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
32 /*
33 * inform the user of the raid configuration
34 */
dump_zones(struct mddev * mddev)35 static 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 += scnprintf(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
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 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 /*
132 * now since we have the hard sector sizes, we can make sure
133 * chunk size is a multiple of that sector size
134 */
135 if ((mddev->chunk_sectors << 9) % blksize) {
136 pr_warn("md/raid0:%s: chunk_size of %d not multiple of block size %d\n",
137 mdname(mddev),
138 mddev->chunk_sectors << 9, blksize);
139 err = -EINVAL;
140 goto abort;
141 }
142
143 err = -ENOMEM;
144 conf->strip_zone = kcalloc(conf->nr_strip_zones,
145 sizeof(struct strip_zone),
146 GFP_KERNEL);
147 if (!conf->strip_zone)
148 goto abort;
149 conf->devlist = kzalloc(array3_size(sizeof(struct md_rdev *),
150 conf->nr_strip_zones,
151 mddev->raid_disks),
152 GFP_KERNEL);
153 if (!conf->devlist)
154 goto abort;
155
156 /* The first zone must contain all devices, so here we check that
157 * there is a proper alignment of slots to devices and find them all
158 */
159 zone = &conf->strip_zone[0];
160 cnt = 0;
161 smallest = NULL;
162 dev = conf->devlist;
163 err = -EINVAL;
164 rdev_for_each(rdev1, mddev) {
165 int j = rdev1->raid_disk;
166
167 if (mddev->level == 10) {
168 /* taking over a raid10-n2 array */
169 j /= 2;
170 rdev1->new_raid_disk = j;
171 }
172
173 if (mddev->level == 1) {
174 /* taiking over a raid1 array-
175 * we have only one active disk
176 */
177 j = 0;
178 rdev1->new_raid_disk = j;
179 }
180
181 if (j < 0) {
182 pr_warn("md/raid0:%s: remove inactive devices before converting to RAID0\n",
183 mdname(mddev));
184 goto abort;
185 }
186 if (j >= mddev->raid_disks) {
187 pr_warn("md/raid0:%s: bad disk number %d - aborting!\n",
188 mdname(mddev), j);
189 goto abort;
190 }
191 if (dev[j]) {
192 pr_warn("md/raid0:%s: multiple devices for %d - aborting!\n",
193 mdname(mddev), j);
194 goto abort;
195 }
196 dev[j] = rdev1;
197
198 if (!smallest || (rdev1->sectors < smallest->sectors))
199 smallest = rdev1;
200 cnt++;
201 }
202 if (cnt != mddev->raid_disks) {
203 pr_warn("md/raid0:%s: too few disks (%d of %d) - aborting!\n",
204 mdname(mddev), cnt, mddev->raid_disks);
205 goto abort;
206 }
207 zone->nb_dev = cnt;
208 zone->zone_end = smallest->sectors * cnt;
209
210 curr_zone_end = zone->zone_end;
211
212 /* now do the other zones */
213 for (i = 1; i < conf->nr_strip_zones; i++)
214 {
215 int j;
216
217 zone = conf->strip_zone + i;
218 dev = conf->devlist + i * mddev->raid_disks;
219
220 pr_debug("md/raid0:%s: zone %d\n", mdname(mddev), i);
221 zone->dev_start = smallest->sectors;
222 smallest = NULL;
223 c = 0;
224
225 for (j=0; j<cnt; j++) {
226 rdev = conf->devlist[j];
227 if (rdev->sectors <= zone->dev_start) {
228 pr_debug("md/raid0:%s: checking %s ... nope\n",
229 mdname(mddev),
230 bdevname(rdev->bdev, b));
231 continue;
232 }
233 pr_debug("md/raid0:%s: checking %s ..."
234 " contained as device %d\n",
235 mdname(mddev),
236 bdevname(rdev->bdev, b), c);
237 dev[c] = rdev;
238 c++;
239 if (!smallest || rdev->sectors < smallest->sectors) {
240 smallest = rdev;
241 pr_debug("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 pr_debug("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 pr_debug("md/raid0:%s: current zone start: %llu\n",
257 mdname(mddev),
258 (unsigned long long)smallest->sectors);
259 }
260
261 if (conf->nr_strip_zones == 1 || conf->strip_zone[1].nb_dev == 1) {
262 conf->layout = RAID0_ORIG_LAYOUT;
263 } else if (mddev->layout == RAID0_ORIG_LAYOUT ||
264 mddev->layout == RAID0_ALT_MULTIZONE_LAYOUT) {
265 conf->layout = mddev->layout;
266 } else if (default_layout == RAID0_ORIG_LAYOUT ||
267 default_layout == RAID0_ALT_MULTIZONE_LAYOUT) {
268 conf->layout = default_layout;
269 } else {
270 pr_err("md/raid0:%s: cannot assemble multi-zone RAID0 with default_layout setting\n",
271 mdname(mddev));
272 pr_err("md/raid0: please set raid0.default_layout to 1 or 2\n");
273 err = -EOPNOTSUPP;
274 goto abort;
275 }
276
277 if (conf->layout == RAID0_ORIG_LAYOUT) {
278 for (i = 1; i < conf->nr_strip_zones; i++) {
279 sector_t first_sector = conf->strip_zone[i-1].zone_end;
280
281 sector_div(first_sector, mddev->chunk_sectors);
282 zone = conf->strip_zone + i;
283 /* disk_shift is first disk index used in the zone */
284 zone->disk_shift = sector_div(first_sector,
285 zone->nb_dev);
286 }
287 }
288
289 pr_debug("md/raid0:%s: done.\n", mdname(mddev));
290 *private_conf = conf;
291
292 return 0;
293 abort:
294 kfree(conf->strip_zone);
295 kfree(conf->devlist);
296 kfree(conf);
297 *private_conf = ERR_PTR(err);
298 return err;
299 }
300
301 /* Find the zone which holds a particular offset
302 * Update *sectorp to be an offset in that zone
303 */
find_zone(struct r0conf * conf,sector_t * sectorp)304 static struct strip_zone *find_zone(struct r0conf *conf,
305 sector_t *sectorp)
306 {
307 int i;
308 struct strip_zone *z = conf->strip_zone;
309 sector_t sector = *sectorp;
310
311 for (i = 0; i < conf->nr_strip_zones; i++)
312 if (sector < z[i].zone_end) {
313 if (i)
314 *sectorp = sector - z[i-1].zone_end;
315 return z + i;
316 }
317 BUG();
318 }
319
320 /*
321 * remaps the bio to the target device. we separate two flows.
322 * power 2 flow and a general flow for the sake of performance
323 */
map_sector(struct mddev * mddev,struct strip_zone * zone,sector_t sector,sector_t * sector_offset)324 static struct md_rdev *map_sector(struct mddev *mddev, struct strip_zone *zone,
325 sector_t sector, sector_t *sector_offset)
326 {
327 unsigned int sect_in_chunk;
328 sector_t chunk;
329 struct r0conf *conf = mddev->private;
330 int raid_disks = conf->strip_zone[0].nb_dev;
331 unsigned int chunk_sects = mddev->chunk_sectors;
332
333 if (is_power_of_2(chunk_sects)) {
334 int chunksect_bits = ffz(~chunk_sects);
335 /* find the sector offset inside the chunk */
336 sect_in_chunk = sector & (chunk_sects - 1);
337 sector >>= chunksect_bits;
338 /* chunk in zone */
339 chunk = *sector_offset;
340 /* quotient is the chunk in real device*/
341 sector_div(chunk, zone->nb_dev << chunksect_bits);
342 } else{
343 sect_in_chunk = sector_div(sector, chunk_sects);
344 chunk = *sector_offset;
345 sector_div(chunk, chunk_sects * zone->nb_dev);
346 }
347 /*
348 * position the bio over the real device
349 * real sector = chunk in device + starting of zone
350 * + the position in the chunk
351 */
352 *sector_offset = (chunk * chunk_sects) + sect_in_chunk;
353 return conf->devlist[(zone - conf->strip_zone)*raid_disks
354 + sector_div(sector, zone->nb_dev)];
355 }
356
raid0_size(struct mddev * mddev,sector_t sectors,int raid_disks)357 static sector_t raid0_size(struct mddev *mddev, sector_t sectors, int raid_disks)
358 {
359 sector_t array_sectors = 0;
360 struct md_rdev *rdev;
361
362 WARN_ONCE(sectors || raid_disks,
363 "%s does not support generic reshape\n", __func__);
364
365 rdev_for_each(rdev, mddev)
366 array_sectors += (rdev->sectors &
367 ~(sector_t)(mddev->chunk_sectors-1));
368
369 return array_sectors;
370 }
371
free_conf(struct mddev * mddev,struct r0conf * conf)372 static void free_conf(struct mddev *mddev, struct r0conf *conf)
373 {
374 kfree(conf->strip_zone);
375 kfree(conf->devlist);
376 kfree(conf);
377 }
378
raid0_free(struct mddev * mddev,void * priv)379 static void raid0_free(struct mddev *mddev, void *priv)
380 {
381 struct r0conf *conf = priv;
382
383 free_conf(mddev, conf);
384 acct_bioset_exit(mddev);
385 }
386
raid0_run(struct mddev * mddev)387 static int raid0_run(struct mddev *mddev)
388 {
389 struct r0conf *conf;
390 int ret;
391
392 if (mddev->chunk_sectors == 0) {
393 pr_warn("md/raid0:%s: chunk size must be set.\n", mdname(mddev));
394 return -EINVAL;
395 }
396 if (md_check_no_bitmap(mddev))
397 return -EINVAL;
398
399 if (acct_bioset_init(mddev)) {
400 pr_err("md/raid0:%s: alloc acct bioset failed.\n", mdname(mddev));
401 return -ENOMEM;
402 }
403
404 /* if private is not null, we are here after takeover */
405 if (mddev->private == NULL) {
406 ret = create_strip_zones(mddev, &conf);
407 if (ret < 0)
408 goto exit_acct_set;
409 mddev->private = conf;
410 }
411 conf = mddev->private;
412 if (mddev->queue) {
413 struct md_rdev *rdev;
414 bool discard_supported = false;
415
416 blk_queue_max_hw_sectors(mddev->queue, mddev->chunk_sectors);
417 blk_queue_max_write_same_sectors(mddev->queue, mddev->chunk_sectors);
418 blk_queue_max_write_zeroes_sectors(mddev->queue, mddev->chunk_sectors);
419 blk_queue_max_discard_sectors(mddev->queue, UINT_MAX);
420
421 blk_queue_io_min(mddev->queue, mddev->chunk_sectors << 9);
422 blk_queue_io_opt(mddev->queue,
423 (mddev->chunk_sectors << 9) * mddev->raid_disks);
424
425 rdev_for_each(rdev, mddev) {
426 disk_stack_limits(mddev->gendisk, rdev->bdev,
427 rdev->data_offset << 9);
428 if (blk_queue_discard(bdev_get_queue(rdev->bdev)))
429 discard_supported = true;
430 }
431 if (!discard_supported)
432 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, mddev->queue);
433 else
434 blk_queue_flag_set(QUEUE_FLAG_DISCARD, mddev->queue);
435 }
436
437 /* calculate array device size */
438 md_set_array_sectors(mddev, raid0_size(mddev, 0, 0));
439
440 pr_debug("md/raid0:%s: md_size is %llu sectors.\n",
441 mdname(mddev),
442 (unsigned long long)mddev->array_sectors);
443
444 dump_zones(mddev);
445
446 ret = md_integrity_register(mddev);
447 if (ret)
448 goto free;
449
450 return ret;
451
452 free:
453 free_conf(mddev, conf);
454 exit_acct_set:
455 acct_bioset_exit(mddev);
456 return ret;
457 }
458
459 /*
460 * Convert disk_index to the disk order in which it is read/written.
461 * For example, if we have 4 disks, they are numbered 0,1,2,3. If we
462 * write the disks starting at disk 3, then the read/write order would
463 * be disk 3, then 0, then 1, and then disk 2 and we want map_disk_shift()
464 * to map the disks as follows 0,1,2,3 => 1,2,3,0. So disk 0 would map
465 * to 1, 1 to 2, 2 to 3, and 3 to 0. That way we can compare disks in
466 * that 'output' space to understand the read/write disk ordering.
467 */
map_disk_shift(int disk_index,int num_disks,int disk_shift)468 static int map_disk_shift(int disk_index, int num_disks, int disk_shift)
469 {
470 return ((disk_index + num_disks - disk_shift) % num_disks);
471 }
472
raid0_handle_discard(struct mddev * mddev,struct bio * bio)473 static void raid0_handle_discard(struct mddev *mddev, struct bio *bio)
474 {
475 struct r0conf *conf = mddev->private;
476 struct strip_zone *zone;
477 sector_t start = bio->bi_iter.bi_sector;
478 sector_t end;
479 unsigned int stripe_size;
480 sector_t first_stripe_index, last_stripe_index;
481 sector_t start_disk_offset;
482 unsigned int start_disk_index;
483 sector_t end_disk_offset;
484 unsigned int end_disk_index;
485 unsigned int disk;
486 sector_t orig_start, orig_end;
487
488 orig_start = start;
489 zone = find_zone(conf, &start);
490
491 if (bio_end_sector(bio) > zone->zone_end) {
492 struct bio *split = bio_split(bio,
493 zone->zone_end - bio->bi_iter.bi_sector, GFP_NOIO,
494 &mddev->bio_set);
495 bio_chain(split, bio);
496 submit_bio_noacct(bio);
497 bio = split;
498 end = zone->zone_end;
499 } else
500 end = bio_end_sector(bio);
501
502 orig_end = end;
503 if (zone != conf->strip_zone)
504 end = end - zone[-1].zone_end;
505
506 /* Now start and end is the offset in zone */
507 stripe_size = zone->nb_dev * mddev->chunk_sectors;
508
509 first_stripe_index = start;
510 sector_div(first_stripe_index, stripe_size);
511 last_stripe_index = end;
512 sector_div(last_stripe_index, stripe_size);
513
514 /* In the first zone the original and alternate layouts are the same */
515 if ((conf->layout == RAID0_ORIG_LAYOUT) && (zone != conf->strip_zone)) {
516 sector_div(orig_start, mddev->chunk_sectors);
517 start_disk_index = sector_div(orig_start, zone->nb_dev);
518 start_disk_index = map_disk_shift(start_disk_index,
519 zone->nb_dev,
520 zone->disk_shift);
521 sector_div(orig_end, mddev->chunk_sectors);
522 end_disk_index = sector_div(orig_end, zone->nb_dev);
523 end_disk_index = map_disk_shift(end_disk_index,
524 zone->nb_dev, zone->disk_shift);
525 } else {
526 start_disk_index = (int)(start - first_stripe_index * stripe_size) /
527 mddev->chunk_sectors;
528 end_disk_index = (int)(end - last_stripe_index * stripe_size) /
529 mddev->chunk_sectors;
530 }
531 start_disk_offset = ((int)(start - first_stripe_index * stripe_size) %
532 mddev->chunk_sectors) +
533 first_stripe_index * mddev->chunk_sectors;
534 end_disk_offset = ((int)(end - last_stripe_index * stripe_size) %
535 mddev->chunk_sectors) +
536 last_stripe_index * mddev->chunk_sectors;
537
538 for (disk = 0; disk < zone->nb_dev; disk++) {
539 sector_t dev_start, dev_end;
540 struct md_rdev *rdev;
541 int compare_disk;
542
543 compare_disk = map_disk_shift(disk, zone->nb_dev,
544 zone->disk_shift);
545
546 if (compare_disk < start_disk_index)
547 dev_start = (first_stripe_index + 1) *
548 mddev->chunk_sectors;
549 else if (compare_disk > start_disk_index)
550 dev_start = first_stripe_index * mddev->chunk_sectors;
551 else
552 dev_start = start_disk_offset;
553
554 if (compare_disk < end_disk_index)
555 dev_end = (last_stripe_index + 1) * mddev->chunk_sectors;
556 else if (compare_disk > end_disk_index)
557 dev_end = last_stripe_index * mddev->chunk_sectors;
558 else
559 dev_end = end_disk_offset;
560
561 if (dev_end <= dev_start)
562 continue;
563
564 rdev = conf->devlist[(zone - conf->strip_zone) *
565 conf->strip_zone[0].nb_dev + disk];
566 md_submit_discard_bio(mddev, rdev, bio,
567 dev_start + zone->dev_start + rdev->data_offset,
568 dev_end - dev_start);
569 }
570 bio_endio(bio);
571 }
572
raid0_map_submit_bio(struct mddev * mddev,struct bio * bio)573 static void raid0_map_submit_bio(struct mddev *mddev, struct bio *bio)
574 {
575 struct r0conf *conf = mddev->private;
576 struct strip_zone *zone;
577 struct md_rdev *tmp_dev;
578 sector_t bio_sector = bio->bi_iter.bi_sector;
579 sector_t sector = bio_sector;
580
581 md_account_bio(mddev, &bio);
582
583 zone = find_zone(mddev->private, §or);
584 switch (conf->layout) {
585 case RAID0_ORIG_LAYOUT:
586 tmp_dev = map_sector(mddev, zone, bio_sector, §or);
587 break;
588 case RAID0_ALT_MULTIZONE_LAYOUT:
589 tmp_dev = map_sector(mddev, zone, sector, §or);
590 break;
591 default:
592 WARN(1, "md/raid0:%s: Invalid layout\n", mdname(mddev));
593 bio_io_error(bio);
594 return;
595 }
596
597 if (unlikely(is_rdev_broken(tmp_dev))) {
598 bio_io_error(bio);
599 md_error(mddev, tmp_dev);
600 return;
601 }
602
603 bio_set_dev(bio, tmp_dev->bdev);
604 bio->bi_iter.bi_sector = sector + zone->dev_start +
605 tmp_dev->data_offset;
606
607 if (mddev->gendisk)
608 trace_block_bio_remap(bio, disk_devt(mddev->gendisk),
609 bio_sector);
610 mddev_check_writesame(mddev, bio);
611 mddev_check_write_zeroes(mddev, bio);
612 submit_bio_noacct(bio);
613 }
614
raid0_make_request(struct mddev * mddev,struct bio * bio)615 static bool raid0_make_request(struct mddev *mddev, struct bio *bio)
616 {
617 sector_t sector;
618 unsigned chunk_sects;
619 unsigned sectors;
620
621 if (unlikely(bio->bi_opf & REQ_PREFLUSH)
622 && md_flush_request(mddev, bio))
623 return true;
624
625 if (unlikely((bio_op(bio) == REQ_OP_DISCARD))) {
626 raid0_handle_discard(mddev, bio);
627 return true;
628 }
629
630 sector = bio->bi_iter.bi_sector;
631 chunk_sects = mddev->chunk_sectors;
632
633 sectors = chunk_sects -
634 (likely(is_power_of_2(chunk_sects))
635 ? (sector & (chunk_sects-1))
636 : sector_div(sector, chunk_sects));
637
638 if (sectors < bio_sectors(bio)) {
639 struct bio *split = bio_split(bio, sectors, GFP_NOIO,
640 &mddev->bio_set);
641 bio_chain(split, bio);
642 raid0_map_submit_bio(mddev, bio);
643 bio = split;
644 }
645
646 raid0_map_submit_bio(mddev, bio);
647 return true;
648 }
649
raid0_status(struct seq_file * seq,struct mddev * mddev)650 static void raid0_status(struct seq_file *seq, struct mddev *mddev)
651 {
652 seq_printf(seq, " %dk chunks", mddev->chunk_sectors / 2);
653 return;
654 }
655
raid0_error(struct mddev * mddev,struct md_rdev * rdev)656 static void raid0_error(struct mddev *mddev, struct md_rdev *rdev)
657 {
658 if (!test_and_set_bit(MD_BROKEN, &mddev->flags)) {
659 char *md_name = mdname(mddev);
660
661 pr_crit("md/raid0%s: Disk failure on %pg detected, failing array.\n",
662 md_name, rdev->bdev);
663 }
664 }
665
raid0_takeover_raid45(struct mddev * mddev)666 static void *raid0_takeover_raid45(struct mddev *mddev)
667 {
668 struct md_rdev *rdev;
669 struct r0conf *priv_conf;
670
671 if (mddev->degraded != 1) {
672 pr_warn("md/raid0:%s: raid5 must be degraded! Degraded disks: %d\n",
673 mdname(mddev),
674 mddev->degraded);
675 return ERR_PTR(-EINVAL);
676 }
677
678 rdev_for_each(rdev, mddev) {
679 /* check slot number for a disk */
680 if (rdev->raid_disk == mddev->raid_disks-1) {
681 pr_warn("md/raid0:%s: raid5 must have missing parity disk!\n",
682 mdname(mddev));
683 return ERR_PTR(-EINVAL);
684 }
685 rdev->sectors = mddev->dev_sectors;
686 }
687
688 /* Set new parameters */
689 mddev->new_level = 0;
690 mddev->new_layout = 0;
691 mddev->new_chunk_sectors = mddev->chunk_sectors;
692 mddev->raid_disks--;
693 mddev->delta_disks = -1;
694 /* make sure it will be not marked as dirty */
695 mddev->recovery_cp = MaxSector;
696 mddev_clear_unsupported_flags(mddev, UNSUPPORTED_MDDEV_FLAGS);
697
698 create_strip_zones(mddev, &priv_conf);
699
700 return priv_conf;
701 }
702
raid0_takeover_raid10(struct mddev * mddev)703 static void *raid0_takeover_raid10(struct mddev *mddev)
704 {
705 struct r0conf *priv_conf;
706
707 /* Check layout:
708 * - far_copies must be 1
709 * - near_copies must be 2
710 * - disks number must be even
711 * - all mirrors must be already degraded
712 */
713 if (mddev->layout != ((1 << 8) + 2)) {
714 pr_warn("md/raid0:%s:: Raid0 cannot takeover layout: 0x%x\n",
715 mdname(mddev),
716 mddev->layout);
717 return ERR_PTR(-EINVAL);
718 }
719 if (mddev->raid_disks & 1) {
720 pr_warn("md/raid0:%s: Raid0 cannot takeover Raid10 with odd disk number.\n",
721 mdname(mddev));
722 return ERR_PTR(-EINVAL);
723 }
724 if (mddev->degraded != (mddev->raid_disks>>1)) {
725 pr_warn("md/raid0:%s: All mirrors must be already degraded!\n",
726 mdname(mddev));
727 return ERR_PTR(-EINVAL);
728 }
729
730 /* Set new parameters */
731 mddev->new_level = 0;
732 mddev->new_layout = 0;
733 mddev->new_chunk_sectors = mddev->chunk_sectors;
734 mddev->delta_disks = - mddev->raid_disks / 2;
735 mddev->raid_disks += mddev->delta_disks;
736 mddev->degraded = 0;
737 /* make sure it will be not marked as dirty */
738 mddev->recovery_cp = MaxSector;
739 mddev_clear_unsupported_flags(mddev, UNSUPPORTED_MDDEV_FLAGS);
740
741 create_strip_zones(mddev, &priv_conf);
742 return priv_conf;
743 }
744
raid0_takeover_raid1(struct mddev * mddev)745 static void *raid0_takeover_raid1(struct mddev *mddev)
746 {
747 struct r0conf *priv_conf;
748 int chunksect;
749
750 /* Check layout:
751 * - (N - 1) mirror drives must be already faulty
752 */
753 if ((mddev->raid_disks - 1) != mddev->degraded) {
754 pr_err("md/raid0:%s: (N - 1) mirrors drives must be already faulty!\n",
755 mdname(mddev));
756 return ERR_PTR(-EINVAL);
757 }
758
759 /*
760 * a raid1 doesn't have the notion of chunk size, so
761 * figure out the largest suitable size we can use.
762 */
763 chunksect = 64 * 2; /* 64K by default */
764
765 /* The array must be an exact multiple of chunksize */
766 while (chunksect && (mddev->array_sectors & (chunksect - 1)))
767 chunksect >>= 1;
768
769 if ((chunksect << 9) < PAGE_SIZE)
770 /* array size does not allow a suitable chunk size */
771 return ERR_PTR(-EINVAL);
772
773 /* Set new parameters */
774 mddev->new_level = 0;
775 mddev->new_layout = 0;
776 mddev->new_chunk_sectors = chunksect;
777 mddev->chunk_sectors = chunksect;
778 mddev->delta_disks = 1 - mddev->raid_disks;
779 mddev->raid_disks = 1;
780 /* make sure it will be not marked as dirty */
781 mddev->recovery_cp = MaxSector;
782 mddev_clear_unsupported_flags(mddev, UNSUPPORTED_MDDEV_FLAGS);
783
784 create_strip_zones(mddev, &priv_conf);
785 return priv_conf;
786 }
787
raid0_takeover(struct mddev * mddev)788 static void *raid0_takeover(struct mddev *mddev)
789 {
790 /* raid0 can take over:
791 * raid4 - if all data disks are active.
792 * raid5 - providing it is Raid4 layout and one disk is faulty
793 * raid10 - assuming we have all necessary active disks
794 * raid1 - with (N -1) mirror drives faulty
795 */
796
797 if (mddev->bitmap) {
798 pr_warn("md/raid0: %s: cannot takeover array with bitmap\n",
799 mdname(mddev));
800 return ERR_PTR(-EBUSY);
801 }
802 if (mddev->level == 4)
803 return raid0_takeover_raid45(mddev);
804
805 if (mddev->level == 5) {
806 if (mddev->layout == ALGORITHM_PARITY_N)
807 return raid0_takeover_raid45(mddev);
808
809 pr_warn("md/raid0:%s: Raid can only takeover Raid5 with layout: %d\n",
810 mdname(mddev), ALGORITHM_PARITY_N);
811 }
812
813 if (mddev->level == 10)
814 return raid0_takeover_raid10(mddev);
815
816 if (mddev->level == 1)
817 return raid0_takeover_raid1(mddev);
818
819 pr_warn("Takeover from raid%i to raid0 not supported\n",
820 mddev->level);
821
822 return ERR_PTR(-EINVAL);
823 }
824
raid0_quiesce(struct mddev * mddev,int quiesce)825 static void raid0_quiesce(struct mddev *mddev, int quiesce)
826 {
827 }
828
829 static struct md_personality raid0_personality=
830 {
831 .name = "raid0",
832 .level = 0,
833 .owner = THIS_MODULE,
834 .make_request = raid0_make_request,
835 .run = raid0_run,
836 .free = raid0_free,
837 .status = raid0_status,
838 .size = raid0_size,
839 .takeover = raid0_takeover,
840 .quiesce = raid0_quiesce,
841 .error_handler = raid0_error,
842 };
843
raid0_init(void)844 static int __init raid0_init (void)
845 {
846 return register_md_personality (&raid0_personality);
847 }
848
raid0_exit(void)849 static void raid0_exit (void)
850 {
851 unregister_md_personality (&raid0_personality);
852 }
853
854 module_init(raid0_init);
855 module_exit(raid0_exit);
856 MODULE_LICENSE("GPL");
857 MODULE_DESCRIPTION("RAID0 (striping) personality for MD");
858 MODULE_ALIAS("md-personality-2"); /* RAID0 */
859 MODULE_ALIAS("md-raid0");
860 MODULE_ALIAS("md-level-0");
861