Re: [PATCH] net: fix race in the receive/select

[Oleg Nesterov <oleg@redhat.com>]

On 06/25, Davide Libenzi wrote:
>
> Not all the code that uses add_wait_queue() does need to have the MB,
> like code that does the most common pattern:
>
> xxx_poll(...) {
> 	poll_wait(...);
> 	lock();
> 	flags = calc_flags(->status);
> 	unlock();
> 	return flags;
> }
>
> xxx_update(...) {
> 	lock();
> 	->status = ...;
> 	unlock();
> 	if (waitqueue_active())
> 		wake_up();
> }
>
> It's the code that does the lockless flags calculation in ->poll that
> might need it.

And if we remove waitqueue_active() in xxx_update(), then lock/unlock is
not needed too.

If xxx_poll() takes q->lock first, it can safely miss the changes in ->status
and schedule(): xxx_update() will take q->lock, notice the sleeper and wake
it up (ok, it will set ->triggered but this doesn't matter).

If xxx_update() takes q->lock first, xxx_poll() must see the changes in
status after poll_wait()->unlock(&q->lock) (in fact, after lock, not unlock).

Oleg.