Loading...
1/*
2 raid0.c : Multiple Devices driver for Linux
3 Copyright (C) 1994-96 Marc ZYNGIER
4 <zyngier@ufr-info-p7.ibp.fr> or
5 <maz@gloups.fdn.fr>
6 Copyright (C) 1999, 2000 Ingo Molnar, Red Hat
7
8
9 RAID-0 management functions.
10
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)
14 any later version.
15
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.
19*/
20
21#include <linux/blkdev.h>
22#include <linux/seq_file.h>
23#include <linux/module.h>
24#include <linux/slab.h>
25#include "md.h"
26#include "raid0.h"
27#include "raid5.h"
28
29static int raid0_congested(void *data, int bits)
30{
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;
35 int i, ret = 0;
36
37 if (mddev_congested(mddev, bits))
38 return 1;
39
40 for (i = 0; i < raid_disks && !ret ; i++) {
41 struct request_queue *q = bdev_get_queue(devlist[i]->bdev);
42
43 ret |= bdi_congested(&q->backing_dev_info, bits);
44 }
45 return ret;
46}
47
48/*
49 * inform the user of the raid configuration
50*/
51static void dump_zones(struct mddev *mddev)
52{
53 int j, k;
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",
60 mdname(mddev),
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
67 + k]->bdev, b));
68 printk(KERN_CONT "]\n");
69
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;
77 }
78 printk(KERN_INFO "\n");
79}
80
81static int create_strip_zones(struct mddev *mddev, struct r0conf **private_conf)
82{
83 int i, c, err;
84 sector_t curr_zone_end, sectors;
85 struct md_rdev *smallest, *rdev1, *rdev2, *rdev, **dev;
86 struct strip_zone *zone;
87 int cnt;
88 char b[BDEVNAME_SIZE];
89 char b2[BDEVNAME_SIZE];
90 struct r0conf *conf = kzalloc(sizeof(*conf), GFP_KERNEL);
91
92 if (!conf)
93 return -ENOMEM;
94 rdev_for_each(rdev1, mddev) {
95 pr_debug("md/raid0:%s: looking at %s\n",
96 mdname(mddev),
97 bdevname(rdev1->bdev, b));
98 c = 0;
99
100 /* round size to chunk_size */
101 sectors = rdev1->sectors;
102 sector_div(sectors, mddev->chunk_sectors);
103 rdev1->sectors = sectors * mddev->chunk_sectors;
104
105 rdev_for_each(rdev2, mddev) {
106 pr_debug("md/raid0:%s: comparing %s(%llu)"
107 " with %s(%llu)\n",
108 mdname(mddev),
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",
115 mdname(mddev));
116 break;
117 }
118 if (rdev2->sectors == rdev1->sectors) {
119 /*
120 * Not unique, don't count it as a new
121 * group
122 */
123 pr_debug("md/raid0:%s: EQUAL\n",
124 mdname(mddev));
125 c = 1;
126 break;
127 }
128 pr_debug("md/raid0:%s: NOT EQUAL\n",
129 mdname(mddev));
130 }
131 if (!c) {
132 pr_debug("md/raid0:%s: ==> UNIQUE\n",
133 mdname(mddev));
134 conf->nr_strip_zones++;
135 pr_debug("md/raid0:%s: %d zones\n",
136 mdname(mddev), conf->nr_strip_zones);
137 }
138 }
139 pr_debug("md/raid0:%s: FINAL %d zones\n",
140 mdname(mddev), conf->nr_strip_zones);
141 err = -ENOMEM;
142 conf->strip_zone = kzalloc(sizeof(struct strip_zone)*
143 conf->nr_strip_zones, GFP_KERNEL);
144 if (!conf->strip_zone)
145 goto abort;
146 conf->devlist = kzalloc(sizeof(struct md_rdev*)*
147 conf->nr_strip_zones*mddev->raid_disks,
148 GFP_KERNEL);
149 if (!conf->devlist)
150 goto abort;
151
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
154 */
155 zone = &conf->strip_zone[0];
156 cnt = 0;
157 smallest = NULL;
158 dev = conf->devlist;
159 err = -EINVAL;
160 rdev_for_each(rdev1, mddev) {
161 int j = rdev1->raid_disk;
162
163 if (mddev->level == 10) {
164 /* taking over a raid10-n2 array */
165 j /= 2;
166 rdev1->new_raid_disk = j;
167 }
168
169 if (mddev->level == 1) {
170 /* taiking over a raid1 array-
171 * we have only one active disk
172 */
173 j = 0;
174 rdev1->new_raid_disk = j;
175 }
176
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);
180 goto abort;
181 }
182 if (dev[j]) {
183 printk(KERN_ERR "md/raid0:%s: multiple devices for %d - "
184 "aborting!\n", mdname(mddev), j);
185 goto abort;
186 }
187 dev[j] = rdev1;
188
189 disk_stack_limits(mddev->gendisk, rdev1->bdev,
190 rdev1->data_offset << 9);
191
192 if (rdev1->bdev->bd_disk->queue->merge_bvec_fn)
193 conf->has_merge_bvec = 1;
194
195 if (!smallest || (rdev1->sectors < smallest->sectors))
196 smallest = rdev1;
197 cnt++;
198 }
199 if (cnt != mddev->raid_disks) {
200 printk(KERN_ERR "md/raid0:%s: too few disks (%d of %d) - "
201 "aborting!\n", mdname(mddev), cnt, mddev->raid_disks);
202 goto abort;
203 }
204 zone->nb_dev = cnt;
205 zone->zone_end = smallest->sectors * cnt;
206
207 curr_zone_end = zone->zone_end;
208
209 /* now do the other zones */
210 for (i = 1; i < conf->nr_strip_zones; i++)
211 {
212 int j;
213
214 zone = conf->strip_zone + i;
215 dev = conf->devlist + i * mddev->raid_disks;
216
217 pr_debug("md/raid0:%s: zone %d\n", mdname(mddev), i);
218 zone->dev_start = smallest->sectors;
219 smallest = NULL;
220 c = 0;
221
222 for (j=0; j<cnt; j++) {
223 rdev = conf->devlist[j];
224 if (rdev->sectors <= zone->dev_start) {
225 pr_debug("md/raid0:%s: checking %s ... nope\n",
226 mdname(mddev),
227 bdevname(rdev->bdev, b));
228 continue;
229 }
230 pr_debug("md/raid0:%s: checking %s ..."
231 " contained as device %d\n",
232 mdname(mddev),
233 bdevname(rdev->bdev, b), c);
234 dev[c] = rdev;
235 c++;
236 if (!smallest || rdev->sectors < smallest->sectors) {
237 smallest = rdev;
238 pr_debug("md/raid0:%s: (%llu) is smallest!.\n",
239 mdname(mddev),
240 (unsigned long long)rdev->sectors);
241 }
242 }
243
244 zone->nb_dev = c;
245 sectors = (smallest->sectors - zone->dev_start) * c;
246 pr_debug("md/raid0:%s: zone->nb_dev: %d, sectors: %llu\n",
247 mdname(mddev),
248 zone->nb_dev, (unsigned long long)sectors);
249
250 curr_zone_end += sectors;
251 zone->zone_end = curr_zone_end;
252
253 pr_debug("md/raid0:%s: current zone start: %llu\n",
254 mdname(mddev),
255 (unsigned long long)smallest->sectors);
256 }
257 mddev->queue->backing_dev_info.congested_fn = raid0_congested;
258 mddev->queue->backing_dev_info.congested_data = mddev;
259
260 /*
261 * now since we have the hard sector sizes, we can make sure
262 * chunk size is a multiple of that sector size
263 */
264 if ((mddev->chunk_sectors << 9) % queue_logical_block_size(mddev->queue)) {
265 printk(KERN_ERR "md/raid0:%s: chunk_size of %d not valid\n",
266 mdname(mddev),
267 mddev->chunk_sectors << 9);
268 goto abort;
269 }
270
271 blk_queue_io_min(mddev->queue, mddev->chunk_sectors << 9);
272 blk_queue_io_opt(mddev->queue,
273 (mddev->chunk_sectors << 9) * mddev->raid_disks);
274
275 pr_debug("md/raid0:%s: done.\n", mdname(mddev));
276 *private_conf = conf;
277
278 return 0;
279abort:
280 kfree(conf->strip_zone);
281 kfree(conf->devlist);
282 kfree(conf);
283 *private_conf = NULL;
284 return err;
285}
286
287/* Find the zone which holds a particular offset
288 * Update *sectorp to be an offset in that zone
289 */
290static struct strip_zone *find_zone(struct r0conf *conf,
291 sector_t *sectorp)
292{
293 int i;
294 struct strip_zone *z = conf->strip_zone;
295 sector_t sector = *sectorp;
296
297 for (i = 0; i < conf->nr_strip_zones; i++)
298 if (sector < z[i].zone_end) {
299 if (i)
300 *sectorp = sector - z[i-1].zone_end;
301 return z + i;
302 }
303 BUG();
304}
305
306/*
307 * remaps the bio to the target device. we separate two flows.
308 * power 2 flow and a general flow for the sake of perfromance
309*/
310static struct md_rdev *map_sector(struct mddev *mddev, struct strip_zone *zone,
311 sector_t sector, sector_t *sector_offset)
312{
313 unsigned int sect_in_chunk;
314 sector_t chunk;
315 struct r0conf *conf = mddev->private;
316 int raid_disks = conf->strip_zone[0].nb_dev;
317 unsigned int chunk_sects = mddev->chunk_sectors;
318
319 if (is_power_of_2(chunk_sects)) {
320 int chunksect_bits = ffz(~chunk_sects);
321 /* find the sector offset inside the chunk */
322 sect_in_chunk = sector & (chunk_sects - 1);
323 sector >>= chunksect_bits;
324 /* chunk in zone */
325 chunk = *sector_offset;
326 /* quotient is the chunk in real device*/
327 sector_div(chunk, zone->nb_dev << chunksect_bits);
328 } else{
329 sect_in_chunk = sector_div(sector, chunk_sects);
330 chunk = *sector_offset;
331 sector_div(chunk, chunk_sects * zone->nb_dev);
332 }
333 /*
334 * position the bio over the real device
335 * real sector = chunk in device + starting of zone
336 * + the position in the chunk
337 */
338 *sector_offset = (chunk * chunk_sects) + sect_in_chunk;
339 return conf->devlist[(zone - conf->strip_zone)*raid_disks
340 + sector_div(sector, zone->nb_dev)];
341}
342
343/**
344 * raid0_mergeable_bvec -- tell bio layer if two requests can be merged
345 * @q: request queue
346 * @bvm: properties of new bio
347 * @biovec: the request that could be merged to it.
348 *
349 * Return amount of bytes we can accept at this offset
350 */
351static int raid0_mergeable_bvec(struct request_queue *q,
352 struct bvec_merge_data *bvm,
353 struct bio_vec *biovec)
354{
355 struct mddev *mddev = q->queuedata;
356 struct r0conf *conf = mddev->private;
357 sector_t sector = bvm->bi_sector + get_start_sect(bvm->bi_bdev);
358 sector_t sector_offset = sector;
359 int max;
360 unsigned int chunk_sectors = mddev->chunk_sectors;
361 unsigned int bio_sectors = bvm->bi_size >> 9;
362 struct strip_zone *zone;
363 struct md_rdev *rdev;
364 struct request_queue *subq;
365
366 if (is_power_of_2(chunk_sectors))
367 max = (chunk_sectors - ((sector & (chunk_sectors-1))
368 + bio_sectors)) << 9;
369 else
370 max = (chunk_sectors - (sector_div(sector, chunk_sectors)
371 + bio_sectors)) << 9;
372 if (max < 0)
373 max = 0; /* bio_add cannot handle a negative return */
374 if (max <= biovec->bv_len && bio_sectors == 0)
375 return biovec->bv_len;
376 if (max < biovec->bv_len)
377 /* too small already, no need to check further */
378 return max;
379 if (!conf->has_merge_bvec)
380 return max;
381
382 /* May need to check subordinate device */
383 sector = sector_offset;
384 zone = find_zone(mddev->private, §or_offset);
385 rdev = map_sector(mddev, zone, sector, §or_offset);
386 subq = bdev_get_queue(rdev->bdev);
387 if (subq->merge_bvec_fn) {
388 bvm->bi_bdev = rdev->bdev;
389 bvm->bi_sector = sector_offset + zone->dev_start +
390 rdev->data_offset;
391 return min(max, subq->merge_bvec_fn(subq, bvm, biovec));
392 } else
393 return max;
394}
395
396static sector_t raid0_size(struct mddev *mddev, sector_t sectors, int raid_disks)
397{
398 sector_t array_sectors = 0;
399 struct md_rdev *rdev;
400
401 WARN_ONCE(sectors || raid_disks,
402 "%s does not support generic reshape\n", __func__);
403
404 rdev_for_each(rdev, mddev)
405 array_sectors += rdev->sectors;
406
407 return array_sectors;
408}
409
410static int raid0_stop(struct mddev *mddev);
411
412static int raid0_run(struct mddev *mddev)
413{
414 struct r0conf *conf;
415 int ret;
416
417 if (mddev->chunk_sectors == 0) {
418 printk(KERN_ERR "md/raid0:%s: chunk size must be set.\n",
419 mdname(mddev));
420 return -EINVAL;
421 }
422 if (md_check_no_bitmap(mddev))
423 return -EINVAL;
424 blk_queue_max_hw_sectors(mddev->queue, mddev->chunk_sectors);
425
426 /* if private is not null, we are here after takeover */
427 if (mddev->private == NULL) {
428 ret = create_strip_zones(mddev, &conf);
429 if (ret < 0)
430 return ret;
431 mddev->private = conf;
432 }
433 conf = mddev->private;
434
435 /* calculate array device size */
436 md_set_array_sectors(mddev, raid0_size(mddev, 0, 0));
437
438 printk(KERN_INFO "md/raid0:%s: md_size is %llu sectors.\n",
439 mdname(mddev),
440 (unsigned long long)mddev->array_sectors);
441 /* calculate the max read-ahead size.
442 * For read-ahead of large files to be effective, we need to
443 * readahead at least twice a whole stripe. i.e. number of devices
444 * multiplied by chunk size times 2.
445 * If an individual device has an ra_pages greater than the
446 * chunk size, then we will not drive that device as hard as it
447 * wants. We consider this a configuration error: a larger
448 * chunksize should be used in that case.
449 */
450 {
451 int stripe = mddev->raid_disks *
452 (mddev->chunk_sectors << 9) / PAGE_SIZE;
453 if (mddev->queue->backing_dev_info.ra_pages < 2* stripe)
454 mddev->queue->backing_dev_info.ra_pages = 2* stripe;
455 }
456
457 blk_queue_merge_bvec(mddev->queue, raid0_mergeable_bvec);
458 dump_zones(mddev);
459
460 ret = md_integrity_register(mddev);
461 if (ret)
462 raid0_stop(mddev);
463
464 return ret;
465}
466
467static int raid0_stop(struct mddev *mddev)
468{
469 struct r0conf *conf = mddev->private;
470
471 blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/
472 kfree(conf->strip_zone);
473 kfree(conf->devlist);
474 kfree(conf);
475 mddev->private = NULL;
476 return 0;
477}
478
479/*
480 * Is io distribute over 1 or more chunks ?
481*/
482static inline int is_io_in_chunk_boundary(struct mddev *mddev,
483 unsigned int chunk_sects, struct bio *bio)
484{
485 if (likely(is_power_of_2(chunk_sects))) {
486 return chunk_sects >= ((bio->bi_sector & (chunk_sects-1))
487 + (bio->bi_size >> 9));
488 } else{
489 sector_t sector = bio->bi_sector;
490 return chunk_sects >= (sector_div(sector, chunk_sects)
491 + (bio->bi_size >> 9));
492 }
493}
494
495static void raid0_make_request(struct mddev *mddev, struct bio *bio)
496{
497 unsigned int chunk_sects;
498 sector_t sector_offset;
499 struct strip_zone *zone;
500 struct md_rdev *tmp_dev;
501
502 if (unlikely(bio->bi_rw & REQ_FLUSH)) {
503 md_flush_request(mddev, bio);
504 return;
505 }
506
507 chunk_sects = mddev->chunk_sectors;
508 if (unlikely(!is_io_in_chunk_boundary(mddev, chunk_sects, bio))) {
509 sector_t sector = bio->bi_sector;
510 struct bio_pair *bp;
511 /* Sanity check -- queue functions should prevent this happening */
512 if (bio->bi_vcnt != 1 ||
513 bio->bi_idx != 0)
514 goto bad_map;
515 /* This is a one page bio that upper layers
516 * refuse to split for us, so we need to split it.
517 */
518 if (likely(is_power_of_2(chunk_sects)))
519 bp = bio_split(bio, chunk_sects - (sector &
520 (chunk_sects-1)));
521 else
522 bp = bio_split(bio, chunk_sects -
523 sector_div(sector, chunk_sects));
524 raid0_make_request(mddev, &bp->bio1);
525 raid0_make_request(mddev, &bp->bio2);
526 bio_pair_release(bp);
527 return;
528 }
529
530 sector_offset = bio->bi_sector;
531 zone = find_zone(mddev->private, §or_offset);
532 tmp_dev = map_sector(mddev, zone, bio->bi_sector,
533 §or_offset);
534 bio->bi_bdev = tmp_dev->bdev;
535 bio->bi_sector = sector_offset + zone->dev_start +
536 tmp_dev->data_offset;
537
538 generic_make_request(bio);
539 return;
540
541bad_map:
542 printk("md/raid0:%s: make_request bug: can't convert block across chunks"
543 " or bigger than %dk %llu %d\n",
544 mdname(mddev), chunk_sects / 2,
545 (unsigned long long)bio->bi_sector, bio->bi_size >> 10);
546
547 bio_io_error(bio);
548 return;
549}
550
551static void raid0_status(struct seq_file *seq, struct mddev *mddev)
552{
553 seq_printf(seq, " %dk chunks", mddev->chunk_sectors / 2);
554 return;
555}
556
557static void *raid0_takeover_raid45(struct mddev *mddev)
558{
559 struct md_rdev *rdev;
560 struct r0conf *priv_conf;
561
562 if (mddev->degraded != 1) {
563 printk(KERN_ERR "md/raid0:%s: raid5 must be degraded! Degraded disks: %d\n",
564 mdname(mddev),
565 mddev->degraded);
566 return ERR_PTR(-EINVAL);
567 }
568
569 rdev_for_each(rdev, mddev) {
570 /* check slot number for a disk */
571 if (rdev->raid_disk == mddev->raid_disks-1) {
572 printk(KERN_ERR "md/raid0:%s: raid5 must have missing parity disk!\n",
573 mdname(mddev));
574 return ERR_PTR(-EINVAL);
575 }
576 }
577
578 /* Set new parameters */
579 mddev->new_level = 0;
580 mddev->new_layout = 0;
581 mddev->new_chunk_sectors = mddev->chunk_sectors;
582 mddev->raid_disks--;
583 mddev->delta_disks = -1;
584 /* make sure it will be not marked as dirty */
585 mddev->recovery_cp = MaxSector;
586
587 create_strip_zones(mddev, &priv_conf);
588 return priv_conf;
589}
590
591static void *raid0_takeover_raid10(struct mddev *mddev)
592{
593 struct r0conf *priv_conf;
594
595 /* Check layout:
596 * - far_copies must be 1
597 * - near_copies must be 2
598 * - disks number must be even
599 * - all mirrors must be already degraded
600 */
601 if (mddev->layout != ((1 << 8) + 2)) {
602 printk(KERN_ERR "md/raid0:%s:: Raid0 cannot takover layout: 0x%x\n",
603 mdname(mddev),
604 mddev->layout);
605 return ERR_PTR(-EINVAL);
606 }
607 if (mddev->raid_disks & 1) {
608 printk(KERN_ERR "md/raid0:%s: Raid0 cannot takover Raid10 with odd disk number.\n",
609 mdname(mddev));
610 return ERR_PTR(-EINVAL);
611 }
612 if (mddev->degraded != (mddev->raid_disks>>1)) {
613 printk(KERN_ERR "md/raid0:%s: All mirrors must be already degraded!\n",
614 mdname(mddev));
615 return ERR_PTR(-EINVAL);
616 }
617
618 /* Set new parameters */
619 mddev->new_level = 0;
620 mddev->new_layout = 0;
621 mddev->new_chunk_sectors = mddev->chunk_sectors;
622 mddev->delta_disks = - mddev->raid_disks / 2;
623 mddev->raid_disks += mddev->delta_disks;
624 mddev->degraded = 0;
625 /* make sure it will be not marked as dirty */
626 mddev->recovery_cp = MaxSector;
627
628 create_strip_zones(mddev, &priv_conf);
629 return priv_conf;
630}
631
632static void *raid0_takeover_raid1(struct mddev *mddev)
633{
634 struct r0conf *priv_conf;
635 int chunksect;
636
637 /* Check layout:
638 * - (N - 1) mirror drives must be already faulty
639 */
640 if ((mddev->raid_disks - 1) != mddev->degraded) {
641 printk(KERN_ERR "md/raid0:%s: (N - 1) mirrors drives must be already faulty!\n",
642 mdname(mddev));
643 return ERR_PTR(-EINVAL);
644 }
645
646 /*
647 * a raid1 doesn't have the notion of chunk size, so
648 * figure out the largest suitable size we can use.
649 */
650 chunksect = 64 * 2; /* 64K by default */
651
652 /* The array must be an exact multiple of chunksize */
653 while (chunksect && (mddev->array_sectors & (chunksect - 1)))
654 chunksect >>= 1;
655
656 if ((chunksect << 9) < PAGE_SIZE)
657 /* array size does not allow a suitable chunk size */
658 return ERR_PTR(-EINVAL);
659
660 /* Set new parameters */
661 mddev->new_level = 0;
662 mddev->new_layout = 0;
663 mddev->new_chunk_sectors = chunksect;
664 mddev->chunk_sectors = chunksect;
665 mddev->delta_disks = 1 - mddev->raid_disks;
666 mddev->raid_disks = 1;
667 /* make sure it will be not marked as dirty */
668 mddev->recovery_cp = MaxSector;
669
670 create_strip_zones(mddev, &priv_conf);
671 return priv_conf;
672}
673
674static void *raid0_takeover(struct mddev *mddev)
675{
676 /* raid0 can take over:
677 * raid4 - if all data disks are active.
678 * raid5 - providing it is Raid4 layout and one disk is faulty
679 * raid10 - assuming we have all necessary active disks
680 * raid1 - with (N -1) mirror drives faulty
681 */
682 if (mddev->level == 4)
683 return raid0_takeover_raid45(mddev);
684
685 if (mddev->level == 5) {
686 if (mddev->layout == ALGORITHM_PARITY_N)
687 return raid0_takeover_raid45(mddev);
688
689 printk(KERN_ERR "md/raid0:%s: Raid can only takeover Raid5 with layout: %d\n",
690 mdname(mddev), ALGORITHM_PARITY_N);
691 }
692
693 if (mddev->level == 10)
694 return raid0_takeover_raid10(mddev);
695
696 if (mddev->level == 1)
697 return raid0_takeover_raid1(mddev);
698
699 printk(KERN_ERR "Takeover from raid%i to raid0 not supported\n",
700 mddev->level);
701
702 return ERR_PTR(-EINVAL);
703}
704
705static void raid0_quiesce(struct mddev *mddev, int state)
706{
707}
708
709static struct md_personality raid0_personality=
710{
711 .name = "raid0",
712 .level = 0,
713 .owner = THIS_MODULE,
714 .make_request = raid0_make_request,
715 .run = raid0_run,
716 .stop = raid0_stop,
717 .status = raid0_status,
718 .size = raid0_size,
719 .takeover = raid0_takeover,
720 .quiesce = raid0_quiesce,
721};
722
723static int __init raid0_init (void)
724{
725 return register_md_personality (&raid0_personality);
726}
727
728static void raid0_exit (void)
729{
730 unregister_md_personality (&raid0_personality);
731}
732
733module_init(raid0_init);
734module_exit(raid0_exit);
735MODULE_LICENSE("GPL");
736MODULE_DESCRIPTION("RAID0 (striping) personality for MD");
737MODULE_ALIAS("md-personality-2"); /* RAID0 */
738MODULE_ALIAS("md-raid0");
739MODULE_ALIAS("md-level-0");
1// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 raid0.c : Multiple Devices driver for Linux
4 Copyright (C) 1994-96 Marc ZYNGIER
5 <zyngier@ufr-info-p7.ibp.fr> or
6 <maz@gloups.fdn.fr>
7 Copyright (C) 1999, 2000 Ingo Molnar, Red Hat
8
9 RAID-0 management functions.
10
11*/
12
13#include <linux/blkdev.h>
14#include <linux/seq_file.h>
15#include <linux/module.h>
16#include <linux/slab.h>
17#include <trace/events/block.h>
18#include "md.h"
19#include "raid0.h"
20#include "raid5.h"
21
22static int default_layout = 0;
23module_param(default_layout, int, 0644);
24
25#define UNSUPPORTED_MDDEV_FLAGS \
26 ((1L << MD_HAS_JOURNAL) | \
27 (1L << MD_JOURNAL_CLEAN) | \
28 (1L << MD_FAILFAST_SUPPORTED) |\
29 (1L << MD_HAS_PPL) | \
30 (1L << MD_HAS_MULTIPLE_PPLS))
31
32/*
33 * inform the user of the raid configuration
34*/
35static void dump_zones(struct mddev *mddev)
36{
37 int j, k;
38 sector_t zone_size = 0;
39 sector_t zone_start = 0;
40 struct r0conf *conf = mddev->private;
41 int raid_disks = conf->strip_zone[0].nb_dev;
42 pr_debug("md: RAID0 configuration for %s - %d zone%s\n",
43 mdname(mddev),
44 conf->nr_strip_zones, conf->nr_strip_zones==1?"":"s");
45 for (j = 0; j < conf->nr_strip_zones; j++) {
46 char line[200];
47 int len = 0;
48
49 for (k = 0; k < conf->strip_zone[j].nb_dev; k++)
50 len += scnprintf(line+len, 200-len, "%s%pg", k?"/":"",
51 conf->devlist[j * raid_disks + k]->bdev);
52 pr_debug("md: zone%d=[%s]\n", j, line);
53
54 zone_size = conf->strip_zone[j].zone_end - zone_start;
55 pr_debug(" zone-offset=%10lluKB, device-offset=%10lluKB, size=%10lluKB\n",
56 (unsigned long long)zone_start>>1,
57 (unsigned long long)conf->strip_zone[j].dev_start>>1,
58 (unsigned long long)zone_size>>1);
59 zone_start = conf->strip_zone[j].zone_end;
60 }
61}
62
63static int create_strip_zones(struct mddev *mddev, struct r0conf **private_conf)
64{
65 int i, c, err;
66 sector_t curr_zone_end, sectors;
67 struct md_rdev *smallest, *rdev1, *rdev2, *rdev, **dev;
68 struct strip_zone *zone;
69 int cnt;
70 struct r0conf *conf = kzalloc(sizeof(*conf), GFP_KERNEL);
71 unsigned blksize = 512;
72
73 *private_conf = ERR_PTR(-ENOMEM);
74 if (!conf)
75 return -ENOMEM;
76 rdev_for_each(rdev1, mddev) {
77 pr_debug("md/raid0:%s: looking at %pg\n",
78 mdname(mddev),
79 rdev1->bdev);
80 c = 0;
81
82 /* round size to chunk_size */
83 sectors = rdev1->sectors;
84 sector_div(sectors, mddev->chunk_sectors);
85 rdev1->sectors = sectors * mddev->chunk_sectors;
86
87 blksize = max(blksize, queue_logical_block_size(
88 rdev1->bdev->bd_disk->queue));
89
90 rdev_for_each(rdev2, mddev) {
91 pr_debug("md/raid0:%s: comparing %pg(%llu)"
92 " with %pg(%llu)\n",
93 mdname(mddev),
94 rdev1->bdev,
95 (unsigned long long)rdev1->sectors,
96 rdev2->bdev,
97 (unsigned long long)rdev2->sectors);
98 if (rdev2 == rdev1) {
99 pr_debug("md/raid0:%s: END\n",
100 mdname(mddev));
101 break;
102 }
103 if (rdev2->sectors == rdev1->sectors) {
104 /*
105 * Not unique, don't count it as a new
106 * group
107 */
108 pr_debug("md/raid0:%s: EQUAL\n",
109 mdname(mddev));
110 c = 1;
111 break;
112 }
113 pr_debug("md/raid0:%s: NOT EQUAL\n",
114 mdname(mddev));
115 }
116 if (!c) {
117 pr_debug("md/raid0:%s: ==> UNIQUE\n",
118 mdname(mddev));
119 conf->nr_strip_zones++;
120 pr_debug("md/raid0:%s: %d zones\n",
121 mdname(mddev), conf->nr_strip_zones);
122 }
123 }
124 pr_debug("md/raid0:%s: FINAL %d zones\n",
125 mdname(mddev), conf->nr_strip_zones);
126
127 /*
128 * now since we have the hard sector sizes, we can make sure
129 * chunk size is a multiple of that sector size
130 */
131 if ((mddev->chunk_sectors << 9) % blksize) {
132 pr_warn("md/raid0:%s: chunk_size of %d not multiple of block size %d\n",
133 mdname(mddev),
134 mddev->chunk_sectors << 9, blksize);
135 err = -EINVAL;
136 goto abort;
137 }
138
139 err = -ENOMEM;
140 conf->strip_zone = kcalloc(conf->nr_strip_zones,
141 sizeof(struct strip_zone),
142 GFP_KERNEL);
143 if (!conf->strip_zone)
144 goto abort;
145 conf->devlist = kzalloc(array3_size(sizeof(struct md_rdev *),
146 conf->nr_strip_zones,
147 mddev->raid_disks),
148 GFP_KERNEL);
149 if (!conf->devlist)
150 goto abort;
151
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
154 */
155 zone = &conf->strip_zone[0];
156 cnt = 0;
157 smallest = NULL;
158 dev = conf->devlist;
159 err = -EINVAL;
160 rdev_for_each(rdev1, mddev) {
161 int j = rdev1->raid_disk;
162
163 if (mddev->level == 10) {
164 /* taking over a raid10-n2 array */
165 j /= 2;
166 rdev1->new_raid_disk = j;
167 }
168
169 if (mddev->level == 1) {
170 /* taiking over a raid1 array-
171 * we have only one active disk
172 */
173 j = 0;
174 rdev1->new_raid_disk = j;
175 }
176
177 if (j < 0) {
178 pr_warn("md/raid0:%s: remove inactive devices before converting to RAID0\n",
179 mdname(mddev));
180 goto abort;
181 }
182 if (j >= mddev->raid_disks) {
183 pr_warn("md/raid0:%s: bad disk number %d - aborting!\n",
184 mdname(mddev), j);
185 goto abort;
186 }
187 if (dev[j]) {
188 pr_warn("md/raid0:%s: multiple devices for %d - aborting!\n",
189 mdname(mddev), j);
190 goto abort;
191 }
192 dev[j] = rdev1;
193
194 if (!smallest || (rdev1->sectors < smallest->sectors))
195 smallest = rdev1;
196 cnt++;
197 }
198 if (cnt != mddev->raid_disks) {
199 pr_warn("md/raid0:%s: too few disks (%d of %d) - aborting!\n",
200 mdname(mddev), cnt, mddev->raid_disks);
201 goto abort;
202 }
203 zone->nb_dev = cnt;
204 zone->zone_end = smallest->sectors * cnt;
205
206 curr_zone_end = zone->zone_end;
207
208 /* now do the other zones */
209 for (i = 1; i < conf->nr_strip_zones; i++)
210 {
211 int j;
212
213 zone = conf->strip_zone + i;
214 dev = conf->devlist + i * mddev->raid_disks;
215
216 pr_debug("md/raid0:%s: zone %d\n", mdname(mddev), i);
217 zone->dev_start = smallest->sectors;
218 smallest = NULL;
219 c = 0;
220
221 for (j=0; j<cnt; j++) {
222 rdev = conf->devlist[j];
223 if (rdev->sectors <= zone->dev_start) {
224 pr_debug("md/raid0:%s: checking %pg ... nope\n",
225 mdname(mddev),
226 rdev->bdev);
227 continue;
228 }
229 pr_debug("md/raid0:%s: checking %pg ..."
230 " contained as device %d\n",
231 mdname(mddev),
232 rdev->bdev, c);
233 dev[c] = rdev;
234 c++;
235 if (!smallest || rdev->sectors < smallest->sectors) {
236 smallest = rdev;
237 pr_debug("md/raid0:%s: (%llu) is smallest!.\n",
238 mdname(mddev),
239 (unsigned long long)rdev->sectors);
240 }
241 }
242
243 zone->nb_dev = c;
244 sectors = (smallest->sectors - zone->dev_start) * c;
245 pr_debug("md/raid0:%s: zone->nb_dev: %d, sectors: %llu\n",
246 mdname(mddev),
247 zone->nb_dev, (unsigned long long)sectors);
248
249 curr_zone_end += sectors;
250 zone->zone_end = curr_zone_end;
251
252 pr_debug("md/raid0:%s: current zone start: %llu\n",
253 mdname(mddev),
254 (unsigned long long)smallest->sectors);
255 }
256
257 if (conf->nr_strip_zones == 1 || conf->strip_zone[1].nb_dev == 1) {
258 conf->layout = RAID0_ORIG_LAYOUT;
259 } else if (mddev->layout == RAID0_ORIG_LAYOUT ||
260 mddev->layout == RAID0_ALT_MULTIZONE_LAYOUT) {
261 conf->layout = mddev->layout;
262 } else if (default_layout == RAID0_ORIG_LAYOUT ||
263 default_layout == RAID0_ALT_MULTIZONE_LAYOUT) {
264 conf->layout = default_layout;
265 } else {
266 pr_err("md/raid0:%s: cannot assemble multi-zone RAID0 with default_layout setting\n",
267 mdname(mddev));
268 pr_err("md/raid0: please set raid0.default_layout to 1 or 2\n");
269 err = -EOPNOTSUPP;
270 goto abort;
271 }
272
273 if (conf->layout == RAID0_ORIG_LAYOUT) {
274 for (i = 1; i < conf->nr_strip_zones; i++) {
275 sector_t first_sector = conf->strip_zone[i-1].zone_end;
276
277 sector_div(first_sector, mddev->chunk_sectors);
278 zone = conf->strip_zone + i;
279 /* disk_shift is first disk index used in the zone */
280 zone->disk_shift = sector_div(first_sector,
281 zone->nb_dev);
282 }
283 }
284
285 pr_debug("md/raid0:%s: done.\n", mdname(mddev));
286 *private_conf = conf;
287
288 return 0;
289abort:
290 kfree(conf->strip_zone);
291 kfree(conf->devlist);
292 kfree(conf);
293 *private_conf = ERR_PTR(err);
294 return err;
295}
296
297/* Find the zone which holds a particular offset
298 * Update *sectorp to be an offset in that zone
299 */
300static struct strip_zone *find_zone(struct r0conf *conf,
301 sector_t *sectorp)
302{
303 int i;
304 struct strip_zone *z = conf->strip_zone;
305 sector_t sector = *sectorp;
306
307 for (i = 0; i < conf->nr_strip_zones; i++)
308 if (sector < z[i].zone_end) {
309 if (i)
310 *sectorp = sector - z[i-1].zone_end;
311 return z + i;
312 }
313 BUG();
314}
315
316/*
317 * remaps the bio to the target device. we separate two flows.
318 * power 2 flow and a general flow for the sake of performance
319*/
320static struct md_rdev *map_sector(struct mddev *mddev, struct strip_zone *zone,
321 sector_t sector, sector_t *sector_offset)
322{
323 unsigned int sect_in_chunk;
324 sector_t chunk;
325 struct r0conf *conf = mddev->private;
326 int raid_disks = conf->strip_zone[0].nb_dev;
327 unsigned int chunk_sects = mddev->chunk_sectors;
328
329 if (is_power_of_2(chunk_sects)) {
330 int chunksect_bits = ffz(~chunk_sects);
331 /* find the sector offset inside the chunk */
332 sect_in_chunk = sector & (chunk_sects - 1);
333 sector >>= chunksect_bits;
334 /* chunk in zone */
335 chunk = *sector_offset;
336 /* quotient is the chunk in real device*/
337 sector_div(chunk, zone->nb_dev << chunksect_bits);
338 } else{
339 sect_in_chunk = sector_div(sector, chunk_sects);
340 chunk = *sector_offset;
341 sector_div(chunk, chunk_sects * zone->nb_dev);
342 }
343 /*
344 * position the bio over the real device
345 * real sector = chunk in device + starting of zone
346 * + the position in the chunk
347 */
348 *sector_offset = (chunk * chunk_sects) + sect_in_chunk;
349 return conf->devlist[(zone - conf->strip_zone)*raid_disks
350 + sector_div(sector, zone->nb_dev)];
351}
352
353static sector_t raid0_size(struct mddev *mddev, sector_t sectors, int raid_disks)
354{
355 sector_t array_sectors = 0;
356 struct md_rdev *rdev;
357
358 WARN_ONCE(sectors || raid_disks,
359 "%s does not support generic reshape\n", __func__);
360
361 rdev_for_each(rdev, mddev)
362 array_sectors += (rdev->sectors &
363 ~(sector_t)(mddev->chunk_sectors-1));
364
365 return array_sectors;
366}
367
368static void free_conf(struct mddev *mddev, struct r0conf *conf)
369{
370 kfree(conf->strip_zone);
371 kfree(conf->devlist);
372 kfree(conf);
373}
374
375static void raid0_free(struct mddev *mddev, void *priv)
376{
377 struct r0conf *conf = priv;
378
379 free_conf(mddev, conf);
380}
381
382static int raid0_set_limits(struct mddev *mddev)
383{
384 struct queue_limits lim;
385
386 blk_set_stacking_limits(&lim);
387 lim.max_hw_sectors = mddev->chunk_sectors;
388 lim.max_write_zeroes_sectors = mddev->chunk_sectors;
389 lim.io_min = mddev->chunk_sectors << 9;
390 lim.io_opt = lim.io_min * mddev->raid_disks;
391 mddev_stack_rdev_limits(mddev, &lim);
392 return queue_limits_set(mddev->gendisk->queue, &lim);
393}
394
395static int raid0_run(struct mddev *mddev)
396{
397 struct r0conf *conf;
398 int ret;
399
400 if (mddev->chunk_sectors == 0) {
401 pr_warn("md/raid0:%s: chunk size must be set.\n", mdname(mddev));
402 return -EINVAL;
403 }
404 if (md_check_no_bitmap(mddev))
405 return -EINVAL;
406
407 /* if private is not null, we are here after takeover */
408 if (mddev->private == NULL) {
409 ret = create_strip_zones(mddev, &conf);
410 if (ret < 0)
411 return ret;
412 mddev->private = conf;
413 }
414 conf = mddev->private;
415 if (!mddev_is_dm(mddev)) {
416 ret = raid0_set_limits(mddev);
417 if (ret)
418 goto out_free_conf;
419 }
420
421 /* calculate array device size */
422 md_set_array_sectors(mddev, raid0_size(mddev, 0, 0));
423
424 pr_debug("md/raid0:%s: md_size is %llu sectors.\n",
425 mdname(mddev),
426 (unsigned long long)mddev->array_sectors);
427
428 dump_zones(mddev);
429
430 ret = md_integrity_register(mddev);
431 if (ret)
432 goto out_free_conf;
433 return 0;
434out_free_conf:
435 free_conf(mddev, conf);
436 return ret;
437}
438
439/*
440 * Convert disk_index to the disk order in which it is read/written.
441 * For example, if we have 4 disks, they are numbered 0,1,2,3. If we
442 * write the disks starting at disk 3, then the read/write order would
443 * be disk 3, then 0, then 1, and then disk 2 and we want map_disk_shift()
444 * to map the disks as follows 0,1,2,3 => 1,2,3,0. So disk 0 would map
445 * to 1, 1 to 2, 2 to 3, and 3 to 0. That way we can compare disks in
446 * that 'output' space to understand the read/write disk ordering.
447 */
448static int map_disk_shift(int disk_index, int num_disks, int disk_shift)
449{
450 return ((disk_index + num_disks - disk_shift) % num_disks);
451}
452
453static void raid0_handle_discard(struct mddev *mddev, struct bio *bio)
454{
455 struct r0conf *conf = mddev->private;
456 struct strip_zone *zone;
457 sector_t start = bio->bi_iter.bi_sector;
458 sector_t end;
459 unsigned int stripe_size;
460 sector_t first_stripe_index, last_stripe_index;
461 sector_t start_disk_offset;
462 unsigned int start_disk_index;
463 sector_t end_disk_offset;
464 unsigned int end_disk_index;
465 unsigned int disk;
466 sector_t orig_start, orig_end;
467
468 orig_start = start;
469 zone = find_zone(conf, &start);
470
471 if (bio_end_sector(bio) > zone->zone_end) {
472 struct bio *split = bio_split(bio,
473 zone->zone_end - bio->bi_iter.bi_sector, GFP_NOIO,
474 &mddev->bio_set);
475 bio_chain(split, bio);
476 submit_bio_noacct(bio);
477 bio = split;
478 end = zone->zone_end;
479 } else
480 end = bio_end_sector(bio);
481
482 orig_end = end;
483 if (zone != conf->strip_zone)
484 end = end - zone[-1].zone_end;
485
486 /* Now start and end is the offset in zone */
487 stripe_size = zone->nb_dev * mddev->chunk_sectors;
488
489 first_stripe_index = start;
490 sector_div(first_stripe_index, stripe_size);
491 last_stripe_index = end;
492 sector_div(last_stripe_index, stripe_size);
493
494 /* In the first zone the original and alternate layouts are the same */
495 if ((conf->layout == RAID0_ORIG_LAYOUT) && (zone != conf->strip_zone)) {
496 sector_div(orig_start, mddev->chunk_sectors);
497 start_disk_index = sector_div(orig_start, zone->nb_dev);
498 start_disk_index = map_disk_shift(start_disk_index,
499 zone->nb_dev,
500 zone->disk_shift);
501 sector_div(orig_end, mddev->chunk_sectors);
502 end_disk_index = sector_div(orig_end, zone->nb_dev);
503 end_disk_index = map_disk_shift(end_disk_index,
504 zone->nb_dev, zone->disk_shift);
505 } else {
506 start_disk_index = (int)(start - first_stripe_index * stripe_size) /
507 mddev->chunk_sectors;
508 end_disk_index = (int)(end - last_stripe_index * stripe_size) /
509 mddev->chunk_sectors;
510 }
511 start_disk_offset = ((int)(start - first_stripe_index * stripe_size) %
512 mddev->chunk_sectors) +
513 first_stripe_index * mddev->chunk_sectors;
514 end_disk_offset = ((int)(end - last_stripe_index * stripe_size) %
515 mddev->chunk_sectors) +
516 last_stripe_index * mddev->chunk_sectors;
517
518 for (disk = 0; disk < zone->nb_dev; disk++) {
519 sector_t dev_start, dev_end;
520 struct md_rdev *rdev;
521 int compare_disk;
522
523 compare_disk = map_disk_shift(disk, zone->nb_dev,
524 zone->disk_shift);
525
526 if (compare_disk < start_disk_index)
527 dev_start = (first_stripe_index + 1) *
528 mddev->chunk_sectors;
529 else if (compare_disk > start_disk_index)
530 dev_start = first_stripe_index * mddev->chunk_sectors;
531 else
532 dev_start = start_disk_offset;
533
534 if (compare_disk < end_disk_index)
535 dev_end = (last_stripe_index + 1) * mddev->chunk_sectors;
536 else if (compare_disk > end_disk_index)
537 dev_end = last_stripe_index * mddev->chunk_sectors;
538 else
539 dev_end = end_disk_offset;
540
541 if (dev_end <= dev_start)
542 continue;
543
544 rdev = conf->devlist[(zone - conf->strip_zone) *
545 conf->strip_zone[0].nb_dev + disk];
546 md_submit_discard_bio(mddev, rdev, bio,
547 dev_start + zone->dev_start + rdev->data_offset,
548 dev_end - dev_start);
549 }
550 bio_endio(bio);
551}
552
553static void raid0_map_submit_bio(struct mddev *mddev, struct bio *bio)
554{
555 struct r0conf *conf = mddev->private;
556 struct strip_zone *zone;
557 struct md_rdev *tmp_dev;
558 sector_t bio_sector = bio->bi_iter.bi_sector;
559 sector_t sector = bio_sector;
560
561 md_account_bio(mddev, &bio);
562
563 zone = find_zone(mddev->private, §or);
564 switch (conf->layout) {
565 case RAID0_ORIG_LAYOUT:
566 tmp_dev = map_sector(mddev, zone, bio_sector, §or);
567 break;
568 case RAID0_ALT_MULTIZONE_LAYOUT:
569 tmp_dev = map_sector(mddev, zone, sector, §or);
570 break;
571 default:
572 WARN(1, "md/raid0:%s: Invalid layout\n", mdname(mddev));
573 bio_io_error(bio);
574 return;
575 }
576
577 if (unlikely(is_rdev_broken(tmp_dev))) {
578 bio_io_error(bio);
579 md_error(mddev, tmp_dev);
580 return;
581 }
582
583 bio_set_dev(bio, tmp_dev->bdev);
584 bio->bi_iter.bi_sector = sector + zone->dev_start +
585 tmp_dev->data_offset;
586 mddev_trace_remap(mddev, bio, bio_sector);
587 mddev_check_write_zeroes(mddev, bio);
588 submit_bio_noacct(bio);
589}
590
591static bool raid0_make_request(struct mddev *mddev, struct bio *bio)
592{
593 sector_t sector;
594 unsigned chunk_sects;
595 unsigned sectors;
596
597 if (unlikely(bio->bi_opf & REQ_PREFLUSH)
598 && md_flush_request(mddev, bio))
599 return true;
600
601 if (unlikely((bio_op(bio) == REQ_OP_DISCARD))) {
602 raid0_handle_discard(mddev, bio);
603 return true;
604 }
605
606 sector = bio->bi_iter.bi_sector;
607 chunk_sects = mddev->chunk_sectors;
608
609 sectors = chunk_sects -
610 (likely(is_power_of_2(chunk_sects))
611 ? (sector & (chunk_sects-1))
612 : sector_div(sector, chunk_sects));
613
614 if (sectors < bio_sectors(bio)) {
615 struct bio *split = bio_split(bio, sectors, GFP_NOIO,
616 &mddev->bio_set);
617 bio_chain(split, bio);
618 raid0_map_submit_bio(mddev, bio);
619 bio = split;
620 }
621
622 raid0_map_submit_bio(mddev, bio);
623 return true;
624}
625
626static void raid0_status(struct seq_file *seq, struct mddev *mddev)
627{
628 seq_printf(seq, " %dk chunks", mddev->chunk_sectors / 2);
629 return;
630}
631
632static void raid0_error(struct mddev *mddev, struct md_rdev *rdev)
633{
634 if (!test_and_set_bit(MD_BROKEN, &mddev->flags)) {
635 char *md_name = mdname(mddev);
636
637 pr_crit("md/raid0%s: Disk failure on %pg detected, failing array.\n",
638 md_name, rdev->bdev);
639 }
640}
641
642static void *raid0_takeover_raid45(struct mddev *mddev)
643{
644 struct md_rdev *rdev;
645 struct r0conf *priv_conf;
646
647 if (mddev->degraded != 1) {
648 pr_warn("md/raid0:%s: raid5 must be degraded! Degraded disks: %d\n",
649 mdname(mddev),
650 mddev->degraded);
651 return ERR_PTR(-EINVAL);
652 }
653
654 rdev_for_each(rdev, mddev) {
655 /* check slot number for a disk */
656 if (rdev->raid_disk == mddev->raid_disks-1) {
657 pr_warn("md/raid0:%s: raid5 must have missing parity disk!\n",
658 mdname(mddev));
659 return ERR_PTR(-EINVAL);
660 }
661 rdev->sectors = mddev->dev_sectors;
662 }
663
664 /* Set new parameters */
665 mddev->new_level = 0;
666 mddev->new_layout = 0;
667 mddev->new_chunk_sectors = mddev->chunk_sectors;
668 mddev->raid_disks--;
669 mddev->delta_disks = -1;
670 /* make sure it will be not marked as dirty */
671 mddev->recovery_cp = MaxSector;
672 mddev_clear_unsupported_flags(mddev, UNSUPPORTED_MDDEV_FLAGS);
673
674 create_strip_zones(mddev, &priv_conf);
675
676 return priv_conf;
677}
678
679static void *raid0_takeover_raid10(struct mddev *mddev)
680{
681 struct r0conf *priv_conf;
682
683 /* Check layout:
684 * - far_copies must be 1
685 * - near_copies must be 2
686 * - disks number must be even
687 * - all mirrors must be already degraded
688 */
689 if (mddev->layout != ((1 << 8) + 2)) {
690 pr_warn("md/raid0:%s:: Raid0 cannot takeover layout: 0x%x\n",
691 mdname(mddev),
692 mddev->layout);
693 return ERR_PTR(-EINVAL);
694 }
695 if (mddev->raid_disks & 1) {
696 pr_warn("md/raid0:%s: Raid0 cannot takeover Raid10 with odd disk number.\n",
697 mdname(mddev));
698 return ERR_PTR(-EINVAL);
699 }
700 if (mddev->degraded != (mddev->raid_disks>>1)) {
701 pr_warn("md/raid0:%s: All mirrors must be already degraded!\n",
702 mdname(mddev));
703 return ERR_PTR(-EINVAL);
704 }
705
706 /* Set new parameters */
707 mddev->new_level = 0;
708 mddev->new_layout = 0;
709 mddev->new_chunk_sectors = mddev->chunk_sectors;
710 mddev->delta_disks = - mddev->raid_disks / 2;
711 mddev->raid_disks += mddev->delta_disks;
712 mddev->degraded = 0;
713 /* make sure it will be not marked as dirty */
714 mddev->recovery_cp = MaxSector;
715 mddev_clear_unsupported_flags(mddev, UNSUPPORTED_MDDEV_FLAGS);
716
717 create_strip_zones(mddev, &priv_conf);
718 return priv_conf;
719}
720
721static void *raid0_takeover_raid1(struct mddev *mddev)
722{
723 struct r0conf *priv_conf;
724 int chunksect;
725
726 /* Check layout:
727 * - (N - 1) mirror drives must be already faulty
728 */
729 if ((mddev->raid_disks - 1) != mddev->degraded) {
730 pr_err("md/raid0:%s: (N - 1) mirrors drives must be already faulty!\n",
731 mdname(mddev));
732 return ERR_PTR(-EINVAL);
733 }
734
735 /*
736 * a raid1 doesn't have the notion of chunk size, so
737 * figure out the largest suitable size we can use.
738 */
739 chunksect = 64 * 2; /* 64K by default */
740
741 /* The array must be an exact multiple of chunksize */
742 while (chunksect && (mddev->array_sectors & (chunksect - 1)))
743 chunksect >>= 1;
744
745 if ((chunksect << 9) < PAGE_SIZE)
746 /* array size does not allow a suitable chunk size */
747 return ERR_PTR(-EINVAL);
748
749 /* Set new parameters */
750 mddev->new_level = 0;
751 mddev->new_layout = 0;
752 mddev->new_chunk_sectors = chunksect;
753 mddev->chunk_sectors = chunksect;
754 mddev->delta_disks = 1 - mddev->raid_disks;
755 mddev->raid_disks = 1;
756 /* make sure it will be not marked as dirty */
757 mddev->recovery_cp = MaxSector;
758 mddev_clear_unsupported_flags(mddev, UNSUPPORTED_MDDEV_FLAGS);
759
760 create_strip_zones(mddev, &priv_conf);
761 return priv_conf;
762}
763
764static void *raid0_takeover(struct mddev *mddev)
765{
766 /* raid0 can take over:
767 * raid4 - if all data disks are active.
768 * raid5 - providing it is Raid4 layout and one disk is faulty
769 * raid10 - assuming we have all necessary active disks
770 * raid1 - with (N -1) mirror drives faulty
771 */
772
773 if (mddev->bitmap) {
774 pr_warn("md/raid0: %s: cannot takeover array with bitmap\n",
775 mdname(mddev));
776 return ERR_PTR(-EBUSY);
777 }
778 if (mddev->level == 4)
779 return raid0_takeover_raid45(mddev);
780
781 if (mddev->level == 5) {
782 if (mddev->layout == ALGORITHM_PARITY_N)
783 return raid0_takeover_raid45(mddev);
784
785 pr_warn("md/raid0:%s: Raid can only takeover Raid5 with layout: %d\n",
786 mdname(mddev), ALGORITHM_PARITY_N);
787 }
788
789 if (mddev->level == 10)
790 return raid0_takeover_raid10(mddev);
791
792 if (mddev->level == 1)
793 return raid0_takeover_raid1(mddev);
794
795 pr_warn("Takeover from raid%i to raid0 not supported\n",
796 mddev->level);
797
798 return ERR_PTR(-EINVAL);
799}
800
801static void raid0_quiesce(struct mddev *mddev, int quiesce)
802{
803}
804
805static struct md_personality raid0_personality=
806{
807 .name = "raid0",
808 .level = 0,
809 .owner = THIS_MODULE,
810 .make_request = raid0_make_request,
811 .run = raid0_run,
812 .free = raid0_free,
813 .status = raid0_status,
814 .size = raid0_size,
815 .takeover = raid0_takeover,
816 .quiesce = raid0_quiesce,
817 .error_handler = raid0_error,
818};
819
820static int __init raid0_init (void)
821{
822 return register_md_personality (&raid0_personality);
823}
824
825static void raid0_exit (void)
826{
827 unregister_md_personality (&raid0_personality);
828}
829
830module_init(raid0_init);
831module_exit(raid0_exit);
832MODULE_LICENSE("GPL");
833MODULE_DESCRIPTION("RAID0 (striping) personality for MD");
834MODULE_ALIAS("md-personality-2"); /* RAID0 */
835MODULE_ALIAS("md-raid0");
836MODULE_ALIAS("md-level-0");