linux-kernel.vger.kernel.org archive mirror
 help / color / mirror / Atom feed
* net/ && prepare_to_wait_exclusive()
@ 2014-05-14 18:40 Oleg Nesterov
  0 siblings, 0 replies; only message in thread
From: Oleg Nesterov @ 2014-05-14 18:40 UTC (permalink / raw)
  To: David Miller; +Cc: Peter Zijlstra, netdev, linux-kernel

Hello,

I know almost nothing about networking and of course I do not understand
this code. So I can be easily wrong, but (at least) unix_wait_for_peer()
looks wrong wrt prepare_to_wait_exclusive(), and ignoring the potential
optimizations it needs

	-	finish_wait(&u->peer_wait, &wait);
	+	abort_exclusive_wait(..., TASK_INTERRUPTIBLE, NULL);

change?

Suppose that unix_recvq_full(X) == true and 2 threads, T1 and T2, are
sleeping in unix_dgram_sendmsg().

Another thread does read(X) and this empties ->sk_receive_queue so that
the next read(X) will block. We should wake up at least one thread.

The reader does wake_up_interruptible_sync_poll() and this is
__wake_up_common(nr_exclusive => 1), so this (say) wakes T1 up.

Suppose that this wake_up() races with (to simplify) SIGKILL and thus
T1 writes nothing and exits.

Now, the reader does another read(X) and blocks, ->sk_receive_queue is
empty, we have the writer T2 but it is still blocked too?

Or I completely misread this code?


(On a related note... looks like ___wait_event() is not complicated^W
 clever enough to implement wait_event_interruptible_exclusive_timeout)

Oleg.


^ permalink raw reply	[flat|nested] only message in thread

only message in thread, other threads:[~2014-05-14 18:41 UTC | newest]

Thread overview: (only message) (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2014-05-14 18:40 net/ && prepare_to_wait_exclusive() Oleg Nesterov

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).