ip6mr: remove synchronize_rcu() in favor of SOCK_RCU_FREE
authorEric Dumazet <edumazet@google.com>
Wed, 7 Mar 2018 16:43:19 +0000 (08:43 -0800)
committerDavid S. Miller <davem@davemloft.net>
Wed, 7 Mar 2018 23:13:41 +0000 (18:13 -0500)
commita366e300ae9fc466d333e6d8f2bc5d58ed248041
tree689aef7cbe9b1df29666ab6a67439a81f2edc7ab
parent3f6be0ebdeb2255534664c4ececc2eac1d5dcb30
ip6mr: remove synchronize_rcu() in favor of SOCK_RCU_FREE

Kirill found that recently added synchronize_rcu() call in
ip6mr_sk_done()
was slowing down netns dismantle and posted a patch to use it only if
the socket
was found.

I instead suggested to get rid of this call, and use instead
SOCK_RCU_FREE

We might later change IPv4 side to use the same technique and unify
both stacks. IPv4 does not use synchronize_rcu() but has a call_rcu()
that could be replaced by SOCK_RCU_FREE.

Tested:
 time for i in {1..1000}; do unshare -n /bin/false;done

 Before : real 7m18.911s
 After : real 10.187s

Fixes: 8571ab479a6e ("ip6mr: Make mroute_sk rcu-based")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reported-by: Kirill Tkhai <ktkhai@virtuozzo.com>
Cc: Yuval Mintz <yuvalm@mellanox.com>
Reviewed-by: Kirill Tkhai <ktkhai@virtuozzo.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
net/ipv6/ip6mr.c