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
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;
73 printk(KERN_INFO "\n");
76 static int create_strip_zones(struct mddev *mddev, struct r0conf **private_conf)
79 sector_t curr_zone_end, sectors;
80 struct md_rdev *smallest, *rdev1, *rdev2, *rdev, **dev;
81 struct strip_zone *zone;
83 char b[BDEVNAME_SIZE];
84 char b2[BDEVNAME_SIZE];
85 struct r0conf *conf = kzalloc(sizeof(*conf), GFP_KERNEL);
86 unsigned short blksize = 512;
88 *private_conf = ERR_PTR(-ENOMEM);
91 rdev_for_each(rdev1, mddev) {
92 pr_debug("md/raid0:%s: looking at %s\n",
94 bdevname(rdev1->bdev, b));
97 /* round size to chunk_size */
98 sectors = rdev1->sectors;
99 sector_div(sectors, mddev->chunk_sectors);
100 rdev1->sectors = sectors * mddev->chunk_sectors;
102 blksize = max(blksize, queue_logical_block_size(
103 rdev1->bdev->bd_disk->queue));
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 * now since we have the hard sector sizes, we can make sure
143 * chunk size is a multiple of that sector size
145 if ((mddev->chunk_sectors << 9) % blksize) {
146 printk(KERN_ERR "md/raid0:%s: chunk_size of %d not multiple of block size %d\n",
148 mddev->chunk_sectors << 9, blksize);
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(struct md_rdev*)*
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 rdev_for_each(rdev1, mddev) {
173 int j = rdev1->raid_disk;
175 if (mddev->level == 10) {
176 /* taking over a raid10-n2 array */
178 rdev1->new_raid_disk = j;
181 if (mddev->level == 1) {
182 /* taiking over a raid1 array-
183 * we have only one active disk
186 rdev1->new_raid_disk = j;
191 "md/raid0:%s: remove inactive devices before converting to RAID0\n",
195 if (j >= mddev->raid_disks) {
196 printk(KERN_ERR "md/raid0:%s: bad disk number %d - "
197 "aborting!\n", mdname(mddev), j);
201 printk(KERN_ERR "md/raid0:%s: multiple devices for %d - "
202 "aborting!\n", mdname(mddev), j);
207 if (!smallest || (rdev1->sectors < smallest->sectors))
211 if (cnt != mddev->raid_disks) {
212 printk(KERN_ERR "md/raid0:%s: too few disks (%d of %d) - "
213 "aborting!\n", mdname(mddev), cnt, mddev->raid_disks);
217 zone->zone_end = smallest->sectors * cnt;
219 curr_zone_end = zone->zone_end;
221 /* now do the other zones */
222 for (i = 1; i < conf->nr_strip_zones; i++)
226 zone = conf->strip_zone + i;
227 dev = conf->devlist + i * mddev->raid_disks;
229 pr_debug("md/raid0:%s: zone %d\n", mdname(mddev), i);
230 zone->dev_start = smallest->sectors;
234 for (j=0; j<cnt; j++) {
235 rdev = conf->devlist[j];
236 if (rdev->sectors <= zone->dev_start) {
237 pr_debug("md/raid0:%s: checking %s ... nope\n",
239 bdevname(rdev->bdev, b));
242 pr_debug("md/raid0:%s: checking %s ..."
243 " contained as device %d\n",
245 bdevname(rdev->bdev, b), c);
248 if (!smallest || rdev->sectors < smallest->sectors) {
250 pr_debug("md/raid0:%s: (%llu) is smallest!.\n",
252 (unsigned long long)rdev->sectors);
257 sectors = (smallest->sectors - zone->dev_start) * c;
258 pr_debug("md/raid0:%s: zone->nb_dev: %d, sectors: %llu\n",
260 zone->nb_dev, (unsigned long long)sectors);
262 curr_zone_end += sectors;
263 zone->zone_end = curr_zone_end;
265 pr_debug("md/raid0:%s: current zone start: %llu\n",
267 (unsigned long long)smallest->sectors);
270 pr_debug("md/raid0:%s: done.\n", mdname(mddev));
271 *private_conf = conf;
275 kfree(conf->strip_zone);
276 kfree(conf->devlist);
278 *private_conf = ERR_PTR(err);
282 /* Find the zone which holds a particular offset
283 * Update *sectorp to be an offset in that zone
285 static struct strip_zone *find_zone(struct r0conf *conf,
289 struct strip_zone *z = conf->strip_zone;
290 sector_t sector = *sectorp;
292 for (i = 0; i < conf->nr_strip_zones; i++)
293 if (sector < z[i].zone_end) {
295 *sectorp = sector - z[i-1].zone_end;
302 * remaps the bio to the target device. we separate two flows.
303 * power 2 flow and a general flow for the sake of performance
305 static struct md_rdev *map_sector(struct mddev *mddev, struct strip_zone *zone,
306 sector_t sector, sector_t *sector_offset)
308 unsigned int sect_in_chunk;
310 struct r0conf *conf = mddev->private;
311 int raid_disks = conf->strip_zone[0].nb_dev;
312 unsigned int chunk_sects = mddev->chunk_sectors;
314 if (is_power_of_2(chunk_sects)) {
315 int chunksect_bits = ffz(~chunk_sects);
316 /* find the sector offset inside the chunk */
317 sect_in_chunk = sector & (chunk_sects - 1);
318 sector >>= chunksect_bits;
320 chunk = *sector_offset;
321 /* quotient is the chunk in real device*/
322 sector_div(chunk, zone->nb_dev << chunksect_bits);
324 sect_in_chunk = sector_div(sector, chunk_sects);
325 chunk = *sector_offset;
326 sector_div(chunk, chunk_sects * zone->nb_dev);
329 * position the bio over the real device
330 * real sector = chunk in device + starting of zone
331 * + the position in the chunk
333 *sector_offset = (chunk * chunk_sects) + sect_in_chunk;
334 return conf->devlist[(zone - conf->strip_zone)*raid_disks
335 + sector_div(sector, zone->nb_dev)];
338 static sector_t raid0_size(struct mddev *mddev, sector_t sectors, int raid_disks)
340 sector_t array_sectors = 0;
341 struct md_rdev *rdev;
343 WARN_ONCE(sectors || raid_disks,
344 "%s does not support generic reshape\n", __func__);
346 rdev_for_each(rdev, mddev)
347 array_sectors += (rdev->sectors &
348 ~(sector_t)(mddev->chunk_sectors-1));
350 return array_sectors;
353 static void raid0_free(struct mddev *mddev, void *priv);
355 static int raid0_run(struct mddev *mddev)
360 if (mddev->chunk_sectors == 0) {
361 printk(KERN_ERR "md/raid0:%s: chunk size must be set.\n",
365 if (md_check_no_bitmap(mddev))
368 /* if private is not null, we are here after takeover */
369 if (mddev->private == NULL) {
370 ret = create_strip_zones(mddev, &conf);
373 mddev->private = conf;
375 conf = mddev->private;
377 struct md_rdev *rdev;
378 bool discard_supported = false;
380 blk_queue_max_hw_sectors(mddev->queue, mddev->chunk_sectors);
381 blk_queue_max_write_same_sectors(mddev->queue, mddev->chunk_sectors);
382 blk_queue_max_discard_sectors(mddev->queue, mddev->chunk_sectors);
384 blk_queue_io_min(mddev->queue, mddev->chunk_sectors << 9);
385 blk_queue_io_opt(mddev->queue,
386 (mddev->chunk_sectors << 9) * mddev->raid_disks);
388 rdev_for_each(rdev, mddev) {
389 disk_stack_limits(mddev->gendisk, rdev->bdev,
390 rdev->data_offset << 9);
391 if (blk_queue_discard(bdev_get_queue(rdev->bdev)))
392 discard_supported = true;
394 if (!discard_supported)
395 queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, mddev->queue);
397 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, mddev->queue);
400 /* calculate array device size */
401 md_set_array_sectors(mddev, raid0_size(mddev, 0, 0));
403 printk(KERN_INFO "md/raid0:%s: md_size is %llu sectors.\n",
405 (unsigned long long)mddev->array_sectors);
408 /* calculate the max read-ahead size.
409 * For read-ahead of large files to be effective, we need to
410 * readahead at least twice a whole stripe. i.e. number of devices
411 * multiplied by chunk size times 2.
412 * If an individual device has an ra_pages greater than the
413 * chunk size, then we will not drive that device as hard as it
414 * wants. We consider this a configuration error: a larger
415 * chunksize should be used in that case.
417 int stripe = mddev->raid_disks *
418 (mddev->chunk_sectors << 9) / PAGE_SIZE;
419 if (mddev->queue->backing_dev_info.ra_pages < 2* stripe)
420 mddev->queue->backing_dev_info.ra_pages = 2* stripe;
425 ret = md_integrity_register(mddev);
430 static void raid0_free(struct mddev *mddev, void *priv)
432 struct r0conf *conf = priv;
434 kfree(conf->strip_zone);
435 kfree(conf->devlist);
440 * Is io distribute over 1 or more chunks ?
442 static inline int is_io_in_chunk_boundary(struct mddev *mddev,
443 unsigned int chunk_sects, struct bio *bio)
445 if (likely(is_power_of_2(chunk_sects))) {
446 return chunk_sects >=
447 ((bio->bi_iter.bi_sector & (chunk_sects-1))
450 sector_t sector = bio->bi_iter.bi_sector;
451 return chunk_sects >= (sector_div(sector, chunk_sects)
456 static void raid0_make_request(struct mddev *mddev, struct bio *bio)
458 struct strip_zone *zone;
459 struct md_rdev *tmp_dev;
462 if (unlikely(bio->bi_rw & REQ_FLUSH)) {
463 md_flush_request(mddev, bio);
468 sector_t sector = bio->bi_iter.bi_sector;
469 unsigned chunk_sects = mddev->chunk_sectors;
471 unsigned sectors = chunk_sects -
472 (likely(is_power_of_2(chunk_sects))
473 ? (sector & (chunk_sects-1))
474 : sector_div(sector, chunk_sects));
476 /* Restore due to sector_div */
477 sector = bio->bi_iter.bi_sector;
479 if (sectors < bio_sectors(bio)) {
480 split = bio_split(bio, sectors, GFP_NOIO, fs_bio_set);
481 bio_chain(split, bio);
486 zone = find_zone(mddev->private, §or);
487 tmp_dev = map_sector(mddev, zone, sector, §or);
488 split->bi_bdev = tmp_dev->bdev;
489 split->bi_iter.bi_sector = sector + zone->dev_start +
490 tmp_dev->data_offset;
492 if (unlikely((split->bi_rw & REQ_DISCARD) &&
493 !blk_queue_discard(bdev_get_queue(split->bi_bdev)))) {
497 generic_make_request(split);
498 } while (split != bio);
501 static void raid0_status(struct seq_file *seq, struct mddev *mddev)
503 seq_printf(seq, " %dk chunks", mddev->chunk_sectors / 2);
507 static void *raid0_takeover_raid45(struct mddev *mddev)
509 struct md_rdev *rdev;
510 struct r0conf *priv_conf;
512 if (mddev->degraded != 1) {
513 printk(KERN_ERR "md/raid0:%s: raid5 must be degraded! Degraded disks: %d\n",
516 return ERR_PTR(-EINVAL);
519 rdev_for_each(rdev, mddev) {
520 /* check slot number for a disk */
521 if (rdev->raid_disk == mddev->raid_disks-1) {
522 printk(KERN_ERR "md/raid0:%s: raid5 must have missing parity disk!\n",
524 return ERR_PTR(-EINVAL);
526 rdev->sectors = mddev->dev_sectors;
529 /* Set new parameters */
530 mddev->new_level = 0;
531 mddev->new_layout = 0;
532 mddev->new_chunk_sectors = mddev->chunk_sectors;
534 mddev->delta_disks = -1;
535 /* make sure it will be not marked as dirty */
536 mddev->recovery_cp = MaxSector;
538 create_strip_zones(mddev, &priv_conf);
542 static void *raid0_takeover_raid10(struct mddev *mddev)
544 struct r0conf *priv_conf;
547 * - far_copies must be 1
548 * - near_copies must be 2
549 * - disks number must be even
550 * - all mirrors must be already degraded
552 if (mddev->layout != ((1 << 8) + 2)) {
553 printk(KERN_ERR "md/raid0:%s:: Raid0 cannot takeover layout: 0x%x\n",
556 return ERR_PTR(-EINVAL);
558 if (mddev->raid_disks & 1) {
559 printk(KERN_ERR "md/raid0:%s: Raid0 cannot takeover Raid10 with odd disk number.\n",
561 return ERR_PTR(-EINVAL);
563 if (mddev->degraded != (mddev->raid_disks>>1)) {
564 printk(KERN_ERR "md/raid0:%s: All mirrors must be already degraded!\n",
566 return ERR_PTR(-EINVAL);
569 /* Set new parameters */
570 mddev->new_level = 0;
571 mddev->new_layout = 0;
572 mddev->new_chunk_sectors = mddev->chunk_sectors;
573 mddev->delta_disks = - mddev->raid_disks / 2;
574 mddev->raid_disks += mddev->delta_disks;
576 /* make sure it will be not marked as dirty */
577 mddev->recovery_cp = MaxSector;
579 create_strip_zones(mddev, &priv_conf);
583 static void *raid0_takeover_raid1(struct mddev *mddev)
585 struct r0conf *priv_conf;
589 * - (N - 1) mirror drives must be already faulty
591 if ((mddev->raid_disks - 1) != mddev->degraded) {
592 printk(KERN_ERR "md/raid0:%s: (N - 1) mirrors drives must be already faulty!\n",
594 return ERR_PTR(-EINVAL);
598 * a raid1 doesn't have the notion of chunk size, so
599 * figure out the largest suitable size we can use.
601 chunksect = 64 * 2; /* 64K by default */
603 /* The array must be an exact multiple of chunksize */
604 while (chunksect && (mddev->array_sectors & (chunksect - 1)))
607 if ((chunksect << 9) < PAGE_SIZE)
608 /* array size does not allow a suitable chunk size */
609 return ERR_PTR(-EINVAL);
611 /* Set new parameters */
612 mddev->new_level = 0;
613 mddev->new_layout = 0;
614 mddev->new_chunk_sectors = chunksect;
615 mddev->chunk_sectors = chunksect;
616 mddev->delta_disks = 1 - mddev->raid_disks;
617 mddev->raid_disks = 1;
618 /* make sure it will be not marked as dirty */
619 mddev->recovery_cp = MaxSector;
621 create_strip_zones(mddev, &priv_conf);
625 static void *raid0_takeover(struct mddev *mddev)
627 /* raid0 can take over:
628 * raid4 - if all data disks are active.
629 * raid5 - providing it is Raid4 layout and one disk is faulty
630 * raid10 - assuming we have all necessary active disks
631 * raid1 - with (N -1) mirror drives faulty
635 printk(KERN_ERR "md/raid0: %s: cannot takeover array with bitmap\n",
637 return ERR_PTR(-EBUSY);
639 if (mddev->level == 4)
640 return raid0_takeover_raid45(mddev);
642 if (mddev->level == 5) {
643 if (mddev->layout == ALGORITHM_PARITY_N)
644 return raid0_takeover_raid45(mddev);
646 printk(KERN_ERR "md/raid0:%s: Raid can only takeover Raid5 with layout: %d\n",
647 mdname(mddev), ALGORITHM_PARITY_N);
650 if (mddev->level == 10)
651 return raid0_takeover_raid10(mddev);
653 if (mddev->level == 1)
654 return raid0_takeover_raid1(mddev);
656 printk(KERN_ERR "Takeover from raid%i to raid0 not supported\n",
659 return ERR_PTR(-EINVAL);
662 static void raid0_quiesce(struct mddev *mddev, int state)
666 static struct md_personality raid0_personality=
670 .owner = THIS_MODULE,
671 .make_request = raid0_make_request,
674 .status = raid0_status,
676 .takeover = raid0_takeover,
677 .quiesce = raid0_quiesce,
678 .congested = raid0_congested,
681 static int __init raid0_init (void)
683 return register_md_personality (&raid0_personality);
686 static void raid0_exit (void)
688 unregister_md_personality (&raid0_personality);
691 module_init(raid0_init);
692 module_exit(raid0_exit);
693 MODULE_LICENSE("GPL");
694 MODULE_DESCRIPTION("RAID0 (striping) personality for MD");
695 MODULE_ALIAS("md-personality-2"); /* RAID0 */
696 MODULE_ALIAS("md-raid0");
697 MODULE_ALIAS("md-level-0");