Neil Horman wrote:

On Fri, Feb 02, 2007 at 11:46:08AM -0800, David Miller wrote:

quoted

From: Neil Horman <nhorman@tuxdriver.com>
Date: Fri, 2 Feb 2007 14:06:34 -0500

quoted

Ok, I'm still testing it, but heres a new patch for review.
Significant changes include the addition of a
CONFIG_IPV6_OPTIMISTIC_DAD option that is dependent on the inclusion
of both IPPV6 and EXPERIMENTAL options, as well as a new method for
redirecting packets from optimistic sources to incomplete neighbors
by instead looking up a default router in ip6_dst_lookup_tail, as I
described in my previous note.

This looks largely fine to me, but I wonder about one bit:

+
+	/*
+	 * Optimistic nodes need to join the anycast address
+	 * right away
+	 */
+	if (ifp->flags & IFA_F_OPTIMISTIC)
+		addrconf_join_anycast(ifp);
+

If something can clear that bit during the DAD, we won't
call addrconf_leave_anycast() later.  Can that happen?

The only way the flag should get cleared once we've started DAD is when it
completes or fails.  In the failure case, we destroy the ifaddr structure, which
I think should force a leave_anycast, while the completed condition calls
addrconf_leave_anycast as part of the completion process before it clears the
flags, so I think we should be ok.  If anyone sees anything to the contrary,
please let me know and I'll be sure to  plug the hole.

I think there is a hole:

Looking for addrconf_leave_anycast() shows:
dev_forward_change():	475 addrconf_leave_anycast(ifa);
__ipv6_ifa_notify():	3613 addrconf_leave_anycast(ifp);

Problem:  __ipv6_ifa_notify only performs the 'leave' if forwarding is enabled.
	  However, OPTIMISTIC, is set when forwarding is _disabled_.


-vlad