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/*
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 RAID-0 management functions.
9
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)
13 any later version.
14
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.
18*/
19
20#include <linux/blkdev.h>
21#include <linux/seq_file.h>
22#include <linux/module.h>
23#include <linux/slab.h>
24#include <trace/events/block.h>
25#include "md.h"
26#include "raid0.h"
27#include "raid5.h"
28
29#define UNSUPPORTED_MDDEV_FLAGS \
30 ((1L << MD_HAS_JOURNAL) | \
31 (1L << MD_JOURNAL_CLEAN) | \
32 (1L << MD_FAILFAST_SUPPORTED) |\
33 (1L << MD_HAS_PPL) | \
34 (1L << MD_HAS_MULTIPLE_PPLS))
35
36static int raid0_congested(struct mddev *mddev, int bits)
37{
38 struct r0conf *conf = mddev->private;
39 struct md_rdev **devlist = conf->devlist;
40 int raid_disks = conf->strip_zone[0].nb_dev;
41 int i, ret = 0;
42
43 for (i = 0; i < raid_disks && !ret ; i++) {
44 struct request_queue *q = bdev_get_queue(devlist[i]->bdev);
45
46 ret |= bdi_congested(q->backing_dev_info, bits);
47 }
48 return ret;
49}
50
51/*
52 * inform the user of the raid configuration
53*/
54static void dump_zones(struct mddev *mddev)
55{
56 int j, k;
57 sector_t zone_size = 0;
58 sector_t zone_start = 0;
59 char b[BDEVNAME_SIZE];
60 struct r0conf *conf = mddev->private;
61 int raid_disks = conf->strip_zone[0].nb_dev;
62 pr_debug("md: RAID0 configuration for %s - %d zone%s\n",
63 mdname(mddev),
64 conf->nr_strip_zones, conf->nr_strip_zones==1?"":"s");
65 for (j = 0; j < conf->nr_strip_zones; j++) {
66 char line[200];
67 int len = 0;
68
69 for (k = 0; k < conf->strip_zone[j].nb_dev; k++)
70 len += snprintf(line+len, 200-len, "%s%s", k?"/":"",
71 bdevname(conf->devlist[j*raid_disks
72 + k]->bdev, b));
73 pr_debug("md: zone%d=[%s]\n", j, line);
74
75 zone_size = conf->strip_zone[j].zone_end - zone_start;
76 pr_debug(" zone-offset=%10lluKB, device-offset=%10lluKB, size=%10lluKB\n",
77 (unsigned long long)zone_start>>1,
78 (unsigned long long)conf->strip_zone[j].dev_start>>1,
79 (unsigned long long)zone_size>>1);
80 zone_start = conf->strip_zone[j].zone_end;
81 }
82}
83
84static int create_strip_zones(struct mddev *mddev, struct r0conf **private_conf)
85{
86 int i, c, err;
87 sector_t curr_zone_end, sectors;
88 struct md_rdev *smallest, *rdev1, *rdev2, *rdev, **dev;
89 struct strip_zone *zone;
90 int cnt;
91 char b[BDEVNAME_SIZE];
92 char b2[BDEVNAME_SIZE];
93 struct r0conf *conf = kzalloc(sizeof(*conf), GFP_KERNEL);
94 unsigned short blksize = 512;
95
96 *private_conf = ERR_PTR(-ENOMEM);
97 if (!conf)
98 return -ENOMEM;
99 rdev_for_each(rdev1, mddev) {
100 pr_debug("md/raid0:%s: looking at %s\n",
101 mdname(mddev),
102 bdevname(rdev1->bdev, b));
103 c = 0;
104
105 /* round size to chunk_size */
106 sectors = rdev1->sectors;
107 sector_div(sectors, mddev->chunk_sectors);
108 rdev1->sectors = sectors * mddev->chunk_sectors;
109
110 blksize = max(blksize, queue_logical_block_size(
111 rdev1->bdev->bd_disk->queue));
112
113 rdev_for_each(rdev2, mddev) {
114 pr_debug("md/raid0:%s: comparing %s(%llu)"
115 " with %s(%llu)\n",
116 mdname(mddev),
117 bdevname(rdev1->bdev,b),
118 (unsigned long long)rdev1->sectors,
119 bdevname(rdev2->bdev,b2),
120 (unsigned long long)rdev2->sectors);
121 if (rdev2 == rdev1) {
122 pr_debug("md/raid0:%s: END\n",
123 mdname(mddev));
124 break;
125 }
126 if (rdev2->sectors == rdev1->sectors) {
127 /*
128 * Not unique, don't count it as a new
129 * group
130 */
131 pr_debug("md/raid0:%s: EQUAL\n",
132 mdname(mddev));
133 c = 1;
134 break;
135 }
136 pr_debug("md/raid0:%s: NOT EQUAL\n",
137 mdname(mddev));
138 }
139 if (!c) {
140 pr_debug("md/raid0:%s: ==> UNIQUE\n",
141 mdname(mddev));
142 conf->nr_strip_zones++;
143 pr_debug("md/raid0:%s: %d zones\n",
144 mdname(mddev), conf->nr_strip_zones);
145 }
146 }
147 pr_debug("md/raid0:%s: FINAL %d zones\n",
148 mdname(mddev), conf->nr_strip_zones);
149 /*
150 * now since we have the hard sector sizes, we can make sure
151 * chunk size is a multiple of that sector size
152 */
153 if ((mddev->chunk_sectors << 9) % blksize) {
154 pr_warn("md/raid0:%s: chunk_size of %d not multiple of block size %d\n",
155 mdname(mddev),
156 mddev->chunk_sectors << 9, blksize);
157 err = -EINVAL;
158 goto abort;
159 }
160
161 err = -ENOMEM;
162 conf->strip_zone = kzalloc(sizeof(struct strip_zone)*
163 conf->nr_strip_zones, GFP_KERNEL);
164 if (!conf->strip_zone)
165 goto abort;
166 conf->devlist = kzalloc(sizeof(struct md_rdev*)*
167 conf->nr_strip_zones*mddev->raid_disks,
168 GFP_KERNEL);
169 if (!conf->devlist)
170 goto abort;
171
172 /* The first zone must contain all devices, so here we check that
173 * there is a proper alignment of slots to devices and find them all
174 */
175 zone = &conf->strip_zone[0];
176 cnt = 0;
177 smallest = NULL;
178 dev = conf->devlist;
179 err = -EINVAL;
180 rdev_for_each(rdev1, mddev) {
181 int j = rdev1->raid_disk;
182
183 if (mddev->level == 10) {
184 /* taking over a raid10-n2 array */
185 j /= 2;
186 rdev1->new_raid_disk = j;
187 }
188
189 if (mddev->level == 1) {
190 /* taiking over a raid1 array-
191 * we have only one active disk
192 */
193 j = 0;
194 rdev1->new_raid_disk = j;
195 }
196
197 if (j < 0) {
198 pr_warn("md/raid0:%s: remove inactive devices before converting to RAID0\n",
199 mdname(mddev));
200 goto abort;
201 }
202 if (j >= mddev->raid_disks) {
203 pr_warn("md/raid0:%s: bad disk number %d - aborting!\n",
204 mdname(mddev), j);
205 goto abort;
206 }
207 if (dev[j]) {
208 pr_warn("md/raid0:%s: multiple devices for %d - aborting!\n",
209 mdname(mddev), j);
210 goto abort;
211 }
212 dev[j] = rdev1;
213
214 if (!smallest || (rdev1->sectors < smallest->sectors))
215 smallest = rdev1;
216 cnt++;
217 }
218 if (cnt != mddev->raid_disks) {
219 pr_warn("md/raid0:%s: too few disks (%d of %d) - aborting!\n",
220 mdname(mddev), cnt, mddev->raid_disks);
221 goto abort;
222 }
223 zone->nb_dev = cnt;
224 zone->zone_end = smallest->sectors * cnt;
225
226 curr_zone_end = zone->zone_end;
227
228 /* now do the other zones */
229 for (i = 1; i < conf->nr_strip_zones; i++)
230 {
231 int j;
232
233 zone = conf->strip_zone + i;
234 dev = conf->devlist + i * mddev->raid_disks;
235
236 pr_debug("md/raid0:%s: zone %d\n", mdname(mddev), i);
237 zone->dev_start = smallest->sectors;
238 smallest = NULL;
239 c = 0;
240
241 for (j=0; j<cnt; j++) {
242 rdev = conf->devlist[j];
243 if (rdev->sectors <= zone->dev_start) {
244 pr_debug("md/raid0:%s: checking %s ... nope\n",
245 mdname(mddev),
246 bdevname(rdev->bdev, b));
247 continue;
248 }
249 pr_debug("md/raid0:%s: checking %s ..."
250 " contained as device %d\n",
251 mdname(mddev),
252 bdevname(rdev->bdev, b), c);
253 dev[c] = rdev;
254 c++;
255 if (!smallest || rdev->sectors < smallest->sectors) {
256 smallest = rdev;
257 pr_debug("md/raid0:%s: (%llu) is smallest!.\n",
258 mdname(mddev),
259 (unsigned long long)rdev->sectors);
260 }
261 }
262
263 zone->nb_dev = c;
264 sectors = (smallest->sectors - zone->dev_start) * c;
265 pr_debug("md/raid0:%s: zone->nb_dev: %d, sectors: %llu\n",
266 mdname(mddev),
267 zone->nb_dev, (unsigned long long)sectors);
268
269 curr_zone_end += sectors;
270 zone->zone_end = curr_zone_end;
271
272 pr_debug("md/raid0:%s: current zone start: %llu\n",
273 mdname(mddev),
274 (unsigned long long)smallest->sectors);
275 }
276
277 pr_debug("md/raid0:%s: done.\n", mdname(mddev));
278 *private_conf = conf;
279
280 return 0;
281abort:
282 kfree(conf->strip_zone);
283 kfree(conf->devlist);
284 kfree(conf);
285 *private_conf = ERR_PTR(err);
286 return err;
287}
288
289/* Find the zone which holds a particular offset
290 * Update *sectorp to be an offset in that zone
291 */
292static struct strip_zone *find_zone(struct r0conf *conf,
293 sector_t *sectorp)
294{
295 int i;
296 struct strip_zone *z = conf->strip_zone;
297 sector_t sector = *sectorp;
298
299 for (i = 0; i < conf->nr_strip_zones; i++)
300 if (sector < z[i].zone_end) {
301 if (i)
302 *sectorp = sector - z[i-1].zone_end;
303 return z + i;
304 }
305 BUG();
306}
307
308/*
309 * remaps the bio to the target device. we separate two flows.
310 * power 2 flow and a general flow for the sake of performance
311*/
312static struct md_rdev *map_sector(struct mddev *mddev, struct strip_zone *zone,
313 sector_t sector, sector_t *sector_offset)
314{
315 unsigned int sect_in_chunk;
316 sector_t chunk;
317 struct r0conf *conf = mddev->private;
318 int raid_disks = conf->strip_zone[0].nb_dev;
319 unsigned int chunk_sects = mddev->chunk_sectors;
320
321 if (is_power_of_2(chunk_sects)) {
322 int chunksect_bits = ffz(~chunk_sects);
323 /* find the sector offset inside the chunk */
324 sect_in_chunk = sector & (chunk_sects - 1);
325 sector >>= chunksect_bits;
326 /* chunk in zone */
327 chunk = *sector_offset;
328 /* quotient is the chunk in real device*/
329 sector_div(chunk, zone->nb_dev << chunksect_bits);
330 } else{
331 sect_in_chunk = sector_div(sector, chunk_sects);
332 chunk = *sector_offset;
333 sector_div(chunk, chunk_sects * zone->nb_dev);
334 }
335 /*
336 * position the bio over the real device
337 * real sector = chunk in device + starting of zone
338 * + the position in the chunk
339 */
340 *sector_offset = (chunk * chunk_sects) + sect_in_chunk;
341 return conf->devlist[(zone - conf->strip_zone)*raid_disks
342 + sector_div(sector, zone->nb_dev)];
343}
344
345static sector_t raid0_size(struct mddev *mddev, sector_t sectors, int raid_disks)
346{
347 sector_t array_sectors = 0;
348 struct md_rdev *rdev;
349
350 WARN_ONCE(sectors || raid_disks,
351 "%s does not support generic reshape\n", __func__);
352
353 rdev_for_each(rdev, mddev)
354 array_sectors += (rdev->sectors &
355 ~(sector_t)(mddev->chunk_sectors-1));
356
357 return array_sectors;
358}
359
360static void raid0_free(struct mddev *mddev, void *priv);
361
362static int raid0_run(struct mddev *mddev)
363{
364 struct r0conf *conf;
365 int ret;
366
367 if (mddev->chunk_sectors == 0) {
368 pr_warn("md/raid0:%s: chunk size must be set.\n", mdname(mddev));
369 return -EINVAL;
370 }
371 if (md_check_no_bitmap(mddev))
372 return -EINVAL;
373
374 /* if private is not null, we are here after takeover */
375 if (mddev->private == NULL) {
376 ret = create_strip_zones(mddev, &conf);
377 if (ret < 0)
378 return ret;
379 mddev->private = conf;
380 }
381 conf = mddev->private;
382 if (mddev->queue) {
383 struct md_rdev *rdev;
384 bool discard_supported = false;
385
386 blk_queue_max_hw_sectors(mddev->queue, mddev->chunk_sectors);
387 blk_queue_max_write_same_sectors(mddev->queue, mddev->chunk_sectors);
388 blk_queue_max_write_zeroes_sectors(mddev->queue, mddev->chunk_sectors);
389 blk_queue_max_discard_sectors(mddev->queue, UINT_MAX);
390
391 blk_queue_io_min(mddev->queue, mddev->chunk_sectors << 9);
392 blk_queue_io_opt(mddev->queue,
393 (mddev->chunk_sectors << 9) * mddev->raid_disks);
394
395 rdev_for_each(rdev, mddev) {
396 disk_stack_limits(mddev->gendisk, rdev->bdev,
397 rdev->data_offset << 9);
398 if (blk_queue_discard(bdev_get_queue(rdev->bdev)))
399 discard_supported = true;
400 }
401 if (!discard_supported)
402 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, mddev->queue);
403 else
404 blk_queue_flag_set(QUEUE_FLAG_DISCARD, mddev->queue);
405 }
406
407 /* calculate array device size */
408 md_set_array_sectors(mddev, raid0_size(mddev, 0, 0));
409
410 pr_debug("md/raid0:%s: md_size is %llu sectors.\n",
411 mdname(mddev),
412 (unsigned long long)mddev->array_sectors);
413
414 if (mddev->queue) {
415 /* calculate the max read-ahead size.
416 * For read-ahead of large files to be effective, we need to
417 * readahead at least twice a whole stripe. i.e. number of devices
418 * multiplied by chunk size times 2.
419 * If an individual device has an ra_pages greater than the
420 * chunk size, then we will not drive that device as hard as it
421 * wants. We consider this a configuration error: a larger
422 * chunksize should be used in that case.
423 */
424 int stripe = mddev->raid_disks *
425 (mddev->chunk_sectors << 9) / PAGE_SIZE;
426 if (mddev->queue->backing_dev_info->ra_pages < 2* stripe)
427 mddev->queue->backing_dev_info->ra_pages = 2* stripe;
428 }
429
430 dump_zones(mddev);
431
432 ret = md_integrity_register(mddev);
433
434 return ret;
435}
436
437static void raid0_free(struct mddev *mddev, void *priv)
438{
439 struct r0conf *conf = priv;
440
441 kfree(conf->strip_zone);
442 kfree(conf->devlist);
443 kfree(conf);
444}
445
446/*
447 * Is io distribute over 1 or more chunks ?
448*/
449static inline int is_io_in_chunk_boundary(struct mddev *mddev,
450 unsigned int chunk_sects, struct bio *bio)
451{
452 if (likely(is_power_of_2(chunk_sects))) {
453 return chunk_sects >=
454 ((bio->bi_iter.bi_sector & (chunk_sects-1))
455 + bio_sectors(bio));
456 } else{
457 sector_t sector = bio->bi_iter.bi_sector;
458 return chunk_sects >= (sector_div(sector, chunk_sects)
459 + bio_sectors(bio));
460 }
461}
462
463static void raid0_handle_discard(struct mddev *mddev, struct bio *bio)
464{
465 struct r0conf *conf = mddev->private;
466 struct strip_zone *zone;
467 sector_t start = bio->bi_iter.bi_sector;
468 sector_t end;
469 unsigned int stripe_size;
470 sector_t first_stripe_index, last_stripe_index;
471 sector_t start_disk_offset;
472 unsigned int start_disk_index;
473 sector_t end_disk_offset;
474 unsigned int end_disk_index;
475 unsigned int disk;
476
477 zone = find_zone(conf, &start);
478
479 if (bio_end_sector(bio) > zone->zone_end) {
480 struct bio *split = bio_split(bio,
481 zone->zone_end - bio->bi_iter.bi_sector, GFP_NOIO,
482 mddev->bio_set);
483 bio_chain(split, bio);
484 generic_make_request(bio);
485 bio = split;
486 end = zone->zone_end;
487 } else
488 end = bio_end_sector(bio);
489
490 if (zone != conf->strip_zone)
491 end = end - zone[-1].zone_end;
492
493 /* Now start and end is the offset in zone */
494 stripe_size = zone->nb_dev * mddev->chunk_sectors;
495
496 first_stripe_index = start;
497 sector_div(first_stripe_index, stripe_size);
498 last_stripe_index = end;
499 sector_div(last_stripe_index, stripe_size);
500
501 start_disk_index = (int)(start - first_stripe_index * stripe_size) /
502 mddev->chunk_sectors;
503 start_disk_offset = ((int)(start - first_stripe_index * stripe_size) %
504 mddev->chunk_sectors) +
505 first_stripe_index * mddev->chunk_sectors;
506 end_disk_index = (int)(end - last_stripe_index * stripe_size) /
507 mddev->chunk_sectors;
508 end_disk_offset = ((int)(end - last_stripe_index * stripe_size) %
509 mddev->chunk_sectors) +
510 last_stripe_index * mddev->chunk_sectors;
511
512 for (disk = 0; disk < zone->nb_dev; disk++) {
513 sector_t dev_start, dev_end;
514 struct bio *discard_bio = NULL;
515 struct md_rdev *rdev;
516
517 if (disk < start_disk_index)
518 dev_start = (first_stripe_index + 1) *
519 mddev->chunk_sectors;
520 else if (disk > start_disk_index)
521 dev_start = first_stripe_index * mddev->chunk_sectors;
522 else
523 dev_start = start_disk_offset;
524
525 if (disk < end_disk_index)
526 dev_end = (last_stripe_index + 1) * mddev->chunk_sectors;
527 else if (disk > end_disk_index)
528 dev_end = last_stripe_index * mddev->chunk_sectors;
529 else
530 dev_end = end_disk_offset;
531
532 if (dev_end <= dev_start)
533 continue;
534
535 rdev = conf->devlist[(zone - conf->strip_zone) *
536 conf->strip_zone[0].nb_dev + disk];
537 if (__blkdev_issue_discard(rdev->bdev,
538 dev_start + zone->dev_start + rdev->data_offset,
539 dev_end - dev_start, GFP_NOIO, 0, &discard_bio) ||
540 !discard_bio)
541 continue;
542 bio_chain(discard_bio, bio);
543 bio_clone_blkcg_association(discard_bio, bio);
544 if (mddev->gendisk)
545 trace_block_bio_remap(bdev_get_queue(rdev->bdev),
546 discard_bio, disk_devt(mddev->gendisk),
547 bio->bi_iter.bi_sector);
548 generic_make_request(discard_bio);
549 }
550 bio_endio(bio);
551}
552
553static bool raid0_make_request(struct mddev *mddev, struct bio *bio)
554{
555 struct strip_zone *zone;
556 struct md_rdev *tmp_dev;
557 sector_t bio_sector;
558 sector_t sector;
559 unsigned chunk_sects;
560 unsigned sectors;
561
562 if (unlikely(bio->bi_opf & REQ_PREFLUSH)) {
563 md_flush_request(mddev, bio);
564 return true;
565 }
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, mddev->bio_set);
586 bio_chain(split, bio);
587 generic_make_request(bio);
588 bio = split;
589 }
590
591 zone = find_zone(mddev->private, §or);
592 tmp_dev = map_sector(mddev, zone, sector, §or);
593 bio_set_dev(bio, tmp_dev->bdev);
594 bio->bi_iter.bi_sector = sector + zone->dev_start +
595 tmp_dev->data_offset;
596
597 if (mddev->gendisk)
598 trace_block_bio_remap(bio->bi_disk->queue, bio,
599 disk_devt(mddev->gendisk), bio_sector);
600 mddev_check_writesame(mddev, bio);
601 mddev_check_write_zeroes(mddev, bio);
602 generic_make_request(bio);
603 return true;
604}
605
606static void raid0_status(struct seq_file *seq, struct mddev *mddev)
607{
608 seq_printf(seq, " %dk chunks", mddev->chunk_sectors / 2);
609 return;
610}
611
612static void *raid0_takeover_raid45(struct mddev *mddev)
613{
614 struct md_rdev *rdev;
615 struct r0conf *priv_conf;
616
617 if (mddev->degraded != 1) {
618 pr_warn("md/raid0:%s: raid5 must be degraded! Degraded disks: %d\n",
619 mdname(mddev),
620 mddev->degraded);
621 return ERR_PTR(-EINVAL);
622 }
623
624 rdev_for_each(rdev, mddev) {
625 /* check slot number for a disk */
626 if (rdev->raid_disk == mddev->raid_disks-1) {
627 pr_warn("md/raid0:%s: raid5 must have missing parity disk!\n",
628 mdname(mddev));
629 return ERR_PTR(-EINVAL);
630 }
631 rdev->sectors = mddev->dev_sectors;
632 }
633
634 /* Set new parameters */
635 mddev->new_level = 0;
636 mddev->new_layout = 0;
637 mddev->new_chunk_sectors = mddev->chunk_sectors;
638 mddev->raid_disks--;
639 mddev->delta_disks = -1;
640 /* make sure it will be not marked as dirty */
641 mddev->recovery_cp = MaxSector;
642 mddev_clear_unsupported_flags(mddev, UNSUPPORTED_MDDEV_FLAGS);
643
644 create_strip_zones(mddev, &priv_conf);
645
646 return priv_conf;
647}
648
649static void *raid0_takeover_raid10(struct mddev *mddev)
650{
651 struct r0conf *priv_conf;
652
653 /* Check layout:
654 * - far_copies must be 1
655 * - near_copies must be 2
656 * - disks number must be even
657 * - all mirrors must be already degraded
658 */
659 if (mddev->layout != ((1 << 8) + 2)) {
660 pr_warn("md/raid0:%s:: Raid0 cannot takeover layout: 0x%x\n",
661 mdname(mddev),
662 mddev->layout);
663 return ERR_PTR(-EINVAL);
664 }
665 if (mddev->raid_disks & 1) {
666 pr_warn("md/raid0:%s: Raid0 cannot takeover Raid10 with odd disk number.\n",
667 mdname(mddev));
668 return ERR_PTR(-EINVAL);
669 }
670 if (mddev->degraded != (mddev->raid_disks>>1)) {
671 pr_warn("md/raid0:%s: All mirrors must be already degraded!\n",
672 mdname(mddev));
673 return ERR_PTR(-EINVAL);
674 }
675
676 /* Set new parameters */
677 mddev->new_level = 0;
678 mddev->new_layout = 0;
679 mddev->new_chunk_sectors = mddev->chunk_sectors;
680 mddev->delta_disks = - mddev->raid_disks / 2;
681 mddev->raid_disks += mddev->delta_disks;
682 mddev->degraded = 0;
683 /* make sure it will be not marked as dirty */
684 mddev->recovery_cp = MaxSector;
685 mddev_clear_unsupported_flags(mddev, UNSUPPORTED_MDDEV_FLAGS);
686
687 create_strip_zones(mddev, &priv_conf);
688 return priv_conf;
689}
690
691static void *raid0_takeover_raid1(struct mddev *mddev)
692{
693 struct r0conf *priv_conf;
694 int chunksect;
695
696 /* Check layout:
697 * - (N - 1) mirror drives must be already faulty
698 */
699 if ((mddev->raid_disks - 1) != mddev->degraded) {
700 pr_err("md/raid0:%s: (N - 1) mirrors drives must be already faulty!\n",
701 mdname(mddev));
702 return ERR_PTR(-EINVAL);
703 }
704
705 /*
706 * a raid1 doesn't have the notion of chunk size, so
707 * figure out the largest suitable size we can use.
708 */
709 chunksect = 64 * 2; /* 64K by default */
710
711 /* The array must be an exact multiple of chunksize */
712 while (chunksect && (mddev->array_sectors & (chunksect - 1)))
713 chunksect >>= 1;
714
715 if ((chunksect << 9) < PAGE_SIZE)
716 /* array size does not allow a suitable chunk size */
717 return ERR_PTR(-EINVAL);
718
719 /* Set new parameters */
720 mddev->new_level = 0;
721 mddev->new_layout = 0;
722 mddev->new_chunk_sectors = chunksect;
723 mddev->chunk_sectors = chunksect;
724 mddev->delta_disks = 1 - mddev->raid_disks;
725 mddev->raid_disks = 1;
726 /* make sure it will be not marked as dirty */
727 mddev->recovery_cp = MaxSector;
728 mddev_clear_unsupported_flags(mddev, UNSUPPORTED_MDDEV_FLAGS);
729
730 create_strip_zones(mddev, &priv_conf);
731 return priv_conf;
732}
733
734static void *raid0_takeover(struct mddev *mddev)
735{
736 /* raid0 can take over:
737 * raid4 - if all data disks are active.
738 * raid5 - providing it is Raid4 layout and one disk is faulty
739 * raid10 - assuming we have all necessary active disks
740 * raid1 - with (N -1) mirror drives faulty
741 */
742
743 if (mddev->bitmap) {
744 pr_warn("md/raid0: %s: cannot takeover array with bitmap\n",
745 mdname(mddev));
746 return ERR_PTR(-EBUSY);
747 }
748 if (mddev->level == 4)
749 return raid0_takeover_raid45(mddev);
750
751 if (mddev->level == 5) {
752 if (mddev->layout == ALGORITHM_PARITY_N)
753 return raid0_takeover_raid45(mddev);
754
755 pr_warn("md/raid0:%s: Raid can only takeover Raid5 with layout: %d\n",
756 mdname(mddev), ALGORITHM_PARITY_N);
757 }
758
759 if (mddev->level == 10)
760 return raid0_takeover_raid10(mddev);
761
762 if (mddev->level == 1)
763 return raid0_takeover_raid1(mddev);
764
765 pr_warn("Takeover from raid%i to raid0 not supported\n",
766 mddev->level);
767
768 return ERR_PTR(-EINVAL);
769}
770
771static void raid0_quiesce(struct mddev *mddev, int quiesce)
772{
773}
774
775static struct md_personality raid0_personality=
776{
777 .name = "raid0",
778 .level = 0,
779 .owner = THIS_MODULE,
780 .make_request = raid0_make_request,
781 .run = raid0_run,
782 .free = raid0_free,
783 .status = raid0_status,
784 .size = raid0_size,
785 .takeover = raid0_takeover,
786 .quiesce = raid0_quiesce,
787 .congested = raid0_congested,
788};
789
790static int __init raid0_init (void)
791{
792 return register_md_personality (&raid0_personality);
793}
794
795static void raid0_exit (void)
796{
797 unregister_md_personality (&raid0_personality);
798}
799
800module_init(raid0_init);
801module_exit(raid0_exit);
802MODULE_LICENSE("GPL");
803MODULE_DESCRIPTION("RAID0 (striping) personality for MD");
804MODULE_ALIAS("md-personality-2"); /* RAID0 */
805MODULE_ALIAS("md-raid0");
806MODULE_ALIAS("md-level-0");