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
71 #define DMF_NOFLUSH_SUSPENDING 5
73 struct mapped_device {
74 struct rw_semaphore io_lock;
75 struct semaphore suspend_lock;
76 spinlock_t pushback_lock;
83 request_queue_t *queue;
90 * A list of ios that arrived while we were suspended.
93 wait_queue_head_t wait;
94 struct bio_list deferred;
95 struct bio_list pushback;
98 * The current mapping.
100 struct dm_table *map;
103 * io objects are allocated from here.
114 wait_queue_head_t eventq;
117 * freeze/thaw support require holding onto a super block
119 struct super_block *frozen_sb;
120 struct block_device *suspended_bdev;
122 /* forced geometry settings */
123 struct hd_geometry geometry;
127 static struct kmem_cache *_io_cache;
128 static struct kmem_cache *_tio_cache;
130 static int __init local_init(void)
134 /* allocate a slab for the dm_ios */
135 _io_cache = kmem_cache_create("dm_io",
136 sizeof(struct dm_io), 0, 0, NULL, NULL);
140 /* allocate a slab for the target ios */
141 _tio_cache = kmem_cache_create("dm_tio", sizeof(struct target_io),
144 kmem_cache_destroy(_io_cache);
149 r = register_blkdev(_major, _name);
151 kmem_cache_destroy(_tio_cache);
152 kmem_cache_destroy(_io_cache);
162 static void local_exit(void)
164 kmem_cache_destroy(_tio_cache);
165 kmem_cache_destroy(_io_cache);
167 if (unregister_blkdev(_major, _name) < 0)
168 DMERR("unregister_blkdev failed");
172 DMINFO("cleaned up");
175 int (*_inits[])(void) __initdata = {
183 void (*_exits[])(void) = {
191 static int __init dm_init(void)
193 const int count = ARRAY_SIZE(_inits);
197 for (i = 0; i < count; i++) {
212 static void __exit dm_exit(void)
214 int i = ARRAY_SIZE(_exits);
221 * Block device functions
223 static int dm_blk_open(struct inode *inode, struct file *file)
225 struct mapped_device *md;
227 spin_lock(&_minor_lock);
229 md = inode->i_bdev->bd_disk->private_data;
233 if (test_bit(DMF_FREEING, &md->flags) ||
234 test_bit(DMF_DELETING, &md->flags)) {
240 atomic_inc(&md->open_count);
243 spin_unlock(&_minor_lock);
245 return md ? 0 : -ENXIO;
248 static int dm_blk_close(struct inode *inode, struct file *file)
250 struct mapped_device *md;
252 md = inode->i_bdev->bd_disk->private_data;
253 atomic_dec(&md->open_count);
258 int dm_open_count(struct mapped_device *md)
260 return atomic_read(&md->open_count);
264 * Guarantees nothing is using the device before it's deleted.
266 int dm_lock_for_deletion(struct mapped_device *md)
270 spin_lock(&_minor_lock);
272 if (dm_open_count(md))
275 set_bit(DMF_DELETING, &md->flags);
277 spin_unlock(&_minor_lock);
282 static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo)
284 struct mapped_device *md = bdev->bd_disk->private_data;
286 return dm_get_geometry(md, geo);
289 static int dm_blk_ioctl(struct inode *inode, struct file *file,
290 unsigned int cmd, unsigned long arg)
292 struct mapped_device *md;
293 struct dm_table *map;
294 struct dm_target *tgt;
297 /* We don't really need this lock, but we do need 'inode'. */
300 md = inode->i_bdev->bd_disk->private_data;
302 map = dm_get_table(md);
304 if (!map || !dm_table_get_size(map))
307 /* We only support devices that have a single target */
308 if (dm_table_get_num_targets(map) != 1)
311 tgt = dm_table_get_target(map, 0);
313 if (dm_suspended(md)) {
318 if (tgt->type->ioctl)
319 r = tgt->type->ioctl(tgt, inode, file, cmd, arg);
328 static inline struct dm_io *alloc_io(struct mapped_device *md)
330 return mempool_alloc(md->io_pool, GFP_NOIO);
333 static inline void free_io(struct mapped_device *md, struct dm_io *io)
335 mempool_free(io, md->io_pool);
338 static inline struct target_io *alloc_tio(struct mapped_device *md)
340 return mempool_alloc(md->tio_pool, GFP_NOIO);
343 static inline void free_tio(struct mapped_device *md, struct target_io *tio)
345 mempool_free(tio, md->tio_pool);
348 static void start_io_acct(struct dm_io *io)
350 struct mapped_device *md = io->md;
352 io->start_time = jiffies;
355 disk_round_stats(dm_disk(md));
357 dm_disk(md)->in_flight = atomic_inc_return(&md->pending);
360 static int end_io_acct(struct dm_io *io)
362 struct mapped_device *md = io->md;
363 struct bio *bio = io->bio;
364 unsigned long duration = jiffies - io->start_time;
366 int rw = bio_data_dir(bio);
369 disk_round_stats(dm_disk(md));
371 dm_disk(md)->in_flight = pending = atomic_dec_return(&md->pending);
373 disk_stat_add(dm_disk(md), ticks[rw], duration);
379 * Add the bio to the list of deferred io.
381 static int queue_io(struct mapped_device *md, struct bio *bio)
383 down_write(&md->io_lock);
385 if (!test_bit(DMF_BLOCK_IO, &md->flags)) {
386 up_write(&md->io_lock);
390 bio_list_add(&md->deferred, bio);
392 up_write(&md->io_lock);
393 return 0; /* deferred successfully */
397 * Everyone (including functions in this file), should use this
398 * function to access the md->map field, and make sure they call
399 * dm_table_put() when finished.
401 struct dm_table *dm_get_table(struct mapped_device *md)
405 read_lock(&md->map_lock);
409 read_unlock(&md->map_lock);
415 * Get the geometry associated with a dm device
417 int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo)
425 * Set the geometry of a device.
427 int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo)
429 sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors;
431 if (geo->start > sz) {
432 DMWARN("Start sector is beyond the geometry limits.");
441 /*-----------------------------------------------------------------
443 * A more elegant soln is in the works that uses the queue
444 * merge fn, unfortunately there are a couple of changes to
445 * the block layer that I want to make for this. So in the
446 * interests of getting something for people to use I give
447 * you this clearly demarcated crap.
448 *---------------------------------------------------------------*/
450 static int __noflush_suspending(struct mapped_device *md)
452 return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
456 * Decrements the number of outstanding ios that a bio has been
457 * cloned into, completing the original io if necc.
459 static void dec_pending(struct dm_io *io, int error)
463 /* Push-back supersedes any I/O errors */
464 if (error && !(io->error > 0 && __noflush_suspending(io->md)))
467 if (atomic_dec_and_test(&io->io_count)) {
468 if (io->error == DM_ENDIO_REQUEUE) {
470 * Target requested pushing back the I/O.
471 * This must be handled before the sleeper on
472 * suspend queue merges the pushback list.
474 spin_lock_irqsave(&io->md->pushback_lock, flags);
475 if (__noflush_suspending(io->md))
476 bio_list_add(&io->md->pushback, io->bio);
478 /* noflush suspend was interrupted. */
480 spin_unlock_irqrestore(&io->md->pushback_lock, flags);
484 /* nudge anyone waiting on suspend queue */
485 wake_up(&io->md->wait);
487 if (io->error != DM_ENDIO_REQUEUE) {
488 blk_add_trace_bio(io->md->queue, io->bio,
491 bio_endio(io->bio, io->bio->bi_size, io->error);
498 static int clone_endio(struct bio *bio, unsigned int done, int error)
501 struct target_io *tio = bio->bi_private;
502 struct mapped_device *md = tio->io->md;
503 dm_endio_fn endio = tio->ti->type->end_io;
508 if (!bio_flagged(bio, BIO_UPTODATE) && !error)
512 r = endio(tio->ti, bio, error, &tio->info);
513 if (r < 0 || r == DM_ENDIO_REQUEUE)
515 * error and requeue request are handled
519 else if (r == DM_ENDIO_INCOMPLETE)
520 /* The target will handle the io */
523 DMWARN("unimplemented target endio return value: %d", r);
528 dec_pending(tio->io, error);
531 * Store md for cleanup instead of tio which is about to get freed.
533 bio->bi_private = md->bs;
540 static sector_t max_io_len(struct mapped_device *md,
541 sector_t sector, struct dm_target *ti)
543 sector_t offset = sector - ti->begin;
544 sector_t len = ti->len - offset;
547 * Does the target need to split even further ?
551 boundary = ((offset + ti->split_io) & ~(ti->split_io - 1))
560 static void __map_bio(struct dm_target *ti, struct bio *clone,
561 struct target_io *tio)
565 struct mapped_device *md;
570 BUG_ON(!clone->bi_size);
572 clone->bi_end_io = clone_endio;
573 clone->bi_private = tio;
576 * Map the clone. If r == 0 we don't need to do
577 * anything, the target has assumed ownership of
580 atomic_inc(&tio->io->io_count);
581 sector = clone->bi_sector;
582 r = ti->type->map(ti, clone, &tio->info);
583 if (r == DM_MAPIO_REMAPPED) {
584 /* the bio has been remapped so dispatch it */
586 blk_add_trace_remap(bdev_get_queue(clone->bi_bdev), clone,
587 tio->io->bio->bi_bdev->bd_dev, sector,
590 generic_make_request(clone);
591 } else if (r < 0 || r == DM_MAPIO_REQUEUE) {
592 /* error the io and bail out, or requeue it if needed */
594 dec_pending(tio->io, r);
596 * Store bio_set for cleanup.
598 clone->bi_private = md->bs;
602 DMWARN("unimplemented target map return value: %d", r);
608 struct mapped_device *md;
609 struct dm_table *map;
613 sector_t sector_count;
617 static void dm_bio_destructor(struct bio *bio)
619 struct bio_set *bs = bio->bi_private;
625 * Creates a little bio that is just does part of a bvec.
627 static struct bio *split_bvec(struct bio *bio, sector_t sector,
628 unsigned short idx, unsigned int offset,
629 unsigned int len, struct bio_set *bs)
632 struct bio_vec *bv = bio->bi_io_vec + idx;
634 clone = bio_alloc_bioset(GFP_NOIO, 1, bs);
635 clone->bi_destructor = dm_bio_destructor;
636 *clone->bi_io_vec = *bv;
638 clone->bi_sector = sector;
639 clone->bi_bdev = bio->bi_bdev;
640 clone->bi_rw = bio->bi_rw;
642 clone->bi_size = to_bytes(len);
643 clone->bi_io_vec->bv_offset = offset;
644 clone->bi_io_vec->bv_len = clone->bi_size;
650 * Creates a bio that consists of range of complete bvecs.
652 static struct bio *clone_bio(struct bio *bio, sector_t sector,
653 unsigned short idx, unsigned short bv_count,
654 unsigned int len, struct bio_set *bs)
658 clone = bio_alloc_bioset(GFP_NOIO, bio->bi_max_vecs, bs);
659 __bio_clone(clone, bio);
660 clone->bi_destructor = dm_bio_destructor;
661 clone->bi_sector = sector;
663 clone->bi_vcnt = idx + bv_count;
664 clone->bi_size = to_bytes(len);
665 clone->bi_flags &= ~(1 << BIO_SEG_VALID);
670 static void __clone_and_map(struct clone_info *ci)
672 struct bio *clone, *bio = ci->bio;
673 struct dm_target *ti = dm_table_find_target(ci->map, ci->sector);
674 sector_t len = 0, max = max_io_len(ci->md, ci->sector, ti);
675 struct target_io *tio;
678 * Allocate a target io object.
680 tio = alloc_tio(ci->md);
683 memset(&tio->info, 0, sizeof(tio->info));
685 if (ci->sector_count <= max) {
687 * Optimise for the simple case where we can do all of
688 * the remaining io with a single clone.
690 clone = clone_bio(bio, ci->sector, ci->idx,
691 bio->bi_vcnt - ci->idx, ci->sector_count,
693 __map_bio(ti, clone, tio);
694 ci->sector_count = 0;
696 } else if (to_sector(bio->bi_io_vec[ci->idx].bv_len) <= max) {
698 * There are some bvecs that don't span targets.
699 * Do as many of these as possible.
702 sector_t remaining = max;
705 for (i = ci->idx; remaining && (i < bio->bi_vcnt); i++) {
706 bv_len = to_sector(bio->bi_io_vec[i].bv_len);
708 if (bv_len > remaining)
715 clone = clone_bio(bio, ci->sector, ci->idx, i - ci->idx, len,
717 __map_bio(ti, clone, tio);
720 ci->sector_count -= len;
725 * Handle a bvec that must be split between two or more targets.
727 struct bio_vec *bv = bio->bi_io_vec + ci->idx;
728 sector_t remaining = to_sector(bv->bv_len);
729 unsigned int offset = 0;
733 ti = dm_table_find_target(ci->map, ci->sector);
734 max = max_io_len(ci->md, ci->sector, ti);
736 tio = alloc_tio(ci->md);
739 memset(&tio->info, 0, sizeof(tio->info));
742 len = min(remaining, max);
744 clone = split_bvec(bio, ci->sector, ci->idx,
745 bv->bv_offset + offset, len,
748 __map_bio(ti, clone, tio);
751 ci->sector_count -= len;
752 offset += to_bytes(len);
753 } while (remaining -= len);
760 * Split the bio into several clones.
762 static void __split_bio(struct mapped_device *md, struct bio *bio)
764 struct clone_info ci;
766 ci.map = dm_get_table(md);
768 bio_io_error(bio, bio->bi_size);
774 ci.io = alloc_io(md);
776 atomic_set(&ci.io->io_count, 1);
779 ci.sector = bio->bi_sector;
780 ci.sector_count = bio_sectors(bio);
781 ci.idx = bio->bi_idx;
783 start_io_acct(ci.io);
784 while (ci.sector_count)
785 __clone_and_map(&ci);
787 /* drop the extra reference count */
788 dec_pending(ci.io, 0);
789 dm_table_put(ci.map);
791 /*-----------------------------------------------------------------
793 *---------------------------------------------------------------*/
796 * The request function that just remaps the bio built up by
799 static int dm_request(request_queue_t *q, struct bio *bio)
802 int rw = bio_data_dir(bio);
803 struct mapped_device *md = q->queuedata;
805 down_read(&md->io_lock);
807 disk_stat_inc(dm_disk(md), ios[rw]);
808 disk_stat_add(dm_disk(md), sectors[rw], bio_sectors(bio));
811 * If we're suspended we have to queue
814 while (test_bit(DMF_BLOCK_IO, &md->flags)) {
815 up_read(&md->io_lock);
817 if (bio_rw(bio) == READA) {
818 bio_io_error(bio, bio->bi_size);
822 r = queue_io(md, bio);
824 bio_io_error(bio, bio->bi_size);
828 return 0; /* deferred successfully */
831 * We're in a while loop, because someone could suspend
832 * before we get to the following read lock.
834 down_read(&md->io_lock);
837 __split_bio(md, bio);
838 up_read(&md->io_lock);
842 static int dm_flush_all(request_queue_t *q, struct gendisk *disk,
843 sector_t *error_sector)
845 struct mapped_device *md = q->queuedata;
846 struct dm_table *map = dm_get_table(md);
850 ret = dm_table_flush_all(map);
857 static void dm_unplug_all(request_queue_t *q)
859 struct mapped_device *md = q->queuedata;
860 struct dm_table *map = dm_get_table(md);
863 dm_table_unplug_all(map);
868 static int dm_any_congested(void *congested_data, int bdi_bits)
871 struct mapped_device *md = (struct mapped_device *) congested_data;
872 struct dm_table *map = dm_get_table(md);
874 if (!map || test_bit(DMF_BLOCK_IO, &md->flags))
877 r = dm_table_any_congested(map, bdi_bits);
883 /*-----------------------------------------------------------------
884 * An IDR is used to keep track of allocated minor numbers.
885 *---------------------------------------------------------------*/
886 static DEFINE_IDR(_minor_idr);
888 static void free_minor(int minor)
890 spin_lock(&_minor_lock);
891 idr_remove(&_minor_idr, minor);
892 spin_unlock(&_minor_lock);
896 * See if the device with a specific minor # is free.
898 static int specific_minor(struct mapped_device *md, int minor)
902 if (minor >= (1 << MINORBITS))
905 r = idr_pre_get(&_minor_idr, GFP_KERNEL);
909 spin_lock(&_minor_lock);
911 if (idr_find(&_minor_idr, minor)) {
916 r = idr_get_new_above(&_minor_idr, MINOR_ALLOCED, minor, &m);
921 idr_remove(&_minor_idr, m);
927 spin_unlock(&_minor_lock);
931 static int next_free_minor(struct mapped_device *md, int *minor)
935 r = idr_pre_get(&_minor_idr, GFP_KERNEL);
939 spin_lock(&_minor_lock);
941 r = idr_get_new(&_minor_idr, MINOR_ALLOCED, &m);
946 if (m >= (1 << MINORBITS)) {
947 idr_remove(&_minor_idr, m);
955 spin_unlock(&_minor_lock);
959 static struct block_device_operations dm_blk_dops;
962 * Allocate and initialise a blank device with a given minor.
964 static struct mapped_device *alloc_dev(int minor)
967 struct mapped_device *md = kmalloc(sizeof(*md), GFP_KERNEL);
971 DMWARN("unable to allocate device, out of memory.");
975 if (!try_module_get(THIS_MODULE))
978 /* get a minor number for the dev */
979 if (minor == DM_ANY_MINOR)
980 r = next_free_minor(md, &minor);
982 r = specific_minor(md, minor);
986 memset(md, 0, sizeof(*md));
987 init_rwsem(&md->io_lock);
988 init_MUTEX(&md->suspend_lock);
989 spin_lock_init(&md->pushback_lock);
990 rwlock_init(&md->map_lock);
991 atomic_set(&md->holders, 1);
992 atomic_set(&md->open_count, 0);
993 atomic_set(&md->event_nr, 0);
995 md->queue = blk_alloc_queue(GFP_KERNEL);
997 goto bad1_free_minor;
999 md->queue->queuedata = md;
1000 md->queue->backing_dev_info.congested_fn = dm_any_congested;
1001 md->queue->backing_dev_info.congested_data = md;
1002 blk_queue_make_request(md->queue, dm_request);
1003 blk_queue_bounce_limit(md->queue, BLK_BOUNCE_ANY);
1004 md->queue->unplug_fn = dm_unplug_all;
1005 md->queue->issue_flush_fn = dm_flush_all;
1007 md->io_pool = mempool_create_slab_pool(MIN_IOS, _io_cache);
1011 md->tio_pool = mempool_create_slab_pool(MIN_IOS, _tio_cache);
1015 md->bs = bioset_create(16, 16, 4);
1019 md->disk = alloc_disk(1);
1023 atomic_set(&md->pending, 0);
1024 init_waitqueue_head(&md->wait);
1025 init_waitqueue_head(&md->eventq);
1027 md->disk->major = _major;
1028 md->disk->first_minor = minor;
1029 md->disk->fops = &dm_blk_dops;
1030 md->disk->queue = md->queue;
1031 md->disk->private_data = md;
1032 sprintf(md->disk->disk_name, "dm-%d", minor);
1034 format_dev_t(md->name, MKDEV(_major, minor));
1036 /* Populate the mapping, nobody knows we exist yet */
1037 spin_lock(&_minor_lock);
1038 old_md = idr_replace(&_minor_idr, md, minor);
1039 spin_unlock(&_minor_lock);
1041 BUG_ON(old_md != MINOR_ALLOCED);
1046 bioset_free(md->bs);
1048 mempool_destroy(md->tio_pool);
1050 mempool_destroy(md->io_pool);
1052 blk_cleanup_queue(md->queue);
1056 module_put(THIS_MODULE);
1062 static void free_dev(struct mapped_device *md)
1064 int minor = md->disk->first_minor;
1066 if (md->suspended_bdev) {
1067 thaw_bdev(md->suspended_bdev, NULL);
1068 bdput(md->suspended_bdev);
1070 mempool_destroy(md->tio_pool);
1071 mempool_destroy(md->io_pool);
1072 bioset_free(md->bs);
1073 del_gendisk(md->disk);
1076 spin_lock(&_minor_lock);
1077 md->disk->private_data = NULL;
1078 spin_unlock(&_minor_lock);
1081 blk_cleanup_queue(md->queue);
1082 module_put(THIS_MODULE);
1087 * Bind a table to the device.
1089 static void event_callback(void *context)
1091 struct mapped_device *md = (struct mapped_device *) context;
1093 atomic_inc(&md->event_nr);
1094 wake_up(&md->eventq);
1097 static void __set_size(struct mapped_device *md, sector_t size)
1099 set_capacity(md->disk, size);
1101 mutex_lock(&md->suspended_bdev->bd_inode->i_mutex);
1102 i_size_write(md->suspended_bdev->bd_inode, (loff_t)size << SECTOR_SHIFT);
1103 mutex_unlock(&md->suspended_bdev->bd_inode->i_mutex);
1106 static int __bind(struct mapped_device *md, struct dm_table *t)
1108 request_queue_t *q = md->queue;
1111 size = dm_table_get_size(t);
1114 * Wipe any geometry if the size of the table changed.
1116 if (size != get_capacity(md->disk))
1117 memset(&md->geometry, 0, sizeof(md->geometry));
1119 __set_size(md, size);
1124 dm_table_event_callback(t, event_callback, md);
1126 write_lock(&md->map_lock);
1128 dm_table_set_restrictions(t, q);
1129 write_unlock(&md->map_lock);
1134 static void __unbind(struct mapped_device *md)
1136 struct dm_table *map = md->map;
1141 dm_table_event_callback(map, NULL, NULL);
1142 write_lock(&md->map_lock);
1144 write_unlock(&md->map_lock);
1149 * Constructor for a new device.
1151 int dm_create(int minor, struct mapped_device **result)
1153 struct mapped_device *md;
1155 md = alloc_dev(minor);
1163 static struct mapped_device *dm_find_md(dev_t dev)
1165 struct mapped_device *md;
1166 unsigned minor = MINOR(dev);
1168 if (MAJOR(dev) != _major || minor >= (1 << MINORBITS))
1171 spin_lock(&_minor_lock);
1173 md = idr_find(&_minor_idr, minor);
1174 if (md && (md == MINOR_ALLOCED ||
1175 (dm_disk(md)->first_minor != minor) ||
1176 test_bit(DMF_FREEING, &md->flags))) {
1182 spin_unlock(&_minor_lock);
1187 struct mapped_device *dm_get_md(dev_t dev)
1189 struct mapped_device *md = dm_find_md(dev);
1197 void *dm_get_mdptr(struct mapped_device *md)
1199 return md->interface_ptr;
1202 void dm_set_mdptr(struct mapped_device *md, void *ptr)
1204 md->interface_ptr = ptr;
1207 void dm_get(struct mapped_device *md)
1209 atomic_inc(&md->holders);
1212 const char *dm_device_name(struct mapped_device *md)
1216 EXPORT_SYMBOL_GPL(dm_device_name);
1218 void dm_put(struct mapped_device *md)
1220 struct dm_table *map;
1222 BUG_ON(test_bit(DMF_FREEING, &md->flags));
1224 if (atomic_dec_and_lock(&md->holders, &_minor_lock)) {
1225 map = dm_get_table(md);
1226 idr_replace(&_minor_idr, MINOR_ALLOCED, dm_disk(md)->first_minor);
1227 set_bit(DMF_FREEING, &md->flags);
1228 spin_unlock(&_minor_lock);
1229 if (!dm_suspended(md)) {
1230 dm_table_presuspend_targets(map);
1231 dm_table_postsuspend_targets(map);
1240 * Process the deferred bios
1242 static void __flush_deferred_io(struct mapped_device *md, struct bio *c)
1255 * Swap in a new table (destroying old one).
1257 int dm_swap_table(struct mapped_device *md, struct dm_table *table)
1261 down(&md->suspend_lock);
1263 /* device must be suspended */
1264 if (!dm_suspended(md))
1268 r = __bind(md, table);
1271 up(&md->suspend_lock);
1276 * Functions to lock and unlock any filesystem running on the
1279 static int lock_fs(struct mapped_device *md)
1283 WARN_ON(md->frozen_sb);
1285 md->frozen_sb = freeze_bdev(md->suspended_bdev);
1286 if (IS_ERR(md->frozen_sb)) {
1287 r = PTR_ERR(md->frozen_sb);
1288 md->frozen_sb = NULL;
1292 set_bit(DMF_FROZEN, &md->flags);
1294 /* don't bdput right now, we don't want the bdev
1295 * to go away while it is locked.
1300 static void unlock_fs(struct mapped_device *md)
1302 if (!test_bit(DMF_FROZEN, &md->flags))
1305 thaw_bdev(md->suspended_bdev, md->frozen_sb);
1306 md->frozen_sb = NULL;
1307 clear_bit(DMF_FROZEN, &md->flags);
1311 * We need to be able to change a mapping table under a mounted
1312 * filesystem. For example we might want to move some data in
1313 * the background. Before the table can be swapped with
1314 * dm_bind_table, dm_suspend must be called to flush any in
1315 * flight bios and ensure that any further io gets deferred.
1317 int dm_suspend(struct mapped_device *md, unsigned suspend_flags)
1319 struct dm_table *map = NULL;
1320 unsigned long flags;
1321 DECLARE_WAITQUEUE(wait, current);
1324 int do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG ? 1 : 0;
1325 int noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG ? 1 : 0;
1327 down(&md->suspend_lock);
1329 if (dm_suspended(md))
1332 map = dm_get_table(md);
1335 * DMF_NOFLUSH_SUSPENDING must be set before presuspend.
1336 * This flag is cleared before dm_suspend returns.
1339 set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
1341 /* This does not get reverted if there's an error later. */
1342 dm_table_presuspend_targets(map);
1344 md->suspended_bdev = bdget_disk(md->disk, 0);
1345 if (!md->suspended_bdev) {
1346 DMWARN("bdget failed in dm_suspend");
1352 * Flush I/O to the device.
1353 * noflush supersedes do_lockfs, because lock_fs() needs to flush I/Os.
1355 if (do_lockfs && !noflush) {
1362 * First we set the BLOCK_IO flag so no more ios will be mapped.
1364 down_write(&md->io_lock);
1365 set_bit(DMF_BLOCK_IO, &md->flags);
1367 add_wait_queue(&md->wait, &wait);
1368 up_write(&md->io_lock);
1372 dm_table_unplug_all(map);
1375 * Then we wait for the already mapped ios to
1379 set_current_state(TASK_INTERRUPTIBLE);
1381 if (!atomic_read(&md->pending) || signal_pending(current))
1386 set_current_state(TASK_RUNNING);
1388 down_write(&md->io_lock);
1389 remove_wait_queue(&md->wait, &wait);
1392 spin_lock_irqsave(&md->pushback_lock, flags);
1393 clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
1394 bio_list_merge_head(&md->deferred, &md->pushback);
1395 bio_list_init(&md->pushback);
1396 spin_unlock_irqrestore(&md->pushback_lock, flags);
1399 /* were we interrupted ? */
1401 if (atomic_read(&md->pending)) {
1402 clear_bit(DMF_BLOCK_IO, &md->flags);
1403 def = bio_list_get(&md->deferred);
1404 __flush_deferred_io(md, def);
1405 up_write(&md->io_lock);
1407 goto out; /* pushback list is already flushed, so skip flush */
1409 up_write(&md->io_lock);
1411 dm_table_postsuspend_targets(map);
1413 set_bit(DMF_SUSPENDED, &md->flags);
1420 * Because there may be already I/Os in the pushback list,
1421 * flush them before return.
1423 down_write(&md->io_lock);
1425 spin_lock_irqsave(&md->pushback_lock, flags);
1426 clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
1427 bio_list_merge_head(&md->deferred, &md->pushback);
1428 bio_list_init(&md->pushback);
1429 spin_unlock_irqrestore(&md->pushback_lock, flags);
1431 def = bio_list_get(&md->deferred);
1432 __flush_deferred_io(md, def);
1433 up_write(&md->io_lock);
1437 if (r && md->suspended_bdev) {
1438 bdput(md->suspended_bdev);
1439 md->suspended_bdev = NULL;
1445 up(&md->suspend_lock);
1449 int dm_resume(struct mapped_device *md)
1453 struct dm_table *map = NULL;
1455 down(&md->suspend_lock);
1456 if (!dm_suspended(md))
1459 map = dm_get_table(md);
1460 if (!map || !dm_table_get_size(map))
1463 r = dm_table_resume_targets(map);
1467 down_write(&md->io_lock);
1468 clear_bit(DMF_BLOCK_IO, &md->flags);
1470 def = bio_list_get(&md->deferred);
1471 __flush_deferred_io(md, def);
1472 up_write(&md->io_lock);
1476 bdput(md->suspended_bdev);
1477 md->suspended_bdev = NULL;
1479 clear_bit(DMF_SUSPENDED, &md->flags);
1481 dm_table_unplug_all(map);
1483 kobject_uevent(&md->disk->kobj, KOBJ_CHANGE);
1489 up(&md->suspend_lock);
1494 /*-----------------------------------------------------------------
1495 * Event notification.
1496 *---------------------------------------------------------------*/
1497 uint32_t dm_get_event_nr(struct mapped_device *md)
1499 return atomic_read(&md->event_nr);
1502 int dm_wait_event(struct mapped_device *md, int event_nr)
1504 return wait_event_interruptible(md->eventq,
1505 (event_nr != atomic_read(&md->event_nr)));
1509 * The gendisk is only valid as long as you have a reference
1512 struct gendisk *dm_disk(struct mapped_device *md)
1517 int dm_suspended(struct mapped_device *md)
1519 return test_bit(DMF_SUSPENDED, &md->flags);
1522 int dm_noflush_suspending(struct dm_target *ti)
1524 struct mapped_device *md = dm_table_get_md(ti->table);
1525 int r = __noflush_suspending(md);
1531 EXPORT_SYMBOL_GPL(dm_noflush_suspending);
1533 static struct block_device_operations dm_blk_dops = {
1534 .open = dm_blk_open,
1535 .release = dm_blk_close,
1536 .ioctl = dm_blk_ioctl,
1537 .getgeo = dm_blk_getgeo,
1538 .owner = THIS_MODULE
1541 EXPORT_SYMBOL(dm_get_mapinfo);
1546 module_init(dm_init);
1547 module_exit(dm_exit);
1549 module_param(major, uint, 0);
1550 MODULE_PARM_DESC(major, "The major number of the device mapper");
1551 MODULE_DESCRIPTION(DM_NAME " driver");
1552 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
1553 MODULE_LICENSE("GPL");