Re: [PATCH] iomap: Fix direct I/O write consistency check

From: Eric Sandeen <sandeen@sandeen.net>
To: Andreas Gruenbacher <agruenba@redhat.com>,
	Christoph Hellwig <hch@infradead.org>,
	"Darrick J. Wong" <darrick.wong@oracle.com>
Cc: linux-xfs@vger.kernel.org, linux-fsdevel@vger.kernel.org,
	linux-kernel@vger.kernel.org, cluster-devel@redhat.com,
	linux-ext4@vger.kernel.org
Subject: Re: [PATCH] iomap: Fix direct I/O write consistency check
Date: Thu, 3 Sep 2020 16:12:33 -0500	[thread overview]
Message-ID: <695a418c-ba6d-d3e9-f521-7dfa059764db@sandeen.net> (raw)
In-Reply-To: <20200903165632.1338996-1-agruenba@redhat.com>

On 9/3/20 11:56 AM, Andreas Gruenbacher wrote:
> When a direct I/O write falls back to buffered I/O entirely, dio->size
> will be 0 in iomap_dio_complete.  Function invalidate_inode_pages2_range
> will try to invalidate the rest of the address space.

(Because if ->size == 0 and offset == 0, then invalidating up to (0+0-1) will
invalidate the entire range.)

                err = invalidate_inode_pages2_range(inode->i_mapping,
                                offset >> PAGE_SHIFT,
                                (offset + dio->size - 1) >> PAGE_SHIFT);

so I guess this behavior is unique to writing to a hole at offset 0?

FWIW, this same test yields the same results on ext3 when it falls back to
buffered.

-Eric

> If there are any
> dirty pages in that range, the write will fail and a "Page cache
> invalidation failure on direct I/O" error will be logged.
> 
> On gfs2, this can be reproduced as follows:
> 
>   xfs_io \
>     -c "open -ft foo" -c "pwrite 4k 4k" -c "close" \
>     -c "open -d foo" -c "pwrite 0 4k"
> 
> Fix this by recognizing 0-length writes.
> 
> Signed-off-by: Andreas Gruenbacher <agruenba@redhat.com>
> ---
>  fs/iomap/direct-io.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/fs/iomap/direct-io.c b/fs/iomap/direct-io.c
> index c1aafb2ab990..c9d6b4eecdb7 100644
> --- a/fs/iomap/direct-io.c
> +++ b/fs/iomap/direct-io.c
> @@ -108,7 +108,7 @@ static ssize_t iomap_dio_complete(struct iomap_dio *dio)
>  	 * ->end_io() when necessary, otherwise a racing buffer read would cache
>  	 * zeros from unwritten extents.
>  	 */
> -	if (!dio->error &&
> +	if (!dio->error && dio->size &&
>  	    (dio->flags & IOMAP_DIO_WRITE) && inode->i_mapping->nrpages) {
>  		int err;
>  		err = invalidate_inode_pages2_range(inode->i_mapping,
> 