Re: [PATCH 0/3] xfs: more verifications!

[Dave Chinner <david@fromorbit.com>]

On Mon, Jun 04, 2018 at 08:38:00PM -0700, Darrick J. Wong wrote:
> On Tue, Jun 05, 2018 at 12:43:10PM +1000, Dave Chinner wrote:
> > Hi folks,
> > 
> > This patchset addresses a series of corruptions noticed in an image
> > provided by Wen Xu. The patches only address detecting corruptions
> > before they can do damage - it does not address/fix the crash the
> > image caused but instead prevents the bad information from getting
> > to the point where it can cause a crash.
> > 
> > The third patch addresses a general btree record validation issue;
> > we should probably dirve this inwards to each btree implementation
> > with a ->verify_record() callout, as this patch does not address
> > all the places that btree records are traversed during searches. It
> > will catch any attempt to use a bad record that a search lands on,
> > however.
> > 
> > Comments?
> 
> So, uh, does xfs_repair catch all these things and fix them?

Don't know - I couldn't test repair on the damaged image because it
was a single AG and repair barfed on the primary superblock and
couldn't recover it.

Yes, both Eric and I have asked for this fuzzing to be done on 2 AG
filesystems so we can reliably test the userspace tools against
them, too, but we're not making any progress there...

Cheers,

Dave.
-- 
Dave Chinner
david@fromorbit.com