Merge tag 'dm-4.11-changes' of git://git.kernel.org/pub/scm/linux/kernel/git/device-mapper/linux-dm

Pull device mapper updates from Mike Snitzer:

 - Fix dm-raid transient device failure processing and other smaller
   tweaks.

 - Add journal support to the DM raid target to close the 'write hole'
   on raid 4/5/6.

 - Fix dm-cache corruption, due to rounding bug, when cache exceeds 2TB.

 - Add 'metadata2' feature to dm-cache to separate the dirty bitset out
   from other cache metadata. This improves speed of shutting down a
   large cache device (which implies writing out dirty bits).

 - Fix a memory leak during dm-stats data structure destruction.

 - Fix a DM multipath round-robin path selector performance regression
   that was caused by less precise balancing across all paths.

 - Lastly, introduce a DM core fix for a long-standing DM snapshot
   deadlock that is rooted in the complexity of the device stack used in
   conjunction with block core maintaining bios on current->bio_list to
   manage recursion in generic_make_request(). A more comprehensive fix
   to block core (and its hook in the cpu scheduler) would be wonderful
   but this DM-specific fix is pragmatic considering how difficult it
   has been to make progress on a generic fix.

* tag 'dm-4.11-changes' of git://git.kernel.org/pub/scm/linux/kernel/git/device-mapper/linux-dm: (22 commits)
  dm: flush queued bios when process blocks to avoid deadlock
  dm round robin: revert "use percpu 'repeat_count' and 'current_path'"
  dm stats: fix a leaked s->histogram_boundaries array
  dm space map metadata: constify dm_space_map structures
  dm cache metadata: use cursor api in blocks_are_clean_separate_dirty()
  dm persistent data: add cursor skip functions to the cursor APIs
  dm cache metadata: use dm_bitset_new() to create the dirty bitset in format 2
  dm bitset: add dm_bitset_new()
  dm cache metadata: name the cache block that couldn't be loaded
  dm cache metadata: add "metadata2" feature
  dm cache metadata: use bitset cursor api to load discard bitset
  dm bitset: introduce cursor api
  dm btree: use GFP_NOFS in dm_btree_del()
  dm space map common: memcpy the disk root to ensure it's arch aligned
  dm block manager: add unlikely() annotations on dm_bufio error paths
  dm cache: fix corruption seen when using cache > 2TB
  dm raid: cleanup awkward branching in raid_message() option processing
  dm raid: use mddev rather than rdev->mddev
  dm raid: use read_disk_sb() throughout
  dm raid: add raid4/5/6 journaling support
  ...

Documentation/device-mapper/cache.txt

@@ -207,6 +207,10 @@ Optional feature arguments are:
 		   block, then the cache block is invalidated.
 		   To enable passthrough mode the cache must be clean.
 
+   metadata2	: use version 2 of the metadata.  This stores the dirty bits
+                  in a separate btree, which improves speed of shutting
+		  down the cache.
+
 A policy called 'default' is always registered.  This is an alias for
 the policy we currently think is giving best all round performance.
 

Documentation/device-mapper/dm-raid.txt

@@ -161,6 +161,15 @@ The target is named "raid" and it accepts the following parameters:
 		the RAID type (i.e. the allocation algorithm) as well, e.g.
 		changing from raid5_ls to raid5_n.
 
+	[journal_dev <dev>]
+		This option adds a journal device to raid4/5/6 raid sets and
+		uses it to close the 'write hole' caused by the non-atomic updates
+		to the component devices which can cause data loss during recovery.
+		The journal device is used as writethrough thus causing writes to
+		be throttled versus non-journaled raid4/5/6 sets.
+		Takeover/reshape is not possible with a raid4/5/6 journal device;
+		it has to be deconfigured before requesting these.
+
 <#raid_devs>: The number of devices composing the array.
 	Each device consists of two entries.  The first is the device
 	containing the metadata (if any); the second is the one containing the
@@ -245,6 +254,9 @@ recovery.  Here is a fuller description of the individual fields:
 	<data_offset>   The current data offset to the start of the user data on
 			each component device of a raid set (see the respective
 			raid parameter to support out-of-place reshaping).
+	<journal_char>	'A' - active raid4/5/6 journal device.
+			'D' - dead journal device.
+			'-' - no journal device.
 
 
 Message Interface
@@ -314,3 +326,8 @@ Version History
 1.9.0   Add support for RAID level takeover/reshape/region size
 	and set size reduction.
 1.9.1   Fix activation of existing RAID 4/10 mapped devices
+1.9.2   Don't emit '- -' on the status table line in case the constructor
+	fails reading a superblock. Correctly emit 'maj:min1 maj:min2' and
+	'D' on the status line.  If '- -' is passed into the constructor, emit
+	'- -' on the table line and '-' as the status line health character.
+1.10.0  Add support for raid4/5/6 journal device

drivers/md/dm-cache-metadata.c

@@ -25,7 +25,7 @@
  * defines a range of metadata versions that this module can handle.
  */
 #define MIN_CACHE_VERSION 1
-#define MAX_CACHE_VERSION 1
+#define MAX_CACHE_VERSION 2
 
 #define CACHE_METADATA_CACHE_SIZE 64
 
@@ -55,6 +55,7 @@ enum mapping_bits {
 
 	/*
 	 * The data on the cache is different from that on the origin.
+	 * This flag is only used by metadata format 1.
 	 */
 	M_DIRTY = 2
 };
@@ -93,12 +94,18 @@ struct cache_disk_superblock {
 	__le32 write_misses;
 
 	__le32 policy_version[CACHE_POLICY_VERSION_SIZE];
+
+	/*
+	 * Metadata format 2 fields.
+	 */
+	__le64 dirty_root;
 } __packed;
 
 struct dm_cache_metadata {
 	atomic_t ref_count;
 	struct list_head list;
 
+	unsigned version;
 	struct block_device *bdev;
 	struct dm_block_manager *bm;
 	struct dm_space_map *metadata_sm;
@@ -141,12 +148,19 @@ struct dm_cache_metadata {
 	 */
 	bool fail_io:1;
 
+	/*
+	 * Metadata format 2 fields.
+	 */
+	dm_block_t dirty_root;
+	struct dm_disk_bitset dirty_info;
+
 	/*
 	 * These structures are used when loading metadata.  They're too
 	 * big to put on the stack.
 	 */
 	struct dm_array_cursor mapping_cursor;
 	struct dm_array_cursor hint_cursor;
+	struct dm_bitset_cursor dirty_cursor;
 };
 
 /*-------------------------------------------------------------------
@@ -170,6 +184,7 @@ static void sb_prepare_for_write(struct dm_block_validator *v,
 static int check_metadata_version(struct cache_disk_superblock *disk_super)
 {
 	uint32_t metadata_version = le32_to_cpu(disk_super->version);
+
 	if (metadata_version < MIN_CACHE_VERSION || metadata_version > MAX_CACHE_VERSION) {
 		DMERR("Cache metadata version %u found, but only versions between %u and %u supported.",
 		      metadata_version, MIN_CACHE_VERSION, MAX_CACHE_VERSION);
@@ -310,6 +325,11 @@ static void __copy_sm_root(struct dm_cache_metadata *cmd,
 	       sizeof(cmd->metadata_space_map_root));
 }
 
+static bool separate_dirty_bits(struct dm_cache_metadata *cmd)
+{
+	return cmd->version >= 2;
+}
+
 static int __write_initial_superblock(struct dm_cache_metadata *cmd)
 {
 	int r;
@@ -341,7 +361,7 @@ static int __write_initial_superblock(struct dm_cache_metadata *cmd)
 	disk_super->flags = 0;
 	memset(disk_super->uuid, 0, sizeof(disk_super->uuid));
 	disk_super->magic = cpu_to_le64(CACHE_SUPERBLOCK_MAGIC);
-	disk_super->version = cpu_to_le32(MAX_CACHE_VERSION);
+	disk_super->version = cpu_to_le32(cmd->version);
 	memset(disk_super->policy_name, 0, sizeof(disk_super->policy_name));
 	memset(disk_super->policy_version, 0, sizeof(disk_super->policy_version));
 	disk_super->policy_hint_size = 0;
@@ -362,6 +382,9 @@ static int __write_initial_superblock(struct dm_cache_metadata *cmd)
 	disk_super->write_hits = cpu_to_le32(0);
 	disk_super->write_misses = cpu_to_le32(0);
 
+	if (separate_dirty_bits(cmd))
+		disk_super->dirty_root = cpu_to_le64(cmd->dirty_root);
+
 	return dm_tm_commit(cmd->tm, sblock);
 }
 
@@ -382,6 +405,13 @@ static int __format_metadata(struct dm_cache_metadata *cmd)
 	if (r < 0)
 		goto bad;
 
+	if (separate_dirty_bits(cmd)) {
+		dm_disk_bitset_init(cmd->tm, &cmd->dirty_info);
+		r = dm_bitset_empty(&cmd->dirty_info, &cmd->dirty_root);
+		if (r < 0)
+			goto bad;
+	}
+
 	dm_disk_bitset_init(cmd->tm, &cmd->discard_info);
 	r = dm_bitset_empty(&cmd->discard_info, &cmd->discard_root);
 	if (r < 0)
@@ -407,9 +437,10 @@ static int __format_metadata(struct dm_cache_metadata *cmd)
 static int __check_incompat_features(struct cache_disk_superblock *disk_super,
 				     struct dm_cache_metadata *cmd)
 {
-	uint32_t features;
+	uint32_t incompat_flags, features;
 
-	features = le32_to_cpu(disk_super->incompat_flags) & ~DM_CACHE_FEATURE_INCOMPAT_SUPP;
+	incompat_flags = le32_to_cpu(disk_super->incompat_flags);
+	features = incompat_flags & ~DM_CACHE_FEATURE_INCOMPAT_SUPP;
 	if (features) {
 		DMERR("could not access metadata due to unsupported optional features (%lx).",
 		      (unsigned long)features);
@@ -470,6 +501,7 @@ static int __open_metadata(struct dm_cache_metadata *cmd)
 	}
 
 	__setup_mapping_info(cmd);
+	dm_disk_bitset_init(cmd->tm, &cmd->dirty_info);
 	dm_disk_bitset_init(cmd->tm, &cmd->discard_info);
 	sb_flags = le32_to_cpu(disk_super->flags);
 	cmd->clean_when_opened = test_bit(CLEAN_SHUTDOWN, &sb_flags);
@@ -548,6 +580,7 @@ static unsigned long clear_clean_shutdown(unsigned long flags)
 static void read_superblock_fields(struct dm_cache_metadata *cmd,
 				   struct cache_disk_superblock *disk_super)
 {
+	cmd->version = le32_to_cpu(disk_super->version);
 	cmd->flags = le32_to_cpu(disk_super->flags);
 	cmd->root = le64_to_cpu(disk_super->mapping_root);
 	cmd->hint_root = le64_to_cpu(disk_super->hint_root);
@@ -567,6 +600,9 @@ static void read_superblock_fields(struct dm_cache_metadata *cmd,
 	cmd->stats.write_hits = le32_to_cpu(disk_super->write_hits);
 	cmd->stats.write_misses = le32_to_cpu(disk_super->write_misses);
 
+	if (separate_dirty_bits(cmd))
+		cmd->dirty_root = le64_to_cpu(disk_super->dirty_root);
+
 	cmd->changed = false;
 }
 
@@ -625,6 +661,13 @@ static int __commit_transaction(struct dm_cache_metadata *cmd,
 	 */
 	BUILD_BUG_ON(sizeof(struct cache_disk_superblock) > 512);
 
+	if (separate_dirty_bits(cmd)) {
+		r = dm_bitset_flush(&cmd->dirty_info, cmd->dirty_root,
+				    &cmd->dirty_root);
+		if (r)
+			return r;
+	}
+
 	r = dm_bitset_flush(&cmd->discard_info, cmd->discard_root,
 			    &cmd->discard_root);
 	if (r)
@@ -649,6 +692,8 @@ static int __commit_transaction(struct dm_cache_metadata *cmd,
 		update_flags(disk_super, mutator);
 
 	disk_super->mapping_root = cpu_to_le64(cmd->root);
+	if (separate_dirty_bits(cmd))
+		disk_super->dirty_root = cpu_to_le64(cmd->dirty_root);
 	disk_super->hint_root = cpu_to_le64(cmd->hint_root);
 	disk_super->discard_root = cpu_to_le64(cmd->discard_root);
 	disk_super->discard_block_size = cpu_to_le64(cmd->discard_block_size);
@@ -698,7 +743,8 @@ static void unpack_value(__le64 value_le, dm_oblock_t *block, unsigned *flags)
 static struct dm_cache_metadata *metadata_open(struct block_device *bdev,
 					       sector_t data_block_size,
 					       bool may_format_device,
-					       size_t policy_hint_size)
+					       size_t policy_hint_size,
+					       unsigned metadata_version)
 {
 	int r;
 	struct dm_cache_metadata *cmd;
@@ -709,6 +755,7 @@ static struct dm_cache_metadata *metadata_open(struct block_device *bdev,
 		return ERR_PTR(-ENOMEM);
 	}
 
+	cmd->version = metadata_version;
 	atomic_set(&cmd->ref_count, 1);
 	init_rwsem(&cmd->root_lock);
 	cmd->bdev = bdev;
@@ -757,7 +804,8 @@ static struct dm_cache_metadata *lookup(struct block_device *bdev)
 static struct dm_cache_metadata *lookup_or_open(struct block_device *bdev,
 						sector_t data_block_size,
 						bool may_format_device,
-						size_t policy_hint_size)
+						size_t policy_hint_size,
+						unsigned metadata_version)
 {
 	struct dm_cache_metadata *cmd, *cmd2;
 
@@ -768,7 +816,8 @@ static struct dm_cache_metadata *lookup_or_open(struct block_device *bdev,
 	if (cmd)
 		return cmd;
 
-	cmd = metadata_open(bdev, data_block_size, may_format_device, policy_hint_size);
+	cmd = metadata_open(bdev, data_block_size, may_format_device,
+			    policy_hint_size, metadata_version);
 	if (!IS_ERR(cmd)) {
 		mutex_lock(&table_lock);
 		cmd2 = lookup(bdev);
@@ -800,10 +849,11 @@ static bool same_params(struct dm_cache_metadata *cmd, sector_t data_block_size)
 struct dm_cache_metadata *dm_cache_metadata_open(struct block_device *bdev,
 						 sector_t data_block_size,
 						 bool may_format_device,
-						 size_t policy_hint_size)
+						 size_t policy_hint_size,
+						 unsigned metadata_version)
 {
-	struct dm_cache_metadata *cmd = lookup_or_open(bdev, data_block_size,
-						       may_format_device, policy_hint_size);
+	struct dm_cache_metadata *cmd = lookup_or_open(bdev, data_block_size, may_format_device,
+						       policy_hint_size, metadata_version);
 
 	if (!IS_ERR(cmd) && !same_params(cmd, data_block_size)) {
 		dm_cache_metadata_close(cmd);
@@ -829,8 +879,8 @@ void dm_cache_metadata_close(struct dm_cache_metadata *cmd)
 /*
  * Checks that the given cache block is either unmapped or clean.
  */
-static int block_unmapped_or_clean(struct dm_cache_metadata *cmd, dm_cblock_t b,
-				   bool *result)
+static int block_clean_combined_dirty(struct dm_cache_metadata *cmd, dm_cblock_t b,
+				      bool *result)
 {
 	int r;
 	__le64 value;
@@ -838,10 +888,8 @@ static int block_unmapped_or_clean(struct dm_cache_metadata *cmd, dm_cblock_t b,
 	unsigned flags;
 
 	r = dm_array_get_value(&cmd->info, cmd->root, from_cblock(b), &value);
-	if (r) {
-		DMERR("block_unmapped_or_clean failed");
+	if (r)
 		return r;
-	}
 
 	unpack_value(value, &ob, &flags);
 	*result = !((flags & M_VALID) && (flags & M_DIRTY));
@@ -849,17 +897,19 @@ static int block_unmapped_or_clean(struct dm_cache_metadata *cmd, dm_cblock_t b,
 	return 0;
 }
 
-static int blocks_are_unmapped_or_clean(struct dm_cache_metadata *cmd,
-					dm_cblock_t begin, dm_cblock_t end,
-					bool *result)
+static int blocks_are_clean_combined_dirty(struct dm_cache_metadata *cmd,
+					   dm_cblock_t begin, dm_cblock_t end,
+					   bool *result)
 {
 	int r;
 	*result = true;
 
 	while (begin != end) {
-		r = block_unmapped_or_clean(cmd, begin, result);
-		if (r)
+		r = block_clean_combined_dirty(cmd, begin, result);
+		if (r) {
+			DMERR("block_clean_combined_dirty failed");
 			return r;
+		}
 
 		if (!*result) {
 			DMERR("cache block %llu is dirty",
@@ -873,6 +923,67 @@ static int blocks_are_unmapped_or_clean(struct dm_cache_metadata *cmd,
 	return 0;
 }
 
+static int blocks_are_clean_separate_dirty(struct dm_cache_metadata *cmd,
+					   dm_cblock_t begin, dm_cblock_t end,
+					   bool *result)
+{
+	int r;
+	bool dirty_flag;
+	*result = true;
+
+	r = dm_bitset_cursor_begin(&cmd->dirty_info, cmd->dirty_root,
+				   from_cblock(begin), &cmd->dirty_cursor);
+	if (r) {
+		DMERR("%s: dm_bitset_cursor_begin for dirty failed", __func__);
+		return r;
+	}
+
+	r = dm_bitset_cursor_skip(&cmd->dirty_cursor, from_cblock(begin));
+	if (r) {
+		DMERR("%s: dm_bitset_cursor_skip for dirty failed", __func__);
+		dm_bitset_cursor_end(&cmd->dirty_cursor);
+		return r;
+	}
+
+	while (begin != end) {
+		/*
+		 * We assume that unmapped blocks have their dirty bit
+		 * cleared.
+		 */
+		dirty_flag = dm_bitset_cursor_get_value(&cmd->dirty_cursor);
+		if (dirty_flag) {
+			DMERR("%s: cache block %llu is dirty", __func__,
+			      (unsigned long long) from_cblock(begin));
+			dm_bitset_cursor_end(&cmd->dirty_cursor);
+			*result = false;
+			return 0;
+		}
+
+		r = dm_bitset_cursor_next(&cmd->dirty_cursor);
+		if (r) {
+			DMERR("%s: dm_bitset_cursor_next for dirty failed", __func__);
+			dm_bitset_cursor_end(&cmd->dirty_cursor);
+			return r;
+		}
+
+		begin = to_cblock(from_cblock(begin) + 1);
+	}
+
+	dm_bitset_cursor_end(&cmd->dirty_cursor);
+
+	return 0;
+}
+
+static int blocks_are_unmapped_or_clean(struct dm_cache_metadata *cmd,
+					dm_cblock_t begin, dm_cblock_t end,
+					bool *result)
+{
+	if (separate_dirty_bits(cmd))
+		return blocks_are_clean_separate_dirty(cmd, begin, end, result);
+	else
+		return blocks_are_clean_combined_dirty(cmd, begin, end, result);
+}
+
 static bool cmd_write_lock(struct dm_cache_metadata *cmd)
 {
 	down_write(&cmd->root_lock);
@@ -950,8 +1061,18 @@ int dm_cache_resize(struct dm_cache_metadata *cmd, dm_cblock_t new_cache_size)
 	r = dm_array_resize(&cmd->info, cmd->root, from_cblock(cmd->cache_blocks),
 			    from_cblock(new_cache_size),
 			    &null_mapping, &cmd->root);
-	if (!r)
-		cmd->cache_blocks = new_cache_size;
+	if (r)
+		goto out;
+
+	if (separate_dirty_bits(cmd)) {
+		r = dm_bitset_resize(&cmd->dirty_info, cmd->dirty_root,
+				     from_cblock(cmd->cache_blocks), from_cblock(new_cache_size),
+				     false, &cmd->dirty_root);
+		if (r)
+			goto out;
+	}
+
+	cmd->cache_blocks = new_cache_size;
 	cmd->changed = true;
 
 out:
@@ -995,14 +1116,6 @@ static int __clear_discard(struct dm_cache_metadata *cmd, dm_dblock_t b)
 				   from_dblock(b), &cmd->discard_root);
 }
 
-static int __is_discarded(struct dm_cache_metadata *cmd, dm_dblock_t b,
-			  bool *is_discarded)
-{
-	return dm_bitset_test_bit(&cmd->discard_info, cmd->discard_root,
-				  from_dblock(b), &cmd->discard_root,
-				  is_discarded);
-}
-
 static int __discard(struct dm_cache_metadata *cmd,
 		     dm_dblock_t dblock, bool discard)
 {
@@ -1032,22 +1145,38 @@ static int __load_discards(struct dm_cache_metadata *cmd,
 			   load_discard_fn fn, void *context)
 {
 	int r = 0;
-	dm_block_t b;
-	bool discard;
+	uint32_t b;
+	struct dm_bitset_cursor c;
 
-	for (b = 0; b < from_dblock(cmd->discard_nr_blocks); b++) {
-		dm_dblock_t dblock = to_dblock(b);
+	if (from_dblock(cmd->discard_nr_blocks) == 0)
+		/* nothing to do */
+		return 0;
 
-		if (cmd->clean_when_opened) {
-			r = __is_discarded(cmd, dblock, &discard);
-			if (r)
-				return r;
-		} else
-			discard = false;
+	if (cmd->clean_when_opened) {
+		r = dm_bitset_flush(&cmd->discard_info, cmd->discard_root, &cmd->discard_root);
+		if (r)
+			return r;
 
-		r = fn(context, cmd->discard_block_size, dblock, discard);
+		r = dm_bitset_cursor_begin(&cmd->discard_info, cmd->discard_root,
+					   from_dblock(cmd->discard_nr_blocks), &c);
 		if (r)
-			break;
+			return r;
+
+		for (b = 0; b < from_dblock(cmd->discard_nr_blocks); b++) {
+			r = fn(context, cmd->discard_block_size, to_dblock(b),
+			       dm_bitset_cursor_get_value(&c));
+			if (r)
+				break;
+		}
+
+		dm_bitset_cursor_end(&c);
+
+	} else {
+		for (b = 0; b < from_dblock(cmd->discard_nr_blocks); b++) {
+			r = fn(context, cmd->discard_block_size, to_dblock(b), false);
+			if (r)
+				return r;
+		}
 	}
 
 	return r;
@@ -1177,11 +1306,11 @@ static bool hints_array_available(struct dm_cache_metadata *cmd,
 		hints_array_initialized(cmd);
 }
 
-static int __load_mapping(struct dm_cache_metadata *cmd,
-			  uint64_t cb, bool hints_valid,
-			  struct dm_array_cursor *mapping_cursor,
-			  struct dm_array_cursor *hint_cursor,
-			  load_mapping_fn fn, void *context)
+static int __load_mapping_v1(struct dm_cache_metadata *cmd,
+			     uint64_t cb, bool hints_valid,
+			     struct dm_array_cursor *mapping_cursor,
+			     struct dm_array_cursor *hint_cursor,
+			     load_mapping_fn fn, void *context)
 {
 	int r = 0;
 
@@ -1206,8 +1335,51 @@ static int __load_mapping(struct dm_cache_metadata *cmd,
 
 		r = fn(context, oblock, to_cblock(cb), flags & M_DIRTY,
 		       le32_to_cpu(hint), hints_valid);
-		if (r)
-			DMERR("policy couldn't load cblock");
+		if (r) {
+			DMERR("policy couldn't load cache block %llu",
+			      (unsigned long long) from_cblock(to_cblock(cb)));
+		}
+	}
+
+	return r;
+}
+
+static int __load_mapping_v2(struct dm_cache_metadata *cmd,
+			     uint64_t cb, bool hints_valid,
+			     struct dm_array_cursor *mapping_cursor,
+			     struct dm_array_cursor *hint_cursor,
+			     struct dm_bitset_cursor *dirty_cursor,
+			     load_mapping_fn fn, void *context)
+{
+	int r = 0;
+
+	__le64 mapping;
+	__le32 hint = 0;
+
+	__le64 *mapping_value_le;
+	__le32 *hint_value_le;
+
+	dm_oblock_t oblock;
+	unsigned flags;
+	bool dirty;
+
+	dm_array_cursor_get_value(mapping_cursor, (void **) &mapping_value_le);
+	memcpy(&mapping, mapping_value_le, sizeof(mapping));
+	unpack_value(mapping, &oblock, &flags);
+
+	if (flags & M_VALID) {
+		if (hints_valid) {
+			dm_array_cursor_get_value(hint_cursor, (void **) &hint_value_le);
+			memcpy(&hint, hint_value_le, sizeof(hint));
+		}
+
+		dirty = dm_bitset_cursor_get_value(dirty_cursor);
+		r = fn(context, oblock, to_cblock(cb), dirty,
+		       le32_to_cpu(hint), hints_valid);
+		if (r) {
+			DMERR("policy couldn't load cache block %llu",
+			      (unsigned long long) from_cblock(to_cblock(cb)));
+		}
 	}
 
 	return r;
@@ -1238,10 +1410,28 @@ static int __load_mappings(struct dm_cache_metadata *cmd,
 		}
 	}
 
+	if (separate_dirty_bits(cmd)) {
+		r = dm_bitset_cursor_begin(&cmd->dirty_info, cmd->dirty_root,
+					   from_cblock(cmd->cache_blocks),
+					   &cmd->dirty_cursor);
+		if (r) {
+			dm_array_cursor_end(&cmd->hint_cursor);
+			dm_array_cursor_end(&cmd->mapping_cursor);
+			return r;
+		}
+	}
+
 	for (cb = 0; ; cb++) {
-		r = __load_mapping(cmd, cb, hints_valid,
-				   &cmd->mapping_cursor, &cmd->hint_cursor,
-				   fn, context);
+		if (separate_dirty_bits(cmd))
+			r = __load_mapping_v2(cmd, cb, hints_valid,
+					      &cmd->mapping_cursor,
+					      &cmd->hint_cursor,
+					      &cmd->dirty_cursor,
+					      fn, context);
+		else
+			r = __load_mapping_v1(cmd, cb, hints_valid,
+					      &cmd->mapping_cursor, &cmd->hint_cursor,
+					      fn, context);
 		if (r)
 			goto out;
 
@@ -1264,12 +1454,23 @@ static int __load_mappings(struct dm_cache_metadata *cmd,
 				goto out;
 			}
 		}
+
+		if (separate_dirty_bits(cmd)) {
+			r = dm_bitset_cursor_next(&cmd->dirty_cursor);
+			if (r) {
+				DMERR("dm_bitset_cursor_next for dirty failed");
+				goto out;
+			}
+		}
 	}
 out:
 	dm_array_cursor_end(&cmd->mapping_cursor);
 	if (hints_valid)
 		dm_array_cursor_end(&cmd->hint_cursor);
 
+	if (separate_dirty_bits(cmd))
+		dm_bitset_cursor_end(&cmd->dirty_cursor);
+
 	return r;
 }
 
@@ -1352,13 +1553,55 @@ static int __dirty(struct dm_cache_metadata *cmd, dm_cblock_t cblock, bool dirty
 
 }
 
-int dm_cache_set_dirty(struct dm_cache_metadata *cmd,
-		       dm_cblock_t cblock, bool dirty)
+static int __set_dirty_bits_v1(struct dm_cache_metadata *cmd, unsigned nr_bits, unsigned long *bits)
+{
+	int r;
+	unsigned i;
+	for (i = 0; i < nr_bits; i++) {
+		r = __dirty(cmd, to_cblock(i), test_bit(i, bits));
+		if (r)
+			return r;
+	}
+
+	return 0;
+}
+
+static int is_dirty_callback(uint32_t index, bool *value, void *context)
+{
+	unsigned long *bits = context;
+	*value = test_bit(index, bits);
+	return 0;
+}
+
+static int __set_dirty_bits_v2(struct dm_cache_metadata *cmd, unsigned nr_bits, unsigned long *bits)
+{
+	int r = 0;
+
+	/* nr_bits is really just a sanity check */
+	if (nr_bits != from_cblock(cmd->cache_blocks)) {
+		DMERR("dirty bitset is wrong size");
+		return -EINVAL;
+	}
+
+	r = dm_bitset_del(&cmd->dirty_info, cmd->dirty_root);
+	if (r)
+		return r;
+
+	cmd->changed = true;
+	return dm_bitset_new(&cmd->dirty_info, &cmd->dirty_root, nr_bits, is_dirty_callback, bits);
+}
+
+int dm_cache_set_dirty_bits(struct dm_cache_metadata *cmd,
+			    unsigned nr_bits,
+			    unsigned long *bits)
 {
 	int r;
 
 	WRITE_LOCK(cmd);
-	r = __dirty(cmd, cblock, dirty);
+	if (separate_dirty_bits(cmd))
+		r = __set_dirty_bits_v2(cmd, nr_bits, bits);
+	else
+		r = __set_dirty_bits_v1(cmd, nr_bits, bits);
 	WRITE_UNLOCK(cmd);
 
 	return r;

drivers/md/dm-cache-metadata.h

@@ -45,18 +45,20 @@
  * As these various flags are defined they should be added to the
  * following masks.
  */
+
 #define DM_CACHE_FEATURE_COMPAT_SUPP	  0UL
 #define DM_CACHE_FEATURE_COMPAT_RO_SUPP	  0UL
 #define DM_CACHE_FEATURE_INCOMPAT_SUPP	  0UL
 
 /*
- * Reopens or creates a new, empty metadata volume.
- * Returns an ERR_PTR on failure.
+ * Reopens or creates a new, empty metadata volume.  Returns an ERR_PTR on
+ * failure.  If reopening then features must match.
  */
 struct dm_cache_metadata *dm_cache_metadata_open(struct block_device *bdev,
 						 sector_t data_block_size,
 						 bool may_format_device,
-						 size_t policy_hint_size);
+						 size_t policy_hint_size,
+						 unsigned metadata_version);
 
 void dm_cache_metadata_close(struct dm_cache_metadata *cmd);
 
@@ -91,7 +93,8 @@ int dm_cache_load_mappings(struct dm_cache_metadata *cmd,
 			   load_mapping_fn fn,
 			   void *context);
 
-int dm_cache_set_dirty(struct dm_cache_metadata *cmd, dm_cblock_t cblock, bool dirty);
+int dm_cache_set_dirty_bits(struct dm_cache_metadata *cmd,
+			    unsigned nr_bits, unsigned long *bits);
 
 struct dm_cache_statistics {
 	uint32_t read_hits;

drivers/md/dm-cache-target.c

@@ -179,6 +179,7 @@ enum cache_io_mode {
 struct cache_features {
 	enum cache_metadata_mode mode;
 	enum cache_io_mode io_mode;
+	unsigned metadata_version;
 };
 
 struct cache_stats {
@@ -248,7 +249,7 @@ struct cache {
 	/*
 	 * Fields for converting from sectors to blocks.
 	 */
-	uint32_t sectors_per_block;
+	sector_t sectors_per_block;
 	int sectors_per_block_shift;
 
 	spinlock_t lock;
@@ -2534,13 +2535,14 @@ static void init_features(struct cache_features *cf)
 {
 	cf->mode = CM_WRITE;
 	cf->io_mode = CM_IO_WRITEBACK;
+	cf->metadata_version = 1;
 }
 
 static int parse_features(struct cache_args *ca, struct dm_arg_set *as,
 			  char **error)
 {
 	static struct dm_arg _args[] = {
-		{0, 1, "Invalid number of cache feature arguments"},
+		{0, 2, "Invalid number of cache feature arguments"},
 	};
 
 	int r;
@@ -2566,6 +2568,9 @@ static int parse_features(struct cache_args *ca, struct dm_arg_set *as,
 		else if (!strcasecmp(arg, "passthrough"))
 			cf->io_mode = CM_IO_PASSTHROUGH;
 
+		else if (!strcasecmp(arg, "metadata2"))
+			cf->metadata_version = 2;
+
 		else {
 			*error = "Unrecognised cache feature requested";
 			return -EINVAL;
@@ -2820,7 +2825,8 @@ static int cache_create(struct cache_args *ca, struct cache **result)
 
 	cmd = dm_cache_metadata_open(cache->metadata_dev->bdev,
 				     ca->block_size, may_format,
-				     dm_cache_policy_get_hint_size(cache->policy));
+				     dm_cache_policy_get_hint_size(cache->policy),
+				     ca->features.metadata_version);
 	if (IS_ERR(cmd)) {
 		*error = "Error creating metadata object";
 		r = PTR_ERR(cmd);
@@ -3165,21 +3171,16 @@ static int cache_end_io(struct dm_target *ti, struct bio *bio, int error)
 
 static int write_dirty_bitset(struct cache *cache)
 {
-	unsigned i, r;
+	int r;
 
 	if (get_cache_mode(cache) >= CM_READ_ONLY)
 		return -EINVAL;
 
-	for (i = 0; i < from_cblock(cache->cache_size); i++) {
-		r = dm_cache_set_dirty(cache->cmd, to_cblock(i),
-				       is_dirty(cache, to_cblock(i)));
-		if (r) {
-			metadata_operation_failed(cache, "dm_cache_set_dirty", r);
-			return r;
-		}
-	}
+	r = dm_cache_set_dirty_bits(cache->cmd, from_cblock(cache->cache_size), cache->dirty_bitset);
+	if (r)
+		metadata_operation_failed(cache, "dm_cache_set_dirty_bits", r);
 
-	return 0;
+	return r;
 }
 
 static int write_discard_bitset(struct cache *cache)
@@ -3540,11 +3541,11 @@ static void cache_status(struct dm_target *ti, status_type_t type,
 
 		residency = policy_residency(cache->policy);
 
-		DMEMIT("%u %llu/%llu %u %llu/%llu %u %u %u %u %u %u %lu ",
+		DMEMIT("%u %llu/%llu %llu %llu/%llu %u %u %u %u %u %u %lu ",
 		       (unsigned)DM_CACHE_METADATA_BLOCK_SIZE,
 		       (unsigned long long)(nr_blocks_metadata - nr_free_blocks_metadata),
 		       (unsigned long long)nr_blocks_metadata,
-		       cache->sectors_per_block,
+		       (unsigned long long)cache->sectors_per_block,
 		       (unsigned long long) from_cblock(residency),
 		       (unsigned long long) from_cblock(cache->cache_size),
 		       (unsigned) atomic_read(&cache->stats.read_hit),
@@ -3555,14 +3556,19 @@ static void cache_status(struct dm_target *ti, status_type_t type,
 		       (unsigned) atomic_read(&cache->stats.promotion),
 		       (unsigned long) atomic_read(&cache->nr_dirty));
 
+		if (cache->features.metadata_version == 2)
+			DMEMIT("2 metadata2 ");
+		else
+			DMEMIT("1 ");
+
 		if (writethrough_mode(&cache->features))
-			DMEMIT("1 writethrough ");
+			DMEMIT("writethrough ");
 
 		else if (passthrough_mode(&cache->features))
-			DMEMIT("1 passthrough ");
+			DMEMIT("passthrough ");
 
 		else if (writeback_mode(&cache->features))
-			DMEMIT("1 writeback ");
+			DMEMIT("writeback ");
 
 		else {
 			DMERR("%s: internal error: unknown io mode: %d",
@@ -3810,7 +3816,7 @@ static void cache_io_hints(struct dm_target *ti, struct queue_limits *limits)
 
 static struct target_type cache_target = {
 	.name = "cache",
-	.version = {1, 9, 0},
+	.version = {1, 10, 0},
 	.module = THIS_MODULE,
 	.ctr = cache_ctr,
 	.dtr = cache_dtr,

drivers/md/dm-raid.c

@@ -24,6 +24,11 @@
  */
 #define	MIN_FREE_RESHAPE_SPACE to_sector(4*4096)
 
+/*
+ * Minimum journal space 4 MiB in sectors.
+ */
+#define	MIN_RAID456_JOURNAL_SPACE (4*2048)
+
 static bool devices_handle_discard_safely = false;
 
 /*
@@ -73,6 +78,9 @@ struct raid_dev {
 #define __CTR_FLAG_DATA_OFFSET		13 /* 2 */ /* Only with reshapable raid4/5/6/10! */
 #define __CTR_FLAG_RAID10_USE_NEAR_SETS 14 /* 2 */ /* Only with raid10! */
 
+/* New for v1.10.0 */
+#define __CTR_FLAG_JOURNAL_DEV		15 /* 2 */ /* Only with raid4/5/6! */
+
 /*
  * Flags for rs->ctr_flags field.
  */
@@ -91,6 +99,7 @@ struct raid_dev {
 #define CTR_FLAG_DELTA_DISKS		(1 << __CTR_FLAG_DELTA_DISKS)
 #define CTR_FLAG_DATA_OFFSET		(1 << __CTR_FLAG_DATA_OFFSET)
 #define CTR_FLAG_RAID10_USE_NEAR_SETS	(1 << __CTR_FLAG_RAID10_USE_NEAR_SETS)
+#define CTR_FLAG_JOURNAL_DEV		(1 << __CTR_FLAG_JOURNAL_DEV)
 
 /*
  * Definitions of various constructor flags to
@@ -163,7 +172,8 @@ struct raid_dev {
 				 CTR_FLAG_STRIPE_CACHE | \
 				 CTR_FLAG_REGION_SIZE | \
 				 CTR_FLAG_DELTA_DISKS | \
-				 CTR_FLAG_DATA_OFFSET)
+				 CTR_FLAG_DATA_OFFSET | \
+				 CTR_FLAG_JOURNAL_DEV)
 
 #define RAID6_VALID_FLAGS	(CTR_FLAG_SYNC | \
 				 CTR_FLAG_REBUILD | \
@@ -173,7 +183,8 @@ struct raid_dev {
 				 CTR_FLAG_STRIPE_CACHE | \
 				 CTR_FLAG_REGION_SIZE | \
 				 CTR_FLAG_DELTA_DISKS | \
-				 CTR_FLAG_DATA_OFFSET)
+				 CTR_FLAG_DATA_OFFSET | \
+				 CTR_FLAG_JOURNAL_DEV)
 /* ...valid options definitions per raid level */
 
 /*
@@ -222,6 +233,12 @@ struct raid_set {
 	struct raid_type *raid_type;
 	struct dm_target_callbacks callbacks;
 
+	/* Optional raid4/5/6 journal device */
+	struct journal_dev {
+		struct dm_dev *dev;
+		struct md_rdev rdev;
+	} journal_dev;
+
 	struct raid_dev dev[0];
 };
 
@@ -306,6 +323,7 @@ static struct arg_name_flag {
 	{ CTR_FLAG_DATA_OFFSET, "data_offset"},
 	{ CTR_FLAG_DELTA_DISKS, "delta_disks"},
 	{ CTR_FLAG_RAID10_USE_NEAR_SETS, "raid10_use_near_sets"},
+	{ CTR_FLAG_JOURNAL_DEV, "journal_dev" },
 };
 
 /* Return argument name string for given @flag */
@@ -370,7 +388,7 @@ static bool rs_is_reshapable(struct raid_set *rs)
 /* Return true, if raid set in @rs is recovering */
 static bool rs_is_recovering(struct raid_set *rs)
 {
-	return rs->md.recovery_cp < rs->dev[0].rdev.sectors;
+	return rs->md.recovery_cp < rs->md.dev_sectors;
 }
 
 /* Return true, if raid set in @rs is reshaping */
@@ -627,7 +645,8 @@ static void rs_set_capacity(struct raid_set *rs)
 	 * is unintended in case of out-of-place reshaping
 	 */
 	rdev_for_each(rdev, mddev)
-		rdev->sectors = mddev->dev_sectors;
+		if (!test_bit(Journal, &rdev->flags))
+			rdev->sectors = mddev->dev_sectors;
 
 	set_capacity(gendisk, mddev->array_sectors);
 	revalidate_disk(gendisk);
@@ -713,6 +732,11 @@ static void raid_set_free(struct raid_set *rs)
 {
 	int i;
 
+	if (rs->journal_dev.dev) {
+		md_rdev_clear(&rs->journal_dev.rdev);
+		dm_put_device(rs->ti, rs->journal_dev.dev);
+	}
+
 	for (i = 0; i < rs->raid_disks; i++) {
 		if (rs->dev[i].meta_dev)
 			dm_put_device(rs->ti, rs->dev[i].meta_dev);
@@ -760,10 +784,11 @@ static int parse_dev_params(struct raid_set *rs, struct dm_arg_set *as)
 		rs->dev[i].data_dev = NULL;
 
 		/*
-		 * There are no offsets, since there is a separate device
-		 * for data and metadata.
+		 * There are no offsets initially.
+		 * Out of place reshape will set them accordingly.
 		 */
 		rs->dev[i].rdev.data_offset = 0;
+		rs->dev[i].rdev.new_data_offset = 0;
 		rs->dev[i].rdev.mddev = &rs->md;
 
 		arg = dm_shift_arg(as);
@@ -821,6 +846,9 @@ static int parse_dev_params(struct raid_set *rs, struct dm_arg_set *as)
 			rebuild++;
 	}
 
+	if (rs->journal_dev.dev)
+		list_add_tail(&rs->journal_dev.rdev.same_set, &rs->md.disks);
+
 	if (metadata_available) {
 		rs->md.external = 0;
 		rs->md.persistent = 1;
@@ -1026,6 +1054,8 @@ static int validate_raid_redundancy(struct raid_set *rs)
  *    [max_write_behind <sectors>]	See '-write-behind=' (man mdadm)
  *    [stripe_cache <sectors>]		Stripe cache size for higher RAIDs
  *    [region_size <sectors>]		Defines granularity of bitmap
+ *    [journal_dev <dev>]		raid4/5/6 journaling deviice
+ *    					(i.e. write hole closing log)
  *
  * RAID10-only options:
  *    [raid10_copies <# copies>]	Number of copies.  (Default: 2)
@@ -1133,7 +1163,7 @@ static int parse_raid_params(struct raid_set *rs, struct dm_arg_set *as,
 		/*
 		 * Parameters that take a string value are checked here.
 		 */
-
+		/* "raid10_format {near|offset|far} */
 		if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_RAID10_FORMAT))) {
 			if (test_and_set_bit(__CTR_FLAG_RAID10_FORMAT, &rs->ctr_flags)) {
 				rs->ti->error = "Only one 'raid10_format' argument pair allowed";
@@ -1151,6 +1181,41 @@ static int parse_raid_params(struct raid_set *rs, struct dm_arg_set *as,
 			continue;
 		}
 
+		/* "journal_dev dev" */
+		if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_JOURNAL_DEV))) {
+			int r;
+			struct md_rdev *jdev;
+
+			if (test_and_set_bit(__CTR_FLAG_JOURNAL_DEV, &rs->ctr_flags)) {
+				rs->ti->error = "Only one raid4/5/6 set journaling device allowed";
+				return -EINVAL;
+			}
+			if (!rt_is_raid456(rt)) {
+				rs->ti->error = "'journal_dev' is an invalid parameter for this RAID type";
+				return -EINVAL;
+			}
+			r = dm_get_device(rs->ti, arg, dm_table_get_mode(rs->ti->table),
+					  &rs->journal_dev.dev);
+			if (r) {
+				rs->ti->error = "raid4/5/6 journal device lookup failure";
+				return r;
+			}
+			jdev = &rs->journal_dev.rdev;
+			md_rdev_init(jdev);
+			jdev->mddev = &rs->md;
+			jdev->bdev = rs->journal_dev.dev->bdev;
+			jdev->sectors = to_sector(i_size_read(jdev->bdev->bd_inode));
+			if (jdev->sectors < MIN_RAID456_JOURNAL_SPACE) {
+				rs->ti->error = "No space for raid4/5/6 journal";
+				return -ENOSPC;
+			}
+			set_bit(Journal, &jdev->flags);
+			continue;
+		}
+
+		/*
+		 * Parameters with number values from here on.
+		 */
 		if (kstrtoint(arg, 10, &value) < 0) {
 			rs->ti->error = "Bad numerical argument given in raid params";
 			return -EINVAL;
@@ -1425,6 +1490,25 @@ static unsigned int rs_data_stripes(struct raid_set *rs)
 	return rs->raid_disks - rs->raid_type->parity_devs;
 }
 
+/*
+ * Retrieve rdev->sectors from any valid raid device of @rs
+ * to allow userpace to pass in arbitray "- -" device tupples.
+ */
+static sector_t __rdev_sectors(struct raid_set *rs)
+{
+	int i;
+
+	for (i = 0; i < rs->md.raid_disks; i++) {
+		struct md_rdev *rdev = &rs->dev[i].rdev;
+
+		if (!test_bit(Journal, &rdev->flags) &&
+		    rdev->bdev && rdev->sectors)
+			return rdev->sectors;
+	}
+
+	BUG(); /* Constructor ensures we got some. */
+}
+
 /* Calculate the sectors per device and per array used for @rs */
 static int rs_set_dev_and_array_sectors(struct raid_set *rs, bool use_mddev)
 {
@@ -1468,7 +1552,8 @@ static int rs_set_dev_and_array_sectors(struct raid_set *rs, bool use_mddev)
 		array_sectors = (data_stripes + delta_disks) * dev_sectors;
 
 	rdev_for_each(rdev, mddev)
-		rdev->sectors = dev_sectors;
+		if (!test_bit(Journal, &rdev->flags))
+			rdev->sectors = dev_sectors;
 
 	mddev->array_sectors = array_sectors;
 	mddev->dev_sectors = dev_sectors;
@@ -1510,9 +1595,9 @@ static void rs_setup_recovery(struct raid_set *rs, sector_t dev_sectors)
 	else if (dev_sectors == MaxSector)
 		/* Prevent recovery */
 		__rs_setup_recovery(rs, MaxSector);
-	else if (rs->dev[0].rdev.sectors < dev_sectors)
+	else if (__rdev_sectors(rs) < dev_sectors)
 		/* Grown raid set */
-		__rs_setup_recovery(rs, rs->dev[0].rdev.sectors);
+		__rs_setup_recovery(rs, __rdev_sectors(rs));
 	else
 		__rs_setup_recovery(rs, MaxSector);
 }
@@ -1851,18 +1936,21 @@ static int rs_check_reshape(struct raid_set *rs)
 	return -EPERM;
 }
 
-static int read_disk_sb(struct md_rdev *rdev, int size)
+static int read_disk_sb(struct md_rdev *rdev, int size, bool force_reload)
 {
 	BUG_ON(!rdev->sb_page);
 
-	if (rdev->sb_loaded)
+	if (rdev->sb_loaded && !force_reload)
 		return 0;
 
+	rdev->sb_loaded = 0;
+
 	if (!sync_page_io(rdev, 0, size, rdev->sb_page, REQ_OP_READ, 0, true)) {
 		DMERR("Failed to read superblock of device at position %d",
 		      rdev->raid_disk);
 		md_error(rdev->mddev, rdev);
-		return -EINVAL;
+		set_bit(Faulty, &rdev->flags);
+		return -EIO;
 	}
 
 	rdev->sb_loaded = 1;
@@ -1990,7 +2078,7 @@ static int super_load(struct md_rdev *rdev, struct md_rdev *refdev)
 		return -EINVAL;
 	}
 
-	r = read_disk_sb(rdev, rdev->sb_size);
+	r = read_disk_sb(rdev, rdev->sb_size, false);
 	if (r)
 		return r;
 
@@ -2146,6 +2234,9 @@ static int super_init_validation(struct raid_set *rs, struct md_rdev *rdev)
 	 */
 	d = 0;
 	rdev_for_each(r, mddev) {
+		if (test_bit(Journal, &rdev->flags))
+			continue;
+
 		if (test_bit(FirstUse, &r->flags))
 			new_devs++;
 
@@ -2201,7 +2292,8 @@ static int super_init_validation(struct raid_set *rs, struct md_rdev *rdev)
 	 */
 	sb_retrieve_failed_devices(sb, failed_devices);
 	rdev_for_each(r, mddev) {
-		if (!r->sb_page)
+		if (test_bit(Journal, &rdev->flags) ||
+		    !r->sb_page)
 			continue;
 		sb2 = page_address(r->sb_page);
 		sb2->failed_devices = 0;
@@ -2253,7 +2345,7 @@ static int super_validate(struct raid_set *rs, struct md_rdev *rdev)
 	struct mddev *mddev = &rs->md;
 	struct dm_raid_superblock *sb;
 
-	if (rs_is_raid0(rs) || !rdev->sb_page)
+	if (rs_is_raid0(rs) || !rdev->sb_page || rdev->raid_disk < 0)
 		return 0;
 
 	sb = page_address(rdev->sb_page);
@@ -2278,7 +2370,7 @@ static int super_validate(struct raid_set *rs, struct md_rdev *rdev)
 
 	/* Enable bitmap creation for RAID levels != 0 */
 	mddev->bitmap_info.offset = rt_is_raid0(rs->raid_type) ? 0 : to_sector(4096);
-	rdev->mddev->bitmap_info.default_offset = mddev->bitmap_info.offset;
+	mddev->bitmap_info.default_offset = mddev->bitmap_info.offset;
 
 	if (!test_and_clear_bit(FirstUse, &rdev->flags)) {
 		/* Retrieve device size stored in superblock to be prepared for shrink */
@@ -2316,21 +2408,22 @@ static int super_validate(struct raid_set *rs, struct md_rdev *rdev)
 static int analyse_superblocks(struct dm_target *ti, struct raid_set *rs)
 {
 	int r;
-	struct raid_dev *dev;
-	struct md_rdev *rdev, *tmp, *freshest;
+	struct md_rdev *rdev, *freshest;
 	struct mddev *mddev = &rs->md;
 
 	freshest = NULL;
-	rdev_for_each_safe(rdev, tmp, mddev) {
+	rdev_for_each(rdev, mddev) {
+		if (test_bit(Journal, &rdev->flags))
+			continue;
+
 		/*
 		 * Skipping super_load due to CTR_FLAG_SYNC will cause
 		 * the array to undergo initialization again as
 		 * though it were new.	This is the intended effect
 		 * of the "sync" directive.
 		 *
-		 * When reshaping capability is added, we must ensure
-		 * that the "sync" directive is disallowed during the
-		 * reshape.
+		 * With reshaping capability added, we must ensure that
+		 * that the "sync" directive is disallowed during the reshape.
 		 */
 		if (test_bit(__CTR_FLAG_SYNC, &rs->ctr_flags))
 			continue;
@@ -2347,6 +2440,7 @@ static int analyse_superblocks(struct dm_target *ti, struct raid_set *rs)
 		case 0:
 			break;
 		default:
+			/* This is a failure to read the superblock from the metadata device. */
 			/*
 			 * We have to keep any raid0 data/metadata device pairs or
 			 * the MD raid0 personality will fail to start the array.
@@ -2354,33 +2448,16 @@ static int analyse_superblocks(struct dm_target *ti, struct raid_set *rs)
 			if (rs_is_raid0(rs))
 				continue;
 
-			dev = container_of(rdev, struct raid_dev, rdev);
-			if (dev->meta_dev)
-				dm_put_device(ti, dev->meta_dev);
-
-			dev->meta_dev = NULL;
-			rdev->meta_bdev = NULL;
-
-			if (rdev->sb_page)
-				put_page(rdev->sb_page);
-
-			rdev->sb_page = NULL;
-
-			rdev->sb_loaded = 0;
-
 			/*
-			 * We might be able to salvage the data device
-			 * even though the meta device has failed.  For
-			 * now, we behave as though '- -' had been
-			 * set for this device in the table.
+			 * We keep the dm_devs to be able to emit the device tuple
+			 * properly on the table line in raid_status() (rather than
+			 * mistakenly acting as if '- -' got passed into the constructor).
+			 *
+			 * The rdev has to stay on the same_set list to allow for
+			 * the attempt to restore faulty devices on second resume.
 			 */
-			if (dev->data_dev)
-				dm_put_device(ti, dev->data_dev);
-
-			dev->data_dev = NULL;
-			rdev->bdev = NULL;
-
-			list_del(&rdev->same_set);
+			rdev->raid_disk = rdev->saved_raid_disk = -1;
+			break;
 		}
 	}
 
@@ -2401,7 +2478,9 @@ static int analyse_superblocks(struct dm_target *ti, struct raid_set *rs)
 		return -EINVAL;
 
 	rdev_for_each(rdev, mddev)
-		if ((rdev != freshest) && super_validate(rs, rdev))
+		if (!test_bit(Journal, &rdev->flags) &&
+		    rdev != freshest &&
+		    super_validate(rs, rdev))
 			return -EINVAL;
 	return 0;
 }
@@ -2488,10 +2567,12 @@ static int rs_adjust_data_offsets(struct raid_set *rs)
 		return -ENOSPC;
 	}
 out:
-	/* Adjust data offsets on all rdevs */
+	/* Adjust data offsets on all rdevs but on any raid4/5/6 journal device */
 	rdev_for_each(rdev, &rs->md) {
-		rdev->data_offset = data_offset;
-		rdev->new_data_offset = new_data_offset;
+		if (!test_bit(Journal, &rdev->flags)) {
+			rdev->data_offset = data_offset;
+			rdev->new_data_offset = new_data_offset;
+		}
 	}
 
 	return 0;
@@ -2504,8 +2585,10 @@ static void __reorder_raid_disk_indexes(struct raid_set *rs)
 	struct md_rdev *rdev;
 
 	rdev_for_each(rdev, &rs->md) {
-		rdev->raid_disk = i++;
-		rdev->saved_raid_disk = rdev->new_raid_disk = -1;
+		if (!test_bit(Journal, &rdev->flags)) {
+			rdev->raid_disk = i++;
+			rdev->saved_raid_disk = rdev->new_raid_disk = -1;
+		}
 	}
 }
 
@@ -2845,7 +2928,7 @@ static int raid_ctr(struct dm_target *ti, unsigned int argc, char **argv)
 	if (r)
 		goto bad;
 
-	calculated_dev_sectors = rs->dev[0].rdev.sectors;
+	calculated_dev_sectors = rs->md.dev_sectors;
 
 	/*
 	 * Backup any new raid set level, layout, ...
@@ -2858,7 +2941,7 @@ static int raid_ctr(struct dm_target *ti, unsigned int argc, char **argv)
 	if (r)
 		goto bad;
 
-	resize = calculated_dev_sectors != rs->dev[0].rdev.sectors;
+	resize = calculated_dev_sectors != __rdev_sectors(rs);
 
 	INIT_WORK(&rs->md.event_work, do_table_event);
 	ti->private = rs;
@@ -2902,6 +2985,13 @@ static int raid_ctr(struct dm_target *ti, unsigned int argc, char **argv)
 			goto bad;
 		}
 
+		/* We can't takeover a journaled raid4/5/6 */
+		if (test_bit(__CTR_FLAG_JOURNAL_DEV, &rs->ctr_flags)) {
+			ti->error = "Can't takeover a journaled raid4/5/6 set";
+			r = -EPERM;
+			goto bad;
+		}
+
 		/*
 		 * If a takeover is needed, userspace sets any additional
 		 * devices to rebuild and we can check for a valid request here.
@@ -2923,6 +3013,18 @@ static int raid_ctr(struct dm_target *ti, unsigned int argc, char **argv)
 		rs_setup_recovery(rs, MaxSector);
 		rs_set_new(rs);
 	} else if (rs_reshape_requested(rs)) {
+		/*
+		 * No need to check for 'ongoing' takeover here, because takeover
+		 * is an instant operation as oposed to an ongoing reshape.
+		 */
+
+		/* We can't reshape a journaled raid4/5/6 */
+		if (test_bit(__CTR_FLAG_JOURNAL_DEV, &rs->ctr_flags)) {
+			ti->error = "Can't reshape a journaled raid4/5/6 set";
+			r = -EPERM;
+			goto bad;
+		}
+
 		/*
 		  * We can only prepare for a reshape here, because the
 		  * raid set needs to run to provide the repective reshape
@@ -3071,18 +3173,23 @@ static const char *decipher_sync_action(struct mddev *mddev)
 }
 
 /*
- * Return status string @rdev
+ * Return status string for @rdev
  *
  * Status characters:
  *
- *  'D' = Dead/Failed device
+ *  'D' = Dead/Failed raid set component or raid4/5/6 journal device
  *  'a' = Alive but not in-sync
- *  'A' = Alive and in-sync
+ *  'A' = Alive and in-sync raid set component or alive raid4/5/6 journal device
+ *  '-' = Non-existing device (i.e. uspace passed '- -' into the ctr)
  */
 static const char *__raid_dev_status(struct md_rdev *rdev, bool array_in_sync)
 {
-	if (test_bit(Faulty, &rdev->flags))
+	if (!rdev->bdev)
+		return "-";
+	else if (test_bit(Faulty, &rdev->flags))
 		return "D";
+	else if (test_bit(Journal, &rdev->flags))
+		return "A";
 	else if (!array_in_sync || !test_bit(In_sync, &rdev->flags))
 		return "a";
 	else
@@ -3151,7 +3258,8 @@ static sector_t rs_get_progress(struct raid_set *rs,
 			 * being initialized.
 			 */
 			rdev_for_each(rdev, mddev)
-				if (!test_bit(In_sync, &rdev->flags))
+				if (!test_bit(Journal, &rdev->flags) &&
+				    !test_bit(In_sync, &rdev->flags))
 					*array_in_sync = true;
 #if 0
 			r = 0; /* HM FIXME: TESTME: https://bugzilla.redhat.com/show_bug.cgi?id=1210637 ? */
@@ -3183,7 +3291,6 @@ static void raid_status(struct dm_target *ti, status_type_t type,
 	sector_t progress, resync_max_sectors, resync_mismatches;
 	const char *sync_action;
 	struct raid_type *rt;
-	struct md_rdev *rdev;
 
 	switch (type) {
 	case STATUSTYPE_INFO:
@@ -3204,9 +3311,9 @@ static void raid_status(struct dm_target *ti, status_type_t type,
 				    atomic64_read(&mddev->resync_mismatches) : 0;
 		sync_action = decipher_sync_action(&rs->md);
 
-		/* HM FIXME: do we want another state char for raid0? It shows 'D' or 'A' now */
-		rdev_for_each(rdev, mddev)
-			DMEMIT(__raid_dev_status(rdev, array_in_sync));
+		/* HM FIXME: do we want another state char for raid0? It shows 'D'/'A'/'-' now */
+		for (i = 0; i < rs->raid_disks; i++)
+			DMEMIT(__raid_dev_status(&rs->dev[i].rdev, array_in_sync));
 
 		/*
 		 * In-sync/Reshape ratio:
@@ -3252,6 +3359,12 @@ static void raid_status(struct dm_target *ti, status_type_t type,
 		 * so retrieving it from the first raid disk is sufficient.
 		 */
 		DMEMIT(" %llu", (unsigned long long) rs->dev[0].rdev.data_offset);
+
+		/*
+		 * v1.10.0+:
+		 */
+		DMEMIT(" %s", test_bit(__CTR_FLAG_JOURNAL_DEV, &rs->ctr_flags) ?
+			      __raid_dev_status(&rs->journal_dev.rdev, 0) : "-");
 		break;
 
 	case STATUSTYPE_TABLE:
@@ -3265,7 +3378,8 @@ static void raid_status(struct dm_target *ti, status_type_t type,
 		raid_param_cnt += rebuild_disks * 2 +
 				  write_mostly_params +
 				  hweight32(rs->ctr_flags & CTR_FLAG_OPTIONS_NO_ARGS) +
-				  hweight32(rs->ctr_flags & CTR_FLAG_OPTIONS_ONE_ARG) * 2;
+				  hweight32(rs->ctr_flags & CTR_FLAG_OPTIONS_ONE_ARG) * 2 +
+				  (test_bit(__CTR_FLAG_JOURNAL_DEV, &rs->ctr_flags) ? 2 : 0);
 		/* Emit table line */
 		DMEMIT("%s %u %u", rs->raid_type->name, raid_param_cnt, mddev->new_chunk_sectors);
 		if (test_bit(__CTR_FLAG_RAID10_FORMAT, &rs->ctr_flags))
@@ -3312,6 +3426,9 @@ static void raid_status(struct dm_target *ti, status_type_t type,
 		if (test_bit(__CTR_FLAG_MIN_RECOVERY_RATE, &rs->ctr_flags))
 			DMEMIT(" %s %d", dm_raid_arg_name_by_flag(CTR_FLAG_MIN_RECOVERY_RATE),
 					 mddev->sync_speed_min);
+		if (test_bit(__CTR_FLAG_JOURNAL_DEV, &rs->ctr_flags))
+			DMEMIT(" %s %s", dm_raid_arg_name_by_flag(CTR_FLAG_JOURNAL_DEV),
+					__get_dev_name(rs->journal_dev.dev));
 		DMEMIT(" %d", rs->raid_disks);
 		for (i = 0; i < rs->raid_disks; i++)
 			DMEMIT(" %s %s", __get_dev_name(rs->dev[i].meta_dev),
@@ -3347,10 +3464,11 @@ static int raid_message(struct dm_target *ti, unsigned int argc, char **argv)
 	else {
 		if (!strcasecmp(argv[0], "check"))
 			set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
-		else if (!!strcasecmp(argv[0], "repair"))
+		else if (!strcasecmp(argv[0], "repair")) {
+			set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
+			set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
+		} else
 			return -EINVAL;
-		set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
-		set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
 	}
 	if (mddev->ro == 2) {
 		/* A write to sync_action is enough to justify
@@ -3427,11 +3545,14 @@ static void attempt_restore_of_faulty_devices(struct raid_set *rs)
 
 	memset(cleared_failed_devices, 0, sizeof(cleared_failed_devices));
 
-	for (i = 0; i < rs->md.raid_disks; i++) {
+	for (i = 0; i < mddev->raid_disks; i++) {
 		r = &rs->dev[i].rdev;
-		if (test_bit(Faulty, &r->flags) && r->sb_page &&
-		    sync_page_io(r, 0, r->sb_size, r->sb_page,
-				 REQ_OP_READ, 0, true)) {
+		/* HM FIXME: enhance journal device recovery processing */
+		if (test_bit(Journal, &r->flags))
+			continue;
+
+		if (test_bit(Faulty, &r->flags) &&
+		    r->meta_bdev && !read_disk_sb(r, r->sb_size, true)) {
 			DMINFO("Faulty %s device #%d has readable super block."
 			       "  Attempting to revive it.",
 			       rs->raid_type->name, i);
@@ -3445,22 +3566,26 @@ static void attempt_restore_of_faulty_devices(struct raid_set *rs)
 			 * '>= 0' - meaning we must call this function
 			 * ourselves.
 			 */
-			if ((r->raid_disk >= 0) &&
-			    (mddev->pers->hot_remove_disk(mddev, r) != 0))
-				/* Failed to revive this device, try next */
-				continue;
-
-			r->raid_disk = i;
-			r->saved_raid_disk = i;
 			flags = r->flags;
+			clear_bit(In_sync, &r->flags); /* Mandatory for hot remove. */
+			if (r->raid_disk >= 0) {
+				if (mddev->pers->hot_remove_disk(mddev, r)) {
+					/* Failed to revive this device, try next */
+					r->flags = flags;
+					continue;
+				}
+			} else
+				r->raid_disk = r->saved_raid_disk = i;
+
 			clear_bit(Faulty, &r->flags);
 			clear_bit(WriteErrorSeen, &r->flags);
-			clear_bit(In_sync, &r->flags);
+
 			if (mddev->pers->hot_add_disk(mddev, r)) {
-				r->raid_disk = -1;
-				r->saved_raid_disk = -1;
+				/* Failed to revive this device, try next */
+				r->raid_disk = r->saved_raid_disk = -1;
 				r->flags = flags;
 			} else {
+				clear_bit(In_sync, &r->flags);
 				r->recovery_offset = 0;
 				set_bit(i, (void *) cleared_failed_devices);
 				cleared = true;
@@ -3473,6 +3598,9 @@ static void attempt_restore_of_faulty_devices(struct raid_set *rs)
 		uint64_t failed_devices[DISKS_ARRAY_ELEMS];
 
 		rdev_for_each(r, &rs->md) {
+			if (test_bit(Journal, &r->flags))
+				continue;
+
 			sb = page_address(r->sb_page);
 			sb_retrieve_failed_devices(sb, failed_devices);
 
@@ -3651,7 +3779,7 @@ static void raid_resume(struct dm_target *ti)
 
 static struct target_type raid_target = {
 	.name = "raid",
-	.version = {1, 9, 1},
+	.version = {1, 10, 0},
 	.module = THIS_MODULE,
 	.ctr = raid_ctr,
 	.dtr = raid_dtr,

drivers/md/dm-round-robin.c

@@ -17,8 +17,8 @@
 #include <linux/module.h>
 
 #define DM_MSG_PREFIX "multipath round-robin"
-#define RR_MIN_IO     1000
-#define RR_VERSION    "1.1.0"
+#define RR_MIN_IO     1
+#define RR_VERSION    "1.2.0"
 
 /*-----------------------------------------------------------------
  * Path-handling code, paths are held in lists
@@ -47,44 +47,19 @@ struct selector {
 	struct list_head valid_paths;
 	struct list_head invalid_paths;
 	spinlock_t lock;
-	struct dm_path * __percpu *current_path;
-	struct percpu_counter repeat_count;
 };
 
-static void set_percpu_current_path(struct selector *s, struct dm_path *path)
-{
-	int cpu;
-
-	for_each_possible_cpu(cpu)
-		*per_cpu_ptr(s->current_path, cpu) = path;
-}
-
 static struct selector *alloc_selector(void)
 {
 	struct selector *s = kmalloc(sizeof(*s), GFP_KERNEL);
 
-	if (!s)
-		return NULL;
-
-	INIT_LIST_HEAD(&s->valid_paths);
-	INIT_LIST_HEAD(&s->invalid_paths);
-	spin_lock_init(&s->lock);
-
-	s->current_path = alloc_percpu(struct dm_path *);
-	if (!s->current_path)
-		goto out_current_path;
-	set_percpu_current_path(s, NULL);
-
-	if (percpu_counter_init(&s->repeat_count, 0, GFP_KERNEL))
-		goto out_repeat_count;
+	if (s) {
+		INIT_LIST_HEAD(&s->valid_paths);
+		INIT_LIST_HEAD(&s->invalid_paths);
+		spin_lock_init(&s->lock);
+	}
 
 	return s;
-
-out_repeat_count:
-	free_percpu(s->current_path);
-out_current_path:
-	kfree(s);
-	return NULL;;
 }
 
 static int rr_create(struct path_selector *ps, unsigned argc, char **argv)
@@ -105,8 +80,6 @@ static void rr_destroy(struct path_selector *ps)
 
 	free_paths(&s->valid_paths);
 	free_paths(&s->invalid_paths);
-	free_percpu(s->current_path);
-	percpu_counter_destroy(&s->repeat_count);
 	kfree(s);
 	ps->context = NULL;
 }
@@ -157,6 +130,11 @@ static int rr_add_path(struct path_selector *ps, struct dm_path *path,
 		return -EINVAL;
 	}
 
+	if (repeat_count > 1) {
+		DMWARN_LIMIT("repeat_count > 1 is deprecated, using 1 instead");
+		repeat_count = 1;
+	}
+
 	/* allocate the path */
 	pi = kmalloc(sizeof(*pi), GFP_KERNEL);
 	if (!pi) {
@@ -183,9 +161,6 @@ static void rr_fail_path(struct path_selector *ps, struct dm_path *p)
 	struct path_info *pi = p->pscontext;
 
 	spin_lock_irqsave(&s->lock, flags);
-	if (p == *this_cpu_ptr(s->current_path))
-		set_percpu_current_path(s, NULL);
-
 	list_move(&pi->list, &s->invalid_paths);
 	spin_unlock_irqrestore(&s->lock, flags);
 }
@@ -208,29 +183,15 @@ static struct dm_path *rr_select_path(struct path_selector *ps, size_t nr_bytes)
 	unsigned long flags;
 	struct selector *s = ps->context;
 	struct path_info *pi = NULL;
-	struct dm_path *current_path = NULL;
-
-	local_irq_save(flags);
-	current_path = *this_cpu_ptr(s->current_path);
-	if (current_path) {
-		percpu_counter_dec(&s->repeat_count);
-		if (percpu_counter_read_positive(&s->repeat_count) > 0) {
-			local_irq_restore(flags);
-			return current_path;
-		}
-	}
 
-	spin_lock(&s->lock);
+	spin_lock_irqsave(&s->lock, flags);
 	if (!list_empty(&s->valid_paths)) {
 		pi = list_entry(s->valid_paths.next, struct path_info, list);
 		list_move_tail(&pi->list, &s->valid_paths);
-		percpu_counter_set(&s->repeat_count, pi->repeat_count);
-		set_percpu_current_path(s, pi->path);
-		current_path = pi->path;
 	}
 	spin_unlock_irqrestore(&s->lock, flags);
 
-	return current_path;
+	return pi ? pi->path : NULL;
 }
 
 static struct path_selector_type rr_ps = {

drivers/md/dm-stats.c

@@ -175,6 +175,7 @@ static void dm_stat_free(struct rcu_head *head)
 	int cpu;
 	struct dm_stat *s = container_of(head, struct dm_stat, rcu_head);
 
+	kfree(s->histogram_boundaries);
 	kfree(s->program_id);
 	kfree(s->aux_data);
 	for_each_possible_cpu(cpu) {

drivers/md/dm.c

@@ -974,10 +974,61 @@ void dm_accept_partial_bio(struct bio *bio, unsigned n_sectors)
 }
 EXPORT_SYMBOL_GPL(dm_accept_partial_bio);
 
+/*
+ * Flush current->bio_list when the target map method blocks.
+ * This fixes deadlocks in snapshot and possibly in other targets.
+ */
+struct dm_offload {
+	struct blk_plug plug;
+	struct blk_plug_cb cb;
+};
+
+static void flush_current_bio_list(struct blk_plug_cb *cb, bool from_schedule)
+{
+	struct dm_offload *o = container_of(cb, struct dm_offload, cb);
+	struct bio_list list;
+	struct bio *bio;
+
+	INIT_LIST_HEAD(&o->cb.list);
+
+	if (unlikely(!current->bio_list))
+		return;
+
+	list = *current->bio_list;
+	bio_list_init(current->bio_list);
+
+	while ((bio = bio_list_pop(&list))) {
+		struct bio_set *bs = bio->bi_pool;
+		if (unlikely(!bs) || bs == fs_bio_set) {
+			bio_list_add(current->bio_list, bio);
+			continue;
+		}
+
+		spin_lock(&bs->rescue_lock);
+		bio_list_add(&bs->rescue_list, bio);
+		queue_work(bs->rescue_workqueue, &bs->rescue_work);
+		spin_unlock(&bs->rescue_lock);
+	}
+}
+
+static void dm_offload_start(struct dm_offload *o)
+{
+	blk_start_plug(&o->plug);
+	o->cb.callback = flush_current_bio_list;
+	list_add(&o->cb.list, &current->plug->cb_list);
+}
+
+static void dm_offload_end(struct dm_offload *o)
+{
+	list_del(&o->cb.list);
+	blk_finish_plug(&o->plug);
+}
+
 static void __map_bio(struct dm_target_io *tio)
 {
 	int r;
 	sector_t sector;
+	struct dm_offload o;
 	struct bio *clone = &tio->clone;
 	struct dm_target *ti = tio->ti;
 
@@ -990,7 +1041,11 @@ static void __map_bio(struct dm_target_io *tio)
 	 */
 	atomic_inc(&tio->io->io_count);
 	sector = clone->bi_iter.bi_sector;
+
+	dm_offload_start(&o);
 	r = ti->type->map(ti, clone);
+	dm_offload_end(&o);
+
 	if (r == DM_MAPIO_REMAPPED) {
 		/* the bio has been remapped so dispatch it */
 

drivers/md/persistent-data/dm-array.c

@@ -976,6 +976,27 @@ int dm_array_cursor_next(struct dm_array_cursor *c)
 }
 EXPORT_SYMBOL_GPL(dm_array_cursor_next);
 
+int dm_array_cursor_skip(struct dm_array_cursor *c, uint32_t count)
+{
+	int r;
+
+	do {
+		uint32_t remaining = le32_to_cpu(c->ab->nr_entries) - c->index;
+
+		if (count < remaining) {
+			c->index += count;
+			return 0;
+		}
+
+		count -= remaining;
+		r = dm_array_cursor_next(c);
+
+	} while (!r);
+
+	return r;
+}
+EXPORT_SYMBOL_GPL(dm_array_cursor_skip);
+
 void dm_array_cursor_get_value(struct dm_array_cursor *c, void **value_le)
 {
 	*value_le = element_at(c->info, c->ab, c->index);

drivers/md/persistent-data/dm-array.h

@@ -207,6 +207,7 @@ void dm_array_cursor_end(struct dm_array_cursor *c);
 
 uint32_t dm_array_cursor_index(struct dm_array_cursor *c);
 int dm_array_cursor_next(struct dm_array_cursor *c);
+int dm_array_cursor_skip(struct dm_array_cursor *c, uint32_t count);
 
 /*
  * value_le is only valid while the cursor points at the current value.

drivers/md/persistent-data/dm-bitset.c

@@ -39,6 +39,48 @@ int dm_bitset_empty(struct dm_disk_bitset *info, dm_block_t *root)
 }
 EXPORT_SYMBOL_GPL(dm_bitset_empty);
 
+struct packer_context {
+	bit_value_fn fn;
+	unsigned nr_bits;
+	void *context;
+};
+
+static int pack_bits(uint32_t index, void *value, void *context)
+{
+	int r;
+	struct packer_context *p = context;
+	unsigned bit, nr = min(64u, p->nr_bits - (index * 64));
+	uint64_t word = 0;
+	bool bv;
+
+	for (bit = 0; bit < nr; bit++) {
+		r = p->fn(index * 64 + bit, &bv, p->context);
+		if (r)
+			return r;
+
+		if (bv)
+			set_bit(bit, (unsigned long *) &word);
+		else
+			clear_bit(bit, (unsigned long *) &word);
+	}
+
+	*((__le64 *) value) = cpu_to_le64(word);
+
+	return 0;
+}
+
+int dm_bitset_new(struct dm_disk_bitset *info, dm_block_t *root,
+		  uint32_t size, bit_value_fn fn, void *context)
+{
+	struct packer_context p;
+	p.fn = fn;
+	p.nr_bits = size;
+	p.context = context;
+
+	return dm_array_new(&info->array_info, root, dm_div_up(size, 64), pack_bits, &p);
+}
+EXPORT_SYMBOL_GPL(dm_bitset_new);
+
 int dm_bitset_resize(struct dm_disk_bitset *info, dm_block_t root,
 		     uint32_t old_nr_entries, uint32_t new_nr_entries,
 		     bool default_value, dm_block_t *new_root)
@@ -168,4 +210,108 @@ int dm_bitset_test_bit(struct dm_disk_bitset *info, dm_block_t root,
 }
 EXPORT_SYMBOL_GPL(dm_bitset_test_bit);
 
+static int cursor_next_array_entry(struct dm_bitset_cursor *c)
+{
+	int r;
+	__le64 *value;
+
+	r = dm_array_cursor_next(&c->cursor);
+	if (r)
+		return r;
+
+	dm_array_cursor_get_value(&c->cursor, (void **) &value);
+	c->array_index++;
+	c->bit_index = 0;
+	c->current_bits = le64_to_cpu(*value);
+	return 0;
+}
+
+int dm_bitset_cursor_begin(struct dm_disk_bitset *info,
+			   dm_block_t root, uint32_t nr_entries,
+			   struct dm_bitset_cursor *c)
+{
+	int r;
+	__le64 *value;
+
+	if (!nr_entries)
+		return -ENODATA;
+
+	c->info = info;
+	c->entries_remaining = nr_entries;
+
+	r = dm_array_cursor_begin(&info->array_info, root, &c->cursor);
+	if (r)
+		return r;
+
+	dm_array_cursor_get_value(&c->cursor, (void **) &value);
+	c->array_index = 0;
+	c->bit_index = 0;
+	c->current_bits = le64_to_cpu(*value);
+
+	return r;
+}
+EXPORT_SYMBOL_GPL(dm_bitset_cursor_begin);
+
+void dm_bitset_cursor_end(struct dm_bitset_cursor *c)
+{
+	return dm_array_cursor_end(&c->cursor);
+}
+EXPORT_SYMBOL_GPL(dm_bitset_cursor_end);
+
+int dm_bitset_cursor_next(struct dm_bitset_cursor *c)
+{
+	int r = 0;
+
+	if (!c->entries_remaining)
+		return -ENODATA;
+
+	c->entries_remaining--;
+	if (++c->bit_index > 63)
+		r = cursor_next_array_entry(c);
+
+	return r;
+}
+EXPORT_SYMBOL_GPL(dm_bitset_cursor_next);
+
+int dm_bitset_cursor_skip(struct dm_bitset_cursor *c, uint32_t count)
+{
+	int r;
+	__le64 *value;
+	uint32_t nr_array_skip;
+	uint32_t remaining_in_word = 64 - c->bit_index;
+
+	if (c->entries_remaining < count)
+		return -ENODATA;
+
+	if (count < remaining_in_word) {
+		c->bit_index += count;
+		c->entries_remaining -= count;
+		return 0;
+
+	} else {
+		c->entries_remaining -= remaining_in_word;
+		count -= remaining_in_word;
+	}
+
+	nr_array_skip = (count / 64) + 1;
+	r = dm_array_cursor_skip(&c->cursor, nr_array_skip);
+	if (r)
+		return r;
+
+	dm_array_cursor_get_value(&c->cursor, (void **) &value);
+	c->entries_remaining -= count;
+	c->array_index += nr_array_skip;
+	c->bit_index = count & 63;
+	c->current_bits = le64_to_cpu(*value);
+
+	return 0;
+}
+EXPORT_SYMBOL_GPL(dm_bitset_cursor_skip);
+
+bool dm_bitset_cursor_get_value(struct dm_bitset_cursor *c)
+{
+	return test_bit(c->bit_index, (unsigned long *) &c->current_bits);
+}
+EXPORT_SYMBOL_GPL(dm_bitset_cursor_get_value);
+
 /*----------------------------------------------------------------*/

drivers/md/persistent-data/dm-bitset.h

@@ -92,6 +92,22 @@ void dm_disk_bitset_init(struct dm_transaction_manager *tm,
  */
 int dm_bitset_empty(struct dm_disk_bitset *info, dm_block_t *new_root);
 
+/*
+ * Creates a new bitset populated with values provided by a callback
+ * function.  This is more efficient than creating an empty bitset,
+ * resizing, and then setting values since that process incurs a lot of
+ * copying.
+ *
+ * info - describes the array
+ * root - the root block of the array on disk
+ * size - the number of entries in the array
+ * fn - the callback
+ * context - passed to the callback
+ */
+typedef int (*bit_value_fn)(uint32_t index, bool *value, void *context);
+int dm_bitset_new(struct dm_disk_bitset *info, dm_block_t *root,
+		  uint32_t size, bit_value_fn fn, void *context);
+
 /*
  * Resize the bitset.
  *
@@ -161,6 +177,29 @@ int dm_bitset_test_bit(struct dm_disk_bitset *info, dm_block_t root,
 int dm_bitset_flush(struct dm_disk_bitset *info, dm_block_t root,
 		    dm_block_t *new_root);
 
+struct dm_bitset_cursor {
+	struct dm_disk_bitset *info;
+	struct dm_array_cursor cursor;
+
+	uint32_t entries_remaining;
+	uint32_t array_index;
+	uint32_t bit_index;
+	uint64_t current_bits;
+};
+
+/*
+ * Make sure you've flush any dm_disk_bitset and updated the root before
+ * using this.
+ */
+int dm_bitset_cursor_begin(struct dm_disk_bitset *info,
+			   dm_block_t root, uint32_t nr_entries,
+			   struct dm_bitset_cursor *c);
+void dm_bitset_cursor_end(struct dm_bitset_cursor *c);
+
+int dm_bitset_cursor_next(struct dm_bitset_cursor *c);
+int dm_bitset_cursor_skip(struct dm_bitset_cursor *c, uint32_t count);
+bool dm_bitset_cursor_get_value(struct dm_bitset_cursor *c);
+
 /*----------------------------------------------------------------*/
 
 #endif /* _LINUX_DM_BITSET_H */

drivers/md/persistent-data/dm-block-manager.c

@@ -462,7 +462,7 @@ int dm_bm_read_lock(struct dm_block_manager *bm, dm_block_t b,
 	int r;
 
 	p = dm_bufio_read(bm->bufio, b, (struct dm_buffer **) result);
-	if (IS_ERR(p))
+	if (unlikely(IS_ERR(p)))
 		return PTR_ERR(p);
 
 	aux = dm_bufio_get_aux_data(to_buffer(*result));
@@ -498,7 +498,7 @@ int dm_bm_write_lock(struct dm_block_manager *bm,
 		return -EPERM;
 
 	p = dm_bufio_read(bm->bufio, b, (struct dm_buffer **) result);
-	if (IS_ERR(p))
+	if (unlikely(IS_ERR(p)))
 		return PTR_ERR(p);
 
 	aux = dm_bufio_get_aux_data(to_buffer(*result));
@@ -531,7 +531,7 @@ int dm_bm_read_try_lock(struct dm_block_manager *bm,
 	int r;
 
 	p = dm_bufio_get(bm->bufio, b, (struct dm_buffer **) result);
-	if (IS_ERR(p))
+	if (unlikely(IS_ERR(p)))
 		return PTR_ERR(p);
 	if (unlikely(!p))
 		return -EWOULDBLOCK;
@@ -567,7 +567,7 @@ int dm_bm_write_lock_zero(struct dm_block_manager *bm,
 		return -EPERM;
 
 	p = dm_bufio_new(bm->bufio, b, (struct dm_buffer **) result);
-	if (IS_ERR(p))
+	if (unlikely(IS_ERR(p)))
 		return PTR_ERR(p);
 
 	memset(p, 0, dm_bm_block_size(bm));

drivers/md/persistent-data/dm-btree.c

@@ -272,7 +272,12 @@ int dm_btree_del(struct dm_btree_info *info, dm_block_t root)
 	int r;
 	struct del_stack *s;
 
-	s = kmalloc(sizeof(*s), GFP_NOIO);
+	/*
+	 * dm_btree_del() is called via an ioctl, as such should be
+	 * considered an FS op.  We can't recurse back into the FS, so we
+	 * allocate GFP_NOFS.
+	 */
+	s = kmalloc(sizeof(*s), GFP_NOFS);
 	if (!s)
 		return -ENOMEM;
 	s->info = info;
@@ -1139,6 +1144,17 @@ int dm_btree_cursor_next(struct dm_btree_cursor *c)
 }
 EXPORT_SYMBOL_GPL(dm_btree_cursor_next);
 
+int dm_btree_cursor_skip(struct dm_btree_cursor *c, uint32_t count)
+{
+	int r = 0;
+
+	while (count-- && !r)
+		r = dm_btree_cursor_next(c);
+
+	return r;
+}
+EXPORT_SYMBOL_GPL(dm_btree_cursor_skip);
+
 int dm_btree_cursor_get_value(struct dm_btree_cursor *c, uint64_t *key, void *value_le)
 {
 	if (c->depth) {

drivers/md/persistent-data/dm-btree.h

@@ -209,6 +209,7 @@ int dm_btree_cursor_begin(struct dm_btree_info *info, dm_block_t root,
 			  bool prefetch_leaves, struct dm_btree_cursor *c);
 void dm_btree_cursor_end(struct dm_btree_cursor *c);
 int dm_btree_cursor_next(struct dm_btree_cursor *c);
+int dm_btree_cursor_skip(struct dm_btree_cursor *c, uint32_t count);
 int dm_btree_cursor_get_value(struct dm_btree_cursor *c, uint64_t *key, void *value_le);
 
 #endif	/* _LINUX_DM_BTREE_H */

drivers/md/persistent-data/dm-space-map-common.c

@@ -626,13 +626,19 @@ int sm_ll_open_metadata(struct ll_disk *ll, struct dm_transaction_manager *tm,
 			void *root_le, size_t len)
 {
 	int r;
-	struct disk_sm_root *smr = root_le;
+	struct disk_sm_root smr;
 
 	if (len < sizeof(struct disk_sm_root)) {
 		DMERR("sm_metadata root too small");
 		return -ENOMEM;
 	}
 
+	/*
+	 * We don't know the alignment of the root_le buffer, so need to
+	 * copy into a new structure.
+	 */
+	memcpy(&smr, root_le, sizeof(smr));
+
 	r = sm_ll_init(ll, tm);
 	if (r < 0)
 		return r;
@@ -644,10 +650,10 @@ int sm_ll_open_metadata(struct ll_disk *ll, struct dm_transaction_manager *tm,
 	ll->max_entries = metadata_ll_max_entries;
 	ll->commit = metadata_ll_commit;
 
-	ll->nr_blocks = le64_to_cpu(smr->nr_blocks);
-	ll->nr_allocated = le64_to_cpu(smr->nr_allocated);
-	ll->bitmap_root = le64_to_cpu(smr->bitmap_root);
-	ll->ref_count_root = le64_to_cpu(smr->ref_count_root);
+	ll->nr_blocks = le64_to_cpu(smr.nr_blocks);
+	ll->nr_allocated = le64_to_cpu(smr.nr_allocated);
+	ll->bitmap_root = le64_to_cpu(smr.bitmap_root);
+	ll->ref_count_root = le64_to_cpu(smr.ref_count_root);
 
 	return ll->open_index(ll);
 }

drivers/md/persistent-data/dm-space-map-metadata.c

@@ -544,7 +544,7 @@ static int sm_metadata_copy_root(struct dm_space_map *sm, void *where_le, size_t
 
 static int sm_metadata_extend(struct dm_space_map *sm, dm_block_t extra_blocks);
 
-static struct dm_space_map ops = {
+static const struct dm_space_map ops = {
 	.destroy = sm_metadata_destroy,
 	.extend = sm_metadata_extend,
 	.get_nr_blocks = sm_metadata_get_nr_blocks,
@@ -671,7 +671,7 @@ static int sm_bootstrap_copy_root(struct dm_space_map *sm, void *where,
 	return -EINVAL;
 }
 
-static struct dm_space_map bootstrap_ops = {
+static const struct dm_space_map bootstrap_ops = {
 	.destroy = sm_bootstrap_destroy,
 	.extend = sm_bootstrap_extend,
 	.get_nr_blocks = sm_bootstrap_get_nr_blocks,