xref: /linux/drivers/md/raid0.c (revision 23b0f90ba871f096474e1c27c3d14f455189d2d9)
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 	 (1L << MD_FAILLAST_DEV) |	\
32 	 (1L << MD_SERIALIZE_POLICY))
33 
34 /*
35  * inform the user of the raid configuration
36 */
37 static void dump_zones(struct mddev *mddev)
38 {
39 	int j, k;
40 	sector_t zone_size = 0;
41 	sector_t zone_start = 0;
42 	struct r0conf *conf = mddev->private;
43 	int raid_disks = conf->strip_zone[0].nb_dev;
44 	pr_debug("md: RAID0 configuration for %s - %d zone%s\n",
45 		 mdname(mddev),
46 		 conf->nr_strip_zones, conf->nr_strip_zones==1?"":"s");
47 	for (j = 0; j < conf->nr_strip_zones; j++) {
48 		char line[200];
49 		int len = 0;
50 
51 		for (k = 0; k < conf->strip_zone[j].nb_dev; k++)
52 			len += scnprintf(line+len, 200-len, "%s%pg", k?"/":"",
53 				conf->devlist[j * raid_disks + k]->bdev);
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 	struct r0conf *conf = kzalloc(sizeof(*conf), GFP_KERNEL);
73 	unsigned int blksize = 512;
74 
75 	if (!mddev_is_dm(mddev))
76 		blksize = queue_logical_block_size(mddev->gendisk->queue);
77 
78 	*private_conf = ERR_PTR(-ENOMEM);
79 	if (!conf)
80 		return -ENOMEM;
81 	rdev_for_each(rdev1, mddev) {
82 		pr_debug("md/raid0:%s: looking at %pg\n",
83 			 mdname(mddev),
84 			 rdev1->bdev);
85 		c = 0;
86 
87 		/* round size to chunk_size */
88 		sectors = rdev1->sectors;
89 		sector_div(sectors, mddev->chunk_sectors);
90 		rdev1->sectors = sectors * mddev->chunk_sectors;
91 
92 		if (mddev_is_dm(mddev))
93 			blksize = max(blksize, queue_logical_block_size(
94 				      rdev1->bdev->bd_disk->queue));
95 
96 		rdev_for_each(rdev2, mddev) {
97 			pr_debug("md/raid0:%s:   comparing %pg(%llu)"
98 				 " with %pg(%llu)\n",
99 				 mdname(mddev),
100 				 rdev1->bdev,
101 				 (unsigned long long)rdev1->sectors,
102 				 rdev2->bdev,
103 				 (unsigned long long)rdev2->sectors);
104 			if (rdev2 == rdev1) {
105 				pr_debug("md/raid0:%s:   END\n",
106 					 mdname(mddev));
107 				break;
108 			}
109 			if (rdev2->sectors == rdev1->sectors) {
110 				/*
111 				 * Not unique, don't count it as a new
112 				 * group
113 				 */
114 				pr_debug("md/raid0:%s:   EQUAL\n",
115 					 mdname(mddev));
116 				c = 1;
117 				break;
118 			}
119 			pr_debug("md/raid0:%s:   NOT EQUAL\n",
120 				 mdname(mddev));
121 		}
122 		if (!c) {
123 			pr_debug("md/raid0:%s:   ==> UNIQUE\n",
124 				 mdname(mddev));
125 			conf->nr_strip_zones++;
126 			pr_debug("md/raid0:%s: %d zones\n",
127 				 mdname(mddev), conf->nr_strip_zones);
128 		}
129 	}
130 	pr_debug("md/raid0:%s: FINAL %d zones\n",
131 		 mdname(mddev), conf->nr_strip_zones);
132 
133 	/*
134 	 * now since we have the hard sector sizes, we can make sure
135 	 * chunk size is a multiple of that sector size
136 	 */
137 	if ((mddev->chunk_sectors << 9) % blksize) {
138 		pr_warn("md/raid0:%s: chunk_size of %d not multiple of block size %d\n",
139 			mdname(mddev),
140 			mddev->chunk_sectors << 9, blksize);
141 		err = -EINVAL;
142 		goto abort;
143 	}
144 
145 	err = -ENOMEM;
146 	conf->strip_zone = kcalloc(conf->nr_strip_zones,
147 				   sizeof(struct strip_zone),
148 				   GFP_KERNEL);
149 	if (!conf->strip_zone)
150 		goto abort;
151 	conf->devlist = kzalloc(array3_size(sizeof(struct md_rdev *),
152 					    conf->nr_strip_zones,
153 					    mddev->raid_disks),
154 				GFP_KERNEL);
155 	if (!conf->devlist)
156 		goto abort;
157 
158 	/* The first zone must contain all devices, so here we check that
159 	 * there is a proper alignment of slots to devices and find them all
160 	 */
161 	zone = &conf->strip_zone[0];
162 	cnt = 0;
163 	smallest = NULL;
164 	dev = conf->devlist;
165 	err = -EINVAL;
166 	rdev_for_each(rdev1, mddev) {
167 		int j = rdev1->raid_disk;
168 
169 		if (mddev->level == 10) {
170 			/* taking over a raid10-n2 array */
171 			j /= 2;
172 			rdev1->new_raid_disk = j;
173 		}
174 
175 		if (mddev->level == 1) {
176 			/* taiking over a raid1 array-
177 			 * we have only one active disk
178 			 */
179 			j = 0;
180 			rdev1->new_raid_disk = j;
181 		}
182 
183 		if (j < 0) {
184 			pr_warn("md/raid0:%s: remove inactive devices before converting to RAID0\n",
185 				mdname(mddev));
186 			goto abort;
187 		}
188 		if (j >= mddev->raid_disks) {
189 			pr_warn("md/raid0:%s: bad disk number %d - aborting!\n",
190 				mdname(mddev), j);
191 			goto abort;
192 		}
193 		if (dev[j]) {
194 			pr_warn("md/raid0:%s: multiple devices for %d - aborting!\n",
195 				mdname(mddev), j);
196 			goto abort;
197 		}
198 		dev[j] = rdev1;
199 
200 		if (!smallest || (rdev1->sectors < smallest->sectors))
201 			smallest = rdev1;
202 		cnt++;
203 	}
204 	if (cnt != mddev->raid_disks) {
205 		pr_warn("md/raid0:%s: too few disks (%d of %d) - aborting!\n",
206 			mdname(mddev), cnt, mddev->raid_disks);
207 		goto abort;
208 	}
209 	zone->nb_dev = cnt;
210 	zone->zone_end = smallest->sectors * cnt;
211 
212 	curr_zone_end = zone->zone_end;
213 
214 	/* now do the other zones */
215 	for (i = 1; i < conf->nr_strip_zones; i++)
216 	{
217 		int j;
218 
219 		zone = conf->strip_zone + i;
220 		dev = conf->devlist + i * mddev->raid_disks;
221 
222 		pr_debug("md/raid0:%s: zone %d\n", mdname(mddev), i);
223 		zone->dev_start = smallest->sectors;
224 		smallest = NULL;
225 		c = 0;
226 
227 		for (j=0; j<cnt; j++) {
228 			rdev = conf->devlist[j];
229 			if (rdev->sectors <= zone->dev_start) {
230 				pr_debug("md/raid0:%s: checking %pg ... nope\n",
231 					 mdname(mddev),
232 					 rdev->bdev);
233 				continue;
234 			}
235 			pr_debug("md/raid0:%s: checking %pg ..."
236 				 " contained as device %d\n",
237 				 mdname(mddev),
238 				 rdev->bdev, c);
239 			dev[c] = rdev;
240 			c++;
241 			if (!smallest || rdev->sectors < smallest->sectors) {
242 				smallest = rdev;
243 				pr_debug("md/raid0:%s:  (%llu) is smallest!.\n",
244 					 mdname(mddev),
245 					 (unsigned long long)rdev->sectors);
246 			}
247 		}
248 
249 		zone->nb_dev = c;
250 		sectors = (smallest->sectors - zone->dev_start) * c;
251 		pr_debug("md/raid0:%s: zone->nb_dev: %d, sectors: %llu\n",
252 			 mdname(mddev),
253 			 zone->nb_dev, (unsigned long long)sectors);
254 
255 		curr_zone_end += sectors;
256 		zone->zone_end = curr_zone_end;
257 
258 		pr_debug("md/raid0:%s: current zone start: %llu\n",
259 			 mdname(mddev),
260 			 (unsigned long long)smallest->sectors);
261 	}
262 
263 	if (conf->nr_strip_zones == 1 || conf->strip_zone[1].nb_dev == 1) {
264 		conf->layout = RAID0_ORIG_LAYOUT;
265 	} else if (mddev->layout == RAID0_ORIG_LAYOUT ||
266 		   mddev->layout == RAID0_ALT_MULTIZONE_LAYOUT) {
267 		conf->layout = mddev->layout;
268 	} else if (default_layout == RAID0_ORIG_LAYOUT ||
269 		   default_layout == RAID0_ALT_MULTIZONE_LAYOUT) {
270 		conf->layout = default_layout;
271 	} else {
272 		pr_err("md/raid0:%s: cannot assemble multi-zone RAID0 with default_layout setting\n",
273 		       mdname(mddev));
274 		pr_err("md/raid0: please set raid0.default_layout to 1 or 2\n");
275 		err = -EOPNOTSUPP;
276 		goto abort;
277 	}
278 
279 	if (conf->layout == RAID0_ORIG_LAYOUT) {
280 		for (i = 1; i < conf->nr_strip_zones; i++) {
281 			sector_t first_sector = conf->strip_zone[i-1].zone_end;
282 
283 			sector_div(first_sector, mddev->chunk_sectors);
284 			zone = conf->strip_zone + i;
285 			/* disk_shift is first disk index used in the zone */
286 			zone->disk_shift = sector_div(first_sector,
287 						      zone->nb_dev);
288 		}
289 	}
290 
291 	pr_debug("md/raid0:%s: done.\n", mdname(mddev));
292 	*private_conf = conf;
293 
294 	return 0;
295 abort:
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  */
306 static 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 */
326 static 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 
359 static 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 
374 static void raid0_free(struct mddev *mddev, void *priv)
375 {
376 	struct r0conf *conf = priv;
377 
378 	kfree(conf->strip_zone);
379 	kfree(conf->devlist);
380 	kfree(conf);
381 }
382 
383 static int raid0_set_limits(struct mddev *mddev)
384 {
385 	struct queue_limits lim;
386 	int err;
387 
388 	md_init_stacking_limits(&lim);
389 	lim.max_hw_sectors = mddev->chunk_sectors;
390 	lim.max_write_zeroes_sectors = mddev->chunk_sectors;
391 	lim.max_hw_wzeroes_unmap_sectors = mddev->chunk_sectors;
392 	lim.logical_block_size = mddev->logical_block_size;
393 	lim.io_min = mddev->chunk_sectors << 9;
394 	lim.io_opt = lim.io_min * mddev->raid_disks;
395 	lim.chunk_sectors = mddev->chunk_sectors;
396 	lim.features |= BLK_FEAT_ATOMIC_WRITES;
397 	err = mddev_stack_rdev_limits(mddev, &lim, MDDEV_STACK_INTEGRITY);
398 	if (err)
399 		return err;
400 	return queue_limits_set(mddev->gendisk->queue, &lim);
401 }
402 
403 static int raid0_run(struct mddev *mddev)
404 {
405 	struct r0conf *conf;
406 	int ret;
407 
408 	if (mddev->chunk_sectors == 0) {
409 		pr_warn("md/raid0:%s: chunk size must be set.\n", mdname(mddev));
410 		return -EINVAL;
411 	}
412 	if (md_check_no_bitmap(mddev))
413 		return -EINVAL;
414 
415 	if (!mddev_is_dm(mddev)) {
416 		ret = raid0_set_limits(mddev);
417 		if (ret)
418 			return ret;
419 	}
420 
421 	/* if private is not null, we are here after takeover */
422 	if (mddev->private == NULL) {
423 		ret = create_strip_zones(mddev, &conf);
424 		if (ret < 0)
425 			return ret;
426 		mddev->private = conf;
427 	}
428 	conf = mddev->private;
429 
430 	/* calculate array device size */
431 	md_set_array_sectors(mddev, raid0_size(mddev, 0, 0));
432 
433 	pr_debug("md/raid0:%s: md_size is %llu sectors.\n",
434 		 mdname(mddev),
435 		 (unsigned long long)mddev->array_sectors);
436 
437 	dump_zones(mddev);
438 
439 	return md_integrity_register(mddev);
440 }
441 
442 /*
443  * Convert disk_index to the disk order in which it is read/written.
444  *  For example, if we have 4 disks, they are numbered 0,1,2,3. If we
445  *  write the disks starting at disk 3, then the read/write order would
446  *  be disk 3, then 0, then 1, and then disk 2 and we want map_disk_shift()
447  *  to map the disks as follows 0,1,2,3 => 1,2,3,0. So disk 0 would map
448  *  to 1, 1 to 2, 2 to 3, and 3 to 0. That way we can compare disks in
449  *  that 'output' space to understand the read/write disk ordering.
450  */
451 static int map_disk_shift(int disk_index, int num_disks, int disk_shift)
452 {
453 	return ((disk_index + num_disks - disk_shift) % num_disks);
454 }
455 
456 static void raid0_handle_discard(struct mddev *mddev, struct bio *bio)
457 {
458 	struct r0conf *conf = mddev->private;
459 	struct strip_zone *zone;
460 	sector_t start = bio->bi_iter.bi_sector;
461 	sector_t end;
462 	unsigned int stripe_size;
463 	sector_t first_stripe_index, last_stripe_index;
464 	sector_t start_disk_offset;
465 	unsigned int start_disk_index;
466 	sector_t end_disk_offset;
467 	unsigned int end_disk_index;
468 	unsigned int disk;
469 	sector_t orig_start, orig_end;
470 
471 	orig_start = start;
472 	zone = find_zone(conf, &start);
473 
474 	if (bio_end_sector(bio) > zone->zone_end) {
475 		bio = bio_submit_split_bioset(bio,
476 				zone->zone_end - bio->bi_iter.bi_sector,
477 				&mddev->bio_set);
478 		if (!bio)
479 			return;
480 
481 		end = zone->zone_end;
482 	} else {
483 		end = bio_end_sector(bio);
484 	}
485 
486 	orig_end = end;
487 	if (zone != conf->strip_zone)
488 		end = end - zone[-1].zone_end;
489 
490 	/* Now start and end is the offset in zone */
491 	stripe_size = zone->nb_dev * mddev->chunk_sectors;
492 
493 	first_stripe_index = start;
494 	sector_div(first_stripe_index, stripe_size);
495 	last_stripe_index = end;
496 	sector_div(last_stripe_index, stripe_size);
497 
498 	/* In the first zone the original and alternate layouts are the same */
499 	if ((conf->layout == RAID0_ORIG_LAYOUT) && (zone != conf->strip_zone)) {
500 		sector_div(orig_start, mddev->chunk_sectors);
501 		start_disk_index = sector_div(orig_start, zone->nb_dev);
502 		start_disk_index = map_disk_shift(start_disk_index,
503 						  zone->nb_dev,
504 						  zone->disk_shift);
505 		sector_div(orig_end, mddev->chunk_sectors);
506 		end_disk_index = sector_div(orig_end, zone->nb_dev);
507 		end_disk_index = map_disk_shift(end_disk_index,
508 						zone->nb_dev, zone->disk_shift);
509 	} else {
510 		start_disk_index = (int)(start - first_stripe_index * stripe_size) /
511 			mddev->chunk_sectors;
512 		end_disk_index = (int)(end - last_stripe_index * stripe_size) /
513 			mddev->chunk_sectors;
514 	}
515 	start_disk_offset = ((int)(start - first_stripe_index * stripe_size) %
516 		mddev->chunk_sectors) +
517 		first_stripe_index * mddev->chunk_sectors;
518 	end_disk_offset = ((int)(end - last_stripe_index * stripe_size) %
519 		mddev->chunk_sectors) +
520 		last_stripe_index * mddev->chunk_sectors;
521 
522 	for (disk = 0; disk < zone->nb_dev; disk++) {
523 		sector_t dev_start, dev_end;
524 		struct md_rdev *rdev;
525 		int compare_disk;
526 
527 		compare_disk = map_disk_shift(disk, zone->nb_dev,
528 					      zone->disk_shift);
529 
530 		if (compare_disk < start_disk_index)
531 			dev_start = (first_stripe_index + 1) *
532 				mddev->chunk_sectors;
533 		else if (compare_disk > start_disk_index)
534 			dev_start = first_stripe_index * mddev->chunk_sectors;
535 		else
536 			dev_start = start_disk_offset;
537 
538 		if (compare_disk < end_disk_index)
539 			dev_end = (last_stripe_index + 1) * mddev->chunk_sectors;
540 		else if (compare_disk > end_disk_index)
541 			dev_end = last_stripe_index * mddev->chunk_sectors;
542 		else
543 			dev_end = end_disk_offset;
544 
545 		if (dev_end <= dev_start)
546 			continue;
547 
548 		rdev = conf->devlist[(zone - conf->strip_zone) *
549 			conf->strip_zone[0].nb_dev + disk];
550 		md_submit_discard_bio(mddev, rdev, bio,
551 			dev_start + zone->dev_start + rdev->data_offset,
552 			dev_end - dev_start);
553 	}
554 	bio_endio(bio);
555 }
556 
557 static void raid0_map_submit_bio(struct mddev *mddev, struct bio *bio)
558 {
559 	struct r0conf *conf = mddev->private;
560 	struct strip_zone *zone;
561 	struct md_rdev *tmp_dev;
562 	sector_t bio_sector = bio->bi_iter.bi_sector;
563 	sector_t sector = bio_sector;
564 
565 	md_account_bio(mddev, &bio);
566 
567 	zone = find_zone(mddev->private, &sector);
568 	switch (conf->layout) {
569 	case RAID0_ORIG_LAYOUT:
570 		tmp_dev = map_sector(mddev, zone, bio_sector, &sector);
571 		break;
572 	case RAID0_ALT_MULTIZONE_LAYOUT:
573 		tmp_dev = map_sector(mddev, zone, sector, &sector);
574 		break;
575 	default:
576 		WARN(1, "md/raid0:%s: Invalid layout\n", mdname(mddev));
577 		bio_io_error(bio);
578 		return;
579 	}
580 
581 	if (unlikely(is_rdev_broken(tmp_dev))) {
582 		bio_io_error(bio);
583 		md_error(mddev, tmp_dev);
584 		return;
585 	}
586 
587 	bio_set_dev(bio, tmp_dev->bdev);
588 	bio->bi_iter.bi_sector = sector + zone->dev_start +
589 		tmp_dev->data_offset;
590 	mddev_trace_remap(mddev, bio, bio_sector);
591 	mddev_check_write_zeroes(mddev, bio);
592 	submit_bio_noacct(bio);
593 }
594 
595 static bool raid0_make_request(struct mddev *mddev, struct bio *bio)
596 {
597 	sector_t sector;
598 	unsigned chunk_sects;
599 	unsigned sectors;
600 
601 	if (unlikely(bio->bi_opf & REQ_PREFLUSH)
602 	    && md_flush_request(mddev, bio))
603 		return true;
604 
605 	if (unlikely((bio_op(bio) == REQ_OP_DISCARD))) {
606 		raid0_handle_discard(mddev, bio);
607 		return true;
608 	}
609 
610 	sector = bio->bi_iter.bi_sector;
611 	chunk_sects = mddev->chunk_sectors;
612 
613 	sectors = chunk_sects -
614 		(likely(is_power_of_2(chunk_sects))
615 		 ? (sector & (chunk_sects-1))
616 		 : sector_div(sector, chunk_sects));
617 
618 	if (sectors < bio_sectors(bio)) {
619 		bio = bio_submit_split_bioset(bio, sectors,
620 					      &mddev->bio_set);
621 		if (!bio)
622 			return true;
623 	}
624 
625 	raid0_map_submit_bio(mddev, bio);
626 	return true;
627 }
628 
629 static void raid0_status(struct seq_file *seq, struct mddev *mddev)
630 {
631 	seq_printf(seq, " %dk chunks", mddev->chunk_sectors / 2);
632 	return;
633 }
634 
635 static void raid0_error(struct mddev *mddev, struct md_rdev *rdev)
636 {
637 	if (!test_and_set_bit(MD_BROKEN, &mddev->flags)) {
638 		char *md_name = mdname(mddev);
639 
640 		pr_crit("md/raid0%s: Disk failure on %pg detected, failing array.\n",
641 			md_name, rdev->bdev);
642 	}
643 }
644 
645 static void *raid0_takeover_raid45(struct mddev *mddev)
646 {
647 	struct md_rdev *rdev;
648 	struct r0conf *priv_conf;
649 
650 	if (mddev->degraded != 1) {
651 		pr_warn("md/raid0:%s: raid5 must be degraded! Degraded disks: %d\n",
652 			mdname(mddev),
653 			mddev->degraded);
654 		return ERR_PTR(-EINVAL);
655 	}
656 
657 	rdev_for_each(rdev, mddev) {
658 		/* check slot number for a disk */
659 		if (rdev->raid_disk == mddev->raid_disks-1) {
660 			pr_warn("md/raid0:%s: raid5 must have missing parity disk!\n",
661 				mdname(mddev));
662 			return ERR_PTR(-EINVAL);
663 		}
664 		rdev->sectors = mddev->dev_sectors;
665 	}
666 
667 	/* Set new parameters */
668 	mddev->new_level = 0;
669 	mddev->new_layout = 0;
670 	mddev->new_chunk_sectors = mddev->chunk_sectors;
671 	mddev->raid_disks--;
672 	mddev->delta_disks = -1;
673 	/* make sure it will be not marked as dirty */
674 	mddev->resync_offset = MaxSector;
675 	mddev_clear_unsupported_flags(mddev, UNSUPPORTED_MDDEV_FLAGS);
676 
677 	create_strip_zones(mddev, &priv_conf);
678 
679 	return priv_conf;
680 }
681 
682 static void *raid0_takeover_raid10(struct mddev *mddev)
683 {
684 	struct r0conf *priv_conf;
685 
686 	/* Check layout:
687 	 *  - far_copies must be 1
688 	 *  - near_copies must be 2
689 	 *  - disks number must be even
690 	 *  - all mirrors must be already degraded
691 	 */
692 	if (mddev->layout != ((1 << 8) + 2)) {
693 		pr_warn("md/raid0:%s:: Raid0 cannot takeover layout: 0x%x\n",
694 			mdname(mddev),
695 			mddev->layout);
696 		return ERR_PTR(-EINVAL);
697 	}
698 	if (mddev->raid_disks & 1) {
699 		pr_warn("md/raid0:%s: Raid0 cannot takeover Raid10 with odd disk number.\n",
700 			mdname(mddev));
701 		return ERR_PTR(-EINVAL);
702 	}
703 	if (mddev->degraded != (mddev->raid_disks>>1)) {
704 		pr_warn("md/raid0:%s: All mirrors must be already degraded!\n",
705 			mdname(mddev));
706 		return ERR_PTR(-EINVAL);
707 	}
708 
709 	/* Set new parameters */
710 	mddev->new_level = 0;
711 	mddev->new_layout = 0;
712 	mddev->new_chunk_sectors = mddev->chunk_sectors;
713 	mddev->delta_disks = - mddev->raid_disks / 2;
714 	mddev->raid_disks += mddev->delta_disks;
715 	mddev->degraded = 0;
716 	/* make sure it will be not marked as dirty */
717 	mddev->resync_offset = MaxSector;
718 	mddev_clear_unsupported_flags(mddev, UNSUPPORTED_MDDEV_FLAGS);
719 
720 	create_strip_zones(mddev, &priv_conf);
721 	return priv_conf;
722 }
723 
724 static void *raid0_takeover_raid1(struct mddev *mddev)
725 {
726 	struct r0conf *priv_conf;
727 	int chunksect;
728 
729 	/* Check layout:
730 	 *  - (N - 1) mirror drives must be already faulty
731 	 */
732 	if ((mddev->raid_disks - 1) != mddev->degraded) {
733 		pr_err("md/raid0:%s: (N - 1) mirrors drives must be already faulty!\n",
734 		       mdname(mddev));
735 		return ERR_PTR(-EINVAL);
736 	}
737 
738 	/*
739 	 * a raid1 doesn't have the notion of chunk size, so
740 	 * figure out the largest suitable size we can use.
741 	 */
742 	chunksect = 64 * 2; /* 64K by default */
743 
744 	/* The array must be an exact multiple of chunksize */
745 	while (chunksect && (mddev->array_sectors & (chunksect - 1)))
746 		chunksect >>= 1;
747 
748 	if ((chunksect << 9) < PAGE_SIZE)
749 		/* array size does not allow a suitable chunk size */
750 		return ERR_PTR(-EINVAL);
751 
752 	/* Set new parameters */
753 	mddev->new_level = 0;
754 	mddev->new_layout = 0;
755 	mddev->new_chunk_sectors = chunksect;
756 	mddev->chunk_sectors = chunksect;
757 	mddev->delta_disks = 1 - mddev->raid_disks;
758 	mddev->raid_disks = 1;
759 	/* make sure it will be not marked as dirty */
760 	mddev->resync_offset = MaxSector;
761 	mddev_clear_unsupported_flags(mddev, UNSUPPORTED_MDDEV_FLAGS);
762 
763 	create_strip_zones(mddev, &priv_conf);
764 	return priv_conf;
765 }
766 
767 static void *raid0_takeover(struct mddev *mddev)
768 {
769 	/* raid0 can take over:
770 	 *  raid4 - if all data disks are active.
771 	 *  raid5 - providing it is Raid4 layout and one disk is faulty
772 	 *  raid10 - assuming we have all necessary active disks
773 	 *  raid1 - with (N -1) mirror drives faulty
774 	 */
775 
776 	if (mddev->bitmap) {
777 		pr_warn("md/raid0: %s: cannot takeover array with bitmap\n",
778 			mdname(mddev));
779 		return ERR_PTR(-EBUSY);
780 	}
781 	if (mddev->level == 4)
782 		return raid0_takeover_raid45(mddev);
783 
784 	if (mddev->level == 5) {
785 		if (mddev->layout == ALGORITHM_PARITY_N)
786 			return raid0_takeover_raid45(mddev);
787 
788 		pr_warn("md/raid0:%s: Raid can only takeover Raid5 with layout: %d\n",
789 			mdname(mddev), ALGORITHM_PARITY_N);
790 	}
791 
792 	if (mddev->level == 10)
793 		return raid0_takeover_raid10(mddev);
794 
795 	if (mddev->level == 1)
796 		return raid0_takeover_raid1(mddev);
797 
798 	pr_warn("Takeover from raid%i to raid0 not supported\n",
799 		mddev->level);
800 
801 	return ERR_PTR(-EINVAL);
802 }
803 
804 static void raid0_quiesce(struct mddev *mddev, int quiesce)
805 {
806 }
807 
808 static struct md_personality raid0_personality=
809 {
810 	.head = {
811 		.type	= MD_PERSONALITY,
812 		.id	= ID_RAID0,
813 		.name	= "raid0",
814 		.owner	= THIS_MODULE,
815 	},
816 
817 	.make_request	= raid0_make_request,
818 	.run		= raid0_run,
819 	.free		= raid0_free,
820 	.status		= raid0_status,
821 	.size		= raid0_size,
822 	.takeover	= raid0_takeover,
823 	.quiesce	= raid0_quiesce,
824 	.error_handler	= raid0_error,
825 };
826 
827 static int __init raid0_init(void)
828 {
829 	return register_md_submodule(&raid0_personality.head);
830 }
831 
832 static void __exit raid0_exit(void)
833 {
834 	unregister_md_submodule(&raid0_personality.head);
835 }
836 
837 module_init(raid0_init);
838 module_exit(raid0_exit);
839 MODULE_LICENSE("GPL");
840 MODULE_DESCRIPTION("RAID0 (striping) personality for MD");
841 MODULE_ALIAS("md-personality-2"); /* RAID0 */
842 MODULE_ALIAS("md-raid0");
843 MODULE_ALIAS("md-level-0");
844