On Sat, Sep 28, 2013 at 10:51:31PM +0200, Steinar H. Gunderson wrote:

On Sat, Sep 28, 2013 at 10:33:18PM +0200, Hannes Frederic Sowa wrote:

quoted

quoted

Could this be related somehow to the packets coming from 2001:67c:29f4::31,
while the default route is to a link-local address? (An RPF issue?) This used
to work (although it was often flaky for me) in 3.10 and before. I can't
easily bisect, though, as I don't boot this machine too often.

This looks like a bug and should definitely get fixed. There should be
no RPF issue. May I have a look at your /proc/net/ipv6_route?

Hi,

I removed all the “weird” routes, and confirmed it fixed the problem.
However, upon adding them back again, the problem was still gone
(despite flushing the route cache).

This means that the issue has gone back to being intermittent, which is of
course the worst kind of bug to trace down. :-) I'll dump
/proc/net/ipv6_route and send you once I see the bug manifest itself again,
OK?

Yes, that would be very helpful.

Also, you can try to churn up your bgp connection a bit so that the fib
serial numbers get incremented a lot (drop and install new routes). When
tcp_ipv6 processes the icmp errors it will drop the in-socket cached
routing entry then and will reinstall a relookuped one.  This is my only
suspect currently. If that would help to reproduce the problem the suspects
would be the changes in the next-hop selection. Sorry, no other idea
currently.

Thanks,

  Hannes