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