signals: fix sigqueue_free() vs __exit_signal() race
authorOleg Nesterov <oleg@tv-sign.ru>
Fri, 23 May 2008 20:04:41 +0000 (13:04 -0700)
committerLinus Torvalds <torvalds@linux-foundation.org>
Sat, 24 May 2008 16:56:10 +0000 (09:56 -0700)
commitda7978b0348d497688541e2d2f5739aa2a2c334f
treef4f55bf4293ff203f598e7a57959da1c5a7ad295
parentdfc7064500061677720fa26352963c772d3ebe6b
signals: fix sigqueue_free() vs __exit_signal() race

__exit_signal() does flush_sigqueue(tsk->pending) outside of ->siglock.
This can race with another thread doing sigqueue_free(), we can free the
same SIGQUEUE_PREALLOC sigqueue twice or corrupt the pending->list.

Note that even sys_exit_group() can trigger this race, not only
sys_timer_delete().

Move the callsite of flush_sigqueue(tsk->pending) under ->siglock.

This patch doesn't touch flush_sigqueue(->shared_pending) below, it is
called when there are no other threads which can play with signals, and
sigqueue_free() can't be used outside of our thread group.

Signed-off-by: Oleg Nesterov <oleg@tv-sign.ru>
Acked-by: Roland McGrath <roland@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
kernel/exit.c
kernel/signal.c