projects / linux-2.6-block.git / blame_incremental
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (non-incremental) | history | HEAD
Merge tag 'arm-fixes-6.11-3' of git://git.kernel.org/pub/scm/linux/kernel/git/soc/soc
[linux-2.6-block.git] / drivers / md / raid0.c
... / ...
CommitLineData
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(line+len, 200-len, "%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 blksize = 512;
72
73 *private_conf = ERR_PTR(-ENOMEM);
74 if (!conf)
75 return -ENOMEM;
76 rdev_for_each(rdev1, mddev) {
77 pr_debug("md/raid0:%s: looking at %pg\n",
78 mdname(mddev),
79 rdev1->bdev);
80 c = 0;
81
82 /* round size to chunk_size */
83 sectors = rdev1->sectors;
84 sector_div(sectors, mddev->chunk_sectors);
85 rdev1->sectors = sectors * mddev->chunk_sectors;
86
87 blksize = max(blksize, queue_logical_block_size(
88 rdev1->bdev->bd_disk->queue));
89
90 rdev_for_each(rdev2, mddev) {
91 pr_debug("md/raid0:%s: comparing %pg(%llu)"
92 " with %pg(%llu)\n",
93 mdname(mddev),
94 rdev1->bdev,
95 (unsigned long long)rdev1->sectors,
96 rdev2->bdev,
97 (unsigned long long)rdev2->sectors);
98 if (rdev2 == rdev1) {
99 pr_debug("md/raid0:%s: END\n",
100 mdname(mddev));
101 break;
102 }
103 if (rdev2->sectors == rdev1->sectors) {
104 /*
105 * Not unique, don't count it as a new
106 * group
107 */
108 pr_debug("md/raid0:%s: EQUAL\n",
109 mdname(mddev));
110 c = 1;
111 break;
112 }
113 pr_debug("md/raid0:%s: NOT EQUAL\n",
114 mdname(mddev));
115 }
116 if (!c) {
117 pr_debug("md/raid0:%s: ==> UNIQUE\n",
118 mdname(mddev));
119 conf->nr_strip_zones++;
120 pr_debug("md/raid0:%s: %d zones\n",
121 mdname(mddev), conf->nr_strip_zones);
122 }
123 }
124 pr_debug("md/raid0:%s: FINAL %d zones\n",
125 mdname(mddev), conf->nr_strip_zones);
126
127 /*
128 * now since we have the hard sector sizes, we can make sure
129 * chunk size is a multiple of that sector size
130 */
131 if ((mddev->chunk_sectors << 9) % blksize) {
132 pr_warn("md/raid0:%s: chunk_size of %d not multiple of block size %d\n",
133 mdname(mddev),
134 mddev->chunk_sectors << 9, blksize);
135 err = -EINVAL;
136 goto abort;
137 }
138
139 err = -ENOMEM;
140 conf->strip_zone = kcalloc(conf->nr_strip_zones,
141 sizeof(struct strip_zone),
142 GFP_KERNEL);
143 if (!conf->strip_zone)
144 goto abort;
145 conf->devlist = kzalloc(array3_size(sizeof(struct md_rdev *),
146 conf->nr_strip_zones,
147 mddev->raid_disks),
148 GFP_KERNEL);
149 if (!conf->devlist)
150 goto abort;
151
152 /* The first zone must contain all devices, so here we check that
153 * there is a proper alignment of slots to devices and find them all
154 */
155 zone = &conf->strip_zone[0];
156 cnt = 0;
157 smallest = NULL;
158 dev = conf->devlist;
159 err = -EINVAL;
160 rdev_for_each(rdev1, mddev) {
161 int j = rdev1->raid_disk;
162
163 if (mddev->level == 10) {
164 /* taking over a raid10-n2 array */
165 j /= 2;
166 rdev1->new_raid_disk = j;
167 }
168
169 if (mddev->level == 1) {
170 /* taiking over a raid1 array-
171 * we have only one active disk
172 */
173 j = 0;
174 rdev1->new_raid_disk = j;
175 }
176
177 if (j < 0) {
178 pr_warn("md/raid0:%s: remove inactive devices before converting to RAID0\n",
179 mdname(mddev));
180 goto abort;
181 }
182 if (j >= mddev->raid_disks) {
183 pr_warn("md/raid0:%s: bad disk number %d - aborting!\n",
184 mdname(mddev), j);
185 goto abort;
186 }
187 if (dev[j]) {
188 pr_warn("md/raid0:%s: multiple devices for %d - aborting!\n",
189 mdname(mddev), j);
190 goto abort;
191 }
192 dev[j] = rdev1;
193
194 if (!smallest || (rdev1->sectors < smallest->sectors))
195 smallest = rdev1;
196 cnt++;
197 }
198 if (cnt != mddev->raid_disks) {
199 pr_warn("md/raid0:%s: too few disks (%d of %d) - aborting!\n",
200 mdname(mddev), cnt, mddev->raid_disks);
201 goto abort;
202 }
203 zone->nb_dev = cnt;
204 zone->zone_end = smallest->sectors * cnt;
205
206 curr_zone_end = zone->zone_end;
207
208 /* now do the other zones */
209 for (i = 1; i < conf->nr_strip_zones; i++)
210 {
211 int j;
212
213 zone = conf->strip_zone + i;
214 dev = conf->devlist + i * mddev->raid_disks;
215
216 pr_debug("md/raid0:%s: zone %d\n", mdname(mddev), i);
217 zone->dev_start = smallest->sectors;
218 smallest = NULL;
219 c = 0;
220
221 for (j=0; j<cnt; j++) {
222 rdev = conf->devlist[j];
223 if (rdev->sectors <= zone->dev_start) {
224 pr_debug("md/raid0:%s: checking %pg ... nope\n",
225 mdname(mddev),
226 rdev->bdev);
227 continue;
228 }
229 pr_debug("md/raid0:%s: checking %pg ..."
230 " contained as device %d\n",
231 mdname(mddev),
232 rdev->bdev, c);
233 dev[c] = rdev;
234 c++;
235 if (!smallest || rdev->sectors < smallest->sectors) {
236 smallest = rdev;
237 pr_debug("md/raid0:%s: (%llu) is smallest!.\n",
238 mdname(mddev),
239 (unsigned long long)rdev->sectors);
240 }
241 }
242
243 zone->nb_dev = c;
244 sectors = (smallest->sectors - zone->dev_start) * c;
245 pr_debug("md/raid0:%s: zone->nb_dev: %d, sectors: %llu\n",
246 mdname(mddev),
247 zone->nb_dev, (unsigned long long)sectors);
248
249 curr_zone_end += sectors;
250 zone->zone_end = curr_zone_end;
251
252 pr_debug("md/raid0:%s: current zone start: %llu\n",
253 mdname(mddev),
254 (unsigned long long)smallest->sectors);
255 }
256
257 if (conf->nr_strip_zones == 1 || conf->strip_zone[1].nb_dev == 1) {
258 conf->layout = RAID0_ORIG_LAYOUT;
259 } else if (mddev->layout == RAID0_ORIG_LAYOUT ||
260 mddev->layout == RAID0_ALT_MULTIZONE_LAYOUT) {
261 conf->layout = mddev->layout;
262 } else if (default_layout == RAID0_ORIG_LAYOUT ||
263 default_layout == RAID0_ALT_MULTIZONE_LAYOUT) {
264 conf->layout = default_layout;
265 } else {
266 pr_err("md/raid0:%s: cannot assemble multi-zone RAID0 with default_layout setting\n",
267 mdname(mddev));
268 pr_err("md/raid0: please set raid0.default_layout to 1 or 2\n");
269 err = -EOPNOTSUPP;
270 goto abort;
271 }
272
273 if (conf->layout == RAID0_ORIG_LAYOUT) {
274 for (i = 1; i < conf->nr_strip_zones; i++) {
275 sector_t first_sector = conf->strip_zone[i-1].zone_end;
276
277 sector_div(first_sector, mddev->chunk_sectors);
278 zone = conf->strip_zone + i;
279 /* disk_shift is first disk index used in the zone */
280 zone->disk_shift = sector_div(first_sector,
281 zone->nb_dev);
282 }
283 }
284
285 pr_debug("md/raid0:%s: done.\n", mdname(mddev));
286 *private_conf = conf;
287
288 return 0;
289abort:
290 kfree(conf->strip_zone);
291 kfree(conf->devlist);
292 kfree(conf);
293 *private_conf = ERR_PTR(err);
294 return err;
295}
296
297/* Find the zone which holds a particular offset
298 * Update *sectorp to be an offset in that zone
299 */
300static struct strip_zone *find_zone(struct r0conf *conf,
301 sector_t *sectorp)
302{
303 int i;
304 struct strip_zone *z = conf->strip_zone;
305 sector_t sector = *sectorp;
306
307 for (i = 0; i < conf->nr_strip_zones; i++)
308 if (sector < z[i].zone_end) {
309 if (i)
310 *sectorp = sector - z[i-1].zone_end;
311 return z + i;
312 }
313 BUG();
314}
315
316/*
317 * remaps the bio to the target device. we separate two flows.
318 * power 2 flow and a general flow for the sake of performance
319*/
320static struct md_rdev *map_sector(struct mddev *mddev, struct strip_zone *zone,
321 sector_t sector, sector_t *sector_offset)
322{
323 unsigned int sect_in_chunk;
324 sector_t chunk;
325 struct r0conf *conf = mddev->private;
326 int raid_disks = conf->strip_zone[0].nb_dev;
327 unsigned int chunk_sects = mddev->chunk_sectors;
328
329 if (is_power_of_2(chunk_sects)) {
330 int chunksect_bits = ffz(~chunk_sects);
331 /* find the sector offset inside the chunk */
332 sect_in_chunk = sector & (chunk_sects - 1);
333 sector >>= chunksect_bits;
334 /* chunk in zone */
335 chunk = *sector_offset;
336 /* quotient is the chunk in real device*/
337 sector_div(chunk, zone->nb_dev << chunksect_bits);
338 } else{
339 sect_in_chunk = sector_div(sector, chunk_sects);
340 chunk = *sector_offset;
341 sector_div(chunk, chunk_sects * zone->nb_dev);
342 }
343 /*
344 * position the bio over the real device
345 * real sector = chunk in device + starting of zone
346 * + the position in the chunk
347 */
348 *sector_offset = (chunk * chunk_sects) + sect_in_chunk;
349 return conf->devlist[(zone - conf->strip_zone)*raid_disks
350 + sector_div(sector, zone->nb_dev)];
351}
352
353static sector_t raid0_size(struct mddev *mddev, sector_t sectors, int raid_disks)
354{
355 sector_t array_sectors = 0;
356 struct md_rdev *rdev;
357
358 WARN_ONCE(sectors || raid_disks,
359 "%s does not support generic reshape\n", __func__);
360
361 rdev_for_each(rdev, mddev)
362 array_sectors += (rdev->sectors &
363 ~(sector_t)(mddev->chunk_sectors-1));
364
365 return array_sectors;
366}
367
368static void raid0_free(struct mddev *mddev, void *priv)
369{
370 struct r0conf *conf = priv;
371
372 kfree(conf->strip_zone);
373 kfree(conf->devlist);
374 kfree(conf);
375}
376
377static int raid0_set_limits(struct mddev *mddev)
378{
379 struct queue_limits lim;
380 int err;
381
382 md_init_stacking_limits(&lim);
383 lim.max_hw_sectors = mddev->chunk_sectors;
384 lim.max_write_zeroes_sectors = mddev->chunk_sectors;
385 lim.io_min = mddev->chunk_sectors << 9;
386 lim.io_opt = lim.io_min * mddev->raid_disks;
387 err = mddev_stack_rdev_limits(mddev, &lim, MDDEV_STACK_INTEGRITY);
388 if (err) {
389 queue_limits_cancel_update(mddev->gendisk->queue);
390 return err;
391 }
392 return queue_limits_set(mddev->gendisk->queue, &lim);
393}
394
395static int raid0_run(struct mddev *mddev)
396{
397 struct r0conf *conf;
398 int ret;
399
400 if (mddev->chunk_sectors == 0) {
401 pr_warn("md/raid0:%s: chunk size must be set.\n", mdname(mddev));
402 return -EINVAL;
403 }
404 if (md_check_no_bitmap(mddev))
405 return -EINVAL;
406
407 /* if private is not null, we are here after takeover */
408 if (mddev->private == NULL) {
409 ret = create_strip_zones(mddev, &conf);
410 if (ret < 0)
411 return ret;
412 mddev->private = conf;
413 }
414 conf = mddev->private;
415 if (!mddev_is_dm(mddev)) {
416 ret = raid0_set_limits(mddev);
417 if (ret)
418 return ret;
419 }
420
421 /* calculate array device size */
422 md_set_array_sectors(mddev, raid0_size(mddev, 0, 0));
423
424 pr_debug("md/raid0:%s: md_size is %llu sectors.\n",
425 mdname(mddev),
426 (unsigned long long)mddev->array_sectors);
427
428 dump_zones(mddev);
429
430 return md_integrity_register(mddev);
431}
432
433/*
434 * Convert disk_index to the disk order in which it is read/written.
435 * For example, if we have 4 disks, they are numbered 0,1,2,3. If we
436 * write the disks starting at disk 3, then the read/write order would
437 * be disk 3, then 0, then 1, and then disk 2 and we want map_disk_shift()
438 * to map the disks as follows 0,1,2,3 => 1,2,3,0. So disk 0 would map
439 * to 1, 1 to 2, 2 to 3, and 3 to 0. That way we can compare disks in
440 * that 'output' space to understand the read/write disk ordering.
441 */
442static int map_disk_shift(int disk_index, int num_disks, int disk_shift)
443{
444 return ((disk_index + num_disks - disk_shift) % num_disks);
445}
446
447static void raid0_handle_discard(struct mddev *mddev, struct bio *bio)
448{
449 struct r0conf *conf = mddev->private;
450 struct strip_zone *zone;
451 sector_t start = bio->bi_iter.bi_sector;
452 sector_t end;
453 unsigned int stripe_size;
454 sector_t first_stripe_index, last_stripe_index;
455 sector_t start_disk_offset;
456 unsigned int start_disk_index;
457 sector_t end_disk_offset;
458 unsigned int end_disk_index;
459 unsigned int disk;
460 sector_t orig_start, orig_end;
461
462 orig_start = start;
463 zone = find_zone(conf, &start);
464
465 if (bio_end_sector(bio) > zone->zone_end) {
466 struct bio *split = bio_split(bio,
467 zone->zone_end - bio->bi_iter.bi_sector, GFP_NOIO,
468 &mddev->bio_set);
469 bio_chain(split, bio);
470 submit_bio_noacct(bio);
471 bio = split;
472 end = zone->zone_end;
473 } else
474 end = bio_end_sector(bio);
475
476 orig_end = end;
477 if (zone != conf->strip_zone)
478 end = end - zone[-1].zone_end;
479
480 /* Now start and end is the offset in zone */
481 stripe_size = zone->nb_dev * mddev->chunk_sectors;
482
483 first_stripe_index = start;
484 sector_div(first_stripe_index, stripe_size);
485 last_stripe_index = end;
486 sector_div(last_stripe_index, stripe_size);
487
488 /* In the first zone the original and alternate layouts are the same */
489 if ((conf->layout == RAID0_ORIG_LAYOUT) && (zone != conf->strip_zone)) {
490 sector_div(orig_start, mddev->chunk_sectors);
491 start_disk_index = sector_div(orig_start, zone->nb_dev);
492 start_disk_index = map_disk_shift(start_disk_index,
493 zone->nb_dev,
494 zone->disk_shift);
495 sector_div(orig_end, mddev->chunk_sectors);
496 end_disk_index = sector_div(orig_end, zone->nb_dev);
497 end_disk_index = map_disk_shift(end_disk_index,
498 zone->nb_dev, zone->disk_shift);
499 } else {
500 start_disk_index = (int)(start - first_stripe_index * stripe_size) /
501 mddev->chunk_sectors;
502 end_disk_index = (int)(end - last_stripe_index * stripe_size) /
503 mddev->chunk_sectors;
504 }
505 start_disk_offset = ((int)(start - first_stripe_index * stripe_size) %
506 mddev->chunk_sectors) +
507 first_stripe_index * mddev->chunk_sectors;
508 end_disk_offset = ((int)(end - last_stripe_index * stripe_size) %
509 mddev->chunk_sectors) +
510 last_stripe_index * mddev->chunk_sectors;
511
512 for (disk = 0; disk < zone->nb_dev; disk++) {
513 sector_t dev_start, dev_end;
514 struct md_rdev *rdev;
515 int compare_disk;
516
517 compare_disk = map_disk_shift(disk, zone->nb_dev,
518 zone->disk_shift);
519
520 if (compare_disk < start_disk_index)
521 dev_start = (first_stripe_index + 1) *
522 mddev->chunk_sectors;
523 else if (compare_disk > start_disk_index)
524 dev_start = first_stripe_index * mddev->chunk_sectors;
525 else
526 dev_start = start_disk_offset;
527
528 if (compare_disk < end_disk_index)
529 dev_end = (last_stripe_index + 1) * mddev->chunk_sectors;
530 else if (compare_disk > end_disk_index)
531 dev_end = last_stripe_index * mddev->chunk_sectors;
532 else
533 dev_end = end_disk_offset;
534
535 if (dev_end <= dev_start)
536 continue;
537
538 rdev = conf->devlist[(zone - conf->strip_zone) *
539 conf->strip_zone[0].nb_dev + disk];
540 md_submit_discard_bio(mddev, rdev, bio,
541 dev_start + zone->dev_start + rdev->data_offset,
542 dev_end - dev_start);
543 }
544 bio_endio(bio);
545}
546
547static void raid0_map_submit_bio(struct mddev *mddev, struct bio *bio)
548{
549 struct r0conf *conf = mddev->private;
550 struct strip_zone *zone;
551 struct md_rdev *tmp_dev;
552 sector_t bio_sector = bio->bi_iter.bi_sector;
553 sector_t sector = bio_sector;
554
555 md_account_bio(mddev, &bio);
556
557 zone = find_zone(mddev->private, &sector);
558 switch (conf->layout) {
559 case RAID0_ORIG_LAYOUT:
560 tmp_dev = map_sector(mddev, zone, bio_sector, &sector);
561 break;
562 case RAID0_ALT_MULTIZONE_LAYOUT:
563 tmp_dev = map_sector(mddev, zone, sector, &sector);
564 break;
565 default:
566 WARN(1, "md/raid0:%s: Invalid layout\n", mdname(mddev));
567 bio_io_error(bio);
568 return;
569 }
570
571 if (unlikely(is_rdev_broken(tmp_dev))) {
572 bio_io_error(bio);
573 md_error(mddev, tmp_dev);
574 return;
575 }
576
577 bio_set_dev(bio, tmp_dev->bdev);
578 bio->bi_iter.bi_sector = sector + zone->dev_start +
579 tmp_dev->data_offset;
580 mddev_trace_remap(mddev, bio, bio_sector);
581 mddev_check_write_zeroes(mddev, bio);
582 submit_bio_noacct(bio);
583}
584
585static bool raid0_make_request(struct mddev *mddev, struct bio *bio)
586{
587 sector_t sector;
588 unsigned chunk_sects;
589 unsigned sectors;
590
591 if (unlikely(bio->bi_opf & REQ_PREFLUSH)
592 && md_flush_request(mddev, bio))
593 return true;
594
595 if (unlikely((bio_op(bio) == REQ_OP_DISCARD))) {
596 raid0_handle_discard(mddev, bio);
597 return true;
598 }
599
600 sector = bio->bi_iter.bi_sector;
601 chunk_sects = mddev->chunk_sectors;
602
603 sectors = chunk_sects -
604 (likely(is_power_of_2(chunk_sects))
605 ? (sector & (chunk_sects-1))
606 : sector_div(sector, chunk_sects));
607
608 if (sectors < bio_sectors(bio)) {
609 struct bio *split = bio_split(bio, sectors, GFP_NOIO,
610 &mddev->bio_set);
611 bio_chain(split, bio);
612 raid0_map_submit_bio(mddev, bio);
613 bio = split;
614 }
615
616 raid0_map_submit_bio(mddev, bio);
617 return true;
618}
619
620static void raid0_status(struct seq_file *seq, struct mddev *mddev)
621{
622 seq_printf(seq, " %dk chunks", mddev->chunk_sectors / 2);
623 return;
624}
625
626static void raid0_error(struct mddev *mddev, struct md_rdev *rdev)
627{
628 if (!test_and_set_bit(MD_BROKEN, &mddev->flags)) {
629 char *md_name = mdname(mddev);
630
631 pr_crit("md/raid0%s: Disk failure on %pg detected, failing array.\n",
632 md_name, rdev->bdev);
633 }
634}
635
636static void *raid0_takeover_raid45(struct mddev *mddev)
637{
638 struct md_rdev *rdev;
639 struct r0conf *priv_conf;
640
641 if (mddev->degraded != 1) {
642 pr_warn("md/raid0:%s: raid5 must be degraded! Degraded disks: %d\n",
643 mdname(mddev),
644 mddev->degraded);
645 return ERR_PTR(-EINVAL);
646 }
647
648 rdev_for_each(rdev, mddev) {
649 /* check slot number for a disk */
650 if (rdev->raid_disk == mddev->raid_disks-1) {
651 pr_warn("md/raid0:%s: raid5 must have missing parity disk!\n",
652 mdname(mddev));
653 return ERR_PTR(-EINVAL);
654 }
655 rdev->sectors = mddev->dev_sectors;
656 }
657
658 /* Set new parameters */
659 mddev->new_level = 0;
660 mddev->new_layout = 0;
661 mddev->new_chunk_sectors = mddev->chunk_sectors;
662 mddev->raid_disks--;
663 mddev->delta_disks = -1;
664 /* make sure it will be not marked as dirty */
665 mddev->recovery_cp = MaxSector;
666 mddev_clear_unsupported_flags(mddev, UNSUPPORTED_MDDEV_FLAGS);
667
668 create_strip_zones(mddev, &priv_conf);
669
670 return priv_conf;
671}
672
673static void *raid0_takeover_raid10(struct mddev *mddev)
674{
675 struct r0conf *priv_conf;
676
677 /* Check layout:
678 * - far_copies must be 1
679 * - near_copies must be 2
680 * - disks number must be even
681 * - all mirrors must be already degraded
682 */
683 if (mddev->layout != ((1 << 8) + 2)) {
684 pr_warn("md/raid0:%s:: Raid0 cannot takeover layout: 0x%x\n",
685 mdname(mddev),
686 mddev->layout);
687 return ERR_PTR(-EINVAL);
688 }
689 if (mddev->raid_disks & 1) {
690 pr_warn("md/raid0:%s: Raid0 cannot takeover Raid10 with odd disk number.\n",
691 mdname(mddev));
692 return ERR_PTR(-EINVAL);
693 }
694 if (mddev->degraded != (mddev->raid_disks>>1)) {
695 pr_warn("md/raid0:%s: All mirrors must be already degraded!\n",
696 mdname(mddev));
697 return ERR_PTR(-EINVAL);
698 }
699
700 /* Set new parameters */
701 mddev->new_level = 0;
702 mddev->new_layout = 0;
703 mddev->new_chunk_sectors = mddev->chunk_sectors;
704 mddev->delta_disks = - mddev->raid_disks / 2;
705 mddev->raid_disks += mddev->delta_disks;
706 mddev->degraded = 0;
707 /* make sure it will be not marked as dirty */
708 mddev->recovery_cp = MaxSector;
709 mddev_clear_unsupported_flags(mddev, UNSUPPORTED_MDDEV_FLAGS);
710
711 create_strip_zones(mddev, &priv_conf);
712 return priv_conf;
713}
714
715static void *raid0_takeover_raid1(struct mddev *mddev)
716{
717 struct r0conf *priv_conf;
718 int chunksect;
719
720 /* Check layout:
721 * - (N - 1) mirror drives must be already faulty
722 */
723 if ((mddev->raid_disks - 1) != mddev->degraded) {
724 pr_err("md/raid0:%s: (N - 1) mirrors drives must be already faulty!\n",
725 mdname(mddev));
726 return ERR_PTR(-EINVAL);
727 }
728
729 /*
730 * a raid1 doesn't have the notion of chunk size, so
731 * figure out the largest suitable size we can use.
732 */
733 chunksect = 64 * 2; /* 64K by default */
734
735 /* The array must be an exact multiple of chunksize */
736 while (chunksect && (mddev->array_sectors & (chunksect - 1)))
737 chunksect >>= 1;
738
739 if ((chunksect << 9) < PAGE_SIZE)
740 /* array size does not allow a suitable chunk size */
741 return ERR_PTR(-EINVAL);
742
743 /* Set new parameters */
744 mddev->new_level = 0;
745 mddev->new_layout = 0;
746 mddev->new_chunk_sectors = chunksect;
747 mddev->chunk_sectors = chunksect;
748 mddev->delta_disks = 1 - mddev->raid_disks;
749 mddev->raid_disks = 1;
750 /* make sure it will be not marked as dirty */
751 mddev->recovery_cp = MaxSector;
752 mddev_clear_unsupported_flags(mddev, UNSUPPORTED_MDDEV_FLAGS);
753
754 create_strip_zones(mddev, &priv_conf);
755 return priv_conf;
756}
757
758static void *raid0_takeover(struct mddev *mddev)
759{
760 /* raid0 can take over:
761 * raid4 - if all data disks are active.
762 * raid5 - providing it is Raid4 layout and one disk is faulty
763 * raid10 - assuming we have all necessary active disks
764 * raid1 - with (N -1) mirror drives faulty
765 */
766
767 if (mddev->bitmap) {
768 pr_warn("md/raid0: %s: cannot takeover array with bitmap\n",
769 mdname(mddev));
770 return ERR_PTR(-EBUSY);
771 }
772 if (mddev->level == 4)
773 return raid0_takeover_raid45(mddev);
774
775 if (mddev->level == 5) {
776 if (mddev->layout == ALGORITHM_PARITY_N)
777 return raid0_takeover_raid45(mddev);
778
779 pr_warn("md/raid0:%s: Raid can only takeover Raid5 with layout: %d\n",
780 mdname(mddev), ALGORITHM_PARITY_N);
781 }
782
783 if (mddev->level == 10)
784 return raid0_takeover_raid10(mddev);
785
786 if (mddev->level == 1)
787 return raid0_takeover_raid1(mddev);
788
789 pr_warn("Takeover from raid%i to raid0 not supported\n",
790 mddev->level);
791
792 return ERR_PTR(-EINVAL);
793}
794
795static void raid0_quiesce(struct mddev *mddev, int quiesce)
796{
797}
798
799static struct md_personality raid0_personality=
800{
801 .name = "raid0",
802 .level = 0,
803 .owner = THIS_MODULE,
804 .make_request = raid0_make_request,
805 .run = raid0_run,
806 .free = raid0_free,
807 .status = raid0_status,
808 .size = raid0_size,
809 .takeover = raid0_takeover,
810 .quiesce = raid0_quiesce,
811 .error_handler = raid0_error,
812};
813
814static int __init raid0_init (void)
815{
816 return register_md_personality (&raid0_personality);
817}
818
819static void raid0_exit (void)
820{
821 unregister_md_personality (&raid0_personality);
822}
823
824module_init(raid0_init);
825module_exit(raid0_exit);
826MODULE_LICENSE("GPL");
827MODULE_DESCRIPTION("RAID0 (striping) personality for MD");
828MODULE_ALIAS("md-personality-2"); /* RAID0 */
829MODULE_ALIAS("md-raid0");
830MODULE_ALIAS("md-level-0");
Linux 6.x block layer and io_uring tree(s)