netns: add schedule point in ops_exit_list()

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: cf86ca59d01c ("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>

diff --git a/net/core/net_namespace.c b/net/core/net_namespace.c
index 9b7171c40434985b869c1477975fc75447d78c3b..a5b5bb99c64462dac2513f7b3e28cb0763844c21 100644 (file)
--- a/net/core/net_namespace.c
+++ b/net/core/net_namespace.c
@@ -164,8 +164,10 @@ static void ops_exit_list(const struct pernet_operations *ops,
 {
        struct net *net;
        if (ops->exit) {
-               list_for_each_entry(net, net_exit_list, exit_list)
+               list_for_each_entry(net, net_exit_list, exit_list) {
                        ops->exit(net);
+                       cond_resched();
+               }
        }
        if (ops->exit_batch)
                ops->exit_batch(net_exit_list);