2022-10-19 13:26:56

by Hoi Pok Wu

Subject: [PATCH] fs: jfs: fix shift-out-of-bounds in dbDiscardAG

Bug reported by syzbot, UBSAN: array-index-out-of-bounds in dbDiscardAG.
Fixed by simply rearranging agno_end smaller than MAXAG. However, I wonder if it
is a better idea to return -EINVAL, rejecting user input. Thank you.

Reported-by: [email protected]
Signed-off-by: Hoi Pok Wu <[email protected]>
---
fs/jfs/jfs_discard.c | 4 ++++
1 file changed, 4 insertions(+)

diff --git a/fs/jfs/jfs_discard.c b/fs/jfs/jfs_discard.c
index 575cb2ba74fc..3101eaf3098a 100644
--- a/fs/jfs/jfs_discard.c
+++ b/fs/jfs/jfs_discard.c
@@ -96,6 +96,10 @@ int jfs_ioc_trim(struct inode *ip, struct fstrim_range *range)
*/
agno = BLKTOAG(start, JFS_SBI(ip->i_sb));
agno_end = BLKTOAG(end, JFS_SBI(ip->i_sb));
+
+ if (agno_end >= MAXAG)
+ agno_end = MAXAG - 1;
+
while (agno <= agno_end) {
trimmed += dbDiscardAG(ip, agno, minlen);
agno++;
--
2.38.0

2022-10-25 00:10:23

by Dave Kleikamp

Subject: Re: [PATCH] fs: jfs: fix shift-out-of-bounds in dbDiscardAG

On 10/19/22 4:38AM, Hoi Pok Wu wrote:
> Bug reported by syzbot, UBSAN: array-index-out-of-bounds in dbDiscardAG.
> Fixed by simply rearranging agno_end smaller than MAXAG. However, I wonder if it
> is a better idea to return -EINVAL, rejecting user input. Thank you.

I don't think the problem is simply that the user input is bad, the
range is already limited:

if (end >= bmp->db_mapsize)
end = bmp->db_mapsize - 1;

It seems syzbot corrupted some on-disk data structure to trigger this
failure. It may be best to call jfs_error() and bail out.

Thanks,
Shaggy

>
> Reported-by: [email protected]
> Signed-off-by: Hoi Pok Wu <[email protected]>
> ---
> fs/jfs/jfs_discard.c | 4 ++++
> 1 file changed, 4 insertions(+)
>
> diff --git a/fs/jfs/jfs_discard.c b/fs/jfs/jfs_discard.c
> index 575cb2ba74fc..3101eaf3098a 100644
> --- a/fs/jfs/jfs_discard.c
> +++ b/fs/jfs/jfs_discard.c
> @@ -96,6 +96,10 @@ int jfs_ioc_trim(struct inode *ip, struct fstrim_range *range)
> */
> agno = BLKTOAG(start, JFS_SBI(ip->i_sb));
> agno_end = BLKTOAG(end, JFS_SBI(ip->i_sb));
> +
> + if (agno_end >= MAXAG)
> + agno_end = MAXAG - 1;
> +
> while (agno <= agno_end) {
> trimmed += dbDiscardAG(ip, agno, minlen);
> agno++;

2022-10-25 05:30:19

by Hoi Pok Wu

Subject: Re: [PATCH] fs: jfs: fix shift-out-of-bounds in dbDiscardAG

Thanks for the suggestion! I will send a Patch V2 soon.


On Tue, Oct 25, 2022 at 5:05 AM Dave Kleikamp <[email protected]> wrote:
>
> On 10/19/22 4:38AM, Hoi Pok Wu wrote:
> > Bug reported by syzbot, UBSAN: array-index-out-of-bounds in dbDiscardAG.
> > Fixed by simply rearranging agno_end smaller than MAXAG. However, I wonder if it
> > is a better idea to return -EINVAL, rejecting user input. Thank you.
>
> I don't think the problem is simply that the user input is bad, the
> range is already limited:
>
> if (end >= bmp->db_mapsize)
> end = bmp->db_mapsize - 1;
>
> It seems syzbot corrupted some on-disk data structure to trigger this
> failure. It may be best to call jfs_error() and bail out.
>
> Thanks,
> Shaggy
>
> >
> > Reported-by: [email protected]
> > Signed-off-by: Hoi Pok Wu <[email protected]>
> > ---
> > fs/jfs/jfs_discard.c | 4 ++++
> > 1 file changed, 4 insertions(+)
> >
> > diff --git a/fs/jfs/jfs_discard.c b/fs/jfs/jfs_discard.c
> > index 575cb2ba74fc..3101eaf3098a 100644
> > --- a/fs/jfs/jfs_discard.c
> > +++ b/fs/jfs/jfs_discard.c
> > @@ -96,6 +96,10 @@ int jfs_ioc_trim(struct inode *ip, struct fstrim_range *range)
> > */
> > agno = BLKTOAG(start, JFS_SBI(ip->i_sb));
> > agno_end = BLKTOAG(end, JFS_SBI(ip->i_sb));
> > +
> > + if (agno_end >= MAXAG)
> > + agno_end = MAXAG - 1;
> > +
> > while (agno <= agno_end) {
> > trimmed += dbDiscardAG(ip, agno, minlen);
> > agno++;