Deal with the case where ifma_protospec may be NULL, during

any IPv4 multicast operations which reference it.

There is a potential race because ifma_protospec is set to NULL
when we discover the underlying ifnet has gone away. This write
is not covered by the IF_ADDR_LOCK, and it's difficult to widen
its scope without making it a recursive lock. It isn't clear why
this manifests more quickly with 802.11 interfaces, but does not
seem to manifest at all with wired interfaces.

With this change, the 802.11 related panics reported by sam@
and cokane@ should go away. It is not the right fix, that requires
more thought before 8.0.

Idea from:	sam
Tested by:	cokane

3 files changed, 37 insertions, 6 deletions

diff --git a/sys/netinet/igmp.c b/sys/netinet/igmp.c
index 31e8306..5952112 100644
--- a/sys/netinet/igmp.c
+++ b/sys/netinet/igmp.c
@@ -183,6 +183,11 @@ static int	vnet_igmp_idetach(const void *);
  * VIMAGE: Each in_multi corresponds to an ifp, and each ifp corresponds
  * to a vnet in ifp->if_vnet.
  *
+ * SMPng: XXX We may potentially race operations on ifma_protospec.
+ * The problem is that we currently lack a clean way of taking the
+ * IF_ADDR_LOCK() between the ifnet and in layers w/o recursing,
+ * as anything which modifies ifma needs to be covered by that lock.
+ * So check for ifma_protospec being NULL before proceeding.
  */
 struct mtx		 igmp_mtx;
 int			 mpsafe_igmp = 0;
@@ -601,6 +606,7 @@ out:
  * is detached, but also before the link layer does its cleanup.
  *
  * SMPNG: igmp_ifdetach() needs to take IF_ADDR_LOCK().
+ * XXX This is also bitten by unlocked ifma_protospec access.
  *
  * VIMAGE: curvnet should have been set by caller, but let's not assume
  * that for now.
@@ -623,8 +629,13 @@ igmp_ifdetach(struct ifnet *ifp)
 	if (igi->igi_version == IGMP_VERSION_3) {
 		IF_ADDR_LOCK(ifp);
 		TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
-			if (ifma->ifma_addr->sa_family != AF_INET)
+			if (ifma->ifma_addr->sa_family != AF_INET ||
+			    ifma->ifma_protospec == NULL)
 				continue;
+#if 0
+			KASSERT(ifma->ifma_protospec != NULL,
+			    ("%s: ifma_protospec is NULL", __func__));
+#endif
 			inm = (struct in_multi *)ifma->ifma_protospec;
 			if (inm->inm_state == IGMP_LEAVING_MEMBER) {
 				SLIST_INSERT_HEAD(&igi->igi_relinmhead,
@@ -783,7 +794,8 @@ igmp_input_v1_query(struct ifnet *ifp, const struct ip *ip)
 	 */
 	IF_ADDR_LOCK(ifp);
 	TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
-		if (ifma->ifma_addr->sa_family != AF_INET)
+		if (ifma->ifma_addr->sa_family != AF_INET ||
+		    ifma->ifma_protospec == NULL)
 			continue;
 		inm = (struct in_multi *)ifma->ifma_protospec;
 		if (inm->inm_timer != 0)
@@ -880,7 +892,8 @@ igmp_input_v2_query(struct ifnet *ifp, const struct ip *ip,
 
 			IF_ADDR_LOCK(ifp);
 			TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
-				if (ifma->ifma_addr->sa_family != AF_INET)
+				if (ifma->ifma_addr->sa_family != AF_INET ||
+				    ifma->ifma_protospec == NULL)
 					continue;
 				inm = (struct in_multi *)ifma->ifma_protospec;
 				igmp_v2_update_group(inm, timer);
@@ -1701,7 +1714,8 @@ igmp_fasttimo_vnet(void)
 		IF_ADDR_LOCK(ifp);
 		TAILQ_FOREACH_SAFE(ifma, &ifp->if_multiaddrs, ifma_link,
 		    tifma) {
-			if (ifma->ifma_addr->sa_family != AF_INET)
+			if (ifma->ifma_addr->sa_family != AF_INET ||
+			    ifma->ifma_protospec == NULL)
 				continue;
 			inm = (struct in_multi *)ifma->ifma_protospec;
 			switch (igi->igi_version) {
@@ -3311,7 +3325,8 @@ igmp_v3_dispatch_general_query(struct igmp_ifinfo *igi)
 
 	IF_ADDR_LOCK(ifp);
 	TAILQ_FOREACH_SAFE(ifma, &ifp->if_multiaddrs, ifma_link, tifma) {
-		if (ifma->ifma_addr->sa_family != AF_INET)
+		if (ifma->ifma_addr->sa_family != AF_INET ||
+		    ifma->ifma_protospec == NULL)
 			continue;
 
 		inm = (struct in_multi *)ifma->ifma_protospec;
diff --git a/sys/netinet/in.c b/sys/netinet/in.c
index 1927b5c..1423e81 100644
--- a/sys/netinet/in.c
+++ b/sys/netinet/in.c
@@ -1011,6 +1011,8 @@ in_ifdetach(struct ifnet *ifp)
  * Delete all IPv4 multicast address records, and associated link-layer
  * multicast address records, associated with ifp.
  * XXX It looks like domifdetach runs AFTER the link layer cleanup.
+ * XXX This should not race with ifma_protospec being set during
+ * a new allocation, if it does, we have bigger problems.
  */
 static void
 in_purgemaddrs(struct ifnet *ifp)
@@ -1031,8 +1033,13 @@ in_purgemaddrs(struct ifnet *ifp)
 	 */
 	IF_ADDR_LOCK(ifp);
 	TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
-		if (ifma->ifma_addr->sa_family != AF_INET)
+		if (ifma->ifma_addr->sa_family != AF_INET ||
+		    ifma->ifma_protospec == NULL)
 			continue;
+#if 0
+		KASSERT(ifma->ifma_protospec != NULL,
+		    ("%s: ifma_protospec is NULL", __func__));
+#endif
 		inm = (struct in_multi *)ifma->ifma_protospec;
 		LIST_INSERT_HEAD(&purgeinms, inm, inm_link);
 	}
diff --git a/sys/netinet/in_mcast.c b/sys/netinet/in_mcast.c
index e91a5bf..9918c95 100644
--- a/sys/netinet/in_mcast.c
+++ b/sys/netinet/in_mcast.c
@@ -432,6 +432,9 @@ in_getmulti(struct ifnet *ifp, const struct in_addr *group,
 	if (error != 0)
 		return (error);
 
+	/* XXX ifma_protospec must be covered by IF_ADDR_LOCK */
+	IF_ADDR_LOCK(ifp);
+
 	/*
 	 * If something other than netinet is occupying the link-layer
 	 * group, print a meaningful error message and back out of
@@ -454,9 +457,12 @@ in_getmulti(struct ifnet *ifp, const struct in_addr *group,
 #endif
 		++inm->inm_refcount;
 		*pinm = inm;
+		IF_ADDR_UNLOCK(ifp);
 		return (0);
 	}
 
+	IF_ADDR_LOCK_ASSERT(ifp);
+
 	/*
 	 * A new in_multi record is needed; allocate and initialize it.
 	 * We DO NOT perform an IGMP join as the in_ layer may need to
@@ -467,6 +473,7 @@ in_getmulti(struct ifnet *ifp, const struct in_addr *group,
 	inm = malloc(sizeof(*inm), M_IPMADDR, M_NOWAIT | M_ZERO);
 	if (inm == NULL) {
 		if_delmulti_ifma(ifma);
+		IF_ADDR_UNLOCK(ifp);
 		return (ENOMEM);
 	}
 	inm->inm_addr = *group;
@@ -489,6 +496,7 @@ in_getmulti(struct ifnet *ifp, const struct in_addr *group,
 
 	*pinm = inm;
 
+	IF_ADDR_UNLOCK(ifp);
 	return (0);
 }
 
@@ -522,6 +530,7 @@ inm_release_locked(struct in_multi *inm)
 
 	ifma = inm->inm_ifma;
 
+	/* XXX this access is not covered by IF_ADDR_LOCK */
 	CTR2(KTR_IGMPV3, "%s: purging ifma %p", __func__, ifma);
 	KASSERT(ifma->ifma_protospec == inm,
 	    ("%s: ifma_protospec != inm", __func__));