Re: WARNING in snd_pcm_hw_params

[Takashi Iwai <tiwai@suse.de>]

On Mon, 09 Apr 2018 17:33:45 +0200,
Takashi Iwai wrote:
> 
> On Mon, 09 Apr 2018 17:25:15 +0200,
> Dmitry Vyukov wrote:
> > 
> > On Mon, Apr 9, 2018 at 5:18 PM, Takashi Iwai <tiwai@suse.de> wrote:
> > > On Fri, 06 Apr 2018 02:02:02 +0200,
> > > syzbot wrote:
> > >>
> > >> WARNING: CPU: 0 PID: 4468 at sound/core/pcm_native.c:620
> > >> snd_pcm_hw_params_choose sound/core/pcm_native.c:620 [inline]
> > >
> > > This is a WARN_ON() to catch the inconsistent hw_params call.
> > > This shouldn't happen in theory, but the reality looks weirder.
> > >
> > > The necessary "fix" is just to get rid of snd_BUG_ON(), but still
> > > wonder which condition it hits.  If it's an empty parameter, the
> > > previous snd_pcm_hw_refine() should have caught and returned the error
> > > beforehand...
> > 
> > Hi Takashi,
> > 
> > The reproducer for this is trivially simple:
> > 
> > https://syzkaller.appspot.com/x/repro.syz?id=4913841943085056
> > 
> > (and I guess perf_event_open is red herring).
> > 
> > But the really important part there is probably this: Fault:true
> > FaultCall:2 FaultNth:18
> > And you can see where it actually happened in the raw console log:
> > 
> > Raw console output: https://syzkaller.appspot.com/x/log.txt?id=6314122077536256
> > 
> > Does it explain things?
> 
> Yes, it does.  So it's the kmalloc error from snd_pcm_hw_refine().
> I overlooked this, as this kmalloc is relatively new, and that
> possibility wasn't considered in the WARN_ON() in question.
> 
> I'll rephrase the changelog text.

... and below is the revised patch.  I'll queue for 4.17-rc1 merge.


Takashi

-- 8< --
From: Takashi Iwai <tiwai@suse.de>
Subject: [PATCH] ALSA: pcm: Remove WARN_ON() at snd_pcm_hw_params() error

snd_pcm_hw_params() (more exactly snd_pcm_hw_params_choose()) contains
a check of the return error from snd_pcm_hw_param_first() and _last()
with snd_BUG_ON() -- i.e. it may trigger WARN_ON() depending on the
kconfig.

This was a valid check in the past, as these functions shouldn't
return any error if the parameters have been already refined via
snd_pcm_hw_refine() beforehand.  However, the recent rewrite
introduced a kmalloc() in snd_pcm_hw_refine() for removing VLA, and
this brought a possibility to trigger an error.  As a result, syzbot
caught lots of superfluous kernel WARN_ON() and paniced via fault
injection.

As the WARN_ON() is no longer valid with the introduction of
kmalloc(), let's drop snd_BUG_ON() check, in order to make the world
peaceful place again.

Reported-by: syzbot+803e0047ac3a3096bb4f@syzkaller.appspotmail.com
Fixes: 5730f9f744cf ("ALSA: pcm: Remove VLA usage")
Signed-off-by: Takashi Iwai <tiwai@suse.de>
---
 sound/core/pcm_native.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/sound/core/pcm_native.c b/sound/core/pcm_native.c
index b84554893fab..35ffccea94c3 100644
--- a/sound/core/pcm_native.c
+++ b/sound/core/pcm_native.c
@@ -617,7 +617,7 @@ static int snd_pcm_hw_params_choose(struct snd_pcm_substream *pcm,
 			changed = snd_pcm_hw_param_first(pcm, params, *v, NULL);
 		else
 			changed = snd_pcm_hw_param_last(pcm, params, *v, NULL);
-		if (snd_BUG_ON(changed < 0))
+		if (changed < 0)
 			return changed;
 		if (changed == 0)
 			continue;
-- 
2.16.3