Skip to content

btrfs: scrub: fix several races which can lead to UAF

Qu Wenruo linux-btrfs stable 2 versions – In review

Get series with b4
b4 am [email protected]

Versions newest first · linked by author and title

  1. v2 Latest2 patches 1 replyIn review
  2. v12 patches 4 repliesSuperseded

Changes in v2 from the cover letter

- Fix the race charts where the sequence is incorrect All functions are shown when there is enough space.

Patches in v2 No applies recorded yet

#Subject and filesSizeRepliesApplied
1/2btrfs: scrub: fix use-after-free when waking up the repair waiterfs/btrfs/scrub.c+8 -110 replies
2/2btrfs: scrub: fix use-after-free when waking up the stripe I/O waiterfs/btrfs/scrub.c+69 -280 replies

Reviews trailers per patch

None of the 2 patches has a Reviewed-by or Acked-by of their own; the series has one for all of them.

PatchFilipe Manana
SeriesR-b
1/2
2/2

Trailers given in replies are filled; those carried in the patches from earlier versions are outlined with dashes.

Read the v2 thread

Cover letter [PATCH v2 0/2] btrfs: scrub: fix several races which can lead to UAF

[CHANGELOG]
v2:
- Fix the race charts where the sequence is incorrect All functions are shown when there is enough space.

Also re-order the functions to properly represent their correct time sequence.

- Do not use non-upstream commit number in the commit message

This is inspired by the fix in raid56:
https://lore.kernel.org/linux-btrfs/[email protected]/

And there are two similar cases inside scrub.c:
- scrub_stripe::repair_wait
- scrub_stripe::io_wait

The only difference is, the window is much smaller compared to RAID56, as most scrub_stripe structure have a very long lifespan, as long as the scrub/replace ioctl. Only the RAID56 data stripes has a shorter lifespan.

But still it's possible in theory, so here comes the same fix for scrub.

For repair_wait it's pretty straight foward.

For io_wait we have several locations that we submit a batch of bios in one go, and pending_io may drop to zero during IOs.
Follow the existing behavior by grabbing one extra ref to ensure the pending_io won't reach zero during IOs.

Qu Wenruo (2):
btrfs: scrub: fix use-after-free when waking up the repair waiter
btrfs: scrub: fix use-after-free when waking up the stripe I/O waiter

 fs/btrfs/scrub.c | 116 +++++++++++++++++++++++++++++++----------------
 1 file changed, 77 insertions(+), 39 deletions(-)
Signature · 1 line
2.56.0