Loading...
1// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 raid0.c : Multiple Devices driver for Linux
4 Copyright (C) 1994-96 Marc ZYNGIER
5 <zyngier@ufr-info-p7.ibp.fr> or
6 <maz@gloups.fdn.fr>
7 Copyright (C) 1999, 2000 Ingo Molnar, Red Hat
8
9 RAID-0 management functions.
10
11*/
12
13#include <linux/blkdev.h>
14#include <linux/seq_file.h>
15#include <linux/module.h>
16#include <linux/slab.h>
17#include <trace/events/block.h>
18#include "md.h"
19#include "raid0.h"
20#include "raid5.h"
21
22static int default_layout = 0;
23module_param(default_layout, int, 0644);
24
25#define UNSUPPORTED_MDDEV_FLAGS \
26 ((1L << MD_HAS_JOURNAL) | \
27 (1L << MD_JOURNAL_CLEAN) | \
28 (1L << MD_FAILFAST_SUPPORTED) |\
29 (1L << MD_HAS_PPL) | \
30 (1L << MD_HAS_MULTIPLE_PPLS))
31
32/*
33 * inform the user of the raid configuration
34*/
35static void dump_zones(struct mddev *mddev)
36{
37 int j, k;
38 sector_t zone_size = 0;
39 sector_t zone_start = 0;
40 char b[BDEVNAME_SIZE];
41 struct r0conf *conf = mddev->private;
42 int raid_disks = conf->strip_zone[0].nb_dev;
43 pr_debug("md: RAID0 configuration for %s - %d zone%s\n",
44 mdname(mddev),
45 conf->nr_strip_zones, conf->nr_strip_zones==1?"":"s");
46 for (j = 0; j < conf->nr_strip_zones; j++) {
47 char line[200];
48 int len = 0;
49
50 for (k = 0; k < conf->strip_zone[j].nb_dev; k++)
51 len += snprintf(line+len, 200-len, "%s%s", k?"/":"",
52 bdevname(conf->devlist[j*raid_disks
53 + k]->bdev, b));
54 pr_debug("md: zone%d=[%s]\n", j, line);
55
56 zone_size = conf->strip_zone[j].zone_end - zone_start;
57 pr_debug(" zone-offset=%10lluKB, device-offset=%10lluKB, size=%10lluKB\n",
58 (unsigned long long)zone_start>>1,
59 (unsigned long long)conf->strip_zone[j].dev_start>>1,
60 (unsigned long long)zone_size>>1);
61 zone_start = conf->strip_zone[j].zone_end;
62 }
63}
64
65static int create_strip_zones(struct mddev *mddev, struct r0conf **private_conf)
66{
67 int i, c, err;
68 sector_t curr_zone_end, sectors;
69 struct md_rdev *smallest, *rdev1, *rdev2, *rdev, **dev;
70 struct strip_zone *zone;
71 int cnt;
72 char b[BDEVNAME_SIZE];
73 char b2[BDEVNAME_SIZE];
74 struct r0conf *conf = kzalloc(sizeof(*conf), GFP_KERNEL);
75 unsigned blksize = 512;
76
77 *private_conf = ERR_PTR(-ENOMEM);
78 if (!conf)
79 return -ENOMEM;
80 rdev_for_each(rdev1, mddev) {
81 pr_debug("md/raid0:%s: looking at %s\n",
82 mdname(mddev),
83 bdevname(rdev1->bdev, b));
84 c = 0;
85
86 /* round size to chunk_size */
87 sectors = rdev1->sectors;
88 sector_div(sectors, mddev->chunk_sectors);
89 rdev1->sectors = sectors * mddev->chunk_sectors;
90
91 blksize = max(blksize, queue_logical_block_size(
92 rdev1->bdev->bd_disk->queue));
93
94 rdev_for_each(rdev2, mddev) {
95 pr_debug("md/raid0:%s: comparing %s(%llu)"
96 " with %s(%llu)\n",
97 mdname(mddev),
98 bdevname(rdev1->bdev,b),
99 (unsigned long long)rdev1->sectors,
100 bdevname(rdev2->bdev,b2),
101 (unsigned long long)rdev2->sectors);
102 if (rdev2 == rdev1) {
103 pr_debug("md/raid0:%s: END\n",
104 mdname(mddev));
105 break;
106 }
107 if (rdev2->sectors == rdev1->sectors) {
108 /*
109 * Not unique, don't count it as a new
110 * group
111 */
112 pr_debug("md/raid0:%s: EQUAL\n",
113 mdname(mddev));
114 c = 1;
115 break;
116 }
117 pr_debug("md/raid0:%s: NOT EQUAL\n",
118 mdname(mddev));
119 }
120 if (!c) {
121 pr_debug("md/raid0:%s: ==> UNIQUE\n",
122 mdname(mddev));
123 conf->nr_strip_zones++;
124 pr_debug("md/raid0:%s: %d zones\n",
125 mdname(mddev), conf->nr_strip_zones);
126 }
127 }
128 pr_debug("md/raid0:%s: FINAL %d zones\n",
129 mdname(mddev), conf->nr_strip_zones);
130
131 if (conf->nr_strip_zones == 1) {
132 conf->layout = RAID0_ORIG_LAYOUT;
133 } else if (mddev->layout == RAID0_ORIG_LAYOUT ||
134 mddev->layout == RAID0_ALT_MULTIZONE_LAYOUT) {
135 conf->layout = mddev->layout;
136 } else if (default_layout == RAID0_ORIG_LAYOUT ||
137 default_layout == RAID0_ALT_MULTIZONE_LAYOUT) {
138 conf->layout = default_layout;
139 } else {
140 pr_err("md/raid0:%s: cannot assemble multi-zone RAID0 with default_layout setting\n",
141 mdname(mddev));
142 pr_err("md/raid0: please set raid0.default_layout to 1 or 2\n");
143 err = -ENOTSUPP;
144 goto abort;
145 }
146 /*
147 * now since we have the hard sector sizes, we can make sure
148 * chunk size is a multiple of that sector size
149 */
150 if ((mddev->chunk_sectors << 9) % blksize) {
151 pr_warn("md/raid0:%s: chunk_size of %d not multiple of block size %d\n",
152 mdname(mddev),
153 mddev->chunk_sectors << 9, blksize);
154 err = -EINVAL;
155 goto abort;
156 }
157
158 err = -ENOMEM;
159 conf->strip_zone = kcalloc(conf->nr_strip_zones,
160 sizeof(struct strip_zone),
161 GFP_KERNEL);
162 if (!conf->strip_zone)
163 goto abort;
164 conf->devlist = kzalloc(array3_size(sizeof(struct md_rdev *),
165 conf->nr_strip_zones,
166 mddev->raid_disks),
167 GFP_KERNEL);
168 if (!conf->devlist)
169 goto abort;
170
171 /* The first zone must contain all devices, so here we check that
172 * there is a proper alignment of slots to devices and find them all
173 */
174 zone = &conf->strip_zone[0];
175 cnt = 0;
176 smallest = NULL;
177 dev = conf->devlist;
178 err = -EINVAL;
179 rdev_for_each(rdev1, mddev) {
180 int j = rdev1->raid_disk;
181
182 if (mddev->level == 10) {
183 /* taking over a raid10-n2 array */
184 j /= 2;
185 rdev1->new_raid_disk = j;
186 }
187
188 if (mddev->level == 1) {
189 /* taiking over a raid1 array-
190 * we have only one active disk
191 */
192 j = 0;
193 rdev1->new_raid_disk = j;
194 }
195
196 if (j < 0) {
197 pr_warn("md/raid0:%s: remove inactive devices before converting to RAID0\n",
198 mdname(mddev));
199 goto abort;
200 }
201 if (j >= mddev->raid_disks) {
202 pr_warn("md/raid0:%s: bad disk number %d - aborting!\n",
203 mdname(mddev), j);
204 goto abort;
205 }
206 if (dev[j]) {
207 pr_warn("md/raid0:%s: multiple devices for %d - aborting!\n",
208 mdname(mddev), j);
209 goto abort;
210 }
211 dev[j] = rdev1;
212
213 if (!smallest || (rdev1->sectors < smallest->sectors))
214 smallest = rdev1;
215 cnt++;
216 }
217 if (cnt != mddev->raid_disks) {
218 pr_warn("md/raid0:%s: too few disks (%d of %d) - aborting!\n",
219 mdname(mddev), cnt, mddev->raid_disks);
220 goto abort;
221 }
222 zone->nb_dev = cnt;
223 zone->zone_end = smallest->sectors * cnt;
224
225 curr_zone_end = zone->zone_end;
226
227 /* now do the other zones */
228 for (i = 1; i < conf->nr_strip_zones; i++)
229 {
230 int j;
231
232 zone = conf->strip_zone + i;
233 dev = conf->devlist + i * mddev->raid_disks;
234
235 pr_debug("md/raid0:%s: zone %d\n", mdname(mddev), i);
236 zone->dev_start = smallest->sectors;
237 smallest = NULL;
238 c = 0;
239
240 for (j=0; j<cnt; j++) {
241 rdev = conf->devlist[j];
242 if (rdev->sectors <= zone->dev_start) {
243 pr_debug("md/raid0:%s: checking %s ... nope\n",
244 mdname(mddev),
245 bdevname(rdev->bdev, b));
246 continue;
247 }
248 pr_debug("md/raid0:%s: checking %s ..."
249 " contained as device %d\n",
250 mdname(mddev),
251 bdevname(rdev->bdev, b), c);
252 dev[c] = rdev;
253 c++;
254 if (!smallest || rdev->sectors < smallest->sectors) {
255 smallest = rdev;
256 pr_debug("md/raid0:%s: (%llu) is smallest!.\n",
257 mdname(mddev),
258 (unsigned long long)rdev->sectors);
259 }
260 }
261
262 zone->nb_dev = c;
263 sectors = (smallest->sectors - zone->dev_start) * c;
264 pr_debug("md/raid0:%s: zone->nb_dev: %d, sectors: %llu\n",
265 mdname(mddev),
266 zone->nb_dev, (unsigned long long)sectors);
267
268 curr_zone_end += sectors;
269 zone->zone_end = curr_zone_end;
270
271 pr_debug("md/raid0:%s: current zone start: %llu\n",
272 mdname(mddev),
273 (unsigned long long)smallest->sectors);
274 }
275
276 pr_debug("md/raid0:%s: done.\n", mdname(mddev));
277 *private_conf = conf;
278
279 return 0;
280abort:
281 kfree(conf->strip_zone);
282 kfree(conf->devlist);
283 kfree(conf);
284 *private_conf = ERR_PTR(err);
285 return err;
286}
287
288/* Find the zone which holds a particular offset
289 * Update *sectorp to be an offset in that zone
290 */
291static struct strip_zone *find_zone(struct r0conf *conf,
292 sector_t *sectorp)
293{
294 int i;
295 struct strip_zone *z = conf->strip_zone;
296 sector_t sector = *sectorp;
297
298 for (i = 0; i < conf->nr_strip_zones; i++)
299 if (sector < z[i].zone_end) {
300 if (i)
301 *sectorp = sector - z[i-1].zone_end;
302 return z + i;
303 }
304 BUG();
305}
306
307/*
308 * remaps the bio to the target device. we separate two flows.
309 * power 2 flow and a general flow for the sake of performance
310*/
311static struct md_rdev *map_sector(struct mddev *mddev, struct strip_zone *zone,
312 sector_t sector, sector_t *sector_offset)
313{
314 unsigned int sect_in_chunk;
315 sector_t chunk;
316 struct r0conf *conf = mddev->private;
317 int raid_disks = conf->strip_zone[0].nb_dev;
318 unsigned int chunk_sects = mddev->chunk_sectors;
319
320 if (is_power_of_2(chunk_sects)) {
321 int chunksect_bits = ffz(~chunk_sects);
322 /* find the sector offset inside the chunk */
323 sect_in_chunk = sector & (chunk_sects - 1);
324 sector >>= chunksect_bits;
325 /* chunk in zone */
326 chunk = *sector_offset;
327 /* quotient is the chunk in real device*/
328 sector_div(chunk, zone->nb_dev << chunksect_bits);
329 } else{
330 sect_in_chunk = sector_div(sector, chunk_sects);
331 chunk = *sector_offset;
332 sector_div(chunk, chunk_sects * zone->nb_dev);
333 }
334 /*
335 * position the bio over the real device
336 * real sector = chunk in device + starting of zone
337 * + the position in the chunk
338 */
339 *sector_offset = (chunk * chunk_sects) + sect_in_chunk;
340 return conf->devlist[(zone - conf->strip_zone)*raid_disks
341 + sector_div(sector, zone->nb_dev)];
342}
343
344static sector_t raid0_size(struct mddev *mddev, sector_t sectors, int raid_disks)
345{
346 sector_t array_sectors = 0;
347 struct md_rdev *rdev;
348
349 WARN_ONCE(sectors || raid_disks,
350 "%s does not support generic reshape\n", __func__);
351
352 rdev_for_each(rdev, mddev)
353 array_sectors += (rdev->sectors &
354 ~(sector_t)(mddev->chunk_sectors-1));
355
356 return array_sectors;
357}
358
359static void raid0_free(struct mddev *mddev, void *priv);
360
361static int raid0_run(struct mddev *mddev)
362{
363 struct r0conf *conf;
364 int ret;
365
366 if (mddev->chunk_sectors == 0) {
367 pr_warn("md/raid0:%s: chunk size must be set.\n", mdname(mddev));
368 return -EINVAL;
369 }
370 if (md_check_no_bitmap(mddev))
371 return -EINVAL;
372
373 /* if private is not null, we are here after takeover */
374 if (mddev->private == NULL) {
375 ret = create_strip_zones(mddev, &conf);
376 if (ret < 0)
377 return ret;
378 mddev->private = conf;
379 }
380 conf = mddev->private;
381 if (mddev->queue) {
382 struct md_rdev *rdev;
383 bool discard_supported = false;
384
385 blk_queue_max_hw_sectors(mddev->queue, mddev->chunk_sectors);
386 blk_queue_max_write_same_sectors(mddev->queue, mddev->chunk_sectors);
387 blk_queue_max_write_zeroes_sectors(mddev->queue, mddev->chunk_sectors);
388 blk_queue_max_discard_sectors(mddev->queue, UINT_MAX);
389
390 blk_queue_io_min(mddev->queue, mddev->chunk_sectors << 9);
391 blk_queue_io_opt(mddev->queue,
392 (mddev->chunk_sectors << 9) * mddev->raid_disks);
393
394 rdev_for_each(rdev, mddev) {
395 disk_stack_limits(mddev->gendisk, rdev->bdev,
396 rdev->data_offset << 9);
397 if (blk_queue_discard(bdev_get_queue(rdev->bdev)))
398 discard_supported = true;
399 }
400 if (!discard_supported)
401 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, mddev->queue);
402 else
403 blk_queue_flag_set(QUEUE_FLAG_DISCARD, mddev->queue);
404 }
405
406 /* calculate array device size */
407 md_set_array_sectors(mddev, raid0_size(mddev, 0, 0));
408
409 pr_debug("md/raid0:%s: md_size is %llu sectors.\n",
410 mdname(mddev),
411 (unsigned long long)mddev->array_sectors);
412
413 if (mddev->queue) {
414 /* calculate the max read-ahead size.
415 * For read-ahead of large files to be effective, we need to
416 * readahead at least twice a whole stripe. i.e. number of devices
417 * multiplied by chunk size times 2.
418 * If an individual device has an ra_pages greater than the
419 * chunk size, then we will not drive that device as hard as it
420 * wants. We consider this a configuration error: a larger
421 * chunksize should be used in that case.
422 */
423 int stripe = mddev->raid_disks *
424 (mddev->chunk_sectors << 9) / PAGE_SIZE;
425 if (mddev->queue->backing_dev_info->ra_pages < 2* stripe)
426 mddev->queue->backing_dev_info->ra_pages = 2* stripe;
427 }
428
429 dump_zones(mddev);
430
431 ret = md_integrity_register(mddev);
432
433 return ret;
434}
435
436static void raid0_free(struct mddev *mddev, void *priv)
437{
438 struct r0conf *conf = priv;
439
440 kfree(conf->strip_zone);
441 kfree(conf->devlist);
442 kfree(conf);
443}
444
445/*
446 * Is io distribute over 1 or more chunks ?
447*/
448static inline int is_io_in_chunk_boundary(struct mddev *mddev,
449 unsigned int chunk_sects, struct bio *bio)
450{
451 if (likely(is_power_of_2(chunk_sects))) {
452 return chunk_sects >=
453 ((bio->bi_iter.bi_sector & (chunk_sects-1))
454 + bio_sectors(bio));
455 } else{
456 sector_t sector = bio->bi_iter.bi_sector;
457 return chunk_sects >= (sector_div(sector, chunk_sects)
458 + bio_sectors(bio));
459 }
460}
461
462static void raid0_handle_discard(struct mddev *mddev, struct bio *bio)
463{
464 struct r0conf *conf = mddev->private;
465 struct strip_zone *zone;
466 sector_t start = bio->bi_iter.bi_sector;
467 sector_t end;
468 unsigned int stripe_size;
469 sector_t first_stripe_index, last_stripe_index;
470 sector_t start_disk_offset;
471 unsigned int start_disk_index;
472 sector_t end_disk_offset;
473 unsigned int end_disk_index;
474 unsigned int disk;
475
476 zone = find_zone(conf, &start);
477
478 if (bio_end_sector(bio) > zone->zone_end) {
479 struct bio *split = bio_split(bio,
480 zone->zone_end - bio->bi_iter.bi_sector, GFP_NOIO,
481 &mddev->bio_set);
482 bio_chain(split, bio);
483 submit_bio_noacct(bio);
484 bio = split;
485 end = zone->zone_end;
486 } else
487 end = bio_end_sector(bio);
488
489 if (zone != conf->strip_zone)
490 end = end - zone[-1].zone_end;
491
492 /* Now start and end is the offset in zone */
493 stripe_size = zone->nb_dev * mddev->chunk_sectors;
494
495 first_stripe_index = start;
496 sector_div(first_stripe_index, stripe_size);
497 last_stripe_index = end;
498 sector_div(last_stripe_index, stripe_size);
499
500 start_disk_index = (int)(start - first_stripe_index * stripe_size) /
501 mddev->chunk_sectors;
502 start_disk_offset = ((int)(start - first_stripe_index * stripe_size) %
503 mddev->chunk_sectors) +
504 first_stripe_index * mddev->chunk_sectors;
505 end_disk_index = (int)(end - last_stripe_index * stripe_size) /
506 mddev->chunk_sectors;
507 end_disk_offset = ((int)(end - last_stripe_index * stripe_size) %
508 mddev->chunk_sectors) +
509 last_stripe_index * mddev->chunk_sectors;
510
511 for (disk = 0; disk < zone->nb_dev; disk++) {
512 sector_t dev_start, dev_end;
513 struct bio *discard_bio = NULL;
514 struct md_rdev *rdev;
515
516 if (disk < start_disk_index)
517 dev_start = (first_stripe_index + 1) *
518 mddev->chunk_sectors;
519 else if (disk > start_disk_index)
520 dev_start = first_stripe_index * mddev->chunk_sectors;
521 else
522 dev_start = start_disk_offset;
523
524 if (disk < end_disk_index)
525 dev_end = (last_stripe_index + 1) * mddev->chunk_sectors;
526 else if (disk > end_disk_index)
527 dev_end = last_stripe_index * mddev->chunk_sectors;
528 else
529 dev_end = end_disk_offset;
530
531 if (dev_end <= dev_start)
532 continue;
533
534 rdev = conf->devlist[(zone - conf->strip_zone) *
535 conf->strip_zone[0].nb_dev + disk];
536 if (__blkdev_issue_discard(rdev->bdev,
537 dev_start + zone->dev_start + rdev->data_offset,
538 dev_end - dev_start, GFP_NOIO, 0, &discard_bio) ||
539 !discard_bio)
540 continue;
541 bio_chain(discard_bio, bio);
542 bio_clone_blkg_association(discard_bio, bio);
543 if (mddev->gendisk)
544 trace_block_bio_remap(bdev_get_queue(rdev->bdev),
545 discard_bio, disk_devt(mddev->gendisk),
546 bio->bi_iter.bi_sector);
547 submit_bio_noacct(discard_bio);
548 }
549 bio_endio(bio);
550}
551
552static bool raid0_make_request(struct mddev *mddev, struct bio *bio)
553{
554 struct r0conf *conf = mddev->private;
555 struct strip_zone *zone;
556 struct md_rdev *tmp_dev;
557 sector_t bio_sector;
558 sector_t sector;
559 sector_t orig_sector;
560 unsigned chunk_sects;
561 unsigned sectors;
562
563 if (unlikely(bio->bi_opf & REQ_PREFLUSH)
564 && md_flush_request(mddev, bio))
565 return true;
566
567 if (unlikely((bio_op(bio) == REQ_OP_DISCARD))) {
568 raid0_handle_discard(mddev, bio);
569 return true;
570 }
571
572 bio_sector = bio->bi_iter.bi_sector;
573 sector = bio_sector;
574 chunk_sects = mddev->chunk_sectors;
575
576 sectors = chunk_sects -
577 (likely(is_power_of_2(chunk_sects))
578 ? (sector & (chunk_sects-1))
579 : sector_div(sector, chunk_sects));
580
581 /* Restore due to sector_div */
582 sector = bio_sector;
583
584 if (sectors < bio_sectors(bio)) {
585 struct bio *split = bio_split(bio, sectors, GFP_NOIO,
586 &mddev->bio_set);
587 bio_chain(split, bio);
588 submit_bio_noacct(bio);
589 bio = split;
590 }
591
592 orig_sector = sector;
593 zone = find_zone(mddev->private, §or);
594 switch (conf->layout) {
595 case RAID0_ORIG_LAYOUT:
596 tmp_dev = map_sector(mddev, zone, orig_sector, §or);
597 break;
598 case RAID0_ALT_MULTIZONE_LAYOUT:
599 tmp_dev = map_sector(mddev, zone, sector, §or);
600 break;
601 default:
602 WARN(1, "md/raid0:%s: Invalid layout\n", mdname(mddev));
603 bio_io_error(bio);
604 return true;
605 }
606
607 if (unlikely(is_mddev_broken(tmp_dev, "raid0"))) {
608 bio_io_error(bio);
609 return true;
610 }
611
612 bio_set_dev(bio, tmp_dev->bdev);
613 bio->bi_iter.bi_sector = sector + zone->dev_start +
614 tmp_dev->data_offset;
615
616 if (mddev->gendisk)
617 trace_block_bio_remap(bio->bi_disk->queue, bio,
618 disk_devt(mddev->gendisk), bio_sector);
619 mddev_check_writesame(mddev, bio);
620 mddev_check_write_zeroes(mddev, bio);
621 submit_bio_noacct(bio);
622 return true;
623}
624
625static void raid0_status(struct seq_file *seq, struct mddev *mddev)
626{
627 seq_printf(seq, " %dk chunks", mddev->chunk_sectors / 2);
628 return;
629}
630
631static void *raid0_takeover_raid45(struct mddev *mddev)
632{
633 struct md_rdev *rdev;
634 struct r0conf *priv_conf;
635
636 if (mddev->degraded != 1) {
637 pr_warn("md/raid0:%s: raid5 must be degraded! Degraded disks: %d\n",
638 mdname(mddev),
639 mddev->degraded);
640 return ERR_PTR(-EINVAL);
641 }
642
643 rdev_for_each(rdev, mddev) {
644 /* check slot number for a disk */
645 if (rdev->raid_disk == mddev->raid_disks-1) {
646 pr_warn("md/raid0:%s: raid5 must have missing parity disk!\n",
647 mdname(mddev));
648 return ERR_PTR(-EINVAL);
649 }
650 rdev->sectors = mddev->dev_sectors;
651 }
652
653 /* Set new parameters */
654 mddev->new_level = 0;
655 mddev->new_layout = 0;
656 mddev->new_chunk_sectors = mddev->chunk_sectors;
657 mddev->raid_disks--;
658 mddev->delta_disks = -1;
659 /* make sure it will be not marked as dirty */
660 mddev->recovery_cp = MaxSector;
661 mddev_clear_unsupported_flags(mddev, UNSUPPORTED_MDDEV_FLAGS);
662
663 create_strip_zones(mddev, &priv_conf);
664
665 return priv_conf;
666}
667
668static void *raid0_takeover_raid10(struct mddev *mddev)
669{
670 struct r0conf *priv_conf;
671
672 /* Check layout:
673 * - far_copies must be 1
674 * - near_copies must be 2
675 * - disks number must be even
676 * - all mirrors must be already degraded
677 */
678 if (mddev->layout != ((1 << 8) + 2)) {
679 pr_warn("md/raid0:%s:: Raid0 cannot takeover layout: 0x%x\n",
680 mdname(mddev),
681 mddev->layout);
682 return ERR_PTR(-EINVAL);
683 }
684 if (mddev->raid_disks & 1) {
685 pr_warn("md/raid0:%s: Raid0 cannot takeover Raid10 with odd disk number.\n",
686 mdname(mddev));
687 return ERR_PTR(-EINVAL);
688 }
689 if (mddev->degraded != (mddev->raid_disks>>1)) {
690 pr_warn("md/raid0:%s: All mirrors must be already degraded!\n",
691 mdname(mddev));
692 return ERR_PTR(-EINVAL);
693 }
694
695 /* Set new parameters */
696 mddev->new_level = 0;
697 mddev->new_layout = 0;
698 mddev->new_chunk_sectors = mddev->chunk_sectors;
699 mddev->delta_disks = - mddev->raid_disks / 2;
700 mddev->raid_disks += mddev->delta_disks;
701 mddev->degraded = 0;
702 /* make sure it will be not marked as dirty */
703 mddev->recovery_cp = MaxSector;
704 mddev_clear_unsupported_flags(mddev, UNSUPPORTED_MDDEV_FLAGS);
705
706 create_strip_zones(mddev, &priv_conf);
707 return priv_conf;
708}
709
710static void *raid0_takeover_raid1(struct mddev *mddev)
711{
712 struct r0conf *priv_conf;
713 int chunksect;
714
715 /* Check layout:
716 * - (N - 1) mirror drives must be already faulty
717 */
718 if ((mddev->raid_disks - 1) != mddev->degraded) {
719 pr_err("md/raid0:%s: (N - 1) mirrors drives must be already faulty!\n",
720 mdname(mddev));
721 return ERR_PTR(-EINVAL);
722 }
723
724 /*
725 * a raid1 doesn't have the notion of chunk size, so
726 * figure out the largest suitable size we can use.
727 */
728 chunksect = 64 * 2; /* 64K by default */
729
730 /* The array must be an exact multiple of chunksize */
731 while (chunksect && (mddev->array_sectors & (chunksect - 1)))
732 chunksect >>= 1;
733
734 if ((chunksect << 9) < PAGE_SIZE)
735 /* array size does not allow a suitable chunk size */
736 return ERR_PTR(-EINVAL);
737
738 /* Set new parameters */
739 mddev->new_level = 0;
740 mddev->new_layout = 0;
741 mddev->new_chunk_sectors = chunksect;
742 mddev->chunk_sectors = chunksect;
743 mddev->delta_disks = 1 - mddev->raid_disks;
744 mddev->raid_disks = 1;
745 /* make sure it will be not marked as dirty */
746 mddev->recovery_cp = MaxSector;
747 mddev_clear_unsupported_flags(mddev, UNSUPPORTED_MDDEV_FLAGS);
748
749 create_strip_zones(mddev, &priv_conf);
750 return priv_conf;
751}
752
753static void *raid0_takeover(struct mddev *mddev)
754{
755 /* raid0 can take over:
756 * raid4 - if all data disks are active.
757 * raid5 - providing it is Raid4 layout and one disk is faulty
758 * raid10 - assuming we have all necessary active disks
759 * raid1 - with (N -1) mirror drives faulty
760 */
761
762 if (mddev->bitmap) {
763 pr_warn("md/raid0: %s: cannot takeover array with bitmap\n",
764 mdname(mddev));
765 return ERR_PTR(-EBUSY);
766 }
767 if (mddev->level == 4)
768 return raid0_takeover_raid45(mddev);
769
770 if (mddev->level == 5) {
771 if (mddev->layout == ALGORITHM_PARITY_N)
772 return raid0_takeover_raid45(mddev);
773
774 pr_warn("md/raid0:%s: Raid can only takeover Raid5 with layout: %d\n",
775 mdname(mddev), ALGORITHM_PARITY_N);
776 }
777
778 if (mddev->level == 10)
779 return raid0_takeover_raid10(mddev);
780
781 if (mddev->level == 1)
782 return raid0_takeover_raid1(mddev);
783
784 pr_warn("Takeover from raid%i to raid0 not supported\n",
785 mddev->level);
786
787 return ERR_PTR(-EINVAL);
788}
789
790static void raid0_quiesce(struct mddev *mddev, int quiesce)
791{
792}
793
794static struct md_personality raid0_personality=
795{
796 .name = "raid0",
797 .level = 0,
798 .owner = THIS_MODULE,
799 .make_request = raid0_make_request,
800 .run = raid0_run,
801 .free = raid0_free,
802 .status = raid0_status,
803 .size = raid0_size,
804 .takeover = raid0_takeover,
805 .quiesce = raid0_quiesce,
806};
807
808static int __init raid0_init (void)
809{
810 return register_md_personality (&raid0_personality);
811}
812
813static void raid0_exit (void)
814{
815 unregister_md_personality (&raid0_personality);
816}
817
818module_init(raid0_init);
819module_exit(raid0_exit);
820MODULE_LICENSE("GPL");
821MODULE_DESCRIPTION("RAID0 (striping) personality for MD");
822MODULE_ALIAS("md-personality-2"); /* RAID0 */
823MODULE_ALIAS("md-raid0");
824MODULE_ALIAS("md-level-0");
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#define UNSUPPORTED_MDDEV_FLAGS \
30 ((1L << MD_HAS_JOURNAL) | \
31 (1L << MD_JOURNAL_CLEAN) | \
32 (1L << MD_FAILFAST_SUPPORTED) |\
33 (1L << MD_HAS_PPL) | \
34 (1L << MD_HAS_MULTIPLE_PPLS))
35
36static int raid0_congested(struct mddev *mddev, int bits)
37{
38 struct r0conf *conf = mddev->private;
39 struct md_rdev **devlist = conf->devlist;
40 int raid_disks = conf->strip_zone[0].nb_dev;
41 int i, ret = 0;
42
43 for (i = 0; i < raid_disks && !ret ; i++) {
44 struct request_queue *q = bdev_get_queue(devlist[i]->bdev);
45
46 ret |= bdi_congested(q->backing_dev_info, bits);
47 }
48 return ret;
49}
50
51/*
52 * inform the user of the raid configuration
53*/
54static void dump_zones(struct mddev *mddev)
55{
56 int j, k;
57 sector_t zone_size = 0;
58 sector_t zone_start = 0;
59 char b[BDEVNAME_SIZE];
60 struct r0conf *conf = mddev->private;
61 int raid_disks = conf->strip_zone[0].nb_dev;
62 pr_debug("md: RAID0 configuration for %s - %d zone%s\n",
63 mdname(mddev),
64 conf->nr_strip_zones, conf->nr_strip_zones==1?"":"s");
65 for (j = 0; j < conf->nr_strip_zones; j++) {
66 char line[200];
67 int len = 0;
68
69 for (k = 0; k < conf->strip_zone[j].nb_dev; k++)
70 len += snprintf(line+len, 200-len, "%s%s", k?"/":"",
71 bdevname(conf->devlist[j*raid_disks
72 + k]->bdev, b));
73 pr_debug("md: zone%d=[%s]\n", j, line);
74
75 zone_size = conf->strip_zone[j].zone_end - zone_start;
76 pr_debug(" zone-offset=%10lluKB, device-offset=%10lluKB, size=%10lluKB\n",
77 (unsigned long long)zone_start>>1,
78 (unsigned long long)conf->strip_zone[j].dev_start>>1,
79 (unsigned long long)zone_size>>1);
80 zone_start = conf->strip_zone[j].zone_end;
81 }
82}
83
84static int create_strip_zones(struct mddev *mddev, struct r0conf **private_conf)
85{
86 int i, c, err;
87 sector_t curr_zone_end, sectors;
88 struct md_rdev *smallest, *rdev1, *rdev2, *rdev, **dev;
89 struct strip_zone *zone;
90 int cnt;
91 char b[BDEVNAME_SIZE];
92 char b2[BDEVNAME_SIZE];
93 struct r0conf *conf = kzalloc(sizeof(*conf), GFP_KERNEL);
94 unsigned short blksize = 512;
95
96 *private_conf = ERR_PTR(-ENOMEM);
97 if (!conf)
98 return -ENOMEM;
99 rdev_for_each(rdev1, mddev) {
100 pr_debug("md/raid0:%s: looking at %s\n",
101 mdname(mddev),
102 bdevname(rdev1->bdev, b));
103 c = 0;
104
105 /* round size to chunk_size */
106 sectors = rdev1->sectors;
107 sector_div(sectors, mddev->chunk_sectors);
108 rdev1->sectors = sectors * mddev->chunk_sectors;
109
110 blksize = max(blksize, queue_logical_block_size(
111 rdev1->bdev->bd_disk->queue));
112
113 rdev_for_each(rdev2, mddev) {
114 pr_debug("md/raid0:%s: comparing %s(%llu)"
115 " with %s(%llu)\n",
116 mdname(mddev),
117 bdevname(rdev1->bdev,b),
118 (unsigned long long)rdev1->sectors,
119 bdevname(rdev2->bdev,b2),
120 (unsigned long long)rdev2->sectors);
121 if (rdev2 == rdev1) {
122 pr_debug("md/raid0:%s: END\n",
123 mdname(mddev));
124 break;
125 }
126 if (rdev2->sectors == rdev1->sectors) {
127 /*
128 * Not unique, don't count it as a new
129 * group
130 */
131 pr_debug("md/raid0:%s: EQUAL\n",
132 mdname(mddev));
133 c = 1;
134 break;
135 }
136 pr_debug("md/raid0:%s: NOT EQUAL\n",
137 mdname(mddev));
138 }
139 if (!c) {
140 pr_debug("md/raid0:%s: ==> UNIQUE\n",
141 mdname(mddev));
142 conf->nr_strip_zones++;
143 pr_debug("md/raid0:%s: %d zones\n",
144 mdname(mddev), conf->nr_strip_zones);
145 }
146 }
147 pr_debug("md/raid0:%s: FINAL %d zones\n",
148 mdname(mddev), conf->nr_strip_zones);
149 /*
150 * now since we have the hard sector sizes, we can make sure
151 * chunk size is a multiple of that sector size
152 */
153 if ((mddev->chunk_sectors << 9) % blksize) {
154 pr_warn("md/raid0:%s: chunk_size of %d not multiple of block size %d\n",
155 mdname(mddev),
156 mddev->chunk_sectors << 9, blksize);
157 err = -EINVAL;
158 goto abort;
159 }
160
161 err = -ENOMEM;
162 conf->strip_zone = kzalloc(sizeof(struct strip_zone)*
163 conf->nr_strip_zones, GFP_KERNEL);
164 if (!conf->strip_zone)
165 goto abort;
166 conf->devlist = kzalloc(sizeof(struct md_rdev*)*
167 conf->nr_strip_zones*mddev->raid_disks,
168 GFP_KERNEL);
169 if (!conf->devlist)
170 goto abort;
171
172 /* The first zone must contain all devices, so here we check that
173 * there is a proper alignment of slots to devices and find them all
174 */
175 zone = &conf->strip_zone[0];
176 cnt = 0;
177 smallest = NULL;
178 dev = conf->devlist;
179 err = -EINVAL;
180 rdev_for_each(rdev1, mddev) {
181 int j = rdev1->raid_disk;
182
183 if (mddev->level == 10) {
184 /* taking over a raid10-n2 array */
185 j /= 2;
186 rdev1->new_raid_disk = j;
187 }
188
189 if (mddev->level == 1) {
190 /* taiking over a raid1 array-
191 * we have only one active disk
192 */
193 j = 0;
194 rdev1->new_raid_disk = j;
195 }
196
197 if (j < 0) {
198 pr_warn("md/raid0:%s: remove inactive devices before converting to RAID0\n",
199 mdname(mddev));
200 goto abort;
201 }
202 if (j >= mddev->raid_disks) {
203 pr_warn("md/raid0:%s: bad disk number %d - aborting!\n",
204 mdname(mddev), j);
205 goto abort;
206 }
207 if (dev[j]) {
208 pr_warn("md/raid0:%s: multiple devices for %d - aborting!\n",
209 mdname(mddev), j);
210 goto abort;
211 }
212 dev[j] = rdev1;
213
214 if (!smallest || (rdev1->sectors < smallest->sectors))
215 smallest = rdev1;
216 cnt++;
217 }
218 if (cnt != mddev->raid_disks) {
219 pr_warn("md/raid0:%s: too few disks (%d of %d) - aborting!\n",
220 mdname(mddev), cnt, mddev->raid_disks);
221 goto abort;
222 }
223 zone->nb_dev = cnt;
224 zone->zone_end = smallest->sectors * cnt;
225
226 curr_zone_end = zone->zone_end;
227
228 /* now do the other zones */
229 for (i = 1; i < conf->nr_strip_zones; i++)
230 {
231 int j;
232
233 zone = conf->strip_zone + i;
234 dev = conf->devlist + i * mddev->raid_disks;
235
236 pr_debug("md/raid0:%s: zone %d\n", mdname(mddev), i);
237 zone->dev_start = smallest->sectors;
238 smallest = NULL;
239 c = 0;
240
241 for (j=0; j<cnt; j++) {
242 rdev = conf->devlist[j];
243 if (rdev->sectors <= zone->dev_start) {
244 pr_debug("md/raid0:%s: checking %s ... nope\n",
245 mdname(mddev),
246 bdevname(rdev->bdev, b));
247 continue;
248 }
249 pr_debug("md/raid0:%s: checking %s ..."
250 " contained as device %d\n",
251 mdname(mddev),
252 bdevname(rdev->bdev, b), c);
253 dev[c] = rdev;
254 c++;
255 if (!smallest || rdev->sectors < smallest->sectors) {
256 smallest = rdev;
257 pr_debug("md/raid0:%s: (%llu) is smallest!.\n",
258 mdname(mddev),
259 (unsigned long long)rdev->sectors);
260 }
261 }
262
263 zone->nb_dev = c;
264 sectors = (smallest->sectors - zone->dev_start) * c;
265 pr_debug("md/raid0:%s: zone->nb_dev: %d, sectors: %llu\n",
266 mdname(mddev),
267 zone->nb_dev, (unsigned long long)sectors);
268
269 curr_zone_end += sectors;
270 zone->zone_end = curr_zone_end;
271
272 pr_debug("md/raid0:%s: current zone start: %llu\n",
273 mdname(mddev),
274 (unsigned long long)smallest->sectors);
275 }
276
277 pr_debug("md/raid0:%s: done.\n", mdname(mddev));
278 *private_conf = conf;
279
280 return 0;
281abort:
282 kfree(conf->strip_zone);
283 kfree(conf->devlist);
284 kfree(conf);
285 *private_conf = ERR_PTR(err);
286 return err;
287}
288
289/* Find the zone which holds a particular offset
290 * Update *sectorp to be an offset in that zone
291 */
292static struct strip_zone *find_zone(struct r0conf *conf,
293 sector_t *sectorp)
294{
295 int i;
296 struct strip_zone *z = conf->strip_zone;
297 sector_t sector = *sectorp;
298
299 for (i = 0; i < conf->nr_strip_zones; i++)
300 if (sector < z[i].zone_end) {
301 if (i)
302 *sectorp = sector - z[i-1].zone_end;
303 return z + i;
304 }
305 BUG();
306}
307
308/*
309 * remaps the bio to the target device. we separate two flows.
310 * power 2 flow and a general flow for the sake of performance
311*/
312static struct md_rdev *map_sector(struct mddev *mddev, struct strip_zone *zone,
313 sector_t sector, sector_t *sector_offset)
314{
315 unsigned int sect_in_chunk;
316 sector_t chunk;
317 struct r0conf *conf = mddev->private;
318 int raid_disks = conf->strip_zone[0].nb_dev;
319 unsigned int chunk_sects = mddev->chunk_sectors;
320
321 if (is_power_of_2(chunk_sects)) {
322 int chunksect_bits = ffz(~chunk_sects);
323 /* find the sector offset inside the chunk */
324 sect_in_chunk = sector & (chunk_sects - 1);
325 sector >>= chunksect_bits;
326 /* chunk in zone */
327 chunk = *sector_offset;
328 /* quotient is the chunk in real device*/
329 sector_div(chunk, zone->nb_dev << chunksect_bits);
330 } else{
331 sect_in_chunk = sector_div(sector, chunk_sects);
332 chunk = *sector_offset;
333 sector_div(chunk, chunk_sects * zone->nb_dev);
334 }
335 /*
336 * position the bio over the real device
337 * real sector = chunk in device + starting of zone
338 * + the position in the chunk
339 */
340 *sector_offset = (chunk * chunk_sects) + sect_in_chunk;
341 return conf->devlist[(zone - conf->strip_zone)*raid_disks
342 + sector_div(sector, zone->nb_dev)];
343}
344
345static sector_t raid0_size(struct mddev *mddev, sector_t sectors, int raid_disks)
346{
347 sector_t array_sectors = 0;
348 struct md_rdev *rdev;
349
350 WARN_ONCE(sectors || raid_disks,
351 "%s does not support generic reshape\n", __func__);
352
353 rdev_for_each(rdev, mddev)
354 array_sectors += (rdev->sectors &
355 ~(sector_t)(mddev->chunk_sectors-1));
356
357 return array_sectors;
358}
359
360static void raid0_free(struct mddev *mddev, void *priv);
361
362static int raid0_run(struct mddev *mddev)
363{
364 struct r0conf *conf;
365 int ret;
366
367 if (mddev->chunk_sectors == 0) {
368 pr_warn("md/raid0:%s: chunk size must be set.\n", mdname(mddev));
369 return -EINVAL;
370 }
371 if (md_check_no_bitmap(mddev))
372 return -EINVAL;
373
374 /* if private is not null, we are here after takeover */
375 if (mddev->private == NULL) {
376 ret = create_strip_zones(mddev, &conf);
377 if (ret < 0)
378 return ret;
379 mddev->private = conf;
380 }
381 conf = mddev->private;
382 if (mddev->queue) {
383 struct md_rdev *rdev;
384 bool discard_supported = false;
385
386 blk_queue_max_hw_sectors(mddev->queue, mddev->chunk_sectors);
387 blk_queue_max_write_same_sectors(mddev->queue, mddev->chunk_sectors);
388 blk_queue_max_write_zeroes_sectors(mddev->queue, mddev->chunk_sectors);
389 blk_queue_max_discard_sectors(mddev->queue, UINT_MAX);
390
391 blk_queue_io_min(mddev->queue, mddev->chunk_sectors << 9);
392 blk_queue_io_opt(mddev->queue,
393 (mddev->chunk_sectors << 9) * mddev->raid_disks);
394
395 rdev_for_each(rdev, mddev) {
396 disk_stack_limits(mddev->gendisk, rdev->bdev,
397 rdev->data_offset << 9);
398 if (blk_queue_discard(bdev_get_queue(rdev->bdev)))
399 discard_supported = true;
400 }
401 if (!discard_supported)
402 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, mddev->queue);
403 else
404 blk_queue_flag_set(QUEUE_FLAG_DISCARD, mddev->queue);
405 }
406
407 /* calculate array device size */
408 md_set_array_sectors(mddev, raid0_size(mddev, 0, 0));
409
410 pr_debug("md/raid0:%s: md_size is %llu sectors.\n",
411 mdname(mddev),
412 (unsigned long long)mddev->array_sectors);
413
414 if (mddev->queue) {
415 /* calculate the max read-ahead size.
416 * For read-ahead of large files to be effective, we need to
417 * readahead at least twice a whole stripe. i.e. number of devices
418 * multiplied by chunk size times 2.
419 * If an individual device has an ra_pages greater than the
420 * chunk size, then we will not drive that device as hard as it
421 * wants. We consider this a configuration error: a larger
422 * chunksize should be used in that case.
423 */
424 int stripe = mddev->raid_disks *
425 (mddev->chunk_sectors << 9) / PAGE_SIZE;
426 if (mddev->queue->backing_dev_info->ra_pages < 2* stripe)
427 mddev->queue->backing_dev_info->ra_pages = 2* stripe;
428 }
429
430 dump_zones(mddev);
431
432 ret = md_integrity_register(mddev);
433
434 return ret;
435}
436
437static void raid0_free(struct mddev *mddev, void *priv)
438{
439 struct r0conf *conf = priv;
440
441 kfree(conf->strip_zone);
442 kfree(conf->devlist);
443 kfree(conf);
444}
445
446/*
447 * Is io distribute over 1 or more chunks ?
448*/
449static inline int is_io_in_chunk_boundary(struct mddev *mddev,
450 unsigned int chunk_sects, struct bio *bio)
451{
452 if (likely(is_power_of_2(chunk_sects))) {
453 return chunk_sects >=
454 ((bio->bi_iter.bi_sector & (chunk_sects-1))
455 + bio_sectors(bio));
456 } else{
457 sector_t sector = bio->bi_iter.bi_sector;
458 return chunk_sects >= (sector_div(sector, chunk_sects)
459 + bio_sectors(bio));
460 }
461}
462
463static void raid0_handle_discard(struct mddev *mddev, struct bio *bio)
464{
465 struct r0conf *conf = mddev->private;
466 struct strip_zone *zone;
467 sector_t start = bio->bi_iter.bi_sector;
468 sector_t end;
469 unsigned int stripe_size;
470 sector_t first_stripe_index, last_stripe_index;
471 sector_t start_disk_offset;
472 unsigned int start_disk_index;
473 sector_t end_disk_offset;
474 unsigned int end_disk_index;
475 unsigned int disk;
476
477 zone = find_zone(conf, &start);
478
479 if (bio_end_sector(bio) > zone->zone_end) {
480 struct bio *split = bio_split(bio,
481 zone->zone_end - bio->bi_iter.bi_sector, GFP_NOIO,
482 mddev->bio_set);
483 bio_chain(split, bio);
484 generic_make_request(bio);
485 bio = split;
486 end = zone->zone_end;
487 } else
488 end = bio_end_sector(bio);
489
490 if (zone != conf->strip_zone)
491 end = end - zone[-1].zone_end;
492
493 /* Now start and end is the offset in zone */
494 stripe_size = zone->nb_dev * mddev->chunk_sectors;
495
496 first_stripe_index = start;
497 sector_div(first_stripe_index, stripe_size);
498 last_stripe_index = end;
499 sector_div(last_stripe_index, stripe_size);
500
501 start_disk_index = (int)(start - first_stripe_index * stripe_size) /
502 mddev->chunk_sectors;
503 start_disk_offset = ((int)(start - first_stripe_index * stripe_size) %
504 mddev->chunk_sectors) +
505 first_stripe_index * mddev->chunk_sectors;
506 end_disk_index = (int)(end - last_stripe_index * stripe_size) /
507 mddev->chunk_sectors;
508 end_disk_offset = ((int)(end - last_stripe_index * stripe_size) %
509 mddev->chunk_sectors) +
510 last_stripe_index * mddev->chunk_sectors;
511
512 for (disk = 0; disk < zone->nb_dev; disk++) {
513 sector_t dev_start, dev_end;
514 struct bio *discard_bio = NULL;
515 struct md_rdev *rdev;
516
517 if (disk < start_disk_index)
518 dev_start = (first_stripe_index + 1) *
519 mddev->chunk_sectors;
520 else if (disk > start_disk_index)
521 dev_start = first_stripe_index * mddev->chunk_sectors;
522 else
523 dev_start = start_disk_offset;
524
525 if (disk < end_disk_index)
526 dev_end = (last_stripe_index + 1) * mddev->chunk_sectors;
527 else if (disk > end_disk_index)
528 dev_end = last_stripe_index * mddev->chunk_sectors;
529 else
530 dev_end = end_disk_offset;
531
532 if (dev_end <= dev_start)
533 continue;
534
535 rdev = conf->devlist[(zone - conf->strip_zone) *
536 conf->strip_zone[0].nb_dev + disk];
537 if (__blkdev_issue_discard(rdev->bdev,
538 dev_start + zone->dev_start + rdev->data_offset,
539 dev_end - dev_start, GFP_NOIO, 0, &discard_bio) ||
540 !discard_bio)
541 continue;
542 bio_chain(discard_bio, bio);
543 bio_clone_blkcg_association(discard_bio, bio);
544 if (mddev->gendisk)
545 trace_block_bio_remap(bdev_get_queue(rdev->bdev),
546 discard_bio, disk_devt(mddev->gendisk),
547 bio->bi_iter.bi_sector);
548 generic_make_request(discard_bio);
549 }
550 bio_endio(bio);
551}
552
553static bool raid0_make_request(struct mddev *mddev, struct bio *bio)
554{
555 struct strip_zone *zone;
556 struct md_rdev *tmp_dev;
557 sector_t bio_sector;
558 sector_t sector;
559 unsigned chunk_sects;
560 unsigned sectors;
561
562 if (unlikely(bio->bi_opf & REQ_PREFLUSH)) {
563 md_flush_request(mddev, bio);
564 return true;
565 }
566
567 if (unlikely((bio_op(bio) == REQ_OP_DISCARD))) {
568 raid0_handle_discard(mddev, bio);
569 return true;
570 }
571
572 bio_sector = bio->bi_iter.bi_sector;
573 sector = bio_sector;
574 chunk_sects = mddev->chunk_sectors;
575
576 sectors = chunk_sects -
577 (likely(is_power_of_2(chunk_sects))
578 ? (sector & (chunk_sects-1))
579 : sector_div(sector, chunk_sects));
580
581 /* Restore due to sector_div */
582 sector = bio_sector;
583
584 if (sectors < bio_sectors(bio)) {
585 struct bio *split = bio_split(bio, sectors, GFP_NOIO, mddev->bio_set);
586 bio_chain(split, bio);
587 generic_make_request(bio);
588 bio = split;
589 }
590
591 zone = find_zone(mddev->private, §or);
592 tmp_dev = map_sector(mddev, zone, sector, §or);
593 bio_set_dev(bio, tmp_dev->bdev);
594 bio->bi_iter.bi_sector = sector + zone->dev_start +
595 tmp_dev->data_offset;
596
597 if (mddev->gendisk)
598 trace_block_bio_remap(bio->bi_disk->queue, bio,
599 disk_devt(mddev->gendisk), bio_sector);
600 mddev_check_writesame(mddev, bio);
601 mddev_check_write_zeroes(mddev, bio);
602 generic_make_request(bio);
603 return true;
604}
605
606static void raid0_status(struct seq_file *seq, struct mddev *mddev)
607{
608 seq_printf(seq, " %dk chunks", mddev->chunk_sectors / 2);
609 return;
610}
611
612static void *raid0_takeover_raid45(struct mddev *mddev)
613{
614 struct md_rdev *rdev;
615 struct r0conf *priv_conf;
616
617 if (mddev->degraded != 1) {
618 pr_warn("md/raid0:%s: raid5 must be degraded! Degraded disks: %d\n",
619 mdname(mddev),
620 mddev->degraded);
621 return ERR_PTR(-EINVAL);
622 }
623
624 rdev_for_each(rdev, mddev) {
625 /* check slot number for a disk */
626 if (rdev->raid_disk == mddev->raid_disks-1) {
627 pr_warn("md/raid0:%s: raid5 must have missing parity disk!\n",
628 mdname(mddev));
629 return ERR_PTR(-EINVAL);
630 }
631 rdev->sectors = mddev->dev_sectors;
632 }
633
634 /* Set new parameters */
635 mddev->new_level = 0;
636 mddev->new_layout = 0;
637 mddev->new_chunk_sectors = mddev->chunk_sectors;
638 mddev->raid_disks--;
639 mddev->delta_disks = -1;
640 /* make sure it will be not marked as dirty */
641 mddev->recovery_cp = MaxSector;
642 mddev_clear_unsupported_flags(mddev, UNSUPPORTED_MDDEV_FLAGS);
643
644 create_strip_zones(mddev, &priv_conf);
645
646 return priv_conf;
647}
648
649static void *raid0_takeover_raid10(struct mddev *mddev)
650{
651 struct r0conf *priv_conf;
652
653 /* Check layout:
654 * - far_copies must be 1
655 * - near_copies must be 2
656 * - disks number must be even
657 * - all mirrors must be already degraded
658 */
659 if (mddev->layout != ((1 << 8) + 2)) {
660 pr_warn("md/raid0:%s:: Raid0 cannot takeover layout: 0x%x\n",
661 mdname(mddev),
662 mddev->layout);
663 return ERR_PTR(-EINVAL);
664 }
665 if (mddev->raid_disks & 1) {
666 pr_warn("md/raid0:%s: Raid0 cannot takeover Raid10 with odd disk number.\n",
667 mdname(mddev));
668 return ERR_PTR(-EINVAL);
669 }
670 if (mddev->degraded != (mddev->raid_disks>>1)) {
671 pr_warn("md/raid0:%s: All mirrors must be already degraded!\n",
672 mdname(mddev));
673 return ERR_PTR(-EINVAL);
674 }
675
676 /* Set new parameters */
677 mddev->new_level = 0;
678 mddev->new_layout = 0;
679 mddev->new_chunk_sectors = mddev->chunk_sectors;
680 mddev->delta_disks = - mddev->raid_disks / 2;
681 mddev->raid_disks += mddev->delta_disks;
682 mddev->degraded = 0;
683 /* make sure it will be not marked as dirty */
684 mddev->recovery_cp = MaxSector;
685 mddev_clear_unsupported_flags(mddev, UNSUPPORTED_MDDEV_FLAGS);
686
687 create_strip_zones(mddev, &priv_conf);
688 return priv_conf;
689}
690
691static void *raid0_takeover_raid1(struct mddev *mddev)
692{
693 struct r0conf *priv_conf;
694 int chunksect;
695
696 /* Check layout:
697 * - (N - 1) mirror drives must be already faulty
698 */
699 if ((mddev->raid_disks - 1) != mddev->degraded) {
700 pr_err("md/raid0:%s: (N - 1) mirrors drives must be already faulty!\n",
701 mdname(mddev));
702 return ERR_PTR(-EINVAL);
703 }
704
705 /*
706 * a raid1 doesn't have the notion of chunk size, so
707 * figure out the largest suitable size we can use.
708 */
709 chunksect = 64 * 2; /* 64K by default */
710
711 /* The array must be an exact multiple of chunksize */
712 while (chunksect && (mddev->array_sectors & (chunksect - 1)))
713 chunksect >>= 1;
714
715 if ((chunksect << 9) < PAGE_SIZE)
716 /* array size does not allow a suitable chunk size */
717 return ERR_PTR(-EINVAL);
718
719 /* Set new parameters */
720 mddev->new_level = 0;
721 mddev->new_layout = 0;
722 mddev->new_chunk_sectors = chunksect;
723 mddev->chunk_sectors = chunksect;
724 mddev->delta_disks = 1 - mddev->raid_disks;
725 mddev->raid_disks = 1;
726 /* make sure it will be not marked as dirty */
727 mddev->recovery_cp = MaxSector;
728 mddev_clear_unsupported_flags(mddev, UNSUPPORTED_MDDEV_FLAGS);
729
730 create_strip_zones(mddev, &priv_conf);
731 return priv_conf;
732}
733
734static void *raid0_takeover(struct mddev *mddev)
735{
736 /* raid0 can take over:
737 * raid4 - if all data disks are active.
738 * raid5 - providing it is Raid4 layout and one disk is faulty
739 * raid10 - assuming we have all necessary active disks
740 * raid1 - with (N -1) mirror drives faulty
741 */
742
743 if (mddev->bitmap) {
744 pr_warn("md/raid0: %s: cannot takeover array with bitmap\n",
745 mdname(mddev));
746 return ERR_PTR(-EBUSY);
747 }
748 if (mddev->level == 4)
749 return raid0_takeover_raid45(mddev);
750
751 if (mddev->level == 5) {
752 if (mddev->layout == ALGORITHM_PARITY_N)
753 return raid0_takeover_raid45(mddev);
754
755 pr_warn("md/raid0:%s: Raid can only takeover Raid5 with layout: %d\n",
756 mdname(mddev), ALGORITHM_PARITY_N);
757 }
758
759 if (mddev->level == 10)
760 return raid0_takeover_raid10(mddev);
761
762 if (mddev->level == 1)
763 return raid0_takeover_raid1(mddev);
764
765 pr_warn("Takeover from raid%i to raid0 not supported\n",
766 mddev->level);
767
768 return ERR_PTR(-EINVAL);
769}
770
771static void raid0_quiesce(struct mddev *mddev, int quiesce)
772{
773}
774
775static struct md_personality raid0_personality=
776{
777 .name = "raid0",
778 .level = 0,
779 .owner = THIS_MODULE,
780 .make_request = raid0_make_request,
781 .run = raid0_run,
782 .free = raid0_free,
783 .status = raid0_status,
784 .size = raid0_size,
785 .takeover = raid0_takeover,
786 .quiesce = raid0_quiesce,
787 .congested = raid0_congested,
788};
789
790static int __init raid0_init (void)
791{
792 return register_md_personality (&raid0_personality);
793}
794
795static void raid0_exit (void)
796{
797 unregister_md_personality (&raid0_personality);
798}
799
800module_init(raid0_init);
801module_exit(raid0_exit);
802MODULE_LICENSE("GPL");
803MODULE_DESCRIPTION("RAID0 (striping) personality for MD");
804MODULE_ALIAS("md-personality-2"); /* RAID0 */
805MODULE_ALIAS("md-raid0");
806MODULE_ALIAS("md-level-0");