xref: /linux/drivers/md/raid0.c (revision 9190d4a263264eabf715f5fc1827da45e3fdc247)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3    raid0.c : Multiple Devices driver for Linux
4 	     Copyright (C) 1994-96 Marc ZYNGIER
5 	     <zyngier@ufr-info-p7.ibp.fr> or
6 	     <maz@gloups.fdn.fr>
7 	     Copyright (C) 1999, 2000 Ingo Molnar, Red Hat
8 
9    RAID-0 management functions.
10 
11 */
12 
13 #include <linux/blkdev.h>
14 #include <linux/seq_file.h>
15 #include <linux/module.h>
16 #include <linux/slab.h>
17 #include <trace/events/block.h>
18 #include "md.h"
19 #include "raid0.h"
20 #include "raid5.h"
21 
22 static int default_layout = 0;
23 module_param(default_layout, int, 0644);
24 
25 #define UNSUPPORTED_MDDEV_FLAGS		\
26 	((1L << MD_HAS_JOURNAL) |	\
27 	 (1L << MD_JOURNAL_CLEAN) |	\
28 	 (1L << MD_FAILFAST_SUPPORTED) |\
29 	 (1L << MD_HAS_PPL) |		\
30 	 (1L << MD_HAS_MULTIPLE_PPLS))
31 
32 /*
33  * inform the user of the raid configuration
34 */
35 static void dump_zones(struct mddev *mddev)
36 {
37 	int j, k;
38 	sector_t zone_size = 0;
39 	sector_t zone_start = 0;
40 	struct r0conf *conf = mddev->private;
41 	int raid_disks = conf->strip_zone[0].nb_dev;
42 	pr_debug("md: RAID0 configuration for %s - %d zone%s\n",
43 		 mdname(mddev),
44 		 conf->nr_strip_zones, conf->nr_strip_zones==1?"":"s");
45 	for (j = 0; j < conf->nr_strip_zones; j++) {
46 		char line[200];
47 		int len = 0;
48 
49 		for (k = 0; k < conf->strip_zone[j].nb_dev; k++)
50 			len += scnprintf(line+len, 200-len, "%s%pg", k?"/":"",
51 				conf->devlist[j * raid_disks + k]->bdev);
52 		pr_debug("md: zone%d=[%s]\n", j, line);
53 
54 		zone_size  = conf->strip_zone[j].zone_end - zone_start;
55 		pr_debug("      zone-offset=%10lluKB, device-offset=%10lluKB, size=%10lluKB\n",
56 			(unsigned long long)zone_start>>1,
57 			(unsigned long long)conf->strip_zone[j].dev_start>>1,
58 			(unsigned long long)zone_size>>1);
59 		zone_start = conf->strip_zone[j].zone_end;
60 	}
61 }
62 
63 static int create_strip_zones(struct mddev *mddev, struct r0conf **private_conf)
64 {
65 	int i, c, err;
66 	sector_t curr_zone_end, sectors;
67 	struct md_rdev *smallest, *rdev1, *rdev2, *rdev, **dev;
68 	struct strip_zone *zone;
69 	int cnt;
70 	struct r0conf *conf = kzalloc(sizeof(*conf), GFP_KERNEL);
71 	unsigned blksize = 512;
72 
73 	*private_conf = ERR_PTR(-ENOMEM);
74 	if (!conf)
75 		return -ENOMEM;
76 	rdev_for_each(rdev1, mddev) {
77 		pr_debug("md/raid0:%s: looking at %pg\n",
78 			 mdname(mddev),
79 			 rdev1->bdev);
80 		c = 0;
81 
82 		/* round size to chunk_size */
83 		sectors = rdev1->sectors;
84 		sector_div(sectors, mddev->chunk_sectors);
85 		rdev1->sectors = sectors * mddev->chunk_sectors;
86 
87 		blksize = max(blksize, queue_logical_block_size(
88 				      rdev1->bdev->bd_disk->queue));
89 
90 		rdev_for_each(rdev2, mddev) {
91 			pr_debug("md/raid0:%s:   comparing %pg(%llu)"
92 				 " with %pg(%llu)\n",
93 				 mdname(mddev),
94 				 rdev1->bdev,
95 				 (unsigned long long)rdev1->sectors,
96 				 rdev2->bdev,
97 				 (unsigned long long)rdev2->sectors);
98 			if (rdev2 == rdev1) {
99 				pr_debug("md/raid0:%s:   END\n",
100 					 mdname(mddev));
101 				break;
102 			}
103 			if (rdev2->sectors == rdev1->sectors) {
104 				/*
105 				 * Not unique, don't count it as a new
106 				 * group
107 				 */
108 				pr_debug("md/raid0:%s:   EQUAL\n",
109 					 mdname(mddev));
110 				c = 1;
111 				break;
112 			}
113 			pr_debug("md/raid0:%s:   NOT EQUAL\n",
114 				 mdname(mddev));
115 		}
116 		if (!c) {
117 			pr_debug("md/raid0:%s:   ==> UNIQUE\n",
118 				 mdname(mddev));
119 			conf->nr_strip_zones++;
120 			pr_debug("md/raid0:%s: %d zones\n",
121 				 mdname(mddev), conf->nr_strip_zones);
122 		}
123 	}
124 	pr_debug("md/raid0:%s: FINAL %d zones\n",
125 		 mdname(mddev), conf->nr_strip_zones);
126 
127 	/*
128 	 * now since we have the hard sector sizes, we can make sure
129 	 * chunk size is a multiple of that sector size
130 	 */
131 	if ((mddev->chunk_sectors << 9) % blksize) {
132 		pr_warn("md/raid0:%s: chunk_size of %d not multiple of block size %d\n",
133 			mdname(mddev),
134 			mddev->chunk_sectors << 9, blksize);
135 		err = -EINVAL;
136 		goto abort;
137 	}
138 
139 	err = -ENOMEM;
140 	conf->strip_zone = kcalloc(conf->nr_strip_zones,
141 				   sizeof(struct strip_zone),
142 				   GFP_KERNEL);
143 	if (!conf->strip_zone)
144 		goto abort;
145 	conf->devlist = kzalloc(array3_size(sizeof(struct md_rdev *),
146 					    conf->nr_strip_zones,
147 					    mddev->raid_disks),
148 				GFP_KERNEL);
149 	if (!conf->devlist)
150 		goto abort;
151 
152 	/* The first zone must contain all devices, so here we check that
153 	 * there is a proper alignment of slots to devices and find them all
154 	 */
155 	zone = &conf->strip_zone[0];
156 	cnt = 0;
157 	smallest = NULL;
158 	dev = conf->devlist;
159 	err = -EINVAL;
160 	rdev_for_each(rdev1, mddev) {
161 		int j = rdev1->raid_disk;
162 
163 		if (mddev->level == 10) {
164 			/* taking over a raid10-n2 array */
165 			j /= 2;
166 			rdev1->new_raid_disk = j;
167 		}
168 
169 		if (mddev->level == 1) {
170 			/* taiking over a raid1 array-
171 			 * we have only one active disk
172 			 */
173 			j = 0;
174 			rdev1->new_raid_disk = j;
175 		}
176 
177 		if (j < 0) {
178 			pr_warn("md/raid0:%s: remove inactive devices before converting to RAID0\n",
179 				mdname(mddev));
180 			goto abort;
181 		}
182 		if (j >= mddev->raid_disks) {
183 			pr_warn("md/raid0:%s: bad disk number %d - aborting!\n",
184 				mdname(mddev), j);
185 			goto abort;
186 		}
187 		if (dev[j]) {
188 			pr_warn("md/raid0:%s: multiple devices for %d - aborting!\n",
189 				mdname(mddev), j);
190 			goto abort;
191 		}
192 		dev[j] = rdev1;
193 
194 		if (!smallest || (rdev1->sectors < smallest->sectors))
195 			smallest = rdev1;
196 		cnt++;
197 	}
198 	if (cnt != mddev->raid_disks) {
199 		pr_warn("md/raid0:%s: too few disks (%d of %d) - aborting!\n",
200 			mdname(mddev), cnt, mddev->raid_disks);
201 		goto abort;
202 	}
203 	zone->nb_dev = cnt;
204 	zone->zone_end = smallest->sectors * cnt;
205 
206 	curr_zone_end = zone->zone_end;
207 
208 	/* now do the other zones */
209 	for (i = 1; i < conf->nr_strip_zones; i++)
210 	{
211 		int j;
212 
213 		zone = conf->strip_zone + i;
214 		dev = conf->devlist + i * mddev->raid_disks;
215 
216 		pr_debug("md/raid0:%s: zone %d\n", mdname(mddev), i);
217 		zone->dev_start = smallest->sectors;
218 		smallest = NULL;
219 		c = 0;
220 
221 		for (j=0; j<cnt; j++) {
222 			rdev = conf->devlist[j];
223 			if (rdev->sectors <= zone->dev_start) {
224 				pr_debug("md/raid0:%s: checking %pg ... nope\n",
225 					 mdname(mddev),
226 					 rdev->bdev);
227 				continue;
228 			}
229 			pr_debug("md/raid0:%s: checking %pg ..."
230 				 " contained as device %d\n",
231 				 mdname(mddev),
232 				 rdev->bdev, c);
233 			dev[c] = rdev;
234 			c++;
235 			if (!smallest || rdev->sectors < smallest->sectors) {
236 				smallest = rdev;
237 				pr_debug("md/raid0:%s:  (%llu) is smallest!.\n",
238 					 mdname(mddev),
239 					 (unsigned long long)rdev->sectors);
240 			}
241 		}
242 
243 		zone->nb_dev = c;
244 		sectors = (smallest->sectors - zone->dev_start) * c;
245 		pr_debug("md/raid0:%s: zone->nb_dev: %d, sectors: %llu\n",
246 			 mdname(mddev),
247 			 zone->nb_dev, (unsigned long long)sectors);
248 
249 		curr_zone_end += sectors;
250 		zone->zone_end = curr_zone_end;
251 
252 		pr_debug("md/raid0:%s: current zone start: %llu\n",
253 			 mdname(mddev),
254 			 (unsigned long long)smallest->sectors);
255 	}
256 
257 	if (conf->nr_strip_zones == 1 || conf->strip_zone[1].nb_dev == 1) {
258 		conf->layout = RAID0_ORIG_LAYOUT;
259 	} else if (mddev->layout == RAID0_ORIG_LAYOUT ||
260 		   mddev->layout == RAID0_ALT_MULTIZONE_LAYOUT) {
261 		conf->layout = mddev->layout;
262 	} else if (default_layout == RAID0_ORIG_LAYOUT ||
263 		   default_layout == RAID0_ALT_MULTIZONE_LAYOUT) {
264 		conf->layout = default_layout;
265 	} else {
266 		pr_err("md/raid0:%s: cannot assemble multi-zone RAID0 with default_layout setting\n",
267 		       mdname(mddev));
268 		pr_err("md/raid0: please set raid0.default_layout to 1 or 2\n");
269 		err = -EOPNOTSUPP;
270 		goto abort;
271 	}
272 
273 	pr_debug("md/raid0:%s: done.\n", mdname(mddev));
274 	*private_conf = conf;
275 
276 	return 0;
277 abort:
278 	kfree(conf->strip_zone);
279 	kfree(conf->devlist);
280 	kfree(conf);
281 	*private_conf = ERR_PTR(err);
282 	return err;
283 }
284 
285 /* Find the zone which holds a particular offset
286  * Update *sectorp to be an offset in that zone
287  */
288 static struct strip_zone *find_zone(struct r0conf *conf,
289 				    sector_t *sectorp)
290 {
291 	int i;
292 	struct strip_zone *z = conf->strip_zone;
293 	sector_t sector = *sectorp;
294 
295 	for (i = 0; i < conf->nr_strip_zones; i++)
296 		if (sector < z[i].zone_end) {
297 			if (i)
298 				*sectorp = sector - z[i-1].zone_end;
299 			return z + i;
300 		}
301 	BUG();
302 }
303 
304 /*
305  * remaps the bio to the target device. we separate two flows.
306  * power 2 flow and a general flow for the sake of performance
307 */
308 static struct md_rdev *map_sector(struct mddev *mddev, struct strip_zone *zone,
309 				sector_t sector, sector_t *sector_offset)
310 {
311 	unsigned int sect_in_chunk;
312 	sector_t chunk;
313 	struct r0conf *conf = mddev->private;
314 	int raid_disks = conf->strip_zone[0].nb_dev;
315 	unsigned int chunk_sects = mddev->chunk_sectors;
316 
317 	if (is_power_of_2(chunk_sects)) {
318 		int chunksect_bits = ffz(~chunk_sects);
319 		/* find the sector offset inside the chunk */
320 		sect_in_chunk  = sector & (chunk_sects - 1);
321 		sector >>= chunksect_bits;
322 		/* chunk in zone */
323 		chunk = *sector_offset;
324 		/* quotient is the chunk in real device*/
325 		sector_div(chunk, zone->nb_dev << chunksect_bits);
326 	} else{
327 		sect_in_chunk = sector_div(sector, chunk_sects);
328 		chunk = *sector_offset;
329 		sector_div(chunk, chunk_sects * zone->nb_dev);
330 	}
331 	/*
332 	*  position the bio over the real device
333 	*  real sector = chunk in device + starting of zone
334 	*	+ the position in the chunk
335 	*/
336 	*sector_offset = (chunk * chunk_sects) + sect_in_chunk;
337 	return conf->devlist[(zone - conf->strip_zone)*raid_disks
338 			     + sector_div(sector, zone->nb_dev)];
339 }
340 
341 static sector_t raid0_size(struct mddev *mddev, sector_t sectors, int raid_disks)
342 {
343 	sector_t array_sectors = 0;
344 	struct md_rdev *rdev;
345 
346 	WARN_ONCE(sectors || raid_disks,
347 		  "%s does not support generic reshape\n", __func__);
348 
349 	rdev_for_each(rdev, mddev)
350 		array_sectors += (rdev->sectors &
351 				  ~(sector_t)(mddev->chunk_sectors-1));
352 
353 	return array_sectors;
354 }
355 
356 static void free_conf(struct mddev *mddev, struct r0conf *conf)
357 {
358 	kfree(conf->strip_zone);
359 	kfree(conf->devlist);
360 	kfree(conf);
361 }
362 
363 static void raid0_free(struct mddev *mddev, void *priv)
364 {
365 	struct r0conf *conf = priv;
366 
367 	free_conf(mddev, conf);
368 	acct_bioset_exit(mddev);
369 }
370 
371 static int raid0_run(struct mddev *mddev)
372 {
373 	struct r0conf *conf;
374 	int ret;
375 
376 	if (mddev->chunk_sectors == 0) {
377 		pr_warn("md/raid0:%s: chunk size must be set.\n", mdname(mddev));
378 		return -EINVAL;
379 	}
380 	if (md_check_no_bitmap(mddev))
381 		return -EINVAL;
382 
383 	if (acct_bioset_init(mddev)) {
384 		pr_err("md/raid0:%s: alloc acct bioset failed.\n", mdname(mddev));
385 		return -ENOMEM;
386 	}
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 			goto exit_acct_set;
393 		mddev->private = conf;
394 	}
395 	conf = mddev->private;
396 	if (mddev->queue) {
397 		struct md_rdev *rdev;
398 
399 		blk_queue_max_hw_sectors(mddev->queue, mddev->chunk_sectors);
400 		blk_queue_max_write_zeroes_sectors(mddev->queue, mddev->chunk_sectors);
401 		blk_queue_max_discard_sectors(mddev->queue, UINT_MAX);
402 
403 		blk_queue_io_min(mddev->queue, mddev->chunk_sectors << 9);
404 		blk_queue_io_opt(mddev->queue,
405 				 (mddev->chunk_sectors << 9) * mddev->raid_disks);
406 
407 		rdev_for_each(rdev, mddev) {
408 			disk_stack_limits(mddev->gendisk, rdev->bdev,
409 					  rdev->data_offset << 9);
410 		}
411 	}
412 
413 	/* calculate array device size */
414 	md_set_array_sectors(mddev, raid0_size(mddev, 0, 0));
415 
416 	pr_debug("md/raid0:%s: md_size is %llu sectors.\n",
417 		 mdname(mddev),
418 		 (unsigned long long)mddev->array_sectors);
419 
420 	dump_zones(mddev);
421 
422 	ret = md_integrity_register(mddev);
423 	if (ret)
424 		goto free;
425 
426 	return ret;
427 
428 free:
429 	free_conf(mddev, conf);
430 exit_acct_set:
431 	acct_bioset_exit(mddev);
432 	return ret;
433 }
434 
435 static void raid0_handle_discard(struct mddev *mddev, struct bio *bio)
436 {
437 	struct r0conf *conf = mddev->private;
438 	struct strip_zone *zone;
439 	sector_t start = bio->bi_iter.bi_sector;
440 	sector_t end;
441 	unsigned int stripe_size;
442 	sector_t first_stripe_index, last_stripe_index;
443 	sector_t start_disk_offset;
444 	unsigned int start_disk_index;
445 	sector_t end_disk_offset;
446 	unsigned int end_disk_index;
447 	unsigned int disk;
448 
449 	zone = find_zone(conf, &start);
450 
451 	if (bio_end_sector(bio) > zone->zone_end) {
452 		struct bio *split = bio_split(bio,
453 			zone->zone_end - bio->bi_iter.bi_sector, GFP_NOIO,
454 			&mddev->bio_set);
455 		bio_chain(split, bio);
456 		submit_bio_noacct(bio);
457 		bio = split;
458 		end = zone->zone_end;
459 	} else
460 		end = bio_end_sector(bio);
461 
462 	if (zone != conf->strip_zone)
463 		end = end - zone[-1].zone_end;
464 
465 	/* Now start and end is the offset in zone */
466 	stripe_size = zone->nb_dev * mddev->chunk_sectors;
467 
468 	first_stripe_index = start;
469 	sector_div(first_stripe_index, stripe_size);
470 	last_stripe_index = end;
471 	sector_div(last_stripe_index, stripe_size);
472 
473 	start_disk_index = (int)(start - first_stripe_index * stripe_size) /
474 		mddev->chunk_sectors;
475 	start_disk_offset = ((int)(start - first_stripe_index * stripe_size) %
476 		mddev->chunk_sectors) +
477 		first_stripe_index * mddev->chunk_sectors;
478 	end_disk_index = (int)(end - last_stripe_index * stripe_size) /
479 		mddev->chunk_sectors;
480 	end_disk_offset = ((int)(end - last_stripe_index * stripe_size) %
481 		mddev->chunk_sectors) +
482 		last_stripe_index * mddev->chunk_sectors;
483 
484 	for (disk = 0; disk < zone->nb_dev; disk++) {
485 		sector_t dev_start, dev_end;
486 		struct md_rdev *rdev;
487 
488 		if (disk < start_disk_index)
489 			dev_start = (first_stripe_index + 1) *
490 				mddev->chunk_sectors;
491 		else if (disk > start_disk_index)
492 			dev_start = first_stripe_index * mddev->chunk_sectors;
493 		else
494 			dev_start = start_disk_offset;
495 
496 		if (disk < end_disk_index)
497 			dev_end = (last_stripe_index + 1) * mddev->chunk_sectors;
498 		else if (disk > end_disk_index)
499 			dev_end = last_stripe_index * mddev->chunk_sectors;
500 		else
501 			dev_end = end_disk_offset;
502 
503 		if (dev_end <= dev_start)
504 			continue;
505 
506 		rdev = conf->devlist[(zone - conf->strip_zone) *
507 			conf->strip_zone[0].nb_dev + disk];
508 		md_submit_discard_bio(mddev, rdev, bio,
509 			dev_start + zone->dev_start + rdev->data_offset,
510 			dev_end - dev_start);
511 	}
512 	bio_endio(bio);
513 }
514 
515 static bool raid0_make_request(struct mddev *mddev, struct bio *bio)
516 {
517 	struct r0conf *conf = mddev->private;
518 	struct strip_zone *zone;
519 	struct md_rdev *tmp_dev;
520 	sector_t bio_sector;
521 	sector_t sector;
522 	sector_t orig_sector;
523 	unsigned chunk_sects;
524 	unsigned sectors;
525 
526 	if (unlikely(bio->bi_opf & REQ_PREFLUSH)
527 	    && md_flush_request(mddev, bio))
528 		return true;
529 
530 	if (unlikely((bio_op(bio) == REQ_OP_DISCARD))) {
531 		raid0_handle_discard(mddev, bio);
532 		return true;
533 	}
534 
535 	bio_sector = bio->bi_iter.bi_sector;
536 	sector = bio_sector;
537 	chunk_sects = mddev->chunk_sectors;
538 
539 	sectors = chunk_sects -
540 		(likely(is_power_of_2(chunk_sects))
541 		 ? (sector & (chunk_sects-1))
542 		 : sector_div(sector, chunk_sects));
543 
544 	/* Restore due to sector_div */
545 	sector = bio_sector;
546 
547 	if (sectors < bio_sectors(bio)) {
548 		struct bio *split = bio_split(bio, sectors, GFP_NOIO,
549 					      &mddev->bio_set);
550 		bio_chain(split, bio);
551 		submit_bio_noacct(bio);
552 		bio = split;
553 	}
554 
555 	if (bio->bi_pool != &mddev->bio_set)
556 		md_account_bio(mddev, &bio);
557 
558 	orig_sector = sector;
559 	zone = find_zone(mddev->private, &sector);
560 	switch (conf->layout) {
561 	case RAID0_ORIG_LAYOUT:
562 		tmp_dev = map_sector(mddev, zone, orig_sector, &sector);
563 		break;
564 	case RAID0_ALT_MULTIZONE_LAYOUT:
565 		tmp_dev = map_sector(mddev, zone, sector, &sector);
566 		break;
567 	default:
568 		WARN(1, "md/raid0:%s: Invalid layout\n", mdname(mddev));
569 		bio_io_error(bio);
570 		return true;
571 	}
572 
573 	if (unlikely(is_mddev_broken(tmp_dev, "raid0"))) {
574 		bio_io_error(bio);
575 		return true;
576 	}
577 
578 	bio_set_dev(bio, tmp_dev->bdev);
579 	bio->bi_iter.bi_sector = sector + zone->dev_start +
580 		tmp_dev->data_offset;
581 
582 	if (mddev->gendisk)
583 		trace_block_bio_remap(bio, disk_devt(mddev->gendisk),
584 				      bio_sector);
585 	mddev_check_write_zeroes(mddev, bio);
586 	submit_bio_noacct(bio);
587 	return true;
588 }
589 
590 static void raid0_status(struct seq_file *seq, struct mddev *mddev)
591 {
592 	seq_printf(seq, " %dk chunks", mddev->chunk_sectors / 2);
593 	return;
594 }
595 
596 static void *raid0_takeover_raid45(struct mddev *mddev)
597 {
598 	struct md_rdev *rdev;
599 	struct r0conf *priv_conf;
600 
601 	if (mddev->degraded != 1) {
602 		pr_warn("md/raid0:%s: raid5 must be degraded! Degraded disks: %d\n",
603 			mdname(mddev),
604 			mddev->degraded);
605 		return ERR_PTR(-EINVAL);
606 	}
607 
608 	rdev_for_each(rdev, mddev) {
609 		/* check slot number for a disk */
610 		if (rdev->raid_disk == mddev->raid_disks-1) {
611 			pr_warn("md/raid0:%s: raid5 must have missing parity disk!\n",
612 				mdname(mddev));
613 			return ERR_PTR(-EINVAL);
614 		}
615 		rdev->sectors = mddev->dev_sectors;
616 	}
617 
618 	/* Set new parameters */
619 	mddev->new_level = 0;
620 	mddev->new_layout = 0;
621 	mddev->new_chunk_sectors = mddev->chunk_sectors;
622 	mddev->raid_disks--;
623 	mddev->delta_disks = -1;
624 	/* make sure it will be not marked as dirty */
625 	mddev->recovery_cp = MaxSector;
626 	mddev_clear_unsupported_flags(mddev, UNSUPPORTED_MDDEV_FLAGS);
627 
628 	create_strip_zones(mddev, &priv_conf);
629 
630 	return priv_conf;
631 }
632 
633 static void *raid0_takeover_raid10(struct mddev *mddev)
634 {
635 	struct r0conf *priv_conf;
636 
637 	/* Check layout:
638 	 *  - far_copies must be 1
639 	 *  - near_copies must be 2
640 	 *  - disks number must be even
641 	 *  - all mirrors must be already degraded
642 	 */
643 	if (mddev->layout != ((1 << 8) + 2)) {
644 		pr_warn("md/raid0:%s:: Raid0 cannot takeover layout: 0x%x\n",
645 			mdname(mddev),
646 			mddev->layout);
647 		return ERR_PTR(-EINVAL);
648 	}
649 	if (mddev->raid_disks & 1) {
650 		pr_warn("md/raid0:%s: Raid0 cannot takeover Raid10 with odd disk number.\n",
651 			mdname(mddev));
652 		return ERR_PTR(-EINVAL);
653 	}
654 	if (mddev->degraded != (mddev->raid_disks>>1)) {
655 		pr_warn("md/raid0:%s: All mirrors must be already degraded!\n",
656 			mdname(mddev));
657 		return ERR_PTR(-EINVAL);
658 	}
659 
660 	/* Set new parameters */
661 	mddev->new_level = 0;
662 	mddev->new_layout = 0;
663 	mddev->new_chunk_sectors = mddev->chunk_sectors;
664 	mddev->delta_disks = - mddev->raid_disks / 2;
665 	mddev->raid_disks += mddev->delta_disks;
666 	mddev->degraded = 0;
667 	/* make sure it will be not marked as dirty */
668 	mddev->recovery_cp = MaxSector;
669 	mddev_clear_unsupported_flags(mddev, UNSUPPORTED_MDDEV_FLAGS);
670 
671 	create_strip_zones(mddev, &priv_conf);
672 	return priv_conf;
673 }
674 
675 static void *raid0_takeover_raid1(struct mddev *mddev)
676 {
677 	struct r0conf *priv_conf;
678 	int chunksect;
679 
680 	/* Check layout:
681 	 *  - (N - 1) mirror drives must be already faulty
682 	 */
683 	if ((mddev->raid_disks - 1) != mddev->degraded) {
684 		pr_err("md/raid0:%s: (N - 1) mirrors drives must be already faulty!\n",
685 		       mdname(mddev));
686 		return ERR_PTR(-EINVAL);
687 	}
688 
689 	/*
690 	 * a raid1 doesn't have the notion of chunk size, so
691 	 * figure out the largest suitable size we can use.
692 	 */
693 	chunksect = 64 * 2; /* 64K by default */
694 
695 	/* The array must be an exact multiple of chunksize */
696 	while (chunksect && (mddev->array_sectors & (chunksect - 1)))
697 		chunksect >>= 1;
698 
699 	if ((chunksect << 9) < PAGE_SIZE)
700 		/* array size does not allow a suitable chunk size */
701 		return ERR_PTR(-EINVAL);
702 
703 	/* Set new parameters */
704 	mddev->new_level = 0;
705 	mddev->new_layout = 0;
706 	mddev->new_chunk_sectors = chunksect;
707 	mddev->chunk_sectors = chunksect;
708 	mddev->delta_disks = 1 - mddev->raid_disks;
709 	mddev->raid_disks = 1;
710 	/* make sure it will be not marked as dirty */
711 	mddev->recovery_cp = MaxSector;
712 	mddev_clear_unsupported_flags(mddev, UNSUPPORTED_MDDEV_FLAGS);
713 
714 	create_strip_zones(mddev, &priv_conf);
715 	return priv_conf;
716 }
717 
718 static void *raid0_takeover(struct mddev *mddev)
719 {
720 	/* raid0 can take over:
721 	 *  raid4 - if all data disks are active.
722 	 *  raid5 - providing it is Raid4 layout and one disk is faulty
723 	 *  raid10 - assuming we have all necessary active disks
724 	 *  raid1 - with (N -1) mirror drives faulty
725 	 */
726 
727 	if (mddev->bitmap) {
728 		pr_warn("md/raid0: %s: cannot takeover array with bitmap\n",
729 			mdname(mddev));
730 		return ERR_PTR(-EBUSY);
731 	}
732 	if (mddev->level == 4)
733 		return raid0_takeover_raid45(mddev);
734 
735 	if (mddev->level == 5) {
736 		if (mddev->layout == ALGORITHM_PARITY_N)
737 			return raid0_takeover_raid45(mddev);
738 
739 		pr_warn("md/raid0:%s: Raid can only takeover Raid5 with layout: %d\n",
740 			mdname(mddev), ALGORITHM_PARITY_N);
741 	}
742 
743 	if (mddev->level == 10)
744 		return raid0_takeover_raid10(mddev);
745 
746 	if (mddev->level == 1)
747 		return raid0_takeover_raid1(mddev);
748 
749 	pr_warn("Takeover from raid%i to raid0 not supported\n",
750 		mddev->level);
751 
752 	return ERR_PTR(-EINVAL);
753 }
754 
755 static void raid0_quiesce(struct mddev *mddev, int quiesce)
756 {
757 }
758 
759 static struct md_personality raid0_personality=
760 {
761 	.name		= "raid0",
762 	.level		= 0,
763 	.owner		= THIS_MODULE,
764 	.make_request	= raid0_make_request,
765 	.run		= raid0_run,
766 	.free		= raid0_free,
767 	.status		= raid0_status,
768 	.size		= raid0_size,
769 	.takeover	= raid0_takeover,
770 	.quiesce	= raid0_quiesce,
771 };
772 
773 static int __init raid0_init (void)
774 {
775 	return register_md_personality (&raid0_personality);
776 }
777 
778 static void raid0_exit (void)
779 {
780 	unregister_md_personality (&raid0_personality);
781 }
782 
783 module_init(raid0_init);
784 module_exit(raid0_exit);
785 MODULE_LICENSE("GPL");
786 MODULE_DESCRIPTION("RAID0 (striping) personality for MD");
787 MODULE_ALIAS("md-personality-2"); /* RAID0 */
788 MODULE_ALIAS("md-raid0");
789 MODULE_ALIAS("md-level-0");
790