Skip to contentSkip to conversation

media: Fix invalid link creation when source entity has 0 pads

PATCH In review [email protected] 3 participants 9 messages Mar 25, 2025 – Apr 11, 2025

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 media: Fix invalid link creation when source entity has 0 pads

Quoted 1 line

From 307209d175be0145e36b9cf95944e2e62afeab11 Mon Sep 17 00:00:00 2001

From: Gabriel Shahrouzi <[email protected]>
Date: Mon, 24 Mar 2025 19:45:55 -0400
Subject: [PATCH] media: Fix invalid link creation when source entity has 0
pads

This patch addresses the warning triggered in the media_create_pad_link() function, specifically related to the check WARN_ON(source_pad >= source->num_pads). The fix proposed adds an additional check to ensure that source->num_pads is non-zero before proceeding with the media_create_pad_link() function.

Reported-by [email protected]
Fixes a3fbc2e6bb05 ("media: mc-entity.c: use WARN_ON, validate link pads")
---
 drivers/media/usb/uvc/uvc_entity.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/media/usb/uvc/uvc_entity.c +1 -1
index cc68dd24eb42..5397ce76c218 100644
@@ -43,7 +43,7 @@ static int uvc_mc_create_links(struct uvc_video_chain *chain,
 		source = (UVC_ENTITY_TYPE(remote) == UVC_TT_STREAMING)
 		       ? (remote->vdev ? &remote->vdev->entity : NULL)
 		       : &remote->subdev.entity;
-		if (source == NULL)
+		if (source == NULL || source->num_pads == 0)
 			continue;
 
 		remote_pad = remote->num_pads - 1;
Signature · 1 line
2.43.0
#1 lore raw

Hi Gabriel

Quoted 7 lines · On Tue, 25 Mar 2025 at 08:55, <[email protected]> wrote:
Quoted 2 lines

From 307209d175be0145e36b9cf95944e2e62afeab11 Mon Sep 17 00:00:00 2001

From: Gabriel Shahrouzi <[email protected]>
Date: Mon, 24 Mar 2025 19:45:55 -0400
Subject: [PATCH] media: Fix invalid link creation when source entity has 0
pads

Quoted 8 lines

This patch addresses the warning triggered in the media_create_pad_link() function, specifically related to the check WARN_ON(source_pad >= source->num_pads). The fix proposed adds an additional check to ensure that source->num_pads is non-zero before proceeding with the media_create_pad_link() function.

I cannot reach that URL

Quoted 2 lines
Fixes a3fbc2e6bb05 ("media: mc-entity.c: use WARN_ON, validate link pads")

Shouldn't it be? :

Fixes 4ffc2d89f38a ("[media] uvcvideo: Register subdevices for each entity")
Quoted 10 lines
---
 drivers/media/usb/uvc/uvc_entity.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/media/usb/uvc/uvc_entity.c +0 -0
index cc68dd24eb42..5397ce76c218 100644

drivers/media/usb/uvc/uvc_entity.c static int uvc_mc_create_links(struct uvc_video_chain *chain, · line 46

Quoted 5 lines
drivers/media/usb/uvc/uvc_entity.c +1 -1
@@ -43,7 +43,7 @@ static int uvc_mc_create_links(struct uvc_video_chain *chain,
                source = (UVC_ENTITY_TYPE(remote) == UVC_TT_STREAMING)
                       ? (remote->vdev ? &remote->vdev->entity : NULL)
                       : &remote->subdev.entity;
-               if (source == NULL)
+               if (source == NULL || source->num_pads == 0)

Shouldn't source->num_pads be the same as remote->num_pads?

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

Regards!

Quoted 6 lines
drivers/media/usb/uvc/uvc_entity.c +0 -1
                        continue;
 
                remote_pad = remote->num_pads - 1;
--

2.43.0

Signature · 1 line
Ricardo Ribalda
author
#2 lore raw

Hi Ricardo,

Quoted 1 line

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.

Quoted 2 lines

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.

Quoted 1 line

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.

Quoted 2 lines

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,
Gabriel

maintainer
#3 lore raw
Quoted 15 lines · Gabriel

Hi Ricardo,

Quoted 1 line

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.

Quoted 2 lines

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.

Quoted 5 lines
Quoted 1 line

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.

Quoted 3 lines
Quoted 2 lines

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.

Signature · 3 lines
Regards,

Laurent Pinchart
author
#4 lore raw

Hi Laurent,

I’ve analyzed the bug report, and the root cause of the "WARNING-media_create_pad_link" issue is a mismatch in terminal references in the USB descriptor.

The format type descriptor references terminal ID 6, while the audio streaming interface descriptor points to terminal ID 5. This discrepancy triggers the warning: "No streaming interface found for terminal 6", followed by the media pad link warning.

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.

Since the USB descriptor is invalid, I believe the warning is necessary and should remain. The code should stay as is.

Regards,
Gabriel

Quoted 33 lines · Laurent Pinchart
Quoted 15 lines · Gabriel

Hi Ricardo,

Quoted 1 line

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.

Quoted 2 lines

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.

Quoted 5 lines
Quoted 1 line

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.

Quoted 3 lines
Quoted 2 lines

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

maintainer
#5 lore raw

Hi Gabriel,

Quoted 10 lines · Gabriel

Hi Laurent,

I’ve analyzed the bug report, and the root cause of the "WARNING-media_create_pad_link" issue is a mismatch in terminal references in the USB descriptor.

The format type descriptor references terminal ID 6, while the audio streaming interface descriptor points to terminal ID 5. This discrepancy triggers the warning: "No streaming interface found for terminal 6", followed by the media pad link warning.

Can you share the USB descriptors.

Quoted 4 lines

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.

Quoted 2 lines

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.

Quoted 29 lines · Laurent Pinchart
Quoted 28 lines · Gabriel
Quoted 15 lines

Hi Ricardo,

Quoted 1 line

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.

Quoted 2 lines

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.

Quoted 5 lines
Quoted 1 line

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.

Quoted 3 lines
Quoted 2 lines

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.

Signature · 3 lines
Regards,

Laurent Pinchart
#6 lore raw
Quoted 10 lines · Laurent Pinchart

Hi Gabriel,

Quoted 6 lines · Gabriel

Hi Laurent,

I’ve analyzed the bug report, and the root cause of the "WARNING-media_create_pad_link" issue is a mismatch in terminal references in the USB descriptor.

Quoted 6 lines
Quoted 4 lines

The format type descriptor references terminal ID 6, while the audio streaming interface descriptor points to terminal ID 5. This discrepancy triggers the warning: "No streaming interface found for terminal 6", followed by the media pad link warning.

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"
Quoted 9 lines
Quoted 5 lines

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

Quoted 6 lines
Quoted 3 lines

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.

Quoted 35 lines
Quoted 25 lines · Laurent Pinchart
Quoted 23 lines · Gabriel
Quoted 20 lines

Hi Ricardo,

Quoted 1 line

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.

Quoted 2 lines

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.

Quoted 1 line

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.

Quoted 5 lines
Quoted 5 lines

This seems like the real bug that should be fixed.

Quoted 3 lines
Quoted 2 lines

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

maintainer
#7 lore raw

Hi Gabriel,

Quoted 15 lines · Gabriel Shahrouzi
Quoted 7 lines · Laurent Pinchart
Quoted 6 lines · Gabriel

Hi Laurent,

I’ve analyzed the bug report, and the root cause of the "WARNING-media_create_pad_link" issue is a mismatch in terminal references in the USB descriptor.

Quoted 6 lines
Quoted 4 lines

The format type descriptor references terminal ID 6, while the audio streaming interface descriptor points to terminal ID 5. This discrepancy triggers the warning: "No streaming interface found for terminal 6", followed by the media pad link warning.

Can you share the USB descriptors.

Quoted 7 lines

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

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

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.

Quoted 47 lines
Quoted 8 lines
Quoted 4 lines

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

Quoted 5 lines
Quoted 2 lines

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.

Quoted 28 lines
Quoted 28 lines · Laurent Pinchart
Quoted 27 lines · Gabriel
Quoted 15 lines

Hi Ricardo,

Quoted 1 line

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.

Quoted 2 lines

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.

Quoted 5 lines
Quoted 1 line

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.

Quoted 3 lines
Quoted 2 lines

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.

Signature · 3 lines
Regards,

Laurent Pinchart
#8 lore raw
Quoted 10 lines
Quoted 9 lines
Quoted 1 line

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"
Quoted 4 lines

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.

Quoted 19 lines
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
Quoted 7 lines
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).

Quoted 28 lines

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.

Quoted 3 lines

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.

Quoted 53 lines
Quoted 12 lines
Quoted 9 lines
Quoted 5 lines

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

Quoted 36 lines
Quoted 5 lines
Quoted 2 lines

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.

Quoted 28 lines
Quoted 28 lines · Laurent Pinchart
Quoted 27 lines · Gabriel
Quoted 15 lines

Hi Ricardo,

Quoted 1 line

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.

Quoted 2 lines

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.

Quoted 5 lines
Quoted 1 line

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.

Quoted 3 lines
Quoted 2 lines

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