2 raid0.c : Multiple Devices driver for Linux
3 Copyright (C) 1994-96 Marc ZYNGIER
4 <zyngier@ufr-info-p7.ibp.fr> or
6 Copyright (C) 1999, 2000 Ingo Molnar, Red Hat
8 RAID-0 management functions.
10 This program is free software; you can redistribute it and/or modify
11 it under the terms of the GNU General Public License as published by
12 the Free Software Foundation; either version 2, or (at your option)
15 You should have received a copy of the GNU General Public License
16 (for example /usr/src/linux/COPYING); if not, write to the Free
17 Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
20 #include <linux/blkdev.h>
21 #include <linux/seq_file.h>
22 #include <linux/module.h>
23 #include <linux/slab.h>
24 #include <trace/events/block.h>
29 #define UNSUPPORTED_MDDEV_FLAGS \
30 ((1L << MD_HAS_JOURNAL) | \
31 (1L << MD_JOURNAL_CLEAN) | \
32 (1L << MD_FAILFAST_SUPPORTED))
34 static int raid0_congested(struct mddev *mddev, int bits)
36 struct r0conf *conf = mddev->private;
37 struct md_rdev **devlist = conf->devlist;
38 int raid_disks = conf->strip_zone[0].nb_dev;
41 for (i = 0; i < raid_disks && !ret ; i++) {
42 struct request_queue *q = bdev_get_queue(devlist[i]->bdev);
44 ret |= bdi_congested(q->backing_dev_info, bits);
50 * inform the user of the raid configuration
52 static void dump_zones(struct mddev *mddev)
55 sector_t zone_size = 0;
56 sector_t zone_start = 0;
57 char b[BDEVNAME_SIZE];
58 struct r0conf *conf = mddev->private;
59 int raid_disks = conf->strip_zone[0].nb_dev;
60 pr_debug("md: RAID0 configuration for %s - %d zone%s\n",
62 conf->nr_strip_zones, conf->nr_strip_zones==1?"":"s");
63 for (j = 0; j < conf->nr_strip_zones; j++) {
67 for (k = 0; k < conf->strip_zone[j].nb_dev; k++)
68 len += snprintf(line+len, 200-len, "%s%s", k?"/":"",
69 bdevname(conf->devlist[j*raid_disks
71 pr_debug("md: zone%d=[%s]\n", j, line);
73 zone_size = conf->strip_zone[j].zone_end - zone_start;
74 pr_debug(" zone-offset=%10lluKB, device-offset=%10lluKB, size=%10lluKB\n",
75 (unsigned long long)zone_start>>1,
76 (unsigned long long)conf->strip_zone[j].dev_start>>1,
77 (unsigned long long)zone_size>>1);
78 zone_start = conf->strip_zone[j].zone_end;
82 static int create_strip_zones(struct mddev *mddev, struct r0conf **private_conf)
85 sector_t curr_zone_end, sectors;
86 struct md_rdev *smallest, *rdev1, *rdev2, *rdev, **dev;
87 struct strip_zone *zone;
89 char b[BDEVNAME_SIZE];
90 char b2[BDEVNAME_SIZE];
91 struct r0conf *conf = kzalloc(sizeof(*conf), GFP_KERNEL);
92 unsigned short blksize = 512;
94 *private_conf = ERR_PTR(-ENOMEM);
97 rdev_for_each(rdev1, mddev) {
98 pr_debug("md/raid0:%s: looking at %s\n",
100 bdevname(rdev1->bdev, b));
103 /* round size to chunk_size */
104 sectors = rdev1->sectors;
105 sector_div(sectors, mddev->chunk_sectors);
106 rdev1->sectors = sectors * mddev->chunk_sectors;
108 blksize = max(blksize, queue_logical_block_size(
109 rdev1->bdev->bd_disk->queue));
111 rdev_for_each(rdev2, mddev) {
112 pr_debug("md/raid0:%s: comparing %s(%llu)"
115 bdevname(rdev1->bdev,b),
116 (unsigned long long)rdev1->sectors,
117 bdevname(rdev2->bdev,b2),
118 (unsigned long long)rdev2->sectors);
119 if (rdev2 == rdev1) {
120 pr_debug("md/raid0:%s: END\n",
124 if (rdev2->sectors == rdev1->sectors) {
126 * Not unique, don't count it as a new
129 pr_debug("md/raid0:%s: EQUAL\n",
134 pr_debug("md/raid0:%s: NOT EQUAL\n",
138 pr_debug("md/raid0:%s: ==> UNIQUE\n",
140 conf->nr_strip_zones++;
141 pr_debug("md/raid0:%s: %d zones\n",
142 mdname(mddev), conf->nr_strip_zones);
145 pr_debug("md/raid0:%s: FINAL %d zones\n",
146 mdname(mddev), conf->nr_strip_zones);
148 * now since we have the hard sector sizes, we can make sure
149 * chunk size is a multiple of that sector size
151 if ((mddev->chunk_sectors << 9) % blksize) {
152 pr_warn("md/raid0:%s: chunk_size of %d not multiple of block size %d\n",
154 mddev->chunk_sectors << 9, blksize);
160 conf->strip_zone = kzalloc(sizeof(struct strip_zone)*
161 conf->nr_strip_zones, GFP_KERNEL);
162 if (!conf->strip_zone)
164 conf->devlist = kzalloc(sizeof(struct md_rdev*)*
165 conf->nr_strip_zones*mddev->raid_disks,
170 /* The first zone must contain all devices, so here we check that
171 * there is a proper alignment of slots to devices and find them all
173 zone = &conf->strip_zone[0];
178 rdev_for_each(rdev1, mddev) {
179 int j = rdev1->raid_disk;
181 if (mddev->level == 10) {
182 /* taking over a raid10-n2 array */
184 rdev1->new_raid_disk = j;
187 if (mddev->level == 1) {
188 /* taiking over a raid1 array-
189 * we have only one active disk
192 rdev1->new_raid_disk = j;
196 pr_warn("md/raid0:%s: remove inactive devices before converting to RAID0\n",
200 if (j >= mddev->raid_disks) {
201 pr_warn("md/raid0:%s: bad disk number %d - aborting!\n",
206 pr_warn("md/raid0:%s: multiple devices for %d - aborting!\n",
212 if (!smallest || (rdev1->sectors < smallest->sectors))
216 if (cnt != mddev->raid_disks) {
217 pr_warn("md/raid0:%s: too few disks (%d of %d) - aborting!\n",
218 mdname(mddev), cnt, mddev->raid_disks);
222 zone->zone_end = smallest->sectors * cnt;
224 curr_zone_end = zone->zone_end;
226 /* now do the other zones */
227 for (i = 1; i < conf->nr_strip_zones; i++)
231 zone = conf->strip_zone + i;
232 dev = conf->devlist + i * mddev->raid_disks;
234 pr_debug("md/raid0:%s: zone %d\n", mdname(mddev), i);
235 zone->dev_start = smallest->sectors;
239 for (j=0; j<cnt; j++) {
240 rdev = conf->devlist[j];
241 if (rdev->sectors <= zone->dev_start) {
242 pr_debug("md/raid0:%s: checking %s ... nope\n",
244 bdevname(rdev->bdev, b));
247 pr_debug("md/raid0:%s: checking %s ..."
248 " contained as device %d\n",
250 bdevname(rdev->bdev, b), c);
253 if (!smallest || rdev->sectors < smallest->sectors) {
255 pr_debug("md/raid0:%s: (%llu) is smallest!.\n",
257 (unsigned long long)rdev->sectors);
262 sectors = (smallest->sectors - zone->dev_start) * c;
263 pr_debug("md/raid0:%s: zone->nb_dev: %d, sectors: %llu\n",
265 zone->nb_dev, (unsigned long long)sectors);
267 curr_zone_end += sectors;
268 zone->zone_end = curr_zone_end;
270 pr_debug("md/raid0:%s: current zone start: %llu\n",
272 (unsigned long long)smallest->sectors);
275 pr_debug("md/raid0:%s: done.\n", mdname(mddev));
276 *private_conf = conf;
280 kfree(conf->strip_zone);
281 kfree(conf->devlist);
283 *private_conf = ERR_PTR(err);
287 /* Find the zone which holds a particular offset
288 * Update *sectorp to be an offset in that zone
290 static struct strip_zone *find_zone(struct r0conf *conf,
294 struct strip_zone *z = conf->strip_zone;
295 sector_t sector = *sectorp;
297 for (i = 0; i < conf->nr_strip_zones; i++)
298 if (sector < z[i].zone_end) {
300 *sectorp = sector - z[i-1].zone_end;
307 * remaps the bio to the target device. we separate two flows.
308 * power 2 flow and a general flow for the sake of performance
310 static struct md_rdev *map_sector(struct mddev *mddev, struct strip_zone *zone,
311 sector_t sector, sector_t *sector_offset)
313 unsigned int sect_in_chunk;
315 struct r0conf *conf = mddev->private;
316 int raid_disks = conf->strip_zone[0].nb_dev;
317 unsigned int chunk_sects = mddev->chunk_sectors;
319 if (is_power_of_2(chunk_sects)) {
320 int chunksect_bits = ffz(~chunk_sects);
321 /* find the sector offset inside the chunk */
322 sect_in_chunk = sector & (chunk_sects - 1);
323 sector >>= chunksect_bits;
325 chunk = *sector_offset;
326 /* quotient is the chunk in real device*/
327 sector_div(chunk, zone->nb_dev << chunksect_bits);
329 sect_in_chunk = sector_div(sector, chunk_sects);
330 chunk = *sector_offset;
331 sector_div(chunk, chunk_sects * zone->nb_dev);
334 * position the bio over the real device
335 * real sector = chunk in device + starting of zone
336 * + the position in the chunk
338 *sector_offset = (chunk * chunk_sects) + sect_in_chunk;
339 return conf->devlist[(zone - conf->strip_zone)*raid_disks
340 + sector_div(sector, zone->nb_dev)];
343 static sector_t raid0_size(struct mddev *mddev, sector_t sectors, int raid_disks)
345 sector_t array_sectors = 0;
346 struct md_rdev *rdev;
348 WARN_ONCE(sectors || raid_disks,
349 "%s does not support generic reshape\n", __func__);
351 rdev_for_each(rdev, mddev)
352 array_sectors += (rdev->sectors &
353 ~(sector_t)(mddev->chunk_sectors-1));
355 return array_sectors;
358 static void raid0_free(struct mddev *mddev, void *priv);
360 static int raid0_run(struct mddev *mddev)
365 if (mddev->chunk_sectors == 0) {
366 pr_warn("md/raid0:%s: chunk size must be set.\n", mdname(mddev));
369 if (md_check_no_bitmap(mddev))
372 /* if private is not null, we are here after takeover */
373 if (mddev->private == NULL) {
374 ret = create_strip_zones(mddev, &conf);
377 mddev->private = conf;
379 conf = mddev->private;
381 struct md_rdev *rdev;
382 bool discard_supported = false;
384 blk_queue_max_hw_sectors(mddev->queue, mddev->chunk_sectors);
385 blk_queue_max_write_same_sectors(mddev->queue, mddev->chunk_sectors);
386 blk_queue_max_write_zeroes_sectors(mddev->queue, mddev->chunk_sectors);
387 blk_queue_max_discard_sectors(mddev->queue, mddev->chunk_sectors);
389 blk_queue_io_min(mddev->queue, mddev->chunk_sectors << 9);
390 blk_queue_io_opt(mddev->queue,
391 (mddev->chunk_sectors << 9) * mddev->raid_disks);
393 rdev_for_each(rdev, mddev) {
394 disk_stack_limits(mddev->gendisk, rdev->bdev,
395 rdev->data_offset << 9);
396 if (blk_queue_discard(bdev_get_queue(rdev->bdev)))
397 discard_supported = true;
399 if (!discard_supported)
400 queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, mddev->queue);
402 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, mddev->queue);
405 /* calculate array device size */
406 md_set_array_sectors(mddev, raid0_size(mddev, 0, 0));
408 pr_debug("md/raid0:%s: md_size is %llu sectors.\n",
410 (unsigned long long)mddev->array_sectors);
413 /* calculate the max read-ahead size.
414 * For read-ahead of large files to be effective, we need to
415 * readahead at least twice a whole stripe. i.e. number of devices
416 * multiplied by chunk size times 2.
417 * If an individual device has an ra_pages greater than the
418 * chunk size, then we will not drive that device as hard as it
419 * wants. We consider this a configuration error: a larger
420 * chunksize should be used in that case.
422 int stripe = mddev->raid_disks *
423 (mddev->chunk_sectors << 9) / PAGE_SIZE;
424 if (mddev->queue->backing_dev_info->ra_pages < 2* stripe)
425 mddev->queue->backing_dev_info->ra_pages = 2* stripe;
430 ret = md_integrity_register(mddev);
435 static void raid0_free(struct mddev *mddev, void *priv)
437 struct r0conf *conf = priv;
439 kfree(conf->strip_zone);
440 kfree(conf->devlist);
445 * Is io distribute over 1 or more chunks ?
447 static inline int is_io_in_chunk_boundary(struct mddev *mddev,
448 unsigned int chunk_sects, struct bio *bio)
450 if (likely(is_power_of_2(chunk_sects))) {
451 return chunk_sects >=
452 ((bio->bi_iter.bi_sector & (chunk_sects-1))
455 sector_t sector = bio->bi_iter.bi_sector;
456 return chunk_sects >= (sector_div(sector, chunk_sects)
461 static void raid0_make_request(struct mddev *mddev, struct bio *bio)
463 struct strip_zone *zone;
464 struct md_rdev *tmp_dev;
467 if (unlikely(bio->bi_opf & REQ_PREFLUSH)) {
468 md_flush_request(mddev, bio);
473 sector_t bio_sector = bio->bi_iter.bi_sector;
474 sector_t sector = bio_sector;
475 unsigned chunk_sects = mddev->chunk_sectors;
477 unsigned sectors = chunk_sects -
478 (likely(is_power_of_2(chunk_sects))
479 ? (sector & (chunk_sects-1))
480 : sector_div(sector, chunk_sects));
482 /* Restore due to sector_div */
485 if (sectors < bio_sectors(bio)) {
486 split = bio_split(bio, sectors, GFP_NOIO, fs_bio_set);
487 bio_chain(split, bio);
492 zone = find_zone(mddev->private, §or);
493 tmp_dev = map_sector(mddev, zone, sector, §or);
494 split->bi_bdev = tmp_dev->bdev;
495 split->bi_iter.bi_sector = sector + zone->dev_start +
496 tmp_dev->data_offset;
498 if (unlikely((bio_op(split) == REQ_OP_DISCARD) &&
499 !blk_queue_discard(bdev_get_queue(split->bi_bdev)))) {
504 trace_block_bio_remap(bdev_get_queue(split->bi_bdev),
505 split, disk_devt(mddev->gendisk),
507 mddev_check_writesame(mddev, split);
508 mddev_check_write_zeroes(mddev, split);
509 generic_make_request(split);
511 } while (split != bio);
514 static void raid0_status(struct seq_file *seq, struct mddev *mddev)
516 seq_printf(seq, " %dk chunks", mddev->chunk_sectors / 2);
520 static void *raid0_takeover_raid45(struct mddev *mddev)
522 struct md_rdev *rdev;
523 struct r0conf *priv_conf;
525 if (mddev->degraded != 1) {
526 pr_warn("md/raid0:%s: raid5 must be degraded! Degraded disks: %d\n",
529 return ERR_PTR(-EINVAL);
532 rdev_for_each(rdev, mddev) {
533 /* check slot number for a disk */
534 if (rdev->raid_disk == mddev->raid_disks-1) {
535 pr_warn("md/raid0:%s: raid5 must have missing parity disk!\n",
537 return ERR_PTR(-EINVAL);
539 rdev->sectors = mddev->dev_sectors;
542 /* Set new parameters */
543 mddev->new_level = 0;
544 mddev->new_layout = 0;
545 mddev->new_chunk_sectors = mddev->chunk_sectors;
547 mddev->delta_disks = -1;
548 /* make sure it will be not marked as dirty */
549 mddev->recovery_cp = MaxSector;
550 mddev_clear_unsupported_flags(mddev, UNSUPPORTED_MDDEV_FLAGS);
552 create_strip_zones(mddev, &priv_conf);
557 static void *raid0_takeover_raid10(struct mddev *mddev)
559 struct r0conf *priv_conf;
562 * - far_copies must be 1
563 * - near_copies must be 2
564 * - disks number must be even
565 * - all mirrors must be already degraded
567 if (mddev->layout != ((1 << 8) + 2)) {
568 pr_warn("md/raid0:%s:: Raid0 cannot takeover layout: 0x%x\n",
571 return ERR_PTR(-EINVAL);
573 if (mddev->raid_disks & 1) {
574 pr_warn("md/raid0:%s: Raid0 cannot takeover Raid10 with odd disk number.\n",
576 return ERR_PTR(-EINVAL);
578 if (mddev->degraded != (mddev->raid_disks>>1)) {
579 pr_warn("md/raid0:%s: All mirrors must be already degraded!\n",
581 return ERR_PTR(-EINVAL);
584 /* Set new parameters */
585 mddev->new_level = 0;
586 mddev->new_layout = 0;
587 mddev->new_chunk_sectors = mddev->chunk_sectors;
588 mddev->delta_disks = - mddev->raid_disks / 2;
589 mddev->raid_disks += mddev->delta_disks;
591 /* make sure it will be not marked as dirty */
592 mddev->recovery_cp = MaxSector;
593 mddev_clear_unsupported_flags(mddev, UNSUPPORTED_MDDEV_FLAGS);
595 create_strip_zones(mddev, &priv_conf);
599 static void *raid0_takeover_raid1(struct mddev *mddev)
601 struct r0conf *priv_conf;
605 * - (N - 1) mirror drives must be already faulty
607 if ((mddev->raid_disks - 1) != mddev->degraded) {
608 pr_err("md/raid0:%s: (N - 1) mirrors drives must be already faulty!\n",
610 return ERR_PTR(-EINVAL);
614 * a raid1 doesn't have the notion of chunk size, so
615 * figure out the largest suitable size we can use.
617 chunksect = 64 * 2; /* 64K by default */
619 /* The array must be an exact multiple of chunksize */
620 while (chunksect && (mddev->array_sectors & (chunksect - 1)))
623 if ((chunksect << 9) < PAGE_SIZE)
624 /* array size does not allow a suitable chunk size */
625 return ERR_PTR(-EINVAL);
627 /* Set new parameters */
628 mddev->new_level = 0;
629 mddev->new_layout = 0;
630 mddev->new_chunk_sectors = chunksect;
631 mddev->chunk_sectors = chunksect;
632 mddev->delta_disks = 1 - mddev->raid_disks;
633 mddev->raid_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);
642 static void *raid0_takeover(struct mddev *mddev)
644 /* raid0 can take over:
645 * raid4 - if all data disks are active.
646 * raid5 - providing it is Raid4 layout and one disk is faulty
647 * raid10 - assuming we have all necessary active disks
648 * raid1 - with (N -1) mirror drives faulty
652 pr_warn("md/raid0: %s: cannot takeover array with bitmap\n",
654 return ERR_PTR(-EBUSY);
656 if (mddev->level == 4)
657 return raid0_takeover_raid45(mddev);
659 if (mddev->level == 5) {
660 if (mddev->layout == ALGORITHM_PARITY_N)
661 return raid0_takeover_raid45(mddev);
663 pr_warn("md/raid0:%s: Raid can only takeover Raid5 with layout: %d\n",
664 mdname(mddev), ALGORITHM_PARITY_N);
667 if (mddev->level == 10)
668 return raid0_takeover_raid10(mddev);
670 if (mddev->level == 1)
671 return raid0_takeover_raid1(mddev);
673 pr_warn("Takeover from raid%i to raid0 not supported\n",
676 return ERR_PTR(-EINVAL);
679 static void raid0_quiesce(struct mddev *mddev, int state)
683 static struct md_personality raid0_personality=
687 .owner = THIS_MODULE,
688 .make_request = raid0_make_request,
691 .status = raid0_status,
693 .takeover = raid0_takeover,
694 .quiesce = raid0_quiesce,
695 .congested = raid0_congested,
698 static int __init raid0_init (void)
700 return register_md_personality (&raid0_personality);
703 static void raid0_exit (void)
705 unregister_md_personality (&raid0_personality);
708 module_init(raid0_init);
709 module_exit(raid0_exit);
710 MODULE_LICENSE("GPL");
711 MODULE_DESCRIPTION("RAID0 (striping) personality for MD");
712 MODULE_ALIAS("md-personality-2"); /* RAID0 */
713 MODULE_ALIAS("md-raid0");
714 MODULE_ALIAS("md-level-0");