1 // SPDX-License-Identifier: GPL-2.0-or-later
3 raid0.c : Multiple Devices driver for Linux
4 Copyright (C) 1994-96 Marc ZYNGIER
5 <zyngier@ufr-info-p7.ibp.fr> or
7 Copyright (C) 1999, 2000 Ingo Molnar, Red Hat
9 RAID-0 management functions.
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>
22 static int default_layout = 0;
23 module_param(default_layout, int, 0644);
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))
33 * inform the user of the raid configuration
35 static void dump_zones(struct mddev *mddev)
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",
44 conf->nr_strip_zones, conf->nr_strip_zones==1?"":"s");
45 for (j = 0; j < conf->nr_strip_zones; j++) {
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);
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;
63 static int create_strip_zones(struct mddev *mddev, struct r0conf **private_conf)
66 sector_t curr_zone_end, sectors;
67 struct md_rdev *smallest, *rdev1, *rdev2, *rdev, **dev;
68 struct strip_zone *zone;
70 struct r0conf *conf = kzalloc(sizeof(*conf), GFP_KERNEL);
71 unsigned blksize = 512;
73 *private_conf = ERR_PTR(-ENOMEM);
76 rdev_for_each(rdev1, mddev) {
77 pr_debug("md/raid0:%s: looking at %pg\n",
82 /* round size to chunk_size */
83 sectors = rdev1->sectors;
84 sector_div(sectors, mddev->chunk_sectors);
85 rdev1->sectors = sectors * mddev->chunk_sectors;
87 blksize = max(blksize, queue_logical_block_size(
88 rdev1->bdev->bd_disk->queue));
90 rdev_for_each(rdev2, mddev) {
91 pr_debug("md/raid0:%s: comparing %pg(%llu)"
95 (unsigned long long)rdev1->sectors,
97 (unsigned long long)rdev2->sectors);
99 pr_debug("md/raid0:%s: END\n",
103 if (rdev2->sectors == rdev1->sectors) {
105 * Not unique, don't count it as a new
108 pr_debug("md/raid0:%s: EQUAL\n",
113 pr_debug("md/raid0:%s: NOT EQUAL\n",
117 pr_debug("md/raid0:%s: ==> UNIQUE\n",
119 conf->nr_strip_zones++;
120 pr_debug("md/raid0:%s: %d zones\n",
121 mdname(mddev), conf->nr_strip_zones);
124 pr_debug("md/raid0:%s: FINAL %d zones\n",
125 mdname(mddev), conf->nr_strip_zones);
128 * now since we have the hard sector sizes, we can make sure
129 * chunk size is a multiple of that sector size
131 if ((mddev->chunk_sectors << 9) % blksize) {
132 pr_warn("md/raid0:%s: chunk_size of %d not multiple of block size %d\n",
134 mddev->chunk_sectors << 9, blksize);
140 conf->strip_zone = kcalloc(conf->nr_strip_zones,
141 sizeof(struct strip_zone),
143 if (!conf->strip_zone)
145 conf->devlist = kzalloc(array3_size(sizeof(struct md_rdev *),
146 conf->nr_strip_zones,
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
155 zone = &conf->strip_zone[0];
160 rdev_for_each(rdev1, mddev) {
161 int j = rdev1->raid_disk;
163 if (mddev->level == 10) {
164 /* taking over a raid10-n2 array */
166 rdev1->new_raid_disk = j;
169 if (mddev->level == 1) {
170 /* taiking over a raid1 array-
171 * we have only one active disk
174 rdev1->new_raid_disk = j;
178 pr_warn("md/raid0:%s: remove inactive devices before converting to RAID0\n",
182 if (j >= mddev->raid_disks) {
183 pr_warn("md/raid0:%s: bad disk number %d - aborting!\n",
188 pr_warn("md/raid0:%s: multiple devices for %d - aborting!\n",
194 if (!smallest || (rdev1->sectors < smallest->sectors))
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);
204 zone->zone_end = smallest->sectors * cnt;
206 curr_zone_end = zone->zone_end;
208 /* now do the other zones */
209 for (i = 1; i < conf->nr_strip_zones; i++)
213 zone = conf->strip_zone + i;
214 dev = conf->devlist + i * mddev->raid_disks;
216 pr_debug("md/raid0:%s: zone %d\n", mdname(mddev), i);
217 zone->dev_start = smallest->sectors;
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",
229 pr_debug("md/raid0:%s: checking %pg ..."
230 " contained as device %d\n",
235 if (!smallest || rdev->sectors < smallest->sectors) {
237 pr_debug("md/raid0:%s: (%llu) is smallest!.\n",
239 (unsigned long long)rdev->sectors);
244 sectors = (smallest->sectors - zone->dev_start) * c;
245 pr_debug("md/raid0:%s: zone->nb_dev: %d, sectors: %llu\n",
247 zone->nb_dev, (unsigned long long)sectors);
249 curr_zone_end += sectors;
250 zone->zone_end = curr_zone_end;
252 pr_debug("md/raid0:%s: current zone start: %llu\n",
254 (unsigned long long)smallest->sectors);
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;
266 pr_err("md/raid0:%s: cannot assemble multi-zone RAID0 with default_layout setting\n",
268 pr_err("md/raid0: please set raid0.default_layout to 1 or 2\n");
273 pr_debug("md/raid0:%s: done.\n", mdname(mddev));
274 *private_conf = conf;
278 kfree(conf->strip_zone);
279 kfree(conf->devlist);
281 *private_conf = ERR_PTR(err);
285 /* Find the zone which holds a particular offset
286 * Update *sectorp to be an offset in that zone
288 static struct strip_zone *find_zone(struct r0conf *conf,
292 struct strip_zone *z = conf->strip_zone;
293 sector_t sector = *sectorp;
295 for (i = 0; i < conf->nr_strip_zones; i++)
296 if (sector < z[i].zone_end) {
298 *sectorp = sector - z[i-1].zone_end;
305 * remaps the bio to the target device. we separate two flows.
306 * power 2 flow and a general flow for the sake of performance
308 static struct md_rdev *map_sector(struct mddev *mddev, struct strip_zone *zone,
309 sector_t sector, sector_t *sector_offset)
311 unsigned int sect_in_chunk;
313 struct r0conf *conf = mddev->private;
314 int raid_disks = conf->strip_zone[0].nb_dev;
315 unsigned int chunk_sects = mddev->chunk_sectors;
317 if (is_power_of_2(chunk_sects)) {
318 int chunksect_bits = ffz(~chunk_sects);
319 /* find the sector offset inside the chunk */
320 sect_in_chunk = sector & (chunk_sects - 1);
321 sector >>= chunksect_bits;
323 chunk = *sector_offset;
324 /* quotient is the chunk in real device*/
325 sector_div(chunk, zone->nb_dev << chunksect_bits);
327 sect_in_chunk = sector_div(sector, chunk_sects);
328 chunk = *sector_offset;
329 sector_div(chunk, chunk_sects * zone->nb_dev);
332 * position the bio over the real device
333 * real sector = chunk in device + starting of zone
334 * + the position in the chunk
336 *sector_offset = (chunk * chunk_sects) + sect_in_chunk;
337 return conf->devlist[(zone - conf->strip_zone)*raid_disks
338 + sector_div(sector, zone->nb_dev)];
341 static sector_t raid0_size(struct mddev *mddev, sector_t sectors, int raid_disks)
343 sector_t array_sectors = 0;
344 struct md_rdev *rdev;
346 WARN_ONCE(sectors || raid_disks,
347 "%s does not support generic reshape\n", __func__);
349 rdev_for_each(rdev, mddev)
350 array_sectors += (rdev->sectors &
351 ~(sector_t)(mddev->chunk_sectors-1));
353 return array_sectors;
356 static void free_conf(struct mddev *mddev, struct r0conf *conf)
358 kfree(conf->strip_zone);
359 kfree(conf->devlist);
363 static void raid0_free(struct mddev *mddev, void *priv)
365 struct r0conf *conf = priv;
367 free_conf(mddev, conf);
368 acct_bioset_exit(mddev);
371 static int raid0_run(struct mddev *mddev)
376 if (mddev->chunk_sectors == 0) {
377 pr_warn("md/raid0:%s: chunk size must be set.\n", mdname(mddev));
380 if (md_check_no_bitmap(mddev))
383 if (acct_bioset_init(mddev)) {
384 pr_err("md/raid0:%s: alloc acct bioset failed.\n", mdname(mddev));
388 /* if private is not null, we are here after takeover */
389 if (mddev->private == NULL) {
390 ret = create_strip_zones(mddev, &conf);
393 mddev->private = conf;
395 conf = mddev->private;
397 struct md_rdev *rdev;
399 blk_queue_max_hw_sectors(mddev->queue, mddev->chunk_sectors);
400 blk_queue_max_write_zeroes_sectors(mddev->queue, mddev->chunk_sectors);
402 blk_queue_io_min(mddev->queue, mddev->chunk_sectors << 9);
403 blk_queue_io_opt(mddev->queue,
404 (mddev->chunk_sectors << 9) * mddev->raid_disks);
406 rdev_for_each(rdev, mddev) {
407 disk_stack_limits(mddev->gendisk, rdev->bdev,
408 rdev->data_offset << 9);
412 /* calculate array device size */
413 md_set_array_sectors(mddev, raid0_size(mddev, 0, 0));
415 pr_debug("md/raid0:%s: md_size is %llu sectors.\n",
417 (unsigned long long)mddev->array_sectors);
421 ret = md_integrity_register(mddev);
428 free_conf(mddev, conf);
430 acct_bioset_exit(mddev);
434 static void raid0_handle_discard(struct mddev *mddev, struct bio *bio)
436 struct r0conf *conf = mddev->private;
437 struct strip_zone *zone;
438 sector_t start = bio->bi_iter.bi_sector;
440 unsigned int stripe_size;
441 sector_t first_stripe_index, last_stripe_index;
442 sector_t start_disk_offset;
443 unsigned int start_disk_index;
444 sector_t end_disk_offset;
445 unsigned int end_disk_index;
448 zone = find_zone(conf, &start);
450 if (bio_end_sector(bio) > zone->zone_end) {
451 struct bio *split = bio_split(bio,
452 zone->zone_end - bio->bi_iter.bi_sector, GFP_NOIO,
454 bio_chain(split, bio);
455 submit_bio_noacct(bio);
457 end = zone->zone_end;
459 end = bio_end_sector(bio);
461 if (zone != conf->strip_zone)
462 end = end - zone[-1].zone_end;
464 /* Now start and end is the offset in zone */
465 stripe_size = zone->nb_dev * mddev->chunk_sectors;
467 first_stripe_index = start;
468 sector_div(first_stripe_index, stripe_size);
469 last_stripe_index = end;
470 sector_div(last_stripe_index, stripe_size);
472 start_disk_index = (int)(start - first_stripe_index * stripe_size) /
473 mddev->chunk_sectors;
474 start_disk_offset = ((int)(start - first_stripe_index * stripe_size) %
475 mddev->chunk_sectors) +
476 first_stripe_index * mddev->chunk_sectors;
477 end_disk_index = (int)(end - last_stripe_index * stripe_size) /
478 mddev->chunk_sectors;
479 end_disk_offset = ((int)(end - last_stripe_index * stripe_size) %
480 mddev->chunk_sectors) +
481 last_stripe_index * mddev->chunk_sectors;
483 for (disk = 0; disk < zone->nb_dev; disk++) {
484 sector_t dev_start, dev_end;
485 struct md_rdev *rdev;
487 if (disk < start_disk_index)
488 dev_start = (first_stripe_index + 1) *
489 mddev->chunk_sectors;
490 else if (disk > start_disk_index)
491 dev_start = first_stripe_index * mddev->chunk_sectors;
493 dev_start = start_disk_offset;
495 if (disk < end_disk_index)
496 dev_end = (last_stripe_index + 1) * mddev->chunk_sectors;
497 else if (disk > end_disk_index)
498 dev_end = last_stripe_index * mddev->chunk_sectors;
500 dev_end = end_disk_offset;
502 if (dev_end <= dev_start)
505 rdev = conf->devlist[(zone - conf->strip_zone) *
506 conf->strip_zone[0].nb_dev + disk];
507 md_submit_discard_bio(mddev, rdev, bio,
508 dev_start + zone->dev_start + rdev->data_offset,
509 dev_end - dev_start);
514 static bool raid0_make_request(struct mddev *mddev, struct bio *bio)
516 struct r0conf *conf = mddev->private;
517 struct strip_zone *zone;
518 struct md_rdev *tmp_dev;
521 sector_t orig_sector;
522 unsigned chunk_sects;
525 if (unlikely(bio->bi_opf & REQ_PREFLUSH)
526 && md_flush_request(mddev, bio))
529 if (unlikely((bio_op(bio) == REQ_OP_DISCARD))) {
530 raid0_handle_discard(mddev, bio);
534 bio_sector = bio->bi_iter.bi_sector;
536 chunk_sects = mddev->chunk_sectors;
538 sectors = chunk_sects -
539 (likely(is_power_of_2(chunk_sects))
540 ? (sector & (chunk_sects-1))
541 : sector_div(sector, chunk_sects));
543 /* Restore due to sector_div */
546 if (sectors < bio_sectors(bio)) {
547 struct bio *split = bio_split(bio, sectors, GFP_NOIO,
549 bio_chain(split, bio);
550 submit_bio_noacct(bio);
554 if (bio->bi_pool != &mddev->bio_set)
555 md_account_bio(mddev, &bio);
557 orig_sector = sector;
558 zone = find_zone(mddev->private, §or);
559 switch (conf->layout) {
560 case RAID0_ORIG_LAYOUT:
561 tmp_dev = map_sector(mddev, zone, orig_sector, §or);
563 case RAID0_ALT_MULTIZONE_LAYOUT:
564 tmp_dev = map_sector(mddev, zone, sector, §or);
567 WARN(1, "md/raid0:%s: Invalid layout\n", mdname(mddev));
572 if (unlikely(is_rdev_broken(tmp_dev))) {
574 md_error(mddev, tmp_dev);
578 bio_set_dev(bio, tmp_dev->bdev);
579 bio->bi_iter.bi_sector = sector + zone->dev_start +
580 tmp_dev->data_offset;
583 trace_block_bio_remap(bio, disk_devt(mddev->gendisk),
585 mddev_check_write_zeroes(mddev, bio);
586 submit_bio_noacct(bio);
590 static void raid0_status(struct seq_file *seq, struct mddev *mddev)
592 seq_printf(seq, " %dk chunks", mddev->chunk_sectors / 2);
596 static void raid0_error(struct mddev *mddev, struct md_rdev *rdev)
598 if (!test_and_set_bit(MD_BROKEN, &mddev->flags)) {
599 char *md_name = mdname(mddev);
601 pr_crit("md/raid0%s: Disk failure on %pg detected, failing array.\n",
602 md_name, rdev->bdev);
606 static void *raid0_takeover_raid45(struct mddev *mddev)
608 struct md_rdev *rdev;
609 struct r0conf *priv_conf;
611 if (mddev->degraded != 1) {
612 pr_warn("md/raid0:%s: raid5 must be degraded! Degraded disks: %d\n",
615 return ERR_PTR(-EINVAL);
618 rdev_for_each(rdev, mddev) {
619 /* check slot number for a disk */
620 if (rdev->raid_disk == mddev->raid_disks-1) {
621 pr_warn("md/raid0:%s: raid5 must have missing parity disk!\n",
623 return ERR_PTR(-EINVAL);
625 rdev->sectors = mddev->dev_sectors;
628 /* Set new parameters */
629 mddev->new_level = 0;
630 mddev->new_layout = 0;
631 mddev->new_chunk_sectors = mddev->chunk_sectors;
633 mddev->delta_disks = -1;
634 /* make sure it will be not marked as dirty */
635 mddev->recovery_cp = MaxSector;
636 mddev_clear_unsupported_flags(mddev, UNSUPPORTED_MDDEV_FLAGS);
638 create_strip_zones(mddev, &priv_conf);
643 static void *raid0_takeover_raid10(struct mddev *mddev)
645 struct r0conf *priv_conf;
648 * - far_copies must be 1
649 * - near_copies must be 2
650 * - disks number must be even
651 * - all mirrors must be already degraded
653 if (mddev->layout != ((1 << 8) + 2)) {
654 pr_warn("md/raid0:%s:: Raid0 cannot takeover layout: 0x%x\n",
657 return ERR_PTR(-EINVAL);
659 if (mddev->raid_disks & 1) {
660 pr_warn("md/raid0:%s: Raid0 cannot takeover Raid10 with odd disk number.\n",
662 return ERR_PTR(-EINVAL);
664 if (mddev->degraded != (mddev->raid_disks>>1)) {
665 pr_warn("md/raid0:%s: All mirrors must be already degraded!\n",
667 return ERR_PTR(-EINVAL);
670 /* Set new parameters */
671 mddev->new_level = 0;
672 mddev->new_layout = 0;
673 mddev->new_chunk_sectors = mddev->chunk_sectors;
674 mddev->delta_disks = - mddev->raid_disks / 2;
675 mddev->raid_disks += mddev->delta_disks;
677 /* make sure it will be not marked as dirty */
678 mddev->recovery_cp = MaxSector;
679 mddev_clear_unsupported_flags(mddev, UNSUPPORTED_MDDEV_FLAGS);
681 create_strip_zones(mddev, &priv_conf);
685 static void *raid0_takeover_raid1(struct mddev *mddev)
687 struct r0conf *priv_conf;
691 * - (N - 1) mirror drives must be already faulty
693 if ((mddev->raid_disks - 1) != mddev->degraded) {
694 pr_err("md/raid0:%s: (N - 1) mirrors drives must be already faulty!\n",
696 return ERR_PTR(-EINVAL);
700 * a raid1 doesn't have the notion of chunk size, so
701 * figure out the largest suitable size we can use.
703 chunksect = 64 * 2; /* 64K by default */
705 /* The array must be an exact multiple of chunksize */
706 while (chunksect && (mddev->array_sectors & (chunksect - 1)))
709 if ((chunksect << 9) < PAGE_SIZE)
710 /* array size does not allow a suitable chunk size */
711 return ERR_PTR(-EINVAL);
713 /* Set new parameters */
714 mddev->new_level = 0;
715 mddev->new_layout = 0;
716 mddev->new_chunk_sectors = chunksect;
717 mddev->chunk_sectors = chunksect;
718 mddev->delta_disks = 1 - mddev->raid_disks;
719 mddev->raid_disks = 1;
720 /* make sure it will be not marked as dirty */
721 mddev->recovery_cp = MaxSector;
722 mddev_clear_unsupported_flags(mddev, UNSUPPORTED_MDDEV_FLAGS);
724 create_strip_zones(mddev, &priv_conf);
728 static void *raid0_takeover(struct mddev *mddev)
730 /* raid0 can take over:
731 * raid4 - if all data disks are active.
732 * raid5 - providing it is Raid4 layout and one disk is faulty
733 * raid10 - assuming we have all necessary active disks
734 * raid1 - with (N -1) mirror drives faulty
738 pr_warn("md/raid0: %s: cannot takeover array with bitmap\n",
740 return ERR_PTR(-EBUSY);
742 if (mddev->level == 4)
743 return raid0_takeover_raid45(mddev);
745 if (mddev->level == 5) {
746 if (mddev->layout == ALGORITHM_PARITY_N)
747 return raid0_takeover_raid45(mddev);
749 pr_warn("md/raid0:%s: Raid can only takeover Raid5 with layout: %d\n",
750 mdname(mddev), ALGORITHM_PARITY_N);
753 if (mddev->level == 10)
754 return raid0_takeover_raid10(mddev);
756 if (mddev->level == 1)
757 return raid0_takeover_raid1(mddev);
759 pr_warn("Takeover from raid%i to raid0 not supported\n",
762 return ERR_PTR(-EINVAL);
765 static void raid0_quiesce(struct mddev *mddev, int quiesce)
769 static struct md_personality raid0_personality=
773 .owner = THIS_MODULE,
774 .make_request = raid0_make_request,
777 .status = raid0_status,
779 .takeover = raid0_takeover,
780 .quiesce = raid0_quiesce,
781 .error_handler = raid0_error,
784 static int __init raid0_init (void)
786 return register_md_personality (&raid0_personality);
789 static void raid0_exit (void)
791 unregister_md_personality (&raid0_personality);
794 module_init(raid0_init);
795 module_exit(raid0_exit);
796 MODULE_LICENSE("GPL");
797 MODULE_DESCRIPTION("RAID0 (striping) personality for MD");
798 MODULE_ALIAS("md-personality-2"); /* RAID0 */
799 MODULE_ALIAS("md-raid0");
800 MODULE_ALIAS("md-level-0");