Re: [PATCH] ocfs2: validate dx extent list bounds during lookup
From: Joseph Qi
Date: Fri Apr 03 2026 - 04:31:04 EST
This makes ocfs2_dx_dir_lookup_rec() messy.
And I think the right place to do this check is ocfs2_validate_dx_root()
and ocfs2_validate_extent_block() during read from disk.
I'll rework it and send out later.
Thanks,
Joseph
On 4/1/26 9:39 AM, ZhengYuan Huang wrote:
> [BUG]
> A corrupted indexed directory can trigger a KASAN use-after-free in
> ocfs2_dx_dir_lookup_rec() when the dx root or leaf extent list carries
> an out-of-range l_count or l_next_free_rec value.
>
> BUG: KASAN: use-after-free in ocfs2_dx_dir_lookup_rec+0x6f7/0x880 fs/ocfs2/dir.c:813
> Read of size 4 at addr ffff888043b7b0e0 by task syz.0.3467/8031
>
> Call Trace:
> <TASK>
> ...
> ocfs2_dx_dir_lookup_rec+0x6f7/0x880 fs/ocfs2/dir.c:813
> ocfs2_dx_dir_lookup+0x100/0x5d0 fs/ocfs2/dir.c:868
> ocfs2_dx_dir_search+0x7bc/0x11c0 fs/ocfs2/dir.c:928
> ocfs2_find_entry_dx fs/ocfs2/dir.c:1042 [inline]
> ocfs2_find_entry+0x97e/0xce0 fs/ocfs2/dir.c:1079
> ocfs2_find_files_on_disk+0xa9/0x2f0 fs/ocfs2/dir.c:2002
> ocfs2_lookup_ino_from_name+0xae/0x110 fs/ocfs2/dir.c:2024
> ocfs2_lookup+0x45e/0x860 fs/ocfs2/namei.c:122
> lookup_open.isra.0+0x4a2/0x1460 fs/namei.c:3774
> open_last_lookups fs/namei.c:3895 [inline]
> path_openat+0x11fe/0x2ce0 fs/namei.c:4131
> do_filp_open+0x1f6/0x430 fs/namei.c:4161
> do_sys_openat2+0x117/0x1c0 fs/open.c:1437
> do_sys_open fs/open.c:1452 [inline]
> __do_sys_openat fs/open.c:1468 [inline]
> __se_sys_openat fs/open.c:1463 [inline]
> __x64_sys_openat+0x15b/0x220 fs/open.c:1463
> ...
>
> [CAUSE]
> ocfs2_dx_dir_lookup_rec() only checked for an empty extent list before
> iterating over the directory index records. It did not verify that the
> root dx list fits within a dx root block, or that the leaf list fits
> within an extent block.
>
> Comparing l_next_free_rec only against the on-disk l_count is also not
> enough because l_count itself can be corrupted.
>
> [FIX]
> Validate both the dx root list and the leaf extent list against their
> physical record capacities before indexing into l_recs[]. Use
> ocfs2_extent_recs_per_dx_root() for the root and
> ocfs2_extent_recs_per_eb() for the leaf block so the bounds check does
> not trust on-disk sizing fields.
>
> Fixes: 9b7895efac90 ("ocfs2: Add a name indexed b-tree to directory inodes")
> Signed-off-by: ZhengYuan Huang <gality369@xxxxxxxxx>
> ---
> fs/ocfs2/dir.c | 41 +++++++++++++++++++++++++++++++++++++++++
> 1 file changed, 41 insertions(+)
>
> diff --git a/fs/ocfs2/dir.c b/fs/ocfs2/dir.c
> index b82fe4431eb1..3403c0d3810e 100644
> --- a/fs/ocfs2/dir.c
> +++ b/fs/ocfs2/dir.c
> @@ -790,15 +790,25 @@ static int ocfs2_dx_dir_lookup_rec(struct inode *inode,
> struct buffer_head *eb_bh = NULL;
> struct ocfs2_extent_block *eb;
> struct ocfs2_extent_rec *rec = NULL;
> + unsigned int max_recs;
>
> - if (le16_to_cpu(el->l_count) !=
> - ocfs2_extent_recs_per_dx_root(inode->i_sb)) {
> + max_recs = ocfs2_extent_recs_per_dx_root(inode->i_sb);
> + if (le16_to_cpu(el->l_count) != max_recs) {
> ret = ocfs2_error(inode->i_sb,
> "Inode %llu has invalid extent list length %u\n",
> inode->i_ino, le16_to_cpu(el->l_count));
> goto out;
> }
>
> + if (le16_to_cpu(el->l_next_free_rec) > max_recs) {
> + ret = ocfs2_error(inode->i_sb,
> + "Inode %llu has invalid dx root next free %u, max %u\n",
> + inode->i_ino,
> + le16_to_cpu(el->l_next_free_rec),
> + max_recs);
> + goto out;
> + }
> +
> if (el->l_tree_depth) {
> ret = ocfs2_find_leaf(INODE_CACHE(inode), el, major_hash,
> &eb_bh);
> @@ -817,6 +827,27 @@ static int ocfs2_dx_dir_lookup_rec(struct inode *inode,
> (unsigned long long)eb_bh->b_blocknr);
> goto out;
> }
> +
> + max_recs = ocfs2_extent_recs_per_eb(inode->i_sb);
> + if (le16_to_cpu(el->l_count) != max_recs) {
> + ret = ocfs2_error(inode->i_sb,
> + "Inode %llu has invalid tree block %llu list count %u, max %u\n",
> + inode->i_ino,
> + (unsigned long long)eb_bh->b_blocknr,
> + le16_to_cpu(el->l_count),
> + max_recs);
> + goto out;
> + }
> +
> + if (le16_to_cpu(el->l_next_free_rec) > max_recs) {
> + ret = ocfs2_error(inode->i_sb,
> + "Inode %llu has invalid tree block %llu next free %u, max %u\n",
> + inode->i_ino,
> + (unsigned long long)eb_bh->b_blocknr,
> + le16_to_cpu(el->l_next_free_rec),
> + max_recs);
> + goto out;
> + }
> }
>
> if (le16_to_cpu(el->l_next_free_rec) == 0) {