dm: update dm_process_bio() to split bio if in ->make_request_fn()

Must call blk_queue_split() otherwise queue_limits for abnormal requests
(e.g. discard, writesame, etc) won't be imposed.

In addition, add dm_queue_split() to simplify DM specific splitting that
is needed for targets that impose ti->max_io_len.

Signed-off-by: Mike Snitzer <snitzer@redhat.com>

diff --git a/drivers/md/dm.c b/drivers/md/dm.c
index 515e6af9bed2c7ab5e94e356215020422316a3ce..7a774fcd01948f62910df3cbae46c0d2ff9e6b1d 100644 (file)
--- a/drivers/md/dm.c
+++ b/drivers/md/dm.c
@@ -1533,6 +1533,22 @@ static int __send_write_zeroes(struct clone_info *ci, struct dm_target *ti)
        return __send_changing_extent_only(ci, ti, get_num_write_zeroes_bios(ti), false);
 }
 
+static bool is_abnormal_io(struct bio *bio)
+{
+       bool r = false;
+
+       switch (bio_op(bio)) {
+       case REQ_OP_DISCARD:
+       case REQ_OP_SECURE_ERASE:
+       case REQ_OP_WRITE_SAME:
+       case REQ_OP_WRITE_ZEROES:
+               r = true;
+               break;
+       }
+
+       return r;
+}
+
 static bool __process_abnormal_io(struct clone_info *ci, struct dm_target *ti,
                                  int *result)
 {
@@ -1565,7 +1581,7 @@ static int __split_and_process_non_flush(struct clone_info *ci)
        if (!dm_target_is_valid(ti))
                return -EIO;
 
-       if (unlikely(__process_abnormal_io(ci, ti, &r)))
+       if (__process_abnormal_io(ci, ti, &r))
                return r;
 
        len = min_t(sector_t, max_io_len(ci->sector, ti), ci->sector_count);
@@ -1601,13 +1617,6 @@ static blk_qc_t __split_and_process_bio(struct mapped_device *md,
        blk_qc_t ret = BLK_QC_T_NONE;
        int error = 0;
 
-       if (unlikely(!map)) {
-               bio_io_error(bio);
-               return ret;
-       }
-
-       blk_queue_split(md->queue, &bio);
-
        init_clone_info(&ci, md, map, bio);
 
        if (bio->bi_opf & REQ_PREFLUSH) {
@@ -1675,18 +1684,13 @@ static blk_qc_t __split_and_process_bio(struct mapped_device *md,
  * Optimized variant of __split_and_process_bio that leverages the
  * fact that targets that use it do _not_ have a need to split bios.
  */
-static blk_qc_t __process_bio(struct mapped_device *md,
-                             struct dm_table *map, struct bio *bio)
+static blk_qc_t __process_bio(struct mapped_device *md, struct dm_table *map,
+                             struct bio *bio, struct dm_target *ti)
 {
        struct clone_info ci;
        blk_qc_t ret = BLK_QC_T_NONE;
        int error = 0;
 
-       if (unlikely(!map)) {
-               bio_io_error(bio);
-               return ret;
-       }
-
        init_clone_info(&ci, md, map, bio);
 
        if (bio->bi_opf & REQ_PREFLUSH) {
@@ -1704,21 +1708,11 @@ static blk_qc_t __process_bio(struct mapped_device *md,
                error = __send_empty_flush(&ci);
                /* dec_pending submits any data associated with flush */
        } else {
-               struct dm_target *ti = md->immutable_target;
                struct dm_target_io *tio;
 
-               /*
-                * Defend against IO still getting in during teardown
-                * - as was seen for a time with nvme-fcloop
-                */
-               if (WARN_ON_ONCE(!ti || !dm_target_is_valid(ti))) {
-                       error = -EIO;
-                       goto out;
-               }
-
                ci.bio = bio;
                ci.sector_count = bio_sectors(bio);
-               if (unlikely(__process_abnormal_io(&ci, ti, &error)))
+               if (__process_abnormal_io(&ci, ti, &error))
                        goto out;
 
                tio = alloc_tio(&ci, ti, 0, GFP_NOIO);
@@ -1730,11 +1724,56 @@ static blk_qc_t __process_bio(struct mapped_device *md,
        return ret;
 }
 
+static void dm_queue_split(struct mapped_device *md, struct dm_target *ti, struct bio **bio)
+{
+       unsigned len, sector_count;
+
+       sector_count = bio_sectors(*bio);
+       len = min_t(sector_t, max_io_len((*bio)->bi_iter.bi_sector, ti), sector_count);
+
+       if (sector_count > len) {
+               struct bio *split = bio_split(*bio, len, GFP_NOIO, &md->queue->bio_split);
+
+               bio_chain(split, *bio);
+               trace_block_split(md->queue, split, (*bio)->bi_iter.bi_sector);
+               generic_make_request(*bio);
+               *bio = split;
+       }
+}
+
 static blk_qc_t dm_process_bio(struct mapped_device *md,
                               struct dm_table *map, struct bio *bio)
 {
+       blk_qc_t ret = BLK_QC_T_NONE;
+       struct dm_target *ti = md->immutable_target;
+
+       if (unlikely(!map)) {
+               bio_io_error(bio);
+               return ret;
+       }
+
+       if (!ti) {
+               ti = dm_table_find_target(map, bio->bi_iter.bi_sector);
+               if (unlikely(!ti || !dm_target_is_valid(ti))) {
+                       bio_io_error(bio);
+                       return ret;
+               }
+       }
+
+       /*
+        * If in ->make_request_fn we need to use blk_queue_split(), otherwise
+        * queue_limits for abnormal requests (e.g. discard, writesame, etc)
+        * won't be imposed.
+        */
+       if (current->bio_list) {
+               if (is_abnormal_io(bio))
+                       blk_queue_split(md->queue, &bio);
+               else
+                       dm_queue_split(md, ti, &bio);
+       }
+
        if (dm_get_md_type(md) == DM_TYPE_NVME_BIO_BASED)
-               return __process_bio(md, map, bio);
+               return __process_bio(md, map, bio, ti);
        else
                return __split_and_process_bio(md, map, bio);
 }