xref: /linux/drivers/md/dm.c (revision 606d099cdd1080bbb50ea50dc52d98252f8f10a1)
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 
11 #include <linux/init.h>
12 #include <linux/module.h>
13 #include <linux/mutex.h>
14 #include <linux/moduleparam.h>
15 #include <linux/blkpg.h>
16 #include <linux/bio.h>
17 #include <linux/buffer_head.h>
18 #include <linux/mempool.h>
19 #include <linux/slab.h>
20 #include <linux/idr.h>
21 #include <linux/hdreg.h>
22 #include <linux/blktrace_api.h>
23 #include <linux/smp_lock.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 	struct bio *bio;
40 	atomic_t io_count;
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 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 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 
72 struct mapped_device {
73 	struct rw_semaphore io_lock;
74 	struct semaphore suspend_lock;
75 	rwlock_t map_lock;
76 	atomic_t holders;
77 	atomic_t open_count;
78 
79 	unsigned long flags;
80 
81 	request_queue_t *queue;
82 	struct gendisk *disk;
83 	char name[16];
84 
85 	void *interface_ptr;
86 
87 	/*
88 	 * A list of ios that arrived while we were suspended.
89 	 */
90 	atomic_t pending;
91 	wait_queue_head_t wait;
92  	struct bio_list deferred;
93 
94 	/*
95 	 * The current mapping.
96 	 */
97 	struct dm_table *map;
98 
99 	/*
100 	 * io objects are allocated from here.
101 	 */
102 	mempool_t *io_pool;
103 	mempool_t *tio_pool;
104 
105 	struct bio_set *bs;
106 
107 	/*
108 	 * Event handling.
109 	 */
110 	atomic_t event_nr;
111 	wait_queue_head_t eventq;
112 
113 	/*
114 	 * freeze/thaw support require holding onto a super block
115 	 */
116 	struct super_block *frozen_sb;
117 	struct block_device *suspended_bdev;
118 
119 	/* forced geometry settings */
120 	struct hd_geometry geometry;
121 };
122 
123 #define MIN_IOS 256
124 static struct kmem_cache *_io_cache;
125 static struct kmem_cache *_tio_cache;
126 
127 static int __init local_init(void)
128 {
129 	int r;
130 
131 	/* allocate a slab for the dm_ios */
132 	_io_cache = kmem_cache_create("dm_io",
133 				      sizeof(struct dm_io), 0, 0, NULL, NULL);
134 	if (!_io_cache)
135 		return -ENOMEM;
136 
137 	/* allocate a slab for the target ios */
138 	_tio_cache = kmem_cache_create("dm_tio", sizeof(struct target_io),
139 				       0, 0, NULL, NULL);
140 	if (!_tio_cache) {
141 		kmem_cache_destroy(_io_cache);
142 		return -ENOMEM;
143 	}
144 
145 	_major = major;
146 	r = register_blkdev(_major, _name);
147 	if (r < 0) {
148 		kmem_cache_destroy(_tio_cache);
149 		kmem_cache_destroy(_io_cache);
150 		return r;
151 	}
152 
153 	if (!_major)
154 		_major = r;
155 
156 	return 0;
157 }
158 
159 static void local_exit(void)
160 {
161 	kmem_cache_destroy(_tio_cache);
162 	kmem_cache_destroy(_io_cache);
163 
164 	if (unregister_blkdev(_major, _name) < 0)
165 		DMERR("unregister_blkdev failed");
166 
167 	_major = 0;
168 
169 	DMINFO("cleaned up");
170 }
171 
172 int (*_inits[])(void) __initdata = {
173 	local_init,
174 	dm_target_init,
175 	dm_linear_init,
176 	dm_stripe_init,
177 	dm_interface_init,
178 };
179 
180 void (*_exits[])(void) = {
181 	local_exit,
182 	dm_target_exit,
183 	dm_linear_exit,
184 	dm_stripe_exit,
185 	dm_interface_exit,
186 };
187 
188 static int __init dm_init(void)
189 {
190 	const int count = ARRAY_SIZE(_inits);
191 
192 	int r, i;
193 
194 	for (i = 0; i < count; i++) {
195 		r = _inits[i]();
196 		if (r)
197 			goto bad;
198 	}
199 
200 	return 0;
201 
202       bad:
203 	while (i--)
204 		_exits[i]();
205 
206 	return r;
207 }
208 
209 static void __exit dm_exit(void)
210 {
211 	int i = ARRAY_SIZE(_exits);
212 
213 	while (i--)
214 		_exits[i]();
215 }
216 
217 /*
218  * Block device functions
219  */
220 static int dm_blk_open(struct inode *inode, struct file *file)
221 {
222 	struct mapped_device *md;
223 
224 	spin_lock(&_minor_lock);
225 
226 	md = inode->i_bdev->bd_disk->private_data;
227 	if (!md)
228 		goto out;
229 
230 	if (test_bit(DMF_FREEING, &md->flags) ||
231 	    test_bit(DMF_DELETING, &md->flags)) {
232 		md = NULL;
233 		goto out;
234 	}
235 
236 	dm_get(md);
237 	atomic_inc(&md->open_count);
238 
239 out:
240 	spin_unlock(&_minor_lock);
241 
242 	return md ? 0 : -ENXIO;
243 }
244 
245 static int dm_blk_close(struct inode *inode, struct file *file)
246 {
247 	struct mapped_device *md;
248 
249 	md = inode->i_bdev->bd_disk->private_data;
250 	atomic_dec(&md->open_count);
251 	dm_put(md);
252 	return 0;
253 }
254 
255 int dm_open_count(struct mapped_device *md)
256 {
257 	return atomic_read(&md->open_count);
258 }
259 
260 /*
261  * Guarantees nothing is using the device before it's deleted.
262  */
263 int dm_lock_for_deletion(struct mapped_device *md)
264 {
265 	int r = 0;
266 
267 	spin_lock(&_minor_lock);
268 
269 	if (dm_open_count(md))
270 		r = -EBUSY;
271 	else
272 		set_bit(DMF_DELETING, &md->flags);
273 
274 	spin_unlock(&_minor_lock);
275 
276 	return r;
277 }
278 
279 static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo)
280 {
281 	struct mapped_device *md = bdev->bd_disk->private_data;
282 
283 	return dm_get_geometry(md, geo);
284 }
285 
286 static int dm_blk_ioctl(struct inode *inode, struct file *file,
287 			unsigned int cmd, unsigned long arg)
288 {
289 	struct mapped_device *md;
290 	struct dm_table *map;
291 	struct dm_target *tgt;
292 	int r = -ENOTTY;
293 
294 	/* We don't really need this lock, but we do need 'inode'. */
295 	unlock_kernel();
296 
297 	md = inode->i_bdev->bd_disk->private_data;
298 
299 	map = dm_get_table(md);
300 
301 	if (!map || !dm_table_get_size(map))
302 		goto out;
303 
304 	/* We only support devices that have a single target */
305 	if (dm_table_get_num_targets(map) != 1)
306 		goto out;
307 
308 	tgt = dm_table_get_target(map, 0);
309 
310 	if (dm_suspended(md)) {
311 		r = -EAGAIN;
312 		goto out;
313 	}
314 
315 	if (tgt->type->ioctl)
316 		r = tgt->type->ioctl(tgt, inode, file, cmd, arg);
317 
318 out:
319 	dm_table_put(map);
320 
321 	lock_kernel();
322 	return r;
323 }
324 
325 static inline struct dm_io *alloc_io(struct mapped_device *md)
326 {
327 	return mempool_alloc(md->io_pool, GFP_NOIO);
328 }
329 
330 static inline void free_io(struct mapped_device *md, struct dm_io *io)
331 {
332 	mempool_free(io, md->io_pool);
333 }
334 
335 static inline struct target_io *alloc_tio(struct mapped_device *md)
336 {
337 	return mempool_alloc(md->tio_pool, GFP_NOIO);
338 }
339 
340 static inline void free_tio(struct mapped_device *md, struct target_io *tio)
341 {
342 	mempool_free(tio, md->tio_pool);
343 }
344 
345 static void start_io_acct(struct dm_io *io)
346 {
347 	struct mapped_device *md = io->md;
348 
349 	io->start_time = jiffies;
350 
351 	preempt_disable();
352 	disk_round_stats(dm_disk(md));
353 	preempt_enable();
354 	dm_disk(md)->in_flight = atomic_inc_return(&md->pending);
355 }
356 
357 static int end_io_acct(struct dm_io *io)
358 {
359 	struct mapped_device *md = io->md;
360 	struct bio *bio = io->bio;
361 	unsigned long duration = jiffies - io->start_time;
362 	int pending;
363 	int rw = bio_data_dir(bio);
364 
365 	preempt_disable();
366 	disk_round_stats(dm_disk(md));
367 	preempt_enable();
368 	dm_disk(md)->in_flight = pending = atomic_dec_return(&md->pending);
369 
370 	disk_stat_add(dm_disk(md), ticks[rw], duration);
371 
372 	return !pending;
373 }
374 
375 /*
376  * Add the bio to the list of deferred io.
377  */
378 static int queue_io(struct mapped_device *md, struct bio *bio)
379 {
380 	down_write(&md->io_lock);
381 
382 	if (!test_bit(DMF_BLOCK_IO, &md->flags)) {
383 		up_write(&md->io_lock);
384 		return 1;
385 	}
386 
387 	bio_list_add(&md->deferred, bio);
388 
389 	up_write(&md->io_lock);
390 	return 0;		/* deferred successfully */
391 }
392 
393 /*
394  * Everyone (including functions in this file), should use this
395  * function to access the md->map field, and make sure they call
396  * dm_table_put() when finished.
397  */
398 struct dm_table *dm_get_table(struct mapped_device *md)
399 {
400 	struct dm_table *t;
401 
402 	read_lock(&md->map_lock);
403 	t = md->map;
404 	if (t)
405 		dm_table_get(t);
406 	read_unlock(&md->map_lock);
407 
408 	return t;
409 }
410 
411 /*
412  * Get the geometry associated with a dm device
413  */
414 int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo)
415 {
416 	*geo = md->geometry;
417 
418 	return 0;
419 }
420 
421 /*
422  * Set the geometry of a device.
423  */
424 int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo)
425 {
426 	sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors;
427 
428 	if (geo->start > sz) {
429 		DMWARN("Start sector is beyond the geometry limits.");
430 		return -EINVAL;
431 	}
432 
433 	md->geometry = *geo;
434 
435 	return 0;
436 }
437 
438 /*-----------------------------------------------------------------
439  * CRUD START:
440  *   A more elegant soln is in the works that uses the queue
441  *   merge fn, unfortunately there are a couple of changes to
442  *   the block layer that I want to make for this.  So in the
443  *   interests of getting something for people to use I give
444  *   you this clearly demarcated crap.
445  *---------------------------------------------------------------*/
446 
447 /*
448  * Decrements the number of outstanding ios that a bio has been
449  * cloned into, completing the original io if necc.
450  */
451 static void dec_pending(struct dm_io *io, int error)
452 {
453 	if (error)
454 		io->error = error;
455 
456 	if (atomic_dec_and_test(&io->io_count)) {
457 		if (end_io_acct(io))
458 			/* nudge anyone waiting on suspend queue */
459 			wake_up(&io->md->wait);
460 
461 		blk_add_trace_bio(io->md->queue, io->bio, BLK_TA_COMPLETE);
462 
463 		bio_endio(io->bio, io->bio->bi_size, io->error);
464 		free_io(io->md, io);
465 	}
466 }
467 
468 static int clone_endio(struct bio *bio, unsigned int done, int error)
469 {
470 	int r = 0;
471 	struct target_io *tio = bio->bi_private;
472 	struct mapped_device *md = tio->io->md;
473 	dm_endio_fn endio = tio->ti->type->end_io;
474 
475 	if (bio->bi_size)
476 		return 1;
477 
478 	if (!bio_flagged(bio, BIO_UPTODATE) && !error)
479 		error = -EIO;
480 
481 	if (endio) {
482 		r = endio(tio->ti, bio, error, &tio->info);
483 		if (r < 0)
484 			error = r;
485 
486 		else if (r > 0)
487 			/* the target wants another shot at the io */
488 			return 1;
489 	}
490 
491 	dec_pending(tio->io, error);
492 
493 	/*
494 	 * Store md for cleanup instead of tio which is about to get freed.
495 	 */
496 	bio->bi_private = md->bs;
497 
498 	bio_put(bio);
499 	free_tio(md, tio);
500 	return r;
501 }
502 
503 static sector_t max_io_len(struct mapped_device *md,
504 			   sector_t sector, struct dm_target *ti)
505 {
506 	sector_t offset = sector - ti->begin;
507 	sector_t len = ti->len - offset;
508 
509 	/*
510 	 * Does the target need to split even further ?
511 	 */
512 	if (ti->split_io) {
513 		sector_t boundary;
514 		boundary = ((offset + ti->split_io) & ~(ti->split_io - 1))
515 			   - offset;
516 		if (len > boundary)
517 			len = boundary;
518 	}
519 
520 	return len;
521 }
522 
523 static void __map_bio(struct dm_target *ti, struct bio *clone,
524 		      struct target_io *tio)
525 {
526 	int r;
527 	sector_t sector;
528 	struct mapped_device *md;
529 
530 	/*
531 	 * Sanity checks.
532 	 */
533 	BUG_ON(!clone->bi_size);
534 
535 	clone->bi_end_io = clone_endio;
536 	clone->bi_private = tio;
537 
538 	/*
539 	 * Map the clone.  If r == 0 we don't need to do
540 	 * anything, the target has assumed ownership of
541 	 * this io.
542 	 */
543 	atomic_inc(&tio->io->io_count);
544 	sector = clone->bi_sector;
545 	r = ti->type->map(ti, clone, &tio->info);
546 	if (r > 0) {
547 		/* the bio has been remapped so dispatch it */
548 
549 		blk_add_trace_remap(bdev_get_queue(clone->bi_bdev), clone,
550 				    tio->io->bio->bi_bdev->bd_dev, sector,
551 				    clone->bi_sector);
552 
553 		generic_make_request(clone);
554 	}
555 
556 	else if (r < 0) {
557 		/* error the io and bail out */
558 		md = tio->io->md;
559 		dec_pending(tio->io, r);
560 		/*
561 		 * Store bio_set for cleanup.
562 		 */
563 		clone->bi_private = md->bs;
564 		bio_put(clone);
565 		free_tio(md, tio);
566 	}
567 }
568 
569 struct clone_info {
570 	struct mapped_device *md;
571 	struct dm_table *map;
572 	struct bio *bio;
573 	struct dm_io *io;
574 	sector_t sector;
575 	sector_t sector_count;
576 	unsigned short idx;
577 };
578 
579 static void dm_bio_destructor(struct bio *bio)
580 {
581 	struct bio_set *bs = bio->bi_private;
582 
583 	bio_free(bio, bs);
584 }
585 
586 /*
587  * Creates a little bio that is just does part of a bvec.
588  */
589 static struct bio *split_bvec(struct bio *bio, sector_t sector,
590 			      unsigned short idx, unsigned int offset,
591 			      unsigned int len, struct bio_set *bs)
592 {
593 	struct bio *clone;
594 	struct bio_vec *bv = bio->bi_io_vec + idx;
595 
596 	clone = bio_alloc_bioset(GFP_NOIO, 1, bs);
597 	clone->bi_destructor = dm_bio_destructor;
598 	*clone->bi_io_vec = *bv;
599 
600 	clone->bi_sector = sector;
601 	clone->bi_bdev = bio->bi_bdev;
602 	clone->bi_rw = bio->bi_rw;
603 	clone->bi_vcnt = 1;
604 	clone->bi_size = to_bytes(len);
605 	clone->bi_io_vec->bv_offset = offset;
606 	clone->bi_io_vec->bv_len = clone->bi_size;
607 
608 	return clone;
609 }
610 
611 /*
612  * Creates a bio that consists of range of complete bvecs.
613  */
614 static struct bio *clone_bio(struct bio *bio, sector_t sector,
615 			     unsigned short idx, unsigned short bv_count,
616 			     unsigned int len, struct bio_set *bs)
617 {
618 	struct bio *clone;
619 
620 	clone = bio_alloc_bioset(GFP_NOIO, bio->bi_max_vecs, bs);
621 	__bio_clone(clone, bio);
622 	clone->bi_destructor = dm_bio_destructor;
623 	clone->bi_sector = sector;
624 	clone->bi_idx = idx;
625 	clone->bi_vcnt = idx + bv_count;
626 	clone->bi_size = to_bytes(len);
627 	clone->bi_flags &= ~(1 << BIO_SEG_VALID);
628 
629 	return clone;
630 }
631 
632 static void __clone_and_map(struct clone_info *ci)
633 {
634 	struct bio *clone, *bio = ci->bio;
635 	struct dm_target *ti = dm_table_find_target(ci->map, ci->sector);
636 	sector_t len = 0, max = max_io_len(ci->md, ci->sector, ti);
637 	struct target_io *tio;
638 
639 	/*
640 	 * Allocate a target io object.
641 	 */
642 	tio = alloc_tio(ci->md);
643 	tio->io = ci->io;
644 	tio->ti = ti;
645 	memset(&tio->info, 0, sizeof(tio->info));
646 
647 	if (ci->sector_count <= max) {
648 		/*
649 		 * Optimise for the simple case where we can do all of
650 		 * the remaining io with a single clone.
651 		 */
652 		clone = clone_bio(bio, ci->sector, ci->idx,
653 				  bio->bi_vcnt - ci->idx, ci->sector_count,
654 				  ci->md->bs);
655 		__map_bio(ti, clone, tio);
656 		ci->sector_count = 0;
657 
658 	} else if (to_sector(bio->bi_io_vec[ci->idx].bv_len) <= max) {
659 		/*
660 		 * There are some bvecs that don't span targets.
661 		 * Do as many of these as possible.
662 		 */
663 		int i;
664 		sector_t remaining = max;
665 		sector_t bv_len;
666 
667 		for (i = ci->idx; remaining && (i < bio->bi_vcnt); i++) {
668 			bv_len = to_sector(bio->bi_io_vec[i].bv_len);
669 
670 			if (bv_len > remaining)
671 				break;
672 
673 			remaining -= bv_len;
674 			len += bv_len;
675 		}
676 
677 		clone = clone_bio(bio, ci->sector, ci->idx, i - ci->idx, len,
678 				  ci->md->bs);
679 		__map_bio(ti, clone, tio);
680 
681 		ci->sector += len;
682 		ci->sector_count -= len;
683 		ci->idx = i;
684 
685 	} else {
686 		/*
687 		 * Handle a bvec that must be split between two or more targets.
688 		 */
689 		struct bio_vec *bv = bio->bi_io_vec + ci->idx;
690 		sector_t remaining = to_sector(bv->bv_len);
691 		unsigned int offset = 0;
692 
693 		do {
694 			if (offset) {
695 				ti = dm_table_find_target(ci->map, ci->sector);
696 				max = max_io_len(ci->md, ci->sector, ti);
697 
698 				tio = alloc_tio(ci->md);
699 				tio->io = ci->io;
700 				tio->ti = ti;
701 				memset(&tio->info, 0, sizeof(tio->info));
702 			}
703 
704 			len = min(remaining, max);
705 
706 			clone = split_bvec(bio, ci->sector, ci->idx,
707 					   bv->bv_offset + offset, len,
708 					   ci->md->bs);
709 
710 			__map_bio(ti, clone, tio);
711 
712 			ci->sector += len;
713 			ci->sector_count -= len;
714 			offset += to_bytes(len);
715 		} while (remaining -= len);
716 
717 		ci->idx++;
718 	}
719 }
720 
721 /*
722  * Split the bio into several clones.
723  */
724 static void __split_bio(struct mapped_device *md, struct bio *bio)
725 {
726 	struct clone_info ci;
727 
728 	ci.map = dm_get_table(md);
729 	if (!ci.map) {
730 		bio_io_error(bio, bio->bi_size);
731 		return;
732 	}
733 
734 	ci.md = md;
735 	ci.bio = bio;
736 	ci.io = alloc_io(md);
737 	ci.io->error = 0;
738 	atomic_set(&ci.io->io_count, 1);
739 	ci.io->bio = bio;
740 	ci.io->md = md;
741 	ci.sector = bio->bi_sector;
742 	ci.sector_count = bio_sectors(bio);
743 	ci.idx = bio->bi_idx;
744 
745 	start_io_acct(ci.io);
746 	while (ci.sector_count)
747 		__clone_and_map(&ci);
748 
749 	/* drop the extra reference count */
750 	dec_pending(ci.io, 0);
751 	dm_table_put(ci.map);
752 }
753 /*-----------------------------------------------------------------
754  * CRUD END
755  *---------------------------------------------------------------*/
756 
757 /*
758  * The request function that just remaps the bio built up by
759  * dm_merge_bvec.
760  */
761 static int dm_request(request_queue_t *q, struct bio *bio)
762 {
763 	int r;
764 	int rw = bio_data_dir(bio);
765 	struct mapped_device *md = q->queuedata;
766 
767 	down_read(&md->io_lock);
768 
769 	disk_stat_inc(dm_disk(md), ios[rw]);
770 	disk_stat_add(dm_disk(md), sectors[rw], bio_sectors(bio));
771 
772 	/*
773 	 * If we're suspended we have to queue
774 	 * this io for later.
775 	 */
776 	while (test_bit(DMF_BLOCK_IO, &md->flags)) {
777 		up_read(&md->io_lock);
778 
779 		if (bio_rw(bio) == READA) {
780 			bio_io_error(bio, bio->bi_size);
781 			return 0;
782 		}
783 
784 		r = queue_io(md, bio);
785 		if (r < 0) {
786 			bio_io_error(bio, bio->bi_size);
787 			return 0;
788 
789 		} else if (r == 0)
790 			return 0;	/* deferred successfully */
791 
792 		/*
793 		 * We're in a while loop, because someone could suspend
794 		 * before we get to the following read lock.
795 		 */
796 		down_read(&md->io_lock);
797 	}
798 
799 	__split_bio(md, bio);
800 	up_read(&md->io_lock);
801 	return 0;
802 }
803 
804 static int dm_flush_all(request_queue_t *q, struct gendisk *disk,
805 			sector_t *error_sector)
806 {
807 	struct mapped_device *md = q->queuedata;
808 	struct dm_table *map = dm_get_table(md);
809 	int ret = -ENXIO;
810 
811 	if (map) {
812 		ret = dm_table_flush_all(map);
813 		dm_table_put(map);
814 	}
815 
816 	return ret;
817 }
818 
819 static void dm_unplug_all(request_queue_t *q)
820 {
821 	struct mapped_device *md = q->queuedata;
822 	struct dm_table *map = dm_get_table(md);
823 
824 	if (map) {
825 		dm_table_unplug_all(map);
826 		dm_table_put(map);
827 	}
828 }
829 
830 static int dm_any_congested(void *congested_data, int bdi_bits)
831 {
832 	int r;
833 	struct mapped_device *md = (struct mapped_device *) congested_data;
834 	struct dm_table *map = dm_get_table(md);
835 
836 	if (!map || test_bit(DMF_BLOCK_IO, &md->flags))
837 		r = bdi_bits;
838 	else
839 		r = dm_table_any_congested(map, bdi_bits);
840 
841 	dm_table_put(map);
842 	return r;
843 }
844 
845 /*-----------------------------------------------------------------
846  * An IDR is used to keep track of allocated minor numbers.
847  *---------------------------------------------------------------*/
848 static DEFINE_IDR(_minor_idr);
849 
850 static void free_minor(int minor)
851 {
852 	spin_lock(&_minor_lock);
853 	idr_remove(&_minor_idr, minor);
854 	spin_unlock(&_minor_lock);
855 }
856 
857 /*
858  * See if the device with a specific minor # is free.
859  */
860 static int specific_minor(struct mapped_device *md, int minor)
861 {
862 	int r, m;
863 
864 	if (minor >= (1 << MINORBITS))
865 		return -EINVAL;
866 
867 	r = idr_pre_get(&_minor_idr, GFP_KERNEL);
868 	if (!r)
869 		return -ENOMEM;
870 
871 	spin_lock(&_minor_lock);
872 
873 	if (idr_find(&_minor_idr, minor)) {
874 		r = -EBUSY;
875 		goto out;
876 	}
877 
878 	r = idr_get_new_above(&_minor_idr, MINOR_ALLOCED, minor, &m);
879 	if (r)
880 		goto out;
881 
882 	if (m != minor) {
883 		idr_remove(&_minor_idr, m);
884 		r = -EBUSY;
885 		goto out;
886 	}
887 
888 out:
889 	spin_unlock(&_minor_lock);
890 	return r;
891 }
892 
893 static int next_free_minor(struct mapped_device *md, int *minor)
894 {
895 	int r, m;
896 
897 	r = idr_pre_get(&_minor_idr, GFP_KERNEL);
898 	if (!r)
899 		return -ENOMEM;
900 
901 	spin_lock(&_minor_lock);
902 
903 	r = idr_get_new(&_minor_idr, MINOR_ALLOCED, &m);
904 	if (r) {
905 		goto out;
906 	}
907 
908 	if (m >= (1 << MINORBITS)) {
909 		idr_remove(&_minor_idr, m);
910 		r = -ENOSPC;
911 		goto out;
912 	}
913 
914 	*minor = m;
915 
916 out:
917 	spin_unlock(&_minor_lock);
918 	return r;
919 }
920 
921 static struct block_device_operations dm_blk_dops;
922 
923 /*
924  * Allocate and initialise a blank device with a given minor.
925  */
926 static struct mapped_device *alloc_dev(int minor)
927 {
928 	int r;
929 	struct mapped_device *md = kmalloc(sizeof(*md), GFP_KERNEL);
930 	void *old_md;
931 
932 	if (!md) {
933 		DMWARN("unable to allocate device, out of memory.");
934 		return NULL;
935 	}
936 
937 	if (!try_module_get(THIS_MODULE))
938 		goto bad0;
939 
940 	/* get a minor number for the dev */
941 	if (minor == DM_ANY_MINOR)
942 		r = next_free_minor(md, &minor);
943 	else
944 		r = specific_minor(md, minor);
945 	if (r < 0)
946 		goto bad1;
947 
948 	memset(md, 0, sizeof(*md));
949 	init_rwsem(&md->io_lock);
950 	init_MUTEX(&md->suspend_lock);
951 	rwlock_init(&md->map_lock);
952 	atomic_set(&md->holders, 1);
953 	atomic_set(&md->open_count, 0);
954 	atomic_set(&md->event_nr, 0);
955 
956 	md->queue = blk_alloc_queue(GFP_KERNEL);
957 	if (!md->queue)
958 		goto bad1_free_minor;
959 
960 	md->queue->queuedata = md;
961 	md->queue->backing_dev_info.congested_fn = dm_any_congested;
962 	md->queue->backing_dev_info.congested_data = md;
963 	blk_queue_make_request(md->queue, dm_request);
964 	blk_queue_bounce_limit(md->queue, BLK_BOUNCE_ANY);
965 	md->queue->unplug_fn = dm_unplug_all;
966 	md->queue->issue_flush_fn = dm_flush_all;
967 
968 	md->io_pool = mempool_create_slab_pool(MIN_IOS, _io_cache);
969  	if (!md->io_pool)
970  		goto bad2;
971 
972 	md->tio_pool = mempool_create_slab_pool(MIN_IOS, _tio_cache);
973 	if (!md->tio_pool)
974 		goto bad3;
975 
976 	md->bs = bioset_create(16, 16, 4);
977 	if (!md->bs)
978 		goto bad_no_bioset;
979 
980 	md->disk = alloc_disk(1);
981 	if (!md->disk)
982 		goto bad4;
983 
984 	atomic_set(&md->pending, 0);
985 	init_waitqueue_head(&md->wait);
986 	init_waitqueue_head(&md->eventq);
987 
988 	md->disk->major = _major;
989 	md->disk->first_minor = minor;
990 	md->disk->fops = &dm_blk_dops;
991 	md->disk->queue = md->queue;
992 	md->disk->private_data = md;
993 	sprintf(md->disk->disk_name, "dm-%d", minor);
994 	add_disk(md->disk);
995 	format_dev_t(md->name, MKDEV(_major, minor));
996 
997 	/* Populate the mapping, nobody knows we exist yet */
998 	spin_lock(&_minor_lock);
999 	old_md = idr_replace(&_minor_idr, md, minor);
1000 	spin_unlock(&_minor_lock);
1001 
1002 	BUG_ON(old_md != MINOR_ALLOCED);
1003 
1004 	return md;
1005 
1006  bad4:
1007 	bioset_free(md->bs);
1008  bad_no_bioset:
1009 	mempool_destroy(md->tio_pool);
1010  bad3:
1011 	mempool_destroy(md->io_pool);
1012  bad2:
1013 	blk_cleanup_queue(md->queue);
1014  bad1_free_minor:
1015 	free_minor(minor);
1016  bad1:
1017 	module_put(THIS_MODULE);
1018  bad0:
1019 	kfree(md);
1020 	return NULL;
1021 }
1022 
1023 static void free_dev(struct mapped_device *md)
1024 {
1025 	int minor = md->disk->first_minor;
1026 
1027 	if (md->suspended_bdev) {
1028 		thaw_bdev(md->suspended_bdev, NULL);
1029 		bdput(md->suspended_bdev);
1030 	}
1031 	mempool_destroy(md->tio_pool);
1032 	mempool_destroy(md->io_pool);
1033 	bioset_free(md->bs);
1034 	del_gendisk(md->disk);
1035 	free_minor(minor);
1036 
1037 	spin_lock(&_minor_lock);
1038 	md->disk->private_data = NULL;
1039 	spin_unlock(&_minor_lock);
1040 
1041 	put_disk(md->disk);
1042 	blk_cleanup_queue(md->queue);
1043 	module_put(THIS_MODULE);
1044 	kfree(md);
1045 }
1046 
1047 /*
1048  * Bind a table to the device.
1049  */
1050 static void event_callback(void *context)
1051 {
1052 	struct mapped_device *md = (struct mapped_device *) context;
1053 
1054 	atomic_inc(&md->event_nr);
1055 	wake_up(&md->eventq);
1056 }
1057 
1058 static void __set_size(struct mapped_device *md, sector_t size)
1059 {
1060 	set_capacity(md->disk, size);
1061 
1062 	mutex_lock(&md->suspended_bdev->bd_inode->i_mutex);
1063 	i_size_write(md->suspended_bdev->bd_inode, (loff_t)size << SECTOR_SHIFT);
1064 	mutex_unlock(&md->suspended_bdev->bd_inode->i_mutex);
1065 }
1066 
1067 static int __bind(struct mapped_device *md, struct dm_table *t)
1068 {
1069 	request_queue_t *q = md->queue;
1070 	sector_t size;
1071 
1072 	size = dm_table_get_size(t);
1073 
1074 	/*
1075 	 * Wipe any geometry if the size of the table changed.
1076 	 */
1077 	if (size != get_capacity(md->disk))
1078 		memset(&md->geometry, 0, sizeof(md->geometry));
1079 
1080 	__set_size(md, size);
1081 	if (size == 0)
1082 		return 0;
1083 
1084 	dm_table_get(t);
1085 	dm_table_event_callback(t, event_callback, md);
1086 
1087 	write_lock(&md->map_lock);
1088 	md->map = t;
1089 	dm_table_set_restrictions(t, q);
1090 	write_unlock(&md->map_lock);
1091 
1092 	return 0;
1093 }
1094 
1095 static void __unbind(struct mapped_device *md)
1096 {
1097 	struct dm_table *map = md->map;
1098 
1099 	if (!map)
1100 		return;
1101 
1102 	dm_table_event_callback(map, NULL, NULL);
1103 	write_lock(&md->map_lock);
1104 	md->map = NULL;
1105 	write_unlock(&md->map_lock);
1106 	dm_table_put(map);
1107 }
1108 
1109 /*
1110  * Constructor for a new device.
1111  */
1112 int dm_create(int minor, struct mapped_device **result)
1113 {
1114 	struct mapped_device *md;
1115 
1116 	md = alloc_dev(minor);
1117 	if (!md)
1118 		return -ENXIO;
1119 
1120 	*result = md;
1121 	return 0;
1122 }
1123 
1124 static struct mapped_device *dm_find_md(dev_t dev)
1125 {
1126 	struct mapped_device *md;
1127 	unsigned minor = MINOR(dev);
1128 
1129 	if (MAJOR(dev) != _major || minor >= (1 << MINORBITS))
1130 		return NULL;
1131 
1132 	spin_lock(&_minor_lock);
1133 
1134 	md = idr_find(&_minor_idr, minor);
1135 	if (md && (md == MINOR_ALLOCED ||
1136 		   (dm_disk(md)->first_minor != minor) ||
1137 		   test_bit(DMF_FREEING, &md->flags))) {
1138 		md = NULL;
1139 		goto out;
1140 	}
1141 
1142 out:
1143 	spin_unlock(&_minor_lock);
1144 
1145 	return md;
1146 }
1147 
1148 struct mapped_device *dm_get_md(dev_t dev)
1149 {
1150 	struct mapped_device *md = dm_find_md(dev);
1151 
1152 	if (md)
1153 		dm_get(md);
1154 
1155 	return md;
1156 }
1157 
1158 void *dm_get_mdptr(struct mapped_device *md)
1159 {
1160 	return md->interface_ptr;
1161 }
1162 
1163 void dm_set_mdptr(struct mapped_device *md, void *ptr)
1164 {
1165 	md->interface_ptr = ptr;
1166 }
1167 
1168 void dm_get(struct mapped_device *md)
1169 {
1170 	atomic_inc(&md->holders);
1171 }
1172 
1173 const char *dm_device_name(struct mapped_device *md)
1174 {
1175 	return md->name;
1176 }
1177 EXPORT_SYMBOL_GPL(dm_device_name);
1178 
1179 void dm_put(struct mapped_device *md)
1180 {
1181 	struct dm_table *map;
1182 
1183 	BUG_ON(test_bit(DMF_FREEING, &md->flags));
1184 
1185 	if (atomic_dec_and_lock(&md->holders, &_minor_lock)) {
1186 		map = dm_get_table(md);
1187 		idr_replace(&_minor_idr, MINOR_ALLOCED, dm_disk(md)->first_minor);
1188 		set_bit(DMF_FREEING, &md->flags);
1189 		spin_unlock(&_minor_lock);
1190 		if (!dm_suspended(md)) {
1191 			dm_table_presuspend_targets(map);
1192 			dm_table_postsuspend_targets(map);
1193 		}
1194 		__unbind(md);
1195 		dm_table_put(map);
1196 		free_dev(md);
1197 	}
1198 }
1199 
1200 /*
1201  * Process the deferred bios
1202  */
1203 static void __flush_deferred_io(struct mapped_device *md, struct bio *c)
1204 {
1205 	struct bio *n;
1206 
1207 	while (c) {
1208 		n = c->bi_next;
1209 		c->bi_next = NULL;
1210 		__split_bio(md, c);
1211 		c = n;
1212 	}
1213 }
1214 
1215 /*
1216  * Swap in a new table (destroying old one).
1217  */
1218 int dm_swap_table(struct mapped_device *md, struct dm_table *table)
1219 {
1220 	int r = -EINVAL;
1221 
1222 	down(&md->suspend_lock);
1223 
1224 	/* device must be suspended */
1225 	if (!dm_suspended(md))
1226 		goto out;
1227 
1228 	__unbind(md);
1229 	r = __bind(md, table);
1230 
1231 out:
1232 	up(&md->suspend_lock);
1233 	return r;
1234 }
1235 
1236 /*
1237  * Functions to lock and unlock any filesystem running on the
1238  * device.
1239  */
1240 static int lock_fs(struct mapped_device *md)
1241 {
1242 	int r;
1243 
1244 	WARN_ON(md->frozen_sb);
1245 
1246 	md->frozen_sb = freeze_bdev(md->suspended_bdev);
1247 	if (IS_ERR(md->frozen_sb)) {
1248 		r = PTR_ERR(md->frozen_sb);
1249 		md->frozen_sb = NULL;
1250 		return r;
1251 	}
1252 
1253 	set_bit(DMF_FROZEN, &md->flags);
1254 
1255 	/* don't bdput right now, we don't want the bdev
1256 	 * to go away while it is locked.
1257 	 */
1258 	return 0;
1259 }
1260 
1261 static void unlock_fs(struct mapped_device *md)
1262 {
1263 	if (!test_bit(DMF_FROZEN, &md->flags))
1264 		return;
1265 
1266 	thaw_bdev(md->suspended_bdev, md->frozen_sb);
1267 	md->frozen_sb = NULL;
1268 	clear_bit(DMF_FROZEN, &md->flags);
1269 }
1270 
1271 /*
1272  * We need to be able to change a mapping table under a mounted
1273  * filesystem.  For example we might want to move some data in
1274  * the background.  Before the table can be swapped with
1275  * dm_bind_table, dm_suspend must be called to flush any in
1276  * flight bios and ensure that any further io gets deferred.
1277  */
1278 int dm_suspend(struct mapped_device *md, int do_lockfs)
1279 {
1280 	struct dm_table *map = NULL;
1281 	DECLARE_WAITQUEUE(wait, current);
1282 	struct bio *def;
1283 	int r = -EINVAL;
1284 
1285 	down(&md->suspend_lock);
1286 
1287 	if (dm_suspended(md))
1288 		goto out_unlock;
1289 
1290 	map = dm_get_table(md);
1291 
1292 	/* This does not get reverted if there's an error later. */
1293 	dm_table_presuspend_targets(map);
1294 
1295 	md->suspended_bdev = bdget_disk(md->disk, 0);
1296 	if (!md->suspended_bdev) {
1297 		DMWARN("bdget failed in dm_suspend");
1298 		r = -ENOMEM;
1299 		goto out;
1300 	}
1301 
1302 	/* Flush I/O to the device. */
1303 	if (do_lockfs) {
1304 		r = lock_fs(md);
1305 		if (r)
1306 			goto out;
1307 	}
1308 
1309 	/*
1310 	 * First we set the BLOCK_IO flag so no more ios will be mapped.
1311 	 */
1312 	down_write(&md->io_lock);
1313 	set_bit(DMF_BLOCK_IO, &md->flags);
1314 
1315 	add_wait_queue(&md->wait, &wait);
1316 	up_write(&md->io_lock);
1317 
1318 	/* unplug */
1319 	if (map)
1320 		dm_table_unplug_all(map);
1321 
1322 	/*
1323 	 * Then we wait for the already mapped ios to
1324 	 * complete.
1325 	 */
1326 	while (1) {
1327 		set_current_state(TASK_INTERRUPTIBLE);
1328 
1329 		if (!atomic_read(&md->pending) || signal_pending(current))
1330 			break;
1331 
1332 		io_schedule();
1333 	}
1334 	set_current_state(TASK_RUNNING);
1335 
1336 	down_write(&md->io_lock);
1337 	remove_wait_queue(&md->wait, &wait);
1338 
1339 	/* were we interrupted ? */
1340 	r = -EINTR;
1341 	if (atomic_read(&md->pending)) {
1342 		clear_bit(DMF_BLOCK_IO, &md->flags);
1343 		def = bio_list_get(&md->deferred);
1344 		__flush_deferred_io(md, def);
1345 		up_write(&md->io_lock);
1346 		unlock_fs(md);
1347 		goto out;
1348 	}
1349 	up_write(&md->io_lock);
1350 
1351 	dm_table_postsuspend_targets(map);
1352 
1353 	set_bit(DMF_SUSPENDED, &md->flags);
1354 
1355 	r = 0;
1356 
1357 out:
1358 	if (r && md->suspended_bdev) {
1359 		bdput(md->suspended_bdev);
1360 		md->suspended_bdev = NULL;
1361 	}
1362 
1363 	dm_table_put(map);
1364 
1365 out_unlock:
1366 	up(&md->suspend_lock);
1367 	return r;
1368 }
1369 
1370 int dm_resume(struct mapped_device *md)
1371 {
1372 	int r = -EINVAL;
1373 	struct bio *def;
1374 	struct dm_table *map = NULL;
1375 
1376 	down(&md->suspend_lock);
1377 	if (!dm_suspended(md))
1378 		goto out;
1379 
1380 	map = dm_get_table(md);
1381 	if (!map || !dm_table_get_size(map))
1382 		goto out;
1383 
1384 	r = dm_table_resume_targets(map);
1385 	if (r)
1386 		goto out;
1387 
1388 	down_write(&md->io_lock);
1389 	clear_bit(DMF_BLOCK_IO, &md->flags);
1390 
1391 	def = bio_list_get(&md->deferred);
1392 	__flush_deferred_io(md, def);
1393 	up_write(&md->io_lock);
1394 
1395 	unlock_fs(md);
1396 
1397 	bdput(md->suspended_bdev);
1398 	md->suspended_bdev = NULL;
1399 
1400 	clear_bit(DMF_SUSPENDED, &md->flags);
1401 
1402 	dm_table_unplug_all(map);
1403 
1404 	kobject_uevent(&md->disk->kobj, KOBJ_CHANGE);
1405 
1406 	r = 0;
1407 
1408 out:
1409 	dm_table_put(map);
1410 	up(&md->suspend_lock);
1411 
1412 	return r;
1413 }
1414 
1415 /*-----------------------------------------------------------------
1416  * Event notification.
1417  *---------------------------------------------------------------*/
1418 uint32_t dm_get_event_nr(struct mapped_device *md)
1419 {
1420 	return atomic_read(&md->event_nr);
1421 }
1422 
1423 int dm_wait_event(struct mapped_device *md, int event_nr)
1424 {
1425 	return wait_event_interruptible(md->eventq,
1426 			(event_nr != atomic_read(&md->event_nr)));
1427 }
1428 
1429 /*
1430  * The gendisk is only valid as long as you have a reference
1431  * count on 'md'.
1432  */
1433 struct gendisk *dm_disk(struct mapped_device *md)
1434 {
1435 	return md->disk;
1436 }
1437 
1438 int dm_suspended(struct mapped_device *md)
1439 {
1440 	return test_bit(DMF_SUSPENDED, &md->flags);
1441 }
1442 
1443 static struct block_device_operations dm_blk_dops = {
1444 	.open = dm_blk_open,
1445 	.release = dm_blk_close,
1446 	.ioctl = dm_blk_ioctl,
1447 	.getgeo = dm_blk_getgeo,
1448 	.owner = THIS_MODULE
1449 };
1450 
1451 EXPORT_SYMBOL(dm_get_mapinfo);
1452 
1453 /*
1454  * module hooks
1455  */
1456 module_init(dm_init);
1457 module_exit(dm_exit);
1458 
1459 module_param(major, uint, 0);
1460 MODULE_PARM_DESC(major, "The major number of the device mapper");
1461 MODULE_DESCRIPTION(DM_NAME " driver");
1462 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
1463 MODULE_LICENSE("GPL");
1464