ALSA: pcm : Call kill_fasync() in stream lock 57/155057/1
authorTakashi Iwai <tiwai@suse.de>
Mon, 12 Dec 2016 16:33:06 +0000 (17:33 +0100)
committerSeung-Woo Kim <sw0312.kim@samsung.com>
Thu, 12 Oct 2017 04:24:22 +0000 (13:24 +0900)
commit254f975ec7b47b8a6375d619b5cba4121936127a
tree3aa0ff7d08965c7cbc55c0f2ada2ea2166c16286
parent0069c946d2a34d041111157b623cddc4b0929bce
ALSA: pcm : Call kill_fasync() in stream lock

commit 3aa02cb664c5fb1042958c8d1aa8c35055a2ebc4 upstream.

Currently kill_fasync() is called outside the stream lock in
snd_pcm_period_elapsed().  This is potentially racy, since the stream
may get released even during the irq handler is running.  Although
snd_pcm_release_substream() calls snd_pcm_drop(), this doesn't
guarantee that the irq handler finishes, thus the kill_fasync() call
outside the stream spin lock may be invoked after the substream is
detached, as recently reported by KASAN.

As a quick workaround, move kill_fasync() call inside the stream
lock.  The fasync is rarely used interface, so this shouldn't have a
big impact from the performance POV.

Ideally, we should implement some sync mechanism for the proper finish
of stream and irq handler.  But this oneliner should suffice for most
cases, so far.

Reported-by: Baozeng Ding <sploving1@gmail.com>
Signed-off-by: Takashi Iwai <tiwai@suse.de>
Signed-off-by: Willy Tarreau <w@1wt.eu>
[sw0312.kim: cherry-pick from linux-3.10.y to fix CVE-2016-9794]
Signed-off-by: Seung-Woo Kim <sw0312.kim@samsung.com>
Change-Id: I1e5a74832b253d89677546175c9855c6844d7675
sound/core/pcm_lib.c