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 bool discard_supported = false;
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 rdev_for_each(rdev2, mddev) {
102 pr_debug("md/raid0:%s: comparing %s(%llu)"
105 bdevname(rdev1->bdev,b),
106 (unsigned long long)rdev1->sectors,
107 bdevname(rdev2->bdev,b2),
108 (unsigned long long)rdev2->sectors);
109 if (rdev2 == rdev1) {
110 pr_debug("md/raid0:%s: END\n",
114 if (rdev2->sectors == rdev1->sectors) {
116 * Not unique, don't count it as a new
119 pr_debug("md/raid0:%s: EQUAL\n",
124 pr_debug("md/raid0:%s: NOT EQUAL\n",
128 pr_debug("md/raid0:%s: ==> UNIQUE\n",
130 conf->nr_strip_zones++;
131 pr_debug("md/raid0:%s: %d zones\n",
132 mdname(mddev), conf->nr_strip_zones);
135 pr_debug("md/raid0:%s: FINAL %d zones\n",
136 mdname(mddev), conf->nr_strip_zones);
138 conf->strip_zone = kzalloc(sizeof(struct strip_zone)*
139 conf->nr_strip_zones, GFP_KERNEL);
140 if (!conf->strip_zone)
142 conf->devlist = kzalloc(sizeof(struct md_rdev*)*
143 conf->nr_strip_zones*mddev->raid_disks,
148 /* The first zone must contain all devices, so here we check that
149 * there is a proper alignment of slots to devices and find them all
151 zone = &conf->strip_zone[0];
156 rdev_for_each(rdev1, mddev) {
157 int j = rdev1->raid_disk;
159 if (mddev->level == 10) {
160 /* taking over a raid10-n2 array */
162 rdev1->new_raid_disk = j;
165 if (mddev->level == 1) {
166 /* taiking over a raid1 array-
167 * we have only one active disk
170 rdev1->new_raid_disk = j;
175 "md/raid0:%s: remove inactive devices before converting to RAID0\n",
179 if (j >= mddev->raid_disks) {
180 printk(KERN_ERR "md/raid0:%s: bad disk number %d - "
181 "aborting!\n", mdname(mddev), j);
185 printk(KERN_ERR "md/raid0:%s: multiple devices for %d - "
186 "aborting!\n", mdname(mddev), j);
191 disk_stack_limits(mddev->gendisk, rdev1->bdev,
192 rdev1->data_offset << 9);
194 if (rdev1->bdev->bd_disk->queue->merge_bvec_fn)
195 conf->has_merge_bvec = 1;
197 if (!smallest || (rdev1->sectors < smallest->sectors))
201 if (blk_queue_discard(bdev_get_queue(rdev1->bdev)))
202 discard_supported = true;
204 if (cnt != mddev->raid_disks) {
205 printk(KERN_ERR "md/raid0:%s: too few disks (%d of %d) - "
206 "aborting!\n", mdname(mddev), cnt, mddev->raid_disks);
210 zone->zone_end = smallest->sectors * cnt;
212 curr_zone_end = zone->zone_end;
214 /* now do the other zones */
215 for (i = 1; i < conf->nr_strip_zones; i++)
219 zone = conf->strip_zone + i;
220 dev = conf->devlist + i * mddev->raid_disks;
222 pr_debug("md/raid0:%s: zone %d\n", mdname(mddev), i);
223 zone->dev_start = smallest->sectors;
227 for (j=0; j<cnt; j++) {
228 rdev = conf->devlist[j];
229 if (rdev->sectors <= zone->dev_start) {
230 pr_debug("md/raid0:%s: checking %s ... nope\n",
232 bdevname(rdev->bdev, b));
235 pr_debug("md/raid0:%s: checking %s ..."
236 " contained as device %d\n",
238 bdevname(rdev->bdev, b), c);
241 if (!smallest || rdev->sectors < smallest->sectors) {
243 pr_debug("md/raid0:%s: (%llu) is smallest!.\n",
245 (unsigned long long)rdev->sectors);
250 sectors = (smallest->sectors - zone->dev_start) * c;
251 pr_debug("md/raid0:%s: zone->nb_dev: %d, sectors: %llu\n",
253 zone->nb_dev, (unsigned long long)sectors);
255 curr_zone_end += sectors;
256 zone->zone_end = curr_zone_end;
258 pr_debug("md/raid0:%s: current zone start: %llu\n",
260 (unsigned long long)smallest->sectors);
264 * now since we have the hard sector sizes, we can make sure
265 * chunk size is a multiple of that sector size
267 if ((mddev->chunk_sectors << 9) % queue_logical_block_size(mddev->queue)) {
268 printk(KERN_ERR "md/raid0:%s: chunk_size of %d not valid\n",
270 mddev->chunk_sectors << 9);
275 blk_queue_io_min(mddev->queue, mddev->chunk_sectors << 9);
276 blk_queue_io_opt(mddev->queue,
277 (mddev->chunk_sectors << 9) * mddev->raid_disks);
279 if (!discard_supported)
280 queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, mddev->queue);
282 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, mddev->queue);
285 pr_debug("md/raid0:%s: done.\n", mdname(mddev));
286 *private_conf = conf;
290 kfree(conf->strip_zone);
291 kfree(conf->devlist);
293 *private_conf = ERR_PTR(err);
297 /* Find the zone which holds a particular offset
298 * Update *sectorp to be an offset in that zone
300 static struct strip_zone *find_zone(struct r0conf *conf,
304 struct strip_zone *z = conf->strip_zone;
305 sector_t sector = *sectorp;
307 for (i = 0; i < conf->nr_strip_zones; i++)
308 if (sector < z[i].zone_end) {
310 *sectorp = sector - z[i-1].zone_end;
317 * remaps the bio to the target device. we separate two flows.
318 * power 2 flow and a general flow for the sake of performance
320 static struct md_rdev *map_sector(struct mddev *mddev, struct strip_zone *zone,
321 sector_t sector, sector_t *sector_offset)
323 unsigned int sect_in_chunk;
325 struct r0conf *conf = mddev->private;
326 int raid_disks = conf->strip_zone[0].nb_dev;
327 unsigned int chunk_sects = mddev->chunk_sectors;
329 if (is_power_of_2(chunk_sects)) {
330 int chunksect_bits = ffz(~chunk_sects);
331 /* find the sector offset inside the chunk */
332 sect_in_chunk = sector & (chunk_sects - 1);
333 sector >>= chunksect_bits;
335 chunk = *sector_offset;
336 /* quotient is the chunk in real device*/
337 sector_div(chunk, zone->nb_dev << chunksect_bits);
339 sect_in_chunk = sector_div(sector, chunk_sects);
340 chunk = *sector_offset;
341 sector_div(chunk, chunk_sects * zone->nb_dev);
344 * position the bio over the real device
345 * real sector = chunk in device + starting of zone
346 * + the position in the chunk
348 *sector_offset = (chunk * chunk_sects) + sect_in_chunk;
349 return conf->devlist[(zone - conf->strip_zone)*raid_disks
350 + sector_div(sector, zone->nb_dev)];
354 * raid0_mergeable_bvec -- tell bio layer if two requests can be merged
355 * @mddev: the md device
356 * @bvm: properties of new bio
357 * @biovec: the request that could be merged to it.
359 * Return amount of bytes we can accept at this offset
361 static int raid0_mergeable_bvec(struct mddev *mddev,
362 struct bvec_merge_data *bvm,
363 struct bio_vec *biovec)
365 struct r0conf *conf = mddev->private;
366 sector_t sector = bvm->bi_sector + get_start_sect(bvm->bi_bdev);
367 sector_t sector_offset = sector;
369 unsigned int chunk_sectors = mddev->chunk_sectors;
370 unsigned int bio_sectors = bvm->bi_size >> 9;
371 struct strip_zone *zone;
372 struct md_rdev *rdev;
373 struct request_queue *subq;
375 if (is_power_of_2(chunk_sectors))
376 max = (chunk_sectors - ((sector & (chunk_sectors-1))
377 + bio_sectors)) << 9;
379 max = (chunk_sectors - (sector_div(sector, chunk_sectors)
380 + bio_sectors)) << 9;
382 max = 0; /* bio_add cannot handle a negative return */
383 if (max <= biovec->bv_len && bio_sectors == 0)
384 return biovec->bv_len;
385 if (max < biovec->bv_len)
386 /* too small already, no need to check further */
388 if (!conf->has_merge_bvec)
391 /* May need to check subordinate device */
392 sector = sector_offset;
393 zone = find_zone(mddev->private, §or_offset);
394 rdev = map_sector(mddev, zone, sector, §or_offset);
395 subq = bdev_get_queue(rdev->bdev);
396 if (subq->merge_bvec_fn) {
397 bvm->bi_bdev = rdev->bdev;
398 bvm->bi_sector = sector_offset + zone->dev_start +
400 return min(max, subq->merge_bvec_fn(subq, bvm, biovec));
405 static sector_t raid0_size(struct mddev *mddev, sector_t sectors, int raid_disks)
407 sector_t array_sectors = 0;
408 struct md_rdev *rdev;
410 WARN_ONCE(sectors || raid_disks,
411 "%s does not support generic reshape\n", __func__);
413 rdev_for_each(rdev, mddev)
414 array_sectors += (rdev->sectors &
415 ~(sector_t)(mddev->chunk_sectors-1));
417 return array_sectors;
420 static void raid0_free(struct mddev *mddev, void *priv);
422 static int raid0_run(struct mddev *mddev)
427 if (mddev->chunk_sectors == 0) {
428 printk(KERN_ERR "md/raid0:%s: chunk size must be set.\n",
432 if (md_check_no_bitmap(mddev))
436 blk_queue_max_hw_sectors(mddev->queue, mddev->chunk_sectors);
437 blk_queue_max_write_same_sectors(mddev->queue, mddev->chunk_sectors);
438 blk_queue_max_discard_sectors(mddev->queue, mddev->chunk_sectors);
441 /* if private is not null, we are here after takeover */
442 if (mddev->private == NULL) {
443 ret = create_strip_zones(mddev, &conf);
446 mddev->private = conf;
448 conf = mddev->private;
450 /* calculate array device size */
451 md_set_array_sectors(mddev, raid0_size(mddev, 0, 0));
453 printk(KERN_INFO "md/raid0:%s: md_size is %llu sectors.\n",
455 (unsigned long long)mddev->array_sectors);
458 /* calculate the max read-ahead size.
459 * For read-ahead of large files to be effective, we need to
460 * readahead at least twice a whole stripe. i.e. number of devices
461 * multiplied by chunk size times 2.
462 * If an individual device has an ra_pages greater than the
463 * chunk size, then we will not drive that device as hard as it
464 * wants. We consider this a configuration error: a larger
465 * chunksize should be used in that case.
467 int stripe = mddev->raid_disks *
468 (mddev->chunk_sectors << 9) / PAGE_SIZE;
469 if (mddev->queue->backing_dev_info.ra_pages < 2* stripe)
470 mddev->queue->backing_dev_info.ra_pages = 2* stripe;
475 ret = md_integrity_register(mddev);
480 static void raid0_free(struct mddev *mddev, void *priv)
482 struct r0conf *conf = priv;
484 kfree(conf->strip_zone);
485 kfree(conf->devlist);
490 * Is io distribute over 1 or more chunks ?
492 static inline int is_io_in_chunk_boundary(struct mddev *mddev,
493 unsigned int chunk_sects, struct bio *bio)
495 if (likely(is_power_of_2(chunk_sects))) {
496 return chunk_sects >=
497 ((bio->bi_iter.bi_sector & (chunk_sects-1))
500 sector_t sector = bio->bi_iter.bi_sector;
501 return chunk_sects >= (sector_div(sector, chunk_sects)
506 static void raid0_make_request(struct mddev *mddev, struct bio *bio)
508 struct strip_zone *zone;
509 struct md_rdev *tmp_dev;
512 if (unlikely(bio->bi_rw & REQ_FLUSH)) {
513 md_flush_request(mddev, bio);
518 sector_t sector = bio->bi_iter.bi_sector;
519 unsigned chunk_sects = mddev->chunk_sectors;
521 unsigned sectors = chunk_sects -
522 (likely(is_power_of_2(chunk_sects))
523 ? (sector & (chunk_sects-1))
524 : sector_div(sector, chunk_sects));
526 if (sectors < bio_sectors(bio)) {
527 split = bio_split(bio, sectors, GFP_NOIO, fs_bio_set);
528 bio_chain(split, bio);
533 sector = bio->bi_iter.bi_sector;
534 zone = find_zone(mddev->private, §or);
535 tmp_dev = map_sector(mddev, zone, sector, §or);
536 split->bi_bdev = tmp_dev->bdev;
537 split->bi_iter.bi_sector = sector + zone->dev_start +
538 tmp_dev->data_offset;
540 if (unlikely((split->bi_rw & REQ_DISCARD) &&
541 !blk_queue_discard(bdev_get_queue(split->bi_bdev)))) {
545 generic_make_request(split);
546 } while (split != bio);
549 static void raid0_status(struct seq_file *seq, struct mddev *mddev)
551 seq_printf(seq, " %dk chunks", mddev->chunk_sectors / 2);
555 static void *raid0_takeover_raid45(struct mddev *mddev)
557 struct md_rdev *rdev;
558 struct r0conf *priv_conf;
560 if (mddev->degraded != 1) {
561 printk(KERN_ERR "md/raid0:%s: raid5 must be degraded! Degraded disks: %d\n",
564 return ERR_PTR(-EINVAL);
567 rdev_for_each(rdev, mddev) {
568 /* check slot number for a disk */
569 if (rdev->raid_disk == mddev->raid_disks-1) {
570 printk(KERN_ERR "md/raid0:%s: raid5 must have missing parity disk!\n",
572 return ERR_PTR(-EINVAL);
574 rdev->sectors = mddev->dev_sectors;
577 /* Set new parameters */
578 mddev->new_level = 0;
579 mddev->new_layout = 0;
580 mddev->new_chunk_sectors = mddev->chunk_sectors;
582 mddev->delta_disks = -1;
583 /* make sure it will be not marked as dirty */
584 mddev->recovery_cp = MaxSector;
586 create_strip_zones(mddev, &priv_conf);
590 static void *raid0_takeover_raid10(struct mddev *mddev)
592 struct r0conf *priv_conf;
595 * - far_copies must be 1
596 * - near_copies must be 2
597 * - disks number must be even
598 * - all mirrors must be already degraded
600 if (mddev->layout != ((1 << 8) + 2)) {
601 printk(KERN_ERR "md/raid0:%s:: Raid0 cannot takover layout: 0x%x\n",
604 return ERR_PTR(-EINVAL);
606 if (mddev->raid_disks & 1) {
607 printk(KERN_ERR "md/raid0:%s: Raid0 cannot takover Raid10 with odd disk number.\n",
609 return ERR_PTR(-EINVAL);
611 if (mddev->degraded != (mddev->raid_disks>>1)) {
612 printk(KERN_ERR "md/raid0:%s: All mirrors must be already degraded!\n",
614 return ERR_PTR(-EINVAL);
617 /* Set new parameters */
618 mddev->new_level = 0;
619 mddev->new_layout = 0;
620 mddev->new_chunk_sectors = mddev->chunk_sectors;
621 mddev->delta_disks = - mddev->raid_disks / 2;
622 mddev->raid_disks += mddev->delta_disks;
624 /* make sure it will be not marked as dirty */
625 mddev->recovery_cp = MaxSector;
627 create_strip_zones(mddev, &priv_conf);
631 static void *raid0_takeover_raid1(struct mddev *mddev)
633 struct r0conf *priv_conf;
637 * - (N - 1) mirror drives must be already faulty
639 if ((mddev->raid_disks - 1) != mddev->degraded) {
640 printk(KERN_ERR "md/raid0:%s: (N - 1) mirrors drives must be already faulty!\n",
642 return ERR_PTR(-EINVAL);
646 * a raid1 doesn't have the notion of chunk size, so
647 * figure out the largest suitable size we can use.
649 chunksect = 64 * 2; /* 64K by default */
651 /* The array must be an exact multiple of chunksize */
652 while (chunksect && (mddev->array_sectors & (chunksect - 1)))
655 if ((chunksect << 9) < PAGE_SIZE)
656 /* array size does not allow a suitable chunk size */
657 return ERR_PTR(-EINVAL);
659 /* Set new parameters */
660 mddev->new_level = 0;
661 mddev->new_layout = 0;
662 mddev->new_chunk_sectors = chunksect;
663 mddev->chunk_sectors = chunksect;
664 mddev->delta_disks = 1 - mddev->raid_disks;
665 mddev->raid_disks = 1;
666 /* make sure it will be not marked as dirty */
667 mddev->recovery_cp = MaxSector;
669 create_strip_zones(mddev, &priv_conf);
673 static void *raid0_takeover(struct mddev *mddev)
675 /* raid0 can take over:
676 * raid4 - if all data disks are active.
677 * raid5 - providing it is Raid4 layout and one disk is faulty
678 * raid10 - assuming we have all necessary active disks
679 * raid1 - with (N -1) mirror drives faulty
683 printk(KERN_ERR "md/raid0: %s: cannot takeover array with bitmap\n",
685 return ERR_PTR(-EBUSY);
687 if (mddev->level == 4)
688 return raid0_takeover_raid45(mddev);
690 if (mddev->level == 5) {
691 if (mddev->layout == ALGORITHM_PARITY_N)
692 return raid0_takeover_raid45(mddev);
694 printk(KERN_ERR "md/raid0:%s: Raid can only takeover Raid5 with layout: %d\n",
695 mdname(mddev), ALGORITHM_PARITY_N);
698 if (mddev->level == 10)
699 return raid0_takeover_raid10(mddev);
701 if (mddev->level == 1)
702 return raid0_takeover_raid1(mddev);
704 printk(KERN_ERR "Takeover from raid%i to raid0 not supported\n",
707 return ERR_PTR(-EINVAL);
710 static void raid0_quiesce(struct mddev *mddev, int state)
714 static struct md_personality raid0_personality=
718 .owner = THIS_MODULE,
719 .make_request = raid0_make_request,
722 .status = raid0_status,
724 .takeover = raid0_takeover,
725 .quiesce = raid0_quiesce,
726 .congested = raid0_congested,
727 .mergeable_bvec = raid0_mergeable_bvec,
730 static int __init raid0_init (void)
732 return register_md_personality (&raid0_personality);
735 static void raid0_exit (void)
737 unregister_md_personality (&raid0_personality);
740 module_init(raid0_init);
741 module_exit(raid0_exit);
742 MODULE_LICENSE("GPL");
743 MODULE_DESCRIPTION("RAID0 (striping) personality for MD");
744 MODULE_ALIAS("md-personality-2"); /* RAID0 */
745 MODULE_ALIAS("md-raid0");
746 MODULE_ALIAS("md-level-0");