
+{
+	timerfd_state_t *state;
+	minor_t minor = getminor(dev);
+	short revents = 0;
+
+	state = ddi_get_soft_state(timerfd_softstate, minor);
+
+	mutex_enter(&state->tfd_lock);
+
+	if (state->tfd_fired > 0)
+		revents |= POLLRDNORM | POLLIN;
+
+	if (!(*reventsp = revents & events) && !anyyet)
+		*phpp = &state->tfd_pollhd;
+
+	mutex_exit(&state->tfd_lock);
+
+	return (0);
+}
+
+static int
+timerfd_copyin(uintptr_t addr, itimerspec_t *dest)
+{
+	if (get_udatamodel() == DATAMODEL_NATIVE) {
+		if (copyin((void *)addr, dest, sizeof (itimerspec_t)) != 0)
+			return (EFAULT);
+	} else {
+		itimerspec32_t dest32;
+
+		if (copyin((void *)addr, &dest32, sizeof (itimerspec32_t)) != 0)
+			return (EFAULT);
+
+		ITIMERSPEC32_TO_ITIMERSPEC(dest, &dest32);
+	}
+
+	if (itimerspecfix(&dest->it_value) ||
+	    (itimerspecfix(&dest->it_interval) &&
+	    timerspecisset(&dest->it_value))) {
+		return (EINVAL);
+	}
+
+	return (0);
+}
+
+static int
+timerfd_copyout(itimerspec_t *src, uintptr_t addr)
+{
+	if (get_udatamodel() == DATAMODEL_NATIVE) {
+		if (copyout(src, (void *)addr, sizeof (itimerspec_t)) != 0)
+			return (EFAULT);
+	} else {
+		itimerspec32_t src32;
+
+		if (ITIMERSPEC_OVERFLOW(src))
+			return (EOVERFLOW);
+
+		ITIMERSPEC_TO_ITIMERSPEC32(&src32, src);
+
+		if (copyout(&src32, (void *)addr, sizeof (itimerspec32_t)) != 0)
+			return (EFAULT);
+	}
+
+	return (0);
+}
+
+/*ARGSUSED*/
+static int
+timerfd_ioctl(dev_t dev, int cmd, intptr_t arg, int md, cred_t *cr, int *rv)
+{
+	itimerspec_t when, oval;
+	timerfd_state_t *state;
+	minor_t minor = getminor(dev);
+	int err;
+	itimer_t *it;
+
+	state = ddi_get_soft_state(timerfd_softstate, minor);
+
+	switch (cmd) {
+	case TIMERFDIOC_CREATE: {
+		if (arg == TIMERFD_MONOTONIC)
+			arg = CLOCK_MONOTONIC;
+
+		it = timerfd_itimer_lock(state);
+
+		if (it->it_backend != NULL) {
+			timerfd_itimer_unlock(state, it);
+			return (EEXIST);
+		}
+
+		if ((it->it_backend = clock_get_backend(arg)) == NULL) {
+			timerfd_itimer_unlock(state, it);
+			return (EINVAL);
+		}
+
+		/*
+		 * We need to provide a proc structure only for purposes
+		 * of locking CLOCK_REALTIME-based timers -- it is safe to
+		 * provide p0 here.
+		 */
+		it->it_proc = &p0;
+
+		err = it->it_backend->clk_timer_create(it, timerfd_fire);
+
+		if (err != 0) {
+			it->it_backend = NULL;
+			timerfd_itimer_unlock(state, it);
+			return (err);
+		}
+
+		it->it_frontend = state;
+		timerfd_itimer_unlock(state, it);
+
+		return (0);
+	}
+
+	case TIMERFDIOC_GETTIME: {
+		it = timerfd_itimer_lock(state);
+
+		if (it->it_backend == NULL) {
+			timerfd_itimer_unlock(state, it);
+			return (ENODEV);
+		}
+
+		err = it->it_backend->clk_timer_gettime(it, &when);
+		timerfd_itimer_unlock(state, it);
+
+		if (err != 0)
+			return (err);
+
+		if ((err = timerfd_copyout(&when, arg)) != 0)
+			return (err);
+
+		return (0);
+	}
+
+	case TIMERFDIOC_SETTIME: {
+		timerfd_settime_t st;
+
+		if (copyin((void *)arg, &st, sizeof (st)) != 0)
+			return (EFAULT);
+
+		if ((err = timerfd_copyin(st.tfd_settime_value, &when)) != 0)
+			return (err);
+
+		it = timerfd_itimer_lock(state);
+
+		if (it->it_backend == NULL) {
+			timerfd_itimer_unlock(state, it);
+			return (ENODEV);
+		}
+
+		if (st.tfd_settime_ovalue != NULL) {
+			err = it->it_backend->clk_timer_gettime(it, &oval);
+
+			if (err != 0) {
+				timerfd_itimer_unlock(state, it);
+				return (err);
+			}
+		}
+
+		err = it->it_backend->clk_timer_settime(it,
+		    st.tfd_settime_flags & TFD_TIMER_ABSTIME ?
+		    TIMER_ABSTIME : TIMER_RELTIME, &when);
+		timerfd_itimer_unlock(state, it);
+
+		if (err != 0 || st.tfd_settime_ovalue == NULL)
+			return (err);
+
+		if ((err = timerfd_copyout(&oval, st.tfd_settime_ovalue)) != 0)
+			return (err);
+
+		return (0);
+	}
+
+	default:
+		break;
+	}
+
+	return (ENOTTY);
+}
+
+/*ARGSUSED*/
+static int
+timerfd_close(dev_t dev, int flag, int otyp, cred_t *cred_p)
+{
+	timerfd_state_t *state, **sp;
+	itimer_t *it;
+	minor_t minor = getminor(dev);
+
+	state = ddi_get_soft_state(timerfd_softstate, minor);
+
+	if (state->tfd_pollhd.ph_list != NULL) {
+		pollwakeup(&state->tfd_pollhd, POLLERR);
+		pollhead_clean(&state->tfd_pollhd);
+	}
+
+	/*
+	 * No one can get to this timer; we don't need to lock it -- we can
+	 * just call on the backend to delete it.
+	 */
+	it = &state->tfd_itimer;
+
+	if (it->it_backend != NULL)
+		it->it_backend->clk_timer_delete(it);
+
+	mutex_enter(&timerfd_lock);
+
+	/*
+	 * Remove our state from our global list.
+	 */
+	for (sp = &timerfd_state; *sp != state; sp = &((*sp)->tfd_next))
+		VERIFY(*sp != NULL);
+
+	*sp = (*sp)->tfd_next;
+
+	ddi_soft_state_free(timerfd_softstate, minor);
+	vmem_free(timerfd_minor, (void *)(uintptr_t)minor, 1);
+
+	mutex_exit(&timerfd_lock);
+
+	return (0);
+}
+
+/*ARGSUSED*/
+static int
+timerfd_attach(dev_info_t *devi, ddi_attach_cmd_t cmd)
+{
+	mutex_enter(&timerfd_lock);
+
+	if (ddi_soft_state_init(&timerfd_softstate,
+	    sizeof (timerfd_state_t), 0) != 0) {
+		cmn_err(CE_NOTE, "/dev/timerfd failed to create soft state");
+		mutex_exit(&timerfd_lock);
+		return (DDI_FAILURE);
+	}
+
+	if (ddi_create_minor_node(devi, "timerfd", S_IFCHR,
+	    TIMERFDMNRN_TIMERFD, DDI_PSEUDO, NULL) == DDI_FAILURE) {
+		cmn_err(CE_NOTE, "/dev/timerfd couldn't create minor node");
+		ddi_soft_state_fini(&timerfd_softstate);
+		mutex_exit(&timerfd_lock);
+		return (DDI_FAILURE);
+	}
+
+	ddi_report_dev(devi);
+	timerfd_devi = devi;
+
+	timerfd_minor = vmem_create("timerfd_minor", (void *)TIMERFDMNRN_CLONE,
+	    UINT32_MAX - TIMERFDMNRN_CLONE, 1, NULL, NULL, NULL, 0,
+	    VM_SLEEP | VMC_IDENTIFIER);
+
+	mutex_exit(&timerfd_lock);
+
+	return (DDI_SUCCESS);
+}
+
+/*ARGSUSED*/
+static int
+timerfd_detach(dev_info_t *dip, ddi_detach_cmd_t cmd)
+{
+	switch (cmd) {
+	case DDI_DETACH:
+		break;
+
+	case DDI_SUSPEND:
+		return (DDI_SUCCESS);
+
+	default:
+		return (DDI_FAILURE);
+	}
+
+	mutex_enter(&timerfd_lock);
+	vmem_destroy(timerfd_minor);
+
+	ddi_remove_minor_node(timerfd_devi, NULL);
+	timerfd_devi = NULL;
+
+	ddi_soft_state_fini(&timerfd_softstate);
+	mutex_exit(&timerfd_lock);
+
+	return (DDI_SUCCESS);
+}
+
+/*ARGSUSED*/
+static int
+timerfd_info(dev_info_t *dip, ddi_info_cmd_t infocmd, void *arg, void **result)
+{
+	int error;
+
+	switch (infocmd) {
+	case DDI_INFO_DEVT2DEVINFO:
+		*result = (void *)timerfd_devi;
+		error = DDI_SUCCESS;
+		break;
+	case DDI_INFO_DEVT2INSTANCE:
+		*result = (void *)0;
+		error = DDI_SUCCESS;
+		break;
+	default:
+		error = DDI_FAILURE;
+	}
+	return (error);
+}
+
+static struct cb_ops timerfd_cb_ops = {
+	timerfd_open,		/* open */
+	timerfd_close,		/* close */
+	nulldev,		/* strategy */
+	nulldev,		/* print */
+	nodev,			/* dump */
+	timerfd_read,		/* read */
+	nodev,			/* write */
+	timerfd_ioctl,		/* ioctl */
+	nodev,			/* devmap */
+	nodev,			/* mmap */
+	nodev,			/* segmap */
+	timerfd_poll,		/* poll */
+	ddi_prop_op,		/* cb_prop_op */
+	0,			/* streamtab  */
+	D_NEW | D_MP		/* Driver compatibility flag */
+};
+
+static struct dev_ops timerfd_ops = {
+	DEVO_REV,		/* devo_rev */
+	0,			/* refcnt */
+	timerfd_info,		/* get_dev_info */
+	nulldev,		/* identify */
+	nulldev,		/* probe */
+	timerfd_attach,		/* attach */
+	timerfd_detach,		/* detach */
+	nodev,			/* reset */
+	&timerfd_cb_ops,	/* driver operations */
+	NULL,			/* bus operations */
+	nodev,			/* dev power */
+	ddi_quiesce_not_needed,	/* quiesce */
+};
+
+static struct modldrv modldrv = {
+	&mod_driverops,		/* module type (this is a pseudo driver) */
+	"timerfd support",	/* name of module */
+	&timerfd_ops,		/* driver ops */
+};
+
+static struct modlinkage modlinkage = {
+	MODREV_1,
+	(void *)&modldrv,
+	NULL
+};
+
+int
+_init(void)
+{
+	return (mod_install(&modlinkage));
+}
+
+int
+_info(struct modinfo *modinfop)
+{
+	return (mod_info(&modlinkage, modinfop));
+}
+
+int
+_fini(void)
+{
+	return (mod_remove(&modlinkage));
+}
diff --git a/usr/src/uts/common/io/timerfd.conf b/usr/src/uts/common/io/timerfd.conf
new file mode 100644
index 0000000000..c6ad86d051
--- /dev/null
+++ b/usr/src/uts/common/io/timerfd.conf
@@ -0,0 +1,16 @@
+#
+# This file and its contents are supplied under the terms of the
+# Common Development and Distribution License ("CDDL"), version 1.0.
+# You may only use this file in accordance with the terms of version
+# 1.0 of the CDDL.
+#
+# A full copy of the text of the CDDL should have accompanied this
+# source.  A copy of the CDDL is also available via the Internet at
+# http://www.illumos.org/license/CDDL.
+#
+
+#
+# Copyright (c) 2015 Joyent, Inc.  All rights reserved.
+#
+
+name="timerfd" parent="pseudo" instance=0;
diff --git a/usr/src/uts/common/io/udmf/dm9601reg.h b/usr/src/uts/common/io/udmf/dm9601reg.h
new file mode 100644
index 0000000000..a36f2b0fc8
--- /dev/null
+++ b/usr/src/uts/common/io/udmf/dm9601reg.h
@@ -0,0 +1,348 @@
+/*
+ * %W% %E%
+ * Macro definitions for Davicom DM9601 USB to fast ethernet controler
+ * based on Davicom DM9601E data sheet
+ * This file is public domain. Coded by M.Murayama (KHF04453@nifty.com)
+ */
+
+#ifndef __DM9601_H__
+#define __DM9601_H__
+
+/*
+ * offset of registers
+ */
+#define	NCR	0x00U	/* network control register */
+#define	NSR	0x01U	/* network status register */
+#define	TCR	0x02U	/* tx control register */
+#define	TSR1	0x03U	/* tx status register 1 */
+#define	TSR2	0x04U	/* tx status register 2 */
+#define	RCR	0x05U	/* rx control register */
+#define	RSR	0x06U	/* rx status register */
+#define	ROCR	0x07U	/* rx overflow counter register */
+#define	BPTR	0x08U	/* back pressure threshold regster */
+#define	FCTR	0x09U	/* flow control threshold regster */
+#define	FCR	0x0aU	/* flow control threshold regster */
+#define	EPCR	0x0bU	/* eeprom & phy control register */
+#define	EPAR	0x0cU	/* eeprom & phy address register */
+#define	EPDR	0x0dU	/* eeprom & phy data register (2byte) */
+#define	WCR	0x0fU	/* wake up control register */
+#define	PAR	0x10U	/* physical address register (6byte) */
+#define	MAR	0x16U	/* multicast address register (8byte) */
+#define	GPCR	0x1eU	/* general purpose control register */
+#define	GPR	0x1fU	/* general purpose register */
+#define	VID	0x28U	/* vendor ID (2byte) */
+#define	PID	0x2aU	/* product ID (2byte) */
+#define	CHIPR	0x2cU	/* chip revision */
+#define	USBDA	0xf0U	/* usb device address register */
+#define	RXC	0xf1U	/* received packet counter register */
+#define	TUSC	0xf2U	/* tx packet counter/usb status register */
+#define	USBC	0xf4U	/* usb control register */
+
+/*
+ * register definitions
+ */
+/* network control register */
+#define	NCR_EXT_PHY	0x80U	/* 1: select external phy */
+#define	NCR_WAKEEN	0x40U	/* 1: wake up event enable */
+#define	NCR_FCOL	0x10U	/* force collision mode for test */
+#define	NCR_FDX		0x08U	/* 1: full duplex mode (for external phy) */
+#define	NCR_LBK		0x06U
+#define		NCR_LBK_SHIFT		1
+#define		NCR_LBK_NORMAL	(0U << NCR_LBK_SHIFT)
+#define		NCR_LBK_MAC	(1U << NCR_LBK_SHIFT)
+#define		NCR_LBK_PHY_D	(2U << NCR_LBK_SHIFT)
+#define		NCR_LBK_PHY_A	(3U << NCR_LBK_SHIFT)
+#define	NCR_RST		0x01U	/* 1: reset, auto clear */
+
+#define	NCR_BITS	\
+	"\020"	\
+	"\010EXT_PHY"	\
+	"\007WAKEEN"	\
+	"\005FCOL"	\
+	"\004FDX"	\
+	"\001RST"
+
+/* network status register */
+#define	NSR_SPEED	0x80U	/* 1:10M 0:100M */
+#define	NSR_LINKST	0x40U	/* 1:ok 0:fail */
+#define	NSR_WAKEST	0x20U	/* 1:enabled */
+#define	NSR_TXFULL	0x10U	/* 1:tx fifo full */
+#define	NSR_TX2END	0x08U	/* tx packet2 complete status */
+#define	NSR_TX1END	0x04U	/* tx packet1 complete status */
+#define	NSR_RXOV	0x02U	/* rx fifo overflow */
+#define	NSR_RXRDY	0x01U	/* rx packet ready */
+
+#define	NSR_BITS	\
+	"\020"	\
+	"\010SPEED_10"	\
+	"\007LINKST_UP"	\
+	"\006WAKEST"	\
+	"\005TXFULL"	\
+	"\004TX2END"	\
+	"\003TX1END"	\
+	"\002RXOV"	\
+	"\001RXRDY"
+
+/* tx control register */
+#define	TCR_TJDIS	0x40U	/* tx jitter control */
+#define	TCR_EXCEDM	0x20U	/* excessive collision mode */
+#define	TCR_PAD_DIS2	0x10U	/* PAD appends disable for pkt2 */
+#define	TCR_CRC_DIS2	0x08U	/* CRC appends disable for pkt2 */
+#define	TCR_PAD_DIS1	0x04U	/* PAD appends disable for pkt1 */
+#define	TCR_CRC_DIS1	0x02U	/* CRC appends disable for pkt1 */
+
+#define	TCR_BITS	\
+	"\020"	\
+	"\007TJDIS"	\
+	"\006EXCEDM"	\
+	"\005PAD_DIS2"	\
+	"\004CRC_DIS2"	\
+	"\003PAD_DIS1"	\
+	"\002CRC_DIS1"
+
+/* tx status register (ro) */
+#define	TSR_TJTO	0x80U	/* tx jabber time out */
+#define	TSR_LC		0x40U	/* loss of carrier */
+#define	TSR_NC		0x20U	/* no carrier */
+#define	TSR_LATEC	0x10U	/* late collision */
+#define	TSR_COL		0x08U	/* late collision */
+#define	TSR_EL		0x04U	/* excessive collision */
+
+#define	TSR_BITS	\
+	"\020"		\
+	"\010TJTO"	\
+	"\007LC"	\
+	"\006NC"	\
+	"\005LATEC"	\
+	"\004COL"	\
+	"\003EL"
+
+/* rx control register */
+#define	RCR_WTDIS	0x40U	/* watch dog timer disable */
+#define	RCR_DIS_LONG	0x20U	/* discard longer packets than 1522 */
+#define	RCR_DIS_CRC	0x10U	/* discard crc error packets */
+#define	RCR_ALL		0x08U	/* pass all multicast */
+#define	RCR_RUNT	0x04U	/* pass runt packets */
+#define	RCR_PRMSC	0x02U	/* promiscuous mode */
+#define	RCR_RXEN	0x01U	/* rx enable */
+
+#define	RCR_BITS	\
+	"\020"		\
+	"\007WTDIS"	\
+	"\006DIS_LONG"	\
+	"\005DIS_CRC"	\
+	"\004ALL"	\
+	"\003RUNT"	\
+	"\002PRMSC"	\
+	"\001RXEN"
+
+/* rx status register */
+#define	RSR_RF		0x80U	/* runt frame */
+#define	RSR_MF		0x40U	/* multicast frame */
+#define	RSR_LCS		0x20U	/* late collision seen */
+#define	RSR_RWTO	0x10U	/* receive watchdog timeout */
+#define	RSR_PLE		0x08U	/* physical layer error */
+#define	RSR_AE		0x04U	/* alignment error */
+#define	RSR_CE		0x02U	/* crc error */
+#define	RSR_FOE		0x01U	/* fifo overflow error */
+
+#define	RSR_BITS	\
+	"\020"		\
+	"\010RF"	\
+	"\007MF"	\
+	"\006LCS"	\
+	"\005RWTO"	\
+	"\004PLE"	\
+	"\003AE"	\
+	"\002CE"	\
+	"\001FOE"
+
+/* receive overflow counter register */
+#define	ROCR_RXFU	0x80U	/* receive overflow counter overflow */
+#define	ROCR_ROC	0x7fU	/* receive overflow counter */
+
+#define	ROCR_BITS	\
+	"\020"		\
+	"\010RXFU"
+
+/* back pressure threshold register */
+#define	BPTR_BPHW	0xf0U	/* high water overflow threshold */
+#define		BPTR_BPHW_SHIFT	4
+#define		BPTR_BPHW_UNIT	1024U
+#define		BPTR_BPHW_DEFAULT	(3 << BPTR_BPHW_SHIFT)	/* 3k */
+#define	BPTR_JPT	0x0fU	/* jam pattern time */
+#define		BPTR_JPT_SHIFT	0
+#define		BPTR_JPT_5us	(0U << BPTR_JPT_SHIFT)
+#define		BPTR_JPT_10us	(1U << BPTR_JPT_SHIFT)
+#define		BPTR_JPT_15us	(2U << BPTR_JPT_SHIFT)
+#define		BPTR_JPT_25us	(3U << BPTR_JPT_SHIFT)
+#define		BPTR_JPT_50us	(4U << BPTR_JPT_SHIFT)
+#define		BPTR_JPT_100us	(5U << BPTR_JPT_SHIFT)
+#define		BPTR_JPT_150us	(6U << BPTR_JPT_SHIFT)
+#define		BPTR_JPT_200us	(7U << BPTR_JPT_SHIFT)
+#define		BPTR_JPT_250us	(8U << BPTR_JPT_SHIFT)
+#define		BPTR_JPT_300us	(9U << BPTR_JPT_SHIFT)
+#define		BPTR_JPT_350us	(10U << BPTR_JPT_SHIFT)
+#define		BPTR_JPT_400us	(11U << BPTR_JPT_SHIFT)
+#define		BPTR_JPT_450us	(12U << BPTR_JPT_SHIFT)
+#define		BPTR_JPT_500us	(13U << BPTR_JPT_SHIFT)
+#define		BPTR_JPT_550us	(14U << BPTR_JPT_SHIFT)
+#define		BPTR_JPT_600us	(15U << BPTR_JPT_SHIFT)
+
+/* flow control threshold register */
+#define	FCTR_HWOT	0xf0U	/* rx fifo high water overflow threshold */
+#define		FCTR_HWOT_SHIFT	4
+#define		FCTR_HWOT_UNIT	1024U
+#define	FCTR_LWOT	0x0fU	/* rx fifo low water overflow threshold */
+#define		FCTR_LWOT_SHIFT	0
+#define		FCTR_LWOT_UNIT	1024U
+
+/* rx/tx flow control register */
+#define	FCR_TXPO	0x80U	/* tx pause packet */
+#define	FCR_TXPF	0x40U	/* tx pause packet */
+#define	FCR_TXPEN	0x20U	/* tx pause packet */
+#define	FCR_BKPA	0x10U	/* back pressure mode */
+#define	FCR_BKPM	0x08U	/* back pressure mode */
+#define	FCR_BKPS	0x04U	/* rx pause packet current status (r/c) */
+#define	FCR_RXPCS	0x02U	/* rx pause packet current status (ro) */
+#define	FCR_FLCE	0x01U	/* flow control enbale */
+
+#define	FCR_BITS	\
+	"\020"		\
+	"\000TXPO"	\
+	"\000TXPF"	\
+	"\000TXPEN"	\
+	"\000BKPA"	\
+	"\000BKPM"	\
+	"\000BKPS"	\
+	"\000RXPCS"	\
+	"\000FLCE"
+
+/* EEPROM & PHY control register (0x0b) */
+#define	EPCR_REEP	0x20U	/* reload eeprom */
+#define	EPCR_WEP	0x10U	/* write eeprom enable */
+#define	EPCR_EPOS	0x08U	/* select device, 0:eeprom, 1:phy */
+#define	EPCR_ERPRR	0x04U	/* read command */
+#define	EPCR_ERPRW	0x02U	/* write command */
+#define	EPCR_ERRE	0x01U	/* eeprom/phy access in progress (ro) */
+
+#define	EPCR_BITS	\
+	"\020"		\
+	"\005REEP"	\
+	"\004WEP"	\
+	"\003EPOS"	\
+	"\002ERPRR"	\
+	"\001ERPRW"	\
+	"\000ERRE"
+
+/* EEPROM & PHY access register (0x0c) */
+#define	EPAR_PHYADR	0xc0U	/* phy address, internal phy(1) or external */
+#define		EPAR_PHYADR_SHIFT	6
+#define	EPAR_EROA	0x3fU	/* eeprom word addr or phy register addr */
+#define		EPAR_EROA_SHIFT	0
+
+/* EEPROM & PHY data register (0x0d(low)-0x0e(hi)) */
+
+/* wake up control register (0x0f) */
+#define	WCR_LINKEN	0x20U	/* enable link status event */
+#define	WCR_SAMPLEEN	0x10U	/* enable sample frame event */
+#define	WCR_MAGICEN	0x08U	/* enable magic pkt event */
+#define	WCR_LINKST	0x04U	/* link status change occur ro */
+#define	WCR_SAMPLEST	0x02U	/* sample frame rx occur ro */
+#define	WCR_MAGICST	0x01U	/* magic pkt rx occur ro */
+
+#define	WCR_BITS	\
+	"\020"		\
+	"\000LINKEN"	\
+	"\000SAMPLEEN"	\
+	"\000MAGICEN"	\
+	"\000LINKST"	\
+	"\000SAMPLEST"	\
+	"\000MAGICST"
+
+/* physical address register (0x10-0x15) */
+/* multicast address register (0x16-0x1c) */
+/* general purpose control register (0x1e) */
+#define	GPCR_GEPCTRL	0x7f
+#define		GPCR_OUT(n)	(1U << (n))
+
+#define	GPCR_BITS	\
+	"\020"		\
+	"\006OUT5"	\
+	"\005OUT4"	\
+	"\004OUT3"	\
+	"\003OUT2"	\
+	"\002OUT1"	\
+	"\001OUT0"
+
+/* general purpose register (0x1f) */
+#define	GPR_GEPIO5	0x20U
+#define	GPR_GEPIO4	0x10U
+#define	GPR_GEPIO3	0x08U
+#define	GPR_GEPIO2	0x04U
+#define	GPR_GEPIO1	0x02U
+#define	GPR_GEPIO0	0x01U
+
+#define	GPR_BITS	\
+	"\020"		\
+	"\006GEPIO5"	\
+	"\005GEPIO4"	\
+	"\004GEPIO3"	\
+	"\003GEPIO2"	\
+	"\002GEPIO1"	\
+	"\001GEPIO0"
+
+/* vendor id register (0x28-0x29) */
+/* product id register (0x2a-0x2b) */
+/* chip revision register (0x2c) */
+
+/* usb device address register (0xf0) */
+#define	USBDA_USBFA	0x3fU	/* usb device address */
+#define		USBDA_USBFA_SHIFT	0
+
+/* receive packet counter register (0xf1) */
+
+/* transmitpacket counter/usb status register (0xf2) */
+#define	TUSR_RXFAULT	0x80U	/* indicate rx has unexpected condition */
+#define	TUSR_SUSFLAG	0x40U	/* indicate device has suspended condition */
+#define	TUSR_EP1RDY	0x20U	/* ready for read from ep1 pipe */
+#define	TUSR_SRAM	0x18U	/* sram size 0:32K, 1:48K, 2:16K, 3:64K */
+#define		TUSR_SRAM_SHIFT	3
+#define		TUSR_SRAM_32K	(0U << TUSR_SRAM_SHIFT)
+#define		TUSR_SRAM_48K	(1U << TUSR_SRAM_SHIFT)
+#define		TUSR_SRAM_16K	(2U << TUSR_SRAM_SHIFT)
+#define		TUSR_SRAM_64K	(3U << TUSR_SRAM_SHIFT)
+#define	TUSR_TXC2	0x04U	/* two or more packets in tx buffer */
+#define	TUSR_TXC1	0x02U	/* one packet in tx buffer */
+#define	TUSR_TXC0	0x01U	/* no packet in tx buffer */
+
+#define	TUSR_BITS	\
+	"\020"		\
+	"\010RXFAULT"	\
+	"\007SUSFLAG"	\
+	"\006EP1RDY"	\
+	"\003TXC2"	\
+	"\002TXC1"	\
+	"\001TXC0"
+
+/* usb control register (0xf4) */
+#define	USBC_EP3ACK	0x20U	/* ep3 will alway return 8byte data if NAK=0*/
+#define	USBC_EP3NACK	0x10U	/* ep3 will alway return NAK */
+#define	USBC_MEMTST	0x01U
+
+/* bulk message format */
+#define	TX_HEADER_SIZE	2
+#define	RX_HEADER_SIZE	3
+
+/* interrupt msg format */
+struct intr_msg {
+	uint8_t	im_nsr;
+	uint8_t	im_tsr1;
+	uint8_t	im_tsr2;
+	uint8_t	im_rsr;
+	uint8_t	im_rocr;
+	uint8_t	im_rxc;
+	uint8_t	im_txc;
+	uint8_t	im_gpr;
+};
+#endif /* __DM9601_H__ */
diff --git a/usr/src/uts/common/io/udmf/udmf_usbgem.c b/usr/src/uts/common/io/udmf/udmf_usbgem.c
new file mode 100644
index 0000000000..0637de054b
--- /dev/null
+++ b/usr/src/uts/common/io/udmf/udmf_usbgem.c
@@ -0,0 +1,1036 @@
+/*
+ * udmfE_usbgem.c : Davicom DM9601E USB to Fast Ethernet Driver for Solaris
+ *
+ * Copyright (c) 2009-2012 Masayuki Murayama.  All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions are met:
+ *
+ * 1. Redistributions of source code must retain the above copyright notice,
+ *    this list of conditions and the following disclaimer.
+ *
+ * 2. Redistributions in binary form must reproduce the above copyright notice,
+ *    this list of conditions and the following disclaimer in the documentation
+ *    and/or other materials provided with the distribution.
+ *
+ * 3. Neither the name of the author nor the names of its contributors may be
+ *    used to endorse or promote products derived from this software without
+ *    specific prior written permission.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
+ * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
+ * COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
+ * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
+ * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS
+ * OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
+ * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+ * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
+ * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH
+ * DAMAGE.
+ */
+
+#pragma ident "%W% %E%"
+
+/*
+ *  Changelog:
+ */
+
+/*
+ * TODO
+ */
+/* ======================================================= */
+
+/*
+ * Solaris system header files and macros
+ */
+
+/* minimum kernel headers for drivers */
+#include <sys/types.h>
+#include <sys/conf.h>
+#include <sys/debug.h>
+#include <sys/kmem.h>
+#include <sys/modctl.h>
+#include <sys/errno.h>
+#include <sys/ddi.h>
+#include <sys/sunddi.h>
+#include <sys/byteorder.h>
+
+/* ethernet stuff */
+#include <sys/ethernet.h>
+
+/* interface card depend stuff */
+#include <sys/stropts.h>
+#include <sys/stream.h>
+#include <sys/strlog.h>
+#include <sys/usb/usba.h>
+#include "usbgem.h"
+
+/* hardware stuff */
+#include "usbgem_mii.h"
+#include "dm9601reg.h"
+
+char	ident[] = "dm9601 usbnic driver v" VERSION;
+
+/*
+ * Useful macros
+ */
+#define	CHECK_AND_JUMP(err, label)	if (err != USB_SUCCESS) goto label
+#define	LE16P(p)	((((uint8_t *)(p))[1] << 8) | ((uint8_t *)(p))[0])
+
+/*
+ * Debugging
+ */
+#ifdef DEBUG_LEVEL
+static int udmf_debug = DEBUG_LEVEL;
+#define	DPRINTF(n, args)	if (udmf_debug > (n)) cmn_err args
+#else
+#define	DPRINTF(n, args)
+#endif
+
+/*
+ * Our configration for dm9601
+ */
+/* timeouts */
+#define	ONESEC	(drv_usectohz(1*1000000))
+
+/*
+ * Local device definitions
+ */
+struct udmf_dev {
+	/*
+	 * Misc HW information
+	 */
+	uint8_t	rcr;
+	uint8_t	last_nsr;
+	uint8_t	mac_addr[ETHERADDRL];
+};
+
+/*
+ * private functions
+ */
+
+/* mii operations */
+static uint16_t udmf_mii_read(struct usbgem_dev *, uint_t, int *errp);
+static void udmf_mii_write(struct usbgem_dev *, uint_t, uint16_t, int *errp);
+
+/* nic operations */
+static int udmf_reset_chip(struct usbgem_dev *);
+static int udmf_init_chip(struct usbgem_dev *);
+static int udmf_start_chip(struct usbgem_dev *);
+static int udmf_stop_chip(struct usbgem_dev *);
+static int udmf_set_media(struct usbgem_dev *);
+static int udmf_set_rx_filter(struct usbgem_dev *);
+static int udmf_get_stats(struct usbgem_dev *);
+static void udmf_interrupt(struct usbgem_dev *, mblk_t *);
+
+/* packet operations */
+static mblk_t *udmf_tx_make_packet(struct usbgem_dev *, mblk_t *);
+static mblk_t *udmf_rx_make_packet(struct usbgem_dev *, mblk_t *);
+
+/* =============================================================== */
+/*
+ * I/O functions
+ */
+/* =============================================================== */
+#define	OUT(dp, ix, len, buf, errp, label)	\
+	if ((*(errp) = usbgem_ctrl_out((dp), 	\
+	/* bmRequestType */ USB_DEV_REQ_HOST_TO_DEV	\
+		    | USB_DEV_REQ_TYPE_VENDOR | USB_DEV_REQ_RCPT_DEV,	\
+	/* bRequest */	1,	\
+	/* wValue */	0,	\
+	/* wIndex */	(ix),	\
+	/* wLength */	(len),	\
+	/* value */	(buf),	\
+	/* size */	(len))) != USB_SUCCESS) goto label
+
+#define	OUTB(dp, ix, val, errp, label)	\
+	if ((*(errp) = usbgem_ctrl_out((dp), 	\
+	/* bmRequestType */ USB_DEV_REQ_HOST_TO_DEV	\
+		    | USB_DEV_REQ_TYPE_VENDOR | USB_DEV_REQ_RCPT_DEV,	\
+	/* bRequest */	3,	\
+	/* wValue */	(val),	\
+	/* wIndex */	(ix),	\
+	/* wLength */	0,	\
+	/* value */	NULL,	\
+	/* size */	0)) != USB_SUCCESS) goto label
+
+#define	IN(dp, ix, len, buf, errp, label)	\
+	if ((*(errp) = usbgem_ctrl_in((dp), 	\
+	/* bmRequestType */ USB_DEV_REQ_DEV_TO_HOST	\
+		    | USB_DEV_REQ_TYPE_VENDOR | USB_DEV_REQ_RCPT_DEV,	\
+	/* bRequest */	0,	\
+	/* wValue */	0,	\
+	/* wIndex */	(ix),	\
+	/* wLength */	(len),	\
+	/* valuep */	(buf),	\
+	/* size */	(len))) != USB_SUCCESS) goto label
+
+/* =============================================================== */
+/*
+ * Hardware manupilation
+ */
+/* =============================================================== */
+static void
+udmf_enable_phy(struct usbgem_dev *dp)
+{
+	int	err = USB_SUCCESS;
+
+	/* de-assert reset signal to phy */
+	OUTB(dp, GPCR, GPCR_OUT(0), &err, usberr);
+	OUTB(dp, GPR, 0, &err, usberr);
+usberr:
+	;
+}
+
+static int
+udmf_reset_chip(struct usbgem_dev *dp)
+{
+	int	err = USB_SUCCESS;
+
+	DPRINTF(2, (CE_CONT, "!%s: %s: called", dp->name, __func__));
+
+	OUTB(dp, NCR, NCR_LBK_NORMAL | NCR_RST, &err, usberr);
+	drv_usecwait(100);
+usberr:
+	return (err);
+}
+
+/*
+ * Setup dm9601
+ */
+static int
+udmf_init_chip(struct usbgem_dev *dp)
+{
+	int		i;
+	uint32_t	val;
+	int		err = USB_SUCCESS;
+	uint16_t	reg;
+	uint8_t		buf[2];
+	struct udmf_dev	*lp = dp->private;
+
+	DPRINTF(2, (CE_CONT, "!%s: %s: called", dp->name, __func__));
+
+	OUTB(dp, NCR, NCR_LBK_NORMAL, &err, usberr);
+
+	/* tx control regiser: enable padding and crc generation */
+	OUTB(dp, TCR, 0, &err, usberr);
+
+	/* rx control register: will be set later by udmf_set_rx_filer() */
+	lp->rcr = RCR_RUNT;
+
+	/* back pressure threshold: */
+	OUTB(dp, BPTR, (2 << BPTR_BPHW_SHIFT) | BPTR_JPT_200us,
+	    &err, usberr);
+
+	/* flow control threshold: same as default */
+	OUTB(dp, FCTR, (3 << FCTR_HWOT_SHIFT) | (8 << FCTR_LWOT_SHIFT),
+	    &err, usberr);
+
+	/* usb control register */
+	OUTB(dp, USBC, USBC_EP3ACK | 0x06, &err, usberr);
+
+	/* flow control: will be set later by udmf_set_media() */
+
+	/* wake up control register: */
+	OUTB(dp, WCR, 0, &err, usberr);
+
+usberr:
+	DPRINTF(2, (CE_CONT, "!%s: %s: end err:%d(%s)",
+	    dp->name, __func__,
+	    err, err == USB_SUCCESS ? "success" : "error"));
+	return (err);
+}
+
+static int
+udmf_start_chip(struct usbgem_dev *dp)
+{
+	int	err = USB_SUCCESS;
+	struct udmf_dev	*lp = dp->private;
+
+	/* enable Rx */
+	lp->rcr |= RCR_RXEN;
+	OUTB(dp, RCR, lp->rcr, &err, usberr);
+
+usberr:
+	DPRINTF(2, (CE_CONT, "!%s: %s: end err:%d(%s)",
+	    dp->name, __func__,
+	    err, err == USB_SUCCESS ? "success" : "error"));
+	return (err);
+}
+
+static int
+udmf_stop_chip(struct usbgem_dev *dp)
+{
+	int	err = USB_SUCCESS;
+	struct udmf_dev	*lp = dp->private;
+
+	/* disable rx */
+	lp->rcr &= ~RCR_RXEN;
+	OUTB(dp, RCR, lp->rcr, &err, usberr);
+
+usberr:
+	DPRINTF(2, (CE_CONT, "!%s: %s: end err:%d(%s)",
+	    dp->name, __func__,
+	    err, err == USB_SUCCESS ? "success" : "error"));
+	return (err);
+}
+
+static int
+udmf_get_stats(struct usbgem_dev *dp)
+{
+	/* EMPTY */
+	return (USB_SUCCESS);
+}
+
+static uint_t
+udmf_mcast_hash(struct usbgem_dev *dp, const uint8_t *addr)
+{
+	return (usbgem_ether_crc_le(addr) & 0x3f);
+}
+
+static int
+udmf_set_rx_filter(struct usbgem_dev *dp)
+{
+	int		i;
+	uint8_t		rcr;
+	uint8_t		mode;
+	uint8_t		mhash[8];
+	uint8_t		*mac;
+	uint_t		h;
+	int		err = USB_SUCCESS;
+	struct udmf_dev	*lp = dp->private;
+	static uint8_t	invalid_mac[ETHERADDRL] = {0, 0, 0, 0, 0, 0};
+
+	DPRINTF(2, (CE_CONT, "!%s: %s: called, rxmode:%x",
+	    dp->name, __func__, dp->rxmode));
+
+	if (lp->rcr & RCR_RXEN) {
+		/* set promiscuous mode before changing rx filter mode */
+		OUTB(dp, RCR, lp->rcr | RCR_PRMSC, &err, usberr);
+	}
+
+	lp->rcr &= ~(RCR_ALL | RCR_PRMSC);
+	mode = 0;
+	bzero(mhash, sizeof (mhash));
+	mac = dp->cur_addr.ether_addr_octet;
+
+	if ((dp->rxmode & RXMODE_ENABLE) == 0) {
+		mac = invalid_mac;
+	} else if (dp->rxmode & RXMODE_PROMISC) {
+		/* promiscious mode implies all multicast and all physical */
+		mode |= RCR_PRMSC;
+	} else if ((dp->rxmode & RXMODE_ALLMULTI) || dp->mc_count > 32) {
+		/* accept all multicast packets */
+		mode |= RCR_ALL;
+	} else if (dp->mc_count > 0) {
+		/*
+		 * make hash table to select interresting
+		 * multicast address only.
+		 */
+		for (i = 0; i < dp->mc_count; i++) {
+			/* hash table is 64 = 2^6 bit width */
+			h = dp->mc_list[i].hash;
+			mhash[h / 8] |= 1 << (h % 8);
+		}
+	}
+
+	/* set node address */
+	if (bcmp(mac, lp->mac_addr, ETHERADDRL) != 0) {
+		OUT(dp, PAR, ETHERADDRL, dp->cur_addr.ether_addr_octet,
+		    &err, usberr);
+		bcopy(mac, lp->mac_addr, ETHERADDRL);
+	}
+
+	/* set multicast hash table */
+	OUT(dp, MAR, sizeof (mhash), &mhash[0], &err, usberr);
+
+	/* update rcr */
+	lp->rcr |= mode;
+	OUTB(dp, RCR, lp->rcr, &err, usberr);
+
+#if DEBUG_LEVEL > 1
+	/* verify rcr */
+	IN(dp, RCR, 1, &rcr, &err, usberr);
+	cmn_err(CE_CONT, "!%s: %s: rcr:%b returned",
+	    dp->name, __func__, rcr, RCR_BITS);
+#endif
+usberr:
+	DPRINTF(2, (CE_CONT, "!%s: %s: end err:%d(%s)",
+	    dp->name, __func__,
+	    err, err == USB_SUCCESS ? "success" : "error"));
+	return (err);
+}
+
+static int
+udmf_set_media(struct usbgem_dev *dp)
+{
+	int	err = USB_SUCCESS;
+	uint8_t	fcr;
+	struct udmf_dev	*lp = dp->private;
+
+	DPRINTF(0, (CE_CONT, "!%s: %s: called", dp->name, __func__));
+
+	/* setup flow control */
+	fcr = 0;
+	if (dp->full_duplex) {
+		/* select flow control */
+		switch (dp->flow_control) {
+		case FLOW_CONTROL_RX_PAUSE:
+			fcr |= FCR_FLCE;
+			break;
+
+		case FLOW_CONTROL_TX_PAUSE:
+			fcr |= FCR_TXPEN;
+			break;
+
+		case FLOW_CONTROL_SYMMETRIC:
+			fcr |= FCR_FLCE | FCR_TXPEN;
+			break;
+		}
+	}
+
+	/* update flow control register */
+	OUTB(dp, FCR, fcr, &err, usberr);
+
+usberr:
+	DPRINTF(2, (CE_CONT, "!%s: %s: end err:%d(%s)",
+	    dp->name, __func__,
+	    err, err == USB_SUCCESS ? "success" : "error"));
+	return (err);
+}
+
+/*
+ * send/receive packet check
+ */
+static mblk_t *
+udmf_tx_make_packet(struct usbgem_dev *dp, mblk_t *mp)
+{
+	int		n;
+	size_t		pkt_size;
+	mblk_t		*new;
+	mblk_t		*tp;
+	uint8_t		*bp;
+	uint8_t		*last_pos;
+	uint_t		align_mask;
+
+	pkt_size = msgdsize(mp);
+	align_mask = 63;
+
+	/*
+	 * re-allocate the mp
+	 */
+
+	/* minimum ethernet packet size of ETHERMIN */
+	pkt_size = max(pkt_size, ETHERMIN);
+
+#if 0 /* CONFIG_ADD_TX_DELIMITOR_ALWAYS */
+	pkt_size += TX_HEADER_SIZE;
+#endif
+	if (((pkt_size + TX_HEADER_SIZE) & align_mask) == 0) {
+		/* padding is required in usb communication */
+		pkt_size += TX_HEADER_SIZE;
+	}
+
+	if ((new = allocb(TX_HEADER_SIZE + pkt_size, 0)) == NULL) {
+		return (NULL);
+	}
+	new->b_wptr = new->b_rptr + TX_HEADER_SIZE + pkt_size;
+
+	/* add a header */
+	bp = new->b_rptr;
+	bp[0] = (uint8_t)pkt_size;
+	bp[1] = (uint8_t)(pkt_size >> 8);
+	bp += TX_HEADER_SIZE;
+
+	/* copy contents of the buffer */
+	for (tp = mp; tp; tp = tp->b_cont) {
+		n = tp->b_wptr - tp->b_rptr;
+		bcopy(tp->b_rptr, bp, n);
+		bp += n;
+	}
+
+	/* clear the rest including the next zero length header */
+	last_pos = new->b_wptr;
+	while (bp < last_pos) {
+		*bp++ = 0;
+	}
+
+	return (new);
+}
+
+static void
+udmf_dump_packet(struct usbgem_dev *dp, uint8_t *bp, int n)
+{
+	int	i;
+
+	for (i = 0; i < n; i += 8, bp += 8) {
+		cmn_err(CE_CONT, "%02x %02x %02x %02x %02x %02x %02x %02x",
+		    bp[0], bp[1], bp[2], bp[3], bp[4], bp[5], bp[6], bp[7]);
+	}
+}
+
+static mblk_t *
+udmf_rx_make_packet(struct usbgem_dev *dp, mblk_t *mp)
+{
+	int	len;
+	uint8_t	rx_stat;
+
+	len = mp->b_wptr - mp->b_rptr;
+
+	if (len <= RX_HEADER_SIZE) {
+		/*
+		 * the usb bulk-in frame doesn't include a valid
+		 * ethernet packet.
+		 */
+		return (NULL);
+	}
+
+	/* remove rx header */
+	rx_stat = mp->b_rptr[0];
+	if (rx_stat & (RSR_RF |  RSR_LCS | RSR_RWTO |
+	    RSR_PLE | RSR_AE | RSR_CE |  RSR_FOE)) {
+		if (rx_stat & RSR_RF) {
+			dp->stats.runt++;
+		}
+		if (rx_stat & RSR_LCS) {
+			/* late collision */
+			dp->stats.rcv_internal_err++;
+		}
+		if (rx_stat & RSR_RWTO) {
+			/* rx timeout */
+			dp->stats.rcv_internal_err++;
+		}
+		if (rx_stat & RSR_PLE) {
+			/* physical layer error */
+			dp->stats.rcv_internal_err++;
+		}
+		if (rx_stat & RSR_AE) {
+			/* alignment error */
+			dp->stats.frame++;
+		}
+		if (rx_stat & RSR_CE) {
+			/* crc error */
+			dp->stats.crc++;
+		}
+		if (rx_stat & RSR_FOE) {
+			/* fifo overflow error */
+			dp->stats.overflow++;
+		}
+		dp->stats.errrcv++;
+	}
+	len = LE16P(&mp->b_rptr[1]);
+	if (len >= ETHERFCSL) {
+		len -= ETHERFCSL;
+	}
+	mp->b_rptr += RX_HEADER_SIZE;
+	mp->b_wptr = mp->b_rptr + len;
+
+	return (mp);
+}
+
+/*
+ * MII Interfaces
+ */
+static uint16_t
+udmf_ep_read(struct usbgem_dev *dp, uint_t which, uint_t addr, int *errp)
+{
+	int	i;
+	uint8_t	epcr;
+	uint16_t	val;
+
+	DPRINTF(4, (CE_CONT, "!%s: %s: called, ix:%d",
+	    dp->name, __func__, addr));
+
+	OUTB(dp, EPAR, addr, errp, usberr);
+	OUTB(dp, EPCR, which | EPCR_ERPRR, errp, usberr);
+
+	for (i = 0; i < 100; i++) {
+		IN(dp, EPCR, sizeof (epcr), &epcr, errp, usberr);
+		if ((epcr & EPCR_ERRE) == 0) {
+			/* done */
+			IN(dp, EPDR, sizeof (val), &val, errp, usberr);
+			val = LE_16(val);
+			goto done;
+		}
+		drv_usecwait(10);
+	}
+	/* timeout */
+	cmn_err(CE_WARN, "!%s: %s: timeout", dp->name, __func__);
+	val = 0;
+done:
+	OUTB(dp, EPCR, 0, errp, usberr);
+	return (val);
+
+usberr:
+	DPRINTF(2, (CE_CONT, "!%s: %s: end err:%d(%s)",
+	    dp->name, __func__,
+	    *errp, *errp == USB_SUCCESS ? "success" : "error"));
+	return (0);
+}
+
+static void
+udmf_ep_write(struct usbgem_dev *dp, uint_t which, uint_t addr,
+    uint16_t val, int *errp)
+{
+	int	i;
+	uint8_t	epcr;
+
+	DPRINTF(5, (CE_CONT, "!%s: %s called", dp->name, __func__));
+
+	val = LE_16(val);
+	OUT(dp, EPDR, sizeof (val), &val, errp, usberr);
+
+	OUTB(dp, EPAR, addr, errp, usberr);
+
+	OUTB(dp, EPCR, which | EPCR_WEP | EPCR_ERPRW, errp, usberr);
+
+	for (i = 0; i < 100; i++) {
+		IN(dp, EPCR, 1, &epcr, errp, usberr);
+		if ((epcr & EPCR_ERRE) == 0) {
+			/* done */
+			goto done;
+		}
+		drv_usecwait(10);
+	}
+	/* timeout */
+	cmn_err(CE_WARN, "!%s: %s: timeout", dp->name, __func__);
+done:
+	OUTB(dp, EPCR, 0, errp, usberr);
+	return;
+
+usberr:
+	DPRINTF(2, (CE_CONT, "!%s: %s: end err:%d(%s)",
+	    dp->name, __func__,
+	    *errp, *errp == USB_SUCCESS ? "success" : "error"));
+}
+
+static uint16_t
+udmf_mii_read(struct usbgem_dev *dp, uint_t index, int *errp)
+{
+	uint16_t	val;
+
+	val = udmf_ep_read(dp, EPCR_EPOS,
+	    (dp->mii_phy_addr << EPAR_PHYADR_SHIFT) | index, errp);
+
+	return (val);
+}
+
+static void
+udmf_mii_write(struct usbgem_dev *dp, uint_t index, uint16_t val, int *errp)
+{
+	udmf_ep_write(dp, EPCR_EPOS,
+	    (dp->mii_phy_addr << EPAR_PHYADR_SHIFT) | index, val, errp);
+}
+
+static void
+udmf_interrupt(struct usbgem_dev *dp, mblk_t *mp)
+{
+	struct intr_msg	*imp;
+	struct udmf_dev	*lp = dp->private;
+
+	imp = (struct intr_msg *)&mp->b_rptr[0];
+
+	DPRINTF(4, (CE_CONT,
+	    "!%s: %s: size:%d, nsr:%b tsr1:%b tsr2:%b"
+	    " rsr:%b rocr:%b rxc:%02x txc:%b gpr:%b",
+	    dp->name, __func__, mp->b_wptr - mp->b_rptr,
+	    imp->im_nsr, NSR_BITS,
+	    imp->im_tsr1, TSR_BITS,
+	    imp->im_tsr2, TSR_BITS,
+	    imp->im_rsr, RSR_BITS,
+	    imp->im_rocr, ROCR_BITS,
+	    imp->im_rxc,
+	    imp->im_txc, TUSR_BITS,
+	    imp->im_gpr, GPR_BITS));
+
+	if ((lp->last_nsr ^ imp->im_nsr) & NSR_LINKST) {
+		usbgem_mii_update_link(dp);
+	}
+
+	lp->last_nsr = imp->im_nsr;
+}
+
+/* ======================================================== */
+/*
+ * OS depend (device driver DKI) routine
+ */
+/* ======================================================== */
+static uint16_t
+udmf_eeprom_read(struct usbgem_dev *dp, uint_t index, int *errp)
+{
+	uint16_t	val;
+
+	val = udmf_ep_read(dp, 0, index, errp);
+
+	return (val);
+}
+
+#ifdef DEBUG_LEVEL
+static void
+udmf_eeprom_dump(struct usbgem_dev *dp, int size)
+{
+	int	i;
+	int	err;
+	uint16_t	w0, w1, w2, w3;
+
+	cmn_err(CE_CONT, "!%s: eeprom dump:", dp->name);
+
+	err = USB_SUCCESS;
+
+	for (i = 0; i < size; i += 4) {
+		w0 = udmf_eeprom_read(dp, i + 0, &err);
+		w1 = udmf_eeprom_read(dp, i + 1, &err);
+		w2 = udmf_eeprom_read(dp, i + 2, &err);
+		w3 = udmf_eeprom_read(dp, i + 3, &err);
+		cmn_err(CE_CONT, "!0x%02x: 0x%04x 0x%04x 0x%04x 0x%04x",
+		    i, w0, w1, w2, w3);
+	}
+usberr:
+	;
+}
+#endif
+
+static int
+udmf_attach_chip(struct usbgem_dev *dp)
+{
+	int	i;
+	uint_t	val;
+	uint8_t	*m;
+	int	err;
+	struct udmf_dev	*lp = dp->private;
+
+	DPRINTF(0, (CE_CONT, "!%s: %s enter", dp->name, __func__));
+
+	/*
+	 * get mac address from EEPROM
+	 */
+	m = dp->dev_addr.ether_addr_octet;
+	for (i = 0; i < ETHERADDRL; i += 2)  {
+		val = udmf_eeprom_read(dp, i/2, &err);
+		m[i + 0] = (uint8_t)val;
+		m[i + 1] = (uint8_t)(val >> 8);
+	}
+
+	/* invalidate a private cache for mac addr */
+	bzero(lp->mac_addr, sizeof (lp->mac_addr));
+#ifdef CONFIG_VLAN
+	dp->misc_flag = USBGEM_VLAN;
+#endif
+#if DEBUG_LEVEL > 0
+	udmf_eeprom_dump(dp, /* 0x3f + 1 */ 128);
+#endif
+{
+	static uint8_t bcst[6] = {0xff, 0xff, 0xff, 0xff, 0xff, 0xff};
+	DPRINTF(0, (CE_CONT, "!%s: %s: hash of bcast:%x",
+	    dp->name, __func__, usbgem_ether_crc_be(bcst)));
+}
+	return (USB_SUCCESS);
+
+usberr:
+	cmn_err(CE_WARN, "%s: %s: usb error detected (%d)",
+	    dp->name, __func__, err);
+	return (USB_FAILURE);
+}
+
+static int
+udmf_mii_probe(struct usbgem_dev *dp)
+{
+	DPRINTF(2, (CE_CONT, "!%s: %s: called", dp->name, __func__));
+
+	udmf_enable_phy(dp);
+	return (usbgem_mii_probe_default(dp));
+}
+
+static int
+udmf_mii_init(struct usbgem_dev *dp)
+{
+	DPRINTF(2, (CE_CONT, "!%s: %s: called", dp->name, __func__));
+	udmf_enable_phy(dp);
+	return (USB_SUCCESS);
+}
+
+static int
+udmfattach(dev_info_t *dip, ddi_attach_cmd_t cmd)
+{
+	int			i;
+	ddi_iblock_cookie_t	c;
+	int			ret;
+	int			revid;
+	int			unit;
+	int			len;
+	const char		*drv_name;
+	struct usbgem_dev	*dp;
+	void			*base;
+	struct usbgem_conf	*ugcp;
+	struct udmf_dev		*lp;
+
+	unit = ddi_get_instance(dip);
+	drv_name = ddi_driver_name(dip);
+
+	DPRINTF(3, (CE_CONT, "!%s%d: %s: called, cmd:%d",
+	    drv_name, unit, __func__, cmd));
+
+	if (cmd == DDI_ATTACH) {
+		/*
+		 * construct usbgem configration
+		 */
+		ugcp = kmem_zalloc(sizeof (*ugcp), KM_SLEEP);
+
+		/* name */
+		/*
+		 * softmac requires that ppa is the instance number
+		 * of the device, otherwise it hangs in seaching the device.
+		 */
+		sprintf(ugcp->usbgc_name, "%s%d", drv_name, unit);
+		ugcp->usbgc_ppa = unit;
+
+		ugcp->usbgc_ifnum = 0;
+		ugcp->usbgc_alt = 0;
+
+		ugcp->usbgc_tx_list_max = 64;
+
+		ugcp->usbgc_rx_header_len = RX_HEADER_SIZE;
+		ugcp->usbgc_rx_list_max = 64;
+
+		/* time out parameters */
+		ugcp->usbgc_tx_timeout = USBGEM_TX_TIMEOUT;
+		ugcp->usbgc_tx_timeout_interval = USBGEM_TX_TIMEOUT_INTERVAL;
+#if 1
+		/* flow control */
+		ugcp->usbgc_flow_control = FLOW_CONTROL_RX_PAUSE;
+#else
+		/*
+		 * XXX - flow control caused link down frequently under
+		 * heavy traffic
+		 */
+		ugcp->usbgc_flow_control = FLOW_CONTROL_NONE;
+#endif
+		/* MII timeout parameters */
+		ugcp->usbgc_mii_link_watch_interval =
+		    USBGEM_LINK_WATCH_INTERVAL;
+		ugcp->usbgc_mii_an_watch_interval = 
+		    USBGEM_LINK_WATCH_INTERVAL/5;
+		ugcp->usbgc_mii_reset_timeout = MII_RESET_TIMEOUT; /* 1 sec */
+		ugcp->usbgc_mii_an_timeout = MII_AN_TIMEOUT;	/* 5 sec */
+		ugcp->usbgc_mii_an_wait = (25*ONESEC)/10;
+		ugcp->usbgc_mii_linkdown_timeout = MII_LINKDOWN_TIMEOUT;
+
+		ugcp->usbgc_mii_an_delay = ONESEC/10;
+		ugcp->usbgc_mii_linkdown_action = MII_ACTION_RSA;
+		ugcp->usbgc_mii_linkdown_timeout_action = MII_ACTION_RESET;
+		ugcp->usbgc_mii_dont_reset = B_FALSE;
+		ugcp->usbgc_mii_hw_link_detection = B_TRUE;
+
+		/* I/O methods */
+
+		/* mac operation */
+		ugcp->usbgc_attach_chip = &udmf_attach_chip;
+		ugcp->usbgc_reset_chip = &udmf_reset_chip;
+		ugcp->usbgc_init_chip = &udmf_init_chip;
+		ugcp->usbgc_start_chip = &udmf_start_chip;
+		ugcp->usbgc_stop_chip = &udmf_stop_chip;
+		ugcp->usbgc_multicast_hash = &udmf_mcast_hash;
+
+		ugcp->usbgc_set_rx_filter = &udmf_set_rx_filter;
+		ugcp->usbgc_set_media = &udmf_set_media;
+		ugcp->usbgc_get_stats = &udmf_get_stats;
+		ugcp->usbgc_interrupt = &udmf_interrupt;
+
+		/* packet operation */
+		ugcp->usbgc_tx_make_packet = &udmf_tx_make_packet;
+		ugcp->usbgc_rx_make_packet = &udmf_rx_make_packet;
+
+		/* mii operations */
+		ugcp->usbgc_mii_probe = &udmf_mii_probe;
+		ugcp->usbgc_mii_init = &udmf_mii_init;
+		ugcp->usbgc_mii_config = &usbgem_mii_config_default;
+		ugcp->usbgc_mii_read = &udmf_mii_read;
+		ugcp->usbgc_mii_write = &udmf_mii_write;
+		ugcp->usbgc_mii_addr_min = 1;
+
+		/* mtu */
+		ugcp->usbgc_min_mtu = ETHERMTU;
+		ugcp->usbgc_max_mtu = ETHERMTU;
+		ugcp->usbgc_default_mtu = ETHERMTU;
+		
+		lp = kmem_zalloc(sizeof (struct udmf_dev), KM_SLEEP);
+		lp->last_nsr;
+
+		ddi_set_driver_private(dip, NULL);
+
+		dp = usbgem_do_attach(dip, ugcp, lp, sizeof (struct udmf_dev));
+
+		kmem_free(ugcp, sizeof (*ugcp));
+
+		if (dp != NULL) {
+			return (DDI_SUCCESS);
+		}
+
+err_free_mem:
+		kmem_free(lp, sizeof (struct udmf_dev));
+err_close_pipe:
+err:
+		return (DDI_FAILURE);
+	}
+
+	if (cmd == DDI_RESUME) {
+		return (usbgem_resume(dip));
+	}
+
+	return (DDI_FAILURE);
+}
+
+static int
+udmfdetach(dev_info_t *dip, ddi_detach_cmd_t cmd)
+{
+	int	ret;
+
+	if (cmd == DDI_DETACH) {
+		ret = usbgem_do_detach(dip);
+		if (ret != DDI_SUCCESS) {
+			return (DDI_FAILURE);
+		}
+		return (DDI_SUCCESS);
+	}
+	if (cmd == DDI_SUSPEND) {
+		return (usbgem_suspend(dip));
+	}
+	return (DDI_FAILURE);
+}
+
+/* ======================================================== */
+/*
+ * OS depend (loadable streams driver) routine
+ */
+/* ======================================================== */
+#ifdef USBGEM_CONFIG_GLDv3
+USBGEM_STREAM_OPS(udmf_ops, udmfattach, udmfdetach);
+#else
+static	struct module_info udmfminfo = {
+	0,			/* mi_idnum */
+	"udmf",			/* mi_idname */
+	0,			/* mi_minpsz */
+	ETHERMTU,		/* mi_maxpsz */
+	ETHERMTU*128,		/* mi_hiwat */
+	1,			/* mi_lowat */
+};
+
+static	struct qinit udmfrinit = {
+	(int (*)()) NULL,	/* qi_putp */
+	usbgem_rsrv,		/* qi_srvp */
+	usbgem_open,		/* qi_qopen */
+	usbgem_close,		/* qi_qclose */
+	(int (*)()) NULL,	/* qi_qadmin */
+	&udmfminfo,		/* qi_minfo */
+	NULL			/* qi_mstat */
+};
+
+static	struct qinit udmfwinit = {
+	usbgem_wput,		/* qi_putp */
+	usbgem_wsrv,		/* qi_srvp */
+	(int (*)()) NULL,	/* qi_qopen */
+	(int (*)()) NULL,	/* qi_qclose */
+	(int (*)()) NULL,	/* qi_qadmin */
+	&udmfminfo,		/* qi_minfo */
+	NULL			/* qi_mstat */
+};
+
+static struct streamtab	udmf_info = {
+	&udmfrinit,	/* st_rdinit */
+	&udmfwinit,	/* st_wrinit */
+	NULL,		/* st_muxrinit */
+	NULL		/* st_muxwrinit */
+};
+
+static	struct cb_ops cb_udmf_ops = {
+	nulldev,	/* cb_open */
+	nulldev,	/* cb_close */
+	nodev,		/* cb_strategy */
+	nodev,		/* cb_print */
+	nodev,		/* cb_dump */
+	nodev,		/* cb_read */
+	nodev,		/* cb_write */
+	nodev,		/* cb_ioctl */
+	nodev,		/* cb_devmap */
+	nodev,		/* cb_mmap */
+	nodev,		/* cb_segmap */
+	nochpoll,	/* cb_chpoll */
+	ddi_prop_op,	/* cb_prop_op */
+	&udmf_info,	/* cb_stream */
+	D_NEW|D_MP	/* cb_flag */
+};
+
+static	struct dev_ops udmf_ops = {
+	DEVO_REV,	/* devo_rev */
+	0,		/* devo_refcnt */
+	usbgem_getinfo,	/* devo_getinfo */
+	nulldev,	/* devo_identify */
+	nulldev,	/* devo_probe */
+	udmfattach,	/* devo_attach */
+	udmfdetach,	/* devo_detach */
+	nodev,		/* devo_reset */
+	&cb_udmf_ops,	/* devo_cb_ops */
+	NULL,		/* devo_bus_ops */
+        usbgem_power,   /* devo_power */
+#if DEVO_REV >= 4
+	usbgem_quiesce, /* devo_quiesce */
+#endif
+};
+#endif
+
+static struct modldrv modldrv = {
+	&mod_driverops,	/* Type of module.  This one is a driver */
+	ident,
+	&udmf_ops,	/* driver ops */
+};
+
+static struct modlinkage modlinkage = {
+	MODREV_1, &modldrv, NULL
+};
+
+/* ======================================================== */
+/*
+ * _init : done
+ */
+/* ======================================================== */
+int
+_init(void)
+{
+	int 	status;
+
+	DPRINTF(2, (CE_CONT, "!udmf: _init: called"));
+
+	status = usbgem_mod_init(&udmf_ops, "udmf");
+	if (status != DDI_SUCCESS) {
+		return (status);
+	}
+	status = mod_install(&modlinkage);
+	if (status != DDI_SUCCESS) {
+		usbgem_mod_fini(&udmf_ops);
+	}
+	return (status);
+}
+
+/*
+ * _fini : done
+ */
+int
+_fini(void)
+{
+	int	status;
+
+	DPRINTF(2, (CE_CONT, "!udmf: _fini: called"));
+	status = mod_remove(&modlinkage);
+	if (status == DDI_SUCCESS) {
+		usbgem_mod_fini(&udmf_ops);
+	}
+	return (status);
+}
+
+int
+_info(struct modinfo *modinfop)
+{
+	return (mod_info(&modlinkage, modinfop));
+}
diff --git a/usr/src/uts/common/io/upf/adm8511reg.h b/usr/src/uts/common/io/upf/adm8511reg.h
new file mode 100644
index 0000000000..68a2207bb5
--- /dev/null
+++ b/usr/src/uts/common/io/upf/adm8511reg.h
@@ -0,0 +1,205 @@
+/*
+ * @(#)adm8511reg.h	1.1 09/06/20
+ * Register dehinitsions of ADMtek ADM8511 Fast Ethernet to USB controller.
+ * Codeded by Masayuki Murayama(KHF04453@nifty.ne.jp)
+ * This file is public domain.
+ */
+
+#define	EC0		0x00	/* B */
+#define	EC1		0x01	/* B */
+#define	EC2		0x02	/* B */
+#define	MA		0x08	/* 8byte array */
+#define	EID		0x10	/* B */
+#define	PAUSETIMER	0x18	/* B pause timer */
+#define	RPNBFC		0x1a	/* B */
+#define	ORFBFC		0x1b	/* B */
+#define	EP1C		0x1c	/* B */
+#define	RXFC		0x1d	/* B */
+#define	BIST		0x1e	/* B */
+#define	EEOFFSET	0x20	/* B */
+#define	EEDATA		0x21	/* W */
+#define	EECTRL		0x23	/* B */
+#define	PHYA		0x25	/* B */
+#define	PHYD		0x26 	/* W */
+#define	PHYAC		0x28	/* B */
+#define	USBSTAT		0x2a	/* B */
+#define	ETHTXSTAT	0x2b	/* W */
+#define	ETHRXSTAT	0x2d	/* B */
+#define	LOSTCNT		0x2e	/* W */
+#define	WF0MASK		0x30	/* 16byte array */
+#define	WF0OFFSET	0x40	/* W */
+#define	WF0CRC		0x41	/* W */
+#define	WF1MASK		0x48	/* 16byte array */
+#define	WF1OFFSET	0x58	/* W */
+#define	WF1CRC		0x59	/* W */
+#define	WF2MASK		0x60	/* 16byte array */
+#define	WF2OFFSET	0x70	/* W */
+#define	WF2CRC		0x71	/* W */
+#define	WCTRL		0x78	/* B */
+#define	WSTAT		0x7a	/* B */
+#define	IPHYC		0x7b	/* B */
+#define	GPIO54		0x7c	/* B */
+#define	GPIO10		0x7e	/* B */
+#define	GPIO32		0x7f	/* B */
+#define	TEST		0x80	/* B */
+#define	TM		0x81	/* B */
+#define	RPN		0x82	/* B */
+
+/* Ethernet control register 0: offset 0 */
+#define	EC0_TXE		0x80U
+#define	EC0_RXE		0x40U
+#define	EC0_RXFCE	0x20U
+#define	EC0_WOE		0x10U
+#define	EC0_RXSA	0x08U
+#define	EC0_SBO		0x04U
+#define	EC0_RXMA	0x02U
+#define	EC0_RXCS	0x01U
+
+#define	EC0_BITS	\
+	"\020"	\
+	"\010TXE"	\
+	"\007RXE"	\
+	"\006RXFCE"	\
+	"\005WOE"	\
+	"\004RXSA"	\
+	"\003SBO"	\
+	"\002RXMA"	\
+	"\001RXCS"
+
+/* Ethernet control register 1: offset 1 */
+#define	EC1_FD		0x20U
+#define	EC1_100M	0x10U	/* 0:10Mbps 1:100Mbps */
+#define	EC1_RM		0x08U	/* reset mac */
+
+#define	EC1_BITS	\
+	"\020"	\
+	"\006FD"	\
+	"\005100M"	\
+	"\004RM"
+
+/* Ethernet control register 2: offset 2 */
+#define	EC2_MEPL	0x80U	/* 8515: MTU 0:1528, 1:1638 */
+#define	EC2_RPNC	0x40U
+#define	EC2_LEEPRS	0x20U
+#define	EC2_EEPRW	0x10U
+#define	EC2_LB		0x08U
+#define	EC2_PROM	0x04U
+#define	EC2_RXBP	0x02U
+#define	EC2_EP3RC	0x01U
+
+#define	EC2_BITS	\
+	"\020"	\
+	"\010MEPS"	\
+	"\007RPNC"	\
+	"\006LEEPRS"	\
+	"\005EEPRW"	\
+	"\004LB"	\
+	"\003PROM"	\
+	"\002RXBP"	\
+	"\001EP3RC"
+
+/* Recieve Packet number based Flow Control register: offset 0x1a */
+#define	RPNBFC_PN	0x7eU	/* */
+#define		RPNBFC_PN_SHIFT	1
+#define	RPNBFC_FCP	0x01U	/* enable rx flow control */
+
+/* Occupied Recieve FIFO based Flow Control register: offset 0x1b */
+#define	ORFBFC_RXS	0x7eU	/* */
+#define		ORFBFC_RXS_SHIFT	1
+#define		ORFBFC_RXS_UNIT	1024U
+#define	ORFBFC_FCRXS	0x01U	/* enable rx flow control */
+
+/* EP1 control register: offset 0x1c */
+#define	EP1C_EP1S0E	0x80U	/* send 0 enable */
+#define	EP1C_ITMA	0x60U	/* internal test mode A */
+#define	EP1C_ITMB	0x1fU	/* internal test mode B */
+
+#define	EP1C_BITS	\
+	"\020"	\
+	"\010EP1S0E"
+
+/* Rx FIFO Control register: offset 0x1d */
+#define	RXFC_EXT_SRAM	0x02	/* enable external 32k sram */
+#define	RXFC_RX32PKT	0x01	/* max 32 packet */
+
+/* EEPROM offset register: offset 0x20 */
+#define	EEOFFSET_MASK	0x3f	/* eeprom offset address in word */
+
+/* EEPROM access control register: offset 0x23 */
+#define	EECTRL_DONE	0x04
+#define	EECTRL_RD	0x02
+#define	EECTRL_WR	0x01
+
+#define	EECTRL_BITS	\
+	"\020"	\
+	"\003DONE"	\
+	"\002RD"	\
+	"\001WR"
+
+/* PHY control register: offset 28 */
+#define	PHYAC_DO	0x80U	/* Done */
+#define	PHYAC_RDPHY	0x40U	/* read phy */
+#define	PHYAC_WRPHY	0x20U	/* write phy */
+#define	PHYAC_PHYRA	0x1fU	/* PHY register address */
+
+#define	PHYCTRL_BITS	\
+	"\020"	\
+	"\010DO"	\
+	"\007RDPHY"	\
+	"\006WRPHY"
+
+/* Internal PHY control register: offset 7b */
+#define	IPHYC_EPHY	0x02
+#define	IPHYC_PHYR	0x01
+
+#define	IPHYC_BITS	\
+	"\020"	\
+	"\002EPHY"	\
+	"\001PHYR"
+
+/* GPIO45 register: offset 7c */
+#define	GPIO54_5OE	0x20
+#define	GPIO54_5O	0x10
+#define	GPIO54_5I	0x08
+#define	GPIO54_4OE	0x04
+#define	GPIO54_4O	0x02
+#define	GPIO54_4I	0x01
+
+/* GPIO01 register: offset 7e */
+#define	GPIO10_1OE	0x20
+#define	GPIO10_1O	0x10
+#define	GPIO10_1I	0x08
+#define	GPIO10_0OE	0x04
+#define	GPIO10_0O	0x02
+#define	GPIO10_0I	0x01
+
+/* GPIO23 register: offset 7f */
+#define	GPIO32_3OE	0x20
+#define	GPIO32_3O	0x10
+#define	GPIO32_3I	0x08
+#define	GPIO32_2OE	0x04
+#define	GPIO32_2O	0x02
+#define	GPIO32_2I	0x01
+
+/* rx status at the end of received packets */
+/* byte 0 and 1 is packet length in little endian */
+/* byte 2 is receive status */
+#define	RSR_DRIBBLE	0x10
+#define	RSR_CRC		0x08
+#define	RSR_RUNT	0x04
+#define	RSR_LONG	0x02
+#define	RSR_MULTI	0x01
+
+#define	RSR_ERRORS	\
+	(RSR_DRIBBLE | RSR_CRC | RSR_RUNT | RSR_LONG | RSR_MULTI)
+
+#define	RSR_BITS	\
+	"\020"	\
+	"\005DRIBBLE"	\
+	"\004CRC"	\
+	"\003RUNT"	\
+	"\002LONG"	\
+	"\001MULTI"
+/* byte 3 is reserved */
+
+/* TEST register: offset 80 */
diff --git a/usr/src/uts/common/io/upf/upf_usbgem.c b/usr/src/uts/common/io/upf/upf_usbgem.c
new file mode 100644
index 0000000000..5614803158
--- /dev/null
+++ b/usr/src/uts/common/io/upf/upf_usbgem.c
@@ -0,0 +1,1213 @@
+/*
+ * upf_usbgem.c : ADMtek an986/adm8511/adm8513/adm8515 USB to
+ * Fast Ethernet Driver for Solaris
+ */
+
+/*
+ * Copyright (c) 2004-2011 Masayuki Murayama.  All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions are met:
+ *
+ * 1. Redistributions of source code must retain the above copyright notice,
+ *    this list of conditions and the following disclaimer.
+ *
+ * 2. Redistributions in binary form must reproduce the above copyright notice,
+ *    this list of conditions and the following disclaimer in the documentation
+ *    and/or other materials provided with the distribution.
+ *
+ * 3. Neither the name of the author nor the names of its contributors may be
+ *    used to endorse or promote products derived from this software without
+ *    specific prior written permission.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
+ * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
+ * COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
+ * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
+ * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS
+ * OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
+ * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+ * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
+ * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH
+ * DAMAGE.
+ */
+
+#pragma ident   "%W% %E%"
+
+/*
+ *  Changelog:
+ */
+
+/*
+ * TODO
+ */
+/* ======================================================= */
+
+/*
+ * Solaris system header files and macros
+ */
+#include <sys/types.h>
+#include <sys/conf.h>
+#include <sys/debug.h>
+#include <sys/kmem.h>
+#include <sys/modctl.h>
+#include <sys/errno.h>
+#include <sys/ddi.h>
+#include <sys/sunddi.h>
+#include <sys/byteorder.h>
+
+/* ethernet stuff */
+#include <sys/ethernet.h>
+
+/* interface card depend stuff */
+#include <sys/stropts.h>
+#include <sys/stream.h>
+#include <sys/strlog.h>
+#include <sys/usb/usba.h>
+#include "usbgem.h"
+
+/* hardware stuff */
+#include "usbgem_mii.h"
+#include "adm8511reg.h"
+
+char	ident[] = "pegasus usbnic driver v" VERSION;
+
+/*
+ * Useful macros
+ */
+#define	CHECK_AND_JUMP(val, label)	\
+	if ((val) != USB_SUCCESS) { goto label; }
+
+/*
+ * Debugging
+ */
+#ifdef DEBUG_LEVEL
+static int upf_debug = DEBUG_LEVEL;
+#define	DPRINTF(n, args)	if (upf_debug > (n)) cmn_err args
+#else
+#define	DPRINTF(n, args)
+#endif
+
+/*
+ * Our configration for ADMtek Pegasus/PegasusII
+ */
+/* timeouts */
+#define	ONESEC		(drv_usectohz(1*1000000))
+
+/*
+ * Local device definitions
+ */
+struct upf_dev {
+	/*
+	 * Misc HW information
+	 */
+	uint8_t		ec[3];
+	uint8_t		mac_addr[ETHERADDRL];
+	int		chip_type;
+#define	CHIP_AN986	1	/* avoid 0 */
+#define	CHIP_ADM8511	2	/* including adm8515 */
+#define	CHIP_ADM8513	3
+	boolean_t	phy_init_done;
+	uint8_t		last_link_state;
+
+	uint16_t	vid;	/* vendor id */
+	uint16_t	pid;	/* product id */
+};
+
+/*
+ * private functions
+ */
+
+/* mii operations */
+static uint16_t upf_mii_read(struct usbgem_dev *, uint_t, int *errp);
+static void upf_mii_write(struct usbgem_dev *, uint_t, uint16_t, int *errp);
+
+/* nic operations */
+static int upf_attach_chip(struct usbgem_dev *);
+static int upf_reset_chip(struct usbgem_dev *);
+static int upf_init_chip(struct usbgem_dev *);
+static int upf_start_chip(struct usbgem_dev *);
+static int upf_stop_chip(struct usbgem_dev *);
+static int upf_set_media(struct usbgem_dev *);
+static int upf_set_rx_filter(struct usbgem_dev *);
+static int upf_get_stats(struct usbgem_dev *);
+
+/* packet operations */
+static mblk_t *upf_tx_make_packet(struct usbgem_dev *, mblk_t *);
+static mblk_t *upf_rx_make_packet(struct usbgem_dev *, mblk_t *);
+
+/* interrupt handler */
+static void upf_interrupt(struct usbgem_dev *, mblk_t *);
+
+/* =============================================================== */
+/*
+ * I/O functions
+ */
+/* =============================================================== */
+#define	UPF_REQ_GET_REGISTER	0xf0
+#define	UPF_REQ_SET_REGISTER	0xf1
+#define	OUTB(dp, p, v, errp, label)	\
+	if ((*(errp) = usbgem_ctrl_out((dp), 	\
+	/* bmRequestType */ USB_DEV_REQ_HOST_TO_DEV	\
+		| USB_DEV_REQ_TYPE_VENDOR | USB_DEV_REQ_RCPT_DEV,	\
+	/* bRequest */	UPF_REQ_SET_REGISTER,	\
+	/* wValue */	(v),	\
+	/* wIndex */	(p),	\
+	/* wLength */	1,	\
+	/* buf */	NULL,	\
+	/* size */	0)) != USB_SUCCESS) goto label;
+
+#define	OUTW(dp, p, v, errp, label)	\
+	if ((*(errp) = usbgem_ctrl_out_val((dp), 	\
+	/* bmRequestType */ USB_DEV_REQ_HOST_TO_DEV	\
+		| USB_DEV_REQ_TYPE_VENDOR | USB_DEV_REQ_RCPT_DEV,	\
+	/* bRequest */	UPF_REQ_SET_REGISTER,	\
+	/* wValue */	0,	\
+	/* wIndex */	(p),	\
+	/* wLength */	2,	\
+	/* value */	(v))) != USB_SUCCESS) goto label
+
+#define	OUTS(dp, p, buf, len, errp, label)	\
+	if ((*(errp) = usbgem_ctrl_out((dp), 	\
+	/* bmRequestType */ USB_DEV_REQ_HOST_TO_DEV	\
+		| USB_DEV_REQ_TYPE_VENDOR | USB_DEV_REQ_RCPT_DEV,	\
+	/* bRequest */	UPF_REQ_SET_REGISTER,	\
+	/* wValue */	0,	\
+	/* wIndex */	(p),	\
+	/* wLength */	(len),	\
+	/* buf */	(buf),	\
+	/* size */	(len))) != USB_SUCCESS) goto label
+
+#define	INB(dp, p, vp, errp, label)	\
+	if ((*(errp) = usbgem_ctrl_in_val((dp), 	\
+	/* bmRequestType */ USB_DEV_REQ_DEV_TO_HOST	\
+		| USB_DEV_REQ_TYPE_VENDOR | USB_DEV_REQ_RCPT_DEV,	\
+	/* bRequest */ UPF_REQ_GET_REGISTER,	\
+	/* wValue */	0,	\
+	/* wIndex */	(p),	\
+	/* wLength */	1,	\
+	/* valuep */	(vp))) != USB_SUCCESS) goto label
+
+#define	INW(dp, p, vp, errp, label)	\
+	if ((*(errp) = usbgem_ctrl_in_val((dp), 	\
+	/* bmRequestType */ USB_DEV_REQ_DEV_TO_HOST	\
+		| USB_DEV_REQ_TYPE_VENDOR | USB_DEV_REQ_RCPT_DEV,	\
+	/* bRequest */ UPF_REQ_GET_REGISTER,	\
+	/* wValue */	0,	\
+	/* wIndex */	(p),	\
+	/* wLength */	2,	\
+	/* valuep */	(vp))) != USB_SUCCESS) goto label
+
+#define	INS(dp, p, buf, len, errp, label)	\
+	if ((*(errp) = usbgem_ctrl_in((dp), 	\
+	/* bmRequestType */ USB_DEV_REQ_DEV_TO_HOST	\
+		    | USB_DEV_REQ_TYPE_VENDOR | USB_DEV_REQ_RCPT_DEV,	\
+	/* bRequest */ UPF_REQ_GET_REGISTER,	\
+	/* wValue */	0,	\
+	/* wIndex */	(p),	\
+	/* wLength */	(len),	\
+	/* buf */	(buf),	\
+	/* size */	(len))) != USB_SUCCESS) goto label
+
+/* =============================================================== */
+/*
+ * Hardware manupilation
+ */
+/* =============================================================== */
+static int
+upf_reset_chip(struct usbgem_dev *dp)
+{
+	int		i;
+	uint8_t		val;
+	int		err;
+	struct upf_dev	*lp = dp->private;
+
+	DPRINTF(0, (CE_CONT, "!%s: %s: called", dp->name, __func__));
+	bzero(lp->mac_addr, sizeof (lp->mac_addr));
+
+	lp->ec[1] = 0;
+	OUTB(dp, EC1, EC1_RM, &err, usberr);
+
+	for (i = 0; i < 1000; i++) {
+		INB(dp, EC1, &val, &err, usberr);
+		if ((val & EC1_RM) == 0) {
+			lp->ec[1] = val;
+			return (USB_SUCCESS);
+		}
+		drv_usecwait(10);
+	}
+
+	/* time out */
+	cmn_err(CE_WARN, "!%s: failed to reset: timeout", dp->name);
+	return (USB_FAILURE);
+
+usberr:
+	cmn_err(CE_NOTE, "!%s: %s: usberr detected", dp->name, __func__);
+	return (USB_FAILURE);
+}
+
+/*
+ * Setup an986/adm8511/adm8513/adm8515
+ */
+static int
+upf_init_chip(struct usbgem_dev *dp)
+{
+	uint64_t	zero64 = 0;
+	int	err = USB_SUCCESS;
+	struct upf_dev	*lp = dp->private;
+
+	DPRINTF(0, (CE_CONT, "!%s: %s: called", dp->name, __func__));
+
+	/* ethernet control register 0 */
+	lp->ec[0] |= EC0_RXSA | EC0_RXCS;
+	OUTB(dp, EC0, lp->ec[0], &err, usberr);
+
+	/* ethernet control reg1: will be set later in set_rx_filter() */
+
+	/* ethernet control register 2: will be set later in set_rx_filter() */
+	INB(dp, EC2, &lp->ec[2], &err, usberr);
+	lp->ec[2] |= EC2_RXBP | EC2_EP3RC;
+#ifdef CONFIG_VLAN
+	if (dp->misc_flag & USBGEM_VLAN) {
+		lp->ec[2] |= EC2_MEPL;
+	}
+#endif
+	OUTB(dp, EC2, lp->ec[2], &err, usberr);
+
+	/* Multicast address hash: clear */
+	OUTS(dp, MA, &zero64, 8, &err, usberr);
+
+	/* Ethernet ID : will be set later in upf_set_rx_filter() */
+
+	/* PAUSE timer */
+	OUTB(dp, PAUSETIMER, 0x1f, &err, usberr);
+
+	/* receive packet number based pause control:set in upf_set_media() */
+
+	/* occupied receive FIFO based pause control:set in upf_set_media() */
+
+	/* EP1 control: default */
+
+	/* Rx FIFO control */
+	if (lp->chip_type != CHIP_AN986) {
+		/* use 24K internal sram, 16pkts in fifo */
+		OUTB(dp, RXFC, 0, &err, usberr);
+	}
+
+	/* BIST contror: do nothing */
+	err = upf_set_media(dp);
+	CHECK_AND_JUMP(err, usberr);
+
+	DPRINTF(2, (CE_CONT, "!%s: %s: end (success)", dp->name, __func__));
+	return (USB_SUCCESS);
+
+usberr:
+	cmn_err(CE_NOTE, "!%s: %s: usberr(%d) detected",
+	    dp->name, __func__, err);
+	return (err);
+}
+
+static int
+upf_start_chip(struct usbgem_dev *dp)
+{
+	int	err = USB_SUCCESS;
+	struct upf_dev	*lp = dp->private;
+
+	DPRINTF(0, (CE_CONT, "!%s: %s: called", dp->name, __func__));
+
+	/* enable RX and TX */
+	lp->ec[0] |= EC0_TXE | EC0_RXE;
+	OUTB(dp, EC0, lp->ec[0], &err, usberr);
+	return (USB_SUCCESS);
+
+usberr:
+	cmn_err(CE_WARN, "!%s: %s: usberr(%d) detected",
+	    dp->name, __func__, err);
+	return (err);
+}
+
+static int
+upf_stop_chip(struct usbgem_dev *dp)
+{
+	int	err;
+	struct upf_dev	*lp = dp->private;
+
+	DPRINTF(0, (CE_CONT, "!%s: %s: called", dp->name, __func__));
+
+	/* disable RX and TX */
+	lp->ec[0] &= ~(EC0_TXE | EC0_RXE);
+	OUTB(dp, EC0, lp->ec[0], &err, usberr);
+
+	return (USB_SUCCESS);
+
+usberr:
+	cmn_err(CE_WARN, "!%s: %s: usberr(%d) detected",
+	    dp->name, __func__, err);
+	return (err);
+}
+
+static int
+upf_get_stats(struct usbgem_dev *dp)
+{
+	/* do nothing */
+	return (USB_SUCCESS);
+}
+
+static uint_t
+upf_mcast_hash(struct usbgem_dev *dp, const uint8_t *addr)
+{
+	/* hash table is 64 = 2^6 bit width */
+	return (usbgem_ether_crc_le(addr) & 0x3f);
+}
+
+static int
+upf_set_rx_filter(struct usbgem_dev *dp)
+{
+	int		i;
+	int		err;
+#ifdef DEBUG_LEVEL
+	uint8_t		reg0;
+	uint8_t		reg1;
+	uint8_t		reg2;
+#endif
+	struct upf_dev	*lp = dp->private;
+
+	DPRINTF(0, (CE_CONT, "!%s: %s: called, rxmode:%b",
+	    dp->name, __func__, dp->rxmode, RXMODE_BITS));
+
+	/* reset rx mode */
+	lp->ec[0] &= ~EC0_RXMA;
+	lp->ec[2] &= ~EC2_PROM;
+
+	if (dp->rxmode & RXMODE_PROMISC) {
+		/* promiscious mode implies all multicast and all physical */
+		lp->ec[0] |= EC0_RXMA;
+		lp->ec[2] |= EC2_PROM;
+	} else if ((dp->rxmode & RXMODE_ALLMULTI) || dp->mc_count > 0) {
+		/* XXX - multicast hash table didin't work */
+		/* accept all multicast packets */
+		lp->ec[0] |= EC0_RXMA;
+	}
+
+	if (bcmp(dp->cur_addr.ether_addr_octet,
+	    lp->mac_addr, ETHERADDRL) != 0) {
+
+		/* need to update mac address */
+		bcopy(dp->cur_addr.ether_addr_octet,
+		    lp->mac_addr, ETHERADDRL);
+		OUTS(dp, EID,
+		    lp->mac_addr, ETHERADDRL, &err, usberr);
+	}
+
+	/* update rx mode */
+	OUTS(dp, EC0, lp->ec, 3, &err, usberr);
+
+#if DEBUG_LEVEL > 0
+	INB(dp, EC0, &reg0, &err, usberr);
+	INB(dp, EC1, &reg1, &err, usberr);
+	INB(dp, EC2, &reg2, &err, usberr);
+
+	cmn_err(CE_CONT, "!%s: %s: returned, ec:%b %b %b",
+	    dp->name, __func__,
+	    reg0, EC0_BITS, reg1, EC1_BITS, reg2, EC2_BITS);
+#endif
+	return (USB_SUCCESS);
+
+usberr:
+	cmn_err(CE_NOTE, "!%s: %s: usberr detected", dp->name, __func__);
+	return (err);
+}
+
+static int
+upf_set_media(struct usbgem_dev *dp)
+{
+	int	err;
+	struct upf_dev	*lp = dp->private;
+
+	DPRINTF(0, (CE_CONT, "!%s: %s: called", dp->name, __func__));
+
+	lp->ec[1] &= ~(EC1_FD | EC1_100M);
+
+	/* select duplex */
+	if (dp->full_duplex) {
+		lp->ec[1] |= EC1_FD;
+	}
+
+	/* select speed */
+	if (dp->speed == USBGEM_SPD_100) {
+		lp->ec[1] |= EC1_100M;
+	}
+
+	/* rx flow control */
+	switch (dp->flow_control) {
+	case FLOW_CONTROL_SYMMETRIC:
+	case FLOW_CONTROL_RX_PAUSE:
+		lp->ec[0] |= EC0_RXFCE;
+		break;
+
+	default:
+		lp->ec[0] &= ~EC0_RXFCE;
+		break;
+	}
+
+	/* tx flow control */
+	switch (dp->flow_control) {
+	case FLOW_CONTROL_SYMMETRIC:
+	case FLOW_CONTROL_TX_PAUSE:
+		if (lp->chip_type != CHIP_AN986) {
+			/* pegasus II has internal 24k fifo */
+			OUTB(dp, ORFBFC,
+			    (12 << ORFBFC_RXS_SHIFT) |  ORFBFC_FCRXS,
+			    &err, usberr);
+
+			/* 16 packts can be stored in rx fifo */
+			OUTB(dp, RPNBFC_PN,
+			    (8 << RPNBFC_PN_SHIFT) |  RPNBFC_FCP,
+			    &err, usberr);
+		} else {
+			/* an986 has external 32k fifo */
+			OUTB(dp, ORFBFC,
+			    (16 << ORFBFC_RXS_SHIFT) |  ORFBFC_FCRXS,
+			    &err, usberr);
+
+			/* AN986 fails to link up when RPNBFC is enabled */
+			OUTB(dp, RPNBFC, 0, &err, usberr);
+		}
+		break;
+
+	default:
+		OUTB(dp, ORFBFC, 0, &err, usberr);
+		OUTB(dp, RPNBFC, 0, &err, usberr);
+		break;
+	}
+
+	/* update ether control registers */
+	OUTS(dp, EC0, lp->ec, 2, &err, usberr);
+	DPRINTF(0, (CE_CONT, "!%s: %s: returned, ec0:%b, ec1:%b",
+	    dp->name, __func__, lp->ec[0], EC0_BITS, lp->ec[1], EC1_BITS));
+
+	return (USB_SUCCESS);
+
+usberr:
+	cmn_err(CE_WARN, "%s: %s: failed to write ec1", dp->name, __func__);
+	return (err);
+}
+
+/*
+ * send/receive packet check
+ */
+static mblk_t *
+upf_tx_make_packet(struct usbgem_dev *dp, mblk_t *mp)
+{
+	size_t		len;
+	mblk_t		*new;
+	mblk_t		*tp;
+	uint8_t		*bp;
+	uint8_t		*last_pos;
+	int		msglen;
+
+	DPRINTF(3, (CE_CONT, "!%s: %s: called", dp->name, __func__));
+
+	len = msgdsize(mp);
+	if (len < ETHERMIN) {
+		len = ETHERMIN;
+	}
+
+	/* allocate msg block */
+	msglen = len + sizeof (uint16_t);
+
+	/* avoid usb controller bug */
+	if ((msglen & 0x3f) == 0) {
+		/* add a header for additional 0-length usb message */
+		msglen += sizeof (uint16_t);
+	}
+
+	if ((new = allocb(msglen, 0)) == NULL) {
+		return (NULL);
+	}
+
+	/* copy contents of the buffer */
+	new->b_wptr = new->b_rptr + msglen;
+	bp = new->b_rptr;
+
+	/* the nic requires a two byte header of the packet size */
+	bp[0] = (uint8_t)len;
+	bp[1] = (uint8_t)(len >> 8);
+	bp += sizeof (uint16_t);
+
+	/* copy the payload */
+	for (tp = mp; tp; tp = tp->b_cont) {
+		len = tp->b_wptr - tp->b_rptr;
+		if (len > 0) {
+			bcopy(tp->b_rptr, bp, len);
+			bp += len;
+		}
+	}
+
+	/* clear ethernet pads and additional usb header if we have */
+	last_pos = new->b_wptr;
+	while (bp < last_pos) {
+		*bp++ = 0;
+	}
+
+	return (new);
+}
+
+static void
+upf_dump_packet(struct usbgem_dev *dp, uint8_t *bp, int n)
+{
+	int	i;
+
+	for (i = 0; i < n; i += 8, bp += 8) {
+		cmn_err(CE_CONT, "%02x %02x %02x %02x %02x %02x %02x %02x",
+		    bp[0], bp[1], bp[2], bp[3], bp[4], bp[5], bp[6], bp[7]);
+	}
+}
+
+static mblk_t *
+upf_rx_make_packet(struct usbgem_dev *dp, mblk_t *mp)
+{
+	uint8_t		*p;
+	uint16_t	rxhd;
+	uint_t		len;
+	uint8_t		rsr;
+	struct upf_dev	*lp = dp->private;
+
+	ASSERT(mp != NULL);
+
+#ifdef DEBUG_LEVEL
+	len = msgdsize(mp);
+	DPRINTF(2, (CE_CONT, "!%s: time:%d %s: cont:%p",
+	    dp->name, ddi_get_lbolt(), __func__, len, mp->b_cont));
+
+	if (upf_debug > 3) {
+		upf_dump_packet(dp, mp->b_rptr, max(6, len));
+	}
+#endif
+	/* get the length of Rx packet */
+	p = mp->b_wptr - 4;
+	rsr = p[3];
+	if (lp->chip_type == CHIP_ADM8513) {
+		/* As Rx packets from ADM8513 have two byte header, remove it */
+		p = mp->b_rptr;
+		len = ((p[1] << 8) | p[0]) & 0x0fff;
+		mp->b_rptr += 2;
+	} else {
+		len = (((p[1] << 8) | p[0]) & 0x0fff) - ETHERFCSL - 4;
+	}
+
+	DPRINTF(2, (CE_CONT, "!%s: %s: rsr:%b len:%d",
+	    dp->name, __func__, rsr, RSR_BITS, len));
+
+	/* check if error happen */
+	if (rsr & RSR_ERRORS) {
+		DPRINTF(0, (CE_CONT, "!%s: rsr:%b", dp->name, rsr, RSR_BITS));
+		if (rsr & (RSR_CRC | RSR_DRIBBLE)) {
+			dp->stats.frame++;
+		}
+		if (rsr & RSR_LONG) {
+			dp->stats.frame_too_long++;
+		}
+		if (rsr & RSR_RUNT) {
+			dp->stats.runt++;
+		}
+
+		dp->stats.errrcv++;
+		return (NULL);
+	}
+#ifndef CONFIG_VLAN
+	/* check packet size */
+	if (len > ETHERMAX) {
+		/* too long */
+		dp->stats.frame_too_long++;
+		dp->stats.errrcv++;
+		return (NULL);
+	} else if (len < ETHERMIN) {
+		dp->stats.runt++;
+		dp->stats.errrcv++;
+		return (NULL);
+	}
+#endif
+	/* remove tailing crc and rx status fields */
+	mp->b_wptr = mp->b_rptr + len;
+	ASSERT(mp->b_next == NULL);
+	return (mp);
+}
+
+/*
+ * Device depend interrupt handler
+ */
+static void
+upf_interrupt(struct usbgem_dev *dp, mblk_t *mp)
+{
+	uint8_t	*bp;
+	struct upf_dev	*lp = dp->private;
+
+	bp = mp->b_rptr;
+
+	DPRINTF(2, (CE_CONT,
+	    "!%s: %s: size:%d, %02x %02x %02x %02x %02x %02x %02x %02x",
+	    dp->name, __func__, mp->b_wptr - mp->b_rptr,
+	    bp[0], bp[1], bp[2], bp[3], bp[4], bp[5], bp[6], bp[7]));
+
+	if ((lp->last_link_state ^ bp[5]) & 1) {
+		DPRINTF(1, (CE_CONT, "!%s:%s link status changed:",
+		    dp->name, __func__));
+		usbgem_mii_update_link(dp);
+	}
+
+	lp->last_link_state = bp[5] & 1;
+}
+
+/*
+ * MII Interfaces
+ */
+static uint16_t
+upf_mii_read(struct usbgem_dev *dp, uint_t index, int *errp)
+{
+	uint8_t		phyctrl;
+	uint16_t	val;
+	int		i;
+
+	DPRINTF(4, (CE_CONT, "!%s: %s: called, ix:%d",
+	    dp->name, __func__, index));
+	ASSERT(index >= 0 && index < 32);
+
+	*errp = USB_SUCCESS;
+
+	/* set PHYADDR */
+	OUTB(dp, PHYA, dp->mii_phy_addr, errp, usberr);
+
+	/* Initiate MII read transaction */
+	OUTB(dp, PHYAC, index | PHYAC_RDPHY, errp, usberr);
+
+	for (i = 0; i < 100; i++) {
+		INB(dp, PHYAC, &phyctrl, errp, usberr);
+		if (phyctrl & PHYAC_DO) {
+			/* done */
+			INW(dp, PHYD, &val, errp, usberr);
+			DPRINTF(4, (CE_CONT, "!%s: %s: return %04x",
+			    dp->name, __func__, val));
+			return (val);
+		}
+		drv_usecwait(10);
+	}
+	/* timeout */
+	cmn_err(CE_WARN, "!%s: %s: timeout detected", dp->name, __func__);
+	*errp = USB_FAILURE;
+	return (0);
+
+usberr:
+	cmn_err(CE_CONT,
+	    "!%s: %s: usberr(%d) detected", dp->name, __func__, *errp);
+	return (0);
+}
+
+static void
+upf_mii_write(struct usbgem_dev *dp, uint_t index, uint16_t val, int *errp)
+{
+	int		i;
+	uint8_t		phyctrl;
+
+	DPRINTF(4, (CE_CONT, "!%s: %s called index:%d val:0x%04x",
+	    dp->name, __func__, index, val));
+	ASSERT(index >= 0 && index < 32);
+
+	*errp = USB_SUCCESS;
+
+	OUTW(dp, PHYD, val, errp, usberr);
+	OUTB(dp, PHYA, dp->mii_phy_addr, errp, usberr);
+	OUTB(dp, PHYAC, index | PHYAC_WRPHY, errp, usberr);
+
+	for (i = 0; i < 100; i++) {
+		INB(dp, PHYAC, &phyctrl, errp, usberr);
+		if (phyctrl & PHYAC_DO) {
+			/* done */
+			return;
+		}
+		drv_usecwait(10);
+	}
+
+	/* time out */
+	cmn_err(CE_WARN, "!%s: %s: timeout detected", dp->name, __func__);
+	*errp = USB_FAILURE;
+	return;
+
+usberr:
+	cmn_err(CE_CONT,
+	    "!%s: %s: usberr(%d) detected", dp->name, __func__, *errp);
+}
+
+
+static int
+upf_enable_phy(struct usbgem_dev *dp)
+{
+	uint8_t	val;
+	int	err;
+	struct upf_dev	*lp = dp->private;
+
+	/*
+	 * first, try to enable internal phy
+	 */
+	INB(dp, IPHYC, &val, &err, usberr);
+	val = (val | IPHYC_EPHY) & ~IPHYC_PHYR;
+	OUTB(dp, IPHYC, val, &err, usberr);
+
+	INB(dp, IPHYC, &val, &err, usberr);
+	DPRINTF(0, (CE_CONT, "!%s: %s: IPHYC: %b",
+	    dp->name, __func__, val, IPHYC_BITS));
+	if (val) {
+		/* reset internal phy */
+		OUTB(dp, IPHYC, val | IPHYC_PHYR, &err, usberr);
+		OUTB(dp, IPHYC, val, &err, usberr);
+		delay(drv_usectohz(10000));
+
+		/* identify the chip generation */
+		OUTB(dp, 0x83, 0xa5, &err, usberr);
+		INB(dp, 0x83, &val, &err, usberr);
+		if (val == 0xa5) {
+			lp->chip_type = CHIP_ADM8513;
+		} else {
+			/* adm8511 or adm8515 */
+			lp->chip_type = CHIP_ADM8511;
+		}
+		dp->ugc.usbgc_mii_hw_link_detection = B_TRUE;
+	} else {
+		/*
+		 * It should be AN986 which doesn't have an internal PHY.
+		 * We need to setup gpio ports in AN986, which are
+		 * connected to external PHY control pins.
+		 */
+		lp->chip_type = CHIP_AN986;
+
+		/* reset external phy */
+		/* output port#0 L, port#1 L */
+		OUTB(dp, GPIO10, GPIO10_0O | GPIO10_0OE, &err, usberr);
+
+		/* output port#0 H, port#1 L */
+		OUTB(dp, GPIO10,
+		    GPIO10_0O | GPIO10_0OE | GPIO10_1OE, &err, usberr);
+
+		/* hw link detection doesn't work correctly */
+		dp->ugc.usbgc_mii_hw_link_detection = B_FALSE;
+	}
+
+	return (USB_SUCCESS);
+
+usberr:
+	cmn_err(CE_NOTE, "!%s: %s: usberr detected", dp->name, __func__);
+	return (USB_FAILURE);
+}
+
+static int
+upf_mii_probe(struct usbgem_dev *dp)
+{
+	int	err;
+	uint16_t	val;
+	struct upf_dev	*lp = dp->private;
+
+	if (!lp->phy_init_done) {
+		upf_enable_phy(dp);
+		lp->phy_init_done = B_TRUE;
+	}
+
+	return (usbgem_mii_probe_default(dp));
+}
+
+static int
+upf_mii_init(struct usbgem_dev *dp)
+{
+	uint16_t	val;
+	int		err = USB_SUCCESS;
+	struct upf_dev	*lp = dp->private;
+
+	if (!lp->phy_init_done) {
+		upf_enable_phy(dp);
+	}
+	lp->phy_init_done = B_FALSE;
+
+	if (lp->chip_type == CHIP_AN986 &&
+	    (lp->vid == 0x0db7 /* elecom */ ||
+	    lp->vid == 0x066b /* linksys */ ||
+	    lp->vid == 0x077b /* linksys */ ||
+	    lp->vid == 0x2001 /* dlink */)) {
+		/* special treatment for Linksys products */
+		val = upf_mii_read(dp, 0x1b, &err) | 0x4;
+		upf_mii_write(dp, 0x1b, val, &err);
+	}
+	return (err);
+}
+
+/* ======================================================== */
+/*
+ * OS depend (device driver DKI) routine
+ */
+/* ======================================================== */
+static uint16_t
+upf_read_eeprom(struct usbgem_dev *dp, int index, int *errp)
+{
+	int		i;
+	uint8_t		eectrl;
+	uint16_t	data;
+
+	*errp = USB_SUCCESS;
+
+	OUTB(dp, EECTRL, 0, errp, usberr);
+
+	OUTB(dp, EEOFFSET, index, errp, usberr);
+	OUTB(dp, EECTRL, EECTRL_RD, errp, usberr);
+
+	for (i = 0; i < 100; i++) {
+		INB(dp, EECTRL, &eectrl, errp, usberr);
+		if (eectrl & EECTRL_DONE) {
+			INW(dp, EEDATA, &data, errp, usberr);
+			return (data);
+		}
+		drv_usecwait(10);
+	}
+
+	/* time out */
+	*errp = USB_FAILURE;
+	return (0);
+
+usberr:
+	cmn_err(CE_CONT,
+	    "!%s: %s: usberr(%d) detected", dp->name, __func__, *errp);
+	return (0);
+}
+
+static void
+upf_eeprom_dump(struct usbgem_dev *dp, int size)
+{
+	int	i;
+	int	err;
+
+	cmn_err(CE_CONT, "!%s: %s dump:", dp->name, __func__);
+
+	for (i = 0; i < size; i += 4) {
+		cmn_err(CE_CONT, "!0x%02x: 0x%04x 0x%04x 0x%04x 0x%04x",
+		    i*2,
+		    upf_read_eeprom(dp, i + 0, &err),
+		    upf_read_eeprom(dp, i + 1, &err),
+		    upf_read_eeprom(dp, i + 2, &err),
+		    upf_read_eeprom(dp, i + 3, &err));
+	}
+}
+
+static int
+upf_attach_chip(struct usbgem_dev *dp)
+{
+	int		i;
+	int		err;
+	uint16_t	val;
+	uint8_t		*mac;
+	struct upf_dev	*lp = dp->private;
+
+	/*
+	 * Read mac address from EEPROM
+	 */
+	mac = dp->dev_addr.ether_addr_octet;
+	for (i = 0; i < 3; i++) {
+		val = upf_read_eeprom(dp, i, &err);
+		if (err != USB_SUCCESS) {
+			goto usberr;
+		}
+		mac[i*2+0] = (uint8_t)val;
+		mac[i*2+1] = (uint8_t)(val >> 8);
+	}
+
+	DPRINTF(0, (CE_CONT,
+	    "%s: %s: mac: %02x:%02x:%02x:%02x:%02x:%02x",
+	    dp->name, __func__,
+	    mac[0], mac[1], mac[2], mac[3], mac[4], mac[5]));
+
+	dp->misc_flag = 0;
+#ifdef CONFIG_VLAN
+	dp->misc_flag |= USBGEM_VLAN;
+#endif
+#if DEBUG_LEVEL > 3
+	upf_eeprom_dump(dp, 0x80);
+#endif
+	return (USB_SUCCESS);
+
+usberr:
+	cmn_err(CE_WARN, "!%s: %s: usb error detected", dp->name, __func__);
+	return (USB_FAILURE);
+}
+
+static int
+upfattach(dev_info_t *dip, ddi_attach_cmd_t cmd)
+{
+	int			i;
+	ddi_iblock_cookie_t	c;
+	int			ret;
+	int			unit;
+	uint32_t		tcr;
+	int			len;
+	const char		*drv_name;
+	struct usbgem_dev	*dp;
+	void			*base;
+	struct usbgem_conf	*ugcp;
+	struct upf_dev		*lp;
+
+	unit = ddi_get_instance(dip);
+	drv_name = ddi_driver_name(dip);
+
+	DPRINTF(3, (CE_CONT, "!%s%d: %s: called, cmd:%d",
+	    drv_name, unit, __func__, cmd));
+
+	if (cmd == DDI_ATTACH) {
+		/*
+		 * construct usbgem configration
+		 */
+		ugcp = kmem_zalloc(sizeof (*ugcp), KM_SLEEP);
+
+		/* name */
+		sprintf(ugcp->usbgc_name, "%s%d", drv_name, unit);
+		ugcp->usbgc_ppa = unit;
+
+		ugcp->usbgc_ifnum = 0;
+		ugcp->usbgc_alt = 0;
+
+		ugcp->usbgc_tx_list_max = 16;
+
+		ugcp->usbgc_rx_header_len = 4;
+		ugcp->usbgc_rx_list_max = 64;
+
+		/* time out parameters */
+		ugcp->usbgc_tx_timeout = USBGEM_TX_TIMEOUT;
+		ugcp->usbgc_tx_timeout_interval = USBGEM_TX_TIMEOUT_INTERVAL;
+
+		/* flow control */
+		ugcp->usbgc_flow_control = FLOW_CONTROL_NONE;
+		ugcp->usbgc_flow_control = FLOW_CONTROL_RX_PAUSE;
+
+		/* MII timeout parameters */
+		ugcp->usbgc_mii_link_watch_interval = ONESEC;
+		ugcp->usbgc_mii_an_watch_interval = ONESEC/5;
+		ugcp->usbgc_mii_reset_timeout = MII_RESET_TIMEOUT; /* 1 sec */
+		ugcp->usbgc_mii_an_timeout = MII_AN_TIMEOUT;	/* 5 sec */
+		ugcp->usbgc_mii_an_wait = MII_AN_TIMEOUT/2;
+		ugcp->usbgc_mii_linkdown_timeout = MII_LINKDOWN_TIMEOUT;
+		ugcp->usbgc_mii_an_delay = ONESEC/10;
+
+		ugcp->usbgc_mii_linkdown_action = MII_ACTION_RESET;
+		ugcp->usbgc_mii_linkdown_timeout_action = MII_ACTION_RESET;
+		ugcp->usbgc_mii_dont_reset = B_FALSE;
+
+		/* I/O methods */
+
+		/* mac operation */
+		ugcp->usbgc_attach_chip = &upf_attach_chip;
+		ugcp->usbgc_reset_chip = &upf_reset_chip;
+		ugcp->usbgc_init_chip = &upf_init_chip;
+		ugcp->usbgc_start_chip = &upf_start_chip;
+		ugcp->usbgc_stop_chip = &upf_stop_chip;
+		ugcp->usbgc_multicast_hash = &upf_mcast_hash;
+
+		ugcp->usbgc_set_rx_filter = &upf_set_rx_filter;
+		ugcp->usbgc_set_media = &upf_set_media;
+		ugcp->usbgc_get_stats = &upf_get_stats;
+		ugcp->usbgc_interrupt = &upf_interrupt;
+
+		/* packet operation */
+		ugcp->usbgc_tx_make_packet = &upf_tx_make_packet;
+		ugcp->usbgc_rx_make_packet = &upf_rx_make_packet;
+
+		/* mii operations */
+		ugcp->usbgc_mii_probe = &upf_mii_probe;
+		ugcp->usbgc_mii_init = &upf_mii_init;
+		ugcp->usbgc_mii_config = &usbgem_mii_config_default;
+		ugcp->usbgc_mii_read = &upf_mii_read;
+		ugcp->usbgc_mii_write = &upf_mii_write;
+
+		/* mtu */
+		ugcp->usbgc_min_mtu = ETHERMTU;
+		ugcp->usbgc_max_mtu = ETHERMTU;
+		ugcp->usbgc_default_mtu = ETHERMTU;
+
+		lp = kmem_zalloc(sizeof (struct upf_dev), KM_SLEEP);
+
+		lp->vid = ddi_prop_get_int(DDI_DEV_T_ANY, dip,
+		    DDI_PROP_DONTPASS, "usb-vendor-id", -1);
+		lp->pid = ddi_prop_get_int(DDI_DEV_T_ANY, dip,
+		    DDI_PROP_DONTPASS, "usb-product-id", -1);
+
+		dp = usbgem_do_attach(dip, ugcp, lp, sizeof (struct upf_dev));
+
+		kmem_free(ugcp, sizeof (*ugcp));
+
+		if (dp != NULL) {
+			return (DDI_SUCCESS);
+		}
+
+err_free_mem:
+		kmem_free(lp, sizeof (struct upf_dev));
+err_close_pipe:
+err:
+		return (DDI_FAILURE);
+	}
+	if (cmd == DDI_RESUME) {
+		dp = USBGEM_GET_DEV(dip);
+		lp = dp->private;
+		lp->phy_init_done = B_FALSE;
+
+		return (usbgem_resume(dip));
+	}
+	return (DDI_FAILURE);
+}
+
+static int
+upfdetach(dev_info_t *dip, ddi_detach_cmd_t cmd)
+{
+	int	ret;
+
+	if (cmd == DDI_DETACH) {
+		ret = usbgem_do_detach(dip);
+		if (ret != DDI_SUCCESS) {
+			return (DDI_FAILURE);
+		}
+		return (DDI_SUCCESS);
+	}
+	if (cmd == DDI_SUSPEND) {
+		return (usbgem_suspend(dip));
+	}
+	return (DDI_FAILURE);
+}
+
+/* ======================================================== */
+/*
+ * OS depend (loadable streams driver) routine
+ */
+/* ======================================================== */
+#ifdef USBGEM_CONFIG_GLDv3
+USBGEM_STREAM_OPS(upf_ops, upfattach, upfdetach);
+#else
+static	struct module_info upfminfo = {
+	0,			/* mi_idnum */
+	"upf",			/* mi_idname */
+	0,			/* mi_minpsz */
+	ETHERMTU,		/* mi_maxpsz */
+	32*1024,		/* mi_hiwat */
+	1,			/* mi_lowat */
+};
+
+static	struct qinit upfrinit = {
+	(int (*)()) NULL,	/* qi_putp */
+	usbgem_rsrv,		/* qi_srvp */
+	usbgem_open,		/* qi_qopen */
+	usbgem_close,		/* qi_qclose */
+	(int (*)()) NULL,	/* qi_qadmin */
+	&upfminfo,		/* qi_minfo */
+	NULL			/* qi_mstat */
+};
+
+static	struct qinit upfwinit = {
+	usbgem_wput,		/* qi_putp */
+	usbgem_wsrv,		/* qi_srvp */
+	(int (*)()) NULL,	/* qi_qopen */
+	(int (*)()) NULL,	/* qi_qclose */
+	(int (*)()) NULL,	/* qi_qadmin */
+	&upfminfo,		/* qi_minfo */
+	NULL			/* qi_mstat */
+};
+
+static struct streamtab	upf_info = {
+	&upfrinit,	/* st_rdinit */
+	&upfwinit,	/* st_wrinit */
+	NULL,		/* st_muxrinit */
+	NULL		/* st_muxwrinit */
+};
+
+static	struct cb_ops cb_upf_ops = {
+	nulldev,	/* cb_open */
+	nulldev,	/* cb_close */
+	nodev,		/* cb_strategy */
+	nodev,		/* cb_print */
+	nodev,		/* cb_dump */
+	nodev,		/* cb_read */
+	nodev,		/* cb_write */
+	nodev,		/* cb_ioctl */
+	nodev,		/* cb_devmap */
+	nodev,		/* cb_mmap */
+	nodev,		/* cb_segmap */
+	nochpoll,	/* cb_chpoll */
+	ddi_prop_op,	/* cb_prop_op */
+	&upf_info,	/* cb_stream */
+	D_MP		/* cb_flag */
+};
+
+static	struct dev_ops upf_ops = {
+	DEVO_REV,	/* devo_rev */
+	0,		/* devo_refcnt */
+	usbgem_getinfo,	/* devo_getinfo */
+	nulldev,	/* devo_identify */
+	nulldev,	/* devo_probe */
+	upfattach,	/* devo_attach */
+	upfdetach,	/* devo_detach */
+	nodev,		/* devo_reset */
+	&cb_upf_ops,	/* devo_cb_ops */
+	NULL,		/* devo_bus_ops */
+	usbgem_power,	/* devo_power */
+#if DEVO_REV >= 4
+	usbgem_quiesce,	/* devo_quiesce */
+#endif
+
+};
+#endif
+static struct modldrv modldrv = {
+	&mod_driverops,	/* Type of module.  This one is a driver */
+	ident,
+	&upf_ops,	/* driver ops */
+};
+
+static struct modlinkage modlinkage = {
+	MODREV_1, &modldrv, NULL
+};
+
+/* ======================================================== */
+/*
+ * _init : done
+ */
+/* ======================================================== */
+int
+_init(void)
+{
+	int 	status;
+
+	DPRINTF(2, (CE_CONT, "!upf: _init: called"));
+
+	status = usbgem_mod_init(&upf_ops, "upf");
+	if (status != DDI_SUCCESS) {
+		return (status);
+	}
+	status = mod_install(&modlinkage);
+	if (status != DDI_SUCCESS) {
+		usbgem_mod_fini(&upf_ops);
+	}
+	return (status);
+}
+
+/*
+ * _fini : done
+ */
+int
+_fini(void)
+{
+	int	status;
+
+	DPRINTF(2, (CE_CONT, "!upf: _fini: called"));
+	status = mod_remove(&modlinkage);
+	if (status == DDI_SUCCESS) {
+		usbgem_mod_fini(&upf_ops);
+	}
+	return (status);
+}
+
+int
+_info(struct modinfo *modinfop)
+{
+	return (mod_info(&modlinkage, modinfop));
+}
diff --git a/usr/src/uts/common/io/urf/rtl8150reg.h b/usr/src/uts/common/io/urf/rtl8150reg.h
new file mode 100644
index 0000000000..7cba53356e
--- /dev/null
+++ b/usr/src/uts/common/io/urf/rtl8150reg.h
@@ -0,0 +1,218 @@
+/*
+ * @(#)rtl8150reg.h	1.1 04/09/16
+ * Macro definitions for Realtek 8150 USB to fast ethernet controller
+ * based on Realtek RTL8150 data sheet
+ * This file is public domain. Coded by M.Murayama (KHF04453@nifty.com)
+ */
+
+/*
+ * Register offset
+ */
+#define	IDR	0x0120	/* Base of ID registers */
+#define	MAR	0x0126	/* Base of multicast registers */
+#define	CR	0x012e	/* Command register */
+#define	TCR	0x012f	/* Transmit Configuration register */
+#define	RCR	0x0130	/* Receive Configuration register */
+#define	TSR	0x0132	/* Transmit Status register */
+#define	RSR	0x0133	/* Receive Status register */
+#define	CON0	0x0135	/* Configuration register 0 */
+#define	CON1	0x0136	/* Configuration register 1 */
+#define	MSR	0x0137	/* Media Status register */
+#define	PHYADD	0x0138	/* PHY address register */
+#define	PHYDAT	0x0139	/* PHY data register */
+#define	PHYCNT	0x013b	/* PHY control register */
+#define	GPPC	0x013d	/* General purpose pin control */
+#define	WAKECNT	0x013e	/* Wake up event control */
+#define	BMCR	0x0140	/* Basic Mode Control register */
+#define	BMSR	0x0142	/* Basic Mode Status register */
+#define	ANAR	0x0144	/* Auto Negotiation Advertisement register */
+#define	ANLP	0x0146	/* Auto Negotiation Link Partner register */
+#define	ANER	0x0148	/* Auto Negotiation Expansion register */
+#define	NWAYT	0x014a	/* Nway test register */
+#define	CSCR	0x014c	/* CS configuration register */
+#define	CRC0	0x014e	/* Power management register for wakeup frame0 */
+#define	CRC1	0x0150	/* Power management register for wakeup frame1 */
+#define	CRC2	0x0152	/* Power management register for wakeup frame2 */
+#define	CRC3	0x0154	/* Power management register for wakeup frame3 */
+#define	CRC4	0x0156	/* Power management register for wakeup frame4 */
+#define	BYTEMASK0 0x0158	/* Power management wakeup frame0 bytemask */
+#define	BYTEMASK1 0x0160	/* Power management wakeup frame1 bytemask */
+#define	BYTEMASK2 0x0168	/* Power management wakeup frame2 bytemask */
+#define	BYTEMASK3 0x0170	/* Power management wakeup frame3 bytemask */
+#define	BYTEMASK4 0x0178	/* Power management wakeup frame4 bytemask */
+#define	PHY1	0x0180	/* PHY parameter 1 */
+#define	PHY2	0x0184	/* PHY parameter 2 */
+#define	TW1	0x0186	/* Twister parameter 1 */
+
+/*
+ * Bit field definitions
+ */
+/* CR : Command register (uint8_t) */
+#define	CR_WEPROM	0x20	/* EEPROM write enable */
+#define	CR_SOFT_RST	0x10	/* Reset */
+#define	CR_RE		0x08	/* Ethernet receive enable */
+#define	CR_TE		0x04	/* Ethernet transmit enable */
+#define	CR_EP3CLREN	0x02	/* clear performance counter after EP3 */
+#define	CR_AUTOLOAD	0x01	/* autoload contents of 93c46 */
+
+#define	CR_BITS	"\020\006WEPROM\005SOFT_RST\004RE\003TE\002EP3CLREN\001AUTOLOAD"
+
+/* TCR: Transmit Configuration register */
+#define	TCR_TXRR	0xc0	/* Tx retry count */
+#define	TCR_TXRR_SHIFT		6
+#define	TCR_IFG		0x18	/* Interframe Gap */
+#define	TCR_IFG_SHIFT		3
+#define	TCR_IFG_802_3		(3 << TCR_IFG_SHIFT)	/* 802.3 standard */
+#define	TCR_NOCRC	0x01	/* Inhibit Appending CRC */
+	
+#define	TCR_BITS	"\020\001NOCRC"
+
+/* Receive Configuration register */
+#define	RCR_TAIL	0x0080	/* Rx header forward to host in CRC field */
+#define	RCR_AER		0x0040	/* Accept Error packet */
+#define	RCR_AR		0x0020	/* Accept runt */
+#define	RCR_AM		0x0010	/* Accept multicast */
+#define	RCR_AB		0x0008	/* Accept broadcast */
+#define	RCR_AD		0x0004	/* Accept physical match */
+#define	RCR_AAM		0x0002	/* Accept all Multicast */
+#define	RCR_AAP		0x0001	/* Accept all physical */
+
+#define	RCR_ACCEPT_MODE		\
+	(RCR_AER | RCR_AR | RCR_AM | RCR_AB | RCR_AD | RCR_AAM | RCR_AAP)
+
+#define	RCR_BITS	\
+	"\020\010TAIL\007AER\006AR\005AM\004AB\003AD\002AAM\001AAP"
+
+/* Transmit Status register */
+
+#define	TSR_ECOL	0x20	/* excessive collision indication */
+#define	TSR_LCOL	0x10	/* late collision indication */
+#define	TSR_LOSS_CRS	0x08	/* lost of carrier indication */
+#define	TSR_JBR		0x04	/* jabber time out indication */
+#define	TSR_BUF_EMPTY	0x02	/* Tx buffer is empty */
+#define	TSR_BUF_FULL	0x01	/* Tx buffer is full */
+
+#define	TSR_BITS	\
+	"\020"		\
+	"\006ECOL"	\
+	"\005LCOL"	\
+	"\004LOSS_CRS"	\
+	"\003JBR"	\
+	"\002BUF_EMPTY"	\
+	"\001BUF_FULL"
+
+/* Receive status register in Rx packet field */
+#define	RSR_WEVENT	0x80	/* Wakeup event indication */
+#define	RSR_RX_BUF_FULL	0x40	/* Receive buffer full indication */
+#define	RSR_LKCHG	0x20	/* Link change indication */
+#define	RSR_RUNT	0x10	/* short packet indication */
+#define	RSR_LONG	0x08	/* Long packet indication*/
+#define	RSR_CRC		0x04	/* CRC error indication*/
+#define	RSR_FAE		0x02	/* Frame alignment error */
+#define	RSR_ROK		0x01	/* Receive OK indication */
+
+#define	RSR_ERRS	(RSR_RUNT | RSR_LONG | RSR_CRC | RSR_FAE)
+#define	RSR_BITS	\
+	"\020"		\
+	"\010WEVENT"	\
+	"\007RX_BUF_FULL"	\
+	"\006LKCHG"	\
+	"\005RUNT"	\
+	"\004LONG"	\
+	"\003CRC"	\
+	"\002FAE"	\
+	"\001ROK"
+
+/* Config 0 */
+
+#define	CON0_SUSLED	0x80
+#define	CON0_PARM_EN	0x40	/* parameter enable */
+#define	CON0_LDPS	0x08
+#define CON0_MSEL	0x04	/* media select 1:MII, 0:auto */
+#define CON0_LEDS	0x03	/* LED pattern */
+
+/* Config 1 */
+#define CON0_BWF	0x40	/* Broadcast wakeup function 1:on 0:off */
+#define CON0_MWF	0x20	/* Multicast wakeup function 1:on 0:off */
+#define CON0_UWF	0x10	/* Unicast wakeup function 1:on 0:off */
+#define CON0_LONGWF1	0x02	/* */
+#define CON0_LONGWF0	0x01	/* */
+
+
+/* MSR : Media Status register */
+#define	MSR_TXFCE	0x80	/* Tx Flow control enable */
+#define	MSR_RXFCE	0x40	/* Rx Flow control enable */
+#define	MSR_DUPLEX	0x10	/* full duplex */
+#define	MSR_SPEED_100	0x08	/* 100Mbps mode */
+#define	MSR_LINK	0x04	/* link status */
+#define	MSR_TXPF	0x02	/* 8150 sends pause packet */
+#define	MSR_RXPF	0x01	/* 8150 is in backoff state*/
+
+#define	MSR_BITS	\
+	"\020"		\
+	"\010TXFCE"	\
+	"\007RXFCE"	\
+	"\005DUPLEX"	\
+	"\004SPEED_100"	\
+	"\003LINK"	\
+	"\002TXPF"	\
+	"\001RXPF"
+
+/* MII PHY Address */
+#define	PHYADD_MASK	0x1f
+
+/* MII PHY Data */
+#define	PHYCNT_OWN	0x40	/* 8150 owns:1 not owns:0 */
+#define	PHYCNT_RWCR	0x20	/* write:1 read:0 */
+#define	PHYCNT_PHYOFF	0x1f
+
+/* BMCR (almost same with MII_CONTROL register) */
+#define	BMCR_RESET	0x8000	/* PHY reset */
+#define	BMCR_Spd_Set	0x2000	/* 100Mbps */
+#define	BMCR_ANE	0x1000	/* auto negotiation enable */
+#define	BMCR_RSA	0x0200	/* restart auto negotiation */
+#define	BMCR_duplex	0x0100	/* 100Mbps */
+
+/* Basic mode status register */
+/* Auto-negotiation Advertisement register */
+/* Auto-negotiation Link Partner Ability register */
+/* Auto-negotiation Expansion register */
+
+/* Nway test register */
+#define	NWAYT_NWLPBK	0x0080
+#define	NWAYT_ENNWLE	0x0008
+#define	NWAYT_FLAGABD	0x0004
+#define	NWAYT_FLAGPDF	0x0002
+#define	NWAYT_FLAGLSC	0x0001
+
+/* CS configuration register */
+#define	CS_TESTFUN	0x8000	/* */
+#define	CS_LD		0x0200	/* */
+#define	CS_HEARTBEAT	0x0100	/* */
+#define	CS_JBEN		0x0080	/* */
+#define	CS_F_LINK100	0x0040	/* */
+#define	CS_F_CONNECT	0x0020	/* */
+#define	CS_CON_STATUS	0x0008	/* */
+#define	CS_CON_STATUS_EN 0x0004	/* */
+#define	CS_PASS_SCR	0x0001	/* bypass scramble function */
+
+/*
+ * header format of rx packet
+ */
+#define	RXHD_MULT	0x8000	/* multicast packet */
+#define	RXHD_PHYS	0x4000	/* physical match packet */
+#define	RXHD_RUNT	0x2000	/* too short */
+#define	RXHD_VALID	0x1000	/* packet is ok */
+#define	RXHD_BYTECNT	0x0fff	/* rx byte count */
+
+#define	RXHD_BITS	\
+	"\020"		\
+	"\020MULT"	\
+	"\017PHYS"	\
+	"\016RUNT"	\
+	"\015VALID"
+/*
+ * Offset to EPROM contents
+ */
+#define	URF_EEPROM_BASE		0x1200
+#define	EPROM_EthernetID	0x0002
diff --git a/usr/src/uts/common/io/urf/urf_usbgem.c b/usr/src/uts/common/io/urf/urf_usbgem.c
new file mode 100644
index 0000000000..f61c8e3502
--- /dev/null
+++ b/usr/src/uts/common/io/urf/urf_usbgem.c
@@ -0,0 +1,1039 @@
+/*
+ * urf_usbgem.c : Realtek RTL8150 USB to Fast Ethernet Driver for Solaris
+ *
+ * Copyright (c) 2003-2012 Masayuki Murayama.  All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions are met:
+ *
+ * 1. Redistributions of source code must retain the above copyright notice,
+ *    this list of conditions and the following disclaimer.
+ *
+ * 2. Redistributions in binary form must reproduce the above copyright notice,
+ *    this list of conditions and the following disclaimer in the documentation
+ *    and/or other materials provided with the distribution.
+ *
+ * 3. Neither the name of the author nor the names of its contributors may be
+ *    used to endorse or promote products derived from this software without
+ *    specific prior written permission.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
+ * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
+ * COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
+ * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
+ * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS
+ * OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
+ * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+ * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
+ * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH
+ * DAMAGE.
+ */
+
+#pragma ident   "%W% %E%"
+
+/*
+ *  Changelog:
+ */
+
+/*
+ * TODO
+ */
+/* ======================================================= */
+
+/*
+ * Solaris system header files and macros
+ */
+
+/* minimum kernel headers for drivers */
+#include <sys/types.h>
+#include <sys/conf.h>
+#include <sys/debug.h>
+#include <sys/kmem.h>
+#include <sys/modctl.h>
+#include <sys/errno.h>
+#include <sys/ddi.h>
+#include <sys/sunddi.h>
+#include <sys/byteorder.h>
+
+/* ethernet stuff */
+#include <sys/ethernet.h>
+
+/* interface card depend stuff */
+#include <sys/stropts.h>
+#include <sys/stream.h>
+#include <sys/strlog.h>
+#include <sys/usb/usba.h>
+#include "usbgem.h"
+#include "usbgem_mii.h"
+#include "rtl8150reg.h"
+
+char	ident[] = "rtl8150 usbnic driver v" VERSION;
+
+/*
+ * Useful macros
+ */
+#define	ROUNDUP2(x, y)	(((x)+(y)-1) & ~((y)-1))
+#define	CHECK_AND_JUMP(err, label)	if (err != USB_SUCCESS) goto label
+
+/*
+ * Debugging
+ */
+#ifdef DEBUG_LEVEL
+static int urf_debug = DEBUG_LEVEL;
+#define	DPRINTF(n, args)	if (urf_debug > (n)) cmn_err args
+#else
+#define	DPRINTF(n, args)
+#endif
+
+/*
+ * Our configration for rtl8150
+ */
+/* timeouts */
+#define	ONESEC			(drv_usectohz(1*1000000))
+
+/*
+ * Local device definitions
+ */
+struct chip_info {
+	int		flags;
+	char		*name;
+	int		type;
+};
+
+#define	CHIPTABLESIZE	(sizeof (chiptbl_8150) / sizeof (struct chip_info))
+
+struct urf_dev {
+	/*
+	 * Misc HW information
+	 */
+	struct chip_info	*chip;
+	uint8_t			cr;
+	uint8_t			tsr;
+	uint16_t		rcr;
+	uint8_t			txok_cnt;
+};
+
+/*
+ * private functions
+ */
+
+/* mii operations */
+static uint16_t  urf_mii_read(struct usbgem_dev *, uint_t, int *errp);
+static void urf_mii_write(struct usbgem_dev *, uint_t, uint16_t, int *errp);
+
+/* nic operations */
+static int urf_attach_chip(struct usbgem_dev *);
+static int urf_reset_chip(struct usbgem_dev *);
+static int urf_init_chip(struct usbgem_dev *);
+static int urf_start_chip(struct usbgem_dev *);
+static int urf_stop_chip(struct usbgem_dev *);
+static int urf_set_media(struct usbgem_dev *);
+static int urf_set_rx_filter(struct usbgem_dev *);
+static int urf_get_stats(struct usbgem_dev *);
+
+/* packet operations */
+static mblk_t *urf_tx_make_packet(struct usbgem_dev *, mblk_t *);
+static mblk_t *urf_rx_make_packet(struct usbgem_dev *, mblk_t *);
+
+/* =============================================================== */
+/*
+ * I/O functions
+ */
+/* =============================================================== */
+#define	OUTB(dp, p, v, errp, label)	\
+	if ((*(errp) = usbgem_ctrl_out_val((dp), 	\
+	/* bmRequestType */ USB_DEV_REQ_HOST_TO_DEV	\
+		    | USB_DEV_REQ_TYPE_VENDOR | USB_DEV_REQ_RCPT_DEV,	\
+	/* bRequest */ USB_REQ_SET_ADDRESS,	\
+	/* wValue */   (p),	\
+	/* wIndex */   0,	\
+	/* wLength */  1,	\
+	/* value */   (v))) != USB_SUCCESS) goto label
+
+#define	OUTW(dp, p, v, errp, label)	\
+	if ((*(errp) = usbgem_ctrl_out_val((dp), 	\
+	/* bmRequestType */ USB_DEV_REQ_HOST_TO_DEV	\
+		    | USB_DEV_REQ_TYPE_VENDOR | USB_DEV_REQ_RCPT_DEV,	\
+	/* bRequest */ USB_REQ_SET_ADDRESS,	\
+	/* wValue */   (p),	\
+	/* wIndex */   0,	\
+	/* wLength */  2,	\
+	/* value */   (v))) != USB_SUCCESS) goto label
+
+#define	OUTS(dp, p, buf, len, errp, label)	\
+	if ((*(errp) = usbgem_ctrl_out((dp), 	\
+	/* bmRequestType */ USB_DEV_REQ_HOST_TO_DEV	\
+		    | USB_DEV_REQ_TYPE_VENDOR | USB_DEV_REQ_RCPT_DEV,	\
+	/* bRequest */ USB_REQ_SET_ADDRESS,	\
+	/* wValue */   (p),	\
+	/* wIndex */   0,	\
+	/* wLength */  (len),	\
+	/* value */    (buf),	\
+	/* size */     (len))) != USB_SUCCESS) goto label
+
+#define	IN(dp, p, vp, errp, label)	\
+	if ((*(errp) = usbgem_ctrl_in_val((dp), 	\
+	/* bmRequestType */ USB_DEV_REQ_DEV_TO_HOST	\
+		    | USB_DEV_REQ_TYPE_VENDOR | USB_DEV_REQ_RCPT_DEV,	\
+	/* bRequest */ USB_REQ_SET_ADDRESS,	\
+	/* wValue */  (p),	\
+	/* wIndex */  0,	\
+	/* wLength */ sizeof ((*vp)),	\
+	/* valuep */  (vp))) != USB_SUCCESS) goto label
+
+#define	INS(dp, p, buf, len, errp, label)	\
+	if ((*(errp) = usbgem_ctrl_in((dp), 	\
+	/* bmRequestType */ USB_DEV_REQ_DEV_TO_HOST	\
+		    | USB_DEV_REQ_TYPE_VENDOR | USB_DEV_REQ_RCPT_DEV,	\
+	/* bRequest */ USB_REQ_SET_ADDRESS,	\
+	/* wValue */   (p),	\
+	/* wIndex */   0,	\
+	/* wLength */  (len),	\
+	/* valuep */  (buf),	\
+	/* size   */  (len))) != USB_SUCCESS) goto label
+
+/* =============================================================== */
+/*
+ * variables
+ */
+/* =============================================================== */
+static int urf_ppa = 0;
+
+/* =============================================================== */
+/*
+ * Hardware manupilation
+ */
+/* =============================================================== */
+static int
+urf_reset_chip(struct usbgem_dev *dp)
+{
+	int		i;
+	int		err;
+	uint8_t		reg;
+	struct urf_dev	*lp = dp->private;
+
+	DPRINTF(0, (CE_CONT, "!%s: %s: called", dp->name, __func__));
+
+	lp->cr = 0;
+	OUTB(dp, CR, lp->cr | CR_SOFT_RST, &err, usberr);
+
+	for (i = 0; i < 100; i++) {
+		IN(dp, CR, &reg, &err, usberr);
+		if ((reg & CR_SOFT_RST) == 0) {
+			return (USB_SUCCESS);
+		}
+	}
+	/* time out */
+	cmn_err(CE_WARN, "%s: failed to reset: timeout", dp->name);
+	return (USB_FAILURE);
+
+usberr:
+	cmn_err(CE_NOTE, "!%s: %s: usberr detected", dp->name, __func__);
+	return (USB_FAILURE);
+}
+
+/*
+ * Setup rtl8150
+ */
+static int
+urf_init_chip(struct usbgem_dev *dp)
+{
+	int		i;
+	uint32_t	val;
+	int		err;
+	struct urf_dev	*lp = dp->private;
+
+	DPRINTF(0, (CE_CONT, "!%s: %s: called", dp->name, __func__));
+
+	/* ID registers: set later by urf_set_rx_filter */
+
+	/* Multicast registers: set later by urf_set_rx_filter */
+
+	/* Command register : Enable Tx and Rx before writing TCR and RCR */
+	lp->cr |= CR_RE | CR_TE;
+	OUTB(dp, CR, lp->cr, &err, usberr);
+
+	/* Transmit configration register : */
+	OUTB(dp, TCR, TCR_IFG_802_3, &err, usberr);
+
+	/* Receive configuration register :  disable rx filter */
+	lp->rcr = RCR_TAIL | RCR_AER | RCR_AR;
+	OUTW(dp, RCR, lp->rcr, &err, usberr);
+#ifdef notdef
+	/* Media status register */
+	err = urf_set_media(dp);
+	CHECK_AND_JUMP(err, usberr);
+#endif
+	/* Configuration register 0: no need to change */
+
+	DPRINTF(2, (CE_CONT, "!%s: %s: end (success)", dp->name, __func__));
+	return (USB_SUCCESS);
+
+usberr:
+	cmn_err(CE_NOTE, "!%s: %s: usberr detected", dp->name, __func__);
+	return (USB_FAILURE);
+}
+
+static int
+urf_start_chip(struct usbgem_dev *dp)
+{
+	struct urf_dev	*lp = dp->private;
+
+	/* do nothing */
+	return (USB_SUCCESS);
+}
+
+static int
+urf_stop_chip(struct usbgem_dev *dp)
+{
+	return (urf_reset_chip(dp));
+}
+
+static int
+urf_get_stats(struct usbgem_dev *dp)
+{
+	/* do nothing */
+	return (USB_SUCCESS);
+}
+
+static uint_t
+urf_mcast_hash(struct usbgem_dev *dp, const uint8_t *addr)
+{
+	return (usbgem_ether_crc_be(addr));
+}
+
+static int
+urf_set_rx_filter(struct usbgem_dev *dp)
+{
+	int		i;
+	uint16_t	mode;
+	uint8_t		mhash[8];
+	int		err;
+	int16_t		rcr;
+	struct urf_dev	*lp = dp->private;
+
+	DPRINTF(2, (CE_CONT, "!%s: %s: called, rxmode:%x",
+	    dp->name, __func__, dp->rxmode));
+
+	if (lp->rcr & (RCR_AB | RCR_AD | RCR_AAM | RCR_AAP | RCR_AM)) {
+#ifdef notdef
+		/* disable rx filter before changing it. */
+		lp->rcr &= ~(RCR_AB | RCR_AD | RCR_AAM | RCR_AAP | RCR_AM);
+		OUTW(dp, RCR, lp->rcr, &err, usberr);
+#else
+		/* receive all packets while we change rx filter*/
+		OUTW(dp, RCR, lp->rcr | RCR_AAM | RCR_AAP, &err, usberr);
+#endif
+	}
+
+	mode = RCR_AB	/* accept broadcast */
+	    | RCR_AD;	/* accept physical match  */
+	bzero(mhash, sizeof (mhash));
+
+	if (dp->rxmode & RXMODE_PROMISC) {
+		/* promiscious mode implies all multicast and all physical */
+		mode |= RCR_AAM | RCR_AAP;
+	} else if ((dp->rxmode & RXMODE_ALLMULTI) || dp->mc_count > 64/2) {
+		/* accept all multicast packets */
+		mode |= RCR_AAM;
+	} else if (dp->mc_count > 0) {
+		/*
+		 * make hash table to select interresting
+		 * multicast address only.
+		 */
+		mode |= RCR_AM;
+		for (i = 0; i < dp->mc_count; i++) {
+			uint_t	h;
+			/* hash table is 64 = 2^6 bit width */
+			h = dp->mc_list[i].hash >> (32 - 6);
+			mhash[h / 8] |= 1 << (h % 8);
+		}
+	}
+	lp->rcr |= mode;
+
+	/* set mac address */
+	OUTS(dp, IDR, dp->cur_addr.ether_addr_octet, ETHERADDRL, &err, usberr);
+
+	/* set multicast hash table */
+	if (mode & RCR_AM) {
+		/* need to set up multicast hash table */
+		OUTS(dp, MAR, mhash, sizeof (mhash), &err, usberr);
+	}
+
+	OUTW(dp, RCR, lp->rcr, &err, usberr);
+
+#if DEBUG_LEVEL > 2
+	IN(dp, RCR, &rcr, &err, usberr);
+	cmn_err(CE_CONT, "!%s: %s: rcr:%b returned",
+	    dp->name, __func__, rcr, RCR_BITS);
+#endif
+	return (USB_SUCCESS);
+
+usberr:
+	cmn_err(CE_NOTE, "!%s: %s: usberr detected", dp->name, __func__);
+	return (USB_FAILURE);
+}
+
+static int
+urf_set_media(struct usbgem_dev *dp)
+{
+	uint8_t		new;
+	uint8_t		old;
+	int		err;
+	struct urf_dev	*lp = dp->private;
+
+	DPRINTF(2, (CE_CONT, "!%s: %s: called", dp->name, __func__));
+
+	/* select duplex: do nothing */
+
+	/* select speed: do nothing */
+
+	/* flow control */
+	IN(dp, MSR, &old, &err, usberr);
+
+
+	/* setup flow control */
+	new = old & ~(MSR_TXFCE | MSR_RXFCE);
+	switch (dp->flow_control) {
+	case FLOW_CONTROL_SYMMETRIC:
+		new |= MSR_TXFCE | MSR_RXFCE;
+		break;
+
+	case FLOW_CONTROL_TX_PAUSE:
+		new |= MSR_TXFCE;
+		break;
+
+	case FLOW_CONTROL_RX_PAUSE:
+		new |= MSR_RXFCE;
+		break;
+
+	case FLOW_CONTROL_NONE:
+	default:
+		break;
+	}
+
+	if (new != old) {
+		OUTB(dp, MSR, new, &err, usberr);
+	}
+	DPRINTF(2, (CE_CONT, "!%s: %s: returned", dp->name, __func__));
+	return (USB_SUCCESS);
+
+usberr:
+	cmn_err(CE_NOTE, "!%s: %s: usberr detected", dp->name, __func__);
+	return (USB_FAILURE);
+}
+
+/*
+ * send/receive packet check
+ */
+static mblk_t *
+urf_tx_make_packet(struct usbgem_dev *dp, mblk_t *mp)
+{
+	size_t		len;
+	mblk_t		*new;
+	mblk_t		*tp;
+	uint8_t		*bp;
+	uint8_t		*last_pos;
+
+	len = msgdsize(mp);
+
+	if (len < ETHERMIN || mp->b_cont != NULL || (len & 0x3f) == 0) {
+		/*
+		 * re-allocate mp
+		 */
+		len = max(len, ETHERMIN);
+
+		if ((len & 0x3f) == 0) {
+			/* workaround for buggy USB hba */
+			len++;
+		}
+
+		if ((new = allocb(len, 0)) == NULL) {
+			return (NULL);
+		}
+
+		/* copy contents of the buffer */
+		new->b_wptr = new->b_rptr + len;
+		bp = new->b_rptr;
+		for (tp = mp; tp; tp = tp->b_cont) {
+			len = tp->b_wptr - tp->b_rptr;
+			bcopy(tp->b_rptr, bp, len);
+			bp += len;
+		}
+
+		last_pos = new->b_wptr;
+		while (bp < last_pos) {
+			*bp++ = 0;
+		}
+
+		mp = new;
+	}
+
+	return (mp);
+}
+
+static void
+urf_dump_packet(struct usbgem_dev *dp, uint8_t *bp, int n)
+{
+	int	i;
+
+	for (i = 0; i < n; i += 8, bp += 8) {
+		cmn_err(CE_CONT, "%02x %02x %02x %02x %02x %02x %02x %02x",
+		    bp[0], bp[1], bp[2], bp[3], bp[4], bp[5], bp[6], bp[7]);
+	}
+}
+
+static mblk_t *
+urf_rx_make_packet(struct usbgem_dev *dp, mblk_t *mp)
+{
+	uint8_t		*p;
+	uint16_t	rxhd;
+	uint_t		len;
+
+	ASSERT(mp != NULL);
+	len = msgdsize(mp);
+#ifdef DEBUG_LEVEL
+	DPRINTF(2, (CE_CONT, "!%s: time:%d %s: len:%d cont:%p",
+	    dp->name, ddi_get_lbolt(), __func__, len, mp->b_cont));
+
+	if (urf_debug > 2) {
+		urf_dump_packet(dp, mp->b_rptr, max(6, len));
+	}
+#endif
+	if (len < ETHERMIN + ETHERFCSL) {
+		/* Too short */
+		dp->stats.runt++;
+		dp->stats.errrcv++;
+		return (NULL);
+	}
+
+	/* get Rx header which is placed at tail of the packet. */
+	p = mp->b_wptr - 4;
+	rxhd = (p[1] << 8) | p[0];
+	len = rxhd & RXHD_BYTECNT;
+
+	DPRINTF(2, (CE_CONT, "!%s: %s: rsr:%b len:%d",
+	    dp->name, __func__, rxhd, RXHD_BITS, len));
+
+	/* check if error happen */
+	if ((rxhd & (RXHD_VALID)) == 0) {
+		DPRINTF(-1, (CE_CONT, "!%s: %s: rxhd:%b",
+		    dp->name, __func__, rxhd, RXHD_BITS));
+		if (rxhd & RXHD_RUNT) {
+			dp->stats.runt++;
+		}
+
+		dp->stats.errrcv++;
+		return (NULL);
+	}
+#ifdef notdef
+	/* check packet size */
+	if (len > ETHERMAX + ETHERFCSL) {
+		/* too long */
+		dp->stats.frame_too_long++;
+		dp->stats.errrcv++;
+		return (NULL);
+	} else if (len < ETHERMIN + ETHERFCSL) {
+		dp->stats.runt++;
+		dp->stats.errrcv++;
+		return (NULL);
+	}
+#endif
+	/* remove tailing crc field */
+	mp->b_wptr -= ETHERFCSL;
+	return (mp);
+}
+
+/*
+ * MII Interfaces
+ */
+static uint16_t
+urf_mii_read(struct usbgem_dev *dp, uint_t index, int *errp)
+{
+	int		reg;
+	uint16_t	val;
+
+	DPRINTF(4, (CE_CONT, "!%s: %s: called, ix:%d",
+	    dp->name, __func__, index));
+
+	*errp = USB_SUCCESS;
+
+	switch (index) {
+	case MII_CONTROL:
+		reg = BMCR;
+		break;
+
+	case MII_STATUS:
+		reg = BMSR;
+		break;
+
+	case MII_AN_ADVERT:
+		reg = ANAR;
+		break;
+
+	case MII_AN_LPABLE:
+		reg = ANLP;
+		break;
+
+	case MII_AN_EXPANSION:
+		reg = ANER;
+		break;
+
+	default:
+		return (0);
+	}
+
+	IN(dp, reg, &val, errp, usberr);
+
+	if (index == MII_STATUS) {
+		uint8_t	msr;
+		/*
+		 * Fix MII status register as it does't have LINKUP and
+		 * MFPRMBLSUPR bits.
+		 */
+		IN(dp, MSR, &msr, errp, usberr);
+
+		val |= (MII_STATUS_MFPRMBLSUPR | MII_STATUS_LINKUP);
+		if ((msr & MSR_LINK) == 0) {
+			val &= ~MII_STATUS_LINKUP;
+		}
+	}
+
+	return (val);
+
+usberr:
+	cmn_err(CE_CONT,
+	    "!%s: %s: usberr(%d) detected", dp->name, __func__, *errp);
+
+	return (0);
+}
+
+static void
+urf_mii_write(struct usbgem_dev *dp, uint_t index, uint16_t val, int *errp)
+{
+	int	reg;
+
+	DPRINTF(5, (CE_CONT, "!%s: %s called", dp->name, __func__));
+
+	*errp = USB_SUCCESS;
+
+	switch (index) {
+	case MII_CONTROL:
+		reg = BMCR;
+		break;
+
+	case MII_STATUS:
+		reg = BMSR;
+		break;
+
+	case MII_AN_ADVERT:
+		reg = ANAR;
+		break;
+
+	case MII_AN_LPABLE:
+		reg = ANLP;
+		break;
+
+	case MII_AN_EXPANSION:
+		reg = ANER;
+		break;
+
+	default:
+		return;
+	}
+
+	OUTW(dp, reg, val, errp, usberr);
+usberr:
+	;
+}
+
+/* ======================================================== */
+/*
+ * OS depend (device driver DKI) routine
+ */
+/* ======================================================== */
+static void
+urf_eeprom_dump(struct usbgem_dev *dp, int size)
+{
+	int		i;
+	int		err;
+	uint16_t	w0, w1, w2, w3;
+
+	cmn_err(CE_CONT, "!%s: eeprom dump:", dp->name);
+	for (i = URF_EEPROM_BASE; i < size + URF_EEPROM_BASE; i += 8) {
+		IN(dp, i + 0, &w0, &err, usberr);
+		IN(dp, i + 2, &w1, &err, usberr);
+		IN(dp, i + 4, &w2, &err, usberr);
+		IN(dp, i + 6, &w3, &err, usberr);
+		cmn_err(CE_CONT, "!0x%02x: 0x%04x 0x%04x 0x%04x 0x%04x",
+		    i - URF_EEPROM_BASE, w0, w1, w2, w3);
+	}
+usberr:
+	;
+}
+
+static int
+urf_attach_chip(struct usbgem_dev *dp)
+{
+	int		i;
+	uint8_t		old;
+	uint_t		new;
+	uint8_t		reg;
+	int		err;
+	struct urf_dev	*lp = dp->private;
+
+	/*
+	 * setup flow control bit in eeprom
+	 */
+	IN(dp, URF_EEPROM_BASE + 9, &old, &err, usberr);
+
+	DPRINTF(0, (CE_CONT, "!%s: eeprom offset 9: %02x", dp->name, old));
+
+	if (dp->ugc.usbgc_flow_control != FLOW_CONTROL_NONE) {
+		/* enable PAUSE bit */
+		new = old | 0x04;
+	} else {
+		/* clear PAUSE bit */
+		new = old & ~0x04;
+	}
+	if (new != old) {
+		/* make eeprom writable */
+		OUTB(dp, CR, lp->cr | CR_WEPROM, &err, usberr);
+
+		/* eerom allows only word access for writing */
+		IN(dp, URF_EEPROM_BASE + 8, &reg, &err, usberr);
+		new = (new << 8) | reg;
+
+		OUTW(dp, URF_EEPROM_BASE + 8, new, &err, usberr);
+
+		/* make eeprom non-writable */
+		OUTB(dp, CR, lp->cr, &err, usberr);
+	}
+
+	/*
+	 * load EEPROM contents into nic
+	 */
+	OUTB(dp, CR, lp->cr | CR_AUTOLOAD, &err, usberr);
+	CHECK_AND_JUMP(err, usberr);
+
+	for (i = 0; i < 100; i++) {
+		IN(dp, CR, &reg, &err, usberr);
+		if ((reg & CR_AUTOLOAD) == 0) {
+			goto autoload_done;
+		}
+	}
+	/* timeout */
+	cmn_err(CE_WARN, "%s: %s: failed to autoload: timeout",
+	    dp->name, __func__);
+	goto usberr;
+
+autoload_done:
+	/*
+	 * mac address in EEPROM has loaded to ID registers.
+	 */
+	INS(dp, IDR, dp->dev_addr.ether_addr_octet, ETHERADDRL, &err, usberr);
+
+	/* no need to scan phy */
+	dp->mii_phy_addr = -1;
+
+#if DEBUG_LEVEL > 2
+	urf_eeprom_dump(dp, 0x80);
+#endif
+
+#ifdef CONFIG_VLAN
+	dp->misc_flag = USBGEM_VLAN;
+#endif
+	return (USB_SUCCESS);
+
+usberr:
+	cmn_err(CE_WARN, "%s: urf_attach_chip: usb error detected", dp->name);
+	return (USB_FAILURE);
+}
+
+static int
+urfattach(dev_info_t *dip, ddi_attach_cmd_t cmd)
+{
+	int			i;
+	ddi_iblock_cookie_t	c;
+	int			ret;
+	int			unit;
+	struct chip_info	*p;
+	const char		*drv_name;
+	struct usbgem_dev	*dp;
+	void			*base;
+	struct usbgem_conf	*ugcp;
+	struct urf_dev		*lp;
+
+	unit =  ddi_get_instance(dip);
+	drv_name = ddi_driver_name(dip);
+
+	DPRINTF(3, (CE_CONT, "!%s%d: %s: called, cmd:%d",
+	    drv_name, __func__, unit, cmd));
+
+	if (cmd == DDI_ATTACH) {
+		/*
+		 * Check if the chip is supported.
+		 */
+
+		/*
+		 * Check the chip if it is really realtek rtl8150
+		 */
+
+		/*
+		 * construct usbgem configration
+		 */
+		ugcp = kmem_zalloc(sizeof (*ugcp), KM_SLEEP);
+
+		/* name */
+		sprintf(ugcp->usbgc_name,
+		    "%s%d(ppa=%d)", drv_name, unit, urf_ppa);
+#ifdef USBGEM_CONFIG_GLDv3
+		ugcp->usbgc_ppa = urf_ppa;
+#else
+		ugcp->usbgc_ppa = unit;
+#endif
+		ugcp->usbgc_ifnum = 0;
+		ugcp->usbgc_alt = 0;
+
+		ugcp->usbgc_tx_list_max = 16;
+
+		/* the rx status partially replaces FCS */
+		ugcp->usbgc_rx_header_len = 0;
+		ugcp->usbgc_rx_list_max = 64;
+
+		/* time out parameters */
+		ugcp->usbgc_tx_timeout = USBGEM_TX_TIMEOUT;
+		ugcp->usbgc_tx_timeout_interval = ONESEC;
+
+		/* flow control */
+		ugcp->usbgc_flow_control = FLOW_CONTROL_RX_PAUSE;
+
+		/* MII timeout parameters */
+		ugcp->usbgc_mii_link_watch_interval = ONESEC;
+		ugcp->usbgc_mii_an_watch_interval = ONESEC/5;
+		ugcp->usbgc_mii_reset_timeout = MII_RESET_TIMEOUT; /* 1 sec */
+		ugcp->usbgc_mii_an_timeout = MII_AN_TIMEOUT;	/* 5 sec */
+		ugcp->usbgc_mii_an_wait = (25*ONESEC)/10;
+		ugcp->usbgc_mii_linkdown_timeout = MII_LINKDOWN_TIMEOUT;
+
+		ugcp->usbgc_mii_an_delay = ONESEC/10;
+		ugcp->usbgc_mii_linkdown_action = MII_ACTION_RSA;
+		ugcp->usbgc_mii_linkdown_timeout_action = MII_ACTION_RESET;
+		ugcp->usbgc_mii_dont_reset = B_FALSE;
+
+		/* I/O methods */
+
+		/* mac operation */
+		ugcp->usbgc_attach_chip = &urf_attach_chip;
+		ugcp->usbgc_reset_chip = &urf_reset_chip;
+		ugcp->usbgc_init_chip = &urf_init_chip;
+		ugcp->usbgc_start_chip = &urf_start_chip;
+		ugcp->usbgc_stop_chip = &urf_stop_chip;
+		ugcp->usbgc_multicast_hash = &urf_mcast_hash;
+
+		ugcp->usbgc_set_rx_filter = &urf_set_rx_filter;
+		ugcp->usbgc_set_media = &urf_set_media;
+		ugcp->usbgc_get_stats = &urf_get_stats;
+#ifdef notdef
+		ugcp->usbgc_interrupt = &urf_interrupt;
+#else
+		ugcp->usbgc_interrupt = NULL;
+#endif
+		/* packet operation */
+		ugcp->usbgc_tx_make_packet = &urf_tx_make_packet;
+		ugcp->usbgc_rx_make_packet = &urf_rx_make_packet;
+
+		/* mii operations */
+		ugcp->usbgc_mii_probe = &usbgem_mii_probe_default;
+		ugcp->usbgc_mii_init = &usbgem_mii_init_default;
+		ugcp->usbgc_mii_config = &usbgem_mii_config_default;
+		ugcp->usbgc_mii_read = &urf_mii_read;
+		ugcp->usbgc_mii_write = &urf_mii_write;
+
+		/* mtu */
+		ugcp->usbgc_min_mtu = ETHERMTU;
+		ugcp->usbgc_max_mtu = ETHERMTU;
+		ugcp->usbgc_default_mtu = ETHERMTU;
+
+		lp = kmem_zalloc(sizeof (struct urf_dev), KM_SLEEP);
+		lp->chip = p;
+
+		ddi_set_driver_private(dip, NULL);
+
+		dp = usbgem_do_attach(dip, ugcp, lp, sizeof (struct urf_dev));
+
+		kmem_free(ugcp, sizeof (*ugcp));
+
+		if (dp != NULL) {
+			urf_ppa++;
+			return (DDI_SUCCESS);
+		}
+
+err_free_mem:
+		kmem_free(lp, sizeof (struct urf_dev));
+err_close_pipe:
+err:
+		return (DDI_FAILURE);
+	}
+	if (cmd == DDI_RESUME) {
+		return (usbgem_resume(dip));
+	}
+	return (DDI_FAILURE);
+}
+
+static int
+urfdetach(dev_info_t *dip, ddi_detach_cmd_t cmd)
+{
+	int	ret;
+
+	if (cmd == DDI_DETACH) {
+		ret = usbgem_do_detach(dip);
+		if (ret != DDI_SUCCESS) {
+			return (DDI_FAILURE);
+		}
+		urf_ppa--;
+		return (DDI_SUCCESS);
+	}
+	if (cmd == DDI_SUSPEND) {
+		return (usbgem_suspend(dip));
+	}
+	return (DDI_FAILURE);
+}
+
+/* ======================================================== */
+/*
+ * OS depend (loadable streams driver) routine
+ */
+/* ======================================================== */
+#ifdef USBGEM_CONFIG_GLDv3
+USBGEM_STREAM_OPS(urf_ops, urfattach, urfdetach);
+#else
+static	struct module_info urfminfo = {
+	0,			/* mi_idnum */
+	"urf",			/* mi_idname */
+	0,			/* mi_minpsz */
+	ETHERMTU,		/* mi_maxpsz */
+	ETHERMTU*128,		/* mi_hiwat */
+	1,			/* mi_lowat */
+};
+
+static	struct qinit urfrinit = {
+	(int (*)()) NULL,	/* qi_putp */
+	usbgem_rsrv,		/* qi_srvp */
+	usbgem_open,		/* qi_qopen */
+	usbgem_close,		/* qi_qclose */
+	(int (*)()) NULL,	/* qi_qadmin */
+	&urfminfo,		/* qi_minfo */
+	NULL			/* qi_mstat */
+};
+
+static	struct qinit urfwinit = {
+	usbgem_wput,		/* qi_putp */
+	usbgem_wsrv,		/* qi_srvp */
+	(int (*)()) NULL,	/* qi_qopen */
+	(int (*)()) NULL,	/* qi_qclose */
+	(int (*)()) NULL,	/* qi_qadmin */
+	&urfminfo,		/* qi_minfo */
+	NULL			/* qi_mstat */
+};
+
+static struct streamtab	urf_info = {
+	&urfrinit,	/* st_rdinit */
+	&urfwinit,	/* st_wrinit */
+	NULL,		/* st_muxrinit */
+	NULL		/* st_muxwrinit */
+};
+
+static	struct cb_ops cb_urf_ops = {
+	nulldev,	/* cb_open */
+	nulldev,	/* cb_close */
+	nodev,		/* cb_strategy */
+	nodev,		/* cb_print */
+	nodev,		/* cb_dump */
+	nodev,		/* cb_read */
+	nodev,		/* cb_write */
+	nodev,		/* cb_ioctl */
+	nodev,		/* cb_devmap */
+	nodev,		/* cb_mmap */
+	nodev,		/* cb_segmap */
+	nochpoll,	/* cb_chpoll */
+	ddi_prop_op,	/* cb_prop_op */
+	&urf_info,	/* cb_stream */
+	D_NEW|D_MP	/* cb_flag */
+};
+
+static	struct dev_ops urf_ops = {
+	DEVO_REV,	/* devo_rev */
+	0,		/* devo_refcnt */
+	usbgem_getinfo,	/* devo_getinfo */
+	nulldev,	/* devo_identify */
+	nulldev,	/* devo_probe */
+	urfattach,	/* devo_attach */
+	urfdetach,	/* devo_detach */
+	nodev,		/* devo_reset */
+	&cb_urf_ops,	/* devo_cb_ops */
+	NULL,		/* devo_bus_ops */
+	usbgem_power,	/* devo_power */
+#if DEVO_REV >= 4
+	usbgem_quiesce, /* devo_quiesce */
+#endif
+
+};
+#endif
+
+static struct modldrv modldrv = {
+	&mod_driverops,	/* Type of module.  This one is a driver */
+	ident,
+	&urf_ops,	/* driver ops */
+};
+
+static struct modlinkage modlinkage = {
+	MODREV_1, &modldrv, NULL
+};
+
+/* ======================================================== */
+/*
+ * _init : done
+ */
+/* ======================================================== */
+int
+_init(void)
+{
+	int 	status;
+
+	DPRINTF(2, (CE_CONT, "!urf: _init: called"));
+
+	status = usbgem_mod_init(&urf_ops, "urf");
+	if (status != DDI_SUCCESS) {
+		return (status);
+	}
+	status = mod_install(&modlinkage);
+	if (status != DDI_SUCCESS) {
+		usbgem_mod_fini(&urf_ops);
+	}
+	return (status);
+}
+
+/*
+ * _fini : done
+ */
+int
+_fini(void)
+{
+	int	status;
+
+	DPRINTF(2, (CE_CONT, "!urf: _fini: called"));
+	status = mod_remove(&modlinkage);
+	if (status == DDI_SUCCESS) {
+		usbgem_mod_fini(&urf_ops);
+	}
+	return (status);
+}
+
+int
+_info(struct modinfo *modinfop)
+{
+	return (mod_info(&modlinkage, modinfop));
+}
diff --git a/usr/src/uts/common/io/usb/usba/usba_ugen.c b/usr/src/uts/common/io/usb/usba/usba_ugen.c
index cb20c24270..5852e40799 100644
--- a/usr/src/uts/common/io/usb/usba/usba_ugen.c
+++ b/usr/src/uts/common/io/usb/usba/usba_ugen.c
@@ -23,6 +23,10 @@
  */
 
 /*
+ * Copyright (c) 2014, Joyent, Inc. All rights reserved.
+ */
+
+/*
  * UGEN: USB Generic Driver support code
  *
  * This code provides entry points called by the ugen driver or other
@@ -1082,7 +1086,10 @@ usb_ugen_poll(usb_ugen_hdl_t usb_ugen_hdl, dev_t dev, short events,
 				    ((epp->ep_state &
 				    UGEN_EP_STATE_INTR_IN_POLLING_ON) == 0)) {
 					*reventsp |= POLLIN;
-				} else if (!anyyet) {
+				}
+
+				if ((!*reventsp && !anyyet) ||
+				    (events & POLLET)) {
 					*phpp = &epp->ep_pollhead;
 					epp->ep_state |=
 					    UGEN_EP_STATE_INTR_IN_POLL_PENDING;
@@ -1101,7 +1108,10 @@ usb_ugen_poll(usb_ugen_hdl_t usb_ugen_hdl, dev_t dev, short events,
 				    ((epp->ep_state &
 				    UGEN_EP_STATE_ISOC_IN_POLLING_ON) == 0)) {
 					*reventsp |= POLLIN;
-				} else if (!anyyet) {
+				}
+
+				if ((!*reventsp && !anyyet) ||
+				    (events & POLLET)) {
 					*phpp = &epp->ep_pollhead;
 					epp->ep_state |=
 					    UGEN_EP_STATE_ISOC_IN_POLL_PENDING;
@@ -1115,9 +1125,10 @@ usb_ugen_poll(usb_ugen_hdl_t usb_ugen_hdl, dev_t dev, short events,
 
 			break;
 		case UGEN_MINOR_DEV_STAT_NODE:
-			if (ugenp->ug_ds.dev_stat & UGEN_DEV_STATUS_CHANGED) {
+			if (ugenp->ug_ds.dev_stat & UGEN_DEV_STATUS_CHANGED)
 				*reventsp |= POLLIN;
-			} else if (!anyyet) {
+
+			if ((!*reventsp && !anyyet) || (events & POLLET)) {
 				*phpp = &ugenp->ug_ds.dev_pollhead;
 				ugenp->ug_ds.dev_stat |=
 				    UGEN_DEV_STATUS_POLL_PENDING;
@@ -1131,9 +1142,10 @@ usb_ugen_poll(usb_ugen_hdl_t usb_ugen_hdl, dev_t dev, short events,
 			break;
 		}
 	} else {
-		if (ugenp->ug_ds.dev_stat & UGEN_DEV_STATUS_CHANGED) {
+		if (ugenp->ug_ds.dev_stat & UGEN_DEV_STATUS_CHANGED)
 			*reventsp |= POLLHUP|POLLIN;
-		} else if (!anyyet) {
+
+		if ((!*reventsp && !anyyet) || (events & POLLET)) {
 			*phpp = &ugenp->ug_ds.dev_pollhead;
 			ugenp->ug_ds.dev_stat |=
 			    UGEN_DEV_STATUS_POLL_PENDING;
diff --git a/usr/src/uts/common/io/usbgem/usbgem.c b/usr/src/uts/common/io/usbgem/usbgem.c
new file mode 100644
index 0000000000..a42f7119ef
--- /dev/null
+++ b/usr/src/uts/common/io/usbgem/usbgem.c
@@ -0,0 +1,6389 @@
+/*
+ * usbgem.c: General USB to Fast Ethernet mac driver framework
+ *
+ * Copyright (c) 2002-2012 Masayuki Murayama.  All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions are met:
+ *
+ * 1. Redistributions of source code must retain the above copyright notice,
+ *    this list of conditions and the following disclaimer.
+ *
+ * 2. Redistributions in binary form must reproduce the above copyright notice,
+ *    this list of conditions and the following disclaimer in the documentation
+ *    and/or other materials provided with the distribution.
+ *
+ * 3. Neither the name of the author nor the names of its contributors may be
+ *    used to endorse or promote products derived from this software without
+ *    specific prior written permission.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
+ * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
+ * COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
+ * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
+ * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS
+ * OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
+ * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+ * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
+ * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH
+ * DAMAGE.
+ */
+
+#pragma	ident	"@(#)usbgem.c 1.6     12/02/09"
+
+/*
+ * Change log
+ */
+
+/*
+ * TODO:
+ * 	implement DELAYED_START
+ */
+
+/*
+ * System Header files.
+ */
+#include <sys/types.h>
+#include <sys/conf.h>
+#include <sys/debug.h>
+#include <sys/kmem.h>
+#include <sys/vtrace.h>
+#include <sys/ethernet.h>
+#include <sys/modctl.h>
+#include <sys/errno.h>
+#include <sys/ddi.h>
+#include <sys/sunddi.h>
+#ifndef USBGEM_CONFIG_GLDv3
+#include <sys/dlpi.h>
+#include <sys/strsubr.h>
+#endif
+#include <sys/stream.h>		/* required for MBLK* */
+#include <sys/strsun.h>		/* required for mionack() */
+#include <sys/byteorder.h>
+
+#include <sys/usb/usba.h>
+#ifdef USBGEM_CONFIG_GLDv3
+#include <inet/common.h>
+#include <inet/led.h>
+#include <inet/mi.h>
+#include <inet/nd.h>
+#endif
+
+/* supplement definitions */
+extern const char *usb_str_cr(usb_cr_t);
+
+#ifndef USBGEM_CONFIG_GLDv3
+#pragma weak	gld_linkstate
+#endif
+#include <sys/note.h>
+
+#include "usbgem_mii.h"
+#include "usbgem.h"
+
+#ifdef MODULE
+char	ident[] = "usb general ethernet mac driver v" VERSION;
+#else
+extern char	ident[];
+#endif
+
+/* Debugging support */
+#ifdef USBGEM_DEBUG_LEVEL
+static int usbgem_debug = USBGEM_DEBUG_LEVEL;
+#define	DPRINTF(n, args)	if (usbgem_debug > (n)) cmn_err args
+#else
+#define	DPRINTF(n, args)
+#endif
+
+/*
+ * Useful macros and typedefs
+ */
+#define	ROUNDUP(x, a)		(((x) + (a) - 1) & ~((a) - 1))
+#define	DEFAULT_PIPE(dp)	((dp)->reg_data->dev_default_ph)
+#define	VTAG_SIZE	4
+#define	BOOLEAN(x)	((x) != 0)
+/*
+ * configuration parameters
+ */
+#define	USBDRV_MAJOR_VER	2
+#define	USBDRV_MINOR_VER	0
+
+#define	ETHERHEADERL	(sizeof (struct ether_header))
+#define	MAXPKTLEN(dp)	((dp)->mtu + ETHERHEADERL)
+#define	MAXPKTBUF(dp)	((dp)->mtu + ETHERHEADERL + ETHERFCSL)
+
+#define	WATCH_INTERVAL_FAST	drv_usectohz(100*1000)
+
+#define	STOP_GRACEFUL	B_TRUE
+
+/*
+ * Private functions
+ */
+static int usbgem_open_pipes(struct usbgem_dev *dp);
+static int usbgem_close_pipes(struct usbgem_dev *dp);
+static void usbgem_intr_cb(usb_pipe_handle_t, usb_intr_req_t *);
+static void usbgem_bulkin_cb(usb_pipe_handle_t, usb_bulk_req_t *);
+static void usbgem_bulkout_cb(usb_pipe_handle_t, usb_bulk_req_t *);
+
+static int usbgem_mii_start(struct usbgem_dev *);
+static void usbgem_mii_stop(struct usbgem_dev *);
+
+/* local buffer management */
+static int usbgem_init_rx_buf(struct usbgem_dev *);
+
+/* internal mac interfaces */
+static void usbgem_tx_timeout(struct usbgem_dev *);
+static void usbgem_mii_link_watcher(struct usbgem_dev *);
+static int usbgem_mac_init(struct usbgem_dev *);
+static int usbgem_mac_start(struct usbgem_dev *);
+static int usbgem_mac_stop(struct usbgem_dev *, int, boolean_t);
+static void usbgem_mac_ioctl(struct usbgem_dev *, queue_t *, mblk_t *);
+
+int usbgem_speed_value[] = {10, 100, 1000};
+
+static int usbgem_ctrl_retry = 5;
+
+/* usb event support */
+static int usbgem_disconnect_cb(dev_info_t *dip);
+static int usbgem_reconnect_cb(dev_info_t *dip);
+int usbgem_suspend(dev_info_t *dip);
+int usbgem_resume(dev_info_t *dip);
+
+static uint8_t usbgem_bcastaddr[] = {
+	0xff, 0xff, 0xff, 0xff, 0xff, 0xff
+};
+
+#ifdef MODULE
+extern struct mod_ops mod_miscops;
+
+static struct modlmisc modlmisc = {
+	&mod_miscops,
+	"usbgem v" VERSION,
+};
+
+static struct modlinkage modlinkage = {
+	MODREV_1, &modlmisc, NULL
+};
+
+/*
+ * _init : done
+ */
+int
+_init(void)
+{
+	int 	status;
+
+	DPRINTF(2, (CE_CONT, "!usbgem: _init: called"));
+	status = mod_install(&modlinkage);
+
+	return (status);
+}
+
+/*
+ * _fini : done
+ */
+int
+_fini(void)
+{
+	int	status;
+
+	DPRINTF(2, (CE_CONT, "!usbgem: _fini: called"));
+	status = mod_remove(&modlinkage);
+	return (status);
+}
+
+int
+_info(struct modinfo *modinfop)
+{
+	return (mod_info(&modlinkage, modinfop));
+}
+#endif /* MODULE */
+
+/* ============================================================== */
+/*
+ * Ether CRC calculation utilities
+ */
+/* ============================================================== */
+/*
+ * Ether CRC calculation according to 21143 data sheet
+ */
+#define	CRC32_POLY_LE	0xedb88320
+uint32_t
+usbgem_ether_crc_le(const uint8_t *addr)
+{
+	int		idx;
+	int		bit;
+	uint_t		data;
+	uint32_t	crc = 0xffffffff;
+
+	crc = 0xffffffff;
+	for (idx = 0; idx < ETHERADDRL; idx++) {
+		for (data = *addr++, bit = 0; bit < 8; bit++, data >>= 1) {
+			crc = (crc >> 1) ^
+			    (((crc ^ data) & 1) ? CRC32_POLY_LE : 0);
+		}
+	}
+	return	(crc);
+}
+
+#define	CRC32_POLY_BE	0x04c11db7
+uint32_t
+usbgem_ether_crc_be(const uint8_t *addr)
+{
+	int		idx;
+	int		bit;
+	uint_t		data;
+	uint32_t	crc;
+
+	crc = 0xffffffff;
+	for (idx = 0; idx < ETHERADDRL; idx++) {
+		for (data = *addr++, bit = 0; bit < 8; bit++, data >>= 1) {
+			crc = (crc << 1) ^
+			    ((((crc >> 31) ^ data) & 1) ? CRC32_POLY_BE : 0);
+		}
+	}
+	return (crc);
+}
+
+int
+usbgem_prop_get_int(struct usbgem_dev *dp, char *prop_template, int def_val)
+{
+	char	propname[32];
+
+	(void) sprintf(propname, prop_template, dp->name);
+
+	return (ddi_prop_get_int(DDI_DEV_T_ANY, dp->dip,
+	    DDI_PROP_DONTPASS, propname, def_val));
+}
+
+static int
+usbgem_population(uint32_t x)
+{
+	int	i;
+	int	cnt;
+
+	cnt = 0;
+	for (i = 0; i < 32; i++) {
+		if (x & (1 << i)) {
+			cnt++;
+		}
+	}
+	return (cnt);
+}
+
+static clock_t
+usbgem_timestamp_nz()
+{
+	clock_t	now;
+	now = ddi_get_lbolt();
+	return (now ? now : (clock_t)1);
+}
+
+#ifdef USBGEM_DEBUG_LEVEL
+#ifdef USBGEM_DEBUG_VLAN
+#ifdef notdef
+#include <netinet/in.h>
+#endif
+static void
+usbgem_dump_packet(struct usbgem_dev *dp, char *title, mblk_t *mp,
+    boolean_t check_cksum)
+{
+	char	msg[180];
+	uint8_t	buf[18+20+20];
+	uint8_t	*p;
+	size_t	offset;
+	uint_t	ethertype;
+	uint_t	proto;
+	uint_t	ipproto = 0;
+	uint_t	iplen;
+	uint_t	iphlen;
+	uint_t	tcplen;
+	uint_t	udplen;
+	uint_t	cksum;
+	int	rest;
+	int	len;
+	char	*bp;
+	mblk_t	*tp;
+	extern uint_t	ip_cksum(mblk_t *, int, uint32_t);
+
+	msg[0] = 0;
+	bp = msg;
+
+	rest = sizeof (buf);
+	offset = 0;
+	for (tp = mp; tp; tp = tp->b_cont) {
+		len = tp->b_wptr - tp->b_rptr;
+		len = min(rest, len);
+		bcopy(tp->b_rptr, &buf[offset], len);
+		rest -= len;
+		offset += len;
+		if (rest == 0) {
+			break;
+		}
+	}
+
+	offset = 0;
+	p = &buf[offset];
+
+	/* ethernet address */
+	sprintf(bp,
+	    "ether: %02x:%02x:%02x:%02x:%02x:%02x"
+	    " -> %02x:%02x:%02x:%02x:%02x:%02x",
+	    p[6], p[7], p[8], p[9], p[10], p[11],
+	    p[0], p[1], p[2], p[3], p[4], p[5]);
+	bp = &msg[strlen(msg)];
+
+	/* vlag tag and etherrtype */
+	ethertype = GET_ETHERTYPE(p);
+	if (ethertype == VTAG_TPID) {
+		sprintf(bp, " vtag:0x%04x", GET_NET16(&p[14]));
+		bp = &msg[strlen(msg)];
+
+		offset += VTAG_SIZE;
+		p = &buf[offset];
+		ethertype = GET_ETHERTYPE(p);
+	}
+	sprintf(bp, " type:%04x", ethertype);
+	bp = &msg[strlen(msg)];
+
+	/* ethernet packet length */
+	sprintf(bp, " mblklen:%d", msgdsize(mp));
+	bp = &msg[strlen(msg)];
+	if (mp->b_cont) {
+		sprintf(bp, "(");
+		bp = &msg[strlen(msg)];
+		for (tp = mp; tp; tp = tp->b_cont) {
+			if (tp == mp) {
+				sprintf(bp, "%d", tp->b_wptr - tp->b_rptr);
+			} else {
+				sprintf(bp, "+%d", tp->b_wptr - tp->b_rptr);
+			}
+			bp = &msg[strlen(msg)];
+		}
+		sprintf(bp, ")");
+		bp = &msg[strlen(msg)];
+	}
+
+	if (ethertype != ETHERTYPE_IP) {
+		goto x;
+	}
+
+	/* ip address */
+	offset += sizeof (struct ether_header);
+	p = &buf[offset];
+	ipproto = p[9];
+	iplen = GET_NET16(&p[2]);
+	sprintf(bp, ", ip: %d.%d.%d.%d -> %d.%d.%d.%d proto:%d iplen:%d",
+	    p[12], p[13], p[14], p[15],
+	    p[16], p[17], p[18], p[19],
+	    ipproto, iplen);
+	bp = (void *)&msg[strlen(msg)];
+
+	iphlen = (p[0] & 0xf) * 4;
+
+	/* cksum for psuedo header */
+	cksum = *(uint16_t *)&p[12];
+	cksum += *(uint16_t *)&p[14];
+	cksum += *(uint16_t *)&p[16];
+	cksum += *(uint16_t *)&p[18];
+	cksum += BE_16(ipproto);
+
+	/* tcp or udp protocol header */
+	offset += iphlen;
+	p = &buf[offset];
+	if (ipproto == IPPROTO_TCP) {
+		tcplen = iplen - iphlen;
+		sprintf(bp, ", tcp: len:%d cksum:%x",
+		    tcplen, GET_NET16(&p[16]));
+		bp = (void *)&msg[strlen(msg)];
+
+		if (check_cksum) {
+			cksum += BE_16(tcplen);
+			cksum = (uint16_t)ip_cksum(mp, offset, cksum);
+			sprintf(bp, " (%s)",
+			    (cksum == 0 || cksum == 0xffff) ? "ok" : "ng");
+			bp = (void *)&msg[strlen(msg)];
+		}
+	} else if (ipproto == IPPROTO_UDP) {
+		udplen = GET_NET16(&p[4]);
+		sprintf(bp, ", udp: len:%d cksum:%x",
+		    udplen, GET_NET16(&p[6]));
+		bp = (void *)&msg[strlen(msg)];
+
+		if (GET_NET16(&p[6]) && check_cksum) {
+			cksum += *(uint16_t *)&p[4];
+			cksum = (uint16_t)ip_cksum(mp, offset, cksum);
+			sprintf(bp, " (%s)",
+			    (cksum == 0 || cksum == 0xffff) ? "ok" : "ng");
+			bp = (void *)&msg[strlen(msg)];
+		}
+	}
+x:
+	cmn_err(CE_CONT, "!%s: %s: %s", dp->name, title, msg);
+}
+#endif /* USBGEM_DEBUG_VLAN */
+#endif /* USBGEM_DEBUG_LEVEL */
+
+#ifdef GEM_GCC_RUNTIME
+/*
+ * gcc3 runtime routines
+ */
+#pragma weak memcmp
+int
+memcmp(const void *s1, const void *s2, size_t n)
+{
+	int	i;
+	int	ret;
+
+	ret = 0;
+	for (i = 0; i < n; i++) {
+		ret = (int)((uint8_t *)s1)[i] - (int)((uint8_t *)s2)[i];
+		if (ret) {
+			return (ret);
+		}
+	}
+	return (0);
+}
+
+#pragma weak memset
+void *
+memset(void *s, int c, size_t n)
+{
+	if ((c & 0xff) == 0) {
+		bzero(s, n);
+	} else {
+		while (n--) {
+			((uint8_t *)s)[n] = c;
+		}
+	}
+	return (s);
+}
+
+#pragma weak _memcpy = memcpy
+#pragma weak memcpy
+void *
+memcpy(void *s1, const void *s2, size_t n)
+{
+	bcopy(s2, s1, n);
+	return (s1);
+}
+#endif /* GEM_GCC_RUNTIME */
+/* ============================================================== */
+/*
+ * hardware operations
+ */
+/* ============================================================== */
+static int
+usbgem_hal_reset_chip(struct usbgem_dev *dp)
+{
+	int	err;
+
+	sema_p(&dp->hal_op_lock);
+	err = (*dp->ugc.usbgc_reset_chip)(dp);
+	sema_v(&dp->hal_op_lock);
+	return (err);
+}
+
+static int
+usbgem_hal_init_chip(struct usbgem_dev *dp)
+{
+	int	err;
+
+	sema_p(&dp->hal_op_lock);
+	err = (*dp->ugc.usbgc_init_chip)(dp);
+	sema_v(&dp->hal_op_lock);
+	return (err);
+}
+
+static int
+usbgem_hal_attach_chip(struct usbgem_dev *dp)
+{
+	int	err;
+
+	sema_p(&dp->hal_op_lock);
+	err = (*dp->ugc.usbgc_attach_chip)(dp);
+	sema_v(&dp->hal_op_lock);
+	return (err);
+}
+
+static int
+usbgem_hal_set_rx_filter(struct usbgem_dev *dp)
+{
+	int	err;
+
+	sema_p(&dp->hal_op_lock);
+	err = (*dp->ugc.usbgc_set_rx_filter)(dp);
+	sema_v(&dp->hal_op_lock);
+	return (err);
+}
+
+static int
+usbgem_hal_set_media(struct usbgem_dev *dp)
+{
+	int	err;
+
+	sema_p(&dp->hal_op_lock);
+	err = (*dp->ugc.usbgc_set_media)(dp);
+	sema_v(&dp->hal_op_lock);
+	return (err);
+}
+
+static int
+usbgem_hal_start_chip(struct usbgem_dev *dp)
+{
+	int	err;
+
+	sema_p(&dp->hal_op_lock);
+	err = (*dp->ugc.usbgc_start_chip)(dp);
+	sema_v(&dp->hal_op_lock);
+	return (err);
+}
+
+static int
+usbgem_hal_stop_chip(struct usbgem_dev *dp)
+{
+	int	err;
+
+	sema_p(&dp->hal_op_lock);
+	err = (*dp->ugc.usbgc_stop_chip)(dp);
+	sema_v(&dp->hal_op_lock);
+	return (err);
+}
+
+static int
+usbgem_hal_get_stats(struct usbgem_dev *dp)
+{
+	int	err;
+
+	sema_p(&dp->hal_op_lock);
+	err = (*dp->ugc.usbgc_get_stats)(dp);
+	sema_v(&dp->hal_op_lock);
+	return (err);
+}
+
+
+/* ============================================================== */
+/*
+ * USB pipe management
+ */
+/* ============================================================== */
+static boolean_t
+usbgem_rx_start_unit(struct usbgem_dev *dp, usb_bulk_req_t *req)
+{
+	mblk_t	*mp;
+	int	err;
+	usb_flags_t	flags;
+
+	ASSERT(req);
+
+	mp = allocb(dp->rx_buf_len, BPRI_MED);
+	if (mp == NULL) {
+		cmn_err(CE_WARN, "!%s: %s: failed to allocate mblk",
+		    dp->name, __func__);
+		goto err;
+	}
+
+	req->bulk_len = dp->rx_buf_len;
+	req->bulk_data = mp;
+	req->bulk_client_private = (usb_opaque_t)dp;
+	req->bulk_timeout = 0;
+	req->bulk_attributes = USB_ATTRS_SHORT_XFER_OK;
+	req->bulk_cb = usbgem_bulkin_cb;
+	req->bulk_exc_cb = usbgem_bulkin_cb;
+	req->bulk_completion_reason = 0;
+	req->bulk_cb_flags = 0;
+
+	flags = 0;
+	err = usb_pipe_bulk_xfer(dp->bulkin_pipe, req, flags);
+
+	if (err != USB_SUCCESS) {
+		cmn_err(CE_WARN, "%s: failed to bulk_xfer for rx, err:%d",
+		    dp->name, err);
+
+		/* free req and mp */
+		usb_free_bulk_req(req);
+		goto err;
+	}
+	return (B_TRUE);
+err:
+	return (B_FALSE);
+}
+
+/* ============================================================== */
+/*
+ * Rx/Tx buffer management
+ */
+/* ============================================================== */
+static int
+usbgem_init_rx_buf(struct usbgem_dev *dp)
+{
+	int	i;
+	usb_bulk_req_t	*req;
+
+	ASSERT(dp->mac_state == MAC_STATE_ONLINE);
+
+	for (i = 0; i < dp->ugc.usbgc_rx_list_max; i++) {
+		req = usb_alloc_bulk_req(dp->dip, 0, USB_FLAGS_SLEEP);
+		if (req == NULL) {
+			cmn_err(CE_WARN,
+			    "!%s: %s: failed to allocate bulkreq for rx",
+			    dp->name, __func__);
+			return (USB_FAILURE);
+		}
+		if (!usbgem_rx_start_unit(dp, req)) {
+			return (USB_FAILURE);
+		}
+		mutex_enter(&dp->rxlock);
+		dp->rx_busy_cnt++;
+		mutex_exit(&dp->rxlock);
+	}
+	return (USB_SUCCESS);
+}
+
+/* ============================================================== */
+/*
+ * memory resource management
+ */
+/* ============================================================== */
+static int
+usbgem_free_memory(struct usbgem_dev *dp)
+{
+	usb_bulk_req_t	*req;
+
+	/* free all tx requst structure */
+	while ((req = dp->tx_free_list) != NULL) {
+		dp->tx_free_list =
+		    (usb_bulk_req_t *)req->bulk_client_private;
+		req->bulk_data = NULL;
+		usb_free_bulk_req(req);
+	}
+	return (USB_SUCCESS);
+}
+
+static int
+usbgem_alloc_memory(struct usbgem_dev *dp)
+{
+	int	i;
+	usb_bulk_req_t	*req;
+
+	/* allocate tx requests */
+	dp->tx_free_list = NULL;
+	for (i = 0; i < dp->ugc.usbgc_tx_list_max; i++) {
+		req = usb_alloc_bulk_req(dp->dip, 0, USB_FLAGS_SLEEP);
+		if (req == NULL) {
+			cmn_err(CE_WARN,
+			    "%s:%s failed to allocate tx requests",
+			    dp->name, __func__);
+
+			/* free partially allocated tx requests */
+			(void) usbgem_free_memory(dp);
+			return (USB_FAILURE);
+		}
+
+		/* add the new one allocated into tx free list */
+		req->bulk_client_private = (usb_opaque_t)dp->tx_free_list;
+		dp->tx_free_list = req;
+	}
+
+	return (USB_SUCCESS);
+}
+
+/* ========================================================== */
+/*
+ * Start transmission.
+ * Return zero on success,
+ */
+/* ========================================================== */
+
+#ifdef TXTIMEOUT_TEST
+static int usbgem_send_cnt = 0;
+#endif
+
+/*
+ * usbgem_send is used only to send data packet into ethernet line.
+ */
+static mblk_t *
+usbgem_send_common(struct usbgem_dev *dp, mblk_t *mp, uint32_t flags)
+{
+	int		err;
+	mblk_t		*new;
+	usb_bulk_req_t	*req;
+	int		mcast;
+	int		bcast;
+	int		len;
+	boolean_t	intr;
+	usb_flags_t	usb_flags = 0;
+#ifdef USBGEM_DEBUG_LEVEL
+	usb_pipe_state_t	p_state;
+#endif
+	DPRINTF(2, (CE_CONT, "!%s: %s: called", dp->name, __func__));
+
+	intr = (flags & 1) != 0;
+	len = msgdsize(mp);
+	bcast = 0;
+	mcast = 0;
+	if (mp->b_rptr[0] & 1) {
+		if (bcmp(mp->b_rptr, &usbgem_bcastaddr, ETHERADDRL) == 0) {
+			bcast = 1;
+		} else {
+			mcast = 1;
+		}
+	}
+	new = (*dp->ugc.usbgc_tx_make_packet)(dp, mp);
+	if (new == NULL) {
+		/*
+		 * no memory resource. we don't stop downstream,
+		 * we just discard the packet.
+		 */
+		DPRINTF(0, (CE_CONT, "!%s: %s: no memory",
+		    dp->name, __func__));
+		freemsg(mp);
+
+		mutex_enter(&dp->txlock);
+		dp->stats.noxmtbuf++;
+		dp->stats.errxmt++;
+		mutex_exit(&dp->txlock);
+
+		return (NULL);
+	}
+
+	ASSERT(new->b_cont == NULL);
+
+	mutex_enter(&dp->txlock);
+	if (dp->tx_free_list == NULL) {
+		/*
+		 * no tx free slot
+		 */
+		ASSERT(dp->tx_busy_cnt == dp->ugc.usbgc_tx_list_max);
+		mutex_exit(&dp->txlock);
+
+		DPRINTF(4, (CE_CONT, "!%s: %s: no free slot",
+		    dp->name, __func__));
+		if (new && new != mp) {
+			/* free reallocated message */
+			freemsg(new);
+		}
+		return (mp);
+	}
+	req = dp->tx_free_list;
+	dp->tx_free_list = (usb_bulk_req_t *)req->bulk_client_private;
+	dp->tx_busy_cnt++;
+	
+	if (dp->tx_free_list == NULL) {
+		intr = B_TRUE;
+	}
+	if (intr) {
+		dp->tx_intr_pended++;
+	}
+	DB_TCI(new) = intr;
+#ifdef USBGEM_DEBUG_LEVEL
+	new->b_datap->db_cksum32 = dp->tx_seq_num;
+	dp->tx_seq_num++;
+#endif
+	dp->stats.obytes += len;
+	dp->stats.opackets++;
+	if (bcast | mcast) {
+		dp->stats.obcast += bcast;
+		dp->stats.omcast += mcast;
+	}
+	mutex_exit(&dp->txlock);
+
+	DPRINTF(2, (CE_CONT, "!%s: %s: sending", dp->name, __func__));
+
+	req->bulk_len = (long)new->b_wptr - (long)new->b_rptr;
+	req->bulk_data = new;
+	req->bulk_client_private = (usb_opaque_t)dp;
+	req->bulk_timeout = dp->bulkout_timeout;	/* in second */
+	req->bulk_attributes = 0;
+	req->bulk_cb = usbgem_bulkout_cb;
+	req->bulk_exc_cb = usbgem_bulkout_cb;
+	req->bulk_completion_reason = 0;
+	req->bulk_cb_flags = 0;
+
+	if (intr) {
+		usb_flags = USB_FLAGS_SLEEP;
+	}
+	if ((err = usb_pipe_bulk_xfer(dp->bulkout_pipe, req, usb_flags))
+	    != USB_SUCCESS) {
+
+		/* failed to transfer the packet, discard it. */
+		freemsg(new);
+		req->bulk_data = NULL;
+
+		/* recycle the request block */
+		mutex_enter(&dp->txlock);
+		dp->tx_busy_cnt--;
+		req->bulk_client_private = (usb_opaque_t)dp->tx_free_list;
+		dp->tx_free_list = req;
+		mutex_exit(&dp->txlock);
+
+		cmn_err(CE_NOTE,
+		    "%s: %s: usb_pipe_bulk_xfer: failed: err:%d",
+		    dp->name, __func__, err);
+
+		/* we use another flag to indicate error state. */
+		if (dp->fatal_error == (clock_t)0) {
+			dp->fatal_error = usbgem_timestamp_nz();
+		}
+	} else {
+		/* record the start time */
+		dp->tx_start_time = ddi_get_lbolt();
+	}
+
+	if (err == USB_SUCCESS && (usb_flags & USB_FLAGS_SLEEP)) {
+		usbgem_bulkout_cb(dp->bulkout_pipe, req);
+	}
+
+	if (new != mp) {
+		freemsg(mp);
+	}
+	return (NULL);
+}
+
+int
+usbgem_restart_nic(struct usbgem_dev *dp)
+{
+	int	ret;
+	int	flags = 0;
+
+	DPRINTF(1, (CE_CONT, "!%s: %s: called", dp->name, __func__));
+
+	ASSERT(dp->mac_state != MAC_STATE_DISCONNECTED);
+
+	/*
+	 * ensure to stop the nic
+	 */
+	if (dp->mac_state == MAC_STATE_ONLINE) {
+		(void) usbgem_mac_stop(dp, MAC_STATE_STOPPED, STOP_GRACEFUL);
+	}
+
+	/* now the nic become quiescent, reset the chip */
+	if (usbgem_hal_reset_chip(dp) != USB_SUCCESS) {
+		cmn_err(CE_WARN, "%s: %s: failed to reset chip",
+		    dp->name, __func__);
+		goto err;
+	}
+
+	/*
+	 * restore the nic state step by step
+	 */
+	if (dp->nic_state < NIC_STATE_INITIALIZED) {
+		goto done;
+	}
+
+	if (usbgem_mac_init(dp) != USB_SUCCESS) {
+		cmn_err(CE_WARN, "%s: %s: failed to initialize chip",
+		    dp->name, __func__);
+		goto err;
+	}
+
+	/* setup mac address and enable rx filter */
+	sema_p(&dp->rxfilter_lock);
+	dp->rxmode |= RXMODE_ENABLE;
+	ret = usbgem_hal_set_rx_filter(dp);
+	sema_v(&dp->rxfilter_lock);
+	if (ret != USB_SUCCESS) {
+		goto err;
+	}
+
+	/*
+	 * update the link state asynchronously
+	 */
+	cv_signal(&dp->link_watcher_wait_cv);
+
+	/*
+	 * XXX - a panic happened because of linkdown.
+	 * We must check mii_state here, because the link can be down just
+	 * before the restart event happen. If the link is down now,
+	 * gem_mac_start() will be called from gem_mii_link_check() when
+	 * the link become up later.
+	 */
+	if (dp->mii_state == MII_STATE_LINKUP) {
+		if (usbgem_hal_set_media(dp) != USB_SUCCESS) {
+			goto err;
+		}
+		if (dp->nic_state < NIC_STATE_ONLINE) {
+			goto done;
+		}
+
+		(void) usbgem_mac_start(dp);
+
+	}
+done:
+	return (USB_SUCCESS);
+err:
+#ifdef GEM_CONFIG_FMA
+	ddi_fm_service_impact(dp->dip, DDI_SERVICE_DEGRADED);
+#endif
+	return (USB_FAILURE);
+}
+
+static void
+usbgem_tx_timeout(struct usbgem_dev *dp)
+{
+	int	ret;
+	uint_t	rwlock;
+	clock_t	now;
+
+	for (; ; ) {
+		mutex_enter(&dp->tx_watcher_lock);
+		ret = cv_timedwait(&dp->tx_watcher_cv, &dp->tx_watcher_lock,
+		    dp->tx_watcher_interval + ddi_get_lbolt());
+		mutex_exit(&dp->tx_watcher_lock);
+
+		if (dp->tx_watcher_stop) {
+			break;
+		}
+
+		now = ddi_get_lbolt();
+
+		rwlock = RW_READER;
+again:
+		rw_enter(&dp->dev_state_lock, rwlock);
+
+		if ((dp->mac_state != MAC_STATE_DISCONNECTED &&
+		    dp->fatal_error &&
+		    now - dp->fatal_error >= dp->ugc.usbgc_tx_timeout) ||
+		    (dp->mac_state == MAC_STATE_ONLINE &&
+		    dp->mii_state == MII_STATE_LINKUP &&
+		    dp->tx_busy_cnt != 0 &&
+		    now - dp->tx_start_time >= dp->ugc.usbgc_tx_timeout)) {
+			if (rwlock == RW_READER) {
+				/*
+				 * Upgrade dev_state_lock from shared mode
+				 * to exclusive mode to restart nic
+				 */
+				rwlock = RW_WRITER;
+				rw_exit(&dp->dev_state_lock);
+				goto again;
+			}
+			cmn_err(CE_WARN, "%s: %s: restarting the nic:"
+			    " fatal_error:%ld nic_state:%d"
+			    " mac_state:%d starttime:%ld",
+			    dp->name, __func__,
+			    dp->fatal_error ? now - dp->fatal_error: 0,
+			    dp->nic_state, dp->mac_state,
+			    dp->tx_busy_cnt ? now - dp->tx_start_time : 0);
+
+			(void) usbgem_restart_nic(dp);
+		}
+
+		rw_exit(&dp->dev_state_lock);
+	}
+}
+
+static int
+usbgem_tx_watcher_start(struct usbgem_dev *dp)
+{
+	int	err;
+	kthread_t	*wdth;
+
+	DPRINTF(0, (CE_CONT, "!%s: %s: called", dp->name, __func__));
+
+	/* make a first call of uwgem_lw_link_check() */
+	dp->tx_watcher_stop = 0;
+	dp->tx_watcher_interval = drv_usectohz(1000*1000);
+
+	wdth = thread_create(NULL, 0, usbgem_tx_timeout, dp, 0, &p0,
+	    TS_RUN, minclsyspri);
+	if (wdth == NULL) {
+		cmn_err(CE_WARN,
+		    "!%s: %s: failed to create a tx_watcher thread",
+		    dp->name, __func__);
+		return (USB_FAILURE);
+	}
+	dp->tx_watcher_did = wdth->t_did;
+
+	return (USB_SUCCESS);
+}
+
+static void
+usbgem_tx_watcher_stop(struct usbgem_dev *dp)
+{
+	DPRINTF(0, (CE_CONT, "!%s: %s: called", dp->name, __func__));
+	if (dp->tx_watcher_did) {
+		/* Ensure timer routine stopped */
+		dp->tx_watcher_stop = 1;
+		cv_signal(&dp->tx_watcher_cv);
+		thread_join(dp->tx_watcher_did);
+		dp->tx_watcher_did = NULL;
+	}
+}
+
+/* ================================================================== */
+/*
+ * Callback handlers
+ */
+/* ================================================================== */
+static void
+usbgem_bulkin_cb(usb_pipe_handle_t pipe, usb_bulk_req_t *req)
+{
+	mblk_t	*newmp;
+	mblk_t	*mp;
+	mblk_t	*tp;
+	int	len = 0;
+	int	pkts = 0;
+	int	bcast = 0;
+	int	mcast = 0;
+	boolean_t	busy;
+	struct usbgem_dev	*dp;
+
+	dp = (struct usbgem_dev *)req->bulk_client_private;
+	mp = req->bulk_data;
+	req->bulk_data = NULL;
+
+	DPRINTF(2, (CE_CONT, "!%s: %s: mp:%p, cr:%s(%d)",
+	    dp->name, __func__, mp,
+	    usb_str_cr(req->bulk_completion_reason),
+	    req->bulk_completion_reason));
+
+	/*
+	 * we cannot acquire dev_state_lock because the routine
+	 * must be executed during usbgem_mac_stop() to avoid
+	 * dead lock.
+	 * we use a simle membar operation to get the state correctly.
+	 */
+	membar_consumer();
+
+	if (req->bulk_completion_reason == USB_CR_OK &&
+	    dp->nic_state == NIC_STATE_ONLINE) {
+		newmp = (*dp->ugc.usbgc_rx_make_packet)(dp, mp);
+
+		if (newmp != mp) {
+			/* the message has been reallocated, free old one */
+			freemsg(mp);
+		}
+
+		/* the message may includes one or more ethernet packets */
+		for (tp = newmp; tp; tp = tp->b_next) {
+			len += tp->b_wptr - tp->b_rptr;
+			pkts++;
+			if (tp->b_rptr[0] & 1) {
+				if (bcmp(tp->b_rptr, &usbgem_bcastaddr,
+				    ETHERADDRL) == 0) {
+					bcast++;
+				} else {
+					mcast++;
+				}
+			}
+		}
+
+		/* send up if it is a valid packet */
+#ifdef USBGEM_CONFIG_GLDv3
+		mac_rx(dp->mh, NULL, newmp);
+#else
+		while (newmp) {
+			tp = newmp;
+			newmp = newmp->b_next;
+			tp->b_next = NULL;
+			gld_recv(dp->macinfo, tp);
+		}
+#endif
+	} else {
+		freemsg(mp);
+		len = 0;
+	}
+
+	mutex_enter(&dp->rxlock);
+	/* update rx_active */
+	if (dp->rx_active) {
+		dp->rx_active = dp->mac_state == MAC_STATE_ONLINE;
+	}
+
+	dp->stats.rbytes += len;
+	dp->stats.rpackets += pkts;
+	if (bcast | mcast) {
+		dp->stats.rbcast += bcast;
+		dp->stats.rmcast += mcast;
+	}
+	mutex_exit(&dp->rxlock);
+
+	if (dp->rx_active) {
+		/* prepare to receive the next packets */
+		if (usbgem_rx_start_unit(dp, req)) {
+			/* we successed */
+			goto done;
+		}
+		cmn_err(CE_WARN,
+		    "!%s: %s: failed to fill next rx packet",
+		    dp->name, __func__);
+		/*
+		 * we use another flag to indicate error state.
+		 * if we acquire dev_state_lock for RW_WRITER here,
+		 * usbgem_mac_stop() may hang.
+		 */
+		if (dp->fatal_error == (clock_t)0) {
+			dp->fatal_error = usbgem_timestamp_nz();
+		}
+	} else {
+		/* no need to prepare the next packets */
+		usb_free_bulk_req(req);
+	}
+
+	mutex_enter(&dp->rxlock);
+	dp->rx_active = B_FALSE;
+	dp->rx_busy_cnt--;
+	if (dp->rx_busy_cnt == 0) {
+		/* wake up someone waits for me */
+		cv_broadcast(&dp->rx_drain_cv);
+	}
+	mutex_exit(&dp->rxlock);
+done:
+	;
+}
+
+static void
+usbgem_bulkout_cb(usb_pipe_handle_t pipe, usb_bulk_req_t *req)
+{
+	boolean_t	intr;
+	boolean_t	tx_sched;
+	struct usbgem_dev	*dp;
+
+	dp = (struct usbgem_dev *)req->bulk_client_private;
+	tx_sched = B_FALSE;
+
+	DPRINTF(2, (CE_CONT,
+	    "!%s: %s: cr:%s(%d) cb_flags:0x%x head:%d tail:%d",
+	    dp->name, __func__,
+	    usb_str_cr(req->bulk_completion_reason),
+	    req->bulk_completion_reason,
+	    req->bulk_cb_flags,
+	    dp->tx_busy_cnt));
+
+	/* we have finished to transfer the packet into tx fifo */
+	intr = DB_TCI(req->bulk_data);
+	freemsg(req->bulk_data);
+
+	if (req->bulk_completion_reason != USB_CR_OK &&
+	    dp->fatal_error == (clock_t)0) {
+		dp->fatal_error = usbgem_timestamp_nz();
+	}
+
+	mutex_enter(&dp->txlock);
+
+	if (intr) {
+		ASSERT(dp->tx_intr_pended > 0);
+		/* find the last interrupt we have scheduled */
+		if (--(dp->tx_intr_pended) == 0) {
+			tx_sched = B_TRUE;
+		}
+	}
+
+	ASSERT(dp->tx_busy_cnt > 0);
+	req->bulk_client_private = (usb_opaque_t)dp->tx_free_list;
+	dp->tx_free_list = req;
+	dp->tx_busy_cnt--;
+
+#ifdef CONFIG_TX_LIMITER
+	if (tx_sched) {
+		dp->tx_max_packets =
+		    min(dp->tx_max_packets + 1, dp->ugc.usbgc_tx_list_max);
+	}
+#endif
+	if (dp->mac_state != MAC_STATE_ONLINE && dp->tx_busy_cnt == 0) {
+		cv_broadcast(&dp->tx_drain_cv);
+	}
+
+	mutex_exit(&dp->txlock);
+
+	if (tx_sched) {
+#ifdef USBGEM_CONFIG_GLDv3
+		mac_tx_update(dp->mh);
+#else
+		gld_sched(dp->macinfo);
+#endif
+	}
+}
+
+static void
+usbgem_intr_cb(usb_pipe_handle_t ph, usb_intr_req_t *req)
+{
+	struct usbgem_dev	*dp;
+
+	dp = (struct usbgem_dev *)req->intr_client_private;
+	dp->stats.intr++;
+
+	if (req->intr_completion_reason == USB_CR_OK) {
+		(*dp->ugc.usbgc_interrupt)(dp, req->intr_data);
+	}
+
+	/* free the request and data */
+	usb_free_intr_req(req);
+}
+
+/* ======================================================================== */
+/*
+ * MII support routines
+ */
+/* ======================================================================== */
+static void
+usbgem_choose_forcedmode(struct usbgem_dev *dp)
+{
+	/* choose media mode */
+	if (dp->anadv_1000fdx || dp->anadv_1000hdx) {
+		dp->speed = USBGEM_SPD_1000;
+		dp->full_duplex = dp->anadv_1000fdx;
+	} else if (dp->anadv_100fdx || dp->anadv_100t4) {
+		dp->speed = USBGEM_SPD_100;
+		dp->full_duplex = B_TRUE;
+	} else if (dp->anadv_100hdx) {
+		dp->speed = USBGEM_SPD_100;
+		dp->full_duplex = B_FALSE;
+	} else {
+		dp->speed = USBGEM_SPD_10;
+		dp->full_duplex = dp->anadv_10fdx;
+	}
+}
+
+static uint16_t
+usbgem_mii_read(struct usbgem_dev *dp, uint_t reg, int *errp)
+{
+	uint16_t	val;
+
+	sema_p(&dp->hal_op_lock);
+	val = (*dp->ugc.usbgc_mii_read)(dp, reg, errp);
+	sema_v(&dp->hal_op_lock);
+
+	return (val);
+}
+
+static void
+usbgem_mii_write(struct usbgem_dev *dp, uint_t reg, uint16_t val, int *errp)
+{
+	sema_p(&dp->hal_op_lock);
+	(*dp->ugc.usbgc_mii_write)(dp, reg, val, errp);
+	sema_v(&dp->hal_op_lock);
+}
+
+static int
+usbgem_mii_probe(struct usbgem_dev *dp)
+{
+	int	err;
+
+	err = (*dp->ugc.usbgc_mii_probe)(dp);
+	return (err);
+}
+
+static int
+usbgem_mii_init(struct usbgem_dev *dp)
+{
+	int	err;
+
+	err = (*dp->ugc.usbgc_mii_init)(dp);
+	return (err);
+}
+
+#define	fc_cap_decode(x)	\
+	((((x) & MII_ABILITY_PAUSE) != 0 ? 1 : 0) |	\
+	(((x) & MII_ABILITY_ASM_DIR) != 0 ? 2 : 0))
+
+int
+usbgem_mii_config_default(struct usbgem_dev *dp, int *errp)
+{
+	uint16_t	mii_stat;
+	uint16_t	val;
+
+	DPRINTF(1, (CE_CONT, "!%s: %s: called", dp->name, __func__));
+
+	/*
+	 * Configure bits in advertisement register
+	 */
+	mii_stat = dp->mii_status;
+
+	DPRINTF(1, (CE_CONT, "!%s: %s: MII_STATUS reg:%b",
+	    dp->name, __func__, mii_stat, MII_STATUS_BITS));
+
+	if ((mii_stat & MII_STATUS_ABILITY_TECH) == 0) {
+		/* it's funny */
+		cmn_err(CE_WARN, "!%s: wrong ability bits: mii_status:%b",
+		    dp->name, mii_stat, MII_STATUS_BITS);
+		return (USB_FAILURE);
+	}
+
+	/* Do not change the rest of ability bits in advert reg */
+	val = usbgem_mii_read(dp, MII_AN_ADVERT, errp) & ~MII_ABILITY_ALL;
+	if (*errp != USB_SUCCESS) {
+		goto usberr;
+	}
+
+	DPRINTF(0, (CE_CONT,
+	    "!%s: %s: 100T4:%d 100F:%d 100H:%d 10F:%d 10H:%d",
+	    dp->name, __func__,
+	    dp->anadv_100t4, dp->anadv_100fdx, dp->anadv_100hdx,
+	    dp->anadv_10fdx, dp->anadv_10hdx));
+
+	/* set technology bits */
+	if (dp->anadv_100t4) {
+		val |= MII_ABILITY_100BASE_T4;
+	}
+	if (dp->anadv_100fdx) {
+		val |= MII_ABILITY_100BASE_TX_FD;
+	}
+	if (dp->anadv_100hdx) {
+		val |= MII_ABILITY_100BASE_TX;
+	}
+	if (dp->anadv_10fdx) {
+		val |= MII_ABILITY_10BASE_T_FD;
+	}
+	if (dp->anadv_10hdx) {
+		val |= MII_ABILITY_10BASE_T;
+	}
+
+	/* set flow control capabilities */
+	if (dp->anadv_pause) {
+		val |= MII_ABILITY_PAUSE;
+	}
+	if (dp->anadv_asmpause) {
+		val |= MII_ABILITY_ASM_DIR;
+	}
+
+	DPRINTF(0, (CE_CONT,
+	    "!%s: %s: setting MII_AN_ADVERT reg:%b, pause:%d, asmpause:%d",
+	    dp->name, __func__, val, MII_ABILITY_BITS,
+	    dp->anadv_pause, dp->anadv_asmpause));
+
+	usbgem_mii_write(dp, MII_AN_ADVERT, val, errp);
+	if (*errp != USB_SUCCESS) {
+		goto usberr;
+	}
+
+	if (dp->mii_status & MII_STATUS_XSTATUS) {
+		/*
+		 * 1000Base-T GMII support
+		 */
+		if (!dp->anadv_autoneg) {
+			/* enable manual configuration */
+			val = MII_1000TC_CFG_EN;
+			if (dp->anadv_1000t_ms == 2) {
+				val |= MII_1000TC_CFG_VAL;
+			}
+		} else {
+			val = 0;
+			if (dp->anadv_1000fdx) {
+				val |= MII_1000TC_ADV_FULL;
+			}
+			if (dp->anadv_1000hdx) {
+				val |= MII_1000TC_ADV_HALF;
+			}
+			switch (dp->anadv_1000t_ms) {
+			case 1:
+				/* slave */
+				val |= MII_1000TC_CFG_EN;
+				break;
+
+			case 2:
+				/* master */
+				val |= MII_1000TC_CFG_EN | MII_1000TC_CFG_VAL;
+				break;
+
+			default:
+				/* auto: do nothing */
+				break;
+			}
+		}
+		DPRINTF(0, (CE_CONT,
+		    "!%s: %s: setting MII_1000TC reg:%b",
+		    dp->name, __func__, val, MII_1000TC_BITS));
+
+		usbgem_mii_write(dp, MII_1000TC, val, errp);
+		if (*errp != USB_SUCCESS) {
+			goto usberr;
+		}
+	}
+	return (USB_SUCCESS);
+
+usberr:
+	return (*errp);
+}
+
+static char *usbgem_fc_type[] = {
+	"without",
+	"with symmetric",
+	"with tx",
+	"with rx",
+};
+
+#ifdef USBGEM_CONFIG_GLDv3
+#define	USBGEM_LINKUP(dp)	mac_link_update((dp)->mh, LINK_STATE_UP)
+#define	USBGEM_LINKDOWN(dp)	mac_link_update((dp)->mh, LINK_STATE_DOWN)
+#else
+#define	USBGEM_LINKUP(dp)	\
+	if (gld_linkstate) {	\
+		gld_linkstate((dp)->macinfo, GLD_LINKSTATE_UP);	\
+	}
+#define	USBGEM_LINKDOWN(dp)	\
+	if (gld_linkstate) {	\
+		gld_linkstate((dp)->macinfo, GLD_LINKSTATE_DOWN);	\
+	}
+#endif
+
+static uint8_t usbgem_fc_result[4 /* my cap */][4 /* lp cap */] = {
+/*	 none	symm	tx	rx/symm */
+/* none */
+	{FLOW_CONTROL_NONE,
+		FLOW_CONTROL_NONE,
+			FLOW_CONTROL_NONE,
+				FLOW_CONTROL_NONE},
+/* sym */
+	{FLOW_CONTROL_NONE,
+		FLOW_CONTROL_SYMMETRIC,
+			FLOW_CONTROL_NONE,
+				FLOW_CONTROL_SYMMETRIC},
+/* tx */
+	{FLOW_CONTROL_NONE,
+		FLOW_CONTROL_NONE,
+			FLOW_CONTROL_NONE,
+				FLOW_CONTROL_TX_PAUSE},
+/* rx/symm */
+	{FLOW_CONTROL_NONE,
+		FLOW_CONTROL_SYMMETRIC,
+			FLOW_CONTROL_RX_PAUSE,
+				FLOW_CONTROL_SYMMETRIC},
+};
+
+static boolean_t
+usbgem_mii_link_check(struct usbgem_dev *dp, int *oldstatep, int *newstatep)
+{
+	boolean_t	tx_sched = B_FALSE;
+	uint16_t	status;
+	uint16_t	advert;
+	uint16_t	lpable;
+	uint16_t	exp;
+	uint16_t	ctl1000;
+	uint16_t	stat1000;
+	uint16_t	val;
+	clock_t		now;
+	clock_t		diff;
+	int		linkdown_action;
+	boolean_t	fix_phy = B_FALSE;
+	int		err;
+	uint_t		rwlock;
+
+	DPRINTF(4, (CE_CONT, "!%s: %s: time:%d state:%d",
+	    dp->name, __func__, ddi_get_lbolt(), dp->mii_state));
+
+	if (dp->mii_state != MII_STATE_LINKUP) {
+		rwlock = RW_WRITER;
+	} else {
+		rwlock = RW_READER;
+	}
+again:
+	rw_enter(&dp->dev_state_lock, rwlock);
+
+	/* save old mii state */
+	*oldstatep = dp->mii_state;
+
+	if (dp->mac_state == MAC_STATE_DISCONNECTED) {
+		/* stop periodic execution of the link watcher */
+		dp->mii_interval = 0;
+		tx_sched = B_FALSE;
+		goto next;
+	}
+
+	now = ddi_get_lbolt();
+	diff = now - dp->mii_last_check;
+	dp->mii_last_check = now;
+
+	/*
+	 * For NWAM, don't show linkdown state right
+	 * when the device is attached.
+	 */
+	if (dp->linkup_delay > 0) {
+		if (dp->linkup_delay > diff) {
+			dp->linkup_delay -= diff;
+		} else {
+			/* link up timeout */
+			dp->linkup_delay = -1;
+		}
+	}
+
+next_nowait:
+	switch (dp->mii_state) {
+	case MII_STATE_UNKNOWN:
+		goto reset_phy;
+
+	case MII_STATE_RESETTING:
+		dp->mii_timer -= diff;
+		if (dp->mii_timer > 0) {
+			/* don't read phy registers in resetting */
+			dp->mii_interval = WATCH_INTERVAL_FAST;
+			goto next;
+		}
+
+		val = usbgem_mii_read(dp, MII_CONTROL, &err);
+		if (err != USB_SUCCESS) {
+			goto usberr;
+		}
+		if (val & MII_CONTROL_RESET) {
+			cmn_err(CE_NOTE,
+			    "!%s: time:%ld resetting phy not complete."
+			    " mii_control:0x%b",
+			    dp->name, ddi_get_lbolt(),
+			    val, MII_CONTROL_BITS);
+		}
+
+		/* ensure neither isolated nor pwrdown nor auto-nego mode */
+		usbgem_mii_write(dp, MII_CONTROL, 0, &err);
+		if (err != USB_SUCCESS) {
+			goto usberr;
+		}
+#if USBGEM_DEBUG_LEVEL > 10
+		val = usbgem_mii_read(dp, MII_CONTROL, &err);
+		cmn_err(CE_CONT, "!%s: readback control %b",
+		    dp->name, val, MII_CONTROL_BITS);
+#endif
+		/* As resetting PHY has completed, configure PHY registers */
+		if ((*dp->ugc.usbgc_mii_config)(dp, &err) != USB_SUCCESS) {
+			/* we failed to configure PHY */
+			goto usberr;
+		}
+
+		/* prepare for forced mode */
+		usbgem_choose_forcedmode(dp);
+
+		dp->mii_lpable = 0;
+		dp->mii_advert = 0;
+		dp->mii_exp = 0;
+		dp->mii_ctl1000 = 0;
+		dp->mii_stat1000 = 0;
+
+		dp->flow_control = FLOW_CONTROL_NONE;
+
+		if (!dp->anadv_autoneg) {
+			/* skip auto-negotiation phase */
+			dp->mii_state = MII_STATE_MEDIA_SETUP;
+			dp->mii_timer = dp->ugc.usbgc_mii_linkdown_timeout;
+			goto next_nowait;
+		}
+
+		/* issue an auto-negotiation command */
+		goto autonego;
+
+	case MII_STATE_AUTONEGOTIATING:
+		/*
+		 * Autonegotiation in progress
+		 */
+		dp->mii_timer -= diff;
+		if (dp->mii_timer -
+		    (dp->ugc.usbgc_mii_an_timeout - dp->ugc.usbgc_mii_an_wait)
+		    > 0) {
+			/* wait for minimum time (2.3 - 2.5 sec) */
+			dp->mii_interval = WATCH_INTERVAL_FAST;
+			goto next;
+		}
+
+		/* read PHY status */
+		status = usbgem_mii_read(dp, MII_STATUS, &err);
+		if (err != USB_SUCCESS) {
+			goto usberr;
+		}
+		DPRINTF(4, (CE_CONT,
+		    "!%s: %s: called: mii_state:%d MII_STATUS reg:%b",
+		    dp->name, __func__, dp->mii_state,
+		    status, MII_STATUS_BITS));
+
+		if (status & MII_STATUS_REMFAULT) {
+			/*
+			 * The link parnert told me something wrong happend.
+			 * What do we do ?
+			 */
+			cmn_err(CE_CONT,
+			    "!%s: auto-negotiation failed: remote fault",
+			    dp->name);
+			goto autonego;
+		}
+
+		if ((status & MII_STATUS_ANDONE) == 0) {
+			if (dp->mii_timer <= 0) {
+				/*
+				 * Auto-negotiation has been timed out,
+				 * Reset PHY and try again.
+				 */
+				if (!dp->mii_supress_msg) {
+					cmn_err(CE_WARN,
+					    "!%s: auto-negotiation failed:"
+					    " timeout",
+					    dp->name);
+					dp->mii_supress_msg = B_TRUE;
+				}
+				goto autonego;
+			}
+			/*
+			 * Auto-negotiation is in progress. Wait for a while.
+			 */
+			dp->mii_interval = dp->ugc.usbgc_mii_an_watch_interval;
+			goto next;
+		}
+
+		/*
+		 * Auto-negotiation has been completed. Let's go to AN_DONE.
+		 */
+		dp->mii_state = MII_STATE_AN_DONE;
+		dp->mii_supress_msg = B_FALSE;
+		DPRINTF(0, (CE_CONT,
+		    "!%s: auto-negotiation completed, MII_STATUS:%b",
+		    dp->name, status, MII_STATUS_BITS));
+
+		if (dp->ugc.usbgc_mii_an_delay > 0) {
+			dp->mii_timer = dp->ugc.usbgc_mii_an_delay;
+			dp->mii_interval = drv_usectohz(20*1000);
+			goto next;
+		}
+
+		dp->mii_timer = 0;
+		diff = 0;
+		goto next_nowait;
+
+	case MII_STATE_AN_DONE:
+		/*
+		 * Auto-negotiation has done. Now we can set up media.
+		 */
+		dp->mii_timer -= diff;
+		if (dp->mii_timer > 0) {
+			/* wait for a while */
+			dp->mii_interval = WATCH_INTERVAL_FAST;
+			goto next;
+		}
+
+		/*
+		 * Setup speed and duplex mode according with
+		 * the result of auto negotiation.
+		 */
+
+		/*
+		 * Read registers required to determin current
+		 * duplex mode and media speed.
+		 */
+		if (dp->ugc.usbgc_mii_an_delay > 0) {
+			/* the 'status' variable is not initialized yet */
+			status = usbgem_mii_read(dp, MII_STATUS, &err);
+			if (err != USB_SUCCESS) {
+				goto usberr;
+			}
+		}
+		advert = usbgem_mii_read(dp, MII_AN_ADVERT, &err);
+		if (err != USB_SUCCESS) {
+			goto usberr;
+		}
+		lpable = usbgem_mii_read(dp, MII_AN_LPABLE, &err);
+		if (err != USB_SUCCESS) {
+			goto usberr;
+		}
+		exp = usbgem_mii_read(dp, MII_AN_EXPANSION, &err);
+		if (err != USB_SUCCESS) {
+			goto usberr;
+		}
+		if (exp == 0xffff) {
+			/* some phys don't have exp register */
+			exp = 0;
+		}
+
+		ctl1000 = 0;
+		stat1000 = 0;
+		if (dp->mii_status & MII_STATUS_XSTATUS) {
+			ctl1000 = usbgem_mii_read(dp, MII_1000TC, &err);
+			if (err != USB_SUCCESS) {
+				goto usberr;
+			}
+			stat1000 = usbgem_mii_read(dp, MII_1000TS, &err);
+			if (err != USB_SUCCESS) {
+				goto usberr;
+			}
+		}
+		dp->mii_lpable = lpable;
+		dp->mii_advert = advert;
+		dp->mii_exp = exp;
+		dp->mii_ctl1000 = ctl1000;
+		dp->mii_stat1000 = stat1000;
+
+		cmn_err(CE_CONT,
+		    "!%s: auto-negotiation done: "
+		    "status:%b, advert:%b, lpable:%b, exp:%b",
+		    dp->name,
+		    status, MII_STATUS_BITS,
+		    advert, MII_ABILITY_BITS,
+		    lpable, MII_ABILITY_BITS,
+		    exp, MII_AN_EXP_BITS);
+
+		DPRINTF(0, (CE_CONT, "!%s: MII_STATUS:%b",
+		    dp->name, status, MII_STATUS_BITS));
+
+		if (dp->mii_status & MII_STATUS_XSTATUS) {
+			cmn_err(CE_CONT,
+			    "! MII_1000TC reg:%b, MII_1000TS reg:%b",
+			    ctl1000, MII_1000TC_BITS,
+			    stat1000, MII_1000TS_BITS);
+		}
+
+		if (usbgem_population(lpable) <= 1 &&
+		    (exp & MII_AN_EXP_LPCANAN) == 0) {
+			if ((advert & MII_ABILITY_TECH) != lpable) {
+				cmn_err(CE_WARN,
+				    "!%s: but the link partner doesn't seem"
+				    " to have auto-negotiation capability."
+				    " please check the link configuration.",
+				    dp->name);
+			}
+			/*
+			 * it should be a result of pararell detection,
+			 * which cannot detect duplex mode.
+			 */
+			if ((advert & lpable) == 0 &&
+			    lpable & MII_ABILITY_10BASE_T) {
+				/* no common technology, try 10M half mode */
+				lpable |= advert & MII_ABILITY_10BASE_T;
+				fix_phy = B_TRUE;
+			}
+		} else if (lpable == 0) {
+			cmn_err(CE_WARN, "!%s: wrong lpable.", dp->name);
+			goto reset_phy;
+		}
+		/*
+		 * configure current link mode according to AN priority.
+		 */
+		val = advert & lpable;
+		if ((ctl1000 & MII_1000TC_ADV_FULL) &&
+		    (stat1000 & MII_1000TS_LP_FULL)) {
+			/* 1000BaseT & full duplex */
+			dp->speed = USBGEM_SPD_1000;
+			dp->full_duplex = B_TRUE;
+		} else if ((ctl1000 & MII_1000TC_ADV_HALF) &&
+		    (stat1000 & MII_1000TS_LP_HALF)) {
+			/* 1000BaseT & half duplex */
+			dp->speed = USBGEM_SPD_1000;
+			dp->full_duplex = B_FALSE;
+		} else if ((val & MII_ABILITY_100BASE_TX_FD)) {
+			/* 100BaseTx & fullduplex */
+			dp->speed = USBGEM_SPD_100;
+			dp->full_duplex = B_TRUE;
+		} else if ((val & MII_ABILITY_100BASE_T4)) {
+			/* 100BaseTx & fullduplex */
+			dp->speed = USBGEM_SPD_100;
+			dp->full_duplex = B_TRUE;
+		} else if ((val & MII_ABILITY_100BASE_TX)) {
+			/* 100BaseTx & half duplex */
+			dp->speed = USBGEM_SPD_100;
+			dp->full_duplex = B_FALSE;
+		} else if ((val & MII_ABILITY_10BASE_T_FD)) {
+			/* 10BaseT & full duplex */
+			dp->speed = USBGEM_SPD_10;
+			dp->full_duplex = B_TRUE;
+		} else if ((val & MII_ABILITY_10BASE_T)) {
+			/* 10BaseT & half duplex */
+			dp->speed = USBGEM_SPD_10;
+			dp->full_duplex = B_FALSE;
+		} else {
+			/*
+			 * the link partner doesn't seem to have
+			 * auto-negotiation capability and our PHY
+			 * could not report current mode correctly.
+			 * We guess current mode by mii_control register.
+			 */
+			val = usbgem_mii_read(dp, MII_CONTROL, &err);
+			if (err != USB_SUCCESS) {
+				goto usberr;
+			}
+
+			/* select 100m half or 10m half */
+			dp->speed = (val & MII_CONTROL_100MB) ?
+			    USBGEM_SPD_100 : USBGEM_SPD_10;
+			dp->full_duplex = B_FALSE;
+			fix_phy = B_TRUE;
+
+			cmn_err(CE_NOTE,
+			    "!%s: auto-negotiation done but "
+			    "common ability not found.\n"
+			    "PHY state: control:%b advert:%b lpable:%b\n"
+			    "guessing %d Mbps %s duplex mode",
+			    dp->name,
+			    val, MII_CONTROL_BITS,
+			    advert, MII_ABILITY_BITS,
+			    lpable, MII_ABILITY_BITS,
+			    usbgem_speed_value[dp->speed],
+			    dp->full_duplex ? "full" : "half");
+		}
+
+		if (dp->full_duplex) {
+			dp->flow_control =
+			    usbgem_fc_result[fc_cap_decode(advert)]
+			    [fc_cap_decode(lpable)];
+		} else {
+			dp->flow_control = FLOW_CONTROL_NONE;
+		}
+		dp->mii_state = MII_STATE_MEDIA_SETUP;
+		dp->mii_timer = dp->ugc.usbgc_mii_linkdown_timeout;
+		goto next_nowait;
+
+	case MII_STATE_MEDIA_SETUP:
+		DPRINTF(2, (CE_CONT, "!%s: setup midia mode", dp->name));
+
+		/* assume the link state is down */
+		dp->mii_state = MII_STATE_LINKDOWN;
+		dp->mii_supress_msg = B_FALSE;
+
+		/* use short interval */
+		dp->mii_interval = WATCH_INTERVAL_FAST;
+
+		if ((!dp->anadv_autoneg) ||
+		    dp->ugc.usbgc_mii_an_oneshot || fix_phy) {
+
+			/*
+			 * write the result of auto negotiation back.
+			 */
+			val = usbgem_mii_read(dp, MII_CONTROL, &err);
+			if (err != USB_SUCCESS) {
+				goto usberr;
+			}
+			val &= ~(MII_CONTROL_SPEED | MII_CONTROL_FDUPLEX |
+			    MII_CONTROL_ANE   | MII_CONTROL_RSAN);
+
+			if (dp->full_duplex) {
+				val |= MII_CONTROL_FDUPLEX;
+			}
+
+			switch (dp->speed) {
+			case USBGEM_SPD_1000:
+				val |= MII_CONTROL_1000MB;
+				break;
+
+			case USBGEM_SPD_100:
+				val |= MII_CONTROL_100MB;
+				break;
+
+			default:
+				cmn_err(CE_WARN, "%s: unknown speed:%d",
+				    dp->name, dp->speed);
+				/* FALLTHROUGH */
+
+			case USBGEM_SPD_10:
+				/* for USBGEM_SPD_10, do nothing */
+				break;
+			}
+
+			if (dp->mii_status & MII_STATUS_XSTATUS) {
+				usbgem_mii_write(dp,
+				    MII_1000TC, MII_1000TC_CFG_EN, &err);
+				if (err != USB_SUCCESS) {
+					goto usberr;
+				}
+			}
+			usbgem_mii_write(dp, MII_CONTROL, val, &err);
+			if (err != USB_SUCCESS) {
+				goto usberr;
+			}
+		}
+		/*
+		 * XXX -- nic state should be one of
+		 * NIC_STATE_DISCONNECTED
+		 * NIC_STATE_STOPPED
+		 * NIC_STATE_INITIALIZED
+		 * NIC_STATE_ONLINE
+		 */
+		if (dp->nic_state >= NIC_STATE_INITIALIZED) {
+			/* notify the result of autonegotiation to mac */
+			if (usbgem_hal_set_media(dp) != USB_SUCCESS) {
+				goto usberr;
+			}
+		}
+		goto next_nowait;
+
+	case MII_STATE_LINKDOWN:
+		status = usbgem_mii_read(dp, MII_STATUS, &err);
+		if (err != USB_SUCCESS) {
+			goto usberr;
+		}
+		if (status & MII_STATUS_LINKUP) {
+			/*
+			 * Link is going up
+			 */
+			dp->mii_state = MII_STATE_LINKUP;
+			dp->mii_supress_msg = B_FALSE;
+
+			DPRINTF(0, (CE_CONT,
+			    "!%s: link up detected: status:%b",
+			    dp->name, status, MII_STATUS_BITS));
+
+			/*
+			 * MII_CONTROL_100MB and  MII_CONTROL_FDUPLEX are
+			 * ignored when MII_CONTROL_ANE is set.
+			 */
+			cmn_err(CE_CONT,
+			    "!%s: Link up: %d Mbps %s duplex %s flow control",
+			    dp->name,
+			    usbgem_speed_value[dp->speed],
+			    dp->full_duplex ? "full" : "half",
+			    usbgem_fc_type[dp->flow_control]);
+
+			dp->mii_interval =
+			    dp->ugc.usbgc_mii_link_watch_interval;
+
+			if (dp->ugc.usbgc_mii_hw_link_detection &&
+			    dp->nic_state == NIC_STATE_ONLINE) {
+				dp->mii_interval = 0;
+			}
+
+			if (dp->nic_state == NIC_STATE_ONLINE) {
+				if (dp->mac_state == MAC_STATE_INITIALIZED) {
+					(void) usbgem_mac_start(dp);
+				}
+				tx_sched = B_TRUE;
+			}
+
+			goto next;
+		}
+
+		dp->mii_supress_msg = B_TRUE;
+		if (dp->anadv_autoneg) {
+			dp->mii_timer -= diff;
+			if (dp->mii_timer <= 0) {
+				/*
+				 * the link down timer expired.
+				 * need to restart auto-negotiation.
+				 */
+				linkdown_action =
+				    dp->ugc.usbgc_mii_linkdown_timeout_action;
+				goto restart_autonego;
+			}
+		}
+		/* don't change mii_state */
+		goto next;
+
+	case MII_STATE_LINKUP:
+		if (rwlock == RW_READER) {
+			/* first pass, read mii status */
+			status = usbgem_mii_read(dp, MII_STATUS, &err);
+			if (err != USB_SUCCESS) {
+				goto usberr;
+			}
+		}
+		if ((status & MII_STATUS_LINKUP) == 0) {
+			/*
+			 * Link is going down
+			 */
+			cmn_err(CE_NOTE,
+			    "!%s: link down detected: status:%b",
+			    dp->name, status, MII_STATUS_BITS);
+			/*
+			 * Acquire exclusive lock to change mii_state
+			 */
+			if (rwlock == RW_READER) {
+				rwlock = RW_WRITER;
+				rw_exit(&dp->dev_state_lock);
+				goto again;
+			}
+
+			dp->mii_state = MII_STATE_LINKDOWN;
+			dp->mii_timer = dp->ugc.usbgc_mii_linkdown_timeout;
+
+			/*
+			 * As we may change the state of the device,
+			 * let us acquire exclusive lock for the state.
+			 */
+			if (dp->nic_state == NIC_STATE_ONLINE &&
+			    dp->mac_state == MAC_STATE_ONLINE &&
+			    dp->ugc.usbgc_mii_stop_mac_on_linkdown) {
+				(void) usbgem_restart_nic(dp);
+				/* drain tx */
+				tx_sched = B_TRUE;
+			}
+
+			if (dp->anadv_autoneg) {
+				/* need to restart auto-negotiation */
+				linkdown_action =
+				    dp->ugc.usbgc_mii_linkdown_action;
+				goto restart_autonego;
+			}
+			/*
+			 * don't use hw link down detection until the link
+			 * status become stable for a while.
+			 */
+			dp->mii_interval =
+			    dp->ugc.usbgc_mii_link_watch_interval;
+
+			goto next;
+		}
+
+		/*
+		 * still link up, no need to change mii_state
+		 */
+		if (dp->ugc.usbgc_mii_hw_link_detection &&
+		    dp->nic_state == NIC_STATE_ONLINE) {
+			/*
+			 * no need to check link status periodicly
+			 * if nic can generate interrupts when link go down.
+			 */
+			dp->mii_interval = 0;
+		}
+		goto next;
+	}
+	/* NOTREACHED */
+	cmn_err(CE_PANIC, "!%s: %s: not reached", dp->name, __func__);
+
+	/*
+	 * Actions for new state.
+	 */
+restart_autonego:
+	switch (linkdown_action) {
+	case MII_ACTION_RESET:
+		if (!dp->mii_supress_msg) {
+			cmn_err(CE_CONT, "!%s: resetting PHY", dp->name);
+		}
+		dp->mii_supress_msg = B_TRUE;
+		goto reset_phy;
+
+	case MII_ACTION_NONE:
+		dp->mii_supress_msg = B_TRUE;
+		if (dp->ugc.usbgc_mii_an_oneshot) {
+			goto autonego;
+		}
+		/* PHY will restart autonego automatically */
+		dp->mii_state = MII_STATE_AUTONEGOTIATING;
+		dp->mii_timer = dp->ugc.usbgc_mii_an_timeout;
+		dp->mii_interval = dp->ugc.usbgc_mii_an_watch_interval;
+		goto next;
+
+	case MII_ACTION_RSA:
+		if (!dp->mii_supress_msg) {
+			cmn_err(CE_CONT, "!%s: restarting auto-negotiation",
+			    dp->name);
+		}
+		dp->mii_supress_msg = B_TRUE;
+		goto autonego;
+
+	default:
+		cmn_err(CE_PANIC, "!%s: unknowm linkdown action: %d",
+		    dp->name, dp->ugc.usbgc_mii_linkdown_action);
+		dp->mii_supress_msg = B_TRUE;
+	}
+	/* NOTREACHED */
+
+reset_phy:
+	if (!dp->mii_supress_msg) {
+		cmn_err(CE_CONT, "!%s: resetting PHY", dp->name);
+	}
+	dp->mii_state = MII_STATE_RESETTING;
+	dp->mii_timer = dp->ugc.usbgc_mii_reset_timeout;
+	if (!dp->ugc.usbgc_mii_dont_reset) {
+		usbgem_mii_write(dp, MII_CONTROL, MII_CONTROL_RESET, &err);
+		if (err != USB_SUCCESS) {
+			goto usberr;
+		}
+	}
+	dp->mii_interval = WATCH_INTERVAL_FAST;
+	goto next;
+
+autonego:
+	if (!dp->mii_supress_msg) {
+		cmn_err(CE_CONT, "!%s: auto-negotiation started", dp->name);
+	}
+	dp->mii_state = MII_STATE_AUTONEGOTIATING;
+	dp->mii_timer = dp->ugc.usbgc_mii_an_timeout;
+
+	/* start/restart autoneg */
+	val = usbgem_mii_read(dp, MII_CONTROL, &err) &
+	    ~(MII_CONTROL_ISOLATE | MII_CONTROL_PWRDN | MII_CONTROL_RESET);
+	if (err != USB_SUCCESS) {
+		goto usberr;
+	}
+	if (val & MII_CONTROL_ANE) {
+		val |= MII_CONTROL_RSAN;
+	}
+	usbgem_mii_write(dp, MII_CONTROL,
+	    val | dp->ugc.usbgc_mii_an_cmd | MII_CONTROL_ANE, &err);
+	if (err != USB_SUCCESS) {
+		goto usberr;
+	}
+
+	dp->mii_interval = dp->ugc.usbgc_mii_an_watch_interval;
+	goto next;
+
+usberr:
+	dp->mii_state = MII_STATE_UNKNOWN;
+	dp->mii_interval = dp->ugc.usbgc_mii_link_watch_interval;
+	tx_sched = B_TRUE;
+
+next:
+	*newstatep = dp->mii_state;
+	rw_exit(&dp->dev_state_lock);
+	return (tx_sched);
+}
+
+static void
+usbgem_mii_link_watcher(struct usbgem_dev *dp)
+{
+	int		old_mii_state;
+	int		new_mii_state;
+	boolean_t	tx_sched;
+
+	DPRINTF(0, (CE_CONT, "!%s: %s: called", dp->name, __func__));
+
+	for (; ; ) {
+
+		mutex_enter(&dp->link_watcher_lock);
+		if (dp->mii_interval) {
+			(void) cv_timedwait(&dp->link_watcher_wait_cv,
+			    &dp->link_watcher_lock,
+			    dp->mii_interval + ddi_get_lbolt());
+		} else {
+			cv_wait(&dp->link_watcher_wait_cv,
+			    &dp->link_watcher_lock);
+		}
+		mutex_exit(&dp->link_watcher_lock);
+
+		if (dp->link_watcher_stop) {
+			break;
+		}
+
+		/* we block callbacks from disconnect/suspend and restart */
+		tx_sched = usbgem_mii_link_check(dp,
+		    &old_mii_state, &new_mii_state);
+
+		/*
+		 * gld v2 notifier functions are not able to
+		 * be called with any locks in this layer.
+		 */
+		if (tx_sched) {
+			/* kick potentially stopped downstream */
+#ifdef USBGEM_CONFIG_GLDv3
+			mac_tx_update(dp->mh);
+#else
+			gld_sched(dp->macinfo);
+#endif
+		}
+
+		if (old_mii_state != new_mii_state) {
+			/* notify new mii link state */
+			if (new_mii_state == MII_STATE_LINKUP) {
+				dp->linkup_delay = 0;
+				USBGEM_LINKUP(dp);
+			} else if (dp->linkup_delay <= 0) {
+				USBGEM_LINKDOWN(dp);
+			}
+		} else if (dp->linkup_delay < 0) {
+			/* first linkup timeout */
+			dp->linkup_delay = 0;
+			USBGEM_LINKDOWN(dp);
+		}
+	}
+
+	thread_exit();
+}
+
+void
+usbgem_mii_update_link(struct usbgem_dev *dp)
+{
+	cv_signal(&dp->link_watcher_wait_cv);
+}
+
+int
+usbgem_mii_probe_default(struct usbgem_dev *dp)
+{
+	int		phy;
+	uint16_t	status;
+	uint16_t	xstatus;
+	int		err;
+	uint16_t	adv;
+	uint16_t	adv_org;
+
+	DPRINTF(3, (CE_CONT, "!%s: %s: called", dp->name, __func__));
+
+	/*
+	 * Scan PHY
+	 */
+	dp->mii_status = 0;
+
+	/* Try default phy first */
+	if (dp->mii_phy_addr) {
+		status = usbgem_mii_read(dp, MII_STATUS, &err);
+		if (err != USB_SUCCESS) {
+			goto usberr;
+		}
+		if (status != 0xffff && status != 0x0000) {
+			goto PHY_found;
+		}
+
+		if (dp->mii_phy_addr < 0) {
+			cmn_err(CE_NOTE,
+		    "!%s: failed to probe default internal and/or non-MII PHY",
+			    dp->name);
+			return (USB_FAILURE);
+		}
+
+		cmn_err(CE_NOTE,
+		    "!%s: failed to probe default MII PHY at %d",
+		    dp->name, dp->mii_phy_addr);
+	}
+
+	/* Try all possible address */
+	for (phy = dp->ugc.usbgc_mii_addr_min; phy < 32; phy++) {
+		dp->mii_phy_addr = phy;
+		status = usbgem_mii_read(dp, MII_STATUS, &err);
+		if (err != USB_SUCCESS) {
+			DPRINTF(0, (CE_CONT,
+			    "!%s: %s: mii_read(status) failed",
+			    dp->name, __func__));
+			goto usberr;
+		}
+
+		if (status != 0xffff && status != 0x0000) {
+			usbgem_mii_write(dp, MII_CONTROL, 0, &err);
+			if (err != USB_SUCCESS) {
+				DPRINTF(0, (CE_CONT,
+				    "!%s: %s: mii_write(control) failed",
+				    dp->name, __func__));
+				goto usberr;
+			}
+			goto PHY_found;
+		}
+	}
+	for (phy = dp->ugc.usbgc_mii_addr_min; phy < 32; phy++) {
+		dp->mii_phy_addr = phy;
+		usbgem_mii_write(dp, MII_CONTROL, 0, &err);
+		if (err != USB_SUCCESS) {
+			DPRINTF(0, (CE_CONT,
+			    "!%s: %s: mii_write(control) failed",
+			    dp->name, __func__));
+			goto usberr;
+		}
+		status = usbgem_mii_read(dp, MII_STATUS, &err);
+		if (err != USB_SUCCESS) {
+			DPRINTF(0, (CE_CONT,
+			    "!%s: %s: mii_read(status) failed",
+			    dp->name, __func__));
+			goto usberr;
+		}
+
+		if (status != 0xffff && status != 0) {
+			goto PHY_found;
+		}
+	}
+
+	cmn_err(CE_NOTE, "!%s: no MII PHY found", dp->name);
+	return (USB_FAILURE);
+
+PHY_found:
+	dp->mii_status = status;
+	dp->mii_status_ro = ~status;
+	dp->mii_phy_id = usbgem_mii_read(dp, MII_PHYIDH, &err) << 16;
+	if (err != USB_SUCCESS) {
+		DPRINTF(0, (CE_CONT,
+		    "!%s: %s: mii_read(PHYIDH) failed",
+		    dp->name, __func__));
+		goto usberr;
+	}
+	dp->mii_phy_id |= usbgem_mii_read(dp, MII_PHYIDL, &err);
+	if (err != USB_SUCCESS) {
+		DPRINTF(0, (CE_CONT,
+		    "!%s: %s: mii_read(PHYIDL) failed",
+		    dp->name, __func__));
+		goto usberr;
+	}
+
+	if (dp->mii_phy_addr < 0) {
+		cmn_err(CE_CONT, "!%s: using internal/non-MII PHY(0x%08x)",
+		    dp->name, dp->mii_phy_id);
+	} else {
+		cmn_err(CE_CONT, "!%s: MII PHY (0x%08x) found at %d",
+		    dp->name, dp->mii_phy_id, dp->mii_phy_addr);
+	}
+
+	cmn_err(CE_CONT,
+	    "!%s: PHY control:%b, status:%b, advert:%b, lpar:%b, exp:%b",
+	    dp->name,
+	    usbgem_mii_read(dp, MII_CONTROL, &err), MII_CONTROL_BITS,
+	    status, MII_STATUS_BITS,
+	    usbgem_mii_read(dp, MII_AN_ADVERT, &err), MII_ABILITY_BITS,
+	    usbgem_mii_read(dp, MII_AN_LPABLE, &err), MII_ABILITY_BITS,
+	    usbgem_mii_read(dp, MII_AN_EXPANSION, &err), MII_AN_EXP_BITS);
+
+	dp->mii_xstatus = 0;
+	if (status & MII_STATUS_XSTATUS) {
+		dp->mii_xstatus = usbgem_mii_read(dp, MII_XSTATUS, &err);
+
+		cmn_err(CE_CONT, "!%s: xstatus:%b",
+		    dp->name, dp->mii_xstatus, MII_XSTATUS_BITS);
+	}
+	dp->mii_xstatus_ro = ~dp->mii_xstatus;
+
+	/* check if the phy can advertize pause abilities */
+	adv_org = usbgem_mii_read(dp, MII_AN_ADVERT, &err);
+	if (err != USB_SUCCESS) {
+		goto usberr;
+	}
+
+	usbgem_mii_write(dp, MII_AN_ADVERT,
+	    MII_ABILITY_PAUSE | MII_ABILITY_ASM_DIR, &err);
+	if (err != USB_SUCCESS) {
+		goto usberr;
+	}
+
+	adv = usbgem_mii_read(dp, MII_AN_ADVERT, &err);
+	if (err != USB_SUCCESS) {
+		goto usberr;
+	}
+
+	if ((adv & MII_ABILITY_PAUSE) == 0) {
+		dp->ugc.usbgc_flow_control &= ~1;
+	}
+
+	if ((adv & MII_ABILITY_ASM_DIR) == 0) {
+		dp->ugc.usbgc_flow_control &= ~2;
+	}
+
+	usbgem_mii_write(dp, MII_AN_ADVERT, adv_org, &err);
+	if (err != USB_SUCCESS) {
+		goto usberr;
+	}
+	return (USB_SUCCESS);
+
+usberr:
+	return (USB_FAILURE);
+}
+
+int
+usbgem_mii_init_default(struct usbgem_dev *dp)
+{
+	/* ENPTY */
+	return (USB_SUCCESS);
+}
+
+static int
+usbgem_mii_start(struct usbgem_dev *dp)
+{
+	int	err;
+	kthread_t	*lwth;
+
+	DPRINTF(0, (CE_CONT, "!%s: %s: called", dp->name, __func__));
+
+	/* make a first call of usbgem_mii_link_check() */
+	dp->link_watcher_stop = 0;
+	dp->mii_state = MII_STATE_UNKNOWN;
+	dp->mii_interval = drv_usectohz(1000*1000); /* 1sec */
+	dp->mii_last_check = ddi_get_lbolt();
+	dp->linkup_delay = 600 * drv_usectohz(1000*1000); /* 10 minutes */
+
+	lwth = thread_create(NULL, 0, usbgem_mii_link_watcher, dp, 0, &p0,
+	    TS_RUN, minclsyspri);
+	if (lwth == NULL) {
+		cmn_err(CE_WARN,
+		    "!%s: %s: failed to create a link watcher thread",
+		    dp->name, __func__);
+		return (USB_FAILURE);
+	}
+	dp->link_watcher_did = lwth->t_did;
+
+	return (USB_SUCCESS);
+}
+
+static void
+usbgem_mii_stop(struct usbgem_dev *dp)
+{
+	DPRINTF(0, (CE_CONT, "!%s: %s: called", dp->name, __func__));
+
+	/* Ensure timer routine stopped */
+	dp->link_watcher_stop = 1;
+	cv_signal(&dp->link_watcher_wait_cv);
+	thread_join(dp->link_watcher_did);
+}
+
+/* ============================================================== */
+/*
+ * internal mac register operation interface
+ */
+/* ============================================================== */
+/*
+ * usbgem_mac_init: cold start
+ */
+static int
+usbgem_mac_init(struct usbgem_dev *dp)
+{
+	int	err;
+
+	DPRINTF(0, (CE_CONT, "!%s: %s: called", dp->name, __func__));
+
+	if (dp->mac_state == MAC_STATE_DISCONNECTED) {
+		/* pretend we succeeded */
+		return (USB_SUCCESS);
+	}
+
+	ASSERT(dp->mac_state == MAC_STATE_STOPPED);
+
+	/* reset fatal error timestamp */
+	dp->fatal_error = (clock_t)0;
+
+	/* reset tx side state */
+	mutex_enter(&dp->txlock);
+	dp->tx_busy_cnt = 0;
+	dp->tx_max_packets = dp->ugc.usbgc_tx_list_max;
+	mutex_exit(&dp->txlock);
+
+	/* reset rx side state */
+	mutex_enter(&dp->rxlock);
+	dp->rx_busy_cnt = 0;
+	mutex_exit(&dp->rxlock);
+
+	err = usbgem_hal_init_chip(dp);
+	if (err == USB_SUCCESS) {
+		dp->mac_state = MAC_STATE_INITIALIZED;
+	}
+
+	return (err);
+}
+
+/*
+ * usbgem_mac_start: warm start
+ */
+static int
+usbgem_mac_start(struct usbgem_dev *dp)
+{
+	int	err;
+	int	i;
+	usb_flags_t	flags = 0;
+	usb_intr_req_t	*req;
+#ifdef USBGEM_DEBUG_LEVEL
+	usb_pipe_state_t	p_state;
+#endif
+	DPRINTF(0, (CE_CONT, "!%s: %s: called", dp->name, __func__));
+
+	if (dp->mac_state == MAC_STATE_DISCONNECTED) {
+		/* do nothing but don't return failure */
+		return (USB_SUCCESS);
+	}
+
+	if (dp->mac_state != MAC_STATE_INITIALIZED) {
+		/* don't return failer */
+		DPRINTF(0, (CE_CONT,
+		    "!%s: %s: mac_state(%d) is not MAC_STATE_INITIALIZED",
+		    dp->name, __func__, dp->mac_state));
+		goto x;
+	}
+
+	dp->mac_state = MAC_STATE_ONLINE;
+
+	if (usbgem_hal_start_chip(dp) != USB_SUCCESS) {
+		cmn_err(CE_NOTE,
+		    "!%s: %s: usb error was detected during start_chip",
+		    dp->name, __func__);
+		goto x;
+	}
+
+#ifdef USBGEM_DEBUG_LEVEL
+	usb_pipe_get_state(dp->intr_pipe, &p_state, 0);
+	ASSERT(p_state == USB_PIPE_STATE_IDLE);
+#endif /* USBGEM_DEBUG_LEVEL */
+
+	if (dp->ugc.usbgc_interrupt && dp->intr_pipe) {
+
+		/* make a request for interrupt */
+
+		req = usb_alloc_intr_req(dp->dip, 0, USB_FLAGS_SLEEP);
+		if (req == NULL) {
+			cmn_err(CE_WARN, "!%s: %s: failed to allocate intreq",
+			    dp->name, __func__);
+			goto x;
+		}
+		req->intr_data = NULL;
+		req->intr_client_private = (usb_opaque_t)dp;
+		req->intr_timeout = 0;
+		req->intr_attributes =
+		    USB_ATTRS_SHORT_XFER_OK | USB_ATTRS_AUTOCLEARING;
+		req->intr_len = dp->ep_intr->wMaxPacketSize;
+		req->intr_cb = usbgem_intr_cb;
+		req->intr_exc_cb = usbgem_intr_cb;
+		req->intr_completion_reason = 0;
+		req->intr_cb_flags = 0;
+
+		err = usb_pipe_intr_xfer(dp->intr_pipe, req, flags);
+		if (err != USB_SUCCESS) {
+			cmn_err(CE_WARN,
+			    "%s: err:%d failed to start polling of intr pipe",
+			    dp->name, err);
+			goto x;
+		}
+	}
+
+	/* kick to receive the first packet */
+	if (usbgem_init_rx_buf(dp) != USB_SUCCESS) {
+		goto err_stop_intr;
+	}
+	dp->rx_active = B_TRUE;
+
+	return (USB_SUCCESS);
+
+err_stop_intr:
+	/* stop the interrupt pipe */
+	DPRINTF(0, (CE_CONT, "!%s: %s: FAULURE", dp->name, __func__));
+	if (dp->ugc.usbgc_interrupt && dp->intr_pipe) {
+		usb_pipe_stop_intr_polling(dp->intr_pipe, USB_FLAGS_SLEEP);
+	}
+x:
+	ASSERT(dp->mac_state == MAC_STATE_ONLINE);
+	/* we use another flag to indicate error state. */
+	if (dp->fatal_error == (clock_t)0) {
+		dp->fatal_error = usbgem_timestamp_nz();
+	}
+	return (USB_FAILURE);
+}
+
+static int
+usbgem_mac_stop(struct usbgem_dev *dp, int new_state, boolean_t graceful)
+{
+	DPRINTF(0, (CE_CONT, "!%s: %s: called", dp->name, __func__));
+
+	/*
+	 * we must have writer lock for dev_state_lock
+	 */
+	ASSERT(new_state == MAC_STATE_STOPPED
+	    || new_state == MAC_STATE_DISCONNECTED);
+
+	/* stop polling interrupt pipe */
+	if (dp->ugc.usbgc_interrupt && dp->intr_pipe) {
+		usb_pipe_stop_intr_polling(dp->intr_pipe, USB_FLAGS_SLEEP);
+	}
+
+	if (new_state == MAC_STATE_STOPPED || graceful) {
+		/* stop the nic hardware completely */
+		if (usbgem_hal_stop_chip(dp) != USB_SUCCESS) {
+			(void) usbgem_hal_reset_chip(dp);
+		}
+	}
+
+	/* stop preparing new rx packets and sending new packets */
+	dp->mac_state = new_state;
+
+	/* other processors must get mac_state correctly after here */
+	membar_producer();
+
+	/* cancel all requests we have sent */
+	usb_pipe_reset(dp->dip, dp->bulkin_pipe, USB_FLAGS_SLEEP, NULL, 0);
+	usb_pipe_reset(dp->dip, dp->bulkout_pipe, USB_FLAGS_SLEEP, NULL, 0);
+
+	DPRINTF(0, (CE_CONT,
+	    "!%s: %s: rx_busy_cnt:%d tx_busy_cnt:%d",
+	    dp->name, __func__, dp->rx_busy_cnt, dp->tx_busy_cnt));
+
+	/*
+	 * Here all rx packets has been cancelled and their call back
+	 * function has been exeuted, because we called usb_pipe_reset
+	 * synchronously.
+	 * So actually we just ensure rx_busy_cnt == 0.
+	 */
+	mutex_enter(&dp->rxlock);
+	while (dp->rx_busy_cnt > 0) {
+		cv_wait(&dp->rx_drain_cv, &dp->rxlock);
+	}
+	mutex_exit(&dp->rxlock);
+
+	DPRINTF(0, (CE_CONT, "!%s: %s: rx_busy_cnt is %d now",
+	    dp->name, __func__, dp->rx_busy_cnt));
+
+	mutex_enter(&dp->txlock);
+	while (dp->tx_busy_cnt > 0) {
+		cv_wait(&dp->tx_drain_cv, &dp->txlock);
+	}
+	mutex_exit(&dp->txlock);
+
+	DPRINTF(0, (CE_CONT, "!%s: %s: tx_busy_cnt is %d now",
+	    dp->name, __func__, dp->tx_busy_cnt));
+
+	return (USB_SUCCESS);
+}
+
+static int
+usbgem_add_multicast(struct usbgem_dev *dp, const uint8_t *ep)
+{
+	int	cnt;
+	int	err;
+
+	DPRINTF(1, (CE_CONT, "!%s: %s: called", dp->name, __func__));
+
+	sema_p(&dp->rxfilter_lock);
+	if (dp->mc_count_req++ < USBGEM_MAXMC) {
+		/* append the new address at the end of the mclist */
+		cnt = dp->mc_count;
+		bcopy(ep, dp->mc_list[cnt].addr.ether_addr_octet,
+		    ETHERADDRL);
+		if (dp->ugc.usbgc_multicast_hash) {
+			dp->mc_list[cnt].hash =
+			    (*dp->ugc.usbgc_multicast_hash)(dp, ep);
+		}
+		dp->mc_count = cnt + 1;
+	}
+
+	if (dp->mc_count_req != dp->mc_count) {
+		/* multicast address list overflow */
+		dp->rxmode |= RXMODE_MULTI_OVF;
+	} else {
+		dp->rxmode &= ~RXMODE_MULTI_OVF;
+	}
+
+	if (dp->mac_state != MAC_STATE_DISCONNECTED) {
+		/* tell new multicast list to the hardware */
+		err = usbgem_hal_set_rx_filter(dp);
+	}
+	sema_v(&dp->rxfilter_lock);
+
+	return (err);
+}
+
+static int
+usbgem_remove_multicast(struct usbgem_dev *dp, const uint8_t *ep)
+{
+	size_t		len;
+	int		i;
+	int		cnt;
+	int		err;
+
+	DPRINTF(1, (CE_CONT, "!%s: %s: called", dp->name, __func__));
+
+	sema_p(&dp->rxfilter_lock);
+	dp->mc_count_req--;
+	cnt = dp->mc_count;
+	for (i = 0; i < cnt; i++) {
+		if (bcmp(ep, &dp->mc_list[i].addr, ETHERADDRL)) {
+			continue;
+		}
+		/* shrink the mclist by copying forward */
+		len = (cnt - (i + 1)) * sizeof (*dp->mc_list);
+		if (len > 0) {
+			bcopy(&dp->mc_list[i+1], &dp->mc_list[i], len);
+		}
+		dp->mc_count--;
+		break;
+	}
+
+	if (dp->mc_count_req != dp->mc_count) {
+		/* multicast address list overflow */
+		dp->rxmode |= RXMODE_MULTI_OVF;
+	} else {
+		dp->rxmode &= ~RXMODE_MULTI_OVF;
+	}
+
+	if (dp->mac_state != MAC_STATE_DISCONNECTED) {
+		err = usbgem_hal_set_rx_filter(dp);
+	}
+	sema_v(&dp->rxfilter_lock);
+
+	return (err);
+}
+
+
+/* ============================================================== */
+/*
+ * ioctl
+ */
+/* ============================================================== */
+enum ioc_reply {
+	IOC_INVAL = -1,				/* bad, NAK with EINVAL	*/
+	IOC_DONE,				/* OK, reply sent	*/
+	IOC_ACK,				/* OK, just send ACK	*/
+	IOC_REPLY,				/* OK, just send reply	*/
+	IOC_RESTART_ACK,			/* OK, restart & ACK	*/
+	IOC_RESTART_REPLY			/* OK, restart & reply	*/
+};
+
+
+#ifdef USBGEM_CONFIG_MAC_PROP
+static int
+usbgem_get_def_val(struct usbgem_dev *dp, mac_prop_id_t pr_num,
+    uint_t pr_valsize, void *pr_val)
+{
+	link_flowctrl_t fl;
+	int err = 0;
+
+	ASSERT(pr_valsize > 0);
+	switch (pr_num) {
+	case MAC_PROP_AUTONEG:
+		*(uint8_t *)pr_val =
+		    BOOLEAN(dp->mii_status & MII_STATUS_CANAUTONEG);
+		break;
+
+	case MAC_PROP_FLOWCTRL:
+		if (pr_valsize < sizeof (link_flowctrl_t)) {
+			return (EINVAL);
+		}
+		switch (dp->ugc.usbgc_flow_control) {
+		case FLOW_CONTROL_NONE:
+			fl = LINK_FLOWCTRL_NONE;
+			break;
+		case FLOW_CONTROL_SYMMETRIC:
+			fl = LINK_FLOWCTRL_BI;
+			break;
+		case FLOW_CONTROL_TX_PAUSE:
+			fl = LINK_FLOWCTRL_TX;
+			break;
+		case FLOW_CONTROL_RX_PAUSE:
+			fl = LINK_FLOWCTRL_RX;
+			break;
+		}
+		bcopy(&fl, pr_val, sizeof (fl));
+		break;
+
+	case MAC_PROP_ADV_1000FDX_CAP:
+	case MAC_PROP_EN_1000FDX_CAP:
+		*(uint8_t *)pr_val =
+		    (dp->mii_xstatus & MII_XSTATUS_1000BASET_FD) ||
+		    (dp->mii_xstatus & MII_XSTATUS_1000BASEX_FD);
+		break;
+
+	case MAC_PROP_ADV_1000HDX_CAP:
+	case MAC_PROP_EN_1000HDX_CAP:
+		*(uint8_t *)pr_val =
+		    (dp->mii_xstatus & MII_XSTATUS_1000BASET) ||
+		    (dp->mii_xstatus & MII_XSTATUS_1000BASEX);
+		break;
+
+	case MAC_PROP_ADV_100T4_CAP:
+	case MAC_PROP_EN_100T4_CAP:
+		*(uint8_t *)pr_val =
+		    BOOLEAN(dp->mii_status & MII_STATUS_100_BASE_T4);
+		break;
+
+	case MAC_PROP_ADV_100FDX_CAP:
+	case MAC_PROP_EN_100FDX_CAP:
+		*(uint8_t *)pr_val =
+		    BOOLEAN(dp->mii_status & MII_STATUS_100_BASEX_FD);
+		break;
+
+	case MAC_PROP_ADV_100HDX_CAP:
+	case MAC_PROP_EN_100HDX_CAP:
+		*(uint8_t *)pr_val =
+		    BOOLEAN(dp->mii_status & MII_STATUS_100_BASEX);
+		break;
+
+	case MAC_PROP_ADV_10FDX_CAP:
+	case MAC_PROP_EN_10FDX_CAP:
+		*(uint8_t *)pr_val =
+		    BOOLEAN(dp->mii_status & MII_STATUS_10_FD);
+		break;
+
+	case MAC_PROP_ADV_10HDX_CAP:
+	case MAC_PROP_EN_10HDX_CAP:
+		*(uint8_t *)pr_val =
+		    BOOLEAN(dp->mii_status & MII_STATUS_10);
+		break;
+
+	default:
+		err = ENOTSUP;
+		break;
+	}
+	return (err);
+}
+
+#ifdef MAC_VERSION_V1
+static void
+usbgem_m_propinfo(void *arg, const char *pr_name, mac_prop_id_t pr_num,
+    mac_prop_info_handle_t prh)
+{
+	struct usbgem_dev *dp = arg;
+	link_flowctrl_t fl;
+
+	/*
+	 * By default permissions are read/write unless specified
+	 * otherwise by the driver.
+	 */
+
+	switch (pr_num) {
+	case MAC_PROP_DUPLEX:
+	case MAC_PROP_SPEED:
+	case MAC_PROP_STATUS:
+	case MAC_PROP_ADV_1000FDX_CAP:
+	case MAC_PROP_ADV_1000HDX_CAP:
+	case MAC_PROP_ADV_100FDX_CAP:
+	case MAC_PROP_ADV_100HDX_CAP:
+	case MAC_PROP_ADV_10FDX_CAP:
+	case MAC_PROP_ADV_10HDX_CAP:
+	case MAC_PROP_ADV_100T4_CAP:
+	case MAC_PROP_EN_100T4_CAP:
+		mac_prop_info_set_perm(prh, MAC_PROP_PERM_READ);
+		break;
+
+	case MAC_PROP_EN_1000FDX_CAP:
+		if ((dp->mii_xstatus_ro & MII_XSTATUS_1000BASET_FD) == 0) {
+			mac_prop_info_set_default_uint8(prh,
+			    BOOLEAN(
+			    dp->mii_xstatus & MII_XSTATUS_1000BASET_FD));
+		} else if ((dp->mii_xstatus_ro & MII_XSTATUS_1000BASEX_FD)
+		    == 0) {
+			mac_prop_info_set_default_uint8(prh,
+			    BOOLEAN(
+			    dp->mii_xstatus & MII_XSTATUS_1000BASEX_FD));
+		} else {
+			mac_prop_info_set_perm(prh, MAC_PROP_PERM_READ);
+		}
+		break;
+
+	case MAC_PROP_EN_1000HDX_CAP:
+		if ((dp->mii_xstatus_ro & MII_XSTATUS_1000BASET) == 0) {
+			mac_prop_info_set_default_uint8(prh,
+			    BOOLEAN(
+			    dp->mii_xstatus & MII_XSTATUS_1000BASET));
+		} else if ((dp->mii_xstatus_ro & MII_XSTATUS_1000BASEX) == 0) {
+			mac_prop_info_set_default_uint8(prh,
+			    BOOLEAN(
+			    dp->mii_xstatus & MII_XSTATUS_1000BASEX));
+		} else {
+			mac_prop_info_set_perm(prh, MAC_PROP_PERM_READ);
+		}
+		break;
+
+	case MAC_PROP_EN_100FDX_CAP:
+		if ((dp->mii_status_ro & MII_STATUS_100_BASEX_FD) == 0) {
+			mac_prop_info_set_default_uint8(prh,
+			    BOOLEAN(dp->mii_status & MII_STATUS_100_BASEX_FD));
+		} else {
+			mac_prop_info_set_perm(prh, MAC_PROP_PERM_READ);
+		}
+		break;
+
+	case MAC_PROP_EN_100HDX_CAP:
+		if ((dp->mii_status_ro & MII_STATUS_100_BASEX) == 0) {
+			mac_prop_info_set_default_uint8(prh,
+			    BOOLEAN(dp->mii_status & MII_STATUS_100_BASEX));
+		} else {
+			mac_prop_info_set_perm(prh, MAC_PROP_PERM_READ);
+		}
+		break;
+
+	case MAC_PROP_EN_10FDX_CAP:
+		if ((dp->mii_status_ro & MII_STATUS_10_FD) == 0) {
+			mac_prop_info_set_default_uint8(prh,
+			    BOOLEAN(dp->mii_status & MII_STATUS_10_FD));
+		} else {
+			mac_prop_info_set_perm(prh, MAC_PROP_PERM_READ);
+		}
+		break;
+
+	case MAC_PROP_EN_10HDX_CAP:
+		if ((dp->mii_status_ro & MII_STATUS_10) == 0) {
+			mac_prop_info_set_default_uint8(prh,
+			    BOOLEAN(dp->mii_status & MII_STATUS_10));
+		} else {
+			mac_prop_info_set_perm(prh, MAC_PROP_PERM_READ);
+		}
+		break;
+
+	case MAC_PROP_AUTONEG:
+		if ((dp->mii_status_ro & MII_STATUS_CANAUTONEG) == 0) {
+			mac_prop_info_set_default_uint8(prh,
+			    BOOLEAN(dp->mii_status & MII_STATUS_CANAUTONEG));
+		} else {
+			mac_prop_info_set_perm(prh, MAC_PROP_PERM_READ);
+		}
+		break;
+
+	case MAC_PROP_FLOWCTRL:
+		switch (dp->ugc.usbgc_flow_control) {
+		case FLOW_CONTROL_NONE:
+			fl = LINK_FLOWCTRL_NONE;
+			break;
+		case FLOW_CONTROL_SYMMETRIC:
+			fl = LINK_FLOWCTRL_BI;
+			break;
+		case FLOW_CONTROL_TX_PAUSE:
+			fl = LINK_FLOWCTRL_TX;
+			break;
+		case FLOW_CONTROL_RX_PAUSE:
+			fl = LINK_FLOWCTRL_RX;
+			break;
+		}
+		mac_prop_info_set_default_link_flowctrl(prh, fl);
+		break;
+
+	case MAC_PROP_MTU:
+		mac_prop_info_set_range_uint32(prh,
+		    dp->ugc.usbgc_min_mtu, dp->ugc.usbgc_max_mtu);
+		break;
+
+	case MAC_PROP_PRIVATE:
+		break;
+	}
+}
+#endif
+
+static int
+usbgem_m_setprop(void *arg, const char *pr_name, mac_prop_id_t pr_num,
+    uint_t pr_valsize, const void *pr_val)
+{
+	struct usbgem_dev *dp = arg;
+	int err = 0;
+	boolean_t	update = B_FALSE;
+	link_flowctrl_t flowctrl;
+	uint32_t cur_mtu, new_mtu;
+
+	rw_enter(&dp->dev_state_lock, RW_WRITER);
+	switch (pr_num) {
+	case MAC_PROP_EN_1000FDX_CAP:
+		if ((dp->mii_xstatus_ro & MII_XSTATUS_1000BASET_FD) == 0 ||
+		    (dp->mii_xstatus_ro & MII_XSTATUS_1000BASEX_FD) == 0) {
+			if (dp->anadv_1000fdx != *(uint8_t *)pr_val) {
+				dp->anadv_1000fdx = *(uint8_t *)pr_val;
+				update = B_TRUE;
+			}
+		} else {
+			err = ENOTSUP;
+		}
+		break;
+
+	case MAC_PROP_EN_1000HDX_CAP:
+		if ((dp->mii_xstatus_ro & MII_XSTATUS_1000BASET) == 0 ||
+		    (dp->mii_xstatus_ro & MII_XSTATUS_1000BASEX) == 0) {
+			if (dp->anadv_1000hdx != *(uint8_t *)pr_val) {
+				dp->anadv_1000hdx = *(uint8_t *)pr_val;
+				update = B_TRUE;
+			}
+		} else {
+			err = ENOTSUP;
+		}
+		break;
+
+	case MAC_PROP_EN_100FDX_CAP:
+		if ((dp->mii_status_ro & MII_STATUS_100_BASEX_FD) == 0) {
+			if (dp->anadv_100fdx != *(uint8_t *)pr_val) {
+				dp->anadv_100fdx = *(uint8_t *)pr_val;
+				update = B_TRUE;
+			}
+		} else {
+			err = ENOTSUP;
+		}
+		break;
+
+	case MAC_PROP_EN_100HDX_CAP:
+		if ((dp->mii_status_ro & MII_STATUS_100_BASEX) == 0) {
+			if (dp->anadv_100hdx != *(uint8_t *)pr_val) {
+				dp->anadv_100hdx = *(uint8_t *)pr_val;
+				update = B_TRUE;
+			}
+		} else {
+			err = ENOTSUP;
+		}
+		break;
+
+	case MAC_PROP_EN_10FDX_CAP:
+		if ((dp->mii_status_ro & MII_STATUS_10_FD) == 0) {
+			if (dp->anadv_10fdx != *(uint8_t *)pr_val) {
+				dp->anadv_10fdx = *(uint8_t *)pr_val;
+				update = B_TRUE;
+			}
+		} else {
+			err = ENOTSUP;
+		}
+		break;
+
+	case MAC_PROP_EN_10HDX_CAP:
+		if ((dp->mii_status_ro & MII_STATUS_10_FD) == 0) {
+			if (dp->anadv_10hdx != *(uint8_t *)pr_val) {
+				dp->anadv_10hdx = *(uint8_t *)pr_val;
+				update = B_TRUE;
+			}
+		} else {
+			err = ENOTSUP;
+		}
+		break;
+
+	case MAC_PROP_AUTONEG:
+		if ((dp->mii_status_ro & MII_STATUS_CANAUTONEG) == 0) {
+			if (dp->anadv_autoneg != *(uint8_t *)pr_val) {
+				dp->anadv_autoneg = *(uint8_t *)pr_val;
+				update = B_TRUE;
+			}
+		} else {
+			err = ENOTSUP;
+		}
+		break;
+
+	case MAC_PROP_FLOWCTRL:
+		bcopy(pr_val, &flowctrl, sizeof (flowctrl));
+
+		switch (flowctrl) {
+		default:
+			err = EINVAL;
+			break;
+
+		case LINK_FLOWCTRL_NONE:
+			if (dp->flow_control != FLOW_CONTROL_NONE) {
+				dp->flow_control = FLOW_CONTROL_NONE;
+				update = B_TRUE;
+			}
+			break;
+
+		case LINK_FLOWCTRL_RX:
+			if (dp->flow_control != FLOW_CONTROL_RX_PAUSE) {
+				dp->flow_control = FLOW_CONTROL_RX_PAUSE;
+				update = B_TRUE;
+			}
+			break;
+
+		case LINK_FLOWCTRL_TX:
+			if (dp->flow_control != FLOW_CONTROL_TX_PAUSE) {
+				dp->flow_control = FLOW_CONTROL_TX_PAUSE;
+				update = B_TRUE;
+			}
+			break;
+
+		case LINK_FLOWCTRL_BI:
+			if (dp->flow_control != FLOW_CONTROL_SYMMETRIC) {
+				dp->flow_control = FLOW_CONTROL_SYMMETRIC;
+				update = B_TRUE;
+			}
+			break;
+		}
+		break;
+
+	case MAC_PROP_ADV_1000FDX_CAP:
+	case MAC_PROP_ADV_1000HDX_CAP:
+	case MAC_PROP_ADV_100FDX_CAP:
+	case MAC_PROP_ADV_100HDX_CAP:
+	case MAC_PROP_ADV_10FDX_CAP:
+	case MAC_PROP_ADV_10HDX_CAP:
+	case MAC_PROP_STATUS:
+	case MAC_PROP_SPEED:
+	case MAC_PROP_DUPLEX:
+		err = ENOTSUP; /* read-only prop. Can't set this. */
+		break;
+
+	case MAC_PROP_MTU:
+		bcopy(pr_val, &new_mtu, sizeof (new_mtu));
+		if (new_mtu != dp->mtu) {
+			err = EINVAL;
+		}
+		break;
+
+	case MAC_PROP_PRIVATE:
+		err = ENOTSUP;
+		break;
+
+	default:
+		err = ENOTSUP;
+		break;
+	}
+
+	if (update) {
+		/* sync with PHY */
+		usbgem_choose_forcedmode(dp);
+		dp->mii_state = MII_STATE_UNKNOWN;
+		cv_signal(&dp->link_watcher_wait_cv);
+	}
+	rw_exit(&dp->dev_state_lock);
+	return (err);
+}
+
+static int
+#ifdef MAC_VERSION_V1
+usbgem_m_getprop(void *arg, const char *pr_name, mac_prop_id_t pr_num,
+    uint_t pr_valsize, void *pr_val)
+#else
+usbgem_m_getprop(void *arg, const char *pr_name, mac_prop_id_t pr_num,
+    uint_t pr_flags, uint_t pr_valsize, void *pr_val, uint_t *perm)
+#endif
+{
+	struct usbgem_dev *dp = arg;
+	int err = 0;
+	link_flowctrl_t flowctrl;
+	uint64_t tmp = 0;
+
+	if (pr_valsize == 0) {
+		return (EINVAL);
+	}
+#ifndef MAC_VERSION_V1
+	*perm = MAC_PROP_PERM_RW;
+#endif
+	bzero(pr_val, pr_valsize);
+#ifndef MAC_VERSION_V1
+	if ((pr_flags & MAC_PROP_DEFAULT) && (pr_num != MAC_PROP_PRIVATE)) {
+		return (usbgem_get_def_val(dp, pr_num, pr_valsize, pr_val));
+	}
+#endif
+	rw_enter(&dp->dev_state_lock, RW_READER);
+	switch (pr_num) {
+	case MAC_PROP_DUPLEX:
+#ifndef MAC_VERSION_V1
+		*perm = MAC_PROP_PERM_READ;
+#endif
+		if (pr_valsize >= sizeof (link_duplex_t)) {
+			if (dp->mii_state != MII_STATE_LINKUP) {
+				*(link_duplex_t *)pr_val = LINK_DUPLEX_UNKNOWN;
+			} else if (dp->full_duplex) {
+				*(link_duplex_t *)pr_val = LINK_DUPLEX_FULL;
+			} else {
+				*(link_duplex_t *)pr_val = LINK_DUPLEX_HALF;
+			}
+		} else {
+			err = EINVAL;
+		}
+		break;
+	case MAC_PROP_SPEED:
+#ifndef MAC_VERSION_V1
+		*perm = MAC_PROP_PERM_READ;
+#endif
+		if (pr_valsize >= sizeof (uint64_t)) {
+			switch (dp->speed) {
+			case USBGEM_SPD_1000:
+				tmp = 1000000000;
+				break;
+			case USBGEM_SPD_100:
+				tmp = 100000000;
+				break;
+			case USBGEM_SPD_10:
+				tmp = 10000000;
+				break;
+			default:
+				tmp = 0;
+			}
+			bcopy(&tmp, pr_val, sizeof (tmp));
+		} else {
+			err = EINVAL;
+		}
+		break;
+
+	case MAC_PROP_AUTONEG:
+#ifndef MAC_VERSION_V1
+		if (dp->mii_status_ro & MII_STATUS_CANAUTONEG) {
+			*perm = MAC_PROP_PERM_READ;
+		}
+#endif
+		*(uint8_t *)pr_val = dp->anadv_autoneg;
+		break;
+
+	case MAC_PROP_FLOWCTRL:
+		if (pr_valsize >= sizeof (link_flowctrl_t)) {
+			switch (dp->flow_control) {
+			case FLOW_CONTROL_NONE:
+				flowctrl = LINK_FLOWCTRL_NONE;
+				break;
+			case FLOW_CONTROL_RX_PAUSE:
+				flowctrl = LINK_FLOWCTRL_RX;
+				break;
+			case FLOW_CONTROL_TX_PAUSE:
+				flowctrl = LINK_FLOWCTRL_TX;
+				break;
+			case FLOW_CONTROL_SYMMETRIC:
+				flowctrl = LINK_FLOWCTRL_BI;
+				break;
+			}
+			bcopy(&flowctrl, pr_val, sizeof (flowctrl));
+		} else {
+			err = EINVAL;
+		}
+		break;
+
+	case MAC_PROP_ADV_1000FDX_CAP:
+	case MAC_PROP_ADV_1000HDX_CAP:
+	case MAC_PROP_ADV_100FDX_CAP:
+	case MAC_PROP_ADV_100HDX_CAP:
+	case MAC_PROP_ADV_10FDX_CAP:
+	case MAC_PROP_ADV_10HDX_CAP:
+	case MAC_PROP_ADV_100T4_CAP:
+		usbgem_get_def_val(dp, pr_num, pr_valsize, pr_val);
+		break;
+
+	case MAC_PROP_EN_1000FDX_CAP:
+#ifndef MAC_VERSION_V1
+		if ((dp->mii_xstatus_ro & MII_XSTATUS_1000BASET_FD) &&
+		    (dp->mii_xstatus_ro & MII_XSTATUS_1000BASEX_FD)) {
+			*perm = MAC_PROP_PERM_READ;
+		}
+#endif
+		*(uint8_t *)pr_val = dp->anadv_1000fdx;
+		break;
+
+	case MAC_PROP_EN_1000HDX_CAP:
+#ifndef MAC_VERSION_V1
+		if ((dp->mii_xstatus_ro & MII_XSTATUS_1000BASET) &&
+		    (dp->mii_xstatus_ro & MII_XSTATUS_1000BASEX)) {
+			*perm = MAC_PROP_PERM_READ;
+		}
+#endif
+		*(uint8_t *)pr_val = dp->anadv_1000hdx;
+		break;
+
+	case MAC_PROP_EN_100FDX_CAP:
+#ifndef MAC_VERSION_V1
+		if (dp->mii_status_ro & MII_STATUS_100_BASEX_FD) {
+			*perm = MAC_PROP_PERM_READ;
+		}
+#endif
+		*(uint8_t *)pr_val = dp->anadv_100fdx;
+		break;
+
+	case MAC_PROP_EN_100HDX_CAP:
+#ifndef MAC_VERSION_V1
+		if (dp->mii_status_ro & MII_STATUS_100_BASEX) {
+			*perm = MAC_PROP_PERM_READ;
+		}
+#endif
+		*(uint8_t *)pr_val = dp->anadv_100hdx;
+		break;
+
+	case MAC_PROP_EN_10FDX_CAP:
+#ifndef MAC_VERSION_V1
+		if (dp->mii_status_ro & MII_STATUS_10_FD) {
+			*perm = MAC_PROP_PERM_READ;
+		}
+#endif
+		*(uint8_t *)pr_val = dp->anadv_10fdx;
+		break;
+
+	case MAC_PROP_EN_10HDX_CAP:
+#ifndef MAC_VERSION_V1
+		if (dp->mii_status_ro & MII_STATUS_10) {
+			*perm = MAC_PROP_PERM_READ;
+		}
+#endif
+		*(uint8_t *)pr_val = dp->anadv_10hdx;
+		break;
+
+	case MAC_PROP_EN_100T4_CAP:
+#ifndef MAC_VERSION_V1
+		if (dp->mii_status_ro & MII_STATUS_100_BASE_T4) {
+			*perm = MAC_PROP_PERM_READ;
+		}
+#endif
+		*(uint8_t *)pr_val = dp->anadv_100t4;
+		break;
+
+	case MAC_PROP_PRIVATE:
+		err = ENOTSUP;
+		break;
+
+#ifndef MAC_VERSION_V1
+	case MAC_PROP_MTU: {
+		mac_propval_range_t range;
+		if (!(pr_flags & MAC_PROP_POSSIBLE)) {
+			err = ENOTSUP;
+			break;
+		}
+		if (pr_valsize < sizeof (mac_propval_range_t)) {
+			err = EINVAL;
+			break;
+		}
+		range.mpr_count = 1;
+		range.mpr_type = MAC_PROPVAL_UINT32;
+		range.range_uint32[0].mpur_min = ETHERMTU;
+		range.range_uint32[0].mpur_max = dp->mtu;
+		bcopy(&range, pr_val, sizeof (range));
+		break;
+	}
+#endif
+	default:
+		err = ENOTSUP;
+		break;
+	}
+
+	rw_exit(&dp->dev_state_lock);
+	return (err);
+}
+#endif /* USBGEM_CONFIG_MAC_PROP */
+
+#ifdef USBGEM_CONFIG_ND
+/* ============================================================== */
+/*
+ * ND interface
+ */
+/* ============================================================== */
+enum {
+	PARAM_AUTONEG_CAP,
+	PARAM_PAUSE_CAP,
+	PARAM_ASYM_PAUSE_CAP,
+	PARAM_1000FDX_CAP,
+	PARAM_1000HDX_CAP,
+	PARAM_100T4_CAP,
+	PARAM_100FDX_CAP,
+	PARAM_100HDX_CAP,
+	PARAM_10FDX_CAP,
+	PARAM_10HDX_CAP,
+
+	PARAM_ADV_AUTONEG_CAP,
+	PARAM_ADV_PAUSE_CAP,
+	PARAM_ADV_ASYM_PAUSE_CAP,
+	PARAM_ADV_1000FDX_CAP,
+	PARAM_ADV_1000HDX_CAP,
+	PARAM_ADV_100T4_CAP,
+	PARAM_ADV_100FDX_CAP,
+	PARAM_ADV_100HDX_CAP,
+	PARAM_ADV_10FDX_CAP,
+	PARAM_ADV_10HDX_CAP,
+	PARAM_ADV_1000T_MS,
+
+	PARAM_LP_AUTONEG_CAP,
+	PARAM_LP_PAUSE_CAP,
+	PARAM_LP_ASYM_PAUSE_CAP,
+	PARAM_LP_1000FDX_CAP,
+	PARAM_LP_1000HDX_CAP,
+	PARAM_LP_100T4_CAP,
+	PARAM_LP_100FDX_CAP,
+	PARAM_LP_100HDX_CAP,
+	PARAM_LP_10FDX_CAP,
+	PARAM_LP_10HDX_CAP,
+
+	PARAM_LINK_STATUS,
+	PARAM_LINK_SPEED,
+	PARAM_LINK_DUPLEX,
+
+	PARAM_LINK_AUTONEG,
+	PARAM_LINK_RX_PAUSE,
+	PARAM_LINK_TX_PAUSE,
+
+	PARAM_LOOP_MODE,
+	PARAM_MSI_CNT,
+#ifdef DEBUG_RESUME
+	PARAM_RESUME_TEST,
+#endif
+
+	PARAM_COUNT
+};
+
+struct usbgem_nd_arg {
+	struct usbgem_dev	*dp;
+	int		item;
+};
+
+static int
+usbgem_param_get(queue_t *q, mblk_t *mp, caddr_t arg, cred_t *credp)
+{
+	struct usbgem_dev	*dp = ((struct usbgem_nd_arg *)(void *)arg)->dp;
+	int		item = ((struct usbgem_nd_arg *)(void *)arg)->item;
+	long		val;
+
+	DPRINTF(1, (CE_CONT, "!%s: %s: called, item:%d",
+	    dp->name, __func__, item));
+
+	switch (item) {
+	case PARAM_AUTONEG_CAP:
+		val = BOOLEAN(dp->mii_status & MII_STATUS_CANAUTONEG);
+		DPRINTF(1, (CE_CONT, "autoneg_cap:%d", val));
+		break;
+
+	case PARAM_PAUSE_CAP:
+		val = dp->ugc.usbgc_flow_control != FLOW_CONTROL_NONE;
+		break;
+
+	case PARAM_ASYM_PAUSE_CAP:
+		val = dp->ugc.usbgc_flow_control > FLOW_CONTROL_SYMMETRIC;
+		break;
+
+	case PARAM_1000FDX_CAP:
+		val = (dp->mii_xstatus & MII_XSTATUS_1000BASET_FD) ||
+		    (dp->mii_xstatus & MII_XSTATUS_1000BASEX_FD);
+		break;
+
+	case PARAM_1000HDX_CAP:
+		val = (dp->mii_xstatus & MII_XSTATUS_1000BASET) ||
+		    (dp->mii_xstatus & MII_XSTATUS_1000BASEX);
+		break;
+
+	case PARAM_100T4_CAP:
+		val = BOOLEAN(dp->mii_status & MII_STATUS_100_BASE_T4);
+		break;
+
+	case PARAM_100FDX_CAP:
+		val = BOOLEAN(dp->mii_status & MII_STATUS_100_BASEX_FD);
+		break;
+
+	case PARAM_100HDX_CAP:
+		val = BOOLEAN(dp->mii_status & MII_STATUS_100_BASEX);
+		break;
+
+	case PARAM_10FDX_CAP:
+		val = BOOLEAN(dp->mii_status & MII_STATUS_10_FD);
+		break;
+
+	case PARAM_10HDX_CAP:
+		val = BOOLEAN(dp->mii_status & MII_STATUS_10);
+		break;
+
+	case PARAM_ADV_AUTONEG_CAP:
+		val = dp->anadv_autoneg;
+		break;
+
+	case PARAM_ADV_PAUSE_CAP:
+		val = dp->anadv_pause;
+		break;
+
+	case PARAM_ADV_ASYM_PAUSE_CAP:
+		val = dp->anadv_asmpause;
+		break;
+
+	case PARAM_ADV_1000FDX_CAP:
+		val = dp->anadv_1000fdx;
+		break;
+
+	case PARAM_ADV_1000HDX_CAP:
+		val = dp->anadv_1000hdx;
+		break;
+
+	case PARAM_ADV_100T4_CAP:
+		val = dp->anadv_100t4;
+		break;
+
+	case PARAM_ADV_100FDX_CAP:
+		val = dp->anadv_100fdx;
+		break;
+
+	case PARAM_ADV_100HDX_CAP:
+		val = dp->anadv_100hdx;
+		break;
+
+	case PARAM_ADV_10FDX_CAP:
+		val = dp->anadv_10fdx;
+		break;
+
+	case PARAM_ADV_10HDX_CAP:
+		val = dp->anadv_10hdx;
+		break;
+
+	case PARAM_ADV_1000T_MS:
+		val = dp->anadv_1000t_ms;
+		break;
+
+	case PARAM_LP_AUTONEG_CAP:
+		val = BOOLEAN(dp->mii_exp & MII_AN_EXP_LPCANAN);
+		break;
+
+	case PARAM_LP_PAUSE_CAP:
+		val = BOOLEAN(dp->mii_lpable & MII_ABILITY_PAUSE);
+		break;
+
+	case PARAM_LP_ASYM_PAUSE_CAP:
+		val = BOOLEAN(dp->mii_lpable & MII_ABILITY_ASM_DIR);
+		break;
+
+	case PARAM_LP_1000FDX_CAP:
+		val = BOOLEAN(dp->mii_stat1000 & MII_1000TS_LP_FULL);
+		break;
+
+	case PARAM_LP_1000HDX_CAP:
+		val = BOOLEAN(dp->mii_stat1000 & MII_1000TS_LP_HALF);
+		break;
+
+	case PARAM_LP_100T4_CAP:
+		val = BOOLEAN(dp->mii_lpable & MII_ABILITY_100BASE_T4);
+		break;
+
+	case PARAM_LP_100FDX_CAP:
+		val = BOOLEAN(dp->mii_lpable & MII_ABILITY_100BASE_TX_FD);
+		break;
+
+	case PARAM_LP_100HDX_CAP:
+		val = BOOLEAN(dp->mii_lpable & MII_ABILITY_100BASE_TX);
+		break;
+
+	case PARAM_LP_10FDX_CAP:
+		val = BOOLEAN(dp->mii_lpable & MII_ABILITY_10BASE_T_FD);
+		break;
+
+	case PARAM_LP_10HDX_CAP:
+		val = BOOLEAN(dp->mii_lpable & MII_ABILITY_10BASE_T);
+		break;
+
+	case PARAM_LINK_STATUS:
+		val = (dp->mii_state == MII_STATE_LINKUP);
+		break;
+
+	case PARAM_LINK_SPEED:
+		val = usbgem_speed_value[dp->speed];
+		break;
+
+	case PARAM_LINK_DUPLEX:
+		val = 0;
+		if (dp->mii_state == MII_STATE_LINKUP) {
+			val = dp->full_duplex ? 2 : 1;
+		}
+		break;
+
+	case PARAM_LINK_AUTONEG:
+		val = BOOLEAN(dp->mii_exp & MII_AN_EXP_LPCANAN);
+		break;
+
+	case PARAM_LINK_RX_PAUSE:
+		val = (dp->flow_control == FLOW_CONTROL_SYMMETRIC) ||
+		    (dp->flow_control == FLOW_CONTROL_RX_PAUSE);
+		break;
+
+	case PARAM_LINK_TX_PAUSE:
+		val = (dp->flow_control == FLOW_CONTROL_SYMMETRIC) ||
+		    (dp->flow_control == FLOW_CONTROL_TX_PAUSE);
+		break;
+
+#ifdef DEBUG_RESUME
+	case PARAM_RESUME_TEST:
+		val = 0;
+		break;
+#endif
+	default:
+		cmn_err(CE_WARN, "%s: unimplemented ndd control (%d)",
+		    dp->name, item);
+		break;
+	}
+
+	(void) mi_mpprintf(mp, "%ld", val);
+
+	return (0);
+}
+
+static int
+usbgem_param_set(queue_t *q,
+    mblk_t *mp, char *value, caddr_t arg, cred_t *credp)
+{
+	struct usbgem_dev	*dp = ((struct usbgem_nd_arg *)(void *)arg)->dp;
+	int		item = ((struct usbgem_nd_arg *)(void *)arg)->item;
+	long		val;
+	char		*end;
+
+	DPRINTF(1, (CE_CONT, "!%s: %s: called", dp->name, __func__));
+	if (ddi_strtol(value, &end, 10, &val)) {
+		return (EINVAL);
+	}
+	if (end == value) {
+		return (EINVAL);
+	}
+
+	switch (item) {
+	case PARAM_ADV_AUTONEG_CAP:
+		if (val != 0 && val != 1) {
+			goto err;
+		}
+		if (val && (dp->mii_status & MII_STATUS_CANAUTONEG) == 0) {
+			goto err;
+		}
+		dp->anadv_autoneg = (int)val;
+		break;
+
+	case PARAM_ADV_PAUSE_CAP:
+		if (val != 0 && val != 1) {
+			goto err;
+		}
+		if (val && dp->ugc.usbgc_flow_control == FLOW_CONTROL_NONE) {
+			goto err;
+		}
+		dp->anadv_pause = (int)val;
+		break;
+
+	case PARAM_ADV_ASYM_PAUSE_CAP:
+		if (val != 0 && val != 1) {
+			goto err;
+		}
+		if (val &&
+		    dp->ugc.usbgc_flow_control <= FLOW_CONTROL_SYMMETRIC) {
+			goto err;
+		}
+		dp->anadv_asmpause = (int)val;
+		break;
+
+	case PARAM_ADV_1000FDX_CAP:
+		if (val != 0 && val != 1) {
+			goto err;
+		}
+		if (val && (dp->mii_xstatus &
+		    (MII_XSTATUS_1000BASET_FD |
+		    MII_XSTATUS_1000BASEX_FD)) == 0) {
+			goto err;
+		}
+		dp->anadv_1000fdx = (int)val;
+		break;
+
+	case PARAM_ADV_1000HDX_CAP:
+		if (val != 0 && val != 1) {
+			goto err;
+		}
+		if (val && (dp->mii_xstatus &
+		    (MII_XSTATUS_1000BASET | MII_XSTATUS_1000BASEX)) == 0) {
+			goto err;
+		}
+		dp->anadv_1000hdx = (int)val;
+		break;
+
+	case PARAM_ADV_100T4_CAP:
+		if (val != 0 && val != 1) {
+			goto err;
+		}
+		if (val && (dp->mii_status & MII_STATUS_100_BASE_T4) == 0) {
+			goto err;
+		}
+		dp->anadv_100t4 = (int)val;
+		break;
+
+	case PARAM_ADV_100FDX_CAP:
+		if (val != 0 && val != 1) {
+			goto err;
+		}
+		if (val && (dp->mii_status & MII_STATUS_100_BASEX_FD) == 0) {
+			goto err;
+		}
+		dp->anadv_100fdx = (int)val;
+		break;
+
+	case PARAM_ADV_100HDX_CAP:
+		if (val != 0 && val != 1) {
+			goto err;
+		}
+		if (val && (dp->mii_status & MII_STATUS_100_BASEX) == 0) {
+			goto err;
+		}
+		dp->anadv_100hdx = (int)val;
+		break;
+
+	case PARAM_ADV_10FDX_CAP:
+		if (val != 0 && val != 1) {
+			goto err;
+		}
+		if (val && (dp->mii_status & MII_STATUS_10_FD) == 0) {
+			goto err;
+		}
+		dp->anadv_10fdx = (int)val;
+		break;
+
+	case PARAM_ADV_10HDX_CAP:
+		if (val != 0 && val != 1) {
+			goto err;
+		}
+		if (val && (dp->mii_status & MII_STATUS_10) == 0) {
+			goto err;
+		}
+		dp->anadv_10hdx = (int)val;
+		break;
+
+	case PARAM_ADV_1000T_MS:
+		if (val != 0 && val != 1 && val != 2) {
+			goto err;
+		}
+		if (val && (dp->mii_xstatus &
+		    (MII_XSTATUS_1000BASET | MII_XSTATUS_1000BASET_FD)) == 0) {
+			goto err;
+		}
+		dp->anadv_1000t_ms = (int)val;
+		break;
+
+#ifdef DEBUG_RESUME
+	case PARAM_RESUME_TEST:
+		mutex_exit(&dp->xmitlock);
+		mutex_exit(&dp->intrlock);
+		gem_suspend(dp->dip);
+		gem_resume(dp->dip);
+		mutex_enter(&dp->intrlock);
+		mutex_enter(&dp->xmitlock);
+		break;
+#endif
+	}
+
+	/* sync with PHY */
+	usbgem_choose_forcedmode(dp);
+
+	dp->mii_state = MII_STATE_UNKNOWN;
+	if (dp->ugc.usbgc_mii_hw_link_detection) {
+		/* wake up link watcher possiblely sleeps */
+		cv_signal(&dp->link_watcher_wait_cv);
+	}
+
+	return (0);
+err:
+	return (EINVAL);
+}
+
+static void
+usbgem_nd_load(struct usbgem_dev *dp,
+    char *name, ndgetf_t gf, ndsetf_t sf, int item)
+{
+	struct usbgem_nd_arg	*arg;
+
+	ASSERT(item >= 0);
+	ASSERT(item < PARAM_COUNT);
+
+	arg = &((struct usbgem_nd_arg *)(void *)dp->nd_arg_p)[item];
+	arg->dp = dp;
+	arg->item = item;
+
+	DPRINTF(2, (CE_CONT, "!%s: %s: name:%s, item:%d",
+	    dp->name, __func__, name, item));
+	(void) nd_load(&dp->nd_data_p, name, gf, sf, (caddr_t)arg);
+}
+
+static void
+usbgem_nd_setup(struct usbgem_dev *dp)
+{
+	DPRINTF(1, (CE_CONT, "!%s: %s: called, mii_status:0x%b",
+	    dp->name, __func__, dp->mii_status, MII_STATUS_BITS));
+
+	ASSERT(dp->nd_arg_p == NULL);
+
+	dp->nd_arg_p =
+	    kmem_zalloc(sizeof (struct usbgem_nd_arg) * PARAM_COUNT, KM_SLEEP);
+
+#define	SETFUNC(x)	((x) ? usbgem_param_set : NULL)
+
+	usbgem_nd_load(dp, "autoneg_cap",
+	    usbgem_param_get, NULL, PARAM_AUTONEG_CAP);
+	usbgem_nd_load(dp, "pause_cap",
+	    usbgem_param_get, NULL, PARAM_PAUSE_CAP);
+	usbgem_nd_load(dp, "asym_pause_cap",
+	    usbgem_param_get, NULL, PARAM_ASYM_PAUSE_CAP);
+	usbgem_nd_load(dp, "1000fdx_cap",
+	    usbgem_param_get, NULL, PARAM_1000FDX_CAP);
+	usbgem_nd_load(dp, "1000hdx_cap",
+	    usbgem_param_get, NULL, PARAM_1000HDX_CAP);
+	usbgem_nd_load(dp, "100T4_cap",
+	    usbgem_param_get, NULL, PARAM_100T4_CAP);
+	usbgem_nd_load(dp, "100fdx_cap",
+	    usbgem_param_get, NULL, PARAM_100FDX_CAP);
+	usbgem_nd_load(dp, "100hdx_cap",
+	    usbgem_param_get, NULL, PARAM_100HDX_CAP);
+	usbgem_nd_load(dp, "10fdx_cap",
+	    usbgem_param_get, NULL, PARAM_10FDX_CAP);
+	usbgem_nd_load(dp, "10hdx_cap",
+	    usbgem_param_get, NULL, PARAM_10HDX_CAP);
+
+	/* Our advertised capabilities */
+	usbgem_nd_load(dp, "adv_autoneg_cap", usbgem_param_get,
+	    SETFUNC(dp->mii_status & MII_STATUS_CANAUTONEG),
+	    PARAM_ADV_AUTONEG_CAP);
+	usbgem_nd_load(dp, "adv_pause_cap", usbgem_param_get,
+	    SETFUNC(dp->ugc.usbgc_flow_control & 1),
+	    PARAM_ADV_PAUSE_CAP);
+	usbgem_nd_load(dp, "adv_asym_pause_cap", usbgem_param_get,
+	    SETFUNC(dp->ugc.usbgc_flow_control & 2),
+	    PARAM_ADV_ASYM_PAUSE_CAP);
+	usbgem_nd_load(dp, "adv_1000fdx_cap", usbgem_param_get,
+	    SETFUNC(dp->mii_xstatus &
+	    (MII_XSTATUS_1000BASEX_FD | MII_XSTATUS_1000BASET_FD)),
+	    PARAM_ADV_1000FDX_CAP);
+	usbgem_nd_load(dp, "adv_1000hdx_cap", usbgem_param_get,
+	    SETFUNC(dp->mii_xstatus &
+	    (MII_XSTATUS_1000BASEX | MII_XSTATUS_1000BASET)),
+	    PARAM_ADV_1000HDX_CAP);
+	usbgem_nd_load(dp, "adv_100T4_cap", usbgem_param_get,
+	    SETFUNC((dp->mii_status & MII_STATUS_100_BASE_T4) &&
+	    !dp->mii_advert_ro),
+	    PARAM_ADV_100T4_CAP);
+	usbgem_nd_load(dp, "adv_100fdx_cap", usbgem_param_get,
+	    SETFUNC((dp->mii_status & MII_STATUS_100_BASEX_FD) &&
+	    !dp->mii_advert_ro),
+	    PARAM_ADV_100FDX_CAP);
+	usbgem_nd_load(dp, "adv_100hdx_cap", usbgem_param_get,
+	    SETFUNC((dp->mii_status & MII_STATUS_100_BASEX) &&
+	    !dp->mii_advert_ro),
+	    PARAM_ADV_100HDX_CAP);
+	usbgem_nd_load(dp, "adv_10fdx_cap", usbgem_param_get,
+	    SETFUNC((dp->mii_status & MII_STATUS_10_FD) &&
+	    !dp->mii_advert_ro),
+	    PARAM_ADV_10FDX_CAP);
+	usbgem_nd_load(dp, "adv_10hdx_cap", usbgem_param_get,
+	    SETFUNC((dp->mii_status & MII_STATUS_10) &&
+	    !dp->mii_advert_ro),
+	    PARAM_ADV_10HDX_CAP);
+	usbgem_nd_load(dp, "adv_1000t_ms", usbgem_param_get,
+	    SETFUNC(dp->mii_xstatus &
+	    (MII_XSTATUS_1000BASET_FD | MII_XSTATUS_1000BASET)),
+	    PARAM_ADV_1000T_MS);
+
+
+	/* Partner's advertised capabilities */
+	usbgem_nd_load(dp, "lp_autoneg_cap",
+	    usbgem_param_get, NULL, PARAM_LP_AUTONEG_CAP);
+	usbgem_nd_load(dp, "lp_pause_cap",
+	    usbgem_param_get, NULL, PARAM_LP_PAUSE_CAP);
+	usbgem_nd_load(dp, "lp_asym_pause_cap",
+	    usbgem_param_get, NULL, PARAM_LP_ASYM_PAUSE_CAP);
+	usbgem_nd_load(dp, "lp_1000fdx_cap",
+	    usbgem_param_get, NULL, PARAM_LP_1000FDX_CAP);
+	usbgem_nd_load(dp, "lp_1000hdx_cap",
+	    usbgem_param_get, NULL, PARAM_LP_1000HDX_CAP);
+	usbgem_nd_load(dp, "lp_100T4_cap",
+	    usbgem_param_get, NULL, PARAM_LP_100T4_CAP);
+	usbgem_nd_load(dp, "lp_100fdx_cap",
+	    usbgem_param_get, NULL, PARAM_LP_100FDX_CAP);
+	usbgem_nd_load(dp, "lp_100hdx_cap",
+	    usbgem_param_get, NULL, PARAM_LP_100HDX_CAP);
+	usbgem_nd_load(dp, "lp_10fdx_cap",
+	    usbgem_param_get, NULL, PARAM_LP_10FDX_CAP);
+	usbgem_nd_load(dp, "lp_10hdx_cap",
+	    usbgem_param_get, NULL, PARAM_LP_10HDX_CAP);
+
+	/* Current operating modes */
+	usbgem_nd_load(dp, "link_status",
+	    usbgem_param_get, NULL, PARAM_LINK_STATUS);
+	usbgem_nd_load(dp, "link_speed",
+	    usbgem_param_get, NULL, PARAM_LINK_SPEED);
+	usbgem_nd_load(dp, "link_duplex",
+	    usbgem_param_get, NULL, PARAM_LINK_DUPLEX);
+	usbgem_nd_load(dp, "link_autoneg",
+	    usbgem_param_get, NULL, PARAM_LINK_AUTONEG);
+	usbgem_nd_load(dp, "link_rx_pause",
+	    usbgem_param_get, NULL, PARAM_LINK_RX_PAUSE);
+	usbgem_nd_load(dp, "link_tx_pause",
+	    usbgem_param_get, NULL, PARAM_LINK_TX_PAUSE);
+#ifdef DEBUG_RESUME
+	usbgem_nd_load(dp, "resume_test",
+	    usbgem_param_get, usbgem_param_set, PARAM_RESUME_TEST);
+#endif
+#undef	SETFUNC
+}
+
+static
+enum ioc_reply
+usbgem_nd_ioctl(struct usbgem_dev *dp,
+    queue_t *wq, mblk_t *mp, struct iocblk *iocp)
+{
+	boolean_t	ok;
+
+	DPRINTF(1, (CE_CONT, "!%s: %s: called", dp->name, __func__));
+
+	switch (iocp->ioc_cmd) {
+	case ND_GET:
+		ok = nd_getset(wq, dp->nd_data_p, mp);
+		DPRINTF(1, (CE_CONT,
+		    "%s: get %s", dp->name, ok ? "OK" : "FAIL"));
+		return (ok ? IOC_REPLY : IOC_INVAL);
+
+	case ND_SET:
+		ok = nd_getset(wq, dp->nd_data_p, mp);
+
+		DPRINTF(1, (CE_CONT, "%s: set %s err %d",
+		    dp->name, ok ? "OK" : "FAIL", iocp->ioc_error));
+
+		if (!ok) {
+			return (IOC_INVAL);
+		}
+
+		if (iocp->ioc_error) {
+			return (IOC_REPLY);
+		}
+
+		return (IOC_RESTART_REPLY);
+	}
+
+	cmn_err(CE_WARN, "%s: invalid cmd 0x%x", dp->name, iocp->ioc_cmd);
+
+	return (IOC_INVAL);
+}
+
+static void
+usbgem_nd_cleanup(struct usbgem_dev *dp)
+{
+	ASSERT(dp->nd_data_p != NULL);
+	ASSERT(dp->nd_arg_p != NULL);
+
+	nd_free(&dp->nd_data_p);
+
+	kmem_free(dp->nd_arg_p, sizeof (struct usbgem_nd_arg) * PARAM_COUNT);
+	dp->nd_arg_p = NULL;
+}
+#endif /* USBGEM_CONFIG_ND */
+
+static void
+usbgem_mac_ioctl(struct usbgem_dev *dp, queue_t *wq, mblk_t *mp)
+{
+	struct iocblk	*iocp;
+	enum ioc_reply	status;
+	int		cmd;
+
+	DPRINTF(1, (CE_CONT, "!%s: %s: called", dp->name, __func__));
+
+	/*
+	 * Validate the command before bothering with the mutex ...
+	 */
+	iocp = (void *)mp->b_rptr;
+	iocp->ioc_error = 0;
+	cmd = iocp->ioc_cmd;
+
+	DPRINTF(1, (CE_CONT, "%s: %s cmd:0x%x", dp->name, __func__, cmd));
+
+#ifdef USBGEM_CONFIG_ND
+	switch (cmd) {
+	default:
+		_NOTE(NOTREACHED)
+		status = IOC_INVAL;
+		break;
+
+	case ND_GET:
+	case ND_SET:
+		status = usbgem_nd_ioctl(dp, wq, mp, iocp);
+		break;
+	}
+
+	/*
+	 * Finally, decide how to reply
+	 */
+	switch (status) {
+	default:
+	case IOC_INVAL:
+		/*
+		 * Error, reply with a NAK and EINVAL or the specified error
+		 */
+		miocnak(wq, mp, 0, iocp->ioc_error == 0 ?
+		    EINVAL : iocp->ioc_error);
+		break;
+
+	case IOC_DONE:
+		/*
+		 * OK, reply already sent
+		 */
+		break;
+
+	case IOC_RESTART_ACK:
+	case IOC_ACK:
+		/*
+		 * OK, reply with an ACK
+		 */
+		miocack(wq, mp, 0, 0);
+		break;
+
+	case IOC_RESTART_REPLY:
+	case IOC_REPLY:
+		/*
+		 * OK, send prepared reply as ACK or NAK
+		 */
+		mp->b_datap->db_type =
+		    iocp->ioc_error == 0 ? M_IOCACK : M_IOCNAK;
+		qreply(wq, mp);
+		break;
+	}
+#else
+	miocnak(wq, mp, 0, EINVAL);
+	return;
+#endif /* USBGEM_CONFIG_GLDv3 */
+}
+
+#ifndef SYS_MAC_H
+#define	XCVR_UNDEFINED	0
+#define	XCVR_NONE	1
+#define	XCVR_10		2
+#define	XCVR_100T4	3
+#define	XCVR_100X	4
+#define	XCVR_100T2	5
+#define	XCVR_1000X	6
+#define	XCVR_1000T	7
+#endif
+static int
+usbgem_mac_xcvr_inuse(struct usbgem_dev *dp)
+{
+	int	val = XCVR_UNDEFINED;
+
+	if ((dp->mii_status & MII_STATUS_XSTATUS) == 0) {
+		if (dp->mii_status & MII_STATUS_100_BASE_T4) {
+			val = XCVR_100T4;
+		} else if (dp->mii_status &
+		    (MII_STATUS_100_BASEX_FD |
+		    MII_STATUS_100_BASEX)) {
+			val = XCVR_100X;
+		} else if (dp->mii_status &
+		    (MII_STATUS_100_BASE_T2_FD |
+		    MII_STATUS_100_BASE_T2)) {
+			val = XCVR_100T2;
+		} else if (dp->mii_status &
+		    (MII_STATUS_10_FD | MII_STATUS_10)) {
+			val = XCVR_10;
+		}
+	} else if (dp->mii_xstatus &
+	    (MII_XSTATUS_1000BASET_FD | MII_XSTATUS_1000BASET)) {
+		val = XCVR_1000T;
+	} else if (dp->mii_xstatus &
+	    (MII_XSTATUS_1000BASEX_FD | MII_XSTATUS_1000BASEX)) {
+		val = XCVR_1000X;
+	}
+
+	return (val);
+}
+
+#ifdef USBGEM_CONFIG_GLDv3
+/* ============================================================== */
+/*
+ * GLDv3 interface
+ */
+/* ============================================================== */
+static int	usbgem_m_getstat(void *, uint_t, uint64_t *);
+static int	usbgem_m_start(void *);
+static void	usbgem_m_stop(void *);
+static int	usbgem_m_setpromisc(void *, boolean_t);
+static int	usbgem_m_multicst(void *, boolean_t, const uint8_t *);
+static int	usbgem_m_unicst(void *, const uint8_t *);
+static mblk_t	*usbgem_m_tx(void *, mblk_t *);
+static void	usbgem_m_ioctl(void *, queue_t *, mblk_t *);
+#ifdef GEM_CONFIG_MAC_PROP
+static int	usbgem_m_setprop(void *, const char *, mac_prop_id_t,
+    uint_t, const void *);
+#ifdef MAC_VERSION_V1
+static int	usbgem_m_getprop(void *, const char *, mac_prop_id_t,
+    uint_t, void *);
+#else
+static int	usbgem_m_getprop(void *, const char *, mac_prop_id_t,
+    uint_t, uint_t, void *, uint_t *);
+#endif
+#endif
+
+#ifdef _SYS_MAC_PROVIDER_H
+#define	GEM_M_CALLBACK_FLAGS	(MC_IOCTL)
+#else
+#define	GEM_M_CALLBACK_FLAGS	(MC_IOCTL)
+#endif
+
+static mac_callbacks_t gem_m_callbacks = {
+#ifdef USBGEM_CONFIG_MAC_PROP
+#ifdef MAC_VERSION_V1
+	GEM_M_CALLBACK_FLAGS | MC_SETPROP | MC_GETPROP | MC_PROPINFO,
+#else
+	GEM_M_CALLBACK_FLAGS | MC_SETPROP | MC_GETPROP,
+#endif
+#else
+	GEM_M_CALLBACK_FLAGS,
+#endif
+	usbgem_m_getstat,
+	usbgem_m_start,
+	usbgem_m_stop,
+	usbgem_m_setpromisc,
+	usbgem_m_multicst,
+	usbgem_m_unicst,
+	usbgem_m_tx,
+#ifdef _SYS_MAC_PROVIDER_H
+#ifdef MAC_VERSION_V1
+	NULL,
+#endif
+#else
+	NULL,	/* m_resources */
+#endif
+	usbgem_m_ioctl,
+	NULL, /* m_getcapab */
+#ifdef USBGEM_CONFIG_MAC_PROP
+	NULL,
+	NULL,
+	usbgem_m_setprop,
+	usbgem_m_getprop,
+#endif
+#ifdef MAC_VERSION_V1
+	usbgem_m_propinfo,
+#endif
+};
+
+static int
+usbgem_m_start(void *arg)
+{
+	int	ret;
+	int	err;
+	struct usbgem_dev *dp = arg;
+
+	DPRINTF(0, (CE_CONT, "!%s: %s: called", dp->name, __func__));
+
+	err = EIO;
+
+	rw_enter(&dp->dev_state_lock, RW_WRITER);
+	dp->nic_state = NIC_STATE_ONLINE;
+
+	if (dp->mac_state == MAC_STATE_DISCONNECTED) {
+		err = 0;
+		goto x;
+	}
+	if (usbgem_mac_init(dp) != USB_SUCCESS) {
+		goto x;
+	}
+
+	/* initialize rx filter state */
+	sema_p(&dp->rxfilter_lock);
+	dp->mc_count = 0;
+	dp->mc_count_req = 0;
+
+	bcopy(dp->dev_addr.ether_addr_octet,
+	    dp->cur_addr.ether_addr_octet, ETHERADDRL);
+	dp->rxmode |= RXMODE_ENABLE;
+
+	ret = usbgem_hal_set_rx_filter(dp);
+	sema_v(&dp->rxfilter_lock);
+
+	if (ret != USB_SUCCESS) {
+		goto x;
+	}
+
+	if (dp->mii_state == MII_STATE_LINKUP) {
+		/* setup media mode if the link have been up */
+		if (usbgem_hal_set_media(dp) != USB_SUCCESS) {
+			goto x;
+		}
+		if (usbgem_mac_start(dp) != USB_SUCCESS) {
+			goto x;
+		}
+	}
+
+	err = 0;
+x:
+	rw_exit(&dp->dev_state_lock);
+	return (err);
+}
+
+static void
+usbgem_m_stop(void *arg)
+{
+	struct usbgem_dev	*dp = arg;
+
+	DPRINTF(0, (CE_CONT, "!%s: %s: called", dp->name, __func__));
+
+	/* stop rx gracefully */
+	rw_enter(&dp->dev_state_lock, RW_READER);
+	sema_p(&dp->rxfilter_lock);
+	dp->rxmode &= ~RXMODE_ENABLE;
+
+	if (dp->mac_state != MAC_STATE_DISCONNECTED) {
+		(void) usbgem_hal_set_rx_filter(dp);
+	}
+	sema_v(&dp->rxfilter_lock);
+	rw_exit(&dp->dev_state_lock);
+
+	/* make the nic state inactive */
+	rw_enter(&dp->dev_state_lock, RW_WRITER);
+	dp->nic_state = NIC_STATE_STOPPED;
+
+	/* stop mac completely */
+	if (dp->mac_state != MAC_STATE_DISCONNECTED) {
+		(void) usbgem_mac_stop(dp, MAC_STATE_STOPPED, STOP_GRACEFUL);
+	}
+	rw_exit(&dp->dev_state_lock);
+}
+
+static int
+usbgem_m_multicst(void *arg, boolean_t add, const uint8_t *ep)
+{
+	int	err;
+	int	ret;
+	struct usbgem_dev	*dp = arg;
+
+	DPRINTF(1, (CE_CONT, "!%s: %s: called", dp->name, __func__));
+
+	rw_enter(&dp->dev_state_lock, RW_READER);
+	if (add) {
+		ret = usbgem_add_multicast(dp, ep);
+	} else {
+		ret = usbgem_remove_multicast(dp, ep);
+	}
+	rw_exit(&dp->dev_state_lock);
+
+	err = 0;
+	if (ret != USB_SUCCESS) {
+#ifdef GEM_CONFIG_FMA
+		ddi_fm_service_impact(dp->dip, DDI_SERVICE_DEGRADED);
+#endif
+		err = EIO;
+	}
+
+	return (err);
+}
+
+static int
+usbgem_m_setpromisc(void *arg, boolean_t on)
+{
+	int	err;
+	struct usbgem_dev	*dp = arg;
+
+	DPRINTF(0, (CE_CONT, "!%s: %s: called", dp->name, __func__));
+
+	rw_enter(&dp->dev_state_lock, RW_READER);
+
+	sema_p(&dp->rxfilter_lock);
+	if (on) {
+		dp->rxmode |= RXMODE_PROMISC;
+	} else {
+		dp->rxmode &= ~RXMODE_PROMISC;
+	}
+
+	err = 0;
+	if (dp->mac_state != MAC_STATE_DISCONNECTED) {
+		if (usbgem_hal_set_rx_filter(dp) != USB_SUCCESS) {
+			err = EIO;
+		}
+	}
+	sema_v(&dp->rxfilter_lock);
+
+	rw_exit(&dp->dev_state_lock);
+
+#ifdef GEM_CONFIG_FMA
+	if (err != 0) {
+		ddi_fm_service_impact(dp->dip, DDI_SERVICE_DEGRADED);
+	}
+#endif
+	return (err);
+}
+
+int
+usbgem_m_getstat(void *arg, uint_t stat, uint64_t *valp)
+{
+	int	ret;
+	uint64_t	val;
+	struct usbgem_dev	*dp = arg;
+	struct usbgem_stats	*gstp = &dp->stats;
+
+	DPRINTF(1, (CE_CONT, "!%s: %s: called", dp->name, __func__));
+
+	rw_enter(&dp->dev_state_lock, RW_READER);
+	if (dp->mac_state == MAC_STATE_DISCONNECTED) {
+		rw_exit(&dp->dev_state_lock);
+		return (0);
+	}
+	ret = usbgem_hal_get_stats(dp);
+	rw_exit(&dp->dev_state_lock);
+
+#ifdef GEM_CONFIG_FMA
+	if (ret != USB_SUCCESS) {
+		ddi_fm_service_impact(dp->dip, DDI_SERVICE_DEGRADED);
+		return (EIO);
+	}
+#endif
+
+	switch (stat) {
+	case MAC_STAT_IFSPEED:
+		val = usbgem_speed_value[dp->speed] *1000000ull;
+		break;
+
+	case MAC_STAT_MULTIRCV:
+		val = gstp->rmcast;
+		break;
+
+	case MAC_STAT_BRDCSTRCV:
+		val = gstp->rbcast;
+		break;
+
+	case MAC_STAT_MULTIXMT:
+		val = gstp->omcast;
+		break;
+
+	case MAC_STAT_BRDCSTXMT:
+		val = gstp->obcast;
+		break;
+
+	case MAC_STAT_NORCVBUF:
+		val = gstp->norcvbuf + gstp->missed;
+		break;
+
+	case MAC_STAT_IERRORS:
+		val = gstp->errrcv;
+		break;
+
+	case MAC_STAT_NOXMTBUF:
+		val = gstp->noxmtbuf;
+		break;
+
+	case MAC_STAT_OERRORS:
+		val = gstp->errxmt;
+		break;
+
+	case MAC_STAT_COLLISIONS:
+		val = gstp->collisions;
+		break;
+
+	case MAC_STAT_RBYTES:
+		val = gstp->rbytes;
+		break;
+
+	case MAC_STAT_IPACKETS:
+		val = gstp->rpackets;
+		break;
+
+	case MAC_STAT_OBYTES:
+		val = gstp->obytes;
+		break;
+
+	case MAC_STAT_OPACKETS:
+		val = gstp->opackets;
+		break;
+
+	case MAC_STAT_UNDERFLOWS:
+		val = gstp->underflow;
+		break;
+
+	case MAC_STAT_OVERFLOWS:
+		val = gstp->overflow;
+		break;
+
+	case ETHER_STAT_ALIGN_ERRORS:
+		val = gstp->frame;
+		break;
+
+	case ETHER_STAT_FCS_ERRORS:
+		val = gstp->crc;
+		break;
+
+	case ETHER_STAT_FIRST_COLLISIONS:
+		val = gstp->first_coll;
+		break;
+
+	case ETHER_STAT_MULTI_COLLISIONS:
+		val = gstp->multi_coll;
+		break;
+
+	case ETHER_STAT_SQE_ERRORS:
+		val = gstp->sqe;
+		break;
+
+	case ETHER_STAT_DEFER_XMTS:
+		val = gstp->defer;
+		break;
+
+	case ETHER_STAT_TX_LATE_COLLISIONS:
+		val = gstp->xmtlatecoll;
+		break;
+
+	case ETHER_STAT_EX_COLLISIONS:
+		val = gstp->excoll;
+		break;
+
+	case ETHER_STAT_MACXMT_ERRORS:
+		val = gstp->xmit_internal_err;
+		break;
+
+	case ETHER_STAT_CARRIER_ERRORS:
+		val = gstp->nocarrier;
+		break;
+
+	case ETHER_STAT_TOOLONG_ERRORS:
+		val = gstp->frame_too_long;
+		break;
+
+	case ETHER_STAT_MACRCV_ERRORS:
+		val = gstp->rcv_internal_err;
+		break;
+
+	case ETHER_STAT_XCVR_ADDR:
+		val = dp->mii_phy_addr;
+		break;
+
+	case ETHER_STAT_XCVR_ID:
+		val = dp->mii_phy_id;
+		break;
+
+	case ETHER_STAT_XCVR_INUSE:
+		val = usbgem_mac_xcvr_inuse(dp);
+		break;
+
+	case ETHER_STAT_CAP_1000FDX:
+		val = (dp->mii_xstatus & MII_XSTATUS_1000BASET_FD) ||
+		    (dp->mii_xstatus & MII_XSTATUS_1000BASEX_FD);
+		break;
+
+	case ETHER_STAT_CAP_1000HDX:
+		val = (dp->mii_xstatus & MII_XSTATUS_1000BASET) ||
+		    (dp->mii_xstatus & MII_XSTATUS_1000BASEX);
+		break;
+
+	case ETHER_STAT_CAP_100FDX:
+		val = BOOLEAN(dp->mii_status & MII_STATUS_100_BASEX_FD);
+		break;
+
+	case ETHER_STAT_CAP_100HDX:
+		val = BOOLEAN(dp->mii_status & MII_STATUS_100_BASEX);
+		break;
+
+	case ETHER_STAT_CAP_10FDX:
+		val = BOOLEAN(dp->mii_status & MII_STATUS_10_FD);
+		break;
+
+	case ETHER_STAT_CAP_10HDX:
+		val = BOOLEAN(dp->mii_status & MII_STATUS_10);
+		break;
+
+	case ETHER_STAT_CAP_ASMPAUSE:
+		val = dp->ugc.usbgc_flow_control > FLOW_CONTROL_SYMMETRIC;
+		break;
+
+	case ETHER_STAT_CAP_PAUSE:
+		val = dp->ugc.usbgc_flow_control != FLOW_CONTROL_NONE;
+		break;
+
+	case ETHER_STAT_CAP_AUTONEG:
+		val = BOOLEAN(dp->mii_status & MII_STATUS_CANAUTONEG);
+		break;
+
+	case ETHER_STAT_ADV_CAP_1000FDX:
+		val = dp->anadv_1000fdx;
+		break;
+
+	case ETHER_STAT_ADV_CAP_1000HDX:
+		val = dp->anadv_1000hdx;
+		break;
+
+	case ETHER_STAT_ADV_CAP_100FDX:
+		val = dp->anadv_100fdx;
+		break;
+
+	case ETHER_STAT_ADV_CAP_100HDX:
+		val = dp->anadv_100hdx;
+		break;
+
+	case ETHER_STAT_ADV_CAP_10FDX:
+		val = dp->anadv_10fdx;
+		break;
+
+	case ETHER_STAT_ADV_CAP_10HDX:
+		val = dp->anadv_10hdx;
+		break;
+
+	case ETHER_STAT_ADV_CAP_ASMPAUSE:
+		val = dp->anadv_asmpause;
+		break;
+
+	case ETHER_STAT_ADV_CAP_PAUSE:
+		val = dp->anadv_pause;
+		break;
+
+	case ETHER_STAT_ADV_CAP_AUTONEG:
+		val = dp->anadv_autoneg;
+		break;
+
+	case ETHER_STAT_LP_CAP_1000FDX:
+		val = BOOLEAN(dp->mii_stat1000 & MII_1000TS_LP_FULL);
+		break;
+
+	case ETHER_STAT_LP_CAP_1000HDX:
+		val = BOOLEAN(dp->mii_stat1000 & MII_1000TS_LP_HALF);
+		break;
+
+	case ETHER_STAT_LP_CAP_100FDX:
+		val = BOOLEAN(dp->mii_lpable & MII_ABILITY_100BASE_TX_FD);
+		break;
+
+	case ETHER_STAT_LP_CAP_100HDX:
+		val = BOOLEAN(dp->mii_lpable & MII_ABILITY_100BASE_TX);
+		break;
+
+	case ETHER_STAT_LP_CAP_10FDX:
+		val = BOOLEAN(dp->mii_lpable & MII_ABILITY_10BASE_T_FD);
+		break;
+
+	case ETHER_STAT_LP_CAP_10HDX:
+		val = BOOLEAN(dp->mii_lpable & MII_ABILITY_10BASE_T);
+		break;
+
+	case ETHER_STAT_LP_CAP_ASMPAUSE:
+		val = BOOLEAN(dp->mii_lpable & MII_ABILITY_ASM_DIR);
+		break;
+
+	case ETHER_STAT_LP_CAP_PAUSE:
+		val = BOOLEAN(dp->mii_lpable & MII_ABILITY_PAUSE);
+		break;
+
+	case ETHER_STAT_LP_CAP_AUTONEG:
+		val = BOOLEAN(dp->mii_exp & MII_AN_EXP_LPCANAN);
+		break;
+
+	case ETHER_STAT_LINK_ASMPAUSE:
+		val = BOOLEAN(dp->flow_control & 2);
+		break;
+
+	case ETHER_STAT_LINK_PAUSE:
+		val = BOOLEAN(dp->flow_control & 1);
+		break;
+
+	case ETHER_STAT_LINK_AUTONEG:
+		val = dp->anadv_autoneg &&
+		    BOOLEAN(dp->mii_exp & MII_AN_EXP_LPCANAN);
+		break;
+
+	case ETHER_STAT_LINK_DUPLEX:
+		val = (dp->mii_state == MII_STATE_LINKUP) ?
+		    (dp->full_duplex ? 2 : 1) : 0;
+		break;
+
+	case ETHER_STAT_TOOSHORT_ERRORS:
+		val = gstp->runt;
+		break;
+#ifdef NEVER	/* it doesn't make sense */
+	case ETHER_STAT_CAP_REMFAULT:
+		val = B_TRUE;
+		break;
+
+	case ETHER_STAT_ADV_REMFAULT:
+		val = dp->anadv_remfault;
+		break;
+#endif
+	case ETHER_STAT_LP_REMFAULT:
+		val = BOOLEAN(dp->mii_lpable & MII_AN_ADVERT_REMFAULT);
+		break;
+
+	case ETHER_STAT_JABBER_ERRORS:
+		val = gstp->jabber;
+		break;
+
+	case ETHER_STAT_CAP_100T4:
+		val = BOOLEAN(dp->mii_status & MII_STATUS_100_BASE_T4);
+		break;
+
+	case ETHER_STAT_ADV_CAP_100T4:
+		val = dp->anadv_100t4;
+		break;
+
+	case ETHER_STAT_LP_CAP_100T4:
+		val = BOOLEAN(dp->mii_lpable & MII_ABILITY_100BASE_T4);
+		break;
+
+	default:
+#if GEM_DEBUG_LEVEL > 2
+		cmn_err(CE_WARN,
+		    "%s: unrecognized parameter value = %d",
+		    __func__, stat);
+#endif
+		*valp = 0;
+		return (ENOTSUP);
+	}
+
+	*valp = val;
+
+	return (0);
+}
+
+static int
+usbgem_m_unicst(void *arg, const uint8_t *mac)
+{
+	int	err;
+	struct usbgem_dev	*dp = arg;
+
+	DPRINTF(0, (CE_CONT, "!%s: %s: called", dp->name, __func__));
+
+	rw_enter(&dp->dev_state_lock, RW_READER);
+
+	sema_p(&dp->rxfilter_lock);
+	bcopy(mac, dp->cur_addr.ether_addr_octet, ETHERADDRL);
+	dp->rxmode |= RXMODE_ENABLE;
+
+	err = 0;
+	if (dp->mac_state != MAC_STATE_DISCONNECTED) {
+		if (usbgem_hal_set_rx_filter(dp) != USB_SUCCESS) {
+			err = EIO;
+		}
+	}
+	sema_v(&dp->rxfilter_lock);
+	rw_exit(&dp->dev_state_lock);
+
+#ifdef GEM_CONFIG_FMA
+	if (err != 0) {
+		ddi_fm_service_impact(dp->dip, DDI_SERVICE_DEGRADED);
+	}
+#endif
+	return (err);
+}
+
+/*
+ * usbgem_m_tx is used only for sending data packets into ethernet wire.
+ */
+static mblk_t *
+usbgem_m_tx(void *arg, mblk_t *mp_head)
+{
+	int	limit;
+	mblk_t	*mp;
+	mblk_t	*nmp;
+	uint32_t	flags;
+	struct usbgem_dev	*dp = arg;
+
+	DPRINTF(4, (CE_CONT, "!%s: %s: called", dp->name, __func__));
+
+	mp = mp_head;
+	flags = 0;
+
+	rw_enter(&dp->dev_state_lock, RW_READER);
+
+	if (dp->mii_state != MII_STATE_LINKUP ||
+	    dp->mac_state != MAC_STATE_ONLINE) {
+		/* some nics hate to send packets during the link is down */
+		for (; mp; mp = nmp) {
+			nmp = mp->b_next;
+			mp->b_next = NULL;
+			freemsg(mp);
+		}
+		goto x;
+	}
+
+	ASSERT(dp->nic_state == NIC_STATE_ONLINE);
+
+	limit = dp->tx_max_packets;
+	for (; limit-- && mp; mp = nmp) {
+		nmp = mp->b_next;
+		mp->b_next = NULL;
+		if (usbgem_send_common(dp, mp,
+		    (limit == 0 && nmp) ? 1 : 0)) {
+			mp->b_next = nmp;
+			break;
+		}
+	}
+#ifdef CONFIG_TX_LIMITER
+	if (mp == mp_head) {
+		/* no packets were sent, descrease allocation limit */
+		mutex_enter(&dp->txlock);
+		dp->tx_max_packets = max(dp->tx_max_packets - 1, 1);
+		mutex_exit(&dp->txlock);
+	}
+#endif
+x:
+	rw_exit(&dp->dev_state_lock);
+
+	return (mp);
+}
+
+static void
+usbgem_m_ioctl(void *arg, queue_t *wq, mblk_t *mp)
+{
+	struct usbgem_dev	*dp = arg;
+
+	DPRINTF(1, (CE_CONT, "!%s: %s: called",
+	    ((struct usbgem_dev *)arg)->name, __func__));
+
+	rw_enter(&dp->dev_state_lock, RW_READER);
+	usbgem_mac_ioctl((struct usbgem_dev *)arg, wq, mp);
+	rw_exit(&dp->dev_state_lock);
+}
+
+static void
+usbgem_gld3_init(struct usbgem_dev *dp, mac_register_t *macp)
+{
+	macp->m_type_ident = MAC_PLUGIN_IDENT_ETHER;
+	macp->m_driver = dp;
+	macp->m_dip = dp->dip;
+	macp->m_src_addr = dp->dev_addr.ether_addr_octet;
+	macp->m_callbacks = &gem_m_callbacks;
+	macp->m_min_sdu = 0;
+	macp->m_max_sdu = dp->mtu;
+
+	if (dp->misc_flag & USBGEM_VLAN) {
+		macp->m_margin = VTAG_SIZE;
+	}
+}
+#else
+/* ============================================================== */
+/*
+ * GLDv2 interface
+ */
+/* ============================================================== */
+static int usbgem_gld_reset(gld_mac_info_t *);
+static int usbgem_gld_start(gld_mac_info_t *);
+static int usbgem_gld_stop(gld_mac_info_t *);
+static int usbgem_gld_set_mac_address(gld_mac_info_t *, uint8_t *);
+static int usbgem_gld_set_multicast(gld_mac_info_t *, uint8_t *, int);
+static int usbgem_gld_set_promiscuous(gld_mac_info_t *, int);
+static int usbgem_gld_get_stats(gld_mac_info_t *, struct gld_stats *);
+static int usbgem_gld_send(gld_mac_info_t *, mblk_t *);
+static int usbgem_gld_send_tagged(gld_mac_info_t *, mblk_t *, uint32_t);
+
+static int
+usbgem_gld_reset(gld_mac_info_t *macinfo)
+{
+	int	err;
+	struct usbgem_dev	*dp;
+
+	err = GLD_SUCCESS;
+	dp = (struct usbgem_dev *)macinfo->gldm_private;
+
+	DPRINTF(1, (CE_CONT, "!%s: %s: called", dp->name, __func__));
+
+	rw_enter(&dp->dev_state_lock, RW_WRITER);
+	if (usbgem_mac_init(dp) != USB_SUCCESS) {
+		err = GLD_FAILURE;
+		goto x;
+	}
+
+	dp->nic_state = NIC_STATE_INITIALIZED;
+
+	/* setup media mode if the link have been up */
+	if (dp->mii_state == MII_STATE_LINKUP) {
+		if (dp->mac_state != MAC_STATE_DISCONNECTED) {
+			(void) usbgem_hal_set_media(dp);
+		}
+	}
+x:
+	rw_exit(&dp->dev_state_lock);
+	return (err);
+}
+
+static int
+usbgem_gld_start(gld_mac_info_t *macinfo)
+{
+	int	err;
+	struct usbgem_dev *dp;
+
+	dp = (struct usbgem_dev *)macinfo->gldm_private;
+
+	DPRINTF(1, (CE_CONT, "!%s: %s: called", dp->name, __func__));
+
+	rw_enter(&dp->dev_state_lock, RW_WRITER);
+
+	dp->nic_state = NIC_STATE_ONLINE;
+
+	if (dp->mii_state == MII_STATE_LINKUP) {
+		if (usbgem_mac_start(dp) != USB_SUCCESS) {
+			/* sema_v(&dp->mii_lock); */
+			err = GLD_FAILURE;
+			goto x;
+		}
+	}
+
+	/*
+	 * XXX - don't call gld_linkstate() here,
+	 * otherwise it cause recursive mutex call.
+	 */
+	err = GLD_SUCCESS;
+x:
+	rw_exit(&dp->dev_state_lock);
+
+	return (err);
+}
+
+static int
+usbgem_gld_stop(gld_mac_info_t *macinfo)
+{
+	int	err = GLD_SUCCESS;
+	struct usbgem_dev	*dp;
+
+	dp = (struct usbgem_dev *)macinfo->gldm_private;
+
+	DPRINTF(1, (CE_CONT, "!%s: %s: called", dp->name, __func__));
+
+	/* try to stop rx gracefully */
+	rw_enter(&dp->dev_state_lock, RW_READER);
+	sema_p(&dp->rxfilter_lock);
+	dp->rxmode &= ~RXMODE_ENABLE;
+
+	if (dp->mac_state != MAC_STATE_DISCONNECTED) {
+		(void) usbgem_hal_set_rx_filter(dp);
+	}
+	sema_v(&dp->rxfilter_lock);
+	rw_exit(&dp->dev_state_lock);
+
+	/* make the nic state inactive */
+	rw_enter(&dp->dev_state_lock, RW_WRITER);
+	dp->nic_state = NIC_STATE_STOPPED;
+
+	if (dp->mac_state != MAC_STATE_DISCONNECTED) {
+		if (usbgem_mac_stop(dp, MAC_STATE_STOPPED, STOP_GRACEFUL)
+		    != USB_SUCCESS) {
+			err = GLD_FAILURE;
+		}
+	}
+	rw_exit(&dp->dev_state_lock);
+
+	return (err);
+}
+
+static int
+usbgem_gld_set_multicast(gld_mac_info_t *macinfo, uint8_t *ep, int flag)
+{
+	int		err;
+	int		ret;
+	struct usbgem_dev	*dp;
+
+	dp = (struct usbgem_dev *)macinfo->gldm_private;
+
+	DPRINTF(1, (CE_CONT, "!%s: %s: called", dp->name, __func__));
+
+	rw_enter(&dp->dev_state_lock, RW_READER);
+	if (flag == GLD_MULTI_ENABLE) {
+		ret = usbgem_add_multicast(dp, ep);
+	} else {
+		ret = usbgem_remove_multicast(dp, ep);
+	}
+	rw_exit(&dp->dev_state_lock);
+
+	err = GLD_SUCCESS;
+	if (ret != USB_SUCCESS) {
+#ifdef GEM_CONFIG_FMA
+		ddi_fm_service_impact(dp->dip, DDI_SERVICE_DEGRADED);
+#endif
+		err = GLD_FAILURE;
+	}
+	return (err);
+}
+
+static int
+usbgem_gld_set_promiscuous(gld_mac_info_t *macinfo, int flag)
+{
+	boolean_t	need_to_change = B_TRUE;
+	struct usbgem_dev	*dp;
+
+	dp = (struct usbgem_dev *)macinfo->gldm_private;
+
+	DPRINTF(1, (CE_CONT, "!%s: %s: called", dp->name, __func__));
+
+	sema_p(&dp->rxfilter_lock);
+	if (flag == GLD_MAC_PROMISC_NONE) {
+		dp->rxmode &= ~(RXMODE_PROMISC | RXMODE_ALLMULTI_REQ);
+	} else if (flag == GLD_MAC_PROMISC_MULTI) {
+		dp->rxmode |= RXMODE_ALLMULTI_REQ;
+	} else if (flag == GLD_MAC_PROMISC_PHYS) {
+		dp->rxmode |= RXMODE_PROMISC;
+	} else {
+		/* mode unchanged */
+		need_to_change = B_FALSE;
+	}
+
+	if (need_to_change) {
+		if (dp->mac_state != MAC_STATE_DISCONNECTED) {
+			(void) usbgem_hal_set_rx_filter(dp);
+		}
+	}
+	sema_v(&dp->rxfilter_lock);
+
+	return (GLD_SUCCESS);
+}
+
+static int
+usbgem_gld_set_mac_address(gld_mac_info_t *macinfo, uint8_t *mac)
+{
+	struct usbgem_dev	*dp;
+	dp = (struct usbgem_dev *)macinfo->gldm_private;
+
+	DPRINTF(1, (CE_CONT, "!%s: %s: called", dp->name, __func__));
+
+	sema_p(&dp->rxfilter_lock);
+	bcopy(mac, dp->cur_addr.ether_addr_octet, ETHERADDRL);
+	dp->rxmode |= RXMODE_ENABLE;
+
+	if (dp->mac_state != MAC_STATE_DISCONNECTED) {
+		(void) usbgem_hal_set_rx_filter(dp);
+	}
+	sema_v(&dp->rxfilter_lock);
+
+	return (GLD_SUCCESS);
+}
+
+static	int
+usbgem_gld_get_stats(gld_mac_info_t *macinfo, struct gld_stats *gs)
+{
+	struct usbgem_dev	*dp;
+	struct usbgem_stats	*vs;
+
+	dp = (struct usbgem_dev *)macinfo->gldm_private;
+
+	if ((*dp->ugc.usbgc_get_stats)(dp) != USB_SUCCESS) {
+#ifdef GEM_CONFIG_FMA
+		ddi_fm_service_impact(dp->dip, DDI_SERVICE_DEGRADED);
+#endif
+		return (USB_FAILURE);
+	}
+
+	vs = &dp->stats;
+
+	gs->glds_errxmt = vs->errxmt;
+	gs->glds_errrcv = vs->errrcv;
+	gs->glds_collisions = vs->collisions;
+
+	gs->glds_excoll = vs->excoll;
+	gs->glds_defer = vs->defer;
+	gs->glds_frame = vs->frame;
+	gs->glds_crc = vs->crc;
+
+	gs->glds_overflow = vs->overflow; /* fifo err,underrun,rbufovf */
+	gs->glds_underflow = vs->underflow;
+	gs->glds_short = vs->runt;
+	gs->glds_missed = vs->missed; /* missed pkts while rbuf ovf */
+	gs->glds_xmtlatecoll = vs->xmtlatecoll;
+	gs->glds_nocarrier = vs->nocarrier;
+	gs->glds_norcvbuf = vs->norcvbuf;	/* OS resource exaust */
+	gs->glds_intr = vs->intr;
+
+	/* all before here must be kept in place for v0 compatibility */
+	gs->glds_speed = usbgem_speed_value[dp->speed] * 1000000;
+	gs->glds_media = GLDM_PHYMII;
+	gs->glds_duplex = dp->full_duplex ? GLD_DUPLEX_FULL : GLD_DUPLEX_HALF;
+
+	/* gs->glds_media_specific */
+	gs->glds_dot3_first_coll = vs->first_coll;
+	gs->glds_dot3_multi_coll = vs->multi_coll;
+	gs->glds_dot3_sqe_error = 0;
+	gs->glds_dot3_mac_xmt_error = 0;
+	gs->glds_dot3_mac_rcv_error = 0;
+	gs->glds_dot3_frame_too_long = vs->frame_too_long;
+
+	return (GLD_SUCCESS);
+}
+
+static int
+usbgem_gld_ioctl(gld_mac_info_t *macinfo, queue_t *wq, mblk_t *mp)
+{
+	struct usbgem_dev	*dp;
+
+	dp = (struct usbgem_dev *)macinfo->gldm_private;
+	usbgem_mac_ioctl(dp, wq, mp);
+
+	return (GLD_SUCCESS);
+}
+
+/*
+ * gem_gld_send is used only for sending data packets into ethernet wire.
+ */
+static int
+usbgem_gld_send(gld_mac_info_t *macinfo, mblk_t *mp)
+{
+	int		ret;
+	uint32_t	flags = 0;
+	struct usbgem_dev	*dp;
+
+	dp = (struct usbgem_dev *)macinfo->gldm_private;
+
+	/* nic state must be online of suspended */
+	rw_enter(&dp->dev_state_lock, RW_READER);
+
+	ASSERT(dp->nic_state == NIC_STATE_ONLINE);
+	ASSERT(mp->b_next == NULL);
+
+	if (dp->mii_state != MII_STATE_LINKUP) {
+		/* Some nics hate to send packets while the link is down. */
+		/* we discard the untransmitted packets silently */
+		rw_exit(&dp->dev_state_lock);
+
+		freemsg(mp);
+#ifdef GEM_CONFIG_FMA
+		/* FIXME - should we ignore the error? */
+		ddi_fm_service_impact(dp->dip, DDI_SERVICE_DEGRADED);
+#endif
+		return (GLD_SUCCESS);
+	}
+
+	ret = (usbgem_send_common(dp, mp, flags) == NULL)
+	    ? GLD_SUCCESS : GLD_NORESOURCES;
+	rw_exit(&dp->dev_state_lock);
+
+	return (ret);
+}
+
+/*
+ * usbgem_gld_send is used only for sending data packets into ethernet wire.
+ */
+static int
+usbgem_gld_send_tagged(gld_mac_info_t *macinfo, mblk_t *mp, uint32_t vtag)
+{
+	uint32_t	flags;
+	struct usbgem_dev	*dp;
+
+	dp = (struct usbgem_dev *)macinfo->gldm_private;
+
+	/*
+	 * Some nics hate to send packets while the link is down.
+	 */
+	if (dp->mii_state != MII_STATE_LINKUP) {
+		/* we dicard the untransmitted packets silently */
+		freemsg(mp);
+#ifdef GEM_CONFIG_FMA
+		/* FIXME - should we ignore the error? */
+		ddi_fm_service_impact(dp->dip, DDI_SERVICE_UNAFFECTED);
+#endif
+		return (GLD_SUCCESS);
+	}
+#ifdef notyet
+	flags = GLD_VTAG_TCI(vtag) << GEM_SEND_VTAG_SHIFT;
+#endif
+	return ((usbgem_send_common(dp, mp, 0) == NULL) ?
+	    GLD_SUCCESS : GLD_NORESOURCES);
+}
+
+static void
+usbgem_gld_init(struct usbgem_dev *dp, gld_mac_info_t *macinfo, char *ident)
+{
+	/*
+	 * configure GLD
+	 */
+	macinfo->gldm_devinfo = dp->dip;
+	macinfo->gldm_private = (caddr_t)dp;
+
+	macinfo->gldm_reset = usbgem_gld_reset;
+	macinfo->gldm_start = usbgem_gld_start;
+	macinfo->gldm_stop = usbgem_gld_stop;
+	macinfo->gldm_set_mac_addr = usbgem_gld_set_mac_address;
+	macinfo->gldm_send = usbgem_gld_send;
+	macinfo->gldm_set_promiscuous = usbgem_gld_set_promiscuous;
+	macinfo->gldm_get_stats = usbgem_gld_get_stats;
+	macinfo->gldm_ioctl = usbgem_gld_ioctl;
+	macinfo->gldm_set_multicast = usbgem_gld_set_multicast;
+	macinfo->gldm_intr = NULL;
+	macinfo->gldm_mctl = NULL;
+
+	macinfo->gldm_ident = ident;
+	macinfo->gldm_type = DL_ETHER;
+	macinfo->gldm_minpkt = 0;
+	macinfo->gldm_maxpkt = dp->mtu;
+	macinfo->gldm_addrlen = ETHERADDRL;
+	macinfo->gldm_saplen = -2;
+	macinfo->gldm_ppa = ddi_get_instance(dp->dip);
+#ifdef GLD_CAP_LINKSTATE
+	macinfo->gldm_capabilities = GLD_CAP_LINKSTATE;
+#endif
+	macinfo->gldm_vendor_addr = dp->dev_addr.ether_addr_octet;
+	macinfo->gldm_broadcast_addr = usbgem_bcastaddr;
+}
+#endif /* USBGEM_CONFIG_GLDv3 */
+
+
+/* ======================================================================== */
+/*
+ * .conf interface
+ */
+/* ======================================================================== */
+void
+usbgem_generate_macaddr(struct usbgem_dev *dp, uint8_t *mac)
+{
+	extern char	hw_serial[];
+	char		*hw_serial_p;
+	int		i;
+	uint64_t	val;
+	uint64_t	key;
+
+	cmn_err(CE_NOTE,
+	    "!%s: using temp ether address,"
+	    " do not use this for long time",
+	    dp->name);
+
+	/* prefer a fixed address for DHCP */
+	hw_serial_p = &hw_serial[0];
+	val = stoi(&hw_serial_p);
+
+	key = 0;
+	for (i = 0; i < USBGEM_NAME_LEN; i++) {
+		if (dp->name[i] == 0) {
+			break;
+		}
+		key ^= dp->name[i];
+	}
+	key ^= ddi_get_instance(dp->dip);
+	val ^= key << 32;
+
+	/* generate a local address */
+	mac[0] = 0x02;
+	mac[1] = (uint8_t)(val >> 32);
+	mac[2] = (uint8_t)(val >> 24);
+	mac[3] = (uint8_t)(val >> 16);
+	mac[4] = (uint8_t)(val >> 8);
+	mac[5] = (uint8_t)val;
+}
+
+boolean_t
+usbgem_get_mac_addr_conf(struct usbgem_dev *dp)
+{
+	char		propname[32];
+	char		*valstr;
+	uint8_t		mac[ETHERADDRL];
+	char		*cp;
+	int		c;
+	int		i;
+	int		j;
+	uint8_t		v;
+	uint8_t		d;
+	uint8_t		ored;
+
+	DPRINTF(3, (CE_CONT, "!%s: %s: called", dp->name, __func__));
+	/*
+	 * Get ethernet address from .conf file
+	 */
+	(void) sprintf(propname, "mac-addr");
+	if ((ddi_prop_lookup_string(DDI_DEV_T_ANY, dp->dip,
+	    DDI_PROP_DONTPASS, propname, &valstr)) != DDI_PROP_SUCCESS) {
+		return (B_FALSE);
+	}
+
+	if (strlen(valstr) != ETHERADDRL*3-1) {
+		goto syntax_err;
+	}
+
+	cp = valstr;
+	j = 0;
+	ored = 0;
+	for (;;) {
+		v = 0;
+		for (i = 0; i < 2; i++) {
+			c = *cp++;
+
+			if (c >= 'a' && c <= 'f') {
+				d = c - 'a' + 10;
+			} else if (c >= 'A' && c <= 'F') {
+				d = c - 'A' + 10;
+			} else if (c >= '0' && c <= '9') {
+				d = c - '0';
+			} else {
+				goto syntax_err;
+			}
+			v = (v << 4) | d;
+		}
+
+		mac[j++] = v;
+		ored |= v;
+		if (j == ETHERADDRL) {
+			/* done */
+			break;
+		}
+
+		c = *cp++;
+		if (c != ':') {
+			goto syntax_err;
+		}
+	}
+
+	if (ored == 0) {
+		usbgem_generate_macaddr(dp, mac);
+	}
+	for (i = 0; i < ETHERADDRL; i++) {
+		dp->dev_addr.ether_addr_octet[i] = mac[i];
+	}
+	ddi_prop_free(valstr);
+	return (B_TRUE);
+
+syntax_err:
+	cmn_err(CE_CONT,
+	    "!%s: read mac addr: trying .conf: syntax err %s",
+	    dp->name, valstr);
+	ddi_prop_free(valstr);
+
+	return (B_FALSE);
+}
+
+static void
+usbgem_read_conf(struct usbgem_dev *dp)
+{
+	int	val;
+
+	DPRINTF(1, (CE_CONT, "!%s: %s: called", dp->name, __func__));
+
+	/*
+	 * Get media mode infomation from .conf file
+	 */
+	dp->anadv_autoneg = usbgem_prop_get_int(dp, "adv_autoneg_cap", 1) != 0;
+	dp->anadv_1000fdx = usbgem_prop_get_int(dp, "adv_1000fdx_cap", 1) != 0;
+	dp->anadv_1000hdx = usbgem_prop_get_int(dp, "adv_1000hdx_cap", 1) != 0;
+	dp->anadv_100t4 = usbgem_prop_get_int(dp, "adv_100T4_cap", 1) != 0;
+	dp->anadv_100fdx = usbgem_prop_get_int(dp, "adv_100fdx_cap", 1) != 0;
+	dp->anadv_100hdx = usbgem_prop_get_int(dp, "adv_100hdx_cap", 1) != 0;
+	dp->anadv_10fdx = usbgem_prop_get_int(dp, "adv_10fdx_cap", 1) != 0;
+	dp->anadv_10hdx = usbgem_prop_get_int(dp, "adv_10hdx_cap", 1) != 0;
+	dp->anadv_1000t_ms = usbgem_prop_get_int(dp, "adv_1000t_ms", 0);
+
+	if ((ddi_prop_exists(DDI_DEV_T_ANY, dp->dip,
+	    DDI_PROP_DONTPASS, "full-duplex"))) {
+		dp->full_duplex =
+		    usbgem_prop_get_int(dp, "full-duplex", 1) != 0;
+		dp->anadv_autoneg = B_FALSE;
+		if (dp->full_duplex) {
+			dp->anadv_1000hdx = B_FALSE;
+			dp->anadv_100hdx = B_FALSE;
+			dp->anadv_10hdx = B_FALSE;
+		} else {
+			dp->anadv_1000fdx = B_FALSE;
+			dp->anadv_100fdx = B_FALSE;
+			dp->anadv_10fdx = B_FALSE;
+		}
+	}
+
+	if ((val = usbgem_prop_get_int(dp, "speed", 0)) > 0) {
+		dp->anadv_autoneg = B_FALSE;
+		switch (val) {
+		case 1000:
+			dp->speed = USBGEM_SPD_1000;
+			dp->anadv_100t4 = B_FALSE;
+			dp->anadv_100fdx = B_FALSE;
+			dp->anadv_100hdx = B_FALSE;
+			dp->anadv_10fdx = B_FALSE;
+			dp->anadv_10hdx = B_FALSE;
+			break;
+		case 100:
+			dp->speed = USBGEM_SPD_100;
+			dp->anadv_1000fdx = B_FALSE;
+			dp->anadv_1000hdx = B_FALSE;
+			dp->anadv_10fdx = B_FALSE;
+			dp->anadv_10hdx = B_FALSE;
+			break;
+		case 10:
+			dp->speed = USBGEM_SPD_10;
+			dp->anadv_1000fdx = B_FALSE;
+			dp->anadv_1000hdx = B_FALSE;
+			dp->anadv_100t4 = B_FALSE;
+			dp->anadv_100fdx = B_FALSE;
+			dp->anadv_100hdx = B_FALSE;
+			break;
+		default:
+			cmn_err(CE_WARN,
+			    "!%s: property %s: illegal value:%d",
+			    dp->name, "speed", val);
+			dp->anadv_autoneg = B_TRUE;
+			break;
+		}
+	}
+	val = usbgem_prop_get_int(dp,
+	    "adv_pause", dp->ugc.usbgc_flow_control & 1);
+	val |= usbgem_prop_get_int(dp,
+	    "adv_asmpause", BOOLEAN(dp->ugc.usbgc_flow_control & 2)) << 1;
+	if (val > FLOW_CONTROL_RX_PAUSE || val < FLOW_CONTROL_NONE) {
+		cmn_err(CE_WARN,
+		    "!%s: property %s: illegal value:%d",
+		    dp->name, "flow-control", val);
+	} else {
+		val = min(val, dp->ugc.usbgc_flow_control);
+	}
+	dp->anadv_pause = BOOLEAN(val & 1);
+	dp->anadv_asmpause = BOOLEAN(val & 2);
+
+	dp->mtu = usbgem_prop_get_int(dp, "mtu", dp->mtu);
+	dp->txthr = usbgem_prop_get_int(dp, "txthr", dp->txthr);
+	dp->rxthr = usbgem_prop_get_int(dp, "rxthr", dp->rxthr);
+	dp->txmaxdma = usbgem_prop_get_int(dp, "txmaxdma", dp->txmaxdma);
+	dp->rxmaxdma = usbgem_prop_get_int(dp, "rxmaxdma", dp->rxmaxdma);
+#ifdef GEM_CONFIG_POLLING
+	dp->poll_pkt_delay =
+	    usbgem_prop_get_int(dp, "pkt_delay", dp->poll_pkt_delay);
+
+	dp->max_poll_interval[GEM_SPD_10] =
+	    usbgem_prop_get_int(dp, "max_poll_interval_10",
+	    dp->max_poll_interval[GEM_SPD_10]);
+	dp->max_poll_interval[GEM_SPD_100] =
+	    usbgem_prop_get_int(dp, "max_poll_interval_100",
+	    dp->max_poll_interval[GEM_SPD_100]);
+	dp->max_poll_interval[GEM_SPD_1000] =
+	    usbgem_prop_get_int(dp, "max_poll_interval_1000",
+	    dp->max_poll_interval[GEM_SPD_1000]);
+
+	dp->min_poll_interval[GEM_SPD_10] =
+	    usbgem_prop_get_int(dp, "min_poll_interval_10",
+	    dp->min_poll_interval[GEM_SPD_10]);
+	dp->min_poll_interval[GEM_SPD_100] =
+	    usbgem_prop_get_int(dp, "min_poll_interval_100",
+	    dp->min_poll_interval[GEM_SPD_100]);
+	dp->min_poll_interval[GEM_SPD_1000] =
+	    usbgem_prop_get_int(dp, "min_poll_interval_1000",
+	    dp->min_poll_interval[GEM_SPD_1000]);
+#endif
+}
+
+/*
+ * usbem kstat support
+ */
+#ifndef GEM_CONFIG_GLDv3
+/* kstat items based from dmfe driver */
+
+struct usbgem_kstat_named {
+	struct kstat_named	ks_xcvr_addr;
+	struct kstat_named	ks_xcvr_id;
+	struct kstat_named	ks_xcvr_inuse;
+	struct kstat_named	ks_link_up;
+	struct kstat_named	ks_link_duplex;	/* 0:unknwon, 1:half, 2:full */
+	struct kstat_named	ks_cap_1000fdx;
+	struct kstat_named	ks_cap_1000hdx;
+	struct kstat_named	ks_cap_100fdx;
+	struct kstat_named	ks_cap_100hdx;
+	struct kstat_named	ks_cap_10fdx;
+	struct kstat_named	ks_cap_10hdx;
+#ifdef NEVER
+	struct kstat_named	ks_cap_remfault;
+#endif
+	struct kstat_named	ks_cap_autoneg;
+
+	struct kstat_named	ks_adv_cap_1000fdx;
+	struct kstat_named	ks_adv_cap_1000hdx;
+	struct kstat_named	ks_adv_cap_100fdx;
+	struct kstat_named	ks_adv_cap_100hdx;
+	struct kstat_named	ks_adv_cap_10fdx;
+	struct kstat_named	ks_adv_cap_10hdx;
+#ifdef NEVER
+	struct kstat_named	ks_adv_cap_remfault;
+#endif
+	struct kstat_named	ks_adv_cap_autoneg;
+	struct kstat_named	ks_lp_cap_1000fdx;
+	struct kstat_named	ks_lp_cap_1000hdx;
+	struct kstat_named	ks_lp_cap_100fdx;
+	struct kstat_named	ks_lp_cap_100hdx;
+	struct kstat_named	ks_lp_cap_10fdx;
+	struct kstat_named	ks_lp_cap_10hdx;
+	struct kstat_named	ks_lp_cap_remfault;
+	struct kstat_named	ks_lp_cap_autoneg;
+};
+
+static int
+usbgem_kstat_update(kstat_t *ksp, int rw)
+{
+	struct usbgem_kstat_named *knp;
+	struct usbgem_dev *dp = (struct usbgem_dev *)ksp->ks_private;
+
+	if (rw != KSTAT_READ) {
+		return (0);
+	}
+
+	knp = (struct usbgem_kstat_named *)ksp->ks_data;
+
+	knp->ks_xcvr_addr.value.ul = dp->mii_phy_addr;
+	knp->ks_xcvr_id.value.ul = dp->mii_phy_id;
+	knp->ks_xcvr_inuse.value.ul = usbgem_mac_xcvr_inuse(dp);
+	knp->ks_link_up.value.ul = dp->mii_state == MII_STATE_LINKUP;
+	knp->ks_link_duplex.value.ul =
+	    (dp->mii_state == MII_STATE_LINKUP) ?
+	    (dp->full_duplex ? 2 : 1) : 0;
+
+	knp->ks_cap_1000fdx.value.ul =
+	    (dp->mii_xstatus & MII_XSTATUS_1000BASET_FD) ||
+	    (dp->mii_xstatus & MII_XSTATUS_1000BASEX_FD);
+	knp->ks_cap_1000hdx.value.ul =
+	    (dp->mii_xstatus & MII_XSTATUS_1000BASET) ||
+	    (dp->mii_xstatus & MII_XSTATUS_1000BASEX);
+	knp->ks_cap_100fdx.value.ul =
+	    BOOLEAN(dp->mii_status & MII_STATUS_100_BASEX_FD);
+	knp->ks_cap_100hdx.value.ul =
+	    BOOLEAN(dp->mii_status & MII_STATUS_100_BASEX);
+	knp->ks_cap_10fdx.value.ul =
+	    BOOLEAN(dp->mii_status & MII_STATUS_10_FD);
+	knp->ks_cap_10hdx.value.ul =
+	    BOOLEAN(dp->mii_status & MII_STATUS_10);
+#ifdef NEVER
+	knp->ks_cap_remfault.value.ul = B_TRUE;
+#endif
+	knp->ks_cap_autoneg.value.ul =
+	    BOOLEAN(dp->mii_status & MII_STATUS_CANAUTONEG);
+
+	knp->ks_adv_cap_1000fdx.value.ul = dp->anadv_1000fdx;
+	knp->ks_adv_cap_1000hdx.value.ul = dp->anadv_1000hdx;
+	knp->ks_adv_cap_100fdx.value.ul	= dp->anadv_100fdx;
+	knp->ks_adv_cap_100hdx.value.ul	= dp->anadv_100hdx;
+	knp->ks_adv_cap_10fdx.value.ul	= dp->anadv_10fdx;
+	knp->ks_adv_cap_10hdx.value.ul	= dp->anadv_10hdx;
+#ifdef NEVER
+	knp->ks_adv_cap_remfault.value.ul = 0;
+#endif
+	knp->ks_adv_cap_autoneg.value.ul = dp->anadv_autoneg;
+
+	knp->ks_lp_cap_1000fdx.value.ul =
+	    BOOLEAN(dp->mii_stat1000 & MII_1000TS_LP_FULL);
+	knp->ks_lp_cap_1000hdx.value.ul =
+	    BOOLEAN(dp->mii_stat1000 & MII_1000TS_LP_HALF);
+	knp->ks_lp_cap_100fdx.value.ul =
+	    BOOLEAN(dp->mii_lpable & MII_ABILITY_100BASE_TX_FD);
+	knp->ks_lp_cap_100hdx.value.ul =
+	    BOOLEAN(dp->mii_lpable & MII_ABILITY_100BASE_TX);
+	knp->ks_lp_cap_10fdx.value.ul =
+	    BOOLEAN(dp->mii_lpable & MII_ABILITY_10BASE_T_FD);
+	knp->ks_lp_cap_10hdx.value.ul =
+	    BOOLEAN(dp->mii_lpable & MII_ABILITY_10BASE_T);
+	knp->ks_lp_cap_remfault.value.ul =
+	    BOOLEAN(dp->mii_exp & MII_AN_EXP_PARFAULT);
+	knp->ks_lp_cap_autoneg.value.ul =
+	    BOOLEAN(dp->mii_exp & MII_AN_EXP_LPCANAN);
+
+	return (0);
+}
+
+
+static int
+usbgem_kstat_init(struct usbgem_dev *dp)
+{
+	int			i;
+	kstat_t			*ksp;
+	struct usbgem_kstat_named	*knp;
+
+	ksp = kstat_create(
+	    (char *)ddi_driver_name(dp->dip), ddi_get_instance(dp->dip),
+	    "mii", "net", KSTAT_TYPE_NAMED,
+	    sizeof (*knp) / sizeof (knp->ks_xcvr_addr), 0);
+
+	if (ksp == NULL) {
+		cmn_err(CE_WARN, "%s: %s() for mii failed",
+		    dp->name, __func__);
+		return (USB_FAILURE);
+	}
+
+	knp = (struct usbgem_kstat_named *)ksp->ks_data;
+
+	kstat_named_init(&knp->ks_xcvr_addr, "xcvr_addr",
+	    KSTAT_DATA_INT32);
+	kstat_named_init(&knp->ks_xcvr_id, "xcvr_id",
+	    KSTAT_DATA_UINT32);
+	kstat_named_init(&knp->ks_xcvr_inuse, "xcvr_inuse",
+	    KSTAT_DATA_UINT32);
+	kstat_named_init(&knp->ks_link_up, "link_up",
+	    KSTAT_DATA_UINT32);
+	kstat_named_init(&knp->ks_link_duplex, "link_duplex",
+	    KSTAT_DATA_UINT32);
+	kstat_named_init(&knp->ks_cap_1000fdx, "cap_1000fdx",
+	    KSTAT_DATA_UINT32);
+	kstat_named_init(&knp->ks_cap_1000hdx, "cap_1000hdx",
+	    KSTAT_DATA_UINT32);
+	kstat_named_init(&knp->ks_cap_100fdx, "cap_100fdx",
+	    KSTAT_DATA_UINT32);
+	kstat_named_init(&knp->ks_cap_100hdx, "cap_100hdx",
+	    KSTAT_DATA_UINT32);
+	kstat_named_init(&knp->ks_cap_10fdx, "cap_10fdx",
+	    KSTAT_DATA_UINT32);
+	kstat_named_init(&knp->ks_cap_10hdx, "cap_10hdx",
+	    KSTAT_DATA_UINT32);
+#ifdef NEVER
+	kstat_named_init(&knp->ks_cap_remfault, "cap_rem_fault",
+	    KSTAT_DATA_UINT32);
+#endif
+	kstat_named_init(&knp->ks_cap_autoneg, "cap_autoneg",
+	    KSTAT_DATA_UINT32);
+	kstat_named_init(&knp->ks_adv_cap_1000fdx, "adv_cap_1000fdx",
+	    KSTAT_DATA_UINT32);
+	kstat_named_init(&knp->ks_adv_cap_1000hdx, "adv_cap_1000hdx",
+	    KSTAT_DATA_UINT32);
+	kstat_named_init(&knp->ks_adv_cap_100fdx, "adv_cap_100fdx",
+	    KSTAT_DATA_UINT32);
+	kstat_named_init(&knp->ks_adv_cap_100hdx, "adv_cap_100hdx",
+	    KSTAT_DATA_UINT32);
+	kstat_named_init(&knp->ks_adv_cap_10fdx, "adv_cap_10fdx",
+	    KSTAT_DATA_UINT32);
+	kstat_named_init(&knp->ks_adv_cap_10hdx, "adv_cap_10hdx",
+	    KSTAT_DATA_UINT32);
+#ifdef NEVER
+	kstat_named_init(&knp->ks_adv_cap_remfault, "adv_rem_fault",
+	    KSTAT_DATA_UINT32);
+#endif
+	kstat_named_init(&knp->ks_adv_cap_autoneg, "adv_cap_autoneg",
+	    KSTAT_DATA_UINT32);
+
+	kstat_named_init(&knp->ks_lp_cap_1000fdx, "lp_cap_1000fdx",
+	    KSTAT_DATA_UINT32);
+	kstat_named_init(&knp->ks_lp_cap_1000hdx, "lp_cap_1000hdx",
+	    KSTAT_DATA_UINT32);
+	kstat_named_init(&knp->ks_lp_cap_100fdx, "lp_cap_100fdx",
+	    KSTAT_DATA_UINT32);
+	kstat_named_init(&knp->ks_lp_cap_100hdx, "lp_cap_100hdx",
+	    KSTAT_DATA_UINT32);
+	kstat_named_init(&knp->ks_lp_cap_10fdx, "lp_cap_10fdx",
+	    KSTAT_DATA_UINT32);
+	kstat_named_init(&knp->ks_lp_cap_10hdx, "lp_cap_10hdx",
+	    KSTAT_DATA_UINT32);
+	kstat_named_init(&knp->ks_lp_cap_remfault, "lp_cap_rem_fault",
+	    KSTAT_DATA_UINT32);
+	kstat_named_init(&knp->ks_lp_cap_autoneg, "lp_cap_autoneg",
+	    KSTAT_DATA_UINT32);
+
+	ksp->ks_private = (void *) dp;
+	ksp->ks_update = usbgem_kstat_update;
+	dp->ksp = ksp;
+
+	kstat_install(ksp);
+
+	return (USB_SUCCESS);
+}
+#endif /* GEM_CONFIG_GLDv3 */
+/* ======================================================================== */
+/*
+ * attach/detatch/usb support
+ */
+/* ======================================================================== */
+int
+usbgem_ctrl_out(struct usbgem_dev *dp,
+	uint8_t reqt, uint8_t req, uint16_t val, uint16_t ix, uint16_t len,
+	void *bp, int size)
+{
+	mblk_t			*data;
+	usb_ctrl_setup_t	setup;
+	usb_cr_t		completion_reason;
+	usb_cb_flags_t		cb_flags;
+	usb_flags_t		flags;
+	int			i;
+	int			ret;
+
+	DPRINTF(4, (CE_CONT, "!%s: %s "
+	    "reqt:0x%02x req:0x%02x val:0x%04x ix:0x%04x len:0x%02x "
+	    "bp:0x%p nic_state:%d",
+	    dp->name, __func__, reqt, req, val, ix, len, bp, dp->nic_state));
+
+	if (dp->mac_state == MAC_STATE_DISCONNECTED) {
+		return (USB_PIPE_ERROR);
+	}
+
+	data = NULL;
+	if (size > 0) {
+		if ((data = allocb(size, 0)) == NULL) {
+			return (USB_FAILURE);
+		}
+
+		bcopy(bp, data->b_rptr, size);
+		data->b_wptr = data->b_rptr + size;
+	}
+
+	setup.bmRequestType = reqt;
+	setup.bRequest = req;
+	setup.wValue = val;
+	setup.wIndex = ix;
+	setup.wLength = len;
+	setup.attrs = 0;	/* attributes */
+
+	for (i = usbgem_ctrl_retry; i > 0; i--) {
+		completion_reason = 0;
+		cb_flags = 0;
+
+		ret = usb_pipe_ctrl_xfer_wait(DEFAULT_PIPE(dp),
+		    &setup, &data, &completion_reason, &cb_flags, 0);
+
+		if (ret == USB_SUCCESS) {
+			break;
+		}
+		if (i == 1) {
+			cmn_err(CE_WARN,
+			    "!%s: %s failed: "
+			    "reqt:0x%x req:0x%x val:0x%x ix:0x%x len:0x%x "
+			    "ret:%d cr:%s(%d), cb_flags:0x%x %s",
+			    dp->name, __func__, reqt, req, val, ix, len,
+			    ret, usb_str_cr(completion_reason),
+			    completion_reason,
+			    cb_flags,
+			    (i > 1) ? "retrying..." : "fatal");
+		}
+	}
+
+	if (data != NULL) {
+		freemsg(data);
+	}
+
+	return (ret);
+}
+
+int
+usbgem_ctrl_in(struct usbgem_dev *dp,
+	uint8_t reqt, uint8_t req, uint16_t val, uint16_t ix, uint16_t len,
+	void *bp, int size)
+{
+	mblk_t			*data;
+	usb_ctrl_setup_t	setup;
+	usb_cr_t		completion_reason;
+	usb_cb_flags_t		cb_flags;
+	int			i;
+	int			ret;
+	int			reclen;
+
+	DPRINTF(4, (CE_CONT,
+	    "!%s: %s:"
+	    " reqt:0x%02x req:0x%02x val:0x%04x ix:0x%04x len:0x%02x"
+	    " bp:x%p mac_state:%d",
+	    dp->name, __func__, reqt, req, val, ix, len, bp, dp->mac_state));
+
+	if (dp->mac_state == MAC_STATE_DISCONNECTED) {
+		return (USB_PIPE_ERROR);
+	}
+
+	data = NULL;
+
+	setup.bmRequestType = reqt;
+	setup.bRequest = req;
+	setup.wValue = val;
+	setup.wIndex = ix;
+	setup.wLength = len;
+	setup.attrs = USB_ATTRS_AUTOCLEARING;	/* XXX */
+
+	for (i = usbgem_ctrl_retry; i > 0; i--) {
+		completion_reason = 0;
+		cb_flags = 0;
+		ret = usb_pipe_ctrl_xfer_wait(DEFAULT_PIPE(dp), &setup, &data,
+		    &completion_reason, &cb_flags, 0);
+
+		if (ret == USB_SUCCESS) {
+			reclen = msgdsize(data);
+			bcopy(data->b_rptr, bp, min(reclen, size));
+			break;
+		}
+		if (i == 1) {
+			cmn_err(CE_WARN,
+			    "!%s: %s failed: "
+			    "reqt:0x%x req:0x%x val:0x%x ix:0x%x len:0x%x "
+			    "ret:%d cr:%s(%d) cb_flags:0x%x %s",
+			    dp->name, __func__,
+			    reqt, req, val, ix, len,
+			    ret, usb_str_cr(completion_reason),
+			    completion_reason,
+			    cb_flags,
+			    (i > 1) ? "retrying..." : "fatal");
+		}
+	}
+
+	if (data) {
+		freemsg(data);
+	}
+
+	return (ret);
+}
+
+int
+usbgem_ctrl_out_val(struct usbgem_dev *dp,
+    uint8_t reqt, uint8_t req, uint16_t val, uint16_t ix, uint16_t len,
+    uint32_t v)
+{
+	uint8_t	buf[4];
+
+	/* convert to little endian from native byte order */
+	switch (len) {
+	case 4:
+		buf[3] = v >> 24;
+		buf[2] = v >> 16;
+		/* fall thru */
+	case 2:
+		buf[1] = v >> 8;
+		/* fall thru */
+	case 1:
+		buf[0] = v;
+	}
+
+	return (usbgem_ctrl_out(dp, reqt, req, val, ix, len, buf, len));
+}
+
+int
+usbgem_ctrl_in_val(struct usbgem_dev *dp,
+    uint8_t reqt, uint8_t req, uint16_t val, uint16_t ix, uint16_t len,
+    void *valp)
+{
+	uint8_t		buf[4];
+	uint_t		v;
+	int		err;
+
+#ifdef SANITY
+	bzero(buf, sizeof (buf));
+#endif
+	err = usbgem_ctrl_in(dp, reqt, req, val, ix, len, buf, len);
+	if (err == USB_SUCCESS) {
+		v = 0;
+		switch (len) {
+		case 4:
+			v |= buf[3] << 24;
+			v |= buf[2] << 16;
+			/* FALLTHROUGH */
+		case 2:
+			v |= buf[1] << 8;
+			/* FALLTHROUGH */
+		case 1:
+			v |= buf[0];
+		}
+
+		switch (len) {
+		case 4:
+			*(uint32_t *)valp = v;
+			break;
+		case 2:
+			*(uint16_t *)valp = v;
+			break;
+		case 1:
+			*(uint8_t *)valp = v;
+			break;
+		}
+	}
+	return (err);
+}
+
+/*
+ * Attach / detach / disconnect / reconnect management
+ */
+static int
+usbgem_open_pipes(struct usbgem_dev *dp)
+{
+	int			i;
+	int			ret;
+	int			ifnum;
+	int			alt;
+	usb_client_dev_data_t	*reg_data;
+	usb_ep_data_t		*ep_tree_node;
+
+	DPRINTF(0, (CE_CONT, "!%s: %s: called", dp->name, __func__));
+
+	ifnum = dp->ugc.usbgc_ifnum;
+	alt = dp->ugc.usbgc_alt;
+
+	ep_tree_node = usb_lookup_ep_data(dp->dip, dp->reg_data, ifnum, alt,
+	    0, USB_EP_ATTR_BULK, USB_EP_DIR_IN);
+	if (ep_tree_node == NULL) {
+		cmn_err(CE_WARN, "!%s: %s: ep_bulkin is NULL",
+		    dp->name, __func__);
+		goto err;
+	}
+	dp->ep_bulkin = &ep_tree_node->ep_descr;
+
+	ep_tree_node = usb_lookup_ep_data(dp->dip, dp->reg_data, ifnum, alt,
+	    0, USB_EP_ATTR_BULK, USB_EP_DIR_OUT);
+	if (ep_tree_node == NULL) {
+		cmn_err(CE_WARN, "!%s: %s: ep_bulkout is NULL",
+		    dp->name, __func__);
+		goto err;
+	}
+	dp->ep_bulkout = &ep_tree_node->ep_descr;
+
+	ep_tree_node = usb_lookup_ep_data(dp->dip, dp->reg_data, ifnum, alt,
+	    0, USB_EP_ATTR_INTR, USB_EP_DIR_IN);
+	if (ep_tree_node) {
+		dp->ep_intr = &ep_tree_node->ep_descr;
+	} else {
+		/* don't care */
+		DPRINTF(1, (CE_CONT, "!%s: %s: ep_intr is NULL",
+		    dp->name, __func__));
+		dp->ep_intr = NULL;
+	}
+
+	/* XXX -- no need to open default pipe */
+
+	/* open bulk out pipe */
+	bzero(&dp->policy_bulkout, sizeof (usb_pipe_policy_t));
+	dp->policy_bulkout.pp_max_async_reqs = 1;
+
+	if ((ret = usb_pipe_open(dp->dip,
+	    dp->ep_bulkout, &dp->policy_bulkout, USB_FLAGS_SLEEP,
+	    &dp->bulkout_pipe)) != USB_SUCCESS) {
+		cmn_err(CE_WARN,
+		    "!%s: %s: err:%x: failed to open bulk-out pipe",
+		    dp->name, __func__, ret);
+		dp->bulkout_pipe = NULL;
+		goto err;
+	}
+	DPRINTF(1, (CE_CONT, "!%s: %s: bulkout_pipe opened successfully",
+	    dp->name, __func__));
+
+	/* open bulk in pipe */
+	bzero(&dp->policy_bulkin, sizeof (usb_pipe_policy_t));
+	dp->policy_bulkin.pp_max_async_reqs = 1;
+	if ((ret = usb_pipe_open(dp->dip,
+	    dp->ep_bulkin, &dp->policy_bulkin, USB_FLAGS_SLEEP,
+	    &dp->bulkin_pipe)) != USB_SUCCESS) {
+		cmn_err(CE_WARN,
+		    "!%s: %s: ret:%x failed to open bulk-in pipe",
+		    dp->name, __func__, ret);
+		dp->bulkin_pipe = NULL;
+		goto err;
+	}
+	DPRINTF(1, (CE_CONT, "!%s: %s: bulkin_pipe opened successfully",
+	    dp->name, __func__));
+
+	if (dp->ep_intr) {
+		/* open interrupt pipe */
+		bzero(&dp->policy_interrupt, sizeof (usb_pipe_policy_t));
+		dp->policy_interrupt.pp_max_async_reqs = 1;
+		if ((ret = usb_pipe_open(dp->dip, dp->ep_intr,
+		    &dp->policy_interrupt, USB_FLAGS_SLEEP,
+		    &dp->intr_pipe)) != USB_SUCCESS) {
+			cmn_err(CE_WARN,
+			    "!%s: %s: ret:%x failed to open interrupt pipe",
+			    dp->name, __func__, ret);
+			dp->intr_pipe = NULL;
+			goto err;
+		}
+	}
+	DPRINTF(1, (CE_CONT, "!%s: %s: intr_pipe opened successfully",
+	    dp->name, __func__));
+
+	return (USB_SUCCESS);
+
+err:
+	if (dp->bulkin_pipe) {
+		usb_pipe_close(dp->dip,
+		    dp->bulkin_pipe, USB_FLAGS_SLEEP, NULL, 0);
+		dp->bulkin_pipe = NULL;
+	}
+	if (dp->bulkout_pipe) {
+		usb_pipe_close(dp->dip,
+		    dp->bulkout_pipe, USB_FLAGS_SLEEP, NULL, 0);
+		dp->bulkout_pipe = NULL;
+	}
+	if (dp->intr_pipe) {
+		usb_pipe_close(dp->dip,
+		    dp->intr_pipe, USB_FLAGS_SLEEP, NULL, 0);
+		dp->intr_pipe = NULL;
+	}
+
+	return (USB_FAILURE);
+}
+
+static int
+usbgem_close_pipes(struct usbgem_dev *dp)
+{
+	DPRINTF(0, (CE_CONT, "!%s: %s: called", dp->name, __func__));
+
+	if (dp->intr_pipe) {
+		usb_pipe_close(dp->dip,
+		    dp->intr_pipe, USB_FLAGS_SLEEP, NULL, 0);
+		dp->intr_pipe = NULL;
+	}
+	DPRINTF(1, (CE_CONT, "!%s: %s: 1", dp->name, __func__));
+
+	ASSERT(dp->bulkin_pipe);
+	usb_pipe_close(dp->dip, dp->bulkin_pipe, USB_FLAGS_SLEEP, NULL, 0);
+	dp->bulkin_pipe = NULL;
+	DPRINTF(1, (CE_CONT, "!%s: %s: 2", dp->name, __func__));
+
+	ASSERT(dp->bulkout_pipe);
+	usb_pipe_close(dp->dip, dp->bulkout_pipe, USB_FLAGS_SLEEP, NULL, 0);
+	dp->bulkout_pipe = NULL;
+	DPRINTF(1, (CE_CONT, "!%s: %s: 3", dp->name, __func__));
+
+	return (USB_SUCCESS);
+}
+
+#define FREEZE_GRACEFUL		(B_TRUE)
+#define FREEZE_NO_GRACEFUL	(B_FALSE)
+static int
+usbgem_freeze_device(struct usbgem_dev *dp, boolean_t graceful)
+{
+	DPRINTF(0, (CE_NOTE, "!%s: %s: called", dp->name, __func__));
+
+	/* stop nic activity */
+	(void) usbgem_mac_stop(dp, MAC_STATE_DISCONNECTED, graceful);
+
+	/*
+	 * Here we free all memory resource allocated, because it will
+	 * cause to panic the system that we free usb_bulk_req objects
+	 * during the usb device is disconnected.
+	 */
+	(void) usbgem_free_memory(dp);
+
+	return (USB_SUCCESS);
+}
+
+static int
+usbgem_disconnect_cb(dev_info_t *dip)
+{
+	int	ret;
+	struct usbgem_dev	*dp;
+
+	dp = USBGEM_GET_DEV(dip);
+
+	cmn_err(CE_NOTE, "!%s: the usb device was disconnected (dp=%p)",
+	    dp->name, dp);
+
+	/* start serialize */
+	rw_enter(&dp->dev_state_lock, RW_WRITER);
+
+	ret = usbgem_freeze_device(dp, 0);
+
+	/* end of serialize */
+	rw_exit(&dp->dev_state_lock);
+
+	return (ret);
+}
+
+static int
+usbgem_recover_device(struct usbgem_dev	*dp)
+{
+	int	err;
+
+	DPRINTF(0, (CE_NOTE, "!%s: %s: called", dp->name, __func__));
+
+	err = USB_SUCCESS;
+
+	/* reinitialize the usb connection */
+	usbgem_close_pipes(dp);
+	if ((err = usbgem_open_pipes(dp)) != USB_SUCCESS) {
+		goto x;
+	}
+
+	/* initialize nic state */
+	dp->mac_state = MAC_STATE_STOPPED;
+	dp->mii_state = MII_STATE_UNKNOWN;
+
+	/* allocate memory resources again */
+	if ((err = usbgem_alloc_memory(dp)) != USB_SUCCESS) {
+		goto x;
+	}
+
+	/* restart nic and recover state */
+	(void) usbgem_restart_nic(dp);
+
+	usbgem_mii_init(dp);
+
+	/* kick potentially stopped house keeping thread */
+	cv_signal(&dp->link_watcher_wait_cv);
+x:
+	return (err);
+}
+
+static int
+usbgem_reconnect_cb(dev_info_t *dip)
+{
+	int	err = USB_SUCCESS;
+	struct usbgem_dev	*dp;
+
+	dp = USBGEM_GET_DEV(dip);
+	DPRINTF(0, (CE_CONT, "!%s: dp=%p", ddi_get_name(dip), dp));
+#ifdef notdef
+	/* check device changes after disconnect */
+	if (usb_check_same_device(dp->dip, NULL, USB_LOG_L2, -1,
+	    USB_CHK_BASIC | USB_CHK_CFG, NULL) != USB_SUCCESS) {
+		cmn_err(CE_CONT,
+		     "!%s: no or different device installed", dp->name);
+		return (DDI_SUCCESS);
+	}
+#endif
+	cmn_err(CE_NOTE, "%s: the usb device was reconnected", dp->name);
+
+	/* start serialize */
+	rw_enter(&dp->dev_state_lock, RW_WRITER);
+
+	if (dp->mac_state == MAC_STATE_DISCONNECTED) {
+		err = usbgem_recover_device(dp);
+	}
+
+	/* end of serialize */
+	rw_exit(&dp->dev_state_lock);
+
+	return (err == USB_SUCCESS ? DDI_SUCCESS : DDI_FAILURE);
+}
+
+int
+usbgem_suspend(dev_info_t *dip)
+{
+	int	err = USB_SUCCESS;
+	struct usbgem_dev	*dp;
+
+	dp = USBGEM_GET_DEV(dip);
+
+	DPRINTF(0, (CE_CONT, "!%s: %s: callded", dp->name, __func__));
+
+	/* start serialize */
+	rw_enter(&dp->dev_state_lock, RW_WRITER);
+
+	if (dp->mac_state == MAC_STATE_DISCONNECTED) {
+		err = usbgem_freeze_device(dp, STOP_GRACEFUL);
+	}
+
+	/* end of serialize */
+	rw_exit(&dp->dev_state_lock);
+
+	return (err == USB_SUCCESS ? DDI_SUCCESS : DDI_FAILURE);
+}
+
+int
+usbgem_resume(dev_info_t *dip)
+{
+	int	err = USB_SUCCESS;
+	struct usbgem_dev	*dp;
+
+	dp = USBGEM_GET_DEV(dip);
+
+	DPRINTF(0, (CE_CONT, "!%s: %s: callded", dp->name, __func__));
+#ifdef notdef
+	/* check device changes after disconnect */
+	if (usb_check_same_device(dp->dip, NULL, USB_LOG_L2, -1,
+	    USB_CHK_BASIC | USB_CHK_CFG, NULL) != USB_SUCCESS) {
+		cmn_err(CE_CONT,
+		     "!%s: no or different device installed", dp->name);
+		return (DDI_SUCCESS);
+	}
+#endif
+	/* start serialize */
+	rw_enter(&dp->dev_state_lock, RW_WRITER);
+
+	if (dp->mac_state == MAC_STATE_DISCONNECTED) {
+		err = usbgem_recover_device(dp);
+	}
+
+	/* end of serialize */
+	rw_exit(&dp->dev_state_lock);
+
+	return (err == USB_SUCCESS ? DDI_SUCCESS : DDI_FAILURE);
+}
+
+#define	USBGEM_LOCAL_DATA_SIZE(gc)	\
+	(sizeof (struct usbgem_dev) + USBGEM_MCALLOC)
+
+struct usbgem_dev *
+usbgem_do_attach(dev_info_t *dip,
+	struct usbgem_conf *gc, void *lp, int lmsize)
+{
+	struct usbgem_dev	*dp;
+	int			i;
+#ifdef USBGEM_CONFIG_GLDv3
+	mac_register_t		*macp = NULL;
+#else
+	gld_mac_info_t		*macinfo;
+	void			*tmp;
+#endif
+	int			ret;
+	int			unit;
+	int			err;
+
+	unit = ddi_get_instance(dip);
+
+	DPRINTF(2, (CE_CONT, "!usbgem%d: %s: called", unit, __func__));
+
+	/*
+	 * Allocate soft data structure
+	 */
+	dp = kmem_zalloc(USBGEM_LOCAL_DATA_SIZE(gc), KM_SLEEP);
+	if (dp == NULL) {
+#ifndef USBGEM_CONFIG_GLDv3
+		gld_mac_free(macinfo);
+#endif
+		return (NULL);
+	}
+#ifdef USBGEM_CONFIG_GLDv3
+	if ((macp = mac_alloc(MAC_VERSION)) == NULL) {
+		cmn_err(CE_WARN, "!gem%d: %s: mac_alloc failed",
+		    unit, __func__);
+		return (NULL);
+	}
+#else
+	macinfo = gld_mac_alloc(dip);
+	dp->macinfo = macinfo;
+#endif
+
+	/* link to private area */
+	dp->private = lp;
+	dp->priv_size = lmsize;
+	dp->mc_list = (struct mcast_addr *)&dp[1];
+
+	dp->dip = dip;
+	bcopy(gc->usbgc_name, dp->name, USBGEM_NAME_LEN);
+
+	/*
+	 * register with usb service
+	 */
+	if (usb_client_attach(dip, USBDRV_VERSION, 0) != USB_SUCCESS) {
+		cmn_err(CE_WARN,
+		    "%s: %s: usb_client_attach failed",
+		    dp->name, __func__);
+		goto err_free_private;
+	}
+
+	if (usb_get_dev_data(dip, &dp->reg_data,
+	    USB_PARSE_LVL_ALL, 0) != USB_SUCCESS) {
+		dp->reg_data = NULL;
+		goto err_unregister_client;
+	}
+#ifdef USBGEM_DEBUG_LEVEL
+	usb_print_descr_tree(dp->dip, dp->reg_data);
+#endif
+
+	if (usbgem_open_pipes(dp) != USB_SUCCESS) {
+		/* failed to open pipes */
+		cmn_err(CE_WARN, "!%s: %s: failed to open pipes",
+		    dp->name, __func__);
+		goto err_unregister_client;
+	}
+
+	/*
+	 * Initialize mutexs and condition variables
+	 */
+	mutex_init(&dp->rxlock, NULL, MUTEX_DRIVER, NULL);
+	mutex_init(&dp->txlock, NULL, MUTEX_DRIVER, NULL);
+	cv_init(&dp->rx_drain_cv, NULL, CV_DRIVER, NULL);
+	cv_init(&dp->tx_drain_cv, NULL, CV_DRIVER, NULL);
+	rw_init(&dp->dev_state_lock, NULL, RW_DRIVER, NULL);
+	mutex_init(&dp->link_watcher_lock, NULL, MUTEX_DRIVER, NULL);
+	cv_init(&dp->link_watcher_wait_cv, NULL, CV_DRIVER, NULL);
+	sema_init(&dp->hal_op_lock, 1, NULL, SEMA_DRIVER, NULL);
+	sema_init(&dp->rxfilter_lock, 1, NULL, SEMA_DRIVER, NULL);
+
+	/*
+	 * Initialize configuration
+	 */
+	dp->ugc = *gc;
+
+	dp->mtu = ETHERMTU;
+	dp->rxmode = 0;
+	dp->speed = USBGEM_SPD_10;	/* default is 10Mbps */
+	dp->full_duplex = B_FALSE;	/* default is half */
+	dp->flow_control = FLOW_CONTROL_NONE;
+
+	dp->nic_state = NIC_STATE_STOPPED;
+	dp->mac_state = MAC_STATE_STOPPED;
+	dp->mii_state = MII_STATE_UNKNOWN;
+
+	/* performance tuning parameters */
+	dp->txthr = ETHERMAX;		/* tx fifo threshoold */
+	dp->txmaxdma = 16*4;		/* tx max dma burst size */
+	dp->rxthr = 128;		/* rx fifo threshoold */
+	dp->rxmaxdma = 16*4;		/* rx max dma burst size */
+
+	/*
+	 * Get media mode infomation from .conf file
+	 */
+	usbgem_read_conf(dp);
+
+	/* rx_buf_len depend on MTU */
+	dp->rx_buf_len = MAXPKTBUF(dp) + dp->ugc.usbgc_rx_header_len;
+
+	/*
+	 * Reset the chip
+	 */
+	if (usbgem_hal_reset_chip(dp) != USB_SUCCESS) {
+		cmn_err(CE_WARN,
+		    "!%s: %s: failed to reset the usb device",
+		    dp->name, __func__);
+		goto err_destroy_locks;
+	}
+
+	/*
+	 * HW dependant paremeter initialization
+	 */
+	if (usbgem_hal_attach_chip(dp) != USB_SUCCESS) {
+		cmn_err(CE_WARN,
+		    "!%s: %s: failed to attach the usb device",
+		    dp->name, __func__);
+		goto err_destroy_locks;
+	}
+
+	/* allocate resources */
+	if (usbgem_alloc_memory(dp) != USB_SUCCESS) {
+		goto err_destroy_locks;
+	}
+
+	DPRINTF(0, (CE_CONT,
+	    "!%s: %02x:%02x:%02x:%02x:%02x:%02x",
+	    dp->name,
+	    dp->dev_addr.ether_addr_octet[0],
+	    dp->dev_addr.ether_addr_octet[1],
+	    dp->dev_addr.ether_addr_octet[2],
+	    dp->dev_addr.ether_addr_octet[3],
+	    dp->dev_addr.ether_addr_octet[4],
+	    dp->dev_addr.ether_addr_octet[5]));
+
+	/* copy mac address */
+	dp->cur_addr = dp->dev_addr;
+
+	/* pre-calculated tx timeout in second for performance */
+	dp->bulkout_timeout =
+	    dp->ugc.usbgc_tx_timeout / drv_usectohz(1000*1000);
+
+#ifdef USBGEM_CONFIG_GLDv3
+	usbgem_gld3_init(dp, macp);
+#else
+	usbgem_gld_init(dp, macinfo, ident);
+#endif
+
+	/* Probe MII phy (scan phy) */
+	dp->mii_lpable = 0;
+	dp->mii_advert = 0;
+	dp->mii_exp = 0;
+	dp->mii_ctl1000 = 0;
+	dp->mii_stat1000 = 0;
+
+	dp->mii_status_ro = 0;
+	dp->mii_xstatus_ro = 0;
+
+	if (usbgem_mii_probe(dp) != USB_SUCCESS) {
+		cmn_err(CE_WARN, "!%s: %s: mii_probe failed",
+		    dp->name, __func__);
+		goto err_free_memory;
+	}
+
+	/* mask unsupported abilities */
+	dp->anadv_autoneg &= BOOLEAN(dp->mii_status & MII_STATUS_CANAUTONEG);
+	dp->anadv_1000fdx &=
+	    BOOLEAN(dp->mii_xstatus &
+	    (MII_XSTATUS_1000BASEX_FD | MII_XSTATUS_1000BASET_FD));
+	dp->anadv_1000hdx &=
+	    BOOLEAN(dp->mii_xstatus &
+	    (MII_XSTATUS_1000BASEX | MII_XSTATUS_1000BASET));
+	dp->anadv_100t4 &= BOOLEAN(dp->mii_status & MII_STATUS_100_BASE_T4);
+	dp->anadv_100fdx &= BOOLEAN(dp->mii_status & MII_STATUS_100_BASEX_FD);
+	dp->anadv_100hdx &= BOOLEAN(dp->mii_status & MII_STATUS_100_BASEX);
+	dp->anadv_10fdx &= BOOLEAN(dp->mii_status & MII_STATUS_10_FD);
+	dp->anadv_10hdx &= BOOLEAN(dp->mii_status & MII_STATUS_10);
+
+	if (usbgem_mii_init(dp) != USB_SUCCESS) {
+		cmn_err(CE_WARN, "!%s: %s: mii_init failed",
+		    dp->name, __func__);
+		goto err_free_memory;
+	}
+
+	/*
+	 * initialize kstats including mii statistics
+	 */
+#ifdef USBGEM_CONFIG_GLDv3
+#ifdef USBGEM_CONFIG_ND
+	usbgem_nd_setup(dp);
+#endif
+#else
+	if (usbgem_kstat_init(dp) != USB_SUCCESS) {
+		goto err_free_memory;
+	}
+#endif
+
+	/*
+	 * Add interrupt to system.
+	 */
+#ifdef USBGEM_CONFIG_GLDv3
+	if (ret = mac_register(macp, &dp->mh)) {
+		cmn_err(CE_WARN, "!%s: mac_register failed, error:%d",
+		    dp->name, ret);
+		goto err_release_stats;
+	}
+	mac_free(macp);
+	macp = NULL;
+#else
+	/* gld_register will corrupts driver_private */
+	tmp = ddi_get_driver_private(dip);
+	if (gld_register(dip,
+	    (char *)ddi_driver_name(dip), macinfo) != DDI_SUCCESS) {
+		cmn_err(CE_WARN, "!%s: %s: gld_register failed",
+		    dp->name, __func__);
+		ddi_set_driver_private(dip, tmp);
+		goto err_release_stats;
+	}
+	/* restore driver private */
+	ddi_set_driver_private(dip, tmp);
+#endif /* USBGEM_CONFIG_GLDv3 */
+	if (usb_register_hotplug_cbs(dip,
+	    usbgem_suspend, usbgem_resume) != USB_SUCCESS) {
+		cmn_err(CE_WARN,
+		    "!%s: %s: failed to register hotplug cbs",
+		    dp->name, __func__);
+		goto err_unregister_gld;
+	}
+
+	/* reset mii and start mii link watcher */
+	if (usbgem_mii_start(dp) != USB_SUCCESS) {
+		goto err_unregister_hotplug;
+	}
+
+	/* start tx watchdow watcher */
+	if (usbgem_tx_watcher_start(dp)) {
+		goto err_usbgem_mii_stop;
+	}
+
+	ddi_set_driver_private(dip, (caddr_t)dp);
+
+	DPRINTF(2, (CE_CONT, "!%s: %s: return: success", dp->name, __func__));
+
+	return (dp);
+
+err_usbgem_mii_stop:
+	usbgem_mii_stop(dp);
+
+err_unregister_hotplug:
+	usb_unregister_hotplug_cbs(dip);
+
+err_unregister_gld:
+#ifdef USBGEM_CONFIG_GLDv3
+	mac_unregister(dp->mh);
+#else
+	gld_unregister(macinfo);
+#endif
+
+err_release_stats:
+#ifdef USBGEM_CONFIG_GLDv3
+#ifdef USBGEM_CONFIG_ND
+	/* release NDD resources */
+	usbgem_nd_cleanup(dp);
+#endif
+#else
+	kstat_delete(dp->ksp);
+#endif
+
+err_free_memory:
+	usbgem_free_memory(dp);
+
+err_destroy_locks:
+	cv_destroy(&dp->tx_drain_cv);
+	cv_destroy(&dp->rx_drain_cv);
+	mutex_destroy(&dp->txlock);
+	mutex_destroy(&dp->rxlock);
+	rw_destroy(&dp->dev_state_lock);
+	mutex_destroy(&dp->link_watcher_lock);
+	cv_destroy(&dp->link_watcher_wait_cv);
+	sema_destroy(&dp->hal_op_lock);
+	sema_destroy(&dp->rxfilter_lock);
+
+err_close_pipes:
+	(void) usbgem_close_pipes(dp);
+
+err_unregister_client:
+	usb_client_detach(dp->dip, dp->reg_data);
+
+err_free_private:
+#ifdef USBGEM_CONFIG_GLDv3
+	if (macp) {
+		mac_free(macp);
+	}
+#else
+	gld_mac_free(macinfo);
+#endif
+	kmem_free((caddr_t)dp, USBGEM_LOCAL_DATA_SIZE(gc));
+
+	return (NULL);
+}
+
+int
+usbgem_do_detach(dev_info_t *dip)
+{
+	struct usbgem_dev	*dp;
+
+	dp = USBGEM_GET_DEV(dip);
+
+#ifdef USBGEM_CONFIG_GLDv3
+	/* unregister with gld v3 */
+	if (mac_unregister(dp->mh) != DDI_SUCCESS) {
+		return (DDI_FAILURE);
+	}
+#else
+	/* unregister with gld v2 */
+	if (gld_unregister(dp->macinfo) != DDI_SUCCESS) {
+		return (DDI_FAILURE);
+	}
+#endif
+	/* unregister with hotplug service */
+	usb_unregister_hotplug_cbs(dip);
+
+	/* stop tx watchdog watcher*/
+	usbgem_tx_watcher_stop(dp);
+
+	/* stop the link manager */
+	usbgem_mii_stop(dp);
+
+	/* unregister with usb service */
+	(void) usbgem_free_memory(dp);
+	(void) usbgem_close_pipes(dp);
+	usb_client_detach(dp->dip, dp->reg_data);
+	dp->reg_data = NULL;
+
+	/* unregister with kernel statistics */
+#ifdef USBGEM_CONFIG_GLDv3
+#ifdef USBGEM_CONFIG_ND
+	/* release ndd resources */
+	usbgem_nd_cleanup(dp);
+#endif
+#else
+	/* destroy kstat objects */
+	kstat_delete(dp->ksp);
+#endif
+
+	/* release locks and condition variables */
+	mutex_destroy(&dp->txlock);
+	mutex_destroy(&dp->rxlock);
+	cv_destroy(&dp->tx_drain_cv);
+	cv_destroy(&dp->rx_drain_cv);
+	rw_destroy(&dp->dev_state_lock);
+	mutex_destroy(&dp->link_watcher_lock);
+	cv_destroy(&dp->link_watcher_wait_cv);
+	sema_destroy(&dp->hal_op_lock);
+	sema_destroy(&dp->rxfilter_lock);
+
+	/* release basic memory resources */
+#ifndef USBGEM_CONFIG_GLDv3
+	gld_mac_free(dp->macinfo);
+#endif
+	kmem_free((caddr_t)(dp->private), dp->priv_size);
+	kmem_free((caddr_t)dp, USBGEM_LOCAL_DATA_SIZE(&dp->ugc));
+
+	DPRINTF(2, (CE_CONT, "!%s: %s: return: success",
+	    ddi_driver_name(dip), __func__));
+
+	return (DDI_SUCCESS);
+}
+
+int
+usbgem_mod_init(struct dev_ops *dop, char *name)
+{
+#ifdef USBGEM_CONFIG_GLDv3
+	major_t	major;
+	major = ddi_name_to_major(name);
+	if (major == DDI_MAJOR_T_NONE) {
+		return (DDI_FAILURE);
+	}
+	mac_init_ops(dop, name);
+#endif
+	return (DDI_SUCCESS);
+}
+
+void
+usbgem_mod_fini(struct dev_ops *dop)
+{
+#ifdef USBGEM_CONFIG_GLDv3
+	mac_fini_ops(dop);
+#endif
+}
+
+int
+usbgem_quiesce(dev_info_t *dip)
+{
+	struct usbgem_dev	*dp;
+
+	dp = USBGEM_GET_DEV(dip);
+
+	ASSERT(dp != NULL);
+
+	if (dp->mac_state != MAC_STATE_DISCONNECTED &&
+	    dp->mac_state != MAC_STATE_STOPPED) {
+		if (usbgem_hal_stop_chip(dp) != USB_SUCCESS) {
+			(void) usbgem_hal_reset_chip(dp);
+		}
+	}
+
+	/* devo_quiesce() must return DDI_SUCCESS always */
+	return (DDI_SUCCESS);
+}
diff --git a/usr/src/uts/common/io/usbgem/usbgem.h b/usr/src/uts/common/io/usbgem/usbgem.h
new file mode 100644
index 0000000000..80b89a260e
--- /dev/null
+++ b/usr/src/uts/common/io/usbgem/usbgem.h
@@ -0,0 +1,428 @@
+/*
+ * usbgem.h: General USB to Ethernet MAC driver framework
+ * @(#)usbgem.h	1.4 12/02/09
+ * (C) Copyright 2003-2009 Masayuki Murayama KHF04453@nifty.ne.jp
+ */
+
+#ifndef __USBGEM_H__
+#define __USBGEM_H__
+
+#pragma	ident	"@(#)usbgem.h	1.4 12/02/09"
+
+#ifdef USBGEM_CONFIG_GLDv3
+#include <sys/mac.h>
+#ifndef MAC_VERSION
+#include <sys/mac_provider.h>
+#endif
+#include <sys/mac_ether.h>
+#else
+#include <sys/gld.h>
+#endif /* GLDv3 */
+
+/*
+ * Useful macros and typedefs
+ */
+#define	USBGEM_NAME_LEN	32
+
+#define USBGEM_TX_TIMEOUT		(drv_usectohz(3*1000000))
+#define USBGEM_TX_TIMEOUT_INTERVAL	(drv_usectohz(1*1000000))
+#define USBGEM_LINK_WATCH_INTERVAL	(drv_usectohz(1*1000000))
+
+/* general return code */
+#define	USBGEM_SUCCESS	0
+#define	USBGEM_FAILURE	1
+
+/* return code of usbgem_tx_done */
+#define	INTR_RESTART_TX	0x80000000U
+
+struct usbgem_stats {
+	uint32_t	intr;
+
+	uint32_t	crc;
+	uint32_t	errrcv;
+	uint32_t	overflow;
+	uint32_t	frame;
+	uint32_t	missed;
+	uint32_t	runt;
+	uint32_t	frame_too_long;
+	uint32_t	norcvbuf;
+	uint32_t	sqe;
+
+	uint32_t	collisions;
+	uint32_t	first_coll;
+	uint32_t	multi_coll;
+	uint32_t	excoll;
+	uint32_t	xmit_internal_err;
+	uint32_t	nocarrier;
+	uint32_t	defer;
+	uint32_t	errxmt;
+	uint32_t	underflow;
+	uint32_t	xmtlatecoll;
+	uint32_t	noxmtbuf;
+	uint32_t	jabber;
+
+
+	uint64_t	rbytes;
+	uint64_t	obytes;
+	uint64_t	rpackets;
+	uint64_t	opackets;
+	uint32_t	rbcast;
+	uint32_t	obcast;
+	uint32_t	rmcast;
+	uint32_t	omcast;
+	uint32_t	rcv_internal_err;
+};
+
+struct mcast_addr {
+	struct ether_addr	addr;
+	uint32_t		hash;
+};
+
+#define	USBGEM_MAXMC	64
+#define	USBGEM_MCALLOC	(sizeof(struct mcast_addr) * USBGEM_MAXMC)
+
+#define	SLOT(dp, n)	((n) % (dp)->ugc.usbgc_tx_list_max)
+
+/*
+ * mac soft state
+ */
+struct usbgem_dev {
+	dev_info_t	*dip;
+#ifdef USBGEM_CONFIG_GLDv3
+	mac_handle_t	mh;
+#else
+	void		*macinfo;	/* opaque handle for upper layer */
+#endif
+	char		name[USBGEM_NAME_LEN];
+
+	/* pointer to usb private data */
+	usb_client_dev_data_t	*reg_data;
+
+	/* usb handles */
+	usb_pipe_handle_t	default_pipe;
+	usb_pipe_handle_t	bulkin_pipe;
+	usb_pipe_handle_t	bulkout_pipe;
+	usb_pipe_handle_t	intr_pipe;
+
+	/* usb endpoints */
+	usb_ep_descr_t		*ep_default;
+	usb_ep_descr_t		*ep_bulkin;
+	usb_ep_descr_t		*ep_bulkout;
+	usb_ep_descr_t		*ep_intr;
+
+	/* usb policies */
+	usb_pipe_policy_t	policy_default;
+	usb_pipe_policy_t	policy_bulkin;
+	usb_pipe_policy_t	policy_bulkout;
+	usb_pipe_policy_t	policy_interrupt;
+
+	/* MAC address information */
+	struct ether_addr	cur_addr;
+	struct ether_addr	dev_addr;
+
+	/* RX state and resource management */
+	kmutex_t		rxlock;
+	int			rx_busy_cnt;
+	boolean_t		rx_active;
+	kcondvar_t		rx_drain_cv;
+
+	/* RX buffer management */
+	int			rx_buf_len;
+
+	/* TX state and resource management */
+	kmutex_t		txlock;
+	int			tx_busy_cnt;
+	usb_bulk_req_t		*tx_free_list;
+	kcondvar_t		tx_drain_cv;
+	clock_t			tx_start_time;
+	int			bulkout_timeout;	/* in second */
+	int			tx_max_packets;
+	int			tx_seq_num;
+	int			tx_intr_pended;
+
+	/* NIC state from OS view */
+	int			nic_state;
+#define	NIC_STATE_UNKNOWN	0
+#define	NIC_STATE_STOPPED	1
+#define	NIC_STATE_INITIALIZED	2
+#define	NIC_STATE_ONLINE	3
+
+	/* MAC state from hardware view */
+	int			mac_state;
+#define	MAC_STATE_DISCONNECTED	0	/* it includes suspended state too */
+#define	MAC_STATE_STOPPED	1	/* powered up / buf not initialized */
+#define	MAC_STATE_INITIALIZED	2	/* initialized */
+#define	MAC_STATE_ONLINE	3	/* working correctly  */
+#define	MAC_STATE_ERROR		4	/* need to restart nic */
+
+	clock_t			fatal_error;
+
+	/* robustness: timer and watchdog */
+	uint_t			tx_watcher_stop;
+	kt_did_t		tx_watcher_did;
+	kcondvar_t		tx_watcher_cv;
+	kmutex_t		tx_watcher_lock;
+	clock_t			tx_watcher_timeout;
+	clock_t			tx_watcher_interval;
+
+	/* MII mamagement */
+	boolean_t		anadv_autoneg:1;
+	boolean_t		anadv_1000fdx:1;
+	boolean_t		anadv_1000hdx:1;
+	boolean_t		anadv_100t4:1;
+	boolean_t		anadv_100fdx:1;
+	boolean_t		anadv_100hdx:1;
+	boolean_t		anadv_10fdx:1;
+	boolean_t		anadv_10hdx:1;
+	boolean_t		anadv_1000t_ms:2;
+	boolean_t		anadv_pause:1;
+	boolean_t		anadv_asmpause:1;
+	boolean_t		mii_advert_ro:1;
+
+	boolean_t		full_duplex:1;
+	int			speed:3;
+#define		USBGEM_SPD_10	0
+#define		USBGEM_SPD_100	1
+#define		USBGEM_SPD_1000	2
+#define		USBGEM_SPD_NUM	3
+	unsigned int		flow_control:2;
+#define		FLOW_CONTROL_NONE	0
+#define		FLOW_CONTROL_SYMMETRIC	1
+#define		FLOW_CONTROL_TX_PAUSE	2
+#define		FLOW_CONTROL_RX_PAUSE	3
+
+	boolean_t		mii_supress_msg:1;
+
+	uint32_t		mii_phy_id;
+	uint16_t		mii_status;
+	uint16_t		mii_advert;
+	uint16_t		mii_lpable;
+	uint16_t		mii_exp;
+	uint16_t		mii_ctl1000;
+	uint16_t		mii_stat1000;
+	uint16_t		mii_xstatus;
+	int8_t			mii_phy_addr;	/* must be signed */
+
+	uint16_t		mii_status_ro;
+	uint16_t		mii_xstatus_ro;
+
+	int			mii_state;
+#define		MII_STATE_UNKNOWN		0
+#define		MII_STATE_RESETTING		1
+#define		MII_STATE_AUTONEGOTIATING	2
+#define		MII_STATE_AN_DONE		3
+#define		MII_STATE_MEDIA_SETUP		4
+#define		MII_STATE_LINKUP		5
+#define		MII_STATE_LINKDOWN		6
+
+	clock_t			mii_last_check;	/* in tick */
+	clock_t			mii_timer;	/* in tick */
+#define		MII_RESET_TIMEOUT	drv_usectohz(1000*1000)
+#define		MII_AN_TIMEOUT		drv_usectohz(5000*1000)
+#define		MII_LINKDOWN_TIMEOUT	drv_usectohz(10000*1000)
+
+	clock_t			mii_interval;	/* in tick */
+	clock_t			linkup_delay;	/* in tick */
+
+	uint_t			link_watcher_stop;
+	kt_did_t		link_watcher_did;
+	kcondvar_t		link_watcher_wait_cv;
+	kmutex_t		link_watcher_lock;
+
+	krwlock_t		dev_state_lock;	/* mac_state and nic_state */
+	ksema_t			hal_op_lock;	/* serialize hw operations */
+	ksema_t			drv_op_lock;	/* hotplug op lock */
+
+	/* multcast list */
+	ksema_t			rxfilter_lock;
+	int			mc_count;
+	int			mc_count_req;
+	struct mcast_addr	*mc_list;
+	int			rxmode;
+#define		RXMODE_PROMISC		0x01
+#define		RXMODE_ALLMULTI_REQ	0x02
+#define		RXMODE_MULTI_OVF	0x04
+#define		RXMODE_ENABLE		0x08
+#define		RXMODE_ALLMULTI		(RXMODE_ALLMULTI_REQ | RXMODE_MULTI_OVF)
+#define		RXMODE_BITS	\
+			"\020"	\
+			"\004ENABLE"	\
+			"\003MULTI_OVF"	\
+			"\002ALLMULTI_REQ"	\
+			"\001PROMISC"
+
+	/* statistcs */
+	struct usbgem_stats		stats;
+
+	/* pointer to local structure */
+	void			*private;
+	int			priv_size;
+
+	/* configuration */
+	struct usbgem_conf {
+		/* name */
+		char		usbgc_name[USBGEM_NAME_LEN];
+		int		usbgc_ppa;
+
+		/* specification on usb */
+		int	usbgc_ifnum;	/* interface number */
+		int	usbgc_alt;	/* alternate */
+
+		/* specification on tx engine */
+		int		usbgc_tx_list_max;
+
+		/* specification on rx engine */
+		int		usbgc_rx_header_len;
+		int		usbgc_rx_list_max;
+
+		/* time out parameters */
+		clock_t		usbgc_tx_timeout;
+		clock_t		usbgc_tx_timeout_interval;
+
+		/* flow control */
+		int		usbgc_flow_control;
+
+		/* MII timeout parameters */
+		clock_t	usbgc_mii_linkdown_timeout;
+		clock_t	usbgc_mii_link_watch_interval;
+		clock_t	usbgc_mii_reset_timeout;
+
+		clock_t	usbgc_mii_an_watch_interval;
+		clock_t	usbgc_mii_an_timeout;
+		clock_t	usbgc_mii_an_wait;	
+		clock_t	usbgc_mii_an_delay;
+
+		/* MII configuration */
+		int	usbgc_mii_addr_min;
+		int	usbgc_mii_linkdown_action;
+		int	usbgc_mii_linkdown_timeout_action;
+#define		MII_ACTION_NONE		0
+#define		MII_ACTION_RESET	1
+#define		MII_ACTION_RSA		2
+		boolean_t	usbgc_mii_dont_reset:1;	
+		boolean_t	usbgc_mii_an_oneshot:1;
+		boolean_t	usbgc_mii_hw_link_detection:1;
+		boolean_t	usbgc_mii_stop_mac_on_linkdown:1;
+		uint16_t	usbgc_mii_an_cmd;
+
+		/* I/O methods */
+
+		/* mac operation */
+		int	(*usbgc_attach_chip)(struct usbgem_dev *dp);
+		int	(*usbgc_reset_chip)(struct usbgem_dev *dp);
+		int	(*usbgc_init_chip)(struct usbgem_dev *dp);
+		int	(*usbgc_start_chip)(struct usbgem_dev *dp);
+		int	(*usbgc_stop_chip)(struct usbgem_dev *dp);
+		uint32_t (*usbgc_multicast_hash)(struct usbgem_dev *dp,
+		    const uint8_t *);
+		int	(*usbgc_set_rx_filter)(struct usbgem_dev *dp);
+		int	(*usbgc_set_media)(struct usbgem_dev *dp);
+		int	(*usbgc_get_stats)(struct usbgem_dev *dp);
+		void	(*usbgc_interrupt)(struct usbgem_dev *dp, mblk_t *mp);
+
+		/* packet manupilation */
+		mblk_t	*(*usbgc_tx_make_packet)(struct usbgem_dev *dp, mblk_t *mp);
+		mblk_t	*(*usbgc_rx_make_packet)(struct usbgem_dev *dp, mblk_t *mp);
+		/* mii operations */
+		int	(*usbgc_mii_probe)(struct usbgem_dev *dp);
+		int	(*usbgc_mii_init)(struct usbgem_dev *dp);
+		int	(*usbgc_mii_config)(struct usbgem_dev *dp, int *errp);
+		uint16_t (*usbgc_mii_read)(struct usbgem_dev *dp, uint_t reg, int *errp);
+		void	(*usbgc_mii_write)(struct usbgem_dev *dp, uint_t reg, uint16_t val, int *errp);
+
+		/* jumbo frame */
+		int     usbgc_max_mtu;
+		int     usbgc_default_mtu;
+		int     usbgc_min_mtu;
+	} ugc;
+
+	int	misc_flag;
+#define USBGEM_VLAN	0x0001
+	timeout_id_t	intr_watcher_id;
+
+	/* buffer size */
+	uint_t	mtu;
+
+	/* performance tuning parameters */
+	uint_t	txthr;		/* tx fifo threshoold */
+	uint_t	txmaxdma;	/* tx max dma burst size */ 
+	uint_t	rxthr;		/* rx fifo threshoold */
+	uint_t	rxmaxdma;	/* tx max dma burst size */ 
+
+	/* kstat stuff */
+	kstat_t	*ksp;
+
+	/* ndd stuff */
+	caddr_t	nd_data_p;
+	caddr_t	nd_arg_p;
+
+#ifdef USBGEM_DEBUG_LEVEL
+	int	tx_cnt;
+#endif
+};
+
+/*
+ * Exported functions
+ */
+int usbgem_ctrl_out(struct usbgem_dev *dp,
+    uint8_t reqt, uint8_t req, uint16_t val, uint16_t ix, uint16_t len,
+    void *bp, int size);
+
+int usbgem_ctrl_in(struct usbgem_dev *dp,
+    uint8_t reqt, uint8_t req, uint16_t val, uint16_t ix, uint16_t len,
+    void *bp, int size);
+
+int usbgem_ctrl_out_val(struct usbgem_dev *dp,
+    uint8_t reqt, uint8_t req, uint16_t val, uint16_t ix, uint16_t len,
+    uint32_t v);
+
+int usbgem_ctrl_in_val(struct usbgem_dev *dp,
+    uint8_t reqt, uint8_t req, uint16_t val, uint16_t ix, uint16_t len,
+    void *valp);
+
+void usbgem_generate_macaddr(struct usbgem_dev *, uint8_t *);
+boolean_t usbgem_get_mac_addr_conf(struct usbgem_dev *);
+int usbgem_mii_probe_default(struct usbgem_dev *);
+int usbgem_mii_init_default(struct usbgem_dev *);
+int usbgem_mii_config_default(struct usbgem_dev *, int *errp);
+void usbgem_mii_update_link(struct usbgem_dev *);
+void usbgem_restart_tx(struct usbgem_dev *);
+boolean_t usbgem_tx_done(struct usbgem_dev *, int);
+void usbgem_receive(struct usbgem_dev *);
+struct usbgem_dev *usbgem_do_attach(dev_info_t *,
+    struct usbgem_conf *, void *, int);
+int usbgem_do_detach(dev_info_t *);
+
+uint32_t usbgem_ether_crc_le(const uint8_t *addr);
+uint32_t usbgem_ether_crc_be(const uint8_t *addr);
+
+int usbgem_resume(dev_info_t *);
+int usbgem_suspend(dev_info_t *);
+int usbgem_quiesce(dev_info_t *);
+
+#ifdef USBGEM_CONFIG_GLDv3
+#if DEVO_REV < 4
+#define USBGEM_STREAM_OPS(dev_ops, attach, detach) \
+        DDI_DEFINE_STREAM_OPS(dev_ops, nulldev, nulldev, attach, detach, \
+        nodev, NULL, D_MP, NULL)
+#else
+#define USBGEM_STREAM_OPS(dev_ops, attach, detach) \
+	DDI_DEFINE_STREAM_OPS(dev_ops, nulldev, nulldev, attach, detach, \
+	nodev, NULL, D_MP, NULL, usbgem_quiesce)
+#endif
+#else
+#define usbgem_getinfo     gld_getinfo
+#define usbgem_open        gld_open
+#define usbgem_close       gld_close
+#define usbgem_wput        gld_wput
+#define usbgem_wsrv        gld_wsrv
+#define usbgem_rsrv        gld_rsrv
+#define usbgem_power       NULL
+#endif
+int usbgem_mod_init(struct dev_ops *, char *);
+void usbgem_mod_fini(struct dev_ops *);
+
+#define USBGEM_GET_DEV(dip) \
+	((struct usbgem_dev *)(ddi_get_driver_private(dip)))
+
+#endif /* __USBGEM_H__ */
diff --git a/usr/src/uts/common/io/usbgem/usbgem_mii.h b/usr/src/uts/common/io/usbgem/usbgem_mii.h
new file mode 100644
index 0000000000..2b4176a340
--- /dev/null
+++ b/usr/src/uts/common/io/usbgem/usbgem_mii.h
@@ -0,0 +1,242 @@
+/*
+ *  gem_mii.h: mii header for gem
+ *
+ * Copyright (c) 2002-2007 Masayuki Murayama.  All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions are met:
+ *
+ * 1. Redistributions of source code must retain the above copyright notice,
+ *    this list of conditions and the following disclaimer.
+ *
+ * 2. Redistributions in binary form must reproduce the above copyright notice,
+ *    this list of conditions and the following disclaimer in the documentation
+ *    and/or other materials provided with the distribution.
+ *
+ * 3. Neither the name of the author nor the names of its contributors may be
+ *    used to endorse or promote products derived from this software without
+ *    specific prior written permission.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
+ * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
+ * COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
+ * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
+ * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS
+ * OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
+ * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+ * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
+ * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH
+ * DAMAGE.
+ */
+#pragma ident	"@(#)gem_mii.h	1.4	07/11/30"
+
+/*
+ * gem_mii.h : MII registers
+ */
+#ifndef _GEM_MII_H_
+#define	_GEM_MII_H_
+
+#ifdef GEM_CONFIG_GLDv3
+#include <sys/miiregs.h>
+#else
+#define	MII_CONTROL 		0
+#define	MII_STATUS 		1
+#define	MII_PHYIDH		2
+#define	MII_PHYIDL		3
+#define	MII_AN_ADVERT		4
+#define	MII_AN_LPABLE		5
+#define	MII_AN_EXPANSION	6
+#define	MII_AN_NXTPGXMIT	7
+#endif /* GEM_CONFIG_GLDv3 */
+
+#define	MII_AN_LPANXT		8
+#define	MII_MS_CONTROL		9
+#define	MII_MS_STATUS		10
+#define	MII_XSTATUS		15
+
+/* for 1000BaseT support */
+#define	MII_1000TC		MII_MS_CONTROL
+#define	MII_1000TS		MII_MS_STATUS
+#ifndef GEM_CONFIG_GLDv3
+#define	MII_CONTROL_RESET	0x8000
+#define	MII_CONTROL_LOOPBACK	0x4000
+#define	MII_CONTROL_100MB	0x2000
+#define	MII_CONTROL_ANE		0x1000
+#define	MII_CONTROL_PWRDN	0x0800
+#define	MII_CONTROL_ISOLATE	0x0400
+#define	MII_CONTROL_RSAN	0x0200
+#define	MII_CONTROL_FDUPLEX	0x0100
+#define	MII_CONTROL_COLTST	0x0080
+#endif /* !GEM_CONFIG_GLDv3 */
+#define	MII_CONTROL_SPEED	0x2040
+
+#define	MII_CONTROL_10MB	0x0000
+#define	MII_CONTROL_1000MB	0x0040
+
+#define	MII_CONTROL_BITS	\
+	"\020"	\
+	"\020RESET"	\
+	"\017LOOPBACK"	\
+	"\016100MB"	\
+	"\015ANE"	\
+	"\014PWRDN"	\
+	"\013ISOLATE"	\
+	"\012RSAN"	\
+	"\011FDUPLEX"	\
+	"\010COLTST"	\
+	"\0071000M"
+#ifndef GEM_CONFIG_GLDv3
+#define	MII_STATUS_100_BASE_T4		0x8000
+#define	MII_STATUS_100_BASEX_FD		0x4000
+#define	MII_STATUS_100_BASEX		0x2000
+#define	MII_STATUS_10_FD		0x1000
+#define	MII_STATUS_10			0x0800
+#define	MII_STATUS_MFPRMBLSUPR		0x0040
+#define	MII_STATUS_ANDONE		0x0020
+#define	MII_STATUS_REMFAULT		0x0010
+#define	MII_STATUS_CANAUTONEG		0x0008
+#define	MII_STATUS_LINKUP		0x0004
+#define	MII_STATUS_JABBERING		0x0002
+#define	MII_STATUS_EXTENDED		0x0001
+#endif /* !GEM_CONFIG_GLDv3 */
+#define	MII_STATUS_XSTATUS		0x0100
+#define	MII_STATUS_100_BASE_T2_FD	0x0400
+#define	MII_STATUS_100_BASE_T2		0x0200
+
+#define	MII_STATUS_ABILITY_TECH	\
+	(MII_STATUS_100_BASE_T4	|	\
+	MII_STATUS_100_BASEX_FD |	\
+	MII_STATUS_100_BASEX |	\
+	MII_STATUS_10 |	\
+	MII_STATUS_10_FD)
+
+
+#define	MII_STATUS_BITS	\
+	"\020"	\
+	"\020100_BASE_T4"	\
+	"\017100_BASEX_FD"	\
+	"\016100_BASEX"	\
+	"\01510_BASE_FD"	\
+	"\01410_BASE"	\
+	"\013100_BASE_T2_FD"	\
+	"\012100_BASE_T2"	\
+	"\011XSTATUS"	\
+	"\007MFPRMBLSUPR"	\
+	"\006ANDONE"	\
+	"\005REMFAULT"	\
+	"\004CANAUTONEG"	\
+	"\003LINKUP"	\
+	"\002JABBERING"	\
+	"\001EXTENDED"
+#ifndef GEM_CONFIG_GLDv3
+#define	MII_AN_ADVERT_NP		0x8000
+#define	MII_AN_ADVERT_REMFAULT		0x2000
+#define	MII_AN_ADVERT_SELECTOR		0x001f
+#endif /* !GEM_CONFIG_GLDv3 */
+
+#define	MII_ABILITY_ASM_DIR		0x0800	/* for annex 28B */
+#ifndef	MII_ABILITY_PAUSE
+#define	MII_ABILITY_PAUSE		0x0400	/* for IEEE 802.3x */
+#endif
+#ifndef GEM_CONFIG_GLDv3
+#define	MII_ABILITY_100BASE_T4		0x0200
+#define	MII_ABILITY_100BASE_TX_FD	0x0100
+#define	MII_ABILITY_100BASE_TX		0x0080
+#define	MII_ABILITY_10BASE_T_FD		0x0040
+#define	MII_ABILITY_10BASE_T		0x0020
+#endif /* !GEM_CONFIG_GLDv3 */
+
+#define	MII_AN_LPABLE_NP	0x8000
+
+#define	MII_ABILITY_TECH	\
+	(MII_ABILITY_100BASE_T4	|	\
+	MII_ABILITY_100BASE_TX_FD |	\
+	MII_ABILITY_100BASE_TX |	\
+	MII_ABILITY_10BASE_T |	\
+	MII_ABILITY_10BASE_T_FD)
+
+#define	MII_ABILITY_ALL	\
+	(MII_AN_ADVERT_REMFAULT |	\
+	MII_ABILITY_ASM_DIR |	\
+	MII_ABILITY_PAUSE |	\
+	MII_ABILITY_TECH)
+
+
+#define	MII_ABILITY_BITS	\
+	"\020"	\
+	"\016REMFAULT"	\
+	"\014ASM_DIR"	\
+	"\013PAUSE"	\
+	"\012100BASE_T4"	\
+	"\011100BASE_TX_FD"	\
+	"\010100BASE_TX"	\
+	"\00710BASE_T_FD"	\
+	"\00610BASE_T"
+#ifndef GEM_CONFIG_GLDv3
+#define	MII_AN_EXP_PARFAULT	0x0010
+#define	MII_AN_EXP_LPCANNXTP	0x0008
+#define	MII_AN_EXP_CANNXTPP	0x0004
+#define	MII_AN_EXP_PAGERCVD 	0x0002
+#define	MII_AN_EXP_LPCANAN 	0x0001
+#endif /* !GEM_CONFIG_GLDv3 */
+
+#define	MII_AN_EXP_BITS	\
+	"\020"	\
+	"\005PARFAULT"	\
+	"\004LPCANNXTP"	\
+	"\003CANNXTPP"	\
+	"\002PAGERCVD"	\
+	"\001LPCANAN"
+
+#define	MII_1000TC_TESTMODE	0xe000
+#define	MII_1000TC_CFG_EN	0x1000
+#define	MII_1000TC_CFG_VAL	0x0800
+#define	MII_1000TC_PORTTYPE	0x0400
+#define	MII_1000TC_ADV_FULL	0x0200
+#define	MII_1000TC_ADV_HALF	0x0100
+
+#define	MII_1000TC_BITS	\
+	"\020"	\
+	"\015CFG_EN"	\
+	"\014CFG_VAL"	\
+	"\013PORTTYPE"	\
+	"\012FULL"	\
+	"\011HALF"
+
+#define	MII_1000TS_CFG_FAULT	0x8000
+#define	MII_1000TS_CFG_MASTER	0x4000
+#define	MII_1000TS_LOCALRXOK	0x2000
+#define	MII_1000TS_REMOTERXOK	0x1000
+#define	MII_1000TS_LP_FULL	0x0800
+#define	MII_1000TS_LP_HALF	0x0400
+
+#define	MII_1000TS_BITS	\
+	"\020"	\
+	"\020CFG_FAULT"	\
+	"\017CFG_MASTER"	\
+	"\014CFG_LOCALRXOK"	\
+	"\013CFG_REMOTERXOK"	\
+	"\012LP_FULL"	\
+	"\011LP_HALF"
+
+#define	MII_XSTATUS_1000BASEX_FD	0x8000
+#define	MII_XSTATUS_1000BASEX		0x4000
+#define	MII_XSTATUS_1000BASET_FD	0x2000
+#define	MII_XSTATUS_1000BASET		0x1000
+
+#define	MII_XSTATUS_BITS	\
+	"\020"	\
+	"\0201000BASEX_FD"	\
+	"\0171000BASEX"		\
+	"\0161000BASET_FD"	\
+	"\0151000BASET"
+
+#define	MII_READ_CMD(p, r)	\
+	((6<<(18+5+5)) | ((p)<<(18+5)) | ((r)<<18))
+
+#define	MII_WRITE_CMD(p, r, v)	\
+	((5<<(18+5+5)) | ((p)<<(18+5)) | ((r)<<18) | (2 << 16) | (v))
+
+#endif /* _GEM_MII_H_ */
diff --git a/usr/src/uts/common/io/vnd/frameio.c b/usr/src/uts/common/io/vnd/frameio.c
new file mode 100644
index 0000000000..e4e700fa12
--- /dev/null
+++ b/usr/src/uts/common/io/vnd/frameio.c
@@ -0,0 +1,464 @@
+/*
+ * This file and its contents are supplied under the terms of the
+ * Common Development and Distribution License ("CDDL"), version 1.0.
+ * You may only use this file in accordance with the terms of version
+ * 1.0 of the CDDL.
+ *
+ * A full copy of the text of the CDDL should have accompanied this
+ * source.  A copy of the CDDL is also available via the Internet at
+ * http://www.illumos.org/license/CDDL.
+ */
+
+/*
+ * Copyright (c) 2014, Joyent, Inc.  All rights reserved.
+ */
+
+/*
+ * Frame I/O utility functions
+ */
+
+#include <sys/frameio.h>
+
+#include <sys/file.h>
+#include <sys/types.h>
+#include <sys/kmem.h>
+#include <sys/ddi.h>
+#include <sys/sunddi.h>
+#include <sys/stream.h>
+#include <sys/strsun.h>
+#include <sys/sysmacros.h>
+#include <sys/inttypes.h>
+
+static kmem_cache_t *frameio_cache;
+
+int
+frameio_init(void)
+{
+	frameio_cache = kmem_cache_create("frameio_cache",
+	    sizeof (frameio_t) + sizeof (framevec_t) * FRAMEIO_NVECS_MAX,
+	    0, NULL, NULL, NULL, NULL, NULL, 0);
+	if (frameio_cache == NULL)
+		return (1);
+
+	return (0);
+}
+
+void
+frameio_fini(void)
+{
+	if (frameio_cache != NULL)
+		kmem_cache_destroy(frameio_cache);
+}
+
+frameio_t *
+frameio_alloc(int kmflags)
+{
+	return (kmem_cache_alloc(frameio_cache, kmflags));
+}
+
+void
+frameio_free(frameio_t *fio)
+{
+	return (kmem_cache_free(frameio_cache, fio));
+}
+
+/*
+ * Ensure that we don't see any garbage in the framevecs that we're nominally
+ * supposed to work with. Specifically we want to make sure that the buflen and
+ * the address are not zero.
+ */
+static int
+frameio_hdr_check_vecs(frameio_t *fio)
+{
+	int i;
+	for (i = 0; i < fio->fio_nvecs; i++)
+		if (fio->fio_vecs[i].fv_buf == NULL ||
+		    fio->fio_vecs[i].fv_buflen == 0)
+			return (EINVAL);
+
+	return (0);
+}
+
+/*
+ * We have to copy in framevec32_t's. To work around the data model issues and
+ * trying not to copy memory we first copy in the framevec32_t data into the
+ * standard fio_vec space. Next we work backwards copying a given framevec32_t
+ * to a temporaory framevec_t and then overwrite the frameio_t's data. Note that
+ * it is important that we do this in reverse so as to ensure that we don't
+ * clobber data as the framevec_t is larger than the framevec32_t.
+ */
+static int
+frameio_hdr_copyin_ilp32(frameio_t *fio, const void *addr)
+{
+	framevec32_t *vec32p;
+	framevec_t fv;
+	int i;
+
+	vec32p = (framevec32_t *)&fio->fio_vecs[0];
+
+	if (ddi_copyin(addr, vec32p, sizeof (framevec32_t) * fio->fio_nvecs,
+	    0) != 0)
+		return (EFAULT);
+
+	for (i = fio->fio_nvecs - 1; i >= 0; i--) {
+		fv.fv_buf = (void *)(uintptr_t)vec32p[i].fv_buf;
+		fv.fv_buflen = vec32p[i].fv_buflen;
+		fv.fv_actlen = vec32p[i].fv_actlen;
+		fio->fio_vecs[i].fv_buf = fv.fv_buf;
+		fio->fio_vecs[i].fv_buflen = fv.fv_buflen;
+		fio->fio_vecs[i].fv_actlen = fv.fv_actlen;
+	}
+
+	return (frameio_hdr_check_vecs(fio));
+}
+
+/*
+ * Copy in a frame io header into fio with space for up to nvecs. If the frameio
+ * contains more vectors than specified it will be ignored. mode should contain
+ * information about the datamodel.
+ */
+int
+frameio_hdr_copyin(frameio_t *fio, int max_vecs, const void *addr, uint_t mode)
+{
+	int model = ddi_model_convert_from(mode & FMODELS);
+	int cpf = mode & FKIOCTL ? FKIOCTL : 0;
+	size_t fsize = model == DDI_MODEL_ILP32 ?
+	    sizeof (frameio32_t) : sizeof (frameio_t);
+
+	/*
+	 * The start of the header is the same in all data models for the
+	 * current verison.
+	 */
+	if (ddi_copyin(addr, fio, fsize, cpf) != 0)
+		return (EFAULT);
+
+	if (fio->fio_version != FRAMEIO_VERSION_ONE)
+		return (EINVAL);
+
+	if (fio->fio_nvecs > FRAMEIO_NVECS_MAX || fio->fio_nvecs == 0)
+		return (EINVAL);
+
+	if (fio->fio_nvpf == 0)
+		return (EINVAL);
+
+	if (fio->fio_nvecs % fio->fio_nvpf != 0)
+		return (EINVAL);
+
+	if (fio->fio_nvecs > max_vecs)
+		return (EOVERFLOW);
+
+	addr = (void *)((uintptr_t)addr + fsize);
+	if (model == DDI_MODEL_ILP32) {
+		if (cpf != 0)
+			return (EINVAL);
+		return (frameio_hdr_copyin_ilp32(fio, addr));
+	}
+
+	if (ddi_copyin(addr, &fio->fio_vecs[0],
+	    sizeof (framevec_t) * fio->fio_nvecs, cpf) != 0)
+		return (EFAULT);
+
+	return (frameio_hdr_check_vecs(fio));
+}
+
+static mblk_t *
+frameio_allocb(size_t sz)
+{
+	mblk_t *mp;
+
+	mp = allocb(sz, 0);
+	if (mp == NULL)
+		return (NULL);
+
+	mp->b_datap->db_type = M_DATA;
+	return (mp);
+}
+
+static int
+framevec_mblk_read(framevec_t *fv, mblk_t **mpp, int cpf)
+{
+	mblk_t *mp;
+	cpf = cpf != 0 ? FKIOCTL : 0;
+
+	mp = frameio_allocb(fv->fv_buflen);
+
+	if (mp == NULL) {
+		freemsg(mp);
+		return (EAGAIN);
+	}
+
+	if (ddi_copyin(fv->fv_buf, mp->b_wptr, fv->fv_buflen,
+	    cpf) != 0) {
+		freemsg(mp);
+		return (EFAULT);
+	}
+
+	mp->b_wptr += fv->fv_buflen;
+	*mpp = mp;
+	return (0);
+}
+
+/*
+ * Read a set of frame vectors that make up a single message boundary and return
+ * that as a single message in *mpp that consists of multiple data parts.
+ */
+static int
+frameio_mblk_read(frameio_t *fio, framevec_t *fv, mblk_t **mpp, int cpf)
+{
+	int nparts = fio->fio_nvpf;
+	int part, error;
+	mblk_t *mp;
+
+	*mpp = NULL;
+	cpf = cpf != 0 ? FKIOCTL : 0;
+
+	/*
+	 * Construct the initial frame
+	 */
+	for (part = 0; part < nparts; part++) {
+		error = framevec_mblk_read(fv, &mp, cpf);
+		if (error != 0) {
+			freemsg(*mpp);
+			return (error);
+		}
+
+		if (*mpp == NULL)
+			*mpp = mp;
+		else
+			linkb(*mpp, mp);
+		fv++;
+	}
+
+	return (0);
+}
+
+/*
+ * Read data from a series of frameio vectors into a message block chain. A
+ * given frameio request has a number of discrete messages divided into
+ * individual vectors based on fio->fio_nvcspframe. Each discrete message will
+ * be constructed into a message block chain pointed to by b_next.
+ *
+ * If we get an EAGAIN while trying to construct a given message block what we
+ * return depends on what else we've done so far. If we have succesfully
+ * completed at least one message then we free everything else we've done so
+ * far and return that. If no messages have been completed we return EAGAIN. If
+ * instead we encounter a different error, say EFAULT, then all of the fv_actlen
+ * entries values are undefined.
+ */
+int
+frameio_mblk_chain_read(frameio_t *fio, mblk_t **mpp, int *nvecs, int cpf)
+{
+	int error = ENOTSUP;
+	int nframes = fio->fio_nvecs / fio->fio_nvpf;
+	int frame;
+	framevec_t *fv;
+	mblk_t *mp, *bmp = NULL;
+
+	/*
+	 * Protect against bogus kernel subsystems.
+	 */
+	VERIFY(fio->fio_nvecs > 0);
+	VERIFY(fio->fio_nvecs % fio->fio_nvpf == 0);
+
+	*mpp = NULL;
+	cpf = cpf != 0 ? FKIOCTL : 0;
+
+	fv = &fio->fio_vecs[0];
+	for (frame = 0; frame < nframes; frame++) {
+		error = frameio_mblk_read(fio, fv, &mp, cpf);
+		if (error != 0)
+			goto failed;
+
+		if (bmp != NULL)
+			bmp->b_next = mp;
+		else
+			*mpp = mp;
+		bmp = mp;
+	}
+
+	*nvecs = nframes;
+	return (0);
+failed:
+	/*
+	 * On EAGAIN we've already taken care of making sure that we have no
+	 * leftover messages, eg. they were never linked in.
+	 */
+	if (error == EAGAIN) {
+		if (frame != 0)
+			error = 0;
+		if (*nvecs != NULL)
+			*nvecs = frame;
+		ASSERT(*mpp != NULL);
+	} else {
+		for (mp = *mpp; mp != NULL; mp = bmp) {
+			bmp = mp->b_next;
+			freemsg(mp);
+		}
+		if (nvecs != NULL)
+			*nvecs = 0;
+		*mpp = NULL;
+	}
+	return (error);
+}
+
+size_t
+frameio_frame_length(frameio_t *fio, framevec_t *fv)
+{
+	int i;
+	size_t len = 0;
+
+	for (i = 0; i < fio->fio_nvpf; i++, fv++)
+		len += fv->fv_buflen;
+
+	return (len);
+}
+
+/*
+ * Write a portion of an mblk to the current.
+ */
+static int
+framevec_write_mblk_part(framevec_t *fv, mblk_t *mp, size_t len, size_t moff,
+    size_t foff, int cpf)
+{
+	ASSERT(len <= MBLKL(mp) - moff);
+	ASSERT(len <= fv->fv_buflen - fv->fv_actlen);
+	cpf = cpf != 0 ? FKIOCTL : 0;
+
+	if (ddi_copyout(mp->b_rptr + moff, fv->fv_buf + foff, len, cpf) != 0)
+		return (EFAULT);
+	fv->fv_actlen += len;
+
+	return (0);
+}
+
+/*
+ * Because copying this out to the user might fail we don't want to update the
+ * b_rptr in case we need to copy it out again.
+ */
+static int
+framevec_map_blk(frameio_t *fio, framevec_t *fv, mblk_t *mp, int cpf)
+{
+	int err;
+	size_t msize, blksize, len, moff, foff;
+
+	msize = msgsize(mp);
+	if (msize > frameio_frame_length(fio, fv))
+		return (EOVERFLOW);
+
+	moff = 0;
+	foff = 0;
+	blksize = MBLKL(mp);
+	fv->fv_actlen = 0;
+	while (msize != 0) {
+		len = MIN(blksize, fv->fv_buflen - fv->fv_actlen);
+		err = framevec_write_mblk_part(fv, mp, len, moff, foff, cpf);
+		if (err != 0)
+			return (err);
+
+		msize -= len;
+		blksize -= len;
+		moff += len;
+		foff += len;
+
+		if (blksize == 0 && msize != 0) {
+			mp = mp->b_cont;
+			ASSERT(mp != NULL);
+			moff = 0;
+			blksize = MBLKL(mp);
+		}
+
+		if (fv->fv_buflen == fv->fv_actlen && msize != 0) {
+			fv++;
+			fv->fv_actlen = 0;
+			foff = 0;
+		}
+	}
+
+	return (0);
+}
+
+int
+frameio_mblk_chain_write(frameio_t *fio, frameio_write_mblk_map_t map,
+    mblk_t *mp, int *nwrite, int cpf)
+{
+	int mcount = 0;
+	int ret = 0;
+
+	if (map != MAP_BLK_FRAME)
+		return (EINVAL);
+
+	while (mp != NULL && mcount < fio->fio_nvecs) {
+		ret = framevec_map_blk(fio, &fio->fio_vecs[mcount], mp, cpf);
+		if (ret != 0)
+			break;
+		mcount += fio->fio_nvpf;
+		mp = mp->b_next;
+	}
+
+	if (ret != 0 && mcount == 0) {
+		if (nwrite != NULL)
+			*nwrite = 0;
+		return (ret);
+	}
+
+	if (nwrite != NULL)
+		*nwrite = mcount / fio->fio_nvpf;
+
+	return (0);
+}
+
+/*
+ * Copy out nframes worth of frameio header data back to userland.
+ */
+int
+frameio_hdr_copyout(frameio_t *fio, int nframes, void *addr, uint_t mode)
+{
+	int i;
+	int model = ddi_model_convert_from(mode & FMODELS);
+	framevec32_t *vec32p;
+	framevec32_t f;
+
+	if (fio->fio_nvecs / fio->fio_nvpf < nframes)
+		return (EINVAL);
+
+	fio->fio_nvecs = nframes * fio->fio_nvpf;
+
+	if (model == DDI_MODEL_NONE) {
+		if (ddi_copyout(fio, addr,
+		    sizeof (frameio_t) + fio->fio_nvecs * sizeof (framevec_t),
+		    mode & FKIOCTL) != 0)
+			return (EFAULT);
+		return (0);
+	}
+
+	ASSERT(model == DDI_MODEL_ILP32);
+
+	vec32p = (framevec32_t *)&fio->fio_vecs[0];
+	for (i = 0; i < fio->fio_nvecs; i++) {
+		f.fv_buf = (caddr32_t)(uintptr_t)fio->fio_vecs[i].fv_buf;
+		if (fio->fio_vecs[i].fv_buflen > UINT_MAX ||
+		    fio->fio_vecs[i].fv_actlen > UINT_MAX)
+			return (EOVERFLOW);
+		f.fv_buflen = fio->fio_vecs[i].fv_buflen;
+		f.fv_actlen = fio->fio_vecs[i].fv_actlen;
+		vec32p[i].fv_buf = f.fv_buf;
+		vec32p[i].fv_buflen = f.fv_buflen;
+		vec32p[i].fv_actlen = f.fv_actlen;
+	}
+
+	if (ddi_copyout(fio, addr,
+	    sizeof (frameio32_t) + fio->fio_nvecs * sizeof (framevec32_t),
+	    mode & FKIOCTL) != 0)
+		return (EFAULT);
+	return (0);
+}
+
+void
+frameio_mark_consumed(frameio_t *fio, int nframes)
+{
+	int i;
+
+	ASSERT(fio->fio_nvecs / fio->fio_nvpf >= nframes);
+	for (i = 0; i < nframes * fio->fio_nvpf; i++)
+		fio->fio_vecs[i].fv_actlen = fio->fio_vecs[i].fv_buflen;
+}
diff --git a/usr/src/uts/common/io/vnd/vnd.c b/usr/src/uts/common/io/vnd/vnd.c
new file mode 100644
index 0000000000..d62c80c1c6
--- /dev/null
+++ b/usr/src/uts/common/io/vnd/vnd.c
@@ -0,0 +1,5584 @@
+/*
+ * This file and its contents are supplied under the terms of the
+ * Common Development and Distribution License ("CDDL"), version 1.0.
+ * You may only use this file in accordance with the terms of version
+ * 1.0 of the CDDL.
+ *
+ * A full copy of the text of the CDDL should have accompanied this
+ * source.  A copy of the CDDL is also available via the Internet at
+ * http://www.illumos.org/license/CDDL.
+ */
+
+/*
+ * Copyright (c) 2014 Joyent, Inc.  All rights reserved.
+ */
+
+/*
+ * vnd - virtual (machine) networking datapath
+ *
+ * vnd's purpose is to provide a highly performant data path for Layer 2 network
+ * traffic and exist side by side an active IP netstack, each servicing
+ * different datalinks. vnd provides many of the same capabilities as the
+ * current TCP/IP stack does and some specific to layer two. Specifically:
+ *
+ * 	o Use of the DLD fastpath
+ * 	o Packet capture hooks
+ * 	o Ability to use hardware capabilities
+ * 	o Useful interfaces for handling multiple frames
+ *
+ * The following image shows where vnd fits into today's networking stack:
+ *
+ *             +---------+----------+----------+
+ *             | libdlpi |  libvnd  | libsocket|
+ *             +---------+----------+----------+
+ *             |         ·          ·    VFS   |
+ *             |   VFS   ·    VFS   +----------+
+ *             |         ·          |  sockfs  |
+ *             +---------+----------+----------+
+ *             |         |    VND   |    IP    |
+ *             |         +----------+----------+
+ *             |            DLD/DLS            |
+ *             +-------------------------------+
+ *             |              MAC              |
+ *             +-------------------------------+
+ *             |             GLDv3             |
+ *             +-------------------------------+
+ *
+ * -----------------------------------------
+ * A Tale of Two Devices - DDI Device Basics
+ * -----------------------------------------
+ *
+ * vnd presents itself to userland as a character device; however, it also is a
+ * STREAMS device so that it can interface with dld and the rest of the
+ * networking stack. Users never interface with the STREAMs devices directly and
+ * they are purely an implementation detail of vnd. Opening the STREAMS device
+ * require kcred and as such userland cannot interact with it or push it onto
+ * the stream head.
+ *
+ * The main vnd character device, /dev/vnd/ctl, is a self-cloning device. Every
+ * clone gets its own minor number; however, minor nodes are not created in the
+ * devices tree for these instances. In this state a user may do two different
+ * things. They may issue ioctls that affect global state or they may issue
+ * ioctls that try to attach it to a given datalink. Once a minor device has
+ * been attached to a datalink, all operations on it are scoped to that context,
+ * therefore subsequent global operations are not permitted.
+ *
+ * A given device can be linked into the /devices and /dev name space via a link
+ * ioctl. That ioctl causes a minor node to be created in /devices and then it
+ * will also appear under /dev/vnd/ due to vnd's sdev plugin. This is similar
+ * to, but simpler than, IP's persistence mechanism.
+ *
+ * ---------------------
+ * Binding to a datalink
+ * ---------------------
+ *
+ * Datalinks are backed by the dld (datalink device) and dls (datalink services)
+ * drivers. These drivers provide a STREAMS device for datalinks on the system
+ * which are exposed through /dev/net. Userland generally manipulates datalinks
+ * through libdlpi. When an IP interface is being plumbed up what actually
+ * happens is that someone does a dlpi_open(3DLPI) of the underlying datalink
+ * and then pushes on the ip STREAMS module with an I_PUSH ioctl.  Modules may
+ * then can negotiate with dld and dls to obtain access to various capabilities
+ * and fast paths via a series of STREAMS messages.
+ *
+ * In vnd, we do the same thing, but we leave our STREAMS module as an
+ * implementation detail of the system. We don't want users to be able to
+ * arbitrarily push vnd STREAMS module onto any stream, so we explicitly require
+ * kcred to manipulate it. Thus, when a user issues a request to attach a
+ * datalink to a minor instance of the character device, that vnd minor instance
+ * itself does a layered open (ldi_open_by_name(9F)) of the specified datalink.
+ * vnd does that open using the passed in credentials from the ioctl, not kcred.
+ * This ensures that users who doesn't have permissions to open the device
+ * cannot. Once that's been opened, we push on the vnd streams module.
+ *
+ * Once the vnd STREAMS instance has been created for this device, eg. the
+ * I_PUSH ioctl returns, we explicitly send a STREAMS ioctl
+ * (VND_STRIOC_ASSOCIATE) to associate the vnd STREAMS and character devices.
+ * This association begins the STREAM device's initialization. We start up an
+ * asynchronous state machine that takes care of all the different aspects of
+ * plumbing up the device with dld and dls and enabling the MAC fast path. We
+ * need to guarantee to consumers of the character device that by the time their
+ * ioctl returns, the data path has been fully initialized.
+ *
+ * The state progression is fairly linear. There are two general steady states.
+ * The first is VND_S_ONLINE, which means that everything is jacked up and good
+ * to go. The alternative is VND_S_ZOMBIE, which means that the streams device
+ * encountered an error or we have finished tearing it down and the character
+ * device can clean it up. The following is our state progression and the
+ * meaning of each state:
+ *
+ *                |
+ *                |
+ *                V
+ *        +---------------+
+ *        | VNS_S_INITIAL |                  This is our initial state. Every
+ *        +---------------+                  vnd STREAMS device starts here.
+ *                |                          While in this state, only dlpi
+ *                |                          M_PROTO and M_IOCTL messages can be
+ *                |                          sent or received. All STREAMS based
+ *                |                          data messages are dropped.
+ *                |                          We transition out of this state by
+ *                |                          sending a DL_INFO_REQ to obtain
+ *                |                          information about the underlying
+ *                |                          link.
+ *                v
+ *        +-----------------+
+ *   +--<-| VNS_S_INFO_SENT |                In this state, we verify and
+ *   |    +-----------------+                record information about the
+ *   |            |                          underlying device. If the device is
+ *   |            |                          not suitable, eg. not of type
+ *   v            |                          DL_ETHER, then we immediately
+ *   |            |                          become a ZOMBIE. To leave this
+ *   |            |                          state we request exclusive active
+ *   |            |                          access to the device via
+ *   v            |                          DL_EXCLUSIVE_REQ.
+ *   |            v
+ *   |    +----------------------+
+ *   +--<-| VNS_S_EXCLUSIVE_SENT |           In this state, we verify whether
+ *   |    +----------------------+           or not we were able to obtain
+ *   |       |             |                 exclusive access to the device. If
+ *   |       |             |                 we were not able to, then we leave,
+ *   v       |             |                 as that means that something like
+ *   |       |             |                 IP is already plumbed up on top of
+ *   |       |             |                 the datalink. We leave this state
+ *   |       |             |                 by progressing through to the
+ *   |       |             |                 appropriate DLPI primitive, either
+ *   v       |             |                 DLPI_ATTACH_REQ or DLPI_BIND_REQ
+ *   |       |             |                 depending on the style of the
+ *   |       |             |                 datalink.
+ *   |       |             v
+ *   |       |    +-------------------+
+ *   +------ |--<-| VNS_S_ATTACH_SENT |      In this state, we verify we were
+ *   |       |    +-------------------+      able to perform a standard DLPI
+ *   |       |          |                    attach and if so, go ahead and
+ *   v       |          |                    send a DLPI_BIND_REQ.
+ *   |       v          v
+ *   |    +-------------------+
+ *   +--<-| VNS_S_BIND_SENT   |              In this state we see the result of
+ *   |    +-------------------+              our attempt to bind to PPA 0 of the
+ *   v             |                         underlying device. Because we're
+ *   |             |                         trying to be a layer two datapath,
+ *   |             |                         the specific attachment point isn't
+ *   |             |                         too important as we're going to
+ *   v             |                         have to enable promiscuous mode. We
+ *   |             |                         transition out of this by sending
+ *   |             |                         our first of three promiscuous mode
+ *   |             |                         requests.
+ *   v             v
+ *   |    +------------------------+
+ *   +--<-| VNS_S_SAP_PROMISC_SENT |         In this state we verify that we
+ *   |    +------------------------+         were able to enable promiscuous
+ *   |             |                         mode at the physical level. We
+ *   |             |                         transition out of this by enabling
+ *   |             |                         multicast and broadcast promiscuous
+ *   v             |                         mode.
+ *   |             v
+ *   |    +--------------------------+
+ *   +--<-| VNS_S_MULTI_PROMISC_SENT |       In this state we verify that we
+ *   |    +--------------------------+       have enabled DL_PROMISC_MULTI and
+ *   v             |                         move onto the final promiscuous
+ *   |             |                         mode request.
+ *   |             v
+ *   |    +----------------------------+
+ *   +--<-| VNS_S_RX_ONLY_PROMISC_SENT |     In this state we verify that we
+ *   |    +----------------------------+     enabled RX_ONLY promiscuous mode.
+ *   |             |                         We specifically do this as we don't
+ *   v             |                         want to receive our own traffic
+ *   |             |                         that we'll send out. We leave this
+ *   |             |                         state by requesting the set of
+ *   |             |                         dld/dls capabilities that we can
+ *   v             |                         process.
+ *   |             |
+ *   |             v
+ *   |    +--------------------+
+ *   +--<-| VNS_S_CAPAB_Q_SENT |             We loop over the set of
+ *   |    +--------------------+             capabilities that dld advertised
+ *   |             |                         and enable the ones that currently
+ *   v             |                         support for use. See the section
+ *   |             |                         later on regarding capabilities
+ *   |             |                         for more information. We leave this
+ *   |             |                         state by sending an enable request.
+ *   v             v
+ *   |    +--------------------+
+ *   +--<-| VNS_S_CAPAB_E_SENT |             Here we finish all capability
+ *   |    +--------------------+             initialization. Once finished, we
+ *   |             |                         transition to the next state. If
+ *   v             |                         the dld fast path is not available,
+ *   |             |                         we become a zombie.
+ *   |             v
+ *   |    +--------------+
+ *   +--<-| VNS_S_ONLINE |                   This is a vnd STREAMS device's
+ *   |    +--------------+                   steady state. It will normally
+ *   |             |                         reside in this state while it is in
+ *   |             |                         active use. It will only transition
+ *   v             |                         to the next state when the STREAMS
+ *   |             |                         device is closed by the character
+ *   |             |                         device. In this state, all data
+ *   |             |                         flows over the dld fast path.
+ *   |             v
+ *   |    +---------------------+
+ *   +--<-| VNS_S_SHUTTING_DOWN |            This vnd state takes care of
+ *   |    +---------------------+            disabling capabilities and then
+ *   |             |                         transitions to zombie state to
+ *   v             |                         indicate that it is finished.
+ *   |             v
+ *   |    +--------------+
+ *   +--->| VNS_S_ZOMBIE |                   In this state, the vnd STREAMS
+ *        +--------------+                   device is waiting to finished being
+ *                                           reaped.
+ *
+ * If the stream association fails for any reason the state machine reaches
+ * VNS_S_ZOMBIE. A more detailed vnd_errno_t will propagate back through the
+ * STREAMS ioctl to the character device. That will fail the user ioctl and
+ * propagate the vnd_errno_t back to userland. If, on the other hand, the
+ * association succeeds, then the vnd STREAMS device will be fully plumbed up
+ * and ready to transmit and receive message blocks. Consumers will be able to
+ * start using the other cbops(9E) entry points once the attach has fully
+ * finished, which will occur after the original user attach ioctl to the
+ * character device returns.
+ *
+ * --------------------
+ * General Architecture
+ * --------------------
+ *
+ * There are several different devices and structures in the vnd driver. There
+ * is a per-netstack component, pieces related to the character device that
+ * consumers see, the internal STREAMS device state, and the data queues
+ * themselves. The following ASCII art picture describes their relationships and
+ * some of the major pieces of data that contain them. These are not exhaustive,
+ * eg. synchronization primitives are left out.
+ *
+ *  +----------------+     +-----------------+
+ *  | global         |     | global          |
+ *  | device list    |     | netstack list   |
+ *  | vnd_dev_list   |     | vnd_nsd_list    |
+ *  +----------------+     +-----------------+
+ *      |                    |
+ *      |                    v
+ *      |    +-------------------+      +-------------------+
+ *      |    | per-netstack data | ---> | per-netstack data | --> ...
+ *      |    | vnd_pnsd_t        |      | vnd_pnsd_t        |
+ *      |    |                   |      +-------------------+
+ *      |    |                   |
+ *      |    | nestackid_t    ---+----> Netstack ID
+ *      |    | vnd_pnsd_flags_t -+----> Status flags
+ *      |    | zoneid_t       ---+----> Zone ID for this netstack
+ *      |    | hook_family_t  ---+----> VND IPv4 Hooks
+ *      |    | hook_family_t  ---+----> VND IPv6 Hooks
+ *      |    | list_t ----+      |
+ *      |    +------------+------+
+ *      |                 |
+ *      |                 v
+ *      |           +------------------+       +------------------+
+ *      |           | character device |  ---> | character device | -> ...
+ *      +---------->| vnd_dev_t        |       | vnd_dev_t        |
+ *                  |                  |       +------------------+
+ *                  |                  |
+ *                  | minor_t       ---+--> device minor number
+ *                  | ldi_handle_t  ---+--> handle to /dev/net/%datalink
+ *                  | vnd_dev_flags_t -+--> device flags, non blocking, etc.
+ *                  | char[]        ---+--> name if linked
+ *                  | vnd_str_t * -+   |
+ *                  +--------------+---+
+ *                                 |
+ *                                 v
+ *          +-------------------------+
+ *          | STREAMS device          |
+ *          | vnd_str_t               |
+ *          |                         |
+ *          | vnd_str_state_t      ---+---> State machine state
+ *          | gsqueue_t *          ---+---> mblk_t Serialization queue
+ *          | vnd_str_stat_t       ---+---> per-device kstats
+ *          | vnd_str_capab_t      ---+----------------------------+
+ *          | vnd_data_queue_t ---+   |                            |
+ *          | vnd_data_queue_t -+ |   |                            v
+ *          +-------------------+-+---+                  +---------------------+
+ *                              | |                      | Stream capabilities |
+ *                              | |                      | vnd_str_capab_t     |
+ *                              | |                      |                     |
+ *                              | |    supported caps <--+-- vnd_capab_flags_t |
+ *                              | |    dld cap handle <--+-- void *            |
+ *                              | |    direct tx func <--+-- vnd_dld_tx_t      |
+ *                              | |                      +---------------------+
+ *                              | |
+ *             +----------------+ +-------------+
+ *             |                                |
+ *             v                                v
+ *  +-------------------+                  +-------------------+
+ *  | Read data queue   |                  | Write data queue  |
+ *  | vnd_data_queue_t  |                  | vnd_data_queue_t  |
+ *  |                   |                  |                   |
+ *  | size_t        ----+--> Current size  | size_t        ----+--> Current size
+ *  | size_t        ----+--> Max size      | size_t        ----+--> Max size
+ *  | mblk_t *      ----+--> Queue head    | mblk_t *      ----+--> Queue head
+ *  | mblk_t *      ----+--> Queue tail    | mblk_t *      ----+--> Queue tail
+ *  +-------------------+                  +-------------------+
+ *
+ *
+ * Globally, we maintain two lists. One list contains all of the character
+ * device soft states. The other maintains a list of all our netstack soft
+ * states. Each netstack maintains a list of active devices that have been
+ * associated with a datalink in its netstack.
+ *
+ * Recall that a given minor instance of the character device exists in one of
+ * two modes. It can either be a cloned open of /dev/vnd/ctl, the control node,
+ * or it can be associated with a given datalink. When minor instances are in
+ * the former state, they do not exist in a given vnd_pnsd_t's list of devices.
+ * As part of attaching to a datalink, the given vnd_dev_t will be inserted into
+ * the appropriate vnd_pnsd_t. In addition, this will cause a STREAMS device, a
+ * vnd_str_t, to be created and associated to a vnd_dev_t.
+ *
+ * The character device, and its vnd_dev_t, is the interface to the rest of the
+ * system. The vnd_dev_t keeps track of various aspects like whether various
+ * operations, such as read, write and the frameio ioctls, are considered
+ * blocking or non-blocking in the O_NONBLOCK sense. It also is responsible for
+ * keeping track of things like the name of the device, if any, in /dev. The
+ * vnd_str_t, on the other hand manages aspects like buffer sizes and the actual
+ * data queues. However, ioctls that manipulate these properties all go through
+ * the vnd_dev_t to its associated vnd_str_t.
+ *
+ * Each of the STREAMS devices, the vnd_str_t, maintains two data queues. One
+ * for frames to transmit (write queue) and one for frames received (read
+ * queue). These data queues have a maximum size and attempting to add data
+ * beyond that maximum size will result in data being dropped. The sizes are
+ * configurable via ioctls VND_IOC_SETTXBUF, VND_IOC_SETRXBUF. Data either sits
+ * in those buffers or has a reservation in those buffers while they are in vnd
+ * and waiting to be consumed by the user or by mac.
+ *
+ * Finally, the vnd_str_t also has a vnd_str_capab_t which we use to manage the
+ * available, negotiated, and currently active features.
+ *
+ * ----------------------
+ * Data Path and gsqueues
+ * ----------------------
+ *
+ * There's a lot of plumbing in vnd to get to the point where we can send data,
+ * but vnd's bread and butter is the data path, so it's worth diving into it in
+ * more detail. Data enters and exits the system from two ends.
+ *
+ * The first end is the vnd consumer. This comes in the form of read and write
+ * system calls as well as the frame I/O ioctls. The read and write system calls
+ * operate on a single frame at a time. Think of a frame as a single message
+ * that has come in off the wire, which may itself comprise multiple mblk_t's
+ * linked together in the kernel. readv(2) and writev(2) have the same
+ * limitations as read(2) and write(2). We enforce this as the system is
+ * required to fill up every uio(9S) buffer before moving onto the next one.
+ * This means that if you have a MTU sized buffer and two frames come in which
+ * are less than half of the MTU they must fill up the given iovec. Even if we
+ * didn't want to do this, we have no way of informing the supplier of the
+ * iovecs that they were only partially filled or where one frame ends and
+ * another begins.  That's life, as such we have frame I/O which solves this
+ * problem. It allows for multiple frames to be consumed as well as for frames
+ * to be broken down into multiple vector components.
+ *
+ * The second end is the mac direct calls. As part of negotiating capabilities
+ * via dld, we give mac a function of ours to call when packets are received
+ * [vnd_mac_input()] and a callback to indicate that flow has been restored
+ * [vnd_mac_flow_control()]. In turn, we also get a function pointer that we can
+ * transmit data with. As part of the contract with mac, mac is allowed to flow
+ * control us by returning a cookie to the transmit function. When that happens,
+ * all outbound traffic is halted until our callback function is called and we
+ * can schedule drains.
+ *
+ * It's worth looking at these in further detail. We'll start with the rx path.
+ *
+ *
+ *                                |
+ *                                * . . . packets from gld
+ *                                |
+ *                                v
+ *                         +-------------+
+ *                         |     mac     |
+ *                         +-------------+
+ *                                |
+ *                                v
+ *                         +-------------+
+ *                         |     dld     |
+ *                         +-------------+
+ *                                |
+ *                                * . . . dld direct callback
+ *                                |
+ *                                v
+ *                        +---------------+
+ *                        | vnd_mac_input |
+ *                        +---------------+
+ *                                |
+ *                                v
+ * +---------+             +-------------+
+ * | dropped |<--*---------|  vnd_hooks  |
+ * |   by    |   .         +-------------+
+ * |  hooks  |   . drop probe     |
+ * +---------+     kstat bump     * . . . Do we have free
+ *                                |         buffer space?
+ *                                |
+ *                          no .  |      . yes
+ *                             .  +      .
+ *                         +---*--+------*-------+
+ *                         |                     |
+ *                         * . . drop probe      * . . recv probe
+ *                         |     kstat bump      |     kstat bump
+ *                         v                     |
+ *                      +---------+              * . . fire pollin
+ *                      | freemsg |              v
+ *                      +---------+   +-----------------------+
+ *                                    | vnd_str_t`vns_dq_read |
+ *                                    +-----------------------+
+ *                                             ^ ^
+ *                             +----------+    | |     +---------+
+ *                             | read(9E) |-->-+ +--<--| frameio |
+ *                             +----------+            +---------+
+ *
+ * The rx path is rather linear. Packets come into us from mac. We always run
+ * them through the various hooks, and if they come out of that, we inspect the
+ * read data queue. If there is not enough space for a packet, we drop it.
+ * Otherwise, we append it to the data queue, and fire read notifications
+ * targetting anyone polling or doing blocking I/O on this device. Those
+ * consumers then drain the head of the data queue.
+ *
+ * The tx path is more complicated due to mac flow control. After any call into
+ * mac, we may have to potentially suspend writes and buffer data for an
+ * arbitrary amount of time. As such, we need to carefully track the total
+ * amount of outstanding data so that we don't waste kernel memory. This is
+ * further complicated by the fact that mac will asynchronously tell us when our
+ * flow has been resumed.
+ *
+ * For data to be able to enter the system, it needs to be able to take a
+ * reservation from the write data queue. Once the reservation has been
+ * obtained, we enter the gsqueue so that we can actually append it. We use
+ * gsqueues (serialization queues) to ensure that packets are manipulated in
+ * order as we deal with the draining and appending packets. We also leverage
+ * its worker thread to help us do draining after mac has restorted our flow.
+ *
+ * The following image describes the flow:
+ *
+ * +-----------+   +--------------+       +-------------------------+   +------+
+ * | write(9E) |-->| Space in the |--*--->| gsqueue_enter_one()     |-->| Done |
+ * | frameio   |   | write queue? |  .    | +->vnd_squeue_tx_append |   +------+
+ * +-----------+   +--------------+  .    +-------------------------+
+ *                         |   ^     .
+ *                         |   |     . reserve space           from gsqueue
+ *                         |   |                                   |
+ *            queue  . . . *   |       space                       v
+ *             full        |   * . . . avail          +------------------------+
+ *                         v   |                      | vnd_squeue_tx_append() |
+ * +--------+          +------------+                 +------------------------+
+ * | EAGAIN |<--*------| Non-block? |<-+                           |
+ * +--------+   .      +------------+  |                           v
+ *              . yes             v    |     wait          +--------------+
+ *                          no . .*    * . . for           | append chain |
+ *                                +----+     space         | to outgoing  |
+ *                                                         |  mblk chain  |
+ *   from gsqueue                                          +--------------+
+ *       |                                                        |
+ *       |      +-------------------------------------------------+
+ *       |      |
+ *       |      |                            yes . . .
+ *       v      v                                    .
+ *  +-----------------------+    +--------------+    .     +------+
+ *  | vnd_squeue_tx_drain() |--->| mac blocked? |----*---->| Done |
+ *  +-----------------------+    +--------------+          +------+
+ *                                       |                     |
+ *     +---------------------------------|---------------------+
+ *     |                                 |           tx        |
+ *     |                          no . . *           queue . . *
+ *     | flow controlled .               |           empty     * . fire pollout
+ *     |                 .               v                     |   if mblk_t's
+ *   +-------------+     .      +---------------------+        |   sent
+ *   | set blocked |<----*------| vnd_squeue_tx_one() |--------^-------+
+ *   | flags       |            +---------------------+                |
+ *   +-------------+    More data       |    |      |      More data   |
+ *                      and limit       ^    v      * . .  and limit   ^
+ *                      not reached . . *    |      |      reached     |
+ *                                      +----+      |                  |
+ *                                                  v                  |
+ *   +----------+          +-------------+    +---------------------------+
+ *   | mac flow |--------->| remove mac  |--->| gsqueue_enter_one() with  |
+ *   | control  |          | block flags |    | vnd_squeue_tx_drain() and |
+ *   | callback |          +-------------+    | GSQUEUE_FILL flag, iff    |
+ *   +----------+                             | not already scheduled     |
+ *                                            +---------------------------+
+ *
+ * The final path taken for a given write(9E)/frameio ioctl depends on whether
+ * or not the vnd_dev_t is non-blocking. That controls the initial path of
+ * trying to take a reservation in write data queue. If the device is in
+ * non-blocking mode, we'll return EAGAIN when there is not enough space
+ * available, otherwise, the calling thread blocks on the data queue.
+ *
+ * Today when we call into vnd_squeue_tx_drain() we will not try to drain the
+ * entire queue, as that could be quite large and we don't want to necessarily
+ * keep the thread that's doing the drain until it's been finished. Not only
+ * could more data be coming in, but the draining thread could be a userland
+ * thread that has more work to do. We have two limits today. There is an upper
+ * bound on the total amount of data and the total number of mblk_t chains. If
+ * we hit either limit, then we will schedule another drain in the gsqueue and
+ * go from there.
+ *
+ * It's worth taking some time to describe how we interact with gsqueues. vnd
+ * has a gsqueue_set_t for itself. It's important that it has its own set, as
+ * the profile of work that vnd does is different from other sub-systems in the
+ * kernel. When we open a STREAMS device in vnd_s_open, we get a random gsqueue.
+ * Unlike TCP/IP which uses an gsqueue for per TCP connection, we end up
+ * maintaining one for a given device. Because of that, we want to use a
+ * pseudo-random one to try and spread out the load, and picking one at random
+ * is likely to be just as good as any fancy algorithm we might come up with,
+ * especially as any two devices could have radically different transmit
+ * profiles.
+ *
+ * While some of the write path may seem complicated, it does allow us to
+ * maintain an important property. Once we have acknowledged a write(9E) or
+ * frameio ioctl, we will not drop the packet, excepting something like ipf via
+ * the firewall hooks.
+ *
+ * There is one other source of flow control that can exist in the system which
+ * is in the form of a barrier. The barrier is an internal mechanism used for
+ * ensuring that an gsqueue is drained for a given device. We use this as part
+ * of tearing down. Specifically we disable the write path so nothing new can be
+ * inserted into the gsqueue and then insert a barrier block. Once the barrier
+ * block comes out of the gsqueue, then we know nothing else in the gsqueue that
+ * could refer to the vnd_str_t, being destroyed, exists.
+ *
+ * ---------------------
+ * vnd, zones, netstacks
+ * ---------------------
+ *
+ * vnd devices are scoped to datalinks and datalinks are scoped to a netstack.
+ * Because of that, vnd is also a netstack module. It registers with the
+ * netstack sub-system and receives callbacks every time a netstack is created,
+ * being shutdown, and destroyed. The netstack callbacks drive the creation and
+ * destruction of the vnd_pnsd_t structures.
+ *
+ * Recall from the earlier architecture diagrams that every vnd device is scoped
+ * to a netstack and known about by a given vnd_pnsd_t. When that netstack is
+ * torn down, we also tear down any vnd devices that are hanging around. When
+ * the netstack is torn down, we know that any zones that are scoped to that
+ * netstack are being shut down and have no processes remaining. This is going
+ * to be the case whether they are shared or exclusive stack zones. We have to
+ * perform a careful dance.
+ *
+ * There are two different callbacks that happen on tear down, the first is a
+ * shutdown callback, the second is a destroy callback. When the shutdown
+ * callback is fired we need to prepare for the netstack to go away and ensure
+ * that nothing can continue to persist itself.
+ *
+ * More specifically, when we get notice of a stack being shutdown we first
+ * remove the netstack from the global netstack list to ensure that no one new
+ * can come in and find the netstack and get a reference to it. After that, we
+ * notify the neti hooks that they're going away. Once that's all done, we get
+ * to the heart of the matter.
+ *
+ * When shutting down there could be any number of outstanding contexts that
+ * have a reference on the vnd_pnsd_t and on the individual links. However, we
+ * know that no one new will be able to find the vnd_pnsd_t. To account for
+ * things that have existing references we mark the vnd_pnsd_t`vpnd_flags with
+ * VND_NS_CONDEMNED. This is checked by code paths that wish to append a device
+ * to the netstack's list. If this is set, then they must not append to it.
+ * Once this is set, we know that the netstack's list of devices can never grow,
+ * only shrink.
+ *
+ * Next, for each device we tag it with VND_D_ZONE_DYING. This indicates that
+ * the container for the device is being destroyed and that we should not allow
+ * additional references to the device to be created, whether via open, or
+ * linking. The presence of this bit also allows things like the list ioctl and
+ * sdev to know not to consider its existence. At the conclusion of this being
+ * set, we know that no one else should be able to obtain a new reference to the
+ * device.
+ *
+ * Once that has been set for all devices, we go through and remove any existing
+ * links that have been established in sdev. Because doing that may cause the
+ * final reference for the device to be dropped, which still has a reference to
+ * the netstack, we have to restart our walk due to dropped locks. We know that
+ * this walk will eventually complete because the device cannot be relinked and
+ * no new devices will be attached in this netstack due to VND_NS_CONDEMNED.
+ * Once that's finished, the shutdown callback returns.
+ *
+ * When we reach the destroy callback, we simply wait for references on the
+ * netstack to disappear. Because the zone has been shut down, all processes in
+ * it that have open references have been terminated and reaped. Any threads
+ * that are newly trying to reference it will fail. However, there is one thing
+ * that can halt this that we have no control over, which is the global zone
+ * holding open a reference to the device. In this case the zone halt will hang
+ * in vnd_stack_destroy. Once the last references is dropped we finish destroy
+ * the netinfo hooks and free the vnd_pnsd_t.
+ *
+ * ----
+ * sdev
+ * ----
+ *
+ * vnd registers a sdev plugin which allows it to dynamically fill out /dev/vnd
+ * for both the global and non-global zones. In any given zone we always supply
+ * a control node via /dev/vnd/ctl. This is the self-cloning node. Each zone
+ * will also have an entry per-link in that zone under /dev/vnd/%datalink, eg.
+ * if a link was named net0, there would be a /dev/vnd/net0. The global zone can
+ * also see every link for every zone, ala /dev/net, under
+ * /dev/vnd/%zonename/%datalink, eg. if a zone named 'turin' had a vnd device
+ * named net0, the global zone would have /dev/vnd/turin/net0.
+ *
+ * The sdev plugin has three interfaces that it supplies back to sdev. One is to
+ * validate that a given node is still valid. The next is a callback from sdev
+ * to say that it is no longer using the node. The third and final one is from
+ * sdev where it asks us to fill a directory. All of the heavy lifting is done
+ * in directory filling and in valiation. We opt not to maintain a reference on
+ * the device while there is an sdev node present. This makes the removal of
+ * nodes much simpler and most of the possible failure modes shouldn't cause any
+ * real problems. For example, the open path has to handle both dev_t's which no
+ * longer exist and which are no longer linked.
+ *
+ * -----
+ * hooks
+ * -----
+ *
+ * Like IP, vnd sends all L3 packets through its firewall hooks. Currently vnd
+ * provides these for L3 IP and IPv6 traffic. Each netstack provides these hooks
+ * in a minimal fashion. While we will allow traffic to be filtered through the
+ * hooks, we do not provide means for packet injection or additional inspection
+ * at this time. There are a total of four different events created:
+ *
+ *   o IPv4 physical in
+ *   o IPv4 physical out
+ *   o IPv6 physical in
+ *   o IPv6 physical out
+ *
+ * ---------------
+ * Synchronization
+ * ---------------
+ *
+ * To make our synchronization simpler, we've put more effort into making the
+ * metadata/setup paths do more work. That work allows the data paths to make
+ * assumptions around synchronization that simplify the general case. Each major
+ * structure, the vnd_pnsd_t, vnd_dev_t, vnd_str_t, and vnd_data_queue_t is
+ * annotated with the protection that its members receives.  The following
+ * annotations are used:
+ *
+ * 	A	Atomics; these values are only modified using atomics values.
+ *		Currently this only applies to kstat values.
+ * 	E	Existence; no lock is needed to access this member, it does not
+ *		change while the structure is valid.
+ * 	GL	Global Lock; these members are protected by the global
+ *		vnd_dev_lock.
+ * 	L	Locked; access to the member is controlled by a lock that is in
+ * 		the structure.
+ * 	NSL	netstack lock; this member is protected by the containing
+ * 		netstack. This only applies to the vnd_dev_t`vdd_nslink.
+ *	X	This member is special, and is discussed in this section.
+ *
+ * In addition to locking, we also have reference counts on the vnd_dev_t and
+ * the vnd_pnsd_t. The reference counts describe the lifetimes of the structure.
+ * With rare exception, once a reference count is decremented, the consumer
+ * should not assume that the data is valid any more. The only exception to this
+ * is the case where we're removing an extant reference count from a link into
+ * /devices or /dev. Reference counts are obtained on these structures as a part
+ * of looking them up.
+ *
+ * 	# Global Lock Ordering
+ * 	######################
+ *
+ * The following is the order that you must take locks in vnd:
+ *
+ * 1) vnd`vnd_dev_lock
+ * 2) vnd_pnsd_t`vpnd_lock
+ * 3) vnd_dev_t`vnd_lock
+ * 4) vnd_str_t`vns_lock
+ * 5) vnd_data_queue_t`vdq_lock
+ *
+ * One must adhere to the following rules:
+ *
+ *   o You must acquire a lower numbered lock before a high numbered lock.
+ *   o It is NOT legal to hold two locks of the same level concurrently, eg. you
+ *     can not hold two different vnd_dev_t's vnd_lock at the same time.
+ *   o You may release locks in any order.
+ *   o If you release a lock, you must honor the locking rules before acquiring
+ *     it again.
+ *   o You should not hold any locks when calling any of the rele functions.
+ *
+ * 	# Special Considerations
+ * 	########################
+ *
+ * While most of the locking is what's expected, it's worth going into the
+ * special nature that a few members hold.  Today, only two structures have
+ * special considerations: the vnd_dev_t and the vnd_str_t. All members with
+ * special considerations have an additional annotation that describes how you
+ * should interact with it.
+ *
+ * vnd_dev_t: The vdd_nsd and vdd_cr are only valid when the minor node is
+ * attached or in the process of attaching. If the code path that goes through
+ * requires an attached vnd_dev_t, eg. the data path and tear down path, then it
+ * is always legal to dereference that member without a lock held. When they are
+ * added to the system, they should be done under the vdd_lock and done as part
+ * of setting the VND_D_ATTACH_INFLIGHT flag. These should not change during the
+ * lifetime of the vnd_dev_t.
+ *
+ * vnd_dev_t: The vdd_ldih is similar to the vdd_nsd and vdd_cr, except that it
+ * always exists as it is a part of the structure. The only time that it's valid
+ * to be using it is during the attach path with the VND_D_ATTACH_INFLIGHT flag
+ * set or during tear down. Outside of those paths which are naturally
+ * serialized, there is no explicit locking around the member.
+ *
+ * vnd_str_t: The vns_dev and vns_nsd work in similar ways. They are not
+ * initially set as part of creating the structure, but are set as part of
+ * responding to the association ioctl. Anything in the data path or metadata
+ * path that requires association may assume that they exist, as we do not kick
+ * off the state machine until they're set.
+ *
+ * vnd_str_t: The vns_drainblk and vns_barrierblk are similarly special. The
+ * members are designed to be used as part of various operations with the
+ * gsqueues. A lock isn't needed to use them, but to work with them, the
+ * appropriate flag in the vnd_str_t`vns_flags must have been set by the current
+ * thread. Otherwise, it is always fair game to refer to their addresses. Their
+ * contents are ignored by vnd, but some members are manipulated by the gsqueue
+ * subsystem.
+ */
+
+#include <sys/conf.h>
+#include <sys/devops.h>
+#include <sys/modctl.h>
+#include <sys/stat.h>
+#include <sys/file.h>
+#include <sys/types.h>
+#include <sys/errno.h>
+#include <sys/open.h>
+#include <sys/ddi.h>
+#include <sys/ethernet.h>
+#include <sys/stropts.h>
+#include <sys/sunddi.h>
+#include <sys/stream.h>
+#include <sys/strsun.h>
+#include <sys/ksynch.h>
+#include <sys/taskq_impl.h>
+#include <sys/sdt.h>
+#include <sys/debug.h>
+#include <sys/sysmacros.h>
+#include <sys/dlpi.h>
+#include <sys/cred.h>
+#include <sys/id_space.h>
+#include <sys/list.h>
+#include <sys/ctype.h>
+#include <sys/policy.h>
+#include <sys/sunldi.h>
+#include <sys/cred.h>
+#include <sys/strsubr.h>
+#include <sys/poll.h>
+#include <sys/neti.h>
+#include <sys/hook.h>
+#include <sys/hook_event.h>
+#include <sys/vlan.h>
+#include <sys/dld.h>
+#include <sys/mac_client.h>
+#include <sys/netstack.h>
+#include <sys/fs/sdev_plugin.h>
+#include <sys/kstat.h>
+#include <sys/atomic.h>
+#include <sys/disp.h>
+#include <sys/random.h>
+#include <sys/gsqueue.h>
+
+#include <inet/ip.h>
+#include <inet/ip6.h>
+
+#include <sys/vnd.h>
+
+/*
+ * Globals
+ */
+static dev_info_t *vnd_dip;
+static taskq_t *vnd_taskq;
+static kmem_cache_t *vnd_str_cache;
+static kmem_cache_t *vnd_dev_cache;
+static kmem_cache_t *vnd_pnsd_cache;
+static id_space_t *vnd_minors;
+static int vnd_list_init = 0;
+static sdev_plugin_hdl_t vnd_sdev_hdl;
+static gsqueue_set_t *vnd_sqset;
+
+static kmutex_t vnd_dev_lock;
+static list_t vnd_dev_list;	/* Protected by the vnd_dev_lock */
+static list_t vnd_nsd_list;	/* Protected by the vnd_dev_lock */
+
+/*
+ * STREAMs ioctls
+ *
+ * The STREAMs ioctls are internal to vnd. No one should be seeing them, as such
+ * they aren't a part of the header file.
+ */
+#define	VND_STRIOC	(('v' << 24) | ('n' << 16) | ('d' << 8) | 0x80)
+
+/*
+ * Private ioctl to associate a given streams instance with a minor instance of
+ * the character device.
+ */
+#define	VND_STRIOC_ASSOCIATE	(VND_STRIOC | 0x1)
+
+typedef struct vnd_strioc_associate {
+	minor_t	vsa_minor;	/* minor device node */
+	netstackid_t vsa_nsid;	/* netstack id */
+	vnd_errno_t vsa_errno;	/* errno */
+} vnd_strioc_associate_t;
+
+typedef enum vnd_strioc_state {
+	VSS_UNKNOWN = 0,
+	VSS_COPYIN = 1,
+	VSS_COPYOUT = 2,
+} vnd_strioc_state_t;
+
+typedef struct vnd_strioc {
+	vnd_strioc_state_t vs_state;
+	caddr_t vs_addr;
+} vnd_strioc_t;
+
+/*
+ * VND SQUEUE TAGS, start at 0x42 so we don't overlap with extent tags. Though
+ * really, overlap is at the end of the day, inevitable.
+ */
+#define	VND_SQUEUE_TAG_TX_DRAIN		0x42
+#define	VND_SQUEUE_TAG_MAC_FLOW_CONTROL	0x43
+#define	VND_SQUEUE_TAG_VND_WRITE	0x44
+#define	VND_SQUEUE_TAG_ND_FRAMEIO_WRITE	0x45
+#define	VND_SQUEUE_TAG_STRBARRIER	0x46
+
+/*
+ * vnd reserved names. These are names which are reserved by vnd and thus
+ * shouldn't be used by some external program.
+ */
+static char *vnd_reserved_names[] = {
+	"ctl",
+	"zone",
+	NULL
+};
+
+/*
+ * vnd's DTrace probe macros
+ *
+ * DTRACE_VND* are all for a stable provider. We also have an unstable internal
+ * set of probes for reference count manipulation.
+ */
+#define	DTRACE_VND3(name, type1, arg1, type2, arg2, type3, arg3) \
+    DTRACE_PROBE3(__vnd_##name, type1, arg1, type2, arg2, type3, arg3);
+
+#define	DTRACE_VND4(name, type1, arg1, type2, arg2, type3, arg3, type4, arg4) \
+    DTRACE_PROBE4(__vnd_##name, type1, arg1, type2, arg2, type3, arg3, \
+	type4, arg4);
+
+#define	DTRACE_VND5(name, type1, arg1, type2, arg2, type3, arg3,	\
+    type4, arg4, type5, arg5)						\
+    DTRACE_PROBE5(__vnd_##name, type1, arg1, type2, arg2, type3, arg3,	\
+	type4, arg4, type5, arg5);
+
+#define	DTRACE_VND_REFINC(vdp) \
+    DTRACE_PROBE2(vnd__ref__inc, vnd_dev_t *, vdp, int, vdp->vdd_ref);
+#define	DTRACE_VND_REFDEC(vdp) \
+    DTRACE_PROBE2(vnd__ref__dec, vnd_dev_t *, vdp, int, vdp->vdd_ref);
+
+
+/*
+ * Tunables
+ */
+size_t vnd_vdq_default_size = 1024 * 64;	/* 64 KB */
+size_t vnd_vdq_hard_max = 1024 * 1024 * 4;	/* 4 MB */
+
+/*
+ * These numbers are designed as per-device tunables that are applied when a new
+ * vnd device is attached. They're a rough stab at what may be a reasonable
+ * amount of work to do in one burst in an squeue.
+ */
+size_t vnd_flush_burst_size = 1520 * 10;	/* 10 1500 MTU packets */
+size_t vnd_flush_nburst = 10;			/* 10 frames */
+
+/*
+ * Constants related to our sdev plugins
+ */
+#define	VND_SDEV_NAME	"vnd"
+#define	VND_SDEV_ROOT	"/dev/vnd"
+#define	VND_SDEV_ZROOT	"/dev/vnd/zone"
+
+/*
+ * Statistic macros
+ */
+#define	VND_STAT_INC(vsp, field, val) \
+    atomic_add_64(&(vsp)->vns_ksdata.field.value.ui64, val)
+#define	VND_LATENCY_1MS		1000000
+#define	VND_LATENCY_10MS	10000000
+#define	VND_LATENCY_100MS	100000000
+#define	VND_LATENCY_1S		1000000000
+#define	VND_LATENCY_10S		10000000000
+
+/*
+ * Constants for vnd hooks
+ */
+static uint8_t vnd_bcast_addr[6] = { 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF };
+#define	IPV4_MCAST_LEN	3
+static uint8_t vnd_ipv4_mcast[3] = { 0x01, 0x00, 0x5E };
+#define	IPV6_MCAST_LEN	2
+static uint8_t vnd_ipv6_mcast[2] = { 0x33, 0x33 };
+
+/*
+ * vnd internal data structures and types
+ */
+
+struct vnd_str;
+struct vnd_dev;
+struct vnd_pnsd;
+
+/*
+ * As part of opening the device stream we need to properly communicate with our
+ * underlying stream. This is a bit of an asynchronous dance and we need to
+ * properly work with dld to get everything set up. We have to initiate the
+ * conversation with dld and as such we keep track of our state here.
+ */
+typedef enum vnd_str_state {
+	VNS_S_INITIAL = 0,
+	VNS_S_INFO_SENT,
+	VNS_S_EXCLUSIVE_SENT,
+	VNS_S_ATTACH_SENT,
+	VNS_S_BIND_SENT,
+	VNS_S_SAP_PROMISC_SENT,
+	VNS_S_MULTI_PROMISC_SENT,
+	VNS_S_RX_ONLY_PROMISC_SENT,
+	VNS_S_CAPAB_Q_SENT,
+	VNS_S_CAPAB_E_SENT,
+	VNS_S_ONLINE,
+	VNS_S_SHUTTING_DOWN,
+	VNS_S_ZOMBIE
+} vnd_str_state_t;
+
+typedef enum vnd_str_flags {
+	VNS_F_NEED_ZONE = 0x1,
+	VNS_F_TASKQ_DISPATCHED = 0x2,
+	VNS_F_CONDEMNED = 0x4,
+	VNS_F_FLOW_CONTROLLED = 0x8,
+	VNS_F_DRAIN_SCHEDULED = 0x10,
+	VNS_F_BARRIER = 0x20,
+	VNS_F_BARRIER_DONE = 0x40
+} vnd_str_flags_t;
+
+typedef enum vnd_capab_flags {
+	VNS_C_HCKSUM = 0x1,
+	VNS_C_DLD = 0x2,
+	VNS_C_DIRECT = 0x4,
+	VNS_C_HCKSUM_BADVERS = 0x8
+} vnd_capab_flags_t;
+
+/*
+ * Definitions to interact with direct callbacks
+ */
+typedef void (*vnd_rx_t)(struct vnd_str *, mac_resource_t *, mblk_t *,
+    mac_header_info_t *);
+typedef uintptr_t vnd_mac_cookie_t;
+/* DLD Direct capability function */
+typedef int (*vnd_dld_cap_t)(void *, uint_t, void *, uint_t);
+/* DLD Direct tx function */
+typedef vnd_mac_cookie_t (*vnd_dld_tx_t)(void *, mblk_t *, uint64_t, uint16_t);
+/* DLD Direct function to set flow control callback */
+typedef void *(*vnd_dld_set_fcb_t)(void *, void (*)(void *, vnd_mac_cookie_t),
+    void *);
+/* DLD Direct function to see if flow controlled still */
+typedef int (*vnd_dld_is_fc_t)(void *, vnd_mac_cookie_t);
+
+/*
+ * The vnd_str_capab_t is always protected by the vnd_str_t it's a member of.
+ */
+typedef struct vnd_str_capab {
+	vnd_capab_flags_t vsc_flags;
+	t_uscalar_t vsc_hcksum_opts;
+	vnd_dld_cap_t vsc_capab_f;
+	void *vsc_capab_hdl;
+	vnd_dld_tx_t vsc_tx_f;
+	void *vsc_tx_hdl;
+	vnd_dld_set_fcb_t vsc_set_fcb_f;
+	void *vsc_set_fcb_hdl;
+	vnd_dld_is_fc_t vsc_is_fc_f;
+	void *vsc_is_fc_hdl;
+	vnd_mac_cookie_t vsc_fc_cookie;
+	void *vsc_tx_fc_hdl;
+} vnd_str_capab_t;
+
+/*
+ * The vnd_data_queue is a simple construct for storing a series of messages in
+ * a queue.
+ *
+ * See synchronization section of the big theory statement for member
+ * annotations.
+ */
+typedef struct vnd_data_queue {
+	struct vnd_str *vdq_vns;	/* E */
+	kmutex_t vdq_lock;
+	kcondvar_t vdq_ready;		/* Uses vdq_lock */
+	ssize_t vdq_max;		/* L */
+	ssize_t vdq_cur;		/* L */
+	mblk_t *vdq_head;		/* L */
+	mblk_t *vdq_tail;		/* L */
+} vnd_data_queue_t;
+
+typedef struct vnd_str_stat {
+	kstat_named_t	vks_rbytes;
+	kstat_named_t	vks_rpackets;
+	kstat_named_t	vks_obytes;
+	kstat_named_t	vks_opackets;
+	kstat_named_t	vks_nhookindrops;
+	kstat_named_t	vks_nhookoutdrops;
+	kstat_named_t	vks_ndlpidrops;
+	kstat_named_t	vks_ndataindrops;
+	kstat_named_t	vks_ndataoutdrops;
+	kstat_named_t	vks_tdrops;
+	kstat_named_t	vks_linkname;
+	kstat_named_t	vks_zonename;
+	kstat_named_t	vks_nmacflow;
+	kstat_named_t	vks_tmacflow;
+	kstat_named_t	vks_mac_flow_1ms;
+	kstat_named_t	vks_mac_flow_10ms;
+	kstat_named_t	vks_mac_flow_100ms;
+	kstat_named_t	vks_mac_flow_1s;
+	kstat_named_t	vks_mac_flow_10s;
+} vnd_str_stat_t;
+
+/*
+ * vnd stream structure
+ *
+ * See synchronization section of the big theory statement for member
+ * annotations.
+ */
+typedef struct vnd_str {
+	kmutex_t 	vns_lock;
+	kcondvar_t	vns_cancelcv;		/* Uses vns_lock */
+	kcondvar_t	vns_barriercv;		/* Uses vns_lock */
+	kcondvar_t	vns_stcv;		/* Uses vns_lock */
+	vnd_str_state_t	vns_state;		/* L */
+	vnd_str_state_t	vns_laststate;		/* L */
+	vnd_errno_t	vns_errno;		/* L */
+	vnd_str_flags_t	vns_flags;		/* L */
+	vnd_str_capab_t vns_caps;		/* L */
+	taskq_ent_t	vns_tqe;		/* L */
+	vnd_data_queue_t vns_dq_read;		/* E */
+	vnd_data_queue_t vns_dq_write;		/* E */
+	mblk_t		*vns_dlpi_inc;		/* L */
+	queue_t		*vns_rq;		/* E */
+	queue_t		*vns_wq;		/* E */
+	queue_t		*vns_lrq;		/* E */
+	t_uscalar_t	vns_dlpi_style;		/* L */
+	t_uscalar_t	vns_minwrite;		/* L */
+	t_uscalar_t	vns_maxwrite;		/* L */
+	hrtime_t	vns_fclatch;		/* L */
+	hrtime_t	vns_fcupdate;		/* L */
+	kstat_t		*vns_kstat;		/* E */
+	gsqueue_t	*vns_squeue;		/* E */
+	mblk_t		vns_drainblk;		/* E + X */
+	mblk_t		vns_barrierblk;		/* E + X */
+	vnd_str_stat_t	vns_ksdata;		/* A */
+	size_t		vns_nflush;		/* L */
+	size_t 		vns_bsize;		/* L */
+	struct vnd_dev	*vns_dev;		/* E + X */
+	struct vnd_pnsd	*vns_nsd;		/* E + X */
+} vnd_str_t;
+
+typedef enum vnd_dev_flags {
+	VND_D_ATTACH_INFLIGHT =	0x001,
+	VND_D_ATTACHED =	0x002,
+	VND_D_LINK_INFLIGHT =	0x004,
+	VND_D_LINKED =		0x008,
+	VND_D_CONDEMNED =	0x010,
+	VND_D_ZONE_DYING =	0x020,
+	VND_D_OPENED =		0x040
+} vnd_dev_flags_t;
+
+/*
+ * This represents the data associated with a minor device instance.
+ *
+ * See synchronization section of the big theory statement for member
+ * annotations.
+ */
+typedef struct vnd_dev {
+	kmutex_t	vdd_lock;
+	list_node_t	vdd_link;			/* GL */
+	list_node_t	vdd_nslink;			/* NSL */
+	int		vdd_ref;			/* L */
+	vnd_dev_flags_t	vdd_flags;			/* L */
+	minor_t		vdd_minor;			/* E */
+	dev_t		vdd_devid;			/* E */
+	ldi_ident_t	vdd_ldiid;			/* E */
+	ldi_handle_t	vdd_ldih;			/* X */
+	cred_t		*vdd_cr;			/* X */
+	vnd_str_t	*vdd_str;			/* L */
+	struct pollhead	vdd_ph;				/* E */
+	struct vnd_pnsd *vdd_nsd;			/* E + X */
+	char		vdd_datalink[VND_NAMELEN];	/* L */
+	char		vdd_lname[VND_NAMELEN];		/* L */
+} vnd_dev_t;
+
+typedef enum vnd_pnsd_flags {
+	VND_NS_CONDEMNED = 0x1
+} vnd_pnsd_flags_t;
+
+/*
+ * Per netstack data structure.
+ *
+ * See synchronization section of the big theory statement for member
+ * annotations.
+ */
+typedef struct vnd_pnsd {
+	list_node_t vpnd_link;	/* protected by global dev lock */
+	zoneid_t vpnd_zid;			/* E */
+	netstackid_t vpnd_nsid;			/* E */
+	boolean_t vpnd_hooked;			/* E */
+	net_handle_t vpnd_neti_v4;		/* E */
+	hook_family_t vpnd_family_v4;		/* E */
+	hook_event_t vpnd_event_in_v4;		/* E */
+	hook_event_t vpnd_event_out_v4;		/* E */
+	hook_event_token_t vpnd_token_in_v4;	/* E */
+	hook_event_token_t vpnd_token_out_v4;	/* E */
+	net_handle_t vpnd_neti_v6;		/* E */
+	hook_family_t vpnd_family_v6;		/* E */
+	hook_event_t vpnd_event_in_v6;		/* E */
+	hook_event_t vpnd_event_out_v6;		/* E */
+	hook_event_token_t vpnd_token_in_v6;	/* E */
+	hook_event_token_t vpnd_token_out_v6;	/* E */
+	kmutex_t vpnd_lock;		/* Protects remaining members */
+	kcondvar_t vpnd_ref_change;		/* Uses vpnd_lock */
+	int vpnd_ref;				/* L */
+	vnd_pnsd_flags_t vpnd_flags;		/* L */
+	list_t vpnd_dev_list;			/* L */
+} vnd_pnsd_t;
+
+static void vnd_squeue_tx_drain(void *, mblk_t *, gsqueue_t *, void *);
+
+/*
+ * Drop function signature.
+ */
+typedef void (*vnd_dropper_f)(vnd_str_t *, mblk_t *, const char *);
+
+static void
+vnd_drop_ctl(vnd_str_t *vsp, mblk_t *mp, const char *reason)
+{
+	DTRACE_VND4(drop__ctl, mblk_t *, mp, vnd_str_t *, vsp, mblk_t *,
+	    mp, const char *, reason);
+	if (mp != NULL) {
+		freemsg(mp);
+	}
+	VND_STAT_INC(vsp, vks_ndlpidrops, 1);
+	VND_STAT_INC(vsp, vks_tdrops, 1);
+}
+
+static void
+vnd_drop_in(vnd_str_t *vsp, mblk_t *mp, const char *reason)
+{
+	DTRACE_VND4(drop__in, mblk_t *, mp, vnd_str_t *, vsp, mblk_t *,
+	    mp, const char *, reason);
+	if (mp != NULL) {
+		freemsg(mp);
+	}
+	VND_STAT_INC(vsp, vks_ndataindrops, 1);
+	VND_STAT_INC(vsp, vks_tdrops, 1);
+}
+
+static void
+vnd_drop_out(vnd_str_t *vsp, mblk_t *mp, const char *reason)
+{
+	DTRACE_VND4(drop__out, mblk_t *, mp, vnd_str_t *, vsp, mblk_t *,
+	    mp, const char *, reason);
+	if (mp != NULL) {
+		freemsg(mp);
+	}
+	VND_STAT_INC(vsp, vks_ndataoutdrops, 1);
+	VND_STAT_INC(vsp, vks_tdrops, 1);
+}
+
+static void
+vnd_drop_hook_in(vnd_str_t *vsp, mblk_t *mp, const char *reason)
+{
+	DTRACE_VND4(drop__in, mblk_t *, mp, vnd_str_t *, vsp, mblk_t *,
+	    mp, const char *, reason);
+	if (mp != NULL) {
+		freemsg(mp);
+	}
+	VND_STAT_INC(vsp, vks_nhookindrops, 1);
+	VND_STAT_INC(vsp, vks_tdrops, 1);
+}
+
+static void
+vnd_drop_hook_out(vnd_str_t *vsp, mblk_t *mp, const char *reason)
+{
+	DTRACE_VND4(drop__out, mblk_t *, mp, vnd_str_t *, vsp, mblk_t *,
+	    mp, const char *, reason);
+	if (mp != NULL) {
+		freemsg(mp);
+	}
+	VND_STAT_INC(vsp, vks_nhookoutdrops, 1);
+	VND_STAT_INC(vsp, vks_tdrops, 1);
+}
+
+static void
+vnd_drop_panic(vnd_str_t *vsp, mblk_t *mp, const char *reason)
+{
+	panic("illegal vnd drop");
+}
+
+static void
+vnd_mac_drop_input(vnd_str_t *vsp, mac_resource_t *unused, mblk_t *mp_chain,
+    mac_header_info_t *mhip)
+{
+	mblk_t *mp;
+
+	while (mp_chain != NULL) {
+		mp = mp_chain;
+		mp_chain = mp->b_next;
+		vnd_drop_hook_in(vsp, mp, "stream not associated");
+	}
+}
+
+static vnd_pnsd_t *
+vnd_nsd_lookup(netstackid_t nsid)
+{
+	vnd_pnsd_t *nsp;
+
+	mutex_enter(&vnd_dev_lock);
+	for (nsp = list_head(&vnd_nsd_list); nsp != NULL;
+	    nsp = list_next(&vnd_nsd_list, nsp)) {
+		if (nsp->vpnd_nsid == nsid) {
+			mutex_enter(&nsp->vpnd_lock);
+			VERIFY(nsp->vpnd_ref >= 0);
+			nsp->vpnd_ref++;
+			mutex_exit(&nsp->vpnd_lock);
+			break;
+		}
+	}
+	mutex_exit(&vnd_dev_lock);
+	return (nsp);
+}
+
+static vnd_pnsd_t *
+vnd_nsd_lookup_by_zid(zoneid_t zid)
+{
+	netstack_t *ns;
+	vnd_pnsd_t *nsp;
+	ns = netstack_find_by_zoneid(zid);
+	if (ns == NULL)
+		return (NULL);
+	nsp = vnd_nsd_lookup(ns->netstack_stackid);
+	netstack_rele(ns);
+	return (nsp);
+}
+
+static vnd_pnsd_t *
+vnd_nsd_lookup_by_zonename(char *zname)
+{
+	zone_t *zonep;
+	vnd_pnsd_t *nsp;
+
+	zonep = zone_find_by_name(zname);
+	if (zonep == NULL)
+		return (NULL);
+
+	nsp = vnd_nsd_lookup_by_zid(zonep->zone_id);
+	zone_rele(zonep);
+	return (nsp);
+}
+
+static void
+vnd_nsd_ref(vnd_pnsd_t *nsp)
+{
+	mutex_enter(&nsp->vpnd_lock);
+	/*
+	 * This can only be used on something that has been obtained through
+	 * some other means. As such, the caller should already have a reference
+	 * before adding another one. This function should not be used as a
+	 * means of creating the initial reference.
+	 */
+	VERIFY(nsp->vpnd_ref > 0);
+	nsp->vpnd_ref++;
+	mutex_exit(&nsp->vpnd_lock);
+	cv_broadcast(&nsp->vpnd_ref_change);
+}
+
+static void
+vnd_nsd_rele(vnd_pnsd_t *nsp)
+{
+	mutex_enter(&nsp->vpnd_lock);
+	VERIFY(nsp->vpnd_ref > 0);
+	nsp->vpnd_ref--;
+	mutex_exit(&nsp->vpnd_lock);
+	cv_broadcast(&nsp->vpnd_ref_change);
+}
+
+static vnd_dev_t *
+vnd_dev_lookup(minor_t m)
+{
+	vnd_dev_t *vdp;
+	mutex_enter(&vnd_dev_lock);
+	for (vdp = list_head(&vnd_dev_list); vdp != NULL;
+	    vdp = list_next(&vnd_dev_list, vdp)) {
+		if (vdp->vdd_minor == m) {
+			mutex_enter(&vdp->vdd_lock);
+			VERIFY(vdp->vdd_ref > 0);
+			vdp->vdd_ref++;
+			DTRACE_VND_REFINC(vdp);
+			mutex_exit(&vdp->vdd_lock);
+			break;
+		}
+	}
+	mutex_exit(&vnd_dev_lock);
+	return (vdp);
+}
+
+static void
+vnd_dev_free(vnd_dev_t *vdp)
+{
+	/*
+	 * When the STREAM exists we need to go through and make sure
+	 * communication gets torn down. As part of closing the stream, we
+	 * guarantee that nothing else should be able to enter the stream layer
+	 * at this point. That means no one should be able to call
+	 * read(),write() or one of the frameio ioctls.
+	 */
+	if (vdp->vdd_flags & VND_D_ATTACHED) {
+		ldi_close(vdp->vdd_ldih, FREAD | FWRITE, vdp->vdd_cr);
+		crfree(vdp->vdd_cr);
+		vdp->vdd_cr = NULL;
+
+		/*
+		 * We have to remove ourselves from our parents list now. It is
+		 * really quite important that we have already set the condemend
+		 * flag here so that our containing netstack basically knows
+		 * that we're on the way down and knows not to wait for us. It's
+		 * also important that we do that before we put a rele on the
+		 * the device as that is the point at which it will check again.
+		 */
+		mutex_enter(&vdp->vdd_nsd->vpnd_lock);
+		list_remove(&vdp->vdd_nsd->vpnd_dev_list, vdp);
+		mutex_exit(&vdp->vdd_nsd->vpnd_lock);
+		vnd_nsd_rele(vdp->vdd_nsd);
+		vdp->vdd_nsd = NULL;
+	}
+	ASSERT(vdp->vdd_flags & VND_D_CONDEMNED);
+	id_free(vnd_minors, vdp->vdd_minor);
+	mutex_destroy(&vdp->vdd_lock);
+	kmem_cache_free(vnd_dev_cache, vdp);
+}
+
+static void
+vnd_dev_ref(vnd_dev_t *vdp)
+{
+	mutex_enter(&vdp->vdd_lock);
+	VERIFY(vdp->vdd_ref > 0);
+	vdp->vdd_ref++;
+	DTRACE_VND_REFINC(vdp);
+	mutex_exit(&vdp->vdd_lock);
+}
+
+/*
+ * As part of releasing the hold on this we may tear down a given vnd_dev_t As
+ * such we need to make sure that we grab the list lock first before grabbing
+ * the vnd_dev_t's lock to ensure proper lock ordering.
+ */
+static void
+vnd_dev_rele(vnd_dev_t *vdp)
+{
+	mutex_enter(&vnd_dev_lock);
+	mutex_enter(&vdp->vdd_lock);
+	VERIFY(vdp->vdd_ref > 0);
+	vdp->vdd_ref--;
+	DTRACE_VND_REFDEC(vdp);
+	if (vdp->vdd_ref > 0) {
+		mutex_exit(&vdp->vdd_lock);
+		mutex_exit(&vnd_dev_lock);
+		return;
+	}
+
+	/*
+	 * Now that we've removed this from the list, we can go ahead and
+	 * drop the list lock. No one else can find this device and reference
+	 * it. As its reference count is zero, it by definition does not have
+	 * any remaining entries in /devices that could lead someone back to
+	 * this.
+	 */
+	vdp->vdd_flags |= VND_D_CONDEMNED;
+	list_remove(&vnd_dev_list, vdp);
+	mutex_exit(&vdp->vdd_lock);
+	mutex_exit(&vnd_dev_lock);
+
+	vnd_dev_free(vdp);
+}
+
+/*
+ * Insert a mesage block chain if there's space, otherwise drop it. Return one
+ * so someone who was waiting for data would now end up having found it. eg.
+ * caller should consider a broadcast.
+ */
+static int
+vnd_dq_push(vnd_data_queue_t *vqp, mblk_t *mp, boolean_t reserved,
+    vnd_dropper_f dropf)
+{
+	size_t msize;
+
+	ASSERT(MUTEX_HELD(&vqp->vdq_lock));
+	if (reserved == B_FALSE) {
+		msize = msgsize(mp);
+		if (vqp->vdq_cur + msize > vqp->vdq_max) {
+			dropf(vqp->vdq_vns, mp, "buffer full");
+			return (0);
+		}
+		vqp->vdq_cur += msize;
+	}
+
+	if (vqp->vdq_head == NULL) {
+		ASSERT(vqp->vdq_tail == NULL);
+		vqp->vdq_head = mp;
+		vqp->vdq_tail = mp;
+	} else {
+		vqp->vdq_tail->b_next = mp;
+		vqp->vdq_tail = mp;
+	}
+
+	return (1);
+}
+
+/*
+ * Remove a message message block chain. If the amount of space in the buffer
+ * has changed we return 1. We have no way of knowing whether or not there is
+ * enough space overall for a given writer who is blocked, so we always end up
+ * having to return true and thus tell consumers that they should consider
+ * signalling.
+ */
+static int
+vnd_dq_pop(vnd_data_queue_t *vqp, mblk_t **mpp)
+{
+	size_t msize;
+	mblk_t *mp;
+
+	ASSERT(MUTEX_HELD(&vqp->vdq_lock));
+	ASSERT(mpp != NULL);
+	if (vqp->vdq_head == NULL) {
+		ASSERT(vqp->vdq_tail == NULL);
+		*mpp = NULL;
+		return (0);
+	}
+
+	mp = vqp->vdq_head;
+	msize = msgsize(mp);
+
+	vqp->vdq_cur -= msize;
+	if (mp->b_next == NULL) {
+		vqp->vdq_head = NULL;
+		vqp->vdq_tail = NULL;
+		/*
+		 * We can't be certain that this is always going to be zero.
+		 * Someone may have basically taken a reservation of space on
+		 * the data queue, eg. claimed spae but not yet pushed it on
+		 * yet.
+		 */
+		ASSERT(vqp->vdq_cur >= 0);
+	} else {
+		vqp->vdq_head = mp->b_next;
+		ASSERT(vqp->vdq_cur > 0);
+	}
+	mp->b_next = NULL;
+	*mpp = mp;
+	return (1);
+}
+
+/*
+ * Reserve space in the queue. This will bump up the size of the queue and
+ * entitle the user to push something on later without bumping the space.
+ */
+static int
+vnd_dq_reserve(vnd_data_queue_t *vqp, ssize_t size)
+{
+	ASSERT(MUTEX_HELD(&vqp->vdq_lock));
+	ASSERT(size >= 0);
+
+	if (size == 0)
+		return (0);
+
+	if (size + vqp->vdq_cur > vqp->vdq_max)
+		return (0);
+
+	vqp->vdq_cur += size;
+	return (1);
+}
+
+static void
+vnd_dq_unreserve(vnd_data_queue_t *vqp, ssize_t size)
+{
+	ASSERT(MUTEX_HELD(&vqp->vdq_lock));
+	ASSERT(size > 0);
+	ASSERT(size <= vqp->vdq_cur);
+
+	vqp->vdq_cur -= size;
+}
+
+static void
+vnd_dq_flush(vnd_data_queue_t *vqp, vnd_dropper_f dropf)
+{
+	mblk_t *mp, *next;
+
+	mutex_enter(&vqp->vdq_lock);
+	for (mp = vqp->vdq_head; mp != NULL; mp = next) {
+		next = mp->b_next;
+		mp->b_next = NULL;
+		dropf(vqp->vdq_vns, mp, "vnd_dq_flush");
+	}
+	vqp->vdq_cur = 0;
+	vqp->vdq_head = NULL;
+	vqp->vdq_tail = NULL;
+	mutex_exit(&vqp->vdq_lock);
+}
+
+static boolean_t
+vnd_dq_is_empty(vnd_data_queue_t *vqp)
+{
+	boolean_t ret;
+
+	mutex_enter(&vqp->vdq_lock);
+	if (vqp->vdq_head == NULL)
+		ret = B_TRUE;
+	else
+		ret = B_FALSE;
+	mutex_exit(&vqp->vdq_lock);
+
+	return (ret);
+}
+
+/*
+ * Get a network uint16_t from the message and translate it into something the
+ * host understands.
+ */
+static int
+vnd_mbc_getu16(mblk_t *mp, off_t off, uint16_t *out)
+{
+	size_t mpsize;
+	uint8_t *bp;
+
+	mpsize = msgsize(mp);
+	/* Check for overflow */
+	if (off + sizeof (uint16_t) > mpsize)
+		return (1);
+
+	mpsize = MBLKL(mp);
+	while (off >= mpsize) {
+		mp = mp->b_cont;
+		off -= mpsize;
+		mpsize = MBLKL(mp);
+	}
+
+	/*
+	 * Data is in network order. Note the second byte of data might be in
+	 * the next mp.
+	 */
+	bp = mp->b_rptr + off;
+	*out = *bp << 8;
+	if (off + 1 == mpsize) {
+		mp = mp->b_cont;
+		bp = mp->b_rptr;
+	} else {
+		bp++;
+	}
+
+	*out |= *bp;
+	return (0);
+}
+
+/*
+ * Given an mblk chain find the mblk and address of a particular offset.
+ */
+static int
+vnd_mbc_getoffset(mblk_t *mp, off_t off, mblk_t **mpp, uintptr_t *offp)
+{
+	size_t mpsize;
+
+	if (off >= msgsize(mp))
+		return (1);
+
+	mpsize = MBLKL(mp);
+	while (off >= mpsize) {
+		mp = mp->b_cont;
+		off -= mpsize;
+		mpsize = MBLKL(mp);
+	}
+	*mpp = mp;
+	*offp = (uintptr_t)mp->b_rptr + off;
+
+	return (0);
+}
+
+/*
+ * Fetch the destination mac address. Set *dstp to that mac address. If the data
+ * is not contiguous in the first mblk_t, fill in datap and set *dstp to it.
+ */
+static int
+vnd_mbc_getdstmac(mblk_t *mp, uint8_t **dstpp, uint8_t *datap)
+{
+	int i;
+
+	if (MBLKL(mp) >= ETHERADDRL) {
+		*dstpp = mp->b_rptr;
+		return (0);
+	}
+
+	*dstpp = datap;
+	for (i = 0; i < ETHERADDRL; i += 2, datap += 2) {
+		if (vnd_mbc_getu16(mp, i, (uint16_t *)datap) != 0)
+			return (1);
+	}
+
+	return (0);
+}
+
+static int
+vnd_hook(vnd_str_t *vsp, mblk_t **mpp, net_handle_t netiv4, hook_event_t hev4,
+    hook_event_token_t hetv4, net_handle_t netiv6, hook_event_t hev6,
+    hook_event_token_t hetv6, vnd_dropper_f hdrop, vnd_dropper_f ddrop)
+{
+	uint16_t etype;
+	int vlan = 0;
+	hook_pkt_event_t info;
+	size_t offset, mblen;
+	uint8_t *dstp;
+	uint8_t dstaddr[6];
+	hook_event_t he;
+	hook_event_token_t het;
+	net_handle_t neti;
+
+	/*
+	 * Before we can ask if we're interested we have to do enough work to
+	 * determine the ethertype.
+	 */
+
+	/* Byte 12 is either the VLAN tag or the ethertype */
+	if (vnd_mbc_getu16(*mpp, 12, &etype) != 0) {
+		ddrop(vsp, *mpp, "packet has incomplete ethernet header");
+		*mpp = NULL;
+		return (1);
+	}
+
+	if (etype == ETHERTYPE_VLAN) {
+		vlan = 1;
+		/* Actual ethertype is another four bytes in */
+		if (vnd_mbc_getu16(*mpp, 16, &etype) != 0) {
+			ddrop(vsp, *mpp,
+			    "packet has incomplete ethernet vlan header");
+			*mpp = NULL;
+			return (1);
+		}
+		offset = sizeof (struct ether_vlan_header);
+	} else {
+		offset = sizeof (struct ether_header);
+	}
+
+	/*
+	 * At the moment we only hook on the kinds of things that the IP module
+	 * would normally.
+	 */
+	if (etype != ETHERTYPE_IP && etype != ETHERTYPE_IPV6)
+		return (0);
+
+	if (etype == ETHERTYPE_IP) {
+		neti = netiv4;
+		he = hev4;
+		het = hetv4;
+	} else {
+		neti = netiv6;
+		he = hev6;
+		het = hetv6;
+	}
+
+	if (!he.he_interested)
+		return (0);
+
+
+	if (vnd_mbc_getdstmac(*mpp, &dstp, dstaddr) != 0) {
+		ddrop(vsp, *mpp, "packet has incomplete ethernet header");
+		*mpp = NULL;
+		return (1);
+	}
+
+	/*
+	 * Now that we know we're interested, we have to do some additional
+	 * sanity checking for IPF's sake, ala ip_check_length(). Specifically
+	 * we need to check to make sure that the remaining packet size,
+	 * excluding MAC, is at least the size of an IP header.
+	 */
+	mblen = msgsize(*mpp);
+	if ((etype == ETHERTYPE_IP &&
+	    mblen - offset < IP_SIMPLE_HDR_LENGTH) ||
+	    (etype == ETHERTYPE_IPV6 && mblen - offset < IPV6_HDR_LEN)) {
+		ddrop(vsp, *mpp, "packet has invalid IP header");
+		*mpp = NULL;
+		return (1);
+	}
+
+	info.hpe_protocol = neti;
+	info.hpe_ifp = (phy_if_t)vsp;
+	info.hpe_ofp = (phy_if_t)vsp;
+	info.hpe_mp = mpp;
+	info.hpe_flags = 0;
+
+	if (bcmp(vnd_bcast_addr, dstp, ETHERADDRL) == 0)
+		info.hpe_flags |= HPE_BROADCAST;
+	else if (etype == ETHERTYPE_IP &&
+	    bcmp(vnd_ipv4_mcast, vnd_bcast_addr, IPV4_MCAST_LEN) == 0)
+		info.hpe_flags |= HPE_MULTICAST;
+	else if (etype == ETHERTYPE_IPV6 &&
+	    bcmp(vnd_ipv6_mcast, vnd_bcast_addr, IPV6_MCAST_LEN) == 0)
+		info.hpe_flags |= HPE_MULTICAST;
+
+	if (vnd_mbc_getoffset(*mpp, offset, &info.hpe_mb,
+	    (uintptr_t *)&info.hpe_hdr) != 0) {
+		ddrop(vsp, *mpp, "packet too small -- "
+		    "unable to find payload");
+		*mpp = NULL;
+		return (1);
+	}
+
+	if (hook_run(neti->netd_hooks, het, (hook_data_t)&info) != 0) {
+		hdrop(vsp, *mpp, "drooped by hooks");
+		return (1);
+	}
+
+	return (0);
+}
+
+/*
+ * This should not be used for DL_INFO_REQ.
+ */
+static mblk_t *
+vnd_dlpi_alloc(size_t len, t_uscalar_t prim)
+{
+	mblk_t *mp;
+	mp = allocb(len, BPRI_MED);
+	if (mp == NULL)
+		return (NULL);
+
+	mp->b_datap->db_type = M_PROTO;
+	mp->b_wptr = mp->b_rptr + len;
+	bzero(mp->b_rptr, len);
+	((dl_unitdata_req_t *)mp->b_rptr)->dl_primitive = prim;
+
+	return (mp);
+}
+
+static void
+vnd_dlpi_inc_push(vnd_str_t *vsp, mblk_t *mp)
+{
+	mblk_t **mpp;
+
+	VERIFY(MUTEX_HELD(&vsp->vns_lock));
+	ASSERT(mp->b_next == NULL);
+	mpp = &vsp->vns_dlpi_inc;
+	while (*mpp != NULL)
+		mpp = &((*mpp)->b_next);
+	*mpp = mp;
+}
+
+static mblk_t *
+vnd_dlpi_inc_pop(vnd_str_t *vsp)
+{
+	mblk_t *mp;
+
+	VERIFY(MUTEX_HELD(&vsp->vns_lock));
+	mp = vsp->vns_dlpi_inc;
+	if (mp != NULL) {
+		VERIFY(mp->b_next == NULL || mp->b_next != mp);
+		vsp->vns_dlpi_inc = mp->b_next;
+		mp->b_next = NULL;
+	}
+	return (mp);
+}
+
+static int
+vnd_st_sinfo(vnd_str_t *vsp)
+{
+	mblk_t *mp;
+	dl_info_req_t *dlir;
+
+	VERIFY(MUTEX_HELD(&vsp->vns_lock));
+	mp = allocb(MAX(sizeof (dl_info_req_t), sizeof (dl_info_ack_t)),
+	    BPRI_HI);
+	if (mp == NULL) {
+		vsp->vns_errno = VND_E_NOMEM;
+		return (1);
+	}
+	vsp->vns_state = VNS_S_INFO_SENT;
+	cv_broadcast(&vsp->vns_stcv);
+
+	mp->b_datap->db_type = M_PCPROTO;
+	dlir = (dl_info_req_t *)mp->b_rptr;
+	mp->b_wptr = (uchar_t *)&dlir[1];
+	dlir->dl_primitive = DL_INFO_REQ;
+	putnext(vsp->vns_wq, mp);
+
+	return (0);
+}
+
+static int
+vnd_st_info(vnd_str_t *vsp)
+{
+	dl_info_ack_t *dlia;
+	mblk_t *mp;
+
+	VERIFY(MUTEX_HELD(&vsp->vns_lock));
+	mp = vnd_dlpi_inc_pop(vsp);
+	dlia = (dl_info_ack_t *)mp->b_rptr;
+	vsp->vns_dlpi_style = dlia->dl_provider_style;
+	vsp->vns_minwrite = dlia->dl_min_sdu;
+	vsp->vns_maxwrite = dlia->dl_max_sdu;
+
+	/*
+	 * At this time we only support DL_ETHER devices.
+	 */
+	if (dlia->dl_mac_type != DL_ETHER) {
+		freemsg(mp);
+		vsp->vns_errno = VND_E_NOTETHER;
+		return (1);
+	}
+
+	/*
+	 * Because vnd operates on entire packets, we need to manually account
+	 * for the ethernet header information. We add the size of the
+	 * ether_vlan_header to account for this, regardless if it is using
+	 * vlans or not.
+	 */
+	vsp->vns_maxwrite += sizeof (struct ether_vlan_header);
+
+	freemsg(mp);
+	return (0);
+}
+
+static int
+vnd_st_sexclusive(vnd_str_t *vsp)
+{
+	mblk_t *mp;
+
+	VERIFY(MUTEX_HELD(&vsp->vns_lock));
+	mp = vnd_dlpi_alloc(sizeof (dl_attach_req_t), DL_EXCLUSIVE_REQ);
+	if (mp == NULL) {
+		vsp->vns_errno = VND_E_NOMEM;
+		return (1);
+	}
+
+	vsp->vns_state = VNS_S_EXCLUSIVE_SENT;
+	cv_broadcast(&vsp->vns_stcv);
+	putnext(vsp->vns_wq, mp);
+	return (0);
+}
+
+static int
+vnd_st_exclusive(vnd_str_t *vsp)
+{
+	mblk_t *mp;
+	t_uscalar_t prim, cprim;
+
+	VERIFY(MUTEX_HELD(&vsp->vns_lock));
+	mp = vnd_dlpi_inc_pop(vsp);
+	prim = ((dl_error_ack_t *)mp->b_rptr)->dl_primitive;
+	cprim = ((dl_ok_ack_t *)mp->b_rptr)->dl_correct_primitive;
+
+	if (prim != DL_OK_ACK && prim != DL_ERROR_ACK) {
+		vnd_drop_ctl(vsp, mp,
+		    "wrong dlpi primitive for vnd_st_exclusive");
+		vsp->vns_errno = VND_E_DLPIINVAL;
+		return (1);
+	}
+
+	if (cprim != DL_EXCLUSIVE_REQ) {
+		vnd_drop_ctl(vsp, mp,
+		    "vnd_st_exclusive: got ack/nack for wrong primitive");
+		vsp->vns_errno = VND_E_DLPIINVAL;
+		return (1);
+	}
+
+	if (prim == DL_ERROR_ACK)
+		vsp->vns_errno = VND_E_DLEXCL;
+
+	freemsg(mp);
+	return (prim == DL_ERROR_ACK);
+}
+
+/*
+ * Send down a DLPI_ATTACH_REQ.
+ */
+static int
+vnd_st_sattach(vnd_str_t *vsp)
+{
+	mblk_t *mp;
+
+	VERIFY(MUTEX_HELD(&vsp->vns_lock));
+	mp = vnd_dlpi_alloc(sizeof (dl_attach_req_t), DL_ATTACH_REQ);
+	if (mp == NULL) {
+		vsp->vns_errno = VND_E_NOMEM;
+		return (1);
+	}
+
+	((dl_attach_req_t *)mp->b_rptr)->dl_ppa = 0;
+	vsp->vns_state = VNS_S_ATTACH_SENT;
+	cv_broadcast(&vsp->vns_stcv);
+	putnext(vsp->vns_wq, mp);
+
+	return (0);
+}
+
+static int
+vnd_st_attach(vnd_str_t *vsp)
+{
+	mblk_t *mp;
+	t_uscalar_t prim, cprim;
+
+	VERIFY(MUTEX_HELD(&vsp->vns_lock));
+	mp = vnd_dlpi_inc_pop(vsp);
+	prim = ((dl_ok_ack_t *)mp->b_rptr)->dl_primitive;
+	cprim = ((dl_ok_ack_t *)mp->b_rptr)->dl_correct_primitive;
+
+
+	if (prim != DL_OK_ACK && prim != DL_ERROR_ACK) {
+		vnd_drop_ctl(vsp, mp, "vnd_st_attach: unknown primitive type");
+		vsp->vns_errno = VND_E_DLPIINVAL;
+		return (1);
+	}
+
+	if (cprim != DL_ATTACH_REQ) {
+		vnd_drop_ctl(vsp, mp,
+		    "vnd_st_attach: Got ack/nack for wrong primitive");
+		vsp->vns_errno = VND_E_DLPIINVAL;
+		return (1);
+	}
+
+	if (prim == DL_ERROR_ACK)
+		vsp->vns_errno = VND_E_ATTACHFAIL;
+
+	freemsg(mp);
+	return (prim == DL_ERROR_ACK);
+}
+
+static int
+vnd_st_sbind(vnd_str_t *vsp)
+{
+	mblk_t *mp;
+	dl_bind_req_t *dbrp;
+
+	VERIFY(MUTEX_HELD(&vsp->vns_lock));
+	mp = vnd_dlpi_alloc(sizeof (dl_bind_req_t) + sizeof (long),
+	    DL_BIND_REQ);
+	if (mp == NULL) {
+		vsp->vns_errno = VND_E_NOMEM;
+		return (1);
+	}
+	dbrp = (dl_bind_req_t *)(mp->b_rptr);
+	dbrp->dl_sap = 0;
+	dbrp->dl_service_mode = DL_CLDLS;
+
+	vsp->vns_state = VNS_S_BIND_SENT;
+	cv_broadcast(&vsp->vns_stcv);
+	putnext(vsp->vns_wq, mp);
+
+	return (0);
+}
+
+static int
+vnd_st_bind(vnd_str_t *vsp)
+{
+	mblk_t *mp;
+	t_uscalar_t prim;
+
+	VERIFY(MUTEX_HELD(&vsp->vns_lock));
+	mp = vnd_dlpi_inc_pop(vsp);
+	prim = ((dl_error_ack_t *)mp->b_rptr)->dl_primitive;
+
+	if (prim != DL_BIND_ACK && prim != DL_ERROR_ACK) {
+		vnd_drop_ctl(vsp, mp, "wrong dlpi primitive for vnd_st_bind");
+		vsp->vns_errno = VND_E_DLPIINVAL;
+		return (1);
+	}
+
+	if (prim == DL_ERROR_ACK)
+		vsp->vns_errno = VND_E_BINDFAIL;
+
+	freemsg(mp);
+	return (prim == DL_ERROR_ACK);
+}
+
+static int
+vnd_st_spromisc(vnd_str_t *vsp, int type, vnd_str_state_t next)
+{
+	mblk_t *mp;
+	dl_promiscon_req_t *dprp;
+
+	VERIFY(MUTEX_HELD(&vsp->vns_lock));
+	mp = vnd_dlpi_alloc(sizeof (dl_promiscon_req_t), DL_PROMISCON_REQ);
+	if (mp == NULL) {
+		vsp->vns_errno = VND_E_NOMEM;
+		return (1);
+	}
+
+	dprp = (dl_promiscon_req_t *)mp->b_rptr;
+	dprp->dl_level = type;
+
+	vsp->vns_state = next;
+	cv_broadcast(&vsp->vns_stcv);
+	putnext(vsp->vns_wq, mp);
+
+	return (0);
+}
+
+static int
+vnd_st_promisc(vnd_str_t *vsp)
+{
+	mblk_t *mp;
+	t_uscalar_t prim, cprim;
+
+	VERIFY(MUTEX_HELD(&vsp->vns_lock));
+	mp = vnd_dlpi_inc_pop(vsp);
+	prim = ((dl_ok_ack_t *)mp->b_rptr)->dl_primitive;
+	cprim = ((dl_ok_ack_t *)mp->b_rptr)->dl_correct_primitive;
+
+	if (prim != DL_OK_ACK && prim != DL_ERROR_ACK) {
+		vnd_drop_ctl(vsp, mp,
+		    "wrong dlpi primitive for vnd_st_promisc");
+		vsp->vns_errno = VND_E_DLPIINVAL;
+		return (1);
+	}
+
+	if (cprim != DL_PROMISCON_REQ) {
+		vnd_drop_ctl(vsp, mp,
+		    "vnd_st_promisc: Got ack/nack for wrong primitive");
+		vsp->vns_errno = VND_E_DLPIINVAL;
+		return (1);
+	}
+
+	if (prim == DL_ERROR_ACK)
+		vsp->vns_errno = VND_E_PROMISCFAIL;
+
+	freemsg(mp);
+	return (prim == DL_ERROR_ACK);
+}
+
+static int
+vnd_st_scapabq(vnd_str_t *vsp)
+{
+	mblk_t *mp;
+
+	VERIFY(MUTEX_HELD(&vsp->vns_lock));
+
+	mp = vnd_dlpi_alloc(sizeof (dl_capability_req_t), DL_CAPABILITY_REQ);
+	if (mp == NULL) {
+		vsp->vns_errno = VND_E_NOMEM;
+		return (1);
+	}
+
+	vsp->vns_state = VNS_S_CAPAB_Q_SENT;
+	cv_broadcast(&vsp->vns_stcv);
+	putnext(vsp->vns_wq, mp);
+
+	return (0);
+}
+
+static void
+vnd_mac_input(vnd_str_t *vsp, mac_resource_t *unused, mblk_t *mp_chain,
+    mac_header_info_t *mhip)
+{
+	int signal = 0;
+	mblk_t *mp;
+	vnd_pnsd_t *nsp = vsp->vns_nsd;
+
+	ASSERT(vsp != NULL);
+	ASSERT(mp_chain != NULL);
+
+	for (mp = mp_chain; mp != NULL; mp = mp_chain) {
+		uint16_t vid;
+		mp_chain = mp->b_next;
+		mp->b_next = NULL;
+
+		/*
+		 * If we were operating in a traditional dlpi context then we
+		 * would have enabled DLIOCRAW and rather than the fast path we
+		 * would come through dld_str_rx_raw. That function does two
+		 * things that we have to consider doing ourselves. The first is
+		 * that it adjusts the b_rptr back to account for dld bumping us
+		 * past the mac header. It also tries to account for cases where
+		 * mac provides an illusion of the mac header. Fortunately, dld
+		 * only allows the fastpath when the media type is the same as
+		 * the native type. Therefore all we have to do here is adjust
+		 * the b_rptr.
+		 */
+		ASSERT(mp->b_rptr >= DB_BASE(mp) + mhip->mhi_hdrsize);
+		mp->b_rptr -= mhip->mhi_hdrsize;
+		vid = VLAN_ID(mhip->mhi_tci);
+		if (mhip->mhi_istagged && vid != VLAN_ID_NONE) {
+			bcopy(mp->b_rptr, mp->b_rptr + 4, 12);
+			mp->b_rptr += 4;
+		}
+
+		if (nsp->vpnd_hooked && vnd_hook(vsp, &mp, nsp->vpnd_neti_v4,
+		    nsp->vpnd_event_in_v4, nsp->vpnd_token_in_v4,
+		    nsp->vpnd_neti_v6, nsp->vpnd_event_in_v6,
+		    nsp->vpnd_token_in_v6, vnd_drop_hook_in, vnd_drop_in) != 0)
+			continue;
+
+		VND_STAT_INC(vsp, vks_rpackets, 1);
+		VND_STAT_INC(vsp, vks_rbytes, msgsize(mp));
+		DTRACE_VND5(recv, mblk_t *, mp, void *, NULL, void *, NULL,
+		    vnd_str_t *, vsp, mblk_t *, mp);
+		mutex_enter(&vsp->vns_dq_read.vdq_lock);
+		signal |= vnd_dq_push(&vsp->vns_dq_read, mp, B_FALSE,
+		    vnd_drop_in);
+		mutex_exit(&vsp->vns_dq_read.vdq_lock);
+
+	}
+
+	if (signal != 0) {
+		cv_broadcast(&vsp->vns_dq_read.vdq_ready);
+		pollwakeup(&vsp->vns_dev->vdd_ph, POLLIN | POLLRDNORM);
+	}
+
+}
+
+static void
+vnd_mac_flow_control_stat(vnd_str_t *vsp, hrtime_t diff)
+{
+	VND_STAT_INC(vsp, vks_nmacflow, 1);
+	VND_STAT_INC(vsp, vks_tmacflow, diff);
+	if (diff >= VND_LATENCY_1MS)
+		VND_STAT_INC(vsp, vks_mac_flow_1ms, 1);
+	if (diff >= VND_LATENCY_10MS)
+		VND_STAT_INC(vsp, vks_mac_flow_10ms, 1);
+	if (diff >= VND_LATENCY_100MS)
+		VND_STAT_INC(vsp, vks_mac_flow_100ms, 1);
+	if (diff >= VND_LATENCY_1S)
+		VND_STAT_INC(vsp, vks_mac_flow_1s, 1);
+	if (diff >= VND_LATENCY_10S)
+		VND_STAT_INC(vsp, vks_mac_flow_10s, 1);
+}
+
+/*
+ * This is a callback from MAC that indicates that we are allowed to send
+ * packets again.
+ */
+static void
+vnd_mac_flow_control(void *arg, vnd_mac_cookie_t cookie)
+{
+	vnd_str_t *vsp = arg;
+	hrtime_t now, diff;
+
+	mutex_enter(&vsp->vns_lock);
+	now = gethrtime();
+
+	/*
+	 * Check for the case that we beat vnd_squeue_tx_one to the punch.
+	 * There's also an additional case here that we got notified because
+	 * we're sharing a device that ran out of tx descriptors, even though it
+	 * wasn't because of us.
+	 */
+	if (!(vsp->vns_flags & VNS_F_FLOW_CONTROLLED)) {
+		vsp->vns_fcupdate = now;
+		mutex_exit(&vsp->vns_lock);
+		return;
+	}
+
+	ASSERT(vsp->vns_flags & VNS_F_FLOW_CONTROLLED);
+	ASSERT(vsp->vns_caps.vsc_fc_cookie == cookie);
+	vsp->vns_flags &= ~VNS_F_FLOW_CONTROLLED;
+	vsp->vns_caps.vsc_fc_cookie = NULL;
+	diff = now - vsp->vns_fclatch;
+	vsp->vns_fclatch = 0;
+	DTRACE_VND3(flow__resumed, vnd_str_t *, vsp, uint64_t,
+	    vsp->vns_dq_write.vdq_cur, uintptr_t, cookie);
+	/*
+	 * If someone has asked to flush the squeue and thus inserted a barrier,
+	 * than we shouldn't schedule a drain.
+	 */
+	if (!(vsp->vns_flags & (VNS_F_DRAIN_SCHEDULED | VNS_F_BARRIER))) {
+		vsp->vns_flags |= VNS_F_DRAIN_SCHEDULED;
+		gsqueue_enter_one(vsp->vns_squeue, &vsp->vns_drainblk,
+		    vnd_squeue_tx_drain, vsp, GSQUEUE_FILL,
+		    VND_SQUEUE_TAG_MAC_FLOW_CONTROL);
+	}
+	mutex_exit(&vsp->vns_lock);
+}
+
+static void
+vnd_mac_enter(vnd_str_t *vsp, mac_perim_handle_t *mphp)
+{
+	ASSERT(MUTEX_HELD(&vsp->vns_lock));
+	VERIFY(vsp->vns_caps.vsc_capab_f(vsp->vns_caps.vsc_capab_hdl,
+	    DLD_CAPAB_PERIM, mphp, DLD_ENABLE) == 0);
+}
+
+static void
+vnd_mac_exit(vnd_str_t *vsp, mac_perim_handle_t mph)
+{
+	ASSERT(MUTEX_HELD(&vsp->vns_lock));
+	VERIFY(vsp->vns_caps.vsc_capab_f(vsp->vns_caps.vsc_capab_hdl,
+	    DLD_CAPAB_PERIM, mph, DLD_DISABLE) == 0);
+}
+
+static int
+vnd_dld_cap_enable(vnd_str_t *vsp, vnd_rx_t rxfunc)
+{
+	int ret;
+	dld_capab_direct_t d;
+	mac_perim_handle_t mph;
+	vnd_str_capab_t *c = &vsp->vns_caps;
+
+	bzero(&d, sizeof (d));
+	d.di_rx_cf = (uintptr_t)rxfunc;
+	d.di_rx_ch = vsp;
+	d.di_flags = DI_DIRECT_RAW;
+
+	vnd_mac_enter(vsp, &mph);
+
+	/*
+	 * If we're coming in here for a second pass, we need to make sure that
+	 * we remove an existing flow control notification callback, otherwise
+	 * we'll create a duplicate that will remain with garbage data.
+	 */
+	if (c->vsc_tx_fc_hdl != NULL) {
+		ASSERT(c->vsc_set_fcb_hdl != NULL);
+		(void) c->vsc_set_fcb_f(c->vsc_set_fcb_hdl, NULL,
+		    c->vsc_tx_fc_hdl);
+		c->vsc_tx_fc_hdl = NULL;
+	}
+
+	if (vsp->vns_caps.vsc_capab_f(c->vsc_capab_hdl,
+	    DLD_CAPAB_DIRECT, &d, DLD_ENABLE) == 0) {
+		c->vsc_tx_f = (vnd_dld_tx_t)d.di_tx_df;
+		c->vsc_tx_hdl = d.di_tx_dh;
+		c->vsc_set_fcb_f = (vnd_dld_set_fcb_t)d.di_tx_cb_df;
+		c->vsc_set_fcb_hdl = d.di_tx_cb_dh;
+		c->vsc_is_fc_f = (vnd_dld_is_fc_t)d.di_tx_fctl_df;
+		c->vsc_is_fc_hdl = d.di_tx_fctl_dh;
+		c->vsc_tx_fc_hdl = c->vsc_set_fcb_f(c->vsc_set_fcb_hdl,
+		    vnd_mac_flow_control, vsp);
+		c->vsc_flags |= VNS_C_DIRECT;
+		ret = 0;
+	} else {
+		vsp->vns_errno = VND_E_DIRECTFAIL;
+		ret = 1;
+	}
+	vnd_mac_exit(vsp, mph);
+	return (ret);
+}
+
+static int
+vnd_st_capabq(vnd_str_t *vsp)
+{
+	mblk_t *mp;
+	dl_capability_ack_t *cap;
+	dl_capability_sub_t *subp;
+	dl_capab_hcksum_t *hck;
+	dl_capab_dld_t *dld;
+	unsigned char *rp;
+	int ret = 0;
+
+	VERIFY(MUTEX_HELD(&vsp->vns_lock));
+	mp = vnd_dlpi_inc_pop(vsp);
+
+	rp = mp->b_rptr;
+	cap = (dl_capability_ack_t *)rp;
+	if (cap->dl_sub_length == 0)
+		goto done;
+
+	/* Don't try to process something too big */
+	if (sizeof (dl_capability_ack_t) + cap->dl_sub_length > MBLKL(mp)) {
+		VND_STAT_INC(vsp, vks_ndlpidrops, 1);
+		VND_STAT_INC(vsp, vks_tdrops, 1);
+		vsp->vns_errno = VND_E_CAPACKINVAL;
+		ret = 1;
+		goto done;
+	}
+
+	rp += cap->dl_sub_offset;
+
+	while (cap->dl_sub_length > 0) {
+		subp = (dl_capability_sub_t *)rp;
+		/* Sanity check something crazy from down below */
+		if (subp->dl_length + sizeof (dl_capability_sub_t) >
+		    cap->dl_sub_length) {
+			VND_STAT_INC(vsp, vks_ndlpidrops, 1);
+			VND_STAT_INC(vsp, vks_tdrops, 1);
+			vsp->vns_errno = VND_E_SUBCAPINVAL;
+			ret = 1;
+			goto done;
+		}
+
+		switch (subp->dl_cap) {
+		case DL_CAPAB_HCKSUM:
+			hck = (dl_capab_hcksum_t *)(rp +
+			    sizeof (dl_capability_sub_t));
+			if (hck->hcksum_version != HCKSUM_CURRENT_VERSION) {
+				vsp->vns_caps.vsc_flags |= VNS_C_HCKSUM_BADVERS;
+				break;
+			}
+			if (dlcapabcheckqid(&hck->hcksum_mid, vsp->vns_lrq) !=
+			    B_TRUE) {
+				vsp->vns_errno = VND_E_CAPABPASS;
+				ret = 1;
+				goto done;
+			}
+			vsp->vns_caps.vsc_flags |= VNS_C_HCKSUM;
+			vsp->vns_caps.vsc_hcksum_opts = hck->hcksum_txflags;
+			break;
+		case DL_CAPAB_DLD:
+			dld = (dl_capab_dld_t *)(rp +
+			    sizeof (dl_capability_sub_t));
+			if (dld->dld_version != DLD_CURRENT_VERSION) {
+				vsp->vns_errno = VND_E_DLDBADVERS;
+				ret = 1;
+				goto done;
+			}
+			if (dlcapabcheckqid(&dld->dld_mid, vsp->vns_lrq) !=
+			    B_TRUE) {
+				vsp->vns_errno = VND_E_CAPABPASS;
+				ret = 1;
+				goto done;
+			}
+			vsp->vns_caps.vsc_flags |= VNS_C_DLD;
+			vsp->vns_caps.vsc_capab_f =
+			    (vnd_dld_cap_t)dld->dld_capab;
+			vsp->vns_caps.vsc_capab_hdl =
+			    (void *)dld->dld_capab_handle;
+			/*
+			 * At this point in time, we have to set up a direct
+			 * function that drops all input. This validates that
+			 * we'll be able to set up direct input and that we can
+			 * easily switch it earlier to the real data function
+			 * when we've plumbed everything up.
+			 */
+			if (vnd_dld_cap_enable(vsp, vnd_mac_drop_input) != 0) {
+				/* vns_errno set by vnd_dld_cap_enable */
+				ret = 1;
+				goto done;
+			}
+			break;
+		default:
+			/* Ignore unsupported cap */
+			break;
+		}
+
+		rp += sizeof (dl_capability_sub_t) + subp->dl_length;
+		cap->dl_sub_length -= sizeof (dl_capability_sub_t) +
+		    subp->dl_length;
+	}
+
+done:
+	/* Make sure we enabled direct callbacks */
+	if (ret == 0 && !(vsp->vns_caps.vsc_flags & VNS_C_DIRECT)) {
+		vsp->vns_errno = VND_E_DIRECTNOTSUP;
+		ret = 1;
+	}
+
+	freemsg(mp);
+	return (ret);
+}
+
+static void
+vnd_st_sonline(vnd_str_t *vsp)
+{
+	VERIFY(MUTEX_HELD(&vsp->vns_lock));
+	vsp->vns_state = VNS_S_ONLINE;
+	cv_broadcast(&vsp->vns_stcv);
+}
+
+static void
+vnd_st_shutdown(vnd_str_t *vsp)
+{
+	mac_perim_handle_t mph;
+	vnd_str_capab_t *vsc = &vsp->vns_caps;
+
+	VERIFY(MUTEX_HELD(&vsp->vns_lock));
+
+	/*
+	 * At this point in time we know that there is no one transmitting as
+	 * our final reference has been torn down and that vnd_s_close inserted
+	 * a barrier to validate that everything is flushed.
+	 */
+	if (vsc->vsc_flags & VNS_C_DIRECT) {
+		vnd_mac_enter(vsp, &mph);
+		vsc->vsc_flags &= ~VNS_C_DIRECT;
+		(void) vsc->vsc_set_fcb_f(vsc->vsc_set_fcb_hdl, NULL,
+		    vsc->vsc_tx_fc_hdl);
+		vsc->vsc_tx_fc_hdl = NULL;
+		(void) vsc->vsc_capab_f(vsc->vsc_capab_hdl, DLD_CAPAB_DIRECT,
+		    NULL, DLD_DISABLE);
+		vnd_mac_exit(vsp, mph);
+	}
+
+	/*
+	 * We could send an unbind, but dld also does that for us. As we add
+	 * more capabilities and the like, we should revisit this.
+	 */
+	vsp->vns_state = VNS_S_ZOMBIE;
+	cv_broadcast(&vsp->vns_stcv);
+}
+
+/*
+ * Perform state transitions. This is a one way shot down the flow chart
+ * described in the big theory statement.
+ */
+static void
+vnd_str_state_transition(void *arg)
+{
+	boolean_t died = B_FALSE;
+	vnd_str_t *vsp = arg;
+	mblk_t *mp;
+
+	mutex_enter(&vsp->vns_lock);
+	if (vsp->vns_dlpi_inc == NULL && (vsp->vns_state != VNS_S_INITIAL &&
+	    vsp->vns_state != VNS_S_SHUTTING_DOWN)) {
+		mutex_exit(&vsp->vns_lock);
+		return;
+	}
+	DTRACE_PROBE2(vnd__state__transition, uintptr_t, vsp,
+	    vnd_str_state_t, vsp->vns_state);
+	switch (vsp->vns_state) {
+	case VNS_S_INITIAL:
+		VERIFY(vsp->vns_dlpi_inc == NULL);
+		if (vnd_st_sinfo(vsp) != 0)
+			died = B_TRUE;
+		break;
+	case VNS_S_INFO_SENT:
+		VERIFY(vsp->vns_dlpi_inc != NULL);
+		if (vnd_st_info(vsp) == 0) {
+			if (vnd_st_sexclusive(vsp) != 0)
+				died = B_TRUE;
+		} else {
+			died = B_TRUE;
+		}
+		break;
+	case VNS_S_EXCLUSIVE_SENT:
+		VERIFY(vsp->vns_dlpi_inc != NULL);
+		if (vnd_st_exclusive(vsp) == 0) {
+			if (vsp->vns_dlpi_style == DL_STYLE2) {
+				if (vnd_st_sattach(vsp) != 0)
+					died = B_TRUE;
+			} else {
+				if (vnd_st_sbind(vsp) != 0)
+					died = B_TRUE;
+			}
+		} else  {
+			died = B_TRUE;
+		}
+		break;
+	case VNS_S_ATTACH_SENT:
+		VERIFY(vsp->vns_dlpi_inc != NULL);
+		if (vnd_st_attach(vsp) == 0) {
+			if (vnd_st_sbind(vsp) != 0)
+				died = B_TRUE;
+		} else {
+			died = B_TRUE;
+		}
+		break;
+	case VNS_S_BIND_SENT:
+		VERIFY(vsp->vns_dlpi_inc != NULL);
+		if (vnd_st_bind(vsp) == 0) {
+			if (vnd_st_spromisc(vsp, DL_PROMISC_SAP,
+			    VNS_S_SAP_PROMISC_SENT) != 0)
+				died = B_TRUE;
+		} else {
+			died = B_TRUE;
+		}
+		break;
+	case VNS_S_SAP_PROMISC_SENT:
+		VERIFY(vsp->vns_dlpi_inc != NULL);
+		if (vnd_st_promisc(vsp) == 0) {
+			if (vnd_st_spromisc(vsp, DL_PROMISC_MULTI,
+			    VNS_S_MULTI_PROMISC_SENT) != 0)
+				died = B_TRUE;
+		} else {
+			died = B_TRUE;
+		}
+		break;
+	case VNS_S_MULTI_PROMISC_SENT:
+		VERIFY(vsp->vns_dlpi_inc != NULL);
+		if (vnd_st_promisc(vsp) == 0) {
+			if (vnd_st_spromisc(vsp, DL_PROMISC_RX_ONLY,
+			    VNS_S_RX_ONLY_PROMISC_SENT) != 0)
+				died = B_TRUE;
+		} else {
+			died = B_TRUE;
+		}
+		break;
+	case VNS_S_RX_ONLY_PROMISC_SENT:
+		VERIFY(vsp->vns_dlpi_inc != NULL);
+		if (vnd_st_promisc(vsp) == 0) {
+			if (vnd_st_scapabq(vsp) != 0)
+				died = B_TRUE;
+		} else {
+			died = B_TRUE;
+		}
+		break;
+	case VNS_S_CAPAB_Q_SENT:
+		if (vnd_st_capabq(vsp) != 0)
+			died = B_TRUE;
+		else
+			vnd_st_sonline(vsp);
+		break;
+	case VNS_S_SHUTTING_DOWN:
+		vnd_st_shutdown(vsp);
+		break;
+	case VNS_S_ZOMBIE:
+		while ((mp = vnd_dlpi_inc_pop(vsp)) != NULL)
+			vnd_drop_ctl(vsp, mp, "vsp committed suicide");
+		break;
+	default:
+		panic("vnd_str_t entered an unknown state");
+	}
+
+	if (died == B_TRUE) {
+		ASSERT(vsp->vns_errno != VND_E_SUCCESS);
+		vsp->vns_laststate = vsp->vns_state;
+		vsp->vns_state = VNS_S_ZOMBIE;
+		cv_broadcast(&vsp->vns_stcv);
+	}
+
+	mutex_exit(&vsp->vns_lock);
+}
+
+static void
+vnd_dlpi_taskq_dispatch(void *arg)
+{
+	vnd_str_t *vsp = arg;
+	int run = 1;
+
+	while (run != 0) {
+		vnd_str_state_transition(vsp);
+		mutex_enter(&vsp->vns_lock);
+		if (vsp->vns_flags & VNS_F_CONDEMNED ||
+		    vsp->vns_dlpi_inc == NULL) {
+			run = 0;
+			vsp->vns_flags &= ~VNS_F_TASKQ_DISPATCHED;
+		}
+		if (vsp->vns_flags & VNS_F_CONDEMNED)
+			cv_signal(&vsp->vns_cancelcv);
+		mutex_exit(&vsp->vns_lock);
+	}
+}
+
+static int
+vnd_neti_getifname(net_handle_t neti, phy_if_t phy, char *buf, const size_t len)
+{
+	return (-1);
+}
+
+static int
+vnd_neti_getmtu(net_handle_t neti, phy_if_t phy, lif_if_t ifdata)
+{
+	return (-1);
+}
+
+static int
+vnd_neti_getptmue(net_handle_t neti)
+{
+	return (-1);
+}
+
+static int
+vnd_neti_getlifaddr(net_handle_t neti, phy_if_t phy, lif_if_t ifdata,
+    size_t nelem, net_ifaddr_t type[], void *storage)
+{
+	return (-1);
+}
+
+static int
+vnd_neti_getlifzone(net_handle_t neti, phy_if_t phy, lif_if_t ifdata,
+    zoneid_t *zid)
+{
+	return (-1);
+}
+
+static int
+vnd_neti_getlifflags(net_handle_t neti, phy_if_t phy, lif_if_t ifdata,
+    uint64_t *flags)
+{
+	return (-1);
+}
+
+static phy_if_t
+vnd_neti_phygetnext(net_handle_t neti, phy_if_t phy)
+{
+	return (-1);
+}
+
+static phy_if_t
+vnd_neti_phylookup(net_handle_t neti, const char *name)
+{
+	return (-1);
+}
+
+static lif_if_t
+vnd_neti_lifgetnext(net_handle_t neti, phy_if_t phy, lif_if_t ifdata)
+{
+	return (-1);
+}
+
+static int
+vnd_neti_inject(net_handle_t neti, inject_t style, net_inject_t *packet)
+{
+	return (-1);
+}
+
+static phy_if_t
+vnd_neti_route(net_handle_t neti, struct sockaddr *address,
+    struct sockaddr *next)
+{
+	return ((phy_if_t)-1);
+}
+
+static int
+vnd_neti_ispchksum(net_handle_t neti, mblk_t *mp)
+{
+	return (-1);
+}
+
+static int
+vnd_neti_isvchksum(net_handle_t neti, mblk_t *mp)
+{
+	return (-1);
+}
+
+static net_protocol_t vnd_neti_info_v4 = {
+	NETINFO_VERSION,
+	NHF_VND_INET,
+	vnd_neti_getifname,
+	vnd_neti_getmtu,
+	vnd_neti_getptmue,
+	vnd_neti_getlifaddr,
+	vnd_neti_getlifzone,
+	vnd_neti_getlifflags,
+	vnd_neti_phygetnext,
+	vnd_neti_phylookup,
+	vnd_neti_lifgetnext,
+	vnd_neti_inject,
+	vnd_neti_route,
+	vnd_neti_ispchksum,
+	vnd_neti_isvchksum
+};
+
+static net_protocol_t vnd_neti_info_v6 = {
+	NETINFO_VERSION,
+	NHF_VND_INET6,
+	vnd_neti_getifname,
+	vnd_neti_getmtu,
+	vnd_neti_getptmue,
+	vnd_neti_getlifaddr,
+	vnd_neti_getlifzone,
+	vnd_neti_getlifflags,
+	vnd_neti_phygetnext,
+	vnd_neti_phylookup,
+	vnd_neti_lifgetnext,
+	vnd_neti_inject,
+	vnd_neti_route,
+	vnd_neti_ispchksum,
+	vnd_neti_isvchksum
+};
+
+
+static int
+vnd_netinfo_init(vnd_pnsd_t *nsp)
+{
+	nsp->vpnd_neti_v4 = net_protocol_register(nsp->vpnd_nsid,
+	    &vnd_neti_info_v4);
+	ASSERT(nsp->vpnd_neti_v4 != NULL);
+
+	nsp->vpnd_neti_v6 = net_protocol_register(nsp->vpnd_nsid,
+	    &vnd_neti_info_v6);
+	ASSERT(nsp->vpnd_neti_v6 != NULL);
+
+	nsp->vpnd_family_v4.hf_version = HOOK_VERSION;
+	nsp->vpnd_family_v4.hf_name = "vnd_inet";
+
+	if (net_family_register(nsp->vpnd_neti_v4, &nsp->vpnd_family_v4) != 0) {
+		net_protocol_unregister(nsp->vpnd_neti_v4);
+		net_protocol_unregister(nsp->vpnd_neti_v6);
+		cmn_err(CE_NOTE, "vnd_netinfo_init: net_family_register "
+		    "failed for stack %d", nsp->vpnd_nsid);
+		return (1);
+	}
+
+	nsp->vpnd_family_v6.hf_version = HOOK_VERSION;
+	nsp->vpnd_family_v6.hf_name = "vnd_inet6";
+
+	if (net_family_register(nsp->vpnd_neti_v6, &nsp->vpnd_family_v6) != 0) {
+		net_family_unregister(nsp->vpnd_neti_v4, &nsp->vpnd_family_v4);
+		net_protocol_unregister(nsp->vpnd_neti_v4);
+		net_protocol_unregister(nsp->vpnd_neti_v6);
+		cmn_err(CE_NOTE, "vnd_netinfo_init: net_family_register "
+		    "failed for stack %d", nsp->vpnd_nsid);
+		return (1);
+	}
+
+	nsp->vpnd_event_in_v4.he_version = HOOK_VERSION;
+	nsp->vpnd_event_in_v4.he_name = NH_PHYSICAL_IN;
+	nsp->vpnd_event_in_v4.he_flags = 0;
+	nsp->vpnd_event_in_v4.he_interested = B_FALSE;
+
+	nsp->vpnd_token_in_v4 = net_event_register(nsp->vpnd_neti_v4,
+	    &nsp->vpnd_event_in_v4);
+	if (nsp->vpnd_token_in_v4 == NULL) {
+		net_family_unregister(nsp->vpnd_neti_v4, &nsp->vpnd_family_v4);
+		net_family_unregister(nsp->vpnd_neti_v6, &nsp->vpnd_family_v6);
+		net_protocol_unregister(nsp->vpnd_neti_v4);
+		net_protocol_unregister(nsp->vpnd_neti_v6);
+		cmn_err(CE_NOTE, "vnd_netinfo_init: net_event_register "
+		    "failed for stack %d", nsp->vpnd_nsid);
+		return (1);
+	}
+
+	nsp->vpnd_event_in_v6.he_version = HOOK_VERSION;
+	nsp->vpnd_event_in_v6.he_name = NH_PHYSICAL_IN;
+	nsp->vpnd_event_in_v6.he_flags = 0;
+	nsp->vpnd_event_in_v6.he_interested = B_FALSE;
+
+	nsp->vpnd_token_in_v6 = net_event_register(nsp->vpnd_neti_v6,
+	    &nsp->vpnd_event_in_v6);
+	if (nsp->vpnd_token_in_v6 == NULL) {
+		net_event_shutdown(nsp->vpnd_neti_v4, &nsp->vpnd_event_in_v4);
+		net_event_unregister(nsp->vpnd_neti_v4, &nsp->vpnd_event_in_v4);
+		net_family_unregister(nsp->vpnd_neti_v4, &nsp->vpnd_family_v4);
+		net_family_unregister(nsp->vpnd_neti_v6, &nsp->vpnd_family_v6);
+		net_protocol_unregister(nsp->vpnd_neti_v4);
+		net_protocol_unregister(nsp->vpnd_neti_v6);
+		cmn_err(CE_NOTE, "vnd_netinfo_init: net_event_register "
+		    "failed for stack %d", nsp->vpnd_nsid);
+		return (1);
+	}
+
+	nsp->vpnd_event_out_v4.he_version = HOOK_VERSION;
+	nsp->vpnd_event_out_v4.he_name = NH_PHYSICAL_OUT;
+	nsp->vpnd_event_out_v4.he_flags = 0;
+	nsp->vpnd_event_out_v4.he_interested = B_FALSE;
+
+	nsp->vpnd_token_out_v4 = net_event_register(nsp->vpnd_neti_v4,
+	    &nsp->vpnd_event_out_v4);
+	if (nsp->vpnd_token_out_v4 == NULL) {
+		net_event_shutdown(nsp->vpnd_neti_v6, &nsp->vpnd_event_in_v6);
+		net_event_unregister(nsp->vpnd_neti_v6, &nsp->vpnd_event_in_v6);
+		net_event_shutdown(nsp->vpnd_neti_v4, &nsp->vpnd_event_in_v4);
+		net_event_unregister(nsp->vpnd_neti_v4, &nsp->vpnd_event_in_v4);
+		net_family_unregister(nsp->vpnd_neti_v4, &nsp->vpnd_family_v4);
+		net_family_unregister(nsp->vpnd_neti_v6, &nsp->vpnd_family_v6);
+		net_protocol_unregister(nsp->vpnd_neti_v4);
+		net_protocol_unregister(nsp->vpnd_neti_v6);
+		cmn_err(CE_NOTE, "vnd_netinfo_init: net_event_register "
+		    "failed for stack %d", nsp->vpnd_nsid);
+		return (1);
+	}
+
+	nsp->vpnd_event_out_v6.he_version = HOOK_VERSION;
+	nsp->vpnd_event_out_v6.he_name = NH_PHYSICAL_OUT;
+	nsp->vpnd_event_out_v6.he_flags = 0;
+	nsp->vpnd_event_out_v6.he_interested = B_FALSE;
+
+	nsp->vpnd_token_out_v6 = net_event_register(nsp->vpnd_neti_v6,
+	    &nsp->vpnd_event_out_v6);
+	if (nsp->vpnd_token_out_v6 == NULL) {
+		net_event_shutdown(nsp->vpnd_neti_v6, &nsp->vpnd_event_in_v6);
+		net_event_unregister(nsp->vpnd_neti_v6, &nsp->vpnd_event_in_v6);
+		net_event_shutdown(nsp->vpnd_neti_v6, &nsp->vpnd_event_in_v6);
+		net_event_unregister(nsp->vpnd_neti_v6, &nsp->vpnd_event_in_v6);
+		net_event_shutdown(nsp->vpnd_neti_v4, &nsp->vpnd_event_in_v4);
+		net_event_unregister(nsp->vpnd_neti_v4, &nsp->vpnd_event_in_v4);
+		net_family_unregister(nsp->vpnd_neti_v4, &nsp->vpnd_family_v4);
+		net_family_unregister(nsp->vpnd_neti_v6, &nsp->vpnd_family_v6);
+		net_protocol_unregister(nsp->vpnd_neti_v4);
+		net_protocol_unregister(nsp->vpnd_neti_v6);
+		cmn_err(CE_NOTE, "vnd_netinfo_init: net_event_register "
+		    "failed for stack %d", nsp->vpnd_nsid);
+		return (1);
+	}
+
+	return (0);
+}
+
+static void
+vnd_netinfo_shutdown(vnd_pnsd_t *nsp)
+{
+	int ret;
+
+	ret = net_event_shutdown(nsp->vpnd_neti_v4, &nsp->vpnd_event_in_v4);
+	VERIFY(ret == 0);
+	ret = net_event_shutdown(nsp->vpnd_neti_v4, &nsp->vpnd_event_out_v4);
+	VERIFY(ret == 0);
+	ret = net_event_shutdown(nsp->vpnd_neti_v6, &nsp->vpnd_event_in_v6);
+	VERIFY(ret == 0);
+	ret = net_event_shutdown(nsp->vpnd_neti_v6, &nsp->vpnd_event_out_v6);
+	VERIFY(ret == 0);
+}
+
+static void
+vnd_netinfo_fini(vnd_pnsd_t *nsp)
+{
+	int ret;
+
+	ret = net_event_unregister(nsp->vpnd_neti_v4, &nsp->vpnd_event_in_v4);
+	VERIFY(ret == 0);
+	ret = net_event_unregister(nsp->vpnd_neti_v4, &nsp->vpnd_event_out_v4);
+	VERIFY(ret == 0);
+	ret = net_event_unregister(nsp->vpnd_neti_v6, &nsp->vpnd_event_in_v6);
+	VERIFY(ret == 0);
+	ret = net_event_unregister(nsp->vpnd_neti_v6, &nsp->vpnd_event_out_v6);
+	VERIFY(ret == 0);
+	ret = net_family_unregister(nsp->vpnd_neti_v4, &nsp->vpnd_family_v4);
+	VERIFY(ret == 0);
+	ret = net_family_unregister(nsp->vpnd_neti_v6, &nsp->vpnd_family_v6);
+	VERIFY(ret == 0);
+	ret = net_protocol_unregister(nsp->vpnd_neti_v4);
+	VERIFY(ret == 0);
+	ret = net_protocol_unregister(nsp->vpnd_neti_v6);
+	VERIFY(ret == 0);
+}
+
+static void
+vnd_strbarrier_cb(void *arg, mblk_t *bmp, gsqueue_t *gsp, void *dummy)
+{
+	vnd_str_t *vsp = arg;
+
+	VERIFY(bmp == &vsp->vns_barrierblk);
+	mutex_enter(&vsp->vns_lock);
+	VERIFY(vsp->vns_flags & VNS_F_BARRIER);
+	VERIFY(!(vsp->vns_flags & VNS_F_BARRIER_DONE));
+	vsp->vns_flags |= VNS_F_BARRIER_DONE;
+	mutex_exit(&vsp->vns_lock);
+
+	/*
+	 * For better or worse, we have to broadcast here as we could have a
+	 * thread that's blocked for completion as well as one that's blocked
+	 * waiting to do a barrier itself.
+	 */
+	cv_broadcast(&vsp->vns_barriercv);
+}
+
+/*
+ * This is a data barrier for the stream while it is in fastpath mode. It blocks
+ * and ensures that there is nothing else in the squeue.
+ */
+static void
+vnd_strbarrier(vnd_str_t *vsp)
+{
+	mutex_enter(&vsp->vns_lock);
+	while (vsp->vns_flags & VNS_F_BARRIER)
+		cv_wait(&vsp->vns_barriercv, &vsp->vns_lock);
+	vsp->vns_flags |= VNS_F_BARRIER;
+	mutex_exit(&vsp->vns_lock);
+
+	gsqueue_enter_one(vsp->vns_squeue, &vsp->vns_barrierblk,
+	    vnd_strbarrier_cb, vsp, GSQUEUE_PROCESS, VND_SQUEUE_TAG_STRBARRIER);
+
+	mutex_enter(&vsp->vns_lock);
+	while (!(vsp->vns_flags & VNS_F_BARRIER_DONE))
+		cv_wait(&vsp->vns_barriercv, &vsp->vns_lock);
+	vsp->vns_flags &= ~VNS_F_BARRIER;
+	vsp->vns_flags &= ~VNS_F_BARRIER_DONE;
+	mutex_exit(&vsp->vns_lock);
+
+	/*
+	 * We have to broadcast in case anyone is waiting for the barrier
+	 * themselves.
+	 */
+	cv_broadcast(&vsp->vns_barriercv);
+}
+
+/*
+ * Based on the type of message that we're dealing with we're going to want to
+ * do one of several things. Basically if it looks like it's something we know
+ * about, we should probably handle it in one of our transition threads.
+ * Otherwise, we should just simply putnext.
+ */
+static int
+vnd_s_rput(queue_t *q, mblk_t *mp)
+{
+	t_uscalar_t prim;
+	int dispatch = 0;
+	vnd_str_t *vsp = q->q_ptr;
+
+	switch (DB_TYPE(mp)) {
+	case M_PROTO:
+	case M_PCPROTO:
+		if (MBLKL(mp) < sizeof (t_uscalar_t)) {
+			vnd_drop_ctl(vsp, mp, "PROTO message too short");
+			break;
+		}
+
+		prim = ((union DL_primitives *)mp->b_rptr)->dl_primitive;
+		if (prim == DL_UNITDATA_REQ || prim == DL_UNITDATA_IND) {
+			vnd_drop_ctl(vsp, mp,
+			    "recieved an unsupported dlpi DATA req");
+			break;
+		}
+
+		/*
+		 * Enqueue the entry and fire off a taskq dispatch.
+		 */
+		mutex_enter(&vsp->vns_lock);
+		vnd_dlpi_inc_push(vsp, mp);
+		if (!(vsp->vns_flags & VNS_F_TASKQ_DISPATCHED)) {
+			dispatch = 1;
+			vsp->vns_flags |= VNS_F_TASKQ_DISPATCHED;
+		}
+		mutex_exit(&vsp->vns_lock);
+		if (dispatch != 0)
+			taskq_dispatch_ent(vnd_taskq, vnd_dlpi_taskq_dispatch,
+			    vsp, 0, &vsp->vns_tqe);
+		break;
+	case M_DATA:
+		vnd_drop_in(vsp, mp, "M_DATA via put(9E)");
+		break;
+	default:
+		putnext(vsp->vns_rq, mp);
+	}
+	return (0);
+}
+
+static void
+vnd_strioctl(queue_t *q, vnd_str_t *vsp, mblk_t *mp, struct iocblk *iocp)
+{
+	int error;
+	vnd_strioc_t *visp;
+
+	if (iocp->ioc_cmd != VND_STRIOC_ASSOCIATE ||
+	    iocp->ioc_count != TRANSPARENT) {
+		error = EINVAL;
+		goto nak;
+	}
+
+	/*
+	 * All streams ioctls that we support must use kcred as a means to
+	 * distinguish that this is a layered open by the kernel as opposed to
+	 * one by a user who has done an I_PUSH of the module.
+	 */
+	if (iocp->ioc_cr != kcred) {
+		error = EPERM;
+		goto nak;
+	}
+
+	if (mp->b_cont == NULL) {
+		error = EAGAIN;
+		goto nak;
+	}
+
+	visp = kmem_alloc(sizeof (vnd_strioc_t), KM_SLEEP);
+	ASSERT(MBLKL(mp->b_cont) == sizeof (caddr_t));
+	visp->vs_addr = *(caddr_t *)mp->b_cont->b_rptr;
+	visp->vs_state = VSS_COPYIN;
+
+	mcopyin(mp, (void *)visp, sizeof (vnd_strioc_associate_t), NULL);
+	qreply(q, mp);
+
+	return;
+
+nak:
+	if (mp->b_cont != NULL) {
+		freemsg(mp->b_cont);
+		mp->b_cont = NULL;
+	}
+
+	iocp->ioc_error = error;
+	mp->b_datap->db_type = M_IOCNAK;
+	iocp->ioc_count = 0;
+	qreply(q, mp);
+}
+
+static void
+vnd_striocdata(queue_t *q, vnd_str_t *vsp, mblk_t *mp, struct copyresp *csp)
+{
+	int error;
+	vnd_str_state_t state;
+	struct copyreq *crp;
+	vnd_strioc_associate_t *vss;
+	vnd_dev_t *vdp = NULL;
+	vnd_pnsd_t *nsp = NULL;
+	char iname[2*VND_NAMELEN];
+	zone_t *zone;
+	vnd_strioc_t *visp;
+
+	visp = (vnd_strioc_t *)csp->cp_private;
+
+	/* If it's not ours, it's not our problem */
+	if (csp->cp_cmd != VND_STRIOC_ASSOCIATE) {
+		if (q->q_next != NULL) {
+			putnext(q, mp);
+		} else {
+			VND_STAT_INC(vsp, vks_ndlpidrops, 1);
+			VND_STAT_INC(vsp, vks_tdrops, 1);
+			vnd_drop_ctl(vsp, mp, "uknown cmd for M_IOCDATA");
+		}
+		kmem_free(visp, sizeof (vnd_strioc_t));
+		return;
+	}
+
+	/* The nak is already sent for us */
+	if (csp->cp_rval != 0) {
+		vnd_drop_ctl(vsp, mp, "M_COPYIN failed");
+		kmem_free(visp, sizeof (vnd_strioc_t));
+		return;
+	}
+
+	/* Data is sitting for us in b_cont */
+	if (mp->b_cont == NULL ||
+	    MBLKL(mp->b_cont) != sizeof (vnd_strioc_associate_t)) {
+		kmem_free(visp, sizeof (vnd_strioc_t));
+		miocnak(q, mp, 0, EINVAL);
+		qreply(q, mp);
+		return;
+	}
+
+	vss = (vnd_strioc_associate_t *)mp->b_cont->b_rptr;
+	vdp = vnd_dev_lookup(vss->vsa_minor);
+	if (vdp == NULL) {
+		error = EIO;
+		vss->vsa_errno = VND_E_NODEV;
+		goto nak;
+	}
+
+	nsp = vnd_nsd_lookup(vss->vsa_nsid);
+	if (nsp == NULL) {
+		error = EIO;
+		vss->vsa_errno = VND_E_NONETSTACK;
+		goto nak;
+	}
+
+	mutex_enter(&vsp->vns_lock);
+	if (!(vsp->vns_flags & VNS_F_NEED_ZONE)) {
+		mutex_exit(&vsp->vns_lock);
+		error = EEXIST;
+		vss->vsa_errno = VND_E_ASSOCIATED;
+		goto nak;
+	}
+
+	vsp->vns_nsd = nsp;
+	vsp->vns_flags &= ~VNS_F_NEED_ZONE;
+	vsp->vns_flags |= VNS_F_TASKQ_DISPATCHED;
+	mutex_exit(&vsp->vns_lock);
+
+	taskq_dispatch_ent(vnd_taskq, vnd_dlpi_taskq_dispatch, vsp, 0,
+	    &vsp->vns_tqe);
+
+
+	/* At this point we need to wait until we have transitioned to ONLINE */
+	mutex_enter(&vsp->vns_lock);
+	while (vsp->vns_state != VNS_S_ONLINE && vsp->vns_state != VNS_S_ZOMBIE)
+		cv_wait(&vsp->vns_stcv, &vsp->vns_lock);
+	state = vsp->vns_state;
+	mutex_exit(&vsp->vns_lock);
+
+	if (state == VNS_S_ZOMBIE) {
+		vss->vsa_errno = vsp->vns_errno;
+		error = EIO;
+		goto nak;
+	}
+
+	mutex_enter(&vdp->vdd_lock);
+	mutex_enter(&vsp->vns_lock);
+	VERIFY(vdp->vdd_str == NULL);
+	/*
+	 * Now initialize the remaining kstat properties and let's go ahead and
+	 * create it.
+	 */
+	(void) snprintf(iname, sizeof (iname), "z%d_%d",
+	    vdp->vdd_nsd->vpnd_zid, vdp->vdd_minor);
+	vsp->vns_kstat = kstat_create_zone("vnd", vdp->vdd_minor, iname, "net",
+	    KSTAT_TYPE_NAMED, sizeof (vnd_str_stat_t) / sizeof (kstat_named_t),
+	    KSTAT_FLAG_VIRTUAL, GLOBAL_ZONEID);
+	if (vsp->vns_kstat == NULL) {
+		error = EIO;
+		vss->vsa_errno = VND_E_KSTATCREATE;
+		mutex_exit(&vsp->vns_lock);
+		mutex_exit(&vdp->vdd_lock);
+		goto nak;
+	}
+	vdp->vdd_str = vsp;
+	vsp->vns_dev = vdp;
+
+	/*
+	 * Now, it's time to do the las thing that can fail, changing out the
+	 * input function. After this we know that we can receive data, so we
+	 * should make sure that we're ready.
+	 */
+	if (vnd_dld_cap_enable(vsp, vnd_mac_input) != 0) {
+		error = EIO;
+		vss->vsa_errno = VND_E_DIRECTFAIL;
+		vdp->vdd_str = NULL;
+		vsp->vns_dev = NULL;
+		mutex_exit(&vsp->vns_lock);
+		mutex_exit(&vdp->vdd_lock);
+		goto nak;
+	}
+
+	zone = zone_find_by_id(vdp->vdd_nsd->vpnd_zid);
+	ASSERT(zone != NULL);
+	vsp->vns_kstat->ks_data = &vsp->vns_ksdata;
+	/* Account for zone name */
+	vsp->vns_kstat->ks_data_size += strlen(zone->zone_name) + 1;
+	/* Account for eventual link name */
+	vsp->vns_kstat->ks_data_size += VND_NAMELEN;
+	kstat_named_setstr(&vsp->vns_ksdata.vks_zonename, zone->zone_name);
+	kstat_named_setstr(&vdp->vdd_str->vns_ksdata.vks_linkname,
+	    vdp->vdd_lname);
+	zone_rele(zone);
+	kstat_install(vsp->vns_kstat);
+
+	mutex_exit(&vsp->vns_lock);
+	mutex_exit(&vdp->vdd_lock);
+
+	/*
+	 * Note that the vnd_str_t does not keep a permanent hold on the
+	 * vnd_pnsd_t. We leave that up to the vnd_dev_t as that's also what
+	 * the nestack goes through to take care of everything.
+	 */
+	vss->vsa_errno = VND_E_SUCCESS;
+nak:
+	if (vdp != NULL)
+		vnd_dev_rele(vdp);
+	if (nsp != NULL)
+		vnd_nsd_rele(nsp);
+	/*
+	 * Change the copyin request to a copyout. Note that we can't use
+	 * mcopyout here as it only works when the DB_TYPE is M_IOCTL. That's
+	 * okay, as the copyin vs. copyout is basically the same.
+	 */
+	DB_TYPE(mp) = M_COPYOUT;
+	visp->vs_state = VSS_COPYOUT;
+	crp = (struct copyreq *)mp->b_rptr;
+	crp->cq_private = (void *)visp;
+	crp->cq_addr = visp->vs_addr;
+	crp->cq_size = sizeof (vnd_strioc_associate_t);
+	qreply(q, mp);
+}
+
+static void
+vnd_stroutdata(queue_t *q, vnd_str_t *vsp, mblk_t *mp, struct copyresp *csp)
+{
+	ASSERT(csp->cp_private != NULL);
+	kmem_free(csp->cp_private, sizeof (vnd_strioc_t));
+	if (csp->cp_cmd != VND_STRIOC_ASSOCIATE) {
+		if (q->q_next != NULL) {
+			putnext(q, mp);
+		} else {
+			VND_STAT_INC(vsp, vks_ndlpidrops, 1);
+			VND_STAT_INC(vsp, vks_tdrops, 1);
+			vnd_drop_ctl(vsp, mp, "uknown cmd for M_IOCDATA");
+		}
+		return;
+	}
+
+	/* The nak is already sent for us */
+	if (csp->cp_rval != 0) {
+		vnd_drop_ctl(vsp, mp, "M_COPYOUT failed");
+		return;
+	}
+
+	/* Ack and let's be done with it all */
+	miocack(q, mp, 0, 0);
+}
+
+static int
+vnd_s_wput(queue_t *q, mblk_t *mp)
+{
+	vnd_str_t *vsp = q->q_ptr;
+	struct copyresp *crp;
+	vnd_strioc_state_t vstate;
+	vnd_strioc_t *visp;
+
+	switch (DB_TYPE(mp)) {
+	case M_IOCTL:
+		vnd_strioctl(q, vsp, mp, (struct iocblk *)mp->b_rptr);
+		return (0);
+	case M_IOCDATA:
+		crp = (struct copyresp *)mp->b_rptr;
+		ASSERT(crp->cp_private != NULL);
+		visp = (vnd_strioc_t *)crp->cp_private;
+		vstate = visp->vs_state;
+		ASSERT(vstate == VSS_COPYIN || vstate == VSS_COPYOUT);
+		if (vstate == VSS_COPYIN)
+			vnd_striocdata(q, vsp, mp,
+			    (struct copyresp *)mp->b_rptr);
+		else
+			vnd_stroutdata(q, vsp, mp,
+			    (struct copyresp *)mp->b_rptr);
+		return (0);
+	default:
+		break;
+	}
+	if (q->q_next != NULL)
+		putnext(q, mp);
+	else
+		vnd_drop_ctl(vsp, mp, "!M_IOCTL in wput");
+
+	return (0);
+}
+
+static int
+vnd_s_open(queue_t *q, dev_t *devp, int oflag, int sflag, cred_t *credp)
+{
+	vnd_str_t *vsp;
+	uint_t rand;
+
+	if (q->q_ptr != NULL)
+		return (EINVAL);
+
+	if (!(sflag & MODOPEN))
+		return (ENXIO);
+
+	if (credp != kcred)
+		return (EPERM);
+
+	vsp = kmem_cache_alloc(vnd_str_cache, KM_SLEEP);
+	bzero(vsp, sizeof (*vsp));
+	mutex_init(&vsp->vns_lock, NULL, MUTEX_DRIVER, NULL);
+	cv_init(&vsp->vns_cancelcv, NULL, CV_DRIVER, NULL);
+	cv_init(&vsp->vns_barriercv, NULL, CV_DRIVER, NULL);
+	cv_init(&vsp->vns_stcv, NULL, CV_DRIVER, NULL);
+	vsp->vns_state = VNS_S_INITIAL;
+
+	mutex_init(&vsp->vns_dq_read.vdq_lock, NULL, MUTEX_DRIVER, NULL);
+	mutex_init(&vsp->vns_dq_write.vdq_lock, NULL, MUTEX_DRIVER, NULL);
+	mutex_enter(&vnd_dev_lock);
+	vsp->vns_dq_read.vdq_max = vnd_vdq_default_size;
+	vsp->vns_dq_read.vdq_vns = vsp;
+	vsp->vns_dq_write.vdq_max = vnd_vdq_default_size;
+	vsp->vns_dq_write.vdq_vns = vsp;
+	mutex_exit(&vnd_dev_lock);
+	vsp->vns_rq = q;
+	vsp->vns_wq = WR(q);
+	q->q_ptr = WR(q)->q_ptr = vsp;
+	vsp->vns_flags = VNS_F_NEED_ZONE;
+	vsp->vns_nflush = vnd_flush_nburst;
+	vsp->vns_bsize = vnd_flush_burst_size;
+
+	(void) random_get_pseudo_bytes((uint8_t *)&rand, sizeof (rand));
+	vsp->vns_squeue = gsqueue_set_get(vnd_sqset, rand);
+
+	/*
+	 * We create our kstat and initialize all of its fields now, but we
+	 * don't install it until we actually do the zone association so we can
+	 * get everything.
+	 */
+	kstat_named_init(&vsp->vns_ksdata.vks_rbytes, "rbytes",
+	    KSTAT_DATA_UINT64);
+	kstat_named_init(&vsp->vns_ksdata.vks_rpackets, "rpackets",
+	    KSTAT_DATA_UINT64);
+	kstat_named_init(&vsp->vns_ksdata.vks_obytes, "obytes",
+	    KSTAT_DATA_UINT64);
+	kstat_named_init(&vsp->vns_ksdata.vks_opackets, "opackets",
+	    KSTAT_DATA_UINT64);
+	kstat_named_init(&vsp->vns_ksdata.vks_nhookindrops, "nhookindrops",
+	    KSTAT_DATA_UINT64);
+	kstat_named_init(&vsp->vns_ksdata.vks_nhookoutdrops, "nhookoutdrops",
+	    KSTAT_DATA_UINT64);
+	kstat_named_init(&vsp->vns_ksdata.vks_ndlpidrops, "ndlpidrops",
+	    KSTAT_DATA_UINT64);
+	kstat_named_init(&vsp->vns_ksdata.vks_ndataindrops, "ndataindrops",
+	    KSTAT_DATA_UINT64);
+	kstat_named_init(&vsp->vns_ksdata.vks_ndataoutdrops, "ndataoutdrops",
+	    KSTAT_DATA_UINT64);
+	kstat_named_init(&vsp->vns_ksdata.vks_tdrops, "total_drops",
+	    KSTAT_DATA_UINT64);
+	kstat_named_init(&vsp->vns_ksdata.vks_linkname, "linkname",
+	    KSTAT_DATA_STRING);
+	kstat_named_init(&vsp->vns_ksdata.vks_zonename, "zonename",
+	    KSTAT_DATA_STRING);
+	kstat_named_init(&vsp->vns_ksdata.vks_nmacflow, "flowcontrol_events",
+	    KSTAT_DATA_UINT64);
+	kstat_named_init(&vsp->vns_ksdata.vks_tmacflow, "flowcontrol_time",
+	    KSTAT_DATA_UINT64);
+	kstat_named_init(&vsp->vns_ksdata.vks_mac_flow_1ms, "flowcontrol_1ms",
+	    KSTAT_DATA_UINT64);
+	kstat_named_init(&vsp->vns_ksdata.vks_mac_flow_10ms, "flowcontrol_10ms",
+	    KSTAT_DATA_UINT64);
+	kstat_named_init(&vsp->vns_ksdata.vks_mac_flow_100ms,
+	    "flowcontrol_100ms", KSTAT_DATA_UINT64);
+	kstat_named_init(&vsp->vns_ksdata.vks_mac_flow_1s, "flowcontrol_1s",
+	    KSTAT_DATA_UINT64);
+	kstat_named_init(&vsp->vns_ksdata.vks_mac_flow_10s, "flowcontrol_10s",
+	    KSTAT_DATA_UINT64);
+	qprocson(q);
+	/*
+	 * Now that we've called qprocson, grab the lower module for making sure
+	 * that we don't have any pass through modules.
+	 */
+	vsp->vns_lrq = RD(vsp->vns_wq->q_next);
+
+	return (0);
+}
+
+static int
+vnd_s_close(queue_t *q, int flag, cred_t *credp)
+{
+	vnd_str_t *vsp;
+	mblk_t *mp;
+
+	VERIFY(WR(q)->q_next != NULL);
+
+	vsp = q->q_ptr;
+	ASSERT(vsp != NULL);
+
+	/*
+	 * We need to transition ourselves down.  This means that we have a few
+	 * important different things to do in the process of tearing down our
+	 * input and output buffers, making sure we've drained the current
+	 * squeue, and disabling the fast path. Before we disable the fast path,
+	 * we should make sure the squeue is drained. Because we're in streams
+	 * close, we know that no packets can come into us from userland, but we
+	 * can receive more. As such, the following is the exact order of things
+	 * that we do:
+	 *
+	 * 1) flush the vns_dq_read
+	 * 2) Insert the drain mblk
+	 * 3) When it's been received, tear down the fast path by kicking
+	 * off the state machine.
+	 * 4) One final flush of both the vns_dq_read,vns_dq_write
+	 */
+
+	vnd_dq_flush(&vsp->vns_dq_read, vnd_drop_in);
+	vnd_strbarrier(vsp);
+	mutex_enter(&vsp->vns_lock);
+	vsp->vns_state = VNS_S_SHUTTING_DOWN;
+	if (!(vsp->vns_flags & VNS_F_TASKQ_DISPATCHED)) {
+		vsp->vns_flags |= VNS_F_TASKQ_DISPATCHED;
+		taskq_dispatch_ent(vnd_taskq, vnd_dlpi_taskq_dispatch, vsp,
+		    0, &vsp->vns_tqe);
+	}
+	while (vsp->vns_state != VNS_S_ZOMBIE)
+		cv_wait(&vsp->vns_stcv, &vsp->vns_lock);
+	mutex_exit(&vsp->vns_lock);
+
+	qprocsoff(q);
+	mutex_enter(&vsp->vns_lock);
+	vsp->vns_flags |= VNS_F_CONDEMNED;
+	while (vsp->vns_flags & VNS_F_TASKQ_DISPATCHED)
+		cv_wait(&vsp->vns_cancelcv, &vsp->vns_lock);
+
+	while ((mp = vnd_dlpi_inc_pop(vsp)) != NULL)
+		vnd_drop_ctl(vsp, mp, "vnd_s_close");
+	mutex_exit(&vsp->vns_lock);
+
+	q->q_ptr = NULL;
+	vnd_dq_flush(&vsp->vns_dq_read, vnd_drop_in);
+	vnd_dq_flush(&vsp->vns_dq_write, vnd_drop_out);
+	mutex_destroy(&vsp->vns_dq_read.vdq_lock);
+	mutex_destroy(&vsp->vns_dq_write.vdq_lock);
+
+	if (vsp->vns_kstat != NULL)
+		kstat_delete(vsp->vns_kstat);
+	mutex_destroy(&vsp->vns_lock);
+	cv_destroy(&vsp->vns_stcv);
+	cv_destroy(&vsp->vns_barriercv);
+	cv_destroy(&vsp->vns_cancelcv);
+	kmem_cache_free(vnd_str_cache, vsp);
+
+	return (0);
+}
+
+static vnd_mac_cookie_t
+vnd_squeue_tx_one(vnd_str_t *vsp, mblk_t *mp)
+{
+	hrtime_t txtime;
+	vnd_mac_cookie_t vc;
+
+	VND_STAT_INC(vsp, vks_opackets, 1);
+	VND_STAT_INC(vsp, vks_obytes, msgsize(mp));
+	DTRACE_VND5(send, mblk_t *, mp, void *, NULL, void *, NULL,
+	    vnd_str_t *, vsp, mblk_t *, mp);
+	/* Actually tx now */
+	txtime = gethrtime();
+	vc = vsp->vns_caps.vsc_tx_f(vsp->vns_caps.vsc_tx_hdl,
+	    mp, 0, MAC_DROP_ON_NO_DESC);
+
+	/*
+	 * We need to check two different conditions before we immediately set
+	 * the flow control lock. The first thing that we need to do is verify
+	 * that this is an instance of hard flow control, so to say. The flow
+	 * control callbacks won't always fire in cases where we still get a
+	 * cookie returned. The explicit check for flow control will guarantee
+	 * us that we'll get a subsequent notification callback.
+	 *
+	 * The second case comes about because we do not hold the
+	 * vnd_str_t`vns_lock across calls to tx, we need to determine if a flow
+	 * control notification already came across for us in a different thread
+	 * calling vnd_mac_flow_control(). To deal with this, we record a
+	 * timestamp every time that we change the flow control state. We grab
+	 * txtime here before we transmit because that guarantees that the
+	 * hrtime_t of the call to vnd_mac_flow_control() will be after txtime.
+	 *
+	 * If the flow control notification beat us to the punch, the value of
+	 * vns_fcupdate will be larger than the value of txtime, and we should
+	 * just record the statistics. However, if we didn't beat it to the
+	 * punch (txtime > vns_fcupdate), then we know that it's safe to wait
+	 * for a notification.
+	 */
+	if (vc != NULL) {
+		hrtime_t diff;
+
+		if (vsp->vns_caps.vsc_is_fc_f(vsp->vns_caps.vsc_is_fc_hdl,
+		    vc) == 0)
+			return (NULL);
+		mutex_enter(&vsp->vns_lock);
+		diff = vsp->vns_fcupdate - txtime;
+		if (diff > 0) {
+			mutex_exit(&vsp->vns_lock);
+			vnd_mac_flow_control_stat(vsp, diff);
+			return (NULL);
+		}
+		vsp->vns_flags |= VNS_F_FLOW_CONTROLLED;
+		vsp->vns_caps.vsc_fc_cookie = vc;
+		vsp->vns_fclatch = txtime;
+		vsp->vns_fcupdate = txtime;
+		DTRACE_VND3(flow__blocked, vnd_str_t *, vsp,
+		    uint64_t, vsp->vns_dq_write.vdq_cur, uintptr_t, vc);
+		mutex_exit(&vsp->vns_lock);
+	}
+
+	return (vc);
+}
+
+static void
+vnd_squeue_tx_drain(void *arg, mblk_t *drain_mp, gsqueue_t *gsp, void *dummy)
+{
+	mblk_t *mp;
+	int nmps;
+	size_t mptot, nflush, bsize;
+	boolean_t blocked, empty;
+	vnd_data_queue_t *vqp;
+	vnd_str_t *vsp = arg;
+
+	mutex_enter(&vsp->vns_lock);
+	/*
+	 * We either enter here via an squeue or via vnd_squeue_tx_append(). In
+	 * the former case we need to mark that there is no longer an active
+	 * user of the drain block.
+	 */
+	if (drain_mp != NULL) {
+		VERIFY(drain_mp == &vsp->vns_drainblk);
+		VERIFY(vsp->vns_flags & VNS_F_DRAIN_SCHEDULED);
+		vsp->vns_flags &= ~VNS_F_DRAIN_SCHEDULED;
+	}
+
+	/*
+	 * If we're still flow controlled or under a flush barrier, nothing to
+	 * do.
+	 */
+	if (vsp->vns_flags & (VNS_F_FLOW_CONTROLLED | VNS_F_BARRIER)) {
+		mutex_exit(&vsp->vns_lock);
+		return;
+	}
+
+	nflush = vsp->vns_nflush;
+	bsize = vsp->vns_bsize;
+	mutex_exit(&vsp->vns_lock);
+
+	nmps = 0;
+	mptot = 0;
+	blocked = B_FALSE;
+	vqp = &vsp->vns_dq_write;
+	while (nmps < nflush && mptot <= bsize) {
+		mutex_enter(&vqp->vdq_lock);
+		if (vnd_dq_pop(vqp, &mp) == 0) {
+			mutex_exit(&vqp->vdq_lock);
+			break;
+		}
+		mutex_exit(&vqp->vdq_lock);
+
+		nmps++;
+		mptot += msgsize(mp);
+		if (vnd_squeue_tx_one(vsp, mp) != NULL) {
+			blocked = B_TRUE;
+			break;
+		}
+	}
+
+	empty = vnd_dq_is_empty(&vsp->vns_dq_write);
+
+	/*
+	 * If the queue is not empty, we're not blocked, and there isn't a drain
+	 * scheduled, put it into the squeue with the drain block and
+	 * GSQUEUE_FILL.
+	 */
+	if (blocked == B_FALSE && empty == B_FALSE) {
+		mutex_enter(&vsp->vns_lock);
+		if (!(vsp->vns_flags & VNS_F_DRAIN_SCHEDULED)) {
+			mblk_t *mp = &vsp->vns_drainblk;
+			vsp->vns_flags |= VNS_F_DRAIN_SCHEDULED;
+			gsqueue_enter_one(vsp->vns_squeue,
+			    mp, vnd_squeue_tx_drain, vsp,
+			    GSQUEUE_FILL, VND_SQUEUE_TAG_TX_DRAIN);
+		}
+		mutex_exit(&vsp->vns_lock);
+	}
+
+	/*
+	 * If we drained some amount of data, we need to signal the data queue.
+	 */
+	if (nmps > 0) {
+		cv_broadcast(&vsp->vns_dq_write.vdq_ready);
+		pollwakeup(&vsp->vns_dev->vdd_ph, POLLOUT);
+	}
+}
+
+static void
+vnd_squeue_tx_append(void *arg, mblk_t *mp, gsqueue_t *gsp, void *dummy)
+{
+	vnd_str_t *vsp = arg;
+	vnd_data_queue_t *vqp = &vsp->vns_dq_write;
+	vnd_pnsd_t *nsp = vsp->vns_nsd;
+	size_t len = msgsize(mp);
+
+	/*
+	 * Before we append this packet, we should run it through the firewall
+	 * rules.
+	 */
+	if (nsp->vpnd_hooked && vnd_hook(vsp, &mp, nsp->vpnd_neti_v4,
+	    nsp->vpnd_event_out_v4, nsp->vpnd_token_out_v4, nsp->vpnd_neti_v6,
+	    nsp->vpnd_event_out_v6, nsp->vpnd_token_out_v6, vnd_drop_hook_out,
+	    vnd_drop_out) != 0) {
+		/*
+		 * Because we earlier reserved space for this packet and it's
+		 * not making the cut, we need to go through and unreserve that
+		 * space. Also note that the message block will likely be freed
+		 * by the time we return from vnd_hook so we cannot rely on it.
+		 */
+		mutex_enter(&vqp->vdq_lock);
+		vnd_dq_unreserve(vqp, len);
+		mutex_exit(&vqp->vdq_lock);
+		return;
+	}
+
+	/*
+	 * We earlier reserved space for this packet. So for now simply append
+	 * it and call drain. We know that no other drain can be going on right
+	 * now thanks to the squeue.
+	 */
+	mutex_enter(&vqp->vdq_lock);
+	(void) vnd_dq_push(&vsp->vns_dq_write, mp, B_TRUE, vnd_drop_panic);
+	mutex_exit(&vqp->vdq_lock);
+	vnd_squeue_tx_drain(vsp, NULL, NULL, NULL);
+}
+
+/*
+ * We need to see if this is a valid name of sorts for us. That means a few
+ * things. First off, we can't assume that what we've been given has actually
+ * been null terminated. More importantly, that it's a valid name as far as
+ * ddi_create_minor_node is concerned (that means no '@', '/', or ' '). We
+ * further constrain ourselves to simply alphanumeric characters and a few
+ * additional ones, ':', '-', and '_'.
+ */
+static int
+vnd_validate_name(const char *buf, size_t buflen)
+{
+	int i, len;
+
+	/* First make sure a null terminator exists */
+	for (i = 0; i < buflen; i++)
+		if (buf[i] == '\0')
+			break;
+	len = i;
+	if (i == 0 || i == buflen)
+		return (0);
+
+	for (i = 0; i < len; i++)
+		if (!isalnum(buf[i]) && buf[i] != ':' && buf[i] != '-' &&
+		    buf[i] != '_')
+			return (0);
+
+	return (1);
+}
+
+static int
+vnd_ioctl_attach(vnd_dev_t *vdp, uintptr_t arg, cred_t *credp, int cpflag)
+{
+	vnd_ioc_attach_t via;
+	vnd_strioc_associate_t vss;
+	vnd_pnsd_t *nsp;
+	zone_t *zonep;
+	zoneid_t zid;
+	char buf[2*VND_NAMELEN];
+	int ret, rp;
+
+	if (secpolicy_net_config(credp, B_FALSE) != 0)
+		return (EPERM);
+
+	if (secpolicy_net_rawaccess(credp) != 0)
+		return (EPERM);
+
+	if (ddi_copyin((void *)arg, &via, sizeof (via), cpflag) != 0)
+		return (EFAULT);
+	via.via_errno = VND_E_SUCCESS;
+
+	if (vnd_validate_name(via.via_name, VND_NAMELEN) == 0) {
+		via.via_errno = VND_E_BADNAME;
+		ret = EIO;
+		goto errcopyout;
+	}
+
+	/*
+	 * Only the global zone can request to create a device in a different
+	 * zone.
+	 */
+	zid = crgetzoneid(credp);
+	if (zid != GLOBAL_ZONEID && via.via_zoneid != -1 &&
+	    zid != via.via_zoneid) {
+		via.via_errno = VND_E_PERM;
+		ret = EIO;
+		goto errcopyout;
+	}
+
+	if (via.via_zoneid == -1)
+		via.via_zoneid = zid;
+
+	/*
+	 * Establish the name we'll use now. We want to be extra paranoid about
+	 * the device we're opening so check that now.
+	 */
+	if (zid == GLOBAL_ZONEID && via.via_zoneid != zid) {
+		zonep = zone_find_by_id(via.via_zoneid);
+		if (zonep == NULL) {
+			via.via_errno = VND_E_NOZONE;
+			ret = EIO;
+			goto errcopyout;
+		}
+		if (snprintf(NULL, 0, "/dev/net/zone/%s/%s", zonep->zone_name,
+		    via.via_name) >= sizeof (buf)) {
+			zone_rele(zonep);
+			via.via_errno = VND_E_BADNAME;
+			ret = EIO;
+			goto errcopyout;
+		}
+		(void) snprintf(buf, sizeof (buf), "/dev/net/zone/%s/%s",
+		    zonep->zone_name, via.via_name);
+		zone_rele(zonep);
+		zonep = NULL;
+	} else {
+		if (snprintf(NULL, 0, "/dev/net/%s", via.via_name) >=
+		    sizeof (buf)) {
+			via.via_errno = VND_E_BADNAME;
+			ret = EIO;
+			goto errcopyout;
+		}
+		(void) snprintf(buf, sizeof (buf), "/dev/net/%s", via.via_name);
+	}
+
+	/*
+	 * If our zone is dying then the netstack will have been removed from
+	 * this list.
+	 */
+	nsp = vnd_nsd_lookup_by_zid(via.via_zoneid);
+	if (nsp == NULL) {
+		via.via_errno = VND_E_NOZONE;
+		ret = EIO;
+		goto errcopyout;
+	}
+
+	/*
+	 * Note we set the attached handle even though we haven't actually
+	 * finished the process of attaching the ldi handle.
+	 */
+	mutex_enter(&vdp->vdd_lock);
+	if (vdp->vdd_flags & (VND_D_ATTACHED | VND_D_ATTACH_INFLIGHT)) {
+		mutex_exit(&vdp->vdd_lock);
+		vnd_nsd_rele(nsp);
+		via.via_errno = VND_E_ATTACHED;
+		ret = EIO;
+		goto errcopyout;
+	}
+	vdp->vdd_flags |= VND_D_ATTACH_INFLIGHT;
+	ASSERT(vdp->vdd_cr == NULL);
+	crhold(credp);
+	vdp->vdd_cr = credp;
+	ASSERT(vdp->vdd_nsd == NULL);
+	vdp->vdd_nsd = nsp;
+	mutex_exit(&vdp->vdd_lock);
+
+	/*
+	 * Place an additional hold on the vnd_pnsd_t as we go through and do
+	 * all of the rest of our work. This will be the hold that we keep for
+	 * as long as this thing is attached.
+	 */
+	vnd_nsd_ref(nsp);
+
+	ret = ldi_open_by_name(buf, FREAD | FWRITE, vdp->vdd_cr,
+	    &vdp->vdd_ldih, vdp->vdd_ldiid);
+	if (ret != 0) {
+		if (ret == ENODEV)
+			via.via_errno = VND_E_NODATALINK;
+		goto err;
+	}
+
+	/*
+	 * Unfortunately the I_PUSH interface doesn't allow us a way to detect
+	 * whether or not we're coming in from a layered device. We really want
+	 * to make sure that a normal user can't push on our streams module.
+	 * Currently the only idea I have for this is to make sure that the
+	 * credp is kcred which is really terrible.
+	 */
+	ret = ldi_ioctl(vdp->vdd_ldih, I_PUSH, (intptr_t)"vnd", FKIOCTL,
+	    kcred, &rp);
+	if (ret != 0) {
+		rp = ldi_close(vdp->vdd_ldih, FREAD | FWRITE, vdp->vdd_cr);
+		VERIFY(rp == 0);
+		via.via_errno = VND_E_STRINIT;
+		ret = EIO;
+		goto err;
+	}
+
+	vss.vsa_minor = vdp->vdd_minor;
+	vss.vsa_nsid = nsp->vpnd_nsid;
+
+	ret = ldi_ioctl(vdp->vdd_ldih, VND_STRIOC_ASSOCIATE, (intptr_t)&vss,
+	    FKIOCTL, kcred, &rp);
+	if (ret != 0 || vss.vsa_errno != VND_E_SUCCESS) {
+		rp = ldi_close(vdp->vdd_ldih, FREAD | FWRITE, vdp->vdd_cr);
+		VERIFY(rp == 0);
+		if (ret == 0) {
+			via.via_errno = vss.vsa_errno;
+			ret = EIO;
+		}
+		goto err;
+	}
+
+	mutex_enter(&vdp->vdd_nsd->vpnd_lock);
+
+	/*
+	 * There's a chance that our netstack was condemned while we've had a
+	 * hold on it. As such we need to check and if so, error out.
+	 */
+	if (vdp->vdd_nsd->vpnd_flags & VND_NS_CONDEMNED) {
+		mutex_exit(&vdp->vdd_nsd->vpnd_lock);
+		rp = ldi_close(vdp->vdd_ldih, FREAD | FWRITE, vdp->vdd_cr);
+		VERIFY(rp == 0);
+		ret = EIO;
+		via.via_errno = VND_E_NOZONE;
+		goto err;
+	}
+
+	mutex_enter(&vdp->vdd_lock);
+	VERIFY(vdp->vdd_str != NULL);
+	vdp->vdd_flags &= ~VND_D_ATTACH_INFLIGHT;
+	vdp->vdd_flags |= VND_D_ATTACHED;
+	(void) strlcpy(vdp->vdd_datalink, via.via_name,
+	    sizeof (vdp->vdd_datalink));
+	list_insert_tail(&vdp->vdd_nsd->vpnd_dev_list, vdp);
+	mutex_exit(&vdp->vdd_lock);
+	mutex_exit(&vdp->vdd_nsd->vpnd_lock);
+	vnd_nsd_rele(nsp);
+
+	return (0);
+
+err:
+	mutex_enter(&vdp->vdd_lock);
+	vdp->vdd_flags &= ~VND_D_ATTACH_INFLIGHT;
+	crfree(vdp->vdd_cr);
+	vdp->vdd_cr = NULL;
+	vdp->vdd_nsd = NULL;
+	mutex_exit(&vdp->vdd_lock);
+
+	/*
+	 * We have two holds to drop here. One for our original reference and
+	 * one for the hold this operation would have represented.
+	 */
+	vnd_nsd_rele(nsp);
+	vnd_nsd_rele(nsp);
+errcopyout:
+	if (ddi_copyout(&via, (void *)arg, sizeof (via), cpflag) != 0)
+		ret = EFAULT;
+
+	return (ret);
+}
+
+static int
+vnd_ioctl_link(vnd_dev_t *vdp, intptr_t arg, cred_t *credp, int cpflag)
+{
+	int ret = 0;
+	vnd_ioc_link_t vil;
+	char mname[2*VND_NAMELEN];
+	char **c;
+	vnd_dev_t *v;
+	zoneid_t zid;
+
+	/* Not anyone can link something */
+	if (secpolicy_net_config(credp, B_FALSE) != 0)
+		return (EPERM);
+
+	if (ddi_copyin((void *)arg, &vil, sizeof (vil), cpflag) != 0)
+		return (EFAULT);
+
+	if (vnd_validate_name(vil.vil_name, VND_NAMELEN) == 0) {
+		ret = EIO;
+		vil.vil_errno = VND_E_BADNAME;
+		goto errcopyout;
+	}
+
+	c = vnd_reserved_names;
+	while (*c != NULL) {
+		if (strcmp(vil.vil_name, *c) == 0) {
+			ret = EIO;
+			vil.vil_errno = VND_E_BADNAME;
+			goto errcopyout;
+		}
+		c++;
+	}
+
+	mutex_enter(&vdp->vdd_lock);
+	if (!(vdp->vdd_flags & VND_D_ATTACHED)) {
+		mutex_exit(&vdp->vdd_lock);
+		ret = EIO;
+		vil.vil_errno = VND_E_NOTATTACHED;
+		goto errcopyout;
+	}
+
+	if (vdp->vdd_flags & VND_D_ZONE_DYING) {
+		mutex_exit(&vdp->vdd_lock);
+		ret = EIO;
+		vil.vil_errno = VND_E_NOZONE;
+		goto errcopyout;
+	}
+
+	if (vdp->vdd_flags & (VND_D_LINK_INFLIGHT | VND_D_LINKED)) {
+		mutex_exit(&vdp->vdd_lock);
+		ret = EIO;
+		vil.vil_errno = VND_E_LINKED;
+		goto errcopyout;
+	}
+	vdp->vdd_flags |= VND_D_LINK_INFLIGHT;
+	zid = vdp->vdd_nsd->vpnd_zid;
+	mutex_exit(&vdp->vdd_lock);
+
+	if (snprintf(NULL, 0, "z%d:%s", zid, vil.vil_name) >=
+	    sizeof (mname)) {
+		ret = EIO;
+		vil.vil_errno = VND_E_BADNAME;
+		goto errcopyout;
+	}
+
+	mutex_enter(&vnd_dev_lock);
+	for (v = list_head(&vnd_dev_list); v != NULL;
+	    v = list_next(&vnd_dev_list, v)) {
+		if (!(v->vdd_flags & VND_D_LINKED))
+			continue;
+
+		if (v->vdd_nsd->vpnd_zid == zid &&
+		    strcmp(v->vdd_lname, vil.vil_name) == 0) {
+			mutex_exit(&vnd_dev_lock);
+			ret = EIO;
+			vil.vil_errno = VND_E_LINKEXISTS;
+			goto error;
+		}
+	}
+
+	/*
+	 * We set the name and mark ourselves attached while holding the list
+	 * lock to ensure that no other user can mistakingly find our name.
+	 */
+	(void) snprintf(mname, sizeof (mname), "z%d:%s", zid,
+	    vil.vil_name);
+	mutex_enter(&vdp->vdd_lock);
+
+	/*
+	 * Because we dropped our lock, we need to double check whether or not
+	 * the zone was marked as dying while we were here. If it hasn't, then
+	 * it's safe for us to link it in.
+	 */
+	if (vdp->vdd_flags & VND_D_ZONE_DYING) {
+		mutex_exit(&vdp->vdd_lock);
+		mutex_exit(&vnd_dev_lock);
+		ret = EIO;
+		vil.vil_errno = VND_E_NOZONE;
+		goto error;
+	}
+
+	(void) strlcpy(vdp->vdd_lname, vil.vil_name, sizeof (vdp->vdd_lname));
+	if (ddi_create_minor_node(vnd_dip, mname, S_IFCHR, vdp->vdd_minor,
+	    DDI_PSEUDO, 0) != DDI_SUCCESS) {
+		ret = EIO;
+		vil.vil_errno = VND_E_MINORNODE;
+	} else {
+		vdp->vdd_flags &= ~VND_D_LINK_INFLIGHT;
+		vdp->vdd_flags |= VND_D_LINKED;
+		kstat_named_setstr(&vdp->vdd_str->vns_ksdata.vks_linkname,
+		    vdp->vdd_lname);
+		ret = 0;
+	}
+	mutex_exit(&vdp->vdd_lock);
+	mutex_exit(&vnd_dev_lock);
+
+	if (ret == 0) {
+		/*
+		 * Add a reference to represent that this device is linked into
+		 * the file system name space to ensure that it doesn't
+		 * disappear.
+		 */
+		vnd_dev_ref(vdp);
+		return (0);
+	}
+
+error:
+	mutex_enter(&vdp->vdd_lock);
+	vdp->vdd_flags &= ~VND_D_LINK_INFLIGHT;
+	vdp->vdd_lname[0] = '\0';
+	mutex_exit(&vdp->vdd_lock);
+
+errcopyout:
+	if (ddi_copyout(&vil, (void *)arg, sizeof (vil), cpflag) != 0)
+		ret = EFAULT;
+	return (ret);
+}
+
+/*
+ * Common unlink function. This is used both from the ioctl path and from the
+ * netstack shutdown path. The caller is required to hold the mutex on the
+ * vnd_dev_t, but they basically will have it relinquished for them. The only
+ * thing the caller is allowed to do afterward is to potentially rele the
+ * vnd_dev_t if they have their own hold. Note that only the ioctl path has its
+ * own hold.
+ */
+static void
+vnd_dev_unlink(vnd_dev_t *vdp)
+{
+	char mname[2*VND_NAMELEN];
+
+	ASSERT(MUTEX_HELD(&vdp->vdd_lock));
+
+	(void) snprintf(mname, sizeof (mname), "z%d:%s",
+	    vdp->vdd_nsd->vpnd_zid, vdp->vdd_lname);
+	ddi_remove_minor_node(vnd_dip, mname);
+	vdp->vdd_lname[0] = '\0';
+	vdp->vdd_flags &= ~VND_D_LINKED;
+	kstat_named_setstr(&vdp->vdd_str->vns_ksdata.vks_linkname,
+	    vdp->vdd_lname);
+	mutex_exit(&vdp->vdd_lock);
+
+	/*
+	 * This rele corresponds to the reference that we took in
+	 * vnd_ioctl_link.
+	 */
+	vnd_dev_rele(vdp);
+}
+
+static int
+vnd_ioctl_unlink(vnd_dev_t *vdp, intptr_t arg, cred_t *credp, int cpflag)
+{
+	int ret;
+	zoneid_t zid;
+	vnd_ioc_unlink_t viu;
+
+	/* Not anyone can unlink something */
+	if (secpolicy_net_config(credp, B_FALSE) != 0)
+		return (EPERM);
+
+	zid = crgetzoneid(credp);
+
+	if (ddi_copyin((void *)arg, &viu, sizeof (viu), cpflag) != 0)
+		return (EFAULT);
+
+	viu.viu_errno = VND_E_SUCCESS;
+
+	mutex_enter(&vdp->vdd_lock);
+	if (!(vdp->vdd_flags & VND_D_LINKED)) {
+		mutex_exit(&vdp->vdd_lock);
+		ret = EIO;
+		viu.viu_errno = VND_E_NOTLINKED;
+		goto err;
+	}
+	VERIFY(vdp->vdd_flags & VND_D_ATTACHED);
+
+	if (zid != GLOBAL_ZONEID && zid != vdp->vdd_nsd->vpnd_zid) {
+		mutex_exit(&vdp->vdd_lock);
+		ret = EIO;
+		viu.viu_errno = VND_E_PERM;
+		goto err;
+	}
+
+	/* vnd_dev_unlink releases the vdp mutex for us */
+	vnd_dev_unlink(vdp);
+	ret = 0;
+err:
+	if (ddi_copyout(&viu, (void *)arg, sizeof (viu), cpflag) != 0)
+		return (EFAULT);
+
+	return (ret);
+}
+
+static int
+vnd_ioctl_setrxbuf(vnd_dev_t *vdp, intptr_t arg, int cpflag)
+{
+	int ret;
+	vnd_ioc_buf_t vib;
+
+	if (ddi_copyin((void *)arg, &vib, sizeof (vib), cpflag) != 0)
+		return (EFAULT);
+
+	mutex_enter(&vnd_dev_lock);
+	if (vib.vib_size > vnd_vdq_hard_max) {
+		mutex_exit(&vnd_dev_lock);
+		vib.vib_errno = VND_E_BUFTOOBIG;
+		ret = EIO;
+		goto err;
+	}
+	mutex_exit(&vnd_dev_lock);
+
+	mutex_enter(&vdp->vdd_lock);
+	if (!(vdp->vdd_flags & VND_D_ATTACHED)) {
+		mutex_exit(&vdp->vdd_lock);
+		vib.vib_errno = VND_E_NOTATTACHED;
+		ret = EIO;
+		goto err;
+	}
+
+	mutex_enter(&vdp->vdd_str->vns_lock);
+	if (vib.vib_size < vdp->vdd_str->vns_minwrite) {
+		mutex_exit(&vdp->vdd_str->vns_lock);
+		mutex_exit(&vdp->vdd_lock);
+		vib.vib_errno = VND_E_BUFTOOSMALL;
+		ret = EIO;
+		goto err;
+	}
+
+	mutex_exit(&vdp->vdd_str->vns_lock);
+	mutex_enter(&vdp->vdd_str->vns_dq_read.vdq_lock);
+	vdp->vdd_str->vns_dq_read.vdq_max = vib.vib_size;
+	mutex_exit(&vdp->vdd_str->vns_dq_read.vdq_lock);
+	mutex_exit(&vdp->vdd_lock);
+	ret = 0;
+
+err:
+	if (ddi_copyout(&vib, (void *)arg, sizeof (vib), cpflag) != 0)
+		return (EFAULT);
+
+	return (ret);
+}
+
+static int
+vnd_ioctl_getrxbuf(vnd_dev_t *vdp, intptr_t arg, int cpflag)
+{
+	int ret;
+	vnd_ioc_buf_t vib;
+
+	mutex_enter(&vdp->vdd_lock);
+	if (!(vdp->vdd_flags & VND_D_ATTACHED)) {
+		mutex_exit(&vdp->vdd_lock);
+		vib.vib_errno = VND_E_NOTATTACHED;
+		ret = EIO;
+		goto err;
+	}
+
+	mutex_enter(&vdp->vdd_str->vns_dq_read.vdq_lock);
+	vib.vib_size = vdp->vdd_str->vns_dq_read.vdq_max;
+	mutex_exit(&vdp->vdd_str->vns_dq_read.vdq_lock);
+	mutex_exit(&vdp->vdd_lock);
+	ret = 0;
+
+err:
+	if (ddi_copyout(&vib, (void *)arg, sizeof (vib), cpflag) != 0)
+		return (EFAULT);
+
+	return (ret);
+}
+
+static int
+vnd_ioctl_getmaxbuf(vnd_dev_t *vdp, intptr_t arg, int cpflag)
+{
+	vnd_ioc_buf_t vib;
+
+	mutex_enter(&vnd_dev_lock);
+	vib.vib_size = vnd_vdq_hard_max;
+	mutex_exit(&vnd_dev_lock);
+
+	if (ddi_copyout(&vib, (void *)arg, sizeof (vib), cpflag) != 0)
+		return (EFAULT);
+
+	return (0);
+}
+
+static int
+vnd_ioctl_gettxbuf(vnd_dev_t *vdp, intptr_t arg, int cpflag)
+{
+	int ret;
+	vnd_ioc_buf_t vib;
+
+	mutex_enter(&vdp->vdd_lock);
+	if (!(vdp->vdd_flags & VND_D_ATTACHED)) {
+		mutex_exit(&vdp->vdd_lock);
+		vib.vib_errno = VND_E_NOTATTACHED;
+		ret = EIO;
+		goto err;
+	}
+
+	mutex_enter(&vdp->vdd_str->vns_dq_write.vdq_lock);
+	vib.vib_size = vdp->vdd_str->vns_dq_write.vdq_max;
+	mutex_exit(&vdp->vdd_str->vns_dq_write.vdq_lock);
+	mutex_exit(&vdp->vdd_lock);
+	ret = 0;
+
+err:
+	if (ddi_copyout(&vib, (void *)arg, sizeof (vib), cpflag) != 0)
+		return (EFAULT);
+
+	return (ret);
+}
+
+static int
+vnd_ioctl_settxbuf(vnd_dev_t *vdp, intptr_t arg, int cpflag)
+{
+	int ret;
+	vnd_ioc_buf_t vib;
+
+	if (ddi_copyin((void *)arg, &vib, sizeof (vib), cpflag) != 0)
+		return (EFAULT);
+
+	mutex_enter(&vnd_dev_lock);
+	if (vib.vib_size > vnd_vdq_hard_max) {
+		mutex_exit(&vnd_dev_lock);
+		vib.vib_errno = VND_E_BUFTOOBIG;
+		ret = EIO;
+		goto err;
+	}
+	mutex_exit(&vnd_dev_lock);
+
+	mutex_enter(&vdp->vdd_lock);
+	if (!(vdp->vdd_flags & VND_D_ATTACHED)) {
+		mutex_exit(&vdp->vdd_lock);
+		vib.vib_errno = VND_E_NOTATTACHED;
+		ret = EIO;
+		goto err;
+	}
+
+	mutex_enter(&vdp->vdd_str->vns_lock);
+	if (vib.vib_size < vdp->vdd_str->vns_minwrite) {
+		mutex_exit(&vdp->vdd_str->vns_lock);
+		mutex_exit(&vdp->vdd_lock);
+		vib.vib_errno = VND_E_BUFTOOSMALL;
+		ret = EIO;
+		goto err;
+	}
+	mutex_exit(&vdp->vdd_str->vns_lock);
+
+	mutex_enter(&vdp->vdd_str->vns_dq_write.vdq_lock);
+	vdp->vdd_str->vns_dq_write.vdq_max = vib.vib_size;
+	mutex_exit(&vdp->vdd_str->vns_dq_write.vdq_lock);
+	mutex_exit(&vdp->vdd_lock);
+	ret = 0;
+
+err:
+	if (ddi_copyout(&vib, (void *)arg, sizeof (vib), cpflag) != 0)
+		return (EFAULT);
+
+	return (ret);
+}
+
+static int
+vnd_ioctl_gettu(vnd_dev_t *vdp, intptr_t arg, int mode, boolean_t min)
+{
+	vnd_ioc_buf_t vib;
+
+	vib.vib_errno = 0;
+	mutex_enter(&vdp->vdd_lock);
+	if (vdp->vdd_flags & VND_D_ATTACHED) {
+		mutex_enter(&vdp->vdd_str->vns_lock);
+		if (min == B_TRUE)
+			vib.vib_size = vdp->vdd_str->vns_minwrite;
+		else
+			vib.vib_size = vdp->vdd_str->vns_maxwrite;
+		mutex_exit(&vdp->vdd_str->vns_lock);
+	} else {
+		vib.vib_errno = VND_E_NOTATTACHED;
+	}
+	mutex_exit(&vdp->vdd_lock);
+
+	if (ddi_copyout(&vib, (void *)arg, sizeof (vib), mode & FKIOCTL) != 0)
+		return (EFAULT);
+
+	return (0);
+}
+
+static int
+vnd_frameio_read(vnd_dev_t *vdp, intptr_t addr, int mode)
+{
+	int ret, nonblock, nwrite;
+	frameio_t *fio;
+	vnd_data_queue_t *vqp;
+	mblk_t *mp;
+
+	fio = frameio_alloc(KM_NOSLEEP | KM_NORMALPRI);
+	if (fio == NULL)
+		return (EAGAIN);
+
+	ret = frameio_hdr_copyin(fio, FRAMEIO_NVECS_MAX, (const void *)addr,
+	    mode);
+	if (ret != 0) {
+		frameio_free(fio);
+		return (ret);
+	}
+
+	mutex_enter(&vdp->vdd_lock);
+	if (!(vdp->vdd_flags & VND_D_ATTACHED)) {
+		mutex_exit(&vdp->vdd_lock);
+		frameio_free(fio);
+		return (ENXIO);
+	}
+	mutex_exit(&vdp->vdd_lock);
+
+	nonblock = mode & (FNONBLOCK | FNDELAY);
+
+	vqp = &vdp->vdd_str->vns_dq_read;
+	mutex_enter(&vqp->vdq_lock);
+
+	/* Check empty case */
+	if (vqp->vdq_cur == 0) {
+		if (nonblock != 0) {
+			mutex_exit(&vqp->vdq_lock);
+			frameio_free(fio);
+			return (EWOULDBLOCK);
+		}
+		while (vqp->vdq_cur == 0) {
+			if (cv_wait_sig(&vqp->vdq_ready, &vqp->vdq_lock) <= 0) {
+				mutex_exit(&vqp->vdq_lock);
+				frameio_free(fio);
+				return (EINTR);
+			}
+		}
+	}
+
+	ret = frameio_mblk_chain_write(fio, MAP_BLK_FRAME, vqp->vdq_head,
+	    &nwrite, mode & FKIOCTL);
+	if (ret != 0) {
+		mutex_exit(&vqp->vdq_lock);
+		frameio_free(fio);
+		return (ret);
+	}
+
+	ret = frameio_hdr_copyout(fio, nwrite, (void *)addr, mode);
+	if (ret != 0) {
+		mutex_exit(&vqp->vdq_lock);
+		frameio_free(fio);
+		return (ret);
+	}
+
+	while (nwrite > 0) {
+		(void) vnd_dq_pop(vqp, &mp);
+		freemsg(mp);
+		nwrite--;
+	}
+	mutex_exit(&vqp->vdq_lock);
+	frameio_free(fio);
+
+	return (0);
+}
+
+static int
+vnd_frameio_write(vnd_dev_t *vdp, intptr_t addr, int mode)
+{
+	frameio_t *fio;
+	int ret, nonblock, nframes, i, nread;
+	size_t maxwrite, minwrite, total, flen;
+	mblk_t *mp_chain, *mp, *nmp;
+	vnd_data_queue_t *vqp;
+
+	fio = frameio_alloc(KM_NOSLEEP | KM_NORMALPRI);
+	if (fio == NULL)
+		return (EAGAIN);
+
+	ret = frameio_hdr_copyin(fio, FRAMEIO_NVECS_MAX, (void *)addr, mode);
+	if (ret != 0) {
+		frameio_free(fio);
+		return (ret);
+	}
+
+	mutex_enter(&vdp->vdd_lock);
+	if (!(vdp->vdd_flags & VND_D_ATTACHED)) {
+		mutex_exit(&vdp->vdd_lock);
+		frameio_free(fio);
+		return (ENXIO);
+	}
+	mutex_exit(&vdp->vdd_lock);
+
+	nonblock = mode & (FNONBLOCK | FNDELAY);
+
+	/*
+	 * Make sure no single frame is larger than we can accept.
+	 */
+	mutex_enter(&vdp->vdd_str->vns_lock);
+	minwrite = vdp->vdd_str->vns_minwrite;
+	maxwrite = vdp->vdd_str->vns_maxwrite;
+	mutex_exit(&vdp->vdd_str->vns_lock);
+
+	nframes = fio->fio_nvpf / fio->fio_nvecs;
+	total = 0;
+	for (i = 0; i < nframes; i++) {
+		flen = frameio_frame_length(fio,
+		    &fio->fio_vecs[i*fio->fio_nvpf]);
+		if (flen < minwrite || flen > maxwrite) {
+			frameio_free(fio);
+			return (ERANGE);
+		}
+		total += flen;
+	}
+
+	vqp = &vdp->vdd_str->vns_dq_write;
+	mutex_enter(&vqp->vdq_lock);
+	while (vnd_dq_reserve(vqp, total) == 0) {
+		if (nonblock != 0) {
+			frameio_free(fio);
+			mutex_exit(&vqp->vdq_lock);
+			return (EAGAIN);
+		}
+		if (cv_wait_sig(&vqp->vdq_ready, &vqp->vdq_lock) <= 0) {
+			mutex_exit(&vqp->vdq_lock);
+			frameio_free(fio);
+			return (EINTR);
+		}
+	}
+	mutex_exit(&vqp->vdq_lock);
+
+	/*
+	 * We've reserved our space, let's copyin and go from here.
+	 */
+	ret = frameio_mblk_chain_read(fio, &mp_chain, &nread, mode & FKIOCTL);
+	if (ret != 0) {
+		frameio_free(fio);
+		vnd_dq_unreserve(vqp, total);
+		cv_broadcast(&vqp->vdq_ready);
+		pollwakeup(&vdp->vdd_ph, POLLOUT);
+		return (ret);
+	}
+
+	for (mp = mp_chain; mp != NULL; mp = nmp) {
+		nmp = mp->b_next;
+		mp->b_next = NULL;
+		gsqueue_enter_one(vdp->vdd_str->vns_squeue, mp,
+		    vnd_squeue_tx_append, vdp->vdd_str, GSQUEUE_PROCESS,
+		    VND_SQUEUE_TAG_VND_WRITE);
+	}
+
+	/*
+	 * Update the frameio structure to indicate that we wrote those frames.
+	 */
+	frameio_mark_consumed(fio, nread);
+	ret = frameio_hdr_copyout(fio, nread, (void *)addr, mode);
+	frameio_free(fio);
+
+	return (ret);
+}
+
+static int
+vnd_ioctl_list_copy_info(vnd_dev_t *vdp, vnd_ioc_info_t *arg, int mode)
+{
+	const char *link;
+	uint32_t vers = 1;
+	ASSERT(MUTEX_HELD(&vdp->vdd_lock));
+
+	/*
+	 * Copy all of the members out to userland.
+	 */
+	if (ddi_copyout(&vers, &arg->vii_version, sizeof (uint32_t),
+	    mode & FKIOCTL) != 0)
+		return (EFAULT);
+
+	if (vdp->vdd_flags & VND_D_LINKED)
+		link = vdp->vdd_lname;
+	else
+		link = "<anonymous>";
+	if (ddi_copyout(link, arg->vii_name, sizeof (arg->vii_name),
+	    mode & FKIOCTL) != 0)
+		return (EFAULT);
+
+	if (ddi_copyout(vdp->vdd_datalink, arg->vii_datalink,
+	    sizeof (arg->vii_datalink), mode & FKIOCTL) != 0)
+		return (EFAULT);
+
+	if (ddi_copyout(&vdp->vdd_nsd->vpnd_zid, &arg->vii_zone,
+	    sizeof (zoneid_t), mode & FKIOCTL) != 0)
+		return (EFAULT);
+	return (0);
+}
+
+static int
+vnd_ioctl_list(intptr_t arg, cred_t *credp, int mode)
+{
+	vnd_ioc_list_t vl;
+	vnd_ioc_list32_t vl32;
+	zoneid_t zid;
+	vnd_dev_t *vdp;
+	vnd_ioc_info_t *vip;
+	int found, cancopy, ret;
+
+	if (ddi_model_convert_from(mode & FMODELS) == DDI_MODEL_ILP32) {
+		if (ddi_copyin((void *)arg, &vl32, sizeof (vnd_ioc_list32_t),
+		    mode & FKIOCTL) != 0)
+			return (EFAULT);
+		vl.vl_nents = vl32.vl_nents;
+		vl.vl_actents = vl32.vl_actents;
+		vl.vl_ents = (void *)(uintptr_t)vl32.vl_ents;
+	} else {
+		if (ddi_copyin((void *)arg, &vl, sizeof (vnd_ioc_list_t),
+		    mode & FKIOCTL) != 0)
+			return (EFAULT);
+	}
+
+	cancopy = vl.vl_nents;
+	vip = vl.vl_ents;
+	found = 0;
+	zid = crgetzoneid(credp);
+	mutex_enter(&vnd_dev_lock);
+	for (vdp = list_head(&vnd_dev_list); vdp != NULL;
+	    vdp = list_next(&vnd_dev_list, vdp)) {
+		mutex_enter(&vdp->vdd_lock);
+		if (vdp->vdd_flags & VND_D_ATTACHED &&
+		    !(vdp->vdd_flags & (VND_D_CONDEMNED | VND_D_ZONE_DYING)) &&
+		    (zid == GLOBAL_ZONEID || zid == vdp->vdd_nsd->vpnd_zid)) {
+			found++;
+			if (cancopy > 0) {
+				ret = vnd_ioctl_list_copy_info(vdp, vip, mode);
+				if (ret != 0) {
+					mutex_exit(&vdp->vdd_lock);
+					mutex_exit(&vnd_dev_lock);
+					return (ret);
+				}
+				cancopy--;
+				vip++;
+			}
+		}
+		mutex_exit(&vdp->vdd_lock);
+	}
+	mutex_exit(&vnd_dev_lock);
+
+	if (ddi_copyout(&found, &((vnd_ioc_list_t *)arg)->vl_actents,
+	    sizeof (uint_t), mode & FKIOCTL) != 0)
+		return (EFAULT);
+
+	return (0);
+}
+
+
+static int
+vnd_ioctl(dev_t dev, int cmd, intptr_t arg, int mode, cred_t *credp,
+    int *rvalp)
+{
+	int ret;
+	minor_t m;
+	vnd_dev_t *vdp;
+
+	m = getminor(dev);
+	ASSERT(m != 0);
+
+	/*
+	 * Make sure no one has come in on an ioctl from the strioc case.
+	 */
+	if ((cmd & VND_STRIOC) == VND_STRIOC)
+		return (ENOTTY);
+
+	/*
+	 * Like close, seems like if this minor isn't found, it's a programmer
+	 * error somehow.
+	 */
+	vdp = vnd_dev_lookup(m);
+	if (vdp == NULL)
+		return (ENXIO);
+
+	switch (cmd) {
+	case VND_IOC_ATTACH:
+		if (!(mode & FWRITE)) {
+			ret = EBADF;
+			break;
+		}
+		ret = vnd_ioctl_attach(vdp, arg, credp, mode);
+		break;
+	case VND_IOC_LINK:
+		if (!(mode & FWRITE)) {
+			ret = EBADF;
+			break;
+		}
+		ret = vnd_ioctl_link(vdp, arg, credp, mode);
+		break;
+	case VND_IOC_UNLINK:
+		if (!(mode & FWRITE)) {
+			ret = EBADF;
+			break;
+		}
+		ret = vnd_ioctl_unlink(vdp, arg, credp, mode);
+		break;
+	case VND_IOC_GETRXBUF:
+		if (!(mode & FREAD)) {
+			ret = EBADF;
+			break;
+		}
+		ret = vnd_ioctl_getrxbuf(vdp, arg, mode);
+		break;
+	case VND_IOC_SETRXBUF:
+		if (!(mode & FWRITE)) {
+			ret = EBADF;
+			break;
+		}
+		ret = vnd_ioctl_setrxbuf(vdp, arg, mode);
+		break;
+	case VND_IOC_GETTXBUF:
+		if (!(mode & FREAD)) {
+			ret = EBADF;
+			break;
+		}
+		ret = vnd_ioctl_gettxbuf(vdp, arg, mode);
+		break;
+	case VND_IOC_SETTXBUF:
+		if (!(mode & FWRITE)) {
+			ret = EBADF;
+			break;
+		}
+		ret = vnd_ioctl_settxbuf(vdp, arg, mode);
+		break;
+	case VND_IOC_GETMAXBUF:
+		if (!(mode & FREAD)) {
+			ret = EBADF;
+			break;
+		}
+		if (crgetzoneid(credp) != GLOBAL_ZONEID) {
+			ret = EPERM;
+			break;
+		}
+		ret = vnd_ioctl_getmaxbuf(vdp, arg, mode);
+		break;
+	case VND_IOC_GETMINTU:
+		if (!(mode & FREAD)) {
+			ret = EBADF;
+			break;
+		}
+		ret = vnd_ioctl_gettu(vdp, arg, mode, B_TRUE);
+		break;
+	case VND_IOC_GETMAXTU:
+		if (!(mode & FREAD)) {
+			ret = EBADF;
+			break;
+		}
+		ret = vnd_ioctl_gettu(vdp, arg, mode, B_FALSE);
+		break;
+	case VND_IOC_FRAMEIO_READ:
+		if (!(mode & FREAD)) {
+			ret = EBADF;
+			break;
+		}
+		ret = vnd_frameio_read(vdp, arg, mode);
+		break;
+	case VND_IOC_FRAMEIO_WRITE:
+		if (!(mode & FWRITE)) {
+			ret = EBADF;
+			break;
+		}
+		ret = vnd_frameio_write(vdp, arg, mode);
+		break;
+	case VND_IOC_LIST:
+		if (!(mode & FREAD)) {
+			ret = EBADF;
+			break;
+		}
+		ret = vnd_ioctl_list(arg, credp, mode);
+		break;
+	default:
+		ret = ENOTTY;
+		break;
+	}
+
+	vnd_dev_rele(vdp);
+	return (ret);
+}
+
+static int
+vnd_open(dev_t *devp, int flag, int otyp, cred_t *credp)
+{
+	vnd_dev_t *vdp;
+	minor_t m;
+	zoneid_t zid;
+
+	if (flag & (FEXCL | FNDELAY))
+		return (ENOTSUP);
+
+	if (otyp & OTYP_BLK)
+		return (ENOTSUP);
+
+	zid = crgetzoneid(credp);
+	m = getminor(*devp);
+
+	/*
+	 * If we have an open of a non-zero instance then we need to look that
+	 * up in our list of entries.
+	 */
+	if (m != 0) {
+
+		/*
+		 * We don't check for rawaccess globally as a user could be
+		 * doing a list ioctl on the control node which doesn't require
+		 * this privilege.
+		 */
+		if (secpolicy_net_rawaccess(credp) != 0)
+			return (EPERM);
+
+
+		vdp = vnd_dev_lookup(m);
+		if (vdp == NULL)
+			return (ENOENT);
+
+		/*
+		 * We need to check to make sure that the user is allowed to
+		 * open this node. At this point it should be an attached handle
+		 * as that's all we're allowed to access.
+		 */
+		mutex_enter(&vdp->vdd_lock);
+		if (!(vdp->vdd_flags & VND_D_LINKED)) {
+			mutex_exit(&vdp->vdd_lock);
+			vnd_dev_rele(vdp);
+			return (ENOENT);
+		}
+
+		if (vdp->vdd_flags & VND_D_ZONE_DYING) {
+			mutex_exit(&vdp->vdd_lock);
+			vnd_dev_rele(vdp);
+			return (ENOENT);
+		}
+
+		if (zid != GLOBAL_ZONEID && zid != vdp->vdd_nsd->vpnd_zid) {
+			mutex_exit(&vdp->vdd_lock);
+			vnd_dev_rele(vdp);
+			return (ENOENT);
+		}
+
+		if ((flag & FEXCL) && (vdp->vdd_flags & VND_D_OPENED)) {
+			mutex_exit(&vdp->vdd_lock);
+			vnd_dev_rele(vdp);
+			return (EBUSY);
+		}
+
+		if (!(vdp->vdd_flags & VND_D_OPENED)) {
+			vdp->vdd_flags |= VND_D_OPENED;
+			vdp->vdd_ref++;
+			DTRACE_VND_REFINC(vdp);
+		}
+		mutex_exit(&vdp->vdd_lock);
+		vnd_dev_rele(vdp);
+
+		return (0);
+	}
+
+	if (flag & FEXCL)
+		return (ENOTSUP);
+
+	/*
+	 * We need to clone ourselves and set up new a state.
+	 */
+	vdp = kmem_cache_alloc(vnd_dev_cache, KM_SLEEP);
+	bzero(vdp, sizeof (vnd_dev_t));
+
+	if (ldi_ident_from_dev(*devp, &vdp->vdd_ldiid) != 0) {
+		kmem_cache_free(vnd_dev_cache, vdp);
+		return (EINVAL);
+	}
+
+	vdp->vdd_minor = id_alloc(vnd_minors);
+	mutex_init(&vdp->vdd_lock, NULL, MUTEX_DRIVER, NULL);
+	list_link_init(&vdp->vdd_link);
+	vdp->vdd_ref = 1;
+	*devp = makedevice(getmajor(*devp), vdp->vdd_minor);
+	vdp->vdd_devid = *devp;
+	DTRACE_VND_REFINC(vdp);
+	vdp->vdd_flags |= VND_D_OPENED;
+
+	mutex_enter(&vnd_dev_lock);
+	list_insert_head(&vnd_dev_list, vdp);
+	mutex_exit(&vnd_dev_lock);
+
+	return (0);
+}
+
+static int
+vnd_close(dev_t dev, int flag, int otyp, cred_t *credp)
+{
+	minor_t m;
+	vnd_dev_t *vdp;
+
+	m = getminor(dev);
+	if (m == 0)
+		return (ENXIO);
+
+	vdp = vnd_dev_lookup(m);
+	if (vdp == NULL)
+		return (ENXIO);
+
+	mutex_enter(&vdp->vdd_lock);
+	VERIFY(vdp->vdd_flags & VND_D_OPENED);
+	vdp->vdd_flags &= ~VND_D_OPENED;
+	mutex_exit(&vdp->vdd_lock);
+
+	/* Remove the hold from the previous open. */
+	vnd_dev_rele(vdp);
+
+	/* And now from lookup */
+	vnd_dev_rele(vdp);
+	return (0);
+}
+
+static int
+vnd_read(dev_t dev, struct uio *uiop, cred_t *credp)
+{
+	int nonblock, error = 0;
+	size_t mpsize;
+	vnd_dev_t *vdp;
+	vnd_data_queue_t *vqp;
+	mblk_t *mp = NULL;
+	offset_t u_loffset;
+
+	/*
+	 * If we have more than one uio we refuse to do anything. That's for
+	 * frameio.
+	 */
+	if (uiop->uio_iovcnt > 1)
+		return (EINVAL);
+
+	vdp = vnd_dev_lookup(getminor(dev));
+	if (vdp == NULL)
+		return (ENXIO);
+
+	mutex_enter(&vdp->vdd_lock);
+	if (!(vdp->vdd_flags & VND_D_ATTACHED)) {
+		mutex_exit(&vdp->vdd_lock);
+		vnd_dev_rele(vdp);
+		return (ENXIO);
+	}
+	mutex_exit(&vdp->vdd_lock);
+	nonblock = uiop->uio_fmode & (FNONBLOCK | FNDELAY);
+
+	vqp = &vdp->vdd_str->vns_dq_read;
+	mutex_enter(&vqp->vdq_lock);
+
+	/* Check empty case */
+	if (vqp->vdq_cur == 0) {
+		if (nonblock != 0) {
+			error = EWOULDBLOCK;
+			goto err;
+		}
+		while (vqp->vdq_cur == 0) {
+			if (cv_wait_sig(&vqp->vdq_ready, &vqp->vdq_lock) <= 0) {
+				error = EINTR;
+				goto err;
+			}
+		}
+	}
+
+	/* Ensure our buffer is big enough */
+	mp = vqp->vdq_head;
+	ASSERT(mp != NULL);
+	mpsize = msgsize(mp);
+	if (mpsize > uiop->uio_resid) {
+		error = EOVERFLOW;
+		goto err;
+	}
+
+	u_loffset = uiop->uio_loffset;
+	while (mp != NULL) {
+		if (uiomove(mp->b_rptr, MBLKL(mp), UIO_READ, uiop) != 0) {
+			error = EFAULT;
+			uiop->uio_loffset = u_loffset;
+			mp = NULL;
+			goto err;
+		}
+		mpsize -= MBLKL(mp);
+		mp = mp->b_cont;
+	}
+	ASSERT(mpsize == 0);
+	(void) vnd_dq_pop(vqp, &mp);
+	freemsg(mp);
+err:
+	mutex_exit(&vqp->vdq_lock);
+	vnd_dev_rele(vdp);
+
+	return (error);
+}
+
+static int
+vnd_write(dev_t dev, struct uio *uiop, cred_t *credp)
+{
+	int nonblock, error;
+	vnd_dev_t *vdp;
+	mblk_t *mp;
+	ssize_t iosize, origsize;
+	vnd_data_queue_t *vqp;
+
+	if (uiop->uio_iovcnt > 1)
+		return (EINVAL);
+
+	vdp = vnd_dev_lookup(getminor(dev));
+	if (vdp == NULL)
+		return (ENXIO);
+
+	mutex_enter(&vdp->vdd_lock);
+	if (!(vdp->vdd_flags & VND_D_ATTACHED)) {
+		mutex_exit(&vdp->vdd_lock);
+		vnd_dev_rele(vdp);
+		return (ENXIO);
+	}
+	mutex_exit(&vdp->vdd_lock);
+	nonblock = uiop->uio_fmode & (FNONBLOCK | FNDELAY);
+
+	mutex_enter(&vdp->vdd_str->vns_lock);
+	if (uiop->uio_resid > vdp->vdd_str->vns_maxwrite ||
+	    uiop->uio_resid < vdp->vdd_str->vns_minwrite) {
+		mutex_exit(&vdp->vdd_str->vns_lock);
+		vnd_dev_rele(vdp);
+		return (ERANGE);
+	}
+	mutex_exit(&vdp->vdd_str->vns_lock);
+	VERIFY(vdp->vdd_str != NULL);
+
+	/*
+	 * Reserve space in the data queue if we can. If we can't, block or
+	 * return EAGAIN. If we can, go and squeue_enter.
+	 */
+	vqp = &vdp->vdd_str->vns_dq_write;
+	mutex_enter(&vqp->vdq_lock);
+	while (vnd_dq_reserve(vqp, uiop->uio_resid) == 0) {
+		if (nonblock != 0) {
+			mutex_exit(&vqp->vdq_lock);
+			vnd_dev_rele(vdp);
+			return (EAGAIN);
+		}
+		if (cv_wait_sig(&vqp->vdq_ready, &vqp->vdq_lock) <= 0) {
+			mutex_exit(&vqp->vdq_lock);
+			vnd_dev_rele(vdp);
+			return (EINTR);
+		}
+	}
+	mutex_exit(&vqp->vdq_lock);
+
+	/*
+	 * Now that we've reserved the space, try to allocate kernel space for
+	 * and copy in the block. To take care of all this we use the
+	 * strmakedata subroutine for now.
+	 */
+	origsize = iosize = uiop->uio_resid;
+	error = strmakedata(&iosize, uiop, vdp->vdd_str->vns_wq->q_stream, 0,
+	    &mp);
+
+	/*
+	 * strmakedata() will return an error or it may only consume a portion
+	 * of the data.
+	 */
+	if (error != 0 || uiop->uio_resid != 0) {
+		vnd_dq_unreserve(vqp, origsize);
+		cv_broadcast(&vqp->vdq_ready);
+		pollwakeup(&vdp->vdd_ph, POLLOUT);
+		vnd_dev_rele(vdp);
+		return (ENOSR);
+	}
+
+	gsqueue_enter_one(vdp->vdd_str->vns_squeue, mp,
+	    vnd_squeue_tx_append, vdp->vdd_str, GSQUEUE_PROCESS,
+	    VND_SQUEUE_TAG_VND_WRITE);
+
+	vnd_dev_rele(vdp);
+	return (0);
+}
+
+static int
+vnd_chpoll(dev_t dev, short events, int anyyet, short *reventsp,
+    struct pollhead **phpp)
+{
+	int ready = 0;
+	vnd_dev_t *vdp;
+	vnd_data_queue_t *vqp;
+
+	vdp = vnd_dev_lookup(getminor(dev));
+	if (vdp == NULL)
+		return (ENXIO);
+
+	mutex_enter(&vdp->vdd_lock);
+	if (!(vdp->vdd_flags & VND_D_ATTACHED)) {
+		mutex_exit(&vdp->vdd_lock);
+		vnd_dev_rele(vdp);
+		return (ENXIO);
+	}
+	mutex_exit(&vdp->vdd_lock);
+
+	if ((events & POLLIN) || (events & POLLRDNORM)) {
+		vqp = &vdp->vdd_str->vns_dq_read;
+		mutex_enter(&vqp->vdq_lock);
+		if (vqp->vdq_head != NULL)
+			ready |= events & (POLLIN | POLLRDNORM);
+		mutex_exit(&vqp->vdq_lock);
+	}
+
+	if (events & POLLOUT) {
+		vqp = &vdp->vdd_str->vns_dq_write;
+		mutex_enter(&vqp->vdq_lock);
+		if (vqp->vdq_cur != vqp->vdq_max)
+			ready |= POLLOUT;
+		mutex_exit(&vqp->vdq_lock);
+	}
+
+	if (ready != 0) {
+		*reventsp = ready;
+		vnd_dev_rele(vdp);
+		return (0);
+	}
+
+	*reventsp = 0;
+	if (!anyyet)
+		*phpp = &vdp->vdd_ph;
+
+	vnd_dev_rele(vdp);
+	return (0);
+}
+
+static void *
+vnd_stack_init(netstackid_t stackid, netstack_t *ns)
+{
+	vnd_pnsd_t *nsp;
+
+	nsp = kmem_cache_alloc(vnd_pnsd_cache, KM_SLEEP);
+	bzero(nsp, sizeof (*nsp));
+	nsp->vpnd_nsid = stackid;
+	nsp->vpnd_zid = netstackid_to_zoneid(stackid);
+	nsp->vpnd_flags = 0;
+	mutex_init(&nsp->vpnd_lock, NULL, MUTEX_DRIVER, NULL);
+	list_create(&nsp->vpnd_dev_list, sizeof (vnd_dev_t),
+	    offsetof(vnd_dev_t, vdd_nslink));
+	if (vnd_netinfo_init(nsp) == 0)
+		nsp->vpnd_hooked = B_TRUE;
+
+	mutex_enter(&vnd_dev_lock);
+	list_insert_tail(&vnd_nsd_list, nsp);
+	mutex_exit(&vnd_dev_lock);
+
+	return (nsp);
+}
+
+static void
+vnd_stack_shutdown(netstackid_t stackid, void *arg)
+{
+	vnd_pnsd_t *nsp = arg;
+	vnd_dev_t *vdp;
+
+	ASSERT(nsp != NULL);
+	/*
+	 * After shut down no one should be able to find their way to this
+	 * netstack again.
+	 */
+	mutex_enter(&vnd_dev_lock);
+	list_remove(&vnd_nsd_list, nsp);
+	mutex_exit(&vnd_dev_lock);
+
+	/*
+	 * Make sure hooks know that they're going away.
+	 */
+	if (nsp->vpnd_hooked == B_TRUE)
+		vnd_netinfo_shutdown(nsp);
+
+	/*
+	 * Now we need to go through and notify each zone that they are in
+	 * teardown phase.  See the big theory statement section on vnd, zones,
+	 * netstacks, and sdev for more information about this.
+	 */
+	mutex_enter(&nsp->vpnd_lock);
+	nsp->vpnd_flags |= VND_NS_CONDEMNED;
+	for (vdp = list_head(&nsp->vpnd_dev_list); vdp != NULL;
+	    vdp = list_next(&nsp->vpnd_dev_list, vdp)) {
+		mutex_enter(&vdp->vdd_lock);
+		if (!(vdp->vdd_flags & VND_D_CONDEMNED))
+			vdp->vdd_flags |= VND_D_ZONE_DYING;
+		mutex_exit(&vdp->vdd_lock);
+	}
+	mutex_exit(&nsp->vpnd_lock);
+
+	/*
+	 * Next we remove all the links as we know nothing new can be added to
+	 * the list and that none of the extent devices can obtain additional
+	 * links.
+	 */
+restart:
+	mutex_enter(&nsp->vpnd_lock);
+	for (vdp = list_head(&nsp->vpnd_dev_list); vdp != NULL;
+	    vdp = list_next(&nsp->vpnd_dev_list, vdp)) {
+		mutex_enter(&vdp->vdd_lock);
+		if ((vdp->vdd_flags & VND_D_CONDEMNED) ||
+		    !(vdp->vdd_flags & VND_D_LINKED)) {
+			mutex_exit(&vdp->vdd_lock);
+			continue;
+		}
+
+		/*
+		 * We drop our lock here and restart afterwards. Note that as
+		 * part of unlinking we end up doing a rele of the vnd_dev_t. If
+		 * this is the final hold on the vnd_dev_t then it might try and
+		 * remove itself. Our locking rules requires not to be holding
+		 * any locks when we call any of the rele functions.
+		 *
+		 * Note that the unlink function requires holders to call into
+		 * it with the vnd_dev_t->vdd_lock held and will take care of it
+		 * for us. Because we don't have a hold on it, we're done at
+		 * this point.
+		 */
+		mutex_exit(&nsp->vpnd_lock);
+		/* Forcibly unlink */
+		vnd_dev_unlink(vdp);
+		goto restart;
+	}
+	mutex_exit(&nsp->vpnd_lock);
+}
+
+static void
+vnd_stack_destroy(netstackid_t stackid, void *arg)
+{
+	vnd_pnsd_t *nsp = arg;
+
+	ASSERT(nsp != NULL);
+
+	/*
+	 * Now that we've unlinked everything we just have to hang out for
+	 * it to finish exiting. Now that it's no longer the kernel itself
+	 * that's doing this we just need to wait for our reference count to
+	 * equal zero and then we're free. If the global zone is holding open a
+	 * reference to a vnd device for another zone, that's bad, but there's
+	 * nothing much we can do. See the section on 'vnd, zones, netstacks' in
+	 * the big theory statement for more information.
+	 */
+	mutex_enter(&nsp->vpnd_lock);
+	while (nsp->vpnd_ref != 0)
+		cv_wait(&nsp->vpnd_ref_change, &nsp->vpnd_lock);
+	mutex_exit(&nsp->vpnd_lock);
+
+	/*
+	 * During shutdown we removed ourselves from the list and now we have no
+	 * more references so we can safely say that there is nothing left and
+	 * destroy everything that we had sitting around.
+	 */
+	if (nsp->vpnd_hooked == B_TRUE)
+		vnd_netinfo_fini(nsp);
+
+	mutex_destroy(&nsp->vpnd_lock);
+	list_destroy(&nsp->vpnd_dev_list);
+	kmem_cache_free(vnd_pnsd_cache, nsp);
+}
+
+/*
+ * Convert a node with a name of the form /dev/vnd/zone/%zonename and
+ * /dev/vnd/zone/%zonename/%linkname to the corresponding vnd netstack.
+ */
+static vnd_pnsd_t *
+vnd_sdev_ctx_to_ns(sdev_ctx_t ctx)
+{
+	enum vtype vt;
+	const char *path = sdev_ctx_path(ctx);
+	char *zstart, *dup;
+	size_t duplen;
+	vnd_pnsd_t *nsp;
+
+	vt = sdev_ctx_vtype(ctx);
+	ASSERT(strncmp(path, VND_SDEV_ZROOT, strlen(VND_SDEV_ZROOT)) == 0);
+
+	if (vt == VDIR) {
+		zstart = strrchr(path, '/');
+		ASSERT(zstart != NULL);
+		zstart++;
+		return (vnd_nsd_lookup_by_zonename(zstart));
+	}
+
+	ASSERT(vt == VCHR);
+
+	dup = strdup(path);
+	duplen = strlen(dup) + 1;
+	zstart = strrchr(dup, '/');
+	*zstart = '\0';
+	zstart--;
+	zstart = strrchr(dup, '/');
+	zstart++;
+	nsp = vnd_nsd_lookup_by_zonename(zstart);
+	kmem_free(dup, duplen);
+
+	return (nsp);
+}
+
+static sdev_plugin_validate_t
+vnd_sdev_validate_dir(sdev_ctx_t ctx)
+{
+	vnd_pnsd_t *nsp;
+
+	if (strcmp(sdev_ctx_path(ctx), VND_SDEV_ROOT) == 0)
+		return (SDEV_VTOR_VALID);
+
+	if (strcmp(sdev_ctx_path(ctx), VND_SDEV_ZROOT) == 0) {
+		ASSERT(getzoneid() == GLOBAL_ZONEID);
+		ASSERT(sdev_ctx_flags(ctx) & SDEV_CTX_GLOBAL);
+		return (SDEV_VTOR_VALID);
+	}
+
+	nsp = vnd_sdev_ctx_to_ns(ctx);
+	if (nsp == NULL)
+		return (SDEV_VTOR_INVALID);
+	vnd_nsd_rele(nsp);
+
+	return (SDEV_VTOR_VALID);
+}
+
+static sdev_plugin_validate_t
+vnd_sdev_validate(sdev_ctx_t ctx)
+{
+	enum vtype vt;
+	dev_t dev;
+	vnd_dev_t *vdp;
+
+	vt = sdev_ctx_vtype(ctx);
+	if (vt == VDIR)
+		return (vnd_sdev_validate_dir(ctx));
+	ASSERT(vt == VCHR);
+
+	if (strcmp("ctl", sdev_ctx_name(ctx)) == 0)
+		return (SDEV_VTOR_VALID);
+
+	dev = (uintptr_t)sdev_ctx_vtype_data(ctx);
+	vdp = vnd_dev_lookup(getminor(dev));
+	if (vdp == NULL)
+		return (SDEV_VTOR_STALE);
+
+	mutex_enter(&vdp->vdd_lock);
+	if (!(vdp->vdd_flags & VND_D_LINKED) ||
+	    (vdp->vdd_flags & (VND_D_CONDEMNED | VND_D_ZONE_DYING))) {
+		mutex_exit(&vdp->vdd_lock);
+		vnd_dev_rele(vdp);
+		return (SDEV_VTOR_STALE);
+	}
+
+	if (strcmp(sdev_ctx_name(ctx), vdp->vdd_lname) != 0) {
+		mutex_exit(&vdp->vdd_lock);
+		vnd_dev_rele(vdp);
+		return (SDEV_VTOR_STALE);
+	}
+
+	mutex_exit(&vdp->vdd_lock);
+	vnd_dev_rele(vdp);
+	return (SDEV_VTOR_VALID);
+}
+
+/*
+ * This function is a no-op. sdev never has holds on our devices as they can go
+ * away at any time and specfs has to deal with that fact.
+ */
+static void
+vnd_sdev_inactive(sdev_ctx_t ctx)
+{
+}
+
+static int
+vnd_sdev_fillzone(vnd_pnsd_t *nsp, sdev_ctx_t ctx)
+{
+	int ret;
+	vnd_dev_t *vdp;
+
+	mutex_enter(&nsp->vpnd_lock);
+	for (vdp = list_head(&nsp->vpnd_dev_list); vdp != NULL;
+	    vdp = list_next(&nsp->vpnd_dev_list, vdp)) {
+		mutex_enter(&vdp->vdd_lock);
+		if ((vdp->vdd_flags & VND_D_LINKED) &&
+		    !(vdp->vdd_flags & (VND_D_CONDEMNED | VND_D_ZONE_DYING))) {
+			ret = sdev_plugin_mknod(ctx, vdp->vdd_lname, S_IFCHR,
+			    vdp->vdd_devid);
+			if (ret != 0 && ret != EEXIST) {
+				mutex_exit(&vdp->vdd_lock);
+				mutex_exit(&nsp->vpnd_lock);
+				vnd_nsd_rele(nsp);
+				return (ret);
+			}
+		}
+		mutex_exit(&vdp->vdd_lock);
+	}
+	mutex_exit(&nsp->vpnd_lock);
+
+	return (0);
+}
+
+static int
+vnd_sdev_filldir_root(sdev_ctx_t ctx)
+{
+	zoneid_t zid;
+	vnd_pnsd_t *nsp;
+	int ret;
+
+	zid = getzoneid();
+	nsp = vnd_nsd_lookup(zoneid_to_netstackid(zid));
+	ASSERT(nsp != NULL);
+	ret = vnd_sdev_fillzone(nsp, ctx);
+	vnd_nsd_rele(nsp);
+	if (ret != 0)
+		return (ret);
+
+	/*
+	 * Checking the zone id is not sufficient as the global zone could be
+	 * reaching down into a non-global zone's mounted /dev.
+	 */
+	if (zid == GLOBAL_ZONEID && (sdev_ctx_flags(ctx) & SDEV_CTX_GLOBAL)) {
+		ret = sdev_plugin_mkdir(ctx, "zone");
+		if (ret != 0 && ret != EEXIST)
+			return (ret);
+	}
+
+	/*
+	 * Always add a reference to the control node. There's no need to
+	 * reference it since it always exists and is always what we clone from.
+	 */
+	ret = sdev_plugin_mknod(ctx, "ctl", S_IFCHR,
+	    makedevice(ddi_driver_major(vnd_dip), 0));
+	if (ret != 0 && ret != EEXIST)
+		return (ret);
+
+	return (0);
+}
+
+static int
+vnd_sdev_filldir_zroot(sdev_ctx_t ctx)
+{
+	int ret;
+	vnd_pnsd_t *nsp;
+	zone_t *zonep;
+
+	ASSERT(getzoneid() == GLOBAL_ZONEID);
+	ASSERT(sdev_ctx_flags(ctx) & SDEV_CTX_GLOBAL);
+
+	mutex_enter(&vnd_dev_lock);
+	for (nsp = list_head(&vnd_nsd_list); nsp != NULL;
+	    nsp = list_next(&vnd_nsd_list, nsp)) {
+		mutex_enter(&nsp->vpnd_lock);
+		if (list_is_empty(&nsp->vpnd_dev_list)) {
+			mutex_exit(&nsp->vpnd_lock);
+			continue;
+		}
+		mutex_exit(&nsp->vpnd_lock);
+		zonep = zone_find_by_id(nsp->vpnd_zid);
+		/*
+		 * This zone must be being torn down, so skip it.
+		 */
+		if (zonep == NULL)
+			continue;
+		ret = sdev_plugin_mkdir(ctx, zonep->zone_name);
+		zone_rele(zonep);
+		if (ret != 0 && ret != EEXIST) {
+			mutex_exit(&vnd_dev_lock);
+			return (ret);
+		}
+	}
+	mutex_exit(&vnd_dev_lock);
+	return (0);
+}
+
+static int
+vnd_sdev_filldir(sdev_ctx_t ctx)
+{
+	int ret;
+	vnd_pnsd_t *nsp;
+
+	ASSERT(sdev_ctx_vtype(ctx) == VDIR);
+	if (strcmp(VND_SDEV_ROOT, sdev_ctx_path(ctx)) == 0)
+		return (vnd_sdev_filldir_root(ctx));
+
+	if (strcmp(VND_SDEV_ZROOT, sdev_ctx_path(ctx)) == 0)
+		return (vnd_sdev_filldir_zroot(ctx));
+
+	ASSERT(strncmp(VND_SDEV_ZROOT, sdev_ctx_path(ctx),
+	    strlen(VND_SDEV_ZROOT)) == 0);
+	nsp = vnd_sdev_ctx_to_ns(ctx);
+	if (nsp == NULL)
+		return (0);
+
+	ret = vnd_sdev_fillzone(nsp, ctx);
+	vnd_nsd_rele(nsp);
+
+	return (ret);
+}
+
+static sdev_plugin_ops_t vnd_sdev_ops = {
+	SDEV_PLUGIN_VERSION,
+	SDEV_PLUGIN_SUBDIR,
+	vnd_sdev_validate,
+	vnd_sdev_filldir,
+	vnd_sdev_inactive
+};
+
+static int
+vnd_attach(dev_info_t *dip, ddi_attach_cmd_t cmd)
+{
+	int errp = 0;
+
+	if (cmd != DDI_ATTACH)
+		return (DDI_FAILURE);
+
+	/*
+	 * Only allow one instance.
+	 */
+	if (vnd_dip != NULL)
+		return (DDI_FAILURE);
+
+	vnd_dip = dip;
+	if (ddi_create_minor_node(vnd_dip, "vnd", S_IFCHR, 0, DDI_PSEUDO, 0) !=
+	    DDI_SUCCESS) {
+		vnd_dip = NULL;
+		return (DDI_FAILURE);
+	}
+
+	if (ddi_prop_create(DDI_DEV_T_NONE, dip, DDI_PROP_CANSLEEP,
+	    DDI_KERNEL_IOCTL, NULL, 0) != DDI_PROP_SUCCESS) {
+		ddi_remove_minor_node(vnd_dip, NULL);
+		vnd_dip = NULL;
+		return (DDI_FAILURE);
+	}
+
+	vnd_sdev_hdl = sdev_plugin_register(VND_SDEV_NAME, &vnd_sdev_ops,
+	    &errp);
+	if (vnd_sdev_hdl == NULL) {
+		ddi_remove_minor_node(vnd_dip, NULL);
+		ddi_prop_remove_all(vnd_dip);
+		vnd_dip = NULL;
+		return (DDI_FAILURE);
+	}
+
+	vnd_sqset = gsqueue_set_create(GSQUEUE_DEFAULT_WAIT,
+	    GSQUEUE_DEFAULT_PRIORITY);
+
+	return (DDI_SUCCESS);
+}
+
+static int
+vnd_detach(dev_info_t *dip, ddi_detach_cmd_t cmd)
+{
+	if (cmd != DDI_DETACH)
+		return (DDI_FAILURE);
+
+	mutex_enter(&vnd_dev_lock);
+	if (!list_is_empty(&vnd_dev_list)) {
+		mutex_exit(&vnd_dev_lock);
+		return (DDI_FAILURE);
+	}
+	mutex_exit(&vnd_dev_lock);
+
+	return (DDI_FAILURE);
+}
+
+static int
+vnd_info(dev_info_t *dip, ddi_info_cmd_t cmd, void *arg, void **result)
+{
+	int error;
+
+	switch (cmd) {
+	case DDI_INFO_DEVT2DEVINFO:
+		*result = (void *)vnd_dip;
+		error = DDI_SUCCESS;
+		break;
+	case DDI_INFO_DEVT2INSTANCE:
+		*result = (void *)0;
+		error = DDI_SUCCESS;
+		break;
+	default:
+		error = DDI_FAILURE;
+		break;
+	}
+	return (error);
+}
+
+
+
+static void
+vnd_ddi_fini(void)
+{
+	netstack_unregister(NS_VND);
+	if (vnd_taskq != NULL)
+		taskq_destroy(vnd_taskq);
+	if (vnd_str_cache != NULL)
+		kmem_cache_destroy(vnd_str_cache);
+	if (vnd_dev_cache != NULL)
+		kmem_cache_destroy(vnd_dev_cache);
+	if (vnd_pnsd_cache != NULL)
+		kmem_cache_destroy(vnd_pnsd_cache);
+	if (vnd_minors != NULL)
+		id_space_destroy(vnd_minors);
+	if (vnd_list_init != 0) {
+		list_destroy(&vnd_nsd_list);
+		list_destroy(&vnd_dev_list);
+		mutex_destroy(&vnd_dev_lock);
+		vnd_list_init = 0;
+	}
+	frameio_fini();
+}
+
+static int
+vnd_ddi_init(void)
+{
+	if (frameio_init() != 0)
+		return (DDI_FAILURE);
+
+	vnd_str_cache = kmem_cache_create("vnd_str_cache", sizeof (vnd_str_t),
+	    0, NULL, NULL, NULL, NULL, NULL, 0);
+	if (vnd_str_cache == NULL) {
+		frameio_fini();
+		return (DDI_FAILURE);
+	}
+	vnd_dev_cache = kmem_cache_create("vnd_dev_cache", sizeof (vnd_dev_t),
+	    0, NULL, NULL, NULL, NULL, NULL, 0);
+	if (vnd_dev_cache == NULL) {
+		kmem_cache_destroy(vnd_str_cache);
+		frameio_fini();
+		return (DDI_FAILURE);
+	}
+	vnd_pnsd_cache = kmem_cache_create("vnd_pnsd_cache",
+	    sizeof (vnd_pnsd_t), 0, NULL, NULL, NULL, NULL, NULL, 0);
+	if (vnd_pnsd_cache == NULL) {
+		kmem_cache_destroy(vnd_dev_cache);
+		kmem_cache_destroy(vnd_str_cache);
+		frameio_fini();
+		return (DDI_FAILURE);
+	}
+
+	vnd_taskq = taskq_create_instance("vnd", -1, 1, minclsyspri, 0, 0, 0);
+	if (vnd_taskq == NULL) {
+		kmem_cache_destroy(vnd_pnsd_cache);
+		kmem_cache_destroy(vnd_dev_cache);
+		kmem_cache_destroy(vnd_str_cache);
+		frameio_fini();
+		return (DDI_FAILURE);
+	}
+
+	vnd_minors = id_space_create("vnd_minors", 1, INT32_MAX);
+	if (vnd_minors == NULL) {
+		taskq_destroy(vnd_taskq);
+		kmem_cache_destroy(vnd_pnsd_cache);
+		kmem_cache_destroy(vnd_dev_cache);
+		kmem_cache_destroy(vnd_str_cache);
+		frameio_fini();
+		return (DDI_FAILURE);
+	}
+
+	mutex_init(&vnd_dev_lock, NULL, MUTEX_DRIVER, NULL);
+	list_create(&vnd_dev_list, sizeof (vnd_dev_t),
+	    offsetof(vnd_dev_t, vdd_link));
+	list_create(&vnd_nsd_list, sizeof (vnd_pnsd_t),
+	    offsetof(vnd_pnsd_t, vpnd_link));
+	vnd_list_init = 1;
+
+	netstack_register(NS_VND, vnd_stack_init, vnd_stack_shutdown,
+	    vnd_stack_destroy);
+
+	return (DDI_SUCCESS);
+}
+
+static struct module_info vnd_minfo = {
+	0,		/* module id */
+	"vnd",		/* module name */
+	1,		/* smallest packet size */
+	INFPSZ,		/* largest packet size (infinite) */
+	1,		/* high watermark */
+	0		/* low watermark */
+};
+
+static struct qinit vnd_r_qinit = {
+	vnd_s_rput,
+	NULL,
+	vnd_s_open,
+	vnd_s_close,
+	NULL,
+	&vnd_minfo,
+	NULL
+};
+
+static struct qinit vnd_w_qinit = {
+	vnd_s_wput,
+	NULL,
+	NULL,
+	NULL,
+	NULL,
+	&vnd_minfo,
+	NULL
+};
+
+static struct streamtab vnd_strtab = {
+	&vnd_r_qinit,
+	&vnd_w_qinit,
+	NULL,
+	NULL
+};
+
+
+static struct cb_ops vnd_cb_ops = {
+	vnd_open,		/* open */
+	vnd_close,		/* close */
+	nulldev,		/* strategy */
+	nulldev,		/* print */
+	nodev,			/* dump */
+	vnd_read,		/* read */
+	vnd_write,		/* write */
+	vnd_ioctl,		/* ioctl */
+	nodev,			/* devmap */
+	nodev,			/* mmap */
+	nodev,			/* segmap */
+	vnd_chpoll,		/* poll */
+	ddi_prop_op,		/* cb_prop_op */
+	NULL,			/* streamtab  */
+	D_MP			/* Driver compatibility flag */
+};
+
+static struct dev_ops vnd_dev_ops = {
+	DEVO_REV,		/* devo_rev */
+	0,			/* refcnt */
+	vnd_info,		/* get_dev_info */
+	nulldev,		/* identify */
+	nulldev,		/* probe */
+	vnd_attach,		/* attach */
+	vnd_detach,		/* detach */
+	nodev,			/* reset */
+	&vnd_cb_ops,		/* driver operations */
+	NULL,			/* bus operations */
+	nodev,			/* dev power */
+	ddi_quiesce_not_needed	/* quiesce */
+};
+
+static struct modldrv vnd_modldrv = {
+	&mod_driverops,
+	"Virtual Networking Datapath Driver",
+	&vnd_dev_ops
+};
+
+static struct fmodsw vnd_fmodfsw = {
+	"vnd",
+	&vnd_strtab,
+	D_NEW | D_MP
+};
+
+static struct modlstrmod vnd_modlstrmod = {
+	&mod_strmodops,
+	"Virtual Networking Datapath Driver",
+	&vnd_fmodfsw
+};
+
+static struct modlinkage vnd_modlinkage = {
+	MODREV_1,
+	&vnd_modldrv,
+	&vnd_modlstrmod,
+	NULL
+};
+
+int
+_init(void)
+{
+	int error;
+
+	/*
+	 * We need to do all of our global initialization in init as opposed to
+	 * attach and detach. The problem here is that because vnd can be used
+	 * from a stream context while being detached, we can not rely on having
+	 * run attach to create everything, alas. so it goes in _init, just like
+	 * our friend ip.
+	 */
+	if ((error = vnd_ddi_init()) != DDI_SUCCESS)
+		return (error);
+	error = mod_install((&vnd_modlinkage));
+	if (error != 0)
+		vnd_ddi_fini();
+	return (error);
+}
+
+int
+_info(struct modinfo *modinfop)
+{
+	return (mod_info(&vnd_modlinkage, modinfop));
+}
+
+int
+_fini(void)
+{
+	int error;
+
+	error = mod_remove(&vnd_modlinkage);
+	if (error == 0)
+		vnd_ddi_fini();
+	return (error);
+}
diff --git a/usr/src/uts/common/io/vnd/vnd.conf b/usr/src/uts/common/io/vnd/vnd.conf
new file mode 100644
index 0000000000..65872e1ddf
--- /dev/null
+++ b/usr/src/uts/common/io/vnd/vnd.conf
@@ -0,0 +1,16 @@
+#
+# This file and its contents are supplied under the terms of the
+# Common Development and Distribution License ("CDDL"), version 1.0.
+# You may only use this file in accordance with the terms of version
+# 1.0 of the CDDL.
+#
+# A full copy of the text of the CDDL should have accompanied this
+# source.  A copy of the CDDL is also available via the Internet at
+# http://www.illumos.org/license/CDDL.
+#
+
+#
+# Copyright (c) 2014, Joyent, Inc.  All rights reserved.
+#
+
+name="vnd" parent="pseudo" instance=0;
diff --git a/usr/src/uts/common/io/vnic/vnic_dev.c b/usr/src/uts/common/io/vnic/vnic_dev.c
index 4615f00e51..c183a9c4f5 100644
--- a/usr/src/uts/common/io/vnic/vnic_dev.c
+++ b/usr/src/uts/common/io/vnic/vnic_dev.c
@@ -52,6 +52,7 @@
 #include <sys/vlan.h>
 #include <sys/vnic.h>
 #include <sys/vnic_impl.h>
+#include <sys/mac_impl.h>
 #include <sys/mac_flow_impl.h>
 #include <inet/ip_impl.h>
 
@@ -1059,6 +1060,18 @@ vnic_m_setprop(void *m_driver, const char *pr_name, mac_prop_id_t pr_num,
 		err = mac_maxsdu_update(vn->vn_mh, mtu);
 		break;
 	}
+	case MAC_PROP_VN_PROMISC_FILTERED: {
+		boolean_t filtered;
+
+		if (pr_valsize < sizeof (filtered)) {
+			err = EINVAL;
+			break;
+		}
+
+		bcopy(pr_val, &filtered, sizeof (filtered));
+		mac_set_promisc_filtered(vn->vn_mch, filtered);
+		break;
+	}
 	case MAC_PROP_SECONDARY_ADDRS: {
 		mac_secondary_addr_t msa;
 
@@ -1080,8 +1093,14 @@ vnic_m_getprop(void *arg, const char *pr_name, mac_prop_id_t pr_num,
 {
 	vnic_t		*vn = arg;
 	int 		ret = 0;
+	boolean_t	out;
 
 	switch (pr_num) {
+	case MAC_PROP_VN_PROMISC_FILTERED:
+		out = mac_get_promisc_filtered(vn->vn_mch);
+		ASSERT(pr_valsize >= sizeof (boolean_t));
+		bcopy(&out, pr_val, sizeof (boolean_t));
+		break;
 	case MAC_PROP_SECONDARY_ADDRS:
 		ret = vnic_get_secondary_macs(vn, pr_valsize, pr_val);
 		break;
diff --git a/usr/src/uts/common/io/vscan/vscan_svc.c b/usr/src/uts/common/io/vscan/vscan_svc.c
index e26fa9e292..b5efbb3a99 100644
--- a/usr/src/uts/common/io/vscan/vscan_svc.c
+++ b/usr/src/uts/common/io/vscan/vscan_svc.c
@@ -22,6 +22,7 @@
 /*
  * Copyright 2009 Sun Microsystems, Inc.  All rights reserved.
  * Use is subject to license terms.
+ * Copyright (c) 2015, Joyent, Inc.
  */
 
 #include <sys/stat.h>
@@ -461,7 +462,7 @@ vscan_svc_scan_file(vnode_t *vp, cred_t *cr, int async)
 	boolean_t allow;
 	clock_t timeout, time_left;
 
-	if ((vp == NULL) || (vp->v_path == NULL) || cr == NULL)
+	if ((vp == NULL) || (vp->v_path == vn_vpath_empty) || cr == NULL)
 		return (0);
 
 	DTRACE_PROBE2(vscan__scan__file, char *, vp->v_path, int, async);
@@ -1080,7 +1081,6 @@ vscan_svc_exempt_file(vnode_t *vp, boolean_t *allow)
 	struct vattr attr;
 
 	ASSERT(vp != NULL);
-	ASSERT(vp->v_path != NULL);
 
 	attr.va_mask = AT_SIZE;
 
diff --git a/usr/src/uts/common/io/zfd.c b/usr/src/uts/common/io/zfd.c
new file mode 100644
index 0000000000..3edaa62bbe
--- /dev/null
+++ b/usr/src/uts/common/io/zfd.c
@@ -0,0 +1,844 @@
+/*
+ * This file and its contents are supplied under the terms of the
+ * Common Development and Distribution License ("CDDL"), version 1.0.
+ * You may only use this file in accordance with the terms of version
+ * 1.0 of the CDDL.
+ *
+ * A full copy of the text of the CDDL should have accompanied this
+ * source.  A copy of the CDDL is also available via the Internet at
+ * http://www.illumos.org/license/CDDL.
+ */
+
+/*
+ * Copyright 2009 Sun Microsystems, Inc.  All rights reserved.
+ * Copyright 2014 Joyent, Inc.  All rights reserved.
+ */
+
+/*
+ * Zone File Descriptor Driver.
+ *
+ * This driver is derived from the zcons driver which is in turn derived from
+ * the pts/ptm drivers. The purpose is to expose file descriptors within the
+ * zone which are connected to zoneadmd and used for logging or an interactive
+ * connection to a process within the zone.
+ *
+ * Its implementation is straightforward. Each instance of the driver
+ * represents a global-zone/local-zone pair. Unlike the zcons device, zoneadmd
+ * uses these devices unidirectionally to provide stdin, stdout and stderr to
+ * the process within the zone.
+ *
+ * Instances of zfd are onlined as children of /pseudo/zfdnex@2/ by zoneadmd,
+ * using the devctl framework; thus the driver does not need to maintain any
+ * sort of "admin" node.
+ *
+ * The driver shuttles I/O from master side to slave side and back.  In a break
+ * from the pts/ptm semantics, if one side is not open, I/O directed towards
+ * it will simply be discarded. This is so that if zoneadmd is not holding the
+ * master side fd open (i.e. it has died somehow), processes in the zone do not
+ * experience any errors and I/O to the fd does not cause the process to hang.
+ */
+
+#include <sys/types.h>
+#include <sys/cmn_err.h>
+#include <sys/conf.h>
+#include <sys/cred.h>
+#include <sys/ddi.h>
+#include <sys/debug.h>
+#include <sys/devops.h>
+#include <sys/errno.h>
+#include <sys/file.h>
+#include <sys/kstr.h>
+#include <sys/modctl.h>
+#include <sys/param.h>
+#include <sys/stat.h>
+#include <sys/stream.h>
+#include <sys/stropts.h>
+#include <sys/strsun.h>
+#include <sys/sunddi.h>
+#include <sys/sysmacros.h>
+#include <sys/systm.h>
+#include <sys/types.h>
+#include <sys/zfd.h>
+#include <sys/vnode.h>
+#include <sys/fs/snode.h>
+#include <sys/zone.h>
+
+static int zfd_getinfo(dev_info_t *, ddi_info_cmd_t, void *, void **);
+static int zfd_attach(dev_info_t *, ddi_attach_cmd_t);
+static int zfd_detach(dev_info_t *, ddi_detach_cmd_t);
+
+static int zfd_open(queue_t *, dev_t *, int, int, cred_t *);
+static int zfd_close(queue_t *, int, cred_t *);
+static void zfd_wput(queue_t *, mblk_t *);
+static void zfd_rsrv(queue_t *);
+static void zfd_wsrv(queue_t *);
+
+/*
+ * The instance number is encoded in the dev_t in the minor number; the lowest
+ * bit of the minor number is used to track the master vs. slave side of the
+ * fd. The rest of the bits in the minor number are the instance.
+ */
+#define	ZFD_MASTER_MINOR		0
+#define	ZFD_SLAVE_MINOR		1
+
+#define	ZFD_INSTANCE(x)		(getminor((x)) >> 1)
+#define	ZFD_NODE(x)		(getminor((x)) & 0x01)
+
+/*
+ * This macro converts a zfd_state_t pointer to the associated slave minor
+ * node's dev_t.
+ */
+#define	ZFD_STATE_TO_SLAVEDEV(x)	\
+	(makedevice(ddi_driver_major((x)->zfd_devinfo), \
+	(minor_t)(ddi_get_instance((x)->zfd_devinfo) << 1 | ZFD_SLAVE_MINOR)))
+
+int zfd_debug = 0;
+#define	DBG(a)		if (zfd_debug) cmn_err(CE_NOTE, a)
+#define	DBG1(a, b)	if (zfd_debug) cmn_err(CE_NOTE, a, b)
+
+/*
+ * ZFD Pseudo Terminal Module: stream data structure definitions,
+ * based on zcons.
+ */
+static struct module_info zfd_info = {
+	0x20FD,	/* ZOFD - 8445 */
+	"zfd",
+	0,		/* min packet size */
+	INFPSZ,		/* max packet size - infinity */
+	2048,		/* high water */
+	128		/* low water */
+};
+
+static struct qinit zfd_rinit = {
+	NULL,
+	(int (*)()) zfd_rsrv,
+	zfd_open,
+	zfd_close,
+	NULL,
+	&zfd_info,
+	NULL
+};
+
+static struct qinit zfd_winit = {
+	(int (*)()) zfd_wput,
+	(int (*)()) zfd_wsrv,
+	NULL,
+	NULL,
+	NULL,
+	&zfd_info,
+	NULL
+};
+
+static struct streamtab zfd_tab_info = {
+	&zfd_rinit,
+	&zfd_winit,
+	NULL,
+	NULL
+};
+
+#define	ZFD_CONF_FLAG	(D_MP | D_MTQPAIR | D_MTOUTPERIM | D_MTOCEXCL)
+
+/*
+ * this will define (struct cb_ops cb_zfd_ops) and (struct dev_ops zfd_ops)
+ */
+DDI_DEFINE_STREAM_OPS(zfd_ops, nulldev, nulldev, zfd_attach, zfd_detach, \
+	nodev, zfd_getinfo, ZFD_CONF_FLAG, &zfd_tab_info, \
+	ddi_quiesce_not_needed);
+
+/*
+ * Module linkage information for the kernel.
+ */
+
+static struct modldrv modldrv = {
+	&mod_driverops, 	/* Type of module (this is a pseudo driver) */
+	"Zone FD driver",	/* description of module */
+	&zfd_ops		/* driver ops */
+};
+
+static struct modlinkage modlinkage = {
+	MODREV_1,
+	&modldrv,
+	NULL
+};
+
+typedef struct zfd_state {
+	dev_info_t *zfd_devinfo;
+	queue_t *zfd_master_rdq;
+	queue_t *zfd_slave_rdq;
+	vnode_t *zfd_slave_vnode;
+	int zfd_state;
+	int zfd_tty;
+} zfd_state_t;
+
+#define	ZFD_STATE_MOPEN	0x01
+#define	ZFD_STATE_SOPEN	0x02
+
+static void *zfd_soft_state;
+
+/*
+ * List of STREAMS modules that is pushed onto a slave instance after the
+ * ZFD_MAKETTY ioctl has been received.
+ */
+static char *zfd_mods[] = {
+	"ptem",
+	"ldterm",
+	"ttcompat",
+	NULL
+};
+
+int
+_init(void)
+{
+	int err;
+
+	if ((err = ddi_soft_state_init(&zfd_soft_state, sizeof (zfd_state_t),
+	    0)) != 0) {
+		return (err);
+	}
+
+	if ((err = mod_install(&modlinkage)) != 0)
+		ddi_soft_state_fini(zfd_soft_state);
+
+	return (err);
+}
+
+
+int
+_fini(void)
+{
+	int err;
+
+	if ((err = mod_remove(&modlinkage)) != 0) {
+		return (err);
+	}
+
+	ddi_soft_state_fini(&zfd_soft_state);
+	return (0);
+}
+
+int
+_info(struct modinfo *modinfop)
+{
+	return (mod_info(&modlinkage, modinfop));
+}
+
+static int
+zfd_attach(dev_info_t *dip, ddi_attach_cmd_t cmd)
+{
+	zfd_state_t *zfds;
+	int instance;
+	char masternm[ZFD_NAME_LEN], slavenm[ZFD_NAME_LEN];
+
+	if (cmd != DDI_ATTACH)
+		return (DDI_FAILURE);
+
+	instance = ddi_get_instance(dip);
+	if (ddi_soft_state_zalloc(zfd_soft_state, instance) != DDI_SUCCESS)
+		return (DDI_FAILURE);
+
+	(void) snprintf(masternm, sizeof (masternm), "%s%d", ZFD_MASTER_NAME,
+	    instance);
+	(void) snprintf(slavenm, sizeof (slavenm), "%s%d", ZFD_SLAVE_NAME,
+	    instance);
+
+	/*
+	 * Create the master and slave minor nodes.
+	 */
+	if ((ddi_create_minor_node(dip, slavenm, S_IFCHR,
+	    instance << 1 | ZFD_SLAVE_MINOR, DDI_PSEUDO, 0) == DDI_FAILURE) ||
+	    (ddi_create_minor_node(dip, masternm, S_IFCHR,
+	    instance << 1 | ZFD_MASTER_MINOR, DDI_PSEUDO, 0) == DDI_FAILURE)) {
+		ddi_remove_minor_node(dip, NULL);
+		ddi_soft_state_free(zfd_soft_state, instance);
+		return (DDI_FAILURE);
+	}
+
+	VERIFY((zfds = ddi_get_soft_state(zfd_soft_state, instance)) != NULL);
+	zfds->zfd_devinfo = dip;
+	zfds->zfd_tty = 0;
+	return (DDI_SUCCESS);
+}
+
+static int
+zfd_detach(dev_info_t *dip, ddi_detach_cmd_t cmd)
+{
+	zfd_state_t *zfds;
+	int instance;
+
+	if (cmd != DDI_DETACH)
+		return (DDI_FAILURE);
+
+	instance = ddi_get_instance(dip);
+	if ((zfds = ddi_get_soft_state(zfd_soft_state, instance)) == NULL)
+		return (DDI_FAILURE);
+
+	if ((zfds->zfd_state & ZFD_STATE_MOPEN) ||
+	    (zfds->zfd_state & ZFD_STATE_SOPEN)) {
+		DBG1("zfd_detach: device (dip=%p) still open\n", (void *)dip);
+		return (DDI_FAILURE);
+	}
+
+	ddi_remove_minor_node(dip, NULL);
+	ddi_soft_state_free(zfd_soft_state, instance);
+
+	return (DDI_SUCCESS);
+}
+
+/*
+ * zfd_getinfo()
+ *	getinfo(9e) entrypoint.
+ */
+/*ARGSUSED*/
+static int
+zfd_getinfo(dev_info_t *dip, ddi_info_cmd_t infocmd, void *arg, void **result)
+{
+	zfd_state_t *zfds;
+	int instance = ZFD_INSTANCE((dev_t)arg);
+
+	switch (infocmd) {
+	case DDI_INFO_DEVT2DEVINFO:
+		if ((zfds = ddi_get_soft_state(zfd_soft_state,
+		    instance)) == NULL)
+			return (DDI_FAILURE);
+		*result = zfds->zfd_devinfo;
+		return (DDI_SUCCESS);
+	case DDI_INFO_DEVT2INSTANCE:
+		*result = (void *)(uintptr_t)instance;
+		return (DDI_SUCCESS);
+	}
+	return (DDI_FAILURE);
+}
+
+/*
+ * Return the equivalent queue from the other side of the relationship.
+ * e.g.: given the slave's write queue, return the master's write queue.
+ */
+static queue_t *
+zfd_switch(queue_t *qp)
+{
+	zfd_state_t *zfds = qp->q_ptr;
+	ASSERT(zfds != NULL);
+
+	if (qp == zfds->zfd_master_rdq)
+		return (zfds->zfd_slave_rdq);
+	else if (OTHERQ(qp) == zfds->zfd_master_rdq && zfds->zfd_slave_rdq
+	    != NULL)
+		return (OTHERQ(zfds->zfd_slave_rdq));
+	else if (qp == zfds->zfd_slave_rdq)
+		return (zfds->zfd_master_rdq);
+	else if (OTHERQ(qp) == zfds->zfd_slave_rdq && zfds->zfd_master_rdq
+	    != NULL)
+		return (OTHERQ(zfds->zfd_master_rdq));
+	else
+		return (NULL);
+}
+
+/*
+ * For debugging and outputting messages.  Returns the name of the side of
+ * the relationship associated with this queue.
+ */
+static const char *
+zfd_side(queue_t *qp)
+{
+	zfd_state_t *zfds = qp->q_ptr;
+	ASSERT(zfds != NULL);
+
+	if (qp == zfds->zfd_master_rdq ||
+	    OTHERQ(qp) == zfds->zfd_master_rdq) {
+		return ("master");
+	}
+	ASSERT(qp == zfds->zfd_slave_rdq || OTHERQ(qp) == zfds->zfd_slave_rdq);
+	return ("slave");
+}
+
+/*ARGSUSED*/
+static int
+zfd_master_open(zfd_state_t *zfds,
+    queue_t	*rqp,	/* pointer to the read side queue */
+    dev_t	*devp,	/* pointer to stream tail's dev */
+    int		oflag,	/* the user open(2) supplied flags */
+    int		sflag,	/* open state flag */
+    cred_t	*credp)	/* credentials */
+{
+	mblk_t *mop;
+	struct stroptions *sop;
+
+	/*
+	 * Enforce exclusivity on the master side; the only consumer should
+	 * be the zoneadmd for the zone.
+	 */
+	if ((zfds->zfd_state & ZFD_STATE_MOPEN) != 0)
+		return (EBUSY);
+
+	if ((mop = allocb(sizeof (struct stroptions), BPRI_MED)) == NULL) {
+		DBG("zfd_master_open(): mop allocation failed\n");
+		return (ENOMEM);
+	}
+
+	zfds->zfd_state |= ZFD_STATE_MOPEN;
+
+	/*
+	 * q_ptr stores driver private data; stash the soft state data on both
+	 * read and write sides of the queue.
+	 */
+	WR(rqp)->q_ptr = rqp->q_ptr = zfds;
+	qprocson(rqp);
+
+	/*
+	 * Following qprocson(), the master side is fully plumbed into the
+	 * STREAM and may send/receive messages.  Setting zfds->zfd_master_rdq
+	 * will allow the slave to send messages to us (the master).
+	 * This cannot occur before qprocson() because the master is not
+	 * ready to process them until that point.
+	 */
+	zfds->zfd_master_rdq = rqp;
+
+	/*
+	 * set up hi/lo water marks on stream head read queue and add
+	 * controlling tty as needed.
+	 */
+	mop->b_datap->db_type = M_SETOPTS;
+	mop->b_wptr += sizeof (struct stroptions);
+	sop = (struct stroptions *)(void *)mop->b_rptr;
+	if (oflag & FNOCTTY)
+		sop->so_flags = SO_HIWAT | SO_LOWAT;
+	else
+		sop->so_flags = SO_HIWAT | SO_LOWAT | SO_ISTTY;
+	sop->so_hiwat = 512;
+	sop->so_lowat = 256;
+	putnext(rqp, mop);
+
+	return (0);
+}
+
+/*ARGSUSED*/
+static int
+zfd_slave_open(zfd_state_t *zfds,
+    queue_t	*rqp,	/* pointer to the read side queue */
+    dev_t	*devp,	/* pointer to stream tail's dev */
+    int		oflag,	/* the user open(2) supplied flags */
+    int		sflag,	/* open state flag */
+    cred_t	*credp)	/* credentials */
+{
+	mblk_t *mop;
+	struct stroptions *sop;
+	/*
+	 * The slave side can be opened as many times as needed.
+	 */
+	if ((zfds->zfd_state & ZFD_STATE_SOPEN) != 0) {
+		ASSERT((rqp != NULL) && (WR(rqp)->q_ptr == zfds));
+		return (0);
+	}
+
+	if (zfds->zfd_tty == 1) {
+		major_t major;
+		minor_t minor;
+		minor_t lastminor;
+		uint_t anchorindex;
+
+		/*
+		 * Set up sad(7D) so that the necessary STREAMS modules will
+		 * be in place.  A wrinkle is that 'ptem' must be anchored
+		 * in place (see streamio(7i)) because we always want the
+		 * fd to have terminal semantics.
+		 */
+		minor =
+		    ddi_get_instance(zfds->zfd_devinfo) << 1 | ZFD_SLAVE_MINOR;
+		major = ddi_driver_major(zfds->zfd_devinfo);
+		lastminor = 0;
+		anchorindex = 1;
+		if (kstr_autopush(SET_AUTOPUSH, &major, &minor, &lastminor,
+		    &anchorindex, zfd_mods) != 0) {
+			DBG("zfd_slave_open(): kstr_autopush() failed\n");
+			return (EIO);
+		}
+	}
+
+	if ((mop = allocb(sizeof (struct stroptions), BPRI_MED)) == NULL) {
+		DBG("zfd_slave_open(): mop allocation failed\n");
+		return (ENOMEM);
+	}
+
+	zfds->zfd_state |= ZFD_STATE_SOPEN;
+
+	/*
+	 * q_ptr stores driver private data; stash the soft state data on both
+	 * read and write sides of the queue.
+	 */
+	WR(rqp)->q_ptr = rqp->q_ptr = zfds;
+
+	qprocson(rqp);
+
+	/*
+	 * Must follow qprocson(), since we aren't ready to process until then.
+	 */
+	zfds->zfd_slave_rdq = rqp;
+
+	/*
+	 * set up hi/lo water marks on stream head read queue and add
+	 * controlling tty as needed.
+	 */
+	mop->b_datap->db_type = M_SETOPTS;
+	mop->b_wptr += sizeof (struct stroptions);
+	sop = (struct stroptions *)(void *)mop->b_rptr;
+	sop->so_flags = SO_HIWAT | SO_LOWAT | SO_ISTTY;
+	sop->so_hiwat = 512;
+	sop->so_lowat = 256;
+	putnext(rqp, mop);
+
+	return (0);
+}
+
+/*
+ * open(9e) entrypoint; checks sflag, and rejects anything unordinary.
+ */
+static int
+zfd_open(queue_t *rqp,		/* pointer to the read side queue */
+	dev_t   *devp,		/* pointer to stream tail's dev */
+	int	oflag,		/* the user open(2) supplied flags */
+	int	sflag,		/* open state flag */
+	cred_t  *credp)		/* credentials */
+{
+	int instance = ZFD_INSTANCE(*devp);
+	int ret;
+	zfd_state_t *zfds;
+
+	if (sflag != 0)
+		return (EINVAL);
+
+	if ((zfds = ddi_get_soft_state(zfd_soft_state, instance)) == NULL)
+		return (ENXIO);
+
+	switch (ZFD_NODE(*devp)) {
+	case ZFD_MASTER_MINOR:
+		ret = zfd_master_open(zfds, rqp, devp, oflag, sflag, credp);
+		break;
+	case ZFD_SLAVE_MINOR:
+		ret = zfd_slave_open(zfds, rqp, devp, oflag, sflag, credp);
+		break;
+	default:
+		ret = ENXIO;
+		break;
+	}
+
+	return (ret);
+}
+
+/*
+ * close(9e) entrypoint.
+ */
+/*ARGSUSED1*/
+static int
+zfd_close(queue_t *rqp, int flag, cred_t *credp)
+{
+	queue_t *wqp;
+	mblk_t	*bp;
+	zfd_state_t *zfds;
+	major_t major;
+	minor_t minor;
+
+	zfds = (zfd_state_t *)rqp->q_ptr;
+
+	if (rqp == zfds->zfd_master_rdq) {
+		DBG("Closing master side");
+
+		zfds->zfd_master_rdq = NULL;
+		zfds->zfd_state &= ~ZFD_STATE_MOPEN;
+
+		/*
+		 * qenable slave side write queue so that it can flush
+		 * its messages as master's read queue is going away
+		 */
+		if (zfds->zfd_slave_rdq != NULL) {
+			qenable(WR(zfds->zfd_slave_rdq));
+		}
+
+		qprocsoff(rqp);
+		WR(rqp)->q_ptr = rqp->q_ptr = NULL;
+
+	} else if (rqp == zfds->zfd_slave_rdq) {
+
+		DBG("Closing slave side");
+		zfds->zfd_state &= ~ZFD_STATE_SOPEN;
+		zfds->zfd_slave_rdq = NULL;
+
+		wqp = WR(rqp);
+		while ((bp = getq(wqp)) != NULL) {
+			if (zfds->zfd_master_rdq != NULL)
+				putnext(zfds->zfd_master_rdq, bp);
+			else if (bp->b_datap->db_type == M_IOCTL)
+				miocnak(wqp, bp, 0, 0);
+			else
+				freemsg(bp);
+		}
+
+		/*
+		 * Qenable master side write queue so that it can flush its
+		 * messages as slaves's read queue is going away.
+		 */
+		if (zfds->zfd_master_rdq != NULL)
+			qenable(WR(zfds->zfd_master_rdq));
+
+		qprocsoff(rqp);
+		WR(rqp)->q_ptr = rqp->q_ptr = NULL;
+
+		if (zfds->zfd_tty == 1) {
+			/*
+			 * Clear the sad configuration so that reopening
+			 * doesn't fail to set up sad configuration.
+			 */
+			major = ddi_driver_major(zfds->zfd_devinfo);
+			minor = ddi_get_instance(zfds->zfd_devinfo) << 1 |
+			    ZFD_SLAVE_MINOR;
+			(void) kstr_autopush(CLR_AUTOPUSH, &major, &minor,
+			    NULL, NULL, NULL);
+		}
+	}
+
+	return (0);
+}
+
+static void
+handle_mflush(queue_t *qp, mblk_t *mp)
+{
+	mblk_t *nmp;
+	DBG1("M_FLUSH on %s side", zfd_side(qp));
+
+	if (*mp->b_rptr & FLUSHW) {
+		DBG1("M_FLUSH, FLUSHW, %s side", zfd_side(qp));
+		flushq(qp, FLUSHDATA);
+		*mp->b_rptr &= ~FLUSHW;
+		if ((*mp->b_rptr & FLUSHR) == 0) {
+			/*
+			 * FLUSHW only. Change to FLUSHR and putnext other side,
+			 * then we are done.
+			 */
+			*mp->b_rptr |= FLUSHR;
+			if (zfd_switch(RD(qp)) != NULL) {
+				putnext(zfd_switch(RD(qp)), mp);
+				return;
+			}
+		} else if ((zfd_switch(RD(qp)) != NULL) &&
+		    (nmp = copyb(mp)) != NULL) {
+			/*
+			 * It is a FLUSHRW; we copy the mblk and send
+			 * it to the other side, since we still need to use
+			 * the mblk in FLUSHR processing, below.
+			 */
+			putnext(zfd_switch(RD(qp)), nmp);
+		}
+	}
+
+	if (*mp->b_rptr & FLUSHR) {
+		DBG("qreply(qp) turning FLUSHR around\n");
+		qreply(qp, mp);
+		return;
+	}
+	freemsg(mp);
+}
+
+/*
+ * wput(9E) is symmetric for master and slave sides, so this handles both
+ * without splitting the codepath.  (The only exception to this is the
+ * processing of zfd ioctls, which is restricted to the master side.)
+ *
+ * zfd_wput() looks at the other side; if there is no process holding that
+ * side open, it frees the message.  This prevents processes from hanging
+ * if no one is holding open the fd.  Otherwise, it putnext's high
+ * priority messages, putnext's normal messages if possible, and otherwise
+ * enqueues the messages; in the case that something is enqueued, wsrv(9E)
+ * will take care of eventually shuttling I/O to the other side.
+ */
+static void
+zfd_wput(queue_t *qp, mblk_t *mp)
+{
+	unsigned char type = mp->b_datap->db_type;
+	zfd_state_t *zfds;
+	struct iocblk *iocbp;
+
+	ASSERT(qp->q_ptr);
+
+	DBG1("entering zfd_wput, %s side", zfd_side(qp));
+
+	/*
+	 * Process zfd ioctl messages if qp is the master side's write queue.
+	 */
+	zfds = (zfd_state_t *)qp->q_ptr;
+	if (zfds->zfd_master_rdq != NULL && qp == WR(zfds->zfd_master_rdq) &&
+	    type == M_IOCTL) {
+		iocbp = (struct iocblk *)(void *)mp->b_rptr;
+		switch (iocbp->ioc_cmd) {
+		case ZFD_MAKETTY:
+			/*
+			 * The process that passed the ioctl must be running in
+			 * the global zone.
+			 */
+			if (crgetzoneid(iocbp->ioc_cr) != GLOBAL_ZONEID) {
+				miocack(qp, mp, 0, EINVAL);
+				return;
+			}
+			zfds->zfd_tty = 1;
+			miocack(qp, mp, 0, 0);
+			return;
+		case ZFD_EOF:
+			/*
+			 * The process that passed the ioctl must be running in
+			 * the global zone.
+			 */
+			if (crgetzoneid(iocbp->ioc_cr) != GLOBAL_ZONEID) {
+				miocack(qp, mp, 0, EINVAL);
+				return;
+			}
+			if (zfds->zfd_slave_rdq != NULL)
+				(void) putnextctl(zfds->zfd_slave_rdq,
+				    M_HANGUP);
+			miocack(qp, mp, 0, 0);
+			return;
+		case ZFD_HAS_SLAVE:
+			/*
+			 * The process that passed the ioctl must be running in
+			 * the global zone.
+			 */
+			if (crgetzoneid(iocbp->ioc_cr) != GLOBAL_ZONEID) {
+				miocack(qp, mp, 0, EINVAL);
+				return;
+			}
+			if ((zfds->zfd_state & ZFD_STATE_SOPEN) != 0) {
+				miocack(qp, mp, 0, 0);
+			} else {
+				miocack(qp, mp, 0, ENOTTY);
+			}
+			return;
+		default:
+			break;
+		}
+	}
+
+	if (zfd_switch(RD(qp)) == NULL) {
+		DBG1("wput to %s side (no one listening)", zfd_side(qp));
+		switch (type) {
+		case M_FLUSH:
+			handle_mflush(qp, mp);
+			break;
+		case M_IOCTL:
+			miocnak(qp, mp, 0, 0);
+			break;
+		default:
+			freemsg(mp);
+			break;
+		}
+		return;
+	}
+
+	if (type >= QPCTL) {
+		DBG1("(hipri) wput, %s side", zfd_side(qp));
+		switch (type) {
+		case M_READ:		/* supposedly from ldterm? */
+			DBG("zfd_wput: tossing M_READ\n");
+			freemsg(mp);
+			break;
+		case M_FLUSH:
+			handle_mflush(qp, mp);
+			break;
+		default:
+			/*
+			 * Put this to the other side.
+			 */
+			ASSERT(zfd_switch(RD(qp)) != NULL);
+			putnext(zfd_switch(RD(qp)), mp);
+			break;
+		}
+		DBG1("done (hipri) wput, %s side", zfd_side(qp));
+		return;
+	}
+
+	/*
+	 * Only putnext if there isn't already something in the queue.
+	 * otherwise things would wind up out of order.
+	 */
+	if (qp->q_first == NULL &&
+	    bcanputnext(RD(zfd_switch(qp)), mp->b_band)) {
+		DBG("wput: putting message to other side\n");
+		putnext(RD(zfd_switch(qp)), mp);
+	} else {
+		DBG("wput: putting msg onto queue\n");
+		(void) putq(qp, mp);
+	}
+	DBG1("done wput, %s side", zfd_side(qp));
+}
+
+/*
+ * rsrv(9E) is symmetric for master and slave, so zfd_rsrv() handles both
+ * without splitting up the codepath.
+ *
+ * Enable the write side of the partner.  This triggers the partner to send
+ * messages queued on its write side to this queue's read side.
+ */
+static void
+zfd_rsrv(queue_t *qp)
+{
+	zfd_state_t *zfds;
+	zfds = (zfd_state_t *)qp->q_ptr;
+
+	/*
+	 * Care must be taken here, as either of the master or slave side
+	 * qptr could be NULL.
+	 */
+	ASSERT(qp == zfds->zfd_master_rdq || qp == zfds->zfd_slave_rdq);
+	if (zfd_switch(qp) == NULL) {
+		DBG("zfd_rsrv: other side isn't listening\n");
+		return;
+	}
+	qenable(WR(zfd_switch(qp)));
+}
+
+/*
+ * This routine is symmetric for master and slave, so it handles both without
+ * splitting up the codepath.
+ *
+ * If there are messages on this queue that can be sent to the other, send
+ * them via putnext(). Else, if queued messages cannot be sent, leave them
+ * on this queue.
+ */
+static void
+zfd_wsrv(queue_t *qp)
+{
+	mblk_t *mp;
+
+	DBG1("zfd_wsrv master (%s) side", zfd_side(qp));
+
+	/*
+	 * Partner has no read queue, so take the data, and throw it away.
+	 */
+	if (zfd_switch(RD(qp)) == NULL) {
+		DBG("zfd_wsrv: other side isn't listening");
+		while ((mp = getq(qp)) != NULL) {
+			if (mp->b_datap->db_type == M_IOCTL)
+				miocnak(qp, mp, 0, 0);
+			else
+				freemsg(mp);
+		}
+		flushq(qp, FLUSHALL);
+		return;
+	}
+
+	/*
+	 * while there are messages on this write queue...
+	 */
+	while ((mp = getq(qp)) != NULL) {
+		/*
+		 * Due to the way zfd_wput is implemented, we should never
+		 * see a control message here.
+		 */
+		ASSERT(mp->b_datap->db_type < QPCTL);
+
+		if (bcanputnext(RD(zfd_switch(qp)), mp->b_band)) {
+			DBG("wsrv: send message to other side\n");
+			putnext(RD(zfd_switch(qp)), mp);
+		} else {
+			DBG("wsrv: putting msg back on queue\n");
+			(void) putbq(qp, mp);
+			break;
+		}
+	}
+}
diff --git a/usr/src/uts/common/krtld/bootrd.c b/usr/src/uts/common/krtld/bootrd.c
index 08e5d98c09..35ad67da96 100644
--- a/usr/src/uts/common/krtld/bootrd.c
+++ b/usr/src/uts/common/krtld/bootrd.c
@@ -21,6 +21,7 @@
 /*
  * Copyright 2008 Sun Microsystems, Inc.  All rights reserved.
  * Use is subject to license terms.
+ * Copyright 2013 Joyent, Inc.  All rights reserved.
  */
 
 
@@ -36,17 +37,95 @@
 extern void (*_kobj_printf)(void *, const char *fmt, ...);
 extern int get_weakish_int(int *);
 extern struct bootops *ops;
-extern struct boot_fs_ops bufs_ops, bhsfs_ops;
+extern struct boot_fs_ops bufs_ops, bhsfs_ops, bbootfs_ops;
 extern int kmem_ready;
 
 static uint64_t rd_start, rd_end;
 struct boot_fs_ops *bfs_ops;
-struct boot_fs_ops *bfs_tab[] = {&bufs_ops, &bhsfs_ops, NULL};
+struct boot_fs_ops *bfs_tab[] = {&bufs_ops, &bhsfs_ops, &bbootfs_ops, NULL};
 
 static uintptr_t scratch_max = 0;
 
 #define	_kmem_ready	get_weakish_int(&kmem_ready)
 
+int
+BRD_MOUNTROOT(struct boot_fs_ops *ops, char *str)
+{
+	return (ops->fsw_mountroot(str));
+}
+
+int
+BRD_UNMOUNTROOT(struct boot_fs_ops *ops)
+{
+	if (bfs_ops != &bbootfs_ops)
+		bbootfs_ops.fsw_closeall(1);
+
+	return (ops->fsw_unmountroot());
+}
+
+int
+BRD_OPEN(struct boot_fs_ops *ops, char *file, int flags)
+{
+	int len = strlen(SYSTEM_BOOT_PATH);
+	int fd;
+
+	/*
+	 * Our policy is that we try bootfs first.  If bootfs is the only
+	 * filesystem, that's the end of it.  Otherwise we will fall back to
+	 * the normal root (i.e., ramdisk) filesystem at this point and try
+	 * again if the file does not exist in bootfs.
+	 */
+	fd = bbootfs_ops.fsw_open(file, flags);
+
+	if (bfs_ops == &bbootfs_ops)
+		return (fd);
+
+	if (strncmp(file, SYSTEM_BOOT_PATH, len) == 0 || fd >= 0)
+		return ((fd < 0) ? fd : (fd | BFD_F_SYSTEM_BOOT));
+
+	return (ops->fsw_open(file, flags));
+}
+
+int
+BRD_CLOSE(struct boot_fs_ops *ops, int fd)
+{
+	if (fd & BFD_F_SYSTEM_BOOT)
+		return (bbootfs_ops.fsw_close(fd & ~BFD_F_SYSTEM_BOOT));
+
+	return (ops->fsw_close(fd));
+}
+
+ssize_t
+BRD_READ(struct boot_fs_ops *ops, int fd, caddr_t buf, size_t len)
+{
+	if (fd & BFD_F_SYSTEM_BOOT) {
+		return (bbootfs_ops.fsw_read(fd & ~BFD_F_SYSTEM_BOOT,
+		    buf, len));
+	}
+
+	return (ops->fsw_read(fd, buf, len));
+}
+
+off_t
+BRD_SEEK(struct boot_fs_ops *ops, int fd, off_t addr, int whence)
+{
+	if (fd & BFD_F_SYSTEM_BOOT) {
+		return (bbootfs_ops.fsw_lseek(fd & ~BFD_F_SYSTEM_BOOT,
+		    addr, whence));
+	}
+
+	return (ops->fsw_lseek(fd, addr, whence));
+}
+
+int
+BRD_FSTAT(struct boot_fs_ops *ops, int fd, struct bootstat *bsp)
+{
+	if (fd & BFD_F_SYSTEM_BOOT)
+		return (bbootfs_ops.fsw_fstat(fd & ~BFD_F_SYSTEM_BOOT, bsp));
+
+	return (ops->fsw_fstat(fd, bsp));
+}
+
 /*
  * This one reads the ramdisk. If fi_memp is set, we copy the
  * ramdisk content to the designated buffer. Otherwise, we
diff --git a/usr/src/uts/common/netinet/in.h b/usr/src/uts/common/netinet/in.h
index d530b7f36e..7927cf5e24 100644
--- a/usr/src/uts/common/netinet/in.h
+++ b/usr/src/uts/common/netinet/in.h
@@ -3,6 +3,7 @@
  * Use is subject to license terms.
  *
  * Copyright 2011 Nexenta Systems, Inc. All rights reserved.
+ * Copyright 2015, Joyent, Inc.
  */
 /*
  * Copyright (c) 1982, 1986 Regents of the University of California.
@@ -225,6 +226,7 @@ typedef	uint16_t	sa_family_t;
 #define	IPPORT_SLP		427
 #define	IPPORT_MIP		434
 #define	IPPORT_SMB		445		/* a.k.a. microsoft-ds */
+#define	IPPORT_VXLAN		4789
 
 /*
  * Internet Key Exchange (IKE) ports
@@ -268,6 +270,11 @@ typedef	uint16_t	sa_family_t;
 #define	IPPORT_RESERVED		1024
 #define	IPPORT_USERRESERVED	5000
 
+#ifdef _KERNEL
+#define	IPPORT_DYNAMIC_MIN	49152
+#define	IPPORT_DYNAMIC_MAX	65535
+#endif
+
 /*
  * Link numbers
  */
diff --git a/usr/src/uts/common/netinet/udp.h b/usr/src/uts/common/netinet/udp.h
index c65a9bad3a..fb9f8a0976 100644
--- a/usr/src/uts/common/netinet/udp.h
+++ b/usr/src/uts/common/netinet/udp.h
@@ -17,9 +17,6 @@
 #ifndef	_NETINET_UDP_H
 #define	_NETINET_UDP_H
 
-#pragma ident	"%Z%%M%	%I%	%E% SMI"
-/* udp.h 1.7 88/08/19 SMI; from UCB 7.1 6/5/86	*/
-
 #ifdef	__cplusplus
 extern "C" {
 #endif
@@ -36,6 +33,15 @@ struct udphdr {
 #define	UDP_EXCLBIND		0x0101		/* for internal use only */
 #define	UDP_RCVHDR		0x0102		/* for internal use only */
 #define	UDP_NAT_T_ENDPOINT	0x0103		/* for internal use only */
+#define	UDP_SRCPORT_HASH	0x0104		/* for internal use only */
+
+/*
+ * Hash definitions for UDP_SRCPORT_HASH that effectively tell UDP how to go
+ * handle UDP_SRCPORT_HASH.
+ */
+#define	UDP_HASH_DISABLE	0x0000		/* for internal use only */
+#define	UDP_HASH_VXLAN		0x0001		/* for internal use only */
+
 /*
  * Following option in UDP_ namespace required to be exposed through
  * <xti.h> (It also requires exposing options not implemented). The options
diff --git a/usr/src/uts/common/os/bio.c b/usr/src/uts/common/os/bio.c
index 96502b8230..b8d2e29058 100644
--- a/usr/src/uts/common/os/bio.c
+++ b/usr/src/uts/common/os/bio.c
@@ -21,6 +21,7 @@
 /*
  * Copyright 2009 Sun Microsystems, Inc.  All rights reserved.
  * Use is subject to license terms.
+ * Copyright 2011 Joyent, Inc.  All rights reserved.
  */
 
 /*	Copyright (c) 1984, 1986, 1987, 1988, 1989 AT&T	*/
@@ -1320,6 +1321,9 @@ pageio_setup(struct page *pp, size_t len, struct vnode *vp, int flags)
 		cpup = CPU;	/* get pointer AFTER preemption is disabled */
 		CPU_STATS_ADDQ(cpup, vm, pgin, 1);
 		CPU_STATS_ADDQ(cpup, vm, pgpgin, btopr(len));
+
+		atomic_add_64(&curzone->zone_pgpgin, btopr(len));
+
 		if ((flags & B_ASYNC) == 0) {
 			klwp_t *lwp = ttolwp(curthread);
 			if (lwp != NULL)
@@ -1336,13 +1340,19 @@ pageio_setup(struct page *pp, size_t len, struct vnode *vp, int flags)
 		if (pp != NULL && pp->p_vnode != NULL) {
 			if (IS_SWAPFSVP(pp->p_vnode)) {
 				CPU_STATS_ADDQ(cpup, vm, anonpgin, btopr(len));
+				atomic_add_64(&curzone->zone_anonpgin,
+				    btopr(len));
 			} else {
 				if (pp->p_vnode->v_flag & VVMEXEC) {
 					CPU_STATS_ADDQ(cpup, vm, execpgin,
 					    btopr(len));
+					atomic_add_64(&curzone->zone_execpgin,
+					    btopr(len));
 				} else {
 					CPU_STATS_ADDQ(cpup, vm, fspgin,
 					    btopr(len));
+					atomic_add_64(&curzone->zone_fspgin,
+					    btopr(len));
 				}
 			}
 		}
diff --git a/usr/src/uts/common/os/brand.c b/usr/src/uts/common/os/brand.c
index eb8c6e730a..00f2ce0440 100644
--- a/usr/src/uts/common/os/brand.c
+++ b/usr/src/uts/common/os/brand.c
@@ -20,6 +20,7 @@
  */
 /*
  * Copyright (c) 2006, 2010, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2015, Joyent, Inc. All rights reserved.
  */
 
 #include <sys/kmem.h>
@@ -45,7 +46,7 @@ struct brand_mach_ops native_mach_ops  = {
 };
 #else /* !__sparcv9 */
 struct brand_mach_ops native_mach_ops  = {
-		NULL, NULL, NULL, NULL
+		NULL, NULL, NULL, NULL, NULL, NULL, NULL
 };
 #endif /* !__sparcv9 */
 
@@ -53,7 +54,8 @@ brand_t native_brand = {
 		BRAND_VER_1,
 		"native",
 		NULL,
-		&native_mach_ops
+		&native_mach_ops,
+		0
 };
 
 /*
@@ -314,42 +316,55 @@ void
 brand_setbrand(proc_t *p)
 {
 	brand_t *bp = p->p_zone->zone_brand;
+	void *brand_data = NULL;
 
-	ASSERT(bp != NULL);
-	ASSERT(p->p_brand == &native_brand);
+	VERIFY(MUTEX_NOT_HELD(&p->p_lock));
+	VERIFY(bp != NULL);
 
 	/*
-	 * We should only be called from exec(), when we know the process
-	 * is single-threaded.
+	 * We should only be called from exec() or getproc(), when we know the
+	 * process has 0 or 1 threads.
 	 */
-	ASSERT(p->p_tlist == p->p_tlist->t_forw);
+	VERIFY((p->p_tlist == NULL) || (p->p_tlist == p->p_tlist->t_forw));
 
+	if (bp->b_data_size > 0) {
+		brand_data = kmem_zalloc(bp->b_data_size, KM_SLEEP);
+	}
+
+	mutex_enter(&p->p_lock);
+	ASSERT(!PROC_IS_BRANDED(p));
 	p->p_brand = bp;
+	p->p_brand_data = brand_data;
 	ASSERT(PROC_IS_BRANDED(p));
 	BROP(p)->b_setbrand(p);
+	mutex_exit(&p->p_lock);
 }
 
 void
-brand_clearbrand(proc_t *p, boolean_t no_lwps)
+brand_clearbrand(proc_t *p)
 {
 	brand_t *bp = p->p_zone->zone_brand;
-	klwp_t *lwp = NULL;
-	ASSERT(bp != NULL);
-	ASSERT(!no_lwps || (p->p_tlist == NULL));
+	void *brand_data;
+
+	VERIFY(MUTEX_NOT_HELD(&p->p_lock));
+	VERIFY(bp != NULL);
+	VERIFY(PROC_IS_BRANDED(p));
 
 	/*
-	 * If called from exec_common() or proc_exit(),
-	 * we know the process is single-threaded.
-	 * If called from fork_fail, p_tlist is NULL.
+	 * There cannot be more than one lwp associated with a process when
+	 * stripping the brand.
 	 */
-	if (!no_lwps) {
-		ASSERT(p->p_tlist == p->p_tlist->t_forw);
-		lwp = p->p_tlist->t_lwp;
-	}
+	VERIFY((p->p_tlist == NULL) || (p->p_tlist == p->p_tlist->t_forw));
 
-	ASSERT(PROC_IS_BRANDED(p));
-	BROP(p)->b_proc_exit(p, lwp);
+	mutex_enter(&p->p_lock);
 	p->p_brand = &native_brand;
+	brand_data = p->p_brand_data;
+	p->p_brand_data = NULL;
+	mutex_exit(&p->p_lock);
+
+	if (brand_data != NULL) {
+		kmem_free(brand_data, bp->b_data_size);
+	}
 }
 
 #if defined(__sparcv9)
@@ -483,7 +498,7 @@ brand_solaris_cmd(int cmd, uintptr_t arg1, uintptr_t arg2, uintptr_t arg3,
 		return (ENOSYS);
 
 	/* For all other operations this must be a branded process. */
-	if (p->p_brand == &native_brand)
+	if (!PROC_IS_BRANDED(p))
 		return (ENOSYS);
 
 	ASSERT(p->p_brand == pbrand);
@@ -601,15 +616,15 @@ restoreexecenv(struct execenv *ep, stack_t *sp)
 int
 brand_solaris_elfexec(vnode_t *vp, execa_t *uap, uarg_t *args,
     intpdata_t *idatap, int level, long *execsz, int setid, caddr_t exec_file,
-    cred_t *cred, int brand_action, struct brand *pbrand, char *bname,
-    char *brandlib, char *brandlib32, char *brandlinker, char *brandlinker32)
+    cred_t *cred, int *brand_action, struct brand *pbrand, char *bname,
+    char *brandlib, char *brandlib32)
 {
 
 	vnode_t		*nvp;
 	Ehdr		ehdr;
 	Addr		uphdr_vaddr;
 	intptr_t	voffset;
-	int		interp;
+	char		*interp;
 	int		i, err;
 	struct execenv	env;
 	struct execenv	origenv;
@@ -619,7 +634,6 @@ brand_solaris_elfexec(vnode_t *vp, execa_t *uap, uarg_t *args,
 	klwp_t		*lwp = ttolwp(curthread);
 	brand_proc_data_t	*spd;
 	brand_elf_data_t sed, *sedp;
-	char		*linker;
 	uintptr_t	lddata; /* lddata of executable's linker */
 
 	ASSERT(curproc->p_brand == pbrand);
@@ -636,12 +650,10 @@ brand_solaris_elfexec(vnode_t *vp, execa_t *uap, uarg_t *args,
 	 */
 	if (args->to_model == DATAMODEL_NATIVE) {
 		args->emulator = brandlib;
-		linker = brandlinker;
 	}
 #if defined(_LP64)
 	else {
 		args->emulator = brandlib32;
-		linker = brandlinker32;
 	}
 #endif  /* _LP64 */
 
@@ -725,7 +737,7 @@ brand_solaris_elfexec(vnode_t *vp, execa_t *uap, uarg_t *args,
 	if (args->to_model == DATAMODEL_NATIVE) {
 		err = mapexec_brand(vp, args, &ehdr, &uphdr_vaddr,
 		    &voffset, exec_file, &interp, &env.ex_bssbase,
-		    &env.ex_brkbase, &env.ex_brksize, NULL);
+		    &env.ex_brkbase, &env.ex_brksize, NULL, NULL);
 	}
 #if defined(_LP64)
 	else {
@@ -733,7 +745,7 @@ brand_solaris_elfexec(vnode_t *vp, execa_t *uap, uarg_t *args,
 		Elf32_Addr uphdr_vaddr32;
 		err = mapexec32_brand(vp, args, &ehdr32, &uphdr_vaddr32,
 		    &voffset, exec_file, &interp, &env.ex_bssbase,
-		    &env.ex_brkbase, &env.ex_brksize, NULL);
+		    &env.ex_brkbase, &env.ex_brksize, NULL, NULL);
 		Ehdr32to64(&ehdr32, &ehdr);
 
 		if (uphdr_vaddr32 == (Elf32_Addr)-1)
@@ -744,6 +756,10 @@ brand_solaris_elfexec(vnode_t *vp, execa_t *uap, uarg_t *args,
 #endif  /* _LP64 */
 	if (err != 0) {
 		restoreexecenv(&origenv, &orig_sigaltstack);
+
+		if (interp != NULL)
+			kmem_free(interp, MAXPATHLEN);
+
 		return (err);
 	}
 
@@ -761,7 +777,7 @@ brand_solaris_elfexec(vnode_t *vp, execa_t *uap, uarg_t *args,
 	sedp->sed_phent = ehdr.e_phentsize;
 	sedp->sed_phnum = ehdr.e_phnum;
 
-	if (interp) {
+	if (interp != NULL) {
 		if (ehdr.e_type == ET_DYN) {
 			/*
 			 * This is a shared object executable, so we
@@ -777,16 +793,20 @@ brand_solaris_elfexec(vnode_t *vp, execa_t *uap, uarg_t *args,
 		 * it in and store relevant information about it in the
 		 * aux vector, where the brand library can find it.
 		 */
-		if ((err = lookupname(linker, UIO_SYSSPACE,
+		if ((err = lookupname(interp, UIO_SYSSPACE,
 		    FOLLOW, NULLVPP, &nvp)) != 0) {
-			uprintf("%s: not found.", brandlinker);
+			uprintf("%s: not found.", interp);
 			restoreexecenv(&origenv, &orig_sigaltstack);
+			kmem_free(interp, MAXPATHLEN);
 			return (err);
 		}
+
+		kmem_free(interp, MAXPATHLEN);
+
 		if (args->to_model == DATAMODEL_NATIVE) {
 			err = mapexec_brand(nvp, args, &ehdr,
 			    &uphdr_vaddr, &voffset, exec_file, &interp,
-			    NULL, NULL, NULL, &lddata);
+			    NULL, NULL, NULL, &lddata, NULL);
 		}
 #if defined(_LP64)
 		else {
@@ -794,7 +814,7 @@ brand_solaris_elfexec(vnode_t *vp, execa_t *uap, uarg_t *args,
 			Elf32_Addr uphdr_vaddr32;
 			err = mapexec32_brand(nvp, args, &ehdr32,
 			    &uphdr_vaddr32, &voffset, exec_file, &interp,
-			    NULL, NULL, NULL, &lddata);
+			    NULL, NULL, NULL, &lddata, NULL);
 			Ehdr32to64(&ehdr32, &ehdr);
 
 			if (uphdr_vaddr32 == (Elf32_Addr)-1)
@@ -934,9 +954,9 @@ brand_solaris_elfexec(vnode_t *vp, execa_t *uap, uarg_t *args,
 
 	/*
 	 * Third, the /proc aux vectors set up by elfexec() point to
-	 * brand emulation library and it's linker.  Copy these to the
+	 * brand emulation library and its linker.  Copy these to the
 	 * /proc brand specific aux vector, and update the regular
-	 * /proc aux vectors to point to the executable (and it's
+	 * /proc aux vectors to point to the executable (and its
 	 * linker).  This will enable debuggers to access the
 	 * executable via the usual /proc or elf notes aux vectors.
 	 *
@@ -1078,55 +1098,31 @@ brand_solaris_freelwp(klwp_t *l, struct brand *pbrand)
 }
 
 /*ARGSUSED*/
-int
+void
 brand_solaris_initlwp(klwp_t *l, struct brand *pbrand)
 {
 	ASSERT(l->lwp_procp->p_brand == pbrand);
 	ASSERT(l->lwp_procp->p_brand_data != NULL);
 	ASSERT(l->lwp_brand == NULL);
 	l->lwp_brand = (void *)-1;
-	return (0);
 }
 
 /*ARGSUSED*/
 void
 brand_solaris_lwpexit(klwp_t *l, struct brand *pbrand)
 {
-	proc_t  *p = l->lwp_procp;
-
 	ASSERT(l->lwp_procp->p_brand == pbrand);
 	ASSERT(l->lwp_procp->p_brand_data != NULL);
 	ASSERT(l->lwp_brand != NULL);
-
-	/*
-	 * We should never be called for the last thread in a process.
-	 * (That case is handled by brand_solaris_proc_exit().)
-	 * Therefore this lwp must be exiting from a multi-threaded
-	 * process.
-	 */
-	ASSERT(p->p_tlist != p->p_tlist->t_forw);
-
-	l->lwp_brand = NULL;
 }
 
 /*ARGSUSED*/
 void
-brand_solaris_proc_exit(struct proc *p, klwp_t *l, struct brand *pbrand)
+brand_solaris_proc_exit(struct proc *p, struct brand *pbrand)
 {
 	ASSERT(p->p_brand == pbrand);
 	ASSERT(p->p_brand_data != NULL);
 
-	/*
-	 * When called from proc_exit(), we know that process is
-	 * single-threaded and free our lwp brand data.
-	 * otherwise just free p_brand_data and return.
-	 */
-	if (l != NULL) {
-		ASSERT(p->p_tlist == p->p_tlist->t_forw);
-		ASSERT(p->p_tlist->t_lwp == l);
-		(void) brand_solaris_freelwp(l, pbrand);
-	}
-
 	/* upon exit, free our proc brand data */
 	kmem_free(p->p_brand_data, sizeof (brand_proc_data_t));
 	p->p_brand_data = NULL;
@@ -1145,5 +1141,4 @@ brand_solaris_setbrand(proc_t *p, struct brand *pbrand)
 	ASSERT(p->p_tlist == p->p_tlist->t_forw);
 
 	p->p_brand_data = kmem_zalloc(sizeof (brand_proc_data_t), KM_SLEEP);
-	(void) brand_solaris_initlwp(p->p_tlist->t_lwp, pbrand);
 }
diff --git a/usr/src/uts/common/os/clock_highres.c b/usr/src/uts/common/os/clock_highres.c
index bcdf20c0bd..805813037d 100644
--- a/usr/src/uts/common/os/clock_highres.c
+++ b/usr/src/uts/common/os/clock_highres.c
@@ -25,7 +25,7 @@
  */
 
 /*
- * Copyright (c) 2012, Joyent Inc. All rights reserved.
+ * Copyright (c) 2015, Joyent Inc. All rights reserved.
  */
 
 #include <sys/timer.h>
@@ -66,7 +66,7 @@ clock_highres_getres(timespec_t *ts)
 
 /*ARGSUSED*/
 static int
-clock_highres_timer_create(itimer_t *it, struct sigevent *ev)
+clock_highres_timer_create(itimer_t *it, void (*fire)(itimer_t *))
 {
 	/*
 	 * CLOCK_HIGHRES timers of sufficiently high resolution can deny
@@ -80,6 +80,7 @@ clock_highres_timer_create(itimer_t *it, struct sigevent *ev)
 	}
 
 	it->it_arg = kmem_zalloc(sizeof (cyclic_id_t), KM_SLEEP);
+	it->it_fire = fire;
 
 	return (0);
 }
@@ -95,7 +96,7 @@ clock_highres_fire(void *arg)
 		old = *addr;
 	} while (atomic_cas_64((uint64_t *)addr, old, new) != old);
 
-	timer_fire(it);
+	it->it_fire(it);
 }
 
 static int
diff --git a/usr/src/uts/common/os/clock_realtime.c b/usr/src/uts/common/os/clock_realtime.c
index ef3383fb28..4a75984b23 100644
--- a/usr/src/uts/common/os/clock_realtime.c
+++ b/usr/src/uts/common/os/clock_realtime.c
@@ -24,7 +24,9 @@
  * Use is subject to license terms.
  */
 
-#pragma ident	"%Z%%M%	%I%	%E% SMI"
+/*
+ * Copyright (c) 2015, Joyent Inc. All rights reserved.
+ */
 
 #include <sys/timer.h>
 #include <sys/systm.h>
@@ -80,8 +82,7 @@ clock_realtime_fire(void *arg)
 	/*
 	 * First call into the timer subsystem to get the signal going.
 	 */
-	timer_fire(it);
-
+	it->it_fire(it);
 	val = &it->it_itime.it_value;
 	interval = &it->it_itime.it_interval;
 
@@ -171,9 +172,10 @@ clock_realtime_fire_first(void *arg)
 
 /*ARGSUSED*/
 static int
-clock_realtime_timer_create(itimer_t *it, struct sigevent *ev)
+clock_realtime_timer_create(itimer_t *it, void (*fire)(itimer_t *))
 {
 	it->it_arg = kmem_zalloc(sizeof (timeout_id_t), KM_SLEEP);
+	it->it_fire = fire;
 
 	return (0);
 }
@@ -184,7 +186,7 @@ clock_realtime_timer_settime(itimer_t *it, int flags,
 {
 	timeout_id_t tid, *tidp = it->it_arg;
 	timespec_t now;
-	proc_t *p = curproc;
+	proc_t *p = it->it_proc;
 	clock_t ticks;
 
 	gethrestime(&now);
@@ -246,7 +248,7 @@ static int
 clock_realtime_timer_gettime(itimer_t *it, struct itimerspec *when)
 {
 	timespec_t now;
-	proc_t *p = curproc;
+	proc_t *p = it->it_proc;
 
 	/*
 	 * We always keep it_itime up to date, so we just need to snapshot
@@ -276,7 +278,7 @@ clock_realtime_timer_gettime(itimer_t *it, struct itimerspec *when)
 static int
 clock_realtime_timer_delete(itimer_t *it)
 {
-	proc_t *p = curproc;
+	proc_t *p = it->it_proc;
 	timeout_id_t tid, *tidp = it->it_arg;
 
 	mutex_enter(&p->p_lock);
diff --git a/usr/src/uts/common/os/core.c b/usr/src/uts/common/os/core.c
index d4dddbe477..3ca17e1f17 100644
--- a/usr/src/uts/common/os/core.c
+++ b/usr/src/uts/common/os/core.c
@@ -64,6 +64,7 @@
 #include <sys/contract/process_impl.h>
 #include <sys/ddi.h>
 
+extern int yield(void);
 /*
  * Processes running within a zone potentially dump core in 3 locations,
  * based on the per-process, per-zone, and the global zone's core settings.
diff --git a/usr/src/uts/common/os/cred.c b/usr/src/uts/common/os/cred.c
index 733fd03a92..b0098946b3 100644
--- a/usr/src/uts/common/os/cred.c
+++ b/usr/src/uts/common/os/cred.c
@@ -727,6 +727,14 @@ crgetzoneid(const cred_t *cr)
 	    cr->cr_zone->zone_id);
 }
 
+zoneid_t
+crgetzonedid(const cred_t *cr)
+{
+	return (cr->cr_zone == NULL ?
+	    (cr->cr_uid == -1 ? (zoneid_t)-1 : GLOBAL_ZONEID) :
+	    cr->cr_zone->zone_did);
+}
+
 projid_t
 crgetprojid(const cred_t *cr)
 {
diff --git a/usr/src/uts/common/os/ddi_intr_irm.c b/usr/src/uts/common/os/ddi_intr_irm.c
index c3c0481e7f..a4b35dcb5b 100644
--- a/usr/src/uts/common/os/ddi_intr_irm.c
+++ b/usr/src/uts/common/os/ddi_intr_irm.c
@@ -1320,7 +1320,7 @@ i_ddi_irm_notify(ddi_irm_pool_t *pool_p, ddi_irm_req_t *req_p)
 
 	/* Log callback errors */
 	if (ret != DDI_SUCCESS) {
-		cmn_err(CE_WARN, "%s%d: failed callback (action=%d, ret=%d)\n",
+		cmn_err(CE_WARN, "!%s%d: failed callback (action=%d, ret=%d)\n",
 		    ddi_driver_name(req_p->ireq_dip),
 		    ddi_get_instance(req_p->ireq_dip), (int)action, ret);
 	}
diff --git a/usr/src/uts/common/os/exec.c b/usr/src/uts/common/os/exec.c
index 7adc5c0c9d..7e32fb0506 100644
--- a/usr/src/uts/common/os/exec.c
+++ b/usr/src/uts/common/os/exec.c
@@ -26,7 +26,7 @@
 /*	Copyright (c) 1988 AT&T	*/
 /*	  All Rights Reserved  	*/
 /*
- * Copyright 2014, Joyent, Inc.  All rights reserved.
+ * Copyright 2015, Joyent, Inc.  All rights reserved.
  */
 
 #include <sys/types.h>
@@ -69,6 +69,7 @@
 #include <sys/sdt.h>
 #include <sys/brand.h>
 #include <sys/klpd.h>
+#include <sys/random.h>
 
 #include <c2/audit.h>
 
@@ -97,6 +98,7 @@ uint_t auxv_hwcap32_2 = 0;	/* 32-bit version of auxv_hwcap2 */
 #endif
 
 #define	PSUIDFLAGS		(SNOCD|SUGID)
+#define	RANDOM_LEN	16	/* 16 bytes for AT_RANDOM aux entry */
 
 /*
  * exece() - system call wrapper around exec_common()
@@ -297,14 +299,30 @@ exec_common(const char *fname, const char **argp, const char **envp,
 	ua.argp = argp;
 	ua.envp = envp;
 
-	/* If necessary, brand this process before we start the exec. */
-	if (brandme)
+	/* If necessary, brand this process/lwp before we start the exec. */
+	if (brandme) {
+		void *brand_data = NULL;
+
 		brand_setbrand(p);
+		if (BROP(p)->b_lwpdata_alloc != NULL &&
+		    (brand_data = BROP(p)->b_lwpdata_alloc(p)) == NULL) {
+			VN_RELE(vp);
+			if (dir != NULL)
+				VN_RELE(dir);
+			pn_free(&resolvepn);
+			goto fail;
+		}
+		mutex_enter(&p->p_lock);
+		BROP(p)->b_initlwp(lwp, brand_data);
+		mutex_exit(&p->p_lock);
+	}
 
 	if ((error = gexec(&vp, &ua, &args, NULL, 0, &execsz,
-	    exec_file, p->p_cred, brand_action)) != 0) {
-		if (brandme)
-			brand_clearbrand(p, B_FALSE);
+	    exec_file, p->p_cred, &brand_action)) != 0) {
+		if (brandme) {
+			BROP(p)->b_freelwp(lwp);
+			brand_clearbrand(p);
+		}
 		VN_RELE(vp);
 		if (dir != NULL)
 			VN_RELE(dir);
@@ -360,6 +378,8 @@ exec_common(const char *fname, const char **argp, const char **envp,
 	 * pending held signals remain held, so don't clear t_hold.
 	 */
 	mutex_enter(&p->p_lock);
+	DTRACE_PROBE3(oldcontext__set, klwp_t *, lwp,
+	    uintptr_t, lwp->lwp_oldcontext, uintptr_t, 0);
 	lwp->lwp_oldcontext = 0;
 	lwp->lwp_ustack = 0;
 	lwp->lwp_old_stk_ctl = 0;
@@ -419,8 +439,10 @@ exec_common(const char *fname, const char **argp, const char **envp,
 	TRACE_2(TR_FAC_PROC, TR_PROC_EXEC, "proc_exec:p %p up %p", p, up);
 
 	/* Unbrand ourself if necessary. */
-	if (PROC_IS_BRANDED(p) && (brand_action == EBA_NATIVE))
-		brand_clearbrand(p, B_FALSE);
+	if (PROC_IS_BRANDED(p) && (brand_action == EBA_NATIVE)) {
+		BROP(p)->b_freelwp(lwp);
+		brand_clearbrand(p);
+	}
 
 	setregs(&args);
 
@@ -544,7 +566,7 @@ gexec(
 	long *execsz,
 	caddr_t exec_file,
 	struct cred *cred,
-	int brand_action)
+	int *brand_action)
 {
 	struct vnode *vp, *execvp = NULL;
 	proc_t *pp = ttoproc(curthread);
@@ -858,8 +880,22 @@ gexec(
 			if (pp->p_plist || (pp->p_proc_flag & P_PR_TRACE))
 				args->traceinval = 1;
 		}
-		if (pp->p_proc_flag & P_PR_PTRACE)
-			psignal(pp, SIGTRAP);
+
+		/*
+		 * If legacy ptrace is enabled, defer to the brand as to the
+		 * behavior as to the SIGTRAP generated during exec().  (If
+		 * we're not branded or the brand isn't interested in changing
+		 * the default behavior, we generate the SIGTRAP.)
+		 */
+		if (pp->p_proc_flag & P_PR_PTRACE) {
+			if (PROC_IS_BRANDED(pp) &&
+			    BROP(pp)->b_ptrace_exectrap != NULL) {
+				BROP(pp)->b_ptrace_exectrap(pp);
+			} else {
+				psignal(pp, SIGTRAP);
+			}
+		}
+
 		if (args->traceinval)
 			prinvalidate(&pp->p_user);
 	}
@@ -1517,6 +1553,27 @@ stk_add(uarg_t *args, const char *sp, enum uio_seg segflg)
 	return (0);
 }
 
+/*
+ * Add a fixed size byte array to the stack (only from kernel space).
+ */
+static int
+stk_byte_add(uarg_t *args, const uint8_t *sp, size_t len)
+{
+	int error;
+
+	if (STK_AVAIL(args) < sizeof (int))
+		return (E2BIG);
+	*--args->stk_offp = args->stk_strp - args->stk_base;
+
+	if (len > STK_AVAIL(args))
+		return (E2BIG);
+	bcopy(sp, args->stk_strp, len);
+
+	args->stk_strp += len;
+
+	return (0);
+}
+
 static int
 stk_getptr(uarg_t *args, char *src, char **dst)
 {
@@ -1553,6 +1610,7 @@ stk_copyin(execa_t *uap, uarg_t *args, intpdata_t *intp, void **auxvpp)
 	size_t size, pad;
 	char *argv = (char *)uap->argp;
 	char *envp = (char *)uap->envp;
+	uint8_t rdata[RANDOM_LEN];
 
 	/*
 	 * Copy interpreter's name and argument to argv[0] and argv[1].
@@ -1622,8 +1680,9 @@ stk_copyin(execa_t *uap, uarg_t *args, intpdata_t *intp, void **auxvpp)
 	args->ne = args->na - argc;
 
 	/*
-	 * Add AT_SUN_PLATFORM, AT_SUN_EXECNAME, AT_SUN_BRANDNAME, and
-	 * AT_SUN_EMULATOR strings to the stack.
+	 * Add AT_SUN_PLATFORM, AT_SUN_EXECNAME, AT_SUN_BRANDNAME,
+	 * AT_SUN_BRAND_NROOT, and AT_SUN_EMULATOR strings, as well as AT_RANDOM
+	 * array, to the stack.
 	 */
 	if (auxvpp != NULL && *auxvpp != NULL) {
 		if ((error = stk_add(args, platform, UIO_SYSSPACE)) != 0)
@@ -1636,6 +1695,20 @@ stk_copyin(execa_t *uap, uarg_t *args, intpdata_t *intp, void **auxvpp)
 		if (args->emulator != NULL &&
 		    (error = stk_add(args, args->emulator, UIO_SYSSPACE)) != 0)
 			return (error);
+
+		/*
+		 * For the AT_RANDOM aux vector we provide 16 bytes of random
+		 * data.
+		 */
+		(void) random_get_pseudo_bytes(rdata, sizeof (rdata));
+
+		if ((error = stk_byte_add(args, rdata, sizeof (rdata))) != 0)
+			return (error);
+
+		if (args->brand_nroot != NULL &&
+		    (error = stk_add(args, args->brand_nroot,
+		    UIO_SYSSPACE)) != 0)
+			return (error);
 	}
 
 	/*
@@ -1742,7 +1815,7 @@ stk_copyout(uarg_t *args, char *usrstack, void **auxvpp, user_t *up)
 	/*
 	 * Fill in the aux vector now that we know the user stack addresses
 	 * for the AT_SUN_PLATFORM, AT_SUN_EXECNAME, AT_SUN_BRANDNAME and
-	 * AT_SUN_EMULATOR strings.
+	 * AT_SUN_EMULATOR strings, as well as the AT_RANDOM array.
 	 */
 	if (auxvpp != NULL && *auxvpp != NULL) {
 		if (args->to_model == DATAMODEL_NATIVE) {
@@ -1755,6 +1828,11 @@ stk_copyout(uarg_t *args, char *usrstack, void **auxvpp, user_t *up)
 			if (args->emulator != NULL)
 				ADDAUX(*a,
 				    AT_SUN_EMULATOR, (long)&ustrp[*--offp])
+			ADDAUX(*a, AT_RANDOM, (long)&ustrp[*--offp])
+			if (args->brand_nroot != NULL) {
+				ADDAUX(*a,
+				    AT_SUN_BRAND_NROOT, (long)&ustrp[*--offp])
+			}
 		} else {
 			auxv32_t **a = (auxv32_t **)auxvpp;
 			ADDAUX(*a,
@@ -1767,6 +1845,11 @@ stk_copyout(uarg_t *args, char *usrstack, void **auxvpp, user_t *up)
 			if (args->emulator != NULL)
 				ADDAUX(*a, AT_SUN_EMULATOR,
 				    (int)(uintptr_t)&ustrp[*--offp])
+			ADDAUX(*a, AT_RANDOM, (int)(uintptr_t)&ustrp[*--offp])
+			if (args->brand_nroot != NULL) {
+				ADDAUX(*a, AT_SUN_BRAND_NROOT,
+				    (int)(uintptr_t)&ustrp[*--offp])
+			}
 		}
 	}
 
@@ -1855,6 +1938,9 @@ exec_args(execa_t *uap, uarg_t *args, intpdata_t *intp, void **auxvpp)
 		usrstack = (char *)USRSTACK32;
 	}
 
+	if (args->maxstack != 0 && (uintptr_t)usrstack > args->maxstack)
+		usrstack = (char *)args->maxstack;
+
 	ASSERT(P2PHASE((uintptr_t)usrstack, args->stk_align) == 0);
 
 #if defined(__sparc)
diff --git a/usr/src/uts/common/os/exit.c b/usr/src/uts/common/os/exit.c
index c5d54b5978..06a3856332 100644
--- a/usr/src/uts/common/os/exit.c
+++ b/usr/src/uts/common/os/exit.c
@@ -21,7 +21,7 @@
 
 /*
  * Copyright (c) 1988, 2010, Oracle and/or its affiliates. All rights reserved.
- * Copyright (c) 2011, Joyent, Inc. All rights reserved.
+ * Copyright 2015 Joyent, Inc. All rights reserved.
  */
 
 /*	Copyright (c) 1984, 1986, 1987, 1988, 1989 AT&T	*/
@@ -366,19 +366,6 @@ proc_exit(int why, int what)
 	}
 	mutex_exit(&p->p_lock);
 
-	DTRACE_PROC(lwp__exit);
-	DTRACE_PROC1(exit, int, why);
-
-	/*
-	 * Will perform any brand specific proc exit processing, since this
-	 * is always the last lwp, will also perform lwp_exit and free brand
-	 * data
-	 */
-	if (PROC_IS_BRANDED(p)) {
-		lwp_detach_brand_hdlrs(lwp);
-		brand_clearbrand(p, B_FALSE);
-	}
-
 	/*
 	 * Don't let init exit unless zone_start_init() failed its exec, or
 	 * we are shutting down the zone or the machine.
@@ -390,12 +377,35 @@ proc_exit(int why, int what)
 		if (z->zone_boot_err == 0 &&
 		    zone_status_get(z) < ZONE_IS_SHUTTING_DOWN &&
 		    zone_status_get(global_zone) < ZONE_IS_SHUTTING_DOWN) {
-			if (z->zone_restart_init == B_TRUE) {
-				if (restart_init(what, why) == 0)
-					return (0);
+
+			/*
+			 * If the init process should be restarted, the
+			 * "zone_restart_init" member will be set.  Some init
+			 * programs in branded zones do not tolerate a restart
+			 * in the traditional manner; setting the
+			 * "zone_reboot_on_init_exit" member will cause the
+			 * entire zone to be rebooted instead.  If neither of
+			 * these flags is set the zone will shut down.
+			 */
+			if (z->zone_reboot_on_init_exit == B_TRUE &&
+			    z->zone_restart_init == B_TRUE) {
+				/*
+				 * Trigger a zone reboot and continue
+				 * with exit processing.
+				 */
+				z->zone_init_status = wstat(why, what);
+				(void) zone_kadmin(A_REBOOT, 0, NULL,
+				    zone_kcred());
+
 			} else {
+				if (z->zone_restart_init == B_TRUE) {
+					if (restart_init(what, why) == 0)
+						return (0);
+				}
+
+				z->zone_init_status = wstat(why, what);
 				(void) zone_kadmin(A_SHUTDOWN, AD_HALT, NULL,
-				    CRED());
+				    zone_kcred());
 			}
 		}
 
@@ -407,6 +417,32 @@ proc_exit(int why, int what)
 		z->zone_proc_initpid = -1;
 	}
 
+	/*
+	 * Delay firing probes (and performing brand cleanup) until after the
+	 * zone_proc_initpid check. Cases which result in zone shutdown or
+	 * restart via zone_kadmin eventually result in a call back to
+	 * proc_exit.
+	 */
+	DTRACE_PROC(lwp__exit);
+	DTRACE_PROC1(exit, int, why);
+
+	/*
+	 * Will perform any brand specific proc exit processing. Since this
+	 * is always the last lwp, will also perform lwp exit/free and proc
+	 * exit. Brand data will be freed when the process is reaped.
+	 */
+	if (PROC_IS_BRANDED(p)) {
+		BROP(p)->b_lwpexit(lwp);
+		BROP(p)->b_proc_exit(p);
+		/*
+		 * To ensure that b_proc_exit has access to brand-specific data
+		 * contained by the one remaining lwp, call the freelwp hook as
+		 * the last part of this clean-up process.
+		 */
+		BROP(p)->b_freelwp(lwp);
+		lwp_detach_brand_hdlrs(lwp);
+	}
+
 	lwp_pcb_exit();
 
 	/*
@@ -650,10 +686,22 @@ proc_exit(int why, int what)
 	if ((q = p->p_child) != NULL && p != proc_init) {
 		struct proc	*np;
 		struct proc	*initp = proc_init;
+		pid_t		zone_initpid = 1;
+		struct proc	*zoneinitp = NULL;
 		boolean_t	setzonetop = B_FALSE;
 
-		if (!INGLOBALZONE(curproc))
-			setzonetop = B_TRUE;
+		if (!INGLOBALZONE(curproc)) {
+			zone_initpid = curproc->p_zone->zone_proc_initpid;
+
+			ASSERT(MUTEX_HELD(&pidlock));
+			zoneinitp = prfind(zone_initpid);
+			if (zoneinitp != NULL) {
+				initp = zoneinitp;
+			} else {
+				zone_initpid = 1;
+				setzonetop = B_TRUE;
+			}
+		}
 
 		pgdetach(p);
 
@@ -665,7 +713,8 @@ proc_exit(int why, int what)
 			 */
 			delete_ns(q->p_parent, q);
 
-			q->p_ppid = 1;
+			q->p_ppid = zone_initpid;
+
 			q->p_pidflag &= ~(CLDNOSIGCHLD | CLDWAITPID);
 			if (setzonetop) {
 				mutex_enter(&q->p_lock);
@@ -839,8 +888,50 @@ proc_exit(int why, int what)
 
 	mutex_exit(&p->p_lock);
 	if (!evaporate) {
-		p->p_pidflag &= ~CLDPEND;
-		sigcld(p, sqp);
+		/*
+		 * The brand specific code only happens when the brand has a
+		 * function to call in place of sigcld and the parent of the
+		 * exiting process is not the global zone init. If the parent
+		 * is the global zone init, then the process was reparented,
+		 * and we don't want brand code delivering possibly strange
+		 * signals to init. Also, init is not branded, so any brand
+		 * specific exit data will not be picked up by init anyway.
+		 */
+		if (PROC_IS_BRANDED(p) &&
+		    BROP(p)->b_exit_with_sig != NULL &&
+		    p->p_ppid != 1) {
+			/*
+			 * The code for _fini that could unload the brand_t
+			 * blocks until the count of zones using the module
+			 * reaches zero. Zones decrement the refcount on their
+			 * brands only after all user tasks in that zone have
+			 * exited and been waited on. The decrement on the
+			 * brand's refcount happen in zone_destroy(). That
+			 * depends on zone_shutdown() having been completed.
+			 * zone_shutdown() includes a call to zone_empty(),
+			 * where the zone waits for itself to reach the state
+			 * ZONE_IS_EMPTY. This state is only set in either
+			 * zone_shutdown(), when there are no user processes as
+			 * the zone enters this function, or in
+			 * zone_task_rele(). zone_task_rele() is called from
+			 * code triggered by waiting on processes, not by the
+			 * processes exiting through proc_exit().  This means
+			 * all the branded processes that could exist for a
+			 * specific brand_t must exit and get reaped before the
+			 * refcount on the brand_t can reach 0. _fini will
+			 * never unload the corresponding brand module before
+			 * proc_exit finishes execution for all processes
+			 * branded with a particular brand_t, which makes the
+			 * operation below safe to do. Brands that wish to use
+			 * this mechanism must wait in _fini as described
+			 * above.
+			 */
+			BROP(p)->b_exit_with_sig(p, sqp);
+		} else {
+			p->p_pidflag &= ~CLDPEND;
+			sigcld(p, sqp);
+		}
+
 	} else {
 		/*
 		 * Do what sigcld() would do if the disposition
@@ -919,10 +1010,9 @@ winfo(proc_t *pp, k_siginfo_t *ip, int waitflag)
 int
 waitid(idtype_t idtype, id_t id, k_siginfo_t *ip, int options)
 {
-	int found;
 	proc_t *cp, *pp;
-	int proc_gone;
 	int waitflag = !(options & WNOWAIT);
+	boolean_t have_brand_helper = B_FALSE;
 
 	/*
 	 * Obsolete flag, defined here only for binary compatibility
@@ -950,7 +1040,8 @@ waitid(idtype_t idtype, id_t id, k_siginfo_t *ip, int options)
 	pp = ttoproc(curthread);
 
 	/*
-	 * lock parent mutex so that sibling chain can be searched.
+	 * Anytime you are looking for a process, you take pidlock to prevent
+	 * things from changing as you look.
 	 */
 	mutex_enter(&pidlock);
 
@@ -970,10 +1061,37 @@ waitid(idtype_t idtype, id_t id, k_siginfo_t *ip, int options)
 		return (ECHILD);
 	}
 
-	while (pp->p_child != NULL) {
+	if (PROC_IS_BRANDED(pp) && BROP(pp)->b_waitid_helper != NULL) {
+		have_brand_helper = B_TRUE;
+	}
+
+	while (pp->p_child != NULL || have_brand_helper) {
+		boolean_t brand_wants_wait = B_FALSE;
+		int proc_gone = 0;
+		int found = 0;
+
+		/*
+		 * Give the brand a chance to return synthetic results from
+		 * this waitid() call before we do the real thing.
+		 */
+		if (have_brand_helper) {
+			int ret;
+
+			if (BROP(pp)->b_waitid_helper(idtype, id, ip, options,
+			    &brand_wants_wait, &ret) == 0) {
+				mutex_exit(&pidlock);
+				return (ret);
+			}
 
-		proc_gone = 0;
+			if (pp->p_child == NULL) {
+				goto no_real_children;
+			}
+		}
 
+		/*
+		 * Look for interesting children in the newstate list.
+		 */
+		VERIFY(pp->p_child != NULL);
 		for (cp = pp->p_child_ns; cp != NULL; cp = cp->p_sibling_ns) {
 			if (idtype != P_PID && (cp->p_pidflag & CLDWAITPID))
 				continue;
@@ -981,6 +1099,11 @@ waitid(idtype_t idtype, id_t id, k_siginfo_t *ip, int options)
 				continue;
 			if (idtype == P_PGID && id != cp->p_pgrp)
 				continue;
+			if (PROC_IS_BRANDED(pp)) {
+				if (BROP(pp)->b_wait_filter != NULL &&
+				    BROP(pp)->b_wait_filter(pp, cp) == B_FALSE)
+					continue;
+			}
 
 			switch (cp->p_wcode) {
 
@@ -1025,12 +1148,16 @@ waitid(idtype_t idtype, id_t id, k_siginfo_t *ip, int options)
 		 * Wow! None of the threads on the p_sibling_ns list were
 		 * interesting threads. Check all the kids!
 		 */
-		found = 0;
 		for (cp = pp->p_child; cp != NULL; cp = cp->p_sibling) {
 			if (idtype == P_PID && id != cp->p_pid)
 				continue;
 			if (idtype == P_PGID && id != cp->p_pgrp)
 				continue;
+			if (PROC_IS_BRANDED(pp)) {
+				if (BROP(pp)->b_wait_filter != NULL &&
+				    BROP(pp)->b_wait_filter(pp, cp) == B_FALSE)
+					continue;
+			}
 
 			switch (cp->p_wcode) {
 			case CLD_TRAPPED:
@@ -1099,11 +1226,12 @@ waitid(idtype_t idtype, id_t id, k_siginfo_t *ip, int options)
 				break;
 		}
 
+no_real_children:
 		/*
 		 * If we found no interesting processes at all,
 		 * break out and return ECHILD.
 		 */
-		if (found + proc_gone == 0)
+		if (!brand_wants_wait && (found + proc_gone == 0))
 			break;
 
 		if (options & WNOHANG) {
@@ -1122,7 +1250,7 @@ waitid(idtype_t idtype, id_t id, k_siginfo_t *ip, int options)
 		 * change state while we wait, we don't wait at all.
 		 * Get out with ECHILD according to SVID.
 		 */
-		if (found == proc_gone)
+		if (!brand_wants_wait && (found == proc_gone))
 			break;
 
 		if (!cv_wait_sig_swap(&pp->p_cv, &pidlock)) {
@@ -1218,6 +1346,12 @@ freeproc(proc_t *p)
 		p->p_killsqp = NULL;
 	}
 
+	/* Clear any remaining brand data */
+	if (PROC_IS_BRANDED(p)) {
+		brand_clearbrand(p);
+	}
+
+
 	prfree(p);	/* inform /proc */
 
 	/*
diff --git a/usr/src/uts/common/os/fio.c b/usr/src/uts/common/os/fio.c
index 6dc0d00011..bfee77130d 100644
--- a/usr/src/uts/common/os/fio.c
+++ b/usr/src/uts/common/os/fio.c
@@ -21,7 +21,7 @@
 
 /*
  * Copyright (c) 1989, 2010, Oracle and/or its affiliates. All rights reserved.
- * Copyright (c) 2012, Joyent Inc. All rights reserved.
+ * Copyright 2015, Joyent Inc.
  */
 
 /*	Copyright (c) 1984, 1986, 1987, 1988, 1989 AT&T	*/
@@ -40,6 +40,7 @@
 #include <sys/vnode.h>
 #include <sys/pathname.h>
 #include <sys/file.h>
+#include <sys/flock.h>
 #include <sys/proc.h>
 #include <sys/var.h>
 #include <sys/cpuvar.h>
@@ -851,7 +852,8 @@ flist_fork(uf_info_t *pfip, uf_info_t *cfip)
 	 */
 	cfip->fi_nfiles = nfiles = flist_minsize(pfip);
 
-	cfip->fi_list = kmem_zalloc(nfiles * sizeof (uf_entry_t), KM_SLEEP);
+	cfip->fi_list = nfiles == 0 ? NULL :
+	    kmem_zalloc(nfiles * sizeof (uf_entry_t), KM_SLEEP);
 
 	for (fd = 0, pufp = pfip->fi_list, cufp = cfip->fi_list; fd < nfiles;
 	    fd++, pufp++, cufp++) {
@@ -952,6 +954,8 @@ closef(file_t *fp)
 		return (error);
 	}
 	ASSERT(fp->f_count == 0);
+	/* Last reference, remove any OFD style lock for the file_t */
+	ofdcleanlock(fp);
 	mutex_exit(&fp->f_tlock);
 
 	/*
@@ -1209,7 +1213,8 @@ f_getfl(int fd, int *flagp)
 			error = EBADF;
 		else {
 			vnode_t *vp = fp->f_vnode;
-			int flag = fp->f_flag | (fp->f_flag2 << 16);
+			int flag = fp->f_flag |
+			    ((fp->f_flag2 & ~FEPOLLED) << 16);
 
 			/*
 			 * BSD fcntl() FASYNC compatibility.
diff --git a/usr/src/uts/common/os/flock.c b/usr/src/uts/common/os/flock.c
index 5dad4abb61..a54d6028d5 100644
--- a/usr/src/uts/common/os/flock.c
+++ b/usr/src/uts/common/os/flock.c
@@ -29,6 +29,7 @@
 
 /*
  * Copyright 2011 Nexenta Systems, Inc.  All rights reserved.
+ * Copyright 2015 Joyent, Inc.
  */
 
 #include <sys/flock_impl.h>
@@ -243,9 +244,293 @@ flk_get_lockmgr_status(void)
 }
 
 /*
- * Routine called from fs_frlock in fs/fs_subr.c
+ * This implements Open File Description (not descriptor) style record locking.
+ * These locks can also be thought of as pid-less since they are not tied to a
+ * specific process, thus they're preserved across fork.
+ *
+ * Called directly from fcntl.
+ *
+ * See reclock() for the implementation of the traditional POSIX style record
+ * locking scheme (pid-ful). This function is derived from reclock() but
+ * simplified and modified to work for OFD style locking.
+ *
+ * The two primary advantages of OFD style of locking are:
+ * 1) It is per-file description, so closing a file descriptor that refers to a
+ *    different file description for the same file will not drop the lock (i.e.
+ *    two open's of the same file get different descriptions but a dup or fork
+ *    will refer to the same description).
+ * 2) Locks are preserved across fork(2).
+ *
+ * Because these locks are per-description a lock ptr lives at the f_filocks
+ * member of the file_t and the lock_descriptor includes a file_t pointer
+ * to enable unique lock identification and management.
+ *
+ * Since these locks are pid-less we cannot do deadlock detection with the
+ * current process-oriented implementation. This is consistent with OFD locking
+ * behavior on other operating systems such as Linux. Since we don't do
+ * deadlock detection we never interact with the process graph that is
+ * maintained for deadlock detection on the traditional POSIX-style locks.
+ *
+ * Future Work:
+ *
+ * The current implementation does not support record locks. That is,
+ * currently the single lock must cover the entire file. This is validated in
+ * fcntl. To support record locks the f_filock pointer in the file_t needs to
+ * be changed to a list of pointers to the locks. That list needs to be
+ * managed independently of the lock list on the vnode itself and it needs to
+ * be maintained as record locks are created, split, coalesced and deleted.
+ *
+ * The current implementation does not support remote file systems (e.g.
+ * NFS or CIFS). This is handled in fs_frlock(). The design of how OFD locks
+ * interact with the NLM is not clear since the NLM protocol/implementation
+ * appears to be oriented around locks associated with a process. A further
+ * problem is that a design is needed for what nlm_send_siglost() should do and
+ * where it will send SIGLOST. More recent versions of Linux apparently try to
+ * emulate OFD locks on NFS by converting them to traditional POSIX style locks
+ * that work with the NLM. It is not clear that this provides the correct
+ * semantics in all cases.
  */
+int
+ofdlock(file_t *fp, int fcmd, flock64_t *lckdat, int flag, u_offset_t offset)
+{
+	int cmd = 0;
+	vnode_t *vp;
+	lock_descriptor_t	stack_lock_request;
+	lock_descriptor_t	*lock_request;
+	int error = 0;
+	graph_t	*gp;
+	int serialize = 0;
+
+	if (fcmd != F_OFD_GETLK)
+		cmd = SETFLCK;
+
+	if (fcmd == F_OFD_SETLKW || fcmd == F_FLOCKW)
+		cmd |= SLPFLCK;
+
+	/* see block comment */
+	VERIFY(lckdat->l_whence == 0);
+	VERIFY(lckdat->l_start == 0);
+	VERIFY(lckdat->l_len == 0);
+
+	vp = fp->f_vnode;
+
+	/*
+	 * For reclock fs_frlock() would normally have set these in a few
+	 * places but for us it's cleaner to centralize it here. Note that
+	 * IGN_PID is -1. We use 0 for our pid-less locks.
+	 */
+	lckdat->l_pid = 0;
+	lckdat->l_sysid = 0;
+
+	/*
+	 * Check access permissions
+	 */
+	if ((fcmd == F_OFD_SETLK || fcmd == F_OFD_SETLKW) &&
+	    ((lckdat->l_type == F_RDLCK && (flag & FREAD) == 0) ||
+	    (lckdat->l_type == F_WRLCK && (flag & FWRITE) == 0)))
+		return (EBADF);
+
+	/*
+	 * for query and unlock we use the stack_lock_request
+	 */
+	if (lckdat->l_type == F_UNLCK || !(cmd & SETFLCK)) {
+		lock_request = &stack_lock_request;
+		(void) bzero((caddr_t)lock_request,
+		    sizeof (lock_descriptor_t));
+
+		/*
+		 * following is added to make the assertions in
+		 * flk_execute_request() pass
+		 */
+		lock_request->l_edge.edge_in_next = &lock_request->l_edge;
+		lock_request->l_edge.edge_in_prev = &lock_request->l_edge;
+		lock_request->l_edge.edge_adj_next = &lock_request->l_edge;
+		lock_request->l_edge.edge_adj_prev = &lock_request->l_edge;
+		lock_request->l_status = FLK_INITIAL_STATE;
+	} else {
+		lock_request = flk_get_lock();
+		fp->f_filock = (struct filock *)lock_request;
+	}
+	lock_request->l_state = 0;
+	lock_request->l_vnode = vp;
+	lock_request->l_zoneid = getzoneid();
+	lock_request->l_ofd = fp;
+
+	/*
+	 * Convert the request range into the canonical start and end
+	 * values then check the validity of the lock range.
+	 */
+	error = flk_convert_lock_data(vp, lckdat, &lock_request->l_start,
+	    &lock_request->l_end, offset);
+	if (error)
+		goto done;
+
+	error = flk_check_lock_data(lock_request->l_start, lock_request->l_end,
+	    MAXEND);
+	if (error)
+		goto done;
+
+	ASSERT(lock_request->l_end >= lock_request->l_start);
+
+	lock_request->l_type = lckdat->l_type;
+	if (cmd & SLPFLCK)
+		lock_request->l_state |= WILLING_TO_SLEEP_LOCK;
+
+	if (!(cmd & SETFLCK)) {
+		if (lock_request->l_type == F_RDLCK ||
+		    lock_request->l_type == F_WRLCK)
+			lock_request->l_state |= QUERY_LOCK;
+	}
+	lock_request->l_flock = (*lckdat);
+
+	/*
+	 * We are ready for processing the request
+	 */
+
+	if (fcmd != F_OFD_GETLK && lock_request->l_type != F_UNLCK &&
+	    nbl_need_check(vp)) {
+		nbl_start_crit(vp, RW_WRITER);
+		serialize = 1;
+	}
 
+	/* Get the lock graph for a particular vnode */
+	gp = flk_get_lock_graph(vp, FLK_INIT_GRAPH);
+
+	mutex_enter(&gp->gp_mutex);
+
+	lock_request->l_state |= REFERENCED_LOCK;
+	lock_request->l_graph = gp;
+
+	switch (lock_request->l_type) {
+	case F_RDLCK:
+	case F_WRLCK:
+		if (IS_QUERY_LOCK(lock_request)) {
+			flk_get_first_blocking_lock(lock_request);
+			if (lock_request->l_ofd != NULL)
+				lock_request->l_flock.l_pid = -1;
+			(*lckdat) = lock_request->l_flock;
+		} else {
+			/* process the request now */
+			error = flk_process_request(lock_request);
+		}
+		break;
+
+	case F_UNLCK:
+		/* unlock request will not block so execute it immediately */
+		error = flk_execute_request(lock_request);
+		break;
+
+	default:
+		error = EINVAL;
+		break;
+	}
+
+	if (lock_request == &stack_lock_request) {
+		flk_set_state(lock_request, FLK_DEAD_STATE);
+	} else {
+		lock_request->l_state &= ~REFERENCED_LOCK;
+		if ((error != 0) || IS_DELETED(lock_request)) {
+			flk_set_state(lock_request, FLK_DEAD_STATE);
+			flk_free_lock(lock_request);
+		}
+	}
+
+	mutex_exit(&gp->gp_mutex);
+	if (serialize)
+		nbl_end_crit(vp);
+
+	return (error);
+
+done:
+	flk_set_state(lock_request, FLK_DEAD_STATE);
+	if (lock_request != &stack_lock_request)
+		flk_free_lock(lock_request);
+	return (error);
+}
+
+/*
+ * Remove any lock on the vnode belonging to the given file_t.
+ * Called from closef on last close, file_t is locked.
+ *
+ * This is modeled on the cleanlocks() function but only removes the single
+ * lock associated with fp.
+ */
+void
+ofdcleanlock(file_t *fp)
+{
+	lock_descriptor_t *fplock, *lock, *nlock;
+	vnode_t *vp;
+	graph_t	*gp;
+
+	ASSERT(MUTEX_HELD(&fp->f_tlock));
+
+	if ((fplock = (lock_descriptor_t *)fp->f_filock) == NULL)
+		return;
+
+	fp->f_filock = NULL;
+	vp = fp->f_vnode;
+
+	gp = flk_get_lock_graph(vp, FLK_USE_GRAPH);
+
+	if (gp == NULL)
+		return;
+	mutex_enter(&gp->gp_mutex);
+
+	CHECK_SLEEPING_LOCKS(gp);
+	CHECK_ACTIVE_LOCKS(gp);
+
+	SET_LOCK_TO_FIRST_SLEEP_VP(gp, lock, vp);
+
+	if (lock) {
+		do {
+			nlock = lock->l_next;
+			if (fplock == lock) {
+				CANCEL_WAKEUP(lock);
+				break;
+			}
+			lock = nlock;
+		} while (lock->l_vnode == vp);
+	}
+
+	SET_LOCK_TO_FIRST_ACTIVE_VP(gp, lock, vp);
+
+	if (lock) {
+		do {
+			nlock = lock->l_next;
+			if (fplock == lock) {
+				flk_delete_active_lock(lock, 0);
+				flk_wakeup(lock, 1);
+				flk_free_lock(lock);
+				break;
+			}
+			lock = nlock;
+		} while (lock->l_vnode == vp);
+	}
+
+	CHECK_SLEEPING_LOCKS(gp);
+	CHECK_ACTIVE_LOCKS(gp);
+	mutex_exit(&gp->gp_mutex);
+}
+
+/*
+ * Routine called from fs_frlock in fs/fs_subr.c
+ *
+ * This implements traditional POSIX style record locking. The two primary
+ * drawbacks to this style of locking are:
+ * 1) It is per-process, so any close of a file descriptor that refers to the
+ *    file will drop the lock (e.g. lock /etc/passwd, call a library function
+ *    which opens /etc/passwd to read the file, when the library closes it's
+ *    file descriptor the application loses its lock and does not know).
+ * 2) Locks are not preserved across fork(2).
+ *
+ * Because these locks are only assoiciated with a pid they are per-process.
+ * This is why any close will drop the lock and is also why once the process
+ * forks then the lock is no long related to the new process. These locks can
+ * be considered as pid-ful.
+ *
+ * See ofdlock() for the implementation of a similar but improved locking
+ * scheme.
+ */
 int
 reclock(vnode_t		*vp,
 	flock64_t	*lckdat,
@@ -424,6 +709,8 @@ reclock(vnode_t		*vp,
 	case F_WRLCK:
 		if (IS_QUERY_LOCK(lock_request)) {
 			flk_get_first_blocking_lock(lock_request);
+			if (lock_request->l_ofd != NULL)
+				lock_request->l_flock.l_pid = -1;
 			(*lckdat) = lock_request->l_flock;
 			break;
 		}
@@ -712,7 +999,13 @@ flk_get_lock(void)
 void
 flk_free_lock(lock_descriptor_t	*lock)
 {
+	file_t *fp;
+
 	ASSERT(IS_DEAD(lock));
+
+	if ((fp = lock->l_ofd) != NULL)
+		fp->f_filock = NULL;
+
 	if (IS_REFERENCED(lock)) {
 		lock->l_state |= DELETED_LOCK;
 		return;
@@ -1214,7 +1507,7 @@ flk_add_edge(lock_descriptor_t *from_lock, lock_descriptor_t *to_lock,
 	from_lock->l_edge.edge_adj_next = edge;
 
 	/*
-	 * put in in list of to vertex
+	 * put in list of to vertex
 	 */
 
 	to_lock->l_edge.edge_in_next->edge_in_prev = edge;
@@ -2601,9 +2894,11 @@ flk_canceled(lock_descriptor_t *request)
 }
 
 /*
- * Remove all the locks for the vnode belonging to the given pid and sysid.
+ * Remove all non-OFD locks for the vnode belonging to the given pid and sysid.
+ * That is, since OFD locks are pid-less we'll never match on the incoming
+ * pid. OFD locks are removed earlier in the close() path via closef() and
+ * ofdcleanlock().
  */
-
 void
 cleanlocks(vnode_t *vp, pid_t pid, int sysid)
 {
@@ -2770,6 +3065,14 @@ flk_check_deadlock(lock_descriptor_t *lock)
 	edge_t	*ep, *nep;
 	proc_vertex_t *process_stack;
 
+	/*
+	 * OFD style locks are not associated with any process so there is
+	 * no proc graph for these. Thus we cannot, and do not, do deadlock
+	 * detection.
+	 */
+	if (lock->l_ofd != NULL)
+		return (0);
+
 	STACK_INIT(process_stack);
 
 	mutex_enter(&flock_lock);
@@ -3062,6 +3365,16 @@ flk_update_proc_graph(edge_t *ep, int delete)
 	proc_edge_t *pep, *prevpep;
 
 	mutex_enter(&flock_lock);
+
+	/*
+	 * OFD style locks are not associated with any process so there is
+	 * no proc graph for these.
+	 */
+	if (ep->from_vertex->l_ofd != NULL) {
+		mutex_exit(&flock_lock);
+		return;
+	}
+
 	toproc = flk_get_proc_vertex(ep->to_vertex);
 	fromproc = flk_get_proc_vertex(ep->from_vertex);
 
@@ -3891,6 +4204,7 @@ report_blocker(lock_descriptor_t *blocker, lock_descriptor_t *request)
 	flrp->l_type = blocker->l_type;
 	flrp->l_pid = blocker->l_flock.l_pid;
 	flrp->l_sysid = blocker->l_flock.l_sysid;
+	request->l_ofd = blocker->l_ofd;
 
 	if (IS_LOCKMGR(request)) {
 		flrp->l_start = blocker->l_start;
@@ -4206,6 +4520,10 @@ check_owner_locks(graph_t *gp, pid_t pid, int sysid, vnode_t *vp)
 {
 	lock_descriptor_t *lock;
 
+	/* Ignore OFD style locks since they're not process-wide. */
+	if (pid == 0)
+		return;
+
 	SET_LOCK_TO_FIRST_ACTIVE_VP(gp, lock, vp);
 
 	if (lock) {
diff --git a/usr/src/uts/common/os/fork.c b/usr/src/uts/common/os/fork.c
index 210a301850..fcd33a7788 100644
--- a/usr/src/uts/common/os/fork.c
+++ b/usr/src/uts/common/os/fork.c
@@ -21,7 +21,7 @@
 
 /*
  * Copyright (c) 1988, 2010, Oracle and/or its affiliates. All rights reserved.
- * Copyright 2013, Joyent, Inc. All rights reserved.
+ * Copyright 2015, Joyent, Inc.
  */
 
 /*	Copyright (c) 1984, 1986, 1987, 1988, 1989 AT&T	*/
@@ -696,7 +696,7 @@ fork_fail(proc_t *cp)
 	if (PTOU(curproc)->u_cwd)
 		refstr_rele(PTOU(curproc)->u_cwd);
 	if (PROC_IS_BRANDED(cp)) {
-		brand_clearbrand(cp, B_TRUE);
+		brand_clearbrand(cp);
 	}
 }
 
@@ -1004,6 +1004,9 @@ getproc(proc_t **cpp, pid_t pid, uint_t flags)
 	cp->p_t1_lgrpid = LGRP_NONE;
 	cp->p_tr_lgrpid = LGRP_NONE;
 
+	/* Default to native brand initially */
+	cp->p_brand = &native_brand;
+
 	if ((newpid = pid_allocate(cp, pid, PID_ALLOC_PROC)) == -1) {
 		if (nproc == v.v_proc) {
 			CPU_STATS_ADDQ(CPU, sys, procovf, 1);
@@ -1071,9 +1074,6 @@ getproc(proc_t **cpp, pid_t pid, uint_t flags)
 	cp->p_flag = pp->p_flag & (SJCTL|SNOWAIT|SNOCD);
 	cp->p_sessp = pp->p_sessp;
 	sess_hold(pp);
-	cp->p_brand = pp->p_brand;
-	if (PROC_IS_BRANDED(pp))
-		BROP(pp)->b_copy_procdata(cp, pp);
 	cp->p_bssbase = pp->p_bssbase;
 	cp->p_brkbase = pp->p_brkbase;
 	cp->p_brksize = pp->p_brksize;
@@ -1153,6 +1153,11 @@ getproc(proc_t **cpp, pid_t pid, uint_t flags)
 	mutex_exit(&cp->p_lock);
 	mutex_exit(&pidlock);
 
+	if (PROC_IS_BRANDED(pp)) {
+		brand_setbrand(cp);
+		BROP(pp)->b_copy_procdata(cp, pp);
+	}
+
 	avl_create(&cp->p_ct_held, contract_compar, sizeof (contract_t),
 	    offsetof(contract_t, ct_ctlist));
 
diff --git a/usr/src/uts/common/os/grow.c b/usr/src/uts/common/os/grow.c
index f5e92cfd94..0c4c0bcad6 100644
--- a/usr/src/uts/common/os/grow.c
+++ b/usr/src/uts/common/os/grow.c
@@ -19,7 +19,10 @@
  * CDDL HEADER END
  */
 
-/* Copyright 2013 OmniTI Computer Consulting, Inc. All rights reserved. */
+/*
+ * Copyright 2013 OmniTI Computer Consulting, Inc. All rights reserved.
+ * Copyright (c) 2014, Joyent, Inc. All rights reserved.
+ */
 
 /*
  * Copyright 2009 Sun Microsystems, Inc.  All rights reserved.
@@ -52,6 +55,7 @@
 #include <sys/fcntl.h>
 #include <sys/lwpchan_impl.h>
 #include <sys/nbmlock.h>
+#include <sys/brand.h>
 
 #include <vm/hat.h>
 #include <vm/as.h>
@@ -522,6 +526,20 @@ choose_addr(struct as *as, caddr_t *addrp, size_t len, offset_t off,
 	return (0);
 }
 
+caddr_t
+map_userlimit(proc_t *pp, struct as *as, int flags)
+{
+	if (flags & _MAP_LOW32) {
+		if (PROC_IS_BRANDED(pp) && BROP(pp)->b_map32limit != NULL) {
+			return ((caddr_t)(uintptr_t)BROP(pp)->b_map32limit(pp));
+		} else {
+			return ((caddr_t)_userlimit32);
+		}
+	}
+
+	return (as->a_userlimit);
+}
+
 
 /*
  * Used for MAP_ANON - fast way to get anonymous pages
@@ -537,8 +555,6 @@ zmap(struct as *as, caddr_t *addrp, size_t len, uint_t uprot, int flags,
 		return (EACCES);
 
 	if ((flags & MAP_FIXED) != 0) {
-		caddr_t userlimit;
-
 		/*
 		 * Use the user address.  First verify that
 		 * the address to be used is page aligned.
@@ -547,9 +563,8 @@ zmap(struct as *as, caddr_t *addrp, size_t len, uint_t uprot, int flags,
 		if (((uintptr_t)*addrp & PAGEOFFSET) != 0)
 			return (EINVAL);
 
-		userlimit = flags & _MAP_LOW32 ?
-		    (caddr_t)USERLIMIT32 : as->a_userlimit;
-		switch (valid_usr_range(*addrp, len, uprot, as, userlimit)) {
+		switch (valid_usr_range(*addrp, len, uprot, as,
+		    map_userlimit(as->a_proc, as, flags))) {
 		case RANGE_OKAY:
 			break;
 		case RANGE_BADPROT:
@@ -717,8 +732,6 @@ smmap_common(caddr_t *addrp, size_t len,
 	 * If the user specified an address, do some simple checks here
 	 */
 	if ((flags & MAP_FIXED) != 0) {
-		caddr_t userlimit;
-
 		/*
 		 * Use the user address.  First verify that
 		 * the address to be used is page aligned.
@@ -726,10 +739,8 @@ smmap_common(caddr_t *addrp, size_t len,
 		 */
 		if (((uintptr_t)*addrp & PAGEOFFSET) != 0)
 			return (EINVAL);
-
-		userlimit = flags & _MAP_LOW32 ?
-		    (caddr_t)USERLIMIT32 : as->a_userlimit;
-		switch (valid_usr_range(*addrp, len, uprot, as, userlimit)) {
+		switch (valid_usr_range(*addrp, len, uprot, as,
+		    map_userlimit(curproc, as, flags))) {
 		case RANGE_OKAY:
 			break;
 		case RANGE_BADPROT:
diff --git a/usr/src/uts/common/os/id_space.c b/usr/src/uts/common/os/id_space.c
deleted file mode 100644
index 2dad0cb940..0000000000
--- a/usr/src/uts/common/os/id_space.c
+++ /dev/null
@@ -1,159 +0,0 @@
-/*
- * CDDL HEADER START
- *
- * The contents of this file are subject to the terms of the
- * Common Development and Distribution License (the "License").
- * You may not use this file except in compliance with the License.
- *
- * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
- * or http://www.opensolaris.org/os/licensing.
- * See the License for the specific language governing permissions
- * and limitations under the License.
- *
- * When distributing Covered Code, include this CDDL HEADER in each
- * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
- * If applicable, add the following below this CDDL HEADER, with the
- * fields enclosed by brackets "[]" replaced with your own identifying
- * information: Portions Copyright [yyyy] [name of copyright owner]
- *
- * CDDL HEADER END
- */
-/*
- * Copyright (c) 2000, 2010, Oracle and/or its affiliates. All rights reserved.
- */
-
-#include <sys/types.h>
-#include <sys/id_space.h>
-#include <sys/debug.h>
-
-/*
- * ID Spaces
- *
- *   The id_space_t provides a simple implementation of a managed range of
- *   integer identifiers using a vmem arena.  An ID space guarantees that the
- *   next identifer returned by an allocation is larger than the previous one,
- *   unless there are no larger slots remaining in the range.  In this case,
- *   the ID space will return the first available slot in the lower part of the
- *   range (viewing the previous identifier as a partitioning element).  If no
- *   slots are available, id_alloc()/id_allocff() will sleep until an
- *   identifier becomes available.  Accordingly, id_space allocations must be
- *   initiated from contexts where sleeping is acceptable.  id_alloc_nosleep()/
- *   id_allocff_nosleep() will return -1 if no slots are available or if the
- *   system is low on memory.  If id_alloc_nosleep() fails, callers should
- *   not try to extend the ID space.  This is to avoid making a possible
- *   low-memory situation worse.
- *
- *   As an ID space is designed for representing a range of id_t's, there
- *   is a preexisting maximal range: [0, MAXUID].  ID space requests outside
- *   that range will fail on a DEBUG kernel.  The id_allocff*() functions
- *   return the first available id, and should be used when there is benefit
- *   to having a compact allocated range.
- *
- *   (Presently, the id_space_t abstraction supports only direct allocations; ID
- *   reservation, in which an ID is allocated but placed in a internal
- *   dictionary for later use, should be added when a consuming subsystem
- *   arrives.)
- */
-
-#define	ID_TO_ADDR(id) ((void *)(uintptr_t)(id + 1))
-#define	ADDR_TO_ID(addr) ((id_t)((uintptr_t)addr - 1))
-
-/*
- * Create an arena to represent the range [low, high).
- * Caller must be in a context in which VM_SLEEP is legal.
- */
-id_space_t *
-id_space_create(const char *name, id_t low, id_t high)
-{
-	ASSERT(low >= 0);
-	ASSERT(low < high);
-
-	return (vmem_create(name, ID_TO_ADDR(low), high - low, 1,
-	    NULL, NULL, NULL, 0, VM_SLEEP | VMC_IDENTIFIER));
-}
-
-/*
- * Destroy a previously created ID space.
- * No restrictions on caller's context.
- */
-void
-id_space_destroy(id_space_t *isp)
-{
-	vmem_destroy(isp);
-}
-
-void
-id_space_extend(id_space_t *isp, id_t low, id_t high)
-{
-	(void) vmem_add(isp, ID_TO_ADDR(low), high - low, VM_SLEEP);
-}
-
-/*
- * Allocate an id_t from specified ID space.
- * Caller must be in a context in which VM_SLEEP is legal.
- */
-id_t
-id_alloc(id_space_t *isp)
-{
-	return (ADDR_TO_ID(vmem_alloc(isp, 1, VM_SLEEP | VM_NEXTFIT)));
-}
-
-/*
- * Allocate an id_t from specified ID space.
- * Returns -1 on failure (see module block comments for more information on
- * failure modes).
- */
-id_t
-id_alloc_nosleep(id_space_t *isp)
-{
-	return (ADDR_TO_ID(vmem_alloc(isp, 1, VM_NOSLEEP | VM_NEXTFIT)));
-}
-
-/*
- * Allocate an id_t from specified ID space using FIRSTFIT.
- * Caller must be in a context in which VM_SLEEP is legal.
- */
-id_t
-id_allocff(id_space_t *isp)
-{
-	return (ADDR_TO_ID(vmem_alloc(isp, 1, VM_SLEEP | VM_FIRSTFIT)));
-}
-
-/*
- * Allocate an id_t from specified ID space using FIRSTFIT
- * Returns -1 on failure (see module block comments for more information on
- * failure modes).
- */
-id_t
-id_allocff_nosleep(id_space_t *isp)
-{
-	return (ADDR_TO_ID(vmem_alloc(isp, 1, VM_NOSLEEP | VM_FIRSTFIT)));
-}
-
-/*
- * Allocate a specific identifier if possible, returning the id if
- * successful, or -1 on failure.
- */
-id_t
-id_alloc_specific_nosleep(id_space_t *isp, id_t id)
-{
-	void *minaddr = ID_TO_ADDR(id);
-	void *maxaddr = ID_TO_ADDR(id + 1);
-
-	/*
-	 * Note that even though we're vmem_free()ing this later, it
-	 * should be OK, since there's no quantum cache.
-	 */
-	return (ADDR_TO_ID(vmem_xalloc(isp, 1, 1, 0, 0,
-	    minaddr, maxaddr, VM_NOSLEEP)));
-}
-
-/*
- * Free a previously allocated ID.
- * No restrictions on caller's context.
- */
-void
-id_free(id_space_t *isp, id_t id)
-{
-	vmem_free(isp, ID_TO_ADDR(id), 1);
-}
diff --git a/usr/src/uts/common/os/kmem.c b/usr/src/uts/common/os/kmem.c
index af94c2d16f..9d38183ce1 100644
--- a/usr/src/uts/common/os/kmem.c
+++ b/usr/src/uts/common/os/kmem.c
@@ -20,6 +20,7 @@
  */
 /*
  * Copyright (c) 1994, 2010, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2015 Joyent, Inc.  All rights reserved.
  */
 
 /*
@@ -1002,6 +1003,13 @@ size_t kmem_lite_maxalign = 1024; /* maximum buffer alignment for KMF_LITE */
 int kmem_lite_pcs = 4;		/* number of PCs to store in KMF_LITE mode */
 size_t kmem_maxverify;		/* maximum bytes to inspect in debug routines */
 size_t kmem_minfirewall;	/* hardware-enforced redzone threshold */
+int kmem_warn_zerosized = 1;	/* whether to warn on zero-sized KM_SLEEP */
+
+#ifdef DEBUG
+int kmem_panic_zerosized = 1;	/* whether to panic on zero-sized KM_SLEEP */
+#else
+int kmem_panic_zerosized = 0;	/* whether to panic on zero-sized KM_SLEEP */
+#endif
 
 #ifdef _LP64
 size_t	kmem_max_cached = KMEM_BIG_MAXBUF;	/* maximum kmem_alloc cache */
@@ -1036,6 +1044,8 @@ static vmem_t		*kmem_default_arena;
 static vmem_t		*kmem_firewall_va_arena;
 static vmem_t		*kmem_firewall_arena;
 
+static int		kmem_zerosized;		/* # of zero-sized allocs */
+
 /*
  * Define KMEM_STATS to turn on statistic gathering. By default, it is only
  * turned on when DEBUG is also defined.
@@ -2925,8 +2935,31 @@ kmem_alloc(size_t size, int kmflag)
 		/* fall through to kmem_cache_alloc() */
 
 	} else {
-		if (size == 0)
+		if (size == 0) {
+			if (kmflag != KM_SLEEP && !(kmflag & KM_PANIC))
+				return (NULL);
+
+			/*
+			 * If this is a sleeping allocation or one that has
+			 * been specified to panic on allocation failure, we
+			 * consider it to be deprecated behavior to allocate
+			 * 0 bytes.  If we have been configured to panic under
+			 * this condition, we panic; if to warn, we warn -- and
+			 * regardless, we bump a counter to at least indicate
+			 * that this condition has occurred.
+			 */
+			if (kmem_panic && kmem_panic_zerosized)
+				panic("attempted to kmem_alloc() size of 0");
+
+			if (kmem_warn_zerosized) {
+				cmn_err(CE_WARN, "kmem_alloc(): sleeping "
+				    "allocation with size of 0");
+			}
+
+			kmem_zerosized++;
+
 			return (NULL);
+		}
 
 		buf = vmem_alloc(kmem_oversize_arena, size,
 		    kmflag & KM_VMFLAGS);
@@ -4120,7 +4153,8 @@ kmem_cache_destroy(kmem_cache_t *cp)
 
 	if (kmem_taskq != NULL)
 		taskq_wait(kmem_taskq);
-	if (kmem_move_taskq != NULL)
+
+	if (kmem_move_taskq != NULL && cp->cache_defrag != NULL)
 		taskq_wait(kmem_move_taskq);
 
 	kmem_cache_magazine_purge(cp);
diff --git a/usr/src/uts/common/os/logsubr.c b/usr/src/uts/common/os/logsubr.c
index 149f5f8a88..6a603c8982 100644
--- a/usr/src/uts/common/os/logsubr.c
+++ b/usr/src/uts/common/os/logsubr.c
@@ -22,6 +22,7 @@
 /*
  * Copyright (c) 2013 Gary Mills
  * Copyright (c) 1998, 2010, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2015, Joyent, Inc.
  */
 
 #include <sys/types.h>
@@ -249,8 +250,7 @@ log_init(void)
 	 */
 	printf("\rSunOS Release %s Version %s %u-bit\n",
 	    utsname.release, utsname.version, NBBY * (uint_t)sizeof (void *));
-	printf("Copyright (c) 1983, 2010, Oracle and/or its affiliates. "
-	    "All rights reserved.\n");
+	printf("Copyright (c) 2010-2015, Joyent Inc. All rights reserved.\n");
 #ifdef DEBUG
 	printf("DEBUG enabled\n");
 #endif
diff --git a/usr/src/uts/common/os/lwp.c b/usr/src/uts/common/os/lwp.c
index feb8e76c42..f5be1b40b3 100644
--- a/usr/src/uts/common/os/lwp.c
+++ b/usr/src/uts/common/os/lwp.c
@@ -25,7 +25,7 @@
  */
 
 /*
- * Copyright (c) 2013, Joyent, Inc. All rights reserved.
+ * Copyright 2015, Joyent, Inc.
  */
 
 #include <sys/param.h>
@@ -115,7 +115,7 @@ lwp_create(void (*proc)(), caddr_t arg, size_t len, proc_t *p,
 	ret_tidhash_t *ret_tidhash = NULL;
 	int i;
 	int rctlfail = 0;
-	boolean_t branded = 0;
+	void *brand_data = NULL;
 	struct ctxop *ctx = NULL;
 
 	ASSERT(cid != sysdccid);	/* system threads must start in SYS */
@@ -283,6 +283,19 @@ lwp_create(void (*proc)(), caddr_t arg, size_t len, proc_t *p,
 	 */
 	lep = kmem_zalloc(sizeof (*lep), KM_SLEEP);
 
+	/*
+	 * If necessary, speculatively allocate lwp brand data.  This is done
+	 * ahead of time so p_lock need not be dropped during lwp branding.
+	 */
+	if (PROC_IS_BRANDED(p) && BROP(p)->b_lwpdata_alloc != NULL) {
+		if ((brand_data = BROP(p)->b_lwpdata_alloc(p)) == NULL) {
+			mutex_enter(&p->p_lock);
+			err = 1;
+			atomic_inc_32(&p->p_zone->zone_ffmisc);
+			goto error;
+		}
+	}
+
 	mutex_enter(&p->p_lock);
 grow:
 	/*
@@ -630,18 +643,6 @@ grow:
 		} while (lwp_hash_lookup(p, t->t_tid) != NULL);
 	}
 
-	/*
-	 * If this is a branded process, let the brand do any necessary lwp
-	 * initialization.
-	 */
-	if (PROC_IS_BRANDED(p)) {
-		if (BROP(p)->b_initlwp(lwp)) {
-			err = 1;
-			atomic_inc_32(&p->p_zone->zone_ffmisc);
-			goto error;
-		}
-		branded = 1;
-	}
 
 	if (t->t_tid == 1) {
 		kpreempt_disable();
@@ -654,7 +655,6 @@ grow:
 		}
 	}
 
-	p->p_lwpcnt++;
 	t->t_waitfor = -1;
 
 	/*
@@ -695,9 +695,21 @@ grow:
 	t->t_pre_sys = 1;
 	t->t_post_sys = 1;
 
+	/* Complete branded lwp initialization */
+	if (PROC_IS_BRANDED(p)) {
+		BROP(p)->b_initlwp(lwp, brand_data);
+		/*
+		 * The b_initlwp hook is expected to consume any preallocated
+		 * brand_data in a way that prepares it for deallocation by the
+		 * b_freelwp hook.
+		 */
+		brand_data = NULL;
+	}
+
 	/*
 	 * Insert the new thread into the list of all threads.
 	 */
+	p->p_lwpcnt++;
 	if ((tx = p->p_tlist) == NULL) {
 		t->t_back = t;
 		t->t_forw = t;
@@ -753,8 +765,9 @@ error:
 		if (cid != NOCLASS && bufp != NULL)
 			CL_FREE(cid, bufp);
 
-		if (branded)
-			BROP(p)->b_freelwp(lwp);
+		if (brand_data != NULL) {
+			BROP(p)->b_lwpdata_free(brand_data);
+		}
 
 		mutex_exit(&p->p_lock);
 		t->t_state = TS_FREE;
@@ -891,13 +904,6 @@ lwp_exit(void)
 	if (t->t_upimutex != NULL)
 		upimutex_cleanup();
 
-	/*
-	 * Perform any brand specific exit processing, then release any
-	 * brand data associated with the lwp
-	 */
-	if (PROC_IS_BRANDED(p))
-		BROP(p)->b_lwpexit(lwp);
-
 	lwp_pcb_exit();
 
 	mutex_enter(&p->p_lock);
@@ -941,6 +947,17 @@ lwp_exit(void)
 	DTRACE_PROC(lwp__exit);
 
 	/*
+	 * Perform any brand specific exit processing, then release any
+	 * brand data associated with the lwp
+	 */
+	if (PROC_IS_BRANDED(p)) {
+		mutex_exit(&p->p_lock);
+		BROP(p)->b_lwpexit(lwp);
+		BROP(p)->b_freelwp(lwp);
+		mutex_enter(&p->p_lock);
+	}
+
+	/*
 	 * If the lwp is a detached lwp or if the process is exiting,
 	 * remove (lwp_hash_out()) the lwp from the lwp directory.
 	 * Otherwise null out the lwp's le_thread pointer in the lwp
diff --git a/usr/src/uts/common/os/main.c b/usr/src/uts/common/os/main.c
index 7afc1cfe00..dda0b3e4a6 100644
--- a/usr/src/uts/common/os/main.c
+++ b/usr/src/uts/common/os/main.c
@@ -27,7 +27,7 @@
 /*	  All Rights Reserved  	*/
 
 /*
- * Copyright (c) 2013, Joyent, Inc. All rights reserved.
+ * Copyright 2015, Joyent, Inc.
  */
 
 #include <sys/types.h>
@@ -157,7 +157,7 @@ exec_init(const char *initpath, const char *args)
 	int error = 0, count = 0;
 	proc_t *p = ttoproc(curthread);
 	klwp_t *lwp = ttolwp(curthread);
-	int brand_action;
+	int brand_action = EBA_NONE;
 
 	if (args == NULL)
 		args = "";
@@ -268,7 +268,15 @@ exec_init(const char *initpath, const char *args)
 	 */
 	sigemptyset(&curthread->t_hold);
 
-	brand_action = ZONE_IS_BRANDED(p->p_zone) ? EBA_BRAND : EBA_NONE;
+	/*
+	 * Only instruct exec_common to brand the process if necessary.  It is
+	 * possible that the init process is already properly branded due to the
+	 * proc_exit -> restart_init -> exec_init call chain.
+	 */
+	if (ZONE_IS_BRANDED(p->p_zone) &&
+	    p->p_brand != p->p_zone->zone_brand) {
+		brand_action = EBA_BRAND;
+	}
 again:
 	error = exec_common((const char *)(uintptr_t)exec_fnamep,
 	    (const char **)(uintptr_t)uap, NULL, brand_action);
diff --git a/usr/src/uts/common/os/msacct.c b/usr/src/uts/common/os/msacct.c
index 928c6b3bb4..66994321f7 100644
--- a/usr/src/uts/common/os/msacct.c
+++ b/usr/src/uts/common/os/msacct.c
@@ -21,6 +21,7 @@
 /*
  * Copyright 2009 Sun Microsystems, Inc.  All rights reserved.
  * Use is subject to license terms.
+ * Copyright 2012 Joyent, Inc.  All rights reserved.
  */
 
 #include <sys/types.h>
@@ -566,27 +567,18 @@ cpu_update_pct(kthread_t *t, hrtime_t newtime)
 	 */
 
 	do {
-		if (T_ONPROC(t) && t->t_waitrq == 0) {
-			hrlb = t->t_hrtime;
+		pctcpu = t->t_pctcpu;
+		hrlb = t->t_hrtime;
+		delta = newtime - hrlb;
+		if (delta < 0) {
+			newtime = gethrtime_unscaled();
 			delta = newtime - hrlb;
-			if (delta < 0) {
-				newtime = gethrtime_unscaled();
-				delta = newtime - hrlb;
-			}
-			t->t_hrtime = newtime;
-			scalehrtime(&delta);
-			pctcpu = t->t_pctcpu;
+		}
+		t->t_hrtime = newtime;
+		scalehrtime(&delta);
+		if (T_ONPROC(t) && t->t_waitrq == 0) {
 			npctcpu = cpu_grow(pctcpu, delta);
 		} else {
-			hrlb = t->t_hrtime;
-			delta = newtime - hrlb;
-			if (delta < 0) {
-				newtime = gethrtime_unscaled();
-				delta = newtime - hrlb;
-			}
-			t->t_hrtime = newtime;
-			scalehrtime(&delta);
-			pctcpu = t->t_pctcpu;
 			npctcpu = cpu_decay(pctcpu, delta);
 		}
 	} while (atomic_cas_32(&t->t_pctcpu, pctcpu, npctcpu) != pctcpu);
diff --git a/usr/src/uts/common/os/pid.c b/usr/src/uts/common/os/pid.c
index b555bb82b7..39db5cb27d 100644
--- a/usr/src/uts/common/os/pid.c
+++ b/usr/src/uts/common/os/pid.c
@@ -112,6 +112,18 @@ pid_lookup(pid_t pid)
 	return (pidp);
 }
 
+struct pid *
+pid_find(pid_t pid)
+{
+	struct pid *pidp;
+
+	mutex_enter(&pidlinklock);
+	pidp = pid_lookup(pid);
+	mutex_exit(&pidlinklock);
+
+	return (pidp);
+}
+
 void
 pid_setmin(void)
 {
diff --git a/usr/src/uts/common/os/policy.c b/usr/src/uts/common/os/policy.c
index 05deaaf43b..d2bdb4ce37 100644
--- a/usr/src/uts/common/os/policy.c
+++ b/usr/src/uts/common/os/policy.c
@@ -20,7 +20,7 @@
  */
 /*
  * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
- * Copyright 2013, Joyent, Inc. All rights reserved.
+ * Copyright 2015 Joyent, Inc.
  */
 
 #include <sys/types.h>
@@ -55,6 +55,7 @@
 #include <sys/mntent.h>
 #include <sys/contract_impl.h>
 #include <sys/dld_ioc.h>
+#include <sys/brand.h>
 
 /*
  * There are two possible layers of privilege routines and two possible
@@ -1243,6 +1244,22 @@ secpolicy_vnode_owner(const cred_t *cr, uid_t owner)
 void
 secpolicy_setid_clear(vattr_t *vap, cred_t *cr)
 {
+	proc_t *p = curproc;
+
+	/*
+	 * Allow the brand to override this behaviour.
+	 */
+	if (PROC_IS_BRANDED(p) && BROP(p)->b_setid_clear != NULL) {
+		/*
+		 * This brand hook will return 0 if handling is complete, or
+		 * some other value if the brand would like us to fall back to
+		 * the usual behaviour.
+		 */
+		if (BROP(p)->b_setid_clear(vap, cr) == 0) {
+			return;
+		}
+	}
+
 	if ((vap->va_mode & (S_ISUID | S_ISGID)) != 0 &&
 	    secpolicy_vnode_setid_retain(cr,
 	    (vap->va_mode & S_ISUID) != 0 &&
@@ -2072,6 +2089,19 @@ secpolicy_tasksys(const cred_t *cr)
 }
 
 int
+secpolicy_meminfo(const cred_t *cr)
+{
+	return (PRIV_POLICY(cr, PRIV_PROC_MEMINFO, B_FALSE, EPERM, NULL));
+}
+
+int
+secpolicy_fs_import(const cred_t *cr)
+{
+	return (PRIV_POLICY(cr, PRIV_SYS_FS_IMPORT, B_FALSE, EPERM, NULL));
+}
+
+
+int
 secpolicy_pfexec_register(const cred_t *cr)
 {
 	return (PRIV_POLICY(cr, PRIV_SYS_ADMIN, B_TRUE, EPERM, NULL));
@@ -2575,3 +2605,11 @@ secpolicy_ppp_config(const cred_t *cr)
 		return (secpolicy_net_config(cr, B_FALSE));
 	return (PRIV_POLICY(cr, PRIV_SYS_PPP_CONFIG, B_FALSE, EPERM, NULL));
 }
+
+int
+secpolicy_hyprlofs_control(const cred_t *cr)
+{
+	if (PRIV_POLICY(cr, PRIV_HYPRLOFS_CONTROL, B_FALSE, EPERM, NULL))
+		return (EPERM);
+	return (0);
+}
diff --git a/usr/src/uts/common/os/priv_defs b/usr/src/uts/common/os/priv_defs
index bfacce1739..cc1c5e03a6 100644
--- a/usr/src/uts/common/os/priv_defs
+++ b/usr/src/uts/common/os/priv_defs
@@ -20,7 +20,7 @@
  */
 /*
  * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
- * Copyright 2013, Joyent, Inc. All rights reserved.
+ * Copyright 2015, Joyent, Inc. All rights reserved.
  *
 INSERT COMMENT
  */
@@ -177,6 +177,10 @@ privilege PRIV_GRAPHICS_MAP
 	Allows a process to perform privileged mappings through a
 	graphics device.
 
+privilege PRIV_HYPRLOFS_CONTROL
+
+	Allows a process to manage hyprlofs entries.
+
 privilege PRIV_IPC_DAC_READ
 
 	Allows a process to read a System V IPC
@@ -291,6 +295,10 @@ privilege PRIV_PROC_LOCK_MEMORY
 
 	Allows a process to lock pages in physical memory.
 
+privilege PRIV_PROC_MEMINFO
+
+	Allows a process to access physical memory information.
+
 privilege PRIV_PROC_OWNER
 
 	Allows a process to send signals to other processes, inspect
@@ -368,6 +376,10 @@ privilege PRIV_SYS_DEVICES
 	Allows a process to open the real console device directly.
 	Allows a process to open devices that have been exclusively opened.
 
+privilege PRIV_SYS_FS_IMPORT
+
+	Allows a process to import a potentially untrusted file system.
+
 privilege PRIV_SYS_IPC_CONFIG
 
 	Allows a process to increase the size of a System V IPC Message
diff --git a/usr/src/uts/common/os/sched.c b/usr/src/uts/common/os/sched.c
index c1d6569f11..15e77d39f7 100644
--- a/usr/src/uts/common/os/sched.c
+++ b/usr/src/uts/common/os/sched.c
@@ -27,6 +27,10 @@
 /*	Copyright (c) 1984, 1986, 1987, 1988, 1989 AT&T	*/
 /*	  All Rights Reserved	*/
 
+/*
+ * Copyright (c) 2015, Joyent, Inc.  All rights reserved.
+ */
+
 #include <sys/param.h>
 #include <sys/types.h>
 #include <sys/sysmacros.h>
@@ -646,16 +650,17 @@ top:
 		klwp_t *lwp = ttolwp(tp);
 
 		/*
-		 * Swapout eligible lwps (specified by the scheduling
-		 * class) which don't have TS_DONT_SWAP set.  Set the
-		 * "intent to swap" flag (TS_SWAPENQ) on threads
-		 * which have TS_DONT_SWAP set so that they can be
+		 * Swapout eligible lwps (specified by the scheduling class)
+		 * which don't have TS_DONT_SWAP set.  Set the "intent to swap"
+		 * flag (TS_SWAPENQ) on threads which have either TS_DONT_SWAP
+		 * set or are currently on a split stack so that they can be
 		 * swapped if and when they reach a safe point.
 		 */
 		thread_lock(tp);
 		thread_pri = CL_SWAPOUT(tp, swapflags);
 		if (thread_pri != -1) {
-			if (tp->t_schedflag & TS_DONT_SWAP) {
+			if ((tp->t_schedflag & TS_DONT_SWAP) ||
+			    (tp->t_flag & T_SPLITSTK)) {
 				tp->t_schedflag |= TS_SWAPENQ;
 				tp->t_trapret = 1;
 				aston(tp);
diff --git a/usr/src/uts/common/os/sig.c b/usr/src/uts/common/os/sig.c
index 0b79c3765a..bc48c6e6e8 100644
--- a/usr/src/uts/common/os/sig.c
+++ b/usr/src/uts/common/os/sig.c
@@ -22,7 +22,7 @@
 /*
  * Copyright 2010 Sun Microsystems, Inc.  All rights reserved.
  * Use is subject to license terms.
- * Copyright (c) 2014, Joyent, Inc.  All rights reserved.
+ * Copyright 2015, Joyent, Inc.
  */
 
 /*	Copyright (c) 1984, 1986, 1987, 1988, 1989 AT&T	*/
@@ -60,6 +60,7 @@
 #include <sys/cyclic.h>
 #include <sys/dtrace.h>
 #include <sys/sdt.h>
+#include <sys/brand.h>
 
 const k_sigset_t nullsmask = {0, 0, 0};
 
@@ -141,6 +142,21 @@ signal_is_blocked(kthread_t *t, int sig)
 }
 
 /*
+ * Return true if the signal can safely be ignored.
+ * That is, if the signal is included in the p_ignore mask and doing so is not
+ * forbidden by any process branding.
+ */
+static int
+sig_ignorable(proc_t *p, int sig)
+{
+	return (sigismember(&p->p_ignore, sig) &&	/* sig in ignore mask */
+	    !(PROC_IS_BRANDED(p) &&			/* allowed by brand */
+	    BROP(p)->b_sig_ignorable != NULL &&
+	    BROP(p)->b_sig_ignorable(p, sig) == B_FALSE));
+
+}
+
+/*
  * Return true if the signal can safely be discarded on generation.
  * That is, if there is no need for the signal on the receiving end.
  * The answer is true if the process is a zombie or
@@ -157,7 +173,7 @@ sig_discardable(proc_t *p, int sig)
 	kthread_t *t = p->p_tlist;
 
 	return (t == NULL ||		/* if zombie or ... */
-	    (sigismember(&p->p_ignore, sig) &&	/* signal is ignored */
+	    (sig_ignorable(p, sig) &&		/* signal is ignored */
 	    t->t_forw == t &&			/* and single-threaded */
 	    !tracing(p, sig) &&			/* and no /proc tracing */
 	    !signal_is_blocked(t, sig) &&	/* and signal not blocked */
@@ -193,7 +209,7 @@ eat_signal(kthread_t *t, int sig)
 		    !(ttoproc(t)->p_proc_flag & P_PR_LOCK)) {
 			ttoproc(t)->p_stopsig = 0;
 			t->t_dtrace_stop = 0;
-			t->t_schedflag |= TS_XSTART | TS_PSTART;
+			t->t_schedflag |= TS_XSTART | TS_PSTART | TS_BSTART;
 			setrun_locked(t);
 		} else if (t != curthread && t->t_state == TS_ONPROC) {
 			aston(t);	/* make it do issig promptly */
@@ -481,7 +497,7 @@ issig_justlooking(void)
 			if (sigismember(&set, sig) &&
 			    (tracing(p, sig) ||
 			    sigismember(&t->t_sigwait, sig) ||
-			    !sigismember(&p->p_ignore, sig))) {
+			    !sig_ignorable(p, sig))) {
 				/*
 				 * Don't promote a signal that will stop
 				 * the process when lwp_nostop is set.
@@ -607,6 +623,21 @@ issig_forreal(void)
 		}
 
 		/*
+		 * Allow the brand the chance to alter (or suppress) delivery
+		 * of this signal.
+		 */
+		if (PROC_IS_BRANDED(p) && BROP(p)->b_issig_stop != NULL) {
+			/*
+			 * The brand hook will return 0 if it would like
+			 * us to drive on, or -1 if we should restart
+			 * the loop to check other conditions.
+			 */
+			if (BROP(p)->b_issig_stop(p, lwp) != 0) {
+				continue;
+			}
+		}
+
+		/*
 		 * Honor requested stop before dealing with the
 		 * current signal; a debugger may change it.
 		 * Do not want to go back to loop here since this is a special
@@ -640,7 +671,7 @@ issig_forreal(void)
 			lwp->lwp_cursig = 0;
 			lwp->lwp_extsig = 0;
 			if (sigismember(&t->t_sigwait, sig) ||
-			    (!sigismember(&p->p_ignore, sig) &&
+			    (!sig_ignorable(p, sig) &&
 			    !isjobstop(sig))) {
 				if (p->p_flag & (SEXITLWPS|SKILLED)) {
 					sig = SIGKILL;
@@ -692,7 +723,7 @@ issig_forreal(void)
 				toproc = 0;
 				if (tracing(p, sig) ||
 				    sigismember(&t->t_sigwait, sig) ||
-				    !sigismember(&p->p_ignore, sig)) {
+				    !sig_ignorable(p, sig)) {
 					if (sigismember(&t->t_extsig, sig))
 						ext = 1;
 					break;
@@ -706,7 +737,7 @@ issig_forreal(void)
 				toproc = 1;
 				if (tracing(p, sig) ||
 				    sigismember(&t->t_sigwait, sig) ||
-				    !sigismember(&p->p_ignore, sig)) {
+				    !sig_ignorable(p, sig)) {
 					if (sigismember(&p->p_extsig, sig))
 						ext = 1;
 					break;
@@ -938,6 +969,16 @@ stop(int why, int what)
 		}
 		break;
 
+	case PR_BRAND:
+		/*
+		 * We have been stopped by the brand code for a brand-private
+		 * reason.  This is an asynchronous stop affecting only this
+		 * LWP.
+		 */
+		VERIFY(PROC_IS_BRANDED(p));
+		flags &= ~TS_BSTART;
+		break;
+
 	default:	/* /proc stop */
 		flags &= ~TS_PSTART;
 		/*
@@ -1049,7 +1090,7 @@ stop(int why, int what)
 		}
 	}
 
-	if (why != PR_JOBCONTROL && why != PR_CHECKPOINT) {
+	if (why != PR_JOBCONTROL && why != PR_CHECKPOINT && why != PR_BRAND) {
 		/*
 		 * Do process-level notification when all lwps are
 		 * either stopped on events of interest to /proc
@@ -1155,6 +1196,13 @@ stop(int why, int what)
 	if (why == PR_CHECKPOINT)
 		del_one_utstop();
 
+	/*
+	 * Allow the brand to post notification of this stop condition.
+	 */
+	if (PROC_IS_BRANDED(p) && BROP(p)->b_stop_notify != NULL) {
+		BROP(p)->b_stop_notify(p, lwp, why, what);
+	}
+
 	thread_lock(t);
 	ASSERT((t->t_schedflag & TS_ALLSTART) == 0);
 	t->t_schedflag |= flags;
@@ -1176,7 +1224,7 @@ stop(int why, int what)
 		    (p->p_flag & (SEXITLWPS|SKILLED))) {
 			p->p_stopsig = 0;
 			thread_lock(t);
-			t->t_schedflag |= TS_XSTART | TS_PSTART;
+			t->t_schedflag |= TS_XSTART | TS_PSTART | TS_BSTART;
 			setrun_locked(t);
 			thread_unlock_nopreempt(t);
 		} else if (why == PR_JOBCONTROL) {
@@ -1311,7 +1359,7 @@ psig(void)
 	 * this signal from pending to current (we dropped p->p_lock).
 	 * This can happen only in a multi-threaded process.
 	 */
-	if (sigismember(&p->p_ignore, sig) ||
+	if (sig_ignorable(p, sig) ||
 	    (func == SIG_DFL && sigismember(&stopdefault, sig))) {
 		lwp->lwp_cursig = 0;
 		lwp->lwp_extsig = 0;
@@ -1413,6 +1461,9 @@ psig(void)
 		DTRACE_PROC3(signal__handle, int, sig, k_siginfo_t *,
 		    sip, void (*)(void), func);
 
+		if (PROC_IS_BRANDED(p) && BROP(p)->b_psig_to_proc)
+			BROP(p)->b_psig_to_proc(p, t, sig);
+
 		lwp->lwp_cursig = 0;
 		lwp->lwp_extsig = 0;
 		if (lwp->lwp_curinfo) {
@@ -1755,9 +1806,12 @@ post_sigcld(proc_t *cp, sigqueue_t *sqp)
 			/*
 			 * This can only happen when the parent is init.
 			 * (See call to sigcld(q, NULL) in exit().)
-			 * Use KM_NOSLEEP to avoid deadlock.
+			 * Use KM_NOSLEEP to avoid deadlock. The child procs
+			 * initpid can be 1 for zlogin.
 			 */
-			ASSERT(pp == proc_init);
+			ASSERT(pp->p_pidp->pid_id ==
+			    cp->p_zone->zone_proc_initpid ||
+			    pp->p_pidp->pid_id == 1);
 			winfo(cp, &info, 0);
 			sigaddq(pp, NULL, &info, KM_NOSLEEP);
 		} else {
@@ -1788,6 +1842,15 @@ sigcld_repost()
 
 	sqp = kmem_zalloc(sizeof (sigqueue_t), KM_SLEEP);
 	mutex_enter(&pidlock);
+	if (PROC_IS_BRANDED(pp) && BROP(pp)->b_sigcld_repost != NULL) {
+		/*
+		 * Allow the brand to inject synthetic SIGCLD signals.
+		 */
+		if (BROP(pp)->b_sigcld_repost(pp, sqp) == 0) {
+			mutex_exit(&pidlock);
+			return;
+		}
+	}
 	for (cp = pp->p_child; cp; cp = cp->p_sibling) {
 		if (cp->p_pidflag & CLDPEND) {
 			post_sigcld(cp, sqp);
diff --git a/usr/src/uts/common/os/smb_subr.c b/usr/src/uts/common/os/smb_subr.c
index 6084676b17..6dc7230bed 100644
--- a/usr/src/uts/common/os/smb_subr.c
+++ b/usr/src/uts/common/os/smb_subr.c
@@ -25,7 +25,9 @@
  * Use is subject to license terms.
  */
 
-#pragma ident	"%Z%%M%	%I%	%E% SMI"
+/*
+ * Copyright (c) 2015 Joyent, Inc.  All rights reserved.
+ */
 
 #include <sys/smbios_impl.h>
 #include <sys/cmn_err.h>
@@ -43,13 +45,13 @@ smb_strerror(int err)
 void *
 smb_alloc(size_t len)
 {
-	return (kmem_alloc(len, KM_SLEEP));
+	return (len > 0 ? kmem_alloc(len, KM_SLEEP) : NULL);
 }
 
 void *
 smb_zalloc(size_t len)
 {
-	return (kmem_zalloc(len, KM_SLEEP));
+	return (len > 0 ? kmem_zalloc(len, KM_SLEEP) : NULL);
 }
 
 void
diff --git a/usr/src/uts/common/os/streamio.c b/usr/src/uts/common/os/streamio.c
index c6ebe8b110..0f3aafcc42 100644
--- a/usr/src/uts/common/os/streamio.c
+++ b/usr/src/uts/common/os/streamio.c
@@ -24,6 +24,7 @@
 
 /*
  * Copyright (c) 1988, 2010, Oracle and/or its affiliates. All rights reserved.
+ * Copyright 2015, Joyent, Inc. All rights reserved.
  */
 
 #include <sys/types.h>
@@ -76,6 +77,7 @@
 #include <sys/policy.h>
 #include <sys/dld.h>
 #include <sys/zone.h>
+#include <sys/limits.h>
 #include <c2/audit.h>
 
 /*
@@ -984,12 +986,20 @@ strget(struct stdata *stp, queue_t *q, struct uio *uiop, int first,
 		 * (registered in sd_wakeq).
 		 */
 		struiod_t uiod;
+		struct iovec buf[IOV_MAX_STACK];
+		int iovlen = 0;
 
 		if (first)
 			stp->sd_wakeq &= ~RSLEEP;
 
-		(void) uiodup(uiop, &uiod.d_uio, uiod.d_iov,
-		    sizeof (uiod.d_iov) / sizeof (*uiod.d_iov));
+		if (uiop->uio_iovcnt > IOV_MAX_STACK) {
+			iovlen = uiop->uio_iovcnt * sizeof (iovec_t);
+			uiod.d_iov = kmem_alloc(iovlen, KM_SLEEP);
+		} else {
+			uiod.d_iov = buf;
+		}
+
+		(void) uiodup(uiop, &uiod.d_uio, uiod.d_iov, uiop->uio_iovcnt);
 		uiod.d_mp = 0;
 		/*
 		 * Mark that a thread is in rwnext on the read side
@@ -1028,6 +1038,8 @@ strget(struct stdata *stp, queue_t *q, struct uio *uiop, int first,
 			if ((bp = uiod.d_mp) != NULL) {
 				*errorp = 0;
 				ASSERT(MUTEX_HELD(&stp->sd_lock));
+				if (iovlen != 0)
+					kmem_free(uiod.d_iov, iovlen);
 				return (bp);
 			}
 			error = 0;
@@ -1047,8 +1059,14 @@ strget(struct stdata *stp, queue_t *q, struct uio *uiop, int first,
 		} else {
 			*errorp = error;
 			ASSERT(MUTEX_HELD(&stp->sd_lock));
+			if (iovlen != 0)
+				kmem_free(uiod.d_iov, iovlen);
 			return (NULL);
 		}
+
+		if (iovlen != 0)
+			kmem_free(uiod.d_iov, iovlen);
+
 		/*
 		 * Try a getq in case a rwnext() generated mblk
 		 * has bubbled up via strrput().
@@ -2543,6 +2561,8 @@ strput(struct stdata *stp, mblk_t *mctl, struct uio *uiop, ssize_t *iosize,
     int b_flag, int pri, int flags)
 {
 	struiod_t uiod;
+	struct iovec buf[IOV_MAX_STACK];
+	int iovlen = 0;
 	mblk_t *mp;
 	queue_t *wqp = stp->sd_wrq;
 	int error = 0;
@@ -2634,13 +2654,21 @@ strput(struct stdata *stp, mblk_t *mctl, struct uio *uiop, ssize_t *iosize,
 	mp->b_flag |= b_flag;
 	mp->b_band = (uchar_t)pri;
 
-	(void) uiodup(uiop, &uiod.d_uio, uiod.d_iov,
-	    sizeof (uiod.d_iov) / sizeof (*uiod.d_iov));
+	if (uiop->uio_iovcnt > IOV_MAX_STACK) {
+		iovlen = uiop->uio_iovcnt * sizeof (iovec_t);
+		uiod.d_iov = (struct iovec *)kmem_alloc(iovlen, KM_SLEEP);
+	} else {
+		uiod.d_iov = buf;
+	}
+
+	(void) uiodup(uiop, &uiod.d_uio, uiod.d_iov, uiop->uio_iovcnt);
 	uiod.d_uio.uio_offset = 0;
 	uiod.d_mp = mp;
 	error = rwnext(wqp, &uiod);
 	if (! uiod.d_mp) {
 		uioskip(uiop, *iosize);
+		if (iovlen != 0)
+			kmem_free(uiod.d_iov, iovlen);
 		return (error);
 	}
 	ASSERT(mp == uiod.d_mp);
@@ -2658,17 +2686,23 @@ strput(struct stdata *stp, mblk_t *mctl, struct uio *uiop, ssize_t *iosize,
 		error = 0;
 	} else {
 		freemsg(mp);
+		if (iovlen != 0)
+			kmem_free(uiod.d_iov, iovlen);
 		return (error);
 	}
 	/* Have to check canput before consuming data from the uio */
 	if (pri == 0) {
 		if (!canputnext(wqp) && !(flags & MSG_IGNFLOW)) {
 			freemsg(mp);
+			if (iovlen != 0)
+				kmem_free(uiod.d_iov, iovlen);
 			return (EWOULDBLOCK);
 		}
 	} else {
 		if (!bcanputnext(wqp, pri) && !(flags & MSG_IGNFLOW)) {
 			freemsg(mp);
+			if (iovlen != 0)
+				kmem_free(uiod.d_iov, iovlen);
 			return (EWOULDBLOCK);
 		}
 	}
@@ -2676,6 +2710,8 @@ strput(struct stdata *stp, mblk_t *mctl, struct uio *uiop, ssize_t *iosize,
 	/* Copyin data from the uio */
 	if ((error = struioget(wqp, mp, &uiod, 0)) != 0) {
 		freemsg(mp);
+		if (iovlen != 0)
+			kmem_free(uiod.d_iov, iovlen);
 		return (error);
 	}
 	uioskip(uiop, *iosize);
@@ -2692,6 +2728,8 @@ strput(struct stdata *stp, mblk_t *mctl, struct uio *uiop, ssize_t *iosize,
 		putnext(wqp, mp);
 		stream_runservice(stp);
 	}
+	if (iovlen != 0)
+		kmem_free(uiod.d_iov, iovlen);
 	return (0);
 }
 
@@ -3177,6 +3215,7 @@ job_control_type(int cmd)
 	case JAGENT:	/* Obsolete */
 	case JTRUN:	/* Obsolete */
 	case JXTPROTO:	/* Obsolete */
+	case TIOCSETLD:
 		return (JCSETP);
 	}
 
@@ -8311,7 +8350,7 @@ chkrd:
 	}
 
 	*reventsp = (short)retevents;
-	if (retevents) {
+	if (retevents && !(events & POLLET)) {
 		if (headlocked)
 			mutex_exit(&stp->sd_lock);
 		return (0);
diff --git a/usr/src/uts/common/os/sysent.c b/usr/src/uts/common/os/sysent.c
index 0d1bb6a8a1..aa44ccf788 100644
--- a/usr/src/uts/common/os/sysent.c
+++ b/usr/src/uts/common/os/sysent.c
@@ -1093,18 +1093,20 @@ char **syscallnames;
 
 systrace_sysent_t *systrace_sysent;
 void (*systrace_probe)(dtrace_id_t, uintptr_t, uintptr_t,
-    uintptr_t, uintptr_t, uintptr_t, uintptr_t);
+    uintptr_t, uintptr_t, uintptr_t, uintptr_t, uintptr_t, uintptr_t);
 
 /*ARGSUSED*/
 void
 systrace_stub(dtrace_id_t id, uintptr_t arg0, uintptr_t arg1,
-    uintptr_t arg2, uintptr_t arg3, uintptr_t arg4, uintptr_t arg5)
+    uintptr_t arg2, uintptr_t arg3, uintptr_t arg4, uintptr_t arg5,
+    uintptr_t arg6, uintptr_t arg7)
 {}
 
 /*ARGSUSED*/
 int64_t
 dtrace_systrace_syscall(uintptr_t arg0, uintptr_t arg1, uintptr_t arg2,
-    uintptr_t arg3, uintptr_t arg4, uintptr_t arg5)
+    uintptr_t arg3, uintptr_t arg4, uintptr_t arg5, uintptr_t arg6,
+    uintptr_t arg7)
 {
 	systrace_sysent_t *sy = &systrace_sysent[curthread->t_sysnum];
 	dtrace_id_t id;
@@ -1112,7 +1114,8 @@ dtrace_systrace_syscall(uintptr_t arg0, uintptr_t arg1, uintptr_t arg2,
 	proc_t *p;
 
 	if ((id = sy->stsy_entry) != DTRACE_IDNONE)
-		(*systrace_probe)(id, arg0, arg1, arg2, arg3, arg4, arg5);
+		(*systrace_probe)(id, arg0, arg1, arg2, arg3, arg4, arg5,
+		    arg6, arg7);
 
 	/*
 	 * We want to explicitly allow DTrace consumers to stop a process
@@ -1126,14 +1129,15 @@ dtrace_systrace_syscall(uintptr_t arg0, uintptr_t arg1, uintptr_t arg2,
 	}
 	mutex_exit(&p->p_lock);
 
-	rval = (*sy->stsy_underlying)(arg0, arg1, arg2, arg3, arg4, arg5);
+	rval = (*sy->stsy_underlying)(arg0, arg1, arg2, arg3, arg4, arg5,
+	    arg6, arg7);
 
 	if (ttolwp(curthread)->lwp_errno != 0)
 		rval = -1;
 
 	if ((id = sy->stsy_return) != DTRACE_IDNONE)
 		(*systrace_probe)(id, (uintptr_t)rval, (uintptr_t)rval,
-		    (uintptr_t)((int64_t)rval >> 32), 0, 0, 0);
+		    (uintptr_t)((int64_t)rval >> 32), 0, 0, 0, 0, 0);
 
 	return (rval);
 }
@@ -1145,7 +1149,8 @@ systrace_sysent_t *systrace_sysent32;
 /*ARGSUSED*/
 int64_t
 dtrace_systrace_syscall32(uintptr_t arg0, uintptr_t arg1, uintptr_t arg2,
-    uintptr_t arg3, uintptr_t arg4, uintptr_t arg5)
+    uintptr_t arg3, uintptr_t arg4, uintptr_t arg5, uintptr_t arg6,
+    uintptr_t arg7)
 {
 	systrace_sysent_t *sy = &systrace_sysent32[curthread->t_sysnum];
 	dtrace_id_t id;
@@ -1153,7 +1158,8 @@ dtrace_systrace_syscall32(uintptr_t arg0, uintptr_t arg1, uintptr_t arg2,
 	proc_t *p;
 
 	if ((id = sy->stsy_entry) != DTRACE_IDNONE)
-		(*systrace_probe)(id, arg0, arg1, arg2, arg3, arg4, arg5);
+		(*systrace_probe)(id, arg0, arg1, arg2, arg3, arg4, arg5, arg6,
+		    arg7);
 
 	/*
 	 * We want to explicitly allow DTrace consumers to stop a process
@@ -1167,14 +1173,15 @@ dtrace_systrace_syscall32(uintptr_t arg0, uintptr_t arg1, uintptr_t arg2,
 	}
 	mutex_exit(&p->p_lock);
 
-	rval = (*sy->stsy_underlying)(arg0, arg1, arg2, arg3, arg4, arg5);
+	rval = (*sy->stsy_underlying)(arg0, arg1, arg2, arg3, arg4, arg5, arg6,
+	    arg7);
 
 	if (ttolwp(curthread)->lwp_errno != 0)
 		rval = -1;
 
 	if ((id = sy->stsy_return) != DTRACE_IDNONE)
 		(*systrace_probe)(id, (uintptr_t)rval, (uintptr_t)rval,
-		    (uintptr_t)((uint64_t)rval >> 32), 0, 0, 0);
+		    (uintptr_t)((uint64_t)rval >> 32), 0, 0, 0, 0, 0);
 
 	return (rval);
 }
@@ -1202,5 +1209,5 @@ dtrace_systrace_rtt(void)
 	}
 
 	if ((id = sy->stsy_return) != DTRACE_IDNONE)
-		(*systrace_probe)(id, 0, 0, 0, 0, 0, 0);
+		(*systrace_probe)(id, 0, 0, 0, 0, 0, 0, 0, 0);
 }
diff --git a/usr/src/uts/common/os/timer.c b/usr/src/uts/common/os/timer.c
index 8559d8736c..b25a6cbcf1 100644
--- a/usr/src/uts/common/os/timer.c
+++ b/usr/src/uts/common/os/timer.c
@@ -269,6 +269,15 @@ clock_add_backend(clockid_t clock, clock_backend_t *backend)
 	clock_backend[clock] = backend;
 }
 
+clock_backend_t *
+clock_get_backend(clockid_t clock)
+{
+	if (clock < 0 || clock >= CLOCK_MAX)
+		return (NULL);
+
+	return (clock_backend[clock]);
+}
+
 int
 clock_settime(clockid_t clock, timespec_t *tp)
 {
@@ -398,7 +407,7 @@ timer_signal(sigqueue_t *sigq)
 /*
  * This routine is called from the clock backend.
  */
-void
+static void
 timer_fire(itimer_t *it)
 {
 	proc_t *p;
@@ -672,7 +681,7 @@ timer_create(clockid_t clock, struct sigevent *evp, timer_t *tid)
 	 * Call on the backend to verify the event argument (or return
 	 * EINVAL if this clock type does not support timers).
 	 */
-	if ((error = backend->clk_timer_create(it, &ev)) != 0)
+	if ((error = backend->clk_timer_create(it, timer_fire)) != 0)
 		goto err;
 
 	it->it_lwp = ttolwp(curthread);
diff --git a/usr/src/uts/common/os/vmem.c b/usr/src/uts/common/os/vmem.c
index a554f8c3f3..0a6fe0ef96 100644
--- a/usr/src/uts/common/os/vmem.c
+++ b/usr/src/uts/common/os/vmem.c
@@ -1618,7 +1618,7 @@ vmem_destroy(vmem_t *vmp)
 
 	leaked = vmem_size(vmp, VMEM_ALLOC);
 	if (leaked != 0)
-		cmn_err(CE_WARN, "vmem_destroy('%s'): leaked %lu %s",
+		cmn_err(CE_WARN, "!vmem_destroy('%s'): leaked %lu %s",
 		    vmp->vm_name, leaked, (vmp->vm_cflags & VMC_IDENTIFIER) ?
 		    "identifiers" : "bytes");
 
diff --git a/usr/src/uts/common/os/zone.c b/usr/src/uts/common/os/zone.c
index 56c654331e..b097b351e6 100644
--- a/usr/src/uts/common/os/zone.c
+++ b/usr/src/uts/common/os/zone.c
@@ -21,7 +21,7 @@
 
 /*
  * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
- * Copyright 2013, Joyent Inc. All rights reserved.
+ * Copyright 2015, Joyent Inc. All rights reserved.
  */
 
 /*
@@ -370,8 +370,12 @@ static char *zone_ref_subsys_names[] = {
 rctl_hndl_t rc_zone_cpu_shares;
 rctl_hndl_t rc_zone_locked_mem;
 rctl_hndl_t rc_zone_max_swap;
+rctl_hndl_t rc_zone_phys_mem;
 rctl_hndl_t rc_zone_max_lofi;
 rctl_hndl_t rc_zone_cpu_cap;
+rctl_hndl_t rc_zone_cpu_baseline;
+rctl_hndl_t rc_zone_cpu_burst_time;
+rctl_hndl_t rc_zone_zfs_io_pri;
 rctl_hndl_t rc_zone_nlwps;
 rctl_hndl_t rc_zone_nprocs;
 rctl_hndl_t rc_zone_shmmax;
@@ -417,8 +421,9 @@ static boolean_t zsd_wait_for_inprogress(zone_t *, struct zsd_entry *,
  * Version 5 alters the zone_boot system call, and converts its old
  *     bootargs parameter to be set by the zone_setattr API instead.
  * Version 6 adds the flag argument to zone_create.
+ * Version 7 adds the requested zoneid to zone_create.
  */
-static const int ZONE_SYSCALL_API_VERSION = 6;
+static const int ZONE_SYSCALL_API_VERSION = 7;
 
 /*
  * Certain filesystems (such as NFS and autofs) need to know which zone
@@ -1377,6 +1382,114 @@ static rctl_ops_t zone_cpu_cap_ops = {
 
 /*ARGSUSED*/
 static rctl_qty_t
+zone_cpu_base_get(rctl_t *rctl, struct proc *p)
+{
+	ASSERT(MUTEX_HELD(&p->p_lock));
+	return (cpucaps_zone_get_base(p->p_zone));
+}
+
+/*
+ * The zone cpu base is used to set the baseline CPU for the zone
+ * so we can track when the zone is bursting.
+ */
+/*ARGSUSED*/
+static int
+zone_cpu_base_set(rctl_t *rctl, struct proc *p, rctl_entity_p_t *e,
+    rctl_qty_t nv)
+{
+	zone_t *zone = e->rcep_p.zone;
+
+	ASSERT(MUTEX_HELD(&p->p_lock));
+	ASSERT(e->rcep_t == RCENTITY_ZONE);
+
+	if (zone == NULL)
+		return (0);
+
+	return (cpucaps_zone_set_base(zone, nv));
+}
+
+static rctl_ops_t zone_cpu_base_ops = {
+	rcop_no_action,
+	zone_cpu_base_get,
+	zone_cpu_base_set,
+	rcop_no_test
+};
+
+/*ARGSUSED*/
+static rctl_qty_t
+zone_cpu_burst_time_get(rctl_t *rctl, struct proc *p)
+{
+	ASSERT(MUTEX_HELD(&p->p_lock));
+	return (cpucaps_zone_get_burst_time(p->p_zone));
+}
+
+/*
+ * The zone cpu burst time is used to set the amount of time CPU(s) can be
+ * bursting for the zone.
+ */
+/*ARGSUSED*/
+static int
+zone_cpu_burst_time_set(rctl_t *rctl, struct proc *p, rctl_entity_p_t *e,
+    rctl_qty_t nv)
+{
+	zone_t *zone = e->rcep_p.zone;
+
+	ASSERT(MUTEX_HELD(&p->p_lock));
+	ASSERT(e->rcep_t == RCENTITY_ZONE);
+
+	if (zone == NULL)
+		return (0);
+
+	return (cpucaps_zone_set_burst_time(zone, nv));
+}
+
+static rctl_ops_t zone_cpu_burst_time_ops = {
+	rcop_no_action,
+	zone_cpu_burst_time_get,
+	zone_cpu_burst_time_set,
+	rcop_no_test
+};
+
+/*
+ * zone.zfs-io-pri resource control support (IO priority).
+ */
+/*ARGSUSED*/
+static rctl_qty_t
+zone_zfs_io_pri_get(rctl_t *rctl, struct proc *p)
+{
+	ASSERT(MUTEX_HELD(&p->p_lock));
+	return (p->p_zone->zone_zfs_io_pri);
+}
+
+/*ARGSUSED*/
+static int
+zone_zfs_io_pri_set(rctl_t *rctl, struct proc *p, rctl_entity_p_t *e,
+    rctl_qty_t nv)
+{
+	zone_t *zone = e->rcep_p.zone;
+
+	ASSERT(MUTEX_HELD(&p->p_lock));
+	ASSERT(e->rcep_t == RCENTITY_ZONE);
+
+	if (zone == NULL)
+		return (0);
+
+	/*
+	 * set priority to the new value.
+	 */
+	zone->zone_zfs_io_pri = nv;
+	return (0);
+}
+
+static rctl_ops_t zone_zfs_io_pri_ops = {
+	rcop_no_action,
+	zone_zfs_io_pri_get,
+	zone_zfs_io_pri_set,
+	rcop_no_test
+};
+
+/*ARGSUSED*/
+static rctl_qty_t
 zone_lwps_usage(rctl_t *r, proc_t *p)
 {
 	rctl_qty_t nlwps;
@@ -1671,6 +1784,39 @@ static rctl_ops_t zone_max_swap_ops = {
 
 /*ARGSUSED*/
 static rctl_qty_t
+zone_phys_mem_usage(rctl_t *rctl, struct proc *p)
+{
+	rctl_qty_t q;
+	zone_t *z = p->p_zone;
+
+	ASSERT(MUTEX_HELD(&p->p_lock));
+	/* No additional lock because not enforced in the kernel */
+	q = z->zone_phys_mem;
+	return (q);
+}
+
+/*ARGSUSED*/
+static int
+zone_phys_mem_set(rctl_t *rctl, struct proc *p, rctl_entity_p_t *e,
+    rctl_qty_t nv)
+{
+	ASSERT(MUTEX_HELD(&p->p_lock));
+	ASSERT(e->rcep_t == RCENTITY_ZONE);
+	if (e->rcep_p.zone == NULL)
+		return (0);
+	e->rcep_p.zone->zone_phys_mem_ctl = nv;
+	return (0);
+}
+
+static rctl_ops_t zone_phys_mem_ops = {
+	rcop_no_action,
+	zone_phys_mem_usage,
+	zone_phys_mem_set,
+	rcop_no_test
+};
+
+/*ARGSUSED*/
+static rctl_qty_t
 zone_max_lofi_usage(rctl_t *rctl, struct proc *p)
 {
 	rctl_qty_t q;
@@ -1764,6 +1910,20 @@ zone_lockedmem_kstat_update(kstat_t *ksp, int rw)
 }
 
 static int
+zone_physmem_kstat_update(kstat_t *ksp, int rw)
+{
+	zone_t *zone = ksp->ks_private;
+	zone_kstat_t *zk = ksp->ks_data;
+
+	if (rw == KSTAT_WRITE)
+		return (EACCES);
+
+	zk->zk_usage.value.ui64 = zone->zone_phys_mem;
+	zk->zk_value.value.ui64 = zone->zone_phys_mem_ctl;
+	return (0);
+}
+
+static int
 zone_nprocs_kstat_update(kstat_t *ksp, int rw)
 {
 	zone_t *zone = ksp->ks_private;
@@ -1792,7 +1952,7 @@ zone_swapresv_kstat_update(kstat_t *ksp, int rw)
 }
 
 static kstat_t *
-zone_kstat_create_common(zone_t *zone, char *name,
+zone_rctl_kstat_create_common(zone_t *zone, char *name,
     int (*updatefunc) (kstat_t *, int))
 {
 	kstat_t *ksp;
@@ -1818,6 +1978,230 @@ zone_kstat_create_common(zone_t *zone, char *name,
 }
 
 static int
+zone_vfs_kstat_update(kstat_t *ksp, int rw)
+{
+	zone_t *zone = ksp->ks_private;
+	zone_vfs_kstat_t *zvp = ksp->ks_data;
+	kstat_io_t *kiop = &zone->zone_vfs_rwstats;
+
+	if (rw == KSTAT_WRITE)
+		return (EACCES);
+
+	/*
+	 * Extract the VFS statistics from the kstat_io_t structure used by
+	 * kstat_runq_enter() and related functions.  Since the slow ops
+	 * counters are updated directly by the VFS layer, there's no need to
+	 * copy those statistics here.
+	 *
+	 * Note that kstat_runq_enter() and the related functions use
+	 * gethrtime_unscaled(), so scale the time here.
+	 */
+	zvp->zv_nread.value.ui64 = kiop->nread;
+	zvp->zv_reads.value.ui64 = kiop->reads;
+	zvp->zv_rtime.value.ui64 = kiop->rtime;
+	zvp->zv_rlentime.value.ui64 = kiop->rlentime;
+	zvp->zv_nwritten.value.ui64 = kiop->nwritten;
+	zvp->zv_writes.value.ui64 = kiop->writes;
+	zvp->zv_wtime.value.ui64 = kiop->wtime;
+	zvp->zv_wlentime.value.ui64 = kiop->wlentime;
+
+	scalehrtime((hrtime_t *)&zvp->zv_rtime.value.ui64);
+	scalehrtime((hrtime_t *)&zvp->zv_rlentime.value.ui64);
+	scalehrtime((hrtime_t *)&zvp->zv_wtime.value.ui64);
+	scalehrtime((hrtime_t *)&zvp->zv_wlentime.value.ui64);
+
+	return (0);
+}
+
+static kstat_t *
+zone_vfs_kstat_create(zone_t *zone)
+{
+	kstat_t *ksp;
+	zone_vfs_kstat_t *zvp;
+
+	if ((ksp = kstat_create_zone("zone_vfs", zone->zone_id,
+	    zone->zone_name, "zone_vfs", KSTAT_TYPE_NAMED,
+	    sizeof (zone_vfs_kstat_t) / sizeof (kstat_named_t),
+	    KSTAT_FLAG_VIRTUAL, zone->zone_id)) == NULL)
+		return (NULL);
+
+	if (zone->zone_id != GLOBAL_ZONEID)
+		kstat_zone_add(ksp, GLOBAL_ZONEID);
+
+	zvp = ksp->ks_data = kmem_zalloc(sizeof (zone_vfs_kstat_t), KM_SLEEP);
+	ksp->ks_data_size += strlen(zone->zone_name) + 1;
+	ksp->ks_lock = &zone->zone_vfs_lock;
+	zone->zone_vfs_stats = zvp;
+
+	/* The kstat "name" field is not large enough for a full zonename */
+	kstat_named_init(&zvp->zv_zonename, "zonename", KSTAT_DATA_STRING);
+	kstat_named_setstr(&zvp->zv_zonename, zone->zone_name);
+	kstat_named_init(&zvp->zv_nread, "nread", KSTAT_DATA_UINT64);
+	kstat_named_init(&zvp->zv_reads, "reads", KSTAT_DATA_UINT64);
+	kstat_named_init(&zvp->zv_rtime, "rtime", KSTAT_DATA_UINT64);
+	kstat_named_init(&zvp->zv_rlentime, "rlentime", KSTAT_DATA_UINT64);
+	kstat_named_init(&zvp->zv_nwritten, "nwritten", KSTAT_DATA_UINT64);
+	kstat_named_init(&zvp->zv_writes, "writes", KSTAT_DATA_UINT64);
+	kstat_named_init(&zvp->zv_wtime, "wtime", KSTAT_DATA_UINT64);
+	kstat_named_init(&zvp->zv_wlentime, "wlentime", KSTAT_DATA_UINT64);
+	kstat_named_init(&zvp->zv_10ms_ops, "10ms_ops", KSTAT_DATA_UINT64);
+	kstat_named_init(&zvp->zv_100ms_ops, "100ms_ops", KSTAT_DATA_UINT64);
+	kstat_named_init(&zvp->zv_1s_ops, "1s_ops", KSTAT_DATA_UINT64);
+	kstat_named_init(&zvp->zv_10s_ops, "10s_ops", KSTAT_DATA_UINT64);
+	kstat_named_init(&zvp->zv_delay_cnt, "delay_cnt", KSTAT_DATA_UINT64);
+	kstat_named_init(&zvp->zv_delay_time, "delay_time", KSTAT_DATA_UINT64);
+
+	ksp->ks_update = zone_vfs_kstat_update;
+	ksp->ks_private = zone;
+
+	kstat_install(ksp);
+	return (ksp);
+}
+
+static int
+zone_zfs_kstat_update(kstat_t *ksp, int rw)
+{
+	zone_t *zone = ksp->ks_private;
+	zone_zfs_kstat_t *zzp = ksp->ks_data;
+	kstat_io_t *kiop = &zone->zone_zfs_rwstats;
+
+	if (rw == KSTAT_WRITE)
+		return (EACCES);
+
+	/*
+	 * Extract the ZFS statistics from the kstat_io_t structure used by
+	 * kstat_runq_enter() and related functions.  Since the I/O throttle
+	 * counters are updated directly by the ZFS layer, there's no need to
+	 * copy those statistics here.
+	 *
+	 * Note that kstat_runq_enter() and the related functions use
+	 * gethrtime_unscaled(), so scale the time here.
+	 */
+	zzp->zz_nread.value.ui64 = kiop->nread;
+	zzp->zz_reads.value.ui64 = kiop->reads;
+	zzp->zz_rtime.value.ui64 = kiop->rtime;
+	zzp->zz_rlentime.value.ui64 = kiop->rlentime;
+	zzp->zz_nwritten.value.ui64 = kiop->nwritten;
+	zzp->zz_writes.value.ui64 = kiop->writes;
+
+	scalehrtime((hrtime_t *)&zzp->zz_rtime.value.ui64);
+	scalehrtime((hrtime_t *)&zzp->zz_rlentime.value.ui64);
+
+	return (0);
+}
+
+static kstat_t *
+zone_zfs_kstat_create(zone_t *zone)
+{
+	kstat_t *ksp;
+	zone_zfs_kstat_t *zzp;
+
+	if ((ksp = kstat_create_zone("zone_zfs", zone->zone_id,
+	    zone->zone_name, "zone_zfs", KSTAT_TYPE_NAMED,
+	    sizeof (zone_zfs_kstat_t) / sizeof (kstat_named_t),
+	    KSTAT_FLAG_VIRTUAL, zone->zone_id)) == NULL)
+		return (NULL);
+
+	if (zone->zone_id != GLOBAL_ZONEID)
+		kstat_zone_add(ksp, GLOBAL_ZONEID);
+
+	zzp = ksp->ks_data = kmem_zalloc(sizeof (zone_zfs_kstat_t), KM_SLEEP);
+	ksp->ks_data_size += strlen(zone->zone_name) + 1;
+	ksp->ks_lock = &zone->zone_zfs_lock;
+	zone->zone_zfs_stats = zzp;
+
+	/* The kstat "name" field is not large enough for a full zonename */
+	kstat_named_init(&zzp->zz_zonename, "zonename", KSTAT_DATA_STRING);
+	kstat_named_setstr(&zzp->zz_zonename, zone->zone_name);
+	kstat_named_init(&zzp->zz_nread, "nread", KSTAT_DATA_UINT64);
+	kstat_named_init(&zzp->zz_reads, "reads", KSTAT_DATA_UINT64);
+	kstat_named_init(&zzp->zz_rtime, "rtime", KSTAT_DATA_UINT64);
+	kstat_named_init(&zzp->zz_rlentime, "rlentime", KSTAT_DATA_UINT64);
+	kstat_named_init(&zzp->zz_nwritten, "nwritten", KSTAT_DATA_UINT64);
+	kstat_named_init(&zzp->zz_writes, "writes", KSTAT_DATA_UINT64);
+	kstat_named_init(&zzp->zz_waittime, "waittime", KSTAT_DATA_UINT64);
+
+	ksp->ks_update = zone_zfs_kstat_update;
+	ksp->ks_private = zone;
+
+	kstat_install(ksp);
+	return (ksp);
+}
+
+static int
+zone_mcap_kstat_update(kstat_t *ksp, int rw)
+{
+	zone_t *zone = ksp->ks_private;
+	zone_mcap_kstat_t *zmp = ksp->ks_data;
+
+	if (rw == KSTAT_WRITE)
+		return (EACCES);
+
+	zmp->zm_rss.value.ui64 = zone->zone_phys_mem;
+	zmp->zm_phys_cap.value.ui64 = zone->zone_phys_mem_ctl;
+	zmp->zm_swap.value.ui64 = zone->zone_max_swap;
+	zmp->zm_swap_cap.value.ui64 = zone->zone_max_swap_ctl;
+	zmp->zm_nover.value.ui64 = zone->zone_mcap_nover;
+	zmp->zm_pagedout.value.ui64 = zone->zone_mcap_pagedout;
+	zmp->zm_pgpgin.value.ui64 = zone->zone_pgpgin;
+	zmp->zm_anonpgin.value.ui64 = zone->zone_anonpgin;
+	zmp->zm_execpgin.value.ui64 = zone->zone_execpgin;
+	zmp->zm_fspgin.value.ui64 = zone->zone_fspgin;
+	zmp->zm_anon_alloc_fail.value.ui64 = zone->zone_anon_alloc_fail;
+	zmp->zm_pf_throttle.value.ui64 = zone->zone_pf_throttle;
+	zmp->zm_pf_throttle_usec.value.ui64 = zone->zone_pf_throttle_usec;
+
+	return (0);
+}
+
+static kstat_t *
+zone_mcap_kstat_create(zone_t *zone)
+{
+	kstat_t *ksp;
+	zone_mcap_kstat_t *zmp;
+
+	if ((ksp = kstat_create_zone("memory_cap", zone->zone_id,
+	    zone->zone_name, "zone_memory_cap", KSTAT_TYPE_NAMED,
+	    sizeof (zone_mcap_kstat_t) / sizeof (kstat_named_t),
+	    KSTAT_FLAG_VIRTUAL, zone->zone_id)) == NULL)
+		return (NULL);
+
+	if (zone->zone_id != GLOBAL_ZONEID)
+		kstat_zone_add(ksp, GLOBAL_ZONEID);
+
+	zmp = ksp->ks_data = kmem_zalloc(sizeof (zone_mcap_kstat_t), KM_SLEEP);
+	ksp->ks_data_size += strlen(zone->zone_name) + 1;
+	ksp->ks_lock = &zone->zone_mcap_lock;
+	zone->zone_mcap_stats = zmp;
+
+	/* The kstat "name" field is not large enough for a full zonename */
+	kstat_named_init(&zmp->zm_zonename, "zonename", KSTAT_DATA_STRING);
+	kstat_named_setstr(&zmp->zm_zonename, zone->zone_name);
+	kstat_named_init(&zmp->zm_rss, "rss", KSTAT_DATA_UINT64);
+	kstat_named_init(&zmp->zm_phys_cap, "physcap", KSTAT_DATA_UINT64);
+	kstat_named_init(&zmp->zm_swap, "swap", KSTAT_DATA_UINT64);
+	kstat_named_init(&zmp->zm_swap_cap, "swapcap", KSTAT_DATA_UINT64);
+	kstat_named_init(&zmp->zm_nover, "nover", KSTAT_DATA_UINT64);
+	kstat_named_init(&zmp->zm_pagedout, "pagedout", KSTAT_DATA_UINT64);
+	kstat_named_init(&zmp->zm_pgpgin, "pgpgin", KSTAT_DATA_UINT64);
+	kstat_named_init(&zmp->zm_anonpgin, "anonpgin", KSTAT_DATA_UINT64);
+	kstat_named_init(&zmp->zm_execpgin, "execpgin", KSTAT_DATA_UINT64);
+	kstat_named_init(&zmp->zm_fspgin, "fspgin", KSTAT_DATA_UINT64);
+	kstat_named_init(&zmp->zm_anon_alloc_fail, "anon_alloc_fail",
+	    KSTAT_DATA_UINT64);
+	kstat_named_init(&zmp->zm_pf_throttle, "n_pf_throttle",
+	    KSTAT_DATA_UINT64);
+	kstat_named_init(&zmp->zm_pf_throttle_usec, "n_pf_throttle_usec",
+	    KSTAT_DATA_UINT64);
+
+	ksp->ks_update = zone_mcap_kstat_update;
+	ksp->ks_private = zone;
+
+	kstat_install(ksp);
+	return (ksp);
+}
+
+static int
 zone_misc_kstat_update(kstat_t *ksp, int rw)
 {
 	zone_t *zone = ksp->ks_private;
@@ -1846,6 +2230,9 @@ zone_misc_kstat_update(kstat_t *ksp, int rw)
 	zmp->zm_ffnomem.value.ui32 = zone->zone_ffnomem;
 	zmp->zm_ffmisc.value.ui32 = zone->zone_ffmisc;
 
+	zmp->zm_init_pid.value.ui32 = zone->zone_proc_initpid;
+	zmp->zm_boot_time.value.ui64 = (uint64_t)zone->zone_boot_time;
+
 	return (0);
 }
 
@@ -1884,7 +2271,8 @@ zone_misc_kstat_create(zone_t *zone)
 	    KSTAT_DATA_UINT32);
 	kstat_named_init(&zmp->zm_ffnomem, "forkfail_nomem", KSTAT_DATA_UINT32);
 	kstat_named_init(&zmp->zm_ffmisc, "forkfail_misc", KSTAT_DATA_UINT32);
-
+	kstat_named_init(&zmp->zm_init_pid, "init_pid", KSTAT_DATA_UINT32);
+	kstat_named_init(&zmp->zm_boot_time, "boot_time", KSTAT_DATA_UINT64);
 
 	ksp->ks_update = zone_misc_kstat_update;
 	ksp->ks_private = zone;
@@ -1896,13 +2284,30 @@ zone_misc_kstat_create(zone_t *zone)
 static void
 zone_kstat_create(zone_t *zone)
 {
-	zone->zone_lockedmem_kstat = zone_kstat_create_common(zone,
+	zone->zone_lockedmem_kstat = zone_rctl_kstat_create_common(zone,
 	    "lockedmem", zone_lockedmem_kstat_update);
-	zone->zone_swapresv_kstat = zone_kstat_create_common(zone,
+	zone->zone_swapresv_kstat = zone_rctl_kstat_create_common(zone,
 	    "swapresv", zone_swapresv_kstat_update);
-	zone->zone_nprocs_kstat = zone_kstat_create_common(zone,
+	zone->zone_physmem_kstat = zone_rctl_kstat_create_common(zone,
+	    "physicalmem", zone_physmem_kstat_update);
+	zone->zone_nprocs_kstat = zone_rctl_kstat_create_common(zone,
 	    "nprocs", zone_nprocs_kstat_update);
 
+	if ((zone->zone_vfs_ksp = zone_vfs_kstat_create(zone)) == NULL) {
+		zone->zone_vfs_stats = kmem_zalloc(
+		    sizeof (zone_vfs_kstat_t), KM_SLEEP);
+	}
+
+	if ((zone->zone_zfs_ksp = zone_zfs_kstat_create(zone)) == NULL) {
+		zone->zone_zfs_stats = kmem_zalloc(
+		    sizeof (zone_zfs_kstat_t), KM_SLEEP);
+	}
+
+	if ((zone->zone_mcap_ksp = zone_mcap_kstat_create(zone)) == NULL) {
+		zone->zone_mcap_stats = kmem_zalloc(
+		    sizeof (zone_mcap_kstat_t), KM_SLEEP);
+	}
+
 	if ((zone->zone_misc_ksp = zone_misc_kstat_create(zone)) == NULL) {
 		zone->zone_misc_stats = kmem_zalloc(
 		    sizeof (zone_misc_kstat_t), KM_SLEEP);
@@ -1929,8 +2334,17 @@ zone_kstat_delete(zone_t *zone)
 	    sizeof (zone_kstat_t));
 	zone_kstat_delete_common(&zone->zone_swapresv_kstat,
 	    sizeof (zone_kstat_t));
+	zone_kstat_delete_common(&zone->zone_physmem_kstat,
+	    sizeof (zone_kstat_t));
 	zone_kstat_delete_common(&zone->zone_nprocs_kstat,
 	    sizeof (zone_kstat_t));
+
+	zone_kstat_delete_common(&zone->zone_vfs_ksp,
+	    sizeof (zone_vfs_kstat_t));
+	zone_kstat_delete_common(&zone->zone_zfs_ksp,
+	    sizeof (zone_zfs_kstat_t));
+	zone_kstat_delete_common(&zone->zone_mcap_ksp,
+	    sizeof (zone_mcap_kstat_t));
 	zone_kstat_delete_common(&zone->zone_misc_ksp,
 	    sizeof (zone_misc_kstat_t));
 }
@@ -1966,6 +2380,8 @@ zone_zsd_init(void)
 	zone0.zone_locked_mem_ctl = UINT64_MAX;
 	ASSERT(zone0.zone_max_swap == 0);
 	zone0.zone_max_swap_ctl = UINT64_MAX;
+	zone0.zone_phys_mem = 0;
+	zone0.zone_phys_mem_ctl = UINT64_MAX;
 	zone0.zone_max_lofi = 0;
 	zone0.zone_max_lofi_ctl = UINT64_MAX;
 	zone0.zone_shmmax = 0;
@@ -1989,8 +2405,9 @@ zone_zsd_init(void)
 	zone0.zone_initname = initname;
 	zone0.zone_lockedmem_kstat = NULL;
 	zone0.zone_swapresv_kstat = NULL;
+	zone0.zone_physmem_kstat = NULL;
 	zone0.zone_nprocs_kstat = NULL;
-
+	zone0.zone_zfs_io_pri = 1;
 	zone0.zone_stime = 0;
 	zone0.zone_utime = 0;
 	zone0.zone_wtime = 0;
@@ -2101,6 +2518,21 @@ zone_init(void)
 	    RCTL_GLOBAL_INFINITE,
 	    MAXCAP, MAXCAP, &zone_cpu_cap_ops);
 
+	rc_zone_cpu_baseline = rctl_register("zone.cpu-baseline",
+	    RCENTITY_ZONE, RCTL_GLOBAL_SIGNAL_NEVER | RCTL_GLOBAL_DENY_NEVER |
+	    RCTL_GLOBAL_NOBASIC | RCTL_GLOBAL_COUNT | RCTL_GLOBAL_SYSLOG_NEVER,
+	    MAXCAP, MAXCAP, &zone_cpu_base_ops);
+
+	rc_zone_cpu_burst_time = rctl_register("zone.cpu-burst-time",
+	    RCENTITY_ZONE, RCTL_GLOBAL_SIGNAL_NEVER | RCTL_GLOBAL_DENY_NEVER |
+	    RCTL_GLOBAL_NOBASIC | RCTL_GLOBAL_COUNT | RCTL_GLOBAL_SYSLOG_NEVER,
+	    INT_MAX, INT_MAX, &zone_cpu_burst_time_ops);
+
+	rc_zone_zfs_io_pri = rctl_register("zone.zfs-io-priority",
+	    RCENTITY_ZONE, RCTL_GLOBAL_SIGNAL_NEVER | RCTL_GLOBAL_DENY_NEVER |
+	    RCTL_GLOBAL_NOBASIC | RCTL_GLOBAL_COUNT | RCTL_GLOBAL_SYSLOG_NEVER,
+	    16384, 16384, &zone_zfs_io_pri_ops);
+
 	rc_zone_nlwps = rctl_register("zone.max-lwps", RCENTITY_ZONE,
 	    RCTL_GLOBAL_NOACTION | RCTL_GLOBAL_NOBASIC | RCTL_GLOBAL_COUNT,
 	    INT_MAX, INT_MAX, &zone_lwps_ops);
@@ -2142,6 +2574,20 @@ zone_init(void)
 	rde = rctl_dict_lookup("zone.cpu-shares");
 	(void) rctl_val_list_insert(&rde->rcd_default_value, dval);
 
+	/*
+	 * Create a rctl_val with PRIVILEGED, NOACTION, value = 1.  Then attach
+	 * this at the head of the rctl_dict_entry for ``zone.zfs-io-priority'.
+	 */
+	dval = kmem_cache_alloc(rctl_val_cache, KM_SLEEP);
+	bzero(dval, sizeof (rctl_val_t));
+	dval->rcv_value = 1;
+	dval->rcv_privilege = RCPRIV_PRIVILEGED;
+	dval->rcv_flagaction = RCTL_LOCAL_NOACTION;
+	dval->rcv_action_recip_pid = -1;
+
+	rde = rctl_dict_lookup("zone.zfs-io-priority");
+	(void) rctl_val_list_insert(&rde->rcd_default_value, dval);
+
 	rc_zone_locked_mem = rctl_register("zone.max-locked-memory",
 	    RCENTITY_ZONE, RCTL_GLOBAL_NOBASIC | RCTL_GLOBAL_BYTES |
 	    RCTL_GLOBAL_DENY_ALWAYS, UINT64_MAX, UINT64_MAX,
@@ -2152,6 +2598,11 @@ zone_init(void)
 	    RCTL_GLOBAL_DENY_ALWAYS, UINT64_MAX, UINT64_MAX,
 	    &zone_max_swap_ops);
 
+	rc_zone_phys_mem = rctl_register("zone.max-physical-memory",
+	    RCENTITY_ZONE, RCTL_GLOBAL_NOBASIC | RCTL_GLOBAL_BYTES |
+	    RCTL_GLOBAL_DENY_ALWAYS, UINT64_MAX, UINT64_MAX,
+	    &zone_phys_mem_ops);
+
 	rc_zone_max_lofi = rctl_register("zone.max-lofi",
 	    RCENTITY_ZONE, RCTL_GLOBAL_NOBASIC | RCTL_GLOBAL_COUNT |
 	    RCTL_GLOBAL_DENY_ALWAYS, UINT64_MAX, UINT64_MAX,
@@ -2173,6 +2624,8 @@ zone_init(void)
 	zone0.zone_ntasks = 1;
 	mutex_exit(&p0.p_lock);
 	zone0.zone_restart_init = B_TRUE;
+	zone0.zone_reboot_on_init_exit = B_FALSE;
+	zone0.zone_init_status = -1;
 	zone0.zone_brand = &native_brand;
 	rctl_prealloc_destroy(gp);
 	/*
@@ -2252,6 +2705,8 @@ zone_init(void)
 static void
 zone_free(zone_t *zone)
 {
+	zone_dl_t *zdl;
+
 	ASSERT(zone != global_zone);
 	ASSERT(zone->zone_ntasks == 0);
 	ASSERT(zone->zone_nlwps == 0);
@@ -2280,6 +2735,19 @@ zone_free(zone_t *zone)
 	list_destroy(&zone->zone_ref_list);
 	zone_free_zsd(zone);
 	zone_free_datasets(zone);
+
+	/*
+	 * While dlmgmtd should have removed all of these, it could have left
+	 * something behind or crashed. In which case it's not safe for us to
+	 * assume that the list is empty which list_destroy() will ASSERT. We
+	 * clean up for our userland comrades which may have crashed, or worse,
+	 * been disabled by SMF.
+	 */
+	while ((zdl = list_remove_head(&zone->zone_dl_list)) != NULL) {
+		if (zdl->zdl_net != NULL)
+			nvlist_free(zdl->zdl_net);
+		kmem_free(zdl, sizeof (zone_dl_t));
+	}
 	list_destroy(&zone->zone_dl_list);
 
 	if (zone->zone_rootvp != NULL)
@@ -2324,12 +2792,18 @@ zone_free(zone_t *zone)
 static void
 zone_status_set(zone_t *zone, zone_status_t status)
 {
+	timestruc_t now;
+	uint64_t t;
 
 	nvlist_t *nvl = NULL;
 	ASSERT(MUTEX_HELD(&zone_status_lock));
 	ASSERT(status > ZONE_MIN_STATE && status <= ZONE_MAX_STATE &&
 	    status >= zone_status_get(zone));
 
+	/* Current time since Jan 1 1970 but consumers expect NS */
+	gethrestime(&now);
+	t = (now.tv_sec * NANOSEC) + now.tv_nsec;
+
 	if (nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP) ||
 	    nvlist_add_string(nvl, ZONE_CB_NAME, zone->zone_name) ||
 	    nvlist_add_string(nvl, ZONE_CB_NEWSTATE,
@@ -2337,7 +2811,7 @@ zone_status_set(zone_t *zone, zone_status_t status)
 	    nvlist_add_string(nvl, ZONE_CB_OLDSTATE,
 	    zone_status_table[zone->zone_status]) ||
 	    nvlist_add_int32(nvl, ZONE_CB_ZONEID, zone->zone_id) ||
-	    nvlist_add_uint64(nvl, ZONE_CB_TIMESTAMP, (uint64_t)gethrtime()) ||
+	    nvlist_add_uint64(nvl, ZONE_CB_TIMESTAMP, t) ||
 	    sysevent_evc_publish(zone_event_chan, ZONE_EVENT_STATUS_CLASS,
 	    ZONE_EVENT_STATUS_SUBCLASS, "sun.com", "kernel", nvl, EVCH_SLEEP)) {
 #ifdef DEBUG
@@ -2463,14 +2937,65 @@ zone_set_initname(zone_t *zone, const char *zone_initname)
 	return (0);
 }
 
+/*
+ * The zone_set_mcap_nover and zone_set_mcap_pageout functions are used
+ * to provide the physical memory capping kstats.  Since physical memory
+ * capping is currently implemented in userland, that code uses the setattr
+ * entry point to increment the kstats.  We always simply increment nover
+ * every time that setattr is called and we always add in the input value
+ * to zone_mcap_pagedout every time that is called.
+ */
+/*ARGSUSED*/
 static int
-zone_set_phys_mcap(zone_t *zone, const uint64_t *zone_mcap)
+zone_set_mcap_nover(zone_t *zone, const uint64_t *zone_nover)
 {
-	uint64_t mcap;
-	int err = 0;
+	zone->zone_mcap_nover++;
+
+	return (0);
+}
+
+static int
+zone_set_mcap_pageout(zone_t *zone, const uint64_t *zone_pageout)
+{
+	uint64_t pageout;
+	int err;
+
+	if ((err = copyin(zone_pageout, &pageout, sizeof (uint64_t))) == 0)
+		zone->zone_mcap_pagedout += pageout;
+
+	return (err);
+}
+
+/*
+ * The zone_set_page_fault_delay function is used to set the number of usecs
+ * to throttle page faults.  This is normally 0 but can be set to a non-0 value
+ * by the user-land memory capping code when the zone is over its physcial
+ * memory cap.
+ */
+static int
+zone_set_page_fault_delay(zone_t *zone, const uint32_t *pfdelay)
+{
+	uint32_t dusec;
+	int err;
+
+	if ((err = copyin(pfdelay, &dusec, sizeof (uint32_t))) == 0)
+		zone->zone_pg_flt_delay = dusec;
+
+	return (err);
+}
+
+/*
+ * The zone_set_rss function is used to set the zone's RSS when we do the
+ * fast, approximate calculation in user-land.
+ */
+static int
+zone_set_rss(zone_t *zone, const uint64_t *prss)
+{
+	uint64_t rss;
+	int err;
 
-	if ((err = copyin(zone_mcap, &mcap, sizeof (uint64_t))) == 0)
-		zone->zone_phys_mcap = mcap;
+	if ((err = copyin(prss, &rss, sizeof (uint64_t))) == 0)
+		zone->zone_phys_mem = rss;
 
 	return (err);
 }
@@ -2882,6 +3407,12 @@ getzoneid(void)
 	return (curproc->p_zone->zone_id);
 }
 
+zoneid_t
+getzonedid(void)
+{
+	return (curproc->p_zone->zone_did);
+}
+
 /*
  * Internal versions of zone_find_by_*().  These don't zone_hold() or
  * check the validity of a zone's state.
@@ -4139,8 +4670,9 @@ parse_rctls(caddr_t ubuf, size_t buflen, nvlist_t **nvlp)
 
 		error = EINVAL;
 		name = nvpair_name(nvp);
-		if (strncmp(nvpair_name(nvp), "zone.", sizeof ("zone.") - 1)
-		    != 0 || nvpair_type(nvp) != DATA_TYPE_NVLIST_ARRAY) {
+		if ((strncmp(name, "zone.", sizeof ("zone.") - 1) != 0 &&
+		    strncmp(name, "project.", sizeof ("project.") - 1) != 0) ||
+		    nvpair_type(nvp) != DATA_TYPE_NVLIST_ARRAY) {
 			goto out;
 		}
 		if ((hndl = rctl_hndl_lookup(name)) == -1) {
@@ -4258,7 +4790,7 @@ zone_create(const char *zone_name, const char *zone_root,
     caddr_t rctlbuf, size_t rctlbufsz,
     caddr_t zfsbuf, size_t zfsbufsz, int *extended_error,
     int match, uint32_t doi, const bslabel_t *label,
-    int flags)
+    int flags, zoneid_t zone_did)
 {
 	struct zsched_arg zarg;
 	nvlist_t *rctls = NULL;
@@ -4281,6 +4813,7 @@ zone_create(const char *zone_name, const char *zone_root,
 
 	zone = kmem_zalloc(sizeof (zone_t), KM_SLEEP);
 	zoneid = zone->zone_id = id_alloc(zoneid_space);
+	zone->zone_did = zone_did;
 	zone->zone_status = ZONE_IS_UNINITIALIZED;
 	zone->zone_pool = pool_default;
 	zone->zone_pool_mod = gethrtime();
@@ -4288,6 +4821,8 @@ zone_create(const char *zone_name, const char *zone_root,
 	zone->zone_ncpus = 0;
 	zone->zone_ncpus_online = 0;
 	zone->zone_restart_init = B_TRUE;
+	zone->zone_reboot_on_init_exit = B_FALSE;
+	zone->zone_init_status = -1;
 	zone->zone_brand = &native_brand;
 	zone->zone_initname = NULL;
 	mutex_init(&zone->zone_lock, NULL, MUTEX_DEFAULT, NULL);
@@ -4349,10 +4884,14 @@ zone_create(const char *zone_name, const char *zone_root,
 	zone->zone_locked_mem_ctl = UINT64_MAX;
 	zone->zone_max_swap = 0;
 	zone->zone_max_swap_ctl = UINT64_MAX;
+	zone->zone_phys_mem = 0;
+	zone->zone_phys_mem_ctl = UINT64_MAX;
 	zone->zone_max_lofi = 0;
 	zone->zone_max_lofi_ctl = UINT64_MAX;
-	zone0.zone_lockedmem_kstat = NULL;
-	zone0.zone_swapresv_kstat = NULL;
+	zone->zone_lockedmem_kstat = NULL;
+	zone->zone_swapresv_kstat = NULL;
+	zone->zone_physmem_kstat = NULL;
+	zone->zone_zfs_io_pri = 1;
 
 	/*
 	 * Zsched initializes the rctls.
@@ -4509,8 +5048,8 @@ zone_create(const char *zone_name, const char *zone_root,
 	/*
 	 * The process, task, and project rctls are probably wrong;
 	 * we need an interface to get the default values of all rctls,
-	 * and initialize zsched appropriately.  I'm not sure that that
-	 * makes much of a difference, though.
+	 * and initialize zsched appropriately. However, we allow zoneadmd
+	 * to pass down both zone and project rctls for the zone's init.
 	 */
 	error = newproc(zsched, (void *)&zarg, syscid, minclsyspri, NULL, 0);
 	if (error != 0) {
@@ -4651,6 +5190,7 @@ zone_boot(zoneid_t zoneid)
 static int
 zone_empty(zone_t *zone)
 {
+	int cnt = 0;
 	int waitstatus;
 
 	/*
@@ -4661,7 +5201,16 @@ zone_empty(zone_t *zone)
 	ASSERT(MUTEX_NOT_HELD(&zonehash_lock));
 	while ((waitstatus = zone_status_timedwait_sig(zone,
 	    ddi_get_lbolt() + hz, ZONE_IS_EMPTY)) == -1) {
-		killall(zone->zone_id);
+		boolean_t force = B_FALSE;
+
+		/* Every 30 seconds, try harder */
+		if (cnt++ >= 30) {
+			cmn_err(CE_WARN, "attempt to force kill zone %d\n",
+			    zone->zone_id);
+			force = B_TRUE;
+			cnt = 0;
+		}
+		killall(zone->zone_id, force);
 	}
 	/*
 	 * return EINTR if we were signaled
@@ -5412,14 +5961,6 @@ zone_getattr(zoneid_t zoneid, int attr, void *buf, size_t bufsize)
 				error = EFAULT;
 		}
 		break;
-	case ZONE_ATTR_PHYS_MCAP:
-		size = sizeof (zone->zone_phys_mcap);
-		if (bufsize > size)
-			bufsize = size;
-		if (buf != NULL &&
-		    copyout(&zone->zone_phys_mcap, buf, bufsize) != 0)
-			error = EFAULT;
-		break;
 	case ZONE_ATTR_SCHED_CLASS:
 		mutex_enter(&class_lock);
 
@@ -5474,6 +6015,14 @@ zone_getattr(zoneid_t zoneid, int attr, void *buf, size_t bufsize)
 		}
 		kmem_free(zbuf, bufsize);
 		break;
+	case ZONE_ATTR_DID:
+		size = sizeof (zoneid_t);
+		if (bufsize > size)
+			bufsize = size;
+
+		if (buf != NULL && copyout(&zone->zone_did, buf, bufsize) != 0)
+			error = EFAULT;
+		break;
 	default:
 		if ((attr >= ZONE_ATTR_BRAND_ATTRS) && ZONE_IS_BRANDED(zone)) {
 			size = bufsize;
@@ -5505,10 +6054,11 @@ zone_setattr(zoneid_t zoneid, int attr, void *buf, size_t bufsize)
 		return (set_errno(EPERM));
 
 	/*
-	 * Only the ZONE_ATTR_PHYS_MCAP attribute can be set on the
-	 * global zone.
+	 * Only the ZONE_ATTR_PMCAP_NOVER and ZONE_ATTR_PMCAP_PAGEOUT
+	 * attributes can be set on the global zone.
 	 */
-	if (zoneid == GLOBAL_ZONEID && attr != ZONE_ATTR_PHYS_MCAP) {
+	if (zoneid == GLOBAL_ZONEID &&
+	    attr != ZONE_ATTR_PMCAP_NOVER && attr != ZONE_ATTR_PMCAP_PAGEOUT) {
 		return (set_errno(EINVAL));
 	}
 
@@ -5525,7 +6075,9 @@ zone_setattr(zoneid_t zoneid, int attr, void *buf, size_t bufsize)
 	 * non-global zones.
 	 */
 	zone_status = zone_status_get(zone);
-	if (attr != ZONE_ATTR_PHYS_MCAP && zone_status > ZONE_IS_READY) {
+	if (attr != ZONE_ATTR_PMCAP_NOVER && attr != ZONE_ATTR_PMCAP_PAGEOUT &&
+	    attr != ZONE_ATTR_PG_FLT_DELAY && attr != ZONE_ATTR_RSS &&
+	    zone_status > ZONE_IS_READY) {
 		err = EINVAL;
 		goto done;
 	}
@@ -5547,8 +6099,17 @@ zone_setattr(zoneid_t zoneid, int attr, void *buf, size_t bufsize)
 	case ZONE_ATTR_FS_ALLOWED:
 		err = zone_set_fs_allowed(zone, (const char *)buf);
 		break;
-	case ZONE_ATTR_PHYS_MCAP:
-		err = zone_set_phys_mcap(zone, (const uint64_t *)buf);
+	case ZONE_ATTR_PMCAP_NOVER:
+		err = zone_set_mcap_nover(zone, (const uint64_t *)buf);
+		break;
+	case ZONE_ATTR_PMCAP_PAGEOUT:
+		err = zone_set_mcap_pageout(zone, (const uint64_t *)buf);
+		break;
+	case ZONE_ATTR_PG_FLT_DELAY:
+		err = zone_set_page_fault_delay(zone, (const uint32_t *)buf);
+		break;
+	case ZONE_ATTR_RSS:
+		err = zone_set_rss(zone, (const uint64_t *)buf);
 		break;
 	case ZONE_ATTR_SCHED_CLASS:
 		err = zone_set_sched_class(zone, (const char *)buf);
@@ -6269,6 +6830,7 @@ zone(int cmd, void *arg1, void *arg2, void *arg3, void *arg4)
 			zs.doi = zs32.doi;
 			zs.label = (const bslabel_t *)(uintptr_t)zs32.label;
 			zs.flags = zs32.flags;
+			zs.zoneid = zs32.zoneid;
 #else
 			panic("get_udatamodel() returned bogus result\n");
 #endif
@@ -6279,7 +6841,7 @@ zone(int cmd, void *arg1, void *arg2, void *arg3, void *arg4)
 		    (caddr_t)zs.rctlbuf, zs.rctlbufsz,
 		    (caddr_t)zs.zfsbuf, zs.zfsbufsz,
 		    zs.extended_error, zs.match, zs.doi,
-		    zs.label, zs.flags));
+		    zs.label, zs.flags, zs.zoneid));
 	case ZONE_BOOT:
 		return (zone_boot((zoneid_t)(uintptr_t)arg1));
 	case ZONE_DESTROY:
@@ -6380,6 +6942,7 @@ zone_ki_call_zoneadmd(struct zarg *zargp)
 	bcopy(zone->zone_name, zone_name, zone_namelen);
 	zoneid = zone->zone_id;
 	uniqid = zone->zone_uniqid;
+	arg.status = zone->zone_init_status;
 	/*
 	 * zoneadmd may be down, but at least we can empty out the zone.
 	 * We can ignore the return value of zone_empty() since we're called
@@ -6557,7 +7120,7 @@ zone_kadmin(int cmd, int fcn, const char *mdep, cred_t *credp)
 	 * zone_ki_call_zoneadmd() will do a more thorough job of this
 	 * later.
 	 */
-	killall(zone->zone_id);
+	killall(zone->zone_id, B_FALSE);
 	/*
 	 * Now, create the thread to contact zoneadmd and do the rest of the
 	 * work.  This thread can't be created in our zone otherwise
diff --git a/usr/src/uts/common/io/scsi/adapters/mpt_sas/mptsas_hash.c b/usr/src/uts/common/refhash/refhash.c
index 8f96c2d9f1..e2de00597e 100644
--- a/usr/src/uts/common/io/scsi/adapters/mpt_sas/mptsas_hash.c
+++ b/usr/src/uts/common/refhash/refhash.c
@@ -10,16 +10,18 @@
  */
 
 /*
- * Copyright 2014 Joyent, Inc.  All rights reserved.
+ * Copyright 2015 Joyent, Inc.
  */
 
-#include <sys/scsi/adapters/mpt_sas/mptsas_hash.h>
+#include <sys/refhash.h>
 #include <sys/sysmacros.h>
 #include <sys/types.h>
 #include <sys/kmem.h>
 #include <sys/list.h>
 #include <sys/ddi.h>
 
+#define	RHL_F_DEAD	0x01
+
 #ifdef lint
 extern refhash_link_t *obj_to_link(refhash_t *, void *);
 extern void *link_to_obj(refhash_t *, refhash_link_t *);
diff --git a/usr/src/uts/common/sys/Makefile b/usr/src/uts/common/sys/Makefile
index c99ef8accb..a6537c9db3 100644
--- a/usr/src/uts/common/sys/Makefile
+++ b/usr/src/uts/common/sys/Makefile
@@ -22,6 +22,7 @@
 # Copyright (c) 1989, 2010, Oracle and/or its affiliates. All rights reserved.
 # Copyright 2013, Joyent, Inc. All rights reserved.
 # Copyright 2013 Garrett D'Amore <garrett@damore.org>
+# Copyright 2015, Joyent, Inc. All rights reserved.
 #
 
 include $(SRC)/uts/Makefile.uts
@@ -212,6 +213,7 @@ CHKHDRS=			\
 	emul64cmd.h		\
 	emul64var.h		\
 	epm.h			\
+	epoll.h			\
 	errno.h			\
 	errorq.h		\
 	errorq_impl.h		\
@@ -219,6 +221,7 @@ CHKHDRS=			\
 	ethernet.h		\
 	euc.h			\
 	eucioctl.h		\
+	eventfd.h		\
 	exacct.h		\
 	exacct_catalog.h	\
 	exacct_impl.h		\
@@ -243,6 +246,7 @@ CHKHDRS=			\
 	flock.h			\
 	flock_impl.h		\
 	fork.h			\
+	frameio.h		\
 	fss.h			\
 	fsspriocntl.h		\
 	fsid.h			\
@@ -268,6 +272,7 @@ CHKHDRS=			\
 	idmap.h 		\
 	ieeefp.h		\
 	id_space.h		\
+	inotify.h		\
 	instance.h		\
 	int_const.h		\
 	int_fmtio.h		\
@@ -336,6 +341,7 @@ CHKHDRS=			\
 	lgrp.h			\
 	lgrp_user.h		\
 	libc_kernel.h		\
+	limits.h		\
 	link.h			\
 	list.h			\
 	list_impl.h		\
@@ -416,6 +422,9 @@ CHKHDRS=			\
 	ontrap.h		\
 	open.h			\
 	openpromio.h		\
+	overlay.h		\
+	overlay_common.h	\
+	overlay_target.h	\
 	panic.h			\
 	param.h			\
 	pathconf.h		\
@@ -581,6 +590,7 @@ CHKHDRS=			\
 	time_std_impl.h		\
 	timeb.h			\
 	timer.h			\
+	timerfd.h		\
 	times.h			\
 	timex.h			\
 	timod.h			\
@@ -635,6 +645,8 @@ CHKHDRS=			\
 	vmem.h			\
 	vmem_impl.h		\
 	vmsystm.h		\
+	vnd.h			\
+	vnd_errno.h		\
 	vnic.h			\
 	vnic_impl.h		\
 	vnode.h			\
@@ -646,12 +658,14 @@ CHKHDRS=			\
 	vuid_queue.h		\
 	vuid_state.h		\
 	vuid_store.h		\
+	vxlan.h			\
 	wait.h			\
 	waitq.h			\
 	wanboot_impl.h		\
 	watchpoint.h		\
 	winlockio.h		\
 	zcons.h			\
+	zfd.h			\
 	zone.h			\
 	xti_inet.h		\
 	xti_osi.h		\
@@ -852,13 +866,14 @@ FSHDRS=				\
 	cachefs_log.h		\
 	decomp.h		\
 	dv_node.h		\
-	sdev_impl.h		\
 	fifonode.h		\
 	hsfs_isospec.h		\
 	hsfs_node.h		\
 	hsfs_rrip.h		\
 	hsfs_spec.h		\
 	hsfs_susp.h		\
+	hyprlofs.h		\
+	hyprlofs_info.h		\
 	lofs_info.h		\
 	lofs_node.h		\
 	mntdata.h		\
@@ -868,6 +883,8 @@ FSHDRS=				\
 	pc_label.h		\
 	pc_node.h		\
 	pxfs_ki.h		\
+	sdev_impl.h		\
+	sdev_plugin.h		\
 	snode.h			\
 	swapnode.h		\
 	tmp.h			\
diff --git a/usr/src/uts/common/sys/aggr_impl.h b/usr/src/uts/common/sys/aggr_impl.h
index 5b3bba08c2..a02240ecea 100644
--- a/usr/src/uts/common/sys/aggr_impl.h
+++ b/usr/src/uts/common/sys/aggr_impl.h
@@ -21,6 +21,7 @@
 /*
  * Copyright 2010 Sun Microsystems, Inc.  All rights reserved.
  * Use is subject to license terms.
+ * Copyright 2012 OmniTI Computer Consulting, Inc  All rights reserved.
  */
 
 #ifndef	_SYS_AGGR_IMPL_H
@@ -307,6 +308,8 @@ extern boolean_t aggr_port_notify_link(aggr_grp_t *, aggr_port_t *);
 extern void aggr_port_init_callbacks(aggr_port_t *);
 
 extern void aggr_recv_cb(void *, mac_resource_handle_t, mblk_t *, boolean_t);
+extern void aggr_recv_promisc_cb(void *, mac_resource_handle_t, mblk_t *,
+    boolean_t);
 
 extern void aggr_tx_ring_update(void *, uintptr_t);
 extern void aggr_tx_notify_thread(void *);
diff --git a/usr/src/uts/common/sys/auxv.h b/usr/src/uts/common/sys/auxv.h
index 3a2e705850..e875cd10d9 100644
--- a/usr/src/uts/common/sys/auxv.h
+++ b/usr/src/uts/common/sys/auxv.h
@@ -29,7 +29,7 @@
  * Use is subject to license terms.
  */
 /*
- * Copyright (c) 2012, Joyent, Inc.  All rights reserved.
+ * Copyright (c) 2014, Joyent, Inc.  All rights reserved.
  */
 
 #ifndef	_SYS_AUXV_H
@@ -78,6 +78,9 @@ typedef struct {
 #define	AT_FLAGS	8	/* processor flags */
 #define	AT_ENTRY	9	/* a.out entry point */
 
+/* First introduced on Linux */
+#define	AT_RANDOM	25	/* address of 16 random bytes */
+
 /*
  * These relate to the original PPC ABI document; Linux reused
  * the values for other things (see below), so disambiguation of
@@ -110,6 +113,16 @@ typedef struct {
  * AT_UCACHEBSIZE	21	(moved from 12)
  *
  * AT_IGNOREPPC		22
+ *
+ * On Linux:
+ * AT_* values 18 through 22 are reserved
+ * AT_SECURE		23	secure mode boolean
+ * AT_BASE_PLATFORM	24	string identifying real platform, may
+ *				differ from AT_PLATFORM.
+ * AT_HWCAP2		26	extension of AT_HWCAP
+ * AT_EXECFN		31	filename of program
+ * AT_SYSINFO		32
+ * AT_SYSINFO_EHDR	33	The vDSO location
  */
 
 /*
@@ -186,6 +199,8 @@ extern uint_t getisax(uint32_t *, uint_t);
 #define	AT_SUN_BRAND_AUX1	2020
 #define	AT_SUN_BRAND_AUX2	2021
 #define	AT_SUN_BRAND_AUX3	2022
+#define	AT_SUN_BRAND_AUX4	2025
+#define	AT_SUN_BRAND_NROOT	2024
 
 /*
  * Note that 2023 is reserved for the AT_SUN_HWCAP2 word defined above.
diff --git a/usr/src/uts/common/sys/brand.h b/usr/src/uts/common/sys/brand.h
index badc3faff8..8183f643a4 100644
--- a/usr/src/uts/common/sys/brand.h
+++ b/usr/src/uts/common/sys/brand.h
@@ -21,6 +21,7 @@
 
 /*
  * Copyright (c) 2006, 2010, Oracle and/or its affiliates. All rights reserved.
+ * Copyright 2015, Joyent, Inc.
  */
 
 #ifndef _SYS_BRAND_H
@@ -102,29 +103,98 @@ struct brand_mach_ops;
 struct intpdata;
 struct execa;
 
+/*
+ * Common structure to define hooks for brand operation.
+ *
+ * Required Fields:
+ * b_init_brand_data - Setup zone brand data during zone_setbrand
+ * b_free_brand_data - Free zone brand data during zone_destroy
+ * b_brandsys - Syscall handler for brandsys
+ * b_setbrand - Initialize process brand data
+ * b_getattr - Get brand-custom zone attribute
+ * b_setattr - Set brand-custom zone attribute
+ * b_copy_procdata - Copy process brand data during fork
+ * b_proc_exit - Perform process brand exit processing
+ * b_exec - Reset branded process state on exec
+ * b_lwp_setrval - Set return code for forked child
+ * b_initlwp - Initialize lwp brand data
+ * b_forklwp - Copy lwp brand data during fork
+ * b_freelwp - Free lwp brand data
+ * b_lwpexit - Perform lwp-specific brand exit processing
+ * b_elfexec - Load and execute ELF binary
+ * b_sigset_native_to_brand - Convert sigset native->brand
+ * b_sigset_brand_to_native - Convert sigset brand->native
+ * b_nsig - Maxiumum signal number
+ * b_sendsig - Update process state after sendsig
+ *
+ * Optional Fields:
+ * b_lwpdata_alloc - Speculatively allocate data for use in b_initlwp
+ * b_lwpdata_free - Free data from allocated by b_lwpdata_alloc if errors occur
+ *                  during lwp creation before b_initlwp could be called.
+ * b_exit_with_sig - Instead of sending SIGCLD, exit with custom behavior
+ * b_psig_to_proc - Custom additional behavior during psig
+ * b_wait_filter - Filter processes from being matched by waitid
+ * b_native_exec - Provide interpreter path prefix for executables
+ * b_ptrace_exectrap - Custom behavior for legacy ptrace traps
+ * b_map32limit - Specify alternate limit for MAP_32BIT mappings
+ * b_stop_notify - Hook process stop events
+ * b_waitid_helper - Generate synthetic results for waitid
+ * b_sigcld_repost - Post synthetic SIGCLD signals
+ * b_issig_stop - Alter/suppress signal delivery during issig
+ * b_sig_ignorable - Disallow discarding of signals
+ * b_savecontext - Alter context during savecontext
+ * b_restorecontext - Alter context during restorecontext
+ * b_sendsig_stack - Override stack used for signal delivery
+ * b_setid_clear - Override setid_clear behavior
+ * b_pagefault - Trap pagefault events
+ */
 struct brand_ops {
 	void	(*b_init_brand_data)(zone_t *);
 	void	(*b_free_brand_data)(zone_t *);
 	int	(*b_brandsys)(int, int64_t *, uintptr_t, uintptr_t, uintptr_t,
-		uintptr_t, uintptr_t, uintptr_t);
+		uintptr_t, uintptr_t);
 	void	(*b_setbrand)(struct proc *);
 	int	(*b_getattr)(zone_t *, int, void *, size_t *);
 	int	(*b_setattr)(zone_t *, int, void *, size_t);
 	void	(*b_copy_procdata)(struct proc *, struct proc *);
-	void	(*b_proc_exit)(struct proc *, klwp_t *);
+	void	(*b_proc_exit)(struct proc *);
 	void	(*b_exec)();
 	void	(*b_lwp_setrval)(klwp_t *, int, int);
-	int	(*b_initlwp)(klwp_t *);
+	void	*(*b_lwpdata_alloc)(struct proc *);
+	void	(*b_lwpdata_free)(void *);
+	void	(*b_initlwp)(klwp_t *, void *);
 	void	(*b_forklwp)(klwp_t *, klwp_t *);
 	void	(*b_freelwp)(klwp_t *);
 	void	(*b_lwpexit)(klwp_t *);
 	int	(*b_elfexec)(struct vnode *vp, struct execa *uap,
 	    struct uarg *args, struct intpdata *idata, int level,
 	    long *execsz, int setid, caddr_t exec_file,
-	    struct cred *cred, int brand_action);
+	    struct cred *cred, int *brand_action);
 	void	(*b_sigset_native_to_brand)(sigset_t *);
 	void	(*b_sigset_brand_to_native)(sigset_t *);
+	void	(*b_psig_to_proc)(proc_t *, kthread_t *, int);
 	int	b_nsig;
+	void	(*b_exit_with_sig)(proc_t *, sigqueue_t *);
+	boolean_t (*b_wait_filter)(proc_t *, proc_t *);
+	boolean_t (*b_native_exec)(uint8_t, const char **);
+	void (*b_ptrace_exectrap)(proc_t *);
+	uint32_t (*b_map32limit)(proc_t *);
+	void	(*b_stop_notify)(proc_t *, klwp_t *, ushort_t, ushort_t);
+	int	(*b_waitid_helper)(idtype_t, id_t, k_siginfo_t *, int,
+	    boolean_t *, int *);
+	int	(*b_sigcld_repost)(proc_t *, sigqueue_t *);
+	int	(*b_issig_stop)(proc_t *, klwp_t *);
+	boolean_t (*b_sig_ignorable)(proc_t *, int);
+	void	(*b_savecontext)(ucontext_t *);
+#if defined(_SYSCALL32_IMPL)
+	void	(*b_savecontext32)(ucontext32_t *);
+#endif
+	void	(*b_restorecontext)(ucontext_t *);
+	caddr_t	(*b_sendsig_stack)(int);
+	void	(*b_sendsig)(int);
+	int	(*b_setid_clear)(vattr_t *vap, cred_t *cr);
+	int	(*b_pagefault)(proc_t *, klwp_t *, caddr_t, enum fault_type,
+	    enum seg_rw);
 };
 
 /*
@@ -135,6 +205,7 @@ typedef struct brand {
 	char    		*b_name;
 	struct brand_ops	*b_ops;
 	struct brand_mach_ops	*b_machops;
+	size_t			b_data_size;
 } brand_t;
 
 extern brand_t native_brand;
@@ -166,7 +237,7 @@ extern brand_t	*brand_find_name(char *);
 extern void	brand_unregister_zone(brand_t *);
 extern int	brand_zone_count(brand_t *);
 extern void	brand_setbrand(proc_t *);
-extern void	brand_clearbrand(proc_t *, boolean_t);
+extern void	brand_clearbrand(proc_t *);
 
 /*
  * The following functions can be shared among kernel brand modules which
@@ -178,17 +249,16 @@ extern int	brand_solaris_cmd(int, uintptr_t, uintptr_t, uintptr_t,
 extern void	brand_solaris_copy_procdata(proc_t *, proc_t *,
 		    struct brand *);
 extern int	brand_solaris_elfexec(vnode_t *, execa_t *, uarg_t *,
-		    intpdata_t *, int, long *, int, caddr_t, cred_t *, int,
-		    struct brand *, char *, char *, char *, char *, char *);
+		    intpdata_t *, int, long *, int, caddr_t, cred_t *, int *,
+		    struct brand *, char *, char *, char *);
 extern void	brand_solaris_exec(struct brand *);
 extern int	brand_solaris_fini(char **, struct modlinkage *,
 		    struct brand *);
 extern void	brand_solaris_forklwp(klwp_t *, klwp_t *, struct brand *);
 extern void	brand_solaris_freelwp(klwp_t *, struct brand *);
-extern int	brand_solaris_initlwp(klwp_t *, struct brand *);
+extern void	brand_solaris_initlwp(klwp_t *, struct brand *);
 extern void	brand_solaris_lwpexit(klwp_t *, struct brand *);
-extern void	brand_solaris_proc_exit(struct proc *, klwp_t *,
-		    struct brand *);
+extern void	brand_solaris_proc_exit(struct proc *, struct brand *);
 extern void	brand_solaris_setbrand(proc_t *, struct brand *);
 
 #if defined(_SYSCALL32)
diff --git a/usr/src/uts/common/sys/buf.h b/usr/src/uts/common/sys/buf.h
index a9191aed7c..cb8a6012fc 100644
--- a/usr/src/uts/common/sys/buf.h
+++ b/usr/src/uts/common/sys/buf.h
@@ -21,6 +21,7 @@
 /*
  * Copyright 2008 Sun Microsystems, Inc.  All rights reserved.
  * Use is subject to license terms.
+ * Copyright 2012 Joyent, Inc.  All rights reserved.
  */
 
 /*	Copyright (c) 1984, 1986, 1987, 1988, 1989 AT&T	*/
@@ -186,6 +187,7 @@ struct biostats {
 #define	B_STARTED	0x2000000	/* io:::start probe called for buf */
 #define	B_ABRWRITE	0x4000000	/* Application based recovery active */
 #define	B_PAGE_NOWAIT	0x8000000	/* Skip the page if it is locked */
+#define	B_INVALCURONLY	0x10000000	/* invalidate only for curproc */
 
 /*
  * There is some confusion over the meaning of B_FREE and B_INVAL and what
@@ -198,6 +200,12 @@ struct biostats {
  * between the sole use of these two flags.  In both cases, IO will be done
  * if the page is not yet committed to storage.
  *
+ * The B_INVALCURONLY flag modifies the behavior of the B_INVAL flag and is
+ * intended to be used in conjunction with B_INVAL.  B_INVALCURONLY has no
+ * meaning on its own.  When both B_INVALCURONLY and B_INVAL are set, then
+ * the mapping for the page is only invalidated for the current process.
+ * In this case, the page is not destroyed unless this was the final mapping.
+ *
  * In order to discard pages without writing them back, (B_INVAL | B_TRUNC)
  * should be used.
  *
diff --git a/usr/src/uts/common/sys/cpucaps.h b/usr/src/uts/common/sys/cpucaps.h
index 6063ff4380..6bc042108c 100644
--- a/usr/src/uts/common/sys/cpucaps.h
+++ b/usr/src/uts/common/sys/cpucaps.h
@@ -22,6 +22,7 @@
 /*
  * Copyright 2007 Sun Microsystems, Inc.  All rights reserved.
  * Use is subject to license terms.
+ * Copyright 2011, 2012, Joyent, Inc.  All rights reserved.
  */
 
 #ifndef	_SYS_CPUCAPS_H
@@ -84,12 +85,16 @@ extern void cpucaps_zone_remove(zone_t *);
  */
 extern int cpucaps_project_set(kproject_t *, rctl_qty_t);
 extern int cpucaps_zone_set(zone_t *, rctl_qty_t);
+extern int cpucaps_zone_set_base(zone_t *, rctl_qty_t);
+extern int cpucaps_zone_set_burst_time(zone_t *, rctl_qty_t);
 
 /*
  * Get current CPU usage for a project/zone.
  */
 extern rctl_qty_t cpucaps_project_get(kproject_t *);
 extern rctl_qty_t cpucaps_zone_get(zone_t *);
+extern rctl_qty_t cpucaps_zone_get_base(zone_t *);
+extern rctl_qty_t cpucaps_zone_get_burst_time(zone_t *);
 
 /*
  * Scheduling class hooks into CPU caps framework.
diff --git a/usr/src/uts/common/sys/cpucaps_impl.h b/usr/src/uts/common/sys/cpucaps_impl.h
index 95afd21827..2cd4ed644d 100644
--- a/usr/src/uts/common/sys/cpucaps_impl.h
+++ b/usr/src/uts/common/sys/cpucaps_impl.h
@@ -22,6 +22,7 @@
 /*
  * Copyright 2009 Sun Microsystems, Inc.  All rights reserved.
  * Use is subject to license terms.
+ * Copyright 2011, 2012, Joyent, Inc.  All rights reserved.
  */
 
 #ifndef	_SYS_CPUCAPS_IMPL_H
@@ -66,8 +67,12 @@ typedef struct cpucap {
 	waitq_t		cap_waitq;	/* waitq for capped threads	*/
 	kstat_t		*cap_kstat;	/* cpucaps specific kstat	*/
 	int64_t		cap_gen;	/* zone cap specific 		*/
+	hrtime_t	cap_chk_value;	/* effective CPU usage cap	*/
 	hrtime_t	cap_value;	/* scaled CPU usage cap		*/
 	hrtime_t	cap_usage;	/* current CPU usage		*/
+	hrtime_t	cap_base;	/* base CPU for burst		*/
+	u_longlong_t	cap_burst_limit; /* max secs (in tics) for a burst */
+	u_longlong_t	cap_bursting;	/* # of ticks currently bursting */
 	disp_lock_t	cap_usagelock;	/* protects cap_usage above	*/
 	/*
 	 * Per cap statistics.
@@ -75,6 +80,7 @@ typedef struct cpucap {
 	hrtime_t	cap_maxusage;	/* maximum cap usage		*/
 	u_longlong_t	cap_below;	/* # of ticks spend below the cap */
 	u_longlong_t	cap_above;	/* # of ticks spend above the cap */
+	u_longlong_t	cap_above_base;	/* # of ticks spent above the base */
 } cpucap_t;
 
 /*
diff --git a/usr/src/uts/common/sys/cred.h b/usr/src/uts/common/sys/cred.h
index 5056f9a511..914f132dc0 100644
--- a/usr/src/uts/common/sys/cred.h
+++ b/usr/src/uts/common/sys/cred.h
@@ -93,6 +93,7 @@ extern gid_t crgetgid(const cred_t *);
 extern gid_t crgetrgid(const cred_t *);
 extern gid_t crgetsgid(const cred_t *);
 extern zoneid_t crgetzoneid(const cred_t *);
+extern zoneid_t crgetzonedid(const cred_t *);
 extern projid_t crgetprojid(const cred_t *);
 
 extern cred_t *crgetmapped(const cred_t *);
diff --git a/usr/src/uts/common/sys/ctf_api.h b/usr/src/uts/common/sys/ctf_api.h
index 04d73c3181..73beb2d244 100644
--- a/usr/src/uts/common/sys/ctf_api.h
+++ b/usr/src/uts/common/sys/ctf_api.h
@@ -24,7 +24,7 @@
  * Use is subject to license terms.
  */
 /*
- * Copyright (c) 2013, Joyent, Inc.  All rights reserved.
+ * Copyright (c) 2015, Joyent, Inc.
  */
 
 /*
@@ -60,6 +60,62 @@ extern "C" {
 typedef struct ctf_file ctf_file_t;
 typedef long ctf_id_t;
 
+#define	ECTF_BASE	1000	/* base value for libctf errnos */
+
+enum {
+	ECTF_FMT = ECTF_BASE,	/* file is not in CTF or ELF format */
+	ECTF_ELFVERS,		/* ELF version is more recent than libctf */
+	ECTF_CTFVERS,		/* CTF version is more recent than libctf */
+	ECTF_ENDIAN,		/* data is different endian-ness than lib */
+	ECTF_SYMTAB,		/* symbol table uses invalid entry size */
+	ECTF_SYMBAD,		/* symbol table data buffer invalid */
+	ECTF_STRBAD,		/* string table data buffer invalid */
+	ECTF_CORRUPT,		/* file data corruption detected */
+	ECTF_NOCTFDATA,		/* ELF file does not contain CTF data */
+	ECTF_NOCTFBUF,		/* buffer does not contain CTF data */
+	ECTF_NOSYMTAB,		/* symbol table data is not available */
+	ECTF_NOPARENT,		/* parent CTF container is not available */
+	ECTF_DMODEL,		/* data model mismatch */
+	ECTF_MMAP,		/* failed to mmap a data section */
+	ECTF_ZMISSING,		/* decompression library not installed */
+	ECTF_ZINIT,		/* failed to initialize decompression library */
+	ECTF_ZALLOC,		/* failed to allocate decompression buffer */
+	ECTF_DECOMPRESS,	/* failed to decompress CTF data */
+	ECTF_STRTAB,		/* string table for this string is missing */
+	ECTF_BADNAME,		/* string offset is corrupt w.r.t. strtab */
+	ECTF_BADID,		/* invalid type ID number */
+	ECTF_NOTSOU,		/* type is not a struct or union */
+	ECTF_NOTENUM,		/* type is not an enum */
+	ECTF_NOTSUE,		/* type is not a struct, union, or enum */
+	ECTF_NOTINTFP,		/* type is not an integer or float */
+	ECTF_NOTARRAY,		/* type is not an array */
+	ECTF_NOTREF,		/* type does not reference another type */
+	ECTF_NAMELEN,		/* buffer is too small to hold type name */
+	ECTF_NOTYPE,		/* no type found corresponding to name */
+	ECTF_SYNTAX,		/* syntax error in type name */
+	ECTF_NOTFUNC,		/* symtab entry does not refer to a function */
+	ECTF_NOFUNCDAT,		/* no func info available for function */
+	ECTF_NOTDATA,		/* symtab entry does not refer to a data obj */
+	ECTF_NOTYPEDAT,		/* no type info available for object */
+	ECTF_NOLABEL,		/* no label found corresponding to name */
+	ECTF_NOLABELDATA,	/* file does not contain any labels */
+	ECTF_NOTSUP,		/* feature not supported */
+	ECTF_NOENUMNAM,		/* enum element name not found */
+	ECTF_NOMEMBNAM,		/* member name not found */
+	ECTF_RDONLY,		/* CTF container is read-only */
+	ECTF_DTFULL,		/* CTF type is full (no more members allowed) */
+	ECTF_FULL,		/* CTF container is full */
+	ECTF_DUPMEMBER,		/* duplicate member name definition */
+	ECTF_CONFLICT,		/* conflicting type definition present */
+	ECTF_REFERENCED,	/* type has outstanding references */
+	ECTF_NOTDYN,		/* type is not a dynamic type */
+	ECTF_ELF,		/* elf library failure */
+	ECTF_MCHILD,		/* cannot merge child container */
+	ECTF_LABELEXISTS,	/* label already exists */
+	ECTF_LCONFLICT,		/* merged labels conflict */
+	ECTF_ZLIB		/* zlib library failure */
+};
+
 /*
  * If the debugger needs to provide the CTF library with a set of raw buffers
  * for use as the CTF data, symbol table, and string table, it can do so by
@@ -143,19 +199,24 @@ typedef struct ctf_lblinfo {
 typedef int ctf_visit_f(const char *, ctf_id_t, ulong_t, int, void *);
 typedef int ctf_member_f(const char *, ctf_id_t, ulong_t, void *);
 typedef int ctf_enum_f(const char *, int, void *);
-typedef int ctf_type_f(ctf_id_t, void *);
+typedef int ctf_type_f(ctf_id_t, boolean_t, void *);
 typedef int ctf_label_f(const char *, const ctf_lblinfo_t *, void *);
+typedef int ctf_function_f(const char *, ulong_t, ctf_funcinfo_t *, void *);
+typedef int ctf_object_f(const char *, ctf_id_t, ulong_t, void *);
+typedef int ctf_string_f(const char *, void *);
 
 extern ctf_file_t *ctf_bufopen(const ctf_sect_t *, const ctf_sect_t *,
     const ctf_sect_t *, int *);
 extern ctf_file_t *ctf_fdopen(int, int *);
 extern ctf_file_t *ctf_open(const char *, int *);
 extern ctf_file_t *ctf_create(int *);
+extern ctf_file_t *ctf_fdcreate(int, int *);
 extern ctf_file_t *ctf_dup(ctf_file_t *);
 extern void ctf_close(ctf_file_t *);
 
 extern ctf_file_t *ctf_parent_file(ctf_file_t *);
 extern const char *ctf_parent_name(ctf_file_t *);
+extern const char *ctf_parent_label(ctf_file_t *);
 
 extern int ctf_import(ctf_file_t *, ctf_file_t *);
 extern int ctf_setmodel(ctf_file_t *, int);
@@ -165,15 +226,20 @@ extern void ctf_setspecific(ctf_file_t *, void *);
 extern void *ctf_getspecific(ctf_file_t *);
 
 extern int ctf_errno(ctf_file_t *);
+extern uint_t ctf_flags(ctf_file_t *);
 extern const char *ctf_errmsg(int);
 extern int ctf_version(int);
 
 extern int ctf_func_info(ctf_file_t *, ulong_t, ctf_funcinfo_t *);
+extern int ctf_func_info_by_id(ctf_file_t *, ctf_id_t, ctf_funcinfo_t *);
 extern int ctf_func_args(ctf_file_t *, ulong_t, uint_t, ctf_id_t *);
+extern int ctf_func_args_by_id(ctf_file_t *, ctf_id_t, uint_t, ctf_id_t *);
 
 extern ctf_id_t ctf_lookup_by_name(ctf_file_t *, const char *);
 extern ctf_id_t ctf_lookup_by_symbol(ctf_file_t *, ulong_t);
 
+extern char *ctf_symbol_name(ctf_file_t *, ulong_t, char *, size_t);
+
 extern ctf_id_t ctf_type_resolve(ctf_file_t *, ctf_id_t);
 extern ssize_t ctf_type_lname(ctf_file_t *, ctf_id_t, char *, size_t);
 extern char *ctf_type_name(ctf_file_t *, ctf_id_t, char *, size_t);
@@ -201,29 +267,39 @@ extern int ctf_label_info(ctf_file_t *, const char *, ctf_lblinfo_t *);
 
 extern int ctf_member_iter(ctf_file_t *, ctf_id_t, ctf_member_f *, void *);
 extern int ctf_enum_iter(ctf_file_t *, ctf_id_t, ctf_enum_f *, void *);
-extern int ctf_type_iter(ctf_file_t *, ctf_type_f *, void *);
+extern int ctf_type_iter(ctf_file_t *, boolean_t, ctf_type_f *, void *);
 extern int ctf_label_iter(ctf_file_t *, ctf_label_f *, void *);
+extern int ctf_function_iter(ctf_file_t *, ctf_function_f *, void *);
+extern int ctf_object_iter(ctf_file_t *, ctf_object_f *, void *);
+extern int ctf_string_iter(ctf_file_t *, ctf_string_f *, void *);
 
 extern ctf_id_t ctf_add_array(ctf_file_t *, uint_t, const ctf_arinfo_t *);
-extern ctf_id_t ctf_add_const(ctf_file_t *, uint_t, ctf_id_t);
+extern ctf_id_t ctf_add_const(ctf_file_t *, uint_t, const char *, ctf_id_t);
 extern ctf_id_t ctf_add_enum(ctf_file_t *, uint_t, const char *);
 extern ctf_id_t ctf_add_float(ctf_file_t *, uint_t,
     const char *, const ctf_encoding_t *);
 extern ctf_id_t ctf_add_forward(ctf_file_t *, uint_t, const char *, uint_t);
-extern ctf_id_t ctf_add_function(ctf_file_t *, uint_t,
-    const ctf_funcinfo_t *, const ctf_id_t *);
+extern ctf_id_t ctf_add_funcptr(ctf_file_t *, uint_t, const ctf_funcinfo_t *,
+    const ctf_id_t *);
 extern ctf_id_t ctf_add_integer(ctf_file_t *, uint_t,
     const char *, const ctf_encoding_t *);
-extern ctf_id_t ctf_add_pointer(ctf_file_t *, uint_t, ctf_id_t);
+extern ctf_id_t ctf_add_pointer(ctf_file_t *, uint_t, const char *, ctf_id_t);
 extern ctf_id_t ctf_add_type(ctf_file_t *, ctf_file_t *, ctf_id_t);
 extern ctf_id_t ctf_add_typedef(ctf_file_t *, uint_t, const char *, ctf_id_t);
-extern ctf_id_t ctf_add_restrict(ctf_file_t *, uint_t, ctf_id_t);
+extern ctf_id_t ctf_add_restrict(ctf_file_t *, uint_t, const char *, ctf_id_t);
 extern ctf_id_t ctf_add_struct(ctf_file_t *, uint_t, const char *);
 extern ctf_id_t ctf_add_union(ctf_file_t *, uint_t, const char *);
-extern ctf_id_t ctf_add_volatile(ctf_file_t *, uint_t, ctf_id_t);
+extern ctf_id_t ctf_add_volatile(ctf_file_t *, uint_t, const char *, ctf_id_t);
 
 extern int ctf_add_enumerator(ctf_file_t *, ctf_id_t, const char *, int);
-extern int ctf_add_member(ctf_file_t *, ctf_id_t, const char *, ctf_id_t);
+extern int ctf_add_member(ctf_file_t *, ctf_id_t, const char *, ctf_id_t,
+    ulong_t);
+
+
+extern int ctf_add_function(ctf_file_t *, ulong_t, const ctf_funcinfo_t *,
+    const ctf_id_t *);
+extern int ctf_add_object(ctf_file_t *, ulong_t, ctf_id_t);
+extern int ctf_add_label(ctf_file_t *, const char *, ctf_id_t, uint_t);
 
 extern int ctf_set_array(ctf_file_t *, ctf_id_t, const ctf_arinfo_t *);
 
@@ -232,6 +308,7 @@ extern int ctf_delete_type(ctf_file_t *, ctf_id_t);
 extern int ctf_update(ctf_file_t *);
 extern int ctf_discard(ctf_file_t *);
 extern int ctf_write(ctf_file_t *, int);
+extern void ctf_dataptr(ctf_file_t *, const void **, size_t *);
 
 #ifdef _KERNEL
 
diff --git a/usr/src/uts/common/sys/devpoll.h b/usr/src/uts/common/sys/devpoll.h
index 36c815c69f..4e4c76d9b0 100644
--- a/usr/src/uts/common/sys/devpoll.h
+++ b/usr/src/uts/common/sys/devpoll.h
@@ -24,11 +24,13 @@
  * All rights reserved.
  */
 
+/*
+ * Copyright (c) 2014, Joyent, Inc. All rights reserved.
+ */
+
 #ifndef	_SYS_DEVPOLL_H
 #define	_SYS_DEVPOLL_H
 
-#pragma ident	"%Z%%M%	%I%	%E% SMI"
-
 #include <sys/poll_impl.h>
 #include <sys/types32.h>
 
@@ -39,8 +41,10 @@ extern "C" {
 
 /* /dev/poll ioctl */
 #define		DPIOC	(0xD0 << 8)
-#define	DP_POLL		(DPIOC | 1)	/* poll on fds in cached in /dev/poll */
+#define	DP_POLL		(DPIOC | 1)	/* poll on fds cached via /dev/poll */
 #define	DP_ISPOLLED	(DPIOC | 2)	/* is this fd cached in /dev/poll */
+#define	DP_PPOLL	(DPIOC | 3)	/* ppoll on fds cached via /dev/poll */
+#define	DP_EPOLLCOMPAT	(DPIOC | 4)	/* turn on epoll compatibility */
 
 #define	DEVPOLLSIZE	1000		/* /dev/poll table size increment */
 
@@ -51,14 +55,21 @@ typedef struct dvpoll {
 	pollfd_t	*dp_fds;	/* pollfd array */
 	nfds_t		dp_nfds;	/* num of pollfd's in dp_fds[] */
 	int		dp_timeout;	/* time out in milisec */
+	sigset_t	*dp_setp;	/* sigset, if any */
 } dvpoll_t;
 
 typedef struct dvpoll32 {
 	caddr32_t	dp_fds;		/* pollfd array */
 	uint32_t	dp_nfds;	/* num of pollfd's in dp_fds[] */
 	int32_t		dp_timeout;	/* time out in milisec */
+	caddr32_t	dp_setp;	/* sigset, if any */
 } dvpoll32_t;
 
+typedef struct dvpoll_epollfd {
+	pollfd_t	dpep_pollfd;	/* must be first member */
+	uint64_t	dpep_data;	/* data payload */
+} dvpoll_epollfd_t;
+
 #ifdef _KERNEL
 
 typedef struct dp_entry {
@@ -71,6 +82,7 @@ typedef struct dp_entry {
 } dp_entry_t;
 
 #define	DP_WRITER_PRESENT	0x1	/* a write is in progress */
+#define	DP_ISEPOLLCOMPAT	0x2	/* epoll compatibility mode */
 
 #define	DP_REFRELE(dpep) {			\
 	mutex_enter(&(dpep)->dpe_lock);		\
diff --git a/usr/src/uts/common/sys/dktp/dadk.h b/usr/src/uts/common/sys/dktp/dadk.h
index f5c990e7c0..2178ad1f0d 100644
--- a/usr/src/uts/common/sys/dktp/dadk.h
+++ b/usr/src/uts/common/sys/dktp/dadk.h
@@ -65,6 +65,8 @@ struct	dadk {
 	kstat_t		*dad_errstats;	/* error stats			*/
 	kmutex_t	dad_cmd_mutex;
 	int		dad_cmd_count;
+	uint32_t	dad_err_cnt;	/* number of recent errors	*/
+	hrtime_t	dad_last_log;	/* time of last error log	*/
 };
 
 #define	DAD_SECSIZ	dad_phyg.g_secsiz
diff --git a/usr/src/uts/common/sys/dld.h b/usr/src/uts/common/sys/dld.h
index fb2a0749d3..4cd93be56e 100644
--- a/usr/src/uts/common/sys/dld.h
+++ b/usr/src/uts/common/sys/dld.h
@@ -21,6 +21,7 @@
 /*
  * Copyright 2010 Sun Microsystems, Inc.  All rights reserved.
  * Use is subject to license terms.
+ * Copyright 2011 Joyent, Inc.  All rights reserved.
  */
 
 #ifndef	_SYS_DLD_H
@@ -191,6 +192,7 @@ typedef struct dld_ioc_rename {
 	datalink_id_t	dir_linkid1;
 	datalink_id_t	dir_linkid2;
 	char		dir_link[MAXLINKNAMELEN];
+	boolean_t	dir_zoneinit;
 } dld_ioc_rename_t;
 
 /*
@@ -203,6 +205,7 @@ typedef struct dld_ioc_rename {
 typedef struct dld_ioc_zid {
 	zoneid_t	diz_zid;
 	datalink_id_t	diz_linkid;
+	boolean_t	diz_transient;
 } dld_ioc_zid_t;
 
 /*
@@ -350,6 +353,7 @@ typedef struct dld_hwgrpinfo {
  */
 typedef	int	(*dld_capab_func_t)(void *, uint_t, void *, uint_t);
 
+#define	DI_DIRECT_RAW	0x1
 /*
  * Direct Tx/Rx capability.
  */
@@ -374,6 +378,9 @@ typedef struct dld_capab_direct_s {
 	/* flow control "can I put on a ring" callback */
 	uintptr_t	di_tx_fctl_df; /* canput-like callback */
 	void		*di_tx_fctl_dh;
+
+	/* flags that control our behavior */
+	uint_t		di_flags;
 } dld_capab_direct_t;
 
 /*
diff --git a/usr/src/uts/common/sys/dld_impl.h b/usr/src/uts/common/sys/dld_impl.h
index a76a927e59..81708aad38 100644
--- a/usr/src/uts/common/sys/dld_impl.h
+++ b/usr/src/uts/common/sys/dld_impl.h
@@ -53,7 +53,8 @@ typedef enum {
 typedef enum {
 	DLD_UNINITIALIZED,
 	DLD_PASSIVE,
-	DLD_ACTIVE
+	DLD_ACTIVE,
+	DLD_EXCLUSIVE
 } dld_passivestate_t;
 
 /*
@@ -256,6 +257,8 @@ extern void		dld_str_rx_unitdata(void *, mac_resource_handle_t,
 extern void		dld_str_notify_ind(dld_str_t *);
 extern mac_tx_cookie_t	str_mdata_fastpath_put(dld_str_t *, mblk_t *,
     uintptr_t, uint16_t);
+extern mac_tx_cookie_t	str_mdata_raw_fastpath_put(dld_str_t *, mblk_t *,
+    uintptr_t, uint16_t);
 extern int		dld_flow_ctl_callb(dld_str_t *, uint64_t,
     int (*func)(), void *);
 
diff --git a/usr/src/uts/common/sys/dld_ioc.h b/usr/src/uts/common/sys/dld_ioc.h
index 2f519a8eda..093a4dc0c3 100644
--- a/usr/src/uts/common/sys/dld_ioc.h
+++ b/usr/src/uts/common/sys/dld_ioc.h
@@ -20,6 +20,7 @@
  */
 /*
  * Copyright (c) 2008, 2010, Oracle and/or its affiliates. All rights reserved.
+ * Copyright 2015 Joyent, Inc.
  */
 
 #ifndef	_SYS_DLD_IOC_H
@@ -59,6 +60,7 @@ extern "C" {
 #define	IPTUN_IOC	0x454A
 #define	BRIDGE_IOC	0xB81D
 #define	IBPART_IOC	0x6171
+#define	OVERLAY_IOC	0x2005
 
 /* GLDv3 modules use these macros to generate unique ioctl commands */
 #define	DLDIOC(cmdid)		DLD_IOC_CMD(DLD_IOC, (cmdid))
@@ -68,6 +70,7 @@ extern "C" {
 #define	IPTUNIOC(cmdid)		DLD_IOC_CMD(IPTUN_IOC, (cmdid))
 #define	BRIDGEIOC(cmdid)	DLD_IOC_CMD(BRIDGE_IOC, (cmdid))
 #define	IBPARTIOC(cmdid)	DLD_IOC_CMD(IBPART_IOC, (cmdid))
+#define	OVERLAYIOC(cmdid)	DLD_IOC_CMD(OVERLAY_IOC, (cmdid))
 
 #ifdef _KERNEL
 
diff --git a/usr/src/uts/common/sys/dlpi.h b/usr/src/uts/common/sys/dlpi.h
index 5bc2bd41c5..dddac5b878 100644
--- a/usr/src/uts/common/sys/dlpi.h
+++ b/usr/src/uts/common/sys/dlpi.h
@@ -107,6 +107,7 @@ typedef struct dl_ipnetinfo {
 #define	DL_PASSIVE_REQ		0x114	/* Allow access to aggregated link */
 #define	DL_INTR_MODE_REQ	0x115	/* Request Rx processing in INTR mode */
 #define	DL_NOTIFY_CONF		0x116	/* Notification from upstream */
+#define	DL_EXCLUSIVE_REQ	0x117	/* Make bind active */
 
 /*
  * Primitives used for Connectionless Service
@@ -388,6 +389,7 @@ typedef struct dl_ipnetinfo {
 #define	DL_PROMISC_PHYS		0x01	/* promiscuous mode at phys level */
 #define	DL_PROMISC_SAP		0x02	/* promiscuous mode at sap level */
 #define	DL_PROMISC_MULTI	0x03	/* promiscuous mode for multicast */
+#define	DL_PROMISC_RX_ONLY	0x04	/* above only enabled for rx */
 
 /*
  * DLPI notification codes for DL_NOTIFY_REQ primitives.
@@ -1107,6 +1109,13 @@ typedef struct {
 } dl_intr_mode_req_t;
 
 /*
+ * DL_EXCLUSIVE_REQ, M_PROTO type
+ */
+typedef struct {
+	t_uscalar_t	dl_primitive;
+} dl_exclusive_req_t;
+
+/*
  *	CONNECTION-ORIENTED SERVICE PRIMITIVES
  */
 
@@ -1528,6 +1537,7 @@ union DL_primitives {
 	dl_control_ack_t	control_ack;
 	dl_passive_req_t	passive_req;
 	dl_intr_mode_req_t	intr_mode_req;
+	dl_exclusive_req_t	exclusive_req;
 };
 
 #define	DL_INFO_REQ_SIZE	sizeof (dl_info_req_t)
@@ -1596,6 +1606,7 @@ union DL_primitives {
 #define	DL_CONTROL_ACK_SIZE	sizeof (dl_control_ack_t)
 #define	DL_PASSIVE_REQ_SIZE	sizeof (dl_passive_req_t)
 #define	DL_INTR_MODE_REQ_SIZE	sizeof (dl_intr_mode_req_t)
+#define	DL_EXCLUSIVE_REQ_SIZE	sizeof (dl_exclusive_req_t)
 
 #ifdef	_KERNEL
 /*
diff --git a/usr/src/uts/common/sys/dls.h b/usr/src/uts/common/sys/dls.h
index 6bd2bbe35a..8e99d0e9d8 100644
--- a/usr/src/uts/common/sys/dls.h
+++ b/usr/src/uts/common/sys/dls.h
@@ -21,6 +21,7 @@
 /*
  * Copyright 2009 Sun Microsystems, Inc.  All rights reserved.
  * Use is subject to license terms.
+ * Copyright 2011 Joyent, Inc.  All rights reserved.
  */
 
 #ifndef	_SYS_DLS_H
@@ -85,6 +86,7 @@ typedef struct dls_link_s	dls_link_t;
 #define	DLS_PROMISC_SAP		0x00000001
 #define	DLS_PROMISC_MULTI	0x00000002
 #define	DLS_PROMISC_PHYS	0x00000004
+#define	DLS_PROMISC_RX_ONLY	0x00000008
 
 extern int	dls_open(dls_link_t *, dls_dl_handle_t, dld_str_t *);
 extern void	dls_close(dld_str_t *);
@@ -106,11 +108,13 @@ extern void	str_notify(void *, mac_notify_type_t);
 
 extern int		dls_devnet_open(const char *,
 			    dls_dl_handle_t *, dev_t *);
+extern int		dls_devnet_open_in_zone(const char *,
+			    dls_dl_handle_t *, dev_t *, zoneid_t);
 extern void		dls_devnet_close(dls_dl_handle_t);
 extern boolean_t	dls_devnet_rebuild();
 
 extern int		dls_devnet_rename(datalink_id_t, datalink_id_t,
-			    const char *);
+			    const char *, boolean_t);
 extern int		dls_devnet_create(mac_handle_t, datalink_id_t,
 			    zoneid_t);
 extern int		dls_devnet_destroy(mac_handle_t, datalink_id_t *,
@@ -127,7 +131,7 @@ extern uint16_t		dls_devnet_vid(dls_dl_handle_t);
 extern datalink_id_t	dls_devnet_linkid(dls_dl_handle_t);
 extern int		dls_devnet_dev2linkid(dev_t, datalink_id_t *);
 extern int		dls_devnet_phydev(datalink_id_t, dev_t *);
-extern int		dls_devnet_setzid(dls_dl_handle_t, zoneid_t);
+extern int		dls_devnet_setzid(dls_dl_handle_t, zoneid_t, boolean_t);
 extern zoneid_t		dls_devnet_getzid(dls_dl_handle_t);
 extern zoneid_t		dls_devnet_getownerzid(dls_dl_handle_t);
 extern boolean_t	dls_devnet_islinkvisible(datalink_id_t, zoneid_t);
@@ -141,6 +145,8 @@ extern int		dls_mgmt_update(const char *, uint32_t, boolean_t,
 extern int		dls_mgmt_get_linkinfo(datalink_id_t, char *,
 			    datalink_class_t *, uint32_t *, uint32_t *);
 extern int		dls_mgmt_get_linkid(const char *, datalink_id_t *);
+extern int		dls_mgmt_get_linkid_in_zone(const char *,
+    datalink_id_t *, zoneid_t);
 extern datalink_id_t	dls_mgmt_get_next(datalink_id_t, datalink_class_t,
 			    datalink_media_t, uint32_t);
 extern int		dls_devnet_macname2linkid(const char *,
diff --git a/usr/src/uts/common/sys/dls_impl.h b/usr/src/uts/common/sys/dls_impl.h
index 60f51c47b5..d502b36a2d 100644
--- a/usr/src/uts/common/sys/dls_impl.h
+++ b/usr/src/uts/common/sys/dls_impl.h
@@ -21,6 +21,7 @@
 /*
  * Copyright 2009 Sun Microsystems, Inc.  All rights reserved.
  * Use is subject to license terms.
+ * Copyright 2011 Joyent, Inc.  All rights reserved.
  */
 
 #ifndef	_SYS_DLS_IMPL_H
@@ -61,6 +62,7 @@ struct dls_link_s {				/* Protected by */
 	uint_t			dl_zone_ref;
 	link_tagmode_t		dl_tagmode;		/* atomic */
 	uint_t			dl_nonip_cnt;		/* SL */
+	uint_t			dl_exclusive;		/* SL */
 };
 
 typedef struct dls_head_s {
@@ -96,7 +98,8 @@ extern void		dls_create_str_kstats(dld_str_t *);
 extern int		dls_stat_update(kstat_t *, dls_link_t *, int);
 extern int		dls_stat_create(const char *, int, const char *,
 			    zoneid_t, int (*)(struct kstat *, int), void *,
-			    kstat_t **);
+			    kstat_t **, zoneid_t);
+extern void	dls_stat_delete(kstat_t *);
 
 extern int		dls_devnet_open_by_dev(dev_t, dls_link_t **,
 			    dls_dl_handle_t *);
@@ -126,6 +129,7 @@ extern void		dls_mgmt_init(void);
 extern void		dls_mgmt_fini(void);
 
 extern int		dls_mgmt_get_phydev(datalink_id_t, dev_t *);
+extern int		dls_exclusive_set(dld_str_t *, boolean_t);
 
 #ifdef	__cplusplus
 }
diff --git a/usr/src/uts/common/sys/dls_mgmt.h b/usr/src/uts/common/sys/dls_mgmt.h
index b4032c24d6..f747e29def 100644
--- a/usr/src/uts/common/sys/dls_mgmt.h
+++ b/usr/src/uts/common/sys/dls_mgmt.h
@@ -20,6 +20,7 @@
  */
 /*
  * Copyright (c) 2008, 2010, Oracle and/or its affiliates. All rights reserved.
+ * Copyright 2015, Joyent, Inc.
  */
 
 #ifndef	_DLS_MGMT_H
@@ -46,13 +47,15 @@ typedef enum {
 	DATALINK_CLASS_SIMNET		= 0x20,
 	DATALINK_CLASS_BRIDGE		= 0x40,
 	DATALINK_CLASS_IPTUN		= 0x80,
-	DATALINK_CLASS_PART		= 0x100
+	DATALINK_CLASS_PART		= 0x100,
+	DATALINK_CLASS_OVERLAY		= 0x200
 } datalink_class_t;
 
 #define	DATALINK_CLASS_ALL	(DATALINK_CLASS_PHYS |	\
 	DATALINK_CLASS_VLAN | DATALINK_CLASS_AGGR | DATALINK_CLASS_VNIC | \
 	DATALINK_CLASS_ETHERSTUB | DATALINK_CLASS_SIMNET | \
-	DATALINK_CLASS_BRIDGE | DATALINK_CLASS_IPTUN | DATALINK_CLASS_PART)
+	DATALINK_CLASS_BRIDGE | DATALINK_CLASS_IPTUN | DATALINK_CLASS_PART | \
+	DATALINK_CLASS_OVERLAY)
 
 /*
  * A combination of flags and media.
@@ -165,6 +168,7 @@ typedef struct dlmgmt_door_getname {
 typedef struct dlmgmt_door_getlinkid {
 	int			ld_cmd;
 	char			ld_link[MAXLINKNAMELEN];
+	zoneid_t		ld_zoneid;
 } dlmgmt_door_getlinkid_t;
 
 typedef struct dlmgmt_door_getnext_s {
diff --git a/usr/src/uts/common/sys/epoll.h b/usr/src/uts/common/sys/epoll.h
new file mode 100644
index 0000000000..f2e4b90ab7
--- /dev/null
+++ b/usr/src/uts/common/sys/epoll.h
@@ -0,0 +1,89 @@
+/*
+ * This file and its contents are supplied under the terms of the
+ * Common Development and Distribution License ("CDDL"), version 1.0.
+ * You may only use this file in accordance with the terms of version
+ * 1.0 of the CDDL.
+ *
+ * A full copy of the text of the CDDL should have accompanied this
+ * source.  A copy of the CDDL is also available via the Internet at
+ * http://www.illumos.org/license/CDDL.
+ */
+
+/*
+ * Copyright (c) 2014, Joyent, Inc.  All rights reserved.
+ */
+
+#ifndef _SYS_EPOLL_H
+#define	_SYS_EPOLL_H
+
+#include <sys/types.h>
+#include <sys/poll.h>
+
+#ifdef	__cplusplus
+extern "C" {
+#endif
+
+typedef union epoll_data {
+	void		*ptr;
+	int		fd;
+	uint32_t	u32;
+	uint64_t	u64;
+} epoll_data_t;
+
+#if _LONG_LONG_ALIGNMENT == 8 && _LONG_LONG_ALIGNMENT_32 == 4
+#pragma pack(4)
+#endif
+
+typedef struct epoll_event {
+	uint32_t	events;		/* events */
+	epoll_data_t	data;		/* user-specified data */
+} epoll_event_t;
+
+#if _LONG_LONG_ALIGNMENT == 8 && _LONG_LONG_ALIGNMENT_32 == 4
+#pragma pack()
+#endif
+
+/*
+ * Define the EPOLL* constants in terms of their poll(2)/poll(7) equivalents.
+ * Note that the values match the equivalents in Linux to allow for any binary
+ * compatibility layers to not need to translate them.
+ */
+#define	EPOLLIN		0x0001
+#define	EPOLLPRI	0x0002
+#define	EPOLLOUT	0x0004
+#define	EPOLLRDNORM	0x0040
+#define	EPOLLRDBAND	0x0080
+#define	EPOLLWRNORM	0x0100
+#define	EPOLLWRBAND	0x0200
+#define	EPOLLMSG	0x0400		/* not used */
+#define	EPOLLERR	0x0008
+#define	EPOLLHUP	0x0010
+#define	EPOLLRDHUP	0x2000
+
+#define	EPOLLWAKEUP	(1UL << 29)	/* no meaning; silently ignored */
+#define	EPOLLONESHOT	(1UL << 30)	/* translated to POLLONESHOT */
+#define	EPOLLET		(1UL << 31)	/* translated to POLLET */
+
+#define	EPOLL_CTL_ADD	1
+#define	EPOLL_CTL_DEL	2
+#define	EPOLL_CTL_MOD	3
+
+#define	EPOLL_CLOEXEC	02000000
+
+#if !defined(_KERNEL)
+
+extern int epoll_create(int size);
+extern int epoll_create1(int flags);
+extern int epoll_ctl(int epfd, int op, int fd, struct epoll_event *event);
+extern int epoll_wait(int epfd, struct epoll_event *events,
+    int maxevents, int timeout);
+extern int epoll_pwait(int epfd, struct epoll_event *events,
+    int maxevents, int timeout, const sigset_t *sigmask);
+
+#endif /* !_KERNEL */
+
+#ifdef	__cplusplus
+}
+#endif
+
+#endif	/* _SYS_EPOLL_H */
diff --git a/usr/src/uts/common/sys/eventfd.h b/usr/src/uts/common/sys/eventfd.h
new file mode 100644
index 0000000000..1b0d961b0b
--- /dev/null
+++ b/usr/src/uts/common/sys/eventfd.h
@@ -0,0 +1,68 @@
+/*
+ * This file and its contents are supplied under the terms of the
+ * Common Development and Distribution License ("CDDL"), version 1.0.
+ * You may only use this file in accordance with the terms of version
+ * 1.0 of the CDDL.
+ *
+ * A full copy of the text of the CDDL should have accompanied this
+ * source.  A copy of the CDDL is also available via the Internet at
+ * http://www.illumos.org/license/CDDL.
+ */
+
+/*
+ * Copyright (c) 2015 Joyent, Inc.  All rights reserved.
+ */
+
+/*
+ * Header file to support for the eventfd facility.  Note that this facility
+ * is designed to be binary compatible with the Linux eventfd facility; values
+ * for constants here should therefore exactly match those found in Linux, and
+ * this facility shouldn't be extended independently of Linux.
+ */
+
+#ifndef _SYS_EVENTFD_H
+#define	_SYS_EVENTFD_H
+
+#include <sys/types.h>
+
+#ifdef	__cplusplus
+extern "C" {
+#endif
+
+typedef uint64_t eventfd_t;
+
+/*
+ * To assure binary compatibility with Linux, these values are fixed at their
+ * Linux equivalents, not their native ones.
+ */
+#define	EFD_CLOEXEC		02000000		/* LX_O_CLOEXEC */
+#define	EFD_NONBLOCK		04000			/* LX_O_NONBLOCK */
+#define	EFD_SEMAPHORE		1
+
+/*
+ * These ioctl values are specific to the native implementation; applications
+ * shouldn't be using them directly, and they should therefore be safe to
+ * change without breaking apps.
+ */
+#define	EVENTFDIOC		(('e' << 24) | ('f' << 16) | ('d' << 8))
+#define	EVENTFDIOC_SEMAPHORE	(EVENTFDIOC | 1)	/* toggle sem state */
+
+#ifndef _KERNEL
+
+extern int eventfd(unsigned int, int);
+extern int eventfd_read(int, eventfd_t *);
+extern int eventfd_write(int, eventfd_t);
+
+#else
+
+#define	EVENTFDMNRN_EVENTFD	0
+#define	EVENTFDMNRN_CLONE	1
+#define	EVENTFD_VALMAX		(ULLONG_MAX - 1ULL)
+
+#endif /* _KERNEL */
+
+#ifdef	__cplusplus
+}
+#endif
+
+#endif	/* _SYS_EVENTFD_H */
diff --git a/usr/src/uts/common/sys/exec.h b/usr/src/uts/common/sys/exec.h
index d36bc20481..de01910b5a 100644
--- a/usr/src/uts/common/sys/exec.h
+++ b/usr/src/uts/common/sys/exec.h
@@ -26,6 +26,10 @@
 /*	Copyright (c) 1984, 1986, 1987, 1988, 1989 AT&T	*/
 /*	  All Rights Reserved  	*/
 
+/*
+ * Copyright 2015, Joyent, Inc.
+ */
+
 #ifndef _SYS_EXEC_H
 #define	_SYS_EXEC_H
 
@@ -102,10 +106,12 @@ typedef struct uarg {
 	vnode_t	*ex_vp;
 	char	*emulator;
 	char	*brandname;
+	const char *brand_nroot;
 	char	*auxp_auxflags; /* addr of auxflags auxv on the user stack */
 	char	*auxp_brand; /* address of first brand auxv on user stack */
 	cred_t	*pfcred;
 	boolean_t scrubenv;
+	uintptr_t maxstack;
 } uarg_t;
 
 /*
@@ -174,7 +180,7 @@ struct execsw {
 	int	(*exec_func)(struct vnode *vp, struct execa *uap,
 		    struct uarg *args, struct intpdata *idata, int level,
 		    long *execsz, int setid, caddr_t exec_file,
-		    struct cred *cred, int brand_action);
+		    struct cred *cred, int *brand_action);
 	int	(*exec_core)(struct vnode *vp, struct proc *p,
 		    struct cred *cred, rlim64_t rlimit, int sig,
 		    core_content_t content);
@@ -212,7 +218,7 @@ extern int exec_common(const char *fname, const char **argp,
     const char **envp, int brand_action);
 extern int gexec(vnode_t **vp, struct execa *uap, struct uarg *args,
     struct intpdata *idata, int level, long *execsz, caddr_t exec_file,
-    struct cred *cred, int brand_action);
+    struct cred *cred, int *brand_action);
 extern struct execsw *allocate_execsw(char *name, char *magic,
     size_t magic_size);
 extern struct execsw *findexecsw(char *magic);
@@ -237,16 +243,18 @@ extern void exec_set_sp(size_t);
  * when compiling the 32-bit compatability elf code in the elfexec module.
  */
 extern int elfexec(vnode_t *, execa_t *, uarg_t *, intpdata_t *, int,
-    long *, int, caddr_t, cred_t *, int);
+    long *, int, caddr_t, cred_t *, int *);
 extern int mapexec_brand(vnode_t *, uarg_t *, Ehdr *, Addr *,
-    intptr_t *, caddr_t, int *, caddr_t *, caddr_t *, size_t *, uintptr_t *);
+    intptr_t *, caddr_t, char **, caddr_t *, caddr_t *, size_t *,
+    uintptr_t *, uintptr_t *);
 #endif /* !_ELF32_COMPAT */
 
 #if defined(_LP64)
 extern int elf32exec(vnode_t *, execa_t *, uarg_t *, intpdata_t *, int,
-    long *, int, caddr_t, cred_t *, int);
+    long *, int, caddr_t, cred_t *, int *);
 extern int mapexec32_brand(vnode_t *, uarg_t *, Elf32_Ehdr *, Elf32_Addr *,
-    intptr_t *, caddr_t, int *, caddr_t *, caddr_t *, size_t *, uintptr_t *);
+    intptr_t *, caddr_t, char **, caddr_t *, caddr_t *, size_t *,
+    uintptr_t *, uintptr_t *);
 #endif  /* _LP64 */
 
 /*
diff --git a/usr/src/uts/common/sys/fcntl.h b/usr/src/uts/common/sys/fcntl.h
index aa74cab8b5..5ddade90e4 100644
--- a/usr/src/uts/common/sys/fcntl.h
+++ b/usr/src/uts/common/sys/fcntl.h
@@ -37,6 +37,7 @@
  */
 
 /* Copyright (c) 2013, OmniTI Computer Consulting, Inc. All rights reserved. */
+/* Copyright 2015, Joyent, Inc. */
 
 #ifndef	_SYS_FCNTL_H
 #define	_SYS_FCNTL_H
@@ -145,6 +146,7 @@ extern "C" {
  * the large and small file environments; therefore, the #defined values must
  * as well.
  * The NBMAND forms are private and should not be used.
+ * The FLOCK forms are also private and should not be used.
  */
 
 #if defined(_LP64) || _FILE_OFFSET_BITS == 32
@@ -155,6 +157,13 @@ extern "C" {
 #define	F_FREESP	11	/* Free file space */
 #define	F_GETLK		14	/* Get file lock */
 #define	F_SETLK_NBMAND	42	/* private */
+#if !defined(_STRICT_SYMBOLS)
+#define	F_OFD_GETLK	47	/* Get file lock owned by file */
+#define	F_OFD_SETLK	48	/* Set file lock owned by file */
+#define	F_OFD_SETLKW	49	/* Set file lock owned by file and wait */
+#define	F_FLOCK		53	/* private - set flock owned by file */
+#define	F_FLOCKW	54	/* private - set flock owned by file and wait */
+#endif	/* _STRICT_SYMBOLS */
 #else
 /* ILP32 large file application compilation environment version */
 #define	F_SETLK		34	/* Set file lock */
@@ -163,6 +172,13 @@ extern "C" {
 #define	F_FREESP	27	/* Free file space */
 #define	F_GETLK		33	/* Get file lock */
 #define	F_SETLK_NBMAND	44	/* private */
+#if !defined(_STRICT_SYMBOLS)
+#define	F_OFD_GETLK	50	/* Get file lock owned by file */
+#define	F_OFD_SETLK	51	/* Set file lock owned by file */
+#define	F_OFD_SETLKW	52	/* Set file lock owned by file and wait */
+#define	F_FLOCK		55	/* private - set flock owned by file */
+#define	F_FLOCKW	56	/* private - set flock owned by file and wait */
+#endif	/* _STRICT_SYMBOLS */
 #endif /* _LP64 || _FILE_OFFSET_BITS == 32 */
 
 #if 	defined(_LARGEFILE64_SOURCE)
@@ -180,6 +196,13 @@ extern "C" {
 #define	F_FREESP64	27	/* Free file space */
 #define	F_GETLK64	33	/* Get file lock */
 #define	F_SETLK64_NBMAND	44	/* private */
+#if !defined(_STRICT_SYMBOLS)
+#define	F_OFD_GETLK64	50	/* Get file lock owned by file */
+#define	F_OFD_SETLK64	51	/* Set file lock owned by file */
+#define	F_OFD_SETLKW64	52	/* Set file lock owned by file and wait */
+#define	F_FLOCK64	55	/* private - set flock owned by file */
+#define	F_FLOCKW64	56	/* private - set flock owned by file and wait */
+#endif	/* _STRICT_SYMBOLS */
 #else
 #define	F_SETLK64	6	/* Set file lock */
 #define	F_SETLKW64	7	/* Set file lock and wait */
@@ -187,6 +210,13 @@ extern "C" {
 #define	F_FREESP64	11	/* Free file space */
 #define	F_GETLK64	14	/* Get file lock */
 #define	F_SETLK64_NBMAND	42	/* private */
+#if !defined(_STRICT_SYMBOLS)
+#define	F_OFD_GETLK64	47	/* Get file lock owned by file */
+#define	F_OFD_SETLK64	48	/* Set file lock owned by file */
+#define	F_OFD_SETLKW64	49	/* Set file lock owned by file and wait */
+#define	F_FLOCK64	53	/* private - set flock owned by file */
+#define	F_FLOCKW64	54	/* private - set flock owned by file and wait */
+#endif /* _STRICT_SYMBOLS */
 #endif /* !_LP64 || _KERNEL */
 
 #endif /* _LARGEFILE64_SOURCE */
diff --git a/usr/src/uts/common/sys/file.h b/usr/src/uts/common/sys/file.h
index 03acc088c2..7e297042af 100644
--- a/usr/src/uts/common/sys/file.h
+++ b/usr/src/uts/common/sys/file.h
@@ -27,6 +27,7 @@
 /*	  All Rights Reserved  	*/
 
 /* Copyright (c) 2013, OmniTI Computer Consulting, Inc. All rights reserved. */
+/* Copyright 2015 Joyent, Inc. */
 
 #ifndef _SYS_FILE_H
 #define	_SYS_FILE_H
@@ -54,7 +55,7 @@ extern "C" {
  */
 /*
  * One file structure is allocated for each open/creat/pipe call.
- * Main use is to hold the read/write pointer associated with
+ * Main use is to hold the read/write pointer (and OFD locks) associated with
  * each open file.
  */
 typedef struct file {
@@ -66,6 +67,7 @@ typedef struct file {
 	struct cred	*f_cred;	/* credentials of user who opened it */
 	struct f_audit_data	*f_audit_data;	/* file audit data */
 	int		f_count;	/* reference count */
+	struct filock *f_filock;	/* ptr to single lock_descriptor_t */
 } file_t;
 
 /*
@@ -120,6 +122,11 @@ typedef struct fpollinfo {
 #ifdef _KERNEL
 
 /*
+ * This is a flag that is set on f_flag2, but is never user-visible
+ */
+#define	FEPOLLED	0x8000
+
+/*
  * Fake flags for driver ioctl calls to inform them of the originating
  * process' model.  See <sys/model.h>
  *
@@ -168,6 +175,19 @@ typedef struct fpollinfo {
 #define	L_SET	0	/* for lseek */
 #endif /* L_SET */
 
+/*
+ * For flock(3C).  These really don't belong here but for historical reasons
+ * the interface defines them to be here.
+ */
+#define	LOCK_SH	1
+#define	LOCK_EX	2
+#define	LOCK_NB	4
+#define	LOCK_UN	8
+
+#if !defined(_STRICT_SYMBOLS)
+extern int flock(int, int);
+#endif
+
 #if defined(_KERNEL)
 
 /*
diff --git a/usr/src/uts/common/sys/flock.h b/usr/src/uts/common/sys/flock.h
index 3ea7afb23b..39542a4c16 100644
--- a/usr/src/uts/common/sys/flock.h
+++ b/usr/src/uts/common/sys/flock.h
@@ -29,6 +29,7 @@
  */
 /*
  * Copyright 2011 Nexenta Systems, Inc.  All rights reserved.
+ * Copyright 2015 Joyent, Inc.
  */
 
 #ifndef _SYS_FLOCK_H
@@ -41,6 +42,9 @@
 #include <sys/callb.h>
 #include <sys/param.h>
 #include <sys/zone.h>
+#if defined(_KERNEL)
+#include <sys/file.h>
+#endif
 
 #ifdef	__cplusplus
 extern "C" {
@@ -221,6 +225,8 @@ typedef struct locklist {
 #define	FLK_QUERY_ACTIVE	0x1
 #define	FLK_QUERY_SLEEPING	0x2
 
+int	ofdlock(file_t *, int, struct flock64 *, int, u_offset_t);
+void	ofdcleanlock(file_t *);
 int	reclock(struct vnode *, struct flock64 *, int, int, u_offset_t,
 		flk_callback_t *);
 int	chklock(struct vnode *, int, u_offset_t, ssize_t, int,
diff --git a/usr/src/uts/common/sys/flock_impl.h b/usr/src/uts/common/sys/flock_impl.h
index d088192025..b1be1d598a 100644
--- a/usr/src/uts/common/sys/flock_impl.h
+++ b/usr/src/uts/common/sys/flock_impl.h
@@ -22,13 +22,12 @@
 /*
  * Copyright 2004 Sun Microsystems, Inc.  All rights reserved.
  * Use is subject to license terms.
+ * Copyright 2015 Joyent, Inc.
  */
 
 #ifndef _SYS_FLOCK_IMPL_H
 #define	_SYS_FLOCK_IMPL_H
 
-#pragma ident	"%Z%%M%	%I%	%E% SMI"
-
 #include <sys/types.h>
 #include <sys/fcntl.h>		/* flock definition */
 #include <sys/file.h>		/* FREAD etc */
@@ -83,6 +82,7 @@ struct lock_descriptor {
 		flk_nlm_status_t l_nlm_state;	/* state of NLM server */
 		flk_callback_t	*l_callbacks;	/* callbacks, or NULL */
 		zoneid_t	l_zoneid;	/* zone of request */
+		file_t		*l_ofd;		/* OFD-style reference */
 };
 
 typedef struct 	lock_descriptor	lock_descriptor_t;
@@ -204,7 +204,8 @@ graph_t *flk_get_lock_graph(vnode_t *, int);
 
 #define	SAME_OWNER(lock1, lock2)	\
 	(((lock1)->l_flock.l_pid == (lock2)->l_flock.l_pid) && \
-		((lock1)->l_flock.l_sysid == (lock2)->l_flock.l_sysid))
+		((lock1)->l_flock.l_sysid == (lock2)->l_flock.l_sysid) && \
+		((lock1)->l_ofd == (lock2)->l_ofd))
 
 #define	COLORED(vertex)		((vertex)->l_color == (vertex)->l_graph->mark)
 #define	COLOR(vertex)		((vertex)->l_color = (vertex)->l_graph->mark)
diff --git a/usr/src/uts/common/sys/frameio.h b/usr/src/uts/common/sys/frameio.h
new file mode 100644
index 0000000000..54e6dbeedf
--- /dev/null
+++ b/usr/src/uts/common/sys/frameio.h
@@ -0,0 +1,107 @@
+/*
+ * This file and its contents are supplied under the terms of the
+ * Common Development and Distribution License ("CDDL"), version 1.0.
+ * You may only use this file in accordance with the terms of version
+ * 1.0 of the CDDL.
+ *
+ * A full copy of the text of the CDDL should have accompanied this
+ * source.  A copy of the CDDL is also available via the Internet at
+ * http://www.illumos.org/license/CDDL.
+ */
+
+/*
+ * Copyright (c) 2014, Joyent, Inc.  All rights reserved.
+ */
+
+#ifndef _SYS_FRAMEIO_H
+#define	_SYS_FRAMEIO_H
+
+/*
+ * Frame I/O definitions
+ */
+
+#include <sys/types.h>
+
+#ifdef _KERNEL
+/* Kernel only headers */
+#include <sys/stream.h>
+#endif	/* _KERNEL */
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+/*
+ * An individual frame vector component. Collections of these are used to make
+ * ioctls.
+ */
+typedef struct framevec {
+	void	*fv_buf;	/* Buffer with data */
+	size_t	fv_buflen;	/* Size of the buffer */
+	size_t	fv_actlen;	/* Amount of buffer consumed, ignore on error */
+} framevec_t;
+
+/*
+ * The base unit used with frameio.
+ */
+typedef struct frameio {
+	uint_t	fio_version;	/* Should always be FRAMEIO_CURRENT_VERSION */
+	uint_t	fio_nvpf;	/* How many vectors make up one frame */
+	uint_t	fio_nvecs;	/* The total number of vectors */
+	framevec_t fio_vecs[];	/* C99 VLA */
+} frameio_t;
+
+
+#define	FRAMEIO_VERSION_ONE	1
+#define	FRAMEIO_CURRENT_VERSION	FRAMEIO_VERSION_ONE
+
+#define	FRAMEIO_NVECS_MAX	32
+
+/*
+ * Definitions for kernel modules to include as helpers. These are consolidation
+ * private.
+ */
+#ifdef _KERNEL
+
+/*
+ * 32-bit versions for 64-bit kernels
+ */
+typedef struct framevec32 {
+	caddr32_t fv_buf;
+	size32_t fv_buflen;
+	size32_t fv_actlen;
+} framevec32_t;
+
+typedef struct frameio32 {
+	uint_t fio_version;
+	uint_t fio_vecspframe;
+	uint_t fio_nvecs;
+	framevec32_t fio_vecs[];
+} frameio32_t;
+
+/*
+ * Describe the different ways that vectors should map to frames.
+ */
+typedef enum frameio_write_mblk_map {
+	MAP_BLK_FRAME
+} frameio_write_mblk_map_t;
+
+int frameio_init(void);
+void frameio_fini(void);
+frameio_t *frameio_alloc(int);
+void frameio_free(frameio_t *);
+int frameio_hdr_copyin(frameio_t *, int, const void *, uint_t);
+int frameio_mblk_chain_read(frameio_t *, mblk_t **, int *, int);
+int frameio_mblk_chain_write(frameio_t *, frameio_write_mblk_map_t, mblk_t *,
+    int *, int);
+int frameio_hdr_copyout(frameio_t *, int, void *, uint_t);
+size_t frameio_frame_length(frameio_t *, framevec_t *);
+void frameio_mark_consumed(frameio_t *, int);
+
+#endif /* _KERNEL */
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif /* _SYS_FRAMEIO_H */
diff --git a/usr/src/uts/common/sys/fs/bootfs_impl.h b/usr/src/uts/common/sys/fs/bootfs_impl.h
new file mode 100644
index 0000000000..5726f1428a
--- /dev/null
+++ b/usr/src/uts/common/sys/fs/bootfs_impl.h
@@ -0,0 +1,81 @@
+/*
+ * This file and its contents are supplied under the terms of the
+ * Common Development and Distribution License ("CDDL"), version 1.0.
+ * You may only use this file in accordance with the terms of version
+ * 1.0 of the CDDL.
+ *
+ * A full copy of the text of the CDDL should have accompanied this
+ * source.  A copy of the CDDL is also available via the Internet at
+ * http://www.illumos.org/license/CDDL.
+ */
+
+/*
+ * Copyright (c) 2014 Joyent, Inc.  All rights reserved.
+ */
+
+#ifndef _SYS_FS_BOOTFS_IMPL_H
+#define	_SYS_FS_BOOTFS_IMPL_H
+
+#include <sys/types.h>
+#include <sys/list.h>
+#include <sys/avl.h>
+#include <sys/vnode.h>
+#include <sys/vfs_opreg.h>
+#include <sys/kstat.h>
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+/*
+ * The bootfs node is the file system specific version of the vnode for the
+ * bootfs file system. Because the bootfs file system is entirely a read-only
+ * file system, this structure requires no locking as the contents are
+ * immutable.
+ */
+typedef struct bootfs_node {
+	char			*bvn_name;	/* entry name */
+	struct vnode		*bvn_vnp;	/* Corresponding vnode */
+	avl_tree_t		bvn_dir;	/* directory entries, if VDIR */
+	avl_node_t		bvn_link;	/* dirent link */
+	list_node_t		bvn_alink;	/* link for all nodes */
+	uint64_t		bvn_addr;	/* Address in pmem */
+	uint64_t		bvn_size;	/* Size of the file */
+	struct bootfs_node	*bvn_parent;	/* .. */
+	vattr_t			bvn_attr;	/* attributes for the node */
+} bootfs_node_t;
+
+typedef struct bootfs_stat {
+	kstat_named_t	bfss_nfiles;
+	kstat_named_t	bfss_ndirs;
+	kstat_named_t	bfss_nbytes;
+	kstat_named_t	bfss_ndups;
+	kstat_named_t	bfss_ndiscards;
+} bootfs_stat_t;
+
+typedef struct bootfs {
+	vfs_t			*bfs_vfsp;
+	char			*bfs_mntpath;
+	bootfs_node_t		*bfs_rootvn;
+	kstat_t			*bfs_kstat;
+	list_t			bfs_nodes;
+	minor_t			bfs_minor;
+	uint_t			bfs_ninode;
+	bootfs_stat_t		bfs_stat;
+} bootfs_t;
+
+extern void bootfs_construct(bootfs_t *);
+extern void bootfs_destruct(bootfs_t *);
+extern int bootfs_node_constructor(void *, void *, int);
+extern void bootfs_node_destructor(void *, void *);
+
+extern struct vnodeops *bootfs_vnodeops;
+extern const fs_operation_def_t bootfs_vnodeops_template[];
+extern kmem_cache_t *bootfs_node_cache;
+extern major_t bootfs_major;
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif /* _SYS_FS_BOOTFS_IMPL_H */
diff --git a/usr/src/uts/common/sys/fs/hyprlofs.h b/usr/src/uts/common/sys/fs/hyprlofs.h
new file mode 100644
index 0000000000..b8c4149df2
--- /dev/null
+++ b/usr/src/uts/common/sys/fs/hyprlofs.h
@@ -0,0 +1,91 @@
+/*
+ * CDDL HEADER START
+ *
+ * The contents of this file are subject to the terms of the
+ * Common Development and Distribution License (the "License").
+ * You may not use this file except in compliance with the License.
+ *
+ * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
+ * or http://www.opensolaris.org/os/licensing.
+ * See the License for the specific language governing permissions
+ * and limitations under the License.
+ *
+ * When distributing Covered Code, include this CDDL HEADER in each
+ * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
+ * If applicable, add the following below this CDDL HEADER, with the
+ * fields enclosed by brackets "[]" replaced with your own identifying
+ * information: Portions Copyright [yyyy] [name of copyright owner]
+ *
+ * CDDL HEADER END
+ */
+
+/*
+ * Copyright 2012, Joyent, Inc.  All rights reserved.
+ */
+
+#ifndef	_SYS_FS_HYPRLOFS_H
+#define	_SYS_FS_HYPRLOFS_H
+
+#include <sys/param.h>
+
+#ifdef	__cplusplus
+extern "C" {
+#endif
+
+/*
+ * hyprlofs ioctl numbers.
+ */
+#define	HYPRLOFS_IOC	('H' << 8)
+
+#define	HYPRLOFS_ADD_ENTRIES	(HYPRLOFS_IOC | 1)
+#define	HYPRLOFS_RM_ENTRIES	(HYPRLOFS_IOC | 2)
+#define	HYPRLOFS_RM_ALL		(HYPRLOFS_IOC | 3)
+#define	HYPRLOFS_GET_ENTRIES	(HYPRLOFS_IOC | 4)
+
+typedef struct {
+	char	*hle_path;
+	uint_t	hle_plen;
+	char	*hle_name;
+	uint_t	hle_nlen;
+} hyprlofs_entry_t;
+
+typedef struct {
+	hyprlofs_entry_t	*hle_entries;
+	uint_t			hle_len;
+} hyprlofs_entries_t;
+
+typedef struct {
+	char		hce_path[MAXPATHLEN];
+	char		hce_name[MAXPATHLEN];
+} hyprlofs_curr_entry_t;
+
+typedef struct {
+	hyprlofs_curr_entry_t	*hce_entries;
+	uint_t			hce_cnt;
+} hyprlofs_curr_entries_t;
+
+#ifdef _KERNEL
+typedef struct {
+	caddr32_t	hle_path;
+	uint_t		hle_plen;
+	caddr32_t	hle_name;
+	uint_t		hle_nlen;
+} hyprlofs_entry32_t;
+
+typedef struct {
+	caddr32_t	hle_entries;
+	uint_t		hle_len;
+} hyprlofs_entries32_t;
+
+typedef struct {
+	caddr32_t	hce_entries;
+	uint_t		hce_cnt;
+} hyprlofs_curr_entries32_t;
+
+#endif /* _KERNEL */
+
+#ifdef	__cplusplus
+}
+#endif
+
+#endif	/* _SYS_FS_HYPRLOFS_H */
diff --git a/usr/src/uts/common/sys/fs/hyprlofs_info.h b/usr/src/uts/common/sys/fs/hyprlofs_info.h
new file mode 100644
index 0000000000..38389f77d9
--- /dev/null
+++ b/usr/src/uts/common/sys/fs/hyprlofs_info.h
@@ -0,0 +1,174 @@
+/*
+ * CDDL HEADER START
+ *
+ * The contents of this file are subject to the terms of the
+ * Common Development and Distribution License (the "License").
+ * You may not use this file except in compliance with the License.
+ *
+ * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
+ * or http://www.opensolaris.org/os/licensing.
+ * See the License for the specific language governing permissions
+ * and limitations under the License.
+ *
+ * When distributing Covered Code, include this CDDL HEADER in each
+ * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
+ * If applicable, add the following below this CDDL HEADER, with the
+ * fields enclosed by brackets "[]" replaced with your own identifying
+ * information: Portions Copyright [yyyy] [name of copyright owner]
+ *
+ * CDDL HEADER END
+ */
+/*
+ * Copyright 2012, Joyent, Inc. All rights reserved.
+ */
+
+#ifndef	_SYS_FS_HYPRLOFS_INFO_H
+#define	_SYS_FS_HYPRLOFS_INFO_H
+
+#include <sys/t_lock.h>
+#include <vm/seg.h>
+#include <vm/seg_vn.h>
+#include <sys/vfs_opreg.h>
+
+#ifdef	__cplusplus
+extern "C" {
+#endif
+
+/*
+ * hlnode is the file system dependent node for hyprlofs.
+ * It is modeled on the tmpfs tmpnode.
+ *
+ *	hln_rwlock protects access of the directory list at hln_dir
+ *	as well as syncronizing read/writes to directory hlnodes.
+ *	hln_tlock protects updates to hln_mode and hln_nlink.
+ *	hln_tlock doesn't require any hlnode locks.
+ */
+typedef struct hlnode {
+	struct hlnode	*hln_back;		/* linked list of hlnodes */
+	struct hlnode	*hln_forw;		/* linked list of hlnodes */
+	union {
+		struct {
+			struct hldirent	*un_dirlist; /* dirent list */
+			uint_t	un_dirents;	/* number of dirents */
+		} un_dirstruct;
+		vnode_t	*un_realvp;		/* real vnode */
+	} un_hlnode;
+	vnode_t 	*hln_vnode;		/* vnode for this hlnode */
+	int 		hln_gen;		/* pseudo gen num for hlfid */
+	int 		hln_looped;		/* flag indicating loopback */
+	vattr_t		hln_attr;		/* attributes */
+	krwlock_t	hln_rwlock;		/* rw - serialize mods and */
+						/* directory updates */
+	kmutex_t	hln_tlock;		/* time, flag, and nlink lock */
+} hlnode_t;
+
+/*
+ * hyprlofs per-mount data structure.
+ * All fields are protected by hlm_contents.
+ */
+typedef struct {
+	vfs_t		*hlm_vfsp;	/* filesystem's vfs struct */
+	hlnode_t	*hlm_rootnode;	/* root hlnode */
+	char 		*hlm_mntpath;	/* name of hyprlofs mount point */
+	dev_t		hlm_dev;	/* unique dev # of mounted `device' */
+	uint_t		hlm_gen;	/* pseudo generation number for files */
+	kmutex_t	hlm_contents;	/* lock for hlfsmount structure */
+} hlfsmount_t;
+
+/*
+ * hyprlofs directories are made up of a linked list of hldirent structures
+ * hanging off directory hlnodes.  File names are not fixed length,
+ * but are null terminated.
+ */
+typedef struct hldirent {
+	hlnode_t	*hld_hlnode;		/* hlnode for this file */
+	struct hldirent	*hld_next;		/* next directory entry */
+	struct hldirent	*hld_prev;		/* prev directory entry */
+	uint_t		hld_offset;		/* "offset" of dir entry */
+	uint_t		hld_hash;		/* a hash of td_name */
+	struct hldirent	*hld_link;		/* linked via the hash table */
+	hlnode_t	*hld_parent;		/* parent, dir we are in */
+	char		*hld_name;		/* must be null terminated */
+						/* max length is MAXNAMELEN */
+} hldirent_t;
+
+/*
+ * hlfid overlays the fid structure (for VFS_VGET)
+ */
+typedef struct {
+	uint16_t hlfid_len;
+	ino32_t	hlfid_ino;
+	int32_t	hlfid_gen;
+} hlfid_t;
+
+/*
+ * File system independent to hyprlofs conversion macros
+ */
+#define	VFSTOHLM(vfsp)		((hlfsmount_t *)(vfsp)->vfs_data)
+#define	VTOHLM(vp)		((hlfsmount_t *)(vp)->v_vfsp->vfs_data)
+#define	VTOHLN(vp)		((hlnode_t *)(vp)->v_data)
+#define	HLNTOV(tp)		((tp)->hln_vnode)
+#define	REALVP(vp)		((vnode_t *)VTOHLN(vp)->hln_realvp)
+#define	hlnode_hold(tp)		VN_HOLD(HLNTOV(tp))
+#define	hlnode_rele(tp)		VN_RELE(HLNTOV(tp))
+
+#define	hln_dir		un_hlnode.un_dirstruct.un_dirlist
+#define	hln_dirents	un_hlnode.un_dirstruct.un_dirents
+#define	hln_realvp	un_hlnode.un_realvp
+
+/*
+ * Attributes
+ */
+#define	hln_mask	hln_attr.va_mask
+#define	hln_type	hln_attr.va_type
+#define	hln_mode	hln_attr.va_mode
+#define	hln_uid		hln_attr.va_uid
+#define	hln_gid		hln_attr.va_gid
+#define	hln_fsid	hln_attr.va_fsid
+#define	hln_nodeid	hln_attr.va_nodeid
+#define	hln_nlink	hln_attr.va_nlink
+#define	hln_size	hln_attr.va_size
+#define	hln_atime	hln_attr.va_atime
+#define	hln_mtime	hln_attr.va_mtime
+#define	hln_ctime	hln_attr.va_ctime
+#define	hln_rdev	hln_attr.va_rdev
+#define	hln_blksize	hln_attr.va_blksize
+#define	hln_nblocks	hln_attr.va_nblocks
+#define	hln_seq		hln_attr.va_seq
+
+/*
+ * enums
+ */
+enum de_op	{ DE_CREATE, DE_MKDIR }; /* direnter ops */
+enum dr_op	{ DR_REMOVE, DR_RMDIR }; /* dirremove ops */
+
+/*
+ * hyprlofs_minfree is the amount (in pages) of anonymous memory that hyprlofs
+ * leaves free for the rest of the system. The default value for
+ * hyprlofs_minfree is btopr(HYPRLOFSMINFREE) but it can be patched to a
+ * different number of pages.  Since hyprlofs doesn't actually use much
+ * memory, its unlikely this ever needs to be patched.
+ */
+#define		HYPRLOFSMINFREE 8 * 1024 * 1024 /* 8 Megabytes */
+
+extern size_t  hyprlofs_minfree;		/* Anonymous memory in pages */
+
+extern	void	hyprlofs_node_init(hlfsmount_t *, hlnode_t *, vattr_t *,
+		    cred_t *);
+extern	int	hyprlofs_dirlookup(hlnode_t *, char *, hlnode_t **, cred_t *);
+extern	int	hyprlofs_dirdelete(hlnode_t *, hlnode_t *, char *, enum dr_op,
+		    cred_t *);
+extern	void	hyprlofs_dirinit(hlnode_t *, hlnode_t *);
+extern	void	hyprlofs_dirtrunc(hlnode_t *);
+extern	int	hyprlofs_taccess(void *, int, cred_t *);
+extern	int	hyprlofs_direnter(hlfsmount_t *, hlnode_t *, char *, enum de_op,
+		    vnode_t *, vattr_t *, hlnode_t **, cred_t *);
+
+extern struct vnodeops *hyprlofs_vnodeops;
+extern const struct fs_operation_def hyprlofs_vnodeops_template[];
+
+#ifdef	__cplusplus
+}
+#endif
+
+#endif	/* _SYS_FS_HYPRLOFS_INFO_H */
diff --git a/usr/src/uts/common/sys/fs/sdev_impl.h b/usr/src/uts/common/sys/fs/sdev_impl.h
index 19a147d88a..d5bd0162c7 100644
--- a/usr/src/uts/common/sys/fs/sdev_impl.h
+++ b/usr/src/uts/common/sys/fs/sdev_impl.h
@@ -35,6 +35,7 @@ extern "C" {
 #include <sys/vfs_opreg.h>
 #include <sys/list.h>
 #include <sys/nvpair.h>
+#include <sys/fs/sdev_plugin.h>
 
 /*
  * sdev_nodes are the file-system specific part of the
@@ -126,6 +127,21 @@ typedef struct sdev_local_data {
 	struct sdev_dprof sdev_lprof;	/* profile for multi-inst */
 } sdev_local_data_t;
 
+/* sdev_flags */
+typedef enum sdev_flags {
+	SDEV_BUILD =		0x0001,	/* directory cache out-of-date */
+	SDEV_GLOBAL =		0x0002,	/* global /dev nodes */
+	SDEV_PERSIST =		0x0004,	/* backing store persisted node */
+	SDEV_NO_NCACHE = 	0x0008,	/* do not include in neg. cache */
+	SDEV_DYNAMIC =		0x0010,	/* special-purpose vnode ops */
+					/* (ex: pts) */
+	SDEV_VTOR =		0x0020,	/* validate sdev_nodes during search */
+	SDEV_ATTR_INVALID =	0x0040,	/* invalid node attributes, */
+					/* need update */
+	SDEV_SUBDIR =		0x0080,	/* match all subdirs under here */
+	SDEV_ZONED =		0x0100	/* zoned subdir */
+} sdev_flags_t;
+
 /*
  * /dev filesystem sdev_node defines
  */
@@ -148,7 +164,7 @@ typedef struct sdev_node {
 	ino64_t		sdev_ino;	/* inode */
 	uint_t		sdev_nlink;	/* link count */
 	int		sdev_state;	/* state of this node */
-	int		sdev_flags;	/* flags bit */
+	sdev_flags_t	sdev_flags;	/* flags bit */
 
 	kmutex_t	sdev_lookup_lock; /* node creation synch lock */
 	kcondvar_t	sdev_lookup_cv;	/* node creation sync cv */
@@ -159,7 +175,7 @@ typedef struct sdev_node {
 		struct sdev_global_data	sdev_globaldata;
 		struct sdev_local_data	sdev_localdata;
 	} sdev_instance_data;
-
+	list_node_t	sdev_plist;	/* link on plugin list */
 	void		*sdev_private;
 } sdev_node_t;
 
@@ -190,29 +206,11 @@ typedef enum {
 	SDEV_READY
 } sdev_node_state_t;
 
-/* sdev_flags */
-#define	SDEV_BUILD		0x0001	/* directory cache out-of-date */
-#define	SDEV_GLOBAL		0x0002	/* global /dev nodes */
-#define	SDEV_PERSIST		0x0004	/* backing store persisted node */
-#define	SDEV_NO_NCACHE		0x0008	/* do not include in neg. cache */
-#define	SDEV_DYNAMIC		0x0010	/* special-purpose vnode ops */
-					/* (ex: pts) */
-#define	SDEV_VTOR		0x0020	/* validate sdev_nodes during search */
-#define	SDEV_ATTR_INVALID	0x0040	/* invalid node attributes, */
-					/* need update */
-#define	SDEV_SUBDIR		0x0080	/* match all subdirs under here */
-#define	SDEV_ZONED		0x0100  /* zoned subdir */
-
 /* sdev_lookup_flags */
 #define	SDEV_LOOKUP	0x0001	/* node creation in progress */
 #define	SDEV_READDIR	0x0002	/* VDIR readdir in progress */
 #define	SDEV_LGWAITING	0x0004	/* waiting for devfsadm completion */
 
-#define	SDEV_VTOR_INVALID	-1
-#define	SDEV_VTOR_SKIP		0
-#define	SDEV_VTOR_VALID		1
-#define	SDEV_VTOR_STALE		2
-
 /* convenient macros */
 #define	SDEV_IS_GLOBAL(dv)	\
 	(dv->sdev_flags & SDEV_GLOBAL)
@@ -364,8 +362,13 @@ extern void sdev_devfsadmd_thread(struct sdev_node *, struct sdev_node *,
 extern int devname_profile_update(char *, size_t);
 extern struct sdev_data *sdev_find_mntinfo(char *);
 void sdev_mntinfo_rele(struct sdev_data *);
+typedef void (*sdev_mnt_walk_f)(struct sdev_node *, void *);
+void sdev_mnt_walk(sdev_mnt_walk_f, void *);
 extern struct vnodeops *devpts_getvnodeops(void);
 extern struct vnodeops *devvt_getvnodeops(void);
+extern void sdev_plugin_nodeready(struct sdev_node *);
+extern int sdev_plugin_init(void);
+extern int sdev_plugin_fini(void);
 
 /*
  * boot states - warning, the ordering here is significant
@@ -510,6 +513,23 @@ extern void sdev_nc_path_exists(sdev_nc_list_t *, char *);
 extern void sdev_modctl_dump_files(void);
 
 /*
+ * plugin and legacy vtab stuff
+ */
+/* directory dependent vop table */
+typedef struct sdev_vop_table {
+	char *vt_name;				/* subdirectory name */
+	const fs_operation_def_t *vt_service;	/* vnodeops table */
+	struct vnodeops **vt_global_vops;	/* global container for vop */
+	int (*vt_vtor)(struct sdev_node *);	/* validate sdev_node */
+	int vt_flags;
+} sdev_vop_table_t;
+
+extern struct sdev_vop_table vtab[];
+extern struct vnodeops *sdev_get_vop(struct sdev_node *);
+extern void sdev_set_no_negcache(struct sdev_node *);
+extern void *sdev_get_vtor(struct sdev_node *dv);
+
+/*
  * globals
  */
 extern kmutex_t sdev_lock;
@@ -522,6 +542,7 @@ extern struct vnodeops		*devipnet_vnodeops;
 extern struct vnodeops		*devvt_vnodeops;
 extern struct sdev_data *sdev_origins; /* mount info for global /dev instance */
 extern struct vnodeops		*devzvol_vnodeops;
+extern int			sdev_vnodeops_tbl_size;
 
 extern const fs_operation_def_t	sdev_vnodeops_tbl[];
 extern const fs_operation_def_t	devpts_vnodeops_tbl[];
diff --git a/usr/src/uts/common/sys/fs/sdev_plugin.h b/usr/src/uts/common/sys/fs/sdev_plugin.h
new file mode 100644
index 0000000000..8783df58e6
--- /dev/null
+++ b/usr/src/uts/common/sys/fs/sdev_plugin.h
@@ -0,0 +1,106 @@
+/*
+ * This file and its contents are supplied under the terms of the
+ * Common Development and Distribution License ("CDDL"), version 1.0.
+ * You may only use this file in accordance with the terms of version
+ * 1.0 of the CDDL.
+ *
+ * A full copy of the text of the CDDL should have accompanied this
+ * source.  A copy of the CDDL is also available via the Internet at
+ * http://www.illumos.org/license/CDDL.
+ */
+
+/*
+ * Copyright (c) 2014 Joyent, Inc.  All rights reserved.
+ */
+
+#ifndef _SYS_SDEV_PLUGIN_H
+#define	_SYS_SDEV_PLUGIN_H
+
+/*
+ * Kernel sdev plugin interface
+ */
+
+#ifdef _KERNEL
+
+#include <sys/types.h>
+#include <sys/stat.h>
+#include <sys/vnode.h>
+
+#endif	/* _KERNEL */
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+#ifdef _KERNEL
+
+typedef uintptr_t sdev_plugin_hdl_t;
+typedef uintptr_t sdev_ctx_t;
+
+/*
+ * Valid return values for sdev_plugin_validate_t.
+ */
+typedef enum sdev_plugin_validate {
+	SDEV_VTOR_INVALID = -1,
+	SDEV_VTOR_SKIP = 0,
+	SDEV_VTOR_VALID	= 1,
+	SDEV_VTOR_STALE	= 2
+} sdev_plugin_validate_t;
+
+/*
+ * Valid flags
+ */
+typedef enum sdev_plugin_flags {
+	SDEV_PLUGIN_NO_NCACHE = 0x1,
+	SDEV_PLUGIN_SUBDIR = 0x2
+} sdev_plugin_flags_t;
+
+#define	SDEV_PLUGIN_FLAGS_MASK	0x3
+
+/*
+ * Functions a module must implement
+ */
+typedef sdev_plugin_validate_t (*sp_valid_f)(sdev_ctx_t);
+typedef int (*sp_filldir_f)(sdev_ctx_t);
+typedef void (*sp_inactive_f)(sdev_ctx_t);
+
+#define	SDEV_PLUGIN_VERSION	1
+
+typedef struct sdev_plugin_ops {
+	int spo_version;
+	sdev_plugin_flags_t spo_flags;
+	sp_valid_f spo_validate;
+	sp_filldir_f spo_filldir;
+	sp_inactive_f spo_inactive;
+} sdev_plugin_ops_t;
+
+extern sdev_plugin_hdl_t sdev_plugin_register(const char *, sdev_plugin_ops_t *,
+    int *);
+extern int sdev_plugin_unregister(sdev_plugin_hdl_t);
+
+typedef enum sdev_ctx_flags {
+	SDEV_CTX_GLOBAL = 0x2	/* node belongs to the GZ */
+} sdev_ctx_flags_t;
+
+/*
+ * Context helper functions
+ */
+extern sdev_ctx_flags_t sdev_ctx_flags(sdev_ctx_t);
+extern const char *sdev_ctx_name(sdev_ctx_t);
+extern const char *sdev_ctx_path(sdev_ctx_t);
+extern enum vtype sdev_ctx_vtype(sdev_ctx_t);
+extern const void *sdev_ctx_vtype_data(sdev_ctx_t);
+
+/*
+ * Callbacks to manipulate nodes
+ */
+extern int sdev_plugin_mkdir(sdev_ctx_t, char *);
+extern int sdev_plugin_mknod(sdev_ctx_t, char *, mode_t, dev_t);
+
+#endif	/* _KERNEL */
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif /* _SYS_SDEV_PLUGIN_H */
diff --git a/usr/src/uts/common/sys/fs/tmp.h b/usr/src/uts/common/sys/fs/tmp.h
index 68dd67c61e..f8740e8873 100644
--- a/usr/src/uts/common/sys/fs/tmp.h
+++ b/usr/src/uts/common/sys/fs/tmp.h
@@ -22,12 +22,13 @@
  * Copyright 2007 Sun Microsystems, Inc.
  * All rights reserved.  Use is subject to license terms.
  */
+/*
+ * Copyright 2015 Joyent, Inc.
+ */
 
 #ifndef	_SYS_FS_TMP_H
 #define	_SYS_FS_TMP_H
 
-#pragma ident	"%Z%%M%	%I%	%E% SMI"
-
 #ifdef	__cplusplus
 extern "C" {
 #endif
@@ -68,29 +69,28 @@ enum dr_op	{ DR_REMOVE, DR_RMDIR, DR_RENAME };	/* dirremove ops */
 
 /*
  * tmpfs_minfree is the amount (in pages) of anonymous memory that tmpfs
- * leaves free for the rest of the system.  E.g. in a system with 32MB of
- * configured swap space, if 16MB were reserved (leaving 16MB free),
- * tmpfs could allocate up to 16MB - tmpfs_minfree.  The default value
- * for tmpfs_minfree is btopr(TMPMINFREE) but it can cautiously patched
- * to a different number of pages.
- * NB: If tmpfs allocates too much swap space, other processes will be
- * unable to execute.
+ * leaves free for the rest of the system.  In antiquity, this number could be
+ * relevant on a system-wide basis, as physical DRAM was routinely exhausted;
+ * however, in more modern times, the relative growth of DRAM with respect to
+ * application footprint means that this number is only likely to become
+ * factor in a virtualized OS environment (e.g., a zone) -- and even then only
+ * when DRAM and swap have both been capped low to allow for maximum tenancy.
+ * TMPMINFREE -- the value from which tmpfs_minfree is derived -- should
+ * therefore be configured to a value that is roughly the smallest practical
+ * value for memory + swap minus the largest reasonable size for tmpfs in such
+ * a configuration.  As of this writing, the smallest practical memory + swap
+ * configuration is 128MB, and it seems reasonable to allow tmpfs to consume
+ * no more than seven-eighths of this, yielding a TMPMINFREE of 16MB.  Care
+ * should be exercised in changing this:  tuning this value too high will
+ * result in spurious ENOSPC errors in tmpfs in small zones (a problem that
+ * can induce cascading failure surprisingly often); tuning this value too low
+ * will result in tmpfs consumption alone to alone induce application-level
+ * memory allocation failure.
  */
-#define	TMPMINFREE	2 * 1024 * 1024	/* 2 Megabytes */
+#define	TMPMINFREE	16 * 1024 * 1024	/* 16 Megabytes */
 
 extern size_t	tmpfs_minfree;		/* Anonymous memory in pages */
 
-/*
- * tmpfs can allocate only a certain percentage of kernel memory,
- * which is used for tmpnodes, directories, file names, etc.
- * This is statically set as TMPMAXFRACKMEM of physical memory.
- * The actual number of allocatable bytes can be patched in tmpfs_maxkmem.
- */
-#define	TMPMAXFRACKMEM	25	/* 1/25 of physical memory */
-
-extern size_t 	tmp_kmemspace;
-extern size_t	tmpfs_maxkmem;	/* Allocatable kernel memory in bytes */
-
 extern	void	tmpnode_init(struct tmount *, struct tmpnode *,
 	struct vattr *, struct cred *);
 extern	int	tmpnode_trunc(struct tmount *, struct tmpnode *, ulong_t);
@@ -101,13 +101,12 @@ extern	int	tdirdelete(struct tmpnode *, struct tmpnode *, char *,
 	enum dr_op, struct cred *);
 extern	void	tdirinit(struct tmpnode *, struct tmpnode *);
 extern	void	tdirtrunc(struct tmpnode *);
-extern	void	*tmp_memalloc(size_t, int);
-extern	void	tmp_memfree(void *, size_t);
 extern	int	tmp_resv(struct tmount *, struct tmpnode *, size_t, int);
 extern	int	tmp_taccess(void *, int, struct cred *);
 extern	int	tmp_sticky_remove_access(struct tmpnode *, struct tmpnode *,
 	struct cred *);
 extern	int	tmp_convnum(char *, pgcnt_t *);
+extern	int	tmp_convmode(char *, mode_t *);
 extern	int	tdirenter(struct tmount *, struct tmpnode *, char *,
 	enum de_op, struct tmpnode *, struct tmpnode *, struct vattr *,
 	struct tmpnode **, struct cred *, caller_context_t *);
diff --git a/usr/src/uts/common/sys/gsqueue.h b/usr/src/uts/common/sys/gsqueue.h
new file mode 100644
index 0000000000..40ef4ce982
--- /dev/null
+++ b/usr/src/uts/common/sys/gsqueue.h
@@ -0,0 +1,65 @@
+/*
+ * This file and its contents are supplied under the terms of the
+ * Common Development and Distribution License ("CDDL"), version 1.0.
+ * You may only use this file in accordance with the terms of version
+ * 1.0 of the CDDL.
+ *
+ * A full copy of the text of the CDDL should have accompanied this
+ * source.  A copy of the CDDL is also available via the Internet at
+ * http://www.illumos.org/license/CDDL.
+ */
+
+/*
+ * Copyright (c) 2014 Joyent, Inc.  All rights reserved.
+ */
+
+#ifndef _SYS_GSQUEUE_H
+#define	_SYS_GSQUEUE_H
+
+/*
+ * Standard interfaces to serializaion queues for everyone (except IP).
+ */
+
+#include <sys/types.h>
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+#ifdef _KERNEL
+
+typedef struct gsqueue gsqueue_t;
+typedef struct gsqueue_set gsqueue_set_t;
+
+typedef void (*gsqueue_cb_f)(gsqueue_set_t *, gsqueue_t *, void *, boolean_t);
+typedef void (*gsqueue_proc_f)(void *, mblk_t *, gsqueue_t *, void *);
+
+extern gsqueue_set_t *gsqueue_set_create(uint_t, pri_t);
+extern void gsqueue_set_destroy(gsqueue_set_t *);
+extern gsqueue_t *gsqueue_set_get(gsqueue_set_t *, uint_t);
+
+extern uintptr_t gsqueue_set_cb_add(gsqueue_set_t *, gsqueue_cb_f, void *);
+extern int gsqueue_set_cb_remove(gsqueue_set_t *, uintptr_t);
+
+#define	GSQUEUE_FILL	0x0001
+#define	GSQUEUE_NODRAIN	0x0002
+#define	GSQUEUE_PROCESS	0x0004
+
+extern void gsqueue_enter_one(gsqueue_t *, mblk_t *, gsqueue_proc_f, void *,
+    int, uint8_t);
+
+/*
+ * The default wait is inherited from IP. This determines the amount of time
+ * that must pass after queuing work, before we wake up the worker thread. This
+ * value is in milliseconds.
+ */
+#define	GSQUEUE_DEFAULT_WAIT	10
+#define	GSQUEUE_DEFAULT_PRIORITY	MAXCLSYSPRI
+
+#endif	/* _KERNEL */
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif /* _SYS_GSQUEUE_H */
diff --git a/usr/src/uts/common/sys/id_space.h b/usr/src/uts/common/sys/id_space.h
index d56fcceb5a..46d25f207f 100644
--- a/usr/src/uts/common/sys/id_space.h
+++ b/usr/src/uts/common/sys/id_space.h
@@ -20,6 +20,7 @@
  */
 /*
  * Copyright (c) 2000, 2010, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2014, Joyent, Inc.  All Rights reserved.
  */
 
 #ifndef	_ID_SPACE_H
@@ -34,8 +35,6 @@ extern "C" {
 #include <sys/mutex.h>
 #include <sys/vmem.h>
 
-#ifdef _KERNEL
-
 typedef vmem_t id_space_t;
 
 id_space_t *id_space_create(const char *, id_t, id_t);
@@ -48,8 +47,6 @@ id_t id_allocff_nosleep(id_space_t *);
 id_t id_alloc_specific_nosleep(id_space_t *, id_t);
 void id_free(id_space_t *, id_t);
 
-#endif /* _KERNEL */
-
 #ifdef	__cplusplus
 }
 #endif
diff --git a/usr/src/uts/common/sys/inotify.h b/usr/src/uts/common/sys/inotify.h
new file mode 100644
index 0000000000..8acc1a7280
--- /dev/null
+++ b/usr/src/uts/common/sys/inotify.h
@@ -0,0 +1,153 @@
+/*
+ * This file and its contents are supplied under the terms of the
+ * Common Development and Distribution License ("CDDL"), version 1.0.
+ * You may only use this file in accordance with the terms of version
+ * 1.0 of the CDDL.
+ *
+ * A full copy of the text of the CDDL should have accompanied this
+ * source.  A copy of the CDDL is also available via the Internet at
+ * http://www.illumos.org/license/CDDL.
+ */
+
+/*
+ * Copyright (c) 2014 Joyent, Inc.  All rights reserved.
+ */
+
+/*
+ * Header file to support for the inotify facility.  Note that this facility
+ * is designed to be binary compatible with the Linux inotify facility; values
+ * for constants here should therefore exactly match those found in Linux, and
+ * this facility shouldn't be extended independently of Linux.
+ */
+
+#ifndef _SYS_INOTIFY_H
+#define	_SYS_INOTIFY_H
+
+#include <sys/types.h>
+
+#ifdef	__cplusplus
+extern "C" {
+#endif
+
+/*
+ * Events that can be explicitly requested on any inotify watch.
+ */
+#define	IN_ACCESS		0x00000001
+#define	IN_MODIFY		0x00000002
+#define	IN_ATTRIB		0x00000004
+#define	IN_CLOSE_WRITE		0x00000008
+#define	IN_CLOSE_NOWRITE	0x00000010
+#define	IN_OPEN			0x00000020
+#define	IN_MOVED_FROM		0x00000040
+#define	IN_MOVED_TO		0x00000080
+#define	IN_CREATE		0x00000100
+#define	IN_DELETE		0x00000200
+#define	IN_DELETE_SELF		0x00000400
+#define	IN_MOVE_SELF		0x00000800
+
+/*
+ * Events that can be sent to an inotify watch -- requested or not.
+ */
+#define	IN_UNMOUNT		0x00002000
+#define	IN_Q_OVERFLOW		0x00004000
+#define	IN_IGNORED		0x00008000
+
+/*
+ * Flags that can modify an inotify event.
+ */
+#define	IN_ONLYDIR		0x01000000
+#define	IN_DONT_FOLLOW		0x02000000
+#define	IN_EXCL_UNLINK		0x04000000
+#define	IN_MASK_ADD		0x20000000
+#define	IN_ISDIR		0x40000000
+#define	IN_ONESHOT		0x80000000
+
+/*
+ * Helpful constants.
+ */
+#define	IN_CLOSE		(IN_CLOSE_WRITE | IN_CLOSE_NOWRITE)
+#define	IN_MOVE			(IN_MOVED_FROM | IN_MOVED_TO)
+#define	IN_ALL_EVENTS		\
+	(IN_ACCESS | IN_MODIFY | IN_ATTRIB | IN_CLOSE_WRITE | \
+	IN_CLOSE_NOWRITE | IN_OPEN | IN_MOVED_FROM | IN_MOVED_TO | \
+	IN_DELETE | IN_CREATE | IN_DELETE_SELF | IN_MOVE_SELF)
+
+#define	IN_CHILD_EVENTS		\
+	(IN_ACCESS | IN_MODIFY | IN_ATTRIB | IN_CLOSE_WRITE | \
+	IN_CLOSE_NOWRITE | IN_MODIFY | IN_OPEN)
+
+/*
+ * To assure binary compatibility with Linux, these values are fixed at their
+ * Linux equivalents, not their native ones.
+ */
+#define	IN_CLOEXEC		02000000		/* LX_O_CLOEXEC */
+#define	IN_NONBLOCK		04000			/* LX_O_NONBLOCK */
+
+struct inotify_event {
+	int32_t		wd;		/* watch descriptor */
+	uint32_t	mask;		/* mask of events */
+	uint32_t	cookie;		/* event association cookie, if any */
+	uint32_t	len;		/* size of name field */
+	char		name[];		/* optional NUL-terminated name */
+};
+
+/*
+ * These ioctl values are specific to the native implementation; applications
+ * shouldn't be using them directly, and they should therefore be safe to
+ * change without breaking apps.
+ */
+#define	INOTIFYIOC		(('i' << 24) | ('n' << 16) | ('y' << 8))
+#define	INOTIFYIOC_ADD_WATCH	(INOTIFYIOC | 1)	/* add watch */
+#define	INOTIFYIOC_RM_WATCH	(INOTIFYIOC | 2)	/* remove watch */
+#define	INOTIFYIOC_ADD_CHILD	(INOTIFYIOC | 3)	/* add child watch */
+#define	INOTIFYIOC_ACTIVATE	(INOTIFYIOC | 4)	/* activate watch */
+
+#ifndef _LP64
+#ifndef _LITTLE_ENDIAN
+#define	INOTIFY_PTR(type, name)	uint32_t name##pad; type *name
+#else
+#define	INOTIFY_PTR(type, name)	type *name; uint32_t name##pad
+#endif
+#else
+#define	INOTIFY_PTR(type, name)	type *name
+#endif
+
+typedef struct inotify_addwatch {
+	int inaw_fd;			/* open fd for object */
+	uint32_t inaw_mask;		/* desired mask */
+} inotify_addwatch_t;
+
+typedef struct inotify_addchild {
+	INOTIFY_PTR(char, inac_name);	/* pointer to name */
+	int inac_fd;			/* open fd for parent */
+} inotify_addchild_t;
+
+#ifndef _KERNEL
+
+extern int inotify_init(void);
+extern int inotify_init1(int);
+extern int inotify_add_watch(int, const char *, uint32_t);
+extern int inotify_rm_watch(int, int);
+
+#else
+
+#define	IN_UNMASKABLE \
+	(IN_UNMOUNT | IN_Q_OVERFLOW | IN_IGNORED | IN_ISDIR)
+
+#define	IN_MODIFIERS \
+	(IN_EXCL_UNLINK | IN_ONESHOT)
+
+#define	IN_FLAGS \
+	(IN_ONLYDIR | IN_DONT_FOLLOW | IN_MASK_ADD)
+
+#define	IN_REMOVAL		(1ULL << 32)
+#define	INOTIFYMNRN_INOTIFY	0
+#define	INOTIFYMNRN_CLONE	1
+
+#endif /* _KERNEL */
+
+#ifdef	__cplusplus
+}
+#endif
+
+#endif	/* _SYS_INOTIFY_H */
diff --git a/usr/src/uts/common/sys/iso/signal_iso.h b/usr/src/uts/common/sys/iso/signal_iso.h
index b1990121b8..0ae64b45d7 100644
--- a/usr/src/uts/common/sys/iso/signal_iso.h
+++ b/usr/src/uts/common/sys/iso/signal_iso.h
@@ -22,6 +22,7 @@
 /*
  * Copyright 2010 Sun Microsystems, Inc.  All rights reserved.
  * Use is subject to license terms.
+ * Copyright 2015, Joyent, Inc.
  */
 
 /*	Copyright (c) 1984, 1986, 1987, 1988, 1989 AT&T	*/
@@ -95,7 +96,7 @@ extern "C" {
 
 /* insert new signals here, and move _SIGRTM* appropriately */
 #define	_SIGRTMIN 42	/* first (highest-priority) realtime signal */
-#define	_SIGRTMAX 73	/* last (lowest-priority) realtime signal */
+#define	_SIGRTMAX 74	/* last (lowest-priority) realtime signal */
 extern long _sysconf(int);	/* System Private interface to sysconf() */
 #define	SIGRTMIN ((int)_sysconf(_SC_SIGRT_MIN))	/* first realtime signal */
 #define	SIGRTMAX ((int)_sysconf(_SC_SIGRT_MAX))	/* last realtime signal */
diff --git a/usr/src/uts/common/sys/klwp.h b/usr/src/uts/common/sys/klwp.h
index 41b70f6a6e..2aceb3a0f6 100644
--- a/usr/src/uts/common/sys/klwp.h
+++ b/usr/src/uts/common/sys/klwp.h
@@ -24,7 +24,7 @@
  */
 
 /*
- * Copyright (c) 2013, Joyent, Inc. All rights reserved.
+ * Copyright 2015 Joyent, Inc.
  */
 
 #ifndef	_SYS_KLWP_H
@@ -192,6 +192,8 @@ typedef struct _klwp {
 	struct contract	*lwp_ct_latest[CTT_MAXTYPE]; /* last created contract */
 
 	void	*lwp_brand;		/* per-lwp brand data */
+	int	(*lwp_brand_syscall)(void); /* brand syscall interposer */
+
 	struct psinfo *lwp_spymaster;	/* if an agent LWP, our spymaster */
 } klwp_t;
 
diff --git a/usr/src/uts/common/sys/ksocket.h b/usr/src/uts/common/sys/ksocket.h
index dfe25eec76..be669cb78d 100644
--- a/usr/src/uts/common/sys/ksocket.h
+++ b/usr/src/uts/common/sys/ksocket.h
@@ -21,6 +21,7 @@
 /*
  * Copyright 2011 Nexenta Systems, Inc.  All rights reserved.
  * Copyright (c) 2008, 2010, Oracle and/or its affiliates. All rights reserved.
+ * Copyright 2015, Joyent, Inc.
  */
 
 #ifndef _SYS_KSOCKET_H_
@@ -121,6 +122,10 @@ extern int 	ksocket_close(ksocket_t, struct cred *);
 extern void	ksocket_hold(ksocket_t);
 extern void	ksocket_rele(ksocket_t);
 
+typedef boolean_t (*ksocket_krecv_f)(ksocket_t, mblk_t *, size_t, int, void *);
+extern int	ksocket_krecv_set(ksocket_t, ksocket_krecv_f, void *);
+extern void	ksocket_krecv_unblock(ksocket_t);
+
 #ifdef	__cplusplus
 }
 #endif
diff --git a/usr/src/uts/common/sys/limits.h b/usr/src/uts/common/sys/limits.h
new file mode 100644
index 0000000000..88625d1829
--- /dev/null
+++ b/usr/src/uts/common/sys/limits.h
@@ -0,0 +1,32 @@
+/*
+ * This file and its contents are supplied under the terms of the
+ * Common Development and Distribution License ("CDDL"), version 1.0.
+ * You may only use this file in accordance with the terms of version
+ * 1.0 of the CDDL.
+ *
+ * A full copy of the text of the CDDL should have accompanied this
+ * source.  A copy of the CDDL is also available via the Internet at
+ * http://www.illumos.org/license/CDDL.
+ */
+/*
+ * Copyright 2015 Joyent, Inc.  All rights reserved.
+ */
+
+#ifndef _SYS_LIMITS_H
+#define	_SYS_LIMITS_H
+
+#ifdef	__cplusplus
+extern "C" {
+#endif
+
+#define	IOV_MAX	1024
+
+#ifdef _KERNEL
+#define	IOV_MAX_STACK	16	/* max. IOV on-stack allocation */
+#endif /* _KERNEL */
+
+#ifdef	__cplusplus
+}
+#endif
+
+#endif	/* _SYS_LIMITS_H */
diff --git a/usr/src/uts/common/sys/mac.h b/usr/src/uts/common/sys/mac.h
index bd09077f1b..3c4e064f06 100644
--- a/usr/src/uts/common/sys/mac.h
+++ b/usr/src/uts/common/sys/mac.h
@@ -21,7 +21,7 @@
 
 /*
  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
- * Copyright (c) 2014, Joyent, Inc.  All rights reserved.
+ * Copyright 2015, Joyent, Inc.
  */
 
 #ifndef	_SYS_MAC_H
@@ -100,6 +100,14 @@ typedef struct mac_propval_uint32_range_s {
 } mac_propval_uint32_range_t;
 
 /*
+ * Defines ranges which are a series of C style strings.
+ */
+typedef struct mac_propval_str_range_s {
+	uint32_t mpur_nextbyte;
+	char mpur_data[1];
+} mac_propval_str_range_t;
+
+/*
  * Data type of property values.
  */
 typedef enum {
@@ -119,6 +127,7 @@ typedef struct mac_propval_range_s {
 	mac_propval_type_t mpr_type;		/* type of value */
 	union {
 		mac_propval_uint32_range_t mpr_uint32[1];
+		mac_propval_str_range_t mpr_str;
 	} u;
 } mac_propval_range_t;
 
@@ -213,6 +222,7 @@ typedef enum {
 	MAC_PROP_MAX_RXHWCLNT_AVAIL,
 	MAC_PROP_MAX_TXHWCLNT_AVAIL,
 	MAC_PROP_IB_LINKMODE,
+	MAC_PROP_VN_PROMISC_FILTERED,
 	MAC_PROP_SECONDARY_ADDRS,
 	MAC_PROP_PRIVATE = -1
 } mac_prop_id_t;
diff --git a/usr/src/uts/common/sys/mac_client_impl.h b/usr/src/uts/common/sys/mac_client_impl.h
index 3c89b7c434..f599e1fcfe 100644
--- a/usr/src/uts/common/sys/mac_client_impl.h
+++ b/usr/src/uts/common/sys/mac_client_impl.h
@@ -328,6 +328,7 @@ extern	int	mac_tx_percpu_cnt;
 
 /* Mac protection flags */
 #define	MPT_FLAG_V6_LOCAL_ADDR_SET	0x0001
+#define	MPT_FLAG_PROMISC_FILTERED	0x0002
 
 /* in mac_client.c */
 extern void mac_promisc_client_dispatch(mac_client_impl_t *, mblk_t *);
diff --git a/usr/src/uts/common/sys/mac_client_priv.h b/usr/src/uts/common/sys/mac_client_priv.h
index bed2e293b3..fa3a1ebfad 100644
--- a/usr/src/uts/common/sys/mac_client_priv.h
+++ b/usr/src/uts/common/sys/mac_client_priv.h
@@ -22,7 +22,7 @@
 /*
  * Copyright 2010 Sun Microsystems, Inc.  All rights reserved.
  * Use is subject to license terms.
- * Copyright 2013 Joyent, Inc.  All rights reserved.
+ * Copyright 2015 Joyent, Inc.
  */
 
 /*
@@ -170,6 +170,7 @@ extern void mac_client_set_intr_cpu(void *, mac_client_handle_t, int32_t);
 extern void *mac_get_devinfo(mac_handle_t);
 
 extern boolean_t mac_is_vnic(mac_handle_t);
+extern boolean_t mac_is_overlay(mac_handle_t);
 extern uint32_t mac_no_notification(mac_handle_t);
 
 extern int mac_set_prop(mac_handle_t, mac_prop_id_t, char *, void *, uint_t);
diff --git a/usr/src/uts/common/sys/mac_impl.h b/usr/src/uts/common/sys/mac_impl.h
index 3c9c6b77a4..d72558c612 100644
--- a/usr/src/uts/common/sys/mac_impl.h
+++ b/usr/src/uts/common/sys/mac_impl.h
@@ -20,7 +20,7 @@
  */
 /*
  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
- * Copyright (c) 2014, Joyent, Inc.  All rights reserved.
+ * Copyright 2015, Joyent, Inc.
  */
 
 #ifndef	_SYS_MAC_IMPL_H
@@ -643,6 +643,7 @@ struct mac_impl_s {
 #define	MIS_LEGACY		0x0040
 #define	MIS_NO_ACTIVE		0x0080
 #define	MIS_POLL_DISABLE	0x0100
+#define	MIS_IS_OVERLAY		0x0200
 
 #define	mi_getstat	mi_callbacks->mc_getstat
 #define	mi_start	mi_callbacks->mc_start
@@ -894,6 +895,8 @@ extern void mac_protect_fini(mac_client_impl_t *);
 extern int mac_set_resources(mac_handle_t, mac_resource_props_t *);
 extern void mac_get_resources(mac_handle_t, mac_resource_props_t *);
 extern void mac_get_effective_resources(mac_handle_t, mac_resource_props_t *);
+extern void mac_set_promisc_filtered(mac_client_handle_t, boolean_t);
+extern boolean_t mac_get_promisc_filtered(mac_client_handle_t);
 
 extern cpupart_t *mac_pset_find(mac_resource_props_t *, boolean_t *);
 extern void mac_set_pool_effective(boolean_t, cpupart_t *,
diff --git a/usr/src/uts/common/sys/mac_provider.h b/usr/src/uts/common/sys/mac_provider.h
index 9f7f2a1a73..5f02451542 100644
--- a/usr/src/uts/common/sys/mac_provider.h
+++ b/usr/src/uts/common/sys/mac_provider.h
@@ -21,6 +21,7 @@
 
 /*
  * Copyright (c) 2008, 2010, Oracle and/or its affiliates. All rights reserved.
+ * Copyright 2015, Joyent, Inc.
  */
 
 #ifndef	_SYS_MAC_PROVIDER_H
@@ -106,7 +107,8 @@ typedef enum {
 	MAC_CAPAB_NO_NATIVEVLAN	= 0x00080000, /* boolean only, no data */
 	MAC_CAPAB_NO_ZCOPY	= 0x00100000, /* boolean only, no data */
 	MAC_CAPAB_LEGACY	= 0x00200000, /* data is mac_capab_legacy_t */
-	MAC_CAPAB_VRRP		= 0x00400000  /* data is mac_capab_vrrp_t */
+	MAC_CAPAB_VRRP		= 0x00400000, /* data is mac_capab_vrrp_t */
+	MAC_CAPAB_OVERLAY	= 0x00800000  /* boolean only, no data */
 } mac_capab_t;
 
 /*
diff --git a/usr/src/uts/common/sys/mman.h b/usr/src/uts/common/sys/mman.h
index 4976bf09ab..34e491fd3b 100644
--- a/usr/src/uts/common/sys/mman.h
+++ b/usr/src/uts/common/sys/mman.h
@@ -318,6 +318,7 @@ struct memcntl_mha32 {
 #define	MADV_ACCESS_DEFAULT	6	/* default access */
 #define	MADV_ACCESS_LWP		7	/* next LWP to access heavily */
 #define	MADV_ACCESS_MANY	8	/* many processes to access heavily */
+#define	MADV_PURGE		9	/* contents will be purged */
 
 #endif	/* (_POSIX_C_SOURCE <= 2) && !defined(_XPG4_2) ...  */
 
@@ -337,6 +338,7 @@ struct memcntl_mha32 {
 #define	MS_SYNC		0x4		/* wait for msync */
 #define	MS_ASYNC	0x1		/* return immediately */
 #define	MS_INVALIDATE	0x2		/* invalidate caches */
+#define	MS_INVALCURPROC	0x8		/* invalidate cache for curproc only */
 
 #if	(_POSIX_C_SOURCE <= 2) && !defined(_XPG4_2) || defined(__EXTENSIONS__)
 /* functions to mctl */
diff --git a/usr/src/uts/common/sys/mntent.h b/usr/src/uts/common/sys/mntent.h
index e95ef3fccc..d215d88790 100644
--- a/usr/src/uts/common/sys/mntent.h
+++ b/usr/src/uts/common/sys/mntent.h
@@ -21,6 +21,7 @@
 /*
  * Copyright 2008 Sun Microsystems, Inc.  All rights reserved.
  * Use is subject to license terms.
+ * Copyright 2012, Joyent, Inc.  All rights reserved.
  *
  *	Copyright (c) 1984, 1986, 1987, 1988, 1989 AT&T
  *		All Rights Reserved
@@ -47,6 +48,7 @@ extern "C" {
 #define	MNTTYPE_PCFS	"pcfs"		/* PC (MSDOS) file system */
 #define	MNTTYPE_PC	MNTTYPE_PCFS	/* Deprecated name; use MNTTYPE_PCFS */
 #define	MNTTYPE_LOFS	"lofs"		/* Loop back file system */
+#define	MNTTYPE_HYPRLOFS "hyprlofs"	/* Hyperlofs file system */
 #define	MNTTYPE_LO	MNTTYPE_LOFS	/* Deprecated name; use MNTTYPE_LOFS */
 #define	MNTTYPE_HSFS	"hsfs"		/* High Sierra (9660) file system */
 #define	MNTTYPE_SWAP	"swap"		/* Swap file system */
diff --git a/usr/src/uts/common/sys/netconfig.h b/usr/src/uts/common/sys/netconfig.h
index 14b1aa55db..883c329aed 100644
--- a/usr/src/uts/common/sys/netconfig.h
+++ b/usr/src/uts/common/sys/netconfig.h
@@ -28,6 +28,7 @@
  *
  * Copyright 2004 Sun Microsystems, Inc.  All rights reserved.
  * Use is subject to license terms.
+ * Copyright 2015 Joyent, Inc.
  */
 
 #ifndef	_SYS_NETCONFIG_H
@@ -147,6 +148,8 @@ extern int		endnetpath(void *);
 extern struct netconfig *getnetpath(void *);
 extern void		nc_perror(const char *);
 extern char		*nc_sperror(void);
+extern void		_nsl_brand_set_hooks(int (*)(void),
+			    struct netconfig *(*)(int));
 
 #ifdef	__cplusplus
 }
diff --git a/usr/src/uts/common/sys/neti.h b/usr/src/uts/common/sys/neti.h
index 93b5fc3e01..ea85c78f6b 100644
--- a/usr/src/uts/common/sys/neti.h
+++ b/usr/src/uts/common/sys/neti.h
@@ -44,6 +44,8 @@ extern "C" {
 #define	NHF_INET	"NHF_INET"
 #define	NHF_INET6	"NHF_INET6"
 #define	NHF_ARP		"NHF_ARP"
+#define	NHF_VND_INET	"NHF_VND_INET"
+#define	NHF_VND_INET6	"NHF_VND_INET6"
 
 /*
  * Event identification
diff --git a/usr/src/uts/common/sys/netstack.h b/usr/src/uts/common/sys/netstack.h
index 2c77e1be96..73f29d1e63 100644
--- a/usr/src/uts/common/sys/netstack.h
+++ b/usr/src/uts/common/sys/netstack.h
@@ -81,7 +81,8 @@ typedef id_t	netstackid_t;
 #define	NS_IPSECESP	16
 #define	NS_IPNET	17
 #define	NS_ILB		18
-#define	NS_MAX		(NS_ILB+1)
+#define	NS_VND		19
+#define	NS_MAX		(NS_VND+1)
 
 /*
  * State maintained for each module which tracks the state of
diff --git a/usr/src/uts/common/sys/overlay.h b/usr/src/uts/common/sys/overlay.h
new file mode 100644
index 0000000000..12d0dbca51
--- /dev/null
+++ b/usr/src/uts/common/sys/overlay.h
@@ -0,0 +1,96 @@
+/*
+ * This file and its contents are supplied under the terms of the
+ * Common Development and Distribution License ("CDDL"), version 1.0.
+ * You may only use this file in accordance with the terms of version
+ * 1.0 of the CDDL.
+ *
+ * A full copy of the text of the CDDL should have accompanied this
+ * source.  A copy of the CDDL is also available via the Internet at
+ * http://www.illumos.org/license/CDDL.
+ */
+
+/*
+ * Copyright 2015, Joyent, Inc.
+ */
+
+#ifndef _SYS_OVERLAY_H
+#define	_SYS_OVERLAY_H
+
+/*
+ * Overlay device support
+ */
+
+#include <sys/param.h>
+#include <sys/dld_ioc.h>
+#include <sys/mac.h>
+#include <sys/overlay_common.h>
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+#define	OVERLAY_IOC_CREATE	OVERLAYIOC(1)
+#define	OVERLAY_IOC_DELETE	OVERLAYIOC(2)
+#define	OVERLAY_IOC_PROPINFO	OVERLAYIOC(3)
+#define	OVERLAY_IOC_GETPROP	OVERLAYIOC(4)
+#define	OVERLAY_IOC_SETPROP	OVERLAYIOC(5)
+#define	OVERLAY_IOC_NPROPS	OVERLAYIOC(6)
+#define	OVERLAY_IOC_ACTIVATE	OVERLAYIOC(7)
+#define	OVERLAY_IOC_STATUS	OVERLAYIOC(8)
+
+typedef struct overlay_ioc_create {
+	datalink_id_t	oic_linkid;
+	uint32_t	oic_filler;
+	uint64_t	oic_vnetid;
+	char		oic_encap[MAXLINKNAMELEN];
+} overlay_ioc_create_t;
+
+typedef struct overlay_ioc_activate {
+	datalink_id_t	oia_linkid;
+} overlay_ioc_activate_t;
+
+typedef struct overlay_ioc_delete {
+	datalink_id_t	oid_linkid;
+} overlay_ioc_delete_t;
+
+typedef struct overlay_ioc_nprops {
+	datalink_id_t	oipn_linkid;
+	int32_t		oipn_nprops;
+} overlay_ioc_nprops_t;
+
+typedef struct overlay_ioc_propinfo {
+	datalink_id_t	oipi_linkid;
+	int32_t		oipi_id;
+	char		oipi_name[OVERLAY_PROP_NAMELEN];
+	uint_t		oipi_type;
+	uint_t		oipi_prot;
+	uint8_t		oipi_default[OVERLAY_PROP_SIZEMAX];
+	uint32_t	oipi_defsize;
+	uint32_t	oipi_posssize;
+	uint8_t		oipi_poss[OVERLAY_PROP_SIZEMAX];
+} overlay_ioc_propinfo_t;
+
+typedef struct overlay_ioc_prop {
+	datalink_id_t	oip_linkid;
+	int32_t		oip_id;
+	char		oip_name[OVERLAY_PROP_NAMELEN];
+	uint8_t		oip_value[OVERLAY_PROP_SIZEMAX];
+	uint32_t	oip_size;
+} overlay_ioc_prop_t;
+
+typedef enum overlay_status {
+	OVERLAY_I_OK		= 0x00,
+	OVERLAY_I_DEGRADED	= 0x01
+} overlay_status_t;
+
+typedef struct overlay_ioc_status {
+	datalink_id_t	ois_linkid;
+	uint_t		ois_status;
+	char		ois_message[OVERLAY_STATUS_BUFLEN];
+} overlay_ioc_status_t;
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif /* _SYS_OVERLAY_H */
diff --git a/usr/src/uts/common/sys/overlay_common.h b/usr/src/uts/common/sys/overlay_common.h
new file mode 100644
index 0000000000..d638096006
--- /dev/null
+++ b/usr/src/uts/common/sys/overlay_common.h
@@ -0,0 +1,65 @@
+/*
+ * This file and its contents are supplied under the terms of the
+ * Common Development and Distribution License ("CDDL"), version 1.0.
+ * You may only use this file in accordance with the terms of version
+ * 1.0 of the CDDL.
+ *
+ * A full copy of the text of the CDDL should have accompanied this
+ * source.  A copy of the CDDL is also available via the Internet at
+ * http://www.illumos.org/license/CDDL.
+ */
+
+/*
+ * Copyright 2015 Joyent, Inc.
+ */
+
+#ifndef _SYS_OVERLAY_COMMON_H
+#define	_SYS_OVERLAY_COMMON_H
+
+/*
+ * Common overlay definitions
+ */
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+typedef enum overlay_target_mode {
+	OVERLAY_TARGET_NONE = 0x0,
+	OVERLAY_TARGET_POINT,
+	OVERLAY_TARGET_DYNAMIC
+} overlay_target_mode_t;
+
+typedef enum overlay_plugin_dest {
+	OVERLAY_PLUGIN_D_INVALID	= 0x0,
+	OVERLAY_PLUGIN_D_ETHERNET	= 0x1,
+	OVERLAY_PLUGIN_D_IP		= 0x2,
+	OVERLAY_PLUGIN_D_PORT 		= 0x4,
+	OVERLAY_PLUGIN_D_MASK		= 0x7
+} overlay_plugin_dest_t;
+
+typedef enum overlay_prop_type {
+	OVERLAY_PROP_T_INT = 0x1,	/* signed int */
+	OVERLAY_PROP_T_UINT,		/* unsigned int */
+	OVERLAY_PROP_T_IP,		/* sinaddr6 */
+	OVERLAY_PROP_T_STRING		/* OVERLAY_PROPS_SIZEMAX */
+} overlay_prop_type_t;
+
+typedef enum overlay_prop_prot {
+	OVERLAY_PROP_PERM_REQ	= 0x1,
+	OVERLAY_PROP_PERM_READ	= 0x2,
+	OVERLAY_PROP_PERM_WRITE	= 0x4,
+	OVERLAY_PROP_PERM_RW 	= 0x6,
+	OVERLAY_PROP_PERM_RRW	= 0x7,
+	OVERLAY_PROP_PERM_MASK	= 0x7
+} overlay_prop_prot_t;
+
+#define	OVERLAY_PROP_NAMELEN	64
+#define	OVERLAY_PROP_SIZEMAX	256
+#define	OVERLAY_STATUS_BUFLEN	256
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif /* _SYS_OVERLAY_COMMON_H */
diff --git a/usr/src/uts/common/sys/overlay_impl.h b/usr/src/uts/common/sys/overlay_impl.h
new file mode 100644
index 0000000000..f62ecff179
--- /dev/null
+++ b/usr/src/uts/common/sys/overlay_impl.h
@@ -0,0 +1,205 @@
+/*
+ * This file and its contents are supplied under the terms of the
+ * Common Development and Distribution License ("CDDL"), version 1.0.
+ * You may only use this file in accordance with the terms of version
+ * 1.0 of the CDDL.
+ *
+ * A full copy of the text of the CDDL should have accompanied this
+ * source.  A copy of the CDDL is also available via the Internet at
+ * http://www.illumos.org/license/CDDL.
+ */
+
+/*
+ * Copyright 2015, Joyent, Inc.
+ */
+
+#ifndef _SYS_OVERLAY_IMPL_H
+#define	_SYS_OVERLAY_IMPL_H
+
+/*
+ * Overlay device support
+ */
+
+#include <sys/overlay.h>
+#include <sys/overlay_common.h>
+#include <sys/overlay_plugin.h>
+#include <sys/overlay_target.h>
+#include <sys/ksynch.h>
+#include <sys/list.h>
+#include <sys/avl.h>
+#include <sys/ksocket.h>
+#include <sys/socket.h>
+#include <sys/refhash.h>
+#include <sys/ethernet.h>
+#include <sys/list.h>
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+#define	OVEP_VERSION_ONE	0x1
+
+typedef struct overlay_plugin {
+	kmutex_t ovp_mutex;
+	list_node_t ovp_link;			/* overlay_plugin_lock */
+	uint_t ovp_active;			/* ovp_mutex */
+	const char *ovp_name;			/* RO */
+	const overlay_plugin_ops_t *ovp_ops;	/* RO */
+	const char *const *ovp_props;		/* RO */
+	uint_t ovp_nprops;			/* RO */
+	uint_t ovp_id_size;			/* RO */
+	overlay_plugin_flags_t ovp_flags;	/* RO */
+	overlay_plugin_dest_t ovp_dest;		/* RO */
+} overlay_plugin_t;
+
+typedef struct overlay_mux {
+	list_node_t		omux_lnode;
+	ksocket_t		omux_ksock;	/* RO */
+	overlay_plugin_t	*omux_plugin;	/* RO: associated encap */
+	int			omux_domain;	/* RO: socket domain */
+	int			omux_family;	/* RO: socket family */
+	int			omux_protocol;	/* RO: socket protocol */
+	struct sockaddr 	*omux_addr;	/* RO: socket address */
+	socklen_t		omux_alen;	/* RO: sockaddr len */
+	kmutex_t		omux_lock;	/* Protects everything below */
+	uint_t			omux_count;	/* Active instances */
+	avl_tree_t		omux_devices;	/* Tree of devices */
+} overlay_mux_t;
+
+typedef enum overlay_target_flag {
+	OVERLAY_T_TEARDOWN	= 0x1
+} overlay_target_flag_t;
+
+typedef struct overlay_target {
+	kmutex_t		ott_lock;
+	kcondvar_t		ott_cond;
+	overlay_target_mode_t	ott_mode;	/* RO */
+	overlay_plugin_dest_t	ott_dest;	/* RO */
+	uint64_t		ott_id;		/* RO */
+	overlay_target_flag_t	ott_flags;	/* ott_lock */
+	uint_t			ott_ocount;	/* ott_lock */
+	union {					/* ott_lock */
+		overlay_target_point_t	ott_point;
+		struct overlay_target_dyn {
+			refhash_t	*ott_dhash;
+			avl_tree_t	ott_tree;
+		} ott_dyn;
+	} ott_u;
+} overlay_target_t;
+
+typedef enum overlay_dev_flag {
+	OVERLAY_F_ACTIVATED	= 0x01, /* Activate ioctl completed */
+	OVERLAY_F_IN_MUX	= 0x02,	/* Currently in a mux */
+	OVERLAY_F_IN_TX		= 0x04,	/* Currently doing tx */
+	OVERLAY_F_IN_RX		= 0x08, /* Currently doing rx */
+	OVERLAY_F_IOMASK	= 0x0c,	/* A mask for rx and tx */
+	OVERLAY_F_MDDROP	= 0x10,	/* Drop traffic for metadata update */
+	OVERLAY_F_STOPMASK	= 0x1e,	/* None set when stopping */
+	OVERLAY_F_VARPD		= 0x20,	/* varpd plugin exists */
+	OVERLAY_F_DEGRADED	= 0x40,	/* device is degraded */
+	OVERLAY_F_MASK		= 0x7f	/* mask of everything */
+} overlay_dev_flag_t;
+
+typedef struct overlay_dev {
+	kmutex_t	odd_lock;
+	kcondvar_t	odd_iowait;
+	list_node_t	odd_link;		/* overlay_dev_lock */
+	mac_handle_t	odd_mh;			/* RO */
+	overlay_plugin_t *odd_plugin;		/* RO */
+	datalink_id_t	odd_linkid;		/* RO */
+	void		*odd_pvoid;		/* RO -- only used by plugin */
+	uint_t		odd_ref;		/* protected by odd_lock */
+	uint_t		odd_mtu;		/* protected by odd_lock */
+	overlay_dev_flag_t odd_flags;		/* protected by odd_lock */
+	uint_t		odd_rxcount;		/* protected by odd_lock */
+	uint_t		odd_txcount;		/* protected by odd_lock */
+	overlay_mux_t	*odd_mux;		/* protected by odd_lock */
+	uint64_t	odd_vid;		/* RO if active else odd_lock */
+	avl_node_t	odd_muxnode;		/* managed by mux */
+	overlay_target_t *odd_target;		/* See big theory statement */
+	char		odd_fmamsg[OVERLAY_STATUS_BUFLEN];	/* odd_lock */
+} overlay_dev_t;
+
+typedef enum overlay_target_entry_flags {
+	OVERLAY_ENTRY_F_PENDING		= 0x01,	/* lookup in progress */
+	OVERLAY_ENTRY_F_VALID		= 0x02,	/* entry is currently valid */
+	OVERLAY_ENTRY_F_DROP		= 0x04,	/* always drop target */
+	OVERLAY_ENTRY_F_VALID_MASK	= 0x06
+} overlay_target_entry_flags_t;
+
+typedef struct overlay_target_entry {
+	kmutex_t		ote_lock;
+	refhash_link_t		ote_reflink;	/* hashtable link */
+	avl_node_t		ote_avllink;	/* iteration link */
+	list_node_t		ote_qlink;
+	overlay_target_entry_flags_t ote_flags;	/* RW: state flags */
+	uint8_t			ote_addr[ETHERADDRL];	/* RO: mac addr */
+	overlay_target_t	*ote_ott;	/* RO */
+	overlay_dev_t		*ote_odd;	/* RO */
+	overlay_target_point_t	ote_dest;	/* RW: destination */
+	mblk_t			*ote_chead;	/* RW: blocked mb chain head */
+	mblk_t			*ote_ctail;	/* RW: blocked mb chain tail */
+	size_t			ote_mbsize;	/* RW: outstanding mblk size */
+	hrtime_t		ote_vtime;	/* RW: valid timestamp */
+} overlay_target_entry_t;
+
+
+#define	OVERLAY_CTL	"overlay"
+
+extern dev_info_t *overlay_dip;
+
+extern mblk_t *overlay_m_tx(void *, mblk_t *);
+
+typedef int (*overlay_dev_iter_f)(overlay_dev_t *, void *);
+extern void overlay_dev_iter(overlay_dev_iter_f, void *);
+
+extern void overlay_plugin_init(void);
+extern overlay_plugin_t *overlay_plugin_lookup(const char *);
+extern void overlay_plugin_rele(overlay_plugin_t *);
+extern void overlay_plugin_fini(void);
+typedef int (*overlay_plugin_walk_f)(overlay_plugin_t *, void *);
+extern void overlay_plugin_walk(overlay_plugin_walk_f, void *);
+
+extern void overlay_io_start(overlay_dev_t *, overlay_dev_flag_t);
+extern void overlay_io_done(overlay_dev_t *, overlay_dev_flag_t);
+
+extern void overlay_mux_init(void);
+extern void overlay_mux_fini(void);
+
+extern overlay_mux_t *overlay_mux_open(overlay_plugin_t *, int, int, int,
+    struct sockaddr *, socklen_t, int *);
+extern void overlay_mux_close(overlay_mux_t *);
+extern void overlay_mux_add_dev(overlay_mux_t *, overlay_dev_t *);
+extern void overlay_mux_remove_dev(overlay_mux_t *, overlay_dev_t *);
+extern int overlay_mux_tx(overlay_mux_t *, struct msghdr *, mblk_t *);
+
+extern void overlay_prop_init(overlay_prop_handle_t);
+
+extern void overlay_target_init(void);
+extern int overlay_target_busy(void);
+extern int overlay_target_open(dev_t *, int, int, cred_t *);
+extern int overlay_target_ioctl(dev_t, int, intptr_t, int, cred_t *, int *);
+extern int overlay_target_close(dev_t, int, int, cred_t *);
+extern void overlay_target_free(overlay_dev_t *);
+
+#define	OVERLAY_TARGET_OK	0
+#define	OVERLAY_TARGET_DROP	1
+#define	OVERLAY_TARGET_ASYNC	2
+extern int overlay_target_lookup(overlay_dev_t *, mblk_t *, struct sockaddr *,
+    socklen_t *);
+extern void overlay_target_quiesce(overlay_target_t *);
+extern void overlay_target_fini(void);
+
+extern void overlay_fm_init(void);
+extern void overlay_fm_fini(void);
+extern void overlay_fm_degrade(overlay_dev_t *, const char *);
+extern void overlay_fm_restore(overlay_dev_t *);
+
+extern overlay_dev_t *overlay_hold_by_dlid(datalink_id_t);
+extern void overlay_hold_rele(overlay_dev_t *);
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif /* _SYS_OVERLAY_IMPL_H */
diff --git a/usr/src/uts/common/sys/overlay_plugin.h b/usr/src/uts/common/sys/overlay_plugin.h
new file mode 100644
index 0000000000..07efaa05df
--- /dev/null
+++ b/usr/src/uts/common/sys/overlay_plugin.h
@@ -0,0 +1,324 @@
+/*
+ * This file and its contents are supplied under the terms of the
+ * Common Development and Distribution License ("CDDL"), version 1.0.
+ * You may only use this file in accordance with the terms of version
+ * 1.0 of the CDDL.
+ *
+ * A full copy of the text of the CDDL should have accompanied this
+ * source.  A copy of the CDDL is also available via the Internet at
+ * http://www.illumos.org/license/CDDL.
+ */
+
+/*
+ * Copyright (c) 2015 Joyent, Inc.
+ */
+
+#ifndef _SYS_OVERLAY_PLUGIN_H
+#define	_SYS_OVERLAY_PLUGIN_H
+
+/*
+ * overlay plugin interface for encapsulation/decapsulation modules
+ *
+ * This header file defines how encapsulation and decapsulation plugins
+ * interact within the broader system. At this time, these interfaces are
+ * considered private to illumos and therefore are subject to change. As we gain
+ * more experience with a few of the different encapsulation formats, say nvgre
+ * or geneve, then we can move to make this a more-stable interface.
+ *
+ * A plugin is a general kernel module that uses the miscellaneous mod-linkage.
+ *
+ * In it's _init(9E) routine, it must register itself with the overlay
+ * subsystem. To do this, it allocates an overlay_plugin_register_t via
+ * overlay_plugin_alloc(), that it then  * fills out with various required
+ * information and then attempts to register with the system via a call to
+ * overlay_plugin_register(). If that succeeds, it should then call
+ * mod_install(9F). If the mod_install(9F) fails, then it should call
+ * overlay_plugin_unregister(). Regardless of success or failure, it should call
+ * overlay_plugin_free() to ensure that any memory that may be associated with
+ * the registration is freed.
+ *
+ * When the module's _fini(9E) is called, overlay_plugin_unregister() should be
+ * called first. It may return an error, such as EBUSY. In such cases, it should
+ * be returned as the return status of _fini(9E). This is quite necessary, it
+ * ensures that if the module is in use it doesn't get unloaded out from under
+ * us the broader subsystem while it's still in use. A driver can use that to
+ * know that there are no current instances of its private data.
+ *
+ * ------------------
+ * Plugin Definitions
+ * ------------------
+ *
+ * A plugin is required to fill in both an operations vector and a series of
+ * information to the callback routine. Here are the routines and their
+ * purposes. The full signatures are available below.
+ *
+ *   overlay_plugin_init_t
+ *
+ * 	This interface is used to create a new instance of a plugin. An instance
+ * 	of a plugin will be created for each overlay device that is created. For
+ * 	example, if a device is created with VXLAN ID 23 and ID 42, then there
+ * 	will be two different calls to this function.
+ *
+ * 	This function gives the plugin a chance to create a private data
+ * 	structure that will be returned on subsequent calls to the system.
+ *
+ *   overlay_plugin_fini_t
+ *
+ *   	This is the opposite of overlay_plugin_init_t. It will be called when it
+ *   	is safe to remove any private data that is associated with this instance
+ *   	of the plugin.
+ *
+ *   overlay_plugin_propinfo_t
+ *
+ *   	This is called with the name of a property that is registered when the
+ *   	plugin is created. This function will be called with the name of the
+ *   	property that information is being requested about. The plugin is
+ *   	responsible for filling out information such as setting the name, the
+ *   	type of property it is, the protection of the property (can a user
+ *   	update it?), whether the property is required, an optional default value
+ *   	for the property, and an optional set of values or ranges that are
+ *   	allowed.
+ *
+ *   overlay_plugin_getprop_t
+ *
+ *	Return the value of the named property from the current instance of the
+ *	plugin.
+ *
+ *   overlay_plugin_setprop_t
+ *
+ *	Set the value of the named property to the specified value for the
+ *	current instance of the plugin. Note, that it is the plugin's
+ *	responsibility to ensure that the value of the property is valid and to
+ *	update state as appropriate.
+ *
+ *   overlay_plugin_socket_t
+ *
+ *   	Every overlay device has a corresponding socket that it uses to send and
+ *   	receive traffic. This routine is used to get the parameters that should
+ *   	be used to define such a socket. The actual socket may be multiplexed
+ *   	with other uses of it.
+ *
+ *   overlay_plugin_sockopt_t
+ *
+ *   	Allow a plugin to set any necessary socket options that it needs on the
+ *   	kernel socket that is being used by a mux. This will only be called once
+ *   	for a given mux, if additional devices are added to a mux, it will not
+ *   	be called additional times.
+ *
+ *   overlay_plugin_encap_t
+ *
+ *   	In this routine you're given a message block and information about the
+ *   	packet, such as the identifier and are asked to fill out a message block
+ *   	that represents the encapsulation header and optionally manipulate the
+ *   	input message if required.
+ *
+ *   overlay_plugin_decap_t
+ *
+ *   	In this routine, you're given the encapsulated message block. The
+ *   	requirement is to decapsulate it and determine what is the correct
+ *   	overlay identifier for this network and to fill in the header size so
+ *   	the broader system knows how much of this data should be considered
+ *   	consumed.
+ *
+ *   ovpo_callbacks
+ *
+ *   	This should be set to zero, it's reserved for future use.
+ *
+ * Once these properties are defined, the module should define the following
+ * members in the overlay_plugin_register_t.
+ *
+ *   ovep_version
+ *
+ *   	Should be set to the value of the macro OVEP_VERSION.
+ *
+ *   ovep_name
+ *
+ *   	Should be set to a character string that has the name of the module.
+ *   	Generally this should match the name of the kernel module; however, this
+ *   	is the name that users will use to refer to this module when creating
+ *   	devices.
+ *
+ *   overlay_plugin_ops_t
+ *
+ *   	Should be set to the functions as described above.
+ *
+ *   ovep_props
+ *
+ *   	This is an array of character strings that holds the names of the
+ *   	properties of the encapsulation plugin.
+ *
+ *
+ *   ovep_id_size
+ *
+ *   	This is the size in bytes of the valid range for the identifier. The
+ *   	valid identifier range is considered a ovep_id_size byte unsigned
+ *   	integer, [ 0, 1 << (ovep_id_size * 8) ).
+ *
+ *   ovep_flags
+ *
+ *   	A series of flags that indicate optional features that are supported.
+ *   	Valid flags include:
+ *
+ *   		OVEP_F_VLAN_TAG
+ *
+ * 			The encapsulation format allows for the encapsulated
+ * 			packet to maintain a VLAN tag.
+ *
+ *   ovep_dest
+ *
+ *   	Describes the kind of destination that the overlay plugin supports for
+ *   	sending traffic. For example, vxlan uses UDP, therefore it requires both
+ *   	an IP address and a port; however, nvgre uses the gre header and
+ *   	therefore only requires an IP address. The following flags may be
+ *   	combined:
+ *
+ *   		OVERLAY_PLUGIN_D_ETHERNET
+ *
+ *   			Indicates that to send a packet to its destination, we
+ *   			require a link-layer ethernet address.
+ *
+ * 		OVERLAY_PLUGIN_D_IP
+ *
+ * 			Indicates that to send a packet to its destination, we
+ * 			require an IP address. Note, all IP addresses are
+ * 			transmitted as IPv6 addresses and for an IPv4
+ * 			destination, using an IPv4-mapped IPv6 address is the
+ * 			expected way to transmit that.
+ *
+ * 		OVERLAY_PLUGIN_D_PORT
+ *
+ * 			Indicates that to send a packet to its destination, a
+ * 			port is required, this usually indicates that the
+ * 			protocol uses something like TCP or UDP.
+ *
+ *
+ * -------------------------------------------------
+ * Downcalls, Upcalls, and Synchronization Guarantees
+ * -------------------------------------------------
+ *
+ * Every instance of a given module is independent. The kernel only guarantees
+ * that it will probably perform downcalls into different instances in parallel
+ * at some point. No locking is provided by the framework for synchronization
+ * across instances. If a module finds itself needing that, it will be up to it
+ * to provide it.
+ *
+ * In a given instance, the kernel may call into entry points in parallel. If
+ * the instance has private data, it should likely synchronize it. The one
+ * guarantee that we do make, is that calls to getprop and setprop will be done
+ * synchronized by a caller holding the MAC perimeter.
+ *
+ * While servicing a downcall from the general overlay device framework, a
+ * kernel module should not make any upcalls, excepting those functions that are
+ * defined in this header file, eg. the property related callbacks. Improtantly,
+ * it cannot make any assumptions about what locks may or may not be held by the
+ * broader system. The only thing that it is safe for it to use are its own
+ * locks.
+ *
+ * ----------------
+ * Downcall Context
+ * ----------------
+ *
+ * For all of the downcalls, excepting the overlay_plugin_encap_t and
+ * overlay_plugin_decap_t, the calls will be made either in kernel or user
+ * context, the module should not assume either way.
+ *
+ * overlay_plugin_encap_t and overlay_plugin_decap_t may be called in user,
+ * kernel or interrupt context; however, it is guaranteed that the interrupt
+ * will be below LOCK_LEVEL, and therefore it is safe to grab locks.
+ */
+
+#include <sys/stream.h>
+#include <sys/mac_provider.h>
+#include <sys/ksocket.h>
+#include <sys/overlay_common.h>
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+#define	OVEP_VERSION	0x1
+
+typedef enum overlay_plugin_flags {
+	OVEP_F_VLAN_TAG	= 0x01	/* Supports VLAN Tags */
+} overlay_plugin_flags_t;
+
+/*
+ * The ID space could easily be more than a 64-bit number, even
+ * though today it's either a 24-64 bit value. How should we future
+ * proof ourselves here?
+ */
+typedef struct ovep_encap_info {
+	uint64_t	ovdi_id;
+	size_t		ovdi_hdr_size;
+} ovep_encap_info_t;
+
+typedef struct __overlay_prop_handle *overlay_prop_handle_t;
+typedef struct __overlay_handle *overlay_handle_t;
+
+/*
+ * Plugins are guaranteed that calls to setprop are serialized. However, any
+ * number of other calls can be going on in parallel otherwise.
+ */
+typedef int (*overlay_plugin_encap_t)(void *, mblk_t *,
+    ovep_encap_info_t *, mblk_t **);
+typedef int (*overlay_plugin_decap_t)(void *, mblk_t *,
+    ovep_encap_info_t *);
+typedef int (*overlay_plugin_init_t)(overlay_handle_t, void **);
+typedef void (*overlay_plugin_fini_t)(void *);
+typedef int (*overlay_plugin_socket_t)(void *, int *, int *, int *,
+    struct sockaddr *, socklen_t *);
+typedef int (*overlay_plugin_sockopt_t)(ksocket_t);
+typedef int (*overlay_plugin_getprop_t)(void *, const char *, void *,
+    uint32_t *);
+typedef int (*overlay_plugin_setprop_t)(void *, const char *, const void *,
+    uint32_t);
+typedef int (*overlay_plugin_propinfo_t)(const char *, overlay_prop_handle_t);
+
+typedef struct overlay_plugin_ops {
+	uint_t			ovpo_callbacks;
+	overlay_plugin_init_t	ovpo_init;
+	overlay_plugin_fini_t	ovpo_fini;
+	overlay_plugin_encap_t	ovpo_encap;
+	overlay_plugin_decap_t	ovpo_decap;
+	overlay_plugin_socket_t ovpo_socket;
+	overlay_plugin_sockopt_t ovpo_sockopt;
+	overlay_plugin_getprop_t ovpo_getprop;
+	overlay_plugin_setprop_t ovpo_setprop;
+	overlay_plugin_propinfo_t ovpo_propinfo;
+} overlay_plugin_ops_t;
+
+typedef struct overlay_plugin_register {
+	uint_t			ovep_version;
+	const char		*ovep_name;
+	const overlay_plugin_ops_t	*ovep_ops;
+	const char 		**ovep_props;
+	uint_t			ovep_id_size;
+	uint_t			ovep_flags;
+	uint_t			ovep_dest;
+} overlay_plugin_register_t;
+
+/*
+ * Functions that interact with registration
+ */
+extern overlay_plugin_register_t *overlay_plugin_alloc(uint_t);
+extern void overlay_plugin_free(overlay_plugin_register_t *);
+extern int overlay_plugin_register(overlay_plugin_register_t *);
+extern int overlay_plugin_unregister(const char *);
+
+/*
+ * Property information callbacks
+ */
+extern void overlay_prop_set_name(overlay_prop_handle_t, const char *);
+extern void overlay_prop_set_prot(overlay_prop_handle_t, overlay_prop_prot_t);
+extern void overlay_prop_set_type(overlay_prop_handle_t, overlay_prop_type_t);
+extern int overlay_prop_set_default(overlay_prop_handle_t, void *, ssize_t);
+extern void overlay_prop_set_nodefault(overlay_prop_handle_t);
+extern void overlay_prop_set_range_uint32(overlay_prop_handle_t, uint32_t,
+    uint32_t);
+extern void overlay_prop_set_range_str(overlay_prop_handle_t, const char *);
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif /* _SYS_OVERLAY_PLUGIN_H */
diff --git a/usr/src/uts/common/sys/overlay_target.h b/usr/src/uts/common/sys/overlay_target.h
new file mode 100644
index 0000000000..cae193c334
--- /dev/null
+++ b/usr/src/uts/common/sys/overlay_target.h
@@ -0,0 +1,292 @@
+/*
+ * This file and its contents are supplied under the terms of the
+ * Common Development and Distribution License ("CDDL"), version 1.0.
+ * You may only use this file in accordance with the terms of version
+ * 1.0 of the CDDL.
+ *
+ * A full copy of the text of the CDDL should have accompanied this
+ * source.  A copy of the CDDL is also available via the Internet at
+ * http://www.illumos.org/license/CDDL.
+ */
+
+/*
+ * Copyright (c) 2015 Joyent, Inc.
+ */
+
+#ifndef _OVERLAY_TARGET_H
+#define	_OVERLAY_TARGET_H
+
+/*
+ * Overlay device varpd ioctl interface (/dev/overlay)
+ */
+
+#include <sys/types.h>
+#include <sys/ethernet.h>
+#include <netinet/in.h>
+#include <sys/overlay_common.h>
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+typedef struct overlay_target_point {
+	uint8_t		otp_mac[ETHERADDRL];
+	struct in6_addr	otp_ip;
+	uint16_t	otp_port;
+} overlay_target_point_t;
+
+#define	OVERLAY_TARG_IOCTL	(('o' << 24) | ('v' << 16) | ('t' << 8))
+
+#define	OVERLAY_TARG_INFO	(OVERLAY_TARG_IOCTL | 0x01)
+
+typedef enum overlay_targ_info_flags {
+	OVERLAY_TARG_INFO_F_ACTIVE = 0x01,
+	OVERLAY_TARG_INFO_F_DEGRADED = 0x02
+} overlay_targ_info_flags_t;
+
+/*
+ * Get target information about an overlay device
+ */
+typedef struct overlay_targ_info {
+	datalink_id_t		oti_linkid;
+	uint32_t		oti_needs;
+	uint64_t		oti_flags;
+	uint64_t		oti_vnetid;
+} overlay_targ_info_t;
+
+/*
+ * Declare an association between a given varpd instance and a datalink.
+ */
+#define	OVERLAY_TARG_ASSOCIATE	(OVERLAY_TARG_IOCTL | 0x02)
+
+typedef struct overlay_targ_associate {
+	datalink_id_t		ota_linkid;
+	uint32_t		ota_mode;
+	uint64_t		ota_id;
+	uint32_t		ota_provides;
+	overlay_target_point_t	ota_point;
+} overlay_targ_associate_t;
+
+/*
+ * Remove an association from a device. If the device has already been started,
+ * this implies OVERLAY_TARG_DEGRADE.
+ */
+#define	OVERLAY_TARG_DISASSOCIATE	(OVERLAY_TARG_IOCTL | 0x3)
+
+/*
+ * Tells the kernel that while a varpd instance still exists, it basically isn't
+ * making any forward progress, so the device should consider itself degraded.
+ */
+#define	OVERLAY_TARG_DEGRADE	(OVERLAY_TARG_IOCTL | 0x4)
+
+typedef struct overlay_targ_degrade {
+	datalink_id_t	otd_linkid;
+	uint32_t	otd_pad;
+	char		otd_buf[OVERLAY_STATUS_BUFLEN];
+} overlay_targ_degrade_t;
+
+/*
+ * Tells the kernel to remove the degraded status that it set on a device.
+ */
+#define	OVERLAY_TARG_RESTORE	(OVERLAY_TARG_IOCTL | 0x5)
+
+typedef struct overlay_targ_id {
+	datalink_id_t	otid_linkid;
+} overlay_targ_id_t;
+
+/*
+ * The following ioctls are all used to support dynamic lookups from userland,
+ * generally serviced by varpd.
+ *
+ * The way this is designed to work is that user land will have threads sitting
+ * in OVERLAY_TARG_LOOKUP ioctls waiting to service requests. A thread will sit
+ * waiting for work for up to approximately one second of time before they will
+ * be sent back out to user land to give user land a chance to clean itself up
+ * or more generally, come back into the kernel for work. Once these threads
+ * return, they will have a request with which more action can be done. The
+ * following ioctls can all be used to answer the request.
+ *
+ *	OVERLAY_TARG_RESPOND - overlay_targ_resp_t
+ *
+ *		The overlay_targ_resp_t has the appropriate information from
+ *		which a reply can be generated. The information is filled into
+ *		an overlay_targ_point_t as appropriate based on the
+ *		overlay_plugin_dest_t type.
+ *
+ *
+ *	OVERLAY_TARG_DROP - overlay_targ_resp_t
+ *
+ *		The overlay_targ_resp_t should identify a request for which to
+ *		drop a packet.
+ *
+ *
+ * 	OVERLAY_TARG_INJECT - overlay_targ_pkt_t
+ *
+ * 		The overlay_targ_pkt_t injects a fully formed packet into the
+ * 		virtual network. It may either be identified by its data link id
+ * 		or by the request id. If both are specified, the
+ * 		datalink id will be used. Note, that an injection is not
+ * 		considered a reply and if this corresponds to a requeset, then
+ * 		that individual packet must still be dropped.
+ *
+ *
+ * 	OVERLAY_TARG_PKT - overlay_targ_pkt_t
+ *
+ * 		This ioctl can be used to copy data from a given request into a
+ * 		user buffer. This can be used in combination with
+ * 		OVERLAY_TARG_INJECT to implemnt services such as a proxy-arp.
+ *
+ *
+ * 	OVERLAY_TARG_RESEND - overlay_targ_pkt_t
+ *
+ * 		This ioctl is similar to the OVERLAY_TARG_INJECT, except instead
+ * 		of receiving it on the local mac handle, it queues it for
+ * 		retransmission again. This is useful if you have a packet that
+ * 		was originally destined for some broadcast or multicast address
+ * 		that you now want to send to a unicast address.
+ */
+#define	OVERLAY_TARG_LOOKUP	(OVERLAY_TARG_IOCTL | 0x10)
+#define	OVERLAY_TARG_RESPOND	(OVERLAY_TARG_IOCTL | 0x11)
+#define	OVERLAY_TARG_DROP	(OVERLAY_TARG_IOCTL | 0x12)
+#define	OVERLAY_TARG_INJECT	(OVERLAY_TARG_IOCTL | 0x13)
+#define	OVERLAY_TARG_PKT	(OVERLAY_TARG_IOCTL | 0x14)
+#define	OVERLAY_TARG_RESEND	(OVERLAY_TARG_IOCTL | 0x15)
+
+typedef struct overlay_targ_lookup {
+	uint64_t	otl_dlid;
+	uint64_t	otl_reqid;
+	uint64_t	otl_varpdid;
+	uint64_t	otl_vnetid;
+	uint64_t	otl_hdrsize;
+	uint64_t	otl_pktsize;
+	uint8_t		otl_srcaddr[ETHERADDRL];
+	uint8_t		otl_dstaddr[ETHERADDRL];
+	uint32_t	otl_dsttype;
+	uint32_t	otl_sap;
+	int32_t		otl_vlan;
+} overlay_targ_lookup_t;
+
+typedef struct overlay_targ_resp {
+	uint64_t	otr_reqid;
+	overlay_target_point_t otr_answer;
+} overlay_targ_resp_t;
+
+typedef struct overlay_targ_pkt {
+	uint64_t	otp_linkid;
+	uint64_t	otp_reqid;
+	uint64_t	otp_size;
+	void		*otp_buf;
+} overlay_targ_pkt_t;
+
+#ifdef _KERNEL
+
+typedef struct overlay_targ_pkt32 {
+	uint64_t	otp_linkid;
+	uint64_t	otp_reqid;
+	uint64_t	otp_size;
+	caddr32_t	otp_buf;
+} overlay_targ_pkt32_t;
+
+#endif /* _KERNEL */
+
+/*
+ * This provides a way to get a list of active overlay devices independently
+ * from dlmgmtd. At the end of the day the kernel always knows what will exist
+ * and this allows varpd which is an implementation of libdladm not to end up
+ * needing to call back into dlmgmtd via libdladm and create an unfortunate
+ * dependency cycle.
+ */
+
+#define	OVERLAY_TARG_LIST	(OVERLAY_TARG_IOCTL | 0x20)
+
+typedef struct overlay_targ_list {
+	uint32_t	otl_nents;
+	uint32_t	otl_ents[];
+} overlay_targ_list_t;
+
+/*
+ * The following family of ioctls all manipulate the target cache of a given
+ * device.
+ *
+ * 	OVERLAY_TARG_CACHE_GET - overlay_targ_cache_t
+ *
+ * 		The overlay_targ_cache_t should be have its link identifier and
+ * 		the desired mac address filled in. On return, it will fill in
+ * 		the otc_dest member, if the entry exists in the table.
+ *
+ *
+ * 	OVERLAY_TARG_CACHE_SET - overlay_targ_cache_t
+ *
+ * 		The cache table entry of the mac address referred to by otc_mac
+ * 		and otd_linkid will be filled in with the details provided by in
+ * 		the otc_dest member.
+ *
+ * 	OVERLAY_TARG_CACHE_REMOVE - overlay_targ_cache_t
+ *
+ * 		Removes the cache entry identified by otc_mac from the table.
+ * 		Note that this does not stop any in-flight lookups or deal with
+ * 		any data that is awaiting a lookup.
+ *
+ *
+ * 	OVERLAY_TARG_CACHE_FLUSH - overlay_targ_cache_t
+ *
+ * 		Similar to OVERLAY_TARG_CACHE_REMOVE, but functions on the
+ * 		entire table identified by otc_linkid. All other parameters are
+ * 		ignored.
+ *
+ *
+ * 	OVERLAY_TARG_CACHE_ITER - overlay_targ_cache_iter_t
+ *
+ * 		Iterates over the contents of a target cache identified by
+ * 		otci_linkid. Iteration is guaranteed to be exactly once for
+ * 		items which are in the hashtable at the beginning and end of
+ * 		iteration. For items which are added or removed after iteration
+ * 		has begun, only at most once semantics are guaranteed. Consumers
+ * 		should ensure that otci_marker is zeroed before starting
+ * 		iteration and should preserve its contents across calls.
+ *
+ * 		Before calling in, otci_count should be set to the number of
+ * 		entries that space has been allocated for in otci_ents. The
+ * 		value will be updated to indicate the total number written out.
+ */
+
+#define	OVERLAY_TARG_CACHE_GET		(OVERLAY_TARG_IOCTL | 0x30)
+#define	OVERLAY_TARG_CACHE_SET		(OVERLAY_TARG_IOCTL | 0x31)
+#define	OVERLAY_TARG_CACHE_REMOVE	(OVERLAY_TARG_IOCTL | 0x32)
+#define	OVERLAY_TARG_CACHE_FLUSH	(OVERLAY_TARG_IOCTL | 0x33)
+#define	OVERLAY_TARG_CACHE_ITER		(OVERLAY_TARG_IOCTL | 0x34)
+
+/*
+ * This is a pretty arbitrary number that we're constraining ourselves to
+ * for iteration. Basically the goal is to make sure that we can't have a user
+ * ask us to allocate too much memory on their behalf at any time. A more
+ * dynamic form may be necessary some day.
+ */
+#define	OVERLAY_TARGET_ITER_MAX	500
+
+#define	OVERLAY_TARGET_CACHE_DROP	0x01
+
+typedef struct overlay_targ_cache_entry {
+	uint8_t			otce_mac[ETHERADDRL];
+	uint16_t		otce_flags;
+	overlay_target_point_t	otce_dest;
+} overlay_targ_cache_entry_t;
+
+typedef struct overlay_targ_cache {
+	datalink_id_t			otc_linkid;
+	overlay_targ_cache_entry_t	otc_entry;
+} overlay_targ_cache_t;
+
+typedef struct overlay_targ_cache_iter {
+	datalink_id_t			otci_linkid;
+	uint32_t			otci_pad;
+	uint64_t			otci_marker;
+	uint16_t			otci_count;
+	overlay_targ_cache_entry_t	otci_ents[];
+} overlay_targ_cache_iter_t;
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif /* _OVERLAY_TARGET_H */
diff --git a/usr/src/uts/common/sys/param.h b/usr/src/uts/common/sys/param.h
index 01c25610dd..12bdc29079 100644
--- a/usr/src/uts/common/sys/param.h
+++ b/usr/src/uts/common/sys/param.h
@@ -103,7 +103,7 @@ extern "C" {
 #define	DEFAULT_MAXPID	999999
 #define	DEFAULT_JUMPPID	100000
 #else
-#define	DEFAULT_MAXPID	30000
+#define	DEFAULT_MAXPID	99999
 #define	DEFAULT_JUMPPID	0
 #endif
 
diff --git a/usr/src/uts/common/sys/policy.h b/usr/src/uts/common/sys/policy.h
index 276c4d386e..5328d02c59 100644
--- a/usr/src/uts/common/sys/policy.h
+++ b/usr/src/uts/common/sys/policy.h
@@ -20,7 +20,7 @@
  */
 /*
  * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
- * Copyright 2013, Joyent, Inc. All rights reserved.
+ * Copyright 2015, Joyent, Inc. All rights reserved.
  */
 
 #ifndef	_SYS_POLICY_H
@@ -107,6 +107,8 @@ int secpolicy_ipc_config(const cred_t *);
 int secpolicy_ipc_owner(const cred_t *, const struct kipc_perm *);
 int secpolicy_kmdb(const cred_t *);
 int secpolicy_lock_memory(const cred_t *);
+int secpolicy_meminfo(const cred_t *);
+int secpolicy_fs_import(const cred_t *);
 int secpolicy_modctl(const cred_t *, int);
 int secpolicy_net(const cred_t *, int, boolean_t);
 int secpolicy_net_bindmlp(const cred_t *);
@@ -173,6 +175,7 @@ int secpolicy_setid_setsticky_clear(vnode_t *, vattr_t *,
     const vattr_t *, cred_t *);
 int secpolicy_xvattr(xvattr_t *, uid_t, cred_t *, vtype_t);
 int secpolicy_xvm_control(const cred_t *);
+int secpolicy_hyprlofs_control(const cred_t *);
 
 int secpolicy_basic_exec(const cred_t *, vnode_t *);
 int secpolicy_basic_fork(const cred_t *);
diff --git a/usr/src/uts/common/sys/poll.h b/usr/src/uts/common/sys/poll.h
index 9fff78a966..efc8457a6a 100644
--- a/usr/src/uts/common/sys/poll.h
+++ b/usr/src/uts/common/sys/poll.h
@@ -30,6 +30,10 @@
  * All rights reserved.
  */
 
+/*
+ * Copyright (c) 2014, Joyent, Inc. All rights reserved.
+ */
+
 #ifndef _SYS_POLL_H
 #define	_SYS_POLL_H
 
@@ -59,6 +63,7 @@ typedef unsigned long	nfds_t;
 #define	POLLWRNORM	POLLOUT
 #define	POLLRDBAND	0x0080		/* out-of-band data is readable */
 #define	POLLWRBAND	0x0100		/* out-of-band data is writeable */
+#define	POLLRDHUP	0x4000		/* read-side hangup */
 
 #define	POLLNORM	POLLRDNORM
 
@@ -70,7 +75,13 @@ typedef unsigned long	nfds_t;
 #define	POLLHUP		0x0010		/* fd has been hung up on */
 #define	POLLNVAL	0x0020		/* invalid pollfd entry */
 
-#define	POLLREMOVE	0x0800	/* remove a cached poll fd from /dev/poll */
+/*
+ * These events will never be specified in revents, but may be specified in
+ * events to control /dev/poll behavior.
+ */
+#define	POLLREMOVE	0x0800		/* remove cached /dev/poll fd */
+#define	POLLONESHOT	0x1000		/* /dev/poll should one-shot this fd */
+#define	POLLET		0x2000		/* edge-triggered /dev/poll fd */
 
 #ifdef _KERNEL
 
diff --git a/usr/src/uts/common/sys/poll_impl.h b/usr/src/uts/common/sys/poll_impl.h
index ede99d0df2..2e866ec4d4 100644
--- a/usr/src/uts/common/sys/poll_impl.h
+++ b/usr/src/uts/common/sys/poll_impl.h
@@ -24,11 +24,13 @@
  * Use is subject to license terms.
  */
 
+/*
+ * Copyright (c) 2014, Joyent, Inc. All rights reserved.
+ */
+
 #ifndef _SYS_POLL_IMPL_H
 #define	_SYS_POLL_IMPL_H
 
-#pragma ident	"%Z%%M%	%I%	%E% SMI"
-
 /*
  * Caching Poll Subsystem:
  *
@@ -160,6 +162,7 @@ typedef struct polldat {
 	int		pd_nsets;	/* num of xref sets, used by poll(2) */
 	xref_t		*pd_ref;	/* ptr to xref info, 1 for each set */
 	struct port_kevent *pd_portev;	/* associated port event struct */
+	uint64_t	pd_epolldata;	/* epoll data, if any */
 } polldat_t;
 
 /*
@@ -187,7 +190,8 @@ typedef struct pollcache {
 } pollcache_t;
 
 /* pc_flag */
-#define	T_POLLWAKE	0x02	/* pollwakeup() occurred */
+#define	PC_POLLWAKE	0x02	/* pollwakeup() occurred */
+#define	PC_WRITEWANTED	0x04	/* writer wishes to modify the pollcache_t */
 
 #if defined(_KERNEL)
 /*
diff --git a/usr/src/uts/common/sys/proc.h b/usr/src/uts/common/sys/proc.h
index f1a2fc5485..e811bebf25 100644
--- a/usr/src/uts/common/sys/proc.h
+++ b/usr/src/uts/common/sys/proc.h
@@ -21,6 +21,7 @@
 
 /*
  * Copyright (c) 1988, 2010, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2015 Joyent, Inc.  All rights reserved.
  */
 
 /*	Copyright (c) 1984, 1986, 1987, 1988, 1989 AT&T	*/
@@ -346,7 +347,9 @@ typedef struct	proc {
 	struct zone	*p_zone;	/* zone in which process lives */
 	struct vnode	*p_execdir;	/* directory that p_exec came from */
 	struct brand	*p_brand;	/* process's brand  */
-	void		*p_brand_data;	/* per-process brand state */
+
+	/* per-process brand state */
+	void		*p_brand_data;
 
 	/* additional lock to protect p_sessp (but not its contents) */
 	kmutex_t p_splock;
@@ -361,7 +364,6 @@ typedef struct	proc {
 	 */
 	struct user p_user;		/* (see sys/user.h) */
 } proc_t;
-
 #define	PROC_T				/* headers relying on proc_t are OK */
 
 #ifdef _KERNEL
@@ -627,6 +629,7 @@ extern int signal_is_blocked(kthread_t *, int);
 extern int sigcheck(proc_t *, kthread_t *);
 extern void sigdefault(proc_t *);
 
+extern struct pid *pid_find(pid_t pid);
 extern void pid_setmin(void);
 extern pid_t pid_allocate(proc_t *, pid_t, int);
 extern int pid_rele(struct pid *);
@@ -716,6 +719,10 @@ extern	kthread_t *thread_unpin(void);
 extern	void	thread_init(void);
 extern	void	thread_load(kthread_t *, void (*)(), caddr_t, size_t);
 
+extern	void	thread_splitstack(void (*)(void *), void *, size_t);
+extern	void	thread_splitstack_run(caddr_t, void (*)(void *), void *);
+extern	void	thread_splitstack_cleanup(void);
+
 extern	void	tsd_create(uint_t *, void (*)(void *));
 extern	void	tsd_destroy(uint_t *);
 extern	void	*tsd_getcreate(uint_t *, void (*)(void *), void *(*)(void));
diff --git a/usr/src/uts/common/sys/procfs.h b/usr/src/uts/common/sys/procfs.h
index f592fd9dcf..501af712ef 100644
--- a/usr/src/uts/common/sys/procfs.h
+++ b/usr/src/uts/common/sys/procfs.h
@@ -25,6 +25,7 @@
  */
 /*
  * Copyright 2012 DEY Storage Systems, Inc.  All rights reserved.
+ * Copyright 2015, Joyent, Inc.
  */
 
 #ifndef _SYS_PROCFS_H
@@ -233,6 +234,7 @@ typedef struct pstatus {
 #define	PR_FAULTED	6
 #define	PR_SUSPENDED	7
 #define	PR_CHECKPOINT	8
+#define	PR_BRAND	9
 
 /*
  * lwp ps(1) information file.  /proc/<pid>/lwp/<lwpid>/lwpsinfo
diff --git a/usr/src/uts/common/sys/ptms.h b/usr/src/uts/common/sys/ptms.h
index 6c79ee266d..ba8b2b1210 100644
--- a/usr/src/uts/common/sys/ptms.h
+++ b/usr/src/uts/common/sys/ptms.h
@@ -126,6 +126,12 @@ extern void ptms_logp(char *, uintptr_t);
 #define	DDBGP(a, b)
 #endif
 
+typedef struct __ptmptsopencb_arg *ptmptsopencb_arg_t;
+typedef struct ptmptsopencb {
+	boolean_t		(*ppocb_func)(ptmptsopencb_arg_t);
+	ptmptsopencb_arg_t	ppocb_arg;
+} ptmptsopencb_t;
+
 #endif /* _KERNEL */
 
 typedef struct pt_own {
@@ -157,6 +163,19 @@ typedef struct pt_own {
 #define	ZONEPT		(('P'<<8)|4)	/* set zone of master/slave pair */
 #define	OWNERPT		(('P'<<8)|5)	/* set owner/group for slave device */
 
+#ifdef _KERNEL
+/*
+ * kernel ioctl commands
+ *
+ * PTMPTSOPENCB: Returns a callback function pointer and opaque argument.
+ *	      The return value of the callback function when it's invoked
+ *	      with the opaque argument passed to it will indicate if the
+ *	      pts slave device is currently open.
+ */
+#define	PTMPTSOPENCB	(('P'<<8)|6)	/* check if the slave is open */
+
+#endif /* _KERNEL */
+
 #ifdef	__cplusplus
 }
 #endif
diff --git a/usr/src/uts/common/sys/scsi/adapters/mpt_sas/mptsas_hash.h b/usr/src/uts/common/sys/refhash.h
index 2069e6d3f1..b7427a454d 100644
--- a/usr/src/uts/common/sys/scsi/adapters/mpt_sas/mptsas_hash.h
+++ b/usr/src/uts/common/sys/refhash.h
@@ -10,11 +10,11 @@
  */
 
 /*
- * Copyright 2014 Joyent, Inc.  All rights reserved.
+ * Copyright 2015 Joyent, Inc.
  */
 
-#ifndef	_SYS_SCSI_ADAPTERS_MPTHASH_H
-#define	_SYS_SCSI_ADAPTERS_MPTHASH_H
+#ifndef	_SYS_REFHASH_H
+#define	_SYS_REFHASH_H
 
 #include <sys/types.h>
 #include <sys/list.h>
@@ -58,4 +58,4 @@ extern void *refhash_first(refhash_t *);
 extern void *refhash_next(refhash_t *, void *);
 extern boolean_t refhash_obj_valid(refhash_t *hp, const void *);
 
-#endif	/* _SYS_SCSI_ADAPTERS_MPTHASH_H */
+#endif	/* _SYS_REFHASH_H */
diff --git a/usr/src/uts/common/sys/resource.h b/usr/src/uts/common/sys/resource.h
index d69b23aade..666fbcd8a4 100644
--- a/usr/src/uts/common/sys/resource.h
+++ b/usr/src/uts/common/sys/resource.h
@@ -23,6 +23,7 @@
  *
  * Copyright 2006 Sun Microsystems, Inc.  All rights reserved.
  * Use is subject to license terms.
+ * Copyright 2014 Joyent, Inc.  All rights reserved.
  */
 
 /*	Copyright (c) 1984, 1986, 1987, 1988, 1989 AT&T	*/
@@ -191,6 +192,7 @@ struct	rusage {
 #define	_RUSAGESYS_GETRUSAGE_CHLD	1	/* rusage child process */
 #define	_RUSAGESYS_GETRUSAGE_LWP	2	/* rusage lwp */
 #define	_RUSAGESYS_GETVMUSAGE		3	/* getvmusage */
+#define	_RUSAGESYS_INVALMAP		4	/* vm_map_inval */
 
 #if defined(_SYSCALL32)
 
diff --git a/usr/src/uts/common/sys/scsi/adapters/mpt_sas/mptsas_var.h b/usr/src/uts/common/sys/scsi/adapters/mpt_sas/mptsas_var.h
index 836548aa30..2689fe27c4 100644
--- a/usr/src/uts/common/sys/scsi/adapters/mpt_sas/mptsas_var.h
+++ b/usr/src/uts/common/sys/scsi/adapters/mpt_sas/mptsas_var.h
@@ -22,7 +22,7 @@
 /*
  * Copyright (c) 2009, 2010, Oracle and/or its affiliates. All rights reserved.
  * Copyright 2014 Nexenta Systems, Inc. All rights reserved.
- * Copyright (c) 2013, Joyent, Inc. All rights reserved.
+ * Copyright 2015, Joyent, Inc.
  * Copyright (c) 2014, Tegile Systems Inc. All rights reserved.
  */
 
@@ -58,10 +58,10 @@
 
 #include <sys/byteorder.h>
 #include <sys/queue.h>
+#include <sys/refhash.h>
 #include <sys/isa_defs.h>
 #include <sys/sunmdi.h>
 #include <sys/mdi_impldefs.h>
-#include <sys/scsi/adapters/mpt_sas/mptsas_hash.h>
 #include <sys/scsi/adapters/mpt_sas/mptsas_ioctl.h>
 #include <sys/scsi/adapters/mpt_sas/mpi/mpi2_tool.h>
 #include <sys/scsi/adapters/mpt_sas/mpi/mpi2_cnfg.h>
diff --git a/usr/src/uts/common/sys/signal.h b/usr/src/uts/common/sys/signal.h
index 8f0e1794f4..139784d578 100644
--- a/usr/src/uts/common/sys/signal.h
+++ b/usr/src/uts/common/sys/signal.h
@@ -22,6 +22,7 @@
 /*
  * Copyright 2010 Sun Microsystems, Inc.  All rights reserved.
  * Use is subject to license terms.
+ * Copyright 2015, Joyent, Inc.
  */
 
 /*	Copyright (c) 1984, 1986, 1987, 1988, 1989 AT&T	*/
@@ -158,8 +159,8 @@ struct sigaction32 {
  * use of these symbols by applications is injurious
  *	to binary compatibility
  */
-#define	NSIG	74	/* valid signals range from 1 to NSIG-1 */
-#define	MAXSIG	73	/* size of u_signal[], NSIG-1 <= MAXSIG */
+#define	NSIG	75	/* valid signals range from 1 to NSIG-1 */
+#define	MAXSIG	74	/* size of u_signal[], NSIG-1 <= MAXSIG */
 #endif /* defined(__EXTENSIONS__) || !defined(_XPG4_2) */
 
 #define	MINSIGSTKSZ	2048
diff --git a/usr/src/uts/common/sys/socket.h b/usr/src/uts/common/sys/socket.h
index 590551f6f2..f5864db7d3 100644
--- a/usr/src/uts/common/sys/socket.h
+++ b/usr/src/uts/common/sys/socket.h
@@ -22,6 +22,7 @@
  * Copyright 2014 Garrett D'Amore <garrett@damore.org>
  *
  * Copyright (c) 1989, 2010, Oracle and/or its affiliates. All rights reserved.
+ * Copyright 2015, Joyent, Inc. All rights reserved.
  */
 
 /*	Copyright (c) 1983, 1984, 1985, 1986, 1987, 1988, 1989 AT&T	*/
@@ -39,6 +40,9 @@
 
 /* Copyright (c) 2013, OmniTI Computer Consulting, Inc. All rights reserved. */
 
+/*
+ * Copyright (c) 2014, Joyent, Inc. All rights reserved.
+ */
 #ifndef	_SYS_SOCKET_H
 #define	_SYS_SOCKET_H
 
@@ -293,8 +297,9 @@ struct	linger {
 #define	AF_INET_OFFLOAD	30		/* Sun private; do not use */
 #define	AF_TRILL	31		/* TRILL interface */
 #define	AF_PACKET	32		/* PF_PACKET Linux socket interface */
+#define	AF_LX_NETLINK	33		/* Linux-compatible netlink */
 
-#define	AF_MAX		32
+#define	AF_MAX		33
 
 /*
  * Protocol families, same as address families for now.
@@ -334,6 +339,7 @@ struct	linger {
 #define	PF_INET_OFFLOAD	AF_INET_OFFLOAD	/* Sun private; do not use */
 #define	PF_TRILL	AF_TRILL
 #define	PF_PACKET	AF_PACKET
+#define	PF_LX_NETLINK	AF_LX_NETLINK
 
 #define	PF_MAX		AF_MAX
 
@@ -420,6 +426,7 @@ struct msghdr32 {
 #define	MSG_NOTIFICATION 0x100		/* Notification, not data */
 #define	MSG_XPG4_2	0x8000		/* Private: XPG4.2 flag */
 
+/* Obsolete but kept for compilation compatability. Use IOV_MAX. */
 #define	MSG_MAXIOVLEN	16
 
 #ifdef _KERNEL
diff --git a/usr/src/uts/common/sys/socketvar.h b/usr/src/uts/common/sys/socketvar.h
index 8221c620a8..b3f51cc0e1 100644
--- a/usr/src/uts/common/sys/socketvar.h
+++ b/usr/src/uts/common/sys/socketvar.h
@@ -21,6 +21,7 @@
 
 /*
  * Copyright (c) 1996, 2010, Oracle and/or its affiliates. All rights reserved.
+ * Copyright 2015 Joyent, Inc.
  */
 
 /*	Copyright (c) 1983, 1984, 1985, 1986, 1987, 1988, 1989 AT&T	*/
@@ -99,6 +100,7 @@ struct sockaddr_ux {
 
 typedef struct sonodeops sonodeops_t;
 typedef struct sonode sonode_t;
+typedef boolean_t (*so_krecv_f)(sonode_t *, mblk_t *, size_t, int, void *);
 
 struct sodirect_s;
 
@@ -241,6 +243,10 @@ struct sonode {
 	struct sof_instance	*so_filter_top;		/* top of stack */
 	struct sof_instance	*so_filter_bottom;	/* bottom of stack */
 	clock_t			so_filter_defertime;	/* time when deferred */
+
+	/* Kernel direct receive callbacks */
+	so_krecv_f		so_krecv_cb;		/* recv callback */
+	void			*so_krecv_arg;		/* recv cb arg */
 };
 
 #define	SO_HAVE_DATA(so)						\
@@ -294,15 +300,16 @@ struct sonode {
 #define	SS_OOBPEND		0x00002000 /* OOB pending or present - poll */
 #define	SS_HAVEOOBDATA		0x00004000 /* OOB data present */
 #define	SS_HADOOBDATA		0x00008000 /* OOB data consumed */
-#define	SS_CLOSING		0x00010000 /* in process of closing */
 
+#define	SS_CLOSING		0x00010000 /* in process of closing */
 #define	SS_FIL_DEFER		0x00020000 /* filter deferred notification */
 #define	SS_FILOP_OK		0x00040000 /* socket can attach filters */
 #define	SS_FIL_RCV_FLOWCTRL	0x00080000 /* filter asserted rcv flow ctrl */
+
 #define	SS_FIL_SND_FLOWCTRL	0x00100000 /* filter asserted snd flow ctrl */
 #define	SS_FIL_STOP		0x00200000 /* no more filter actions */
-
 #define	SS_SODIRECT		0x00400000 /* transport supports sodirect */
+#define	SS_FILOP_UNSF		0x00800000 /* block attaching unsafe filters */
 
 #define	SS_SENTLASTREADSIG	0x01000000 /* last rx signal has been sent */
 #define	SS_SENTLASTWRITESIG	0x02000000 /* last tx signal has been sent */
@@ -318,7 +325,8 @@ struct sonode {
 
 /*
  * Sockets that can fall back to TPI must ensure that fall back is not
- * initiated while a thread is using a socket.
+ * initiated while a thread is using a socket. Otherwise this disables all
+ * future filter attachment.
  */
 #define	SO_BLOCK_FALLBACK(so, fn)				\
 	ASSERT(MUTEX_NOT_HELD(&(so)->so_lock));			\
@@ -334,6 +342,24 @@ struct sonode {
 		}						\
 	}
 
+/*
+ * Sockets that can fall back to TPI must ensure that fall back is not
+ * initiated while a thread is using a socket. Otherwise this disables all
+ * future unsafe filter attachment. Safe filters can still attach after
+ * we execute the function in which this macro is used.
+ */
+#define	SO_BLOCK_FALLBACK_SAFE(so, fn)				\
+	ASSERT(MUTEX_NOT_HELD(&(so)->so_lock));			\
+	rw_enter(&(so)->so_fallback_rwlock, RW_READER);		\
+	if ((so)->so_state & SS_FALLBACK_COMP) {		\
+		rw_exit(&(so)->so_fallback_rwlock);		\
+		return (fn);					\
+	} else if (((so)->so_state & SS_FILOP_UNSF) == 0) {	\
+		mutex_enter(&(so)->so_lock);			\
+		(so)->so_state |= SS_FILOP_UNSF;		\
+		mutex_exit(&(so)->so_lock);			\
+	}
+
 #define	SO_UNBLOCK_FALLBACK(so)	{			\
 	rw_exit(&(so)->so_fallback_rwlock);		\
 }
@@ -942,6 +968,15 @@ extern struct sonode	*socreate(struct sockparams *, int, int, int, int,
 extern int	so_copyin(const void *, void *, size_t, int);
 extern int	so_copyout(const void *, void *, size_t, int);
 
+/*
+ * Functions to manipulate the use of direct receive callbacks. This should not
+ * be used outside of sockfs and ksocket. These are generally considered a use
+ * once interface for a socket and will cause all outstanding data on the socket
+ * to be flushed.
+ */
+extern int	so_krecv_set(sonode_t *, so_krecv_f, void *);
+extern void	so_krecv_unblock(sonode_t *);
+
 #endif
 
 /*
diff --git a/usr/src/uts/common/sys/sockfilter.h b/usr/src/uts/common/sys/sockfilter.h
index 9f6d8b499b..c4dd6539de 100644
--- a/usr/src/uts/common/sys/sockfilter.h
+++ b/usr/src/uts/common/sys/sockfilter.h
@@ -20,6 +20,7 @@
  */
 /*
  * Copyright (c) 2010, Oracle and/or its affiliates. All rights reserved.
+ * Copyright 2015 Joyent, Inc.
  */
 
 #ifndef	_SYS_SOCKFILTER_H
@@ -129,6 +130,15 @@ typedef struct sof_ops {
 
 #define	SOF_VERSION	1
 
+/*
+ * Flag indicating that the filter module is safe to attach after bind,
+ * getsockname, getsockopt or setsockopt calls. By default filters are unsafe
+ * so may not be attached after any socket operation. However, a safe filter
+ * can still be attached after one of the above calls. This makes attaching
+ * the filter less dependent on the initial socket setup order.
+ */
+#define	SOF_ATT_SAFE	0x1
+
 extern int	sof_register(int, const char *, const sof_ops_t *, int);
 extern int	sof_unregister(const char *);
 
diff --git a/usr/src/uts/common/sys/squeue.h b/usr/src/uts/common/sys/squeue.h
index f1bd429815..35e1cf64c7 100644
--- a/usr/src/uts/common/sys/squeue.h
+++ b/usr/src/uts/common/sys/squeue.h
@@ -29,6 +29,17 @@
 extern "C" {
 #endif
 
+/*
+ * Originally in illumos, we had an IP-centric view of the serialization queue
+ * abstraction. While that has useful properties, the implementation of squeues
+ * hardcodes various parts of the implementation of IP into it which makes it
+ * unsuitable for other consumers. To enable them, we created another interface,
+ * but opted not to port all of the functionality that IP uses in the form of
+ * ip_squeue.c As other consumers need the functionality that IP has in squeues,
+ * then we'll come up with more genericized methods and add that functionality
+ * to <sys/gsqueue.h>. Please do not continue to use this header.
+ */
+
 #include <sys/types.h>
 #include <sys/processor.h>
 #include <sys/stream.h>
@@ -76,12 +87,13 @@ typedef enum {
 
 struct ip_recv_attr_s;
 extern void squeue_init(void);
-extern squeue_t *squeue_create(clock_t, pri_t);
+extern squeue_t *squeue_create(clock_t, pri_t, boolean_t);
 extern void squeue_bind(squeue_t *, processorid_t);
 extern void squeue_unbind(squeue_t *);
 extern void squeue_enter(squeue_t *, mblk_t *, mblk_t *,
     uint32_t, struct ip_recv_attr_s *, int, uint8_t);
 extern uintptr_t *squeue_getprivate(squeue_t *, sqprivate_t);
+extern void squeue_destroy(squeue_t *);
 
 struct conn_s;
 extern int squeue_synch_enter(struct conn_s *, mblk_t *);
diff --git a/usr/src/uts/common/sys/squeue_impl.h b/usr/src/uts/common/sys/squeue_impl.h
index 22550886eb..d2418bbc15 100644
--- a/usr/src/uts/common/sys/squeue_impl.h
+++ b/usr/src/uts/common/sys/squeue_impl.h
@@ -117,6 +117,7 @@ struct squeue_s {
 	squeue_set_t	*sq_set;	/* managed by squeue creator */
 
 	pri_t		sq_priority;	/* squeue thread priority */
+	boolean_t	sq_isip;	/* use IP-centric features */
 
 	/* Keep the debug-only fields at the end of the structure */
 #ifdef DEBUG
@@ -165,6 +166,7 @@ struct squeue_s {
 #define	SQS_POLL_RESTART_DONE	0x01000000
 #define	SQS_POLL_THR_QUIESCE	0x02000000
 #define	SQS_PAUSE		0x04000000 /* The squeue has been paused */
+#define	SQS_EXIT		0x08000000 /* squeue is being torn down */
 
 #define	SQS_WORKER_THR_CONTROL          \
 	(SQS_POLL_QUIESCE | SQS_POLL_RESTART | SQS_POLL_CLEANUP)
diff --git a/usr/src/uts/common/sys/stream.h b/usr/src/uts/common/sys/stream.h
index 7a3b4e3448..4151cc0208 100644
--- a/usr/src/uts/common/sys/stream.h
+++ b/usr/src/uts/common/sys/stream.h
@@ -21,6 +21,7 @@
 /*
  * Copyright 2009 Sun Microsystems, Inc.  All rights reserved.
  * Use is subject to license terms.
+ * Copyright 2015 Joyent, Inc.  All rights reserved.
  */
 
 /*	Copyright (c) 1984, 1986, 1987, 1988, 1989 AT&T	*/
@@ -627,16 +628,11 @@ struct stroptions {
 /*
  * Structure for rw (read/write) procedure calls. A pointer
  * to a struiod_t is passed as a parameter to the rwnext() call.
- *
- * Note: DEF_IOV_MAX is defined and used as it is in "fs/vncalls.c"
- *	 as there isn't a formal definition of IOV_MAX ???
  */
-#define	DEF_IOV_MAX	16
-
 typedef struct struiod {
 	mblk_t		*d_mp;		/* pointer to mblk (chain) */
 	uio_t		d_uio;		/* uio info */
-	iovec_t d_iov[DEF_IOV_MAX];	/* iov referenced by uio */
+	iovec_t 	*d_iov;		/* iov referenced by uio */
 } struiod_t;
 
 /*
diff --git a/usr/src/uts/common/sys/sunddi.h b/usr/src/uts/common/sys/sunddi.h
index 8d1fd5e730..18d2284b4a 100644
--- a/usr/src/uts/common/sys/sunddi.h
+++ b/usr/src/uts/common/sys/sunddi.h
@@ -462,6 +462,7 @@ extern size_t strlcat(char *, const char *, size_t);
 extern size_t strlcpy(char *, const char *, size_t);
 extern size_t strspn(const char *, const char *);
 extern size_t strcspn(const char *, const char *);
+extern char *strsep(char **, const char *);
 extern int bcmp(const void *, const void *, size_t) __PURE;
 extern int stoi(char **);
 extern void numtos(ulong_t, char *);
diff --git a/usr/src/uts/common/sys/syscall.h b/usr/src/uts/common/sys/syscall.h
index e365668f85..7d86565564 100644
--- a/usr/src/uts/common/sys/syscall.h
+++ b/usr/src/uts/common/sys/syscall.h
@@ -511,8 +511,8 @@ typedef struct {	/* return values from system call */
 
 #if !defined(_KERNEL)
 
-extern int	syscall(int, ...);
-extern int	__systemcall(sysret_t *, int, ...);
+extern long	syscall(int, ...);
+extern long	__systemcall(sysret_t *, int, ...);
 extern int	__set_errno(int);
 
 #endif	/* _KERNEL */
diff --git a/usr/src/uts/common/sys/systrace.h b/usr/src/uts/common/sys/systrace.h
index d43974451e..17e509d4d8 100644
--- a/usr/src/uts/common/sys/systrace.h
+++ b/usr/src/uts/common/sys/systrace.h
@@ -22,13 +22,12 @@
 /*
  * Copyright 2006 Sun Microsystems, Inc.  All rights reserved.
  * Use is subject to license terms.
+ * Copyright 2014 Joyent, Inc.  All rights reserved.
  */
 
 #ifndef _SYS_SYSTRACE_H
 #define	_SYS_SYSTRACE_H
 
-#pragma ident	"%Z%%M%	%I%	%E% SMI"
-
 #include <sys/dtrace.h>
 
 #ifdef	__cplusplus
@@ -47,16 +46,18 @@ extern systrace_sysent_t *systrace_sysent;
 extern systrace_sysent_t *systrace_sysent32;
 
 extern void (*systrace_probe)(dtrace_id_t, uintptr_t, uintptr_t,
-    uintptr_t, uintptr_t, uintptr_t, uintptr_t);
+    uintptr_t, uintptr_t, uintptr_t, uintptr_t, uintptr_t, uintptr_t);
 extern void systrace_stub(dtrace_id_t, uintptr_t, uintptr_t,
-    uintptr_t, uintptr_t, uintptr_t, uintptr_t);
+    uintptr_t, uintptr_t, uintptr_t, uintptr_t, uintptr_t, uintptr_t);
 
 extern int64_t dtrace_systrace_syscall(uintptr_t arg0, uintptr_t arg1,
-    uintptr_t arg2, uintptr_t arg3, uintptr_t arg4, uintptr_t arg5);
+    uintptr_t arg2, uintptr_t arg3, uintptr_t arg4, uintptr_t arg5,
+    uintptr_t arg6, uintptr_t arg7);
 
 #ifdef _SYSCALL32_IMPL
 extern int64_t dtrace_systrace_syscall32(uintptr_t arg0, uintptr_t arg1,
-    uintptr_t arg2, uintptr_t arg3, uintptr_t arg4, uintptr_t arg5);
+    uintptr_t arg2, uintptr_t arg3, uintptr_t arg4, uintptr_t arg5,
+    uintptr_t arg6, uintptr_t arg7);
 #endif
 
 #endif
diff --git a/usr/src/uts/common/sys/termios.h b/usr/src/uts/common/sys/termios.h
index 09be20858d..889a7096cd 100644
--- a/usr/src/uts/common/sys/termios.h
+++ b/usr/src/uts/common/sys/termios.h
@@ -361,6 +361,24 @@ extern pid_t tcgetsid(int);
 #define	TCSETSF		(_TIOC|16)
 
 /*
+ * linux terminal ioctls we need to be aware of
+ */
+#define	TIOCSETLD	(_TIOC|123)	/* set line discipline parms */
+#define	TIOCGETLD	(_TIOC|124)	/* get line discipline parms */
+
+/*
+ * The VMIN and VTIME and solaris overlap with VEOF and VEOL - This is
+ * perfectly legal except, linux expects them to be separate. So we keep
+ * them separately.
+ */
+struct lx_cc {
+	unsigned char veof;	/* veof value */
+	unsigned char veol;	/* veol value */
+	unsigned char vmin;	/* vmin value */
+	unsigned char vtime;	/* vtime value */
+};
+
+/*
  * NTP PPS ioctls
  */
 #define	TIOCGPPS	(_TIOC|125)
diff --git a/usr/src/uts/common/sys/thread.h b/usr/src/uts/common/sys/thread.h
index 188230d61e..41ea2331df 100644
--- a/usr/src/uts/common/sys/thread.h
+++ b/usr/src/uts/common/sys/thread.h
@@ -24,6 +24,10 @@
  * Use is subject to license terms.
  */
 
+/*
+ * Copyright (c) 2015, Joyent, Inc.  All rights reserved.
+ */
+
 #ifndef	_SYS_THREAD_H
 #define	_SYS_THREAD_H
 
@@ -68,6 +72,8 @@ typedef struct ctxop {
 	void	(*free_op)(void *, int); /* function which frees the context */
 	void	*arg;		/* argument to above functions, ctx pointer */
 	struct ctxop *next;	/* next context ops */
+	hrtime_t save_ts;		/* timestamp of last save */
+	hrtime_t restore_ts;		/* timestamp of last restore */
 } ctxop_t;
 
 /*
@@ -365,7 +371,7 @@ typedef struct _kthread {
 #define	T_WOULDBLOCK	0x0020	/* for lockfs */
 #define	T_DONTBLOCK	0x0040	/* for lockfs */
 #define	T_DONTPEND	0x0080	/* for lockfs */
-#define	T_SYS_PROF	0x0100	/* profiling on for duration of system call */
+#define	T_SPLITSTK	0x0100	/* kernel stack is currently split */
 #define	T_WAITCVSEM	0x0200	/* waiting for a lwp_cv or lwp_sema on sleepq */
 #define	T_WATCHPT	0x0400	/* thread undergoing a watchpoint emulation */
 #define	T_PANIC		0x0800	/* thread initiated a system panic */
@@ -413,8 +419,9 @@ typedef struct _kthread {
 #define	TS_RESUME	0x1000	/* setrun() by CPR resume process */
 #define	TS_CREATE	0x2000	/* setrun() by syslwp_create() */
 #define	TS_RUNQMATCH	0x4000	/* exact run queue balancing by setbackdq() */
+#define	TS_BSTART	0x8000	/* setrun() by brand */
 #define	TS_ALLSTART	\
-	(TS_CSTART|TS_UNPAUSE|TS_XSTART|TS_PSTART|TS_RESUME|TS_CREATE)
+	(TS_CSTART|TS_UNPAUSE|TS_XSTART|TS_PSTART|TS_RESUME|TS_CREATE|TS_BSTART)
 #define	TS_ANYWAITQ	(TS_PROJWAITQ|TS_ZONEWAITQ)
 
 /*
@@ -442,6 +449,10 @@ typedef struct _kthread {
 #define	ISTOPPED(t) ((t)->t_state == TS_STOPPED && \
 			!((t)->t_schedflag & TS_PSTART))
 
+/* True if thread is stopped for a brand-specific reason */
+#define	BSTOPPED(t)	((t)->t_state == TS_STOPPED && \
+			    !((t)->t_schedflag & TS_BSTART))
+
 /* True if thread is asleep and wakeable */
 #define	ISWAKEABLE(t) (((t)->t_state == TS_SLEEP && \
 			((t)->t_flag & T_WAKEABLE)))
diff --git a/usr/src/uts/common/sys/timer.h b/usr/src/uts/common/sys/timer.h
index 604ddf5d83..ec349c962f 100644
--- a/usr/src/uts/common/sys/timer.h
+++ b/usr/src/uts/common/sys/timer.h
@@ -24,6 +24,10 @@
  * Use is subject to license terms.
  */
 
+/*
+ * Copyright (c) 2015, Joyent, Inc. All rights reserved.
+ */
+
 #ifndef	_SYS_TIMER_H
 #define	_SYS_TIMER_H
 
@@ -55,32 +59,45 @@ extern	int	timer_max;		/* patchable via /etc/system */
 
 struct clock_backend;
 
-typedef struct itimer {
+struct itimer;
+typedef struct itimer itimer_t;
+
+struct itimer {
 	itimerspec_t	it_itime;
 	hrtime_t	it_hrtime;
 	ushort_t	it_flags;
 	ushort_t	it_lock;
-	void		*it_arg;
-	sigqueue_t	*it_sigq;
-	klwp_t		*it_lwp;
+	void		*it_arg;	/* clock backend-specific data */
 	struct proc	*it_proc;
+	union {
+		struct {
+			sigqueue_t	*__it_sigq;
+			klwp_t		*__it_lwp;
+		} __proc;
+		void *__it_frontend;
+	} __data;			/* timer frontend-specific data */
 	kcondvar_t	it_cv;
 	int		it_blockers;
 	int		it_pending;
 	int		it_overrun;
 	struct clock_backend *it_backend;
+	void		(*it_fire)(itimer_t *);
 	kmutex_t	it_mutex;
 	void		*it_portev;	/* port_kevent_t pointer */
 	void		*it_portsrc;	/* port_source_t pointer */
 	int		it_portfd;	/* port file descriptor */
-} itimer_t;
+};
+
+#define	it_sigq		__data.__proc.__it_sigq
+#define	it_lwp		__data.__proc.__it_lwp
+#define	it_frontend	__data.__it_frontend
 
 typedef struct clock_backend {
 	struct sigevent clk_default;
 	int (*clk_clock_settime)(timespec_t *);
 	int (*clk_clock_gettime)(timespec_t *);
 	int (*clk_clock_getres)(timespec_t *);
-	int (*clk_timer_create)(itimer_t *, struct sigevent *);
+	int (*clk_timer_create)(itimer_t *, void (*)(itimer_t *));
 	int (*clk_timer_settime)(itimer_t *, int, const struct itimerspec *);
 	int (*clk_timer_gettime)(itimer_t *, struct itimerspec *);
 	int (*clk_timer_delete)(itimer_t *);
@@ -88,8 +105,8 @@ typedef struct clock_backend {
 } clock_backend_t;
 
 extern void clock_add_backend(clockid_t clock, clock_backend_t *backend);
+extern clock_backend_t *clock_get_backend(clockid_t clock);
 
-extern void timer_fire(itimer_t *);
 extern void timer_lwpbind();
 
 extern	void	timer_func(sigqueue_t *);
diff --git a/usr/src/uts/common/sys/timerfd.h b/usr/src/uts/common/sys/timerfd.h
new file mode 100644
index 0000000000..66cb50ac88
--- /dev/null
+++ b/usr/src/uts/common/sys/timerfd.h
@@ -0,0 +1,81 @@
+/*
+ * This file and its contents are supplied under the terms of the
+ * Common Development and Distribution License ("CDDL"), version 1.0.
+ * You may only use this file in accordance with the terms of version
+ * 1.0 of the CDDL.
+ *
+ * A full copy of the text of the CDDL should have accompanied this
+ * source.  A copy of the CDDL is also available via the Internet at
+ * http://www.illumos.org/license/CDDL.
+ */
+
+/*
+ * Copyright (c) 2015 Joyent, Inc.  All rights reserved.
+ */
+
+/*
+ * Header file to support for the timerfd facility.
+ */
+
+#ifndef _SYS_TIMERFD_H
+#define	_SYS_TIMERFD_H
+
+#include <sys/types.h>
+#include <sys/time_impl.h>
+
+#ifdef	__cplusplus
+extern "C" {
+#endif
+
+/*
+ * To assure binary compatibility with Linux, these values are fixed at their
+ * Linux equivalents, not their native ones.
+ */
+#define	TFD_CLOEXEC		02000000		/* LX_O_CLOEXEC */
+#define	TFD_NONBLOCK		04000			/* LX_O_NONBLOCK */
+#define	TFD_TIMER_ABSTIME	(1 << 0)
+#define	TFD_TIMER_CANCEL_ON_SET	(1 << 1)
+
+/*
+ * These ioctl values are specific to the native implementation; applications
+ * shouldn't be using them directly, and they should therefore be safe to
+ * change without breaking apps.
+ */
+#define	TIMERFDIOC		(('t' << 24) | ('f' << 16) | ('d' << 8))
+#define	TIMERFDIOC_CREATE	(TIMERFDIOC | 1)	/* create timer */
+#define	TIMERFDIOC_SETTIME	(TIMERFDIOC | 2)	/* timerfd_settime() */
+#define	TIMERFDIOC_GETTIME	(TIMERFDIOC | 3)	/* timerfd_gettime() */
+
+typedef struct timerfd_settime {
+	uint64_t tfd_settime_flags;	/* flags (e.g., TFD_TIMER_ABSTIME) */
+	uint64_t tfd_settime_value;	/* pointer to value */
+	uint64_t tfd_settime_ovalue;	/* pointer to old value, if any */
+} timerfd_settime_t;
+
+#ifndef _KERNEL
+
+extern int timerfd_create(int, int);
+extern int timerfd_settime(int, int,
+    const struct itimerspec *, struct itimerspec *);
+extern int timerfd_gettime(int, struct itimerspec *);
+
+#else
+
+#define	TIMERFDMNRN_TIMERFD	0
+#define	TIMERFDMNRN_CLONE	1
+#define	TIMERFD_VALMAX		(ULLONG_MAX - 1ULL)
+
+/*
+ * Fortunately, the values for the Linux clocks that are valid for timerfd
+ * (namely, CLOCK_REALTIME and CLOCK_MONOTONIC) don't overlap with our values
+ * the same.
+ */
+#define	TIMERFD_MONOTONIC	1	/* Linux value for CLOCK_MONOTONIC */
+
+#endif /* _KERNEL */
+
+#ifdef	__cplusplus
+}
+#endif
+
+#endif	/* _SYS_TIMERFD_H */
diff --git a/usr/src/uts/common/sys/uadmin.h b/usr/src/uts/common/sys/uadmin.h
index d5168c9b2c..c14a3bf11e 100644
--- a/usr/src/uts/common/sys/uadmin.h
+++ b/usr/src/uts/common/sys/uadmin.h
@@ -23,6 +23,7 @@
  *
  * Copyright 2009 Sun Microsystems, Inc.  All rights reserved.
  * Use is subject to license terms.
+ * Copyright 2011 Joyent, Inc.  All rights reserved.
  */
 
 /*	Copyright (c) 1984, 1986, 1987, 1988, 1989 AT&T	*/
@@ -159,7 +160,7 @@ extern kmutex_t ualock;
 extern void mdboot(int, int, char *, boolean_t);
 extern void mdpreboot(int, int, char *);
 extern int kadmin(int, int, void *, cred_t *);
-extern void killall(zoneid_t);
+extern void killall(zoneid_t, boolean_t);
 #endif
 
 extern int uadmin(int, int, uintptr_t);
diff --git a/usr/src/uts/common/sys/vm_usage.h b/usr/src/uts/common/sys/vm_usage.h
index 1aa4a8ee6d..c2954cbc29 100644
--- a/usr/src/uts/common/sys/vm_usage.h
+++ b/usr/src/uts/common/sys/vm_usage.h
@@ -21,6 +21,7 @@
 /*
  * Copyright 2008 Sun Microsystems, Inc.  All rights reserved.
  * Use is subject to license terms.
+ * Copyright 2014 Joyent, Inc.  All rights reserved.
  */
 
 #ifndef	_SYS_VM_USAGE_H
@@ -79,8 +80,9 @@ extern "C" {
 					/* zoneid */
 #define	VMUSAGE_COL_EUSERS	0x2000	/* same as VMUSAGE_COL_RUSERS, but by */
 					/* euser */
+#define	VMUSAGE_A_ZONE		0x4000	/* rss/swap for a specified zone */
 
-#define	VMUSAGE_MASK		0x3fff  /* all valid flags for getvmusage() */
+#define	VMUSAGE_MASK		0x7fff  /* all valid flags for getvmusage() */
 
 typedef struct vmusage {
 	id_t	vmu_zoneid;		/* zoneid, or ALL_ZONES for */
@@ -108,6 +110,7 @@ extern int getvmusage(uint_t flags, time_t age, vmusage_t *buf, size_t *nres);
 
 int vm_getusage(uint_t, time_t, vmusage_t *, size_t *, int);
 void vm_usage_init();
+int vm_map_inval(pid_t, caddr_t, size_t);
 
 #endif	/* _KERNEL */
 
diff --git a/usr/src/uts/common/sys/vmsystm.h b/usr/src/uts/common/sys/vmsystm.h
index 6122b6cd2f..c7b41730b6 100644
--- a/usr/src/uts/common/sys/vmsystm.h
+++ b/usr/src/uts/common/sys/vmsystm.h
@@ -19,6 +19,9 @@
  * CDDL HEADER END
  */
 /*
+ * Copyright (c) 2014, Joyent, Inc. All rights reserved.
+ */
+/*
  * Copyright 2009 Sun Microsystems, Inc.  All rights reserved.
  * Use is subject to license terms.
  */
@@ -159,6 +162,8 @@ extern	void	*boot_virt_alloc(void *addr, size_t size);
 
 extern	size_t	exec_get_spslew(void);
 
+extern	caddr_t	map_userlimit(proc_t *pp, struct as *as, int flags);
+
 #endif	/* _KERNEL */
 
 #ifdef	__cplusplus
diff --git a/usr/src/uts/common/sys/vnd.h b/usr/src/uts/common/sys/vnd.h
new file mode 100644
index 0000000000..bc7c9c3122
--- /dev/null
+++ b/usr/src/uts/common/sys/vnd.h
@@ -0,0 +1,141 @@
+/*
+ * This file and its contents are supplied under the terms of the
+ * Common Development and Distribution License ("CDDL"), version 1.0.
+ * You may only use this file in accordance with the terms of version
+ * 1.0 of the CDDL.
+ *
+ * A full copy of the text of the CDDL should have accompanied this
+ * source.  A copy of the CDDL is also available via the Internet at
+ * http://www.illumos.org/license/CDDL.
+ */
+
+/*
+ * Copyright (c) 2014 Joyent, Inc. All rights reserved.
+ */
+
+#ifndef _SYS_VND_H
+#define	_SYS_VND_H
+
+#include <sys/types.h>
+#include <sys/vnd_errno.h>
+#include <sys/frameio.h>
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+/*
+ * We distinguish between normal ioctls and private ioctls we issues to out
+ * streams version. Streams ioctls have the upper bit set in the lowest byte.
+ * Note that there are no STREAMs ioctls for userland and all definitions
+ * related to them are not present in this file.
+ */
+#define	VND_IOC		(('v' << 24) | ('n' << 16) | ('d' << 8))
+
+/*
+ * Attach the current minor instance to a given dlpi datalink identified by a
+ * vnd_ioc_name_t argument. This fails if it's already been attached. Note that
+ * unlike the other ioctls, this is passed directly as opposed to every other
+ * function which is passed as a pointer to the value.
+ */
+#define	VND_IOC_ATTACH		(VND_IOC | 0x1)
+
+#define	VND_NAMELEN	32
+
+typedef struct vnd_ioc_attach {
+	char		via_name[VND_NAMELEN];
+	zoneid_t	via_zoneid;
+	uint32_t	via_errno;
+} vnd_ioc_attach_t;
+
+/*
+ * Link the current minor instance into the /devices name space.
+ *
+ * This ioctl adds entries into /devices with a name of the form z%d:%s vil_zid,
+ * vil_name. The device will be namespaced to the zone. The global zone will be
+ * able to see all minor nodes. In the zone, only the /dev entries will exist.
+ * At this time, a given device can only have one link at a time. Note that a
+ * user cannot specify the zone to pass in, rather it is the zone that the
+ * device was attached in.
+ */
+#define	VND_IOC_LINK		(VND_IOC | 0x2)
+
+typedef struct vnd_ioc_link {
+	char		vil_name[VND_NAMELEN];
+	uint32_t	vil_errno;
+} vnd_ioc_link_t;
+
+/*
+ * Unlink the opened minor instance from the /devices name space. A zone may use
+ * this to unlink an extent entry in /dev; however, they will not be able to
+ * link it in again.
+ */
+#define	VND_IOC_UNLINK		(VND_IOC | 0x3)
+typedef struct vnd_ioc_unlink {
+	uint32_t viu_errno;
+} vnd_ioc_unlink_t;
+
+/*
+ * Controls to get and set the current buffer recieve buffer size.
+ */
+typedef struct vnd_ioc_buf {
+	uint64_t	vib_size;
+	uint32_t	vib_filler;
+	uint32_t	vib_errno;
+} vnd_ioc_buf_t;
+
+#define	VND_IOC_GETRXBUF	(VND_IOC | 0x04)
+#define	VND_IOC_SETRXBUF	(VND_IOC | 0x05)
+#define	VND_IOC_GETMAXBUF	(VND_IOC | 0x06)
+#define	VND_IOC_GETTXBUF	(VND_IOC | 0x07)
+#define	VND_IOC_SETTXBUF	(VND_IOC | 0x08)
+#define	VND_IOC_GETMINTU	(VND_IOC | 0x09)
+#define	VND_IOC_GETMAXTU	(VND_IOC | 0x0a)
+
+/*
+ * Information and listing ioctls
+ *
+ * This gets information about all of the active vnd instances. vl_actents is
+ * always updated to the number around and vl_nents is the number of
+ * vnd_ioc_info_t elements are allocated in vl_ents.
+ */
+typedef struct vnd_ioc_info {
+	uint32_t vii_version;
+	zoneid_t vii_zone;
+	char vii_name[VND_NAMELEN];
+	char vii_datalink[VND_NAMELEN];
+} vnd_ioc_info_t;
+
+typedef struct vnd_ioc_list {
+	uint_t vl_nents;
+	uint_t vl_actents;
+	vnd_ioc_info_t *vl_ents;
+} vnd_ioc_list_t;
+
+#ifdef _KERNEL
+
+typedef struct vnd_ioc_list32 {
+	uint_t vl_nents;
+	uint_t vl_actents;
+	caddr32_t vl_ents;
+} vnd_ioc_list32_t;
+
+#endif	/* _KERNEL */
+
+#define	VND_IOC_LIST		(VND_IOC | 0x20)
+
+/*
+ * Framed I/O ioctls
+ *
+ * Users should use the standard frameio_t as opposed to a vnd specific type.
+ * This is a consolidation private ioctl pending futher stability in the form of
+ * specific system work.
+ */
+#define	VND_IOC_FRAMEIO_READ	(VND_IOC | 0x30)
+#define	VND_IOC_FRAMEIO_WRITE	(VND_IOC | 0x31)
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif /* _SYS_VND_H */
diff --git a/usr/src/uts/common/sys/vnd_errno.h b/usr/src/uts/common/sys/vnd_errno.h
new file mode 100644
index 0000000000..89e5fc2543
--- /dev/null
+++ b/usr/src/uts/common/sys/vnd_errno.h
@@ -0,0 +1,72 @@
+/*
+ * This file and its contents are supplied under the terms of the
+ * Common Development and Distribution License ("CDDL"), version 1.0.
+ * You may only use this file in accordance with the terms of version
+ * 1.0 of the CDDL.
+ *
+ * A full copy of the text of the CDDL should have accompanied this
+ * source.  A copy of the CDDL is also available via the Internet at
+ * http://www.illumos.org/license/CDDL.
+ */
+
+/*
+ * Copyright (c) 2014 Joyent, Inc.  All rights reserved.
+ */
+
+#ifndef _SYS_VND_ERRNO_H
+#define	_SYS_VND_ERRNO_H
+
+/*
+ * This header contains all of the available vnd errors.
+ */
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+typedef enum vnd_errno {
+	VND_E_SUCCESS = 0,		/* no error */
+	VND_E_NOMEM,			/* no memory */
+	VND_E_NODATALINK,		/* no such datalink */
+	VND_E_NOTETHER,			/* not DL_ETHER */
+	VND_E_DLPIINVAL,		/* Unknown DLPI failures */
+	VND_E_ATTACHFAIL,		/* DL_ATTACH_REQ failed */
+	VND_E_BINDFAIL,			/* DL_BIND_REQ failed */
+	VND_E_PROMISCFAIL,		/* DL_PROMISCON_REQ failed */
+	VND_E_DIRECTFAIL,		/* DLD_CAPAB_DIRECT enable failed */
+	VND_E_CAPACKINVAL,		/* bad dl_capability_ack_t */
+	VND_E_SUBCAPINVAL,		/* bad dl_capability_sub_t */
+	VND_E_DLDBADVERS,		/* bad dld version */
+	VND_E_KSTATCREATE,		/* failed to create kstats */
+	VND_E_NODEV,			/* no such vnd link */
+	VND_E_NONETSTACK,		/* netstack doesn't exist */
+	VND_E_ASSOCIATED,		/* device already associated */
+	VND_E_ATTACHED,			/* device already attached */
+	VND_E_LINKED,			/* device already linked */
+	VND_E_BADNAME,			/* invalid name */
+	VND_E_PERM,			/* can't touch this */
+	VND_E_NOZONE,			/* no such zone */
+	VND_E_STRINIT,		/* failed to initialize vnd stream module */
+	VND_E_NOTATTACHED,		/* device not attached */
+	VND_E_NOTLINKED,		/* device not linked */
+	VND_E_LINKEXISTS,	/* another device has the same link name */
+	VND_E_MINORNODE,		/* failed to create minor node */
+	VND_E_BUFTOOBIG,		/* requested buffer size is too large */
+	VND_E_BUFTOOSMALL,		/* requested buffer size is too small */
+	VND_E_DLEXCL,			/* unable to get dlpi excl access */
+	VND_E_DIRECTNOTSUP,
+			/* DLD direct capability not suported over data link */
+	VND_E_BADPROPSIZE,		/* invalid property size */
+	VND_E_BADPROP,			/* invalid property */
+	VND_E_PROPRDONLY,		/* property is read only */
+	VND_E_SYS,			/* unexpected system error */
+	VND_E_CAPABPASS,
+			/* capabilities invalid, pass-through module detected */
+	VND_E_UNKNOWN			/* unknown error */
+} vnd_errno_t;
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif /* _SYS_VND_ERRNO_H */
diff --git a/usr/src/uts/common/sys/vnode.h b/usr/src/uts/common/sys/vnode.h
index af9516fe52..28145636f3 100644
--- a/usr/src/uts/common/sys/vnode.h
+++ b/usr/src/uts/common/sys/vnode.h
@@ -21,7 +21,7 @@
 
 /*
  * Copyright (c) 1988, 2010, Oracle and/or its affiliates. All rights reserved.
- * Copyright (c) 2013, Joyent, Inc. All rights reserved.
+ * Copyright (c) 2015, Joyent, Inc.
  */
 
 /*	Copyright (c) 1983, 1984, 1985, 1986, 1987, 1988, 1989 AT&T	*/
@@ -735,7 +735,8 @@ typedef enum vnevent	{
 	VE_LINK		= 6, 	/* Link with vnode's name as source */
 	VE_RENAME_DEST_DIR	= 7, 	/* Rename with vnode as target dir */
 	VE_MOUNTEDOVER	= 8, 	/* File or Filesystem got mounted over vnode */
-	VE_TRUNCATE = 9		/* Truncate */
+	VE_TRUNCATE = 9,	/* Truncate */
+	VE_RENAME_SRC_DIR = 10	/* Rename with vnode as source dir */
 } vnevent_t;
 
 /*
@@ -1290,7 +1291,8 @@ void	vnevent_remove(vnode_t *, vnode_t *, char *, caller_context_t *);
 void	vnevent_rmdir(vnode_t *, vnode_t *, char *, caller_context_t *);
 void	vnevent_create(vnode_t *, caller_context_t *);
 void	vnevent_link(vnode_t *, caller_context_t *);
-void	vnevent_rename_dest_dir(vnode_t *, caller_context_t *ct);
+void	vnevent_rename_dest_dir(vnode_t *, vnode_t *, char *,
+    caller_context_t *ct);
 void	vnevent_mountedover(vnode_t *, caller_context_t *);
 void	vnevent_truncate(vnode_t *, caller_context_t *);
 int	vnevent_support(vnode_t *, caller_context_t *);
@@ -1323,6 +1325,9 @@ u_longlong_t	fs_new_caller_id();
 
 int	vn_vmpss_usepageio(vnode_t *);
 
+/* Empty v_path placeholder */
+extern char *vn_vpath_empty;
+
 /*
  * Needed for use of IS_VMODSORT() in kernel.
  */
diff --git a/usr/src/uts/common/sys/vxlan.h b/usr/src/uts/common/sys/vxlan.h
new file mode 100644
index 0000000000..d87786b507
--- /dev/null
+++ b/usr/src/uts/common/sys/vxlan.h
@@ -0,0 +1,47 @@
+/*
+ * This file and its contents are supplied under the terms of the
+ * Common Development and Distribution License ("CDDL"), version 1.0.
+ * You may only use this file in accordance with the terms of version
+ * 1.0 of the CDDL.
+ *
+ * A full copy of the text of the CDDL should have accompanied this
+ * source.  A copy of the CDDL is also available via the Internet at
+ * http://www.illumos.org/license/CDDL.
+ */
+
+/*
+ * Copyright 2015 Joyent, Inc.
+ */
+
+#ifndef _SYS_VXLAN_H
+#define	_SYS_VXLAN_H
+
+/*
+ * Common VXLAN information
+ */
+
+#include <sys/inttypes.h>
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+/* Sizes in bytes */
+#define	VXLAN_HDR_LEN	8
+#define	VXLAN_ID_LEN	3
+
+#define	VXLAN_F_VDI	0x08000000
+#define	VXLAN_ID_SHIFT	8
+
+#pragma pack(1)
+typedef struct vxlan_hdr {
+	uint32_t vxlan_flags;
+	uint32_t vxlan_id;
+} vxlan_hdr_t;
+#pragma pack()
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif /* _SYS_VXLAN_H */
diff --git a/usr/src/uts/common/sys/zfd.h b/usr/src/uts/common/sys/zfd.h
new file mode 100644
index 0000000000..acf9ce8d33
--- /dev/null
+++ b/usr/src/uts/common/sys/zfd.h
@@ -0,0 +1,63 @@
+/*
+ * This file and its contents are supplied under the terms of the
+ * Common Development and Distribution License ("CDDL"), version 1.0.
+ * You may only use this file in accordance with the terms of version
+ * 1.0 of the CDDL.
+ *
+ * A full copy of the text of the CDDL should have accompanied this
+ * source.  A copy of the CDDL is also available via the Internet at
+ * http://www.illumos.org/license/CDDL.
+ */
+/*
+ * Copyright 2015 Joyent, Inc.
+ */
+
+#ifndef _SYS_ZFD_H
+#define	_SYS_ZFD_H
+
+#include <sys/types.h>
+
+#ifdef	__cplusplus
+extern "C" {
+#endif
+
+/*
+ * Minor node name of the global zone side (often called the "master" side)
+ * of the zfd dev.
+ */
+#define	ZFD_MASTER_NAME	"master"
+
+/*
+ * Minor node name of the non-global zone side (often called the "slave"
+ * side) of the zfd dev.
+ */
+#define	ZFD_SLAVE_NAME	"slave"
+
+#define	ZFD_NAME_LEN	16
+
+/*
+ * ZFD_IOC forms the base for all zfd ioctls.
+ */
+#define	ZFD_IOC		(('Z' << 24) | ('f' << 16) | ('d' << 8))
+
+/*
+ * This ioctl tells the slave side it should push the TTY stream modules
+ * so that the fd looks like a tty.
+ */
+#define	ZFD_MAKETTY		(ZFD_IOC | 0)
+
+/*
+ * This ioctl puts a hangup into the stream so that the slave side sees EOF.
+ */
+#define	ZFD_EOF			(ZFD_IOC | 1)
+
+/*
+ * This ioctl succeeds if the slave side is open.
+ */
+#define	ZFD_HAS_SLAVE		(ZFD_IOC | 2)
+
+#ifdef	__cplusplus
+}
+#endif
+
+#endif	/* _SYS_ZFD_H */
diff --git a/usr/src/uts/common/sys/zone.h b/usr/src/uts/common/sys/zone.h
index 7aab7c8d32..a5d1610842 100644
--- a/usr/src/uts/common/sys/zone.h
+++ b/usr/src/uts/common/sys/zone.h
@@ -23,6 +23,7 @@
  * Copyright 2013, Joyent, Inc. All rights reserved.
  * Copyright 2014 Nexenta Systems, Inc. All rights reserved.
  * Copyright 2014 Igor Kozhukhov <ikozhukhov@gmail.com>.
+ * Copyright 2015, Joyent, Inc. All rights reserved.
  */
 
 #ifndef _SYS_ZONE_H
@@ -97,13 +98,17 @@ extern "C" {
 #define	ZONE_ATTR_INITNAME	9
 #define	ZONE_ATTR_BOOTARGS	10
 #define	ZONE_ATTR_BRAND		11
-#define	ZONE_ATTR_PHYS_MCAP	12
+#define	ZONE_ATTR_PMCAP_NOVER	12
 #define	ZONE_ATTR_SCHED_CLASS	13
 #define	ZONE_ATTR_FLAGS		14
 #define	ZONE_ATTR_HOSTID	15
 #define	ZONE_ATTR_FS_ALLOWED	16
 #define	ZONE_ATTR_NETWORK	17
+#define	ZONE_ATTR_DID		18
+#define	ZONE_ATTR_PMCAP_PAGEOUT	19
 #define	ZONE_ATTR_INITNORESTART	20
+#define	ZONE_ATTR_PG_FLT_DELAY	21
+#define	ZONE_ATTR_RSS		22
 
 /* Start of the brand-specific attribute namespace */
 #define	ZONE_ATTR_BRAND_ATTRS	32768
@@ -184,6 +189,7 @@ typedef struct {
 	uint32_t doi;			/* DOI for label */
 	caddr32_t label;		/* label associated with zone */
 	int flags;
+	zoneid_t zoneid;		/* requested zoneid */
 } zone_def32;
 #endif
 typedef struct {
@@ -200,6 +206,7 @@ typedef struct {
 	uint32_t doi;			/* DOI for label */
 	const bslabel_t *label;		/* label associated with zone */
 	int flags;
+	zoneid_t zoneid;		/* requested zoneid */
 } zone_def;
 
 /* extended error information */
@@ -244,9 +251,12 @@ typedef enum zone_cmd {
 typedef struct zone_cmd_arg {
 	uint64_t	uniqid;		/* unique "generation number" */
 	zone_cmd_t	cmd;		/* requested action */
-	uint32_t	_pad;		/* need consistent 32/64 bit alignmt */
+	int		status;		/* init status on shutdown */
+	uint32_t	debug;		/* enable brand hook debug */
 	char locale[MAXPATHLEN];	/* locale in which to render messages */
 	char bootbuf[BOOTARGS_MAX];	/* arguments passed to zone_boot() */
+	/* Needed for 32/64 zoneadm -> zoneadmd door arg size check. */
+	int		pad;
 } zone_cmd_arg_t;
 
 /*
@@ -372,7 +382,7 @@ typedef struct zone_dataset {
 } zone_dataset_t;
 
 /*
- * structure for zone kstats
+ * structure for rctl zone kstats
  */
 typedef struct zone_kstat {
 	kstat_named_t zk_zonename;
@@ -383,6 +393,58 @@ typedef struct zone_kstat {
 struct cpucap;
 
 typedef struct {
+	hrtime_t	cycle_start;
+	uint_t		cycle_cnt;
+	hrtime_t	zone_avg_cnt;
+} sys_zio_cntr_t;
+
+typedef struct {
+	kstat_named_t	zv_zonename;
+	kstat_named_t	zv_nread;
+	kstat_named_t	zv_reads;
+	kstat_named_t	zv_rtime;
+	kstat_named_t	zv_rlentime;
+	kstat_named_t	zv_nwritten;
+	kstat_named_t	zv_writes;
+	kstat_named_t	zv_wtime;
+	kstat_named_t	zv_wlentime;
+	kstat_named_t	zv_10ms_ops;
+	kstat_named_t	zv_100ms_ops;
+	kstat_named_t	zv_1s_ops;
+	kstat_named_t	zv_10s_ops;
+	kstat_named_t 	zv_delay_cnt;
+	kstat_named_t	zv_delay_time;
+} zone_vfs_kstat_t;
+
+typedef struct {
+	kstat_named_t	zz_zonename;
+	kstat_named_t	zz_nread;
+	kstat_named_t	zz_reads;
+	kstat_named_t	zz_rtime;
+	kstat_named_t	zz_rlentime;
+	kstat_named_t	zz_nwritten;
+	kstat_named_t	zz_writes;
+	kstat_named_t	zz_waittime;
+} zone_zfs_kstat_t;
+
+typedef struct {
+	kstat_named_t	zm_zonename;
+	kstat_named_t	zm_rss;
+	kstat_named_t	zm_phys_cap;
+	kstat_named_t	zm_swap;
+	kstat_named_t	zm_swap_cap;
+	kstat_named_t	zm_nover;
+	kstat_named_t	zm_pagedout;
+	kstat_named_t	zm_pgpgin;
+	kstat_named_t	zm_anonpgin;
+	kstat_named_t	zm_execpgin;
+	kstat_named_t	zm_fspgin;
+	kstat_named_t	zm_anon_alloc_fail;
+	kstat_named_t	zm_pf_throttle;
+	kstat_named_t	zm_pf_throttle_usec;
+} zone_mcap_kstat_t;
+
+typedef struct {
 	kstat_named_t	zm_zonename;	/* full name, kstat truncates name */
 	kstat_named_t	zm_utime;
 	kstat_named_t	zm_stime;
@@ -390,15 +452,12 @@ typedef struct {
 	kstat_named_t	zm_avenrun1;
 	kstat_named_t	zm_avenrun5;
 	kstat_named_t	zm_avenrun15;
-	kstat_named_t	zm_run_ticks;
-	kstat_named_t	zm_run_wait;
-	kstat_named_t	zm_fss_shr_pct;
-	kstat_named_t	zm_fss_pri_hi;
-	kstat_named_t	zm_fss_pri_avg;
 	kstat_named_t	zm_ffcap;
 	kstat_named_t	zm_ffnoproc;
 	kstat_named_t	zm_ffnomem;
 	kstat_named_t	zm_ffmisc;
+	kstat_named_t	zm_init_pid;
+	kstat_named_t	zm_boot_time;
 } zone_misc_kstat_t;
 
 typedef struct zone {
@@ -440,6 +499,7 @@ typedef struct zone {
 	 */
 	list_node_t	zone_linkage;
 	zoneid_t	zone_id;	/* ID of zone */
+	zoneid_t	zone_did;	/* persistent debug ID of zone */
 	uint_t		zone_ref;	/* count of zone_hold()s on zone */
 	uint_t		zone_cred_ref;	/* count of zone_hold_cred()s on zone */
 	/*
@@ -492,10 +552,11 @@ typedef struct zone {
 	kcondvar_t	zone_cv;	/* used to signal state changes */
 	struct proc	*zone_zsched;	/* Dummy kernel "zsched" process */
 	pid_t		zone_proc_initpid; /* pid of "init" for this zone */
-	char		*zone_initname;	/* fs path to 'init' */
+	char		*zone_initname;		/* fs path to 'init' */
+	int		zone_init_status;	/* init's exit status */
 	int		zone_boot_err;  /* for zone_boot() if boot fails */
 	char		*zone_bootargs;	/* arguments passed via zone_boot() */
-	uint64_t	zone_phys_mcap;	/* physical memory cap */
+	rctl_qty_t	zone_phys_mem_ctl;	/* current phys. memory limit */
 	/*
 	 * zone_kthreads is protected by zone_status_lock.
 	 */
@@ -533,6 +594,7 @@ typedef struct zone {
 	tsol_mlp_list_t zone_mlps;	/* MLPs on zone-private addresses */
 
 	boolean_t	zone_restart_init;	/* Restart init if it dies? */
+	boolean_t	zone_reboot_on_init_exit; /* Reboot if init dies? */
 	struct brand	*zone_brand;		/* zone's brand */
 	void 		*zone_brand_data;	/* store brand specific data */
 	id_t		zone_defaultcid;	/* dflt scheduling class id */
@@ -544,6 +606,37 @@ typedef struct zone {
 	list_t		zone_dl_list;
 	netstack_t	*zone_netstack;
 	struct cpucap	*zone_cpucap;	/* CPU caps data */
+
+	/*
+	 * Data and counters used for ZFS fair-share disk IO.
+	 */
+	rctl_qty_t	zone_zfs_io_pri;	/* ZFS IO priority */
+	uint_t		zone_zfs_queued[2];	/* sync I/O enqueued count */
+	uint64_t	zone_zfs_weight;	/* used to prevent starvation */
+	uint64_t	zone_io_util;		/* IO utilization metric */
+	boolean_t	zone_io_util_above_avg;	/* IO util percent > avg. */
+	uint16_t	zone_io_delay;		/* IO delay on logical r/w */
+	kmutex_t	zone_stg_io_lock;	/* protects IO window data */
+	sys_zio_cntr_t	zone_rd_ops;		/* Counters for ZFS reads, */
+	sys_zio_cntr_t	zone_wr_ops;		/* writes and */
+	sys_zio_cntr_t	zone_lwr_ops;		/* logical writes. */
+
+	/*
+	 * kstats and counters for VFS ops and bytes.
+	 */
+	kmutex_t	zone_vfs_lock;		/* protects VFS statistics */
+	kstat_t		*zone_vfs_ksp;
+	kstat_io_t	zone_vfs_rwstats;
+	zone_vfs_kstat_t *zone_vfs_stats;
+
+	/*
+	 * kstats for ZFS I/O ops and bytes.
+	 */
+	kmutex_t	zone_zfs_lock;		/* protects ZFS statistics */
+	kstat_t		*zone_zfs_ksp;
+	kstat_io_t	zone_zfs_rwstats;
+	zone_zfs_kstat_t *zone_zfs_stats;
+
 	/*
 	 * Solaris Auditing per-zone audit context
 	 */
@@ -563,6 +656,27 @@ typedef struct zone {
 	kstat_t		*zone_nprocs_kstat;
 
 	/*
+	 * kstats and counters for physical memory capping.
+	 */
+	rctl_qty_t	zone_phys_mem;	/* current bytes of phys. mem. (RSS) */
+	kstat_t		*zone_physmem_kstat;
+	uint64_t	zone_mcap_nover;	/* # of times over phys. cap */
+	uint64_t	zone_mcap_pagedout;	/* bytes of mem. paged out */
+	kmutex_t	zone_mcap_lock;	/* protects mcap statistics */
+	kstat_t		*zone_mcap_ksp;
+	zone_mcap_kstat_t *zone_mcap_stats;
+	uint64_t	zone_pgpgin;		/* pages paged in */
+	uint64_t	zone_anonpgin;		/* anon pages paged in */
+	uint64_t	zone_execpgin;		/* exec pages paged in */
+	uint64_t	zone_fspgin;		/* fs pages paged in */
+	uint64_t	zone_anon_alloc_fail;	/* cnt of anon alloc fails */
+	uint64_t	zone_pf_throttle;	/* cnt of page flt throttles */
+	uint64_t	zone_pf_throttle_usec;	/* time of page flt throttles */
+
+	/* Num usecs to throttle page fault when zone is over phys. mem cap */
+	uint32_t	zone_pg_flt_delay;
+
+	/*
 	 * Misc. kstats and counters for zone cpu-usage aggregation.
 	 * The zone_Xtime values are the sum of the micro-state accounting
 	 * values for all threads that are running or have run in the zone.
@@ -640,6 +754,7 @@ extern zone_t *zone_find_by_name(char *);
 extern zone_t *zone_find_by_any_path(const char *, boolean_t);
 extern zone_t *zone_find_by_path(const char *);
 extern zoneid_t getzoneid(void);
+extern zoneid_t getzonedid(void);
 extern zone_t *zone_find_by_id_nolock(zoneid_t);
 extern int zone_datalink_walk(zoneid_t, int (*)(datalink_id_t, void *), void *);
 extern int zone_check_datalink(zoneid_t *, datalink_id_t);
@@ -834,6 +949,7 @@ extern int zone_walk(int (*)(zone_t *, void *), void *);
 
 extern rctl_hndl_t rc_zone_locked_mem;
 extern rctl_hndl_t rc_zone_max_swap;
+extern rctl_hndl_t rc_zone_phys_mem;
 extern rctl_hndl_t rc_zone_max_lofi;
 
 #endif	/* _KERNEL */
diff --git a/usr/src/uts/common/syscall/brandsys.c b/usr/src/uts/common/syscall/brandsys.c
index 9b4bd38baa..8ee5511fd0 100644
--- a/usr/src/uts/common/syscall/brandsys.c
+++ b/usr/src/uts/common/syscall/brandsys.c
@@ -23,7 +23,9 @@
  * Use is subject to license terms.
  */
 
-#pragma ident	"%Z%%M%	%I%	%E% SMI"
+/*
+ * Copyright (c) 2014, Joyent, Inc. All rights reserved.
+ */
 
 #include <sys/brand.h>
 #include <sys/systm.h>
@@ -35,7 +37,7 @@
  */
 int64_t
 brandsys(int cmd, uintptr_t arg1, uintptr_t arg2, uintptr_t arg3,
-    uintptr_t arg4, uintptr_t arg5, uintptr_t arg6)
+    uintptr_t arg4, uintptr_t arg5)
 {
 	struct proc *p = curthread->t_procp;
 	int64_t rval = 0;
@@ -49,7 +51,7 @@ brandsys(int cmd, uintptr_t arg1, uintptr_t arg2, uintptr_t arg3,
 		return (set_errno(ENOSYS));
 
 	if ((err = ZBROP(p->p_zone)->b_brandsys(cmd, &rval, arg1, arg2, arg3,
-	    arg4, arg5, arg6)) != 0)
+	    arg4, arg5)) != 0)
 		return (set_errno(err));
 
 	return (rval);
diff --git a/usr/src/uts/common/syscall/fcntl.c b/usr/src/uts/common/syscall/fcntl.c
index 7421957235..d631fe62f6 100644
--- a/usr/src/uts/common/syscall/fcntl.c
+++ b/usr/src/uts/common/syscall/fcntl.c
@@ -22,6 +22,7 @@
 /*
  * Copyright (c) 1994, 2010, Oracle and/or its affiliates. All rights reserved.
  * Copyright (c) 2013, OmniTI Computer Consulting, Inc. All rights reserved.
+ * Copyright 2015, Joyent, Inc.
  */
 
 /*	Copyright (c) 1983, 1984, 1985, 1986, 1987, 1988, 1989 AT&T	*/
@@ -53,7 +54,8 @@
 
 #include <sys/cmn_err.h>
 
-static int flock_check(vnode_t *, flock64_t *, offset_t, offset_t);
+/* This is global so that it can be used by brand emulation. */
+int flock_check(vnode_t *, flock64_t *, offset_t, offset_t);
 static int flock_get_start(vnode_t *, flock64_t *, offset_t, u_offset_t *);
 static void fd_too_big(proc_t *);
 
@@ -271,11 +273,12 @@ fcntl(int fdes, int cmd, intptr_t arg)
 	 * The file system and vnode layers understand and implement
 	 * locking with flock64 structures. So here once we pass through
 	 * the test for compatibility as defined by LFS API, (for F_SETLK,
-	 * F_SETLKW, F_GETLK, F_GETLKW, F_FREESP) we transform
-	 * the flock structure to a flock64 structure and send it to the
-	 * lower layers. Similarly in case of GETLK the returned flock64
-	 * structure is transformed to a flock structure if everything fits
-	 * in nicely, otherwise we return EOVERFLOW.
+	 * F_SETLKW, F_GETLK, F_GETLKW, F_OFD_GETLK, F_OFD_SETLK, F_OFD_SETLKW,
+	 * F_FREESP) we transform the flock structure to a flock64 structure
+	 * and send it to the lower layers. Similarly in case of GETLK and
+	 * OFD_GETLK the returned flock64 structure is transformed to a flock
+	 * structure if everything fits in nicely, otherwise we return
+	 * EOVERFLOW.
 	 */
 
 	case F_GETLK:
@@ -283,6 +286,11 @@ fcntl(int fdes, int cmd, intptr_t arg)
 	case F_SETLK:
 	case F_SETLKW:
 	case F_SETLK_NBMAND:
+	case F_OFD_GETLK:
+	case F_OFD_SETLK:
+	case F_OFD_SETLKW:
+	case F_FLOCK:
+	case F_FLOCKW:
 
 		/*
 		 * Copy in input fields only.
@@ -345,20 +353,65 @@ fcntl(int fdes, int cmd, intptr_t arg)
 		if ((error = flock_check(vp, &bf, offset, maxoffset)) != 0)
 			break;
 
+		if (cmd == F_FLOCK || cmd == F_FLOCKW) {
+			/* FLOCK* locking is always over the entire file. */
+			if (bf.l_whence != 0 || bf.l_start != 0 ||
+			    bf.l_len != 0) {
+				error = EINVAL;
+				break;
+			}
+			if (bf.l_type < F_RDLCK || bf.l_type > F_UNLCK) {
+				error = EINVAL;
+				break;
+			}
+		}
+
+		if (cmd == F_OFD_SETLK || cmd == F_OFD_SETLKW) {
+			/*
+			 * TBD OFD-style locking is currently limited to
+			 * covering the entire file.
+			 */
+			if (bf.l_whence != 0 || bf.l_start != 0 ||
+			    bf.l_len != 0) {
+				error = EINVAL;
+				break;
+			}
+		}
+
 		/*
 		 * Not all of the filesystems understand F_O_GETLK, and
 		 * there's no need for them to know.  Map it to F_GETLK.
+		 *
+		 * The *_frlock functions in the various file systems basically
+		 * do some validation and then funnel everything through the
+		 * fs_frlock function. For OFD-style locks fs_frlock will do
+		 * nothing so that once control returns here we can call the
+		 * ofdlock function with the correct fp. For OFD-style locks
+		 * the unsupported remote file systems, such as NFS, detect and
+		 * reject the OFD-style cmd argument.
 		 */
 		if ((error = VOP_FRLOCK(vp, (cmd == F_O_GETLK) ? F_GETLK : cmd,
 		    &bf, flag, offset, NULL, fp->f_cred, NULL)) != 0)
 			break;
 
+		if (cmd == F_FLOCK || cmd == F_FLOCKW || cmd == F_OFD_GETLK ||
+		    cmd == F_OFD_SETLK || cmd == F_OFD_SETLKW) {
+			/*
+			 * This is an OFD-style lock so we need to handle it
+			 * here. Because OFD-style locks are associated with
+			 * the file_t we didn't have enough info down the
+			 * VOP_FRLOCK path immediately above.
+			 */
+			if ((error = ofdlock(fp, cmd, &bf, flag, offset)) != 0)
+				break;
+		}
+
 		/*
 		 * If command is GETLK and no lock is found, only
 		 * the type field is changed.
 		 */
-		if ((cmd == F_O_GETLK || cmd == F_GETLK) &&
-		    bf.l_type == F_UNLCK) {
+		if ((cmd == F_O_GETLK || cmd == F_GETLK ||
+		    cmd == F_OFD_GETLK) && bf.l_type == F_UNLCK) {
 			/* l_type always first entry, always a short */
 			if (copyout(&bf.l_type, &((struct flock *)arg)->l_type,
 			    sizeof (bf.l_type)))
@@ -387,7 +440,7 @@ fcntl(int fdes, int cmd, intptr_t arg)
 			obf.l_pid = (int16_t)bf.l_pid;
 			if (copyout(&obf, (void *)arg, sizeof (obf)))
 				error = EFAULT;
-		} else if (cmd == F_GETLK) {
+		} else if (cmd == F_GETLK || cmd == F_OFD_GETLK) {
 			/*
 			 * Copy out SVR4 flock.
 			 */
@@ -591,6 +644,11 @@ fcntl(int fdes, int cmd, intptr_t arg)
 	case F_SETLK64:
 	case F_SETLKW64:
 	case F_SETLK64_NBMAND:
+	case F_OFD_GETLK64:
+	case F_OFD_SETLK64:
+	case F_OFD_SETLKW64:
+	case F_FLOCK64:
+	case F_FLOCKW64:
 		/*
 		 * Large Files: Here we set cmd as *LK and send it to
 		 * lower layers. *LK64 is only for the user land.
@@ -611,6 +669,16 @@ fcntl(int fdes, int cmd, intptr_t arg)
 			cmd = F_SETLKW;
 		else if (cmd == F_SETLK64_NBMAND)
 			cmd = F_SETLK_NBMAND;
+		else if (cmd == F_OFD_GETLK64)
+			cmd = F_OFD_GETLK;
+		else if (cmd == F_OFD_SETLK64)
+			cmd = F_OFD_SETLK;
+		else if (cmd == F_OFD_SETLKW64)
+			cmd = F_OFD_SETLKW;
+		else if (cmd == F_FLOCK64)
+			cmd = F_FLOCK;
+		else if (cmd == F_FLOCKW64)
+			cmd = F_FLOCKW;
 
 		/*
 		 * Note that the size of flock64 is different in the ILP32
@@ -636,18 +704,65 @@ fcntl(int fdes, int cmd, intptr_t arg)
 		if ((error = flock_check(vp, &bf, offset, MAXOFFSET_T)) != 0)
 			break;
 
+		if (cmd == F_FLOCK || cmd == F_FLOCKW) {
+			/* FLOCK* locking is always over the entire file. */
+			if (bf.l_whence != 0 || bf.l_start != 0 ||
+			    bf.l_len != 0) {
+				error = EINVAL;
+				break;
+			}
+			if (bf.l_type < F_RDLCK || bf.l_type > F_UNLCK) {
+				error = EINVAL;
+				break;
+			}
+		}
+
+		if (cmd == F_OFD_SETLK || cmd == F_OFD_SETLKW) {
+			/*
+			 * TBD OFD-style locking is currently limited to
+			 * covering the entire file.
+			 */
+			if (bf.l_whence != 0 || bf.l_start != 0 ||
+			    bf.l_len != 0) {
+				error = EINVAL;
+				break;
+			}
+		}
+
+		/*
+		 * The *_frlock functions in the various file systems basically
+		 * do some validation and then funnel everything through the
+		 * fs_frlock function. For OFD-style locks fs_frlock will do
+		 * nothing so that once control returns here we can call the
+		 * ofdlock function with the correct fp. For OFD-style locks
+		 * the unsupported remote file systems, such as NFS, detect and
+		 * reject the OFD-style cmd argument.
+		 */
 		if ((error = VOP_FRLOCK(vp, cmd, &bf, flag, offset,
 		    NULL, fp->f_cred, NULL)) != 0)
 			break;
 
-		if ((cmd == F_GETLK) && bf.l_type == F_UNLCK) {
+		if (cmd == F_FLOCK || cmd == F_FLOCKW || cmd == F_OFD_GETLK ||
+		    cmd == F_OFD_SETLK || cmd == F_OFD_SETLKW) {
+			/*
+			 * This is an OFD-style lock so we need to handle it
+			 * here. Because OFD-style locks are associated with
+			 * the file_t we didn't have enough info down the
+			 * VOP_FRLOCK path immediately above.
+			 */
+			if ((error = ofdlock(fp, cmd, &bf, flag, offset)) != 0)
+				break;
+		}
+
+		if ((cmd == F_GETLK || cmd == F_OFD_GETLK) &&
+		    bf.l_type == F_UNLCK) {
 			if (copyout(&bf.l_type, &((struct flock *)arg)->l_type,
 			    sizeof (bf.l_type)))
 				error = EFAULT;
 			break;
 		}
 
-		if (cmd == F_GETLK) {
+		if (cmd == F_GETLK || cmd == F_OFD_GETLK) {
 			int i;
 
 			/*
diff --git a/usr/src/uts/common/syscall/lgrpsys.c b/usr/src/uts/common/syscall/lgrpsys.c
index 44d616fa34..1066b94986 100644
--- a/usr/src/uts/common/syscall/lgrpsys.c
+++ b/usr/src/uts/common/syscall/lgrpsys.c
@@ -22,10 +22,9 @@
 /*
  * Copyright 2006 Sun Microsystems, Inc.  All rights reserved.
  * Use is subject to license terms.
+ * Copyright 2015 Joyent, Inc.
  */
 
-#pragma ident	"%Z%%M%	%I%	%E% SMI"
-
 /*
  * lgroup system calls
  */
@@ -40,6 +39,7 @@
 #include <sys/lgrp_user.h>
 #include <sys/promif.h>		/* for prom_printf() */
 #include <sys/sysmacros.h>
+#include <sys/policy.h>
 
 #include <vm/as.h>
 
@@ -141,6 +141,24 @@ meminfo(int addr_count, struct meminfo *mip)
 	}
 
 	/*
+	 * Validate privs for each req.
+	 */
+	for (i = 0; i < info_count; i++) {
+		switch (req_array[i] & MEMINFO_MASK) {
+		case MEMINFO_VLGRP:
+		case MEMINFO_VPAGESIZE:
+			break;
+		default:
+			if (secpolicy_meminfo(CRED()) != 0) {
+				kmem_free(req_array, req_size);
+				kmem_free(in_array, in_size);
+				return (set_errno(EPERM));
+			}
+			break;
+		}
+	}
+
+	/*
 	 * allocate buffer out_array which holds the results and will have
 	 * to be copied out later
 	 */
diff --git a/usr/src/uts/common/syscall/memcntl.c b/usr/src/uts/common/syscall/memcntl.c
index ae2a0cc45c..721f884a7e 100644
--- a/usr/src/uts/common/syscall/memcntl.c
+++ b/usr/src/uts/common/syscall/memcntl.c
@@ -115,13 +115,17 @@ memcntl(caddr_t addr, size_t len, int cmd, caddr_t arg, int attr, int mask)
 		 * MS_SYNC used to be defined to be zero but is now non-zero.
 		 * For binary compatibility we still accept zero
 		 * (the absence of MS_ASYNC) to mean the same thing.
+		 * Binary compatibility is not an issue for MS_INVALCURPROC.
 		 */
 		iarg = (uintptr_t)arg;
 		if ((iarg & ~MS_INVALIDATE) == 0)
 			iarg |= MS_SYNC;
 
-		if (((iarg & ~(MS_SYNC|MS_ASYNC|MS_INVALIDATE)) != 0) ||
-		    ((iarg & (MS_SYNC|MS_ASYNC)) == (MS_SYNC|MS_ASYNC))) {
+		if (((iarg &
+		    ~(MS_SYNC|MS_ASYNC|MS_INVALIDATE|MS_INVALCURPROC)) != 0) ||
+		    ((iarg & (MS_SYNC|MS_ASYNC)) == (MS_SYNC|MS_ASYNC)) ||
+		    ((iarg & (MS_INVALIDATE|MS_INVALCURPROC)) ==
+		    (MS_INVALIDATE|MS_INVALCURPROC))) {
 			error = set_errno(EINVAL);
 		} else {
 			error = as_ctl(as, addr, len, cmd, attr, iarg, NULL, 0);
@@ -347,7 +351,8 @@ memcntl(caddr_t addr, size_t len, int cmd, caddr_t arg, int attr, int mask)
 		}
 		return (error);
 	case MC_ADVISE:
-		if ((uintptr_t)arg == MADV_FREE) {
+		if ((uintptr_t)arg == MADV_FREE ||
+		    (uintptr_t)arg == MADV_PURGE) {
 			len &= PAGEMASK;
 		}
 		switch ((uintptr_t)arg) {
diff --git a/usr/src/uts/common/syscall/open.c b/usr/src/uts/common/syscall/open.c
index edb04c824b..874e31869c 100644
--- a/usr/src/uts/common/syscall/open.c
+++ b/usr/src/uts/common/syscall/open.c
@@ -74,12 +74,12 @@ copen(int startfd, char *fname, int filemode, int createmode)
 
 	if (filemode & (FSEARCH|FEXEC)) {
 		/*
-		 * Must be one or the other and neither FREAD nor FWRITE
+		 * Must be one or the other.
 		 * Must not be any of FAPPEND FCREAT FTRUNC FXATTR FXATTRDIROPEN
-		 * XXX: Should these just be silently ignored?
+		 * XXX: Should these just be silently ignored like we
+		 * silently ignore FREAD|FWRITE?
 		 */
-		if ((filemode & (FREAD|FWRITE)) ||
-		    (filemode & (FSEARCH|FEXEC)) == (FSEARCH|FEXEC) ||
+		if ((filemode & (FSEARCH|FEXEC)) == (FSEARCH|FEXEC) ||
 		    (filemode & (FAPPEND|FCREAT|FTRUNC|FXATTR|FXATTRDIROPEN)))
 			return (set_errno(EINVAL));
 	}
diff --git a/usr/src/uts/common/syscall/poll.c b/usr/src/uts/common/syscall/poll.c
index 7f37529941..c33156a4fc 100644
--- a/usr/src/uts/common/syscall/poll.c
+++ b/usr/src/uts/common/syscall/poll.c
@@ -29,6 +29,7 @@
 
 /*
  * Copyright (c) 2012 by Delphix. All rights reserved.
+ * Copyright (c) 2014, Joyent, Inc. All rights reserved.
  */
 
 /*
@@ -525,13 +526,13 @@ poll_common(pollfd_t *fds, nfds_t nfds, timespec_t *tsp, k_sigset_t *ksetp)
 		}
 
 		/*
-		 * If T_POLLWAKE is set, a pollwakeup() was performed on
+		 * If PC_POLLWAKE is set, a pollwakeup() was performed on
 		 * one of the file descriptors.  This can happen only if
 		 * one of the VOP_POLL() functions dropped pcp->pc_lock.
 		 * The only current cases of this is in procfs (prpoll())
 		 * and STREAMS (strpoll()).
 		 */
-		if (pcp->pc_flag & T_POLLWAKE)
+		if (pcp->pc_flag & PC_POLLWAKE)
 			continue;
 
 		/*
@@ -886,9 +887,9 @@ retry:
 }
 
 /*
- * This function is called to inform a thread that
- * an event being polled for has occurred.
- * The pollstate lock on the thread should be held on entry.
+ * This function is called to inform a thread (or threads) that an event being
+ * polled on has occurred.  The pollstate lock on the thread should be held
+ * on entry.
  */
 void
 pollnotify(pollcache_t *pcp, int fd)
@@ -896,8 +897,8 @@ pollnotify(pollcache_t *pcp, int fd)
 	ASSERT(fd < pcp->pc_mapsize);
 	ASSERT(MUTEX_HELD(&pcp->pc_lock));
 	BT_SET(pcp->pc_bitmap, fd);
-	pcp->pc_flag |= T_POLLWAKE;
-	cv_signal(&pcp->pc_cv);
+	pcp->pc_flag |= PC_POLLWAKE;
+	cv_broadcast(&pcp->pc_cv);
 }
 
 /*
@@ -2024,7 +2025,7 @@ retry:
 				 */
 				if ((pdp->pd_php != NULL) &&
 				    (pollfdp[entry].events == pdp->pd_events) &&
-				    ((pcp->pc_flag & T_POLLWAKE) == 0)) {
+				    ((pcp->pc_flag & PC_POLLWAKE) == 0)) {
 					BT_CLEAR(pcp->pc_bitmap, fd);
 				}
 				/*
@@ -2251,7 +2252,7 @@ pollstate_destroy(pollstate_t *ps)
 	pcacheset_destroy(ps->ps_pcacheset, ps->ps_nsets);
 	ps->ps_pcacheset = NULL;
 	if (ps->ps_dpbuf != NULL) {
-		kmem_free(ps->ps_dpbuf, ps->ps_dpbufsize * sizeof (pollfd_t));
+		kmem_free(ps->ps_dpbuf, ps->ps_dpbufsize);
 		ps->ps_dpbuf = NULL;
 	}
 	mutex_destroy(&ps->ps_lock);
diff --git a/usr/src/uts/common/syscall/rusagesys.c b/usr/src/uts/common/syscall/rusagesys.c
index 3e0e63f4c0..417c629168 100644
--- a/usr/src/uts/common/syscall/rusagesys.c
+++ b/usr/src/uts/common/syscall/rusagesys.c
@@ -21,6 +21,7 @@
 /*
  * Copyright 2008 Sun Microsystems, Inc.  All rights reserved.
  * Use is subject to license terms.
+ * Copyright 2014 Joyent, Inc.  All rights reserved.
  */
 
 /*
@@ -257,6 +258,19 @@ rusagesys(int code, void *arg1, void *arg2, void *arg3, void *arg4)
 	case _RUSAGESYS_GETVMUSAGE:
 		return (vm_getusage((uint_t)(uintptr_t)arg1, (time_t)arg2,
 		    (vmusage_t *)arg3, (size_t *)arg4, 0));
+	case _RUSAGESYS_INVALMAP:
+		/*
+		 * SPARC sfmmu hat does not support HAT_CURPROC_PGUNLOAD
+		 * handling so callers on SPARC should get simple sync
+		 * handling with invalidation to all processes.
+		 */
+#if defined(__sparc)
+		return (memcntl((caddr_t)arg2, (size_t)arg3, MC_SYNC,
+		    (caddr_t)(MS_ASYNC | MS_INVALIDATE), 0, 0));
+#else
+		return (vm_map_inval((pid_t)(uintptr_t)arg1, (caddr_t)arg2,
+		    (size_t)arg3));
+#endif
 	default:
 		return (set_errno(EINVAL));
 	}
diff --git a/usr/src/uts/common/syscall/rw.c b/usr/src/uts/common/syscall/rw.c
index a28894b2c9..943b7d244e 100644
--- a/usr/src/uts/common/syscall/rw.c
+++ b/usr/src/uts/common/syscall/rw.c
@@ -22,7 +22,7 @@
 /*
  * Copyright 2008 Sun Microsystems, Inc.  All rights reserved.
  * Use is subject to license terms.
- * Copyright (c) 2015, Joyent, Inc.  All rights reserved.
+ * Copyright 2015, Joyent, Inc.  All rights reserved.
  */
 
 /*	Copyright (c) 1983, 1984, 1985, 1986, 1987, 1988, 1989 AT&T	*/
@@ -50,6 +50,7 @@
 #include <sys/debug.h>
 #include <sys/rctl.h>
 #include <sys/nbmlock.h>
+#include <sys/limits.h>
 
 #define	COPYOUT_MAX_CACHE	(1<<17)		/* 128K */
 
@@ -607,19 +608,12 @@ out:
 	return (bcount);
 }
 
-/*
- * XXX -- The SVID refers to IOV_MAX, but doesn't define it.  Grrrr....
- * XXX -- However, SVVS expects readv() and writev() to fail if
- * XXX -- iovcnt > 16 (yes, it's hard-coded in the SVVS source),
- * XXX -- so I guess that's the "interface".
- */
-#define	DEF_IOV_MAX	16
-
 ssize_t
 readv(int fdes, struct iovec *iovp, int iovcnt)
 {
 	struct uio auio;
-	struct iovec aiov[DEF_IOV_MAX];
+	struct iovec buf[IOV_MAX_STACK], *aiov = buf;
+	int aiovlen = 0;
 	file_t *fp;
 	register vnode_t *vp;
 	struct cpu *cp;
@@ -630,9 +624,14 @@ readv(int fdes, struct iovec *iovp, int iovcnt)
 	u_offset_t fileoff;
 	int in_crit = 0;
 
-	if (iovcnt <= 0 || iovcnt > DEF_IOV_MAX)
+	if (iovcnt <= 0 || iovcnt > IOV_MAX)
 		return (set_errno(EINVAL));
 
+	if (iovcnt > IOV_MAX_STACK) {
+		aiovlen = iovcnt * sizeof (iovec_t);
+		aiov = kmem_alloc(aiovlen, KM_SLEEP);
+	}
+
 #ifdef _SYSCALL32_IMPL
 	/*
 	 * 32-bit callers need to have their iovec expanded,
@@ -640,36 +639,63 @@ readv(int fdes, struct iovec *iovp, int iovcnt)
 	 * of data in a single call.
 	 */
 	if (get_udatamodel() == DATAMODEL_ILP32) {
-		struct iovec32 aiov32[DEF_IOV_MAX];
+		struct iovec32 buf32[IOV_MAX_STACK], *aiov32 = buf32;
+		int aiov32len;
 		ssize32_t count32;
 
-		if (copyin(iovp, aiov32, iovcnt * sizeof (struct iovec32)))
+		aiov32len = iovcnt * sizeof (iovec32_t);
+		if (aiovlen != 0)
+			aiov32 = kmem_alloc(aiov32len, KM_SLEEP);
+
+		if (copyin(iovp, aiov32, aiov32len)) {
+			if (aiovlen != 0) {
+				kmem_free(aiov32, aiov32len);
+				kmem_free(aiov, aiovlen);
+			}
 			return (set_errno(EFAULT));
+		}
 
 		count32 = 0;
 		for (i = 0; i < iovcnt; i++) {
 			ssize32_t iovlen32 = aiov32[i].iov_len;
 			count32 += iovlen32;
-			if (iovlen32 < 0 || count32 < 0)
+			if (iovlen32 < 0 || count32 < 0) {
+				if (aiovlen != 0) {
+					kmem_free(aiov32, aiov32len);
+					kmem_free(aiov, aiovlen);
+				}
 				return (set_errno(EINVAL));
+			}
 			aiov[i].iov_len = iovlen32;
 			aiov[i].iov_base =
 			    (caddr_t)(uintptr_t)aiov32[i].iov_base;
 		}
+
+		if (aiovlen != 0)
+			kmem_free(aiov32, aiov32len);
 	} else
 #endif
-	if (copyin(iovp, aiov, iovcnt * sizeof (struct iovec)))
+	if (copyin(iovp, aiov, iovcnt * sizeof (iovec_t))) {
+		if (aiovlen != 0)
+			kmem_free(aiov, aiovlen);
 		return (set_errno(EFAULT));
+	}
 
 	count = 0;
 	for (i = 0; i < iovcnt; i++) {
 		ssize_t iovlen = aiov[i].iov_len;
 		count += iovlen;
-		if (iovlen < 0 || count < 0)
+		if (iovlen < 0 || count < 0) {
+			if (aiovlen != 0)
+				kmem_free(aiov, aiovlen);
 			return (set_errno(EINVAL));
+		}
 	}
-	if ((fp = getf(fdes)) == NULL)
+	if ((fp = getf(fdes)) == NULL) {
+		if (aiovlen != 0)
+			kmem_free(aiov, aiovlen);
 		return (set_errno(EBADF));
+	}
 	if (((fflag = fp->f_flag) & FREAD) == 0) {
 		error = EBADF;
 		goto out;
@@ -768,6 +794,8 @@ out:
 	if (in_crit)
 		nbl_end_crit(vp);
 	releasef(fdes);
+	if (aiovlen != 0)
+		kmem_free(aiov, aiovlen);
 	if (error)
 		return (set_errno(error));
 	return (count);
@@ -777,7 +805,8 @@ ssize_t
 writev(int fdes, struct iovec *iovp, int iovcnt)
 {
 	struct uio auio;
-	struct iovec aiov[DEF_IOV_MAX];
+	struct iovec buf[IOV_MAX_STACK], *aiov = buf;
+	int aiovlen = 0;
 	file_t *fp;
 	register vnode_t *vp;
 	struct cpu *cp;
@@ -788,9 +817,14 @@ writev(int fdes, struct iovec *iovp, int iovcnt)
 	u_offset_t fileoff;
 	int in_crit = 0;
 
-	if (iovcnt <= 0 || iovcnt > DEF_IOV_MAX)
+	if (iovcnt <= 0 || iovcnt > IOV_MAX)
 		return (set_errno(EINVAL));
 
+	if (iovcnt > IOV_MAX_STACK) {
+		aiovlen = iovcnt * sizeof (iovec_t);
+		aiov = kmem_alloc(aiovlen, KM_SLEEP);
+	}
+
 #ifdef _SYSCALL32_IMPL
 	/*
 	 * 32-bit callers need to have their iovec expanded,
@@ -798,36 +832,62 @@ writev(int fdes, struct iovec *iovp, int iovcnt)
 	 * of data in a single call.
 	 */
 	if (get_udatamodel() == DATAMODEL_ILP32) {
-		struct iovec32 aiov32[DEF_IOV_MAX];
+		struct iovec32 buf32[IOV_MAX_STACK], *aiov32 = buf32;
+		int aiov32len;
 		ssize32_t count32;
 
-		if (copyin(iovp, aiov32, iovcnt * sizeof (struct iovec32)))
+		aiov32len = iovcnt * sizeof (iovec32_t);
+		if (aiovlen != 0)
+			aiov32 = kmem_alloc(aiov32len, KM_SLEEP);
+
+		if (copyin(iovp, aiov32, aiov32len)) {
+			if (aiovlen != 0) {
+				kmem_free(aiov32, aiov32len);
+				kmem_free(aiov, aiovlen);
+			}
 			return (set_errno(EFAULT));
+		}
 
 		count32 = 0;
 		for (i = 0; i < iovcnt; i++) {
 			ssize32_t iovlen = aiov32[i].iov_len;
 			count32 += iovlen;
-			if (iovlen < 0 || count32 < 0)
+			if (iovlen < 0 || count32 < 0) {
+				if (aiovlen != 0) {
+					kmem_free(aiov32, aiov32len);
+					kmem_free(aiov, aiovlen);
+				}
 				return (set_errno(EINVAL));
+			}
 			aiov[i].iov_len = iovlen;
 			aiov[i].iov_base =
 			    (caddr_t)(uintptr_t)aiov32[i].iov_base;
 		}
+		if (aiovlen != 0)
+			kmem_free(aiov32, aiov32len);
 	} else
 #endif
-	if (copyin(iovp, aiov, iovcnt * sizeof (struct iovec)))
+	if (copyin(iovp, aiov, iovcnt * sizeof (iovec_t))) {
+		if (aiovlen != 0)
+			kmem_free(aiov, aiovlen);
 		return (set_errno(EFAULT));
+	}
 
 	count = 0;
 	for (i = 0; i < iovcnt; i++) {
 		ssize_t iovlen = aiov[i].iov_len;
 		count += iovlen;
-		if (iovlen < 0 || count < 0)
+		if (iovlen < 0 || count < 0) {
+			if (aiovlen != 0)
+				kmem_free(aiov, aiovlen);
 			return (set_errno(EINVAL));
+		}
 	}
-	if ((fp = getf(fdes)) == NULL)
+	if ((fp = getf(fdes)) == NULL) {
+		if (aiovlen != 0)
+			kmem_free(aiov, aiovlen);
 		return (set_errno(EBADF));
+	}
 	if (((fflag = fp->f_flag) & FWRITE) == 0) {
 		error = EBADF;
 		goto out;
@@ -917,6 +977,8 @@ out:
 	if (in_crit)
 		nbl_end_crit(vp);
 	releasef(fdes);
+	if (aiovlen != 0)
+		kmem_free(aiov, aiovlen);
 	if (error)
 		return (set_errno(error));
 	return (count);
@@ -927,7 +989,8 @@ preadv(int fdes, struct iovec *iovp, int iovcnt, off_t offset,
     off_t extended_offset)
 {
 	struct uio auio;
-	struct iovec aiov[DEF_IOV_MAX];
+	struct iovec buf[IOV_MAX_STACK], *aiov = buf;
+	int aiovlen = 0;
 	file_t *fp;
 	register vnode_t *vp;
 	struct cpu *cp;
@@ -952,9 +1015,14 @@ preadv(int fdes, struct iovec *iovp, int iovcnt, off_t offset,
 
 	int in_crit = 0;
 
-	if (iovcnt <= 0 || iovcnt > DEF_IOV_MAX)
+	if (iovcnt <= 0 || iovcnt > IOV_MAX)
 		return (set_errno(EINVAL));
 
+	if (iovcnt > IOV_MAX_STACK) {
+		aiovlen = iovcnt * sizeof (iovec_t);
+		aiov = kmem_alloc(aiovlen, KM_SLEEP);
+	}
+
 #ifdef _SYSCALL32_IMPL
 	/*
 	 * 32-bit callers need to have their iovec expanded,
@@ -962,39 +1030,68 @@ preadv(int fdes, struct iovec *iovp, int iovcnt, off_t offset,
 	 * of data in a single call.
 	 */
 	if (get_udatamodel() == DATAMODEL_ILP32) {
-		struct iovec32 aiov32[DEF_IOV_MAX];
+		struct iovec32 buf32[IOV_MAX_STACK], *aiov32 = buf32;
+		int aiov32len;
 		ssize32_t count32;
 
-		if (copyin(iovp, aiov32, iovcnt * sizeof (struct iovec32)))
+		aiov32len = iovcnt * sizeof (iovec32_t);
+		if (aiovlen != 0)
+			aiov32 = kmem_alloc(aiov32len, KM_SLEEP);
+
+		if (copyin(iovp, aiov32, aiov32len)) {
+			if (aiovlen != 0) {
+				kmem_free(aiov32, aiov32len);
+				kmem_free(aiov, aiovlen);
+			}
 			return (set_errno(EFAULT));
+		}
 
 		count32 = 0;
 		for (i = 0; i < iovcnt; i++) {
 			ssize32_t iovlen32 = aiov32[i].iov_len;
 			count32 += iovlen32;
-			if (iovlen32 < 0 || count32 < 0)
+			if (iovlen32 < 0 || count32 < 0) {
+				if (aiovlen != 0) {
+					kmem_free(aiov32, aiov32len);
+					kmem_free(aiov, aiovlen);
+				}
 				return (set_errno(EINVAL));
+			}
 			aiov[i].iov_len = iovlen32;
 			aiov[i].iov_base =
 			    (caddr_t)(uintptr_t)aiov32[i].iov_base;
 		}
+		if (aiovlen != 0)
+			kmem_free(aiov32, aiov32len);
 	} else
 #endif /* _SYSCALL32_IMPL */
-		if (copyin(iovp, aiov, iovcnt * sizeof (struct iovec)))
+		if (copyin(iovp, aiov, iovcnt * sizeof (iovec_t))) {
+			if (aiovlen != 0)
+				kmem_free(aiov, aiovlen);
 			return (set_errno(EFAULT));
+		}
 
 	count = 0;
 	for (i = 0; i < iovcnt; i++) {
 		ssize_t iovlen = aiov[i].iov_len;
 		count += iovlen;
-		if (iovlen < 0 || count < 0)
+		if (iovlen < 0 || count < 0) {
+			if (aiovlen != 0)
+				kmem_free(aiov, aiovlen);
 			return (set_errno(EINVAL));
+		}
 	}
 
-	if ((bcount = (ssize_t)count) < 0)
+	if ((bcount = (ssize_t)count) < 0) {
+		if (aiovlen != 0)
+			kmem_free(aiov, aiovlen);
 		return (set_errno(EINVAL));
-	if ((fp = getf(fdes)) == NULL)
+	}
+	if ((fp = getf(fdes)) == NULL) {
+		if (aiovlen != 0)
+			kmem_free(aiov, aiovlen);
 		return (set_errno(EBADF));
+	}
 	if (((fflag = fp->f_flag) & FREAD) == 0) {
 		error = EBADF;
 		goto out;
@@ -1099,6 +1196,8 @@ out:
 	if (in_crit)
 		nbl_end_crit(vp);
 	releasef(fdes);
+	if (aiovlen != 0)
+		kmem_free(aiov, aiovlen);
 	if (error)
 		return (set_errno(error));
 	return (count);
@@ -1109,7 +1208,8 @@ pwritev(int fdes, struct iovec *iovp, int iovcnt, off_t offset,
     off_t extended_offset)
 {
 	struct uio auio;
-	struct iovec aiov[DEF_IOV_MAX];
+	struct iovec buf[IOV_MAX_STACK], *aiov = buf;
+	int aiovlen = 0;
 	file_t *fp;
 	register vnode_t *vp;
 	struct cpu *cp;
@@ -1134,9 +1234,14 @@ pwritev(int fdes, struct iovec *iovp, int iovcnt, off_t offset,
 
 	int in_crit = 0;
 
-	if (iovcnt <= 0 || iovcnt > DEF_IOV_MAX)
+	if (iovcnt <= 0 || iovcnt > IOV_MAX)
 		return (set_errno(EINVAL));
 
+	if (iovcnt > IOV_MAX_STACK) {
+		aiovlen = iovcnt * sizeof (iovec_t);
+		aiov = kmem_alloc(aiovlen, KM_SLEEP);
+	}
+
 #ifdef _SYSCALL32_IMPL
 	/*
 	 * 32-bit callers need to have their iovec expanded,
@@ -1144,39 +1249,68 @@ pwritev(int fdes, struct iovec *iovp, int iovcnt, off_t offset,
 	 * of data in a single call.
 	 */
 	if (get_udatamodel() == DATAMODEL_ILP32) {
-		struct iovec32 aiov32[DEF_IOV_MAX];
+		struct iovec32 buf32[IOV_MAX_STACK], *aiov32 = buf32;
+		int aiov32len;
 		ssize32_t count32;
 
-		if (copyin(iovp, aiov32, iovcnt * sizeof (struct iovec32)))
+		aiov32len = iovcnt * sizeof (iovec32_t);
+		if (aiovlen != 0)
+			aiov32 = kmem_alloc(aiov32len, KM_SLEEP);
+
+		if (copyin(iovp, aiov32, aiov32len)) {
+			if (aiovlen != 0) {
+				kmem_free(aiov32, aiov32len);
+				kmem_free(aiov, aiovlen);
+			}
 			return (set_errno(EFAULT));
+		}
 
 		count32 = 0;
 		for (i = 0; i < iovcnt; i++) {
 			ssize32_t iovlen32 = aiov32[i].iov_len;
 			count32 += iovlen32;
-			if (iovlen32 < 0 || count32 < 0)
+			if (iovlen32 < 0 || count32 < 0) {
+				if (aiovlen != 0) {
+					kmem_free(aiov32, aiov32len);
+					kmem_free(aiov, aiovlen);
+				}
 				return (set_errno(EINVAL));
+			}
 			aiov[i].iov_len = iovlen32;
 			aiov[i].iov_base =
 			    (caddr_t)(uintptr_t)aiov32[i].iov_base;
 		}
+		if (aiovlen != 0)
+			kmem_free(aiov32, aiov32len);
 	} else
 #endif /* _SYSCALL32_IMPL */
-		if (copyin(iovp, aiov, iovcnt * sizeof (struct iovec)))
+		if (copyin(iovp, aiov, iovcnt * sizeof (iovec_t))) {
+			if (aiovlen != 0)
+				kmem_free(aiov, aiovlen);
 			return (set_errno(EFAULT));
+		}
 
 	count = 0;
 	for (i = 0; i < iovcnt; i++) {
 		ssize_t iovlen = aiov[i].iov_len;
 		count += iovlen;
-		if (iovlen < 0 || count < 0)
+		if (iovlen < 0 || count < 0) {
+			if (aiovlen != 0)
+				kmem_free(aiov, aiovlen);
 			return (set_errno(EINVAL));
+		}
 	}
 
-	if ((bcount = (ssize_t)count) < 0)
+	if ((bcount = (ssize_t)count) < 0) {
+		if (aiovlen != 0)
+			kmem_free(aiov, aiovlen);
 		return (set_errno(EINVAL));
-	if ((fp = getf(fdes)) == NULL)
+	}
+	if ((fp = getf(fdes)) == NULL) {
+		if (aiovlen != 0)
+			kmem_free(aiov, aiovlen);
 		return (set_errno(EBADF));
+	}
 	if (((fflag = fp->f_flag) & FWRITE) == 0) {
 		error = EBADF;
 		goto out;
@@ -1308,6 +1442,8 @@ out:
 	if (in_crit)
 		nbl_end_crit(vp);
 	releasef(fdes);
+	if (aiovlen != 0)
+		kmem_free(aiov, aiovlen);
 	if (error)
 		return (set_errno(error));
 	return (count);
diff --git a/usr/src/uts/common/syscall/sendfile.c b/usr/src/uts/common/syscall/sendfile.c
index cb8246f584..ccceca7c6d 100644
--- a/usr/src/uts/common/syscall/sendfile.c
+++ b/usr/src/uts/common/syscall/sendfile.c
@@ -82,7 +82,7 @@ extern sotpi_info_t *sotpi_sototpi(struct sonode *);
  * 64 bit kernel or 32 bit kernel. For 32 bit apps, we can't transfer
  * more than 2GB of data.
  */
-int
+static int
 sendvec_chunk64(file_t *fp, u_offset_t *fileoff, struct ksendfilevec64 *sfv,
     int copy_cnt, ssize32_t *count)
 {
@@ -343,7 +343,7 @@ sendvec_chunk64(file_t *fp, u_offset_t *fileoff, struct ksendfilevec64 *sfv,
 	return (0);
 }
 
-ssize32_t
+static ssize32_t
 sendvec64(file_t *fp, const struct ksendfilevec64 *vec, int sfvcnt,
 	size32_t *xferred, int fildes)
 {
@@ -390,7 +390,7 @@ sendvec64(file_t *fp, const struct ksendfilevec64 *vec, int sfvcnt,
 }
 #endif
 
-int
+static int
 sendvec_small_chunk(file_t *fp, u_offset_t *fileoff, struct sendfilevec *sfv,
     int copy_cnt, ssize_t total_size, int maxblk, ssize_t *count)
 {
@@ -680,7 +680,7 @@ sendvec_small_chunk(file_t *fp, u_offset_t *fileoff, struct sendfilevec *sfv,
 }
 
 
-int
+static int
 sendvec_chunk(file_t *fp, u_offset_t *fileoff, struct sendfilevec *sfv,
     int copy_cnt, ssize_t *count)
 {
@@ -1160,6 +1160,17 @@ sendfilev(int opcode, int fildes, const struct sendfilevec *vec, int sfvcnt,
 		} else {
 			maxblk = (int)vp->v_stream->sd_maxblk;
 		}
+
+		/*
+		 * We need to make sure that the socket that we're sending on
+		 * supports sendfile behavior. sockfs doesn't know that the APIs
+		 * we want to use are coming from sendfile, so we can't rely on
+		 * it to check for us.
+		 */
+		if ((so->so_mode & SM_SENDFILESUPP) == 0) {
+			error = EOPNOTSUPP;
+			goto err;
+		}
 		break;
 	case VREG:
 		break;
diff --git a/usr/src/uts/common/syscall/sysconfig.c b/usr/src/uts/common/syscall/sysconfig.c
index 03f2fabe13..26ea859224 100644
--- a/usr/src/uts/common/syscall/sysconfig.c
+++ b/usr/src/uts/common/syscall/sysconfig.c
@@ -22,6 +22,7 @@
 /*
  * Copyright 2008 Sun Microsystems, Inc.  All rights reserved.
  * Use is subject to license terms.
+ * Copyright 2012 Joyent, Inc.  All rights reserved.
  */
 
 /*	Copyright (c) 1984, 1986, 1987, 1988, 1989 AT&T	*/
@@ -170,8 +171,8 @@ sysconfig(int which)
 		 * even though rcapd can be used on the global zone too.
 		 */
 		if (!INGLOBALZONE(curproc) &&
-		    curproc->p_zone->zone_phys_mcap != 0)
-			return (MIN(btop(curproc->p_zone->zone_phys_mcap),
+		    curproc->p_zone->zone_phys_mem_ctl != UINT64_MAX)
+			return (MIN(btop(curproc->p_zone->zone_phys_mem_ctl),
 			    physinstalled));
 
 		return (physinstalled);
@@ -179,26 +180,23 @@ sysconfig(int which)
 	case _CONFIG_AVPHYS_PAGES:
 		/*
 		 * If the non-global zone has a phys. memory cap, use
-		 * the phys. memory cap - zone's current rss.  We always
+		 * the phys. memory cap - zone's rss.  We always
 		 * report the system-wide value for the global zone, even
-		 * though rcapd can be used on the global zone too.
+		 * though memory capping can be used on the global zone too.
+		 * We use the cached value for the RSS since vm_getusage()
+		 * is so expensive and we don't need this value to be exact.
 		 */
 		if (!INGLOBALZONE(curproc) &&
-		    curproc->p_zone->zone_phys_mcap != 0) {
+		    curproc->p_zone->zone_phys_mem_ctl != UINT64_MAX) {
 			pgcnt_t cap, rss, free;
-			vmusage_t in_use;
-			size_t cnt = 1;
 
-			cap = btop(curproc->p_zone->zone_phys_mcap);
+			cap = btop(curproc->p_zone->zone_phys_mem_ctl);
 			if (cap > physinstalled)
 				return (freemem);
 
-			if (vm_getusage(VMUSAGE_ZONE, 1, &in_use, &cnt,
-			    FKIOCTL) != 0)
-				in_use.vmu_rss_all = 0;
-			rss = btop(in_use.vmu_rss_all);
+			rss = btop(curproc->p_zone->zone_phys_mem);
 			/*
-			 * Because rcapd implements a soft cap, it is possible
+			 * Because this is a soft cap, it is possible
 			 * for rss to be temporarily over the cap.
 			 */
 			if (cap > rss)
diff --git a/usr/src/uts/common/syscall/uadmin.c b/usr/src/uts/common/syscall/uadmin.c
index 2dda4001bf..68aa1a95f5 100644
--- a/usr/src/uts/common/syscall/uadmin.c
+++ b/usr/src/uts/common/syscall/uadmin.c
@@ -78,7 +78,7 @@ volatile int fastreboot_dryrun = 0;
  * system with many zones.
  */
 void
-killall(zoneid_t zoneid)
+killall(zoneid_t zoneid, boolean_t force)
 {
 	proc_t *p;
 
@@ -108,7 +108,7 @@ killall(zoneid_t zoneid)
 		    p->p_stat != SIDL &&
 		    p->p_stat != SZOMB) {
 			mutex_enter(&p->p_lock);
-			if (sigismember(&p->p_sig, SIGKILL)) {
+			if (!force && sigismember(&p->p_sig, SIGKILL)) {
 				mutex_exit(&p->p_lock);
 				p = p->p_next;
 			} else {
@@ -245,7 +245,7 @@ kadmin(int cmd, int fcn, void *mdep, cred_t *credp)
 		 */
 		zone_shutdown_global();
 
-		killall(ALL_ZONES);
+		killall(ALL_ZONES, B_FALSE);
 		/*
 		 * If we are calling kadmin() from a kernel context then we
 		 * do not release these resources.
diff --git a/usr/src/uts/common/vm/anon.h b/usr/src/uts/common/vm/anon.h
index 2a39c896bc..1273b984fb 100644
--- a/usr/src/uts/common/vm/anon.h
+++ b/usr/src/uts/common/vm/anon.h
@@ -20,6 +20,7 @@
  */
 /*
  * Copyright (c) 1986, 2010, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2015, Joyent, Inc. All rights reserved.
  */
 
 /*	Copyright (c) 1984, 1986, 1987, 1988, 1989 AT&T */
@@ -390,7 +391,8 @@ extern int	anon_fill_cow_holes(struct seg *, caddr_t, struct anon_hdr *,
 		    uint_t, struct vpage [], struct cred *);
 extern void	anon_free(struct anon_hdr *, ulong_t, size_t);
 extern void	anon_free_pages(struct anon_hdr *, ulong_t, size_t, uint_t);
-extern void	anon_disclaim(struct anon_map *, ulong_t, size_t);
+extern int	anon_disclaim(struct anon_map *,
+		    ulong_t, size_t, uint_t, pgcnt_t *);
 extern int	anon_getpage(struct anon **, uint_t *, struct page **,
 		    size_t, struct seg *, caddr_t, enum seg_rw, struct cred *);
 extern int	swap_getconpage(struct vnode *, u_offset_t, size_t,
diff --git a/usr/src/uts/common/vm/hat.h b/usr/src/uts/common/vm/hat.h
index 1d91475e38..c908a9e16c 100644
--- a/usr/src/uts/common/vm/hat.h
+++ b/usr/src/uts/common/vm/hat.h
@@ -21,6 +21,7 @@
 /*
  * Copyright 2009 Sun Microsystems, Inc.  All rights reserved.
  * Use is subject to license terms.
+ * Copyright 2014 Joyent, Inc.  All rights reserved.
  */
 
 /*	Copyright (c) 1984, 1986, 1987, 1988, 1989 AT&T	*/
@@ -269,7 +270,12 @@ void	hat_kpm_walk(void (*)(void *, void *, size_t), void *);
  *	call.
  *
  * int hat_pageunload(pp, forceflag)
- *	unload all translations attached to pp.
+ *	Unload all translations attached to pp. On x86 the bulk of the work is
+ *	done by hat_page_inval.
+ *
+ * void	hat_page_inval(pp, pgsz, curhat)
+ *	Unload translations attached to pp. If curhat is provided, only the
+ *	translation for that process is unloaded, otherwise all are unloaded.
  *
  * uint_t hat_pagesync(pp, flags)
  *	get hw stats from hardware into page struct and reset hw stats
@@ -291,6 +297,7 @@ void	hat_page_setattr(struct page *, uint_t);
 void	hat_page_clrattr(struct page *, uint_t);
 uint_t	hat_page_getattr(struct page *, uint_t);
 int	hat_pageunload(struct page *, uint_t);
+void	hat_page_inval(struct page *, uint_t, struct hat *);
 uint_t	hat_pagesync(struct page *, uint_t);
 ulong_t	hat_page_getshare(struct page *);
 int	hat_page_checkshare(struct page *, ulong_t);
@@ -460,6 +467,7 @@ void	hat_setstat(struct as *, caddr_t, size_t, uint_t);
  */
 #define	HAT_ADV_PGUNLOAD	0x00
 #define	HAT_FORCE_PGUNLOAD	0x01
+#define	HAT_CURPROC_PGUNLOAD	0x02
 
 /*
  * Attributes for hat_page_*attr, hat_setstats and
diff --git a/usr/src/uts/common/vm/seg_spt.c b/usr/src/uts/common/vm/seg_spt.c
index 8d85fbaef7..2b0b7d60b9 100644
--- a/usr/src/uts/common/vm/seg_spt.c
+++ b/usr/src/uts/common/vm/seg_spt.c
@@ -20,6 +20,7 @@
  */
 /*
  * Copyright (c) 1993, 2010, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2015, Joyent, Inc. All rights reserved.
  */
 
 #include <sys/param.h>
@@ -2880,7 +2881,7 @@ segspt_shmadvise(struct seg *seg, caddr_t addr, size_t len, uint_t behav)
 
 	ASSERT(seg->s_as && AS_LOCK_HELD(seg->s_as, &seg->s_as->a_lock));
 
-	if (behav == MADV_FREE) {
+	if (behav == MADV_FREE || behav == MADV_PURGE) {
 		if ((sptd->spt_flags & SHM_PAGEABLE) == 0)
 			return (0);
 
@@ -2891,7 +2892,7 @@ segspt_shmadvise(struct seg *seg, caddr_t addr, size_t len, uint_t behav)
 		if ((ppa = sptd->spt_ppa) == NULL) {
 			mutex_exit(&sptd->spt_lock);
 			ANON_LOCK_ENTER(&amp->a_rwlock, RW_READER);
-			anon_disclaim(amp, pg_idx, len);
+			(void) anon_disclaim(amp, pg_idx, len, behav, NULL);
 			ANON_LOCK_EXIT(&amp->a_rwlock);
 			return (0);
 		}
@@ -2950,7 +2951,7 @@ segspt_shmadvise(struct seg *seg, caddr_t addr, size_t len, uint_t behav)
 		}
 
 		ANON_LOCK_ENTER(&amp->a_rwlock, RW_READER);
-		anon_disclaim(amp, pg_idx, len);
+		(void) anon_disclaim(amp, pg_idx, len, behav, NULL);
 		ANON_LOCK_EXIT(&amp->a_rwlock);
 	} else if (lgrp_optimizations() && (behav == MADV_ACCESS_LWP ||
 	    behav == MADV_ACCESS_MANY || behav == MADV_ACCESS_DEFAULT)) {
diff --git a/usr/src/uts/common/vm/seg_vn.c b/usr/src/uts/common/vm/seg_vn.c
index 7e514ba1c9..8f7afb7ba0 100644
--- a/usr/src/uts/common/vm/seg_vn.c
+++ b/usr/src/uts/common/vm/seg_vn.c
@@ -7329,7 +7329,8 @@ segvn_sync(struct seg *seg, caddr_t addr, size_t len, int attr, uint_t flags)
 	vpp = svd->vpage;
 	offset = svd->offset + (uintptr_t)(addr - seg->s_base);
 	bflags = ((flags & MS_ASYNC) ? B_ASYNC : 0) |
-	    ((flags & MS_INVALIDATE) ? B_INVAL : 0);
+	    ((flags & MS_INVALIDATE) ? B_INVAL : 0) |
+	    ((flags & MS_INVALCURPROC) ? (B_INVALCURONLY | B_INVAL) : 0);
 
 	if (attr) {
 		pageprot = attr & ~(SHARED|PRIVATE);
@@ -7354,11 +7355,11 @@ segvn_sync(struct seg *seg, caddr_t addr, size_t len, int attr, uint_t flags)
 			vpp = &svd->vpage[seg_page(seg, addr)];
 
 	} else if (svd->vp && svd->amp == NULL &&
-	    (flags & MS_INVALIDATE) == 0) {
+	    (flags & (MS_INVALIDATE | MS_INVALCURPROC)) == 0) {
 
 		/*
-		 * No attributes, no anonymous pages and MS_INVALIDATE flag
-		 * is not on, just use one big request.
+		 * No attributes, no anonymous pages and MS_INVAL* flags
+		 * are not on, just use one big request.
 		 */
 		err = VOP_PUTPAGE(svd->vp, (offset_t)offset, len,
 		    bflags, svd->cred, NULL);
@@ -7410,7 +7411,7 @@ segvn_sync(struct seg *seg, caddr_t addr, size_t len, int attr, uint_t flags)
 		 * might race in and lock the page after we unlock and before
 		 * we do the PUTPAGE, then PUTPAGE simply does nothing.
 		 */
-		if (flags & MS_INVALIDATE) {
+		if (flags & (MS_INVALIDATE | MS_INVALCURPROC)) {
 			if ((pp = page_lookup(vp, off, SE_SHARED)) != NULL) {
 				if (pp->p_lckcnt != 0 || pp->p_cowcnt != 0) {
 					page_unlock(pp);
@@ -8064,7 +8065,7 @@ out:
 
 /*
  * Set advice from user for specified pages
- * There are 9 types of advice:
+ * There are 10 types of advice:
  *	MADV_NORMAL	- Normal (default) behavior (whatever that is)
  *	MADV_RANDOM	- Random page references
  *				do not allow readahead or 'klustering'
@@ -8078,6 +8079,7 @@ out:
  *	MADV_ACCESS_DEFAULT- Default access
  *	MADV_ACCESS_LWP	- Next LWP will access heavily
  *	MADV_ACCESS_MANY- Many LWPs or processes will access heavily
+ *	MADV_PURGE	- Contents will be immediately discarded
  */
 static int
 segvn_advise(struct seg *seg, caddr_t addr, size_t len, uint_t behav)
@@ -8096,10 +8098,10 @@ segvn_advise(struct seg *seg, caddr_t addr, size_t len, uint_t behav)
 	ASSERT(seg->s_as && AS_LOCK_HELD(seg->s_as, &seg->s_as->a_lock));
 
 	/*
-	 * In case of MADV_FREE, we won't be modifying any segment private
-	 * data structures; so, we only need to grab READER's lock
+	 * In case of MADV_FREE/MADV_PURGE, we won't be modifying any segment
+	 * private data structures; so, we only need to grab READER's lock
 	 */
-	if (behav != MADV_FREE) {
+	if (behav != MADV_FREE && behav != MADV_PURGE) {
 		SEGVN_LOCK_ENTER(seg->s_as, &svd->lock, RW_WRITER);
 		if (svd->tr_state != SEGVN_TR_OFF) {
 			SEGVN_LOCK_EXIT(seg->s_as, &svd->lock);
@@ -8175,27 +8177,65 @@ segvn_advise(struct seg *seg, caddr_t addr, size_t len, uint_t behav)
 
 	amp = svd->amp;
 	vp = svd->vp;
-	if (behav == MADV_FREE) {
-		/*
-		 * MADV_FREE is not supported for segments with
-		 * underlying object; if anonmap is NULL, anon slots
-		 * are not yet populated and there is nothing for
-		 * us to do. As MADV_FREE is advisory, we don't
-		 * return error in either case.
-		 */
-		if (vp != NULL || amp == NULL) {
+	if (behav == MADV_FREE || behav == MADV_PURGE) {
+		pgcnt_t purged;
+
+		if (behav == MADV_FREE && (vp != NULL || amp == NULL)) {
+			/*
+			 * MADV_FREE is not supported for segments with an
+			 * underlying object; if anonmap is NULL, anon slots
+			 * are not yet populated and there is nothing for us
+			 * to do. As MADV_FREE is advisory, we don't return an
+			 * error in either case.
+			 */
 			SEGVN_LOCK_EXIT(seg->s_as, &svd->lock);
 			return (0);
 		}
 
+		if (amp == NULL) {
+			/*
+			 * If we're here with a NULL anonmap, it's because we
+			 * are doing a MADV_PURGE.  We have nothing to do, but
+			 * because MADV_PURGE isn't merely advisory, we return
+			 * an error in this case.
+			 */
+			SEGVN_LOCK_EXIT(seg->s_as, &svd->lock);
+			return (EBUSY);
+		}
+
 		segvn_purge(seg);
 
 		page = seg_page(seg, addr);
 		ANON_LOCK_ENTER(&amp->a_rwlock, RW_READER);
-		anon_disclaim(amp, svd->anon_index + page, len);
+		err = anon_disclaim(amp,
+		    svd->anon_index + page, len, behav, &purged);
+
+		if (purged != 0 && (svd->flags & MAP_NORESERVE)) {
+			/*
+			 * If we purged pages on a MAP_NORESERVE mapping, we
+			 * need to be sure to now unreserve our reserved swap.
+			 * (We use the atomic operations to manipulate our
+			 * segment and address space counters because we only
+			 * have the corresponding locks held as reader, not
+			 * writer.)
+			 */
+			ssize_t bytes = ptob(purged);
+
+			anon_unresv_zone(bytes, seg->s_as->a_proc->p_zone);
+			atomic_add_long(&svd->swresv, -bytes);
+			atomic_add_long(&seg->s_as->a_resvsize, -bytes);
+		}
+
 		ANON_LOCK_EXIT(&amp->a_rwlock);
 		SEGVN_LOCK_EXIT(seg->s_as, &svd->lock);
-		return (0);
+
+		/*
+		 * MADV_PURGE and MADV_FREE differ in their return semantics:
+		 * because MADV_PURGE is designed to be bug-for-bug compatible
+		 * with its clumsy Linux forebear, it will fail where MADV_FREE
+		 * does not.
+		 */
+		return (behav == MADV_PURGE ? err : 0);
 	}
 
 	/*
@@ -8301,6 +8341,7 @@ segvn_advise(struct seg *seg, caddr_t addr, size_t len, uint_t behav)
 		case MADV_WILLNEED:	/* handled in memcntl */
 		case MADV_DONTNEED:	/* handled in memcntl */
 		case MADV_FREE:		/* handled above */
+		case MADV_PURGE:	/* handled above */
 			break;
 		default:
 			err = EINVAL;
@@ -8536,6 +8577,7 @@ segvn_advise(struct seg *seg, caddr_t addr, size_t len, uint_t behav)
 		case MADV_WILLNEED:	/* handled in memcntl */
 		case MADV_DONTNEED:	/* handled in memcntl */
 		case MADV_FREE:		/* handled above */
+		case MADV_PURGE:	/* handled above */
 			break;
 		default:
 			err = EINVAL;
diff --git a/usr/src/uts/common/vm/vm_anon.c b/usr/src/uts/common/vm/vm_anon.c
index fdf9f7790c..01db9b23d7 100644
--- a/usr/src/uts/common/vm/vm_anon.c
+++ b/usr/src/uts/common/vm/vm_anon.c
@@ -20,6 +20,7 @@
  */
 /*
  * Copyright (c) 1986, 2010, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2015, Joyent, Inc. All rights reserved.
  */
 
 /*	Copyright (c) 1984, 1986, 1987, 1988, 1989 AT&T	*/
@@ -787,13 +788,21 @@ anon_resvmem(size_t size, boolean_t takemem, zone_t *zone, int tryhard)
 	pgcnt_t pswap_pages = 0;
 	proc_t *p = curproc;
 
-	if (zone != NULL && takemem) {
+	if (zone != NULL) {
 		/* test zone.max-swap resource control */
 		mutex_enter(&p->p_lock);
 		if (rctl_incr_swap(p, zone, ptob(npages)) != 0) {
 			mutex_exit(&p->p_lock);
+
+			if (takemem)
+				atomic_add_64(&zone->zone_anon_alloc_fail, 1);
+
 			return (0);
 		}
+
+		if (!takemem)
+			rctl_decr_swap(zone, ptob(npages));
+
 		mutex_exit(&p->p_lock);
 	}
 	mutex_enter(&anoninfo_lock);
@@ -1660,8 +1669,9 @@ anon_free_pages(
 /*
  * Make anonymous pages discardable
  */
-void
-anon_disclaim(struct anon_map *amp, ulong_t index, size_t size)
+int
+anon_disclaim(struct anon_map *amp, ulong_t index, size_t size,
+    uint_t behav, pgcnt_t *purged)
 {
 	spgcnt_t npages = btopr(size);
 	struct anon *ap;
@@ -1669,11 +1679,13 @@ anon_disclaim(struct anon_map *amp, ulong_t index, size_t size)
 	anoff_t off;
 	page_t *pp, *root_pp;
 	kmutex_t *ahm;
-	pgcnt_t pgcnt;
+	pgcnt_t pgcnt, npurged = 0;
 	ulong_t old_idx, idx, i;
 	struct anon_hdr *ahp = amp->ahp;
 	anon_sync_obj_t cookie;
+	int err = 0;
 
+	VERIFY(behav == MADV_FREE || behav == MADV_PURGE);
 	ASSERT(RW_READ_HELD(&amp->a_rwlock));
 	pgcnt = 1;
 	for (; npages > 0; index = (pgcnt == 1) ? index + 1 :
@@ -1721,6 +1733,7 @@ anon_disclaim(struct anon_map *amp, ulong_t index, size_t size)
 			page_unlock(pp);
 			segadvstat.MADV_FREE_miss.value.ul++;
 			anon_array_exit(&cookie);
+			err = EBUSY;
 			continue;
 		}
 
@@ -1738,6 +1751,15 @@ anon_disclaim(struct anon_map *amp, ulong_t index, size_t size)
 			continue;
 		}
 
+		if (behav == MADV_PURGE && pp->p_szc != 0) {
+			/*
+			 * If we're purging and we have a large page, simplify
+			 * things a bit by demoting ourselves into the base
+			 * page case.
+			 */
+			(void) page_try_demote_pages(pp);
+		}
+
 		if (pp->p_szc == 0) {
 			pgcnt = 1;
 
@@ -1750,7 +1772,27 @@ anon_disclaim(struct anon_map *amp, ulong_t index, size_t size)
 				ap->an_pvp = NULL;
 				ap->an_poff = 0;
 			}
-			mutex_exit(ahm);
+
+			if (behav == MADV_PURGE) {
+				/*
+				 * If we're purging (instead of merely freeing),
+				 * rip out this anon structure entirely to
+				 * assure that any subsequent fault pulls from
+				 * the backing vnode (if any).
+				 */
+				if (--ap->an_refcnt == 0)
+					anon_rmhash(ap);
+
+				mutex_exit(ahm);
+				(void) anon_set_ptr(ahp, index,
+				    NULL, ANON_SLEEP);
+				npurged++;
+				ANI_ADD(1);
+				kmem_cache_free(anon_cache, ap);
+			} else {
+				mutex_exit(ahm);
+			}
+
 			segadvstat.MADV_FREE_hit.value.ul++;
 
 			/*
@@ -1759,7 +1801,9 @@ anon_disclaim(struct anon_map *amp, ulong_t index, size_t size)
 			 */
 			(void) hat_pageunload(pp, HAT_FORCE_PGUNLOAD);
 			/*LINTED: constant in conditional context */
-			VN_DISPOSE(pp, B_FREE, 0, kcred);
+			VN_DISPOSE(pp,
+			    behav == MADV_FREE ? B_FREE : B_INVAL, 0, kcred);
+
 			anon_array_exit(&cookie);
 			continue;
 		}
@@ -1771,6 +1815,7 @@ anon_disclaim(struct anon_map *amp, ulong_t index, size_t size)
 				page_unlock(pp);
 				segadvstat.MADV_FREE_miss.value.ul++;
 				anon_array_exit(&cookie);
+				err = EBUSY;
 				continue;
 			} else {
 				pgcnt = 1;
@@ -1842,6 +1887,11 @@ skiplp:
 			page_unlock(pp);
 		anon_array_exit(&cookie);
 	}
+
+	if (purged != NULL)
+		*purged = npurged;
+
+	return (err);
 }
 
 /*
diff --git a/usr/src/uts/common/vm/vm_as.c b/usr/src/uts/common/vm/vm_as.c
index e0432f5b42..992254938f 100644
--- a/usr/src/uts/common/vm/vm_as.c
+++ b/usr/src/uts/common/vm/vm_as.c
@@ -57,6 +57,7 @@
 #include <sys/debug.h>
 #include <sys/tnf_probe.h>
 #include <sys/vtrace.h>
+#include <sys/ddi.h>
 
 #include <vm/hat.h>
 #include <vm/xhat.h>
@@ -880,6 +881,7 @@ as_fault(struct hat *hat, struct as *as, caddr_t addr, size_t size,
 	struct seg *segsav;
 	int as_lock_held;
 	klwp_t *lwp = ttolwp(curthread);
+	zone_t *zonep = curzone;
 	int is_xhat = 0;
 	int holding_wpage = 0;
 	extern struct seg_ops   segdev_ops;
@@ -929,6 +931,23 @@ retry:
 			if (as == &kas)
 				CPU_STATS_ADDQ(CPU, vm, kernel_asflt, 1);
 			CPU_STATS_EXIT_K();
+			if (zonep->zone_pg_flt_delay != 0) {
+				/*
+				 * The zone in which this process is running
+				 * is currently over it's physical memory cap.
+				 * Throttle page faults to help the user-land
+				 * memory capper catch up. Note that
+				 * drv_usectohz() rounds up.
+				 */
+				atomic_add_64(&zonep->zone_pf_throttle, 1);
+				atomic_add_64(&zonep->zone_pf_throttle_usec,
+				    zonep->zone_pg_flt_delay);
+				if (zonep->zone_pg_flt_delay < TICK_TO_USEC(1))
+					drv_usecwait(zonep->zone_pg_flt_delay);
+				else
+					delay(drv_usectohz(
+					    zonep->zone_pg_flt_delay));
+			}
 			break;
 		}
 	}
diff --git a/usr/src/uts/common/vm/vm_pvn.c b/usr/src/uts/common/vm/vm_pvn.c
index 1b8d12eb8d..a206320a30 100644
--- a/usr/src/uts/common/vm/vm_pvn.c
+++ b/usr/src/uts/common/vm/vm_pvn.c
@@ -20,6 +20,7 @@
  */
 /*
  * Copyright (c) 1986, 2010, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2012, Joyent, Inc. All rights reserved.
  * Copyright 2015 Nexenta Systems, Inc.  All rights reserved.
  */
 
@@ -432,7 +433,14 @@ pvn_write_done(page_t *plist, int flags)
 				page_io_unlock(pp);
 				page_unlock(pp);
 			}
-		} else if (flags & B_INVAL) {
+		} else if ((flags & (B_INVAL | B_INVALCURONLY)) == B_INVAL) {
+			/*
+			 * If B_INVALCURONLY is set, then we handle that case
+			 * in the next conditional if hat_page_is_mapped()
+			 * indicates that there are no additional mappings
+			 * to the page.
+			 */
+
 			/*
 			 * XXX - Failed writes with B_INVAL set are
 			 * not handled appropriately.
@@ -573,8 +581,9 @@ pvn_write_done(page_t *plist, int flags)
 }
 
 /*
- * Flags are composed of {B_ASYNC, B_INVAL, B_FREE, B_DONTNEED, B_DELWRI,
- * B_TRUNC, B_FORCE}.  B_DELWRI indicates that this page is part of a kluster
+ * Flags are composed of {B_ASYNC, B_INVAL, B_INVALCURONLY, B_FREE,
+ * B_DONTNEED, B_DELWRI, B_TRUNC, B_FORCE}.
+ * B_DELWRI indicates that this page is part of a kluster
  * operation and is only to be considered if it doesn't involve any
  * waiting here.  B_TRUNC indicates that the file is being truncated
  * and so no i/o needs to be done. B_FORCE indicates that the page
@@ -628,13 +637,17 @@ pvn_getdirty(page_t *pp, int flags)
 	 * If we want to free or invalidate the page then
 	 * we need to unload it so that anyone who wants
 	 * it will have to take a minor fault to get it.
+	 * If we are only invalidating the page for the
+	 * current process, then pass in a different flag.
 	 * Otherwise, we're just writing the page back so we
 	 * need to sync up the hardwre and software mod bit to
 	 * detect any future modifications.  We clear the
 	 * software mod bit when we put the page on the dirty
 	 * list.
 	 */
-	if (flags & (B_INVAL | B_FREE)) {
+	if (flags & B_INVALCURONLY) {
+		(void) hat_pageunload(pp, HAT_CURPROC_PGUNLOAD);
+	} else if (flags & (B_INVAL | B_FREE)) {
 		(void) hat_pageunload(pp, HAT_FORCE_PGUNLOAD);
 	} else {
 		(void) hat_pagesync(pp, HAT_SYNC_ZERORM);
@@ -646,7 +659,7 @@ pvn_getdirty(page_t *pp, int flags)
 		 * list after all.
 		 */
 		page_io_unlock(pp);
-		if (flags & B_INVAL) {
+		if ((flags & (B_INVAL | B_INVALCURONLY)) == B_INVAL) {
 			/*LINTED: constant in conditional context*/
 			VN_DISPOSE(pp, B_INVAL, 0, kcred);
 		} else if (flags & B_FREE) {
@@ -658,6 +671,9 @@ pvn_getdirty(page_t *pp, int flags)
 			 * of VOP_PUTPAGE() who prefer freeing the
 			 * page _only_ if no one else is accessing it.
 			 * E.g. segmap_release()
+			 * We also take this path for B_INVALCURONLY and
+			 * let page_release call VN_DISPOSE if no one else is
+			 * using the page.
 			 *
 			 * The above hat_ismod() check is useless because:
 			 * (1) we may not be holding SE_EXCL lock;
@@ -682,7 +698,7 @@ pvn_getdirty(page_t *pp, int flags)
 	 * We'll detect the fact that they used it when the
 	 * i/o is done and avoid freeing the page.
 	 */
-	if (flags & B_FREE)
+	if (flags & (B_FREE | B_INVALCURONLY))
 		page_downgrade(pp);
 
 
diff --git a/usr/src/uts/common/vm/vm_usage.c b/usr/src/uts/common/vm/vm_usage.c
index d422f8d0e8..33c5383f68 100644
--- a/usr/src/uts/common/vm/vm_usage.c
+++ b/usr/src/uts/common/vm/vm_usage.c
@@ -25,6 +25,10 @@
  */
 
 /*
+ * Copyright (c) 2014, Joyent, Inc. All rights reserved.
+ */
+
+/*
  * vm_usage
  *
  * This file implements the getvmusage() private system call.
@@ -114,7 +118,7 @@
  *	For accurate counting of map-shared and COW-shared pages.
  *
  *    - visited private anons (refcnt > 1) for each collective.
- *	(entity->vme_anon_hash)
+ *	(entity->vme_anon)
  *	For accurate counting of COW-shared pages.
  *
  * The common accounting structure is the vmu_entity_t, which represents
@@ -152,6 +156,7 @@
 #include <sys/vm_usage.h>
 #include <sys/zone.h>
 #include <sys/sunddi.h>
+#include <sys/sysmacros.h>
 #include <sys/avl.h>
 #include <vm/anon.h>
 #include <vm/as.h>
@@ -199,6 +204,14 @@ typedef struct vmu_object {
 } vmu_object_t;
 
 /*
+ * Node for tree of visited COW anons.
+ */
+typedef struct vmu_anon {
+	avl_node_t vma_node;
+	uintptr_t vma_addr;
+} vmu_anon_t;
+
+/*
  * Entity by which to count results.
  *
  * The entity structure keeps the current rss/swap counts for each entity
@@ -221,7 +234,7 @@ typedef struct vmu_entity {
 	struct vmu_entity *vme_next_calc;
 	mod_hash_t	*vme_vnode_hash; /* vnodes visited for entity */
 	mod_hash_t	*vme_amp_hash;	 /* shared amps visited for entity */
-	mod_hash_t	*vme_anon_hash;	 /* COW anons visited for entity */
+	avl_tree_t	vme_anon;	 /* COW anons visited for entity */
 	vmusage_t	vme_result;	 /* identifies entity and results */
 } vmu_entity_t;
 
@@ -324,6 +337,23 @@ bounds_cmp(const void *bnd1, const void *bnd2)
 }
 
 /*
+ * Comparison routine for our AVL tree of anon structures.
+ */
+static int
+vmu_anon_cmp(const void *lhs, const void *rhs)
+{
+	const vmu_anon_t *l = lhs, *r = rhs;
+
+	if (l->vma_addr == r->vma_addr)
+		return (0);
+
+	if (l->vma_addr < r->vma_addr)
+		return (-1);
+
+	return (1);
+}
+
+/*
  * Save a bound on the free list.
  */
 static void
@@ -363,13 +393,18 @@ static void
 vmu_free_entity(mod_hash_val_t val)
 {
 	vmu_entity_t *entity = (vmu_entity_t *)val;
+	vmu_anon_t *anon;
+	void *cookie = NULL;
 
 	if (entity->vme_vnode_hash != NULL)
 		i_mod_hash_clear_nosync(entity->vme_vnode_hash);
 	if (entity->vme_amp_hash != NULL)
 		i_mod_hash_clear_nosync(entity->vme_amp_hash);
-	if (entity->vme_anon_hash != NULL)
-		i_mod_hash_clear_nosync(entity->vme_anon_hash);
+
+	while ((anon = avl_destroy_nodes(&entity->vme_anon, &cookie)) != NULL)
+		kmem_free(anon, sizeof (vmu_anon_t));
+
+	avl_destroy(&entity->vme_anon);
 
 	entity->vme_next = vmu_data.vmu_free_entities;
 	vmu_data.vmu_free_entities = entity;
@@ -485,10 +520,10 @@ vmu_alloc_entity(id_t id, int type, id_t zoneid)
 		    "vmusage amp hash", VMUSAGE_HASH_SIZE, vmu_free_object,
 		    sizeof (struct anon_map));
 
-	if (entity->vme_anon_hash == NULL)
-		entity->vme_anon_hash = mod_hash_create_ptrhash(
-		    "vmusage anon hash", VMUSAGE_HASH_SIZE,
-		    mod_hash_null_valdtor, sizeof (struct anon));
+	VERIFY(avl_first(&entity->vme_anon) == NULL);
+
+	avl_create(&entity->vme_anon, vmu_anon_cmp, sizeof (struct vmu_anon),
+	    offsetof(struct vmu_anon, vma_node));
 
 	entity->vme_next = vmu_data.vmu_entities;
 	vmu_data.vmu_entities = entity;
@@ -518,7 +553,8 @@ vmu_alloc_zone(id_t id)
 
 	zone->vmz_id = id;
 
-	if ((vmu_data.vmu_calc_flags & (VMUSAGE_ZONE | VMUSAGE_ALL_ZONES)) != 0)
+	if ((vmu_data.vmu_calc_flags &
+	    (VMUSAGE_ZONE | VMUSAGE_ALL_ZONES | VMUSAGE_A_ZONE)) != 0)
 		zone->vmz_zone = vmu_alloc_entity(id, VMUSAGE_ZONE, id);
 
 	if ((vmu_data.vmu_calc_flags & (VMUSAGE_PROJECTS |
@@ -613,21 +649,19 @@ vmu_find_insert_object(mod_hash_t *hash, caddr_t key, uint_t type)
 }
 
 static int
-vmu_find_insert_anon(mod_hash_t *hash, caddr_t key)
+vmu_find_insert_anon(vmu_entity_t *entity, void *key)
 {
-	int ret;
-	caddr_t val;
+	vmu_anon_t anon, *ap;
 
-	ret = i_mod_hash_find_nosync(hash, (mod_hash_key_t)key,
-	    (mod_hash_val_t *)&val);
+	anon.vma_addr = (uintptr_t)key;
 
-	if (ret == 0)
+	if (avl_find(&entity->vme_anon, &anon, NULL) != NULL)
 		return (0);
 
-	ret = i_mod_hash_insert_nosync(hash, (mod_hash_key_t)key,
-	    (mod_hash_val_t)key, (mod_hash_hndl_t)0);
+	ap = kmem_alloc(sizeof (vmu_anon_t), KM_SLEEP);
+	ap->vma_addr = (uintptr_t)key;
 
-	ASSERT(ret == 0);
+	avl_add(&entity->vme_anon, ap);
 
 	return (1);
 }
@@ -918,6 +952,8 @@ vmu_amp_update_incore_bounds(avl_tree_t *tree, struct anon_map *amp,
 			next = AVL_NEXT(tree, next);
 			continue;
 		}
+
+		ASSERT(next->vmb_type == VMUSAGE_BOUND_UNKNOWN);
 		bound_type = next->vmb_type;
 		index = next->vmb_start;
 		while (index <= next->vmb_end) {
@@ -937,7 +973,10 @@ vmu_amp_update_incore_bounds(avl_tree_t *tree, struct anon_map *amp,
 
 			if (ap != NULL && vn != NULL && vn->v_pages != NULL &&
 			    (page = page_exists(vn, off)) != NULL) {
-				page_type = VMUSAGE_BOUND_INCORE;
+				if (PP_ISFREE(page))
+					page_type = VMUSAGE_BOUND_NOT_INCORE;
+				else
+					page_type = VMUSAGE_BOUND_INCORE;
 				if (page->p_szc > 0) {
 					pgcnt = page_get_pagecnt(page->p_szc);
 					pgshft = page_get_shift(page->p_szc);
@@ -947,8 +986,10 @@ vmu_amp_update_incore_bounds(avl_tree_t *tree, struct anon_map *amp,
 			} else {
 				page_type = VMUSAGE_BOUND_NOT_INCORE;
 			}
+
 			if (bound_type == VMUSAGE_BOUND_UNKNOWN) {
 				next->vmb_type = page_type;
+				bound_type = page_type;
 			} else if (next->vmb_type != page_type) {
 				/*
 				 * If current bound type does not match page
@@ -1009,6 +1050,7 @@ vmu_vnode_update_incore_bounds(avl_tree_t *tree, vnode_t *vnode,
 			continue;
 		}
 
+		ASSERT(next->vmb_type == VMUSAGE_BOUND_UNKNOWN);
 		bound_type = next->vmb_type;
 		index = next->vmb_start;
 		while (index <= next->vmb_end) {
@@ -1024,7 +1066,10 @@ vmu_vnode_update_incore_bounds(avl_tree_t *tree, vnode_t *vnode,
 
 			if (vnode->v_pages != NULL &&
 			    (page = page_exists(vnode, ptob(index))) != NULL) {
-				page_type = VMUSAGE_BOUND_INCORE;
+				if (PP_ISFREE(page))
+					page_type = VMUSAGE_BOUND_NOT_INCORE;
+				else
+					page_type = VMUSAGE_BOUND_INCORE;
 				if (page->p_szc > 0) {
 					pgcnt = page_get_pagecnt(page->p_szc);
 					pgshft = page_get_shift(page->p_szc);
@@ -1034,8 +1079,10 @@ vmu_vnode_update_incore_bounds(avl_tree_t *tree, vnode_t *vnode,
 			} else {
 				page_type = VMUSAGE_BOUND_NOT_INCORE;
 			}
+
 			if (bound_type == VMUSAGE_BOUND_UNKNOWN) {
 				next->vmb_type = page_type;
+				bound_type = page_type;
 			} else if (next->vmb_type != page_type) {
 				/*
 				 * If current bound type does not match page
@@ -1304,6 +1351,12 @@ vmu_calculate_seg(vmu_entity_t *vmu_entities, struct seg *seg)
 			}
 
 			/*
+			 * Pages on the free list aren't counted for the rss.
+			 */
+			if (PP_ISFREE(page))
+				continue;
+
+			/*
 			 * Assume anon structs with a refcnt
 			 * of 1 are not COW shared, so there
 			 * is no reason to track them per entity.
@@ -1320,8 +1373,7 @@ vmu_calculate_seg(vmu_entity_t *vmu_entities, struct seg *seg)
 				 * Track COW anons per entity so
 				 * they are not double counted.
 				 */
-				if (vmu_find_insert_anon(entity->vme_anon_hash,
-				    (caddr_t)ap) == 0)
+				if (vmu_find_insert_anon(entity, ap) == 0)
 					continue;
 
 				result->vmu_rss_all += (pgcnt << PAGESHIFT);
@@ -1461,8 +1513,9 @@ vmu_calculate_proc(proc_t *p)
 		entities = tmp;
 	}
 	if (vmu_data.vmu_calc_flags &
-	    (VMUSAGE_ZONE | VMUSAGE_ALL_ZONES | VMUSAGE_PROJECTS |
-	    VMUSAGE_ALL_PROJECTS | VMUSAGE_TASKS | VMUSAGE_ALL_TASKS |
+	    (VMUSAGE_ZONE | VMUSAGE_ALL_ZONES | VMUSAGE_A_ZONE |
+	    VMUSAGE_PROJECTS | VMUSAGE_ALL_PROJECTS |
+	    VMUSAGE_TASKS | VMUSAGE_ALL_TASKS |
 	    VMUSAGE_RUSERS | VMUSAGE_ALL_RUSERS | VMUSAGE_EUSERS |
 	    VMUSAGE_ALL_EUSERS)) {
 		ret = i_mod_hash_find_nosync(vmu_data.vmu_zones_hash,
@@ -1594,8 +1647,7 @@ vmu_free_extra()
 			mod_hash_destroy_hash(te->vme_vnode_hash);
 		if (te->vme_amp_hash != NULL)
 			mod_hash_destroy_hash(te->vme_amp_hash);
-		if (te->vme_anon_hash != NULL)
-			mod_hash_destroy_hash(te->vme_anon_hash);
+		VERIFY(avl_first(&te->vme_anon) == NULL);
 		kmem_free(te, sizeof (vmu_entity_t));
 	}
 	while (vmu_data.vmu_free_zones != NULL) {
@@ -1739,12 +1791,34 @@ vmu_cache_rele(vmu_cache_t *cache)
 }
 
 /*
+ * When new data is calculated, update the phys_mem rctl usage value in the
+ * zones.
+ */
+static void
+vmu_update_zone_rctls(vmu_cache_t *cache)
+{
+	vmusage_t	*rp;
+	size_t		i = 0;
+	zone_t		*zp;
+
+	for (rp = cache->vmc_results; i < cache->vmc_nresults; rp++, i++) {
+		if (rp->vmu_type == VMUSAGE_ZONE &&
+		    rp->vmu_zoneid != ALL_ZONES) {
+			if ((zp = zone_find_by_id(rp->vmu_zoneid)) != NULL) {
+				zp->zone_phys_mem = rp->vmu_rss_all;
+				zone_rele(zp);
+			}
+		}
+	}
+}
+
+/*
  * Copy out the cached results to a caller.  Inspect the callers flags
  * and zone to determine which cached results should be copied.
  */
 static int
 vmu_copyout_results(vmu_cache_t *cache, vmusage_t *buf, size_t *nres,
-    uint_t flags, int cpflg)
+    uint_t flags, id_t req_zone_id, int cpflg)
 {
 	vmusage_t *result, *out_result;
 	vmusage_t dummy;
@@ -1763,7 +1837,7 @@ vmu_copyout_results(vmu_cache_t *cache, vmusage_t *buf, size_t *nres,
 	/* figure out what results the caller is interested in. */
 	if ((flags & VMUSAGE_SYSTEM) && curproc->p_zone == global_zone)
 		types |= VMUSAGE_SYSTEM;
-	if (flags & (VMUSAGE_ZONE | VMUSAGE_ALL_ZONES))
+	if (flags & (VMUSAGE_ZONE | VMUSAGE_ALL_ZONES | VMUSAGE_A_ZONE))
 		types |= VMUSAGE_ZONE;
 	if (flags & (VMUSAGE_PROJECTS | VMUSAGE_ALL_PROJECTS |
 	    VMUSAGE_COL_PROJECTS))
@@ -1826,26 +1900,33 @@ vmu_copyout_results(vmu_cache_t *cache, vmusage_t *buf, size_t *nres,
 				continue;
 		}
 
-		/* Skip "other zone" results if not requested */
-		if (result->vmu_zoneid != curproc->p_zone->zone_id) {
-			if (result->vmu_type == VMUSAGE_ZONE &&
-			    (flags & VMUSAGE_ALL_ZONES) == 0)
-				continue;
-			if (result->vmu_type == VMUSAGE_PROJECTS &&
-			    (flags & (VMUSAGE_ALL_PROJECTS |
-			    VMUSAGE_COL_PROJECTS)) == 0)
-				continue;
-			if (result->vmu_type == VMUSAGE_TASKS &&
-			    (flags & VMUSAGE_ALL_TASKS) == 0)
-				continue;
-			if (result->vmu_type == VMUSAGE_RUSERS &&
-			    (flags & (VMUSAGE_ALL_RUSERS |
-			    VMUSAGE_COL_RUSERS)) == 0)
-				continue;
-			if (result->vmu_type == VMUSAGE_EUSERS &&
-			    (flags & (VMUSAGE_ALL_EUSERS |
-			    VMUSAGE_COL_EUSERS)) == 0)
+		if (result->vmu_type == VMUSAGE_ZONE &&
+		    flags & VMUSAGE_A_ZONE) {
+			/* Skip non-requested zone results */
+			if (result->vmu_zoneid != req_zone_id)
 				continue;
+		} else {
+			/* Skip "other zone" results if not requested */
+			if (result->vmu_zoneid != curproc->p_zone->zone_id) {
+				if (result->vmu_type == VMUSAGE_ZONE &&
+				    (flags & VMUSAGE_ALL_ZONES) == 0)
+					continue;
+				if (result->vmu_type == VMUSAGE_PROJECTS &&
+				    (flags & (VMUSAGE_ALL_PROJECTS |
+				    VMUSAGE_COL_PROJECTS)) == 0)
+					continue;
+				if (result->vmu_type == VMUSAGE_TASKS &&
+				    (flags & VMUSAGE_ALL_TASKS) == 0)
+					continue;
+				if (result->vmu_type == VMUSAGE_RUSERS &&
+				    (flags & (VMUSAGE_ALL_RUSERS |
+				    VMUSAGE_COL_RUSERS)) == 0)
+					continue;
+				if (result->vmu_type == VMUSAGE_EUSERS &&
+				    (flags & (VMUSAGE_ALL_EUSERS |
+				    VMUSAGE_COL_EUSERS)) == 0)
+					continue;
+			}
 		}
 		count++;
 		if (out_result != NULL) {
@@ -1901,10 +1982,12 @@ vm_getusage(uint_t flags, time_t age, vmusage_t *buf, size_t *nres, int cpflg)
 	int cacherecent = 0;
 	hrtime_t now;
 	uint_t flags_orig;
+	id_t req_zone_id;
 
 	/*
 	 * Non-global zones cannot request system wide and/or collated
-	 * results, or the system result, so munge the flags accordingly.
+	 * results, or the system result, or usage of another zone, so munge
+	 * the flags accordingly.
 	 */
 	flags_orig = flags;
 	if (curproc->p_zone != global_zone) {
@@ -1924,6 +2007,10 @@ vm_getusage(uint_t flags, time_t age, vmusage_t *buf, size_t *nres, int cpflg)
 			flags &= ~VMUSAGE_SYSTEM;
 			flags |= VMUSAGE_ZONE;
 		}
+		if (flags & VMUSAGE_A_ZONE) {
+			flags &= ~VMUSAGE_A_ZONE;
+			flags |= VMUSAGE_ZONE;
+		}
 	}
 
 	/* Check for unknown flags */
@@ -1934,6 +2021,21 @@ vm_getusage(uint_t flags, time_t age, vmusage_t *buf, size_t *nres, int cpflg)
 	if ((flags & VMUSAGE_MASK) == 0)
 		return (set_errno(EINVAL));
 
+	/* If requesting results for a specific zone, get the zone ID */
+	if (flags & VMUSAGE_A_ZONE) {
+		size_t bufsize;
+		vmusage_t zreq;
+
+		if (ddi_copyin((caddr_t)nres, &bufsize, sizeof (size_t), cpflg))
+			return (set_errno(EFAULT));
+		/* Requested zone ID is passed in buf, so 0 len not allowed */
+		if (bufsize == 0)
+			return (set_errno(EINVAL));
+		if (ddi_copyin((caddr_t)buf, &zreq, sizeof (vmusage_t), cpflg))
+			return (set_errno(EFAULT));
+		req_zone_id = zreq.vmu_id;
+	}
+
 	mutex_enter(&vmu_data.vmu_lock);
 	now = gethrtime();
 
@@ -1953,7 +2055,7 @@ start:
 			mutex_exit(&vmu_data.vmu_lock);
 
 			ret = vmu_copyout_results(cache, buf, nres, flags_orig,
-			    cpflg);
+			    req_zone_id, cpflg);
 			mutex_enter(&vmu_data.vmu_lock);
 			vmu_cache_rele(cache);
 			if (vmu_data.vmu_pending_waiters > 0)
@@ -2009,8 +2111,11 @@ start:
 
 		mutex_exit(&vmu_data.vmu_lock);
 
+		/* update zone's phys. mem. rctl usage */
+		vmu_update_zone_rctls(cache);
 		/* copy cache */
-		ret = vmu_copyout_results(cache, buf, nres, flags_orig, cpflg);
+		ret = vmu_copyout_results(cache, buf, nres, flags_orig,
+		    req_zone_id, cpflg);
 		mutex_enter(&vmu_data.vmu_lock);
 		vmu_cache_rele(cache);
 		mutex_exit(&vmu_data.vmu_lock);
@@ -2030,3 +2135,185 @@ start:
 	vmu_data.vmu_pending_waiters--;
 	goto start;
 }
+
+#if defined(__x86)
+/*
+ * Attempt to invalidate all of the pages in the mapping for the given process.
+ */
+static void
+map_inval(proc_t *p, struct seg *seg, caddr_t addr, size_t size)
+{
+	page_t		*pp;
+	size_t		psize;
+	u_offset_t	off;
+	caddr_t		eaddr;
+	struct vnode	*vp;
+	struct segvn_data *svd;
+	struct hat	*victim_hat;
+
+	ASSERT((addr + size) <= (seg->s_base + seg->s_size));
+
+	victim_hat = p->p_as->a_hat;
+	svd = (struct segvn_data *)seg->s_data;
+	vp = svd->vp;
+	psize = page_get_pagesize(seg->s_szc);
+
+	off = svd->offset + (uintptr_t)(addr - seg->s_base);
+
+	for (eaddr = addr + size; addr < eaddr; addr += psize, off += psize) {
+		pp = page_lookup_nowait(vp, off, SE_SHARED);
+
+		if (pp != NULL) {
+			/* following logic based on pvn_getdirty() */
+
+			if (pp->p_lckcnt != 0 || pp->p_cowcnt != 0) {
+				page_unlock(pp);
+				continue;
+			}
+
+			page_io_lock(pp);
+			hat_page_inval(pp, 0, victim_hat);
+			page_io_unlock(pp);
+
+			/*
+			 * For B_INVALCURONLY-style handling we let
+			 * page_release call VN_DISPOSE if no one else is using
+			 * the page.
+			 *
+			 * A hat_ismod() check would be useless because:
+			 * (1) we are not be holding SE_EXCL lock
+			 * (2) we've not unloaded _all_ translations
+			 *
+			 * Let page_release() do the heavy-lifting.
+			 */
+			(void) page_release(pp, 1);
+		}
+	}
+}
+
+/*
+ * vm_map_inval()
+ *
+ * Invalidate as many pages as possible within the given mapping for the given
+ * process. addr is expected to be the base address of the mapping and size is
+ * the length of the mapping. In some cases a mapping will encompass an
+ * entire segment, but at least for anon or stack mappings, these will be
+ * regions within a single large segment. Thus, the invalidation is oriented
+ * around a single mapping and not an entire segment.
+ *
+ * SPARC sfmmu hat does not support HAT_CURPROC_PGUNLOAD-style handling so
+ * this code is only applicable to x86.
+ */
+int
+vm_map_inval(pid_t pid, caddr_t addr, size_t size)
+{
+	int ret;
+	int error = 0;
+	proc_t *p;		/* target proc */
+	struct as *as;		/* target proc's address space */
+	struct seg *seg;	/* working segment */
+
+	if (curproc->p_zone != global_zone || crgetruid(curproc->p_cred) != 0)
+		return (set_errno(EPERM));
+
+	/* If not a valid mapping address, return an error */
+	if ((caddr_t)((uintptr_t)addr & (uintptr_t)PAGEMASK) != addr)
+		return (set_errno(EINVAL));
+
+again:
+	mutex_enter(&pidlock);
+	p = prfind(pid);
+	if (p == NULL) {
+		mutex_exit(&pidlock);
+		return (set_errno(ESRCH));
+	}
+
+	mutex_enter(&p->p_lock);
+	mutex_exit(&pidlock);
+
+	if (panicstr != NULL) {
+		mutex_exit(&p->p_lock);
+		return (0);
+	}
+
+	as = p->p_as;
+
+	/*
+	 * Try to set P_PR_LOCK - prevents process "changing shape"
+	 * - blocks fork
+	 * - blocks sigkill
+	 * - cannot be a system proc
+	 * - must be fully created proc
+	 */
+	ret = sprtrylock_proc(p);
+	if (ret == -1) {
+		/* Process in invalid state */
+		mutex_exit(&p->p_lock);
+		return (set_errno(ESRCH));
+	}
+
+	if (ret == 1) {
+		/*
+		 * P_PR_LOCK is already set. Wait and try again. This also
+		 * drops p_lock so p may no longer be valid since the proc may
+		 * have exited.
+		 */
+		sprwaitlock_proc(p);
+		goto again;
+	}
+
+	/* P_PR_LOCK is now set */
+	mutex_exit(&p->p_lock);
+
+	AS_LOCK_ENTER(as, &as->a_lock, RW_READER);
+	if ((seg = as_segat(as, addr)) == NULL) {
+		AS_LOCK_EXIT(as, &as->a_lock);
+		mutex_enter(&p->p_lock);
+		sprunlock(p);
+		return (set_errno(ENOMEM));
+	}
+
+	/*
+	 * The invalidation behavior only makes sense for vnode-backed segments.
+	 */
+	if (seg->s_ops != &segvn_ops) {
+		AS_LOCK_EXIT(as, &as->a_lock);
+		mutex_enter(&p->p_lock);
+		sprunlock(p);
+		return (0);
+	}
+
+	/*
+	 * If the mapping is out of bounds of the segement return an error.
+	 */
+	if ((addr + size) > (seg->s_base + seg->s_size)) {
+		AS_LOCK_EXIT(as, &as->a_lock);
+		mutex_enter(&p->p_lock);
+		sprunlock(p);
+		return (set_errno(EINVAL));
+	}
+
+	/*
+	 * Don't use MS_INVALCURPROC flag here since that would eventually
+	 * initiate hat invalidation based on curthread. Since we're doing this
+	 * on behalf of a different process, that would erroneously invalidate
+	 * our own process mappings.
+	 */
+	error = SEGOP_SYNC(seg, addr, size, 0, (uint_t)MS_ASYNC);
+	if (error == 0) {
+		/*
+		 * Since we didn't invalidate during the sync above, we now
+		 * try to invalidate all of the pages in the mapping.
+		 */
+		map_inval(p, seg, addr, size);
+	}
+	AS_LOCK_EXIT(as, &as->a_lock);
+
+	mutex_enter(&p->p_lock);
+	sprunlock(p);
+
+	if (error)
+		(void) set_errno(error);
+	return (error);
+}
+#endif