[PATCH] Fix in6p_leave_group() panic by misbehaving apps

[Bruce Simpson]

Hi,

If anyone is experiencing panics with IPv6 in the kernel, and multicast 
applications active, please test this patch. I think some folk here saw 
this with VLC.

re@: If this patch is good (I'll try to test locally) then it should go 
into HEAD ASAP.

    Some poorly behaved IPv6 multicast applications don't specify an 
interface for the join, and this triggers a KASSERT I put in to catch 
such corner cases.

    Multicast doesn't work unless apps are aware of the links active in 
the system they're running on, and this is a glaring hole in the 
Boost.ASIO API, sadly.  This was caught by a Boost regression run on 
ref8.freebsd.org.

Thanks to simon@ for logging the panic from the cluster console servers.

cheers,
BMS

-------------- next part --------------
Index: in6_mcast.c
===================================================================
--- in6_mcast.c	(revision 195752)
+++ in6_mcast.c	(working copy)
@@ -2166,8 +2166,23 @@
 		 * refactored, assuming the scope IDs are the way to go.
 		 */
 		ifindex = ntohs(gsa->sin6.sin6_addr.s6_addr16[1]);
+#if 0
 		KASSERT(ifindex != 0, ("%s: bad zone ID", __func__));
 		ifp = ifnet_byindex(ifindex);
+#else
+		/*
+		 * Some badly behaved applications don't pass an ifindex
+		 * or a scope ID, which is an API violation.
+		 */
+		if (ifindex == 0) {
+			CTR2(KTR_MLD, "%s: warning: no ifindex, looking up "
+			    "ifp for group %s.", __func__,
+			    ip6_sprintf(ip6tbuf, &gsa->sin6.sin6_addr));
+			ifp = in6p_lookup_mcast_ifp(inp, &gsa->sin6);
+		} else {
+			ifp = ifnet_byindex(ifindex);
+		}
+#endif
 		if (ifp == NULL)
 			return (EADDRNOTAVAIL);
 	}