ocfs2: validate inline xattr header before ibody lookups

Patch series "ocfs2: validate inline xattr header consumers".

Corrupt i_xattr_inline_size can move the computed inode-body xattr header
outside the dinode block. Several OCFS2 paths then trust xh_count or
xattr entry geometry from that unchecked header.

The reported KASAN splat hits the ibody lookup path:

  BUG: KASAN: use-after-free in ocfs2_xattr_find_entry+0x37b/0x3a0
  ocfs2_xattr_ibody_get()
  ocfs2_xattr_get_nolock()
  ocfs2_calc_xattr_init()

The same unchecked header derivation also exists in the outside-value
probe, ibody remove, inline refcount attach, and inline reflink paths.

This series factors the existing ibody list validation into a shared
helper and then converts the remaining inline-header consumers one at a
time.

Patch layout:

1. validate ibody get/find and reuse the helper in ibody list
2. validate the outside-value probe
3. validate ibody remove
4. validate inline refcount attach
5. validate inline reflink


This patch (of 5):

[BUG]
mknodat() can read past the end of a dinode block when ACL inheritance
walks a corrupted inode-body xattr header. Another report shows the same
unchecked lookup later faulting in the VFS open path after create
returns a garbage status.

KASAN: use-after-free in
ocfs2_xattr_find_entry+0x37b/0x3a0 fs/ocfs2/xattr.c:1078
Read of size 2 at addr ffff88801c520300 by task syz.0.10/360

Trace:
 ...
 ocfs2_xattr_find_entry+0x37b/0x3a0 fs/ocfs2/xattr.c:1078
 ocfs2_xattr_ibody_get fs/ocfs2/xattr.c:1178 [inline]
 ocfs2_xattr_get_nolock+0x2ee/0x1110 fs/ocfs2/xattr.c:1309
 ocfs2_calc_xattr_init+0x716/0xac0 fs/ocfs2/xattr.c:628
 ocfs2_mknod+0x935/0x2400 fs/ocfs2/namei.c:333
 ocfs2_create+0x158/0x390 fs/ocfs2/namei.c:676
 vfs_create fs/namei.c:3493 [inline]
 vfs_create+0x445/0x6f0 fs/namei.c:3477
 do_mknodat+0x2d8/0x5e0 fs/namei.c:4372
 __do_sys_mknodat fs/namei.c:4400 [inline]
 __se_sys_mknodat fs/namei.c:4397 [inline]
 __x64_sys_mknodat+0xb6/0xf0 fs/namei.c:4397
 ...

Another report:
 BUG: unable to handle page fault for address: fffffbfff3e40ec0
 RIP: 0010:__d_entry_type include/linux/dcache.h:414 [inline]
 RIP: 0010:d_can_lookup include/linux/dcache.h:429 [inline]
 RIP: 0010:d_is_dir include/linux/dcache.h:439 [inline]
 RIP: 0010:path_openat+0xe2f/0x2ce0 fs/namei.c:4134

Trace:
 ...
 do_filp_open+0x1f6/0x430 fs/namei.c:4161
 do_sys_openat2+0x117/0x1c0 fs/open.c:1437
 __x64_sys_openat+0x15b/0x220 fs/open.c:1463
 ...

[CAUSE]
ocfs2_xattr_ibody_list() already validates the inline xattr size and
entry count, but ocfs2_xattr_ibody_get() and ocfs2_xattr_ibody_find()
still derive the inline header directly from di->i_xattr_inline_size and
then trust xh_count. A corrupted inline size or entry count can therefore
move the computed header outside the dinode block before get/find start
walking it. That can either make ocfs2_xattr_find_entry() dereference
xs->header->xh_count outside the block or make ocfs2_xattr_get_nolock()
bubble a garbage status back through ocfs2_calc_xattr_init() into the
create/open path.

[FIX]
Factor the existing ibody header geometry checks into a shared helper.
Use it in ocfs2_xattr_ibody_get() and ocfs2_xattr_ibody_find(), and have
ocfs2_xattr_ibody_list() reuse the same helper instead of open-coding
the validation. Reject corrupt ibody metadata with -EFSCORRUPTED before
the lookup path can walk bogus xattr geometry or return a garbage status.

Link: https://lore.kernel.org/20260508085914.61647-1-gality369@gmail.com
Link: https://lore.kernel.org/20260508085914.61647-2-gality369@gmail.com
Signed-off-by: ZhengYuan Huang <gality369@gmail.com>
Reviewed-by: Joseph Qi <joseph.qi@linux.alibaba.com>
Cc: Jia-Ju Bai <baijiaju1990@gmail.com>
Cc: Zixuan Fu <r33s3n6@gmail.com>
Cc: Mark Fasheh <mark@fasheh.com>
Cc: Joel Becker <jlbec@evilplan.org>
Cc: Junxiao Bi <junxiao.bi@oracle.com>
Cc: Changwei Ge <gechangwei@live.cn>
Cc: Jun Piao <piaojun@huawei.com>
Cc: Heming Zhao <heming.zhao@suse.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>

0 files changed, 0 insertions, 0 deletions