Re: [PATCH 1/3] xfs_repair: set NEEDSREPAIR the first time we write to a filesystem

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

On Tue, Feb 16, 2021 at 06:55:34AM -0500, Brian Foster wrote:
> On Fri, Feb 12, 2021 at 09:46:50PM -0800, Darrick J. Wong wrote:
> > From: Darrick J. Wong <djwong@kernel.org>
> > 
> > Add a hook to the buffer cache so that xfs_repair can intercept the
> > first write to a V5 filesystem to set the NEEDSREPAIR flag.  In the
> > event that xfs_repair dirties the filesystem and goes down, this ensures
> > that the sysadmin will have to re-start repair before mounting.
> > 
> > Signed-off-by: Darrick J. Wong <djwong@kernel.org>
> > ---
> >  include/xfs_mount.h |    4 ++
> >  libxfs/rdwr.c       |    4 ++
> >  repair/xfs_repair.c |   95 +++++++++++++++++++++++++++++++++++++++++++++++++++
> >  3 files changed, 103 insertions(+)
> > 
> > 
> ...
> > diff --git a/repair/xfs_repair.c b/repair/xfs_repair.c
> > index 90d1a95a..12e319ae 100644
> > --- a/repair/xfs_repair.c
> > +++ b/repair/xfs_repair.c
> > @@ -720,6 +720,8 @@ clear_needsrepair(
> >  	struct xfs_buf		*bp;
> >  	int			error;
> >  
> > +	mp->m_buf_writeback_fn = NULL;
> > +
> 
> Slight impedence mismatch in that we set the callback unconditionally
> (assuming crc=1) but we only get to this call if needsrepair is set.

<nod> I'll remove that.

> >  	/*
> >  	 * If we're going to clear NEEDSREPAIR, we need to make absolutely sure
> >  	 * that everything is ok with the ondisk filesystem.  Make sure any
> > @@ -751,6 +753,95 @@ clear_needsrepair(
> >  		libxfs_buf_relse(bp);
> >  }
> >  
> > +static void
> > +update_sb_crc_only(
> > +	struct xfs_buf		*bp)
> > +{
> > +	xfs_buf_update_cksum(bp, XFS_SB_CRC_OFF);
> > +}
> > +
> > +/* Forcibly write the primary superblock with the NEEDSREPAIR flag set. */
> > +static void
> > +force_needsrepair(
> > +	struct xfs_mount	*mp)
> > +{
> > +	struct xfs_buf_ops	fake_ops;
> > +	struct xfs_buf		*bp;
> > +	int			error;
> > +
> > +	if (!xfs_sb_version_hascrc(&mp->m_sb) ||
> > +	    xfs_sb_version_needsrepair(&mp->m_sb))
> > +		return;
> > +
> > +	bp = libxfs_getsb(mp);
> > +	if (!bp || bp->b_error) {
> > +		do_log(
> > +	_("couldn't get superblock to set needsrepair, err=%d\n"),
> > +				bp ? bp->b_error : ENOMEM);
> > +		return;
> > +	} else {
> > +		/*
> > +		 * It's possible that we need to set NEEDSREPAIR before we've
> > +		 * had a chance to fix the summary counters in the primary sb.
> > +		 * With the exception of those counters, phase 1 already
> > +		 * ensured that the geometry makes sense.
> > +		 *
> > +		 * Bad summary counters in the primary super can cause the
> > +		 * write verifier to fail, so substitute a dummy that only sets
> > +		 * the CRC.  In the event of a crash, NEEDSREPAIR will prevent
> > +		 * the kernel from mounting our potentially damaged filesystem
> > +		 * until repair is run again, so it's ok to bypass the usual
> > +		 * verification in this one case.
> > +		 */
> > +		fake_ops = xfs_sb_buf_ops; /* struct copy */
> > +		fake_ops.verify_write = update_sb_crc_only;
> > +
> > +		mp->m_sb.sb_features_incompat |=
> > +				XFS_SB_FEAT_INCOMPAT_NEEDSREPAIR;
> > +		libxfs_sb_to_disk(bp->b_addr, &mp->m_sb);
> > +
> > +		/* Force the primary super to disk immediately. */
> > +		bp->b_ops = &fake_ops;
> > +		error = -libxfs_bwrite(bp);
> 
> This looks like it calls back into the writeback hook before it's been
> cleared.

Yes, it does...

> I'm guessing the xfs_sb_version_needsrepair() check above cuts
> off any recursion issues,

...but as soon as we end up in the writeback hook, the b_ops and
XFS_SB_DADDR checks cause the hook to return without doing anything.

> but it might be cleaner to clear the callback pointer first.

We need to block all other writers until /after/ we've written the
primary super, which means that we can't clear the callback until some
time after the bwrite.

> 
> > +		bp->b_ops = &xfs_sb_buf_ops;
> > +		if (error)
> > +			do_log(_("couldn't force needsrepair, err=%d\n"), error);
> > +	}
> > +	if (bp)
> > +		libxfs_buf_relse(bp);
> 
> Doesn't appear we can get here with bp == NULL. Otherwise the rest looks
> reasonable to me.

Yep, will fix.

--D

> Brian
> 
> > +}
> > +
> > +/* Intercept the first write to the filesystem so we can set NEEDSREPAIR. */
> > +static void
> > +repair_capture_writeback(
> > +	struct xfs_buf		*bp)
> > +{
> > +	struct xfs_mount	*mp = bp->b_mount;
> > +	static pthread_mutex_t	wb_mutex = PTHREAD_MUTEX_INITIALIZER;
> > +
> > +	/* Higher level code modifying a superblock must set NEEDSREPAIR. */
> > +	if (bp->b_ops == &xfs_sb_buf_ops || bp->b_bn == XFS_SB_DADDR)
> > +		return;
> > +
> > +	pthread_mutex_lock(&wb_mutex);
> > +
> > +	/*
> > +	 * If someone else already dropped the hook, then needsrepair has
> > +	 * already been set on the filesystem and we can unlock.
> > +	 */
> > +	if (mp->m_buf_writeback_fn != repair_capture_writeback)
> > +		goto unlock;
> > +
> > +	/*
> > +	 * If we get here, the buffer being written is not a superblock, and
> > +	 * needsrepair needs to be set.
> > +	 */
> > +	force_needsrepair(mp);
> > +	mp->m_buf_writeback_fn = NULL;
> > +unlock:
> > +	pthread_mutex_unlock(&wb_mutex);
> > +}
> > +
> >  int
> >  main(int argc, char **argv)
> >  {
> > @@ -847,6 +938,10 @@ main(int argc, char **argv)
> >  	if (verbose > 2)
> >  		mp->m_flags |= LIBXFS_MOUNT_WANT_CORRUPTED;
> >  
> > +	/* Capture the first writeback so that we can set needsrepair. */
> > +	if (xfs_sb_version_hascrc(&mp->m_sb))
> > +		mp->m_buf_writeback_fn = repair_capture_writeback;
> > +
> >  	/*
> >  	 * set XFS-independent status vars from the mount/sb structure
> >  	 */
> > 
>