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
22 static int default_layout = 0;
23 module_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 */
dump_zones(struct mddev *mddev)35 static void dump_zones(struct mddev *mddev)
36 {
37 int j, k;
38 sector_t zone_size = 0;
39 sector_t zone_start = 0;
40 char b[BDEVNAME_SIZE];
41 struct r0conf *conf = mddev->private;
42 int raid_disks = conf->strip_zone[0].nb_dev;
43 pr_debug("md: RAID0 configuration for %s - %d zone%s\n",
44 mdname(mddev),
45 conf->nr_strip_zones, conf->nr_strip_zones==1?"":"s");
46 for (j = 0; j < conf->nr_strip_zones; j++) {
47 char line[200];
48 int len = 0;
49
50 for (k = 0; k < conf->strip_zone[j].nb_dev; k++)
51 len += scnprintf(line+len, 200-len, "%s%s", k?"/":"",
52 bdevname(conf->devlist[j*raid_disks
53 + k]->bdev, b));
54 pr_debug("md: zone%d=[%s]\n", j, line);
55
56 zone_size = conf->strip_zone[j].zone_end - zone_start;
57 pr_debug(" zone-offset=%10lluKB, device-offset=%10lluKB, size=%10lluKB\n",
58 (unsigned long long)zone_start>>1,
59 (unsigned long long)conf->strip_zone[j].dev_start>>1,
60 (unsigned long long)zone_size>>1);
61 zone_start = conf->strip_zone[j].zone_end;
62 }
63 }
64
create_strip_zones(struct mddev *mddev, struct r0conf **private_conf)65 static int create_strip_zones(struct mddev *mddev, struct r0conf **private_conf)
66 {
67 int i, c, err;
68 sector_t curr_zone_end, sectors;
69 struct md_rdev *smallest, *rdev1, *rdev2, *rdev, **dev;
70 struct strip_zone *zone;
71 int cnt;
72 char b[BDEVNAME_SIZE];
73 char b2[BDEVNAME_SIZE];
74 struct r0conf *conf = kzalloc(sizeof(*conf), GFP_KERNEL);
75 unsigned blksize = 512;
76
77 *private_conf = ERR_PTR(-ENOMEM);
78 if (!conf)
79 return -ENOMEM;
80 rdev_for_each(rdev1, mddev) {
81 pr_debug("md/raid0:%s: looking at %s\n",
82 mdname(mddev),
83 bdevname(rdev1->bdev, b));
84 c = 0;
85
86 /* round size to chunk_size */
87 sectors = rdev1->sectors;
88 sector_div(sectors, mddev->chunk_sectors);
89 rdev1->sectors = sectors * mddev->chunk_sectors;
90
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 %s(%llu)"
96 " with %s(%llu)\n",
97 mdname(mddev),
98 bdevname(rdev1->bdev,b),
99 (unsigned long long)rdev1->sectors,
100 bdevname(rdev2->bdev,b2),
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 %s ... nope\n",
229 mdname(mddev),
230 bdevname(rdev->bdev, b));
231 continue;
232 }
233 pr_debug("md/raid0:%s: checking %s ..."
234 " contained as device %d\n",
235 mdname(mddev),
236 bdevname(rdev->bdev, b), 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;
293 abort:
294 kfree(conf->strip_zone);
295 kfree(conf->devlist);
296 kfree(conf);
297 *private_conf = ERR_PTR(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 */
find_zone(struct r0conf *conf, sector_t *sectorp)304 static 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 */
map_sector(struct mddev *mddev, struct strip_zone *zone, sector_t sector, sector_t *sector_offset)324 static 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(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
raid0_size(struct mddev *mddev, sector_t sectors, int raid_disks)357 static 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
372 static void raid0_free(struct mddev *mddev, void *priv);
373
raid0_run(struct mddev *mddev)374 static int raid0_run(struct mddev *mddev)
375 {
376 struct r0conf *conf;
377 int ret;
378
379 if (mddev->chunk_sectors == 0) {
380 pr_warn("md/raid0:%s: chunk size must be set.\n", mdname(mddev));
381 return -EINVAL;
382 }
383 if (md_check_no_bitmap(mddev))
384 return -EINVAL;
385
386 /* if private is not null, we are here after takeover */
387 if (mddev->private == NULL) {
388 ret = create_strip_zones(mddev, &conf);
389 if (ret < 0)
390 return ret;
391 mddev->private = conf;
392 }
393 conf = mddev->private;
394 if (mddev->queue) {
395 struct md_rdev *rdev;
396 bool discard_supported = false;
397
398 blk_queue_max_hw_sectors(mddev->queue, mddev->chunk_sectors);
399 blk_queue_max_write_same_sectors(mddev->queue, mddev->chunk_sectors);
400 blk_queue_max_write_zeroes_sectors(mddev->queue, mddev->chunk_sectors);
401 blk_queue_max_discard_sectors(mddev->queue, UINT_MAX);
402
403 blk_queue_io_min(mddev->queue, mddev->chunk_sectors << 9);
404 blk_queue_io_opt(mddev->queue,
405 (mddev->chunk_sectors << 9) * mddev->raid_disks);
406
407 rdev_for_each(rdev, mddev) {
408 disk_stack_limits(mddev->gendisk, rdev->bdev,
409 rdev->data_offset << 9);
410 if (blk_queue_discard(bdev_get_queue(rdev->bdev)))
411 discard_supported = true;
412 }
413 if (!discard_supported)
414 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, mddev->queue);
415 else
416 blk_queue_flag_set(QUEUE_FLAG_DISCARD, mddev->queue);
417 }
418
419 /* calculate array device size */
420 md_set_array_sectors(mddev, raid0_size(mddev, 0, 0));
421
422 pr_debug("md/raid0:%s: md_size is %llu sectors.\n",
423 mdname(mddev),
424 (unsigned long long)mddev->array_sectors);
425
426 dump_zones(mddev);
427
428 ret = md_integrity_register(mddev);
429
430 return ret;
431 }
432
raid0_free(struct mddev *mddev, void *priv)433 static void raid0_free(struct mddev *mddev, void *priv)
434 {
435 struct r0conf *conf = priv;
436
437 kfree(conf->strip_zone);
438 kfree(conf->devlist);
439 kfree(conf);
440 }
441
442 /*
443 * Convert disk_index to the disk order in which it is read/written.
444 * For example, if we have 4 disks, they are numbered 0,1,2,3. If we
445 * write the disks starting at disk 3, then the read/write order would
446 * be disk 3, then 0, then 1, and then disk 2 and we want map_disk_shift()
447 * to map the disks as follows 0,1,2,3 => 1,2,3,0. So disk 0 would map
448 * to 1, 1 to 2, 2 to 3, and 3 to 0. That way we can compare disks in
449 * that 'output' space to understand the read/write disk ordering.
450 */
map_disk_shift(int disk_index, int num_disks, int disk_shift)451 static int map_disk_shift(int disk_index, int num_disks, int disk_shift)
452 {
453 return ((disk_index + num_disks - disk_shift) % num_disks);
454 }
455
raid0_handle_discard(struct mddev *mddev, struct bio *bio)456 static void raid0_handle_discard(struct mddev *mddev, struct bio *bio)
457 {
458 struct r0conf *conf = mddev->private;
459 struct strip_zone *zone;
460 sector_t start = bio->bi_iter.bi_sector;
461 sector_t end;
462 unsigned int stripe_size;
463 sector_t first_stripe_index, last_stripe_index;
464 sector_t start_disk_offset;
465 unsigned int start_disk_index;
466 sector_t end_disk_offset;
467 unsigned int end_disk_index;
468 unsigned int disk;
469 sector_t orig_start, orig_end;
470
471 orig_start = start;
472 zone = find_zone(conf, &start);
473
474 if (bio_end_sector(bio) > zone->zone_end) {
475 struct bio *split = bio_split(bio,
476 zone->zone_end - bio->bi_iter.bi_sector, GFP_NOIO,
477 &mddev->bio_set);
478 bio_chain(split, bio);
479 submit_bio_noacct(bio);
480 bio = split;
481 end = zone->zone_end;
482 } else
483 end = bio_end_sector(bio);
484
485 orig_end = end;
486 if (zone != conf->strip_zone)
487 end = end - zone[-1].zone_end;
488
489 /* Now start and end is the offset in zone */
490 stripe_size = zone->nb_dev * mddev->chunk_sectors;
491
492 first_stripe_index = start;
493 sector_div(first_stripe_index, stripe_size);
494 last_stripe_index = end;
495 sector_div(last_stripe_index, stripe_size);
496
497 /* In the first zone the original and alternate layouts are the same */
498 if ((conf->layout == RAID0_ORIG_LAYOUT) && (zone != conf->strip_zone)) {
499 sector_div(orig_start, mddev->chunk_sectors);
500 start_disk_index = sector_div(orig_start, zone->nb_dev);
501 start_disk_index = map_disk_shift(start_disk_index,
502 zone->nb_dev,
503 zone->disk_shift);
504 sector_div(orig_end, mddev->chunk_sectors);
505 end_disk_index = sector_div(orig_end, zone->nb_dev);
506 end_disk_index = map_disk_shift(end_disk_index,
507 zone->nb_dev, zone->disk_shift);
508 } else {
509 start_disk_index = (int)(start - first_stripe_index * stripe_size) /
510 mddev->chunk_sectors;
511 end_disk_index = (int)(end - last_stripe_index * stripe_size) /
512 mddev->chunk_sectors;
513 }
514 start_disk_offset = ((int)(start - first_stripe_index * stripe_size) %
515 mddev->chunk_sectors) +
516 first_stripe_index * mddev->chunk_sectors;
517 end_disk_offset = ((int)(end - last_stripe_index * stripe_size) %
518 mddev->chunk_sectors) +
519 last_stripe_index * mddev->chunk_sectors;
520
521 for (disk = 0; disk < zone->nb_dev; disk++) {
522 sector_t dev_start, dev_end;
523 struct bio *discard_bio = NULL;
524 struct md_rdev *rdev;
525 int compare_disk;
526
527 compare_disk = map_disk_shift(disk, zone->nb_dev,
528 zone->disk_shift);
529
530 if (compare_disk < start_disk_index)
531 dev_start = (first_stripe_index + 1) *
532 mddev->chunk_sectors;
533 else if (compare_disk > start_disk_index)
534 dev_start = first_stripe_index * mddev->chunk_sectors;
535 else
536 dev_start = start_disk_offset;
537
538 if (compare_disk < end_disk_index)
539 dev_end = (last_stripe_index + 1) * mddev->chunk_sectors;
540 else if (compare_disk > end_disk_index)
541 dev_end = last_stripe_index * mddev->chunk_sectors;
542 else
543 dev_end = end_disk_offset;
544
545 if (dev_end <= dev_start)
546 continue;
547
548 rdev = conf->devlist[(zone - conf->strip_zone) *
549 conf->strip_zone[0].nb_dev + disk];
550 if (__blkdev_issue_discard(rdev->bdev,
551 dev_start + zone->dev_start + rdev->data_offset,
552 dev_end - dev_start, GFP_NOIO, 0, &discard_bio) ||
553 !discard_bio)
554 continue;
555 bio_chain(discard_bio, bio);
556 bio_clone_blkg_association(discard_bio, bio);
557 if (mddev->gendisk)
558 trace_block_bio_remap(bdev_get_queue(rdev->bdev),
559 discard_bio, disk_devt(mddev->gendisk),
560 bio->bi_iter.bi_sector);
561 submit_bio_noacct(discard_bio);
562 }
563 bio_endio(bio);
564 }
565
raid0_make_request(struct mddev *mddev, struct bio *bio)566 static bool raid0_make_request(struct mddev *mddev, struct bio *bio)
567 {
568 struct r0conf *conf = mddev->private;
569 struct strip_zone *zone;
570 struct md_rdev *tmp_dev;
571 sector_t bio_sector;
572 sector_t sector;
573 sector_t orig_sector;
574 unsigned chunk_sects;
575 unsigned sectors;
576
577 if (unlikely(bio->bi_opf & REQ_PREFLUSH)
578 && md_flush_request(mddev, bio))
579 return true;
580
581 if (unlikely((bio_op(bio) == REQ_OP_DISCARD))) {
582 raid0_handle_discard(mddev, bio);
583 return true;
584 }
585
586 bio_sector = bio->bi_iter.bi_sector;
587 sector = bio_sector;
588 chunk_sects = mddev->chunk_sectors;
589
590 sectors = chunk_sects -
591 (likely(is_power_of_2(chunk_sects))
592 ? (sector & (chunk_sects-1))
593 : sector_div(sector, chunk_sects));
594
595 /* Restore due to sector_div */
596 sector = bio_sector;
597
598 if (sectors < bio_sectors(bio)) {
599 struct bio *split = bio_split(bio, sectors, GFP_NOIO,
600 &mddev->bio_set);
601 bio_chain(split, bio);
602 submit_bio_noacct(bio);
603 bio = split;
604 }
605
606 orig_sector = sector;
607 zone = find_zone(mddev->private, §or);
608 switch (conf->layout) {
609 case RAID0_ORIG_LAYOUT:
610 tmp_dev = map_sector(mddev, zone, orig_sector, §or);
611 break;
612 case RAID0_ALT_MULTIZONE_LAYOUT:
613 tmp_dev = map_sector(mddev, zone, sector, §or);
614 break;
615 default:
616 WARN(1, "md/raid0:%s: Invalid layout\n", mdname(mddev));
617 bio_io_error(bio);
618 return true;
619 }
620
621 if (unlikely(is_mddev_broken(tmp_dev, "raid0"))) {
622 bio_io_error(bio);
623 return true;
624 }
625
626 bio_set_dev(bio, tmp_dev->bdev);
627 bio->bi_iter.bi_sector = sector + zone->dev_start +
628 tmp_dev->data_offset;
629
630 if (mddev->gendisk)
631 trace_block_bio_remap(bio->bi_disk->queue, bio,
632 disk_devt(mddev->gendisk), bio_sector);
633 mddev_check_writesame(mddev, bio);
634 mddev_check_write_zeroes(mddev, bio);
635 submit_bio_noacct(bio);
636 return true;
637 }
638
raid0_status(struct seq_file *seq, struct mddev *mddev)639 static void raid0_status(struct seq_file *seq, struct mddev *mddev)
640 {
641 seq_printf(seq, " %dk chunks", mddev->chunk_sectors / 2);
642 return;
643 }
644
raid0_takeover_raid45(struct mddev *mddev)645 static void *raid0_takeover_raid45(struct mddev *mddev)
646 {
647 struct md_rdev *rdev;
648 struct r0conf *priv_conf;
649
650 if (mddev->degraded != 1) {
651 pr_warn("md/raid0:%s: raid5 must be degraded! Degraded disks: %d\n",
652 mdname(mddev),
653 mddev->degraded);
654 return ERR_PTR(-EINVAL);
655 }
656
657 rdev_for_each(rdev, mddev) {
658 /* check slot number for a disk */
659 if (rdev->raid_disk == mddev->raid_disks-1) {
660 pr_warn("md/raid0:%s: raid5 must have missing parity disk!\n",
661 mdname(mddev));
662 return ERR_PTR(-EINVAL);
663 }
664 rdev->sectors = mddev->dev_sectors;
665 }
666
667 /* Set new parameters */
668 mddev->new_level = 0;
669 mddev->new_layout = 0;
670 mddev->new_chunk_sectors = mddev->chunk_sectors;
671 mddev->raid_disks--;
672 mddev->delta_disks = -1;
673 /* make sure it will be not marked as dirty */
674 mddev->recovery_cp = MaxSector;
675 mddev_clear_unsupported_flags(mddev, UNSUPPORTED_MDDEV_FLAGS);
676
677 create_strip_zones(mddev, &priv_conf);
678
679 return priv_conf;
680 }
681
raid0_takeover_raid10(struct mddev *mddev)682 static void *raid0_takeover_raid10(struct mddev *mddev)
683 {
684 struct r0conf *priv_conf;
685
686 /* Check layout:
687 * - far_copies must be 1
688 * - near_copies must be 2
689 * - disks number must be even
690 * - all mirrors must be already degraded
691 */
692 if (mddev->layout != ((1 << 8) + 2)) {
693 pr_warn("md/raid0:%s:: Raid0 cannot takeover layout: 0x%x\n",
694 mdname(mddev),
695 mddev->layout);
696 return ERR_PTR(-EINVAL);
697 }
698 if (mddev->raid_disks & 1) {
699 pr_warn("md/raid0:%s: Raid0 cannot takeover Raid10 with odd disk number.\n",
700 mdname(mddev));
701 return ERR_PTR(-EINVAL);
702 }
703 if (mddev->degraded != (mddev->raid_disks>>1)) {
704 pr_warn("md/raid0:%s: All mirrors must be already degraded!\n",
705 mdname(mddev));
706 return ERR_PTR(-EINVAL);
707 }
708
709 /* Set new parameters */
710 mddev->new_level = 0;
711 mddev->new_layout = 0;
712 mddev->new_chunk_sectors = mddev->chunk_sectors;
713 mddev->delta_disks = - mddev->raid_disks / 2;
714 mddev->raid_disks += mddev->delta_disks;
715 mddev->degraded = 0;
716 /* make sure it will be not marked as dirty */
717 mddev->recovery_cp = MaxSector;
718 mddev_clear_unsupported_flags(mddev, UNSUPPORTED_MDDEV_FLAGS);
719
720 create_strip_zones(mddev, &priv_conf);
721 return priv_conf;
722 }
723
raid0_takeover_raid1(struct mddev *mddev)724 static void *raid0_takeover_raid1(struct mddev *mddev)
725 {
726 struct r0conf *priv_conf;
727 int chunksect;
728
729 /* Check layout:
730 * - (N - 1) mirror drives must be already faulty
731 */
732 if ((mddev->raid_disks - 1) != mddev->degraded) {
733 pr_err("md/raid0:%s: (N - 1) mirrors drives must be already faulty!\n",
734 mdname(mddev));
735 return ERR_PTR(-EINVAL);
736 }
737
738 /*
739 * a raid1 doesn't have the notion of chunk size, so
740 * figure out the largest suitable size we can use.
741 */
742 chunksect = 64 * 2; /* 64K by default */
743
744 /* The array must be an exact multiple of chunksize */
745 while (chunksect && (mddev->array_sectors & (chunksect - 1)))
746 chunksect >>= 1;
747
748 if ((chunksect << 9) < PAGE_SIZE)
749 /* array size does not allow a suitable chunk size */
750 return ERR_PTR(-EINVAL);
751
752 /* Set new parameters */
753 mddev->new_level = 0;
754 mddev->new_layout = 0;
755 mddev->new_chunk_sectors = chunksect;
756 mddev->chunk_sectors = chunksect;
757 mddev->delta_disks = 1 - mddev->raid_disks;
758 mddev->raid_disks = 1;
759 /* make sure it will be not marked as dirty */
760 mddev->recovery_cp = MaxSector;
761 mddev_clear_unsupported_flags(mddev, UNSUPPORTED_MDDEV_FLAGS);
762
763 create_strip_zones(mddev, &priv_conf);
764 return priv_conf;
765 }
766
raid0_takeover(struct mddev *mddev)767 static void *raid0_takeover(struct mddev *mddev)
768 {
769 /* raid0 can take over:
770 * raid4 - if all data disks are active.
771 * raid5 - providing it is Raid4 layout and one disk is faulty
772 * raid10 - assuming we have all necessary active disks
773 * raid1 - with (N -1) mirror drives faulty
774 */
775
776 if (mddev->bitmap) {
777 pr_warn("md/raid0: %s: cannot takeover array with bitmap\n",
778 mdname(mddev));
779 return ERR_PTR(-EBUSY);
780 }
781 if (mddev->level == 4)
782 return raid0_takeover_raid45(mddev);
783
784 if (mddev->level == 5) {
785 if (mddev->layout == ALGORITHM_PARITY_N)
786 return raid0_takeover_raid45(mddev);
787
788 pr_warn("md/raid0:%s: Raid can only takeover Raid5 with layout: %d\n",
789 mdname(mddev), ALGORITHM_PARITY_N);
790 }
791
792 if (mddev->level == 10)
793 return raid0_takeover_raid10(mddev);
794
795 if (mddev->level == 1)
796 return raid0_takeover_raid1(mddev);
797
798 pr_warn("Takeover from raid%i to raid0 not supported\n",
799 mddev->level);
800
801 return ERR_PTR(-EINVAL);
802 }
803
raid0_quiesce(struct mddev *mddev, int quiesce)804 static void raid0_quiesce(struct mddev *mddev, int quiesce)
805 {
806 }
807
808 static struct md_personality raid0_personality=
809 {
810 .name = "raid0",
811 .level = 0,
812 .owner = THIS_MODULE,
813 .make_request = raid0_make_request,
814 .run = raid0_run,
815 .free = raid0_free,
816 .status = raid0_status,
817 .size = raid0_size,
818 .takeover = raid0_takeover,
819 .quiesce = raid0_quiesce,
820 };
821
raid0_init(void)822 static int __init raid0_init (void)
823 {
824 return register_md_personality (&raid0_personality);
825 }
826
raid0_exit(void)827 static void raid0_exit (void)
828 {
829 unregister_md_personality (&raid0_personality);
830 }
831
832 module_init(raid0_init);
833 module_exit(raid0_exit);
834 MODULE_LICENSE("GPL");
835 MODULE_DESCRIPTION("RAID0 (striping) personality for MD");
836 MODULE_ALIAS("md-personality-2"); /* RAID0 */
837 MODULE_ALIAS("md-raid0");
838 MODULE_ALIAS("md-level-0");
839