Re: [PATCH v3] jfs: UBSAN: shift-out-of-bounds in dbFindBits

[Dave Kleikamp <dave.kleikamp@oracle.com>]

On 10/25/24 12:00PM, Matt Jan wrote:
> Return immediately if the needed free bits span a full word to avoid
> out-of-bounds shifting.

dbFindBits really shouldn't be called with l2nb == BUDMIN. Something in 
the dmap is corrupt and this patch just lets things continue as if 
nothing is wrong. I think a sanity check in dbAllocDmapLev where we can 
return -EIO makes more sense.

Shaggy

> 
> #syz test
> 
> Reported-by: syzbot+9e90a1c5eedb9dc4c6cc@syzkaller.appspotmail.com
> Signed-off-by: Matt Jan <zoo868e@gmail.com>
> ---
> Changes in v3: Return the result earlier instead of assert it
> Changes in v2: Test if the patch resolve the issue through syzbot and
> reference the reporter.
> 
>   fs/jfs/jfs_dmap.c | 5 +++++
>   1 file changed, 5 insertions(+)
> 
> diff --git a/fs/jfs/jfs_dmap.c b/fs/jfs/jfs_dmap.c
> index 974ecf5e0d95..45b7a393b769 100644
> --- a/fs/jfs/jfs_dmap.c
> +++ b/fs/jfs/jfs_dmap.c
> @@ -3012,6 +3012,11 @@ static int dbFindBits(u32 word, int l2nb)
>   	int bitno, nb;
>   	u32 mask;
>   
> +	/* return immediately if the number of free bits is a word
> +	 */
> +	if (l2nb == BUDMIN)
> +		return (!!word) << BUDMIN;
> +
>   	/* get the number of bits.
>   	 */
>   	nb = 1 << l2nb;