* [PATCH] ALSA: pcm: keep linked runtime alive while draining
@ 2026-06-05 5:16 Guangshuo Li
2026-06-05 6:24 ` Takashi Iwai
0 siblings, 1 reply; 3+ messages in thread
From: Guangshuo Li @ 2026-06-05 5:16 UTC (permalink / raw)
To: Jaroslav Kysela, Takashi Iwai, Christophe Leroy, Mehul Rao,
Nathan Chancellor, Cássio Gabriel, Guangshuo Li, Cen Zhang,
Sebastian Andrzej Siewior, linux-sound, linux-kernel
snd_pcm_drain() may select a runtime from a linked playback stream and
queue a stack wait entry on its embedded sleep waitqueue. The function
then drops the stream lock and sleeps before removing the wait entry.
commit 9b1dbd69ba6f ("ALSA: pcm: fix use-after-free on linked stream
runtime in snd_pcm_drain()") cached scalar fields from the linked
runtime before dropping the stream lock, because a concurrent close can
unlink the stream and free its runtime. The same lifetime issue remains
for runtime->sleep. If the linked runtime is detached while the wait
entry is queued, remove_wait_queue() can later operate on a freed or
reused waitqueue, or the stack wait entry can remain linked to freed
memory.
Take snd_pcm_link_rwsem for read while the wait entry is queued on the
linked runtime. Acquire it before re-taking the stream lock, matching the
link/unlink lock order, and drop it only after the wait entry has been
removed.
Fixes: 21cb2a2ec581 ("[ALSA] Fix races between PCM drain and other ops")
Signed-off-by: Guangshuo Li <lgs201920130244@gmail.com>
---
sound/core/pcm_native.c | 36 +++++++++++++++++++++++++++---------
1 file changed, 27 insertions(+), 9 deletions(-)
diff --git a/sound/core/pcm_native.c b/sound/core/pcm_native.c
index a541bb235cfa..1f7aba988e45 100644
--- a/sound/core/pcm_native.c
+++ b/sound/core/pcm_native.c
@@ -2175,6 +2175,22 @@ static int snd_pcm_drain(struct snd_pcm_substream *substream,
result = -ERESTARTSYS;
break;
}
+
+ /*
+ * The wait below is queued on to_check->sleep, which is embedded
+ * in a runtime that may belong to a linked substream. Link/unlink
+ * paths take snd_pcm_link_rwsem for write before moving streams
+ * between groups and detaching runtimes, so take it for read while
+ * the wait entry is queued.
+ *
+ * Drop and re-take the stream lock so the lock order matches
+ * snd_pcm_link() and snd_pcm_unlink(): link_rwsem first, then the
+ * stream/group lock.
+ */
+ snd_pcm_stream_unlock_irq(substream);
+ down_read(&snd_pcm_link_rwsem);
+ snd_pcm_stream_lock_irq(substream);
+
/* find a substream to drain */
to_check = NULL;
group = snd_pcm_stream_group_ref(substream);
@@ -2188,8 +2204,14 @@ static int snd_pcm_drain(struct snd_pcm_substream *substream,
}
}
snd_pcm_group_unref(group, substream);
- if (!to_check)
+
+ if (!to_check) {
+ snd_pcm_stream_unlock_irq(substream);
+ up_read(&snd_pcm_link_rwsem);
+ snd_pcm_stream_lock_irq(substream);
break; /* all drained */
+ }
+
/*
* Cache the runtime fields needed after unlock.
* A concurrent close() on the linked stream may free
@@ -2216,14 +2238,10 @@ static int snd_pcm_drain(struct snd_pcm_substream *substream,
tout = schedule_timeout(tout);
snd_pcm_stream_lock_irq(substream);
- group = snd_pcm_stream_group_ref(substream);
- snd_pcm_group_for_each_entry(s, substream) {
- if (s->runtime == to_check) {
- remove_wait_queue(&to_check->sleep, &wait);
- break;
- }
- }
- snd_pcm_group_unref(group, substream);
+ remove_wait_queue(&to_check->sleep, &wait);
+ snd_pcm_stream_unlock_irq(substream);
+ up_read(&snd_pcm_link_rwsem);
+ snd_pcm_stream_lock_irq(substream);
if (card->shutdown) {
result = -ENODEV;
--
2.43.0
^ permalink raw reply [flat|nested] 3+ messages in thread
* Re: [PATCH] ALSA: pcm: keep linked runtime alive while draining
2026-06-05 5:16 [PATCH] ALSA: pcm: keep linked runtime alive while draining Guangshuo Li
@ 2026-06-05 6:24 ` Takashi Iwai
2026-06-09 15:23 ` Guangshuo Li
0 siblings, 1 reply; 3+ messages in thread
From: Takashi Iwai @ 2026-06-05 6:24 UTC (permalink / raw)
To: Guangshuo Li
Cc: Jaroslav Kysela, Takashi Iwai, Christophe Leroy, Mehul Rao,
Nathan Chancellor, Cássio Gabriel, Cen Zhang,
Sebastian Andrzej Siewior, linux-sound, linux-kernel
On Fri, 05 Jun 2026 07:16:37 +0200,
Guangshuo Li wrote:
>
> snd_pcm_drain() may select a runtime from a linked playback stream and
> queue a stack wait entry on its embedded sleep waitqueue. The function
> then drops the stream lock and sleeps before removing the wait entry.
>
> commit 9b1dbd69ba6f ("ALSA: pcm: fix use-after-free on linked stream
> runtime in snd_pcm_drain()") cached scalar fields from the linked
> runtime before dropping the stream lock, because a concurrent close can
> unlink the stream and free its runtime. The same lifetime issue remains
> for runtime->sleep. If the linked runtime is detached while the wait
> entry is queued, remove_wait_queue() can later operate on a freed or
> reused waitqueue, or the stack wait entry can remain linked to freed
> memory.
>
> Take snd_pcm_link_rwsem for read while the wait entry is queued on the
> linked runtime. Acquire it before re-taking the stream lock, matching the
> link/unlink lock order, and drop it only after the wait entry has been
> removed.
>
> Fixes: 21cb2a2ec581 ("[ALSA] Fix races between PCM drain and other ops")
> Signed-off-by: Guangshuo Li <lgs201920130244@gmail.com>
This should have been already fixed by the commit 88fe2e365872
ALSA: PCM: Fix wait queue list corruption in snd_pcm_drain() on
linked streams
in for-linus branch of sound.git tree.
Please check whether the issue still occurs.
thanks,
Takashi
^ permalink raw reply [flat|nested] 3+ messages in thread
* Re: [PATCH] ALSA: pcm: keep linked runtime alive while draining
2026-06-05 6:24 ` Takashi Iwai
@ 2026-06-09 15:23 ` Guangshuo Li
0 siblings, 0 replies; 3+ messages in thread
From: Guangshuo Li @ 2026-06-09 15:23 UTC (permalink / raw)
To: Takashi Iwai
Cc: Jaroslav Kysela, Takashi Iwai, Christophe Leroy, Mehul Rao,
Nathan Chancellor, Cássio Gabriel, Cen Zhang,
Sebastian Andrzej Siewior, linux-sound, linux-kernel
Hi Takashi,
Thanks for the review and the pointer.
On Fri, 5 Jun 2026 at 14:24, Takashi Iwai <tiwai@suse.de> wrote:
>
> On Fri, 05 Jun 2026 07:16:37 +0200,
> Guangshuo Li wrote:
> >
> > snd_pcm_drain() may select a runtime from a linked playback stream and
> > queue a stack wait entry on its embedded sleep waitqueue. The function
> > then drops the stream lock and sleeps before removing the wait entry.
> >
> > commit 9b1dbd69ba6f ("ALSA: pcm: fix use-after-free on linked stream
> > runtime in snd_pcm_drain()") cached scalar fields from the linked
> > runtime before dropping the stream lock, because a concurrent close can
> > unlink the stream and free its runtime. The same lifetime issue remains
> > for runtime->sleep. If the linked runtime is detached while the wait
> > entry is queued, remove_wait_queue() can later operate on a freed or
> > reused waitqueue, or the stack wait entry can remain linked to freed
> > memory.
> >
> > Take snd_pcm_link_rwsem for read while the wait entry is queued on the
> > linked runtime. Acquire it before re-taking the stream lock, matching the
> > link/unlink lock order, and drop it only after the wait entry has been
> > removed.
> >
> > Fixes: 21cb2a2ec581 ("[ALSA] Fix races between PCM drain and other ops")
> > Signed-off-by: Guangshuo Li <lgs201920130244@gmail.com>
>
> This should have been already fixed by the commit 88fe2e365872
> ALSA: PCM: Fix wait queue list corruption in snd_pcm_drain() on
> linked streams
> in for-linus branch of sound.git tree.
> Please check whether the issue still occurs.
>
>
> thanks,
>
> Takashi
I missed commit 88fe2e365872 ("ALSA: PCM: Fix wait queue list corruption
in snd_pcm_drain() on linked streams") in the sound.git for-linus branch.
After checking it, it addresses the waitqueue corruption case I was
looking at.
Thanks,
Guangshuo
^ permalink raw reply [flat|nested] 3+ messages in thread
end of thread, other threads:[~2026-06-09 15:23 UTC | newest]
Thread overview: 3+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2026-06-05 5:16 [PATCH] ALSA: pcm: keep linked runtime alive while draining Guangshuo Li
2026-06-05 6:24 ` Takashi Iwai
2026-06-09 15:23 ` Guangshuo Li
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox
all inboxes | Powered by JetHome®