[PATCH] ALSA: pcm: Test for "silence" field in struct "pcm_format_data"

[Takashi Iwai]

On Tue, 14 Jun 2022 11:58:51 +0200,
Eugeniu Rosca wrote:
> 
> Hello Fabio, hello All,
> 
> On Sa, Apr 09, 2022 at 03:26:55 +0200, Fabio M. De Francesco wrote:
> > Syzbot reports "KASAN: null-ptr-deref Write in
> > snd_pcm_format_set_silence".[1]
> > 
> > It is due to missing validation of the "silence" field of struct
> > "pcm_format_data" in "pcm_formats" array.
> > 
> > Add a test for valid "pat" and, if it is not so, return -EINVAL.
> > 
> > [1] https://lore.kernel.org/lkml/000000000000d188ef05dc2c7279@google.com/
> > 
> > Reported-and-tested-by: syzbot+205eb15961852c2c5974 at syzkaller.appspotmail.com
> > Signed-off-by: Fabio M. De Francesco <fmdefrancesco at gmail.com>
> > ---
> > 
> > I wasn't able to figure out the commit for the "Fixes:" tag. If this patch
> > is good, can someone please help with providing this missing information?
> > 
> >  sound/core/pcm_misc.c | 2 +-
> >  1 file changed, 1 insertion(+), 1 deletion(-)
> > 
> > diff --git a/sound/core/pcm_misc.c b/sound/core/pcm_misc.c
> > index 4866aed97aac..5588b6a1ee8b 100644
> > --- a/sound/core/pcm_misc.c
> > +++ b/sound/core/pcm_misc.c
> > @@ -433,7 +433,7 @@ int snd_pcm_format_set_silence(snd_pcm_format_t format, void *data, unsigned int
> >  		return 0;
> >  	width = pcm_formats[(INT)format].phys; /* physical width */
> >  	pat = pcm_formats[(INT)format].silence;
> > -	if (! width)
> > +	if (!width || !pat)
> >  		return -EINVAL;
> >  	/* signed or 1 byte data */
> >  	if (pcm_formats[(INT)format].signd == 1 || width <= 8) {
> 
> JFYI, PVS-Studio 7.19 reports:
> 
> sound/core/pcm_misc.c	409	warn	V560 A part of conditional expression is always false: !pat.
> 
> I haven't fully validated the finding, but it appears to be legit,
> since the pointer variable (as opposed to the contents behind the
> pointer) is always non-null, hence !pat always evaluating to false.
> 
> If the above is true, then the patch likely hasn't introduced any
> regression, but also likely hasn't fixed the original KASAN problem.
> 
> Or are there alternative views?

Indeed the fix looks bogus, and maybe better to revert.

Looking at the original syzkaller report again, it points rather to
the *write* at the address 1, and it means not the source (silence[])
but the target pointer (data) is invalid; i.e. it's a problem in the
caller side, likely some race between the OSS temporary buffer removal
and other operation.

Thanks for checking this.


Takashi