Re: [PATCH 2/3] xfs: only end a COW operation in xfs_zero_remaining_bytes if we started one

["Darrick J. Wong" <darrick.wong@xxxxxxxxxx>]

On Sun, Jan 03, 2016 at 01:07:52PM +0100, Christoph Hellwig wrote:
> Without these we can see NULL pointer dereferences due to a non-existing
> COW fork during xfstests runs.
> 
> Signed-off-by: Christoph Hellwig <hch@xxxxxx>
> ---
>  fs/xfs/xfs_bmap_util.c | 12 +++++++-----
>  1 file changed, 7 insertions(+), 5 deletions(-)
> 
> diff --git a/fs/xfs/xfs_bmap_util.c b/fs/xfs/xfs_bmap_util.c
> index 7bee3c7..e777095 100644
> --- a/fs/xfs/xfs_bmap_util.c
> +++ b/fs/xfs/xfs_bmap_util.c
> @@ -1226,13 +1226,15 @@ xfs_zero_remaining_bytes(
>  
>               error = xfs_bwrite(bp);
>               xfs_buf_relse(bp);
> -             if (error) {
> -                     err2 = xfs_reflink_end_cow_failed(ip, offset,
> +             if (should_fork) {
> +                     if (error) {
> +                             err2 = xfs_reflink_end_cow_failed(ip, offset,
> +                                             lastoffset - offset + 1);
> +                             return error;
> +                     }
> +                     error = xfs_reflink_end_cow(ip, offset,
>                                       lastoffset - offset + 1);
> -                     return error;
>               }

Good catch!  There's a buffer leak in the error case for xfs_map_cow_blocks()
so I'll fold them both into the original patch.

--D

> -             error = xfs_reflink_end_cow(ip, offset,
> -                             lastoffset - offset + 1);
>               if (error)
>                       return error;
>       }
> -- 
> 1.9.1
> 
> _______________________________________________
> xfs mailing list
> xfs@xxxxxxxxxxx
> http://oss.sgi.com/mailman/listinfo/xfs