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);
94 rdev_for_each(rdev1, mddev) {
95 pr_debug("md/raid0:%s: looking at %s\n",
97 bdevname(rdev1->bdev, b));
100 /* round size to chunk_size */
101 sectors = rdev1->sectors;
102 sector_div(sectors, mddev->chunk_sectors);
103 rdev1->sectors = sectors * mddev->chunk_sectors;
105 rdev_for_each(rdev2, mddev) {
106 pr_debug("md/raid0:%s: comparing %s(%llu)"
109 bdevname(rdev1->bdev,b),
110 (unsigned long long)rdev1->sectors,
111 bdevname(rdev2->bdev,b2),
112 (unsigned long long)rdev2->sectors);
113 if (rdev2 == rdev1) {
114 pr_debug("md/raid0:%s: END\n",
118 if (rdev2->sectors == rdev1->sectors) {
120 * Not unique, don't count it as a new
123 pr_debug("md/raid0:%s: EQUAL\n",
128 pr_debug("md/raid0:%s: NOT EQUAL\n",
132 pr_debug("md/raid0:%s: ==> UNIQUE\n",
134 conf->nr_strip_zones++;
135 pr_debug("md/raid0:%s: %d zones\n",
136 mdname(mddev), conf->nr_strip_zones);
139 pr_debug("md/raid0:%s: FINAL %d zones\n",
140 mdname(mddev), conf->nr_strip_zones);
142 conf->strip_zone = kzalloc(sizeof(struct strip_zone)*
143 conf->nr_strip_zones, GFP_KERNEL);
144 if (!conf->strip_zone)
146 conf->devlist = kzalloc(sizeof(struct md_rdev*)*
147 conf->nr_strip_zones*mddev->raid_disks,
152 /* The first zone must contain all devices, so here we check that
153 * there is a proper alignment of slots to devices and find them all
155 zone = &conf->strip_zone[0];
160 rdev_for_each(rdev1, mddev) {
161 int j = rdev1->raid_disk;
163 if (mddev->level == 10) {
164 /* taking over a raid10-n2 array */
166 rdev1->new_raid_disk = j;
169 if (mddev->level == 1) {
170 /* taiking over a raid1 array-
171 * we have only one active disk
174 rdev1->new_raid_disk = j;
177 if (j < 0 || j >= mddev->raid_disks) {
178 printk(KERN_ERR "md/raid0:%s: bad disk number %d - "
179 "aborting!\n", mdname(mddev), j);
183 printk(KERN_ERR "md/raid0:%s: multiple devices for %d - "
184 "aborting!\n", mdname(mddev), j);
189 disk_stack_limits(mddev->gendisk, rdev1->bdev,
190 rdev1->data_offset << 9);
192 if (rdev1->bdev->bd_disk->queue->merge_bvec_fn)
193 conf->has_merge_bvec = 1;
195 if (!smallest || (rdev1->sectors < smallest->sectors))
199 if (cnt != mddev->raid_disks) {
200 printk(KERN_ERR "md/raid0:%s: too few disks (%d of %d) - "
201 "aborting!\n", mdname(mddev), cnt, mddev->raid_disks);
205 zone->zone_end = smallest->sectors * cnt;
207 curr_zone_end = zone->zone_end;
209 /* now do the other zones */
210 for (i = 1; i < conf->nr_strip_zones; i++)
214 zone = conf->strip_zone + i;
215 dev = conf->devlist + i * mddev->raid_disks;
217 pr_debug("md/raid0:%s: zone %d\n", mdname(mddev), i);
218 zone->dev_start = smallest->sectors;
222 for (j=0; j<cnt; j++) {
223 rdev = conf->devlist[j];
224 if (rdev->sectors <= zone->dev_start) {
225 pr_debug("md/raid0:%s: checking %s ... nope\n",
227 bdevname(rdev->bdev, b));
230 pr_debug("md/raid0:%s: checking %s ..."
231 " contained as device %d\n",
233 bdevname(rdev->bdev, b), c);
236 if (!smallest || rdev->sectors < smallest->sectors) {
238 pr_debug("md/raid0:%s: (%llu) is smallest!.\n",
240 (unsigned long long)rdev->sectors);
245 sectors = (smallest->sectors - zone->dev_start) * c;
246 pr_debug("md/raid0:%s: zone->nb_dev: %d, sectors: %llu\n",
248 zone->nb_dev, (unsigned long long)sectors);
250 curr_zone_end += sectors;
251 zone->zone_end = curr_zone_end;
253 pr_debug("md/raid0:%s: current zone start: %llu\n",
255 (unsigned long long)smallest->sectors);
257 mddev->queue->backing_dev_info.congested_fn = raid0_congested;
258 mddev->queue->backing_dev_info.congested_data = mddev;
261 * now since we have the hard sector sizes, we can make sure
262 * chunk size is a multiple of that sector size
264 if ((mddev->chunk_sectors << 9) % queue_logical_block_size(mddev->queue)) {
265 printk(KERN_ERR "md/raid0:%s: chunk_size of %d not valid\n",
267 mddev->chunk_sectors << 9);
271 blk_queue_io_min(mddev->queue, mddev->chunk_sectors << 9);
272 blk_queue_io_opt(mddev->queue,
273 (mddev->chunk_sectors << 9) * mddev->raid_disks);
275 pr_debug("md/raid0:%s: done.\n", mdname(mddev));
276 *private_conf = conf;
280 kfree(conf->strip_zone);
281 kfree(conf->devlist);
283 *private_conf = NULL;
287 /* Find the zone which holds a particular offset
288 * Update *sectorp to be an offset in that zone
290 static struct strip_zone *find_zone(struct r0conf *conf,
294 struct strip_zone *z = conf->strip_zone;
295 sector_t sector = *sectorp;
297 for (i = 0; i < conf->nr_strip_zones; i++)
298 if (sector < z[i].zone_end) {
300 *sectorp = sector - z[i-1].zone_end;
307 * remaps the bio to the target device. we separate two flows.
308 * power 2 flow and a general flow for the sake of perfromance
310 static struct md_rdev *map_sector(struct mddev *mddev, struct strip_zone *zone,
311 sector_t sector, sector_t *sector_offset)
313 unsigned int sect_in_chunk;
315 struct r0conf *conf = mddev->private;
316 int raid_disks = conf->strip_zone[0].nb_dev;
317 unsigned int chunk_sects = mddev->chunk_sectors;
319 if (is_power_of_2(chunk_sects)) {
320 int chunksect_bits = ffz(~chunk_sects);
321 /* find the sector offset inside the chunk */
322 sect_in_chunk = sector & (chunk_sects - 1);
323 sector >>= chunksect_bits;
325 chunk = *sector_offset;
326 /* quotient is the chunk in real device*/
327 sector_div(chunk, zone->nb_dev << chunksect_bits);
329 sect_in_chunk = sector_div(sector, chunk_sects);
330 chunk = *sector_offset;
331 sector_div(chunk, chunk_sects * zone->nb_dev);
334 * position the bio over the real device
335 * real sector = chunk in device + starting of zone
336 * + the position in the chunk
338 *sector_offset = (chunk * chunk_sects) + sect_in_chunk;
339 return conf->devlist[(zone - conf->strip_zone)*raid_disks
340 + sector_div(sector, zone->nb_dev)];
344 * raid0_mergeable_bvec -- tell bio layer if two requests can be merged
346 * @bvm: properties of new bio
347 * @biovec: the request that could be merged to it.
349 * Return amount of bytes we can accept at this offset
351 static int raid0_mergeable_bvec(struct request_queue *q,
352 struct bvec_merge_data *bvm,
353 struct bio_vec *biovec)
355 struct mddev *mddev = q->queuedata;
356 struct r0conf *conf = mddev->private;
357 sector_t sector = bvm->bi_sector + get_start_sect(bvm->bi_bdev);
358 sector_t sector_offset = sector;
360 unsigned int chunk_sectors = mddev->chunk_sectors;
361 unsigned int bio_sectors = bvm->bi_size >> 9;
362 struct strip_zone *zone;
363 struct md_rdev *rdev;
364 struct request_queue *subq;
366 if (is_power_of_2(chunk_sectors))
367 max = (chunk_sectors - ((sector & (chunk_sectors-1))
368 + bio_sectors)) << 9;
370 max = (chunk_sectors - (sector_div(sector, chunk_sectors)
371 + bio_sectors)) << 9;
373 max = 0; /* bio_add cannot handle a negative return */
374 if (max <= biovec->bv_len && bio_sectors == 0)
375 return biovec->bv_len;
376 if (max < biovec->bv_len)
377 /* too small already, no need to check further */
379 if (!conf->has_merge_bvec)
382 /* May need to check subordinate device */
383 sector = sector_offset;
384 zone = find_zone(mddev->private, §or_offset);
385 rdev = map_sector(mddev, zone, sector, §or_offset);
386 subq = bdev_get_queue(rdev->bdev);
387 if (subq->merge_bvec_fn) {
388 bvm->bi_bdev = rdev->bdev;
389 bvm->bi_sector = sector_offset + zone->dev_start +
391 return min(max, subq->merge_bvec_fn(subq, bvm, biovec));
396 static sector_t raid0_size(struct mddev *mddev, sector_t sectors, int raid_disks)
398 sector_t array_sectors = 0;
399 struct md_rdev *rdev;
401 WARN_ONCE(sectors || raid_disks,
402 "%s does not support generic reshape\n", __func__);
404 rdev_for_each(rdev, mddev)
405 array_sectors += rdev->sectors;
407 return array_sectors;
410 static int raid0_stop(struct mddev *mddev);
412 static int raid0_run(struct mddev *mddev)
417 if (mddev->chunk_sectors == 0) {
418 printk(KERN_ERR "md/raid0:%s: chunk size must be set.\n",
422 if (md_check_no_bitmap(mddev))
424 blk_queue_max_hw_sectors(mddev->queue, mddev->chunk_sectors);
425 blk_queue_max_write_same_sectors(mddev->queue, mddev->chunk_sectors);
427 /* if private is not null, we are here after takeover */
428 if (mddev->private == NULL) {
429 ret = create_strip_zones(mddev, &conf);
432 mddev->private = conf;
434 conf = mddev->private;
436 /* calculate array device size */
437 md_set_array_sectors(mddev, raid0_size(mddev, 0, 0));
439 printk(KERN_INFO "md/raid0:%s: md_size is %llu sectors.\n",
441 (unsigned long long)mddev->array_sectors);
442 /* calculate the max read-ahead size.
443 * For read-ahead of large files to be effective, we need to
444 * readahead at least twice a whole stripe. i.e. number of devices
445 * multiplied by chunk size times 2.
446 * If an individual device has an ra_pages greater than the
447 * chunk size, then we will not drive that device as hard as it
448 * wants. We consider this a configuration error: a larger
449 * chunksize should be used in that case.
452 int stripe = mddev->raid_disks *
453 (mddev->chunk_sectors << 9) / PAGE_SIZE;
454 if (mddev->queue->backing_dev_info.ra_pages < 2* stripe)
455 mddev->queue->backing_dev_info.ra_pages = 2* stripe;
458 blk_queue_merge_bvec(mddev->queue, raid0_mergeable_bvec);
461 ret = md_integrity_register(mddev);
468 static int raid0_stop(struct mddev *mddev)
470 struct r0conf *conf = mddev->private;
472 blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/
473 kfree(conf->strip_zone);
474 kfree(conf->devlist);
476 mddev->private = NULL;
481 * Is io distribute over 1 or more chunks ?
483 static inline int is_io_in_chunk_boundary(struct mddev *mddev,
484 unsigned int chunk_sects, struct bio *bio)
486 if (likely(is_power_of_2(chunk_sects))) {
487 return chunk_sects >= ((bio->bi_sector & (chunk_sects-1))
488 + (bio->bi_size >> 9));
490 sector_t sector = bio->bi_sector;
491 return chunk_sects >= (sector_div(sector, chunk_sects)
492 + (bio->bi_size >> 9));
496 static void raid0_make_request(struct mddev *mddev, struct bio *bio)
498 unsigned int chunk_sects;
499 sector_t sector_offset;
500 struct strip_zone *zone;
501 struct md_rdev *tmp_dev;
503 if (unlikely(bio->bi_rw & REQ_FLUSH)) {
504 md_flush_request(mddev, bio);
508 chunk_sects = mddev->chunk_sectors;
509 if (unlikely(!is_io_in_chunk_boundary(mddev, chunk_sects, bio))) {
510 sector_t sector = bio->bi_sector;
512 /* Sanity check -- queue functions should prevent this happening */
513 if (bio->bi_vcnt != 1 ||
516 /* This is a one page bio that upper layers
517 * refuse to split for us, so we need to split it.
519 if (likely(is_power_of_2(chunk_sects)))
520 bp = bio_split(bio, chunk_sects - (sector &
523 bp = bio_split(bio, chunk_sects -
524 sector_div(sector, chunk_sects));
525 raid0_make_request(mddev, &bp->bio1);
526 raid0_make_request(mddev, &bp->bio2);
527 bio_pair_release(bp);
531 sector_offset = bio->bi_sector;
532 zone = find_zone(mddev->private, §or_offset);
533 tmp_dev = map_sector(mddev, zone, bio->bi_sector,
535 bio->bi_bdev = tmp_dev->bdev;
536 bio->bi_sector = sector_offset + zone->dev_start +
537 tmp_dev->data_offset;
539 generic_make_request(bio);
543 printk("md/raid0:%s: make_request bug: can't convert block across chunks"
544 " or bigger than %dk %llu %d\n",
545 mdname(mddev), chunk_sects / 2,
546 (unsigned long long)bio->bi_sector, bio->bi_size >> 10);
552 static void raid0_status(struct seq_file *seq, struct mddev *mddev)
554 seq_printf(seq, " %dk chunks", mddev->chunk_sectors / 2);
558 static void *raid0_takeover_raid45(struct mddev *mddev)
560 struct md_rdev *rdev;
561 struct r0conf *priv_conf;
563 if (mddev->degraded != 1) {
564 printk(KERN_ERR "md/raid0:%s: raid5 must be degraded! Degraded disks: %d\n",
567 return ERR_PTR(-EINVAL);
570 rdev_for_each(rdev, mddev) {
571 /* check slot number for a disk */
572 if (rdev->raid_disk == mddev->raid_disks-1) {
573 printk(KERN_ERR "md/raid0:%s: raid5 must have missing parity disk!\n",
575 return ERR_PTR(-EINVAL);
579 /* Set new parameters */
580 mddev->new_level = 0;
581 mddev->new_layout = 0;
582 mddev->new_chunk_sectors = mddev->chunk_sectors;
584 mddev->delta_disks = -1;
585 /* make sure it will be not marked as dirty */
586 mddev->recovery_cp = MaxSector;
588 create_strip_zones(mddev, &priv_conf);
592 static void *raid0_takeover_raid10(struct mddev *mddev)
594 struct r0conf *priv_conf;
597 * - far_copies must be 1
598 * - near_copies must be 2
599 * - disks number must be even
600 * - all mirrors must be already degraded
602 if (mddev->layout != ((1 << 8) + 2)) {
603 printk(KERN_ERR "md/raid0:%s:: Raid0 cannot takover layout: 0x%x\n",
606 return ERR_PTR(-EINVAL);
608 if (mddev->raid_disks & 1) {
609 printk(KERN_ERR "md/raid0:%s: Raid0 cannot takover Raid10 with odd disk number.\n",
611 return ERR_PTR(-EINVAL);
613 if (mddev->degraded != (mddev->raid_disks>>1)) {
614 printk(KERN_ERR "md/raid0:%s: All mirrors must be already degraded!\n",
616 return ERR_PTR(-EINVAL);
619 /* Set new parameters */
620 mddev->new_level = 0;
621 mddev->new_layout = 0;
622 mddev->new_chunk_sectors = mddev->chunk_sectors;
623 mddev->delta_disks = - mddev->raid_disks / 2;
624 mddev->raid_disks += mddev->delta_disks;
626 /* make sure it will be not marked as dirty */
627 mddev->recovery_cp = MaxSector;
629 create_strip_zones(mddev, &priv_conf);
633 static void *raid0_takeover_raid1(struct mddev *mddev)
635 struct r0conf *priv_conf;
639 * - (N - 1) mirror drives must be already faulty
641 if ((mddev->raid_disks - 1) != mddev->degraded) {
642 printk(KERN_ERR "md/raid0:%s: (N - 1) mirrors drives must be already faulty!\n",
644 return ERR_PTR(-EINVAL);
648 * a raid1 doesn't have the notion of chunk size, so
649 * figure out the largest suitable size we can use.
651 chunksect = 64 * 2; /* 64K by default */
653 /* The array must be an exact multiple of chunksize */
654 while (chunksect && (mddev->array_sectors & (chunksect - 1)))
657 if ((chunksect << 9) < PAGE_SIZE)
658 /* array size does not allow a suitable chunk size */
659 return ERR_PTR(-EINVAL);
661 /* Set new parameters */
662 mddev->new_level = 0;
663 mddev->new_layout = 0;
664 mddev->new_chunk_sectors = chunksect;
665 mddev->chunk_sectors = chunksect;
666 mddev->delta_disks = 1 - mddev->raid_disks;
667 mddev->raid_disks = 1;
668 /* make sure it will be not marked as dirty */
669 mddev->recovery_cp = MaxSector;
671 create_strip_zones(mddev, &priv_conf);
675 static void *raid0_takeover(struct mddev *mddev)
677 /* raid0 can take over:
678 * raid4 - if all data disks are active.
679 * raid5 - providing it is Raid4 layout and one disk is faulty
680 * raid10 - assuming we have all necessary active disks
681 * raid1 - with (N -1) mirror drives faulty
683 if (mddev->level == 4)
684 return raid0_takeover_raid45(mddev);
686 if (mddev->level == 5) {
687 if (mddev->layout == ALGORITHM_PARITY_N)
688 return raid0_takeover_raid45(mddev);
690 printk(KERN_ERR "md/raid0:%s: Raid can only takeover Raid5 with layout: %d\n",
691 mdname(mddev), ALGORITHM_PARITY_N);
694 if (mddev->level == 10)
695 return raid0_takeover_raid10(mddev);
697 if (mddev->level == 1)
698 return raid0_takeover_raid1(mddev);
700 printk(KERN_ERR "Takeover from raid%i to raid0 not supported\n",
703 return ERR_PTR(-EINVAL);
706 static void raid0_quiesce(struct mddev *mddev, int state)
710 static struct md_personality raid0_personality=
714 .owner = THIS_MODULE,
715 .make_request = raid0_make_request,
718 .status = raid0_status,
720 .takeover = raid0_takeover,
721 .quiesce = raid0_quiesce,
724 static int __init raid0_init (void)
726 return register_md_personality (&raid0_personality);
729 static void raid0_exit (void)
731 unregister_md_personality (&raid0_personality);
734 module_init(raid0_init);
735 module_exit(raid0_exit);
736 MODULE_LICENSE("GPL");
737 MODULE_DESCRIPTION("RAID0 (striping) personality for MD");
738 MODULE_ALIAS("md-personality-2"); /* RAID0 */
739 MODULE_ALIAS("md-raid0");
740 MODULE_ALIAS("md-level-0");