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
9 RAID-0 management functions.
11 This program is free software; you can redistribute it and/or modify
12 it under the terms of the GNU General Public License as published by
13 the Free Software Foundation; either version 2, or (at your option)
16 You should have received a copy of the GNU General Public License
17 (for example /usr/src/linux/COPYING); if not, write to the Free
18 Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
21 #include <linux/blkdev.h>
22 #include <linux/seq_file.h>
23 #include <linux/module.h>
24 #include <linux/slab.h>
29 static int raid0_congested(void *data, int bits)
31 struct mddev *mddev = data;
32 struct r0conf *conf = mddev->private;
33 struct md_rdev **devlist = conf->devlist;
34 int raid_disks = conf->strip_zone[0].nb_dev;
37 if (mddev_congested(mddev, bits))
40 for (i = 0; i < raid_disks && !ret ; i++) {
41 struct request_queue *q = bdev_get_queue(devlist[i]->bdev);
43 ret |= bdi_congested(&q->backing_dev_info, bits);
49 * inform the user of the raid configuration
51 static void dump_zones(struct mddev *mddev)
54 sector_t zone_size = 0;
55 sector_t zone_start = 0;
56 char b[BDEVNAME_SIZE];
57 struct r0conf *conf = mddev->private;
58 int raid_disks = conf->strip_zone[0].nb_dev;
59 printk(KERN_INFO "md: RAID0 configuration for %s - %d zone%s\n",
61 conf->nr_strip_zones, conf->nr_strip_zones==1?"":"s");
62 for (j = 0; j < conf->nr_strip_zones; j++) {
63 printk(KERN_INFO "md: zone%d=[", j);
64 for (k = 0; k < conf->strip_zone[j].nb_dev; k++)
65 printk(KERN_CONT "%s%s", k?"/":"",
66 bdevname(conf->devlist[j*raid_disks
68 printk(KERN_CONT "]\n");
70 zone_size = conf->strip_zone[j].zone_end - zone_start;
71 printk(KERN_INFO " zone-offset=%10lluKB, "
72 "device-offset=%10lluKB, size=%10lluKB\n",
73 (unsigned long long)zone_start>>1,
74 (unsigned long long)conf->strip_zone[j].dev_start>>1,
75 (unsigned long long)zone_size>>1);
76 zone_start = conf->strip_zone[j].zone_end;
78 printk(KERN_INFO "\n");
81 static int create_strip_zones(struct mddev *mddev, struct r0conf **private_conf)
84 sector_t curr_zone_end, sectors;
85 struct md_rdev *smallest, *rdev1, *rdev2, *rdev, **dev;
86 struct strip_zone *zone;
88 char b[BDEVNAME_SIZE];
89 char b2[BDEVNAME_SIZE];
90 struct r0conf *conf = kzalloc(sizeof(*conf), GFP_KERNEL);
91 bool discard_supported = false;
95 rdev_for_each(rdev1, mddev) {
96 pr_debug("md/raid0:%s: looking at %s\n",
98 bdevname(rdev1->bdev, b));
101 /* round size to chunk_size */
102 sectors = rdev1->sectors;
103 sector_div(sectors, mddev->chunk_sectors);
104 rdev1->sectors = sectors * mddev->chunk_sectors;
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 conf->strip_zone = kzalloc(sizeof(struct strip_zone)*
144 conf->nr_strip_zones, GFP_KERNEL);
145 if (!conf->strip_zone)
147 conf->devlist = kzalloc(sizeof(struct md_rdev*)*
148 conf->nr_strip_zones*mddev->raid_disks,
153 /* The first zone must contain all devices, so here we check that
154 * there is a proper alignment of slots to devices and find them all
156 zone = &conf->strip_zone[0];
161 rdev_for_each(rdev1, mddev) {
162 int j = rdev1->raid_disk;
164 if (mddev->level == 10) {
165 /* taking over a raid10-n2 array */
167 rdev1->new_raid_disk = j;
170 if (mddev->level == 1) {
171 /* taiking over a raid1 array-
172 * we have only one active disk
175 rdev1->new_raid_disk = j;
180 "md/raid0:%s: remove inactive devices before converting to RAID0\n",
184 if (j >= mddev->raid_disks) {
185 printk(KERN_ERR "md/raid0:%s: bad disk number %d - "
186 "aborting!\n", mdname(mddev), j);
190 printk(KERN_ERR "md/raid0:%s: multiple devices for %d - "
191 "aborting!\n", mdname(mddev), j);
196 disk_stack_limits(mddev->gendisk, rdev1->bdev,
197 rdev1->data_offset << 9);
199 if (rdev1->bdev->bd_disk->queue->merge_bvec_fn)
200 conf->has_merge_bvec = 1;
202 if (!smallest || (rdev1->sectors < smallest->sectors))
206 if (blk_queue_discard(bdev_get_queue(rdev1->bdev)))
207 discard_supported = true;
209 if (cnt != mddev->raid_disks) {
210 printk(KERN_ERR "md/raid0:%s: too few disks (%d of %d) - "
211 "aborting!\n", mdname(mddev), cnt, mddev->raid_disks);
215 zone->zone_end = smallest->sectors * cnt;
217 curr_zone_end = zone->zone_end;
219 /* now do the other zones */
220 for (i = 1; i < conf->nr_strip_zones; i++)
224 zone = conf->strip_zone + i;
225 dev = conf->devlist + i * mddev->raid_disks;
227 pr_debug("md/raid0:%s: zone %d\n", mdname(mddev), i);
228 zone->dev_start = smallest->sectors;
232 for (j=0; j<cnt; j++) {
233 rdev = conf->devlist[j];
234 if (rdev->sectors <= zone->dev_start) {
235 pr_debug("md/raid0:%s: checking %s ... nope\n",
237 bdevname(rdev->bdev, b));
240 pr_debug("md/raid0:%s: checking %s ..."
241 " contained as device %d\n",
243 bdevname(rdev->bdev, b), c);
246 if (!smallest || rdev->sectors < smallest->sectors) {
248 pr_debug("md/raid0:%s: (%llu) is smallest!.\n",
250 (unsigned long long)rdev->sectors);
255 sectors = (smallest->sectors - zone->dev_start) * c;
256 pr_debug("md/raid0:%s: zone->nb_dev: %d, sectors: %llu\n",
258 zone->nb_dev, (unsigned long long)sectors);
260 curr_zone_end += sectors;
261 zone->zone_end = curr_zone_end;
263 pr_debug("md/raid0:%s: current zone start: %llu\n",
265 (unsigned long long)smallest->sectors);
267 mddev->queue->backing_dev_info.congested_fn = raid0_congested;
268 mddev->queue->backing_dev_info.congested_data = mddev;
271 * now since we have the hard sector sizes, we can make sure
272 * chunk size is a multiple of that sector size
274 if ((mddev->chunk_sectors << 9) % queue_logical_block_size(mddev->queue)) {
275 printk(KERN_ERR "md/raid0:%s: chunk_size of %d not valid\n",
277 mddev->chunk_sectors << 9);
281 blk_queue_io_min(mddev->queue, mddev->chunk_sectors << 9);
282 blk_queue_io_opt(mddev->queue,
283 (mddev->chunk_sectors << 9) * mddev->raid_disks);
285 if (!discard_supported)
286 queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, mddev->queue);
288 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, mddev->queue);
290 pr_debug("md/raid0:%s: done.\n", mdname(mddev));
291 *private_conf = conf;
295 kfree(conf->strip_zone);
296 kfree(conf->devlist);
298 *private_conf = ERR_PTR(err);
302 /* Find the zone which holds a particular offset
303 * Update *sectorp to be an offset in that zone
305 static struct strip_zone *find_zone(struct r0conf *conf,
309 struct strip_zone *z = conf->strip_zone;
310 sector_t sector = *sectorp;
312 for (i = 0; i < conf->nr_strip_zones; i++)
313 if (sector < z[i].zone_end) {
315 *sectorp = sector - z[i-1].zone_end;
322 * remaps the bio to the target device. we separate two flows.
323 * power 2 flow and a general flow for the sake of perfromance
325 static struct md_rdev *map_sector(struct mddev *mddev, struct strip_zone *zone,
326 sector_t sector, sector_t *sector_offset)
328 unsigned int sect_in_chunk;
330 struct r0conf *conf = mddev->private;
331 int raid_disks = conf->strip_zone[0].nb_dev;
332 unsigned int chunk_sects = mddev->chunk_sectors;
334 if (is_power_of_2(chunk_sects)) {
335 int chunksect_bits = ffz(~chunk_sects);
336 /* find the sector offset inside the chunk */
337 sect_in_chunk = sector & (chunk_sects - 1);
338 sector >>= chunksect_bits;
340 chunk = *sector_offset;
341 /* quotient is the chunk in real device*/
342 sector_div(chunk, zone->nb_dev << chunksect_bits);
344 sect_in_chunk = sector_div(sector, chunk_sects);
345 chunk = *sector_offset;
346 sector_div(chunk, chunk_sects * zone->nb_dev);
349 * position the bio over the real device
350 * real sector = chunk in device + starting of zone
351 * + the position in the chunk
353 *sector_offset = (chunk * chunk_sects) + sect_in_chunk;
354 return conf->devlist[(zone - conf->strip_zone)*raid_disks
355 + sector_div(sector, zone->nb_dev)];
359 * raid0_mergeable_bvec -- tell bio layer if two requests can be merged
361 * @bvm: properties of new bio
362 * @biovec: the request that could be merged to it.
364 * Return amount of bytes we can accept at this offset
366 static int raid0_mergeable_bvec(struct request_queue *q,
367 struct bvec_merge_data *bvm,
368 struct bio_vec *biovec)
370 struct mddev *mddev = q->queuedata;
371 struct r0conf *conf = mddev->private;
372 sector_t sector = bvm->bi_sector + get_start_sect(bvm->bi_bdev);
373 sector_t sector_offset = sector;
375 unsigned int chunk_sectors = mddev->chunk_sectors;
376 unsigned int bio_sectors = bvm->bi_size >> 9;
377 struct strip_zone *zone;
378 struct md_rdev *rdev;
379 struct request_queue *subq;
381 if (is_power_of_2(chunk_sectors))
382 max = (chunk_sectors - ((sector & (chunk_sectors-1))
383 + bio_sectors)) << 9;
385 max = (chunk_sectors - (sector_div(sector, chunk_sectors)
386 + bio_sectors)) << 9;
388 max = 0; /* bio_add cannot handle a negative return */
389 if (max <= biovec->bv_len && bio_sectors == 0)
390 return biovec->bv_len;
391 if (max < biovec->bv_len)
392 /* too small already, no need to check further */
394 if (!conf->has_merge_bvec)
397 /* May need to check subordinate device */
398 sector = sector_offset;
399 zone = find_zone(mddev->private, §or_offset);
400 rdev = map_sector(mddev, zone, sector, §or_offset);
401 subq = bdev_get_queue(rdev->bdev);
402 if (subq->merge_bvec_fn) {
403 bvm->bi_bdev = rdev->bdev;
404 bvm->bi_sector = sector_offset + zone->dev_start +
406 return min(max, subq->merge_bvec_fn(subq, bvm, biovec));
411 static sector_t raid0_size(struct mddev *mddev, sector_t sectors, int raid_disks)
413 sector_t array_sectors = 0;
414 struct md_rdev *rdev;
416 WARN_ONCE(sectors || raid_disks,
417 "%s does not support generic reshape\n", __func__);
419 rdev_for_each(rdev, mddev)
420 array_sectors += (rdev->sectors &
421 ~(sector_t)(mddev->chunk_sectors-1));
423 return array_sectors;
426 static int raid0_stop(struct mddev *mddev);
428 static int raid0_run(struct mddev *mddev)
433 if (mddev->chunk_sectors == 0) {
434 printk(KERN_ERR "md/raid0:%s: chunk size must be set.\n",
438 if (md_check_no_bitmap(mddev))
440 blk_queue_max_hw_sectors(mddev->queue, mddev->chunk_sectors);
441 blk_queue_max_write_same_sectors(mddev->queue, mddev->chunk_sectors);
442 blk_queue_max_discard_sectors(mddev->queue, mddev->chunk_sectors);
444 /* if private is not null, we are here after takeover */
445 if (mddev->private == NULL) {
446 ret = create_strip_zones(mddev, &conf);
449 mddev->private = conf;
451 conf = mddev->private;
453 /* calculate array device size */
454 md_set_array_sectors(mddev, raid0_size(mddev, 0, 0));
456 printk(KERN_INFO "md/raid0:%s: md_size is %llu sectors.\n",
458 (unsigned long long)mddev->array_sectors);
459 /* calculate the max read-ahead size.
460 * For read-ahead of large files to be effective, we need to
461 * readahead at least twice a whole stripe. i.e. number of devices
462 * multiplied by chunk size times 2.
463 * If an individual device has an ra_pages greater than the
464 * chunk size, then we will not drive that device as hard as it
465 * wants. We consider this a configuration error: a larger
466 * chunksize should be used in that case.
469 int stripe = mddev->raid_disks *
470 (mddev->chunk_sectors << 9) / PAGE_SIZE;
471 if (mddev->queue->backing_dev_info.ra_pages < 2* stripe)
472 mddev->queue->backing_dev_info.ra_pages = 2* stripe;
475 blk_queue_merge_bvec(mddev->queue, raid0_mergeable_bvec);
478 ret = md_integrity_register(mddev);
485 static int raid0_stop(struct mddev *mddev)
487 struct r0conf *conf = mddev->private;
489 blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/
490 kfree(conf->strip_zone);
491 kfree(conf->devlist);
493 mddev->private = NULL;
498 * Is io distribute over 1 or more chunks ?
500 static inline int is_io_in_chunk_boundary(struct mddev *mddev,
501 unsigned int chunk_sects, struct bio *bio)
503 if (likely(is_power_of_2(chunk_sects))) {
504 return chunk_sects >= ((bio->bi_sector & (chunk_sects-1))
507 sector_t sector = bio->bi_sector;
508 return chunk_sects >= (sector_div(sector, chunk_sects)
513 static void raid0_make_request(struct mddev *mddev, struct bio *bio)
515 unsigned int chunk_sects;
516 sector_t sector_offset;
517 struct strip_zone *zone;
518 struct md_rdev *tmp_dev;
520 if (unlikely(bio->bi_rw & REQ_FLUSH)) {
521 md_flush_request(mddev, bio);
525 chunk_sects = mddev->chunk_sectors;
526 if (unlikely(!is_io_in_chunk_boundary(mddev, chunk_sects, bio))) {
527 sector_t sector = bio->bi_sector;
529 /* Sanity check -- queue functions should prevent this happening */
530 if (bio_segments(bio) > 1)
532 /* This is a one page bio that upper layers
533 * refuse to split for us, so we need to split it.
535 if (likely(is_power_of_2(chunk_sects)))
536 bp = bio_split(bio, chunk_sects - (sector &
539 bp = bio_split(bio, chunk_sects -
540 sector_div(sector, chunk_sects));
541 raid0_make_request(mddev, &bp->bio1);
542 raid0_make_request(mddev, &bp->bio2);
543 bio_pair_release(bp);
547 sector_offset = bio->bi_sector;
548 zone = find_zone(mddev->private, §or_offset);
549 tmp_dev = map_sector(mddev, zone, bio->bi_sector,
551 bio->bi_bdev = tmp_dev->bdev;
552 bio->bi_sector = sector_offset + zone->dev_start +
553 tmp_dev->data_offset;
555 if (unlikely((bio->bi_rw & REQ_DISCARD) &&
556 !blk_queue_discard(bdev_get_queue(bio->bi_bdev)))) {
562 generic_make_request(bio);
566 printk("md/raid0:%s: make_request bug: can't convert block across chunks"
567 " or bigger than %dk %llu %d\n",
568 mdname(mddev), chunk_sects / 2,
569 (unsigned long long)bio->bi_sector, bio_sectors(bio) / 2);
575 static void raid0_status(struct seq_file *seq, struct mddev *mddev)
577 seq_printf(seq, " %dk chunks", mddev->chunk_sectors / 2);
581 static void *raid0_takeover_raid45(struct mddev *mddev)
583 struct md_rdev *rdev;
584 struct r0conf *priv_conf;
586 if (mddev->degraded != 1) {
587 printk(KERN_ERR "md/raid0:%s: raid5 must be degraded! Degraded disks: %d\n",
590 return ERR_PTR(-EINVAL);
593 rdev_for_each(rdev, mddev) {
594 /* check slot number for a disk */
595 if (rdev->raid_disk == mddev->raid_disks-1) {
596 printk(KERN_ERR "md/raid0:%s: raid5 must have missing parity disk!\n",
598 return ERR_PTR(-EINVAL);
600 rdev->sectors = mddev->dev_sectors;
603 /* Set new parameters */
604 mddev->new_level = 0;
605 mddev->new_layout = 0;
606 mddev->new_chunk_sectors = mddev->chunk_sectors;
608 mddev->delta_disks = -1;
609 /* make sure it will be not marked as dirty */
610 mddev->recovery_cp = MaxSector;
612 create_strip_zones(mddev, &priv_conf);
616 static void *raid0_takeover_raid10(struct mddev *mddev)
618 struct r0conf *priv_conf;
621 * - far_copies must be 1
622 * - near_copies must be 2
623 * - disks number must be even
624 * - all mirrors must be already degraded
626 if (mddev->layout != ((1 << 8) + 2)) {
627 printk(KERN_ERR "md/raid0:%s:: Raid0 cannot takover layout: 0x%x\n",
630 return ERR_PTR(-EINVAL);
632 if (mddev->raid_disks & 1) {
633 printk(KERN_ERR "md/raid0:%s: Raid0 cannot takover Raid10 with odd disk number.\n",
635 return ERR_PTR(-EINVAL);
637 if (mddev->degraded != (mddev->raid_disks>>1)) {
638 printk(KERN_ERR "md/raid0:%s: All mirrors must be already degraded!\n",
640 return ERR_PTR(-EINVAL);
643 /* Set new parameters */
644 mddev->new_level = 0;
645 mddev->new_layout = 0;
646 mddev->new_chunk_sectors = mddev->chunk_sectors;
647 mddev->delta_disks = - mddev->raid_disks / 2;
648 mddev->raid_disks += mddev->delta_disks;
650 /* make sure it will be not marked as dirty */
651 mddev->recovery_cp = MaxSector;
653 create_strip_zones(mddev, &priv_conf);
657 static void *raid0_takeover_raid1(struct mddev *mddev)
659 struct r0conf *priv_conf;
663 * - (N - 1) mirror drives must be already faulty
665 if ((mddev->raid_disks - 1) != mddev->degraded) {
666 printk(KERN_ERR "md/raid0:%s: (N - 1) mirrors drives must be already faulty!\n",
668 return ERR_PTR(-EINVAL);
672 * a raid1 doesn't have the notion of chunk size, so
673 * figure out the largest suitable size we can use.
675 chunksect = 64 * 2; /* 64K by default */
677 /* The array must be an exact multiple of chunksize */
678 while (chunksect && (mddev->array_sectors & (chunksect - 1)))
681 if ((chunksect << 9) < PAGE_SIZE)
682 /* array size does not allow a suitable chunk size */
683 return ERR_PTR(-EINVAL);
685 /* Set new parameters */
686 mddev->new_level = 0;
687 mddev->new_layout = 0;
688 mddev->new_chunk_sectors = chunksect;
689 mddev->chunk_sectors = chunksect;
690 mddev->delta_disks = 1 - mddev->raid_disks;
691 mddev->raid_disks = 1;
692 /* make sure it will be not marked as dirty */
693 mddev->recovery_cp = MaxSector;
695 create_strip_zones(mddev, &priv_conf);
699 static void *raid0_takeover(struct mddev *mddev)
701 /* raid0 can take over:
702 * raid4 - if all data disks are active.
703 * raid5 - providing it is Raid4 layout and one disk is faulty
704 * raid10 - assuming we have all necessary active disks
705 * raid1 - with (N -1) mirror drives faulty
707 if (mddev->level == 4)
708 return raid0_takeover_raid45(mddev);
710 if (mddev->level == 5) {
711 if (mddev->layout == ALGORITHM_PARITY_N)
712 return raid0_takeover_raid45(mddev);
714 printk(KERN_ERR "md/raid0:%s: Raid can only takeover Raid5 with layout: %d\n",
715 mdname(mddev), ALGORITHM_PARITY_N);
718 if (mddev->level == 10)
719 return raid0_takeover_raid10(mddev);
721 if (mddev->level == 1)
722 return raid0_takeover_raid1(mddev);
724 printk(KERN_ERR "Takeover from raid%i to raid0 not supported\n",
727 return ERR_PTR(-EINVAL);
730 static void raid0_quiesce(struct mddev *mddev, int state)
734 static struct md_personality raid0_personality=
738 .owner = THIS_MODULE,
739 .make_request = raid0_make_request,
742 .status = raid0_status,
744 .takeover = raid0_takeover,
745 .quiesce = raid0_quiesce,
748 static int __init raid0_init (void)
750 return register_md_personality (&raid0_personality);
753 static void raid0_exit (void)
755 unregister_md_personality (&raid0_personality);
758 module_init(raid0_init);
759 module_exit(raid0_exit);
760 MODULE_LICENSE("GPL");
761 MODULE_DESCRIPTION("RAID0 (striping) personality for MD");
762 MODULE_ALIAS("md-personality-2"); /* RAID0 */
763 MODULE_ALIAS("md-raid0");
764 MODULE_ALIAS("md-level-0");