Re: [PATCH v2.1 2/4] xfs: avoid buffer deadlocks when walking fs inodes

["Darrick J. Wong" <djwong@kernel.org>]

On Mon, Mar 08, 2021 at 09:37:03AM +1100, Dave Chinner wrote:
> On Sun, Mar 07, 2021 at 12:36:38PM -0800, Darrick J. Wong wrote:
> > From: Darrick J. Wong <djwong@kernel.org>
> > 
> > When we're servicing an INUMBERS or BULKSTAT request or running
> > quotacheck, grab an empty transaction so that we can use its inherent
> > recursive buffer locking abilities to detect inode btree cycles without
> > hitting ABBA buffer deadlocks.
> > 
> > Found by fuzzing an inode btree pointer to introduce a cycle into the
> > tree (xfs/365).
> > 
> > Signed-off-by: Darrick J. Wong <djwong@kernel.org>
> > ---
> > v2.1: actually pass tp in the bulkstat_single case
> > ---
> >  fs/xfs/xfs_itable.c |   42 +++++++++++++++++++++++++++++++++++++-----
> >  fs/xfs/xfs_iwalk.c  |   32 +++++++++++++++++++++++++++-----
> >  2 files changed, 64 insertions(+), 10 deletions(-)
> 
> Looks ok, but I can't help but wonder if this case should flag
> corruption if lock recursion does actually occur...

Hm.  Scrub will, since it checks the level and block number of every
pointer it follows.  IIRC the rest of the btree code isn't as paranoid
about things like that.

> Reviewed-by: Dave Chinner <dchinner@redhat.com>

Anyway, thanks for the reviews.

--D

> -- 
> Dave Chinner
> david@fromorbit.com