Hi Takashi,
Thank you for the review. Except for one question below, this patch
looks good to me. Would you like me to send your patch as a revised
patch to this mailing list?
On Thu, 21 Nov 2024 at 15:53, Takashi Iwai <tiwai@xxxxxxx> wrote:
>
> On Thu, 21 Nov 2024 15:06:13 +0100,
> Benoît Sevens wrote:
> >
> > A bogus device can provide a clock selector descriptor that contains a
> > bNrInPins that is larger than the actual size of baCSourceID. This can
> > lead to out-of-bound reads in __uac_clock_find_source. These out-of-bound
> > values can be leaked back to the device via the uac_clock_selector_get_val
> > calls.
> >
> > Fixes: 79f920fbff56 ("ALSA: usb-audio: parse clock topology of UAC2 devices")
> > CC: stable@xxxxxxxxxx
> > Signed-off-by: Benoît Sevens <bsevens@xxxxxxxxxx>
> > ---
> > sound/usb/clock.c | 10 +++++++++-
> > 1 file changed, 9 insertions(+), 1 deletion(-)
> >
> > diff --git a/sound/usb/clock.c b/sound/usb/clock.c
> > index 8f85200292f3..94fb628f116e 100644
> > --- a/sound/usb/clock.c
> > +++ b/sound/usb/clock.c
> > @@ -270,7 +270,7 @@ static int __uac_clock_find_source(struct snd_usb_audio *chip,
> > union uac23_clock_source_desc *source;
> > union uac23_clock_selector_desc *selector;
> > union uac23_clock_multiplier_desc *multiplier;
> > - int ret, i, cur, err, pins, clock_id;
> > + int ret, i, cur, err, length, pins, clock_id;
> > const u8 *sources;
> > int proto = fmt->protocol;
> > bool readable, writeable;
> > @@ -301,11 +301,19 @@ static int __uac_clock_find_source(struct snd_usb_audio *chip,
> >
> > selector = snd_usb_find_clock_selector(chip, entity_id, fmt);
> > if (selector) {
> > + length = GET_VAL(selector, proto, bLength);
> > pins = GET_VAL(selector, proto, bNrInPins);
> > clock_id = GET_VAL(selector, proto, bClockID);
> > sources = GET_VAL(selector, proto, baCSourceID);
> > cur = 0;
> >
> > + if (length < sizeof(selector) + pins) {
>
> This happens to be correct because the size of both
> uac_clock_selector_descriptor and uac3_clock_selector_descriptor are
> same. Otherwise we'd have to pass sizeof() of the corresponding
> struct depending on the proto value. So it's worth to comment here.
>
> BTW, now I noticed that we didn't check the size at traversing the
> descriptors. This means that the driver might read at a wrong
> position even before the point this patch addresses. For covering it,
> we need additional checks in the validator code. And this size check
> of clock selectors can be put there, too.
>
> That is, something like below (totally untested).
>
> It'll lead to a different behavior from your patch, though; currently,
> when no suitable clock selector is found, it check the multiplier and
> this might match. So I'm not entirely sure, but just wanted to tell
> you that it's another option.
>
>
> thanks,
>
> Takashi
>
> --- a/sound/usb/clock.c
> +++ b/sound/usb/clock.c
> @@ -36,6 +36,12 @@ union uac23_clock_multiplier_desc {
> struct uac_clock_multiplier_descriptor v3;
> };
>
> +/* check whether the descriptor bLength has the minimal length */
> +#define DESC_LENGTH_CHECK(p, proto) \
> + ((proto) == UAC_VERSION_3 ? \
> + ((p)->v3.bLength >= sizeof((p)->v3)) : \
> + ((p)->v2.bLength >= sizeof((p)->v2)))
> +
> #define GET_VAL(p, proto, field) \
> ((proto) == UAC_VERSION_3 ? (p)->v3.field : (p)->v2.field)
>
> @@ -58,6 +64,8 @@ static bool validate_clock_source(void *p, int id, int proto)
> {
> union uac23_clock_source_desc *cs = p;
>
> + if (!DESC_LENGTH_CHECK(cs, proto))
> + return false;
> return GET_VAL(cs, proto, bClockID) == id;
> }
>
> @@ -65,13 +73,23 @@ static bool validate_clock_selector(void *p, int id, int proto)
> {
> union uac23_clock_selector_desc *cs = p;
>
> - return GET_VAL(cs, proto, bClockID) == id;
> + if (!DESC_LENGTH_CHECK(cs, proto))
> + return false;
> + if (GET_VAL(cs, proto, bClockID) != id)
> + return false;
> + /* additional length check for baCSourceID, bmControls and iClockSelector */
> + if (proto == UAC_VERSION_3)
> + return cs->v3.bLength >= sizeof(cs->v3.bLength) + cs->v3.bNrInPins + 6;
Why can't we just do this here?
return cs->v3.bLength >= sizeof(cs->v3) + cs->v3.bNrInPins;
> + else
> + return cs->v2.bLength >= sizeof(cs->v2.bLength) + cs->v2.bNrInPins + 2;
And same question here, why not:
return cs->v2.bLength >= sizeof(cs->v2) + cs->v2.bNrInPins;
> }
>
> static bool validate_clock_multiplier(void *p, int id, int proto)
> {
> union uac23_clock_multiplier_desc *cs = p;
>
> + if (!DESC_LENGTH_CHECK(cs, proto))
> + return false;
> return GET_VAL(cs, proto, bClockID) == id;
> }
>
Thanks,
Benoit
[Index of Archives]
[Pulseaudio]
[Linux Audio Users]
[ALSA Devel]
[Fedora Desktop]
[Fedora SELinux]
[Big List of Linux Books]
[Yosemite News]
[KDE Users]