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