2 raid0.c : Multiple Devices driver for Linux
3 Copyright (C) 1994-96 Marc ZYNGIER
6 Copyright (C) 1999, 2000 Ingo Molnar, Red Hat
8 RAID-0 management functions.
10 This program is free software; you can redistribute it and/or modify
11 it under the terms of the GNU General Public License as published by
12 the Free Software Foundation; either version 2, or (at your option)
15 You should have received a copy of the GNU General Public License
16 (for example /usr/src/linux/COPYING); if not, write to the Free
17 Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
20 #include <linux/blkdev.h>
21 #include <linux/seq_file.h>
22 #include <linux/module.h>
23 #include <linux/slab.h>
28 static int raid0_congested(struct mddev *mddev, int bits)
30 struct r0conf *conf = mddev->private;
31 struct md_rdev **devlist = conf->devlist;
32 int raid_disks = conf->strip_zone[0].nb_dev;
35 for (i = 0; i < raid_disks && !ret ; i++) {
36 struct request_queue *q = bdev_get_queue(devlist[i]->bdev);
38 ret |= bdi_congested(&q->backing_dev_info, bits);
44 * inform the user of the raid configuration
46 static void dump_zones(struct mddev *mddev)
49 sector_t zone_size = 0;
50 sector_t zone_start = 0;
51 char b[BDEVNAME_SIZE];
52 struct r0conf *conf = mddev->private;
53 int raid_disks = conf->strip_zone[0].nb_dev;
54 printk(KERN_INFO "md: RAID0 configuration for %s - %d zone%s\n",
56 conf->nr_strip_zones, conf->nr_strip_zones==1?"":"s");
57 for (j = 0; j < conf->nr_strip_zones; j++) {
58 printk(KERN_INFO "md: zone%d=[", j);
59 for (k = 0; k < conf->strip_zone[j].nb_dev; k++)
60 printk(KERN_CONT "%s%s", k?"/":"",
61 bdevname(conf->devlist[j*raid_disks
63 printk(KERN_CONT "]\n");
65 zone_size = conf->strip_zone[j].zone_end - zone_start;
66 printk(KERN_INFO " zone-offset=%10lluKB, "
67 "device-offset=%10lluKB, size=%10lluKB\n",
68 (unsigned long long)zone_start>>1,
69 (unsigned long long)conf->strip_zone[j].dev_start>>1,
70 (unsigned long long)zone_size>>1);
71 zone_start = conf->strip_zone[j].zone_end;
75 static int create_strip_zones(struct mddev *mddev, struct r0conf **private_conf)
78 sector_t curr_zone_end, sectors;
79 struct md_rdev *smallest, *rdev1, *rdev2, *rdev, **dev;
80 struct strip_zone *zone;
82 char b[BDEVNAME_SIZE];
83 char b2[BDEVNAME_SIZE];
84 struct r0conf *conf = kzalloc(sizeof(*conf), GFP_KERNEL);
85 unsigned short blksize = 512;
87 *private_conf = ERR_PTR(-ENOMEM);
90 rdev_for_each(rdev1, mddev) {
91 pr_debug("md/raid0:%s: looking at %s\n",
93 bdevname(rdev1->bdev, b));
96 /* round size to chunk_size */
97 sectors = rdev1->sectors;
98 sector_div(sectors, mddev->chunk_sectors);
99 rdev1->sectors = sectors * mddev->chunk_sectors;
101 blksize = max(blksize, queue_logical_block_size(
102 rdev1->bdev->bd_disk->queue));
104 rdev_for_each(rdev2, mddev) {
105 pr_debug("md/raid0:%s: comparing %s(%llu)"
108 bdevname(rdev1->bdev,b),
109 (unsigned long long)rdev1->sectors,
110 bdevname(rdev2->bdev,b2),
111 (unsigned long long)rdev2->sectors);
112 if (rdev2 == rdev1) {
113 pr_debug("md/raid0:%s: END\n",
117 if (rdev2->sectors == rdev1->sectors) {
119 * Not unique, don't count it as a new
122 pr_debug("md/raid0:%s: EQUAL\n",
127 pr_debug("md/raid0:%s: NOT EQUAL\n",
131 pr_debug("md/raid0:%s: ==> UNIQUE\n",
133 conf->nr_strip_zones++;
134 pr_debug("md/raid0:%s: %d zones\n",
135 mdname(mddev), conf->nr_strip_zones);
138 pr_debug("md/raid0:%s: FINAL %d zones\n",
139 mdname(mddev), conf->nr_strip_zones);
141 * now since we have the hard sector sizes, we can make sure
142 * chunk size is a multiple of that sector size
144 if ((mddev->chunk_sectors << 9) % blksize) {
145 printk(KERN_ERR "md/raid0:%s: chunk_size of %d not multiple of block size %d\n",
147 mddev->chunk_sectors << 9, blksize);
153 conf->strip_zone = kzalloc(sizeof(struct strip_zone)*
154 conf->nr_strip_zones, GFP_KERNEL);
155 if (!conf->strip_zone)
157 conf->devlist = kzalloc(sizeof(struct md_rdev*)*
158 conf->nr_strip_zones*mddev->raid_disks,
163 /* The first zone must contain all devices, so here we check that
164 * there is a proper alignment of slots to devices and find them all
166 zone = &conf->strip_zone[0];
171 rdev_for_each(rdev1, mddev) {
172 int j = rdev1->raid_disk;
174 if (mddev->level == 10) {
175 /* taking over a raid10-n2 array */
177 rdev1->new_raid_disk = j;
180 if (mddev->level == 1) {
181 /* taiking over a raid1 array-
182 * we have only one active disk
185 rdev1->new_raid_disk = j;
190 "md/raid0:%s: remove inactive devices before converting to RAID0\n",
194 if (j >= mddev->raid_disks) {
195 printk(KERN_ERR "md/raid0:%s: bad disk number %d - "
196 "aborting!\n", mdname(mddev), j);
200 printk(KERN_ERR "md/raid0:%s: multiple devices for %d - "
201 "aborting!\n", mdname(mddev), j);
206 if (!smallest || (rdev1->sectors < smallest->sectors))
210 if (cnt != mddev->raid_disks) {
211 printk(KERN_ERR "md/raid0:%s: too few disks (%d of %d) - "
212 "aborting!\n", mdname(mddev), cnt, mddev->raid_disks);
216 zone->zone_end = smallest->sectors * cnt;
218 curr_zone_end = zone->zone_end;
220 /* now do the other zones */
221 for (i = 1; i < conf->nr_strip_zones; i++)
225 zone = conf->strip_zone + i;
226 dev = conf->devlist + i * mddev->raid_disks;
228 pr_debug("md/raid0:%s: zone %d\n", mdname(mddev), i);
229 zone->dev_start = smallest->sectors;
233 for (j=0; j<cnt; j++) {
234 rdev = conf->devlist[j];
235 if (rdev->sectors <= zone->dev_start) {
236 pr_debug("md/raid0:%s: checking %s ... nope\n",
238 bdevname(rdev->bdev, b));
241 pr_debug("md/raid0:%s: checking %s ..."
242 " contained as device %d\n",
244 bdevname(rdev->bdev, b), c);
247 if (!smallest || rdev->sectors < smallest->sectors) {
249 pr_debug("md/raid0:%s: (%llu) is smallest!.\n",
251 (unsigned long long)rdev->sectors);
256 sectors = (smallest->sectors - zone->dev_start) * c;
257 pr_debug("md/raid0:%s: zone->nb_dev: %d, sectors: %llu\n",
259 zone->nb_dev, (unsigned long long)sectors);
261 curr_zone_end += sectors;
262 zone->zone_end = curr_zone_end;
264 pr_debug("md/raid0:%s: current zone start: %llu\n",
266 (unsigned long long)smallest->sectors);
269 pr_debug("md/raid0:%s: done.\n", mdname(mddev));
270 *private_conf = conf;
274 kfree(conf->strip_zone);
275 kfree(conf->devlist);
277 *private_conf = ERR_PTR(err);
281 /* Find the zone which holds a particular offset
282 * Update *sectorp to be an offset in that zone
284 static struct strip_zone *find_zone(struct r0conf *conf,
288 struct strip_zone *z = conf->strip_zone;
289 sector_t sector = *sectorp;
291 for (i = 0; i < conf->nr_strip_zones; i++)
292 if (sector < z[i].zone_end) {
294 *sectorp = sector - z[i-1].zone_end;
301 * remaps the bio to the target device. we separate two flows.
302 * power 2 flow and a general flow for the sake of performance
304 static struct md_rdev *map_sector(struct mddev *mddev, struct strip_zone *zone,
305 sector_t sector, sector_t *sector_offset)
307 unsigned int sect_in_chunk;
309 struct r0conf *conf = mddev->private;
310 int raid_disks = conf->strip_zone[0].nb_dev;
311 unsigned int chunk_sects = mddev->chunk_sectors;
313 if (is_power_of_2(chunk_sects)) {
314 int chunksect_bits = ffz(~chunk_sects);
315 /* find the sector offset inside the chunk */
316 sect_in_chunk = sector & (chunk_sects - 1);
317 sector >>= chunksect_bits;
319 chunk = *sector_offset;
320 /* quotient is the chunk in real device*/
321 sector_div(chunk, zone->nb_dev << chunksect_bits);
323 sect_in_chunk = sector_div(sector, chunk_sects);
324 chunk = *sector_offset;
325 sector_div(chunk, chunk_sects * zone->nb_dev);
328 * position the bio over the real device
329 * real sector = chunk in device + starting of zone
330 * + the position in the chunk
332 *sector_offset = (chunk * chunk_sects) + sect_in_chunk;
333 return conf->devlist[(zone - conf->strip_zone)*raid_disks
334 + sector_div(sector, zone->nb_dev)];
337 static sector_t raid0_size(struct mddev *mddev, sector_t sectors, int raid_disks)
339 sector_t array_sectors = 0;
340 struct md_rdev *rdev;
342 WARN_ONCE(sectors || raid_disks,
343 "%s does not support generic reshape\n", __func__);
345 rdev_for_each(rdev, mddev)
346 array_sectors += (rdev->sectors &
347 ~(sector_t)(mddev->chunk_sectors-1));
349 return array_sectors;
352 static void raid0_free(struct mddev *mddev, void *priv);
354 static int raid0_run(struct mddev *mddev)
359 if (mddev->chunk_sectors == 0) {
360 printk(KERN_ERR "md/raid0:%s: chunk size must be set.\n",
364 if (md_check_no_bitmap(mddev))
367 /* if private is not null, we are here after takeover */
368 if (mddev->private == NULL) {
369 ret = create_strip_zones(mddev, &conf);
372 mddev->private = conf;
374 conf = mddev->private;
376 struct md_rdev *rdev;
377 bool discard_supported = false;
379 blk_queue_max_hw_sectors(mddev->queue, mddev->chunk_sectors);
380 blk_queue_max_write_same_sectors(mddev->queue, mddev->chunk_sectors);
381 blk_queue_max_discard_sectors(mddev->queue, mddev->chunk_sectors);
383 blk_queue_io_min(mddev->queue, mddev->chunk_sectors << 9);
384 blk_queue_io_opt(mddev->queue,
385 (mddev->chunk_sectors << 9) * mddev->raid_disks);
387 rdev_for_each(rdev, mddev) {
388 disk_stack_limits(mddev->gendisk, rdev->bdev,
389 rdev->data_offset << 9);
390 if (blk_queue_discard(bdev_get_queue(rdev->bdev)))
391 discard_supported = true;
393 if (!discard_supported)
394 queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, mddev->queue);
396 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, mddev->queue);
399 /* calculate array device size */
400 md_set_array_sectors(mddev, raid0_size(mddev, 0, 0));
402 printk(KERN_INFO "md/raid0:%s: md_size is %llu sectors.\n",
404 (unsigned long long)mddev->array_sectors);
407 /* calculate the max read-ahead size.
408 * For read-ahead of large files to be effective, we need to
409 * readahead at least twice a whole stripe. i.e. number of devices
410 * multiplied by chunk size times 2.
411 * If an individual device has an ra_pages greater than the
412 * chunk size, then we will not drive that device as hard as it
413 * wants. We consider this a configuration error: a larger
414 * chunksize should be used in that case.
416 int stripe = mddev->raid_disks *
417 (mddev->chunk_sectors << 9) / PAGE_SIZE;
418 if (mddev->queue->backing_dev_info.ra_pages < 2* stripe)
419 mddev->queue->backing_dev_info.ra_pages = 2* stripe;
424 ret = md_integrity_register(mddev);
429 static void raid0_free(struct mddev *mddev, void *priv)
431 struct r0conf *conf = priv;
433 kfree(conf->strip_zone);
434 kfree(conf->devlist);
439 * Is io distribute over 1 or more chunks ?
441 static inline int is_io_in_chunk_boundary(struct mddev *mddev,
442 unsigned int chunk_sects, struct bio *bio)
444 if (likely(is_power_of_2(chunk_sects))) {
445 return chunk_sects >=
446 ((bio->bi_iter.bi_sector & (chunk_sects-1))
449 sector_t sector = bio->bi_iter.bi_sector;
450 return chunk_sects >= (sector_div(sector, chunk_sects)
455 static void raid0_make_request(struct mddev *mddev, struct bio *bio)
457 struct strip_zone *zone;
458 struct md_rdev *tmp_dev;
461 if (unlikely(bio->bi_rw & REQ_PREFLUSH)) {
462 md_flush_request(mddev, bio);
467 sector_t sector = bio->bi_iter.bi_sector;
468 unsigned chunk_sects = mddev->chunk_sectors;
470 unsigned sectors = chunk_sects -
471 (likely(is_power_of_2(chunk_sects))
472 ? (sector & (chunk_sects-1))
473 : sector_div(sector, chunk_sects));
475 /* Restore due to sector_div */
476 sector = bio->bi_iter.bi_sector;
478 if (sectors < bio_sectors(bio)) {
479 split = bio_split(bio, sectors, GFP_NOIO, fs_bio_set);
480 bio_chain(split, bio);
485 zone = find_zone(mddev->private, §or);
486 tmp_dev = map_sector(mddev, zone, sector, §or);
487 split->bi_bdev = tmp_dev->bdev;
488 split->bi_iter.bi_sector = sector + zone->dev_start +
489 tmp_dev->data_offset;
491 if (unlikely((bio_op(split) == REQ_OP_DISCARD) &&
492 !blk_queue_discard(bdev_get_queue(split->bi_bdev)))) {
496 generic_make_request(split);
497 } while (split != bio);
500 static void raid0_status(struct seq_file *seq, struct mddev *mddev)
502 seq_printf(seq, " %dk chunks", mddev->chunk_sectors / 2);
506 static void *raid0_takeover_raid45(struct mddev *mddev)
508 struct md_rdev *rdev;
509 struct r0conf *priv_conf;
511 if (mddev->degraded != 1) {
512 printk(KERN_ERR "md/raid0:%s: raid5 must be degraded! Degraded disks: %d\n",
515 return ERR_PTR(-EINVAL);
518 rdev_for_each(rdev, mddev) {
519 /* check slot number for a disk */
520 if (rdev->raid_disk == mddev->raid_disks-1) {
521 printk(KERN_ERR "md/raid0:%s: raid5 must have missing parity disk!\n",
523 return ERR_PTR(-EINVAL);
525 rdev->sectors = mddev->dev_sectors;
528 /* Set new parameters */
529 mddev->new_level = 0;
530 mddev->new_layout = 0;
531 mddev->new_chunk_sectors = mddev->chunk_sectors;
533 mddev->delta_disks = -1;
534 /* make sure it will be not marked as dirty */
535 mddev->recovery_cp = MaxSector;
537 create_strip_zones(mddev, &priv_conf);
541 static void *raid0_takeover_raid10(struct mddev *mddev)
543 struct r0conf *priv_conf;
546 * - far_copies must be 1
547 * - near_copies must be 2
548 * - disks number must be even
549 * - all mirrors must be already degraded
551 if (mddev->layout != ((1 << 8) + 2)) {
552 printk(KERN_ERR "md/raid0:%s:: Raid0 cannot takeover layout: 0x%x\n",
555 return ERR_PTR(-EINVAL);
557 if (mddev->raid_disks & 1) {
558 printk(KERN_ERR "md/raid0:%s: Raid0 cannot takeover Raid10 with odd disk number.\n",
560 return ERR_PTR(-EINVAL);
562 if (mddev->degraded != (mddev->raid_disks>>1)) {
563 printk(KERN_ERR "md/raid0:%s: All mirrors must be already degraded!\n",
565 return ERR_PTR(-EINVAL);
568 /* Set new parameters */
569 mddev->new_level = 0;
570 mddev->new_layout = 0;
571 mddev->new_chunk_sectors = mddev->chunk_sectors;
572 mddev->delta_disks = - mddev->raid_disks / 2;
573 mddev->raid_disks += mddev->delta_disks;
575 /* make sure it will be not marked as dirty */
576 mddev->recovery_cp = MaxSector;
578 create_strip_zones(mddev, &priv_conf);
582 static void *raid0_takeover_raid1(struct mddev *mddev)
584 struct r0conf *priv_conf;
588 * - (N - 1) mirror drives must be already faulty
590 if ((mddev->raid_disks - 1) != mddev->degraded) {
591 printk(KERN_ERR "md/raid0:%s: (N - 1) mirrors drives must be already faulty!\n",
593 return ERR_PTR(-EINVAL);
597 * a raid1 doesn't have the notion of chunk size, so
598 * figure out the largest suitable size we can use.
600 chunksect = 64 * 2; /* 64K by default */
602 /* The array must be an exact multiple of chunksize */
603 while (chunksect && (mddev->array_sectors & (chunksect - 1)))
606 if ((chunksect << 9) < PAGE_SIZE)
607 /* array size does not allow a suitable chunk size */
608 return ERR_PTR(-EINVAL);
610 /* Set new parameters */
611 mddev->new_level = 0;
612 mddev->new_layout = 0;
613 mddev->new_chunk_sectors = chunksect;
614 mddev->chunk_sectors = chunksect;
615 mddev->delta_disks = 1 - mddev->raid_disks;
616 mddev->raid_disks = 1;
617 /* make sure it will be not marked as dirty */
618 mddev->recovery_cp = MaxSector;
620 create_strip_zones(mddev, &priv_conf);
624 static void *raid0_takeover(struct mddev *mddev)
626 /* raid0 can take over:
627 * raid4 - if all data disks are active.
628 * raid5 - providing it is Raid4 layout and one disk is faulty
629 * raid10 - assuming we have all necessary active disks
630 * raid1 - with (N -1) mirror drives faulty
634 printk(KERN_ERR "md/raid0: %s: cannot takeover array with bitmap\n",
636 return ERR_PTR(-EBUSY);
638 if (mddev->level == 4)
639 return raid0_takeover_raid45(mddev);
641 if (mddev->level == 5) {
642 if (mddev->layout == ALGORITHM_PARITY_N)
643 return raid0_takeover_raid45(mddev);
645 printk(KERN_ERR "md/raid0:%s: Raid can only takeover Raid5 with layout: %d\n",
646 mdname(mddev), ALGORITHM_PARITY_N);
649 if (mddev->level == 10)
650 return raid0_takeover_raid10(mddev);
652 if (mddev->level == 1)
653 return raid0_takeover_raid1(mddev);
655 printk(KERN_ERR "Takeover from raid%i to raid0 not supported\n",
658 return ERR_PTR(-EINVAL);
661 static void raid0_quiesce(struct mddev *mddev, int state)
665 static struct md_personality raid0_personality=
669 .owner = THIS_MODULE,
670 .make_request = raid0_make_request,
673 .status = raid0_status,
675 .takeover = raid0_takeover,
676 .quiesce = raid0_quiesce,
677 .congested = raid0_congested,
680 static int __init raid0_init (void)
682 return register_md_personality (&raid0_personality);
685 static void raid0_exit (void)
687 unregister_md_personality (&raid0_personality);
690 module_init(raid0_init);
691 module_exit(raid0_exit);
692 MODULE_LICENSE("GPL");
693 MODULE_DESCRIPTION("RAID0 (striping) personality for MD");
694 MODULE_ALIAS("md-personality-2"); /* RAID0 */
695 MODULE_ALIAS("md-raid0");
696 MODULE_ALIAS("md-level-0");