[PATCH] ocfs2: fix listxattr handling when the buffer is full

[PATCH] ocfs2: fix listxattr handling when the buffer is full

[ZhengYuan Huang]

[BUG]
If an OCFS2 inode has both inline and block-based xattrs, listxattr()
can return a size larger than the caller's buffer when the inline names
consume that buffer exactly.

kernel BUG at mm/usercopy.c:102!
Oops: invalid opcode: 0000 [#1] SMP KASAN NOPTI
RIP: 0010:usercopy_abort+0xb7/0xd0 mm/usercopy.c:102
Call Trace:
 __check_heap_object+0xe3/0x120 mm/slub.c:8243
 check_heap_object mm/usercopy.c:196 [inline]
 __check_object_size mm/usercopy.c:250 [inline]
 __check_object_size+0x5c5/0x780 mm/usercopy.c:215
 check_object_size include/linux/ucopysize.h:22 [inline]
 check_copy_size include/linux/ucopysize.h:59 [inline]
 copy_to_user include/linux/uaccess.h:219 [inline]
 listxattr+0xb0/0x170 fs/xattr.c:926
 filename_listxattr fs/xattr.c:958 [inline]
 path_listxattrat+0x137/0x320 fs/xattr.c:988
 __do_sys_listxattr fs/xattr.c:1001 [inline]
 __se_sys_listxattr fs/xattr.c:998 [inline]
 __x64_sys_listxattr+0x7f/0xd0 fs/xattr.c:998
 ...

[CAUSE]
Commit 936b8834366e ("ocfs2: Refactor xattr list and remove
ocfs2_xattr_handler().") replaced the old per-handler list accounting
with ocfs2_xattr_list_entry(), but it kept using size == 0 to detect
probe mode.

That assumption stops being true once ocfs2_listxattr() finishes the
inline-xattr pass. If the inline names fill the caller buffer exactly,
the block-xattr pass runs with a non-NULL buffer and a remaining size of
zero. ocfs2_xattr_list_entry() then skips the bounds check, keeps
counting block names, and returns a positive size larger than the
supplied buffer.

[FIX]
Detect probe mode by testing whether the destination buffer pointer is
NULL instead of whether the remaining size is zero.

That restores the pre-refactor behavior and matches the OCFS2 getxattr
helpers. Once the remaining buffer reaches zero while more names are
left, the block-xattr pass now returns -ERANGE instead of reporting a
size larger than the allocated list buffer.

Fixes: 936b8834366e ("ocfs2: Refactor xattr list and remove ocfs2_xattr_handler().")
Signed-off-by: ZhengYuan Huang <gality369@gmail.com>
---
 fs/ocfs2/xattr.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/fs/ocfs2/xattr.c b/fs/ocfs2/xattr.c
--- a/fs/ocfs2/xattr.c
+++ b/fs/ocfs2/xattr.c
@@ -906,8 +906,8 @@ static int ocfs2_xattr_list_entry(struct super_block *sb,
 	prefix_len = strlen(prefix);
 	total_len = prefix_len + name_len + 1;
 	*result += total_len;

-	/* we are just looking for how big our buffer needs to be */
-	if (!size)
+	/* No buffer means we are only looking for the required size. */
+	if (!buffer)
 		return 0;
 
 	if (*result > size)
-- 
2.49.0

Re: [PATCH] ocfs2: fix listxattr handling when the buffer is full

[Joseph Qi]

On 4/10/26 12:03 PM, ZhengYuan Huang wrote:
> [BUG]
> If an OCFS2 inode has both inline and block-based xattrs, listxattr()
> can return a size larger than the caller's buffer when the inline names
> consume that buffer exactly.
> 
> kernel BUG at mm/usercopy.c:102!
> Oops: invalid opcode: 0000 [#1] SMP KASAN NOPTI
> RIP: 0010:usercopy_abort+0xb7/0xd0 mm/usercopy.c:102
> Call Trace:
>  __check_heap_object+0xe3/0x120 mm/slub.c:8243
>  check_heap_object mm/usercopy.c:196 [inline]
>  __check_object_size mm/usercopy.c:250 [inline]
>  __check_object_size+0x5c5/0x780 mm/usercopy.c:215
>  check_object_size include/linux/ucopysize.h:22 [inline]
>  check_copy_size include/linux/ucopysize.h:59 [inline]
>  copy_to_user include/linux/uaccess.h:219 [inline]
>  listxattr+0xb0/0x170 fs/xattr.c:926
>  filename_listxattr fs/xattr.c:958 [inline]
>  path_listxattrat+0x137/0x320 fs/xattr.c:988
>  __do_sys_listxattr fs/xattr.c:1001 [inline]
>  __se_sys_listxattr fs/xattr.c:998 [inline]
>  __x64_sys_listxattr+0x7f/0xd0 fs/xattr.c:998
>  ...
> 
> [CAUSE]
> Commit 936b8834366e ("ocfs2: Refactor xattr list and remove
> ocfs2_xattr_handler().") replaced the old per-handler list accounting
> with ocfs2_xattr_list_entry(), but it kept using size == 0 to detect
> probe mode.
> 
> That assumption stops being true once ocfs2_listxattr() finishes the
> inline-xattr pass. If the inline names fill the caller buffer exactly,
> the block-xattr pass runs with a non-NULL buffer and a remaining size of
> zero. ocfs2_xattr_list_entry() then skips the bounds check, keeps
> counting block names, and returns a positive size larger than the
> supplied buffer.
> 
> [FIX]
> Detect probe mode by testing whether the destination buffer pointer is
> NULL instead of whether the remaining size is zero.
> 
> That restores the pre-refactor behavior and matches the OCFS2 getxattr
> helpers. Once the remaining buffer reaches zero while more names are
> left, the block-xattr pass now returns -ERANGE instead of reporting a
> size larger than the allocated list buffer.
> 
> Fixes: 936b8834366e ("ocfs2: Refactor xattr list and remove ocfs2_xattr_handler().")
> Signed-off-by: ZhengYuan Huang <gality369@gmail.com>

Looks good.
Reviewed-by: Joseph Qi <joseph.qi@linux.alibaba.com>

> ---
>  fs/ocfs2/xattr.c | 4 ++--
>  1 file changed, 2 insertions(+), 2 deletions(-)
> 
> diff --git a/fs/ocfs2/xattr.c b/fs/ocfs2/xattr.c
> --- a/fs/ocfs2/xattr.c
> +++ b/fs/ocfs2/xattr.c
> @@ -906,8 +906,8 @@ static int ocfs2_xattr_list_entry(struct super_block *sb,
>  	prefix_len = strlen(prefix);
>  	total_len = prefix_len + name_len + 1;
>  	*result += total_len;
> 
> -	/* we are just looking for how big our buffer needs to be */
> -	if (!size)
> +	/* No buffer means we are only looking for the required size. */
> +	if (!buffer)
>  		return 0;
>  
>  	if (*result > size)