bcache: use (REQ_META|REQ_PRIO) to indicate bio for metadata

commit dc7292a5bcb4c878b076fca2ac3fc22f81b8f8df upstream.

In 'commit 752f66a75aba ("bcache: use REQ_PRIO to indicate bio for
metadata")' REQ_META is replaced by REQ_PRIO to indicate metadata bio.
This assumption is not always correct, e.g. XFS uses REQ_META to mark
metadata bio other than REQ_PRIO. This is why Nix noticed that bcache
does not cache metadata for XFS after the above commit.

Thanks to Dave Chinner, he explains the difference between REQ_META and
REQ_PRIO from view of file system developer. Here I quote part of his
explanation from mailing list,
   REQ_META is used for metadata. REQ_PRIO is used to communicate to
   the lower layers that the submitter considers this IO to be more
   important that non REQ_PRIO IO and so dispatch should be expedited.

   IOWs, if the filesystem considers metadata IO to be more important
   that user data IO, then it will use REQ_PRIO | REQ_META rather than
   just REQ_META.

Then it seems bios with REQ_META or REQ_PRIO should both be cached for
performance optimation, because they are all probably low I/O latency
demand by upper layer (e.g. file system).

So in this patch, when we want to decide whether to bypass the cache,
REQ_META and REQ_PRIO are both checked. Then both metadata and
high priority I/O requests will be handled properly.

Reported-by: Nix <nix@esperi.org.uk>
Signed-off-by: Coly Li <colyli@suse.de>
Reviewed-by: Andre Noll <maan@tuebingen.mpg.de>
Tested-by: Nix <nix@esperi.org.uk>
Cc: stable@vger.kernel.org
Cc: Dave Chinner <david@fromorbit.com>
Cc: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>

diff --git a/drivers/md/bcache/request.c b/drivers/md/bcache/request.c
index 22944aa..4ca3e3d 100644 (file)
--- a/drivers/md/bcache/request.c
+++ b/drivers/md/bcache/request.c
@@ -392,10 +392,11 @@ static bool check_should_bypass(struct cached_dev *dc, struct bio *bio)
 
        /*
         * Flag for bypass if the IO is for read-ahead or background,
-        * unless the read-ahead request is for metadata (eg, for gfs2).
+        * unless the read-ahead request is for metadata
+        * (eg, for gfs2 or xfs).
         */
        if (bio->bi_opf & (REQ_RAHEAD|REQ_BACKGROUND) &&
-           !(bio->bi_opf & REQ_META))
+           !(bio->bi_opf & (REQ_META|REQ_PRIO)))
                goto skip;
 
        if (bio->bi_iter.bi_sector & (c->sb.block_size - 1) ||
@@ -877,7 +878,7 @@ static int cached_dev_cache_miss(struct btree *b, struct search *s,
        }
 
        if (!(bio->bi_opf & REQ_RAHEAD) &&
-           !(bio->bi_opf & REQ_META) &&
+           !(bio->bi_opf & (REQ_META|REQ_PRIO)) &&
            s->iop.c->gc_stats.in_use < CUTOFF_CACHE_READA)
                reada = min_t(sector_t, dc->readahead >> 9,
                              get_capacity(bio->bi_disk) - bio_end_sector(bio));