As far as I can tell, this bug affects every Linux 2.6+ release ever. Reduced Cc: from original commit, as this fixes a different bug from what I intended to fix(!) and limited to epoll. Big thanks to Andreas and Jason for bringing this to our attention and testing.
----------------------------- 8< ------------------------------ >From 128dd1759d96ad36c379240f8b9463e8acfd37a1 Mon Sep 17 00:00:00 2001 From: Eric Wong <[email protected]> Date: Tue, 1 Jan 2013 21:20:27 +0000 Subject: [PATCH] epoll: prevent missed events on EPOLL_CTL_MOD commit 128dd1759d96ad36c379240f8b9463e8acfd37a1 upstream. EPOLL_CTL_MOD sets the interest mask before calling f_op->poll() to ensure events are not missed. Since the modifications to the interest mask are not protected by the same lock as ep_poll_callback, we need to ensure the change is visible to other CPUs calling ep_poll_callback. We also need to ensure f_op->poll() has an up-to-date view of past events which occured before we modified the interest mask. So this barrier also pairs with the barrier in wq_has_sleeper(). This should guarantee either ep_poll_callback or f_op->poll() (or both) will notice the readiness of a recently-ready/modified item. This issue was encountered by Andreas Voellmy and Junchang(Jason) Wang in: http://thread.gmane.org/gmane.linux.kernel/1408782/ Signed-off-by: Eric Wong <[email protected]> Cc: Al Viro <[email protected]> Cc: Davide Libenzi <[email protected]> Cc: Eric Dumazet <[email protected]> Cc: Andrew Morton <[email protected]> Cc: Andreas Voellmy <[email protected]> Tested-by: "Junchang(Jason) Wang" <[email protected]> Signed-off-by: Linus Torvalds <[email protected]> --- fs/eventpoll.c | 22 +++++++++++++++++++++- 1 file changed, 21 insertions(+), 1 deletion(-) diff --git a/fs/eventpoll.c b/fs/eventpoll.c index be56b21..9fec183 100644 --- a/fs/eventpoll.c +++ b/fs/eventpoll.c @@ -1313,7 +1313,7 @@ static int ep_modify(struct eventpoll *ep, struct epitem *epi, struct epoll_even * otherwise we might miss an event that happens between the * f_op->poll() call and the new event set registering. */ - epi->event.events = event->events; + epi->event.events = event->events; /* need barrier below */ pt._key = event->events; epi->event.data = event->data; /* protected by mtx */ if (epi->event.events & EPOLLWAKEUP) { @@ -1324,6 +1324,26 @@ static int ep_modify(struct eventpoll *ep, struct epitem *epi, struct epoll_even } /* + * The following barrier has two effects: + * + * 1) Flush epi changes above to other CPUs. This ensures + * we do not miss events from ep_poll_callback if an + * event occurs immediately after we call f_op->poll(). + * We need this because we did not take ep->lock while + * changing epi above (but ep_poll_callback does take + * ep->lock). + * + * 2) We also need to ensure we do not miss _past_ events + * when calling f_op->poll(). This barrier also + * pairs with the barrier in wq_has_sleeper (see + * comments for wq_has_sleeper). + * + * This barrier will now guarantee ep_poll_callback or f_op->poll + * (or both) will notice the readiness of an item. + */ + smp_mb(); + + /* * Get current event bits. We can safely use the file* here because * its usage count has been increased by the caller of this function. */ -- Eric Wong (!) http://mid.gmane.org/[email protected] http://mid.gmane.org/[email protected] -- To unsubscribe from this list: send the line "unsubscribe stable" in the body of a message to [email protected] More majordomo info at http://vger.kernel.org/majordomo-info.html
