sound: hdmi: avoid dereferencing uninitialized 'jack' pointer

Commit Message

Takashi Iwai Feb. 16, 2016, 5:10 p.m. UTC

On Tue, 16 Feb 2016 17:59:04 +0100,
Arnd Bergmann wrote:
> 
> On Tuesday 16 February 2016 16:38:40 Mark Brown wrote:
> > On Tue, Feb 16, 2016 at 05:18:29PM +0100, Takashi Iwai wrote:
> > > Arnd Bergmann wrote:
> > 
> > > > Another option might would be to change snd_jack_new() to return
> > > > an error if that SND_JACK is disabled, and then require all users
> > > > to handle the error gracefully, i.e. not fail the probe() function
> > > > but just not use the jack.
> > 
> > > Yes, I thought of that, too.  If select is no good option, it's a good
> > > alternative, indeed.
> > 
> > It's going to be a bunch of work to implement though.
> > 
> 
> I've already sent a v2 to change the snd_jack_new() function, feel free
> to ignore that. I also saw now that the same bug is present in hda_jack.c,
> but I think the other drivers are fine.
> 
> How about this approach below? That should also make it possible to
> use the jack APIs without using a error return.

I prefer setting NULL explicitly in snd_jack_new(), and let callers
checking NULL.  Then we can avoid ugly IS_ENABLE() while the compiler
should be still capable to optimize out.

OTOH, I think it'd be a waste of time to bikeshedding too much, so I
don't care so much which to take :)


Takashi

---

Comments

Arnd Bergmann Feb. 16, 2016, 5:26 p.m. UTC | #1

On Tuesday 16 February 2016 18:10:02 Takashi Iwai wrote:
> diff --git a/include/sound/jack.h b/include/sound/jack.h
> index 23bede121c78..a27c253a3207 100644
> --- a/include/sound/jack.h
> +++ b/include/sound/jack.h
> @@ -99,6 +99,7 @@ void snd_jack_report(struct snd_jack *jack, int status);
>  static inline int snd_jack_new(struct snd_card *card, const char *id, int type,
>                                struct snd_jack **jack, bool initial_kctl, bool phantom_jack)
>  {
> +       *jack = NULL;
>         return 0;
>  }
>  
> diff --git a/sound/pci/hda/hda_jack.c b/sound/pci/hda/hda_jack.c
> index a33234e04d4f..babd3a8864a1 100644
> --- a/sound/pci/hda/hda_jack.c
> +++ b/sound/pci/hda/hda_jack.c
> @@ -403,10 +403,12 @@ int snd_hda_jack_add_kctl(struct hda_codec *codec, hda_nid_t nid,
>  
>         jack->phantom_jack = !!phantom_jack;
>         jack->type = type;
> -       jack->jack->private_data = jack;
> -       jack->jack->private_free = hda_free_jack_priv;
> -       state = snd_hda_jack_detect(codec, nid);
> -       snd_jack_report(jack->jack, state ? jack->type : 0);
> +       if (jack->jack) {
> +               jack->jack->private_data = jack;
> +               jack->jack->private_free = hda_free_jack_priv;
> +               state = snd_hda_jack_detect(codec, nid);
> +               snd_jack_report(jack->jack, state ? jack->type : 0);
> +       }
>  
>         return 0;
>  }
> diff --git a/sound/pci/hda/patch_hdmi.c b/sound/pci/hda/patch_hdmi.c
> index 8ee78dbd4c60..34a7b3aaba11 100644
> --- a/sound/pci/hda/patch_hdmi.c
> +++ b/sound/pci/hda/patch_hdmi.c
> @@ -2158,8 +2158,10 @@ static int add_acomp_jack_kctl(struct hda_codec *codec,
>         if (err < 0)
>                 return err;
>         per_pin->acomp_jack = jack;
> -       jack->private_data = per_pin;
> -       jack->private_free = free_acomp_jack_priv;
> +       if (jack) {
> +               jack->private_data = per_pin;
> +               jack->private_free = free_acomp_jack_priv;
> +       }
>         return 0;
>  }
> 
Looks good to me.

Acked-by: Arnd Bergmann <[e-mail]>

Arnd Bergmann Feb. 16, 2016, 10:08 p.m. UTC | #2

On Tuesday 16 February 2016 18:26:09 Arnd Bergmann wrote:
> >                 return err;
> >         per_pin->acomp_jack = jack;
> > -       jack->private_data = per_pin;
> > -       jack->private_free = free_acomp_jack_priv;
> > +       if (jack) {
> > +               jack->private_data = per_pin;
> > +               jack->private_free = free_acomp_jack_priv;
> > +       }
> >         return 0;
> >  }
> > 
> Looks good to me.
> 
> Acked-by: Arnd Bergmann <[e-mail]>
> 

Actually, on second thought, there is one notable downside
of this approach: 'jack' is guaranteed to be NULL here, so gcc
won't produce a warning if another driver makes the same mistake:

By default, we only get a warning for uses of uninitialized
variables, but not those that are known to be NULL. I think
some compiler warnings are able to warn about that too, but I
forgot the details.

	Arnd

Patch

diff --git a/include/sound/jack.h b/include/sound/jack.h
index 23bede121c78..a27c253a3207 100644
--- a/include/sound/jack.h
+++ b/include/sound/jack.h
@@ -99,6 +99,7 @@  void snd_jack_report(struct snd_jack *jack, int status);
 static inline int snd_jack_new(struct snd_card *card, const char *id, int type,
 			       struct snd_jack **jack, bool initial_kctl, bool phantom_jack)
 {
+	*jack = NULL;
 	return 0;
 }
 
diff --git a/sound/pci/hda/hda_jack.c b/sound/pci/hda/hda_jack.c
index a33234e04d4f..babd3a8864a1 100644
--- a/sound/pci/hda/hda_jack.c
+++ b/sound/pci/hda/hda_jack.c
@@ -403,10 +403,12 @@  int snd_hda_jack_add_kctl(struct hda_codec *codec, hda_nid_t nid,
 
 	jack->phantom_jack = !!phantom_jack;
 	jack->type = type;
-	jack->jack->private_data = jack;
-	jack->jack->private_free = hda_free_jack_priv;
-	state = snd_hda_jack_detect(codec, nid);
-	snd_jack_report(jack->jack, state ? jack->type : 0);
+	if (jack->jack) {
+		jack->jack->private_data = jack;
+		jack->jack->private_free = hda_free_jack_priv;
+		state = snd_hda_jack_detect(codec, nid);
+		snd_jack_report(jack->jack, state ? jack->type : 0);
+	}
 
 	return 0;
 }
diff --git a/sound/pci/hda/patch_hdmi.c b/sound/pci/hda/patch_hdmi.c
index 8ee78dbd4c60..34a7b3aaba11 100644
--- a/sound/pci/hda/patch_hdmi.c
+++ b/sound/pci/hda/patch_hdmi.c
@@ -2158,8 +2158,10 @@  static int add_acomp_jack_kctl(struct hda_codec *codec,
 	if (err < 0)
 		return err;
 	per_pin->acomp_jack = jack;
-	jack->private_data = per_pin;
-	jack->private_free = free_acomp_jack_priv;
+	if (jack) {
+		jack->private_data = per_pin;
+		jack->private_free = free_acomp_jack_priv;
+	}
 	return 0;
 }