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 unsigned short blksize = 512;
95 list_for_each_entry(rdev1, &mddev->disks, same_set) {
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 blksize = max(blksize, queue_logical_block_size(
107 rdev1->bdev->bd_disk->queue));
109 list_for_each_entry(rdev2, &mddev->disks, same_set) {
110 pr_debug("md/raid0:%s: comparing %s(%llu)"
113 bdevname(rdev1->bdev,b),
114 (unsigned long long)rdev1->sectors,
115 bdevname(rdev2->bdev,b2),
116 (unsigned long long)rdev2->sectors);
117 if (rdev2 == rdev1) {
118 pr_debug("md/raid0:%s: END\n",
122 if (rdev2->sectors == rdev1->sectors) {
124 * Not unique, don't count it as a new
127 pr_debug("md/raid0:%s: EQUAL\n",
132 pr_debug("md/raid0:%s: NOT EQUAL\n",
136 pr_debug("md/raid0:%s: ==> UNIQUE\n",
138 conf->nr_strip_zones++;
139 pr_debug("md/raid0:%s: %d zones\n",
140 mdname(mddev), conf->nr_strip_zones);
143 pr_debug("md/raid0:%s: FINAL %d zones\n",
144 mdname(mddev), conf->nr_strip_zones);
146 * now since we have the hard sector sizes, we can make sure
147 * chunk size is a multiple of that sector size
149 if ((mddev->chunk_sectors << 9) % blksize) {
150 printk(KERN_ERR "md/raid0:%s: chunk_size of %d not multiple of block size %d\n",
152 mddev->chunk_sectors << 9, blksize);
158 conf->strip_zone = kzalloc(sizeof(struct strip_zone)*
159 conf->nr_strip_zones, GFP_KERNEL);
160 if (!conf->strip_zone)
162 conf->devlist = kzalloc(sizeof(struct md_rdev*)*
163 conf->nr_strip_zones*mddev->raid_disks,
168 /* The first zone must contain all devices, so here we check that
169 * there is a proper alignment of slots to devices and find them all
171 zone = &conf->strip_zone[0];
176 list_for_each_entry(rdev1, &mddev->disks, same_set) {
177 int j = rdev1->raid_disk;
179 if (mddev->level == 10) {
180 /* taking over a raid10-n2 array */
182 rdev1->new_raid_disk = j;
185 if (mddev->level == 1) {
186 /* taiking over a raid1 array-
187 * we have only one active disk
190 rdev1->new_raid_disk = j;
193 if (j < 0 || j >= mddev->raid_disks) {
194 printk(KERN_ERR "md/raid0:%s: bad disk number %d - "
195 "aborting!\n", mdname(mddev), j);
199 printk(KERN_ERR "md/raid0:%s: multiple devices for %d - "
200 "aborting!\n", mdname(mddev), j);
205 /* as we don't honour merge_bvec_fn, we must never risk
206 * violating it, so limit ->max_segments to 1, lying within
210 if (rdev1->bdev->bd_disk->queue->merge_bvec_fn) {
211 blk_queue_max_segments(mddev->queue, 1);
212 blk_queue_segment_boundary(mddev->queue,
213 PAGE_CACHE_SIZE - 1);
215 if (!smallest || (rdev1->sectors < smallest->sectors))
219 if (cnt != mddev->raid_disks) {
220 printk(KERN_ERR "md/raid0:%s: too few disks (%d of %d) - "
221 "aborting!\n", mdname(mddev), cnt, mddev->raid_disks);
225 zone->zone_end = smallest->sectors * cnt;
227 curr_zone_end = zone->zone_end;
229 /* now do the other zones */
230 for (i = 1; i < conf->nr_strip_zones; i++)
234 zone = conf->strip_zone + i;
235 dev = conf->devlist + i * mddev->raid_disks;
237 pr_debug("md/raid0:%s: zone %d\n", mdname(mddev), i);
238 zone->dev_start = smallest->sectors;
242 for (j=0; j<cnt; j++) {
243 rdev = conf->devlist[j];
244 if (rdev->sectors <= zone->dev_start) {
245 pr_debug("md/raid0:%s: checking %s ... nope\n",
247 bdevname(rdev->bdev, b));
250 pr_debug("md/raid0:%s: checking %s ..."
251 " contained as device %d\n",
253 bdevname(rdev->bdev, b), c);
256 if (!smallest || rdev->sectors < smallest->sectors) {
258 pr_debug("md/raid0:%s: (%llu) is smallest!.\n",
260 (unsigned long long)rdev->sectors);
265 sectors = (smallest->sectors - zone->dev_start) * c;
266 pr_debug("md/raid0:%s: zone->nb_dev: %d, sectors: %llu\n",
268 zone->nb_dev, (unsigned long long)sectors);
270 curr_zone_end += sectors;
271 zone->zone_end = curr_zone_end;
273 pr_debug("md/raid0:%s: current zone start: %llu\n",
275 (unsigned long long)smallest->sectors);
277 mddev->queue->backing_dev_info.congested_fn = raid0_congested;
278 mddev->queue->backing_dev_info.congested_data = mddev;
280 pr_debug("md/raid0:%s: done.\n", mdname(mddev));
281 *private_conf = conf;
285 kfree(conf->strip_zone);
286 kfree(conf->devlist);
288 *private_conf = ERR_PTR(err);
293 * raid0_mergeable_bvec -- tell bio layer if a two requests can be merged
295 * @bvm: properties of new bio
296 * @biovec: the request that could be merged to it.
298 * Return amount of bytes we can accept at this offset
300 static int raid0_mergeable_bvec(struct request_queue *q,
301 struct bvec_merge_data *bvm,
302 struct bio_vec *biovec)
304 struct mddev *mddev = q->queuedata;
305 sector_t sector = bvm->bi_sector + get_start_sect(bvm->bi_bdev);
307 unsigned int chunk_sectors = mddev->chunk_sectors;
308 unsigned int bio_sectors = bvm->bi_size >> 9;
310 if (is_power_of_2(chunk_sectors))
311 max = (chunk_sectors - ((sector & (chunk_sectors-1))
312 + bio_sectors)) << 9;
314 max = (chunk_sectors - (sector_div(sector, chunk_sectors)
315 + bio_sectors)) << 9;
316 if (max < 0) max = 0; /* bio_add cannot handle a negative return */
317 if (max <= biovec->bv_len && bio_sectors == 0)
318 return biovec->bv_len;
323 static sector_t raid0_size(struct mddev *mddev, sector_t sectors, int raid_disks)
325 sector_t array_sectors = 0;
326 struct md_rdev *rdev;
328 WARN_ONCE(sectors || raid_disks,
329 "%s does not support generic reshape\n", __func__);
331 list_for_each_entry(rdev, &mddev->disks, same_set)
332 array_sectors += (rdev->sectors &
333 ~(sector_t)(mddev->chunk_sectors-1));
335 return array_sectors;
338 static int raid0_run(struct mddev *mddev)
342 struct md_rdev *rdev;
344 if (mddev->chunk_sectors == 0) {
345 printk(KERN_ERR "md/raid0:%s: chunk size must be set.\n",
349 if (md_check_no_bitmap(mddev))
352 /* if private is not null, we are here after takeover */
353 if (mddev->private == NULL) {
354 ret = create_strip_zones(mddev, &conf);
357 mddev->private = conf;
359 conf = mddev->private;
361 blk_queue_max_hw_sectors(mddev->queue, mddev->chunk_sectors);
363 blk_queue_io_min(mddev->queue, mddev->chunk_sectors << 9);
364 blk_queue_io_opt(mddev->queue,
365 (mddev->chunk_sectors << 9) * mddev->raid_disks);
367 list_for_each_entry(rdev, &mddev->disks, same_set) {
368 disk_stack_limits(mddev->gendisk, rdev->bdev,
369 rdev->data_offset << 9);
372 /* calculate array device size */
373 md_set_array_sectors(mddev, raid0_size(mddev, 0, 0));
375 printk(KERN_INFO "md/raid0:%s: md_size is %llu sectors.\n",
377 (unsigned long long)mddev->array_sectors);
378 /* calculate the max read-ahead size.
379 * For read-ahead of large files to be effective, we need to
380 * readahead at least twice a whole stripe. i.e. number of devices
381 * multiplied by chunk size times 2.
382 * If an individual device has an ra_pages greater than the
383 * chunk size, then we will not drive that device as hard as it
384 * wants. We consider this a configuration error: a larger
385 * chunksize should be used in that case.
388 int stripe = mddev->raid_disks *
389 (mddev->chunk_sectors << 9) / PAGE_SIZE;
390 if (mddev->queue->backing_dev_info.ra_pages < 2* stripe)
391 mddev->queue->backing_dev_info.ra_pages = 2* stripe;
394 blk_queue_merge_bvec(mddev->queue, raid0_mergeable_bvec);
396 return md_integrity_register(mddev);
399 static int raid0_stop(struct mddev *mddev)
401 struct r0conf *conf = mddev->private;
403 blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/
404 kfree(conf->strip_zone);
405 kfree(conf->devlist);
407 mddev->private = NULL;
411 /* Find the zone which holds a particular offset
412 * Update *sectorp to be an offset in that zone
414 static struct strip_zone *find_zone(struct r0conf *conf,
418 struct strip_zone *z = conf->strip_zone;
419 sector_t sector = *sectorp;
421 for (i = 0; i < conf->nr_strip_zones; i++)
422 if (sector < z[i].zone_end) {
424 *sectorp = sector - z[i-1].zone_end;
431 * remaps the bio to the target device. we separate two flows.
432 * power 2 flow and a general flow for the sake of perfromance
434 static struct md_rdev *map_sector(struct mddev *mddev, struct strip_zone *zone,
435 sector_t sector, sector_t *sector_offset)
437 unsigned int sect_in_chunk;
439 struct r0conf *conf = mddev->private;
440 int raid_disks = conf->strip_zone[0].nb_dev;
441 unsigned int chunk_sects = mddev->chunk_sectors;
443 if (is_power_of_2(chunk_sects)) {
444 int chunksect_bits = ffz(~chunk_sects);
445 /* find the sector offset inside the chunk */
446 sect_in_chunk = sector & (chunk_sects - 1);
447 sector >>= chunksect_bits;
449 chunk = *sector_offset;
450 /* quotient is the chunk in real device*/
451 sector_div(chunk, zone->nb_dev << chunksect_bits);
453 sect_in_chunk = sector_div(sector, chunk_sects);
454 chunk = *sector_offset;
455 sector_div(chunk, chunk_sects * zone->nb_dev);
458 * position the bio over the real device
459 * real sector = chunk in device + starting of zone
460 * + the position in the chunk
462 *sector_offset = (chunk * chunk_sects) + sect_in_chunk;
463 return conf->devlist[(zone - conf->strip_zone)*raid_disks
464 + sector_div(sector, zone->nb_dev)];
468 * Is io distribute over 1 or more chunks ?
470 static inline int is_io_in_chunk_boundary(struct mddev *mddev,
471 unsigned int chunk_sects, struct bio *bio)
473 if (likely(is_power_of_2(chunk_sects))) {
474 return chunk_sects >= ((bio->bi_sector & (chunk_sects-1))
475 + (bio->bi_size >> 9));
477 sector_t sector = bio->bi_sector;
478 return chunk_sects >= (sector_div(sector, chunk_sects)
479 + (bio->bi_size >> 9));
483 static void raid0_make_request(struct mddev *mddev, struct bio *bio)
485 unsigned int chunk_sects;
486 sector_t sector_offset;
487 struct strip_zone *zone;
488 struct md_rdev *tmp_dev;
490 if (unlikely(bio->bi_rw & REQ_FLUSH)) {
491 md_flush_request(mddev, bio);
495 chunk_sects = mddev->chunk_sectors;
496 if (unlikely(!is_io_in_chunk_boundary(mddev, chunk_sects, bio))) {
497 sector_t sector = bio->bi_sector;
499 /* Sanity check -- queue functions should prevent this happening */
500 if (bio->bi_vcnt != 1 ||
503 /* This is a one page bio that upper layers
504 * refuse to split for us, so we need to split it.
506 if (likely(is_power_of_2(chunk_sects)))
507 bp = bio_split(bio, chunk_sects - (sector &
510 bp = bio_split(bio, chunk_sects -
511 sector_div(sector, chunk_sects));
512 raid0_make_request(mddev, &bp->bio1);
513 raid0_make_request(mddev, &bp->bio2);
514 bio_pair_release(bp);
518 sector_offset = bio->bi_sector;
519 zone = find_zone(mddev->private, §or_offset);
520 tmp_dev = map_sector(mddev, zone, bio->bi_sector,
522 bio->bi_bdev = tmp_dev->bdev;
523 bio->bi_sector = sector_offset + zone->dev_start +
524 tmp_dev->data_offset;
526 generic_make_request(bio);
530 printk("md/raid0:%s: make_request bug: can't convert block across chunks"
531 " or bigger than %dk %llu %d\n",
532 mdname(mddev), chunk_sects / 2,
533 (unsigned long long)bio->bi_sector, bio->bi_size >> 10);
539 static void raid0_status(struct seq_file *seq, struct mddev *mddev)
541 seq_printf(seq, " %dk chunks", mddev->chunk_sectors / 2);
545 static void *raid0_takeover_raid45(struct mddev *mddev)
547 struct md_rdev *rdev;
548 struct r0conf *priv_conf;
550 if (mddev->degraded != 1) {
551 printk(KERN_ERR "md/raid0:%s: raid5 must be degraded! Degraded disks: %d\n",
554 return ERR_PTR(-EINVAL);
557 list_for_each_entry(rdev, &mddev->disks, same_set) {
558 /* check slot number for a disk */
559 if (rdev->raid_disk == mddev->raid_disks-1) {
560 printk(KERN_ERR "md/raid0:%s: raid5 must have missing parity disk!\n",
562 return ERR_PTR(-EINVAL);
566 /* Set new parameters */
567 mddev->new_level = 0;
568 mddev->new_layout = 0;
569 mddev->new_chunk_sectors = mddev->chunk_sectors;
571 mddev->delta_disks = -1;
572 /* make sure it will be not marked as dirty */
573 mddev->recovery_cp = MaxSector;
575 create_strip_zones(mddev, &priv_conf);
579 static void *raid0_takeover_raid10(struct mddev *mddev)
581 struct r0conf *priv_conf;
584 * - far_copies must be 1
585 * - near_copies must be 2
586 * - disks number must be even
587 * - all mirrors must be already degraded
589 if (mddev->layout != ((1 << 8) + 2)) {
590 printk(KERN_ERR "md/raid0:%s:: Raid0 cannot takover layout: 0x%x\n",
593 return ERR_PTR(-EINVAL);
595 if (mddev->raid_disks & 1) {
596 printk(KERN_ERR "md/raid0:%s: Raid0 cannot takover Raid10 with odd disk number.\n",
598 return ERR_PTR(-EINVAL);
600 if (mddev->degraded != (mddev->raid_disks>>1)) {
601 printk(KERN_ERR "md/raid0:%s: All mirrors must be already degraded!\n",
603 return ERR_PTR(-EINVAL);
606 /* Set new parameters */
607 mddev->new_level = 0;
608 mddev->new_layout = 0;
609 mddev->new_chunk_sectors = mddev->chunk_sectors;
610 mddev->delta_disks = - mddev->raid_disks / 2;
611 mddev->raid_disks += mddev->delta_disks;
613 /* make sure it will be not marked as dirty */
614 mddev->recovery_cp = MaxSector;
616 create_strip_zones(mddev, &priv_conf);
620 static void *raid0_takeover_raid1(struct mddev *mddev)
622 struct r0conf *priv_conf;
625 * - (N - 1) mirror drives must be already faulty
627 if ((mddev->raid_disks - 1) != mddev->degraded) {
628 printk(KERN_ERR "md/raid0:%s: (N - 1) mirrors drives must be already faulty!\n",
630 return ERR_PTR(-EINVAL);
633 /* Set new parameters */
634 mddev->new_level = 0;
635 mddev->new_layout = 0;
636 mddev->new_chunk_sectors = 128; /* by default set chunk size to 64k */
637 mddev->delta_disks = 1 - mddev->raid_disks;
638 mddev->raid_disks = 1;
639 /* make sure it will be not marked as dirty */
640 mddev->recovery_cp = MaxSector;
642 create_strip_zones(mddev, &priv_conf);
646 static void *raid0_takeover(struct mddev *mddev)
648 /* raid0 can take over:
649 * raid4 - if all data disks are active.
650 * raid5 - providing it is Raid4 layout and one disk is faulty
651 * raid10 - assuming we have all necessary active disks
652 * raid1 - with (N -1) mirror drives faulty
654 if (mddev->level == 4)
655 return raid0_takeover_raid45(mddev);
657 if (mddev->level == 5) {
658 if (mddev->layout == ALGORITHM_PARITY_N)
659 return raid0_takeover_raid45(mddev);
661 printk(KERN_ERR "md/raid0:%s: Raid can only takeover Raid5 with layout: %d\n",
662 mdname(mddev), ALGORITHM_PARITY_N);
665 if (mddev->level == 10)
666 return raid0_takeover_raid10(mddev);
668 if (mddev->level == 1)
669 return raid0_takeover_raid1(mddev);
671 printk(KERN_ERR "Takeover from raid%i to raid0 not supported\n",
674 return ERR_PTR(-EINVAL);
677 static void raid0_quiesce(struct mddev *mddev, int state)
681 static struct md_personality raid0_personality=
685 .owner = THIS_MODULE,
686 .make_request = raid0_make_request,
689 .status = raid0_status,
691 .takeover = raid0_takeover,
692 .quiesce = raid0_quiesce,
695 static int __init raid0_init (void)
697 return register_md_personality (&raid0_personality);
700 static void raid0_exit (void)
702 unregister_md_personality (&raid0_personality);
705 module_init(raid0_init);
706 module_exit(raid0_exit);
707 MODULE_LICENSE("GPL");
708 MODULE_DESCRIPTION("RAID0 (striping) personality for MD");
709 MODULE_ALIAS("md-personality-2"); /* RAID0 */
710 MODULE_ALIAS("md-raid0");
711 MODULE_ALIAS("md-level-0");