xref: /linux/drivers/md/dm.c (revision 93d546399c2b7d66a54d5fbd5eee17de19246bf6)
1 /*
2  * Copyright (C) 2001, 2002 Sistina Software (UK) Limited.
3  * Copyright (C) 2004-2006 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 #include "dm-uevent.h"
11 
12 #include <linux/init.h>
13 #include <linux/module.h>
14 #include <linux/mutex.h>
15 #include <linux/moduleparam.h>
16 #include <linux/blkpg.h>
17 #include <linux/bio.h>
18 #include <linux/buffer_head.h>
19 #include <linux/mempool.h>
20 #include <linux/slab.h>
21 #include <linux/idr.h>
22 #include <linux/hdreg.h>
23 #include <linux/blktrace_api.h>
24 
25 #define DM_MSG_PREFIX "core"
26 
27 static const char *_name = DM_NAME;
28 
29 static unsigned int major = 0;
30 static unsigned int _major = 0;
31 
32 static DEFINE_SPINLOCK(_minor_lock);
33 /*
34  * One of these is allocated per bio.
35  */
36 struct dm_io {
37 	struct mapped_device *md;
38 	int error;
39 	atomic_t io_count;
40 	struct bio *bio;
41 	unsigned long start_time;
42 };
43 
44 /*
45  * One of these is allocated per target within a bio.  Hopefully
46  * this will be simplified out one day.
47  */
48 struct dm_target_io {
49 	struct dm_io *io;
50 	struct dm_target *ti;
51 	union map_info info;
52 };
53 
54 union map_info *dm_get_mapinfo(struct bio *bio)
55 {
56 	if (bio && bio->bi_private)
57 		return &((struct dm_target_io *)bio->bi_private)->info;
58 	return NULL;
59 }
60 
61 #define MINOR_ALLOCED ((void *)-1)
62 
63 /*
64  * Bits for the md->flags field.
65  */
66 #define DMF_BLOCK_IO 0
67 #define DMF_SUSPENDED 1
68 #define DMF_FROZEN 2
69 #define DMF_FREEING 3
70 #define DMF_DELETING 4
71 #define DMF_NOFLUSH_SUSPENDING 5
72 
73 /*
74  * Work processed by per-device workqueue.
75  */
76 struct dm_wq_req {
77 	enum {
78 		DM_WQ_FLUSH_DEFERRED,
79 	} type;
80 	struct work_struct work;
81 	struct mapped_device *md;
82 	void *context;
83 };
84 
85 struct mapped_device {
86 	struct rw_semaphore io_lock;
87 	struct mutex suspend_lock;
88 	spinlock_t pushback_lock;
89 	rwlock_t map_lock;
90 	atomic_t holders;
91 	atomic_t open_count;
92 
93 	unsigned long flags;
94 
95 	struct request_queue *queue;
96 	struct gendisk *disk;
97 	char name[16];
98 
99 	void *interface_ptr;
100 
101 	/*
102 	 * A list of ios that arrived while we were suspended.
103 	 */
104 	atomic_t pending;
105 	wait_queue_head_t wait;
106 	struct bio_list deferred;
107 	struct bio_list pushback;
108 
109 	/*
110 	 * Processing queue (flush/barriers)
111 	 */
112 	struct workqueue_struct *wq;
113 
114 	/*
115 	 * The current mapping.
116 	 */
117 	struct dm_table *map;
118 
119 	/*
120 	 * io objects are allocated from here.
121 	 */
122 	mempool_t *io_pool;
123 	mempool_t *tio_pool;
124 
125 	struct bio_set *bs;
126 
127 	/*
128 	 * Event handling.
129 	 */
130 	atomic_t event_nr;
131 	wait_queue_head_t eventq;
132 	atomic_t uevent_seq;
133 	struct list_head uevent_list;
134 	spinlock_t uevent_lock; /* Protect access to uevent_list */
135 
136 	/*
137 	 * freeze/thaw support require holding onto a super block
138 	 */
139 	struct super_block *frozen_sb;
140 	struct block_device *suspended_bdev;
141 
142 	/* forced geometry settings */
143 	struct hd_geometry geometry;
144 };
145 
146 #define MIN_IOS 256
147 static struct kmem_cache *_io_cache;
148 static struct kmem_cache *_tio_cache;
149 
150 static int __init local_init(void)
151 {
152 	int r = -ENOMEM;
153 
154 	/* allocate a slab for the dm_ios */
155 	_io_cache = KMEM_CACHE(dm_io, 0);
156 	if (!_io_cache)
157 		return r;
158 
159 	/* allocate a slab for the target ios */
160 	_tio_cache = KMEM_CACHE(dm_target_io, 0);
161 	if (!_tio_cache)
162 		goto out_free_io_cache;
163 
164 	r = dm_uevent_init();
165 	if (r)
166 		goto out_free_tio_cache;
167 
168 	_major = major;
169 	r = register_blkdev(_major, _name);
170 	if (r < 0)
171 		goto out_uevent_exit;
172 
173 	if (!_major)
174 		_major = r;
175 
176 	return 0;
177 
178 out_uevent_exit:
179 	dm_uevent_exit();
180 out_free_tio_cache:
181 	kmem_cache_destroy(_tio_cache);
182 out_free_io_cache:
183 	kmem_cache_destroy(_io_cache);
184 
185 	return r;
186 }
187 
188 static void local_exit(void)
189 {
190 	kmem_cache_destroy(_tio_cache);
191 	kmem_cache_destroy(_io_cache);
192 	unregister_blkdev(_major, _name);
193 	dm_uevent_exit();
194 
195 	_major = 0;
196 
197 	DMINFO("cleaned up");
198 }
199 
200 static int (*_inits[])(void) __initdata = {
201 	local_init,
202 	dm_target_init,
203 	dm_linear_init,
204 	dm_stripe_init,
205 	dm_kcopyd_init,
206 	dm_interface_init,
207 };
208 
209 static void (*_exits[])(void) = {
210 	local_exit,
211 	dm_target_exit,
212 	dm_linear_exit,
213 	dm_stripe_exit,
214 	dm_kcopyd_exit,
215 	dm_interface_exit,
216 };
217 
218 static int __init dm_init(void)
219 {
220 	const int count = ARRAY_SIZE(_inits);
221 
222 	int r, i;
223 
224 	for (i = 0; i < count; i++) {
225 		r = _inits[i]();
226 		if (r)
227 			goto bad;
228 	}
229 
230 	return 0;
231 
232       bad:
233 	while (i--)
234 		_exits[i]();
235 
236 	return r;
237 }
238 
239 static void __exit dm_exit(void)
240 {
241 	int i = ARRAY_SIZE(_exits);
242 
243 	while (i--)
244 		_exits[i]();
245 }
246 
247 /*
248  * Block device functions
249  */
250 static int dm_blk_open(struct block_device *bdev, fmode_t mode)
251 {
252 	struct mapped_device *md;
253 
254 	spin_lock(&_minor_lock);
255 
256 	md = bdev->bd_disk->private_data;
257 	if (!md)
258 		goto out;
259 
260 	if (test_bit(DMF_FREEING, &md->flags) ||
261 	    test_bit(DMF_DELETING, &md->flags)) {
262 		md = NULL;
263 		goto out;
264 	}
265 
266 	dm_get(md);
267 	atomic_inc(&md->open_count);
268 
269 out:
270 	spin_unlock(&_minor_lock);
271 
272 	return md ? 0 : -ENXIO;
273 }
274 
275 static int dm_blk_close(struct gendisk *disk, fmode_t mode)
276 {
277 	struct mapped_device *md = disk->private_data;
278 	atomic_dec(&md->open_count);
279 	dm_put(md);
280 	return 0;
281 }
282 
283 int dm_open_count(struct mapped_device *md)
284 {
285 	return atomic_read(&md->open_count);
286 }
287 
288 /*
289  * Guarantees nothing is using the device before it's deleted.
290  */
291 int dm_lock_for_deletion(struct mapped_device *md)
292 {
293 	int r = 0;
294 
295 	spin_lock(&_minor_lock);
296 
297 	if (dm_open_count(md))
298 		r = -EBUSY;
299 	else
300 		set_bit(DMF_DELETING, &md->flags);
301 
302 	spin_unlock(&_minor_lock);
303 
304 	return r;
305 }
306 
307 static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo)
308 {
309 	struct mapped_device *md = bdev->bd_disk->private_data;
310 
311 	return dm_get_geometry(md, geo);
312 }
313 
314 static int dm_blk_ioctl(struct block_device *bdev, fmode_t mode,
315 			unsigned int cmd, unsigned long arg)
316 {
317 	struct mapped_device *md = bdev->bd_disk->private_data;
318 	struct dm_table *map = dm_get_table(md);
319 	struct dm_target *tgt;
320 	int r = -ENOTTY;
321 
322 	if (!map || !dm_table_get_size(map))
323 		goto out;
324 
325 	/* We only support devices that have a single target */
326 	if (dm_table_get_num_targets(map) != 1)
327 		goto out;
328 
329 	tgt = dm_table_get_target(map, 0);
330 
331 	if (dm_suspended(md)) {
332 		r = -EAGAIN;
333 		goto out;
334 	}
335 
336 	if (tgt->type->ioctl)
337 		r = tgt->type->ioctl(tgt, cmd, arg);
338 
339 out:
340 	dm_table_put(map);
341 
342 	return r;
343 }
344 
345 static struct dm_io *alloc_io(struct mapped_device *md)
346 {
347 	return mempool_alloc(md->io_pool, GFP_NOIO);
348 }
349 
350 static void free_io(struct mapped_device *md, struct dm_io *io)
351 {
352 	mempool_free(io, md->io_pool);
353 }
354 
355 static struct dm_target_io *alloc_tio(struct mapped_device *md)
356 {
357 	return mempool_alloc(md->tio_pool, GFP_NOIO);
358 }
359 
360 static void free_tio(struct mapped_device *md, struct dm_target_io *tio)
361 {
362 	mempool_free(tio, md->tio_pool);
363 }
364 
365 static void start_io_acct(struct dm_io *io)
366 {
367 	struct mapped_device *md = io->md;
368 	int cpu;
369 
370 	io->start_time = jiffies;
371 
372 	cpu = part_stat_lock();
373 	part_round_stats(cpu, &dm_disk(md)->part0);
374 	part_stat_unlock();
375 	dm_disk(md)->part0.in_flight = atomic_inc_return(&md->pending);
376 }
377 
378 static void end_io_acct(struct dm_io *io)
379 {
380 	struct mapped_device *md = io->md;
381 	struct bio *bio = io->bio;
382 	unsigned long duration = jiffies - io->start_time;
383 	int pending, cpu;
384 	int rw = bio_data_dir(bio);
385 
386 	cpu = part_stat_lock();
387 	part_round_stats(cpu, &dm_disk(md)->part0);
388 	part_stat_add(cpu, &dm_disk(md)->part0, ticks[rw], duration);
389 	part_stat_unlock();
390 
391 	dm_disk(md)->part0.in_flight = pending =
392 		atomic_dec_return(&md->pending);
393 
394 	/* nudge anyone waiting on suspend queue */
395 	if (!pending)
396 		wake_up(&md->wait);
397 }
398 
399 /*
400  * Add the bio to the list of deferred io.
401  */
402 static int queue_io(struct mapped_device *md, struct bio *bio)
403 {
404 	down_write(&md->io_lock);
405 
406 	if (!test_bit(DMF_BLOCK_IO, &md->flags)) {
407 		up_write(&md->io_lock);
408 		return 1;
409 	}
410 
411 	bio_list_add(&md->deferred, bio);
412 
413 	up_write(&md->io_lock);
414 	return 0;		/* deferred successfully */
415 }
416 
417 /*
418  * Everyone (including functions in this file), should use this
419  * function to access the md->map field, and make sure they call
420  * dm_table_put() when finished.
421  */
422 struct dm_table *dm_get_table(struct mapped_device *md)
423 {
424 	struct dm_table *t;
425 
426 	read_lock(&md->map_lock);
427 	t = md->map;
428 	if (t)
429 		dm_table_get(t);
430 	read_unlock(&md->map_lock);
431 
432 	return t;
433 }
434 
435 /*
436  * Get the geometry associated with a dm device
437  */
438 int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo)
439 {
440 	*geo = md->geometry;
441 
442 	return 0;
443 }
444 
445 /*
446  * Set the geometry of a device.
447  */
448 int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo)
449 {
450 	sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors;
451 
452 	if (geo->start > sz) {
453 		DMWARN("Start sector is beyond the geometry limits.");
454 		return -EINVAL;
455 	}
456 
457 	md->geometry = *geo;
458 
459 	return 0;
460 }
461 
462 /*-----------------------------------------------------------------
463  * CRUD START:
464  *   A more elegant soln is in the works that uses the queue
465  *   merge fn, unfortunately there are a couple of changes to
466  *   the block layer that I want to make for this.  So in the
467  *   interests of getting something for people to use I give
468  *   you this clearly demarcated crap.
469  *---------------------------------------------------------------*/
470 
471 static int __noflush_suspending(struct mapped_device *md)
472 {
473 	return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
474 }
475 
476 /*
477  * Decrements the number of outstanding ios that a bio has been
478  * cloned into, completing the original io if necc.
479  */
480 static void dec_pending(struct dm_io *io, int error)
481 {
482 	unsigned long flags;
483 
484 	/* Push-back supersedes any I/O errors */
485 	if (error && !(io->error > 0 && __noflush_suspending(io->md)))
486 		io->error = error;
487 
488 	if (atomic_dec_and_test(&io->io_count)) {
489 		if (io->error == DM_ENDIO_REQUEUE) {
490 			/*
491 			 * Target requested pushing back the I/O.
492 			 * This must be handled before the sleeper on
493 			 * suspend queue merges the pushback list.
494 			 */
495 			spin_lock_irqsave(&io->md->pushback_lock, flags);
496 			if (__noflush_suspending(io->md))
497 				bio_list_add(&io->md->pushback, io->bio);
498 			else
499 				/* noflush suspend was interrupted. */
500 				io->error = -EIO;
501 			spin_unlock_irqrestore(&io->md->pushback_lock, flags);
502 		}
503 
504 		end_io_acct(io);
505 
506 		if (io->error != DM_ENDIO_REQUEUE) {
507 			blk_add_trace_bio(io->md->queue, io->bio,
508 					  BLK_TA_COMPLETE);
509 
510 			bio_endio(io->bio, io->error);
511 		}
512 
513 		free_io(io->md, io);
514 	}
515 }
516 
517 static void clone_endio(struct bio *bio, int error)
518 {
519 	int r = 0;
520 	struct dm_target_io *tio = bio->bi_private;
521 	struct mapped_device *md = tio->io->md;
522 	dm_endio_fn endio = tio->ti->type->end_io;
523 
524 	if (!bio_flagged(bio, BIO_UPTODATE) && !error)
525 		error = -EIO;
526 
527 	if (endio) {
528 		r = endio(tio->ti, bio, error, &tio->info);
529 		if (r < 0 || r == DM_ENDIO_REQUEUE)
530 			/*
531 			 * error and requeue request are handled
532 			 * in dec_pending().
533 			 */
534 			error = r;
535 		else if (r == DM_ENDIO_INCOMPLETE)
536 			/* The target will handle the io */
537 			return;
538 		else if (r) {
539 			DMWARN("unimplemented target endio return value: %d", r);
540 			BUG();
541 		}
542 	}
543 
544 	dec_pending(tio->io, error);
545 
546 	/*
547 	 * Store md for cleanup instead of tio which is about to get freed.
548 	 */
549 	bio->bi_private = md->bs;
550 
551 	bio_put(bio);
552 	free_tio(md, tio);
553 }
554 
555 static sector_t max_io_len(struct mapped_device *md,
556 			   sector_t sector, struct dm_target *ti)
557 {
558 	sector_t offset = sector - ti->begin;
559 	sector_t len = ti->len - offset;
560 
561 	/*
562 	 * Does the target need to split even further ?
563 	 */
564 	if (ti->split_io) {
565 		sector_t boundary;
566 		boundary = ((offset + ti->split_io) & ~(ti->split_io - 1))
567 			   - offset;
568 		if (len > boundary)
569 			len = boundary;
570 	}
571 
572 	return len;
573 }
574 
575 static void __map_bio(struct dm_target *ti, struct bio *clone,
576 		      struct dm_target_io *tio)
577 {
578 	int r;
579 	sector_t sector;
580 	struct mapped_device *md;
581 
582 	/*
583 	 * Sanity checks.
584 	 */
585 	BUG_ON(!clone->bi_size);
586 
587 	clone->bi_end_io = clone_endio;
588 	clone->bi_private = tio;
589 
590 	/*
591 	 * Map the clone.  If r == 0 we don't need to do
592 	 * anything, the target has assumed ownership of
593 	 * this io.
594 	 */
595 	atomic_inc(&tio->io->io_count);
596 	sector = clone->bi_sector;
597 	r = ti->type->map(ti, clone, &tio->info);
598 	if (r == DM_MAPIO_REMAPPED) {
599 		/* the bio has been remapped so dispatch it */
600 
601 		blk_add_trace_remap(bdev_get_queue(clone->bi_bdev), clone,
602 				    tio->io->bio->bi_bdev->bd_dev,
603 				    clone->bi_sector, sector);
604 
605 		generic_make_request(clone);
606 	} else if (r < 0 || r == DM_MAPIO_REQUEUE) {
607 		/* error the io and bail out, or requeue it if needed */
608 		md = tio->io->md;
609 		dec_pending(tio->io, r);
610 		/*
611 		 * Store bio_set for cleanup.
612 		 */
613 		clone->bi_private = md->bs;
614 		bio_put(clone);
615 		free_tio(md, tio);
616 	} else if (r) {
617 		DMWARN("unimplemented target map return value: %d", r);
618 		BUG();
619 	}
620 }
621 
622 struct clone_info {
623 	struct mapped_device *md;
624 	struct dm_table *map;
625 	struct bio *bio;
626 	struct dm_io *io;
627 	sector_t sector;
628 	sector_t sector_count;
629 	unsigned short idx;
630 };
631 
632 static void dm_bio_destructor(struct bio *bio)
633 {
634 	struct bio_set *bs = bio->bi_private;
635 
636 	bio_free(bio, bs);
637 }
638 
639 /*
640  * Creates a little bio that is just does part of a bvec.
641  */
642 static struct bio *split_bvec(struct bio *bio, sector_t sector,
643 			      unsigned short idx, unsigned int offset,
644 			      unsigned int len, struct bio_set *bs)
645 {
646 	struct bio *clone;
647 	struct bio_vec *bv = bio->bi_io_vec + idx;
648 
649 	clone = bio_alloc_bioset(GFP_NOIO, 1, bs);
650 	clone->bi_destructor = dm_bio_destructor;
651 	*clone->bi_io_vec = *bv;
652 
653 	clone->bi_sector = sector;
654 	clone->bi_bdev = bio->bi_bdev;
655 	clone->bi_rw = bio->bi_rw;
656 	clone->bi_vcnt = 1;
657 	clone->bi_size = to_bytes(len);
658 	clone->bi_io_vec->bv_offset = offset;
659 	clone->bi_io_vec->bv_len = clone->bi_size;
660 	clone->bi_flags |= 1 << BIO_CLONED;
661 
662 	return clone;
663 }
664 
665 /*
666  * Creates a bio that consists of range of complete bvecs.
667  */
668 static struct bio *clone_bio(struct bio *bio, sector_t sector,
669 			     unsigned short idx, unsigned short bv_count,
670 			     unsigned int len, struct bio_set *bs)
671 {
672 	struct bio *clone;
673 
674 	clone = bio_alloc_bioset(GFP_NOIO, bio->bi_max_vecs, bs);
675 	__bio_clone(clone, bio);
676 	clone->bi_destructor = dm_bio_destructor;
677 	clone->bi_sector = sector;
678 	clone->bi_idx = idx;
679 	clone->bi_vcnt = idx + bv_count;
680 	clone->bi_size = to_bytes(len);
681 	clone->bi_flags &= ~(1 << BIO_SEG_VALID);
682 
683 	return clone;
684 }
685 
686 static int __clone_and_map(struct clone_info *ci)
687 {
688 	struct bio *clone, *bio = ci->bio;
689 	struct dm_target *ti;
690 	sector_t len = 0, max;
691 	struct dm_target_io *tio;
692 
693 	ti = dm_table_find_target(ci->map, ci->sector);
694 	if (!dm_target_is_valid(ti))
695 		return -EIO;
696 
697 	max = max_io_len(ci->md, ci->sector, ti);
698 
699 	/*
700 	 * Allocate a target io object.
701 	 */
702 	tio = alloc_tio(ci->md);
703 	tio->io = ci->io;
704 	tio->ti = ti;
705 	memset(&tio->info, 0, sizeof(tio->info));
706 
707 	if (ci->sector_count <= max) {
708 		/*
709 		 * Optimise for the simple case where we can do all of
710 		 * the remaining io with a single clone.
711 		 */
712 		clone = clone_bio(bio, ci->sector, ci->idx,
713 				  bio->bi_vcnt - ci->idx, ci->sector_count,
714 				  ci->md->bs);
715 		__map_bio(ti, clone, tio);
716 		ci->sector_count = 0;
717 
718 	} else if (to_sector(bio->bi_io_vec[ci->idx].bv_len) <= max) {
719 		/*
720 		 * There are some bvecs that don't span targets.
721 		 * Do as many of these as possible.
722 		 */
723 		int i;
724 		sector_t remaining = max;
725 		sector_t bv_len;
726 
727 		for (i = ci->idx; remaining && (i < bio->bi_vcnt); i++) {
728 			bv_len = to_sector(bio->bi_io_vec[i].bv_len);
729 
730 			if (bv_len > remaining)
731 				break;
732 
733 			remaining -= bv_len;
734 			len += bv_len;
735 		}
736 
737 		clone = clone_bio(bio, ci->sector, ci->idx, i - ci->idx, len,
738 				  ci->md->bs);
739 		__map_bio(ti, clone, tio);
740 
741 		ci->sector += len;
742 		ci->sector_count -= len;
743 		ci->idx = i;
744 
745 	} else {
746 		/*
747 		 * Handle a bvec that must be split between two or more targets.
748 		 */
749 		struct bio_vec *bv = bio->bi_io_vec + ci->idx;
750 		sector_t remaining = to_sector(bv->bv_len);
751 		unsigned int offset = 0;
752 
753 		do {
754 			if (offset) {
755 				ti = dm_table_find_target(ci->map, ci->sector);
756 				if (!dm_target_is_valid(ti))
757 					return -EIO;
758 
759 				max = max_io_len(ci->md, ci->sector, ti);
760 
761 				tio = alloc_tio(ci->md);
762 				tio->io = ci->io;
763 				tio->ti = ti;
764 				memset(&tio->info, 0, sizeof(tio->info));
765 			}
766 
767 			len = min(remaining, max);
768 
769 			clone = split_bvec(bio, ci->sector, ci->idx,
770 					   bv->bv_offset + offset, len,
771 					   ci->md->bs);
772 
773 			__map_bio(ti, clone, tio);
774 
775 			ci->sector += len;
776 			ci->sector_count -= len;
777 			offset += to_bytes(len);
778 		} while (remaining -= len);
779 
780 		ci->idx++;
781 	}
782 
783 	return 0;
784 }
785 
786 /*
787  * Split the bio into several clones.
788  */
789 static int __split_bio(struct mapped_device *md, struct bio *bio)
790 {
791 	struct clone_info ci;
792 	int error = 0;
793 
794 	ci.map = dm_get_table(md);
795 	if (unlikely(!ci.map))
796 		return -EIO;
797 
798 	ci.md = md;
799 	ci.bio = bio;
800 	ci.io = alloc_io(md);
801 	ci.io->error = 0;
802 	atomic_set(&ci.io->io_count, 1);
803 	ci.io->bio = bio;
804 	ci.io->md = md;
805 	ci.sector = bio->bi_sector;
806 	ci.sector_count = bio_sectors(bio);
807 	ci.idx = bio->bi_idx;
808 
809 	start_io_acct(ci.io);
810 	while (ci.sector_count && !error)
811 		error = __clone_and_map(&ci);
812 
813 	/* drop the extra reference count */
814 	dec_pending(ci.io, error);
815 	dm_table_put(ci.map);
816 
817 	return 0;
818 }
819 /*-----------------------------------------------------------------
820  * CRUD END
821  *---------------------------------------------------------------*/
822 
823 static int dm_merge_bvec(struct request_queue *q,
824 			 struct bvec_merge_data *bvm,
825 			 struct bio_vec *biovec)
826 {
827 	struct mapped_device *md = q->queuedata;
828 	struct dm_table *map = dm_get_table(md);
829 	struct dm_target *ti;
830 	sector_t max_sectors;
831 	int max_size = 0;
832 
833 	if (unlikely(!map))
834 		goto out;
835 
836 	ti = dm_table_find_target(map, bvm->bi_sector);
837 	if (!dm_target_is_valid(ti))
838 		goto out_table;
839 
840 	/*
841 	 * Find maximum amount of I/O that won't need splitting
842 	 */
843 	max_sectors = min(max_io_len(md, bvm->bi_sector, ti),
844 			  (sector_t) BIO_MAX_SECTORS);
845 	max_size = (max_sectors << SECTOR_SHIFT) - bvm->bi_size;
846 	if (max_size < 0)
847 		max_size = 0;
848 
849 	/*
850 	 * merge_bvec_fn() returns number of bytes
851 	 * it can accept at this offset
852 	 * max is precomputed maximal io size
853 	 */
854 	if (max_size && ti->type->merge)
855 		max_size = ti->type->merge(ti, bvm, biovec, max_size);
856 
857 out_table:
858 	dm_table_put(map);
859 
860 out:
861 	/*
862 	 * Always allow an entire first page
863 	 */
864 	if (max_size <= biovec->bv_len && !(bvm->bi_size >> SECTOR_SHIFT))
865 		max_size = biovec->bv_len;
866 
867 	return max_size;
868 }
869 
870 /*
871  * The request function that just remaps the bio built up by
872  * dm_merge_bvec.
873  */
874 static int dm_request(struct request_queue *q, struct bio *bio)
875 {
876 	int r = -EIO;
877 	int rw = bio_data_dir(bio);
878 	struct mapped_device *md = q->queuedata;
879 	int cpu;
880 
881 	/*
882 	 * There is no use in forwarding any barrier request since we can't
883 	 * guarantee it is (or can be) handled by the targets correctly.
884 	 */
885 	if (unlikely(bio_barrier(bio))) {
886 		bio_endio(bio, -EOPNOTSUPP);
887 		return 0;
888 	}
889 
890 	down_read(&md->io_lock);
891 
892 	cpu = part_stat_lock();
893 	part_stat_inc(cpu, &dm_disk(md)->part0, ios[rw]);
894 	part_stat_add(cpu, &dm_disk(md)->part0, sectors[rw], bio_sectors(bio));
895 	part_stat_unlock();
896 
897 	/*
898 	 * If we're suspended we have to queue
899 	 * this io for later.
900 	 */
901 	while (test_bit(DMF_BLOCK_IO, &md->flags)) {
902 		up_read(&md->io_lock);
903 
904 		if (bio_rw(bio) != READA)
905 			r = queue_io(md, bio);
906 
907 		if (r <= 0)
908 			goto out_req;
909 
910 		/*
911 		 * We're in a while loop, because someone could suspend
912 		 * before we get to the following read lock.
913 		 */
914 		down_read(&md->io_lock);
915 	}
916 
917 	r = __split_bio(md, bio);
918 	up_read(&md->io_lock);
919 
920 out_req:
921 	if (r < 0)
922 		bio_io_error(bio);
923 
924 	return 0;
925 }
926 
927 static void dm_unplug_all(struct request_queue *q)
928 {
929 	struct mapped_device *md = q->queuedata;
930 	struct dm_table *map = dm_get_table(md);
931 
932 	if (map) {
933 		dm_table_unplug_all(map);
934 		dm_table_put(map);
935 	}
936 }
937 
938 static int dm_any_congested(void *congested_data, int bdi_bits)
939 {
940 	int r = bdi_bits;
941 	struct mapped_device *md = congested_data;
942 	struct dm_table *map;
943 
944 	atomic_inc(&md->pending);
945 
946 	if (!test_bit(DMF_BLOCK_IO, &md->flags)) {
947 		map = dm_get_table(md);
948 		if (map) {
949 			r = dm_table_any_congested(map, bdi_bits);
950 			dm_table_put(map);
951 		}
952 	}
953 
954 	if (!atomic_dec_return(&md->pending))
955 		/* nudge anyone waiting on suspend queue */
956 		wake_up(&md->wait);
957 
958 	return r;
959 }
960 
961 /*-----------------------------------------------------------------
962  * An IDR is used to keep track of allocated minor numbers.
963  *---------------------------------------------------------------*/
964 static DEFINE_IDR(_minor_idr);
965 
966 static void free_minor(int minor)
967 {
968 	spin_lock(&_minor_lock);
969 	idr_remove(&_minor_idr, minor);
970 	spin_unlock(&_minor_lock);
971 }
972 
973 /*
974  * See if the device with a specific minor # is free.
975  */
976 static int specific_minor(int minor)
977 {
978 	int r, m;
979 
980 	if (minor >= (1 << MINORBITS))
981 		return -EINVAL;
982 
983 	r = idr_pre_get(&_minor_idr, GFP_KERNEL);
984 	if (!r)
985 		return -ENOMEM;
986 
987 	spin_lock(&_minor_lock);
988 
989 	if (idr_find(&_minor_idr, minor)) {
990 		r = -EBUSY;
991 		goto out;
992 	}
993 
994 	r = idr_get_new_above(&_minor_idr, MINOR_ALLOCED, minor, &m);
995 	if (r)
996 		goto out;
997 
998 	if (m != minor) {
999 		idr_remove(&_minor_idr, m);
1000 		r = -EBUSY;
1001 		goto out;
1002 	}
1003 
1004 out:
1005 	spin_unlock(&_minor_lock);
1006 	return r;
1007 }
1008 
1009 static int next_free_minor(int *minor)
1010 {
1011 	int r, m;
1012 
1013 	r = idr_pre_get(&_minor_idr, GFP_KERNEL);
1014 	if (!r)
1015 		return -ENOMEM;
1016 
1017 	spin_lock(&_minor_lock);
1018 
1019 	r = idr_get_new(&_minor_idr, MINOR_ALLOCED, &m);
1020 	if (r)
1021 		goto out;
1022 
1023 	if (m >= (1 << MINORBITS)) {
1024 		idr_remove(&_minor_idr, m);
1025 		r = -ENOSPC;
1026 		goto out;
1027 	}
1028 
1029 	*minor = m;
1030 
1031 out:
1032 	spin_unlock(&_minor_lock);
1033 	return r;
1034 }
1035 
1036 static struct block_device_operations dm_blk_dops;
1037 
1038 /*
1039  * Allocate and initialise a blank device with a given minor.
1040  */
1041 static struct mapped_device *alloc_dev(int minor)
1042 {
1043 	int r;
1044 	struct mapped_device *md = kzalloc(sizeof(*md), GFP_KERNEL);
1045 	void *old_md;
1046 
1047 	if (!md) {
1048 		DMWARN("unable to allocate device, out of memory.");
1049 		return NULL;
1050 	}
1051 
1052 	if (!try_module_get(THIS_MODULE))
1053 		goto bad_module_get;
1054 
1055 	/* get a minor number for the dev */
1056 	if (minor == DM_ANY_MINOR)
1057 		r = next_free_minor(&minor);
1058 	else
1059 		r = specific_minor(minor);
1060 	if (r < 0)
1061 		goto bad_minor;
1062 
1063 	init_rwsem(&md->io_lock);
1064 	mutex_init(&md->suspend_lock);
1065 	spin_lock_init(&md->pushback_lock);
1066 	rwlock_init(&md->map_lock);
1067 	atomic_set(&md->holders, 1);
1068 	atomic_set(&md->open_count, 0);
1069 	atomic_set(&md->event_nr, 0);
1070 	atomic_set(&md->uevent_seq, 0);
1071 	INIT_LIST_HEAD(&md->uevent_list);
1072 	spin_lock_init(&md->uevent_lock);
1073 
1074 	md->queue = blk_alloc_queue(GFP_KERNEL);
1075 	if (!md->queue)
1076 		goto bad_queue;
1077 
1078 	md->queue->queuedata = md;
1079 	md->queue->backing_dev_info.congested_fn = dm_any_congested;
1080 	md->queue->backing_dev_info.congested_data = md;
1081 	blk_queue_make_request(md->queue, dm_request);
1082 	blk_queue_bounce_limit(md->queue, BLK_BOUNCE_ANY);
1083 	md->queue->unplug_fn = dm_unplug_all;
1084 	blk_queue_merge_bvec(md->queue, dm_merge_bvec);
1085 
1086 	md->io_pool = mempool_create_slab_pool(MIN_IOS, _io_cache);
1087 	if (!md->io_pool)
1088 		goto bad_io_pool;
1089 
1090 	md->tio_pool = mempool_create_slab_pool(MIN_IOS, _tio_cache);
1091 	if (!md->tio_pool)
1092 		goto bad_tio_pool;
1093 
1094 	md->bs = bioset_create(16, 16);
1095 	if (!md->bs)
1096 		goto bad_no_bioset;
1097 
1098 	md->disk = alloc_disk(1);
1099 	if (!md->disk)
1100 		goto bad_disk;
1101 
1102 	atomic_set(&md->pending, 0);
1103 	init_waitqueue_head(&md->wait);
1104 	init_waitqueue_head(&md->eventq);
1105 
1106 	md->disk->major = _major;
1107 	md->disk->first_minor = minor;
1108 	md->disk->fops = &dm_blk_dops;
1109 	md->disk->queue = md->queue;
1110 	md->disk->private_data = md;
1111 	sprintf(md->disk->disk_name, "dm-%d", minor);
1112 	add_disk(md->disk);
1113 	format_dev_t(md->name, MKDEV(_major, minor));
1114 
1115 	md->wq = create_singlethread_workqueue("kdmflush");
1116 	if (!md->wq)
1117 		goto bad_thread;
1118 
1119 	/* Populate the mapping, nobody knows we exist yet */
1120 	spin_lock(&_minor_lock);
1121 	old_md = idr_replace(&_minor_idr, md, minor);
1122 	spin_unlock(&_minor_lock);
1123 
1124 	BUG_ON(old_md != MINOR_ALLOCED);
1125 
1126 	return md;
1127 
1128 bad_thread:
1129 	put_disk(md->disk);
1130 bad_disk:
1131 	bioset_free(md->bs);
1132 bad_no_bioset:
1133 	mempool_destroy(md->tio_pool);
1134 bad_tio_pool:
1135 	mempool_destroy(md->io_pool);
1136 bad_io_pool:
1137 	blk_cleanup_queue(md->queue);
1138 bad_queue:
1139 	free_minor(minor);
1140 bad_minor:
1141 	module_put(THIS_MODULE);
1142 bad_module_get:
1143 	kfree(md);
1144 	return NULL;
1145 }
1146 
1147 static void unlock_fs(struct mapped_device *md);
1148 
1149 static void free_dev(struct mapped_device *md)
1150 {
1151 	int minor = MINOR(disk_devt(md->disk));
1152 
1153 	if (md->suspended_bdev) {
1154 		unlock_fs(md);
1155 		bdput(md->suspended_bdev);
1156 	}
1157 	destroy_workqueue(md->wq);
1158 	mempool_destroy(md->tio_pool);
1159 	mempool_destroy(md->io_pool);
1160 	bioset_free(md->bs);
1161 	del_gendisk(md->disk);
1162 	free_minor(minor);
1163 
1164 	spin_lock(&_minor_lock);
1165 	md->disk->private_data = NULL;
1166 	spin_unlock(&_minor_lock);
1167 
1168 	put_disk(md->disk);
1169 	blk_cleanup_queue(md->queue);
1170 	module_put(THIS_MODULE);
1171 	kfree(md);
1172 }
1173 
1174 /*
1175  * Bind a table to the device.
1176  */
1177 static void event_callback(void *context)
1178 {
1179 	unsigned long flags;
1180 	LIST_HEAD(uevents);
1181 	struct mapped_device *md = (struct mapped_device *) context;
1182 
1183 	spin_lock_irqsave(&md->uevent_lock, flags);
1184 	list_splice_init(&md->uevent_list, &uevents);
1185 	spin_unlock_irqrestore(&md->uevent_lock, flags);
1186 
1187 	dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj);
1188 
1189 	atomic_inc(&md->event_nr);
1190 	wake_up(&md->eventq);
1191 }
1192 
1193 static void __set_size(struct mapped_device *md, sector_t size)
1194 {
1195 	set_capacity(md->disk, size);
1196 
1197 	mutex_lock(&md->suspended_bdev->bd_inode->i_mutex);
1198 	i_size_write(md->suspended_bdev->bd_inode, (loff_t)size << SECTOR_SHIFT);
1199 	mutex_unlock(&md->suspended_bdev->bd_inode->i_mutex);
1200 }
1201 
1202 static int __bind(struct mapped_device *md, struct dm_table *t)
1203 {
1204 	struct request_queue *q = md->queue;
1205 	sector_t size;
1206 
1207 	size = dm_table_get_size(t);
1208 
1209 	/*
1210 	 * Wipe any geometry if the size of the table changed.
1211 	 */
1212 	if (size != get_capacity(md->disk))
1213 		memset(&md->geometry, 0, sizeof(md->geometry));
1214 
1215 	if (md->suspended_bdev)
1216 		__set_size(md, size);
1217 	if (size == 0)
1218 		return 0;
1219 
1220 	dm_table_get(t);
1221 	dm_table_event_callback(t, event_callback, md);
1222 
1223 	write_lock(&md->map_lock);
1224 	md->map = t;
1225 	dm_table_set_restrictions(t, q);
1226 	write_unlock(&md->map_lock);
1227 
1228 	return 0;
1229 }
1230 
1231 static void __unbind(struct mapped_device *md)
1232 {
1233 	struct dm_table *map = md->map;
1234 
1235 	if (!map)
1236 		return;
1237 
1238 	dm_table_event_callback(map, NULL, NULL);
1239 	write_lock(&md->map_lock);
1240 	md->map = NULL;
1241 	write_unlock(&md->map_lock);
1242 	dm_table_put(map);
1243 }
1244 
1245 /*
1246  * Constructor for a new device.
1247  */
1248 int dm_create(int minor, struct mapped_device **result)
1249 {
1250 	struct mapped_device *md;
1251 
1252 	md = alloc_dev(minor);
1253 	if (!md)
1254 		return -ENXIO;
1255 
1256 	*result = md;
1257 	return 0;
1258 }
1259 
1260 static struct mapped_device *dm_find_md(dev_t dev)
1261 {
1262 	struct mapped_device *md;
1263 	unsigned minor = MINOR(dev);
1264 
1265 	if (MAJOR(dev) != _major || minor >= (1 << MINORBITS))
1266 		return NULL;
1267 
1268 	spin_lock(&_minor_lock);
1269 
1270 	md = idr_find(&_minor_idr, minor);
1271 	if (md && (md == MINOR_ALLOCED ||
1272 		   (MINOR(disk_devt(dm_disk(md))) != minor) ||
1273 		   test_bit(DMF_FREEING, &md->flags))) {
1274 		md = NULL;
1275 		goto out;
1276 	}
1277 
1278 out:
1279 	spin_unlock(&_minor_lock);
1280 
1281 	return md;
1282 }
1283 
1284 struct mapped_device *dm_get_md(dev_t dev)
1285 {
1286 	struct mapped_device *md = dm_find_md(dev);
1287 
1288 	if (md)
1289 		dm_get(md);
1290 
1291 	return md;
1292 }
1293 
1294 void *dm_get_mdptr(struct mapped_device *md)
1295 {
1296 	return md->interface_ptr;
1297 }
1298 
1299 void dm_set_mdptr(struct mapped_device *md, void *ptr)
1300 {
1301 	md->interface_ptr = ptr;
1302 }
1303 
1304 void dm_get(struct mapped_device *md)
1305 {
1306 	atomic_inc(&md->holders);
1307 }
1308 
1309 const char *dm_device_name(struct mapped_device *md)
1310 {
1311 	return md->name;
1312 }
1313 EXPORT_SYMBOL_GPL(dm_device_name);
1314 
1315 void dm_put(struct mapped_device *md)
1316 {
1317 	struct dm_table *map;
1318 
1319 	BUG_ON(test_bit(DMF_FREEING, &md->flags));
1320 
1321 	if (atomic_dec_and_lock(&md->holders, &_minor_lock)) {
1322 		map = dm_get_table(md);
1323 		idr_replace(&_minor_idr, MINOR_ALLOCED,
1324 			    MINOR(disk_devt(dm_disk(md))));
1325 		set_bit(DMF_FREEING, &md->flags);
1326 		spin_unlock(&_minor_lock);
1327 		if (!dm_suspended(md)) {
1328 			dm_table_presuspend_targets(map);
1329 			dm_table_postsuspend_targets(map);
1330 		}
1331 		__unbind(md);
1332 		dm_table_put(map);
1333 		free_dev(md);
1334 	}
1335 }
1336 EXPORT_SYMBOL_GPL(dm_put);
1337 
1338 static int dm_wait_for_completion(struct mapped_device *md)
1339 {
1340 	int r = 0;
1341 
1342 	while (1) {
1343 		set_current_state(TASK_INTERRUPTIBLE);
1344 
1345 		smp_mb();
1346 		if (!atomic_read(&md->pending))
1347 			break;
1348 
1349 		if (signal_pending(current)) {
1350 			r = -EINTR;
1351 			break;
1352 		}
1353 
1354 		io_schedule();
1355 	}
1356 	set_current_state(TASK_RUNNING);
1357 
1358 	return r;
1359 }
1360 
1361 /*
1362  * Process the deferred bios
1363  */
1364 static void __flush_deferred_io(struct mapped_device *md)
1365 {
1366 	struct bio *c;
1367 
1368 	while ((c = bio_list_pop(&md->deferred))) {
1369 		if (__split_bio(md, c))
1370 			bio_io_error(c);
1371 	}
1372 
1373 	clear_bit(DMF_BLOCK_IO, &md->flags);
1374 }
1375 
1376 static void __merge_pushback_list(struct mapped_device *md)
1377 {
1378 	unsigned long flags;
1379 
1380 	spin_lock_irqsave(&md->pushback_lock, flags);
1381 	clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
1382 	bio_list_merge_head(&md->deferred, &md->pushback);
1383 	bio_list_init(&md->pushback);
1384 	spin_unlock_irqrestore(&md->pushback_lock, flags);
1385 }
1386 
1387 static void dm_wq_work(struct work_struct *work)
1388 {
1389 	struct dm_wq_req *req = container_of(work, struct dm_wq_req, work);
1390 	struct mapped_device *md = req->md;
1391 
1392 	down_write(&md->io_lock);
1393 	switch (req->type) {
1394 	case DM_WQ_FLUSH_DEFERRED:
1395 		__flush_deferred_io(md);
1396 		break;
1397 	default:
1398 		DMERR("dm_wq_work: unrecognised work type %d", req->type);
1399 		BUG();
1400 	}
1401 	up_write(&md->io_lock);
1402 }
1403 
1404 static void dm_wq_queue(struct mapped_device *md, int type, void *context,
1405 			struct dm_wq_req *req)
1406 {
1407 	req->type = type;
1408 	req->md = md;
1409 	req->context = context;
1410 	INIT_WORK(&req->work, dm_wq_work);
1411 	queue_work(md->wq, &req->work);
1412 }
1413 
1414 static void dm_queue_flush(struct mapped_device *md, int type, void *context)
1415 {
1416 	struct dm_wq_req req;
1417 
1418 	dm_wq_queue(md, type, context, &req);
1419 	flush_workqueue(md->wq);
1420 }
1421 
1422 /*
1423  * Swap in a new table (destroying old one).
1424  */
1425 int dm_swap_table(struct mapped_device *md, struct dm_table *table)
1426 {
1427 	int r = -EINVAL;
1428 
1429 	mutex_lock(&md->suspend_lock);
1430 
1431 	/* device must be suspended */
1432 	if (!dm_suspended(md))
1433 		goto out;
1434 
1435 	/* without bdev, the device size cannot be changed */
1436 	if (!md->suspended_bdev)
1437 		if (get_capacity(md->disk) != dm_table_get_size(table))
1438 			goto out;
1439 
1440 	__unbind(md);
1441 	r = __bind(md, table);
1442 
1443 out:
1444 	mutex_unlock(&md->suspend_lock);
1445 	return r;
1446 }
1447 
1448 /*
1449  * Functions to lock and unlock any filesystem running on the
1450  * device.
1451  */
1452 static int lock_fs(struct mapped_device *md)
1453 {
1454 	int r;
1455 
1456 	WARN_ON(md->frozen_sb);
1457 
1458 	md->frozen_sb = freeze_bdev(md->suspended_bdev);
1459 	if (IS_ERR(md->frozen_sb)) {
1460 		r = PTR_ERR(md->frozen_sb);
1461 		md->frozen_sb = NULL;
1462 		return r;
1463 	}
1464 
1465 	set_bit(DMF_FROZEN, &md->flags);
1466 
1467 	/* don't bdput right now, we don't want the bdev
1468 	 * to go away while it is locked.
1469 	 */
1470 	return 0;
1471 }
1472 
1473 static void unlock_fs(struct mapped_device *md)
1474 {
1475 	if (!test_bit(DMF_FROZEN, &md->flags))
1476 		return;
1477 
1478 	thaw_bdev(md->suspended_bdev, md->frozen_sb);
1479 	md->frozen_sb = NULL;
1480 	clear_bit(DMF_FROZEN, &md->flags);
1481 }
1482 
1483 /*
1484  * We need to be able to change a mapping table under a mounted
1485  * filesystem.  For example we might want to move some data in
1486  * the background.  Before the table can be swapped with
1487  * dm_bind_table, dm_suspend must be called to flush any in
1488  * flight bios and ensure that any further io gets deferred.
1489  */
1490 int dm_suspend(struct mapped_device *md, unsigned suspend_flags)
1491 {
1492 	struct dm_table *map = NULL;
1493 	DECLARE_WAITQUEUE(wait, current);
1494 	int r = 0;
1495 	int do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG ? 1 : 0;
1496 	int noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG ? 1 : 0;
1497 
1498 	mutex_lock(&md->suspend_lock);
1499 
1500 	if (dm_suspended(md)) {
1501 		r = -EINVAL;
1502 		goto out_unlock;
1503 	}
1504 
1505 	map = dm_get_table(md);
1506 
1507 	/*
1508 	 * DMF_NOFLUSH_SUSPENDING must be set before presuspend.
1509 	 * This flag is cleared before dm_suspend returns.
1510 	 */
1511 	if (noflush)
1512 		set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
1513 
1514 	/* This does not get reverted if there's an error later. */
1515 	dm_table_presuspend_targets(map);
1516 
1517 	/* bdget() can stall if the pending I/Os are not flushed */
1518 	if (!noflush) {
1519 		md->suspended_bdev = bdget_disk(md->disk, 0);
1520 		if (!md->suspended_bdev) {
1521 			DMWARN("bdget failed in dm_suspend");
1522 			r = -ENOMEM;
1523 			goto out;
1524 		}
1525 
1526 		/*
1527 		 * Flush I/O to the device. noflush supersedes do_lockfs,
1528 		 * because lock_fs() needs to flush I/Os.
1529 		 */
1530 		if (do_lockfs) {
1531 			r = lock_fs(md);
1532 			if (r)
1533 				goto out;
1534 		}
1535 	}
1536 
1537 	/*
1538 	 * First we set the BLOCK_IO flag so no more ios will be mapped.
1539 	 */
1540 	down_write(&md->io_lock);
1541 	set_bit(DMF_BLOCK_IO, &md->flags);
1542 
1543 	add_wait_queue(&md->wait, &wait);
1544 	up_write(&md->io_lock);
1545 
1546 	/* unplug */
1547 	if (map)
1548 		dm_table_unplug_all(map);
1549 
1550 	/*
1551 	 * Wait for the already-mapped ios to complete.
1552 	 */
1553 	r = dm_wait_for_completion(md);
1554 
1555 	down_write(&md->io_lock);
1556 	remove_wait_queue(&md->wait, &wait);
1557 
1558 	if (noflush)
1559 		__merge_pushback_list(md);
1560 	up_write(&md->io_lock);
1561 
1562 	/* were we interrupted ? */
1563 	if (r < 0) {
1564 		dm_queue_flush(md, DM_WQ_FLUSH_DEFERRED, NULL);
1565 
1566 		unlock_fs(md);
1567 		goto out; /* pushback list is already flushed, so skip flush */
1568 	}
1569 
1570 	dm_table_postsuspend_targets(map);
1571 
1572 	set_bit(DMF_SUSPENDED, &md->flags);
1573 
1574 out:
1575 	if (r && md->suspended_bdev) {
1576 		bdput(md->suspended_bdev);
1577 		md->suspended_bdev = NULL;
1578 	}
1579 
1580 	dm_table_put(map);
1581 
1582 out_unlock:
1583 	mutex_unlock(&md->suspend_lock);
1584 	return r;
1585 }
1586 
1587 int dm_resume(struct mapped_device *md)
1588 {
1589 	int r = -EINVAL;
1590 	struct dm_table *map = NULL;
1591 
1592 	mutex_lock(&md->suspend_lock);
1593 	if (!dm_suspended(md))
1594 		goto out;
1595 
1596 	map = dm_get_table(md);
1597 	if (!map || !dm_table_get_size(map))
1598 		goto out;
1599 
1600 	r = dm_table_resume_targets(map);
1601 	if (r)
1602 		goto out;
1603 
1604 	dm_queue_flush(md, DM_WQ_FLUSH_DEFERRED, NULL);
1605 
1606 	unlock_fs(md);
1607 
1608 	if (md->suspended_bdev) {
1609 		bdput(md->suspended_bdev);
1610 		md->suspended_bdev = NULL;
1611 	}
1612 
1613 	clear_bit(DMF_SUSPENDED, &md->flags);
1614 
1615 	dm_table_unplug_all(map);
1616 
1617 	dm_kobject_uevent(md);
1618 
1619 	r = 0;
1620 
1621 out:
1622 	dm_table_put(map);
1623 	mutex_unlock(&md->suspend_lock);
1624 
1625 	return r;
1626 }
1627 
1628 /*-----------------------------------------------------------------
1629  * Event notification.
1630  *---------------------------------------------------------------*/
1631 void dm_kobject_uevent(struct mapped_device *md)
1632 {
1633 	kobject_uevent(&disk_to_dev(md->disk)->kobj, KOBJ_CHANGE);
1634 }
1635 
1636 uint32_t dm_next_uevent_seq(struct mapped_device *md)
1637 {
1638 	return atomic_add_return(1, &md->uevent_seq);
1639 }
1640 
1641 uint32_t dm_get_event_nr(struct mapped_device *md)
1642 {
1643 	return atomic_read(&md->event_nr);
1644 }
1645 
1646 int dm_wait_event(struct mapped_device *md, int event_nr)
1647 {
1648 	return wait_event_interruptible(md->eventq,
1649 			(event_nr != atomic_read(&md->event_nr)));
1650 }
1651 
1652 void dm_uevent_add(struct mapped_device *md, struct list_head *elist)
1653 {
1654 	unsigned long flags;
1655 
1656 	spin_lock_irqsave(&md->uevent_lock, flags);
1657 	list_add(elist, &md->uevent_list);
1658 	spin_unlock_irqrestore(&md->uevent_lock, flags);
1659 }
1660 
1661 /*
1662  * The gendisk is only valid as long as you have a reference
1663  * count on 'md'.
1664  */
1665 struct gendisk *dm_disk(struct mapped_device *md)
1666 {
1667 	return md->disk;
1668 }
1669 
1670 int dm_suspended(struct mapped_device *md)
1671 {
1672 	return test_bit(DMF_SUSPENDED, &md->flags);
1673 }
1674 
1675 int dm_noflush_suspending(struct dm_target *ti)
1676 {
1677 	struct mapped_device *md = dm_table_get_md(ti->table);
1678 	int r = __noflush_suspending(md);
1679 
1680 	dm_put(md);
1681 
1682 	return r;
1683 }
1684 EXPORT_SYMBOL_GPL(dm_noflush_suspending);
1685 
1686 static struct block_device_operations dm_blk_dops = {
1687 	.open = dm_blk_open,
1688 	.release = dm_blk_close,
1689 	.ioctl = dm_blk_ioctl,
1690 	.getgeo = dm_blk_getgeo,
1691 	.owner = THIS_MODULE
1692 };
1693 
1694 EXPORT_SYMBOL(dm_get_mapinfo);
1695 
1696 /*
1697  * module hooks
1698  */
1699 module_init(dm_init);
1700 module_exit(dm_exit);
1701 
1702 module_param(major, uint, 0);
1703 MODULE_PARM_DESC(major, "The major number of the device mapper");
1704 MODULE_DESCRIPTION(DM_NAME " driver");
1705 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
1706 MODULE_LICENSE("GPL");
1707