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 free_conf(struct mddev *mddev, struct r0conf *conf)
369{
370 kfree(conf->strip_zone);
371 kfree(conf->devlist);
372 kfree(conf);
373}
374
375static void raid0_free(struct mddev *mddev, void *priv)
376{
377 struct r0conf *conf = priv;
378
379 free_conf(mddev, conf);
380}
381
382static int raid0_run(struct mddev *mddev)
383{
384 struct r0conf *conf;
385 int ret;
386
387 if (mddev->chunk_sectors == 0) {
388 pr_warn("md/raid0:%s: chunk size must be set.\n", mdname(mddev));
389 return -EINVAL;
390 }
391 if (md_check_no_bitmap(mddev))
392 return -EINVAL;
393
394 /* if private is not null, we are here after takeover */
395 if (mddev->private == NULL) {
396 ret = create_strip_zones(mddev, &conf);
397 if (ret < 0)
398 return ret;
399 mddev->private = conf;
400 }
401 conf = mddev->private;
402 if (mddev->queue) {
403 struct md_rdev *rdev;
404
405 blk_queue_max_hw_sectors(mddev->queue, mddev->chunk_sectors);
406 blk_queue_max_write_zeroes_sectors(mddev->queue, mddev->chunk_sectors);
407
408 blk_queue_io_min(mddev->queue, mddev->chunk_sectors << 9);
409 blk_queue_io_opt(mddev->queue,
410 (mddev->chunk_sectors << 9) * mddev->raid_disks);
411
412 rdev_for_each(rdev, mddev) {
413 disk_stack_limits(mddev->gendisk, rdev->bdev,
414 rdev->data_offset << 9);
415 }
416 }
417
418 /* calculate array device size */
419 md_set_array_sectors(mddev, raid0_size(mddev, 0, 0));
420
421 pr_debug("md/raid0:%s: md_size is %llu sectors.\n",
422 mdname(mddev),
423 (unsigned long long)mddev->array_sectors);
424
425 dump_zones(mddev);
426
427 ret = md_integrity_register(mddev);
428 if (ret)
429 free_conf(mddev, conf);
430
431 return ret;
432}
433
434/*
435 * Convert disk_index to the disk order in which it is read/written.
436 * For example, if we have 4 disks, they are numbered 0,1,2,3. If we
437 * write the disks starting at disk 3, then the read/write order would
438 * be disk 3, then 0, then 1, and then disk 2 and we want map_disk_shift()
439 * to map the disks as follows 0,1,2,3 => 1,2,3,0. So disk 0 would map
440 * to 1, 1 to 2, 2 to 3, and 3 to 0. That way we can compare disks in
441 * that 'output' space to understand the read/write disk ordering.
442 */
443static int map_disk_shift(int disk_index, int num_disks, int disk_shift)
444{
445 return ((disk_index + num_disks - disk_shift) % num_disks);
446}
447
448static void raid0_handle_discard(struct mddev *mddev, struct bio *bio)
449{
450 struct r0conf *conf = mddev->private;
451 struct strip_zone *zone;
452 sector_t start = bio->bi_iter.bi_sector;
453 sector_t end;
454 unsigned int stripe_size;
455 sector_t first_stripe_index, last_stripe_index;
456 sector_t start_disk_offset;
457 unsigned int start_disk_index;
458 sector_t end_disk_offset;
459 unsigned int end_disk_index;
460 unsigned int disk;
461 sector_t orig_start, orig_end;
462
463 orig_start = start;
464 zone = find_zone(conf, &start);
465
466 if (bio_end_sector(bio) > zone->zone_end) {
467 struct bio *split = bio_split(bio,
468 zone->zone_end - bio->bi_iter.bi_sector, GFP_NOIO,
469 &mddev->bio_set);
470 bio_chain(split, bio);
471 submit_bio_noacct(bio);
472 bio = split;
473 end = zone->zone_end;
474 } else
475 end = bio_end_sector(bio);
476
477 orig_end = end;
478 if (zone != conf->strip_zone)
479 end = end - zone[-1].zone_end;
480
481 /* Now start and end is the offset in zone */
482 stripe_size = zone->nb_dev * mddev->chunk_sectors;
483
484 first_stripe_index = start;
485 sector_div(first_stripe_index, stripe_size);
486 last_stripe_index = end;
487 sector_div(last_stripe_index, stripe_size);
488
489 /* In the first zone the original and alternate layouts are the same */
490 if ((conf->layout == RAID0_ORIG_LAYOUT) && (zone != conf->strip_zone)) {
491 sector_div(orig_start, mddev->chunk_sectors);
492 start_disk_index = sector_div(orig_start, zone->nb_dev);
493 start_disk_index = map_disk_shift(start_disk_index,
494 zone->nb_dev,
495 zone->disk_shift);
496 sector_div(orig_end, mddev->chunk_sectors);
497 end_disk_index = sector_div(orig_end, zone->nb_dev);
498 end_disk_index = map_disk_shift(end_disk_index,
499 zone->nb_dev, zone->disk_shift);
500 } else {
501 start_disk_index = (int)(start - first_stripe_index * stripe_size) /
502 mddev->chunk_sectors;
503 end_disk_index = (int)(end - last_stripe_index * stripe_size) /
504 mddev->chunk_sectors;
505 }
506 start_disk_offset = ((int)(start - first_stripe_index * stripe_size) %
507 mddev->chunk_sectors) +
508 first_stripe_index * mddev->chunk_sectors;
509 end_disk_offset = ((int)(end - last_stripe_index * stripe_size) %
510 mddev->chunk_sectors) +
511 last_stripe_index * mddev->chunk_sectors;
512
513 for (disk = 0; disk < zone->nb_dev; disk++) {
514 sector_t dev_start, dev_end;
515 struct md_rdev *rdev;
516 int compare_disk;
517
518 compare_disk = map_disk_shift(disk, zone->nb_dev,
519 zone->disk_shift);
520
521 if (compare_disk < start_disk_index)
522 dev_start = (first_stripe_index + 1) *
523 mddev->chunk_sectors;
524 else if (compare_disk > start_disk_index)
525 dev_start = first_stripe_index * mddev->chunk_sectors;
526 else
527 dev_start = start_disk_offset;
528
529 if (compare_disk < end_disk_index)
530 dev_end = (last_stripe_index + 1) * mddev->chunk_sectors;
531 else if (compare_disk > end_disk_index)
532 dev_end = last_stripe_index * mddev->chunk_sectors;
533 else
534 dev_end = end_disk_offset;
535
536 if (dev_end <= dev_start)
537 continue;
538
539 rdev = conf->devlist[(zone - conf->strip_zone) *
540 conf->strip_zone[0].nb_dev + disk];
541 md_submit_discard_bio(mddev, rdev, bio,
542 dev_start + zone->dev_start + rdev->data_offset,
543 dev_end - dev_start);
544 }
545 bio_endio(bio);
546}
547
548static void raid0_map_submit_bio(struct mddev *mddev, struct bio *bio)
549{
550 struct r0conf *conf = mddev->private;
551 struct strip_zone *zone;
552 struct md_rdev *tmp_dev;
553 sector_t bio_sector = bio->bi_iter.bi_sector;
554 sector_t sector = bio_sector;
555
556 md_account_bio(mddev, &bio);
557
558 zone = find_zone(mddev->private, §or);
559 switch (conf->layout) {
560 case RAID0_ORIG_LAYOUT:
561 tmp_dev = map_sector(mddev, zone, bio_sector, §or);
562 break;
563 case RAID0_ALT_MULTIZONE_LAYOUT:
564 tmp_dev = map_sector(mddev, zone, sector, §or);
565 break;
566 default:
567 WARN(1, "md/raid0:%s: Invalid layout\n", mdname(mddev));
568 bio_io_error(bio);
569 return;
570 }
571
572 if (unlikely(is_rdev_broken(tmp_dev))) {
573 bio_io_error(bio);
574 md_error(mddev, tmp_dev);
575 return;
576 }
577
578 bio_set_dev(bio, tmp_dev->bdev);
579 bio->bi_iter.bi_sector = sector + zone->dev_start +
580 tmp_dev->data_offset;
581
582 if (mddev->gendisk)
583 trace_block_bio_remap(bio, disk_devt(mddev->gendisk),
584 bio_sector);
585 mddev_check_write_zeroes(mddev, bio);
586 submit_bio_noacct(bio);
587}
588
589static bool raid0_make_request(struct mddev *mddev, struct bio *bio)
590{
591 sector_t sector;
592 unsigned chunk_sects;
593 unsigned sectors;
594
595 if (unlikely(bio->bi_opf & REQ_PREFLUSH)
596 && md_flush_request(mddev, bio))
597 return true;
598
599 if (unlikely((bio_op(bio) == REQ_OP_DISCARD))) {
600 raid0_handle_discard(mddev, bio);
601 return true;
602 }
603
604 sector = bio->bi_iter.bi_sector;
605 chunk_sects = mddev->chunk_sectors;
606
607 sectors = chunk_sects -
608 (likely(is_power_of_2(chunk_sects))
609 ? (sector & (chunk_sects-1))
610 : sector_div(sector, chunk_sects));
611
612 if (sectors < bio_sectors(bio)) {
613 struct bio *split = bio_split(bio, sectors, GFP_NOIO,
614 &mddev->bio_set);
615 bio_chain(split, bio);
616 raid0_map_submit_bio(mddev, bio);
617 bio = split;
618 }
619
620 raid0_map_submit_bio(mddev, bio);
621 return true;
622}
623
624static void raid0_status(struct seq_file *seq, struct mddev *mddev)
625{
626 seq_printf(seq, " %dk chunks", mddev->chunk_sectors / 2);
627 return;
628}
629
630static void raid0_error(struct mddev *mddev, struct md_rdev *rdev)
631{
632 if (!test_and_set_bit(MD_BROKEN, &mddev->flags)) {
633 char *md_name = mdname(mddev);
634
635 pr_crit("md/raid0%s: Disk failure on %pg detected, failing array.\n",
636 md_name, rdev->bdev);
637 }
638}
639
640static void *raid0_takeover_raid45(struct mddev *mddev)
641{
642 struct md_rdev *rdev;
643 struct r0conf *priv_conf;
644
645 if (mddev->degraded != 1) {
646 pr_warn("md/raid0:%s: raid5 must be degraded! Degraded disks: %d\n",
647 mdname(mddev),
648 mddev->degraded);
649 return ERR_PTR(-EINVAL);
650 }
651
652 rdev_for_each(rdev, mddev) {
653 /* check slot number for a disk */
654 if (rdev->raid_disk == mddev->raid_disks-1) {
655 pr_warn("md/raid0:%s: raid5 must have missing parity disk!\n",
656 mdname(mddev));
657 return ERR_PTR(-EINVAL);
658 }
659 rdev->sectors = mddev->dev_sectors;
660 }
661
662 /* Set new parameters */
663 mddev->new_level = 0;
664 mddev->new_layout = 0;
665 mddev->new_chunk_sectors = mddev->chunk_sectors;
666 mddev->raid_disks--;
667 mddev->delta_disks = -1;
668 /* make sure it will be not marked as dirty */
669 mddev->recovery_cp = MaxSector;
670 mddev_clear_unsupported_flags(mddev, UNSUPPORTED_MDDEV_FLAGS);
671
672 create_strip_zones(mddev, &priv_conf);
673
674 return priv_conf;
675}
676
677static void *raid0_takeover_raid10(struct mddev *mddev)
678{
679 struct r0conf *priv_conf;
680
681 /* Check layout:
682 * - far_copies must be 1
683 * - near_copies must be 2
684 * - disks number must be even
685 * - all mirrors must be already degraded
686 */
687 if (mddev->layout != ((1 << 8) + 2)) {
688 pr_warn("md/raid0:%s:: Raid0 cannot takeover layout: 0x%x\n",
689 mdname(mddev),
690 mddev->layout);
691 return ERR_PTR(-EINVAL);
692 }
693 if (mddev->raid_disks & 1) {
694 pr_warn("md/raid0:%s: Raid0 cannot takeover Raid10 with odd disk number.\n",
695 mdname(mddev));
696 return ERR_PTR(-EINVAL);
697 }
698 if (mddev->degraded != (mddev->raid_disks>>1)) {
699 pr_warn("md/raid0:%s: All mirrors must be already degraded!\n",
700 mdname(mddev));
701 return ERR_PTR(-EINVAL);
702 }
703
704 /* Set new parameters */
705 mddev->new_level = 0;
706 mddev->new_layout = 0;
707 mddev->new_chunk_sectors = mddev->chunk_sectors;
708 mddev->delta_disks = - mddev->raid_disks / 2;
709 mddev->raid_disks += mddev->delta_disks;
710 mddev->degraded = 0;
711 /* make sure it will be not marked as dirty */
712 mddev->recovery_cp = MaxSector;
713 mddev_clear_unsupported_flags(mddev, UNSUPPORTED_MDDEV_FLAGS);
714
715 create_strip_zones(mddev, &priv_conf);
716 return priv_conf;
717}
718
719static void *raid0_takeover_raid1(struct mddev *mddev)
720{
721 struct r0conf *priv_conf;
722 int chunksect;
723
724 /* Check layout:
725 * - (N - 1) mirror drives must be already faulty
726 */
727 if ((mddev->raid_disks - 1) != mddev->degraded) {
728 pr_err("md/raid0:%s: (N - 1) mirrors drives must be already faulty!\n",
729 mdname(mddev));
730 return ERR_PTR(-EINVAL);
731 }
732
733 /*
734 * a raid1 doesn't have the notion of chunk size, so
735 * figure out the largest suitable size we can use.
736 */
737 chunksect = 64 * 2; /* 64K by default */
738
739 /* The array must be an exact multiple of chunksize */
740 while (chunksect && (mddev->array_sectors & (chunksect - 1)))
741 chunksect >>= 1;
742
743 if ((chunksect << 9) < PAGE_SIZE)
744 /* array size does not allow a suitable chunk size */
745 return ERR_PTR(-EINVAL);
746
747 /* Set new parameters */
748 mddev->new_level = 0;
749 mddev->new_layout = 0;
750 mddev->new_chunk_sectors = chunksect;
751 mddev->chunk_sectors = chunksect;
752 mddev->delta_disks = 1 - mddev->raid_disks;
753 mddev->raid_disks = 1;
754 /* make sure it will be not marked as dirty */
755 mddev->recovery_cp = MaxSector;
756 mddev_clear_unsupported_flags(mddev, UNSUPPORTED_MDDEV_FLAGS);
757
758 create_strip_zones(mddev, &priv_conf);
759 return priv_conf;
760}
761
762static void *raid0_takeover(struct mddev *mddev)
763{
764 /* raid0 can take over:
765 * raid4 - if all data disks are active.
766 * raid5 - providing it is Raid4 layout and one disk is faulty
767 * raid10 - assuming we have all necessary active disks
768 * raid1 - with (N -1) mirror drives faulty
769 */
770
771 if (mddev->bitmap) {
772 pr_warn("md/raid0: %s: cannot takeover array with bitmap\n",
773 mdname(mddev));
774 return ERR_PTR(-EBUSY);
775 }
776 if (mddev->level == 4)
777 return raid0_takeover_raid45(mddev);
778
779 if (mddev->level == 5) {
780 if (mddev->layout == ALGORITHM_PARITY_N)
781 return raid0_takeover_raid45(mddev);
782
783 pr_warn("md/raid0:%s: Raid can only takeover Raid5 with layout: %d\n",
784 mdname(mddev), ALGORITHM_PARITY_N);
785 }
786
787 if (mddev->level == 10)
788 return raid0_takeover_raid10(mddev);
789
790 if (mddev->level == 1)
791 return raid0_takeover_raid1(mddev);
792
793 pr_warn("Takeover from raid%i to raid0 not supported\n",
794 mddev->level);
795
796 return ERR_PTR(-EINVAL);
797}
798
799static void raid0_quiesce(struct mddev *mddev, int quiesce)
800{
801}
802
803static struct md_personality raid0_personality=
804{
805 .name = "raid0",
806 .level = 0,
807 .owner = THIS_MODULE,
808 .make_request = raid0_make_request,
809 .run = raid0_run,
810 .free = raid0_free,
811 .status = raid0_status,
812 .size = raid0_size,
813 .takeover = raid0_takeover,
814 .quiesce = raid0_quiesce,
815 .error_handler = raid0_error,
816};
817
818static int __init raid0_init (void)
819{
820 return register_md_personality (&raid0_personality);
821}
822
823static void raid0_exit (void)
824{
825 unregister_md_personality (&raid0_personality);
826}
827
828module_init(raid0_init);
829module_exit(raid0_exit);
830MODULE_LICENSE("GPL");
831MODULE_DESCRIPTION("RAID0 (striping) personality for MD");
832MODULE_ALIAS("md-personality-2"); /* RAID0 */
833MODULE_ALIAS("md-raid0");
834MODULE_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
9 RAID-0 management functions.
10
11 This program is free software; you can redistribute it and/or modify
12 it under the terms of the GNU General Public License as published by
13 the Free Software Foundation; either version 2, or (at your option)
14 any later version.
15
16 You should have received a copy of the GNU General Public License
17 (for example /usr/src/linux/COPYING); if not, write to the Free
18 Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
19*/
20
21#include <linux/blkdev.h>
22#include <linux/seq_file.h>
23#include <linux/module.h>
24#include <linux/slab.h>
25#include "md.h"
26#include "raid0.h"
27#include "raid5.h"
28
29static int raid0_congested(void *data, int bits)
30{
31 struct mddev *mddev = data;
32 struct r0conf *conf = mddev->private;
33 struct md_rdev **devlist = conf->devlist;
34 int raid_disks = conf->strip_zone[0].nb_dev;
35 int i, ret = 0;
36
37 if (mddev_congested(mddev, bits))
38 return 1;
39
40 for (i = 0; i < raid_disks && !ret ; i++) {
41 struct request_queue *q = bdev_get_queue(devlist[i]->bdev);
42
43 ret |= bdi_congested(&q->backing_dev_info, bits);
44 }
45 return ret;
46}
47
48/*
49 * inform the user of the raid configuration
50*/
51static void dump_zones(struct mddev *mddev)
52{
53 int j, k;
54 sector_t zone_size = 0;
55 sector_t zone_start = 0;
56 char b[BDEVNAME_SIZE];
57 struct r0conf *conf = mddev->private;
58 int raid_disks = conf->strip_zone[0].nb_dev;
59 printk(KERN_INFO "md: RAID0 configuration for %s - %d zone%s\n",
60 mdname(mddev),
61 conf->nr_strip_zones, conf->nr_strip_zones==1?"":"s");
62 for (j = 0; j < conf->nr_strip_zones; j++) {
63 printk(KERN_INFO "md: zone%d=[", j);
64 for (k = 0; k < conf->strip_zone[j].nb_dev; k++)
65 printk(KERN_CONT "%s%s", k?"/":"",
66 bdevname(conf->devlist[j*raid_disks
67 + k]->bdev, b));
68 printk(KERN_CONT "]\n");
69
70 zone_size = conf->strip_zone[j].zone_end - zone_start;
71 printk(KERN_INFO " zone-offset=%10lluKB, "
72 "device-offset=%10lluKB, size=%10lluKB\n",
73 (unsigned long long)zone_start>>1,
74 (unsigned long long)conf->strip_zone[j].dev_start>>1,
75 (unsigned long long)zone_size>>1);
76 zone_start = conf->strip_zone[j].zone_end;
77 }
78 printk(KERN_INFO "\n");
79}
80
81static int create_strip_zones(struct mddev *mddev, struct r0conf **private_conf)
82{
83 int i, c, err;
84 sector_t curr_zone_end, sectors;
85 struct md_rdev *smallest, *rdev1, *rdev2, *rdev, **dev;
86 struct strip_zone *zone;
87 int cnt;
88 char b[BDEVNAME_SIZE];
89 char b2[BDEVNAME_SIZE];
90 struct r0conf *conf = kzalloc(sizeof(*conf), GFP_KERNEL);
91 bool discard_supported = false;
92
93 if (!conf)
94 return -ENOMEM;
95 rdev_for_each(rdev1, mddev) {
96 pr_debug("md/raid0:%s: looking at %s\n",
97 mdname(mddev),
98 bdevname(rdev1->bdev, b));
99 c = 0;
100
101 /* round size to chunk_size */
102 sectors = rdev1->sectors;
103 sector_div(sectors, mddev->chunk_sectors);
104 rdev1->sectors = sectors * mddev->chunk_sectors;
105
106 rdev_for_each(rdev2, mddev) {
107 pr_debug("md/raid0:%s: comparing %s(%llu)"
108 " with %s(%llu)\n",
109 mdname(mddev),
110 bdevname(rdev1->bdev,b),
111 (unsigned long long)rdev1->sectors,
112 bdevname(rdev2->bdev,b2),
113 (unsigned long long)rdev2->sectors);
114 if (rdev2 == rdev1) {
115 pr_debug("md/raid0:%s: END\n",
116 mdname(mddev));
117 break;
118 }
119 if (rdev2->sectors == rdev1->sectors) {
120 /*
121 * Not unique, don't count it as a new
122 * group
123 */
124 pr_debug("md/raid0:%s: EQUAL\n",
125 mdname(mddev));
126 c = 1;
127 break;
128 }
129 pr_debug("md/raid0:%s: NOT EQUAL\n",
130 mdname(mddev));
131 }
132 if (!c) {
133 pr_debug("md/raid0:%s: ==> UNIQUE\n",
134 mdname(mddev));
135 conf->nr_strip_zones++;
136 pr_debug("md/raid0:%s: %d zones\n",
137 mdname(mddev), conf->nr_strip_zones);
138 }
139 }
140 pr_debug("md/raid0:%s: FINAL %d zones\n",
141 mdname(mddev), conf->nr_strip_zones);
142 err = -ENOMEM;
143 conf->strip_zone = kzalloc(sizeof(struct strip_zone)*
144 conf->nr_strip_zones, GFP_KERNEL);
145 if (!conf->strip_zone)
146 goto abort;
147 conf->devlist = kzalloc(sizeof(struct md_rdev*)*
148 conf->nr_strip_zones*mddev->raid_disks,
149 GFP_KERNEL);
150 if (!conf->devlist)
151 goto abort;
152
153 /* The first zone must contain all devices, so here we check that
154 * there is a proper alignment of slots to devices and find them all
155 */
156 zone = &conf->strip_zone[0];
157 cnt = 0;
158 smallest = NULL;
159 dev = conf->devlist;
160 err = -EINVAL;
161 rdev_for_each(rdev1, mddev) {
162 int j = rdev1->raid_disk;
163
164 if (mddev->level == 10) {
165 /* taking over a raid10-n2 array */
166 j /= 2;
167 rdev1->new_raid_disk = j;
168 }
169
170 if (mddev->level == 1) {
171 /* taiking over a raid1 array-
172 * we have only one active disk
173 */
174 j = 0;
175 rdev1->new_raid_disk = j;
176 }
177
178 if (j < 0) {
179 printk(KERN_ERR
180 "md/raid0:%s: remove inactive devices before converting to RAID0\n",
181 mdname(mddev));
182 goto abort;
183 }
184 if (j >= mddev->raid_disks) {
185 printk(KERN_ERR "md/raid0:%s: bad disk number %d - "
186 "aborting!\n", mdname(mddev), j);
187 goto abort;
188 }
189 if (dev[j]) {
190 printk(KERN_ERR "md/raid0:%s: multiple devices for %d - "
191 "aborting!\n", mdname(mddev), j);
192 goto abort;
193 }
194 dev[j] = rdev1;
195
196 disk_stack_limits(mddev->gendisk, rdev1->bdev,
197 rdev1->data_offset << 9);
198
199 if (rdev1->bdev->bd_disk->queue->merge_bvec_fn)
200 conf->has_merge_bvec = 1;
201
202 if (!smallest || (rdev1->sectors < smallest->sectors))
203 smallest = rdev1;
204 cnt++;
205
206 if (blk_queue_discard(bdev_get_queue(rdev1->bdev)))
207 discard_supported = true;
208 }
209 if (cnt != mddev->raid_disks) {
210 printk(KERN_ERR "md/raid0:%s: too few disks (%d of %d) - "
211 "aborting!\n", mdname(mddev), cnt, mddev->raid_disks);
212 goto abort;
213 }
214 zone->nb_dev = cnt;
215 zone->zone_end = smallest->sectors * cnt;
216
217 curr_zone_end = zone->zone_end;
218
219 /* now do the other zones */
220 for (i = 1; i < conf->nr_strip_zones; i++)
221 {
222 int j;
223
224 zone = conf->strip_zone + i;
225 dev = conf->devlist + i * mddev->raid_disks;
226
227 pr_debug("md/raid0:%s: zone %d\n", mdname(mddev), i);
228 zone->dev_start = smallest->sectors;
229 smallest = NULL;
230 c = 0;
231
232 for (j=0; j<cnt; j++) {
233 rdev = conf->devlist[j];
234 if (rdev->sectors <= zone->dev_start) {
235 pr_debug("md/raid0:%s: checking %s ... nope\n",
236 mdname(mddev),
237 bdevname(rdev->bdev, b));
238 continue;
239 }
240 pr_debug("md/raid0:%s: checking %s ..."
241 " contained as device %d\n",
242 mdname(mddev),
243 bdevname(rdev->bdev, b), c);
244 dev[c] = rdev;
245 c++;
246 if (!smallest || rdev->sectors < smallest->sectors) {
247 smallest = rdev;
248 pr_debug("md/raid0:%s: (%llu) is smallest!.\n",
249 mdname(mddev),
250 (unsigned long long)rdev->sectors);
251 }
252 }
253
254 zone->nb_dev = c;
255 sectors = (smallest->sectors - zone->dev_start) * c;
256 pr_debug("md/raid0:%s: zone->nb_dev: %d, sectors: %llu\n",
257 mdname(mddev),
258 zone->nb_dev, (unsigned long long)sectors);
259
260 curr_zone_end += sectors;
261 zone->zone_end = curr_zone_end;
262
263 pr_debug("md/raid0:%s: current zone start: %llu\n",
264 mdname(mddev),
265 (unsigned long long)smallest->sectors);
266 }
267 mddev->queue->backing_dev_info.congested_fn = raid0_congested;
268 mddev->queue->backing_dev_info.congested_data = mddev;
269
270 /*
271 * now since we have the hard sector sizes, we can make sure
272 * chunk size is a multiple of that sector size
273 */
274 if ((mddev->chunk_sectors << 9) % queue_logical_block_size(mddev->queue)) {
275 printk(KERN_ERR "md/raid0:%s: chunk_size of %d not valid\n",
276 mdname(mddev),
277 mddev->chunk_sectors << 9);
278 goto abort;
279 }
280
281 blk_queue_io_min(mddev->queue, mddev->chunk_sectors << 9);
282 blk_queue_io_opt(mddev->queue,
283 (mddev->chunk_sectors << 9) * mddev->raid_disks);
284
285 if (!discard_supported)
286 queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, mddev->queue);
287 else
288 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, mddev->queue);
289
290 pr_debug("md/raid0:%s: done.\n", mdname(mddev));
291 *private_conf = conf;
292
293 return 0;
294abort:
295 kfree(conf->strip_zone);
296 kfree(conf->devlist);
297 kfree(conf);
298 *private_conf = ERR_PTR(err);
299 return err;
300}
301
302/* Find the zone which holds a particular offset
303 * Update *sectorp to be an offset in that zone
304 */
305static struct strip_zone *find_zone(struct r0conf *conf,
306 sector_t *sectorp)
307{
308 int i;
309 struct strip_zone *z = conf->strip_zone;
310 sector_t sector = *sectorp;
311
312 for (i = 0; i < conf->nr_strip_zones; i++)
313 if (sector < z[i].zone_end) {
314 if (i)
315 *sectorp = sector - z[i-1].zone_end;
316 return z + i;
317 }
318 BUG();
319}
320
321/*
322 * remaps the bio to the target device. we separate two flows.
323 * power 2 flow and a general flow for the sake of perfromance
324*/
325static struct md_rdev *map_sector(struct mddev *mddev, struct strip_zone *zone,
326 sector_t sector, sector_t *sector_offset)
327{
328 unsigned int sect_in_chunk;
329 sector_t chunk;
330 struct r0conf *conf = mddev->private;
331 int raid_disks = conf->strip_zone[0].nb_dev;
332 unsigned int chunk_sects = mddev->chunk_sectors;
333
334 if (is_power_of_2(chunk_sects)) {
335 int chunksect_bits = ffz(~chunk_sects);
336 /* find the sector offset inside the chunk */
337 sect_in_chunk = sector & (chunk_sects - 1);
338 sector >>= chunksect_bits;
339 /* chunk in zone */
340 chunk = *sector_offset;
341 /* quotient is the chunk in real device*/
342 sector_div(chunk, zone->nb_dev << chunksect_bits);
343 } else{
344 sect_in_chunk = sector_div(sector, chunk_sects);
345 chunk = *sector_offset;
346 sector_div(chunk, chunk_sects * zone->nb_dev);
347 }
348 /*
349 * position the bio over the real device
350 * real sector = chunk in device + starting of zone
351 * + the position in the chunk
352 */
353 *sector_offset = (chunk * chunk_sects) + sect_in_chunk;
354 return conf->devlist[(zone - conf->strip_zone)*raid_disks
355 + sector_div(sector, zone->nb_dev)];
356}
357
358/**
359 * raid0_mergeable_bvec -- tell bio layer if two requests can be merged
360 * @q: request queue
361 * @bvm: properties of new bio
362 * @biovec: the request that could be merged to it.
363 *
364 * Return amount of bytes we can accept at this offset
365 */
366static int raid0_mergeable_bvec(struct request_queue *q,
367 struct bvec_merge_data *bvm,
368 struct bio_vec *biovec)
369{
370 struct mddev *mddev = q->queuedata;
371 struct r0conf *conf = mddev->private;
372 sector_t sector = bvm->bi_sector + get_start_sect(bvm->bi_bdev);
373 sector_t sector_offset = sector;
374 int max;
375 unsigned int chunk_sectors = mddev->chunk_sectors;
376 unsigned int bio_sectors = bvm->bi_size >> 9;
377 struct strip_zone *zone;
378 struct md_rdev *rdev;
379 struct request_queue *subq;
380
381 if (is_power_of_2(chunk_sectors))
382 max = (chunk_sectors - ((sector & (chunk_sectors-1))
383 + bio_sectors)) << 9;
384 else
385 max = (chunk_sectors - (sector_div(sector, chunk_sectors)
386 + bio_sectors)) << 9;
387 if (max < 0)
388 max = 0; /* bio_add cannot handle a negative return */
389 if (max <= biovec->bv_len && bio_sectors == 0)
390 return biovec->bv_len;
391 if (max < biovec->bv_len)
392 /* too small already, no need to check further */
393 return max;
394 if (!conf->has_merge_bvec)
395 return max;
396
397 /* May need to check subordinate device */
398 sector = sector_offset;
399 zone = find_zone(mddev->private, §or_offset);
400 rdev = map_sector(mddev, zone, sector, §or_offset);
401 subq = bdev_get_queue(rdev->bdev);
402 if (subq->merge_bvec_fn) {
403 bvm->bi_bdev = rdev->bdev;
404 bvm->bi_sector = sector_offset + zone->dev_start +
405 rdev->data_offset;
406 return min(max, subq->merge_bvec_fn(subq, bvm, biovec));
407 } else
408 return max;
409}
410
411static sector_t raid0_size(struct mddev *mddev, sector_t sectors, int raid_disks)
412{
413 sector_t array_sectors = 0;
414 struct md_rdev *rdev;
415
416 WARN_ONCE(sectors || raid_disks,
417 "%s does not support generic reshape\n", __func__);
418
419 rdev_for_each(rdev, mddev)
420 array_sectors += (rdev->sectors &
421 ~(sector_t)(mddev->chunk_sectors-1));
422
423 return array_sectors;
424}
425
426static int raid0_stop(struct mddev *mddev);
427
428static int raid0_run(struct mddev *mddev)
429{
430 struct r0conf *conf;
431 int ret;
432
433 if (mddev->chunk_sectors == 0) {
434 printk(KERN_ERR "md/raid0:%s: chunk size must be set.\n",
435 mdname(mddev));
436 return -EINVAL;
437 }
438 if (md_check_no_bitmap(mddev))
439 return -EINVAL;
440 blk_queue_max_hw_sectors(mddev->queue, mddev->chunk_sectors);
441 blk_queue_max_write_same_sectors(mddev->queue, mddev->chunk_sectors);
442 blk_queue_max_discard_sectors(mddev->queue, mddev->chunk_sectors);
443
444 /* if private is not null, we are here after takeover */
445 if (mddev->private == NULL) {
446 ret = create_strip_zones(mddev, &conf);
447 if (ret < 0)
448 return ret;
449 mddev->private = conf;
450 }
451 conf = mddev->private;
452
453 /* calculate array device size */
454 md_set_array_sectors(mddev, raid0_size(mddev, 0, 0));
455
456 printk(KERN_INFO "md/raid0:%s: md_size is %llu sectors.\n",
457 mdname(mddev),
458 (unsigned long long)mddev->array_sectors);
459 /* calculate the max read-ahead size.
460 * For read-ahead of large files to be effective, we need to
461 * readahead at least twice a whole stripe. i.e. number of devices
462 * multiplied by chunk size times 2.
463 * If an individual device has an ra_pages greater than the
464 * chunk size, then we will not drive that device as hard as it
465 * wants. We consider this a configuration error: a larger
466 * chunksize should be used in that case.
467 */
468 {
469 int stripe = mddev->raid_disks *
470 (mddev->chunk_sectors << 9) / PAGE_SIZE;
471 if (mddev->queue->backing_dev_info.ra_pages < 2* stripe)
472 mddev->queue->backing_dev_info.ra_pages = 2* stripe;
473 }
474
475 blk_queue_merge_bvec(mddev->queue, raid0_mergeable_bvec);
476 dump_zones(mddev);
477
478 ret = md_integrity_register(mddev);
479 if (ret)
480 raid0_stop(mddev);
481
482 return ret;
483}
484
485static int raid0_stop(struct mddev *mddev)
486{
487 struct r0conf *conf = mddev->private;
488
489 blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/
490 kfree(conf->strip_zone);
491 kfree(conf->devlist);
492 kfree(conf);
493 mddev->private = NULL;
494 return 0;
495}
496
497/*
498 * Is io distribute over 1 or more chunks ?
499*/
500static inline int is_io_in_chunk_boundary(struct mddev *mddev,
501 unsigned int chunk_sects, struct bio *bio)
502{
503 if (likely(is_power_of_2(chunk_sects))) {
504 return chunk_sects >=
505 ((bio->bi_iter.bi_sector & (chunk_sects-1))
506 + bio_sectors(bio));
507 } else{
508 sector_t sector = bio->bi_iter.bi_sector;
509 return chunk_sects >= (sector_div(sector, chunk_sects)
510 + bio_sectors(bio));
511 }
512}
513
514static void raid0_make_request(struct mddev *mddev, struct bio *bio)
515{
516 struct strip_zone *zone;
517 struct md_rdev *tmp_dev;
518 struct bio *split;
519
520 if (unlikely(bio->bi_rw & REQ_FLUSH)) {
521 md_flush_request(mddev, bio);
522 return;
523 }
524
525 do {
526 sector_t sector = bio->bi_iter.bi_sector;
527 unsigned chunk_sects = mddev->chunk_sectors;
528
529 unsigned sectors = chunk_sects -
530 (likely(is_power_of_2(chunk_sects))
531 ? (sector & (chunk_sects-1))
532 : sector_div(sector, chunk_sects));
533
534 if (sectors < bio_sectors(bio)) {
535 split = bio_split(bio, sectors, GFP_NOIO, fs_bio_set);
536 bio_chain(split, bio);
537 } else {
538 split = bio;
539 }
540
541 zone = find_zone(mddev->private, §or);
542 tmp_dev = map_sector(mddev, zone, sector, §or);
543 split->bi_bdev = tmp_dev->bdev;
544 split->bi_iter.bi_sector = sector + zone->dev_start +
545 tmp_dev->data_offset;
546
547 if (unlikely((split->bi_rw & REQ_DISCARD) &&
548 !blk_queue_discard(bdev_get_queue(split->bi_bdev)))) {
549 /* Just ignore it */
550 bio_endio(split, 0);
551 } else
552 generic_make_request(split);
553 } while (split != bio);
554}
555
556static void raid0_status(struct seq_file *seq, struct mddev *mddev)
557{
558 seq_printf(seq, " %dk chunks", mddev->chunk_sectors / 2);
559 return;
560}
561
562static void *raid0_takeover_raid45(struct mddev *mddev)
563{
564 struct md_rdev *rdev;
565 struct r0conf *priv_conf;
566
567 if (mddev->degraded != 1) {
568 printk(KERN_ERR "md/raid0:%s: raid5 must be degraded! Degraded disks: %d\n",
569 mdname(mddev),
570 mddev->degraded);
571 return ERR_PTR(-EINVAL);
572 }
573
574 rdev_for_each(rdev, mddev) {
575 /* check slot number for a disk */
576 if (rdev->raid_disk == mddev->raid_disks-1) {
577 printk(KERN_ERR "md/raid0:%s: raid5 must have missing parity disk!\n",
578 mdname(mddev));
579 return ERR_PTR(-EINVAL);
580 }
581 rdev->sectors = mddev->dev_sectors;
582 }
583
584 /* Set new parameters */
585 mddev->new_level = 0;
586 mddev->new_layout = 0;
587 mddev->new_chunk_sectors = mddev->chunk_sectors;
588 mddev->raid_disks--;
589 mddev->delta_disks = -1;
590 /* make sure it will be not marked as dirty */
591 mddev->recovery_cp = MaxSector;
592
593 create_strip_zones(mddev, &priv_conf);
594 return priv_conf;
595}
596
597static void *raid0_takeover_raid10(struct mddev *mddev)
598{
599 struct r0conf *priv_conf;
600
601 /* Check layout:
602 * - far_copies must be 1
603 * - near_copies must be 2
604 * - disks number must be even
605 * - all mirrors must be already degraded
606 */
607 if (mddev->layout != ((1 << 8) + 2)) {
608 printk(KERN_ERR "md/raid0:%s:: Raid0 cannot takover layout: 0x%x\n",
609 mdname(mddev),
610 mddev->layout);
611 return ERR_PTR(-EINVAL);
612 }
613 if (mddev->raid_disks & 1) {
614 printk(KERN_ERR "md/raid0:%s: Raid0 cannot takover Raid10 with odd disk number.\n",
615 mdname(mddev));
616 return ERR_PTR(-EINVAL);
617 }
618 if (mddev->degraded != (mddev->raid_disks>>1)) {
619 printk(KERN_ERR "md/raid0:%s: All mirrors must be already degraded!\n",
620 mdname(mddev));
621 return ERR_PTR(-EINVAL);
622 }
623
624 /* Set new parameters */
625 mddev->new_level = 0;
626 mddev->new_layout = 0;
627 mddev->new_chunk_sectors = mddev->chunk_sectors;
628 mddev->delta_disks = - mddev->raid_disks / 2;
629 mddev->raid_disks += mddev->delta_disks;
630 mddev->degraded = 0;
631 /* make sure it will be not marked as dirty */
632 mddev->recovery_cp = MaxSector;
633
634 create_strip_zones(mddev, &priv_conf);
635 return priv_conf;
636}
637
638static void *raid0_takeover_raid1(struct mddev *mddev)
639{
640 struct r0conf *priv_conf;
641 int chunksect;
642
643 /* Check layout:
644 * - (N - 1) mirror drives must be already faulty
645 */
646 if ((mddev->raid_disks - 1) != mddev->degraded) {
647 printk(KERN_ERR "md/raid0:%s: (N - 1) mirrors drives must be already faulty!\n",
648 mdname(mddev));
649 return ERR_PTR(-EINVAL);
650 }
651
652 /*
653 * a raid1 doesn't have the notion of chunk size, so
654 * figure out the largest suitable size we can use.
655 */
656 chunksect = 64 * 2; /* 64K by default */
657
658 /* The array must be an exact multiple of chunksize */
659 while (chunksect && (mddev->array_sectors & (chunksect - 1)))
660 chunksect >>= 1;
661
662 if ((chunksect << 9) < PAGE_SIZE)
663 /* array size does not allow a suitable chunk size */
664 return ERR_PTR(-EINVAL);
665
666 /* Set new parameters */
667 mddev->new_level = 0;
668 mddev->new_layout = 0;
669 mddev->new_chunk_sectors = chunksect;
670 mddev->chunk_sectors = chunksect;
671 mddev->delta_disks = 1 - mddev->raid_disks;
672 mddev->raid_disks = 1;
673 /* make sure it will be not marked as dirty */
674 mddev->recovery_cp = MaxSector;
675
676 create_strip_zones(mddev, &priv_conf);
677 return priv_conf;
678}
679
680static void *raid0_takeover(struct mddev *mddev)
681{
682 /* raid0 can take over:
683 * raid4 - if all data disks are active.
684 * raid5 - providing it is Raid4 layout and one disk is faulty
685 * raid10 - assuming we have all necessary active disks
686 * raid1 - with (N -1) mirror drives faulty
687 */
688 if (mddev->level == 4)
689 return raid0_takeover_raid45(mddev);
690
691 if (mddev->level == 5) {
692 if (mddev->layout == ALGORITHM_PARITY_N)
693 return raid0_takeover_raid45(mddev);
694
695 printk(KERN_ERR "md/raid0:%s: Raid can only takeover Raid5 with layout: %d\n",
696 mdname(mddev), ALGORITHM_PARITY_N);
697 }
698
699 if (mddev->level == 10)
700 return raid0_takeover_raid10(mddev);
701
702 if (mddev->level == 1)
703 return raid0_takeover_raid1(mddev);
704
705 printk(KERN_ERR "Takeover from raid%i to raid0 not supported\n",
706 mddev->level);
707
708 return ERR_PTR(-EINVAL);
709}
710
711static void raid0_quiesce(struct mddev *mddev, int state)
712{
713}
714
715static struct md_personality raid0_personality=
716{
717 .name = "raid0",
718 .level = 0,
719 .owner = THIS_MODULE,
720 .make_request = raid0_make_request,
721 .run = raid0_run,
722 .stop = raid0_stop,
723 .status = raid0_status,
724 .size = raid0_size,
725 .takeover = raid0_takeover,
726 .quiesce = raid0_quiesce,
727};
728
729static int __init raid0_init (void)
730{
731 return register_md_personality (&raid0_personality);
732}
733
734static void raid0_exit (void)
735{
736 unregister_md_personality (&raid0_personality);
737}
738
739module_init(raid0_init);
740module_exit(raid0_exit);
741MODULE_LICENSE("GPL");
742MODULE_DESCRIPTION("RAID0 (striping) personality for MD");
743MODULE_ALIAS("md-personality-2"); /* RAID0 */
744MODULE_ALIAS("md-raid0");
745MODULE_ALIAS("md-level-0");