xref: /openbmc/linux/drivers/md/dm.c (revision 2761e95fe40ca0d01864310fa4d488d7c5e34e18)
1 /*
2  * Copyright (C) 2001, 2002 Sistina Software (UK) Limited.
3  * Copyright (C) 2004-2008 Red Hat, Inc. All rights reserved.
4  *
5  * This file is released under the GPL.
6  */
7 
8 #include "dm.h"
9 #include "dm-uevent.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 
23 #include <trace/events/block.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  * For bio-based dm.
35  * One of these is allocated per bio.
36  */
37 struct dm_io {
38 	struct mapped_device *md;
39 	int error;
40 	atomic_t io_count;
41 	struct bio *bio;
42 	unsigned long start_time;
43 };
44 
45 /*
46  * For bio-based dm.
47  * One of these is allocated per target within a bio.  Hopefully
48  * this will be simplified out one day.
49  */
50 struct dm_target_io {
51 	struct dm_io *io;
52 	struct dm_target *ti;
53 	union map_info info;
54 };
55 
56 /*
57  * For request-based dm.
58  * One of these is allocated per request.
59  */
60 struct dm_rq_target_io {
61 	struct mapped_device *md;
62 	struct dm_target *ti;
63 	struct request *orig, clone;
64 	int error;
65 	union map_info info;
66 };
67 
68 /*
69  * For request-based dm.
70  * One of these is allocated per bio.
71  */
72 struct dm_rq_clone_bio_info {
73 	struct bio *orig;
74 	struct request *rq;
75 };
76 
77 union map_info *dm_get_mapinfo(struct bio *bio)
78 {
79 	if (bio && bio->bi_private)
80 		return &((struct dm_target_io *)bio->bi_private)->info;
81 	return NULL;
82 }
83 
84 #define MINOR_ALLOCED ((void *)-1)
85 
86 /*
87  * Bits for the md->flags field.
88  */
89 #define DMF_BLOCK_IO_FOR_SUSPEND 0
90 #define DMF_SUSPENDED 1
91 #define DMF_FROZEN 2
92 #define DMF_FREEING 3
93 #define DMF_DELETING 4
94 #define DMF_NOFLUSH_SUSPENDING 5
95 #define DMF_QUEUE_IO_TO_THREAD 6
96 
97 /*
98  * Work processed by per-device workqueue.
99  */
100 struct mapped_device {
101 	struct rw_semaphore io_lock;
102 	struct mutex suspend_lock;
103 	rwlock_t map_lock;
104 	atomic_t holders;
105 	atomic_t open_count;
106 
107 	unsigned long flags;
108 
109 	struct request_queue *queue;
110 	struct gendisk *disk;
111 	char name[16];
112 
113 	void *interface_ptr;
114 
115 	/*
116 	 * A list of ios that arrived while we were suspended.
117 	 */
118 	atomic_t pending;
119 	wait_queue_head_t wait;
120 	struct work_struct work;
121 	struct bio_list deferred;
122 	spinlock_t deferred_lock;
123 
124 	/*
125 	 * An error from the barrier request currently being processed.
126 	 */
127 	int barrier_error;
128 
129 	/*
130 	 * Processing queue (flush/barriers)
131 	 */
132 	struct workqueue_struct *wq;
133 
134 	/*
135 	 * The current mapping.
136 	 */
137 	struct dm_table *map;
138 
139 	/*
140 	 * io objects are allocated from here.
141 	 */
142 	mempool_t *io_pool;
143 	mempool_t *tio_pool;
144 
145 	struct bio_set *bs;
146 
147 	/*
148 	 * Event handling.
149 	 */
150 	atomic_t event_nr;
151 	wait_queue_head_t eventq;
152 	atomic_t uevent_seq;
153 	struct list_head uevent_list;
154 	spinlock_t uevent_lock; /* Protect access to uevent_list */
155 
156 	/*
157 	 * freeze/thaw support require holding onto a super block
158 	 */
159 	struct super_block *frozen_sb;
160 	struct block_device *bdev;
161 
162 	/* forced geometry settings */
163 	struct hd_geometry geometry;
164 
165 	/* sysfs handle */
166 	struct kobject kobj;
167 };
168 
169 #define MIN_IOS 256
170 static struct kmem_cache *_io_cache;
171 static struct kmem_cache *_tio_cache;
172 static struct kmem_cache *_rq_tio_cache;
173 static struct kmem_cache *_rq_bio_info_cache;
174 
175 static int __init local_init(void)
176 {
177 	int r = -ENOMEM;
178 
179 	/* allocate a slab for the dm_ios */
180 	_io_cache = KMEM_CACHE(dm_io, 0);
181 	if (!_io_cache)
182 		return r;
183 
184 	/* allocate a slab for the target ios */
185 	_tio_cache = KMEM_CACHE(dm_target_io, 0);
186 	if (!_tio_cache)
187 		goto out_free_io_cache;
188 
189 	_rq_tio_cache = KMEM_CACHE(dm_rq_target_io, 0);
190 	if (!_rq_tio_cache)
191 		goto out_free_tio_cache;
192 
193 	_rq_bio_info_cache = KMEM_CACHE(dm_rq_clone_bio_info, 0);
194 	if (!_rq_bio_info_cache)
195 		goto out_free_rq_tio_cache;
196 
197 	r = dm_uevent_init();
198 	if (r)
199 		goto out_free_rq_bio_info_cache;
200 
201 	_major = major;
202 	r = register_blkdev(_major, _name);
203 	if (r < 0)
204 		goto out_uevent_exit;
205 
206 	if (!_major)
207 		_major = r;
208 
209 	return 0;
210 
211 out_uevent_exit:
212 	dm_uevent_exit();
213 out_free_rq_bio_info_cache:
214 	kmem_cache_destroy(_rq_bio_info_cache);
215 out_free_rq_tio_cache:
216 	kmem_cache_destroy(_rq_tio_cache);
217 out_free_tio_cache:
218 	kmem_cache_destroy(_tio_cache);
219 out_free_io_cache:
220 	kmem_cache_destroy(_io_cache);
221 
222 	return r;
223 }
224 
225 static void local_exit(void)
226 {
227 	kmem_cache_destroy(_rq_bio_info_cache);
228 	kmem_cache_destroy(_rq_tio_cache);
229 	kmem_cache_destroy(_tio_cache);
230 	kmem_cache_destroy(_io_cache);
231 	unregister_blkdev(_major, _name);
232 	dm_uevent_exit();
233 
234 	_major = 0;
235 
236 	DMINFO("cleaned up");
237 }
238 
239 static int (*_inits[])(void) __initdata = {
240 	local_init,
241 	dm_target_init,
242 	dm_linear_init,
243 	dm_stripe_init,
244 	dm_kcopyd_init,
245 	dm_interface_init,
246 };
247 
248 static void (*_exits[])(void) = {
249 	local_exit,
250 	dm_target_exit,
251 	dm_linear_exit,
252 	dm_stripe_exit,
253 	dm_kcopyd_exit,
254 	dm_interface_exit,
255 };
256 
257 static int __init dm_init(void)
258 {
259 	const int count = ARRAY_SIZE(_inits);
260 
261 	int r, i;
262 
263 	for (i = 0; i < count; i++) {
264 		r = _inits[i]();
265 		if (r)
266 			goto bad;
267 	}
268 
269 	return 0;
270 
271       bad:
272 	while (i--)
273 		_exits[i]();
274 
275 	return r;
276 }
277 
278 static void __exit dm_exit(void)
279 {
280 	int i = ARRAY_SIZE(_exits);
281 
282 	while (i--)
283 		_exits[i]();
284 }
285 
286 /*
287  * Block device functions
288  */
289 static int dm_blk_open(struct block_device *bdev, fmode_t mode)
290 {
291 	struct mapped_device *md;
292 
293 	spin_lock(&_minor_lock);
294 
295 	md = bdev->bd_disk->private_data;
296 	if (!md)
297 		goto out;
298 
299 	if (test_bit(DMF_FREEING, &md->flags) ||
300 	    test_bit(DMF_DELETING, &md->flags)) {
301 		md = NULL;
302 		goto out;
303 	}
304 
305 	dm_get(md);
306 	atomic_inc(&md->open_count);
307 
308 out:
309 	spin_unlock(&_minor_lock);
310 
311 	return md ? 0 : -ENXIO;
312 }
313 
314 static int dm_blk_close(struct gendisk *disk, fmode_t mode)
315 {
316 	struct mapped_device *md = disk->private_data;
317 	atomic_dec(&md->open_count);
318 	dm_put(md);
319 	return 0;
320 }
321 
322 int dm_open_count(struct mapped_device *md)
323 {
324 	return atomic_read(&md->open_count);
325 }
326 
327 /*
328  * Guarantees nothing is using the device before it's deleted.
329  */
330 int dm_lock_for_deletion(struct mapped_device *md)
331 {
332 	int r = 0;
333 
334 	spin_lock(&_minor_lock);
335 
336 	if (dm_open_count(md))
337 		r = -EBUSY;
338 	else
339 		set_bit(DMF_DELETING, &md->flags);
340 
341 	spin_unlock(&_minor_lock);
342 
343 	return r;
344 }
345 
346 static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo)
347 {
348 	struct mapped_device *md = bdev->bd_disk->private_data;
349 
350 	return dm_get_geometry(md, geo);
351 }
352 
353 static int dm_blk_ioctl(struct block_device *bdev, fmode_t mode,
354 			unsigned int cmd, unsigned long arg)
355 {
356 	struct mapped_device *md = bdev->bd_disk->private_data;
357 	struct dm_table *map = dm_get_table(md);
358 	struct dm_target *tgt;
359 	int r = -ENOTTY;
360 
361 	if (!map || !dm_table_get_size(map))
362 		goto out;
363 
364 	/* We only support devices that have a single target */
365 	if (dm_table_get_num_targets(map) != 1)
366 		goto out;
367 
368 	tgt = dm_table_get_target(map, 0);
369 
370 	if (dm_suspended(md)) {
371 		r = -EAGAIN;
372 		goto out;
373 	}
374 
375 	if (tgt->type->ioctl)
376 		r = tgt->type->ioctl(tgt, cmd, arg);
377 
378 out:
379 	dm_table_put(map);
380 
381 	return r;
382 }
383 
384 static struct dm_io *alloc_io(struct mapped_device *md)
385 {
386 	return mempool_alloc(md->io_pool, GFP_NOIO);
387 }
388 
389 static void free_io(struct mapped_device *md, struct dm_io *io)
390 {
391 	mempool_free(io, md->io_pool);
392 }
393 
394 static struct dm_target_io *alloc_tio(struct mapped_device *md)
395 {
396 	return mempool_alloc(md->tio_pool, GFP_NOIO);
397 }
398 
399 static void free_tio(struct mapped_device *md, struct dm_target_io *tio)
400 {
401 	mempool_free(tio, md->tio_pool);
402 }
403 
404 static void start_io_acct(struct dm_io *io)
405 {
406 	struct mapped_device *md = io->md;
407 	int cpu;
408 
409 	io->start_time = jiffies;
410 
411 	cpu = part_stat_lock();
412 	part_round_stats(cpu, &dm_disk(md)->part0);
413 	part_stat_unlock();
414 	dm_disk(md)->part0.in_flight = atomic_inc_return(&md->pending);
415 }
416 
417 static void end_io_acct(struct dm_io *io)
418 {
419 	struct mapped_device *md = io->md;
420 	struct bio *bio = io->bio;
421 	unsigned long duration = jiffies - io->start_time;
422 	int pending, cpu;
423 	int rw = bio_data_dir(bio);
424 
425 	cpu = part_stat_lock();
426 	part_round_stats(cpu, &dm_disk(md)->part0);
427 	part_stat_add(cpu, &dm_disk(md)->part0, ticks[rw], duration);
428 	part_stat_unlock();
429 
430 	/*
431 	 * After this is decremented the bio must not be touched if it is
432 	 * a barrier.
433 	 */
434 	dm_disk(md)->part0.in_flight = pending =
435 		atomic_dec_return(&md->pending);
436 
437 	/* nudge anyone waiting on suspend queue */
438 	if (!pending)
439 		wake_up(&md->wait);
440 }
441 
442 /*
443  * Add the bio to the list of deferred io.
444  */
445 static void queue_io(struct mapped_device *md, struct bio *bio)
446 {
447 	down_write(&md->io_lock);
448 
449 	spin_lock_irq(&md->deferred_lock);
450 	bio_list_add(&md->deferred, bio);
451 	spin_unlock_irq(&md->deferred_lock);
452 
453 	if (!test_and_set_bit(DMF_QUEUE_IO_TO_THREAD, &md->flags))
454 		queue_work(md->wq, &md->work);
455 
456 	up_write(&md->io_lock);
457 }
458 
459 /*
460  * Everyone (including functions in this file), should use this
461  * function to access the md->map field, and make sure they call
462  * dm_table_put() when finished.
463  */
464 struct dm_table *dm_get_table(struct mapped_device *md)
465 {
466 	struct dm_table *t;
467 
468 	read_lock(&md->map_lock);
469 	t = md->map;
470 	if (t)
471 		dm_table_get(t);
472 	read_unlock(&md->map_lock);
473 
474 	return t;
475 }
476 
477 /*
478  * Get the geometry associated with a dm device
479  */
480 int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo)
481 {
482 	*geo = md->geometry;
483 
484 	return 0;
485 }
486 
487 /*
488  * Set the geometry of a device.
489  */
490 int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo)
491 {
492 	sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors;
493 
494 	if (geo->start > sz) {
495 		DMWARN("Start sector is beyond the geometry limits.");
496 		return -EINVAL;
497 	}
498 
499 	md->geometry = *geo;
500 
501 	return 0;
502 }
503 
504 /*-----------------------------------------------------------------
505  * CRUD START:
506  *   A more elegant soln is in the works that uses the queue
507  *   merge fn, unfortunately there are a couple of changes to
508  *   the block layer that I want to make for this.  So in the
509  *   interests of getting something for people to use I give
510  *   you this clearly demarcated crap.
511  *---------------------------------------------------------------*/
512 
513 static int __noflush_suspending(struct mapped_device *md)
514 {
515 	return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
516 }
517 
518 /*
519  * Decrements the number of outstanding ios that a bio has been
520  * cloned into, completing the original io if necc.
521  */
522 static void dec_pending(struct dm_io *io, int error)
523 {
524 	unsigned long flags;
525 	int io_error;
526 	struct bio *bio;
527 	struct mapped_device *md = io->md;
528 
529 	/* Push-back supersedes any I/O errors */
530 	if (error && !(io->error > 0 && __noflush_suspending(md)))
531 		io->error = error;
532 
533 	if (atomic_dec_and_test(&io->io_count)) {
534 		if (io->error == DM_ENDIO_REQUEUE) {
535 			/*
536 			 * Target requested pushing back the I/O.
537 			 */
538 			spin_lock_irqsave(&md->deferred_lock, flags);
539 			if (__noflush_suspending(md)) {
540 				if (!bio_barrier(io->bio))
541 					bio_list_add_head(&md->deferred,
542 							  io->bio);
543 			} else
544 				/* noflush suspend was interrupted. */
545 				io->error = -EIO;
546 			spin_unlock_irqrestore(&md->deferred_lock, flags);
547 		}
548 
549 		io_error = io->error;
550 		bio = io->bio;
551 
552 		if (bio_barrier(bio)) {
553 			/*
554 			 * There can be just one barrier request so we use
555 			 * a per-device variable for error reporting.
556 			 * Note that you can't touch the bio after end_io_acct
557 			 */
558 			md->barrier_error = io_error;
559 			end_io_acct(io);
560 		} else {
561 			end_io_acct(io);
562 
563 			if (io_error != DM_ENDIO_REQUEUE) {
564 				trace_block_bio_complete(md->queue, bio);
565 
566 				bio_endio(bio, io_error);
567 			}
568 		}
569 
570 		free_io(md, io);
571 	}
572 }
573 
574 static void clone_endio(struct bio *bio, int error)
575 {
576 	int r = 0;
577 	struct dm_target_io *tio = bio->bi_private;
578 	struct dm_io *io = tio->io;
579 	struct mapped_device *md = tio->io->md;
580 	dm_endio_fn endio = tio->ti->type->end_io;
581 
582 	if (!bio_flagged(bio, BIO_UPTODATE) && !error)
583 		error = -EIO;
584 
585 	if (endio) {
586 		r = endio(tio->ti, bio, error, &tio->info);
587 		if (r < 0 || r == DM_ENDIO_REQUEUE)
588 			/*
589 			 * error and requeue request are handled
590 			 * in dec_pending().
591 			 */
592 			error = r;
593 		else if (r == DM_ENDIO_INCOMPLETE)
594 			/* The target will handle the io */
595 			return;
596 		else if (r) {
597 			DMWARN("unimplemented target endio return value: %d", r);
598 			BUG();
599 		}
600 	}
601 
602 	/*
603 	 * Store md for cleanup instead of tio which is about to get freed.
604 	 */
605 	bio->bi_private = md->bs;
606 
607 	free_tio(md, tio);
608 	bio_put(bio);
609 	dec_pending(io, error);
610 }
611 
612 static sector_t max_io_len(struct mapped_device *md,
613 			   sector_t sector, struct dm_target *ti)
614 {
615 	sector_t offset = sector - ti->begin;
616 	sector_t len = ti->len - offset;
617 
618 	/*
619 	 * Does the target need to split even further ?
620 	 */
621 	if (ti->split_io) {
622 		sector_t boundary;
623 		boundary = ((offset + ti->split_io) & ~(ti->split_io - 1))
624 			   - offset;
625 		if (len > boundary)
626 			len = boundary;
627 	}
628 
629 	return len;
630 }
631 
632 static void __map_bio(struct dm_target *ti, struct bio *clone,
633 		      struct dm_target_io *tio)
634 {
635 	int r;
636 	sector_t sector;
637 	struct mapped_device *md;
638 
639 	/*
640 	 * Sanity checks.
641 	 */
642 	BUG_ON(!clone->bi_size);
643 
644 	clone->bi_end_io = clone_endio;
645 	clone->bi_private = tio;
646 
647 	/*
648 	 * Map the clone.  If r == 0 we don't need to do
649 	 * anything, the target has assumed ownership of
650 	 * this io.
651 	 */
652 	atomic_inc(&tio->io->io_count);
653 	sector = clone->bi_sector;
654 	r = ti->type->map(ti, clone, &tio->info);
655 	if (r == DM_MAPIO_REMAPPED) {
656 		/* the bio has been remapped so dispatch it */
657 
658 		trace_block_remap(bdev_get_queue(clone->bi_bdev), clone,
659 				    tio->io->bio->bi_bdev->bd_dev, sector);
660 
661 		generic_make_request(clone);
662 	} else if (r < 0 || r == DM_MAPIO_REQUEUE) {
663 		/* error the io and bail out, or requeue it if needed */
664 		md = tio->io->md;
665 		dec_pending(tio->io, r);
666 		/*
667 		 * Store bio_set for cleanup.
668 		 */
669 		clone->bi_private = md->bs;
670 		bio_put(clone);
671 		free_tio(md, tio);
672 	} else if (r) {
673 		DMWARN("unimplemented target map return value: %d", r);
674 		BUG();
675 	}
676 }
677 
678 struct clone_info {
679 	struct mapped_device *md;
680 	struct dm_table *map;
681 	struct bio *bio;
682 	struct dm_io *io;
683 	sector_t sector;
684 	sector_t sector_count;
685 	unsigned short idx;
686 };
687 
688 static void dm_bio_destructor(struct bio *bio)
689 {
690 	struct bio_set *bs = bio->bi_private;
691 
692 	bio_free(bio, bs);
693 }
694 
695 /*
696  * Creates a little bio that is just does part of a bvec.
697  */
698 static struct bio *split_bvec(struct bio *bio, sector_t sector,
699 			      unsigned short idx, unsigned int offset,
700 			      unsigned int len, struct bio_set *bs)
701 {
702 	struct bio *clone;
703 	struct bio_vec *bv = bio->bi_io_vec + idx;
704 
705 	clone = bio_alloc_bioset(GFP_NOIO, 1, bs);
706 	clone->bi_destructor = dm_bio_destructor;
707 	*clone->bi_io_vec = *bv;
708 
709 	clone->bi_sector = sector;
710 	clone->bi_bdev = bio->bi_bdev;
711 	clone->bi_rw = bio->bi_rw & ~(1 << BIO_RW_BARRIER);
712 	clone->bi_vcnt = 1;
713 	clone->bi_size = to_bytes(len);
714 	clone->bi_io_vec->bv_offset = offset;
715 	clone->bi_io_vec->bv_len = clone->bi_size;
716 	clone->bi_flags |= 1 << BIO_CLONED;
717 
718 	if (bio_integrity(bio)) {
719 		bio_integrity_clone(clone, bio, GFP_NOIO);
720 		bio_integrity_trim(clone,
721 				   bio_sector_offset(bio, idx, offset), len);
722 	}
723 
724 	return clone;
725 }
726 
727 /*
728  * Creates a bio that consists of range of complete bvecs.
729  */
730 static struct bio *clone_bio(struct bio *bio, sector_t sector,
731 			     unsigned short idx, unsigned short bv_count,
732 			     unsigned int len, struct bio_set *bs)
733 {
734 	struct bio *clone;
735 
736 	clone = bio_alloc_bioset(GFP_NOIO, bio->bi_max_vecs, bs);
737 	__bio_clone(clone, bio);
738 	clone->bi_rw &= ~(1 << BIO_RW_BARRIER);
739 	clone->bi_destructor = dm_bio_destructor;
740 	clone->bi_sector = sector;
741 	clone->bi_idx = idx;
742 	clone->bi_vcnt = idx + bv_count;
743 	clone->bi_size = to_bytes(len);
744 	clone->bi_flags &= ~(1 << BIO_SEG_VALID);
745 
746 	if (bio_integrity(bio)) {
747 		bio_integrity_clone(clone, bio, GFP_NOIO);
748 
749 		if (idx != bio->bi_idx || clone->bi_size < bio->bi_size)
750 			bio_integrity_trim(clone,
751 					   bio_sector_offset(bio, idx, 0), len);
752 	}
753 
754 	return clone;
755 }
756 
757 static int __clone_and_map(struct clone_info *ci)
758 {
759 	struct bio *clone, *bio = ci->bio;
760 	struct dm_target *ti;
761 	sector_t len = 0, max;
762 	struct dm_target_io *tio;
763 
764 	ti = dm_table_find_target(ci->map, ci->sector);
765 	if (!dm_target_is_valid(ti))
766 		return -EIO;
767 
768 	max = max_io_len(ci->md, ci->sector, ti);
769 
770 	/*
771 	 * Allocate a target io object.
772 	 */
773 	tio = alloc_tio(ci->md);
774 	tio->io = ci->io;
775 	tio->ti = ti;
776 	memset(&tio->info, 0, sizeof(tio->info));
777 
778 	if (ci->sector_count <= max) {
779 		/*
780 		 * Optimise for the simple case where we can do all of
781 		 * the remaining io with a single clone.
782 		 */
783 		clone = clone_bio(bio, ci->sector, ci->idx,
784 				  bio->bi_vcnt - ci->idx, ci->sector_count,
785 				  ci->md->bs);
786 		__map_bio(ti, clone, tio);
787 		ci->sector_count = 0;
788 
789 	} else if (to_sector(bio->bi_io_vec[ci->idx].bv_len) <= max) {
790 		/*
791 		 * There are some bvecs that don't span targets.
792 		 * Do as many of these as possible.
793 		 */
794 		int i;
795 		sector_t remaining = max;
796 		sector_t bv_len;
797 
798 		for (i = ci->idx; remaining && (i < bio->bi_vcnt); i++) {
799 			bv_len = to_sector(bio->bi_io_vec[i].bv_len);
800 
801 			if (bv_len > remaining)
802 				break;
803 
804 			remaining -= bv_len;
805 			len += bv_len;
806 		}
807 
808 		clone = clone_bio(bio, ci->sector, ci->idx, i - ci->idx, len,
809 				  ci->md->bs);
810 		__map_bio(ti, clone, tio);
811 
812 		ci->sector += len;
813 		ci->sector_count -= len;
814 		ci->idx = i;
815 
816 	} else {
817 		/*
818 		 * Handle a bvec that must be split between two or more targets.
819 		 */
820 		struct bio_vec *bv = bio->bi_io_vec + ci->idx;
821 		sector_t remaining = to_sector(bv->bv_len);
822 		unsigned int offset = 0;
823 
824 		do {
825 			if (offset) {
826 				ti = dm_table_find_target(ci->map, ci->sector);
827 				if (!dm_target_is_valid(ti))
828 					return -EIO;
829 
830 				max = max_io_len(ci->md, ci->sector, ti);
831 
832 				tio = alloc_tio(ci->md);
833 				tio->io = ci->io;
834 				tio->ti = ti;
835 				memset(&tio->info, 0, sizeof(tio->info));
836 			}
837 
838 			len = min(remaining, max);
839 
840 			clone = split_bvec(bio, ci->sector, ci->idx,
841 					   bv->bv_offset + offset, len,
842 					   ci->md->bs);
843 
844 			__map_bio(ti, clone, tio);
845 
846 			ci->sector += len;
847 			ci->sector_count -= len;
848 			offset += to_bytes(len);
849 		} while (remaining -= len);
850 
851 		ci->idx++;
852 	}
853 
854 	return 0;
855 }
856 
857 /*
858  * Split the bio into several clones and submit it to targets.
859  */
860 static void __split_and_process_bio(struct mapped_device *md, struct bio *bio)
861 {
862 	struct clone_info ci;
863 	int error = 0;
864 
865 	ci.map = dm_get_table(md);
866 	if (unlikely(!ci.map)) {
867 		if (!bio_barrier(bio))
868 			bio_io_error(bio);
869 		else
870 			md->barrier_error = -EIO;
871 		return;
872 	}
873 
874 	ci.md = md;
875 	ci.bio = bio;
876 	ci.io = alloc_io(md);
877 	ci.io->error = 0;
878 	atomic_set(&ci.io->io_count, 1);
879 	ci.io->bio = bio;
880 	ci.io->md = md;
881 	ci.sector = bio->bi_sector;
882 	ci.sector_count = bio_sectors(bio);
883 	ci.idx = bio->bi_idx;
884 
885 	start_io_acct(ci.io);
886 	while (ci.sector_count && !error)
887 		error = __clone_and_map(&ci);
888 
889 	/* drop the extra reference count */
890 	dec_pending(ci.io, error);
891 	dm_table_put(ci.map);
892 }
893 /*-----------------------------------------------------------------
894  * CRUD END
895  *---------------------------------------------------------------*/
896 
897 static int dm_merge_bvec(struct request_queue *q,
898 			 struct bvec_merge_data *bvm,
899 			 struct bio_vec *biovec)
900 {
901 	struct mapped_device *md = q->queuedata;
902 	struct dm_table *map = dm_get_table(md);
903 	struct dm_target *ti;
904 	sector_t max_sectors;
905 	int max_size = 0;
906 
907 	if (unlikely(!map))
908 		goto out;
909 
910 	ti = dm_table_find_target(map, bvm->bi_sector);
911 	if (!dm_target_is_valid(ti))
912 		goto out_table;
913 
914 	/*
915 	 * Find maximum amount of I/O that won't need splitting
916 	 */
917 	max_sectors = min(max_io_len(md, bvm->bi_sector, ti),
918 			  (sector_t) BIO_MAX_SECTORS);
919 	max_size = (max_sectors << SECTOR_SHIFT) - bvm->bi_size;
920 	if (max_size < 0)
921 		max_size = 0;
922 
923 	/*
924 	 * merge_bvec_fn() returns number of bytes
925 	 * it can accept at this offset
926 	 * max is precomputed maximal io size
927 	 */
928 	if (max_size && ti->type->merge)
929 		max_size = ti->type->merge(ti, bvm, biovec, max_size);
930 	/*
931 	 * If the target doesn't support merge method and some of the devices
932 	 * provided their merge_bvec method (we know this by looking at
933 	 * queue_max_hw_sectors), then we can't allow bios with multiple vector
934 	 * entries.  So always set max_size to 0, and the code below allows
935 	 * just one page.
936 	 */
937 	else if (queue_max_hw_sectors(q) <= PAGE_SIZE >> 9)
938 
939 		max_size = 0;
940 
941 out_table:
942 	dm_table_put(map);
943 
944 out:
945 	/*
946 	 * Always allow an entire first page
947 	 */
948 	if (max_size <= biovec->bv_len && !(bvm->bi_size >> SECTOR_SHIFT))
949 		max_size = biovec->bv_len;
950 
951 	return max_size;
952 }
953 
954 /*
955  * The request function that just remaps the bio built up by
956  * dm_merge_bvec.
957  */
958 static int dm_request(struct request_queue *q, struct bio *bio)
959 {
960 	int rw = bio_data_dir(bio);
961 	struct mapped_device *md = q->queuedata;
962 	int cpu;
963 
964 	down_read(&md->io_lock);
965 
966 	cpu = part_stat_lock();
967 	part_stat_inc(cpu, &dm_disk(md)->part0, ios[rw]);
968 	part_stat_add(cpu, &dm_disk(md)->part0, sectors[rw], bio_sectors(bio));
969 	part_stat_unlock();
970 
971 	/*
972 	 * If we're suspended or the thread is processing barriers
973 	 * we have to queue this io for later.
974 	 */
975 	if (unlikely(test_bit(DMF_QUEUE_IO_TO_THREAD, &md->flags)) ||
976 	    unlikely(bio_barrier(bio))) {
977 		up_read(&md->io_lock);
978 
979 		if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) &&
980 		    bio_rw(bio) == READA) {
981 			bio_io_error(bio);
982 			return 0;
983 		}
984 
985 		queue_io(md, bio);
986 
987 		return 0;
988 	}
989 
990 	__split_and_process_bio(md, bio);
991 	up_read(&md->io_lock);
992 	return 0;
993 }
994 
995 static void dm_unplug_all(struct request_queue *q)
996 {
997 	struct mapped_device *md = q->queuedata;
998 	struct dm_table *map = dm_get_table(md);
999 
1000 	if (map) {
1001 		dm_table_unplug_all(map);
1002 		dm_table_put(map);
1003 	}
1004 }
1005 
1006 static int dm_any_congested(void *congested_data, int bdi_bits)
1007 {
1008 	int r = bdi_bits;
1009 	struct mapped_device *md = congested_data;
1010 	struct dm_table *map;
1011 
1012 	if (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
1013 		map = dm_get_table(md);
1014 		if (map) {
1015 			r = dm_table_any_congested(map, bdi_bits);
1016 			dm_table_put(map);
1017 		}
1018 	}
1019 
1020 	return r;
1021 }
1022 
1023 /*-----------------------------------------------------------------
1024  * An IDR is used to keep track of allocated minor numbers.
1025  *---------------------------------------------------------------*/
1026 static DEFINE_IDR(_minor_idr);
1027 
1028 static void free_minor(int minor)
1029 {
1030 	spin_lock(&_minor_lock);
1031 	idr_remove(&_minor_idr, minor);
1032 	spin_unlock(&_minor_lock);
1033 }
1034 
1035 /*
1036  * See if the device with a specific minor # is free.
1037  */
1038 static int specific_minor(int minor)
1039 {
1040 	int r, m;
1041 
1042 	if (minor >= (1 << MINORBITS))
1043 		return -EINVAL;
1044 
1045 	r = idr_pre_get(&_minor_idr, GFP_KERNEL);
1046 	if (!r)
1047 		return -ENOMEM;
1048 
1049 	spin_lock(&_minor_lock);
1050 
1051 	if (idr_find(&_minor_idr, minor)) {
1052 		r = -EBUSY;
1053 		goto out;
1054 	}
1055 
1056 	r = idr_get_new_above(&_minor_idr, MINOR_ALLOCED, minor, &m);
1057 	if (r)
1058 		goto out;
1059 
1060 	if (m != minor) {
1061 		idr_remove(&_minor_idr, m);
1062 		r = -EBUSY;
1063 		goto out;
1064 	}
1065 
1066 out:
1067 	spin_unlock(&_minor_lock);
1068 	return r;
1069 }
1070 
1071 static int next_free_minor(int *minor)
1072 {
1073 	int r, m;
1074 
1075 	r = idr_pre_get(&_minor_idr, GFP_KERNEL);
1076 	if (!r)
1077 		return -ENOMEM;
1078 
1079 	spin_lock(&_minor_lock);
1080 
1081 	r = idr_get_new(&_minor_idr, MINOR_ALLOCED, &m);
1082 	if (r)
1083 		goto out;
1084 
1085 	if (m >= (1 << MINORBITS)) {
1086 		idr_remove(&_minor_idr, m);
1087 		r = -ENOSPC;
1088 		goto out;
1089 	}
1090 
1091 	*minor = m;
1092 
1093 out:
1094 	spin_unlock(&_minor_lock);
1095 	return r;
1096 }
1097 
1098 static struct block_device_operations dm_blk_dops;
1099 
1100 static void dm_wq_work(struct work_struct *work);
1101 
1102 /*
1103  * Allocate and initialise a blank device with a given minor.
1104  */
1105 static struct mapped_device *alloc_dev(int minor)
1106 {
1107 	int r;
1108 	struct mapped_device *md = kzalloc(sizeof(*md), GFP_KERNEL);
1109 	void *old_md;
1110 
1111 	if (!md) {
1112 		DMWARN("unable to allocate device, out of memory.");
1113 		return NULL;
1114 	}
1115 
1116 	if (!try_module_get(THIS_MODULE))
1117 		goto bad_module_get;
1118 
1119 	/* get a minor number for the dev */
1120 	if (minor == DM_ANY_MINOR)
1121 		r = next_free_minor(&minor);
1122 	else
1123 		r = specific_minor(minor);
1124 	if (r < 0)
1125 		goto bad_minor;
1126 
1127 	init_rwsem(&md->io_lock);
1128 	mutex_init(&md->suspend_lock);
1129 	spin_lock_init(&md->deferred_lock);
1130 	rwlock_init(&md->map_lock);
1131 	atomic_set(&md->holders, 1);
1132 	atomic_set(&md->open_count, 0);
1133 	atomic_set(&md->event_nr, 0);
1134 	atomic_set(&md->uevent_seq, 0);
1135 	INIT_LIST_HEAD(&md->uevent_list);
1136 	spin_lock_init(&md->uevent_lock);
1137 
1138 	md->queue = blk_alloc_queue(GFP_KERNEL);
1139 	if (!md->queue)
1140 		goto bad_queue;
1141 
1142 	md->queue->queuedata = md;
1143 	md->queue->backing_dev_info.congested_fn = dm_any_congested;
1144 	md->queue->backing_dev_info.congested_data = md;
1145 	blk_queue_make_request(md->queue, dm_request);
1146 	blk_queue_ordered(md->queue, QUEUE_ORDERED_DRAIN, NULL);
1147 	blk_queue_bounce_limit(md->queue, BLK_BOUNCE_ANY);
1148 	md->queue->unplug_fn = dm_unplug_all;
1149 	blk_queue_merge_bvec(md->queue, dm_merge_bvec);
1150 
1151 	md->io_pool = mempool_create_slab_pool(MIN_IOS, _io_cache);
1152 	if (!md->io_pool)
1153 		goto bad_io_pool;
1154 
1155 	md->tio_pool = mempool_create_slab_pool(MIN_IOS, _tio_cache);
1156 	if (!md->tio_pool)
1157 		goto bad_tio_pool;
1158 
1159 	md->bs = bioset_create(16, 0);
1160 	if (!md->bs)
1161 		goto bad_no_bioset;
1162 
1163 	md->disk = alloc_disk(1);
1164 	if (!md->disk)
1165 		goto bad_disk;
1166 
1167 	atomic_set(&md->pending, 0);
1168 	init_waitqueue_head(&md->wait);
1169 	INIT_WORK(&md->work, dm_wq_work);
1170 	init_waitqueue_head(&md->eventq);
1171 
1172 	md->disk->major = _major;
1173 	md->disk->first_minor = minor;
1174 	md->disk->fops = &dm_blk_dops;
1175 	md->disk->queue = md->queue;
1176 	md->disk->private_data = md;
1177 	sprintf(md->disk->disk_name, "dm-%d", minor);
1178 	add_disk(md->disk);
1179 	format_dev_t(md->name, MKDEV(_major, minor));
1180 
1181 	md->wq = create_singlethread_workqueue("kdmflush");
1182 	if (!md->wq)
1183 		goto bad_thread;
1184 
1185 	md->bdev = bdget_disk(md->disk, 0);
1186 	if (!md->bdev)
1187 		goto bad_bdev;
1188 
1189 	/* Populate the mapping, nobody knows we exist yet */
1190 	spin_lock(&_minor_lock);
1191 	old_md = idr_replace(&_minor_idr, md, minor);
1192 	spin_unlock(&_minor_lock);
1193 
1194 	BUG_ON(old_md != MINOR_ALLOCED);
1195 
1196 	return md;
1197 
1198 bad_bdev:
1199 	destroy_workqueue(md->wq);
1200 bad_thread:
1201 	put_disk(md->disk);
1202 bad_disk:
1203 	bioset_free(md->bs);
1204 bad_no_bioset:
1205 	mempool_destroy(md->tio_pool);
1206 bad_tio_pool:
1207 	mempool_destroy(md->io_pool);
1208 bad_io_pool:
1209 	blk_cleanup_queue(md->queue);
1210 bad_queue:
1211 	free_minor(minor);
1212 bad_minor:
1213 	module_put(THIS_MODULE);
1214 bad_module_get:
1215 	kfree(md);
1216 	return NULL;
1217 }
1218 
1219 static void unlock_fs(struct mapped_device *md);
1220 
1221 static void free_dev(struct mapped_device *md)
1222 {
1223 	int minor = MINOR(disk_devt(md->disk));
1224 
1225 	unlock_fs(md);
1226 	bdput(md->bdev);
1227 	destroy_workqueue(md->wq);
1228 	mempool_destroy(md->tio_pool);
1229 	mempool_destroy(md->io_pool);
1230 	bioset_free(md->bs);
1231 	blk_integrity_unregister(md->disk);
1232 	del_gendisk(md->disk);
1233 	free_minor(minor);
1234 
1235 	spin_lock(&_minor_lock);
1236 	md->disk->private_data = NULL;
1237 	spin_unlock(&_minor_lock);
1238 
1239 	put_disk(md->disk);
1240 	blk_cleanup_queue(md->queue);
1241 	module_put(THIS_MODULE);
1242 	kfree(md);
1243 }
1244 
1245 /*
1246  * Bind a table to the device.
1247  */
1248 static void event_callback(void *context)
1249 {
1250 	unsigned long flags;
1251 	LIST_HEAD(uevents);
1252 	struct mapped_device *md = (struct mapped_device *) context;
1253 
1254 	spin_lock_irqsave(&md->uevent_lock, flags);
1255 	list_splice_init(&md->uevent_list, &uevents);
1256 	spin_unlock_irqrestore(&md->uevent_lock, flags);
1257 
1258 	dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj);
1259 
1260 	atomic_inc(&md->event_nr);
1261 	wake_up(&md->eventq);
1262 }
1263 
1264 static void __set_size(struct mapped_device *md, sector_t size)
1265 {
1266 	set_capacity(md->disk, size);
1267 
1268 	mutex_lock(&md->bdev->bd_inode->i_mutex);
1269 	i_size_write(md->bdev->bd_inode, (loff_t)size << SECTOR_SHIFT);
1270 	mutex_unlock(&md->bdev->bd_inode->i_mutex);
1271 }
1272 
1273 static int __bind(struct mapped_device *md, struct dm_table *t)
1274 {
1275 	struct request_queue *q = md->queue;
1276 	sector_t size;
1277 
1278 	size = dm_table_get_size(t);
1279 
1280 	/*
1281 	 * Wipe any geometry if the size of the table changed.
1282 	 */
1283 	if (size != get_capacity(md->disk))
1284 		memset(&md->geometry, 0, sizeof(md->geometry));
1285 
1286 	__set_size(md, size);
1287 
1288 	if (!size) {
1289 		dm_table_destroy(t);
1290 		return 0;
1291 	}
1292 
1293 	dm_table_event_callback(t, event_callback, md);
1294 
1295 	write_lock(&md->map_lock);
1296 	md->map = t;
1297 	dm_table_set_restrictions(t, q);
1298 	write_unlock(&md->map_lock);
1299 
1300 	return 0;
1301 }
1302 
1303 static void __unbind(struct mapped_device *md)
1304 {
1305 	struct dm_table *map = md->map;
1306 
1307 	if (!map)
1308 		return;
1309 
1310 	dm_table_event_callback(map, NULL, NULL);
1311 	write_lock(&md->map_lock);
1312 	md->map = NULL;
1313 	write_unlock(&md->map_lock);
1314 	dm_table_destroy(map);
1315 }
1316 
1317 /*
1318  * Constructor for a new device.
1319  */
1320 int dm_create(int minor, struct mapped_device **result)
1321 {
1322 	struct mapped_device *md;
1323 
1324 	md = alloc_dev(minor);
1325 	if (!md)
1326 		return -ENXIO;
1327 
1328 	dm_sysfs_init(md);
1329 
1330 	*result = md;
1331 	return 0;
1332 }
1333 
1334 static struct mapped_device *dm_find_md(dev_t dev)
1335 {
1336 	struct mapped_device *md;
1337 	unsigned minor = MINOR(dev);
1338 
1339 	if (MAJOR(dev) != _major || minor >= (1 << MINORBITS))
1340 		return NULL;
1341 
1342 	spin_lock(&_minor_lock);
1343 
1344 	md = idr_find(&_minor_idr, minor);
1345 	if (md && (md == MINOR_ALLOCED ||
1346 		   (MINOR(disk_devt(dm_disk(md))) != minor) ||
1347 		   test_bit(DMF_FREEING, &md->flags))) {
1348 		md = NULL;
1349 		goto out;
1350 	}
1351 
1352 out:
1353 	spin_unlock(&_minor_lock);
1354 
1355 	return md;
1356 }
1357 
1358 struct mapped_device *dm_get_md(dev_t dev)
1359 {
1360 	struct mapped_device *md = dm_find_md(dev);
1361 
1362 	if (md)
1363 		dm_get(md);
1364 
1365 	return md;
1366 }
1367 
1368 void *dm_get_mdptr(struct mapped_device *md)
1369 {
1370 	return md->interface_ptr;
1371 }
1372 
1373 void dm_set_mdptr(struct mapped_device *md, void *ptr)
1374 {
1375 	md->interface_ptr = ptr;
1376 }
1377 
1378 void dm_get(struct mapped_device *md)
1379 {
1380 	atomic_inc(&md->holders);
1381 }
1382 
1383 const char *dm_device_name(struct mapped_device *md)
1384 {
1385 	return md->name;
1386 }
1387 EXPORT_SYMBOL_GPL(dm_device_name);
1388 
1389 void dm_put(struct mapped_device *md)
1390 {
1391 	struct dm_table *map;
1392 
1393 	BUG_ON(test_bit(DMF_FREEING, &md->flags));
1394 
1395 	if (atomic_dec_and_lock(&md->holders, &_minor_lock)) {
1396 		map = dm_get_table(md);
1397 		idr_replace(&_minor_idr, MINOR_ALLOCED,
1398 			    MINOR(disk_devt(dm_disk(md))));
1399 		set_bit(DMF_FREEING, &md->flags);
1400 		spin_unlock(&_minor_lock);
1401 		if (!dm_suspended(md)) {
1402 			dm_table_presuspend_targets(map);
1403 			dm_table_postsuspend_targets(map);
1404 		}
1405 		dm_sysfs_exit(md);
1406 		dm_table_put(map);
1407 		__unbind(md);
1408 		free_dev(md);
1409 	}
1410 }
1411 EXPORT_SYMBOL_GPL(dm_put);
1412 
1413 static int dm_wait_for_completion(struct mapped_device *md, int interruptible)
1414 {
1415 	int r = 0;
1416 	DECLARE_WAITQUEUE(wait, current);
1417 
1418 	dm_unplug_all(md->queue);
1419 
1420 	add_wait_queue(&md->wait, &wait);
1421 
1422 	while (1) {
1423 		set_current_state(interruptible);
1424 
1425 		smp_mb();
1426 		if (!atomic_read(&md->pending))
1427 			break;
1428 
1429 		if (interruptible == TASK_INTERRUPTIBLE &&
1430 		    signal_pending(current)) {
1431 			r = -EINTR;
1432 			break;
1433 		}
1434 
1435 		io_schedule();
1436 	}
1437 	set_current_state(TASK_RUNNING);
1438 
1439 	remove_wait_queue(&md->wait, &wait);
1440 
1441 	return r;
1442 }
1443 
1444 static void dm_flush(struct mapped_device *md)
1445 {
1446 	dm_wait_for_completion(md, TASK_UNINTERRUPTIBLE);
1447 }
1448 
1449 static void process_barrier(struct mapped_device *md, struct bio *bio)
1450 {
1451 	dm_flush(md);
1452 
1453 	if (bio_empty_barrier(bio)) {
1454 		bio_endio(bio, 0);
1455 		return;
1456 	}
1457 
1458 	__split_and_process_bio(md, bio);
1459 	dm_flush(md);
1460 
1461 	if (md->barrier_error != DM_ENDIO_REQUEUE)
1462 		bio_endio(bio, md->barrier_error);
1463 	else {
1464 		spin_lock_irq(&md->deferred_lock);
1465 		bio_list_add_head(&md->deferred, bio);
1466 		spin_unlock_irq(&md->deferred_lock);
1467 	}
1468 }
1469 
1470 /*
1471  * Process the deferred bios
1472  */
1473 static void dm_wq_work(struct work_struct *work)
1474 {
1475 	struct mapped_device *md = container_of(work, struct mapped_device,
1476 						work);
1477 	struct bio *c;
1478 
1479 	down_write(&md->io_lock);
1480 
1481 	while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
1482 		spin_lock_irq(&md->deferred_lock);
1483 		c = bio_list_pop(&md->deferred);
1484 		spin_unlock_irq(&md->deferred_lock);
1485 
1486 		if (!c) {
1487 			clear_bit(DMF_QUEUE_IO_TO_THREAD, &md->flags);
1488 			break;
1489 		}
1490 
1491 		up_write(&md->io_lock);
1492 
1493 		if (bio_barrier(c))
1494 			process_barrier(md, c);
1495 		else
1496 			__split_and_process_bio(md, c);
1497 
1498 		down_write(&md->io_lock);
1499 	}
1500 
1501 	up_write(&md->io_lock);
1502 }
1503 
1504 static void dm_queue_flush(struct mapped_device *md)
1505 {
1506 	clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
1507 	smp_mb__after_clear_bit();
1508 	queue_work(md->wq, &md->work);
1509 }
1510 
1511 /*
1512  * Swap in a new table (destroying old one).
1513  */
1514 int dm_swap_table(struct mapped_device *md, struct dm_table *table)
1515 {
1516 	int r = -EINVAL;
1517 
1518 	mutex_lock(&md->suspend_lock);
1519 
1520 	/* device must be suspended */
1521 	if (!dm_suspended(md))
1522 		goto out;
1523 
1524 	__unbind(md);
1525 	r = __bind(md, table);
1526 
1527 out:
1528 	mutex_unlock(&md->suspend_lock);
1529 	return r;
1530 }
1531 
1532 /*
1533  * Functions to lock and unlock any filesystem running on the
1534  * device.
1535  */
1536 static int lock_fs(struct mapped_device *md)
1537 {
1538 	int r;
1539 
1540 	WARN_ON(md->frozen_sb);
1541 
1542 	md->frozen_sb = freeze_bdev(md->bdev);
1543 	if (IS_ERR(md->frozen_sb)) {
1544 		r = PTR_ERR(md->frozen_sb);
1545 		md->frozen_sb = NULL;
1546 		return r;
1547 	}
1548 
1549 	set_bit(DMF_FROZEN, &md->flags);
1550 
1551 	return 0;
1552 }
1553 
1554 static void unlock_fs(struct mapped_device *md)
1555 {
1556 	if (!test_bit(DMF_FROZEN, &md->flags))
1557 		return;
1558 
1559 	thaw_bdev(md->bdev, md->frozen_sb);
1560 	md->frozen_sb = NULL;
1561 	clear_bit(DMF_FROZEN, &md->flags);
1562 }
1563 
1564 /*
1565  * We need to be able to change a mapping table under a mounted
1566  * filesystem.  For example we might want to move some data in
1567  * the background.  Before the table can be swapped with
1568  * dm_bind_table, dm_suspend must be called to flush any in
1569  * flight bios and ensure that any further io gets deferred.
1570  */
1571 int dm_suspend(struct mapped_device *md, unsigned suspend_flags)
1572 {
1573 	struct dm_table *map = NULL;
1574 	int r = 0;
1575 	int do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG ? 1 : 0;
1576 	int noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG ? 1 : 0;
1577 
1578 	mutex_lock(&md->suspend_lock);
1579 
1580 	if (dm_suspended(md)) {
1581 		r = -EINVAL;
1582 		goto out_unlock;
1583 	}
1584 
1585 	map = dm_get_table(md);
1586 
1587 	/*
1588 	 * DMF_NOFLUSH_SUSPENDING must be set before presuspend.
1589 	 * This flag is cleared before dm_suspend returns.
1590 	 */
1591 	if (noflush)
1592 		set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
1593 
1594 	/* This does not get reverted if there's an error later. */
1595 	dm_table_presuspend_targets(map);
1596 
1597 	/*
1598 	 * Flush I/O to the device. noflush supersedes do_lockfs,
1599 	 * because lock_fs() needs to flush I/Os.
1600 	 */
1601 	if (!noflush && do_lockfs) {
1602 		r = lock_fs(md);
1603 		if (r)
1604 			goto out;
1605 	}
1606 
1607 	/*
1608 	 * Here we must make sure that no processes are submitting requests
1609 	 * to target drivers i.e. no one may be executing
1610 	 * __split_and_process_bio. This is called from dm_request and
1611 	 * dm_wq_work.
1612 	 *
1613 	 * To get all processes out of __split_and_process_bio in dm_request,
1614 	 * we take the write lock. To prevent any process from reentering
1615 	 * __split_and_process_bio from dm_request, we set
1616 	 * DMF_QUEUE_IO_TO_THREAD.
1617 	 *
1618 	 * To quiesce the thread (dm_wq_work), we set DMF_BLOCK_IO_FOR_SUSPEND
1619 	 * and call flush_workqueue(md->wq). flush_workqueue will wait until
1620 	 * dm_wq_work exits and DMF_BLOCK_IO_FOR_SUSPEND will prevent any
1621 	 * further calls to __split_and_process_bio from dm_wq_work.
1622 	 */
1623 	down_write(&md->io_lock);
1624 	set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
1625 	set_bit(DMF_QUEUE_IO_TO_THREAD, &md->flags);
1626 	up_write(&md->io_lock);
1627 
1628 	flush_workqueue(md->wq);
1629 
1630 	/*
1631 	 * At this point no more requests are entering target request routines.
1632 	 * We call dm_wait_for_completion to wait for all existing requests
1633 	 * to finish.
1634 	 */
1635 	r = dm_wait_for_completion(md, TASK_INTERRUPTIBLE);
1636 
1637 	down_write(&md->io_lock);
1638 	if (noflush)
1639 		clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
1640 	up_write(&md->io_lock);
1641 
1642 	/* were we interrupted ? */
1643 	if (r < 0) {
1644 		dm_queue_flush(md);
1645 
1646 		unlock_fs(md);
1647 		goto out; /* pushback list is already flushed, so skip flush */
1648 	}
1649 
1650 	/*
1651 	 * If dm_wait_for_completion returned 0, the device is completely
1652 	 * quiescent now. There is no request-processing activity. All new
1653 	 * requests are being added to md->deferred list.
1654 	 */
1655 
1656 	dm_table_postsuspend_targets(map);
1657 
1658 	set_bit(DMF_SUSPENDED, &md->flags);
1659 
1660 out:
1661 	dm_table_put(map);
1662 
1663 out_unlock:
1664 	mutex_unlock(&md->suspend_lock);
1665 	return r;
1666 }
1667 
1668 int dm_resume(struct mapped_device *md)
1669 {
1670 	int r = -EINVAL;
1671 	struct dm_table *map = NULL;
1672 
1673 	mutex_lock(&md->suspend_lock);
1674 	if (!dm_suspended(md))
1675 		goto out;
1676 
1677 	map = dm_get_table(md);
1678 	if (!map || !dm_table_get_size(map))
1679 		goto out;
1680 
1681 	r = dm_table_resume_targets(map);
1682 	if (r)
1683 		goto out;
1684 
1685 	dm_queue_flush(md);
1686 
1687 	unlock_fs(md);
1688 
1689 	clear_bit(DMF_SUSPENDED, &md->flags);
1690 
1691 	dm_table_unplug_all(map);
1692 
1693 	dm_kobject_uevent(md);
1694 
1695 	r = 0;
1696 
1697 out:
1698 	dm_table_put(map);
1699 	mutex_unlock(&md->suspend_lock);
1700 
1701 	return r;
1702 }
1703 
1704 /*-----------------------------------------------------------------
1705  * Event notification.
1706  *---------------------------------------------------------------*/
1707 void dm_kobject_uevent(struct mapped_device *md)
1708 {
1709 	kobject_uevent(&disk_to_dev(md->disk)->kobj, KOBJ_CHANGE);
1710 }
1711 
1712 uint32_t dm_next_uevent_seq(struct mapped_device *md)
1713 {
1714 	return atomic_add_return(1, &md->uevent_seq);
1715 }
1716 
1717 uint32_t dm_get_event_nr(struct mapped_device *md)
1718 {
1719 	return atomic_read(&md->event_nr);
1720 }
1721 
1722 int dm_wait_event(struct mapped_device *md, int event_nr)
1723 {
1724 	return wait_event_interruptible(md->eventq,
1725 			(event_nr != atomic_read(&md->event_nr)));
1726 }
1727 
1728 void dm_uevent_add(struct mapped_device *md, struct list_head *elist)
1729 {
1730 	unsigned long flags;
1731 
1732 	spin_lock_irqsave(&md->uevent_lock, flags);
1733 	list_add(elist, &md->uevent_list);
1734 	spin_unlock_irqrestore(&md->uevent_lock, flags);
1735 }
1736 
1737 /*
1738  * The gendisk is only valid as long as you have a reference
1739  * count on 'md'.
1740  */
1741 struct gendisk *dm_disk(struct mapped_device *md)
1742 {
1743 	return md->disk;
1744 }
1745 
1746 struct kobject *dm_kobject(struct mapped_device *md)
1747 {
1748 	return &md->kobj;
1749 }
1750 
1751 /*
1752  * struct mapped_device should not be exported outside of dm.c
1753  * so use this check to verify that kobj is part of md structure
1754  */
1755 struct mapped_device *dm_get_from_kobject(struct kobject *kobj)
1756 {
1757 	struct mapped_device *md;
1758 
1759 	md = container_of(kobj, struct mapped_device, kobj);
1760 	if (&md->kobj != kobj)
1761 		return NULL;
1762 
1763 	if (test_bit(DMF_FREEING, &md->flags) ||
1764 	    test_bit(DMF_DELETING, &md->flags))
1765 		return NULL;
1766 
1767 	dm_get(md);
1768 	return md;
1769 }
1770 
1771 int dm_suspended(struct mapped_device *md)
1772 {
1773 	return test_bit(DMF_SUSPENDED, &md->flags);
1774 }
1775 
1776 int dm_noflush_suspending(struct dm_target *ti)
1777 {
1778 	struct mapped_device *md = dm_table_get_md(ti->table);
1779 	int r = __noflush_suspending(md);
1780 
1781 	dm_put(md);
1782 
1783 	return r;
1784 }
1785 EXPORT_SYMBOL_GPL(dm_noflush_suspending);
1786 
1787 static struct block_device_operations dm_blk_dops = {
1788 	.open = dm_blk_open,
1789 	.release = dm_blk_close,
1790 	.ioctl = dm_blk_ioctl,
1791 	.getgeo = dm_blk_getgeo,
1792 	.owner = THIS_MODULE
1793 };
1794 
1795 EXPORT_SYMBOL(dm_get_mapinfo);
1796 
1797 /*
1798  * module hooks
1799  */
1800 module_init(dm_init);
1801 module_exit(dm_exit);
1802 
1803 module_param(major, uint, 0);
1804 MODULE_PARM_DESC(major, "The major number of the device mapper");
1805 MODULE_DESCRIPTION(DM_NAME " driver");
1806 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
1807 MODULE_LICENSE("GPL");
1808