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
27static const char *_name = DM_NAME;
28
29static unsigned int major = 0;
30static unsigned int _major = 0;
31
32static DEFINE_SPINLOCK(_minor_lock);
33/*
34 * One of these is allocated per bio.
35 */
36struct 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 */
48struct target_io {
49	struct dm_io *io;
50	struct dm_target *ti;
51	union map_info info;
52};
53
54union 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#define DMF_NOFLUSH_SUSPENDING 5
72
73struct mapped_device {
74	struct rw_semaphore io_lock;
75	struct semaphore suspend_lock;
76	spinlock_t pushback_lock;
77	rwlock_t map_lock;
78	atomic_t holders;
79	atomic_t open_count;
80
81	unsigned long flags;
82
83	request_queue_t *queue;
84	struct gendisk *disk;
85	char name[16];
86
87	void *interface_ptr;
88
89	/*
90	 * A list of ios that arrived while we were suspended.
91	 */
92	atomic_t pending;
93	wait_queue_head_t wait;
94	struct bio_list deferred;
95	struct bio_list pushback;
96
97	/*
98	 * The current mapping.
99	 */
100	struct dm_table *map;
101
102	/*
103	 * io objects are allocated from here.
104	 */
105	mempool_t *io_pool;
106	mempool_t *tio_pool;
107
108	struct bio_set *bs;
109
110	/*
111	 * Event handling.
112	 */
113	atomic_t event_nr;
114	wait_queue_head_t eventq;
115
116	/*
117	 * freeze/thaw support require holding onto a super block
118	 */
119	struct super_block *frozen_sb;
120	struct block_device *suspended_bdev;
121
122	/* forced geometry settings */
123	struct hd_geometry geometry;
124};
125
126#define MIN_IOS 256
127static struct kmem_cache *_io_cache;
128static struct kmem_cache *_tio_cache;
129
130static int __init local_init(void)
131{
132	int r;
133
134	/* allocate a slab for the dm_ios */
135	_io_cache = kmem_cache_create("dm_io",
136				      sizeof(struct dm_io), 0, 0, NULL, NULL);
137	if (!_io_cache)
138		return -ENOMEM;
139
140	/* allocate a slab for the target ios */
141	_tio_cache = kmem_cache_create("dm_tio", sizeof(struct target_io),
142				       0, 0, NULL, NULL);
143	if (!_tio_cache) {
144		kmem_cache_destroy(_io_cache);
145		return -ENOMEM;
146	}
147
148	_major = major;
149	r = register_blkdev(_major, _name);
150	if (r < 0) {
151		kmem_cache_destroy(_tio_cache);
152		kmem_cache_destroy(_io_cache);
153		return r;
154	}
155
156	if (!_major)
157		_major = r;
158
159	return 0;
160}
161
162static void local_exit(void)
163{
164	kmem_cache_destroy(_tio_cache);
165	kmem_cache_destroy(_io_cache);
166
167	if (unregister_blkdev(_major, _name) < 0)
168		DMERR("unregister_blkdev failed");
169
170	_major = 0;
171
172	DMINFO("cleaned up");
173}
174
175int (*_inits[])(void) __initdata = {
176	local_init,
177	dm_target_init,
178	dm_linear_init,
179	dm_stripe_init,
180	dm_interface_init,
181};
182
183void (*_exits[])(void) = {
184	local_exit,
185	dm_target_exit,
186	dm_linear_exit,
187	dm_stripe_exit,
188	dm_interface_exit,
189};
190
191static int __init dm_init(void)
192{
193	const int count = ARRAY_SIZE(_inits);
194
195	int r, i;
196
197	for (i = 0; i < count; i++) {
198		r = _inits[i]();
199		if (r)
200			goto bad;
201	}
202
203	return 0;
204
205      bad:
206	while (i--)
207		_exits[i]();
208
209	return r;
210}
211
212static void __exit dm_exit(void)
213{
214	int i = ARRAY_SIZE(_exits);
215
216	while (i--)
217		_exits[i]();
218}
219
220/*
221 * Block device functions
222 */
223static int dm_blk_open(struct inode *inode, struct file *file)
224{
225	struct mapped_device *md;
226
227	spin_lock(&_minor_lock);
228
229	md = inode->i_bdev->bd_disk->private_data;
230	if (!md)
231		goto out;
232
233	if (test_bit(DMF_FREEING, &md->flags) ||
234	    test_bit(DMF_DELETING, &md->flags)) {
235		md = NULL;
236		goto out;
237	}
238
239	dm_get(md);
240	atomic_inc(&md->open_count);
241
242out:
243	spin_unlock(&_minor_lock);
244
245	return md ? 0 : -ENXIO;
246}
247
248static int dm_blk_close(struct inode *inode, struct file *file)
249{
250	struct mapped_device *md;
251
252	md = inode->i_bdev->bd_disk->private_data;
253	atomic_dec(&md->open_count);
254	dm_put(md);
255	return 0;
256}
257
258int dm_open_count(struct mapped_device *md)
259{
260	return atomic_read(&md->open_count);
261}
262
263/*
264 * Guarantees nothing is using the device before it's deleted.
265 */
266int dm_lock_for_deletion(struct mapped_device *md)
267{
268	int r = 0;
269
270	spin_lock(&_minor_lock);
271
272	if (dm_open_count(md))
273		r = -EBUSY;
274	else
275		set_bit(DMF_DELETING, &md->flags);
276
277	spin_unlock(&_minor_lock);
278
279	return r;
280}
281
282static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo)
283{
284	struct mapped_device *md = bdev->bd_disk->private_data;
285
286	return dm_get_geometry(md, geo);
287}
288
289static int dm_blk_ioctl(struct inode *inode, struct file *file,
290			unsigned int cmd, unsigned long arg)
291{
292	struct mapped_device *md;
293	struct dm_table *map;
294	struct dm_target *tgt;
295	int r = -ENOTTY;
296
297	/* We don't really need this lock, but we do need 'inode'. */
298	unlock_kernel();
299
300	md = inode->i_bdev->bd_disk->private_data;
301
302	map = dm_get_table(md);
303
304	if (!map || !dm_table_get_size(map))
305		goto out;
306
307	/* We only support devices that have a single target */
308	if (dm_table_get_num_targets(map) != 1)
309		goto out;
310
311	tgt = dm_table_get_target(map, 0);
312
313	if (dm_suspended(md)) {
314		r = -EAGAIN;
315		goto out;
316	}
317
318	if (tgt->type->ioctl)
319		r = tgt->type->ioctl(tgt, inode, file, cmd, arg);
320
321out:
322	dm_table_put(map);
323
324	lock_kernel();
325	return r;
326}
327
328static inline struct dm_io *alloc_io(struct mapped_device *md)
329{
330	return mempool_alloc(md->io_pool, GFP_NOIO);
331}
332
333static inline void free_io(struct mapped_device *md, struct dm_io *io)
334{
335	mempool_free(io, md->io_pool);
336}
337
338static inline struct target_io *alloc_tio(struct mapped_device *md)
339{
340	return mempool_alloc(md->tio_pool, GFP_NOIO);
341}
342
343static inline void free_tio(struct mapped_device *md, struct target_io *tio)
344{
345	mempool_free(tio, md->tio_pool);
346}
347
348static void start_io_acct(struct dm_io *io)
349{
350	struct mapped_device *md = io->md;
351
352	io->start_time = jiffies;
353
354	preempt_disable();
355	disk_round_stats(dm_disk(md));
356	preempt_enable();
357	dm_disk(md)->in_flight = atomic_inc_return(&md->pending);
358}
359
360static int end_io_acct(struct dm_io *io)
361{
362	struct mapped_device *md = io->md;
363	struct bio *bio = io->bio;
364	unsigned long duration = jiffies - io->start_time;
365	int pending;
366	int rw = bio_data_dir(bio);
367
368	preempt_disable();
369	disk_round_stats(dm_disk(md));
370	preempt_enable();
371	dm_disk(md)->in_flight = pending = atomic_dec_return(&md->pending);
372
373	disk_stat_add(dm_disk(md), ticks[rw], duration);
374
375	return !pending;
376}
377
378/*
379 * Add the bio to the list of deferred io.
380 */
381static int queue_io(struct mapped_device *md, struct bio *bio)
382{
383	down_write(&md->io_lock);
384
385	if (!test_bit(DMF_BLOCK_IO, &md->flags)) {
386		up_write(&md->io_lock);
387		return 1;
388	}
389
390	bio_list_add(&md->deferred, bio);
391
392	up_write(&md->io_lock);
393	return 0;		/* deferred successfully */
394}
395
396/*
397 * Everyone (including functions in this file), should use this
398 * function to access the md->map field, and make sure they call
399 * dm_table_put() when finished.
400 */
401struct dm_table *dm_get_table(struct mapped_device *md)
402{
403	struct dm_table *t;
404
405	read_lock(&md->map_lock);
406	t = md->map;
407	if (t)
408		dm_table_get(t);
409	read_unlock(&md->map_lock);
410
411	return t;
412}
413
414/*
415 * Get the geometry associated with a dm device
416 */
417int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo)
418{
419	*geo = md->geometry;
420
421	return 0;
422}
423
424/*
425 * Set the geometry of a device.
426 */
427int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo)
428{
429	sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors;
430
431	if (geo->start > sz) {
432		DMWARN("Start sector is beyond the geometry limits.");
433		return -EINVAL;
434	}
435
436	md->geometry = *geo;
437
438	return 0;
439}
440
441/*-----------------------------------------------------------------
442 * CRUD START:
443 *   A more elegant soln is in the works that uses the queue
444 *   merge fn, unfortunately there are a couple of changes to
445 *   the block layer that I want to make for this.  So in the
446 *   interests of getting something for people to use I give
447 *   you this clearly demarcated crap.
448 *---------------------------------------------------------------*/
449
450static int __noflush_suspending(struct mapped_device *md)
451{
452	return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
453}
454
455/*
456 * Decrements the number of outstanding ios that a bio has been
457 * cloned into, completing the original io if necc.
458 */
459static void dec_pending(struct dm_io *io, int error)
460{
461	unsigned long flags;
462
463	/* Push-back supersedes any I/O errors */
464	if (error && !(io->error > 0 && __noflush_suspending(io->md)))
465		io->error = error;
466
467	if (atomic_dec_and_test(&io->io_count)) {
468		if (io->error == DM_ENDIO_REQUEUE) {
469			/*
470			 * Target requested pushing back the I/O.
471			 * This must be handled before the sleeper on
472			 * suspend queue merges the pushback list.
473			 */
474			spin_lock_irqsave(&io->md->pushback_lock, flags);
475			if (__noflush_suspending(io->md))
476				bio_list_add(&io->md->pushback, io->bio);
477			else
478				/* noflush suspend was interrupted. */
479				io->error = -EIO;
480			spin_unlock_irqrestore(&io->md->pushback_lock, flags);
481		}
482
483		if (end_io_acct(io))
484			/* nudge anyone waiting on suspend queue */
485			wake_up(&io->md->wait);
486
487		if (io->error != DM_ENDIO_REQUEUE) {
488			blk_add_trace_bio(io->md->queue, io->bio,
489					  BLK_TA_COMPLETE);
490
491			bio_endio(io->bio, io->bio->bi_size, io->error);
492		}
493
494		free_io(io->md, io);
495	}
496}
497
498static int clone_endio(struct bio *bio, unsigned int done, int error)
499{
500	int r = 0;
501	struct target_io *tio = bio->bi_private;
502	struct mapped_device *md = tio->io->md;
503	dm_endio_fn endio = tio->ti->type->end_io;
504
505	if (bio->bi_size)
506		return 1;
507
508	if (!bio_flagged(bio, BIO_UPTODATE) && !error)
509		error = -EIO;
510
511	if (endio) {
512		r = endio(tio->ti, bio, error, &tio->info);
513		if (r < 0 || r == DM_ENDIO_REQUEUE)
514			/*
515			 * error and requeue request are handled
516			 * in dec_pending().
517			 */
518			error = r;
519		else if (r == DM_ENDIO_INCOMPLETE)
520			/* The target will handle the io */
521			return 1;
522		else if (r) {
523			DMWARN("unimplemented target endio return value: %d", r);
524			BUG();
525		}
526	}
527
528	dec_pending(tio->io, error);
529
530	/*
531	 * Store md for cleanup instead of tio which is about to get freed.
532	 */
533	bio->bi_private = md->bs;
534
535	bio_put(bio);
536	free_tio(md, tio);
537	return r;
538}
539
540static sector_t max_io_len(struct mapped_device *md,
541			   sector_t sector, struct dm_target *ti)
542{
543	sector_t offset = sector - ti->begin;
544	sector_t len = ti->len - offset;
545
546	/*
547	 * Does the target need to split even further ?
548	 */
549	if (ti->split_io) {
550		sector_t boundary;
551		boundary = ((offset + ti->split_io) & ~(ti->split_io - 1))
552			   - offset;
553		if (len > boundary)
554			len = boundary;
555	}
556
557	return len;
558}
559
560static void __map_bio(struct dm_target *ti, struct bio *clone,
561		      struct target_io *tio)
562{
563	int r;
564	sector_t sector;
565	struct mapped_device *md;
566
567	/*
568	 * Sanity checks.
569	 */
570	BUG_ON(!clone->bi_size);
571
572	clone->bi_end_io = clone_endio;
573	clone->bi_private = tio;
574
575	/*
576	 * Map the clone.  If r == 0 we don't need to do
577	 * anything, the target has assumed ownership of
578	 * this io.
579	 */
580	atomic_inc(&tio->io->io_count);
581	sector = clone->bi_sector;
582	r = ti->type->map(ti, clone, &tio->info);
583	if (r == DM_MAPIO_REMAPPED) {
584		/* the bio has been remapped so dispatch it */
585
586		blk_add_trace_remap(bdev_get_queue(clone->bi_bdev), clone,
587				    tio->io->bio->bi_bdev->bd_dev, sector,
588				    clone->bi_sector);
589
590		generic_make_request(clone);
591	} else if (r < 0 || r == DM_MAPIO_REQUEUE) {
592		/* error the io and bail out, or requeue it if needed */
593		md = tio->io->md;
594		dec_pending(tio->io, r);
595		/*
596		 * Store bio_set for cleanup.
597		 */
598		clone->bi_private = md->bs;
599		bio_put(clone);
600		free_tio(md, tio);
601	} else if (r) {
602		DMWARN("unimplemented target map return value: %d", r);
603		BUG();
604	}
605}
606
607struct clone_info {
608	struct mapped_device *md;
609	struct dm_table *map;
610	struct bio *bio;
611	struct dm_io *io;
612	sector_t sector;
613	sector_t sector_count;
614	unsigned short idx;
615};
616
617static void dm_bio_destructor(struct bio *bio)
618{
619	struct bio_set *bs = bio->bi_private;
620
621	bio_free(bio, bs);
622}
623
624/*
625 * Creates a little bio that is just does part of a bvec.
626 */
627static struct bio *split_bvec(struct bio *bio, sector_t sector,
628			      unsigned short idx, unsigned int offset,
629			      unsigned int len, struct bio_set *bs)
630{
631	struct bio *clone;
632	struct bio_vec *bv = bio->bi_io_vec + idx;
633
634	clone = bio_alloc_bioset(GFP_NOIO, 1, bs);
635	clone->bi_destructor = dm_bio_destructor;
636	*clone->bi_io_vec = *bv;
637
638	clone->bi_sector = sector;
639	clone->bi_bdev = bio->bi_bdev;
640	clone->bi_rw = bio->bi_rw;
641	clone->bi_vcnt = 1;
642	clone->bi_size = to_bytes(len);
643	clone->bi_io_vec->bv_offset = offset;
644	clone->bi_io_vec->bv_len = clone->bi_size;
645
646	return clone;
647}
648
649/*
650 * Creates a bio that consists of range of complete bvecs.
651 */
652static struct bio *clone_bio(struct bio *bio, sector_t sector,
653			     unsigned short idx, unsigned short bv_count,
654			     unsigned int len, struct bio_set *bs)
655{
656	struct bio *clone;
657
658	clone = bio_alloc_bioset(GFP_NOIO, bio->bi_max_vecs, bs);
659	__bio_clone(clone, bio);
660	clone->bi_destructor = dm_bio_destructor;
661	clone->bi_sector = sector;
662	clone->bi_idx = idx;
663	clone->bi_vcnt = idx + bv_count;
664	clone->bi_size = to_bytes(len);
665	clone->bi_flags &= ~(1 << BIO_SEG_VALID);
666
667	return clone;
668}
669
670static void __clone_and_map(struct clone_info *ci)
671{
672	struct bio *clone, *bio = ci->bio;
673	struct dm_target *ti = dm_table_find_target(ci->map, ci->sector);
674	sector_t len = 0, max = max_io_len(ci->md, ci->sector, ti);
675	struct target_io *tio;
676
677	/*
678	 * Allocate a target io object.
679	 */
680	tio = alloc_tio(ci->md);
681	tio->io = ci->io;
682	tio->ti = ti;
683	memset(&tio->info, 0, sizeof(tio->info));
684
685	if (ci->sector_count <= max) {
686		/*
687		 * Optimise for the simple case where we can do all of
688		 * the remaining io with a single clone.
689		 */
690		clone = clone_bio(bio, ci->sector, ci->idx,
691				  bio->bi_vcnt - ci->idx, ci->sector_count,
692				  ci->md->bs);
693		__map_bio(ti, clone, tio);
694		ci->sector_count = 0;
695
696	} else if (to_sector(bio->bi_io_vec[ci->idx].bv_len) <= max) {
697		/*
698		 * There are some bvecs that don't span targets.
699		 * Do as many of these as possible.
700		 */
701		int i;
702		sector_t remaining = max;
703		sector_t bv_len;
704
705		for (i = ci->idx; remaining && (i < bio->bi_vcnt); i++) {
706			bv_len = to_sector(bio->bi_io_vec[i].bv_len);
707
708			if (bv_len > remaining)
709				break;
710
711			remaining -= bv_len;
712			len += bv_len;
713		}
714
715		clone = clone_bio(bio, ci->sector, ci->idx, i - ci->idx, len,
716				  ci->md->bs);
717		__map_bio(ti, clone, tio);
718
719		ci->sector += len;
720		ci->sector_count -= len;
721		ci->idx = i;
722
723	} else {
724		/*
725		 * Handle a bvec that must be split between two or more targets.
726		 */
727		struct bio_vec *bv = bio->bi_io_vec + ci->idx;
728		sector_t remaining = to_sector(bv->bv_len);
729		unsigned int offset = 0;
730
731		do {
732			if (offset) {
733				ti = dm_table_find_target(ci->map, ci->sector);
734				max = max_io_len(ci->md, ci->sector, ti);
735
736				tio = alloc_tio(ci->md);
737				tio->io = ci->io;
738				tio->ti = ti;
739				memset(&tio->info, 0, sizeof(tio->info));
740			}
741
742			len = min(remaining, max);
743
744			clone = split_bvec(bio, ci->sector, ci->idx,
745					   bv->bv_offset + offset, len,
746					   ci->md->bs);
747
748			__map_bio(ti, clone, tio);
749
750			ci->sector += len;
751			ci->sector_count -= len;
752			offset += to_bytes(len);
753		} while (remaining -= len);
754
755		ci->idx++;
756	}
757}
758
759/*
760 * Split the bio into several clones.
761 */
762static void __split_bio(struct mapped_device *md, struct bio *bio)
763{
764	struct clone_info ci;
765
766	ci.map = dm_get_table(md);
767	if (!ci.map) {
768		bio_io_error(bio, bio->bi_size);
769		return;
770	}
771
772	ci.md = md;
773	ci.bio = bio;
774	ci.io = alloc_io(md);
775	ci.io->error = 0;
776	atomic_set(&ci.io->io_count, 1);
777	ci.io->bio = bio;
778	ci.io->md = md;
779	ci.sector = bio->bi_sector;
780	ci.sector_count = bio_sectors(bio);
781	ci.idx = bio->bi_idx;
782
783	start_io_acct(ci.io);
784	while (ci.sector_count)
785		__clone_and_map(&ci);
786
787	/* drop the extra reference count */
788	dec_pending(ci.io, 0);
789	dm_table_put(ci.map);
790}
791/*-----------------------------------------------------------------
792 * CRUD END
793 *---------------------------------------------------------------*/
794
795/*
796 * The request function that just remaps the bio built up by
797 * dm_merge_bvec.
798 */
799static int dm_request(request_queue_t *q, struct bio *bio)
800{
801	int r;
802	int rw = bio_data_dir(bio);
803	struct mapped_device *md = q->queuedata;
804
805	down_read(&md->io_lock);
806
807	disk_stat_inc(dm_disk(md), ios[rw]);
808	disk_stat_add(dm_disk(md), sectors[rw], bio_sectors(bio));
809
810	/*
811	 * If we're suspended we have to queue
812	 * this io for later.
813	 */
814	while (test_bit(DMF_BLOCK_IO, &md->flags)) {
815		up_read(&md->io_lock);
816
817		if (bio_rw(bio) == READA) {
818			bio_io_error(bio, bio->bi_size);
819			return 0;
820		}
821
822		r = queue_io(md, bio);
823		if (r < 0) {
824			bio_io_error(bio, bio->bi_size);
825			return 0;
826
827		} else if (r == 0)
828			return 0;	/* deferred successfully */
829
830		/*
831		 * We're in a while loop, because someone could suspend
832		 * before we get to the following read lock.
833		 */
834		down_read(&md->io_lock);
835	}
836
837	__split_bio(md, bio);
838	up_read(&md->io_lock);
839	return 0;
840}
841
842static int dm_flush_all(request_queue_t *q, struct gendisk *disk,
843			sector_t *error_sector)
844{
845	struct mapped_device *md = q->queuedata;
846	struct dm_table *map = dm_get_table(md);
847	int ret = -ENXIO;
848
849	if (map) {
850		ret = dm_table_flush_all(map);
851		dm_table_put(map);
852	}
853
854	return ret;
855}
856
857static void dm_unplug_all(request_queue_t *q)
858{
859	struct mapped_device *md = q->queuedata;
860	struct dm_table *map = dm_get_table(md);
861
862	if (map) {
863		dm_table_unplug_all(map);
864		dm_table_put(map);
865	}
866}
867
868static int dm_any_congested(void *congested_data, int bdi_bits)
869{
870	int r;
871	struct mapped_device *md = (struct mapped_device *) congested_data;
872	struct dm_table *map = dm_get_table(md);
873
874	if (!map || test_bit(DMF_BLOCK_IO, &md->flags))
875		r = bdi_bits;
876	else
877		r = dm_table_any_congested(map, bdi_bits);
878
879	dm_table_put(map);
880	return r;
881}
882
883/*-----------------------------------------------------------------
884 * An IDR is used to keep track of allocated minor numbers.
885 *---------------------------------------------------------------*/
886static DEFINE_IDR(_minor_idr);
887
888static void free_minor(int minor)
889{
890	spin_lock(&_minor_lock);
891	idr_remove(&_minor_idr, minor);
892	spin_unlock(&_minor_lock);
893}
894
895/*
896 * See if the device with a specific minor # is free.
897 */
898static int specific_minor(struct mapped_device *md, int minor)
899{
900	int r, m;
901
902	if (minor >= (1 << MINORBITS))
903		return -EINVAL;
904
905	r = idr_pre_get(&_minor_idr, GFP_KERNEL);
906	if (!r)
907		return -ENOMEM;
908
909	spin_lock(&_minor_lock);
910
911	if (idr_find(&_minor_idr, minor)) {
912		r = -EBUSY;
913		goto out;
914	}
915
916	r = idr_get_new_above(&_minor_idr, MINOR_ALLOCED, minor, &m);
917	if (r)
918		goto out;
919
920	if (m != minor) {
921		idr_remove(&_minor_idr, m);
922		r = -EBUSY;
923		goto out;
924	}
925
926out:
927	spin_unlock(&_minor_lock);
928	return r;
929}
930
931static int next_free_minor(struct mapped_device *md, int *minor)
932{
933	int r, m;
934
935	r = idr_pre_get(&_minor_idr, GFP_KERNEL);
936	if (!r)
937		return -ENOMEM;
938
939	spin_lock(&_minor_lock);
940
941	r = idr_get_new(&_minor_idr, MINOR_ALLOCED, &m);
942	if (r) {
943		goto out;
944	}
945
946	if (m >= (1 << MINORBITS)) {
947		idr_remove(&_minor_idr, m);
948		r = -ENOSPC;
949		goto out;
950	}
951
952	*minor = m;
953
954out:
955	spin_unlock(&_minor_lock);
956	return r;
957}
958
959static struct block_device_operations dm_blk_dops;
960
961/*
962 * Allocate and initialise a blank device with a given minor.
963 */
964static struct mapped_device *alloc_dev(int minor)
965{
966	int r;
967	struct mapped_device *md = kmalloc(sizeof(*md), GFP_KERNEL);
968	void *old_md;
969
970	if (!md) {
971		DMWARN("unable to allocate device, out of memory.");
972		return NULL;
973	}
974
975	if (!try_module_get(THIS_MODULE))
976		goto bad0;
977
978	/* get a minor number for the dev */
979	if (minor == DM_ANY_MINOR)
980		r = next_free_minor(md, &minor);
981	else
982		r = specific_minor(md, minor);
983	if (r < 0)
984		goto bad1;
985
986	memset(md, 0, sizeof(*md));
987	init_rwsem(&md->io_lock);
988	init_MUTEX(&md->suspend_lock);
989	spin_lock_init(&md->pushback_lock);
990	rwlock_init(&md->map_lock);
991	atomic_set(&md->holders, 1);
992	atomic_set(&md->open_count, 0);
993	atomic_set(&md->event_nr, 0);
994
995	md->queue = blk_alloc_queue(GFP_KERNEL);
996	if (!md->queue)
997		goto bad1_free_minor;
998
999	md->queue->queuedata = md;
1000	md->queue->backing_dev_info.congested_fn = dm_any_congested;
1001	md->queue->backing_dev_info.congested_data = md;
1002	blk_queue_make_request(md->queue, dm_request);
1003	blk_queue_bounce_limit(md->queue, BLK_BOUNCE_ANY);
1004	md->queue->unplug_fn = dm_unplug_all;
1005	md->queue->issue_flush_fn = dm_flush_all;
1006
1007	md->io_pool = mempool_create_slab_pool(MIN_IOS, _io_cache);
1008	if (!md->io_pool)
1009		goto bad2;
1010
1011	md->tio_pool = mempool_create_slab_pool(MIN_IOS, _tio_cache);
1012	if (!md->tio_pool)
1013		goto bad3;
1014
1015	md->bs = bioset_create(16, 16);
1016	if (!md->bs)
1017		goto bad_no_bioset;
1018
1019	md->disk = alloc_disk(1);
1020	if (!md->disk)
1021		goto bad4;
1022
1023	atomic_set(&md->pending, 0);
1024	init_waitqueue_head(&md->wait);
1025	init_waitqueue_head(&md->eventq);
1026
1027	md->disk->major = _major;
1028	md->disk->first_minor = minor;
1029	md->disk->fops = &dm_blk_dops;
1030	md->disk->queue = md->queue;
1031	md->disk->private_data = md;
1032	sprintf(md->disk->disk_name, "dm-%d", minor);
1033	add_disk(md->disk);
1034	format_dev_t(md->name, MKDEV(_major, minor));
1035
1036	/* Populate the mapping, nobody knows we exist yet */
1037	spin_lock(&_minor_lock);
1038	old_md = idr_replace(&_minor_idr, md, minor);
1039	spin_unlock(&_minor_lock);
1040
1041	BUG_ON(old_md != MINOR_ALLOCED);
1042
1043	return md;
1044
1045 bad4:
1046	bioset_free(md->bs);
1047 bad_no_bioset:
1048	mempool_destroy(md->tio_pool);
1049 bad3:
1050	mempool_destroy(md->io_pool);
1051 bad2:
1052	blk_cleanup_queue(md->queue);
1053 bad1_free_minor:
1054	free_minor(minor);
1055 bad1:
1056	module_put(THIS_MODULE);
1057 bad0:
1058	kfree(md);
1059	return NULL;
1060}
1061
1062static void free_dev(struct mapped_device *md)
1063{
1064	int minor = md->disk->first_minor;
1065
1066	if (md->suspended_bdev) {
1067		thaw_bdev(md->suspended_bdev, NULL);
1068		bdput(md->suspended_bdev);
1069	}
1070	mempool_destroy(md->tio_pool);
1071	mempool_destroy(md->io_pool);
1072	bioset_free(md->bs);
1073	del_gendisk(md->disk);
1074	free_minor(minor);
1075
1076	spin_lock(&_minor_lock);
1077	md->disk->private_data = NULL;
1078	spin_unlock(&_minor_lock);
1079
1080	put_disk(md->disk);
1081	blk_cleanup_queue(md->queue);
1082	module_put(THIS_MODULE);
1083	kfree(md);
1084}
1085
1086/*
1087 * Bind a table to the device.
1088 */
1089static void event_callback(void *context)
1090{
1091	struct mapped_device *md = (struct mapped_device *) context;
1092
1093	atomic_inc(&md->event_nr);
1094	wake_up(&md->eventq);
1095}
1096
1097static void __set_size(struct mapped_device *md, sector_t size)
1098{
1099	set_capacity(md->disk, size);
1100
1101	mutex_lock(&md->suspended_bdev->bd_inode->i_mutex);
1102	i_size_write(md->suspended_bdev->bd_inode, (loff_t)size << SECTOR_SHIFT);
1103	mutex_unlock(&md->suspended_bdev->bd_inode->i_mutex);
1104}
1105
1106static int __bind(struct mapped_device *md, struct dm_table *t)
1107{
1108	request_queue_t *q = md->queue;
1109	sector_t size;
1110
1111	size = dm_table_get_size(t);
1112
1113	/*
1114	 * Wipe any geometry if the size of the table changed.
1115	 */
1116	if (size != get_capacity(md->disk))
1117		memset(&md->geometry, 0, sizeof(md->geometry));
1118
1119	if (md->suspended_bdev)
1120		__set_size(md, size);
1121	if (size == 0)
1122		return 0;
1123
1124	dm_table_get(t);
1125	dm_table_event_callback(t, event_callback, md);
1126
1127	write_lock(&md->map_lock);
1128	md->map = t;
1129	dm_table_set_restrictions(t, q);
1130	write_unlock(&md->map_lock);
1131
1132	return 0;
1133}
1134
1135static void __unbind(struct mapped_device *md)
1136{
1137	struct dm_table *map = md->map;
1138
1139	if (!map)
1140		return;
1141
1142	dm_table_event_callback(map, NULL, NULL);
1143	write_lock(&md->map_lock);
1144	md->map = NULL;
1145	write_unlock(&md->map_lock);
1146	dm_table_put(map);
1147}
1148
1149/*
1150 * Constructor for a new device.
1151 */
1152int dm_create(int minor, struct mapped_device **result)
1153{
1154	struct mapped_device *md;
1155
1156	md = alloc_dev(minor);
1157	if (!md)
1158		return -ENXIO;
1159
1160	*result = md;
1161	return 0;
1162}
1163
1164static struct mapped_device *dm_find_md(dev_t dev)
1165{
1166	struct mapped_device *md;
1167	unsigned minor = MINOR(dev);
1168
1169	if (MAJOR(dev) != _major || minor >= (1 << MINORBITS))
1170		return NULL;
1171
1172	spin_lock(&_minor_lock);
1173
1174	md = idr_find(&_minor_idr, minor);
1175	if (md && (md == MINOR_ALLOCED ||
1176		   (dm_disk(md)->first_minor != minor) ||
1177		   test_bit(DMF_FREEING, &md->flags))) {
1178		md = NULL;
1179		goto out;
1180	}
1181
1182out:
1183	spin_unlock(&_minor_lock);
1184
1185	return md;
1186}
1187
1188struct mapped_device *dm_get_md(dev_t dev)
1189{
1190	struct mapped_device *md = dm_find_md(dev);
1191
1192	if (md)
1193		dm_get(md);
1194
1195	return md;
1196}
1197
1198void *dm_get_mdptr(struct mapped_device *md)
1199{
1200	return md->interface_ptr;
1201}
1202
1203void dm_set_mdptr(struct mapped_device *md, void *ptr)
1204{
1205	md->interface_ptr = ptr;
1206}
1207
1208void dm_get(struct mapped_device *md)
1209{
1210	atomic_inc(&md->holders);
1211}
1212
1213const char *dm_device_name(struct mapped_device *md)
1214{
1215	return md->name;
1216}
1217EXPORT_SYMBOL_GPL(dm_device_name);
1218
1219void dm_put(struct mapped_device *md)
1220{
1221	struct dm_table *map;
1222
1223	BUG_ON(test_bit(DMF_FREEING, &md->flags));
1224
1225	if (atomic_dec_and_lock(&md->holders, &_minor_lock)) {
1226		map = dm_get_table(md);
1227		idr_replace(&_minor_idr, MINOR_ALLOCED, dm_disk(md)->first_minor);
1228		set_bit(DMF_FREEING, &md->flags);
1229		spin_unlock(&_minor_lock);
1230		if (!dm_suspended(md)) {
1231			dm_table_presuspend_targets(map);
1232			dm_table_postsuspend_targets(map);
1233		}
1234		__unbind(md);
1235		dm_table_put(map);
1236		free_dev(md);
1237	}
1238}
1239EXPORT_SYMBOL_GPL(dm_put);
1240
1241/*
1242 * Process the deferred bios
1243 */
1244static void __flush_deferred_io(struct mapped_device *md, struct bio *c)
1245{
1246	struct bio *n;
1247
1248	while (c) {
1249		n = c->bi_next;
1250		c->bi_next = NULL;
1251		__split_bio(md, c);
1252		c = n;
1253	}
1254}
1255
1256/*
1257 * Swap in a new table (destroying old one).
1258 */
1259int dm_swap_table(struct mapped_device *md, struct dm_table *table)
1260{
1261	int r = -EINVAL;
1262
1263	down(&md->suspend_lock);
1264
1265	/* device must be suspended */
1266	if (!dm_suspended(md))
1267		goto out;
1268
1269	/* without bdev, the device size cannot be changed */
1270	if (!md->suspended_bdev)
1271		if (get_capacity(md->disk) != dm_table_get_size(table))
1272			goto out;
1273
1274	__unbind(md);
1275	r = __bind(md, table);
1276
1277out:
1278	up(&md->suspend_lock);
1279	return r;
1280}
1281
1282/*
1283 * Functions to lock and unlock any filesystem running on the
1284 * device.
1285 */
1286static int lock_fs(struct mapped_device *md)
1287{
1288	int r;
1289
1290	WARN_ON(md->frozen_sb);
1291
1292	md->frozen_sb = freeze_bdev(md->suspended_bdev);
1293	if (IS_ERR(md->frozen_sb)) {
1294		r = PTR_ERR(md->frozen_sb);
1295		md->frozen_sb = NULL;
1296		return r;
1297	}
1298
1299	set_bit(DMF_FROZEN, &md->flags);
1300
1301	/* don't bdput right now, we don't want the bdev
1302	 * to go away while it is locked.
1303	 */
1304	return 0;
1305}
1306
1307static void unlock_fs(struct mapped_device *md)
1308{
1309	if (!test_bit(DMF_FROZEN, &md->flags))
1310		return;
1311
1312	thaw_bdev(md->suspended_bdev, md->frozen_sb);
1313	md->frozen_sb = NULL;
1314	clear_bit(DMF_FROZEN, &md->flags);
1315}
1316
1317/*
1318 * We need to be able to change a mapping table under a mounted
1319 * filesystem.  For example we might want to move some data in
1320 * the background.  Before the table can be swapped with
1321 * dm_bind_table, dm_suspend must be called to flush any in
1322 * flight bios and ensure that any further io gets deferred.
1323 */
1324int dm_suspend(struct mapped_device *md, unsigned suspend_flags)
1325{
1326	struct dm_table *map = NULL;
1327	unsigned long flags;
1328	DECLARE_WAITQUEUE(wait, current);
1329	struct bio *def;
1330	int r = -EINVAL;
1331	int do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG ? 1 : 0;
1332	int noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG ? 1 : 0;
1333
1334	down(&md->suspend_lock);
1335
1336	if (dm_suspended(md))
1337		goto out_unlock;
1338
1339	map = dm_get_table(md);
1340
1341	/*
1342	 * DMF_NOFLUSH_SUSPENDING must be set before presuspend.
1343	 * This flag is cleared before dm_suspend returns.
1344	 */
1345	if (noflush)
1346		set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
1347
1348	/* This does not get reverted if there's an error later. */
1349	dm_table_presuspend_targets(map);
1350
1351	/* bdget() can stall if the pending I/Os are not flushed */
1352	if (!noflush) {
1353		md->suspended_bdev = bdget_disk(md->disk, 0);
1354		if (!md->suspended_bdev) {
1355			DMWARN("bdget failed in dm_suspend");
1356			r = -ENOMEM;
1357			goto flush_and_out;
1358		}
1359	}
1360
1361	/*
1362	 * Flush I/O to the device.
1363	 * noflush supersedes do_lockfs, because lock_fs() needs to flush I/Os.
1364	 */
1365	if (do_lockfs && !noflush) {
1366		r = lock_fs(md);
1367		if (r)
1368			goto out;
1369	}
1370
1371	/*
1372	 * First we set the BLOCK_IO flag so no more ios will be mapped.
1373	 */
1374	down_write(&md->io_lock);
1375	set_bit(DMF_BLOCK_IO, &md->flags);
1376
1377	add_wait_queue(&md->wait, &wait);
1378	up_write(&md->io_lock);
1379
1380	/* unplug */
1381	if (map)
1382		dm_table_unplug_all(map);
1383
1384	/*
1385	 * Then we wait for the already mapped ios to
1386	 * complete.
1387	 */
1388	while (1) {
1389		set_current_state(TASK_INTERRUPTIBLE);
1390
1391		if (!atomic_read(&md->pending) || signal_pending(current))
1392			break;
1393
1394		io_schedule();
1395	}
1396	set_current_state(TASK_RUNNING);
1397
1398	down_write(&md->io_lock);
1399	remove_wait_queue(&md->wait, &wait);
1400
1401	if (noflush) {
1402		spin_lock_irqsave(&md->pushback_lock, flags);
1403		clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
1404		bio_list_merge_head(&md->deferred, &md->pushback);
1405		bio_list_init(&md->pushback);
1406		spin_unlock_irqrestore(&md->pushback_lock, flags);
1407	}
1408
1409	/* were we interrupted ? */
1410	r = -EINTR;
1411	if (atomic_read(&md->pending)) {
1412		clear_bit(DMF_BLOCK_IO, &md->flags);
1413		def = bio_list_get(&md->deferred);
1414		__flush_deferred_io(md, def);
1415		up_write(&md->io_lock);
1416		unlock_fs(md);
1417		goto out; /* pushback list is already flushed, so skip flush */
1418	}
1419	up_write(&md->io_lock);
1420
1421	dm_table_postsuspend_targets(map);
1422
1423	set_bit(DMF_SUSPENDED, &md->flags);
1424
1425	r = 0;
1426
1427flush_and_out:
1428	if (r && noflush) {
1429		/*
1430		 * Because there may be already I/Os in the pushback list,
1431		 * flush them before return.
1432		 */
1433		down_write(&md->io_lock);
1434
1435		spin_lock_irqsave(&md->pushback_lock, flags);
1436		clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
1437		bio_list_merge_head(&md->deferred, &md->pushback);
1438		bio_list_init(&md->pushback);
1439		spin_unlock_irqrestore(&md->pushback_lock, flags);
1440
1441		def = bio_list_get(&md->deferred);
1442		__flush_deferred_io(md, def);
1443		up_write(&md->io_lock);
1444	}
1445
1446out:
1447	if (r && md->suspended_bdev) {
1448		bdput(md->suspended_bdev);
1449		md->suspended_bdev = NULL;
1450	}
1451
1452	dm_table_put(map);
1453
1454out_unlock:
1455	up(&md->suspend_lock);
1456	return r;
1457}
1458
1459int dm_resume(struct mapped_device *md)
1460{
1461	int r = -EINVAL;
1462	struct bio *def;
1463	struct dm_table *map = NULL;
1464
1465	down(&md->suspend_lock);
1466	if (!dm_suspended(md))
1467		goto out;
1468
1469	map = dm_get_table(md);
1470	if (!map || !dm_table_get_size(map))
1471		goto out;
1472
1473	r = dm_table_resume_targets(map);
1474	if (r)
1475		goto out;
1476
1477	down_write(&md->io_lock);
1478	clear_bit(DMF_BLOCK_IO, &md->flags);
1479
1480	def = bio_list_get(&md->deferred);
1481	__flush_deferred_io(md, def);
1482	up_write(&md->io_lock);
1483
1484	unlock_fs(md);
1485
1486	if (md->suspended_bdev) {
1487		bdput(md->suspended_bdev);
1488		md->suspended_bdev = NULL;
1489	}
1490
1491	clear_bit(DMF_SUSPENDED, &md->flags);
1492
1493	dm_table_unplug_all(map);
1494
1495	kobject_uevent(&md->disk->kobj, KOBJ_CHANGE);
1496
1497	r = 0;
1498
1499out:
1500	dm_table_put(map);
1501	up(&md->suspend_lock);
1502
1503	return r;
1504}
1505
1506/*-----------------------------------------------------------------
1507 * Event notification.
1508 *---------------------------------------------------------------*/
1509uint32_t dm_get_event_nr(struct mapped_device *md)
1510{
1511	return atomic_read(&md->event_nr);
1512}
1513
1514int dm_wait_event(struct mapped_device *md, int event_nr)
1515{
1516	return wait_event_interruptible(md->eventq,
1517			(event_nr != atomic_read(&md->event_nr)));
1518}
1519
1520/*
1521 * The gendisk is only valid as long as you have a reference
1522 * count on 'md'.
1523 */
1524struct gendisk *dm_disk(struct mapped_device *md)
1525{
1526	return md->disk;
1527}
1528
1529int dm_suspended(struct mapped_device *md)
1530{
1531	return test_bit(DMF_SUSPENDED, &md->flags);
1532}
1533
1534int dm_noflush_suspending(struct dm_target *ti)
1535{
1536	struct mapped_device *md = dm_table_get_md(ti->table);
1537	int r = __noflush_suspending(md);
1538
1539	dm_put(md);
1540
1541	return r;
1542}
1543EXPORT_SYMBOL_GPL(dm_noflush_suspending);
1544
1545static struct block_device_operations dm_blk_dops = {
1546	.open = dm_blk_open,
1547	.release = dm_blk_close,
1548	.ioctl = dm_blk_ioctl,
1549	.getgeo = dm_blk_getgeo,
1550	.owner = THIS_MODULE
1551};
1552
1553EXPORT_SYMBOL(dm_get_mapinfo);
1554
1555/*
1556 * module hooks
1557 */
1558module_init(dm_init);
1559module_exit(dm_exit);
1560
1561module_param(major, uint, 0);
1562MODULE_PARM_DESC(major, "The major number of the device mapper");
1563MODULE_DESCRIPTION(DM_NAME " driver");
1564MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
1565MODULE_LICENSE("GPL");
1566