xref: /linux/drivers/md/raid0.c (revision 2ba9268dd603d23e17643437b2246acb6844953b)
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 
28 static 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 */
46 static 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 	printk(KERN_INFO "\n");
74 }
75 
76 static int create_strip_zones(struct mddev *mddev, struct r0conf **private_conf)
77 {
78 	int i, c, err;
79 	sector_t curr_zone_end, sectors;
80 	struct md_rdev *smallest, *rdev1, *rdev2, *rdev, **dev;
81 	struct strip_zone *zone;
82 	int cnt;
83 	char b[BDEVNAME_SIZE];
84 	char b2[BDEVNAME_SIZE];
85 	struct r0conf *conf = kzalloc(sizeof(*conf), GFP_KERNEL);
86 	bool discard_supported = false;
87 
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 		rdev_for_each(rdev2, mddev) {
102 			pr_debug("md/raid0:%s:   comparing %s(%llu)"
103 				 " with %s(%llu)\n",
104 				 mdname(mddev),
105 				 bdevname(rdev1->bdev,b),
106 				 (unsigned long long)rdev1->sectors,
107 				 bdevname(rdev2->bdev,b2),
108 				 (unsigned long long)rdev2->sectors);
109 			if (rdev2 == rdev1) {
110 				pr_debug("md/raid0:%s:   END\n",
111 					 mdname(mddev));
112 				break;
113 			}
114 			if (rdev2->sectors == rdev1->sectors) {
115 				/*
116 				 * Not unique, don't count it as a new
117 				 * group
118 				 */
119 				pr_debug("md/raid0:%s:   EQUAL\n",
120 					 mdname(mddev));
121 				c = 1;
122 				break;
123 			}
124 			pr_debug("md/raid0:%s:   NOT EQUAL\n",
125 				 mdname(mddev));
126 		}
127 		if (!c) {
128 			pr_debug("md/raid0:%s:   ==> UNIQUE\n",
129 				 mdname(mddev));
130 			conf->nr_strip_zones++;
131 			pr_debug("md/raid0:%s: %d zones\n",
132 				 mdname(mddev), conf->nr_strip_zones);
133 		}
134 	}
135 	pr_debug("md/raid0:%s: FINAL %d zones\n",
136 		 mdname(mddev), conf->nr_strip_zones);
137 	err = -ENOMEM;
138 	conf->strip_zone = kzalloc(sizeof(struct strip_zone)*
139 				conf->nr_strip_zones, GFP_KERNEL);
140 	if (!conf->strip_zone)
141 		goto abort;
142 	conf->devlist = kzalloc(sizeof(struct md_rdev*)*
143 				conf->nr_strip_zones*mddev->raid_disks,
144 				GFP_KERNEL);
145 	if (!conf->devlist)
146 		goto abort;
147 
148 	/* The first zone must contain all devices, so here we check that
149 	 * there is a proper alignment of slots to devices and find them all
150 	 */
151 	zone = &conf->strip_zone[0];
152 	cnt = 0;
153 	smallest = NULL;
154 	dev = conf->devlist;
155 	err = -EINVAL;
156 	rdev_for_each(rdev1, mddev) {
157 		int j = rdev1->raid_disk;
158 
159 		if (mddev->level == 10) {
160 			/* taking over a raid10-n2 array */
161 			j /= 2;
162 			rdev1->new_raid_disk = j;
163 		}
164 
165 		if (mddev->level == 1) {
166 			/* taiking over a raid1 array-
167 			 * we have only one active disk
168 			 */
169 			j = 0;
170 			rdev1->new_raid_disk = j;
171 		}
172 
173 		if (j < 0) {
174 			printk(KERN_ERR
175 			       "md/raid0:%s: remove inactive devices before converting to RAID0\n",
176 			       mdname(mddev));
177 			goto abort;
178 		}
179 		if (j >= mddev->raid_disks) {
180 			printk(KERN_ERR "md/raid0:%s: bad disk number %d - "
181 			       "aborting!\n", mdname(mddev), j);
182 			goto abort;
183 		}
184 		if (dev[j]) {
185 			printk(KERN_ERR "md/raid0:%s: multiple devices for %d - "
186 			       "aborting!\n", mdname(mddev), j);
187 			goto abort;
188 		}
189 		dev[j] = rdev1;
190 
191 		disk_stack_limits(mddev->gendisk, rdev1->bdev,
192 				  rdev1->data_offset << 9);
193 
194 		if (rdev1->bdev->bd_disk->queue->merge_bvec_fn)
195 			conf->has_merge_bvec = 1;
196 
197 		if (!smallest || (rdev1->sectors < smallest->sectors))
198 			smallest = rdev1;
199 		cnt++;
200 
201 		if (blk_queue_discard(bdev_get_queue(rdev1->bdev)))
202 			discard_supported = true;
203 	}
204 	if (cnt != mddev->raid_disks) {
205 		printk(KERN_ERR "md/raid0:%s: too few disks (%d of %d) - "
206 		       "aborting!\n", 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 %s ... nope\n",
231 					 mdname(mddev),
232 					 bdevname(rdev->bdev, b));
233 				continue;
234 			}
235 			pr_debug("md/raid0:%s: checking %s ..."
236 				 " contained as device %d\n",
237 				 mdname(mddev),
238 				 bdevname(rdev->bdev, b), 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 	/*
264 	 * now since we have the hard sector sizes, we can make sure
265 	 * chunk size is a multiple of that sector size
266 	 */
267 	if ((mddev->chunk_sectors << 9) % queue_logical_block_size(mddev->queue)) {
268 		printk(KERN_ERR "md/raid0:%s: chunk_size of %d not valid\n",
269 		       mdname(mddev),
270 		       mddev->chunk_sectors << 9);
271 		goto abort;
272 	}
273 
274 	blk_queue_io_min(mddev->queue, mddev->chunk_sectors << 9);
275 	blk_queue_io_opt(mddev->queue,
276 			 (mddev->chunk_sectors << 9) * mddev->raid_disks);
277 
278 	if (!discard_supported)
279 		queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, mddev->queue);
280 	else
281 		queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, mddev->queue);
282 
283 	pr_debug("md/raid0:%s: done.\n", mdname(mddev));
284 	*private_conf = conf;
285 
286 	return 0;
287 abort:
288 	kfree(conf->strip_zone);
289 	kfree(conf->devlist);
290 	kfree(conf);
291 	*private_conf = ERR_PTR(err);
292 	return err;
293 }
294 
295 /* Find the zone which holds a particular offset
296  * Update *sectorp to be an offset in that zone
297  */
298 static struct strip_zone *find_zone(struct r0conf *conf,
299 				    sector_t *sectorp)
300 {
301 	int i;
302 	struct strip_zone *z = conf->strip_zone;
303 	sector_t sector = *sectorp;
304 
305 	for (i = 0; i < conf->nr_strip_zones; i++)
306 		if (sector < z[i].zone_end) {
307 			if (i)
308 				*sectorp = sector - z[i-1].zone_end;
309 			return z + i;
310 		}
311 	BUG();
312 }
313 
314 /*
315  * remaps the bio to the target device. we separate two flows.
316  * power 2 flow and a general flow for the sake of perfromance
317 */
318 static struct md_rdev *map_sector(struct mddev *mddev, struct strip_zone *zone,
319 				sector_t sector, sector_t *sector_offset)
320 {
321 	unsigned int sect_in_chunk;
322 	sector_t chunk;
323 	struct r0conf *conf = mddev->private;
324 	int raid_disks = conf->strip_zone[0].nb_dev;
325 	unsigned int chunk_sects = mddev->chunk_sectors;
326 
327 	if (is_power_of_2(chunk_sects)) {
328 		int chunksect_bits = ffz(~chunk_sects);
329 		/* find the sector offset inside the chunk */
330 		sect_in_chunk  = sector & (chunk_sects - 1);
331 		sector >>= chunksect_bits;
332 		/* chunk in zone */
333 		chunk = *sector_offset;
334 		/* quotient is the chunk in real device*/
335 		sector_div(chunk, zone->nb_dev << chunksect_bits);
336 	} else{
337 		sect_in_chunk = sector_div(sector, chunk_sects);
338 		chunk = *sector_offset;
339 		sector_div(chunk, chunk_sects * zone->nb_dev);
340 	}
341 	/*
342 	*  position the bio over the real device
343 	*  real sector = chunk in device + starting of zone
344 	*	+ the position in the chunk
345 	*/
346 	*sector_offset = (chunk * chunk_sects) + sect_in_chunk;
347 	return conf->devlist[(zone - conf->strip_zone)*raid_disks
348 			     + sector_div(sector, zone->nb_dev)];
349 }
350 
351 /**
352  *	raid0_mergeable_bvec -- tell bio layer if two requests can be merged
353  *	@mddev: the md device
354  *	@bvm: properties of new bio
355  *	@biovec: the request that could be merged to it.
356  *
357  *	Return amount of bytes we can accept at this offset
358  */
359 static int raid0_mergeable_bvec(struct mddev *mddev,
360 				struct bvec_merge_data *bvm,
361 				struct bio_vec *biovec)
362 {
363 	struct r0conf *conf = mddev->private;
364 	sector_t sector = bvm->bi_sector + get_start_sect(bvm->bi_bdev);
365 	sector_t sector_offset = sector;
366 	int max;
367 	unsigned int chunk_sectors = mddev->chunk_sectors;
368 	unsigned int bio_sectors = bvm->bi_size >> 9;
369 	struct strip_zone *zone;
370 	struct md_rdev *rdev;
371 	struct request_queue *subq;
372 
373 	if (is_power_of_2(chunk_sectors))
374 		max =  (chunk_sectors - ((sector & (chunk_sectors-1))
375 						+ bio_sectors)) << 9;
376 	else
377 		max =  (chunk_sectors - (sector_div(sector, chunk_sectors)
378 						+ bio_sectors)) << 9;
379 	if (max < 0)
380 		max = 0; /* bio_add cannot handle a negative return */
381 	if (max <= biovec->bv_len && bio_sectors == 0)
382 		return biovec->bv_len;
383 	if (max < biovec->bv_len)
384 		/* too small already, no need to check further */
385 		return max;
386 	if (!conf->has_merge_bvec)
387 		return max;
388 
389 	/* May need to check subordinate device */
390 	sector = sector_offset;
391 	zone = find_zone(mddev->private, &sector_offset);
392 	rdev = map_sector(mddev, zone, sector, &sector_offset);
393 	subq = bdev_get_queue(rdev->bdev);
394 	if (subq->merge_bvec_fn) {
395 		bvm->bi_bdev = rdev->bdev;
396 		bvm->bi_sector = sector_offset + zone->dev_start +
397 			rdev->data_offset;
398 		return min(max, subq->merge_bvec_fn(subq, bvm, biovec));
399 	} else
400 		return max;
401 }
402 
403 static sector_t raid0_size(struct mddev *mddev, sector_t sectors, int raid_disks)
404 {
405 	sector_t array_sectors = 0;
406 	struct md_rdev *rdev;
407 
408 	WARN_ONCE(sectors || raid_disks,
409 		  "%s does not support generic reshape\n", __func__);
410 
411 	rdev_for_each(rdev, mddev)
412 		array_sectors += (rdev->sectors &
413 				  ~(sector_t)(mddev->chunk_sectors-1));
414 
415 	return array_sectors;
416 }
417 
418 static void raid0_free(struct mddev *mddev, void *priv);
419 
420 static int raid0_run(struct mddev *mddev)
421 {
422 	struct r0conf *conf;
423 	int ret;
424 
425 	if (mddev->chunk_sectors == 0) {
426 		printk(KERN_ERR "md/raid0:%s: chunk size must be set.\n",
427 		       mdname(mddev));
428 		return -EINVAL;
429 	}
430 	if (md_check_no_bitmap(mddev))
431 		return -EINVAL;
432 	blk_queue_max_hw_sectors(mddev->queue, mddev->chunk_sectors);
433 	blk_queue_max_write_same_sectors(mddev->queue, mddev->chunk_sectors);
434 	blk_queue_max_discard_sectors(mddev->queue, mddev->chunk_sectors);
435 
436 	/* if private is not null, we are here after takeover */
437 	if (mddev->private == NULL) {
438 		ret = create_strip_zones(mddev, &conf);
439 		if (ret < 0)
440 			return ret;
441 		mddev->private = conf;
442 	}
443 	conf = mddev->private;
444 
445 	/* calculate array device size */
446 	md_set_array_sectors(mddev, raid0_size(mddev, 0, 0));
447 
448 	printk(KERN_INFO "md/raid0:%s: md_size is %llu sectors.\n",
449 	       mdname(mddev),
450 	       (unsigned long long)mddev->array_sectors);
451 	/* calculate the max read-ahead size.
452 	 * For read-ahead of large files to be effective, we need to
453 	 * readahead at least twice a whole stripe. i.e. number of devices
454 	 * multiplied by chunk size times 2.
455 	 * If an individual device has an ra_pages greater than the
456 	 * chunk size, then we will not drive that device as hard as it
457 	 * wants.  We consider this a configuration error: a larger
458 	 * chunksize should be used in that case.
459 	 */
460 	{
461 		int stripe = mddev->raid_disks *
462 			(mddev->chunk_sectors << 9) / PAGE_SIZE;
463 		if (mddev->queue->backing_dev_info.ra_pages < 2* stripe)
464 			mddev->queue->backing_dev_info.ra_pages = 2* stripe;
465 	}
466 
467 	dump_zones(mddev);
468 
469 	ret = md_integrity_register(mddev);
470 
471 	return ret;
472 }
473 
474 static void raid0_free(struct mddev *mddev, void *priv)
475 {
476 	struct r0conf *conf = priv;
477 
478 	kfree(conf->strip_zone);
479 	kfree(conf->devlist);
480 	kfree(conf);
481 }
482 
483 /*
484  * Is io distribute over 1 or more chunks ?
485 */
486 static inline int is_io_in_chunk_boundary(struct mddev *mddev,
487 			unsigned int chunk_sects, struct bio *bio)
488 {
489 	if (likely(is_power_of_2(chunk_sects))) {
490 		return chunk_sects >=
491 			((bio->bi_iter.bi_sector & (chunk_sects-1))
492 					+ bio_sectors(bio));
493 	} else{
494 		sector_t sector = bio->bi_iter.bi_sector;
495 		return chunk_sects >= (sector_div(sector, chunk_sects)
496 						+ bio_sectors(bio));
497 	}
498 }
499 
500 static void raid0_make_request(struct mddev *mddev, struct bio *bio)
501 {
502 	struct strip_zone *zone;
503 	struct md_rdev *tmp_dev;
504 	struct bio *split;
505 
506 	if (unlikely(bio->bi_rw & REQ_FLUSH)) {
507 		md_flush_request(mddev, bio);
508 		return;
509 	}
510 
511 	do {
512 		sector_t sector = bio->bi_iter.bi_sector;
513 		unsigned chunk_sects = mddev->chunk_sectors;
514 
515 		unsigned sectors = chunk_sects -
516 			(likely(is_power_of_2(chunk_sects))
517 			 ? (sector & (chunk_sects-1))
518 			 : sector_div(sector, chunk_sects));
519 
520 		if (sectors < bio_sectors(bio)) {
521 			split = bio_split(bio, sectors, GFP_NOIO, fs_bio_set);
522 			bio_chain(split, bio);
523 		} else {
524 			split = bio;
525 		}
526 
527 		zone = find_zone(mddev->private, &sector);
528 		tmp_dev = map_sector(mddev, zone, sector, &sector);
529 		split->bi_bdev = tmp_dev->bdev;
530 		split->bi_iter.bi_sector = sector + zone->dev_start +
531 			tmp_dev->data_offset;
532 
533 		if (unlikely((split->bi_rw & REQ_DISCARD) &&
534 			 !blk_queue_discard(bdev_get_queue(split->bi_bdev)))) {
535 			/* Just ignore it */
536 			bio_endio(split, 0);
537 		} else
538 			generic_make_request(split);
539 	} while (split != bio);
540 }
541 
542 static void raid0_status(struct seq_file *seq, struct mddev *mddev)
543 {
544 	seq_printf(seq, " %dk chunks", mddev->chunk_sectors / 2);
545 	return;
546 }
547 
548 static void *raid0_takeover_raid45(struct mddev *mddev)
549 {
550 	struct md_rdev *rdev;
551 	struct r0conf *priv_conf;
552 
553 	if (mddev->degraded != 1) {
554 		printk(KERN_ERR "md/raid0:%s: raid5 must be degraded! Degraded disks: %d\n",
555 		       mdname(mddev),
556 		       mddev->degraded);
557 		return ERR_PTR(-EINVAL);
558 	}
559 
560 	rdev_for_each(rdev, mddev) {
561 		/* check slot number for a disk */
562 		if (rdev->raid_disk == mddev->raid_disks-1) {
563 			printk(KERN_ERR "md/raid0:%s: raid5 must have missing parity disk!\n",
564 			       mdname(mddev));
565 			return ERR_PTR(-EINVAL);
566 		}
567 		rdev->sectors = mddev->dev_sectors;
568 	}
569 
570 	/* Set new parameters */
571 	mddev->new_level = 0;
572 	mddev->new_layout = 0;
573 	mddev->new_chunk_sectors = mddev->chunk_sectors;
574 	mddev->raid_disks--;
575 	mddev->delta_disks = -1;
576 	/* make sure it will be not marked as dirty */
577 	mddev->recovery_cp = MaxSector;
578 
579 	create_strip_zones(mddev, &priv_conf);
580 	return priv_conf;
581 }
582 
583 static void *raid0_takeover_raid10(struct mddev *mddev)
584 {
585 	struct r0conf *priv_conf;
586 
587 	/* Check layout:
588 	 *  - far_copies must be 1
589 	 *  - near_copies must be 2
590 	 *  - disks number must be even
591 	 *  - all mirrors must be already degraded
592 	 */
593 	if (mddev->layout != ((1 << 8) + 2)) {
594 		printk(KERN_ERR "md/raid0:%s:: Raid0 cannot takover layout: 0x%x\n",
595 		       mdname(mddev),
596 		       mddev->layout);
597 		return ERR_PTR(-EINVAL);
598 	}
599 	if (mddev->raid_disks & 1) {
600 		printk(KERN_ERR "md/raid0:%s: Raid0 cannot takover Raid10 with odd disk number.\n",
601 		       mdname(mddev));
602 		return ERR_PTR(-EINVAL);
603 	}
604 	if (mddev->degraded != (mddev->raid_disks>>1)) {
605 		printk(KERN_ERR "md/raid0:%s: All mirrors must be already degraded!\n",
606 		       mdname(mddev));
607 		return ERR_PTR(-EINVAL);
608 	}
609 
610 	/* Set new parameters */
611 	mddev->new_level = 0;
612 	mddev->new_layout = 0;
613 	mddev->new_chunk_sectors = mddev->chunk_sectors;
614 	mddev->delta_disks = - mddev->raid_disks / 2;
615 	mddev->raid_disks += mddev->delta_disks;
616 	mddev->degraded = 0;
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 
624 static void *raid0_takeover_raid1(struct mddev *mddev)
625 {
626 	struct r0conf *priv_conf;
627 	int chunksect;
628 
629 	/* Check layout:
630 	 *  - (N - 1) mirror drives must be already faulty
631 	 */
632 	if ((mddev->raid_disks - 1) != mddev->degraded) {
633 		printk(KERN_ERR "md/raid0:%s: (N - 1) mirrors drives must be already faulty!\n",
634 		       mdname(mddev));
635 		return ERR_PTR(-EINVAL);
636 	}
637 
638 	/*
639 	 * a raid1 doesn't have the notion of chunk size, so
640 	 * figure out the largest suitable size we can use.
641 	 */
642 	chunksect = 64 * 2; /* 64K by default */
643 
644 	/* The array must be an exact multiple of chunksize */
645 	while (chunksect && (mddev->array_sectors & (chunksect - 1)))
646 		chunksect >>= 1;
647 
648 	if ((chunksect << 9) < PAGE_SIZE)
649 		/* array size does not allow a suitable chunk size */
650 		return ERR_PTR(-EINVAL);
651 
652 	/* Set new parameters */
653 	mddev->new_level = 0;
654 	mddev->new_layout = 0;
655 	mddev->new_chunk_sectors = chunksect;
656 	mddev->chunk_sectors = chunksect;
657 	mddev->delta_disks = 1 - mddev->raid_disks;
658 	mddev->raid_disks = 1;
659 	/* make sure it will be not marked as dirty */
660 	mddev->recovery_cp = MaxSector;
661 
662 	create_strip_zones(mddev, &priv_conf);
663 	return priv_conf;
664 }
665 
666 static void *raid0_takeover(struct mddev *mddev)
667 {
668 	/* raid0 can take over:
669 	 *  raid4 - if all data disks are active.
670 	 *  raid5 - providing it is Raid4 layout and one disk is faulty
671 	 *  raid10 - assuming we have all necessary active disks
672 	 *  raid1 - with (N -1) mirror drives faulty
673 	 */
674 
675 	if (mddev->bitmap) {
676 		printk(KERN_ERR "md/raid0: %s: cannot takeover array with bitmap\n",
677 		       mdname(mddev));
678 		return ERR_PTR(-EBUSY);
679 	}
680 	if (mddev->level == 4)
681 		return raid0_takeover_raid45(mddev);
682 
683 	if (mddev->level == 5) {
684 		if (mddev->layout == ALGORITHM_PARITY_N)
685 			return raid0_takeover_raid45(mddev);
686 
687 		printk(KERN_ERR "md/raid0:%s: Raid can only takeover Raid5 with layout: %d\n",
688 		       mdname(mddev), ALGORITHM_PARITY_N);
689 	}
690 
691 	if (mddev->level == 10)
692 		return raid0_takeover_raid10(mddev);
693 
694 	if (mddev->level == 1)
695 		return raid0_takeover_raid1(mddev);
696 
697 	printk(KERN_ERR "Takeover from raid%i to raid0 not supported\n",
698 		mddev->level);
699 
700 	return ERR_PTR(-EINVAL);
701 }
702 
703 static void raid0_quiesce(struct mddev *mddev, int state)
704 {
705 }
706 
707 static struct md_personality raid0_personality=
708 {
709 	.name		= "raid0",
710 	.level		= 0,
711 	.owner		= THIS_MODULE,
712 	.make_request	= raid0_make_request,
713 	.run		= raid0_run,
714 	.free		= raid0_free,
715 	.status		= raid0_status,
716 	.size		= raid0_size,
717 	.takeover	= raid0_takeover,
718 	.quiesce	= raid0_quiesce,
719 	.congested	= raid0_congested,
720 	.mergeable_bvec	= raid0_mergeable_bvec,
721 };
722 
723 static int __init raid0_init (void)
724 {
725 	return register_md_personality (&raid0_personality);
726 }
727 
728 static void raid0_exit (void)
729 {
730 	unregister_md_personality (&raid0_personality);
731 }
732 
733 module_init(raid0_init);
734 module_exit(raid0_exit);
735 MODULE_LICENSE("GPL");
736 MODULE_DESCRIPTION("RAID0 (striping) personality for MD");
737 MODULE_ALIAS("md-personality-2"); /* RAID0 */
738 MODULE_ALIAS("md-raid0");
739 MODULE_ALIAS("md-level-0");
740