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 0
93 #define DMF_SUSPENDED 1
96 #define DMF_DELETING 4
97 #define DMF_NOFLUSH_SUSPENDING 5
100 * Work processed by per-device workqueue.
102 struct mapped_device {
103 struct rw_semaphore io_lock;
104 struct mutex suspend_lock;
111 struct request_queue *queue;
112 struct gendisk *disk;
118 * A list of ios that arrived while we were suspended.
121 wait_queue_head_t wait;
122 struct work_struct work;
123 struct bio_list deferred;
124 spinlock_t deferred_lock;
127 * Processing queue (flush/barriers)
129 struct workqueue_struct *wq;
132 * The current mapping.
134 struct dm_table *map;
137 * io objects are allocated from here.
148 wait_queue_head_t eventq;
150 struct list_head uevent_list;
151 spinlock_t uevent_lock; /* Protect access to uevent_list */
154 * freeze/thaw support require holding onto a super block
156 struct super_block *frozen_sb;
157 struct block_device *suspended_bdev;
159 /* forced geometry settings */
160 struct hd_geometry geometry;
167 static struct kmem_cache *_io_cache;
168 static struct kmem_cache *_tio_cache;
169 static struct kmem_cache *_rq_tio_cache;
170 static struct kmem_cache *_rq_bio_info_cache;
172 static int __init local_init(void)
176 /* allocate a slab for the dm_ios */
177 _io_cache = KMEM_CACHE(dm_io, 0);
181 /* allocate a slab for the target ios */
182 _tio_cache = KMEM_CACHE(dm_target_io, 0);
184 goto out_free_io_cache;
186 _rq_tio_cache = KMEM_CACHE(dm_rq_target_io, 0);
188 goto out_free_tio_cache;
190 _rq_bio_info_cache = KMEM_CACHE(dm_rq_clone_bio_info, 0);
191 if (!_rq_bio_info_cache)
192 goto out_free_rq_tio_cache;
194 r = dm_uevent_init();
196 goto out_free_rq_bio_info_cache;
199 r = register_blkdev(_major, _name);
201 goto out_uevent_exit;
210 out_free_rq_bio_info_cache:
211 kmem_cache_destroy(_rq_bio_info_cache);
212 out_free_rq_tio_cache:
213 kmem_cache_destroy(_rq_tio_cache);
215 kmem_cache_destroy(_tio_cache);
217 kmem_cache_destroy(_io_cache);
222 static void local_exit(void)
224 kmem_cache_destroy(_rq_bio_info_cache);
225 kmem_cache_destroy(_rq_tio_cache);
226 kmem_cache_destroy(_tio_cache);
227 kmem_cache_destroy(_io_cache);
228 unregister_blkdev(_major, _name);
233 DMINFO("cleaned up");
236 static int (*_inits[])(void) __initdata = {
245 static void (*_exits[])(void) = {
254 static int __init dm_init(void)
256 const int count = ARRAY_SIZE(_inits);
260 for (i = 0; i < count; i++) {
275 static void __exit dm_exit(void)
277 int i = ARRAY_SIZE(_exits);
284 * Block device functions
286 static int dm_blk_open(struct block_device *bdev, fmode_t mode)
288 struct mapped_device *md;
290 spin_lock(&_minor_lock);
292 md = bdev->bd_disk->private_data;
296 if (test_bit(DMF_FREEING, &md->flags) ||
297 test_bit(DMF_DELETING, &md->flags)) {
303 atomic_inc(&md->open_count);
306 spin_unlock(&_minor_lock);
308 return md ? 0 : -ENXIO;
311 static int dm_blk_close(struct gendisk *disk, fmode_t mode)
313 struct mapped_device *md = disk->private_data;
314 atomic_dec(&md->open_count);
319 int dm_open_count(struct mapped_device *md)
321 return atomic_read(&md->open_count);
325 * Guarantees nothing is using the device before it's deleted.
327 int dm_lock_for_deletion(struct mapped_device *md)
331 spin_lock(&_minor_lock);
333 if (dm_open_count(md))
336 set_bit(DMF_DELETING, &md->flags);
338 spin_unlock(&_minor_lock);
343 static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo)
345 struct mapped_device *md = bdev->bd_disk->private_data;
347 return dm_get_geometry(md, geo);
350 static int dm_blk_ioctl(struct block_device *bdev, fmode_t mode,
351 unsigned int cmd, unsigned long arg)
353 struct mapped_device *md = bdev->bd_disk->private_data;
354 struct dm_table *map = dm_get_table(md);
355 struct dm_target *tgt;
358 if (!map || !dm_table_get_size(map))
361 /* We only support devices that have a single target */
362 if (dm_table_get_num_targets(map) != 1)
365 tgt = dm_table_get_target(map, 0);
367 if (dm_suspended(md)) {
372 if (tgt->type->ioctl)
373 r = tgt->type->ioctl(tgt, cmd, arg);
381 static struct dm_io *alloc_io(struct mapped_device *md)
383 return mempool_alloc(md->io_pool, GFP_NOIO);
386 static void free_io(struct mapped_device *md, struct dm_io *io)
388 mempool_free(io, md->io_pool);
391 static struct dm_target_io *alloc_tio(struct mapped_device *md)
393 return mempool_alloc(md->tio_pool, GFP_NOIO);
396 static void free_tio(struct mapped_device *md, struct dm_target_io *tio)
398 mempool_free(tio, md->tio_pool);
401 static void start_io_acct(struct dm_io *io)
403 struct mapped_device *md = io->md;
406 io->start_time = jiffies;
408 cpu = part_stat_lock();
409 part_round_stats(cpu, &dm_disk(md)->part0);
411 dm_disk(md)->part0.in_flight = atomic_inc_return(&md->pending);
414 static void end_io_acct(struct dm_io *io)
416 struct mapped_device *md = io->md;
417 struct bio *bio = io->bio;
418 unsigned long duration = jiffies - io->start_time;
420 int rw = bio_data_dir(bio);
422 cpu = part_stat_lock();
423 part_round_stats(cpu, &dm_disk(md)->part0);
424 part_stat_add(cpu, &dm_disk(md)->part0, ticks[rw], duration);
427 dm_disk(md)->part0.in_flight = pending =
428 atomic_dec_return(&md->pending);
430 /* nudge anyone waiting on suspend queue */
436 * Add the bio to the list of deferred io.
438 static int queue_io(struct mapped_device *md, struct bio *bio)
440 down_write(&md->io_lock);
442 if (!test_bit(DMF_BLOCK_IO, &md->flags)) {
443 up_write(&md->io_lock);
447 spin_lock_irq(&md->deferred_lock);
448 bio_list_add(&md->deferred, bio);
449 spin_unlock_irq(&md->deferred_lock);
451 up_write(&md->io_lock);
452 return 0; /* deferred successfully */
456 * Everyone (including functions in this file), should use this
457 * function to access the md->map field, and make sure they call
458 * dm_table_put() when finished.
460 struct dm_table *dm_get_table(struct mapped_device *md)
464 read_lock(&md->map_lock);
468 read_unlock(&md->map_lock);
474 * Get the geometry associated with a dm device
476 int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo)
484 * Set the geometry of a device.
486 int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo)
488 sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors;
490 if (geo->start > sz) {
491 DMWARN("Start sector is beyond the geometry limits.");
500 /*-----------------------------------------------------------------
502 * A more elegant soln is in the works that uses the queue
503 * merge fn, unfortunately there are a couple of changes to
504 * the block layer that I want to make for this. So in the
505 * interests of getting something for people to use I give
506 * you this clearly demarcated crap.
507 *---------------------------------------------------------------*/
509 static int __noflush_suspending(struct mapped_device *md)
511 return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
515 * Decrements the number of outstanding ios that a bio has been
516 * cloned into, completing the original io if necc.
518 static void dec_pending(struct dm_io *io, int error)
523 struct mapped_device *md = io->md;
525 /* Push-back supersedes any I/O errors */
526 if (error && !(io->error > 0 && __noflush_suspending(md)))
529 if (atomic_dec_and_test(&io->io_count)) {
530 if (io->error == DM_ENDIO_REQUEUE) {
532 * Target requested pushing back the I/O.
534 spin_lock_irqsave(&md->deferred_lock, flags);
535 if (__noflush_suspending(md))
536 bio_list_add(&md->deferred, io->bio);
538 /* noflush suspend was interrupted. */
540 spin_unlock_irqrestore(&md->deferred_lock, flags);
545 io_error = io->error;
550 if (io_error != DM_ENDIO_REQUEUE) {
551 trace_block_bio_complete(md->queue, bio);
553 bio_endio(bio, io_error);
558 static void clone_endio(struct bio *bio, int error)
561 struct dm_target_io *tio = bio->bi_private;
562 struct dm_io *io = tio->io;
563 struct mapped_device *md = tio->io->md;
564 dm_endio_fn endio = tio->ti->type->end_io;
566 if (!bio_flagged(bio, BIO_UPTODATE) && !error)
570 r = endio(tio->ti, bio, error, &tio->info);
571 if (r < 0 || r == DM_ENDIO_REQUEUE)
573 * error and requeue request are handled
577 else if (r == DM_ENDIO_INCOMPLETE)
578 /* The target will handle the io */
581 DMWARN("unimplemented target endio return value: %d", r);
587 * Store md for cleanup instead of tio which is about to get freed.
589 bio->bi_private = md->bs;
593 dec_pending(io, error);
596 static sector_t max_io_len(struct mapped_device *md,
597 sector_t sector, struct dm_target *ti)
599 sector_t offset = sector - ti->begin;
600 sector_t len = ti->len - offset;
603 * Does the target need to split even further ?
607 boundary = ((offset + ti->split_io) & ~(ti->split_io - 1))
616 static void __map_bio(struct dm_target *ti, struct bio *clone,
617 struct dm_target_io *tio)
621 struct mapped_device *md;
626 BUG_ON(!clone->bi_size);
628 clone->bi_end_io = clone_endio;
629 clone->bi_private = tio;
632 * Map the clone. If r == 0 we don't need to do
633 * anything, the target has assumed ownership of
636 atomic_inc(&tio->io->io_count);
637 sector = clone->bi_sector;
638 r = ti->type->map(ti, clone, &tio->info);
639 if (r == DM_MAPIO_REMAPPED) {
640 /* the bio has been remapped so dispatch it */
642 trace_block_remap(bdev_get_queue(clone->bi_bdev), clone,
643 tio->io->bio->bi_bdev->bd_dev,
644 clone->bi_sector, sector);
646 generic_make_request(clone);
647 } else if (r < 0 || r == DM_MAPIO_REQUEUE) {
648 /* error the io and bail out, or requeue it if needed */
650 dec_pending(tio->io, r);
652 * Store bio_set for cleanup.
654 clone->bi_private = md->bs;
658 DMWARN("unimplemented target map return value: %d", r);
664 struct mapped_device *md;
665 struct dm_table *map;
669 sector_t sector_count;
673 static void dm_bio_destructor(struct bio *bio)
675 struct bio_set *bs = bio->bi_private;
681 * Creates a little bio that is just does part of a bvec.
683 static struct bio *split_bvec(struct bio *bio, sector_t sector,
684 unsigned short idx, unsigned int offset,
685 unsigned int len, struct bio_set *bs)
688 struct bio_vec *bv = bio->bi_io_vec + idx;
690 clone = bio_alloc_bioset(GFP_NOIO, 1, bs);
691 clone->bi_destructor = dm_bio_destructor;
692 *clone->bi_io_vec = *bv;
694 clone->bi_sector = sector;
695 clone->bi_bdev = bio->bi_bdev;
696 clone->bi_rw = bio->bi_rw;
698 clone->bi_size = to_bytes(len);
699 clone->bi_io_vec->bv_offset = offset;
700 clone->bi_io_vec->bv_len = clone->bi_size;
701 clone->bi_flags |= 1 << BIO_CLONED;
703 if (bio_integrity(bio)) {
704 bio_integrity_clone(clone, bio, GFP_NOIO);
705 bio_integrity_trim(clone,
706 bio_sector_offset(bio, idx, offset), len);
713 * Creates a bio that consists of range of complete bvecs.
715 static struct bio *clone_bio(struct bio *bio, sector_t sector,
716 unsigned short idx, unsigned short bv_count,
717 unsigned int len, struct bio_set *bs)
721 clone = bio_alloc_bioset(GFP_NOIO, bio->bi_max_vecs, bs);
722 __bio_clone(clone, bio);
723 clone->bi_destructor = dm_bio_destructor;
724 clone->bi_sector = sector;
726 clone->bi_vcnt = idx + bv_count;
727 clone->bi_size = to_bytes(len);
728 clone->bi_flags &= ~(1 << BIO_SEG_VALID);
730 if (bio_integrity(bio)) {
731 bio_integrity_clone(clone, bio, GFP_NOIO);
733 if (idx != bio->bi_idx || clone->bi_size < bio->bi_size)
734 bio_integrity_trim(clone,
735 bio_sector_offset(bio, idx, 0), len);
741 static int __clone_and_map(struct clone_info *ci)
743 struct bio *clone, *bio = ci->bio;
744 struct dm_target *ti;
745 sector_t len = 0, max;
746 struct dm_target_io *tio;
748 ti = dm_table_find_target(ci->map, ci->sector);
749 if (!dm_target_is_valid(ti))
752 max = max_io_len(ci->md, ci->sector, ti);
755 * Allocate a target io object.
757 tio = alloc_tio(ci->md);
760 memset(&tio->info, 0, sizeof(tio->info));
762 if (ci->sector_count <= max) {
764 * Optimise for the simple case where we can do all of
765 * the remaining io with a single clone.
767 clone = clone_bio(bio, ci->sector, ci->idx,
768 bio->bi_vcnt - ci->idx, ci->sector_count,
770 __map_bio(ti, clone, tio);
771 ci->sector_count = 0;
773 } else if (to_sector(bio->bi_io_vec[ci->idx].bv_len) <= max) {
775 * There are some bvecs that don't span targets.
776 * Do as many of these as possible.
779 sector_t remaining = max;
782 for (i = ci->idx; remaining && (i < bio->bi_vcnt); i++) {
783 bv_len = to_sector(bio->bi_io_vec[i].bv_len);
785 if (bv_len > remaining)
792 clone = clone_bio(bio, ci->sector, ci->idx, i - ci->idx, len,
794 __map_bio(ti, clone, tio);
797 ci->sector_count -= len;
802 * Handle a bvec that must be split between two or more targets.
804 struct bio_vec *bv = bio->bi_io_vec + ci->idx;
805 sector_t remaining = to_sector(bv->bv_len);
806 unsigned int offset = 0;
810 ti = dm_table_find_target(ci->map, ci->sector);
811 if (!dm_target_is_valid(ti))
814 max = max_io_len(ci->md, ci->sector, ti);
816 tio = alloc_tio(ci->md);
819 memset(&tio->info, 0, sizeof(tio->info));
822 len = min(remaining, max);
824 clone = split_bvec(bio, ci->sector, ci->idx,
825 bv->bv_offset + offset, len,
828 __map_bio(ti, clone, tio);
831 ci->sector_count -= len;
832 offset += to_bytes(len);
833 } while (remaining -= len);
842 * Split the bio into several clones and submit it to targets.
844 static void __split_and_process_bio(struct mapped_device *md, struct bio *bio)
846 struct clone_info ci;
849 ci.map = dm_get_table(md);
850 if (unlikely(!ci.map)) {
857 ci.io = alloc_io(md);
859 atomic_set(&ci.io->io_count, 1);
862 ci.sector = bio->bi_sector;
863 ci.sector_count = bio_sectors(bio);
864 ci.idx = bio->bi_idx;
866 start_io_acct(ci.io);
867 while (ci.sector_count && !error)
868 error = __clone_and_map(&ci);
870 /* drop the extra reference count */
871 dec_pending(ci.io, error);
872 dm_table_put(ci.map);
874 /*-----------------------------------------------------------------
876 *---------------------------------------------------------------*/
878 static int dm_merge_bvec(struct request_queue *q,
879 struct bvec_merge_data *bvm,
880 struct bio_vec *biovec)
882 struct mapped_device *md = q->queuedata;
883 struct dm_table *map = dm_get_table(md);
884 struct dm_target *ti;
885 sector_t max_sectors;
891 ti = dm_table_find_target(map, bvm->bi_sector);
892 if (!dm_target_is_valid(ti))
896 * Find maximum amount of I/O that won't need splitting
898 max_sectors = min(max_io_len(md, bvm->bi_sector, ti),
899 (sector_t) BIO_MAX_SECTORS);
900 max_size = (max_sectors << SECTOR_SHIFT) - bvm->bi_size;
905 * merge_bvec_fn() returns number of bytes
906 * it can accept at this offset
907 * max is precomputed maximal io size
909 if (max_size && ti->type->merge)
910 max_size = ti->type->merge(ti, bvm, biovec, max_size);
917 * Always allow an entire first page
919 if (max_size <= biovec->bv_len && !(bvm->bi_size >> SECTOR_SHIFT))
920 max_size = biovec->bv_len;
926 * The request function that just remaps the bio built up by
929 static int dm_request(struct request_queue *q, struct bio *bio)
932 int rw = bio_data_dir(bio);
933 struct mapped_device *md = q->queuedata;
937 * There is no use in forwarding any barrier request since we can't
938 * guarantee it is (or can be) handled by the targets correctly.
940 if (unlikely(bio_barrier(bio))) {
941 bio_endio(bio, -EOPNOTSUPP);
945 down_read(&md->io_lock);
947 cpu = part_stat_lock();
948 part_stat_inc(cpu, &dm_disk(md)->part0, ios[rw]);
949 part_stat_add(cpu, &dm_disk(md)->part0, sectors[rw], bio_sectors(bio));
953 * If we're suspended we have to queue
956 while (test_bit(DMF_BLOCK_IO, &md->flags)) {
957 up_read(&md->io_lock);
959 if (bio_rw(bio) != READA)
960 r = queue_io(md, bio);
966 * We're in a while loop, because someone could suspend
967 * before we get to the following read lock.
969 down_read(&md->io_lock);
972 __split_and_process_bio(md, bio);
973 up_read(&md->io_lock);
983 static void dm_unplug_all(struct request_queue *q)
985 struct mapped_device *md = q->queuedata;
986 struct dm_table *map = dm_get_table(md);
989 dm_table_unplug_all(map);
994 static int dm_any_congested(void *congested_data, int bdi_bits)
997 struct mapped_device *md = congested_data;
998 struct dm_table *map;
1000 if (!test_bit(DMF_BLOCK_IO, &md->flags)) {
1001 map = dm_get_table(md);
1003 r = dm_table_any_congested(map, bdi_bits);
1011 /*-----------------------------------------------------------------
1012 * An IDR is used to keep track of allocated minor numbers.
1013 *---------------------------------------------------------------*/
1014 static DEFINE_IDR(_minor_idr);
1016 static void free_minor(int minor)
1018 spin_lock(&_minor_lock);
1019 idr_remove(&_minor_idr, minor);
1020 spin_unlock(&_minor_lock);
1024 * See if the device with a specific minor # is free.
1026 static int specific_minor(int minor)
1030 if (minor >= (1 << MINORBITS))
1033 r = idr_pre_get(&_minor_idr, GFP_KERNEL);
1037 spin_lock(&_minor_lock);
1039 if (idr_find(&_minor_idr, minor)) {
1044 r = idr_get_new_above(&_minor_idr, MINOR_ALLOCED, minor, &m);
1049 idr_remove(&_minor_idr, m);
1055 spin_unlock(&_minor_lock);
1059 static int next_free_minor(int *minor)
1063 r = idr_pre_get(&_minor_idr, GFP_KERNEL);
1067 spin_lock(&_minor_lock);
1069 r = idr_get_new(&_minor_idr, MINOR_ALLOCED, &m);
1073 if (m >= (1 << MINORBITS)) {
1074 idr_remove(&_minor_idr, m);
1082 spin_unlock(&_minor_lock);
1086 static struct block_device_operations dm_blk_dops;
1088 static void dm_wq_work(struct work_struct *work);
1091 * Allocate and initialise a blank device with a given minor.
1093 static struct mapped_device *alloc_dev(int minor)
1096 struct mapped_device *md = kzalloc(sizeof(*md), GFP_KERNEL);
1100 DMWARN("unable to allocate device, out of memory.");
1104 if (!try_module_get(THIS_MODULE))
1105 goto bad_module_get;
1107 /* get a minor number for the dev */
1108 if (minor == DM_ANY_MINOR)
1109 r = next_free_minor(&minor);
1111 r = specific_minor(minor);
1115 init_rwsem(&md->io_lock);
1116 mutex_init(&md->suspend_lock);
1117 spin_lock_init(&md->deferred_lock);
1118 rwlock_init(&md->map_lock);
1119 atomic_set(&md->holders, 1);
1120 atomic_set(&md->open_count, 0);
1121 atomic_set(&md->event_nr, 0);
1122 atomic_set(&md->uevent_seq, 0);
1123 INIT_LIST_HEAD(&md->uevent_list);
1124 spin_lock_init(&md->uevent_lock);
1126 md->queue = blk_alloc_queue(GFP_KERNEL);
1130 md->queue->queuedata = md;
1131 md->queue->backing_dev_info.congested_fn = dm_any_congested;
1132 md->queue->backing_dev_info.congested_data = md;
1133 blk_queue_make_request(md->queue, dm_request);
1134 blk_queue_ordered(md->queue, QUEUE_ORDERED_DRAIN, NULL);
1135 blk_queue_bounce_limit(md->queue, BLK_BOUNCE_ANY);
1136 md->queue->unplug_fn = dm_unplug_all;
1137 blk_queue_merge_bvec(md->queue, dm_merge_bvec);
1139 md->io_pool = mempool_create_slab_pool(MIN_IOS, _io_cache);
1143 md->tio_pool = mempool_create_slab_pool(MIN_IOS, _tio_cache);
1147 md->bs = bioset_create(16, 0);
1151 md->disk = alloc_disk(1);
1155 atomic_set(&md->pending, 0);
1156 init_waitqueue_head(&md->wait);
1157 INIT_WORK(&md->work, dm_wq_work);
1158 init_waitqueue_head(&md->eventq);
1160 md->disk->major = _major;
1161 md->disk->first_minor = minor;
1162 md->disk->fops = &dm_blk_dops;
1163 md->disk->queue = md->queue;
1164 md->disk->private_data = md;
1165 sprintf(md->disk->disk_name, "dm-%d", minor);
1167 format_dev_t(md->name, MKDEV(_major, minor));
1169 md->wq = create_singlethread_workqueue("kdmflush");
1173 /* Populate the mapping, nobody knows we exist yet */
1174 spin_lock(&_minor_lock);
1175 old_md = idr_replace(&_minor_idr, md, minor);
1176 spin_unlock(&_minor_lock);
1178 BUG_ON(old_md != MINOR_ALLOCED);
1185 bioset_free(md->bs);
1187 mempool_destroy(md->tio_pool);
1189 mempool_destroy(md->io_pool);
1191 blk_cleanup_queue(md->queue);
1195 module_put(THIS_MODULE);
1201 static void unlock_fs(struct mapped_device *md);
1203 static void free_dev(struct mapped_device *md)
1205 int minor = MINOR(disk_devt(md->disk));
1207 if (md->suspended_bdev) {
1209 bdput(md->suspended_bdev);
1211 destroy_workqueue(md->wq);
1212 mempool_destroy(md->tio_pool);
1213 mempool_destroy(md->io_pool);
1214 bioset_free(md->bs);
1215 blk_integrity_unregister(md->disk);
1216 del_gendisk(md->disk);
1219 spin_lock(&_minor_lock);
1220 md->disk->private_data = NULL;
1221 spin_unlock(&_minor_lock);
1224 blk_cleanup_queue(md->queue);
1225 module_put(THIS_MODULE);
1230 * Bind a table to the device.
1232 static void event_callback(void *context)
1234 unsigned long flags;
1236 struct mapped_device *md = (struct mapped_device *) context;
1238 spin_lock_irqsave(&md->uevent_lock, flags);
1239 list_splice_init(&md->uevent_list, &uevents);
1240 spin_unlock_irqrestore(&md->uevent_lock, flags);
1242 dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj);
1244 atomic_inc(&md->event_nr);
1245 wake_up(&md->eventq);
1248 static void __set_size(struct mapped_device *md, sector_t size)
1250 set_capacity(md->disk, size);
1252 mutex_lock(&md->suspended_bdev->bd_inode->i_mutex);
1253 i_size_write(md->suspended_bdev->bd_inode, (loff_t)size << SECTOR_SHIFT);
1254 mutex_unlock(&md->suspended_bdev->bd_inode->i_mutex);
1257 static int __bind(struct mapped_device *md, struct dm_table *t)
1259 struct request_queue *q = md->queue;
1262 size = dm_table_get_size(t);
1265 * Wipe any geometry if the size of the table changed.
1267 if (size != get_capacity(md->disk))
1268 memset(&md->geometry, 0, sizeof(md->geometry));
1270 if (md->suspended_bdev)
1271 __set_size(md, size);
1274 dm_table_destroy(t);
1278 dm_table_event_callback(t, event_callback, md);
1280 write_lock(&md->map_lock);
1282 dm_table_set_restrictions(t, q);
1283 write_unlock(&md->map_lock);
1288 static void __unbind(struct mapped_device *md)
1290 struct dm_table *map = md->map;
1295 dm_table_event_callback(map, NULL, NULL);
1296 write_lock(&md->map_lock);
1298 write_unlock(&md->map_lock);
1299 dm_table_destroy(map);
1303 * Constructor for a new device.
1305 int dm_create(int minor, struct mapped_device **result)
1307 struct mapped_device *md;
1309 md = alloc_dev(minor);
1319 static struct mapped_device *dm_find_md(dev_t dev)
1321 struct mapped_device *md;
1322 unsigned minor = MINOR(dev);
1324 if (MAJOR(dev) != _major || minor >= (1 << MINORBITS))
1327 spin_lock(&_minor_lock);
1329 md = idr_find(&_minor_idr, minor);
1330 if (md && (md == MINOR_ALLOCED ||
1331 (MINOR(disk_devt(dm_disk(md))) != minor) ||
1332 test_bit(DMF_FREEING, &md->flags))) {
1338 spin_unlock(&_minor_lock);
1343 struct mapped_device *dm_get_md(dev_t dev)
1345 struct mapped_device *md = dm_find_md(dev);
1353 void *dm_get_mdptr(struct mapped_device *md)
1355 return md->interface_ptr;
1358 void dm_set_mdptr(struct mapped_device *md, void *ptr)
1360 md->interface_ptr = ptr;
1363 void dm_get(struct mapped_device *md)
1365 atomic_inc(&md->holders);
1368 const char *dm_device_name(struct mapped_device *md)
1372 EXPORT_SYMBOL_GPL(dm_device_name);
1374 void dm_put(struct mapped_device *md)
1376 struct dm_table *map;
1378 BUG_ON(test_bit(DMF_FREEING, &md->flags));
1380 if (atomic_dec_and_lock(&md->holders, &_minor_lock)) {
1381 map = dm_get_table(md);
1382 idr_replace(&_minor_idr, MINOR_ALLOCED,
1383 MINOR(disk_devt(dm_disk(md))));
1384 set_bit(DMF_FREEING, &md->flags);
1385 spin_unlock(&_minor_lock);
1386 if (!dm_suspended(md)) {
1387 dm_table_presuspend_targets(map);
1388 dm_table_postsuspend_targets(map);
1396 EXPORT_SYMBOL_GPL(dm_put);
1398 static int dm_wait_for_completion(struct mapped_device *md, int interruptible)
1401 DECLARE_WAITQUEUE(wait, current);
1403 dm_unplug_all(md->queue);
1405 add_wait_queue(&md->wait, &wait);
1408 set_current_state(interruptible);
1411 if (!atomic_read(&md->pending))
1414 if (interruptible == TASK_INTERRUPTIBLE &&
1415 signal_pending(current)) {
1422 set_current_state(TASK_RUNNING);
1424 remove_wait_queue(&md->wait, &wait);
1430 * Process the deferred bios
1432 static void dm_wq_work(struct work_struct *work)
1434 struct mapped_device *md = container_of(work, struct mapped_device,
1438 down_write(&md->io_lock);
1441 spin_lock_irq(&md->deferred_lock);
1442 c = bio_list_pop(&md->deferred);
1443 spin_unlock_irq(&md->deferred_lock);
1446 __split_and_process_bio(md, c);
1450 clear_bit(DMF_BLOCK_IO, &md->flags);
1452 up_write(&md->io_lock);
1455 static void dm_queue_flush(struct mapped_device *md)
1457 queue_work(md->wq, &md->work);
1458 flush_workqueue(md->wq);
1462 * Swap in a new table (destroying old one).
1464 int dm_swap_table(struct mapped_device *md, struct dm_table *table)
1468 mutex_lock(&md->suspend_lock);
1470 /* device must be suspended */
1471 if (!dm_suspended(md))
1474 /* without bdev, the device size cannot be changed */
1475 if (!md->suspended_bdev)
1476 if (get_capacity(md->disk) != dm_table_get_size(table))
1480 r = __bind(md, table);
1483 mutex_unlock(&md->suspend_lock);
1488 * Functions to lock and unlock any filesystem running on the
1491 static int lock_fs(struct mapped_device *md)
1495 WARN_ON(md->frozen_sb);
1497 md->frozen_sb = freeze_bdev(md->suspended_bdev);
1498 if (IS_ERR(md->frozen_sb)) {
1499 r = PTR_ERR(md->frozen_sb);
1500 md->frozen_sb = NULL;
1504 set_bit(DMF_FROZEN, &md->flags);
1506 /* don't bdput right now, we don't want the bdev
1507 * to go away while it is locked.
1512 static void unlock_fs(struct mapped_device *md)
1514 if (!test_bit(DMF_FROZEN, &md->flags))
1517 thaw_bdev(md->suspended_bdev, md->frozen_sb);
1518 md->frozen_sb = NULL;
1519 clear_bit(DMF_FROZEN, &md->flags);
1523 * We need to be able to change a mapping table under a mounted
1524 * filesystem. For example we might want to move some data in
1525 * the background. Before the table can be swapped with
1526 * dm_bind_table, dm_suspend must be called to flush any in
1527 * flight bios and ensure that any further io gets deferred.
1529 int dm_suspend(struct mapped_device *md, unsigned suspend_flags)
1531 struct dm_table *map = NULL;
1533 int do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG ? 1 : 0;
1534 int noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG ? 1 : 0;
1536 mutex_lock(&md->suspend_lock);
1538 if (dm_suspended(md)) {
1543 map = dm_get_table(md);
1546 * DMF_NOFLUSH_SUSPENDING must be set before presuspend.
1547 * This flag is cleared before dm_suspend returns.
1550 set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
1552 /* This does not get reverted if there's an error later. */
1553 dm_table_presuspend_targets(map);
1555 /* bdget() can stall if the pending I/Os are not flushed */
1557 md->suspended_bdev = bdget_disk(md->disk, 0);
1558 if (!md->suspended_bdev) {
1559 DMWARN("bdget failed in dm_suspend");
1565 * Flush I/O to the device. noflush supersedes do_lockfs,
1566 * because lock_fs() needs to flush I/Os.
1576 * First we set the BLOCK_IO flag so no more ios will be mapped.
1578 down_write(&md->io_lock);
1579 set_bit(DMF_BLOCK_IO, &md->flags);
1581 up_write(&md->io_lock);
1584 * Wait for the already-mapped ios to complete.
1586 r = dm_wait_for_completion(md, TASK_INTERRUPTIBLE);
1588 down_write(&md->io_lock);
1591 clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
1592 up_write(&md->io_lock);
1594 /* were we interrupted ? */
1599 goto out; /* pushback list is already flushed, so skip flush */
1602 dm_table_postsuspend_targets(map);
1604 set_bit(DMF_SUSPENDED, &md->flags);
1607 if (r && md->suspended_bdev) {
1608 bdput(md->suspended_bdev);
1609 md->suspended_bdev = NULL;
1615 mutex_unlock(&md->suspend_lock);
1619 int dm_resume(struct mapped_device *md)
1622 struct dm_table *map = NULL;
1624 mutex_lock(&md->suspend_lock);
1625 if (!dm_suspended(md))
1628 map = dm_get_table(md);
1629 if (!map || !dm_table_get_size(map))
1632 r = dm_table_resume_targets(map);
1640 if (md->suspended_bdev) {
1641 bdput(md->suspended_bdev);
1642 md->suspended_bdev = NULL;
1645 clear_bit(DMF_SUSPENDED, &md->flags);
1647 dm_table_unplug_all(map);
1649 dm_kobject_uevent(md);
1655 mutex_unlock(&md->suspend_lock);
1660 /*-----------------------------------------------------------------
1661 * Event notification.
1662 *---------------------------------------------------------------*/
1663 void dm_kobject_uevent(struct mapped_device *md)
1665 kobject_uevent(&disk_to_dev(md->disk)->kobj, KOBJ_CHANGE);
1668 uint32_t dm_next_uevent_seq(struct mapped_device *md)
1670 return atomic_add_return(1, &md->uevent_seq);
1673 uint32_t dm_get_event_nr(struct mapped_device *md)
1675 return atomic_read(&md->event_nr);
1678 int dm_wait_event(struct mapped_device *md, int event_nr)
1680 return wait_event_interruptible(md->eventq,
1681 (event_nr != atomic_read(&md->event_nr)));
1684 void dm_uevent_add(struct mapped_device *md, struct list_head *elist)
1686 unsigned long flags;
1688 spin_lock_irqsave(&md->uevent_lock, flags);
1689 list_add(elist, &md->uevent_list);
1690 spin_unlock_irqrestore(&md->uevent_lock, flags);
1694 * The gendisk is only valid as long as you have a reference
1697 struct gendisk *dm_disk(struct mapped_device *md)
1702 struct kobject *dm_kobject(struct mapped_device *md)
1708 * struct mapped_device should not be exported outside of dm.c
1709 * so use this check to verify that kobj is part of md structure
1711 struct mapped_device *dm_get_from_kobject(struct kobject *kobj)
1713 struct mapped_device *md;
1715 md = container_of(kobj, struct mapped_device, kobj);
1716 if (&md->kobj != kobj)
1723 int dm_suspended(struct mapped_device *md)
1725 return test_bit(DMF_SUSPENDED, &md->flags);
1728 int dm_noflush_suspending(struct dm_target *ti)
1730 struct mapped_device *md = dm_table_get_md(ti->table);
1731 int r = __noflush_suspending(md);
1737 EXPORT_SYMBOL_GPL(dm_noflush_suspending);
1739 static struct block_device_operations dm_blk_dops = {
1740 .open = dm_blk_open,
1741 .release = dm_blk_close,
1742 .ioctl = dm_blk_ioctl,
1743 .getgeo = dm_blk_getgeo,
1744 .owner = THIS_MODULE
1747 EXPORT_SYMBOL(dm_get_mapinfo);
1752 module_init(dm_init);
1753 module_exit(dm_exit);
1755 module_param(major, uint, 0);
1756 MODULE_PARM_DESC(major, "The major number of the device mapper");
1757 MODULE_DESCRIPTION(DM_NAME " driver");
1759 MODULE_LICENSE("GPL");