Loading...
1/*
2 raid0.c : Multiple Devices driver for Linux
3 Copyright (C) 1994-96 Marc ZYNGIER
4 <zyngier@ufr-info-p7.ibp.fr> or
5 <maz@gloups.fdn.fr>
6 Copyright (C) 1999, 2000 Ingo Molnar, Red Hat
7
8 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 "md.h"
25#include "raid0.h"
26#include "raid5.h"
27
28static int raid0_congested(struct mddev *mddev, int bits)
29{
30 struct r0conf *conf = mddev->private;
31 struct md_rdev **devlist = conf->devlist;
32 int raid_disks = conf->strip_zone[0].nb_dev;
33 int i, ret = 0;
34
35 for (i = 0; i < raid_disks && !ret ; i++) {
36 struct request_queue *q = bdev_get_queue(devlist[i]->bdev);
37
38 ret |= bdi_congested(&q->backing_dev_info, bits);
39 }
40 return ret;
41}
42
43/*
44 * inform the user of the raid configuration
45*/
46static void dump_zones(struct mddev *mddev)
47{
48 int j, k;
49 sector_t zone_size = 0;
50 sector_t zone_start = 0;
51 char b[BDEVNAME_SIZE];
52 struct r0conf *conf = mddev->private;
53 int raid_disks = conf->strip_zone[0].nb_dev;
54 printk(KERN_INFO "md: RAID0 configuration for %s - %d zone%s\n",
55 mdname(mddev),
56 conf->nr_strip_zones, conf->nr_strip_zones==1?"":"s");
57 for (j = 0; j < conf->nr_strip_zones; j++) {
58 printk(KERN_INFO "md: zone%d=[", j);
59 for (k = 0; k < conf->strip_zone[j].nb_dev; k++)
60 printk(KERN_CONT "%s%s", k?"/":"",
61 bdevname(conf->devlist[j*raid_disks
62 + k]->bdev, b));
63 printk(KERN_CONT "]\n");
64
65 zone_size = conf->strip_zone[j].zone_end - zone_start;
66 printk(KERN_INFO " zone-offset=%10lluKB, "
67 "device-offset=%10lluKB, size=%10lluKB\n",
68 (unsigned long long)zone_start>>1,
69 (unsigned long long)conf->strip_zone[j].dev_start>>1,
70 (unsigned long long)zone_size>>1);
71 zone_start = conf->strip_zone[j].zone_end;
72 }
73}
74
75static int create_strip_zones(struct mddev *mddev, struct r0conf **private_conf)
76{
77 int i, c, err;
78 sector_t curr_zone_end, sectors;
79 struct md_rdev *smallest, *rdev1, *rdev2, *rdev, **dev;
80 struct strip_zone *zone;
81 int cnt;
82 char b[BDEVNAME_SIZE];
83 char b2[BDEVNAME_SIZE];
84 struct r0conf *conf = kzalloc(sizeof(*conf), GFP_KERNEL);
85 unsigned short blksize = 512;
86
87 *private_conf = ERR_PTR(-ENOMEM);
88 if (!conf)
89 return -ENOMEM;
90 rdev_for_each(rdev1, mddev) {
91 pr_debug("md/raid0:%s: looking at %s\n",
92 mdname(mddev),
93 bdevname(rdev1->bdev, b));
94 c = 0;
95
96 /* round size to chunk_size */
97 sectors = rdev1->sectors;
98 sector_div(sectors, mddev->chunk_sectors);
99 rdev1->sectors = sectors * mddev->chunk_sectors;
100
101 blksize = max(blksize, queue_logical_block_size(
102 rdev1->bdev->bd_disk->queue));
103
104 rdev_for_each(rdev2, mddev) {
105 pr_debug("md/raid0:%s: comparing %s(%llu)"
106 " with %s(%llu)\n",
107 mdname(mddev),
108 bdevname(rdev1->bdev,b),
109 (unsigned long long)rdev1->sectors,
110 bdevname(rdev2->bdev,b2),
111 (unsigned long long)rdev2->sectors);
112 if (rdev2 == rdev1) {
113 pr_debug("md/raid0:%s: END\n",
114 mdname(mddev));
115 break;
116 }
117 if (rdev2->sectors == rdev1->sectors) {
118 /*
119 * Not unique, don't count it as a new
120 * group
121 */
122 pr_debug("md/raid0:%s: EQUAL\n",
123 mdname(mddev));
124 c = 1;
125 break;
126 }
127 pr_debug("md/raid0:%s: NOT EQUAL\n",
128 mdname(mddev));
129 }
130 if (!c) {
131 pr_debug("md/raid0:%s: ==> UNIQUE\n",
132 mdname(mddev));
133 conf->nr_strip_zones++;
134 pr_debug("md/raid0:%s: %d zones\n",
135 mdname(mddev), conf->nr_strip_zones);
136 }
137 }
138 pr_debug("md/raid0:%s: FINAL %d zones\n",
139 mdname(mddev), conf->nr_strip_zones);
140 /*
141 * now since we have the hard sector sizes, we can make sure
142 * chunk size is a multiple of that sector size
143 */
144 if ((mddev->chunk_sectors << 9) % blksize) {
145 printk(KERN_ERR "md/raid0:%s: chunk_size of %d not multiple of block size %d\n",
146 mdname(mddev),
147 mddev->chunk_sectors << 9, blksize);
148 err = -EINVAL;
149 goto abort;
150 }
151
152 err = -ENOMEM;
153 conf->strip_zone = kzalloc(sizeof(struct strip_zone)*
154 conf->nr_strip_zones, GFP_KERNEL);
155 if (!conf->strip_zone)
156 goto abort;
157 conf->devlist = kzalloc(sizeof(struct md_rdev*)*
158 conf->nr_strip_zones*mddev->raid_disks,
159 GFP_KERNEL);
160 if (!conf->devlist)
161 goto abort;
162
163 /* The first zone must contain all devices, so here we check that
164 * there is a proper alignment of slots to devices and find them all
165 */
166 zone = &conf->strip_zone[0];
167 cnt = 0;
168 smallest = NULL;
169 dev = conf->devlist;
170 err = -EINVAL;
171 rdev_for_each(rdev1, mddev) {
172 int j = rdev1->raid_disk;
173
174 if (mddev->level == 10) {
175 /* taking over a raid10-n2 array */
176 j /= 2;
177 rdev1->new_raid_disk = j;
178 }
179
180 if (mddev->level == 1) {
181 /* taiking over a raid1 array-
182 * we have only one active disk
183 */
184 j = 0;
185 rdev1->new_raid_disk = j;
186 }
187
188 if (j < 0) {
189 printk(KERN_ERR
190 "md/raid0:%s: remove inactive devices before converting to RAID0\n",
191 mdname(mddev));
192 goto abort;
193 }
194 if (j >= mddev->raid_disks) {
195 printk(KERN_ERR "md/raid0:%s: bad disk number %d - "
196 "aborting!\n", mdname(mddev), j);
197 goto abort;
198 }
199 if (dev[j]) {
200 printk(KERN_ERR "md/raid0:%s: multiple devices for %d - "
201 "aborting!\n", mdname(mddev), j);
202 goto abort;
203 }
204 dev[j] = rdev1;
205
206 if (!smallest || (rdev1->sectors < smallest->sectors))
207 smallest = rdev1;
208 cnt++;
209 }
210 if (cnt != mddev->raid_disks) {
211 printk(KERN_ERR "md/raid0:%s: too few disks (%d of %d) - "
212 "aborting!\n", mdname(mddev), cnt, mddev->raid_disks);
213 goto abort;
214 }
215 zone->nb_dev = cnt;
216 zone->zone_end = smallest->sectors * cnt;
217
218 curr_zone_end = zone->zone_end;
219
220 /* now do the other zones */
221 for (i = 1; i < conf->nr_strip_zones; i++)
222 {
223 int j;
224
225 zone = conf->strip_zone + i;
226 dev = conf->devlist + i * mddev->raid_disks;
227
228 pr_debug("md/raid0:%s: zone %d\n", mdname(mddev), i);
229 zone->dev_start = smallest->sectors;
230 smallest = NULL;
231 c = 0;
232
233 for (j=0; j<cnt; j++) {
234 rdev = conf->devlist[j];
235 if (rdev->sectors <= zone->dev_start) {
236 pr_debug("md/raid0:%s: checking %s ... nope\n",
237 mdname(mddev),
238 bdevname(rdev->bdev, b));
239 continue;
240 }
241 pr_debug("md/raid0:%s: checking %s ..."
242 " contained as device %d\n",
243 mdname(mddev),
244 bdevname(rdev->bdev, b), c);
245 dev[c] = rdev;
246 c++;
247 if (!smallest || rdev->sectors < smallest->sectors) {
248 smallest = rdev;
249 pr_debug("md/raid0:%s: (%llu) is smallest!.\n",
250 mdname(mddev),
251 (unsigned long long)rdev->sectors);
252 }
253 }
254
255 zone->nb_dev = c;
256 sectors = (smallest->sectors - zone->dev_start) * c;
257 pr_debug("md/raid0:%s: zone->nb_dev: %d, sectors: %llu\n",
258 mdname(mddev),
259 zone->nb_dev, (unsigned long long)sectors);
260
261 curr_zone_end += sectors;
262 zone->zone_end = curr_zone_end;
263
264 pr_debug("md/raid0:%s: current zone start: %llu\n",
265 mdname(mddev),
266 (unsigned long long)smallest->sectors);
267 }
268
269 pr_debug("md/raid0:%s: done.\n", mdname(mddev));
270 *private_conf = conf;
271
272 return 0;
273abort:
274 kfree(conf->strip_zone);
275 kfree(conf->devlist);
276 kfree(conf);
277 *private_conf = ERR_PTR(err);
278 return err;
279}
280
281/* Find the zone which holds a particular offset
282 * Update *sectorp to be an offset in that zone
283 */
284static struct strip_zone *find_zone(struct r0conf *conf,
285 sector_t *sectorp)
286{
287 int i;
288 struct strip_zone *z = conf->strip_zone;
289 sector_t sector = *sectorp;
290
291 for (i = 0; i < conf->nr_strip_zones; i++)
292 if (sector < z[i].zone_end) {
293 if (i)
294 *sectorp = sector - z[i-1].zone_end;
295 return z + i;
296 }
297 BUG();
298}
299
300/*
301 * remaps the bio to the target device. we separate two flows.
302 * power 2 flow and a general flow for the sake of performance
303*/
304static struct md_rdev *map_sector(struct mddev *mddev, struct strip_zone *zone,
305 sector_t sector, sector_t *sector_offset)
306{
307 unsigned int sect_in_chunk;
308 sector_t chunk;
309 struct r0conf *conf = mddev->private;
310 int raid_disks = conf->strip_zone[0].nb_dev;
311 unsigned int chunk_sects = mddev->chunk_sectors;
312
313 if (is_power_of_2(chunk_sects)) {
314 int chunksect_bits = ffz(~chunk_sects);
315 /* find the sector offset inside the chunk */
316 sect_in_chunk = sector & (chunk_sects - 1);
317 sector >>= chunksect_bits;
318 /* chunk in zone */
319 chunk = *sector_offset;
320 /* quotient is the chunk in real device*/
321 sector_div(chunk, zone->nb_dev << chunksect_bits);
322 } else{
323 sect_in_chunk = sector_div(sector, chunk_sects);
324 chunk = *sector_offset;
325 sector_div(chunk, chunk_sects * zone->nb_dev);
326 }
327 /*
328 * position the bio over the real device
329 * real sector = chunk in device + starting of zone
330 * + the position in the chunk
331 */
332 *sector_offset = (chunk * chunk_sects) + sect_in_chunk;
333 return conf->devlist[(zone - conf->strip_zone)*raid_disks
334 + sector_div(sector, zone->nb_dev)];
335}
336
337static sector_t raid0_size(struct mddev *mddev, sector_t sectors, int raid_disks)
338{
339 sector_t array_sectors = 0;
340 struct md_rdev *rdev;
341
342 WARN_ONCE(sectors || raid_disks,
343 "%s does not support generic reshape\n", __func__);
344
345 rdev_for_each(rdev, mddev)
346 array_sectors += (rdev->sectors &
347 ~(sector_t)(mddev->chunk_sectors-1));
348
349 return array_sectors;
350}
351
352static void raid0_free(struct mddev *mddev, void *priv);
353
354static int raid0_run(struct mddev *mddev)
355{
356 struct r0conf *conf;
357 int ret;
358
359 if (mddev->chunk_sectors == 0) {
360 printk(KERN_ERR "md/raid0:%s: chunk size must be set.\n",
361 mdname(mddev));
362 return -EINVAL;
363 }
364 if (md_check_no_bitmap(mddev))
365 return -EINVAL;
366
367 /* if private is not null, we are here after takeover */
368 if (mddev->private == NULL) {
369 ret = create_strip_zones(mddev, &conf);
370 if (ret < 0)
371 return ret;
372 mddev->private = conf;
373 }
374 conf = mddev->private;
375 if (mddev->queue) {
376 struct md_rdev *rdev;
377 bool discard_supported = false;
378
379 blk_queue_max_hw_sectors(mddev->queue, mddev->chunk_sectors);
380 blk_queue_max_write_same_sectors(mddev->queue, mddev->chunk_sectors);
381 blk_queue_max_discard_sectors(mddev->queue, mddev->chunk_sectors);
382
383 blk_queue_io_min(mddev->queue, mddev->chunk_sectors << 9);
384 blk_queue_io_opt(mddev->queue,
385 (mddev->chunk_sectors << 9) * mddev->raid_disks);
386
387 rdev_for_each(rdev, mddev) {
388 disk_stack_limits(mddev->gendisk, rdev->bdev,
389 rdev->data_offset << 9);
390 if (blk_queue_discard(bdev_get_queue(rdev->bdev)))
391 discard_supported = true;
392 }
393 if (!discard_supported)
394 queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, mddev->queue);
395 else
396 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, mddev->queue);
397 }
398
399 /* calculate array device size */
400 md_set_array_sectors(mddev, raid0_size(mddev, 0, 0));
401
402 printk(KERN_INFO "md/raid0:%s: md_size is %llu sectors.\n",
403 mdname(mddev),
404 (unsigned long long)mddev->array_sectors);
405
406 if (mddev->queue) {
407 /* calculate the max read-ahead size.
408 * For read-ahead of large files to be effective, we need to
409 * readahead at least twice a whole stripe. i.e. number of devices
410 * multiplied by chunk size times 2.
411 * If an individual device has an ra_pages greater than the
412 * chunk size, then we will not drive that device as hard as it
413 * wants. We consider this a configuration error: a larger
414 * chunksize should be used in that case.
415 */
416 int stripe = mddev->raid_disks *
417 (mddev->chunk_sectors << 9) / PAGE_SIZE;
418 if (mddev->queue->backing_dev_info.ra_pages < 2* stripe)
419 mddev->queue->backing_dev_info.ra_pages = 2* stripe;
420 }
421
422 dump_zones(mddev);
423
424 ret = md_integrity_register(mddev);
425
426 return ret;
427}
428
429static void raid0_free(struct mddev *mddev, void *priv)
430{
431 struct r0conf *conf = priv;
432
433 kfree(conf->strip_zone);
434 kfree(conf->devlist);
435 kfree(conf);
436}
437
438/*
439 * Is io distribute over 1 or more chunks ?
440*/
441static inline int is_io_in_chunk_boundary(struct mddev *mddev,
442 unsigned int chunk_sects, struct bio *bio)
443{
444 if (likely(is_power_of_2(chunk_sects))) {
445 return chunk_sects >=
446 ((bio->bi_iter.bi_sector & (chunk_sects-1))
447 + bio_sectors(bio));
448 } else{
449 sector_t sector = bio->bi_iter.bi_sector;
450 return chunk_sects >= (sector_div(sector, chunk_sects)
451 + bio_sectors(bio));
452 }
453}
454
455static void raid0_make_request(struct mddev *mddev, struct bio *bio)
456{
457 struct strip_zone *zone;
458 struct md_rdev *tmp_dev;
459 struct bio *split;
460
461 if (unlikely(bio->bi_rw & REQ_FLUSH)) {
462 md_flush_request(mddev, bio);
463 return;
464 }
465
466 do {
467 sector_t sector = bio->bi_iter.bi_sector;
468 unsigned chunk_sects = mddev->chunk_sectors;
469
470 unsigned sectors = chunk_sects -
471 (likely(is_power_of_2(chunk_sects))
472 ? (sector & (chunk_sects-1))
473 : sector_div(sector, chunk_sects));
474
475 /* Restore due to sector_div */
476 sector = bio->bi_iter.bi_sector;
477
478 if (sectors < bio_sectors(bio)) {
479 split = bio_split(bio, sectors, GFP_NOIO, fs_bio_set);
480 bio_chain(split, bio);
481 } else {
482 split = bio;
483 }
484
485 zone = find_zone(mddev->private, §or);
486 tmp_dev = map_sector(mddev, zone, sector, §or);
487 split->bi_bdev = tmp_dev->bdev;
488 split->bi_iter.bi_sector = sector + zone->dev_start +
489 tmp_dev->data_offset;
490
491 if (unlikely((split->bi_rw & REQ_DISCARD) &&
492 !blk_queue_discard(bdev_get_queue(split->bi_bdev)))) {
493 /* Just ignore it */
494 bio_endio(split);
495 } else
496 generic_make_request(split);
497 } while (split != bio);
498}
499
500static void raid0_status(struct seq_file *seq, struct mddev *mddev)
501{
502 seq_printf(seq, " %dk chunks", mddev->chunk_sectors / 2);
503 return;
504}
505
506static void *raid0_takeover_raid45(struct mddev *mddev)
507{
508 struct md_rdev *rdev;
509 struct r0conf *priv_conf;
510
511 if (mddev->degraded != 1) {
512 printk(KERN_ERR "md/raid0:%s: raid5 must be degraded! Degraded disks: %d\n",
513 mdname(mddev),
514 mddev->degraded);
515 return ERR_PTR(-EINVAL);
516 }
517
518 rdev_for_each(rdev, mddev) {
519 /* check slot number for a disk */
520 if (rdev->raid_disk == mddev->raid_disks-1) {
521 printk(KERN_ERR "md/raid0:%s: raid5 must have missing parity disk!\n",
522 mdname(mddev));
523 return ERR_PTR(-EINVAL);
524 }
525 rdev->sectors = mddev->dev_sectors;
526 }
527
528 /* Set new parameters */
529 mddev->new_level = 0;
530 mddev->new_layout = 0;
531 mddev->new_chunk_sectors = mddev->chunk_sectors;
532 mddev->raid_disks--;
533 mddev->delta_disks = -1;
534 /* make sure it will be not marked as dirty */
535 mddev->recovery_cp = MaxSector;
536
537 create_strip_zones(mddev, &priv_conf);
538 return priv_conf;
539}
540
541static void *raid0_takeover_raid10(struct mddev *mddev)
542{
543 struct r0conf *priv_conf;
544
545 /* Check layout:
546 * - far_copies must be 1
547 * - near_copies must be 2
548 * - disks number must be even
549 * - all mirrors must be already degraded
550 */
551 if (mddev->layout != ((1 << 8) + 2)) {
552 printk(KERN_ERR "md/raid0:%s:: Raid0 cannot takeover layout: 0x%x\n",
553 mdname(mddev),
554 mddev->layout);
555 return ERR_PTR(-EINVAL);
556 }
557 if (mddev->raid_disks & 1) {
558 printk(KERN_ERR "md/raid0:%s: Raid0 cannot takeover Raid10 with odd disk number.\n",
559 mdname(mddev));
560 return ERR_PTR(-EINVAL);
561 }
562 if (mddev->degraded != (mddev->raid_disks>>1)) {
563 printk(KERN_ERR "md/raid0:%s: All mirrors must be already degraded!\n",
564 mdname(mddev));
565 return ERR_PTR(-EINVAL);
566 }
567
568 /* Set new parameters */
569 mddev->new_level = 0;
570 mddev->new_layout = 0;
571 mddev->new_chunk_sectors = mddev->chunk_sectors;
572 mddev->delta_disks = - mddev->raid_disks / 2;
573 mddev->raid_disks += mddev->delta_disks;
574 mddev->degraded = 0;
575 /* make sure it will be not marked as dirty */
576 mddev->recovery_cp = MaxSector;
577
578 create_strip_zones(mddev, &priv_conf);
579 return priv_conf;
580}
581
582static void *raid0_takeover_raid1(struct mddev *mddev)
583{
584 struct r0conf *priv_conf;
585 int chunksect;
586
587 /* Check layout:
588 * - (N - 1) mirror drives must be already faulty
589 */
590 if ((mddev->raid_disks - 1) != mddev->degraded) {
591 printk(KERN_ERR "md/raid0:%s: (N - 1) mirrors drives must be already faulty!\n",
592 mdname(mddev));
593 return ERR_PTR(-EINVAL);
594 }
595
596 /*
597 * a raid1 doesn't have the notion of chunk size, so
598 * figure out the largest suitable size we can use.
599 */
600 chunksect = 64 * 2; /* 64K by default */
601
602 /* The array must be an exact multiple of chunksize */
603 while (chunksect && (mddev->array_sectors & (chunksect - 1)))
604 chunksect >>= 1;
605
606 if ((chunksect << 9) < PAGE_SIZE)
607 /* array size does not allow a suitable chunk size */
608 return ERR_PTR(-EINVAL);
609
610 /* Set new parameters */
611 mddev->new_level = 0;
612 mddev->new_layout = 0;
613 mddev->new_chunk_sectors = chunksect;
614 mddev->chunk_sectors = chunksect;
615 mddev->delta_disks = 1 - mddev->raid_disks;
616 mddev->raid_disks = 1;
617 /* make sure it will be not marked as dirty */
618 mddev->recovery_cp = MaxSector;
619
620 create_strip_zones(mddev, &priv_conf);
621 return priv_conf;
622}
623
624static void *raid0_takeover(struct mddev *mddev)
625{
626 /* raid0 can take over:
627 * raid4 - if all data disks are active.
628 * raid5 - providing it is Raid4 layout and one disk is faulty
629 * raid10 - assuming we have all necessary active disks
630 * raid1 - with (N -1) mirror drives faulty
631 */
632
633 if (mddev->bitmap) {
634 printk(KERN_ERR "md/raid0: %s: cannot takeover array with bitmap\n",
635 mdname(mddev));
636 return ERR_PTR(-EBUSY);
637 }
638 if (mddev->level == 4)
639 return raid0_takeover_raid45(mddev);
640
641 if (mddev->level == 5) {
642 if (mddev->layout == ALGORITHM_PARITY_N)
643 return raid0_takeover_raid45(mddev);
644
645 printk(KERN_ERR "md/raid0:%s: Raid can only takeover Raid5 with layout: %d\n",
646 mdname(mddev), ALGORITHM_PARITY_N);
647 }
648
649 if (mddev->level == 10)
650 return raid0_takeover_raid10(mddev);
651
652 if (mddev->level == 1)
653 return raid0_takeover_raid1(mddev);
654
655 printk(KERN_ERR "Takeover from raid%i to raid0 not supported\n",
656 mddev->level);
657
658 return ERR_PTR(-EINVAL);
659}
660
661static void raid0_quiesce(struct mddev *mddev, int state)
662{
663}
664
665static struct md_personality raid0_personality=
666{
667 .name = "raid0",
668 .level = 0,
669 .owner = THIS_MODULE,
670 .make_request = raid0_make_request,
671 .run = raid0_run,
672 .free = raid0_free,
673 .status = raid0_status,
674 .size = raid0_size,
675 .takeover = raid0_takeover,
676 .quiesce = raid0_quiesce,
677 .congested = raid0_congested,
678};
679
680static int __init raid0_init (void)
681{
682 return register_md_personality (&raid0_personality);
683}
684
685static void raid0_exit (void)
686{
687 unregister_md_personality (&raid0_personality);
688}
689
690module_init(raid0_init);
691module_exit(raid0_exit);
692MODULE_LICENSE("GPL");
693MODULE_DESCRIPTION("RAID0 (striping) personality for MD");
694MODULE_ALIAS("md-personality-2"); /* RAID0 */
695MODULE_ALIAS("md-raid0");
696MODULE_ALIAS("md-level-0");
1// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 raid0.c : Multiple Devices driver for Linux
4 Copyright (C) 1994-96 Marc ZYNGIER
5 <zyngier@ufr-info-p7.ibp.fr> or
6 <maz@gloups.fdn.fr>
7 Copyright (C) 1999, 2000 Ingo Molnar, Red Hat
8
9 RAID-0 management functions.
10
11*/
12
13#include <linux/blkdev.h>
14#include <linux/seq_file.h>
15#include <linux/module.h>
16#include <linux/slab.h>
17#include <trace/events/block.h>
18#include "md.h"
19#include "raid0.h"
20#include "raid5.h"
21
22static int default_layout = 0;
23module_param(default_layout, int, 0644);
24
25#define UNSUPPORTED_MDDEV_FLAGS \
26 ((1L << MD_HAS_JOURNAL) | \
27 (1L << MD_JOURNAL_CLEAN) | \
28 (1L << MD_FAILFAST_SUPPORTED) |\
29 (1L << MD_HAS_PPL) | \
30 (1L << MD_HAS_MULTIPLE_PPLS))
31
32static int raid0_congested(struct mddev *mddev, int bits)
33{
34 struct r0conf *conf = mddev->private;
35 struct md_rdev **devlist = conf->devlist;
36 int raid_disks = conf->strip_zone[0].nb_dev;
37 int i, ret = 0;
38
39 for (i = 0; i < raid_disks && !ret ; i++) {
40 struct request_queue *q = bdev_get_queue(devlist[i]->bdev);
41
42 ret |= bdi_congested(q->backing_dev_info, bits);
43 }
44 return ret;
45}
46
47/*
48 * inform the user of the raid configuration
49*/
50static void dump_zones(struct mddev *mddev)
51{
52 int j, k;
53 sector_t zone_size = 0;
54 sector_t zone_start = 0;
55 char b[BDEVNAME_SIZE];
56 struct r0conf *conf = mddev->private;
57 int raid_disks = conf->strip_zone[0].nb_dev;
58 pr_debug("md: RAID0 configuration for %s - %d zone%s\n",
59 mdname(mddev),
60 conf->nr_strip_zones, conf->nr_strip_zones==1?"":"s");
61 for (j = 0; j < conf->nr_strip_zones; j++) {
62 char line[200];
63 int len = 0;
64
65 for (k = 0; k < conf->strip_zone[j].nb_dev; k++)
66 len += snprintf(line+len, 200-len, "%s%s", k?"/":"",
67 bdevname(conf->devlist[j*raid_disks
68 + k]->bdev, b));
69 pr_debug("md: zone%d=[%s]\n", j, line);
70
71 zone_size = conf->strip_zone[j].zone_end - zone_start;
72 pr_debug(" zone-offset=%10lluKB, device-offset=%10lluKB, size=%10lluKB\n",
73 (unsigned long long)zone_start>>1,
74 (unsigned long long)conf->strip_zone[j].dev_start>>1,
75 (unsigned long long)zone_size>>1);
76 zone_start = conf->strip_zone[j].zone_end;
77 }
78}
79
80static int create_strip_zones(struct mddev *mddev, struct r0conf **private_conf)
81{
82 int i, c, err;
83 sector_t curr_zone_end, sectors;
84 struct md_rdev *smallest, *rdev1, *rdev2, *rdev, **dev;
85 struct strip_zone *zone;
86 int cnt;
87 char b[BDEVNAME_SIZE];
88 char b2[BDEVNAME_SIZE];
89 struct r0conf *conf = kzalloc(sizeof(*conf), GFP_KERNEL);
90 unsigned short blksize = 512;
91
92 *private_conf = ERR_PTR(-ENOMEM);
93 if (!conf)
94 return -ENOMEM;
95 rdev_for_each(rdev1, mddev) {
96 pr_debug("md/raid0:%s: looking at %s\n",
97 mdname(mddev),
98 bdevname(rdev1->bdev, b));
99 c = 0;
100
101 /* round size to chunk_size */
102 sectors = rdev1->sectors;
103 sector_div(sectors, mddev->chunk_sectors);
104 rdev1->sectors = sectors * mddev->chunk_sectors;
105
106 blksize = max(blksize, queue_logical_block_size(
107 rdev1->bdev->bd_disk->queue));
108
109 rdev_for_each(rdev2, mddev) {
110 pr_debug("md/raid0:%s: comparing %s(%llu)"
111 " with %s(%llu)\n",
112 mdname(mddev),
113 bdevname(rdev1->bdev,b),
114 (unsigned long long)rdev1->sectors,
115 bdevname(rdev2->bdev,b2),
116 (unsigned long long)rdev2->sectors);
117 if (rdev2 == rdev1) {
118 pr_debug("md/raid0:%s: END\n",
119 mdname(mddev));
120 break;
121 }
122 if (rdev2->sectors == rdev1->sectors) {
123 /*
124 * Not unique, don't count it as a new
125 * group
126 */
127 pr_debug("md/raid0:%s: EQUAL\n",
128 mdname(mddev));
129 c = 1;
130 break;
131 }
132 pr_debug("md/raid0:%s: NOT EQUAL\n",
133 mdname(mddev));
134 }
135 if (!c) {
136 pr_debug("md/raid0:%s: ==> UNIQUE\n",
137 mdname(mddev));
138 conf->nr_strip_zones++;
139 pr_debug("md/raid0:%s: %d zones\n",
140 mdname(mddev), conf->nr_strip_zones);
141 }
142 }
143 pr_debug("md/raid0:%s: FINAL %d zones\n",
144 mdname(mddev), conf->nr_strip_zones);
145
146 if (conf->nr_strip_zones == 1) {
147 conf->layout = RAID0_ORIG_LAYOUT;
148 } else if (mddev->layout == RAID0_ORIG_LAYOUT ||
149 mddev->layout == RAID0_ALT_MULTIZONE_LAYOUT) {
150 conf->layout = mddev->layout;
151 } else if (default_layout == RAID0_ORIG_LAYOUT ||
152 default_layout == RAID0_ALT_MULTIZONE_LAYOUT) {
153 conf->layout = default_layout;
154 } else {
155 pr_err("md/raid0:%s: cannot assemble multi-zone RAID0 with default_layout setting\n",
156 mdname(mddev));
157 pr_err("md/raid0: please set raid0.default_layout to 1 or 2\n");
158 err = -ENOTSUPP;
159 goto abort;
160 }
161 /*
162 * now since we have the hard sector sizes, we can make sure
163 * chunk size is a multiple of that sector size
164 */
165 if ((mddev->chunk_sectors << 9) % blksize) {
166 pr_warn("md/raid0:%s: chunk_size of %d not multiple of block size %d\n",
167 mdname(mddev),
168 mddev->chunk_sectors << 9, blksize);
169 err = -EINVAL;
170 goto abort;
171 }
172
173 err = -ENOMEM;
174 conf->strip_zone = kcalloc(conf->nr_strip_zones,
175 sizeof(struct strip_zone),
176 GFP_KERNEL);
177 if (!conf->strip_zone)
178 goto abort;
179 conf->devlist = kzalloc(array3_size(sizeof(struct md_rdev *),
180 conf->nr_strip_zones,
181 mddev->raid_disks),
182 GFP_KERNEL);
183 if (!conf->devlist)
184 goto abort;
185
186 /* The first zone must contain all devices, so here we check that
187 * there is a proper alignment of slots to devices and find them all
188 */
189 zone = &conf->strip_zone[0];
190 cnt = 0;
191 smallest = NULL;
192 dev = conf->devlist;
193 err = -EINVAL;
194 rdev_for_each(rdev1, mddev) {
195 int j = rdev1->raid_disk;
196
197 if (mddev->level == 10) {
198 /* taking over a raid10-n2 array */
199 j /= 2;
200 rdev1->new_raid_disk = j;
201 }
202
203 if (mddev->level == 1) {
204 /* taiking over a raid1 array-
205 * we have only one active disk
206 */
207 j = 0;
208 rdev1->new_raid_disk = j;
209 }
210
211 if (j < 0) {
212 pr_warn("md/raid0:%s: remove inactive devices before converting to RAID0\n",
213 mdname(mddev));
214 goto abort;
215 }
216 if (j >= mddev->raid_disks) {
217 pr_warn("md/raid0:%s: bad disk number %d - aborting!\n",
218 mdname(mddev), j);
219 goto abort;
220 }
221 if (dev[j]) {
222 pr_warn("md/raid0:%s: multiple devices for %d - aborting!\n",
223 mdname(mddev), j);
224 goto abort;
225 }
226 dev[j] = rdev1;
227
228 if (!smallest || (rdev1->sectors < smallest->sectors))
229 smallest = rdev1;
230 cnt++;
231 }
232 if (cnt != mddev->raid_disks) {
233 pr_warn("md/raid0:%s: too few disks (%d of %d) - aborting!\n",
234 mdname(mddev), cnt, mddev->raid_disks);
235 goto abort;
236 }
237 zone->nb_dev = cnt;
238 zone->zone_end = smallest->sectors * cnt;
239
240 curr_zone_end = zone->zone_end;
241
242 /* now do the other zones */
243 for (i = 1; i < conf->nr_strip_zones; i++)
244 {
245 int j;
246
247 zone = conf->strip_zone + i;
248 dev = conf->devlist + i * mddev->raid_disks;
249
250 pr_debug("md/raid0:%s: zone %d\n", mdname(mddev), i);
251 zone->dev_start = smallest->sectors;
252 smallest = NULL;
253 c = 0;
254
255 for (j=0; j<cnt; j++) {
256 rdev = conf->devlist[j];
257 if (rdev->sectors <= zone->dev_start) {
258 pr_debug("md/raid0:%s: checking %s ... nope\n",
259 mdname(mddev),
260 bdevname(rdev->bdev, b));
261 continue;
262 }
263 pr_debug("md/raid0:%s: checking %s ..."
264 " contained as device %d\n",
265 mdname(mddev),
266 bdevname(rdev->bdev, b), c);
267 dev[c] = rdev;
268 c++;
269 if (!smallest || rdev->sectors < smallest->sectors) {
270 smallest = rdev;
271 pr_debug("md/raid0:%s: (%llu) is smallest!.\n",
272 mdname(mddev),
273 (unsigned long long)rdev->sectors);
274 }
275 }
276
277 zone->nb_dev = c;
278 sectors = (smallest->sectors - zone->dev_start) * c;
279 pr_debug("md/raid0:%s: zone->nb_dev: %d, sectors: %llu\n",
280 mdname(mddev),
281 zone->nb_dev, (unsigned long long)sectors);
282
283 curr_zone_end += sectors;
284 zone->zone_end = curr_zone_end;
285
286 pr_debug("md/raid0:%s: current zone start: %llu\n",
287 mdname(mddev),
288 (unsigned long long)smallest->sectors);
289 }
290
291 pr_debug("md/raid0:%s: done.\n", mdname(mddev));
292 *private_conf = conf;
293
294 return 0;
295abort:
296 kfree(conf->strip_zone);
297 kfree(conf->devlist);
298 kfree(conf);
299 *private_conf = ERR_PTR(err);
300 return err;
301}
302
303/* Find the zone which holds a particular offset
304 * Update *sectorp to be an offset in that zone
305 */
306static struct strip_zone *find_zone(struct r0conf *conf,
307 sector_t *sectorp)
308{
309 int i;
310 struct strip_zone *z = conf->strip_zone;
311 sector_t sector = *sectorp;
312
313 for (i = 0; i < conf->nr_strip_zones; i++)
314 if (sector < z[i].zone_end) {
315 if (i)
316 *sectorp = sector - z[i-1].zone_end;
317 return z + i;
318 }
319 BUG();
320}
321
322/*
323 * remaps the bio to the target device. we separate two flows.
324 * power 2 flow and a general flow for the sake of performance
325*/
326static struct md_rdev *map_sector(struct mddev *mddev, struct strip_zone *zone,
327 sector_t sector, sector_t *sector_offset)
328{
329 unsigned int sect_in_chunk;
330 sector_t chunk;
331 struct r0conf *conf = mddev->private;
332 int raid_disks = conf->strip_zone[0].nb_dev;
333 unsigned int chunk_sects = mddev->chunk_sectors;
334
335 if (is_power_of_2(chunk_sects)) {
336 int chunksect_bits = ffz(~chunk_sects);
337 /* find the sector offset inside the chunk */
338 sect_in_chunk = sector & (chunk_sects - 1);
339 sector >>= chunksect_bits;
340 /* chunk in zone */
341 chunk = *sector_offset;
342 /* quotient is the chunk in real device*/
343 sector_div(chunk, zone->nb_dev << chunksect_bits);
344 } else{
345 sect_in_chunk = sector_div(sector, chunk_sects);
346 chunk = *sector_offset;
347 sector_div(chunk, chunk_sects * zone->nb_dev);
348 }
349 /*
350 * position the bio over the real device
351 * real sector = chunk in device + starting of zone
352 * + the position in the chunk
353 */
354 *sector_offset = (chunk * chunk_sects) + sect_in_chunk;
355 return conf->devlist[(zone - conf->strip_zone)*raid_disks
356 + sector_div(sector, zone->nb_dev)];
357}
358
359static sector_t raid0_size(struct mddev *mddev, sector_t sectors, int raid_disks)
360{
361 sector_t array_sectors = 0;
362 struct md_rdev *rdev;
363
364 WARN_ONCE(sectors || raid_disks,
365 "%s does not support generic reshape\n", __func__);
366
367 rdev_for_each(rdev, mddev)
368 array_sectors += (rdev->sectors &
369 ~(sector_t)(mddev->chunk_sectors-1));
370
371 return array_sectors;
372}
373
374static void raid0_free(struct mddev *mddev, void *priv);
375
376static int raid0_run(struct mddev *mddev)
377{
378 struct r0conf *conf;
379 int ret;
380
381 if (mddev->chunk_sectors == 0) {
382 pr_warn("md/raid0:%s: chunk size must be set.\n", mdname(mddev));
383 return -EINVAL;
384 }
385 if (md_check_no_bitmap(mddev))
386 return -EINVAL;
387
388 /* if private is not null, we are here after takeover */
389 if (mddev->private == NULL) {
390 ret = create_strip_zones(mddev, &conf);
391 if (ret < 0)
392 return ret;
393 mddev->private = conf;
394 }
395 conf = mddev->private;
396 if (mddev->queue) {
397 struct md_rdev *rdev;
398 bool discard_supported = false;
399
400 blk_queue_max_hw_sectors(mddev->queue, mddev->chunk_sectors);
401 blk_queue_max_write_same_sectors(mddev->queue, mddev->chunk_sectors);
402 blk_queue_max_write_zeroes_sectors(mddev->queue, mddev->chunk_sectors);
403 blk_queue_max_discard_sectors(mddev->queue, UINT_MAX);
404
405 blk_queue_io_min(mddev->queue, mddev->chunk_sectors << 9);
406 blk_queue_io_opt(mddev->queue,
407 (mddev->chunk_sectors << 9) * mddev->raid_disks);
408
409 rdev_for_each(rdev, mddev) {
410 disk_stack_limits(mddev->gendisk, rdev->bdev,
411 rdev->data_offset << 9);
412 if (blk_queue_discard(bdev_get_queue(rdev->bdev)))
413 discard_supported = true;
414 }
415 if (!discard_supported)
416 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, mddev->queue);
417 else
418 blk_queue_flag_set(QUEUE_FLAG_DISCARD, mddev->queue);
419 }
420
421 /* calculate array device size */
422 md_set_array_sectors(mddev, raid0_size(mddev, 0, 0));
423
424 pr_debug("md/raid0:%s: md_size is %llu sectors.\n",
425 mdname(mddev),
426 (unsigned long long)mddev->array_sectors);
427
428 if (mddev->queue) {
429 /* calculate the max read-ahead size.
430 * For read-ahead of large files to be effective, we need to
431 * readahead at least twice a whole stripe. i.e. number of devices
432 * multiplied by chunk size times 2.
433 * If an individual device has an ra_pages greater than the
434 * chunk size, then we will not drive that device as hard as it
435 * wants. We consider this a configuration error: a larger
436 * chunksize should be used in that case.
437 */
438 int stripe = mddev->raid_disks *
439 (mddev->chunk_sectors << 9) / PAGE_SIZE;
440 if (mddev->queue->backing_dev_info->ra_pages < 2* stripe)
441 mddev->queue->backing_dev_info->ra_pages = 2* stripe;
442 }
443
444 dump_zones(mddev);
445
446 ret = md_integrity_register(mddev);
447
448 return ret;
449}
450
451static void raid0_free(struct mddev *mddev, void *priv)
452{
453 struct r0conf *conf = priv;
454
455 kfree(conf->strip_zone);
456 kfree(conf->devlist);
457 kfree(conf);
458}
459
460/*
461 * Is io distribute over 1 or more chunks ?
462*/
463static inline int is_io_in_chunk_boundary(struct mddev *mddev,
464 unsigned int chunk_sects, struct bio *bio)
465{
466 if (likely(is_power_of_2(chunk_sects))) {
467 return chunk_sects >=
468 ((bio->bi_iter.bi_sector & (chunk_sects-1))
469 + bio_sectors(bio));
470 } else{
471 sector_t sector = bio->bi_iter.bi_sector;
472 return chunk_sects >= (sector_div(sector, chunk_sects)
473 + bio_sectors(bio));
474 }
475}
476
477static void raid0_handle_discard(struct mddev *mddev, struct bio *bio)
478{
479 struct r0conf *conf = mddev->private;
480 struct strip_zone *zone;
481 sector_t start = bio->bi_iter.bi_sector;
482 sector_t end;
483 unsigned int stripe_size;
484 sector_t first_stripe_index, last_stripe_index;
485 sector_t start_disk_offset;
486 unsigned int start_disk_index;
487 sector_t end_disk_offset;
488 unsigned int end_disk_index;
489 unsigned int disk;
490
491 zone = find_zone(conf, &start);
492
493 if (bio_end_sector(bio) > zone->zone_end) {
494 struct bio *split = bio_split(bio,
495 zone->zone_end - bio->bi_iter.bi_sector, GFP_NOIO,
496 &mddev->bio_set);
497 bio_chain(split, bio);
498 generic_make_request(bio);
499 bio = split;
500 end = zone->zone_end;
501 } else
502 end = bio_end_sector(bio);
503
504 if (zone != conf->strip_zone)
505 end = end - zone[-1].zone_end;
506
507 /* Now start and end is the offset in zone */
508 stripe_size = zone->nb_dev * mddev->chunk_sectors;
509
510 first_stripe_index = start;
511 sector_div(first_stripe_index, stripe_size);
512 last_stripe_index = end;
513 sector_div(last_stripe_index, stripe_size);
514
515 start_disk_index = (int)(start - first_stripe_index * stripe_size) /
516 mddev->chunk_sectors;
517 start_disk_offset = ((int)(start - first_stripe_index * stripe_size) %
518 mddev->chunk_sectors) +
519 first_stripe_index * mddev->chunk_sectors;
520 end_disk_index = (int)(end - last_stripe_index * stripe_size) /
521 mddev->chunk_sectors;
522 end_disk_offset = ((int)(end - last_stripe_index * stripe_size) %
523 mddev->chunk_sectors) +
524 last_stripe_index * mddev->chunk_sectors;
525
526 for (disk = 0; disk < zone->nb_dev; disk++) {
527 sector_t dev_start, dev_end;
528 struct bio *discard_bio = NULL;
529 struct md_rdev *rdev;
530
531 if (disk < start_disk_index)
532 dev_start = (first_stripe_index + 1) *
533 mddev->chunk_sectors;
534 else if (disk > start_disk_index)
535 dev_start = first_stripe_index * mddev->chunk_sectors;
536 else
537 dev_start = start_disk_offset;
538
539 if (disk < end_disk_index)
540 dev_end = (last_stripe_index + 1) * mddev->chunk_sectors;
541 else if (disk > end_disk_index)
542 dev_end = last_stripe_index * mddev->chunk_sectors;
543 else
544 dev_end = end_disk_offset;
545
546 if (dev_end <= dev_start)
547 continue;
548
549 rdev = conf->devlist[(zone - conf->strip_zone) *
550 conf->strip_zone[0].nb_dev + disk];
551 if (__blkdev_issue_discard(rdev->bdev,
552 dev_start + zone->dev_start + rdev->data_offset,
553 dev_end - dev_start, GFP_NOIO, 0, &discard_bio) ||
554 !discard_bio)
555 continue;
556 bio_chain(discard_bio, bio);
557 bio_clone_blkg_association(discard_bio, bio);
558 if (mddev->gendisk)
559 trace_block_bio_remap(bdev_get_queue(rdev->bdev),
560 discard_bio, disk_devt(mddev->gendisk),
561 bio->bi_iter.bi_sector);
562 generic_make_request(discard_bio);
563 }
564 bio_endio(bio);
565}
566
567static bool raid0_make_request(struct mddev *mddev, struct bio *bio)
568{
569 struct r0conf *conf = mddev->private;
570 struct strip_zone *zone;
571 struct md_rdev *tmp_dev;
572 sector_t bio_sector;
573 sector_t sector;
574 sector_t orig_sector;
575 unsigned chunk_sects;
576 unsigned sectors;
577
578 if (unlikely(bio->bi_opf & REQ_PREFLUSH)) {
579 md_flush_request(mddev, bio);
580 return true;
581 }
582
583 if (unlikely((bio_op(bio) == REQ_OP_DISCARD))) {
584 raid0_handle_discard(mddev, bio);
585 return true;
586 }
587
588 bio_sector = bio->bi_iter.bi_sector;
589 sector = bio_sector;
590 chunk_sects = mddev->chunk_sectors;
591
592 sectors = chunk_sects -
593 (likely(is_power_of_2(chunk_sects))
594 ? (sector & (chunk_sects-1))
595 : sector_div(sector, chunk_sects));
596
597 /* Restore due to sector_div */
598 sector = bio_sector;
599
600 if (sectors < bio_sectors(bio)) {
601 struct bio *split = bio_split(bio, sectors, GFP_NOIO,
602 &mddev->bio_set);
603 bio_chain(split, bio);
604 generic_make_request(bio);
605 bio = split;
606 }
607
608 orig_sector = sector;
609 zone = find_zone(mddev->private, §or);
610 switch (conf->layout) {
611 case RAID0_ORIG_LAYOUT:
612 tmp_dev = map_sector(mddev, zone, orig_sector, §or);
613 break;
614 case RAID0_ALT_MULTIZONE_LAYOUT:
615 tmp_dev = map_sector(mddev, zone, sector, §or);
616 break;
617 default:
618 WARN("md/raid0:%s: Invalid layout\n", mdname(mddev));
619 bio_io_error(bio);
620 return true;
621 }
622
623 if (unlikely(is_mddev_broken(tmp_dev, "raid0"))) {
624 bio_io_error(bio);
625 return true;
626 }
627
628 bio_set_dev(bio, tmp_dev->bdev);
629 bio->bi_iter.bi_sector = sector + zone->dev_start +
630 tmp_dev->data_offset;
631
632 if (mddev->gendisk)
633 trace_block_bio_remap(bio->bi_disk->queue, bio,
634 disk_devt(mddev->gendisk), bio_sector);
635 mddev_check_writesame(mddev, bio);
636 mddev_check_write_zeroes(mddev, bio);
637 generic_make_request(bio);
638 return true;
639}
640
641static void raid0_status(struct seq_file *seq, struct mddev *mddev)
642{
643 seq_printf(seq, " %dk chunks", mddev->chunk_sectors / 2);
644 return;
645}
646
647static void *raid0_takeover_raid45(struct mddev *mddev)
648{
649 struct md_rdev *rdev;
650 struct r0conf *priv_conf;
651
652 if (mddev->degraded != 1) {
653 pr_warn("md/raid0:%s: raid5 must be degraded! Degraded disks: %d\n",
654 mdname(mddev),
655 mddev->degraded);
656 return ERR_PTR(-EINVAL);
657 }
658
659 rdev_for_each(rdev, mddev) {
660 /* check slot number for a disk */
661 if (rdev->raid_disk == mddev->raid_disks-1) {
662 pr_warn("md/raid0:%s: raid5 must have missing parity disk!\n",
663 mdname(mddev));
664 return ERR_PTR(-EINVAL);
665 }
666 rdev->sectors = mddev->dev_sectors;
667 }
668
669 /* Set new parameters */
670 mddev->new_level = 0;
671 mddev->new_layout = 0;
672 mddev->new_chunk_sectors = mddev->chunk_sectors;
673 mddev->raid_disks--;
674 mddev->delta_disks = -1;
675 /* make sure it will be not marked as dirty */
676 mddev->recovery_cp = MaxSector;
677 mddev_clear_unsupported_flags(mddev, UNSUPPORTED_MDDEV_FLAGS);
678
679 create_strip_zones(mddev, &priv_conf);
680
681 return priv_conf;
682}
683
684static void *raid0_takeover_raid10(struct mddev *mddev)
685{
686 struct r0conf *priv_conf;
687
688 /* Check layout:
689 * - far_copies must be 1
690 * - near_copies must be 2
691 * - disks number must be even
692 * - all mirrors must be already degraded
693 */
694 if (mddev->layout != ((1 << 8) + 2)) {
695 pr_warn("md/raid0:%s:: Raid0 cannot takeover layout: 0x%x\n",
696 mdname(mddev),
697 mddev->layout);
698 return ERR_PTR(-EINVAL);
699 }
700 if (mddev->raid_disks & 1) {
701 pr_warn("md/raid0:%s: Raid0 cannot takeover Raid10 with odd disk number.\n",
702 mdname(mddev));
703 return ERR_PTR(-EINVAL);
704 }
705 if (mddev->degraded != (mddev->raid_disks>>1)) {
706 pr_warn("md/raid0:%s: All mirrors must be already degraded!\n",
707 mdname(mddev));
708 return ERR_PTR(-EINVAL);
709 }
710
711 /* Set new parameters */
712 mddev->new_level = 0;
713 mddev->new_layout = 0;
714 mddev->new_chunk_sectors = mddev->chunk_sectors;
715 mddev->delta_disks = - mddev->raid_disks / 2;
716 mddev->raid_disks += mddev->delta_disks;
717 mddev->degraded = 0;
718 /* make sure it will be not marked as dirty */
719 mddev->recovery_cp = MaxSector;
720 mddev_clear_unsupported_flags(mddev, UNSUPPORTED_MDDEV_FLAGS);
721
722 create_strip_zones(mddev, &priv_conf);
723 return priv_conf;
724}
725
726static void *raid0_takeover_raid1(struct mddev *mddev)
727{
728 struct r0conf *priv_conf;
729 int chunksect;
730
731 /* Check layout:
732 * - (N - 1) mirror drives must be already faulty
733 */
734 if ((mddev->raid_disks - 1) != mddev->degraded) {
735 pr_err("md/raid0:%s: (N - 1) mirrors drives must be already faulty!\n",
736 mdname(mddev));
737 return ERR_PTR(-EINVAL);
738 }
739
740 /*
741 * a raid1 doesn't have the notion of chunk size, so
742 * figure out the largest suitable size we can use.
743 */
744 chunksect = 64 * 2; /* 64K by default */
745
746 /* The array must be an exact multiple of chunksize */
747 while (chunksect && (mddev->array_sectors & (chunksect - 1)))
748 chunksect >>= 1;
749
750 if ((chunksect << 9) < PAGE_SIZE)
751 /* array size does not allow a suitable chunk size */
752 return ERR_PTR(-EINVAL);
753
754 /* Set new parameters */
755 mddev->new_level = 0;
756 mddev->new_layout = 0;
757 mddev->new_chunk_sectors = chunksect;
758 mddev->chunk_sectors = chunksect;
759 mddev->delta_disks = 1 - mddev->raid_disks;
760 mddev->raid_disks = 1;
761 /* make sure it will be not marked as dirty */
762 mddev->recovery_cp = MaxSector;
763 mddev_clear_unsupported_flags(mddev, UNSUPPORTED_MDDEV_FLAGS);
764
765 create_strip_zones(mddev, &priv_conf);
766 return priv_conf;
767}
768
769static void *raid0_takeover(struct mddev *mddev)
770{
771 /* raid0 can take over:
772 * raid4 - if all data disks are active.
773 * raid5 - providing it is Raid4 layout and one disk is faulty
774 * raid10 - assuming we have all necessary active disks
775 * raid1 - with (N -1) mirror drives faulty
776 */
777
778 if (mddev->bitmap) {
779 pr_warn("md/raid0: %s: cannot takeover array with bitmap\n",
780 mdname(mddev));
781 return ERR_PTR(-EBUSY);
782 }
783 if (mddev->level == 4)
784 return raid0_takeover_raid45(mddev);
785
786 if (mddev->level == 5) {
787 if (mddev->layout == ALGORITHM_PARITY_N)
788 return raid0_takeover_raid45(mddev);
789
790 pr_warn("md/raid0:%s: Raid can only takeover Raid5 with layout: %d\n",
791 mdname(mddev), ALGORITHM_PARITY_N);
792 }
793
794 if (mddev->level == 10)
795 return raid0_takeover_raid10(mddev);
796
797 if (mddev->level == 1)
798 return raid0_takeover_raid1(mddev);
799
800 pr_warn("Takeover from raid%i to raid0 not supported\n",
801 mddev->level);
802
803 return ERR_PTR(-EINVAL);
804}
805
806static void raid0_quiesce(struct mddev *mddev, int quiesce)
807{
808}
809
810static struct md_personality raid0_personality=
811{
812 .name = "raid0",
813 .level = 0,
814 .owner = THIS_MODULE,
815 .make_request = raid0_make_request,
816 .run = raid0_run,
817 .free = raid0_free,
818 .status = raid0_status,
819 .size = raid0_size,
820 .takeover = raid0_takeover,
821 .quiesce = raid0_quiesce,
822 .congested = raid0_congested,
823};
824
825static int __init raid0_init (void)
826{
827 return register_md_personality (&raid0_personality);
828}
829
830static void raid0_exit (void)
831{
832 unregister_md_personality (&raid0_personality);
833}
834
835module_init(raid0_init);
836module_exit(raid0_exit);
837MODULE_LICENSE("GPL");
838MODULE_DESCRIPTION("RAID0 (striping) personality for MD");
839MODULE_ALIAS("md-personality-2"); /* RAID0 */
840MODULE_ALIAS("md-raid0");
841MODULE_ALIAS("md-level-0");