Re: [PATCH RFC] ipv6: fix route selection if kernel is not compiled with CONFIG_IPV6_ROUTER_PREF

[Hannes Frederic Sowa <hannes@stressinduktion.org>]

On Wed, Jul 10, 2013 at 02:22:55PM +0200, Nicolas Dichtel wrote:
> Le 10/07/2013 12:53, Hannes Frederic Sowa a écrit :
> >On Wed, Jul 10, 2013 at 11:28:57AM +0200, Nicolas Dichtel wrote:
> >>Le 10/07/2013 09:54, Nicolas Dichtel a écrit :
> >>>Le 09/07/2013 23:57, Hannes Frederic Sowa a écrit :
> >>>>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,
> >                          * but not the same gateway, then the route we 
> >                          try to
> >                          * add is sibling to this route, increment our 
> >                          counter
> >                          * of siblings, and later we will add our route 
> >                          to the
> >                          * list.
> >                          * Only static routes (which don't have flag
> >                          * RTF_EXPIRES) are used for ECMPv6.
> >                          *
> >                          * To avoid long list, we only had siblings if the
> >                          * route have a gateway.
> >                          */
> >                         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 
> >2001:db8::32, siblings 2, metric 0, expires 0 gateway 2 idev6 
> >ffff8801139ddc00 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 
> >ffff880117edc400 dev ffff8801185cb000
> >[   42.527241] (iter): ffff88011796d380 dst 2000::1 plen 128 gateway 
> >2001:db8::33, siblings 2, metric 0, expires 0 gateway 2 idev6 
> >ffff8801139ddc00 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 
> >ffff880117edc400 dev ffff8801185cb000
> >
> > From my understanding these two routes should not be aggregated in one 
> > ecmp
> >route set. Am I seeing this correct? (My configuration is like in the mail
> >before.)
> Hmm, why?
> Routes have the same destination, same metric, are static (expires == 0) 
> and have a gateway.

The route with rt6i_gateway does actually expire because I got it from
autoconf and ip -6 r l confirms this, too. It seems this is only the cached
route (I will confirm shortly). Is this still ok?

> nsiblings counts the number of siblings and does not contains ourself, 
> hence both iter should be 1, not 2.

Ok.

Thanks for helping,

  Hannes