]> git.ipfire.org Git - thirdparty/kernel/stable.git/commit
netns: add schedule point in ops_exit_list()
authorEric Dumazet <edumazet@google.com>
Tue, 18 Jan 2022 11:43:40 +0000 (03:43 -0800)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Thu, 27 Jan 2022 08:04:33 +0000 (09:04 +0100)
commit4048cedfd16a995b2ef4294b9539da17e2fea750
tree6ef6e525c32bbf4bfd8143196825dc7f1e77e072
parent9f9ba0b65031c68384d88580d987f1fdf2b9478d
netns: add schedule point in ops_exit_list()

commit 2836615aa22de55b8fca5e32fe1b27a67cda625e upstream.

When under stress, cleanup_net() can have to dismantle
netns in big numbers. ops_exit_list() currently calls
many helpers [1] that have no schedule point, and we can
end up with soft lockups, particularly on hosts
with many cpus.

Even for moderate amount of netns processed by cleanup_net()
this patch avoids latency spikes.

[1] Some of these helpers like fib_sync_up() and fib_sync_down_dev()
are very slow because net/ipv4/fib_semantics.c uses host-wide hash tables,
and ifindex is used as the only input of two hash functions.
    ifindexes tend to be the same for all netns (lo.ifindex==1 per instance)
    This will be fixed in a separate patch.

Fixes: 72ad937abd0a ("net: Add support for batching network namespace cleanups")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Eric W. Biederman <ebiederm@xmission.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
net/core/net_namespace.c