2 * Copyright (C) 2010-2011 Neil Brown
3 * Copyright (C) 2010-2011 Red Hat, Inc. All rights reserved.
5 * This file is released under the GPL.
8 #include <linux/slab.h>
9 #include <linux/module.h>
17 #include <linux/device-mapper.h>
19 #define DM_MSG_PREFIX "raid"
22 * The following flags are used by dm-raid.c to set up the array state.
23 * They must be cleared before md_run is called.
25 #define FirstUse 10 /* rdev flag */
29 * Two DM devices, one to hold metadata and one to hold the
30 * actual data/parity. The reason for this is to not confuse
31 * ti->len and give more flexibility in altering size and
34 * While it is possible for this device to be associated
35 * with a different physical device than the data_dev, it
36 * is intended for it to be the same.
37 * |--------- Physical Device ---------|
38 * |- meta_dev -|------ data_dev ------|
40 struct dm_dev *meta_dev;
41 struct dm_dev *data_dev;
46 * Flags for rs->print_flags field.
49 #define DMPF_NOSYNC 0x2
50 #define DMPF_REBUILD 0x4
51 #define DMPF_DAEMON_SLEEP 0x8
52 #define DMPF_MIN_RECOVERY_RATE 0x10
53 #define DMPF_MAX_RECOVERY_RATE 0x20
54 #define DMPF_MAX_WRITE_BEHIND 0x40
55 #define DMPF_STRIPE_CACHE 0x80
56 #define DMPF_REGION_SIZE 0x100
57 #define DMPF_RAID10_COPIES 0x200
58 #define DMPF_RAID10_FORMAT 0x400
63 uint32_t bitmap_loaded;
67 struct raid_type *raid_type;
68 struct dm_target_callbacks callbacks;
70 struct raid_dev dev[0];
73 /* Supported raid types and properties. */
74 static struct raid_type {
75 const char *name; /* RAID algorithm. */
76 const char *descr; /* Descriptor text for logging. */
77 const unsigned parity_devs; /* # of parity devices. */
78 const unsigned minimal_devs; /* minimal # of devices in set. */
79 const unsigned level; /* RAID level. */
80 const unsigned algorithm; /* RAID algorithm. */
82 {"raid1", "RAID1 (mirroring)", 0, 2, 1, 0 /* NONE */},
83 {"raid10", "RAID10 (striped mirrors)", 0, 2, 10, UINT_MAX /* Varies */},
84 {"raid4", "RAID4 (dedicated parity disk)", 1, 2, 5, ALGORITHM_PARITY_0},
85 {"raid5_la", "RAID5 (left asymmetric)", 1, 2, 5, ALGORITHM_LEFT_ASYMMETRIC},
86 {"raid5_ra", "RAID5 (right asymmetric)", 1, 2, 5, ALGORITHM_RIGHT_ASYMMETRIC},
87 {"raid5_ls", "RAID5 (left symmetric)", 1, 2, 5, ALGORITHM_LEFT_SYMMETRIC},
88 {"raid5_rs", "RAID5 (right symmetric)", 1, 2, 5, ALGORITHM_RIGHT_SYMMETRIC},
89 {"raid6_zr", "RAID6 (zero restart)", 2, 4, 6, ALGORITHM_ROTATING_ZERO_RESTART},
90 {"raid6_nr", "RAID6 (N restart)", 2, 4, 6, ALGORITHM_ROTATING_N_RESTART},
91 {"raid6_nc", "RAID6 (N continue)", 2, 4, 6, ALGORITHM_ROTATING_N_CONTINUE}
94 static unsigned raid10_md_layout_to_copies(int layout)
99 static int raid10_format_to_md_layout(char *format, unsigned copies)
101 /* 1 "far" copy, and 'copies' "near" copies */
102 return (1 << 8) | (copies & 0xFF);
105 static struct raid_type *get_raid_type(char *name)
109 for (i = 0; i < ARRAY_SIZE(raid_types); i++)
110 if (!strcmp(raid_types[i].name, name))
111 return &raid_types[i];
116 static struct raid_set *context_alloc(struct dm_target *ti, struct raid_type *raid_type, unsigned raid_devs)
121 if (raid_devs <= raid_type->parity_devs) {
122 ti->error = "Insufficient number of devices";
123 return ERR_PTR(-EINVAL);
126 rs = kzalloc(sizeof(*rs) + raid_devs * sizeof(rs->dev[0]), GFP_KERNEL);
128 ti->error = "Cannot allocate raid context";
129 return ERR_PTR(-ENOMEM);
135 rs->raid_type = raid_type;
136 rs->md.raid_disks = raid_devs;
137 rs->md.level = raid_type->level;
138 rs->md.new_level = rs->md.level;
139 rs->md.layout = raid_type->algorithm;
140 rs->md.new_layout = rs->md.layout;
141 rs->md.delta_disks = 0;
142 rs->md.recovery_cp = 0;
144 for (i = 0; i < raid_devs; i++)
145 md_rdev_init(&rs->dev[i].rdev);
148 * Remaining items to be initialized by further RAID params:
151 * rs->md.chunk_sectors
152 * rs->md.new_chunk_sectors
159 static void context_free(struct raid_set *rs)
163 for (i = 0; i < rs->md.raid_disks; i++) {
164 if (rs->dev[i].meta_dev)
165 dm_put_device(rs->ti, rs->dev[i].meta_dev);
166 md_rdev_clear(&rs->dev[i].rdev);
167 if (rs->dev[i].data_dev)
168 dm_put_device(rs->ti, rs->dev[i].data_dev);
175 * For every device we have two words
176 * <meta_dev>: meta device name or '-' if missing
177 * <data_dev>: data device name or '-' if missing
179 * The following are permitted:
182 * <meta_dev> <data_dev>
184 * The following is not allowed:
187 * This code parses those words. If there is a failure,
188 * the caller must use context_free to unwind the operations.
190 static int dev_parms(struct raid_set *rs, char **argv)
194 int metadata_available = 0;
197 for (i = 0; i < rs->md.raid_disks; i++, argv += 2) {
198 rs->dev[i].rdev.raid_disk = i;
200 rs->dev[i].meta_dev = NULL;
201 rs->dev[i].data_dev = NULL;
204 * There are no offsets, since there is a separate device
205 * for data and metadata.
207 rs->dev[i].rdev.data_offset = 0;
208 rs->dev[i].rdev.mddev = &rs->md;
210 if (strcmp(argv[0], "-")) {
211 ret = dm_get_device(rs->ti, argv[0],
212 dm_table_get_mode(rs->ti->table),
213 &rs->dev[i].meta_dev);
214 rs->ti->error = "RAID metadata device lookup failure";
218 rs->dev[i].rdev.sb_page = alloc_page(GFP_KERNEL);
219 if (!rs->dev[i].rdev.sb_page)
223 if (!strcmp(argv[1], "-")) {
224 if (!test_bit(In_sync, &rs->dev[i].rdev.flags) &&
225 (!rs->dev[i].rdev.recovery_offset)) {
226 rs->ti->error = "Drive designated for rebuild not specified";
230 rs->ti->error = "No data device supplied with metadata device";
231 if (rs->dev[i].meta_dev)
237 ret = dm_get_device(rs->ti, argv[1],
238 dm_table_get_mode(rs->ti->table),
239 &rs->dev[i].data_dev);
241 rs->ti->error = "RAID device lookup failure";
245 if (rs->dev[i].meta_dev) {
246 metadata_available = 1;
247 rs->dev[i].rdev.meta_bdev = rs->dev[i].meta_dev->bdev;
249 rs->dev[i].rdev.bdev = rs->dev[i].data_dev->bdev;
250 list_add(&rs->dev[i].rdev.same_set, &rs->md.disks);
251 if (!test_bit(In_sync, &rs->dev[i].rdev.flags))
255 if (metadata_available) {
257 rs->md.persistent = 1;
258 rs->md.major_version = 2;
259 } else if (rebuild && !rs->md.recovery_cp) {
261 * Without metadata, we will not be able to tell if the array
262 * is in-sync or not - we must assume it is not. Therefore,
263 * it is impossible to rebuild a drive.
265 * Even if there is metadata, the on-disk information may
266 * indicate that the array is not in-sync and it will then
269 * User could specify 'nosync' option if desperate.
271 DMERR("Unable to rebuild drive while array is not in-sync");
272 rs->ti->error = "RAID device lookup failure";
280 * validate_region_size
282 * @region_size: region size in sectors. If 0, pick a size (4MiB default).
284 * Set rs->md.bitmap_info.chunksize (which really refers to 'region size').
285 * Ensure that (ti->len/region_size < 2^21) - required by MD bitmap.
287 * Returns: 0 on success, -EINVAL on failure.
289 static int validate_region_size(struct raid_set *rs, unsigned long region_size)
291 unsigned long min_region_size = rs->ti->len / (1 << 21);
295 * Choose a reasonable default. All figures in sectors.
297 if (min_region_size > (1 << 13)) {
298 /* If not a power of 2, make it the next power of 2 */
299 if (min_region_size & (min_region_size - 1))
300 region_size = 1 << fls(region_size);
301 DMINFO("Choosing default region size of %lu sectors",
304 DMINFO("Choosing default region size of 4MiB");
305 region_size = 1 << 13; /* sectors */
309 * Validate user-supplied value.
311 if (region_size > rs->ti->len) {
312 rs->ti->error = "Supplied region size is too large";
316 if (region_size < min_region_size) {
317 DMERR("Supplied region_size (%lu sectors) below minimum (%lu)",
318 region_size, min_region_size);
319 rs->ti->error = "Supplied region size is too small";
323 if (!is_power_of_2(region_size)) {
324 rs->ti->error = "Region size is not a power of 2";
328 if (region_size < rs->md.chunk_sectors) {
329 rs->ti->error = "Region size is smaller than the chunk size";
335 * Convert sectors to bytes.
337 rs->md.bitmap_info.chunksize = (region_size << 9);
343 * validate_raid_redundancy
346 * Determine if there are enough devices in the array that haven't
347 * failed (or are being rebuilt) to form a usable array.
349 * Returns: 0 on success, -EINVAL on failure.
351 static int validate_raid_redundancy(struct raid_set *rs)
353 unsigned i, rebuild_cnt = 0;
354 unsigned rebuilds_per_group, copies, d;
356 for (i = 0; i < rs->md.raid_disks; i++)
357 if (!test_bit(In_sync, &rs->dev[i].rdev.flags) ||
358 !rs->dev[i].rdev.sb_page)
361 switch (rs->raid_type->level) {
363 if (rebuild_cnt >= rs->md.raid_disks)
369 if (rebuild_cnt > rs->raid_type->parity_devs)
373 copies = raid10_md_layout_to_copies(rs->md.layout);
374 if (rebuild_cnt < copies)
378 * It is possible to have a higher rebuild count for RAID10,
379 * as long as the failed devices occur in different mirror
380 * groups (i.e. different stripes).
382 * Right now, we only allow for "near" copies. When other
383 * formats are added, we will have to check those too.
385 * When checking "near" format, make sure no adjacent devices
386 * have failed beyond what can be handled. In addition to the
387 * simple case where the number of devices is a multiple of the
388 * number of copies, we must also handle cases where the number
389 * of devices is not a multiple of the number of copies.
390 * E.g. dev1 dev2 dev3 dev4 dev5
394 for (i = 0; i < rs->md.raid_disks * copies; i++) {
396 rebuilds_per_group = 0;
397 d = i % rs->md.raid_disks;
398 if ((!rs->dev[d].rdev.sb_page ||
399 !test_bit(In_sync, &rs->dev[d].rdev.flags)) &&
400 (++rebuilds_per_group >= copies))
416 * Possible arguments are...
417 * <chunk_size> [optional_args]
419 * Argument definitions
420 * <chunk_size> The number of sectors per disk that
421 * will form the "stripe"
422 * [[no]sync] Force or prevent recovery of the
424 * [rebuild <idx>] Rebuild the drive indicated by the index
425 * [daemon_sleep <ms>] Time between bitmap daemon work to
427 * [min_recovery_rate <kB/sec/disk>] Throttle RAID initialization
428 * [max_recovery_rate <kB/sec/disk>] Throttle RAID initialization
429 * [write_mostly <idx>] Indicate a write mostly drive via index
430 * [max_write_behind <sectors>] See '-write-behind=' (man mdadm)
431 * [stripe_cache <sectors>] Stripe cache size for higher RAIDs
432 * [region_size <sectors>] Defines granularity of bitmap
434 * RAID10-only options:
435 * [raid10_copies <# copies>] Number of copies. (Default: 2)
436 * [raid10_format <near>] Layout algorithm. (Default: near)
438 static int parse_raid_params(struct raid_set *rs, char **argv,
439 unsigned num_raid_params)
441 char *raid10_format = "near";
442 unsigned raid10_copies = 2;
444 unsigned long value, region_size = 0;
445 sector_t sectors_per_dev = rs->ti->len;
450 * First, parse the in-order required arguments
451 * "chunk_size" is the only argument of this type.
453 if ((strict_strtoul(argv[0], 10, &value) < 0)) {
454 rs->ti->error = "Bad chunk size";
456 } else if (rs->raid_type->level == 1) {
458 DMERR("Ignoring chunk size parameter for RAID 1");
460 } else if (!is_power_of_2(value)) {
461 rs->ti->error = "Chunk size must be a power of 2";
463 } else if (value < 8) {
464 rs->ti->error = "Chunk size value is too small";
468 rs->md.new_chunk_sectors = rs->md.chunk_sectors = value;
473 * We set each individual device as In_sync with a completed
474 * 'recovery_offset'. If there has been a device failure or
475 * replacement then one of the following cases applies:
477 * 1) User specifies 'rebuild'.
478 * - Device is reset when param is read.
479 * 2) A new device is supplied.
480 * - No matching superblock found, resets device.
481 * 3) Device failure was transient and returns on reload.
482 * - Failure noticed, resets device for bitmap replay.
483 * 4) Device hadn't completed recovery after previous failure.
484 * - Superblock is read and overrides recovery_offset.
486 * What is found in the superblocks of the devices is always
487 * authoritative, unless 'rebuild' or '[no]sync' was specified.
489 for (i = 0; i < rs->md.raid_disks; i++) {
490 set_bit(In_sync, &rs->dev[i].rdev.flags);
491 rs->dev[i].rdev.recovery_offset = MaxSector;
495 * Second, parse the unordered optional arguments
497 for (i = 0; i < num_raid_params; i++) {
498 if (!strcasecmp(argv[i], "nosync")) {
499 rs->md.recovery_cp = MaxSector;
500 rs->print_flags |= DMPF_NOSYNC;
503 if (!strcasecmp(argv[i], "sync")) {
504 rs->md.recovery_cp = 0;
505 rs->print_flags |= DMPF_SYNC;
509 /* The rest of the optional arguments come in key/value pairs */
510 if ((i + 1) >= num_raid_params) {
511 rs->ti->error = "Wrong number of raid parameters given";
517 /* Parameters that take a string value are checked here. */
518 if (!strcasecmp(key, "raid10_format")) {
519 if (rs->raid_type->level != 10) {
520 rs->ti->error = "'raid10_format' is an invalid parameter for this RAID type";
523 if (strcmp("near", argv[i])) {
524 rs->ti->error = "Invalid 'raid10_format' value given";
527 raid10_format = argv[i];
528 rs->print_flags |= DMPF_RAID10_FORMAT;
532 if (strict_strtoul(argv[i], 10, &value) < 0) {
533 rs->ti->error = "Bad numerical argument given in raid params";
537 /* Parameters that take a numeric value are checked here */
538 if (!strcasecmp(key, "rebuild")) {
539 if (value >= rs->md.raid_disks) {
540 rs->ti->error = "Invalid rebuild index given";
543 clear_bit(In_sync, &rs->dev[value].rdev.flags);
544 rs->dev[value].rdev.recovery_offset = 0;
545 rs->print_flags |= DMPF_REBUILD;
546 } else if (!strcasecmp(key, "write_mostly")) {
547 if (rs->raid_type->level != 1) {
548 rs->ti->error = "write_mostly option is only valid for RAID1";
551 if (value >= rs->md.raid_disks) {
552 rs->ti->error = "Invalid write_mostly drive index given";
555 set_bit(WriteMostly, &rs->dev[value].rdev.flags);
556 } else if (!strcasecmp(key, "max_write_behind")) {
557 if (rs->raid_type->level != 1) {
558 rs->ti->error = "max_write_behind option is only valid for RAID1";
561 rs->print_flags |= DMPF_MAX_WRITE_BEHIND;
564 * In device-mapper, we specify things in sectors, but
565 * MD records this value in kB
568 if (value > COUNTER_MAX) {
569 rs->ti->error = "Max write-behind limit out of range";
572 rs->md.bitmap_info.max_write_behind = value;
573 } else if (!strcasecmp(key, "daemon_sleep")) {
574 rs->print_flags |= DMPF_DAEMON_SLEEP;
575 if (!value || (value > MAX_SCHEDULE_TIMEOUT)) {
576 rs->ti->error = "daemon sleep period out of range";
579 rs->md.bitmap_info.daemon_sleep = value;
580 } else if (!strcasecmp(key, "stripe_cache")) {
581 rs->print_flags |= DMPF_STRIPE_CACHE;
584 * In device-mapper, we specify things in sectors, but
585 * MD records this value in kB
589 if ((rs->raid_type->level != 5) &&
590 (rs->raid_type->level != 6)) {
591 rs->ti->error = "Inappropriate argument: stripe_cache";
594 if (raid5_set_cache_size(&rs->md, (int)value)) {
595 rs->ti->error = "Bad stripe_cache size";
598 } else if (!strcasecmp(key, "min_recovery_rate")) {
599 rs->print_flags |= DMPF_MIN_RECOVERY_RATE;
600 if (value > INT_MAX) {
601 rs->ti->error = "min_recovery_rate out of range";
604 rs->md.sync_speed_min = (int)value;
605 } else if (!strcasecmp(key, "max_recovery_rate")) {
606 rs->print_flags |= DMPF_MAX_RECOVERY_RATE;
607 if (value > INT_MAX) {
608 rs->ti->error = "max_recovery_rate out of range";
611 rs->md.sync_speed_max = (int)value;
612 } else if (!strcasecmp(key, "region_size")) {
613 rs->print_flags |= DMPF_REGION_SIZE;
615 } else if (!strcasecmp(key, "raid10_copies") &&
616 (rs->raid_type->level == 10)) {
617 if ((value < 2) || (value > 0xFF)) {
618 rs->ti->error = "Bad value for 'raid10_copies'";
621 rs->print_flags |= DMPF_RAID10_COPIES;
622 raid10_copies = value;
624 DMERR("Unable to parse RAID parameter: %s", key);
625 rs->ti->error = "Unable to parse RAID parameters";
630 if (validate_region_size(rs, region_size))
633 if (rs->md.chunk_sectors)
634 max_io_len = rs->md.chunk_sectors;
636 max_io_len = region_size;
638 if (dm_set_target_max_io_len(rs->ti, max_io_len))
641 if (rs->raid_type->level == 10) {
642 if (raid10_copies > rs->md.raid_disks) {
643 rs->ti->error = "Not enough devices to satisfy specification";
647 /* (Len * #mirrors) / #devices */
648 sectors_per_dev = rs->ti->len * raid10_copies;
649 sector_div(sectors_per_dev, rs->md.raid_disks);
651 rs->md.layout = raid10_format_to_md_layout(raid10_format,
653 rs->md.new_layout = rs->md.layout;
654 } else if ((rs->raid_type->level > 1) &&
655 sector_div(sectors_per_dev,
656 (rs->md.raid_disks - rs->raid_type->parity_devs))) {
657 rs->ti->error = "Target length not divisible by number of data devices";
660 rs->md.dev_sectors = sectors_per_dev;
662 /* Assume there are no metadata devices until the drives are parsed */
663 rs->md.persistent = 0;
669 static void do_table_event(struct work_struct *ws)
671 struct raid_set *rs = container_of(ws, struct raid_set, md.event_work);
673 dm_table_event(rs->ti->table);
676 static int raid_is_congested(struct dm_target_callbacks *cb, int bits)
678 struct raid_set *rs = container_of(cb, struct raid_set, callbacks);
680 if (rs->raid_type->level == 1)
681 return md_raid1_congested(&rs->md, bits);
683 if (rs->raid_type->level == 10)
684 return md_raid10_congested(&rs->md, bits);
686 return md_raid5_congested(&rs->md, bits);
690 * This structure is never routinely used by userspace, unlike md superblocks.
691 * Devices with this superblock should only ever be accessed via device-mapper.
693 #define DM_RAID_MAGIC 0x64526D44
694 struct dm_raid_superblock {
695 __le32 magic; /* "DmRd" */
696 __le32 features; /* Used to indicate possible future changes */
698 __le32 num_devices; /* Number of devices in this array. (Max 64) */
699 __le32 array_position; /* The position of this drive in the array */
701 __le64 events; /* Incremented by md when superblock updated */
702 __le64 failed_devices; /* Bit field of devices to indicate failures */
705 * This offset tracks the progress of the repair or replacement of
706 * an individual drive.
708 __le64 disk_recovery_offset;
711 * This offset tracks the progress of the initial array
712 * synchronisation/parity calculation.
714 __le64 array_resync_offset;
717 * RAID characteristics
721 __le32 stripe_sectors;
723 __u8 pad[452]; /* Round struct to 512 bytes. */
724 /* Always set to 0 when writing. */
727 static int read_disk_sb(struct md_rdev *rdev, int size)
729 BUG_ON(!rdev->sb_page);
734 if (!sync_page_io(rdev, 0, size, rdev->sb_page, READ, 1)) {
735 DMERR("Failed to read superblock of device at position %d",
737 md_error(rdev->mddev, rdev);
746 static void super_sync(struct mddev *mddev, struct md_rdev *rdev)
749 uint64_t failed_devices;
750 struct dm_raid_superblock *sb;
751 struct raid_set *rs = container_of(mddev, struct raid_set, md);
753 sb = page_address(rdev->sb_page);
754 failed_devices = le64_to_cpu(sb->failed_devices);
756 for (i = 0; i < mddev->raid_disks; i++)
757 if (!rs->dev[i].data_dev ||
758 test_bit(Faulty, &(rs->dev[i].rdev.flags)))
759 failed_devices |= (1ULL << i);
761 memset(sb, 0, sizeof(*sb));
763 sb->magic = cpu_to_le32(DM_RAID_MAGIC);
764 sb->features = cpu_to_le32(0); /* No features yet */
766 sb->num_devices = cpu_to_le32(mddev->raid_disks);
767 sb->array_position = cpu_to_le32(rdev->raid_disk);
769 sb->events = cpu_to_le64(mddev->events);
770 sb->failed_devices = cpu_to_le64(failed_devices);
772 sb->disk_recovery_offset = cpu_to_le64(rdev->recovery_offset);
773 sb->array_resync_offset = cpu_to_le64(mddev->recovery_cp);
775 sb->level = cpu_to_le32(mddev->level);
776 sb->layout = cpu_to_le32(mddev->layout);
777 sb->stripe_sectors = cpu_to_le32(mddev->chunk_sectors);
783 * This function creates a superblock if one is not found on the device
784 * and will decide which superblock to use if there's a choice.
786 * Return: 1 if use rdev, 0 if use refdev, -Exxx otherwise
788 static int super_load(struct md_rdev *rdev, struct md_rdev *refdev)
791 struct dm_raid_superblock *sb;
792 struct dm_raid_superblock *refsb;
793 uint64_t events_sb, events_refsb;
796 rdev->sb_size = sizeof(*sb);
798 ret = read_disk_sb(rdev, rdev->sb_size);
802 sb = page_address(rdev->sb_page);
805 * Two cases that we want to write new superblocks and rebuild:
806 * 1) New device (no matching magic number)
807 * 2) Device specified for rebuild (!In_sync w/ offset == 0)
809 if ((sb->magic != cpu_to_le32(DM_RAID_MAGIC)) ||
810 (!test_bit(In_sync, &rdev->flags) && !rdev->recovery_offset)) {
811 super_sync(rdev->mddev, rdev);
813 set_bit(FirstUse, &rdev->flags);
815 /* Force writing of superblocks to disk */
816 set_bit(MD_CHANGE_DEVS, &rdev->mddev->flags);
818 /* Any superblock is better than none, choose that if given */
819 return refdev ? 0 : 1;
825 events_sb = le64_to_cpu(sb->events);
827 refsb = page_address(refdev->sb_page);
828 events_refsb = le64_to_cpu(refsb->events);
830 return (events_sb > events_refsb) ? 1 : 0;
833 static int super_init_validation(struct mddev *mddev, struct md_rdev *rdev)
836 struct raid_set *rs = container_of(mddev, struct raid_set, md);
838 uint64_t failed_devices;
839 struct dm_raid_superblock *sb;
840 uint32_t new_devs = 0;
841 uint32_t rebuilds = 0;
843 struct dm_raid_superblock *sb2;
845 sb = page_address(rdev->sb_page);
846 events_sb = le64_to_cpu(sb->events);
847 failed_devices = le64_to_cpu(sb->failed_devices);
850 * Initialise to 1 if this is a new superblock.
852 mddev->events = events_sb ? : 1;
855 * Reshaping is not currently allowed
857 if ((le32_to_cpu(sb->level) != mddev->level) ||
858 (le32_to_cpu(sb->layout) != mddev->layout) ||
859 (le32_to_cpu(sb->stripe_sectors) != mddev->chunk_sectors)) {
860 DMERR("Reshaping arrays not yet supported.");
864 /* We can only change the number of devices in RAID1 right now */
865 if ((rs->raid_type->level != 1) &&
866 (le32_to_cpu(sb->num_devices) != mddev->raid_disks)) {
867 DMERR("Reshaping arrays not yet supported.");
871 if (!(rs->print_flags & (DMPF_SYNC | DMPF_NOSYNC)))
872 mddev->recovery_cp = le64_to_cpu(sb->array_resync_offset);
875 * During load, we set FirstUse if a new superblock was written.
876 * There are two reasons we might not have a superblock:
877 * 1) The array is brand new - in which case, all of the
878 * devices must have their In_sync bit set. Also,
879 * recovery_cp must be 0, unless forced.
880 * 2) This is a new device being added to an old array
881 * and the new device needs to be rebuilt - in which
882 * case the In_sync bit will /not/ be set and
883 * recovery_cp must be MaxSector.
885 rdev_for_each(r, mddev) {
886 if (!test_bit(In_sync, &r->flags)) {
887 DMINFO("Device %d specified for rebuild: "
888 "Clearing superblock", r->raid_disk);
890 } else if (test_bit(FirstUse, &r->flags))
895 if (new_devs == mddev->raid_disks) {
896 DMINFO("Superblocks created for new array");
897 set_bit(MD_ARRAY_FIRST_USE, &mddev->flags);
898 } else if (new_devs) {
899 DMERR("New device injected "
900 "into existing array without 'rebuild' "
901 "parameter specified");
904 } else if (new_devs) {
905 DMERR("'rebuild' devices cannot be "
906 "injected into an array with other first-time devices");
908 } else if (mddev->recovery_cp != MaxSector) {
909 DMERR("'rebuild' specified while array is not in-sync");
914 * Now we set the Faulty bit for those devices that are
915 * recorded in the superblock as failed.
917 rdev_for_each(r, mddev) {
920 sb2 = page_address(r->sb_page);
921 sb2->failed_devices = 0;
924 * Check for any device re-ordering.
926 if (!test_bit(FirstUse, &r->flags) && (r->raid_disk >= 0)) {
927 role = le32_to_cpu(sb2->array_position);
928 if (role != r->raid_disk) {
929 if (rs->raid_type->level != 1) {
930 rs->ti->error = "Cannot change device "
931 "positions in RAID array";
934 DMINFO("RAID1 device #%d now at position #%d",
939 * Partial recovery is performed on
940 * returning failed devices.
942 if (failed_devices & (1 << role))
943 set_bit(Faulty, &r->flags);
950 static int super_validate(struct mddev *mddev, struct md_rdev *rdev)
952 struct dm_raid_superblock *sb = page_address(rdev->sb_page);
955 * If mddev->events is not set, we know we have not yet initialized
958 if (!mddev->events && super_init_validation(mddev, rdev))
961 mddev->bitmap_info.offset = 4096 >> 9; /* Enable bitmap creation */
962 rdev->mddev->bitmap_info.default_offset = 4096 >> 9;
963 if (!test_bit(FirstUse, &rdev->flags)) {
964 rdev->recovery_offset = le64_to_cpu(sb->disk_recovery_offset);
965 if (rdev->recovery_offset != MaxSector)
966 clear_bit(In_sync, &rdev->flags);
970 * If a device comes back, set it as not In_sync and no longer faulty.
972 if (test_bit(Faulty, &rdev->flags)) {
973 clear_bit(Faulty, &rdev->flags);
974 clear_bit(In_sync, &rdev->flags);
975 rdev->saved_raid_disk = rdev->raid_disk;
976 rdev->recovery_offset = 0;
979 clear_bit(FirstUse, &rdev->flags);
985 * Analyse superblocks and select the freshest.
987 static int analyse_superblocks(struct dm_target *ti, struct raid_set *rs)
990 struct raid_dev *dev;
991 struct md_rdev *rdev, *tmp, *freshest;
992 struct mddev *mddev = &rs->md;
995 rdev_for_each_safe(rdev, tmp, mddev) {
997 * Skipping super_load due to DMPF_SYNC will cause
998 * the array to undergo initialization again as
999 * though it were new. This is the intended effect
1000 * of the "sync" directive.
1002 * When reshaping capability is added, we must ensure
1003 * that the "sync" directive is disallowed during the
1006 if (rs->print_flags & DMPF_SYNC)
1009 if (!rdev->meta_bdev)
1012 ret = super_load(rdev, freshest);
1021 dev = container_of(rdev, struct raid_dev, rdev);
1023 dm_put_device(ti, dev->meta_dev);
1025 dev->meta_dev = NULL;
1026 rdev->meta_bdev = NULL;
1029 put_page(rdev->sb_page);
1031 rdev->sb_page = NULL;
1033 rdev->sb_loaded = 0;
1036 * We might be able to salvage the data device
1037 * even though the meta device has failed. For
1038 * now, we behave as though '- -' had been
1039 * set for this device in the table.
1042 dm_put_device(ti, dev->data_dev);
1044 dev->data_dev = NULL;
1047 list_del(&rdev->same_set);
1054 if (validate_raid_redundancy(rs)) {
1055 rs->ti->error = "Insufficient redundancy to activate array";
1060 * Validation of the freshest device provides the source of
1061 * validation for the remaining devices.
1063 ti->error = "Unable to assemble array: Invalid superblocks";
1064 if (super_validate(mddev, freshest))
1067 rdev_for_each(rdev, mddev)
1068 if ((rdev != freshest) && super_validate(mddev, rdev))
1075 * Construct a RAID4/5/6 mapping:
1077 * <raid_type> <#raid_params> <raid_params> \
1078 * <#raid_devs> { <meta_dev1> <dev1> .. <meta_devN> <devN> }
1080 * <raid_params> varies by <raid_type>. See 'parse_raid_params' for
1081 * details on possible <raid_params>.
1083 static int raid_ctr(struct dm_target *ti, unsigned argc, char **argv)
1086 struct raid_type *rt;
1087 unsigned long num_raid_params, num_raid_devs;
1088 struct raid_set *rs = NULL;
1090 /* Must have at least <raid_type> <#raid_params> */
1092 ti->error = "Too few arguments";
1097 rt = get_raid_type(argv[0]);
1099 ti->error = "Unrecognised raid_type";
1105 /* number of RAID parameters */
1106 if (strict_strtoul(argv[0], 10, &num_raid_params) < 0) {
1107 ti->error = "Cannot understand number of RAID parameters";
1113 /* Skip over RAID params for now and find out # of devices */
1114 if (num_raid_params + 1 > argc) {
1115 ti->error = "Arguments do not agree with counts given";
1119 if ((strict_strtoul(argv[num_raid_params], 10, &num_raid_devs) < 0) ||
1120 (num_raid_devs >= INT_MAX)) {
1121 ti->error = "Cannot understand number of raid devices";
1125 rs = context_alloc(ti, rt, (unsigned)num_raid_devs);
1129 ret = parse_raid_params(rs, argv, (unsigned)num_raid_params);
1135 argc -= num_raid_params + 1; /* +1: we already have num_raid_devs */
1136 argv += num_raid_params + 1;
1138 if (argc != (num_raid_devs * 2)) {
1139 ti->error = "Supplied RAID devices does not match the count given";
1143 ret = dev_parms(rs, argv);
1147 rs->md.sync_super = super_sync;
1148 ret = analyse_superblocks(ti, rs);
1152 INIT_WORK(&rs->md.event_work, do_table_event);
1154 ti->num_flush_requests = 1;
1156 mutex_lock(&rs->md.reconfig_mutex);
1157 ret = md_run(&rs->md);
1158 rs->md.in_sync = 0; /* Assume already marked dirty */
1159 mutex_unlock(&rs->md.reconfig_mutex);
1162 ti->error = "Fail to run raid array";
1166 if (ti->len != rs->md.array_sectors) {
1167 ti->error = "Array size does not match requested target length";
1171 rs->callbacks.congested_fn = raid_is_congested;
1172 dm_table_add_target_callbacks(ti->table, &rs->callbacks);
1174 mddev_suspend(&rs->md);
1185 static void raid_dtr(struct dm_target *ti)
1187 struct raid_set *rs = ti->private;
1189 list_del_init(&rs->callbacks.list);
1194 static int raid_map(struct dm_target *ti, struct bio *bio)
1196 struct raid_set *rs = ti->private;
1197 struct mddev *mddev = &rs->md;
1199 mddev->pers->make_request(mddev, bio);
1201 return DM_MAPIO_SUBMITTED;
1204 static int raid_status(struct dm_target *ti, status_type_t type,
1205 unsigned status_flags, char *result, unsigned maxlen)
1207 struct raid_set *rs = ti->private;
1208 unsigned raid_param_cnt = 1; /* at least 1 for chunksize */
1210 int i, array_in_sync = 0;
1214 case STATUSTYPE_INFO:
1215 DMEMIT("%s %d ", rs->raid_type->name, rs->md.raid_disks);
1217 if (test_bit(MD_RECOVERY_RUNNING, &rs->md.recovery))
1218 sync = rs->md.curr_resync_completed;
1220 sync = rs->md.recovery_cp;
1222 if (sync >= rs->md.resync_max_sectors) {
1224 sync = rs->md.resync_max_sectors;
1227 * The array may be doing an initial sync, or it may
1228 * be rebuilding individual components. If all the
1229 * devices are In_sync, then it is the array that is
1230 * being initialized.
1232 for (i = 0; i < rs->md.raid_disks; i++)
1233 if (!test_bit(In_sync, &rs->dev[i].rdev.flags))
1237 * Status characters:
1238 * 'D' = Dead/Failed device
1239 * 'a' = Alive but not in-sync
1240 * 'A' = Alive and in-sync
1242 for (i = 0; i < rs->md.raid_disks; i++) {
1243 if (test_bit(Faulty, &rs->dev[i].rdev.flags))
1245 else if (!array_in_sync ||
1246 !test_bit(In_sync, &rs->dev[i].rdev.flags))
1254 * The in-sync ratio shows the progress of:
1255 * - Initializing the array
1256 * - Rebuilding a subset of devices of the array
1257 * The user can distinguish between the two by referring
1258 * to the status characters.
1260 DMEMIT(" %llu/%llu",
1261 (unsigned long long) sync,
1262 (unsigned long long) rs->md.resync_max_sectors);
1265 case STATUSTYPE_TABLE:
1266 /* The string you would use to construct this array */
1267 for (i = 0; i < rs->md.raid_disks; i++) {
1268 if ((rs->print_flags & DMPF_REBUILD) &&
1269 rs->dev[i].data_dev &&
1270 !test_bit(In_sync, &rs->dev[i].rdev.flags))
1271 raid_param_cnt += 2; /* for rebuilds */
1272 if (rs->dev[i].data_dev &&
1273 test_bit(WriteMostly, &rs->dev[i].rdev.flags))
1274 raid_param_cnt += 2;
1277 raid_param_cnt += (hweight32(rs->print_flags & ~DMPF_REBUILD) * 2);
1278 if (rs->print_flags & (DMPF_SYNC | DMPF_NOSYNC))
1281 DMEMIT("%s %u %u", rs->raid_type->name,
1282 raid_param_cnt, rs->md.chunk_sectors);
1284 if ((rs->print_flags & DMPF_SYNC) &&
1285 (rs->md.recovery_cp == MaxSector))
1287 if (rs->print_flags & DMPF_NOSYNC)
1290 for (i = 0; i < rs->md.raid_disks; i++)
1291 if ((rs->print_flags & DMPF_REBUILD) &&
1292 rs->dev[i].data_dev &&
1293 !test_bit(In_sync, &rs->dev[i].rdev.flags))
1294 DMEMIT(" rebuild %u", i);
1296 if (rs->print_flags & DMPF_DAEMON_SLEEP)
1297 DMEMIT(" daemon_sleep %lu",
1298 rs->md.bitmap_info.daemon_sleep);
1300 if (rs->print_flags & DMPF_MIN_RECOVERY_RATE)
1301 DMEMIT(" min_recovery_rate %d", rs->md.sync_speed_min);
1303 if (rs->print_flags & DMPF_MAX_RECOVERY_RATE)
1304 DMEMIT(" max_recovery_rate %d", rs->md.sync_speed_max);
1306 for (i = 0; i < rs->md.raid_disks; i++)
1307 if (rs->dev[i].data_dev &&
1308 test_bit(WriteMostly, &rs->dev[i].rdev.flags))
1309 DMEMIT(" write_mostly %u", i);
1311 if (rs->print_flags & DMPF_MAX_WRITE_BEHIND)
1312 DMEMIT(" max_write_behind %lu",
1313 rs->md.bitmap_info.max_write_behind);
1315 if (rs->print_flags & DMPF_STRIPE_CACHE) {
1316 struct r5conf *conf = rs->md.private;
1318 /* convert from kiB to sectors */
1319 DMEMIT(" stripe_cache %d",
1320 conf ? conf->max_nr_stripes * 2 : 0);
1323 if (rs->print_flags & DMPF_REGION_SIZE)
1324 DMEMIT(" region_size %lu",
1325 rs->md.bitmap_info.chunksize >> 9);
1327 if (rs->print_flags & DMPF_RAID10_COPIES)
1328 DMEMIT(" raid10_copies %u",
1329 raid10_md_layout_to_copies(rs->md.layout));
1331 if (rs->print_flags & DMPF_RAID10_FORMAT)
1332 DMEMIT(" raid10_format near");
1334 DMEMIT(" %d", rs->md.raid_disks);
1335 for (i = 0; i < rs->md.raid_disks; i++) {
1336 if (rs->dev[i].meta_dev)
1337 DMEMIT(" %s", rs->dev[i].meta_dev->name);
1341 if (rs->dev[i].data_dev)
1342 DMEMIT(" %s", rs->dev[i].data_dev->name);
1351 static int raid_iterate_devices(struct dm_target *ti, iterate_devices_callout_fn fn, void *data)
1353 struct raid_set *rs = ti->private;
1357 for (i = 0; !ret && i < rs->md.raid_disks; i++)
1358 if (rs->dev[i].data_dev)
1360 rs->dev[i].data_dev,
1361 0, /* No offset on data devs */
1368 static void raid_io_hints(struct dm_target *ti, struct queue_limits *limits)
1370 struct raid_set *rs = ti->private;
1371 unsigned chunk_size = rs->md.chunk_sectors << 9;
1372 struct r5conf *conf = rs->md.private;
1374 blk_limits_io_min(limits, chunk_size);
1375 blk_limits_io_opt(limits, chunk_size * (conf->raid_disks - conf->max_degraded));
1378 static void raid_presuspend(struct dm_target *ti)
1380 struct raid_set *rs = ti->private;
1382 md_stop_writes(&rs->md);
1385 static void raid_postsuspend(struct dm_target *ti)
1387 struct raid_set *rs = ti->private;
1389 mddev_suspend(&rs->md);
1392 static void raid_resume(struct dm_target *ti)
1394 struct raid_set *rs = ti->private;
1396 set_bit(MD_CHANGE_DEVS, &rs->md.flags);
1397 if (!rs->bitmap_loaded) {
1398 bitmap_load(&rs->md);
1399 rs->bitmap_loaded = 1;
1402 clear_bit(MD_RECOVERY_FROZEN, &rs->md.recovery);
1403 mddev_resume(&rs->md);
1406 static struct target_type raid_target = {
1408 .version = {1, 4, 1},
1409 .module = THIS_MODULE,
1413 .status = raid_status,
1414 .iterate_devices = raid_iterate_devices,
1415 .io_hints = raid_io_hints,
1416 .presuspend = raid_presuspend,
1417 .postsuspend = raid_postsuspend,
1418 .resume = raid_resume,
1421 static int __init dm_raid_init(void)
1423 return dm_register_target(&raid_target);
1426 static void __exit dm_raid_exit(void)
1428 dm_unregister_target(&raid_target);
1431 module_init(dm_raid_init);
1432 module_exit(dm_raid_exit);
1434 MODULE_DESCRIPTION(DM_NAME " raid4/5/6 target");
1435 MODULE_ALIAS("dm-raid1");
1436 MODULE_ALIAS("dm-raid10");
1437 MODULE_ALIAS("dm-raid4");
1438 MODULE_ALIAS("dm-raid5");
1439 MODULE_ALIAS("dm-raid6");
1440 MODULE_AUTHOR("Neil Brown <dm-devel@redhat.com>");
1441 MODULE_LICENSE("GPL");