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 char b[BDEVNAME_SIZE];
41 struct r0conf *conf = mddev->private;
42 int raid_disks = conf->strip_zone[0].nb_dev;
43 pr_debug("md: RAID0 configuration for %s - %d zone%s\n",
45 conf->nr_strip_zones, conf->nr_strip_zones==1?"":"s");
46 for (j = 0; j < conf->nr_strip_zones; j++) {
50 for (k = 0; k < conf->strip_zone[j].nb_dev; k++)
51 len += scnprintf(line+len, 200-len, "%s%s", k?"/":"",
52 bdevname(conf->devlist[j*raid_disks
54 pr_debug("md: zone%d=[%s]\n", j, line);
56 zone_size = conf->strip_zone[j].zone_end - zone_start;
57 pr_debug(" zone-offset=%10lluKB, device-offset=%10lluKB, size=%10lluKB\n",
58 (unsigned long long)zone_start>>1,
59 (unsigned long long)conf->strip_zone[j].dev_start>>1,
60 (unsigned long long)zone_size>>1);
61 zone_start = conf->strip_zone[j].zone_end;
65 static int create_strip_zones(struct mddev *mddev, struct r0conf **private_conf)
68 sector_t curr_zone_end, sectors;
69 struct md_rdev *smallest, *rdev1, *rdev2, *rdev, **dev;
70 struct strip_zone *zone;
72 char b[BDEVNAME_SIZE];
73 char b2[BDEVNAME_SIZE];
74 struct r0conf *conf = kzalloc(sizeof(*conf), GFP_KERNEL);
75 unsigned blksize = 512;
77 *private_conf = ERR_PTR(-ENOMEM);
80 rdev_for_each(rdev1, mddev) {
81 pr_debug("md/raid0:%s: looking at %s\n",
83 bdevname(rdev1->bdev, b));
86 /* round size to chunk_size */
87 sectors = rdev1->sectors;
88 sector_div(sectors, mddev->chunk_sectors);
89 rdev1->sectors = sectors * mddev->chunk_sectors;
91 blksize = max(blksize, queue_logical_block_size(
92 rdev1->bdev->bd_disk->queue));
94 rdev_for_each(rdev2, mddev) {
95 pr_debug("md/raid0:%s: comparing %s(%llu)"
98 bdevname(rdev1->bdev,b),
99 (unsigned long long)rdev1->sectors,
100 bdevname(rdev2->bdev,b2),
101 (unsigned long long)rdev2->sectors);
102 if (rdev2 == rdev1) {
103 pr_debug("md/raid0:%s: END\n",
107 if (rdev2->sectors == rdev1->sectors) {
109 * Not unique, don't count it as a new
112 pr_debug("md/raid0:%s: EQUAL\n",
117 pr_debug("md/raid0:%s: NOT EQUAL\n",
121 pr_debug("md/raid0:%s: ==> UNIQUE\n",
123 conf->nr_strip_zones++;
124 pr_debug("md/raid0:%s: %d zones\n",
125 mdname(mddev), conf->nr_strip_zones);
128 pr_debug("md/raid0:%s: FINAL %d zones\n",
129 mdname(mddev), conf->nr_strip_zones);
132 * now since we have the hard sector sizes, we can make sure
133 * chunk size is a multiple of that sector size
135 if ((mddev->chunk_sectors << 9) % blksize) {
136 pr_warn("md/raid0:%s: chunk_size of %d not multiple of block size %d\n",
138 mddev->chunk_sectors << 9, blksize);
144 conf->strip_zone = kcalloc(conf->nr_strip_zones,
145 sizeof(struct strip_zone),
147 if (!conf->strip_zone)
149 conf->devlist = kzalloc(array3_size(sizeof(struct md_rdev *),
150 conf->nr_strip_zones,
156 /* The first zone must contain all devices, so here we check that
157 * there is a proper alignment of slots to devices and find them all
159 zone = &conf->strip_zone[0];
164 rdev_for_each(rdev1, mddev) {
165 int j = rdev1->raid_disk;
167 if (mddev->level == 10) {
168 /* taking over a raid10-n2 array */
170 rdev1->new_raid_disk = j;
173 if (mddev->level == 1) {
174 /* taiking over a raid1 array-
175 * we have only one active disk
178 rdev1->new_raid_disk = j;
182 pr_warn("md/raid0:%s: remove inactive devices before converting to RAID0\n",
186 if (j >= mddev->raid_disks) {
187 pr_warn("md/raid0:%s: bad disk number %d - aborting!\n",
192 pr_warn("md/raid0:%s: multiple devices for %d - aborting!\n",
198 if (!smallest || (rdev1->sectors < smallest->sectors))
202 if (cnt != mddev->raid_disks) {
203 pr_warn("md/raid0:%s: too few disks (%d of %d) - aborting!\n",
204 mdname(mddev), cnt, mddev->raid_disks);
208 zone->zone_end = smallest->sectors * cnt;
210 curr_zone_end = zone->zone_end;
212 /* now do the other zones */
213 for (i = 1; i < conf->nr_strip_zones; i++)
217 zone = conf->strip_zone + i;
218 dev = conf->devlist + i * mddev->raid_disks;
220 pr_debug("md/raid0:%s: zone %d\n", mdname(mddev), i);
221 zone->dev_start = smallest->sectors;
225 for (j=0; j<cnt; j++) {
226 rdev = conf->devlist[j];
227 if (rdev->sectors <= zone->dev_start) {
228 pr_debug("md/raid0:%s: checking %s ... nope\n",
230 bdevname(rdev->bdev, b));
233 pr_debug("md/raid0:%s: checking %s ..."
234 " contained as device %d\n",
236 bdevname(rdev->bdev, b), c);
239 if (!smallest || rdev->sectors < smallest->sectors) {
241 pr_debug("md/raid0:%s: (%llu) is smallest!.\n",
243 (unsigned long long)rdev->sectors);
248 sectors = (smallest->sectors - zone->dev_start) * c;
249 pr_debug("md/raid0:%s: zone->nb_dev: %d, sectors: %llu\n",
251 zone->nb_dev, (unsigned long long)sectors);
253 curr_zone_end += sectors;
254 zone->zone_end = curr_zone_end;
256 pr_debug("md/raid0:%s: current zone start: %llu\n",
258 (unsigned long long)smallest->sectors);
261 if (conf->nr_strip_zones == 1 || conf->strip_zone[1].nb_dev == 1) {
262 conf->layout = RAID0_ORIG_LAYOUT;
263 } else if (mddev->layout == RAID0_ORIG_LAYOUT ||
264 mddev->layout == RAID0_ALT_MULTIZONE_LAYOUT) {
265 conf->layout = mddev->layout;
266 } else if (default_layout == RAID0_ORIG_LAYOUT ||
267 default_layout == RAID0_ALT_MULTIZONE_LAYOUT) {
268 conf->layout = default_layout;
270 pr_err("md/raid0:%s: cannot assemble multi-zone RAID0 with default_layout setting\n",
272 pr_err("md/raid0: please set raid0.default_layout to 1 or 2\n");
277 pr_debug("md/raid0:%s: done.\n", mdname(mddev));
278 *private_conf = conf;
282 kfree(conf->strip_zone);
283 kfree(conf->devlist);
285 *private_conf = ERR_PTR(err);
289 /* Find the zone which holds a particular offset
290 * Update *sectorp to be an offset in that zone
292 static struct strip_zone *find_zone(struct r0conf *conf,
296 struct strip_zone *z = conf->strip_zone;
297 sector_t sector = *sectorp;
299 for (i = 0; i < conf->nr_strip_zones; i++)
300 if (sector < z[i].zone_end) {
302 *sectorp = sector - z[i-1].zone_end;
309 * remaps the bio to the target device. we separate two flows.
310 * power 2 flow and a general flow for the sake of performance
312 static struct md_rdev *map_sector(struct mddev *mddev, struct strip_zone *zone,
313 sector_t sector, sector_t *sector_offset)
315 unsigned int sect_in_chunk;
317 struct r0conf *conf = mddev->private;
318 int raid_disks = conf->strip_zone[0].nb_dev;
319 unsigned int chunk_sects = mddev->chunk_sectors;
321 if (is_power_of_2(chunk_sects)) {
322 int chunksect_bits = ffz(~chunk_sects);
323 /* find the sector offset inside the chunk */
324 sect_in_chunk = sector & (chunk_sects - 1);
325 sector >>= chunksect_bits;
327 chunk = *sector_offset;
328 /* quotient is the chunk in real device*/
329 sector_div(chunk, zone->nb_dev << chunksect_bits);
331 sect_in_chunk = sector_div(sector, chunk_sects);
332 chunk = *sector_offset;
333 sector_div(chunk, chunk_sects * zone->nb_dev);
336 * position the bio over the real device
337 * real sector = chunk in device + starting of zone
338 * + the position in the chunk
340 *sector_offset = (chunk * chunk_sects) + sect_in_chunk;
341 return conf->devlist[(zone - conf->strip_zone)*raid_disks
342 + sector_div(sector, zone->nb_dev)];
345 static sector_t raid0_size(struct mddev *mddev, sector_t sectors, int raid_disks)
347 sector_t array_sectors = 0;
348 struct md_rdev *rdev;
350 WARN_ONCE(sectors || raid_disks,
351 "%s does not support generic reshape\n", __func__);
353 rdev_for_each(rdev, mddev)
354 array_sectors += (rdev->sectors &
355 ~(sector_t)(mddev->chunk_sectors-1));
357 return array_sectors;
360 static void free_conf(struct mddev *mddev, struct r0conf *conf)
362 kfree(conf->strip_zone);
363 kfree(conf->devlist);
367 static void raid0_free(struct mddev *mddev, void *priv)
369 struct r0conf *conf = priv;
371 free_conf(mddev, conf);
372 acct_bioset_exit(mddev);
375 static int raid0_run(struct mddev *mddev)
380 if (mddev->chunk_sectors == 0) {
381 pr_warn("md/raid0:%s: chunk size must be set.\n", mdname(mddev));
384 if (md_check_no_bitmap(mddev))
387 if (acct_bioset_init(mddev)) {
388 pr_err("md/raid0:%s: alloc acct bioset failed.\n", mdname(mddev));
392 /* if private is not null, we are here after takeover */
393 if (mddev->private == NULL) {
394 ret = create_strip_zones(mddev, &conf);
397 mddev->private = conf;
399 conf = mddev->private;
401 struct md_rdev *rdev;
402 bool discard_supported = false;
404 blk_queue_max_hw_sectors(mddev->queue, mddev->chunk_sectors);
405 blk_queue_max_write_same_sectors(mddev->queue, mddev->chunk_sectors);
406 blk_queue_max_write_zeroes_sectors(mddev->queue, mddev->chunk_sectors);
407 blk_queue_max_discard_sectors(mddev->queue, UINT_MAX);
409 blk_queue_io_min(mddev->queue, mddev->chunk_sectors << 9);
410 blk_queue_io_opt(mddev->queue,
411 (mddev->chunk_sectors << 9) * mddev->raid_disks);
413 rdev_for_each(rdev, mddev) {
414 disk_stack_limits(mddev->gendisk, rdev->bdev,
415 rdev->data_offset << 9);
416 if (blk_queue_discard(bdev_get_queue(rdev->bdev)))
417 discard_supported = true;
419 if (!discard_supported)
420 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, mddev->queue);
422 blk_queue_flag_set(QUEUE_FLAG_DISCARD, mddev->queue);
425 /* calculate array device size */
426 md_set_array_sectors(mddev, raid0_size(mddev, 0, 0));
428 pr_debug("md/raid0:%s: md_size is %llu sectors.\n",
430 (unsigned long long)mddev->array_sectors);
434 ret = md_integrity_register(mddev);
441 free_conf(mddev, conf);
443 acct_bioset_exit(mddev);
447 static void raid0_handle_discard(struct mddev *mddev, struct bio *bio)
449 struct r0conf *conf = mddev->private;
450 struct strip_zone *zone;
451 sector_t start = bio->bi_iter.bi_sector;
453 unsigned int stripe_size;
454 sector_t first_stripe_index, last_stripe_index;
455 sector_t start_disk_offset;
456 unsigned int start_disk_index;
457 sector_t end_disk_offset;
458 unsigned int end_disk_index;
461 zone = find_zone(conf, &start);
463 if (bio_end_sector(bio) > zone->zone_end) {
464 struct bio *split = bio_split(bio,
465 zone->zone_end - bio->bi_iter.bi_sector, GFP_NOIO,
467 bio_chain(split, bio);
468 submit_bio_noacct(bio);
470 end = zone->zone_end;
472 end = bio_end_sector(bio);
474 if (zone != conf->strip_zone)
475 end = end - zone[-1].zone_end;
477 /* Now start and end is the offset in zone */
478 stripe_size = zone->nb_dev * mddev->chunk_sectors;
480 first_stripe_index = start;
481 sector_div(first_stripe_index, stripe_size);
482 last_stripe_index = end;
483 sector_div(last_stripe_index, stripe_size);
485 start_disk_index = (int)(start - first_stripe_index * stripe_size) /
486 mddev->chunk_sectors;
487 start_disk_offset = ((int)(start - first_stripe_index * stripe_size) %
488 mddev->chunk_sectors) +
489 first_stripe_index * mddev->chunk_sectors;
490 end_disk_index = (int)(end - last_stripe_index * stripe_size) /
491 mddev->chunk_sectors;
492 end_disk_offset = ((int)(end - last_stripe_index * stripe_size) %
493 mddev->chunk_sectors) +
494 last_stripe_index * mddev->chunk_sectors;
496 for (disk = 0; disk < zone->nb_dev; disk++) {
497 sector_t dev_start, dev_end;
498 struct md_rdev *rdev;
500 if (disk < start_disk_index)
501 dev_start = (first_stripe_index + 1) *
502 mddev->chunk_sectors;
503 else if (disk > start_disk_index)
504 dev_start = first_stripe_index * mddev->chunk_sectors;
506 dev_start = start_disk_offset;
508 if (disk < end_disk_index)
509 dev_end = (last_stripe_index + 1) * mddev->chunk_sectors;
510 else if (disk > end_disk_index)
511 dev_end = last_stripe_index * mddev->chunk_sectors;
513 dev_end = end_disk_offset;
515 if (dev_end <= dev_start)
518 rdev = conf->devlist[(zone - conf->strip_zone) *
519 conf->strip_zone[0].nb_dev + disk];
520 md_submit_discard_bio(mddev, rdev, bio,
521 dev_start + zone->dev_start + rdev->data_offset,
522 dev_end - dev_start);
527 static bool raid0_make_request(struct mddev *mddev, struct bio *bio)
529 struct r0conf *conf = mddev->private;
530 struct strip_zone *zone;
531 struct md_rdev *tmp_dev;
534 sector_t orig_sector;
535 unsigned chunk_sects;
538 if (unlikely(bio->bi_opf & REQ_PREFLUSH)
539 && md_flush_request(mddev, bio))
542 if (unlikely((bio_op(bio) == REQ_OP_DISCARD))) {
543 raid0_handle_discard(mddev, bio);
547 bio_sector = bio->bi_iter.bi_sector;
549 chunk_sects = mddev->chunk_sectors;
551 sectors = chunk_sects -
552 (likely(is_power_of_2(chunk_sects))
553 ? (sector & (chunk_sects-1))
554 : sector_div(sector, chunk_sects));
556 /* Restore due to sector_div */
559 if (sectors < bio_sectors(bio)) {
560 struct bio *split = bio_split(bio, sectors, GFP_NOIO,
562 bio_chain(split, bio);
563 submit_bio_noacct(bio);
567 if (bio->bi_pool != &mddev->bio_set)
568 md_account_bio(mddev, &bio);
570 orig_sector = sector;
571 zone = find_zone(mddev->private, §or);
572 switch (conf->layout) {
573 case RAID0_ORIG_LAYOUT:
574 tmp_dev = map_sector(mddev, zone, orig_sector, §or);
576 case RAID0_ALT_MULTIZONE_LAYOUT:
577 tmp_dev = map_sector(mddev, zone, sector, §or);
580 WARN(1, "md/raid0:%s: Invalid layout\n", mdname(mddev));
585 if (unlikely(is_mddev_broken(tmp_dev, "raid0"))) {
590 bio_set_dev(bio, tmp_dev->bdev);
591 bio->bi_iter.bi_sector = sector + zone->dev_start +
592 tmp_dev->data_offset;
595 trace_block_bio_remap(bio, disk_devt(mddev->gendisk),
597 mddev_check_writesame(mddev, bio);
598 mddev_check_write_zeroes(mddev, bio);
599 submit_bio_noacct(bio);
603 static void raid0_status(struct seq_file *seq, struct mddev *mddev)
605 seq_printf(seq, " %dk chunks", mddev->chunk_sectors / 2);
609 static void *raid0_takeover_raid45(struct mddev *mddev)
611 struct md_rdev *rdev;
612 struct r0conf *priv_conf;
614 if (mddev->degraded != 1) {
615 pr_warn("md/raid0:%s: raid5 must be degraded! Degraded disks: %d\n",
618 return ERR_PTR(-EINVAL);
621 rdev_for_each(rdev, mddev) {
622 /* check slot number for a disk */
623 if (rdev->raid_disk == mddev->raid_disks-1) {
624 pr_warn("md/raid0:%s: raid5 must have missing parity disk!\n",
626 return ERR_PTR(-EINVAL);
628 rdev->sectors = mddev->dev_sectors;
631 /* Set new parameters */
632 mddev->new_level = 0;
633 mddev->new_layout = 0;
634 mddev->new_chunk_sectors = mddev->chunk_sectors;
636 mddev->delta_disks = -1;
637 /* make sure it will be not marked as dirty */
638 mddev->recovery_cp = MaxSector;
639 mddev_clear_unsupported_flags(mddev, UNSUPPORTED_MDDEV_FLAGS);
641 create_strip_zones(mddev, &priv_conf);
646 static void *raid0_takeover_raid10(struct mddev *mddev)
648 struct r0conf *priv_conf;
651 * - far_copies must be 1
652 * - near_copies must be 2
653 * - disks number must be even
654 * - all mirrors must be already degraded
656 if (mddev->layout != ((1 << 8) + 2)) {
657 pr_warn("md/raid0:%s:: Raid0 cannot takeover layout: 0x%x\n",
660 return ERR_PTR(-EINVAL);
662 if (mddev->raid_disks & 1) {
663 pr_warn("md/raid0:%s: Raid0 cannot takeover Raid10 with odd disk number.\n",
665 return ERR_PTR(-EINVAL);
667 if (mddev->degraded != (mddev->raid_disks>>1)) {
668 pr_warn("md/raid0:%s: All mirrors must be already degraded!\n",
670 return ERR_PTR(-EINVAL);
673 /* Set new parameters */
674 mddev->new_level = 0;
675 mddev->new_layout = 0;
676 mddev->new_chunk_sectors = mddev->chunk_sectors;
677 mddev->delta_disks = - mddev->raid_disks / 2;
678 mddev->raid_disks += mddev->delta_disks;
680 /* make sure it will be not marked as dirty */
681 mddev->recovery_cp = MaxSector;
682 mddev_clear_unsupported_flags(mddev, UNSUPPORTED_MDDEV_FLAGS);
684 create_strip_zones(mddev, &priv_conf);
688 static void *raid0_takeover_raid1(struct mddev *mddev)
690 struct r0conf *priv_conf;
694 * - (N - 1) mirror drives must be already faulty
696 if ((mddev->raid_disks - 1) != mddev->degraded) {
697 pr_err("md/raid0:%s: (N - 1) mirrors drives must be already faulty!\n",
699 return ERR_PTR(-EINVAL);
703 * a raid1 doesn't have the notion of chunk size, so
704 * figure out the largest suitable size we can use.
706 chunksect = 64 * 2; /* 64K by default */
708 /* The array must be an exact multiple of chunksize */
709 while (chunksect && (mddev->array_sectors & (chunksect - 1)))
712 if ((chunksect << 9) < PAGE_SIZE)
713 /* array size does not allow a suitable chunk size */
714 return ERR_PTR(-EINVAL);
716 /* Set new parameters */
717 mddev->new_level = 0;
718 mddev->new_layout = 0;
719 mddev->new_chunk_sectors = chunksect;
720 mddev->chunk_sectors = chunksect;
721 mddev->delta_disks = 1 - mddev->raid_disks;
722 mddev->raid_disks = 1;
723 /* make sure it will be not marked as dirty */
724 mddev->recovery_cp = MaxSector;
725 mddev_clear_unsupported_flags(mddev, UNSUPPORTED_MDDEV_FLAGS);
727 create_strip_zones(mddev, &priv_conf);
731 static void *raid0_takeover(struct mddev *mddev)
733 /* raid0 can take over:
734 * raid4 - if all data disks are active.
735 * raid5 - providing it is Raid4 layout and one disk is faulty
736 * raid10 - assuming we have all necessary active disks
737 * raid1 - with (N -1) mirror drives faulty
741 pr_warn("md/raid0: %s: cannot takeover array with bitmap\n",
743 return ERR_PTR(-EBUSY);
745 if (mddev->level == 4)
746 return raid0_takeover_raid45(mddev);
748 if (mddev->level == 5) {
749 if (mddev->layout == ALGORITHM_PARITY_N)
750 return raid0_takeover_raid45(mddev);
752 pr_warn("md/raid0:%s: Raid can only takeover Raid5 with layout: %d\n",
753 mdname(mddev), ALGORITHM_PARITY_N);
756 if (mddev->level == 10)
757 return raid0_takeover_raid10(mddev);
759 if (mddev->level == 1)
760 return raid0_takeover_raid1(mddev);
762 pr_warn("Takeover from raid%i to raid0 not supported\n",
765 return ERR_PTR(-EINVAL);
768 static void raid0_quiesce(struct mddev *mddev, int quiesce)
772 static struct md_personality raid0_personality=
776 .owner = THIS_MODULE,
777 .make_request = raid0_make_request,
780 .status = raid0_status,
782 .takeover = raid0_takeover,
783 .quiesce = raid0_quiesce,
786 static int __init raid0_init (void)
788 return register_md_personality (&raid0_personality);
791 static void raid0_exit (void)
793 unregister_md_personality (&raid0_personality);
796 module_init(raid0_init);
797 module_exit(raid0_exit);
798 MODULE_LICENSE("GPL");
799 MODULE_DESCRIPTION("RAID0 (striping) personality for MD");
800 MODULE_ALIAS("md-personality-2"); /* RAID0 */
801 MODULE_ALIAS("md-raid0");
802 MODULE_ALIAS("md-level-0");