[PATCH v2] ASoC: soc-compress: fix UAF in soc_compr_trigger_fe()
"Shinhyung Kang" <[email protected]> Tue, 23 Jun 2026 20:31:57 +0900
| Newsgroups | org.alsa-project.alsa-devel |
|---|---|
| Message-ID | <[email protected]> |
The DPCM compress trigger path traverses the FE's BE client list in
dpcm_be_dai_trigger() without holding card->pcm_mutex, while
dpcm_be_disconnect() can concurrently remove and free entries from
that same list under pcm_mutex protection.
This causes a use-after-free when for_each_dpcm_be() advances to the
next list node after releasing a BE's stream lock between iterations,
and the snd_soc_dpcm entry has already been kfree()'d by a concurrent
dpcm_be_disconnect() call.
Crash signature observed:
Unable to handle kernel paging request at virtual address dead0000000000e8
Call trace:
dpcm_be_dai_trigger+0x90/0x3f0
soc_compr_trigger_fe+0xa8/0x144
snd_compr_ioctl+0xc98/0x2010
Race condition timeline:
Thread A(soc_compr_trigger_fe):
snd_soc_card_mutex_lock() <- holds card->mutex only
dpcm_be_dai_trigger()
for_each_dpcm_be(fe, stream, dpcm) {
snd_pcm_stream_lock_irqsave_nested(be_substream);
...
snd_pcm_stream_unlock_irqrestore(be_substream);
/* WINDOW: next iteration reads dpcm->list_be.next */
}
Thread B(snd_soc_dpcm_runtime_update via DAPM):
snd_soc_dpcm_mutex_lock() <- holds card->pcm_mutex
dpcm_be_disconnect()
snd_pcm_stream_lock_irq(fe_substream);
list_del(&dpcm->list_be); <- removes from list
snd_pcm_stream_unlock_irq();
kfree(dpcm); <- frees the struct
The PCM trigger path (dpcm_fe_dai_trigger) is protected against this
race by checking runtime_update and deferring to trigger_pending when
a concurrent update is in progress. The compress trigger path
(soc_compr_trigger_fe) lacks this deferred-trigger mechanism, so the
only correct fix is to hold pcm_mutex for the duration of the BE list
traversal, as is done in all other compress FE operations such as
soc_compr_open_fe() and soc_compr_set_params().
Signed-off-by: Shinhyung Kang <[email protected]>
---
Changes in v2:
- Reworded commit message for clarity.
- Resend due to mail client corruption in v1.
- No functional changes.
Link to v1:
https://lore.kernel.org/alsa-devel/[email protected]
om
---
sound/soc/soc-compress.c | 2 ++
1 file changed, 2 insertions(+)
diff --git a/sound/soc/soc-compress.c b/sound/soc/soc-compress.c
index b8402802ae78..615ce7a0e8d9 100644
--- a/sound/soc/soc-compress.c
+++ b/sound/soc/soc-compress.c
@@ -285,6 +285,7 @@ static int soc_compr_trigger_fe(struct snd_compr_stream
*cstream, int cmd)
return snd_soc_component_compr_trigger(cstream, cmd);
snd_soc_card_mutex_lock(fe->card);
+ snd_soc_dpcm_mutex_lock(fe);
ret = snd_soc_dai_compr_trigger(cpu_dai, cstream, cmd);
if (ret < 0)
@@ -315,6 +316,7 @@ static int soc_compr_trigger_fe(struct snd_compr_stream
*cstream, int cmd)
out:
fe->dpcm[stream].runtime_update = SND_SOC_DPCM_UPDATE_NO;
+ snd_soc_dpcm_mutex_unlock(fe);
snd_soc_card_mutex_unlock(fe->card);
return ret;
}
--
2.21.0