From mboxrd@z Thu Jan 1 00:00:00 1970 From: Hannes Frederic Sowa Subject: Re: [PATCH RFC] ipv6: fix route selection if kernel is not compiled with CONFIG_IPV6_ROUTER_PREF Date: Wed, 10 Jul 2013 12:53:16 +0200 Message-ID: <20130710105316.GA5735@order.stressinduktion.org> References: <20130707173031.GC9625@order.stressinduktion.org> <20130709215701.GD9763@order.stressinduktion.org> <51DD1352.8000705@6wind.com> <51DD2959.9060206@6wind.com> Mime-Version: 1.0 Content-Type: text/plain; charset=utf-8 Content-Transfer-Encoding: QUOTED-PRINTABLE Cc: netdev@vger.kernel.org, yoshfuji@linux-ipv6.org, petrus.lt@gmail.com, davem@davemloft.net To: Nicolas Dichtel Return-path: Received: from s15338416.onlinehome-server.info ([87.106.68.36]:36515 "EHLO order.stressinduktion.org" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1754357Ab3GJKxS (ORCPT ); Wed, 10 Jul 2013 06:53:18 -0400 Content-Disposition: inline In-Reply-To: <51DD2959.9060206@6wind.com> Sender: netdev-owner@vger.kernel.org List-ID: On Wed, Jul 10, 2013 at 11:28:57AM +0200, Nicolas Dichtel wrote: > Le 10/07/2013 09:54, Nicolas Dichtel a =C3=A9crit : > >Le 09/07/2013 23:57, Hannes Frederic Sowa a =C3=A9crit : > >>After starting a ping6 2000::1 the box should panic soon, after the > >>first nexthop entry times out. > >> > >>Perhaps you could give me a hint? > >I will run some tests with your patch. Will see. > I don't reproduce this panic. I just dumped the routes for which it does increase the rt6i_nsiblings counter in this condition: /* If we have the same destination and the same= metric, = = = =20 * but not the same gateway, then the route we = try to = = = =20 * add is sibling to this route, increment our = counter = = = =20 * of siblings, and later we will add our route= to the = = = =20 * list. = = = = =20 * Only static routes (which don't have flag = = = = =20 * RTF_EXPIRES) are used for ECMPv6. = = = = =20 * = = = = =20 * To avoid long list, we only had siblings if = the = = = =20 * route have a gateway. = = = = =20 */ if (rt->rt6i_flags & RTF_GATEWAY && !(rt->rt6i_flags & RTF_EXPIRES) && !(iter->rt6i_flags & RTF_EXPIRES)) rt->rt6i_nsiblings++; dump_route(iter, "(iter)"); dump_route(rt, "(rt)"); } Here: [ 42.497470] (iter): ffff88011796cc00 dst 2000::1 plen 128 gateway 20= 01:db8::32, siblings 2, metric 0, expires 0 gateway 2 idev6 ffff8801139= ddc00 dev ffff880117e83000 [ 42.505912] (rt): ffff88011796d800 dst 2000::1 plen 128 gateway fe80= ::5054:ff:fe82:e153, siblings 1, metric 0, expires 0 gateway 2 idev6 ff= ff880117edc400 dev ffff8801185cb000 [ 42.527241] (iter): ffff88011796d380 dst 2000::1 plen 128 gateway 20= 01:db8::33, siblings 2, metric 0, expires 0 gateway 2 idev6 ffff8801139= ddc00 dev ffff880117e83000 [ 42.536440] (rt): ffff88011796d800 dst 2000::1 plen 128 gateway fe80= ::5054:ff:fe82:e153, siblings 2, metric 0, expires 0 gateway 2 idev6 ff= ff880117edc400 dev ffff8801185cb000 =46rom my understanding these two routes should not be aggregated in on= e ecmp route set. Am I seeing this correct? (My configuration is like in the m= ail before.) I wonder why the '(rt)' route does not have the expires flag, but it se= ems we have to special-case RTF_CACHE routes here which derive from different levels of the fib6_tree. Does that make sense? Greetings, Hannes