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 list_for_each_entry(rdev1, &mddev->disks, same_set) {
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 list_for_each_entry(rdev2, &mddev->disks, same_set) {
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 list_for_each_entry(rdev1, &mddev->disks, same_set) {
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);
191 /* as we don't honour merge_bvec_fn, we must never risk
192 * violating it, so limit ->max_segments to 1, lying within
196 if (rdev1->bdev->bd_disk->queue->merge_bvec_fn) {
197 blk_queue_max_segments(mddev->queue, 1);
198 blk_queue_segment_boundary(mddev->queue,
199 PAGE_CACHE_SIZE - 1);
201 if (!smallest || (rdev1->sectors < smallest->sectors))
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);
263 mddev->queue->backing_dev_info.congested_fn = raid0_congested;
264 mddev->queue->backing_dev_info.congested_data = mddev;
267 * now since we have the hard sector sizes, we can make sure
268 * chunk size is a multiple of that sector size
270 if ((mddev->chunk_sectors << 9) % queue_logical_block_size(mddev->queue)) {
271 printk(KERN_ERR "md/raid0:%s: chunk_size of %d not valid\n",
273 mddev->chunk_sectors << 9);
277 blk_queue_io_min(mddev->queue, mddev->chunk_sectors << 9);
278 blk_queue_io_opt(mddev->queue,
279 (mddev->chunk_sectors << 9) * mddev->raid_disks);
281 pr_debug("md/raid0:%s: done.\n", mdname(mddev));
282 *private_conf = conf;
286 kfree(conf->strip_zone);
287 kfree(conf->devlist);
289 *private_conf = ERR_PTR(err);
294 * raid0_mergeable_bvec -- tell bio layer if a two requests can be merged
296 * @bvm: properties of new bio
297 * @biovec: the request that could be merged to it.
299 * Return amount of bytes we can accept at this offset
301 static int raid0_mergeable_bvec(struct request_queue *q,
302 struct bvec_merge_data *bvm,
303 struct bio_vec *biovec)
305 struct mddev *mddev = q->queuedata;
306 sector_t sector = bvm->bi_sector + get_start_sect(bvm->bi_bdev);
308 unsigned int chunk_sectors = mddev->chunk_sectors;
309 unsigned int bio_sectors = bvm->bi_size >> 9;
311 if (is_power_of_2(chunk_sectors))
312 max = (chunk_sectors - ((sector & (chunk_sectors-1))
313 + bio_sectors)) << 9;
315 max = (chunk_sectors - (sector_div(sector, chunk_sectors)
316 + bio_sectors)) << 9;
317 if (max < 0) max = 0; /* bio_add cannot handle a negative return */
318 if (max <= biovec->bv_len && bio_sectors == 0)
319 return biovec->bv_len;
324 static sector_t raid0_size(struct mddev *mddev, sector_t sectors, int raid_disks)
326 sector_t array_sectors = 0;
327 struct md_rdev *rdev;
329 WARN_ONCE(sectors || raid_disks,
330 "%s does not support generic reshape\n", __func__);
332 list_for_each_entry(rdev, &mddev->disks, same_set)
333 array_sectors += (rdev->sectors &
334 ~(sector_t)(mddev->chunk_sectors-1));
336 return array_sectors;
339 static int raid0_run(struct mddev *mddev)
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))
351 blk_queue_max_hw_sectors(mddev->queue, mddev->chunk_sectors);
353 /* if private is not null, we are here after takeover */
354 if (mddev->private == NULL) {
355 ret = create_strip_zones(mddev, &conf);
358 mddev->private = conf;
360 conf = mddev->private;
362 /* calculate array device size */
363 md_set_array_sectors(mddev, raid0_size(mddev, 0, 0));
365 printk(KERN_INFO "md/raid0:%s: md_size is %llu sectors.\n",
367 (unsigned long long)mddev->array_sectors);
368 /* calculate the max read-ahead size.
369 * For read-ahead of large files to be effective, we need to
370 * readahead at least twice a whole stripe. i.e. number of devices
371 * multiplied by chunk size times 2.
372 * If an individual device has an ra_pages greater than the
373 * chunk size, then we will not drive that device as hard as it
374 * wants. We consider this a configuration error: a larger
375 * chunksize should be used in that case.
378 int stripe = mddev->raid_disks *
379 (mddev->chunk_sectors << 9) / PAGE_SIZE;
380 if (mddev->queue->backing_dev_info.ra_pages < 2* stripe)
381 mddev->queue->backing_dev_info.ra_pages = 2* stripe;
384 blk_queue_merge_bvec(mddev->queue, raid0_mergeable_bvec);
386 return md_integrity_register(mddev);
389 static int raid0_stop(struct mddev *mddev)
391 struct r0conf *conf = mddev->private;
393 blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/
394 kfree(conf->strip_zone);
395 kfree(conf->devlist);
397 mddev->private = NULL;
401 /* Find the zone which holds a particular offset
402 * Update *sectorp to be an offset in that zone
404 static struct strip_zone *find_zone(struct r0conf *conf,
408 struct strip_zone *z = conf->strip_zone;
409 sector_t sector = *sectorp;
411 for (i = 0; i < conf->nr_strip_zones; i++)
412 if (sector < z[i].zone_end) {
414 *sectorp = sector - z[i-1].zone_end;
421 * remaps the bio to the target device. we separate two flows.
422 * power 2 flow and a general flow for the sake of perfromance
424 static struct md_rdev *map_sector(struct mddev *mddev, struct strip_zone *zone,
425 sector_t sector, sector_t *sector_offset)
427 unsigned int sect_in_chunk;
429 struct r0conf *conf = mddev->private;
430 int raid_disks = conf->strip_zone[0].nb_dev;
431 unsigned int chunk_sects = mddev->chunk_sectors;
433 if (is_power_of_2(chunk_sects)) {
434 int chunksect_bits = ffz(~chunk_sects);
435 /* find the sector offset inside the chunk */
436 sect_in_chunk = sector & (chunk_sects - 1);
437 sector >>= chunksect_bits;
439 chunk = *sector_offset;
440 /* quotient is the chunk in real device*/
441 sector_div(chunk, zone->nb_dev << chunksect_bits);
443 sect_in_chunk = sector_div(sector, chunk_sects);
444 chunk = *sector_offset;
445 sector_div(chunk, chunk_sects * zone->nb_dev);
448 * position the bio over the real device
449 * real sector = chunk in device + starting of zone
450 * + the position in the chunk
452 *sector_offset = (chunk * chunk_sects) + sect_in_chunk;
453 return conf->devlist[(zone - conf->strip_zone)*raid_disks
454 + sector_div(sector, zone->nb_dev)];
458 * Is io distribute over 1 or more chunks ?
460 static inline int is_io_in_chunk_boundary(struct mddev *mddev,
461 unsigned int chunk_sects, struct bio *bio)
463 if (likely(is_power_of_2(chunk_sects))) {
464 return chunk_sects >= ((bio->bi_sector & (chunk_sects-1))
465 + (bio->bi_size >> 9));
467 sector_t sector = bio->bi_sector;
468 return chunk_sects >= (sector_div(sector, chunk_sects)
469 + (bio->bi_size >> 9));
473 static void raid0_make_request(struct mddev *mddev, struct bio *bio)
475 unsigned int chunk_sects;
476 sector_t sector_offset;
477 struct strip_zone *zone;
478 struct md_rdev *tmp_dev;
480 if (unlikely(bio->bi_rw & REQ_FLUSH)) {
481 md_flush_request(mddev, bio);
485 chunk_sects = mddev->chunk_sectors;
486 if (unlikely(!is_io_in_chunk_boundary(mddev, chunk_sects, bio))) {
487 sector_t sector = bio->bi_sector;
489 /* Sanity check -- queue functions should prevent this happening */
490 if (bio->bi_vcnt != 1 ||
493 /* This is a one page bio that upper layers
494 * refuse to split for us, so we need to split it.
496 if (likely(is_power_of_2(chunk_sects)))
497 bp = bio_split(bio, chunk_sects - (sector &
500 bp = bio_split(bio, chunk_sects -
501 sector_div(sector, chunk_sects));
502 raid0_make_request(mddev, &bp->bio1);
503 raid0_make_request(mddev, &bp->bio2);
504 bio_pair_release(bp);
508 sector_offset = bio->bi_sector;
509 zone = find_zone(mddev->private, §or_offset);
510 tmp_dev = map_sector(mddev, zone, bio->bi_sector,
512 bio->bi_bdev = tmp_dev->bdev;
513 bio->bi_sector = sector_offset + zone->dev_start +
514 tmp_dev->data_offset;
516 generic_make_request(bio);
520 printk("md/raid0:%s: make_request bug: can't convert block across chunks"
521 " or bigger than %dk %llu %d\n",
522 mdname(mddev), chunk_sects / 2,
523 (unsigned long long)bio->bi_sector, bio->bi_size >> 10);
529 static void raid0_status(struct seq_file *seq, struct mddev *mddev)
531 seq_printf(seq, " %dk chunks", mddev->chunk_sectors / 2);
535 static void *raid0_takeover_raid45(struct mddev *mddev)
537 struct md_rdev *rdev;
538 struct r0conf *priv_conf;
540 if (mddev->degraded != 1) {
541 printk(KERN_ERR "md/raid0:%s: raid5 must be degraded! Degraded disks: %d\n",
544 return ERR_PTR(-EINVAL);
547 list_for_each_entry(rdev, &mddev->disks, same_set) {
548 /* check slot number for a disk */
549 if (rdev->raid_disk == mddev->raid_disks-1) {
550 printk(KERN_ERR "md/raid0:%s: raid5 must have missing parity disk!\n",
552 return ERR_PTR(-EINVAL);
556 /* Set new parameters */
557 mddev->new_level = 0;
558 mddev->new_layout = 0;
559 mddev->new_chunk_sectors = mddev->chunk_sectors;
561 mddev->delta_disks = -1;
562 /* make sure it will be not marked as dirty */
563 mddev->recovery_cp = MaxSector;
565 create_strip_zones(mddev, &priv_conf);
569 static void *raid0_takeover_raid10(struct mddev *mddev)
571 struct r0conf *priv_conf;
574 * - far_copies must be 1
575 * - near_copies must be 2
576 * - disks number must be even
577 * - all mirrors must be already degraded
579 if (mddev->layout != ((1 << 8) + 2)) {
580 printk(KERN_ERR "md/raid0:%s:: Raid0 cannot takover layout: 0x%x\n",
583 return ERR_PTR(-EINVAL);
585 if (mddev->raid_disks & 1) {
586 printk(KERN_ERR "md/raid0:%s: Raid0 cannot takover Raid10 with odd disk number.\n",
588 return ERR_PTR(-EINVAL);
590 if (mddev->degraded != (mddev->raid_disks>>1)) {
591 printk(KERN_ERR "md/raid0:%s: All mirrors must be already degraded!\n",
593 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;
600 mddev->delta_disks = - mddev->raid_disks / 2;
601 mddev->raid_disks += mddev->delta_disks;
603 /* make sure it will be not marked as dirty */
604 mddev->recovery_cp = MaxSector;
606 create_strip_zones(mddev, &priv_conf);
610 static void *raid0_takeover_raid1(struct mddev *mddev)
612 struct r0conf *priv_conf;
615 * - (N - 1) mirror drives must be already faulty
617 if ((mddev->raid_disks - 1) != mddev->degraded) {
618 printk(KERN_ERR "md/raid0:%s: (N - 1) mirrors drives must be already faulty!\n",
620 return ERR_PTR(-EINVAL);
623 /* Set new parameters */
624 mddev->new_level = 0;
625 mddev->new_layout = 0;
626 mddev->new_chunk_sectors = 128; /* by default set chunk size to 64k */
627 mddev->delta_disks = 1 - mddev->raid_disks;
628 mddev->raid_disks = 1;
629 /* make sure it will be not marked as dirty */
630 mddev->recovery_cp = MaxSector;
632 create_strip_zones(mddev, &priv_conf);
636 static void *raid0_takeover(struct mddev *mddev)
638 /* raid0 can take over:
639 * raid4 - if all data disks are active.
640 * raid5 - providing it is Raid4 layout and one disk is faulty
641 * raid10 - assuming we have all necessary active disks
642 * raid1 - with (N -1) mirror drives faulty
644 if (mddev->level == 4)
645 return raid0_takeover_raid45(mddev);
647 if (mddev->level == 5) {
648 if (mddev->layout == ALGORITHM_PARITY_N)
649 return raid0_takeover_raid45(mddev);
651 printk(KERN_ERR "md/raid0:%s: Raid can only takeover Raid5 with layout: %d\n",
652 mdname(mddev), ALGORITHM_PARITY_N);
655 if (mddev->level == 10)
656 return raid0_takeover_raid10(mddev);
658 if (mddev->level == 1)
659 return raid0_takeover_raid1(mddev);
661 printk(KERN_ERR "Takeover from raid%i to raid0 not supported\n",
664 return ERR_PTR(-EINVAL);
667 static void raid0_quiesce(struct mddev *mddev, int state)
671 static struct md_personality raid0_personality=
675 .owner = THIS_MODULE,
676 .make_request = raid0_make_request,
679 .status = raid0_status,
681 .takeover = raid0_takeover,
682 .quiesce = raid0_quiesce,
685 static int __init raid0_init (void)
687 return register_md_personality (&raid0_personality);
690 static void raid0_exit (void)
692 unregister_md_personality (&raid0_personality);
695 module_init(raid0_init);
696 module_exit(raid0_exit);
697 MODULE_LICENSE("GPL");
698 MODULE_DESCRIPTION("RAID0 (striping) personality for MD");
699 MODULE_ALIAS("md-personality-2"); /* RAID0 */
700 MODULE_ALIAS("md-raid0");
701 MODULE_ALIAS("md-level-0");