[v7,0/3] uvcvideo: Attempt N to land UVC race conditions fixes

Message

Ricardo Ribalda Sept. 26, 2024, 5:49 a.m. UTC

Back in 2020 Guenter published a set of patches to fix some race
conditions in UVC:
https://lore.kernel.org/all/[e-mail]/

That kind of race conditions are not only seen in UVC, but are a common
seen in almost all the kernel, so this is what it was decided back then
that we should try to fix them at higher levels.

After that. A lot of video_is_registered() were added to the core:

```
ribalda@alco:~/work/linux$ git grep is_registered drivers/media/v4l2-core/
drivers/media/v4l2-core/v4l2-compat-ioctl32.c:  if (!video_is_registered(vdev))
drivers/media/v4l2-core/v4l2-dev.c:     if (video_is_registered(vdev))
drivers/media/v4l2-core/v4l2-dev.c:     if (video_is_registered(vdev))
drivers/media/v4l2-core/v4l2-dev.c:     if (video_is_registered(vdev)) {
drivers/media/v4l2-core/v4l2-dev.c:             if (video_is_registered(vdev))
drivers/media/v4l2-core/v4l2-dev.c:     if (!video_is_registered(vdev))
drivers/media/v4l2-core/v4l2-dev.c:     if (video_is_registered(vdev))
drivers/media/v4l2-core/v4l2-dev.c:     if (vdev == NULL || !video_is_registered(vdev)) {
drivers/media/v4l2-core/v4l2-dev.c:             if (video_is_registered(vdev))
drivers/media/v4l2-core/v4l2-dev.c:     if (!vdev || !video_is_registered(vdev))
drivers/media/v4l2-core/v4l2-ioctl.c:   if (!video_is_registered(vfd)) {
drivers/media/v4l2-core/v4l2-subdev.c:  if (video_is_registered(vdev)) {
```

And recently Sakari is trying to land:
https://lore.kernel.org/linux-media/[e-mail]/

Which will make obsolete a lot off (all?) of the video_is_registered() checks in
Guenter's patches.

Besides those checks, there were some other valid races fixed in his
patches.

This patchset tries to fix the races still present in our code.

Thanks!

Signed-off-by: Ricardo Ribalda <[e-mail]>
---
Changes in v7: Thanks Laurent
- Split "media: uvcvideo: Stop stream during unregister" into a separate
  patchset.
- Use guards.
- Improve naming and comments.
- Link to v6: https://lore.kernel.org/r/[e-mail]

Changes in v6: Thanks Hans
- s/uvc_queue_streamoff/uvc_queue_release/
- Link to v5: https://lore.kernel.org/r/[e-mail]

Changes in v5: Thanks Hans!
- Refactor unregister as vb2_video_unregister_device
- I have tested the first patch independently from the others, so it
  could be merged in two steps if needed.
- Link to v4: https://lore.kernel.org/r/[e-mail]

Changes in v4: Thanks Sergey and Guenter
- Fix typos
- Move location of mutex_init
- Split patch to make the suspend change explicit
- Link to v3: https://lore.kernel.org/r/[e-mail]

Changes in v3: Thanks Hans!
- Stop streaming during uvc_unregister()
- Refactor the uvc_status code
- Link to v2: https://lore.kernel.org/r/[e-mail]

Changes in v2:
- Actually send the series to the ML an not only to individuals.
- Link to v1: https://lore.kernel.org/r/[e-mail]

---
Ricardo Ribalda (3):
      media: uvcvideo: Refactor the status irq API
      media: uvcvideo: Avoid race condition during unregister
      media: uvcvideo: Exit early if there is not int_urb

 drivers/media/usb/uvc/uvc_driver.c | 13 ++-------
 drivers/media/usb/uvc/uvc_status.c | 60 +++++++++++++++++++++++++++++++++++---
 drivers/media/usb/uvc/uvc_v4l2.c   | 22 ++++----------
 drivers/media/usb/uvc/uvcvideo.h   | 10 ++++---
 4 files changed, 70 insertions(+), 35 deletions(-)
---
base-commit: 81ee62e8d09ee3c7107d11c8bbfd64073ab601ad
change-id: 20230309-guenter-mini-89861b084ef1

Best regards,