Really ripngd should either leave groups on down interfaces, or keep

track of on which interfaces joins have occurred.  This is complicated
by the possible lack of a clear behavior spec on whether interfaces
keep their joined group lists across a down/up/down transition.  (On
NetBSD, they do.)

2004-01-03  Greg Troxel  <gdt@ahi.ir.bbn.com>

        * ripng_interface.c (ripng_multicast_join): If IPV6_JOIN_GROUP
        returns EADDRINUSE, consider the join successful.  This happens
        when an interface goes down and comes back because
        ripng_multicast_leave does not invoke the LEAVE_GROUP operation if
        the interface is down.  Solves problem of ripng stopping working
        on an interface which goes down and then comes back up (on NetBSD).

ripngd/ChangeLog

@@ -1,3 +1,12 @@
+2004-01-03  Greg Troxel  <gdt@ahi.ir.bbn.com>
+
+	* ripng_interface.c (ripng_multicast_join): If IPV6_JOIN_GROUP
+	returns EADDRINUSE, consider the join successful.  This happens
+	when an interface goes down and comes back because
+	ripng_multicast_leave does not invoke the LEAVE_GROUP operation if
+	the interface is down.  Solves problem of ripng stopping working
+	on an interface which goes down and then comes back up (on NetBSD).
+
 2003-05-25 Hasso Tepper <hasso@estpak.ee>
   
   * Revert ripng_ifrmap stuff because ripd uses it now as well.

ripngd/ripng_interface.c

@@ -67,6 +67,18 @@ ripng_multicast_join (struct interface *ifp)
 
     ret = setsockopt (ripng->sock, IPPROTO_IPV6, IPV6_JOIN_GROUP,
 		      (char *) &mreq, sizeof (mreq));
+
+    if (ret < 0 && errno == EADDRINUSE)
+      {
+	/*
+	 * Group is already joined.  This occurs due to sloppy group
+	 * management, in particular declining to leave the group on
+	 * an interface that has just gone down.
+	 */
+	zlog_warn ("ripng join on %s EADDRINUSE (ignoring)\n", ifp->name);
+	return 0;		/* not an error */
+      }
+
     if (ret < 0)
       zlog_warn ("can't setsockopt IPV6_JOIN_GROUP: %s", strerror (errno));