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 #define UNSUPPORTED_MDDEV_FLAGS \
23 ((1L << MD_HAS_JOURNAL) | \
24 (1L << MD_JOURNAL_CLEAN) | \
25 (1L << MD_FAILFAST_SUPPORTED) |\
26 (1L << MD_HAS_PPL) | \
27 (1L << MD_HAS_MULTIPLE_PPLS))
29 static int raid0_congested(struct mddev *mddev, int bits)
31 struct r0conf *conf = mddev->private;
32 struct md_rdev **devlist = conf->devlist;
33 int raid_disks = conf->strip_zone[0].nb_dev;
36 for (i = 0; i < raid_disks && !ret ; i++) {
37 struct request_queue *q = bdev_get_queue(devlist[i]->bdev);
39 ret |= bdi_congested(q->backing_dev_info, bits);
45 * inform the user of the raid configuration
47 static void dump_zones(struct mddev *mddev)
50 sector_t zone_size = 0;
51 sector_t zone_start = 0;
52 char b[BDEVNAME_SIZE];
53 struct r0conf *conf = mddev->private;
54 int raid_disks = conf->strip_zone[0].nb_dev;
55 pr_debug("md: RAID0 configuration for %s - %d zone%s\n",
57 conf->nr_strip_zones, conf->nr_strip_zones==1?"":"s");
58 for (j = 0; j < conf->nr_strip_zones; j++) {
62 for (k = 0; k < conf->strip_zone[j].nb_dev; k++)
63 len += snprintf(line+len, 200-len, "%s%s", k?"/":"",
64 bdevname(conf->devlist[j*raid_disks
66 pr_debug("md: zone%d=[%s]\n", j, line);
68 zone_size = conf->strip_zone[j].zone_end - zone_start;
69 pr_debug(" zone-offset=%10lluKB, device-offset=%10lluKB, size=%10lluKB\n",
70 (unsigned long long)zone_start>>1,
71 (unsigned long long)conf->strip_zone[j].dev_start>>1,
72 (unsigned long long)zone_size>>1);
73 zone_start = conf->strip_zone[j].zone_end;
77 static int create_strip_zones(struct mddev *mddev, struct r0conf **private_conf)
80 sector_t curr_zone_end, sectors;
81 struct md_rdev *smallest, *rdev1, *rdev2, *rdev, **dev;
82 struct strip_zone *zone;
84 char b[BDEVNAME_SIZE];
85 char b2[BDEVNAME_SIZE];
86 struct r0conf *conf = kzalloc(sizeof(*conf), GFP_KERNEL);
87 unsigned short blksize = 512;
89 *private_conf = ERR_PTR(-ENOMEM);
92 rdev_for_each(rdev1, mddev) {
93 pr_debug("md/raid0:%s: looking at %s\n",
95 bdevname(rdev1->bdev, b));
98 /* round size to chunk_size */
99 sectors = rdev1->sectors;
100 sector_div(sectors, mddev->chunk_sectors);
101 rdev1->sectors = sectors * mddev->chunk_sectors;
103 blksize = max(blksize, queue_logical_block_size(
104 rdev1->bdev->bd_disk->queue));
106 rdev_for_each(rdev2, mddev) {
107 pr_debug("md/raid0:%s: comparing %s(%llu)"
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",
119 if (rdev2->sectors == rdev1->sectors) {
121 * Not unique, don't count it as a new
124 pr_debug("md/raid0:%s: EQUAL\n",
129 pr_debug("md/raid0:%s: NOT EQUAL\n",
133 pr_debug("md/raid0:%s: ==> UNIQUE\n",
135 conf->nr_strip_zones++;
136 pr_debug("md/raid0:%s: %d zones\n",
137 mdname(mddev), conf->nr_strip_zones);
140 pr_debug("md/raid0:%s: FINAL %d zones\n",
141 mdname(mddev), conf->nr_strip_zones);
143 * now since we have the hard sector sizes, we can make sure
144 * chunk size is a multiple of that sector size
146 if ((mddev->chunk_sectors << 9) % blksize) {
147 pr_warn("md/raid0:%s: chunk_size of %d not multiple of block size %d\n",
149 mddev->chunk_sectors << 9, blksize);
155 conf->strip_zone = kcalloc(conf->nr_strip_zones,
156 sizeof(struct strip_zone),
158 if (!conf->strip_zone)
160 conf->devlist = kzalloc(array3_size(sizeof(struct md_rdev *),
161 conf->nr_strip_zones,
167 /* The first zone must contain all devices, so here we check that
168 * there is a proper alignment of slots to devices and find them all
170 zone = &conf->strip_zone[0];
175 rdev_for_each(rdev1, mddev) {
176 int j = rdev1->raid_disk;
178 if (mddev->level == 10) {
179 /* taking over a raid10-n2 array */
181 rdev1->new_raid_disk = j;
184 if (mddev->level == 1) {
185 /* taiking over a raid1 array-
186 * we have only one active disk
189 rdev1->new_raid_disk = j;
193 pr_warn("md/raid0:%s: remove inactive devices before converting to RAID0\n",
197 if (j >= mddev->raid_disks) {
198 pr_warn("md/raid0:%s: bad disk number %d - aborting!\n",
203 pr_warn("md/raid0:%s: multiple devices for %d - aborting!\n",
209 if (!smallest || (rdev1->sectors < smallest->sectors))
213 if (cnt != mddev->raid_disks) {
214 pr_warn("md/raid0:%s: too few disks (%d of %d) - aborting!\n",
215 mdname(mddev), cnt, mddev->raid_disks);
219 zone->zone_end = smallest->sectors * cnt;
221 curr_zone_end = zone->zone_end;
223 /* now do the other zones */
224 for (i = 1; i < conf->nr_strip_zones; i++)
228 zone = conf->strip_zone + i;
229 dev = conf->devlist + i * mddev->raid_disks;
231 pr_debug("md/raid0:%s: zone %d\n", mdname(mddev), i);
232 zone->dev_start = smallest->sectors;
236 for (j=0; j<cnt; j++) {
237 rdev = conf->devlist[j];
238 if (rdev->sectors <= zone->dev_start) {
239 pr_debug("md/raid0:%s: checking %s ... nope\n",
241 bdevname(rdev->bdev, b));
244 pr_debug("md/raid0:%s: checking %s ..."
245 " contained as device %d\n",
247 bdevname(rdev->bdev, b), c);
250 if (!smallest || rdev->sectors < smallest->sectors) {
252 pr_debug("md/raid0:%s: (%llu) is smallest!.\n",
254 (unsigned long long)rdev->sectors);
259 sectors = (smallest->sectors - zone->dev_start) * c;
260 pr_debug("md/raid0:%s: zone->nb_dev: %d, sectors: %llu\n",
262 zone->nb_dev, (unsigned long long)sectors);
264 curr_zone_end += sectors;
265 zone->zone_end = curr_zone_end;
267 pr_debug("md/raid0:%s: current zone start: %llu\n",
269 (unsigned long long)smallest->sectors);
272 pr_debug("md/raid0:%s: done.\n", mdname(mddev));
273 *private_conf = conf;
277 kfree(conf->strip_zone);
278 kfree(conf->devlist);
280 *private_conf = ERR_PTR(err);
284 /* Find the zone which holds a particular offset
285 * Update *sectorp to be an offset in that zone
287 static struct strip_zone *find_zone(struct r0conf *conf,
291 struct strip_zone *z = conf->strip_zone;
292 sector_t sector = *sectorp;
294 for (i = 0; i < conf->nr_strip_zones; i++)
295 if (sector < z[i].zone_end) {
297 *sectorp = sector - z[i-1].zone_end;
304 * remaps the bio to the target device. we separate two flows.
305 * power 2 flow and a general flow for the sake of performance
307 static struct md_rdev *map_sector(struct mddev *mddev, struct strip_zone *zone,
308 sector_t sector, sector_t *sector_offset)
310 unsigned int sect_in_chunk;
312 struct r0conf *conf = mddev->private;
313 int raid_disks = conf->strip_zone[0].nb_dev;
314 unsigned int chunk_sects = mddev->chunk_sectors;
316 if (is_power_of_2(chunk_sects)) {
317 int chunksect_bits = ffz(~chunk_sects);
318 /* find the sector offset inside the chunk */
319 sect_in_chunk = sector & (chunk_sects - 1);
320 sector >>= chunksect_bits;
322 chunk = *sector_offset;
323 /* quotient is the chunk in real device*/
324 sector_div(chunk, zone->nb_dev << chunksect_bits);
326 sect_in_chunk = sector_div(sector, chunk_sects);
327 chunk = *sector_offset;
328 sector_div(chunk, chunk_sects * zone->nb_dev);
331 * position the bio over the real device
332 * real sector = chunk in device + starting of zone
333 * + the position in the chunk
335 *sector_offset = (chunk * chunk_sects) + sect_in_chunk;
336 return conf->devlist[(zone - conf->strip_zone)*raid_disks
337 + sector_div(sector, zone->nb_dev)];
340 static sector_t raid0_size(struct mddev *mddev, sector_t sectors, int raid_disks)
342 sector_t array_sectors = 0;
343 struct md_rdev *rdev;
345 WARN_ONCE(sectors || raid_disks,
346 "%s does not support generic reshape\n", __func__);
348 rdev_for_each(rdev, mddev)
349 array_sectors += (rdev->sectors &
350 ~(sector_t)(mddev->chunk_sectors-1));
352 return array_sectors;
355 static void raid0_free(struct mddev *mddev, void *priv);
357 static int raid0_run(struct mddev *mddev)
362 if (mddev->chunk_sectors == 0) {
363 pr_warn("md/raid0:%s: chunk size must be set.\n", mdname(mddev));
366 if (md_check_no_bitmap(mddev))
369 /* if private is not null, we are here after takeover */
370 if (mddev->private == NULL) {
371 ret = create_strip_zones(mddev, &conf);
374 mddev->private = conf;
376 conf = mddev->private;
378 struct md_rdev *rdev;
379 bool discard_supported = false;
381 blk_queue_max_hw_sectors(mddev->queue, mddev->chunk_sectors);
382 blk_queue_max_write_same_sectors(mddev->queue, mddev->chunk_sectors);
383 blk_queue_max_write_zeroes_sectors(mddev->queue, mddev->chunk_sectors);
384 blk_queue_max_discard_sectors(mddev->queue, UINT_MAX);
386 blk_queue_io_min(mddev->queue, mddev->chunk_sectors << 9);
387 blk_queue_io_opt(mddev->queue,
388 (mddev->chunk_sectors << 9) * mddev->raid_disks);
390 rdev_for_each(rdev, mddev) {
391 disk_stack_limits(mddev->gendisk, rdev->bdev,
392 rdev->data_offset << 9);
393 if (blk_queue_discard(bdev_get_queue(rdev->bdev)))
394 discard_supported = true;
396 if (!discard_supported)
397 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, mddev->queue);
399 blk_queue_flag_set(QUEUE_FLAG_DISCARD, mddev->queue);
402 /* calculate array device size */
403 md_set_array_sectors(mddev, raid0_size(mddev, 0, 0));
405 pr_debug("md/raid0:%s: md_size is %llu sectors.\n",
407 (unsigned long long)mddev->array_sectors);
410 /* calculate the max read-ahead size.
411 * For read-ahead of large files to be effective, we need to
412 * readahead at least twice a whole stripe. i.e. number of devices
413 * multiplied by chunk size times 2.
414 * If an individual device has an ra_pages greater than the
415 * chunk size, then we will not drive that device as hard as it
416 * wants. We consider this a configuration error: a larger
417 * chunksize should be used in that case.
419 int stripe = mddev->raid_disks *
420 (mddev->chunk_sectors << 9) / PAGE_SIZE;
421 if (mddev->queue->backing_dev_info->ra_pages < 2* stripe)
422 mddev->queue->backing_dev_info->ra_pages = 2* stripe;
427 ret = md_integrity_register(mddev);
432 static void raid0_free(struct mddev *mddev, void *priv)
434 struct r0conf *conf = priv;
436 kfree(conf->strip_zone);
437 kfree(conf->devlist);
442 * Is io distribute over 1 or more chunks ?
444 static inline int is_io_in_chunk_boundary(struct mddev *mddev,
445 unsigned int chunk_sects, struct bio *bio)
447 if (likely(is_power_of_2(chunk_sects))) {
448 return chunk_sects >=
449 ((bio->bi_iter.bi_sector & (chunk_sects-1))
452 sector_t sector = bio->bi_iter.bi_sector;
453 return chunk_sects >= (sector_div(sector, chunk_sects)
458 static void raid0_handle_discard(struct mddev *mddev, struct bio *bio)
460 struct r0conf *conf = mddev->private;
461 struct strip_zone *zone;
462 sector_t start = bio->bi_iter.bi_sector;
464 unsigned int stripe_size;
465 sector_t first_stripe_index, last_stripe_index;
466 sector_t start_disk_offset;
467 unsigned int start_disk_index;
468 sector_t end_disk_offset;
469 unsigned int end_disk_index;
472 zone = find_zone(conf, &start);
474 if (bio_end_sector(bio) > zone->zone_end) {
475 struct bio *split = bio_split(bio,
476 zone->zone_end - bio->bi_iter.bi_sector, GFP_NOIO,
478 bio_chain(split, bio);
479 generic_make_request(bio);
481 end = zone->zone_end;
483 end = bio_end_sector(bio);
485 if (zone != conf->strip_zone)
486 end = end - zone[-1].zone_end;
488 /* Now start and end is the offset in zone */
489 stripe_size = zone->nb_dev * mddev->chunk_sectors;
491 first_stripe_index = start;
492 sector_div(first_stripe_index, stripe_size);
493 last_stripe_index = end;
494 sector_div(last_stripe_index, stripe_size);
496 start_disk_index = (int)(start - first_stripe_index * stripe_size) /
497 mddev->chunk_sectors;
498 start_disk_offset = ((int)(start - first_stripe_index * stripe_size) %
499 mddev->chunk_sectors) +
500 first_stripe_index * mddev->chunk_sectors;
501 end_disk_index = (int)(end - last_stripe_index * stripe_size) /
502 mddev->chunk_sectors;
503 end_disk_offset = ((int)(end - last_stripe_index * stripe_size) %
504 mddev->chunk_sectors) +
505 last_stripe_index * mddev->chunk_sectors;
507 for (disk = 0; disk < zone->nb_dev; disk++) {
508 sector_t dev_start, dev_end;
509 struct bio *discard_bio = NULL;
510 struct md_rdev *rdev;
512 if (disk < start_disk_index)
513 dev_start = (first_stripe_index + 1) *
514 mddev->chunk_sectors;
515 else if (disk > start_disk_index)
516 dev_start = first_stripe_index * mddev->chunk_sectors;
518 dev_start = start_disk_offset;
520 if (disk < end_disk_index)
521 dev_end = (last_stripe_index + 1) * mddev->chunk_sectors;
522 else if (disk > end_disk_index)
523 dev_end = last_stripe_index * mddev->chunk_sectors;
525 dev_end = end_disk_offset;
527 if (dev_end <= dev_start)
530 rdev = conf->devlist[(zone - conf->strip_zone) *
531 conf->strip_zone[0].nb_dev + disk];
532 if (__blkdev_issue_discard(rdev->bdev,
533 dev_start + zone->dev_start + rdev->data_offset,
534 dev_end - dev_start, GFP_NOIO, 0, &discard_bio) ||
537 bio_chain(discard_bio, bio);
538 bio_clone_blkg_association(discard_bio, bio);
540 trace_block_bio_remap(bdev_get_queue(rdev->bdev),
541 discard_bio, disk_devt(mddev->gendisk),
542 bio->bi_iter.bi_sector);
543 generic_make_request(discard_bio);
548 static bool raid0_make_request(struct mddev *mddev, struct bio *bio)
550 struct strip_zone *zone;
551 struct md_rdev *tmp_dev;
554 unsigned chunk_sects;
557 if (unlikely(bio->bi_opf & REQ_PREFLUSH)) {
558 md_flush_request(mddev, bio);
562 if (unlikely((bio_op(bio) == REQ_OP_DISCARD))) {
563 raid0_handle_discard(mddev, bio);
567 bio_sector = bio->bi_iter.bi_sector;
569 chunk_sects = mddev->chunk_sectors;
571 sectors = chunk_sects -
572 (likely(is_power_of_2(chunk_sects))
573 ? (sector & (chunk_sects-1))
574 : sector_div(sector, chunk_sects));
576 /* Restore due to sector_div */
579 if (sectors < bio_sectors(bio)) {
580 struct bio *split = bio_split(bio, sectors, GFP_NOIO,
582 bio_chain(split, bio);
583 generic_make_request(bio);
587 zone = find_zone(mddev->private, §or);
588 tmp_dev = map_sector(mddev, zone, sector, §or);
589 bio_set_dev(bio, tmp_dev->bdev);
590 bio->bi_iter.bi_sector = sector + zone->dev_start +
591 tmp_dev->data_offset;
594 trace_block_bio_remap(bio->bi_disk->queue, bio,
595 disk_devt(mddev->gendisk), bio_sector);
596 mddev_check_writesame(mddev, bio);
597 mddev_check_write_zeroes(mddev, bio);
598 generic_make_request(bio);
602 static void raid0_status(struct seq_file *seq, struct mddev *mddev)
604 seq_printf(seq, " %dk chunks", mddev->chunk_sectors / 2);
608 static void *raid0_takeover_raid45(struct mddev *mddev)
610 struct md_rdev *rdev;
611 struct r0conf *priv_conf;
613 if (mddev->degraded != 1) {
614 pr_warn("md/raid0:%s: raid5 must be degraded! Degraded disks: %d\n",
617 return ERR_PTR(-EINVAL);
620 rdev_for_each(rdev, mddev) {
621 /* check slot number for a disk */
622 if (rdev->raid_disk == mddev->raid_disks-1) {
623 pr_warn("md/raid0:%s: raid5 must have missing parity disk!\n",
625 return ERR_PTR(-EINVAL);
627 rdev->sectors = mddev->dev_sectors;
630 /* Set new parameters */
631 mddev->new_level = 0;
632 mddev->new_layout = 0;
633 mddev->new_chunk_sectors = mddev->chunk_sectors;
635 mddev->delta_disks = -1;
636 /* make sure it will be not marked as dirty */
637 mddev->recovery_cp = MaxSector;
638 mddev_clear_unsupported_flags(mddev, UNSUPPORTED_MDDEV_FLAGS);
640 create_strip_zones(mddev, &priv_conf);
645 static void *raid0_takeover_raid10(struct mddev *mddev)
647 struct r0conf *priv_conf;
650 * - far_copies must be 1
651 * - near_copies must be 2
652 * - disks number must be even
653 * - all mirrors must be already degraded
655 if (mddev->layout != ((1 << 8) + 2)) {
656 pr_warn("md/raid0:%s:: Raid0 cannot takeover layout: 0x%x\n",
659 return ERR_PTR(-EINVAL);
661 if (mddev->raid_disks & 1) {
662 pr_warn("md/raid0:%s: Raid0 cannot takeover Raid10 with odd disk number.\n",
664 return ERR_PTR(-EINVAL);
666 if (mddev->degraded != (mddev->raid_disks>>1)) {
667 pr_warn("md/raid0:%s: All mirrors must be already degraded!\n",
669 return ERR_PTR(-EINVAL);
672 /* Set new parameters */
673 mddev->new_level = 0;
674 mddev->new_layout = 0;
675 mddev->new_chunk_sectors = mddev->chunk_sectors;
676 mddev->delta_disks = - mddev->raid_disks / 2;
677 mddev->raid_disks += mddev->delta_disks;
679 /* make sure it will be not marked as dirty */
680 mddev->recovery_cp = MaxSector;
681 mddev_clear_unsupported_flags(mddev, UNSUPPORTED_MDDEV_FLAGS);
683 create_strip_zones(mddev, &priv_conf);
687 static void *raid0_takeover_raid1(struct mddev *mddev)
689 struct r0conf *priv_conf;
693 * - (N - 1) mirror drives must be already faulty
695 if ((mddev->raid_disks - 1) != mddev->degraded) {
696 pr_err("md/raid0:%s: (N - 1) mirrors drives must be already faulty!\n",
698 return ERR_PTR(-EINVAL);
702 * a raid1 doesn't have the notion of chunk size, so
703 * figure out the largest suitable size we can use.
705 chunksect = 64 * 2; /* 64K by default */
707 /* The array must be an exact multiple of chunksize */
708 while (chunksect && (mddev->array_sectors & (chunksect - 1)))
711 if ((chunksect << 9) < PAGE_SIZE)
712 /* array size does not allow a suitable chunk size */
713 return ERR_PTR(-EINVAL);
715 /* Set new parameters */
716 mddev->new_level = 0;
717 mddev->new_layout = 0;
718 mddev->new_chunk_sectors = chunksect;
719 mddev->chunk_sectors = chunksect;
720 mddev->delta_disks = 1 - mddev->raid_disks;
721 mddev->raid_disks = 1;
722 /* make sure it will be not marked as dirty */
723 mddev->recovery_cp = MaxSector;
724 mddev_clear_unsupported_flags(mddev, UNSUPPORTED_MDDEV_FLAGS);
726 create_strip_zones(mddev, &priv_conf);
730 static void *raid0_takeover(struct mddev *mddev)
732 /* raid0 can take over:
733 * raid4 - if all data disks are active.
734 * raid5 - providing it is Raid4 layout and one disk is faulty
735 * raid10 - assuming we have all necessary active disks
736 * raid1 - with (N -1) mirror drives faulty
740 pr_warn("md/raid0: %s: cannot takeover array with bitmap\n",
742 return ERR_PTR(-EBUSY);
744 if (mddev->level == 4)
745 return raid0_takeover_raid45(mddev);
747 if (mddev->level == 5) {
748 if (mddev->layout == ALGORITHM_PARITY_N)
749 return raid0_takeover_raid45(mddev);
751 pr_warn("md/raid0:%s: Raid can only takeover Raid5 with layout: %d\n",
752 mdname(mddev), ALGORITHM_PARITY_N);
755 if (mddev->level == 10)
756 return raid0_takeover_raid10(mddev);
758 if (mddev->level == 1)
759 return raid0_takeover_raid1(mddev);
761 pr_warn("Takeover from raid%i to raid0 not supported\n",
764 return ERR_PTR(-EINVAL);
767 static void raid0_quiesce(struct mddev *mddev, int quiesce)
771 static struct md_personality raid0_personality=
775 .owner = THIS_MODULE,
776 .make_request = raid0_make_request,
779 .status = raid0_status,
781 .takeover = raid0_takeover,
782 .quiesce = raid0_quiesce,
783 .congested = raid0_congested,
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");