xref: /linux/drivers/md/dm.c (revision f3d9478b2ce468c3115b02ecae7e975990697f15)
1 /*
2  * Copyright (C) 2001, 2002 Sistina Software (UK) Limited.
3  * Copyright (C) 2004 Red Hat, Inc. All rights reserved.
4  *
5  * This file is released under the GPL.
6  */
7 
8 #include "dm.h"
9 #include "dm-bio-list.h"
10 
11 #include <linux/init.h>
12 #include <linux/module.h>
13 #include <linux/mutex.h>
14 #include <linux/moduleparam.h>
15 #include <linux/blkpg.h>
16 #include <linux/bio.h>
17 #include <linux/buffer_head.h>
18 #include <linux/mempool.h>
19 #include <linux/slab.h>
20 #include <linux/idr.h>
21 #include <linux/hdreg.h>
22 #include <linux/blktrace_api.h>
23 
24 static const char *_name = DM_NAME;
25 
26 static unsigned int major = 0;
27 static unsigned int _major = 0;
28 
29 /*
30  * One of these is allocated per bio.
31  */
32 struct dm_io {
33 	struct mapped_device *md;
34 	int error;
35 	struct bio *bio;
36 	atomic_t io_count;
37 	unsigned long start_time;
38 };
39 
40 /*
41  * One of these is allocated per target within a bio.  Hopefully
42  * this will be simplified out one day.
43  */
44 struct target_io {
45 	struct dm_io *io;
46 	struct dm_target *ti;
47 	union map_info info;
48 };
49 
50 union map_info *dm_get_mapinfo(struct bio *bio)
51 {
52         if (bio && bio->bi_private)
53                 return &((struct target_io *)bio->bi_private)->info;
54         return NULL;
55 }
56 
57 /*
58  * Bits for the md->flags field.
59  */
60 #define DMF_BLOCK_IO 0
61 #define DMF_SUSPENDED 1
62 #define DMF_FROZEN 2
63 
64 struct mapped_device {
65 	struct rw_semaphore io_lock;
66 	struct semaphore suspend_lock;
67 	rwlock_t map_lock;
68 	atomic_t holders;
69 
70 	unsigned long flags;
71 
72 	request_queue_t *queue;
73 	struct gendisk *disk;
74 	char name[16];
75 
76 	void *interface_ptr;
77 
78 	/*
79 	 * A list of ios that arrived while we were suspended.
80 	 */
81 	atomic_t pending;
82 	wait_queue_head_t wait;
83  	struct bio_list deferred;
84 
85 	/*
86 	 * The current mapping.
87 	 */
88 	struct dm_table *map;
89 
90 	/*
91 	 * io objects are allocated from here.
92 	 */
93 	mempool_t *io_pool;
94 	mempool_t *tio_pool;
95 
96 	/*
97 	 * Event handling.
98 	 */
99 	atomic_t event_nr;
100 	wait_queue_head_t eventq;
101 
102 	/*
103 	 * freeze/thaw support require holding onto a super block
104 	 */
105 	struct super_block *frozen_sb;
106 	struct block_device *suspended_bdev;
107 
108 	/* forced geometry settings */
109 	struct hd_geometry geometry;
110 };
111 
112 #define MIN_IOS 256
113 static kmem_cache_t *_io_cache;
114 static kmem_cache_t *_tio_cache;
115 
116 static struct bio_set *dm_set;
117 
118 static int __init local_init(void)
119 {
120 	int r;
121 
122 	dm_set = bioset_create(16, 16, 4);
123 	if (!dm_set)
124 		return -ENOMEM;
125 
126 	/* allocate a slab for the dm_ios */
127 	_io_cache = kmem_cache_create("dm_io",
128 				      sizeof(struct dm_io), 0, 0, NULL, NULL);
129 	if (!_io_cache)
130 		return -ENOMEM;
131 
132 	/* allocate a slab for the target ios */
133 	_tio_cache = kmem_cache_create("dm_tio", sizeof(struct target_io),
134 				       0, 0, NULL, NULL);
135 	if (!_tio_cache) {
136 		kmem_cache_destroy(_io_cache);
137 		return -ENOMEM;
138 	}
139 
140 	_major = major;
141 	r = register_blkdev(_major, _name);
142 	if (r < 0) {
143 		kmem_cache_destroy(_tio_cache);
144 		kmem_cache_destroy(_io_cache);
145 		return r;
146 	}
147 
148 	if (!_major)
149 		_major = r;
150 
151 	return 0;
152 }
153 
154 static void local_exit(void)
155 {
156 	kmem_cache_destroy(_tio_cache);
157 	kmem_cache_destroy(_io_cache);
158 
159 	bioset_free(dm_set);
160 
161 	if (unregister_blkdev(_major, _name) < 0)
162 		DMERR("devfs_unregister_blkdev failed");
163 
164 	_major = 0;
165 
166 	DMINFO("cleaned up");
167 }
168 
169 int (*_inits[])(void) __initdata = {
170 	local_init,
171 	dm_target_init,
172 	dm_linear_init,
173 	dm_stripe_init,
174 	dm_interface_init,
175 };
176 
177 void (*_exits[])(void) = {
178 	local_exit,
179 	dm_target_exit,
180 	dm_linear_exit,
181 	dm_stripe_exit,
182 	dm_interface_exit,
183 };
184 
185 static int __init dm_init(void)
186 {
187 	const int count = ARRAY_SIZE(_inits);
188 
189 	int r, i;
190 
191 	for (i = 0; i < count; i++) {
192 		r = _inits[i]();
193 		if (r)
194 			goto bad;
195 	}
196 
197 	return 0;
198 
199       bad:
200 	while (i--)
201 		_exits[i]();
202 
203 	return r;
204 }
205 
206 static void __exit dm_exit(void)
207 {
208 	int i = ARRAY_SIZE(_exits);
209 
210 	while (i--)
211 		_exits[i]();
212 }
213 
214 /*
215  * Block device functions
216  */
217 static int dm_blk_open(struct inode *inode, struct file *file)
218 {
219 	struct mapped_device *md;
220 
221 	md = inode->i_bdev->bd_disk->private_data;
222 	dm_get(md);
223 	return 0;
224 }
225 
226 static int dm_blk_close(struct inode *inode, struct file *file)
227 {
228 	struct mapped_device *md;
229 
230 	md = inode->i_bdev->bd_disk->private_data;
231 	dm_put(md);
232 	return 0;
233 }
234 
235 static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo)
236 {
237 	struct mapped_device *md = bdev->bd_disk->private_data;
238 
239 	return dm_get_geometry(md, geo);
240 }
241 
242 static inline struct dm_io *alloc_io(struct mapped_device *md)
243 {
244 	return mempool_alloc(md->io_pool, GFP_NOIO);
245 }
246 
247 static inline void free_io(struct mapped_device *md, struct dm_io *io)
248 {
249 	mempool_free(io, md->io_pool);
250 }
251 
252 static inline struct target_io *alloc_tio(struct mapped_device *md)
253 {
254 	return mempool_alloc(md->tio_pool, GFP_NOIO);
255 }
256 
257 static inline void free_tio(struct mapped_device *md, struct target_io *tio)
258 {
259 	mempool_free(tio, md->tio_pool);
260 }
261 
262 static void start_io_acct(struct dm_io *io)
263 {
264 	struct mapped_device *md = io->md;
265 
266 	io->start_time = jiffies;
267 
268 	preempt_disable();
269 	disk_round_stats(dm_disk(md));
270 	preempt_enable();
271 	dm_disk(md)->in_flight = atomic_inc_return(&md->pending);
272 }
273 
274 static int end_io_acct(struct dm_io *io)
275 {
276 	struct mapped_device *md = io->md;
277 	struct bio *bio = io->bio;
278 	unsigned long duration = jiffies - io->start_time;
279 	int pending;
280 	int rw = bio_data_dir(bio);
281 
282 	preempt_disable();
283 	disk_round_stats(dm_disk(md));
284 	preempt_enable();
285 	dm_disk(md)->in_flight = pending = atomic_dec_return(&md->pending);
286 
287 	disk_stat_add(dm_disk(md), ticks[rw], duration);
288 
289 	return !pending;
290 }
291 
292 /*
293  * Add the bio to the list of deferred io.
294  */
295 static int queue_io(struct mapped_device *md, struct bio *bio)
296 {
297 	down_write(&md->io_lock);
298 
299 	if (!test_bit(DMF_BLOCK_IO, &md->flags)) {
300 		up_write(&md->io_lock);
301 		return 1;
302 	}
303 
304 	bio_list_add(&md->deferred, bio);
305 
306 	up_write(&md->io_lock);
307 	return 0;		/* deferred successfully */
308 }
309 
310 /*
311  * Everyone (including functions in this file), should use this
312  * function to access the md->map field, and make sure they call
313  * dm_table_put() when finished.
314  */
315 struct dm_table *dm_get_table(struct mapped_device *md)
316 {
317 	struct dm_table *t;
318 
319 	read_lock(&md->map_lock);
320 	t = md->map;
321 	if (t)
322 		dm_table_get(t);
323 	read_unlock(&md->map_lock);
324 
325 	return t;
326 }
327 
328 /*
329  * Get the geometry associated with a dm device
330  */
331 int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo)
332 {
333 	*geo = md->geometry;
334 
335 	return 0;
336 }
337 
338 /*
339  * Set the geometry of a device.
340  */
341 int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo)
342 {
343 	sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors;
344 
345 	if (geo->start > sz) {
346 		DMWARN("Start sector is beyond the geometry limits.");
347 		return -EINVAL;
348 	}
349 
350 	md->geometry = *geo;
351 
352 	return 0;
353 }
354 
355 /*-----------------------------------------------------------------
356  * CRUD START:
357  *   A more elegant soln is in the works that uses the queue
358  *   merge fn, unfortunately there are a couple of changes to
359  *   the block layer that I want to make for this.  So in the
360  *   interests of getting something for people to use I give
361  *   you this clearly demarcated crap.
362  *---------------------------------------------------------------*/
363 
364 /*
365  * Decrements the number of outstanding ios that a bio has been
366  * cloned into, completing the original io if necc.
367  */
368 static void dec_pending(struct dm_io *io, int error)
369 {
370 	if (error)
371 		io->error = error;
372 
373 	if (atomic_dec_and_test(&io->io_count)) {
374 		if (end_io_acct(io))
375 			/* nudge anyone waiting on suspend queue */
376 			wake_up(&io->md->wait);
377 
378 		blk_add_trace_bio(io->md->queue, io->bio, BLK_TA_COMPLETE);
379 
380 		bio_endio(io->bio, io->bio->bi_size, io->error);
381 		free_io(io->md, io);
382 	}
383 }
384 
385 static int clone_endio(struct bio *bio, unsigned int done, int error)
386 {
387 	int r = 0;
388 	struct target_io *tio = bio->bi_private;
389 	struct dm_io *io = tio->io;
390 	dm_endio_fn endio = tio->ti->type->end_io;
391 
392 	if (bio->bi_size)
393 		return 1;
394 
395 	if (!bio_flagged(bio, BIO_UPTODATE) && !error)
396 		error = -EIO;
397 
398 	if (endio) {
399 		r = endio(tio->ti, bio, error, &tio->info);
400 		if (r < 0)
401 			error = r;
402 
403 		else if (r > 0)
404 			/* the target wants another shot at the io */
405 			return 1;
406 	}
407 
408 	free_tio(io->md, tio);
409 	dec_pending(io, error);
410 	bio_put(bio);
411 	return r;
412 }
413 
414 static sector_t max_io_len(struct mapped_device *md,
415 			   sector_t sector, struct dm_target *ti)
416 {
417 	sector_t offset = sector - ti->begin;
418 	sector_t len = ti->len - offset;
419 
420 	/*
421 	 * Does the target need to split even further ?
422 	 */
423 	if (ti->split_io) {
424 		sector_t boundary;
425 		boundary = ((offset + ti->split_io) & ~(ti->split_io - 1))
426 			   - offset;
427 		if (len > boundary)
428 			len = boundary;
429 	}
430 
431 	return len;
432 }
433 
434 static void __map_bio(struct dm_target *ti, struct bio *clone,
435 		      struct target_io *tio)
436 {
437 	int r;
438 	sector_t sector;
439 
440 	/*
441 	 * Sanity checks.
442 	 */
443 	BUG_ON(!clone->bi_size);
444 
445 	clone->bi_end_io = clone_endio;
446 	clone->bi_private = tio;
447 
448 	/*
449 	 * Map the clone.  If r == 0 we don't need to do
450 	 * anything, the target has assumed ownership of
451 	 * this io.
452 	 */
453 	atomic_inc(&tio->io->io_count);
454 	sector = clone->bi_sector;
455 	r = ti->type->map(ti, clone, &tio->info);
456 	if (r > 0) {
457 		/* the bio has been remapped so dispatch it */
458 
459 		blk_add_trace_remap(bdev_get_queue(clone->bi_bdev), clone,
460 				    tio->io->bio->bi_bdev->bd_dev, sector,
461 				    clone->bi_sector);
462 
463 		generic_make_request(clone);
464 	}
465 
466 	else if (r < 0) {
467 		/* error the io and bail out */
468 		struct dm_io *io = tio->io;
469 		free_tio(tio->io->md, tio);
470 		dec_pending(io, r);
471 		bio_put(clone);
472 	}
473 }
474 
475 struct clone_info {
476 	struct mapped_device *md;
477 	struct dm_table *map;
478 	struct bio *bio;
479 	struct dm_io *io;
480 	sector_t sector;
481 	sector_t sector_count;
482 	unsigned short idx;
483 };
484 
485 static void dm_bio_destructor(struct bio *bio)
486 {
487 	bio_free(bio, dm_set);
488 }
489 
490 /*
491  * Creates a little bio that is just does part of a bvec.
492  */
493 static struct bio *split_bvec(struct bio *bio, sector_t sector,
494 			      unsigned short idx, unsigned int offset,
495 			      unsigned int len)
496 {
497 	struct bio *clone;
498 	struct bio_vec *bv = bio->bi_io_vec + idx;
499 
500 	clone = bio_alloc_bioset(GFP_NOIO, 1, dm_set);
501 	clone->bi_destructor = dm_bio_destructor;
502 	*clone->bi_io_vec = *bv;
503 
504 	clone->bi_sector = sector;
505 	clone->bi_bdev = bio->bi_bdev;
506 	clone->bi_rw = bio->bi_rw;
507 	clone->bi_vcnt = 1;
508 	clone->bi_size = to_bytes(len);
509 	clone->bi_io_vec->bv_offset = offset;
510 	clone->bi_io_vec->bv_len = clone->bi_size;
511 
512 	return clone;
513 }
514 
515 /*
516  * Creates a bio that consists of range of complete bvecs.
517  */
518 static struct bio *clone_bio(struct bio *bio, sector_t sector,
519 			     unsigned short idx, unsigned short bv_count,
520 			     unsigned int len)
521 {
522 	struct bio *clone;
523 
524 	clone = bio_clone(bio, GFP_NOIO);
525 	clone->bi_sector = sector;
526 	clone->bi_idx = idx;
527 	clone->bi_vcnt = idx + bv_count;
528 	clone->bi_size = to_bytes(len);
529 	clone->bi_flags &= ~(1 << BIO_SEG_VALID);
530 
531 	return clone;
532 }
533 
534 static void __clone_and_map(struct clone_info *ci)
535 {
536 	struct bio *clone, *bio = ci->bio;
537 	struct dm_target *ti = dm_table_find_target(ci->map, ci->sector);
538 	sector_t len = 0, max = max_io_len(ci->md, ci->sector, ti);
539 	struct target_io *tio;
540 
541 	/*
542 	 * Allocate a target io object.
543 	 */
544 	tio = alloc_tio(ci->md);
545 	tio->io = ci->io;
546 	tio->ti = ti;
547 	memset(&tio->info, 0, sizeof(tio->info));
548 
549 	if (ci->sector_count <= max) {
550 		/*
551 		 * Optimise for the simple case where we can do all of
552 		 * the remaining io with a single clone.
553 		 */
554 		clone = clone_bio(bio, ci->sector, ci->idx,
555 				  bio->bi_vcnt - ci->idx, ci->sector_count);
556 		__map_bio(ti, clone, tio);
557 		ci->sector_count = 0;
558 
559 	} else if (to_sector(bio->bi_io_vec[ci->idx].bv_len) <= max) {
560 		/*
561 		 * There are some bvecs that don't span targets.
562 		 * Do as many of these as possible.
563 		 */
564 		int i;
565 		sector_t remaining = max;
566 		sector_t bv_len;
567 
568 		for (i = ci->idx; remaining && (i < bio->bi_vcnt); i++) {
569 			bv_len = to_sector(bio->bi_io_vec[i].bv_len);
570 
571 			if (bv_len > remaining)
572 				break;
573 
574 			remaining -= bv_len;
575 			len += bv_len;
576 		}
577 
578 		clone = clone_bio(bio, ci->sector, ci->idx, i - ci->idx, len);
579 		__map_bio(ti, clone, tio);
580 
581 		ci->sector += len;
582 		ci->sector_count -= len;
583 		ci->idx = i;
584 
585 	} else {
586 		/*
587 		 * Handle a bvec that must be split between two or more targets.
588 		 */
589 		struct bio_vec *bv = bio->bi_io_vec + ci->idx;
590 		sector_t remaining = to_sector(bv->bv_len);
591 		unsigned int offset = 0;
592 
593 		do {
594 			if (offset) {
595 				ti = dm_table_find_target(ci->map, ci->sector);
596 				max = max_io_len(ci->md, ci->sector, ti);
597 
598 				tio = alloc_tio(ci->md);
599 				tio->io = ci->io;
600 				tio->ti = ti;
601 				memset(&tio->info, 0, sizeof(tio->info));
602 			}
603 
604 			len = min(remaining, max);
605 
606 			clone = split_bvec(bio, ci->sector, ci->idx,
607 					   bv->bv_offset + offset, len);
608 
609 			__map_bio(ti, clone, tio);
610 
611 			ci->sector += len;
612 			ci->sector_count -= len;
613 			offset += to_bytes(len);
614 		} while (remaining -= len);
615 
616 		ci->idx++;
617 	}
618 }
619 
620 /*
621  * Split the bio into several clones.
622  */
623 static void __split_bio(struct mapped_device *md, struct bio *bio)
624 {
625 	struct clone_info ci;
626 
627 	ci.map = dm_get_table(md);
628 	if (!ci.map) {
629 		bio_io_error(bio, bio->bi_size);
630 		return;
631 	}
632 
633 	ci.md = md;
634 	ci.bio = bio;
635 	ci.io = alloc_io(md);
636 	ci.io->error = 0;
637 	atomic_set(&ci.io->io_count, 1);
638 	ci.io->bio = bio;
639 	ci.io->md = md;
640 	ci.sector = bio->bi_sector;
641 	ci.sector_count = bio_sectors(bio);
642 	ci.idx = bio->bi_idx;
643 
644 	start_io_acct(ci.io);
645 	while (ci.sector_count)
646 		__clone_and_map(&ci);
647 
648 	/* drop the extra reference count */
649 	dec_pending(ci.io, 0);
650 	dm_table_put(ci.map);
651 }
652 /*-----------------------------------------------------------------
653  * CRUD END
654  *---------------------------------------------------------------*/
655 
656 /*
657  * The request function that just remaps the bio built up by
658  * dm_merge_bvec.
659  */
660 static int dm_request(request_queue_t *q, struct bio *bio)
661 {
662 	int r;
663 	int rw = bio_data_dir(bio);
664 	struct mapped_device *md = q->queuedata;
665 
666 	down_read(&md->io_lock);
667 
668 	disk_stat_inc(dm_disk(md), ios[rw]);
669 	disk_stat_add(dm_disk(md), sectors[rw], bio_sectors(bio));
670 
671 	/*
672 	 * If we're suspended we have to queue
673 	 * this io for later.
674 	 */
675 	while (test_bit(DMF_BLOCK_IO, &md->flags)) {
676 		up_read(&md->io_lock);
677 
678 		if (bio_rw(bio) == READA) {
679 			bio_io_error(bio, bio->bi_size);
680 			return 0;
681 		}
682 
683 		r = queue_io(md, bio);
684 		if (r < 0) {
685 			bio_io_error(bio, bio->bi_size);
686 			return 0;
687 
688 		} else if (r == 0)
689 			return 0;	/* deferred successfully */
690 
691 		/*
692 		 * We're in a while loop, because someone could suspend
693 		 * before we get to the following read lock.
694 		 */
695 		down_read(&md->io_lock);
696 	}
697 
698 	__split_bio(md, bio);
699 	up_read(&md->io_lock);
700 	return 0;
701 }
702 
703 static int dm_flush_all(request_queue_t *q, struct gendisk *disk,
704 			sector_t *error_sector)
705 {
706 	struct mapped_device *md = q->queuedata;
707 	struct dm_table *map = dm_get_table(md);
708 	int ret = -ENXIO;
709 
710 	if (map) {
711 		ret = dm_table_flush_all(map);
712 		dm_table_put(map);
713 	}
714 
715 	return ret;
716 }
717 
718 static void dm_unplug_all(request_queue_t *q)
719 {
720 	struct mapped_device *md = q->queuedata;
721 	struct dm_table *map = dm_get_table(md);
722 
723 	if (map) {
724 		dm_table_unplug_all(map);
725 		dm_table_put(map);
726 	}
727 }
728 
729 static int dm_any_congested(void *congested_data, int bdi_bits)
730 {
731 	int r;
732 	struct mapped_device *md = (struct mapped_device *) congested_data;
733 	struct dm_table *map = dm_get_table(md);
734 
735 	if (!map || test_bit(DMF_BLOCK_IO, &md->flags))
736 		r = bdi_bits;
737 	else
738 		r = dm_table_any_congested(map, bdi_bits);
739 
740 	dm_table_put(map);
741 	return r;
742 }
743 
744 /*-----------------------------------------------------------------
745  * An IDR is used to keep track of allocated minor numbers.
746  *---------------------------------------------------------------*/
747 static DEFINE_MUTEX(_minor_lock);
748 static DEFINE_IDR(_minor_idr);
749 
750 static void free_minor(unsigned int minor)
751 {
752 	mutex_lock(&_minor_lock);
753 	idr_remove(&_minor_idr, minor);
754 	mutex_unlock(&_minor_lock);
755 }
756 
757 /*
758  * See if the device with a specific minor # is free.
759  */
760 static int specific_minor(struct mapped_device *md, unsigned int minor)
761 {
762 	int r, m;
763 
764 	if (minor >= (1 << MINORBITS))
765 		return -EINVAL;
766 
767 	mutex_lock(&_minor_lock);
768 
769 	if (idr_find(&_minor_idr, minor)) {
770 		r = -EBUSY;
771 		goto out;
772 	}
773 
774 	r = idr_pre_get(&_minor_idr, GFP_KERNEL);
775 	if (!r) {
776 		r = -ENOMEM;
777 		goto out;
778 	}
779 
780 	r = idr_get_new_above(&_minor_idr, md, minor, &m);
781 	if (r) {
782 		goto out;
783 	}
784 
785 	if (m != minor) {
786 		idr_remove(&_minor_idr, m);
787 		r = -EBUSY;
788 		goto out;
789 	}
790 
791 out:
792 	mutex_unlock(&_minor_lock);
793 	return r;
794 }
795 
796 static int next_free_minor(struct mapped_device *md, unsigned int *minor)
797 {
798 	int r;
799 	unsigned int m;
800 
801 	mutex_lock(&_minor_lock);
802 
803 	r = idr_pre_get(&_minor_idr, GFP_KERNEL);
804 	if (!r) {
805 		r = -ENOMEM;
806 		goto out;
807 	}
808 
809 	r = idr_get_new(&_minor_idr, md, &m);
810 	if (r) {
811 		goto out;
812 	}
813 
814 	if (m >= (1 << MINORBITS)) {
815 		idr_remove(&_minor_idr, m);
816 		r = -ENOSPC;
817 		goto out;
818 	}
819 
820 	*minor = m;
821 
822 out:
823 	mutex_unlock(&_minor_lock);
824 	return r;
825 }
826 
827 static struct block_device_operations dm_blk_dops;
828 
829 /*
830  * Allocate and initialise a blank device with a given minor.
831  */
832 static struct mapped_device *alloc_dev(unsigned int minor, int persistent)
833 {
834 	int r;
835 	struct mapped_device *md = kmalloc(sizeof(*md), GFP_KERNEL);
836 
837 	if (!md) {
838 		DMWARN("unable to allocate device, out of memory.");
839 		return NULL;
840 	}
841 
842 	/* get a minor number for the dev */
843 	r = persistent ? specific_minor(md, minor) : next_free_minor(md, &minor);
844 	if (r < 0)
845 		goto bad1;
846 
847 	memset(md, 0, sizeof(*md));
848 	init_rwsem(&md->io_lock);
849 	init_MUTEX(&md->suspend_lock);
850 	rwlock_init(&md->map_lock);
851 	atomic_set(&md->holders, 1);
852 	atomic_set(&md->event_nr, 0);
853 
854 	md->queue = blk_alloc_queue(GFP_KERNEL);
855 	if (!md->queue)
856 		goto bad1;
857 
858 	md->queue->queuedata = md;
859 	md->queue->backing_dev_info.congested_fn = dm_any_congested;
860 	md->queue->backing_dev_info.congested_data = md;
861 	blk_queue_make_request(md->queue, dm_request);
862 	blk_queue_bounce_limit(md->queue, BLK_BOUNCE_ANY);
863 	md->queue->unplug_fn = dm_unplug_all;
864 	md->queue->issue_flush_fn = dm_flush_all;
865 
866 	md->io_pool = mempool_create_slab_pool(MIN_IOS, _io_cache);
867  	if (!md->io_pool)
868  		goto bad2;
869 
870 	md->tio_pool = mempool_create_slab_pool(MIN_IOS, _tio_cache);
871 	if (!md->tio_pool)
872 		goto bad3;
873 
874 	md->disk = alloc_disk(1);
875 	if (!md->disk)
876 		goto bad4;
877 
878 	md->disk->major = _major;
879 	md->disk->first_minor = minor;
880 	md->disk->fops = &dm_blk_dops;
881 	md->disk->queue = md->queue;
882 	md->disk->private_data = md;
883 	sprintf(md->disk->disk_name, "dm-%d", minor);
884 	add_disk(md->disk);
885 	format_dev_t(md->name, MKDEV(_major, minor));
886 
887 	atomic_set(&md->pending, 0);
888 	init_waitqueue_head(&md->wait);
889 	init_waitqueue_head(&md->eventq);
890 
891 	return md;
892 
893  bad4:
894 	mempool_destroy(md->tio_pool);
895  bad3:
896 	mempool_destroy(md->io_pool);
897  bad2:
898 	blk_cleanup_queue(md->queue);
899 	free_minor(minor);
900  bad1:
901 	kfree(md);
902 	return NULL;
903 }
904 
905 static void free_dev(struct mapped_device *md)
906 {
907 	unsigned int minor = md->disk->first_minor;
908 
909 	if (md->suspended_bdev) {
910 		thaw_bdev(md->suspended_bdev, NULL);
911 		bdput(md->suspended_bdev);
912 	}
913 	mempool_destroy(md->tio_pool);
914 	mempool_destroy(md->io_pool);
915 	del_gendisk(md->disk);
916 	free_minor(minor);
917 	put_disk(md->disk);
918 	blk_cleanup_queue(md->queue);
919 	kfree(md);
920 }
921 
922 /*
923  * Bind a table to the device.
924  */
925 static void event_callback(void *context)
926 {
927 	struct mapped_device *md = (struct mapped_device *) context;
928 
929 	atomic_inc(&md->event_nr);
930 	wake_up(&md->eventq);
931 }
932 
933 static void __set_size(struct mapped_device *md, sector_t size)
934 {
935 	set_capacity(md->disk, size);
936 
937 	mutex_lock(&md->suspended_bdev->bd_inode->i_mutex);
938 	i_size_write(md->suspended_bdev->bd_inode, (loff_t)size << SECTOR_SHIFT);
939 	mutex_unlock(&md->suspended_bdev->bd_inode->i_mutex);
940 }
941 
942 static int __bind(struct mapped_device *md, struct dm_table *t)
943 {
944 	request_queue_t *q = md->queue;
945 	sector_t size;
946 
947 	size = dm_table_get_size(t);
948 
949 	/*
950 	 * Wipe any geometry if the size of the table changed.
951 	 */
952 	if (size != get_capacity(md->disk))
953 		memset(&md->geometry, 0, sizeof(md->geometry));
954 
955 	__set_size(md, size);
956 	if (size == 0)
957 		return 0;
958 
959 	dm_table_get(t);
960 	dm_table_event_callback(t, event_callback, md);
961 
962 	write_lock(&md->map_lock);
963 	md->map = t;
964 	dm_table_set_restrictions(t, q);
965 	write_unlock(&md->map_lock);
966 
967 	return 0;
968 }
969 
970 static void __unbind(struct mapped_device *md)
971 {
972 	struct dm_table *map = md->map;
973 
974 	if (!map)
975 		return;
976 
977 	dm_table_event_callback(map, NULL, NULL);
978 	write_lock(&md->map_lock);
979 	md->map = NULL;
980 	write_unlock(&md->map_lock);
981 	dm_table_put(map);
982 }
983 
984 /*
985  * Constructor for a new device.
986  */
987 static int create_aux(unsigned int minor, int persistent,
988 		      struct mapped_device **result)
989 {
990 	struct mapped_device *md;
991 
992 	md = alloc_dev(minor, persistent);
993 	if (!md)
994 		return -ENXIO;
995 
996 	*result = md;
997 	return 0;
998 }
999 
1000 int dm_create(struct mapped_device **result)
1001 {
1002 	return create_aux(0, 0, result);
1003 }
1004 
1005 int dm_create_with_minor(unsigned int minor, struct mapped_device **result)
1006 {
1007 	return create_aux(minor, 1, result);
1008 }
1009 
1010 static struct mapped_device *dm_find_md(dev_t dev)
1011 {
1012 	struct mapped_device *md;
1013 	unsigned minor = MINOR(dev);
1014 
1015 	if (MAJOR(dev) != _major || minor >= (1 << MINORBITS))
1016 		return NULL;
1017 
1018 	mutex_lock(&_minor_lock);
1019 
1020 	md = idr_find(&_minor_idr, minor);
1021 	if (!md || (dm_disk(md)->first_minor != minor))
1022 		md = NULL;
1023 
1024 	mutex_unlock(&_minor_lock);
1025 
1026 	return md;
1027 }
1028 
1029 struct mapped_device *dm_get_md(dev_t dev)
1030 {
1031 	struct mapped_device *md = dm_find_md(dev);
1032 
1033 	if (md)
1034 		dm_get(md);
1035 
1036 	return md;
1037 }
1038 
1039 void *dm_get_mdptr(struct mapped_device *md)
1040 {
1041 	return md->interface_ptr;
1042 }
1043 
1044 void dm_set_mdptr(struct mapped_device *md, void *ptr)
1045 {
1046 	md->interface_ptr = ptr;
1047 }
1048 
1049 void dm_get(struct mapped_device *md)
1050 {
1051 	atomic_inc(&md->holders);
1052 }
1053 
1054 void dm_put(struct mapped_device *md)
1055 {
1056 	struct dm_table *map;
1057 
1058 	if (atomic_dec_and_test(&md->holders)) {
1059 		map = dm_get_table(md);
1060 		if (!dm_suspended(md)) {
1061 			dm_table_presuspend_targets(map);
1062 			dm_table_postsuspend_targets(map);
1063 		}
1064 		__unbind(md);
1065 		dm_table_put(map);
1066 		free_dev(md);
1067 	}
1068 }
1069 
1070 /*
1071  * Process the deferred bios
1072  */
1073 static void __flush_deferred_io(struct mapped_device *md, struct bio *c)
1074 {
1075 	struct bio *n;
1076 
1077 	while (c) {
1078 		n = c->bi_next;
1079 		c->bi_next = NULL;
1080 		__split_bio(md, c);
1081 		c = n;
1082 	}
1083 }
1084 
1085 /*
1086  * Swap in a new table (destroying old one).
1087  */
1088 int dm_swap_table(struct mapped_device *md, struct dm_table *table)
1089 {
1090 	int r = -EINVAL;
1091 
1092 	down(&md->suspend_lock);
1093 
1094 	/* device must be suspended */
1095 	if (!dm_suspended(md))
1096 		goto out;
1097 
1098 	__unbind(md);
1099 	r = __bind(md, table);
1100 
1101 out:
1102 	up(&md->suspend_lock);
1103 	return r;
1104 }
1105 
1106 /*
1107  * Functions to lock and unlock any filesystem running on the
1108  * device.
1109  */
1110 static int lock_fs(struct mapped_device *md)
1111 {
1112 	int r;
1113 
1114 	WARN_ON(md->frozen_sb);
1115 
1116 	md->frozen_sb = freeze_bdev(md->suspended_bdev);
1117 	if (IS_ERR(md->frozen_sb)) {
1118 		r = PTR_ERR(md->frozen_sb);
1119 		md->frozen_sb = NULL;
1120 		return r;
1121 	}
1122 
1123 	set_bit(DMF_FROZEN, &md->flags);
1124 
1125 	/* don't bdput right now, we don't want the bdev
1126 	 * to go away while it is locked.
1127 	 */
1128 	return 0;
1129 }
1130 
1131 static void unlock_fs(struct mapped_device *md)
1132 {
1133 	if (!test_bit(DMF_FROZEN, &md->flags))
1134 		return;
1135 
1136 	thaw_bdev(md->suspended_bdev, md->frozen_sb);
1137 	md->frozen_sb = NULL;
1138 	clear_bit(DMF_FROZEN, &md->flags);
1139 }
1140 
1141 /*
1142  * We need to be able to change a mapping table under a mounted
1143  * filesystem.  For example we might want to move some data in
1144  * the background.  Before the table can be swapped with
1145  * dm_bind_table, dm_suspend must be called to flush any in
1146  * flight bios and ensure that any further io gets deferred.
1147  */
1148 int dm_suspend(struct mapped_device *md, int do_lockfs)
1149 {
1150 	struct dm_table *map = NULL;
1151 	DECLARE_WAITQUEUE(wait, current);
1152 	struct bio *def;
1153 	int r = -EINVAL;
1154 
1155 	down(&md->suspend_lock);
1156 
1157 	if (dm_suspended(md))
1158 		goto out;
1159 
1160 	map = dm_get_table(md);
1161 
1162 	/* This does not get reverted if there's an error later. */
1163 	dm_table_presuspend_targets(map);
1164 
1165 	md->suspended_bdev = bdget_disk(md->disk, 0);
1166 	if (!md->suspended_bdev) {
1167 		DMWARN("bdget failed in dm_suspend");
1168 		r = -ENOMEM;
1169 		goto out;
1170 	}
1171 
1172 	/* Flush I/O to the device. */
1173 	if (do_lockfs) {
1174 		r = lock_fs(md);
1175 		if (r)
1176 			goto out;
1177 	}
1178 
1179 	/*
1180 	 * First we set the BLOCK_IO flag so no more ios will be mapped.
1181 	 */
1182 	down_write(&md->io_lock);
1183 	set_bit(DMF_BLOCK_IO, &md->flags);
1184 
1185 	add_wait_queue(&md->wait, &wait);
1186 	up_write(&md->io_lock);
1187 
1188 	/* unplug */
1189 	if (map)
1190 		dm_table_unplug_all(map);
1191 
1192 	/*
1193 	 * Then we wait for the already mapped ios to
1194 	 * complete.
1195 	 */
1196 	while (1) {
1197 		set_current_state(TASK_INTERRUPTIBLE);
1198 
1199 		if (!atomic_read(&md->pending) || signal_pending(current))
1200 			break;
1201 
1202 		io_schedule();
1203 	}
1204 	set_current_state(TASK_RUNNING);
1205 
1206 	down_write(&md->io_lock);
1207 	remove_wait_queue(&md->wait, &wait);
1208 
1209 	/* were we interrupted ? */
1210 	r = -EINTR;
1211 	if (atomic_read(&md->pending)) {
1212 		clear_bit(DMF_BLOCK_IO, &md->flags);
1213 		def = bio_list_get(&md->deferred);
1214 		__flush_deferred_io(md, def);
1215 		up_write(&md->io_lock);
1216 		unlock_fs(md);
1217 		goto out;
1218 	}
1219 	up_write(&md->io_lock);
1220 
1221 	dm_table_postsuspend_targets(map);
1222 
1223 	set_bit(DMF_SUSPENDED, &md->flags);
1224 
1225 	r = 0;
1226 
1227 out:
1228 	if (r && md->suspended_bdev) {
1229 		bdput(md->suspended_bdev);
1230 		md->suspended_bdev = NULL;
1231 	}
1232 
1233 	dm_table_put(map);
1234 	up(&md->suspend_lock);
1235 	return r;
1236 }
1237 
1238 int dm_resume(struct mapped_device *md)
1239 {
1240 	int r = -EINVAL;
1241 	struct bio *def;
1242 	struct dm_table *map = NULL;
1243 
1244 	down(&md->suspend_lock);
1245 	if (!dm_suspended(md))
1246 		goto out;
1247 
1248 	map = dm_get_table(md);
1249 	if (!map || !dm_table_get_size(map))
1250 		goto out;
1251 
1252 	dm_table_resume_targets(map);
1253 
1254 	down_write(&md->io_lock);
1255 	clear_bit(DMF_BLOCK_IO, &md->flags);
1256 
1257 	def = bio_list_get(&md->deferred);
1258 	__flush_deferred_io(md, def);
1259 	up_write(&md->io_lock);
1260 
1261 	unlock_fs(md);
1262 
1263 	bdput(md->suspended_bdev);
1264 	md->suspended_bdev = NULL;
1265 
1266 	clear_bit(DMF_SUSPENDED, &md->flags);
1267 
1268 	dm_table_unplug_all(map);
1269 
1270 	r = 0;
1271 
1272 out:
1273 	dm_table_put(map);
1274 	up(&md->suspend_lock);
1275 
1276 	return r;
1277 }
1278 
1279 /*-----------------------------------------------------------------
1280  * Event notification.
1281  *---------------------------------------------------------------*/
1282 uint32_t dm_get_event_nr(struct mapped_device *md)
1283 {
1284 	return atomic_read(&md->event_nr);
1285 }
1286 
1287 int dm_wait_event(struct mapped_device *md, int event_nr)
1288 {
1289 	return wait_event_interruptible(md->eventq,
1290 			(event_nr != atomic_read(&md->event_nr)));
1291 }
1292 
1293 /*
1294  * The gendisk is only valid as long as you have a reference
1295  * count on 'md'.
1296  */
1297 struct gendisk *dm_disk(struct mapped_device *md)
1298 {
1299 	return md->disk;
1300 }
1301 
1302 int dm_suspended(struct mapped_device *md)
1303 {
1304 	return test_bit(DMF_SUSPENDED, &md->flags);
1305 }
1306 
1307 static struct block_device_operations dm_blk_dops = {
1308 	.open = dm_blk_open,
1309 	.release = dm_blk_close,
1310 	.getgeo = dm_blk_getgeo,
1311 	.owner = THIS_MODULE
1312 };
1313 
1314 EXPORT_SYMBOL(dm_get_mapinfo);
1315 
1316 /*
1317  * module hooks
1318  */
1319 module_init(dm_init);
1320 module_exit(dm_exit);
1321 
1322 module_param(major, uint, 0);
1323 MODULE_PARM_DESC(major, "The major number of the device mapper");
1324 MODULE_DESCRIPTION(DM_NAME " driver");
1325 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
1326 MODULE_LICENSE("GPL");
1327