Merge tag 'md/4.1' of git://neil.brown.name/md

Pull md updates from Neil Brown:
"More updates that usual this time. A few have performance impacts
which hould mostly be positive, but RAID5 (in particular) can be very
work-load ensitive... We'll have to wait and see.

Highlights:

- "experimental" code for managing md/raid1 across a cluster using
DLM. Code is not ready for general use and triggers a WARNING if
used. However it is looking good and mostly done and having in
mainline will help co-ordinate development.

- RAID5/6 can now batch multiple (4K wide) stripe_heads so as to
handle a full (chunk wide) stripe as a single unit.

- RAID6 can now perform read-modify-write cycles which should help
performance on larger arrays: 6 or more devices.

- RAID5/6 stripe cache now grows and shrinks dynamically. The value
set is used as a minimum.

- Resync is now allowed to go a little faster than the 'mininum' when
there is competing IO. How much faster depends on the speed of the
devices, so the effective minimum should scale with device speed to
some extent"

* tag 'md/4.1' of git://neil.brown.name/md: (58 commits)
md/raid5: don't do chunk aligned read on degraded array.
md/raid5: allow the stripe_cache to grow and shrink.
md/raid5: change ->inactive_blocked to a bit-flag.
md/raid5: move max_nr_stripes management into grow_one_stripe and drop_one_stripe
md/raid5: pass gfp_t arg to grow_one_stripe()
md/raid5: introduce configuration option rmw_level
md/raid5: activate raid6 rmw feature
md/raid6 algorithms: xor_syndrome() for SSE2
md/raid6 algorithms: xor_syndrome() for generic int
md/raid6 algorithms: improve test program
md/raid6 algorithms: delta syndrome functions
raid5: handle expansion/resync case with stripe batching
raid5: handle io error of batch list
RAID5: batch adjacent full stripe write
raid5: track overwrite disk count
raid5: add a new flag to track if a stripe can be batched
raid5: use flex_array for scribble data
md raid0: access mddev->queue (request queue member) conditionally because it is not set when accessed from dm-raid
md: allow resync to go faster when there is competing IO.
md: remove 'go_faster' option from ->sync_request()
...

Documentation/md-cluster.txt

···
+The cluster MD is a shared-device RAID for a cluster.
+
+
+1. On-disk format
+
+Separate write-intent-bitmap are used for each cluster node.
+The bitmaps record all writes that may have been started on that node,
+and may not yet have finished. The on-disk layout is:
+
+0                    4k                     8k                    12k
+-------------------------------------------------------------------
+| idle                | md super            | bm super [0] + bits |
+| bm bits[0, contd]   | bm super[1] + bits  | bm bits[1, contd]   |
+| bm super[2] + bits  | bm bits [2, contd]  | bm super[3] + bits  |
+| bm bits [3, contd]  |                     |                     |
+
+During "normal" functioning we assume the filesystem ensures that only one
+node writes to any given block at a time, so a write
+request will
+ - set the appropriate bit (if not already set)
+ - commit the write to all mirrors
+ - schedule the bit to be cleared after a timeout.
+
+Reads are just handled normally.  It is up to the filesystem to
+ensure one node doesn't read from a location where another node (or the same
+node) is writing.
+
+
+2. DLM Locks for management
+
+There are two locks for managing the device:
+
+2.1 Bitmap lock resource (bm_lockres)
+
+ The bm_lockres protects individual node bitmaps. They are named in the
+ form bitmap001 for node 1, bitmap002 for node and so on. When a node
+ joins the cluster, it acquires the lock in PW mode and it stays so
+ during the lifetime the node is part of the cluster. The lock resource
+ number is based on the slot number returned by the DLM subsystem. Since
+ DLM starts node count from one and bitmap slots start from zero, one is
+ subtracted from the DLM slot number to arrive at the bitmap slot number.
+
+3. Communication
+
+Each node has to communicate with other nodes when starting or ending
+resync, and metadata superblock updates.
+
+3.1 Message Types
+
+ There are 3 types, of messages which are passed
+
+ 3.1.1 METADATA_UPDATED: informs other nodes that the metadata has been
+   updated, and the node must re-read the md superblock. This is performed
+   synchronously.
+
+ 3.1.2 RESYNC: informs other nodes that a resync is initiated or ended
+   so that each node may suspend or resume the region.
+
+3.2 Communication mechanism
+
+ The DLM LVB is used to communicate within nodes of the cluster. There
+ are three resources used for the purpose:
+
+  3.2.1 Token: The resource which protects the entire communication
+   system. The node having the token resource is allowed to
+   communicate.
+
+  3.2.2 Message: The lock resource which carries the data to
+   communicate.
+
+  3.2.3 Ack: The resource, acquiring which means the message has been
+   acknowledged by all nodes in the cluster. The BAST of the resource
+   is used to inform the receive node that a node wants to communicate.
+
+The algorithm is:
+
+ 1. receive status
+
+   sender                         receiver                   receiver
+   ACK:CR                          ACK:CR                     ACK:CR
+
+ 2. sender get EX of TOKEN
+    sender get EX of MESSAGE
+    sender                        receiver                 receiver
+    TOKEN:EX                       ACK:CR                   ACK:CR
+    MESSAGE:EX
+    ACK:CR
+
+    Sender checks that it still needs to send a message. Messages received
+    or other events that happened while waiting for the TOKEN may have made
+    this message inappropriate or redundant.
+
+ 3. sender write LVB.
+    sender down-convert MESSAGE from EX to CR
+    sender try to get EX of ACK
+    [ wait until all receiver has *processed* the MESSAGE ]
+
+                                     [ triggered by bast of ACK ]
+                                     receiver get CR of MESSAGE
+                                     receiver read LVB
+                                     receiver processes the message
+                                     [ wait finish ]
+                                     receiver release ACK
+
+   sender                         receiver                   receiver
+   TOKEN:EX                       MESSAGE:CR                 MESSAGE:CR
+   MESSAGE:CR
+   ACK:EX
+
+ 4. triggered by grant of EX on ACK (indicating all receivers have processed
+    message)
+    sender down-convert ACK from EX to CR
+    sender release MESSAGE
+    sender release TOKEN
+                               receiver upconvert to EX of MESSAGE
+                               receiver get CR of ACK
+                               receiver release MESSAGE
+
+   sender                      receiver                   receiver
+   ACK:CR                       ACK:CR                     ACK:CR
+
+
+4. Handling Failures
+
+4.1 Node Failure
+ When a node fails, the DLM informs the cluster with the slot. The node
+ starts a cluster recovery thread. The cluster recovery thread:
+	- acquires the bitmap<number> lock of the failed node
+	- opens the bitmap
+	- reads the bitmap of the failed node
+	- copies the set bitmap to local node
+	- cleans the bitmap of the failed node
+	- releases bitmap<number> lock of the failed node
+	- initiates resync of the bitmap on the current node
+
+ The resync process, is the regular md resync. However, in a clustered
+ environment when a resync is performed, it needs to tell other nodes
+ of the areas which are suspended. Before a resync starts, the node
+ send out RESYNC_START with the (lo,hi) range of the area which needs
+ to be suspended. Each node maintains a suspend_list, which contains
+ the list  of ranges which are currently suspended. On receiving
+ RESYNC_START, the node adds the range to the suspend_list. Similarly,
+ when the node performing resync finishes, it send RESYNC_FINISHED
+ to other nodes and other nodes remove the corresponding entry from
+ the suspend_list.
+
+ A helper function, should_suspend() can be used to check if a particular
+ I/O range should be suspended or not.
+
+4.2 Device Failure
+ Device failures are handled and communicated with the metadata update
+ routine.
+
+5. Adding a new Device
+For adding a new device, it is necessary that all nodes "see" the new device
+to be added. For this, the following algorithm is used:
+
+    1. Node 1 issues mdadm --manage /dev/mdX --add /dev/sdYY which issues
+       ioctl(ADD_NEW_DISC with disc.state set to MD_DISK_CLUSTER_ADD)
+    2. Node 1 sends NEWDISK with uuid and slot number
+    3. Other nodes issue kobject_uevent_env with uuid and slot number
+       (Steps 4,5 could be a udev rule)
+    4. In userspace, the node searches for the disk, perhaps
+       using blkid -t SUB_UUID=""
+    5. Other nodes issue either of the following depending on whether the disk
+       was found:
+       ioctl(ADD_NEW_DISK with disc.state set to MD_DISK_CANDIDATE and
+                disc.number set to slot number)
+       ioctl(CLUSTERED_DISK_NACK)
+    6. Other nodes drop lock on no-new-devs (CR) if device is found
+    7. Node 1 attempts EX lock on no-new-devs
+    8. If node 1 gets the lock, it sends METADATA_UPDATED after unmarking the disk
+       as SpareLocal
+    9. If not (get no-new-dev lock), it fails the operation and sends METADATA_UPDATED
+    10. Other nodes get the information whether a disk is added or not
+	by the following METADATA_UPDATED.

crypto/async_tx/async_pq.c

···
 {
 	void **srcs;
 	int i;
+	int start = -1, stop = disks - 3;
 
 	if (submit->scribble)
 		srcs = submit->scribble;
···
 		if (blocks[i] == NULL) {
 			BUG_ON(i > disks - 3); /* P or Q can't be zero */
 			srcs[i] = (void*)raid6_empty_zero_page;
-		} else
+		} else {
 			srcs[i] = page_address(blocks[i]) + offset;
+			if (i < disks - 2) {
+				stop = i;
+				if (start == -1)
+					start = i;
+			}
+		}
 	}
-	raid6_call.gen_syndrome(disks, len, srcs);
+	if (submit->flags & ASYNC_TX_PQ_XOR_DST) {
+		BUG_ON(!raid6_call.xor_syndrome);
+		if (start >= 0)
+			raid6_call.xor_syndrome(disks, start, stop, len, srcs);
+	} else
+		raid6_call.gen_syndrome(disks, len, srcs);
 	async_tx_sync_epilog(submit);
 }
 
···
 	if (device)
 		unmap = dmaengine_get_unmap_data(device->dev, disks, GFP_NOIO);
 
-	if (unmap &&
+	/* XORing P/Q is only implemented in software */
+	if (unmap && !(submit->flags & ASYNC_TX_PQ_XOR_DST) &&
 	    (src_cnt <= dma_maxpq(device, 0) ||
 	     dma_maxpq(device, DMA_PREP_CONTINUE) > 0) &&
 	    is_dma_pq_aligned(device, offset, 0, len)) {

drivers/md/Kconfig

···
 
 	  In unsure, say N.
 
+
+config MD_CLUSTER
+	tristate "Cluster Support for MD (EXPERIMENTAL)"
+	depends on BLK_DEV_MD
+	depends on DLM
+	default n
+	---help---
+	Clustering support for MD devices. This enables locking and
+	synchronization across multiple systems on the cluster, so all
+	nodes in the cluster can access the MD devices simultaneously.
+
+	This brings the redundancy (and uptime) of RAID levels across the
+	nodes of the cluster.
+
+	If unsure, say N.
+
 source "drivers/md/bcache/Kconfig"
 
 config BLK_DEV_DM_BUILTIN

drivers/md/Makefile

···
 obj-$(CONFIG_MD_RAID456)	+= raid456.o
 obj-$(CONFIG_MD_MULTIPATH)	+= multipath.o
 obj-$(CONFIG_MD_FAULTY)		+= faulty.o
+obj-$(CONFIG_MD_CLUSTER)	+= md-cluster.o
 obj-$(CONFIG_BCACHE)		+= bcache/
 obj-$(CONFIG_BLK_DEV_MD)	+= md-mod.o
 obj-$(CONFIG_BLK_DEV_DM)	+= dm-mod.o

drivers/md/bitmap.c

···
 	struct block_device *bdev;
 	struct mddev *mddev = bitmap->mddev;
 	struct bitmap_storage *store = &bitmap->storage;
+	int node_offset = 0;
+
+	if (mddev_is_clustered(bitmap->mddev))
+		node_offset = bitmap->cluster_slot * store->file_pages;
 
 	while ((rdev = next_active_rdev(rdev, mddev)) != NULL) {
 		int size = PAGE_SIZE;
···
 	/* This might have been changed by a reshape */
 	sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors);
 	sb->chunksize = cpu_to_le32(bitmap->mddev->bitmap_info.chunksize);
+	sb->nodes = cpu_to_le32(bitmap->mddev->bitmap_info.nodes);
 	sb->sectors_reserved = cpu_to_le32(bitmap->mddev->
 					   bitmap_info.space);
 	kunmap_atomic(sb);
···
 	bitmap_super_t *sb;
 	unsigned long chunksize, daemon_sleep, write_behind;
 	unsigned long long events;
+	int nodes = 0;
 	unsigned long sectors_reserved = 0;
 	int err = -EINVAL;
 	struct page *sb_page;
···
 		return -ENOMEM;
 	bitmap->storage.sb_page = sb_page;
 
+re_read:
+	/* If cluster_slot is set, the cluster is setup */
+	if (bitmap->cluster_slot >= 0) {
+		sector_t bm_blocks = bitmap->mddev->resync_max_sectors;
+
+		sector_div(bm_blocks,
+			   bitmap->mddev->bitmap_info.chunksize >> 9);
+		/* bits to bytes */
+		bm_blocks = ((bm_blocks+7) >> 3) + sizeof(bitmap_super_t);
+		/* to 4k blocks */
+		bm_blocks = DIV_ROUND_UP_SECTOR_T(bm_blocks, 4096);
+		bitmap->mddev->bitmap_info.offset += bitmap->cluster_slot * (bm_blocks << 3);
+		pr_info("%s:%d bm slot: %d offset: %llu\n", __func__, __LINE__,
+			bitmap->cluster_slot, (unsigned long long)bitmap->mddev->bitmap_info.offset);
+	}
+
 	if (bitmap->storage.file) {
 		loff_t isize = i_size_read(bitmap->storage.file->f_mapping->host);
 		int bytes = isize > PAGE_SIZE ? PAGE_SIZE : isize;
···
 	if (err)
 		return err;
 
+	err = -EINVAL;
 	sb = kmap_atomic(sb_page);
 
 	chunksize = le32_to_cpu(sb->chunksize);
 	daemon_sleep = le32_to_cpu(sb->daemon_sleep) * HZ;
 	write_behind = le32_to_cpu(sb->write_behind);
 	sectors_reserved = le32_to_cpu(sb->sectors_reserved);
+	nodes = le32_to_cpu(sb->nodes);
+	strlcpy(bitmap->mddev->bitmap_info.cluster_name, sb->cluster_name, 64);
 
 	/* verify that the bitmap-specific fields are valid */
 	if (sb->magic != cpu_to_le32(BITMAP_MAGIC))
···
 			goto out;
 		}
 		events = le64_to_cpu(sb->events);
-		if (events < bitmap->mddev->events) {
+		if (!nodes && (events < bitmap->mddev->events)) {
 			printk(KERN_INFO
 			       "%s: bitmap file is out of date (%llu < %llu) "
 			       "-- forcing full recovery\n",
···
 	if (le32_to_cpu(sb->version) == BITMAP_MAJOR_HOSTENDIAN)
 		set_bit(BITMAP_HOSTENDIAN, &bitmap->flags);
 	bitmap->events_cleared = le64_to_cpu(sb->events_cleared);
+	strlcpy(bitmap->mddev->bitmap_info.cluster_name, sb->cluster_name, 64);
 	err = 0;
+
 out:
 	kunmap_atomic(sb);
+	/* Assiging chunksize is required for "re_read" */
+	bitmap->mddev->bitmap_info.chunksize = chunksize;
+	if (nodes && (bitmap->cluster_slot < 0)) {
+		err = md_setup_cluster(bitmap->mddev, nodes);
+		if (err) {
+			pr_err("%s: Could not setup cluster service (%d)\n",
+					bmname(bitmap), err);
+			goto out_no_sb;
+		}
+		bitmap->cluster_slot = md_cluster_ops->slot_number(bitmap->mddev);
+		goto re_read;
+	}
+
+
 out_no_sb:
 	if (test_bit(BITMAP_STALE, &bitmap->flags))
 		bitmap->events_cleared = bitmap->mddev->events;
 	bitmap->mddev->bitmap_info.chunksize = chunksize;
 	bitmap->mddev->bitmap_info.daemon_sleep = daemon_sleep;
 	bitmap->mddev->bitmap_info.max_write_behind = write_behind;
+	bitmap->mddev->bitmap_info.nodes = nodes;
 	if (bitmap->mddev->bitmap_info.space == 0 ||
 	    bitmap->mddev->bitmap_info.space > sectors_reserved)
 		bitmap->mddev->bitmap_info.space = sectors_reserved;
-	if (err)
+	if (err) {
 		bitmap_print_sb(bitmap);
+		if (bitmap->cluster_slot < 0)
+			md_cluster_stop(bitmap->mddev);
+	}
 	return err;
 }
 
···
 }
 
 static int bitmap_storage_alloc(struct bitmap_storage *store,
-				unsigned long chunks, int with_super)
+				unsigned long chunks, int with_super,
+				int slot_number)
 {
-	int pnum;
+	int pnum, offset = 0;
 	unsigned long num_pages;
 	unsigned long bytes;
 
···
 		bytes += sizeof(bitmap_super_t);
 
 	num_pages = DIV_ROUND_UP(bytes, PAGE_SIZE);
+	offset = slot_number * (num_pages - 1);
 
 	store->filemap = kmalloc(sizeof(struct page *)
 				 * num_pages, GFP_KERNEL);
···
 		store->sb_page = alloc_page(GFP_KERNEL|__GFP_ZERO);
 		if (store->sb_page == NULL)
 			return -ENOMEM;
-		store->sb_page->index = 0;
 	}
+
 	pnum = 0;
 	if (store->sb_page) {
 		store->filemap[0] = store->sb_page;
 		pnum = 1;
+		store->sb_page->index = offset;
 	}
+
 	for ( ; pnum < num_pages; pnum++) {
 		store->filemap[pnum] = alloc_page(GFP_KERNEL|__GFP_ZERO);
 		if (!store->filemap[pnum]) {
 			store->file_pages = pnum;
 			return -ENOMEM;
 		}
-		store->filemap[pnum]->index = pnum;
+		store->filemap[pnum]->index = pnum + offset;
 	}
 	store->file_pages = pnum;
 
···
 	}
 }
 
+static int bitmap_file_test_bit(struct bitmap *bitmap, sector_t block)
+{
+	unsigned long bit;
+	struct page *page;
+	void *paddr;
+	unsigned long chunk = block >> bitmap->counts.chunkshift;
+	int set = 0;
+
+	page = filemap_get_page(&bitmap->storage, chunk);
+	if (!page)
+		return -EINVAL;
+	bit = file_page_offset(&bitmap->storage, chunk);
+	paddr = kmap_atomic(page);
+	if (test_bit(BITMAP_HOSTENDIAN, &bitmap->flags))
+		set = test_bit(bit, paddr);
+	else
+		set = test_bit_le(bit, paddr);
+	kunmap_atomic(paddr);
+	return set;
+}
+
+
 /* this gets called when the md device is ready to unplug its underlying
  * (slave) device queues -- before we let any writes go down, we need to
  * sync the dirty pages of the bitmap file to disk */
···
  */
 static int bitmap_init_from_disk(struct bitmap *bitmap, sector_t start)
 {
-	unsigned long i, chunks, index, oldindex, bit;
+	unsigned long i, chunks, index, oldindex, bit, node_offset = 0;
 	struct page *page = NULL;
 	unsigned long bit_cnt = 0;
 	struct file *file;
···
 	if (!bitmap->mddev->bitmap_info.external)
 		offset = sizeof(bitmap_super_t);
 
+	if (mddev_is_clustered(bitmap->mddev))
+		node_offset = bitmap->cluster_slot * (DIV_ROUND_UP(store->bytes, PAGE_SIZE));
+
 	for (i = 0; i < chunks; i++) {
 		int b;
 		index = file_page_index(&bitmap->storage, i);
···
 					bitmap->mddev,
 					bitmap->mddev->bitmap_info.offset,
 					page,
-					index, count);
+					index + node_offset, count);
 
 			if (ret)
 				goto err;
···
 	     j < bitmap->storage.file_pages
 		     && !test_bit(BITMAP_STALE, &bitmap->flags);
 	     j++) {
-
 		if (test_page_attr(bitmap, j,
 				   BITMAP_PAGE_DIRTY))
 			/* bitmap_unplug will handle the rest */
···
 		return;
 	}
 	if (!*bmc) {
-		*bmc = 2 | (needed ? NEEDED_MASK : 0);
+		*bmc = 2;
 		bitmap_count_page(&bitmap->counts, offset, 1);
 		bitmap_set_pending(&bitmap->counts, offset);
 		bitmap->allclean = 0;
 	}
+	if (needed)
+		*bmc |= NEEDED_MASK;
 	spin_unlock_irq(&bitmap->counts.lock);
 }
 
···
 	if (!bitmap) /* there was no bitmap */
 		return;
 
+	if (mddev_is_clustered(bitmap->mddev) && bitmap->mddev->cluster_info &&
+		bitmap->cluster_slot == md_cluster_ops->slot_number(bitmap->mddev))
+		md_cluster_stop(bitmap->mddev);
+
 	/* Shouldn't be needed - but just in case.... */
 	wait_event(bitmap->write_wait,
 		   atomic_read(&bitmap->pending_writes) == 0);
···
  * initialize the bitmap structure
  * if this returns an error, bitmap_destroy must be called to do clean up
  */
-int bitmap_create(struct mddev *mddev)
+struct bitmap *bitmap_create(struct mddev *mddev, int slot)
 {
 	struct bitmap *bitmap;
 	sector_t blocks = mddev->resync_max_sectors;
···
 
 	bitmap = kzalloc(sizeof(*bitmap), GFP_KERNEL);
 	if (!bitmap)
-		return -ENOMEM;
+		return ERR_PTR(-ENOMEM);
 
 	spin_lock_init(&bitmap->counts.lock);
 	atomic_set(&bitmap->pending_writes, 0);
···
 	init_waitqueue_head(&bitmap->behind_wait);
 
 	bitmap->mddev = mddev;
+	bitmap->cluster_slot = slot;
 
 	if (mddev->kobj.sd)
 		bm = sysfs_get_dirent(mddev->kobj.sd, "bitmap");
···
 	printk(KERN_INFO "created bitmap (%lu pages) for device %s\n",
 	       bitmap->counts.pages, bmname(bitmap));
 
-	mddev->bitmap = bitmap;
-	return test_bit(BITMAP_WRITE_ERROR, &bitmap->flags) ? -EIO : 0;
+	err = test_bit(BITMAP_WRITE_ERROR, &bitmap->flags) ? -EIO : 0;
+	if (err)
+		goto error;
 
+	return bitmap;
  error:
 	bitmap_free(bitmap);
-	return err;
+	return ERR_PTR(err);
 }
 
 int bitmap_load(struct mddev *mddev)
···
 	return err;
 }
 EXPORT_SYMBOL_GPL(bitmap_load);
+
+/* Loads the bitmap associated with slot and copies the resync information
+ * to our bitmap
+ */
+int bitmap_copy_from_slot(struct mddev *mddev, int slot,
+		sector_t *low, sector_t *high, bool clear_bits)
+{
+	int rv = 0, i, j;
+	sector_t block, lo = 0, hi = 0;
+	struct bitmap_counts *counts;
+	struct bitmap *bitmap = bitmap_create(mddev, slot);
+
+	if (IS_ERR(bitmap))
+		return PTR_ERR(bitmap);
+
+	rv = bitmap_read_sb(bitmap);
+	if (rv)
+		goto err;
+
+	rv = bitmap_init_from_disk(bitmap, 0);
+	if (rv)
+		goto err;
+
+	counts = &bitmap->counts;
+	for (j = 0; j < counts->chunks; j++) {
+		block = (sector_t)j << counts->chunkshift;
+		if (bitmap_file_test_bit(bitmap, block)) {
+			if (!lo)
+				lo = block;
+			hi = block;
+			bitmap_file_clear_bit(bitmap, block);
+			bitmap_set_memory_bits(mddev->bitmap, block, 1);
+			bitmap_file_set_bit(mddev->bitmap, block);
+		}
+	}
+
+	if (clear_bits) {
+		bitmap_update_sb(bitmap);
+		/* Setting this for the ev_page should be enough.
+		 * And we do not require both write_all and PAGE_DIRT either
+		 */
+		for (i = 0; i < bitmap->storage.file_pages; i++)
+			set_page_attr(bitmap, i, BITMAP_PAGE_DIRTY);
+		bitmap_write_all(bitmap);
+		bitmap_unplug(bitmap);
+	}
+	*low = lo;
+	*high = hi;
+err:
+	bitmap_free(bitmap);
+	return rv;
+}
+EXPORT_SYMBOL_GPL(bitmap_copy_from_slot);
+
 
 void bitmap_status(struct seq_file *seq, struct bitmap *bitmap)
 {
···
 	memset(&store, 0, sizeof(store));
 	if (bitmap->mddev->bitmap_info.offset || bitmap->mddev->bitmap_info.file)
 		ret = bitmap_storage_alloc(&store, chunks,
-					   !bitmap->mddev->bitmap_info.external);
+					   !bitmap->mddev->bitmap_info.external,
+					   bitmap->cluster_slot);
 	if (ret)
 		goto err;
 
···
 				return -EINVAL;
 			mddev->bitmap_info.offset = offset;
 			if (mddev->pers) {
+				struct bitmap *bitmap;
 				mddev->pers->quiesce(mddev, 1);
-				rv = bitmap_create(mddev);
-				if (!rv)
+				bitmap = bitmap_create(mddev, -1);
+				if (IS_ERR(bitmap))
+					rv = PTR_ERR(bitmap);
+				else {
+					mddev->bitmap = bitmap;
 					rv = bitmap_load(mddev);
-				if (rv) {
-					bitmap_destroy(mddev);
-					mddev->bitmap_info.offset = 0;
+					if (rv) {
+						bitmap_destroy(mddev);
+						mddev->bitmap_info.offset = 0;
+					}
 				}
 				mddev->pers->quiesce(mddev, 0);
 				if (rv)
···
 
 static ssize_t metadata_show(struct mddev *mddev, char *page)
 {
+	if (mddev_is_clustered(mddev))
+		return sprintf(page, "clustered\n");
 	return sprintf(page, "%s\n", (mddev->bitmap_info.external
 				      ? "external" : "internal"));
 }
···
 		return -EBUSY;
 	if (strncmp(buf, "external", 8) == 0)
 		mddev->bitmap_info.external = 1;
-	else if (strncmp(buf, "internal", 8) == 0)
+	else if ((strncmp(buf, "internal", 8) == 0) ||
+			(strncmp(buf, "clustered", 9) == 0))
 		mddev->bitmap_info.external = 0;
 	else
 		return -EINVAL;

drivers/md/bitmap.h

···
 	__le32 write_behind; /* 60  number of outstanding write-behind writes */
 	__le32 sectors_reserved; /* 64 number of 512-byte sectors that are
 				  * reserved for the bitmap. */
-
-	__u8  pad[256 - 68]; /* set to zero */
+	__le32 nodes;        /* 68 the maximum number of nodes in cluster. */
+	__u8 cluster_name[64]; /* 72 cluster name to which this md belongs */
+	__u8  pad[256 - 136]; /* set to zero */
 } bitmap_super_t;
 
 /* notes:
···
 	wait_queue_head_t behind_wait;
 
 	struct kernfs_node *sysfs_can_clear;
+	int cluster_slot;		/* Slot offset for clustered env */
 };
 
 /* the bitmap API */
 
 /* these are used only by md/bitmap */
-int  bitmap_create(struct mddev *mddev);
+struct bitmap *bitmap_create(struct mddev *mddev, int slot);
 int bitmap_load(struct mddev *mddev);
 void bitmap_flush(struct mddev *mddev);
 void bitmap_destroy(struct mddev *mddev);
···
 
 int bitmap_resize(struct bitmap *bitmap, sector_t blocks,
 		  int chunksize, int init);
+int bitmap_copy_from_slot(struct mddev *mddev, int slot,
+				sector_t *lo, sector_t *hi, bool clear_bits);
 #endif
 
 #endif

drivers/md/md-cluster.c

···
+/*
+ * Copyright (C) 2015, SUSE
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; either version 2, or (at your option)
+ * any later version.
+ *
+ */
+
+
+#include <linux/module.h>
+#include <linux/dlm.h>
+#include <linux/sched.h>
+#include <linux/raid/md_p.h>
+#include "md.h"
+#include "bitmap.h"
+#include "md-cluster.h"
+
+#define LVB_SIZE	64
+#define NEW_DEV_TIMEOUT 5000
+
+struct dlm_lock_resource {
+	dlm_lockspace_t *ls;
+	struct dlm_lksb lksb;
+	char *name; /* lock name. */
+	uint32_t flags; /* flags to pass to dlm_lock() */
+	struct completion completion; /* completion for synchronized locking */
+	void (*bast)(void *arg, int mode); /* blocking AST function pointer*/
+	struct mddev *mddev; /* pointing back to mddev. */
+};
+
+struct suspend_info {
+	int slot;
+	sector_t lo;
+	sector_t hi;
+	struct list_head list;
+};
+
+struct resync_info {
+	__le64 lo;
+	__le64 hi;
+};
+
+/* md_cluster_info flags */
+#define		MD_CLUSTER_WAITING_FOR_NEWDISK		1
+
+
+struct md_cluster_info {
+	/* dlm lock space and resources for clustered raid. */
+	dlm_lockspace_t *lockspace;
+	int slot_number;
+	struct completion completion;
+	struct dlm_lock_resource *sb_lock;
+	struct mutex sb_mutex;
+	struct dlm_lock_resource *bitmap_lockres;
+	struct list_head suspend_list;
+	spinlock_t suspend_lock;
+	struct md_thread *recovery_thread;
+	unsigned long recovery_map;
+	/* communication loc resources */
+	struct dlm_lock_resource *ack_lockres;
+	struct dlm_lock_resource *message_lockres;
+	struct dlm_lock_resource *token_lockres;
+	struct dlm_lock_resource *no_new_dev_lockres;
+	struct md_thread *recv_thread;
+	struct completion newdisk_completion;
+	unsigned long state;
+};
+
+enum msg_type {
+	METADATA_UPDATED = 0,
+	RESYNCING,
+	NEWDISK,
+	REMOVE,
+	RE_ADD,
+};
+
+struct cluster_msg {
+	int type;
+	int slot;
+	/* TODO: Unionize this for smaller footprint */
+	sector_t low;
+	sector_t high;
+	char uuid[16];
+	int raid_slot;
+};
+
+static void sync_ast(void *arg)
+{
+	struct dlm_lock_resource *res;
+
+	res = (struct dlm_lock_resource *) arg;
+	complete(&res->completion);
+}
+
+static int dlm_lock_sync(struct dlm_lock_resource *res, int mode)
+{
+	int ret = 0;
+
+	init_completion(&res->completion);
+	ret = dlm_lock(res->ls, mode, &res->lksb,
+			res->flags, res->name, strlen(res->name),
+			0, sync_ast, res, res->bast);
+	if (ret)
+		return ret;
+	wait_for_completion(&res->completion);
+	return res->lksb.sb_status;
+}
+
+static int dlm_unlock_sync(struct dlm_lock_resource *res)
+{
+	return dlm_lock_sync(res, DLM_LOCK_NL);
+}
+
+static struct dlm_lock_resource *lockres_init(struct mddev *mddev,
+		char *name, void (*bastfn)(void *arg, int mode), int with_lvb)
+{
+	struct dlm_lock_resource *res = NULL;
+	int ret, namelen;
+	struct md_cluster_info *cinfo = mddev->cluster_info;
+
+	res = kzalloc(sizeof(struct dlm_lock_resource), GFP_KERNEL);
+	if (!res)
+		return NULL;
+	res->ls = cinfo->lockspace;
+	res->mddev = mddev;
+	namelen = strlen(name);
+	res->name = kzalloc(namelen + 1, GFP_KERNEL);
+	if (!res->name) {
+		pr_err("md-cluster: Unable to allocate resource name for resource %s\n", name);
+		goto out_err;
+	}
+	strlcpy(res->name, name, namelen + 1);
+	if (with_lvb) {
+		res->lksb.sb_lvbptr = kzalloc(LVB_SIZE, GFP_KERNEL);
+		if (!res->lksb.sb_lvbptr) {
+			pr_err("md-cluster: Unable to allocate LVB for resource %s\n", name);
+			goto out_err;
+		}
+		res->flags = DLM_LKF_VALBLK;
+	}
+
+	if (bastfn)
+		res->bast = bastfn;
+
+	res->flags |= DLM_LKF_EXPEDITE;
+
+	ret = dlm_lock_sync(res, DLM_LOCK_NL);
+	if (ret) {
+		pr_err("md-cluster: Unable to lock NL on new lock resource %s\n", name);
+		goto out_err;
+	}
+	res->flags &= ~DLM_LKF_EXPEDITE;
+	res->flags |= DLM_LKF_CONVERT;
+
+	return res;
+out_err:
+	kfree(res->lksb.sb_lvbptr);
+	kfree(res->name);
+	kfree(res);
+	return NULL;
+}
+
+static void lockres_free(struct dlm_lock_resource *res)
+{
+	if (!res)
+		return;
+
+	init_completion(&res->completion);
+	dlm_unlock(res->ls, res->lksb.sb_lkid, 0, &res->lksb, res);
+	wait_for_completion(&res->completion);
+
+	kfree(res->name);
+	kfree(res->lksb.sb_lvbptr);
+	kfree(res);
+}
+
+static char *pretty_uuid(char *dest, char *src)
+{
+	int i, len = 0;
+
+	for (i = 0; i < 16; i++) {
+		if (i == 4 || i == 6 || i == 8 || i == 10)
+			len += sprintf(dest + len, "-");
+		len += sprintf(dest + len, "%02x", (__u8)src[i]);
+	}
+	return dest;
+}
+
+static void add_resync_info(struct mddev *mddev, struct dlm_lock_resource *lockres,
+		sector_t lo, sector_t hi)
+{
+	struct resync_info *ri;
+
+	ri = (struct resync_info *)lockres->lksb.sb_lvbptr;
+	ri->lo = cpu_to_le64(lo);
+	ri->hi = cpu_to_le64(hi);
+}
+
+static struct suspend_info *read_resync_info(struct mddev *mddev, struct dlm_lock_resource *lockres)
+{
+	struct resync_info ri;
+	struct suspend_info *s = NULL;
+	sector_t hi = 0;
+
+	dlm_lock_sync(lockres, DLM_LOCK_CR);
+	memcpy(&ri, lockres->lksb.sb_lvbptr, sizeof(struct resync_info));
+	hi = le64_to_cpu(ri.hi);
+	if (ri.hi > 0) {
+		s = kzalloc(sizeof(struct suspend_info), GFP_KERNEL);
+		if (!s)
+			goto out;
+		s->hi = hi;
+		s->lo = le64_to_cpu(ri.lo);
+	}
+	dlm_unlock_sync(lockres);
+out:
+	return s;
+}
+
+static void recover_bitmaps(struct md_thread *thread)
+{
+	struct mddev *mddev = thread->mddev;
+	struct md_cluster_info *cinfo = mddev->cluster_info;
+	struct dlm_lock_resource *bm_lockres;
+	char str[64];
+	int slot, ret;
+	struct suspend_info *s, *tmp;
+	sector_t lo, hi;
+
+	while (cinfo->recovery_map) {
+		slot = fls64((u64)cinfo->recovery_map) - 1;
+
+		/* Clear suspend_area associated with the bitmap */
+		spin_lock_irq(&cinfo->suspend_lock);
+		list_for_each_entry_safe(s, tmp, &cinfo->suspend_list, list)
+			if (slot == s->slot) {
+				list_del(&s->list);
+				kfree(s);
+			}
+		spin_unlock_irq(&cinfo->suspend_lock);
+
+		snprintf(str, 64, "bitmap%04d", slot);
+		bm_lockres = lockres_init(mddev, str, NULL, 1);
+		if (!bm_lockres) {
+			pr_err("md-cluster: Cannot initialize bitmaps\n");
+			goto clear_bit;
+		}
+
+		ret = dlm_lock_sync(bm_lockres, DLM_LOCK_PW);
+		if (ret) {
+			pr_err("md-cluster: Could not DLM lock %s: %d\n",
+					str, ret);
+			goto clear_bit;
+		}
+		ret = bitmap_copy_from_slot(mddev, slot, &lo, &hi, true);
+		if (ret) {
+			pr_err("md-cluster: Could not copy data from bitmap %d\n", slot);
+			goto dlm_unlock;
+		}
+		if (hi > 0) {
+			/* TODO:Wait for current resync to get over */
+			set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
+			if (lo < mddev->recovery_cp)
+				mddev->recovery_cp = lo;
+			md_check_recovery(mddev);
+		}
+dlm_unlock:
+		dlm_unlock_sync(bm_lockres);
+clear_bit:
+		clear_bit(slot, &cinfo->recovery_map);
+	}
+}
+
+static void recover_prep(void *arg)
+{
+}
+
+static void recover_slot(void *arg, struct dlm_slot *slot)
+{
+	struct mddev *mddev = arg;
+	struct md_cluster_info *cinfo = mddev->cluster_info;
+
+	pr_info("md-cluster: %s Node %d/%d down. My slot: %d. Initiating recovery.\n",
+			mddev->bitmap_info.cluster_name,
+			slot->nodeid, slot->slot,
+			cinfo->slot_number);
+	set_bit(slot->slot - 1, &cinfo->recovery_map);
+	if (!cinfo->recovery_thread) {
+		cinfo->recovery_thread = md_register_thread(recover_bitmaps,
+				mddev, "recover");
+		if (!cinfo->recovery_thread) {
+			pr_warn("md-cluster: Could not create recovery thread\n");
+			return;
+		}
+	}
+	md_wakeup_thread(cinfo->recovery_thread);
+}
+
+static void recover_done(void *arg, struct dlm_slot *slots,
+		int num_slots, int our_slot,
+		uint32_t generation)
+{
+	struct mddev *mddev = arg;
+	struct md_cluster_info *cinfo = mddev->cluster_info;
+
+	cinfo->slot_number = our_slot;
+	complete(&cinfo->completion);
+}
+
+static const struct dlm_lockspace_ops md_ls_ops = {
+	.recover_prep = recover_prep,
+	.recover_slot = recover_slot,
+	.recover_done = recover_done,
+};
+
+/*
+ * The BAST function for the ack lock resource
+ * This function wakes up the receive thread in
+ * order to receive and process the message.
+ */
+static void ack_bast(void *arg, int mode)
+{
+	struct dlm_lock_resource *res = (struct dlm_lock_resource *)arg;
+	struct md_cluster_info *cinfo = res->mddev->cluster_info;
+
+	if (mode == DLM_LOCK_EX)
+		md_wakeup_thread(cinfo->recv_thread);
+}
+
+static void __remove_suspend_info(struct md_cluster_info *cinfo, int slot)
+{
+	struct suspend_info *s, *tmp;
+
+	list_for_each_entry_safe(s, tmp, &cinfo->suspend_list, list)
+		if (slot == s->slot) {
+			pr_info("%s:%d Deleting suspend_info: %d\n",
+					__func__, __LINE__, slot);
+			list_del(&s->list);
+			kfree(s);
+			break;
+		}
+}
+
+static void remove_suspend_info(struct md_cluster_info *cinfo, int slot)
+{
+	spin_lock_irq(&cinfo->suspend_lock);
+	__remove_suspend_info(cinfo, slot);
+	spin_unlock_irq(&cinfo->suspend_lock);
+}
+
+
+static void process_suspend_info(struct md_cluster_info *cinfo,
+		int slot, sector_t lo, sector_t hi)
+{
+	struct suspend_info *s;
+
+	if (!hi) {
+		remove_suspend_info(cinfo, slot);
+		return;
+	}
+	s = kzalloc(sizeof(struct suspend_info), GFP_KERNEL);
+	if (!s)
+		return;
+	s->slot = slot;
+	s->lo = lo;
+	s->hi = hi;
+	spin_lock_irq(&cinfo->suspend_lock);
+	/* Remove existing entry (if exists) before adding */
+	__remove_suspend_info(cinfo, slot);
+	list_add(&s->list, &cinfo->suspend_list);
+	spin_unlock_irq(&cinfo->suspend_lock);
+}
+
+static void process_add_new_disk(struct mddev *mddev, struct cluster_msg *cmsg)
+{
+	char disk_uuid[64];
+	struct md_cluster_info *cinfo = mddev->cluster_info;
+	char event_name[] = "EVENT=ADD_DEVICE";
+	char raid_slot[16];
+	char *envp[] = {event_name, disk_uuid, raid_slot, NULL};
+	int len;
+
+	len = snprintf(disk_uuid, 64, "DEVICE_UUID=");
+	pretty_uuid(disk_uuid + len, cmsg->uuid);
+	snprintf(raid_slot, 16, "RAID_DISK=%d", cmsg->raid_slot);
+	pr_info("%s:%d Sending kobject change with %s and %s\n", __func__, __LINE__, disk_uuid, raid_slot);
+	init_completion(&cinfo->newdisk_completion);
+	set_bit(MD_CLUSTER_WAITING_FOR_NEWDISK, &cinfo->state);
+	kobject_uevent_env(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE, envp);
+	wait_for_completion_timeout(&cinfo->newdisk_completion,
+			NEW_DEV_TIMEOUT);
+	clear_bit(MD_CLUSTER_WAITING_FOR_NEWDISK, &cinfo->state);
+}
+
+
+static void process_metadata_update(struct mddev *mddev, struct cluster_msg *msg)
+{
+	struct md_cluster_info *cinfo = mddev->cluster_info;
+
+	md_reload_sb(mddev);
+	dlm_lock_sync(cinfo->no_new_dev_lockres, DLM_LOCK_CR);
+}
+
+static void process_remove_disk(struct mddev *mddev, struct cluster_msg *msg)
+{
+	struct md_rdev *rdev = md_find_rdev_nr_rcu(mddev, msg->raid_slot);
+
+	if (rdev)
+		md_kick_rdev_from_array(rdev);
+	else
+		pr_warn("%s: %d Could not find disk(%d) to REMOVE\n", __func__, __LINE__, msg->raid_slot);
+}
+
+static void process_readd_disk(struct mddev *mddev, struct cluster_msg *msg)
+{
+	struct md_rdev *rdev = md_find_rdev_nr_rcu(mddev, msg->raid_slot);
+
+	if (rdev && test_bit(Faulty, &rdev->flags))
+		clear_bit(Faulty, &rdev->flags);
+	else
+		pr_warn("%s: %d Could not find disk(%d) which is faulty", __func__, __LINE__, msg->raid_slot);
+}
+
+static void process_recvd_msg(struct mddev *mddev, struct cluster_msg *msg)
+{
+	switch (msg->type) {
+	case METADATA_UPDATED:
+		pr_info("%s: %d Received message: METADATA_UPDATE from %d\n",
+			__func__, __LINE__, msg->slot);
+		process_metadata_update(mddev, msg);
+		break;
+	case RESYNCING:
+		pr_info("%s: %d Received message: RESYNCING from %d\n",
+			__func__, __LINE__, msg->slot);
+		process_suspend_info(mddev->cluster_info, msg->slot,
+				msg->low, msg->high);
+		break;
+	case NEWDISK:
+		pr_info("%s: %d Received message: NEWDISK from %d\n",
+			__func__, __LINE__, msg->slot);
+		process_add_new_disk(mddev, msg);
+		break;
+	case REMOVE:
+		pr_info("%s: %d Received REMOVE from %d\n",
+			__func__, __LINE__, msg->slot);
+		process_remove_disk(mddev, msg);
+		break;
+	case RE_ADD:
+		pr_info("%s: %d Received RE_ADD from %d\n",
+			__func__, __LINE__, msg->slot);
+		process_readd_disk(mddev, msg);
+		break;
+	default:
+		pr_warn("%s:%d Received unknown message from %d\n",
+			__func__, __LINE__, msg->slot);
+	}
+}
+
+/*
+ * thread for receiving message
+ */
+static void recv_daemon(struct md_thread *thread)
+{
+	struct md_cluster_info *cinfo = thread->mddev->cluster_info;
+	struct dlm_lock_resource *ack_lockres = cinfo->ack_lockres;
+	struct dlm_lock_resource *message_lockres = cinfo->message_lockres;
+	struct cluster_msg msg;
+
+	/*get CR on Message*/
+	if (dlm_lock_sync(message_lockres, DLM_LOCK_CR)) {
+		pr_err("md/raid1:failed to get CR on MESSAGE\n");
+		return;
+	}
+
+	/* read lvb and wake up thread to process this message_lockres */
+	memcpy(&msg, message_lockres->lksb.sb_lvbptr, sizeof(struct cluster_msg));
+	process_recvd_msg(thread->mddev, &msg);
+
+	/*release CR on ack_lockres*/
+	dlm_unlock_sync(ack_lockres);
+	/*up-convert to EX on message_lockres*/
+	dlm_lock_sync(message_lockres, DLM_LOCK_EX);
+	/*get CR on ack_lockres again*/
+	dlm_lock_sync(ack_lockres, DLM_LOCK_CR);
+	/*release CR on message_lockres*/
+	dlm_unlock_sync(message_lockres);
+}
+
+/* lock_comm()
+ * Takes the lock on the TOKEN lock resource so no other
+ * node can communicate while the operation is underway.
+ */
+static int lock_comm(struct md_cluster_info *cinfo)
+{
+	int error;
+
+	error = dlm_lock_sync(cinfo->token_lockres, DLM_LOCK_EX);
+	if (error)
+		pr_err("md-cluster(%s:%d): failed to get EX on TOKEN (%d)\n",
+				__func__, __LINE__, error);
+	return error;
+}
+
+static void unlock_comm(struct md_cluster_info *cinfo)
+{
+	dlm_unlock_sync(cinfo->token_lockres);
+}
+
+/* __sendmsg()
+ * This function performs the actual sending of the message. This function is
+ * usually called after performing the encompassing operation
+ * The function:
+ * 1. Grabs the message lockresource in EX mode
+ * 2. Copies the message to the message LVB
+ * 3. Downconverts message lockresource to CR
+ * 4. Upconverts ack lock resource from CR to EX. This forces the BAST on other nodes
+ *    and the other nodes read the message. The thread will wait here until all other
+ *    nodes have released ack lock resource.
+ * 5. Downconvert ack lockresource to CR
+ */
+static int __sendmsg(struct md_cluster_info *cinfo, struct cluster_msg *cmsg)
+{
+	int error;
+	int slot = cinfo->slot_number - 1;
+
+	cmsg->slot = cpu_to_le32(slot);
+	/*get EX on Message*/
+	error = dlm_lock_sync(cinfo->message_lockres, DLM_LOCK_EX);
+	if (error) {
+		pr_err("md-cluster: failed to get EX on MESSAGE (%d)\n", error);
+		goto failed_message;
+	}
+
+	memcpy(cinfo->message_lockres->lksb.sb_lvbptr, (void *)cmsg,
+			sizeof(struct cluster_msg));
+	/*down-convert EX to CR on Message*/
+	error = dlm_lock_sync(cinfo->message_lockres, DLM_LOCK_CR);
+	if (error) {
+		pr_err("md-cluster: failed to convert EX to CR on MESSAGE(%d)\n",
+				error);
+		goto failed_message;
+	}
+
+	/*up-convert CR to EX on Ack*/
+	error = dlm_lock_sync(cinfo->ack_lockres, DLM_LOCK_EX);
+	if (error) {
+		pr_err("md-cluster: failed to convert CR to EX on ACK(%d)\n",
+				error);
+		goto failed_ack;
+	}
+
+	/*down-convert EX to CR on Ack*/
+	error = dlm_lock_sync(cinfo->ack_lockres, DLM_LOCK_CR);
+	if (error) {
+		pr_err("md-cluster: failed to convert EX to CR on ACK(%d)\n",
+				error);
+		goto failed_ack;
+	}
+
+failed_ack:
+	dlm_unlock_sync(cinfo->message_lockres);
+failed_message:
+	return error;
+}
+
+static int sendmsg(struct md_cluster_info *cinfo, struct cluster_msg *cmsg)
+{
+	int ret;
+
+	lock_comm(cinfo);
+	ret = __sendmsg(cinfo, cmsg);
+	unlock_comm(cinfo);
+	return ret;
+}
+
+static int gather_all_resync_info(struct mddev *mddev, int total_slots)
+{
+	struct md_cluster_info *cinfo = mddev->cluster_info;
+	int i, ret = 0;
+	struct dlm_lock_resource *bm_lockres;
+	struct suspend_info *s;
+	char str[64];
+
+
+	for (i = 0; i < total_slots; i++) {
+		memset(str, '\0', 64);
+		snprintf(str, 64, "bitmap%04d", i);
+		bm_lockres = lockres_init(mddev, str, NULL, 1);
+		if (!bm_lockres)
+			return -ENOMEM;
+		if (i == (cinfo->slot_number - 1))
+			continue;
+
+		bm_lockres->flags |= DLM_LKF_NOQUEUE;
+		ret = dlm_lock_sync(bm_lockres, DLM_LOCK_PW);
+		if (ret == -EAGAIN) {
+			memset(bm_lockres->lksb.sb_lvbptr, '\0', LVB_SIZE);
+			s = read_resync_info(mddev, bm_lockres);
+			if (s) {
+				pr_info("%s:%d Resync[%llu..%llu] in progress on %d\n",
+						__func__, __LINE__,
+						(unsigned long long) s->lo,
+						(unsigned long long) s->hi, i);
+				spin_lock_irq(&cinfo->suspend_lock);
+				s->slot = i;
+				list_add(&s->list, &cinfo->suspend_list);
+				spin_unlock_irq(&cinfo->suspend_lock);
+			}
+			ret = 0;
+			lockres_free(bm_lockres);
+			continue;
+		}
+		if (ret)
+			goto out;
+		/* TODO: Read the disk bitmap sb and check if it needs recovery */
+		dlm_unlock_sync(bm_lockres);
+		lockres_free(bm_lockres);
+	}
+out:
+	return ret;
+}
+
+static int join(struct mddev *mddev, int nodes)
+{
+	struct md_cluster_info *cinfo;
+	int ret, ops_rv;
+	char str[64];
+
+	if (!try_module_get(THIS_MODULE))
+		return -ENOENT;
+
+	cinfo = kzalloc(sizeof(struct md_cluster_info), GFP_KERNEL);
+	if (!cinfo)
+		return -ENOMEM;
+
+	init_completion(&cinfo->completion);
+
+	mutex_init(&cinfo->sb_mutex);
+	mddev->cluster_info = cinfo;
+
+	memset(str, 0, 64);
+	pretty_uuid(str, mddev->uuid);
+	ret = dlm_new_lockspace(str, mddev->bitmap_info.cluster_name,
+				DLM_LSFL_FS, LVB_SIZE,
+				&md_ls_ops, mddev, &ops_rv, &cinfo->lockspace);
+	if (ret)
+		goto err;
+	wait_for_completion(&cinfo->completion);
+	if (nodes < cinfo->slot_number) {
+		pr_err("md-cluster: Slot allotted(%d) is greater than available slots(%d).",
+			cinfo->slot_number, nodes);
+		ret = -ERANGE;
+		goto err;
+	}
+	cinfo->sb_lock = lockres_init(mddev, "cmd-super",
+					NULL, 0);
+	if (!cinfo->sb_lock) {
+		ret = -ENOMEM;
+		goto err;
+	}
+	/* Initiate the communication resources */
+	ret = -ENOMEM;
+	cinfo->recv_thread = md_register_thread(recv_daemon, mddev, "cluster_recv");
+	if (!cinfo->recv_thread) {
+		pr_err("md-cluster: cannot allocate memory for recv_thread!\n");
+		goto err;
+	}
+	cinfo->message_lockres = lockres_init(mddev, "message", NULL, 1);
+	if (!cinfo->message_lockres)
+		goto err;
+	cinfo->token_lockres = lockres_init(mddev, "token", NULL, 0);
+	if (!cinfo->token_lockres)
+		goto err;
+	cinfo->ack_lockres = lockres_init(mddev, "ack", ack_bast, 0);
+	if (!cinfo->ack_lockres)
+		goto err;
+	cinfo->no_new_dev_lockres = lockres_init(mddev, "no-new-dev", NULL, 0);
+	if (!cinfo->no_new_dev_lockres)
+		goto err;
+
+	/* get sync CR lock on ACK. */
+	if (dlm_lock_sync(cinfo->ack_lockres, DLM_LOCK_CR))
+		pr_err("md-cluster: failed to get a sync CR lock on ACK!(%d)\n",
+				ret);
+	/* get sync CR lock on no-new-dev. */
+	if (dlm_lock_sync(cinfo->no_new_dev_lockres, DLM_LOCK_CR))
+		pr_err("md-cluster: failed to get a sync CR lock on no-new-dev!(%d)\n", ret);
+
+
+	pr_info("md-cluster: Joined cluster %s slot %d\n", str, cinfo->slot_number);
+	snprintf(str, 64, "bitmap%04d", cinfo->slot_number - 1);
+	cinfo->bitmap_lockres = lockres_init(mddev, str, NULL, 1);
+	if (!cinfo->bitmap_lockres)
+		goto err;
+	if (dlm_lock_sync(cinfo->bitmap_lockres, DLM_LOCK_PW)) {
+		pr_err("Failed to get bitmap lock\n");
+		ret = -EINVAL;
+		goto err;
+	}
+
+	INIT_LIST_HEAD(&cinfo->suspend_list);
+	spin_lock_init(&cinfo->suspend_lock);
+
+	ret = gather_all_resync_info(mddev, nodes);
+	if (ret)
+		goto err;
+
+	return 0;
+err:
+	lockres_free(cinfo->message_lockres);
+	lockres_free(cinfo->token_lockres);
+	lockres_free(cinfo->ack_lockres);
+	lockres_free(cinfo->no_new_dev_lockres);
+	lockres_free(cinfo->bitmap_lockres);
+	lockres_free(cinfo->sb_lock);
+	if (cinfo->lockspace)
+		dlm_release_lockspace(cinfo->lockspace, 2);
+	mddev->cluster_info = NULL;
+	kfree(cinfo);
+	module_put(THIS_MODULE);
+	return ret;
+}
+
+static int leave(struct mddev *mddev)
+{
+	struct md_cluster_info *cinfo = mddev->cluster_info;
+
+	if (!cinfo)
+		return 0;
+	md_unregister_thread(&cinfo->recovery_thread);
+	md_unregister_thread(&cinfo->recv_thread);
+	lockres_free(cinfo->message_lockres);
+	lockres_free(cinfo->token_lockres);
+	lockres_free(cinfo->ack_lockres);
+	lockres_free(cinfo->no_new_dev_lockres);
+	lockres_free(cinfo->sb_lock);
+	lockres_free(cinfo->bitmap_lockres);
+	dlm_release_lockspace(cinfo->lockspace, 2);
+	return 0;
+}
+
+/* slot_number(): Returns the MD slot number to use
+ * DLM starts the slot numbers from 1, wheras cluster-md
+ * wants the number to be from zero, so we deduct one
+ */
+static int slot_number(struct mddev *mddev)
+{
+	struct md_cluster_info *cinfo = mddev->cluster_info;
+
+	return cinfo->slot_number - 1;
+}
+
+static void resync_info_update(struct mddev *mddev, sector_t lo, sector_t hi)
+{
+	struct md_cluster_info *cinfo = mddev->cluster_info;
+
+	add_resync_info(mddev, cinfo->bitmap_lockres, lo, hi);
+	/* Re-acquire the lock to refresh LVB */
+	dlm_lock_sync(cinfo->bitmap_lockres, DLM_LOCK_PW);
+}
+
+static int metadata_update_start(struct mddev *mddev)
+{
+	return lock_comm(mddev->cluster_info);
+}
+
+static int metadata_update_finish(struct mddev *mddev)
+{
+	struct md_cluster_info *cinfo = mddev->cluster_info;
+	struct cluster_msg cmsg;
+	int ret;
+
+	memset(&cmsg, 0, sizeof(cmsg));
+	cmsg.type = cpu_to_le32(METADATA_UPDATED);
+	ret = __sendmsg(cinfo, &cmsg);
+	unlock_comm(cinfo);
+	return ret;
+}
+
+static int metadata_update_cancel(struct mddev *mddev)
+{
+	struct md_cluster_info *cinfo = mddev->cluster_info;
+
+	return dlm_unlock_sync(cinfo->token_lockres);
+}
+
+static int resync_send(struct mddev *mddev, enum msg_type type,
+		sector_t lo, sector_t hi)
+{
+	struct md_cluster_info *cinfo = mddev->cluster_info;
+	struct cluster_msg cmsg;
+	int slot = cinfo->slot_number - 1;
+
+	pr_info("%s:%d lo: %llu hi: %llu\n", __func__, __LINE__,
+			(unsigned long long)lo,
+			(unsigned long long)hi);
+	resync_info_update(mddev, lo, hi);
+	cmsg.type = cpu_to_le32(type);
+	cmsg.slot = cpu_to_le32(slot);
+	cmsg.low = cpu_to_le64(lo);
+	cmsg.high = cpu_to_le64(hi);
+	return sendmsg(cinfo, &cmsg);
+}
+
+static int resync_start(struct mddev *mddev, sector_t lo, sector_t hi)
+{
+	pr_info("%s:%d\n", __func__, __LINE__);
+	return resync_send(mddev, RESYNCING, lo, hi);
+}
+
+static void resync_finish(struct mddev *mddev)
+{
+	pr_info("%s:%d\n", __func__, __LINE__);
+	resync_send(mddev, RESYNCING, 0, 0);
+}
+
+static int area_resyncing(struct mddev *mddev, sector_t lo, sector_t hi)
+{
+	struct md_cluster_info *cinfo = mddev->cluster_info;
+	int ret = 0;
+	struct suspend_info *s;
+
+	spin_lock_irq(&cinfo->suspend_lock);
+	if (list_empty(&cinfo->suspend_list))
+		goto out;
+	list_for_each_entry(s, &cinfo->suspend_list, list)
+		if (hi > s->lo && lo < s->hi) {
+			ret = 1;
+			break;
+		}
+out:
+	spin_unlock_irq(&cinfo->suspend_lock);
+	return ret;
+}
+
+static int add_new_disk_start(struct mddev *mddev, struct md_rdev *rdev)
+{
+	struct md_cluster_info *cinfo = mddev->cluster_info;
+	struct cluster_msg cmsg;
+	int ret = 0;
+	struct mdp_superblock_1 *sb = page_address(rdev->sb_page);
+	char *uuid = sb->device_uuid;
+
+	memset(&cmsg, 0, sizeof(cmsg));
+	cmsg.type = cpu_to_le32(NEWDISK);
+	memcpy(cmsg.uuid, uuid, 16);
+	cmsg.raid_slot = rdev->desc_nr;
+	lock_comm(cinfo);
+	ret = __sendmsg(cinfo, &cmsg);
+	if (ret)
+		return ret;
+	cinfo->no_new_dev_lockres->flags |= DLM_LKF_NOQUEUE;
+	ret = dlm_lock_sync(cinfo->no_new_dev_lockres, DLM_LOCK_EX);
+	cinfo->no_new_dev_lockres->flags &= ~DLM_LKF_NOQUEUE;
+	/* Some node does not "see" the device */
+	if (ret == -EAGAIN)
+		ret = -ENOENT;
+	else
+		dlm_lock_sync(cinfo->no_new_dev_lockres, DLM_LOCK_CR);
+	return ret;
+}
+
+static int add_new_disk_finish(struct mddev *mddev)
+{
+	struct cluster_msg cmsg;
+	struct md_cluster_info *cinfo = mddev->cluster_info;
+	int ret;
+	/* Write sb and inform others */
+	md_update_sb(mddev, 1);
+	cmsg.type = METADATA_UPDATED;
+	ret = __sendmsg(cinfo, &cmsg);
+	unlock_comm(cinfo);
+	return ret;
+}
+
+static int new_disk_ack(struct mddev *mddev, bool ack)
+{
+	struct md_cluster_info *cinfo = mddev->cluster_info;
+
+	if (!test_bit(MD_CLUSTER_WAITING_FOR_NEWDISK, &cinfo->state)) {
+		pr_warn("md-cluster(%s): Spurious cluster confirmation\n", mdname(mddev));
+		return -EINVAL;
+	}
+
+	if (ack)
+		dlm_unlock_sync(cinfo->no_new_dev_lockres);
+	complete(&cinfo->newdisk_completion);
+	return 0;
+}
+
+static int remove_disk(struct mddev *mddev, struct md_rdev *rdev)
+{
+	struct cluster_msg cmsg;
+	struct md_cluster_info *cinfo = mddev->cluster_info;
+	cmsg.type = REMOVE;
+	cmsg.raid_slot = rdev->desc_nr;
+	return __sendmsg(cinfo, &cmsg);
+}
+
+static int gather_bitmaps(struct md_rdev *rdev)
+{
+	int sn, err;
+	sector_t lo, hi;
+	struct cluster_msg cmsg;
+	struct mddev *mddev = rdev->mddev;
+	struct md_cluster_info *cinfo = mddev->cluster_info;
+
+	cmsg.type = RE_ADD;
+	cmsg.raid_slot = rdev->desc_nr;
+	err = sendmsg(cinfo, &cmsg);
+	if (err)
+		goto out;
+
+	for (sn = 0; sn < mddev->bitmap_info.nodes; sn++) {
+		if (sn == (cinfo->slot_number - 1))
+			continue;
+		err = bitmap_copy_from_slot(mddev, sn, &lo, &hi, false);
+		if (err) {
+			pr_warn("md-cluster: Could not gather bitmaps from slot %d", sn);
+			goto out;
+		}
+		if ((hi > 0) && (lo < mddev->recovery_cp))
+			mddev->recovery_cp = lo;
+	}
+out:
+	return err;
+}
+
+static struct md_cluster_operations cluster_ops = {
+	.join   = join,
+	.leave  = leave,
+	.slot_number = slot_number,
+	.resync_info_update = resync_info_update,
+	.resync_start = resync_start,
+	.resync_finish = resync_finish,
+	.metadata_update_start = metadata_update_start,
+	.metadata_update_finish = metadata_update_finish,
+	.metadata_update_cancel = metadata_update_cancel,
+	.area_resyncing = area_resyncing,
+	.add_new_disk_start = add_new_disk_start,
+	.add_new_disk_finish = add_new_disk_finish,
+	.new_disk_ack = new_disk_ack,
+	.remove_disk = remove_disk,
+	.gather_bitmaps = gather_bitmaps,
+};
+
+static int __init cluster_init(void)
+{
+	pr_warn("md-cluster: EXPERIMENTAL. Use with caution\n");
+	pr_info("Registering Cluster MD functions\n");
+	register_md_cluster_operations(&cluster_ops, THIS_MODULE);
+	return 0;
+}
+
+static void cluster_exit(void)
+{
+	unregister_md_cluster_operations();
+}
+
+module_init(cluster_init);
+module_exit(cluster_exit);
+MODULE_LICENSE("GPL");
+MODULE_DESCRIPTION("Clustering support for MD");

drivers/md/md-cluster.h

···
+
+
+#ifndef _MD_CLUSTER_H
+#define _MD_CLUSTER_H
+
+#include "md.h"
+
+struct mddev;
+struct md_rdev;
+
+struct md_cluster_operations {
+	int (*join)(struct mddev *mddev, int nodes);
+	int (*leave)(struct mddev *mddev);
+	int (*slot_number)(struct mddev *mddev);
+	void (*resync_info_update)(struct mddev *mddev, sector_t lo, sector_t hi);
+	int (*resync_start)(struct mddev *mddev, sector_t lo, sector_t hi);
+	void (*resync_finish)(struct mddev *mddev);
+	int (*metadata_update_start)(struct mddev *mddev);
+	int (*metadata_update_finish)(struct mddev *mddev);
+	int (*metadata_update_cancel)(struct mddev *mddev);
+	int (*area_resyncing)(struct mddev *mddev, sector_t lo, sector_t hi);
+	int (*add_new_disk_start)(struct mddev *mddev, struct md_rdev *rdev);
+	int (*add_new_disk_finish)(struct mddev *mddev);
+	int (*new_disk_ack)(struct mddev *mddev, bool ack);
+	int (*remove_disk)(struct mddev *mddev, struct md_rdev *rdev);
+	int (*gather_bitmaps)(struct md_rdev *rdev);
+};
+
+#endif /* _MD_CLUSTER_H */

drivers/md/md.c

···
 #include <linux/slab.h>
 #include "md.h"
 #include "bitmap.h"
+#include "md-cluster.h"
 
 #ifndef MODULE
 static void autostart_arrays(int part);
···
  */
 static LIST_HEAD(pers_list);
 static DEFINE_SPINLOCK(pers_lock);
+
+struct md_cluster_operations *md_cluster_ops;
+EXPORT_SYMBOL(md_cluster_ops);
+struct module *md_cluster_mod;
+EXPORT_SYMBOL(md_cluster_mod);
 
 static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
 static struct workqueue_struct *md_wq;
···
 }
 EXPORT_SYMBOL_GPL(mddev_unlock);
 
-static struct md_rdev *find_rdev_nr_rcu(struct mddev *mddev, int nr)
+struct md_rdev *md_find_rdev_nr_rcu(struct mddev *mddev, int nr)
 {
 	struct md_rdev *rdev;
 
···
 
 	return NULL;
 }
+EXPORT_SYMBOL_GPL(md_find_rdev_nr_rcu);
 
 static struct md_rdev *find_rdev(struct mddev *mddev, dev_t dev)
 {
···
 		int choice = 0;
 		if (mddev->pers)
 			choice = mddev->raid_disks;
-		while (find_rdev_nr_rcu(mddev, choice))
+		while (md_find_rdev_nr_rcu(mddev, choice))
 			choice++;
 		rdev->desc_nr = choice;
 	} else {
-		if (find_rdev_nr_rcu(mddev, rdev->desc_nr)) {
+		if (md_find_rdev_nr_rcu(mddev, rdev->desc_nr)) {
 			rcu_read_unlock();
 			return -EBUSY;
 		}
···
 	kobject_put(&rdev->kobj);
 }
 
-static void kick_rdev_from_array(struct md_rdev *rdev)
+void md_kick_rdev_from_array(struct md_rdev *rdev)
 {
 	unbind_rdev_from_array(rdev);
 	export_rdev(rdev);
 }
+EXPORT_SYMBOL_GPL(md_kick_rdev_from_array);
 
 static void export_array(struct mddev *mddev)
 {
···
 	while (!list_empty(&mddev->disks)) {
 		rdev = list_first_entry(&mddev->disks, struct md_rdev,
 					same_set);
-		kick_rdev_from_array(rdev);
+		md_kick_rdev_from_array(rdev);
 	}
 	mddev->raid_disks = 0;
 	mddev->major_version = 0;
···
 	}
 }
 
-static void md_update_sb(struct mddev *mddev, int force_change)
+void md_update_sb(struct mddev *mddev, int force_change)
 {
 	struct md_rdev *rdev;
 	int sync_req;
···
 		wake_up(&rdev->blocked_wait);
 	}
 }
+EXPORT_SYMBOL(md_update_sb);
+
+static int add_bound_rdev(struct md_rdev *rdev)
+{
+	struct mddev *mddev = rdev->mddev;
+	int err = 0;
+
+	if (!mddev->pers->hot_remove_disk) {
+		/* If there is hot_add_disk but no hot_remove_disk
+		 * then added disks for geometry changes,
+		 * and should be added immediately.
+		 */
+		super_types[mddev->major_version].
+			validate_super(mddev, rdev);
+		err = mddev->pers->hot_add_disk(mddev, rdev);
+		if (err) {
+			unbind_rdev_from_array(rdev);
+			export_rdev(rdev);
+			return err;
+		}
+	}
+	sysfs_notify_dirent_safe(rdev->sysfs_state);
+
+	set_bit(MD_CHANGE_DEVS, &mddev->flags);
+	if (mddev->degraded)
+		set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
+	set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
+	md_new_event(mddev);
+	md_wakeup_thread(mddev->thread);
+	return 0;
+}
 
 /* words written to sysfs files may, or may not, be \n terminated.
  * We want to accept with case. For this we use cmd_match.
···
 			err = -EBUSY;
 		else {
 			struct mddev *mddev = rdev->mddev;
-			kick_rdev_from_array(rdev);
+			if (mddev_is_clustered(mddev))
+				md_cluster_ops->remove_disk(mddev, rdev);
+			md_kick_rdev_from_array(rdev);
+			if (mddev_is_clustered(mddev))
+				md_cluster_ops->metadata_update_start(mddev);
 			if (mddev->pers)
 				md_update_sb(mddev, 1);
 			md_new_event(mddev);
+			if (mddev_is_clustered(mddev))
+				md_cluster_ops->metadata_update_finish(mddev);
 			err = 0;
 		}
 	} else if (cmd_match(buf, "writemostly")) {
···
 			clear_bit(Replacement, &rdev->flags);
 			err = 0;
 		}
+	} else if (cmd_match(buf, "re-add")) {
+		if (test_bit(Faulty, &rdev->flags) && (rdev->raid_disk == -1)) {
+			/* clear_bit is performed _after_ all the devices
+			 * have their local Faulty bit cleared. If any writes
+			 * happen in the meantime in the local node, they
+			 * will land in the local bitmap, which will be synced
+			 * by this node eventually
+			 */
+			if (!mddev_is_clustered(rdev->mddev) ||
+			    (err = md_cluster_ops->gather_bitmaps(rdev)) == 0) {
+				clear_bit(Faulty, &rdev->flags);
+				err = add_bound_rdev(rdev);
+			}
+		} else
+			err = -EBUSY;
 	}
 	if (!err)
 		sysfs_notify_dirent_safe(rdev->sysfs_state);
···
 				"md: fatal superblock inconsistency in %s"
 				" -- removing from array\n",
 				bdevname(rdev->bdev,b));
-			kick_rdev_from_array(rdev);
+			md_kick_rdev_from_array(rdev);
 		}
 
 	super_types[mddev->major_version].
···
 			       "md: %s: %s: only %d devices permitted\n",
 			       mdname(mddev), bdevname(rdev->bdev, b),
 			       mddev->max_disks);
-			kick_rdev_from_array(rdev);
+			md_kick_rdev_from_array(rdev);
 			continue;
 		}
-		if (rdev != freshest)
+		if (rdev != freshest) {
 			if (super_types[mddev->major_version].
 			    validate_super(mddev, rdev)) {
 				printk(KERN_WARNING "md: kicking non-fresh %s"
 					" from array!\n",
 					bdevname(rdev->bdev,b));
-				kick_rdev_from_array(rdev);
+				md_kick_rdev_from_array(rdev);
 				continue;
 			}
+			/* No device should have a Candidate flag
+			 * when reading devices
+			 */
+			if (test_bit(Candidate, &rdev->flags)) {
+				pr_info("md: kicking Cluster Candidate %s from array!\n",
+					bdevname(rdev->bdev, b));
+				md_kick_rdev_from_array(rdev);
+			}
+		}
 		if (mddev->level == LEVEL_MULTIPATH) {
 			rdev->desc_nr = i++;
 			rdev->raid_disk = rdev->desc_nr;
···
 	if (err)
 		return err;
 	if (mddev->pers) {
+		if (mddev_is_clustered(mddev))
+			md_cluster_ops->metadata_update_start(mddev);
 		err = update_size(mddev, sectors);
 		md_update_sb(mddev, 1);
+		if (mddev_is_clustered(mddev))
+			md_cluster_ops->metadata_update_finish(mddev);
 	} else {
 		if (mddev->dev_sectors == 0 ||
 		    mddev->dev_sectors > sectors)
···
 {
 	unsigned long long min;
 	int err;
-	int chunk;
 
 	if (kstrtoull(buf, 10, &min))
 		return -EINVAL;
···
 	if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
 		goto out_unlock;
 
-	/* Must be a multiple of chunk_size */
-	chunk = mddev->chunk_sectors;
-	if (chunk) {
-		sector_t temp = min;
-
-		err = -EINVAL;
-		if (sector_div(temp, chunk))
-			goto out_unlock;
-	}
-	mddev->resync_min = min;
+	/* Round down to multiple of 4K for safety */
+	mddev->resync_min = round_down(min, 8);
 	err = 0;
 
 out_unlock:
···
 	}
 	if (err == 0 && pers->sync_request &&
 	    (mddev->bitmap_info.file || mddev->bitmap_info.offset)) {
-		err = bitmap_create(mddev);
-		if (err)
+		struct bitmap *bitmap;
+
+		bitmap = bitmap_create(mddev, -1);
+		if (IS_ERR(bitmap)) {
+			err = PTR_ERR(bitmap);
 			printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
 			       mdname(mddev), err);
+		} else
+			mddev->bitmap = bitmap;
+
 	}
 	if (err) {
 		mddev_detach(mddev);
···
 
 static void __md_stop_writes(struct mddev *mddev)
 {
+	if (mddev_is_clustered(mddev))
+		md_cluster_ops->metadata_update_start(mddev);
 	set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
 	flush_workqueue(md_misc_wq);
 	if (mddev->sync_thread) {
···
 		mddev->in_sync = 1;
 		md_update_sb(mddev, 1);
 	}
+	if (mddev_is_clustered(mddev))
+		md_cluster_ops->metadata_update_finish(mddev);
 }
 
 void md_stop_writes(struct mddev *mddev)
···
 		info.state = (1<<MD_SB_CLEAN);
 	if (mddev->bitmap && mddev->bitmap_info.offset)
 		info.state |= (1<<MD_SB_BITMAP_PRESENT);
+	if (mddev_is_clustered(mddev))
+		info.state |= (1<<MD_SB_CLUSTERED);
 	info.active_disks  = insync;
 	info.working_disks = working;
 	info.failed_disks  = failed;
···
 		return -EFAULT;
 
 	rcu_read_lock();
-	rdev = find_rdev_nr_rcu(mddev, info.number);
+	rdev = md_find_rdev_nr_rcu(mddev, info.number);
 	if (rdev) {
 		info.major = MAJOR(rdev->bdev->bd_dev);
 		info.minor = MINOR(rdev->bdev->bd_dev);
···
 	char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
 	struct md_rdev *rdev;
 	dev_t dev = MKDEV(info->major,info->minor);
+
+	if (mddev_is_clustered(mddev) &&
+		!(info->state & ((1 << MD_DISK_CLUSTER_ADD) | (1 << MD_DISK_CANDIDATE)))) {
+		pr_err("%s: Cannot add to clustered mddev.\n",
+			       mdname(mddev));
+		return -EINVAL;
+	}
 
 	if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
 		return -EOVERFLOW;
···
 		else
 			clear_bit(WriteMostly, &rdev->flags);
 
+		/*
+		 * check whether the device shows up in other nodes
+		 */
+		if (mddev_is_clustered(mddev)) {
+			if (info->state & (1 << MD_DISK_CANDIDATE)) {
+				/* Through --cluster-confirm */
+				set_bit(Candidate, &rdev->flags);
+				err = md_cluster_ops->new_disk_ack(mddev, true);
+				if (err) {
+					export_rdev(rdev);
+					return err;
+				}
+			} else if (info->state & (1 << MD_DISK_CLUSTER_ADD)) {
+				/* --add initiated by this node */
+				err = md_cluster_ops->add_new_disk_start(mddev, rdev);
+				if (err) {
+					md_cluster_ops->add_new_disk_finish(mddev);
+					export_rdev(rdev);
+					return err;
+				}
+			}
+		}
+
 		rdev->raid_disk = -1;
 		err = bind_rdev_to_array(rdev, mddev);
-		if (!err && !mddev->pers->hot_remove_disk) {
-			/* If there is hot_add_disk but no hot_remove_disk
-			 * then added disks for geometry changes,
-			 * and should be added immediately.
-			 */
-			super_types[mddev->major_version].
-				validate_super(mddev, rdev);
-			err = mddev->pers->hot_add_disk(mddev, rdev);
-			if (err)
-				unbind_rdev_from_array(rdev);
-		}
 		if (err)
 			export_rdev(rdev);
 		else
-			sysfs_notify_dirent_safe(rdev->sysfs_state);
-
-		set_bit(MD_CHANGE_DEVS, &mddev->flags);
-		if (mddev->degraded)
-			set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
-		set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
-		if (!err)
-			md_new_event(mddev);
-		md_wakeup_thread(mddev->thread);
+			err = add_bound_rdev(rdev);
+		if (mddev_is_clustered(mddev) &&
+				(info->state & (1 << MD_DISK_CLUSTER_ADD)))
+			md_cluster_ops->add_new_disk_finish(mddev);
 		return err;
 	}
 
···
 	if (!rdev)
 		return -ENXIO;
 
+	if (mddev_is_clustered(mddev))
+		md_cluster_ops->metadata_update_start(mddev);
+
 	clear_bit(Blocked, &rdev->flags);
 	remove_and_add_spares(mddev, rdev);
 
 	if (rdev->raid_disk >= 0)
 		goto busy;
 
-	kick_rdev_from_array(rdev);
+	if (mddev_is_clustered(mddev))
+		md_cluster_ops->remove_disk(mddev, rdev);
+
+	md_kick_rdev_from_array(rdev);
 	md_update_sb(mddev, 1);
 	md_new_event(mddev);
 
+	if (mddev_is_clustered(mddev))
+		md_cluster_ops->metadata_update_finish(mddev);
+
 	return 0;
 busy:
+	if (mddev_is_clustered(mddev))
+		md_cluster_ops->metadata_update_cancel(mddev);
 	printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n",
 		bdevname(rdev->bdev,b), mdname(mddev));
 	return -EBUSY;
···
 		err = -EINVAL;
 		goto abort_export;
 	}
+
+	if (mddev_is_clustered(mddev))
+		md_cluster_ops->metadata_update_start(mddev);
 	clear_bit(In_sync, &rdev->flags);
 	rdev->desc_nr = -1;
 	rdev->saved_raid_disk = -1;
 	err = bind_rdev_to_array(rdev, mddev);
 	if (err)
-		goto abort_export;
+		goto abort_clustered;
 
 	/*
 	 * The rest should better be atomic, we can have disk failures
···
 
 	md_update_sb(mddev, 1);
 
+	if (mddev_is_clustered(mddev))
+		md_cluster_ops->metadata_update_finish(mddev);
 	/*
 	 * Kick recovery, maybe this spare has to be added to the
 	 * array immediately.
···
 	md_new_event(mddev);
 	return 0;
 
+abort_clustered:
+	if (mddev_is_clustered(mddev))
+		md_cluster_ops->metadata_update_cancel(mddev);
 abort_export:
 	export_rdev(rdev);
 	return err;
···
 	if (mddev->pers) {
 		mddev->pers->quiesce(mddev, 1);
 		if (fd >= 0) {
-			err = bitmap_create(mddev);
-			if (!err)
+			struct bitmap *bitmap;
+
+			bitmap = bitmap_create(mddev, -1);
+			if (!IS_ERR(bitmap)) {
+				mddev->bitmap = bitmap;
 				err = bitmap_load(mddev);
+			} else
+				err = PTR_ERR(bitmap);
 		}
 		if (fd < 0 || err) {
 			bitmap_destroy(mddev);
···
 			return rv;
 		}
 	}
+	if (mddev_is_clustered(mddev))
+		md_cluster_ops->metadata_update_start(mddev);
 	if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
 		rv = update_size(mddev, (sector_t)info->size * 2);
 
···
 		rv = update_raid_disks(mddev, info->raid_disks);
 
 	if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
-		if (mddev->pers->quiesce == NULL || mddev->thread == NULL)
-			return -EINVAL;
-		if (mddev->recovery || mddev->sync_thread)
-			return -EBUSY;
+		if (mddev->pers->quiesce == NULL || mddev->thread == NULL) {
+			rv = -EINVAL;
+			goto err;
+		}
+		if (mddev->recovery || mddev->sync_thread) {
+			rv = -EBUSY;
+			goto err;
+		}
 		if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
+			struct bitmap *bitmap;
 			/* add the bitmap */
-			if (mddev->bitmap)
-				return -EEXIST;
-			if (mddev->bitmap_info.default_offset == 0)
-				return -EINVAL;
+			if (mddev->bitmap) {
+				rv = -EEXIST;
+				goto err;
+			}
+			if (mddev->bitmap_info.default_offset == 0) {
+				rv = -EINVAL;
+				goto err;
+			}
 			mddev->bitmap_info.offset =
 				mddev->bitmap_info.default_offset;
 			mddev->bitmap_info.space =
 				mddev->bitmap_info.default_space;
 			mddev->pers->quiesce(mddev, 1);
-			rv = bitmap_create(mddev);
-			if (!rv)
+			bitmap = bitmap_create(mddev, -1);
+			if (!IS_ERR(bitmap)) {
+				mddev->bitmap = bitmap;
 				rv = bitmap_load(mddev);
+			} else
+				rv = PTR_ERR(bitmap);
 			if (rv)
 				bitmap_destroy(mddev);
 			mddev->pers->quiesce(mddev, 0);
 		} else {
 			/* remove the bitmap */
-			if (!mddev->bitmap)
-				return -ENOENT;
-			if (mddev->bitmap->storage.file)
-				return -EINVAL;
+			if (!mddev->bitmap) {
+				rv = -ENOENT;
+				goto err;
+			}
+			if (mddev->bitmap->storage.file) {
+				rv = -EINVAL;
+				goto err;
+			}
 			mddev->pers->quiesce(mddev, 1);
 			bitmap_destroy(mddev);
 			mddev->pers->quiesce(mddev, 0);
···
 		}
 	}
 	md_update_sb(mddev, 1);
+	if (mddev_is_clustered(mddev))
+		md_cluster_ops->metadata_update_finish(mddev);
+	return rv;
+err:
+	if (mddev_is_clustered(mddev))
+		md_cluster_ops->metadata_update_cancel(mddev);
 	return rv;
 }
 
···
 	case SET_DISK_FAULTY:
 	case STOP_ARRAY:
 	case STOP_ARRAY_RO:
+	case CLUSTERED_DISK_NACK:
 		return true;
 	default:
 		return false;
···
 			err = add_new_disk(mddev, &info);
 		goto unlock;
 	}
+
+	case CLUSTERED_DISK_NACK:
+		if (mddev_is_clustered(mddev))
+			md_cluster_ops->new_disk_ack(mddev, false);
+		else
+			err = -EINVAL;
+		goto unlock;
 
 	case HOT_ADD_DISK:
 		err = hot_add_disk(mddev, new_decode_dev(arg));
···
 }
 EXPORT_SYMBOL(unregister_md_personality);
 
+int register_md_cluster_operations(struct md_cluster_operations *ops, struct module *module)
+{
+	if (md_cluster_ops != NULL)
+		return -EALREADY;
+	spin_lock(&pers_lock);
+	md_cluster_ops = ops;
+	md_cluster_mod = module;
+	spin_unlock(&pers_lock);
+	return 0;
+}
+EXPORT_SYMBOL(register_md_cluster_operations);
+
+int unregister_md_cluster_operations(void)
+{
+	spin_lock(&pers_lock);
+	md_cluster_ops = NULL;
+	spin_unlock(&pers_lock);
+	return 0;
+}
+EXPORT_SYMBOL(unregister_md_cluster_operations);
+
+int md_setup_cluster(struct mddev *mddev, int nodes)
+{
+	int err;
+
+	err = request_module("md-cluster");
+	if (err) {
+		pr_err("md-cluster module not found.\n");
+		return err;
+	}
+
+	spin_lock(&pers_lock);
+	if (!md_cluster_ops || !try_module_get(md_cluster_mod)) {
+		spin_unlock(&pers_lock);
+		return -ENOENT;
+	}
+	spin_unlock(&pers_lock);
+
+	return md_cluster_ops->join(mddev, nodes);
+}
+
+void md_cluster_stop(struct mddev *mddev)
+{
+	if (!md_cluster_ops)
+		return;
+	md_cluster_ops->leave(mddev);
+	module_put(md_cluster_mod);
+}
+
 static int is_mddev_idle(struct mddev *mddev, int init)
 {
 	struct md_rdev *rdev;
···
 		    mddev->safemode == 0)
 			mddev->safemode = 1;
 		spin_unlock(&mddev->lock);
+		if (mddev_is_clustered(mddev))
+			md_cluster_ops->metadata_update_start(mddev);
 		md_update_sb(mddev, 0);
+		if (mddev_is_clustered(mddev))
+			md_cluster_ops->metadata_update_finish(mddev);
 		sysfs_notify_dirent_safe(mddev->sysfs_state);
 	} else
 		spin_unlock(&mddev->lock);
···
 	md_new_event(mddev);
 	update_time = jiffies;
 
+	if (mddev_is_clustered(mddev))
+		md_cluster_ops->resync_start(mddev, j, max_sectors);
+
 	blk_start_plug(&plug);
 	while (j < max_sectors) {
 		sector_t sectors;
···
 		if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
 			break;
 
-		sectors = mddev->pers->sync_request(mddev, j, &skipped,
-						  currspeed < speed_min(mddev));
+		sectors = mddev->pers->sync_request(mddev, j, &skipped);
 		if (sectors == 0) {
 			set_bit(MD_RECOVERY_INTR, &mddev->recovery);
 			break;
···
 		j += sectors;
 		if (j > 2)
 			mddev->curr_resync = j;
+		if (mddev_is_clustered(mddev))
+			md_cluster_ops->resync_info_update(mddev, j, max_sectors);
 		mddev->curr_mark_cnt = io_sectors;
 		if (last_check == 0)
 			/* this is the earliest that rebuild will be
···
 			/((jiffies-mddev->resync_mark)/HZ +1) +1;
 
 		if (currspeed > speed_min(mddev)) {
-			if ((currspeed > speed_max(mddev)) ||
-					!is_mddev_idle(mddev, 0)) {
+			if (currspeed > speed_max(mddev)) {
 				msleep(500);
 				goto repeat;
+			}
+			if (!is_mddev_idle(mddev, 0)) {
+				/*
+				 * Give other IO more of a chance.
+				 * The faster the devices, the less we wait.
+				 */
+				wait_event(mddev->recovery_wait,
+					   !atomic_read(&mddev->recovery_active));
 			}
 		}
 	}
···
 	wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
 
 	/* tell personality that we are finished */
-	mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
+	mddev->pers->sync_request(mddev, max_sectors, &skipped);
+
+	if (mddev_is_clustered(mddev))
+		md_cluster_ops->resync_finish(mddev);
 
 	if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
 	    mddev->curr_resync > 2) {
···
 				sysfs_notify_dirent_safe(mddev->sysfs_state);
 		}
 
-		if (mddev->flags & MD_UPDATE_SB_FLAGS)
+		if (mddev->flags & MD_UPDATE_SB_FLAGS) {
+			if (mddev_is_clustered(mddev))
+				md_cluster_ops->metadata_update_start(mddev);
 			md_update_sb(mddev, 0);
+			if (mddev_is_clustered(mddev))
+				md_cluster_ops->metadata_update_finish(mddev);
+		}
 
 		if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
 		    !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
···
 			set_bit(MD_CHANGE_DEVS, &mddev->flags);
 		}
 	}
+	if (mddev_is_clustered(mddev))
+		md_cluster_ops->metadata_update_start(mddev);
 	if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
 	    mddev->pers->finish_reshape)
 		mddev->pers->finish_reshape(mddev);
···
 			rdev->saved_raid_disk = -1;
 
 	md_update_sb(mddev, 1);
+	if (mddev_is_clustered(mddev))
+		md_cluster_ops->metadata_update_finish(mddev);
 	clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
 	clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
 	clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
···
 err_wq:
 	return ret;
 }
+
+void md_reload_sb(struct mddev *mddev)
+{
+	struct md_rdev *rdev, *tmp;
+
+	rdev_for_each_safe(rdev, tmp, mddev) {
+		rdev->sb_loaded = 0;
+		ClearPageUptodate(rdev->sb_page);
+	}
+	mddev->raid_disks = 0;
+	analyze_sbs(mddev);
+	rdev_for_each_safe(rdev, tmp, mddev) {
+		struct mdp_superblock_1 *sb = page_address(rdev->sb_page);
+		/* since we don't write to faulty devices, we figure out if the
+		 *  disk is faulty by comparing events
+		 */
+		if (mddev->events > sb->events)
+			set_bit(Faulty, &rdev->flags);
+	}
+
+}
+EXPORT_SYMBOL(md_reload_sb);
 
 #ifndef MODULE
 

drivers/md/md.h

···
 #include <linux/timer.h>
 #include <linux/wait.h>
 #include <linux/workqueue.h>
+#include "md-cluster.h"
 
 #define MaxSector (~(sector_t)0)
 
···
 				 * a want_replacement device with same
 				 * raid_disk number.
 				 */
+	Candidate,		/* For clustered environments only:
+				 * This device is seen locally but not
+				 * by the whole cluster
+				 */
 };
 
 #define BB_LEN_MASK	(0x00000000000001FFULL)
···
 extern int rdev_clear_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
 				int is_new);
 extern void md_ack_all_badblocks(struct badblocks *bb);
+
+struct md_cluster_info;
 
 struct mddev {
 	void				*private;
···
 		unsigned long		daemon_sleep; /* how many jiffies between updates? */
 		unsigned long		max_write_behind; /* write-behind mode */
 		int			external;
+		int			nodes; /* Maximum number of nodes in the cluster */
+		char                    cluster_name[64]; /* Name of the cluster */
 	} bitmap_info;
 
 	atomic_t			max_corr_read_errors; /* max read retries */
···
 	struct work_struct flush_work;
 	struct work_struct event_work;	/* used by dm to report failure event */
 	void (*sync_super)(struct mddev *mddev, struct md_rdev *rdev);
+	struct md_cluster_info		*cluster_info;
 };
 
 static inline int __must_check mddev_lock(struct mddev *mddev)
···
 	int (*hot_add_disk) (struct mddev *mddev, struct md_rdev *rdev);
 	int (*hot_remove_disk) (struct mddev *mddev, struct md_rdev *rdev);
 	int (*spare_active) (struct mddev *mddev);
-	sector_t (*sync_request)(struct mddev *mddev, sector_t sector_nr, int *skipped, int go_faster);
+	sector_t (*sync_request)(struct mddev *mddev, sector_t sector_nr, int *skipped);
 	int (*resize) (struct mddev *mddev, sector_t sectors);
 	sector_t (*size) (struct mddev *mddev, sector_t sectors, int raid_disks);
 	int (*check_reshape) (struct mddev *mddev);
···
 
 extern int register_md_personality(struct md_personality *p);
 extern int unregister_md_personality(struct md_personality *p);
+extern int register_md_cluster_operations(struct md_cluster_operations *ops,
+		struct module *module);
+extern int unregister_md_cluster_operations(void);
+extern int md_setup_cluster(struct mddev *mddev, int nodes);
+extern void md_cluster_stop(struct mddev *mddev);
 extern struct md_thread *md_register_thread(
 	void (*run)(struct md_thread *thread),
 	struct mddev *mddev,
···
 				   struct mddev *mddev);
 
 extern void md_unplug(struct blk_plug_cb *cb, bool from_schedule);
+extern void md_reload_sb(struct mddev *mddev);
+extern void md_update_sb(struct mddev *mddev, int force);
+extern void md_kick_rdev_from_array(struct md_rdev * rdev);
+struct md_rdev *md_find_rdev_nr_rcu(struct mddev *mddev, int nr);
 static inline int mddev_check_plugged(struct mddev *mddev)
 {
 	return !!blk_check_plugged(md_unplug, mddev,
···
 	}
 }
 
+extern struct md_cluster_operations *md_cluster_ops;
+static inline int mddev_is_clustered(struct mddev *mddev)
+{
+	return mddev->cluster_info && mddev->bitmap_info.nodes > 1;
+}
 #endif /* _MD_MD_H */

drivers/md/raid0.c

···
 		goto abort;
 	}
 
-	blk_queue_io_min(mddev->queue, mddev->chunk_sectors << 9);
-	blk_queue_io_opt(mddev->queue,
-			 (mddev->chunk_sectors << 9) * mddev->raid_disks);
+	if (mddev->queue) {
+		blk_queue_io_min(mddev->queue, mddev->chunk_sectors << 9);
+		blk_queue_io_opt(mddev->queue,
+				 (mddev->chunk_sectors << 9) * mddev->raid_disks);
 
-	if (!discard_supported)
-		queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, mddev->queue);
-	else
-		queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, mddev->queue);
+		if (!discard_supported)
+			queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, mddev->queue);
+		else
+			queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, mddev->queue);
+	}
 
 	pr_debug("md/raid0:%s: done.\n", mdname(mddev));
 	*private_conf = conf;
···
 	}
 	if (md_check_no_bitmap(mddev))
 		return -EINVAL;
-	blk_queue_max_hw_sectors(mddev->queue, mddev->chunk_sectors);
-	blk_queue_max_write_same_sectors(mddev->queue, mddev->chunk_sectors);
-	blk_queue_max_discard_sectors(mddev->queue, mddev->chunk_sectors);
+
+	if (mddev->queue) {
+		blk_queue_max_hw_sectors(mddev->queue, mddev->chunk_sectors);
+		blk_queue_max_write_same_sectors(mddev->queue, mddev->chunk_sectors);
+		blk_queue_max_discard_sectors(mddev->queue, mddev->chunk_sectors);
+	}
 
 	/* if private is not null, we are here after takeover */
 	if (mddev->private == NULL) {
···
 	printk(KERN_INFO "md/raid0:%s: md_size is %llu sectors.\n",
 	       mdname(mddev),
 	       (unsigned long long)mddev->array_sectors);
-	/* calculate the max read-ahead size.
-	 * For read-ahead of large files to be effective, we need to
-	 * readahead at least twice a whole stripe. i.e. number of devices
-	 * multiplied by chunk size times 2.
-	 * If an individual device has an ra_pages greater than the
-	 * chunk size, then we will not drive that device as hard as it
-	 * wants.  We consider this a configuration error: a larger
-	 * chunksize should be used in that case.
-	 */
-	{
+
+	if (mddev->queue) {
+		/* calculate the max read-ahead size.
+		 * For read-ahead of large files to be effective, we need to
+		 * readahead at least twice a whole stripe. i.e. number of devices
+		 * multiplied by chunk size times 2.
+		 * If an individual device has an ra_pages greater than the
+		 * chunk size, then we will not drive that device as hard as it
+		 * wants.  We consider this a configuration error: a larger
+		 * chunksize should be used in that case.
+		 */
 		int stripe = mddev->raid_disks *
 			(mddev->chunk_sectors << 9) / PAGE_SIZE;
 		if (mddev->queue->backing_dev_info.ra_pages < 2* stripe)

drivers/md/raid1.c

···
 	has_nonrot_disk = 0;
 	choose_next_idle = 0;
 
-	choose_first = (conf->mddev->recovery_cp < this_sector + sectors);
+	if ((conf->mddev->recovery_cp < this_sector + sectors) ||
+	    (mddev_is_clustered(conf->mddev) &&
+	    md_cluster_ops->area_resyncing(conf->mddev, this_sector,
+		    this_sector + sectors)))
+		choose_first = 1;
+	else
+		choose_first = 0;
 
 	for (disk = 0 ; disk < conf->raid_disks * 2 ; disk++) {
 		sector_t dist;
···
 	md_write_start(mddev, bio); /* wait on superblock update early */
 
 	if (bio_data_dir(bio) == WRITE &&
-	    bio_end_sector(bio) > mddev->suspend_lo &&
-	    bio->bi_iter.bi_sector < mddev->suspend_hi) {
+	    ((bio_end_sector(bio) > mddev->suspend_lo &&
+	    bio->bi_iter.bi_sector < mddev->suspend_hi) ||
+	    (mddev_is_clustered(mddev) &&
+	     md_cluster_ops->area_resyncing(mddev, bio->bi_iter.bi_sector, bio_end_sector(bio))))) {
 		/* As the suspend_* range is controlled by
 		 * userspace, we want an interruptible
 		 * wait.
···
 			prepare_to_wait(&conf->wait_barrier,
 					&w, TASK_INTERRUPTIBLE);
 			if (bio_end_sector(bio) <= mddev->suspend_lo ||
-			    bio->bi_iter.bi_sector >= mddev->suspend_hi)
+			    bio->bi_iter.bi_sector >= mddev->suspend_hi ||
+			    (mddev_is_clustered(mddev) &&
+			     !md_cluster_ops->area_resyncing(mddev,
+				     bio->bi_iter.bi_sector, bio_end_sector(bio))))
 				break;
 			schedule();
 		}
···
 		struct md_rdev *rdev = conf->mirrors[i].rdev;
 		struct md_rdev *repl = conf->mirrors[conf->raid_disks + i].rdev;
 		if (repl
+		    && !test_bit(Candidate, &repl->flags)
 		    && repl->recovery_offset == MaxSector
 		    && !test_bit(Faulty, &repl->flags)
 		    && !test_and_set_bit(In_sync, &repl->flags)) {
···
  * that can be installed to exclude normal IO requests.
  */
 
-static sector_t sync_request(struct mddev *mddev, sector_t sector_nr, int *skipped, int go_faster)
+static sector_t sync_request(struct mddev *mddev, sector_t sector_nr, int *skipped)
 {
 	struct r1conf *conf = mddev->private;
 	struct r1bio *r1_bio;
···
 		*skipped = 1;
 		return sync_blocks;
 	}
-	/*
-	 * If there is non-resync activity waiting for a turn,
-	 * and resync is going fast enough,
-	 * then let it though before starting on this new sync request.
-	 */
-	if (!go_faster && conf->nr_waiting)
-		msleep_interruptible(1000);
 
 	bitmap_cond_end_sync(mddev->bitmap, sector_nr);
 	r1_bio = mempool_alloc(conf->r1buf_pool, GFP_NOIO);

drivers/md/raid10.c

···
  */
 
 static sector_t sync_request(struct mddev *mddev, sector_t sector_nr,
-			     int *skipped, int go_faster)
+			     int *skipped)
 {
 	struct r10conf *conf = mddev->private;
 	struct r10bio *r10_bio;
···
 	if (conf->geo.near_copies < conf->geo.raid_disks &&
 	    max_sector > (sector_nr | chunk_mask))
 		max_sector = (sector_nr | chunk_mask) + 1;
-	/*
-	 * If there is non-resync activity waiting for us then
-	 * put in a delay to throttle resync.
-	 */
-	if (!go_faster && conf->nr_waiting)
-		msleep_interruptible(1000);
 
 	/* Again, very different code for resync and recovery.
 	 * Both must result in an r10bio with a list of bios that

drivers/md/raid5.c

···
 #include <linux/slab.h>
 #include <linux/ratelimit.h>
 #include <linux/nodemask.h>
+#include <linux/flex_array.h>
 #include <trace/events/block.h>
 
 #include "md.h"
···
 	}
 }
 
-static int grow_buffers(struct stripe_head *sh)
+static int grow_buffers(struct stripe_head *sh, gfp_t gfp)
 {
 	int i;
 	int num = sh->raid_conf->pool_size;
···
 	for (i = 0; i < num; i++) {
 		struct page *page;
 
-		if (!(page = alloc_page(GFP_KERNEL))) {
+		if (!(page = alloc_page(gfp))) {
 			return 1;
 		}
 		sh->dev[i].page = page;
···
 	BUG_ON(atomic_read(&sh->count) != 0);
 	BUG_ON(test_bit(STRIPE_HANDLE, &sh->state));
 	BUG_ON(stripe_operations_active(sh));
+	BUG_ON(sh->batch_head);
 
 	pr_debug("init_stripe called, stripe %llu\n",
 		(unsigned long long)sector);
···
 	}
 	if (read_seqcount_retry(&conf->gen_lock, seq))
 		goto retry;
+	sh->overwrite_disks = 0;
 	insert_hash(conf, sh);
 	sh->cpu = smp_processor_id();
+	set_bit(STRIPE_BATCH_READY, &sh->state);
 }
 
 static struct stripe_head *__find_stripe(struct r5conf *conf, sector_t sector,
···
 				    *(conf->hash_locks + hash));
 		sh = __find_stripe(conf, sector, conf->generation - previous);
 		if (!sh) {
-			if (!conf->inactive_blocked)
+			if (!test_bit(R5_INACTIVE_BLOCKED, &conf->cache_state)) {
 				sh = get_free_stripe(conf, hash);
+				if (!sh && llist_empty(&conf->released_stripes) &&
+				    !test_bit(R5_DID_ALLOC, &conf->cache_state))
+					set_bit(R5_ALLOC_MORE,
+						&conf->cache_state);
+			}
 			if (noblock && sh == NULL)
 				break;
 			if (!sh) {
-				conf->inactive_blocked = 1;
+				set_bit(R5_INACTIVE_BLOCKED,
+					&conf->cache_state);
 				wait_event_lock_irq(
 					conf->wait_for_stripe,
 					!list_empty(conf->inactive_list + hash) &&
 					(atomic_read(&conf->active_stripes)
 					 < (conf->max_nr_stripes * 3 / 4)
-					 || !conf->inactive_blocked),
+					 || !test_bit(R5_INACTIVE_BLOCKED,
+						      &conf->cache_state)),
 					*(conf->hash_locks + hash));
-				conf->inactive_blocked = 0;
+				clear_bit(R5_INACTIVE_BLOCKED,
+					  &conf->cache_state);
 			} else {
 				init_stripe(sh, sector, previous);
 				atomic_inc(&sh->count);
···
 
 	spin_unlock_irq(conf->hash_locks + hash);
 	return sh;
+}
+
+static bool is_full_stripe_write(struct stripe_head *sh)
+{
+	BUG_ON(sh->overwrite_disks > (sh->disks - sh->raid_conf->max_degraded));
+	return sh->overwrite_disks == (sh->disks - sh->raid_conf->max_degraded);
+}
+
+static void lock_two_stripes(struct stripe_head *sh1, struct stripe_head *sh2)
+{
+	local_irq_disable();
+	if (sh1 > sh2) {
+		spin_lock(&sh2->stripe_lock);
+		spin_lock_nested(&sh1->stripe_lock, 1);
+	} else {
+		spin_lock(&sh1->stripe_lock);
+		spin_lock_nested(&sh2->stripe_lock, 1);
+	}
+}
+
+static void unlock_two_stripes(struct stripe_head *sh1, struct stripe_head *sh2)
+{
+	spin_unlock(&sh1->stripe_lock);
+	spin_unlock(&sh2->stripe_lock);
+	local_irq_enable();
+}
+
+/* Only freshly new full stripe normal write stripe can be added to a batch list */
+static bool stripe_can_batch(struct stripe_head *sh)
+{
+	return test_bit(STRIPE_BATCH_READY, &sh->state) &&
+		is_full_stripe_write(sh);
+}
+
+/* we only do back search */
+static void stripe_add_to_batch_list(struct r5conf *conf, struct stripe_head *sh)
+{
+	struct stripe_head *head;
+	sector_t head_sector, tmp_sec;
+	int hash;
+	int dd_idx;
+
+	if (!stripe_can_batch(sh))
+		return;
+	/* Don't cross chunks, so stripe pd_idx/qd_idx is the same */
+	tmp_sec = sh->sector;
+	if (!sector_div(tmp_sec, conf->chunk_sectors))
+		return;
+	head_sector = sh->sector - STRIPE_SECTORS;
+
+	hash = stripe_hash_locks_hash(head_sector);
+	spin_lock_irq(conf->hash_locks + hash);
+	head = __find_stripe(conf, head_sector, conf->generation);
+	if (head && !atomic_inc_not_zero(&head->count)) {
+		spin_lock(&conf->device_lock);
+		if (!atomic_read(&head->count)) {
+			if (!test_bit(STRIPE_HANDLE, &head->state))
+				atomic_inc(&conf->active_stripes);
+			BUG_ON(list_empty(&head->lru) &&
+			       !test_bit(STRIPE_EXPANDING, &head->state));
+			list_del_init(&head->lru);
+			if (head->group) {
+				head->group->stripes_cnt--;
+				head->group = NULL;
+			}
+		}
+		atomic_inc(&head->count);
+		spin_unlock(&conf->device_lock);
+	}
+	spin_unlock_irq(conf->hash_locks + hash);
+
+	if (!head)
+		return;
+	if (!stripe_can_batch(head))
+		goto out;
+
+	lock_two_stripes(head, sh);
+	/* clear_batch_ready clear the flag */
+	if (!stripe_can_batch(head) || !stripe_can_batch(sh))
+		goto unlock_out;
+
+	if (sh->batch_head)
+		goto unlock_out;
+
+	dd_idx = 0;
+	while (dd_idx == sh->pd_idx || dd_idx == sh->qd_idx)
+		dd_idx++;
+	if (head->dev[dd_idx].towrite->bi_rw != sh->dev[dd_idx].towrite->bi_rw)
+		goto unlock_out;
+
+	if (head->batch_head) {
+		spin_lock(&head->batch_head->batch_lock);
+		/* This batch list is already running */
+		if (!stripe_can_batch(head)) {
+			spin_unlock(&head->batch_head->batch_lock);
+			goto unlock_out;
+		}
+
+		/*
+		 * at this point, head's BATCH_READY could be cleared, but we
+		 * can still add the stripe to batch list
+		 */
+		list_add(&sh->batch_list, &head->batch_list);
+		spin_unlock(&head->batch_head->batch_lock);
+
+		sh->batch_head = head->batch_head;
+	} else {
+		head->batch_head = head;
+		sh->batch_head = head->batch_head;
+		spin_lock(&head->batch_lock);
+		list_add_tail(&sh->batch_list, &head->batch_list);
+		spin_unlock(&head->batch_lock);
+	}
+
+	if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE, &sh->state))
+		if (atomic_dec_return(&conf->preread_active_stripes)
+		    < IO_THRESHOLD)
+			md_wakeup_thread(conf->mddev->thread);
+
+	atomic_inc(&sh->count);
+unlock_out:
+	unlock_two_stripes(head, sh);
+out:
+	release_stripe(head);
 }
 
 /* Determine if 'data_offset' or 'new_data_offset' should be used
···
 {
 	struct r5conf *conf = sh->raid_conf;
 	int i, disks = sh->disks;
+	struct stripe_head *head_sh = sh;
 
 	might_sleep();
 
···
 		int replace_only = 0;
 		struct bio *bi, *rbi;
 		struct md_rdev *rdev, *rrdev = NULL;
+
+		sh = head_sh;
 		if (test_and_clear_bit(R5_Wantwrite, &sh->dev[i].flags)) {
 			if (test_and_clear_bit(R5_WantFUA, &sh->dev[i].flags))
 				rw = WRITE_FUA;
···
 		if (test_and_clear_bit(R5_SyncIO, &sh->dev[i].flags))
 			rw |= REQ_SYNC;
 
+again:
 		bi = &sh->dev[i].req;
 		rbi = &sh->dev[i].rreq; /* For writing to replacement */
 
···
 				/* We raced and saw duplicates */
 				rrdev = NULL;
 		} else {
-			if (test_bit(R5_ReadRepl, &sh->dev[i].flags) && rrdev)
+			if (test_bit(R5_ReadRepl, &head_sh->dev[i].flags) && rrdev)
 				rdev = rrdev;
 			rrdev = NULL;
 		}
···
 				__func__, (unsigned long long)sh->sector,
 				bi->bi_rw, i);
 			atomic_inc(&sh->count);
+			if (sh != head_sh)
+				atomic_inc(&head_sh->count);
 			if (use_new_offset(conf, sh))
 				bi->bi_iter.bi_sector = (sh->sector
 						 + rdev->new_data_offset);
 			else
 				bi->bi_iter.bi_sector = (sh->sector
 						 + rdev->data_offset);
-			if (test_bit(R5_ReadNoMerge, &sh->dev[i].flags))
+			if (test_bit(R5_ReadNoMerge, &head_sh->dev[i].flags))
 				bi->bi_rw |= REQ_NOMERGE;
 
 			if (test_bit(R5_SkipCopy, &sh->dev[i].flags))
···
 				__func__, (unsigned long long)sh->sector,
 				rbi->bi_rw, i);
 			atomic_inc(&sh->count);
+			if (sh != head_sh)
+				atomic_inc(&head_sh->count);
 			if (use_new_offset(conf, sh))
 				rbi->bi_iter.bi_sector = (sh->sector
 						  + rrdev->new_data_offset);
···
 			pr_debug("skip op %ld on disc %d for sector %llu\n",
 				bi->bi_rw, i, (unsigned long long)sh->sector);
 			clear_bit(R5_LOCKED, &sh->dev[i].flags);
+			if (sh->batch_head)
+				set_bit(STRIPE_BATCH_ERR,
+					&sh->batch_head->state);
 			set_bit(STRIPE_HANDLE, &sh->state);
 		}
+
+		if (!head_sh->batch_head)
+			continue;
+		sh = list_first_entry(&sh->batch_list, struct stripe_head,
+				      batch_list);
+		if (sh != head_sh)
+			goto again;
 	}
 }
 
···
 	struct async_submit_ctl submit;
 	int i;
 
+	BUG_ON(sh->batch_head);
 	pr_debug("%s: stripe %llu\n", __func__,
 		(unsigned long long)sh->sector);
 
···
 
 /* return a pointer to the address conversion region of the scribble buffer */
 static addr_conv_t *to_addr_conv(struct stripe_head *sh,
-				 struct raid5_percpu *percpu)
+				 struct raid5_percpu *percpu, int i)
 {
-	return percpu->scribble + sizeof(struct page *) * (sh->disks + 2);
+	void *addr;
+
+	addr = flex_array_get(percpu->scribble, i);
+	return addr + sizeof(struct page *) * (sh->disks + 2);
+}
+
+/* return a pointer to the address conversion region of the scribble buffer */
+static struct page **to_addr_page(struct raid5_percpu *percpu, int i)
+{
+	void *addr;
+
+	addr = flex_array_get(percpu->scribble, i);
+	return addr;
 }
 
 static struct dma_async_tx_descriptor *
 ops_run_compute5(struct stripe_head *sh, struct raid5_percpu *percpu)
 {
 	int disks = sh->disks;
-	struct page **xor_srcs = percpu->scribble;
+	struct page **xor_srcs = to_addr_page(percpu, 0);
 	int target = sh->ops.target;
 	struct r5dev *tgt = &sh->dev[target];
 	struct page *xor_dest = tgt->page;
···
 	struct dma_async_tx_descriptor *tx;
 	struct async_submit_ctl submit;
 	int i;
+
+	BUG_ON(sh->batch_head);
 
 	pr_debug("%s: stripe %llu block: %d\n",
 		__func__, (unsigned long long)sh->sector, target);
···
 	atomic_inc(&sh->count);
 
 	init_async_submit(&submit, ASYNC_TX_FENCE|ASYNC_TX_XOR_ZERO_DST, NULL,
-			  ops_complete_compute, sh, to_addr_conv(sh, percpu));
+			  ops_complete_compute, sh, to_addr_conv(sh, percpu, 0));
 	if (unlikely(count == 1))
 		tx = async_memcpy(xor_dest, xor_srcs[0], 0, 0, STRIPE_SIZE, &submit);
 	else
···
  * destination buffer is recorded in srcs[count] and the Q destination
  * is recorded in srcs[count+1]].
  */
-static int set_syndrome_sources(struct page **srcs, struct stripe_head *sh)
+static int set_syndrome_sources(struct page **srcs,
+				struct stripe_head *sh,
+				int srctype)
 {
 	int disks = sh->disks;
 	int syndrome_disks = sh->ddf_layout ? disks : (disks - 2);
···
 	i = d0_idx;
 	do {
 		int slot = raid6_idx_to_slot(i, sh, &count, syndrome_disks);
+		struct r5dev *dev = &sh->dev[i];
 
-		srcs[slot] = sh->dev[i].page;
+		if (i == sh->qd_idx || i == sh->pd_idx ||
+		    (srctype == SYNDROME_SRC_ALL) ||
+		    (srctype == SYNDROME_SRC_WANT_DRAIN &&
+		     test_bit(R5_Wantdrain, &dev->flags)) ||
+		    (srctype == SYNDROME_SRC_WRITTEN &&
+		     dev->written))
+			srcs[slot] = sh->dev[i].page;
 		i = raid6_next_disk(i, disks);
 	} while (i != d0_idx);
 
···
 ops_run_compute6_1(struct stripe_head *sh, struct raid5_percpu *percpu)
 {
 	int disks = sh->disks;
-	struct page **blocks = percpu->scribble;
+	struct page **blocks = to_addr_page(percpu, 0);
 	int target;
 	int qd_idx = sh->qd_idx;
 	struct dma_async_tx_descriptor *tx;
···
 	int i;
 	int count;
 
+	BUG_ON(sh->batch_head);
 	if (sh->ops.target < 0)
 		target = sh->ops.target2;
 	else if (sh->ops.target2 < 0)
···
 	atomic_inc(&sh->count);
 
 	if (target == qd_idx) {
-		count = set_syndrome_sources(blocks, sh);
+		count = set_syndrome_sources(blocks, sh, SYNDROME_SRC_ALL);
 		blocks[count] = NULL; /* regenerating p is not necessary */
 		BUG_ON(blocks[count+1] != dest); /* q should already be set */
 		init_async_submit(&submit, ASYNC_TX_FENCE, NULL,
 				  ops_complete_compute, sh,
-				  to_addr_conv(sh, percpu));
+				  to_addr_conv(sh, percpu, 0));
 		tx = async_gen_syndrome(blocks, 0, count+2, STRIPE_SIZE, &submit);
 	} else {
 		/* Compute any data- or p-drive using XOR */
···
 
 		init_async_submit(&submit, ASYNC_TX_FENCE|ASYNC_TX_XOR_ZERO_DST,
 				  NULL, ops_complete_compute, sh,
-				  to_addr_conv(sh, percpu));
+				  to_addr_conv(sh, percpu, 0));
 		tx = async_xor(dest, blocks, 0, count, STRIPE_SIZE, &submit);
 	}
 
···
 	struct r5dev *tgt = &sh->dev[target];
 	struct r5dev *tgt2 = &sh->dev[target2];
 	struct dma_async_tx_descriptor *tx;
-	struct page **blocks = percpu->scribble;
+	struct page **blocks = to_addr_page(percpu, 0);
 	struct async_submit_ctl submit;
 
+	BUG_ON(sh->batch_head);
 	pr_debug("%s: stripe %llu block1: %d block2: %d\n",
 		 __func__, (unsigned long long)sh->sector, target, target2);
 	BUG_ON(target < 0 || target2 < 0);
···
 			/* Missing P+Q, just recompute */
 			init_async_submit(&submit, ASYNC_TX_FENCE, NULL,
 					  ops_complete_compute, sh,
-					  to_addr_conv(sh, percpu));
+					  to_addr_conv(sh, percpu, 0));
 			return async_gen_syndrome(blocks, 0, syndrome_disks+2,
 						  STRIPE_SIZE, &submit);
 		} else {
···
 			init_async_submit(&submit,
 					  ASYNC_TX_FENCE|ASYNC_TX_XOR_ZERO_DST,
 					  NULL, NULL, NULL,
-					  to_addr_conv(sh, percpu));
+					  to_addr_conv(sh, percpu, 0));
 			tx = async_xor(dest, blocks, 0, count, STRIPE_SIZE,
 				       &submit);
 
-			count = set_syndrome_sources(blocks, sh);
+			count = set_syndrome_sources(blocks, sh, SYNDROME_SRC_ALL);
 			init_async_submit(&submit, ASYNC_TX_FENCE, tx,
 					  ops_complete_compute, sh,
-					  to_addr_conv(sh, percpu));
+					  to_addr_conv(sh, percpu, 0));
 			return async_gen_syndrome(blocks, 0, count+2,
 						  STRIPE_SIZE, &submit);
 		}
 	} else {
 		init_async_submit(&submit, ASYNC_TX_FENCE, NULL,
 				  ops_complete_compute, sh,
-				  to_addr_conv(sh, percpu));
+				  to_addr_conv(sh, percpu, 0));
 		if (failb == syndrome_disks) {
 			/* We're missing D+P. */
 			return async_raid6_datap_recov(syndrome_disks+2,
···
 }
 
 static struct dma_async_tx_descriptor *
-ops_run_prexor(struct stripe_head *sh, struct raid5_percpu *percpu,
-	       struct dma_async_tx_descriptor *tx)
+ops_run_prexor5(struct stripe_head *sh, struct raid5_percpu *percpu,
+		struct dma_async_tx_descriptor *tx)
 {
 	int disks = sh->disks;
-	struct page **xor_srcs = percpu->scribble;
+	struct page **xor_srcs = to_addr_page(percpu, 0);
 	int count = 0, pd_idx = sh->pd_idx, i;
 	struct async_submit_ctl submit;
 
 	/* existing parity data subtracted */
 	struct page *xor_dest = xor_srcs[count++] = sh->dev[pd_idx].page;
 
+	BUG_ON(sh->batch_head);
 	pr_debug("%s: stripe %llu\n", __func__,
 		(unsigned long long)sh->sector);
 
···
 	}
 
 	init_async_submit(&submit, ASYNC_TX_FENCE|ASYNC_TX_XOR_DROP_DST, tx,
-			  ops_complete_prexor, sh, to_addr_conv(sh, percpu));
+			  ops_complete_prexor, sh, to_addr_conv(sh, percpu, 0));
 	tx = async_xor(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, &submit);
+
+	return tx;
+}
+
+static struct dma_async_tx_descriptor *
+ops_run_prexor6(struct stripe_head *sh, struct raid5_percpu *percpu,
+		struct dma_async_tx_descriptor *tx)
+{
+	struct page **blocks = to_addr_page(percpu, 0);
+	int count;
+	struct async_submit_ctl submit;
+
+	pr_debug("%s: stripe %llu\n", __func__,
+		(unsigned long long)sh->sector);
+
+	count = set_syndrome_sources(blocks, sh, SYNDROME_SRC_WANT_DRAIN);
+
+	init_async_submit(&submit, ASYNC_TX_FENCE|ASYNC_TX_PQ_XOR_DST, tx,
+			  ops_complete_prexor, sh, to_addr_conv(sh, percpu, 0));
+	tx = async_gen_syndrome(blocks, 0, count+2, STRIPE_SIZE,  &submit);
 
 	return tx;
 }
···
 {
 	int disks = sh->disks;
 	int i;
+	struct stripe_head *head_sh = sh;
 
 	pr_debug("%s: stripe %llu\n", __func__,
 		(unsigned long long)sh->sector);
 
 	for (i = disks; i--; ) {
-		struct r5dev *dev = &sh->dev[i];
+		struct r5dev *dev;
 		struct bio *chosen;
 
-		if (test_and_clear_bit(R5_Wantdrain, &dev->flags)) {
+		sh = head_sh;
+		if (test_and_clear_bit(R5_Wantdrain, &head_sh->dev[i].flags)) {
 			struct bio *wbi;
 
+again:
+			dev = &sh->dev[i];
 			spin_lock_irq(&sh->stripe_lock);
 			chosen = dev->towrite;
 			dev->towrite = NULL;
+			sh->overwrite_disks = 0;
 			BUG_ON(dev->written);
 			wbi = dev->written = chosen;
 			spin_unlock_irq(&sh->stripe_lock);
···
 					}
 				}
 				wbi = r5_next_bio(wbi, dev->sector);
+			}
+
+			if (head_sh->batch_head) {
+				sh = list_first_entry(&sh->batch_list,
+						      struct stripe_head,
+						      batch_list);
+				if (sh == head_sh)
+					continue;
+				goto again;
 			}
 		}
 	}
···
 		     struct dma_async_tx_descriptor *tx)
 {
 	int disks = sh->disks;
-	struct page **xor_srcs = percpu->scribble;
+	struct page **xor_srcs;
 	struct async_submit_ctl submit;
-	int count = 0, pd_idx = sh->pd_idx, i;
+	int count, pd_idx = sh->pd_idx, i;
 	struct page *xor_dest;
 	int prexor = 0;
 	unsigned long flags;
+	int j = 0;
+	struct stripe_head *head_sh = sh;
+	int last_stripe;
 
 	pr_debug("%s: stripe %llu\n", __func__,
 		(unsigned long long)sh->sector);
···
 		ops_complete_reconstruct(sh);
 		return;
 	}
+again:
+	count = 0;
+	xor_srcs = to_addr_page(percpu, j);
 	/* check if prexor is active which means only process blocks
 	 * that are part of a read-modify-write (written)
 	 */
-	if (sh->reconstruct_state == reconstruct_state_prexor_drain_run) {
+	if (head_sh->reconstruct_state == reconstruct_state_prexor_drain_run) {
 		prexor = 1;
 		xor_dest = xor_srcs[count++] = sh->dev[pd_idx].page;
 		for (i = disks; i--; ) {
 			struct r5dev *dev = &sh->dev[i];
-			if (dev->written)
+			if (head_sh->dev[i].written)
 				xor_srcs[count++] = dev->page;
 		}
 	} else {
···
 	 * set ASYNC_TX_XOR_DROP_DST and ASYNC_TX_XOR_ZERO_DST
 	 * for the synchronous xor case
 	 */
-	flags = ASYNC_TX_ACK |
-		(prexor ? ASYNC_TX_XOR_DROP_DST : ASYNC_TX_XOR_ZERO_DST);
+	last_stripe = !head_sh->batch_head ||
+		list_first_entry(&sh->batch_list,
+				 struct stripe_head, batch_list) == head_sh;
+	if (last_stripe) {
+		flags = ASYNC_TX_ACK |
+			(prexor ? ASYNC_TX_XOR_DROP_DST : ASYNC_TX_XOR_ZERO_DST);
 
-	atomic_inc(&sh->count);
+		atomic_inc(&head_sh->count);
+		init_async_submit(&submit, flags, tx, ops_complete_reconstruct, head_sh,
+				  to_addr_conv(sh, percpu, j));
+	} else {
+		flags = prexor ? ASYNC_TX_XOR_DROP_DST : ASYNC_TX_XOR_ZERO_DST;
+		init_async_submit(&submit, flags, tx, NULL, NULL,
+				  to_addr_conv(sh, percpu, j));
+	}
 
-	init_async_submit(&submit, flags, tx, ops_complete_reconstruct, sh,
-			  to_addr_conv(sh, percpu));
 	if (unlikely(count == 1))
 		tx = async_memcpy(xor_dest, xor_srcs[0], 0, 0, STRIPE_SIZE, &submit);
 	else
 		tx = async_xor(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, &submit);
+	if (!last_stripe) {
+		j++;
+		sh = list_first_entry(&sh->batch_list, struct stripe_head,
+				      batch_list);
+		goto again;
+	}
 }
 
 static void
···
 		     struct dma_async_tx_descriptor *tx)
 {
 	struct async_submit_ctl submit;
-	struct page **blocks = percpu->scribble;
-	int count, i;
+	struct page **blocks;
+	int count, i, j = 0;
+	struct stripe_head *head_sh = sh;
+	int last_stripe;
+	int synflags;
+	unsigned long txflags;
 
 	pr_debug("%s: stripe %llu\n", __func__, (unsigned long long)sh->sector);
 
···
 		return;
 	}
 
-	count = set_syndrome_sources(blocks, sh);
+again:
+	blocks = to_addr_page(percpu, j);
 
-	atomic_inc(&sh->count);
+	if (sh->reconstruct_state == reconstruct_state_prexor_drain_run) {
+		synflags = SYNDROME_SRC_WRITTEN;
+		txflags = ASYNC_TX_ACK | ASYNC_TX_PQ_XOR_DST;
+	} else {
+		synflags = SYNDROME_SRC_ALL;
+		txflags = ASYNC_TX_ACK;
+	}
 
-	init_async_submit(&submit, ASYNC_TX_ACK, tx, ops_complete_reconstruct,
-			  sh, to_addr_conv(sh, percpu));
+	count = set_syndrome_sources(blocks, sh, synflags);
+	last_stripe = !head_sh->batch_head ||
+		list_first_entry(&sh->batch_list,
+				 struct stripe_head, batch_list) == head_sh;
+
+	if (last_stripe) {
+		atomic_inc(&head_sh->count);
+		init_async_submit(&submit, txflags, tx, ops_complete_reconstruct,
+				  head_sh, to_addr_conv(sh, percpu, j));
+	} else
+		init_async_submit(&submit, 0, tx, NULL, NULL,
+				  to_addr_conv(sh, percpu, j));
 	async_gen_syndrome(blocks, 0, count+2, STRIPE_SIZE,  &submit);
+	if (!last_stripe) {
+		j++;
+		sh = list_first_entry(&sh->batch_list, struct stripe_head,
+				      batch_list);
+		goto again;
+	}
 }
 
 static void ops_complete_check(void *stripe_head_ref)
···
 	int pd_idx = sh->pd_idx;
 	int qd_idx = sh->qd_idx;
 	struct page *xor_dest;
-	struct page **xor_srcs = percpu->scribble;
+	struct page **xor_srcs = to_addr_page(percpu, 0);
 	struct dma_async_tx_descriptor *tx;
 	struct async_submit_ctl submit;
 	int count;
···
 	pr_debug("%s: stripe %llu\n", __func__,
 		(unsigned long long)sh->sector);
 
+	BUG_ON(sh->batch_head);
 	count = 0;
 	xor_dest = sh->dev[pd_idx].page;
 	xor_srcs[count++] = xor_dest;
···
 	}
 
 	init_async_submit(&submit, 0, NULL, NULL, NULL,
-			  to_addr_conv(sh, percpu));
+			  to_addr_conv(sh, percpu, 0));
 	tx = async_xor_val(xor_dest, xor_srcs, 0, count, STRIPE_SIZE,
 			   &sh->ops.zero_sum_result, &submit);
 
···
 
 static void ops_run_check_pq(struct stripe_head *sh, struct raid5_percpu *percpu, int checkp)
 {
-	struct page **srcs = percpu->scribble;
+	struct page **srcs = to_addr_page(percpu, 0);
 	struct async_submit_ctl submit;
 	int count;
 
 	pr_debug("%s: stripe %llu checkp: %d\n", __func__,
 		(unsigned long long)sh->sector, checkp);
 
-	count = set_syndrome_sources(srcs, sh);
+	BUG_ON(sh->batch_head);
+	count = set_syndrome_sources(srcs, sh, SYNDROME_SRC_ALL);
 	if (!checkp)
 		srcs[count] = NULL;
 
 	atomic_inc(&sh->count);
 	init_async_submit(&submit, ASYNC_TX_ACK, NULL, ops_complete_check,
-			  sh, to_addr_conv(sh, percpu));
+			  sh, to_addr_conv(sh, percpu, 0));
 	async_syndrome_val(srcs, 0, count+2, STRIPE_SIZE,
 			   &sh->ops.zero_sum_result, percpu->spare_page, &submit);
 }
···
 			async_tx_ack(tx);
 	}
 
-	if (test_bit(STRIPE_OP_PREXOR, &ops_request))
-		tx = ops_run_prexor(sh, percpu, tx);
+	if (test_bit(STRIPE_OP_PREXOR, &ops_request)) {
+		if (level < 6)
+			tx = ops_run_prexor5(sh, percpu, tx);
+		else
+			tx = ops_run_prexor6(sh, percpu, tx);
+	}
 
 	if (test_bit(STRIPE_OP_BIODRAIN, &ops_request)) {
 		tx = ops_run_biodrain(sh, tx);
···
 			BUG();
 	}
 
-	if (overlap_clear)
+	if (overlap_clear && !sh->batch_head)
 		for (i = disks; i--; ) {
 			struct r5dev *dev = &sh->dev[i];
 			if (test_and_clear_bit(R5_Overlap, &dev->flags))
···
 	put_cpu();
 }
 
-static int grow_one_stripe(struct r5conf *conf, int hash)
+static int grow_one_stripe(struct r5conf *conf, gfp_t gfp)
 {
 	struct stripe_head *sh;
-	sh = kmem_cache_zalloc(conf->slab_cache, GFP_KERNEL);
+	sh = kmem_cache_zalloc(conf->slab_cache, gfp);
 	if (!sh)
 		return 0;
 
···
 
 	spin_lock_init(&sh->stripe_lock);
 
-	if (grow_buffers(sh)) {
+	if (grow_buffers(sh, gfp)) {
 		shrink_buffers(sh);
 		kmem_cache_free(conf->slab_cache, sh);
 		return 0;
 	}
-	sh->hash_lock_index = hash;
+	sh->hash_lock_index =
+		conf->max_nr_stripes % NR_STRIPE_HASH_LOCKS;
 	/* we just created an active stripe so... */
 	atomic_set(&sh->count, 1);
 	atomic_inc(&conf->active_stripes);
 	INIT_LIST_HEAD(&sh->lru);
+
+	spin_lock_init(&sh->batch_lock);
+	INIT_LIST_HEAD(&sh->batch_list);
+	sh->batch_head = NULL;
 	release_stripe(sh);
+	conf->max_nr_stripes++;
 	return 1;
 }
 
···
 {
 	struct kmem_cache *sc;
 	int devs = max(conf->raid_disks, conf->previous_raid_disks);
-	int hash;
 
 	if (conf->mddev->gendisk)
 		sprintf(conf->cache_name[0],
···
 		return 1;
 	conf->slab_cache = sc;
 	conf->pool_size = devs;
-	hash = conf->max_nr_stripes % NR_STRIPE_HASH_LOCKS;
-	while (num--) {
-		if (!grow_one_stripe(conf, hash))
+	while (num--)
+		if (!grow_one_stripe(conf, GFP_KERNEL))
 			return 1;
-		conf->max_nr_stripes++;
-		hash = (hash + 1) % NR_STRIPE_HASH_LOCKS;
-	}
+
 	return 0;
 }
 
···
  * calculate over all devices (not just the data blocks), using zeros in place
  * of the P and Q blocks.
  */
-static size_t scribble_len(int num)
+static struct flex_array *scribble_alloc(int num, int cnt, gfp_t flags)
 {
+	struct flex_array *ret;
 	size_t len;
 
 	len = sizeof(struct page *) * (num+2) + sizeof(addr_conv_t) * (num+2);
-
-	return len;
+	ret = flex_array_alloc(len, cnt, flags);
+	if (!ret)
+		return NULL;
+	/* always prealloc all elements, so no locking is required */
+	if (flex_array_prealloc(ret, 0, cnt, flags)) {
+		flex_array_free(ret);
+		return NULL;
+	}
+	return ret;
 }
 
 static int resize_stripes(struct r5conf *conf, int newsize)
···
 		err = -ENOMEM;
 
 	get_online_cpus();
-	conf->scribble_len = scribble_len(newsize);
 	for_each_present_cpu(cpu) {
 		struct raid5_percpu *percpu;
-		void *scribble;
+		struct flex_array *scribble;
 
 		percpu = per_cpu_ptr(conf->percpu, cpu);
-		scribble = kmalloc(conf->scribble_len, GFP_NOIO);
+		scribble = scribble_alloc(newsize, conf->chunk_sectors /
+			STRIPE_SECTORS, GFP_NOIO);
 
 		if (scribble) {
-			kfree(percpu->scribble);
+			flex_array_free(percpu->scribble);
 			percpu->scribble = scribble;
 		} else {
 			err = -ENOMEM;
···
 	return err;
 }
 
-static int drop_one_stripe(struct r5conf *conf, int hash)
+static int drop_one_stripe(struct r5conf *conf)
 {
 	struct stripe_head *sh;
+	int hash = (conf->max_nr_stripes - 1) % NR_STRIPE_HASH_LOCKS;
 
 	spin_lock_irq(conf->hash_locks + hash);
 	sh = get_free_stripe(conf, hash);
···
 	shrink_buffers(sh);
 	kmem_cache_free(conf->slab_cache, sh);
 	atomic_dec(&conf->active_stripes);
+	conf->max_nr_stripes--;
 	return 1;
 }
 
 static void shrink_stripes(struct r5conf *conf)
 {
-	int hash;
-	for (hash = 0; hash < NR_STRIPE_HASH_LOCKS; hash++)
-		while (drop_one_stripe(conf, hash))
-			;
+	while (conf->max_nr_stripes &&
+	       drop_one_stripe(conf))
+		;
 
 	if (conf->slab_cache)
 		kmem_cache_destroy(conf->slab_cache);
···
 	}
 	rdev_dec_pending(rdev, conf->mddev);
 
+	if (sh->batch_head && !uptodate)
+		set_bit(STRIPE_BATCH_ERR, &sh->batch_head->state);
+
 	if (!test_and_clear_bit(R5_DOUBLE_LOCKED, &sh->dev[i].flags))
 		clear_bit(R5_LOCKED, &sh->dev[i].flags);
 	set_bit(STRIPE_HANDLE, &sh->state);
 	release_stripe(sh);
+
+	if (sh->batch_head && sh != sh->batch_head)
+		release_stripe(sh->batch_head);
 }
 
 static sector_t compute_blocknr(struct stripe_head *sh, int i, int previous);
···
 schedule_reconstruction(struct stripe_head *sh, struct stripe_head_state *s,
 			 int rcw, int expand)
 {
-	int i, pd_idx = sh->pd_idx, disks = sh->disks;
+	int i, pd_idx = sh->pd_idx, qd_idx = sh->qd_idx, disks = sh->disks;
 	struct r5conf *conf = sh->raid_conf;
 	int level = conf->level;
 
···
 			if (!test_and_set_bit(STRIPE_FULL_WRITE, &sh->state))
 				atomic_inc(&conf->pending_full_writes);
 	} else {
-		BUG_ON(level == 6);
 		BUG_ON(!(test_bit(R5_UPTODATE, &sh->dev[pd_idx].flags) ||
 			test_bit(R5_Wantcompute, &sh->dev[pd_idx].flags)));
+		BUG_ON(level == 6 &&
+			(!(test_bit(R5_UPTODATE, &sh->dev[qd_idx].flags) ||
+			   test_bit(R5_Wantcompute, &sh->dev[qd_idx].flags))));
 
 		for (i = disks; i--; ) {
 			struct r5dev *dev = &sh->dev[i];
-			if (i == pd_idx)
+			if (i == pd_idx || i == qd_idx)
 				continue;
 
 			if (dev->towrite &&
···
  * toread/towrite point to the first in a chain.
  * The bi_next chain must be in order.
  */
-static int add_stripe_bio(struct stripe_head *sh, struct bio *bi, int dd_idx, int forwrite)
+static int add_stripe_bio(struct stripe_head *sh, struct bio *bi, int dd_idx,
+			  int forwrite, int previous)
 {
 	struct bio **bip;
 	struct r5conf *conf = sh->raid_conf;
···
 	 * protect it.
 	 */
 	spin_lock_irq(&sh->stripe_lock);
+	/* Don't allow new IO added to stripes in batch list */
+	if (sh->batch_head)
+		goto overlap;
 	if (forwrite) {
 		bip = &sh->dev[dd_idx].towrite;
 		if (*bip == NULL)
···
 	}
 	if (*bip && (*bip)->bi_iter.bi_sector < bio_end_sector(bi))
 		goto overlap;
+
+	if (!forwrite || previous)
+		clear_bit(STRIPE_BATCH_READY, &sh->state);
 
 	BUG_ON(*bip && bi->bi_next && (*bip) != bi->bi_next);
 	if (*bip)
···
 				sector = bio_end_sector(bi);
 		}
 		if (sector >= sh->dev[dd_idx].sector + STRIPE_SECTORS)
-			set_bit(R5_OVERWRITE, &sh->dev[dd_idx].flags);
+			if (!test_and_set_bit(R5_OVERWRITE, &sh->dev[dd_idx].flags))
+				sh->overwrite_disks++;
 	}
 
 	pr_debug("added bi b#%llu to stripe s#%llu, disk %d.\n",
···
 		sh->bm_seq = conf->seq_flush+1;
 		set_bit(STRIPE_BIT_DELAY, &sh->state);
 	}
+
+	if (stripe_can_batch(sh))
+		stripe_add_to_batch_list(conf, sh);
 	return 1;
 
  overlap:
···
 				struct bio **return_bi)
 {
 	int i;
+	BUG_ON(sh->batch_head);
 	for (i = disks; i--; ) {
 		struct bio *bi;
 		int bitmap_end = 0;
···
 		/* fail all writes first */
 		bi = sh->dev[i].towrite;
 		sh->dev[i].towrite = NULL;
+		sh->overwrite_disks = 0;
 		spin_unlock_irq(&sh->stripe_lock);
 		if (bi)
 			bitmap_end = 1;
···
 	int abort = 0;
 	int i;
 
+	BUG_ON(sh->batch_head);
 	clear_bit(STRIPE_SYNCING, &sh->state);
 	if (test_and_clear_bit(R5_Overlap, &sh->dev[sh->pd_idx].flags))
 		wake_up(&conf->wait_for_overlap);
···
 {
 	int i;
 
+	BUG_ON(sh->batch_head);
 	/* look for blocks to read/compute, skip this if a compute
 	 * is already in flight, or if the stripe contents are in the
 	 * midst of changing due to a write
···
 	int i;
 	struct r5dev *dev;
 	int discard_pending = 0;
+	struct stripe_head *head_sh = sh;
+	bool do_endio = false;
+	int wakeup_nr = 0;
 
 	for (i = disks; i--; )
 		if (sh->dev[i].written) {
···
 					clear_bit(R5_UPTODATE, &dev->flags);
 				if (test_and_clear_bit(R5_SkipCopy, &dev->flags)) {
 					WARN_ON(test_bit(R5_UPTODATE, &dev->flags));
-					dev->page = dev->orig_page;
 				}
+				do_endio = true;
+
+returnbi:
+				dev->page = dev->orig_page;
 				wbi = dev->written;
 				dev->written = NULL;
 				while (wbi && wbi->bi_iter.bi_sector <
···
 						STRIPE_SECTORS,
 					 !test_bit(STRIPE_DEGRADED, &sh->state),
 						0);
+				if (head_sh->batch_head) {
+					sh = list_first_entry(&sh->batch_list,
+							      struct stripe_head,
+							      batch_list);
+					if (sh != head_sh) {
+						dev = &sh->dev[i];
+						goto returnbi;
+					}
+				}
+				sh = head_sh;
+				dev = &sh->dev[i];
 			} else if (test_bit(R5_Discard, &dev->flags))
 				discard_pending = 1;
 			WARN_ON(test_bit(R5_SkipCopy, &dev->flags));
···
 		 * will be reinitialized
 		 */
 		spin_lock_irq(&conf->device_lock);
+unhash:
 		remove_hash(sh);
+		if (head_sh->batch_head) {
+			sh = list_first_entry(&sh->batch_list,
+					      struct stripe_head, batch_list);
+			if (sh != head_sh)
+					goto unhash;
+		}
 		spin_unlock_irq(&conf->device_lock);
+		sh = head_sh;
+
 		if (test_bit(STRIPE_SYNC_REQUESTED, &sh->state))
 			set_bit(STRIPE_HANDLE, &sh->state);
 
···
 	if (test_and_clear_bit(STRIPE_FULL_WRITE, &sh->state))
 		if (atomic_dec_and_test(&conf->pending_full_writes))
 			md_wakeup_thread(conf->mddev->thread);
+
+	if (!head_sh->batch_head || !do_endio)
+		return;
+	for (i = 0; i < head_sh->disks; i++) {
+		if (test_and_clear_bit(R5_Overlap, &head_sh->dev[i].flags))
+			wakeup_nr++;
+	}
+	while (!list_empty(&head_sh->batch_list)) {
+		int i;
+		sh = list_first_entry(&head_sh->batch_list,
+				      struct stripe_head, batch_list);
+		list_del_init(&sh->batch_list);
+
+		set_mask_bits(&sh->state, ~STRIPE_EXPAND_SYNC_FLAG,
+			      head_sh->state & ~((1 << STRIPE_ACTIVE) |
+						 (1 << STRIPE_PREREAD_ACTIVE) |
+						 STRIPE_EXPAND_SYNC_FLAG));
+		sh->check_state = head_sh->check_state;
+		sh->reconstruct_state = head_sh->reconstruct_state;
+		for (i = 0; i < sh->disks; i++) {
+			if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags))
+				wakeup_nr++;
+			sh->dev[i].flags = head_sh->dev[i].flags;
+		}
+
+		spin_lock_irq(&sh->stripe_lock);
+		sh->batch_head = NULL;
+		spin_unlock_irq(&sh->stripe_lock);
+		if (sh->state & STRIPE_EXPAND_SYNC_FLAG)
+			set_bit(STRIPE_HANDLE, &sh->state);
+		release_stripe(sh);
+	}
+
+	spin_lock_irq(&head_sh->stripe_lock);
+	head_sh->batch_head = NULL;
+	spin_unlock_irq(&head_sh->stripe_lock);
+	wake_up_nr(&conf->wait_for_overlap, wakeup_nr);
+	if (head_sh->state & STRIPE_EXPAND_SYNC_FLAG)
+		set_bit(STRIPE_HANDLE, &head_sh->state);
 }
 
 static void handle_stripe_dirtying(struct r5conf *conf,
···
 	int rmw = 0, rcw = 0, i;
 	sector_t recovery_cp = conf->mddev->recovery_cp;
 
-	/* RAID6 requires 'rcw' in current implementation.
-	 * Otherwise, check whether resync is now happening or should start.
+	/* Check whether resync is now happening or should start.
 	 * If yes, then the array is dirty (after unclean shutdown or
 	 * initial creation), so parity in some stripes might be inconsistent.
 	 * In this case, we need to always do reconstruct-write, to ensure
 	 * that in case of drive failure or read-error correction, we
 	 * generate correct data from the parity.
 	 */
-	if (conf->max_degraded == 2 ||
+	if (conf->rmw_level == PARITY_DISABLE_RMW ||
 	    (recovery_cp < MaxSector && sh->sector >= recovery_cp &&
 	     s->failed == 0)) {
 		/* Calculate the real rcw later - for now make it
 		 * look like rcw is cheaper
 		 */
 		rcw = 1; rmw = 2;
-		pr_debug("force RCW max_degraded=%u, recovery_cp=%llu sh->sector=%llu\n",
-			 conf->max_degraded, (unsigned long long)recovery_cp,
+		pr_debug("force RCW rmw_level=%u, recovery_cp=%llu sh->sector=%llu\n",
+			 conf->rmw_level, (unsigned long long)recovery_cp,
 			 (unsigned long long)sh->sector);
 	} else for (i = disks; i--; ) {
 		/* would I have to read this buffer for read_modify_write */
 		struct r5dev *dev = &sh->dev[i];
-		if ((dev->towrite || i == sh->pd_idx) &&
+		if ((dev->towrite || i == sh->pd_idx || i == sh->qd_idx) &&
 		    !test_bit(R5_LOCKED, &dev->flags) &&
 		    !(test_bit(R5_UPTODATE, &dev->flags) ||
 		      test_bit(R5_Wantcompute, &dev->flags))) {
···
 				rmw += 2*disks;  /* cannot read it */
 		}
 		/* Would I have to read this buffer for reconstruct_write */
-		if (!test_bit(R5_OVERWRITE, &dev->flags) && i != sh->pd_idx &&
+		if (!test_bit(R5_OVERWRITE, &dev->flags) &&
+		    i != sh->pd_idx && i != sh->qd_idx &&
 		    !test_bit(R5_LOCKED, &dev->flags) &&
 		    !(test_bit(R5_UPTODATE, &dev->flags) ||
 		    test_bit(R5_Wantcompute, &dev->flags))) {
···
 	pr_debug("for sector %llu, rmw=%d rcw=%d\n",
 		(unsigned long long)sh->sector, rmw, rcw);
 	set_bit(STRIPE_HANDLE, &sh->state);
-	if (rmw < rcw && rmw > 0) {
+	if ((rmw < rcw || (rmw == rcw && conf->rmw_level == PARITY_ENABLE_RMW)) && rmw > 0) {
 		/* prefer read-modify-write, but need to get some data */
 		if (conf->mddev->queue)
 			blk_add_trace_msg(conf->mddev->queue,
···
 					  (unsigned long long)sh->sector, rmw);
 		for (i = disks; i--; ) {
 			struct r5dev *dev = &sh->dev[i];
-			if ((dev->towrite || i == sh->pd_idx) &&
+			if ((dev->towrite || i == sh->pd_idx || i == sh->qd_idx) &&
 			    !test_bit(R5_LOCKED, &dev->flags) &&
 			    !(test_bit(R5_UPTODATE, &dev->flags) ||
 			    test_bit(R5_Wantcompute, &dev->flags)) &&
···
 			}
 		}
 	}
-	if (rcw <= rmw && rcw > 0) {
+	if ((rcw < rmw || (rcw == rmw && conf->rmw_level != PARITY_ENABLE_RMW)) && rcw > 0) {
 		/* want reconstruct write, but need to get some data */
 		int qread =0;
 		rcw = 0;
···
 {
 	struct r5dev *dev = NULL;
 
+	BUG_ON(sh->batch_head);
 	set_bit(STRIPE_HANDLE, &sh->state);
 
 	switch (sh->check_state) {
···
 	int qd_idx = sh->qd_idx;
 	struct r5dev *dev;
 
+	BUG_ON(sh->batch_head);
 	set_bit(STRIPE_HANDLE, &sh->state);
 
 	BUG_ON(s->failed > 2);
···
 	 * copy some of them into a target stripe for expand.
 	 */
 	struct dma_async_tx_descriptor *tx = NULL;
+	BUG_ON(sh->batch_head);
 	clear_bit(STRIPE_EXPAND_SOURCE, &sh->state);
 	for (i = 0; i < sh->disks; i++)
 		if (i != sh->pd_idx && i != sh->qd_idx) {
···
 
 	memset(s, 0, sizeof(*s));
 
-	s->expanding = test_bit(STRIPE_EXPAND_SOURCE, &sh->state);
-	s->expanded = test_bit(STRIPE_EXPAND_READY, &sh->state);
+	s->expanding = test_bit(STRIPE_EXPAND_SOURCE, &sh->state) && !sh->batch_head;
+	s->expanded = test_bit(STRIPE_EXPAND_READY, &sh->state) && !sh->batch_head;
 	s->failed_num[0] = -1;
 	s->failed_num[1] = -1;
 
···
 	rcu_read_unlock();
 }
 
+static int clear_batch_ready(struct stripe_head *sh)
+{
+	struct stripe_head *tmp;
+	if (!test_and_clear_bit(STRIPE_BATCH_READY, &sh->state))
+		return 0;
+	spin_lock(&sh->stripe_lock);
+	if (!sh->batch_head) {
+		spin_unlock(&sh->stripe_lock);
+		return 0;
+	}
+
+	/*
+	 * this stripe could be added to a batch list before we check
+	 * BATCH_READY, skips it
+	 */
+	if (sh->batch_head != sh) {
+		spin_unlock(&sh->stripe_lock);
+		return 1;
+	}
+	spin_lock(&sh->batch_lock);
+	list_for_each_entry(tmp, &sh->batch_list, batch_list)
+		clear_bit(STRIPE_BATCH_READY, &tmp->state);
+	spin_unlock(&sh->batch_lock);
+	spin_unlock(&sh->stripe_lock);
+
+	/*
+	 * BATCH_READY is cleared, no new stripes can be added.
+	 * batch_list can be accessed without lock
+	 */
+	return 0;
+}
+
+static void check_break_stripe_batch_list(struct stripe_head *sh)
+{
+	struct stripe_head *head_sh, *next;
+	int i;
+
+	if (!test_and_clear_bit(STRIPE_BATCH_ERR, &sh->state))
+		return;
+
+	head_sh = sh;
+	do {
+		sh = list_first_entry(&sh->batch_list,
+				      struct stripe_head, batch_list);
+		BUG_ON(sh == head_sh);
+	} while (!test_bit(STRIPE_DEGRADED, &sh->state));
+
+	while (sh != head_sh) {
+		next = list_first_entry(&sh->batch_list,
+					struct stripe_head, batch_list);
+		list_del_init(&sh->batch_list);
+
+		set_mask_bits(&sh->state, ~STRIPE_EXPAND_SYNC_FLAG,
+			      head_sh->state & ~((1 << STRIPE_ACTIVE) |
+						 (1 << STRIPE_PREREAD_ACTIVE) |
+						 (1 << STRIPE_DEGRADED) |
+						 STRIPE_EXPAND_SYNC_FLAG));
+		sh->check_state = head_sh->check_state;
+		sh->reconstruct_state = head_sh->reconstruct_state;
+		for (i = 0; i < sh->disks; i++)
+			sh->dev[i].flags = head_sh->dev[i].flags &
+				(~((1 << R5_WriteError) | (1 << R5_Overlap)));
+
+		spin_lock_irq(&sh->stripe_lock);
+		sh->batch_head = NULL;
+		spin_unlock_irq(&sh->stripe_lock);
+
+		set_bit(STRIPE_HANDLE, &sh->state);
+		release_stripe(sh);
+
+		sh = next;
+	}
+}
+
 static void handle_stripe(struct stripe_head *sh)
 {
 	struct stripe_head_state s;
···
 		return;
 	}
 
-	if (test_bit(STRIPE_SYNC_REQUESTED, &sh->state)) {
+	if (clear_batch_ready(sh) ) {
+		clear_bit_unlock(STRIPE_ACTIVE, &sh->state);
+		return;
+	}
+
+	check_break_stripe_batch_list(sh);
+
+	if (test_bit(STRIPE_SYNC_REQUESTED, &sh->state) && !sh->batch_head) {
 		spin_lock(&sh->stripe_lock);
 		/* Cannot process 'sync' concurrently with 'discard' */
 		if (!test_bit(STRIPE_DISCARD, &sh->state) &&
···
 	 * how busy the stripe_cache is
 	 */
 
-	if (conf->inactive_blocked)
+	if (test_bit(R5_INACTIVE_BLOCKED, &conf->cache_state))
 		return 1;
 	if (conf->quiesce)
 		return 1;
···
 	unsigned int chunk_sectors = mddev->chunk_sectors;
 	unsigned int bio_sectors = bvm->bi_size >> 9;
 
-	if ((bvm->bi_rw & 1) == WRITE)
-		return biovec->bv_len; /* always allow writes to be mergeable */
+	/*
+	 * always allow writes to be mergeable, read as well if array
+	 * is degraded as we'll go through stripe cache anyway.
+	 */
+	if ((bvm->bi_rw & 1) == WRITE || mddev->degraded)
+		return biovec->bv_len;
 
 	if (mddev->new_chunk_sectors < mddev->chunk_sectors)
 		chunk_sectors = mddev->new_chunk_sectors;
···
 		}
 		set_bit(STRIPE_DISCARD, &sh->state);
 		finish_wait(&conf->wait_for_overlap, &w);
+		sh->overwrite_disks = 0;
 		for (d = 0; d < conf->raid_disks; d++) {
 			if (d == sh->pd_idx || d == sh->qd_idx)
 				continue;
 			sh->dev[d].towrite = bi;
 			set_bit(R5_OVERWRITE, &sh->dev[d].flags);
 			raid5_inc_bi_active_stripes(bi);
+			sh->overwrite_disks++;
 		}
 		spin_unlock_irq(&sh->stripe_lock);
 		if (conf->mddev->bitmap) {
···
 
 	md_write_start(mddev, bi);
 
-	if (rw == READ &&
+	/*
+	 * If array is degraded, better not do chunk aligned read because
+	 * later we might have to read it again in order to reconstruct
+	 * data on failed drives.
+	 */
+	if (rw == READ && mddev->degraded == 0 &&
 	     mddev->reshape_position == MaxSector &&
 	     chunk_aligned_read(mddev,bi))
 		return;
···
 			}
 
 			if (test_bit(STRIPE_EXPANDING, &sh->state) ||
-			    !add_stripe_bio(sh, bi, dd_idx, rw)) {
+			    !add_stripe_bio(sh, bi, dd_idx, rw, previous)) {
 				/* Stripe is busy expanding or
 				 * add failed due to overlap.  Flush everything
 				 * and wait a while
···
 			}
 			set_bit(STRIPE_HANDLE, &sh->state);
 			clear_bit(STRIPE_DELAYED, &sh->state);
-			if ((bi->bi_rw & REQ_SYNC) &&
+			if ((!sh->batch_head || sh == sh->batch_head) &&
+			    (bi->bi_rw & REQ_SYNC) &&
 			    !test_and_set_bit(STRIPE_PREREAD_ACTIVE, &sh->state))
 				atomic_inc(&conf->preread_active_stripes);
 			release_stripe_plug(mddev, sh);
···
 	return reshape_sectors;
 }
 
-/* FIXME go_faster isn't used */
-static inline sector_t sync_request(struct mddev *mddev, sector_t sector_nr, int *skipped, int go_faster)
+static inline sector_t sync_request(struct mddev *mddev, sector_t sector_nr, int *skipped)
 {
 	struct r5conf *conf = mddev->private;
 	struct stripe_head *sh;
···
 			return handled;
 		}
 
-		if (!add_stripe_bio(sh, raid_bio, dd_idx, 0)) {
+		if (!add_stripe_bio(sh, raid_bio, dd_idx, 0, 0)) {
 			release_stripe(sh);
 			raid5_set_bi_processed_stripes(raid_bio, scnt);
 			conf->retry_read_aligned = raid_bio;
···
 		int batch_size, released;
 
 		released = release_stripe_list(conf, conf->temp_inactive_list);
+		if (released)
+			clear_bit(R5_DID_ALLOC, &conf->cache_state);
 
 		if (
 		    !list_empty(&conf->bitmap_list)) {
···
 	pr_debug("%d stripes handled\n", handled);
 
 	spin_unlock_irq(&conf->device_lock);
+	if (test_and_clear_bit(R5_ALLOC_MORE, &conf->cache_state)) {
+		grow_one_stripe(conf, __GFP_NOWARN);
+		/* Set flag even if allocation failed.  This helps
+		 * slow down allocation requests when mem is short
+		 */
+		set_bit(R5_DID_ALLOC, &conf->cache_state);
+	}
 
 	async_tx_issue_pending_all();
 	blk_finish_plug(&plug);
···
 	spin_lock(&mddev->lock);
 	conf = mddev->private;
 	if (conf)
-		ret = sprintf(page, "%d\n", conf->max_nr_stripes);
+		ret = sprintf(page, "%d\n", conf->min_nr_stripes);
 	spin_unlock(&mddev->lock);
 	return ret;
 }
···
 {
 	struct r5conf *conf = mddev->private;
 	int err;
-	int hash;
 
 	if (size <= 16 || size > 32768)
 		return -EINVAL;
-	hash = (conf->max_nr_stripes - 1) % NR_STRIPE_HASH_LOCKS;
-	while (size < conf->max_nr_stripes) {
-		if (drop_one_stripe(conf, hash))
-			conf->max_nr_stripes--;
-		else
-			break;
-		hash--;
-		if (hash < 0)
-			hash = NR_STRIPE_HASH_LOCKS - 1;
-	}
+
+	conf->min_nr_stripes = size;
+	while (size < conf->max_nr_stripes &&
+	       drop_one_stripe(conf))
+		;
+
+
 	err = md_allow_write(mddev);
 	if (err)
 		return err;
-	hash = conf->max_nr_stripes % NR_STRIPE_HASH_LOCKS;
-	while (size > conf->max_nr_stripes) {
-		if (grow_one_stripe(conf, hash))
-			conf->max_nr_stripes++;
-		else break;
-		hash = (hash + 1) % NR_STRIPE_HASH_LOCKS;
-	}
+
+	while (size > conf->max_nr_stripes)
+		if (!grow_one_stripe(conf, GFP_KERNEL))
+			break;
+
 	return 0;
 }
 EXPORT_SYMBOL(raid5_set_cache_size);
···
 				raid5_store_stripe_cache_size);
 
 static ssize_t
+raid5_show_rmw_level(struct mddev  *mddev, char *page)
+{
+	struct r5conf *conf = mddev->private;
+	if (conf)
+		return sprintf(page, "%d\n", conf->rmw_level);
+	else
+		return 0;
+}
+
+static ssize_t
+raid5_store_rmw_level(struct mddev  *mddev, const char *page, size_t len)
+{
+	struct r5conf *conf = mddev->private;
+	unsigned long new;
+
+	if (!conf)
+		return -ENODEV;
+
+	if (len >= PAGE_SIZE)
+		return -EINVAL;
+
+	if (kstrtoul(page, 10, &new))
+		return -EINVAL;
+
+	if (new != PARITY_DISABLE_RMW && !raid6_call.xor_syndrome)
+		return -EINVAL;
+
+	if (new != PARITY_DISABLE_RMW &&
+	    new != PARITY_ENABLE_RMW &&
+	    new != PARITY_PREFER_RMW)
+		return -EINVAL;
+
+	conf->rmw_level = new;
+	return len;
+}
+
+static struct md_sysfs_entry
+raid5_rmw_level = __ATTR(rmw_level, S_IRUGO | S_IWUSR,
+			 raid5_show_rmw_level,
+			 raid5_store_rmw_level);
+
+
+static ssize_t
 raid5_show_preread_threshold(struct mddev *mddev, char *page)
 {
 	struct r5conf *conf;
···
 	conf = mddev->private;
 	if (!conf)
 		err = -ENODEV;
-	else if (new > conf->max_nr_stripes)
+	else if (new > conf->min_nr_stripes)
 		err = -EINVAL;
 	else
 		conf->bypass_threshold = new;
···
 	&raid5_preread_bypass_threshold.attr,
 	&raid5_group_thread_cnt.attr,
 	&raid5_skip_copy.attr,
+	&raid5_rmw_level.attr,
 	NULL,
 };
 static struct attribute_group raid5_attrs_group = {
···
 static void free_scratch_buffer(struct r5conf *conf, struct raid5_percpu *percpu)
 {
 	safe_put_page(percpu->spare_page);
-	kfree(percpu->scribble);
+	if (percpu->scribble)
+		flex_array_free(percpu->scribble);
 	percpu->spare_page = NULL;
 	percpu->scribble = NULL;
 }
···
 	if (conf->level == 6 && !percpu->spare_page)
 		percpu->spare_page = alloc_page(GFP_KERNEL);
 	if (!percpu->scribble)
-		percpu->scribble = kmalloc(conf->scribble_len, GFP_KERNEL);
+		percpu->scribble = scribble_alloc(max(conf->raid_disks,
+			conf->previous_raid_disks), conf->chunk_sectors /
+			STRIPE_SECTORS, GFP_KERNEL);
 
 	if (!percpu->scribble || (conf->level == 6 && !percpu->spare_page)) {
 		free_scratch_buffer(conf, percpu);
···
 
 static void free_conf(struct r5conf *conf)
 {
+	if (conf->shrinker.seeks)
+		unregister_shrinker(&conf->shrinker);
 	free_thread_groups(conf);
 	shrink_stripes(conf);
 	raid5_free_percpu(conf);
···
 	put_online_cpus();
 
 	return err;
+}
+
+static unsigned long raid5_cache_scan(struct shrinker *shrink,
+				      struct shrink_control *sc)
+{
+	struct r5conf *conf = container_of(shrink, struct r5conf, shrinker);
+	int ret = 0;
+	while (ret < sc->nr_to_scan) {
+		if (drop_one_stripe(conf) == 0)
+			return SHRINK_STOP;
+		ret++;
+	}
+	return ret;
+}
+
+static unsigned long raid5_cache_count(struct shrinker *shrink,
+				       struct shrink_control *sc)
+{
+	struct r5conf *conf = container_of(shrink, struct r5conf, shrinker);
+
+	if (conf->max_nr_stripes < conf->min_nr_stripes)
+		/* unlikely, but not impossible */
+		return 0;
+	return conf->max_nr_stripes - conf->min_nr_stripes;
 }
 
 static struct r5conf *setup_conf(struct mddev *mddev)
···
 	else
 		conf->previous_raid_disks = mddev->raid_disks - mddev->delta_disks;
 	max_disks = max(conf->raid_disks, conf->previous_raid_disks);
-	conf->scribble_len = scribble_len(max_disks);
 
 	conf->disks = kzalloc(max_disks * sizeof(struct disk_info),
 			      GFP_KERNEL);
···
 		INIT_LIST_HEAD(conf->temp_inactive_list + i);
 
 	conf->level = mddev->new_level;
+	conf->chunk_sectors = mddev->new_chunk_sectors;
 	if (raid5_alloc_percpu(conf) != 0)
 		goto abort;
 
···
 			conf->fullsync = 1;
 	}
 
-	conf->chunk_sectors = mddev->new_chunk_sectors;
 	conf->level = mddev->new_level;
-	if (conf->level == 6)
+	if (conf->level == 6) {
 		conf->max_degraded = 2;
-	else
+		if (raid6_call.xor_syndrome)
+			conf->rmw_level = PARITY_ENABLE_RMW;
+		else
+			conf->rmw_level = PARITY_DISABLE_RMW;
+	} else {
 		conf->max_degraded = 1;
+		conf->rmw_level = PARITY_ENABLE_RMW;
+	}
 	conf->algorithm = mddev->new_layout;
 	conf->reshape_progress = mddev->reshape_position;
 	if (conf->reshape_progress != MaxSector) {
···
 		conf->prev_algo = mddev->layout;
 	}
 
-	memory = conf->max_nr_stripes * (sizeof(struct stripe_head) +
+	conf->min_nr_stripes = NR_STRIPES;
+	memory = conf->min_nr_stripes * (sizeof(struct stripe_head) +
 		 max_disks * ((sizeof(struct bio) + PAGE_SIZE))) / 1024;
 	atomic_set(&conf->empty_inactive_list_nr, NR_STRIPE_HASH_LOCKS);
-	if (grow_stripes(conf, NR_STRIPES)) {
+	if (grow_stripes(conf, conf->min_nr_stripes)) {
 		printk(KERN_ERR
 		       "md/raid:%s: couldn't allocate %dkB for buffers\n",
 		       mdname(mddev), memory);
···
 	} else
 		printk(KERN_INFO "md/raid:%s: allocated %dkB\n",
 		       mdname(mddev), memory);
+	/*
+	 * Losing a stripe head costs more than the time to refill it,
+	 * it reduces the queue depth and so can hurt throughput.
+	 * So set it rather large, scaled by number of devices.
+	 */
+	conf->shrinker.seeks = DEFAULT_SEEKS * conf->raid_disks * 4;
+	conf->shrinker.scan_objects = raid5_cache_scan;
+	conf->shrinker.count_objects = raid5_cache_count;
+	conf->shrinker.batch = 128;
+	conf->shrinker.flags = 0;
+	register_shrinker(&conf->shrinker);
 
 	sprintf(pers_name, "raid%d", mddev->new_level);
 	conf->thread = md_register_thread(raid5d, mddev, pers_name);
···
 	 */
 	struct r5conf *conf = mddev->private;
 	if (((mddev->chunk_sectors << 9) / STRIPE_SIZE) * 4
-	    > conf->max_nr_stripes ||
+	    > conf->min_nr_stripes ||
 	    ((mddev->new_chunk_sectors << 9) / STRIPE_SIZE) * 4
-	    > conf->max_nr_stripes) {
+	    > conf->min_nr_stripes) {
 		printk(KERN_WARNING "md/raid:%s: reshape: not enough stripes.  Needed %lu\n",
 		       mdname(mddev),
 		       ((max(mddev->chunk_sectors, mddev->new_chunk_sectors) << 9)

drivers/md/raid5.h

···
 	atomic_t		count;	      /* nr of active thread/requests */
 	int			bm_seq;	/* sequence number for bitmap flushes */
 	int			disks;		/* disks in stripe */
+	int			overwrite_disks; /* total overwrite disks in stripe,
+						  * this is only checked when stripe
+						  * has STRIPE_BATCH_READY
+						  */
 	enum check_states	check_state;
 	enum reconstruct_states reconstruct_state;
 	spinlock_t		stripe_lock;
 	int			cpu;
 	struct r5worker_group	*group;
+
+	struct stripe_head	*batch_head; /* protected by stripe lock */
+	spinlock_t		batch_lock; /* only header's lock is useful */
+	struct list_head	batch_list; /* protected by head's batch lock*/
 	/**
 	 * struct stripe_operations
 	 * @target - STRIPE_OP_COMPUTE_BLK target
···
 	STRIPE_ON_UNPLUG_LIST,
 	STRIPE_DISCARD,
 	STRIPE_ON_RELEASE_LIST,
+	STRIPE_BATCH_READY,
+	STRIPE_BATCH_ERR,
 };
 
+#define STRIPE_EXPAND_SYNC_FLAG \
+	((1 << STRIPE_EXPAND_SOURCE) |\
+	(1 << STRIPE_EXPAND_READY) |\
+	(1 << STRIPE_EXPANDING) |\
+	(1 << STRIPE_SYNC_REQUESTED))
 /*
  * Operation request flags
  */
···
 	STRIPE_OP_BIODRAIN,
 	STRIPE_OP_RECONSTRUCT,
 	STRIPE_OP_CHECK,
+};
+
+/*
+ * RAID parity calculation preferences
+ */
+enum {
+	PARITY_DISABLE_RMW = 0,
+	PARITY_ENABLE_RMW,
+	PARITY_PREFER_RMW,
+};
+
+/*
+ * Pages requested from set_syndrome_sources()
+ */
+enum {
+	SYNDROME_SRC_ALL,
+	SYNDROME_SRC_WANT_DRAIN,
+	SYNDROME_SRC_WRITTEN,
 };
 /*
  * Plugging:
···
 	spinlock_t		hash_locks[NR_STRIPE_HASH_LOCKS];
 	struct mddev		*mddev;
 	int			chunk_sectors;
-	int			level, algorithm;
+	int			level, algorithm, rmw_level;
 	int			max_degraded;
 	int			raid_disks;
 	int			max_nr_stripes;
+	int			min_nr_stripes;
 
 	/* reshape_progress is the leading edge of a 'reshape'
 	 * It has value MaxSector when no reshape is happening
···
 	/* per cpu variables */
 	struct raid5_percpu {
 		struct page	*spare_page; /* Used when checking P/Q in raid6 */
-		void		*scribble;   /* space for constructing buffer
+		struct flex_array *scribble;   /* space for constructing buffer
 					      * lists and performing address
 					      * conversions
 					      */
 	} __percpu *percpu;
-	size_t			scribble_len; /* size of scribble region must be
-					       * associated with conf to handle
-					       * cpu hotplug while reshaping
-					       */
 #ifdef CONFIG_HOTPLUG_CPU
 	struct notifier_block	cpu_notify;
 #endif
···
 	struct llist_head	released_stripes;
 	wait_queue_head_t	wait_for_stripe;
 	wait_queue_head_t	wait_for_overlap;
-	int			inactive_blocked;	/* release of inactive stripes blocked,
-							 * waiting for 25% to be free
-							 */
+	unsigned long		cache_state;
+#define R5_INACTIVE_BLOCKED	1	/* release of inactive stripes blocked,
+					 * waiting for 25% to be free
+					 */
+#define R5_ALLOC_MORE		2	/* It might help to allocate another
+					 * stripe.
+					 */
+#define R5_DID_ALLOC		4	/* A stripe was allocated, don't allocate
+					 * more until at least one has been
+					 * released.  This avoids flooding
+					 * the cache.
+					 */
+	struct shrinker		shrinker;
 	int			pool_size; /* number of disks in stripeheads in pool */
 	spinlock_t		device_lock;
 	struct disk_info	*disks;
···
 	int			group_cnt;
 	int			worker_cnt_per_group;
 };
+
 
 /*
  * Our supported algorithms

include/linux/async_tx.h

···
  * dependency chain
  * @ASYNC_TX_FENCE: specify that the next operation in the dependency
  * chain uses this operation's result as an input
+ * @ASYNC_TX_PQ_XOR_DST: do not overwrite the syndrome but XOR it with the
+ * input data. Required for rmw case.
  */
 enum async_tx_flags {
 	ASYNC_TX_XOR_ZERO_DST	 = (1 << 0),
 	ASYNC_TX_XOR_DROP_DST	 = (1 << 1),
 	ASYNC_TX_ACK		 = (1 << 2),
 	ASYNC_TX_FENCE		 = (1 << 3),
+	ASYNC_TX_PQ_XOR_DST	 = (1 << 4),
 };
 
 /**

include/linux/raid/pq.h

···
 /* Routine choices */
 struct raid6_calls {
 	void (*gen_syndrome)(int, size_t, void **);
+	void (*xor_syndrome)(int, int, int, size_t, void **);
 	int  (*valid)(void);	/* Returns 1 if this routine set is usable */
 	const char *name;	/* Name of this routine set */
 	int prefer;		/* Has special performance attribute */

include/uapi/linux/raid/md_p.h

···
 #define MD_DISK_ACTIVE		1 /* disk is running or spare disk */
 #define MD_DISK_SYNC		2 /* disk is in sync with the raid set */
 #define MD_DISK_REMOVED		3 /* disk is in sync with the raid set */
+#define MD_DISK_CLUSTER_ADD     4 /* Initiate a disk add across the cluster
+				   * For clustered enviroments only.
+				   */
+#define MD_DISK_CANDIDATE	5 /* disk is added as spare (local) until confirmed
+				   * For clustered enviroments only.
+				   */
 
 #define	MD_DISK_WRITEMOSTLY	9 /* disk is "write-mostly" is RAID1 config.
 				   * read requests will only be sent here in
···
 #define MD_SB_CLEAN		0
 #define MD_SB_ERRORS		1
 
+#define	MD_SB_CLUSTERED		5 /* MD is clustered */
 #define	MD_SB_BITMAP_PRESENT	8 /* bitmap may be present nearby */
 
 /*

include/uapi/linux/raid/md_u.h

···
 #define STOP_ARRAY		_IO (MD_MAJOR, 0x32)
 #define STOP_ARRAY_RO		_IO (MD_MAJOR, 0x33)
 #define RESTART_ARRAY_RW	_IO (MD_MAJOR, 0x34)
+#define CLUSTERED_DISK_NACK	_IO (MD_MAJOR, 0x35)
 
 /* 63 partitions with the alternate major number (mdp) */
 #define MdpMinorShift 6

lib/raid6/algos.c

···
 static inline const struct raid6_calls *raid6_choose_gen(
 	void *(*const dptrs)[(65536/PAGE_SIZE)+2], const int disks)
 {
-	unsigned long perf, bestperf, j0, j1;
+	unsigned long perf, bestgenperf, bestxorperf, j0, j1;
+	int start = (disks>>1)-1, stop = disks-3;	/* work on the second half of the disks */
 	const struct raid6_calls *const *algo;
 	const struct raid6_calls *best;
 
-	for (bestperf = 0, best = NULL, algo = raid6_algos; *algo; algo++) {
+	for (bestgenperf = 0, bestxorperf = 0, best = NULL, algo = raid6_algos; *algo; algo++) {
 		if (!best || (*algo)->prefer >= best->prefer) {
 			if ((*algo)->valid && !(*algo)->valid())
 				continue;
···
 			}
 			preempt_enable();
 
-			if (perf > bestperf) {
-				bestperf = perf;
+			if (perf > bestgenperf) {
+				bestgenperf = perf;
 				best = *algo;
 			}
-			pr_info("raid6: %-8s %5ld MB/s\n", (*algo)->name,
+			pr_info("raid6: %-8s gen() %5ld MB/s\n", (*algo)->name,
 			       (perf*HZ) >> (20-16+RAID6_TIME_JIFFIES_LG2));
+
+			if (!(*algo)->xor_syndrome)
+				continue;
+
+			perf = 0;
+
+			preempt_disable();
+			j0 = jiffies;
+			while ((j1 = jiffies) == j0)
+				cpu_relax();
+			while (time_before(jiffies,
+					    j1 + (1<<RAID6_TIME_JIFFIES_LG2))) {
+				(*algo)->xor_syndrome(disks, start, stop,
+						      PAGE_SIZE, *dptrs);
+				perf++;
+			}
+			preempt_enable();
+
+			if (best == *algo)
+				bestxorperf = perf;
+
+			pr_info("raid6: %-8s xor() %5ld MB/s\n", (*algo)->name,
+				(perf*HZ) >> (20-16+RAID6_TIME_JIFFIES_LG2+1));
 		}
 	}
 
 	if (best) {
-		pr_info("raid6: using algorithm %s (%ld MB/s)\n",
+		pr_info("raid6: using algorithm %s gen() %ld MB/s\n",
 		       best->name,
-		       (bestperf*HZ) >> (20-16+RAID6_TIME_JIFFIES_LG2));
+		       (bestgenperf*HZ) >> (20-16+RAID6_TIME_JIFFIES_LG2));
+		if (best->xor_syndrome)
+			pr_info("raid6: .... xor() %ld MB/s, rmw enabled\n",
+			       (bestxorperf*HZ) >> (20-16+RAID6_TIME_JIFFIES_LG2+1));
 		raid6_call = *best;
 	} else
 		pr_err("raid6: Yikes!  No algorithm found!\n");

lib/raid6/altivec.uc

···
 
 const struct raid6_calls raid6_altivec$# = {
 	raid6_altivec$#_gen_syndrome,
+	NULL,			/* XOR not yet implemented */
 	raid6_have_altivec,
 	"altivecx$#",
 	0

lib/raid6/avx2.c

···
 
 const struct raid6_calls raid6_avx2x1 = {
 	raid6_avx21_gen_syndrome,
+	NULL,			/* XOR not yet implemented */
 	raid6_have_avx2,
 	"avx2x1",
 	1			/* Has cache hints */
···
 
 const struct raid6_calls raid6_avx2x2 = {
 	raid6_avx22_gen_syndrome,
+	NULL,			/* XOR not yet implemented */
 	raid6_have_avx2,
 	"avx2x2",
 	1			/* Has cache hints */
···
 
 const struct raid6_calls raid6_avx2x4 = {
 	raid6_avx24_gen_syndrome,
+	NULL,			/* XOR not yet implemented */
 	raid6_have_avx2,
 	"avx2x4",
 	1			/* Has cache hints */

lib/raid6/int.uc

···
 	}
 }
 
+static void raid6_int$#_xor_syndrome(int disks, int start, int stop,
+				     size_t bytes, void **ptrs)
+{
+	u8 **dptr = (u8 **)ptrs;
+	u8 *p, *q;
+	int d, z, z0;
+
+	unative_t wd$$, wq$$, wp$$, w1$$, w2$$;
+
+	z0 = stop;		/* P/Q right side optimization */
+	p = dptr[disks-2];	/* XOR parity */
+	q = dptr[disks-1];	/* RS syndrome */
+
+	for ( d = 0 ; d < bytes ; d += NSIZE*$# ) {
+		/* P/Q data pages */
+		wq$$ = wp$$ = *(unative_t *)&dptr[z0][d+$$*NSIZE];
+		for ( z = z0-1 ; z >= start ; z-- ) {
+			wd$$ = *(unative_t *)&dptr[z][d+$$*NSIZE];
+			wp$$ ^= wd$$;
+			w2$$ = MASK(wq$$);
+			w1$$ = SHLBYTE(wq$$);
+			w2$$ &= NBYTES(0x1d);
+			w1$$ ^= w2$$;
+			wq$$ = w1$$ ^ wd$$;
+		}
+		/* P/Q left side optimization */
+		for ( z = start-1 ; z >= 0 ; z-- ) {
+			w2$$ = MASK(wq$$);
+			w1$$ = SHLBYTE(wq$$);
+			w2$$ &= NBYTES(0x1d);
+			wq$$ = w1$$ ^ w2$$;
+		}
+		*(unative_t *)&p[d+NSIZE*$$] ^= wp$$;
+		*(unative_t *)&q[d+NSIZE*$$] ^= wq$$;
+	}
+
+}
+
 const struct raid6_calls raid6_intx$# = {
 	raid6_int$#_gen_syndrome,
-	NULL,		/* always valid */
+	raid6_int$#_xor_syndrome,
+	NULL,			/* always valid */
 	"int" NSTRING "x$#",
 	0
 };

lib/raid6/mmx.c

···
 
 const struct raid6_calls raid6_mmxx1 = {
 	raid6_mmx1_gen_syndrome,
+	NULL,			/* XOR not yet implemented */
 	raid6_have_mmx,
 	"mmxx1",
 	0
···
 
 const struct raid6_calls raid6_mmxx2 = {
 	raid6_mmx2_gen_syndrome,
+	NULL,			/* XOR not yet implemented */
 	raid6_have_mmx,
 	"mmxx2",
 	0

lib/raid6/neon.c

···
 	}								\
 	struct raid6_calls const raid6_neonx ## _n = {			\
 		raid6_neon ## _n ## _gen_syndrome,			\
+		NULL,		/* XOR not yet implemented */		\
 		raid6_have_neon,					\
 		"neonx" #_n,						\
 		0							\

lib/raid6/sse1.c

···
 
 const struct raid6_calls raid6_sse1x1 = {
 	raid6_sse11_gen_syndrome,
+	NULL,			/* XOR not yet implemented */
 	raid6_have_sse1_or_mmxext,
 	"sse1x1",
 	1			/* Has cache hints */
···
 
 const struct raid6_calls raid6_sse1x2 = {
 	raid6_sse12_gen_syndrome,
+	NULL,			/* XOR not yet implemented */
 	raid6_have_sse1_or_mmxext,
 	"sse1x2",
 	1			/* Has cache hints */

lib/raid6/sse2.c

···
 	kernel_fpu_end();
 }
 
+
+static void raid6_sse21_xor_syndrome(int disks, int start, int stop,
+				     size_t bytes, void **ptrs)
+ {
+	u8 **dptr = (u8 **)ptrs;
+	u8 *p, *q;
+	int d, z, z0;
+
+	z0 = stop;		/* P/Q right side optimization */
+	p = dptr[disks-2];	/* XOR parity */
+	q = dptr[disks-1];	/* RS syndrome */
+
+	kernel_fpu_begin();
+
+	asm volatile("movdqa %0,%%xmm0" : : "m" (raid6_sse_constants.x1d[0]));
+
+	for ( d = 0 ; d < bytes ; d += 16 ) {
+		asm volatile("movdqa %0,%%xmm4" :: "m" (dptr[z0][d]));
+		asm volatile("movdqa %0,%%xmm2" : : "m" (p[d]));
+		asm volatile("pxor %xmm4,%xmm2");
+		/* P/Q data pages */
+		for ( z = z0-1 ; z >= start ; z-- ) {
+			asm volatile("pxor %xmm5,%xmm5");
+			asm volatile("pcmpgtb %xmm4,%xmm5");
+			asm volatile("paddb %xmm4,%xmm4");
+			asm volatile("pand %xmm0,%xmm5");
+			asm volatile("pxor %xmm5,%xmm4");
+			asm volatile("movdqa %0,%%xmm5" :: "m" (dptr[z][d]));
+			asm volatile("pxor %xmm5,%xmm2");
+			asm volatile("pxor %xmm5,%xmm4");
+		}
+		/* P/Q left side optimization */
+		for ( z = start-1 ; z >= 0 ; z-- ) {
+			asm volatile("pxor %xmm5,%xmm5");
+			asm volatile("pcmpgtb %xmm4,%xmm5");
+			asm volatile("paddb %xmm4,%xmm4");
+			asm volatile("pand %xmm0,%xmm5");
+			asm volatile("pxor %xmm5,%xmm4");
+		}
+		asm volatile("pxor %0,%%xmm4" : : "m" (q[d]));
+		/* Don't use movntdq for r/w memory area < cache line */
+		asm volatile("movdqa %%xmm4,%0" : "=m" (q[d]));
+		asm volatile("movdqa %%xmm2,%0" : "=m" (p[d]));
+	}
+
+	asm volatile("sfence" : : : "memory");
+	kernel_fpu_end();
+}
+
 const struct raid6_calls raid6_sse2x1 = {
 	raid6_sse21_gen_syndrome,
+	raid6_sse21_xor_syndrome,
 	raid6_have_sse2,
 	"sse2x1",
 	1			/* Has cache hints */
···
 	kernel_fpu_end();
 }
 
+ static void raid6_sse22_xor_syndrome(int disks, int start, int stop,
+				     size_t bytes, void **ptrs)
+ {
+	u8 **dptr = (u8 **)ptrs;
+	u8 *p, *q;
+	int d, z, z0;
+
+	z0 = stop;		/* P/Q right side optimization */
+	p = dptr[disks-2];	/* XOR parity */
+	q = dptr[disks-1];	/* RS syndrome */
+
+	kernel_fpu_begin();
+
+	asm volatile("movdqa %0,%%xmm0" : : "m" (raid6_sse_constants.x1d[0]));
+
+	for ( d = 0 ; d < bytes ; d += 32 ) {
+		asm volatile("movdqa %0,%%xmm4" :: "m" (dptr[z0][d]));
+		asm volatile("movdqa %0,%%xmm6" :: "m" (dptr[z0][d+16]));
+		asm volatile("movdqa %0,%%xmm2" : : "m" (p[d]));
+		asm volatile("movdqa %0,%%xmm3" : : "m" (p[d+16]));
+		asm volatile("pxor %xmm4,%xmm2");
+		asm volatile("pxor %xmm6,%xmm3");
+		/* P/Q data pages */
+		for ( z = z0-1 ; z >= start ; z-- ) {
+			asm volatile("pxor %xmm5,%xmm5");
+			asm volatile("pxor %xmm7,%xmm7");
+			asm volatile("pcmpgtb %xmm4,%xmm5");
+			asm volatile("pcmpgtb %xmm6,%xmm7");
+			asm volatile("paddb %xmm4,%xmm4");
+			asm volatile("paddb %xmm6,%xmm6");
+			asm volatile("pand %xmm0,%xmm5");
+			asm volatile("pand %xmm0,%xmm7");
+			asm volatile("pxor %xmm5,%xmm4");
+			asm volatile("pxor %xmm7,%xmm6");
+			asm volatile("movdqa %0,%%xmm5" :: "m" (dptr[z][d]));
+			asm volatile("movdqa %0,%%xmm7" :: "m" (dptr[z][d+16]));
+			asm volatile("pxor %xmm5,%xmm2");
+			asm volatile("pxor %xmm7,%xmm3");
+			asm volatile("pxor %xmm5,%xmm4");
+			asm volatile("pxor %xmm7,%xmm6");
+		}
+		/* P/Q left side optimization */
+		for ( z = start-1 ; z >= 0 ; z-- ) {
+			asm volatile("pxor %xmm5,%xmm5");
+			asm volatile("pxor %xmm7,%xmm7");
+			asm volatile("pcmpgtb %xmm4,%xmm5");
+			asm volatile("pcmpgtb %xmm6,%xmm7");
+			asm volatile("paddb %xmm4,%xmm4");
+			asm volatile("paddb %xmm6,%xmm6");
+			asm volatile("pand %xmm0,%xmm5");
+			asm volatile("pand %xmm0,%xmm7");
+			asm volatile("pxor %xmm5,%xmm4");
+			asm volatile("pxor %xmm7,%xmm6");
+		}
+		asm volatile("pxor %0,%%xmm4" : : "m" (q[d]));
+		asm volatile("pxor %0,%%xmm6" : : "m" (q[d+16]));
+		/* Don't use movntdq for r/w memory area < cache line */
+		asm volatile("movdqa %%xmm4,%0" : "=m" (q[d]));
+		asm volatile("movdqa %%xmm6,%0" : "=m" (q[d+16]));
+		asm volatile("movdqa %%xmm2,%0" : "=m" (p[d]));
+		asm volatile("movdqa %%xmm3,%0" : "=m" (p[d+16]));
+	}
+
+	asm volatile("sfence" : : : "memory");
+	kernel_fpu_end();
+ }
+
 const struct raid6_calls raid6_sse2x2 = {
 	raid6_sse22_gen_syndrome,
+	raid6_sse22_xor_syndrome,
 	raid6_have_sse2,
 	"sse2x2",
 	1			/* Has cache hints */
···
 	kernel_fpu_end();
 }
 
+ static void raid6_sse24_xor_syndrome(int disks, int start, int stop,
+				     size_t bytes, void **ptrs)
+ {
+	u8 **dptr = (u8 **)ptrs;
+	u8 *p, *q;
+	int d, z, z0;
+
+	z0 = stop;		/* P/Q right side optimization */
+	p = dptr[disks-2];	/* XOR parity */
+	q = dptr[disks-1];	/* RS syndrome */
+
+	kernel_fpu_begin();
+
+	asm volatile("movdqa %0,%%xmm0" :: "m" (raid6_sse_constants.x1d[0]));
+
+	for ( d = 0 ; d < bytes ; d += 64 ) {
+		asm volatile("movdqa %0,%%xmm4" :: "m" (dptr[z0][d]));
+		asm volatile("movdqa %0,%%xmm6" :: "m" (dptr[z0][d+16]));
+		asm volatile("movdqa %0,%%xmm12" :: "m" (dptr[z0][d+32]));
+		asm volatile("movdqa %0,%%xmm14" :: "m" (dptr[z0][d+48]));
+		asm volatile("movdqa %0,%%xmm2" : : "m" (p[d]));
+		asm volatile("movdqa %0,%%xmm3" : : "m" (p[d+16]));
+		asm volatile("movdqa %0,%%xmm10" : : "m" (p[d+32]));
+		asm volatile("movdqa %0,%%xmm11" : : "m" (p[d+48]));
+		asm volatile("pxor %xmm4,%xmm2");
+		asm volatile("pxor %xmm6,%xmm3");
+		asm volatile("pxor %xmm12,%xmm10");
+		asm volatile("pxor %xmm14,%xmm11");
+		/* P/Q data pages */
+		for ( z = z0-1 ; z >= start ; z-- ) {
+			asm volatile("prefetchnta %0" :: "m" (dptr[z][d]));
+			asm volatile("prefetchnta %0" :: "m" (dptr[z][d+32]));
+			asm volatile("pxor %xmm5,%xmm5");
+			asm volatile("pxor %xmm7,%xmm7");
+			asm volatile("pxor %xmm13,%xmm13");
+			asm volatile("pxor %xmm15,%xmm15");
+			asm volatile("pcmpgtb %xmm4,%xmm5");
+			asm volatile("pcmpgtb %xmm6,%xmm7");
+			asm volatile("pcmpgtb %xmm12,%xmm13");
+			asm volatile("pcmpgtb %xmm14,%xmm15");
+			asm volatile("paddb %xmm4,%xmm4");
+			asm volatile("paddb %xmm6,%xmm6");
+			asm volatile("paddb %xmm12,%xmm12");
+			asm volatile("paddb %xmm14,%xmm14");
+			asm volatile("pand %xmm0,%xmm5");
+			asm volatile("pand %xmm0,%xmm7");
+			asm volatile("pand %xmm0,%xmm13");
+			asm volatile("pand %xmm0,%xmm15");
+			asm volatile("pxor %xmm5,%xmm4");
+			asm volatile("pxor %xmm7,%xmm6");
+			asm volatile("pxor %xmm13,%xmm12");
+			asm volatile("pxor %xmm15,%xmm14");
+			asm volatile("movdqa %0,%%xmm5" :: "m" (dptr[z][d]));
+			asm volatile("movdqa %0,%%xmm7" :: "m" (dptr[z][d+16]));
+			asm volatile("movdqa %0,%%xmm13" :: "m" (dptr[z][d+32]));
+			asm volatile("movdqa %0,%%xmm15" :: "m" (dptr[z][d+48]));
+			asm volatile("pxor %xmm5,%xmm2");
+			asm volatile("pxor %xmm7,%xmm3");
+			asm volatile("pxor %xmm13,%xmm10");
+			asm volatile("pxor %xmm15,%xmm11");
+			asm volatile("pxor %xmm5,%xmm4");
+			asm volatile("pxor %xmm7,%xmm6");
+			asm volatile("pxor %xmm13,%xmm12");
+			asm volatile("pxor %xmm15,%xmm14");
+		}
+		asm volatile("prefetchnta %0" :: "m" (q[d]));
+		asm volatile("prefetchnta %0" :: "m" (q[d+32]));
+		/* P/Q left side optimization */
+		for ( z = start-1 ; z >= 0 ; z-- ) {
+			asm volatile("pxor %xmm5,%xmm5");
+			asm volatile("pxor %xmm7,%xmm7");
+			asm volatile("pxor %xmm13,%xmm13");
+			asm volatile("pxor %xmm15,%xmm15");
+			asm volatile("pcmpgtb %xmm4,%xmm5");
+			asm volatile("pcmpgtb %xmm6,%xmm7");
+			asm volatile("pcmpgtb %xmm12,%xmm13");
+			asm volatile("pcmpgtb %xmm14,%xmm15");
+			asm volatile("paddb %xmm4,%xmm4");
+			asm volatile("paddb %xmm6,%xmm6");
+			asm volatile("paddb %xmm12,%xmm12");
+			asm volatile("paddb %xmm14,%xmm14");
+			asm volatile("pand %xmm0,%xmm5");
+			asm volatile("pand %xmm0,%xmm7");
+			asm volatile("pand %xmm0,%xmm13");
+			asm volatile("pand %xmm0,%xmm15");
+			asm volatile("pxor %xmm5,%xmm4");
+			asm volatile("pxor %xmm7,%xmm6");
+			asm volatile("pxor %xmm13,%xmm12");
+			asm volatile("pxor %xmm15,%xmm14");
+		}
+		asm volatile("movntdq %%xmm2,%0" : "=m" (p[d]));
+		asm volatile("movntdq %%xmm3,%0" : "=m" (p[d+16]));
+		asm volatile("movntdq %%xmm10,%0" : "=m" (p[d+32]));
+		asm volatile("movntdq %%xmm11,%0" : "=m" (p[d+48]));
+		asm volatile("pxor %0,%%xmm4" : : "m" (q[d]));
+		asm volatile("pxor %0,%%xmm6" : : "m" (q[d+16]));
+		asm volatile("pxor %0,%%xmm12" : : "m" (q[d+32]));
+		asm volatile("pxor %0,%%xmm14" : : "m" (q[d+48]));
+		asm volatile("movntdq %%xmm4,%0" : "=m" (q[d]));
+		asm volatile("movntdq %%xmm6,%0" : "=m" (q[d+16]));
+		asm volatile("movntdq %%xmm12,%0" : "=m" (q[d+32]));
+		asm volatile("movntdq %%xmm14,%0" : "=m" (q[d+48]));
+	}
+	asm volatile("sfence" : : : "memory");
+	kernel_fpu_end();
+ }
+
+
 const struct raid6_calls raid6_sse2x4 = {
 	raid6_sse24_gen_syndrome,
+	raid6_sse24_xor_syndrome,
 	raid6_have_sse2,
 	"sse2x4",
 	1			/* Has cache hints */

lib/raid6/test/test.c

···
 char data[NDISKS][PAGE_SIZE];
 char recovi[PAGE_SIZE], recovj[PAGE_SIZE];
 
-static void makedata(void)
+static void makedata(int start, int stop)
 {
 	int i, j;
 
-	for (i = 0; i < NDISKS; i++) {
+	for (i = start; i <= stop; i++) {
 		for (j = 0; j < PAGE_SIZE; j++)
 			data[i][j] = rand();
 
···
 {
 	const struct raid6_calls *const *algo;
 	const struct raid6_recov_calls *const *ra;
-	int i, j;
+	int i, j, p1, p2;
 	int err = 0;
 
-	makedata();
+	makedata(0, NDISKS-1);
 
 	for (ra = raid6_recov_algos; *ra; ra++) {
 		if ((*ra)->valid  && !(*ra)->valid())
 			continue;
+
 		raid6_2data_recov = (*ra)->data2;
 		raid6_datap_recov = (*ra)->datap;
 
 		printf("using recovery %s\n", (*ra)->name);
 
 		for (algo = raid6_algos; *algo; algo++) {
-			if (!(*algo)->valid || (*algo)->valid()) {
-				raid6_call = **algo;
+			if ((*algo)->valid && !(*algo)->valid())
+				continue;
 
-				/* Nuke syndromes */
-				memset(data[NDISKS-2], 0xee, 2*PAGE_SIZE);
+			raid6_call = **algo;
 
-				/* Generate assumed good syndrome */
-				raid6_call.gen_syndrome(NDISKS, PAGE_SIZE,
-							(void **)&dataptrs);
+			/* Nuke syndromes */
+			memset(data[NDISKS-2], 0xee, 2*PAGE_SIZE);
 
-				for (i = 0; i < NDISKS-1; i++)
-					for (j = i+1; j < NDISKS; j++)
-						err += test_disks(i, j);
-			}
+			/* Generate assumed good syndrome */
+			raid6_call.gen_syndrome(NDISKS, PAGE_SIZE,
+						(void **)&dataptrs);
+
+			for (i = 0; i < NDISKS-1; i++)
+				for (j = i+1; j < NDISKS; j++)
+					err += test_disks(i, j);
+
+			if (!raid6_call.xor_syndrome)
+				continue;
+
+			for (p1 = 0; p1 < NDISKS-2; p1++)
+				for (p2 = p1; p2 < NDISKS-2; p2++) {
+
+					/* Simulate rmw run */
+					raid6_call.xor_syndrome(NDISKS, p1, p2, PAGE_SIZE,
+								(void **)&dataptrs);
+					makedata(p1, p2);
+					raid6_call.xor_syndrome(NDISKS, p1, p2, PAGE_SIZE,
+                                                                (void **)&dataptrs);
+
+					for (i = 0; i < NDISKS-1; i++)
+						for (j = i+1; j < NDISKS; j++)
+							err += test_disks(i, j);
+				}
+
 		}
 		printf("\n");
 	}

lib/raid6/tilegx.uc

···
 
 const struct raid6_calls raid6_tilegx$# = {
 	raid6_tilegx$#_gen_syndrome,
+	NULL,			/* XOR not yet implemented */
 	NULL,
 	"tilegx$#",
 	0