Re: [PATCH 05/12] xfs: make xfs_buf_read_map return an error code

From: "Darrick J. Wong" <darrick.wong@oracle.com>
To: Dave Chinner <david@fromorbit.com>
Cc: linux-xfs@vger.kernel.org, hch@infradead.org
Subject: Re: [PATCH 05/12] xfs: make xfs_buf_read_map return an error code
Date: Thu, 23 Jan 2020 20:34:23 -0800	[thread overview]
Message-ID: <20200124043423.GC8247@magnolia> (raw)
In-Reply-To: <20200124013152.GF7090@dread.disaster.area>

On Fri, Jan 24, 2020 at 12:31:52PM +1100, Dave Chinner wrote:
> On Wed, Jan 22, 2020 at 11:42:22PM -0800, Darrick J. Wong wrote:
> > From: Darrick J. Wong <darrick.wong@oracle.com>
> > 
> > Convert xfs_buf_read_map() to return numeric error codes like most
> > everywhere else in xfs.  This involves moving the open-coded logic that
> > reports metadata IO read / corruption errors and stales the buffer into
> > xfs_buf_read_map so that the logic is all in one place.
> > 
> > Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
> .....
> 
> > diff --git a/fs/xfs/xfs_trans_buf.c b/fs/xfs/xfs_trans_buf.c
> > index b5b3a78ef31c..56e7f8126cd7 100644
> > --- a/fs/xfs/xfs_trans_buf.c
> > +++ b/fs/xfs/xfs_trans_buf.c
> > @@ -298,36 +298,17 @@ xfs_trans_read_buf_map(
> >  		return 0;
> >  	}
> >  
> > -	bp = xfs_buf_read_map(target, map, nmaps, flags, ops);
> > -	if (!bp) {
> > -		if (!(flags & XBF_TRYLOCK))
> > -			return -ENOMEM;
> > -		return tp ? 0 : -EAGAIN;
> > -	}
> > -
> > -	/*
> > -	 * If we've had a read error, then the contents of the buffer are
> > -	 * invalid and should not be used. To ensure that a followup read tries
> > -	 * to pull the buffer from disk again, we clear the XBF_DONE flag and
> > -	 * mark the buffer stale. This ensures that anyone who has a current
> > -	 * reference to the buffer will interpret it's contents correctly and
> > -	 * future cache lookups will also treat it as an empty, uninitialised
> > -	 * buffer.
> > -	 */
> > -	if (bp->b_error) {
> > -		error = bp->b_error;
> > -		if (!XFS_FORCED_SHUTDOWN(mp))
> > -			xfs_buf_ioerror_alert(bp, __func__);
> > -		bp->b_flags &= ~XBF_DONE;
> > -		xfs_buf_stale(bp);
> > -
> > +	error = xfs_buf_read_map(target, map, nmaps, flags, &bp, ops);
> > +	switch (error) {
> > +	case 0:
> > +		break;
> > +	case -EFSCORRUPTED:
> > +	case -EIO:
> >  		if (tp && (tp->t_flags & XFS_TRANS_DIRTY))
> > -			xfs_force_shutdown(tp->t_mountp, SHUTDOWN_META_IO_ERROR);
> > -		xfs_buf_relse(bp);
> > -
> > -		/* bad CRC means corrupted metadata */
> > -		if (error == -EFSBADCRC)
> > -			error = -EFSCORRUPTED;
> > +			xfs_force_shutdown(tp->t_mountp,
> > +					SHUTDOWN_META_IO_ERROR);
> > +		/* fall through */
> > +	default:
> >  		return error;
> >  	}
> 
> Same question as Christoph - we're only trying to avoid ENOMEM and
> EAGAIN errors from shutting down the filesystem here, right?
> Every other type of IO error that could end up on bp->b_error would
> result in a shutdown, so perhaps this should be the other way
> around:
> 
> 	switch (error) {
> 	case 0:
> 		break;
> 	default:
> 		/* shutdown stuff */
> 		/* fall through */
> 	case -ENOMEM:
> 	case -EAGAIN:
> 		return error;
> 	}

I agree that ENOMEM ought to be on the list of things that don't
immediately cause a shutdown if the transaction is dirty.

--D

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