Merge tag 'for-5.15/drivers-2021-08-30' of git://git.kernel.dk/linux-block

Pull block driver updates from Jens Axboe:
 "Sitting on top of the core block changes, here are the driver changes
  for the 5.15 merge window:

   - NVMe updates via Christoph:
       - suspend improvements for devices with an HMB (Keith Busch)
       - handle double completions more gacefull (Sagi Grimberg)
       - cleanup the selects for the nvme core code a bit (Sagi Grimberg)
       - don't update queue count when failing to set io queues (Ruozhu Li)
       - various nvmet connect fixes (Amit Engel)
       - cleanup lightnvm leftovers (Keith Busch, me)
       - small cleanups (Colin Ian King, Hou Pu)
       - add tracing for the Set Features command (Hou Pu)
       - CMB sysfs cleanups (Keith Busch)
       - add a mutex_destroy call (Keith Busch)

   - remove lightnvm subsystem. It's served its purpose and ultimately
     led to zoned nvme support, we no longer need it (Christoph)

   - revert floppy O_NDELAY fix (Denis)

   - nbd fixes (Hou, Pavel, Baokun)

   - nbd locking fixes (Tetsuo)

   - nbd device removal fixes (Christoph)

   - raid10 rcu warning fix (Xiao)

   - raid1 write behind fix (Guoqing)

   - rnbd fixes (Gioh, Md Haris)

   - misc fixes (Colin)"

* tag 'for-5.15/drivers-2021-08-30' of git://git.kernel.dk/linux-block: (42 commits)
  Revert "floppy: reintroduce O_NDELAY fix"
  raid1: ensure write behind bio has less than BIO_MAX_VECS sectors
  md/raid10: Remove unnecessary rcu_dereference in raid10_handle_discard
  nbd: remove nbd->destroy_complete
  nbd: only return usable devices from nbd_find_unused
  nbd: set nbd->index before releasing nbd_index_mutex
  nbd: prevent IDR lookups from finding partially initialized devices
  nbd: reset NBD to NULL when restarting in nbd_genl_connect
  nbd: add missing locking to the nbd_dev_add error path
  nvme: remove the unused NVME_NS_* enum
  nvme: remove nvm_ndev from ns
  nvme: Have NVME_FABRICS select NVME_CORE instead of transport drivers
  block: nbd: add sanity check for first_minor
  nvmet: check that host sqsize does not exceed ctrl MQES
  nvmet: avoid duplicate qid in connect cmd
  nvmet: pass back cntlid on successful completion
  nvme-rdma: don't update queue count when failing to set io queues
  nvme-tcp: don't update queue count when failing to set io queues
  nvme-tcp: pair send_mutex init with destroy
  nvme: allow user toggling hmb usage
  ...

2 files changed, 29 insertions, 4 deletions

diff --git a/drivers/md/raid1.c b/drivers/md/raid1.c
index 3c44c4bb40fc..19598bd38939 100644
--- a/drivers/md/raid1.c
+++ b/drivers/md/raid1.c
@@ -1329,6 +1329,7 @@ static void raid1_write_request(struct mddev *mddev, struct bio *bio,
 	struct raid1_plug_cb *plug = NULL;
 	int first_clone;
 	int max_sectors;
+	bool write_behind = false;
 
 	if (mddev_is_clustered(mddev) &&
 	     md_cluster_ops->area_resyncing(mddev, WRITE,
@@ -1381,6 +1382,15 @@ static void raid1_write_request(struct mddev *mddev, struct bio *bio,
 	max_sectors = r1_bio->sectors;
 	for (i = 0;  i < disks; i++) {
 		struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev);
+
+		/*
+		 * The write-behind io is only attempted on drives marked as
+		 * write-mostly, which means we could allocate write behind
+		 * bio later.
+		 */
+		if (rdev && test_bit(WriteMostly, &rdev->flags))
+			write_behind = true;
+
 		if (rdev && unlikely(test_bit(Blocked, &rdev->flags))) {
 			atomic_inc(&rdev->nr_pending);
 			blocked_rdev = rdev;
@@ -1454,6 +1464,15 @@ static void raid1_write_request(struct mddev *mddev, struct bio *bio,
 		goto retry_write;
 	}
 
+	/*
+	 * When using a bitmap, we may call alloc_behind_master_bio below.
+	 * alloc_behind_master_bio allocates a copy of the data payload a page
+	 * at a time and thus needs a new bio that can fit the whole payload
+	 * this bio in page sized chunks.
+	 */
+	if (write_behind && bitmap)
+		max_sectors = min_t(int, max_sectors,
+				    BIO_MAX_VECS * (PAGE_SIZE >> 9));
 	if (max_sectors < bio_sectors(bio)) {
 		struct bio *split = bio_split(bio, max_sectors,
 					      GFP_NOIO, &conf->bio_split);
diff --git a/drivers/md/raid10.c b/drivers/md/raid10.c
index 07119d7e0fdf..aa2636582841 100644
--- a/drivers/md/raid10.c
+++ b/drivers/md/raid10.c
@@ -1712,6 +1712,11 @@ retry_discard:
 	} else
 		r10_bio->master_bio = (struct bio *)first_r10bio;
 
+	/*
+	 * first select target devices under rcu_lock and
+	 * inc refcount on their rdev.  Record them by setting
+	 * bios[x] to bio
+	 */
 	rcu_read_lock();
 	for (disk = 0; disk < geo->raid_disks; disk++) {
 		struct md_rdev *rdev = rcu_dereference(conf->mirrors[disk].rdev);
@@ -1743,9 +1748,6 @@ retry_discard:
 	for (disk = 0; disk < geo->raid_disks; disk++) {
 		sector_t dev_start, dev_end;
 		struct bio *mbio, *rbio = NULL;
-		struct md_rdev *rdev = rcu_dereference(conf->mirrors[disk].rdev);
-		struct md_rdev *rrdev = rcu_dereference(
-			conf->mirrors[disk].replacement);
 
 		/*
 		 * Now start to calculate the start and end address for each disk.
@@ -1775,9 +1777,12 @@ retry_discard:
 
 		/*
 		 * It only handles discard bio which size is >= stripe size, so
-		 * dev_end > dev_start all the time
+		 * dev_end > dev_start all the time.
+		 * It doesn't need to use rcu lock to get rdev here. We already
+		 * add rdev->nr_pending in the first loop.
 		 */
 		if (r10_bio->devs[disk].bio) {
+			struct md_rdev *rdev = conf->mirrors[disk].rdev;
 			mbio = bio_clone_fast(bio, GFP_NOIO, &mddev->bio_set);
 			mbio->bi_end_io = raid10_end_discard_request;
 			mbio->bi_private = r10_bio;
@@ -1790,6 +1795,7 @@ retry_discard:
 			bio_endio(mbio);
 		}
 		if (r10_bio->devs[disk].repl_bio) {
+			struct md_rdev *rrdev = conf->mirrors[disk].replacement;
 			rbio = bio_clone_fast(bio, GFP_NOIO, &mddev->bio_set);
 			rbio->bi_end_io = raid10_end_discard_request;
 			rbio->bi_private = r10_bio;