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// 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");