PSARC/2007/272 Project Clearview: IPMP Rearchitecture

PSARC/2008/773 IPQoS if_groupname Selector Removal
PSARC/2009/001 Move in.mpathd into /lib/inet
6783149 Clearview IPMP Rearchitecture
4472956 libipmp should provide administrative interfaces
4494577 ipmp is opaque - there's no way to get current status
4509788 IPMP's usage of interface flags is not backward compatible
4509869 IPMP's address move mechanism needs to be transparent to applications
4531232 "in.rdiscd: sendto: Bad file number" seen during IPMP DR
4533876 new instances of interfaces under ipmp are generated with each dr/op
4699003 in.mpathd should squawk if interfaces in a group have the same hwaddr
4704937 SUNW_ip_rcm.so is sloppy with strings
4713308 IPMP shouldn't failover unconfigured logical interfaces
4785694 non-homogeneous IPMP group does not do failback
4850407 if_mpadm and IPMP DR failure
5015757 ip can panic with ASSERT(attach_ill == ipif->ipif_ill) failure
5086201 in.ndpd's phyint_reach_random() spews "SIOCSLIFLNKINFO Invalid argument"
6184000 routes cannot be created on failed interfaces
6246564 if_mpadm -r <ifname> doesn't bring up IPv6 link-local data address
6359058 SIOCLIFFAILBACK repeatedly fails with EAGAIN; in.mpathd fills logs
6359536 enabling STANDBY on an interface with no test address acts oddly
6378487 in.dhcpd doesn't work well in an IPMP setup
6462335 cannot offline to IPMP interfaces that have no probe targets
6516992 in.routed spews "Address already in use" during IPMP address move
6518460 ip_rcm`update_pif() must remain calm when logical interfaces disappear
6549957 failed IP interfaces at boot may go unreported
6591186 rpcbind can't deal with indirect calls if all addresses are deprecated
6667143 NCE_F_UNSOL_ADV broken
6698480 IGMP version not retained during IPMP failover
6726235 IPMP group failure can sometimes lead to an extra failover
6726645 in.routed skips DEPRECATED addresses even when no others exist
6738310 ip_ndp_recover() checks IPIF_CONDEMNED on the wrong ipif flags field
6739454 system panics at sdpib`sdp_rts_announce
6739531 IPv6 DAD doesn't work well with IPMP
6740719 in.mpathd may fail to switch to router target mode
6743260 ipif_resolver_up() can fail and leave ARP bringup pending
6746613 ip's DL_NOTE_SDU_SIZE logic mishandles ill_max_frag < ill_max_mtu
6748145 in.ndpd's IPv6 link-local hardware address mappings can go stale
6753560 ilg_delete_all() can race with ill_delete_tail() when ilg_ill changes
6755987 stillborn IFF_POINTOPOINT in.mpathd logic should be hauled out
6775126 SUBDIRS ipsecutils element does not in order be
6775811 NCEs can get stuck in ND_INCOMPLETE if ARP fails when IPMP is in-use
6777496 receive-side ILL_HCKSUM_CAPABLE checks look at the wrong ill
6781488 IPSQ timer restart logic can deadlock under stress
6781883 ip_ndp_find_solicitation() can be passed adverts, and has API issues
6784852 RPCIB, SDP, and RDS all break when vanity naming is used
6786048 IPv6 ND probes create IREs with incorrect source addresses
6786091 I_PLINK handling in IP must not request reentry via ipsq_try_enter()
6786711 IPQoS if_groupname selector needs to go
6787091 assertion failure in ipcl_conn_cleanup() due to non-NULL conn_ilg
6789235 INADDR_ANY ILMs can trigger an assertion failure in IPMP environments
6789502 ipif_resolver_up() calls after ipif_ndp_up() clobber ipif_addr_ready
6789718 ip6.tun0 cannot be plumbed in a non-global-zone post-6745288
6789732 libdlpi may get stuck in i_dlpi_strgetmsg()
6789870 ipif6_dup_recovery() may operate on a freed ipif, corrupting memory
6789874 ipnet_nicevent_cb() may call taskq_dispatch() on a bogus taskq
6790310 in.mpathd may core with "phyint_inst_timer: invalid state 4"

--HG--
rename : usr/src/lib/libinetutil/common/inetutil4.c => usr/src/lib/libinetutil/common/inetutil.c
rename : usr/src/uts/common/inet/vni/vni.c => usr/src/uts/common/inet/dlpistub/dlpistub.c
rename : usr/src/uts/common/inet/vni/vni.conf => usr/src/uts/common/inet/dlpistub/dlpistub.conf
rename : usr/src/uts/common/inet/vni/vni_impl.h => usr/src/uts/common/inet/dlpistub/dlpistub_impl.h
rename : usr/src/uts/intel/vni/Makefile => usr/src/uts/intel/dlpistub/Makefile
rename : usr/src/uts/sparc/vni/Makefile => usr/src/uts/sparc/dlpistub/Makefile

1 files changed, 162 insertions, 44 deletions

diff --git a/usr/src/cmd/cmd-inet/sbin/dhcpagent/interface.c b/usr/src/cmd/cmd-inet/sbin/dhcpagent/interface.c
index 0cfdad40e3..5d2d5fb99e 100644
--- a/usr/src/cmd/cmd-inet/sbin/dhcpagent/interface.c
+++ b/usr/src/cmd/cmd-inet/sbin/dhcpagent/interface.c
@@ -19,7 +19,7 @@
  * CDDL HEADER END
  */
 /*
- * Copyright 2008 Sun Microsystems, Inc.  All rights reserved.
+ * Copyright 2009 Sun Microsystems, Inc.  All rights reserved.
  * Use is subject to license terms.
  */
 
@@ -76,6 +76,7 @@ insert_pif(const char *pname, boolean_t isv6, int *error)
 {
 	dhcp_pif_t *pif;
 	struct lifreq lifr;
+	lifgroupinfo_t lifgr;
 	dlpi_handle_t dh = NULL;
 	int fd = isv6 ? v6_sock_fd : v4_sock_fd;
 
@@ -127,12 +128,60 @@ insert_pif(const char *pname, boolean_t isv6, int *error)
 	}
 
 	/*
-	 * For IPv4, use DLPI to determine the hardware type, hardware
-	 * address, and hardware address length.
+	 * Check if the pif is in an IPMP group.  Interfaces using IPMP don't
+	 * have dedicated hardware addresses, and get their hardware type from
+	 * the SIOCGLIFGROUPINFO ioctl rather than DLPI.
 	 */
-	if (!isv6) {
-		int		rc;
-		dlpi_info_t	dlinfo;
+	if (ioctl(fd, SIOCGLIFGROUPNAME, &lifr) == -1) {
+		*error = DHCP_IPC_E_INT;
+		dhcpmsg(MSG_ERR, "insert_pif: SIOCGLIFGROUPNAME for %s", pname);
+		goto failure;
+	}
+
+	if (lifr.lifr_groupname[0] != '\0') {
+		(void) strlcpy(lifgr.gi_grname, lifr.lifr_groupname,
+		    LIFGRNAMSIZ);
+		if (ioctl(fd, SIOCGLIFGROUPINFO, &lifgr) == -1) {
+			*error = DHCP_IPC_E_INT;
+			dhcpmsg(MSG_ERR, "insert_pif: SIOCGLIFGROUPINFO for %s",
+			    lifgr.gi_grname);
+			goto failure;
+		}
+
+		pif->pif_hwtype = dlpi_arptype(lifgr.gi_mactype);
+		pif->pif_under_ipmp = (strcmp(pname, lifgr.gi_grifname) != 0);
+		(void) strlcpy(pif->pif_grifname, lifgr.gi_grifname, LIFNAMSIZ);
+
+		/*
+		 * For IPMP underlying interfaces, stash the interface index
+		 * of the IPMP meta-interface; we'll use it to send/receive
+		 * traffic.  This is both necessary (since IP_BOUND_IF for
+		 * non-unicast traffic won't work on underlying interfaces)
+		 * and preferred (since a test address lease will be able to
+		 * be maintained as long as another interface in the group is
+		 * still functioning).
+		 */
+		if (pif->pif_under_ipmp) {
+			(void) strlcpy(lifr.lifr_name, pif->pif_grifname,
+			    LIFNAMSIZ);
+
+			if (ioctl(fd, SIOCGLIFINDEX, &lifr) == -1) {
+				*error = DHCP_IPC_E_INT;
+				dhcpmsg(MSG_ERR, "insert_pif: SIOCGLIFINDEX "
+				    "for %s", lifr.lifr_name);
+				goto failure;
+			}
+			pif->pif_grindex = lifr.lifr_index;
+		}
+	}
+
+	/*
+	 * For IPv4, if the hardware type is still unknown, use DLPI to
+	 * determine it, the hardware address, and hardware address length.
+	 */
+	if (!isv6 && pif->pif_hwtype == 0) {
+		int			rc;
+		dlpi_info_t		dlinfo;
 
 		if ((rc = dlpi_open(pname, &dh, 0)) != DLPI_SUCCESS) {
 			dhcpmsg(MSG_ERROR, "insert_pif: dlpi_open: %s",
@@ -661,11 +710,12 @@ verify_lif(const dhcp_lif_t *lif)
 	boolean_t isv6;
 	int fd;
 	struct lifreq lifr;
+	dhcp_pif_t *pif = lif->lif_pif;
 
 	(void) memset(&lifr, 0, sizeof (struct lifreq));
 	(void) strlcpy(lifr.lifr_name, lif->lif_name, LIFNAMSIZ);
 
-	isv6 = lif->lif_pif->pif_isv6;
+	isv6 = pif->pif_isv6;
 	fd = isv6 ? v6_sock_fd : v4_sock_fd;
 
 	if (ioctl(fd, SIOCGLIFFLAGS, &lifr) == -1) {
@@ -689,43 +739,41 @@ verify_lif(const dhcp_lif_t *lif)
 	}
 
 	/*
-	 * Special case: if the interface has gone down as a duplicate, then
-	 * this alone does _not_ mean that we're abandoning it just yet.  Allow
-	 * the state machine to handle this normally by trying to get a new
-	 * lease.
-	 */
-	if ((lifr.lifr_flags & (IFF_UP|IFF_DUPLICATE)) == IFF_DUPLICATE) {
-		dhcpmsg(MSG_DEBUG, "verify_lif: duplicate address on %s",
-		    lif->lif_name);
-		return (B_TRUE);
-	}
-
-	/*
-	 * If the user has torn down or started up the interface manually, then
-	 * abandon the lease.
-	 */
-	if ((lif->lif_flags ^ lifr.lifr_flags) & IFF_UP) {
-		dhcpmsg(MSG_DEBUG, "verify_lif: user has %s %s",
-		    lifr.lifr_flags & IFF_UP ? "started up" : "shut down",
-		    lif->lif_name);
-		return (B_FALSE);
-	}
-
-	/*
 	 * Check for delete and recreate.
 	 */
 	if (ioctl(fd, SIOCGLIFINDEX, &lifr) == -1) {
-		dhcpmsg(MSG_ERR, "verify_lif: SIOCGLIFINDEX failed on %s",
-		    lif->lif_name);
+		if (errno != ENXIO) {
+			dhcpmsg(MSG_ERR, "verify_lif: SIOCGLIFINDEX failed "
+			    "on %s", lif->lif_name);
+		}
 		return (B_FALSE);
 	}
-	if (lifr.lifr_index != lif->lif_pif->pif_index) {
+	if (lifr.lifr_index != pif->pif_index) {
 		dhcpmsg(MSG_DEBUG,
 		    "verify_lif: ifindex on %s changed: %u to %u",
-		    lif->lif_name, lif->lif_pif->pif_index, lifr.lifr_index);
+		    lif->lif_name, pif->pif_index, lifr.lifr_index);
 		return (B_FALSE);
 	}
 
+	if (pif->pif_under_ipmp) {
+		(void) strlcpy(lifr.lifr_name, pif->pif_grifname, LIFNAMSIZ);
+
+		if (ioctl(fd, SIOCGLIFINDEX, &lifr) == -1) {
+			if (errno != ENXIO) {
+				dhcpmsg(MSG_ERR, "verify_lif: SIOCGLIFINDEX "
+				    "failed on %s", lifr.lifr_name);
+			}
+			return (B_FALSE);
+		}
+
+		if (lifr.lifr_index != pif->pif_grindex) {
+			dhcpmsg(MSG_DEBUG, "verify_lif: IPMP group ifindex "
+			    "on %s changed: %u to %u", lifr.lifr_name,
+			    pif->pif_grindex, lifr.lifr_index);
+			return (B_FALSE);
+		}
+	}
+
 	/*
 	 * If the IP address, netmask, or broadcast address have changed, or
 	 * the interface has been unplumbed, then we act like there has been an
@@ -934,6 +982,13 @@ plumb_lif(dhcp_pif_t *pif, const in6_addr_t *addr)
 		    lifr.lifr_name);
 		goto failure;
 	}
+
+	/*
+	 * See comment in set_lif_dhcp().
+	 */
+	if (pif->pif_under_ipmp && !(lifr.lifr_flags & IFF_NOFAILOVER))
+		lifr.lifr_flags |= IFF_NOFAILOVER | IFF_DEPRECATED;
+
 	lifr.lifr_flags |= IFF_UP | IFF_DHCPRUNNING;
 	if (ioctl(v6_sock_fd, SIOCSLIFFLAGS, &lifr) == -1) {
 		dhcpmsg(MSG_ERR, "plumb_lif: SIOCSLIFFLAGS %s",
@@ -1060,8 +1115,9 @@ set_lif_dhcp(dhcp_lif_t *lif, boolean_t is_adopting)
 	int fd;
 	int err;
 	struct lifreq lifr;
+	dhcp_pif_t *pif = lif->lif_pif;
 
-	fd = lif->lif_pif->pif_isv6 ? v6_sock_fd : v4_sock_fd;
+	fd = pif->pif_isv6 ? v6_sock_fd : v4_sock_fd;
 
 	(void) strlcpy(lifr.lifr_name, lif->lif_name, LIFNAMSIZ);
 
@@ -1098,6 +1154,17 @@ set_lif_dhcp(dhcp_lif_t *lif, boolean_t is_adopting)
 			    "set on %s", lif->lif_name);
 		}
 	} else {
+		/*
+		 * If the lif is on an interface under IPMP, IFF_NOFAILOVER
+		 * must be set or the kernel will prevent us from setting
+		 * IFF_DHCPRUNNING (since the subsequent IFF_UP would lead to
+		 * migration).  We set IFF_DEPRECATED too since the kernel
+		 * will set it automatically when setting IFF_NOFAILOVER,
+		 * causing our lif_flags value to grow stale.
+		 */
+		if (pif->pif_under_ipmp && !(lifr.lifr_flags & IFF_NOFAILOVER))
+			lifr.lifr_flags |= IFF_NOFAILOVER | IFF_DEPRECATED;
+
 		lifr.lifr_flags |= IFF_DHCPRUNNING;
 		if (ioctl(fd, SIOCSLIFFLAGS, &lifr) == -1) {
 			dhcpmsg(MSG_ERR, "set_lif_dhcp: SIOCSLIFFLAGS for %s",
@@ -1207,6 +1274,13 @@ clear_lif_deprecated(dhcp_lif_t *lif)
 		return (B_FALSE);
 	}
 
+	/*
+	 * Don't try to clear IFF_DEPRECATED if this is a test address,
+	 * since IPMP's use of IFF_DEPRECATED is not compatible with ours.
+	 */
+	if (lifr.lifr_flags & IFF_NOFAILOVER)
+		return (B_TRUE);
+
 	if (!(lifr.lifr_flags & IFF_DEPRECATED))
 		return (B_TRUE);
 
@@ -1226,16 +1300,19 @@ clear_lif_deprecated(dhcp_lif_t *lif)
  *
  *   input: dhcp_lif_t *: the logical interface to operate on
  *	    in_addr_t: the address the socket will be bound to (in hbo)
+ *	    boolean_t: B_TRUE if the address should be brought up (if needed)
  *  output: boolean_t: B_TRUE if the socket was opened successfully.
  */
 
 boolean_t
-open_ip_lif(dhcp_lif_t *lif, in_addr_t addr_hbo)
+open_ip_lif(dhcp_lif_t *lif, in_addr_t addr_hbo, boolean_t bringup)
 {
 	const char *errmsg;
 	struct lifreq lifr;
 	int on = 1;
 	uchar_t ttl = 255;
+	uint32_t ifindex;
+	dhcp_pif_t *pif = lif->lif_pif;
 
 	if (lif->lif_sock_ip_fd != -1) {
 		dhcpmsg(MSG_WARNING, "open_ip_lif: socket already open on %s",
@@ -1270,7 +1347,7 @@ open_ip_lif(dhcp_lif_t *lif, in_addr_t addr_hbo)
 		}
 
 		if (setsockopt(lif->lif_sock_ip_fd, IPPROTO_IP, IP_DHCPINIT_IF,
-		    &lif->lif_pif->pif_index, sizeof (int)) == -1) {
+		    &pif->pif_index, sizeof (int)) == -1) {
 			errmsg = "cannot set IP_DHCPINIT_IF";
 			goto failure;
 		}
@@ -1288,23 +1365,40 @@ open_ip_lif(dhcp_lif_t *lif, in_addr_t addr_hbo)
 		goto failure;
 	}
 
-	if (setsockopt(lif->lif_sock_ip_fd, IPPROTO_IP, IP_BOUND_IF,
-	    &lif->lif_pif->pif_index, sizeof (int)) == -1) {
+	ifindex = pif->pif_under_ipmp ? pif->pif_grindex : pif->pif_index;
+	if (setsockopt(lif->lif_sock_ip_fd, IPPROTO_IP, IP_BOUND_IF, &ifindex,
+	    sizeof (int)) == -1) {
 		errmsg = "cannot set IP_BOUND_IF";
 		goto failure;
 	}
 
-	/*
-	 * Make sure at least one lif on the interface we used in IP_BOUND_IF
-	 * is IFF_UP so that we can send and receive IP packets.
-	 */
 	(void) strlcpy(lifr.lifr_name, lif->lif_name, LIFNAMSIZ);
 	if (ioctl(v4_sock_fd, SIOCGLIFFLAGS, &lifr) == -1) {
 		errmsg = "cannot get interface flags";
 		goto failure;
 	}
 
-	if (!(lifr.lifr_flags & IFF_UP)) {
+	/*
+	 * If the lif is part of an interface under IPMP, IFF_NOFAILOVER must
+	 * be set or the kernel will prevent us from setting IFF_DHCPRUNNING
+	 * (since the subsequent IFF_UP would lead to migration).  We set
+	 * IFF_DEPRECATED too since the kernel will set it automatically when
+	 * setting IFF_NOFAILOVER, causing our lif_flags value to grow stale.
+	 */
+	if (pif->pif_under_ipmp && !(lifr.lifr_flags & IFF_NOFAILOVER)) {
+		lifr.lifr_flags |= IFF_NOFAILOVER | IFF_DEPRECATED;
+		if (ioctl(v4_sock_fd, SIOCSLIFFLAGS, &lifr) == -1) {
+			errmsg = "cannot set IFF_NOFAILOVER";
+			goto failure;
+		}
+	}
+	lif->lif_flags = lifr.lifr_flags;
+
+	/*
+	 * If this is initial bringup, make sure the address we're acquiring a
+	 * lease on is IFF_UP.
+	 */
+	if (bringup && !(lifr.lifr_flags & IFF_UP)) {
 		/*
 		 * Start from a clean slate.
 		 */
@@ -1330,6 +1424,30 @@ open_ip_lif(dhcp_lif_t *lif, in_addr_t addr_hbo)
 		    ((struct sockaddr_in *)&lifr.lifr_addr)->sin_addr.s_addr;
 	}
 
+	/*
+	 * Usually, bringing up the address we're acquiring a lease on is
+	 * sufficient to allow packets to be sent and received via the
+	 * IP_BOUND_IF we did earlier.  However, if we're acquiring a lease on
+	 * an underlying IPMP interface, the group interface will be used for
+	 * sending and receiving IP packets via IP_BOUND_IF.  Thus, ensure at
+	 * least one address on the group interface is IFF_UP.
+	 */
+	if (bringup && pif->pif_under_ipmp) {
+		(void) strlcpy(lifr.lifr_name, pif->pif_grifname, LIFNAMSIZ);
+		if (ioctl(v4_sock_fd, SIOCGLIFFLAGS, &lifr) == -1) {
+			errmsg = "cannot get IPMP group interface flags";
+			goto failure;
+		}
+
+		if (!(lifr.lifr_flags & IFF_UP)) {
+			lifr.lifr_flags |= IFF_UP;
+			if (ioctl(v4_sock_fd, SIOCSLIFFLAGS, &lifr) == -1) {
+				errmsg = "cannot bring up IPMP group interface";
+				goto failure;
+			}
+		}
+	}
+
 	lif->lif_packet_id = iu_register_event(eh, lif->lif_sock_ip_fd, POLLIN,
 	    dhcp_packet_lif, lif);
 	if (lif->lif_packet_id == -1) {