dm raid: support metadata devices

Add the ability to parse and use metadata devices to dm-raid.  Although
not strictly required, without the metadata devices, many features of
RAID are unavailable.  They are used to store a superblock and bitmap.

The role, or position in the array, of each device must be recorded in
its superblock.  This is to help with fault handling, array reshaping,
and sanity checks.  RAID 4/5/6 devices must be loaded in a specific order:
in this way, the 'array_position' field helps validate the correctness
of the mapping when it is loaded.  It can be used during reshaping to
identify which devices are added/removed.  Fault handling is impossible
without this field.  For example, when a device fails it is recorded in
the superblock.  If this is a RAID1 device and the offending device is
removed from the array, there must be a way during subsequent array
assembly to determine that the failed device was the one removed.  This
is done by correlating the 'array_position' field and the bit-field
variable 'failed_devices'.

Signed-off-by: Jonathan Brassow <jbrassow@redhat.com>
Signed-off-by: Alasdair G Kergon <agk@redhat.com>

3 files changed, 412 insertions, 24 deletions

diff --git a/Documentation/device-mapper/dm-raid.txt b/Documentation/device-mapper/dm-raid.txt
index a7d1c4abc927..2a8c11331d2d 100644
--- a/Documentation/device-mapper/dm-raid.txt
+++ b/Documentation/device-mapper/dm-raid.txt
@@ -11,6 +11,7 @@ The target is named "raid" and it accepts the following parameters:
     <#raid_devs> <metadata_dev0> <dev0> [.. <metadata_devN> <devN>]
 
 <raid_type>:
+  raid1		RAID1 mirroring
   raid4		RAID4 dedicated parity disk
   raid5_la	RAID5 left asymmetric
 		- rotating parity 0 with data continuation
@@ -61,8 +62,7 @@ The target is named "raid" and it accepts the following parameters:
 <#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
-	data.  Currently, separate metadata devices are not supported and '-'
-	is required in place of the metadata device.
+	data.
 
 	If a drive has failed or is missing at creation time, a '-' can be
 	given for both the metadata and data drives for a given position.
@@ -70,7 +70,7 @@ The target is named "raid" and it accepts the following parameters:
 
 Example tables
 --------------
-# RAID4 - 4 data drives, 1 parity
+# RAID4 - 4 data drives, 1 parity (no metadata devices)
 # No metadata devices specified to hold superblock/bitmap info
 # Chunk size of 1MiB
 # (Lines separated for easy reading)
@@ -79,13 +79,13 @@ Example tables
         raid4 1 2048 \
         5 - 8:17 - 8:33 - 8:49 - 8:65 - 8:81
 
-# RAID4 - 4 data drives, 1 parity (no metadata devices)
+# RAID4 - 4 data drives, 1 parity (with metadata devices)
 # Chunk size of 1MiB, force RAID initialization,
 #       min recovery rate at 20 kiB/sec/disk
 
 0 1960893648 raid \
-        raid4 4 2048 min_recovery_rate 20 sync\
-        5 - 8:17 - 8:33 - 8:49 - 8:65 - 8:81
+        raid4 4 2048 sync min_recovery_rate 20 \
+        5 8:17 8:18 8:33 8:34 8:49 8:50 8:65 8:66 8:81 8:82
 
 'dmsetup table' displays the table used to construct the mapping.
 The optional parameters are always printed in the order listed
diff --git a/drivers/md/Kconfig b/drivers/md/Kconfig
index 8420129fc5ee..f75a66e7d312 100644
--- a/drivers/md/Kconfig
+++ b/drivers/md/Kconfig
@@ -241,12 +241,13 @@ config DM_MIRROR
          needed for live data migration tools such as 'pvmove'.
 
 config DM_RAID
-       tristate "RAID 4/5/6 target (EXPERIMENTAL)"
+       tristate "RAID 1/4/5/6 target (EXPERIMENTAL)"
        depends on BLK_DEV_DM && EXPERIMENTAL
+       select MD_RAID1
        select MD_RAID456
        select BLK_DEV_MD
        ---help---
-	 A dm target that supports RAID4, RAID5 and RAID6 mappings
+	 A dm target that supports RAID1, RAID4, RAID5 and RAID6 mappings
 
 	 A RAID-5 set of N drives with a capacity of C MB per drive provides
 	 the capacity of C * (N - 1) MB, and protects against a failure
diff --git a/drivers/md/dm-raid.c b/drivers/md/dm-raid.c
index 88143a0303d2..69873806c50c 100644
--- a/drivers/md/dm-raid.c
+++ b/drivers/md/dm-raid.c
@@ -16,12 +16,10 @@
 #define DM_MSG_PREFIX "raid"
 
 /*
- * If the MD doesn't support MD_SYNC_STATE_FORCED yet, then
- * make it so the flag doesn't set anything.
+ * The following flags are used by dm-raid.c to set up the array state.
+ * They must be cleared before md_run is called.
  */
-#ifndef MD_SYNC_STATE_FORCED
-#define MD_SYNC_STATE_FORCED 0
-#endif
+#define FirstUse 10             /* rdev flag */
 
 struct raid_dev {
 	/*
@@ -149,9 +147,16 @@ static void context_free(struct raid_set *rs)
 {
 	int i;
 
-	for (i = 0; i < rs->md.raid_disks; i++)
+	for (i = 0; i < rs->md.raid_disks; i++) {
+		if (rs->dev[i].meta_dev)
+			dm_put_device(rs->ti, rs->dev[i].meta_dev);
+		if (rs->dev[i].rdev.sb_page)
+			put_page(rs->dev[i].rdev.sb_page);
+		rs->dev[i].rdev.sb_page = NULL;
+		rs->dev[i].rdev.sb_loaded = 0;
 		if (rs->dev[i].data_dev)
 			dm_put_device(rs->ti, rs->dev[i].data_dev);
+	}
 
 	kfree(rs);
 }
@@ -161,7 +166,16 @@ static void context_free(struct raid_set *rs)
  *  <meta_dev>: meta device name or '-' if missing
  *  <data_dev>: data device name or '-' if missing
  *
- * This code parses those words.
+ * The following are permitted:
+ *    - -
+ *    - <data_dev>
+ *    <meta_dev> <data_dev>
+ *
+ * The following is not allowed:
+ *    <meta_dev> -
+ *
+ * This code parses those words.  If there is a failure,
+ * the caller must use context_free to unwind the operations.
  */
 static int dev_parms(struct raid_set *rs, char **argv)
 {
@@ -184,8 +198,16 @@ static int dev_parms(struct raid_set *rs, char **argv)
 		rs->dev[i].rdev.mddev = &rs->md;
 
 		if (strcmp(argv[0], "-")) {
-			rs->ti->error = "Metadata devices not supported";
-			return -EINVAL;
+			ret = dm_get_device(rs->ti, argv[0],
+					    dm_table_get_mode(rs->ti->table),
+					    &rs->dev[i].meta_dev);
+			rs->ti->error = "RAID metadata device lookup failure";
+			if (ret)
+				return ret;
+
+			rs->dev[i].rdev.sb_page = alloc_page(GFP_KERNEL);
+			if (!rs->dev[i].rdev.sb_page)
+				return -ENOMEM;
 		}
 
 		if (!strcmp(argv[1], "-")) {
@@ -195,6 +217,10 @@ static int dev_parms(struct raid_set *rs, char **argv)
 				return -EINVAL;
 			}
 
+			rs->ti->error = "No data device supplied with metadata device";
+			if (rs->dev[i].meta_dev)
+				return -EINVAL;
+
 			continue;
 		}
 
@@ -206,6 +232,10 @@ static int dev_parms(struct raid_set *rs, char **argv)
 			return ret;
 		}
 
+		if (rs->dev[i].meta_dev) {
+			metadata_available = 1;
+			rs->dev[i].rdev.meta_bdev = rs->dev[i].meta_dev->bdev;
+		}
 		rs->dev[i].rdev.bdev = rs->dev[i].data_dev->bdev;
 		list_add(&rs->dev[i].rdev.same_set, &rs->md.disks);
 		if (!test_bit(In_sync, &rs->dev[i].rdev.flags))
@@ -334,22 +364,39 @@ static int parse_raid_params(struct raid_set *rs, char **argv,
 	num_raid_params--;
 
 	/*
-	 * Second, parse the unordered optional arguments
+	 * We set each individual device as In_sync with a completed
+	 * 'recovery_offset'.  If there has been a device failure or
+	 * replacement then one of the following cases applies:
+	 *
+	 *   1) User specifies 'rebuild'.
+	 *      - Device is reset when param is read.
+	 *   2) A new device is supplied.
+	 *      - No matching superblock found, resets device.
+	 *   3) Device failure was transient and returns on reload.
+	 *      - Failure noticed, resets device for bitmap replay.
+	 *   4) Device hadn't completed recovery after previous failure.
+	 *      - Superblock is read and overrides recovery_offset.
+	 *
+	 * What is found in the superblocks of the devices is always
+	 * authoritative, unless 'rebuild' or '[no]sync' was specified.
 	 */
-	for (i = 0; i < rs->md.raid_disks; i++)
+	for (i = 0; i < rs->md.raid_disks; i++) {
 		set_bit(In_sync, &rs->dev[i].rdev.flags);
+		rs->dev[i].rdev.recovery_offset = MaxSector;
+	}
 
+	/*
+	 * Second, parse the unordered optional arguments
+	 */
 	for (i = 0; i < num_raid_params; i++) {
 		if (!strcasecmp(argv[i], "nosync")) {
 			rs->md.recovery_cp = MaxSector;
 			rs->print_flags |= DMPF_NOSYNC;
-			rs->md.flags |= MD_SYNC_STATE_FORCED;
 			continue;
 		}
 		if (!strcasecmp(argv[i], "sync")) {
 			rs->md.recovery_cp = 0;
 			rs->print_flags |= DMPF_SYNC;
-			rs->md.flags |= MD_SYNC_STATE_FORCED;
 			continue;
 		}
 
@@ -482,13 +529,344 @@ static int raid_is_congested(struct dm_target_callbacks *cb, int bits)
 }
 
 /*
+ * This structure is never routinely used by userspace, unlike md superblocks.
+ * Devices with this superblock should only ever be accessed via device-mapper.
+ */
+#define DM_RAID_MAGIC 0x64526D44
+struct dm_raid_superblock {
+	__le32 magic;		/* "DmRd" */
+	__le32 features;	/* Used to indicate possible future changes */
+
+	__le32 num_devices;	/* Number of devices in this array. (Max 64) */
+	__le32 array_position;	/* The position of this drive in the array */
+
+	__le64 events;		/* Incremented by md when superblock updated */
+	__le64 failed_devices;	/* Bit field of devices to indicate failures */
+
+	/*
+	 * This offset tracks the progress of the repair or replacement of
+	 * an individual drive.
+	 */
+	__le64 disk_recovery_offset;
+
+	/*
+	 * This offset tracks the progress of the initial array
+	 * synchronisation/parity calculation.
+	 */
+	__le64 array_resync_offset;
+
+	/*
+	 * RAID characteristics
+	 */
+	__le32 level;
+	__le32 layout;
+	__le32 stripe_sectors;
+
+	__u8 pad[452];		/* Round struct to 512 bytes. */
+				/* Always set to 0 when writing. */
+} __packed;
+
+static int read_disk_sb(mdk_rdev_t *rdev, int size)
+{
+	BUG_ON(!rdev->sb_page);
+
+	if (rdev->sb_loaded)
+		return 0;
+
+	if (!sync_page_io(rdev, 0, size, rdev->sb_page, READ, 1)) {
+		DMERR("Failed to read device superblock");
+		return -EINVAL;
+	}
+
+	rdev->sb_loaded = 1;
+
+	return 0;
+}
+
+static void super_sync(mddev_t *mddev, mdk_rdev_t *rdev)
+{
+	mdk_rdev_t *r, *t;
+	uint64_t failed_devices;
+	struct dm_raid_superblock *sb;
+
+	sb = page_address(rdev->sb_page);
+	failed_devices = le64_to_cpu(sb->failed_devices);
+
+	rdev_for_each(r, t, mddev)
+		if ((r->raid_disk >= 0) && test_bit(Faulty, &r->flags))
+			failed_devices |= (1ULL << r->raid_disk);
+
+	memset(sb, 0, sizeof(*sb));
+
+	sb->magic = cpu_to_le32(DM_RAID_MAGIC);
+	sb->features = cpu_to_le32(0);	/* No features yet */
+
+	sb->num_devices = cpu_to_le32(mddev->raid_disks);
+	sb->array_position = cpu_to_le32(rdev->raid_disk);
+
+	sb->events = cpu_to_le64(mddev->events);
+	sb->failed_devices = cpu_to_le64(failed_devices);
+
+	sb->disk_recovery_offset = cpu_to_le64(rdev->recovery_offset);
+	sb->array_resync_offset = cpu_to_le64(mddev->recovery_cp);
+
+	sb->level = cpu_to_le32(mddev->level);
+	sb->layout = cpu_to_le32(mddev->layout);
+	sb->stripe_sectors = cpu_to_le32(mddev->chunk_sectors);
+}
+
+/*
+ * super_load
+ *
+ * This function creates a superblock if one is not found on the device
+ * and will decide which superblock to use if there's a choice.
+ *
+ * Return: 1 if use rdev, 0 if use refdev, -Exxx otherwise
+ */
+static int super_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev)
+{
+	int ret;
+	struct dm_raid_superblock *sb;
+	struct dm_raid_superblock *refsb;
+	uint64_t events_sb, events_refsb;
+
+	rdev->sb_start = 0;
+	rdev->sb_size = sizeof(*sb);
+
+	ret = read_disk_sb(rdev, rdev->sb_size);
+	if (ret)
+		return ret;
+
+	sb = page_address(rdev->sb_page);
+	if (sb->magic != cpu_to_le32(DM_RAID_MAGIC)) {
+		super_sync(rdev->mddev, rdev);
+
+		set_bit(FirstUse, &rdev->flags);
+
+		/* Force writing of superblocks to disk */
+		set_bit(MD_CHANGE_DEVS, &rdev->mddev->flags);
+
+		/* Any superblock is better than none, choose that if given */
+		return refdev ? 0 : 1;
+	}
+
+	if (!refdev)
+		return 1;
+
+	events_sb = le64_to_cpu(sb->events);
+
+	refsb = page_address(refdev->sb_page);
+	events_refsb = le64_to_cpu(refsb->events);
+
+	return (events_sb > events_refsb) ? 1 : 0;
+}
+
+static int super_init_validation(mddev_t *mddev, mdk_rdev_t *rdev)
+{
+	int role;
+	struct raid_set *rs = container_of(mddev, struct raid_set, md);
+	uint64_t events_sb;
+	uint64_t failed_devices;
+	struct dm_raid_superblock *sb;
+	uint32_t new_devs = 0;
+	uint32_t rebuilds = 0;
+	mdk_rdev_t *r, *t;
+	struct dm_raid_superblock *sb2;
+
+	sb = page_address(rdev->sb_page);
+	events_sb = le64_to_cpu(sb->events);
+	failed_devices = le64_to_cpu(sb->failed_devices);
+
+	/*
+	 * Initialise to 1 if this is a new superblock.
+	 */
+	mddev->events = events_sb ? : 1;
+
+	/*
+	 * Reshaping is not currently allowed
+	 */
+	if ((le32_to_cpu(sb->level) != mddev->level) ||
+	    (le32_to_cpu(sb->layout) != mddev->layout) ||
+	    (le32_to_cpu(sb->stripe_sectors) != mddev->chunk_sectors)) {
+		DMERR("Reshaping arrays not yet supported.");
+		return -EINVAL;
+	}
+
+	/* We can only change the number of devices in RAID1 right now */
+	if ((rs->raid_type->level != 1) &&
+	    (le32_to_cpu(sb->num_devices) != mddev->raid_disks)) {
+		DMERR("Reshaping arrays not yet supported.");
+		return -EINVAL;
+	}
+
+	if (!(rs->print_flags & (DMPF_SYNC | DMPF_NOSYNC)))
+		mddev->recovery_cp = le64_to_cpu(sb->array_resync_offset);
+
+	/*
+	 * During load, we set FirstUse if a new superblock was written.
+	 * There are two reasons we might not have a superblock:
+	 * 1) The array is brand new - in which case, all of the
+	 *    devices must have their In_sync bit set.  Also,
+	 *    recovery_cp must be 0, unless forced.
+	 * 2) This is a new device being added to an old array
+	 *    and the new device needs to be rebuilt - in which
+	 *    case the In_sync bit will /not/ be set and
+	 *    recovery_cp must be MaxSector.
+	 */
+	rdev_for_each(r, t, mddev) {
+		if (!test_bit(In_sync, &r->flags)) {
+			if (!test_bit(FirstUse, &r->flags))
+				DMERR("Superblock area of "
+				      "rebuild device %d should have been "
+				      "cleared.", r->raid_disk);
+			set_bit(FirstUse, &r->flags);
+			rebuilds++;
+		} else if (test_bit(FirstUse, &r->flags))
+			new_devs++;
+	}
+
+	if (!rebuilds) {
+		if (new_devs == mddev->raid_disks) {
+			DMINFO("Superblocks created for new array");
+			set_bit(MD_ARRAY_FIRST_USE, &mddev->flags);
+		} else if (new_devs) {
+			DMERR("New device injected "
+			      "into existing array without 'rebuild' "
+			      "parameter specified");
+			return -EINVAL;
+		}
+	} else if (new_devs) {
+		DMERR("'rebuild' devices cannot be "
+		      "injected into an array with other first-time devices");
+		return -EINVAL;
+	} else if (mddev->recovery_cp != MaxSector) {
+		DMERR("'rebuild' specified while array is not in-sync");
+		return -EINVAL;
+	}
+
+	/*
+	 * Now we set the Faulty bit for those devices that are
+	 * recorded in the superblock as failed.
+	 */
+	rdev_for_each(r, t, mddev) {
+		if (!r->sb_page)
+			continue;
+		sb2 = page_address(r->sb_page);
+		sb2->failed_devices = 0;
+
+		/*
+		 * Check for any device re-ordering.
+		 */
+		if (!test_bit(FirstUse, &r->flags) && (r->raid_disk >= 0)) {
+			role = le32_to_cpu(sb2->array_position);
+			if (role != r->raid_disk) {
+				if (rs->raid_type->level != 1) {
+					rs->ti->error = "Cannot change device "
+						"positions in RAID array";
+					return -EINVAL;
+				}
+				DMINFO("RAID1 device #%d now at position #%d",
+				       role, r->raid_disk);
+			}
+
+			/*
+			 * Partial recovery is performed on
+			 * returning failed devices.
+			 */
+			if (failed_devices & (1 << role))
+				set_bit(Faulty, &r->flags);
+		}
+	}
+
+	return 0;
+}
+
+static int super_validate(mddev_t *mddev, mdk_rdev_t *rdev)
+{
+	struct dm_raid_superblock *sb = page_address(rdev->sb_page);
+
+	/*
+	 * If mddev->events is not set, we know we have not yet initialized
+	 * the array.
+	 */
+	if (!mddev->events && super_init_validation(mddev, rdev))
+		return -EINVAL;
+
+	mddev->bitmap_info.offset = 4096 >> 9; /* Enable bitmap creation */
+	rdev->mddev->bitmap_info.default_offset = 4096 >> 9;
+	if (!test_bit(FirstUse, &rdev->flags)) {
+		rdev->recovery_offset = le64_to_cpu(sb->disk_recovery_offset);
+		if (rdev->recovery_offset != MaxSector)
+			clear_bit(In_sync, &rdev->flags);
+	}
+
+	/*
+	 * If a device comes back, set it as not In_sync and no longer faulty.
+	 */
+	if (test_bit(Faulty, &rdev->flags)) {
+		clear_bit(Faulty, &rdev->flags);
+		clear_bit(In_sync, &rdev->flags);
+		rdev->saved_raid_disk = rdev->raid_disk;
+		rdev->recovery_offset = 0;
+	}
+
+	clear_bit(FirstUse, &rdev->flags);
+
+	return 0;
+}
+
+/*
+ * Analyse superblocks and select the freshest.
+ */
+static int analyse_superblocks(struct dm_target *ti, struct raid_set *rs)
+{
+	int ret;
+	mdk_rdev_t *rdev, *freshest, *tmp;
+	mddev_t *mddev = &rs->md;
+
+	freshest = NULL;
+	rdev_for_each(rdev, tmp, mddev) {
+		if (!rdev->meta_bdev)
+			continue;
+
+		ret = super_load(rdev, freshest);
+
+		switch (ret) {
+		case 1:
+			freshest = rdev;
+			break;
+		case 0:
+			break;
+		default:
+			ti->error = "Failed to load superblock";
+			return ret;
+		}
+	}
+
+	if (!freshest)
+		return 0;
+
+	/*
+	 * Validation of the freshest device provides the source of
+	 * validation for the remaining devices.
+	 */
+	ti->error = "Unable to assemble array: Invalid superblocks";
+	if (super_validate(mddev, freshest))
+		return -EINVAL;
+
+	rdev_for_each(rdev, tmp, mddev)
+		if ((rdev != freshest) && super_validate(mddev, rdev))
+			return -EINVAL;
+
+	return 0;
+}
+
+/*
  * Construct a RAID4/5/6 mapping:
  * Args:
  *	<raid_type> <#raid_params> <raid_params>		\
  *	<#raid_devs> { <meta_dev1> <dev1> .. <meta_devN> <devN> }
  *
- * ** metadata devices are not supported yet, use '-' instead **
- *
  * <raid_params> varies by <raid_type>.  See 'parse_raid_params' for
  * details on possible <raid_params>.
  */
@@ -556,6 +934,11 @@ static int raid_ctr(struct dm_target *ti, unsigned argc, char **argv)
 	if (ret)
 		goto bad;
 
+	rs->md.sync_super = super_sync;
+	ret = analyse_superblocks(ti, rs);
+	if (ret)
+		goto bad;
+
 	INIT_WORK(&rs->md.event_work, do_table_event);
 	ti->private = rs;
 
@@ -698,7 +1081,10 @@ static int raid_status(struct dm_target *ti, status_type_t type,
 
 		DMEMIT(" %d", rs->md.raid_disks);
 		for (i = 0; i < rs->md.raid_disks; i++) {
-			DMEMIT(" -"); /* metadata device */
+			if (rs->dev[i].meta_dev)
+				DMEMIT(" %s", rs->dev[i].meta_dev->name);
+			else
+				DMEMIT(" -");
 
 			if (rs->dev[i].data_dev)
 				DMEMIT(" %s", rs->dev[i].data_dev->name);
@@ -755,6 +1141,7 @@ static void raid_resume(struct dm_target *ti)
 {
 	struct raid_set *rs = ti->private;
 
+	bitmap_load(&rs->md);
 	mddev_resume(&rs->md);
 }