1// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 raid0.c : Multiple Devices driver for Linux
4 Copyright (C) 1994-96 Marc ZYNGIER
5 <zyngier@ufr-info-p7.ibp.fr> or
6 <maz@gloups.fdn.fr>
7 Copyright (C) 1999, 2000 Ingo Molnar, Red Hat
8
9 RAID-0 management functions.
10
11*/
12
13#include <linux/blkdev.h>
14#include <linux/seq_file.h>
15#include <linux/module.h>
16#include <linux/slab.h>
17#include <trace/events/block.h>
18#include "md.h"
19#include "raid0.h"
20#include "raid5.h"
21
22static int default_layout = 0;
23module_param(default_layout, int, 0644);
24
25#define UNSUPPORTED_MDDEV_FLAGS \
26 ((1L << MD_HAS_JOURNAL) | \
27 (1L << MD_JOURNAL_CLEAN) | \
28 (1L << MD_FAILFAST_SUPPORTED) |\
29 (1L << MD_HAS_PPL) | \
30 (1L << MD_HAS_MULTIPLE_PPLS))
31
32/*
33 * inform the user of the raid configuration
34*/
35static void dump_zones(struct mddev *mddev)
36{
37 int j, k;
38 sector_t zone_size = 0;
39 sector_t zone_start = 0;
40 struct r0conf *conf = mddev->private;
41 int raid_disks = conf->strip_zone[0].nb_dev;
42 pr_debug("md: RAID0 configuration for %s - %d zone%s\n",
43 mdname(mddev),
44 conf->nr_strip_zones, conf->nr_strip_zones==1?"":"s");
45 for (j = 0; j < conf->nr_strip_zones; j++) {
46 char line[200];
47 int len = 0;
48
49 for (k = 0; k < conf->strip_zone[j].nb_dev; k++)
50 len += scnprintf(buf: line+len, size: 200-len, fmt: "%s%pg", k?"/":"",
51 conf->devlist[j * raid_disks + k]->bdev);
52 pr_debug("md: zone%d=[%s]\n", j, line);
53
54 zone_size = conf->strip_zone[j].zone_end - zone_start;
55 pr_debug(" zone-offset=%10lluKB, device-offset=%10lluKB, size=%10lluKB\n",
56 (unsigned long long)zone_start>>1,
57 (unsigned long long)conf->strip_zone[j].dev_start>>1,
58 (unsigned long long)zone_size>>1);
59 zone_start = conf->strip_zone[j].zone_end;
60 }
61}
62
63static int create_strip_zones(struct mddev *mddev, struct r0conf **private_conf)
64{
65 int i, c, err;
66 sector_t curr_zone_end, sectors;
67 struct md_rdev *smallest, *rdev1, *rdev2, *rdev, **dev;
68 struct strip_zone *zone;
69 int cnt;
70 struct r0conf *conf = kzalloc(sizeof(*conf), GFP_KERNEL);
71 unsigned int blksize = 512;
72
73 if (!mddev_is_dm(mddev))
74 blksize = queue_logical_block_size(q: mddev->gendisk->queue);
75
76 *private_conf = ERR_PTR(error: -ENOMEM);
77 if (!conf)
78 return -ENOMEM;
79 rdev_for_each(rdev1, mddev) {
80 pr_debug("md/raid0:%s: looking at %pg\n",
81 mdname(mddev),
82 rdev1->bdev);
83 c = 0;
84
85 /* round size to chunk_size */
86 sectors = rdev1->sectors;
87 sector_div(sectors, mddev->chunk_sectors);
88 rdev1->sectors = sectors * mddev->chunk_sectors;
89
90 if (mddev_is_dm(mddev))
91 blksize = max(blksize, queue_logical_block_size(
92 rdev1->bdev->bd_disk->queue));
93
94 rdev_for_each(rdev2, mddev) {
95 pr_debug("md/raid0:%s: comparing %pg(%llu)"
96 " with %pg(%llu)\n",
97 mdname(mddev),
98 rdev1->bdev,
99 (unsigned long long)rdev1->sectors,
100 rdev2->bdev,
101 (unsigned long long)rdev2->sectors);
102 if (rdev2 == rdev1) {
103 pr_debug("md/raid0:%s: END\n",
104 mdname(mddev));
105 break;
106 }
107 if (rdev2->sectors == rdev1->sectors) {
108 /*
109 * Not unique, don't count it as a new
110 * group
111 */
112 pr_debug("md/raid0:%s: EQUAL\n",
113 mdname(mddev));
114 c = 1;
115 break;
116 }
117 pr_debug("md/raid0:%s: NOT EQUAL\n",
118 mdname(mddev));
119 }
120 if (!c) {
121 pr_debug("md/raid0:%s: ==> UNIQUE\n",
122 mdname(mddev));
123 conf->nr_strip_zones++;
124 pr_debug("md/raid0:%s: %d zones\n",
125 mdname(mddev), conf->nr_strip_zones);
126 }
127 }
128 pr_debug("md/raid0:%s: FINAL %d zones\n",
129 mdname(mddev), conf->nr_strip_zones);
130
131 /*
132 * now since we have the hard sector sizes, we can make sure
133 * chunk size is a multiple of that sector size
134 */
135 if ((mddev->chunk_sectors << 9) % blksize) {
136 pr_warn("md/raid0:%s: chunk_size of %d not multiple of block size %d\n",
137 mdname(mddev),
138 mddev->chunk_sectors << 9, blksize);
139 err = -EINVAL;
140 goto abort;
141 }
142
143 err = -ENOMEM;
144 conf->strip_zone = kcalloc(conf->nr_strip_zones,
145 sizeof(struct strip_zone),
146 GFP_KERNEL);
147 if (!conf->strip_zone)
148 goto abort;
149 conf->devlist = kzalloc(array3_size(sizeof(struct md_rdev *),
150 conf->nr_strip_zones,
151 mddev->raid_disks),
152 GFP_KERNEL);
153 if (!conf->devlist)
154 goto abort;
155
156 /* The first zone must contain all devices, so here we check that
157 * there is a proper alignment of slots to devices and find them all
158 */
159 zone = &conf->strip_zone[0];
160 cnt = 0;
161 smallest = NULL;
162 dev = conf->devlist;
163 err = -EINVAL;
164 rdev_for_each(rdev1, mddev) {
165 int j = rdev1->raid_disk;
166
167 if (mddev->level == 10) {
168 /* taking over a raid10-n2 array */
169 j /= 2;
170 rdev1->new_raid_disk = j;
171 }
172
173 if (mddev->level == 1) {
174 /* taiking over a raid1 array-
175 * we have only one active disk
176 */
177 j = 0;
178 rdev1->new_raid_disk = j;
179 }
180
181 if (j < 0) {
182 pr_warn("md/raid0:%s: remove inactive devices before converting to RAID0\n",
183 mdname(mddev));
184 goto abort;
185 }
186 if (j >= mddev->raid_disks) {
187 pr_warn("md/raid0:%s: bad disk number %d - aborting!\n",
188 mdname(mddev), j);
189 goto abort;
190 }
191 if (dev[j]) {
192 pr_warn("md/raid0:%s: multiple devices for %d - aborting!\n",
193 mdname(mddev), j);
194 goto abort;
195 }
196 dev[j] = rdev1;
197
198 if (!smallest || (rdev1->sectors < smallest->sectors))
199 smallest = rdev1;
200 cnt++;
201 }
202 if (cnt != mddev->raid_disks) {
203 pr_warn("md/raid0:%s: too few disks (%d of %d) - aborting!\n",
204 mdname(mddev), cnt, mddev->raid_disks);
205 goto abort;
206 }
207 zone->nb_dev = cnt;
208 zone->zone_end = smallest->sectors * cnt;
209
210 curr_zone_end = zone->zone_end;
211
212 /* now do the other zones */
213 for (i = 1; i < conf->nr_strip_zones; i++)
214 {
215 int j;
216
217 zone = conf->strip_zone + i;
218 dev = conf->devlist + i * mddev->raid_disks;
219
220 pr_debug("md/raid0:%s: zone %d\n", mdname(mddev), i);
221 zone->dev_start = smallest->sectors;
222 smallest = NULL;
223 c = 0;
224
225 for (j=0; j<cnt; j++) {
226 rdev = conf->devlist[j];
227 if (rdev->sectors <= zone->dev_start) {
228 pr_debug("md/raid0:%s: checking %pg ... nope\n",
229 mdname(mddev),
230 rdev->bdev);
231 continue;
232 }
233 pr_debug("md/raid0:%s: checking %pg ..."
234 " contained as device %d\n",
235 mdname(mddev),
236 rdev->bdev, c);
237 dev[c] = rdev;
238 c++;
239 if (!smallest || rdev->sectors < smallest->sectors) {
240 smallest = rdev;
241 pr_debug("md/raid0:%s: (%llu) is smallest!.\n",
242 mdname(mddev),
243 (unsigned long long)rdev->sectors);
244 }
245 }
246
247 zone->nb_dev = c;
248 sectors = (smallest->sectors - zone->dev_start) * c;
249 pr_debug("md/raid0:%s: zone->nb_dev: %d, sectors: %llu\n",
250 mdname(mddev),
251 zone->nb_dev, (unsigned long long)sectors);
252
253 curr_zone_end += sectors;
254 zone->zone_end = curr_zone_end;
255
256 pr_debug("md/raid0:%s: current zone start: %llu\n",
257 mdname(mddev),
258 (unsigned long long)smallest->sectors);
259 }
260
261 if (conf->nr_strip_zones == 1 || conf->strip_zone[1].nb_dev == 1) {
262 conf->layout = RAID0_ORIG_LAYOUT;
263 } else if (mddev->layout == RAID0_ORIG_LAYOUT ||
264 mddev->layout == RAID0_ALT_MULTIZONE_LAYOUT) {
265 conf->layout = mddev->layout;
266 } else if (default_layout == RAID0_ORIG_LAYOUT ||
267 default_layout == RAID0_ALT_MULTIZONE_LAYOUT) {
268 conf->layout = default_layout;
269 } else {
270 pr_err("md/raid0:%s: cannot assemble multi-zone RAID0 with default_layout setting\n",
271 mdname(mddev));
272 pr_err("md/raid0: please set raid0.default_layout to 1 or 2\n");
273 err = -EOPNOTSUPP;
274 goto abort;
275 }
276
277 if (conf->layout == RAID0_ORIG_LAYOUT) {
278 for (i = 1; i < conf->nr_strip_zones; i++) {
279 sector_t first_sector = conf->strip_zone[i-1].zone_end;
280
281 sector_div(first_sector, mddev->chunk_sectors);
282 zone = conf->strip_zone + i;
283 /* disk_shift is first disk index used in the zone */
284 zone->disk_shift = sector_div(first_sector,
285 zone->nb_dev);
286 }
287 }
288
289 pr_debug("md/raid0:%s: done.\n", mdname(mddev));
290 *private_conf = conf;
291
292 return 0;
293abort:
294 kfree(objp: conf->strip_zone);
295 kfree(objp: conf->devlist);
296 kfree(objp: conf);
297 *private_conf = ERR_PTR(error: err);
298 return err;
299}
300
301/* Find the zone which holds a particular offset
302 * Update *sectorp to be an offset in that zone
303 */
304static struct strip_zone *find_zone(struct r0conf *conf,
305 sector_t *sectorp)
306{
307 int i;
308 struct strip_zone *z = conf->strip_zone;
309 sector_t sector = *sectorp;
310
311 for (i = 0; i < conf->nr_strip_zones; i++)
312 if (sector < z[i].zone_end) {
313 if (i)
314 *sectorp = sector - z[i-1].zone_end;
315 return z + i;
316 }
317 BUG();
318}
319
320/*
321 * remaps the bio to the target device. we separate two flows.
322 * power 2 flow and a general flow for the sake of performance
323*/
324static struct md_rdev *map_sector(struct mddev *mddev, struct strip_zone *zone,
325 sector_t sector, sector_t *sector_offset)
326{
327 unsigned int sect_in_chunk;
328 sector_t chunk;
329 struct r0conf *conf = mddev->private;
330 int raid_disks = conf->strip_zone[0].nb_dev;
331 unsigned int chunk_sects = mddev->chunk_sectors;
332
333 if (is_power_of_2(n: chunk_sects)) {
334 int chunksect_bits = ffz(~chunk_sects);
335 /* find the sector offset inside the chunk */
336 sect_in_chunk = sector & (chunk_sects - 1);
337 sector >>= chunksect_bits;
338 /* chunk in zone */
339 chunk = *sector_offset;
340 /* quotient is the chunk in real device*/
341 sector_div(chunk, zone->nb_dev << chunksect_bits);
342 } else{
343 sect_in_chunk = sector_div(sector, chunk_sects);
344 chunk = *sector_offset;
345 sector_div(chunk, chunk_sects * zone->nb_dev);
346 }
347 /*
348 * position the bio over the real device
349 * real sector = chunk in device + starting of zone
350 * + the position in the chunk
351 */
352 *sector_offset = (chunk * chunk_sects) + sect_in_chunk;
353 return conf->devlist[(zone - conf->strip_zone)*raid_disks
354 + sector_div(sector, zone->nb_dev)];
355}
356
357static sector_t raid0_size(struct mddev *mddev, sector_t sectors, int raid_disks)
358{
359 sector_t array_sectors = 0;
360 struct md_rdev *rdev;
361
362 WARN_ONCE(sectors || raid_disks,
363 "%s does not support generic reshape\n", __func__);
364
365 rdev_for_each(rdev, mddev)
366 array_sectors += (rdev->sectors &
367 ~(sector_t)(mddev->chunk_sectors-1));
368
369 return array_sectors;
370}
371
372static void raid0_free(struct mddev *mddev, void *priv)
373{
374 struct r0conf *conf = priv;
375
376 kfree(objp: conf->strip_zone);
377 kfree(objp: conf->devlist);
378 kfree(objp: conf);
379}
380
381static int raid0_set_limits(struct mddev *mddev)
382{
383 struct queue_limits lim;
384 int err;
385
386 md_init_stacking_limits(lim: &lim);
387 lim.max_hw_sectors = mddev->chunk_sectors;
388 lim.max_write_zeroes_sectors = mddev->chunk_sectors;
389 lim.max_hw_wzeroes_unmap_sectors = mddev->chunk_sectors;
390 lim.logical_block_size = mddev->logical_block_size;
391 lim.io_min = mddev->chunk_sectors << 9;
392 lim.io_opt = lim.io_min * mddev->raid_disks;
393 lim.chunk_sectors = mddev->chunk_sectors;
394 lim.features |= BLK_FEAT_ATOMIC_WRITES;
395 err = mddev_stack_rdev_limits(mddev, lim: &lim, MDDEV_STACK_INTEGRITY);
396 if (err)
397 return err;
398 return queue_limits_set(q: mddev->gendisk->queue, lim: &lim);
399}
400
401static int raid0_run(struct mddev *mddev)
402{
403 struct r0conf *conf;
404 int ret;
405
406 if (mddev->chunk_sectors == 0) {
407 pr_warn("md/raid0:%s: chunk size must be set.\n", mdname(mddev));
408 return -EINVAL;
409 }
410 if (md_check_no_bitmap(mddev))
411 return -EINVAL;
412
413 if (!mddev_is_dm(mddev)) {
414 ret = raid0_set_limits(mddev);
415 if (ret)
416 return ret;
417 }
418
419 /* if private is not null, we are here after takeover */
420 if (mddev->private == NULL) {
421 ret = create_strip_zones(mddev, private_conf: &conf);
422 if (ret < 0)
423 return ret;
424 mddev->private = conf;
425 }
426 conf = mddev->private;
427
428 /* calculate array device size */
429 md_set_array_sectors(mddev, array_sectors: raid0_size(mddev, sectors: 0, raid_disks: 0));
430
431 pr_debug("md/raid0:%s: md_size is %llu sectors.\n",
432 mdname(mddev),
433 (unsigned long long)mddev->array_sectors);
434
435 dump_zones(mddev);
436
437 return md_integrity_register(mddev);
438}
439
440/*
441 * Convert disk_index to the disk order in which it is read/written.
442 * For example, if we have 4 disks, they are numbered 0,1,2,3. If we
443 * write the disks starting at disk 3, then the read/write order would
444 * be disk 3, then 0, then 1, and then disk 2 and we want map_disk_shift()
445 * to map the disks as follows 0,1,2,3 => 1,2,3,0. So disk 0 would map
446 * to 1, 1 to 2, 2 to 3, and 3 to 0. That way we can compare disks in
447 * that 'output' space to understand the read/write disk ordering.
448 */
449static int map_disk_shift(int disk_index, int num_disks, int disk_shift)
450{
451 return ((disk_index + num_disks - disk_shift) % num_disks);
452}
453
454static void raid0_handle_discard(struct mddev *mddev, struct bio *bio)
455{
456 struct r0conf *conf = mddev->private;
457 struct strip_zone *zone;
458 sector_t start = bio->bi_iter.bi_sector;
459 sector_t end;
460 unsigned int stripe_size;
461 sector_t first_stripe_index, last_stripe_index;
462 sector_t start_disk_offset;
463 unsigned int start_disk_index;
464 sector_t end_disk_offset;
465 unsigned int end_disk_index;
466 unsigned int disk;
467 sector_t orig_start, orig_end;
468
469 orig_start = start;
470 zone = find_zone(conf, sectorp: &start);
471
472 if (bio_end_sector(bio) > zone->zone_end) {
473 bio = bio_submit_split_bioset(bio,
474 split_sectors: zone->zone_end - bio->bi_iter.bi_sector,
475 bs: &mddev->bio_set);
476 if (!bio)
477 return;
478
479 end = zone->zone_end;
480 } else {
481 end = bio_end_sector(bio);
482 }
483
484 orig_end = end;
485 if (zone != conf->strip_zone)
486 end = end - zone[-1].zone_end;
487
488 /* Now start and end is the offset in zone */
489 stripe_size = zone->nb_dev * mddev->chunk_sectors;
490
491 first_stripe_index = start;
492 sector_div(first_stripe_index, stripe_size);
493 last_stripe_index = end;
494 sector_div(last_stripe_index, stripe_size);
495
496 /* In the first zone the original and alternate layouts are the same */
497 if ((conf->layout == RAID0_ORIG_LAYOUT) && (zone != conf->strip_zone)) {
498 sector_div(orig_start, mddev->chunk_sectors);
499 start_disk_index = sector_div(orig_start, zone->nb_dev);
500 start_disk_index = map_disk_shift(disk_index: start_disk_index,
501 num_disks: zone->nb_dev,
502 disk_shift: zone->disk_shift);
503 sector_div(orig_end, mddev->chunk_sectors);
504 end_disk_index = sector_div(orig_end, zone->nb_dev);
505 end_disk_index = map_disk_shift(disk_index: end_disk_index,
506 num_disks: zone->nb_dev, disk_shift: zone->disk_shift);
507 } else {
508 start_disk_index = (int)(start - first_stripe_index * stripe_size) /
509 mddev->chunk_sectors;
510 end_disk_index = (int)(end - last_stripe_index * stripe_size) /
511 mddev->chunk_sectors;
512 }
513 start_disk_offset = ((int)(start - first_stripe_index * stripe_size) %
514 mddev->chunk_sectors) +
515 first_stripe_index * mddev->chunk_sectors;
516 end_disk_offset = ((int)(end - last_stripe_index * stripe_size) %
517 mddev->chunk_sectors) +
518 last_stripe_index * mddev->chunk_sectors;
519
520 for (disk = 0; disk < zone->nb_dev; disk++) {
521 sector_t dev_start, dev_end;
522 struct md_rdev *rdev;
523 int compare_disk;
524
525 compare_disk = map_disk_shift(disk_index: disk, num_disks: zone->nb_dev,
526 disk_shift: zone->disk_shift);
527
528 if (compare_disk < start_disk_index)
529 dev_start = (first_stripe_index + 1) *
530 mddev->chunk_sectors;
531 else if (compare_disk > start_disk_index)
532 dev_start = first_stripe_index * mddev->chunk_sectors;
533 else
534 dev_start = start_disk_offset;
535
536 if (compare_disk < end_disk_index)
537 dev_end = (last_stripe_index + 1) * mddev->chunk_sectors;
538 else if (compare_disk > end_disk_index)
539 dev_end = last_stripe_index * mddev->chunk_sectors;
540 else
541 dev_end = end_disk_offset;
542
543 if (dev_end <= dev_start)
544 continue;
545
546 rdev = conf->devlist[(zone - conf->strip_zone) *
547 conf->strip_zone[0].nb_dev + disk];
548 md_submit_discard_bio(mddev, rdev, bio,
549 start: dev_start + zone->dev_start + rdev->data_offset,
550 size: dev_end - dev_start);
551 }
552 bio_endio(bio);
553}
554
555static void raid0_map_submit_bio(struct mddev *mddev, struct bio *bio)
556{
557 struct r0conf *conf = mddev->private;
558 struct strip_zone *zone;
559 struct md_rdev *tmp_dev;
560 sector_t bio_sector = bio->bi_iter.bi_sector;
561 sector_t sector = bio_sector;
562
563 md_account_bio(mddev, bio: &bio);
564
565 zone = find_zone(conf: mddev->private, sectorp: &sector);
566 switch (conf->layout) {
567 case RAID0_ORIG_LAYOUT:
568 tmp_dev = map_sector(mddev, zone, sector: bio_sector, sector_offset: &sector);
569 break;
570 case RAID0_ALT_MULTIZONE_LAYOUT:
571 tmp_dev = map_sector(mddev, zone, sector, sector_offset: &sector);
572 break;
573 default:
574 WARN(1, "md/raid0:%s: Invalid layout\n", mdname(mddev));
575 bio_io_error(bio);
576 return;
577 }
578
579 if (unlikely(is_rdev_broken(tmp_dev))) {
580 bio_io_error(bio);
581 md_error(mddev, rdev: tmp_dev);
582 return;
583 }
584
585 bio_set_dev(bio, bdev: tmp_dev->bdev);
586 bio->bi_iter.bi_sector = sector + zone->dev_start +
587 tmp_dev->data_offset;
588 mddev_trace_remap(mddev, bio, sector: bio_sector);
589 mddev_check_write_zeroes(mddev, bio);
590 submit_bio_noacct(bio);
591}
592
593static bool raid0_make_request(struct mddev *mddev, struct bio *bio)
594{
595 sector_t sector;
596 unsigned chunk_sects;
597 unsigned sectors;
598
599 if (unlikely(bio->bi_opf & REQ_PREFLUSH)
600 && md_flush_request(mddev, bio))
601 return true;
602
603 if (unlikely((bio_op(bio) == REQ_OP_DISCARD))) {
604 raid0_handle_discard(mddev, bio);
605 return true;
606 }
607
608 sector = bio->bi_iter.bi_sector;
609 chunk_sects = mddev->chunk_sectors;
610
611 sectors = chunk_sects -
612 (likely(is_power_of_2(chunk_sects))
613 ? (sector & (chunk_sects-1))
614 : sector_div(sector, chunk_sects));
615
616 if (sectors < bio_sectors(bio)) {
617 bio = bio_submit_split_bioset(bio, split_sectors: sectors,
618 bs: &mddev->bio_set);
619 if (!bio)
620 return true;
621 }
622
623 raid0_map_submit_bio(mddev, bio);
624 return true;
625}
626
627static void raid0_status(struct seq_file *seq, struct mddev *mddev)
628{
629 seq_printf(m: seq, fmt: " %dk chunks", mddev->chunk_sectors / 2);
630 return;
631}
632
633static void raid0_error(struct mddev *mddev, struct md_rdev *rdev)
634{
635 if (!test_and_set_bit(nr: MD_BROKEN, addr: &mddev->flags)) {
636 char *md_name = mdname(mddev);
637
638 pr_crit("md/raid0%s: Disk failure on %pg detected, failing array.\n",
639 md_name, rdev->bdev);
640 }
641}
642
643static void *raid0_takeover_raid45(struct mddev *mddev)
644{
645 struct md_rdev *rdev;
646 struct r0conf *priv_conf;
647
648 if (mddev->degraded != 1) {
649 pr_warn("md/raid0:%s: raid5 must be degraded! Degraded disks: %d\n",
650 mdname(mddev),
651 mddev->degraded);
652 return ERR_PTR(error: -EINVAL);
653 }
654
655 rdev_for_each(rdev, mddev) {
656 /* check slot number for a disk */
657 if (rdev->raid_disk == mddev->raid_disks-1) {
658 pr_warn("md/raid0:%s: raid5 must have missing parity disk!\n",
659 mdname(mddev));
660 return ERR_PTR(error: -EINVAL);
661 }
662 rdev->sectors = mddev->dev_sectors;
663 }
664
665 /* Set new parameters */
666 mddev->new_level = 0;
667 mddev->new_layout = 0;
668 mddev->new_chunk_sectors = mddev->chunk_sectors;
669 mddev->raid_disks--;
670 mddev->delta_disks = -1;
671 /* make sure it will be not marked as dirty */
672 mddev->resync_offset = MaxSector;
673 mddev_clear_unsupported_flags(mddev, UNSUPPORTED_MDDEV_FLAGS);
674
675 create_strip_zones(mddev, private_conf: &priv_conf);
676
677 return priv_conf;
678}
679
680static void *raid0_takeover_raid10(struct mddev *mddev)
681{
682 struct r0conf *priv_conf;
683
684 /* Check layout:
685 * - far_copies must be 1
686 * - near_copies must be 2
687 * - disks number must be even
688 * - all mirrors must be already degraded
689 */
690 if (mddev->layout != ((1 << 8) + 2)) {
691 pr_warn("md/raid0:%s:: Raid0 cannot takeover layout: 0x%x\n",
692 mdname(mddev),
693 mddev->layout);
694 return ERR_PTR(error: -EINVAL);
695 }
696 if (mddev->raid_disks & 1) {
697 pr_warn("md/raid0:%s: Raid0 cannot takeover Raid10 with odd disk number.\n",
698 mdname(mddev));
699 return ERR_PTR(error: -EINVAL);
700 }
701 if (mddev->degraded != (mddev->raid_disks>>1)) {
702 pr_warn("md/raid0:%s: All mirrors must be already degraded!\n",
703 mdname(mddev));
704 return ERR_PTR(error: -EINVAL);
705 }
706
707 /* Set new parameters */
708 mddev->new_level = 0;
709 mddev->new_layout = 0;
710 mddev->new_chunk_sectors = mddev->chunk_sectors;
711 mddev->delta_disks = - mddev->raid_disks / 2;
712 mddev->raid_disks += mddev->delta_disks;
713 mddev->degraded = 0;
714 /* make sure it will be not marked as dirty */
715 mddev->resync_offset = MaxSector;
716 mddev_clear_unsupported_flags(mddev, UNSUPPORTED_MDDEV_FLAGS);
717
718 create_strip_zones(mddev, private_conf: &priv_conf);
719 return priv_conf;
720}
721
722static void *raid0_takeover_raid1(struct mddev *mddev)
723{
724 struct r0conf *priv_conf;
725 int chunksect;
726
727 /* Check layout:
728 * - (N - 1) mirror drives must be already faulty
729 */
730 if ((mddev->raid_disks - 1) != mddev->degraded) {
731 pr_err("md/raid0:%s: (N - 1) mirrors drives must be already faulty!\n",
732 mdname(mddev));
733 return ERR_PTR(error: -EINVAL);
734 }
735
736 /*
737 * a raid1 doesn't have the notion of chunk size, so
738 * figure out the largest suitable size we can use.
739 */
740 chunksect = 64 * 2; /* 64K by default */
741
742 /* The array must be an exact multiple of chunksize */
743 while (chunksect && (mddev->array_sectors & (chunksect - 1)))
744 chunksect >>= 1;
745
746 if ((chunksect << 9) < PAGE_SIZE)
747 /* array size does not allow a suitable chunk size */
748 return ERR_PTR(error: -EINVAL);
749
750 /* Set new parameters */
751 mddev->new_level = 0;
752 mddev->new_layout = 0;
753 mddev->new_chunk_sectors = chunksect;
754 mddev->chunk_sectors = chunksect;
755 mddev->delta_disks = 1 - mddev->raid_disks;
756 mddev->raid_disks = 1;
757 /* make sure it will be not marked as dirty */
758 mddev->resync_offset = MaxSector;
759 mddev_clear_unsupported_flags(mddev, UNSUPPORTED_MDDEV_FLAGS);
760
761 create_strip_zones(mddev, private_conf: &priv_conf);
762 return priv_conf;
763}
764
765static void *raid0_takeover(struct mddev *mddev)
766{
767 /* raid0 can take over:
768 * raid4 - if all data disks are active.
769 * raid5 - providing it is Raid4 layout and one disk is faulty
770 * raid10 - assuming we have all necessary active disks
771 * raid1 - with (N -1) mirror drives faulty
772 */
773
774 if (mddev->bitmap) {
775 pr_warn("md/raid0: %s: cannot takeover array with bitmap\n",
776 mdname(mddev));
777 return ERR_PTR(error: -EBUSY);
778 }
779 if (mddev->level == 4)
780 return raid0_takeover_raid45(mddev);
781
782 if (mddev->level == 5) {
783 if (mddev->layout == ALGORITHM_PARITY_N)
784 return raid0_takeover_raid45(mddev);
785
786 pr_warn("md/raid0:%s: Raid can only takeover Raid5 with layout: %d\n",
787 mdname(mddev), ALGORITHM_PARITY_N);
788 }
789
790 if (mddev->level == 10)
791 return raid0_takeover_raid10(mddev);
792
793 if (mddev->level == 1)
794 return raid0_takeover_raid1(mddev);
795
796 pr_warn("Takeover from raid%i to raid0 not supported\n",
797 mddev->level);
798
799 return ERR_PTR(error: -EINVAL);
800}
801
802static void raid0_quiesce(struct mddev *mddev, int quiesce)
803{
804}
805
806static struct md_personality raid0_personality=
807{
808 .head = {
809 .type = MD_PERSONALITY,
810 .id = ID_RAID0,
811 .name = "raid0",
812 .owner = THIS_MODULE,
813 },
814
815 .make_request = raid0_make_request,
816 .run = raid0_run,
817 .free = raid0_free,
818 .status = raid0_status,
819 .size = raid0_size,
820 .takeover = raid0_takeover,
821 .quiesce = raid0_quiesce,
822 .error_handler = raid0_error,
823};
824
825static int __init raid0_init(void)
826{
827 return register_md_submodule(msh: &raid0_personality.head);
828}
829
830static void __exit raid0_exit(void)
831{
832 unregister_md_submodule(msh: &raid0_personality.head);
833}
834
835module_init(raid0_init);
836module_exit(raid0_exit);
837MODULE_LICENSE("GPL");
838MODULE_DESCRIPTION("RAID0 (striping) personality for MD");
839MODULE_ALIAS("md-personality-2"); /* RAID0 */
840MODULE_ALIAS("md-raid0");
841MODULE_ALIAS("md-level-0");
842

source code of linux/drivers/md/raid0.c