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;
178 if (j < 0 || j >= mddev->raid_disks) {
179 printk(KERN_ERR "md/raid0:%s: bad disk number %d - "
180 "aborting!\n", mdname(mddev), j);
184 printk(KERN_ERR "md/raid0:%s: multiple devices for %d - "
185 "aborting!\n", mdname(mddev), j);
190 disk_stack_limits(mddev->gendisk, rdev1->bdev,
191 rdev1->data_offset << 9);
193 if (rdev1->bdev->bd_disk->queue->merge_bvec_fn)
194 conf->has_merge_bvec = 1;
196 if (!smallest || (rdev1->sectors < smallest->sectors))
200 if (blk_queue_discard(bdev_get_queue(rdev1->bdev)))
201 discard_supported = true;
203 if (cnt != mddev->raid_disks) {
204 printk(KERN_ERR "md/raid0:%s: too few disks (%d of %d) - "
205 "aborting!\n", mdname(mddev), cnt, mddev->raid_disks);
209 zone->zone_end = smallest->sectors * cnt;
211 curr_zone_end = zone->zone_end;
213 /* now do the other zones */
214 for (i = 1; i < conf->nr_strip_zones; i++)
218 zone = conf->strip_zone + i;
219 dev = conf->devlist + i * mddev->raid_disks;
221 pr_debug("md/raid0:%s: zone %d\n", mdname(mddev), i);
222 zone->dev_start = smallest->sectors;
226 for (j=0; j<cnt; j++) {
227 rdev = conf->devlist[j];
228 if (rdev->sectors <= zone->dev_start) {
229 pr_debug("md/raid0:%s: checking %s ... nope\n",
231 bdevname(rdev->bdev, b));
234 pr_debug("md/raid0:%s: checking %s ..."
235 " contained as device %d\n",
237 bdevname(rdev->bdev, b), c);
240 if (!smallest || rdev->sectors < smallest->sectors) {
242 pr_debug("md/raid0:%s: (%llu) is smallest!.\n",
244 (unsigned long long)rdev->sectors);
249 sectors = (smallest->sectors - zone->dev_start) * c;
250 pr_debug("md/raid0:%s: zone->nb_dev: %d, sectors: %llu\n",
252 zone->nb_dev, (unsigned long long)sectors);
254 curr_zone_end += sectors;
255 zone->zone_end = curr_zone_end;
257 pr_debug("md/raid0:%s: current zone start: %llu\n",
259 (unsigned long long)smallest->sectors);
261 mddev->queue->backing_dev_info.congested_fn = raid0_congested;
262 mddev->queue->backing_dev_info.congested_data = mddev;
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);
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);
284 pr_debug("md/raid0:%s: done.\n", mdname(mddev));
285 *private_conf = conf;
289 kfree(conf->strip_zone);
290 kfree(conf->devlist);
292 *private_conf = NULL;
296 /* Find the zone which holds a particular offset
297 * Update *sectorp to be an offset in that zone
299 static struct strip_zone *find_zone(struct r0conf *conf,
303 struct strip_zone *z = conf->strip_zone;
304 sector_t sector = *sectorp;
306 for (i = 0; i < conf->nr_strip_zones; i++)
307 if (sector < z[i].zone_end) {
309 *sectorp = sector - z[i-1].zone_end;
316 * remaps the bio to the target device. we separate two flows.
317 * power 2 flow and a general flow for the sake of perfromance
319 static struct md_rdev *map_sector(struct mddev *mddev, struct strip_zone *zone,
320 sector_t sector, sector_t *sector_offset)
322 unsigned int sect_in_chunk;
324 struct r0conf *conf = mddev->private;
325 int raid_disks = conf->strip_zone[0].nb_dev;
326 unsigned int chunk_sects = mddev->chunk_sectors;
328 if (is_power_of_2(chunk_sects)) {
329 int chunksect_bits = ffz(~chunk_sects);
330 /* find the sector offset inside the chunk */
331 sect_in_chunk = sector & (chunk_sects - 1);
332 sector >>= chunksect_bits;
334 chunk = *sector_offset;
335 /* quotient is the chunk in real device*/
336 sector_div(chunk, zone->nb_dev << chunksect_bits);
338 sect_in_chunk = sector_div(sector, chunk_sects);
339 chunk = *sector_offset;
340 sector_div(chunk, chunk_sects * zone->nb_dev);
343 * position the bio over the real device
344 * real sector = chunk in device + starting of zone
345 * + the position in the chunk
347 *sector_offset = (chunk * chunk_sects) + sect_in_chunk;
348 return conf->devlist[(zone - conf->strip_zone)*raid_disks
349 + sector_div(sector, zone->nb_dev)];
353 * raid0_mergeable_bvec -- tell bio layer if two requests can be merged
355 * @bvm: properties of new bio
356 * @biovec: the request that could be merged to it.
358 * Return amount of bytes we can accept at this offset
360 static int raid0_mergeable_bvec(struct request_queue *q,
361 struct bvec_merge_data *bvm,
362 struct bio_vec *biovec)
364 struct mddev *mddev = q->queuedata;
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;
416 return array_sectors;
419 static int raid0_stop(struct mddev *mddev);
421 static int raid0_run(struct mddev *mddev)
426 if (mddev->chunk_sectors == 0) {
427 printk(KERN_ERR "md/raid0:%s: chunk size must be set.\n",
431 if (md_check_no_bitmap(mddev))
433 blk_queue_max_hw_sectors(mddev->queue, mddev->chunk_sectors);
434 blk_queue_max_write_same_sectors(mddev->queue, mddev->chunk_sectors);
435 blk_queue_max_discard_sectors(mddev->queue, mddev->chunk_sectors);
437 /* if private is not null, we are here after takeover */
438 if (mddev->private == NULL) {
439 ret = create_strip_zones(mddev, &conf);
442 mddev->private = conf;
444 conf = mddev->private;
446 /* calculate array device size */
447 md_set_array_sectors(mddev, raid0_size(mddev, 0, 0));
449 printk(KERN_INFO "md/raid0:%s: md_size is %llu sectors.\n",
451 (unsigned long long)mddev->array_sectors);
452 /* calculate the max read-ahead size.
453 * For read-ahead of large files to be effective, we need to
454 * readahead at least twice a whole stripe. i.e. number of devices
455 * multiplied by chunk size times 2.
456 * If an individual device has an ra_pages greater than the
457 * chunk size, then we will not drive that device as hard as it
458 * wants. We consider this a configuration error: a larger
459 * chunksize should be used in that case.
462 int stripe = mddev->raid_disks *
463 (mddev->chunk_sectors << 9) / PAGE_SIZE;
464 if (mddev->queue->backing_dev_info.ra_pages < 2* stripe)
465 mddev->queue->backing_dev_info.ra_pages = 2* stripe;
468 blk_queue_merge_bvec(mddev->queue, raid0_mergeable_bvec);
471 ret = md_integrity_register(mddev);
478 static int raid0_stop(struct mddev *mddev)
480 struct r0conf *conf = mddev->private;
482 blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/
483 kfree(conf->strip_zone);
484 kfree(conf->devlist);
486 mddev->private = NULL;
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 >= ((bio->bi_sector & (chunk_sects-1))
498 + (bio->bi_size >> 9));
500 sector_t sector = bio->bi_sector;
501 return chunk_sects >= (sector_div(sector, chunk_sects)
502 + (bio->bi_size >> 9));
506 static void raid0_make_request(struct mddev *mddev, struct bio *bio)
508 unsigned int chunk_sects;
509 sector_t sector_offset;
510 struct strip_zone *zone;
511 struct md_rdev *tmp_dev;
513 if (unlikely(bio->bi_rw & REQ_FLUSH)) {
514 md_flush_request(mddev, bio);
518 chunk_sects = mddev->chunk_sectors;
519 if (unlikely(!is_io_in_chunk_boundary(mddev, chunk_sects, bio))) {
520 sector_t sector = bio->bi_sector;
522 /* Sanity check -- queue functions should prevent this happening */
523 if ((bio->bi_vcnt != 1 && bio->bi_vcnt != 0) ||
526 /* This is a one page bio that upper layers
527 * refuse to split for us, so we need to split it.
529 if (likely(is_power_of_2(chunk_sects)))
530 bp = bio_split(bio, chunk_sects - (sector &
533 bp = bio_split(bio, chunk_sects -
534 sector_div(sector, chunk_sects));
535 raid0_make_request(mddev, &bp->bio1);
536 raid0_make_request(mddev, &bp->bio2);
537 bio_pair_release(bp);
541 sector_offset = bio->bi_sector;
542 zone = find_zone(mddev->private, §or_offset);
543 tmp_dev = map_sector(mddev, zone, bio->bi_sector,
545 bio->bi_bdev = tmp_dev->bdev;
546 bio->bi_sector = sector_offset + zone->dev_start +
547 tmp_dev->data_offset;
549 if (unlikely((bio->bi_rw & REQ_DISCARD) &&
550 !blk_queue_discard(bdev_get_queue(bio->bi_bdev)))) {
556 generic_make_request(bio);
560 printk("md/raid0:%s: make_request bug: can't convert block across chunks"
561 " or bigger than %dk %llu %d\n",
562 mdname(mddev), chunk_sects / 2,
563 (unsigned long long)bio->bi_sector, bio->bi_size >> 10);
569 static void raid0_status(struct seq_file *seq, struct mddev *mddev)
571 seq_printf(seq, " %dk chunks", mddev->chunk_sectors / 2);
575 static void *raid0_takeover_raid45(struct mddev *mddev)
577 struct md_rdev *rdev;
578 struct r0conf *priv_conf;
580 if (mddev->degraded != 1) {
581 printk(KERN_ERR "md/raid0:%s: raid5 must be degraded! Degraded disks: %d\n",
584 return ERR_PTR(-EINVAL);
587 rdev_for_each(rdev, mddev) {
588 /* check slot number for a disk */
589 if (rdev->raid_disk == mddev->raid_disks-1) {
590 printk(KERN_ERR "md/raid0:%s: raid5 must have missing parity disk!\n",
592 return ERR_PTR(-EINVAL);
596 /* Set new parameters */
597 mddev->new_level = 0;
598 mddev->new_layout = 0;
599 mddev->new_chunk_sectors = mddev->chunk_sectors;
601 mddev->delta_disks = -1;
602 /* make sure it will be not marked as dirty */
603 mddev->recovery_cp = MaxSector;
605 create_strip_zones(mddev, &priv_conf);
609 static void *raid0_takeover_raid10(struct mddev *mddev)
611 struct r0conf *priv_conf;
614 * - far_copies must be 1
615 * - near_copies must be 2
616 * - disks number must be even
617 * - all mirrors must be already degraded
619 if (mddev->layout != ((1 << 8) + 2)) {
620 printk(KERN_ERR "md/raid0:%s:: Raid0 cannot takover layout: 0x%x\n",
623 return ERR_PTR(-EINVAL);
625 if (mddev->raid_disks & 1) {
626 printk(KERN_ERR "md/raid0:%s: Raid0 cannot takover Raid10 with odd disk number.\n",
628 return ERR_PTR(-EINVAL);
630 if (mddev->degraded != (mddev->raid_disks>>1)) {
631 printk(KERN_ERR "md/raid0:%s: All mirrors must be already degraded!\n",
633 return ERR_PTR(-EINVAL);
636 /* Set new parameters */
637 mddev->new_level = 0;
638 mddev->new_layout = 0;
639 mddev->new_chunk_sectors = mddev->chunk_sectors;
640 mddev->delta_disks = - mddev->raid_disks / 2;
641 mddev->raid_disks += mddev->delta_disks;
643 /* make sure it will be not marked as dirty */
644 mddev->recovery_cp = MaxSector;
646 create_strip_zones(mddev, &priv_conf);
650 static void *raid0_takeover_raid1(struct mddev *mddev)
652 struct r0conf *priv_conf;
656 * - (N - 1) mirror drives must be already faulty
658 if ((mddev->raid_disks - 1) != mddev->degraded) {
659 printk(KERN_ERR "md/raid0:%s: (N - 1) mirrors drives must be already faulty!\n",
661 return ERR_PTR(-EINVAL);
665 * a raid1 doesn't have the notion of chunk size, so
666 * figure out the largest suitable size we can use.
668 chunksect = 64 * 2; /* 64K by default */
670 /* The array must be an exact multiple of chunksize */
671 while (chunksect && (mddev->array_sectors & (chunksect - 1)))
674 if ((chunksect << 9) < PAGE_SIZE)
675 /* array size does not allow a suitable chunk size */
676 return ERR_PTR(-EINVAL);
678 /* Set new parameters */
679 mddev->new_level = 0;
680 mddev->new_layout = 0;
681 mddev->new_chunk_sectors = chunksect;
682 mddev->chunk_sectors = chunksect;
683 mddev->delta_disks = 1 - mddev->raid_disks;
684 mddev->raid_disks = 1;
685 /* make sure it will be not marked as dirty */
686 mddev->recovery_cp = MaxSector;
688 create_strip_zones(mddev, &priv_conf);
692 static void *raid0_takeover(struct mddev *mddev)
694 /* raid0 can take over:
695 * raid4 - if all data disks are active.
696 * raid5 - providing it is Raid4 layout and one disk is faulty
697 * raid10 - assuming we have all necessary active disks
698 * raid1 - with (N -1) mirror drives faulty
700 if (mddev->level == 4)
701 return raid0_takeover_raid45(mddev);
703 if (mddev->level == 5) {
704 if (mddev->layout == ALGORITHM_PARITY_N)
705 return raid0_takeover_raid45(mddev);
707 printk(KERN_ERR "md/raid0:%s: Raid can only takeover Raid5 with layout: %d\n",
708 mdname(mddev), ALGORITHM_PARITY_N);
711 if (mddev->level == 10)
712 return raid0_takeover_raid10(mddev);
714 if (mddev->level == 1)
715 return raid0_takeover_raid1(mddev);
717 printk(KERN_ERR "Takeover from raid%i to raid0 not supported\n",
720 return ERR_PTR(-EINVAL);
723 static void raid0_quiesce(struct mddev *mddev, int state)
727 static struct md_personality raid0_personality=
731 .owner = THIS_MODULE,
732 .make_request = raid0_make_request,
735 .status = raid0_status,
737 .takeover = raid0_takeover,
738 .quiesce = raid0_quiesce,
741 static int __init raid0_init (void)
743 return register_md_personality (&raid0_personality);
746 static void raid0_exit (void)
748 unregister_md_personality (&raid0_personality);
751 module_init(raid0_init);
752 module_exit(raid0_exit);
753 MODULE_LICENSE("GPL");
754 MODULE_DESCRIPTION("RAID0 (striping) personality for MD");
755 MODULE_ALIAS("md-personality-2"); /* RAID0 */
756 MODULE_ALIAS("md-raid0");
757 MODULE_ALIAS("md-level-0");