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/*
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");