Loading...
1/*
2 * Copyright (C) 2010-2011 Neil Brown
3 * Copyright (C) 2010-2011 Red Hat, Inc. All rights reserved.
4 *
5 * This file is released under the GPL.
6 */
7
8#include <linux/slab.h>
9
10#include "md.h"
11#include "raid1.h"
12#include "raid5.h"
13#include "bitmap.h"
14
15#include <linux/device-mapper.h>
16
17#define DM_MSG_PREFIX "raid"
18
19/*
20 * The following flags are used by dm-raid.c to set up the array state.
21 * They must be cleared before md_run is called.
22 */
23#define FirstUse 10 /* rdev flag */
24
25struct raid_dev {
26 /*
27 * Two DM devices, one to hold metadata and one to hold the
28 * actual data/parity. The reason for this is to not confuse
29 * ti->len and give more flexibility in altering size and
30 * characteristics.
31 *
32 * While it is possible for this device to be associated
33 * with a different physical device than the data_dev, it
34 * is intended for it to be the same.
35 * |--------- Physical Device ---------|
36 * |- meta_dev -|------ data_dev ------|
37 */
38 struct dm_dev *meta_dev;
39 struct dm_dev *data_dev;
40 struct mdk_rdev_s rdev;
41};
42
43/*
44 * Flags for rs->print_flags field.
45 */
46#define DMPF_SYNC 0x1
47#define DMPF_NOSYNC 0x2
48#define DMPF_REBUILD 0x4
49#define DMPF_DAEMON_SLEEP 0x8
50#define DMPF_MIN_RECOVERY_RATE 0x10
51#define DMPF_MAX_RECOVERY_RATE 0x20
52#define DMPF_MAX_WRITE_BEHIND 0x40
53#define DMPF_STRIPE_CACHE 0x80
54#define DMPF_REGION_SIZE 0X100
55struct raid_set {
56 struct dm_target *ti;
57
58 uint64_t print_flags;
59
60 struct mddev_s md;
61 struct raid_type *raid_type;
62 struct dm_target_callbacks callbacks;
63
64 struct raid_dev dev[0];
65};
66
67/* Supported raid types and properties. */
68static struct raid_type {
69 const char *name; /* RAID algorithm. */
70 const char *descr; /* Descriptor text for logging. */
71 const unsigned parity_devs; /* # of parity devices. */
72 const unsigned minimal_devs; /* minimal # of devices in set. */
73 const unsigned level; /* RAID level. */
74 const unsigned algorithm; /* RAID algorithm. */
75} raid_types[] = {
76 {"raid1", "RAID1 (mirroring)", 0, 2, 1, 0 /* NONE */},
77 {"raid4", "RAID4 (dedicated parity disk)", 1, 2, 5, ALGORITHM_PARITY_0},
78 {"raid5_la", "RAID5 (left asymmetric)", 1, 2, 5, ALGORITHM_LEFT_ASYMMETRIC},
79 {"raid5_ra", "RAID5 (right asymmetric)", 1, 2, 5, ALGORITHM_RIGHT_ASYMMETRIC},
80 {"raid5_ls", "RAID5 (left symmetric)", 1, 2, 5, ALGORITHM_LEFT_SYMMETRIC},
81 {"raid5_rs", "RAID5 (right symmetric)", 1, 2, 5, ALGORITHM_RIGHT_SYMMETRIC},
82 {"raid6_zr", "RAID6 (zero restart)", 2, 4, 6, ALGORITHM_ROTATING_ZERO_RESTART},
83 {"raid6_nr", "RAID6 (N restart)", 2, 4, 6, ALGORITHM_ROTATING_N_RESTART},
84 {"raid6_nc", "RAID6 (N continue)", 2, 4, 6, ALGORITHM_ROTATING_N_CONTINUE}
85};
86
87static struct raid_type *get_raid_type(char *name)
88{
89 int i;
90
91 for (i = 0; i < ARRAY_SIZE(raid_types); i++)
92 if (!strcmp(raid_types[i].name, name))
93 return &raid_types[i];
94
95 return NULL;
96}
97
98static struct raid_set *context_alloc(struct dm_target *ti, struct raid_type *raid_type, unsigned raid_devs)
99{
100 unsigned i;
101 struct raid_set *rs;
102 sector_t sectors_per_dev;
103
104 if (raid_devs <= raid_type->parity_devs) {
105 ti->error = "Insufficient number of devices";
106 return ERR_PTR(-EINVAL);
107 }
108
109 sectors_per_dev = ti->len;
110 if ((raid_type->level > 1) &&
111 sector_div(sectors_per_dev, (raid_devs - raid_type->parity_devs))) {
112 ti->error = "Target length not divisible by number of data devices";
113 return ERR_PTR(-EINVAL);
114 }
115
116 rs = kzalloc(sizeof(*rs) + raid_devs * sizeof(rs->dev[0]), GFP_KERNEL);
117 if (!rs) {
118 ti->error = "Cannot allocate raid context";
119 return ERR_PTR(-ENOMEM);
120 }
121
122 mddev_init(&rs->md);
123
124 rs->ti = ti;
125 rs->raid_type = raid_type;
126 rs->md.raid_disks = raid_devs;
127 rs->md.level = raid_type->level;
128 rs->md.new_level = rs->md.level;
129 rs->md.dev_sectors = sectors_per_dev;
130 rs->md.layout = raid_type->algorithm;
131 rs->md.new_layout = rs->md.layout;
132 rs->md.delta_disks = 0;
133 rs->md.recovery_cp = 0;
134
135 for (i = 0; i < raid_devs; i++)
136 md_rdev_init(&rs->dev[i].rdev);
137
138 /*
139 * Remaining items to be initialized by further RAID params:
140 * rs->md.persistent
141 * rs->md.external
142 * rs->md.chunk_sectors
143 * rs->md.new_chunk_sectors
144 */
145
146 return rs;
147}
148
149static void context_free(struct raid_set *rs)
150{
151 int i;
152
153 for (i = 0; i < rs->md.raid_disks; i++) {
154 if (rs->dev[i].meta_dev)
155 dm_put_device(rs->ti, rs->dev[i].meta_dev);
156 if (rs->dev[i].rdev.sb_page)
157 put_page(rs->dev[i].rdev.sb_page);
158 rs->dev[i].rdev.sb_page = NULL;
159 rs->dev[i].rdev.sb_loaded = 0;
160 if (rs->dev[i].data_dev)
161 dm_put_device(rs->ti, rs->dev[i].data_dev);
162 }
163
164 kfree(rs);
165}
166
167/*
168 * For every device we have two words
169 * <meta_dev>: meta device name or '-' if missing
170 * <data_dev>: data device name or '-' if missing
171 *
172 * The following are permitted:
173 * - -
174 * - <data_dev>
175 * <meta_dev> <data_dev>
176 *
177 * The following is not allowed:
178 * <meta_dev> -
179 *
180 * This code parses those words. If there is a failure,
181 * the caller must use context_free to unwind the operations.
182 */
183static int dev_parms(struct raid_set *rs, char **argv)
184{
185 int i;
186 int rebuild = 0;
187 int metadata_available = 0;
188 int ret = 0;
189
190 for (i = 0; i < rs->md.raid_disks; i++, argv += 2) {
191 rs->dev[i].rdev.raid_disk = i;
192
193 rs->dev[i].meta_dev = NULL;
194 rs->dev[i].data_dev = NULL;
195
196 /*
197 * There are no offsets, since there is a separate device
198 * for data and metadata.
199 */
200 rs->dev[i].rdev.data_offset = 0;
201 rs->dev[i].rdev.mddev = &rs->md;
202
203 if (strcmp(argv[0], "-")) {
204 ret = dm_get_device(rs->ti, argv[0],
205 dm_table_get_mode(rs->ti->table),
206 &rs->dev[i].meta_dev);
207 rs->ti->error = "RAID metadata device lookup failure";
208 if (ret)
209 return ret;
210
211 rs->dev[i].rdev.sb_page = alloc_page(GFP_KERNEL);
212 if (!rs->dev[i].rdev.sb_page)
213 return -ENOMEM;
214 }
215
216 if (!strcmp(argv[1], "-")) {
217 if (!test_bit(In_sync, &rs->dev[i].rdev.flags) &&
218 (!rs->dev[i].rdev.recovery_offset)) {
219 rs->ti->error = "Drive designated for rebuild not specified";
220 return -EINVAL;
221 }
222
223 rs->ti->error = "No data device supplied with metadata device";
224 if (rs->dev[i].meta_dev)
225 return -EINVAL;
226
227 continue;
228 }
229
230 ret = dm_get_device(rs->ti, argv[1],
231 dm_table_get_mode(rs->ti->table),
232 &rs->dev[i].data_dev);
233 if (ret) {
234 rs->ti->error = "RAID device lookup failure";
235 return ret;
236 }
237
238 if (rs->dev[i].meta_dev) {
239 metadata_available = 1;
240 rs->dev[i].rdev.meta_bdev = rs->dev[i].meta_dev->bdev;
241 }
242 rs->dev[i].rdev.bdev = rs->dev[i].data_dev->bdev;
243 list_add(&rs->dev[i].rdev.same_set, &rs->md.disks);
244 if (!test_bit(In_sync, &rs->dev[i].rdev.flags))
245 rebuild++;
246 }
247
248 if (metadata_available) {
249 rs->md.external = 0;
250 rs->md.persistent = 1;
251 rs->md.major_version = 2;
252 } else if (rebuild && !rs->md.recovery_cp) {
253 /*
254 * Without metadata, we will not be able to tell if the array
255 * is in-sync or not - we must assume it is not. Therefore,
256 * it is impossible to rebuild a drive.
257 *
258 * Even if there is metadata, the on-disk information may
259 * indicate that the array is not in-sync and it will then
260 * fail at that time.
261 *
262 * User could specify 'nosync' option if desperate.
263 */
264 DMERR("Unable to rebuild drive while array is not in-sync");
265 rs->ti->error = "RAID device lookup failure";
266 return -EINVAL;
267 }
268
269 return 0;
270}
271
272/*
273 * validate_region_size
274 * @rs
275 * @region_size: region size in sectors. If 0, pick a size (4MiB default).
276 *
277 * Set rs->md.bitmap_info.chunksize (which really refers to 'region size').
278 * Ensure that (ti->len/region_size < 2^21) - required by MD bitmap.
279 *
280 * Returns: 0 on success, -EINVAL on failure.
281 */
282static int validate_region_size(struct raid_set *rs, unsigned long region_size)
283{
284 unsigned long min_region_size = rs->ti->len / (1 << 21);
285
286 if (!region_size) {
287 /*
288 * Choose a reasonable default. All figures in sectors.
289 */
290 if (min_region_size > (1 << 13)) {
291 DMINFO("Choosing default region size of %lu sectors",
292 region_size);
293 region_size = min_region_size;
294 } else {
295 DMINFO("Choosing default region size of 4MiB");
296 region_size = 1 << 13; /* sectors */
297 }
298 } else {
299 /*
300 * Validate user-supplied value.
301 */
302 if (region_size > rs->ti->len) {
303 rs->ti->error = "Supplied region size is too large";
304 return -EINVAL;
305 }
306
307 if (region_size < min_region_size) {
308 DMERR("Supplied region_size (%lu sectors) below minimum (%lu)",
309 region_size, min_region_size);
310 rs->ti->error = "Supplied region size is too small";
311 return -EINVAL;
312 }
313
314 if (!is_power_of_2(region_size)) {
315 rs->ti->error = "Region size is not a power of 2";
316 return -EINVAL;
317 }
318
319 if (region_size < rs->md.chunk_sectors) {
320 rs->ti->error = "Region size is smaller than the chunk size";
321 return -EINVAL;
322 }
323 }
324
325 /*
326 * Convert sectors to bytes.
327 */
328 rs->md.bitmap_info.chunksize = (region_size << 9);
329
330 return 0;
331}
332
333/*
334 * Possible arguments are...
335 * <chunk_size> [optional_args]
336 *
337 * Argument definitions
338 * <chunk_size> The number of sectors per disk that
339 * will form the "stripe"
340 * [[no]sync] Force or prevent recovery of the
341 * entire array
342 * [rebuild <idx>] Rebuild the drive indicated by the index
343 * [daemon_sleep <ms>] Time between bitmap daemon work to
344 * clear bits
345 * [min_recovery_rate <kB/sec/disk>] Throttle RAID initialization
346 * [max_recovery_rate <kB/sec/disk>] Throttle RAID initialization
347 * [write_mostly <idx>] Indicate a write mostly drive via index
348 * [max_write_behind <sectors>] See '-write-behind=' (man mdadm)
349 * [stripe_cache <sectors>] Stripe cache size for higher RAIDs
350 * [region_size <sectors>] Defines granularity of bitmap
351 */
352static int parse_raid_params(struct raid_set *rs, char **argv,
353 unsigned num_raid_params)
354{
355 unsigned i, rebuild_cnt = 0;
356 unsigned long value, region_size = 0;
357 char *key;
358
359 /*
360 * First, parse the in-order required arguments
361 * "chunk_size" is the only argument of this type.
362 */
363 if ((strict_strtoul(argv[0], 10, &value) < 0)) {
364 rs->ti->error = "Bad chunk size";
365 return -EINVAL;
366 } else if (rs->raid_type->level == 1) {
367 if (value)
368 DMERR("Ignoring chunk size parameter for RAID 1");
369 value = 0;
370 } else if (!is_power_of_2(value)) {
371 rs->ti->error = "Chunk size must be a power of 2";
372 return -EINVAL;
373 } else if (value < 8) {
374 rs->ti->error = "Chunk size value is too small";
375 return -EINVAL;
376 }
377
378 rs->md.new_chunk_sectors = rs->md.chunk_sectors = value;
379 argv++;
380 num_raid_params--;
381
382 /*
383 * We set each individual device as In_sync with a completed
384 * 'recovery_offset'. If there has been a device failure or
385 * replacement then one of the following cases applies:
386 *
387 * 1) User specifies 'rebuild'.
388 * - Device is reset when param is read.
389 * 2) A new device is supplied.
390 * - No matching superblock found, resets device.
391 * 3) Device failure was transient and returns on reload.
392 * - Failure noticed, resets device for bitmap replay.
393 * 4) Device hadn't completed recovery after previous failure.
394 * - Superblock is read and overrides recovery_offset.
395 *
396 * What is found in the superblocks of the devices is always
397 * authoritative, unless 'rebuild' or '[no]sync' was specified.
398 */
399 for (i = 0; i < rs->md.raid_disks; i++) {
400 set_bit(In_sync, &rs->dev[i].rdev.flags);
401 rs->dev[i].rdev.recovery_offset = MaxSector;
402 }
403
404 /*
405 * Second, parse the unordered optional arguments
406 */
407 for (i = 0; i < num_raid_params; i++) {
408 if (!strcasecmp(argv[i], "nosync")) {
409 rs->md.recovery_cp = MaxSector;
410 rs->print_flags |= DMPF_NOSYNC;
411 continue;
412 }
413 if (!strcasecmp(argv[i], "sync")) {
414 rs->md.recovery_cp = 0;
415 rs->print_flags |= DMPF_SYNC;
416 continue;
417 }
418
419 /* The rest of the optional arguments come in key/value pairs */
420 if ((i + 1) >= num_raid_params) {
421 rs->ti->error = "Wrong number of raid parameters given";
422 return -EINVAL;
423 }
424
425 key = argv[i++];
426 if (strict_strtoul(argv[i], 10, &value) < 0) {
427 rs->ti->error = "Bad numerical argument given in raid params";
428 return -EINVAL;
429 }
430
431 if (!strcasecmp(key, "rebuild")) {
432 rebuild_cnt++;
433 if (((rs->raid_type->level != 1) &&
434 (rebuild_cnt > rs->raid_type->parity_devs)) ||
435 ((rs->raid_type->level == 1) &&
436 (rebuild_cnt > (rs->md.raid_disks - 1)))) {
437 rs->ti->error = "Too many rebuild devices specified for given RAID type";
438 return -EINVAL;
439 }
440 if (value > rs->md.raid_disks) {
441 rs->ti->error = "Invalid rebuild index given";
442 return -EINVAL;
443 }
444 clear_bit(In_sync, &rs->dev[value].rdev.flags);
445 rs->dev[value].rdev.recovery_offset = 0;
446 rs->print_flags |= DMPF_REBUILD;
447 } else if (!strcasecmp(key, "write_mostly")) {
448 if (rs->raid_type->level != 1) {
449 rs->ti->error = "write_mostly option is only valid for RAID1";
450 return -EINVAL;
451 }
452 if (value >= rs->md.raid_disks) {
453 rs->ti->error = "Invalid write_mostly drive index given";
454 return -EINVAL;
455 }
456 set_bit(WriteMostly, &rs->dev[value].rdev.flags);
457 } else if (!strcasecmp(key, "max_write_behind")) {
458 if (rs->raid_type->level != 1) {
459 rs->ti->error = "max_write_behind option is only valid for RAID1";
460 return -EINVAL;
461 }
462 rs->print_flags |= DMPF_MAX_WRITE_BEHIND;
463
464 /*
465 * In device-mapper, we specify things in sectors, but
466 * MD records this value in kB
467 */
468 value /= 2;
469 if (value > COUNTER_MAX) {
470 rs->ti->error = "Max write-behind limit out of range";
471 return -EINVAL;
472 }
473 rs->md.bitmap_info.max_write_behind = value;
474 } else if (!strcasecmp(key, "daemon_sleep")) {
475 rs->print_flags |= DMPF_DAEMON_SLEEP;
476 if (!value || (value > MAX_SCHEDULE_TIMEOUT)) {
477 rs->ti->error = "daemon sleep period out of range";
478 return -EINVAL;
479 }
480 rs->md.bitmap_info.daemon_sleep = value;
481 } else if (!strcasecmp(key, "stripe_cache")) {
482 rs->print_flags |= DMPF_STRIPE_CACHE;
483
484 /*
485 * In device-mapper, we specify things in sectors, but
486 * MD records this value in kB
487 */
488 value /= 2;
489
490 if (rs->raid_type->level < 5) {
491 rs->ti->error = "Inappropriate argument: stripe_cache";
492 return -EINVAL;
493 }
494 if (raid5_set_cache_size(&rs->md, (int)value)) {
495 rs->ti->error = "Bad stripe_cache size";
496 return -EINVAL;
497 }
498 } else if (!strcasecmp(key, "min_recovery_rate")) {
499 rs->print_flags |= DMPF_MIN_RECOVERY_RATE;
500 if (value > INT_MAX) {
501 rs->ti->error = "min_recovery_rate out of range";
502 return -EINVAL;
503 }
504 rs->md.sync_speed_min = (int)value;
505 } else if (!strcasecmp(key, "max_recovery_rate")) {
506 rs->print_flags |= DMPF_MAX_RECOVERY_RATE;
507 if (value > INT_MAX) {
508 rs->ti->error = "max_recovery_rate out of range";
509 return -EINVAL;
510 }
511 rs->md.sync_speed_max = (int)value;
512 } else if (!strcasecmp(key, "region_size")) {
513 rs->print_flags |= DMPF_REGION_SIZE;
514 region_size = value;
515 } else {
516 DMERR("Unable to parse RAID parameter: %s", key);
517 rs->ti->error = "Unable to parse RAID parameters";
518 return -EINVAL;
519 }
520 }
521
522 if (validate_region_size(rs, region_size))
523 return -EINVAL;
524
525 if (rs->md.chunk_sectors)
526 rs->ti->split_io = rs->md.chunk_sectors;
527 else
528 rs->ti->split_io = region_size;
529
530 if (rs->md.chunk_sectors)
531 rs->ti->split_io = rs->md.chunk_sectors;
532 else
533 rs->ti->split_io = region_size;
534
535 /* Assume there are no metadata devices until the drives are parsed */
536 rs->md.persistent = 0;
537 rs->md.external = 1;
538
539 return 0;
540}
541
542static void do_table_event(struct work_struct *ws)
543{
544 struct raid_set *rs = container_of(ws, struct raid_set, md.event_work);
545
546 dm_table_event(rs->ti->table);
547}
548
549static int raid_is_congested(struct dm_target_callbacks *cb, int bits)
550{
551 struct raid_set *rs = container_of(cb, struct raid_set, callbacks);
552
553 if (rs->raid_type->level == 1)
554 return md_raid1_congested(&rs->md, bits);
555
556 return md_raid5_congested(&rs->md, bits);
557}
558
559/*
560 * This structure is never routinely used by userspace, unlike md superblocks.
561 * Devices with this superblock should only ever be accessed via device-mapper.
562 */
563#define DM_RAID_MAGIC 0x64526D44
564struct dm_raid_superblock {
565 __le32 magic; /* "DmRd" */
566 __le32 features; /* Used to indicate possible future changes */
567
568 __le32 num_devices; /* Number of devices in this array. (Max 64) */
569 __le32 array_position; /* The position of this drive in the array */
570
571 __le64 events; /* Incremented by md when superblock updated */
572 __le64 failed_devices; /* Bit field of devices to indicate failures */
573
574 /*
575 * This offset tracks the progress of the repair or replacement of
576 * an individual drive.
577 */
578 __le64 disk_recovery_offset;
579
580 /*
581 * This offset tracks the progress of the initial array
582 * synchronisation/parity calculation.
583 */
584 __le64 array_resync_offset;
585
586 /*
587 * RAID characteristics
588 */
589 __le32 level;
590 __le32 layout;
591 __le32 stripe_sectors;
592
593 __u8 pad[452]; /* Round struct to 512 bytes. */
594 /* Always set to 0 when writing. */
595} __packed;
596
597static int read_disk_sb(mdk_rdev_t *rdev, int size)
598{
599 BUG_ON(!rdev->sb_page);
600
601 if (rdev->sb_loaded)
602 return 0;
603
604 if (!sync_page_io(rdev, 0, size, rdev->sb_page, READ, 1)) {
605 DMERR("Failed to read device superblock");
606 return -EINVAL;
607 }
608
609 rdev->sb_loaded = 1;
610
611 return 0;
612}
613
614static void super_sync(mddev_t *mddev, mdk_rdev_t *rdev)
615{
616 mdk_rdev_t *r, *t;
617 uint64_t failed_devices;
618 struct dm_raid_superblock *sb;
619
620 sb = page_address(rdev->sb_page);
621 failed_devices = le64_to_cpu(sb->failed_devices);
622
623 rdev_for_each(r, t, mddev)
624 if ((r->raid_disk >= 0) && test_bit(Faulty, &r->flags))
625 failed_devices |= (1ULL << r->raid_disk);
626
627 memset(sb, 0, sizeof(*sb));
628
629 sb->magic = cpu_to_le32(DM_RAID_MAGIC);
630 sb->features = cpu_to_le32(0); /* No features yet */
631
632 sb->num_devices = cpu_to_le32(mddev->raid_disks);
633 sb->array_position = cpu_to_le32(rdev->raid_disk);
634
635 sb->events = cpu_to_le64(mddev->events);
636 sb->failed_devices = cpu_to_le64(failed_devices);
637
638 sb->disk_recovery_offset = cpu_to_le64(rdev->recovery_offset);
639 sb->array_resync_offset = cpu_to_le64(mddev->recovery_cp);
640
641 sb->level = cpu_to_le32(mddev->level);
642 sb->layout = cpu_to_le32(mddev->layout);
643 sb->stripe_sectors = cpu_to_le32(mddev->chunk_sectors);
644}
645
646/*
647 * super_load
648 *
649 * This function creates a superblock if one is not found on the device
650 * and will decide which superblock to use if there's a choice.
651 *
652 * Return: 1 if use rdev, 0 if use refdev, -Exxx otherwise
653 */
654static int super_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev)
655{
656 int ret;
657 struct dm_raid_superblock *sb;
658 struct dm_raid_superblock *refsb;
659 uint64_t events_sb, events_refsb;
660
661 rdev->sb_start = 0;
662 rdev->sb_size = sizeof(*sb);
663
664 ret = read_disk_sb(rdev, rdev->sb_size);
665 if (ret)
666 return ret;
667
668 sb = page_address(rdev->sb_page);
669 if (sb->magic != cpu_to_le32(DM_RAID_MAGIC)) {
670 super_sync(rdev->mddev, rdev);
671
672 set_bit(FirstUse, &rdev->flags);
673
674 /* Force writing of superblocks to disk */
675 set_bit(MD_CHANGE_DEVS, &rdev->mddev->flags);
676
677 /* Any superblock is better than none, choose that if given */
678 return refdev ? 0 : 1;
679 }
680
681 if (!refdev)
682 return 1;
683
684 events_sb = le64_to_cpu(sb->events);
685
686 refsb = page_address(refdev->sb_page);
687 events_refsb = le64_to_cpu(refsb->events);
688
689 return (events_sb > events_refsb) ? 1 : 0;
690}
691
692static int super_init_validation(mddev_t *mddev, mdk_rdev_t *rdev)
693{
694 int role;
695 struct raid_set *rs = container_of(mddev, struct raid_set, md);
696 uint64_t events_sb;
697 uint64_t failed_devices;
698 struct dm_raid_superblock *sb;
699 uint32_t new_devs = 0;
700 uint32_t rebuilds = 0;
701 mdk_rdev_t *r, *t;
702 struct dm_raid_superblock *sb2;
703
704 sb = page_address(rdev->sb_page);
705 events_sb = le64_to_cpu(sb->events);
706 failed_devices = le64_to_cpu(sb->failed_devices);
707
708 /*
709 * Initialise to 1 if this is a new superblock.
710 */
711 mddev->events = events_sb ? : 1;
712
713 /*
714 * Reshaping is not currently allowed
715 */
716 if ((le32_to_cpu(sb->level) != mddev->level) ||
717 (le32_to_cpu(sb->layout) != mddev->layout) ||
718 (le32_to_cpu(sb->stripe_sectors) != mddev->chunk_sectors)) {
719 DMERR("Reshaping arrays not yet supported.");
720 return -EINVAL;
721 }
722
723 /* We can only change the number of devices in RAID1 right now */
724 if ((rs->raid_type->level != 1) &&
725 (le32_to_cpu(sb->num_devices) != mddev->raid_disks)) {
726 DMERR("Reshaping arrays not yet supported.");
727 return -EINVAL;
728 }
729
730 if (!(rs->print_flags & (DMPF_SYNC | DMPF_NOSYNC)))
731 mddev->recovery_cp = le64_to_cpu(sb->array_resync_offset);
732
733 /*
734 * During load, we set FirstUse if a new superblock was written.
735 * There are two reasons we might not have a superblock:
736 * 1) The array is brand new - in which case, all of the
737 * devices must have their In_sync bit set. Also,
738 * recovery_cp must be 0, unless forced.
739 * 2) This is a new device being added to an old array
740 * and the new device needs to be rebuilt - in which
741 * case the In_sync bit will /not/ be set and
742 * recovery_cp must be MaxSector.
743 */
744 rdev_for_each(r, t, mddev) {
745 if (!test_bit(In_sync, &r->flags)) {
746 if (!test_bit(FirstUse, &r->flags))
747 DMERR("Superblock area of "
748 "rebuild device %d should have been "
749 "cleared.", r->raid_disk);
750 set_bit(FirstUse, &r->flags);
751 rebuilds++;
752 } else if (test_bit(FirstUse, &r->flags))
753 new_devs++;
754 }
755
756 if (!rebuilds) {
757 if (new_devs == mddev->raid_disks) {
758 DMINFO("Superblocks created for new array");
759 set_bit(MD_ARRAY_FIRST_USE, &mddev->flags);
760 } else if (new_devs) {
761 DMERR("New device injected "
762 "into existing array without 'rebuild' "
763 "parameter specified");
764 return -EINVAL;
765 }
766 } else if (new_devs) {
767 DMERR("'rebuild' devices cannot be "
768 "injected into an array with other first-time devices");
769 return -EINVAL;
770 } else if (mddev->recovery_cp != MaxSector) {
771 DMERR("'rebuild' specified while array is not in-sync");
772 return -EINVAL;
773 }
774
775 /*
776 * Now we set the Faulty bit for those devices that are
777 * recorded in the superblock as failed.
778 */
779 rdev_for_each(r, t, mddev) {
780 if (!r->sb_page)
781 continue;
782 sb2 = page_address(r->sb_page);
783 sb2->failed_devices = 0;
784
785 /*
786 * Check for any device re-ordering.
787 */
788 if (!test_bit(FirstUse, &r->flags) && (r->raid_disk >= 0)) {
789 role = le32_to_cpu(sb2->array_position);
790 if (role != r->raid_disk) {
791 if (rs->raid_type->level != 1) {
792 rs->ti->error = "Cannot change device "
793 "positions in RAID array";
794 return -EINVAL;
795 }
796 DMINFO("RAID1 device #%d now at position #%d",
797 role, r->raid_disk);
798 }
799
800 /*
801 * Partial recovery is performed on
802 * returning failed devices.
803 */
804 if (failed_devices & (1 << role))
805 set_bit(Faulty, &r->flags);
806 }
807 }
808
809 return 0;
810}
811
812static int super_validate(mddev_t *mddev, mdk_rdev_t *rdev)
813{
814 struct dm_raid_superblock *sb = page_address(rdev->sb_page);
815
816 /*
817 * If mddev->events is not set, we know we have not yet initialized
818 * the array.
819 */
820 if (!mddev->events && super_init_validation(mddev, rdev))
821 return -EINVAL;
822
823 mddev->bitmap_info.offset = 4096 >> 9; /* Enable bitmap creation */
824 rdev->mddev->bitmap_info.default_offset = 4096 >> 9;
825 if (!test_bit(FirstUse, &rdev->flags)) {
826 rdev->recovery_offset = le64_to_cpu(sb->disk_recovery_offset);
827 if (rdev->recovery_offset != MaxSector)
828 clear_bit(In_sync, &rdev->flags);
829 }
830
831 /*
832 * If a device comes back, set it as not In_sync and no longer faulty.
833 */
834 if (test_bit(Faulty, &rdev->flags)) {
835 clear_bit(Faulty, &rdev->flags);
836 clear_bit(In_sync, &rdev->flags);
837 rdev->saved_raid_disk = rdev->raid_disk;
838 rdev->recovery_offset = 0;
839 }
840
841 clear_bit(FirstUse, &rdev->flags);
842
843 return 0;
844}
845
846/*
847 * Analyse superblocks and select the freshest.
848 */
849static int analyse_superblocks(struct dm_target *ti, struct raid_set *rs)
850{
851 int ret;
852 mdk_rdev_t *rdev, *freshest, *tmp;
853 mddev_t *mddev = &rs->md;
854
855 freshest = NULL;
856 rdev_for_each(rdev, tmp, mddev) {
857 if (!rdev->meta_bdev)
858 continue;
859
860 ret = super_load(rdev, freshest);
861
862 switch (ret) {
863 case 1:
864 freshest = rdev;
865 break;
866 case 0:
867 break;
868 default:
869 ti->error = "Failed to load superblock";
870 return ret;
871 }
872 }
873
874 if (!freshest)
875 return 0;
876
877 /*
878 * Validation of the freshest device provides the source of
879 * validation for the remaining devices.
880 */
881 ti->error = "Unable to assemble array: Invalid superblocks";
882 if (super_validate(mddev, freshest))
883 return -EINVAL;
884
885 rdev_for_each(rdev, tmp, mddev)
886 if ((rdev != freshest) && super_validate(mddev, rdev))
887 return -EINVAL;
888
889 return 0;
890}
891
892/*
893 * Construct a RAID4/5/6 mapping:
894 * Args:
895 * <raid_type> <#raid_params> <raid_params> \
896 * <#raid_devs> { <meta_dev1> <dev1> .. <meta_devN> <devN> }
897 *
898 * <raid_params> varies by <raid_type>. See 'parse_raid_params' for
899 * details on possible <raid_params>.
900 */
901static int raid_ctr(struct dm_target *ti, unsigned argc, char **argv)
902{
903 int ret;
904 struct raid_type *rt;
905 unsigned long num_raid_params, num_raid_devs;
906 struct raid_set *rs = NULL;
907
908 /* Must have at least <raid_type> <#raid_params> */
909 if (argc < 2) {
910 ti->error = "Too few arguments";
911 return -EINVAL;
912 }
913
914 /* raid type */
915 rt = get_raid_type(argv[0]);
916 if (!rt) {
917 ti->error = "Unrecognised raid_type";
918 return -EINVAL;
919 }
920 argc--;
921 argv++;
922
923 /* number of RAID parameters */
924 if (strict_strtoul(argv[0], 10, &num_raid_params) < 0) {
925 ti->error = "Cannot understand number of RAID parameters";
926 return -EINVAL;
927 }
928 argc--;
929 argv++;
930
931 /* Skip over RAID params for now and find out # of devices */
932 if (num_raid_params + 1 > argc) {
933 ti->error = "Arguments do not agree with counts given";
934 return -EINVAL;
935 }
936
937 if ((strict_strtoul(argv[num_raid_params], 10, &num_raid_devs) < 0) ||
938 (num_raid_devs >= INT_MAX)) {
939 ti->error = "Cannot understand number of raid devices";
940 return -EINVAL;
941 }
942
943 rs = context_alloc(ti, rt, (unsigned)num_raid_devs);
944 if (IS_ERR(rs))
945 return PTR_ERR(rs);
946
947 ret = parse_raid_params(rs, argv, (unsigned)num_raid_params);
948 if (ret)
949 goto bad;
950
951 ret = -EINVAL;
952
953 argc -= num_raid_params + 1; /* +1: we already have num_raid_devs */
954 argv += num_raid_params + 1;
955
956 if (argc != (num_raid_devs * 2)) {
957 ti->error = "Supplied RAID devices does not match the count given";
958 goto bad;
959 }
960
961 ret = dev_parms(rs, argv);
962 if (ret)
963 goto bad;
964
965 rs->md.sync_super = super_sync;
966 ret = analyse_superblocks(ti, rs);
967 if (ret)
968 goto bad;
969
970 INIT_WORK(&rs->md.event_work, do_table_event);
971 ti->private = rs;
972
973 mutex_lock(&rs->md.reconfig_mutex);
974 ret = md_run(&rs->md);
975 rs->md.in_sync = 0; /* Assume already marked dirty */
976 mutex_unlock(&rs->md.reconfig_mutex);
977
978 if (ret) {
979 ti->error = "Fail to run raid array";
980 goto bad;
981 }
982
983 rs->callbacks.congested_fn = raid_is_congested;
984 dm_table_add_target_callbacks(ti->table, &rs->callbacks);
985
986 mddev_suspend(&rs->md);
987 return 0;
988
989bad:
990 context_free(rs);
991
992 return ret;
993}
994
995static void raid_dtr(struct dm_target *ti)
996{
997 struct raid_set *rs = ti->private;
998
999 list_del_init(&rs->callbacks.list);
1000 md_stop(&rs->md);
1001 context_free(rs);
1002}
1003
1004static int raid_map(struct dm_target *ti, struct bio *bio, union map_info *map_context)
1005{
1006 struct raid_set *rs = ti->private;
1007 mddev_t *mddev = &rs->md;
1008
1009 mddev->pers->make_request(mddev, bio);
1010
1011 return DM_MAPIO_SUBMITTED;
1012}
1013
1014static int raid_status(struct dm_target *ti, status_type_t type,
1015 char *result, unsigned maxlen)
1016{
1017 struct raid_set *rs = ti->private;
1018 unsigned raid_param_cnt = 1; /* at least 1 for chunksize */
1019 unsigned sz = 0;
1020 int i;
1021 sector_t sync;
1022
1023 switch (type) {
1024 case STATUSTYPE_INFO:
1025 DMEMIT("%s %d ", rs->raid_type->name, rs->md.raid_disks);
1026
1027 for (i = 0; i < rs->md.raid_disks; i++) {
1028 if (test_bit(Faulty, &rs->dev[i].rdev.flags))
1029 DMEMIT("D");
1030 else if (test_bit(In_sync, &rs->dev[i].rdev.flags))
1031 DMEMIT("A");
1032 else
1033 DMEMIT("a");
1034 }
1035
1036 if (test_bit(MD_RECOVERY_RUNNING, &rs->md.recovery))
1037 sync = rs->md.curr_resync_completed;
1038 else
1039 sync = rs->md.recovery_cp;
1040
1041 if (sync > rs->md.resync_max_sectors)
1042 sync = rs->md.resync_max_sectors;
1043
1044 DMEMIT(" %llu/%llu",
1045 (unsigned long long) sync,
1046 (unsigned long long) rs->md.resync_max_sectors);
1047
1048 break;
1049 case STATUSTYPE_TABLE:
1050 /* The string you would use to construct this array */
1051 for (i = 0; i < rs->md.raid_disks; i++) {
1052 if ((rs->print_flags & DMPF_REBUILD) &&
1053 rs->dev[i].data_dev &&
1054 !test_bit(In_sync, &rs->dev[i].rdev.flags))
1055 raid_param_cnt += 2; /* for rebuilds */
1056 if (rs->dev[i].data_dev &&
1057 test_bit(WriteMostly, &rs->dev[i].rdev.flags))
1058 raid_param_cnt += 2;
1059 }
1060
1061 raid_param_cnt += (hweight64(rs->print_flags & ~DMPF_REBUILD) * 2);
1062 if (rs->print_flags & (DMPF_SYNC | DMPF_NOSYNC))
1063 raid_param_cnt--;
1064
1065 DMEMIT("%s %u %u", rs->raid_type->name,
1066 raid_param_cnt, rs->md.chunk_sectors);
1067
1068 if ((rs->print_flags & DMPF_SYNC) &&
1069 (rs->md.recovery_cp == MaxSector))
1070 DMEMIT(" sync");
1071 if (rs->print_flags & DMPF_NOSYNC)
1072 DMEMIT(" nosync");
1073
1074 for (i = 0; i < rs->md.raid_disks; i++)
1075 if ((rs->print_flags & DMPF_REBUILD) &&
1076 rs->dev[i].data_dev &&
1077 !test_bit(In_sync, &rs->dev[i].rdev.flags))
1078 DMEMIT(" rebuild %u", i);
1079
1080 if (rs->print_flags & DMPF_DAEMON_SLEEP)
1081 DMEMIT(" daemon_sleep %lu",
1082 rs->md.bitmap_info.daemon_sleep);
1083
1084 if (rs->print_flags & DMPF_MIN_RECOVERY_RATE)
1085 DMEMIT(" min_recovery_rate %d", rs->md.sync_speed_min);
1086
1087 if (rs->print_flags & DMPF_MAX_RECOVERY_RATE)
1088 DMEMIT(" max_recovery_rate %d", rs->md.sync_speed_max);
1089
1090 for (i = 0; i < rs->md.raid_disks; i++)
1091 if (rs->dev[i].data_dev &&
1092 test_bit(WriteMostly, &rs->dev[i].rdev.flags))
1093 DMEMIT(" write_mostly %u", i);
1094
1095 if (rs->print_flags & DMPF_MAX_WRITE_BEHIND)
1096 DMEMIT(" max_write_behind %lu",
1097 rs->md.bitmap_info.max_write_behind);
1098
1099 if (rs->print_flags & DMPF_STRIPE_CACHE) {
1100 raid5_conf_t *conf = rs->md.private;
1101
1102 /* convert from kiB to sectors */
1103 DMEMIT(" stripe_cache %d",
1104 conf ? conf->max_nr_stripes * 2 : 0);
1105 }
1106
1107 if (rs->print_flags & DMPF_REGION_SIZE)
1108 DMEMIT(" region_size %lu",
1109 rs->md.bitmap_info.chunksize >> 9);
1110
1111 DMEMIT(" %d", rs->md.raid_disks);
1112 for (i = 0; i < rs->md.raid_disks; i++) {
1113 if (rs->dev[i].meta_dev)
1114 DMEMIT(" %s", rs->dev[i].meta_dev->name);
1115 else
1116 DMEMIT(" -");
1117
1118 if (rs->dev[i].data_dev)
1119 DMEMIT(" %s", rs->dev[i].data_dev->name);
1120 else
1121 DMEMIT(" -");
1122 }
1123 }
1124
1125 return 0;
1126}
1127
1128static int raid_iterate_devices(struct dm_target *ti, iterate_devices_callout_fn fn, void *data)
1129{
1130 struct raid_set *rs = ti->private;
1131 unsigned i;
1132 int ret = 0;
1133
1134 for (i = 0; !ret && i < rs->md.raid_disks; i++)
1135 if (rs->dev[i].data_dev)
1136 ret = fn(ti,
1137 rs->dev[i].data_dev,
1138 0, /* No offset on data devs */
1139 rs->md.dev_sectors,
1140 data);
1141
1142 return ret;
1143}
1144
1145static void raid_io_hints(struct dm_target *ti, struct queue_limits *limits)
1146{
1147 struct raid_set *rs = ti->private;
1148 unsigned chunk_size = rs->md.chunk_sectors << 9;
1149 raid5_conf_t *conf = rs->md.private;
1150
1151 blk_limits_io_min(limits, chunk_size);
1152 blk_limits_io_opt(limits, chunk_size * (conf->raid_disks - conf->max_degraded));
1153}
1154
1155static void raid_presuspend(struct dm_target *ti)
1156{
1157 struct raid_set *rs = ti->private;
1158
1159 md_stop_writes(&rs->md);
1160}
1161
1162static void raid_postsuspend(struct dm_target *ti)
1163{
1164 struct raid_set *rs = ti->private;
1165
1166 mddev_suspend(&rs->md);
1167}
1168
1169static void raid_resume(struct dm_target *ti)
1170{
1171 struct raid_set *rs = ti->private;
1172
1173 bitmap_load(&rs->md);
1174 mddev_resume(&rs->md);
1175}
1176
1177static struct target_type raid_target = {
1178 .name = "raid",
1179 .version = {1, 1, 0},
1180 .module = THIS_MODULE,
1181 .ctr = raid_ctr,
1182 .dtr = raid_dtr,
1183 .map = raid_map,
1184 .status = raid_status,
1185 .iterate_devices = raid_iterate_devices,
1186 .io_hints = raid_io_hints,
1187 .presuspend = raid_presuspend,
1188 .postsuspend = raid_postsuspend,
1189 .resume = raid_resume,
1190};
1191
1192static int __init dm_raid_init(void)
1193{
1194 return dm_register_target(&raid_target);
1195}
1196
1197static void __exit dm_raid_exit(void)
1198{
1199 dm_unregister_target(&raid_target);
1200}
1201
1202module_init(dm_raid_init);
1203module_exit(dm_raid_exit);
1204
1205MODULE_DESCRIPTION(DM_NAME " raid4/5/6 target");
1206MODULE_ALIAS("dm-raid4");
1207MODULE_ALIAS("dm-raid5");
1208MODULE_ALIAS("dm-raid6");
1209MODULE_AUTHOR("Neil Brown <dm-devel@redhat.com>");
1210MODULE_LICENSE("GPL");
1/*
2 * Copyright (C) 2010-2011 Neil Brown
3 * Copyright (C) 2010-2018 Red Hat, Inc. All rights reserved.
4 *
5 * This file is released under the GPL.
6 */
7
8#include <linux/slab.h>
9#include <linux/module.h>
10
11#include "md.h"
12#include "raid1.h"
13#include "raid5.h"
14#include "raid10.h"
15#include "md-bitmap.h"
16
17#include <linux/device-mapper.h>
18
19#define DM_MSG_PREFIX "raid"
20#define MAX_RAID_DEVICES 253 /* md-raid kernel limit */
21
22/*
23 * Minimum sectors of free reshape space per raid device
24 */
25#define MIN_FREE_RESHAPE_SPACE to_sector(4*4096)
26
27/*
28 * Minimum journal space 4 MiB in sectors.
29 */
30#define MIN_RAID456_JOURNAL_SPACE (4*2048)
31
32static bool devices_handle_discard_safely = false;
33
34/*
35 * The following flags are used by dm-raid.c to set up the array state.
36 * They must be cleared before md_run is called.
37 */
38#define FirstUse 10 /* rdev flag */
39
40struct raid_dev {
41 /*
42 * Two DM devices, one to hold metadata and one to hold the
43 * actual data/parity. The reason for this is to not confuse
44 * ti->len and give more flexibility in altering size and
45 * characteristics.
46 *
47 * While it is possible for this device to be associated
48 * with a different physical device than the data_dev, it
49 * is intended for it to be the same.
50 * |--------- Physical Device ---------|
51 * |- meta_dev -|------ data_dev ------|
52 */
53 struct dm_dev *meta_dev;
54 struct dm_dev *data_dev;
55 struct md_rdev rdev;
56};
57
58/*
59 * Bits for establishing rs->ctr_flags
60 *
61 * 1 = no flag value
62 * 2 = flag with value
63 */
64#define __CTR_FLAG_SYNC 0 /* 1 */ /* Not with raid0! */
65#define __CTR_FLAG_NOSYNC 1 /* 1 */ /* Not with raid0! */
66#define __CTR_FLAG_REBUILD 2 /* 2 */ /* Not with raid0! */
67#define __CTR_FLAG_DAEMON_SLEEP 3 /* 2 */ /* Not with raid0! */
68#define __CTR_FLAG_MIN_RECOVERY_RATE 4 /* 2 */ /* Not with raid0! */
69#define __CTR_FLAG_MAX_RECOVERY_RATE 5 /* 2 */ /* Not with raid0! */
70#define __CTR_FLAG_MAX_WRITE_BEHIND 6 /* 2 */ /* Only with raid1! */
71#define __CTR_FLAG_WRITE_MOSTLY 7 /* 2 */ /* Only with raid1! */
72#define __CTR_FLAG_STRIPE_CACHE 8 /* 2 */ /* Only with raid4/5/6! */
73#define __CTR_FLAG_REGION_SIZE 9 /* 2 */ /* Not with raid0! */
74#define __CTR_FLAG_RAID10_COPIES 10 /* 2 */ /* Only with raid10 */
75#define __CTR_FLAG_RAID10_FORMAT 11 /* 2 */ /* Only with raid10 */
76/* New for v1.9.0 */
77#define __CTR_FLAG_DELTA_DISKS 12 /* 2 */ /* Only with reshapable raid1/4/5/6/10! */
78#define __CTR_FLAG_DATA_OFFSET 13 /* 2 */ /* Only with reshapable raid4/5/6/10! */
79#define __CTR_FLAG_RAID10_USE_NEAR_SETS 14 /* 2 */ /* Only with raid10! */
80
81/* New for v1.10.0 */
82#define __CTR_FLAG_JOURNAL_DEV 15 /* 2 */ /* Only with raid4/5/6 (journal device)! */
83
84/* New for v1.11.1 */
85#define __CTR_FLAG_JOURNAL_MODE 16 /* 2 */ /* Only with raid4/5/6 (journal mode)! */
86
87/*
88 * Flags for rs->ctr_flags field.
89 */
90#define CTR_FLAG_SYNC (1 << __CTR_FLAG_SYNC)
91#define CTR_FLAG_NOSYNC (1 << __CTR_FLAG_NOSYNC)
92#define CTR_FLAG_REBUILD (1 << __CTR_FLAG_REBUILD)
93#define CTR_FLAG_DAEMON_SLEEP (1 << __CTR_FLAG_DAEMON_SLEEP)
94#define CTR_FLAG_MIN_RECOVERY_RATE (1 << __CTR_FLAG_MIN_RECOVERY_RATE)
95#define CTR_FLAG_MAX_RECOVERY_RATE (1 << __CTR_FLAG_MAX_RECOVERY_RATE)
96#define CTR_FLAG_MAX_WRITE_BEHIND (1 << __CTR_FLAG_MAX_WRITE_BEHIND)
97#define CTR_FLAG_WRITE_MOSTLY (1 << __CTR_FLAG_WRITE_MOSTLY)
98#define CTR_FLAG_STRIPE_CACHE (1 << __CTR_FLAG_STRIPE_CACHE)
99#define CTR_FLAG_REGION_SIZE (1 << __CTR_FLAG_REGION_SIZE)
100#define CTR_FLAG_RAID10_COPIES (1 << __CTR_FLAG_RAID10_COPIES)
101#define CTR_FLAG_RAID10_FORMAT (1 << __CTR_FLAG_RAID10_FORMAT)
102#define CTR_FLAG_DELTA_DISKS (1 << __CTR_FLAG_DELTA_DISKS)
103#define CTR_FLAG_DATA_OFFSET (1 << __CTR_FLAG_DATA_OFFSET)
104#define CTR_FLAG_RAID10_USE_NEAR_SETS (1 << __CTR_FLAG_RAID10_USE_NEAR_SETS)
105#define CTR_FLAG_JOURNAL_DEV (1 << __CTR_FLAG_JOURNAL_DEV)
106#define CTR_FLAG_JOURNAL_MODE (1 << __CTR_FLAG_JOURNAL_MODE)
107
108/*
109 * Definitions of various constructor flags to
110 * be used in checks of valid / invalid flags
111 * per raid level.
112 */
113/* Define all any sync flags */
114#define CTR_FLAGS_ANY_SYNC (CTR_FLAG_SYNC | CTR_FLAG_NOSYNC)
115
116/* Define flags for options without argument (e.g. 'nosync') */
117#define CTR_FLAG_OPTIONS_NO_ARGS (CTR_FLAGS_ANY_SYNC | \
118 CTR_FLAG_RAID10_USE_NEAR_SETS)
119
120/* Define flags for options with one argument (e.g. 'delta_disks +2') */
121#define CTR_FLAG_OPTIONS_ONE_ARG (CTR_FLAG_REBUILD | \
122 CTR_FLAG_WRITE_MOSTLY | \
123 CTR_FLAG_DAEMON_SLEEP | \
124 CTR_FLAG_MIN_RECOVERY_RATE | \
125 CTR_FLAG_MAX_RECOVERY_RATE | \
126 CTR_FLAG_MAX_WRITE_BEHIND | \
127 CTR_FLAG_STRIPE_CACHE | \
128 CTR_FLAG_REGION_SIZE | \
129 CTR_FLAG_RAID10_COPIES | \
130 CTR_FLAG_RAID10_FORMAT | \
131 CTR_FLAG_DELTA_DISKS | \
132 CTR_FLAG_DATA_OFFSET)
133
134/* Valid options definitions per raid level... */
135
136/* "raid0" does only accept data offset */
137#define RAID0_VALID_FLAGS (CTR_FLAG_DATA_OFFSET)
138
139/* "raid1" does not accept stripe cache, data offset, delta_disks or any raid10 options */
140#define RAID1_VALID_FLAGS (CTR_FLAGS_ANY_SYNC | \
141 CTR_FLAG_REBUILD | \
142 CTR_FLAG_WRITE_MOSTLY | \
143 CTR_FLAG_DAEMON_SLEEP | \
144 CTR_FLAG_MIN_RECOVERY_RATE | \
145 CTR_FLAG_MAX_RECOVERY_RATE | \
146 CTR_FLAG_MAX_WRITE_BEHIND | \
147 CTR_FLAG_REGION_SIZE | \
148 CTR_FLAG_DELTA_DISKS | \
149 CTR_FLAG_DATA_OFFSET)
150
151/* "raid10" does not accept any raid1 or stripe cache options */
152#define RAID10_VALID_FLAGS (CTR_FLAGS_ANY_SYNC | \
153 CTR_FLAG_REBUILD | \
154 CTR_FLAG_DAEMON_SLEEP | \
155 CTR_FLAG_MIN_RECOVERY_RATE | \
156 CTR_FLAG_MAX_RECOVERY_RATE | \
157 CTR_FLAG_REGION_SIZE | \
158 CTR_FLAG_RAID10_COPIES | \
159 CTR_FLAG_RAID10_FORMAT | \
160 CTR_FLAG_DELTA_DISKS | \
161 CTR_FLAG_DATA_OFFSET | \
162 CTR_FLAG_RAID10_USE_NEAR_SETS)
163
164/*
165 * "raid4/5/6" do not accept any raid1 or raid10 specific options
166 *
167 * "raid6" does not accept "nosync", because it is not guaranteed
168 * that both parity and q-syndrome are being written properly with
169 * any writes
170 */
171#define RAID45_VALID_FLAGS (CTR_FLAGS_ANY_SYNC | \
172 CTR_FLAG_REBUILD | \
173 CTR_FLAG_DAEMON_SLEEP | \
174 CTR_FLAG_MIN_RECOVERY_RATE | \
175 CTR_FLAG_MAX_RECOVERY_RATE | \
176 CTR_FLAG_STRIPE_CACHE | \
177 CTR_FLAG_REGION_SIZE | \
178 CTR_FLAG_DELTA_DISKS | \
179 CTR_FLAG_DATA_OFFSET | \
180 CTR_FLAG_JOURNAL_DEV | \
181 CTR_FLAG_JOURNAL_MODE)
182
183#define RAID6_VALID_FLAGS (CTR_FLAG_SYNC | \
184 CTR_FLAG_REBUILD | \
185 CTR_FLAG_DAEMON_SLEEP | \
186 CTR_FLAG_MIN_RECOVERY_RATE | \
187 CTR_FLAG_MAX_RECOVERY_RATE | \
188 CTR_FLAG_STRIPE_CACHE | \
189 CTR_FLAG_REGION_SIZE | \
190 CTR_FLAG_DELTA_DISKS | \
191 CTR_FLAG_DATA_OFFSET | \
192 CTR_FLAG_JOURNAL_DEV | \
193 CTR_FLAG_JOURNAL_MODE)
194/* ...valid options definitions per raid level */
195
196/*
197 * Flags for rs->runtime_flags field
198 * (RT_FLAG prefix meaning "runtime flag")
199 *
200 * These are all internal and used to define runtime state,
201 * e.g. to prevent another resume from preresume processing
202 * the raid set all over again.
203 */
204#define RT_FLAG_RS_PRERESUMED 0
205#define RT_FLAG_RS_RESUMED 1
206#define RT_FLAG_RS_BITMAP_LOADED 2
207#define RT_FLAG_UPDATE_SBS 3
208#define RT_FLAG_RESHAPE_RS 4
209#define RT_FLAG_RS_SUSPENDED 5
210#define RT_FLAG_RS_IN_SYNC 6
211#define RT_FLAG_RS_RESYNCING 7
212
213/* Array elements of 64 bit needed for rebuild/failed disk bits */
214#define DISKS_ARRAY_ELEMS ((MAX_RAID_DEVICES + (sizeof(uint64_t) * 8 - 1)) / sizeof(uint64_t) / 8)
215
216/*
217 * raid set level, layout and chunk sectors backup/restore
218 */
219struct rs_layout {
220 int new_level;
221 int new_layout;
222 int new_chunk_sectors;
223};
224
225struct raid_set {
226 struct dm_target *ti;
227
228 uint32_t stripe_cache_entries;
229 unsigned long ctr_flags;
230 unsigned long runtime_flags;
231
232 uint64_t rebuild_disks[DISKS_ARRAY_ELEMS];
233
234 int raid_disks;
235 int delta_disks;
236 int data_offset;
237 int raid10_copies;
238 int requested_bitmap_chunk_sectors;
239
240 struct mddev md;
241 struct raid_type *raid_type;
242 struct dm_target_callbacks callbacks;
243
244 /* Optional raid4/5/6 journal device */
245 struct journal_dev {
246 struct dm_dev *dev;
247 struct md_rdev rdev;
248 int mode;
249 } journal_dev;
250
251 struct raid_dev dev[0];
252};
253
254static void rs_config_backup(struct raid_set *rs, struct rs_layout *l)
255{
256 struct mddev *mddev = &rs->md;
257
258 l->new_level = mddev->new_level;
259 l->new_layout = mddev->new_layout;
260 l->new_chunk_sectors = mddev->new_chunk_sectors;
261}
262
263static void rs_config_restore(struct raid_set *rs, struct rs_layout *l)
264{
265 struct mddev *mddev = &rs->md;
266
267 mddev->new_level = l->new_level;
268 mddev->new_layout = l->new_layout;
269 mddev->new_chunk_sectors = l->new_chunk_sectors;
270}
271
272/* raid10 algorithms (i.e. formats) */
273#define ALGORITHM_RAID10_DEFAULT 0
274#define ALGORITHM_RAID10_NEAR 1
275#define ALGORITHM_RAID10_OFFSET 2
276#define ALGORITHM_RAID10_FAR 3
277
278/* Supported raid types and properties. */
279static struct raid_type {
280 const char *name; /* RAID algorithm. */
281 const char *descr; /* Descriptor text for logging. */
282 const unsigned int parity_devs; /* # of parity devices. */
283 const unsigned int minimal_devs;/* minimal # of devices in set. */
284 const unsigned int level; /* RAID level. */
285 const unsigned int algorithm; /* RAID algorithm. */
286} raid_types[] = {
287 {"raid0", "raid0 (striping)", 0, 2, 0, 0 /* NONE */},
288 {"raid1", "raid1 (mirroring)", 0, 2, 1, 0 /* NONE */},
289 {"raid10_far", "raid10 far (striped mirrors)", 0, 2, 10, ALGORITHM_RAID10_FAR},
290 {"raid10_offset", "raid10 offset (striped mirrors)", 0, 2, 10, ALGORITHM_RAID10_OFFSET},
291 {"raid10_near", "raid10 near (striped mirrors)", 0, 2, 10, ALGORITHM_RAID10_NEAR},
292 {"raid10", "raid10 (striped mirrors)", 0, 2, 10, ALGORITHM_RAID10_DEFAULT},
293 {"raid4", "raid4 (dedicated first parity disk)", 1, 2, 5, ALGORITHM_PARITY_0}, /* raid4 layout = raid5_0 */
294 {"raid5_n", "raid5 (dedicated last parity disk)", 1, 2, 5, ALGORITHM_PARITY_N},
295 {"raid5_ls", "raid5 (left symmetric)", 1, 2, 5, ALGORITHM_LEFT_SYMMETRIC},
296 {"raid5_rs", "raid5 (right symmetric)", 1, 2, 5, ALGORITHM_RIGHT_SYMMETRIC},
297 {"raid5_la", "raid5 (left asymmetric)", 1, 2, 5, ALGORITHM_LEFT_ASYMMETRIC},
298 {"raid5_ra", "raid5 (right asymmetric)", 1, 2, 5, ALGORITHM_RIGHT_ASYMMETRIC},
299 {"raid6_zr", "raid6 (zero restart)", 2, 4, 6, ALGORITHM_ROTATING_ZERO_RESTART},
300 {"raid6_nr", "raid6 (N restart)", 2, 4, 6, ALGORITHM_ROTATING_N_RESTART},
301 {"raid6_nc", "raid6 (N continue)", 2, 4, 6, ALGORITHM_ROTATING_N_CONTINUE},
302 {"raid6_n_6", "raid6 (dedicated parity/Q n/6)", 2, 4, 6, ALGORITHM_PARITY_N_6},
303 {"raid6_ls_6", "raid6 (left symmetric dedicated Q 6)", 2, 4, 6, ALGORITHM_LEFT_SYMMETRIC_6},
304 {"raid6_rs_6", "raid6 (right symmetric dedicated Q 6)", 2, 4, 6, ALGORITHM_RIGHT_SYMMETRIC_6},
305 {"raid6_la_6", "raid6 (left asymmetric dedicated Q 6)", 2, 4, 6, ALGORITHM_LEFT_ASYMMETRIC_6},
306 {"raid6_ra_6", "raid6 (right asymmetric dedicated Q 6)", 2, 4, 6, ALGORITHM_RIGHT_ASYMMETRIC_6}
307};
308
309/* True, if @v is in inclusive range [@min, @max] */
310static bool __within_range(long v, long min, long max)
311{
312 return v >= min && v <= max;
313}
314
315/* All table line arguments are defined here */
316static struct arg_name_flag {
317 const unsigned long flag;
318 const char *name;
319} __arg_name_flags[] = {
320 { CTR_FLAG_SYNC, "sync"},
321 { CTR_FLAG_NOSYNC, "nosync"},
322 { CTR_FLAG_REBUILD, "rebuild"},
323 { CTR_FLAG_DAEMON_SLEEP, "daemon_sleep"},
324 { CTR_FLAG_MIN_RECOVERY_RATE, "min_recovery_rate"},
325 { CTR_FLAG_MAX_RECOVERY_RATE, "max_recovery_rate"},
326 { CTR_FLAG_MAX_WRITE_BEHIND, "max_write_behind"},
327 { CTR_FLAG_WRITE_MOSTLY, "write_mostly"},
328 { CTR_FLAG_STRIPE_CACHE, "stripe_cache"},
329 { CTR_FLAG_REGION_SIZE, "region_size"},
330 { CTR_FLAG_RAID10_COPIES, "raid10_copies"},
331 { CTR_FLAG_RAID10_FORMAT, "raid10_format"},
332 { CTR_FLAG_DATA_OFFSET, "data_offset"},
333 { CTR_FLAG_DELTA_DISKS, "delta_disks"},
334 { CTR_FLAG_RAID10_USE_NEAR_SETS, "raid10_use_near_sets"},
335 { CTR_FLAG_JOURNAL_DEV, "journal_dev" },
336 { CTR_FLAG_JOURNAL_MODE, "journal_mode" },
337};
338
339/* Return argument name string for given @flag */
340static const char *dm_raid_arg_name_by_flag(const uint32_t flag)
341{
342 if (hweight32(flag) == 1) {
343 struct arg_name_flag *anf = __arg_name_flags + ARRAY_SIZE(__arg_name_flags);
344
345 while (anf-- > __arg_name_flags)
346 if (flag & anf->flag)
347 return anf->name;
348
349 } else
350 DMERR("%s called with more than one flag!", __func__);
351
352 return NULL;
353}
354
355/* Define correlation of raid456 journal cache modes and dm-raid target line parameters */
356static struct {
357 const int mode;
358 const char *param;
359} _raid456_journal_mode[] = {
360 { R5C_JOURNAL_MODE_WRITE_THROUGH , "writethrough" },
361 { R5C_JOURNAL_MODE_WRITE_BACK , "writeback" }
362};
363
364/* Return MD raid4/5/6 journal mode for dm @journal_mode one */
365static int dm_raid_journal_mode_to_md(const char *mode)
366{
367 int m = ARRAY_SIZE(_raid456_journal_mode);
368
369 while (m--)
370 if (!strcasecmp(mode, _raid456_journal_mode[m].param))
371 return _raid456_journal_mode[m].mode;
372
373 return -EINVAL;
374}
375
376/* Return dm-raid raid4/5/6 journal mode string for @mode */
377static const char *md_journal_mode_to_dm_raid(const int mode)
378{
379 int m = ARRAY_SIZE(_raid456_journal_mode);
380
381 while (m--)
382 if (mode == _raid456_journal_mode[m].mode)
383 return _raid456_journal_mode[m].param;
384
385 return "unknown";
386}
387
388/*
389 * Bool helpers to test for various raid levels of a raid set.
390 * It's level as reported by the superblock rather than
391 * the requested raid_type passed to the constructor.
392 */
393/* Return true, if raid set in @rs is raid0 */
394static bool rs_is_raid0(struct raid_set *rs)
395{
396 return !rs->md.level;
397}
398
399/* Return true, if raid set in @rs is raid1 */
400static bool rs_is_raid1(struct raid_set *rs)
401{
402 return rs->md.level == 1;
403}
404
405/* Return true, if raid set in @rs is raid10 */
406static bool rs_is_raid10(struct raid_set *rs)
407{
408 return rs->md.level == 10;
409}
410
411/* Return true, if raid set in @rs is level 6 */
412static bool rs_is_raid6(struct raid_set *rs)
413{
414 return rs->md.level == 6;
415}
416
417/* Return true, if raid set in @rs is level 4, 5 or 6 */
418static bool rs_is_raid456(struct raid_set *rs)
419{
420 return __within_range(rs->md.level, 4, 6);
421}
422
423/* Return true, if raid set in @rs is reshapable */
424static bool __is_raid10_far(int layout);
425static bool rs_is_reshapable(struct raid_set *rs)
426{
427 return rs_is_raid456(rs) ||
428 (rs_is_raid10(rs) && !__is_raid10_far(rs->md.new_layout));
429}
430
431/* Return true, if raid set in @rs is recovering */
432static bool rs_is_recovering(struct raid_set *rs)
433{
434 return rs->md.recovery_cp < rs->md.dev_sectors;
435}
436
437/* Return true, if raid set in @rs is reshaping */
438static bool rs_is_reshaping(struct raid_set *rs)
439{
440 return rs->md.reshape_position != MaxSector;
441}
442
443/*
444 * bool helpers to test for various raid levels of a raid type @rt
445 */
446
447/* Return true, if raid type in @rt is raid0 */
448static bool rt_is_raid0(struct raid_type *rt)
449{
450 return !rt->level;
451}
452
453/* Return true, if raid type in @rt is raid1 */
454static bool rt_is_raid1(struct raid_type *rt)
455{
456 return rt->level == 1;
457}
458
459/* Return true, if raid type in @rt is raid10 */
460static bool rt_is_raid10(struct raid_type *rt)
461{
462 return rt->level == 10;
463}
464
465/* Return true, if raid type in @rt is raid4/5 */
466static bool rt_is_raid45(struct raid_type *rt)
467{
468 return __within_range(rt->level, 4, 5);
469}
470
471/* Return true, if raid type in @rt is raid6 */
472static bool rt_is_raid6(struct raid_type *rt)
473{
474 return rt->level == 6;
475}
476
477/* Return true, if raid type in @rt is raid4/5/6 */
478static bool rt_is_raid456(struct raid_type *rt)
479{
480 return __within_range(rt->level, 4, 6);
481}
482/* END: raid level bools */
483
484/* Return valid ctr flags for the raid level of @rs */
485static unsigned long __valid_flags(struct raid_set *rs)
486{
487 if (rt_is_raid0(rs->raid_type))
488 return RAID0_VALID_FLAGS;
489 else if (rt_is_raid1(rs->raid_type))
490 return RAID1_VALID_FLAGS;
491 else if (rt_is_raid10(rs->raid_type))
492 return RAID10_VALID_FLAGS;
493 else if (rt_is_raid45(rs->raid_type))
494 return RAID45_VALID_FLAGS;
495 else if (rt_is_raid6(rs->raid_type))
496 return RAID6_VALID_FLAGS;
497
498 return 0;
499}
500
501/*
502 * Check for valid flags set on @rs
503 *
504 * Has to be called after parsing of the ctr flags!
505 */
506static int rs_check_for_valid_flags(struct raid_set *rs)
507{
508 if (rs->ctr_flags & ~__valid_flags(rs)) {
509 rs->ti->error = "Invalid flags combination";
510 return -EINVAL;
511 }
512
513 return 0;
514}
515
516/* MD raid10 bit definitions and helpers */
517#define RAID10_OFFSET (1 << 16) /* stripes with data copies area adjacent on devices */
518#define RAID10_BROCKEN_USE_FAR_SETS (1 << 17) /* Broken in raid10.c: use sets instead of whole stripe rotation */
519#define RAID10_USE_FAR_SETS (1 << 18) /* Use sets instead of whole stripe rotation */
520#define RAID10_FAR_COPIES_SHIFT 8 /* raid10 # far copies shift (2nd byte of layout) */
521
522/* Return md raid10 near copies for @layout */
523static unsigned int __raid10_near_copies(int layout)
524{
525 return layout & 0xFF;
526}
527
528/* Return md raid10 far copies for @layout */
529static unsigned int __raid10_far_copies(int layout)
530{
531 return __raid10_near_copies(layout >> RAID10_FAR_COPIES_SHIFT);
532}
533
534/* Return true if md raid10 offset for @layout */
535static bool __is_raid10_offset(int layout)
536{
537 return !!(layout & RAID10_OFFSET);
538}
539
540/* Return true if md raid10 near for @layout */
541static bool __is_raid10_near(int layout)
542{
543 return !__is_raid10_offset(layout) && __raid10_near_copies(layout) > 1;
544}
545
546/* Return true if md raid10 far for @layout */
547static bool __is_raid10_far(int layout)
548{
549 return !__is_raid10_offset(layout) && __raid10_far_copies(layout) > 1;
550}
551
552/* Return md raid10 layout string for @layout */
553static const char *raid10_md_layout_to_format(int layout)
554{
555 /*
556 * Bit 16 stands for "offset"
557 * (i.e. adjacent stripes hold copies)
558 *
559 * Refer to MD's raid10.c for details
560 */
561 if (__is_raid10_offset(layout))
562 return "offset";
563
564 if (__raid10_near_copies(layout) > 1)
565 return "near";
566
567 if (__raid10_far_copies(layout) > 1)
568 return "far";
569
570 return "unknown";
571}
572
573/* Return md raid10 algorithm for @name */
574static int raid10_name_to_format(const char *name)
575{
576 if (!strcasecmp(name, "near"))
577 return ALGORITHM_RAID10_NEAR;
578 else if (!strcasecmp(name, "offset"))
579 return ALGORITHM_RAID10_OFFSET;
580 else if (!strcasecmp(name, "far"))
581 return ALGORITHM_RAID10_FAR;
582
583 return -EINVAL;
584}
585
586/* Return md raid10 copies for @layout */
587static unsigned int raid10_md_layout_to_copies(int layout)
588{
589 return max(__raid10_near_copies(layout), __raid10_far_copies(layout));
590}
591
592/* Return md raid10 format id for @format string */
593static int raid10_format_to_md_layout(struct raid_set *rs,
594 unsigned int algorithm,
595 unsigned int copies)
596{
597 unsigned int n = 1, f = 1, r = 0;
598
599 /*
600 * MD resilienece flaw:
601 *
602 * enabling use_far_sets for far/offset formats causes copies
603 * to be colocated on the same devs together with their origins!
604 *
605 * -> disable it for now in the definition above
606 */
607 if (algorithm == ALGORITHM_RAID10_DEFAULT ||
608 algorithm == ALGORITHM_RAID10_NEAR)
609 n = copies;
610
611 else if (algorithm == ALGORITHM_RAID10_OFFSET) {
612 f = copies;
613 r = RAID10_OFFSET;
614 if (!test_bit(__CTR_FLAG_RAID10_USE_NEAR_SETS, &rs->ctr_flags))
615 r |= RAID10_USE_FAR_SETS;
616
617 } else if (algorithm == ALGORITHM_RAID10_FAR) {
618 f = copies;
619 r = !RAID10_OFFSET;
620 if (!test_bit(__CTR_FLAG_RAID10_USE_NEAR_SETS, &rs->ctr_flags))
621 r |= RAID10_USE_FAR_SETS;
622
623 } else
624 return -EINVAL;
625
626 return r | (f << RAID10_FAR_COPIES_SHIFT) | n;
627}
628/* END: MD raid10 bit definitions and helpers */
629
630/* Check for any of the raid10 algorithms */
631static bool __got_raid10(struct raid_type *rtp, const int layout)
632{
633 if (rtp->level == 10) {
634 switch (rtp->algorithm) {
635 case ALGORITHM_RAID10_DEFAULT:
636 case ALGORITHM_RAID10_NEAR:
637 return __is_raid10_near(layout);
638 case ALGORITHM_RAID10_OFFSET:
639 return __is_raid10_offset(layout);
640 case ALGORITHM_RAID10_FAR:
641 return __is_raid10_far(layout);
642 default:
643 break;
644 }
645 }
646
647 return false;
648}
649
650/* Return raid_type for @name */
651static struct raid_type *get_raid_type(const char *name)
652{
653 struct raid_type *rtp = raid_types + ARRAY_SIZE(raid_types);
654
655 while (rtp-- > raid_types)
656 if (!strcasecmp(rtp->name, name))
657 return rtp;
658
659 return NULL;
660}
661
662/* Return raid_type for @name based derived from @level and @layout */
663static struct raid_type *get_raid_type_by_ll(const int level, const int layout)
664{
665 struct raid_type *rtp = raid_types + ARRAY_SIZE(raid_types);
666
667 while (rtp-- > raid_types) {
668 /* RAID10 special checks based on @layout flags/properties */
669 if (rtp->level == level &&
670 (__got_raid10(rtp, layout) || rtp->algorithm == layout))
671 return rtp;
672 }
673
674 return NULL;
675}
676
677/* Adjust rdev sectors */
678static void rs_set_rdev_sectors(struct raid_set *rs)
679{
680 struct mddev *mddev = &rs->md;
681 struct md_rdev *rdev;
682
683 /*
684 * raid10 sets rdev->sector to the device size, which
685 * is unintended in case of out-of-place reshaping
686 */
687 rdev_for_each(rdev, mddev)
688 if (!test_bit(Journal, &rdev->flags))
689 rdev->sectors = mddev->dev_sectors;
690}
691
692/*
693 * Change bdev capacity of @rs in case of a disk add/remove reshape
694 */
695static void rs_set_capacity(struct raid_set *rs)
696{
697 struct gendisk *gendisk = dm_disk(dm_table_get_md(rs->ti->table));
698
699 set_capacity(gendisk, rs->md.array_sectors);
700 revalidate_disk(gendisk);
701}
702
703/*
704 * Set the mddev properties in @rs to the current
705 * ones retrieved from the freshest superblock
706 */
707static void rs_set_cur(struct raid_set *rs)
708{
709 struct mddev *mddev = &rs->md;
710
711 mddev->new_level = mddev->level;
712 mddev->new_layout = mddev->layout;
713 mddev->new_chunk_sectors = mddev->chunk_sectors;
714}
715
716/*
717 * Set the mddev properties in @rs to the new
718 * ones requested by the ctr
719 */
720static void rs_set_new(struct raid_set *rs)
721{
722 struct mddev *mddev = &rs->md;
723
724 mddev->level = mddev->new_level;
725 mddev->layout = mddev->new_layout;
726 mddev->chunk_sectors = mddev->new_chunk_sectors;
727 mddev->raid_disks = rs->raid_disks;
728 mddev->delta_disks = 0;
729}
730
731static struct raid_set *raid_set_alloc(struct dm_target *ti, struct raid_type *raid_type,
732 unsigned int raid_devs)
733{
734 unsigned int i;
735 struct raid_set *rs;
736
737 if (raid_devs <= raid_type->parity_devs) {
738 ti->error = "Insufficient number of devices";
739 return ERR_PTR(-EINVAL);
740 }
741
742 rs = kzalloc(struct_size(rs, dev, raid_devs), GFP_KERNEL);
743 if (!rs) {
744 ti->error = "Cannot allocate raid context";
745 return ERR_PTR(-ENOMEM);
746 }
747
748 mddev_init(&rs->md);
749
750 rs->raid_disks = raid_devs;
751 rs->delta_disks = 0;
752
753 rs->ti = ti;
754 rs->raid_type = raid_type;
755 rs->stripe_cache_entries = 256;
756 rs->md.raid_disks = raid_devs;
757 rs->md.level = raid_type->level;
758 rs->md.new_level = rs->md.level;
759 rs->md.layout = raid_type->algorithm;
760 rs->md.new_layout = rs->md.layout;
761 rs->md.delta_disks = 0;
762 rs->md.recovery_cp = MaxSector;
763
764 for (i = 0; i < raid_devs; i++)
765 md_rdev_init(&rs->dev[i].rdev);
766
767 /*
768 * Remaining items to be initialized by further RAID params:
769 * rs->md.persistent
770 * rs->md.external
771 * rs->md.chunk_sectors
772 * rs->md.new_chunk_sectors
773 * rs->md.dev_sectors
774 */
775
776 return rs;
777}
778
779/* Free all @rs allocations */
780static void raid_set_free(struct raid_set *rs)
781{
782 int i;
783
784 if (rs->journal_dev.dev) {
785 md_rdev_clear(&rs->journal_dev.rdev);
786 dm_put_device(rs->ti, rs->journal_dev.dev);
787 }
788
789 for (i = 0; i < rs->raid_disks; i++) {
790 if (rs->dev[i].meta_dev)
791 dm_put_device(rs->ti, rs->dev[i].meta_dev);
792 md_rdev_clear(&rs->dev[i].rdev);
793 if (rs->dev[i].data_dev)
794 dm_put_device(rs->ti, rs->dev[i].data_dev);
795 }
796
797 kfree(rs);
798}
799
800/*
801 * For every device we have two words
802 * <meta_dev>: meta device name or '-' if missing
803 * <data_dev>: data device name or '-' if missing
804 *
805 * The following are permitted:
806 * - -
807 * - <data_dev>
808 * <meta_dev> <data_dev>
809 *
810 * The following is not allowed:
811 * <meta_dev> -
812 *
813 * This code parses those words. If there is a failure,
814 * the caller must use raid_set_free() to unwind the operations.
815 */
816static int parse_dev_params(struct raid_set *rs, struct dm_arg_set *as)
817{
818 int i;
819 int rebuild = 0;
820 int metadata_available = 0;
821 int r = 0;
822 const char *arg;
823
824 /* Put off the number of raid devices argument to get to dev pairs */
825 arg = dm_shift_arg(as);
826 if (!arg)
827 return -EINVAL;
828
829 for (i = 0; i < rs->raid_disks; i++) {
830 rs->dev[i].rdev.raid_disk = i;
831
832 rs->dev[i].meta_dev = NULL;
833 rs->dev[i].data_dev = NULL;
834
835 /*
836 * There are no offsets initially.
837 * Out of place reshape will set them accordingly.
838 */
839 rs->dev[i].rdev.data_offset = 0;
840 rs->dev[i].rdev.new_data_offset = 0;
841 rs->dev[i].rdev.mddev = &rs->md;
842
843 arg = dm_shift_arg(as);
844 if (!arg)
845 return -EINVAL;
846
847 if (strcmp(arg, "-")) {
848 r = dm_get_device(rs->ti, arg, dm_table_get_mode(rs->ti->table),
849 &rs->dev[i].meta_dev);
850 if (r) {
851 rs->ti->error = "RAID metadata device lookup failure";
852 return r;
853 }
854
855 rs->dev[i].rdev.sb_page = alloc_page(GFP_KERNEL);
856 if (!rs->dev[i].rdev.sb_page) {
857 rs->ti->error = "Failed to allocate superblock page";
858 return -ENOMEM;
859 }
860 }
861
862 arg = dm_shift_arg(as);
863 if (!arg)
864 return -EINVAL;
865
866 if (!strcmp(arg, "-")) {
867 if (!test_bit(In_sync, &rs->dev[i].rdev.flags) &&
868 (!rs->dev[i].rdev.recovery_offset)) {
869 rs->ti->error = "Drive designated for rebuild not specified";
870 return -EINVAL;
871 }
872
873 if (rs->dev[i].meta_dev) {
874 rs->ti->error = "No data device supplied with metadata device";
875 return -EINVAL;
876 }
877
878 continue;
879 }
880
881 r = dm_get_device(rs->ti, arg, dm_table_get_mode(rs->ti->table),
882 &rs->dev[i].data_dev);
883 if (r) {
884 rs->ti->error = "RAID device lookup failure";
885 return r;
886 }
887
888 if (rs->dev[i].meta_dev) {
889 metadata_available = 1;
890 rs->dev[i].rdev.meta_bdev = rs->dev[i].meta_dev->bdev;
891 }
892 rs->dev[i].rdev.bdev = rs->dev[i].data_dev->bdev;
893 list_add_tail(&rs->dev[i].rdev.same_set, &rs->md.disks);
894 if (!test_bit(In_sync, &rs->dev[i].rdev.flags))
895 rebuild++;
896 }
897
898 if (rs->journal_dev.dev)
899 list_add_tail(&rs->journal_dev.rdev.same_set, &rs->md.disks);
900
901 if (metadata_available) {
902 rs->md.external = 0;
903 rs->md.persistent = 1;
904 rs->md.major_version = 2;
905 } else if (rebuild && !rs->md.recovery_cp) {
906 /*
907 * Without metadata, we will not be able to tell if the array
908 * is in-sync or not - we must assume it is not. Therefore,
909 * it is impossible to rebuild a drive.
910 *
911 * Even if there is metadata, the on-disk information may
912 * indicate that the array is not in-sync and it will then
913 * fail at that time.
914 *
915 * User could specify 'nosync' option if desperate.
916 */
917 rs->ti->error = "Unable to rebuild drive while array is not in-sync";
918 return -EINVAL;
919 }
920
921 return 0;
922}
923
924/*
925 * validate_region_size
926 * @rs
927 * @region_size: region size in sectors. If 0, pick a size (4MiB default).
928 *
929 * Set rs->md.bitmap_info.chunksize (which really refers to 'region size').
930 * Ensure that (ti->len/region_size < 2^21) - required by MD bitmap.
931 *
932 * Returns: 0 on success, -EINVAL on failure.
933 */
934static int validate_region_size(struct raid_set *rs, unsigned long region_size)
935{
936 unsigned long min_region_size = rs->ti->len / (1 << 21);
937
938 if (rs_is_raid0(rs))
939 return 0;
940
941 if (!region_size) {
942 /*
943 * Choose a reasonable default. All figures in sectors.
944 */
945 if (min_region_size > (1 << 13)) {
946 /* If not a power of 2, make it the next power of 2 */
947 region_size = roundup_pow_of_two(min_region_size);
948 DMINFO("Choosing default region size of %lu sectors",
949 region_size);
950 } else {
951 DMINFO("Choosing default region size of 4MiB");
952 region_size = 1 << 13; /* sectors */
953 }
954 } else {
955 /*
956 * Validate user-supplied value.
957 */
958 if (region_size > rs->ti->len) {
959 rs->ti->error = "Supplied region size is too large";
960 return -EINVAL;
961 }
962
963 if (region_size < min_region_size) {
964 DMERR("Supplied region_size (%lu sectors) below minimum (%lu)",
965 region_size, min_region_size);
966 rs->ti->error = "Supplied region size is too small";
967 return -EINVAL;
968 }
969
970 if (!is_power_of_2(region_size)) {
971 rs->ti->error = "Region size is not a power of 2";
972 return -EINVAL;
973 }
974
975 if (region_size < rs->md.chunk_sectors) {
976 rs->ti->error = "Region size is smaller than the chunk size";
977 return -EINVAL;
978 }
979 }
980
981 /*
982 * Convert sectors to bytes.
983 */
984 rs->md.bitmap_info.chunksize = to_bytes(region_size);
985
986 return 0;
987}
988
989/*
990 * validate_raid_redundancy
991 * @rs
992 *
993 * Determine if there are enough devices in the array that haven't
994 * failed (or are being rebuilt) to form a usable array.
995 *
996 * Returns: 0 on success, -EINVAL on failure.
997 */
998static int validate_raid_redundancy(struct raid_set *rs)
999{
1000 unsigned int i, rebuild_cnt = 0;
1001 unsigned int rebuilds_per_group = 0, copies;
1002 unsigned int group_size, last_group_start;
1003
1004 for (i = 0; i < rs->md.raid_disks; i++)
1005 if (!test_bit(In_sync, &rs->dev[i].rdev.flags) ||
1006 !rs->dev[i].rdev.sb_page)
1007 rebuild_cnt++;
1008
1009 switch (rs->md.level) {
1010 case 0:
1011 break;
1012 case 1:
1013 if (rebuild_cnt >= rs->md.raid_disks)
1014 goto too_many;
1015 break;
1016 case 4:
1017 case 5:
1018 case 6:
1019 if (rebuild_cnt > rs->raid_type->parity_devs)
1020 goto too_many;
1021 break;
1022 case 10:
1023 copies = raid10_md_layout_to_copies(rs->md.new_layout);
1024 if (copies < 2) {
1025 DMERR("Bogus raid10 data copies < 2!");
1026 return -EINVAL;
1027 }
1028
1029 if (rebuild_cnt < copies)
1030 break;
1031
1032 /*
1033 * It is possible to have a higher rebuild count for RAID10,
1034 * as long as the failed devices occur in different mirror
1035 * groups (i.e. different stripes).
1036 *
1037 * When checking "near" format, make sure no adjacent devices
1038 * have failed beyond what can be handled. In addition to the
1039 * simple case where the number of devices is a multiple of the
1040 * number of copies, we must also handle cases where the number
1041 * of devices is not a multiple of the number of copies.
1042 * E.g. dev1 dev2 dev3 dev4 dev5
1043 * A A B B C
1044 * C D D E E
1045 */
1046 if (__is_raid10_near(rs->md.new_layout)) {
1047 for (i = 0; i < rs->md.raid_disks; i++) {
1048 if (!(i % copies))
1049 rebuilds_per_group = 0;
1050 if ((!rs->dev[i].rdev.sb_page ||
1051 !test_bit(In_sync, &rs->dev[i].rdev.flags)) &&
1052 (++rebuilds_per_group >= copies))
1053 goto too_many;
1054 }
1055 break;
1056 }
1057
1058 /*
1059 * When checking "far" and "offset" formats, we need to ensure
1060 * that the device that holds its copy is not also dead or
1061 * being rebuilt. (Note that "far" and "offset" formats only
1062 * support two copies right now. These formats also only ever
1063 * use the 'use_far_sets' variant.)
1064 *
1065 * This check is somewhat complicated by the need to account
1066 * for arrays that are not a multiple of (far) copies. This
1067 * results in the need to treat the last (potentially larger)
1068 * set differently.
1069 */
1070 group_size = (rs->md.raid_disks / copies);
1071 last_group_start = (rs->md.raid_disks / group_size) - 1;
1072 last_group_start *= group_size;
1073 for (i = 0; i < rs->md.raid_disks; i++) {
1074 if (!(i % copies) && !(i > last_group_start))
1075 rebuilds_per_group = 0;
1076 if ((!rs->dev[i].rdev.sb_page ||
1077 !test_bit(In_sync, &rs->dev[i].rdev.flags)) &&
1078 (++rebuilds_per_group >= copies))
1079 goto too_many;
1080 }
1081 break;
1082 default:
1083 if (rebuild_cnt)
1084 return -EINVAL;
1085 }
1086
1087 return 0;
1088
1089too_many:
1090 return -EINVAL;
1091}
1092
1093/*
1094 * Possible arguments are...
1095 * <chunk_size> [optional_args]
1096 *
1097 * Argument definitions
1098 * <chunk_size> The number of sectors per disk that
1099 * will form the "stripe"
1100 * [[no]sync] Force or prevent recovery of the
1101 * entire array
1102 * [rebuild <idx>] Rebuild the drive indicated by the index
1103 * [daemon_sleep <ms>] Time between bitmap daemon work to
1104 * clear bits
1105 * [min_recovery_rate <kB/sec/disk>] Throttle RAID initialization
1106 * [max_recovery_rate <kB/sec/disk>] Throttle RAID initialization
1107 * [write_mostly <idx>] Indicate a write mostly drive via index
1108 * [max_write_behind <sectors>] See '-write-behind=' (man mdadm)
1109 * [stripe_cache <sectors>] Stripe cache size for higher RAIDs
1110 * [region_size <sectors>] Defines granularity of bitmap
1111 * [journal_dev <dev>] raid4/5/6 journaling deviice
1112 * (i.e. write hole closing log)
1113 *
1114 * RAID10-only options:
1115 * [raid10_copies <# copies>] Number of copies. (Default: 2)
1116 * [raid10_format <near|far|offset>] Layout algorithm. (Default: near)
1117 */
1118static int parse_raid_params(struct raid_set *rs, struct dm_arg_set *as,
1119 unsigned int num_raid_params)
1120{
1121 int value, raid10_format = ALGORITHM_RAID10_DEFAULT;
1122 unsigned int raid10_copies = 2;
1123 unsigned int i, write_mostly = 0;
1124 unsigned int region_size = 0;
1125 sector_t max_io_len;
1126 const char *arg, *key;
1127 struct raid_dev *rd;
1128 struct raid_type *rt = rs->raid_type;
1129
1130 arg = dm_shift_arg(as);
1131 num_raid_params--; /* Account for chunk_size argument */
1132
1133 if (kstrtoint(arg, 10, &value) < 0) {
1134 rs->ti->error = "Bad numerical argument given for chunk_size";
1135 return -EINVAL;
1136 }
1137
1138 /*
1139 * First, parse the in-order required arguments
1140 * "chunk_size" is the only argument of this type.
1141 */
1142 if (rt_is_raid1(rt)) {
1143 if (value)
1144 DMERR("Ignoring chunk size parameter for RAID 1");
1145 value = 0;
1146 } else if (!is_power_of_2(value)) {
1147 rs->ti->error = "Chunk size must be a power of 2";
1148 return -EINVAL;
1149 } else if (value < 8) {
1150 rs->ti->error = "Chunk size value is too small";
1151 return -EINVAL;
1152 }
1153
1154 rs->md.new_chunk_sectors = rs->md.chunk_sectors = value;
1155
1156 /*
1157 * We set each individual device as In_sync with a completed
1158 * 'recovery_offset'. If there has been a device failure or
1159 * replacement then one of the following cases applies:
1160 *
1161 * 1) User specifies 'rebuild'.
1162 * - Device is reset when param is read.
1163 * 2) A new device is supplied.
1164 * - No matching superblock found, resets device.
1165 * 3) Device failure was transient and returns on reload.
1166 * - Failure noticed, resets device for bitmap replay.
1167 * 4) Device hadn't completed recovery after previous failure.
1168 * - Superblock is read and overrides recovery_offset.
1169 *
1170 * What is found in the superblocks of the devices is always
1171 * authoritative, unless 'rebuild' or '[no]sync' was specified.
1172 */
1173 for (i = 0; i < rs->raid_disks; i++) {
1174 set_bit(In_sync, &rs->dev[i].rdev.flags);
1175 rs->dev[i].rdev.recovery_offset = MaxSector;
1176 }
1177
1178 /*
1179 * Second, parse the unordered optional arguments
1180 */
1181 for (i = 0; i < num_raid_params; i++) {
1182 key = dm_shift_arg(as);
1183 if (!key) {
1184 rs->ti->error = "Not enough raid parameters given";
1185 return -EINVAL;
1186 }
1187
1188 if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_NOSYNC))) {
1189 if (test_and_set_bit(__CTR_FLAG_NOSYNC, &rs->ctr_flags)) {
1190 rs->ti->error = "Only one 'nosync' argument allowed";
1191 return -EINVAL;
1192 }
1193 continue;
1194 }
1195 if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_SYNC))) {
1196 if (test_and_set_bit(__CTR_FLAG_SYNC, &rs->ctr_flags)) {
1197 rs->ti->error = "Only one 'sync' argument allowed";
1198 return -EINVAL;
1199 }
1200 continue;
1201 }
1202 if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_RAID10_USE_NEAR_SETS))) {
1203 if (test_and_set_bit(__CTR_FLAG_RAID10_USE_NEAR_SETS, &rs->ctr_flags)) {
1204 rs->ti->error = "Only one 'raid10_use_new_sets' argument allowed";
1205 return -EINVAL;
1206 }
1207 continue;
1208 }
1209
1210 arg = dm_shift_arg(as);
1211 i++; /* Account for the argument pairs */
1212 if (!arg) {
1213 rs->ti->error = "Wrong number of raid parameters given";
1214 return -EINVAL;
1215 }
1216
1217 /*
1218 * Parameters that take a string value are checked here.
1219 */
1220 /* "raid10_format {near|offset|far} */
1221 if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_RAID10_FORMAT))) {
1222 if (test_and_set_bit(__CTR_FLAG_RAID10_FORMAT, &rs->ctr_flags)) {
1223 rs->ti->error = "Only one 'raid10_format' argument pair allowed";
1224 return -EINVAL;
1225 }
1226 if (!rt_is_raid10(rt)) {
1227 rs->ti->error = "'raid10_format' is an invalid parameter for this RAID type";
1228 return -EINVAL;
1229 }
1230 raid10_format = raid10_name_to_format(arg);
1231 if (raid10_format < 0) {
1232 rs->ti->error = "Invalid 'raid10_format' value given";
1233 return raid10_format;
1234 }
1235 continue;
1236 }
1237
1238 /* "journal_dev <dev>" */
1239 if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_JOURNAL_DEV))) {
1240 int r;
1241 struct md_rdev *jdev;
1242
1243 if (test_and_set_bit(__CTR_FLAG_JOURNAL_DEV, &rs->ctr_flags)) {
1244 rs->ti->error = "Only one raid4/5/6 set journaling device allowed";
1245 return -EINVAL;
1246 }
1247 if (!rt_is_raid456(rt)) {
1248 rs->ti->error = "'journal_dev' is an invalid parameter for this RAID type";
1249 return -EINVAL;
1250 }
1251 r = dm_get_device(rs->ti, arg, dm_table_get_mode(rs->ti->table),
1252 &rs->journal_dev.dev);
1253 if (r) {
1254 rs->ti->error = "raid4/5/6 journal device lookup failure";
1255 return r;
1256 }
1257 jdev = &rs->journal_dev.rdev;
1258 md_rdev_init(jdev);
1259 jdev->mddev = &rs->md;
1260 jdev->bdev = rs->journal_dev.dev->bdev;
1261 jdev->sectors = to_sector(i_size_read(jdev->bdev->bd_inode));
1262 if (jdev->sectors < MIN_RAID456_JOURNAL_SPACE) {
1263 rs->ti->error = "No space for raid4/5/6 journal";
1264 return -ENOSPC;
1265 }
1266 rs->journal_dev.mode = R5C_JOURNAL_MODE_WRITE_THROUGH;
1267 set_bit(Journal, &jdev->flags);
1268 continue;
1269 }
1270
1271 /* "journal_mode <mode>" ("journal_dev" mandatory!) */
1272 if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_JOURNAL_MODE))) {
1273 int r;
1274
1275 if (!test_bit(__CTR_FLAG_JOURNAL_DEV, &rs->ctr_flags)) {
1276 rs->ti->error = "raid4/5/6 'journal_mode' is invalid without 'journal_dev'";
1277 return -EINVAL;
1278 }
1279 if (test_and_set_bit(__CTR_FLAG_JOURNAL_MODE, &rs->ctr_flags)) {
1280 rs->ti->error = "Only one raid4/5/6 'journal_mode' argument allowed";
1281 return -EINVAL;
1282 }
1283 r = dm_raid_journal_mode_to_md(arg);
1284 if (r < 0) {
1285 rs->ti->error = "Invalid 'journal_mode' argument";
1286 return r;
1287 }
1288 rs->journal_dev.mode = r;
1289 continue;
1290 }
1291
1292 /*
1293 * Parameters with number values from here on.
1294 */
1295 if (kstrtoint(arg, 10, &value) < 0) {
1296 rs->ti->error = "Bad numerical argument given in raid params";
1297 return -EINVAL;
1298 }
1299
1300 if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_REBUILD))) {
1301 /*
1302 * "rebuild" is being passed in by userspace to provide
1303 * indexes of replaced devices and to set up additional
1304 * devices on raid level takeover.
1305 */
1306 if (!__within_range(value, 0, rs->raid_disks - 1)) {
1307 rs->ti->error = "Invalid rebuild index given";
1308 return -EINVAL;
1309 }
1310
1311 if (test_and_set_bit(value, (void *) rs->rebuild_disks)) {
1312 rs->ti->error = "rebuild for this index already given";
1313 return -EINVAL;
1314 }
1315
1316 rd = rs->dev + value;
1317 clear_bit(In_sync, &rd->rdev.flags);
1318 clear_bit(Faulty, &rd->rdev.flags);
1319 rd->rdev.recovery_offset = 0;
1320 set_bit(__CTR_FLAG_REBUILD, &rs->ctr_flags);
1321 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_WRITE_MOSTLY))) {
1322 if (!rt_is_raid1(rt)) {
1323 rs->ti->error = "write_mostly option is only valid for RAID1";
1324 return -EINVAL;
1325 }
1326
1327 if (!__within_range(value, 0, rs->md.raid_disks - 1)) {
1328 rs->ti->error = "Invalid write_mostly index given";
1329 return -EINVAL;
1330 }
1331
1332 write_mostly++;
1333 set_bit(WriteMostly, &rs->dev[value].rdev.flags);
1334 set_bit(__CTR_FLAG_WRITE_MOSTLY, &rs->ctr_flags);
1335 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_MAX_WRITE_BEHIND))) {
1336 if (!rt_is_raid1(rt)) {
1337 rs->ti->error = "max_write_behind option is only valid for RAID1";
1338 return -EINVAL;
1339 }
1340
1341 if (test_and_set_bit(__CTR_FLAG_MAX_WRITE_BEHIND, &rs->ctr_flags)) {
1342 rs->ti->error = "Only one max_write_behind argument pair allowed";
1343 return -EINVAL;
1344 }
1345
1346 /*
1347 * In device-mapper, we specify things in sectors, but
1348 * MD records this value in kB
1349 */
1350 if (value < 0 || value / 2 > COUNTER_MAX) {
1351 rs->ti->error = "Max write-behind limit out of range";
1352 return -EINVAL;
1353 }
1354
1355 rs->md.bitmap_info.max_write_behind = value / 2;
1356 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_DAEMON_SLEEP))) {
1357 if (test_and_set_bit(__CTR_FLAG_DAEMON_SLEEP, &rs->ctr_flags)) {
1358 rs->ti->error = "Only one daemon_sleep argument pair allowed";
1359 return -EINVAL;
1360 }
1361 if (value < 0) {
1362 rs->ti->error = "daemon sleep period out of range";
1363 return -EINVAL;
1364 }
1365 rs->md.bitmap_info.daemon_sleep = value;
1366 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_DATA_OFFSET))) {
1367 /* Userspace passes new data_offset after having extended the the data image LV */
1368 if (test_and_set_bit(__CTR_FLAG_DATA_OFFSET, &rs->ctr_flags)) {
1369 rs->ti->error = "Only one data_offset argument pair allowed";
1370 return -EINVAL;
1371 }
1372 /* Ensure sensible data offset */
1373 if (value < 0 ||
1374 (value && (value < MIN_FREE_RESHAPE_SPACE || value % to_sector(PAGE_SIZE)))) {
1375 rs->ti->error = "Bogus data_offset value";
1376 return -EINVAL;
1377 }
1378 rs->data_offset = value;
1379 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_DELTA_DISKS))) {
1380 /* Define the +/-# of disks to add to/remove from the given raid set */
1381 if (test_and_set_bit(__CTR_FLAG_DELTA_DISKS, &rs->ctr_flags)) {
1382 rs->ti->error = "Only one delta_disks argument pair allowed";
1383 return -EINVAL;
1384 }
1385 /* Ensure MAX_RAID_DEVICES and raid type minimal_devs! */
1386 if (!__within_range(abs(value), 1, MAX_RAID_DEVICES - rt->minimal_devs)) {
1387 rs->ti->error = "Too many delta_disk requested";
1388 return -EINVAL;
1389 }
1390
1391 rs->delta_disks = value;
1392 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_STRIPE_CACHE))) {
1393 if (test_and_set_bit(__CTR_FLAG_STRIPE_CACHE, &rs->ctr_flags)) {
1394 rs->ti->error = "Only one stripe_cache argument pair allowed";
1395 return -EINVAL;
1396 }
1397
1398 if (!rt_is_raid456(rt)) {
1399 rs->ti->error = "Inappropriate argument: stripe_cache";
1400 return -EINVAL;
1401 }
1402
1403 if (value < 0) {
1404 rs->ti->error = "Bogus stripe cache entries value";
1405 return -EINVAL;
1406 }
1407 rs->stripe_cache_entries = value;
1408 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_MIN_RECOVERY_RATE))) {
1409 if (test_and_set_bit(__CTR_FLAG_MIN_RECOVERY_RATE, &rs->ctr_flags)) {
1410 rs->ti->error = "Only one min_recovery_rate argument pair allowed";
1411 return -EINVAL;
1412 }
1413
1414 if (value < 0) {
1415 rs->ti->error = "min_recovery_rate out of range";
1416 return -EINVAL;
1417 }
1418 rs->md.sync_speed_min = value;
1419 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_MAX_RECOVERY_RATE))) {
1420 if (test_and_set_bit(__CTR_FLAG_MAX_RECOVERY_RATE, &rs->ctr_flags)) {
1421 rs->ti->error = "Only one max_recovery_rate argument pair allowed";
1422 return -EINVAL;
1423 }
1424
1425 if (value < 0) {
1426 rs->ti->error = "max_recovery_rate out of range";
1427 return -EINVAL;
1428 }
1429 rs->md.sync_speed_max = value;
1430 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_REGION_SIZE))) {
1431 if (test_and_set_bit(__CTR_FLAG_REGION_SIZE, &rs->ctr_flags)) {
1432 rs->ti->error = "Only one region_size argument pair allowed";
1433 return -EINVAL;
1434 }
1435
1436 region_size = value;
1437 rs->requested_bitmap_chunk_sectors = value;
1438 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_RAID10_COPIES))) {
1439 if (test_and_set_bit(__CTR_FLAG_RAID10_COPIES, &rs->ctr_flags)) {
1440 rs->ti->error = "Only one raid10_copies argument pair allowed";
1441 return -EINVAL;
1442 }
1443
1444 if (!__within_range(value, 2, rs->md.raid_disks)) {
1445 rs->ti->error = "Bad value for 'raid10_copies'";
1446 return -EINVAL;
1447 }
1448
1449 raid10_copies = value;
1450 } else {
1451 DMERR("Unable to parse RAID parameter: %s", key);
1452 rs->ti->error = "Unable to parse RAID parameter";
1453 return -EINVAL;
1454 }
1455 }
1456
1457 if (test_bit(__CTR_FLAG_SYNC, &rs->ctr_flags) &&
1458 test_bit(__CTR_FLAG_NOSYNC, &rs->ctr_flags)) {
1459 rs->ti->error = "sync and nosync are mutually exclusive";
1460 return -EINVAL;
1461 }
1462
1463 if (test_bit(__CTR_FLAG_REBUILD, &rs->ctr_flags) &&
1464 (test_bit(__CTR_FLAG_SYNC, &rs->ctr_flags) ||
1465 test_bit(__CTR_FLAG_NOSYNC, &rs->ctr_flags))) {
1466 rs->ti->error = "sync/nosync and rebuild are mutually exclusive";
1467 return -EINVAL;
1468 }
1469
1470 if (write_mostly >= rs->md.raid_disks) {
1471 rs->ti->error = "Can't set all raid1 devices to write_mostly";
1472 return -EINVAL;
1473 }
1474
1475 if (rs->md.sync_speed_max &&
1476 rs->md.sync_speed_min > rs->md.sync_speed_max) {
1477 rs->ti->error = "Bogus recovery rates";
1478 return -EINVAL;
1479 }
1480
1481 if (validate_region_size(rs, region_size))
1482 return -EINVAL;
1483
1484 if (rs->md.chunk_sectors)
1485 max_io_len = rs->md.chunk_sectors;
1486 else
1487 max_io_len = region_size;
1488
1489 if (dm_set_target_max_io_len(rs->ti, max_io_len))
1490 return -EINVAL;
1491
1492 if (rt_is_raid10(rt)) {
1493 if (raid10_copies > rs->md.raid_disks) {
1494 rs->ti->error = "Not enough devices to satisfy specification";
1495 return -EINVAL;
1496 }
1497
1498 rs->md.new_layout = raid10_format_to_md_layout(rs, raid10_format, raid10_copies);
1499 if (rs->md.new_layout < 0) {
1500 rs->ti->error = "Error getting raid10 format";
1501 return rs->md.new_layout;
1502 }
1503
1504 rt = get_raid_type_by_ll(10, rs->md.new_layout);
1505 if (!rt) {
1506 rs->ti->error = "Failed to recognize new raid10 layout";
1507 return -EINVAL;
1508 }
1509
1510 if ((rt->algorithm == ALGORITHM_RAID10_DEFAULT ||
1511 rt->algorithm == ALGORITHM_RAID10_NEAR) &&
1512 test_bit(__CTR_FLAG_RAID10_USE_NEAR_SETS, &rs->ctr_flags)) {
1513 rs->ti->error = "RAID10 format 'near' and 'raid10_use_near_sets' are incompatible";
1514 return -EINVAL;
1515 }
1516 }
1517
1518 rs->raid10_copies = raid10_copies;
1519
1520 /* Assume there are no metadata devices until the drives are parsed */
1521 rs->md.persistent = 0;
1522 rs->md.external = 1;
1523
1524 /* Check, if any invalid ctr arguments have been passed in for the raid level */
1525 return rs_check_for_valid_flags(rs);
1526}
1527
1528/* Set raid4/5/6 cache size */
1529static int rs_set_raid456_stripe_cache(struct raid_set *rs)
1530{
1531 int r;
1532 struct r5conf *conf;
1533 struct mddev *mddev = &rs->md;
1534 uint32_t min_stripes = max(mddev->chunk_sectors, mddev->new_chunk_sectors) / 2;
1535 uint32_t nr_stripes = rs->stripe_cache_entries;
1536
1537 if (!rt_is_raid456(rs->raid_type)) {
1538 rs->ti->error = "Inappropriate raid level; cannot change stripe_cache size";
1539 return -EINVAL;
1540 }
1541
1542 if (nr_stripes < min_stripes) {
1543 DMINFO("Adjusting requested %u stripe cache entries to %u to suit stripe size",
1544 nr_stripes, min_stripes);
1545 nr_stripes = min_stripes;
1546 }
1547
1548 conf = mddev->private;
1549 if (!conf) {
1550 rs->ti->error = "Cannot change stripe_cache size on inactive RAID set";
1551 return -EINVAL;
1552 }
1553
1554 /* Try setting number of stripes in raid456 stripe cache */
1555 if (conf->min_nr_stripes != nr_stripes) {
1556 r = raid5_set_cache_size(mddev, nr_stripes);
1557 if (r) {
1558 rs->ti->error = "Failed to set raid4/5/6 stripe cache size";
1559 return r;
1560 }
1561
1562 DMINFO("%u stripe cache entries", nr_stripes);
1563 }
1564
1565 return 0;
1566}
1567
1568/* Return # of data stripes as kept in mddev as of @rs (i.e. as of superblock) */
1569static unsigned int mddev_data_stripes(struct raid_set *rs)
1570{
1571 return rs->md.raid_disks - rs->raid_type->parity_devs;
1572}
1573
1574/* Return # of data stripes of @rs (i.e. as of ctr) */
1575static unsigned int rs_data_stripes(struct raid_set *rs)
1576{
1577 return rs->raid_disks - rs->raid_type->parity_devs;
1578}
1579
1580/*
1581 * Retrieve rdev->sectors from any valid raid device of @rs
1582 * to allow userpace to pass in arbitray "- -" device tupples.
1583 */
1584static sector_t __rdev_sectors(struct raid_set *rs)
1585{
1586 int i;
1587
1588 for (i = 0; i < rs->md.raid_disks; i++) {
1589 struct md_rdev *rdev = &rs->dev[i].rdev;
1590
1591 if (!test_bit(Journal, &rdev->flags) &&
1592 rdev->bdev && rdev->sectors)
1593 return rdev->sectors;
1594 }
1595
1596 return 0;
1597}
1598
1599/* Check that calculated dev_sectors fits all component devices. */
1600static int _check_data_dev_sectors(struct raid_set *rs)
1601{
1602 sector_t ds = ~0;
1603 struct md_rdev *rdev;
1604
1605 rdev_for_each(rdev, &rs->md)
1606 if (!test_bit(Journal, &rdev->flags) && rdev->bdev) {
1607 ds = min(ds, to_sector(i_size_read(rdev->bdev->bd_inode)));
1608 if (ds < rs->md.dev_sectors) {
1609 rs->ti->error = "Component device(s) too small";
1610 return -EINVAL;
1611 }
1612 }
1613
1614 return 0;
1615}
1616
1617/* Calculate the sectors per device and per array used for @rs */
1618static int rs_set_dev_and_array_sectors(struct raid_set *rs, bool use_mddev)
1619{
1620 int delta_disks;
1621 unsigned int data_stripes;
1622 struct mddev *mddev = &rs->md;
1623 struct md_rdev *rdev;
1624 sector_t array_sectors = rs->ti->len, dev_sectors = rs->ti->len;
1625
1626 if (use_mddev) {
1627 delta_disks = mddev->delta_disks;
1628 data_stripes = mddev_data_stripes(rs);
1629 } else {
1630 delta_disks = rs->delta_disks;
1631 data_stripes = rs_data_stripes(rs);
1632 }
1633
1634 /* Special raid1 case w/o delta_disks support (yet) */
1635 if (rt_is_raid1(rs->raid_type))
1636 ;
1637 else if (rt_is_raid10(rs->raid_type)) {
1638 if (rs->raid10_copies < 2 ||
1639 delta_disks < 0) {
1640 rs->ti->error = "Bogus raid10 data copies or delta disks";
1641 return -EINVAL;
1642 }
1643
1644 dev_sectors *= rs->raid10_copies;
1645 if (sector_div(dev_sectors, data_stripes))
1646 goto bad;
1647
1648 array_sectors = (data_stripes + delta_disks) * dev_sectors;
1649 if (sector_div(array_sectors, rs->raid10_copies))
1650 goto bad;
1651
1652 } else if (sector_div(dev_sectors, data_stripes))
1653 goto bad;
1654
1655 else
1656 /* Striped layouts */
1657 array_sectors = (data_stripes + delta_disks) * dev_sectors;
1658
1659 rdev_for_each(rdev, mddev)
1660 if (!test_bit(Journal, &rdev->flags))
1661 rdev->sectors = dev_sectors;
1662
1663 mddev->array_sectors = array_sectors;
1664 mddev->dev_sectors = dev_sectors;
1665
1666 return _check_data_dev_sectors(rs);
1667bad:
1668 rs->ti->error = "Target length not divisible by number of data devices";
1669 return -EINVAL;
1670}
1671
1672/* Setup recovery on @rs */
1673static void __rs_setup_recovery(struct raid_set *rs, sector_t dev_sectors)
1674{
1675 /* raid0 does not recover */
1676 if (rs_is_raid0(rs))
1677 rs->md.recovery_cp = MaxSector;
1678 /*
1679 * A raid6 set has to be recovered either
1680 * completely or for the grown part to
1681 * ensure proper parity and Q-Syndrome
1682 */
1683 else if (rs_is_raid6(rs))
1684 rs->md.recovery_cp = dev_sectors;
1685 /*
1686 * Other raid set types may skip recovery
1687 * depending on the 'nosync' flag.
1688 */
1689 else
1690 rs->md.recovery_cp = test_bit(__CTR_FLAG_NOSYNC, &rs->ctr_flags)
1691 ? MaxSector : dev_sectors;
1692}
1693
1694/* Setup recovery on @rs based on raid type, device size and 'nosync' flag */
1695static void rs_setup_recovery(struct raid_set *rs, sector_t dev_sectors)
1696{
1697 if (!dev_sectors)
1698 /* New raid set or 'sync' flag provided */
1699 __rs_setup_recovery(rs, 0);
1700 else if (dev_sectors == MaxSector)
1701 /* Prevent recovery */
1702 __rs_setup_recovery(rs, MaxSector);
1703 else if (__rdev_sectors(rs) < dev_sectors)
1704 /* Grown raid set */
1705 __rs_setup_recovery(rs, __rdev_sectors(rs));
1706 else
1707 __rs_setup_recovery(rs, MaxSector);
1708}
1709
1710static void do_table_event(struct work_struct *ws)
1711{
1712 struct raid_set *rs = container_of(ws, struct raid_set, md.event_work);
1713
1714 smp_rmb(); /* Make sure we access most actual mddev properties */
1715 if (!rs_is_reshaping(rs)) {
1716 if (rs_is_raid10(rs))
1717 rs_set_rdev_sectors(rs);
1718 rs_set_capacity(rs);
1719 }
1720 dm_table_event(rs->ti->table);
1721}
1722
1723static int raid_is_congested(struct dm_target_callbacks *cb, int bits)
1724{
1725 struct raid_set *rs = container_of(cb, struct raid_set, callbacks);
1726
1727 return mddev_congested(&rs->md, bits);
1728}
1729
1730/*
1731 * Make sure a valid takover (level switch) is being requested on @rs
1732 *
1733 * Conversions of raid sets from one MD personality to another
1734 * have to conform to restrictions which are enforced here.
1735 */
1736static int rs_check_takeover(struct raid_set *rs)
1737{
1738 struct mddev *mddev = &rs->md;
1739 unsigned int near_copies;
1740
1741 if (rs->md.degraded) {
1742 rs->ti->error = "Can't takeover degraded raid set";
1743 return -EPERM;
1744 }
1745
1746 if (rs_is_reshaping(rs)) {
1747 rs->ti->error = "Can't takeover reshaping raid set";
1748 return -EPERM;
1749 }
1750
1751 switch (mddev->level) {
1752 case 0:
1753 /* raid0 -> raid1/5 with one disk */
1754 if ((mddev->new_level == 1 || mddev->new_level == 5) &&
1755 mddev->raid_disks == 1)
1756 return 0;
1757
1758 /* raid0 -> raid10 */
1759 if (mddev->new_level == 10 &&
1760 !(rs->raid_disks % mddev->raid_disks))
1761 return 0;
1762
1763 /* raid0 with multiple disks -> raid4/5/6 */
1764 if (__within_range(mddev->new_level, 4, 6) &&
1765 mddev->new_layout == ALGORITHM_PARITY_N &&
1766 mddev->raid_disks > 1)
1767 return 0;
1768
1769 break;
1770
1771 case 10:
1772 /* Can't takeover raid10_offset! */
1773 if (__is_raid10_offset(mddev->layout))
1774 break;
1775
1776 near_copies = __raid10_near_copies(mddev->layout);
1777
1778 /* raid10* -> raid0 */
1779 if (mddev->new_level == 0) {
1780 /* Can takeover raid10_near with raid disks divisable by data copies! */
1781 if (near_copies > 1 &&
1782 !(mddev->raid_disks % near_copies)) {
1783 mddev->raid_disks /= near_copies;
1784 mddev->delta_disks = mddev->raid_disks;
1785 return 0;
1786 }
1787
1788 /* Can takeover raid10_far */
1789 if (near_copies == 1 &&
1790 __raid10_far_copies(mddev->layout) > 1)
1791 return 0;
1792
1793 break;
1794 }
1795
1796 /* raid10_{near,far} -> raid1 */
1797 if (mddev->new_level == 1 &&
1798 max(near_copies, __raid10_far_copies(mddev->layout)) == mddev->raid_disks)
1799 return 0;
1800
1801 /* raid10_{near,far} with 2 disks -> raid4/5 */
1802 if (__within_range(mddev->new_level, 4, 5) &&
1803 mddev->raid_disks == 2)
1804 return 0;
1805 break;
1806
1807 case 1:
1808 /* raid1 with 2 disks -> raid4/5 */
1809 if (__within_range(mddev->new_level, 4, 5) &&
1810 mddev->raid_disks == 2) {
1811 mddev->degraded = 1;
1812 return 0;
1813 }
1814
1815 /* raid1 -> raid0 */
1816 if (mddev->new_level == 0 &&
1817 mddev->raid_disks == 1)
1818 return 0;
1819
1820 /* raid1 -> raid10 */
1821 if (mddev->new_level == 10)
1822 return 0;
1823 break;
1824
1825 case 4:
1826 /* raid4 -> raid0 */
1827 if (mddev->new_level == 0)
1828 return 0;
1829
1830 /* raid4 -> raid1/5 with 2 disks */
1831 if ((mddev->new_level == 1 || mddev->new_level == 5) &&
1832 mddev->raid_disks == 2)
1833 return 0;
1834
1835 /* raid4 -> raid5/6 with parity N */
1836 if (__within_range(mddev->new_level, 5, 6) &&
1837 mddev->layout == ALGORITHM_PARITY_N)
1838 return 0;
1839 break;
1840
1841 case 5:
1842 /* raid5 with parity N -> raid0 */
1843 if (mddev->new_level == 0 &&
1844 mddev->layout == ALGORITHM_PARITY_N)
1845 return 0;
1846
1847 /* raid5 with parity N -> raid4 */
1848 if (mddev->new_level == 4 &&
1849 mddev->layout == ALGORITHM_PARITY_N)
1850 return 0;
1851
1852 /* raid5 with 2 disks -> raid1/4/10 */
1853 if ((mddev->new_level == 1 || mddev->new_level == 4 || mddev->new_level == 10) &&
1854 mddev->raid_disks == 2)
1855 return 0;
1856
1857 /* raid5_* -> raid6_*_6 with Q-Syndrome N (e.g. raid5_ra -> raid6_ra_6 */
1858 if (mddev->new_level == 6 &&
1859 ((mddev->layout == ALGORITHM_PARITY_N && mddev->new_layout == ALGORITHM_PARITY_N) ||
1860 __within_range(mddev->new_layout, ALGORITHM_LEFT_ASYMMETRIC_6, ALGORITHM_RIGHT_SYMMETRIC_6)))
1861 return 0;
1862 break;
1863
1864 case 6:
1865 /* raid6 with parity N -> raid0 */
1866 if (mddev->new_level == 0 &&
1867 mddev->layout == ALGORITHM_PARITY_N)
1868 return 0;
1869
1870 /* raid6 with parity N -> raid4 */
1871 if (mddev->new_level == 4 &&
1872 mddev->layout == ALGORITHM_PARITY_N)
1873 return 0;
1874
1875 /* raid6_*_n with Q-Syndrome N -> raid5_* */
1876 if (mddev->new_level == 5 &&
1877 ((mddev->layout == ALGORITHM_PARITY_N && mddev->new_layout == ALGORITHM_PARITY_N) ||
1878 __within_range(mddev->new_layout, ALGORITHM_LEFT_ASYMMETRIC, ALGORITHM_RIGHT_SYMMETRIC)))
1879 return 0;
1880
1881 default:
1882 break;
1883 }
1884
1885 rs->ti->error = "takeover not possible";
1886 return -EINVAL;
1887}
1888
1889/* True if @rs requested to be taken over */
1890static bool rs_takeover_requested(struct raid_set *rs)
1891{
1892 return rs->md.new_level != rs->md.level;
1893}
1894
1895/* True if @rs is requested to reshape by ctr */
1896static bool rs_reshape_requested(struct raid_set *rs)
1897{
1898 bool change;
1899 struct mddev *mddev = &rs->md;
1900
1901 if (rs_takeover_requested(rs))
1902 return false;
1903
1904 if (rs_is_raid0(rs))
1905 return false;
1906
1907 change = mddev->new_layout != mddev->layout ||
1908 mddev->new_chunk_sectors != mddev->chunk_sectors ||
1909 rs->delta_disks;
1910
1911 /* Historical case to support raid1 reshape without delta disks */
1912 if (rs_is_raid1(rs)) {
1913 if (rs->delta_disks)
1914 return !!rs->delta_disks;
1915
1916 return !change &&
1917 mddev->raid_disks != rs->raid_disks;
1918 }
1919
1920 if (rs_is_raid10(rs))
1921 return change &&
1922 !__is_raid10_far(mddev->new_layout) &&
1923 rs->delta_disks >= 0;
1924
1925 return change;
1926}
1927
1928/* Features */
1929#define FEATURE_FLAG_SUPPORTS_V190 0x1 /* Supports extended superblock */
1930
1931/* State flags for sb->flags */
1932#define SB_FLAG_RESHAPE_ACTIVE 0x1
1933#define SB_FLAG_RESHAPE_BACKWARDS 0x2
1934
1935/*
1936 * This structure is never routinely used by userspace, unlike md superblocks.
1937 * Devices with this superblock should only ever be accessed via device-mapper.
1938 */
1939#define DM_RAID_MAGIC 0x64526D44
1940struct dm_raid_superblock {
1941 __le32 magic; /* "DmRd" */
1942 __le32 compat_features; /* Used to indicate compatible features (like 1.9.0 ondisk metadata extension) */
1943
1944 __le32 num_devices; /* Number of devices in this raid set. (Max 64) */
1945 __le32 array_position; /* The position of this drive in the raid set */
1946
1947 __le64 events; /* Incremented by md when superblock updated */
1948 __le64 failed_devices; /* Pre 1.9.0 part of bit field of devices to */
1949 /* indicate failures (see extension below) */
1950
1951 /*
1952 * This offset tracks the progress of the repair or replacement of
1953 * an individual drive.
1954 */
1955 __le64 disk_recovery_offset;
1956
1957 /*
1958 * This offset tracks the progress of the initial raid set
1959 * synchronisation/parity calculation.
1960 */
1961 __le64 array_resync_offset;
1962
1963 /*
1964 * raid characteristics
1965 */
1966 __le32 level;
1967 __le32 layout;
1968 __le32 stripe_sectors;
1969
1970 /********************************************************************
1971 * BELOW FOLLOW V1.9.0 EXTENSIONS TO THE PRISTINE SUPERBLOCK FORMAT!!!
1972 *
1973 * FEATURE_FLAG_SUPPORTS_V190 in the compat_features member indicates that those exist
1974 */
1975
1976 __le32 flags; /* Flags defining array states for reshaping */
1977
1978 /*
1979 * This offset tracks the progress of a raid
1980 * set reshape in order to be able to restart it
1981 */
1982 __le64 reshape_position;
1983
1984 /*
1985 * These define the properties of the array in case of an interrupted reshape
1986 */
1987 __le32 new_level;
1988 __le32 new_layout;
1989 __le32 new_stripe_sectors;
1990 __le32 delta_disks;
1991
1992 __le64 array_sectors; /* Array size in sectors */
1993
1994 /*
1995 * Sector offsets to data on devices (reshaping).
1996 * Needed to support out of place reshaping, thus
1997 * not writing over any stripes whilst converting
1998 * them from old to new layout
1999 */
2000 __le64 data_offset;
2001 __le64 new_data_offset;
2002
2003 __le64 sectors; /* Used device size in sectors */
2004
2005 /*
2006 * Additonal Bit field of devices indicating failures to support
2007 * up to 256 devices with the 1.9.0 on-disk metadata format
2008 */
2009 __le64 extended_failed_devices[DISKS_ARRAY_ELEMS - 1];
2010
2011 __le32 incompat_features; /* Used to indicate any incompatible features */
2012
2013 /* Always set rest up to logical block size to 0 when writing (see get_metadata_device() below). */
2014} __packed;
2015
2016/*
2017 * Check for reshape constraints on raid set @rs:
2018 *
2019 * - reshape function non-existent
2020 * - degraded set
2021 * - ongoing recovery
2022 * - ongoing reshape
2023 *
2024 * Returns 0 if none or -EPERM if given constraint
2025 * and error message reference in @errmsg
2026 */
2027static int rs_check_reshape(struct raid_set *rs)
2028{
2029 struct mddev *mddev = &rs->md;
2030
2031 if (!mddev->pers || !mddev->pers->check_reshape)
2032 rs->ti->error = "Reshape not supported";
2033 else if (mddev->degraded)
2034 rs->ti->error = "Can't reshape degraded raid set";
2035 else if (rs_is_recovering(rs))
2036 rs->ti->error = "Convert request on recovering raid set prohibited";
2037 else if (rs_is_reshaping(rs))
2038 rs->ti->error = "raid set already reshaping!";
2039 else if (!(rs_is_raid1(rs) || rs_is_raid10(rs) || rs_is_raid456(rs)))
2040 rs->ti->error = "Reshaping only supported for raid1/4/5/6/10";
2041 else
2042 return 0;
2043
2044 return -EPERM;
2045}
2046
2047static int read_disk_sb(struct md_rdev *rdev, int size, bool force_reload)
2048{
2049 BUG_ON(!rdev->sb_page);
2050
2051 if (rdev->sb_loaded && !force_reload)
2052 return 0;
2053
2054 rdev->sb_loaded = 0;
2055
2056 if (!sync_page_io(rdev, 0, size, rdev->sb_page, REQ_OP_READ, 0, true)) {
2057 DMERR("Failed to read superblock of device at position %d",
2058 rdev->raid_disk);
2059 md_error(rdev->mddev, rdev);
2060 set_bit(Faulty, &rdev->flags);
2061 return -EIO;
2062 }
2063
2064 rdev->sb_loaded = 1;
2065
2066 return 0;
2067}
2068
2069static void sb_retrieve_failed_devices(struct dm_raid_superblock *sb, uint64_t *failed_devices)
2070{
2071 failed_devices[0] = le64_to_cpu(sb->failed_devices);
2072 memset(failed_devices + 1, 0, sizeof(sb->extended_failed_devices));
2073
2074 if (le32_to_cpu(sb->compat_features) & FEATURE_FLAG_SUPPORTS_V190) {
2075 int i = ARRAY_SIZE(sb->extended_failed_devices);
2076
2077 while (i--)
2078 failed_devices[i+1] = le64_to_cpu(sb->extended_failed_devices[i]);
2079 }
2080}
2081
2082static void sb_update_failed_devices(struct dm_raid_superblock *sb, uint64_t *failed_devices)
2083{
2084 int i = ARRAY_SIZE(sb->extended_failed_devices);
2085
2086 sb->failed_devices = cpu_to_le64(failed_devices[0]);
2087 while (i--)
2088 sb->extended_failed_devices[i] = cpu_to_le64(failed_devices[i+1]);
2089}
2090
2091/*
2092 * Synchronize the superblock members with the raid set properties
2093 *
2094 * All superblock data is little endian.
2095 */
2096static void super_sync(struct mddev *mddev, struct md_rdev *rdev)
2097{
2098 bool update_failed_devices = false;
2099 unsigned int i;
2100 uint64_t failed_devices[DISKS_ARRAY_ELEMS];
2101 struct dm_raid_superblock *sb;
2102 struct raid_set *rs = container_of(mddev, struct raid_set, md);
2103
2104 /* No metadata device, no superblock */
2105 if (!rdev->meta_bdev)
2106 return;
2107
2108 BUG_ON(!rdev->sb_page);
2109
2110 sb = page_address(rdev->sb_page);
2111
2112 sb_retrieve_failed_devices(sb, failed_devices);
2113
2114 for (i = 0; i < rs->raid_disks; i++)
2115 if (!rs->dev[i].data_dev || test_bit(Faulty, &rs->dev[i].rdev.flags)) {
2116 update_failed_devices = true;
2117 set_bit(i, (void *) failed_devices);
2118 }
2119
2120 if (update_failed_devices)
2121 sb_update_failed_devices(sb, failed_devices);
2122
2123 sb->magic = cpu_to_le32(DM_RAID_MAGIC);
2124 sb->compat_features = cpu_to_le32(FEATURE_FLAG_SUPPORTS_V190);
2125
2126 sb->num_devices = cpu_to_le32(mddev->raid_disks);
2127 sb->array_position = cpu_to_le32(rdev->raid_disk);
2128
2129 sb->events = cpu_to_le64(mddev->events);
2130
2131 sb->disk_recovery_offset = cpu_to_le64(rdev->recovery_offset);
2132 sb->array_resync_offset = cpu_to_le64(mddev->recovery_cp);
2133
2134 sb->level = cpu_to_le32(mddev->level);
2135 sb->layout = cpu_to_le32(mddev->layout);
2136 sb->stripe_sectors = cpu_to_le32(mddev->chunk_sectors);
2137
2138 /********************************************************************
2139 * BELOW FOLLOW V1.9.0 EXTENSIONS TO THE PRISTINE SUPERBLOCK FORMAT!!!
2140 *
2141 * FEATURE_FLAG_SUPPORTS_V190 in the compat_features member indicates that those exist
2142 */
2143 sb->new_level = cpu_to_le32(mddev->new_level);
2144 sb->new_layout = cpu_to_le32(mddev->new_layout);
2145 sb->new_stripe_sectors = cpu_to_le32(mddev->new_chunk_sectors);
2146
2147 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
2148
2149 smp_rmb(); /* Make sure we access most recent reshape position */
2150 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
2151 if (le64_to_cpu(sb->reshape_position) != MaxSector) {
2152 /* Flag ongoing reshape */
2153 sb->flags |= cpu_to_le32(SB_FLAG_RESHAPE_ACTIVE);
2154
2155 if (mddev->delta_disks < 0 || mddev->reshape_backwards)
2156 sb->flags |= cpu_to_le32(SB_FLAG_RESHAPE_BACKWARDS);
2157 } else {
2158 /* Clear reshape flags */
2159 sb->flags &= ~(cpu_to_le32(SB_FLAG_RESHAPE_ACTIVE|SB_FLAG_RESHAPE_BACKWARDS));
2160 }
2161
2162 sb->array_sectors = cpu_to_le64(mddev->array_sectors);
2163 sb->data_offset = cpu_to_le64(rdev->data_offset);
2164 sb->new_data_offset = cpu_to_le64(rdev->new_data_offset);
2165 sb->sectors = cpu_to_le64(rdev->sectors);
2166 sb->incompat_features = cpu_to_le32(0);
2167
2168 /* Zero out the rest of the payload after the size of the superblock */
2169 memset(sb + 1, 0, rdev->sb_size - sizeof(*sb));
2170}
2171
2172/*
2173 * super_load
2174 *
2175 * This function creates a superblock if one is not found on the device
2176 * and will decide which superblock to use if there's a choice.
2177 *
2178 * Return: 1 if use rdev, 0 if use refdev, -Exxx otherwise
2179 */
2180static int super_load(struct md_rdev *rdev, struct md_rdev *refdev)
2181{
2182 int r;
2183 struct dm_raid_superblock *sb;
2184 struct dm_raid_superblock *refsb;
2185 uint64_t events_sb, events_refsb;
2186
2187 r = read_disk_sb(rdev, rdev->sb_size, false);
2188 if (r)
2189 return r;
2190
2191 sb = page_address(rdev->sb_page);
2192
2193 /*
2194 * Two cases that we want to write new superblocks and rebuild:
2195 * 1) New device (no matching magic number)
2196 * 2) Device specified for rebuild (!In_sync w/ offset == 0)
2197 */
2198 if ((sb->magic != cpu_to_le32(DM_RAID_MAGIC)) ||
2199 (!test_bit(In_sync, &rdev->flags) && !rdev->recovery_offset)) {
2200 super_sync(rdev->mddev, rdev);
2201
2202 set_bit(FirstUse, &rdev->flags);
2203 sb->compat_features = cpu_to_le32(FEATURE_FLAG_SUPPORTS_V190);
2204
2205 /* Force writing of superblocks to disk */
2206 set_bit(MD_SB_CHANGE_DEVS, &rdev->mddev->sb_flags);
2207
2208 /* Any superblock is better than none, choose that if given */
2209 return refdev ? 0 : 1;
2210 }
2211
2212 if (!refdev)
2213 return 1;
2214
2215 events_sb = le64_to_cpu(sb->events);
2216
2217 refsb = page_address(refdev->sb_page);
2218 events_refsb = le64_to_cpu(refsb->events);
2219
2220 return (events_sb > events_refsb) ? 1 : 0;
2221}
2222
2223static int super_init_validation(struct raid_set *rs, struct md_rdev *rdev)
2224{
2225 int role;
2226 unsigned int d;
2227 struct mddev *mddev = &rs->md;
2228 uint64_t events_sb;
2229 uint64_t failed_devices[DISKS_ARRAY_ELEMS];
2230 struct dm_raid_superblock *sb;
2231 uint32_t new_devs = 0, rebuild_and_new = 0, rebuilds = 0;
2232 struct md_rdev *r;
2233 struct dm_raid_superblock *sb2;
2234
2235 sb = page_address(rdev->sb_page);
2236 events_sb = le64_to_cpu(sb->events);
2237
2238 /*
2239 * Initialise to 1 if this is a new superblock.
2240 */
2241 mddev->events = events_sb ? : 1;
2242
2243 mddev->reshape_position = MaxSector;
2244
2245 mddev->raid_disks = le32_to_cpu(sb->num_devices);
2246 mddev->level = le32_to_cpu(sb->level);
2247 mddev->layout = le32_to_cpu(sb->layout);
2248 mddev->chunk_sectors = le32_to_cpu(sb->stripe_sectors);
2249
2250 /*
2251 * Reshaping is supported, e.g. reshape_position is valid
2252 * in superblock and superblock content is authoritative.
2253 */
2254 if (le32_to_cpu(sb->compat_features) & FEATURE_FLAG_SUPPORTS_V190) {
2255 /* Superblock is authoritative wrt given raid set layout! */
2256 mddev->new_level = le32_to_cpu(sb->new_level);
2257 mddev->new_layout = le32_to_cpu(sb->new_layout);
2258 mddev->new_chunk_sectors = le32_to_cpu(sb->new_stripe_sectors);
2259 mddev->delta_disks = le32_to_cpu(sb->delta_disks);
2260 mddev->array_sectors = le64_to_cpu(sb->array_sectors);
2261
2262 /* raid was reshaping and got interrupted */
2263 if (le32_to_cpu(sb->flags) & SB_FLAG_RESHAPE_ACTIVE) {
2264 if (test_bit(__CTR_FLAG_DELTA_DISKS, &rs->ctr_flags)) {
2265 DMERR("Reshape requested but raid set is still reshaping");
2266 return -EINVAL;
2267 }
2268
2269 if (mddev->delta_disks < 0 ||
2270 (!mddev->delta_disks && (le32_to_cpu(sb->flags) & SB_FLAG_RESHAPE_BACKWARDS)))
2271 mddev->reshape_backwards = 1;
2272 else
2273 mddev->reshape_backwards = 0;
2274
2275 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
2276 rs->raid_type = get_raid_type_by_ll(mddev->level, mddev->layout);
2277 }
2278
2279 } else {
2280 /*
2281 * No takeover/reshaping, because we don't have the extended v1.9.0 metadata
2282 */
2283 struct raid_type *rt_cur = get_raid_type_by_ll(mddev->level, mddev->layout);
2284 struct raid_type *rt_new = get_raid_type_by_ll(mddev->new_level, mddev->new_layout);
2285
2286 if (rs_takeover_requested(rs)) {
2287 if (rt_cur && rt_new)
2288 DMERR("Takeover raid sets from %s to %s not yet supported by metadata. (raid level change)",
2289 rt_cur->name, rt_new->name);
2290 else
2291 DMERR("Takeover raid sets not yet supported by metadata. (raid level change)");
2292 return -EINVAL;
2293 } else if (rs_reshape_requested(rs)) {
2294 DMERR("Reshaping raid sets not yet supported by metadata. (raid layout change keeping level)");
2295 if (mddev->layout != mddev->new_layout) {
2296 if (rt_cur && rt_new)
2297 DMERR(" current layout %s vs new layout %s",
2298 rt_cur->name, rt_new->name);
2299 else
2300 DMERR(" current layout 0x%X vs new layout 0x%X",
2301 le32_to_cpu(sb->layout), mddev->new_layout);
2302 }
2303 if (mddev->chunk_sectors != mddev->new_chunk_sectors)
2304 DMERR(" current stripe sectors %u vs new stripe sectors %u",
2305 mddev->chunk_sectors, mddev->new_chunk_sectors);
2306 if (rs->delta_disks)
2307 DMERR(" current %u disks vs new %u disks",
2308 mddev->raid_disks, mddev->raid_disks + rs->delta_disks);
2309 if (rs_is_raid10(rs)) {
2310 DMERR(" Old layout: %s w/ %u copies",
2311 raid10_md_layout_to_format(mddev->layout),
2312 raid10_md_layout_to_copies(mddev->layout));
2313 DMERR(" New layout: %s w/ %u copies",
2314 raid10_md_layout_to_format(mddev->new_layout),
2315 raid10_md_layout_to_copies(mddev->new_layout));
2316 }
2317 return -EINVAL;
2318 }
2319
2320 DMINFO("Discovered old metadata format; upgrading to extended metadata format");
2321 }
2322
2323 if (!test_bit(__CTR_FLAG_NOSYNC, &rs->ctr_flags))
2324 mddev->recovery_cp = le64_to_cpu(sb->array_resync_offset);
2325
2326 /*
2327 * During load, we set FirstUse if a new superblock was written.
2328 * There are two reasons we might not have a superblock:
2329 * 1) The raid set is brand new - in which case, all of the
2330 * devices must have their In_sync bit set. Also,
2331 * recovery_cp must be 0, unless forced.
2332 * 2) This is a new device being added to an old raid set
2333 * and the new device needs to be rebuilt - in which
2334 * case the In_sync bit will /not/ be set and
2335 * recovery_cp must be MaxSector.
2336 * 3) This is/are a new device(s) being added to an old
2337 * raid set during takeover to a higher raid level
2338 * to provide capacity for redundancy or during reshape
2339 * to add capacity to grow the raid set.
2340 */
2341 d = 0;
2342 rdev_for_each(r, mddev) {
2343 if (test_bit(Journal, &rdev->flags))
2344 continue;
2345
2346 if (test_bit(FirstUse, &r->flags))
2347 new_devs++;
2348
2349 if (!test_bit(In_sync, &r->flags)) {
2350 DMINFO("Device %d specified for rebuild; clearing superblock",
2351 r->raid_disk);
2352 rebuilds++;
2353
2354 if (test_bit(FirstUse, &r->flags))
2355 rebuild_and_new++;
2356 }
2357
2358 d++;
2359 }
2360
2361 if (new_devs == rs->raid_disks || !rebuilds) {
2362 /* Replace a broken device */
2363 if (new_devs == 1 && !rs->delta_disks)
2364 ;
2365 if (new_devs == rs->raid_disks) {
2366 DMINFO("Superblocks created for new raid set");
2367 set_bit(MD_ARRAY_FIRST_USE, &mddev->flags);
2368 } else if (new_devs != rebuilds &&
2369 new_devs != rs->delta_disks) {
2370 DMERR("New device injected into existing raid set without "
2371 "'delta_disks' or 'rebuild' parameter specified");
2372 return -EINVAL;
2373 }
2374 } else if (new_devs && new_devs != rebuilds) {
2375 DMERR("%u 'rebuild' devices cannot be injected into"
2376 " a raid set with %u other first-time devices",
2377 rebuilds, new_devs);
2378 return -EINVAL;
2379 } else if (rebuilds) {
2380 if (rebuild_and_new && rebuilds != rebuild_and_new) {
2381 DMERR("new device%s provided without 'rebuild'",
2382 new_devs > 1 ? "s" : "");
2383 return -EINVAL;
2384 } else if (!test_bit(__CTR_FLAG_REBUILD, &rs->ctr_flags) && rs_is_recovering(rs)) {
2385 DMERR("'rebuild' specified while raid set is not in-sync (recovery_cp=%llu)",
2386 (unsigned long long) mddev->recovery_cp);
2387 return -EINVAL;
2388 } else if (rs_is_reshaping(rs)) {
2389 DMERR("'rebuild' specified while raid set is being reshaped (reshape_position=%llu)",
2390 (unsigned long long) mddev->reshape_position);
2391 return -EINVAL;
2392 }
2393 }
2394
2395 /*
2396 * Now we set the Faulty bit for those devices that are
2397 * recorded in the superblock as failed.
2398 */
2399 sb_retrieve_failed_devices(sb, failed_devices);
2400 rdev_for_each(r, mddev) {
2401 if (test_bit(Journal, &rdev->flags) ||
2402 !r->sb_page)
2403 continue;
2404 sb2 = page_address(r->sb_page);
2405 sb2->failed_devices = 0;
2406 memset(sb2->extended_failed_devices, 0, sizeof(sb2->extended_failed_devices));
2407
2408 /*
2409 * Check for any device re-ordering.
2410 */
2411 if (!test_bit(FirstUse, &r->flags) && (r->raid_disk >= 0)) {
2412 role = le32_to_cpu(sb2->array_position);
2413 if (role < 0)
2414 continue;
2415
2416 if (role != r->raid_disk) {
2417 if (rs_is_raid10(rs) && __is_raid10_near(mddev->layout)) {
2418 if (mddev->raid_disks % __raid10_near_copies(mddev->layout) ||
2419 rs->raid_disks % rs->raid10_copies) {
2420 rs->ti->error =
2421 "Cannot change raid10 near set to odd # of devices!";
2422 return -EINVAL;
2423 }
2424
2425 sb2->array_position = cpu_to_le32(r->raid_disk);
2426
2427 } else if (!(rs_is_raid10(rs) && rt_is_raid0(rs->raid_type)) &&
2428 !(rs_is_raid0(rs) && rt_is_raid10(rs->raid_type)) &&
2429 !rt_is_raid1(rs->raid_type)) {
2430 rs->ti->error = "Cannot change device positions in raid set";
2431 return -EINVAL;
2432 }
2433
2434 DMINFO("raid device #%d now at position #%d", role, r->raid_disk);
2435 }
2436
2437 /*
2438 * Partial recovery is performed on
2439 * returning failed devices.
2440 */
2441 if (test_bit(role, (void *) failed_devices))
2442 set_bit(Faulty, &r->flags);
2443 }
2444 }
2445
2446 return 0;
2447}
2448
2449static int super_validate(struct raid_set *rs, struct md_rdev *rdev)
2450{
2451 struct mddev *mddev = &rs->md;
2452 struct dm_raid_superblock *sb;
2453
2454 if (rs_is_raid0(rs) || !rdev->sb_page || rdev->raid_disk < 0)
2455 return 0;
2456
2457 sb = page_address(rdev->sb_page);
2458
2459 /*
2460 * If mddev->events is not set, we know we have not yet initialized
2461 * the array.
2462 */
2463 if (!mddev->events && super_init_validation(rs, rdev))
2464 return -EINVAL;
2465
2466 if (le32_to_cpu(sb->compat_features) &&
2467 le32_to_cpu(sb->compat_features) != FEATURE_FLAG_SUPPORTS_V190) {
2468 rs->ti->error = "Unable to assemble array: Unknown flag(s) in compatible feature flags";
2469 return -EINVAL;
2470 }
2471
2472 if (sb->incompat_features) {
2473 rs->ti->error = "Unable to assemble array: No incompatible feature flags supported yet";
2474 return -EINVAL;
2475 }
2476
2477 /* Enable bitmap creation for RAID levels != 0 */
2478 mddev->bitmap_info.offset = (rt_is_raid0(rs->raid_type) || rs->journal_dev.dev) ? 0 : to_sector(4096);
2479 mddev->bitmap_info.default_offset = mddev->bitmap_info.offset;
2480
2481 if (!test_and_clear_bit(FirstUse, &rdev->flags)) {
2482 /*
2483 * Retrieve rdev size stored in superblock to be prepared for shrink.
2484 * Check extended superblock members are present otherwise the size
2485 * will not be set!
2486 */
2487 if (le32_to_cpu(sb->compat_features) & FEATURE_FLAG_SUPPORTS_V190)
2488 rdev->sectors = le64_to_cpu(sb->sectors);
2489
2490 rdev->recovery_offset = le64_to_cpu(sb->disk_recovery_offset);
2491 if (rdev->recovery_offset == MaxSector)
2492 set_bit(In_sync, &rdev->flags);
2493 /*
2494 * If no reshape in progress -> we're recovering single
2495 * disk(s) and have to set the device(s) to out-of-sync
2496 */
2497 else if (!rs_is_reshaping(rs))
2498 clear_bit(In_sync, &rdev->flags); /* Mandatory for recovery */
2499 }
2500
2501 /*
2502 * If a device comes back, set it as not In_sync and no longer faulty.
2503 */
2504 if (test_and_clear_bit(Faulty, &rdev->flags)) {
2505 rdev->recovery_offset = 0;
2506 clear_bit(In_sync, &rdev->flags);
2507 rdev->saved_raid_disk = rdev->raid_disk;
2508 }
2509
2510 /* Reshape support -> restore repective data offsets */
2511 rdev->data_offset = le64_to_cpu(sb->data_offset);
2512 rdev->new_data_offset = le64_to_cpu(sb->new_data_offset);
2513
2514 return 0;
2515}
2516
2517/*
2518 * Analyse superblocks and select the freshest.
2519 */
2520static int analyse_superblocks(struct dm_target *ti, struct raid_set *rs)
2521{
2522 int r;
2523 struct md_rdev *rdev, *freshest;
2524 struct mddev *mddev = &rs->md;
2525
2526 freshest = NULL;
2527 rdev_for_each(rdev, mddev) {
2528 if (test_bit(Journal, &rdev->flags))
2529 continue;
2530
2531 if (!rdev->meta_bdev)
2532 continue;
2533
2534 /* Set superblock offset/size for metadata device. */
2535 rdev->sb_start = 0;
2536 rdev->sb_size = bdev_logical_block_size(rdev->meta_bdev);
2537 if (rdev->sb_size < sizeof(struct dm_raid_superblock) || rdev->sb_size > PAGE_SIZE) {
2538 DMERR("superblock size of a logical block is no longer valid");
2539 return -EINVAL;
2540 }
2541
2542 /*
2543 * Skipping super_load due to CTR_FLAG_SYNC will cause
2544 * the array to undergo initialization again as
2545 * though it were new. This is the intended effect
2546 * of the "sync" directive.
2547 *
2548 * With reshaping capability added, we must ensure that
2549 * that the "sync" directive is disallowed during the reshape.
2550 */
2551 if (test_bit(__CTR_FLAG_SYNC, &rs->ctr_flags))
2552 continue;
2553
2554 r = super_load(rdev, freshest);
2555
2556 switch (r) {
2557 case 1:
2558 freshest = rdev;
2559 break;
2560 case 0:
2561 break;
2562 default:
2563 /* This is a failure to read the superblock from the metadata device. */
2564 /*
2565 * We have to keep any raid0 data/metadata device pairs or
2566 * the MD raid0 personality will fail to start the array.
2567 */
2568 if (rs_is_raid0(rs))
2569 continue;
2570
2571 /*
2572 * We keep the dm_devs to be able to emit the device tuple
2573 * properly on the table line in raid_status() (rather than
2574 * mistakenly acting as if '- -' got passed into the constructor).
2575 *
2576 * The rdev has to stay on the same_set list to allow for
2577 * the attempt to restore faulty devices on second resume.
2578 */
2579 rdev->raid_disk = rdev->saved_raid_disk = -1;
2580 break;
2581 }
2582 }
2583
2584 if (!freshest)
2585 return 0;
2586
2587 /*
2588 * Validation of the freshest device provides the source of
2589 * validation for the remaining devices.
2590 */
2591 rs->ti->error = "Unable to assemble array: Invalid superblocks";
2592 if (super_validate(rs, freshest))
2593 return -EINVAL;
2594
2595 if (validate_raid_redundancy(rs)) {
2596 rs->ti->error = "Insufficient redundancy to activate array";
2597 return -EINVAL;
2598 }
2599
2600 rdev_for_each(rdev, mddev)
2601 if (!test_bit(Journal, &rdev->flags) &&
2602 rdev != freshest &&
2603 super_validate(rs, rdev))
2604 return -EINVAL;
2605 return 0;
2606}
2607
2608/*
2609 * Adjust data_offset and new_data_offset on all disk members of @rs
2610 * for out of place reshaping if requested by contructor
2611 *
2612 * We need free space at the beginning of each raid disk for forward
2613 * and at the end for backward reshapes which userspace has to provide
2614 * via remapping/reordering of space.
2615 */
2616static int rs_adjust_data_offsets(struct raid_set *rs)
2617{
2618 sector_t data_offset = 0, new_data_offset = 0;
2619 struct md_rdev *rdev;
2620
2621 /* Constructor did not request data offset change */
2622 if (!test_bit(__CTR_FLAG_DATA_OFFSET, &rs->ctr_flags)) {
2623 if (!rs_is_reshapable(rs))
2624 goto out;
2625
2626 return 0;
2627 }
2628
2629 /* HM FIXME: get In_Sync raid_dev? */
2630 rdev = &rs->dev[0].rdev;
2631
2632 if (rs->delta_disks < 0) {
2633 /*
2634 * Removing disks (reshaping backwards):
2635 *
2636 * - before reshape: data is at offset 0 and free space
2637 * is at end of each component LV
2638 *
2639 * - after reshape: data is at offset rs->data_offset != 0 on each component LV
2640 */
2641 data_offset = 0;
2642 new_data_offset = rs->data_offset;
2643
2644 } else if (rs->delta_disks > 0) {
2645 /*
2646 * Adding disks (reshaping forwards):
2647 *
2648 * - before reshape: data is at offset rs->data_offset != 0 and
2649 * free space is at begin of each component LV
2650 *
2651 * - after reshape: data is at offset 0 on each component LV
2652 */
2653 data_offset = rs->data_offset;
2654 new_data_offset = 0;
2655
2656 } else {
2657 /*
2658 * User space passes in 0 for data offset after having removed reshape space
2659 *
2660 * - or - (data offset != 0)
2661 *
2662 * Changing RAID layout or chunk size -> toggle offsets
2663 *
2664 * - before reshape: data is at offset rs->data_offset 0 and
2665 * free space is at end of each component LV
2666 * -or-
2667 * data is at offset rs->data_offset != 0 and
2668 * free space is at begin of each component LV
2669 *
2670 * - after reshape: data is at offset 0 if it was at offset != 0
2671 * or at offset != 0 if it was at offset 0
2672 * on each component LV
2673 *
2674 */
2675 data_offset = rs->data_offset ? rdev->data_offset : 0;
2676 new_data_offset = data_offset ? 0 : rs->data_offset;
2677 set_bit(RT_FLAG_UPDATE_SBS, &rs->runtime_flags);
2678 }
2679
2680 /*
2681 * Make sure we got a minimum amount of free sectors per device
2682 */
2683 if (rs->data_offset &&
2684 to_sector(i_size_read(rdev->bdev->bd_inode)) - rs->md.dev_sectors < MIN_FREE_RESHAPE_SPACE) {
2685 rs->ti->error = data_offset ? "No space for forward reshape" :
2686 "No space for backward reshape";
2687 return -ENOSPC;
2688 }
2689out:
2690 /*
2691 * Raise recovery_cp in case data_offset != 0 to
2692 * avoid false recovery positives in the constructor.
2693 */
2694 if (rs->md.recovery_cp < rs->md.dev_sectors)
2695 rs->md.recovery_cp += rs->dev[0].rdev.data_offset;
2696
2697 /* Adjust data offsets on all rdevs but on any raid4/5/6 journal device */
2698 rdev_for_each(rdev, &rs->md) {
2699 if (!test_bit(Journal, &rdev->flags)) {
2700 rdev->data_offset = data_offset;
2701 rdev->new_data_offset = new_data_offset;
2702 }
2703 }
2704
2705 return 0;
2706}
2707
2708/* Userpace reordered disks -> adjust raid_disk indexes in @rs */
2709static void __reorder_raid_disk_indexes(struct raid_set *rs)
2710{
2711 int i = 0;
2712 struct md_rdev *rdev;
2713
2714 rdev_for_each(rdev, &rs->md) {
2715 if (!test_bit(Journal, &rdev->flags)) {
2716 rdev->raid_disk = i++;
2717 rdev->saved_raid_disk = rdev->new_raid_disk = -1;
2718 }
2719 }
2720}
2721
2722/*
2723 * Setup @rs for takeover by a different raid level
2724 */
2725static int rs_setup_takeover(struct raid_set *rs)
2726{
2727 struct mddev *mddev = &rs->md;
2728 struct md_rdev *rdev;
2729 unsigned int d = mddev->raid_disks = rs->raid_disks;
2730 sector_t new_data_offset = rs->dev[0].rdev.data_offset ? 0 : rs->data_offset;
2731
2732 if (rt_is_raid10(rs->raid_type)) {
2733 if (rs_is_raid0(rs)) {
2734 /* Userpace reordered disks -> adjust raid_disk indexes */
2735 __reorder_raid_disk_indexes(rs);
2736
2737 /* raid0 -> raid10_far layout */
2738 mddev->layout = raid10_format_to_md_layout(rs, ALGORITHM_RAID10_FAR,
2739 rs->raid10_copies);
2740 } else if (rs_is_raid1(rs))
2741 /* raid1 -> raid10_near layout */
2742 mddev->layout = raid10_format_to_md_layout(rs, ALGORITHM_RAID10_NEAR,
2743 rs->raid_disks);
2744 else
2745 return -EINVAL;
2746
2747 }
2748
2749 clear_bit(MD_ARRAY_FIRST_USE, &mddev->flags);
2750 mddev->recovery_cp = MaxSector;
2751
2752 while (d--) {
2753 rdev = &rs->dev[d].rdev;
2754
2755 if (test_bit(d, (void *) rs->rebuild_disks)) {
2756 clear_bit(In_sync, &rdev->flags);
2757 clear_bit(Faulty, &rdev->flags);
2758 mddev->recovery_cp = rdev->recovery_offset = 0;
2759 /* Bitmap has to be created when we do an "up" takeover */
2760 set_bit(MD_ARRAY_FIRST_USE, &mddev->flags);
2761 }
2762
2763 rdev->new_data_offset = new_data_offset;
2764 }
2765
2766 return 0;
2767}
2768
2769/* Prepare @rs for reshape */
2770static int rs_prepare_reshape(struct raid_set *rs)
2771{
2772 bool reshape;
2773 struct mddev *mddev = &rs->md;
2774
2775 if (rs_is_raid10(rs)) {
2776 if (rs->raid_disks != mddev->raid_disks &&
2777 __is_raid10_near(mddev->layout) &&
2778 rs->raid10_copies &&
2779 rs->raid10_copies != __raid10_near_copies(mddev->layout)) {
2780 /*
2781 * raid disk have to be multiple of data copies to allow this conversion,
2782 *
2783 * This is actually not a reshape it is a
2784 * rebuild of any additional mirrors per group
2785 */
2786 if (rs->raid_disks % rs->raid10_copies) {
2787 rs->ti->error = "Can't reshape raid10 mirror groups";
2788 return -EINVAL;
2789 }
2790
2791 /* Userpace reordered disks to add/remove mirrors -> adjust raid_disk indexes */
2792 __reorder_raid_disk_indexes(rs);
2793 mddev->layout = raid10_format_to_md_layout(rs, ALGORITHM_RAID10_NEAR,
2794 rs->raid10_copies);
2795 mddev->new_layout = mddev->layout;
2796 reshape = false;
2797 } else
2798 reshape = true;
2799
2800 } else if (rs_is_raid456(rs))
2801 reshape = true;
2802
2803 else if (rs_is_raid1(rs)) {
2804 if (rs->delta_disks) {
2805 /* Process raid1 via delta_disks */
2806 mddev->degraded = rs->delta_disks < 0 ? -rs->delta_disks : rs->delta_disks;
2807 reshape = true;
2808 } else {
2809 /* Process raid1 without delta_disks */
2810 mddev->raid_disks = rs->raid_disks;
2811 reshape = false;
2812 }
2813 } else {
2814 rs->ti->error = "Called with bogus raid type";
2815 return -EINVAL;
2816 }
2817
2818 if (reshape) {
2819 set_bit(RT_FLAG_RESHAPE_RS, &rs->runtime_flags);
2820 set_bit(RT_FLAG_UPDATE_SBS, &rs->runtime_flags);
2821 } else if (mddev->raid_disks < rs->raid_disks)
2822 /* Create new superblocks and bitmaps, if any new disks */
2823 set_bit(RT_FLAG_UPDATE_SBS, &rs->runtime_flags);
2824
2825 return 0;
2826}
2827
2828/* Get reshape sectors from data_offsets or raid set */
2829static sector_t _get_reshape_sectors(struct raid_set *rs)
2830{
2831 struct md_rdev *rdev;
2832 sector_t reshape_sectors = 0;
2833
2834 rdev_for_each(rdev, &rs->md)
2835 if (!test_bit(Journal, &rdev->flags)) {
2836 reshape_sectors = (rdev->data_offset > rdev->new_data_offset) ?
2837 rdev->data_offset - rdev->new_data_offset :
2838 rdev->new_data_offset - rdev->data_offset;
2839 break;
2840 }
2841
2842 return max(reshape_sectors, (sector_t) rs->data_offset);
2843}
2844
2845/*
2846 *
2847 * - change raid layout
2848 * - change chunk size
2849 * - add disks
2850 * - remove disks
2851 */
2852static int rs_setup_reshape(struct raid_set *rs)
2853{
2854 int r = 0;
2855 unsigned int cur_raid_devs, d;
2856 sector_t reshape_sectors = _get_reshape_sectors(rs);
2857 struct mddev *mddev = &rs->md;
2858 struct md_rdev *rdev;
2859
2860 mddev->delta_disks = rs->delta_disks;
2861 cur_raid_devs = mddev->raid_disks;
2862
2863 /* Ignore impossible layout change whilst adding/removing disks */
2864 if (mddev->delta_disks &&
2865 mddev->layout != mddev->new_layout) {
2866 DMINFO("Ignoring invalid layout change with delta_disks=%d", rs->delta_disks);
2867 mddev->new_layout = mddev->layout;
2868 }
2869
2870 /*
2871 * Adjust array size:
2872 *
2873 * - in case of adding disk(s), array size has
2874 * to grow after the disk adding reshape,
2875 * which'll hapen in the event handler;
2876 * reshape will happen forward, so space has to
2877 * be available at the beginning of each disk
2878 *
2879 * - in case of removing disk(s), array size
2880 * has to shrink before starting the reshape,
2881 * which'll happen here;
2882 * reshape will happen backward, so space has to
2883 * be available at the end of each disk
2884 *
2885 * - data_offset and new_data_offset are
2886 * adjusted for aforementioned out of place
2887 * reshaping based on userspace passing in
2888 * the "data_offset <sectors>" key/value
2889 * pair via the constructor
2890 */
2891
2892 /* Add disk(s) */
2893 if (rs->delta_disks > 0) {
2894 /* Prepare disks for check in raid4/5/6/10 {check|start}_reshape */
2895 for (d = cur_raid_devs; d < rs->raid_disks; d++) {
2896 rdev = &rs->dev[d].rdev;
2897 clear_bit(In_sync, &rdev->flags);
2898
2899 /*
2900 * save_raid_disk needs to be -1, or recovery_offset will be set to 0
2901 * by md, which'll store that erroneously in the superblock on reshape
2902 */
2903 rdev->saved_raid_disk = -1;
2904 rdev->raid_disk = d;
2905
2906 rdev->sectors = mddev->dev_sectors;
2907 rdev->recovery_offset = rs_is_raid1(rs) ? 0 : MaxSector;
2908 }
2909
2910 mddev->reshape_backwards = 0; /* adding disk(s) -> forward reshape */
2911
2912 /* Remove disk(s) */
2913 } else if (rs->delta_disks < 0) {
2914 r = rs_set_dev_and_array_sectors(rs, true);
2915 mddev->reshape_backwards = 1; /* removing disk(s) -> backward reshape */
2916
2917 /* Change layout and/or chunk size */
2918 } else {
2919 /*
2920 * Reshape layout (e.g. raid5_ls -> raid5_n) and/or chunk size:
2921 *
2922 * keeping number of disks and do layout change ->
2923 *
2924 * toggle reshape_backward depending on data_offset:
2925 *
2926 * - free space upfront -> reshape forward
2927 *
2928 * - free space at the end -> reshape backward
2929 *
2930 *
2931 * This utilizes free reshape space avoiding the need
2932 * for userspace to move (parts of) LV segments in
2933 * case of layout/chunksize change (for disk
2934 * adding/removing reshape space has to be at
2935 * the proper address (see above with delta_disks):
2936 *
2937 * add disk(s) -> begin
2938 * remove disk(s)-> end
2939 */
2940 mddev->reshape_backwards = rs->dev[0].rdev.data_offset ? 0 : 1;
2941 }
2942
2943 /*
2944 * Adjust device size for forward reshape
2945 * because md_finish_reshape() reduces it.
2946 */
2947 if (!mddev->reshape_backwards)
2948 rdev_for_each(rdev, &rs->md)
2949 if (!test_bit(Journal, &rdev->flags))
2950 rdev->sectors += reshape_sectors;
2951
2952 return r;
2953}
2954
2955/*
2956 * Enable/disable discard support on RAID set depending on
2957 * RAID level and discard properties of underlying RAID members.
2958 */
2959static void configure_discard_support(struct raid_set *rs)
2960{
2961 int i;
2962 bool raid456;
2963 struct dm_target *ti = rs->ti;
2964
2965 /*
2966 * XXX: RAID level 4,5,6 require zeroing for safety.
2967 */
2968 raid456 = rs_is_raid456(rs);
2969
2970 for (i = 0; i < rs->raid_disks; i++) {
2971 struct request_queue *q;
2972
2973 if (!rs->dev[i].rdev.bdev)
2974 continue;
2975
2976 q = bdev_get_queue(rs->dev[i].rdev.bdev);
2977 if (!q || !blk_queue_discard(q))
2978 return;
2979
2980 if (raid456) {
2981 if (!devices_handle_discard_safely) {
2982 DMERR("raid456 discard support disabled due to discard_zeroes_data uncertainty.");
2983 DMERR("Set dm-raid.devices_handle_discard_safely=Y to override.");
2984 return;
2985 }
2986 }
2987 }
2988
2989 ti->num_discard_bios = 1;
2990}
2991
2992/*
2993 * Construct a RAID0/1/10/4/5/6 mapping:
2994 * Args:
2995 * <raid_type> <#raid_params> <raid_params>{0,} \
2996 * <#raid_devs> [<meta_dev1> <dev1>]{1,}
2997 *
2998 * <raid_params> varies by <raid_type>. See 'parse_raid_params' for
2999 * details on possible <raid_params>.
3000 *
3001 * Userspace is free to initialize the metadata devices, hence the superblocks to
3002 * enforce recreation based on the passed in table parameters.
3003 *
3004 */
3005static int raid_ctr(struct dm_target *ti, unsigned int argc, char **argv)
3006{
3007 int r;
3008 bool resize = false;
3009 struct raid_type *rt;
3010 unsigned int num_raid_params, num_raid_devs;
3011 sector_t calculated_dev_sectors, rdev_sectors, reshape_sectors;
3012 struct raid_set *rs = NULL;
3013 const char *arg;
3014 struct rs_layout rs_layout;
3015 struct dm_arg_set as = { argc, argv }, as_nrd;
3016 struct dm_arg _args[] = {
3017 { 0, as.argc, "Cannot understand number of raid parameters" },
3018 { 1, 254, "Cannot understand number of raid devices parameters" }
3019 };
3020
3021 /* Must have <raid_type> */
3022 arg = dm_shift_arg(&as);
3023 if (!arg) {
3024 ti->error = "No arguments";
3025 return -EINVAL;
3026 }
3027
3028 rt = get_raid_type(arg);
3029 if (!rt) {
3030 ti->error = "Unrecognised raid_type";
3031 return -EINVAL;
3032 }
3033
3034 /* Must have <#raid_params> */
3035 if (dm_read_arg_group(_args, &as, &num_raid_params, &ti->error))
3036 return -EINVAL;
3037
3038 /* number of raid device tupples <meta_dev data_dev> */
3039 as_nrd = as;
3040 dm_consume_args(&as_nrd, num_raid_params);
3041 _args[1].max = (as_nrd.argc - 1) / 2;
3042 if (dm_read_arg(_args + 1, &as_nrd, &num_raid_devs, &ti->error))
3043 return -EINVAL;
3044
3045 if (!__within_range(num_raid_devs, 1, MAX_RAID_DEVICES)) {
3046 ti->error = "Invalid number of supplied raid devices";
3047 return -EINVAL;
3048 }
3049
3050 rs = raid_set_alloc(ti, rt, num_raid_devs);
3051 if (IS_ERR(rs))
3052 return PTR_ERR(rs);
3053
3054 r = parse_raid_params(rs, &as, num_raid_params);
3055 if (r)
3056 goto bad;
3057
3058 r = parse_dev_params(rs, &as);
3059 if (r)
3060 goto bad;
3061
3062 rs->md.sync_super = super_sync;
3063
3064 /*
3065 * Calculate ctr requested array and device sizes to allow
3066 * for superblock analysis needing device sizes defined.
3067 *
3068 * Any existing superblock will overwrite the array and device sizes
3069 */
3070 r = rs_set_dev_and_array_sectors(rs, false);
3071 if (r)
3072 goto bad;
3073
3074 calculated_dev_sectors = rs->md.dev_sectors;
3075
3076 /*
3077 * Backup any new raid set level, layout, ...
3078 * requested to be able to compare to superblock
3079 * members for conversion decisions.
3080 */
3081 rs_config_backup(rs, &rs_layout);
3082
3083 r = analyse_superblocks(ti, rs);
3084 if (r)
3085 goto bad;
3086
3087 rdev_sectors = __rdev_sectors(rs);
3088 if (!rdev_sectors) {
3089 ti->error = "Invalid rdev size";
3090 r = -EINVAL;
3091 goto bad;
3092 }
3093
3094
3095 reshape_sectors = _get_reshape_sectors(rs);
3096 if (calculated_dev_sectors != rdev_sectors)
3097 resize = calculated_dev_sectors != (reshape_sectors ? rdev_sectors - reshape_sectors : rdev_sectors);
3098
3099 INIT_WORK(&rs->md.event_work, do_table_event);
3100 ti->private = rs;
3101 ti->num_flush_bios = 1;
3102
3103 /* Restore any requested new layout for conversion decision */
3104 rs_config_restore(rs, &rs_layout);
3105
3106 /*
3107 * Now that we have any superblock metadata available,
3108 * check for new, recovering, reshaping, to be taken over,
3109 * to be reshaped or an existing, unchanged raid set to
3110 * run in sequence.
3111 */
3112 if (test_bit(MD_ARRAY_FIRST_USE, &rs->md.flags)) {
3113 /* A new raid6 set has to be recovered to ensure proper parity and Q-Syndrome */
3114 if (rs_is_raid6(rs) &&
3115 test_bit(__CTR_FLAG_NOSYNC, &rs->ctr_flags)) {
3116 ti->error = "'nosync' not allowed for new raid6 set";
3117 r = -EINVAL;
3118 goto bad;
3119 }
3120 rs_setup_recovery(rs, 0);
3121 set_bit(RT_FLAG_UPDATE_SBS, &rs->runtime_flags);
3122 rs_set_new(rs);
3123 } else if (rs_is_recovering(rs)) {
3124 /* Rebuild particular devices */
3125 if (test_bit(__CTR_FLAG_REBUILD, &rs->ctr_flags)) {
3126 set_bit(RT_FLAG_UPDATE_SBS, &rs->runtime_flags);
3127 rs_setup_recovery(rs, MaxSector);
3128 }
3129 /* A recovering raid set may be resized */
3130 ; /* skip setup rs */
3131 } else if (rs_is_reshaping(rs)) {
3132 /* Have to reject size change request during reshape */
3133 if (resize) {
3134 ti->error = "Can't resize a reshaping raid set";
3135 r = -EPERM;
3136 goto bad;
3137 }
3138 /* skip setup rs */
3139 } else if (rs_takeover_requested(rs)) {
3140 if (rs_is_reshaping(rs)) {
3141 ti->error = "Can't takeover a reshaping raid set";
3142 r = -EPERM;
3143 goto bad;
3144 }
3145
3146 /* We can't takeover a journaled raid4/5/6 */
3147 if (test_bit(__CTR_FLAG_JOURNAL_DEV, &rs->ctr_flags)) {
3148 ti->error = "Can't takeover a journaled raid4/5/6 set";
3149 r = -EPERM;
3150 goto bad;
3151 }
3152
3153 /*
3154 * If a takeover is needed, userspace sets any additional
3155 * devices to rebuild and we can check for a valid request here.
3156 *
3157 * If acceptible, set the level to the new requested
3158 * one, prohibit requesting recovery, allow the raid
3159 * set to run and store superblocks during resume.
3160 */
3161 r = rs_check_takeover(rs);
3162 if (r)
3163 goto bad;
3164
3165 r = rs_setup_takeover(rs);
3166 if (r)
3167 goto bad;
3168
3169 set_bit(RT_FLAG_UPDATE_SBS, &rs->runtime_flags);
3170 /* Takeover ain't recovery, so disable recovery */
3171 rs_setup_recovery(rs, MaxSector);
3172 rs_set_new(rs);
3173 } else if (rs_reshape_requested(rs)) {
3174 /*
3175 * No need to check for 'ongoing' takeover here, because takeover
3176 * is an instant operation as oposed to an ongoing reshape.
3177 */
3178
3179 /* We can't reshape a journaled raid4/5/6 */
3180 if (test_bit(__CTR_FLAG_JOURNAL_DEV, &rs->ctr_flags)) {
3181 ti->error = "Can't reshape a journaled raid4/5/6 set";
3182 r = -EPERM;
3183 goto bad;
3184 }
3185
3186 /* Out-of-place space has to be available to allow for a reshape unless raid1! */
3187 if (reshape_sectors || rs_is_raid1(rs)) {
3188 /*
3189 * We can only prepare for a reshape here, because the
3190 * raid set needs to run to provide the repective reshape
3191 * check functions via its MD personality instance.
3192 *
3193 * So do the reshape check after md_run() succeeded.
3194 */
3195 r = rs_prepare_reshape(rs);
3196 if (r)
3197 goto bad;
3198
3199 /* Reshaping ain't recovery, so disable recovery */
3200 rs_setup_recovery(rs, MaxSector);
3201 }
3202 rs_set_cur(rs);
3203 } else {
3204 /* May not set recovery when a device rebuild is requested */
3205 if (test_bit(__CTR_FLAG_REBUILD, &rs->ctr_flags)) {
3206 rs_setup_recovery(rs, MaxSector);
3207 set_bit(RT_FLAG_UPDATE_SBS, &rs->runtime_flags);
3208 } else
3209 rs_setup_recovery(rs, test_bit(__CTR_FLAG_SYNC, &rs->ctr_flags) ?
3210 0 : (resize ? calculated_dev_sectors : MaxSector));
3211 rs_set_cur(rs);
3212 }
3213
3214 /* If constructor requested it, change data and new_data offsets */
3215 r = rs_adjust_data_offsets(rs);
3216 if (r)
3217 goto bad;
3218
3219 /* Start raid set read-only and assumed clean to change in raid_resume() */
3220 rs->md.ro = 1;
3221 rs->md.in_sync = 1;
3222
3223 /* Keep array frozen */
3224 set_bit(MD_RECOVERY_FROZEN, &rs->md.recovery);
3225
3226 /* Has to be held on running the array */
3227 mddev_lock_nointr(&rs->md);
3228 r = md_run(&rs->md);
3229 rs->md.in_sync = 0; /* Assume already marked dirty */
3230 if (r) {
3231 ti->error = "Failed to run raid array";
3232 mddev_unlock(&rs->md);
3233 goto bad;
3234 }
3235
3236 r = md_start(&rs->md);
3237
3238 if (r) {
3239 ti->error = "Failed to start raid array";
3240 mddev_unlock(&rs->md);
3241 goto bad_md_start;
3242 }
3243
3244 rs->callbacks.congested_fn = raid_is_congested;
3245 dm_table_add_target_callbacks(ti->table, &rs->callbacks);
3246
3247 /* If raid4/5/6 journal mode explicitly requested (only possible with journal dev) -> set it */
3248 if (test_bit(__CTR_FLAG_JOURNAL_MODE, &rs->ctr_flags)) {
3249 r = r5c_journal_mode_set(&rs->md, rs->journal_dev.mode);
3250 if (r) {
3251 ti->error = "Failed to set raid4/5/6 journal mode";
3252 mddev_unlock(&rs->md);
3253 goto bad_journal_mode_set;
3254 }
3255 }
3256
3257 mddev_suspend(&rs->md);
3258 set_bit(RT_FLAG_RS_SUSPENDED, &rs->runtime_flags);
3259
3260 /* Try to adjust the raid4/5/6 stripe cache size to the stripe size */
3261 if (rs_is_raid456(rs)) {
3262 r = rs_set_raid456_stripe_cache(rs);
3263 if (r)
3264 goto bad_stripe_cache;
3265 }
3266
3267 /* Now do an early reshape check */
3268 if (test_bit(RT_FLAG_RESHAPE_RS, &rs->runtime_flags)) {
3269 r = rs_check_reshape(rs);
3270 if (r)
3271 goto bad_check_reshape;
3272
3273 /* Restore new, ctr requested layout to perform check */
3274 rs_config_restore(rs, &rs_layout);
3275
3276 if (rs->md.pers->start_reshape) {
3277 r = rs->md.pers->check_reshape(&rs->md);
3278 if (r) {
3279 ti->error = "Reshape check failed";
3280 goto bad_check_reshape;
3281 }
3282 }
3283 }
3284
3285 /* Disable/enable discard support on raid set. */
3286 configure_discard_support(rs);
3287
3288 mddev_unlock(&rs->md);
3289 return 0;
3290
3291bad_md_start:
3292bad_journal_mode_set:
3293bad_stripe_cache:
3294bad_check_reshape:
3295 md_stop(&rs->md);
3296bad:
3297 raid_set_free(rs);
3298
3299 return r;
3300}
3301
3302static void raid_dtr(struct dm_target *ti)
3303{
3304 struct raid_set *rs = ti->private;
3305
3306 list_del_init(&rs->callbacks.list);
3307 md_stop(&rs->md);
3308 raid_set_free(rs);
3309}
3310
3311static int raid_map(struct dm_target *ti, struct bio *bio)
3312{
3313 struct raid_set *rs = ti->private;
3314 struct mddev *mddev = &rs->md;
3315
3316 /*
3317 * If we're reshaping to add disk(s)), ti->len and
3318 * mddev->array_sectors will differ during the process
3319 * (ti->len > mddev->array_sectors), so we have to requeue
3320 * bios with addresses > mddev->array_sectors here or
3321 * there will occur accesses past EOD of the component
3322 * data images thus erroring the raid set.
3323 */
3324 if (unlikely(bio_end_sector(bio) > mddev->array_sectors))
3325 return DM_MAPIO_REQUEUE;
3326
3327 md_handle_request(mddev, bio);
3328
3329 return DM_MAPIO_SUBMITTED;
3330}
3331
3332/* Return sync state string for @state */
3333enum sync_state { st_frozen, st_reshape, st_resync, st_check, st_repair, st_recover, st_idle };
3334static const char *sync_str(enum sync_state state)
3335{
3336 /* Has to be in above sync_state order! */
3337 static const char *sync_strs[] = {
3338 "frozen",
3339 "reshape",
3340 "resync",
3341 "check",
3342 "repair",
3343 "recover",
3344 "idle"
3345 };
3346
3347 return __within_range(state, 0, ARRAY_SIZE(sync_strs) - 1) ? sync_strs[state] : "undef";
3348};
3349
3350/* Return enum sync_state for @mddev derived from @recovery flags */
3351static enum sync_state decipher_sync_action(struct mddev *mddev, unsigned long recovery)
3352{
3353 if (test_bit(MD_RECOVERY_FROZEN, &recovery))
3354 return st_frozen;
3355
3356 /* The MD sync thread can be done with io or be interrupted but still be running */
3357 if (!test_bit(MD_RECOVERY_DONE, &recovery) &&
3358 (test_bit(MD_RECOVERY_RUNNING, &recovery) ||
3359 (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &recovery)))) {
3360 if (test_bit(MD_RECOVERY_RESHAPE, &recovery))
3361 return st_reshape;
3362
3363 if (test_bit(MD_RECOVERY_SYNC, &recovery)) {
3364 if (!test_bit(MD_RECOVERY_REQUESTED, &recovery))
3365 return st_resync;
3366 if (test_bit(MD_RECOVERY_CHECK, &recovery))
3367 return st_check;
3368 return st_repair;
3369 }
3370
3371 if (test_bit(MD_RECOVERY_RECOVER, &recovery))
3372 return st_recover;
3373
3374 if (mddev->reshape_position != MaxSector)
3375 return st_reshape;
3376 }
3377
3378 return st_idle;
3379}
3380
3381/*
3382 * Return status string for @rdev
3383 *
3384 * Status characters:
3385 *
3386 * 'D' = Dead/Failed raid set component or raid4/5/6 journal device
3387 * 'a' = Alive but not in-sync raid set component _or_ alive raid4/5/6 'write_back' journal device
3388 * 'A' = Alive and in-sync raid set component _or_ alive raid4/5/6 'write_through' journal device
3389 * '-' = Non-existing device (i.e. uspace passed '- -' into the ctr)
3390 */
3391static const char *__raid_dev_status(struct raid_set *rs, struct md_rdev *rdev)
3392{
3393 if (!rdev->bdev)
3394 return "-";
3395 else if (test_bit(Faulty, &rdev->flags))
3396 return "D";
3397 else if (test_bit(Journal, &rdev->flags))
3398 return (rs->journal_dev.mode == R5C_JOURNAL_MODE_WRITE_THROUGH) ? "A" : "a";
3399 else if (test_bit(RT_FLAG_RS_RESYNCING, &rs->runtime_flags) ||
3400 (!test_bit(RT_FLAG_RS_IN_SYNC, &rs->runtime_flags) &&
3401 !test_bit(In_sync, &rdev->flags)))
3402 return "a";
3403 else
3404 return "A";
3405}
3406
3407/* Helper to return resync/reshape progress for @rs and runtime flags for raid set in sync / resynching */
3408static sector_t rs_get_progress(struct raid_set *rs, unsigned long recovery,
3409 sector_t resync_max_sectors)
3410{
3411 sector_t r;
3412 enum sync_state state;
3413 struct mddev *mddev = &rs->md;
3414
3415 clear_bit(RT_FLAG_RS_IN_SYNC, &rs->runtime_flags);
3416 clear_bit(RT_FLAG_RS_RESYNCING, &rs->runtime_flags);
3417
3418 if (rs_is_raid0(rs)) {
3419 r = resync_max_sectors;
3420 set_bit(RT_FLAG_RS_IN_SYNC, &rs->runtime_flags);
3421
3422 } else {
3423 state = decipher_sync_action(mddev, recovery);
3424
3425 if (state == st_idle && !test_bit(MD_RECOVERY_INTR, &recovery))
3426 r = mddev->recovery_cp;
3427 else
3428 r = mddev->curr_resync_completed;
3429
3430 if (state == st_idle && r >= resync_max_sectors) {
3431 /*
3432 * Sync complete.
3433 */
3434 /* In case we have finished recovering, the array is in sync. */
3435 if (test_bit(MD_RECOVERY_RECOVER, &recovery))
3436 set_bit(RT_FLAG_RS_IN_SYNC, &rs->runtime_flags);
3437
3438 } else if (state == st_recover)
3439 /*
3440 * In case we are recovering, the array is not in sync
3441 * and health chars should show the recovering legs.
3442 */
3443 ;
3444 else if (state == st_resync)
3445 /*
3446 * If "resync" is occurring, the raid set
3447 * is or may be out of sync hence the health
3448 * characters shall be 'a'.
3449 */
3450 set_bit(RT_FLAG_RS_RESYNCING, &rs->runtime_flags);
3451 else if (state == st_reshape)
3452 /*
3453 * If "reshape" is occurring, the raid set
3454 * is or may be out of sync hence the health
3455 * characters shall be 'a'.
3456 */
3457 set_bit(RT_FLAG_RS_RESYNCING, &rs->runtime_flags);
3458
3459 else if (state == st_check || state == st_repair)
3460 /*
3461 * If "check" or "repair" is occurring, the raid set has
3462 * undergone an initial sync and the health characters
3463 * should not be 'a' anymore.
3464 */
3465 set_bit(RT_FLAG_RS_IN_SYNC, &rs->runtime_flags);
3466
3467 else {
3468 struct md_rdev *rdev;
3469
3470 /*
3471 * We are idle and recovery is needed, prevent 'A' chars race
3472 * caused by components still set to in-sync by constructor.
3473 */
3474 if (test_bit(MD_RECOVERY_NEEDED, &recovery))
3475 set_bit(RT_FLAG_RS_RESYNCING, &rs->runtime_flags);
3476
3477 /*
3478 * The raid set may be doing an initial sync, or it may
3479 * be rebuilding individual components. If all the
3480 * devices are In_sync, then it is the raid set that is
3481 * being initialized.
3482 */
3483 set_bit(RT_FLAG_RS_IN_SYNC, &rs->runtime_flags);
3484 rdev_for_each(rdev, mddev)
3485 if (!test_bit(Journal, &rdev->flags) &&
3486 !test_bit(In_sync, &rdev->flags)) {
3487 clear_bit(RT_FLAG_RS_IN_SYNC, &rs->runtime_flags);
3488 break;
3489 }
3490 }
3491 }
3492
3493 return min(r, resync_max_sectors);
3494}
3495
3496/* Helper to return @dev name or "-" if !@dev */
3497static const char *__get_dev_name(struct dm_dev *dev)
3498{
3499 return dev ? dev->name : "-";
3500}
3501
3502static void raid_status(struct dm_target *ti, status_type_t type,
3503 unsigned int status_flags, char *result, unsigned int maxlen)
3504{
3505 struct raid_set *rs = ti->private;
3506 struct mddev *mddev = &rs->md;
3507 struct r5conf *conf = mddev->private;
3508 int i, max_nr_stripes = conf ? conf->max_nr_stripes : 0;
3509 unsigned long recovery;
3510 unsigned int raid_param_cnt = 1; /* at least 1 for chunksize */
3511 unsigned int sz = 0;
3512 unsigned int rebuild_disks;
3513 unsigned int write_mostly_params = 0;
3514 sector_t progress, resync_max_sectors, resync_mismatches;
3515 const char *sync_action;
3516 struct raid_type *rt;
3517
3518 switch (type) {
3519 case STATUSTYPE_INFO:
3520 /* *Should* always succeed */
3521 rt = get_raid_type_by_ll(mddev->new_level, mddev->new_layout);
3522 if (!rt)
3523 return;
3524
3525 DMEMIT("%s %d ", rt->name, mddev->raid_disks);
3526
3527 /* Access most recent mddev properties for status output */
3528 smp_rmb();
3529 recovery = rs->md.recovery;
3530 /* Get sensible max sectors even if raid set not yet started */
3531 resync_max_sectors = test_bit(RT_FLAG_RS_PRERESUMED, &rs->runtime_flags) ?
3532 mddev->resync_max_sectors : mddev->dev_sectors;
3533 progress = rs_get_progress(rs, recovery, resync_max_sectors);
3534 resync_mismatches = (mddev->last_sync_action && !strcasecmp(mddev->last_sync_action, "check")) ?
3535 atomic64_read(&mddev->resync_mismatches) : 0;
3536 sync_action = sync_str(decipher_sync_action(&rs->md, recovery));
3537
3538 /* HM FIXME: do we want another state char for raid0? It shows 'D'/'A'/'-' now */
3539 for (i = 0; i < rs->raid_disks; i++)
3540 DMEMIT(__raid_dev_status(rs, &rs->dev[i].rdev));
3541
3542 /*
3543 * In-sync/Reshape ratio:
3544 * The in-sync ratio shows the progress of:
3545 * - Initializing the raid set
3546 * - Rebuilding a subset of devices of the raid set
3547 * The user can distinguish between the two by referring
3548 * to the status characters.
3549 *
3550 * The reshape ratio shows the progress of
3551 * changing the raid layout or the number of
3552 * disks of a raid set
3553 */
3554 DMEMIT(" %llu/%llu", (unsigned long long) progress,
3555 (unsigned long long) resync_max_sectors);
3556
3557 /*
3558 * v1.5.0+:
3559 *
3560 * Sync action:
3561 * See Documentation/admin-guide/device-mapper/dm-raid.rst for
3562 * information on each of these states.
3563 */
3564 DMEMIT(" %s", sync_action);
3565
3566 /*
3567 * v1.5.0+:
3568 *
3569 * resync_mismatches/mismatch_cnt
3570 * This field shows the number of discrepancies found when
3571 * performing a "check" of the raid set.
3572 */
3573 DMEMIT(" %llu", (unsigned long long) resync_mismatches);
3574
3575 /*
3576 * v1.9.0+:
3577 *
3578 * data_offset (needed for out of space reshaping)
3579 * This field shows the data offset into the data
3580 * image LV where the first stripes data starts.
3581 *
3582 * We keep data_offset equal on all raid disks of the set,
3583 * so retrieving it from the first raid disk is sufficient.
3584 */
3585 DMEMIT(" %llu", (unsigned long long) rs->dev[0].rdev.data_offset);
3586
3587 /*
3588 * v1.10.0+:
3589 */
3590 DMEMIT(" %s", test_bit(__CTR_FLAG_JOURNAL_DEV, &rs->ctr_flags) ?
3591 __raid_dev_status(rs, &rs->journal_dev.rdev) : "-");
3592 break;
3593
3594 case STATUSTYPE_TABLE:
3595 /* Report the table line string you would use to construct this raid set */
3596
3597 /* Calculate raid parameter count */
3598 for (i = 0; i < rs->raid_disks; i++)
3599 if (test_bit(WriteMostly, &rs->dev[i].rdev.flags))
3600 write_mostly_params += 2;
3601 rebuild_disks = memweight(rs->rebuild_disks, DISKS_ARRAY_ELEMS * sizeof(*rs->rebuild_disks));
3602 raid_param_cnt += rebuild_disks * 2 +
3603 write_mostly_params +
3604 hweight32(rs->ctr_flags & CTR_FLAG_OPTIONS_NO_ARGS) +
3605 hweight32(rs->ctr_flags & CTR_FLAG_OPTIONS_ONE_ARG) * 2 +
3606 (test_bit(__CTR_FLAG_JOURNAL_DEV, &rs->ctr_flags) ? 2 : 0) +
3607 (test_bit(__CTR_FLAG_JOURNAL_MODE, &rs->ctr_flags) ? 2 : 0);
3608
3609 /* Emit table line */
3610 /* This has to be in the documented order for userspace! */
3611 DMEMIT("%s %u %u", rs->raid_type->name, raid_param_cnt, mddev->new_chunk_sectors);
3612 if (test_bit(__CTR_FLAG_SYNC, &rs->ctr_flags))
3613 DMEMIT(" %s", dm_raid_arg_name_by_flag(CTR_FLAG_SYNC));
3614 if (test_bit(__CTR_FLAG_NOSYNC, &rs->ctr_flags))
3615 DMEMIT(" %s", dm_raid_arg_name_by_flag(CTR_FLAG_NOSYNC));
3616 if (rebuild_disks)
3617 for (i = 0; i < rs->raid_disks; i++)
3618 if (test_bit(rs->dev[i].rdev.raid_disk, (void *) rs->rebuild_disks))
3619 DMEMIT(" %s %u", dm_raid_arg_name_by_flag(CTR_FLAG_REBUILD),
3620 rs->dev[i].rdev.raid_disk);
3621 if (test_bit(__CTR_FLAG_DAEMON_SLEEP, &rs->ctr_flags))
3622 DMEMIT(" %s %lu", dm_raid_arg_name_by_flag(CTR_FLAG_DAEMON_SLEEP),
3623 mddev->bitmap_info.daemon_sleep);
3624 if (test_bit(__CTR_FLAG_MIN_RECOVERY_RATE, &rs->ctr_flags))
3625 DMEMIT(" %s %d", dm_raid_arg_name_by_flag(CTR_FLAG_MIN_RECOVERY_RATE),
3626 mddev->sync_speed_min);
3627 if (test_bit(__CTR_FLAG_MAX_RECOVERY_RATE, &rs->ctr_flags))
3628 DMEMIT(" %s %d", dm_raid_arg_name_by_flag(CTR_FLAG_MAX_RECOVERY_RATE),
3629 mddev->sync_speed_max);
3630 if (write_mostly_params)
3631 for (i = 0; i < rs->raid_disks; i++)
3632 if (test_bit(WriteMostly, &rs->dev[i].rdev.flags))
3633 DMEMIT(" %s %d", dm_raid_arg_name_by_flag(CTR_FLAG_WRITE_MOSTLY),
3634 rs->dev[i].rdev.raid_disk);
3635 if (test_bit(__CTR_FLAG_MAX_WRITE_BEHIND, &rs->ctr_flags))
3636 DMEMIT(" %s %lu", dm_raid_arg_name_by_flag(CTR_FLAG_MAX_WRITE_BEHIND),
3637 mddev->bitmap_info.max_write_behind);
3638 if (test_bit(__CTR_FLAG_STRIPE_CACHE, &rs->ctr_flags))
3639 DMEMIT(" %s %d", dm_raid_arg_name_by_flag(CTR_FLAG_STRIPE_CACHE),
3640 max_nr_stripes);
3641 if (test_bit(__CTR_FLAG_REGION_SIZE, &rs->ctr_flags))
3642 DMEMIT(" %s %llu", dm_raid_arg_name_by_flag(CTR_FLAG_REGION_SIZE),
3643 (unsigned long long) to_sector(mddev->bitmap_info.chunksize));
3644 if (test_bit(__CTR_FLAG_RAID10_COPIES, &rs->ctr_flags))
3645 DMEMIT(" %s %d", dm_raid_arg_name_by_flag(CTR_FLAG_RAID10_COPIES),
3646 raid10_md_layout_to_copies(mddev->layout));
3647 if (test_bit(__CTR_FLAG_RAID10_FORMAT, &rs->ctr_flags))
3648 DMEMIT(" %s %s", dm_raid_arg_name_by_flag(CTR_FLAG_RAID10_FORMAT),
3649 raid10_md_layout_to_format(mddev->layout));
3650 if (test_bit(__CTR_FLAG_DELTA_DISKS, &rs->ctr_flags))
3651 DMEMIT(" %s %d", dm_raid_arg_name_by_flag(CTR_FLAG_DELTA_DISKS),
3652 max(rs->delta_disks, mddev->delta_disks));
3653 if (test_bit(__CTR_FLAG_DATA_OFFSET, &rs->ctr_flags))
3654 DMEMIT(" %s %llu", dm_raid_arg_name_by_flag(CTR_FLAG_DATA_OFFSET),
3655 (unsigned long long) rs->data_offset);
3656 if (test_bit(__CTR_FLAG_JOURNAL_DEV, &rs->ctr_flags))
3657 DMEMIT(" %s %s", dm_raid_arg_name_by_flag(CTR_FLAG_JOURNAL_DEV),
3658 __get_dev_name(rs->journal_dev.dev));
3659 if (test_bit(__CTR_FLAG_JOURNAL_MODE, &rs->ctr_flags))
3660 DMEMIT(" %s %s", dm_raid_arg_name_by_flag(CTR_FLAG_JOURNAL_MODE),
3661 md_journal_mode_to_dm_raid(rs->journal_dev.mode));
3662 DMEMIT(" %d", rs->raid_disks);
3663 for (i = 0; i < rs->raid_disks; i++)
3664 DMEMIT(" %s %s", __get_dev_name(rs->dev[i].meta_dev),
3665 __get_dev_name(rs->dev[i].data_dev));
3666 }
3667}
3668
3669static int raid_message(struct dm_target *ti, unsigned int argc, char **argv,
3670 char *result, unsigned maxlen)
3671{
3672 struct raid_set *rs = ti->private;
3673 struct mddev *mddev = &rs->md;
3674
3675 if (!mddev->pers || !mddev->pers->sync_request)
3676 return -EINVAL;
3677
3678 if (!strcasecmp(argv[0], "frozen"))
3679 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3680 else
3681 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3682
3683 if (!strcasecmp(argv[0], "idle") || !strcasecmp(argv[0], "frozen")) {
3684 if (mddev->sync_thread) {
3685 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3686 md_reap_sync_thread(mddev);
3687 }
3688 } else if (decipher_sync_action(mddev, mddev->recovery) != st_idle)
3689 return -EBUSY;
3690 else if (!strcasecmp(argv[0], "resync"))
3691 ; /* MD_RECOVERY_NEEDED set below */
3692 else if (!strcasecmp(argv[0], "recover"))
3693 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
3694 else {
3695 if (!strcasecmp(argv[0], "check")) {
3696 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
3697 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
3698 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
3699 } else if (!strcasecmp(argv[0], "repair")) {
3700 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
3701 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
3702 } else
3703 return -EINVAL;
3704 }
3705 if (mddev->ro == 2) {
3706 /* A write to sync_action is enough to justify
3707 * canceling read-auto mode
3708 */
3709 mddev->ro = 0;
3710 if (!mddev->suspended && mddev->sync_thread)
3711 md_wakeup_thread(mddev->sync_thread);
3712 }
3713 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3714 if (!mddev->suspended && mddev->thread)
3715 md_wakeup_thread(mddev->thread);
3716
3717 return 0;
3718}
3719
3720static int raid_iterate_devices(struct dm_target *ti,
3721 iterate_devices_callout_fn fn, void *data)
3722{
3723 struct raid_set *rs = ti->private;
3724 unsigned int i;
3725 int r = 0;
3726
3727 for (i = 0; !r && i < rs->md.raid_disks; i++)
3728 if (rs->dev[i].data_dev)
3729 r = fn(ti,
3730 rs->dev[i].data_dev,
3731 0, /* No offset on data devs */
3732 rs->md.dev_sectors,
3733 data);
3734
3735 return r;
3736}
3737
3738static void raid_io_hints(struct dm_target *ti, struct queue_limits *limits)
3739{
3740 struct raid_set *rs = ti->private;
3741 unsigned int chunk_size_bytes = to_bytes(rs->md.chunk_sectors);
3742
3743 blk_limits_io_min(limits, chunk_size_bytes);
3744 blk_limits_io_opt(limits, chunk_size_bytes * mddev_data_stripes(rs));
3745
3746 /*
3747 * RAID1 and RAID10 personalities require bio splitting,
3748 * RAID0/4/5/6 don't and process large discard bios properly.
3749 */
3750 if (rs_is_raid1(rs) || rs_is_raid10(rs)) {
3751 limits->discard_granularity = chunk_size_bytes;
3752 limits->max_discard_sectors = rs->md.chunk_sectors;
3753 }
3754}
3755
3756static void raid_postsuspend(struct dm_target *ti)
3757{
3758 struct raid_set *rs = ti->private;
3759
3760 if (!test_and_set_bit(RT_FLAG_RS_SUSPENDED, &rs->runtime_flags)) {
3761 /* Writes have to be stopped before suspending to avoid deadlocks. */
3762 if (!test_bit(MD_RECOVERY_FROZEN, &rs->md.recovery))
3763 md_stop_writes(&rs->md);
3764
3765 mddev_lock_nointr(&rs->md);
3766 mddev_suspend(&rs->md);
3767 mddev_unlock(&rs->md);
3768 }
3769}
3770
3771static void attempt_restore_of_faulty_devices(struct raid_set *rs)
3772{
3773 int i;
3774 uint64_t cleared_failed_devices[DISKS_ARRAY_ELEMS];
3775 unsigned long flags;
3776 bool cleared = false;
3777 struct dm_raid_superblock *sb;
3778 struct mddev *mddev = &rs->md;
3779 struct md_rdev *r;
3780
3781 /* RAID personalities have to provide hot add/remove methods or we need to bail out. */
3782 if (!mddev->pers || !mddev->pers->hot_add_disk || !mddev->pers->hot_remove_disk)
3783 return;
3784
3785 memset(cleared_failed_devices, 0, sizeof(cleared_failed_devices));
3786
3787 for (i = 0; i < mddev->raid_disks; i++) {
3788 r = &rs->dev[i].rdev;
3789 /* HM FIXME: enhance journal device recovery processing */
3790 if (test_bit(Journal, &r->flags))
3791 continue;
3792
3793 if (test_bit(Faulty, &r->flags) &&
3794 r->meta_bdev && !read_disk_sb(r, r->sb_size, true)) {
3795 DMINFO("Faulty %s device #%d has readable super block."
3796 " Attempting to revive it.",
3797 rs->raid_type->name, i);
3798
3799 /*
3800 * Faulty bit may be set, but sometimes the array can
3801 * be suspended before the personalities can respond
3802 * by removing the device from the array (i.e. calling
3803 * 'hot_remove_disk'). If they haven't yet removed
3804 * the failed device, its 'raid_disk' number will be
3805 * '>= 0' - meaning we must call this function
3806 * ourselves.
3807 */
3808 flags = r->flags;
3809 clear_bit(In_sync, &r->flags); /* Mandatory for hot remove. */
3810 if (r->raid_disk >= 0) {
3811 if (mddev->pers->hot_remove_disk(mddev, r)) {
3812 /* Failed to revive this device, try next */
3813 r->flags = flags;
3814 continue;
3815 }
3816 } else
3817 r->raid_disk = r->saved_raid_disk = i;
3818
3819 clear_bit(Faulty, &r->flags);
3820 clear_bit(WriteErrorSeen, &r->flags);
3821
3822 if (mddev->pers->hot_add_disk(mddev, r)) {
3823 /* Failed to revive this device, try next */
3824 r->raid_disk = r->saved_raid_disk = -1;
3825 r->flags = flags;
3826 } else {
3827 clear_bit(In_sync, &r->flags);
3828 r->recovery_offset = 0;
3829 set_bit(i, (void *) cleared_failed_devices);
3830 cleared = true;
3831 }
3832 }
3833 }
3834
3835 /* If any failed devices could be cleared, update all sbs failed_devices bits */
3836 if (cleared) {
3837 uint64_t failed_devices[DISKS_ARRAY_ELEMS];
3838
3839 rdev_for_each(r, &rs->md) {
3840 if (test_bit(Journal, &r->flags))
3841 continue;
3842
3843 sb = page_address(r->sb_page);
3844 sb_retrieve_failed_devices(sb, failed_devices);
3845
3846 for (i = 0; i < DISKS_ARRAY_ELEMS; i++)
3847 failed_devices[i] &= ~cleared_failed_devices[i];
3848
3849 sb_update_failed_devices(sb, failed_devices);
3850 }
3851 }
3852}
3853
3854static int __load_dirty_region_bitmap(struct raid_set *rs)
3855{
3856 int r = 0;
3857
3858 /* Try loading the bitmap unless "raid0", which does not have one */
3859 if (!rs_is_raid0(rs) &&
3860 !test_and_set_bit(RT_FLAG_RS_BITMAP_LOADED, &rs->runtime_flags)) {
3861 r = md_bitmap_load(&rs->md);
3862 if (r)
3863 DMERR("Failed to load bitmap");
3864 }
3865
3866 return r;
3867}
3868
3869/* Enforce updating all superblocks */
3870static void rs_update_sbs(struct raid_set *rs)
3871{
3872 struct mddev *mddev = &rs->md;
3873 int ro = mddev->ro;
3874
3875 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
3876 mddev->ro = 0;
3877 md_update_sb(mddev, 1);
3878 mddev->ro = ro;
3879}
3880
3881/*
3882 * Reshape changes raid algorithm of @rs to new one within personality
3883 * (e.g. raid6_zr -> raid6_nc), changes stripe size, adds/removes
3884 * disks from a raid set thus growing/shrinking it or resizes the set
3885 *
3886 * Call mddev_lock_nointr() before!
3887 */
3888static int rs_start_reshape(struct raid_set *rs)
3889{
3890 int r;
3891 struct mddev *mddev = &rs->md;
3892 struct md_personality *pers = mddev->pers;
3893
3894 /* Don't allow the sync thread to work until the table gets reloaded. */
3895 set_bit(MD_RECOVERY_WAIT, &mddev->recovery);
3896
3897 r = rs_setup_reshape(rs);
3898 if (r)
3899 return r;
3900
3901 /*
3902 * Check any reshape constraints enforced by the personalility
3903 *
3904 * May as well already kick the reshape off so that * pers->start_reshape() becomes optional.
3905 */
3906 r = pers->check_reshape(mddev);
3907 if (r) {
3908 rs->ti->error = "pers->check_reshape() failed";
3909 return r;
3910 }
3911
3912 /*
3913 * Personality may not provide start reshape method in which
3914 * case check_reshape above has already covered everything
3915 */
3916 if (pers->start_reshape) {
3917 r = pers->start_reshape(mddev);
3918 if (r) {
3919 rs->ti->error = "pers->start_reshape() failed";
3920 return r;
3921 }
3922 }
3923
3924 /*
3925 * Now reshape got set up, update superblocks to
3926 * reflect the fact so that a table reload will
3927 * access proper superblock content in the ctr.
3928 */
3929 rs_update_sbs(rs);
3930
3931 return 0;
3932}
3933
3934static int raid_preresume(struct dm_target *ti)
3935{
3936 int r;
3937 struct raid_set *rs = ti->private;
3938 struct mddev *mddev = &rs->md;
3939
3940 /* This is a resume after a suspend of the set -> it's already started. */
3941 if (test_and_set_bit(RT_FLAG_RS_PRERESUMED, &rs->runtime_flags))
3942 return 0;
3943
3944 /*
3945 * The superblocks need to be updated on disk if the
3946 * array is new or new devices got added (thus zeroed
3947 * out by userspace) or __load_dirty_region_bitmap
3948 * will overwrite them in core with old data or fail.
3949 */
3950 if (test_bit(RT_FLAG_UPDATE_SBS, &rs->runtime_flags))
3951 rs_update_sbs(rs);
3952
3953 /* Load the bitmap from disk unless raid0 */
3954 r = __load_dirty_region_bitmap(rs);
3955 if (r)
3956 return r;
3957
3958 /* Resize bitmap to adjust to changed region size (aka MD bitmap chunksize) */
3959 if (test_bit(RT_FLAG_RS_BITMAP_LOADED, &rs->runtime_flags) && mddev->bitmap &&
3960 mddev->bitmap_info.chunksize != to_bytes(rs->requested_bitmap_chunk_sectors)) {
3961 r = md_bitmap_resize(mddev->bitmap, mddev->dev_sectors,
3962 to_bytes(rs->requested_bitmap_chunk_sectors), 0);
3963 if (r)
3964 DMERR("Failed to resize bitmap");
3965 }
3966
3967 /* Check for any resize/reshape on @rs and adjust/initiate */
3968 /* Be prepared for mddev_resume() in raid_resume() */
3969 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3970 if (mddev->recovery_cp && mddev->recovery_cp < MaxSector) {
3971 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
3972 mddev->resync_min = mddev->recovery_cp;
3973 }
3974
3975 /* Check for any reshape request unless new raid set */
3976 if (test_bit(RT_FLAG_RESHAPE_RS, &rs->runtime_flags)) {
3977 /* Initiate a reshape. */
3978 rs_set_rdev_sectors(rs);
3979 mddev_lock_nointr(mddev);
3980 r = rs_start_reshape(rs);
3981 mddev_unlock(mddev);
3982 if (r)
3983 DMWARN("Failed to check/start reshape, continuing without change");
3984 r = 0;
3985 }
3986
3987 return r;
3988}
3989
3990static void raid_resume(struct dm_target *ti)
3991{
3992 struct raid_set *rs = ti->private;
3993 struct mddev *mddev = &rs->md;
3994
3995 if (test_and_set_bit(RT_FLAG_RS_RESUMED, &rs->runtime_flags)) {
3996 /*
3997 * A secondary resume while the device is active.
3998 * Take this opportunity to check whether any failed
3999 * devices are reachable again.
4000 */
4001 attempt_restore_of_faulty_devices(rs);
4002 }
4003
4004 if (test_and_clear_bit(RT_FLAG_RS_SUSPENDED, &rs->runtime_flags)) {
4005 /* Only reduce raid set size before running a disk removing reshape. */
4006 if (mddev->delta_disks < 0)
4007 rs_set_capacity(rs);
4008
4009 mddev_lock_nointr(mddev);
4010 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4011 mddev->ro = 0;
4012 mddev->in_sync = 0;
4013 mddev_resume(mddev);
4014 mddev_unlock(mddev);
4015 }
4016}
4017
4018static struct target_type raid_target = {
4019 .name = "raid",
4020 .version = {1, 14, 0},
4021 .module = THIS_MODULE,
4022 .ctr = raid_ctr,
4023 .dtr = raid_dtr,
4024 .map = raid_map,
4025 .status = raid_status,
4026 .message = raid_message,
4027 .iterate_devices = raid_iterate_devices,
4028 .io_hints = raid_io_hints,
4029 .postsuspend = raid_postsuspend,
4030 .preresume = raid_preresume,
4031 .resume = raid_resume,
4032};
4033
4034static int __init dm_raid_init(void)
4035{
4036 DMINFO("Loading target version %u.%u.%u",
4037 raid_target.version[0],
4038 raid_target.version[1],
4039 raid_target.version[2]);
4040 return dm_register_target(&raid_target);
4041}
4042
4043static void __exit dm_raid_exit(void)
4044{
4045 dm_unregister_target(&raid_target);
4046}
4047
4048module_init(dm_raid_init);
4049module_exit(dm_raid_exit);
4050
4051module_param(devices_handle_discard_safely, bool, 0644);
4052MODULE_PARM_DESC(devices_handle_discard_safely,
4053 "Set to Y if all devices in each array reliably return zeroes on reads from discarded regions");
4054
4055MODULE_DESCRIPTION(DM_NAME " raid0/1/10/4/5/6 target");
4056MODULE_ALIAS("dm-raid0");
4057MODULE_ALIAS("dm-raid1");
4058MODULE_ALIAS("dm-raid10");
4059MODULE_ALIAS("dm-raid4");
4060MODULE_ALIAS("dm-raid5");
4061MODULE_ALIAS("dm-raid6");
4062MODULE_AUTHOR("Neil Brown <dm-devel@redhat.com>");
4063MODULE_AUTHOR("Heinz Mauelshagen <dm-devel@redhat.com>");
4064MODULE_LICENSE("GPL");