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.
109 wait_queue_head_t eventq;
112 * freeze/thaw support require holding onto a super block
114 struct super_block *frozen_sb;
115 struct block_device *suspended_bdev;
117 /* forced geometry settings */
118 struct hd_geometry geometry;
122 static kmem_cache_t *_io_cache;
123 static kmem_cache_t *_tio_cache;
125 static struct bio_set *dm_set;
127 static int __init local_init(void)
131 dm_set = bioset_create(16, 16, 4);
135 /* allocate a slab for the dm_ios */
136 _io_cache = kmem_cache_create("dm_io",
137 sizeof(struct dm_io), 0, 0, NULL, NULL);
141 /* allocate a slab for the target ios */
142 _tio_cache = kmem_cache_create("dm_tio", sizeof(struct target_io),
145 kmem_cache_destroy(_io_cache);
150 r = register_blkdev(_major, _name);
152 kmem_cache_destroy(_tio_cache);
153 kmem_cache_destroy(_io_cache);
163 static void local_exit(void)
165 kmem_cache_destroy(_tio_cache);
166 kmem_cache_destroy(_io_cache);
170 if (unregister_blkdev(_major, _name) < 0)
171 DMERR("unregister_blkdev failed");
175 DMINFO("cleaned up");
178 int (*_inits[])(void) __initdata = {
186 void (*_exits[])(void) = {
194 static int __init dm_init(void)
196 const int count = ARRAY_SIZE(_inits);
200 for (i = 0; i < count; i++) {
215 static void __exit dm_exit(void)
217 int i = ARRAY_SIZE(_exits);
224 * Block device functions
226 static int dm_blk_open(struct inode *inode, struct file *file)
228 struct mapped_device *md;
230 spin_lock(&_minor_lock);
232 md = inode->i_bdev->bd_disk->private_data;
236 if (test_bit(DMF_FREEING, &md->flags) ||
237 test_bit(DMF_DELETING, &md->flags)) {
243 atomic_inc(&md->open_count);
246 spin_unlock(&_minor_lock);
248 return md ? 0 : -ENXIO;
251 static int dm_blk_close(struct inode *inode, struct file *file)
253 struct mapped_device *md;
255 md = inode->i_bdev->bd_disk->private_data;
256 atomic_dec(&md->open_count);
261 int dm_open_count(struct mapped_device *md)
263 return atomic_read(&md->open_count);
267 * Guarantees nothing is using the device before it's deleted.
269 int dm_lock_for_deletion(struct mapped_device *md)
273 spin_lock(&_minor_lock);
275 if (dm_open_count(md))
278 set_bit(DMF_DELETING, &md->flags);
280 spin_unlock(&_minor_lock);
285 static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo)
287 struct mapped_device *md = bdev->bd_disk->private_data;
289 return dm_get_geometry(md, geo);
292 static int dm_blk_ioctl(struct inode *inode, struct file *file,
293 unsigned int cmd, unsigned long arg)
295 struct mapped_device *md;
296 struct dm_table *map;
297 struct dm_target *tgt;
300 /* We don't really need this lock, but we do need 'inode'. */
303 md = inode->i_bdev->bd_disk->private_data;
305 map = dm_get_table(md);
307 if (!map || !dm_table_get_size(map))
310 /* We only support devices that have a single target */
311 if (dm_table_get_num_targets(map) != 1)
314 tgt = dm_table_get_target(map, 0);
316 if (dm_suspended(md)) {
321 if (tgt->type->ioctl)
322 r = tgt->type->ioctl(tgt, inode, file, cmd, arg);
331 static inline struct dm_io *alloc_io(struct mapped_device *md)
333 return mempool_alloc(md->io_pool, GFP_NOIO);
336 static inline void free_io(struct mapped_device *md, struct dm_io *io)
338 mempool_free(io, md->io_pool);
341 static inline struct target_io *alloc_tio(struct mapped_device *md)
343 return mempool_alloc(md->tio_pool, GFP_NOIO);
346 static inline void free_tio(struct mapped_device *md, struct target_io *tio)
348 mempool_free(tio, md->tio_pool);
351 static void start_io_acct(struct dm_io *io)
353 struct mapped_device *md = io->md;
355 io->start_time = jiffies;
358 disk_round_stats(dm_disk(md));
360 dm_disk(md)->in_flight = atomic_inc_return(&md->pending);
363 static int end_io_acct(struct dm_io *io)
365 struct mapped_device *md = io->md;
366 struct bio *bio = io->bio;
367 unsigned long duration = jiffies - io->start_time;
369 int rw = bio_data_dir(bio);
372 disk_round_stats(dm_disk(md));
374 dm_disk(md)->in_flight = pending = atomic_dec_return(&md->pending);
376 disk_stat_add(dm_disk(md), ticks[rw], duration);
382 * Add the bio to the list of deferred io.
384 static int queue_io(struct mapped_device *md, struct bio *bio)
386 down_write(&md->io_lock);
388 if (!test_bit(DMF_BLOCK_IO, &md->flags)) {
389 up_write(&md->io_lock);
393 bio_list_add(&md->deferred, bio);
395 up_write(&md->io_lock);
396 return 0; /* deferred successfully */
400 * Everyone (including functions in this file), should use this
401 * function to access the md->map field, and make sure they call
402 * dm_table_put() when finished.
404 struct dm_table *dm_get_table(struct mapped_device *md)
408 read_lock(&md->map_lock);
412 read_unlock(&md->map_lock);
418 * Get the geometry associated with a dm device
420 int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo)
428 * Set the geometry of a device.
430 int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo)
432 sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors;
434 if (geo->start > sz) {
435 DMWARN("Start sector is beyond the geometry limits.");
444 /*-----------------------------------------------------------------
446 * A more elegant soln is in the works that uses the queue
447 * merge fn, unfortunately there are a couple of changes to
448 * the block layer that I want to make for this. So in the
449 * interests of getting something for people to use I give
450 * you this clearly demarcated crap.
451 *---------------------------------------------------------------*/
454 * Decrements the number of outstanding ios that a bio has been
455 * cloned into, completing the original io if necc.
457 static void dec_pending(struct dm_io *io, int error)
462 if (atomic_dec_and_test(&io->io_count)) {
464 /* nudge anyone waiting on suspend queue */
465 wake_up(&io->md->wait);
467 blk_add_trace_bio(io->md->queue, io->bio, BLK_TA_COMPLETE);
469 bio_endio(io->bio, io->bio->bi_size, io->error);
474 static int clone_endio(struct bio *bio, unsigned int done, int error)
477 struct target_io *tio = bio->bi_private;
478 struct dm_io *io = tio->io;
479 dm_endio_fn endio = tio->ti->type->end_io;
484 if (!bio_flagged(bio, BIO_UPTODATE) && !error)
488 r = endio(tio->ti, bio, error, &tio->info);
493 /* the target wants another shot at the io */
497 free_tio(io->md, tio);
498 dec_pending(io, error);
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)
532 BUG_ON(!clone->bi_size);
534 clone->bi_end_io = clone_endio;
535 clone->bi_private = tio;
538 * Map the clone. If r == 0 we don't need to do
539 * anything, the target has assumed ownership of
542 atomic_inc(&tio->io->io_count);
543 sector = clone->bi_sector;
544 r = ti->type->map(ti, clone, &tio->info);
546 /* the bio has been remapped so dispatch it */
548 blk_add_trace_remap(bdev_get_queue(clone->bi_bdev), clone,
549 tio->io->bio->bi_bdev->bd_dev, sector,
552 generic_make_request(clone);
556 /* error the io and bail out */
557 struct dm_io *io = tio->io;
558 free_tio(tio->io->md, tio);
565 struct mapped_device *md;
566 struct dm_table *map;
570 sector_t sector_count;
574 static void dm_bio_destructor(struct bio *bio)
576 bio_free(bio, dm_set);
580 * Creates a little bio that is just does part of a bvec.
582 static struct bio *split_bvec(struct bio *bio, sector_t sector,
583 unsigned short idx, unsigned int offset,
587 struct bio_vec *bv = bio->bi_io_vec + idx;
589 clone = bio_alloc_bioset(GFP_NOIO, 1, dm_set);
590 clone->bi_destructor = dm_bio_destructor;
591 *clone->bi_io_vec = *bv;
593 clone->bi_sector = sector;
594 clone->bi_bdev = bio->bi_bdev;
595 clone->bi_rw = bio->bi_rw;
597 clone->bi_size = to_bytes(len);
598 clone->bi_io_vec->bv_offset = offset;
599 clone->bi_io_vec->bv_len = clone->bi_size;
605 * Creates a bio that consists of range of complete bvecs.
607 static struct bio *clone_bio(struct bio *bio, sector_t sector,
608 unsigned short idx, unsigned short bv_count,
613 clone = bio_clone(bio, GFP_NOIO);
614 clone->bi_sector = sector;
616 clone->bi_vcnt = idx + bv_count;
617 clone->bi_size = to_bytes(len);
618 clone->bi_flags &= ~(1 << BIO_SEG_VALID);
623 static void __clone_and_map(struct clone_info *ci)
625 struct bio *clone, *bio = ci->bio;
626 struct dm_target *ti = dm_table_find_target(ci->map, ci->sector);
627 sector_t len = 0, max = max_io_len(ci->md, ci->sector, ti);
628 struct target_io *tio;
631 * Allocate a target io object.
633 tio = alloc_tio(ci->md);
636 memset(&tio->info, 0, sizeof(tio->info));
638 if (ci->sector_count <= max) {
640 * Optimise for the simple case where we can do all of
641 * the remaining io with a single clone.
643 clone = clone_bio(bio, ci->sector, ci->idx,
644 bio->bi_vcnt - ci->idx, ci->sector_count);
645 __map_bio(ti, clone, tio);
646 ci->sector_count = 0;
648 } else if (to_sector(bio->bi_io_vec[ci->idx].bv_len) <= max) {
650 * There are some bvecs that don't span targets.
651 * Do as many of these as possible.
654 sector_t remaining = max;
657 for (i = ci->idx; remaining && (i < bio->bi_vcnt); i++) {
658 bv_len = to_sector(bio->bi_io_vec[i].bv_len);
660 if (bv_len > remaining)
667 clone = clone_bio(bio, ci->sector, ci->idx, i - ci->idx, len);
668 __map_bio(ti, clone, tio);
671 ci->sector_count -= len;
676 * Handle a bvec that must be split between two or more targets.
678 struct bio_vec *bv = bio->bi_io_vec + ci->idx;
679 sector_t remaining = to_sector(bv->bv_len);
680 unsigned int offset = 0;
684 ti = dm_table_find_target(ci->map, ci->sector);
685 max = max_io_len(ci->md, ci->sector, ti);
687 tio = alloc_tio(ci->md);
690 memset(&tio->info, 0, sizeof(tio->info));
693 len = min(remaining, max);
695 clone = split_bvec(bio, ci->sector, ci->idx,
696 bv->bv_offset + offset, len);
698 __map_bio(ti, clone, tio);
701 ci->sector_count -= len;
702 offset += to_bytes(len);
703 } while (remaining -= len);
710 * Split the bio into several clones.
712 static void __split_bio(struct mapped_device *md, struct bio *bio)
714 struct clone_info ci;
716 ci.map = dm_get_table(md);
718 bio_io_error(bio, bio->bi_size);
724 ci.io = alloc_io(md);
726 atomic_set(&ci.io->io_count, 1);
729 ci.sector = bio->bi_sector;
730 ci.sector_count = bio_sectors(bio);
731 ci.idx = bio->bi_idx;
733 start_io_acct(ci.io);
734 while (ci.sector_count)
735 __clone_and_map(&ci);
737 /* drop the extra reference count */
738 dec_pending(ci.io, 0);
739 dm_table_put(ci.map);
741 /*-----------------------------------------------------------------
743 *---------------------------------------------------------------*/
746 * The request function that just remaps the bio built up by
749 static int dm_request(request_queue_t *q, struct bio *bio)
752 int rw = bio_data_dir(bio);
753 struct mapped_device *md = q->queuedata;
755 down_read(&md->io_lock);
757 disk_stat_inc(dm_disk(md), ios[rw]);
758 disk_stat_add(dm_disk(md), sectors[rw], bio_sectors(bio));
761 * If we're suspended we have to queue
764 while (test_bit(DMF_BLOCK_IO, &md->flags)) {
765 up_read(&md->io_lock);
767 if (bio_rw(bio) == READA) {
768 bio_io_error(bio, bio->bi_size);
772 r = queue_io(md, bio);
774 bio_io_error(bio, bio->bi_size);
778 return 0; /* deferred successfully */
781 * We're in a while loop, because someone could suspend
782 * before we get to the following read lock.
784 down_read(&md->io_lock);
787 __split_bio(md, bio);
788 up_read(&md->io_lock);
792 static int dm_flush_all(request_queue_t *q, struct gendisk *disk,
793 sector_t *error_sector)
795 struct mapped_device *md = q->queuedata;
796 struct dm_table *map = dm_get_table(md);
800 ret = dm_table_flush_all(map);
807 static void dm_unplug_all(request_queue_t *q)
809 struct mapped_device *md = q->queuedata;
810 struct dm_table *map = dm_get_table(md);
813 dm_table_unplug_all(map);
818 static int dm_any_congested(void *congested_data, int bdi_bits)
821 struct mapped_device *md = (struct mapped_device *) congested_data;
822 struct dm_table *map = dm_get_table(md);
824 if (!map || test_bit(DMF_BLOCK_IO, &md->flags))
827 r = dm_table_any_congested(map, bdi_bits);
833 /*-----------------------------------------------------------------
834 * An IDR is used to keep track of allocated minor numbers.
835 *---------------------------------------------------------------*/
836 static DEFINE_IDR(_minor_idr);
838 static void free_minor(int minor)
840 spin_lock(&_minor_lock);
841 idr_remove(&_minor_idr, minor);
842 spin_unlock(&_minor_lock);
846 * See if the device with a specific minor # is free.
848 static int specific_minor(struct mapped_device *md, int minor)
852 if (minor >= (1 << MINORBITS))
855 r = idr_pre_get(&_minor_idr, GFP_KERNEL);
859 spin_lock(&_minor_lock);
861 if (idr_find(&_minor_idr, minor)) {
866 r = idr_get_new_above(&_minor_idr, MINOR_ALLOCED, minor, &m);
871 idr_remove(&_minor_idr, m);
877 spin_unlock(&_minor_lock);
881 static int next_free_minor(struct mapped_device *md, int *minor)
885 r = idr_pre_get(&_minor_idr, GFP_KERNEL);
889 spin_lock(&_minor_lock);
891 r = idr_get_new(&_minor_idr, MINOR_ALLOCED, &m);
896 if (m >= (1 << MINORBITS)) {
897 idr_remove(&_minor_idr, m);
905 spin_unlock(&_minor_lock);
909 static struct block_device_operations dm_blk_dops;
912 * Allocate and initialise a blank device with a given minor.
914 static struct mapped_device *alloc_dev(int minor)
917 struct mapped_device *md = kmalloc(sizeof(*md), GFP_KERNEL);
921 DMWARN("unable to allocate device, out of memory.");
925 if (!try_module_get(THIS_MODULE))
928 /* get a minor number for the dev */
929 if (minor == DM_ANY_MINOR)
930 r = next_free_minor(md, &minor);
932 r = specific_minor(md, minor);
936 memset(md, 0, sizeof(*md));
937 init_rwsem(&md->io_lock);
938 init_MUTEX(&md->suspend_lock);
939 rwlock_init(&md->map_lock);
940 atomic_set(&md->holders, 1);
941 atomic_set(&md->open_count, 0);
942 atomic_set(&md->event_nr, 0);
944 md->queue = blk_alloc_queue(GFP_KERNEL);
948 md->queue->queuedata = md;
949 md->queue->backing_dev_info.congested_fn = dm_any_congested;
950 md->queue->backing_dev_info.congested_data = md;
951 blk_queue_make_request(md->queue, dm_request);
952 blk_queue_bounce_limit(md->queue, BLK_BOUNCE_ANY);
953 md->queue->unplug_fn = dm_unplug_all;
954 md->queue->issue_flush_fn = dm_flush_all;
956 md->io_pool = mempool_create_slab_pool(MIN_IOS, _io_cache);
960 md->tio_pool = mempool_create_slab_pool(MIN_IOS, _tio_cache);
964 md->disk = alloc_disk(1);
968 atomic_set(&md->pending, 0);
969 init_waitqueue_head(&md->wait);
970 init_waitqueue_head(&md->eventq);
972 md->disk->major = _major;
973 md->disk->first_minor = minor;
974 md->disk->fops = &dm_blk_dops;
975 md->disk->queue = md->queue;
976 md->disk->private_data = md;
977 sprintf(md->disk->disk_name, "dm-%d", minor);
979 format_dev_t(md->name, MKDEV(_major, minor));
981 /* Populate the mapping, nobody knows we exist yet */
982 spin_lock(&_minor_lock);
983 old_md = idr_replace(&_minor_idr, md, minor);
984 spin_unlock(&_minor_lock);
986 BUG_ON(old_md != MINOR_ALLOCED);
991 mempool_destroy(md->tio_pool);
993 mempool_destroy(md->io_pool);
995 blk_cleanup_queue(md->queue);
998 module_put(THIS_MODULE);
1004 static void free_dev(struct mapped_device *md)
1006 int minor = md->disk->first_minor;
1008 if (md->suspended_bdev) {
1009 thaw_bdev(md->suspended_bdev, NULL);
1010 bdput(md->suspended_bdev);
1012 mempool_destroy(md->tio_pool);
1013 mempool_destroy(md->io_pool);
1014 del_gendisk(md->disk);
1017 spin_lock(&_minor_lock);
1018 md->disk->private_data = NULL;
1019 spin_unlock(&_minor_lock);
1022 blk_cleanup_queue(md->queue);
1023 module_put(THIS_MODULE);
1028 * Bind a table to the device.
1030 static void event_callback(void *context)
1032 struct mapped_device *md = (struct mapped_device *) context;
1034 atomic_inc(&md->event_nr);
1035 wake_up(&md->eventq);
1038 static void __set_size(struct mapped_device *md, sector_t size)
1040 set_capacity(md->disk, size);
1042 mutex_lock(&md->suspended_bdev->bd_inode->i_mutex);
1043 i_size_write(md->suspended_bdev->bd_inode, (loff_t)size << SECTOR_SHIFT);
1044 mutex_unlock(&md->suspended_bdev->bd_inode->i_mutex);
1047 static int __bind(struct mapped_device *md, struct dm_table *t)
1049 request_queue_t *q = md->queue;
1052 size = dm_table_get_size(t);
1055 * Wipe any geometry if the size of the table changed.
1057 if (size != get_capacity(md->disk))
1058 memset(&md->geometry, 0, sizeof(md->geometry));
1060 __set_size(md, size);
1065 dm_table_event_callback(t, event_callback, md);
1067 write_lock(&md->map_lock);
1069 dm_table_set_restrictions(t, q);
1070 write_unlock(&md->map_lock);
1075 static void __unbind(struct mapped_device *md)
1077 struct dm_table *map = md->map;
1082 dm_table_event_callback(map, NULL, NULL);
1083 write_lock(&md->map_lock);
1085 write_unlock(&md->map_lock);
1090 * Constructor for a new device.
1092 int dm_create(int minor, struct mapped_device **result)
1094 struct mapped_device *md;
1096 md = alloc_dev(minor);
1104 static struct mapped_device *dm_find_md(dev_t dev)
1106 struct mapped_device *md;
1107 unsigned minor = MINOR(dev);
1109 if (MAJOR(dev) != _major || minor >= (1 << MINORBITS))
1112 spin_lock(&_minor_lock);
1114 md = idr_find(&_minor_idr, minor);
1115 if (md && (md == MINOR_ALLOCED ||
1116 (dm_disk(md)->first_minor != minor) ||
1117 test_bit(DMF_FREEING, &md->flags))) {
1123 spin_unlock(&_minor_lock);
1128 struct mapped_device *dm_get_md(dev_t dev)
1130 struct mapped_device *md = dm_find_md(dev);
1138 void *dm_get_mdptr(struct mapped_device *md)
1140 return md->interface_ptr;
1143 void dm_set_mdptr(struct mapped_device *md, void *ptr)
1145 md->interface_ptr = ptr;
1148 void dm_get(struct mapped_device *md)
1150 atomic_inc(&md->holders);
1153 const char *dm_device_name(struct mapped_device *md)
1157 EXPORT_SYMBOL_GPL(dm_device_name);
1159 void dm_put(struct mapped_device *md)
1161 struct dm_table *map;
1163 BUG_ON(test_bit(DMF_FREEING, &md->flags));
1165 if (atomic_dec_and_lock(&md->holders, &_minor_lock)) {
1166 map = dm_get_table(md);
1167 idr_replace(&_minor_idr, MINOR_ALLOCED, dm_disk(md)->first_minor);
1168 set_bit(DMF_FREEING, &md->flags);
1169 spin_unlock(&_minor_lock);
1170 if (!dm_suspended(md)) {
1171 dm_table_presuspend_targets(map);
1172 dm_table_postsuspend_targets(map);
1181 * Process the deferred bios
1183 static void __flush_deferred_io(struct mapped_device *md, struct bio *c)
1196 * Swap in a new table (destroying old one).
1198 int dm_swap_table(struct mapped_device *md, struct dm_table *table)
1202 down(&md->suspend_lock);
1204 /* device must be suspended */
1205 if (!dm_suspended(md))
1209 r = __bind(md, table);
1212 up(&md->suspend_lock);
1217 * Functions to lock and unlock any filesystem running on the
1220 static int lock_fs(struct mapped_device *md)
1224 WARN_ON(md->frozen_sb);
1226 md->frozen_sb = freeze_bdev(md->suspended_bdev);
1227 if (IS_ERR(md->frozen_sb)) {
1228 r = PTR_ERR(md->frozen_sb);
1229 md->frozen_sb = NULL;
1233 set_bit(DMF_FROZEN, &md->flags);
1235 /* don't bdput right now, we don't want the bdev
1236 * to go away while it is locked.
1241 static void unlock_fs(struct mapped_device *md)
1243 if (!test_bit(DMF_FROZEN, &md->flags))
1246 thaw_bdev(md->suspended_bdev, md->frozen_sb);
1247 md->frozen_sb = NULL;
1248 clear_bit(DMF_FROZEN, &md->flags);
1252 * We need to be able to change a mapping table under a mounted
1253 * filesystem. For example we might want to move some data in
1254 * the background. Before the table can be swapped with
1255 * dm_bind_table, dm_suspend must be called to flush any in
1256 * flight bios and ensure that any further io gets deferred.
1258 int dm_suspend(struct mapped_device *md, int do_lockfs)
1260 struct dm_table *map = NULL;
1261 DECLARE_WAITQUEUE(wait, current);
1265 down(&md->suspend_lock);
1267 if (dm_suspended(md))
1270 map = dm_get_table(md);
1272 /* This does not get reverted if there's an error later. */
1273 dm_table_presuspend_targets(map);
1275 md->suspended_bdev = bdget_disk(md->disk, 0);
1276 if (!md->suspended_bdev) {
1277 DMWARN("bdget failed in dm_suspend");
1282 /* Flush I/O to the device. */
1290 * First we set the BLOCK_IO flag so no more ios will be mapped.
1292 down_write(&md->io_lock);
1293 set_bit(DMF_BLOCK_IO, &md->flags);
1295 add_wait_queue(&md->wait, &wait);
1296 up_write(&md->io_lock);
1300 dm_table_unplug_all(map);
1303 * Then we wait for the already mapped ios to
1307 set_current_state(TASK_INTERRUPTIBLE);
1309 if (!atomic_read(&md->pending) || signal_pending(current))
1314 set_current_state(TASK_RUNNING);
1316 down_write(&md->io_lock);
1317 remove_wait_queue(&md->wait, &wait);
1319 /* were we interrupted ? */
1321 if (atomic_read(&md->pending)) {
1322 clear_bit(DMF_BLOCK_IO, &md->flags);
1323 def = bio_list_get(&md->deferred);
1324 __flush_deferred_io(md, def);
1325 up_write(&md->io_lock);
1329 up_write(&md->io_lock);
1331 dm_table_postsuspend_targets(map);
1333 set_bit(DMF_SUSPENDED, &md->flags);
1338 if (r && md->suspended_bdev) {
1339 bdput(md->suspended_bdev);
1340 md->suspended_bdev = NULL;
1344 up(&md->suspend_lock);
1348 int dm_resume(struct mapped_device *md)
1352 struct dm_table *map = NULL;
1354 down(&md->suspend_lock);
1355 if (!dm_suspended(md))
1358 map = dm_get_table(md);
1359 if (!map || !dm_table_get_size(map))
1362 dm_table_resume_targets(map);
1364 down_write(&md->io_lock);
1365 clear_bit(DMF_BLOCK_IO, &md->flags);
1367 def = bio_list_get(&md->deferred);
1368 __flush_deferred_io(md, def);
1369 up_write(&md->io_lock);
1373 bdput(md->suspended_bdev);
1374 md->suspended_bdev = NULL;
1376 clear_bit(DMF_SUSPENDED, &md->flags);
1378 dm_table_unplug_all(map);
1384 up(&md->suspend_lock);
1389 /*-----------------------------------------------------------------
1390 * Event notification.
1391 *---------------------------------------------------------------*/
1392 uint32_t dm_get_event_nr(struct mapped_device *md)
1394 return atomic_read(&md->event_nr);
1397 int dm_wait_event(struct mapped_device *md, int event_nr)
1399 return wait_event_interruptible(md->eventq,
1400 (event_nr != atomic_read(&md->event_nr)));
1404 * The gendisk is only valid as long as you have a reference
1407 struct gendisk *dm_disk(struct mapped_device *md)
1412 int dm_suspended(struct mapped_device *md)
1414 return test_bit(DMF_SUSPENDED, &md->flags);
1417 static struct block_device_operations dm_blk_dops = {
1418 .open = dm_blk_open,
1419 .release = dm_blk_close,
1420 .ioctl = dm_blk_ioctl,
1421 .getgeo = dm_blk_getgeo,
1422 .owner = THIS_MODULE
1425 EXPORT_SYMBOL(dm_get_mapinfo);
1430 module_init(dm_init);
1431 module_exit(dm_exit);
1433 module_param(major, uint, 0);
1434 MODULE_PARM_DESC(major, "The major number of the device mapper");
1435 MODULE_DESCRIPTION(DM_NAME " driver");
1436 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
1437 MODULE_LICENSE("GPL");