[PATCH] ALSA: usb-audio: fix race condition to UAF in snd_usbmidi_free

[PATCH] ALSA: usb-audio: fix race condition to UAF in snd_usbmidi_free

[Jeongjun Park]

The previous commit 0718a78f6a9f ("ALSA: usb-audio: Kill timer properly at
removal") patched a UAF issue caused by the error timer.

However, because the error timer kill added in this patch occurs after the
endpoint delete, a race condition to UAF still occurs, albeit rarely.

Therefore, to prevent this, the error timer must be killed before freeing
the heap memory.

Cc: <stable@vger.kernel.org>
Fixes: 0718a78f6a9f ("ALSA: usb-audio: Kill timer properly at removal")
Signed-off-by: Jeongjun Park <aha310510@gmail.com>
---
 sound/usb/midi.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/sound/usb/midi.c b/sound/usb/midi.c
index acb3bf92857c..8d15f1caa92b 100644
--- a/sound/usb/midi.c
+++ b/sound/usb/midi.c
@@ -1522,6 +1522,8 @@ static void snd_usbmidi_free(struct snd_usb_midi *umidi)
 {
 	int i;
 
+	timer_shutdown_sync(&umidi->error_timer);
+
 	for (i = 0; i < MIDI_MAX_ENDPOINTS; ++i) {
 		struct snd_usb_midi_endpoint *ep = &umidi->endpoints[i];
 		if (ep->out)
@@ -1530,7 +1532,6 @@ static void snd_usbmidi_free(struct snd_usb_midi *umidi)
 			snd_usbmidi_in_endpoint_delete(ep->in);
 	}
 	mutex_destroy(&umidi->mutex);
-	timer_shutdown_sync(&umidi->error_timer);
 	kfree(umidi);
 }
 
--

Re: [PATCH] ALSA: usb-audio: fix race condition to UAF in snd_usbmidi_free

[Takashi Iwai]

On Sat, 27 Sep 2025 06:41:06 +0200,
Jeongjun Park wrote:
> 
> The previous commit 0718a78f6a9f ("ALSA: usb-audio: Kill timer properly at
> removal") patched a UAF issue caused by the error timer.
> 
> However, because the error timer kill added in this patch occurs after the
> endpoint delete, a race condition to UAF still occurs, albeit rarely.
> 
> Therefore, to prevent this, the error timer must be killed before freeing
> the heap memory.
> 
> Cc: <stable@vger.kernel.org>
> Fixes: 0718a78f6a9f ("ALSA: usb-audio: Kill timer properly at removal")
> Signed-off-by: Jeongjun Park <aha310510@gmail.com>

I suppose it's a fix for the recent syzbot reports?
  https://lore.kernel.org/68d17f44.050a0220.13cd81.05b7.GAE@google.com
  https://lore.kernel.org/68d38327.a70a0220.1b52b.02be.GAE@google.com

I had the very same fix in mind, as posted in
  https://lore.kernel.org/87plbhn16a.wl-tiwai@suse.de
so I'll happily apply if that's the case (and it was verified to
work).  I'm just back from vacation and trying to catch up things.


thanks,

Takashi

> ---
>  sound/usb/midi.c | 3 ++-
>  1 file changed, 2 insertions(+), 1 deletion(-)
> 
> diff --git a/sound/usb/midi.c b/sound/usb/midi.c
> index acb3bf92857c..8d15f1caa92b 100644
> --- a/sound/usb/midi.c
> +++ b/sound/usb/midi.c
> @@ -1522,6 +1522,8 @@ static void snd_usbmidi_free(struct snd_usb_midi *umidi)
>  {
>  	int i;
>  
> +	timer_shutdown_sync(&umidi->error_timer);
> +
>  	for (i = 0; i < MIDI_MAX_ENDPOINTS; ++i) {
>  		struct snd_usb_midi_endpoint *ep = &umidi->endpoints[i];
>  		if (ep->out)
> @@ -1530,7 +1532,6 @@ static void snd_usbmidi_free(struct snd_usb_midi *umidi)
>  			snd_usbmidi_in_endpoint_delete(ep->in);
>  	}
>  	mutex_destroy(&umidi->mutex);
> -	timer_shutdown_sync(&umidi->error_timer);
>  	kfree(umidi);
>  }
>  
> --

Re: [PATCH] ALSA: usb-audio: fix race condition to UAF in snd_usbmidi_free

[Jeongjun Park]

Hi,

Takashi Iwai <tiwai@suse.de> wrote:
>
> On Sat, 27 Sep 2025 06:41:06 +0200,
> Jeongjun Park wrote:
> >
> > The previous commit 0718a78f6a9f ("ALSA: usb-audio: Kill timer properly at
> > removal") patched a UAF issue caused by the error timer.
> >
> > However, because the error timer kill added in this patch occurs after the
> > endpoint delete, a race condition to UAF still occurs, albeit rarely.
> >
> > Therefore, to prevent this, the error timer must be killed before freeing
> > the heap memory.
> >
> > Cc: <stable@vger.kernel.org>
> > Fixes: 0718a78f6a9f ("ALSA: usb-audio: Kill timer properly at removal")
> > Signed-off-by: Jeongjun Park <aha310510@gmail.com>
>
> I suppose it's a fix for the recent syzbot reports?
>   https://lore.kernel.org/68d17f44.050a0220.13cd81.05b7.GAE@google.com
>   https://lore.kernel.org/68d38327.a70a0220.1b52b.02be.GAE@google.com
>

Oh, I didn't know it was already reported on syzbot.

> I had the very same fix in mind, as posted in
>   https://lore.kernel.org/87plbhn16a.wl-tiwai@suse.de
> so I'll happily apply if that's the case (and it was verified to
> work).  I'm just back from vacation and trying to catch up things.
>

Although it's difficult to disclose right now, I have already completed
writing a PoC that triggers a UAF due to the error timer in a slightly
different way than the backtrace reported to syzbot, and I have confirmed
that no bugs occur when testing this patch through this PoC.

>
> thanks,
>
> Takashi
>
> > ---
> >  sound/usb/midi.c | 3 ++-
> >  1 file changed, 2 insertions(+), 1 deletion(-)
> >
> > diff --git a/sound/usb/midi.c b/sound/usb/midi.c
> > index acb3bf92857c..8d15f1caa92b 100644
> > --- a/sound/usb/midi.c
> > +++ b/sound/usb/midi.c
> > @@ -1522,6 +1522,8 @@ static void snd_usbmidi_free(struct snd_usb_midi *umidi)
> >  {
> >       int i;
> >
> > +     timer_shutdown_sync(&umidi->error_timer);
> > +
> >       for (i = 0; i < MIDI_MAX_ENDPOINTS; ++i) {
> >               struct snd_usb_midi_endpoint *ep = &umidi->endpoints[i];
> >               if (ep->out)
> > @@ -1530,7 +1532,6 @@ static void snd_usbmidi_free(struct snd_usb_midi *umidi)
> >                       snd_usbmidi_in_endpoint_delete(ep->in);
> >       }
> >       mutex_destroy(&umidi->mutex);
> > -     timer_shutdown_sync(&umidi->error_timer);
> >       kfree(umidi);
> >  }
> >
> > --

Regards,
Jeongjun Park

Re: [PATCH] ALSA: usb-audio: fix race condition to UAF in snd_usbmidi_free

[Takashi Iwai]

On Sat, 27 Sep 2025 10:48:02 +0200,
Jeongjun Park wrote:
> 
> Hi,
> 
> Takashi Iwai <tiwai@suse.de> wrote:
> >
> > On Sat, 27 Sep 2025 06:41:06 +0200,
> > Jeongjun Park wrote:
> > >
> > > The previous commit 0718a78f6a9f ("ALSA: usb-audio: Kill timer properly at
> > > removal") patched a UAF issue caused by the error timer.
> > >
> > > However, because the error timer kill added in this patch occurs after the
> > > endpoint delete, a race condition to UAF still occurs, albeit rarely.
> > >
> > > Therefore, to prevent this, the error timer must be killed before freeing
> > > the heap memory.
> > >
> > > Cc: <stable@vger.kernel.org>
> > > Fixes: 0718a78f6a9f ("ALSA: usb-audio: Kill timer properly at removal")
> > > Signed-off-by: Jeongjun Park <aha310510@gmail.com>
> >
> > I suppose it's a fix for the recent syzbot reports?
> >   https://lore.kernel.org/68d17f44.050a0220.13cd81.05b7.GAE@google.com
> >   https://lore.kernel.org/68d38327.a70a0220.1b52b.02be.GAE@google.com
> >
> 
> Oh, I didn't know it was already reported on syzbot.
> 
> > I had the very same fix in mind, as posted in
> >   https://lore.kernel.org/87plbhn16a.wl-tiwai@suse.de
> > so I'll happily apply if that's the case (and it was verified to
> > work).  I'm just back from vacation and trying to catch up things.
> >
> 
> Although it's difficult to disclose right now, I have already completed
> writing a PoC that triggers a UAF due to the error timer in a slightly
> different way than the backtrace reported to syzbot, and I have confirmed
> that no bugs occur when testing this patch through this PoC.

OK, so this sounds like a coincidence, but it's very likely the same
issue, so I'm going to put mark those syzbot reports.


thanks,

Takashi