xref: /openbmc/linux/drivers/md/raid0.c (revision 6774def6)
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(void *data, int bits)
29 {
30 	struct mddev *mddev = data;
31 	struct r0conf *conf = mddev->private;
32 	struct md_rdev **devlist = conf->devlist;
33 	int raid_disks = conf->strip_zone[0].nb_dev;
34 	int i, ret = 0;
35 
36 	if (mddev_congested(mddev, bits))
37 		return 1;
38 
39 	for (i = 0; i < raid_disks && !ret ; i++) {
40 		struct request_queue *q = bdev_get_queue(devlist[i]->bdev);
41 
42 		ret |= bdi_congested(&q->backing_dev_info, bits);
43 	}
44 	return ret;
45 }
46 
47 /*
48  * inform the user of the raid configuration
49 */
50 static void dump_zones(struct mddev *mddev)
51 {
52 	int j, k;
53 	sector_t zone_size = 0;
54 	sector_t zone_start = 0;
55 	char b[BDEVNAME_SIZE];
56 	struct r0conf *conf = mddev->private;
57 	int raid_disks = conf->strip_zone[0].nb_dev;
58 	printk(KERN_INFO "md: RAID0 configuration for %s - %d zone%s\n",
59 	       mdname(mddev),
60 	       conf->nr_strip_zones, conf->nr_strip_zones==1?"":"s");
61 	for (j = 0; j < conf->nr_strip_zones; j++) {
62 		printk(KERN_INFO "md: zone%d=[", j);
63 		for (k = 0; k < conf->strip_zone[j].nb_dev; k++)
64 			printk(KERN_CONT "%s%s", k?"/":"",
65 			bdevname(conf->devlist[j*raid_disks
66 						+ k]->bdev, b));
67 		printk(KERN_CONT "]\n");
68 
69 		zone_size  = conf->strip_zone[j].zone_end - zone_start;
70 		printk(KERN_INFO "      zone-offset=%10lluKB, "
71 				"device-offset=%10lluKB, size=%10lluKB\n",
72 			(unsigned long long)zone_start>>1,
73 			(unsigned long long)conf->strip_zone[j].dev_start>>1,
74 			(unsigned long long)zone_size>>1);
75 		zone_start = conf->strip_zone[j].zone_end;
76 	}
77 	printk(KERN_INFO "\n");
78 }
79 
80 static int create_strip_zones(struct mddev *mddev, struct r0conf **private_conf)
81 {
82 	int i, c, err;
83 	sector_t curr_zone_end, sectors;
84 	struct md_rdev *smallest, *rdev1, *rdev2, *rdev, **dev;
85 	struct strip_zone *zone;
86 	int cnt;
87 	char b[BDEVNAME_SIZE];
88 	char b2[BDEVNAME_SIZE];
89 	struct r0conf *conf = kzalloc(sizeof(*conf), GFP_KERNEL);
90 	bool discard_supported = false;
91 
92 	if (!conf)
93 		return -ENOMEM;
94 	rdev_for_each(rdev1, mddev) {
95 		pr_debug("md/raid0:%s: looking at %s\n",
96 			 mdname(mddev),
97 			 bdevname(rdev1->bdev, b));
98 		c = 0;
99 
100 		/* round size to chunk_size */
101 		sectors = rdev1->sectors;
102 		sector_div(sectors, mddev->chunk_sectors);
103 		rdev1->sectors = sectors * mddev->chunk_sectors;
104 
105 		rdev_for_each(rdev2, mddev) {
106 			pr_debug("md/raid0:%s:   comparing %s(%llu)"
107 				 " with %s(%llu)\n",
108 				 mdname(mddev),
109 				 bdevname(rdev1->bdev,b),
110 				 (unsigned long long)rdev1->sectors,
111 				 bdevname(rdev2->bdev,b2),
112 				 (unsigned long long)rdev2->sectors);
113 			if (rdev2 == rdev1) {
114 				pr_debug("md/raid0:%s:   END\n",
115 					 mdname(mddev));
116 				break;
117 			}
118 			if (rdev2->sectors == rdev1->sectors) {
119 				/*
120 				 * Not unique, don't count it as a new
121 				 * group
122 				 */
123 				pr_debug("md/raid0:%s:   EQUAL\n",
124 					 mdname(mddev));
125 				c = 1;
126 				break;
127 			}
128 			pr_debug("md/raid0:%s:   NOT EQUAL\n",
129 				 mdname(mddev));
130 		}
131 		if (!c) {
132 			pr_debug("md/raid0:%s:   ==> UNIQUE\n",
133 				 mdname(mddev));
134 			conf->nr_strip_zones++;
135 			pr_debug("md/raid0:%s: %d zones\n",
136 				 mdname(mddev), conf->nr_strip_zones);
137 		}
138 	}
139 	pr_debug("md/raid0:%s: FINAL %d zones\n",
140 		 mdname(mddev), conf->nr_strip_zones);
141 	err = -ENOMEM;
142 	conf->strip_zone = kzalloc(sizeof(struct strip_zone)*
143 				conf->nr_strip_zones, GFP_KERNEL);
144 	if (!conf->strip_zone)
145 		goto abort;
146 	conf->devlist = kzalloc(sizeof(struct md_rdev*)*
147 				conf->nr_strip_zones*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 			printk(KERN_ERR
179 			       "md/raid0:%s: remove inactive devices before converting to RAID0\n",
180 			       mdname(mddev));
181 			goto abort;
182 		}
183 		if (j >= mddev->raid_disks) {
184 			printk(KERN_ERR "md/raid0:%s: bad disk number %d - "
185 			       "aborting!\n", mdname(mddev), j);
186 			goto abort;
187 		}
188 		if (dev[j]) {
189 			printk(KERN_ERR "md/raid0:%s: multiple devices for %d - "
190 			       "aborting!\n", mdname(mddev), j);
191 			goto abort;
192 		}
193 		dev[j] = rdev1;
194 
195 		disk_stack_limits(mddev->gendisk, rdev1->bdev,
196 				  rdev1->data_offset << 9);
197 
198 		if (rdev1->bdev->bd_disk->queue->merge_bvec_fn)
199 			conf->has_merge_bvec = 1;
200 
201 		if (!smallest || (rdev1->sectors < smallest->sectors))
202 			smallest = rdev1;
203 		cnt++;
204 
205 		if (blk_queue_discard(bdev_get_queue(rdev1->bdev)))
206 			discard_supported = true;
207 	}
208 	if (cnt != mddev->raid_disks) {
209 		printk(KERN_ERR "md/raid0:%s: too few disks (%d of %d) - "
210 		       "aborting!\n", mdname(mddev), cnt, mddev->raid_disks);
211 		goto abort;
212 	}
213 	zone->nb_dev = cnt;
214 	zone->zone_end = smallest->sectors * cnt;
215 
216 	curr_zone_end = zone->zone_end;
217 
218 	/* now do the other zones */
219 	for (i = 1; i < conf->nr_strip_zones; i++)
220 	{
221 		int j;
222 
223 		zone = conf->strip_zone + i;
224 		dev = conf->devlist + i * mddev->raid_disks;
225 
226 		pr_debug("md/raid0:%s: zone %d\n", mdname(mddev), i);
227 		zone->dev_start = smallest->sectors;
228 		smallest = NULL;
229 		c = 0;
230 
231 		for (j=0; j<cnt; j++) {
232 			rdev = conf->devlist[j];
233 			if (rdev->sectors <= zone->dev_start) {
234 				pr_debug("md/raid0:%s: checking %s ... nope\n",
235 					 mdname(mddev),
236 					 bdevname(rdev->bdev, b));
237 				continue;
238 			}
239 			pr_debug("md/raid0:%s: checking %s ..."
240 				 " contained as device %d\n",
241 				 mdname(mddev),
242 				 bdevname(rdev->bdev, b), c);
243 			dev[c] = rdev;
244 			c++;
245 			if (!smallest || rdev->sectors < smallest->sectors) {
246 				smallest = rdev;
247 				pr_debug("md/raid0:%s:  (%llu) is smallest!.\n",
248 					 mdname(mddev),
249 					 (unsigned long long)rdev->sectors);
250 			}
251 		}
252 
253 		zone->nb_dev = c;
254 		sectors = (smallest->sectors - zone->dev_start) * c;
255 		pr_debug("md/raid0:%s: zone->nb_dev: %d, sectors: %llu\n",
256 			 mdname(mddev),
257 			 zone->nb_dev, (unsigned long long)sectors);
258 
259 		curr_zone_end += sectors;
260 		zone->zone_end = curr_zone_end;
261 
262 		pr_debug("md/raid0:%s: current zone start: %llu\n",
263 			 mdname(mddev),
264 			 (unsigned long long)smallest->sectors);
265 	}
266 	mddev->queue->backing_dev_info.congested_fn = raid0_congested;
267 	mddev->queue->backing_dev_info.congested_data = mddev;
268 
269 	/*
270 	 * now since we have the hard sector sizes, we can make sure
271 	 * chunk size is a multiple of that sector size
272 	 */
273 	if ((mddev->chunk_sectors << 9) % queue_logical_block_size(mddev->queue)) {
274 		printk(KERN_ERR "md/raid0:%s: chunk_size of %d not valid\n",
275 		       mdname(mddev),
276 		       mddev->chunk_sectors << 9);
277 		goto abort;
278 	}
279 
280 	blk_queue_io_min(mddev->queue, mddev->chunk_sectors << 9);
281 	blk_queue_io_opt(mddev->queue,
282 			 (mddev->chunk_sectors << 9) * mddev->raid_disks);
283 
284 	if (!discard_supported)
285 		queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, mddev->queue);
286 	else
287 		queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, mddev->queue);
288 
289 	pr_debug("md/raid0:%s: done.\n", mdname(mddev));
290 	*private_conf = conf;
291 
292 	return 0;
293 abort:
294 	kfree(conf->strip_zone);
295 	kfree(conf->devlist);
296 	kfree(conf);
297 	*private_conf = ERR_PTR(err);
298 	return err;
299 }
300 
301 /* Find the zone which holds a particular offset
302  * Update *sectorp to be an offset in that zone
303  */
304 static struct strip_zone *find_zone(struct r0conf *conf,
305 				    sector_t *sectorp)
306 {
307 	int i;
308 	struct strip_zone *z = conf->strip_zone;
309 	sector_t sector = *sectorp;
310 
311 	for (i = 0; i < conf->nr_strip_zones; i++)
312 		if (sector < z[i].zone_end) {
313 			if (i)
314 				*sectorp = sector - z[i-1].zone_end;
315 			return z + i;
316 		}
317 	BUG();
318 }
319 
320 /*
321  * remaps the bio to the target device. we separate two flows.
322  * power 2 flow and a general flow for the sake of perfromance
323 */
324 static struct md_rdev *map_sector(struct mddev *mddev, struct strip_zone *zone,
325 				sector_t sector, sector_t *sector_offset)
326 {
327 	unsigned int sect_in_chunk;
328 	sector_t chunk;
329 	struct r0conf *conf = mddev->private;
330 	int raid_disks = conf->strip_zone[0].nb_dev;
331 	unsigned int chunk_sects = mddev->chunk_sectors;
332 
333 	if (is_power_of_2(chunk_sects)) {
334 		int chunksect_bits = ffz(~chunk_sects);
335 		/* find the sector offset inside the chunk */
336 		sect_in_chunk  = sector & (chunk_sects - 1);
337 		sector >>= chunksect_bits;
338 		/* chunk in zone */
339 		chunk = *sector_offset;
340 		/* quotient is the chunk in real device*/
341 		sector_div(chunk, zone->nb_dev << chunksect_bits);
342 	} else{
343 		sect_in_chunk = sector_div(sector, chunk_sects);
344 		chunk = *sector_offset;
345 		sector_div(chunk, chunk_sects * zone->nb_dev);
346 	}
347 	/*
348 	*  position the bio over the real device
349 	*  real sector = chunk in device + starting of zone
350 	*	+ the position in the chunk
351 	*/
352 	*sector_offset = (chunk * chunk_sects) + sect_in_chunk;
353 	return conf->devlist[(zone - conf->strip_zone)*raid_disks
354 			     + sector_div(sector, zone->nb_dev)];
355 }
356 
357 /**
358  *	raid0_mergeable_bvec -- tell bio layer if two requests can be merged
359  *	@q: request queue
360  *	@bvm: properties of new bio
361  *	@biovec: the request that could be merged to it.
362  *
363  *	Return amount of bytes we can accept at this offset
364  */
365 static int raid0_mergeable_bvec(struct request_queue *q,
366 				struct bvec_merge_data *bvm,
367 				struct bio_vec *biovec)
368 {
369 	struct mddev *mddev = q->queuedata;
370 	struct r0conf *conf = mddev->private;
371 	sector_t sector = bvm->bi_sector + get_start_sect(bvm->bi_bdev);
372 	sector_t sector_offset = sector;
373 	int max;
374 	unsigned int chunk_sectors = mddev->chunk_sectors;
375 	unsigned int bio_sectors = bvm->bi_size >> 9;
376 	struct strip_zone *zone;
377 	struct md_rdev *rdev;
378 	struct request_queue *subq;
379 
380 	if (is_power_of_2(chunk_sectors))
381 		max =  (chunk_sectors - ((sector & (chunk_sectors-1))
382 						+ bio_sectors)) << 9;
383 	else
384 		max =  (chunk_sectors - (sector_div(sector, chunk_sectors)
385 						+ bio_sectors)) << 9;
386 	if (max < 0)
387 		max = 0; /* bio_add cannot handle a negative return */
388 	if (max <= biovec->bv_len && bio_sectors == 0)
389 		return biovec->bv_len;
390 	if (max < biovec->bv_len)
391 		/* too small already, no need to check further */
392 		return max;
393 	if (!conf->has_merge_bvec)
394 		return max;
395 
396 	/* May need to check subordinate device */
397 	sector = sector_offset;
398 	zone = find_zone(mddev->private, &sector_offset);
399 	rdev = map_sector(mddev, zone, sector, &sector_offset);
400 	subq = bdev_get_queue(rdev->bdev);
401 	if (subq->merge_bvec_fn) {
402 		bvm->bi_bdev = rdev->bdev;
403 		bvm->bi_sector = sector_offset + zone->dev_start +
404 			rdev->data_offset;
405 		return min(max, subq->merge_bvec_fn(subq, bvm, biovec));
406 	} else
407 		return max;
408 }
409 
410 static sector_t raid0_size(struct mddev *mddev, sector_t sectors, int raid_disks)
411 {
412 	sector_t array_sectors = 0;
413 	struct md_rdev *rdev;
414 
415 	WARN_ONCE(sectors || raid_disks,
416 		  "%s does not support generic reshape\n", __func__);
417 
418 	rdev_for_each(rdev, mddev)
419 		array_sectors += (rdev->sectors &
420 				  ~(sector_t)(mddev->chunk_sectors-1));
421 
422 	return array_sectors;
423 }
424 
425 static int raid0_stop(struct mddev *mddev);
426 
427 static int raid0_run(struct mddev *mddev)
428 {
429 	struct r0conf *conf;
430 	int ret;
431 
432 	if (mddev->chunk_sectors == 0) {
433 		printk(KERN_ERR "md/raid0:%s: chunk size must be set.\n",
434 		       mdname(mddev));
435 		return -EINVAL;
436 	}
437 	if (md_check_no_bitmap(mddev))
438 		return -EINVAL;
439 	blk_queue_max_hw_sectors(mddev->queue, mddev->chunk_sectors);
440 	blk_queue_max_write_same_sectors(mddev->queue, mddev->chunk_sectors);
441 	blk_queue_max_discard_sectors(mddev->queue, mddev->chunk_sectors);
442 
443 	/* if private is not null, we are here after takeover */
444 	if (mddev->private == NULL) {
445 		ret = create_strip_zones(mddev, &conf);
446 		if (ret < 0)
447 			return ret;
448 		mddev->private = conf;
449 	}
450 	conf = mddev->private;
451 
452 	/* calculate array device size */
453 	md_set_array_sectors(mddev, raid0_size(mddev, 0, 0));
454 
455 	printk(KERN_INFO "md/raid0:%s: md_size is %llu sectors.\n",
456 	       mdname(mddev),
457 	       (unsigned long long)mddev->array_sectors);
458 	/* calculate the max read-ahead size.
459 	 * For read-ahead of large files to be effective, we need to
460 	 * readahead at least twice a whole stripe. i.e. number of devices
461 	 * multiplied by chunk size times 2.
462 	 * If an individual device has an ra_pages greater than the
463 	 * chunk size, then we will not drive that device as hard as it
464 	 * wants.  We consider this a configuration error: a larger
465 	 * chunksize should be used in that case.
466 	 */
467 	{
468 		int stripe = mddev->raid_disks *
469 			(mddev->chunk_sectors << 9) / PAGE_SIZE;
470 		if (mddev->queue->backing_dev_info.ra_pages < 2* stripe)
471 			mddev->queue->backing_dev_info.ra_pages = 2* stripe;
472 	}
473 
474 	blk_queue_merge_bvec(mddev->queue, raid0_mergeable_bvec);
475 	dump_zones(mddev);
476 
477 	ret = md_integrity_register(mddev);
478 	if (ret)
479 		raid0_stop(mddev);
480 
481 	return ret;
482 }
483 
484 static int raid0_stop(struct mddev *mddev)
485 {
486 	struct r0conf *conf = mddev->private;
487 
488 	blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/
489 	kfree(conf->strip_zone);
490 	kfree(conf->devlist);
491 	kfree(conf);
492 	mddev->private = NULL;
493 	return 0;
494 }
495 
496 /*
497  * Is io distribute over 1 or more chunks ?
498 */
499 static inline int is_io_in_chunk_boundary(struct mddev *mddev,
500 			unsigned int chunk_sects, struct bio *bio)
501 {
502 	if (likely(is_power_of_2(chunk_sects))) {
503 		return chunk_sects >=
504 			((bio->bi_iter.bi_sector & (chunk_sects-1))
505 					+ bio_sectors(bio));
506 	} else{
507 		sector_t sector = bio->bi_iter.bi_sector;
508 		return chunk_sects >= (sector_div(sector, chunk_sects)
509 						+ bio_sectors(bio));
510 	}
511 }
512 
513 static void raid0_make_request(struct mddev *mddev, struct bio *bio)
514 {
515 	struct strip_zone *zone;
516 	struct md_rdev *tmp_dev;
517 	struct bio *split;
518 
519 	if (unlikely(bio->bi_rw & REQ_FLUSH)) {
520 		md_flush_request(mddev, bio);
521 		return;
522 	}
523 
524 	do {
525 		sector_t sector = bio->bi_iter.bi_sector;
526 		unsigned chunk_sects = mddev->chunk_sectors;
527 
528 		unsigned sectors = chunk_sects -
529 			(likely(is_power_of_2(chunk_sects))
530 			 ? (sector & (chunk_sects-1))
531 			 : sector_div(sector, chunk_sects));
532 
533 		if (sectors < bio_sectors(bio)) {
534 			split = bio_split(bio, sectors, GFP_NOIO, fs_bio_set);
535 			bio_chain(split, bio);
536 		} else {
537 			split = bio;
538 		}
539 
540 		zone = find_zone(mddev->private, &sector);
541 		tmp_dev = map_sector(mddev, zone, sector, &sector);
542 		split->bi_bdev = tmp_dev->bdev;
543 		split->bi_iter.bi_sector = sector + zone->dev_start +
544 			tmp_dev->data_offset;
545 
546 		if (unlikely((split->bi_rw & REQ_DISCARD) &&
547 			 !blk_queue_discard(bdev_get_queue(split->bi_bdev)))) {
548 			/* Just ignore it */
549 			bio_endio(split, 0);
550 		} else
551 			generic_make_request(split);
552 	} while (split != bio);
553 }
554 
555 static void raid0_status(struct seq_file *seq, struct mddev *mddev)
556 {
557 	seq_printf(seq, " %dk chunks", mddev->chunk_sectors / 2);
558 	return;
559 }
560 
561 static void *raid0_takeover_raid45(struct mddev *mddev)
562 {
563 	struct md_rdev *rdev;
564 	struct r0conf *priv_conf;
565 
566 	if (mddev->degraded != 1) {
567 		printk(KERN_ERR "md/raid0:%s: raid5 must be degraded! Degraded disks: %d\n",
568 		       mdname(mddev),
569 		       mddev->degraded);
570 		return ERR_PTR(-EINVAL);
571 	}
572 
573 	rdev_for_each(rdev, mddev) {
574 		/* check slot number for a disk */
575 		if (rdev->raid_disk == mddev->raid_disks-1) {
576 			printk(KERN_ERR "md/raid0:%s: raid5 must have missing parity disk!\n",
577 			       mdname(mddev));
578 			return ERR_PTR(-EINVAL);
579 		}
580 		rdev->sectors = mddev->dev_sectors;
581 	}
582 
583 	/* Set new parameters */
584 	mddev->new_level = 0;
585 	mddev->new_layout = 0;
586 	mddev->new_chunk_sectors = mddev->chunk_sectors;
587 	mddev->raid_disks--;
588 	mddev->delta_disks = -1;
589 	/* make sure it will be not marked as dirty */
590 	mddev->recovery_cp = MaxSector;
591 
592 	create_strip_zones(mddev, &priv_conf);
593 	return priv_conf;
594 }
595 
596 static void *raid0_takeover_raid10(struct mddev *mddev)
597 {
598 	struct r0conf *priv_conf;
599 
600 	/* Check layout:
601 	 *  - far_copies must be 1
602 	 *  - near_copies must be 2
603 	 *  - disks number must be even
604 	 *  - all mirrors must be already degraded
605 	 */
606 	if (mddev->layout != ((1 << 8) + 2)) {
607 		printk(KERN_ERR "md/raid0:%s:: Raid0 cannot takover layout: 0x%x\n",
608 		       mdname(mddev),
609 		       mddev->layout);
610 		return ERR_PTR(-EINVAL);
611 	}
612 	if (mddev->raid_disks & 1) {
613 		printk(KERN_ERR "md/raid0:%s: Raid0 cannot takover Raid10 with odd disk number.\n",
614 		       mdname(mddev));
615 		return ERR_PTR(-EINVAL);
616 	}
617 	if (mddev->degraded != (mddev->raid_disks>>1)) {
618 		printk(KERN_ERR "md/raid0:%s: All mirrors must be already degraded!\n",
619 		       mdname(mddev));
620 		return ERR_PTR(-EINVAL);
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->delta_disks = - mddev->raid_disks / 2;
628 	mddev->raid_disks += mddev->delta_disks;
629 	mddev->degraded = 0;
630 	/* make sure it will be not marked as dirty */
631 	mddev->recovery_cp = MaxSector;
632 
633 	create_strip_zones(mddev, &priv_conf);
634 	return priv_conf;
635 }
636 
637 static void *raid0_takeover_raid1(struct mddev *mddev)
638 {
639 	struct r0conf *priv_conf;
640 	int chunksect;
641 
642 	/* Check layout:
643 	 *  - (N - 1) mirror drives must be already faulty
644 	 */
645 	if ((mddev->raid_disks - 1) != mddev->degraded) {
646 		printk(KERN_ERR "md/raid0:%s: (N - 1) mirrors drives must be already faulty!\n",
647 		       mdname(mddev));
648 		return ERR_PTR(-EINVAL);
649 	}
650 
651 	/*
652 	 * a raid1 doesn't have the notion of chunk size, so
653 	 * figure out the largest suitable size we can use.
654 	 */
655 	chunksect = 64 * 2; /* 64K by default */
656 
657 	/* The array must be an exact multiple of chunksize */
658 	while (chunksect && (mddev->array_sectors & (chunksect - 1)))
659 		chunksect >>= 1;
660 
661 	if ((chunksect << 9) < PAGE_SIZE)
662 		/* array size does not allow a suitable chunk size */
663 		return ERR_PTR(-EINVAL);
664 
665 	/* Set new parameters */
666 	mddev->new_level = 0;
667 	mddev->new_layout = 0;
668 	mddev->new_chunk_sectors = chunksect;
669 	mddev->chunk_sectors = chunksect;
670 	mddev->delta_disks = 1 - mddev->raid_disks;
671 	mddev->raid_disks = 1;
672 	/* make sure it will be not marked as dirty */
673 	mddev->recovery_cp = MaxSector;
674 
675 	create_strip_zones(mddev, &priv_conf);
676 	return priv_conf;
677 }
678 
679 static void *raid0_takeover(struct mddev *mddev)
680 {
681 	/* raid0 can take over:
682 	 *  raid4 - if all data disks are active.
683 	 *  raid5 - providing it is Raid4 layout and one disk is faulty
684 	 *  raid10 - assuming we have all necessary active disks
685 	 *  raid1 - with (N -1) mirror drives faulty
686 	 */
687 
688 	if (mddev->bitmap) {
689 		printk(KERN_ERR "md/raid0: %s: cannot takeover array with bitmap\n",
690 		       mdname(mddev));
691 		return ERR_PTR(-EBUSY);
692 	}
693 	if (mddev->level == 4)
694 		return raid0_takeover_raid45(mddev);
695 
696 	if (mddev->level == 5) {
697 		if (mddev->layout == ALGORITHM_PARITY_N)
698 			return raid0_takeover_raid45(mddev);
699 
700 		printk(KERN_ERR "md/raid0:%s: Raid can only takeover Raid5 with layout: %d\n",
701 		       mdname(mddev), ALGORITHM_PARITY_N);
702 	}
703 
704 	if (mddev->level == 10)
705 		return raid0_takeover_raid10(mddev);
706 
707 	if (mddev->level == 1)
708 		return raid0_takeover_raid1(mddev);
709 
710 	printk(KERN_ERR "Takeover from raid%i to raid0 not supported\n",
711 		mddev->level);
712 
713 	return ERR_PTR(-EINVAL);
714 }
715 
716 static void raid0_quiesce(struct mddev *mddev, int state)
717 {
718 }
719 
720 static struct md_personality raid0_personality=
721 {
722 	.name		= "raid0",
723 	.level		= 0,
724 	.owner		= THIS_MODULE,
725 	.make_request	= raid0_make_request,
726 	.run		= raid0_run,
727 	.stop		= raid0_stop,
728 	.status		= raid0_status,
729 	.size		= raid0_size,
730 	.takeover	= raid0_takeover,
731 	.quiesce	= raid0_quiesce,
732 };
733 
734 static int __init raid0_init (void)
735 {
736 	return register_md_personality (&raid0_personality);
737 }
738 
739 static void raid0_exit (void)
740 {
741 	unregister_md_personality (&raid0_personality);
742 }
743 
744 module_init(raid0_init);
745 module_exit(raid0_exit);
746 MODULE_LICENSE("GPL");
747 MODULE_DESCRIPTION("RAID0 (striping) personality for MD");
748 MODULE_ALIAS("md-personality-2"); /* RAID0 */
749 MODULE_ALIAS("md-raid0");
750 MODULE_ALIAS("md-level-0");
751