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
22 static int default_layout = 0;
23 module_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 (1L << MD_FAILLAST_DEV) | \
32 (1L << MD_SERIALIZE_POLICY))
33
34 /*
35 * inform the user of the raid configuration
36 */
dump_zones(struct mddev * mddev)37 static void dump_zones(struct mddev *mddev)
38 {
39 int j, k;
40 sector_t zone_size = 0;
41 sector_t zone_start = 0;
42 struct r0conf *conf = mddev->private;
43 int raid_disks = conf->strip_zone[0].nb_dev;
44 pr_debug("md: RAID0 configuration for %s - %d zone%s\n",
45 mdname(mddev),
46 conf->nr_strip_zones, conf->nr_strip_zones==1?"":"s");
47 for (j = 0; j < conf->nr_strip_zones; j++) {
48 char line[200];
49 int len = 0;
50
51 for (k = 0; k < conf->strip_zone[j].nb_dev; k++)
52 len += scnprintf(line+len, 200-len, "%s%pg", k?"/":"",
53 conf->devlist[j * raid_disks + k]->bdev);
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
create_strip_zones(struct mddev * mddev,struct r0conf ** private_conf)65 static 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 struct r0conf *conf = kzalloc_obj(*conf, GFP_KERNEL);
73 unsigned int blksize = 512;
74
75 if (!mddev_is_dm(mddev))
76 blksize = queue_logical_block_size(mddev->gendisk->queue);
77
78 *private_conf = ERR_PTR(-ENOMEM);
79 if (!conf)
80 return -ENOMEM;
81 rdev_for_each(rdev1, mddev) {
82 pr_debug("md/raid0:%s: looking at %pg\n",
83 mdname(mddev),
84 rdev1->bdev);
85 c = 0;
86
87 /* round size to chunk_size */
88 sectors = rdev1->sectors;
89 sector_div(sectors, mddev->chunk_sectors);
90 rdev1->sectors = sectors * mddev->chunk_sectors;
91
92 if (mddev_is_dm(mddev))
93 blksize = max(blksize, queue_logical_block_size(
94 rdev1->bdev->bd_disk->queue));
95
96 rdev_for_each(rdev2, mddev) {
97 pr_debug("md/raid0:%s: comparing %pg(%llu)"
98 " with %pg(%llu)\n",
99 mdname(mddev),
100 rdev1->bdev,
101 (unsigned long long)rdev1->sectors,
102 rdev2->bdev,
103 (unsigned long long)rdev2->sectors);
104 if (rdev2 == rdev1) {
105 pr_debug("md/raid0:%s: END\n",
106 mdname(mddev));
107 break;
108 }
109 if (rdev2->sectors == rdev1->sectors) {
110 /*
111 * Not unique, don't count it as a new
112 * group
113 */
114 pr_debug("md/raid0:%s: EQUAL\n",
115 mdname(mddev));
116 c = 1;
117 break;
118 }
119 pr_debug("md/raid0:%s: NOT EQUAL\n",
120 mdname(mddev));
121 }
122 if (!c) {
123 pr_debug("md/raid0:%s: ==> UNIQUE\n",
124 mdname(mddev));
125 conf->nr_strip_zones++;
126 pr_debug("md/raid0:%s: %d zones\n",
127 mdname(mddev), conf->nr_strip_zones);
128 }
129 }
130 pr_debug("md/raid0:%s: FINAL %d zones\n",
131 mdname(mddev), conf->nr_strip_zones);
132
133 /*
134 * now since we have the hard sector sizes, we can make sure
135 * chunk size is a multiple of that sector size
136 */
137 if ((mddev->chunk_sectors << 9) % blksize) {
138 pr_warn("md/raid0:%s: chunk_size of %d not multiple of block size %d\n",
139 mdname(mddev),
140 mddev->chunk_sectors << 9, blksize);
141 err = -EINVAL;
142 goto abort;
143 }
144
145 err = -ENOMEM;
146 conf->strip_zone = kzalloc_objs(struct strip_zone, conf->nr_strip_zones,
147 GFP_KERNEL);
148 if (!conf->strip_zone)
149 goto abort;
150 conf->devlist = kzalloc(array3_size(sizeof(struct md_rdev *),
151 conf->nr_strip_zones,
152 mddev->raid_disks),
153 GFP_KERNEL);
154 if (!conf->devlist)
155 goto abort;
156
157 /* The first zone must contain all devices, so here we check that
158 * there is a proper alignment of slots to devices and find them all
159 */
160 zone = &conf->strip_zone[0];
161 cnt = 0;
162 smallest = NULL;
163 dev = conf->devlist;
164 err = -EINVAL;
165 rdev_for_each(rdev1, mddev) {
166 int j = rdev1->raid_disk;
167
168 if (mddev->level == 10) {
169 /* taking over a raid10-n2 array */
170 j /= 2;
171 rdev1->new_raid_disk = j;
172 }
173
174 if (mddev->level == 1) {
175 /* taiking over a raid1 array-
176 * we have only one active disk
177 */
178 j = 0;
179 rdev1->new_raid_disk = j;
180 }
181
182 if (j < 0) {
183 pr_warn("md/raid0:%s: remove inactive devices before converting to RAID0\n",
184 mdname(mddev));
185 goto abort;
186 }
187 if (j >= mddev->raid_disks) {
188 pr_warn("md/raid0:%s: bad disk number %d - aborting!\n",
189 mdname(mddev), j);
190 goto abort;
191 }
192 if (dev[j]) {
193 pr_warn("md/raid0:%s: multiple devices for %d - aborting!\n",
194 mdname(mddev), j);
195 goto abort;
196 }
197 dev[j] = rdev1;
198
199 if (!smallest || (rdev1->sectors < smallest->sectors))
200 smallest = rdev1;
201 cnt++;
202 }
203 if (cnt != mddev->raid_disks) {
204 pr_warn("md/raid0:%s: too few disks (%d of %d) - aborting!\n",
205 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 pr_debug("md/raid0:%s: zone %d\n", mdname(mddev), i);
222 zone->dev_start = smallest->sectors;
223 smallest = NULL;
224 c = 0;
225
226 for (j=0; j<cnt; j++) {
227 rdev = conf->devlist[j];
228 if (rdev->sectors <= zone->dev_start) {
229 pr_debug("md/raid0:%s: checking %pg ... nope\n",
230 mdname(mddev),
231 rdev->bdev);
232 continue;
233 }
234 pr_debug("md/raid0:%s: checking %pg ..."
235 " contained as device %d\n",
236 mdname(mddev),
237 rdev->bdev, c);
238 dev[c] = rdev;
239 c++;
240 if (!smallest || rdev->sectors < smallest->sectors) {
241 smallest = rdev;
242 pr_debug("md/raid0:%s: (%llu) is smallest!.\n",
243 mdname(mddev),
244 (unsigned long long)rdev->sectors);
245 }
246 }
247
248 zone->nb_dev = c;
249 sectors = (smallest->sectors - zone->dev_start) * c;
250 pr_debug("md/raid0:%s: zone->nb_dev: %d, sectors: %llu\n",
251 mdname(mddev),
252 zone->nb_dev, (unsigned long long)sectors);
253
254 curr_zone_end += sectors;
255 zone->zone_end = curr_zone_end;
256
257 pr_debug("md/raid0:%s: current zone start: %llu\n",
258 mdname(mddev),
259 (unsigned long long)smallest->sectors);
260 }
261
262 if (conf->nr_strip_zones == 1 || conf->strip_zone[1].nb_dev == 1) {
263 conf->layout = RAID0_ORIG_LAYOUT;
264 } else if (mddev->layout == RAID0_ORIG_LAYOUT ||
265 mddev->layout == RAID0_ALT_MULTIZONE_LAYOUT) {
266 conf->layout = mddev->layout;
267 } else if (default_layout == RAID0_ORIG_LAYOUT ||
268 default_layout == RAID0_ALT_MULTIZONE_LAYOUT) {
269 conf->layout = default_layout;
270 } else {
271 pr_err("md/raid0:%s: cannot assemble multi-zone RAID0 with default_layout setting\n",
272 mdname(mddev));
273 pr_err("md/raid0: please set raid0.default_layout to 1 or 2\n");
274 err = -EOPNOTSUPP;
275 goto abort;
276 }
277
278 if (conf->layout == RAID0_ORIG_LAYOUT) {
279 for (i = 1; i < conf->nr_strip_zones; i++) {
280 sector_t first_sector = conf->strip_zone[i-1].zone_end;
281
282 sector_div(first_sector, mddev->chunk_sectors);
283 zone = conf->strip_zone + i;
284 /* disk_shift is first disk index used in the zone */
285 zone->disk_shift = sector_div(first_sector,
286 zone->nb_dev);
287 }
288 }
289
290 pr_debug("md/raid0:%s: done.\n", mdname(mddev));
291 *private_conf = conf;
292
293 return 0;
294 abort:
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 */
find_zone(struct r0conf * conf,sector_t * sectorp)305 static 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 performance
324 */
map_sector(struct mddev * mddev,struct strip_zone * zone,sector_t sector,sector_t * sector_offset)325 static 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
raid0_size(struct mddev * mddev,sector_t sectors,int raid_disks)358 static sector_t raid0_size(struct mddev *mddev, sector_t sectors, int raid_disks)
359 {
360 sector_t array_sectors = 0;
361 struct md_rdev *rdev;
362
363 WARN_ONCE(sectors || raid_disks,
364 "%s does not support generic reshape\n", __func__);
365
366 rdev_for_each(rdev, mddev)
367 array_sectors += (rdev->sectors &
368 ~(sector_t)(mddev->chunk_sectors-1));
369
370 return array_sectors;
371 }
372
raid0_free(struct mddev * mddev,void * priv)373 static void raid0_free(struct mddev *mddev, void *priv)
374 {
375 struct r0conf *conf = priv;
376
377 kfree(conf->strip_zone);
378 kfree(conf->devlist);
379 kfree(conf);
380 }
381
raid0_set_limits(struct mddev * mddev)382 static int raid0_set_limits(struct mddev *mddev)
383 {
384 struct queue_limits lim;
385 int err;
386
387 md_init_stacking_limits(&lim);
388 lim.max_hw_sectors = mddev->chunk_sectors;
389 lim.max_write_zeroes_sectors = mddev->chunk_sectors;
390 lim.max_hw_wzeroes_unmap_sectors = mddev->chunk_sectors;
391 lim.logical_block_size = mddev->logical_block_size;
392 lim.io_min = mddev->chunk_sectors << 9;
393 lim.io_opt = lim.io_min * mddev->raid_disks;
394 lim.chunk_sectors = mddev->chunk_sectors;
395 lim.features |= BLK_FEAT_ATOMIC_WRITES;
396 err = mddev_stack_rdev_limits(mddev, &lim, MDDEV_STACK_INTEGRITY);
397 if (err)
398 return err;
399 return queue_limits_set(mddev->gendisk->queue, &lim);
400 }
401
raid0_run(struct mddev * mddev)402 static int raid0_run(struct mddev *mddev)
403 {
404 struct r0conf *conf;
405 int ret;
406
407 if (mddev->chunk_sectors == 0) {
408 pr_warn("md/raid0:%s: chunk size must be set.\n", mdname(mddev));
409 return -EINVAL;
410 }
411 if (md_check_no_bitmap(mddev))
412 return -EINVAL;
413
414 if (!mddev_is_dm(mddev)) {
415 ret = raid0_set_limits(mddev);
416 if (ret)
417 return ret;
418 }
419
420 /* if private is not null, we are here after takeover */
421 if (mddev->private == NULL) {
422 ret = create_strip_zones(mddev, &conf);
423 if (ret < 0)
424 return ret;
425 mddev->private = conf;
426 }
427 conf = mddev->private;
428
429 /* calculate array device size */
430 md_set_array_sectors(mddev, raid0_size(mddev, 0, 0));
431
432 pr_debug("md/raid0:%s: md_size is %llu sectors.\n",
433 mdname(mddev),
434 (unsigned long long)mddev->array_sectors);
435
436 dump_zones(mddev);
437
438 return md_integrity_register(mddev);
439 }
440
441 /*
442 * Convert disk_index to the disk order in which it is read/written.
443 * For example, if we have 4 disks, they are numbered 0,1,2,3. If we
444 * write the disks starting at disk 3, then the read/write order would
445 * be disk 3, then 0, then 1, and then disk 2 and we want map_disk_shift()
446 * to map the disks as follows 0,1,2,3 => 1,2,3,0. So disk 0 would map
447 * to 1, 1 to 2, 2 to 3, and 3 to 0. That way we can compare disks in
448 * that 'output' space to understand the read/write disk ordering.
449 */
map_disk_shift(int disk_index,int num_disks,int disk_shift)450 static int map_disk_shift(int disk_index, int num_disks, int disk_shift)
451 {
452 return ((disk_index + num_disks - disk_shift) % num_disks);
453 }
454
raid0_handle_discard(struct mddev * mddev,struct bio * bio)455 static void raid0_handle_discard(struct mddev *mddev, struct bio *bio)
456 {
457 struct r0conf *conf = mddev->private;
458 struct strip_zone *zone;
459 sector_t start = bio->bi_iter.bi_sector;
460 sector_t end;
461 unsigned int stripe_size;
462 sector_t first_stripe_index, last_stripe_index;
463 sector_t start_disk_offset;
464 unsigned int start_disk_index;
465 sector_t end_disk_offset;
466 unsigned int end_disk_index;
467 unsigned int disk;
468 sector_t orig_start, orig_end;
469
470 orig_start = start;
471 zone = find_zone(conf, &start);
472
473 if (bio_end_sector(bio) > zone->zone_end) {
474 bio = bio_submit_split_bioset(bio,
475 zone->zone_end - bio->bi_iter.bi_sector,
476 &mddev->bio_set);
477 if (!bio)
478 return;
479
480 end = zone->zone_end;
481 } else {
482 end = bio_end_sector(bio);
483 }
484
485 orig_end = end;
486 if (zone != conf->strip_zone)
487 end = end - zone[-1].zone_end;
488
489 /* Now start and end is the offset in zone */
490 stripe_size = zone->nb_dev * mddev->chunk_sectors;
491
492 first_stripe_index = start;
493 sector_div(first_stripe_index, stripe_size);
494 last_stripe_index = end;
495 sector_div(last_stripe_index, stripe_size);
496
497 /* In the first zone the original and alternate layouts are the same */
498 if ((conf->layout == RAID0_ORIG_LAYOUT) && (zone != conf->strip_zone)) {
499 sector_div(orig_start, mddev->chunk_sectors);
500 start_disk_index = sector_div(orig_start, zone->nb_dev);
501 start_disk_index = map_disk_shift(start_disk_index,
502 zone->nb_dev,
503 zone->disk_shift);
504 sector_div(orig_end, mddev->chunk_sectors);
505 end_disk_index = sector_div(orig_end, zone->nb_dev);
506 end_disk_index = map_disk_shift(end_disk_index,
507 zone->nb_dev, zone->disk_shift);
508 } else {
509 start_disk_index = (int)(start - first_stripe_index * stripe_size) /
510 mddev->chunk_sectors;
511 end_disk_index = (int)(end - last_stripe_index * stripe_size) /
512 mddev->chunk_sectors;
513 }
514 start_disk_offset = ((int)(start - first_stripe_index * stripe_size) %
515 mddev->chunk_sectors) +
516 first_stripe_index * mddev->chunk_sectors;
517 end_disk_offset = ((int)(end - last_stripe_index * stripe_size) %
518 mddev->chunk_sectors) +
519 last_stripe_index * mddev->chunk_sectors;
520
521 for (disk = 0; disk < zone->nb_dev; disk++) {
522 sector_t dev_start, dev_end;
523 struct md_rdev *rdev;
524 int compare_disk;
525
526 compare_disk = map_disk_shift(disk, zone->nb_dev,
527 zone->disk_shift);
528
529 if (compare_disk < start_disk_index)
530 dev_start = (first_stripe_index + 1) *
531 mddev->chunk_sectors;
532 else if (compare_disk > start_disk_index)
533 dev_start = first_stripe_index * mddev->chunk_sectors;
534 else
535 dev_start = start_disk_offset;
536
537 if (compare_disk < end_disk_index)
538 dev_end = (last_stripe_index + 1) * mddev->chunk_sectors;
539 else if (compare_disk > end_disk_index)
540 dev_end = last_stripe_index * mddev->chunk_sectors;
541 else
542 dev_end = end_disk_offset;
543
544 if (dev_end <= dev_start)
545 continue;
546
547 rdev = conf->devlist[(zone - conf->strip_zone) *
548 conf->strip_zone[0].nb_dev + disk];
549 md_submit_discard_bio(mddev, rdev, bio,
550 dev_start + zone->dev_start + rdev->data_offset,
551 dev_end - dev_start);
552 }
553 bio_endio(bio);
554 }
555
raid0_map_submit_bio(struct mddev * mddev,struct bio * bio)556 static void raid0_map_submit_bio(struct mddev *mddev, struct bio *bio)
557 {
558 struct r0conf *conf = mddev->private;
559 struct strip_zone *zone;
560 struct md_rdev *tmp_dev;
561 sector_t bio_sector = bio->bi_iter.bi_sector;
562 sector_t sector = bio_sector;
563
564 md_account_bio(mddev, &bio);
565
566 zone = find_zone(mddev->private, §or);
567 switch (conf->layout) {
568 case RAID0_ORIG_LAYOUT:
569 tmp_dev = map_sector(mddev, zone, bio_sector, §or);
570 break;
571 case RAID0_ALT_MULTIZONE_LAYOUT:
572 tmp_dev = map_sector(mddev, zone, sector, §or);
573 break;
574 default:
575 WARN(1, "md/raid0:%s: Invalid layout\n", mdname(mddev));
576 bio_io_error(bio);
577 return;
578 }
579
580 if (unlikely(is_rdev_broken(tmp_dev))) {
581 bio_io_error(bio);
582 md_error(mddev, tmp_dev);
583 return;
584 }
585
586 bio_set_dev(bio, tmp_dev->bdev);
587 bio->bi_iter.bi_sector = sector + zone->dev_start +
588 tmp_dev->data_offset;
589 mddev_trace_remap(mddev, bio, bio_sector);
590 mddev_check_write_zeroes(mddev, bio);
591 submit_bio_noacct(bio);
592 }
593
raid0_make_request(struct mddev * mddev,struct bio * bio)594 static bool raid0_make_request(struct mddev *mddev, struct bio *bio)
595 {
596 sector_t sector;
597 unsigned chunk_sects;
598 unsigned sectors;
599
600 if (unlikely(bio->bi_opf & REQ_PREFLUSH)
601 && md_flush_request(mddev, bio))
602 return true;
603
604 if (unlikely((bio_op(bio) == REQ_OP_DISCARD))) {
605 raid0_handle_discard(mddev, bio);
606 return true;
607 }
608
609 sector = bio->bi_iter.bi_sector;
610 chunk_sects = mddev->chunk_sectors;
611
612 sectors = chunk_sects -
613 (likely(is_power_of_2(chunk_sects))
614 ? (sector & (chunk_sects-1))
615 : sector_div(sector, chunk_sects));
616
617 if (sectors < bio_sectors(bio)) {
618 bio = bio_submit_split_bioset(bio, sectors,
619 &mddev->bio_set);
620 if (!bio)
621 return true;
622 }
623
624 raid0_map_submit_bio(mddev, bio);
625 return true;
626 }
627
raid0_status(struct seq_file * seq,struct mddev * mddev)628 static void raid0_status(struct seq_file *seq, struct mddev *mddev)
629 {
630 seq_printf(seq, " %dk chunks", mddev->chunk_sectors / 2);
631 return;
632 }
633
raid0_error(struct mddev * mddev,struct md_rdev * rdev)634 static void raid0_error(struct mddev *mddev, struct md_rdev *rdev)
635 {
636 if (!test_and_set_bit(MD_BROKEN, &mddev->flags)) {
637 char *md_name = mdname(mddev);
638
639 pr_crit("md/raid0%s: Disk failure on %pg detected, failing array.\n",
640 md_name, rdev->bdev);
641 }
642 }
643
raid0_takeover_raid45(struct mddev * mddev)644 static void *raid0_takeover_raid45(struct mddev *mddev)
645 {
646 struct md_rdev *rdev;
647 struct r0conf *priv_conf;
648
649 if (mddev->degraded != 1) {
650 pr_warn("md/raid0:%s: raid5 must be degraded! Degraded disks: %d\n",
651 mdname(mddev),
652 mddev->degraded);
653 return ERR_PTR(-EINVAL);
654 }
655
656 rdev_for_each(rdev, mddev) {
657 /* check slot number for a disk */
658 if (rdev->raid_disk == mddev->raid_disks-1) {
659 pr_warn("md/raid0:%s: raid5 must have missing parity disk!\n",
660 mdname(mddev));
661 return ERR_PTR(-EINVAL);
662 }
663 rdev->sectors = mddev->dev_sectors;
664 }
665
666 /* Set new parameters */
667 mddev->new_level = 0;
668 mddev->new_layout = 0;
669 mddev->new_chunk_sectors = mddev->chunk_sectors;
670 mddev->raid_disks--;
671 mddev->delta_disks = -1;
672 /* make sure it will be not marked as dirty */
673 mddev->resync_offset = MaxSector;
674 mddev_clear_unsupported_flags(mddev, UNSUPPORTED_MDDEV_FLAGS);
675
676 create_strip_zones(mddev, &priv_conf);
677
678 return priv_conf;
679 }
680
raid0_takeover_raid10(struct mddev * mddev)681 static void *raid0_takeover_raid10(struct mddev *mddev)
682 {
683 struct r0conf *priv_conf;
684
685 /* Check layout:
686 * - far_copies must be 1
687 * - near_copies must be 2
688 * - disks number must be even
689 * - all mirrors must be already degraded
690 */
691 if (mddev->layout != ((1 << 8) + 2)) {
692 pr_warn("md/raid0:%s:: Raid0 cannot takeover layout: 0x%x\n",
693 mdname(mddev),
694 mddev->layout);
695 return ERR_PTR(-EINVAL);
696 }
697 if (mddev->raid_disks & 1) {
698 pr_warn("md/raid0:%s: Raid0 cannot takeover Raid10 with odd disk number.\n",
699 mdname(mddev));
700 return ERR_PTR(-EINVAL);
701 }
702 if (mddev->degraded != (mddev->raid_disks>>1)) {
703 pr_warn("md/raid0:%s: All mirrors must be already degraded!\n",
704 mdname(mddev));
705 return ERR_PTR(-EINVAL);
706 }
707
708 /* Set new parameters */
709 mddev->new_level = 0;
710 mddev->new_layout = 0;
711 mddev->new_chunk_sectors = mddev->chunk_sectors;
712 mddev->delta_disks = - mddev->raid_disks / 2;
713 mddev->raid_disks += mddev->delta_disks;
714 mddev->degraded = 0;
715 /* make sure it will be not marked as dirty */
716 mddev->resync_offset = MaxSector;
717 mddev_clear_unsupported_flags(mddev, UNSUPPORTED_MDDEV_FLAGS);
718
719 create_strip_zones(mddev, &priv_conf);
720 return priv_conf;
721 }
722
raid0_takeover_raid1(struct mddev * mddev)723 static void *raid0_takeover_raid1(struct mddev *mddev)
724 {
725 struct r0conf *priv_conf;
726 int chunksect;
727
728 /* Check layout:
729 * - (N - 1) mirror drives must be already faulty
730 */
731 if ((mddev->raid_disks - 1) != mddev->degraded) {
732 pr_err("md/raid0:%s: (N - 1) mirrors drives must be already faulty!\n",
733 mdname(mddev));
734 return ERR_PTR(-EINVAL);
735 }
736
737 /*
738 * a raid1 doesn't have the notion of chunk size, so
739 * figure out the largest suitable size we can use.
740 */
741 chunksect = 64 * 2; /* 64K by default */
742
743 /* The array must be an exact multiple of chunksize */
744 while (chunksect && (mddev->array_sectors & (chunksect - 1)))
745 chunksect >>= 1;
746
747 if ((chunksect << 9) < PAGE_SIZE)
748 /* array size does not allow a suitable chunk size */
749 return ERR_PTR(-EINVAL);
750
751 /* Set new parameters */
752 mddev->new_level = 0;
753 mddev->new_layout = 0;
754 mddev->new_chunk_sectors = chunksect;
755 mddev->chunk_sectors = chunksect;
756 mddev->delta_disks = 1 - mddev->raid_disks;
757 mddev->raid_disks = 1;
758 /* make sure it will be not marked as dirty */
759 mddev->resync_offset = MaxSector;
760 mddev_clear_unsupported_flags(mddev, UNSUPPORTED_MDDEV_FLAGS);
761
762 create_strip_zones(mddev, &priv_conf);
763 return priv_conf;
764 }
765
raid0_takeover(struct mddev * mddev)766 static void *raid0_takeover(struct mddev *mddev)
767 {
768 /* raid0 can take over:
769 * raid4 - if all data disks are active.
770 * raid5 - providing it is Raid4 layout and one disk is faulty
771 * raid10 - assuming we have all necessary active disks
772 * raid1 - with (N -1) mirror drives faulty
773 */
774
775 if (mddev->bitmap) {
776 pr_warn("md/raid0: %s: cannot takeover array with bitmap\n",
777 mdname(mddev));
778 return ERR_PTR(-EBUSY);
779 }
780 if (mddev->level == 4)
781 return raid0_takeover_raid45(mddev);
782
783 if (mddev->level == 5) {
784 if (mddev->layout == ALGORITHM_PARITY_N)
785 return raid0_takeover_raid45(mddev);
786
787 pr_warn("md/raid0:%s: Raid can only takeover Raid5 with layout: %d\n",
788 mdname(mddev), ALGORITHM_PARITY_N);
789 }
790
791 if (mddev->level == 10)
792 return raid0_takeover_raid10(mddev);
793
794 if (mddev->level == 1)
795 return raid0_takeover_raid1(mddev);
796
797 pr_warn("Takeover from raid%i to raid0 not supported\n",
798 mddev->level);
799
800 return ERR_PTR(-EINVAL);
801 }
802
raid0_quiesce(struct mddev * mddev,int quiesce)803 static void raid0_quiesce(struct mddev *mddev, int quiesce)
804 {
805 }
806
807 static struct md_personality raid0_personality=
808 {
809 .head = {
810 .type = MD_PERSONALITY,
811 .id = ID_RAID0,
812 .name = "raid0",
813 .owner = THIS_MODULE,
814 },
815
816 .make_request = raid0_make_request,
817 .run = raid0_run,
818 .free = raid0_free,
819 .status = raid0_status,
820 .size = raid0_size,
821 .takeover = raid0_takeover,
822 .quiesce = raid0_quiesce,
823 .error_handler = raid0_error,
824 };
825
raid0_init(void)826 static int __init raid0_init(void)
827 {
828 return register_md_submodule(&raid0_personality.head);
829 }
830
raid0_exit(void)831 static void __exit raid0_exit(void)
832 {
833 unregister_md_submodule(&raid0_personality.head);
834 }
835
836 module_init(raid0_init);
837 module_exit(raid0_exit);
838 MODULE_LICENSE("GPL");
839 MODULE_DESCRIPTION("RAID0 (striping) personality for MD");
840 MODULE_ALIAS("md-personality-2"); /* RAID0 */
841 MODULE_ALIAS("md-raid0");
842 MODULE_ALIAS("md-level-0");
843