summaryrefslogtreecommitdiff
path: root/usr/src/uts/common/sys
diff options
context:
space:
mode:
Diffstat (limited to 'usr/src/uts/common/sys')
-rw-r--r--usr/src/uts/common/sys/Makefile13
-rw-r--r--usr/src/uts/common/sys/aggr_impl.h3
-rw-r--r--usr/src/uts/common/sys/auxv.h17
-rw-r--r--usr/src/uts/common/sys/brand.h12
-rw-r--r--usr/src/uts/common/sys/buf.h8
-rw-r--r--usr/src/uts/common/sys/cpucaps.h5
-rw-r--r--usr/src/uts/common/sys/cpucaps_impl.h6
-rw-r--r--usr/src/uts/common/sys/cred.h1
-rw-r--r--usr/src/uts/common/sys/ctf_api.h30
-rw-r--r--usr/src/uts/common/sys/devpoll.h18
-rw-r--r--usr/src/uts/common/sys/dktp/dadk.h2
-rw-r--r--usr/src/uts/common/sys/dld.h7
-rw-r--r--usr/src/uts/common/sys/dld_impl.h5
-rw-r--r--usr/src/uts/common/sys/dlpi.h11
-rw-r--r--usr/src/uts/common/sys/dls.h10
-rw-r--r--usr/src/uts/common/sys/dls_impl.h6
-rw-r--r--usr/src/uts/common/sys/dls_mgmt.h2
-rw-r--r--usr/src/uts/common/sys/epoll.h89
-rw-r--r--usr/src/uts/common/sys/eventfd.h68
-rw-r--r--usr/src/uts/common/sys/exec.h12
-rw-r--r--usr/src/uts/common/sys/frameio.h107
-rw-r--r--usr/src/uts/common/sys/fs/bootfs_impl.h81
-rw-r--r--usr/src/uts/common/sys/fs/hyprlofs.h91
-rw-r--r--usr/src/uts/common/sys/fs/hyprlofs_info.h174
-rw-r--r--usr/src/uts/common/sys/fs/sdev_impl.h61
-rw-r--r--usr/src/uts/common/sys/fs/sdev_plugin.h106
-rw-r--r--usr/src/uts/common/sys/fs/tmp.h39
-rw-r--r--usr/src/uts/common/sys/gsqueue.h65
-rw-r--r--usr/src/uts/common/sys/inotify.h153
-rw-r--r--usr/src/uts/common/sys/mac.h1
-rw-r--r--usr/src/uts/common/sys/mac_client_impl.h27
-rw-r--r--usr/src/uts/common/sys/mac_impl.h2
-rw-r--r--usr/src/uts/common/sys/mman.h2
-rw-r--r--usr/src/uts/common/sys/mntent.h2
-rw-r--r--usr/src/uts/common/sys/neti.h2
-rw-r--r--usr/src/uts/common/sys/netstack.h3
-rw-r--r--usr/src/uts/common/sys/param.h2
-rw-r--r--usr/src/uts/common/sys/policy.h1
-rw-r--r--usr/src/uts/common/sys/poll.h13
-rw-r--r--usr/src/uts/common/sys/poll_impl.h10
-rw-r--r--usr/src/uts/common/sys/proc.h15
-rw-r--r--usr/src/uts/common/sys/ptms.h19
-rw-r--r--usr/src/uts/common/sys/resource.h2
-rw-r--r--usr/src/uts/common/sys/socket.h7
-rw-r--r--usr/src/uts/common/sys/squeue.h14
-rw-r--r--usr/src/uts/common/sys/squeue_impl.h2
-rw-r--r--usr/src/uts/common/sys/sunddi.h1
-rw-r--r--usr/src/uts/common/sys/systrace.h13
-rw-r--r--usr/src/uts/common/sys/termios.h18
-rw-r--r--usr/src/uts/common/sys/thread.h8
-rw-r--r--usr/src/uts/common/sys/uadmin.h3
-rw-r--r--usr/src/uts/common/sys/vm_usage.h5
-rw-r--r--usr/src/uts/common/sys/vmsystm.h5
-rw-r--r--usr/src/uts/common/sys/vnd.h141
-rw-r--r--usr/src/uts/common/sys/vnd_errno.h72
-rw-r--r--usr/src/uts/common/sys/vnode.h8
-rw-r--r--usr/src/uts/common/sys/zfd.h53
-rw-r--r--usr/src/uts/common/sys/zone.h128
58 files changed, 1700 insertions, 81 deletions
diff --git a/usr/src/uts/common/sys/Makefile b/usr/src/uts/common/sys/Makefile
index 656494c228..6f28942704 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 (c) 2014, 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 \
@@ -635,6 +640,8 @@ CHKHDRS= \
vmem.h \
vmem_impl.h \
vmsystm.h \
+ vnd.h \
+ vnd_errno.h \
vnic.h \
vnic_impl.h \
vnode.h \
@@ -652,6 +659,7 @@ CHKHDRS= \
watchpoint.h \
winlockio.h \
zcons.h \
+ zfd.h \
zone.h \
xti_inet.h \
xti_osi.h \
@@ -852,13 +860,14 @@ 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 +877,8 @@ 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..3486ae864d 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 (c) 2014, Joyent, Inc. All rights reserved.
*/
#ifndef _SYS_BRAND_H
@@ -106,7 +107,7 @@ 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);
@@ -124,7 +125,13 @@ struct brand_ops {
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 *, void *);
+ 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 *);
};
/*
@@ -135,6 +142,7 @@ 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;
@@ -179,7 +187,7 @@ 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 *);
+ struct brand *, char *, char *, char *);
extern void brand_solaris_exec(struct brand *);
extern int brand_solaris_fini(char **, struct modlinkage *,
struct brand *);
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..ab648d5489 100644
--- a/usr/src/uts/common/sys/ctf_api.h
+++ b/usr/src/uts/common/sys/ctf_api.h
@@ -169,7 +169,9 @@ 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);
@@ -240,6 +242,34 @@ extern ctf_file_t *ctf_modopen(struct module *, int *);
#endif
+typedef struct ctf_ihelem {
+ ushort_t ih_type; /* type ID number */
+ ushort_t ih_value; /* type ID number in a different fp */
+ ushort_t ih_next; /* index of next element in hash chain */
+} ctf_ihelem_t;
+
+typedef struct ctf_idhash {
+ ushort_t *ih_buckets; /* hash bucket array (chain indices) */
+ ctf_ihelem_t *ih_chains; /* hash chains buffer */
+ ushort_t ih_nbuckets; /* number of elements in bucket array */
+ ushort_t ih_nelems; /* number of elements in hash table */
+ uint_t ih_free; /* index of next free hash element */
+} ctf_idhash_t;
+
+extern int ctf_idhash_create(ctf_idhash_t *, ulong_t);
+extern void ctf_idhash_clear(ctf_idhash_t *);
+extern int ctf_idhash_define(ctf_idhash_t *, ushort_t, ushort_t);
+extern int ctf_idhash_insert(ctf_idhash_t *, ushort_t, ushort_t);
+extern ctf_ihelem_t *ctf_idhash_lookup(ctf_idhash_t *, ushort_t);
+extern uint_t ctf_idhash_size(const ctf_idhash_t *);
+extern void ctf_idhash_destroy(ctf_idhash_t *);
+
+typedef struct ctf_idhash_iter ctf_idhash_iter_t;
+extern int ctf_idhash_iter_init(ctf_idhash_t *, ctf_idhash_iter_t **);
+extern const ctf_ihelem_t *ctf_idhash_iter(ctf_idhash_t *,
+ ctf_idhash_iter_t *);
+extern void ctf_idhash_iter_fini(ctf_idhash_t *, ctf_idhash_iter_t *);
+
#ifdef __cplusplus
}
#endif
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/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..4f73d92118 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 (c) 2011, Joyent Inc. All rights reserved.
*/
#ifndef _DLS_MGMT_H
@@ -165,6 +166,7 @@ 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..c7dbaf947f
--- /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) 2014 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_INOTIFY 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..50786aca30 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 (c) 2014, Joyent, Inc. All rights reserved.
+ */
+
#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;
/*
@@ -239,14 +245,16 @@ extern void exec_set_sp(size_t);
extern int elfexec(vnode_t *, execa_t *, uarg_t *, intpdata_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);
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/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..4a6e86bdd9 100644
--- a/usr/src/uts/common/sys/fs/tmp.h
+++ b/usr/src/uts/common/sys/fs/tmp.h
@@ -68,29 +68,28 @@ 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);
@@ -101,8 +100,6 @@ 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 *,
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/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/mac.h b/usr/src/uts/common/sys/mac.h
index bd09077f1b..5803ad58d4 100644
--- a/usr/src/uts/common/sys/mac.h
+++ b/usr/src/uts/common/sys/mac.h
@@ -213,6 +213,7 @@ 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 ae25df6a0d..ce1a6f1d7c 100644
--- a/usr/src/uts/common/sys/mac_client_impl.h
+++ b/usr/src/uts/common/sys/mac_client_impl.h
@@ -22,6 +22,9 @@
* Copyright 2010 Sun Microsystems, Inc. All rights reserved.
* Use is subject to license terms.
*/
+/*
+ * Copyright (c) 2013, Joyent, Inc. All rights reserved.
+ */
#ifndef _SYS_MAC_CLIENT_IMPL_H
#define _SYS_MAC_CLIENT_IMPL_H
@@ -137,6 +140,7 @@ struct mac_client_impl_s { /* Protected by */
flow_entry_t *mci_flent_list; /* mci_rw_lock */
uint_t mci_nflents; /* mci_rw_lock */
uint_t mci_nvids; /* mci_rw_lock */
+ volatile uint32_t mci_vidcache; /* VID cache */
/* Resource Management Functions */
mac_resource_add_t mci_resource_add; /* SL */
@@ -280,6 +284,28 @@ extern int mac_tx_percpu_cnt;
} \
}
+/*
+ * To allow the hot path to not grab any additional locks, we keep a single
+ * entry VLAD ID cache that caches whether or not a given VID belongs to a
+ * MAC client.
+ */
+#define MCIP_VIDCACHE_VALIDSHIFT 31
+#define MCIP_VIDCACHE_VIDSHIFT 1
+#define MCIP_VIDCACHE_VIDMASK (UINT16_MAX << MCIP_VIDCACHE_VIDSHIFT)
+#define MCIP_VIDCACHE_BOOLSHIFT 0
+
+#define MCIP_VIDCACHE_INVALID 0
+
+#define MCIP_VIDCACHE_CACHE(vid, bool) \
+ ((1U << MCIP_VIDCACHE_VALIDSHIFT) | \
+ ((vid) << MCIP_VIDCACHE_VIDSHIFT) | \
+ ((bool) ? (1U << MCIP_VIDCACHE_BOOLSHIFT) : 0))
+
+#define MCIP_VIDCACHE_ISVALID(v) ((v) & (1U << MCIP_VIDCACHE_VALIDSHIFT))
+#define MCIP_VIDCACHE_VID(v) \
+ (((v) & MCIP_VIDCACHE_VIDMASK) >> MCIP_VIDCACHE_VIDSHIFT)
+#define MCIP_VIDCACHE_BOOL(v) ((v) & (1U << MCIP_VIDCACHE_BOOLSHIFT))
+
#define MAC_TAG_NEEDED(mcip) \
(((mcip)->mci_state_flags & MCIS_TAG_DISABLE) == 0 && \
(mcip)->mci_nvids == 1) \
@@ -302,6 +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_impl.h b/usr/src/uts/common/sys/mac_impl.h
index 3c9c6b77a4..d19c19e43d 100644
--- a/usr/src/uts/common/sys/mac_impl.h
+++ b/usr/src/uts/common/sys/mac_impl.h
@@ -894,6 +894,8 @@ 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/mman.h b/usr/src/uts/common/sys/mman.h
index a4cca68098..8f4cd1639f 100644
--- a/usr/src/uts/common/sys/mman.h
+++ b/usr/src/uts/common/sys/mman.h
@@ -25,6 +25,7 @@
*
* Copyright 2008 Sun Microsystems, Inc. All rights reserved.
* Use is subject to license terms.
+ * Copyright 2012 Joyent, Inc. All rights reserved.
*/
/* Copyright (c) 1983, 1984, 1985, 1986, 1987, 1988, 1989 AT&T */
@@ -330,6 +331,7 @@ 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/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/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..597c166b31 100644
--- a/usr/src/uts/common/sys/policy.h
+++ b/usr/src/uts/common/sys/policy.h
@@ -173,6 +173,7 @@ 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..134cb9132b 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 +347,12 @@ 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 */
+ union {
+ void *__brand_data;
+ int __exit_data;
+ } __p_brand_data;
/* additional lock to protect p_sessp (but not its contents) */
kmutex_t p_splock;
@@ -361,7 +367,8 @@ typedef struct proc {
*/
struct user p_user; /* (see sys/user.h) */
} proc_t;
-
+#define p_brand_data __p_brand_data.__brand_data
+#define p_exit_data __p_brand_data.__exit_data
#define PROC_T /* headers relying on proc_t are OK */
#ifdef _KERNEL
@@ -716,6 +723,10 @@ 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/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/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/socket.h b/usr/src/uts/common/sys/socket.h
index 590551f6f2..bbe2121d0e 100644
--- a/usr/src/uts/common/sys/socket.h
+++ b/usr/src/uts/common/sys/socket.h
@@ -39,6 +39,9 @@
/* 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 +296,9 @@ 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 +338,7 @@ 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
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/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/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..9f2e166fea 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 */
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..c1c12a084e 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) 2014, Joyent, Inc. All rights reserved.
*/
/* 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 *);
diff --git a/usr/src/uts/common/sys/zfd.h b/usr/src/uts/common/sys/zfd.h
new file mode 100644
index 0000000000..c676f143ac
--- /dev/null
+++ b/usr/src/uts/common/sys/zfd.h
@@ -0,0 +1,53 @@
+/*
+ * 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 2014 Joyent, Inc. All rights reserved.
+ */
+
+#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)
+
+#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..ce7f2cc93d 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 2014, 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;
@@ -440,6 +502,7 @@ 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 +555,11 @@ 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.
*/
@@ -544,6 +608,37 @@ 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 +658,27 @@ 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 +756,7 @@ 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 +951,7 @@ 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 */