Re: [PATCH v2] ocfs2: fix kernel BUG in ocfs2_write_block

[Joseph Qi <joseph.qi@linux.alibaba.com>]

On 2025/12/11 03:32, Prithvi Tambewagh wrote:
> When the filesystem is being mounted, the kernel panics while the data
> regarding slot map allocation to the local node, is being written to the
> disk. This occurs because the value of slot map buffer head block
> number, which should have been greater than or equal to
> `OCFS2_SUPER_BLOCK_BLKNO` (evaluating to 2) is less than it, indicative
> of disk metadata corruption. This triggers
> BUG_ON(bh->b_blocknr < OCFS2_SUPER_BLOCK_BLKNO) in ocfs2_write_block(),
> causing the kernel to panic.
> 
> This is fixed by introducing an if condition block in
> ocfs2_update_disk_slot(), right before calling ocfs2_write_block(), which
> checks if `bh->b_blocknr` is lesser than `OCFS2_SUPER_BLOCK_BLKNO`; if
> yes, then ocfs2_error is called, which prints the error log, for
> debugging purposes, and the return value of ocfs2_error() is returned
> back to caller of ocfs2_update_disk_slot() i.e. ocfs2_find_slot(). If
> the return value is zero. then error code EIO is returned.
> 
> Reported-by: syzbot+c818e5c4559444f88aa0@syzkaller.appspotmail.com
> Closes: https://syzkaller.appspot.com/bug?extid=c818e5c4559444f88aa0
> Tested-by: syzbot+c818e5c4559444f88aa0@syzkaller.appspotmail.com
> Cc: stable@vger.kernel.org
> Signed-off-by: Prithvi Tambewagh <activprithvi@gmail.com>
> ---
> v1->v2:
>  - Remove usage of le16_to_cpu() from ocfs2_error()
>  - Cast bh->b_blocknr to unsigned long long
>  - Remove type casting for OCFS2_SUPER_BLOCK_BLKNO
>  - Fix Sparse warnings reported in v1 by kernel test robot
>  - Update title from 'ocfs2: Fix kernel BUG in ocfs2_write_block' to
>    'ocfs2: fix kernel BUG in ocfs2_write_block'
> 
> v1 link: https://lore.kernel.org/all/20251206154819.175479-1-activprithvi@gmail.com/T/
> 
>  fs/ocfs2/slot_map.c | 10 ++++++++++
>  1 file changed, 10 insertions(+)
> 
> diff --git a/fs/ocfs2/slot_map.c b/fs/ocfs2/slot_map.c
> index e544c704b583..e916a2e8f92d 100644
> --- a/fs/ocfs2/slot_map.c
> +++ b/fs/ocfs2/slot_map.c
> @@ -193,6 +193,16 @@ static int ocfs2_update_disk_slot(struct ocfs2_super *osb,
>  	else
>  		ocfs2_update_disk_slot_old(si, slot_num, &bh);
>  	spin_unlock(&osb->osb_lock);
> +	if (bh->b_blocknr < OCFS2_SUPER_BLOCK_BLKNO) {
> +		status = ocfs2_error(osb->sb,
> +				     "Invalid Slot Map Buffer Head "
> +				     "Block Number : %llu, Should be >= %d",
> +				     (unsigned long long)bh->b_blocknr,
> +				     OCFS2_SUPER_BLOCK_BLKNO);
> +		if (!status)
> +			return -EIO;
> +		return status;
> +	}
>  
>  	status = ocfs2_write_block(osb, bh, INODE_CACHE(si->si_inode));
>  	if (status < 0)
> 

Ummm... The 'bh' is from ocfs2_slot_info, which is load from crafted
image during mount.
So IIUC, the root cause is we read slot info without validating, see
ocfs2_refresh_slot_info().
So I'd prefer to implement a validate func and pass it into
ocfs2_read_blocks() to do this job.

Thanks,
Joseph