]> git.baikalelectronics.ru Git - kernel.git/commit
tcp/dccp: block bh before arming time_wait timer
authorEric Dumazet <edumazet@google.com>
Fri, 1 Dec 2017 18:06:56 +0000 (10:06 -0800)
committerDavid S. Miller <davem@davemloft.net>
Fri, 1 Dec 2017 20:07:43 +0000 (15:07 -0500)
commit1b6ae69b32f9ee2670d53e597e40d1e652f61865
tree129ca3745010edb380f18e9420ea50ede3b8b1bc
parent38b563c4cc3518ed8f29643dd41e2613ce69caf8
tcp/dccp: block bh before arming time_wait timer

Maciej Żenczykowski reported some panics in tcp_twsk_destructor()
that might be caused by the following bug.

timewait timer is pinned to the cpu, because we want to transition
timwewait refcount from 0 to 4 in one go, once everything has been
initialized.

At the time commit ae418b4f601b ("tcp/dccp: fix timewait races in timer
handling") was merged, TCP was always running from BH habdler.

After commit 54a874cf32f4 ("net: do not block BH while processing
socket backlog") we definitely can run tcp_time_wait() from process
context.

We need to block BH in the critical section so that the pinned timer
has still its purpose.

This bug is more likely to happen under stress and when very small RTO
are used in datacenter flows.

Fixes: 54a874cf32f4 ("net: do not block BH while processing socket backlog")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reported-by: Maciej Żenczykowski <maze@google.com>
Acked-by: Maciej Żenczykowski <maze@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
net/dccp/minisocks.c
net/ipv4/tcp_minisocks.c