[<prev] [next>] [thread-next>] [day] [month] [year] [list]
Message-Id: <1471355943-21288-1-git-send-email-ayourtch@gmail.com>
Date: Tue, 16 Aug 2016 15:59:01 +0200
From: Andrew Yourtchenko <ayourtch@...il.com>
To: netdev@...r.kernel.org
Cc: Andrew Yourtchenko <ayourtch@...il.com>
Subject: [PATCH 0/2] ipv6: fix stuck RA-derived route in container after migration with criu
This patchset fixes the connectivity problem for containers
with RA-derived default route, after they were migrated using criu:
the default routes would lose their "expires" value and become
stuck forever. The corresponding criu issue with the discussion
is at https://github.com/xemul/criu/issues/177
The latter uses "ip -6 route save" to save the routes during
the migration, and "ip -6 route restore" during restore, so the problem
is easily reproducible even without criu.
There are two problems, hence two patches in this patchset:
1) the expiry time for the route is saved in
RTA_CACHEINFO via rtnl_put_cacheinfo, but the code in rtm_to_fib6_config
expects the RTA_EXPIRES. Rather than cherrypicking in the restore code path
from RTA_CACHEINFO, adding RTA_EXPIRES upon save seemed like
a better option.
2) the restored route, even with the properly restored expires,
does not have the correct RTF_* flags set (RTF_ADDRCONF|RTF_DEFAULT),
preventing the incoming router advertisements from updating it.
During the code review I noticed RTF_ROUTEINFO would be also lost
during save/restore. This can be viewed as an operation symmetric
to that done in f0396f60d7c165018c9b203fb9b89fb224835578.
Tested both net and net-next with the patches.
Andrew Yourtchenko (2):
ipv6: save route expiry in RTA_EXPIRES if RTF_EXPIRES set
ipv6: fixup RTF_* flags when restoring RTPROT_RA route from rtnetlink
net/ipv6/route.c | 18 ++++++++++++++++++
1 file changed, 18 insertions(+)
--
2.7.4
Powered by blists - more mailing lists