tcp: Combat per-cpu skew in orphan tests.
authorDavid S. Miller <davem@davemloft.net>
Wed, 25 Aug 2010 09:27:49 +0000 (02:27 -0700)
committerDavid S. Miller <davem@davemloft.net>
Wed, 25 Aug 2010 09:27:49 +0000 (02:27 -0700)
commitad1af0fedba14f82b240a03fe20eb9b2fdbd0357
tree4d53aa8bc2d9df782aa792e52670ab55c7a44d5b
parentb2bc85631e72485b984bcd202a104591874babba
tcp: Combat per-cpu skew in orphan tests.

As reported by Anton Blanchard when we use
percpu_counter_read_positive() to make our orphan socket limit checks,
the check can be off by up to num_cpus_online() * batch (which is 32
by default) which on a 128 cpu machine can be as large as the default
orphan limit itself.

Fix this by doing the full expensive sum check if the optimized check
triggers.

Reported-by: Anton Blanchard <anton@samba.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
Acked-by: Eric Dumazet <eric.dumazet@gmail.com>
include/net/tcp.h
net/ipv4/tcp.c
net/ipv4/tcp_timer.c