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