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>
28 static int raid0_congested(struct mddev *mddev, int bits)
30 struct r0conf *conf = mddev->private;
31 struct md_rdev **devlist = conf->devlist;
32 int raid_disks = conf->strip_zone[0].nb_dev;
35 for (i = 0; i < raid_disks && !ret ; i++) {
36 struct request_queue *q = bdev_get_queue(devlist[i]->bdev);
38 ret |= bdi_congested(&q->backing_dev_info, bits);
44 * inform the user of the raid configuration
46 static void dump_zones(struct mddev *mddev)
49 sector_t zone_size = 0;
50 sector_t zone_start = 0;
51 char b[BDEVNAME_SIZE];
52 struct r0conf *conf = mddev->private;
53 int raid_disks = conf->strip_zone[0].nb_dev;
54 printk(KERN_INFO "md: RAID0 configuration for %s - %d zone%s\n",
56 conf->nr_strip_zones, conf->nr_strip_zones==1?"":"s");
57 for (j = 0; j < conf->nr_strip_zones; j++) {
58 printk(KERN_INFO "md: zone%d=[", j);
59 for (k = 0; k < conf->strip_zone[j].nb_dev; k++)
60 printk(KERN_CONT "%s%s", k?"/":"",
61 bdevname(conf->devlist[j*raid_disks
63 printk(KERN_CONT "]\n");
65 zone_size = conf->strip_zone[j].zone_end - zone_start;
66 printk(KERN_INFO " zone-offset=%10lluKB, "
67 "device-offset=%10lluKB, size=%10lluKB\n",
68 (unsigned long long)zone_start>>1,
69 (unsigned long long)conf->strip_zone[j].dev_start>>1,
70 (unsigned long long)zone_size>>1);
71 zone_start = conf->strip_zone[j].zone_end;
73 printk(KERN_INFO "\n");
76 static int create_strip_zones(struct mddev *mddev, struct r0conf **private_conf)
79 sector_t curr_zone_end, sectors;
80 struct md_rdev *smallest, *rdev1, *rdev2, *rdev, **dev;
81 struct strip_zone *zone;
83 char b[BDEVNAME_SIZE];
84 char b2[BDEVNAME_SIZE];
85 struct r0conf *conf = kzalloc(sizeof(*conf), GFP_KERNEL);
86 unsigned short blksize = 512;
90 rdev_for_each(rdev1, mddev) {
91 pr_debug("md/raid0:%s: looking at %s\n",
93 bdevname(rdev1->bdev, b));
96 /* round size to chunk_size */
97 sectors = rdev1->sectors;
98 sector_div(sectors, mddev->chunk_sectors);
99 rdev1->sectors = sectors * mddev->chunk_sectors;
101 blksize = max(blksize, queue_logical_block_size(
102 rdev1->bdev->bd_disk->queue));
104 rdev_for_each(rdev2, mddev) {
105 pr_debug("md/raid0:%s: comparing %s(%llu)"
108 bdevname(rdev1->bdev,b),
109 (unsigned long long)rdev1->sectors,
110 bdevname(rdev2->bdev,b2),
111 (unsigned long long)rdev2->sectors);
112 if (rdev2 == rdev1) {
113 pr_debug("md/raid0:%s: END\n",
117 if (rdev2->sectors == rdev1->sectors) {
119 * Not unique, don't count it as a new
122 pr_debug("md/raid0:%s: EQUAL\n",
127 pr_debug("md/raid0:%s: NOT EQUAL\n",
131 pr_debug("md/raid0:%s: ==> UNIQUE\n",
133 conf->nr_strip_zones++;
134 pr_debug("md/raid0:%s: %d zones\n",
135 mdname(mddev), conf->nr_strip_zones);
138 pr_debug("md/raid0:%s: FINAL %d zones\n",
139 mdname(mddev), conf->nr_strip_zones);
141 * now since we have the hard sector sizes, we can make sure
142 * chunk size is a multiple of that sector size
144 if ((mddev->chunk_sectors << 9) % blksize) {
145 printk(KERN_ERR "md/raid0:%s: chunk_size of %d not multiple of block size %d\n",
147 mddev->chunk_sectors << 9, blksize);
153 conf->strip_zone = kzalloc(sizeof(struct strip_zone)*
154 conf->nr_strip_zones, GFP_KERNEL);
155 if (!conf->strip_zone)
157 conf->devlist = kzalloc(sizeof(struct md_rdev*)*
158 conf->nr_strip_zones*mddev->raid_disks,
163 /* The first zone must contain all devices, so here we check that
164 * there is a proper alignment of slots to devices and find them all
166 zone = &conf->strip_zone[0];
171 rdev_for_each(rdev1, mddev) {
172 int j = rdev1->raid_disk;
174 if (mddev->level == 10) {
175 /* taking over a raid10-n2 array */
177 rdev1->new_raid_disk = j;
180 if (mddev->level == 1) {
181 /* taiking over a raid1 array-
182 * we have only one active disk
185 rdev1->new_raid_disk = j;
190 "md/raid0:%s: remove inactive devices before converting to RAID0\n",
194 if (j >= mddev->raid_disks) {
195 printk(KERN_ERR "md/raid0:%s: bad disk number %d - "
196 "aborting!\n", mdname(mddev), j);
200 printk(KERN_ERR "md/raid0:%s: multiple devices for %d - "
201 "aborting!\n", mdname(mddev), j);
206 if (rdev1->bdev->bd_disk->queue->merge_bvec_fn)
207 conf->has_merge_bvec = 1;
209 if (!smallest || (rdev1->sectors < smallest->sectors))
213 if (cnt != mddev->raid_disks) {
214 printk(KERN_ERR "md/raid0:%s: too few disks (%d of %d) - "
215 "aborting!\n", 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)];
341 * raid0_mergeable_bvec -- tell bio layer if two requests can be merged
342 * @mddev: the md device
343 * @bvm: properties of new bio
344 * @biovec: the request that could be merged to it.
346 * Return amount of bytes we can accept at this offset
348 static int raid0_mergeable_bvec(struct mddev *mddev,
349 struct bvec_merge_data *bvm,
350 struct bio_vec *biovec)
352 struct r0conf *conf = mddev->private;
353 sector_t sector = bvm->bi_sector + get_start_sect(bvm->bi_bdev);
354 sector_t sector_offset = sector;
356 unsigned int chunk_sectors = mddev->chunk_sectors;
357 unsigned int bio_sectors = bvm->bi_size >> 9;
358 struct strip_zone *zone;
359 struct md_rdev *rdev;
360 struct request_queue *subq;
362 if (is_power_of_2(chunk_sectors))
363 max = (chunk_sectors - ((sector & (chunk_sectors-1))
364 + bio_sectors)) << 9;
366 max = (chunk_sectors - (sector_div(sector, chunk_sectors)
367 + bio_sectors)) << 9;
369 max = 0; /* bio_add cannot handle a negative return */
370 if (max <= biovec->bv_len && bio_sectors == 0)
371 return biovec->bv_len;
372 if (max < biovec->bv_len)
373 /* too small already, no need to check further */
375 if (!conf->has_merge_bvec)
378 /* May need to check subordinate device */
379 sector = sector_offset;
380 zone = find_zone(mddev->private, §or_offset);
381 rdev = map_sector(mddev, zone, sector, §or_offset);
382 subq = bdev_get_queue(rdev->bdev);
383 if (subq->merge_bvec_fn) {
384 bvm->bi_bdev = rdev->bdev;
385 bvm->bi_sector = sector_offset + zone->dev_start +
387 return min(max, subq->merge_bvec_fn(subq, bvm, biovec));
392 static sector_t raid0_size(struct mddev *mddev, sector_t sectors, int raid_disks)
394 sector_t array_sectors = 0;
395 struct md_rdev *rdev;
397 WARN_ONCE(sectors || raid_disks,
398 "%s does not support generic reshape\n", __func__);
400 rdev_for_each(rdev, mddev)
401 array_sectors += (rdev->sectors &
402 ~(sector_t)(mddev->chunk_sectors-1));
404 return array_sectors;
407 static void raid0_free(struct mddev *mddev, void *priv);
409 static int raid0_run(struct mddev *mddev)
414 if (mddev->chunk_sectors == 0) {
415 printk(KERN_ERR "md/raid0:%s: chunk size must be set.\n",
419 if (md_check_no_bitmap(mddev))
422 /* if private is not null, we are here after takeover */
423 if (mddev->private == NULL) {
424 ret = create_strip_zones(mddev, &conf);
427 mddev->private = conf;
429 conf = mddev->private;
431 struct md_rdev *rdev;
432 bool discard_supported = false;
434 rdev_for_each(rdev, mddev) {
435 disk_stack_limits(mddev->gendisk, rdev->bdev,
436 rdev->data_offset << 9);
437 if (blk_queue_discard(bdev_get_queue(rdev->bdev)))
438 discard_supported = true;
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 blk_queue_io_min(mddev->queue, mddev->chunk_sectors << 9);
445 blk_queue_io_opt(mddev->queue,
446 (mddev->chunk_sectors << 9) * mddev->raid_disks);
448 if (!discard_supported)
449 queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, mddev->queue);
451 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, mddev->queue);
454 /* calculate array device size */
455 md_set_array_sectors(mddev, raid0_size(mddev, 0, 0));
457 printk(KERN_INFO "md/raid0:%s: md_size is %llu sectors.\n",
459 (unsigned long long)mddev->array_sectors);
462 /* calculate the max read-ahead size.
463 * For read-ahead of large files to be effective, we need to
464 * readahead at least twice a whole stripe. i.e. number of devices
465 * multiplied by chunk size times 2.
466 * If an individual device has an ra_pages greater than the
467 * chunk size, then we will not drive that device as hard as it
468 * wants. We consider this a configuration error: a larger
469 * chunksize should be used in that case.
471 int stripe = mddev->raid_disks *
472 (mddev->chunk_sectors << 9) / PAGE_SIZE;
473 if (mddev->queue->backing_dev_info.ra_pages < 2* stripe)
474 mddev->queue->backing_dev_info.ra_pages = 2* stripe;
479 ret = md_integrity_register(mddev);
484 static void raid0_free(struct mddev *mddev, void *priv)
486 struct r0conf *conf = priv;
488 kfree(conf->strip_zone);
489 kfree(conf->devlist);
494 * Is io distribute over 1 or more chunks ?
496 static inline int is_io_in_chunk_boundary(struct mddev *mddev,
497 unsigned int chunk_sects, struct bio *bio)
499 if (likely(is_power_of_2(chunk_sects))) {
500 return chunk_sects >=
501 ((bio->bi_iter.bi_sector & (chunk_sects-1))
504 sector_t sector = bio->bi_iter.bi_sector;
505 return chunk_sects >= (sector_div(sector, chunk_sects)
510 static void raid0_make_request(struct mddev *mddev, struct bio *bio)
512 struct strip_zone *zone;
513 struct md_rdev *tmp_dev;
516 if (unlikely(bio->bi_rw & REQ_FLUSH)) {
517 md_flush_request(mddev, bio);
522 sector_t sector = bio->bi_iter.bi_sector;
523 unsigned chunk_sects = mddev->chunk_sectors;
525 unsigned sectors = chunk_sects -
526 (likely(is_power_of_2(chunk_sects))
527 ? (sector & (chunk_sects-1))
528 : sector_div(sector, chunk_sects));
530 /* Restore due to sector_div */
531 sector = bio->bi_iter.bi_sector;
533 if (sectors < bio_sectors(bio)) {
534 split = bio_split(bio, sectors, GFP_NOIO, fs_bio_set);
535 bio_chain(split, bio);
540 zone = find_zone(mddev->private, §or);
541 tmp_dev = map_sector(mddev, zone, sector, §or);
542 split->bi_bdev = tmp_dev->bdev;
543 split->bi_iter.bi_sector = sector + zone->dev_start +
544 tmp_dev->data_offset;
546 if (unlikely((split->bi_rw & REQ_DISCARD) &&
547 !blk_queue_discard(bdev_get_queue(split->bi_bdev)))) {
551 generic_make_request(split);
552 } while (split != bio);
555 static void raid0_status(struct seq_file *seq, struct mddev *mddev)
557 seq_printf(seq, " %dk chunks", mddev->chunk_sectors / 2);
561 static void *raid0_takeover_raid45(struct mddev *mddev)
563 struct md_rdev *rdev;
564 struct r0conf *priv_conf;
566 if (mddev->degraded != 1) {
567 printk(KERN_ERR "md/raid0:%s: raid5 must be degraded! Degraded disks: %d\n",
570 return ERR_PTR(-EINVAL);
573 rdev_for_each(rdev, mddev) {
574 /* check slot number for a disk */
575 if (rdev->raid_disk == mddev->raid_disks-1) {
576 printk(KERN_ERR "md/raid0:%s: raid5 must have missing parity disk!\n",
578 return ERR_PTR(-EINVAL);
580 rdev->sectors = mddev->dev_sectors;
583 /* Set new parameters */
584 mddev->new_level = 0;
585 mddev->new_layout = 0;
586 mddev->new_chunk_sectors = mddev->chunk_sectors;
588 mddev->delta_disks = -1;
589 /* make sure it will be not marked as dirty */
590 mddev->recovery_cp = MaxSector;
592 create_strip_zones(mddev, &priv_conf);
596 static void *raid0_takeover_raid10(struct mddev *mddev)
598 struct r0conf *priv_conf;
601 * - far_copies must be 1
602 * - near_copies must be 2
603 * - disks number must be even
604 * - all mirrors must be already degraded
606 if (mddev->layout != ((1 << 8) + 2)) {
607 printk(KERN_ERR "md/raid0:%s:: Raid0 cannot takover layout: 0x%x\n",
610 return ERR_PTR(-EINVAL);
612 if (mddev->raid_disks & 1) {
613 printk(KERN_ERR "md/raid0:%s: Raid0 cannot takover Raid10 with odd disk number.\n",
615 return ERR_PTR(-EINVAL);
617 if (mddev->degraded != (mddev->raid_disks>>1)) {
618 printk(KERN_ERR "md/raid0:%s: All mirrors must be already degraded!\n",
620 return ERR_PTR(-EINVAL);
623 /* Set new parameters */
624 mddev->new_level = 0;
625 mddev->new_layout = 0;
626 mddev->new_chunk_sectors = mddev->chunk_sectors;
627 mddev->delta_disks = - mddev->raid_disks / 2;
628 mddev->raid_disks += mddev->delta_disks;
630 /* make sure it will be not marked as dirty */
631 mddev->recovery_cp = MaxSector;
633 create_strip_zones(mddev, &priv_conf);
637 static void *raid0_takeover_raid1(struct mddev *mddev)
639 struct r0conf *priv_conf;
643 * - (N - 1) mirror drives must be already faulty
645 if ((mddev->raid_disks - 1) != mddev->degraded) {
646 printk(KERN_ERR "md/raid0:%s: (N - 1) mirrors drives must be already faulty!\n",
648 return ERR_PTR(-EINVAL);
652 * a raid1 doesn't have the notion of chunk size, so
653 * figure out the largest suitable size we can use.
655 chunksect = 64 * 2; /* 64K by default */
657 /* The array must be an exact multiple of chunksize */
658 while (chunksect && (mddev->array_sectors & (chunksect - 1)))
661 if ((chunksect << 9) < PAGE_SIZE)
662 /* array size does not allow a suitable chunk size */
663 return ERR_PTR(-EINVAL);
665 /* Set new parameters */
666 mddev->new_level = 0;
667 mddev->new_layout = 0;
668 mddev->new_chunk_sectors = chunksect;
669 mddev->chunk_sectors = chunksect;
670 mddev->delta_disks = 1 - mddev->raid_disks;
671 mddev->raid_disks = 1;
672 /* make sure it will be not marked as dirty */
673 mddev->recovery_cp = MaxSector;
675 create_strip_zones(mddev, &priv_conf);
679 static void *raid0_takeover(struct mddev *mddev)
681 /* raid0 can take over:
682 * raid4 - if all data disks are active.
683 * raid5 - providing it is Raid4 layout and one disk is faulty
684 * raid10 - assuming we have all necessary active disks
685 * raid1 - with (N -1) mirror drives faulty
689 printk(KERN_ERR "md/raid0: %s: cannot takeover array with bitmap\n",
691 return ERR_PTR(-EBUSY);
693 if (mddev->level == 4)
694 return raid0_takeover_raid45(mddev);
696 if (mddev->level == 5) {
697 if (mddev->layout == ALGORITHM_PARITY_N)
698 return raid0_takeover_raid45(mddev);
700 printk(KERN_ERR "md/raid0:%s: Raid can only takeover Raid5 with layout: %d\n",
701 mdname(mddev), ALGORITHM_PARITY_N);
704 if (mddev->level == 10)
705 return raid0_takeover_raid10(mddev);
707 if (mddev->level == 1)
708 return raid0_takeover_raid1(mddev);
710 printk(KERN_ERR "Takeover from raid%i to raid0 not supported\n",
713 return ERR_PTR(-EINVAL);
716 static void raid0_quiesce(struct mddev *mddev, int state)
720 static struct md_personality raid0_personality=
724 .owner = THIS_MODULE,
725 .make_request = raid0_make_request,
728 .status = raid0_status,
730 .takeover = raid0_takeover,
731 .quiesce = raid0_quiesce,
732 .congested = raid0_congested,
733 .mergeable_bvec = raid0_mergeable_bvec,
736 static int __init raid0_init (void)
738 return register_md_personality (&raid0_personality);
741 static void raid0_exit (void)
743 unregister_md_personality (&raid0_personality);
746 module_init(raid0_init);
747 module_exit(raid0_exit);
748 MODULE_LICENSE("GPL");
749 MODULE_DESCRIPTION("RAID0 (striping) personality for MD");
750 MODULE_ALIAS("md-personality-2"); /* RAID0 */
751 MODULE_ALIAS("md-raid0");
752 MODULE_ALIAS("md-level-0");