Linux Audio

Check our new training course

Loading...
v3.15
  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
  9   RAID-0 management functions.
 10
 11   This program is free software; you can redistribute it and/or modify
 12   it under the terms of the GNU General Public License as published by
 13   the Free Software Foundation; either version 2, or (at your option)
 14   any later version.
 15   
 16   You should have received a copy of the GNU General Public License
 17   (for example /usr/src/linux/COPYING); if not, write to the Free
 18   Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.  
 19*/
 20
 21#include <linux/blkdev.h>
 22#include <linux/seq_file.h>
 23#include <linux/module.h>
 24#include <linux/slab.h>
 25#include "md.h"
 26#include "raid0.h"
 27#include "raid5.h"
 28
 29static int raid0_congested(void *data, int bits)
 30{
 31	struct mddev *mddev = data;
 32	struct r0conf *conf = mddev->private;
 33	struct md_rdev **devlist = conf->devlist;
 34	int raid_disks = conf->strip_zone[0].nb_dev;
 35	int i, ret = 0;
 36
 37	if (mddev_congested(mddev, bits))
 38		return 1;
 39
 40	for (i = 0; i < raid_disks && !ret ; i++) {
 41		struct request_queue *q = bdev_get_queue(devlist[i]->bdev);
 42
 43		ret |= bdi_congested(&q->backing_dev_info, bits);
 44	}
 45	return ret;
 46}
 47
 48/*
 49 * inform the user of the raid configuration
 50*/
 51static void dump_zones(struct mddev *mddev)
 52{
 53	int j, k;
 54	sector_t zone_size = 0;
 55	sector_t zone_start = 0;
 56	char b[BDEVNAME_SIZE];
 57	struct r0conf *conf = mddev->private;
 58	int raid_disks = conf->strip_zone[0].nb_dev;
 59	printk(KERN_INFO "md: RAID0 configuration for %s - %d zone%s\n",
 60	       mdname(mddev),
 61	       conf->nr_strip_zones, conf->nr_strip_zones==1?"":"s");
 62	for (j = 0; j < conf->nr_strip_zones; j++) {
 63		printk(KERN_INFO "md: zone%d=[", j);
 64		for (k = 0; k < conf->strip_zone[j].nb_dev; k++)
 65			printk(KERN_CONT "%s%s", k?"/":"",
 66			bdevname(conf->devlist[j*raid_disks
 67						+ k]->bdev, b));
 68		printk(KERN_CONT "]\n");
 69
 70		zone_size  = conf->strip_zone[j].zone_end - zone_start;
 71		printk(KERN_INFO "      zone-offset=%10lluKB, "
 72				"device-offset=%10lluKB, size=%10lluKB\n",
 73			(unsigned long long)zone_start>>1,
 74			(unsigned long long)conf->strip_zone[j].dev_start>>1,
 75			(unsigned long long)zone_size>>1);
 76		zone_start = conf->strip_zone[j].zone_end;
 77	}
 78	printk(KERN_INFO "\n");
 79}
 80
 81static int create_strip_zones(struct mddev *mddev, struct r0conf **private_conf)
 82{
 83	int i, c, err;
 84	sector_t curr_zone_end, sectors;
 85	struct md_rdev *smallest, *rdev1, *rdev2, *rdev, **dev;
 86	struct strip_zone *zone;
 87	int cnt;
 88	char b[BDEVNAME_SIZE];
 89	char b2[BDEVNAME_SIZE];
 90	struct r0conf *conf = kzalloc(sizeof(*conf), GFP_KERNEL);
 91	bool discard_supported = false;
 92
 
 93	if (!conf)
 94		return -ENOMEM;
 95	rdev_for_each(rdev1, mddev) {
 96		pr_debug("md/raid0:%s: looking at %s\n",
 97			 mdname(mddev),
 98			 bdevname(rdev1->bdev, b));
 99		c = 0;
100
101		/* round size to chunk_size */
102		sectors = rdev1->sectors;
103		sector_div(sectors, mddev->chunk_sectors);
104		rdev1->sectors = sectors * mddev->chunk_sectors;
105
 
 
 
106		rdev_for_each(rdev2, mddev) {
107			pr_debug("md/raid0:%s:   comparing %s(%llu)"
108				 " with %s(%llu)\n",
109				 mdname(mddev),
110				 bdevname(rdev1->bdev,b),
111				 (unsigned long long)rdev1->sectors,
112				 bdevname(rdev2->bdev,b2),
113				 (unsigned long long)rdev2->sectors);
114			if (rdev2 == rdev1) {
115				pr_debug("md/raid0:%s:   END\n",
116					 mdname(mddev));
117				break;
118			}
119			if (rdev2->sectors == rdev1->sectors) {
120				/*
121				 * Not unique, don't count it as a new
122				 * group
123				 */
124				pr_debug("md/raid0:%s:   EQUAL\n",
125					 mdname(mddev));
126				c = 1;
127				break;
128			}
129			pr_debug("md/raid0:%s:   NOT EQUAL\n",
130				 mdname(mddev));
131		}
132		if (!c) {
133			pr_debug("md/raid0:%s:   ==> UNIQUE\n",
134				 mdname(mddev));
135			conf->nr_strip_zones++;
136			pr_debug("md/raid0:%s: %d zones\n",
137				 mdname(mddev), conf->nr_strip_zones);
138		}
139	}
140	pr_debug("md/raid0:%s: FINAL %d zones\n",
141		 mdname(mddev), conf->nr_strip_zones);
 
 
 
 
 
 
 
 
 
 
 
 
142	err = -ENOMEM;
143	conf->strip_zone = kzalloc(sizeof(struct strip_zone)*
144				conf->nr_strip_zones, GFP_KERNEL);
145	if (!conf->strip_zone)
146		goto abort;
147	conf->devlist = kzalloc(sizeof(struct md_rdev*)*
148				conf->nr_strip_zones*mddev->raid_disks,
149				GFP_KERNEL);
150	if (!conf->devlist)
151		goto abort;
152
153	/* The first zone must contain all devices, so here we check that
154	 * there is a proper alignment of slots to devices and find them all
155	 */
156	zone = &conf->strip_zone[0];
157	cnt = 0;
158	smallest = NULL;
159	dev = conf->devlist;
160	err = -EINVAL;
161	rdev_for_each(rdev1, mddev) {
162		int j = rdev1->raid_disk;
163
164		if (mddev->level == 10) {
165			/* taking over a raid10-n2 array */
166			j /= 2;
167			rdev1->new_raid_disk = j;
168		}
169
170		if (mddev->level == 1) {
171			/* taiking over a raid1 array-
172			 * we have only one active disk
173			 */
174			j = 0;
175			rdev1->new_raid_disk = j;
176		}
177
178		if (j < 0) {
179			printk(KERN_ERR
180			       "md/raid0:%s: remove inactive devices before converting to RAID0\n",
181			       mdname(mddev));
182			goto abort;
183		}
184		if (j >= mddev->raid_disks) {
185			printk(KERN_ERR "md/raid0:%s: bad disk number %d - "
186			       "aborting!\n", mdname(mddev), j);
187			goto abort;
188		}
189		if (dev[j]) {
190			printk(KERN_ERR "md/raid0:%s: multiple devices for %d - "
191			       "aborting!\n", mdname(mddev), j);
192			goto abort;
193		}
194		dev[j] = rdev1;
195
196		disk_stack_limits(mddev->gendisk, rdev1->bdev,
197				  rdev1->data_offset << 9);
198
199		if (rdev1->bdev->bd_disk->queue->merge_bvec_fn)
200			conf->has_merge_bvec = 1;
201
202		if (!smallest || (rdev1->sectors < smallest->sectors))
203			smallest = rdev1;
204		cnt++;
205
206		if (blk_queue_discard(bdev_get_queue(rdev1->bdev)))
207			discard_supported = true;
208	}
209	if (cnt != mddev->raid_disks) {
210		printk(KERN_ERR "md/raid0:%s: too few disks (%d of %d) - "
211		       "aborting!\n", mdname(mddev), cnt, mddev->raid_disks);
212		goto abort;
213	}
214	zone->nb_dev = cnt;
215	zone->zone_end = smallest->sectors * cnt;
216
217	curr_zone_end = zone->zone_end;
218
219	/* now do the other zones */
220	for (i = 1; i < conf->nr_strip_zones; i++)
221	{
222		int j;
223
224		zone = conf->strip_zone + i;
225		dev = conf->devlist + i * mddev->raid_disks;
226
227		pr_debug("md/raid0:%s: zone %d\n", mdname(mddev), i);
228		zone->dev_start = smallest->sectors;
229		smallest = NULL;
230		c = 0;
231
232		for (j=0; j<cnt; j++) {
233			rdev = conf->devlist[j];
234			if (rdev->sectors <= zone->dev_start) {
235				pr_debug("md/raid0:%s: checking %s ... nope\n",
236					 mdname(mddev),
237					 bdevname(rdev->bdev, b));
238				continue;
239			}
240			pr_debug("md/raid0:%s: checking %s ..."
241				 " contained as device %d\n",
242				 mdname(mddev),
243				 bdevname(rdev->bdev, b), c);
244			dev[c] = rdev;
245			c++;
246			if (!smallest || rdev->sectors < smallest->sectors) {
247				smallest = rdev;
248				pr_debug("md/raid0:%s:  (%llu) is smallest!.\n",
249					 mdname(mddev),
250					 (unsigned long long)rdev->sectors);
251			}
252		}
253
254		zone->nb_dev = c;
255		sectors = (smallest->sectors - zone->dev_start) * c;
256		pr_debug("md/raid0:%s: zone->nb_dev: %d, sectors: %llu\n",
257			 mdname(mddev),
258			 zone->nb_dev, (unsigned long long)sectors);
259
260		curr_zone_end += sectors;
261		zone->zone_end = curr_zone_end;
262
263		pr_debug("md/raid0:%s: current zone start: %llu\n",
264			 mdname(mddev),
265			 (unsigned long long)smallest->sectors);
266	}
267	mddev->queue->backing_dev_info.congested_fn = raid0_congested;
268	mddev->queue->backing_dev_info.congested_data = mddev;
269
270	/*
271	 * now since we have the hard sector sizes, we can make sure
272	 * chunk size is a multiple of that sector size
273	 */
274	if ((mddev->chunk_sectors << 9) % queue_logical_block_size(mddev->queue)) {
275		printk(KERN_ERR "md/raid0:%s: chunk_size of %d not valid\n",
276		       mdname(mddev),
277		       mddev->chunk_sectors << 9);
278		goto abort;
279	}
280
281	blk_queue_io_min(mddev->queue, mddev->chunk_sectors << 9);
282	blk_queue_io_opt(mddev->queue,
283			 (mddev->chunk_sectors << 9) * mddev->raid_disks);
284
285	if (!discard_supported)
286		queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, mddev->queue);
287	else
288		queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, mddev->queue);
289
290	pr_debug("md/raid0:%s: done.\n", mdname(mddev));
291	*private_conf = conf;
292
293	return 0;
294abort:
295	kfree(conf->strip_zone);
296	kfree(conf->devlist);
297	kfree(conf);
298	*private_conf = ERR_PTR(err);
299	return err;
300}
301
302/* Find the zone which holds a particular offset
303 * Update *sectorp to be an offset in that zone
304 */
305static struct strip_zone *find_zone(struct r0conf *conf,
306				    sector_t *sectorp)
307{
308	int i;
309	struct strip_zone *z = conf->strip_zone;
310	sector_t sector = *sectorp;
311
312	for (i = 0; i < conf->nr_strip_zones; i++)
313		if (sector < z[i].zone_end) {
314			if (i)
315				*sectorp = sector - z[i-1].zone_end;
316			return z + i;
317		}
318	BUG();
319}
320
321/*
322 * remaps the bio to the target device. we separate two flows.
323 * power 2 flow and a general flow for the sake of perfromance
324*/
325static struct md_rdev *map_sector(struct mddev *mddev, struct strip_zone *zone,
326				sector_t sector, sector_t *sector_offset)
327{
328	unsigned int sect_in_chunk;
329	sector_t chunk;
330	struct r0conf *conf = mddev->private;
331	int raid_disks = conf->strip_zone[0].nb_dev;
332	unsigned int chunk_sects = mddev->chunk_sectors;
333
334	if (is_power_of_2(chunk_sects)) {
335		int chunksect_bits = ffz(~chunk_sects);
336		/* find the sector offset inside the chunk */
337		sect_in_chunk  = sector & (chunk_sects - 1);
338		sector >>= chunksect_bits;
339		/* chunk in zone */
340		chunk = *sector_offset;
341		/* quotient is the chunk in real device*/
342		sector_div(chunk, zone->nb_dev << chunksect_bits);
343	} else{
344		sect_in_chunk = sector_div(sector, chunk_sects);
345		chunk = *sector_offset;
346		sector_div(chunk, chunk_sects * zone->nb_dev);
347	}
348	/*
349	*  position the bio over the real device
350	*  real sector = chunk in device + starting of zone
351	*	+ the position in the chunk
352	*/
353	*sector_offset = (chunk * chunk_sects) + sect_in_chunk;
354	return conf->devlist[(zone - conf->strip_zone)*raid_disks
355			     + sector_div(sector, zone->nb_dev)];
356}
357
358/**
359 *	raid0_mergeable_bvec -- tell bio layer if two requests can be merged
360 *	@q: request queue
361 *	@bvm: properties of new bio
362 *	@biovec: the request that could be merged to it.
363 *
364 *	Return amount of bytes we can accept at this offset
365 */
366static int raid0_mergeable_bvec(struct request_queue *q,
367				struct bvec_merge_data *bvm,
368				struct bio_vec *biovec)
369{
370	struct mddev *mddev = q->queuedata;
371	struct r0conf *conf = mddev->private;
372	sector_t sector = bvm->bi_sector + get_start_sect(bvm->bi_bdev);
373	sector_t sector_offset = sector;
374	int max;
375	unsigned int chunk_sectors = mddev->chunk_sectors;
376	unsigned int bio_sectors = bvm->bi_size >> 9;
377	struct strip_zone *zone;
378	struct md_rdev *rdev;
379	struct request_queue *subq;
380
381	if (is_power_of_2(chunk_sectors))
382		max =  (chunk_sectors - ((sector & (chunk_sectors-1))
383						+ bio_sectors)) << 9;
384	else
385		max =  (chunk_sectors - (sector_div(sector, chunk_sectors)
386						+ bio_sectors)) << 9;
387	if (max < 0)
388		max = 0; /* bio_add cannot handle a negative return */
389	if (max <= biovec->bv_len && bio_sectors == 0)
390		return biovec->bv_len;
391	if (max < biovec->bv_len)
392		/* too small already, no need to check further */
393		return max;
394	if (!conf->has_merge_bvec)
395		return max;
396
397	/* May need to check subordinate device */
398	sector = sector_offset;
399	zone = find_zone(mddev->private, &sector_offset);
400	rdev = map_sector(mddev, zone, sector, &sector_offset);
401	subq = bdev_get_queue(rdev->bdev);
402	if (subq->merge_bvec_fn) {
403		bvm->bi_bdev = rdev->bdev;
404		bvm->bi_sector = sector_offset + zone->dev_start +
405			rdev->data_offset;
406		return min(max, subq->merge_bvec_fn(subq, bvm, biovec));
407	} else
408		return max;
409}
410
411static sector_t raid0_size(struct mddev *mddev, sector_t sectors, int raid_disks)
412{
413	sector_t array_sectors = 0;
414	struct md_rdev *rdev;
415
416	WARN_ONCE(sectors || raid_disks,
417		  "%s does not support generic reshape\n", __func__);
418
419	rdev_for_each(rdev, mddev)
420		array_sectors += (rdev->sectors &
421				  ~(sector_t)(mddev->chunk_sectors-1));
422
423	return array_sectors;
424}
425
426static int raid0_stop(struct mddev *mddev);
427
428static int raid0_run(struct mddev *mddev)
429{
430	struct r0conf *conf;
431	int ret;
432
433	if (mddev->chunk_sectors == 0) {
434		printk(KERN_ERR "md/raid0:%s: chunk size must be set.\n",
435		       mdname(mddev));
436		return -EINVAL;
437	}
438	if (md_check_no_bitmap(mddev))
439		return -EINVAL;
440	blk_queue_max_hw_sectors(mddev->queue, mddev->chunk_sectors);
441	blk_queue_max_write_same_sectors(mddev->queue, mddev->chunk_sectors);
442	blk_queue_max_discard_sectors(mddev->queue, mddev->chunk_sectors);
443
444	/* if private is not null, we are here after takeover */
445	if (mddev->private == NULL) {
446		ret = create_strip_zones(mddev, &conf);
447		if (ret < 0)
448			return ret;
449		mddev->private = conf;
450	}
451	conf = mddev->private;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
452
453	/* calculate array device size */
454	md_set_array_sectors(mddev, raid0_size(mddev, 0, 0));
455
456	printk(KERN_INFO "md/raid0:%s: md_size is %llu sectors.\n",
457	       mdname(mddev),
458	       (unsigned long long)mddev->array_sectors);
459	/* calculate the max read-ahead size.
460	 * For read-ahead of large files to be effective, we need to
461	 * readahead at least twice a whole stripe. i.e. number of devices
462	 * multiplied by chunk size times 2.
463	 * If an individual device has an ra_pages greater than the
464	 * chunk size, then we will not drive that device as hard as it
465	 * wants.  We consider this a configuration error: a larger
466	 * chunksize should be used in that case.
467	 */
468	{
 
469		int stripe = mddev->raid_disks *
470			(mddev->chunk_sectors << 9) / PAGE_SIZE;
471		if (mddev->queue->backing_dev_info.ra_pages < 2* stripe)
472			mddev->queue->backing_dev_info.ra_pages = 2* stripe;
473	}
474
475	blk_queue_merge_bvec(mddev->queue, raid0_mergeable_bvec);
476	dump_zones(mddev);
477
478	ret = md_integrity_register(mddev);
479	if (ret)
480		raid0_stop(mddev);
481
482	return ret;
483}
484
485static int raid0_stop(struct mddev *mddev)
486{
487	struct r0conf *conf = mddev->private;
488
489	blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/
490	kfree(conf->strip_zone);
491	kfree(conf->devlist);
492	kfree(conf);
493	mddev->private = NULL;
494	return 0;
495}
496
497/*
498 * Is io distribute over 1 or more chunks ?
499*/
500static inline int is_io_in_chunk_boundary(struct mddev *mddev,
501			unsigned int chunk_sects, struct bio *bio)
502{
503	if (likely(is_power_of_2(chunk_sects))) {
504		return chunk_sects >=
505			((bio->bi_iter.bi_sector & (chunk_sects-1))
506					+ bio_sectors(bio));
507	} else{
508		sector_t sector = bio->bi_iter.bi_sector;
509		return chunk_sects >= (sector_div(sector, chunk_sects)
510						+ bio_sectors(bio));
511	}
512}
513
514static void raid0_make_request(struct mddev *mddev, struct bio *bio)
515{
516	struct strip_zone *zone;
517	struct md_rdev *tmp_dev;
518	struct bio *split;
519
520	if (unlikely(bio->bi_rw & REQ_FLUSH)) {
521		md_flush_request(mddev, bio);
522		return;
523	}
524
525	do {
526		sector_t sector = bio->bi_iter.bi_sector;
527		unsigned chunk_sects = mddev->chunk_sectors;
528
529		unsigned sectors = chunk_sects -
530			(likely(is_power_of_2(chunk_sects))
531			 ? (sector & (chunk_sects-1))
532			 : sector_div(sector, chunk_sects));
533
 
 
 
534		if (sectors < bio_sectors(bio)) {
535			split = bio_split(bio, sectors, GFP_NOIO, fs_bio_set);
536			bio_chain(split, bio);
537		} else {
538			split = bio;
539		}
540
541		zone = find_zone(mddev->private, &sector);
542		tmp_dev = map_sector(mddev, zone, sector, &sector);
543		split->bi_bdev = tmp_dev->bdev;
544		split->bi_iter.bi_sector = sector + zone->dev_start +
545			tmp_dev->data_offset;
546
547		if (unlikely((split->bi_rw & REQ_DISCARD) &&
548			 !blk_queue_discard(bdev_get_queue(split->bi_bdev)))) {
549			/* Just ignore it */
550			bio_endio(split, 0);
551		} else
552			generic_make_request(split);
553	} while (split != bio);
554}
555
556static void raid0_status(struct seq_file *seq, struct mddev *mddev)
557{
558	seq_printf(seq, " %dk chunks", mddev->chunk_sectors / 2);
559	return;
560}
561
562static void *raid0_takeover_raid45(struct mddev *mddev)
563{
564	struct md_rdev *rdev;
565	struct r0conf *priv_conf;
566
567	if (mddev->degraded != 1) {
568		printk(KERN_ERR "md/raid0:%s: raid5 must be degraded! Degraded disks: %d\n",
569		       mdname(mddev),
570		       mddev->degraded);
571		return ERR_PTR(-EINVAL);
572	}
573
574	rdev_for_each(rdev, mddev) {
575		/* check slot number for a disk */
576		if (rdev->raid_disk == mddev->raid_disks-1) {
577			printk(KERN_ERR "md/raid0:%s: raid5 must have missing parity disk!\n",
578			       mdname(mddev));
579			return ERR_PTR(-EINVAL);
580		}
581		rdev->sectors = mddev->dev_sectors;
582	}
583
584	/* Set new parameters */
585	mddev->new_level = 0;
586	mddev->new_layout = 0;
587	mddev->new_chunk_sectors = mddev->chunk_sectors;
588	mddev->raid_disks--;
589	mddev->delta_disks = -1;
590	/* make sure it will be not marked as dirty */
591	mddev->recovery_cp = MaxSector;
592
593	create_strip_zones(mddev, &priv_conf);
594	return priv_conf;
595}
596
597static void *raid0_takeover_raid10(struct mddev *mddev)
598{
599	struct r0conf *priv_conf;
600
601	/* Check layout:
602	 *  - far_copies must be 1
603	 *  - near_copies must be 2
604	 *  - disks number must be even
605	 *  - all mirrors must be already degraded
606	 */
607	if (mddev->layout != ((1 << 8) + 2)) {
608		printk(KERN_ERR "md/raid0:%s:: Raid0 cannot takover layout: 0x%x\n",
609		       mdname(mddev),
610		       mddev->layout);
611		return ERR_PTR(-EINVAL);
612	}
613	if (mddev->raid_disks & 1) {
614		printk(KERN_ERR "md/raid0:%s: Raid0 cannot takover Raid10 with odd disk number.\n",
615		       mdname(mddev));
616		return ERR_PTR(-EINVAL);
617	}
618	if (mddev->degraded != (mddev->raid_disks>>1)) {
619		printk(KERN_ERR "md/raid0:%s: All mirrors must be already degraded!\n",
620		       mdname(mddev));
621		return ERR_PTR(-EINVAL);
622	}
623
624	/* Set new parameters */
625	mddev->new_level = 0;
626	mddev->new_layout = 0;
627	mddev->new_chunk_sectors = mddev->chunk_sectors;
628	mddev->delta_disks = - mddev->raid_disks / 2;
629	mddev->raid_disks += mddev->delta_disks;
630	mddev->degraded = 0;
631	/* make sure it will be not marked as dirty */
632	mddev->recovery_cp = MaxSector;
633
634	create_strip_zones(mddev, &priv_conf);
635	return priv_conf;
636}
637
638static void *raid0_takeover_raid1(struct mddev *mddev)
639{
640	struct r0conf *priv_conf;
641	int chunksect;
642
643	/* Check layout:
644	 *  - (N - 1) mirror drives must be already faulty
645	 */
646	if ((mddev->raid_disks - 1) != mddev->degraded) {
647		printk(KERN_ERR "md/raid0:%s: (N - 1) mirrors drives must be already faulty!\n",
648		       mdname(mddev));
649		return ERR_PTR(-EINVAL);
650	}
651
652	/*
653	 * a raid1 doesn't have the notion of chunk size, so
654	 * figure out the largest suitable size we can use.
655	 */
656	chunksect = 64 * 2; /* 64K by default */
657
658	/* The array must be an exact multiple of chunksize */
659	while (chunksect && (mddev->array_sectors & (chunksect - 1)))
660		chunksect >>= 1;
661
662	if ((chunksect << 9) < PAGE_SIZE)
663		/* array size does not allow a suitable chunk size */
664		return ERR_PTR(-EINVAL);
665
666	/* Set new parameters */
667	mddev->new_level = 0;
668	mddev->new_layout = 0;
669	mddev->new_chunk_sectors = chunksect;
670	mddev->chunk_sectors = chunksect;
671	mddev->delta_disks = 1 - mddev->raid_disks;
672	mddev->raid_disks = 1;
673	/* make sure it will be not marked as dirty */
674	mddev->recovery_cp = MaxSector;
675
676	create_strip_zones(mddev, &priv_conf);
677	return priv_conf;
678}
679
680static void *raid0_takeover(struct mddev *mddev)
681{
682	/* raid0 can take over:
683	 *  raid4 - if all data disks are active.
684	 *  raid5 - providing it is Raid4 layout and one disk is faulty
685	 *  raid10 - assuming we have all necessary active disks
686	 *  raid1 - with (N -1) mirror drives faulty
687	 */
 
 
 
 
 
 
688	if (mddev->level == 4)
689		return raid0_takeover_raid45(mddev);
690
691	if (mddev->level == 5) {
692		if (mddev->layout == ALGORITHM_PARITY_N)
693			return raid0_takeover_raid45(mddev);
694
695		printk(KERN_ERR "md/raid0:%s: Raid can only takeover Raid5 with layout: %d\n",
696		       mdname(mddev), ALGORITHM_PARITY_N);
697	}
698
699	if (mddev->level == 10)
700		return raid0_takeover_raid10(mddev);
701
702	if (mddev->level == 1)
703		return raid0_takeover_raid1(mddev);
704
705	printk(KERN_ERR "Takeover from raid%i to raid0 not supported\n",
706		mddev->level);
707
708	return ERR_PTR(-EINVAL);
709}
710
711static void raid0_quiesce(struct mddev *mddev, int state)
712{
713}
714
715static struct md_personality raid0_personality=
716{
717	.name		= "raid0",
718	.level		= 0,
719	.owner		= THIS_MODULE,
720	.make_request	= raid0_make_request,
721	.run		= raid0_run,
722	.stop		= raid0_stop,
723	.status		= raid0_status,
724	.size		= raid0_size,
725	.takeover	= raid0_takeover,
726	.quiesce	= raid0_quiesce,
 
727};
728
729static int __init raid0_init (void)
730{
731	return register_md_personality (&raid0_personality);
732}
733
734static void raid0_exit (void)
735{
736	unregister_md_personality (&raid0_personality);
737}
738
739module_init(raid0_init);
740module_exit(raid0_exit);
741MODULE_LICENSE("GPL");
742MODULE_DESCRIPTION("RAID0 (striping) personality for MD");
743MODULE_ALIAS("md-personality-2"); /* RAID0 */
744MODULE_ALIAS("md-raid0");
745MODULE_ALIAS("md-level-0");
v4.6
  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
 28static 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*/
 46static 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}
 74
 75static int create_strip_zones(struct mddev *mddev, struct r0conf **private_conf)
 76{
 77	int i, c, err;
 78	sector_t curr_zone_end, sectors;
 79	struct md_rdev *smallest, *rdev1, *rdev2, *rdev, **dev;
 80	struct strip_zone *zone;
 81	int cnt;
 82	char b[BDEVNAME_SIZE];
 83	char b2[BDEVNAME_SIZE];
 84	struct r0conf *conf = kzalloc(sizeof(*conf), GFP_KERNEL);
 85	unsigned short blksize = 512;
 86
 87	*private_conf = ERR_PTR(-ENOMEM);
 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		blksize = max(blksize, queue_logical_block_size(
102				      rdev1->bdev->bd_disk->queue));
103
104		rdev_for_each(rdev2, mddev) {
105			pr_debug("md/raid0:%s:   comparing %s(%llu)"
106				 " with %s(%llu)\n",
107				 mdname(mddev),
108				 bdevname(rdev1->bdev,b),
109				 (unsigned long long)rdev1->sectors,
110				 bdevname(rdev2->bdev,b2),
111				 (unsigned long long)rdev2->sectors);
112			if (rdev2 == rdev1) {
113				pr_debug("md/raid0:%s:   END\n",
114					 mdname(mddev));
115				break;
116			}
117			if (rdev2->sectors == rdev1->sectors) {
118				/*
119				 * Not unique, don't count it as a new
120				 * group
121				 */
122				pr_debug("md/raid0:%s:   EQUAL\n",
123					 mdname(mddev));
124				c = 1;
125				break;
126			}
127			pr_debug("md/raid0:%s:   NOT EQUAL\n",
128				 mdname(mddev));
129		}
130		if (!c) {
131			pr_debug("md/raid0:%s:   ==> UNIQUE\n",
132				 mdname(mddev));
133			conf->nr_strip_zones++;
134			pr_debug("md/raid0:%s: %d zones\n",
135				 mdname(mddev), conf->nr_strip_zones);
136		}
137	}
138	pr_debug("md/raid0:%s: FINAL %d zones\n",
139		 mdname(mddev), conf->nr_strip_zones);
140	/*
141	 * now since we have the hard sector sizes, we can make sure
142	 * chunk size is a multiple of that sector size
143	 */
144	if ((mddev->chunk_sectors << 9) % blksize) {
145		printk(KERN_ERR "md/raid0:%s: chunk_size of %d not multiple of block size %d\n",
146		       mdname(mddev),
147		       mddev->chunk_sectors << 9, blksize);
148		err = -EINVAL;
149		goto abort;
150	}
151
152	err = -ENOMEM;
153	conf->strip_zone = kzalloc(sizeof(struct strip_zone)*
154				conf->nr_strip_zones, GFP_KERNEL);
155	if (!conf->strip_zone)
156		goto abort;
157	conf->devlist = kzalloc(sizeof(struct md_rdev*)*
158				conf->nr_strip_zones*mddev->raid_disks,
159				GFP_KERNEL);
160	if (!conf->devlist)
161		goto abort;
162
163	/* The first zone must contain all devices, so here we check that
164	 * there is a proper alignment of slots to devices and find them all
165	 */
166	zone = &conf->strip_zone[0];
167	cnt = 0;
168	smallest = NULL;
169	dev = conf->devlist;
170	err = -EINVAL;
171	rdev_for_each(rdev1, mddev) {
172		int j = rdev1->raid_disk;
173
174		if (mddev->level == 10) {
175			/* taking over a raid10-n2 array */
176			j /= 2;
177			rdev1->new_raid_disk = j;
178		}
179
180		if (mddev->level == 1) {
181			/* taiking over a raid1 array-
182			 * we have only one active disk
183			 */
184			j = 0;
185			rdev1->new_raid_disk = j;
186		}
187
188		if (j < 0) {
189			printk(KERN_ERR
190			       "md/raid0:%s: remove inactive devices before converting to RAID0\n",
191			       mdname(mddev));
192			goto abort;
193		}
194		if (j >= mddev->raid_disks) {
195			printk(KERN_ERR "md/raid0:%s: bad disk number %d - "
196			       "aborting!\n", mdname(mddev), j);
197			goto abort;
198		}
199		if (dev[j]) {
200			printk(KERN_ERR "md/raid0:%s: multiple devices for %d - "
201			       "aborting!\n", mdname(mddev), j);
202			goto abort;
203		}
204		dev[j] = rdev1;
205
 
 
 
 
 
 
206		if (!smallest || (rdev1->sectors < smallest->sectors))
207			smallest = rdev1;
208		cnt++;
 
 
 
209	}
210	if (cnt != mddev->raid_disks) {
211		printk(KERN_ERR "md/raid0:%s: too few disks (%d of %d) - "
212		       "aborting!\n", mdname(mddev), cnt, mddev->raid_disks);
213		goto abort;
214	}
215	zone->nb_dev = cnt;
216	zone->zone_end = smallest->sectors * cnt;
217
218	curr_zone_end = zone->zone_end;
219
220	/* now do the other zones */
221	for (i = 1; i < conf->nr_strip_zones; i++)
222	{
223		int j;
224
225		zone = conf->strip_zone + i;
226		dev = conf->devlist + i * mddev->raid_disks;
227
228		pr_debug("md/raid0:%s: zone %d\n", mdname(mddev), i);
229		zone->dev_start = smallest->sectors;
230		smallest = NULL;
231		c = 0;
232
233		for (j=0; j<cnt; j++) {
234			rdev = conf->devlist[j];
235			if (rdev->sectors <= zone->dev_start) {
236				pr_debug("md/raid0:%s: checking %s ... nope\n",
237					 mdname(mddev),
238					 bdevname(rdev->bdev, b));
239				continue;
240			}
241			pr_debug("md/raid0:%s: checking %s ..."
242				 " contained as device %d\n",
243				 mdname(mddev),
244				 bdevname(rdev->bdev, b), c);
245			dev[c] = rdev;
246			c++;
247			if (!smallest || rdev->sectors < smallest->sectors) {
248				smallest = rdev;
249				pr_debug("md/raid0:%s:  (%llu) is smallest!.\n",
250					 mdname(mddev),
251					 (unsigned long long)rdev->sectors);
252			}
253		}
254
255		zone->nb_dev = c;
256		sectors = (smallest->sectors - zone->dev_start) * c;
257		pr_debug("md/raid0:%s: zone->nb_dev: %d, sectors: %llu\n",
258			 mdname(mddev),
259			 zone->nb_dev, (unsigned long long)sectors);
260
261		curr_zone_end += sectors;
262		zone->zone_end = curr_zone_end;
263
264		pr_debug("md/raid0:%s: current zone start: %llu\n",
265			 mdname(mddev),
266			 (unsigned long long)smallest->sectors);
267	}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
268
269	pr_debug("md/raid0:%s: done.\n", mdname(mddev));
270	*private_conf = conf;
271
272	return 0;
273abort:
274	kfree(conf->strip_zone);
275	kfree(conf->devlist);
276	kfree(conf);
277	*private_conf = ERR_PTR(err);
278	return err;
279}
280
281/* Find the zone which holds a particular offset
282 * Update *sectorp to be an offset in that zone
283 */
284static struct strip_zone *find_zone(struct r0conf *conf,
285				    sector_t *sectorp)
286{
287	int i;
288	struct strip_zone *z = conf->strip_zone;
289	sector_t sector = *sectorp;
290
291	for (i = 0; i < conf->nr_strip_zones; i++)
292		if (sector < z[i].zone_end) {
293			if (i)
294				*sectorp = sector - z[i-1].zone_end;
295			return z + i;
296		}
297	BUG();
298}
299
300/*
301 * remaps the bio to the target device. we separate two flows.
302 * power 2 flow and a general flow for the sake of performance
303*/
304static struct md_rdev *map_sector(struct mddev *mddev, struct strip_zone *zone,
305				sector_t sector, sector_t *sector_offset)
306{
307	unsigned int sect_in_chunk;
308	sector_t chunk;
309	struct r0conf *conf = mddev->private;
310	int raid_disks = conf->strip_zone[0].nb_dev;
311	unsigned int chunk_sects = mddev->chunk_sectors;
312
313	if (is_power_of_2(chunk_sects)) {
314		int chunksect_bits = ffz(~chunk_sects);
315		/* find the sector offset inside the chunk */
316		sect_in_chunk  = sector & (chunk_sects - 1);
317		sector >>= chunksect_bits;
318		/* chunk in zone */
319		chunk = *sector_offset;
320		/* quotient is the chunk in real device*/
321		sector_div(chunk, zone->nb_dev << chunksect_bits);
322	} else{
323		sect_in_chunk = sector_div(sector, chunk_sects);
324		chunk = *sector_offset;
325		sector_div(chunk, chunk_sects * zone->nb_dev);
326	}
327	/*
328	*  position the bio over the real device
329	*  real sector = chunk in device + starting of zone
330	*	+ the position in the chunk
331	*/
332	*sector_offset = (chunk * chunk_sects) + sect_in_chunk;
333	return conf->devlist[(zone - conf->strip_zone)*raid_disks
334			     + sector_div(sector, zone->nb_dev)];
335}
336
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
337static sector_t raid0_size(struct mddev *mddev, sector_t sectors, int raid_disks)
338{
339	sector_t array_sectors = 0;
340	struct md_rdev *rdev;
341
342	WARN_ONCE(sectors || raid_disks,
343		  "%s does not support generic reshape\n", __func__);
344
345	rdev_for_each(rdev, mddev)
346		array_sectors += (rdev->sectors &
347				  ~(sector_t)(mddev->chunk_sectors-1));
348
349	return array_sectors;
350}
351
352static void raid0_free(struct mddev *mddev, void *priv);
353
354static int raid0_run(struct mddev *mddev)
355{
356	struct r0conf *conf;
357	int ret;
358
359	if (mddev->chunk_sectors == 0) {
360		printk(KERN_ERR "md/raid0:%s: chunk size must be set.\n",
361		       mdname(mddev));
362		return -EINVAL;
363	}
364	if (md_check_no_bitmap(mddev))
365		return -EINVAL;
 
 
 
366
367	/* if private is not null, we are here after takeover */
368	if (mddev->private == NULL) {
369		ret = create_strip_zones(mddev, &conf);
370		if (ret < 0)
371			return ret;
372		mddev->private = conf;
373	}
374	conf = mddev->private;
375	if (mddev->queue) {
376		struct md_rdev *rdev;
377		bool discard_supported = false;
378
379		blk_queue_max_hw_sectors(mddev->queue, mddev->chunk_sectors);
380		blk_queue_max_write_same_sectors(mddev->queue, mddev->chunk_sectors);
381		blk_queue_max_discard_sectors(mddev->queue, mddev->chunk_sectors);
382
383		blk_queue_io_min(mddev->queue, mddev->chunk_sectors << 9);
384		blk_queue_io_opt(mddev->queue,
385				 (mddev->chunk_sectors << 9) * mddev->raid_disks);
386
387		rdev_for_each(rdev, mddev) {
388			disk_stack_limits(mddev->gendisk, rdev->bdev,
389					  rdev->data_offset << 9);
390			if (blk_queue_discard(bdev_get_queue(rdev->bdev)))
391				discard_supported = true;
392		}
393		if (!discard_supported)
394			queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, mddev->queue);
395		else
396			queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, mddev->queue);
397	}
398
399	/* calculate array device size */
400	md_set_array_sectors(mddev, raid0_size(mddev, 0, 0));
401
402	printk(KERN_INFO "md/raid0:%s: md_size is %llu sectors.\n",
403	       mdname(mddev),
404	       (unsigned long long)mddev->array_sectors);
405
406	if (mddev->queue) {
407		/* calculate the max read-ahead size.
408		 * For read-ahead of large files to be effective, we need to
409		 * readahead at least twice a whole stripe. i.e. number of devices
410		 * multiplied by chunk size times 2.
411		 * If an individual device has an ra_pages greater than the
412		 * chunk size, then we will not drive that device as hard as it
413		 * wants.  We consider this a configuration error: a larger
414		 * chunksize should be used in that case.
415		 */
416		int stripe = mddev->raid_disks *
417			(mddev->chunk_sectors << 9) / PAGE_SIZE;
418		if (mddev->queue->backing_dev_info.ra_pages < 2* stripe)
419			mddev->queue->backing_dev_info.ra_pages = 2* stripe;
420	}
421
 
422	dump_zones(mddev);
423
424	ret = md_integrity_register(mddev);
 
 
425
426	return ret;
427}
428
429static void raid0_free(struct mddev *mddev, void *priv)
430{
431	struct r0conf *conf = priv;
432
 
433	kfree(conf->strip_zone);
434	kfree(conf->devlist);
435	kfree(conf);
 
 
436}
437
438/*
439 * Is io distribute over 1 or more chunks ?
440*/
441static inline int is_io_in_chunk_boundary(struct mddev *mddev,
442			unsigned int chunk_sects, struct bio *bio)
443{
444	if (likely(is_power_of_2(chunk_sects))) {
445		return chunk_sects >=
446			((bio->bi_iter.bi_sector & (chunk_sects-1))
447					+ bio_sectors(bio));
448	} else{
449		sector_t sector = bio->bi_iter.bi_sector;
450		return chunk_sects >= (sector_div(sector, chunk_sects)
451						+ bio_sectors(bio));
452	}
453}
454
455static void raid0_make_request(struct mddev *mddev, struct bio *bio)
456{
457	struct strip_zone *zone;
458	struct md_rdev *tmp_dev;
459	struct bio *split;
460
461	if (unlikely(bio->bi_rw & REQ_FLUSH)) {
462		md_flush_request(mddev, bio);
463		return;
464	}
465
466	do {
467		sector_t sector = bio->bi_iter.bi_sector;
468		unsigned chunk_sects = mddev->chunk_sectors;
469
470		unsigned sectors = chunk_sects -
471			(likely(is_power_of_2(chunk_sects))
472			 ? (sector & (chunk_sects-1))
473			 : sector_div(sector, chunk_sects));
474
475		/* Restore due to sector_div */
476		sector = bio->bi_iter.bi_sector;
477
478		if (sectors < bio_sectors(bio)) {
479			split = bio_split(bio, sectors, GFP_NOIO, fs_bio_set);
480			bio_chain(split, bio);
481		} else {
482			split = bio;
483		}
484
485		zone = find_zone(mddev->private, &sector);
486		tmp_dev = map_sector(mddev, zone, sector, &sector);
487		split->bi_bdev = tmp_dev->bdev;
488		split->bi_iter.bi_sector = sector + zone->dev_start +
489			tmp_dev->data_offset;
490
491		if (unlikely((split->bi_rw & REQ_DISCARD) &&
492			 !blk_queue_discard(bdev_get_queue(split->bi_bdev)))) {
493			/* Just ignore it */
494			bio_endio(split);
495		} else
496			generic_make_request(split);
497	} while (split != bio);
498}
499
500static void raid0_status(struct seq_file *seq, struct mddev *mddev)
501{
502	seq_printf(seq, " %dk chunks", mddev->chunk_sectors / 2);
503	return;
504}
505
506static void *raid0_takeover_raid45(struct mddev *mddev)
507{
508	struct md_rdev *rdev;
509	struct r0conf *priv_conf;
510
511	if (mddev->degraded != 1) {
512		printk(KERN_ERR "md/raid0:%s: raid5 must be degraded! Degraded disks: %d\n",
513		       mdname(mddev),
514		       mddev->degraded);
515		return ERR_PTR(-EINVAL);
516	}
517
518	rdev_for_each(rdev, mddev) {
519		/* check slot number for a disk */
520		if (rdev->raid_disk == mddev->raid_disks-1) {
521			printk(KERN_ERR "md/raid0:%s: raid5 must have missing parity disk!\n",
522			       mdname(mddev));
523			return ERR_PTR(-EINVAL);
524		}
525		rdev->sectors = mddev->dev_sectors;
526	}
527
528	/* Set new parameters */
529	mddev->new_level = 0;
530	mddev->new_layout = 0;
531	mddev->new_chunk_sectors = mddev->chunk_sectors;
532	mddev->raid_disks--;
533	mddev->delta_disks = -1;
534	/* make sure it will be not marked as dirty */
535	mddev->recovery_cp = MaxSector;
536
537	create_strip_zones(mddev, &priv_conf);
538	return priv_conf;
539}
540
541static void *raid0_takeover_raid10(struct mddev *mddev)
542{
543	struct r0conf *priv_conf;
544
545	/* Check layout:
546	 *  - far_copies must be 1
547	 *  - near_copies must be 2
548	 *  - disks number must be even
549	 *  - all mirrors must be already degraded
550	 */
551	if (mddev->layout != ((1 << 8) + 2)) {
552		printk(KERN_ERR "md/raid0:%s:: Raid0 cannot takeover layout: 0x%x\n",
553		       mdname(mddev),
554		       mddev->layout);
555		return ERR_PTR(-EINVAL);
556	}
557	if (mddev->raid_disks & 1) {
558		printk(KERN_ERR "md/raid0:%s: Raid0 cannot takeover Raid10 with odd disk number.\n",
559		       mdname(mddev));
560		return ERR_PTR(-EINVAL);
561	}
562	if (mddev->degraded != (mddev->raid_disks>>1)) {
563		printk(KERN_ERR "md/raid0:%s: All mirrors must be already degraded!\n",
564		       mdname(mddev));
565		return ERR_PTR(-EINVAL);
566	}
567
568	/* Set new parameters */
569	mddev->new_level = 0;
570	mddev->new_layout = 0;
571	mddev->new_chunk_sectors = mddev->chunk_sectors;
572	mddev->delta_disks = - mddev->raid_disks / 2;
573	mddev->raid_disks += mddev->delta_disks;
574	mddev->degraded = 0;
575	/* make sure it will be not marked as dirty */
576	mddev->recovery_cp = MaxSector;
577
578	create_strip_zones(mddev, &priv_conf);
579	return priv_conf;
580}
581
582static void *raid0_takeover_raid1(struct mddev *mddev)
583{
584	struct r0conf *priv_conf;
585	int chunksect;
586
587	/* Check layout:
588	 *  - (N - 1) mirror drives must be already faulty
589	 */
590	if ((mddev->raid_disks - 1) != mddev->degraded) {
591		printk(KERN_ERR "md/raid0:%s: (N - 1) mirrors drives must be already faulty!\n",
592		       mdname(mddev));
593		return ERR_PTR(-EINVAL);
594	}
595
596	/*
597	 * a raid1 doesn't have the notion of chunk size, so
598	 * figure out the largest suitable size we can use.
599	 */
600	chunksect = 64 * 2; /* 64K by default */
601
602	/* The array must be an exact multiple of chunksize */
603	while (chunksect && (mddev->array_sectors & (chunksect - 1)))
604		chunksect >>= 1;
605
606	if ((chunksect << 9) < PAGE_SIZE)
607		/* array size does not allow a suitable chunk size */
608		return ERR_PTR(-EINVAL);
609
610	/* Set new parameters */
611	mddev->new_level = 0;
612	mddev->new_layout = 0;
613	mddev->new_chunk_sectors = chunksect;
614	mddev->chunk_sectors = chunksect;
615	mddev->delta_disks = 1 - mddev->raid_disks;
616	mddev->raid_disks = 1;
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
624static void *raid0_takeover(struct mddev *mddev)
625{
626	/* raid0 can take over:
627	 *  raid4 - if all data disks are active.
628	 *  raid5 - providing it is Raid4 layout and one disk is faulty
629	 *  raid10 - assuming we have all necessary active disks
630	 *  raid1 - with (N -1) mirror drives faulty
631	 */
632
633	if (mddev->bitmap) {
634		printk(KERN_ERR "md/raid0: %s: cannot takeover array with bitmap\n",
635		       mdname(mddev));
636		return ERR_PTR(-EBUSY);
637	}
638	if (mddev->level == 4)
639		return raid0_takeover_raid45(mddev);
640
641	if (mddev->level == 5) {
642		if (mddev->layout == ALGORITHM_PARITY_N)
643			return raid0_takeover_raid45(mddev);
644
645		printk(KERN_ERR "md/raid0:%s: Raid can only takeover Raid5 with layout: %d\n",
646		       mdname(mddev), ALGORITHM_PARITY_N);
647	}
648
649	if (mddev->level == 10)
650		return raid0_takeover_raid10(mddev);
651
652	if (mddev->level == 1)
653		return raid0_takeover_raid1(mddev);
654
655	printk(KERN_ERR "Takeover from raid%i to raid0 not supported\n",
656		mddev->level);
657
658	return ERR_PTR(-EINVAL);
659}
660
661static void raid0_quiesce(struct mddev *mddev, int state)
662{
663}
664
665static struct md_personality raid0_personality=
666{
667	.name		= "raid0",
668	.level		= 0,
669	.owner		= THIS_MODULE,
670	.make_request	= raid0_make_request,
671	.run		= raid0_run,
672	.free		= raid0_free,
673	.status		= raid0_status,
674	.size		= raid0_size,
675	.takeover	= raid0_takeover,
676	.quiesce	= raid0_quiesce,
677	.congested	= raid0_congested,
678};
679
680static int __init raid0_init (void)
681{
682	return register_md_personality (&raid0_personality);
683}
684
685static void raid0_exit (void)
686{
687	unregister_md_personality (&raid0_personality);
688}
689
690module_init(raid0_init);
691module_exit(raid0_exit);
692MODULE_LICENSE("GPL");
693MODULE_DESCRIPTION("RAID0 (striping) personality for MD");
694MODULE_ALIAS("md-personality-2"); /* RAID0 */
695MODULE_ALIAS("md-raid0");
696MODULE_ALIAS("md-level-0");