dropped icmp6 too big messages that reported an MTU less than the min…

…imum

MTU (1280).
we believe such an error is only meaningful when we're acting as an SIIT
client, which we do not support.

kame/sys/netinet6/icmp6.c

@@ -1,4 +1,4 @@
-/*	$KAME: icmp6.c,v 1.296 2002/04/08 14:15:03 jinmei Exp $	*/
+/*	$KAME: icmp6.c,v 1.297 2002/04/09 12:17:53 jinmei Exp $	*/
 
 /*
  * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project.
@@ -672,6 +672,17 @@ icmp6_input(mp, offp, proto)
 		if (code != 0)
 			goto badcode;
 
+		/*
+		 * RFC 1981 says that we may receive a Too Big message
+		 * reporting a next-hop MTU that is less than the minimum MTU,
+		 * but this can only happen when using an SIIT style
+		 * translation.
+		 * Since we do not support SIIT, it should be better to ignore
+		 * those messages.
+		 */
+		if (ntohl(icmp6->icmp6_mtu) < IPV6_MMTU)
+			break;	/* or just drop it? */
+
 		code = PRC_MSGSIZE;
 
 		/*
@@ -1381,16 +1392,13 @@ icmp6_mtudisc_update(ip6cp, dst, validated)
 #endif
 
 	/*
-	 * RFC 1981 says that we may receive a Too Big message reporting a
-	 * next-hop MTU that is less than the minimum MTU, but the behavior
-	 * which should be taken in that case is not very clear.
-	 * We can perhaps discard such a message, but we dare to reduce the
-	 * MTU to the minimum MTU; using a too small MTU reduces performance,
-	 * whereas using a too large MTU causes reachability problem.
-	 * The former should be better.
+	 * Ignore MTU less than the minimum link MTU.  See comments in
+	 * icmp6_input().  Although the function drops such a case from the
+	 * wire, we may also fall into the situation if we're called from
+	 * ip6_output() via pfctlinput2(), perhaps due to an internal bug.
 	 */
 	if (mtu < IPV6_MMTU)
-		mtu = IPV6_MMTU;
+		return;
 
 #if defined(__NetBSD__) || defined(__OpenBSD__)
 	/*
@@ -1444,7 +1452,6 @@ icmp6_mtudisc_update(ip6cp, dst, validated)
 
 	if (rt && (rt->rt_flags & RTF_HOST) &&
 	    !(rt->rt_rmx.rmx_locks & RTV_MTU) &&
-	    mtu >= IPV6_MMTU &&	/* this is actually ensured already */
 	    mtu < rt->rt_rmx.rmx_mtu) {
 		if (mtu < nd_ifinfo[rt->rt_ifp->if_index].linkmtu) {
 			icmp6stat.icp6s_pmtuchg++;