Message-ID: <CAKUZ0zKow-wstck4nq7JFbXfxYXR94phRG_bEuWBY2H7p3EWyg@mail.gmail.com>
Date: Fri, 11 Apr 2025 01:05:00 -0400
From: Gabriel Shahrouzi <gshahrouzi@...il.com>
To: Laurent Pinchart <laurent.pinchart@...asonboard.com>
Cc: Ricardo Ribalda <ribalda@...omium.org>, hdegoede@...hat.com, mchehab@...nel.org, 
	linux-media@...r.kernel.org, linux-kernel@...r.kernel.org, 
	syzbot+701fc9cc0cb44e2b0fe9@...kaller.appspotmail.com, 
	skhan@...uxfoundation.org, kernelmentees@...ts.linuxfoundation.org
Subject: Re: [PATCH] media: Fix invalid link creation when source entity has 0 pads

> > > Can you share the USB descriptors.
> >
> > The USB descriptors via the Syzkaller reproducer:
> >          "\x12\x01\x00\x00\xfb\x5d\x7d\x08\x6d\x04\xc3\x08\x16\x6b\x01\x02\x03"
> >          "\x01\x09\x02\x50\x00\x01\x00\x00\x00\x00\x09\x04\x1f\x00\x00\xff\x01"
> >          "\x00\x00\x0a\x24\x02\x00\x00\x05\x02\x01\x02\x07\x24\x07\x05\x00\x00"
> >          "\x18\xc2\x24\x08\x05\x04\x00\x04\x96\x0d\x24\x06\x01\x01\x03\x02\x00"
> >          "\x01\x00\x06\x00\x06\x09\x24\x03\x05\x05\x03\x06\x05\x81\x09\x24\x03"
> >          "\x06\x01\x01\x04\x05\x05\x07\x24\x04\x05\x01\x00\x9c\xbd\x89"
>
> If I haven't made any mistake in the manual decode process (is there any
> Linux tool that can decode a binary descriptors dump the same way lsusb
> decodes descriptors from a device ?), the relevant UVC descriptors there
> are
I tried looking around and couldn’t find anything. The closest I could
find was defining the structure and then using a parser or reverse
engineering it using wireshark. Do you think there is a need for such
a tool? I would be interested in making one.

>
> 0x0a            bLength
> 0x24            bDescriptorType USB_DT_CS_INTERFACE
> 0x02            bDescriptorSubtype VC_INPUT_TERMINAL
> 0x00            bTerminalID 0 (invalid)
> 0x00, 0x05      bTerminalType 0x0500 (invalid)
> 0x02            bAssocTerminal 2 (invalid)
> 0x01            iTerminal 1
> 0x02, 0x07
>
> 0x09            bLength
> 0x24            bDescriptorType USB_DT_CS_INTERFACE
> 0x03            bDescriptorSubtype VC_OUTPUT_TERMINAL
> 0x06            bTerminalID 6
> 0x01, 0x01      bTerminalType TT_STREAMING
> 0x04            bAssocTerminal 4 (invalid)
> 0x05            bSourceID 5
> 0x05            iTerminal 5
>
> 0x07            bLength
> 0x24            bDescriptorType USB_DT_CS_INTERFACE
> 0x04            bDescriptorSubtype VC_SELECTOR_UNIT
> 0x05            bUnitID 5
> 0x01            bNrInPins 1
> 0x00            baSourceID(1) 0
> 0x9c            iSelector 156
This looks correct. I compared it with the Universal Serial Bus Device
Class Definition for Video Devices: Video Device Examples (revision
1.5 2012). I only got slightly different results (only formatting
related - w prefix instead of b for 2 bytes).

>
> Ignoring a few invalid values (bTerminalID shouldb't be 0, bTerminalType
> 0x0500 is defined by the specification, and the two bAssocTerminal ids
> are also invalid), this creates the following chain:
>
> VC_INPUT_TERMINAL (0) -> VC_SELECTOR_UNIT (5) -> VC_OUTPUT_TERMINAL (6)
>
> Looking at uvc_mc_init_entity() where the media_entity->num_pads field
> gets assigned by calling media_entity_pads_init(), a media entity is
> only initialized when the entity type is not TT_STREAMING (so it's a
> subdev), or when the entity has an associated video device. I think
> that what is happening here is that the second entity in the above list
> (VC_OUTPUT_TERMINAL, id 6) fails to initialize properly in
> uvc_register_terms() is there is no corresponding streaming interface in
> the device. This is confirmed by the
>
> usb 1-1: No streaming interface found for terminal 6.
>
> message in the syzbot kernel log. No video device is created for the
> terminal, and no media_entity is initialized. Trying to later link the
> entity in uvc_mc_create_links() then fails.
>
> I don't want to address this in uvc_mc_create_links() as the invalid
> terminal in the chain means we could have other issues elsewhere. One
> option is to fail turn the missing streaming interface check in a hard
> failure, at least for the chain being registered. The driver could still
> proceed to registering other chains.
>
> There's a small risk of regression for buggy devices. If that's a
> problem, we could instead remove invalid terminals from the device
> entities list before we proceed to scanning chains.
Which solution do you recommend? I can try implementing the one for
the hard failure while letting it proceed to register other chains. It
seems like testing for a regression would require feedback from the
community which might be hard since it would most likely only affect a
few devices.





>
> > > > I confirmed this by changing the terminal ID in the format descriptor
> > > > from 6 to 5, which eliminates both warnings. This shows the warning is
> > > > correctly identifying an invalid descriptor configuration, not a
> > > > kernel bug.
> > >
> > > There's still something not quite right. uvc_entity->num_pads should
> > > always be equal to the corresponding media_entity->num_pads. That's not
> > > the case here, and I think it indicates a bug.
> >
> > Ah ok - the mismatch itself shouldn't happen regardless of the descriptor
> >
> > > > Since the USB descriptor is invalid, I believe the warning is
> > > > necessary and should remain. The code should stay as is.
> > >
> > > There should be a warning, but I think it needs to be caught in a
> > > different place, earlier.
> >
> > Got it.
> >
> > > > On Tue, Mar 25, 2025 at 8:13 PM Laurent Pinchart wrote:
> > > > > On Tue, Mar 25, 2025 at 06:05:00PM -0400, Gabriel wrote:
> > > > > > Hi Ricardo,
> > > > > >
> > > > > > > I cannot reach that URL
> > > > > > I was unable to access the URL from my email client when I initially
> > > > > > sent the email, but a couple of hours later, I was able to. Initially,
> > > > > > copying and pasting the URL into the browser provided a workaround.
> > > > > >
> > > > > > > Shouldn't it be?:
> > > > > > > Fixes: 4ffc2d89f38a ("[media] uvcvideo: Register subdevices for each entity")
> > > > > > You're right, I incorrectly referenced the wrong commit. However, I’m
> > > > > > not certain if it should reference a96aa5342d57 (Fixes: a96aa5342d57 -
> > > > > > '[media] uvcvideo: Ignore entities for terminals with no supported
> > > > > > format') as it's the latest commit affecting the line I'm changing or
> > > > > > the one you mentioned.
> > > > > >
> > > > > > > Shouldn't source->num_pads be the same as remote->num_pads?
> > > > > > The fuzzer (Syzkaller) that triggered the warning appears to have
> > > > > > encountered a case where source->num_pads and remote->num_pads were
> > > > > > different. When analyzing the case in GDB, remote->num_pads was 1,
> > > > > > while source->num_pads was 0.
> > > > >
> > > > > This seems like the real bug that should be fixed.
> > > > >
> > > > > > > Are you sure that your kernel does not contain?
> > > > > > > https://web.git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/drivers/media/usb/uvc/uvc_entity.c?id=41ddb251c68ac75c101d3a50a68c4629c9055e4c
> > > > > > Yes, it should be included since I am running the upstream kernel.
>
> --
> Regards,
>
> Laurent Pinchart

Powered by blists - more mailing lists