Skip to content

btrfs: Enhanced runtime defence against fuzzed images

Qu Wenruo linux-btrfs 4 versions – In review

Get series with b4
b4 am [email protected]

Versions newest first · linked by author and title

  1. v5 Latest4 patches 8 repliesIn review
  2. v4 posting 2/24 patches 11 repliesSuperseded
  3. v4 posting 1/21 of 4 patches 0 repliesSuperseded
Earlier versions · 1
VersionPostedPatchesRepliesStatus
v35 patches17 repliesSuperseded

Changes in v5 from the cover letter

- Properly inline the check while make the report code into another function for the 1st patch

- Keep btrfs_abort_transaction() call where it is for the 2nd patch To make the line number correct and abort transaction asap.

- Function naming update for the 4th patch

Patches in v5 No applies recorded yet

#Subject and filesSizeRepliesApplied
1/4btrfs: extent_io: do extra check for extent buffer read write functionsfs/btrfs/extent_io.c+45 -377 replies
2/4btrfs: extent-tree: kill BUG_ON() in __btrfs_free_extent() and do better commentfs/btrfs/extent-tree.c+137 -130 replies
3/4btrfs: extent-tree: kill the BUG_ON() in insert_inline_extent_backref()fs/btrfs/extent-tree.c+19 -10 replies
4/4btrfs: ctree: checking key orders before merged tree blocksfs/btrfs/ctree.c+71 -00 replies

Reviews trailers per patch

All 4 patches have a Reviewed-by or Acked-by.

PatchJosef BacikNikolay Borisov
1/4R-b
2/4R-b R-b
3/4R-b R-b
4/4R-b R-b

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

Read the v5 thread

Cover letter [PATCH v5 0/4] btrfs: Enhanced runtime defence against fuzzed images

This patch is revived after one year, as one internal report has hit one BUG_ON() with real world fs, so I believe this patchset still makes sense.

- Enhanced eb accessors
Not really needed for the fuzzed images, as 448de471cd4c ("btrfs: Check the first key and level for cached extent buffer") already fixed half of the reported images.
Just add a final layer of safe net.

Just to complain here, two experienced btrfs developer have got confused by @start, @len in functions like read_extent_buffer() with logical address.
The best example to solve the confusion is to check the read_extent_buffer() call in btree_read_extent_buffer_pages().

I'm not sure why this confusion happens or even get spread. My guess is the extent_buffer::start naming causing the problem.

If so, I would definitely rename extent_buffer::start to extent_buffer::bytenr at any cost.
Hopes the new commend will address the problem for now.

- BUG_ON() hunt in __btrfs_free_extent()
Kill BUG_ON()s in __btrfs_free_extent(), replace with error reporting and why it shouldn't happen.

Also add comment on what __btrfs_free_extent() is designed to do, with two dump-tree examples for newcomers.

- BUG_ON() hunt in __btrfs_inc_extent_ref()
Just like __btrfs_free_extent(), but less comment as comment for __btrfs_free_extent() should also work for __btrfs_inc_extent_ref(), and __btrfs_inc_extent_ref() has a better structure than __btrfs_free_extent().

- Defence against unbalanced empty leaf

- Defence against bad key order across two tree blocks

The last two cases can't be rejected by tree-checker and they are all cross-eb cases.
Thankfully we can reuse existing first_key check against unbalanced empty leaf, but needs extra check deep into ctree.c for tree block merging time check.

[ Not to mail bombarding the report, thus only RB tag in cover letter ]

Changelog:
v2:
- Remove duplicated error message in WARN() call. Changed to WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG)) Also move WARN() after btrfs error message.

- Fix a comment error in __btrfs_free_extent()
It's not adding refs to a tree block, but adding the same refs to an existing tree block ref.
It's impossible a btrfs tree owning the same tree block directly twice.

- Add comment for eb accessors about @start and @len
If anyone could tell me why such confusion between @start @len and logical address is here, I will definitely solve the root cause no matter how many codes need to be modified.

- Use bool to replace int where only two values are returned Also rename to follow the bool type.

- Remove one unrelated change for the error handler in btrfs_inc_extent_ref()

- Add Reviewed-by tag

v3:
- Rebased to latest misc-next branch
All conflicts can be auto-merged.

v4:
- Remove one patch which is already merged
A little surprised by the fact that git can't detecth such case.

- Add new reviewed-by tags from Josef

v5:
- Properly inline the check while make the report code into another function for the 1st patch

- Keep btrfs_abort_transaction() call where it is for the 2nd patch To make the line number correct and abort transaction asap.

- Function naming update for the 4th patch

Qu Wenruo (4):
btrfs: extent_io: do extra check for extent buffer read write
functions
btrfs: extent-tree: kill BUG_ON() in __btrfs_free_extent() and do
better comment
btrfs: extent-tree: kill the BUG_ON() in
insert_inline_extent_backref()
btrfs: ctree: checking key orders before merged tree blocks

 fs/btrfs/ctree.c       |  71 +++++++++++++++++
 fs/btrfs/extent-tree.c | 170 +++++++++++++++++++++++++++++++++++++----
 fs/btrfs/extent_io.c   |  82 +++++++++++---------
 3 files changed, 272 insertions(+), 51 deletions(-)
Signature · 1 line
2.28.0