xref: /linux/drivers/md/raid0.c (revision 0be3ff0c)
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 	char b[BDEVNAME_SIZE];
41 	struct r0conf *conf = mddev->private;
42 	int raid_disks = conf->strip_zone[0].nb_dev;
43 	pr_debug("md: RAID0 configuration for %s - %d zone%s\n",
44 		 mdname(mddev),
45 		 conf->nr_strip_zones, conf->nr_strip_zones==1?"":"s");
46 	for (j = 0; j < conf->nr_strip_zones; j++) {
47 		char line[200];
48 		int len = 0;
49 
50 		for (k = 0; k < conf->strip_zone[j].nb_dev; k++)
51 			len += snprintf(line+len, 200-len, "%s%s", k?"/":"",
52 					bdevname(conf->devlist[j*raid_disks
53 							       + k]->bdev, b));
54 		pr_debug("md: zone%d=[%s]\n", j, line);
55 
56 		zone_size  = conf->strip_zone[j].zone_end - zone_start;
57 		pr_debug("      zone-offset=%10lluKB, device-offset=%10lluKB, size=%10lluKB\n",
58 			(unsigned long long)zone_start>>1,
59 			(unsigned long long)conf->strip_zone[j].dev_start>>1,
60 			(unsigned long long)zone_size>>1);
61 		zone_start = conf->strip_zone[j].zone_end;
62 	}
63 }
64 
65 static int create_strip_zones(struct mddev *mddev, struct r0conf **private_conf)
66 {
67 	int i, c, err;
68 	sector_t curr_zone_end, sectors;
69 	struct md_rdev *smallest, *rdev1, *rdev2, *rdev, **dev;
70 	struct strip_zone *zone;
71 	int cnt;
72 	char b[BDEVNAME_SIZE];
73 	char b2[BDEVNAME_SIZE];
74 	struct r0conf *conf = kzalloc(sizeof(*conf), GFP_KERNEL);
75 	unsigned blksize = 512;
76 
77 	*private_conf = ERR_PTR(-ENOMEM);
78 	if (!conf)
79 		return -ENOMEM;
80 	rdev_for_each(rdev1, mddev) {
81 		pr_debug("md/raid0:%s: looking at %s\n",
82 			 mdname(mddev),
83 			 bdevname(rdev1->bdev, b));
84 		c = 0;
85 
86 		/* round size to chunk_size */
87 		sectors = rdev1->sectors;
88 		sector_div(sectors, mddev->chunk_sectors);
89 		rdev1->sectors = sectors * mddev->chunk_sectors;
90 
91 		blksize = max(blksize, queue_logical_block_size(
92 				      rdev1->bdev->bd_disk->queue));
93 
94 		rdev_for_each(rdev2, mddev) {
95 			pr_debug("md/raid0:%s:   comparing %s(%llu)"
96 				 " with %s(%llu)\n",
97 				 mdname(mddev),
98 				 bdevname(rdev1->bdev,b),
99 				 (unsigned long long)rdev1->sectors,
100 				 bdevname(rdev2->bdev,b2),
101 				 (unsigned long long)rdev2->sectors);
102 			if (rdev2 == rdev1) {
103 				pr_debug("md/raid0:%s:   END\n",
104 					 mdname(mddev));
105 				break;
106 			}
107 			if (rdev2->sectors == rdev1->sectors) {
108 				/*
109 				 * Not unique, don't count it as a new
110 				 * group
111 				 */
112 				pr_debug("md/raid0:%s:   EQUAL\n",
113 					 mdname(mddev));
114 				c = 1;
115 				break;
116 			}
117 			pr_debug("md/raid0:%s:   NOT EQUAL\n",
118 				 mdname(mddev));
119 		}
120 		if (!c) {
121 			pr_debug("md/raid0:%s:   ==> UNIQUE\n",
122 				 mdname(mddev));
123 			conf->nr_strip_zones++;
124 			pr_debug("md/raid0:%s: %d zones\n",
125 				 mdname(mddev), conf->nr_strip_zones);
126 		}
127 	}
128 	pr_debug("md/raid0:%s: FINAL %d zones\n",
129 		 mdname(mddev), conf->nr_strip_zones);
130 
131 	/*
132 	 * now since we have the hard sector sizes, we can make sure
133 	 * chunk size is a multiple of that sector size
134 	 */
135 	if ((mddev->chunk_sectors << 9) % blksize) {
136 		pr_warn("md/raid0:%s: chunk_size of %d not multiple of block size %d\n",
137 			mdname(mddev),
138 			mddev->chunk_sectors << 9, blksize);
139 		err = -EINVAL;
140 		goto abort;
141 	}
142 
143 	err = -ENOMEM;
144 	conf->strip_zone = kcalloc(conf->nr_strip_zones,
145 				   sizeof(struct strip_zone),
146 				   GFP_KERNEL);
147 	if (!conf->strip_zone)
148 		goto abort;
149 	conf->devlist = kzalloc(array3_size(sizeof(struct md_rdev *),
150 					    conf->nr_strip_zones,
151 					    mddev->raid_disks),
152 				GFP_KERNEL);
153 	if (!conf->devlist)
154 		goto abort;
155 
156 	/* The first zone must contain all devices, so here we check that
157 	 * there is a proper alignment of slots to devices and find them all
158 	 */
159 	zone = &conf->strip_zone[0];
160 	cnt = 0;
161 	smallest = NULL;
162 	dev = conf->devlist;
163 	err = -EINVAL;
164 	rdev_for_each(rdev1, mddev) {
165 		int j = rdev1->raid_disk;
166 
167 		if (mddev->level == 10) {
168 			/* taking over a raid10-n2 array */
169 			j /= 2;
170 			rdev1->new_raid_disk = j;
171 		}
172 
173 		if (mddev->level == 1) {
174 			/* taiking over a raid1 array-
175 			 * we have only one active disk
176 			 */
177 			j = 0;
178 			rdev1->new_raid_disk = j;
179 		}
180 
181 		if (j < 0) {
182 			pr_warn("md/raid0:%s: remove inactive devices before converting to RAID0\n",
183 				mdname(mddev));
184 			goto abort;
185 		}
186 		if (j >= mddev->raid_disks) {
187 			pr_warn("md/raid0:%s: bad disk number %d - aborting!\n",
188 				mdname(mddev), j);
189 			goto abort;
190 		}
191 		if (dev[j]) {
192 			pr_warn("md/raid0:%s: multiple devices for %d - aborting!\n",
193 				mdname(mddev), j);
194 			goto abort;
195 		}
196 		dev[j] = rdev1;
197 
198 		if (!smallest || (rdev1->sectors < smallest->sectors))
199 			smallest = rdev1;
200 		cnt++;
201 	}
202 	if (cnt != mddev->raid_disks) {
203 		pr_warn("md/raid0:%s: too few disks (%d of %d) - aborting!\n",
204 			mdname(mddev), cnt, mddev->raid_disks);
205 		goto abort;
206 	}
207 	zone->nb_dev = cnt;
208 	zone->zone_end = smallest->sectors * cnt;
209 
210 	curr_zone_end = zone->zone_end;
211 
212 	/* now do the other zones */
213 	for (i = 1; i < conf->nr_strip_zones; i++)
214 	{
215 		int j;
216 
217 		zone = conf->strip_zone + i;
218 		dev = conf->devlist + i * mddev->raid_disks;
219 
220 		pr_debug("md/raid0:%s: zone %d\n", mdname(mddev), i);
221 		zone->dev_start = smallest->sectors;
222 		smallest = NULL;
223 		c = 0;
224 
225 		for (j=0; j<cnt; j++) {
226 			rdev = conf->devlist[j];
227 			if (rdev->sectors <= zone->dev_start) {
228 				pr_debug("md/raid0:%s: checking %s ... nope\n",
229 					 mdname(mddev),
230 					 bdevname(rdev->bdev, b));
231 				continue;
232 			}
233 			pr_debug("md/raid0:%s: checking %s ..."
234 				 " contained as device %d\n",
235 				 mdname(mddev),
236 				 bdevname(rdev->bdev, b), c);
237 			dev[c] = rdev;
238 			c++;
239 			if (!smallest || rdev->sectors < smallest->sectors) {
240 				smallest = rdev;
241 				pr_debug("md/raid0:%s:  (%llu) is smallest!.\n",
242 					 mdname(mddev),
243 					 (unsigned long long)rdev->sectors);
244 			}
245 		}
246 
247 		zone->nb_dev = c;
248 		sectors = (smallest->sectors - zone->dev_start) * c;
249 		pr_debug("md/raid0:%s: zone->nb_dev: %d, sectors: %llu\n",
250 			 mdname(mddev),
251 			 zone->nb_dev, (unsigned long long)sectors);
252 
253 		curr_zone_end += sectors;
254 		zone->zone_end = curr_zone_end;
255 
256 		pr_debug("md/raid0:%s: current zone start: %llu\n",
257 			 mdname(mddev),
258 			 (unsigned long long)smallest->sectors);
259 	}
260 
261 	if (conf->nr_strip_zones == 1 || conf->strip_zone[1].nb_dev == 1) {
262 		conf->layout = RAID0_ORIG_LAYOUT;
263 	} else if (mddev->layout == RAID0_ORIG_LAYOUT ||
264 		   mddev->layout == RAID0_ALT_MULTIZONE_LAYOUT) {
265 		conf->layout = mddev->layout;
266 	} else if (default_layout == RAID0_ORIG_LAYOUT ||
267 		   default_layout == RAID0_ALT_MULTIZONE_LAYOUT) {
268 		conf->layout = default_layout;
269 	} else {
270 		pr_err("md/raid0:%s: cannot assemble multi-zone RAID0 with default_layout setting\n",
271 		       mdname(mddev));
272 		pr_err("md/raid0: please set raid0.default_layout to 1 or 2\n");
273 		err = -EOPNOTSUPP;
274 		goto abort;
275 	}
276 
277 	pr_debug("md/raid0:%s: done.\n", mdname(mddev));
278 	*private_conf = conf;
279 
280 	return 0;
281 abort:
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  */
292 static 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 */
312 static 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 
345 static 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 
360 static void free_conf(struct mddev *mddev, struct r0conf *conf)
361 {
362 	kfree(conf->strip_zone);
363 	kfree(conf->devlist);
364 	kfree(conf);
365 	mddev->private = NULL;
366 }
367 
368 static void raid0_free(struct mddev *mddev, void *priv)
369 {
370 	struct r0conf *conf = priv;
371 
372 	free_conf(mddev, conf);
373 	acct_bioset_exit(mddev);
374 }
375 
376 static 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 (acct_bioset_init(mddev)) {
389 		pr_err("md/raid0:%s: alloc acct bioset failed.\n", mdname(mddev));
390 		return -ENOMEM;
391 	}
392 
393 	/* if private is not null, we are here after takeover */
394 	if (mddev->private == NULL) {
395 		ret = create_strip_zones(mddev, &conf);
396 		if (ret < 0)
397 			goto exit_acct_set;
398 		mddev->private = conf;
399 	}
400 	conf = mddev->private;
401 	if (mddev->queue) {
402 		struct md_rdev *rdev;
403 
404 		blk_queue_max_hw_sectors(mddev->queue, mddev->chunk_sectors);
405 		blk_queue_max_write_zeroes_sectors(mddev->queue, mddev->chunk_sectors);
406 		blk_queue_max_discard_sectors(mddev->queue, UINT_MAX);
407 
408 		blk_queue_io_min(mddev->queue, mddev->chunk_sectors << 9);
409 		blk_queue_io_opt(mddev->queue,
410 				 (mddev->chunk_sectors << 9) * mddev->raid_disks);
411 
412 		rdev_for_each(rdev, mddev) {
413 			disk_stack_limits(mddev->gendisk, rdev->bdev,
414 					  rdev->data_offset << 9);
415 		}
416 	}
417 
418 	/* calculate array device size */
419 	md_set_array_sectors(mddev, raid0_size(mddev, 0, 0));
420 
421 	pr_debug("md/raid0:%s: md_size is %llu sectors.\n",
422 		 mdname(mddev),
423 		 (unsigned long long)mddev->array_sectors);
424 
425 	dump_zones(mddev);
426 
427 	ret = md_integrity_register(mddev);
428 	if (ret)
429 		goto free;
430 
431 	return ret;
432 
433 free:
434 	free_conf(mddev, conf);
435 exit_acct_set:
436 	acct_bioset_exit(mddev);
437 	return ret;
438 }
439 
440 static void raid0_handle_discard(struct mddev *mddev, struct bio *bio)
441 {
442 	struct r0conf *conf = mddev->private;
443 	struct strip_zone *zone;
444 	sector_t start = bio->bi_iter.bi_sector;
445 	sector_t end;
446 	unsigned int stripe_size;
447 	sector_t first_stripe_index, last_stripe_index;
448 	sector_t start_disk_offset;
449 	unsigned int start_disk_index;
450 	sector_t end_disk_offset;
451 	unsigned int end_disk_index;
452 	unsigned int disk;
453 
454 	zone = find_zone(conf, &start);
455 
456 	if (bio_end_sector(bio) > zone->zone_end) {
457 		struct bio *split = bio_split(bio,
458 			zone->zone_end - bio->bi_iter.bi_sector, GFP_NOIO,
459 			&mddev->bio_set);
460 		bio_chain(split, bio);
461 		submit_bio_noacct(bio);
462 		bio = split;
463 		end = zone->zone_end;
464 	} else
465 		end = bio_end_sector(bio);
466 
467 	if (zone != conf->strip_zone)
468 		end = end - zone[-1].zone_end;
469 
470 	/* Now start and end is the offset in zone */
471 	stripe_size = zone->nb_dev * mddev->chunk_sectors;
472 
473 	first_stripe_index = start;
474 	sector_div(first_stripe_index, stripe_size);
475 	last_stripe_index = end;
476 	sector_div(last_stripe_index, stripe_size);
477 
478 	start_disk_index = (int)(start - first_stripe_index * stripe_size) /
479 		mddev->chunk_sectors;
480 	start_disk_offset = ((int)(start - first_stripe_index * stripe_size) %
481 		mddev->chunk_sectors) +
482 		first_stripe_index * mddev->chunk_sectors;
483 	end_disk_index = (int)(end - last_stripe_index * stripe_size) /
484 		mddev->chunk_sectors;
485 	end_disk_offset = ((int)(end - last_stripe_index * stripe_size) %
486 		mddev->chunk_sectors) +
487 		last_stripe_index * mddev->chunk_sectors;
488 
489 	for (disk = 0; disk < zone->nb_dev; disk++) {
490 		sector_t dev_start, dev_end;
491 		struct md_rdev *rdev;
492 
493 		if (disk < start_disk_index)
494 			dev_start = (first_stripe_index + 1) *
495 				mddev->chunk_sectors;
496 		else if (disk > start_disk_index)
497 			dev_start = first_stripe_index * mddev->chunk_sectors;
498 		else
499 			dev_start = start_disk_offset;
500 
501 		if (disk < end_disk_index)
502 			dev_end = (last_stripe_index + 1) * mddev->chunk_sectors;
503 		else if (disk > end_disk_index)
504 			dev_end = last_stripe_index * mddev->chunk_sectors;
505 		else
506 			dev_end = end_disk_offset;
507 
508 		if (dev_end <= dev_start)
509 			continue;
510 
511 		rdev = conf->devlist[(zone - conf->strip_zone) *
512 			conf->strip_zone[0].nb_dev + disk];
513 		md_submit_discard_bio(mddev, rdev, bio,
514 			dev_start + zone->dev_start + rdev->data_offset,
515 			dev_end - dev_start);
516 	}
517 	bio_endio(bio);
518 }
519 
520 static bool raid0_make_request(struct mddev *mddev, struct bio *bio)
521 {
522 	struct r0conf *conf = mddev->private;
523 	struct strip_zone *zone;
524 	struct md_rdev *tmp_dev;
525 	sector_t bio_sector;
526 	sector_t sector;
527 	sector_t orig_sector;
528 	unsigned chunk_sects;
529 	unsigned sectors;
530 
531 	if (unlikely(bio->bi_opf & REQ_PREFLUSH)
532 	    && md_flush_request(mddev, bio))
533 		return true;
534 
535 	if (unlikely((bio_op(bio) == REQ_OP_DISCARD))) {
536 		raid0_handle_discard(mddev, bio);
537 		return true;
538 	}
539 
540 	bio_sector = bio->bi_iter.bi_sector;
541 	sector = bio_sector;
542 	chunk_sects = mddev->chunk_sectors;
543 
544 	sectors = chunk_sects -
545 		(likely(is_power_of_2(chunk_sects))
546 		 ? (sector & (chunk_sects-1))
547 		 : sector_div(sector, chunk_sects));
548 
549 	/* Restore due to sector_div */
550 	sector = bio_sector;
551 
552 	if (sectors < bio_sectors(bio)) {
553 		struct bio *split = bio_split(bio, sectors, GFP_NOIO,
554 					      &mddev->bio_set);
555 		bio_chain(split, bio);
556 		submit_bio_noacct(bio);
557 		bio = split;
558 	}
559 
560 	if (bio->bi_pool != &mddev->bio_set)
561 		md_account_bio(mddev, &bio);
562 
563 	orig_sector = sector;
564 	zone = find_zone(mddev->private, &sector);
565 	switch (conf->layout) {
566 	case RAID0_ORIG_LAYOUT:
567 		tmp_dev = map_sector(mddev, zone, orig_sector, &sector);
568 		break;
569 	case RAID0_ALT_MULTIZONE_LAYOUT:
570 		tmp_dev = map_sector(mddev, zone, sector, &sector);
571 		break;
572 	default:
573 		WARN(1, "md/raid0:%s: Invalid layout\n", mdname(mddev));
574 		bio_io_error(bio);
575 		return true;
576 	}
577 
578 	if (unlikely(is_mddev_broken(tmp_dev, "raid0"))) {
579 		bio_io_error(bio);
580 		return true;
581 	}
582 
583 	bio_set_dev(bio, tmp_dev->bdev);
584 	bio->bi_iter.bi_sector = sector + zone->dev_start +
585 		tmp_dev->data_offset;
586 
587 	if (mddev->gendisk)
588 		trace_block_bio_remap(bio, disk_devt(mddev->gendisk),
589 				      bio_sector);
590 	mddev_check_write_zeroes(mddev, bio);
591 	submit_bio_noacct(bio);
592 	return true;
593 }
594 
595 static void raid0_status(struct seq_file *seq, struct mddev *mddev)
596 {
597 	seq_printf(seq, " %dk chunks", mddev->chunk_sectors / 2);
598 	return;
599 }
600 
601 static void *raid0_takeover_raid45(struct mddev *mddev)
602 {
603 	struct md_rdev *rdev;
604 	struct r0conf *priv_conf;
605 
606 	if (mddev->degraded != 1) {
607 		pr_warn("md/raid0:%s: raid5 must be degraded! Degraded disks: %d\n",
608 			mdname(mddev),
609 			mddev->degraded);
610 		return ERR_PTR(-EINVAL);
611 	}
612 
613 	rdev_for_each(rdev, mddev) {
614 		/* check slot number for a disk */
615 		if (rdev->raid_disk == mddev->raid_disks-1) {
616 			pr_warn("md/raid0:%s: raid5 must have missing parity disk!\n",
617 				mdname(mddev));
618 			return ERR_PTR(-EINVAL);
619 		}
620 		rdev->sectors = mddev->dev_sectors;
621 	}
622 
623 	/* Set new parameters */
624 	mddev->new_level = 0;
625 	mddev->new_layout = 0;
626 	mddev->new_chunk_sectors = mddev->chunk_sectors;
627 	mddev->raid_disks--;
628 	mddev->delta_disks = -1;
629 	/* make sure it will be not marked as dirty */
630 	mddev->recovery_cp = MaxSector;
631 	mddev_clear_unsupported_flags(mddev, UNSUPPORTED_MDDEV_FLAGS);
632 
633 	create_strip_zones(mddev, &priv_conf);
634 
635 	return priv_conf;
636 }
637 
638 static void *raid0_takeover_raid10(struct mddev *mddev)
639 {
640 	struct r0conf *priv_conf;
641 
642 	/* Check layout:
643 	 *  - far_copies must be 1
644 	 *  - near_copies must be 2
645 	 *  - disks number must be even
646 	 *  - all mirrors must be already degraded
647 	 */
648 	if (mddev->layout != ((1 << 8) + 2)) {
649 		pr_warn("md/raid0:%s:: Raid0 cannot takeover layout: 0x%x\n",
650 			mdname(mddev),
651 			mddev->layout);
652 		return ERR_PTR(-EINVAL);
653 	}
654 	if (mddev->raid_disks & 1) {
655 		pr_warn("md/raid0:%s: Raid0 cannot takeover Raid10 with odd disk number.\n",
656 			mdname(mddev));
657 		return ERR_PTR(-EINVAL);
658 	}
659 	if (mddev->degraded != (mddev->raid_disks>>1)) {
660 		pr_warn("md/raid0:%s: All mirrors must be already degraded!\n",
661 			mdname(mddev));
662 		return ERR_PTR(-EINVAL);
663 	}
664 
665 	/* Set new parameters */
666 	mddev->new_level = 0;
667 	mddev->new_layout = 0;
668 	mddev->new_chunk_sectors = mddev->chunk_sectors;
669 	mddev->delta_disks = - mddev->raid_disks / 2;
670 	mddev->raid_disks += mddev->delta_disks;
671 	mddev->degraded = 0;
672 	/* make sure it will be not marked as dirty */
673 	mddev->recovery_cp = MaxSector;
674 	mddev_clear_unsupported_flags(mddev, UNSUPPORTED_MDDEV_FLAGS);
675 
676 	create_strip_zones(mddev, &priv_conf);
677 	return priv_conf;
678 }
679 
680 static void *raid0_takeover_raid1(struct mddev *mddev)
681 {
682 	struct r0conf *priv_conf;
683 	int chunksect;
684 
685 	/* Check layout:
686 	 *  - (N - 1) mirror drives must be already faulty
687 	 */
688 	if ((mddev->raid_disks - 1) != mddev->degraded) {
689 		pr_err("md/raid0:%s: (N - 1) mirrors drives must be already faulty!\n",
690 		       mdname(mddev));
691 		return ERR_PTR(-EINVAL);
692 	}
693 
694 	/*
695 	 * a raid1 doesn't have the notion of chunk size, so
696 	 * figure out the largest suitable size we can use.
697 	 */
698 	chunksect = 64 * 2; /* 64K by default */
699 
700 	/* The array must be an exact multiple of chunksize */
701 	while (chunksect && (mddev->array_sectors & (chunksect - 1)))
702 		chunksect >>= 1;
703 
704 	if ((chunksect << 9) < PAGE_SIZE)
705 		/* array size does not allow a suitable chunk size */
706 		return ERR_PTR(-EINVAL);
707 
708 	/* Set new parameters */
709 	mddev->new_level = 0;
710 	mddev->new_layout = 0;
711 	mddev->new_chunk_sectors = chunksect;
712 	mddev->chunk_sectors = chunksect;
713 	mddev->delta_disks = 1 - mddev->raid_disks;
714 	mddev->raid_disks = 1;
715 	/* make sure it will be not marked as dirty */
716 	mddev->recovery_cp = MaxSector;
717 	mddev_clear_unsupported_flags(mddev, UNSUPPORTED_MDDEV_FLAGS);
718 
719 	create_strip_zones(mddev, &priv_conf);
720 	return priv_conf;
721 }
722 
723 static void *raid0_takeover(struct mddev *mddev)
724 {
725 	/* raid0 can take over:
726 	 *  raid4 - if all data disks are active.
727 	 *  raid5 - providing it is Raid4 layout and one disk is faulty
728 	 *  raid10 - assuming we have all necessary active disks
729 	 *  raid1 - with (N -1) mirror drives faulty
730 	 */
731 
732 	if (mddev->bitmap) {
733 		pr_warn("md/raid0: %s: cannot takeover array with bitmap\n",
734 			mdname(mddev));
735 		return ERR_PTR(-EBUSY);
736 	}
737 	if (mddev->level == 4)
738 		return raid0_takeover_raid45(mddev);
739 
740 	if (mddev->level == 5) {
741 		if (mddev->layout == ALGORITHM_PARITY_N)
742 			return raid0_takeover_raid45(mddev);
743 
744 		pr_warn("md/raid0:%s: Raid can only takeover Raid5 with layout: %d\n",
745 			mdname(mddev), ALGORITHM_PARITY_N);
746 	}
747 
748 	if (mddev->level == 10)
749 		return raid0_takeover_raid10(mddev);
750 
751 	if (mddev->level == 1)
752 		return raid0_takeover_raid1(mddev);
753 
754 	pr_warn("Takeover from raid%i to raid0 not supported\n",
755 		mddev->level);
756 
757 	return ERR_PTR(-EINVAL);
758 }
759 
760 static void raid0_quiesce(struct mddev *mddev, int quiesce)
761 {
762 }
763 
764 static struct md_personality raid0_personality=
765 {
766 	.name		= "raid0",
767 	.level		= 0,
768 	.owner		= THIS_MODULE,
769 	.make_request	= raid0_make_request,
770 	.run		= raid0_run,
771 	.free		= raid0_free,
772 	.status		= raid0_status,
773 	.size		= raid0_size,
774 	.takeover	= raid0_takeover,
775 	.quiesce	= raid0_quiesce,
776 };
777 
778 static int __init raid0_init (void)
779 {
780 	return register_md_personality (&raid0_personality);
781 }
782 
783 static void raid0_exit (void)
784 {
785 	unregister_md_personality (&raid0_personality);
786 }
787 
788 module_init(raid0_init);
789 module_exit(raid0_exit);
790 MODULE_LICENSE("GPL");
791 MODULE_DESCRIPTION("RAID0 (striping) personality for MD");
792 MODULE_ALIAS("md-personality-2"); /* RAID0 */
793 MODULE_ALIAS("md-raid0");
794 MODULE_ALIAS("md-level-0");
795