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.
104 DM_WQ_FLUSH_DEFERRED,
106 struct work_struct work;
107 struct mapped_device *md;
111 struct mapped_device {
112 struct rw_semaphore io_lock;
113 struct mutex suspend_lock;
114 spinlock_t pushback_lock;
121 struct request_queue *queue;
122 struct gendisk *disk;
128 * A list of ios that arrived while we were suspended.
131 wait_queue_head_t wait;
132 struct bio_list deferred;
133 struct bio_list pushback;
136 * Processing queue (flush/barriers)
138 struct workqueue_struct *wq;
141 * The current mapping.
143 struct dm_table *map;
146 * io objects are allocated from here.
157 wait_queue_head_t eventq;
159 struct list_head uevent_list;
160 spinlock_t uevent_lock; /* Protect access to uevent_list */
163 * freeze/thaw support require holding onto a super block
165 struct super_block *frozen_sb;
166 struct block_device *suspended_bdev;
168 /* forced geometry settings */
169 struct hd_geometry geometry;
176 static struct kmem_cache *_io_cache;
177 static struct kmem_cache *_tio_cache;
178 static struct kmem_cache *_rq_tio_cache;
179 static struct kmem_cache *_rq_bio_info_cache;
181 static int __init local_init(void)
185 /* allocate a slab for the dm_ios */
186 _io_cache = KMEM_CACHE(dm_io, 0);
190 /* allocate a slab for the target ios */
191 _tio_cache = KMEM_CACHE(dm_target_io, 0);
193 goto out_free_io_cache;
195 _rq_tio_cache = KMEM_CACHE(dm_rq_target_io, 0);
197 goto out_free_tio_cache;
199 _rq_bio_info_cache = KMEM_CACHE(dm_rq_clone_bio_info, 0);
200 if (!_rq_bio_info_cache)
201 goto out_free_rq_tio_cache;
203 r = dm_uevent_init();
205 goto out_free_rq_bio_info_cache;
208 r = register_blkdev(_major, _name);
210 goto out_uevent_exit;
219 out_free_rq_bio_info_cache:
220 kmem_cache_destroy(_rq_bio_info_cache);
221 out_free_rq_tio_cache:
222 kmem_cache_destroy(_rq_tio_cache);
224 kmem_cache_destroy(_tio_cache);
226 kmem_cache_destroy(_io_cache);
231 static void local_exit(void)
233 kmem_cache_destroy(_rq_bio_info_cache);
234 kmem_cache_destroy(_rq_tio_cache);
235 kmem_cache_destroy(_tio_cache);
236 kmem_cache_destroy(_io_cache);
237 unregister_blkdev(_major, _name);
242 DMINFO("cleaned up");
245 static int (*_inits[])(void) __initdata = {
254 static void (*_exits[])(void) = {
263 static int __init dm_init(void)
265 const int count = ARRAY_SIZE(_inits);
269 for (i = 0; i < count; i++) {
284 static void __exit dm_exit(void)
286 int i = ARRAY_SIZE(_exits);
293 * Block device functions
295 static int dm_blk_open(struct block_device *bdev, fmode_t mode)
297 struct mapped_device *md;
299 spin_lock(&_minor_lock);
301 md = bdev->bd_disk->private_data;
305 if (test_bit(DMF_FREEING, &md->flags) ||
306 test_bit(DMF_DELETING, &md->flags)) {
312 atomic_inc(&md->open_count);
315 spin_unlock(&_minor_lock);
317 return md ? 0 : -ENXIO;
320 static int dm_blk_close(struct gendisk *disk, fmode_t mode)
322 struct mapped_device *md = disk->private_data;
323 atomic_dec(&md->open_count);
328 int dm_open_count(struct mapped_device *md)
330 return atomic_read(&md->open_count);
334 * Guarantees nothing is using the device before it's deleted.
336 int dm_lock_for_deletion(struct mapped_device *md)
340 spin_lock(&_minor_lock);
342 if (dm_open_count(md))
345 set_bit(DMF_DELETING, &md->flags);
347 spin_unlock(&_minor_lock);
352 static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo)
354 struct mapped_device *md = bdev->bd_disk->private_data;
356 return dm_get_geometry(md, geo);
359 static int dm_blk_ioctl(struct block_device *bdev, fmode_t mode,
360 unsigned int cmd, unsigned long arg)
362 struct mapped_device *md = bdev->bd_disk->private_data;
363 struct dm_table *map = dm_get_table(md);
364 struct dm_target *tgt;
367 if (!map || !dm_table_get_size(map))
370 /* We only support devices that have a single target */
371 if (dm_table_get_num_targets(map) != 1)
374 tgt = dm_table_get_target(map, 0);
376 if (dm_suspended(md)) {
381 if (tgt->type->ioctl)
382 r = tgt->type->ioctl(tgt, cmd, arg);
390 static struct dm_io *alloc_io(struct mapped_device *md)
392 return mempool_alloc(md->io_pool, GFP_NOIO);
395 static void free_io(struct mapped_device *md, struct dm_io *io)
397 mempool_free(io, md->io_pool);
400 static struct dm_target_io *alloc_tio(struct mapped_device *md)
402 return mempool_alloc(md->tio_pool, GFP_NOIO);
405 static void free_tio(struct mapped_device *md, struct dm_target_io *tio)
407 mempool_free(tio, md->tio_pool);
410 static void start_io_acct(struct dm_io *io)
412 struct mapped_device *md = io->md;
415 io->start_time = jiffies;
417 cpu = part_stat_lock();
418 part_round_stats(cpu, &dm_disk(md)->part0);
420 dm_disk(md)->part0.in_flight = atomic_inc_return(&md->pending);
423 static void end_io_acct(struct dm_io *io)
425 struct mapped_device *md = io->md;
426 struct bio *bio = io->bio;
427 unsigned long duration = jiffies - io->start_time;
429 int rw = bio_data_dir(bio);
431 cpu = part_stat_lock();
432 part_round_stats(cpu, &dm_disk(md)->part0);
433 part_stat_add(cpu, &dm_disk(md)->part0, ticks[rw], duration);
436 dm_disk(md)->part0.in_flight = pending =
437 atomic_dec_return(&md->pending);
439 /* nudge anyone waiting on suspend queue */
445 * Add the bio to the list of deferred io.
447 static int queue_io(struct mapped_device *md, struct bio *bio)
449 down_write(&md->io_lock);
451 if (!test_bit(DMF_BLOCK_IO, &md->flags)) {
452 up_write(&md->io_lock);
456 bio_list_add(&md->deferred, bio);
458 up_write(&md->io_lock);
459 return 0; /* deferred successfully */
463 * Everyone (including functions in this file), should use this
464 * function to access the md->map field, and make sure they call
465 * dm_table_put() when finished.
467 struct dm_table *dm_get_table(struct mapped_device *md)
471 read_lock(&md->map_lock);
475 read_unlock(&md->map_lock);
481 * Get the geometry associated with a dm device
483 int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo)
491 * Set the geometry of a device.
493 int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo)
495 sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors;
497 if (geo->start > sz) {
498 DMWARN("Start sector is beyond the geometry limits.");
507 /*-----------------------------------------------------------------
509 * A more elegant soln is in the works that uses the queue
510 * merge fn, unfortunately there are a couple of changes to
511 * the block layer that I want to make for this. So in the
512 * interests of getting something for people to use I give
513 * you this clearly demarcated crap.
514 *---------------------------------------------------------------*/
516 static int __noflush_suspending(struct mapped_device *md)
518 return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
522 * Decrements the number of outstanding ios that a bio has been
523 * cloned into, completing the original io if necc.
525 static void dec_pending(struct dm_io *io, int error)
529 /* Push-back supersedes any I/O errors */
530 if (error && !(io->error > 0 && __noflush_suspending(io->md)))
533 if (atomic_dec_and_test(&io->io_count)) {
534 if (io->error == DM_ENDIO_REQUEUE) {
536 * Target requested pushing back the I/O.
537 * This must be handled before the sleeper on
538 * suspend queue merges the pushback list.
540 spin_lock_irqsave(&io->md->pushback_lock, flags);
541 if (__noflush_suspending(io->md))
542 bio_list_add(&io->md->pushback, io->bio);
544 /* noflush suspend was interrupted. */
546 spin_unlock_irqrestore(&io->md->pushback_lock, flags);
551 if (io->error != DM_ENDIO_REQUEUE) {
552 trace_block_bio_complete(io->md->queue, io->bio);
554 bio_endio(io->bio, io->error);
561 static void clone_endio(struct bio *bio, int error)
564 struct dm_target_io *tio = bio->bi_private;
565 struct mapped_device *md = tio->io->md;
566 dm_endio_fn endio = tio->ti->type->end_io;
568 if (!bio_flagged(bio, BIO_UPTODATE) && !error)
572 r = endio(tio->ti, bio, error, &tio->info);
573 if (r < 0 || r == DM_ENDIO_REQUEUE)
575 * error and requeue request are handled
579 else if (r == DM_ENDIO_INCOMPLETE)
580 /* The target will handle the io */
583 DMWARN("unimplemented target endio return value: %d", r);
588 dec_pending(tio->io, error);
591 * Store md for cleanup instead of tio which is about to get freed.
593 bio->bi_private = md->bs;
599 static sector_t max_io_len(struct mapped_device *md,
600 sector_t sector, struct dm_target *ti)
602 sector_t offset = sector - ti->begin;
603 sector_t len = ti->len - offset;
606 * Does the target need to split even further ?
610 boundary = ((offset + ti->split_io) & ~(ti->split_io - 1))
619 static void __map_bio(struct dm_target *ti, struct bio *clone,
620 struct dm_target_io *tio)
624 struct mapped_device *md;
629 BUG_ON(!clone->bi_size);
631 clone->bi_end_io = clone_endio;
632 clone->bi_private = tio;
635 * Map the clone. If r == 0 we don't need to do
636 * anything, the target has assumed ownership of
639 atomic_inc(&tio->io->io_count);
640 sector = clone->bi_sector;
641 r = ti->type->map(ti, clone, &tio->info);
642 if (r == DM_MAPIO_REMAPPED) {
643 /* the bio has been remapped so dispatch it */
645 trace_block_remap(bdev_get_queue(clone->bi_bdev), clone,
646 tio->io->bio->bi_bdev->bd_dev,
647 clone->bi_sector, sector);
649 generic_make_request(clone);
650 } else if (r < 0 || r == DM_MAPIO_REQUEUE) {
651 /* error the io and bail out, or requeue it if needed */
653 dec_pending(tio->io, r);
655 * Store bio_set for cleanup.
657 clone->bi_private = md->bs;
661 DMWARN("unimplemented target map return value: %d", r);
667 struct mapped_device *md;
668 struct dm_table *map;
672 sector_t sector_count;
676 static void dm_bio_destructor(struct bio *bio)
678 struct bio_set *bs = bio->bi_private;
684 * Creates a little bio that is just does part of a bvec.
686 static struct bio *split_bvec(struct bio *bio, sector_t sector,
687 unsigned short idx, unsigned int offset,
688 unsigned int len, struct bio_set *bs)
691 struct bio_vec *bv = bio->bi_io_vec + idx;
693 clone = bio_alloc_bioset(GFP_NOIO, 1, bs);
694 clone->bi_destructor = dm_bio_destructor;
695 *clone->bi_io_vec = *bv;
697 clone->bi_sector = sector;
698 clone->bi_bdev = bio->bi_bdev;
699 clone->bi_rw = bio->bi_rw;
701 clone->bi_size = to_bytes(len);
702 clone->bi_io_vec->bv_offset = offset;
703 clone->bi_io_vec->bv_len = clone->bi_size;
704 clone->bi_flags |= 1 << BIO_CLONED;
710 * Creates a bio that consists of range of complete bvecs.
712 static struct bio *clone_bio(struct bio *bio, sector_t sector,
713 unsigned short idx, unsigned short bv_count,
714 unsigned int len, struct bio_set *bs)
718 clone = bio_alloc_bioset(GFP_NOIO, bio->bi_max_vecs, bs);
719 __bio_clone(clone, bio);
720 clone->bi_destructor = dm_bio_destructor;
721 clone->bi_sector = sector;
723 clone->bi_vcnt = idx + bv_count;
724 clone->bi_size = to_bytes(len);
725 clone->bi_flags &= ~(1 << BIO_SEG_VALID);
730 static int __clone_and_map(struct clone_info *ci)
732 struct bio *clone, *bio = ci->bio;
733 struct dm_target *ti;
734 sector_t len = 0, max;
735 struct dm_target_io *tio;
737 ti = dm_table_find_target(ci->map, ci->sector);
738 if (!dm_target_is_valid(ti))
741 max = max_io_len(ci->md, ci->sector, ti);
744 * Allocate a target io object.
746 tio = alloc_tio(ci->md);
749 memset(&tio->info, 0, sizeof(tio->info));
751 if (ci->sector_count <= max) {
753 * Optimise for the simple case where we can do all of
754 * the remaining io with a single clone.
756 clone = clone_bio(bio, ci->sector, ci->idx,
757 bio->bi_vcnt - ci->idx, ci->sector_count,
759 __map_bio(ti, clone, tio);
760 ci->sector_count = 0;
762 } else if (to_sector(bio->bi_io_vec[ci->idx].bv_len) <= max) {
764 * There are some bvecs that don't span targets.
765 * Do as many of these as possible.
768 sector_t remaining = max;
771 for (i = ci->idx; remaining && (i < bio->bi_vcnt); i++) {
772 bv_len = to_sector(bio->bi_io_vec[i].bv_len);
774 if (bv_len > remaining)
781 clone = clone_bio(bio, ci->sector, ci->idx, i - ci->idx, len,
783 __map_bio(ti, clone, tio);
786 ci->sector_count -= len;
791 * Handle a bvec that must be split between two or more targets.
793 struct bio_vec *bv = bio->bi_io_vec + ci->idx;
794 sector_t remaining = to_sector(bv->bv_len);
795 unsigned int offset = 0;
799 ti = dm_table_find_target(ci->map, ci->sector);
800 if (!dm_target_is_valid(ti))
803 max = max_io_len(ci->md, ci->sector, ti);
805 tio = alloc_tio(ci->md);
808 memset(&tio->info, 0, sizeof(tio->info));
811 len = min(remaining, max);
813 clone = split_bvec(bio, ci->sector, ci->idx,
814 bv->bv_offset + offset, len,
817 __map_bio(ti, clone, tio);
820 ci->sector_count -= len;
821 offset += to_bytes(len);
822 } while (remaining -= len);
831 * Split the bio into several clones.
833 static int __split_bio(struct mapped_device *md, struct bio *bio)
835 struct clone_info ci;
838 ci.map = dm_get_table(md);
839 if (unlikely(!ci.map))
841 if (unlikely(bio_barrier(bio) && !dm_table_barrier_ok(ci.map))) {
842 dm_table_put(ci.map);
843 bio_endio(bio, -EOPNOTSUPP);
848 ci.io = alloc_io(md);
850 atomic_set(&ci.io->io_count, 1);
853 ci.sector = bio->bi_sector;
854 ci.sector_count = bio_sectors(bio);
855 ci.idx = bio->bi_idx;
857 start_io_acct(ci.io);
858 while (ci.sector_count && !error)
859 error = __clone_and_map(&ci);
861 /* drop the extra reference count */
862 dec_pending(ci.io, error);
863 dm_table_put(ci.map);
867 /*-----------------------------------------------------------------
869 *---------------------------------------------------------------*/
871 static int dm_merge_bvec(struct request_queue *q,
872 struct bvec_merge_data *bvm,
873 struct bio_vec *biovec)
875 struct mapped_device *md = q->queuedata;
876 struct dm_table *map = dm_get_table(md);
877 struct dm_target *ti;
878 sector_t max_sectors;
884 ti = dm_table_find_target(map, bvm->bi_sector);
885 if (!dm_target_is_valid(ti))
889 * Find maximum amount of I/O that won't need splitting
891 max_sectors = min(max_io_len(md, bvm->bi_sector, ti),
892 (sector_t) BIO_MAX_SECTORS);
893 max_size = (max_sectors << SECTOR_SHIFT) - bvm->bi_size;
898 * merge_bvec_fn() returns number of bytes
899 * it can accept at this offset
900 * max is precomputed maximal io size
902 if (max_size && ti->type->merge)
903 max_size = ti->type->merge(ti, bvm, biovec, max_size);
910 * Always allow an entire first page
912 if (max_size <= biovec->bv_len && !(bvm->bi_size >> SECTOR_SHIFT))
913 max_size = biovec->bv_len;
919 * The request function that just remaps the bio built up by
922 static int dm_request(struct request_queue *q, struct bio *bio)
925 int rw = bio_data_dir(bio);
926 struct mapped_device *md = q->queuedata;
929 down_read(&md->io_lock);
931 cpu = part_stat_lock();
932 part_stat_inc(cpu, &dm_disk(md)->part0, ios[rw]);
933 part_stat_add(cpu, &dm_disk(md)->part0, sectors[rw], bio_sectors(bio));
937 * If we're suspended we have to queue
940 while (test_bit(DMF_BLOCK_IO, &md->flags)) {
941 up_read(&md->io_lock);
943 if (bio_rw(bio) != READA)
944 r = queue_io(md, bio);
950 * We're in a while loop, because someone could suspend
951 * before we get to the following read lock.
953 down_read(&md->io_lock);
956 r = __split_bio(md, bio);
957 up_read(&md->io_lock);
966 static void dm_unplug_all(struct request_queue *q)
968 struct mapped_device *md = q->queuedata;
969 struct dm_table *map = dm_get_table(md);
972 dm_table_unplug_all(map);
977 static int dm_any_congested(void *congested_data, int bdi_bits)
980 struct mapped_device *md = congested_data;
981 struct dm_table *map;
983 if (!test_bit(DMF_BLOCK_IO, &md->flags)) {
984 map = dm_get_table(md);
986 r = dm_table_any_congested(map, bdi_bits);
994 /*-----------------------------------------------------------------
995 * An IDR is used to keep track of allocated minor numbers.
996 *---------------------------------------------------------------*/
997 static DEFINE_IDR(_minor_idr);
999 static void free_minor(int minor)
1001 spin_lock(&_minor_lock);
1002 idr_remove(&_minor_idr, minor);
1003 spin_unlock(&_minor_lock);
1007 * See if the device with a specific minor # is free.
1009 static int specific_minor(int minor)
1013 if (minor >= (1 << MINORBITS))
1016 r = idr_pre_get(&_minor_idr, GFP_KERNEL);
1020 spin_lock(&_minor_lock);
1022 if (idr_find(&_minor_idr, minor)) {
1027 r = idr_get_new_above(&_minor_idr, MINOR_ALLOCED, minor, &m);
1032 idr_remove(&_minor_idr, m);
1038 spin_unlock(&_minor_lock);
1042 static int next_free_minor(int *minor)
1046 r = idr_pre_get(&_minor_idr, GFP_KERNEL);
1050 spin_lock(&_minor_lock);
1052 r = idr_get_new(&_minor_idr, MINOR_ALLOCED, &m);
1056 if (m >= (1 << MINORBITS)) {
1057 idr_remove(&_minor_idr, m);
1065 spin_unlock(&_minor_lock);
1069 static struct block_device_operations dm_blk_dops;
1072 * Allocate and initialise a blank device with a given minor.
1074 static struct mapped_device *alloc_dev(int minor)
1077 struct mapped_device *md = kzalloc(sizeof(*md), GFP_KERNEL);
1081 DMWARN("unable to allocate device, out of memory.");
1085 if (!try_module_get(THIS_MODULE))
1086 goto bad_module_get;
1088 /* get a minor number for the dev */
1089 if (minor == DM_ANY_MINOR)
1090 r = next_free_minor(&minor);
1092 r = specific_minor(minor);
1096 init_rwsem(&md->io_lock);
1097 mutex_init(&md->suspend_lock);
1098 spin_lock_init(&md->pushback_lock);
1099 rwlock_init(&md->map_lock);
1100 atomic_set(&md->holders, 1);
1101 atomic_set(&md->open_count, 0);
1102 atomic_set(&md->event_nr, 0);
1103 atomic_set(&md->uevent_seq, 0);
1104 INIT_LIST_HEAD(&md->uevent_list);
1105 spin_lock_init(&md->uevent_lock);
1107 md->queue = blk_alloc_queue(GFP_KERNEL);
1111 md->queue->queuedata = md;
1112 md->queue->backing_dev_info.congested_fn = dm_any_congested;
1113 md->queue->backing_dev_info.congested_data = md;
1114 blk_queue_make_request(md->queue, dm_request);
1115 blk_queue_bounce_limit(md->queue, BLK_BOUNCE_ANY);
1116 md->queue->unplug_fn = dm_unplug_all;
1117 blk_queue_merge_bvec(md->queue, dm_merge_bvec);
1119 md->io_pool = mempool_create_slab_pool(MIN_IOS, _io_cache);
1123 md->tio_pool = mempool_create_slab_pool(MIN_IOS, _tio_cache);
1127 md->bs = bioset_create(16, 0);
1131 md->disk = alloc_disk(1);
1135 atomic_set(&md->pending, 0);
1136 init_waitqueue_head(&md->wait);
1137 init_waitqueue_head(&md->eventq);
1139 md->disk->major = _major;
1140 md->disk->first_minor = minor;
1141 md->disk->fops = &dm_blk_dops;
1142 md->disk->queue = md->queue;
1143 md->disk->private_data = md;
1144 sprintf(md->disk->disk_name, "dm-%d", minor);
1146 format_dev_t(md->name, MKDEV(_major, minor));
1148 md->wq = create_singlethread_workqueue("kdmflush");
1152 /* Populate the mapping, nobody knows we exist yet */
1153 spin_lock(&_minor_lock);
1154 old_md = idr_replace(&_minor_idr, md, minor);
1155 spin_unlock(&_minor_lock);
1157 BUG_ON(old_md != MINOR_ALLOCED);
1164 bioset_free(md->bs);
1166 mempool_destroy(md->tio_pool);
1168 mempool_destroy(md->io_pool);
1170 blk_cleanup_queue(md->queue);
1174 module_put(THIS_MODULE);
1180 static void unlock_fs(struct mapped_device *md);
1182 static void free_dev(struct mapped_device *md)
1184 int minor = MINOR(disk_devt(md->disk));
1186 if (md->suspended_bdev) {
1188 bdput(md->suspended_bdev);
1190 destroy_workqueue(md->wq);
1191 mempool_destroy(md->tio_pool);
1192 mempool_destroy(md->io_pool);
1193 bioset_free(md->bs);
1194 del_gendisk(md->disk);
1197 spin_lock(&_minor_lock);
1198 md->disk->private_data = NULL;
1199 spin_unlock(&_minor_lock);
1202 blk_cleanup_queue(md->queue);
1203 module_put(THIS_MODULE);
1208 * Bind a table to the device.
1210 static void event_callback(void *context)
1212 unsigned long flags;
1214 struct mapped_device *md = (struct mapped_device *) context;
1216 spin_lock_irqsave(&md->uevent_lock, flags);
1217 list_splice_init(&md->uevent_list, &uevents);
1218 spin_unlock_irqrestore(&md->uevent_lock, flags);
1220 dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj);
1222 atomic_inc(&md->event_nr);
1223 wake_up(&md->eventq);
1226 static void __set_size(struct mapped_device *md, sector_t size)
1228 set_capacity(md->disk, size);
1230 mutex_lock(&md->suspended_bdev->bd_inode->i_mutex);
1231 i_size_write(md->suspended_bdev->bd_inode, (loff_t)size << SECTOR_SHIFT);
1232 mutex_unlock(&md->suspended_bdev->bd_inode->i_mutex);
1235 static int __bind(struct mapped_device *md, struct dm_table *t)
1237 struct request_queue *q = md->queue;
1240 size = dm_table_get_size(t);
1243 * Wipe any geometry if the size of the table changed.
1245 if (size != get_capacity(md->disk))
1246 memset(&md->geometry, 0, sizeof(md->geometry));
1248 if (md->suspended_bdev)
1249 __set_size(md, size);
1252 dm_table_destroy(t);
1256 dm_table_event_callback(t, event_callback, md);
1258 write_lock(&md->map_lock);
1260 dm_table_set_restrictions(t, q);
1261 write_unlock(&md->map_lock);
1266 static void __unbind(struct mapped_device *md)
1268 struct dm_table *map = md->map;
1273 dm_table_event_callback(map, NULL, NULL);
1274 write_lock(&md->map_lock);
1276 write_unlock(&md->map_lock);
1277 dm_table_destroy(map);
1281 * Constructor for a new device.
1283 int dm_create(int minor, struct mapped_device **result)
1285 struct mapped_device *md;
1287 md = alloc_dev(minor);
1297 static struct mapped_device *dm_find_md(dev_t dev)
1299 struct mapped_device *md;
1300 unsigned minor = MINOR(dev);
1302 if (MAJOR(dev) != _major || minor >= (1 << MINORBITS))
1305 spin_lock(&_minor_lock);
1307 md = idr_find(&_minor_idr, minor);
1308 if (md && (md == MINOR_ALLOCED ||
1309 (MINOR(disk_devt(dm_disk(md))) != minor) ||
1310 test_bit(DMF_FREEING, &md->flags))) {
1316 spin_unlock(&_minor_lock);
1321 struct mapped_device *dm_get_md(dev_t dev)
1323 struct mapped_device *md = dm_find_md(dev);
1331 void *dm_get_mdptr(struct mapped_device *md)
1333 return md->interface_ptr;
1336 void dm_set_mdptr(struct mapped_device *md, void *ptr)
1338 md->interface_ptr = ptr;
1341 void dm_get(struct mapped_device *md)
1343 atomic_inc(&md->holders);
1346 const char *dm_device_name(struct mapped_device *md)
1350 EXPORT_SYMBOL_GPL(dm_device_name);
1352 void dm_put(struct mapped_device *md)
1354 struct dm_table *map;
1356 BUG_ON(test_bit(DMF_FREEING, &md->flags));
1358 if (atomic_dec_and_lock(&md->holders, &_minor_lock)) {
1359 map = dm_get_table(md);
1360 idr_replace(&_minor_idr, MINOR_ALLOCED,
1361 MINOR(disk_devt(dm_disk(md))));
1362 set_bit(DMF_FREEING, &md->flags);
1363 spin_unlock(&_minor_lock);
1364 if (!dm_suspended(md)) {
1365 dm_table_presuspend_targets(map);
1366 dm_table_postsuspend_targets(map);
1374 EXPORT_SYMBOL_GPL(dm_put);
1376 static int dm_wait_for_completion(struct mapped_device *md)
1381 set_current_state(TASK_INTERRUPTIBLE);
1384 if (!atomic_read(&md->pending))
1387 if (signal_pending(current)) {
1394 set_current_state(TASK_RUNNING);
1400 * Process the deferred bios
1402 static void __flush_deferred_io(struct mapped_device *md)
1406 while ((c = bio_list_pop(&md->deferred))) {
1407 if (__split_bio(md, c))
1411 clear_bit(DMF_BLOCK_IO, &md->flags);
1414 static void __merge_pushback_list(struct mapped_device *md)
1416 unsigned long flags;
1418 spin_lock_irqsave(&md->pushback_lock, flags);
1419 clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
1420 bio_list_merge_head(&md->deferred, &md->pushback);
1421 bio_list_init(&md->pushback);
1422 spin_unlock_irqrestore(&md->pushback_lock, flags);
1425 static void dm_wq_work(struct work_struct *work)
1427 struct dm_wq_req *req = container_of(work, struct dm_wq_req, work);
1428 struct mapped_device *md = req->md;
1430 down_write(&md->io_lock);
1431 switch (req->type) {
1432 case DM_WQ_FLUSH_DEFERRED:
1433 __flush_deferred_io(md);
1436 DMERR("dm_wq_work: unrecognised work type %d", req->type);
1439 up_write(&md->io_lock);
1442 static void dm_wq_queue(struct mapped_device *md, int type, void *context,
1443 struct dm_wq_req *req)
1447 req->context = context;
1448 INIT_WORK(&req->work, dm_wq_work);
1449 queue_work(md->wq, &req->work);
1452 static void dm_queue_flush(struct mapped_device *md, int type, void *context)
1454 struct dm_wq_req req;
1456 dm_wq_queue(md, type, context, &req);
1457 flush_workqueue(md->wq);
1461 * Swap in a new table (destroying old one).
1463 int dm_swap_table(struct mapped_device *md, struct dm_table *table)
1467 mutex_lock(&md->suspend_lock);
1469 /* device must be suspended */
1470 if (!dm_suspended(md))
1473 /* without bdev, the device size cannot be changed */
1474 if (!md->suspended_bdev)
1475 if (get_capacity(md->disk) != dm_table_get_size(table))
1479 r = __bind(md, table);
1482 mutex_unlock(&md->suspend_lock);
1487 * Functions to lock and unlock any filesystem running on the
1490 static int lock_fs(struct mapped_device *md)
1494 WARN_ON(md->frozen_sb);
1496 md->frozen_sb = freeze_bdev(md->suspended_bdev);
1497 if (IS_ERR(md->frozen_sb)) {
1498 r = PTR_ERR(md->frozen_sb);
1499 md->frozen_sb = NULL;
1503 set_bit(DMF_FROZEN, &md->flags);
1505 /* don't bdput right now, we don't want the bdev
1506 * to go away while it is locked.
1511 static void unlock_fs(struct mapped_device *md)
1513 if (!test_bit(DMF_FROZEN, &md->flags))
1516 thaw_bdev(md->suspended_bdev, md->frozen_sb);
1517 md->frozen_sb = NULL;
1518 clear_bit(DMF_FROZEN, &md->flags);
1522 * We need to be able to change a mapping table under a mounted
1523 * filesystem. For example we might want to move some data in
1524 * the background. Before the table can be swapped with
1525 * dm_bind_table, dm_suspend must be called to flush any in
1526 * flight bios and ensure that any further io gets deferred.
1528 int dm_suspend(struct mapped_device *md, unsigned suspend_flags)
1530 struct dm_table *map = NULL;
1531 DECLARE_WAITQUEUE(wait, current);
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 add_wait_queue(&md->wait, &wait);
1582 up_write(&md->io_lock);
1586 dm_table_unplug_all(map);
1589 * Wait for the already-mapped ios to complete.
1591 r = dm_wait_for_completion(md);
1593 down_write(&md->io_lock);
1594 remove_wait_queue(&md->wait, &wait);
1597 __merge_pushback_list(md);
1598 up_write(&md->io_lock);
1600 /* were we interrupted ? */
1602 dm_queue_flush(md, DM_WQ_FLUSH_DEFERRED, NULL);
1605 goto out; /* pushback list is already flushed, so skip flush */
1608 dm_table_postsuspend_targets(map);
1610 set_bit(DMF_SUSPENDED, &md->flags);
1613 if (r && md->suspended_bdev) {
1614 bdput(md->suspended_bdev);
1615 md->suspended_bdev = NULL;
1621 mutex_unlock(&md->suspend_lock);
1625 int dm_resume(struct mapped_device *md)
1628 struct dm_table *map = NULL;
1630 mutex_lock(&md->suspend_lock);
1631 if (!dm_suspended(md))
1634 map = dm_get_table(md);
1635 if (!map || !dm_table_get_size(map))
1638 r = dm_table_resume_targets(map);
1642 dm_queue_flush(md, DM_WQ_FLUSH_DEFERRED, NULL);
1646 if (md->suspended_bdev) {
1647 bdput(md->suspended_bdev);
1648 md->suspended_bdev = NULL;
1651 clear_bit(DMF_SUSPENDED, &md->flags);
1653 dm_table_unplug_all(map);
1655 dm_kobject_uevent(md);
1661 mutex_unlock(&md->suspend_lock);
1666 /*-----------------------------------------------------------------
1667 * Event notification.
1668 *---------------------------------------------------------------*/
1669 void dm_kobject_uevent(struct mapped_device *md)
1671 kobject_uevent(&disk_to_dev(md->disk)->kobj, KOBJ_CHANGE);
1674 uint32_t dm_next_uevent_seq(struct mapped_device *md)
1676 return atomic_add_return(1, &md->uevent_seq);
1679 uint32_t dm_get_event_nr(struct mapped_device *md)
1681 return atomic_read(&md->event_nr);
1684 int dm_wait_event(struct mapped_device *md, int event_nr)
1686 return wait_event_interruptible(md->eventq,
1687 (event_nr != atomic_read(&md->event_nr)));
1690 void dm_uevent_add(struct mapped_device *md, struct list_head *elist)
1692 unsigned long flags;
1694 spin_lock_irqsave(&md->uevent_lock, flags);
1695 list_add(elist, &md->uevent_list);
1696 spin_unlock_irqrestore(&md->uevent_lock, flags);
1700 * The gendisk is only valid as long as you have a reference
1703 struct gendisk *dm_disk(struct mapped_device *md)
1708 struct kobject *dm_kobject(struct mapped_device *md)
1714 * struct mapped_device should not be exported outside of dm.c
1715 * so use this check to verify that kobj is part of md structure
1717 struct mapped_device *dm_get_from_kobject(struct kobject *kobj)
1719 struct mapped_device *md;
1721 md = container_of(kobj, struct mapped_device, kobj);
1722 if (&md->kobj != kobj)
1729 int dm_suspended(struct mapped_device *md)
1731 return test_bit(DMF_SUSPENDED, &md->flags);
1734 int dm_noflush_suspending(struct dm_target *ti)
1736 struct mapped_device *md = dm_table_get_md(ti->table);
1737 int r = __noflush_suspending(md);
1743 EXPORT_SYMBOL_GPL(dm_noflush_suspending);
1745 static struct block_device_operations dm_blk_dops = {
1746 .open = dm_blk_open,
1747 .release = dm_blk_close,
1748 .ioctl = dm_blk_ioctl,
1749 .getgeo = dm_blk_getgeo,
1750 .owner = THIS_MODULE
1753 EXPORT_SYMBOL(dm_get_mapinfo);
1758 module_init(dm_init);
1759 module_exit(dm_exit);
1761 module_param(major, uint, 0);
1762 MODULE_PARM_DESC(major, "The major number of the device mapper");
1763 MODULE_DESCRIPTION(DM_NAME " driver");
1764 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
1765 MODULE_LICENSE("GPL");