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 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;
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,
285 pr_debug("md/raid0:%s: done.\n", mdname(mddev));
286 *private_conf = conf;
290 kfree(conf->strip_zone);
291 kfree(conf->devlist);
293 *private_conf = ERR_PTR(err);
297 /* Find the zone which holds a particular offset
298 * Update *sectorp to be an offset in that zone
300 static struct strip_zone *find_zone(struct r0conf *conf,
304 struct strip_zone *z = conf->strip_zone;
305 sector_t sector = *sectorp;
307 for (i = 0; i < conf->nr_strip_zones; i++)
308 if (sector < z[i].zone_end) {
310 *sectorp = sector - z[i-1].zone_end;
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
320 static struct md_rdev *map_sector(struct mddev *mddev, struct strip_zone *zone,
321 sector_t sector, sector_t *sector_offset)
323 unsigned int sect_in_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;
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;
335 chunk = *sector_offset;
336 /* quotient is the chunk in real device*/
337 sector_div(chunk, zone->nb_dev << chunksect_bits);
339 sect_in_chunk = sector_div(sector, chunk_sects);
340 chunk = *sector_offset;
341 sector_div(chunk, chunk_sects * zone->nb_dev);
344 * position the bio over the real device
345 * real sector = chunk in device + starting of zone
346 * + the position in the chunk
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)];
353 static sector_t raid0_size(struct mddev *mddev, sector_t sectors, int raid_disks)
355 sector_t array_sectors = 0;
356 struct md_rdev *rdev;
358 WARN_ONCE(sectors || raid_disks,
359 "%s does not support generic reshape\n", __func__);
361 rdev_for_each(rdev, mddev)
362 array_sectors += (rdev->sectors &
363 ~(sector_t)(mddev->chunk_sectors-1));
365 return array_sectors;
368 static void free_conf(struct mddev *mddev, struct r0conf *conf)
370 kfree(conf->strip_zone);
371 kfree(conf->devlist);
375 static void raid0_free(struct mddev *mddev, void *priv)
377 struct r0conf *conf = priv;
379 free_conf(mddev, conf);
382 static int raid0_set_limits(struct mddev *mddev)
384 struct queue_limits lim;
386 blk_set_stacking_limits(&lim);
387 lim.max_hw_sectors = mddev->chunk_sectors;
388 lim.max_write_zeroes_sectors = mddev->chunk_sectors;
389 lim.io_min = mddev->chunk_sectors << 9;
390 lim.io_opt = lim.io_min * mddev->raid_disks;
391 mddev_stack_rdev_limits(mddev, &lim);
392 return queue_limits_set(mddev->gendisk->queue, &lim);
395 static int raid0_run(struct mddev *mddev)
400 if (mddev->chunk_sectors == 0) {
401 pr_warn("md/raid0:%s: chunk size must be set.\n", mdname(mddev));
404 if (md_check_no_bitmap(mddev))
407 /* if private is not null, we are here after takeover */
408 if (mddev->private == NULL) {
409 ret = create_strip_zones(mddev, &conf);
412 mddev->private = conf;
414 conf = mddev->private;
415 if (!mddev_is_dm(mddev)) {
416 ret = raid0_set_limits(mddev);
421 /* calculate array device size */
422 md_set_array_sectors(mddev, raid0_size(mddev, 0, 0));
424 pr_debug("md/raid0:%s: md_size is %llu sectors.\n",
426 (unsigned long long)mddev->array_sectors);
430 ret = md_integrity_register(mddev);
435 free_conf(mddev, conf);
440 * Convert disk_index to the disk order in which it is read/written.
441 * For example, if we have 4 disks, they are numbered 0,1,2,3. If we
442 * write the disks starting at disk 3, then the read/write order would
443 * be disk 3, then 0, then 1, and then disk 2 and we want map_disk_shift()
444 * to map the disks as follows 0,1,2,3 => 1,2,3,0. So disk 0 would map
445 * to 1, 1 to 2, 2 to 3, and 3 to 0. That way we can compare disks in
446 * that 'output' space to understand the read/write disk ordering.
448 static int map_disk_shift(int disk_index, int num_disks, int disk_shift)
450 return ((disk_index + num_disks - disk_shift) % num_disks);
453 static void raid0_handle_discard(struct mddev *mddev, struct bio *bio)
455 struct r0conf *conf = mddev->private;
456 struct strip_zone *zone;
457 sector_t start = bio->bi_iter.bi_sector;
459 unsigned int stripe_size;
460 sector_t first_stripe_index, last_stripe_index;
461 sector_t start_disk_offset;
462 unsigned int start_disk_index;
463 sector_t end_disk_offset;
464 unsigned int end_disk_index;
466 sector_t orig_start, orig_end;
469 zone = find_zone(conf, &start);
471 if (bio_end_sector(bio) > zone->zone_end) {
472 struct bio *split = bio_split(bio,
473 zone->zone_end - bio->bi_iter.bi_sector, GFP_NOIO,
475 bio_chain(split, bio);
476 submit_bio_noacct(bio);
478 end = zone->zone_end;
480 end = bio_end_sector(bio);
483 if (zone != conf->strip_zone)
484 end = end - zone[-1].zone_end;
486 /* Now start and end is the offset in zone */
487 stripe_size = zone->nb_dev * mddev->chunk_sectors;
489 first_stripe_index = start;
490 sector_div(first_stripe_index, stripe_size);
491 last_stripe_index = end;
492 sector_div(last_stripe_index, stripe_size);
494 /* In the first zone the original and alternate layouts are the same */
495 if ((conf->layout == RAID0_ORIG_LAYOUT) && (zone != conf->strip_zone)) {
496 sector_div(orig_start, mddev->chunk_sectors);
497 start_disk_index = sector_div(orig_start, zone->nb_dev);
498 start_disk_index = map_disk_shift(start_disk_index,
501 sector_div(orig_end, mddev->chunk_sectors);
502 end_disk_index = sector_div(orig_end, zone->nb_dev);
503 end_disk_index = map_disk_shift(end_disk_index,
504 zone->nb_dev, zone->disk_shift);
506 start_disk_index = (int)(start - first_stripe_index * stripe_size) /
507 mddev->chunk_sectors;
508 end_disk_index = (int)(end - last_stripe_index * stripe_size) /
509 mddev->chunk_sectors;
511 start_disk_offset = ((int)(start - first_stripe_index * stripe_size) %
512 mddev->chunk_sectors) +
513 first_stripe_index * mddev->chunk_sectors;
514 end_disk_offset = ((int)(end - last_stripe_index * stripe_size) %
515 mddev->chunk_sectors) +
516 last_stripe_index * mddev->chunk_sectors;
518 for (disk = 0; disk < zone->nb_dev; disk++) {
519 sector_t dev_start, dev_end;
520 struct md_rdev *rdev;
523 compare_disk = map_disk_shift(disk, zone->nb_dev,
526 if (compare_disk < start_disk_index)
527 dev_start = (first_stripe_index + 1) *
528 mddev->chunk_sectors;
529 else if (compare_disk > start_disk_index)
530 dev_start = first_stripe_index * mddev->chunk_sectors;
532 dev_start = start_disk_offset;
534 if (compare_disk < end_disk_index)
535 dev_end = (last_stripe_index + 1) * mddev->chunk_sectors;
536 else if (compare_disk > end_disk_index)
537 dev_end = last_stripe_index * mddev->chunk_sectors;
539 dev_end = end_disk_offset;
541 if (dev_end <= dev_start)
544 rdev = conf->devlist[(zone - conf->strip_zone) *
545 conf->strip_zone[0].nb_dev + disk];
546 md_submit_discard_bio(mddev, rdev, bio,
547 dev_start + zone->dev_start + rdev->data_offset,
548 dev_end - dev_start);
553 static void raid0_map_submit_bio(struct mddev *mddev, struct bio *bio)
555 struct r0conf *conf = mddev->private;
556 struct strip_zone *zone;
557 struct md_rdev *tmp_dev;
558 sector_t bio_sector = bio->bi_iter.bi_sector;
559 sector_t sector = bio_sector;
561 md_account_bio(mddev, &bio);
563 zone = find_zone(mddev->private, §or);
564 switch (conf->layout) {
565 case RAID0_ORIG_LAYOUT:
566 tmp_dev = map_sector(mddev, zone, bio_sector, §or);
568 case RAID0_ALT_MULTIZONE_LAYOUT:
569 tmp_dev = map_sector(mddev, zone, sector, §or);
572 WARN(1, "md/raid0:%s: Invalid layout\n", mdname(mddev));
577 if (unlikely(is_rdev_broken(tmp_dev))) {
579 md_error(mddev, tmp_dev);
583 bio_set_dev(bio, tmp_dev->bdev);
584 bio->bi_iter.bi_sector = sector + zone->dev_start +
585 tmp_dev->data_offset;
586 mddev_trace_remap(mddev, bio, bio_sector);
587 mddev_check_write_zeroes(mddev, bio);
588 submit_bio_noacct(bio);
591 static bool raid0_make_request(struct mddev *mddev, struct bio *bio)
594 unsigned chunk_sects;
597 if (unlikely(bio->bi_opf & REQ_PREFLUSH)
598 && md_flush_request(mddev, bio))
601 if (unlikely((bio_op(bio) == REQ_OP_DISCARD))) {
602 raid0_handle_discard(mddev, bio);
606 sector = bio->bi_iter.bi_sector;
607 chunk_sects = mddev->chunk_sectors;
609 sectors = chunk_sects -
610 (likely(is_power_of_2(chunk_sects))
611 ? (sector & (chunk_sects-1))
612 : sector_div(sector, chunk_sects));
614 if (sectors < bio_sectors(bio)) {
615 struct bio *split = bio_split(bio, sectors, GFP_NOIO,
617 bio_chain(split, bio);
618 raid0_map_submit_bio(mddev, bio);
622 raid0_map_submit_bio(mddev, bio);
626 static void raid0_status(struct seq_file *seq, struct mddev *mddev)
628 seq_printf(seq, " %dk chunks", mddev->chunk_sectors / 2);
632 static void raid0_error(struct mddev *mddev, struct md_rdev *rdev)
634 if (!test_and_set_bit(MD_BROKEN, &mddev->flags)) {
635 char *md_name = mdname(mddev);
637 pr_crit("md/raid0%s: Disk failure on %pg detected, failing array.\n",
638 md_name, rdev->bdev);
642 static void *raid0_takeover_raid45(struct mddev *mddev)
644 struct md_rdev *rdev;
645 struct r0conf *priv_conf;
647 if (mddev->degraded != 1) {
648 pr_warn("md/raid0:%s: raid5 must be degraded! Degraded disks: %d\n",
651 return ERR_PTR(-EINVAL);
654 rdev_for_each(rdev, mddev) {
655 /* check slot number for a disk */
656 if (rdev->raid_disk == mddev->raid_disks-1) {
657 pr_warn("md/raid0:%s: raid5 must have missing parity disk!\n",
659 return ERR_PTR(-EINVAL);
661 rdev->sectors = mddev->dev_sectors;
664 /* Set new parameters */
665 mddev->new_level = 0;
666 mddev->new_layout = 0;
667 mddev->new_chunk_sectors = mddev->chunk_sectors;
669 mddev->delta_disks = -1;
670 /* make sure it will be not marked as dirty */
671 mddev->recovery_cp = MaxSector;
672 mddev_clear_unsupported_flags(mddev, UNSUPPORTED_MDDEV_FLAGS);
674 create_strip_zones(mddev, &priv_conf);
679 static void *raid0_takeover_raid10(struct mddev *mddev)
681 struct r0conf *priv_conf;
684 * - far_copies must be 1
685 * - near_copies must be 2
686 * - disks number must be even
687 * - all mirrors must be already degraded
689 if (mddev->layout != ((1 << 8) + 2)) {
690 pr_warn("md/raid0:%s:: Raid0 cannot takeover layout: 0x%x\n",
693 return ERR_PTR(-EINVAL);
695 if (mddev->raid_disks & 1) {
696 pr_warn("md/raid0:%s: Raid0 cannot takeover Raid10 with odd disk number.\n",
698 return ERR_PTR(-EINVAL);
700 if (mddev->degraded != (mddev->raid_disks>>1)) {
701 pr_warn("md/raid0:%s: All mirrors must be already degraded!\n",
703 return ERR_PTR(-EINVAL);
706 /* Set new parameters */
707 mddev->new_level = 0;
708 mddev->new_layout = 0;
709 mddev->new_chunk_sectors = mddev->chunk_sectors;
710 mddev->delta_disks = - mddev->raid_disks / 2;
711 mddev->raid_disks += mddev->delta_disks;
713 /* make sure it will be not marked as dirty */
714 mddev->recovery_cp = MaxSector;
715 mddev_clear_unsupported_flags(mddev, UNSUPPORTED_MDDEV_FLAGS);
717 create_strip_zones(mddev, &priv_conf);
721 static void *raid0_takeover_raid1(struct mddev *mddev)
723 struct r0conf *priv_conf;
727 * - (N - 1) mirror drives must be already faulty
729 if ((mddev->raid_disks - 1) != mddev->degraded) {
730 pr_err("md/raid0:%s: (N - 1) mirrors drives must be already faulty!\n",
732 return ERR_PTR(-EINVAL);
736 * a raid1 doesn't have the notion of chunk size, so
737 * figure out the largest suitable size we can use.
739 chunksect = 64 * 2; /* 64K by default */
741 /* The array must be an exact multiple of chunksize */
742 while (chunksect && (mddev->array_sectors & (chunksect - 1)))
745 if ((chunksect << 9) < PAGE_SIZE)
746 /* array size does not allow a suitable chunk size */
747 return ERR_PTR(-EINVAL);
749 /* Set new parameters */
750 mddev->new_level = 0;
751 mddev->new_layout = 0;
752 mddev->new_chunk_sectors = chunksect;
753 mddev->chunk_sectors = chunksect;
754 mddev->delta_disks = 1 - mddev->raid_disks;
755 mddev->raid_disks = 1;
756 /* make sure it will be not marked as dirty */
757 mddev->recovery_cp = MaxSector;
758 mddev_clear_unsupported_flags(mddev, UNSUPPORTED_MDDEV_FLAGS);
760 create_strip_zones(mddev, &priv_conf);
764 static void *raid0_takeover(struct mddev *mddev)
766 /* raid0 can take over:
767 * raid4 - if all data disks are active.
768 * raid5 - providing it is Raid4 layout and one disk is faulty
769 * raid10 - assuming we have all necessary active disks
770 * raid1 - with (N -1) mirror drives faulty
774 pr_warn("md/raid0: %s: cannot takeover array with bitmap\n",
776 return ERR_PTR(-EBUSY);
778 if (mddev->level == 4)
779 return raid0_takeover_raid45(mddev);
781 if (mddev->level == 5) {
782 if (mddev->layout == ALGORITHM_PARITY_N)
783 return raid0_takeover_raid45(mddev);
785 pr_warn("md/raid0:%s: Raid can only takeover Raid5 with layout: %d\n",
786 mdname(mddev), ALGORITHM_PARITY_N);
789 if (mddev->level == 10)
790 return raid0_takeover_raid10(mddev);
792 if (mddev->level == 1)
793 return raid0_takeover_raid1(mddev);
795 pr_warn("Takeover from raid%i to raid0 not supported\n",
798 return ERR_PTR(-EINVAL);
801 static void raid0_quiesce(struct mddev *mddev, int quiesce)
805 static struct md_personality raid0_personality=
809 .owner = THIS_MODULE,
810 .make_request = raid0_make_request,
813 .status = raid0_status,
815 .takeover = raid0_takeover,
816 .quiesce = raid0_quiesce,
817 .error_handler = raid0_error,
820 static int __init raid0_init (void)
822 return register_md_personality (&raid0_personality);
825 static void raid0_exit (void)
827 unregister_md_personality (&raid0_personality);
830 module_init(raid0_init);
831 module_exit(raid0_exit);
832 MODULE_LICENSE("GPL");
833 MODULE_DESCRIPTION("RAID0 (striping) personality for MD");
834 MODULE_ALIAS("md-personality-2"); /* RAID0 */
835 MODULE_ALIAS("md-raid0");
836 MODULE_ALIAS("md-level-0");