Re: [PATCH 06/12] xfs: remove duplicate verification from xfs_qm_dqflush()

[Brian Foster <bfoster@redhat.com>]

On Mon, Apr 20, 2020 at 01:53:22PM +1000, Dave Chinner wrote:
> On Fri, Apr 17, 2020 at 11:08:53AM -0400, Brian Foster wrote:
> > The dquot read/write verifier calls xfs_dqblk_verify() on every
> > dquot in the buffer. Remove the duplicate call from
> > xfs_qm_dqflush().
> 
> Ah, I think there's a bug here - it's not supposed to be a
> duplicate....
> 
> > Signed-off-by: Brian Foster <bfoster@redhat.com>
> > ---
> >  fs/xfs/xfs_dquot.c | 14 --------------
> >  1 file changed, 14 deletions(-)
> > 
> > diff --git a/fs/xfs/xfs_dquot.c b/fs/xfs/xfs_dquot.c
> > index af2c8e5ceea0..73032c18a94a 100644
> > --- a/fs/xfs/xfs_dquot.c
> > +++ b/fs/xfs/xfs_dquot.c
> > @@ -1071,7 +1071,6 @@ xfs_qm_dqflush(
> >  	struct xfs_buf		*bp;
> >  	struct xfs_dqblk	*dqb;
> >  	struct xfs_disk_dquot	*ddqp;
> > -	xfs_failaddr_t		fa;
> >  	int			error;
> >  
> >  	ASSERT(XFS_DQ_IS_LOCKED(dqp));
> > @@ -1116,19 +1115,6 @@ xfs_qm_dqflush(
> >  	dqb = bp->b_addr + dqp->q_bufoffset;
> >  	ddqp = &dqb->dd_diskdq;
> >  
> > -	/*
> > -	 * A simple sanity check in case we got a corrupted dquot.
> > -	 */
> > -	fa = xfs_dqblk_verify(mp, dqb, be32_to_cpu(ddqp->d_id), 0);
> 
> So this verifies the on disk dquot ....
> 
> > -	if (fa) {
> > -		xfs_alert(mp, "corrupt dquot ID 0x%x in memory at %pS",
> 
> ...which issues an "in memory corruption" alert on failure...
> 
> > -				be32_to_cpu(ddqp->d_id), fa);
> > -		xfs_buf_relse(bp);
> > -		xfs_dqfunlock(dqp);
> > -		xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
> > -		return -EFSCORRUPTED;
> > -	}
> > -
> >  	/* This is the only portion of data that needs to persist */
> >  	memcpy(ddqp, &dqp->q_core, sizeof(struct xfs_disk_dquot));
> 
> .... and on success we immediately overwrite the on-disk copy with
> the unchecked in-memory copy of the dquot. 
> 
> IOWs, I think that verification call here should be checking the
> in-memory dquot core, not the on disk buffer that is about to get
> trashed.  i.e. something like this:
> 
> -	fa = xfs_dqblk_verify(mp, dqb, be32_to_cpu(ddqp->d_id), 0);
> +	fa = xfs_dquot_verify(mp, &dqp->q_core, be32_to_cpu(ddqp->d_id), 0);
> 

Isn't this still essentially duplicated by the write verifier? I don't
feel strongly about changing it as above vs. removing it, but it does
still seem unnecessary to me..

Brian

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