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_and_notify(gendisk, rs->md.array_sectors);
704}
705
706/*
707 * Set the mddev properties in @rs to the current
708 * ones retrieved from the freshest superblock
709 */
710static void rs_set_cur(struct raid_set *rs)
711{
712 struct mddev *mddev = &rs->md;
713
714 mddev->new_level = mddev->level;
715 mddev->new_layout = mddev->layout;
716 mddev->new_chunk_sectors = mddev->chunk_sectors;
717}
718
719/*
720 * Set the mddev properties in @rs to the new
721 * ones requested by the ctr
722 */
723static void rs_set_new(struct raid_set *rs)
724{
725 struct mddev *mddev = &rs->md;
726
727 mddev->level = mddev->new_level;
728 mddev->layout = mddev->new_layout;
729 mddev->chunk_sectors = mddev->new_chunk_sectors;
730 mddev->raid_disks = rs->raid_disks;
731 mddev->delta_disks = 0;
732}
733
734static struct raid_set *raid_set_alloc(struct dm_target *ti, struct raid_type *raid_type,
735 unsigned int raid_devs)
736{
737 unsigned int i;
738 struct raid_set *rs;
739
740 if (raid_devs <= raid_type->parity_devs) {
741 ti->error = "Insufficient number of devices";
742 return ERR_PTR(-EINVAL);
743 }
744
745 rs = kzalloc(struct_size(rs, dev, raid_devs), GFP_KERNEL);
746 if (!rs) {
747 ti->error = "Cannot allocate raid context";
748 return ERR_PTR(-ENOMEM);
749 }
750
751 mddev_init(&rs->md);
752
753 rs->raid_disks = raid_devs;
754 rs->delta_disks = 0;
755
756 rs->ti = ti;
757 rs->raid_type = raid_type;
758 rs->stripe_cache_entries = 256;
759 rs->md.raid_disks = raid_devs;
760 rs->md.level = raid_type->level;
761 rs->md.new_level = rs->md.level;
762 rs->md.layout = raid_type->algorithm;
763 rs->md.new_layout = rs->md.layout;
764 rs->md.delta_disks = 0;
765 rs->md.recovery_cp = MaxSector;
766
767 for (i = 0; i < raid_devs; i++)
768 md_rdev_init(&rs->dev[i].rdev);
769
770 /*
771 * Remaining items to be initialized by further RAID params:
772 * rs->md.persistent
773 * rs->md.external
774 * rs->md.chunk_sectors
775 * rs->md.new_chunk_sectors
776 * rs->md.dev_sectors
777 */
778
779 return rs;
780}
781
782/* Free all @rs allocations */
783static void raid_set_free(struct raid_set *rs)
784{
785 int i;
786
787 if (rs->journal_dev.dev) {
788 md_rdev_clear(&rs->journal_dev.rdev);
789 dm_put_device(rs->ti, rs->journal_dev.dev);
790 }
791
792 for (i = 0; i < rs->raid_disks; i++) {
793 if (rs->dev[i].meta_dev)
794 dm_put_device(rs->ti, rs->dev[i].meta_dev);
795 md_rdev_clear(&rs->dev[i].rdev);
796 if (rs->dev[i].data_dev)
797 dm_put_device(rs->ti, rs->dev[i].data_dev);
798 }
799
800 kfree(rs);
801}
802
803/*
804 * For every device we have two words
805 * <meta_dev>: meta device name or '-' if missing
806 * <data_dev>: data device name or '-' if missing
807 *
808 * The following are permitted:
809 * - -
810 * - <data_dev>
811 * <meta_dev> <data_dev>
812 *
813 * The following is not allowed:
814 * <meta_dev> -
815 *
816 * This code parses those words. If there is a failure,
817 * the caller must use raid_set_free() to unwind the operations.
818 */
819static int parse_dev_params(struct raid_set *rs, struct dm_arg_set *as)
820{
821 int i;
822 int rebuild = 0;
823 int metadata_available = 0;
824 int r = 0;
825 const char *arg;
826
827 /* Put off the number of raid devices argument to get to dev pairs */
828 arg = dm_shift_arg(as);
829 if (!arg)
830 return -EINVAL;
831
832 for (i = 0; i < rs->raid_disks; i++) {
833 rs->dev[i].rdev.raid_disk = i;
834
835 rs->dev[i].meta_dev = NULL;
836 rs->dev[i].data_dev = NULL;
837
838 /*
839 * There are no offsets initially.
840 * Out of place reshape will set them accordingly.
841 */
842 rs->dev[i].rdev.data_offset = 0;
843 rs->dev[i].rdev.new_data_offset = 0;
844 rs->dev[i].rdev.mddev = &rs->md;
845
846 arg = dm_shift_arg(as);
847 if (!arg)
848 return -EINVAL;
849
850 if (strcmp(arg, "-")) {
851 r = dm_get_device(rs->ti, arg, dm_table_get_mode(rs->ti->table),
852 &rs->dev[i].meta_dev);
853 if (r) {
854 rs->ti->error = "RAID metadata device lookup failure";
855 return r;
856 }
857
858 rs->dev[i].rdev.sb_page = alloc_page(GFP_KERNEL);
859 if (!rs->dev[i].rdev.sb_page) {
860 rs->ti->error = "Failed to allocate superblock page";
861 return -ENOMEM;
862 }
863 }
864
865 arg = dm_shift_arg(as);
866 if (!arg)
867 return -EINVAL;
868
869 if (!strcmp(arg, "-")) {
870 if (!test_bit(In_sync, &rs->dev[i].rdev.flags) &&
871 (!rs->dev[i].rdev.recovery_offset)) {
872 rs->ti->error = "Drive designated for rebuild not specified";
873 return -EINVAL;
874 }
875
876 if (rs->dev[i].meta_dev) {
877 rs->ti->error = "No data device supplied with metadata device";
878 return -EINVAL;
879 }
880
881 continue;
882 }
883
884 r = dm_get_device(rs->ti, arg, dm_table_get_mode(rs->ti->table),
885 &rs->dev[i].data_dev);
886 if (r) {
887 rs->ti->error = "RAID device lookup failure";
888 return r;
889 }
890
891 if (rs->dev[i].meta_dev) {
892 metadata_available = 1;
893 rs->dev[i].rdev.meta_bdev = rs->dev[i].meta_dev->bdev;
894 }
895 rs->dev[i].rdev.bdev = rs->dev[i].data_dev->bdev;
896 list_add_tail(&rs->dev[i].rdev.same_set, &rs->md.disks);
897 if (!test_bit(In_sync, &rs->dev[i].rdev.flags))
898 rebuild++;
899 }
900
901 if (rs->journal_dev.dev)
902 list_add_tail(&rs->journal_dev.rdev.same_set, &rs->md.disks);
903
904 if (metadata_available) {
905 rs->md.external = 0;
906 rs->md.persistent = 1;
907 rs->md.major_version = 2;
908 } else if (rebuild && !rs->md.recovery_cp) {
909 /*
910 * Without metadata, we will not be able to tell if the array
911 * is in-sync or not - we must assume it is not. Therefore,
912 * it is impossible to rebuild a drive.
913 *
914 * Even if there is metadata, the on-disk information may
915 * indicate that the array is not in-sync and it will then
916 * fail at that time.
917 *
918 * User could specify 'nosync' option if desperate.
919 */
920 rs->ti->error = "Unable to rebuild drive while array is not in-sync";
921 return -EINVAL;
922 }
923
924 return 0;
925}
926
927/*
928 * validate_region_size
929 * @rs
930 * @region_size: region size in sectors. If 0, pick a size (4MiB default).
931 *
932 * Set rs->md.bitmap_info.chunksize (which really refers to 'region size').
933 * Ensure that (ti->len/region_size < 2^21) - required by MD bitmap.
934 *
935 * Returns: 0 on success, -EINVAL on failure.
936 */
937static int validate_region_size(struct raid_set *rs, unsigned long region_size)
938{
939 unsigned long min_region_size = rs->ti->len / (1 << 21);
940
941 if (rs_is_raid0(rs))
942 return 0;
943
944 if (!region_size) {
945 /*
946 * Choose a reasonable default. All figures in sectors.
947 */
948 if (min_region_size > (1 << 13)) {
949 /* If not a power of 2, make it the next power of 2 */
950 region_size = roundup_pow_of_two(min_region_size);
951 DMINFO("Choosing default region size of %lu sectors",
952 region_size);
953 } else {
954 DMINFO("Choosing default region size of 4MiB");
955 region_size = 1 << 13; /* sectors */
956 }
957 } else {
958 /*
959 * Validate user-supplied value.
960 */
961 if (region_size > rs->ti->len) {
962 rs->ti->error = "Supplied region size is too large";
963 return -EINVAL;
964 }
965
966 if (region_size < min_region_size) {
967 DMERR("Supplied region_size (%lu sectors) below minimum (%lu)",
968 region_size, min_region_size);
969 rs->ti->error = "Supplied region size is too small";
970 return -EINVAL;
971 }
972
973 if (!is_power_of_2(region_size)) {
974 rs->ti->error = "Region size is not a power of 2";
975 return -EINVAL;
976 }
977
978 if (region_size < rs->md.chunk_sectors) {
979 rs->ti->error = "Region size is smaller than the chunk size";
980 return -EINVAL;
981 }
982 }
983
984 /*
985 * Convert sectors to bytes.
986 */
987 rs->md.bitmap_info.chunksize = to_bytes(region_size);
988
989 return 0;
990}
991
992/*
993 * validate_raid_redundancy
994 * @rs
995 *
996 * Determine if there are enough devices in the array that haven't
997 * failed (or are being rebuilt) to form a usable array.
998 *
999 * Returns: 0 on success, -EINVAL on failure.
1000 */
1001static int validate_raid_redundancy(struct raid_set *rs)
1002{
1003 unsigned int i, rebuild_cnt = 0;
1004 unsigned int rebuilds_per_group = 0, copies, raid_disks;
1005 unsigned int group_size, last_group_start;
1006
1007 for (i = 0; i < rs->raid_disks; i++)
1008 if (!test_bit(FirstUse, &rs->dev[i].rdev.flags) &&
1009 ((!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 raid_disks = min(rs->raid_disks, rs->md.raid_disks);
1051 if (__is_raid10_near(rs->md.new_layout)) {
1052 for (i = 0; i < raid_disks; i++) {
1053 if (!(i % copies))
1054 rebuilds_per_group = 0;
1055 if ((!rs->dev[i].rdev.sb_page ||
1056 !test_bit(In_sync, &rs->dev[i].rdev.flags)) &&
1057 (++rebuilds_per_group >= copies))
1058 goto too_many;
1059 }
1060 break;
1061 }
1062
1063 /*
1064 * When checking "far" and "offset" formats, we need to ensure
1065 * that the device that holds its copy is not also dead or
1066 * being rebuilt. (Note that "far" and "offset" formats only
1067 * support two copies right now. These formats also only ever
1068 * use the 'use_far_sets' variant.)
1069 *
1070 * This check is somewhat complicated by the need to account
1071 * for arrays that are not a multiple of (far) copies. This
1072 * results in the need to treat the last (potentially larger)
1073 * set differently.
1074 */
1075 group_size = (raid_disks / copies);
1076 last_group_start = (raid_disks / group_size) - 1;
1077 last_group_start *= group_size;
1078 for (i = 0; i < raid_disks; i++) {
1079 if (!(i % copies) && !(i > last_group_start))
1080 rebuilds_per_group = 0;
1081 if ((!rs->dev[i].rdev.sb_page ||
1082 !test_bit(In_sync, &rs->dev[i].rdev.flags)) &&
1083 (++rebuilds_per_group >= copies))
1084 goto too_many;
1085 }
1086 break;
1087 default:
1088 if (rebuild_cnt)
1089 return -EINVAL;
1090 }
1091
1092 return 0;
1093
1094too_many:
1095 return -EINVAL;
1096}
1097
1098/*
1099 * Possible arguments are...
1100 * <chunk_size> [optional_args]
1101 *
1102 * Argument definitions
1103 * <chunk_size> The number of sectors per disk that
1104 * will form the "stripe"
1105 * [[no]sync] Force or prevent recovery of the
1106 * entire array
1107 * [rebuild <idx>] Rebuild the drive indicated by the index
1108 * [daemon_sleep <ms>] Time between bitmap daemon work to
1109 * clear bits
1110 * [min_recovery_rate <kB/sec/disk>] Throttle RAID initialization
1111 * [max_recovery_rate <kB/sec/disk>] Throttle RAID initialization
1112 * [write_mostly <idx>] Indicate a write mostly drive via index
1113 * [max_write_behind <sectors>] See '-write-behind=' (man mdadm)
1114 * [stripe_cache <sectors>] Stripe cache size for higher RAIDs
1115 * [region_size <sectors>] Defines granularity of bitmap
1116 * [journal_dev <dev>] raid4/5/6 journaling deviice
1117 * (i.e. write hole closing log)
1118 *
1119 * RAID10-only options:
1120 * [raid10_copies <# copies>] Number of copies. (Default: 2)
1121 * [raid10_format <near|far|offset>] Layout algorithm. (Default: near)
1122 */
1123static int parse_raid_params(struct raid_set *rs, struct dm_arg_set *as,
1124 unsigned int num_raid_params)
1125{
1126 int value, raid10_format = ALGORITHM_RAID10_DEFAULT;
1127 unsigned int raid10_copies = 2;
1128 unsigned int i, write_mostly = 0;
1129 unsigned int region_size = 0;
1130 sector_t max_io_len;
1131 const char *arg, *key;
1132 struct raid_dev *rd;
1133 struct raid_type *rt = rs->raid_type;
1134
1135 arg = dm_shift_arg(as);
1136 num_raid_params--; /* Account for chunk_size argument */
1137
1138 if (kstrtoint(arg, 10, &value) < 0) {
1139 rs->ti->error = "Bad numerical argument given for chunk_size";
1140 return -EINVAL;
1141 }
1142
1143 /*
1144 * First, parse the in-order required arguments
1145 * "chunk_size" is the only argument of this type.
1146 */
1147 if (rt_is_raid1(rt)) {
1148 if (value)
1149 DMERR("Ignoring chunk size parameter for RAID 1");
1150 value = 0;
1151 } else if (!is_power_of_2(value)) {
1152 rs->ti->error = "Chunk size must be a power of 2";
1153 return -EINVAL;
1154 } else if (value < 8) {
1155 rs->ti->error = "Chunk size value is too small";
1156 return -EINVAL;
1157 }
1158
1159 rs->md.new_chunk_sectors = rs->md.chunk_sectors = value;
1160
1161 /*
1162 * We set each individual device as In_sync with a completed
1163 * 'recovery_offset'. If there has been a device failure or
1164 * replacement then one of the following cases applies:
1165 *
1166 * 1) User specifies 'rebuild'.
1167 * - Device is reset when param is read.
1168 * 2) A new device is supplied.
1169 * - No matching superblock found, resets device.
1170 * 3) Device failure was transient and returns on reload.
1171 * - Failure noticed, resets device for bitmap replay.
1172 * 4) Device hadn't completed recovery after previous failure.
1173 * - Superblock is read and overrides recovery_offset.
1174 *
1175 * What is found in the superblocks of the devices is always
1176 * authoritative, unless 'rebuild' or '[no]sync' was specified.
1177 */
1178 for (i = 0; i < rs->raid_disks; i++) {
1179 set_bit(In_sync, &rs->dev[i].rdev.flags);
1180 rs->dev[i].rdev.recovery_offset = MaxSector;
1181 }
1182
1183 /*
1184 * Second, parse the unordered optional arguments
1185 */
1186 for (i = 0; i < num_raid_params; i++) {
1187 key = dm_shift_arg(as);
1188 if (!key) {
1189 rs->ti->error = "Not enough raid parameters given";
1190 return -EINVAL;
1191 }
1192
1193 if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_NOSYNC))) {
1194 if (test_and_set_bit(__CTR_FLAG_NOSYNC, &rs->ctr_flags)) {
1195 rs->ti->error = "Only one 'nosync' argument allowed";
1196 return -EINVAL;
1197 }
1198 continue;
1199 }
1200 if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_SYNC))) {
1201 if (test_and_set_bit(__CTR_FLAG_SYNC, &rs->ctr_flags)) {
1202 rs->ti->error = "Only one 'sync' argument allowed";
1203 return -EINVAL;
1204 }
1205 continue;
1206 }
1207 if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_RAID10_USE_NEAR_SETS))) {
1208 if (test_and_set_bit(__CTR_FLAG_RAID10_USE_NEAR_SETS, &rs->ctr_flags)) {
1209 rs->ti->error = "Only one 'raid10_use_new_sets' argument allowed";
1210 return -EINVAL;
1211 }
1212 continue;
1213 }
1214
1215 arg = dm_shift_arg(as);
1216 i++; /* Account for the argument pairs */
1217 if (!arg) {
1218 rs->ti->error = "Wrong number of raid parameters given";
1219 return -EINVAL;
1220 }
1221
1222 /*
1223 * Parameters that take a string value are checked here.
1224 */
1225 /* "raid10_format {near|offset|far} */
1226 if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_RAID10_FORMAT))) {
1227 if (test_and_set_bit(__CTR_FLAG_RAID10_FORMAT, &rs->ctr_flags)) {
1228 rs->ti->error = "Only one 'raid10_format' argument pair allowed";
1229 return -EINVAL;
1230 }
1231 if (!rt_is_raid10(rt)) {
1232 rs->ti->error = "'raid10_format' is an invalid parameter for this RAID type";
1233 return -EINVAL;
1234 }
1235 raid10_format = raid10_name_to_format(arg);
1236 if (raid10_format < 0) {
1237 rs->ti->error = "Invalid 'raid10_format' value given";
1238 return raid10_format;
1239 }
1240 continue;
1241 }
1242
1243 /* "journal_dev <dev>" */
1244 if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_JOURNAL_DEV))) {
1245 int r;
1246 struct md_rdev *jdev;
1247
1248 if (test_and_set_bit(__CTR_FLAG_JOURNAL_DEV, &rs->ctr_flags)) {
1249 rs->ti->error = "Only one raid4/5/6 set journaling device allowed";
1250 return -EINVAL;
1251 }
1252 if (!rt_is_raid456(rt)) {
1253 rs->ti->error = "'journal_dev' is an invalid parameter for this RAID type";
1254 return -EINVAL;
1255 }
1256 r = dm_get_device(rs->ti, arg, dm_table_get_mode(rs->ti->table),
1257 &rs->journal_dev.dev);
1258 if (r) {
1259 rs->ti->error = "raid4/5/6 journal device lookup failure";
1260 return r;
1261 }
1262 jdev = &rs->journal_dev.rdev;
1263 md_rdev_init(jdev);
1264 jdev->mddev = &rs->md;
1265 jdev->bdev = rs->journal_dev.dev->bdev;
1266 jdev->sectors = bdev_nr_sectors(jdev->bdev);
1267 if (jdev->sectors < MIN_RAID456_JOURNAL_SPACE) {
1268 rs->ti->error = "No space for raid4/5/6 journal";
1269 return -ENOSPC;
1270 }
1271 rs->journal_dev.mode = R5C_JOURNAL_MODE_WRITE_THROUGH;
1272 set_bit(Journal, &jdev->flags);
1273 continue;
1274 }
1275
1276 /* "journal_mode <mode>" ("journal_dev" mandatory!) */
1277 if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_JOURNAL_MODE))) {
1278 int r;
1279
1280 if (!test_bit(__CTR_FLAG_JOURNAL_DEV, &rs->ctr_flags)) {
1281 rs->ti->error = "raid4/5/6 'journal_mode' is invalid without 'journal_dev'";
1282 return -EINVAL;
1283 }
1284 if (test_and_set_bit(__CTR_FLAG_JOURNAL_MODE, &rs->ctr_flags)) {
1285 rs->ti->error = "Only one raid4/5/6 'journal_mode' argument allowed";
1286 return -EINVAL;
1287 }
1288 r = dm_raid_journal_mode_to_md(arg);
1289 if (r < 0) {
1290 rs->ti->error = "Invalid 'journal_mode' argument";
1291 return r;
1292 }
1293 rs->journal_dev.mode = r;
1294 continue;
1295 }
1296
1297 /*
1298 * Parameters with number values from here on.
1299 */
1300 if (kstrtoint(arg, 10, &value) < 0) {
1301 rs->ti->error = "Bad numerical argument given in raid params";
1302 return -EINVAL;
1303 }
1304
1305 if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_REBUILD))) {
1306 /*
1307 * "rebuild" is being passed in by userspace to provide
1308 * indexes of replaced devices and to set up additional
1309 * devices on raid level takeover.
1310 */
1311 if (!__within_range(value, 0, rs->raid_disks - 1)) {
1312 rs->ti->error = "Invalid rebuild index given";
1313 return -EINVAL;
1314 }
1315
1316 if (test_and_set_bit(value, (void *) rs->rebuild_disks)) {
1317 rs->ti->error = "rebuild for this index already given";
1318 return -EINVAL;
1319 }
1320
1321 rd = rs->dev + value;
1322 clear_bit(In_sync, &rd->rdev.flags);
1323 clear_bit(Faulty, &rd->rdev.flags);
1324 rd->rdev.recovery_offset = 0;
1325 set_bit(__CTR_FLAG_REBUILD, &rs->ctr_flags);
1326 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_WRITE_MOSTLY))) {
1327 if (!rt_is_raid1(rt)) {
1328 rs->ti->error = "write_mostly option is only valid for RAID1";
1329 return -EINVAL;
1330 }
1331
1332 if (!__within_range(value, 0, rs->md.raid_disks - 1)) {
1333 rs->ti->error = "Invalid write_mostly index given";
1334 return -EINVAL;
1335 }
1336
1337 write_mostly++;
1338 set_bit(WriteMostly, &rs->dev[value].rdev.flags);
1339 set_bit(__CTR_FLAG_WRITE_MOSTLY, &rs->ctr_flags);
1340 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_MAX_WRITE_BEHIND))) {
1341 if (!rt_is_raid1(rt)) {
1342 rs->ti->error = "max_write_behind option is only valid for RAID1";
1343 return -EINVAL;
1344 }
1345
1346 if (test_and_set_bit(__CTR_FLAG_MAX_WRITE_BEHIND, &rs->ctr_flags)) {
1347 rs->ti->error = "Only one max_write_behind argument pair allowed";
1348 return -EINVAL;
1349 }
1350
1351 /*
1352 * In device-mapper, we specify things in sectors, but
1353 * MD records this value in kB
1354 */
1355 if (value < 0 || value / 2 > COUNTER_MAX) {
1356 rs->ti->error = "Max write-behind limit out of range";
1357 return -EINVAL;
1358 }
1359
1360 rs->md.bitmap_info.max_write_behind = value / 2;
1361 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_DAEMON_SLEEP))) {
1362 if (test_and_set_bit(__CTR_FLAG_DAEMON_SLEEP, &rs->ctr_flags)) {
1363 rs->ti->error = "Only one daemon_sleep argument pair allowed";
1364 return -EINVAL;
1365 }
1366 if (value < 0) {
1367 rs->ti->error = "daemon sleep period out of range";
1368 return -EINVAL;
1369 }
1370 rs->md.bitmap_info.daemon_sleep = value;
1371 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_DATA_OFFSET))) {
1372 /* Userspace passes new data_offset after having extended the data image LV */
1373 if (test_and_set_bit(__CTR_FLAG_DATA_OFFSET, &rs->ctr_flags)) {
1374 rs->ti->error = "Only one data_offset argument pair allowed";
1375 return -EINVAL;
1376 }
1377 /* Ensure sensible data offset */
1378 if (value < 0 ||
1379 (value && (value < MIN_FREE_RESHAPE_SPACE || value % to_sector(PAGE_SIZE)))) {
1380 rs->ti->error = "Bogus data_offset value";
1381 return -EINVAL;
1382 }
1383 rs->data_offset = value;
1384 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_DELTA_DISKS))) {
1385 /* Define the +/-# of disks to add to/remove from the given raid set */
1386 if (test_and_set_bit(__CTR_FLAG_DELTA_DISKS, &rs->ctr_flags)) {
1387 rs->ti->error = "Only one delta_disks argument pair allowed";
1388 return -EINVAL;
1389 }
1390 /* Ensure MAX_RAID_DEVICES and raid type minimal_devs! */
1391 if (!__within_range(abs(value), 1, MAX_RAID_DEVICES - rt->minimal_devs)) {
1392 rs->ti->error = "Too many delta_disk requested";
1393 return -EINVAL;
1394 }
1395
1396 rs->delta_disks = value;
1397 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_STRIPE_CACHE))) {
1398 if (test_and_set_bit(__CTR_FLAG_STRIPE_CACHE, &rs->ctr_flags)) {
1399 rs->ti->error = "Only one stripe_cache argument pair allowed";
1400 return -EINVAL;
1401 }
1402
1403 if (!rt_is_raid456(rt)) {
1404 rs->ti->error = "Inappropriate argument: stripe_cache";
1405 return -EINVAL;
1406 }
1407
1408 if (value < 0) {
1409 rs->ti->error = "Bogus stripe cache entries value";
1410 return -EINVAL;
1411 }
1412 rs->stripe_cache_entries = value;
1413 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_MIN_RECOVERY_RATE))) {
1414 if (test_and_set_bit(__CTR_FLAG_MIN_RECOVERY_RATE, &rs->ctr_flags)) {
1415 rs->ti->error = "Only one min_recovery_rate argument pair allowed";
1416 return -EINVAL;
1417 }
1418
1419 if (value < 0) {
1420 rs->ti->error = "min_recovery_rate out of range";
1421 return -EINVAL;
1422 }
1423 rs->md.sync_speed_min = value;
1424 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_MAX_RECOVERY_RATE))) {
1425 if (test_and_set_bit(__CTR_FLAG_MAX_RECOVERY_RATE, &rs->ctr_flags)) {
1426 rs->ti->error = "Only one max_recovery_rate argument pair allowed";
1427 return -EINVAL;
1428 }
1429
1430 if (value < 0) {
1431 rs->ti->error = "max_recovery_rate out of range";
1432 return -EINVAL;
1433 }
1434 rs->md.sync_speed_max = value;
1435 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_REGION_SIZE))) {
1436 if (test_and_set_bit(__CTR_FLAG_REGION_SIZE, &rs->ctr_flags)) {
1437 rs->ti->error = "Only one region_size argument pair allowed";
1438 return -EINVAL;
1439 }
1440
1441 region_size = value;
1442 rs->requested_bitmap_chunk_sectors = value;
1443 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_RAID10_COPIES))) {
1444 if (test_and_set_bit(__CTR_FLAG_RAID10_COPIES, &rs->ctr_flags)) {
1445 rs->ti->error = "Only one raid10_copies argument pair allowed";
1446 return -EINVAL;
1447 }
1448
1449 if (!__within_range(value, 2, rs->md.raid_disks)) {
1450 rs->ti->error = "Bad value for 'raid10_copies'";
1451 return -EINVAL;
1452 }
1453
1454 raid10_copies = value;
1455 } else {
1456 DMERR("Unable to parse RAID parameter: %s", key);
1457 rs->ti->error = "Unable to parse RAID parameter";
1458 return -EINVAL;
1459 }
1460 }
1461
1462 if (test_bit(__CTR_FLAG_SYNC, &rs->ctr_flags) &&
1463 test_bit(__CTR_FLAG_NOSYNC, &rs->ctr_flags)) {
1464 rs->ti->error = "sync and nosync are mutually exclusive";
1465 return -EINVAL;
1466 }
1467
1468 if (test_bit(__CTR_FLAG_REBUILD, &rs->ctr_flags) &&
1469 (test_bit(__CTR_FLAG_SYNC, &rs->ctr_flags) ||
1470 test_bit(__CTR_FLAG_NOSYNC, &rs->ctr_flags))) {
1471 rs->ti->error = "sync/nosync and rebuild are mutually exclusive";
1472 return -EINVAL;
1473 }
1474
1475 if (write_mostly >= rs->md.raid_disks) {
1476 rs->ti->error = "Can't set all raid1 devices to write_mostly";
1477 return -EINVAL;
1478 }
1479
1480 if (rs->md.sync_speed_max &&
1481 rs->md.sync_speed_min > rs->md.sync_speed_max) {
1482 rs->ti->error = "Bogus recovery rates";
1483 return -EINVAL;
1484 }
1485
1486 if (validate_region_size(rs, region_size))
1487 return -EINVAL;
1488
1489 if (rs->md.chunk_sectors)
1490 max_io_len = rs->md.chunk_sectors;
1491 else
1492 max_io_len = region_size;
1493
1494 if (dm_set_target_max_io_len(rs->ti, max_io_len))
1495 return -EINVAL;
1496
1497 if (rt_is_raid10(rt)) {
1498 if (raid10_copies > rs->md.raid_disks) {
1499 rs->ti->error = "Not enough devices to satisfy specification";
1500 return -EINVAL;
1501 }
1502
1503 rs->md.new_layout = raid10_format_to_md_layout(rs, raid10_format, raid10_copies);
1504 if (rs->md.new_layout < 0) {
1505 rs->ti->error = "Error getting raid10 format";
1506 return rs->md.new_layout;
1507 }
1508
1509 rt = get_raid_type_by_ll(10, rs->md.new_layout);
1510 if (!rt) {
1511 rs->ti->error = "Failed to recognize new raid10 layout";
1512 return -EINVAL;
1513 }
1514
1515 if ((rt->algorithm == ALGORITHM_RAID10_DEFAULT ||
1516 rt->algorithm == ALGORITHM_RAID10_NEAR) &&
1517 test_bit(__CTR_FLAG_RAID10_USE_NEAR_SETS, &rs->ctr_flags)) {
1518 rs->ti->error = "RAID10 format 'near' and 'raid10_use_near_sets' are incompatible";
1519 return -EINVAL;
1520 }
1521 }
1522
1523 rs->raid10_copies = raid10_copies;
1524
1525 /* Assume there are no metadata devices until the drives are parsed */
1526 rs->md.persistent = 0;
1527 rs->md.external = 1;
1528
1529 /* Check, if any invalid ctr arguments have been passed in for the raid level */
1530 return rs_check_for_valid_flags(rs);
1531}
1532
1533/* Set raid4/5/6 cache size */
1534static int rs_set_raid456_stripe_cache(struct raid_set *rs)
1535{
1536 int r;
1537 struct r5conf *conf;
1538 struct mddev *mddev = &rs->md;
1539 uint32_t min_stripes = max(mddev->chunk_sectors, mddev->new_chunk_sectors) / 2;
1540 uint32_t nr_stripes = rs->stripe_cache_entries;
1541
1542 if (!rt_is_raid456(rs->raid_type)) {
1543 rs->ti->error = "Inappropriate raid level; cannot change stripe_cache size";
1544 return -EINVAL;
1545 }
1546
1547 if (nr_stripes < min_stripes) {
1548 DMINFO("Adjusting requested %u stripe cache entries to %u to suit stripe size",
1549 nr_stripes, min_stripes);
1550 nr_stripes = min_stripes;
1551 }
1552
1553 conf = mddev->private;
1554 if (!conf) {
1555 rs->ti->error = "Cannot change stripe_cache size on inactive RAID set";
1556 return -EINVAL;
1557 }
1558
1559 /* Try setting number of stripes in raid456 stripe cache */
1560 if (conf->min_nr_stripes != nr_stripes) {
1561 r = raid5_set_cache_size(mddev, nr_stripes);
1562 if (r) {
1563 rs->ti->error = "Failed to set raid4/5/6 stripe cache size";
1564 return r;
1565 }
1566
1567 DMINFO("%u stripe cache entries", nr_stripes);
1568 }
1569
1570 return 0;
1571}
1572
1573/* Return # of data stripes as kept in mddev as of @rs (i.e. as of superblock) */
1574static unsigned int mddev_data_stripes(struct raid_set *rs)
1575{
1576 return rs->md.raid_disks - rs->raid_type->parity_devs;
1577}
1578
1579/* Return # of data stripes of @rs (i.e. as of ctr) */
1580static unsigned int rs_data_stripes(struct raid_set *rs)
1581{
1582 return rs->raid_disks - rs->raid_type->parity_devs;
1583}
1584
1585/*
1586 * Retrieve rdev->sectors from any valid raid device of @rs
1587 * to allow userpace to pass in arbitray "- -" device tupples.
1588 */
1589static sector_t __rdev_sectors(struct raid_set *rs)
1590{
1591 int i;
1592
1593 for (i = 0; i < rs->raid_disks; i++) {
1594 struct md_rdev *rdev = &rs->dev[i].rdev;
1595
1596 if (!test_bit(Journal, &rdev->flags) &&
1597 rdev->bdev && rdev->sectors)
1598 return rdev->sectors;
1599 }
1600
1601 return 0;
1602}
1603
1604/* Check that calculated dev_sectors fits all component devices. */
1605static int _check_data_dev_sectors(struct raid_set *rs)
1606{
1607 sector_t ds = ~0;
1608 struct md_rdev *rdev;
1609
1610 rdev_for_each(rdev, &rs->md)
1611 if (!test_bit(Journal, &rdev->flags) && rdev->bdev) {
1612 ds = min(ds, bdev_nr_sectors(rdev->bdev));
1613 if (ds < rs->md.dev_sectors) {
1614 rs->ti->error = "Component device(s) too small";
1615 return -EINVAL;
1616 }
1617 }
1618
1619 return 0;
1620}
1621
1622/* Calculate the sectors per device and per array used for @rs */
1623static int rs_set_dev_and_array_sectors(struct raid_set *rs, sector_t sectors, bool use_mddev)
1624{
1625 int delta_disks;
1626 unsigned int data_stripes;
1627 sector_t array_sectors = sectors, dev_sectors = sectors;
1628 struct mddev *mddev = &rs->md;
1629
1630 if (use_mddev) {
1631 delta_disks = mddev->delta_disks;
1632 data_stripes = mddev_data_stripes(rs);
1633 } else {
1634 delta_disks = rs->delta_disks;
1635 data_stripes = rs_data_stripes(rs);
1636 }
1637
1638 /* Special raid1 case w/o delta_disks support (yet) */
1639 if (rt_is_raid1(rs->raid_type))
1640 ;
1641 else if (rt_is_raid10(rs->raid_type)) {
1642 if (rs->raid10_copies < 2 ||
1643 delta_disks < 0) {
1644 rs->ti->error = "Bogus raid10 data copies or delta disks";
1645 return -EINVAL;
1646 }
1647
1648 dev_sectors *= rs->raid10_copies;
1649 if (sector_div(dev_sectors, data_stripes))
1650 goto bad;
1651
1652 array_sectors = (data_stripes + delta_disks) * dev_sectors;
1653 if (sector_div(array_sectors, rs->raid10_copies))
1654 goto bad;
1655
1656 } else if (sector_div(dev_sectors, data_stripes))
1657 goto bad;
1658
1659 else
1660 /* Striped layouts */
1661 array_sectors = (data_stripes + delta_disks) * dev_sectors;
1662
1663 mddev->array_sectors = array_sectors;
1664 mddev->dev_sectors = dev_sectors;
1665 rs_set_rdev_sectors(rs);
1666
1667 return _check_data_dev_sectors(rs);
1668bad:
1669 rs->ti->error = "Target length not divisible by number of data devices";
1670 return -EINVAL;
1671}
1672
1673/* Setup recovery on @rs */
1674static void rs_setup_recovery(struct raid_set *rs, sector_t dev_sectors)
1675{
1676 /* raid0 does not recover */
1677 if (rs_is_raid0(rs))
1678 rs->md.recovery_cp = MaxSector;
1679 /*
1680 * A raid6 set has to be recovered either
1681 * completely or for the grown part to
1682 * ensure proper parity and Q-Syndrome
1683 */
1684 else if (rs_is_raid6(rs))
1685 rs->md.recovery_cp = dev_sectors;
1686 /*
1687 * Other raid set types may skip recovery
1688 * depending on the 'nosync' flag.
1689 */
1690 else
1691 rs->md.recovery_cp = test_bit(__CTR_FLAG_NOSYNC, &rs->ctr_flags)
1692 ? MaxSector : dev_sectors;
1693}
1694
1695static void do_table_event(struct work_struct *ws)
1696{
1697 struct raid_set *rs = container_of(ws, struct raid_set, md.event_work);
1698
1699 smp_rmb(); /* Make sure we access most actual mddev properties */
1700 if (!rs_is_reshaping(rs)) {
1701 if (rs_is_raid10(rs))
1702 rs_set_rdev_sectors(rs);
1703 rs_set_capacity(rs);
1704 }
1705 dm_table_event(rs->ti->table);
1706}
1707
1708/*
1709 * Make sure a valid takover (level switch) is being requested on @rs
1710 *
1711 * Conversions of raid sets from one MD personality to another
1712 * have to conform to restrictions which are enforced here.
1713 */
1714static int rs_check_takeover(struct raid_set *rs)
1715{
1716 struct mddev *mddev = &rs->md;
1717 unsigned int near_copies;
1718
1719 if (rs->md.degraded) {
1720 rs->ti->error = "Can't takeover degraded raid set";
1721 return -EPERM;
1722 }
1723
1724 if (rs_is_reshaping(rs)) {
1725 rs->ti->error = "Can't takeover reshaping raid set";
1726 return -EPERM;
1727 }
1728
1729 switch (mddev->level) {
1730 case 0:
1731 /* raid0 -> raid1/5 with one disk */
1732 if ((mddev->new_level == 1 || mddev->new_level == 5) &&
1733 mddev->raid_disks == 1)
1734 return 0;
1735
1736 /* raid0 -> raid10 */
1737 if (mddev->new_level == 10 &&
1738 !(rs->raid_disks % mddev->raid_disks))
1739 return 0;
1740
1741 /* raid0 with multiple disks -> raid4/5/6 */
1742 if (__within_range(mddev->new_level, 4, 6) &&
1743 mddev->new_layout == ALGORITHM_PARITY_N &&
1744 mddev->raid_disks > 1)
1745 return 0;
1746
1747 break;
1748
1749 case 10:
1750 /* Can't takeover raid10_offset! */
1751 if (__is_raid10_offset(mddev->layout))
1752 break;
1753
1754 near_copies = __raid10_near_copies(mddev->layout);
1755
1756 /* raid10* -> raid0 */
1757 if (mddev->new_level == 0) {
1758 /* Can takeover raid10_near with raid disks divisable by data copies! */
1759 if (near_copies > 1 &&
1760 !(mddev->raid_disks % near_copies)) {
1761 mddev->raid_disks /= near_copies;
1762 mddev->delta_disks = mddev->raid_disks;
1763 return 0;
1764 }
1765
1766 /* Can takeover raid10_far */
1767 if (near_copies == 1 &&
1768 __raid10_far_copies(mddev->layout) > 1)
1769 return 0;
1770
1771 break;
1772 }
1773
1774 /* raid10_{near,far} -> raid1 */
1775 if (mddev->new_level == 1 &&
1776 max(near_copies, __raid10_far_copies(mddev->layout)) == mddev->raid_disks)
1777 return 0;
1778
1779 /* raid10_{near,far} with 2 disks -> raid4/5 */
1780 if (__within_range(mddev->new_level, 4, 5) &&
1781 mddev->raid_disks == 2)
1782 return 0;
1783 break;
1784
1785 case 1:
1786 /* raid1 with 2 disks -> raid4/5 */
1787 if (__within_range(mddev->new_level, 4, 5) &&
1788 mddev->raid_disks == 2) {
1789 mddev->degraded = 1;
1790 return 0;
1791 }
1792
1793 /* raid1 -> raid0 */
1794 if (mddev->new_level == 0 &&
1795 mddev->raid_disks == 1)
1796 return 0;
1797
1798 /* raid1 -> raid10 */
1799 if (mddev->new_level == 10)
1800 return 0;
1801 break;
1802
1803 case 4:
1804 /* raid4 -> raid0 */
1805 if (mddev->new_level == 0)
1806 return 0;
1807
1808 /* raid4 -> raid1/5 with 2 disks */
1809 if ((mddev->new_level == 1 || mddev->new_level == 5) &&
1810 mddev->raid_disks == 2)
1811 return 0;
1812
1813 /* raid4 -> raid5/6 with parity N */
1814 if (__within_range(mddev->new_level, 5, 6) &&
1815 mddev->layout == ALGORITHM_PARITY_N)
1816 return 0;
1817 break;
1818
1819 case 5:
1820 /* raid5 with parity N -> raid0 */
1821 if (mddev->new_level == 0 &&
1822 mddev->layout == ALGORITHM_PARITY_N)
1823 return 0;
1824
1825 /* raid5 with parity N -> raid4 */
1826 if (mddev->new_level == 4 &&
1827 mddev->layout == ALGORITHM_PARITY_N)
1828 return 0;
1829
1830 /* raid5 with 2 disks -> raid1/4/10 */
1831 if ((mddev->new_level == 1 || mddev->new_level == 4 || mddev->new_level == 10) &&
1832 mddev->raid_disks == 2)
1833 return 0;
1834
1835 /* raid5_* -> raid6_*_6 with Q-Syndrome N (e.g. raid5_ra -> raid6_ra_6 */
1836 if (mddev->new_level == 6 &&
1837 ((mddev->layout == ALGORITHM_PARITY_N && mddev->new_layout == ALGORITHM_PARITY_N) ||
1838 __within_range(mddev->new_layout, ALGORITHM_LEFT_ASYMMETRIC_6, ALGORITHM_RIGHT_SYMMETRIC_6)))
1839 return 0;
1840 break;
1841
1842 case 6:
1843 /* raid6 with parity N -> raid0 */
1844 if (mddev->new_level == 0 &&
1845 mddev->layout == ALGORITHM_PARITY_N)
1846 return 0;
1847
1848 /* raid6 with parity N -> raid4 */
1849 if (mddev->new_level == 4 &&
1850 mddev->layout == ALGORITHM_PARITY_N)
1851 return 0;
1852
1853 /* raid6_*_n with Q-Syndrome N -> raid5_* */
1854 if (mddev->new_level == 5 &&
1855 ((mddev->layout == ALGORITHM_PARITY_N && mddev->new_layout == ALGORITHM_PARITY_N) ||
1856 __within_range(mddev->new_layout, ALGORITHM_LEFT_ASYMMETRIC, ALGORITHM_RIGHT_SYMMETRIC)))
1857 return 0;
1858 break;
1859
1860 default:
1861 break;
1862 }
1863
1864 rs->ti->error = "takeover not possible";
1865 return -EINVAL;
1866}
1867
1868/* True if @rs requested to be taken over */
1869static bool rs_takeover_requested(struct raid_set *rs)
1870{
1871 return rs->md.new_level != rs->md.level;
1872}
1873
1874/* True if layout is set to reshape. */
1875static bool rs_is_layout_change(struct raid_set *rs, bool use_mddev)
1876{
1877 return (use_mddev ? rs->md.delta_disks : rs->delta_disks) ||
1878 rs->md.new_layout != rs->md.layout ||
1879 rs->md.new_chunk_sectors != rs->md.chunk_sectors;
1880}
1881
1882/* True if @rs is requested to reshape by ctr */
1883static bool rs_reshape_requested(struct raid_set *rs)
1884{
1885 bool change;
1886 struct mddev *mddev = &rs->md;
1887
1888 if (rs_takeover_requested(rs))
1889 return false;
1890
1891 if (rs_is_raid0(rs))
1892 return false;
1893
1894 change = rs_is_layout_change(rs, false);
1895
1896 /* Historical case to support raid1 reshape without delta disks */
1897 if (rs_is_raid1(rs)) {
1898 if (rs->delta_disks)
1899 return !!rs->delta_disks;
1900
1901 return !change &&
1902 mddev->raid_disks != rs->raid_disks;
1903 }
1904
1905 if (rs_is_raid10(rs))
1906 return change &&
1907 !__is_raid10_far(mddev->new_layout) &&
1908 rs->delta_disks >= 0;
1909
1910 return change;
1911}
1912
1913/* Features */
1914#define FEATURE_FLAG_SUPPORTS_V190 0x1 /* Supports extended superblock */
1915
1916/* State flags for sb->flags */
1917#define SB_FLAG_RESHAPE_ACTIVE 0x1
1918#define SB_FLAG_RESHAPE_BACKWARDS 0x2
1919
1920/*
1921 * This structure is never routinely used by userspace, unlike md superblocks.
1922 * Devices with this superblock should only ever be accessed via device-mapper.
1923 */
1924#define DM_RAID_MAGIC 0x64526D44
1925struct dm_raid_superblock {
1926 __le32 magic; /* "DmRd" */
1927 __le32 compat_features; /* Used to indicate compatible features (like 1.9.0 ondisk metadata extension) */
1928
1929 __le32 num_devices; /* Number of devices in this raid set. (Max 64) */
1930 __le32 array_position; /* The position of this drive in the raid set */
1931
1932 __le64 events; /* Incremented by md when superblock updated */
1933 __le64 failed_devices; /* Pre 1.9.0 part of bit field of devices to */
1934 /* indicate failures (see extension below) */
1935
1936 /*
1937 * This offset tracks the progress of the repair or replacement of
1938 * an individual drive.
1939 */
1940 __le64 disk_recovery_offset;
1941
1942 /*
1943 * This offset tracks the progress of the initial raid set
1944 * synchronisation/parity calculation.
1945 */
1946 __le64 array_resync_offset;
1947
1948 /*
1949 * raid characteristics
1950 */
1951 __le32 level;
1952 __le32 layout;
1953 __le32 stripe_sectors;
1954
1955 /********************************************************************
1956 * BELOW FOLLOW V1.9.0 EXTENSIONS TO THE PRISTINE SUPERBLOCK FORMAT!!!
1957 *
1958 * FEATURE_FLAG_SUPPORTS_V190 in the compat_features member indicates that those exist
1959 */
1960
1961 __le32 flags; /* Flags defining array states for reshaping */
1962
1963 /*
1964 * This offset tracks the progress of a raid
1965 * set reshape in order to be able to restart it
1966 */
1967 __le64 reshape_position;
1968
1969 /*
1970 * These define the properties of the array in case of an interrupted reshape
1971 */
1972 __le32 new_level;
1973 __le32 new_layout;
1974 __le32 new_stripe_sectors;
1975 __le32 delta_disks;
1976
1977 __le64 array_sectors; /* Array size in sectors */
1978
1979 /*
1980 * Sector offsets to data on devices (reshaping).
1981 * Needed to support out of place reshaping, thus
1982 * not writing over any stripes whilst converting
1983 * them from old to new layout
1984 */
1985 __le64 data_offset;
1986 __le64 new_data_offset;
1987
1988 __le64 sectors; /* Used device size in sectors */
1989
1990 /*
1991 * Additonal Bit field of devices indicating failures to support
1992 * up to 256 devices with the 1.9.0 on-disk metadata format
1993 */
1994 __le64 extended_failed_devices[DISKS_ARRAY_ELEMS - 1];
1995
1996 __le32 incompat_features; /* Used to indicate any incompatible features */
1997
1998 /* Always set rest up to logical block size to 0 when writing (see get_metadata_device() below). */
1999} __packed;
2000
2001/*
2002 * Check for reshape constraints on raid set @rs:
2003 *
2004 * - reshape function non-existent
2005 * - degraded set
2006 * - ongoing recovery
2007 * - ongoing reshape
2008 *
2009 * Returns 0 if none or -EPERM if given constraint
2010 * and error message reference in @errmsg
2011 */
2012static int rs_check_reshape(struct raid_set *rs)
2013{
2014 struct mddev *mddev = &rs->md;
2015
2016 if (!mddev->pers || !mddev->pers->check_reshape)
2017 rs->ti->error = "Reshape not supported";
2018 else if (mddev->degraded)
2019 rs->ti->error = "Can't reshape degraded raid set";
2020 else if (rs_is_recovering(rs))
2021 rs->ti->error = "Convert request on recovering raid set prohibited";
2022 else if (rs_is_reshaping(rs))
2023 rs->ti->error = "raid set already reshaping!";
2024 else if (!(rs_is_raid1(rs) || rs_is_raid10(rs) || rs_is_raid456(rs)))
2025 rs->ti->error = "Reshaping only supported for raid1/4/5/6/10";
2026 else
2027 return 0;
2028
2029 return -EPERM;
2030}
2031
2032static int read_disk_sb(struct md_rdev *rdev, int size, bool force_reload)
2033{
2034 BUG_ON(!rdev->sb_page);
2035
2036 if (rdev->sb_loaded && !force_reload)
2037 return 0;
2038
2039 rdev->sb_loaded = 0;
2040
2041 if (!sync_page_io(rdev, 0, size, rdev->sb_page, REQ_OP_READ, true)) {
2042 DMERR("Failed to read superblock of device at position %d",
2043 rdev->raid_disk);
2044 md_error(rdev->mddev, rdev);
2045 set_bit(Faulty, &rdev->flags);
2046 return -EIO;
2047 }
2048
2049 rdev->sb_loaded = 1;
2050
2051 return 0;
2052}
2053
2054static void sb_retrieve_failed_devices(struct dm_raid_superblock *sb, uint64_t *failed_devices)
2055{
2056 failed_devices[0] = le64_to_cpu(sb->failed_devices);
2057 memset(failed_devices + 1, 0, sizeof(sb->extended_failed_devices));
2058
2059 if (le32_to_cpu(sb->compat_features) & FEATURE_FLAG_SUPPORTS_V190) {
2060 int i = ARRAY_SIZE(sb->extended_failed_devices);
2061
2062 while (i--)
2063 failed_devices[i+1] = le64_to_cpu(sb->extended_failed_devices[i]);
2064 }
2065}
2066
2067static void sb_update_failed_devices(struct dm_raid_superblock *sb, uint64_t *failed_devices)
2068{
2069 int i = ARRAY_SIZE(sb->extended_failed_devices);
2070
2071 sb->failed_devices = cpu_to_le64(failed_devices[0]);
2072 while (i--)
2073 sb->extended_failed_devices[i] = cpu_to_le64(failed_devices[i+1]);
2074}
2075
2076/*
2077 * Synchronize the superblock members with the raid set properties
2078 *
2079 * All superblock data is little endian.
2080 */
2081static void super_sync(struct mddev *mddev, struct md_rdev *rdev)
2082{
2083 bool update_failed_devices = false;
2084 unsigned int i;
2085 uint64_t failed_devices[DISKS_ARRAY_ELEMS];
2086 struct dm_raid_superblock *sb;
2087 struct raid_set *rs = container_of(mddev, struct raid_set, md);
2088
2089 /* No metadata device, no superblock */
2090 if (!rdev->meta_bdev)
2091 return;
2092
2093 BUG_ON(!rdev->sb_page);
2094
2095 sb = page_address(rdev->sb_page);
2096
2097 sb_retrieve_failed_devices(sb, failed_devices);
2098
2099 for (i = 0; i < rs->raid_disks; i++)
2100 if (!rs->dev[i].data_dev || test_bit(Faulty, &rs->dev[i].rdev.flags)) {
2101 update_failed_devices = true;
2102 set_bit(i, (void *) failed_devices);
2103 }
2104
2105 if (update_failed_devices)
2106 sb_update_failed_devices(sb, failed_devices);
2107
2108 sb->magic = cpu_to_le32(DM_RAID_MAGIC);
2109 sb->compat_features = cpu_to_le32(FEATURE_FLAG_SUPPORTS_V190);
2110
2111 sb->num_devices = cpu_to_le32(mddev->raid_disks);
2112 sb->array_position = cpu_to_le32(rdev->raid_disk);
2113
2114 sb->events = cpu_to_le64(mddev->events);
2115
2116 sb->disk_recovery_offset = cpu_to_le64(rdev->recovery_offset);
2117 sb->array_resync_offset = cpu_to_le64(mddev->recovery_cp);
2118
2119 sb->level = cpu_to_le32(mddev->level);
2120 sb->layout = cpu_to_le32(mddev->layout);
2121 sb->stripe_sectors = cpu_to_le32(mddev->chunk_sectors);
2122
2123 /********************************************************************
2124 * BELOW FOLLOW V1.9.0 EXTENSIONS TO THE PRISTINE SUPERBLOCK FORMAT!!!
2125 *
2126 * FEATURE_FLAG_SUPPORTS_V190 in the compat_features member indicates that those exist
2127 */
2128 sb->new_level = cpu_to_le32(mddev->new_level);
2129 sb->new_layout = cpu_to_le32(mddev->new_layout);
2130 sb->new_stripe_sectors = cpu_to_le32(mddev->new_chunk_sectors);
2131
2132 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
2133
2134 smp_rmb(); /* Make sure we access most recent reshape position */
2135 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
2136 if (le64_to_cpu(sb->reshape_position) != MaxSector) {
2137 /* Flag ongoing reshape */
2138 sb->flags |= cpu_to_le32(SB_FLAG_RESHAPE_ACTIVE);
2139
2140 if (mddev->delta_disks < 0 || mddev->reshape_backwards)
2141 sb->flags |= cpu_to_le32(SB_FLAG_RESHAPE_BACKWARDS);
2142 } else {
2143 /* Clear reshape flags */
2144 sb->flags &= ~(cpu_to_le32(SB_FLAG_RESHAPE_ACTIVE|SB_FLAG_RESHAPE_BACKWARDS));
2145 }
2146
2147 sb->array_sectors = cpu_to_le64(mddev->array_sectors);
2148 sb->data_offset = cpu_to_le64(rdev->data_offset);
2149 sb->new_data_offset = cpu_to_le64(rdev->new_data_offset);
2150 sb->sectors = cpu_to_le64(rdev->sectors);
2151 sb->incompat_features = cpu_to_le32(0);
2152
2153 /* Zero out the rest of the payload after the size of the superblock */
2154 memset(sb + 1, 0, rdev->sb_size - sizeof(*sb));
2155}
2156
2157/*
2158 * super_load
2159 *
2160 * This function creates a superblock if one is not found on the device
2161 * and will decide which superblock to use if there's a choice.
2162 *
2163 * Return: 1 if use rdev, 0 if use refdev, -Exxx otherwise
2164 */
2165static int super_load(struct md_rdev *rdev, struct md_rdev *refdev)
2166{
2167 int r;
2168 struct dm_raid_superblock *sb;
2169 struct dm_raid_superblock *refsb;
2170 uint64_t events_sb, events_refsb;
2171
2172 r = read_disk_sb(rdev, rdev->sb_size, false);
2173 if (r)
2174 return r;
2175
2176 sb = page_address(rdev->sb_page);
2177
2178 /*
2179 * Two cases that we want to write new superblocks and rebuild:
2180 * 1) New device (no matching magic number)
2181 * 2) Device specified for rebuild (!In_sync w/ offset == 0)
2182 */
2183 if ((sb->magic != cpu_to_le32(DM_RAID_MAGIC)) ||
2184 (!test_bit(In_sync, &rdev->flags) && !rdev->recovery_offset)) {
2185 super_sync(rdev->mddev, rdev);
2186
2187 set_bit(FirstUse, &rdev->flags);
2188 sb->compat_features = cpu_to_le32(FEATURE_FLAG_SUPPORTS_V190);
2189
2190 /* Force writing of superblocks to disk */
2191 set_bit(MD_SB_CHANGE_DEVS, &rdev->mddev->sb_flags);
2192
2193 /* Any superblock is better than none, choose that if given */
2194 return refdev ? 0 : 1;
2195 }
2196
2197 if (!refdev)
2198 return 1;
2199
2200 events_sb = le64_to_cpu(sb->events);
2201
2202 refsb = page_address(refdev->sb_page);
2203 events_refsb = le64_to_cpu(refsb->events);
2204
2205 return (events_sb > events_refsb) ? 1 : 0;
2206}
2207
2208static int super_init_validation(struct raid_set *rs, struct md_rdev *rdev)
2209{
2210 int role;
2211 unsigned int d;
2212 struct mddev *mddev = &rs->md;
2213 uint64_t events_sb;
2214 uint64_t failed_devices[DISKS_ARRAY_ELEMS];
2215 struct dm_raid_superblock *sb;
2216 uint32_t new_devs = 0, rebuild_and_new = 0, rebuilds = 0;
2217 struct md_rdev *r;
2218 struct dm_raid_superblock *sb2;
2219
2220 sb = page_address(rdev->sb_page);
2221 events_sb = le64_to_cpu(sb->events);
2222
2223 /*
2224 * Initialise to 1 if this is a new superblock.
2225 */
2226 mddev->events = events_sb ? : 1;
2227
2228 mddev->reshape_position = MaxSector;
2229
2230 mddev->raid_disks = le32_to_cpu(sb->num_devices);
2231 mddev->level = le32_to_cpu(sb->level);
2232 mddev->layout = le32_to_cpu(sb->layout);
2233 mddev->chunk_sectors = le32_to_cpu(sb->stripe_sectors);
2234
2235 /*
2236 * Reshaping is supported, e.g. reshape_position is valid
2237 * in superblock and superblock content is authoritative.
2238 */
2239 if (le32_to_cpu(sb->compat_features) & FEATURE_FLAG_SUPPORTS_V190) {
2240 /* Superblock is authoritative wrt given raid set layout! */
2241 mddev->new_level = le32_to_cpu(sb->new_level);
2242 mddev->new_layout = le32_to_cpu(sb->new_layout);
2243 mddev->new_chunk_sectors = le32_to_cpu(sb->new_stripe_sectors);
2244 mddev->delta_disks = le32_to_cpu(sb->delta_disks);
2245 mddev->array_sectors = le64_to_cpu(sb->array_sectors);
2246
2247 /* raid was reshaping and got interrupted */
2248 if (le32_to_cpu(sb->flags) & SB_FLAG_RESHAPE_ACTIVE) {
2249 if (test_bit(__CTR_FLAG_DELTA_DISKS, &rs->ctr_flags)) {
2250 DMERR("Reshape requested but raid set is still reshaping");
2251 return -EINVAL;
2252 }
2253
2254 if (mddev->delta_disks < 0 ||
2255 (!mddev->delta_disks && (le32_to_cpu(sb->flags) & SB_FLAG_RESHAPE_BACKWARDS)))
2256 mddev->reshape_backwards = 1;
2257 else
2258 mddev->reshape_backwards = 0;
2259
2260 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
2261 rs->raid_type = get_raid_type_by_ll(mddev->level, mddev->layout);
2262 }
2263
2264 } else {
2265 /*
2266 * No takeover/reshaping, because we don't have the extended v1.9.0 metadata
2267 */
2268 struct raid_type *rt_cur = get_raid_type_by_ll(mddev->level, mddev->layout);
2269 struct raid_type *rt_new = get_raid_type_by_ll(mddev->new_level, mddev->new_layout);
2270
2271 if (rs_takeover_requested(rs)) {
2272 if (rt_cur && rt_new)
2273 DMERR("Takeover raid sets from %s to %s not yet supported by metadata. (raid level change)",
2274 rt_cur->name, rt_new->name);
2275 else
2276 DMERR("Takeover raid sets not yet supported by metadata. (raid level change)");
2277 return -EINVAL;
2278 } else if (rs_reshape_requested(rs)) {
2279 DMERR("Reshaping raid sets not yet supported by metadata. (raid layout change keeping level)");
2280 if (mddev->layout != mddev->new_layout) {
2281 if (rt_cur && rt_new)
2282 DMERR(" current layout %s vs new layout %s",
2283 rt_cur->name, rt_new->name);
2284 else
2285 DMERR(" current layout 0x%X vs new layout 0x%X",
2286 le32_to_cpu(sb->layout), mddev->new_layout);
2287 }
2288 if (mddev->chunk_sectors != mddev->new_chunk_sectors)
2289 DMERR(" current stripe sectors %u vs new stripe sectors %u",
2290 mddev->chunk_sectors, mddev->new_chunk_sectors);
2291 if (rs->delta_disks)
2292 DMERR(" current %u disks vs new %u disks",
2293 mddev->raid_disks, mddev->raid_disks + rs->delta_disks);
2294 if (rs_is_raid10(rs)) {
2295 DMERR(" Old layout: %s w/ %u copies",
2296 raid10_md_layout_to_format(mddev->layout),
2297 raid10_md_layout_to_copies(mddev->layout));
2298 DMERR(" New layout: %s w/ %u copies",
2299 raid10_md_layout_to_format(mddev->new_layout),
2300 raid10_md_layout_to_copies(mddev->new_layout));
2301 }
2302 return -EINVAL;
2303 }
2304
2305 DMINFO("Discovered old metadata format; upgrading to extended metadata format");
2306 }
2307
2308 if (!test_bit(__CTR_FLAG_NOSYNC, &rs->ctr_flags))
2309 mddev->recovery_cp = le64_to_cpu(sb->array_resync_offset);
2310
2311 /*
2312 * During load, we set FirstUse if a new superblock was written.
2313 * There are two reasons we might not have a superblock:
2314 * 1) The raid set is brand new - in which case, all of the
2315 * devices must have their In_sync bit set. Also,
2316 * recovery_cp must be 0, unless forced.
2317 * 2) This is a new device being added to an old raid set
2318 * and the new device needs to be rebuilt - in which
2319 * case the In_sync bit will /not/ be set and
2320 * recovery_cp must be MaxSector.
2321 * 3) This is/are a new device(s) being added to an old
2322 * raid set during takeover to a higher raid level
2323 * to provide capacity for redundancy or during reshape
2324 * to add capacity to grow the raid set.
2325 */
2326 d = 0;
2327 rdev_for_each(r, mddev) {
2328 if (test_bit(Journal, &rdev->flags))
2329 continue;
2330
2331 if (test_bit(FirstUse, &r->flags))
2332 new_devs++;
2333
2334 if (!test_bit(In_sync, &r->flags)) {
2335 DMINFO("Device %d specified for rebuild; clearing superblock",
2336 r->raid_disk);
2337 rebuilds++;
2338
2339 if (test_bit(FirstUse, &r->flags))
2340 rebuild_and_new++;
2341 }
2342
2343 d++;
2344 }
2345
2346 if (new_devs == rs->raid_disks || !rebuilds) {
2347 /* Replace a broken device */
2348 if (new_devs == rs->raid_disks) {
2349 DMINFO("Superblocks created for new raid set");
2350 set_bit(MD_ARRAY_FIRST_USE, &mddev->flags);
2351 } else if (new_devs != rebuilds &&
2352 new_devs != rs->delta_disks) {
2353 DMERR("New device injected into existing raid set without "
2354 "'delta_disks' or 'rebuild' parameter specified");
2355 return -EINVAL;
2356 }
2357 } else if (new_devs && new_devs != rebuilds) {
2358 DMERR("%u 'rebuild' devices cannot be injected into"
2359 " a raid set with %u other first-time devices",
2360 rebuilds, new_devs);
2361 return -EINVAL;
2362 } else if (rebuilds) {
2363 if (rebuild_and_new && rebuilds != rebuild_and_new) {
2364 DMERR("new device%s provided without 'rebuild'",
2365 new_devs > 1 ? "s" : "");
2366 return -EINVAL;
2367 } else if (!test_bit(__CTR_FLAG_REBUILD, &rs->ctr_flags) && rs_is_recovering(rs)) {
2368 DMERR("'rebuild' specified while raid set is not in-sync (recovery_cp=%llu)",
2369 (unsigned long long) mddev->recovery_cp);
2370 return -EINVAL;
2371 } else if (rs_is_reshaping(rs)) {
2372 DMERR("'rebuild' specified while raid set is being reshaped (reshape_position=%llu)",
2373 (unsigned long long) mddev->reshape_position);
2374 return -EINVAL;
2375 }
2376 }
2377
2378 /*
2379 * Now we set the Faulty bit for those devices that are
2380 * recorded in the superblock as failed.
2381 */
2382 sb_retrieve_failed_devices(sb, failed_devices);
2383 rdev_for_each(r, mddev) {
2384 if (test_bit(Journal, &rdev->flags) ||
2385 !r->sb_page)
2386 continue;
2387 sb2 = page_address(r->sb_page);
2388 sb2->failed_devices = 0;
2389 memset(sb2->extended_failed_devices, 0, sizeof(sb2->extended_failed_devices));
2390
2391 /*
2392 * Check for any device re-ordering.
2393 */
2394 if (!test_bit(FirstUse, &r->flags) && (r->raid_disk >= 0)) {
2395 role = le32_to_cpu(sb2->array_position);
2396 if (role < 0)
2397 continue;
2398
2399 if (role != r->raid_disk) {
2400 if (rs_is_raid10(rs) && __is_raid10_near(mddev->layout)) {
2401 if (mddev->raid_disks % __raid10_near_copies(mddev->layout) ||
2402 rs->raid_disks % rs->raid10_copies) {
2403 rs->ti->error =
2404 "Cannot change raid10 near set to odd # of devices!";
2405 return -EINVAL;
2406 }
2407
2408 sb2->array_position = cpu_to_le32(r->raid_disk);
2409
2410 } else if (!(rs_is_raid10(rs) && rt_is_raid0(rs->raid_type)) &&
2411 !(rs_is_raid0(rs) && rt_is_raid10(rs->raid_type)) &&
2412 !rt_is_raid1(rs->raid_type)) {
2413 rs->ti->error = "Cannot change device positions in raid set";
2414 return -EINVAL;
2415 }
2416
2417 DMINFO("raid device #%d now at position #%d", role, r->raid_disk);
2418 }
2419
2420 /*
2421 * Partial recovery is performed on
2422 * returning failed devices.
2423 */
2424 if (test_bit(role, (void *) failed_devices))
2425 set_bit(Faulty, &r->flags);
2426 }
2427 }
2428
2429 return 0;
2430}
2431
2432static int super_validate(struct raid_set *rs, struct md_rdev *rdev)
2433{
2434 struct mddev *mddev = &rs->md;
2435 struct dm_raid_superblock *sb;
2436
2437 if (rs_is_raid0(rs) || !rdev->sb_page || rdev->raid_disk < 0)
2438 return 0;
2439
2440 sb = page_address(rdev->sb_page);
2441
2442 /*
2443 * If mddev->events is not set, we know we have not yet initialized
2444 * the array.
2445 */
2446 if (!mddev->events && super_init_validation(rs, rdev))
2447 return -EINVAL;
2448
2449 if (le32_to_cpu(sb->compat_features) &&
2450 le32_to_cpu(sb->compat_features) != FEATURE_FLAG_SUPPORTS_V190) {
2451 rs->ti->error = "Unable to assemble array: Unknown flag(s) in compatible feature flags";
2452 return -EINVAL;
2453 }
2454
2455 if (sb->incompat_features) {
2456 rs->ti->error = "Unable to assemble array: No incompatible feature flags supported yet";
2457 return -EINVAL;
2458 }
2459
2460 /* Enable bitmap creation on @rs unless no metadevs or raid0 or journaled raid4/5/6 set. */
2461 mddev->bitmap_info.offset = (rt_is_raid0(rs->raid_type) || rs->journal_dev.dev) ? 0 : to_sector(4096);
2462 mddev->bitmap_info.default_offset = mddev->bitmap_info.offset;
2463
2464 if (!test_and_clear_bit(FirstUse, &rdev->flags)) {
2465 /*
2466 * Retrieve rdev size stored in superblock to be prepared for shrink.
2467 * Check extended superblock members are present otherwise the size
2468 * will not be set!
2469 */
2470 if (le32_to_cpu(sb->compat_features) & FEATURE_FLAG_SUPPORTS_V190)
2471 rdev->sectors = le64_to_cpu(sb->sectors);
2472
2473 rdev->recovery_offset = le64_to_cpu(sb->disk_recovery_offset);
2474 if (rdev->recovery_offset == MaxSector)
2475 set_bit(In_sync, &rdev->flags);
2476 /*
2477 * If no reshape in progress -> we're recovering single
2478 * disk(s) and have to set the device(s) to out-of-sync
2479 */
2480 else if (!rs_is_reshaping(rs))
2481 clear_bit(In_sync, &rdev->flags); /* Mandatory for recovery */
2482 }
2483
2484 /*
2485 * If a device comes back, set it as not In_sync and no longer faulty.
2486 */
2487 if (test_and_clear_bit(Faulty, &rdev->flags)) {
2488 rdev->recovery_offset = 0;
2489 clear_bit(In_sync, &rdev->flags);
2490 rdev->saved_raid_disk = rdev->raid_disk;
2491 }
2492
2493 /* Reshape support -> restore repective data offsets */
2494 rdev->data_offset = le64_to_cpu(sb->data_offset);
2495 rdev->new_data_offset = le64_to_cpu(sb->new_data_offset);
2496
2497 return 0;
2498}
2499
2500/*
2501 * Analyse superblocks and select the freshest.
2502 */
2503static int analyse_superblocks(struct dm_target *ti, struct raid_set *rs)
2504{
2505 int r;
2506 struct md_rdev *rdev, *freshest;
2507 struct mddev *mddev = &rs->md;
2508
2509 freshest = NULL;
2510 rdev_for_each(rdev, mddev) {
2511 if (test_bit(Journal, &rdev->flags))
2512 continue;
2513
2514 if (!rdev->meta_bdev)
2515 continue;
2516
2517 /* Set superblock offset/size for metadata device. */
2518 rdev->sb_start = 0;
2519 rdev->sb_size = bdev_logical_block_size(rdev->meta_bdev);
2520 if (rdev->sb_size < sizeof(struct dm_raid_superblock) || rdev->sb_size > PAGE_SIZE) {
2521 DMERR("superblock size of a logical block is no longer valid");
2522 return -EINVAL;
2523 }
2524
2525 /*
2526 * Skipping super_load due to CTR_FLAG_SYNC will cause
2527 * the array to undergo initialization again as
2528 * though it were new. This is the intended effect
2529 * of the "sync" directive.
2530 *
2531 * With reshaping capability added, we must ensure that
2532 * the "sync" directive is disallowed during the reshape.
2533 */
2534 if (test_bit(__CTR_FLAG_SYNC, &rs->ctr_flags))
2535 continue;
2536
2537 r = super_load(rdev, freshest);
2538
2539 switch (r) {
2540 case 1:
2541 freshest = rdev;
2542 break;
2543 case 0:
2544 break;
2545 default:
2546 /* This is a failure to read the superblock from the metadata device. */
2547 /*
2548 * We have to keep any raid0 data/metadata device pairs or
2549 * the MD raid0 personality will fail to start the array.
2550 */
2551 if (rs_is_raid0(rs))
2552 continue;
2553
2554 /*
2555 * We keep the dm_devs to be able to emit the device tuple
2556 * properly on the table line in raid_status() (rather than
2557 * mistakenly acting as if '- -' got passed into the constructor).
2558 *
2559 * The rdev has to stay on the same_set list to allow for
2560 * the attempt to restore faulty devices on second resume.
2561 */
2562 rdev->raid_disk = rdev->saved_raid_disk = -1;
2563 break;
2564 }
2565 }
2566
2567 if (!freshest)
2568 return 0;
2569
2570 /*
2571 * Validation of the freshest device provides the source of
2572 * validation for the remaining devices.
2573 */
2574 rs->ti->error = "Unable to assemble array: Invalid superblocks";
2575 if (super_validate(rs, freshest))
2576 return -EINVAL;
2577
2578 if (validate_raid_redundancy(rs)) {
2579 rs->ti->error = "Insufficient redundancy to activate array";
2580 return -EINVAL;
2581 }
2582
2583 rdev_for_each(rdev, mddev)
2584 if (!test_bit(Journal, &rdev->flags) &&
2585 rdev != freshest &&
2586 super_validate(rs, rdev))
2587 return -EINVAL;
2588 return 0;
2589}
2590
2591/*
2592 * Adjust data_offset and new_data_offset on all disk members of @rs
2593 * for out of place reshaping if requested by constructor
2594 *
2595 * We need free space at the beginning of each raid disk for forward
2596 * and at the end for backward reshapes which userspace has to provide
2597 * via remapping/reordering of space.
2598 */
2599static int rs_adjust_data_offsets(struct raid_set *rs)
2600{
2601 sector_t data_offset = 0, new_data_offset = 0;
2602 struct md_rdev *rdev;
2603
2604 /* Constructor did not request data offset change */
2605 if (!test_bit(__CTR_FLAG_DATA_OFFSET, &rs->ctr_flags)) {
2606 if (!rs_is_reshapable(rs))
2607 goto out;
2608
2609 return 0;
2610 }
2611
2612 /* HM FIXME: get In_Sync raid_dev? */
2613 rdev = &rs->dev[0].rdev;
2614
2615 if (rs->delta_disks < 0) {
2616 /*
2617 * Removing disks (reshaping backwards):
2618 *
2619 * - before reshape: data is at offset 0 and free space
2620 * is at end of each component LV
2621 *
2622 * - after reshape: data is at offset rs->data_offset != 0 on each component LV
2623 */
2624 data_offset = 0;
2625 new_data_offset = rs->data_offset;
2626
2627 } else if (rs->delta_disks > 0) {
2628 /*
2629 * Adding disks (reshaping forwards):
2630 *
2631 * - before reshape: data is at offset rs->data_offset != 0 and
2632 * free space is at begin of each component LV
2633 *
2634 * - after reshape: data is at offset 0 on each component LV
2635 */
2636 data_offset = rs->data_offset;
2637 new_data_offset = 0;
2638
2639 } else {
2640 /*
2641 * User space passes in 0 for data offset after having removed reshape space
2642 *
2643 * - or - (data offset != 0)
2644 *
2645 * Changing RAID layout or chunk size -> toggle offsets
2646 *
2647 * - before reshape: data is at offset rs->data_offset 0 and
2648 * free space is at end of each component LV
2649 * -or-
2650 * data is at offset rs->data_offset != 0 and
2651 * free space is at begin of each component LV
2652 *
2653 * - after reshape: data is at offset 0 if it was at offset != 0
2654 * or at offset != 0 if it was at offset 0
2655 * on each component LV
2656 *
2657 */
2658 data_offset = rs->data_offset ? rdev->data_offset : 0;
2659 new_data_offset = data_offset ? 0 : rs->data_offset;
2660 set_bit(RT_FLAG_UPDATE_SBS, &rs->runtime_flags);
2661 }
2662
2663 /*
2664 * Make sure we got a minimum amount of free sectors per device
2665 */
2666 if (rs->data_offset &&
2667 bdev_nr_sectors(rdev->bdev) - rs->md.dev_sectors < MIN_FREE_RESHAPE_SPACE) {
2668 rs->ti->error = data_offset ? "No space for forward reshape" :
2669 "No space for backward reshape";
2670 return -ENOSPC;
2671 }
2672out:
2673 /*
2674 * Raise recovery_cp in case data_offset != 0 to
2675 * avoid false recovery positives in the constructor.
2676 */
2677 if (rs->md.recovery_cp < rs->md.dev_sectors)
2678 rs->md.recovery_cp += rs->dev[0].rdev.data_offset;
2679
2680 /* Adjust data offsets on all rdevs but on any raid4/5/6 journal device */
2681 rdev_for_each(rdev, &rs->md) {
2682 if (!test_bit(Journal, &rdev->flags)) {
2683 rdev->data_offset = data_offset;
2684 rdev->new_data_offset = new_data_offset;
2685 }
2686 }
2687
2688 return 0;
2689}
2690
2691/* Userpace reordered disks -> adjust raid_disk indexes in @rs */
2692static void __reorder_raid_disk_indexes(struct raid_set *rs)
2693{
2694 int i = 0;
2695 struct md_rdev *rdev;
2696
2697 rdev_for_each(rdev, &rs->md) {
2698 if (!test_bit(Journal, &rdev->flags)) {
2699 rdev->raid_disk = i++;
2700 rdev->saved_raid_disk = rdev->new_raid_disk = -1;
2701 }
2702 }
2703}
2704
2705/*
2706 * Setup @rs for takeover by a different raid level
2707 */
2708static int rs_setup_takeover(struct raid_set *rs)
2709{
2710 struct mddev *mddev = &rs->md;
2711 struct md_rdev *rdev;
2712 unsigned int d = mddev->raid_disks = rs->raid_disks;
2713 sector_t new_data_offset = rs->dev[0].rdev.data_offset ? 0 : rs->data_offset;
2714
2715 if (rt_is_raid10(rs->raid_type)) {
2716 if (rs_is_raid0(rs)) {
2717 /* Userpace reordered disks -> adjust raid_disk indexes */
2718 __reorder_raid_disk_indexes(rs);
2719
2720 /* raid0 -> raid10_far layout */
2721 mddev->layout = raid10_format_to_md_layout(rs, ALGORITHM_RAID10_FAR,
2722 rs->raid10_copies);
2723 } else if (rs_is_raid1(rs))
2724 /* raid1 -> raid10_near layout */
2725 mddev->layout = raid10_format_to_md_layout(rs, ALGORITHM_RAID10_NEAR,
2726 rs->raid_disks);
2727 else
2728 return -EINVAL;
2729
2730 }
2731
2732 clear_bit(MD_ARRAY_FIRST_USE, &mddev->flags);
2733 mddev->recovery_cp = MaxSector;
2734
2735 while (d--) {
2736 rdev = &rs->dev[d].rdev;
2737
2738 if (test_bit(d, (void *) rs->rebuild_disks)) {
2739 clear_bit(In_sync, &rdev->flags);
2740 clear_bit(Faulty, &rdev->flags);
2741 mddev->recovery_cp = rdev->recovery_offset = 0;
2742 /* Bitmap has to be created when we do an "up" takeover */
2743 set_bit(MD_ARRAY_FIRST_USE, &mddev->flags);
2744 }
2745
2746 rdev->new_data_offset = new_data_offset;
2747 }
2748
2749 return 0;
2750}
2751
2752/* Prepare @rs for reshape */
2753static int rs_prepare_reshape(struct raid_set *rs)
2754{
2755 bool reshape;
2756 struct mddev *mddev = &rs->md;
2757
2758 if (rs_is_raid10(rs)) {
2759 if (rs->raid_disks != mddev->raid_disks &&
2760 __is_raid10_near(mddev->layout) &&
2761 rs->raid10_copies &&
2762 rs->raid10_copies != __raid10_near_copies(mddev->layout)) {
2763 /*
2764 * raid disk have to be multiple of data copies to allow this conversion,
2765 *
2766 * This is actually not a reshape it is a
2767 * rebuild of any additional mirrors per group
2768 */
2769 if (rs->raid_disks % rs->raid10_copies) {
2770 rs->ti->error = "Can't reshape raid10 mirror groups";
2771 return -EINVAL;
2772 }
2773
2774 /* Userpace reordered disks to add/remove mirrors -> adjust raid_disk indexes */
2775 __reorder_raid_disk_indexes(rs);
2776 mddev->layout = raid10_format_to_md_layout(rs, ALGORITHM_RAID10_NEAR,
2777 rs->raid10_copies);
2778 mddev->new_layout = mddev->layout;
2779 reshape = false;
2780 } else
2781 reshape = true;
2782
2783 } else if (rs_is_raid456(rs))
2784 reshape = true;
2785
2786 else if (rs_is_raid1(rs)) {
2787 if (rs->delta_disks) {
2788 /* Process raid1 via delta_disks */
2789 mddev->degraded = rs->delta_disks < 0 ? -rs->delta_disks : rs->delta_disks;
2790 reshape = true;
2791 } else {
2792 /* Process raid1 without delta_disks */
2793 mddev->raid_disks = rs->raid_disks;
2794 reshape = false;
2795 }
2796 } else {
2797 rs->ti->error = "Called with bogus raid type";
2798 return -EINVAL;
2799 }
2800
2801 if (reshape) {
2802 set_bit(RT_FLAG_RESHAPE_RS, &rs->runtime_flags);
2803 set_bit(RT_FLAG_UPDATE_SBS, &rs->runtime_flags);
2804 } else if (mddev->raid_disks < rs->raid_disks)
2805 /* Create new superblocks and bitmaps, if any new disks */
2806 set_bit(RT_FLAG_UPDATE_SBS, &rs->runtime_flags);
2807
2808 return 0;
2809}
2810
2811/* Get reshape sectors from data_offsets or raid set */
2812static sector_t _get_reshape_sectors(struct raid_set *rs)
2813{
2814 struct md_rdev *rdev;
2815 sector_t reshape_sectors = 0;
2816
2817 rdev_for_each(rdev, &rs->md)
2818 if (!test_bit(Journal, &rdev->flags)) {
2819 reshape_sectors = (rdev->data_offset > rdev->new_data_offset) ?
2820 rdev->data_offset - rdev->new_data_offset :
2821 rdev->new_data_offset - rdev->data_offset;
2822 break;
2823 }
2824
2825 return max(reshape_sectors, (sector_t) rs->data_offset);
2826}
2827
2828/*
2829 * Reshape:
2830 * - change raid layout
2831 * - change chunk size
2832 * - add disks
2833 * - remove disks
2834 */
2835static int rs_setup_reshape(struct raid_set *rs)
2836{
2837 int r = 0;
2838 unsigned int cur_raid_devs, d;
2839 sector_t reshape_sectors = _get_reshape_sectors(rs);
2840 struct mddev *mddev = &rs->md;
2841 struct md_rdev *rdev;
2842
2843 mddev->delta_disks = rs->delta_disks;
2844 cur_raid_devs = mddev->raid_disks;
2845
2846 /* Ignore impossible layout change whilst adding/removing disks */
2847 if (mddev->delta_disks &&
2848 mddev->layout != mddev->new_layout) {
2849 DMINFO("Ignoring invalid layout change with delta_disks=%d", rs->delta_disks);
2850 mddev->new_layout = mddev->layout;
2851 }
2852
2853 /*
2854 * Adjust array size:
2855 *
2856 * - in case of adding disk(s), array size has
2857 * to grow after the disk adding reshape,
2858 * which'll hapen in the event handler;
2859 * reshape will happen forward, so space has to
2860 * be available at the beginning of each disk
2861 *
2862 * - in case of removing disk(s), array size
2863 * has to shrink before starting the reshape,
2864 * which'll happen here;
2865 * reshape will happen backward, so space has to
2866 * be available at the end of each disk
2867 *
2868 * - data_offset and new_data_offset are
2869 * adjusted for aforementioned out of place
2870 * reshaping based on userspace passing in
2871 * the "data_offset <sectors>" key/value
2872 * pair via the constructor
2873 */
2874
2875 /* Add disk(s) */
2876 if (rs->delta_disks > 0) {
2877 /* Prepare disks for check in raid4/5/6/10 {check|start}_reshape */
2878 for (d = cur_raid_devs; d < rs->raid_disks; d++) {
2879 rdev = &rs->dev[d].rdev;
2880 clear_bit(In_sync, &rdev->flags);
2881
2882 /*
2883 * save_raid_disk needs to be -1, or recovery_offset will be set to 0
2884 * by md, which'll store that erroneously in the superblock on reshape
2885 */
2886 rdev->saved_raid_disk = -1;
2887 rdev->raid_disk = d;
2888
2889 rdev->sectors = mddev->dev_sectors;
2890 rdev->recovery_offset = rs_is_raid1(rs) ? 0 : MaxSector;
2891 }
2892
2893 mddev->reshape_backwards = 0; /* adding disk(s) -> forward reshape */
2894
2895 /* Remove disk(s) */
2896 } else if (rs->delta_disks < 0) {
2897 r = rs_set_dev_and_array_sectors(rs, rs->ti->len, true);
2898 mddev->reshape_backwards = 1; /* removing disk(s) -> backward reshape */
2899
2900 /* Change layout and/or chunk size */
2901 } else {
2902 /*
2903 * Reshape layout (e.g. raid5_ls -> raid5_n) and/or chunk size:
2904 *
2905 * keeping number of disks and do layout change ->
2906 *
2907 * toggle reshape_backward depending on data_offset:
2908 *
2909 * - free space upfront -> reshape forward
2910 *
2911 * - free space at the end -> reshape backward
2912 *
2913 *
2914 * This utilizes free reshape space avoiding the need
2915 * for userspace to move (parts of) LV segments in
2916 * case of layout/chunksize change (for disk
2917 * adding/removing reshape space has to be at
2918 * the proper address (see above with delta_disks):
2919 *
2920 * add disk(s) -> begin
2921 * remove disk(s)-> end
2922 */
2923 mddev->reshape_backwards = rs->dev[0].rdev.data_offset ? 0 : 1;
2924 }
2925
2926 /*
2927 * Adjust device size for forward reshape
2928 * because md_finish_reshape() reduces it.
2929 */
2930 if (!mddev->reshape_backwards)
2931 rdev_for_each(rdev, &rs->md)
2932 if (!test_bit(Journal, &rdev->flags))
2933 rdev->sectors += reshape_sectors;
2934
2935 return r;
2936}
2937
2938/*
2939 * If the md resync thread has updated superblock with max reshape position
2940 * at the end of a reshape but not (yet) reset the layout configuration
2941 * changes -> reset the latter.
2942 */
2943static void rs_reset_inconclusive_reshape(struct raid_set *rs)
2944{
2945 if (!rs_is_reshaping(rs) && rs_is_layout_change(rs, true)) {
2946 rs_set_cur(rs);
2947 rs->md.delta_disks = 0;
2948 rs->md.reshape_backwards = 0;
2949 }
2950}
2951
2952/*
2953 * Enable/disable discard support on RAID set depending on
2954 * RAID level and discard properties of underlying RAID members.
2955 */
2956static void configure_discard_support(struct raid_set *rs)
2957{
2958 int i;
2959 bool raid456;
2960 struct dm_target *ti = rs->ti;
2961
2962 /*
2963 * XXX: RAID level 4,5,6 require zeroing for safety.
2964 */
2965 raid456 = rs_is_raid456(rs);
2966
2967 for (i = 0; i < rs->raid_disks; i++) {
2968 if (!rs->dev[i].rdev.bdev ||
2969 !bdev_max_discard_sectors(rs->dev[i].rdev.bdev))
2970 return;
2971
2972 if (raid456) {
2973 if (!devices_handle_discard_safely) {
2974 DMERR("raid456 discard support disabled due to discard_zeroes_data uncertainty.");
2975 DMERR("Set dm-raid.devices_handle_discard_safely=Y to override.");
2976 return;
2977 }
2978 }
2979 }
2980
2981 ti->num_discard_bios = 1;
2982}
2983
2984/*
2985 * Construct a RAID0/1/10/4/5/6 mapping:
2986 * Args:
2987 * <raid_type> <#raid_params> <raid_params>{0,} \
2988 * <#raid_devs> [<meta_dev1> <dev1>]{1,}
2989 *
2990 * <raid_params> varies by <raid_type>. See 'parse_raid_params' for
2991 * details on possible <raid_params>.
2992 *
2993 * Userspace is free to initialize the metadata devices, hence the superblocks to
2994 * enforce recreation based on the passed in table parameters.
2995 *
2996 */
2997static int raid_ctr(struct dm_target *ti, unsigned int argc, char **argv)
2998{
2999 int r;
3000 bool resize = false;
3001 struct raid_type *rt;
3002 unsigned int num_raid_params, num_raid_devs;
3003 sector_t sb_array_sectors, rdev_sectors, reshape_sectors;
3004 struct raid_set *rs = NULL;
3005 const char *arg;
3006 struct rs_layout rs_layout;
3007 struct dm_arg_set as = { argc, argv }, as_nrd;
3008 struct dm_arg _args[] = {
3009 { 0, as.argc, "Cannot understand number of raid parameters" },
3010 { 1, 254, "Cannot understand number of raid devices parameters" }
3011 };
3012
3013 arg = dm_shift_arg(&as);
3014 if (!arg) {
3015 ti->error = "No arguments";
3016 return -EINVAL;
3017 }
3018
3019 rt = get_raid_type(arg);
3020 if (!rt) {
3021 ti->error = "Unrecognised raid_type";
3022 return -EINVAL;
3023 }
3024
3025 /* Must have <#raid_params> */
3026 if (dm_read_arg_group(_args, &as, &num_raid_params, &ti->error))
3027 return -EINVAL;
3028
3029 /* number of raid device tupples <meta_dev data_dev> */
3030 as_nrd = as;
3031 dm_consume_args(&as_nrd, num_raid_params);
3032 _args[1].max = (as_nrd.argc - 1) / 2;
3033 if (dm_read_arg(_args + 1, &as_nrd, &num_raid_devs, &ti->error))
3034 return -EINVAL;
3035
3036 if (!__within_range(num_raid_devs, 1, MAX_RAID_DEVICES)) {
3037 ti->error = "Invalid number of supplied raid devices";
3038 return -EINVAL;
3039 }
3040
3041 rs = raid_set_alloc(ti, rt, num_raid_devs);
3042 if (IS_ERR(rs))
3043 return PTR_ERR(rs);
3044
3045 r = parse_raid_params(rs, &as, num_raid_params);
3046 if (r)
3047 goto bad;
3048
3049 r = parse_dev_params(rs, &as);
3050 if (r)
3051 goto bad;
3052
3053 rs->md.sync_super = super_sync;
3054
3055 /*
3056 * Calculate ctr requested array and device sizes to allow
3057 * for superblock analysis needing device sizes defined.
3058 *
3059 * Any existing superblock will overwrite the array and device sizes
3060 */
3061 r = rs_set_dev_and_array_sectors(rs, rs->ti->len, false);
3062 if (r)
3063 goto bad;
3064
3065 /* Memorize just calculated, potentially larger sizes to grow the raid set in preresume */
3066 rs->array_sectors = rs->md.array_sectors;
3067 rs->dev_sectors = rs->md.dev_sectors;
3068
3069 /*
3070 * Backup any new raid set level, layout, ...
3071 * requested to be able to compare to superblock
3072 * members for conversion decisions.
3073 */
3074 rs_config_backup(rs, &rs_layout);
3075
3076 r = analyse_superblocks(ti, rs);
3077 if (r)
3078 goto bad;
3079
3080 /* All in-core metadata now as of current superblocks after calling analyse_superblocks() */
3081 sb_array_sectors = rs->md.array_sectors;
3082 rdev_sectors = __rdev_sectors(rs);
3083 if (!rdev_sectors) {
3084 ti->error = "Invalid rdev size";
3085 r = -EINVAL;
3086 goto bad;
3087 }
3088
3089
3090 reshape_sectors = _get_reshape_sectors(rs);
3091 if (rs->dev_sectors != rdev_sectors) {
3092 resize = (rs->dev_sectors != rdev_sectors - reshape_sectors);
3093 if (rs->dev_sectors > rdev_sectors - reshape_sectors)
3094 set_bit(RT_FLAG_RS_GROW, &rs->runtime_flags);
3095 }
3096
3097 INIT_WORK(&rs->md.event_work, do_table_event);
3098 ti->private = rs;
3099 ti->num_flush_bios = 1;
3100 ti->needs_bio_set_dev = true;
3101
3102 /* Restore any requested new layout for conversion decision */
3103 rs_config_restore(rs, &rs_layout);
3104
3105 /*
3106 * Now that we have any superblock metadata available,
3107 * check for new, recovering, reshaping, to be taken over,
3108 * to be reshaped or an existing, unchanged raid set to
3109 * run in sequence.
3110 */
3111 if (test_bit(MD_ARRAY_FIRST_USE, &rs->md.flags)) {
3112 /* A new raid6 set has to be recovered to ensure proper parity and Q-Syndrome */
3113 if (rs_is_raid6(rs) &&
3114 test_bit(__CTR_FLAG_NOSYNC, &rs->ctr_flags)) {
3115 ti->error = "'nosync' not allowed for new raid6 set";
3116 r = -EINVAL;
3117 goto bad;
3118 }
3119 rs_setup_recovery(rs, 0);
3120 set_bit(RT_FLAG_UPDATE_SBS, &rs->runtime_flags);
3121 rs_set_new(rs);
3122 } else if (rs_is_recovering(rs)) {
3123 /* A recovering raid set may be resized */
3124 goto size_check;
3125 } else if (rs_is_reshaping(rs)) {
3126 /* Have to reject size change request during reshape */
3127 if (resize) {
3128 ti->error = "Can't resize a reshaping raid set";
3129 r = -EPERM;
3130 goto bad;
3131 }
3132 /* skip setup rs */
3133 } else if (rs_takeover_requested(rs)) {
3134 if (rs_is_reshaping(rs)) {
3135 ti->error = "Can't takeover a reshaping raid set";
3136 r = -EPERM;
3137 goto bad;
3138 }
3139
3140 /* We can't takeover a journaled raid4/5/6 */
3141 if (test_bit(__CTR_FLAG_JOURNAL_DEV, &rs->ctr_flags)) {
3142 ti->error = "Can't takeover a journaled raid4/5/6 set";
3143 r = -EPERM;
3144 goto bad;
3145 }
3146
3147 /*
3148 * If a takeover is needed, userspace sets any additional
3149 * devices to rebuild and we can check for a valid request here.
3150 *
3151 * If acceptible, set the level to the new requested
3152 * one, prohibit requesting recovery, allow the raid
3153 * set to run and store superblocks during resume.
3154 */
3155 r = rs_check_takeover(rs);
3156 if (r)
3157 goto bad;
3158
3159 r = rs_setup_takeover(rs);
3160 if (r)
3161 goto bad;
3162
3163 set_bit(RT_FLAG_UPDATE_SBS, &rs->runtime_flags);
3164 /* Takeover ain't recovery, so disable recovery */
3165 rs_setup_recovery(rs, MaxSector);
3166 rs_set_new(rs);
3167 } else if (rs_reshape_requested(rs)) {
3168 /* Only request grow on raid set size extensions, not on reshapes. */
3169 clear_bit(RT_FLAG_RS_GROW, &rs->runtime_flags);
3170
3171 /*
3172 * No need to check for 'ongoing' takeover here, because takeover
3173 * is an instant operation as oposed to an ongoing reshape.
3174 */
3175
3176 /* We can't reshape a journaled raid4/5/6 */
3177 if (test_bit(__CTR_FLAG_JOURNAL_DEV, &rs->ctr_flags)) {
3178 ti->error = "Can't reshape a journaled raid4/5/6 set";
3179 r = -EPERM;
3180 goto bad;
3181 }
3182
3183 /* Out-of-place space has to be available to allow for a reshape unless raid1! */
3184 if (reshape_sectors || rs_is_raid1(rs)) {
3185 /*
3186 * We can only prepare for a reshape here, because the
3187 * raid set needs to run to provide the repective reshape
3188 * check functions via its MD personality instance.
3189 *
3190 * So do the reshape check after md_run() succeeded.
3191 */
3192 r = rs_prepare_reshape(rs);
3193 if (r)
3194 goto bad;
3195
3196 /* Reshaping ain't recovery, so disable recovery */
3197 rs_setup_recovery(rs, MaxSector);
3198 }
3199 rs_set_cur(rs);
3200 } else {
3201size_check:
3202 /* May not set recovery when a device rebuild is requested */
3203 if (test_bit(__CTR_FLAG_REBUILD, &rs->ctr_flags)) {
3204 clear_bit(RT_FLAG_RS_GROW, &rs->runtime_flags);
3205 set_bit(RT_FLAG_UPDATE_SBS, &rs->runtime_flags);
3206 rs_setup_recovery(rs, MaxSector);
3207 } else if (test_bit(RT_FLAG_RS_GROW, &rs->runtime_flags)) {
3208 /*
3209 * Set raid set to current size, i.e. size as of
3210 * superblocks to grow to larger size in preresume.
3211 */
3212 r = rs_set_dev_and_array_sectors(rs, sb_array_sectors, false);
3213 if (r)
3214 goto bad;
3215
3216 rs_setup_recovery(rs, rs->md.recovery_cp < rs->md.dev_sectors ? rs->md.recovery_cp : rs->md.dev_sectors);
3217 } else {
3218 /* This is no size change or it is shrinking, update size and record in superblocks */
3219 r = rs_set_dev_and_array_sectors(rs, rs->ti->len, false);
3220 if (r)
3221 goto bad;
3222
3223 if (sb_array_sectors > rs->array_sectors)
3224 set_bit(RT_FLAG_UPDATE_SBS, &rs->runtime_flags);
3225 }
3226 rs_set_cur(rs);
3227 }
3228
3229 /* If constructor requested it, change data and new_data offsets */
3230 r = rs_adjust_data_offsets(rs);
3231 if (r)
3232 goto bad;
3233
3234 /* Catch any inconclusive reshape superblock content. */
3235 rs_reset_inconclusive_reshape(rs);
3236
3237 /* Start raid set read-only and assumed clean to change in raid_resume() */
3238 rs->md.ro = 1;
3239 rs->md.in_sync = 1;
3240
3241 /* Keep array frozen until resume. */
3242 set_bit(MD_RECOVERY_FROZEN, &rs->md.recovery);
3243
3244 /* Has to be held on running the array */
3245 mddev_lock_nointr(&rs->md);
3246 r = md_run(&rs->md);
3247 rs->md.in_sync = 0; /* Assume already marked dirty */
3248 if (r) {
3249 ti->error = "Failed to run raid array";
3250 mddev_unlock(&rs->md);
3251 goto bad;
3252 }
3253
3254 r = md_start(&rs->md);
3255 if (r) {
3256 ti->error = "Failed to start raid array";
3257 mddev_unlock(&rs->md);
3258 goto bad_md_start;
3259 }
3260
3261 /* If raid4/5/6 journal mode explicitly requested (only possible with journal dev) -> set it */
3262 if (test_bit(__CTR_FLAG_JOURNAL_MODE, &rs->ctr_flags)) {
3263 r = r5c_journal_mode_set(&rs->md, rs->journal_dev.mode);
3264 if (r) {
3265 ti->error = "Failed to set raid4/5/6 journal mode";
3266 mddev_unlock(&rs->md);
3267 goto bad_journal_mode_set;
3268 }
3269 }
3270
3271 mddev_suspend(&rs->md);
3272 set_bit(RT_FLAG_RS_SUSPENDED, &rs->runtime_flags);
3273
3274 /* Try to adjust the raid4/5/6 stripe cache size to the stripe size */
3275 if (rs_is_raid456(rs)) {
3276 r = rs_set_raid456_stripe_cache(rs);
3277 if (r)
3278 goto bad_stripe_cache;
3279 }
3280
3281 /* Now do an early reshape check */
3282 if (test_bit(RT_FLAG_RESHAPE_RS, &rs->runtime_flags)) {
3283 r = rs_check_reshape(rs);
3284 if (r)
3285 goto bad_check_reshape;
3286
3287 /* Restore new, ctr requested layout to perform check */
3288 rs_config_restore(rs, &rs_layout);
3289
3290 if (rs->md.pers->start_reshape) {
3291 r = rs->md.pers->check_reshape(&rs->md);
3292 if (r) {
3293 ti->error = "Reshape check failed";
3294 goto bad_check_reshape;
3295 }
3296 }
3297 }
3298
3299 /* Disable/enable discard support on raid set. */
3300 configure_discard_support(rs);
3301
3302 mddev_unlock(&rs->md);
3303 return 0;
3304
3305bad_md_start:
3306bad_journal_mode_set:
3307bad_stripe_cache:
3308bad_check_reshape:
3309 md_stop(&rs->md);
3310bad:
3311 raid_set_free(rs);
3312
3313 return r;
3314}
3315
3316static void raid_dtr(struct dm_target *ti)
3317{
3318 struct raid_set *rs = ti->private;
3319
3320 md_stop(&rs->md);
3321 raid_set_free(rs);
3322}
3323
3324static int raid_map(struct dm_target *ti, struct bio *bio)
3325{
3326 struct raid_set *rs = ti->private;
3327 struct mddev *mddev = &rs->md;
3328
3329 /*
3330 * If we're reshaping to add disk(s)), ti->len and
3331 * mddev->array_sectors will differ during the process
3332 * (ti->len > mddev->array_sectors), so we have to requeue
3333 * bios with addresses > mddev->array_sectors here or
3334 * there will occur accesses past EOD of the component
3335 * data images thus erroring the raid set.
3336 */
3337 if (unlikely(bio_end_sector(bio) > mddev->array_sectors))
3338 return DM_MAPIO_REQUEUE;
3339
3340 md_handle_request(mddev, bio);
3341
3342 return DM_MAPIO_SUBMITTED;
3343}
3344
3345/* Return sync state string for @state */
3346enum sync_state { st_frozen, st_reshape, st_resync, st_check, st_repair, st_recover, st_idle };
3347static const char *sync_str(enum sync_state state)
3348{
3349 /* Has to be in above sync_state order! */
3350 static const char *sync_strs[] = {
3351 "frozen",
3352 "reshape",
3353 "resync",
3354 "check",
3355 "repair",
3356 "recover",
3357 "idle"
3358 };
3359
3360 return __within_range(state, 0, ARRAY_SIZE(sync_strs) - 1) ? sync_strs[state] : "undef";
3361};
3362
3363/* Return enum sync_state for @mddev derived from @recovery flags */
3364static enum sync_state decipher_sync_action(struct mddev *mddev, unsigned long recovery)
3365{
3366 if (test_bit(MD_RECOVERY_FROZEN, &recovery))
3367 return st_frozen;
3368
3369 /* The MD sync thread can be done with io or be interrupted but still be running */
3370 if (!test_bit(MD_RECOVERY_DONE, &recovery) &&
3371 (test_bit(MD_RECOVERY_RUNNING, &recovery) ||
3372 (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &recovery)))) {
3373 if (test_bit(MD_RECOVERY_RESHAPE, &recovery))
3374 return st_reshape;
3375
3376 if (test_bit(MD_RECOVERY_SYNC, &recovery)) {
3377 if (!test_bit(MD_RECOVERY_REQUESTED, &recovery))
3378 return st_resync;
3379 if (test_bit(MD_RECOVERY_CHECK, &recovery))
3380 return st_check;
3381 return st_repair;
3382 }
3383
3384 if (test_bit(MD_RECOVERY_RECOVER, &recovery))
3385 return st_recover;
3386
3387 if (mddev->reshape_position != MaxSector)
3388 return st_reshape;
3389 }
3390
3391 return st_idle;
3392}
3393
3394/*
3395 * Return status string for @rdev
3396 *
3397 * Status characters:
3398 *
3399 * 'D' = Dead/Failed raid set component or raid4/5/6 journal device
3400 * 'a' = Alive but not in-sync raid set component _or_ alive raid4/5/6 'write_back' journal device
3401 * 'A' = Alive and in-sync raid set component _or_ alive raid4/5/6 'write_through' journal device
3402 * '-' = Non-existing device (i.e. uspace passed '- -' into the ctr)
3403 */
3404static const char *__raid_dev_status(struct raid_set *rs, struct md_rdev *rdev)
3405{
3406 if (!rdev->bdev)
3407 return "-";
3408 else if (test_bit(Faulty, &rdev->flags))
3409 return "D";
3410 else if (test_bit(Journal, &rdev->flags))
3411 return (rs->journal_dev.mode == R5C_JOURNAL_MODE_WRITE_THROUGH) ? "A" : "a";
3412 else if (test_bit(RT_FLAG_RS_RESYNCING, &rs->runtime_flags) ||
3413 (!test_bit(RT_FLAG_RS_IN_SYNC, &rs->runtime_flags) &&
3414 !test_bit(In_sync, &rdev->flags)))
3415 return "a";
3416 else
3417 return "A";
3418}
3419
3420/* Helper to return resync/reshape progress for @rs and runtime flags for raid set in sync / resynching */
3421static sector_t rs_get_progress(struct raid_set *rs, unsigned long recovery,
3422 enum sync_state state, sector_t resync_max_sectors)
3423{
3424 sector_t r;
3425 struct mddev *mddev = &rs->md;
3426
3427 clear_bit(RT_FLAG_RS_IN_SYNC, &rs->runtime_flags);
3428 clear_bit(RT_FLAG_RS_RESYNCING, &rs->runtime_flags);
3429
3430 if (rs_is_raid0(rs)) {
3431 r = resync_max_sectors;
3432 set_bit(RT_FLAG_RS_IN_SYNC, &rs->runtime_flags);
3433
3434 } else {
3435 if (state == st_idle && !test_bit(MD_RECOVERY_INTR, &recovery))
3436 r = mddev->recovery_cp;
3437 else
3438 r = mddev->curr_resync_completed;
3439
3440 if (state == st_idle && r >= resync_max_sectors) {
3441 /*
3442 * Sync complete.
3443 */
3444 /* In case we have finished recovering, the array is in sync. */
3445 if (test_bit(MD_RECOVERY_RECOVER, &recovery))
3446 set_bit(RT_FLAG_RS_IN_SYNC, &rs->runtime_flags);
3447
3448 } else if (state == st_recover)
3449 /*
3450 * In case we are recovering, the array is not in sync
3451 * and health chars should show the recovering legs.
3452 *
3453 * Already retrieved recovery offset from curr_resync_completed above.
3454 */
3455 ;
3456
3457 else if (state == st_resync || state == st_reshape)
3458 /*
3459 * If "resync/reshape" is occurring, the raid set
3460 * is or may be out of sync hence the health
3461 * characters shall be 'a'.
3462 */
3463 set_bit(RT_FLAG_RS_RESYNCING, &rs->runtime_flags);
3464
3465 else if (state == st_check || state == st_repair)
3466 /*
3467 * If "check" or "repair" is occurring, the raid set has
3468 * undergone an initial sync and the health characters
3469 * should not be 'a' anymore.
3470 */
3471 set_bit(RT_FLAG_RS_IN_SYNC, &rs->runtime_flags);
3472
3473 else if (test_bit(MD_RECOVERY_NEEDED, &recovery))
3474 /*
3475 * We are idle and recovery is needed, prevent 'A' chars race
3476 * caused by components still set to in-sync by constructor.
3477 */
3478 set_bit(RT_FLAG_RS_RESYNCING, &rs->runtime_flags);
3479
3480 else {
3481 /*
3482 * We are idle and the raid set may be doing an initial
3483 * sync, or it may be rebuilding individual components.
3484 * If all the devices are In_sync, then it is the raid set
3485 * that is being initialized.
3486 */
3487 struct md_rdev *rdev;
3488
3489 set_bit(RT_FLAG_RS_IN_SYNC, &rs->runtime_flags);
3490 rdev_for_each(rdev, mddev)
3491 if (!test_bit(Journal, &rdev->flags) &&
3492 !test_bit(In_sync, &rdev->flags)) {
3493 clear_bit(RT_FLAG_RS_IN_SYNC, &rs->runtime_flags);
3494 break;
3495 }
3496 }
3497 }
3498
3499 return min(r, resync_max_sectors);
3500}
3501
3502/* Helper to return @dev name or "-" if !@dev */
3503static const char *__get_dev_name(struct dm_dev *dev)
3504{
3505 return dev ? dev->name : "-";
3506}
3507
3508static void raid_status(struct dm_target *ti, status_type_t type,
3509 unsigned int status_flags, char *result, unsigned int maxlen)
3510{
3511 struct raid_set *rs = ti->private;
3512 struct mddev *mddev = &rs->md;
3513 struct r5conf *conf = rs_is_raid456(rs) ? mddev->private : NULL;
3514 int i, max_nr_stripes = conf ? conf->max_nr_stripes : 0;
3515 unsigned long recovery;
3516 unsigned int raid_param_cnt = 1; /* at least 1 for chunksize */
3517 unsigned int sz = 0;
3518 unsigned int rebuild_writemostly_count = 0;
3519 sector_t progress, resync_max_sectors, resync_mismatches;
3520 enum sync_state state;
3521 struct raid_type *rt;
3522
3523 switch (type) {
3524 case STATUSTYPE_INFO:
3525 /* *Should* always succeed */
3526 rt = get_raid_type_by_ll(mddev->new_level, mddev->new_layout);
3527 if (!rt)
3528 return;
3529
3530 DMEMIT("%s %d ", rt->name, mddev->raid_disks);
3531
3532 /* Access most recent mddev properties for status output */
3533 smp_rmb();
3534 /* Get sensible max sectors even if raid set not yet started */
3535 resync_max_sectors = test_bit(RT_FLAG_RS_PRERESUMED, &rs->runtime_flags) ?
3536 mddev->resync_max_sectors : mddev->dev_sectors;
3537 recovery = rs->md.recovery;
3538 state = decipher_sync_action(mddev, recovery);
3539 progress = rs_get_progress(rs, recovery, state, resync_max_sectors);
3540 resync_mismatches = (mddev->last_sync_action && !strcasecmp(mddev->last_sync_action, "check")) ?
3541 atomic64_read(&mddev->resync_mismatches) : 0;
3542
3543 /* HM FIXME: do we want another state char for raid0? It shows 'D'/'A'/'-' now */
3544 for (i = 0; i < rs->raid_disks; i++)
3545 DMEMIT(__raid_dev_status(rs, &rs->dev[i].rdev));
3546
3547 /*
3548 * In-sync/Reshape ratio:
3549 * The in-sync ratio shows the progress of:
3550 * - Initializing the raid set
3551 * - Rebuilding a subset of devices of the raid set
3552 * The user can distinguish between the two by referring
3553 * to the status characters.
3554 *
3555 * The reshape ratio shows the progress of
3556 * changing the raid layout or the number of
3557 * disks of a raid set
3558 */
3559 DMEMIT(" %llu/%llu", (unsigned long long) progress,
3560 (unsigned long long) resync_max_sectors);
3561
3562 /*
3563 * v1.5.0+:
3564 *
3565 * Sync action:
3566 * See Documentation/admin-guide/device-mapper/dm-raid.rst for
3567 * information on each of these states.
3568 */
3569 DMEMIT(" %s", sync_str(state));
3570
3571 /*
3572 * v1.5.0+:
3573 *
3574 * resync_mismatches/mismatch_cnt
3575 * This field shows the number of discrepancies found when
3576 * performing a "check" of the raid set.
3577 */
3578 DMEMIT(" %llu", (unsigned long long) resync_mismatches);
3579
3580 /*
3581 * v1.9.0+:
3582 *
3583 * data_offset (needed for out of space reshaping)
3584 * This field shows the data offset into the data
3585 * image LV where the first stripes data starts.
3586 *
3587 * We keep data_offset equal on all raid disks of the set,
3588 * so retrieving it from the first raid disk is sufficient.
3589 */
3590 DMEMIT(" %llu", (unsigned long long) rs->dev[0].rdev.data_offset);
3591
3592 /*
3593 * v1.10.0+:
3594 */
3595 DMEMIT(" %s", test_bit(__CTR_FLAG_JOURNAL_DEV, &rs->ctr_flags) ?
3596 __raid_dev_status(rs, &rs->journal_dev.rdev) : "-");
3597 break;
3598
3599 case STATUSTYPE_TABLE:
3600 /* Report the table line string you would use to construct this raid set */
3601
3602 /*
3603 * Count any rebuild or writemostly argument pairs and subtract the
3604 * hweight count being added below of any rebuild and writemostly ctr flags.
3605 */
3606 for (i = 0; i < rs->raid_disks; i++) {
3607 rebuild_writemostly_count += (test_bit(i, (void *) rs->rebuild_disks) ? 2 : 0) +
3608 (test_bit(WriteMostly, &rs->dev[i].rdev.flags) ? 2 : 0);
3609 }
3610 rebuild_writemostly_count -= (test_bit(__CTR_FLAG_REBUILD, &rs->ctr_flags) ? 2 : 0) +
3611 (test_bit(__CTR_FLAG_WRITE_MOSTLY, &rs->ctr_flags) ? 2 : 0);
3612 /* Calculate raid parameter count based on ^ rebuild/writemostly argument counts and ctr flags set. */
3613 raid_param_cnt += rebuild_writemostly_count +
3614 hweight32(rs->ctr_flags & CTR_FLAG_OPTIONS_NO_ARGS) +
3615 hweight32(rs->ctr_flags & CTR_FLAG_OPTIONS_ONE_ARG) * 2;
3616 /* Emit table line */
3617 /* This has to be in the documented order for userspace! */
3618 DMEMIT("%s %u %u", rs->raid_type->name, raid_param_cnt, mddev->new_chunk_sectors);
3619 if (test_bit(__CTR_FLAG_SYNC, &rs->ctr_flags))
3620 DMEMIT(" %s", dm_raid_arg_name_by_flag(CTR_FLAG_SYNC));
3621 if (test_bit(__CTR_FLAG_NOSYNC, &rs->ctr_flags))
3622 DMEMIT(" %s", dm_raid_arg_name_by_flag(CTR_FLAG_NOSYNC));
3623 if (test_bit(__CTR_FLAG_REBUILD, &rs->ctr_flags))
3624 for (i = 0; i < rs->raid_disks; i++)
3625 if (test_bit(i, (void *) rs->rebuild_disks))
3626 DMEMIT(" %s %u", dm_raid_arg_name_by_flag(CTR_FLAG_REBUILD), i);
3627 if (test_bit(__CTR_FLAG_DAEMON_SLEEP, &rs->ctr_flags))
3628 DMEMIT(" %s %lu", dm_raid_arg_name_by_flag(CTR_FLAG_DAEMON_SLEEP),
3629 mddev->bitmap_info.daemon_sleep);
3630 if (test_bit(__CTR_FLAG_MIN_RECOVERY_RATE, &rs->ctr_flags))
3631 DMEMIT(" %s %d", dm_raid_arg_name_by_flag(CTR_FLAG_MIN_RECOVERY_RATE),
3632 mddev->sync_speed_min);
3633 if (test_bit(__CTR_FLAG_MAX_RECOVERY_RATE, &rs->ctr_flags))
3634 DMEMIT(" %s %d", dm_raid_arg_name_by_flag(CTR_FLAG_MAX_RECOVERY_RATE),
3635 mddev->sync_speed_max);
3636 if (test_bit(__CTR_FLAG_WRITE_MOSTLY, &rs->ctr_flags))
3637 for (i = 0; i < rs->raid_disks; i++)
3638 if (test_bit(WriteMostly, &rs->dev[i].rdev.flags))
3639 DMEMIT(" %s %d", dm_raid_arg_name_by_flag(CTR_FLAG_WRITE_MOSTLY),
3640 rs->dev[i].rdev.raid_disk);
3641 if (test_bit(__CTR_FLAG_MAX_WRITE_BEHIND, &rs->ctr_flags))
3642 DMEMIT(" %s %lu", dm_raid_arg_name_by_flag(CTR_FLAG_MAX_WRITE_BEHIND),
3643 mddev->bitmap_info.max_write_behind);
3644 if (test_bit(__CTR_FLAG_STRIPE_CACHE, &rs->ctr_flags))
3645 DMEMIT(" %s %d", dm_raid_arg_name_by_flag(CTR_FLAG_STRIPE_CACHE),
3646 max_nr_stripes);
3647 if (test_bit(__CTR_FLAG_REGION_SIZE, &rs->ctr_flags))
3648 DMEMIT(" %s %llu", dm_raid_arg_name_by_flag(CTR_FLAG_REGION_SIZE),
3649 (unsigned long long) to_sector(mddev->bitmap_info.chunksize));
3650 if (test_bit(__CTR_FLAG_RAID10_COPIES, &rs->ctr_flags))
3651 DMEMIT(" %s %d", dm_raid_arg_name_by_flag(CTR_FLAG_RAID10_COPIES),
3652 raid10_md_layout_to_copies(mddev->layout));
3653 if (test_bit(__CTR_FLAG_RAID10_FORMAT, &rs->ctr_flags))
3654 DMEMIT(" %s %s", dm_raid_arg_name_by_flag(CTR_FLAG_RAID10_FORMAT),
3655 raid10_md_layout_to_format(mddev->layout));
3656 if (test_bit(__CTR_FLAG_DELTA_DISKS, &rs->ctr_flags))
3657 DMEMIT(" %s %d", dm_raid_arg_name_by_flag(CTR_FLAG_DELTA_DISKS),
3658 max(rs->delta_disks, mddev->delta_disks));
3659 if (test_bit(__CTR_FLAG_DATA_OFFSET, &rs->ctr_flags))
3660 DMEMIT(" %s %llu", dm_raid_arg_name_by_flag(CTR_FLAG_DATA_OFFSET),
3661 (unsigned long long) rs->data_offset);
3662 if (test_bit(__CTR_FLAG_JOURNAL_DEV, &rs->ctr_flags))
3663 DMEMIT(" %s %s", dm_raid_arg_name_by_flag(CTR_FLAG_JOURNAL_DEV),
3664 __get_dev_name(rs->journal_dev.dev));
3665 if (test_bit(__CTR_FLAG_JOURNAL_MODE, &rs->ctr_flags))
3666 DMEMIT(" %s %s", dm_raid_arg_name_by_flag(CTR_FLAG_JOURNAL_MODE),
3667 md_journal_mode_to_dm_raid(rs->journal_dev.mode));
3668 DMEMIT(" %d", rs->raid_disks);
3669 for (i = 0; i < rs->raid_disks; i++)
3670 DMEMIT(" %s %s", __get_dev_name(rs->dev[i].meta_dev),
3671 __get_dev_name(rs->dev[i].data_dev));
3672 break;
3673
3674 case STATUSTYPE_IMA:
3675 rt = get_raid_type_by_ll(mddev->new_level, mddev->new_layout);
3676 if (!rt)
3677 return;
3678
3679 DMEMIT_TARGET_NAME_VERSION(ti->type);
3680 DMEMIT(",raid_type=%s,raid_disks=%d", rt->name, mddev->raid_disks);
3681
3682 /* Access most recent mddev properties for status output */
3683 smp_rmb();
3684 recovery = rs->md.recovery;
3685 state = decipher_sync_action(mddev, recovery);
3686 DMEMIT(",raid_state=%s", sync_str(state));
3687
3688 for (i = 0; i < rs->raid_disks; i++) {
3689 DMEMIT(",raid_device_%d_status=", i);
3690 DMEMIT(__raid_dev_status(rs, &rs->dev[i].rdev));
3691 }
3692
3693 if (rt_is_raid456(rt)) {
3694 DMEMIT(",journal_dev_mode=");
3695 switch (rs->journal_dev.mode) {
3696 case R5C_JOURNAL_MODE_WRITE_THROUGH:
3697 DMEMIT("%s",
3698 _raid456_journal_mode[R5C_JOURNAL_MODE_WRITE_THROUGH].param);
3699 break;
3700 case R5C_JOURNAL_MODE_WRITE_BACK:
3701 DMEMIT("%s",
3702 _raid456_journal_mode[R5C_JOURNAL_MODE_WRITE_BACK].param);
3703 break;
3704 default:
3705 DMEMIT("invalid");
3706 break;
3707 }
3708 }
3709 DMEMIT(";");
3710 break;
3711 }
3712}
3713
3714static int raid_message(struct dm_target *ti, unsigned int argc, char **argv,
3715 char *result, unsigned maxlen)
3716{
3717 struct raid_set *rs = ti->private;
3718 struct mddev *mddev = &rs->md;
3719
3720 if (!mddev->pers || !mddev->pers->sync_request)
3721 return -EINVAL;
3722
3723 if (!strcasecmp(argv[0], "frozen"))
3724 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3725 else
3726 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3727
3728 if (!strcasecmp(argv[0], "idle") || !strcasecmp(argv[0], "frozen")) {
3729 if (mddev->sync_thread) {
3730 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3731 md_unregister_thread(&mddev->sync_thread);
3732 md_reap_sync_thread(mddev);
3733 }
3734 } else if (decipher_sync_action(mddev, mddev->recovery) != st_idle)
3735 return -EBUSY;
3736 else if (!strcasecmp(argv[0], "resync"))
3737 ; /* MD_RECOVERY_NEEDED set below */
3738 else if (!strcasecmp(argv[0], "recover"))
3739 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
3740 else {
3741 if (!strcasecmp(argv[0], "check")) {
3742 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
3743 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
3744 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
3745 } else if (!strcasecmp(argv[0], "repair")) {
3746 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
3747 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
3748 } else
3749 return -EINVAL;
3750 }
3751 if (mddev->ro == 2) {
3752 /* A write to sync_action is enough to justify
3753 * canceling read-auto mode
3754 */
3755 mddev->ro = 0;
3756 if (!mddev->suspended && mddev->sync_thread)
3757 md_wakeup_thread(mddev->sync_thread);
3758 }
3759 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3760 if (!mddev->suspended && mddev->thread)
3761 md_wakeup_thread(mddev->thread);
3762
3763 return 0;
3764}
3765
3766static int raid_iterate_devices(struct dm_target *ti,
3767 iterate_devices_callout_fn fn, void *data)
3768{
3769 struct raid_set *rs = ti->private;
3770 unsigned int i;
3771 int r = 0;
3772
3773 for (i = 0; !r && i < rs->raid_disks; i++) {
3774 if (rs->dev[i].data_dev) {
3775 r = fn(ti, rs->dev[i].data_dev,
3776 0, /* No offset on data devs */
3777 rs->md.dev_sectors, data);
3778 }
3779 }
3780
3781 return r;
3782}
3783
3784static void raid_io_hints(struct dm_target *ti, struct queue_limits *limits)
3785{
3786 struct raid_set *rs = ti->private;
3787 unsigned int chunk_size_bytes = to_bytes(rs->md.chunk_sectors);
3788
3789 blk_limits_io_min(limits, chunk_size_bytes);
3790 blk_limits_io_opt(limits, chunk_size_bytes * mddev_data_stripes(rs));
3791}
3792
3793static void raid_postsuspend(struct dm_target *ti)
3794{
3795 struct raid_set *rs = ti->private;
3796
3797 if (!test_and_set_bit(RT_FLAG_RS_SUSPENDED, &rs->runtime_flags)) {
3798 /* Writes have to be stopped before suspending to avoid deadlocks. */
3799 if (!test_bit(MD_RECOVERY_FROZEN, &rs->md.recovery))
3800 md_stop_writes(&rs->md);
3801
3802 mddev_lock_nointr(&rs->md);
3803 mddev_suspend(&rs->md);
3804 mddev_unlock(&rs->md);
3805 }
3806}
3807
3808static void attempt_restore_of_faulty_devices(struct raid_set *rs)
3809{
3810 int i;
3811 uint64_t cleared_failed_devices[DISKS_ARRAY_ELEMS];
3812 unsigned long flags;
3813 bool cleared = false;
3814 struct dm_raid_superblock *sb;
3815 struct mddev *mddev = &rs->md;
3816 struct md_rdev *r;
3817
3818 /* RAID personalities have to provide hot add/remove methods or we need to bail out. */
3819 if (!mddev->pers || !mddev->pers->hot_add_disk || !mddev->pers->hot_remove_disk)
3820 return;
3821
3822 memset(cleared_failed_devices, 0, sizeof(cleared_failed_devices));
3823
3824 for (i = 0; i < rs->raid_disks; i++) {
3825 r = &rs->dev[i].rdev;
3826 /* HM FIXME: enhance journal device recovery processing */
3827 if (test_bit(Journal, &r->flags))
3828 continue;
3829
3830 if (test_bit(Faulty, &r->flags) &&
3831 r->meta_bdev && !read_disk_sb(r, r->sb_size, true)) {
3832 DMINFO("Faulty %s device #%d has readable super block."
3833 " Attempting to revive it.",
3834 rs->raid_type->name, i);
3835
3836 /*
3837 * Faulty bit may be set, but sometimes the array can
3838 * be suspended before the personalities can respond
3839 * by removing the device from the array (i.e. calling
3840 * 'hot_remove_disk'). If they haven't yet removed
3841 * the failed device, its 'raid_disk' number will be
3842 * '>= 0' - meaning we must call this function
3843 * ourselves.
3844 */
3845 flags = r->flags;
3846 clear_bit(In_sync, &r->flags); /* Mandatory for hot remove. */
3847 if (r->raid_disk >= 0) {
3848 if (mddev->pers->hot_remove_disk(mddev, r)) {
3849 /* Failed to revive this device, try next */
3850 r->flags = flags;
3851 continue;
3852 }
3853 } else
3854 r->raid_disk = r->saved_raid_disk = i;
3855
3856 clear_bit(Faulty, &r->flags);
3857 clear_bit(WriteErrorSeen, &r->flags);
3858
3859 if (mddev->pers->hot_add_disk(mddev, r)) {
3860 /* Failed to revive this device, try next */
3861 r->raid_disk = r->saved_raid_disk = -1;
3862 r->flags = flags;
3863 } else {
3864 clear_bit(In_sync, &r->flags);
3865 r->recovery_offset = 0;
3866 set_bit(i, (void *) cleared_failed_devices);
3867 cleared = true;
3868 }
3869 }
3870 }
3871
3872 /* If any failed devices could be cleared, update all sbs failed_devices bits */
3873 if (cleared) {
3874 uint64_t failed_devices[DISKS_ARRAY_ELEMS];
3875
3876 rdev_for_each(r, &rs->md) {
3877 if (test_bit(Journal, &r->flags))
3878 continue;
3879
3880 sb = page_address(r->sb_page);
3881 sb_retrieve_failed_devices(sb, failed_devices);
3882
3883 for (i = 0; i < DISKS_ARRAY_ELEMS; i++)
3884 failed_devices[i] &= ~cleared_failed_devices[i];
3885
3886 sb_update_failed_devices(sb, failed_devices);
3887 }
3888 }
3889}
3890
3891static int __load_dirty_region_bitmap(struct raid_set *rs)
3892{
3893 int r = 0;
3894
3895 /* Try loading the bitmap unless "raid0", which does not have one */
3896 if (!rs_is_raid0(rs) &&
3897 !test_and_set_bit(RT_FLAG_RS_BITMAP_LOADED, &rs->runtime_flags)) {
3898 r = md_bitmap_load(&rs->md);
3899 if (r)
3900 DMERR("Failed to load bitmap");
3901 }
3902
3903 return r;
3904}
3905
3906/* Enforce updating all superblocks */
3907static void rs_update_sbs(struct raid_set *rs)
3908{
3909 struct mddev *mddev = &rs->md;
3910 int ro = mddev->ro;
3911
3912 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
3913 mddev->ro = 0;
3914 md_update_sb(mddev, 1);
3915 mddev->ro = ro;
3916}
3917
3918/*
3919 * Reshape changes raid algorithm of @rs to new one within personality
3920 * (e.g. raid6_zr -> raid6_nc), changes stripe size, adds/removes
3921 * disks from a raid set thus growing/shrinking it or resizes the set
3922 *
3923 * Call mddev_lock_nointr() before!
3924 */
3925static int rs_start_reshape(struct raid_set *rs)
3926{
3927 int r;
3928 struct mddev *mddev = &rs->md;
3929 struct md_personality *pers = mddev->pers;
3930
3931 /* Don't allow the sync thread to work until the table gets reloaded. */
3932 set_bit(MD_RECOVERY_WAIT, &mddev->recovery);
3933
3934 r = rs_setup_reshape(rs);
3935 if (r)
3936 return r;
3937
3938 /*
3939 * Check any reshape constraints enforced by the personalility
3940 *
3941 * May as well already kick the reshape off so that * pers->start_reshape() becomes optional.
3942 */
3943 r = pers->check_reshape(mddev);
3944 if (r) {
3945 rs->ti->error = "pers->check_reshape() failed";
3946 return r;
3947 }
3948
3949 /*
3950 * Personality may not provide start reshape method in which
3951 * case check_reshape above has already covered everything
3952 */
3953 if (pers->start_reshape) {
3954 r = pers->start_reshape(mddev);
3955 if (r) {
3956 rs->ti->error = "pers->start_reshape() failed";
3957 return r;
3958 }
3959 }
3960
3961 /*
3962 * Now reshape got set up, update superblocks to
3963 * reflect the fact so that a table reload will
3964 * access proper superblock content in the ctr.
3965 */
3966 rs_update_sbs(rs);
3967
3968 return 0;
3969}
3970
3971static int raid_preresume(struct dm_target *ti)
3972{
3973 int r;
3974 struct raid_set *rs = ti->private;
3975 struct mddev *mddev = &rs->md;
3976
3977 /* This is a resume after a suspend of the set -> it's already started. */
3978 if (test_and_set_bit(RT_FLAG_RS_PRERESUMED, &rs->runtime_flags))
3979 return 0;
3980
3981 /*
3982 * The superblocks need to be updated on disk if the
3983 * array is new or new devices got added (thus zeroed
3984 * out by userspace) or __load_dirty_region_bitmap
3985 * will overwrite them in core with old data or fail.
3986 */
3987 if (test_bit(RT_FLAG_UPDATE_SBS, &rs->runtime_flags))
3988 rs_update_sbs(rs);
3989
3990 /* Load the bitmap from disk unless raid0 */
3991 r = __load_dirty_region_bitmap(rs);
3992 if (r)
3993 return r;
3994
3995 /* We are extending the raid set size, adjust mddev/md_rdev sizes and set capacity. */
3996 if (test_bit(RT_FLAG_RS_GROW, &rs->runtime_flags)) {
3997 mddev->array_sectors = rs->array_sectors;
3998 mddev->dev_sectors = rs->dev_sectors;
3999 rs_set_rdev_sectors(rs);
4000 rs_set_capacity(rs);
4001 }
4002
4003 /* Resize bitmap to adjust to changed region size (aka MD bitmap chunksize) or grown device size */
4004 if (test_bit(RT_FLAG_RS_BITMAP_LOADED, &rs->runtime_flags) && mddev->bitmap &&
4005 (test_bit(RT_FLAG_RS_GROW, &rs->runtime_flags) ||
4006 (rs->requested_bitmap_chunk_sectors &&
4007 mddev->bitmap_info.chunksize != to_bytes(rs->requested_bitmap_chunk_sectors)))) {
4008 int chunksize = to_bytes(rs->requested_bitmap_chunk_sectors) ?: mddev->bitmap_info.chunksize;
4009
4010 r = md_bitmap_resize(mddev->bitmap, mddev->dev_sectors, chunksize, 0);
4011 if (r)
4012 DMERR("Failed to resize bitmap");
4013 }
4014
4015 /* Check for any resize/reshape on @rs and adjust/initiate */
4016 /* Be prepared for mddev_resume() in raid_resume() */
4017 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4018 if (mddev->recovery_cp && mddev->recovery_cp < MaxSector) {
4019 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
4020 mddev->resync_min = mddev->recovery_cp;
4021 if (test_bit(RT_FLAG_RS_GROW, &rs->runtime_flags))
4022 mddev->resync_max_sectors = mddev->dev_sectors;
4023 }
4024
4025 /* Check for any reshape request unless new raid set */
4026 if (test_bit(RT_FLAG_RESHAPE_RS, &rs->runtime_flags)) {
4027 /* Initiate a reshape. */
4028 rs_set_rdev_sectors(rs);
4029 mddev_lock_nointr(mddev);
4030 r = rs_start_reshape(rs);
4031 mddev_unlock(mddev);
4032 if (r)
4033 DMWARN("Failed to check/start reshape, continuing without change");
4034 r = 0;
4035 }
4036
4037 return r;
4038}
4039
4040static void raid_resume(struct dm_target *ti)
4041{
4042 struct raid_set *rs = ti->private;
4043 struct mddev *mddev = &rs->md;
4044
4045 if (test_and_set_bit(RT_FLAG_RS_RESUMED, &rs->runtime_flags)) {
4046 /*
4047 * A secondary resume while the device is active.
4048 * Take this opportunity to check whether any failed
4049 * devices are reachable again.
4050 */
4051 attempt_restore_of_faulty_devices(rs);
4052 }
4053
4054 if (test_and_clear_bit(RT_FLAG_RS_SUSPENDED, &rs->runtime_flags)) {
4055 /* Only reduce raid set size before running a disk removing reshape. */
4056 if (mddev->delta_disks < 0)
4057 rs_set_capacity(rs);
4058
4059 mddev_lock_nointr(mddev);
4060 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4061 mddev->ro = 0;
4062 mddev->in_sync = 0;
4063 mddev_resume(mddev);
4064 mddev_unlock(mddev);
4065 }
4066}
4067
4068static struct target_type raid_target = {
4069 .name = "raid",
4070 .version = {1, 15, 1},
4071 .module = THIS_MODULE,
4072 .ctr = raid_ctr,
4073 .dtr = raid_dtr,
4074 .map = raid_map,
4075 .status = raid_status,
4076 .message = raid_message,
4077 .iterate_devices = raid_iterate_devices,
4078 .io_hints = raid_io_hints,
4079 .postsuspend = raid_postsuspend,
4080 .preresume = raid_preresume,
4081 .resume = raid_resume,
4082};
4083
4084static int __init dm_raid_init(void)
4085{
4086 DMINFO("Loading target version %u.%u.%u",
4087 raid_target.version[0],
4088 raid_target.version[1],
4089 raid_target.version[2]);
4090 return dm_register_target(&raid_target);
4091}
4092
4093static void __exit dm_raid_exit(void)
4094{
4095 dm_unregister_target(&raid_target);
4096}
4097
4098module_init(dm_raid_init);
4099module_exit(dm_raid_exit);
4100
4101module_param(devices_handle_discard_safely, bool, 0644);
4102MODULE_PARM_DESC(devices_handle_discard_safely,
4103 "Set to Y if all devices in each array reliably return zeroes on reads from discarded regions");
4104
4105MODULE_DESCRIPTION(DM_NAME " raid0/1/10/4/5/6 target");
4106MODULE_ALIAS("dm-raid0");
4107MODULE_ALIAS("dm-raid1");
4108MODULE_ALIAS("dm-raid10");
4109MODULE_ALIAS("dm-raid4");
4110MODULE_ALIAS("dm-raid5");
4111MODULE_ALIAS("dm-raid6");
4112MODULE_AUTHOR("Neil Brown <dm-devel@redhat.com>");
4113MODULE_AUTHOR("Heinz Mauelshagen <dm-devel@redhat.com>");
4114MODULE_LICENSE("GPL");