dm thin: split discards on block boundary
authorMikulas Patocka <mpatocka@redhat.com>
Fri, 27 Jul 2012 14:08:03 +0000 (15:08 +0100)
committerAlasdair G Kergon <agk@redhat.com>
Fri, 27 Jul 2012 14:08:03 +0000 (15:08 +0100)
This patch sets the variable "ti->split_discard_requests" for the dm thin
target so that device mapper core splits discard requests on a block
boundary.

Consequently, a discard request that spans multiple blocks is never sent
to dm-thin. The patch also removes some code in process_discard that
deals with discards that span multiple blocks.

Signed-off-by: Mikulas Patocka <mpatocka@redhat.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>
Signed-off-by: Alasdair G Kergon <agk@redhat.com>
drivers/md/dm-thin.c

index f21d318d98f0041befc095f8aecf62fcb6977507..828649256902f5745718b542f55761648bf190ac 100644 (file)
@@ -1238,15 +1238,10 @@ static void process_discard(struct thin_c *tc, struct bio *bio)
                        }
                } else {
                        /*
-                        * This path is hit if people are ignoring
-                        * limits->discard_granularity.  It ignores any
-                        * part of the discard that is in a subsequent
-                        * block.
+                        * The DM core makes sure that the discard doesn't span
+                        * a block boundary.  So we submit the discard of a
+                        * partial block appropriately.
                         */
-                       sector_t offset = bio->bi_sector - (block * pool->sectors_per_block);
-                       unsigned remaining = (pool->sectors_per_block - offset) << SECTOR_SHIFT;
-                       bio->bi_size = min(bio->bi_size, remaining);
-
                        cell_release_singleton(cell, bio);
                        cell_release_singleton(cell2, bio);
                        if ((!lookup_result.shared) && pool->pf.discard_passdown)
@@ -2509,7 +2504,8 @@ static void set_discard_limits(struct pool *pool, struct queue_limits *limits)
 
        /*
         * This is just a hint, and not enforced.  We have to cope with
-        * bios that overlap 2 blocks.
+        * bios that cover a block partially.  A discard that spans a block
+        * boundary is not sent to this target.
         */
        limits->discard_granularity = pool->sectors_per_block << SECTOR_SHIFT;
        limits->discard_zeroes_data = pool->pf.zero_new_blocks;
@@ -2652,6 +2648,8 @@ static int thin_ctr(struct dm_target *ti, unsigned argc, char **argv)
                ti->discards_supported = 1;
                ti->num_discard_requests = 1;
                ti->discard_zeroes_data_unsupported = 1;
+               /* Discard requests must be split on a block boundary */
+               ti->split_discard_requests = 1;
        }
 
        dm_put(pool_md);