Re: [PATCH v2] ocfs2: validate global bitmap cl_bpc before resize

On 8/4/26 2:44 PM, ZhengYuan Huang wrote:
> [BUG]
> A corrupted global bitmap inode can make online group extension scan past
> the end of a group descriptor bitmap:
> 
> BUG: KASAN: use-after-free in _find_next_bit+0xef/0x120 lib/find_bit.c:157
> Read of size 8 at addr ffff888021b52000 by task syz.0.34/409
> Call Trace:
>  ...
>  _find_next_bit+0xef/0x120 lib/find_bit.c:157
>  find_next_bit include/linux/find.h:73 [inline]
>  find_next_bit_le include/linux/find.h:518 [inline]
>  ocfs2_find_max_contig_free_bits+0x53/0xb0 fs/ocfs2/suballoc.c:1292
>  ocfs2_update_last_group_and_inode fs/ocfs2/resize.c:127 [inline]
>  ocfs2_group_extend+0x83e/0x1ae0 fs/ocfs2/resize.c:350
>  ocfs2_ioctl+0x175/0x6e0 fs/ocfs2/ioctl.c:869
>  vfs_ioctl fs/ioctl.c:51 [inline]
>  __do_sys_ioctl fs/ioctl.c:597 [inline]
>  __se_sys_ioctl fs/ioctl.c:583 [inline]
>  __x64_sys_ioctl+0x197/0x1e0 fs/ioctl.c:583
>  ...
> 
> [CAUSE]
> ocfs2_group_extend() consumes the global bitmap dinode's cl_bpc value in
> resize arithmetic. The existing inode validation checked cl_bpc only for
> non-global chain allocators, so a corrupted global bitmap value could reach
> the resize path. With cl_bpc changed from 1 to 51457, extending by seven
> clusters wraps the u16 bit count and grows a 2048-bit group to 34567 bits,
> exceeding its 32256-bit bitmap capacity.
> 
> [FIX]
> Validate cl_bpc in ocfs2_validate_inode_block() for every chain allocator,
> including the global bitmap, against the value derived from the
> filesystem's cluster and block sizes. This rejects the corrupted dinode
> when it is read and removes the resize-local check that incorrectly assumed
> cl_bpc is always one. The resize path still uses the validated value for
> its arithmetic.
> 
> Fixes: d659072f7368 ("[PATCH 1/2] ocfs2: Add group extend for online resize")
> Signed-off-by: ZhengYuan Huang <[email protected]>
> ---
> v2:
> - Derive the expected cl_bpc from the filesystem block and cluster sizes.
> - Extend the existing inode-block validation to cover the global bitmap.
> - Remove the resize-local hardcoded cl_bpc check.
> ---
> ---
>  fs/ocfs2/inode.c | 9 ++++-----
>  1 file changed, 4 insertions(+), 5 deletions(-)
> 
> diff --git a/fs/ocfs2/inode.c b/fs/ocfs2/inode.c
> index 41db7dd39ed9..0e1f9ae0eb73 100644
> --- a/fs/ocfs2/inode.c
> +++ b/fs/ocfs2/inode.c
> @@ -1683,12 +1683,11 @@ int ocfs2_validate_inode_block(struct super_block *sb,
>  					 le16_to_cpu(cl->cl_next_free_rec));
>  			goto bail;
>  		}
> -		if (OCFS2_SB(sb)->bitmap_blkno &&
> -		    OCFS2_SB(sb)->bitmap_blkno != le64_to_cpu(di->i_blkno) &&
> -		    le16_to_cpu(cl->cl_bpc) != bpc) {
> -			rc = ocfs2_error(sb, "Invalid dinode %llu: bits per cluster %u\n",
> +		if (le16_to_cpu(cl->cl_bpc) != bpc) {

For global bitmap inode, its cl_bpc is legitimately 1.
So this is wrong and will break mount during fill super.
NAK.

Thanks,
Joseph

> +			rc = ocfs2_error(sb,
> +					 "Invalid dinode %llu: bits per cluster %u (expected %u)\n",
>  					 (unsigned long long)bh->b_blocknr,
> -					 le16_to_cpu(cl->cl_bpc));
> +					 le16_to_cpu(cl->cl_bpc), bpc);
>  			goto bail;
>  		}
>  	}