On Wed, 2013-01-02 at 18:40 +0000, Eric Wong wrote:

Eric Dumazet [off-list ref] wrote:

quoted

First, thanks for working on this issue.

No problem!

quoted

It seems the real problem is the epi->event.events = event->events;
which is done without taking ep->lock

Yes.  I am hoping it is possible to do it without a lock there,
but your change is more obviously correct.

quoted

While a smp_mb() could reduce the race window, I believe there is still
a race, and the following patch would close it.

I'm not an experienced kernel hacker, can you describe where the race
would be?

It would be for example in ep_send_events_proc() doing :

if (epi->event.events & EPOLLONESHOT)
    epi->event.events &= EP_PRIVATE_BITS;

And this could happen at the same time.