Re: [PATCH] ALSA: usb-audio: Prevent urb from writing out of bounds

[Takashi Iwai <tiwai@suse.de>]

On Fri, 07 Nov 2025 02:13:27 +0100,
Lizhi Xu wrote:
> 
> On Fri, 7 Nov 2025 08:54:20 +0800, Lizhi Xu wrote:
> > > > > > > > The calculation rule for the actual data length written to the URB's
> > > > > > > > transfer buffer differs from that used to allocate the URB's transfer
> > > > > > > > buffer, and in this problem, the value used during allocation is smaller.
> > > > > > > >
> > > > > > > > This ultimately leads to write out-of-bounds errors when writing data to
> > > > > > > > the transfer buffer.
> > > > > > > >
> > > > > > > > To prevent out-of-bounds writes to the transfer buffer, a check between
> > > > > > > > the size of the bytes to be written and the size of the allocated bytes
> > > > > > > > should be added before performing the write operation.
> > > > > > > >
> > > > > > > > When the written bytes are too large, -EPIPE is returned instead of
> > > > > > > > -EAGAIN, because returning -EAGAIN might result in push back to ready
> > > > > > > > list again.
> > > > > > > >
> > > > > > > > Based on the context of calculating the bytes to be written here, both
> > > > > > > > copy_to_urb() and copy_to_urb_quirk() require a check for the size of
> > > > > > > > the bytes to be written before execution.
> > > > > > > >
> > > > > > > > syzbot reported:
> > > > > > > > BUG: KASAN: slab-out-of-bounds in copy_to_urb+0x261/0x460 sound/usb/pcm.c:1487
> > > > > > > > Write of size 264 at addr ffff88801107b400 by task syz.0.17/5461
> > > > > > > >
> > > > > > > > Call Trace:
> > > > > > > >  copy_to_urb+0x261/0x460 sound/usb/pcm.c:1487
> > > > > > > >  prepare_playback_urb+0x953/0x13d0 sound/usb/pcm.c:1611
> > > > > > > >
> > > > > > > > Reported-by: syzbot+bfd77469c8966de076f7@syzkaller.appspotmail.com
> > > > > > > > Closes: https://syzkaller.appspot.com/bug?extid=bfd77469c8966de076f7
> > > > > > > > Tested-by: syzbot+bfd77469c8966de076f7@syzkaller.appspotmail.com
> > > > > > > > Signed-off-by: Lizhi Xu <lizhi.xu@windriver.com>
> > > > > > >
> > > > > > > I'm afraid that this doesn't address the root cause at all.
> > > > > > > The description above sounds plausible, but not pointing to "why".
> > > > > > >
> > > > > > > The bytes is frames * stride, so the question is why a too large
> > > > > > > frames is calculated.  I couldn't have time to check the details, but
> > > > > > > there should be rather some weird condition / parameters to trigger
> > > > > > > this, and we should check that at first.
> > > > > > During debugging, I discovered that the value of ep->packsize[0] is 22,
> > > > > > which causes the counts calculated by
> > > > > > counts = snd_usb_endpoint_next_packet_size(ep, ctx, i, avail);
> > > > > > to be 22, resulting in a frames value of 22 * 6 = 132;
> > > > > > Meanwhile, the stride value is 2, which ultimately results in
> > > > > > bytes = frames * stride = 132 * 2 = 264;
> > > > > > @@ -1241,6 +1252,10 @@ static int data_ep_set_params(struct snd_usb_endpoint *ep)
> > > > > > 	u->buffer_size = maxsize * u->packets;
> > > > > > 	...
> > > > > > 	u->urb->transfer_buffer =
> > > > > >                 usb_alloc_coherent(chip->dev, u->buffer_size,
> > > > > >                                    GFP_KERNEL, &u->urb->transfer_dma);
> > > > > >
> > > > > > Here, when calculating u->buffer_size = maxsize * u->packets;
> > > > > > maxsize = 9, packets = 6, which results in only 54 bytes allocated to
> > > > > > transfer_buffer;
> > > > >
> > > > > Hm, so the problem is rather the calculation of the buffer size.
> > > > > The size sounds extremely small.  Which parameters (rates, formats,
> > > > > etc) are used for achieving this?
> > > > rates: 22050
> > > > format: 2
> > > > channels: 1
> > > > /////////////////////////////
> > > > stride: 2
> > > > packets: 6
> > > > data interval: 0
> > > > frame_bits: 16
> > > > >
> > > > > The calculation of u->buffer_size is a bit complex, as maxsize is
> > > > > adjusted in many different ways.  Is it limited due to wMaxPacketSize
> > > > > setup?
> > > > Yes, it's because the value of ep->maxpacksize is 9 that the maxsize
> > > > value is 9.
> > > 
> > > OK, then a fix like below would work?
> > > 
> > > 
> > > thanks,
> > > 
> > > Takashi
> > > 
> > > --- a/sound/usb/endpoint.c
> > > +++ b/sound/usb/endpoint.c
> > > @@ -1362,6 +1362,11 @@ int snd_usb_endpoint_set_params(struct snd_usb_audio *chip,
> > >  	ep->sample_rem = ep->cur_rate % ep->pps;
> > >  	ep->packsize[0] = ep->cur_rate / ep->pps;
> > >  	ep->packsize[1] = (ep->cur_rate + (ep->pps - 1)) / ep->pps;
> > > +	if (ep->packsize[1] > ep->maxpacksize) {
> > > +		usb_audio_dbg(chip, "Too small maxpacksize %u for rate %u / pps %u\n",
> > > +			      ep->maxpacksize, ep->cur_rate, ep->pps);
> > > +		return -EINVAL;
> > > +	}
> > > 
> > >  	/* calculate the frequency in 16.16 format */
> > >  	ep->freqm = ep->freqn;
> > Of course, this fix was added after packsize[0] was assigned a value,
> > and Hillf Danton has already tested it.
> > 
> > However, to be more precise, although both packsize[1] and packsize[0]
> > exceed maxpacksize, this example is about packsize[0], so judging packsize[0]
> > is more rigorous.
> Of course, since packsize[1] is always greater than packsize[0] when pps
> is greater than 1, judging only packsize[1] is sufficient to avoid judging
> packsize[0] and packsize[1].

Right, that's the reason it checks only packsize[1].

I think it's fine to add another check like your patch in addition for
more safety, too.  But then it should better in copy_to_urb() for
both size and offset.  And, it should be with WARN_ON() or such, as
the OOB must not happen after packsize[] sanity check.


thanks,

Takashi