xref: /openbmc/linux/drivers/md/dm.c (revision 32a926da5a16c01a8213331e5764472ce2f14a8d)
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 				bio_list_add_head(&md->deferred, io->bio);
541 			else
542 				/* noflush suspend was interrupted. */
543 				io->error = -EIO;
544 			spin_unlock_irqrestore(&md->deferred_lock, flags);
545 		}
546 
547 		io_error = io->error;
548 		bio = io->bio;
549 
550 		if (bio_barrier(bio)) {
551 			/*
552 			 * There can be just one barrier request so we use
553 			 * a per-device variable for error reporting.
554 			 * Note that you can't touch the bio after end_io_acct
555 			 */
556 			md->barrier_error = io_error;
557 			end_io_acct(io);
558 		} else {
559 			end_io_acct(io);
560 
561 			if (io_error != DM_ENDIO_REQUEUE) {
562 				trace_block_bio_complete(md->queue, bio);
563 
564 				bio_endio(bio, io_error);
565 			}
566 		}
567 
568 		free_io(md, io);
569 	}
570 }
571 
572 static void clone_endio(struct bio *bio, int error)
573 {
574 	int r = 0;
575 	struct dm_target_io *tio = bio->bi_private;
576 	struct dm_io *io = tio->io;
577 	struct mapped_device *md = tio->io->md;
578 	dm_endio_fn endio = tio->ti->type->end_io;
579 
580 	if (!bio_flagged(bio, BIO_UPTODATE) && !error)
581 		error = -EIO;
582 
583 	if (endio) {
584 		r = endio(tio->ti, bio, error, &tio->info);
585 		if (r < 0 || r == DM_ENDIO_REQUEUE)
586 			/*
587 			 * error and requeue request are handled
588 			 * in dec_pending().
589 			 */
590 			error = r;
591 		else if (r == DM_ENDIO_INCOMPLETE)
592 			/* The target will handle the io */
593 			return;
594 		else if (r) {
595 			DMWARN("unimplemented target endio return value: %d", r);
596 			BUG();
597 		}
598 	}
599 
600 	/*
601 	 * Store md for cleanup instead of tio which is about to get freed.
602 	 */
603 	bio->bi_private = md->bs;
604 
605 	free_tio(md, tio);
606 	bio_put(bio);
607 	dec_pending(io, error);
608 }
609 
610 static sector_t max_io_len(struct mapped_device *md,
611 			   sector_t sector, struct dm_target *ti)
612 {
613 	sector_t offset = sector - ti->begin;
614 	sector_t len = ti->len - offset;
615 
616 	/*
617 	 * Does the target need to split even further ?
618 	 */
619 	if (ti->split_io) {
620 		sector_t boundary;
621 		boundary = ((offset + ti->split_io) & ~(ti->split_io - 1))
622 			   - offset;
623 		if (len > boundary)
624 			len = boundary;
625 	}
626 
627 	return len;
628 }
629 
630 static void __map_bio(struct dm_target *ti, struct bio *clone,
631 		      struct dm_target_io *tio)
632 {
633 	int r;
634 	sector_t sector;
635 	struct mapped_device *md;
636 
637 	/*
638 	 * Sanity checks.
639 	 */
640 	BUG_ON(!clone->bi_size);
641 
642 	clone->bi_end_io = clone_endio;
643 	clone->bi_private = tio;
644 
645 	/*
646 	 * Map the clone.  If r == 0 we don't need to do
647 	 * anything, the target has assumed ownership of
648 	 * this io.
649 	 */
650 	atomic_inc(&tio->io->io_count);
651 	sector = clone->bi_sector;
652 	r = ti->type->map(ti, clone, &tio->info);
653 	if (r == DM_MAPIO_REMAPPED) {
654 		/* the bio has been remapped so dispatch it */
655 
656 		trace_block_remap(bdev_get_queue(clone->bi_bdev), clone,
657 				    tio->io->bio->bi_bdev->bd_dev, sector);
658 
659 		generic_make_request(clone);
660 	} else if (r < 0 || r == DM_MAPIO_REQUEUE) {
661 		/* error the io and bail out, or requeue it if needed */
662 		md = tio->io->md;
663 		dec_pending(tio->io, r);
664 		/*
665 		 * Store bio_set for cleanup.
666 		 */
667 		clone->bi_private = md->bs;
668 		bio_put(clone);
669 		free_tio(md, tio);
670 	} else if (r) {
671 		DMWARN("unimplemented target map return value: %d", r);
672 		BUG();
673 	}
674 }
675 
676 struct clone_info {
677 	struct mapped_device *md;
678 	struct dm_table *map;
679 	struct bio *bio;
680 	struct dm_io *io;
681 	sector_t sector;
682 	sector_t sector_count;
683 	unsigned short idx;
684 };
685 
686 static void dm_bio_destructor(struct bio *bio)
687 {
688 	struct bio_set *bs = bio->bi_private;
689 
690 	bio_free(bio, bs);
691 }
692 
693 /*
694  * Creates a little bio that is just does part of a bvec.
695  */
696 static struct bio *split_bvec(struct bio *bio, sector_t sector,
697 			      unsigned short idx, unsigned int offset,
698 			      unsigned int len, struct bio_set *bs)
699 {
700 	struct bio *clone;
701 	struct bio_vec *bv = bio->bi_io_vec + idx;
702 
703 	clone = bio_alloc_bioset(GFP_NOIO, 1, bs);
704 	clone->bi_destructor = dm_bio_destructor;
705 	*clone->bi_io_vec = *bv;
706 
707 	clone->bi_sector = sector;
708 	clone->bi_bdev = bio->bi_bdev;
709 	clone->bi_rw = bio->bi_rw & ~(1 << BIO_RW_BARRIER);
710 	clone->bi_vcnt = 1;
711 	clone->bi_size = to_bytes(len);
712 	clone->bi_io_vec->bv_offset = offset;
713 	clone->bi_io_vec->bv_len = clone->bi_size;
714 	clone->bi_flags |= 1 << BIO_CLONED;
715 
716 	if (bio_integrity(bio)) {
717 		bio_integrity_clone(clone, bio, GFP_NOIO);
718 		bio_integrity_trim(clone,
719 				   bio_sector_offset(bio, idx, offset), len);
720 	}
721 
722 	return clone;
723 }
724 
725 /*
726  * Creates a bio that consists of range of complete bvecs.
727  */
728 static struct bio *clone_bio(struct bio *bio, sector_t sector,
729 			     unsigned short idx, unsigned short bv_count,
730 			     unsigned int len, struct bio_set *bs)
731 {
732 	struct bio *clone;
733 
734 	clone = bio_alloc_bioset(GFP_NOIO, bio->bi_max_vecs, bs);
735 	__bio_clone(clone, bio);
736 	clone->bi_rw &= ~(1 << BIO_RW_BARRIER);
737 	clone->bi_destructor = dm_bio_destructor;
738 	clone->bi_sector = sector;
739 	clone->bi_idx = idx;
740 	clone->bi_vcnt = idx + bv_count;
741 	clone->bi_size = to_bytes(len);
742 	clone->bi_flags &= ~(1 << BIO_SEG_VALID);
743 
744 	if (bio_integrity(bio)) {
745 		bio_integrity_clone(clone, bio, GFP_NOIO);
746 
747 		if (idx != bio->bi_idx || clone->bi_size < bio->bi_size)
748 			bio_integrity_trim(clone,
749 					   bio_sector_offset(bio, idx, 0), len);
750 	}
751 
752 	return clone;
753 }
754 
755 static int __clone_and_map(struct clone_info *ci)
756 {
757 	struct bio *clone, *bio = ci->bio;
758 	struct dm_target *ti;
759 	sector_t len = 0, max;
760 	struct dm_target_io *tio;
761 
762 	ti = dm_table_find_target(ci->map, ci->sector);
763 	if (!dm_target_is_valid(ti))
764 		return -EIO;
765 
766 	max = max_io_len(ci->md, ci->sector, ti);
767 
768 	/*
769 	 * Allocate a target io object.
770 	 */
771 	tio = alloc_tio(ci->md);
772 	tio->io = ci->io;
773 	tio->ti = ti;
774 	memset(&tio->info, 0, sizeof(tio->info));
775 
776 	if (ci->sector_count <= max) {
777 		/*
778 		 * Optimise for the simple case where we can do all of
779 		 * the remaining io with a single clone.
780 		 */
781 		clone = clone_bio(bio, ci->sector, ci->idx,
782 				  bio->bi_vcnt - ci->idx, ci->sector_count,
783 				  ci->md->bs);
784 		__map_bio(ti, clone, tio);
785 		ci->sector_count = 0;
786 
787 	} else if (to_sector(bio->bi_io_vec[ci->idx].bv_len) <= max) {
788 		/*
789 		 * There are some bvecs that don't span targets.
790 		 * Do as many of these as possible.
791 		 */
792 		int i;
793 		sector_t remaining = max;
794 		sector_t bv_len;
795 
796 		for (i = ci->idx; remaining && (i < bio->bi_vcnt); i++) {
797 			bv_len = to_sector(bio->bi_io_vec[i].bv_len);
798 
799 			if (bv_len > remaining)
800 				break;
801 
802 			remaining -= bv_len;
803 			len += bv_len;
804 		}
805 
806 		clone = clone_bio(bio, ci->sector, ci->idx, i - ci->idx, len,
807 				  ci->md->bs);
808 		__map_bio(ti, clone, tio);
809 
810 		ci->sector += len;
811 		ci->sector_count -= len;
812 		ci->idx = i;
813 
814 	} else {
815 		/*
816 		 * Handle a bvec that must be split between two or more targets.
817 		 */
818 		struct bio_vec *bv = bio->bi_io_vec + ci->idx;
819 		sector_t remaining = to_sector(bv->bv_len);
820 		unsigned int offset = 0;
821 
822 		do {
823 			if (offset) {
824 				ti = dm_table_find_target(ci->map, ci->sector);
825 				if (!dm_target_is_valid(ti))
826 					return -EIO;
827 
828 				max = max_io_len(ci->md, ci->sector, ti);
829 
830 				tio = alloc_tio(ci->md);
831 				tio->io = ci->io;
832 				tio->ti = ti;
833 				memset(&tio->info, 0, sizeof(tio->info));
834 			}
835 
836 			len = min(remaining, max);
837 
838 			clone = split_bvec(bio, ci->sector, ci->idx,
839 					   bv->bv_offset + offset, len,
840 					   ci->md->bs);
841 
842 			__map_bio(ti, clone, tio);
843 
844 			ci->sector += len;
845 			ci->sector_count -= len;
846 			offset += to_bytes(len);
847 		} while (remaining -= len);
848 
849 		ci->idx++;
850 	}
851 
852 	return 0;
853 }
854 
855 /*
856  * Split the bio into several clones and submit it to targets.
857  */
858 static void __split_and_process_bio(struct mapped_device *md, struct bio *bio)
859 {
860 	struct clone_info ci;
861 	int error = 0;
862 
863 	ci.map = dm_get_table(md);
864 	if (unlikely(!ci.map)) {
865 		if (!bio_barrier(bio))
866 			bio_io_error(bio);
867 		else
868 			md->barrier_error = -EIO;
869 		return;
870 	}
871 
872 	ci.md = md;
873 	ci.bio = bio;
874 	ci.io = alloc_io(md);
875 	ci.io->error = 0;
876 	atomic_set(&ci.io->io_count, 1);
877 	ci.io->bio = bio;
878 	ci.io->md = md;
879 	ci.sector = bio->bi_sector;
880 	ci.sector_count = bio_sectors(bio);
881 	ci.idx = bio->bi_idx;
882 
883 	start_io_acct(ci.io);
884 	while (ci.sector_count && !error)
885 		error = __clone_and_map(&ci);
886 
887 	/* drop the extra reference count */
888 	dec_pending(ci.io, error);
889 	dm_table_put(ci.map);
890 }
891 /*-----------------------------------------------------------------
892  * CRUD END
893  *---------------------------------------------------------------*/
894 
895 static int dm_merge_bvec(struct request_queue *q,
896 			 struct bvec_merge_data *bvm,
897 			 struct bio_vec *biovec)
898 {
899 	struct mapped_device *md = q->queuedata;
900 	struct dm_table *map = dm_get_table(md);
901 	struct dm_target *ti;
902 	sector_t max_sectors;
903 	int max_size = 0;
904 
905 	if (unlikely(!map))
906 		goto out;
907 
908 	ti = dm_table_find_target(map, bvm->bi_sector);
909 	if (!dm_target_is_valid(ti))
910 		goto out_table;
911 
912 	/*
913 	 * Find maximum amount of I/O that won't need splitting
914 	 */
915 	max_sectors = min(max_io_len(md, bvm->bi_sector, ti),
916 			  (sector_t) BIO_MAX_SECTORS);
917 	max_size = (max_sectors << SECTOR_SHIFT) - bvm->bi_size;
918 	if (max_size < 0)
919 		max_size = 0;
920 
921 	/*
922 	 * merge_bvec_fn() returns number of bytes
923 	 * it can accept at this offset
924 	 * max is precomputed maximal io size
925 	 */
926 	if (max_size && ti->type->merge)
927 		max_size = ti->type->merge(ti, bvm, biovec, max_size);
928 	/*
929 	 * If the target doesn't support merge method and some of the devices
930 	 * provided their merge_bvec method (we know this by looking at
931 	 * queue_max_hw_sectors), then we can't allow bios with multiple vector
932 	 * entries.  So always set max_size to 0, and the code below allows
933 	 * just one page.
934 	 */
935 	else if (queue_max_hw_sectors(q) <= PAGE_SIZE >> 9)
936 
937 		max_size = 0;
938 
939 out_table:
940 	dm_table_put(map);
941 
942 out:
943 	/*
944 	 * Always allow an entire first page
945 	 */
946 	if (max_size <= biovec->bv_len && !(bvm->bi_size >> SECTOR_SHIFT))
947 		max_size = biovec->bv_len;
948 
949 	return max_size;
950 }
951 
952 /*
953  * The request function that just remaps the bio built up by
954  * dm_merge_bvec.
955  */
956 static int dm_request(struct request_queue *q, struct bio *bio)
957 {
958 	int rw = bio_data_dir(bio);
959 	struct mapped_device *md = q->queuedata;
960 	int cpu;
961 
962 	down_read(&md->io_lock);
963 
964 	cpu = part_stat_lock();
965 	part_stat_inc(cpu, &dm_disk(md)->part0, ios[rw]);
966 	part_stat_add(cpu, &dm_disk(md)->part0, sectors[rw], bio_sectors(bio));
967 	part_stat_unlock();
968 
969 	/*
970 	 * If we're suspended or the thread is processing barriers
971 	 * we have to queue this io for later.
972 	 */
973 	if (unlikely(test_bit(DMF_QUEUE_IO_TO_THREAD, &md->flags)) ||
974 	    unlikely(bio_barrier(bio))) {
975 		up_read(&md->io_lock);
976 
977 		if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) &&
978 		    bio_rw(bio) == READA) {
979 			bio_io_error(bio);
980 			return 0;
981 		}
982 
983 		queue_io(md, bio);
984 
985 		return 0;
986 	}
987 
988 	__split_and_process_bio(md, bio);
989 	up_read(&md->io_lock);
990 	return 0;
991 }
992 
993 static void dm_unplug_all(struct request_queue *q)
994 {
995 	struct mapped_device *md = q->queuedata;
996 	struct dm_table *map = dm_get_table(md);
997 
998 	if (map) {
999 		dm_table_unplug_all(map);
1000 		dm_table_put(map);
1001 	}
1002 }
1003 
1004 static int dm_any_congested(void *congested_data, int bdi_bits)
1005 {
1006 	int r = bdi_bits;
1007 	struct mapped_device *md = congested_data;
1008 	struct dm_table *map;
1009 
1010 	if (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
1011 		map = dm_get_table(md);
1012 		if (map) {
1013 			r = dm_table_any_congested(map, bdi_bits);
1014 			dm_table_put(map);
1015 		}
1016 	}
1017 
1018 	return r;
1019 }
1020 
1021 /*-----------------------------------------------------------------
1022  * An IDR is used to keep track of allocated minor numbers.
1023  *---------------------------------------------------------------*/
1024 static DEFINE_IDR(_minor_idr);
1025 
1026 static void free_minor(int minor)
1027 {
1028 	spin_lock(&_minor_lock);
1029 	idr_remove(&_minor_idr, minor);
1030 	spin_unlock(&_minor_lock);
1031 }
1032 
1033 /*
1034  * See if the device with a specific minor # is free.
1035  */
1036 static int specific_minor(int minor)
1037 {
1038 	int r, m;
1039 
1040 	if (minor >= (1 << MINORBITS))
1041 		return -EINVAL;
1042 
1043 	r = idr_pre_get(&_minor_idr, GFP_KERNEL);
1044 	if (!r)
1045 		return -ENOMEM;
1046 
1047 	spin_lock(&_minor_lock);
1048 
1049 	if (idr_find(&_minor_idr, minor)) {
1050 		r = -EBUSY;
1051 		goto out;
1052 	}
1053 
1054 	r = idr_get_new_above(&_minor_idr, MINOR_ALLOCED, minor, &m);
1055 	if (r)
1056 		goto out;
1057 
1058 	if (m != minor) {
1059 		idr_remove(&_minor_idr, m);
1060 		r = -EBUSY;
1061 		goto out;
1062 	}
1063 
1064 out:
1065 	spin_unlock(&_minor_lock);
1066 	return r;
1067 }
1068 
1069 static int next_free_minor(int *minor)
1070 {
1071 	int r, m;
1072 
1073 	r = idr_pre_get(&_minor_idr, GFP_KERNEL);
1074 	if (!r)
1075 		return -ENOMEM;
1076 
1077 	spin_lock(&_minor_lock);
1078 
1079 	r = idr_get_new(&_minor_idr, MINOR_ALLOCED, &m);
1080 	if (r)
1081 		goto out;
1082 
1083 	if (m >= (1 << MINORBITS)) {
1084 		idr_remove(&_minor_idr, m);
1085 		r = -ENOSPC;
1086 		goto out;
1087 	}
1088 
1089 	*minor = m;
1090 
1091 out:
1092 	spin_unlock(&_minor_lock);
1093 	return r;
1094 }
1095 
1096 static struct block_device_operations dm_blk_dops;
1097 
1098 static void dm_wq_work(struct work_struct *work);
1099 
1100 /*
1101  * Allocate and initialise a blank device with a given minor.
1102  */
1103 static struct mapped_device *alloc_dev(int minor)
1104 {
1105 	int r;
1106 	struct mapped_device *md = kzalloc(sizeof(*md), GFP_KERNEL);
1107 	void *old_md;
1108 
1109 	if (!md) {
1110 		DMWARN("unable to allocate device, out of memory.");
1111 		return NULL;
1112 	}
1113 
1114 	if (!try_module_get(THIS_MODULE))
1115 		goto bad_module_get;
1116 
1117 	/* get a minor number for the dev */
1118 	if (minor == DM_ANY_MINOR)
1119 		r = next_free_minor(&minor);
1120 	else
1121 		r = specific_minor(minor);
1122 	if (r < 0)
1123 		goto bad_minor;
1124 
1125 	init_rwsem(&md->io_lock);
1126 	mutex_init(&md->suspend_lock);
1127 	spin_lock_init(&md->deferred_lock);
1128 	rwlock_init(&md->map_lock);
1129 	atomic_set(&md->holders, 1);
1130 	atomic_set(&md->open_count, 0);
1131 	atomic_set(&md->event_nr, 0);
1132 	atomic_set(&md->uevent_seq, 0);
1133 	INIT_LIST_HEAD(&md->uevent_list);
1134 	spin_lock_init(&md->uevent_lock);
1135 
1136 	md->queue = blk_alloc_queue(GFP_KERNEL);
1137 	if (!md->queue)
1138 		goto bad_queue;
1139 
1140 	md->queue->queuedata = md;
1141 	md->queue->backing_dev_info.congested_fn = dm_any_congested;
1142 	md->queue->backing_dev_info.congested_data = md;
1143 	blk_queue_make_request(md->queue, dm_request);
1144 	blk_queue_ordered(md->queue, QUEUE_ORDERED_DRAIN, NULL);
1145 	blk_queue_bounce_limit(md->queue, BLK_BOUNCE_ANY);
1146 	md->queue->unplug_fn = dm_unplug_all;
1147 	blk_queue_merge_bvec(md->queue, dm_merge_bvec);
1148 
1149 	md->io_pool = mempool_create_slab_pool(MIN_IOS, _io_cache);
1150 	if (!md->io_pool)
1151 		goto bad_io_pool;
1152 
1153 	md->tio_pool = mempool_create_slab_pool(MIN_IOS, _tio_cache);
1154 	if (!md->tio_pool)
1155 		goto bad_tio_pool;
1156 
1157 	md->bs = bioset_create(16, 0);
1158 	if (!md->bs)
1159 		goto bad_no_bioset;
1160 
1161 	md->disk = alloc_disk(1);
1162 	if (!md->disk)
1163 		goto bad_disk;
1164 
1165 	atomic_set(&md->pending, 0);
1166 	init_waitqueue_head(&md->wait);
1167 	INIT_WORK(&md->work, dm_wq_work);
1168 	init_waitqueue_head(&md->eventq);
1169 
1170 	md->disk->major = _major;
1171 	md->disk->first_minor = minor;
1172 	md->disk->fops = &dm_blk_dops;
1173 	md->disk->queue = md->queue;
1174 	md->disk->private_data = md;
1175 	sprintf(md->disk->disk_name, "dm-%d", minor);
1176 	add_disk(md->disk);
1177 	format_dev_t(md->name, MKDEV(_major, minor));
1178 
1179 	md->wq = create_singlethread_workqueue("kdmflush");
1180 	if (!md->wq)
1181 		goto bad_thread;
1182 
1183 	md->bdev = bdget_disk(md->disk, 0);
1184 	if (!md->bdev)
1185 		goto bad_bdev;
1186 
1187 	/* Populate the mapping, nobody knows we exist yet */
1188 	spin_lock(&_minor_lock);
1189 	old_md = idr_replace(&_minor_idr, md, minor);
1190 	spin_unlock(&_minor_lock);
1191 
1192 	BUG_ON(old_md != MINOR_ALLOCED);
1193 
1194 	return md;
1195 
1196 bad_bdev:
1197 	destroy_workqueue(md->wq);
1198 bad_thread:
1199 	put_disk(md->disk);
1200 bad_disk:
1201 	bioset_free(md->bs);
1202 bad_no_bioset:
1203 	mempool_destroy(md->tio_pool);
1204 bad_tio_pool:
1205 	mempool_destroy(md->io_pool);
1206 bad_io_pool:
1207 	blk_cleanup_queue(md->queue);
1208 bad_queue:
1209 	free_minor(minor);
1210 bad_minor:
1211 	module_put(THIS_MODULE);
1212 bad_module_get:
1213 	kfree(md);
1214 	return NULL;
1215 }
1216 
1217 static void unlock_fs(struct mapped_device *md);
1218 
1219 static void free_dev(struct mapped_device *md)
1220 {
1221 	int minor = MINOR(disk_devt(md->disk));
1222 
1223 	unlock_fs(md);
1224 	bdput(md->bdev);
1225 	destroy_workqueue(md->wq);
1226 	mempool_destroy(md->tio_pool);
1227 	mempool_destroy(md->io_pool);
1228 	bioset_free(md->bs);
1229 	blk_integrity_unregister(md->disk);
1230 	del_gendisk(md->disk);
1231 	free_minor(minor);
1232 
1233 	spin_lock(&_minor_lock);
1234 	md->disk->private_data = NULL;
1235 	spin_unlock(&_minor_lock);
1236 
1237 	put_disk(md->disk);
1238 	blk_cleanup_queue(md->queue);
1239 	module_put(THIS_MODULE);
1240 	kfree(md);
1241 }
1242 
1243 /*
1244  * Bind a table to the device.
1245  */
1246 static void event_callback(void *context)
1247 {
1248 	unsigned long flags;
1249 	LIST_HEAD(uevents);
1250 	struct mapped_device *md = (struct mapped_device *) context;
1251 
1252 	spin_lock_irqsave(&md->uevent_lock, flags);
1253 	list_splice_init(&md->uevent_list, &uevents);
1254 	spin_unlock_irqrestore(&md->uevent_lock, flags);
1255 
1256 	dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj);
1257 
1258 	atomic_inc(&md->event_nr);
1259 	wake_up(&md->eventq);
1260 }
1261 
1262 static void __set_size(struct mapped_device *md, sector_t size)
1263 {
1264 	set_capacity(md->disk, size);
1265 
1266 	mutex_lock(&md->bdev->bd_inode->i_mutex);
1267 	i_size_write(md->bdev->bd_inode, (loff_t)size << SECTOR_SHIFT);
1268 	mutex_unlock(&md->bdev->bd_inode->i_mutex);
1269 }
1270 
1271 static int __bind(struct mapped_device *md, struct dm_table *t)
1272 {
1273 	struct request_queue *q = md->queue;
1274 	sector_t size;
1275 
1276 	size = dm_table_get_size(t);
1277 
1278 	/*
1279 	 * Wipe any geometry if the size of the table changed.
1280 	 */
1281 	if (size != get_capacity(md->disk))
1282 		memset(&md->geometry, 0, sizeof(md->geometry));
1283 
1284 	__set_size(md, size);
1285 
1286 	if (!size) {
1287 		dm_table_destroy(t);
1288 		return 0;
1289 	}
1290 
1291 	dm_table_event_callback(t, event_callback, md);
1292 
1293 	write_lock(&md->map_lock);
1294 	md->map = t;
1295 	dm_table_set_restrictions(t, q);
1296 	write_unlock(&md->map_lock);
1297 
1298 	return 0;
1299 }
1300 
1301 static void __unbind(struct mapped_device *md)
1302 {
1303 	struct dm_table *map = md->map;
1304 
1305 	if (!map)
1306 		return;
1307 
1308 	dm_table_event_callback(map, NULL, NULL);
1309 	write_lock(&md->map_lock);
1310 	md->map = NULL;
1311 	write_unlock(&md->map_lock);
1312 	dm_table_destroy(map);
1313 }
1314 
1315 /*
1316  * Constructor for a new device.
1317  */
1318 int dm_create(int minor, struct mapped_device **result)
1319 {
1320 	struct mapped_device *md;
1321 
1322 	md = alloc_dev(minor);
1323 	if (!md)
1324 		return -ENXIO;
1325 
1326 	dm_sysfs_init(md);
1327 
1328 	*result = md;
1329 	return 0;
1330 }
1331 
1332 static struct mapped_device *dm_find_md(dev_t dev)
1333 {
1334 	struct mapped_device *md;
1335 	unsigned minor = MINOR(dev);
1336 
1337 	if (MAJOR(dev) != _major || minor >= (1 << MINORBITS))
1338 		return NULL;
1339 
1340 	spin_lock(&_minor_lock);
1341 
1342 	md = idr_find(&_minor_idr, minor);
1343 	if (md && (md == MINOR_ALLOCED ||
1344 		   (MINOR(disk_devt(dm_disk(md))) != minor) ||
1345 		   test_bit(DMF_FREEING, &md->flags))) {
1346 		md = NULL;
1347 		goto out;
1348 	}
1349 
1350 out:
1351 	spin_unlock(&_minor_lock);
1352 
1353 	return md;
1354 }
1355 
1356 struct mapped_device *dm_get_md(dev_t dev)
1357 {
1358 	struct mapped_device *md = dm_find_md(dev);
1359 
1360 	if (md)
1361 		dm_get(md);
1362 
1363 	return md;
1364 }
1365 
1366 void *dm_get_mdptr(struct mapped_device *md)
1367 {
1368 	return md->interface_ptr;
1369 }
1370 
1371 void dm_set_mdptr(struct mapped_device *md, void *ptr)
1372 {
1373 	md->interface_ptr = ptr;
1374 }
1375 
1376 void dm_get(struct mapped_device *md)
1377 {
1378 	atomic_inc(&md->holders);
1379 }
1380 
1381 const char *dm_device_name(struct mapped_device *md)
1382 {
1383 	return md->name;
1384 }
1385 EXPORT_SYMBOL_GPL(dm_device_name);
1386 
1387 void dm_put(struct mapped_device *md)
1388 {
1389 	struct dm_table *map;
1390 
1391 	BUG_ON(test_bit(DMF_FREEING, &md->flags));
1392 
1393 	if (atomic_dec_and_lock(&md->holders, &_minor_lock)) {
1394 		map = dm_get_table(md);
1395 		idr_replace(&_minor_idr, MINOR_ALLOCED,
1396 			    MINOR(disk_devt(dm_disk(md))));
1397 		set_bit(DMF_FREEING, &md->flags);
1398 		spin_unlock(&_minor_lock);
1399 		if (!dm_suspended(md)) {
1400 			dm_table_presuspend_targets(map);
1401 			dm_table_postsuspend_targets(map);
1402 		}
1403 		dm_sysfs_exit(md);
1404 		dm_table_put(map);
1405 		__unbind(md);
1406 		free_dev(md);
1407 	}
1408 }
1409 EXPORT_SYMBOL_GPL(dm_put);
1410 
1411 static int dm_wait_for_completion(struct mapped_device *md, int interruptible)
1412 {
1413 	int r = 0;
1414 	DECLARE_WAITQUEUE(wait, current);
1415 
1416 	dm_unplug_all(md->queue);
1417 
1418 	add_wait_queue(&md->wait, &wait);
1419 
1420 	while (1) {
1421 		set_current_state(interruptible);
1422 
1423 		smp_mb();
1424 		if (!atomic_read(&md->pending))
1425 			break;
1426 
1427 		if (interruptible == TASK_INTERRUPTIBLE &&
1428 		    signal_pending(current)) {
1429 			r = -EINTR;
1430 			break;
1431 		}
1432 
1433 		io_schedule();
1434 	}
1435 	set_current_state(TASK_RUNNING);
1436 
1437 	remove_wait_queue(&md->wait, &wait);
1438 
1439 	return r;
1440 }
1441 
1442 static int dm_flush(struct mapped_device *md)
1443 {
1444 	dm_wait_for_completion(md, TASK_UNINTERRUPTIBLE);
1445 	return 0;
1446 }
1447 
1448 static void process_barrier(struct mapped_device *md, struct bio *bio)
1449 {
1450 	int error = dm_flush(md);
1451 
1452 	if (unlikely(error)) {
1453 		bio_endio(bio, error);
1454 		return;
1455 	}
1456 	if (bio_empty_barrier(bio)) {
1457 		bio_endio(bio, 0);
1458 		return;
1459 	}
1460 
1461 	__split_and_process_bio(md, bio);
1462 
1463 	error = dm_flush(md);
1464 
1465 	if (!error && md->barrier_error)
1466 		error = md->barrier_error;
1467 
1468 	if (md->barrier_error != DM_ENDIO_REQUEUE)
1469 		bio_endio(bio, error);
1470 }
1471 
1472 /*
1473  * Process the deferred bios
1474  */
1475 static void dm_wq_work(struct work_struct *work)
1476 {
1477 	struct mapped_device *md = container_of(work, struct mapped_device,
1478 						work);
1479 	struct bio *c;
1480 
1481 	down_write(&md->io_lock);
1482 
1483 	while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
1484 		spin_lock_irq(&md->deferred_lock);
1485 		c = bio_list_pop(&md->deferred);
1486 		spin_unlock_irq(&md->deferred_lock);
1487 
1488 		if (!c) {
1489 			clear_bit(DMF_QUEUE_IO_TO_THREAD, &md->flags);
1490 			break;
1491 		}
1492 
1493 		up_write(&md->io_lock);
1494 
1495 		if (bio_barrier(c))
1496 			process_barrier(md, c);
1497 		else
1498 			__split_and_process_bio(md, c);
1499 
1500 		down_write(&md->io_lock);
1501 	}
1502 
1503 	up_write(&md->io_lock);
1504 }
1505 
1506 static void dm_queue_flush(struct mapped_device *md)
1507 {
1508 	clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
1509 	smp_mb__after_clear_bit();
1510 	queue_work(md->wq, &md->work);
1511 }
1512 
1513 /*
1514  * Swap in a new table (destroying old one).
1515  */
1516 int dm_swap_table(struct mapped_device *md, struct dm_table *table)
1517 {
1518 	int r = -EINVAL;
1519 
1520 	mutex_lock(&md->suspend_lock);
1521 
1522 	/* device must be suspended */
1523 	if (!dm_suspended(md))
1524 		goto out;
1525 
1526 	__unbind(md);
1527 	r = __bind(md, table);
1528 
1529 out:
1530 	mutex_unlock(&md->suspend_lock);
1531 	return r;
1532 }
1533 
1534 /*
1535  * Functions to lock and unlock any filesystem running on the
1536  * device.
1537  */
1538 static int lock_fs(struct mapped_device *md)
1539 {
1540 	int r;
1541 
1542 	WARN_ON(md->frozen_sb);
1543 
1544 	md->frozen_sb = freeze_bdev(md->bdev);
1545 	if (IS_ERR(md->frozen_sb)) {
1546 		r = PTR_ERR(md->frozen_sb);
1547 		md->frozen_sb = NULL;
1548 		return r;
1549 	}
1550 
1551 	set_bit(DMF_FROZEN, &md->flags);
1552 
1553 	return 0;
1554 }
1555 
1556 static void unlock_fs(struct mapped_device *md)
1557 {
1558 	if (!test_bit(DMF_FROZEN, &md->flags))
1559 		return;
1560 
1561 	thaw_bdev(md->bdev, md->frozen_sb);
1562 	md->frozen_sb = NULL;
1563 	clear_bit(DMF_FROZEN, &md->flags);
1564 }
1565 
1566 /*
1567  * We need to be able to change a mapping table under a mounted
1568  * filesystem.  For example we might want to move some data in
1569  * the background.  Before the table can be swapped with
1570  * dm_bind_table, dm_suspend must be called to flush any in
1571  * flight bios and ensure that any further io gets deferred.
1572  */
1573 int dm_suspend(struct mapped_device *md, unsigned suspend_flags)
1574 {
1575 	struct dm_table *map = NULL;
1576 	int r = 0;
1577 	int do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG ? 1 : 0;
1578 	int noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG ? 1 : 0;
1579 
1580 	mutex_lock(&md->suspend_lock);
1581 
1582 	if (dm_suspended(md)) {
1583 		r = -EINVAL;
1584 		goto out_unlock;
1585 	}
1586 
1587 	map = dm_get_table(md);
1588 
1589 	/*
1590 	 * DMF_NOFLUSH_SUSPENDING must be set before presuspend.
1591 	 * This flag is cleared before dm_suspend returns.
1592 	 */
1593 	if (noflush)
1594 		set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
1595 
1596 	/* This does not get reverted if there's an error later. */
1597 	dm_table_presuspend_targets(map);
1598 
1599 	/*
1600 	 * Flush I/O to the device. noflush supersedes do_lockfs,
1601 	 * because lock_fs() needs to flush I/Os.
1602 	 */
1603 	if (!noflush && do_lockfs) {
1604 		r = lock_fs(md);
1605 		if (r)
1606 			goto out;
1607 	}
1608 
1609 	/*
1610 	 * Here we must make sure that no processes are submitting requests
1611 	 * to target drivers i.e. no one may be executing
1612 	 * __split_and_process_bio. This is called from dm_request and
1613 	 * dm_wq_work.
1614 	 *
1615 	 * To get all processes out of __split_and_process_bio in dm_request,
1616 	 * we take the write lock. To prevent any process from reentering
1617 	 * __split_and_process_bio from dm_request, we set
1618 	 * DMF_QUEUE_IO_TO_THREAD.
1619 	 *
1620 	 * To quiesce the thread (dm_wq_work), we set DMF_BLOCK_IO_FOR_SUSPEND
1621 	 * and call flush_workqueue(md->wq). flush_workqueue will wait until
1622 	 * dm_wq_work exits and DMF_BLOCK_IO_FOR_SUSPEND will prevent any
1623 	 * further calls to __split_and_process_bio from dm_wq_work.
1624 	 */
1625 	down_write(&md->io_lock);
1626 	set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
1627 	set_bit(DMF_QUEUE_IO_TO_THREAD, &md->flags);
1628 	up_write(&md->io_lock);
1629 
1630 	flush_workqueue(md->wq);
1631 
1632 	/*
1633 	 * At this point no more requests are entering target request routines.
1634 	 * We call dm_wait_for_completion to wait for all existing requests
1635 	 * to finish.
1636 	 */
1637 	r = dm_wait_for_completion(md, TASK_INTERRUPTIBLE);
1638 
1639 	down_write(&md->io_lock);
1640 	if (noflush)
1641 		clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
1642 	up_write(&md->io_lock);
1643 
1644 	/* were we interrupted ? */
1645 	if (r < 0) {
1646 		dm_queue_flush(md);
1647 
1648 		unlock_fs(md);
1649 		goto out; /* pushback list is already flushed, so skip flush */
1650 	}
1651 
1652 	/*
1653 	 * If dm_wait_for_completion returned 0, the device is completely
1654 	 * quiescent now. There is no request-processing activity. All new
1655 	 * requests are being added to md->deferred list.
1656 	 */
1657 
1658 	dm_table_postsuspend_targets(map);
1659 
1660 	set_bit(DMF_SUSPENDED, &md->flags);
1661 
1662 out:
1663 	dm_table_put(map);
1664 
1665 out_unlock:
1666 	mutex_unlock(&md->suspend_lock);
1667 	return r;
1668 }
1669 
1670 int dm_resume(struct mapped_device *md)
1671 {
1672 	int r = -EINVAL;
1673 	struct dm_table *map = NULL;
1674 
1675 	mutex_lock(&md->suspend_lock);
1676 	if (!dm_suspended(md))
1677 		goto out;
1678 
1679 	map = dm_get_table(md);
1680 	if (!map || !dm_table_get_size(map))
1681 		goto out;
1682 
1683 	r = dm_table_resume_targets(map);
1684 	if (r)
1685 		goto out;
1686 
1687 	dm_queue_flush(md);
1688 
1689 	unlock_fs(md);
1690 
1691 	clear_bit(DMF_SUSPENDED, &md->flags);
1692 
1693 	dm_table_unplug_all(map);
1694 
1695 	dm_kobject_uevent(md);
1696 
1697 	r = 0;
1698 
1699 out:
1700 	dm_table_put(map);
1701 	mutex_unlock(&md->suspend_lock);
1702 
1703 	return r;
1704 }
1705 
1706 /*-----------------------------------------------------------------
1707  * Event notification.
1708  *---------------------------------------------------------------*/
1709 void dm_kobject_uevent(struct mapped_device *md)
1710 {
1711 	kobject_uevent(&disk_to_dev(md->disk)->kobj, KOBJ_CHANGE);
1712 }
1713 
1714 uint32_t dm_next_uevent_seq(struct mapped_device *md)
1715 {
1716 	return atomic_add_return(1, &md->uevent_seq);
1717 }
1718 
1719 uint32_t dm_get_event_nr(struct mapped_device *md)
1720 {
1721 	return atomic_read(&md->event_nr);
1722 }
1723 
1724 int dm_wait_event(struct mapped_device *md, int event_nr)
1725 {
1726 	return wait_event_interruptible(md->eventq,
1727 			(event_nr != atomic_read(&md->event_nr)));
1728 }
1729 
1730 void dm_uevent_add(struct mapped_device *md, struct list_head *elist)
1731 {
1732 	unsigned long flags;
1733 
1734 	spin_lock_irqsave(&md->uevent_lock, flags);
1735 	list_add(elist, &md->uevent_list);
1736 	spin_unlock_irqrestore(&md->uevent_lock, flags);
1737 }
1738 
1739 /*
1740  * The gendisk is only valid as long as you have a reference
1741  * count on 'md'.
1742  */
1743 struct gendisk *dm_disk(struct mapped_device *md)
1744 {
1745 	return md->disk;
1746 }
1747 
1748 struct kobject *dm_kobject(struct mapped_device *md)
1749 {
1750 	return &md->kobj;
1751 }
1752 
1753 /*
1754  * struct mapped_device should not be exported outside of dm.c
1755  * so use this check to verify that kobj is part of md structure
1756  */
1757 struct mapped_device *dm_get_from_kobject(struct kobject *kobj)
1758 {
1759 	struct mapped_device *md;
1760 
1761 	md = container_of(kobj, struct mapped_device, kobj);
1762 	if (&md->kobj != kobj)
1763 		return NULL;
1764 
1765 	if (test_bit(DMF_FREEING, &md->flags) ||
1766 	    test_bit(DMF_DELETING, &md->flags))
1767 		return NULL;
1768 
1769 	dm_get(md);
1770 	return md;
1771 }
1772 
1773 int dm_suspended(struct mapped_device *md)
1774 {
1775 	return test_bit(DMF_SUSPENDED, &md->flags);
1776 }
1777 
1778 int dm_noflush_suspending(struct dm_target *ti)
1779 {
1780 	struct mapped_device *md = dm_table_get_md(ti->table);
1781 	int r = __noflush_suspending(md);
1782 
1783 	dm_put(md);
1784 
1785 	return r;
1786 }
1787 EXPORT_SYMBOL_GPL(dm_noflush_suspending);
1788 
1789 static struct block_device_operations dm_blk_dops = {
1790 	.open = dm_blk_open,
1791 	.release = dm_blk_close,
1792 	.ioctl = dm_blk_ioctl,
1793 	.getgeo = dm_blk_getgeo,
1794 	.owner = THIS_MODULE
1795 };
1796 
1797 EXPORT_SYMBOL(dm_get_mapinfo);
1798 
1799 /*
1800  * module hooks
1801  */
1802 module_init(dm_init);
1803 module_exit(dm_exit);
1804 
1805 module_param(major, uint, 0);
1806 MODULE_PARM_DESC(major, "The major number of the device mapper");
1807 MODULE_DESCRIPTION(DM_NAME " driver");
1808 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
1809 MODULE_LICENSE("GPL");
1810