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);
192 disk_stack_limits(mddev->gendisk, rdev1->bdev,
193 rdev1->data_offset << 9);
195 if (rdev1->bdev->bd_disk->queue->merge_bvec_fn)
196 conf->has_merge_bvec = 1;
198 if (!smallest || (rdev1->sectors < smallest->sectors))
202 if (blk_queue_discard(bdev_get_queue(rdev1->bdev)))
203 discard_supported = true;
205 if (cnt != mddev->raid_disks) {
206 printk(KERN_ERR "md/raid0:%s: too few disks (%d of %d) - "
207 "aborting!\n", mdname(mddev), cnt, mddev->raid_disks);
211 zone->zone_end = smallest->sectors * cnt;
213 curr_zone_end = zone->zone_end;
215 /* now do the other zones */
216 for (i = 1; i < conf->nr_strip_zones; i++)
220 zone = conf->strip_zone + i;
221 dev = conf->devlist + i * mddev->raid_disks;
223 pr_debug("md/raid0:%s: zone %d\n", mdname(mddev), i);
224 zone->dev_start = smallest->sectors;
228 for (j=0; j<cnt; j++) {
229 rdev = conf->devlist[j];
230 if (rdev->sectors <= zone->dev_start) {
231 pr_debug("md/raid0:%s: checking %s ... nope\n",
233 bdevname(rdev->bdev, b));
236 pr_debug("md/raid0:%s: checking %s ..."
237 " contained as device %d\n",
239 bdevname(rdev->bdev, b), c);
242 if (!smallest || rdev->sectors < smallest->sectors) {
244 pr_debug("md/raid0:%s: (%llu) is smallest!.\n",
246 (unsigned long long)rdev->sectors);
251 sectors = (smallest->sectors - zone->dev_start) * c;
252 pr_debug("md/raid0:%s: zone->nb_dev: %d, sectors: %llu\n",
254 zone->nb_dev, (unsigned long long)sectors);
256 curr_zone_end += sectors;
257 zone->zone_end = curr_zone_end;
259 pr_debug("md/raid0:%s: current zone start: %llu\n",
261 (unsigned long long)smallest->sectors);
265 * now since we have the hard sector sizes, we can make sure
266 * chunk size is a multiple of that sector size
268 if ((mddev->chunk_sectors << 9) % queue_logical_block_size(mddev->queue)) {
269 printk(KERN_ERR "md/raid0:%s: chunk_size of %d not valid\n",
271 mddev->chunk_sectors << 9);
276 blk_queue_io_min(mddev->queue, mddev->chunk_sectors << 9);
277 blk_queue_io_opt(mddev->queue,
278 (mddev->chunk_sectors << 9) * mddev->raid_disks);
280 if (!discard_supported)
281 queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, mddev->queue);
283 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, mddev->queue);
286 pr_debug("md/raid0:%s: done.\n", mdname(mddev));
287 *private_conf = conf;
291 kfree(conf->strip_zone);
292 kfree(conf->devlist);
294 *private_conf = ERR_PTR(err);
298 /* Find the zone which holds a particular offset
299 * Update *sectorp to be an offset in that zone
301 static struct strip_zone *find_zone(struct r0conf *conf,
305 struct strip_zone *z = conf->strip_zone;
306 sector_t sector = *sectorp;
308 for (i = 0; i < conf->nr_strip_zones; i++)
309 if (sector < z[i].zone_end) {
311 *sectorp = sector - z[i-1].zone_end;
318 * remaps the bio to the target device. we separate two flows.
319 * power 2 flow and a general flow for the sake of performance
321 static struct md_rdev *map_sector(struct mddev *mddev, struct strip_zone *zone,
322 sector_t sector, sector_t *sector_offset)
324 unsigned int sect_in_chunk;
326 struct r0conf *conf = mddev->private;
327 int raid_disks = conf->strip_zone[0].nb_dev;
328 unsigned int chunk_sects = mddev->chunk_sectors;
330 if (is_power_of_2(chunk_sects)) {
331 int chunksect_bits = ffz(~chunk_sects);
332 /* find the sector offset inside the chunk */
333 sect_in_chunk = sector & (chunk_sects - 1);
334 sector >>= chunksect_bits;
336 chunk = *sector_offset;
337 /* quotient is the chunk in real device*/
338 sector_div(chunk, zone->nb_dev << chunksect_bits);
340 sect_in_chunk = sector_div(sector, chunk_sects);
341 chunk = *sector_offset;
342 sector_div(chunk, chunk_sects * zone->nb_dev);
345 * position the bio over the real device
346 * real sector = chunk in device + starting of zone
347 * + the position in the chunk
349 *sector_offset = (chunk * chunk_sects) + sect_in_chunk;
350 return conf->devlist[(zone - conf->strip_zone)*raid_disks
351 + sector_div(sector, zone->nb_dev)];
355 * raid0_mergeable_bvec -- tell bio layer if two requests can be merged
356 * @mddev: the md device
357 * @bvm: properties of new bio
358 * @biovec: the request that could be merged to it.
360 * Return amount of bytes we can accept at this offset
362 static int raid0_mergeable_bvec(struct mddev *mddev,
363 struct bvec_merge_data *bvm,
364 struct bio_vec *biovec)
366 struct r0conf *conf = mddev->private;
367 sector_t sector = bvm->bi_sector + get_start_sect(bvm->bi_bdev);
368 sector_t sector_offset = sector;
370 unsigned int chunk_sectors = mddev->chunk_sectors;
371 unsigned int bio_sectors = bvm->bi_size >> 9;
372 struct strip_zone *zone;
373 struct md_rdev *rdev;
374 struct request_queue *subq;
376 if (is_power_of_2(chunk_sectors))
377 max = (chunk_sectors - ((sector & (chunk_sectors-1))
378 + bio_sectors)) << 9;
380 max = (chunk_sectors - (sector_div(sector, chunk_sectors)
381 + bio_sectors)) << 9;
383 max = 0; /* bio_add cannot handle a negative return */
384 if (max <= biovec->bv_len && bio_sectors == 0)
385 return biovec->bv_len;
386 if (max < biovec->bv_len)
387 /* too small already, no need to check further */
389 if (!conf->has_merge_bvec)
392 /* May need to check subordinate device */
393 sector = sector_offset;
394 zone = find_zone(mddev->private, §or_offset);
395 rdev = map_sector(mddev, zone, sector, §or_offset);
396 subq = bdev_get_queue(rdev->bdev);
397 if (subq->merge_bvec_fn) {
398 bvm->bi_bdev = rdev->bdev;
399 bvm->bi_sector = sector_offset + zone->dev_start +
401 return min(max, subq->merge_bvec_fn(subq, bvm, biovec));
406 static sector_t raid0_size(struct mddev *mddev, sector_t sectors, int raid_disks)
408 sector_t array_sectors = 0;
409 struct md_rdev *rdev;
411 WARN_ONCE(sectors || raid_disks,
412 "%s does not support generic reshape\n", __func__);
414 rdev_for_each(rdev, mddev)
415 array_sectors += (rdev->sectors &
416 ~(sector_t)(mddev->chunk_sectors-1));
418 return array_sectors;
421 static void raid0_free(struct mddev *mddev, void *priv);
423 static int raid0_run(struct mddev *mddev)
428 if (mddev->chunk_sectors == 0) {
429 printk(KERN_ERR "md/raid0:%s: chunk size must be set.\n",
433 if (md_check_no_bitmap(mddev))
437 blk_queue_max_hw_sectors(mddev->queue, mddev->chunk_sectors);
438 blk_queue_max_write_same_sectors(mddev->queue, mddev->chunk_sectors);
439 blk_queue_max_discard_sectors(mddev->queue, mddev->chunk_sectors);
442 /* if private is not null, we are here after takeover */
443 if (mddev->private == NULL) {
444 ret = create_strip_zones(mddev, &conf);
447 mddev->private = conf;
449 conf = mddev->private;
451 /* calculate array device size */
452 md_set_array_sectors(mddev, raid0_size(mddev, 0, 0));
454 printk(KERN_INFO "md/raid0:%s: md_size is %llu sectors.\n",
456 (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.
468 int stripe = mddev->raid_disks *
469 (mddev->chunk_sectors << 9) / PAGE_SIZE;
470 if (mddev->queue->backing_dev_info.ra_pages < 2* stripe)
471 mddev->queue->backing_dev_info.ra_pages = 2* stripe;
476 ret = md_integrity_register(mddev);
481 static void raid0_free(struct mddev *mddev, void *priv)
483 struct r0conf *conf = priv;
485 kfree(conf->strip_zone);
486 kfree(conf->devlist);
491 * Is io distribute over 1 or more chunks ?
493 static inline int is_io_in_chunk_boundary(struct mddev *mddev,
494 unsigned int chunk_sects, struct bio *bio)
496 if (likely(is_power_of_2(chunk_sects))) {
497 return chunk_sects >=
498 ((bio->bi_iter.bi_sector & (chunk_sects-1))
501 sector_t sector = bio->bi_iter.bi_sector;
502 return chunk_sects >= (sector_div(sector, chunk_sects)
507 static void raid0_make_request(struct mddev *mddev, struct bio *bio)
509 struct strip_zone *zone;
510 struct md_rdev *tmp_dev;
513 if (unlikely(bio->bi_rw & REQ_FLUSH)) {
514 md_flush_request(mddev, bio);
519 sector_t sector = bio->bi_iter.bi_sector;
520 unsigned chunk_sects = mddev->chunk_sectors;
522 unsigned sectors = chunk_sects -
523 (likely(is_power_of_2(chunk_sects))
524 ? (sector & (chunk_sects-1))
525 : sector_div(sector, chunk_sects));
527 if (sectors < bio_sectors(bio)) {
528 split = bio_split(bio, sectors, GFP_NOIO, fs_bio_set);
529 bio_chain(split, bio);
534 sector = bio->bi_iter.bi_sector;
535 zone = find_zone(mddev->private, §or);
536 tmp_dev = map_sector(mddev, zone, sector, §or);
537 split->bi_bdev = tmp_dev->bdev;
538 split->bi_iter.bi_sector = sector + zone->dev_start +
539 tmp_dev->data_offset;
541 if (unlikely((split->bi_rw & REQ_DISCARD) &&
542 !blk_queue_discard(bdev_get_queue(split->bi_bdev)))) {
546 generic_make_request(split);
547 } while (split != bio);
550 static void raid0_status(struct seq_file *seq, struct mddev *mddev)
552 seq_printf(seq, " %dk chunks", mddev->chunk_sectors / 2);
556 static void *raid0_takeover_raid45(struct mddev *mddev)
558 struct md_rdev *rdev;
559 struct r0conf *priv_conf;
561 if (mddev->degraded != 1) {
562 printk(KERN_ERR "md/raid0:%s: raid5 must be degraded! Degraded disks: %d\n",
565 return ERR_PTR(-EINVAL);
568 rdev_for_each(rdev, mddev) {
569 /* check slot number for a disk */
570 if (rdev->raid_disk == mddev->raid_disks-1) {
571 printk(KERN_ERR "md/raid0:%s: raid5 must have missing parity disk!\n",
573 return ERR_PTR(-EINVAL);
575 rdev->sectors = mddev->dev_sectors;
578 /* Set new parameters */
579 mddev->new_level = 0;
580 mddev->new_layout = 0;
581 mddev->new_chunk_sectors = mddev->chunk_sectors;
583 mddev->delta_disks = -1;
584 /* make sure it will be not marked as dirty */
585 mddev->recovery_cp = MaxSector;
587 create_strip_zones(mddev, &priv_conf);
591 static void *raid0_takeover_raid10(struct mddev *mddev)
593 struct r0conf *priv_conf;
596 * - far_copies must be 1
597 * - near_copies must be 2
598 * - disks number must be even
599 * - all mirrors must be already degraded
601 if (mddev->layout != ((1 << 8) + 2)) {
602 printk(KERN_ERR "md/raid0:%s:: Raid0 cannot takover layout: 0x%x\n",
605 return ERR_PTR(-EINVAL);
607 if (mddev->raid_disks & 1) {
608 printk(KERN_ERR "md/raid0:%s: Raid0 cannot takover Raid10 with odd disk number.\n",
610 return ERR_PTR(-EINVAL);
612 if (mddev->degraded != (mddev->raid_disks>>1)) {
613 printk(KERN_ERR "md/raid0:%s: All mirrors must be already degraded!\n",
615 return ERR_PTR(-EINVAL);
618 /* Set new parameters */
619 mddev->new_level = 0;
620 mddev->new_layout = 0;
621 mddev->new_chunk_sectors = mddev->chunk_sectors;
622 mddev->delta_disks = - mddev->raid_disks / 2;
623 mddev->raid_disks += mddev->delta_disks;
625 /* make sure it will be not marked as dirty */
626 mddev->recovery_cp = MaxSector;
628 create_strip_zones(mddev, &priv_conf);
632 static void *raid0_takeover_raid1(struct mddev *mddev)
634 struct r0conf *priv_conf;
638 * - (N - 1) mirror drives must be already faulty
640 if ((mddev->raid_disks - 1) != mddev->degraded) {
641 printk(KERN_ERR "md/raid0:%s: (N - 1) mirrors drives must be already faulty!\n",
643 return ERR_PTR(-EINVAL);
647 * a raid1 doesn't have the notion of chunk size, so
648 * figure out the largest suitable size we can use.
650 chunksect = 64 * 2; /* 64K by default */
652 /* The array must be an exact multiple of chunksize */
653 while (chunksect && (mddev->array_sectors & (chunksect - 1)))
656 if ((chunksect << 9) < PAGE_SIZE)
657 /* array size does not allow a suitable chunk size */
658 return ERR_PTR(-EINVAL);
660 /* Set new parameters */
661 mddev->new_level = 0;
662 mddev->new_layout = 0;
663 mddev->new_chunk_sectors = chunksect;
664 mddev->chunk_sectors = chunksect;
665 mddev->delta_disks = 1 - mddev->raid_disks;
666 mddev->raid_disks = 1;
667 /* make sure it will be not marked as dirty */
668 mddev->recovery_cp = MaxSector;
670 create_strip_zones(mddev, &priv_conf);
674 static void *raid0_takeover(struct mddev *mddev)
676 /* raid0 can take over:
677 * raid4 - if all data disks are active.
678 * raid5 - providing it is Raid4 layout and one disk is faulty
679 * raid10 - assuming we have all necessary active disks
680 * raid1 - with (N -1) mirror drives faulty
684 printk(KERN_ERR "md/raid0: %s: cannot takeover array with bitmap\n",
686 return ERR_PTR(-EBUSY);
688 if (mddev->level == 4)
689 return raid0_takeover_raid45(mddev);
691 if (mddev->level == 5) {
692 if (mddev->layout == ALGORITHM_PARITY_N)
693 return raid0_takeover_raid45(mddev);
695 printk(KERN_ERR "md/raid0:%s: Raid can only takeover Raid5 with layout: %d\n",
696 mdname(mddev), ALGORITHM_PARITY_N);
699 if (mddev->level == 10)
700 return raid0_takeover_raid10(mddev);
702 if (mddev->level == 1)
703 return raid0_takeover_raid1(mddev);
705 printk(KERN_ERR "Takeover from raid%i to raid0 not supported\n",
708 return ERR_PTR(-EINVAL);
711 static void raid0_quiesce(struct mddev *mddev, int state)
715 static struct md_personality raid0_personality=
719 .owner = THIS_MODULE,
720 .make_request = raid0_make_request,
723 .status = raid0_status,
725 .takeover = raid0_takeover,
726 .quiesce = raid0_quiesce,
727 .congested = raid0_congested,
728 .mergeable_bvec = raid0_mergeable_bvec,
731 static int __init raid0_init (void)
733 return register_md_personality (&raid0_personality);
736 static void raid0_exit (void)
738 unregister_md_personality (&raid0_personality);
741 module_init(raid0_init);
742 module_exit(raid0_exit);
743 MODULE_LICENSE("GPL");
744 MODULE_DESCRIPTION("RAID0 (striping) personality for MD");
745 MODULE_ALIAS("md-personality-2"); /* RAID0 */
746 MODULE_ALIAS("md-raid0");
747 MODULE_ALIAS("md-level-0");