2 * Copyright (C) 2001, 2002 Sistina Software (UK) Limited.
3 * Copyright (C) 2004-2006 Red Hat, Inc. All rights reserved.
5 * This file is released under the GPL.
9 #include "dm-bio-list.h"
11 #include <linux/init.h>
12 #include <linux/module.h>
13 #include <linux/mutex.h>
14 #include <linux/moduleparam.h>
15 #include <linux/blkpg.h>
16 #include <linux/bio.h>
17 #include <linux/buffer_head.h>
18 #include <linux/mempool.h>
19 #include <linux/slab.h>
20 #include <linux/idr.h>
21 #include <linux/hdreg.h>
22 #include <linux/blktrace_api.h>
23 #include <linux/smp_lock.h>
25 #define DM_MSG_PREFIX "core"
27 static const char *_name = DM_NAME;
29 static unsigned int major = 0;
30 static unsigned int _major = 0;
32 static DEFINE_SPINLOCK(_minor_lock);
34 * One of these is allocated per bio.
37 struct mapped_device *md;
41 unsigned long start_time;
45 * One of these is allocated per target within a bio. Hopefully
46 * this will be simplified out one day.
54 union map_info *dm_get_mapinfo(struct bio *bio)
56 if (bio && bio->bi_private)
57 return &((struct target_io *)bio->bi_private)->info;
61 #define MINOR_ALLOCED ((void *)-1)
64 * Bits for the md->flags field.
66 #define DMF_BLOCK_IO 0
67 #define DMF_SUSPENDED 1
70 #define DMF_DELETING 4
72 struct mapped_device {
73 struct rw_semaphore io_lock;
74 struct semaphore suspend_lock;
81 request_queue_t *queue;
88 * A list of ios that arrived while we were suspended.
91 wait_queue_head_t wait;
92 struct bio_list deferred;
95 * The current mapping.
100 * io objects are allocated from here.
111 wait_queue_head_t eventq;
114 * freeze/thaw support require holding onto a super block
116 struct super_block *frozen_sb;
117 struct block_device *suspended_bdev;
119 /* forced geometry settings */
120 struct hd_geometry geometry;
124 static kmem_cache_t *_io_cache;
125 static kmem_cache_t *_tio_cache;
127 static int __init local_init(void)
131 /* allocate a slab for the dm_ios */
132 _io_cache = kmem_cache_create("dm_io",
133 sizeof(struct dm_io), 0, 0, NULL, NULL);
137 /* allocate a slab for the target ios */
138 _tio_cache = kmem_cache_create("dm_tio", sizeof(struct target_io),
141 kmem_cache_destroy(_io_cache);
146 r = register_blkdev(_major, _name);
148 kmem_cache_destroy(_tio_cache);
149 kmem_cache_destroy(_io_cache);
159 static void local_exit(void)
161 kmem_cache_destroy(_tio_cache);
162 kmem_cache_destroy(_io_cache);
164 if (unregister_blkdev(_major, _name) < 0)
165 DMERR("unregister_blkdev failed");
169 DMINFO("cleaned up");
172 int (*_inits[])(void) __initdata = {
180 void (*_exits[])(void) = {
188 static int __init dm_init(void)
190 const int count = ARRAY_SIZE(_inits);
194 for (i = 0; i < count; i++) {
209 static void __exit dm_exit(void)
211 int i = ARRAY_SIZE(_exits);
218 * Block device functions
220 static int dm_blk_open(struct inode *inode, struct file *file)
222 struct mapped_device *md;
224 spin_lock(&_minor_lock);
226 md = inode->i_bdev->bd_disk->private_data;
230 if (test_bit(DMF_FREEING, &md->flags) ||
231 test_bit(DMF_DELETING, &md->flags)) {
237 atomic_inc(&md->open_count);
240 spin_unlock(&_minor_lock);
242 return md ? 0 : -ENXIO;
245 static int dm_blk_close(struct inode *inode, struct file *file)
247 struct mapped_device *md;
249 md = inode->i_bdev->bd_disk->private_data;
250 atomic_dec(&md->open_count);
255 int dm_open_count(struct mapped_device *md)
257 return atomic_read(&md->open_count);
261 * Guarantees nothing is using the device before it's deleted.
263 int dm_lock_for_deletion(struct mapped_device *md)
267 spin_lock(&_minor_lock);
269 if (dm_open_count(md))
272 set_bit(DMF_DELETING, &md->flags);
274 spin_unlock(&_minor_lock);
279 static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo)
281 struct mapped_device *md = bdev->bd_disk->private_data;
283 return dm_get_geometry(md, geo);
286 static int dm_blk_ioctl(struct inode *inode, struct file *file,
287 unsigned int cmd, unsigned long arg)
289 struct mapped_device *md;
290 struct dm_table *map;
291 struct dm_target *tgt;
294 /* We don't really need this lock, but we do need 'inode'. */
297 md = inode->i_bdev->bd_disk->private_data;
299 map = dm_get_table(md);
301 if (!map || !dm_table_get_size(map))
304 /* We only support devices that have a single target */
305 if (dm_table_get_num_targets(map) != 1)
308 tgt = dm_table_get_target(map, 0);
310 if (dm_suspended(md)) {
315 if (tgt->type->ioctl)
316 r = tgt->type->ioctl(tgt, inode, file, cmd, arg);
325 static inline struct dm_io *alloc_io(struct mapped_device *md)
327 return mempool_alloc(md->io_pool, GFP_NOIO);
330 static inline void free_io(struct mapped_device *md, struct dm_io *io)
332 mempool_free(io, md->io_pool);
335 static inline struct target_io *alloc_tio(struct mapped_device *md)
337 return mempool_alloc(md->tio_pool, GFP_NOIO);
340 static inline void free_tio(struct mapped_device *md, struct target_io *tio)
342 mempool_free(tio, md->tio_pool);
345 static void start_io_acct(struct dm_io *io)
347 struct mapped_device *md = io->md;
349 io->start_time = jiffies;
352 disk_round_stats(dm_disk(md));
354 dm_disk(md)->in_flight = atomic_inc_return(&md->pending);
357 static int end_io_acct(struct dm_io *io)
359 struct mapped_device *md = io->md;
360 struct bio *bio = io->bio;
361 unsigned long duration = jiffies - io->start_time;
363 int rw = bio_data_dir(bio);
366 disk_round_stats(dm_disk(md));
368 dm_disk(md)->in_flight = pending = atomic_dec_return(&md->pending);
370 disk_stat_add(dm_disk(md), ticks[rw], duration);
376 * Add the bio to the list of deferred io.
378 static int queue_io(struct mapped_device *md, struct bio *bio)
380 down_write(&md->io_lock);
382 if (!test_bit(DMF_BLOCK_IO, &md->flags)) {
383 up_write(&md->io_lock);
387 bio_list_add(&md->deferred, bio);
389 up_write(&md->io_lock);
390 return 0; /* deferred successfully */
394 * Everyone (including functions in this file), should use this
395 * function to access the md->map field, and make sure they call
396 * dm_table_put() when finished.
398 struct dm_table *dm_get_table(struct mapped_device *md)
402 read_lock(&md->map_lock);
406 read_unlock(&md->map_lock);
412 * Get the geometry associated with a dm device
414 int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo)
422 * Set the geometry of a device.
424 int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo)
426 sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors;
428 if (geo->start > sz) {
429 DMWARN("Start sector is beyond the geometry limits.");
438 /*-----------------------------------------------------------------
440 * A more elegant soln is in the works that uses the queue
441 * merge fn, unfortunately there are a couple of changes to
442 * the block layer that I want to make for this. So in the
443 * interests of getting something for people to use I give
444 * you this clearly demarcated crap.
445 *---------------------------------------------------------------*/
448 * Decrements the number of outstanding ios that a bio has been
449 * cloned into, completing the original io if necc.
451 static void dec_pending(struct dm_io *io, int error)
456 if (atomic_dec_and_test(&io->io_count)) {
458 /* nudge anyone waiting on suspend queue */
459 wake_up(&io->md->wait);
461 blk_add_trace_bio(io->md->queue, io->bio, BLK_TA_COMPLETE);
463 bio_endio(io->bio, io->bio->bi_size, io->error);
468 static int clone_endio(struct bio *bio, unsigned int done, int error)
471 struct target_io *tio = bio->bi_private;
472 struct mapped_device *md = tio->io->md;
473 dm_endio_fn endio = tio->ti->type->end_io;
478 if (!bio_flagged(bio, BIO_UPTODATE) && !error)
482 r = endio(tio->ti, bio, error, &tio->info);
487 /* the target wants another shot at the io */
491 dec_pending(tio->io, error);
494 * Store md for cleanup instead of tio which is about to get freed.
496 bio->bi_private = md->bs;
503 static sector_t max_io_len(struct mapped_device *md,
504 sector_t sector, struct dm_target *ti)
506 sector_t offset = sector - ti->begin;
507 sector_t len = ti->len - offset;
510 * Does the target need to split even further ?
514 boundary = ((offset + ti->split_io) & ~(ti->split_io - 1))
523 static void __map_bio(struct dm_target *ti, struct bio *clone,
524 struct target_io *tio)
528 struct mapped_device *md;
533 BUG_ON(!clone->bi_size);
535 clone->bi_end_io = clone_endio;
536 clone->bi_private = tio;
539 * Map the clone. If r == 0 we don't need to do
540 * anything, the target has assumed ownership of
543 atomic_inc(&tio->io->io_count);
544 sector = clone->bi_sector;
545 r = ti->type->map(ti, clone, &tio->info);
547 /* the bio has been remapped so dispatch it */
549 blk_add_trace_remap(bdev_get_queue(clone->bi_bdev), clone,
550 tio->io->bio->bi_bdev->bd_dev, sector,
553 generic_make_request(clone);
557 /* error the io and bail out */
559 dec_pending(tio->io, r);
561 * Store bio_set for cleanup.
563 clone->bi_private = md->bs;
570 struct mapped_device *md;
571 struct dm_table *map;
575 sector_t sector_count;
579 static void dm_bio_destructor(struct bio *bio)
581 struct bio_set *bs = bio->bi_private;
587 * Creates a little bio that is just does part of a bvec.
589 static struct bio *split_bvec(struct bio *bio, sector_t sector,
590 unsigned short idx, unsigned int offset,
591 unsigned int len, struct bio_set *bs)
594 struct bio_vec *bv = bio->bi_io_vec + idx;
596 clone = bio_alloc_bioset(GFP_NOIO, 1, bs);
597 clone->bi_destructor = dm_bio_destructor;
598 *clone->bi_io_vec = *bv;
600 clone->bi_sector = sector;
601 clone->bi_bdev = bio->bi_bdev;
602 clone->bi_rw = bio->bi_rw;
604 clone->bi_size = to_bytes(len);
605 clone->bi_io_vec->bv_offset = offset;
606 clone->bi_io_vec->bv_len = clone->bi_size;
612 * Creates a bio that consists of range of complete bvecs.
614 static struct bio *clone_bio(struct bio *bio, sector_t sector,
615 unsigned short idx, unsigned short bv_count,
616 unsigned int len, struct bio_set *bs)
620 clone = bio_alloc_bioset(GFP_NOIO, bio->bi_max_vecs, bs);
621 __bio_clone(clone, bio);
622 clone->bi_destructor = dm_bio_destructor;
623 clone->bi_sector = sector;
625 clone->bi_vcnt = idx + bv_count;
626 clone->bi_size = to_bytes(len);
627 clone->bi_flags &= ~(1 << BIO_SEG_VALID);
632 static void __clone_and_map(struct clone_info *ci)
634 struct bio *clone, *bio = ci->bio;
635 struct dm_target *ti = dm_table_find_target(ci->map, ci->sector);
636 sector_t len = 0, max = max_io_len(ci->md, ci->sector, ti);
637 struct target_io *tio;
640 * Allocate a target io object.
642 tio = alloc_tio(ci->md);
645 memset(&tio->info, 0, sizeof(tio->info));
647 if (ci->sector_count <= max) {
649 * Optimise for the simple case where we can do all of
650 * the remaining io with a single clone.
652 clone = clone_bio(bio, ci->sector, ci->idx,
653 bio->bi_vcnt - ci->idx, ci->sector_count,
655 __map_bio(ti, clone, tio);
656 ci->sector_count = 0;
658 } else if (to_sector(bio->bi_io_vec[ci->idx].bv_len) <= max) {
660 * There are some bvecs that don't span targets.
661 * Do as many of these as possible.
664 sector_t remaining = max;
667 for (i = ci->idx; remaining && (i < bio->bi_vcnt); i++) {
668 bv_len = to_sector(bio->bi_io_vec[i].bv_len);
670 if (bv_len > remaining)
677 clone = clone_bio(bio, ci->sector, ci->idx, i - ci->idx, len,
679 __map_bio(ti, clone, tio);
682 ci->sector_count -= len;
687 * Handle a bvec that must be split between two or more targets.
689 struct bio_vec *bv = bio->bi_io_vec + ci->idx;
690 sector_t remaining = to_sector(bv->bv_len);
691 unsigned int offset = 0;
695 ti = dm_table_find_target(ci->map, ci->sector);
696 max = max_io_len(ci->md, ci->sector, ti);
698 tio = alloc_tio(ci->md);
701 memset(&tio->info, 0, sizeof(tio->info));
704 len = min(remaining, max);
706 clone = split_bvec(bio, ci->sector, ci->idx,
707 bv->bv_offset + offset, len,
710 __map_bio(ti, clone, tio);
713 ci->sector_count -= len;
714 offset += to_bytes(len);
715 } while (remaining -= len);
722 * Split the bio into several clones.
724 static void __split_bio(struct mapped_device *md, struct bio *bio)
726 struct clone_info ci;
728 ci.map = dm_get_table(md);
730 bio_io_error(bio, bio->bi_size);
736 ci.io = alloc_io(md);
738 atomic_set(&ci.io->io_count, 1);
741 ci.sector = bio->bi_sector;
742 ci.sector_count = bio_sectors(bio);
743 ci.idx = bio->bi_idx;
745 start_io_acct(ci.io);
746 while (ci.sector_count)
747 __clone_and_map(&ci);
749 /* drop the extra reference count */
750 dec_pending(ci.io, 0);
751 dm_table_put(ci.map);
753 /*-----------------------------------------------------------------
755 *---------------------------------------------------------------*/
758 * The request function that just remaps the bio built up by
761 static int dm_request(request_queue_t *q, struct bio *bio)
764 int rw = bio_data_dir(bio);
765 struct mapped_device *md = q->queuedata;
767 down_read(&md->io_lock);
769 disk_stat_inc(dm_disk(md), ios[rw]);
770 disk_stat_add(dm_disk(md), sectors[rw], bio_sectors(bio));
773 * If we're suspended we have to queue
776 while (test_bit(DMF_BLOCK_IO, &md->flags)) {
777 up_read(&md->io_lock);
779 if (bio_rw(bio) == READA) {
780 bio_io_error(bio, bio->bi_size);
784 r = queue_io(md, bio);
786 bio_io_error(bio, bio->bi_size);
790 return 0; /* deferred successfully */
793 * We're in a while loop, because someone could suspend
794 * before we get to the following read lock.
796 down_read(&md->io_lock);
799 __split_bio(md, bio);
800 up_read(&md->io_lock);
804 static int dm_flush_all(request_queue_t *q, struct gendisk *disk,
805 sector_t *error_sector)
807 struct mapped_device *md = q->queuedata;
808 struct dm_table *map = dm_get_table(md);
812 ret = dm_table_flush_all(map);
819 static void dm_unplug_all(request_queue_t *q)
821 struct mapped_device *md = q->queuedata;
822 struct dm_table *map = dm_get_table(md);
825 dm_table_unplug_all(map);
830 static int dm_any_congested(void *congested_data, int bdi_bits)
833 struct mapped_device *md = (struct mapped_device *) congested_data;
834 struct dm_table *map = dm_get_table(md);
836 if (!map || test_bit(DMF_BLOCK_IO, &md->flags))
839 r = dm_table_any_congested(map, bdi_bits);
845 /*-----------------------------------------------------------------
846 * An IDR is used to keep track of allocated minor numbers.
847 *---------------------------------------------------------------*/
848 static DEFINE_IDR(_minor_idr);
850 static void free_minor(int minor)
852 spin_lock(&_minor_lock);
853 idr_remove(&_minor_idr, minor);
854 spin_unlock(&_minor_lock);
858 * See if the device with a specific minor # is free.
860 static int specific_minor(struct mapped_device *md, int minor)
864 if (minor >= (1 << MINORBITS))
867 r = idr_pre_get(&_minor_idr, GFP_KERNEL);
871 spin_lock(&_minor_lock);
873 if (idr_find(&_minor_idr, minor)) {
878 r = idr_get_new_above(&_minor_idr, MINOR_ALLOCED, minor, &m);
883 idr_remove(&_minor_idr, m);
889 spin_unlock(&_minor_lock);
893 static int next_free_minor(struct mapped_device *md, int *minor)
897 r = idr_pre_get(&_minor_idr, GFP_KERNEL);
901 spin_lock(&_minor_lock);
903 r = idr_get_new(&_minor_idr, MINOR_ALLOCED, &m);
908 if (m >= (1 << MINORBITS)) {
909 idr_remove(&_minor_idr, m);
917 spin_unlock(&_minor_lock);
921 static struct block_device_operations dm_blk_dops;
924 * Allocate and initialise a blank device with a given minor.
926 static struct mapped_device *alloc_dev(int minor)
929 struct mapped_device *md = kmalloc(sizeof(*md), GFP_KERNEL);
933 DMWARN("unable to allocate device, out of memory.");
937 if (!try_module_get(THIS_MODULE))
940 /* get a minor number for the dev */
941 if (minor == DM_ANY_MINOR)
942 r = next_free_minor(md, &minor);
944 r = specific_minor(md, minor);
948 memset(md, 0, sizeof(*md));
949 init_rwsem(&md->io_lock);
950 init_MUTEX(&md->suspend_lock);
951 rwlock_init(&md->map_lock);
952 atomic_set(&md->holders, 1);
953 atomic_set(&md->open_count, 0);
954 atomic_set(&md->event_nr, 0);
956 md->queue = blk_alloc_queue(GFP_KERNEL);
958 goto bad1_free_minor;
960 md->queue->queuedata = md;
961 md->queue->backing_dev_info.congested_fn = dm_any_congested;
962 md->queue->backing_dev_info.congested_data = md;
963 blk_queue_make_request(md->queue, dm_request);
964 blk_queue_bounce_limit(md->queue, BLK_BOUNCE_ANY);
965 md->queue->unplug_fn = dm_unplug_all;
966 md->queue->issue_flush_fn = dm_flush_all;
968 md->io_pool = mempool_create_slab_pool(MIN_IOS, _io_cache);
972 md->tio_pool = mempool_create_slab_pool(MIN_IOS, _tio_cache);
976 md->bs = bioset_create(16, 16, 4);
980 md->disk = alloc_disk(1);
984 atomic_set(&md->pending, 0);
985 init_waitqueue_head(&md->wait);
986 init_waitqueue_head(&md->eventq);
988 md->disk->major = _major;
989 md->disk->first_minor = minor;
990 md->disk->fops = &dm_blk_dops;
991 md->disk->queue = md->queue;
992 md->disk->private_data = md;
993 sprintf(md->disk->disk_name, "dm-%d", minor);
995 format_dev_t(md->name, MKDEV(_major, minor));
997 /* Populate the mapping, nobody knows we exist yet */
998 spin_lock(&_minor_lock);
999 old_md = idr_replace(&_minor_idr, md, minor);
1000 spin_unlock(&_minor_lock);
1002 BUG_ON(old_md != MINOR_ALLOCED);
1007 bioset_free(md->bs);
1009 mempool_destroy(md->tio_pool);
1011 mempool_destroy(md->io_pool);
1013 blk_cleanup_queue(md->queue);
1017 module_put(THIS_MODULE);
1023 static void free_dev(struct mapped_device *md)
1025 int minor = md->disk->first_minor;
1027 if (md->suspended_bdev) {
1028 thaw_bdev(md->suspended_bdev, NULL);
1029 bdput(md->suspended_bdev);
1031 mempool_destroy(md->tio_pool);
1032 mempool_destroy(md->io_pool);
1033 bioset_free(md->bs);
1034 del_gendisk(md->disk);
1037 spin_lock(&_minor_lock);
1038 md->disk->private_data = NULL;
1039 spin_unlock(&_minor_lock);
1042 blk_cleanup_queue(md->queue);
1043 module_put(THIS_MODULE);
1048 * Bind a table to the device.
1050 static void event_callback(void *context)
1052 struct mapped_device *md = (struct mapped_device *) context;
1054 atomic_inc(&md->event_nr);
1055 wake_up(&md->eventq);
1058 static void __set_size(struct mapped_device *md, sector_t size)
1060 set_capacity(md->disk, size);
1062 mutex_lock(&md->suspended_bdev->bd_inode->i_mutex);
1063 i_size_write(md->suspended_bdev->bd_inode, (loff_t)size << SECTOR_SHIFT);
1064 mutex_unlock(&md->suspended_bdev->bd_inode->i_mutex);
1067 static int __bind(struct mapped_device *md, struct dm_table *t)
1069 request_queue_t *q = md->queue;
1072 size = dm_table_get_size(t);
1075 * Wipe any geometry if the size of the table changed.
1077 if (size != get_capacity(md->disk))
1078 memset(&md->geometry, 0, sizeof(md->geometry));
1080 __set_size(md, size);
1085 dm_table_event_callback(t, event_callback, md);
1087 write_lock(&md->map_lock);
1089 dm_table_set_restrictions(t, q);
1090 write_unlock(&md->map_lock);
1095 static void __unbind(struct mapped_device *md)
1097 struct dm_table *map = md->map;
1102 dm_table_event_callback(map, NULL, NULL);
1103 write_lock(&md->map_lock);
1105 write_unlock(&md->map_lock);
1110 * Constructor for a new device.
1112 int dm_create(int minor, struct mapped_device **result)
1114 struct mapped_device *md;
1116 md = alloc_dev(minor);
1124 static struct mapped_device *dm_find_md(dev_t dev)
1126 struct mapped_device *md;
1127 unsigned minor = MINOR(dev);
1129 if (MAJOR(dev) != _major || minor >= (1 << MINORBITS))
1132 spin_lock(&_minor_lock);
1134 md = idr_find(&_minor_idr, minor);
1135 if (md && (md == MINOR_ALLOCED ||
1136 (dm_disk(md)->first_minor != minor) ||
1137 test_bit(DMF_FREEING, &md->flags))) {
1143 spin_unlock(&_minor_lock);
1148 struct mapped_device *dm_get_md(dev_t dev)
1150 struct mapped_device *md = dm_find_md(dev);
1158 void *dm_get_mdptr(struct mapped_device *md)
1160 return md->interface_ptr;
1163 void dm_set_mdptr(struct mapped_device *md, void *ptr)
1165 md->interface_ptr = ptr;
1168 void dm_get(struct mapped_device *md)
1170 atomic_inc(&md->holders);
1173 const char *dm_device_name(struct mapped_device *md)
1177 EXPORT_SYMBOL_GPL(dm_device_name);
1179 void dm_put(struct mapped_device *md)
1181 struct dm_table *map;
1183 BUG_ON(test_bit(DMF_FREEING, &md->flags));
1185 if (atomic_dec_and_lock(&md->holders, &_minor_lock)) {
1186 map = dm_get_table(md);
1187 idr_replace(&_minor_idr, MINOR_ALLOCED, dm_disk(md)->first_minor);
1188 set_bit(DMF_FREEING, &md->flags);
1189 spin_unlock(&_minor_lock);
1190 if (!dm_suspended(md)) {
1191 dm_table_presuspend_targets(map);
1192 dm_table_postsuspend_targets(map);
1201 * Process the deferred bios
1203 static void __flush_deferred_io(struct mapped_device *md, struct bio *c)
1216 * Swap in a new table (destroying old one).
1218 int dm_swap_table(struct mapped_device *md, struct dm_table *table)
1222 down(&md->suspend_lock);
1224 /* device must be suspended */
1225 if (!dm_suspended(md))
1229 r = __bind(md, table);
1232 up(&md->suspend_lock);
1237 * Functions to lock and unlock any filesystem running on the
1240 static int lock_fs(struct mapped_device *md)
1244 WARN_ON(md->frozen_sb);
1246 md->frozen_sb = freeze_bdev(md->suspended_bdev);
1247 if (IS_ERR(md->frozen_sb)) {
1248 r = PTR_ERR(md->frozen_sb);
1249 md->frozen_sb = NULL;
1253 set_bit(DMF_FROZEN, &md->flags);
1255 /* don't bdput right now, we don't want the bdev
1256 * to go away while it is locked.
1261 static void unlock_fs(struct mapped_device *md)
1263 if (!test_bit(DMF_FROZEN, &md->flags))
1266 thaw_bdev(md->suspended_bdev, md->frozen_sb);
1267 md->frozen_sb = NULL;
1268 clear_bit(DMF_FROZEN, &md->flags);
1272 * We need to be able to change a mapping table under a mounted
1273 * filesystem. For example we might want to move some data in
1274 * the background. Before the table can be swapped with
1275 * dm_bind_table, dm_suspend must be called to flush any in
1276 * flight bios and ensure that any further io gets deferred.
1278 int dm_suspend(struct mapped_device *md, int do_lockfs)
1280 struct dm_table *map = NULL;
1281 DECLARE_WAITQUEUE(wait, current);
1285 down(&md->suspend_lock);
1287 if (dm_suspended(md))
1290 map = dm_get_table(md);
1292 /* This does not get reverted if there's an error later. */
1293 dm_table_presuspend_targets(map);
1295 md->suspended_bdev = bdget_disk(md->disk, 0);
1296 if (!md->suspended_bdev) {
1297 DMWARN("bdget failed in dm_suspend");
1302 /* Flush I/O to the device. */
1310 * First we set the BLOCK_IO flag so no more ios will be mapped.
1312 down_write(&md->io_lock);
1313 set_bit(DMF_BLOCK_IO, &md->flags);
1315 add_wait_queue(&md->wait, &wait);
1316 up_write(&md->io_lock);
1320 dm_table_unplug_all(map);
1323 * Then we wait for the already mapped ios to
1327 set_current_state(TASK_INTERRUPTIBLE);
1329 if (!atomic_read(&md->pending) || signal_pending(current))
1334 set_current_state(TASK_RUNNING);
1336 down_write(&md->io_lock);
1337 remove_wait_queue(&md->wait, &wait);
1339 /* were we interrupted ? */
1341 if (atomic_read(&md->pending)) {
1342 clear_bit(DMF_BLOCK_IO, &md->flags);
1343 def = bio_list_get(&md->deferred);
1344 __flush_deferred_io(md, def);
1345 up_write(&md->io_lock);
1349 up_write(&md->io_lock);
1351 dm_table_postsuspend_targets(map);
1353 set_bit(DMF_SUSPENDED, &md->flags);
1358 if (r && md->suspended_bdev) {
1359 bdput(md->suspended_bdev);
1360 md->suspended_bdev = NULL;
1366 up(&md->suspend_lock);
1370 int dm_resume(struct mapped_device *md)
1374 struct dm_table *map = NULL;
1376 down(&md->suspend_lock);
1377 if (!dm_suspended(md))
1380 map = dm_get_table(md);
1381 if (!map || !dm_table_get_size(map))
1384 r = dm_table_resume_targets(map);
1388 down_write(&md->io_lock);
1389 clear_bit(DMF_BLOCK_IO, &md->flags);
1391 def = bio_list_get(&md->deferred);
1392 __flush_deferred_io(md, def);
1393 up_write(&md->io_lock);
1397 bdput(md->suspended_bdev);
1398 md->suspended_bdev = NULL;
1400 clear_bit(DMF_SUSPENDED, &md->flags);
1402 dm_table_unplug_all(map);
1404 kobject_uevent(&md->disk->kobj, KOBJ_CHANGE);
1410 up(&md->suspend_lock);
1415 /*-----------------------------------------------------------------
1416 * Event notification.
1417 *---------------------------------------------------------------*/
1418 uint32_t dm_get_event_nr(struct mapped_device *md)
1420 return atomic_read(&md->event_nr);
1423 int dm_wait_event(struct mapped_device *md, int event_nr)
1425 return wait_event_interruptible(md->eventq,
1426 (event_nr != atomic_read(&md->event_nr)));
1430 * The gendisk is only valid as long as you have a reference
1433 struct gendisk *dm_disk(struct mapped_device *md)
1438 int dm_suspended(struct mapped_device *md)
1440 return test_bit(DMF_SUSPENDED, &md->flags);
1443 static struct block_device_operations dm_blk_dops = {
1444 .open = dm_blk_open,
1445 .release = dm_blk_close,
1446 .ioctl = dm_blk_ioctl,
1447 .getgeo = dm_blk_getgeo,
1448 .owner = THIS_MODULE
1451 EXPORT_SYMBOL(dm_get_mapinfo);
1456 module_init(dm_init);
1457 module_exit(dm_exit);
1459 module_param(major, uint, 0);
1460 MODULE_PARM_DESC(major, "The major number of the device mapper");
1461 MODULE_DESCRIPTION(DM_NAME " driver");
1462 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
1463 MODULE_LICENSE("GPL");