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>
27 static void raid0_unplug(struct request_queue *q)
29 mddev_t *mddev = q->queuedata;
30 raid0_conf_t *conf = mddev->private;
31 mdk_rdev_t **devlist = conf->devlist;
32 int raid_disks = conf->strip_zone[0].nb_dev;
35 for (i=0; i < raid_disks; i++) {
36 struct request_queue *r_queue = bdev_get_queue(devlist[i]->bdev);
42 static int raid0_congested(void *data, int bits)
44 mddev_t *mddev = data;
45 raid0_conf_t *conf = mddev->private;
46 mdk_rdev_t **devlist = conf->devlist;
47 int raid_disks = conf->strip_zone[0].nb_dev;
50 if (mddev_congested(mddev, bits))
53 for (i = 0; i < raid_disks && !ret ; i++) {
54 struct request_queue *q = bdev_get_queue(devlist[i]->bdev);
56 ret |= bdi_congested(&q->backing_dev_info, bits);
62 * inform the user of the raid configuration
64 static void dump_zones(mddev_t *mddev)
67 sector_t zone_size = 0;
68 sector_t zone_start = 0;
69 char b[BDEVNAME_SIZE];
70 raid0_conf_t *conf = mddev->private;
71 int raid_disks = conf->strip_zone[0].nb_dev;
72 printk(KERN_INFO "******* %s configuration *********\n",
75 for (j = 0; j < conf->nr_strip_zones; j++) {
76 printk(KERN_INFO "zone%d=[", j);
77 for (k = 0; k < conf->strip_zone[j].nb_dev; k++)
78 printk(KERN_CONT "%s/",
79 bdevname(conf->devlist[j*raid_disks
81 printk(KERN_CONT "]\n");
83 zone_size = conf->strip_zone[j].zone_end - zone_start;
84 printk(KERN_INFO " zone offset=%llukb "
85 "device offset=%llukb size=%llukb\n",
86 (unsigned long long)zone_start>>1,
87 (unsigned long long)conf->strip_zone[j].dev_start>>1,
88 (unsigned long long)zone_size>>1);
89 zone_start = conf->strip_zone[j].zone_end;
91 printk(KERN_INFO "**********************************\n\n");
94 static int create_strip_zones(mddev_t *mddev, raid0_conf_t **private_conf)
97 sector_t curr_zone_end, sectors;
98 mdk_rdev_t *smallest, *rdev1, *rdev2, *rdev, **dev;
99 struct strip_zone *zone;
101 char b[BDEVNAME_SIZE];
102 raid0_conf_t *conf = kzalloc(sizeof(*conf), GFP_KERNEL);
106 list_for_each_entry(rdev1, &mddev->disks, same_set) {
107 printk(KERN_INFO "md/raid0:%s: looking at %s\n",
109 bdevname(rdev1->bdev, b));
112 /* round size to chunk_size */
113 sectors = rdev1->sectors;
114 sector_div(sectors, mddev->chunk_sectors);
115 rdev1->sectors = sectors * mddev->chunk_sectors;
117 list_for_each_entry(rdev2, &mddev->disks, same_set) {
118 printk(KERN_INFO "md/raid0:%s: comparing %s(%llu)",
120 bdevname(rdev1->bdev,b),
121 (unsigned long long)rdev1->sectors);
122 printk(KERN_CONT " with %s(%llu)\n",
123 bdevname(rdev2->bdev,b),
124 (unsigned long long)rdev2->sectors);
125 if (rdev2 == rdev1) {
126 printk(KERN_INFO "md/raid0:%s: END\n",
130 if (rdev2->sectors == rdev1->sectors) {
132 * Not unique, don't count it as a new
135 printk(KERN_INFO "md/raid0:%s: EQUAL\n",
140 printk(KERN_INFO "md/raid0:%s: NOT EQUAL\n",
144 printk(KERN_INFO "md/raid0:%s: ==> UNIQUE\n",
146 conf->nr_strip_zones++;
147 printk(KERN_INFO "md/raid0:%s: %d zones\n",
148 mdname(mddev), conf->nr_strip_zones);
151 printk(KERN_INFO "md/raid0:%s: FINAL %d zones\n",
152 mdname(mddev), conf->nr_strip_zones);
154 conf->strip_zone = kzalloc(sizeof(struct strip_zone)*
155 conf->nr_strip_zones, GFP_KERNEL);
156 if (!conf->strip_zone)
158 conf->devlist = kzalloc(sizeof(mdk_rdev_t*)*
159 conf->nr_strip_zones*mddev->raid_disks,
164 /* The first zone must contain all devices, so here we check that
165 * there is a proper alignment of slots to devices and find them all
167 zone = &conf->strip_zone[0];
172 list_for_each_entry(rdev1, &mddev->disks, same_set) {
173 int j = rdev1->raid_disk;
175 if (mddev->level == 10)
176 /* taking over a raid10-n2 array */
179 if (j < 0 || 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);
191 disk_stack_limits(mddev->gendisk, rdev1->bdev,
192 rdev1->data_offset << 9);
193 /* as we don't honour merge_bvec_fn, we must never risk
194 * violating it, so limit ->max_segments to 1, lying within
198 if (rdev1->bdev->bd_disk->queue->merge_bvec_fn) {
199 blk_queue_max_segments(mddev->queue, 1);
200 blk_queue_segment_boundary(mddev->queue,
201 PAGE_CACHE_SIZE - 1);
203 if (!smallest || (rdev1->sectors < smallest->sectors))
207 if (cnt != mddev->raid_disks) {
208 printk(KERN_ERR "md/raid0:%s: too few disks (%d of %d) - "
209 "aborting!\n", mdname(mddev), cnt, mddev->raid_disks);
213 zone->zone_end = smallest->sectors * cnt;
215 curr_zone_end = zone->zone_end;
217 /* now do the other zones */
218 for (i = 1; i < conf->nr_strip_zones; i++)
222 zone = conf->strip_zone + i;
223 dev = conf->devlist + i * mddev->raid_disks;
225 printk(KERN_INFO "md/raid0:%s: zone %d\n",
227 zone->dev_start = smallest->sectors;
231 for (j=0; j<cnt; j++) {
232 rdev = conf->devlist[j];
233 printk(KERN_INFO "md/raid0:%s: checking %s ...",
235 bdevname(rdev->bdev, b));
236 if (rdev->sectors <= zone->dev_start) {
237 printk(KERN_CONT " nope.\n");
240 printk(KERN_CONT " contained as device %d\n", c);
243 if (!smallest || rdev->sectors < smallest->sectors) {
245 printk(KERN_INFO "md/raid0:%s: (%llu) is smallest!.\n",
247 (unsigned long long)rdev->sectors);
252 sectors = (smallest->sectors - zone->dev_start) * c;
253 printk(KERN_INFO "md/raid0:%s: zone->nb_dev: %d, sectors: %llu\n",
255 zone->nb_dev, (unsigned long long)sectors);
257 curr_zone_end += sectors;
258 zone->zone_end = curr_zone_end;
260 printk(KERN_INFO "md/raid0:%s: current zone start: %llu\n",
262 (unsigned long long)smallest->sectors);
264 mddev->queue->unplug_fn = raid0_unplug;
265 mddev->queue->backing_dev_info.congested_fn = raid0_congested;
266 mddev->queue->backing_dev_info.congested_data = mddev;
269 * now since we have the hard sector sizes, we can make sure
270 * chunk size is a multiple of that sector size
272 if ((mddev->chunk_sectors << 9) % queue_logical_block_size(mddev->queue)) {
273 printk(KERN_ERR "md/raid0:%s: chunk_size of %d not valid\n",
275 mddev->chunk_sectors << 9);
279 blk_queue_io_min(mddev->queue, mddev->chunk_sectors << 9);
280 blk_queue_io_opt(mddev->queue,
281 (mddev->chunk_sectors << 9) * mddev->raid_disks);
283 printk(KERN_INFO "md/raid0:%s: done.\n", mdname(mddev));
284 *private_conf = conf;
288 kfree(conf->strip_zone);
289 kfree(conf->devlist);
291 *private_conf = NULL;
296 * raid0_mergeable_bvec -- tell bio layer if a two requests can be merged
298 * @bvm: properties of new bio
299 * @biovec: the request that could be merged to it.
301 * Return amount of bytes we can accept at this offset
303 static int raid0_mergeable_bvec(struct request_queue *q,
304 struct bvec_merge_data *bvm,
305 struct bio_vec *biovec)
307 mddev_t *mddev = q->queuedata;
308 sector_t sector = bvm->bi_sector + get_start_sect(bvm->bi_bdev);
310 unsigned int chunk_sectors = mddev->chunk_sectors;
311 unsigned int bio_sectors = bvm->bi_size >> 9;
313 if (is_power_of_2(chunk_sectors))
314 max = (chunk_sectors - ((sector & (chunk_sectors-1))
315 + bio_sectors)) << 9;
317 max = (chunk_sectors - (sector_div(sector, chunk_sectors)
318 + bio_sectors)) << 9;
319 if (max < 0) max = 0; /* bio_add cannot handle a negative return */
320 if (max <= biovec->bv_len && bio_sectors == 0)
321 return biovec->bv_len;
326 static sector_t raid0_size(mddev_t *mddev, sector_t sectors, int raid_disks)
328 sector_t array_sectors = 0;
331 WARN_ONCE(sectors || raid_disks,
332 "%s does not support generic reshape\n", __func__);
334 list_for_each_entry(rdev, &mddev->disks, same_set)
335 array_sectors += rdev->sectors;
337 return array_sectors;
340 static int raid0_run(mddev_t *mddev)
345 if (mddev->chunk_sectors == 0) {
346 printk(KERN_ERR "md/raid0:%s: chunk size must be set.\n",
350 if (md_check_no_bitmap(mddev))
352 blk_queue_max_hw_sectors(mddev->queue, mddev->chunk_sectors);
353 mddev->queue->queue_lock = &mddev->queue->__queue_lock;
355 /* if private is not null, we are here after takeover */
356 if (mddev->private == NULL) {
357 ret = create_strip_zones(mddev, &conf);
360 mddev->private = conf;
362 conf = mddev->private;
363 if (conf->scale_raid_disks) {
365 for (i=0; i < conf->strip_zone[0].nb_dev; i++)
366 conf->devlist[i]->raid_disk /= conf->scale_raid_disks;
367 /* FIXME update sysfs rd links */
370 /* calculate array device size */
371 md_set_array_sectors(mddev, raid0_size(mddev, 0, 0));
373 printk(KERN_INFO "md/raid0:%s: md_size is %llu sectors.\n",
375 (unsigned long long)mddev->array_sectors);
376 /* calculate the max read-ahead size.
377 * For read-ahead of large files to be effective, we need to
378 * readahead at least twice a whole stripe. i.e. number of devices
379 * multiplied by chunk size times 2.
380 * If an individual device has an ra_pages greater than the
381 * chunk size, then we will not drive that device as hard as it
382 * wants. We consider this a configuration error: a larger
383 * chunksize should be used in that case.
386 int stripe = mddev->raid_disks *
387 (mddev->chunk_sectors << 9) / PAGE_SIZE;
388 if (mddev->queue->backing_dev_info.ra_pages < 2* stripe)
389 mddev->queue->backing_dev_info.ra_pages = 2* stripe;
392 blk_queue_merge_bvec(mddev->queue, raid0_mergeable_bvec);
394 md_integrity_register(mddev);
398 static int raid0_stop(mddev_t *mddev)
400 raid0_conf_t *conf = mddev->private;
402 blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/
403 kfree(conf->strip_zone);
404 kfree(conf->devlist);
406 mddev->private = NULL;
410 /* Find the zone which holds a particular offset
411 * Update *sectorp to be an offset in that zone
413 static struct strip_zone *find_zone(struct raid0_private_data *conf,
417 struct strip_zone *z = conf->strip_zone;
418 sector_t sector = *sectorp;
420 for (i = 0; i < conf->nr_strip_zones; i++)
421 if (sector < z[i].zone_end) {
423 *sectorp = sector - z[i-1].zone_end;
430 * remaps the bio to the target device. we separate two flows.
431 * power 2 flow and a general flow for the sake of perfromance
433 static mdk_rdev_t *map_sector(mddev_t *mddev, struct strip_zone *zone,
434 sector_t sector, sector_t *sector_offset)
436 unsigned int sect_in_chunk;
438 raid0_conf_t *conf = mddev->private;
439 int raid_disks = conf->strip_zone[0].nb_dev;
440 unsigned int chunk_sects = mddev->chunk_sectors;
442 if (is_power_of_2(chunk_sects)) {
443 int chunksect_bits = ffz(~chunk_sects);
444 /* find the sector offset inside the chunk */
445 sect_in_chunk = sector & (chunk_sects - 1);
446 sector >>= chunksect_bits;
448 chunk = *sector_offset;
449 /* quotient is the chunk in real device*/
450 sector_div(chunk, zone->nb_dev << chunksect_bits);
452 sect_in_chunk = sector_div(sector, chunk_sects);
453 chunk = *sector_offset;
454 sector_div(chunk, chunk_sects * zone->nb_dev);
457 * position the bio over the real device
458 * real sector = chunk in device + starting of zone
459 * + the position in the chunk
461 *sector_offset = (chunk * chunk_sects) + sect_in_chunk;
462 return conf->devlist[(zone - conf->strip_zone)*raid_disks
463 + sector_div(sector, zone->nb_dev)];
467 * Is io distribute over 1 or more chunks ?
469 static inline int is_io_in_chunk_boundary(mddev_t *mddev,
470 unsigned int chunk_sects, struct bio *bio)
472 if (likely(is_power_of_2(chunk_sects))) {
473 return chunk_sects >= ((bio->bi_sector & (chunk_sects-1))
474 + (bio->bi_size >> 9));
476 sector_t sector = bio->bi_sector;
477 return chunk_sects >= (sector_div(sector, chunk_sects)
478 + (bio->bi_size >> 9));
482 static int raid0_make_request(mddev_t *mddev, struct bio *bio)
484 unsigned int chunk_sects;
485 sector_t sector_offset;
486 struct strip_zone *zone;
489 if (unlikely(bio_rw_flagged(bio, BIO_RW_BARRIER))) {
490 md_barrier_request(mddev, bio);
494 chunk_sects = mddev->chunk_sectors;
495 if (unlikely(!is_io_in_chunk_boundary(mddev, chunk_sects, bio))) {
496 sector_t sector = bio->bi_sector;
498 /* Sanity check -- queue functions should prevent this happening */
499 if (bio->bi_vcnt != 1 ||
502 /* This is a one page bio that upper layers
503 * refuse to split for us, so we need to split it.
505 if (likely(is_power_of_2(chunk_sects)))
506 bp = bio_split(bio, chunk_sects - (sector &
509 bp = bio_split(bio, chunk_sects -
510 sector_div(sector, chunk_sects));
511 if (raid0_make_request(mddev, &bp->bio1))
512 generic_make_request(&bp->bio1);
513 if (raid0_make_request(mddev, &bp->bio2))
514 generic_make_request(&bp->bio2);
516 bio_pair_release(bp);
520 sector_offset = bio->bi_sector;
521 zone = find_zone(mddev->private, §or_offset);
522 tmp_dev = map_sector(mddev, zone, bio->bi_sector,
524 bio->bi_bdev = tmp_dev->bdev;
525 bio->bi_sector = sector_offset + zone->dev_start +
526 tmp_dev->data_offset;
528 * Let the main block layer submit the IO and resolve recursion:
533 printk("md/raid0:%s: make_request bug: can't convert block across chunks"
534 " or bigger than %dk %llu %d\n",
535 mdname(mddev), chunk_sects / 2,
536 (unsigned long long)bio->bi_sector, bio->bi_size >> 10);
542 static void raid0_status(struct seq_file *seq, mddev_t *mddev)
547 char b[BDEVNAME_SIZE];
548 raid0_conf_t *conf = mddev->private;
549 int raid_disks = conf->strip_zone[0].nb_dev;
552 sector_t zone_start = 0;
555 for (j = 0; j < conf->nr_strip_zones; j++) {
556 seq_printf(seq, " z%d", j);
557 seq_printf(seq, "=[");
558 for (k = 0; k < conf->strip_zone[j].nb_dev; k++)
559 seq_printf(seq, "%s/", bdevname(
560 conf->devlist[j*raid_disks + k]
563 zone_size = conf->strip_zone[j].zone_end - zone_start;
564 seq_printf(seq, "] ze=%lld ds=%lld s=%lld\n",
565 (unsigned long long)zone_start>>1,
566 (unsigned long long)conf->strip_zone[j].dev_start>>1,
567 (unsigned long long)zone_size>>1);
568 zone_start = conf->strip_zone[j].zone_end;
571 seq_printf(seq, " %dk chunks", mddev->chunk_sectors / 2);
575 static void *raid0_takeover_raid5(mddev_t *mddev)
578 raid0_conf_t *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 list_for_each_entry(rdev, &mddev->disks, same_set) {
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_chunk_sectors = mddev->chunk_sectors;
600 mddev->delta_disks = -1;
601 /* make sure it will be not marked as dirty */
602 mddev->recovery_cp = MaxSector;
604 create_strip_zones(mddev, &priv_conf);
608 static void *raid0_takeover_raid10(mddev_t *mddev)
610 raid0_conf_t *priv_conf;
613 * - far_copies must be 1
614 * - near_copies must be 2
615 * - disks number must be even
616 * - all mirrors must be already degraded
618 if (mddev->layout != ((1 << 8) + 2)) {
619 printk(KERN_ERR "md/raid0:%s:: Raid0 cannot takover layout: 0x%x\n",
622 return ERR_PTR(-EINVAL);
624 if (mddev->raid_disks & 1) {
625 printk(KERN_ERR "md/raid0:%s: Raid0 cannot takover Raid10 with odd disk number.\n",
627 return ERR_PTR(-EINVAL);
629 if (mddev->degraded != (mddev->raid_disks>>1)) {
630 printk(KERN_ERR "md/raid0:%s: All mirrors must be already degraded!\n",
632 return ERR_PTR(-EINVAL);
635 /* Set new parameters */
636 mddev->new_level = 0;
637 mddev->new_chunk_sectors = mddev->chunk_sectors;
638 mddev->delta_disks = - mddev->raid_disks / 2;
639 mddev->raid_disks += mddev->delta_disks;
641 /* make sure it will be not marked as dirty */
642 mddev->recovery_cp = MaxSector;
644 create_strip_zones(mddev, &priv_conf);
645 priv_conf->scale_raid_disks = 2;
649 static void *raid0_takeover(mddev_t *mddev)
651 /* raid0 can take over:
652 * raid5 - providing it is Raid4 layout and one disk is faulty
653 * raid10 - assuming we have all necessary active disks
655 if (mddev->level == 5) {
656 if (mddev->layout == ALGORITHM_PARITY_N)
657 return raid0_takeover_raid5(mddev);
659 printk(KERN_ERR "md/raid0:%s: Raid can only takeover Raid5 with layout: %d\n",
660 mdname(mddev), ALGORITHM_PARITY_N);
663 if (mddev->level == 10)
664 return raid0_takeover_raid10(mddev);
666 return ERR_PTR(-EINVAL);
669 static void raid0_quiesce(mddev_t *mddev, int state)
673 static struct mdk_personality raid0_personality=
677 .owner = THIS_MODULE,
678 .make_request = raid0_make_request,
681 .status = raid0_status,
683 .takeover = raid0_takeover,
684 .quiesce = raid0_quiesce,
687 static int __init raid0_init (void)
689 return register_md_personality (&raid0_personality);
692 static void raid0_exit (void)
694 unregister_md_personality (&raid0_personality);
697 module_init(raid0_init);
698 module_exit(raid0_exit);
699 MODULE_LICENSE("GPL");
700 MODULE_DESCRIPTION("RAID0 (striping) personality for MD");
701 MODULE_ALIAS("md-personality-2"); /* RAID0 */
702 MODULE_ALIAS("md-raid0");
703 MODULE_ALIAS("md-level-0");