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 dm_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 struct request_queue *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(dm_io, 0);
139 /* allocate a slab for the target ios */
140 _tio_cache = KMEM_CACHE(dm_target_io, 0);
142 kmem_cache_destroy(_io_cache);
147 r = register_blkdev(_major, _name);
149 kmem_cache_destroy(_tio_cache);
150 kmem_cache_destroy(_io_cache);
160 static void local_exit(void)
162 kmem_cache_destroy(_tio_cache);
163 kmem_cache_destroy(_io_cache);
164 unregister_blkdev(_major, _name);
168 DMINFO("cleaned up");
171 int (*_inits[])(void) __initdata = {
179 void (*_exits[])(void) = {
187 static int __init dm_init(void)
189 const int count = ARRAY_SIZE(_inits);
193 for (i = 0; i < count; i++) {
208 static void __exit dm_exit(void)
210 int i = ARRAY_SIZE(_exits);
217 * Block device functions
219 static int dm_blk_open(struct inode *inode, struct file *file)
221 struct mapped_device *md;
223 spin_lock(&_minor_lock);
225 md = inode->i_bdev->bd_disk->private_data;
229 if (test_bit(DMF_FREEING, &md->flags) ||
230 test_bit(DMF_DELETING, &md->flags)) {
236 atomic_inc(&md->open_count);
239 spin_unlock(&_minor_lock);
241 return md ? 0 : -ENXIO;
244 static int dm_blk_close(struct inode *inode, struct file *file)
246 struct mapped_device *md;
248 md = inode->i_bdev->bd_disk->private_data;
249 atomic_dec(&md->open_count);
254 int dm_open_count(struct mapped_device *md)
256 return atomic_read(&md->open_count);
260 * Guarantees nothing is using the device before it's deleted.
262 int dm_lock_for_deletion(struct mapped_device *md)
266 spin_lock(&_minor_lock);
268 if (dm_open_count(md))
271 set_bit(DMF_DELETING, &md->flags);
273 spin_unlock(&_minor_lock);
278 static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo)
280 struct mapped_device *md = bdev->bd_disk->private_data;
282 return dm_get_geometry(md, geo);
285 static int dm_blk_ioctl(struct inode *inode, struct file *file,
286 unsigned int cmd, unsigned long arg)
288 struct mapped_device *md;
289 struct dm_table *map;
290 struct dm_target *tgt;
293 /* We don't really need this lock, but we do need 'inode'. */
296 md = inode->i_bdev->bd_disk->private_data;
298 map = dm_get_table(md);
300 if (!map || !dm_table_get_size(map))
303 /* We only support devices that have a single target */
304 if (dm_table_get_num_targets(map) != 1)
307 tgt = dm_table_get_target(map, 0);
309 if (dm_suspended(md)) {
314 if (tgt->type->ioctl)
315 r = tgt->type->ioctl(tgt, inode, file, cmd, arg);
324 static struct dm_io *alloc_io(struct mapped_device *md)
326 return mempool_alloc(md->io_pool, GFP_NOIO);
329 static void free_io(struct mapped_device *md, struct dm_io *io)
331 mempool_free(io, md->io_pool);
334 static struct dm_target_io *alloc_tio(struct mapped_device *md)
336 return mempool_alloc(md->tio_pool, GFP_NOIO);
339 static void free_tio(struct mapped_device *md, struct dm_target_io *tio)
341 mempool_free(tio, md->tio_pool);
344 static void start_io_acct(struct dm_io *io)
346 struct mapped_device *md = io->md;
348 io->start_time = jiffies;
351 disk_round_stats(dm_disk(md));
353 dm_disk(md)->in_flight = atomic_inc_return(&md->pending);
356 static int end_io_acct(struct dm_io *io)
358 struct mapped_device *md = io->md;
359 struct bio *bio = io->bio;
360 unsigned long duration = jiffies - io->start_time;
362 int rw = bio_data_dir(bio);
365 disk_round_stats(dm_disk(md));
367 dm_disk(md)->in_flight = pending = atomic_dec_return(&md->pending);
369 disk_stat_add(dm_disk(md), ticks[rw], duration);
375 * Add the bio to the list of deferred io.
377 static int queue_io(struct mapped_device *md, struct bio *bio)
379 down_write(&md->io_lock);
381 if (!test_bit(DMF_BLOCK_IO, &md->flags)) {
382 up_write(&md->io_lock);
386 bio_list_add(&md->deferred, bio);
388 up_write(&md->io_lock);
389 return 0; /* deferred successfully */
393 * Everyone (including functions in this file), should use this
394 * function to access the md->map field, and make sure they call
395 * dm_table_put() when finished.
397 struct dm_table *dm_get_table(struct mapped_device *md)
401 read_lock(&md->map_lock);
405 read_unlock(&md->map_lock);
411 * Get the geometry associated with a dm device
413 int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo)
421 * Set the geometry of a device.
423 int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo)
425 sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors;
427 if (geo->start > sz) {
428 DMWARN("Start sector is beyond the geometry limits.");
437 /*-----------------------------------------------------------------
439 * A more elegant soln is in the works that uses the queue
440 * merge fn, unfortunately there are a couple of changes to
441 * the block layer that I want to make for this. So in the
442 * interests of getting something for people to use I give
443 * you this clearly demarcated crap.
444 *---------------------------------------------------------------*/
446 static int __noflush_suspending(struct mapped_device *md)
448 return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
452 * Decrements the number of outstanding ios that a bio has been
453 * cloned into, completing the original io if necc.
455 static void dec_pending(struct dm_io *io, int error)
459 /* Push-back supersedes any I/O errors */
460 if (error && !(io->error > 0 && __noflush_suspending(io->md)))
463 if (atomic_dec_and_test(&io->io_count)) {
464 if (io->error == DM_ENDIO_REQUEUE) {
466 * Target requested pushing back the I/O.
467 * This must be handled before the sleeper on
468 * suspend queue merges the pushback list.
470 spin_lock_irqsave(&io->md->pushback_lock, flags);
471 if (__noflush_suspending(io->md))
472 bio_list_add(&io->md->pushback, io->bio);
474 /* noflush suspend was interrupted. */
476 spin_unlock_irqrestore(&io->md->pushback_lock, flags);
480 /* nudge anyone waiting on suspend queue */
481 wake_up(&io->md->wait);
483 if (io->error != DM_ENDIO_REQUEUE) {
484 blk_add_trace_bio(io->md->queue, io->bio,
487 bio_endio(io->bio, io->bio->bi_size, io->error);
494 static int clone_endio(struct bio *bio, unsigned int done, int error)
497 struct dm_target_io *tio = bio->bi_private;
498 struct mapped_device *md = tio->io->md;
499 dm_endio_fn endio = tio->ti->type->end_io;
504 if (!bio_flagged(bio, BIO_UPTODATE) && !error)
508 r = endio(tio->ti, bio, error, &tio->info);
509 if (r < 0 || r == DM_ENDIO_REQUEUE)
511 * error and requeue request are handled
515 else if (r == DM_ENDIO_INCOMPLETE)
516 /* The target will handle the io */
519 DMWARN("unimplemented target endio return value: %d", r);
524 dec_pending(tio->io, error);
527 * Store md for cleanup instead of tio which is about to get freed.
529 bio->bi_private = md->bs;
536 static sector_t max_io_len(struct mapped_device *md,
537 sector_t sector, struct dm_target *ti)
539 sector_t offset = sector - ti->begin;
540 sector_t len = ti->len - offset;
543 * Does the target need to split even further ?
547 boundary = ((offset + ti->split_io) & ~(ti->split_io - 1))
556 static void __map_bio(struct dm_target *ti, struct bio *clone,
557 struct dm_target_io *tio)
561 struct mapped_device *md;
566 BUG_ON(!clone->bi_size);
568 clone->bi_end_io = clone_endio;
569 clone->bi_private = tio;
572 * Map the clone. If r == 0 we don't need to do
573 * anything, the target has assumed ownership of
576 atomic_inc(&tio->io->io_count);
577 sector = clone->bi_sector;
578 r = ti->type->map(ti, clone, &tio->info);
579 if (r == DM_MAPIO_REMAPPED) {
580 /* the bio has been remapped so dispatch it */
582 blk_add_trace_remap(bdev_get_queue(clone->bi_bdev), clone,
583 tio->io->bio->bi_bdev->bd_dev,
584 clone->bi_sector, sector);
586 generic_make_request(clone);
587 } else if (r < 0 || r == DM_MAPIO_REQUEUE) {
588 /* error the io and bail out, or requeue it if needed */
590 dec_pending(tio->io, r);
592 * Store bio_set for cleanup.
594 clone->bi_private = md->bs;
598 DMWARN("unimplemented target map return value: %d", r);
604 struct mapped_device *md;
605 struct dm_table *map;
609 sector_t sector_count;
613 static void dm_bio_destructor(struct bio *bio)
615 struct bio_set *bs = bio->bi_private;
621 * Creates a little bio that is just does part of a bvec.
623 static struct bio *split_bvec(struct bio *bio, sector_t sector,
624 unsigned short idx, unsigned int offset,
625 unsigned int len, struct bio_set *bs)
628 struct bio_vec *bv = bio->bi_io_vec + idx;
630 clone = bio_alloc_bioset(GFP_NOIO, 1, bs);
631 clone->bi_destructor = dm_bio_destructor;
632 *clone->bi_io_vec = *bv;
634 clone->bi_sector = sector;
635 clone->bi_bdev = bio->bi_bdev;
636 clone->bi_rw = bio->bi_rw;
638 clone->bi_size = to_bytes(len);
639 clone->bi_io_vec->bv_offset = offset;
640 clone->bi_io_vec->bv_len = clone->bi_size;
646 * Creates a bio that consists of range of complete bvecs.
648 static struct bio *clone_bio(struct bio *bio, sector_t sector,
649 unsigned short idx, unsigned short bv_count,
650 unsigned int len, struct bio_set *bs)
654 clone = bio_alloc_bioset(GFP_NOIO, bio->bi_max_vecs, bs);
655 __bio_clone(clone, bio);
656 clone->bi_destructor = dm_bio_destructor;
657 clone->bi_sector = sector;
659 clone->bi_vcnt = idx + bv_count;
660 clone->bi_size = to_bytes(len);
661 clone->bi_flags &= ~(1 << BIO_SEG_VALID);
666 static void __clone_and_map(struct clone_info *ci)
668 struct bio *clone, *bio = ci->bio;
669 struct dm_target *ti = dm_table_find_target(ci->map, ci->sector);
670 sector_t len = 0, max = max_io_len(ci->md, ci->sector, ti);
671 struct dm_target_io *tio;
674 * Allocate a target io object.
676 tio = alloc_tio(ci->md);
679 memset(&tio->info, 0, sizeof(tio->info));
681 if (ci->sector_count <= max) {
683 * Optimise for the simple case where we can do all of
684 * the remaining io with a single clone.
686 clone = clone_bio(bio, ci->sector, ci->idx,
687 bio->bi_vcnt - ci->idx, ci->sector_count,
689 __map_bio(ti, clone, tio);
690 ci->sector_count = 0;
692 } else if (to_sector(bio->bi_io_vec[ci->idx].bv_len) <= max) {
694 * There are some bvecs that don't span targets.
695 * Do as many of these as possible.
698 sector_t remaining = max;
701 for (i = ci->idx; remaining && (i < bio->bi_vcnt); i++) {
702 bv_len = to_sector(bio->bi_io_vec[i].bv_len);
704 if (bv_len > remaining)
711 clone = clone_bio(bio, ci->sector, ci->idx, i - ci->idx, len,
713 __map_bio(ti, clone, tio);
716 ci->sector_count -= len;
721 * Handle a bvec that must be split between two or more targets.
723 struct bio_vec *bv = bio->bi_io_vec + ci->idx;
724 sector_t remaining = to_sector(bv->bv_len);
725 unsigned int offset = 0;
729 ti = dm_table_find_target(ci->map, ci->sector);
730 max = max_io_len(ci->md, ci->sector, ti);
732 tio = alloc_tio(ci->md);
735 memset(&tio->info, 0, sizeof(tio->info));
738 len = min(remaining, max);
740 clone = split_bvec(bio, ci->sector, ci->idx,
741 bv->bv_offset + offset, len,
744 __map_bio(ti, clone, tio);
747 ci->sector_count -= len;
748 offset += to_bytes(len);
749 } while (remaining -= len);
756 * Split the bio into several clones.
758 static void __split_bio(struct mapped_device *md, struct bio *bio)
760 struct clone_info ci;
762 ci.map = dm_get_table(md);
764 bio_io_error(bio, bio->bi_size);
770 ci.io = alloc_io(md);
772 atomic_set(&ci.io->io_count, 1);
775 ci.sector = bio->bi_sector;
776 ci.sector_count = bio_sectors(bio);
777 ci.idx = bio->bi_idx;
779 start_io_acct(ci.io);
780 while (ci.sector_count)
781 __clone_and_map(&ci);
783 /* drop the extra reference count */
784 dec_pending(ci.io, 0);
785 dm_table_put(ci.map);
787 /*-----------------------------------------------------------------
789 *---------------------------------------------------------------*/
792 * The request function that just remaps the bio built up by
795 static int dm_request(struct request_queue *q, struct bio *bio)
798 int rw = bio_data_dir(bio);
799 struct mapped_device *md = q->queuedata;
802 * There is no use in forwarding any barrier request since we can't
803 * guarantee it is (or can be) handled by the targets correctly.
805 if (unlikely(bio_barrier(bio))) {
806 bio_endio(bio, bio->bi_size, -EOPNOTSUPP);
810 down_read(&md->io_lock);
812 disk_stat_inc(dm_disk(md), ios[rw]);
813 disk_stat_add(dm_disk(md), sectors[rw], bio_sectors(bio));
816 * If we're suspended we have to queue
819 while (test_bit(DMF_BLOCK_IO, &md->flags)) {
820 up_read(&md->io_lock);
822 if (bio_rw(bio) == READA) {
823 bio_io_error(bio, bio->bi_size);
827 r = queue_io(md, bio);
829 bio_io_error(bio, bio->bi_size);
833 return 0; /* deferred successfully */
836 * We're in a while loop, because someone could suspend
837 * before we get to the following read lock.
839 down_read(&md->io_lock);
842 __split_bio(md, bio);
843 up_read(&md->io_lock);
847 static int dm_flush_all(struct request_queue *q, struct gendisk *disk,
848 sector_t *error_sector)
850 struct mapped_device *md = q->queuedata;
851 struct dm_table *map = dm_get_table(md);
855 ret = dm_table_flush_all(map);
862 static void dm_unplug_all(struct request_queue *q)
864 struct mapped_device *md = q->queuedata;
865 struct dm_table *map = dm_get_table(md);
868 dm_table_unplug_all(map);
873 static int dm_any_congested(void *congested_data, int bdi_bits)
876 struct mapped_device *md = (struct mapped_device *) congested_data;
877 struct dm_table *map = dm_get_table(md);
879 if (!map || test_bit(DMF_BLOCK_IO, &md->flags))
882 r = dm_table_any_congested(map, bdi_bits);
888 /*-----------------------------------------------------------------
889 * An IDR is used to keep track of allocated minor numbers.
890 *---------------------------------------------------------------*/
891 static DEFINE_IDR(_minor_idr);
893 static void free_minor(int minor)
895 spin_lock(&_minor_lock);
896 idr_remove(&_minor_idr, minor);
897 spin_unlock(&_minor_lock);
901 * See if the device with a specific minor # is free.
903 static int specific_minor(struct mapped_device *md, int minor)
907 if (minor >= (1 << MINORBITS))
910 r = idr_pre_get(&_minor_idr, GFP_KERNEL);
914 spin_lock(&_minor_lock);
916 if (idr_find(&_minor_idr, minor)) {
921 r = idr_get_new_above(&_minor_idr, MINOR_ALLOCED, minor, &m);
926 idr_remove(&_minor_idr, m);
932 spin_unlock(&_minor_lock);
936 static int next_free_minor(struct mapped_device *md, int *minor)
940 r = idr_pre_get(&_minor_idr, GFP_KERNEL);
944 spin_lock(&_minor_lock);
946 r = idr_get_new(&_minor_idr, MINOR_ALLOCED, &m);
951 if (m >= (1 << MINORBITS)) {
952 idr_remove(&_minor_idr, m);
960 spin_unlock(&_minor_lock);
964 static struct block_device_operations dm_blk_dops;
967 * Allocate and initialise a blank device with a given minor.
969 static struct mapped_device *alloc_dev(int minor)
972 struct mapped_device *md = kmalloc(sizeof(*md), GFP_KERNEL);
976 DMWARN("unable to allocate device, out of memory.");
980 if (!try_module_get(THIS_MODULE))
983 /* get a minor number for the dev */
984 if (minor == DM_ANY_MINOR)
985 r = next_free_minor(md, &minor);
987 r = specific_minor(md, minor);
991 memset(md, 0, sizeof(*md));
992 init_rwsem(&md->io_lock);
993 init_MUTEX(&md->suspend_lock);
994 spin_lock_init(&md->pushback_lock);
995 rwlock_init(&md->map_lock);
996 atomic_set(&md->holders, 1);
997 atomic_set(&md->open_count, 0);
998 atomic_set(&md->event_nr, 0);
1000 md->queue = blk_alloc_queue(GFP_KERNEL);
1002 goto bad1_free_minor;
1004 md->queue->queuedata = md;
1005 md->queue->backing_dev_info.congested_fn = dm_any_congested;
1006 md->queue->backing_dev_info.congested_data = md;
1007 blk_queue_make_request(md->queue, dm_request);
1008 blk_queue_bounce_limit(md->queue, BLK_BOUNCE_ANY);
1009 md->queue->unplug_fn = dm_unplug_all;
1010 md->queue->issue_flush_fn = dm_flush_all;
1012 md->io_pool = mempool_create_slab_pool(MIN_IOS, _io_cache);
1016 md->tio_pool = mempool_create_slab_pool(MIN_IOS, _tio_cache);
1020 md->bs = bioset_create(16, 16);
1024 md->disk = alloc_disk(1);
1028 atomic_set(&md->pending, 0);
1029 init_waitqueue_head(&md->wait);
1030 init_waitqueue_head(&md->eventq);
1032 md->disk->major = _major;
1033 md->disk->first_minor = minor;
1034 md->disk->fops = &dm_blk_dops;
1035 md->disk->queue = md->queue;
1036 md->disk->private_data = md;
1037 sprintf(md->disk->disk_name, "dm-%d", minor);
1039 format_dev_t(md->name, MKDEV(_major, minor));
1041 /* Populate the mapping, nobody knows we exist yet */
1042 spin_lock(&_minor_lock);
1043 old_md = idr_replace(&_minor_idr, md, minor);
1044 spin_unlock(&_minor_lock);
1046 BUG_ON(old_md != MINOR_ALLOCED);
1051 bioset_free(md->bs);
1053 mempool_destroy(md->tio_pool);
1055 mempool_destroy(md->io_pool);
1057 blk_cleanup_queue(md->queue);
1061 module_put(THIS_MODULE);
1067 static void free_dev(struct mapped_device *md)
1069 int minor = md->disk->first_minor;
1071 if (md->suspended_bdev) {
1072 thaw_bdev(md->suspended_bdev, NULL);
1073 bdput(md->suspended_bdev);
1075 mempool_destroy(md->tio_pool);
1076 mempool_destroy(md->io_pool);
1077 bioset_free(md->bs);
1078 del_gendisk(md->disk);
1081 spin_lock(&_minor_lock);
1082 md->disk->private_data = NULL;
1083 spin_unlock(&_minor_lock);
1086 blk_cleanup_queue(md->queue);
1087 module_put(THIS_MODULE);
1092 * Bind a table to the device.
1094 static void event_callback(void *context)
1096 struct mapped_device *md = (struct mapped_device *) context;
1098 atomic_inc(&md->event_nr);
1099 wake_up(&md->eventq);
1102 static void __set_size(struct mapped_device *md, sector_t size)
1104 set_capacity(md->disk, size);
1106 mutex_lock(&md->suspended_bdev->bd_inode->i_mutex);
1107 i_size_write(md->suspended_bdev->bd_inode, (loff_t)size << SECTOR_SHIFT);
1108 mutex_unlock(&md->suspended_bdev->bd_inode->i_mutex);
1111 static int __bind(struct mapped_device *md, struct dm_table *t)
1113 struct request_queue *q = md->queue;
1116 size = dm_table_get_size(t);
1119 * Wipe any geometry if the size of the table changed.
1121 if (size != get_capacity(md->disk))
1122 memset(&md->geometry, 0, sizeof(md->geometry));
1124 if (md->suspended_bdev)
1125 __set_size(md, size);
1130 dm_table_event_callback(t, event_callback, md);
1132 write_lock(&md->map_lock);
1134 dm_table_set_restrictions(t, q);
1135 write_unlock(&md->map_lock);
1140 static void __unbind(struct mapped_device *md)
1142 struct dm_table *map = md->map;
1147 dm_table_event_callback(map, NULL, NULL);
1148 write_lock(&md->map_lock);
1150 write_unlock(&md->map_lock);
1155 * Constructor for a new device.
1157 int dm_create(int minor, struct mapped_device **result)
1159 struct mapped_device *md;
1161 md = alloc_dev(minor);
1169 static struct mapped_device *dm_find_md(dev_t dev)
1171 struct mapped_device *md;
1172 unsigned minor = MINOR(dev);
1174 if (MAJOR(dev) != _major || minor >= (1 << MINORBITS))
1177 spin_lock(&_minor_lock);
1179 md = idr_find(&_minor_idr, minor);
1180 if (md && (md == MINOR_ALLOCED ||
1181 (dm_disk(md)->first_minor != minor) ||
1182 test_bit(DMF_FREEING, &md->flags))) {
1188 spin_unlock(&_minor_lock);
1193 struct mapped_device *dm_get_md(dev_t dev)
1195 struct mapped_device *md = dm_find_md(dev);
1203 void *dm_get_mdptr(struct mapped_device *md)
1205 return md->interface_ptr;
1208 void dm_set_mdptr(struct mapped_device *md, void *ptr)
1210 md->interface_ptr = ptr;
1213 void dm_get(struct mapped_device *md)
1215 atomic_inc(&md->holders);
1218 const char *dm_device_name(struct mapped_device *md)
1222 EXPORT_SYMBOL_GPL(dm_device_name);
1224 void dm_put(struct mapped_device *md)
1226 struct dm_table *map;
1228 BUG_ON(test_bit(DMF_FREEING, &md->flags));
1230 if (atomic_dec_and_lock(&md->holders, &_minor_lock)) {
1231 map = dm_get_table(md);
1232 idr_replace(&_minor_idr, MINOR_ALLOCED, dm_disk(md)->first_minor);
1233 set_bit(DMF_FREEING, &md->flags);
1234 spin_unlock(&_minor_lock);
1235 if (!dm_suspended(md)) {
1236 dm_table_presuspend_targets(map);
1237 dm_table_postsuspend_targets(map);
1244 EXPORT_SYMBOL_GPL(dm_put);
1247 * Process the deferred bios
1249 static void __flush_deferred_io(struct mapped_device *md, struct bio *c)
1262 * Swap in a new table (destroying old one).
1264 int dm_swap_table(struct mapped_device *md, struct dm_table *table)
1268 down(&md->suspend_lock);
1270 /* device must be suspended */
1271 if (!dm_suspended(md))
1274 /* without bdev, the device size cannot be changed */
1275 if (!md->suspended_bdev)
1276 if (get_capacity(md->disk) != dm_table_get_size(table))
1280 r = __bind(md, table);
1283 up(&md->suspend_lock);
1288 * Functions to lock and unlock any filesystem running on the
1291 static int lock_fs(struct mapped_device *md)
1295 WARN_ON(md->frozen_sb);
1297 md->frozen_sb = freeze_bdev(md->suspended_bdev);
1298 if (IS_ERR(md->frozen_sb)) {
1299 r = PTR_ERR(md->frozen_sb);
1300 md->frozen_sb = NULL;
1304 set_bit(DMF_FROZEN, &md->flags);
1306 /* don't bdput right now, we don't want the bdev
1307 * to go away while it is locked.
1312 static void unlock_fs(struct mapped_device *md)
1314 if (!test_bit(DMF_FROZEN, &md->flags))
1317 thaw_bdev(md->suspended_bdev, md->frozen_sb);
1318 md->frozen_sb = NULL;
1319 clear_bit(DMF_FROZEN, &md->flags);
1323 * We need to be able to change a mapping table under a mounted
1324 * filesystem. For example we might want to move some data in
1325 * the background. Before the table can be swapped with
1326 * dm_bind_table, dm_suspend must be called to flush any in
1327 * flight bios and ensure that any further io gets deferred.
1329 int dm_suspend(struct mapped_device *md, unsigned suspend_flags)
1331 struct dm_table *map = NULL;
1332 unsigned long flags;
1333 DECLARE_WAITQUEUE(wait, current);
1336 int do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG ? 1 : 0;
1337 int noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG ? 1 : 0;
1339 down(&md->suspend_lock);
1341 if (dm_suspended(md))
1344 map = dm_get_table(md);
1347 * DMF_NOFLUSH_SUSPENDING must be set before presuspend.
1348 * This flag is cleared before dm_suspend returns.
1351 set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
1353 /* This does not get reverted if there's an error later. */
1354 dm_table_presuspend_targets(map);
1356 /* bdget() can stall if the pending I/Os are not flushed */
1358 md->suspended_bdev = bdget_disk(md->disk, 0);
1359 if (!md->suspended_bdev) {
1360 DMWARN("bdget failed in dm_suspend");
1367 * Flush I/O to the device.
1368 * noflush supersedes do_lockfs, because lock_fs() needs to flush I/Os.
1370 if (do_lockfs && !noflush) {
1377 * First we set the BLOCK_IO flag so no more ios will be mapped.
1379 down_write(&md->io_lock);
1380 set_bit(DMF_BLOCK_IO, &md->flags);
1382 add_wait_queue(&md->wait, &wait);
1383 up_write(&md->io_lock);
1387 dm_table_unplug_all(map);
1390 * Then we wait for the already mapped ios to
1394 set_current_state(TASK_INTERRUPTIBLE);
1396 if (!atomic_read(&md->pending) || signal_pending(current))
1401 set_current_state(TASK_RUNNING);
1403 down_write(&md->io_lock);
1404 remove_wait_queue(&md->wait, &wait);
1407 spin_lock_irqsave(&md->pushback_lock, flags);
1408 clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
1409 bio_list_merge_head(&md->deferred, &md->pushback);
1410 bio_list_init(&md->pushback);
1411 spin_unlock_irqrestore(&md->pushback_lock, flags);
1414 /* were we interrupted ? */
1416 if (atomic_read(&md->pending)) {
1417 clear_bit(DMF_BLOCK_IO, &md->flags);
1418 def = bio_list_get(&md->deferred);
1419 __flush_deferred_io(md, def);
1420 up_write(&md->io_lock);
1422 goto out; /* pushback list is already flushed, so skip flush */
1424 up_write(&md->io_lock);
1426 dm_table_postsuspend_targets(map);
1428 set_bit(DMF_SUSPENDED, &md->flags);
1435 * Because there may be already I/Os in the pushback list,
1436 * flush them before return.
1438 down_write(&md->io_lock);
1440 spin_lock_irqsave(&md->pushback_lock, flags);
1441 clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
1442 bio_list_merge_head(&md->deferred, &md->pushback);
1443 bio_list_init(&md->pushback);
1444 spin_unlock_irqrestore(&md->pushback_lock, flags);
1446 def = bio_list_get(&md->deferred);
1447 __flush_deferred_io(md, def);
1448 up_write(&md->io_lock);
1452 if (r && md->suspended_bdev) {
1453 bdput(md->suspended_bdev);
1454 md->suspended_bdev = NULL;
1460 up(&md->suspend_lock);
1464 int dm_resume(struct mapped_device *md)
1468 struct dm_table *map = NULL;
1470 down(&md->suspend_lock);
1471 if (!dm_suspended(md))
1474 map = dm_get_table(md);
1475 if (!map || !dm_table_get_size(map))
1478 r = dm_table_resume_targets(map);
1482 down_write(&md->io_lock);
1483 clear_bit(DMF_BLOCK_IO, &md->flags);
1485 def = bio_list_get(&md->deferred);
1486 __flush_deferred_io(md, def);
1487 up_write(&md->io_lock);
1491 if (md->suspended_bdev) {
1492 bdput(md->suspended_bdev);
1493 md->suspended_bdev = NULL;
1496 clear_bit(DMF_SUSPENDED, &md->flags);
1498 dm_table_unplug_all(map);
1500 kobject_uevent(&md->disk->kobj, KOBJ_CHANGE);
1506 up(&md->suspend_lock);
1511 /*-----------------------------------------------------------------
1512 * Event notification.
1513 *---------------------------------------------------------------*/
1514 uint32_t dm_get_event_nr(struct mapped_device *md)
1516 return atomic_read(&md->event_nr);
1519 int dm_wait_event(struct mapped_device *md, int event_nr)
1521 return wait_event_interruptible(md->eventq,
1522 (event_nr != atomic_read(&md->event_nr)));
1526 * The gendisk is only valid as long as you have a reference
1529 struct gendisk *dm_disk(struct mapped_device *md)
1534 int dm_suspended(struct mapped_device *md)
1536 return test_bit(DMF_SUSPENDED, &md->flags);
1539 int dm_noflush_suspending(struct dm_target *ti)
1541 struct mapped_device *md = dm_table_get_md(ti->table);
1542 int r = __noflush_suspending(md);
1548 EXPORT_SYMBOL_GPL(dm_noflush_suspending);
1550 static struct block_device_operations dm_blk_dops = {
1551 .open = dm_blk_open,
1552 .release = dm_blk_close,
1553 .ioctl = dm_blk_ioctl,
1554 .getgeo = dm_blk_getgeo,
1555 .owner = THIS_MODULE
1558 EXPORT_SYMBOL(dm_get_mapinfo);
1563 module_init(dm_init);
1564 module_exit(dm_exit);
1566 module_param(major, uint, 0);
1567 MODULE_PARM_DESC(major, "The major number of the device mapper");
1568 MODULE_DESCRIPTION(DM_NAME " driver");
1569 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
1570 MODULE_LICENSE("GPL");