O_DIRECT: fix the splitting up of contiguous I/O
authorJeff Moyer <jmoyer@redhat.com>
Thu, 9 Sep 2010 23:37:33 +0000 (16:37 -0700)
committerLinus Torvalds <torvalds@linux-foundation.org>
Fri, 10 Sep 2010 01:57:22 +0000 (18:57 -0700)
commit c2c6ca4 (direct-io: do not merge logically non-contiguous requests)
introduced a bug whereby all O_DIRECT I/Os were submitted a page at a time
to the block layer.  The problem is that the code expected
dio->block_in_file to correspond to the current page in the dio.  In fact,
it corresponds to the previous page submitted via submit_page_section.
This was purely an oversight, as the dio->cur_page_fs_offset field was
introduced for just this purpose.  This patch simply uses the correct
variable when calculating whether there is a mismatch between contiguous
logical blocks and contiguous physical blocks (as described in the
comments).

I also switched the if conditional following this check to an else if, to
ensure that we never call dio_bio_submit twice for the same dio (in
theory, this should not happen, anyway).

I've tested this by running blktrace and verifying that a 64KB I/O was
submitted as a single I/O.  I also ran the patched kernel through
xfstests' aio tests using xfs, ext4 (with 1k and 4k block sizes) and btrfs
and verified that there were no regressions as compared to an unpatched
kernel.

Signed-off-by: Jeff Moyer <jmoyer@redhat.com>
Acked-by: Josef Bacik <jbacik@redhat.com>
Cc: Christoph Hellwig <hch@infradead.org>
Cc: Chris Mason <chris.mason@oracle.com>
Cc: <stable@kernel.org> [2.6.35.x]
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
fs/direct-io.c

index 51f270b479b6938a4a730ea56f9011c30f99563f..48d74c7391d13f4f07393c45d19825e937ecbcd1 100644 (file)
@@ -634,7 +634,7 @@ static int dio_send_cur_page(struct dio *dio)
        int ret = 0;
 
        if (dio->bio) {
-               loff_t cur_offset = dio->block_in_file << dio->blkbits;
+               loff_t cur_offset = dio->cur_page_fs_offset;
                loff_t bio_next_offset = dio->logical_offset_in_bio +
                        dio->bio->bi_size;
 
@@ -659,7 +659,7 @@ static int dio_send_cur_page(struct dio *dio)
                 * Submit now if the underlying fs is about to perform a
                 * metadata read
                 */
-               if (dio->boundary)
+               else if (dio->boundary)
                        dio_bio_submit(dio);
        }