Re: [PATCH 15/27] ext4: check multi-block overlap on left traversal in add_system_zone()

[sashiko-bot@kernel.org]

> When walking system_blks->root to insert [start_blk, start_blk + count),
> add_system_zone() takes the left branch whenever start_blk <
> entry->start_blk instead of checking that the entire range ends at or
> before entry->start_blk (unlike ext4_sb_block_valid(), which checks
> start_blk + count - 1 < entry->start_blk).
> 
> As a result, a corrupt multi-block metadata range (such as a group's
> inode table or a reserved journal inode extent mapped in
> ext4_protect_reserved_inode() before sbi->s_system_blks is published)
> that starts strictly before an existing system zone and extends into or
> across it traverses rb_left instead of returning -EFSCORRUPTED. This
> inserts overlapping intervals into system_blks->root and breaks the
> disjoint-interval invariant relied upon by ext4_sb_block_valid() (for
> example, a lookup on blocks past entry->start_blk + entry->count
> branches right and misses the overlapping zone stored in rb_left). A
> corrupt 64-bit start_blk whose start_blk + count wraps around u64 can
> similarly corrupt the rbtree ordering.
> 
> Reject zero-length or u64-wrapping ranges up front and require
> start_blk + count <= entry->start_blk before taking rb_left in
> add_system_zone(). Add KUnit regression tests in
> fs/ext4/block_validity-test.c covering group metadata and journal
> extents that partially overlap from the left, completely enclose an
> existing system zone, or wrap 64-bit block numbers.
> 
> Fixes: bf9a379d0980 ("ext4: don't allow overlapping system zones")
> Signed-off-by: Theodore Ts'o <tytso@mit.edu>

Sashiko has reviewed this patch and found no issues.

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/20261112040638.1576836-1-tytso@mit.edu?part=15