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 RAID-0 management functions.
9
10 This program is free software; you can redistribute it and/or modify
11 it under the terms of the GNU General Public License as published by
12 the Free Software Foundation; either version 2, or (at your option)
13 any later version.
14
15 You should have received a copy of the GNU General Public License
16 (for example /usr/src/linux/COPYING); if not, write to the Free
17 Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
18 */
19
20 #include <linux/blkdev.h>
21 #include <linux/seq_file.h>
22 #include <linux/module.h>
23 #include <linux/slab.h>
24 #include <trace/events/block.h>
25 #include "md.h"
26 #include "raid0.h"
27 #include "raid5.h"
28
29 static int default_layout = 0;
30 module_param(default_layout, int, 0644);
31
32 #define UNSUPPORTED_MDDEV_FLAGS \
33 ((1L << MD_HAS_JOURNAL) | \
34 (1L << MD_JOURNAL_CLEAN) | \
35 (1L << MD_FAILFAST_SUPPORTED) |\
36 (1L << MD_HAS_PPL) | \
37 (1L << MD_HAS_MULTIPLE_PPLS))
38
raid0_congested(struct mddev * mddev,int bits)39 static int raid0_congested(struct mddev *mddev, int bits)
40 {
41 struct r0conf *conf = mddev->private;
42 struct md_rdev **devlist = conf->devlist;
43 int raid_disks = conf->strip_zone[0].nb_dev;
44 int i, ret = 0;
45
46 for (i = 0; i < raid_disks && !ret ; i++) {
47 struct request_queue *q = bdev_get_queue(devlist[i]->bdev);
48
49 ret |= bdi_congested(q->backing_dev_info, bits);
50 }
51 return ret;
52 }
53
54 /*
55 * inform the user of the raid configuration
56 */
dump_zones(struct mddev * mddev)57 static void dump_zones(struct mddev *mddev)
58 {
59 int j, k;
60 sector_t zone_size = 0;
61 sector_t zone_start = 0;
62 char b[BDEVNAME_SIZE];
63 struct r0conf *conf = mddev->private;
64 int raid_disks = conf->strip_zone[0].nb_dev;
65 pr_debug("md: RAID0 configuration for %s - %d zone%s\n",
66 mdname(mddev),
67 conf->nr_strip_zones, conf->nr_strip_zones==1?"":"s");
68 for (j = 0; j < conf->nr_strip_zones; j++) {
69 char line[200];
70 int len = 0;
71
72 for (k = 0; k < conf->strip_zone[j].nb_dev; k++)
73 len += snprintf(line+len, 200-len, "%s%s", k?"/":"",
74 bdevname(conf->devlist[j*raid_disks
75 + k]->bdev, b));
76 pr_debug("md: zone%d=[%s]\n", j, line);
77
78 zone_size = conf->strip_zone[j].zone_end - zone_start;
79 pr_debug(" zone-offset=%10lluKB, device-offset=%10lluKB, size=%10lluKB\n",
80 (unsigned long long)zone_start>>1,
81 (unsigned long long)conf->strip_zone[j].dev_start>>1,
82 (unsigned long long)zone_size>>1);
83 zone_start = conf->strip_zone[j].zone_end;
84 }
85 }
86
create_strip_zones(struct mddev * mddev,struct r0conf ** private_conf)87 static int create_strip_zones(struct mddev *mddev, struct r0conf **private_conf)
88 {
89 int i, c, err;
90 sector_t curr_zone_end, sectors;
91 struct md_rdev *smallest, *rdev1, *rdev2, *rdev, **dev;
92 struct strip_zone *zone;
93 int cnt;
94 char b[BDEVNAME_SIZE];
95 char b2[BDEVNAME_SIZE];
96 struct r0conf *conf = kzalloc(sizeof(*conf), GFP_KERNEL);
97 unsigned blksize = 512;
98
99 *private_conf = ERR_PTR(-ENOMEM);
100 if (!conf)
101 return -ENOMEM;
102 rdev_for_each(rdev1, mddev) {
103 pr_debug("md/raid0:%s: looking at %s\n",
104 mdname(mddev),
105 bdevname(rdev1->bdev, b));
106 c = 0;
107
108 /* round size to chunk_size */
109 sectors = rdev1->sectors;
110 sector_div(sectors, mddev->chunk_sectors);
111 rdev1->sectors = sectors * mddev->chunk_sectors;
112
113 blksize = max(blksize, queue_logical_block_size(
114 rdev1->bdev->bd_disk->queue));
115
116 rdev_for_each(rdev2, mddev) {
117 pr_debug("md/raid0:%s: comparing %s(%llu)"
118 " with %s(%llu)\n",
119 mdname(mddev),
120 bdevname(rdev1->bdev,b),
121 (unsigned long long)rdev1->sectors,
122 bdevname(rdev2->bdev,b2),
123 (unsigned long long)rdev2->sectors);
124 if (rdev2 == rdev1) {
125 pr_debug("md/raid0:%s: END\n",
126 mdname(mddev));
127 break;
128 }
129 if (rdev2->sectors == rdev1->sectors) {
130 /*
131 * Not unique, don't count it as a new
132 * group
133 */
134 pr_debug("md/raid0:%s: EQUAL\n",
135 mdname(mddev));
136 c = 1;
137 break;
138 }
139 pr_debug("md/raid0:%s: NOT EQUAL\n",
140 mdname(mddev));
141 }
142 if (!c) {
143 pr_debug("md/raid0:%s: ==> UNIQUE\n",
144 mdname(mddev));
145 conf->nr_strip_zones++;
146 pr_debug("md/raid0:%s: %d zones\n",
147 mdname(mddev), conf->nr_strip_zones);
148 }
149 }
150 pr_debug("md/raid0:%s: FINAL %d zones\n",
151 mdname(mddev), conf->nr_strip_zones);
152
153 if (conf->nr_strip_zones == 1) {
154 conf->layout = RAID0_ORIG_LAYOUT;
155 } else if (mddev->layout == RAID0_ORIG_LAYOUT ||
156 mddev->layout == RAID0_ALT_MULTIZONE_LAYOUT) {
157 conf->layout = mddev->layout;
158 } else if (default_layout == RAID0_ORIG_LAYOUT ||
159 default_layout == RAID0_ALT_MULTIZONE_LAYOUT) {
160 conf->layout = default_layout;
161 } else {
162 pr_err("md/raid0:%s: cannot assemble multi-zone RAID0 with default_layout setting\n",
163 mdname(mddev));
164 pr_err("md/raid0: please set raid0.default_layout to 1 or 2\n");
165 err = -ENOTSUPP;
166 goto abort;
167 }
168 /*
169 * now since we have the hard sector sizes, we can make sure
170 * chunk size is a multiple of that sector size
171 */
172 if ((mddev->chunk_sectors << 9) % blksize) {
173 pr_warn("md/raid0:%s: chunk_size of %d not multiple of block size %d\n",
174 mdname(mddev),
175 mddev->chunk_sectors << 9, blksize);
176 err = -EINVAL;
177 goto abort;
178 }
179
180 err = -ENOMEM;
181 conf->strip_zone = kcalloc(conf->nr_strip_zones,
182 sizeof(struct strip_zone),
183 GFP_KERNEL);
184 if (!conf->strip_zone)
185 goto abort;
186 conf->devlist = kzalloc(array3_size(sizeof(struct md_rdev *),
187 conf->nr_strip_zones,
188 mddev->raid_disks),
189 GFP_KERNEL);
190 if (!conf->devlist)
191 goto abort;
192
193 /* The first zone must contain all devices, so here we check that
194 * there is a proper alignment of slots to devices and find them all
195 */
196 zone = &conf->strip_zone[0];
197 cnt = 0;
198 smallest = NULL;
199 dev = conf->devlist;
200 err = -EINVAL;
201 rdev_for_each(rdev1, mddev) {
202 int j = rdev1->raid_disk;
203
204 if (mddev->level == 10) {
205 /* taking over a raid10-n2 array */
206 j /= 2;
207 rdev1->new_raid_disk = j;
208 }
209
210 if (mddev->level == 1) {
211 /* taiking over a raid1 array-
212 * we have only one active disk
213 */
214 j = 0;
215 rdev1->new_raid_disk = j;
216 }
217
218 if (j < 0) {
219 pr_warn("md/raid0:%s: remove inactive devices before converting to RAID0\n",
220 mdname(mddev));
221 goto abort;
222 }
223 if (j >= mddev->raid_disks) {
224 pr_warn("md/raid0:%s: bad disk number %d - aborting!\n",
225 mdname(mddev), j);
226 goto abort;
227 }
228 if (dev[j]) {
229 pr_warn("md/raid0:%s: multiple devices for %d - aborting!\n",
230 mdname(mddev), j);
231 goto abort;
232 }
233 dev[j] = rdev1;
234
235 if (!smallest || (rdev1->sectors < smallest->sectors))
236 smallest = rdev1;
237 cnt++;
238 }
239 if (cnt != mddev->raid_disks) {
240 pr_warn("md/raid0:%s: too few disks (%d of %d) - aborting!\n",
241 mdname(mddev), cnt, mddev->raid_disks);
242 goto abort;
243 }
244 zone->nb_dev = cnt;
245 zone->zone_end = smallest->sectors * cnt;
246
247 curr_zone_end = zone->zone_end;
248
249 /* now do the other zones */
250 for (i = 1; i < conf->nr_strip_zones; i++)
251 {
252 int j;
253
254 zone = conf->strip_zone + i;
255 dev = conf->devlist + i * mddev->raid_disks;
256
257 pr_debug("md/raid0:%s: zone %d\n", mdname(mddev), i);
258 zone->dev_start = smallest->sectors;
259 smallest = NULL;
260 c = 0;
261
262 for (j=0; j<cnt; j++) {
263 rdev = conf->devlist[j];
264 if (rdev->sectors <= zone->dev_start) {
265 pr_debug("md/raid0:%s: checking %s ... nope\n",
266 mdname(mddev),
267 bdevname(rdev->bdev, b));
268 continue;
269 }
270 pr_debug("md/raid0:%s: checking %s ..."
271 " contained as device %d\n",
272 mdname(mddev),
273 bdevname(rdev->bdev, b), c);
274 dev[c] = rdev;
275 c++;
276 if (!smallest || rdev->sectors < smallest->sectors) {
277 smallest = rdev;
278 pr_debug("md/raid0:%s: (%llu) is smallest!.\n",
279 mdname(mddev),
280 (unsigned long long)rdev->sectors);
281 }
282 }
283
284 zone->nb_dev = c;
285 sectors = (smallest->sectors - zone->dev_start) * c;
286 pr_debug("md/raid0:%s: zone->nb_dev: %d, sectors: %llu\n",
287 mdname(mddev),
288 zone->nb_dev, (unsigned long long)sectors);
289
290 curr_zone_end += sectors;
291 zone->zone_end = curr_zone_end;
292
293 pr_debug("md/raid0:%s: current zone start: %llu\n",
294 mdname(mddev),
295 (unsigned long long)smallest->sectors);
296 }
297
298 pr_debug("md/raid0:%s: done.\n", mdname(mddev));
299 *private_conf = conf;
300
301 return 0;
302 abort:
303 kfree(conf->strip_zone);
304 kfree(conf->devlist);
305 kfree(conf);
306 *private_conf = ERR_PTR(err);
307 return err;
308 }
309
310 /* Find the zone which holds a particular offset
311 * Update *sectorp to be an offset in that zone
312 */
find_zone(struct r0conf * conf,sector_t * sectorp)313 static struct strip_zone *find_zone(struct r0conf *conf,
314 sector_t *sectorp)
315 {
316 int i;
317 struct strip_zone *z = conf->strip_zone;
318 sector_t sector = *sectorp;
319
320 for (i = 0; i < conf->nr_strip_zones; i++)
321 if (sector < z[i].zone_end) {
322 if (i)
323 *sectorp = sector - z[i-1].zone_end;
324 return z + i;
325 }
326 BUG();
327 }
328
329 /*
330 * remaps the bio to the target device. we separate two flows.
331 * power 2 flow and a general flow for the sake of performance
332 */
map_sector(struct mddev * mddev,struct strip_zone * zone,sector_t sector,sector_t * sector_offset)333 static struct md_rdev *map_sector(struct mddev *mddev, struct strip_zone *zone,
334 sector_t sector, sector_t *sector_offset)
335 {
336 unsigned int sect_in_chunk;
337 sector_t chunk;
338 struct r0conf *conf = mddev->private;
339 int raid_disks = conf->strip_zone[0].nb_dev;
340 unsigned int chunk_sects = mddev->chunk_sectors;
341
342 if (is_power_of_2(chunk_sects)) {
343 int chunksect_bits = ffz(~chunk_sects);
344 /* find the sector offset inside the chunk */
345 sect_in_chunk = sector & (chunk_sects - 1);
346 sector >>= chunksect_bits;
347 /* chunk in zone */
348 chunk = *sector_offset;
349 /* quotient is the chunk in real device*/
350 sector_div(chunk, zone->nb_dev << chunksect_bits);
351 } else{
352 sect_in_chunk = sector_div(sector, chunk_sects);
353 chunk = *sector_offset;
354 sector_div(chunk, chunk_sects * zone->nb_dev);
355 }
356 /*
357 * position the bio over the real device
358 * real sector = chunk in device + starting of zone
359 * + the position in the chunk
360 */
361 *sector_offset = (chunk * chunk_sects) + sect_in_chunk;
362 return conf->devlist[(zone - conf->strip_zone)*raid_disks
363 + sector_div(sector, zone->nb_dev)];
364 }
365
raid0_size(struct mddev * mddev,sector_t sectors,int raid_disks)366 static sector_t raid0_size(struct mddev *mddev, sector_t sectors, int raid_disks)
367 {
368 sector_t array_sectors = 0;
369 struct md_rdev *rdev;
370
371 WARN_ONCE(sectors || raid_disks,
372 "%s does not support generic reshape\n", __func__);
373
374 rdev_for_each(rdev, mddev)
375 array_sectors += (rdev->sectors &
376 ~(sector_t)(mddev->chunk_sectors-1));
377
378 return array_sectors;
379 }
380
381 static void raid0_free(struct mddev *mddev, void *priv);
382
raid0_run(struct mddev * mddev)383 static int raid0_run(struct mddev *mddev)
384 {
385 struct r0conf *conf;
386 int ret;
387
388 if (mddev->chunk_sectors == 0) {
389 pr_warn("md/raid0:%s: chunk size must be set.\n", mdname(mddev));
390 return -EINVAL;
391 }
392 if (md_check_no_bitmap(mddev))
393 return -EINVAL;
394
395 /* if private is not null, we are here after takeover */
396 if (mddev->private == NULL) {
397 ret = create_strip_zones(mddev, &conf);
398 if (ret < 0)
399 return ret;
400 mddev->private = conf;
401 }
402 conf = mddev->private;
403 if (mddev->queue) {
404 struct md_rdev *rdev;
405 bool discard_supported = false;
406
407 blk_queue_max_hw_sectors(mddev->queue, mddev->chunk_sectors);
408 blk_queue_max_write_same_sectors(mddev->queue, mddev->chunk_sectors);
409 blk_queue_max_write_zeroes_sectors(mddev->queue, mddev->chunk_sectors);
410 blk_queue_max_discard_sectors(mddev->queue, UINT_MAX);
411
412 blk_queue_io_min(mddev->queue, mddev->chunk_sectors << 9);
413 blk_queue_io_opt(mddev->queue,
414 (mddev->chunk_sectors << 9) * mddev->raid_disks);
415
416 rdev_for_each(rdev, mddev) {
417 disk_stack_limits(mddev->gendisk, rdev->bdev,
418 rdev->data_offset << 9);
419 if (blk_queue_discard(bdev_get_queue(rdev->bdev)))
420 discard_supported = true;
421 }
422 if (!discard_supported)
423 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, mddev->queue);
424 else
425 blk_queue_flag_set(QUEUE_FLAG_DISCARD, mddev->queue);
426 }
427
428 /* calculate array device size */
429 md_set_array_sectors(mddev, raid0_size(mddev, 0, 0));
430
431 pr_debug("md/raid0:%s: md_size is %llu sectors.\n",
432 mdname(mddev),
433 (unsigned long long)mddev->array_sectors);
434
435 if (mddev->queue) {
436 /* calculate the max read-ahead size.
437 * For read-ahead of large files to be effective, we need to
438 * readahead at least twice a whole stripe. i.e. number of devices
439 * multiplied by chunk size times 2.
440 * If an individual device has an ra_pages greater than the
441 * chunk size, then we will not drive that device as hard as it
442 * wants. We consider this a configuration error: a larger
443 * chunksize should be used in that case.
444 */
445 int stripe = mddev->raid_disks *
446 (mddev->chunk_sectors << 9) / PAGE_SIZE;
447 if (mddev->queue->backing_dev_info->ra_pages < 2* stripe)
448 mddev->queue->backing_dev_info->ra_pages = 2* stripe;
449 }
450
451 dump_zones(mddev);
452
453 ret = md_integrity_register(mddev);
454
455 return ret;
456 }
457
raid0_free(struct mddev * mddev,void * priv)458 static void raid0_free(struct mddev *mddev, void *priv)
459 {
460 struct r0conf *conf = priv;
461
462 kfree(conf->strip_zone);
463 kfree(conf->devlist);
464 kfree(conf);
465 }
466
467 /*
468 * Is io distribute over 1 or more chunks ?
469 */
is_io_in_chunk_boundary(struct mddev * mddev,unsigned int chunk_sects,struct bio * bio)470 static inline int is_io_in_chunk_boundary(struct mddev *mddev,
471 unsigned int chunk_sects, struct bio *bio)
472 {
473 if (likely(is_power_of_2(chunk_sects))) {
474 return chunk_sects >=
475 ((bio->bi_iter.bi_sector & (chunk_sects-1))
476 + bio_sectors(bio));
477 } else{
478 sector_t sector = bio->bi_iter.bi_sector;
479 return chunk_sects >= (sector_div(sector, chunk_sects)
480 + bio_sectors(bio));
481 }
482 }
483
raid0_handle_discard(struct mddev * mddev,struct bio * bio)484 static void raid0_handle_discard(struct mddev *mddev, struct bio *bio)
485 {
486 struct r0conf *conf = mddev->private;
487 struct strip_zone *zone;
488 sector_t start = bio->bi_iter.bi_sector;
489 sector_t end;
490 unsigned int stripe_size;
491 sector_t first_stripe_index, last_stripe_index;
492 sector_t start_disk_offset;
493 unsigned int start_disk_index;
494 sector_t end_disk_offset;
495 unsigned int end_disk_index;
496 unsigned int disk;
497
498 zone = find_zone(conf, &start);
499
500 if (bio_end_sector(bio) > zone->zone_end) {
501 struct bio *split = bio_split(bio,
502 zone->zone_end - bio->bi_iter.bi_sector, GFP_NOIO,
503 &mddev->bio_set);
504 bio_chain(split, bio);
505 generic_make_request(bio);
506 bio = split;
507 end = zone->zone_end;
508 } else
509 end = bio_end_sector(bio);
510
511 if (zone != conf->strip_zone)
512 end = end - zone[-1].zone_end;
513
514 /* Now start and end is the offset in zone */
515 stripe_size = zone->nb_dev * mddev->chunk_sectors;
516
517 first_stripe_index = start;
518 sector_div(first_stripe_index, stripe_size);
519 last_stripe_index = end;
520 sector_div(last_stripe_index, stripe_size);
521
522 start_disk_index = (int)(start - first_stripe_index * stripe_size) /
523 mddev->chunk_sectors;
524 start_disk_offset = ((int)(start - first_stripe_index * stripe_size) %
525 mddev->chunk_sectors) +
526 first_stripe_index * mddev->chunk_sectors;
527 end_disk_index = (int)(end - last_stripe_index * stripe_size) /
528 mddev->chunk_sectors;
529 end_disk_offset = ((int)(end - last_stripe_index * stripe_size) %
530 mddev->chunk_sectors) +
531 last_stripe_index * mddev->chunk_sectors;
532
533 for (disk = 0; disk < zone->nb_dev; disk++) {
534 sector_t dev_start, dev_end;
535 struct bio *discard_bio = NULL;
536 struct md_rdev *rdev;
537
538 if (disk < start_disk_index)
539 dev_start = (first_stripe_index + 1) *
540 mddev->chunk_sectors;
541 else if (disk > start_disk_index)
542 dev_start = first_stripe_index * mddev->chunk_sectors;
543 else
544 dev_start = start_disk_offset;
545
546 if (disk < end_disk_index)
547 dev_end = (last_stripe_index + 1) * mddev->chunk_sectors;
548 else if (disk > end_disk_index)
549 dev_end = last_stripe_index * mddev->chunk_sectors;
550 else
551 dev_end = end_disk_offset;
552
553 if (dev_end <= dev_start)
554 continue;
555
556 rdev = conf->devlist[(zone - conf->strip_zone) *
557 conf->strip_zone[0].nb_dev + disk];
558 if (__blkdev_issue_discard(rdev->bdev,
559 dev_start + zone->dev_start + rdev->data_offset,
560 dev_end - dev_start, GFP_NOIO, 0, &discard_bio) ||
561 !discard_bio)
562 continue;
563 bio_chain(discard_bio, bio);
564 bio_clone_blkcg_association(discard_bio, bio);
565 if (mddev->gendisk)
566 trace_block_bio_remap(bdev_get_queue(rdev->bdev),
567 discard_bio, disk_devt(mddev->gendisk),
568 bio->bi_iter.bi_sector);
569 bio_clear_flag(bio, BIO_QUEUE_ENTERED);
570 generic_make_request(discard_bio);
571 }
572 bio_endio(bio);
573 }
574
raid0_make_request(struct mddev * mddev,struct bio * bio)575 static bool raid0_make_request(struct mddev *mddev, struct bio *bio)
576 {
577 struct r0conf *conf = mddev->private;
578 struct strip_zone *zone;
579 struct md_rdev *tmp_dev;
580 sector_t bio_sector;
581 sector_t sector;
582 sector_t orig_sector;
583 unsigned chunk_sects;
584 unsigned sectors;
585
586 if (unlikely(bio->bi_opf & REQ_PREFLUSH)
587 && md_flush_request(mddev, bio))
588 return true;
589
590 if (unlikely((bio_op(bio) == REQ_OP_DISCARD))) {
591 raid0_handle_discard(mddev, bio);
592 return true;
593 }
594
595 bio_sector = bio->bi_iter.bi_sector;
596 sector = bio_sector;
597 chunk_sects = mddev->chunk_sectors;
598
599 sectors = chunk_sects -
600 (likely(is_power_of_2(chunk_sects))
601 ? (sector & (chunk_sects-1))
602 : sector_div(sector, chunk_sects));
603
604 /* Restore due to sector_div */
605 sector = bio_sector;
606
607 if (sectors < bio_sectors(bio)) {
608 struct bio *split = bio_split(bio, sectors, GFP_NOIO,
609 &mddev->bio_set);
610 bio_chain(split, bio);
611 generic_make_request(bio);
612 bio = split;
613 }
614
615 orig_sector = sector;
616 zone = find_zone(mddev->private, §or);
617 switch (conf->layout) {
618 case RAID0_ORIG_LAYOUT:
619 tmp_dev = map_sector(mddev, zone, orig_sector, §or);
620 break;
621 case RAID0_ALT_MULTIZONE_LAYOUT:
622 tmp_dev = map_sector(mddev, zone, sector, §or);
623 break;
624 default:
625 WARN(1, "md/raid0:%s: Invalid layout\n", mdname(mddev));
626 bio_io_error(bio);
627 return true;
628 }
629
630 bio_set_dev(bio, tmp_dev->bdev);
631 bio->bi_iter.bi_sector = sector + zone->dev_start +
632 tmp_dev->data_offset;
633
634 if (mddev->gendisk)
635 trace_block_bio_remap(bio->bi_disk->queue, bio,
636 disk_devt(mddev->gendisk), bio_sector);
637 mddev_check_writesame(mddev, bio);
638 mddev_check_write_zeroes(mddev, bio);
639 bio_clear_flag(bio, BIO_QUEUE_ENTERED);
640 generic_make_request(bio);
641 return true;
642 }
643
raid0_status(struct seq_file * seq,struct mddev * mddev)644 static void raid0_status(struct seq_file *seq, struct mddev *mddev)
645 {
646 seq_printf(seq, " %dk chunks", mddev->chunk_sectors / 2);
647 return;
648 }
649
raid0_takeover_raid45(struct mddev * mddev)650 static void *raid0_takeover_raid45(struct mddev *mddev)
651 {
652 struct md_rdev *rdev;
653 struct r0conf *priv_conf;
654
655 if (mddev->degraded != 1) {
656 pr_warn("md/raid0:%s: raid5 must be degraded! Degraded disks: %d\n",
657 mdname(mddev),
658 mddev->degraded);
659 return ERR_PTR(-EINVAL);
660 }
661
662 rdev_for_each(rdev, mddev) {
663 /* check slot number for a disk */
664 if (rdev->raid_disk == mddev->raid_disks-1) {
665 pr_warn("md/raid0:%s: raid5 must have missing parity disk!\n",
666 mdname(mddev));
667 return ERR_PTR(-EINVAL);
668 }
669 rdev->sectors = mddev->dev_sectors;
670 }
671
672 /* Set new parameters */
673 mddev->new_level = 0;
674 mddev->new_layout = 0;
675 mddev->new_chunk_sectors = mddev->chunk_sectors;
676 mddev->raid_disks--;
677 mddev->delta_disks = -1;
678 /* make sure it will be not marked as dirty */
679 mddev->recovery_cp = MaxSector;
680 mddev_clear_unsupported_flags(mddev, UNSUPPORTED_MDDEV_FLAGS);
681
682 create_strip_zones(mddev, &priv_conf);
683
684 return priv_conf;
685 }
686
raid0_takeover_raid10(struct mddev * mddev)687 static void *raid0_takeover_raid10(struct mddev *mddev)
688 {
689 struct r0conf *priv_conf;
690
691 /* Check layout:
692 * - far_copies must be 1
693 * - near_copies must be 2
694 * - disks number must be even
695 * - all mirrors must be already degraded
696 */
697 if (mddev->layout != ((1 << 8) + 2)) {
698 pr_warn("md/raid0:%s:: Raid0 cannot takeover layout: 0x%x\n",
699 mdname(mddev),
700 mddev->layout);
701 return ERR_PTR(-EINVAL);
702 }
703 if (mddev->raid_disks & 1) {
704 pr_warn("md/raid0:%s: Raid0 cannot takeover Raid10 with odd disk number.\n",
705 mdname(mddev));
706 return ERR_PTR(-EINVAL);
707 }
708 if (mddev->degraded != (mddev->raid_disks>>1)) {
709 pr_warn("md/raid0:%s: All mirrors must be already degraded!\n",
710 mdname(mddev));
711 return ERR_PTR(-EINVAL);
712 }
713
714 /* Set new parameters */
715 mddev->new_level = 0;
716 mddev->new_layout = 0;
717 mddev->new_chunk_sectors = mddev->chunk_sectors;
718 mddev->delta_disks = - mddev->raid_disks / 2;
719 mddev->raid_disks += mddev->delta_disks;
720 mddev->degraded = 0;
721 /* make sure it will be not marked as dirty */
722 mddev->recovery_cp = MaxSector;
723 mddev_clear_unsupported_flags(mddev, UNSUPPORTED_MDDEV_FLAGS);
724
725 create_strip_zones(mddev, &priv_conf);
726 return priv_conf;
727 }
728
raid0_takeover_raid1(struct mddev * mddev)729 static void *raid0_takeover_raid1(struct mddev *mddev)
730 {
731 struct r0conf *priv_conf;
732 int chunksect;
733
734 /* Check layout:
735 * - (N - 1) mirror drives must be already faulty
736 */
737 if ((mddev->raid_disks - 1) != mddev->degraded) {
738 pr_err("md/raid0:%s: (N - 1) mirrors drives must be already faulty!\n",
739 mdname(mddev));
740 return ERR_PTR(-EINVAL);
741 }
742
743 /*
744 * a raid1 doesn't have the notion of chunk size, so
745 * figure out the largest suitable size we can use.
746 */
747 chunksect = 64 * 2; /* 64K by default */
748
749 /* The array must be an exact multiple of chunksize */
750 while (chunksect && (mddev->array_sectors & (chunksect - 1)))
751 chunksect >>= 1;
752
753 if ((chunksect << 9) < PAGE_SIZE)
754 /* array size does not allow a suitable chunk size */
755 return ERR_PTR(-EINVAL);
756
757 /* Set new parameters */
758 mddev->new_level = 0;
759 mddev->new_layout = 0;
760 mddev->new_chunk_sectors = chunksect;
761 mddev->chunk_sectors = chunksect;
762 mddev->delta_disks = 1 - mddev->raid_disks;
763 mddev->raid_disks = 1;
764 /* make sure it will be not marked as dirty */
765 mddev->recovery_cp = MaxSector;
766 mddev_clear_unsupported_flags(mddev, UNSUPPORTED_MDDEV_FLAGS);
767
768 create_strip_zones(mddev, &priv_conf);
769 return priv_conf;
770 }
771
raid0_takeover(struct mddev * mddev)772 static void *raid0_takeover(struct mddev *mddev)
773 {
774 /* raid0 can take over:
775 * raid4 - if all data disks are active.
776 * raid5 - providing it is Raid4 layout and one disk is faulty
777 * raid10 - assuming we have all necessary active disks
778 * raid1 - with (N -1) mirror drives faulty
779 */
780
781 if (mddev->bitmap) {
782 pr_warn("md/raid0: %s: cannot takeover array with bitmap\n",
783 mdname(mddev));
784 return ERR_PTR(-EBUSY);
785 }
786 if (mddev->level == 4)
787 return raid0_takeover_raid45(mddev);
788
789 if (mddev->level == 5) {
790 if (mddev->layout == ALGORITHM_PARITY_N)
791 return raid0_takeover_raid45(mddev);
792
793 pr_warn("md/raid0:%s: Raid can only takeover Raid5 with layout: %d\n",
794 mdname(mddev), ALGORITHM_PARITY_N);
795 }
796
797 if (mddev->level == 10)
798 return raid0_takeover_raid10(mddev);
799
800 if (mddev->level == 1)
801 return raid0_takeover_raid1(mddev);
802
803 pr_warn("Takeover from raid%i to raid0 not supported\n",
804 mddev->level);
805
806 return ERR_PTR(-EINVAL);
807 }
808
raid0_quiesce(struct mddev * mddev,int quiesce)809 static void raid0_quiesce(struct mddev *mddev, int quiesce)
810 {
811 }
812
813 static struct md_personality raid0_personality=
814 {
815 .name = "raid0",
816 .level = 0,
817 .owner = THIS_MODULE,
818 .make_request = raid0_make_request,
819 .run = raid0_run,
820 .free = raid0_free,
821 .status = raid0_status,
822 .size = raid0_size,
823 .takeover = raid0_takeover,
824 .quiesce = raid0_quiesce,
825 .congested = raid0_congested,
826 };
827
raid0_init(void)828 static int __init raid0_init (void)
829 {
830 return register_md_personality (&raid0_personality);
831 }
832
raid0_exit(void)833 static void raid0_exit (void)
834 {
835 unregister_md_personality (&raid0_personality);
836 }
837
838 module_init(raid0_init);
839 module_exit(raid0_exit);
840 MODULE_LICENSE("GPL");
841 MODULE_DESCRIPTION("RAID0 (striping) personality for MD");
842 MODULE_ALIAS("md-personality-2"); /* RAID0 */
843 MODULE_ALIAS("md-raid0");
844 MODULE_ALIAS("md-level-0");
845