exec: kill unsafe BUG_ON(sig->count) checks
authorOleg Nesterov <oleg@tv-sign.ru>
Wed, 22 Aug 2007 21:01:58 +0000 (14:01 -0700)
committerLinus Torvalds <torvalds@woody.linux-foundation.org>
Thu, 23 Aug 2007 02:52:47 +0000 (19:52 -0700)
commitabd96ecb298675a21c412a29f5de2f80174d5f18
tree5fb722fcfc224bf3e69a06a9e87d8fe56e1a6548
parent5c076fce2e217240b44bc753a5ec8ecd379c6eb9
exec: kill unsafe BUG_ON(sig->count) checks

de_thread:

if (atomic_read(&oldsighand->count) <= 1)
BUG_ON(atomic_read(&sig->count) != 1);

This is not safe without the rmb() in between.  The results of two
correctly ordered __exit_signal()->atomic_dec_and_test()'s could be seen
out of order on our CPU.

The same is true for the "thread_group_empty()" case, __unhash_process()'s
changes could be seen before atomic_dec_and_test(&sig->count).

On some platforms (including i386) atomic_read() doesn't provide even the
compiler barrier, in that case these checks are simply racy.

Remove these BUG_ON()'s. Alternatively, we can do something like

BUG_ON( ({ smp_rmb(); atomic_read(&sig->count) != 1; }) );

Signed-off-by: Oleg Nesterov <oleg@tv-sign.ru>
Acked-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Cc: Roland McGrath <roland@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
fs/exec.c