On Sat 21-07-12 16:35:17, Ashish Sangwan wrote:
> If s_lvid_bh is not freed and set to NULL before re-scanning partition
> with default block size, we might end up using wrong lvid in case
> s_lvid_bh is not updated in udf_load_logicalvolint during rescan.
>
> Signed-off-by: Ashish Sangwan <[email protected]>
> Signed-off-by: Namjae Jeon <[email protected]>
Yeah, I don't think this happens in practice but in theory it could. BTW,
did you check whether we don't need to free other information (like VAT
inode etc.) when rescanning the filesystem? I think we do but currently I'm
catching up after a long vacation and this doesn't have high priority.
Anyway, I've added your patch to my tree. Thanks.
Honza
>
> ---
> fs/udf/super.c | 2 ++
> 1 files changed, 2 insertions(+), 0 deletions(-)
>
> diff --git a/fs/udf/super.c b/fs/udf/super.c
> index ac8a348..4a95c56 100644
> --- a/fs/udf/super.c
> +++ b/fs/udf/super.c
> @@ -1961,6 +1961,8 @@ static int udf_fill_super(struct super_block *sb, void *options, int silent)
> if (!silent)
> pr_notice("Rescanning with blocksize %d\n",
> UDF_DEFAULT_BLOCKSIZE);
> + brelse(sbi->s_lvid_bh);
> + sbi->s_lvid_bh = NULL;
> uopt.blocksize = UDF_DEFAULT_BLOCKSIZE;
> ret = udf_load_vrs(sb, &uopt, silent, &fileset);
> }
> --
> 1.7.2.3
>
--
Jan Kara <[email protected]>
SUSE Labs, CR
Hi Jan,
> Yeah, I don't think this happens in practice but in theory it could. BTW,
> did you check whether we don't need to free other information (like VAT
> inode etc.) when rescanning the filesystem? I think we do but currently I'm
> catching up after a long vacation and this doesn't have high priority.
>
Actually, it did happen in practice. That's how we discovered it.
Currently, it seems a bug.
We formatted a usb stick, sector size 512bytes, using mkudffs with
default block size 2KB.
While writing to this media we unplug the USB which left lvid in
corrupted state.
On remounting, first UDF tries to mount the media with sector size and
somehow it managed to fill lvid bh
but failed to load metadata/mirror fe because of wrong block size.
While rescanning with 2KB block size it failed to load the correct
lvid as it was corrupted earlier
and ended up using the wrong one.
After noticing this problem, we did check other info too. Everything
else seems to be correct.
> Anyway, I've added your patch to my tree. Thanks.
>
Thanks for looking into the patch.
Ashish