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/slab.h>
24#include "md.h"
25#include "raid0.h"
26#include "raid5.h"
27
28static int raid0_congested(void *data, int bits)
29{
30 mddev_t *mddev = data;
31 raid0_conf_t *conf = mddev->private;
32 mdk_rdev_t **devlist = conf->devlist;
33 int raid_disks = conf->strip_zone[0].nb_dev;
34 int i, ret = 0;
35
36 if (mddev_congested(mddev, bits))
37 return 1;
38
39 for (i = 0; i < raid_disks && !ret ; i++) {
40 struct request_queue *q = bdev_get_queue(devlist[i]->bdev);
41
42 ret |= bdi_congested(&q->backing_dev_info, bits);
43 }
44 return ret;
45}
46
47/*
48 * inform the user of the raid configuration
49*/
50static void dump_zones(mddev_t *mddev)
51{
52 int j, k, h;
53 sector_t zone_size = 0;
54 sector_t zone_start = 0;
55 char b[BDEVNAME_SIZE];
56 raid0_conf_t *conf = mddev->private;
57 int raid_disks = conf->strip_zone[0].nb_dev;
58 printk(KERN_INFO "******* %s configuration *********\n",
59 mdname(mddev));
60 h = 0;
61 for (j = 0; j < conf->nr_strip_zones; j++) {
62 printk(KERN_INFO "zone%d=[", j);
63 for (k = 0; k < conf->strip_zone[j].nb_dev; k++)
64 printk(KERN_CONT "%s/",
65 bdevname(conf->devlist[j*raid_disks
66 + k]->bdev, b));
67 printk(KERN_CONT "]\n");
68
69 zone_size = conf->strip_zone[j].zone_end - zone_start;
70 printk(KERN_INFO " zone offset=%llukb "
71 "device offset=%llukb size=%llukb\n",
72 (unsigned long long)zone_start>>1,
73 (unsigned long long)conf->strip_zone[j].dev_start>>1,
74 (unsigned long long)zone_size>>1);
75 zone_start = conf->strip_zone[j].zone_end;
76 }
77 printk(KERN_INFO "**********************************\n\n");
78}
79
80static int create_strip_zones(mddev_t *mddev, raid0_conf_t **private_conf)
81{
82 int i, c, err;
83 sector_t curr_zone_end, sectors;
84 mdk_rdev_t *smallest, *rdev1, *rdev2, *rdev, **dev;
85 struct strip_zone *zone;
86 int cnt;
87 char b[BDEVNAME_SIZE];
88 raid0_conf_t *conf = kzalloc(sizeof(*conf), GFP_KERNEL);
89
90 if (!conf)
91 return -ENOMEM;
92 list_for_each_entry(rdev1, &mddev->disks, same_set) {
93 printk(KERN_INFO "md/raid0:%s: looking at %s\n",
94 mdname(mddev),
95 bdevname(rdev1->bdev, b));
96 c = 0;
97
98 /* round size to chunk_size */
99 sectors = rdev1->sectors;
100 sector_div(sectors, mddev->chunk_sectors);
101 rdev1->sectors = sectors * mddev->chunk_sectors;
102
103 list_for_each_entry(rdev2, &mddev->disks, same_set) {
104 printk(KERN_INFO "md/raid0:%s: comparing %s(%llu)",
105 mdname(mddev),
106 bdevname(rdev1->bdev,b),
107 (unsigned long long)rdev1->sectors);
108 printk(KERN_CONT " with %s(%llu)\n",
109 bdevname(rdev2->bdev,b),
110 (unsigned long long)rdev2->sectors);
111 if (rdev2 == rdev1) {
112 printk(KERN_INFO "md/raid0:%s: END\n",
113 mdname(mddev));
114 break;
115 }
116 if (rdev2->sectors == rdev1->sectors) {
117 /*
118 * Not unique, don't count it as a new
119 * group
120 */
121 printk(KERN_INFO "md/raid0:%s: EQUAL\n",
122 mdname(mddev));
123 c = 1;
124 break;
125 }
126 printk(KERN_INFO "md/raid0:%s: NOT EQUAL\n",
127 mdname(mddev));
128 }
129 if (!c) {
130 printk(KERN_INFO "md/raid0:%s: ==> UNIQUE\n",
131 mdname(mddev));
132 conf->nr_strip_zones++;
133 printk(KERN_INFO "md/raid0:%s: %d zones\n",
134 mdname(mddev), conf->nr_strip_zones);
135 }
136 }
137 printk(KERN_INFO "md/raid0:%s: FINAL %d zones\n",
138 mdname(mddev), conf->nr_strip_zones);
139 err = -ENOMEM;
140 conf->strip_zone = kzalloc(sizeof(struct strip_zone)*
141 conf->nr_strip_zones, GFP_KERNEL);
142 if (!conf->strip_zone)
143 goto abort;
144 conf->devlist = kzalloc(sizeof(mdk_rdev_t*)*
145 conf->nr_strip_zones*mddev->raid_disks,
146 GFP_KERNEL);
147 if (!conf->devlist)
148 goto abort;
149
150 /* The first zone must contain all devices, so here we check that
151 * there is a proper alignment of slots to devices and find them all
152 */
153 zone = &conf->strip_zone[0];
154 cnt = 0;
155 smallest = NULL;
156 dev = conf->devlist;
157 err = -EINVAL;
158 list_for_each_entry(rdev1, &mddev->disks, same_set) {
159 int j = rdev1->raid_disk;
160
161 if (mddev->level == 10) {
162 /* taking over a raid10-n2 array */
163 j /= 2;
164 rdev1->new_raid_disk = j;
165 }
166
167 if (mddev->level == 1) {
168 /* taiking over a raid1 array-
169 * we have only one active disk
170 */
171 j = 0;
172 rdev1->new_raid_disk = j;
173 }
174
175 if (j < 0 || j >= mddev->raid_disks) {
176 printk(KERN_ERR "md/raid0:%s: bad disk number %d - "
177 "aborting!\n", mdname(mddev), j);
178 goto abort;
179 }
180 if (dev[j]) {
181 printk(KERN_ERR "md/raid0:%s: multiple devices for %d - "
182 "aborting!\n", mdname(mddev), j);
183 goto abort;
184 }
185 dev[j] = rdev1;
186
187 disk_stack_limits(mddev->gendisk, rdev1->bdev,
188 rdev1->data_offset << 9);
189 /* as we don't honour merge_bvec_fn, we must never risk
190 * violating it, so limit ->max_segments to 1, lying within
191 * a single page.
192 */
193
194 if (rdev1->bdev->bd_disk->queue->merge_bvec_fn) {
195 blk_queue_max_segments(mddev->queue, 1);
196 blk_queue_segment_boundary(mddev->queue,
197 PAGE_CACHE_SIZE - 1);
198 }
199 if (!smallest || (rdev1->sectors < smallest->sectors))
200 smallest = rdev1;
201 cnt++;
202 }
203 if (cnt != mddev->raid_disks) {
204 printk(KERN_ERR "md/raid0:%s: too few disks (%d of %d) - "
205 "aborting!\n", mdname(mddev), cnt, mddev->raid_disks);
206 goto abort;
207 }
208 zone->nb_dev = cnt;
209 zone->zone_end = smallest->sectors * cnt;
210
211 curr_zone_end = zone->zone_end;
212
213 /* now do the other zones */
214 for (i = 1; i < conf->nr_strip_zones; i++)
215 {
216 int j;
217
218 zone = conf->strip_zone + i;
219 dev = conf->devlist + i * mddev->raid_disks;
220
221 printk(KERN_INFO "md/raid0:%s: zone %d\n",
222 mdname(mddev), i);
223 zone->dev_start = smallest->sectors;
224 smallest = NULL;
225 c = 0;
226
227 for (j=0; j<cnt; j++) {
228 rdev = conf->devlist[j];
229 printk(KERN_INFO "md/raid0:%s: checking %s ...",
230 mdname(mddev),
231 bdevname(rdev->bdev, b));
232 if (rdev->sectors <= zone->dev_start) {
233 printk(KERN_CONT " nope.\n");
234 continue;
235 }
236 printk(KERN_CONT " contained as device %d\n", c);
237 dev[c] = rdev;
238 c++;
239 if (!smallest || rdev->sectors < smallest->sectors) {
240 smallest = rdev;
241 printk(KERN_INFO "md/raid0:%s: (%llu) is smallest!.\n",
242 mdname(mddev),
243 (unsigned long long)rdev->sectors);
244 }
245 }
246
247 zone->nb_dev = c;
248 sectors = (smallest->sectors - zone->dev_start) * c;
249 printk(KERN_INFO "md/raid0:%s: zone->nb_dev: %d, sectors: %llu\n",
250 mdname(mddev),
251 zone->nb_dev, (unsigned long long)sectors);
252
253 curr_zone_end += sectors;
254 zone->zone_end = curr_zone_end;
255
256 printk(KERN_INFO "md/raid0:%s: current zone start: %llu\n",
257 mdname(mddev),
258 (unsigned long long)smallest->sectors);
259 }
260 mddev->queue->backing_dev_info.congested_fn = raid0_congested;
261 mddev->queue->backing_dev_info.congested_data = mddev;
262
263 /*
264 * now since we have the hard sector sizes, we can make sure
265 * chunk size is a multiple of that sector size
266 */
267 if ((mddev->chunk_sectors << 9) % queue_logical_block_size(mddev->queue)) {
268 printk(KERN_ERR "md/raid0:%s: chunk_size of %d not valid\n",
269 mdname(mddev),
270 mddev->chunk_sectors << 9);
271 goto abort;
272 }
273
274 blk_queue_io_min(mddev->queue, mddev->chunk_sectors << 9);
275 blk_queue_io_opt(mddev->queue,
276 (mddev->chunk_sectors << 9) * mddev->raid_disks);
277
278 printk(KERN_INFO "md/raid0:%s: done.\n", mdname(mddev));
279 *private_conf = conf;
280
281 return 0;
282abort:
283 kfree(conf->strip_zone);
284 kfree(conf->devlist);
285 kfree(conf);
286 *private_conf = NULL;
287 return err;
288}
289
290/**
291 * raid0_mergeable_bvec -- tell bio layer if a two requests can be merged
292 * @q: request queue
293 * @bvm: properties of new bio
294 * @biovec: the request that could be merged to it.
295 *
296 * Return amount of bytes we can accept at this offset
297 */
298static int raid0_mergeable_bvec(struct request_queue *q,
299 struct bvec_merge_data *bvm,
300 struct bio_vec *biovec)
301{
302 mddev_t *mddev = q->queuedata;
303 sector_t sector = bvm->bi_sector + get_start_sect(bvm->bi_bdev);
304 int max;
305 unsigned int chunk_sectors = mddev->chunk_sectors;
306 unsigned int bio_sectors = bvm->bi_size >> 9;
307
308 if (is_power_of_2(chunk_sectors))
309 max = (chunk_sectors - ((sector & (chunk_sectors-1))
310 + bio_sectors)) << 9;
311 else
312 max = (chunk_sectors - (sector_div(sector, chunk_sectors)
313 + bio_sectors)) << 9;
314 if (max < 0) max = 0; /* bio_add cannot handle a negative return */
315 if (max <= biovec->bv_len && bio_sectors == 0)
316 return biovec->bv_len;
317 else
318 return max;
319}
320
321static sector_t raid0_size(mddev_t *mddev, sector_t sectors, int raid_disks)
322{
323 sector_t array_sectors = 0;
324 mdk_rdev_t *rdev;
325
326 WARN_ONCE(sectors || raid_disks,
327 "%s does not support generic reshape\n", __func__);
328
329 list_for_each_entry(rdev, &mddev->disks, same_set)
330 array_sectors += rdev->sectors;
331
332 return array_sectors;
333}
334
335static int raid0_run(mddev_t *mddev)
336{
337 raid0_conf_t *conf;
338 int ret;
339
340 if (mddev->chunk_sectors == 0) {
341 printk(KERN_ERR "md/raid0:%s: chunk size must be set.\n",
342 mdname(mddev));
343 return -EINVAL;
344 }
345 if (md_check_no_bitmap(mddev))
346 return -EINVAL;
347 blk_queue_max_hw_sectors(mddev->queue, mddev->chunk_sectors);
348
349 /* if private is not null, we are here after takeover */
350 if (mddev->private == NULL) {
351 ret = create_strip_zones(mddev, &conf);
352 if (ret < 0)
353 return ret;
354 mddev->private = conf;
355 }
356 conf = mddev->private;
357
358 /* calculate array device size */
359 md_set_array_sectors(mddev, raid0_size(mddev, 0, 0));
360
361 printk(KERN_INFO "md/raid0:%s: md_size is %llu sectors.\n",
362 mdname(mddev),
363 (unsigned long long)mddev->array_sectors);
364 /* calculate the max read-ahead size.
365 * For read-ahead of large files to be effective, we need to
366 * readahead at least twice a whole stripe. i.e. number of devices
367 * multiplied by chunk size times 2.
368 * If an individual device has an ra_pages greater than the
369 * chunk size, then we will not drive that device as hard as it
370 * wants. We consider this a configuration error: a larger
371 * chunksize should be used in that case.
372 */
373 {
374 int stripe = mddev->raid_disks *
375 (mddev->chunk_sectors << 9) / PAGE_SIZE;
376 if (mddev->queue->backing_dev_info.ra_pages < 2* stripe)
377 mddev->queue->backing_dev_info.ra_pages = 2* stripe;
378 }
379
380 blk_queue_merge_bvec(mddev->queue, raid0_mergeable_bvec);
381 dump_zones(mddev);
382 return md_integrity_register(mddev);
383}
384
385static int raid0_stop(mddev_t *mddev)
386{
387 raid0_conf_t *conf = mddev->private;
388
389 blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/
390 kfree(conf->strip_zone);
391 kfree(conf->devlist);
392 kfree(conf);
393 mddev->private = NULL;
394 return 0;
395}
396
397/* Find the zone which holds a particular offset
398 * Update *sectorp to be an offset in that zone
399 */
400static struct strip_zone *find_zone(struct raid0_private_data *conf,
401 sector_t *sectorp)
402{
403 int i;
404 struct strip_zone *z = conf->strip_zone;
405 sector_t sector = *sectorp;
406
407 for (i = 0; i < conf->nr_strip_zones; i++)
408 if (sector < z[i].zone_end) {
409 if (i)
410 *sectorp = sector - z[i-1].zone_end;
411 return z + i;
412 }
413 BUG();
414}
415
416/*
417 * remaps the bio to the target device. we separate two flows.
418 * power 2 flow and a general flow for the sake of perfromance
419*/
420static mdk_rdev_t *map_sector(mddev_t *mddev, struct strip_zone *zone,
421 sector_t sector, sector_t *sector_offset)
422{
423 unsigned int sect_in_chunk;
424 sector_t chunk;
425 raid0_conf_t *conf = mddev->private;
426 int raid_disks = conf->strip_zone[0].nb_dev;
427 unsigned int chunk_sects = mddev->chunk_sectors;
428
429 if (is_power_of_2(chunk_sects)) {
430 int chunksect_bits = ffz(~chunk_sects);
431 /* find the sector offset inside the chunk */
432 sect_in_chunk = sector & (chunk_sects - 1);
433 sector >>= chunksect_bits;
434 /* chunk in zone */
435 chunk = *sector_offset;
436 /* quotient is the chunk in real device*/
437 sector_div(chunk, zone->nb_dev << chunksect_bits);
438 } else{
439 sect_in_chunk = sector_div(sector, chunk_sects);
440 chunk = *sector_offset;
441 sector_div(chunk, chunk_sects * zone->nb_dev);
442 }
443 /*
444 * position the bio over the real device
445 * real sector = chunk in device + starting of zone
446 * + the position in the chunk
447 */
448 *sector_offset = (chunk * chunk_sects) + sect_in_chunk;
449 return conf->devlist[(zone - conf->strip_zone)*raid_disks
450 + sector_div(sector, zone->nb_dev)];
451}
452
453/*
454 * Is io distribute over 1 or more chunks ?
455*/
456static inline int is_io_in_chunk_boundary(mddev_t *mddev,
457 unsigned int chunk_sects, struct bio *bio)
458{
459 if (likely(is_power_of_2(chunk_sects))) {
460 return chunk_sects >= ((bio->bi_sector & (chunk_sects-1))
461 + (bio->bi_size >> 9));
462 } else{
463 sector_t sector = bio->bi_sector;
464 return chunk_sects >= (sector_div(sector, chunk_sects)
465 + (bio->bi_size >> 9));
466 }
467}
468
469static int raid0_make_request(mddev_t *mddev, struct bio *bio)
470{
471 unsigned int chunk_sects;
472 sector_t sector_offset;
473 struct strip_zone *zone;
474 mdk_rdev_t *tmp_dev;
475
476 if (unlikely(bio->bi_rw & REQ_FLUSH)) {
477 md_flush_request(mddev, bio);
478 return 0;
479 }
480
481 chunk_sects = mddev->chunk_sectors;
482 if (unlikely(!is_io_in_chunk_boundary(mddev, chunk_sects, bio))) {
483 sector_t sector = bio->bi_sector;
484 struct bio_pair *bp;
485 /* Sanity check -- queue functions should prevent this happening */
486 if (bio->bi_vcnt != 1 ||
487 bio->bi_idx != 0)
488 goto bad_map;
489 /* This is a one page bio that upper layers
490 * refuse to split for us, so we need to split it.
491 */
492 if (likely(is_power_of_2(chunk_sects)))
493 bp = bio_split(bio, chunk_sects - (sector &
494 (chunk_sects-1)));
495 else
496 bp = bio_split(bio, chunk_sects -
497 sector_div(sector, chunk_sects));
498 if (raid0_make_request(mddev, &bp->bio1))
499 generic_make_request(&bp->bio1);
500 if (raid0_make_request(mddev, &bp->bio2))
501 generic_make_request(&bp->bio2);
502
503 bio_pair_release(bp);
504 return 0;
505 }
506
507 sector_offset = bio->bi_sector;
508 zone = find_zone(mddev->private, §or_offset);
509 tmp_dev = map_sector(mddev, zone, bio->bi_sector,
510 §or_offset);
511 bio->bi_bdev = tmp_dev->bdev;
512 bio->bi_sector = sector_offset + zone->dev_start +
513 tmp_dev->data_offset;
514 /*
515 * Let the main block layer submit the IO and resolve recursion:
516 */
517 return 1;
518
519bad_map:
520 printk("md/raid0:%s: make_request bug: can't convert block across chunks"
521 " or bigger than %dk %llu %d\n",
522 mdname(mddev), chunk_sects / 2,
523 (unsigned long long)bio->bi_sector, bio->bi_size >> 10);
524
525 bio_io_error(bio);
526 return 0;
527}
528
529static void raid0_status(struct seq_file *seq, mddev_t *mddev)
530{
531#undef MD_DEBUG
532#ifdef MD_DEBUG
533 int j, k, h;
534 char b[BDEVNAME_SIZE];
535 raid0_conf_t *conf = mddev->private;
536 int raid_disks = conf->strip_zone[0].nb_dev;
537
538 sector_t zone_size;
539 sector_t zone_start = 0;
540 h = 0;
541
542 for (j = 0; j < conf->nr_strip_zones; j++) {
543 seq_printf(seq, " z%d", j);
544 seq_printf(seq, "=[");
545 for (k = 0; k < conf->strip_zone[j].nb_dev; k++)
546 seq_printf(seq, "%s/", bdevname(
547 conf->devlist[j*raid_disks + k]
548 ->bdev, b));
549
550 zone_size = conf->strip_zone[j].zone_end - zone_start;
551 seq_printf(seq, "] ze=%lld ds=%lld s=%lld\n",
552 (unsigned long long)zone_start>>1,
553 (unsigned long long)conf->strip_zone[j].dev_start>>1,
554 (unsigned long long)zone_size>>1);
555 zone_start = conf->strip_zone[j].zone_end;
556 }
557#endif
558 seq_printf(seq, " %dk chunks", mddev->chunk_sectors / 2);
559 return;
560}
561
562static void *raid0_takeover_raid45(mddev_t *mddev)
563{
564 mdk_rdev_t *rdev;
565 raid0_conf_t *priv_conf;
566
567 if (mddev->degraded != 1) {
568 printk(KERN_ERR "md/raid0:%s: raid5 must be degraded! Degraded disks: %d\n",
569 mdname(mddev),
570 mddev->degraded);
571 return ERR_PTR(-EINVAL);
572 }
573
574 list_for_each_entry(rdev, &mddev->disks, same_set) {
575 /* check slot number for a disk */
576 if (rdev->raid_disk == mddev->raid_disks-1) {
577 printk(KERN_ERR "md/raid0:%s: raid5 must have missing parity disk!\n",
578 mdname(mddev));
579 return ERR_PTR(-EINVAL);
580 }
581 }
582
583 /* Set new parameters */
584 mddev->new_level = 0;
585 mddev->new_layout = 0;
586 mddev->new_chunk_sectors = mddev->chunk_sectors;
587 mddev->raid_disks--;
588 mddev->delta_disks = -1;
589 /* make sure it will be not marked as dirty */
590 mddev->recovery_cp = MaxSector;
591
592 create_strip_zones(mddev, &priv_conf);
593 return priv_conf;
594}
595
596static void *raid0_takeover_raid10(mddev_t *mddev)
597{
598 raid0_conf_t *priv_conf;
599
600 /* Check layout:
601 * - far_copies must be 1
602 * - near_copies must be 2
603 * - disks number must be even
604 * - all mirrors must be already degraded
605 */
606 if (mddev->layout != ((1 << 8) + 2)) {
607 printk(KERN_ERR "md/raid0:%s:: Raid0 cannot takover layout: 0x%x\n",
608 mdname(mddev),
609 mddev->layout);
610 return ERR_PTR(-EINVAL);
611 }
612 if (mddev->raid_disks & 1) {
613 printk(KERN_ERR "md/raid0:%s: Raid0 cannot takover Raid10 with odd disk number.\n",
614 mdname(mddev));
615 return ERR_PTR(-EINVAL);
616 }
617 if (mddev->degraded != (mddev->raid_disks>>1)) {
618 printk(KERN_ERR "md/raid0:%s: All mirrors must be already degraded!\n",
619 mdname(mddev));
620 return ERR_PTR(-EINVAL);
621 }
622
623 /* Set new parameters */
624 mddev->new_level = 0;
625 mddev->new_layout = 0;
626 mddev->new_chunk_sectors = mddev->chunk_sectors;
627 mddev->delta_disks = - mddev->raid_disks / 2;
628 mddev->raid_disks += mddev->delta_disks;
629 mddev->degraded = 0;
630 /* make sure it will be not marked as dirty */
631 mddev->recovery_cp = MaxSector;
632
633 create_strip_zones(mddev, &priv_conf);
634 return priv_conf;
635}
636
637static void *raid0_takeover_raid1(mddev_t *mddev)
638{
639 raid0_conf_t *priv_conf;
640
641 /* Check layout:
642 * - (N - 1) mirror drives must be already faulty
643 */
644 if ((mddev->raid_disks - 1) != mddev->degraded) {
645 printk(KERN_ERR "md/raid0:%s: (N - 1) mirrors drives must be already faulty!\n",
646 mdname(mddev));
647 return ERR_PTR(-EINVAL);
648 }
649
650 /* Set new parameters */
651 mddev->new_level = 0;
652 mddev->new_layout = 0;
653 mddev->new_chunk_sectors = 128; /* by default set chunk size to 64k */
654 mddev->delta_disks = 1 - mddev->raid_disks;
655 mddev->raid_disks = 1;
656 /* make sure it will be not marked as dirty */
657 mddev->recovery_cp = MaxSector;
658
659 create_strip_zones(mddev, &priv_conf);
660 return priv_conf;
661}
662
663static void *raid0_takeover(mddev_t *mddev)
664{
665 /* raid0 can take over:
666 * raid4 - if all data disks are active.
667 * raid5 - providing it is Raid4 layout and one disk is faulty
668 * raid10 - assuming we have all necessary active disks
669 * raid1 - with (N -1) mirror drives faulty
670 */
671 if (mddev->level == 4)
672 return raid0_takeover_raid45(mddev);
673
674 if (mddev->level == 5) {
675 if (mddev->layout == ALGORITHM_PARITY_N)
676 return raid0_takeover_raid45(mddev);
677
678 printk(KERN_ERR "md/raid0:%s: Raid can only takeover Raid5 with layout: %d\n",
679 mdname(mddev), ALGORITHM_PARITY_N);
680 }
681
682 if (mddev->level == 10)
683 return raid0_takeover_raid10(mddev);
684
685 if (mddev->level == 1)
686 return raid0_takeover_raid1(mddev);
687
688 printk(KERN_ERR "Takeover from raid%i to raid0 not supported\n",
689 mddev->level);
690
691 return ERR_PTR(-EINVAL);
692}
693
694static void raid0_quiesce(mddev_t *mddev, int state)
695{
696}
697
698static struct mdk_personality raid0_personality=
699{
700 .name = "raid0",
701 .level = 0,
702 .owner = THIS_MODULE,
703 .make_request = raid0_make_request,
704 .run = raid0_run,
705 .stop = raid0_stop,
706 .status = raid0_status,
707 .size = raid0_size,
708 .takeover = raid0_takeover,
709 .quiesce = raid0_quiesce,
710};
711
712static int __init raid0_init (void)
713{
714 return register_md_personality (&raid0_personality);
715}
716
717static void raid0_exit (void)
718{
719 unregister_md_personality (&raid0_personality);
720}
721
722module_init(raid0_init);
723module_exit(raid0_exit);
724MODULE_LICENSE("GPL");
725MODULE_DESCRIPTION("RAID0 (striping) personality for MD");
726MODULE_ALIAS("md-personality-2"); /* RAID0 */
727MODULE_ALIAS("md-raid0");
728MODULE_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
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 bool discard_supported = false;
92
93 if (!conf)
94 return -ENOMEM;
95 rdev_for_each(rdev1, mddev) {
96 pr_debug("md/raid0:%s: looking at %s\n",
97 mdname(mddev),
98 bdevname(rdev1->bdev, b));
99 c = 0;
100
101 /* round size to chunk_size */
102 sectors = rdev1->sectors;
103 sector_div(sectors, mddev->chunk_sectors);
104 rdev1->sectors = sectors * mddev->chunk_sectors;
105
106 rdev_for_each(rdev2, mddev) {
107 pr_debug("md/raid0:%s: comparing %s(%llu)"
108 " with %s(%llu)\n",
109 mdname(mddev),
110 bdevname(rdev1->bdev,b),
111 (unsigned long long)rdev1->sectors,
112 bdevname(rdev2->bdev,b2),
113 (unsigned long long)rdev2->sectors);
114 if (rdev2 == rdev1) {
115 pr_debug("md/raid0:%s: END\n",
116 mdname(mddev));
117 break;
118 }
119 if (rdev2->sectors == rdev1->sectors) {
120 /*
121 * Not unique, don't count it as a new
122 * group
123 */
124 pr_debug("md/raid0:%s: EQUAL\n",
125 mdname(mddev));
126 c = 1;
127 break;
128 }
129 pr_debug("md/raid0:%s: NOT EQUAL\n",
130 mdname(mddev));
131 }
132 if (!c) {
133 pr_debug("md/raid0:%s: ==> UNIQUE\n",
134 mdname(mddev));
135 conf->nr_strip_zones++;
136 pr_debug("md/raid0:%s: %d zones\n",
137 mdname(mddev), conf->nr_strip_zones);
138 }
139 }
140 pr_debug("md/raid0:%s: FINAL %d zones\n",
141 mdname(mddev), conf->nr_strip_zones);
142 err = -ENOMEM;
143 conf->strip_zone = kzalloc(sizeof(struct strip_zone)*
144 conf->nr_strip_zones, GFP_KERNEL);
145 if (!conf->strip_zone)
146 goto abort;
147 conf->devlist = kzalloc(sizeof(struct md_rdev*)*
148 conf->nr_strip_zones*mddev->raid_disks,
149 GFP_KERNEL);
150 if (!conf->devlist)
151 goto abort;
152
153 /* The first zone must contain all devices, so here we check that
154 * there is a proper alignment of slots to devices and find them all
155 */
156 zone = &conf->strip_zone[0];
157 cnt = 0;
158 smallest = NULL;
159 dev = conf->devlist;
160 err = -EINVAL;
161 rdev_for_each(rdev1, mddev) {
162 int j = rdev1->raid_disk;
163
164 if (mddev->level == 10) {
165 /* taking over a raid10-n2 array */
166 j /= 2;
167 rdev1->new_raid_disk = j;
168 }
169
170 if (mddev->level == 1) {
171 /* taiking over a raid1 array-
172 * we have only one active disk
173 */
174 j = 0;
175 rdev1->new_raid_disk = j;
176 }
177
178 if (j < 0) {
179 printk(KERN_ERR
180 "md/raid0:%s: remove inactive devices before converting to RAID0\n",
181 mdname(mddev));
182 goto abort;
183 }
184 if (j >= mddev->raid_disks) {
185 printk(KERN_ERR "md/raid0:%s: bad disk number %d - "
186 "aborting!\n", mdname(mddev), j);
187 goto abort;
188 }
189 if (dev[j]) {
190 printk(KERN_ERR "md/raid0:%s: multiple devices for %d - "
191 "aborting!\n", mdname(mddev), j);
192 goto abort;
193 }
194 dev[j] = rdev1;
195
196 disk_stack_limits(mddev->gendisk, rdev1->bdev,
197 rdev1->data_offset << 9);
198
199 if (rdev1->bdev->bd_disk->queue->merge_bvec_fn)
200 conf->has_merge_bvec = 1;
201
202 if (!smallest || (rdev1->sectors < smallest->sectors))
203 smallest = rdev1;
204 cnt++;
205
206 if (blk_queue_discard(bdev_get_queue(rdev1->bdev)))
207 discard_supported = true;
208 }
209 if (cnt != mddev->raid_disks) {
210 printk(KERN_ERR "md/raid0:%s: too few disks (%d of %d) - "
211 "aborting!\n", mdname(mddev), cnt, mddev->raid_disks);
212 goto abort;
213 }
214 zone->nb_dev = cnt;
215 zone->zone_end = smallest->sectors * cnt;
216
217 curr_zone_end = zone->zone_end;
218
219 /* now do the other zones */
220 for (i = 1; i < conf->nr_strip_zones; i++)
221 {
222 int j;
223
224 zone = conf->strip_zone + i;
225 dev = conf->devlist + i * mddev->raid_disks;
226
227 pr_debug("md/raid0:%s: zone %d\n", mdname(mddev), i);
228 zone->dev_start = smallest->sectors;
229 smallest = NULL;
230 c = 0;
231
232 for (j=0; j<cnt; j++) {
233 rdev = conf->devlist[j];
234 if (rdev->sectors <= zone->dev_start) {
235 pr_debug("md/raid0:%s: checking %s ... nope\n",
236 mdname(mddev),
237 bdevname(rdev->bdev, b));
238 continue;
239 }
240 pr_debug("md/raid0:%s: checking %s ..."
241 " contained as device %d\n",
242 mdname(mddev),
243 bdevname(rdev->bdev, b), c);
244 dev[c] = rdev;
245 c++;
246 if (!smallest || rdev->sectors < smallest->sectors) {
247 smallest = rdev;
248 pr_debug("md/raid0:%s: (%llu) is smallest!.\n",
249 mdname(mddev),
250 (unsigned long long)rdev->sectors);
251 }
252 }
253
254 zone->nb_dev = c;
255 sectors = (smallest->sectors - zone->dev_start) * c;
256 pr_debug("md/raid0:%s: zone->nb_dev: %d, sectors: %llu\n",
257 mdname(mddev),
258 zone->nb_dev, (unsigned long long)sectors);
259
260 curr_zone_end += sectors;
261 zone->zone_end = curr_zone_end;
262
263 pr_debug("md/raid0:%s: current zone start: %llu\n",
264 mdname(mddev),
265 (unsigned long long)smallest->sectors);
266 }
267 mddev->queue->backing_dev_info.congested_fn = raid0_congested;
268 mddev->queue->backing_dev_info.congested_data = mddev;
269
270 /*
271 * now since we have the hard sector sizes, we can make sure
272 * chunk size is a multiple of that sector size
273 */
274 if ((mddev->chunk_sectors << 9) % queue_logical_block_size(mddev->queue)) {
275 printk(KERN_ERR "md/raid0:%s: chunk_size of %d not valid\n",
276 mdname(mddev),
277 mddev->chunk_sectors << 9);
278 goto abort;
279 }
280
281 blk_queue_io_min(mddev->queue, mddev->chunk_sectors << 9);
282 blk_queue_io_opt(mddev->queue,
283 (mddev->chunk_sectors << 9) * mddev->raid_disks);
284
285 if (!discard_supported)
286 queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, mddev->queue);
287 else
288 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, mddev->queue);
289
290 pr_debug("md/raid0:%s: done.\n", mdname(mddev));
291 *private_conf = conf;
292
293 return 0;
294abort:
295 kfree(conf->strip_zone);
296 kfree(conf->devlist);
297 kfree(conf);
298 *private_conf = ERR_PTR(err);
299 return err;
300}
301
302/* Find the zone which holds a particular offset
303 * Update *sectorp to be an offset in that zone
304 */
305static struct strip_zone *find_zone(struct r0conf *conf,
306 sector_t *sectorp)
307{
308 int i;
309 struct strip_zone *z = conf->strip_zone;
310 sector_t sector = *sectorp;
311
312 for (i = 0; i < conf->nr_strip_zones; i++)
313 if (sector < z[i].zone_end) {
314 if (i)
315 *sectorp = sector - z[i-1].zone_end;
316 return z + i;
317 }
318 BUG();
319}
320
321/*
322 * remaps the bio to the target device. we separate two flows.
323 * power 2 flow and a general flow for the sake of perfromance
324*/
325static struct md_rdev *map_sector(struct mddev *mddev, struct strip_zone *zone,
326 sector_t sector, sector_t *sector_offset)
327{
328 unsigned int sect_in_chunk;
329 sector_t chunk;
330 struct r0conf *conf = mddev->private;
331 int raid_disks = conf->strip_zone[0].nb_dev;
332 unsigned int chunk_sects = mddev->chunk_sectors;
333
334 if (is_power_of_2(chunk_sects)) {
335 int chunksect_bits = ffz(~chunk_sects);
336 /* find the sector offset inside the chunk */
337 sect_in_chunk = sector & (chunk_sects - 1);
338 sector >>= chunksect_bits;
339 /* chunk in zone */
340 chunk = *sector_offset;
341 /* quotient is the chunk in real device*/
342 sector_div(chunk, zone->nb_dev << chunksect_bits);
343 } else{
344 sect_in_chunk = sector_div(sector, chunk_sects);
345 chunk = *sector_offset;
346 sector_div(chunk, chunk_sects * zone->nb_dev);
347 }
348 /*
349 * position the bio over the real device
350 * real sector = chunk in device + starting of zone
351 * + the position in the chunk
352 */
353 *sector_offset = (chunk * chunk_sects) + sect_in_chunk;
354 return conf->devlist[(zone - conf->strip_zone)*raid_disks
355 + sector_div(sector, zone->nb_dev)];
356}
357
358/**
359 * raid0_mergeable_bvec -- tell bio layer if two requests can be merged
360 * @q: request queue
361 * @bvm: properties of new bio
362 * @biovec: the request that could be merged to it.
363 *
364 * Return amount of bytes we can accept at this offset
365 */
366static int raid0_mergeable_bvec(struct request_queue *q,
367 struct bvec_merge_data *bvm,
368 struct bio_vec *biovec)
369{
370 struct mddev *mddev = q->queuedata;
371 struct r0conf *conf = mddev->private;
372 sector_t sector = bvm->bi_sector + get_start_sect(bvm->bi_bdev);
373 sector_t sector_offset = sector;
374 int max;
375 unsigned int chunk_sectors = mddev->chunk_sectors;
376 unsigned int bio_sectors = bvm->bi_size >> 9;
377 struct strip_zone *zone;
378 struct md_rdev *rdev;
379 struct request_queue *subq;
380
381 if (is_power_of_2(chunk_sectors))
382 max = (chunk_sectors - ((sector & (chunk_sectors-1))
383 + bio_sectors)) << 9;
384 else
385 max = (chunk_sectors - (sector_div(sector, chunk_sectors)
386 + bio_sectors)) << 9;
387 if (max < 0)
388 max = 0; /* bio_add cannot handle a negative return */
389 if (max <= biovec->bv_len && bio_sectors == 0)
390 return biovec->bv_len;
391 if (max < biovec->bv_len)
392 /* too small already, no need to check further */
393 return max;
394 if (!conf->has_merge_bvec)
395 return max;
396
397 /* May need to check subordinate device */
398 sector = sector_offset;
399 zone = find_zone(mddev->private, §or_offset);
400 rdev = map_sector(mddev, zone, sector, §or_offset);
401 subq = bdev_get_queue(rdev->bdev);
402 if (subq->merge_bvec_fn) {
403 bvm->bi_bdev = rdev->bdev;
404 bvm->bi_sector = sector_offset + zone->dev_start +
405 rdev->data_offset;
406 return min(max, subq->merge_bvec_fn(subq, bvm, biovec));
407 } else
408 return max;
409}
410
411static sector_t raid0_size(struct mddev *mddev, sector_t sectors, int raid_disks)
412{
413 sector_t array_sectors = 0;
414 struct md_rdev *rdev;
415
416 WARN_ONCE(sectors || raid_disks,
417 "%s does not support generic reshape\n", __func__);
418
419 rdev_for_each(rdev, mddev)
420 array_sectors += (rdev->sectors &
421 ~(sector_t)(mddev->chunk_sectors-1));
422
423 return array_sectors;
424}
425
426static int raid0_stop(struct mddev *mddev);
427
428static int raid0_run(struct mddev *mddev)
429{
430 struct r0conf *conf;
431 int ret;
432
433 if (mddev->chunk_sectors == 0) {
434 printk(KERN_ERR "md/raid0:%s: chunk size must be set.\n",
435 mdname(mddev));
436 return -EINVAL;
437 }
438 if (md_check_no_bitmap(mddev))
439 return -EINVAL;
440 blk_queue_max_hw_sectors(mddev->queue, mddev->chunk_sectors);
441 blk_queue_max_write_same_sectors(mddev->queue, mddev->chunk_sectors);
442 blk_queue_max_discard_sectors(mddev->queue, mddev->chunk_sectors);
443
444 /* if private is not null, we are here after takeover */
445 if (mddev->private == NULL) {
446 ret = create_strip_zones(mddev, &conf);
447 if (ret < 0)
448 return ret;
449 mddev->private = conf;
450 }
451 conf = mddev->private;
452
453 /* calculate array device size */
454 md_set_array_sectors(mddev, raid0_size(mddev, 0, 0));
455
456 printk(KERN_INFO "md/raid0:%s: md_size is %llu sectors.\n",
457 mdname(mddev),
458 (unsigned long long)mddev->array_sectors);
459 /* calculate the max read-ahead size.
460 * For read-ahead of large files to be effective, we need to
461 * readahead at least twice a whole stripe. i.e. number of devices
462 * multiplied by chunk size times 2.
463 * If an individual device has an ra_pages greater than the
464 * chunk size, then we will not drive that device as hard as it
465 * wants. We consider this a configuration error: a larger
466 * chunksize should be used in that case.
467 */
468 {
469 int stripe = mddev->raid_disks *
470 (mddev->chunk_sectors << 9) / PAGE_SIZE;
471 if (mddev->queue->backing_dev_info.ra_pages < 2* stripe)
472 mddev->queue->backing_dev_info.ra_pages = 2* stripe;
473 }
474
475 blk_queue_merge_bvec(mddev->queue, raid0_mergeable_bvec);
476 dump_zones(mddev);
477
478 ret = md_integrity_register(mddev);
479 if (ret)
480 raid0_stop(mddev);
481
482 return ret;
483}
484
485static int raid0_stop(struct mddev *mddev)
486{
487 struct r0conf *conf = mddev->private;
488
489 blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/
490 kfree(conf->strip_zone);
491 kfree(conf->devlist);
492 kfree(conf);
493 mddev->private = NULL;
494 return 0;
495}
496
497/*
498 * Is io distribute over 1 or more chunks ?
499*/
500static inline int is_io_in_chunk_boundary(struct mddev *mddev,
501 unsigned int chunk_sects, struct bio *bio)
502{
503 if (likely(is_power_of_2(chunk_sects))) {
504 return chunk_sects >=
505 ((bio->bi_iter.bi_sector & (chunk_sects-1))
506 + bio_sectors(bio));
507 } else{
508 sector_t sector = bio->bi_iter.bi_sector;
509 return chunk_sects >= (sector_div(sector, chunk_sects)
510 + bio_sectors(bio));
511 }
512}
513
514static void raid0_make_request(struct mddev *mddev, struct bio *bio)
515{
516 struct strip_zone *zone;
517 struct md_rdev *tmp_dev;
518 struct bio *split;
519
520 if (unlikely(bio->bi_rw & REQ_FLUSH)) {
521 md_flush_request(mddev, bio);
522 return;
523 }
524
525 do {
526 sector_t sector = bio->bi_iter.bi_sector;
527 unsigned chunk_sects = mddev->chunk_sectors;
528
529 unsigned sectors = chunk_sects -
530 (likely(is_power_of_2(chunk_sects))
531 ? (sector & (chunk_sects-1))
532 : sector_div(sector, chunk_sects));
533
534 if (sectors < bio_sectors(bio)) {
535 split = bio_split(bio, sectors, GFP_NOIO, fs_bio_set);
536 bio_chain(split, bio);
537 } else {
538 split = bio;
539 }
540
541 zone = find_zone(mddev->private, §or);
542 tmp_dev = map_sector(mddev, zone, sector, §or);
543 split->bi_bdev = tmp_dev->bdev;
544 split->bi_iter.bi_sector = sector + zone->dev_start +
545 tmp_dev->data_offset;
546
547 if (unlikely((split->bi_rw & REQ_DISCARD) &&
548 !blk_queue_discard(bdev_get_queue(split->bi_bdev)))) {
549 /* Just ignore it */
550 bio_endio(split, 0);
551 } else
552 generic_make_request(split);
553 } while (split != bio);
554}
555
556static void raid0_status(struct seq_file *seq, struct mddev *mddev)
557{
558 seq_printf(seq, " %dk chunks", mddev->chunk_sectors / 2);
559 return;
560}
561
562static void *raid0_takeover_raid45(struct mddev *mddev)
563{
564 struct md_rdev *rdev;
565 struct r0conf *priv_conf;
566
567 if (mddev->degraded != 1) {
568 printk(KERN_ERR "md/raid0:%s: raid5 must be degraded! Degraded disks: %d\n",
569 mdname(mddev),
570 mddev->degraded);
571 return ERR_PTR(-EINVAL);
572 }
573
574 rdev_for_each(rdev, mddev) {
575 /* check slot number for a disk */
576 if (rdev->raid_disk == mddev->raid_disks-1) {
577 printk(KERN_ERR "md/raid0:%s: raid5 must have missing parity disk!\n",
578 mdname(mddev));
579 return ERR_PTR(-EINVAL);
580 }
581 rdev->sectors = mddev->dev_sectors;
582 }
583
584 /* Set new parameters */
585 mddev->new_level = 0;
586 mddev->new_layout = 0;
587 mddev->new_chunk_sectors = mddev->chunk_sectors;
588 mddev->raid_disks--;
589 mddev->delta_disks = -1;
590 /* make sure it will be not marked as dirty */
591 mddev->recovery_cp = MaxSector;
592
593 create_strip_zones(mddev, &priv_conf);
594 return priv_conf;
595}
596
597static void *raid0_takeover_raid10(struct mddev *mddev)
598{
599 struct r0conf *priv_conf;
600
601 /* Check layout:
602 * - far_copies must be 1
603 * - near_copies must be 2
604 * - disks number must be even
605 * - all mirrors must be already degraded
606 */
607 if (mddev->layout != ((1 << 8) + 2)) {
608 printk(KERN_ERR "md/raid0:%s:: Raid0 cannot takover layout: 0x%x\n",
609 mdname(mddev),
610 mddev->layout);
611 return ERR_PTR(-EINVAL);
612 }
613 if (mddev->raid_disks & 1) {
614 printk(KERN_ERR "md/raid0:%s: Raid0 cannot takover Raid10 with odd disk number.\n",
615 mdname(mddev));
616 return ERR_PTR(-EINVAL);
617 }
618 if (mddev->degraded != (mddev->raid_disks>>1)) {
619 printk(KERN_ERR "md/raid0:%s: All mirrors must be already degraded!\n",
620 mdname(mddev));
621 return ERR_PTR(-EINVAL);
622 }
623
624 /* Set new parameters */
625 mddev->new_level = 0;
626 mddev->new_layout = 0;
627 mddev->new_chunk_sectors = mddev->chunk_sectors;
628 mddev->delta_disks = - mddev->raid_disks / 2;
629 mddev->raid_disks += mddev->delta_disks;
630 mddev->degraded = 0;
631 /* make sure it will be not marked as dirty */
632 mddev->recovery_cp = MaxSector;
633
634 create_strip_zones(mddev, &priv_conf);
635 return priv_conf;
636}
637
638static void *raid0_takeover_raid1(struct mddev *mddev)
639{
640 struct r0conf *priv_conf;
641 int chunksect;
642
643 /* Check layout:
644 * - (N - 1) mirror drives must be already faulty
645 */
646 if ((mddev->raid_disks - 1) != mddev->degraded) {
647 printk(KERN_ERR "md/raid0:%s: (N - 1) mirrors drives must be already faulty!\n",
648 mdname(mddev));
649 return ERR_PTR(-EINVAL);
650 }
651
652 /*
653 * a raid1 doesn't have the notion of chunk size, so
654 * figure out the largest suitable size we can use.
655 */
656 chunksect = 64 * 2; /* 64K by default */
657
658 /* The array must be an exact multiple of chunksize */
659 while (chunksect && (mddev->array_sectors & (chunksect - 1)))
660 chunksect >>= 1;
661
662 if ((chunksect << 9) < PAGE_SIZE)
663 /* array size does not allow a suitable chunk size */
664 return ERR_PTR(-EINVAL);
665
666 /* Set new parameters */
667 mddev->new_level = 0;
668 mddev->new_layout = 0;
669 mddev->new_chunk_sectors = chunksect;
670 mddev->chunk_sectors = chunksect;
671 mddev->delta_disks = 1 - mddev->raid_disks;
672 mddev->raid_disks = 1;
673 /* make sure it will be not marked as dirty */
674 mddev->recovery_cp = MaxSector;
675
676 create_strip_zones(mddev, &priv_conf);
677 return priv_conf;
678}
679
680static void *raid0_takeover(struct mddev *mddev)
681{
682 /* raid0 can take over:
683 * raid4 - if all data disks are active.
684 * raid5 - providing it is Raid4 layout and one disk is faulty
685 * raid10 - assuming we have all necessary active disks
686 * raid1 - with (N -1) mirror drives faulty
687 */
688 if (mddev->level == 4)
689 return raid0_takeover_raid45(mddev);
690
691 if (mddev->level == 5) {
692 if (mddev->layout == ALGORITHM_PARITY_N)
693 return raid0_takeover_raid45(mddev);
694
695 printk(KERN_ERR "md/raid0:%s: Raid can only takeover Raid5 with layout: %d\n",
696 mdname(mddev), ALGORITHM_PARITY_N);
697 }
698
699 if (mddev->level == 10)
700 return raid0_takeover_raid10(mddev);
701
702 if (mddev->level == 1)
703 return raid0_takeover_raid1(mddev);
704
705 printk(KERN_ERR "Takeover from raid%i to raid0 not supported\n",
706 mddev->level);
707
708 return ERR_PTR(-EINVAL);
709}
710
711static void raid0_quiesce(struct mddev *mddev, int state)
712{
713}
714
715static struct md_personality raid0_personality=
716{
717 .name = "raid0",
718 .level = 0,
719 .owner = THIS_MODULE,
720 .make_request = raid0_make_request,
721 .run = raid0_run,
722 .stop = raid0_stop,
723 .status = raid0_status,
724 .size = raid0_size,
725 .takeover = raid0_takeover,
726 .quiesce = raid0_quiesce,
727};
728
729static int __init raid0_init (void)
730{
731 return register_md_personality (&raid0_personality);
732}
733
734static void raid0_exit (void)
735{
736 unregister_md_personality (&raid0_personality);
737}
738
739module_init(raid0_init);
740module_exit(raid0_exit);
741MODULE_LICENSE("GPL");
742MODULE_DESCRIPTION("RAID0 (striping) personality for MD");
743MODULE_ALIAS("md-personality-2"); /* RAID0 */
744MODULE_ALIAS("md-raid0");
745MODULE_ALIAS("md-level-0");