[lvc-project] [PATCH] fs/jfs: Prevent integer overflow in AG size calculation

[Dave Kleikamp]

On 2/20/25 3:52AM, Rand Deeb wrote:
> The JFS filesystem calculates allocation group (AG) size using 1 <<
> l2agsize in dbExtendFS(). When l2agsize exceeds 31 (possible with >2TB
> aggregates on 32-bit systems), this 32-bit shift operation causes undefined
> behavior and improper AG sizing.
> 
> On 32-bit architectures:
> - Left-shifting 1 by 32+ bits results in 0 due to integer overflow
> - This creates invalid AG sizes (0 or garbage values) in
> sbi->bmap->db_agsize
> - Subsequent block allocations would reference invalid AG structures
> - Could lead to:
>    - Filesystem corruption during extend operations
>    - Kernel crashes due to invalid memory accesses
>    - Security vulnerabilities via malformed on-disk structures
> 
> Fix by casting to s64 before shifting:
> bmp->db_agsize = (s64)1 << l2agsize;
> 
> This ensures 64-bit arithmetic even on 32-bit architectures. The cast
> matches the data type of db_agsize (s64) and follows similar patterns in
> JFS block calculation code.

Adding this to jfs-next.

Thanks,
Shaggy

> 
> Found by Linux Verification Center (linuxtesting.org) with SVACE.
> 
> Signed-off-by: Rand Deeb <rand.sec96 at gmail.com>
> ---
>   fs/jfs/jfs_dmap.c | 2 +-
>   1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/fs/jfs/jfs_dmap.c b/fs/jfs/jfs_dmap.c
> index f9009e4f9ffd..edb22cf9521a 100644
> --- a/fs/jfs/jfs_dmap.c
> +++ b/fs/jfs/jfs_dmap.c
> @@ -3403,7 +3403,7 @@ int dbExtendFS(struct inode *ipbmap, s64 blkno,	s64 nblocks)
>   	oldl2agsize = bmp->db_agl2size;
>   
>   	bmp->db_agl2size = l2agsize;
> -	bmp->db_agsize = 1 << l2agsize;
> +	bmp->db_agsize = (s64)1 << l2agsize;
>   
>   	/* compute new number of AG */
>   	agno = bmp->db_numag;