2 * Copyright (C) 2001, 2002 Sistina Software (UK) Limited.
3 * Copyright (C) 2004-2008 Red Hat, Inc. All rights reserved.
5 * This file is released under the GPL.
9 #include "dm-bio-list.h"
10 #include "dm-uevent.h"
12 #include <linux/init.h>
13 #include <linux/module.h>
14 #include <linux/mutex.h>
15 #include <linux/moduleparam.h>
16 #include <linux/blkpg.h>
17 #include <linux/bio.h>
18 #include <linux/buffer_head.h>
19 #include <linux/mempool.h>
20 #include <linux/slab.h>
21 #include <linux/idr.h>
22 #include <linux/hdreg.h>
23 #include <linux/blktrace_api.h>
24 #include <trace/block.h>
26 #define DM_MSG_PREFIX "core"
28 static const char *_name = DM_NAME;
30 static unsigned int major = 0;
31 static unsigned int _major = 0;
33 static DEFINE_SPINLOCK(_minor_lock);
36 * One of these is allocated per bio.
39 struct mapped_device *md;
43 unsigned long start_time;
48 * One of these is allocated per target within a bio. Hopefully
49 * this will be simplified out one day.
57 DEFINE_TRACE(block_bio_complete);
60 * For request-based dm.
61 * One of these is allocated per request.
63 struct dm_rq_target_io {
64 struct mapped_device *md;
66 struct request *orig, clone;
72 * For request-based dm.
73 * One of these is allocated per bio.
75 struct dm_rq_clone_bio_info {
80 union map_info *dm_get_mapinfo(struct bio *bio)
82 if (bio && bio->bi_private)
83 return &((struct dm_target_io *)bio->bi_private)->info;
87 #define MINOR_ALLOCED ((void *)-1)
90 * Bits for the md->flags field.
92 #define DMF_BLOCK_IO_FOR_SUSPEND 0
93 #define DMF_SUSPENDED 1
96 #define DMF_DELETING 4
97 #define DMF_NOFLUSH_SUSPENDING 5
98 #define DMF_QUEUE_IO_TO_THREAD 6
101 * Work processed by per-device workqueue.
103 struct mapped_device {
104 struct rw_semaphore io_lock;
105 struct mutex suspend_lock;
112 struct request_queue *queue;
113 struct gendisk *disk;
119 * A list of ios that arrived while we were suspended.
122 wait_queue_head_t wait;
123 struct work_struct work;
124 struct bio_list deferred;
125 spinlock_t deferred_lock;
128 * Processing queue (flush/barriers)
130 struct workqueue_struct *wq;
133 * The current mapping.
135 struct dm_table *map;
138 * io objects are allocated from here.
149 wait_queue_head_t eventq;
151 struct list_head uevent_list;
152 spinlock_t uevent_lock; /* Protect access to uevent_list */
155 * freeze/thaw support require holding onto a super block
157 struct super_block *frozen_sb;
158 struct block_device *suspended_bdev;
160 /* forced geometry settings */
161 struct hd_geometry geometry;
168 static struct kmem_cache *_io_cache;
169 static struct kmem_cache *_tio_cache;
170 static struct kmem_cache *_rq_tio_cache;
171 static struct kmem_cache *_rq_bio_info_cache;
173 static int __init local_init(void)
177 /* allocate a slab for the dm_ios */
178 _io_cache = KMEM_CACHE(dm_io, 0);
182 /* allocate a slab for the target ios */
183 _tio_cache = KMEM_CACHE(dm_target_io, 0);
185 goto out_free_io_cache;
187 _rq_tio_cache = KMEM_CACHE(dm_rq_target_io, 0);
189 goto out_free_tio_cache;
191 _rq_bio_info_cache = KMEM_CACHE(dm_rq_clone_bio_info, 0);
192 if (!_rq_bio_info_cache)
193 goto out_free_rq_tio_cache;
195 r = dm_uevent_init();
197 goto out_free_rq_bio_info_cache;
200 r = register_blkdev(_major, _name);
202 goto out_uevent_exit;
211 out_free_rq_bio_info_cache:
212 kmem_cache_destroy(_rq_bio_info_cache);
213 out_free_rq_tio_cache:
214 kmem_cache_destroy(_rq_tio_cache);
216 kmem_cache_destroy(_tio_cache);
218 kmem_cache_destroy(_io_cache);
223 static void local_exit(void)
225 kmem_cache_destroy(_rq_bio_info_cache);
226 kmem_cache_destroy(_rq_tio_cache);
227 kmem_cache_destroy(_tio_cache);
228 kmem_cache_destroy(_io_cache);
229 unregister_blkdev(_major, _name);
234 DMINFO("cleaned up");
237 static int (*_inits[])(void) __initdata = {
246 static void (*_exits[])(void) = {
255 static int __init dm_init(void)
257 const int count = ARRAY_SIZE(_inits);
261 for (i = 0; i < count; i++) {
276 static void __exit dm_exit(void)
278 int i = ARRAY_SIZE(_exits);
285 * Block device functions
287 static int dm_blk_open(struct block_device *bdev, fmode_t mode)
289 struct mapped_device *md;
291 spin_lock(&_minor_lock);
293 md = bdev->bd_disk->private_data;
297 if (test_bit(DMF_FREEING, &md->flags) ||
298 test_bit(DMF_DELETING, &md->flags)) {
304 atomic_inc(&md->open_count);
307 spin_unlock(&_minor_lock);
309 return md ? 0 : -ENXIO;
312 static int dm_blk_close(struct gendisk *disk, fmode_t mode)
314 struct mapped_device *md = disk->private_data;
315 atomic_dec(&md->open_count);
320 int dm_open_count(struct mapped_device *md)
322 return atomic_read(&md->open_count);
326 * Guarantees nothing is using the device before it's deleted.
328 int dm_lock_for_deletion(struct mapped_device *md)
332 spin_lock(&_minor_lock);
334 if (dm_open_count(md))
337 set_bit(DMF_DELETING, &md->flags);
339 spin_unlock(&_minor_lock);
344 static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo)
346 struct mapped_device *md = bdev->bd_disk->private_data;
348 return dm_get_geometry(md, geo);
351 static int dm_blk_ioctl(struct block_device *bdev, fmode_t mode,
352 unsigned int cmd, unsigned long arg)
354 struct mapped_device *md = bdev->bd_disk->private_data;
355 struct dm_table *map = dm_get_table(md);
356 struct dm_target *tgt;
359 if (!map || !dm_table_get_size(map))
362 /* We only support devices that have a single target */
363 if (dm_table_get_num_targets(map) != 1)
366 tgt = dm_table_get_target(map, 0);
368 if (dm_suspended(md)) {
373 if (tgt->type->ioctl)
374 r = tgt->type->ioctl(tgt, cmd, arg);
382 static struct dm_io *alloc_io(struct mapped_device *md)
384 return mempool_alloc(md->io_pool, GFP_NOIO);
387 static void free_io(struct mapped_device *md, struct dm_io *io)
389 mempool_free(io, md->io_pool);
392 static struct dm_target_io *alloc_tio(struct mapped_device *md)
394 return mempool_alloc(md->tio_pool, GFP_NOIO);
397 static void free_tio(struct mapped_device *md, struct dm_target_io *tio)
399 mempool_free(tio, md->tio_pool);
402 static void start_io_acct(struct dm_io *io)
404 struct mapped_device *md = io->md;
407 io->start_time = jiffies;
409 cpu = part_stat_lock();
410 part_round_stats(cpu, &dm_disk(md)->part0);
412 dm_disk(md)->part0.in_flight = atomic_inc_return(&md->pending);
415 static void end_io_acct(struct dm_io *io)
417 struct mapped_device *md = io->md;
418 struct bio *bio = io->bio;
419 unsigned long duration = jiffies - io->start_time;
421 int rw = bio_data_dir(bio);
423 cpu = part_stat_lock();
424 part_round_stats(cpu, &dm_disk(md)->part0);
425 part_stat_add(cpu, &dm_disk(md)->part0, ticks[rw], duration);
428 dm_disk(md)->part0.in_flight = pending =
429 atomic_dec_return(&md->pending);
431 /* nudge anyone waiting on suspend queue */
437 * Add the bio to the list of deferred io.
439 static void queue_io(struct mapped_device *md, struct bio *bio)
441 down_write(&md->io_lock);
443 spin_lock_irq(&md->deferred_lock);
444 bio_list_add(&md->deferred, bio);
445 spin_unlock_irq(&md->deferred_lock);
447 if (!test_and_set_bit(DMF_QUEUE_IO_TO_THREAD, &md->flags))
448 queue_work(md->wq, &md->work);
450 up_write(&md->io_lock);
454 * Everyone (including functions in this file), should use this
455 * function to access the md->map field, and make sure they call
456 * dm_table_put() when finished.
458 struct dm_table *dm_get_table(struct mapped_device *md)
462 read_lock(&md->map_lock);
466 read_unlock(&md->map_lock);
472 * Get the geometry associated with a dm device
474 int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo)
482 * Set the geometry of a device.
484 int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo)
486 sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors;
488 if (geo->start > sz) {
489 DMWARN("Start sector is beyond the geometry limits.");
498 /*-----------------------------------------------------------------
500 * A more elegant soln is in the works that uses the queue
501 * merge fn, unfortunately there are a couple of changes to
502 * the block layer that I want to make for this. So in the
503 * interests of getting something for people to use I give
504 * you this clearly demarcated crap.
505 *---------------------------------------------------------------*/
507 static int __noflush_suspending(struct mapped_device *md)
509 return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
513 * Decrements the number of outstanding ios that a bio has been
514 * cloned into, completing the original io if necc.
516 static void dec_pending(struct dm_io *io, int error)
521 struct mapped_device *md = io->md;
523 /* Push-back supersedes any I/O errors */
524 if (error && !(io->error > 0 && __noflush_suspending(md)))
527 if (atomic_dec_and_test(&io->io_count)) {
528 if (io->error == DM_ENDIO_REQUEUE) {
530 * Target requested pushing back the I/O.
532 spin_lock_irqsave(&md->deferred_lock, flags);
533 if (__noflush_suspending(md))
534 bio_list_add(&md->deferred, io->bio);
536 /* noflush suspend was interrupted. */
538 spin_unlock_irqrestore(&md->deferred_lock, flags);
543 io_error = io->error;
548 if (io_error != DM_ENDIO_REQUEUE) {
549 trace_block_bio_complete(md->queue, bio);
551 bio_endio(bio, io_error);
556 static void clone_endio(struct bio *bio, int error)
559 struct dm_target_io *tio = bio->bi_private;
560 struct dm_io *io = tio->io;
561 struct mapped_device *md = tio->io->md;
562 dm_endio_fn endio = tio->ti->type->end_io;
564 if (!bio_flagged(bio, BIO_UPTODATE) && !error)
568 r = endio(tio->ti, bio, error, &tio->info);
569 if (r < 0 || r == DM_ENDIO_REQUEUE)
571 * error and requeue request are handled
575 else if (r == DM_ENDIO_INCOMPLETE)
576 /* The target will handle the io */
579 DMWARN("unimplemented target endio return value: %d", r);
585 * Store md for cleanup instead of tio which is about to get freed.
587 bio->bi_private = md->bs;
591 dec_pending(io, error);
594 static sector_t max_io_len(struct mapped_device *md,
595 sector_t sector, struct dm_target *ti)
597 sector_t offset = sector - ti->begin;
598 sector_t len = ti->len - offset;
601 * Does the target need to split even further ?
605 boundary = ((offset + ti->split_io) & ~(ti->split_io - 1))
614 static void __map_bio(struct dm_target *ti, struct bio *clone,
615 struct dm_target_io *tio)
619 struct mapped_device *md;
624 BUG_ON(!clone->bi_size);
626 clone->bi_end_io = clone_endio;
627 clone->bi_private = tio;
630 * Map the clone. If r == 0 we don't need to do
631 * anything, the target has assumed ownership of
634 atomic_inc(&tio->io->io_count);
635 sector = clone->bi_sector;
636 r = ti->type->map(ti, clone, &tio->info);
637 if (r == DM_MAPIO_REMAPPED) {
638 /* the bio has been remapped so dispatch it */
640 trace_block_remap(bdev_get_queue(clone->bi_bdev), clone,
641 tio->io->bio->bi_bdev->bd_dev,
642 clone->bi_sector, sector);
644 generic_make_request(clone);
645 } else if (r < 0 || r == DM_MAPIO_REQUEUE) {
646 /* error the io and bail out, or requeue it if needed */
648 dec_pending(tio->io, r);
650 * Store bio_set for cleanup.
652 clone->bi_private = md->bs;
656 DMWARN("unimplemented target map return value: %d", r);
662 struct mapped_device *md;
663 struct dm_table *map;
667 sector_t sector_count;
671 static void dm_bio_destructor(struct bio *bio)
673 struct bio_set *bs = bio->bi_private;
679 * Creates a little bio that is just does part of a bvec.
681 static struct bio *split_bvec(struct bio *bio, sector_t sector,
682 unsigned short idx, unsigned int offset,
683 unsigned int len, struct bio_set *bs)
686 struct bio_vec *bv = bio->bi_io_vec + idx;
688 clone = bio_alloc_bioset(GFP_NOIO, 1, bs);
689 clone->bi_destructor = dm_bio_destructor;
690 *clone->bi_io_vec = *bv;
692 clone->bi_sector = sector;
693 clone->bi_bdev = bio->bi_bdev;
694 clone->bi_rw = bio->bi_rw;
696 clone->bi_size = to_bytes(len);
697 clone->bi_io_vec->bv_offset = offset;
698 clone->bi_io_vec->bv_len = clone->bi_size;
699 clone->bi_flags |= 1 << BIO_CLONED;
701 if (bio_integrity(bio)) {
702 bio_integrity_clone(clone, bio, GFP_NOIO);
703 bio_integrity_trim(clone,
704 bio_sector_offset(bio, idx, offset), len);
711 * Creates a bio that consists of range of complete bvecs.
713 static struct bio *clone_bio(struct bio *bio, sector_t sector,
714 unsigned short idx, unsigned short bv_count,
715 unsigned int len, struct bio_set *bs)
719 clone = bio_alloc_bioset(GFP_NOIO, bio->bi_max_vecs, bs);
720 __bio_clone(clone, bio);
721 clone->bi_destructor = dm_bio_destructor;
722 clone->bi_sector = sector;
724 clone->bi_vcnt = idx + bv_count;
725 clone->bi_size = to_bytes(len);
726 clone->bi_flags &= ~(1 << BIO_SEG_VALID);
728 if (bio_integrity(bio)) {
729 bio_integrity_clone(clone, bio, GFP_NOIO);
731 if (idx != bio->bi_idx || clone->bi_size < bio->bi_size)
732 bio_integrity_trim(clone,
733 bio_sector_offset(bio, idx, 0), len);
739 static int __clone_and_map(struct clone_info *ci)
741 struct bio *clone, *bio = ci->bio;
742 struct dm_target *ti;
743 sector_t len = 0, max;
744 struct dm_target_io *tio;
746 ti = dm_table_find_target(ci->map, ci->sector);
747 if (!dm_target_is_valid(ti))
750 max = max_io_len(ci->md, ci->sector, ti);
753 * Allocate a target io object.
755 tio = alloc_tio(ci->md);
758 memset(&tio->info, 0, sizeof(tio->info));
760 if (ci->sector_count <= max) {
762 * Optimise for the simple case where we can do all of
763 * the remaining io with a single clone.
765 clone = clone_bio(bio, ci->sector, ci->idx,
766 bio->bi_vcnt - ci->idx, ci->sector_count,
768 __map_bio(ti, clone, tio);
769 ci->sector_count = 0;
771 } else if (to_sector(bio->bi_io_vec[ci->idx].bv_len) <= max) {
773 * There are some bvecs that don't span targets.
774 * Do as many of these as possible.
777 sector_t remaining = max;
780 for (i = ci->idx; remaining && (i < bio->bi_vcnt); i++) {
781 bv_len = to_sector(bio->bi_io_vec[i].bv_len);
783 if (bv_len > remaining)
790 clone = clone_bio(bio, ci->sector, ci->idx, i - ci->idx, len,
792 __map_bio(ti, clone, tio);
795 ci->sector_count -= len;
800 * Handle a bvec that must be split between two or more targets.
802 struct bio_vec *bv = bio->bi_io_vec + ci->idx;
803 sector_t remaining = to_sector(bv->bv_len);
804 unsigned int offset = 0;
808 ti = dm_table_find_target(ci->map, ci->sector);
809 if (!dm_target_is_valid(ti))
812 max = max_io_len(ci->md, ci->sector, ti);
814 tio = alloc_tio(ci->md);
817 memset(&tio->info, 0, sizeof(tio->info));
820 len = min(remaining, max);
822 clone = split_bvec(bio, ci->sector, ci->idx,
823 bv->bv_offset + offset, len,
826 __map_bio(ti, clone, tio);
829 ci->sector_count -= len;
830 offset += to_bytes(len);
831 } while (remaining -= len);
840 * Split the bio into several clones and submit it to targets.
842 static void __split_and_process_bio(struct mapped_device *md, struct bio *bio)
844 struct clone_info ci;
847 ci.map = dm_get_table(md);
848 if (unlikely(!ci.map)) {
855 ci.io = alloc_io(md);
857 atomic_set(&ci.io->io_count, 1);
860 ci.sector = bio->bi_sector;
861 ci.sector_count = bio_sectors(bio);
862 ci.idx = bio->bi_idx;
864 start_io_acct(ci.io);
865 while (ci.sector_count && !error)
866 error = __clone_and_map(&ci);
868 /* drop the extra reference count */
869 dec_pending(ci.io, error);
870 dm_table_put(ci.map);
872 /*-----------------------------------------------------------------
874 *---------------------------------------------------------------*/
876 static int dm_merge_bvec(struct request_queue *q,
877 struct bvec_merge_data *bvm,
878 struct bio_vec *biovec)
880 struct mapped_device *md = q->queuedata;
881 struct dm_table *map = dm_get_table(md);
882 struct dm_target *ti;
883 sector_t max_sectors;
889 ti = dm_table_find_target(map, bvm->bi_sector);
890 if (!dm_target_is_valid(ti))
894 * Find maximum amount of I/O that won't need splitting
896 max_sectors = min(max_io_len(md, bvm->bi_sector, ti),
897 (sector_t) BIO_MAX_SECTORS);
898 max_size = (max_sectors << SECTOR_SHIFT) - bvm->bi_size;
903 * merge_bvec_fn() returns number of bytes
904 * it can accept at this offset
905 * max is precomputed maximal io size
907 if (max_size && ti->type->merge)
908 max_size = ti->type->merge(ti, bvm, biovec, max_size);
915 * Always allow an entire first page
917 if (max_size <= biovec->bv_len && !(bvm->bi_size >> SECTOR_SHIFT))
918 max_size = biovec->bv_len;
924 * The request function that just remaps the bio built up by
927 static int dm_request(struct request_queue *q, struct bio *bio)
929 int rw = bio_data_dir(bio);
930 struct mapped_device *md = q->queuedata;
934 * There is no use in forwarding any barrier request since we can't
935 * guarantee it is (or can be) handled by the targets correctly.
937 if (unlikely(bio_barrier(bio))) {
938 bio_endio(bio, -EOPNOTSUPP);
942 down_read(&md->io_lock);
944 cpu = part_stat_lock();
945 part_stat_inc(cpu, &dm_disk(md)->part0, ios[rw]);
946 part_stat_add(cpu, &dm_disk(md)->part0, sectors[rw], bio_sectors(bio));
950 * If we're suspended or the thread is processing barriers
951 * we have to queue this io for later.
953 if (unlikely(test_bit(DMF_QUEUE_IO_TO_THREAD, &md->flags))) {
954 up_read(&md->io_lock);
956 if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) &&
957 bio_rw(bio) == READA) {
967 __split_and_process_bio(md, bio);
968 up_read(&md->io_lock);
972 static void dm_unplug_all(struct request_queue *q)
974 struct mapped_device *md = q->queuedata;
975 struct dm_table *map = dm_get_table(md);
978 dm_table_unplug_all(map);
983 static int dm_any_congested(void *congested_data, int bdi_bits)
986 struct mapped_device *md = congested_data;
987 struct dm_table *map;
989 if (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
990 map = dm_get_table(md);
992 r = dm_table_any_congested(map, bdi_bits);
1000 /*-----------------------------------------------------------------
1001 * An IDR is used to keep track of allocated minor numbers.
1002 *---------------------------------------------------------------*/
1003 static DEFINE_IDR(_minor_idr);
1005 static void free_minor(int minor)
1007 spin_lock(&_minor_lock);
1008 idr_remove(&_minor_idr, minor);
1009 spin_unlock(&_minor_lock);
1013 * See if the device with a specific minor # is free.
1015 static int specific_minor(int minor)
1019 if (minor >= (1 << MINORBITS))
1022 r = idr_pre_get(&_minor_idr, GFP_KERNEL);
1026 spin_lock(&_minor_lock);
1028 if (idr_find(&_minor_idr, minor)) {
1033 r = idr_get_new_above(&_minor_idr, MINOR_ALLOCED, minor, &m);
1038 idr_remove(&_minor_idr, m);
1044 spin_unlock(&_minor_lock);
1048 static int next_free_minor(int *minor)
1052 r = idr_pre_get(&_minor_idr, GFP_KERNEL);
1056 spin_lock(&_minor_lock);
1058 r = idr_get_new(&_minor_idr, MINOR_ALLOCED, &m);
1062 if (m >= (1 << MINORBITS)) {
1063 idr_remove(&_minor_idr, m);
1071 spin_unlock(&_minor_lock);
1075 static struct block_device_operations dm_blk_dops;
1077 static void dm_wq_work(struct work_struct *work);
1080 * Allocate and initialise a blank device with a given minor.
1082 static struct mapped_device *alloc_dev(int minor)
1085 struct mapped_device *md = kzalloc(sizeof(*md), GFP_KERNEL);
1089 DMWARN("unable to allocate device, out of memory.");
1093 if (!try_module_get(THIS_MODULE))
1094 goto bad_module_get;
1096 /* get a minor number for the dev */
1097 if (minor == DM_ANY_MINOR)
1098 r = next_free_minor(&minor);
1100 r = specific_minor(minor);
1104 init_rwsem(&md->io_lock);
1105 mutex_init(&md->suspend_lock);
1106 spin_lock_init(&md->deferred_lock);
1107 rwlock_init(&md->map_lock);
1108 atomic_set(&md->holders, 1);
1109 atomic_set(&md->open_count, 0);
1110 atomic_set(&md->event_nr, 0);
1111 atomic_set(&md->uevent_seq, 0);
1112 INIT_LIST_HEAD(&md->uevent_list);
1113 spin_lock_init(&md->uevent_lock);
1115 md->queue = blk_alloc_queue(GFP_KERNEL);
1119 md->queue->queuedata = md;
1120 md->queue->backing_dev_info.congested_fn = dm_any_congested;
1121 md->queue->backing_dev_info.congested_data = md;
1122 blk_queue_make_request(md->queue, dm_request);
1123 blk_queue_ordered(md->queue, QUEUE_ORDERED_DRAIN, NULL);
1124 blk_queue_bounce_limit(md->queue, BLK_BOUNCE_ANY);
1125 md->queue->unplug_fn = dm_unplug_all;
1126 blk_queue_merge_bvec(md->queue, dm_merge_bvec);
1128 md->io_pool = mempool_create_slab_pool(MIN_IOS, _io_cache);
1132 md->tio_pool = mempool_create_slab_pool(MIN_IOS, _tio_cache);
1136 md->bs = bioset_create(16, 0);
1140 md->disk = alloc_disk(1);
1144 atomic_set(&md->pending, 0);
1145 init_waitqueue_head(&md->wait);
1146 INIT_WORK(&md->work, dm_wq_work);
1147 init_waitqueue_head(&md->eventq);
1149 md->disk->major = _major;
1150 md->disk->first_minor = minor;
1151 md->disk->fops = &dm_blk_dops;
1152 md->disk->queue = md->queue;
1153 md->disk->private_data = md;
1154 sprintf(md->disk->disk_name, "dm-%d", minor);
1156 format_dev_t(md->name, MKDEV(_major, minor));
1158 md->wq = create_singlethread_workqueue("kdmflush");
1162 /* Populate the mapping, nobody knows we exist yet */
1163 spin_lock(&_minor_lock);
1164 old_md = idr_replace(&_minor_idr, md, minor);
1165 spin_unlock(&_minor_lock);
1167 BUG_ON(old_md != MINOR_ALLOCED);
1174 bioset_free(md->bs);
1176 mempool_destroy(md->tio_pool);
1178 mempool_destroy(md->io_pool);
1180 blk_cleanup_queue(md->queue);
1184 module_put(THIS_MODULE);
1190 static void unlock_fs(struct mapped_device *md);
1192 static void free_dev(struct mapped_device *md)
1194 int minor = MINOR(disk_devt(md->disk));
1196 if (md->suspended_bdev) {
1198 bdput(md->suspended_bdev);
1200 destroy_workqueue(md->wq);
1201 mempool_destroy(md->tio_pool);
1202 mempool_destroy(md->io_pool);
1203 bioset_free(md->bs);
1204 blk_integrity_unregister(md->disk);
1205 del_gendisk(md->disk);
1208 spin_lock(&_minor_lock);
1209 md->disk->private_data = NULL;
1210 spin_unlock(&_minor_lock);
1213 blk_cleanup_queue(md->queue);
1214 module_put(THIS_MODULE);
1219 * Bind a table to the device.
1221 static void event_callback(void *context)
1223 unsigned long flags;
1225 struct mapped_device *md = (struct mapped_device *) context;
1227 spin_lock_irqsave(&md->uevent_lock, flags);
1228 list_splice_init(&md->uevent_list, &uevents);
1229 spin_unlock_irqrestore(&md->uevent_lock, flags);
1231 dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj);
1233 atomic_inc(&md->event_nr);
1234 wake_up(&md->eventq);
1237 static void __set_size(struct mapped_device *md, sector_t size)
1239 set_capacity(md->disk, size);
1241 mutex_lock(&md->suspended_bdev->bd_inode->i_mutex);
1242 i_size_write(md->suspended_bdev->bd_inode, (loff_t)size << SECTOR_SHIFT);
1243 mutex_unlock(&md->suspended_bdev->bd_inode->i_mutex);
1246 static int __bind(struct mapped_device *md, struct dm_table *t)
1248 struct request_queue *q = md->queue;
1251 size = dm_table_get_size(t);
1254 * Wipe any geometry if the size of the table changed.
1256 if (size != get_capacity(md->disk))
1257 memset(&md->geometry, 0, sizeof(md->geometry));
1259 if (md->suspended_bdev)
1260 __set_size(md, size);
1263 dm_table_destroy(t);
1267 dm_table_event_callback(t, event_callback, md);
1269 write_lock(&md->map_lock);
1271 dm_table_set_restrictions(t, q);
1272 write_unlock(&md->map_lock);
1277 static void __unbind(struct mapped_device *md)
1279 struct dm_table *map = md->map;
1284 dm_table_event_callback(map, NULL, NULL);
1285 write_lock(&md->map_lock);
1287 write_unlock(&md->map_lock);
1288 dm_table_destroy(map);
1292 * Constructor for a new device.
1294 int dm_create(int minor, struct mapped_device **result)
1296 struct mapped_device *md;
1298 md = alloc_dev(minor);
1308 static struct mapped_device *dm_find_md(dev_t dev)
1310 struct mapped_device *md;
1311 unsigned minor = MINOR(dev);
1313 if (MAJOR(dev) != _major || minor >= (1 << MINORBITS))
1316 spin_lock(&_minor_lock);
1318 md = idr_find(&_minor_idr, minor);
1319 if (md && (md == MINOR_ALLOCED ||
1320 (MINOR(disk_devt(dm_disk(md))) != minor) ||
1321 test_bit(DMF_FREEING, &md->flags))) {
1327 spin_unlock(&_minor_lock);
1332 struct mapped_device *dm_get_md(dev_t dev)
1334 struct mapped_device *md = dm_find_md(dev);
1342 void *dm_get_mdptr(struct mapped_device *md)
1344 return md->interface_ptr;
1347 void dm_set_mdptr(struct mapped_device *md, void *ptr)
1349 md->interface_ptr = ptr;
1352 void dm_get(struct mapped_device *md)
1354 atomic_inc(&md->holders);
1357 const char *dm_device_name(struct mapped_device *md)
1361 EXPORT_SYMBOL_GPL(dm_device_name);
1363 void dm_put(struct mapped_device *md)
1365 struct dm_table *map;
1367 BUG_ON(test_bit(DMF_FREEING, &md->flags));
1369 if (atomic_dec_and_lock(&md->holders, &_minor_lock)) {
1370 map = dm_get_table(md);
1371 idr_replace(&_minor_idr, MINOR_ALLOCED,
1372 MINOR(disk_devt(dm_disk(md))));
1373 set_bit(DMF_FREEING, &md->flags);
1374 spin_unlock(&_minor_lock);
1375 if (!dm_suspended(md)) {
1376 dm_table_presuspend_targets(map);
1377 dm_table_postsuspend_targets(map);
1385 EXPORT_SYMBOL_GPL(dm_put);
1387 static int dm_wait_for_completion(struct mapped_device *md, int interruptible)
1390 DECLARE_WAITQUEUE(wait, current);
1392 dm_unplug_all(md->queue);
1394 add_wait_queue(&md->wait, &wait);
1397 set_current_state(interruptible);
1400 if (!atomic_read(&md->pending))
1403 if (interruptible == TASK_INTERRUPTIBLE &&
1404 signal_pending(current)) {
1411 set_current_state(TASK_RUNNING);
1413 remove_wait_queue(&md->wait, &wait);
1419 * Process the deferred bios
1421 static void dm_wq_work(struct work_struct *work)
1423 struct mapped_device *md = container_of(work, struct mapped_device,
1427 down_write(&md->io_lock);
1429 while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
1430 spin_lock_irq(&md->deferred_lock);
1431 c = bio_list_pop(&md->deferred);
1432 spin_unlock_irq(&md->deferred_lock);
1435 clear_bit(DMF_QUEUE_IO_TO_THREAD, &md->flags);
1439 up_write(&md->io_lock);
1441 __split_and_process_bio(md, c);
1443 down_write(&md->io_lock);
1446 up_write(&md->io_lock);
1449 static void dm_queue_flush(struct mapped_device *md)
1451 clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
1452 smp_mb__after_clear_bit();
1453 queue_work(md->wq, &md->work);
1457 * Swap in a new table (destroying old one).
1459 int dm_swap_table(struct mapped_device *md, struct dm_table *table)
1463 mutex_lock(&md->suspend_lock);
1465 /* device must be suspended */
1466 if (!dm_suspended(md))
1469 /* without bdev, the device size cannot be changed */
1470 if (!md->suspended_bdev)
1471 if (get_capacity(md->disk) != dm_table_get_size(table))
1475 r = __bind(md, table);
1478 mutex_unlock(&md->suspend_lock);
1483 * Functions to lock and unlock any filesystem running on the
1486 static int lock_fs(struct mapped_device *md)
1490 WARN_ON(md->frozen_sb);
1492 md->frozen_sb = freeze_bdev(md->suspended_bdev);
1493 if (IS_ERR(md->frozen_sb)) {
1494 r = PTR_ERR(md->frozen_sb);
1495 md->frozen_sb = NULL;
1499 set_bit(DMF_FROZEN, &md->flags);
1501 /* don't bdput right now, we don't want the bdev
1502 * to go away while it is locked.
1507 static void unlock_fs(struct mapped_device *md)
1509 if (!test_bit(DMF_FROZEN, &md->flags))
1512 thaw_bdev(md->suspended_bdev, md->frozen_sb);
1513 md->frozen_sb = NULL;
1514 clear_bit(DMF_FROZEN, &md->flags);
1518 * We need to be able to change a mapping table under a mounted
1519 * filesystem. For example we might want to move some data in
1520 * the background. Before the table can be swapped with
1521 * dm_bind_table, dm_suspend must be called to flush any in
1522 * flight bios and ensure that any further io gets deferred.
1524 int dm_suspend(struct mapped_device *md, unsigned suspend_flags)
1526 struct dm_table *map = NULL;
1528 int do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG ? 1 : 0;
1529 int noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG ? 1 : 0;
1531 mutex_lock(&md->suspend_lock);
1533 if (dm_suspended(md)) {
1538 map = dm_get_table(md);
1541 * DMF_NOFLUSH_SUSPENDING must be set before presuspend.
1542 * This flag is cleared before dm_suspend returns.
1545 set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
1547 /* This does not get reverted if there's an error later. */
1548 dm_table_presuspend_targets(map);
1550 /* bdget() can stall if the pending I/Os are not flushed */
1552 md->suspended_bdev = bdget_disk(md->disk, 0);
1553 if (!md->suspended_bdev) {
1554 DMWARN("bdget failed in dm_suspend");
1560 * Flush I/O to the device. noflush supersedes do_lockfs,
1561 * because lock_fs() needs to flush I/Os.
1571 * Here we must make sure that no processes are submitting requests
1572 * to target drivers i.e. no one may be executing
1573 * __split_and_process_bio. This is called from dm_request and
1576 * To get all processes out of __split_and_process_bio in dm_request,
1577 * we take the write lock. To prevent any process from reentering
1578 * __split_and_process_bio from dm_request, we set
1579 * DMF_QUEUE_IO_TO_THREAD.
1581 * To quiesce the thread (dm_wq_work), we set DMF_BLOCK_IO_FOR_SUSPEND
1582 * and call flush_workqueue(md->wq). flush_workqueue will wait until
1583 * dm_wq_work exits and DMF_BLOCK_IO_FOR_SUSPEND will prevent any
1584 * further calls to __split_and_process_bio from dm_wq_work.
1586 down_write(&md->io_lock);
1587 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
1588 set_bit(DMF_QUEUE_IO_TO_THREAD, &md->flags);
1589 up_write(&md->io_lock);
1591 flush_workqueue(md->wq);
1594 * At this point no more requests are entering target request routines.
1595 * We call dm_wait_for_completion to wait for all existing requests
1598 r = dm_wait_for_completion(md, TASK_INTERRUPTIBLE);
1600 down_write(&md->io_lock);
1602 clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
1603 up_write(&md->io_lock);
1605 /* were we interrupted ? */
1610 goto out; /* pushback list is already flushed, so skip flush */
1614 * If dm_wait_for_completion returned 0, the device is completely
1615 * quiescent now. There is no request-processing activity. All new
1616 * requests are being added to md->deferred list.
1619 dm_table_postsuspend_targets(map);
1621 set_bit(DMF_SUSPENDED, &md->flags);
1624 if (r && md->suspended_bdev) {
1625 bdput(md->suspended_bdev);
1626 md->suspended_bdev = NULL;
1632 mutex_unlock(&md->suspend_lock);
1636 int dm_resume(struct mapped_device *md)
1639 struct dm_table *map = NULL;
1641 mutex_lock(&md->suspend_lock);
1642 if (!dm_suspended(md))
1645 map = dm_get_table(md);
1646 if (!map || !dm_table_get_size(map))
1649 r = dm_table_resume_targets(map);
1657 if (md->suspended_bdev) {
1658 bdput(md->suspended_bdev);
1659 md->suspended_bdev = NULL;
1662 clear_bit(DMF_SUSPENDED, &md->flags);
1664 dm_table_unplug_all(map);
1666 dm_kobject_uevent(md);
1672 mutex_unlock(&md->suspend_lock);
1677 /*-----------------------------------------------------------------
1678 * Event notification.
1679 *---------------------------------------------------------------*/
1680 void dm_kobject_uevent(struct mapped_device *md)
1682 kobject_uevent(&disk_to_dev(md->disk)->kobj, KOBJ_CHANGE);
1685 uint32_t dm_next_uevent_seq(struct mapped_device *md)
1687 return atomic_add_return(1, &md->uevent_seq);
1690 uint32_t dm_get_event_nr(struct mapped_device *md)
1692 return atomic_read(&md->event_nr);
1695 int dm_wait_event(struct mapped_device *md, int event_nr)
1697 return wait_event_interruptible(md->eventq,
1698 (event_nr != atomic_read(&md->event_nr)));
1701 void dm_uevent_add(struct mapped_device *md, struct list_head *elist)
1703 unsigned long flags;
1705 spin_lock_irqsave(&md->uevent_lock, flags);
1706 list_add(elist, &md->uevent_list);
1707 spin_unlock_irqrestore(&md->uevent_lock, flags);
1711 * The gendisk is only valid as long as you have a reference
1714 struct gendisk *dm_disk(struct mapped_device *md)
1719 struct kobject *dm_kobject(struct mapped_device *md)
1725 * struct mapped_device should not be exported outside of dm.c
1726 * so use this check to verify that kobj is part of md structure
1728 struct mapped_device *dm_get_from_kobject(struct kobject *kobj)
1730 struct mapped_device *md;
1732 md = container_of(kobj, struct mapped_device, kobj);
1733 if (&md->kobj != kobj)
1740 int dm_suspended(struct mapped_device *md)
1742 return test_bit(DMF_SUSPENDED, &md->flags);
1745 int dm_noflush_suspending(struct dm_target *ti)
1747 struct mapped_device *md = dm_table_get_md(ti->table);
1748 int r = __noflush_suspending(md);
1754 EXPORT_SYMBOL_GPL(dm_noflush_suspending);
1756 static struct block_device_operations dm_blk_dops = {
1757 .open = dm_blk_open,
1758 .release = dm_blk_close,
1759 .ioctl = dm_blk_ioctl,
1760 .getgeo = dm_blk_getgeo,
1761 .owner = THIS_MODULE
1764 EXPORT_SYMBOL(dm_get_mapinfo);
1769 module_init(dm_init);
1770 module_exit(dm_exit);
1772 module_param(major, uint, 0);
1773 MODULE_PARM_DESC(major, "The major number of the device mapper");
1774 MODULE_DESCRIPTION(DM_NAME " driver");
1775 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
1776 MODULE_LICENSE("GPL");