Loading...
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 lim.features |= BLK_FEAT_ATOMIC_WRITES_STACKED;
388 err = mddev_stack_rdev_limits(mddev, &lim, MDDEV_STACK_INTEGRITY);
389 if (err)
390 return err;
391 return queue_limits_set(mddev->gendisk->queue, &lim);
392}
393
394static int raid0_run(struct mddev *mddev)
395{
396 struct r0conf *conf;
397 int ret;
398
399 if (mddev->chunk_sectors == 0) {
400 pr_warn("md/raid0:%s: chunk size must be set.\n", mdname(mddev));
401 return -EINVAL;
402 }
403 if (md_check_no_bitmap(mddev))
404 return -EINVAL;
405
406 /* if private is not null, we are here after takeover */
407 if (mddev->private == NULL) {
408 ret = create_strip_zones(mddev, &conf);
409 if (ret < 0)
410 return ret;
411 mddev->private = conf;
412 }
413 conf = mddev->private;
414 if (!mddev_is_dm(mddev)) {
415 ret = raid0_set_limits(mddev);
416 if (ret)
417 return ret;
418 }
419
420 /* calculate array device size */
421 md_set_array_sectors(mddev, raid0_size(mddev, 0, 0));
422
423 pr_debug("md/raid0:%s: md_size is %llu sectors.\n",
424 mdname(mddev),
425 (unsigned long long)mddev->array_sectors);
426
427 dump_zones(mddev);
428
429 return md_integrity_register(mddev);
430}
431
432/*
433 * Convert disk_index to the disk order in which it is read/written.
434 * For example, if we have 4 disks, they are numbered 0,1,2,3. If we
435 * write the disks starting at disk 3, then the read/write order would
436 * be disk 3, then 0, then 1, and then disk 2 and we want map_disk_shift()
437 * to map the disks as follows 0,1,2,3 => 1,2,3,0. So disk 0 would map
438 * to 1, 1 to 2, 2 to 3, and 3 to 0. That way we can compare disks in
439 * that 'output' space to understand the read/write disk ordering.
440 */
441static int map_disk_shift(int disk_index, int num_disks, int disk_shift)
442{
443 return ((disk_index + num_disks - disk_shift) % num_disks);
444}
445
446static void raid0_handle_discard(struct mddev *mddev, struct bio *bio)
447{
448 struct r0conf *conf = mddev->private;
449 struct strip_zone *zone;
450 sector_t start = bio->bi_iter.bi_sector;
451 sector_t end;
452 unsigned int stripe_size;
453 sector_t first_stripe_index, last_stripe_index;
454 sector_t start_disk_offset;
455 unsigned int start_disk_index;
456 sector_t end_disk_offset;
457 unsigned int end_disk_index;
458 unsigned int disk;
459 sector_t orig_start, orig_end;
460
461 orig_start = start;
462 zone = find_zone(conf, &start);
463
464 if (bio_end_sector(bio) > zone->zone_end) {
465 struct bio *split = bio_split(bio,
466 zone->zone_end - bio->bi_iter.bi_sector, GFP_NOIO,
467 &mddev->bio_set);
468
469 if (IS_ERR(split)) {
470 bio->bi_status = errno_to_blk_status(PTR_ERR(split));
471 bio_endio(bio);
472 return;
473 }
474 bio_chain(split, bio);
475 submit_bio_noacct(bio);
476 bio = split;
477 end = zone->zone_end;
478 } else
479 end = bio_end_sector(bio);
480
481 orig_end = end;
482 if (zone != conf->strip_zone)
483 end = end - zone[-1].zone_end;
484
485 /* Now start and end is the offset in zone */
486 stripe_size = zone->nb_dev * mddev->chunk_sectors;
487
488 first_stripe_index = start;
489 sector_div(first_stripe_index, stripe_size);
490 last_stripe_index = end;
491 sector_div(last_stripe_index, stripe_size);
492
493 /* In the first zone the original and alternate layouts are the same */
494 if ((conf->layout == RAID0_ORIG_LAYOUT) && (zone != conf->strip_zone)) {
495 sector_div(orig_start, mddev->chunk_sectors);
496 start_disk_index = sector_div(orig_start, zone->nb_dev);
497 start_disk_index = map_disk_shift(start_disk_index,
498 zone->nb_dev,
499 zone->disk_shift);
500 sector_div(orig_end, mddev->chunk_sectors);
501 end_disk_index = sector_div(orig_end, zone->nb_dev);
502 end_disk_index = map_disk_shift(end_disk_index,
503 zone->nb_dev, zone->disk_shift);
504 } else {
505 start_disk_index = (int)(start - first_stripe_index * stripe_size) /
506 mddev->chunk_sectors;
507 end_disk_index = (int)(end - last_stripe_index * stripe_size) /
508 mddev->chunk_sectors;
509 }
510 start_disk_offset = ((int)(start - first_stripe_index * stripe_size) %
511 mddev->chunk_sectors) +
512 first_stripe_index * mddev->chunk_sectors;
513 end_disk_offset = ((int)(end - last_stripe_index * stripe_size) %
514 mddev->chunk_sectors) +
515 last_stripe_index * mddev->chunk_sectors;
516
517 for (disk = 0; disk < zone->nb_dev; disk++) {
518 sector_t dev_start, dev_end;
519 struct md_rdev *rdev;
520 int compare_disk;
521
522 compare_disk = map_disk_shift(disk, zone->nb_dev,
523 zone->disk_shift);
524
525 if (compare_disk < start_disk_index)
526 dev_start = (first_stripe_index + 1) *
527 mddev->chunk_sectors;
528 else if (compare_disk > start_disk_index)
529 dev_start = first_stripe_index * mddev->chunk_sectors;
530 else
531 dev_start = start_disk_offset;
532
533 if (compare_disk < end_disk_index)
534 dev_end = (last_stripe_index + 1) * mddev->chunk_sectors;
535 else if (compare_disk > end_disk_index)
536 dev_end = last_stripe_index * mddev->chunk_sectors;
537 else
538 dev_end = end_disk_offset;
539
540 if (dev_end <= dev_start)
541 continue;
542
543 rdev = conf->devlist[(zone - conf->strip_zone) *
544 conf->strip_zone[0].nb_dev + disk];
545 md_submit_discard_bio(mddev, rdev, bio,
546 dev_start + zone->dev_start + rdev->data_offset,
547 dev_end - dev_start);
548 }
549 bio_endio(bio);
550}
551
552static void raid0_map_submit_bio(struct mddev *mddev, struct bio *bio)
553{
554 struct r0conf *conf = mddev->private;
555 struct strip_zone *zone;
556 struct md_rdev *tmp_dev;
557 sector_t bio_sector = bio->bi_iter.bi_sector;
558 sector_t sector = bio_sector;
559
560 md_account_bio(mddev, &bio);
561
562 zone = find_zone(mddev->private, §or);
563 switch (conf->layout) {
564 case RAID0_ORIG_LAYOUT:
565 tmp_dev = map_sector(mddev, zone, bio_sector, §or);
566 break;
567 case RAID0_ALT_MULTIZONE_LAYOUT:
568 tmp_dev = map_sector(mddev, zone, sector, §or);
569 break;
570 default:
571 WARN(1, "md/raid0:%s: Invalid layout\n", mdname(mddev));
572 bio_io_error(bio);
573 return;
574 }
575
576 if (unlikely(is_rdev_broken(tmp_dev))) {
577 bio_io_error(bio);
578 md_error(mddev, tmp_dev);
579 return;
580 }
581
582 bio_set_dev(bio, tmp_dev->bdev);
583 bio->bi_iter.bi_sector = sector + zone->dev_start +
584 tmp_dev->data_offset;
585 mddev_trace_remap(mddev, bio, bio_sector);
586 mddev_check_write_zeroes(mddev, bio);
587 submit_bio_noacct(bio);
588}
589
590static bool raid0_make_request(struct mddev *mddev, struct bio *bio)
591{
592 sector_t sector;
593 unsigned chunk_sects;
594 unsigned sectors;
595
596 if (unlikely(bio->bi_opf & REQ_PREFLUSH)
597 && md_flush_request(mddev, bio))
598 return true;
599
600 if (unlikely((bio_op(bio) == REQ_OP_DISCARD))) {
601 raid0_handle_discard(mddev, bio);
602 return true;
603 }
604
605 sector = bio->bi_iter.bi_sector;
606 chunk_sects = mddev->chunk_sectors;
607
608 sectors = chunk_sects -
609 (likely(is_power_of_2(chunk_sects))
610 ? (sector & (chunk_sects-1))
611 : sector_div(sector, chunk_sects));
612
613 if (sectors < bio_sectors(bio)) {
614 struct bio *split = bio_split(bio, sectors, GFP_NOIO,
615 &mddev->bio_set);
616
617 if (IS_ERR(split)) {
618 bio->bi_status = errno_to_blk_status(PTR_ERR(split));
619 bio_endio(bio);
620 return true;
621 }
622 bio_chain(split, bio);
623 raid0_map_submit_bio(mddev, bio);
624 bio = split;
625 }
626
627 raid0_map_submit_bio(mddev, bio);
628 return true;
629}
630
631static void raid0_status(struct seq_file *seq, struct mddev *mddev)
632{
633 seq_printf(seq, " %dk chunks", mddev->chunk_sectors / 2);
634 return;
635}
636
637static void raid0_error(struct mddev *mddev, struct md_rdev *rdev)
638{
639 if (!test_and_set_bit(MD_BROKEN, &mddev->flags)) {
640 char *md_name = mdname(mddev);
641
642 pr_crit("md/raid0%s: Disk failure on %pg detected, failing array.\n",
643 md_name, rdev->bdev);
644 }
645}
646
647static void *raid0_takeover_raid45(struct mddev *mddev)
648{
649 struct md_rdev *rdev;
650 struct r0conf *priv_conf;
651
652 if (mddev->degraded != 1) {
653 pr_warn("md/raid0:%s: raid5 must be degraded! Degraded disks: %d\n",
654 mdname(mddev),
655 mddev->degraded);
656 return ERR_PTR(-EINVAL);
657 }
658
659 rdev_for_each(rdev, mddev) {
660 /* check slot number for a disk */
661 if (rdev->raid_disk == mddev->raid_disks-1) {
662 pr_warn("md/raid0:%s: raid5 must have missing parity disk!\n",
663 mdname(mddev));
664 return ERR_PTR(-EINVAL);
665 }
666 rdev->sectors = mddev->dev_sectors;
667 }
668
669 /* Set new parameters */
670 mddev->new_level = 0;
671 mddev->new_layout = 0;
672 mddev->new_chunk_sectors = mddev->chunk_sectors;
673 mddev->raid_disks--;
674 mddev->delta_disks = -1;
675 /* make sure it will be not marked as dirty */
676 mddev->recovery_cp = MaxSector;
677 mddev_clear_unsupported_flags(mddev, UNSUPPORTED_MDDEV_FLAGS);
678
679 create_strip_zones(mddev, &priv_conf);
680
681 return priv_conf;
682}
683
684static void *raid0_takeover_raid10(struct mddev *mddev)
685{
686 struct r0conf *priv_conf;
687
688 /* Check layout:
689 * - far_copies must be 1
690 * - near_copies must be 2
691 * - disks number must be even
692 * - all mirrors must be already degraded
693 */
694 if (mddev->layout != ((1 << 8) + 2)) {
695 pr_warn("md/raid0:%s:: Raid0 cannot takeover layout: 0x%x\n",
696 mdname(mddev),
697 mddev->layout);
698 return ERR_PTR(-EINVAL);
699 }
700 if (mddev->raid_disks & 1) {
701 pr_warn("md/raid0:%s: Raid0 cannot takeover Raid10 with odd disk number.\n",
702 mdname(mddev));
703 return ERR_PTR(-EINVAL);
704 }
705 if (mddev->degraded != (mddev->raid_disks>>1)) {
706 pr_warn("md/raid0:%s: All mirrors must be already degraded!\n",
707 mdname(mddev));
708 return ERR_PTR(-EINVAL);
709 }
710
711 /* Set new parameters */
712 mddev->new_level = 0;
713 mddev->new_layout = 0;
714 mddev->new_chunk_sectors = mddev->chunk_sectors;
715 mddev->delta_disks = - mddev->raid_disks / 2;
716 mddev->raid_disks += mddev->delta_disks;
717 mddev->degraded = 0;
718 /* make sure it will be not marked as dirty */
719 mddev->recovery_cp = MaxSector;
720 mddev_clear_unsupported_flags(mddev, UNSUPPORTED_MDDEV_FLAGS);
721
722 create_strip_zones(mddev, &priv_conf);
723 return priv_conf;
724}
725
726static void *raid0_takeover_raid1(struct mddev *mddev)
727{
728 struct r0conf *priv_conf;
729 int chunksect;
730
731 /* Check layout:
732 * - (N - 1) mirror drives must be already faulty
733 */
734 if ((mddev->raid_disks - 1) != mddev->degraded) {
735 pr_err("md/raid0:%s: (N - 1) mirrors drives must be already faulty!\n",
736 mdname(mddev));
737 return ERR_PTR(-EINVAL);
738 }
739
740 /*
741 * a raid1 doesn't have the notion of chunk size, so
742 * figure out the largest suitable size we can use.
743 */
744 chunksect = 64 * 2; /* 64K by default */
745
746 /* The array must be an exact multiple of chunksize */
747 while (chunksect && (mddev->array_sectors & (chunksect - 1)))
748 chunksect >>= 1;
749
750 if ((chunksect << 9) < PAGE_SIZE)
751 /* array size does not allow a suitable chunk size */
752 return ERR_PTR(-EINVAL);
753
754 /* Set new parameters */
755 mddev->new_level = 0;
756 mddev->new_layout = 0;
757 mddev->new_chunk_sectors = chunksect;
758 mddev->chunk_sectors = chunksect;
759 mddev->delta_disks = 1 - mddev->raid_disks;
760 mddev->raid_disks = 1;
761 /* make sure it will be not marked as dirty */
762 mddev->recovery_cp = MaxSector;
763 mddev_clear_unsupported_flags(mddev, UNSUPPORTED_MDDEV_FLAGS);
764
765 create_strip_zones(mddev, &priv_conf);
766 return priv_conf;
767}
768
769static void *raid0_takeover(struct mddev *mddev)
770{
771 /* raid0 can take over:
772 * raid4 - if all data disks are active.
773 * raid5 - providing it is Raid4 layout and one disk is faulty
774 * raid10 - assuming we have all necessary active disks
775 * raid1 - with (N -1) mirror drives faulty
776 */
777
778 if (mddev->bitmap) {
779 pr_warn("md/raid0: %s: cannot takeover array with bitmap\n",
780 mdname(mddev));
781 return ERR_PTR(-EBUSY);
782 }
783 if (mddev->level == 4)
784 return raid0_takeover_raid45(mddev);
785
786 if (mddev->level == 5) {
787 if (mddev->layout == ALGORITHM_PARITY_N)
788 return raid0_takeover_raid45(mddev);
789
790 pr_warn("md/raid0:%s: Raid can only takeover Raid5 with layout: %d\n",
791 mdname(mddev), ALGORITHM_PARITY_N);
792 }
793
794 if (mddev->level == 10)
795 return raid0_takeover_raid10(mddev);
796
797 if (mddev->level == 1)
798 return raid0_takeover_raid1(mddev);
799
800 pr_warn("Takeover from raid%i to raid0 not supported\n",
801 mddev->level);
802
803 return ERR_PTR(-EINVAL);
804}
805
806static void raid0_quiesce(struct mddev *mddev, int quiesce)
807{
808}
809
810static struct md_personality raid0_personality=
811{
812 .name = "raid0",
813 .level = 0,
814 .owner = THIS_MODULE,
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_personality (&raid0_personality);
828}
829
830static void raid0_exit (void)
831{
832 unregister_md_personality (&raid0_personality);
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");
1/*
2 raid0.c : Multiple Devices driver for Linux
3 Copyright (C) 1994-96 Marc ZYNGIER
4 <zyngier@ufr-info-p7.ibp.fr> or
5 <maz@gloups.fdn.fr>
6 Copyright (C) 1999, 2000 Ingo Molnar, Red Hat
7
8 RAID-0 management functions.
9
10 This program is free software; you can redistribute it and/or modify
11 it under the terms of the GNU General Public License as published by
12 the Free Software Foundation; either version 2, or (at your option)
13 any later version.
14
15 You should have received a copy of the GNU General Public License
16 (for example /usr/src/linux/COPYING); if not, write to the Free
17 Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
18*/
19
20#include <linux/blkdev.h>
21#include <linux/seq_file.h>
22#include <linux/module.h>
23#include <linux/slab.h>
24#include <trace/events/block.h>
25#include "md.h"
26#include "raid0.h"
27#include "raid5.h"
28
29#define UNSUPPORTED_MDDEV_FLAGS \
30 ((1L << MD_HAS_JOURNAL) | \
31 (1L << MD_JOURNAL_CLEAN) | \
32 (1L << MD_FAILFAST_SUPPORTED))
33
34static int raid0_congested(struct mddev *mddev, int bits)
35{
36 struct r0conf *conf = mddev->private;
37 struct md_rdev **devlist = conf->devlist;
38 int raid_disks = conf->strip_zone[0].nb_dev;
39 int i, ret = 0;
40
41 for (i = 0; i < raid_disks && !ret ; i++) {
42 struct request_queue *q = bdev_get_queue(devlist[i]->bdev);
43
44 ret |= bdi_congested(&q->backing_dev_info, bits);
45 }
46 return ret;
47}
48
49/*
50 * inform the user of the raid configuration
51*/
52static void dump_zones(struct mddev *mddev)
53{
54 int j, k;
55 sector_t zone_size = 0;
56 sector_t zone_start = 0;
57 char b[BDEVNAME_SIZE];
58 struct r0conf *conf = mddev->private;
59 int raid_disks = conf->strip_zone[0].nb_dev;
60 pr_debug("md: RAID0 configuration for %s - %d zone%s\n",
61 mdname(mddev),
62 conf->nr_strip_zones, conf->nr_strip_zones==1?"":"s");
63 for (j = 0; j < conf->nr_strip_zones; j++) {
64 char line[200];
65 int len = 0;
66
67 for (k = 0; k < conf->strip_zone[j].nb_dev; k++)
68 len += snprintf(line+len, 200-len, "%s%s", k?"/":"",
69 bdevname(conf->devlist[j*raid_disks
70 + k]->bdev, b));
71 pr_debug("md: zone%d=[%s]\n", j, line);
72
73 zone_size = conf->strip_zone[j].zone_end - zone_start;
74 pr_debug(" zone-offset=%10lluKB, device-offset=%10lluKB, size=%10lluKB\n",
75 (unsigned long long)zone_start>>1,
76 (unsigned long long)conf->strip_zone[j].dev_start>>1,
77 (unsigned long long)zone_size>>1);
78 zone_start = conf->strip_zone[j].zone_end;
79 }
80}
81
82static int create_strip_zones(struct mddev *mddev, struct r0conf **private_conf)
83{
84 int i, c, err;
85 sector_t curr_zone_end, sectors;
86 struct md_rdev *smallest, *rdev1, *rdev2, *rdev, **dev;
87 struct strip_zone *zone;
88 int cnt;
89 char b[BDEVNAME_SIZE];
90 char b2[BDEVNAME_SIZE];
91 struct r0conf *conf = kzalloc(sizeof(*conf), GFP_KERNEL);
92 unsigned short blksize = 512;
93
94 *private_conf = ERR_PTR(-ENOMEM);
95 if (!conf)
96 return -ENOMEM;
97 rdev_for_each(rdev1, mddev) {
98 pr_debug("md/raid0:%s: looking at %s\n",
99 mdname(mddev),
100 bdevname(rdev1->bdev, b));
101 c = 0;
102
103 /* round size to chunk_size */
104 sectors = rdev1->sectors;
105 sector_div(sectors, mddev->chunk_sectors);
106 rdev1->sectors = sectors * mddev->chunk_sectors;
107
108 blksize = max(blksize, queue_logical_block_size(
109 rdev1->bdev->bd_disk->queue));
110
111 rdev_for_each(rdev2, mddev) {
112 pr_debug("md/raid0:%s: comparing %s(%llu)"
113 " with %s(%llu)\n",
114 mdname(mddev),
115 bdevname(rdev1->bdev,b),
116 (unsigned long long)rdev1->sectors,
117 bdevname(rdev2->bdev,b2),
118 (unsigned long long)rdev2->sectors);
119 if (rdev2 == rdev1) {
120 pr_debug("md/raid0:%s: END\n",
121 mdname(mddev));
122 break;
123 }
124 if (rdev2->sectors == rdev1->sectors) {
125 /*
126 * Not unique, don't count it as a new
127 * group
128 */
129 pr_debug("md/raid0:%s: EQUAL\n",
130 mdname(mddev));
131 c = 1;
132 break;
133 }
134 pr_debug("md/raid0:%s: NOT EQUAL\n",
135 mdname(mddev));
136 }
137 if (!c) {
138 pr_debug("md/raid0:%s: ==> UNIQUE\n",
139 mdname(mddev));
140 conf->nr_strip_zones++;
141 pr_debug("md/raid0:%s: %d zones\n",
142 mdname(mddev), conf->nr_strip_zones);
143 }
144 }
145 pr_debug("md/raid0:%s: FINAL %d zones\n",
146 mdname(mddev), conf->nr_strip_zones);
147 /*
148 * now since we have the hard sector sizes, we can make sure
149 * chunk size is a multiple of that sector size
150 */
151 if ((mddev->chunk_sectors << 9) % blksize) {
152 pr_warn("md/raid0:%s: chunk_size of %d not multiple of block size %d\n",
153 mdname(mddev),
154 mddev->chunk_sectors << 9, blksize);
155 err = -EINVAL;
156 goto abort;
157 }
158
159 err = -ENOMEM;
160 conf->strip_zone = kzalloc(sizeof(struct strip_zone)*
161 conf->nr_strip_zones, GFP_KERNEL);
162 if (!conf->strip_zone)
163 goto abort;
164 conf->devlist = kzalloc(sizeof(struct md_rdev*)*
165 conf->nr_strip_zones*mddev->raid_disks,
166 GFP_KERNEL);
167 if (!conf->devlist)
168 goto abort;
169
170 /* The first zone must contain all devices, so here we check that
171 * there is a proper alignment of slots to devices and find them all
172 */
173 zone = &conf->strip_zone[0];
174 cnt = 0;
175 smallest = NULL;
176 dev = conf->devlist;
177 err = -EINVAL;
178 rdev_for_each(rdev1, mddev) {
179 int j = rdev1->raid_disk;
180
181 if (mddev->level == 10) {
182 /* taking over a raid10-n2 array */
183 j /= 2;
184 rdev1->new_raid_disk = j;
185 }
186
187 if (mddev->level == 1) {
188 /* taiking over a raid1 array-
189 * we have only one active disk
190 */
191 j = 0;
192 rdev1->new_raid_disk = j;
193 }
194
195 if (j < 0) {
196 pr_warn("md/raid0:%s: remove inactive devices before converting to RAID0\n",
197 mdname(mddev));
198 goto abort;
199 }
200 if (j >= mddev->raid_disks) {
201 pr_warn("md/raid0:%s: bad disk number %d - aborting!\n",
202 mdname(mddev), j);
203 goto abort;
204 }
205 if (dev[j]) {
206 pr_warn("md/raid0:%s: multiple devices for %d - aborting!\n",
207 mdname(mddev), j);
208 goto abort;
209 }
210 dev[j] = rdev1;
211
212 if (!smallest || (rdev1->sectors < smallest->sectors))
213 smallest = rdev1;
214 cnt++;
215 }
216 if (cnt != mddev->raid_disks) {
217 pr_warn("md/raid0:%s: too few disks (%d of %d) - aborting!\n",
218 mdname(mddev), cnt, mddev->raid_disks);
219 goto abort;
220 }
221 zone->nb_dev = cnt;
222 zone->zone_end = smallest->sectors * cnt;
223
224 curr_zone_end = zone->zone_end;
225
226 /* now do the other zones */
227 for (i = 1; i < conf->nr_strip_zones; i++)
228 {
229 int j;
230
231 zone = conf->strip_zone + i;
232 dev = conf->devlist + i * mddev->raid_disks;
233
234 pr_debug("md/raid0:%s: zone %d\n", mdname(mddev), i);
235 zone->dev_start = smallest->sectors;
236 smallest = NULL;
237 c = 0;
238
239 for (j=0; j<cnt; j++) {
240 rdev = conf->devlist[j];
241 if (rdev->sectors <= zone->dev_start) {
242 pr_debug("md/raid0:%s: checking %s ... nope\n",
243 mdname(mddev),
244 bdevname(rdev->bdev, b));
245 continue;
246 }
247 pr_debug("md/raid0:%s: checking %s ..."
248 " contained as device %d\n",
249 mdname(mddev),
250 bdevname(rdev->bdev, b), c);
251 dev[c] = rdev;
252 c++;
253 if (!smallest || rdev->sectors < smallest->sectors) {
254 smallest = rdev;
255 pr_debug("md/raid0:%s: (%llu) is smallest!.\n",
256 mdname(mddev),
257 (unsigned long long)rdev->sectors);
258 }
259 }
260
261 zone->nb_dev = c;
262 sectors = (smallest->sectors - zone->dev_start) * c;
263 pr_debug("md/raid0:%s: zone->nb_dev: %d, sectors: %llu\n",
264 mdname(mddev),
265 zone->nb_dev, (unsigned long long)sectors);
266
267 curr_zone_end += sectors;
268 zone->zone_end = curr_zone_end;
269
270 pr_debug("md/raid0:%s: current zone start: %llu\n",
271 mdname(mddev),
272 (unsigned long long)smallest->sectors);
273 }
274
275 pr_debug("md/raid0:%s: done.\n", mdname(mddev));
276 *private_conf = conf;
277
278 return 0;
279abort:
280 kfree(conf->strip_zone);
281 kfree(conf->devlist);
282 kfree(conf);
283 *private_conf = ERR_PTR(err);
284 return err;
285}
286
287/* Find the zone which holds a particular offset
288 * Update *sectorp to be an offset in that zone
289 */
290static struct strip_zone *find_zone(struct r0conf *conf,
291 sector_t *sectorp)
292{
293 int i;
294 struct strip_zone *z = conf->strip_zone;
295 sector_t sector = *sectorp;
296
297 for (i = 0; i < conf->nr_strip_zones; i++)
298 if (sector < z[i].zone_end) {
299 if (i)
300 *sectorp = sector - z[i-1].zone_end;
301 return z + i;
302 }
303 BUG();
304}
305
306/*
307 * remaps the bio to the target device. we separate two flows.
308 * power 2 flow and a general flow for the sake of performance
309*/
310static struct md_rdev *map_sector(struct mddev *mddev, struct strip_zone *zone,
311 sector_t sector, sector_t *sector_offset)
312{
313 unsigned int sect_in_chunk;
314 sector_t chunk;
315 struct r0conf *conf = mddev->private;
316 int raid_disks = conf->strip_zone[0].nb_dev;
317 unsigned int chunk_sects = mddev->chunk_sectors;
318
319 if (is_power_of_2(chunk_sects)) {
320 int chunksect_bits = ffz(~chunk_sects);
321 /* find the sector offset inside the chunk */
322 sect_in_chunk = sector & (chunk_sects - 1);
323 sector >>= chunksect_bits;
324 /* chunk in zone */
325 chunk = *sector_offset;
326 /* quotient is the chunk in real device*/
327 sector_div(chunk, zone->nb_dev << chunksect_bits);
328 } else{
329 sect_in_chunk = sector_div(sector, chunk_sects);
330 chunk = *sector_offset;
331 sector_div(chunk, chunk_sects * zone->nb_dev);
332 }
333 /*
334 * position the bio over the real device
335 * real sector = chunk in device + starting of zone
336 * + the position in the chunk
337 */
338 *sector_offset = (chunk * chunk_sects) + sect_in_chunk;
339 return conf->devlist[(zone - conf->strip_zone)*raid_disks
340 + sector_div(sector, zone->nb_dev)];
341}
342
343static sector_t raid0_size(struct mddev *mddev, sector_t sectors, int raid_disks)
344{
345 sector_t array_sectors = 0;
346 struct md_rdev *rdev;
347
348 WARN_ONCE(sectors || raid_disks,
349 "%s does not support generic reshape\n", __func__);
350
351 rdev_for_each(rdev, mddev)
352 array_sectors += (rdev->sectors &
353 ~(sector_t)(mddev->chunk_sectors-1));
354
355 return array_sectors;
356}
357
358static void raid0_free(struct mddev *mddev, void *priv);
359
360static int raid0_run(struct mddev *mddev)
361{
362 struct r0conf *conf;
363 int ret;
364
365 if (mddev->chunk_sectors == 0) {
366 pr_warn("md/raid0:%s: chunk size must be set.\n", mdname(mddev));
367 return -EINVAL;
368 }
369 if (md_check_no_bitmap(mddev))
370 return -EINVAL;
371
372 /* if private is not null, we are here after takeover */
373 if (mddev->private == NULL) {
374 ret = create_strip_zones(mddev, &conf);
375 if (ret < 0)
376 return ret;
377 mddev->private = conf;
378 }
379 conf = mddev->private;
380 if (mddev->queue) {
381 struct md_rdev *rdev;
382 bool discard_supported = false;
383
384 blk_queue_max_hw_sectors(mddev->queue, mddev->chunk_sectors);
385 blk_queue_max_write_same_sectors(mddev->queue, mddev->chunk_sectors);
386 blk_queue_max_discard_sectors(mddev->queue, mddev->chunk_sectors);
387
388 blk_queue_io_min(mddev->queue, mddev->chunk_sectors << 9);
389 blk_queue_io_opt(mddev->queue,
390 (mddev->chunk_sectors << 9) * mddev->raid_disks);
391
392 rdev_for_each(rdev, mddev) {
393 disk_stack_limits(mddev->gendisk, rdev->bdev,
394 rdev->data_offset << 9);
395 if (blk_queue_discard(bdev_get_queue(rdev->bdev)))
396 discard_supported = true;
397 }
398 if (!discard_supported)
399 queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, mddev->queue);
400 else
401 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, mddev->queue);
402 }
403
404 /* calculate array device size */
405 md_set_array_sectors(mddev, raid0_size(mddev, 0, 0));
406
407 pr_debug("md/raid0:%s: md_size is %llu sectors.\n",
408 mdname(mddev),
409 (unsigned long long)mddev->array_sectors);
410
411 if (mddev->queue) {
412 /* calculate the max read-ahead size.
413 * For read-ahead of large files to be effective, we need to
414 * readahead at least twice a whole stripe. i.e. number of devices
415 * multiplied by chunk size times 2.
416 * If an individual device has an ra_pages greater than the
417 * chunk size, then we will not drive that device as hard as it
418 * wants. We consider this a configuration error: a larger
419 * chunksize should be used in that case.
420 */
421 int stripe = mddev->raid_disks *
422 (mddev->chunk_sectors << 9) / PAGE_SIZE;
423 if (mddev->queue->backing_dev_info.ra_pages < 2* stripe)
424 mddev->queue->backing_dev_info.ra_pages = 2* stripe;
425 }
426
427 dump_zones(mddev);
428
429 ret = md_integrity_register(mddev);
430
431 return ret;
432}
433
434static void raid0_free(struct mddev *mddev, void *priv)
435{
436 struct r0conf *conf = priv;
437
438 kfree(conf->strip_zone);
439 kfree(conf->devlist);
440 kfree(conf);
441}
442
443/*
444 * Is io distribute over 1 or more chunks ?
445*/
446static inline int is_io_in_chunk_boundary(struct mddev *mddev,
447 unsigned int chunk_sects, struct bio *bio)
448{
449 if (likely(is_power_of_2(chunk_sects))) {
450 return chunk_sects >=
451 ((bio->bi_iter.bi_sector & (chunk_sects-1))
452 + bio_sectors(bio));
453 } else{
454 sector_t sector = bio->bi_iter.bi_sector;
455 return chunk_sects >= (sector_div(sector, chunk_sects)
456 + bio_sectors(bio));
457 }
458}
459
460static void raid0_make_request(struct mddev *mddev, struct bio *bio)
461{
462 struct strip_zone *zone;
463 struct md_rdev *tmp_dev;
464 struct bio *split;
465
466 if (unlikely(bio->bi_opf & REQ_PREFLUSH)) {
467 md_flush_request(mddev, bio);
468 return;
469 }
470
471 do {
472 sector_t bio_sector = bio->bi_iter.bi_sector;
473 sector_t sector = bio_sector;
474 unsigned chunk_sects = mddev->chunk_sectors;
475
476 unsigned sectors = chunk_sects -
477 (likely(is_power_of_2(chunk_sects))
478 ? (sector & (chunk_sects-1))
479 : sector_div(sector, chunk_sects));
480
481 /* Restore due to sector_div */
482 sector = bio_sector;
483
484 if (sectors < bio_sectors(bio)) {
485 split = bio_split(bio, sectors, GFP_NOIO, fs_bio_set);
486 bio_chain(split, bio);
487 } else {
488 split = bio;
489 }
490
491 zone = find_zone(mddev->private, §or);
492 tmp_dev = map_sector(mddev, zone, sector, §or);
493 split->bi_bdev = tmp_dev->bdev;
494 split->bi_iter.bi_sector = sector + zone->dev_start +
495 tmp_dev->data_offset;
496
497 if (unlikely((bio_op(split) == REQ_OP_DISCARD) &&
498 !blk_queue_discard(bdev_get_queue(split->bi_bdev)))) {
499 /* Just ignore it */
500 bio_endio(split);
501 } else {
502 if (mddev->gendisk)
503 trace_block_bio_remap(bdev_get_queue(split->bi_bdev),
504 split, disk_devt(mddev->gendisk),
505 bio_sector);
506 generic_make_request(split);
507 }
508 } while (split != bio);
509}
510
511static void raid0_status(struct seq_file *seq, struct mddev *mddev)
512{
513 seq_printf(seq, " %dk chunks", mddev->chunk_sectors / 2);
514 return;
515}
516
517static void *raid0_takeover_raid45(struct mddev *mddev)
518{
519 struct md_rdev *rdev;
520 struct r0conf *priv_conf;
521
522 if (mddev->degraded != 1) {
523 pr_warn("md/raid0:%s: raid5 must be degraded! Degraded disks: %d\n",
524 mdname(mddev),
525 mddev->degraded);
526 return ERR_PTR(-EINVAL);
527 }
528
529 rdev_for_each(rdev, mddev) {
530 /* check slot number for a disk */
531 if (rdev->raid_disk == mddev->raid_disks-1) {
532 pr_warn("md/raid0:%s: raid5 must have missing parity disk!\n",
533 mdname(mddev));
534 return ERR_PTR(-EINVAL);
535 }
536 rdev->sectors = mddev->dev_sectors;
537 }
538
539 /* Set new parameters */
540 mddev->new_level = 0;
541 mddev->new_layout = 0;
542 mddev->new_chunk_sectors = mddev->chunk_sectors;
543 mddev->raid_disks--;
544 mddev->delta_disks = -1;
545 /* make sure it will be not marked as dirty */
546 mddev->recovery_cp = MaxSector;
547 mddev_clear_unsupported_flags(mddev, UNSUPPORTED_MDDEV_FLAGS);
548
549 create_strip_zones(mddev, &priv_conf);
550
551 return priv_conf;
552}
553
554static void *raid0_takeover_raid10(struct mddev *mddev)
555{
556 struct r0conf *priv_conf;
557
558 /* Check layout:
559 * - far_copies must be 1
560 * - near_copies must be 2
561 * - disks number must be even
562 * - all mirrors must be already degraded
563 */
564 if (mddev->layout != ((1 << 8) + 2)) {
565 pr_warn("md/raid0:%s:: Raid0 cannot takeover layout: 0x%x\n",
566 mdname(mddev),
567 mddev->layout);
568 return ERR_PTR(-EINVAL);
569 }
570 if (mddev->raid_disks & 1) {
571 pr_warn("md/raid0:%s: Raid0 cannot takeover Raid10 with odd disk number.\n",
572 mdname(mddev));
573 return ERR_PTR(-EINVAL);
574 }
575 if (mddev->degraded != (mddev->raid_disks>>1)) {
576 pr_warn("md/raid0:%s: All mirrors must be already degraded!\n",
577 mdname(mddev));
578 return ERR_PTR(-EINVAL);
579 }
580
581 /* Set new parameters */
582 mddev->new_level = 0;
583 mddev->new_layout = 0;
584 mddev->new_chunk_sectors = mddev->chunk_sectors;
585 mddev->delta_disks = - mddev->raid_disks / 2;
586 mddev->raid_disks += mddev->delta_disks;
587 mddev->degraded = 0;
588 /* make sure it will be not marked as dirty */
589 mddev->recovery_cp = MaxSector;
590 mddev_clear_unsupported_flags(mddev, UNSUPPORTED_MDDEV_FLAGS);
591
592 create_strip_zones(mddev, &priv_conf);
593 return priv_conf;
594}
595
596static void *raid0_takeover_raid1(struct mddev *mddev)
597{
598 struct r0conf *priv_conf;
599 int chunksect;
600
601 /* Check layout:
602 * - (N - 1) mirror drives must be already faulty
603 */
604 if ((mddev->raid_disks - 1) != mddev->degraded) {
605 pr_err("md/raid0:%s: (N - 1) mirrors drives must be already faulty!\n",
606 mdname(mddev));
607 return ERR_PTR(-EINVAL);
608 }
609
610 /*
611 * a raid1 doesn't have the notion of chunk size, so
612 * figure out the largest suitable size we can use.
613 */
614 chunksect = 64 * 2; /* 64K by default */
615
616 /* The array must be an exact multiple of chunksize */
617 while (chunksect && (mddev->array_sectors & (chunksect - 1)))
618 chunksect >>= 1;
619
620 if ((chunksect << 9) < PAGE_SIZE)
621 /* array size does not allow a suitable chunk size */
622 return ERR_PTR(-EINVAL);
623
624 /* Set new parameters */
625 mddev->new_level = 0;
626 mddev->new_layout = 0;
627 mddev->new_chunk_sectors = chunksect;
628 mddev->chunk_sectors = chunksect;
629 mddev->delta_disks = 1 - mddev->raid_disks;
630 mddev->raid_disks = 1;
631 /* make sure it will be not marked as dirty */
632 mddev->recovery_cp = MaxSector;
633 mddev_clear_unsupported_flags(mddev, UNSUPPORTED_MDDEV_FLAGS);
634
635 create_strip_zones(mddev, &priv_conf);
636 return priv_conf;
637}
638
639static void *raid0_takeover(struct mddev *mddev)
640{
641 /* raid0 can take over:
642 * raid4 - if all data disks are active.
643 * raid5 - providing it is Raid4 layout and one disk is faulty
644 * raid10 - assuming we have all necessary active disks
645 * raid1 - with (N -1) mirror drives faulty
646 */
647
648 if (mddev->bitmap) {
649 pr_warn("md/raid0: %s: cannot takeover array with bitmap\n",
650 mdname(mddev));
651 return ERR_PTR(-EBUSY);
652 }
653 if (mddev->level == 4)
654 return raid0_takeover_raid45(mddev);
655
656 if (mddev->level == 5) {
657 if (mddev->layout == ALGORITHM_PARITY_N)
658 return raid0_takeover_raid45(mddev);
659
660 pr_warn("md/raid0:%s: Raid can only takeover Raid5 with layout: %d\n",
661 mdname(mddev), ALGORITHM_PARITY_N);
662 }
663
664 if (mddev->level == 10)
665 return raid0_takeover_raid10(mddev);
666
667 if (mddev->level == 1)
668 return raid0_takeover_raid1(mddev);
669
670 pr_warn("Takeover from raid%i to raid0 not supported\n",
671 mddev->level);
672
673 return ERR_PTR(-EINVAL);
674}
675
676static void raid0_quiesce(struct mddev *mddev, int state)
677{
678}
679
680static struct md_personality raid0_personality=
681{
682 .name = "raid0",
683 .level = 0,
684 .owner = THIS_MODULE,
685 .make_request = raid0_make_request,
686 .run = raid0_run,
687 .free = raid0_free,
688 .status = raid0_status,
689 .size = raid0_size,
690 .takeover = raid0_takeover,
691 .quiesce = raid0_quiesce,
692 .congested = raid0_congested,
693};
694
695static int __init raid0_init (void)
696{
697 return register_md_personality (&raid0_personality);
698}
699
700static void raid0_exit (void)
701{
702 unregister_md_personality (&raid0_personality);
703}
704
705module_init(raid0_init);
706module_exit(raid0_exit);
707MODULE_LICENSE("GPL");
708MODULE_DESCRIPTION("RAID0 (striping) personality for MD");
709MODULE_ALIAS("md-personality-2"); /* RAID0 */
710MODULE_ALIAS("md-raid0");
711MODULE_ALIAS("md-level-0");