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->error);
494 static void clone_endio(struct bio *bio, 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;
501 if (!bio_flagged(bio, BIO_UPTODATE) && !error)
505 r = endio(tio->ti, bio, error, &tio->info);
506 if (r < 0 || r == DM_ENDIO_REQUEUE)
508 * error and requeue request are handled
512 else if (r == DM_ENDIO_INCOMPLETE)
513 /* The target will handle the io */
516 DMWARN("unimplemented target endio return value: %d", r);
521 dec_pending(tio->io, error);
524 * Store md for cleanup instead of tio which is about to get freed.
526 bio->bi_private = md->bs;
532 static sector_t max_io_len(struct mapped_device *md,
533 sector_t sector, struct dm_target *ti)
535 sector_t offset = sector - ti->begin;
536 sector_t len = ti->len - offset;
539 * Does the target need to split even further ?
543 boundary = ((offset + ti->split_io) & ~(ti->split_io - 1))
552 static void __map_bio(struct dm_target *ti, struct bio *clone,
553 struct dm_target_io *tio)
557 struct mapped_device *md;
562 BUG_ON(!clone->bi_size);
564 clone->bi_end_io = clone_endio;
565 clone->bi_private = tio;
568 * Map the clone. If r == 0 we don't need to do
569 * anything, the target has assumed ownership of
572 atomic_inc(&tio->io->io_count);
573 sector = clone->bi_sector;
574 r = ti->type->map(ti, clone, &tio->info);
575 if (r == DM_MAPIO_REMAPPED) {
576 /* the bio has been remapped so dispatch it */
578 blk_add_trace_remap(bdev_get_queue(clone->bi_bdev), clone,
579 tio->io->bio->bi_bdev->bd_dev,
580 clone->bi_sector, sector);
582 generic_make_request(clone);
583 } else if (r < 0 || r == DM_MAPIO_REQUEUE) {
584 /* error the io and bail out, or requeue it if needed */
586 dec_pending(tio->io, r);
588 * Store bio_set for cleanup.
590 clone->bi_private = md->bs;
594 DMWARN("unimplemented target map return value: %d", r);
600 struct mapped_device *md;
601 struct dm_table *map;
605 sector_t sector_count;
609 static void dm_bio_destructor(struct bio *bio)
611 struct bio_set *bs = bio->bi_private;
617 * Creates a little bio that is just does part of a bvec.
619 static struct bio *split_bvec(struct bio *bio, sector_t sector,
620 unsigned short idx, unsigned int offset,
621 unsigned int len, struct bio_set *bs)
624 struct bio_vec *bv = bio->bi_io_vec + idx;
626 clone = bio_alloc_bioset(GFP_NOIO, 1, bs);
627 clone->bi_destructor = dm_bio_destructor;
628 *clone->bi_io_vec = *bv;
630 clone->bi_sector = sector;
631 clone->bi_bdev = bio->bi_bdev;
632 clone->bi_rw = bio->bi_rw;
634 clone->bi_size = to_bytes(len);
635 clone->bi_io_vec->bv_offset = offset;
636 clone->bi_io_vec->bv_len = clone->bi_size;
642 * Creates a bio that consists of range of complete bvecs.
644 static struct bio *clone_bio(struct bio *bio, sector_t sector,
645 unsigned short idx, unsigned short bv_count,
646 unsigned int len, struct bio_set *bs)
650 clone = bio_alloc_bioset(GFP_NOIO, bio->bi_max_vecs, bs);
651 __bio_clone(clone, bio);
652 clone->bi_destructor = dm_bio_destructor;
653 clone->bi_sector = sector;
655 clone->bi_vcnt = idx + bv_count;
656 clone->bi_size = to_bytes(len);
657 clone->bi_flags &= ~(1 << BIO_SEG_VALID);
662 static void __clone_and_map(struct clone_info *ci)
664 struct bio *clone, *bio = ci->bio;
665 struct dm_target *ti = dm_table_find_target(ci->map, ci->sector);
666 sector_t len = 0, max = max_io_len(ci->md, ci->sector, ti);
667 struct dm_target_io *tio;
670 * Allocate a target io object.
672 tio = alloc_tio(ci->md);
675 memset(&tio->info, 0, sizeof(tio->info));
677 if (ci->sector_count <= max) {
679 * Optimise for the simple case where we can do all of
680 * the remaining io with a single clone.
682 clone = clone_bio(bio, ci->sector, ci->idx,
683 bio->bi_vcnt - ci->idx, ci->sector_count,
685 __map_bio(ti, clone, tio);
686 ci->sector_count = 0;
688 } else if (to_sector(bio->bi_io_vec[ci->idx].bv_len) <= max) {
690 * There are some bvecs that don't span targets.
691 * Do as many of these as possible.
694 sector_t remaining = max;
697 for (i = ci->idx; remaining && (i < bio->bi_vcnt); i++) {
698 bv_len = to_sector(bio->bi_io_vec[i].bv_len);
700 if (bv_len > remaining)
707 clone = clone_bio(bio, ci->sector, ci->idx, i - ci->idx, len,
709 __map_bio(ti, clone, tio);
712 ci->sector_count -= len;
717 * Handle a bvec that must be split between two or more targets.
719 struct bio_vec *bv = bio->bi_io_vec + ci->idx;
720 sector_t remaining = to_sector(bv->bv_len);
721 unsigned int offset = 0;
725 ti = dm_table_find_target(ci->map, ci->sector);
726 max = max_io_len(ci->md, ci->sector, ti);
728 tio = alloc_tio(ci->md);
731 memset(&tio->info, 0, sizeof(tio->info));
734 len = min(remaining, max);
736 clone = split_bvec(bio, ci->sector, ci->idx,
737 bv->bv_offset + offset, len,
740 __map_bio(ti, clone, tio);
743 ci->sector_count -= len;
744 offset += to_bytes(len);
745 } while (remaining -= len);
752 * Split the bio into several clones.
754 static void __split_bio(struct mapped_device *md, struct bio *bio)
756 struct clone_info ci;
758 ci.map = dm_get_table(md);
766 ci.io = alloc_io(md);
768 atomic_set(&ci.io->io_count, 1);
771 ci.sector = bio->bi_sector;
772 ci.sector_count = bio_sectors(bio);
773 ci.idx = bio->bi_idx;
775 start_io_acct(ci.io);
776 while (ci.sector_count)
777 __clone_and_map(&ci);
779 /* drop the extra reference count */
780 dec_pending(ci.io, 0);
781 dm_table_put(ci.map);
783 /*-----------------------------------------------------------------
785 *---------------------------------------------------------------*/
788 * The request function that just remaps the bio built up by
791 static int dm_request(struct request_queue *q, struct bio *bio)
794 int rw = bio_data_dir(bio);
795 struct mapped_device *md = q->queuedata;
798 * There is no use in forwarding any barrier request since we can't
799 * guarantee it is (or can be) handled by the targets correctly.
801 if (unlikely(bio_barrier(bio))) {
802 bio_endio(bio, -EOPNOTSUPP);
806 down_read(&md->io_lock);
808 disk_stat_inc(dm_disk(md), ios[rw]);
809 disk_stat_add(dm_disk(md), sectors[rw], bio_sectors(bio));
812 * If we're suspended we have to queue
815 while (test_bit(DMF_BLOCK_IO, &md->flags)) {
816 up_read(&md->io_lock);
818 if (bio_rw(bio) == READA) {
823 r = queue_io(md, bio);
829 return 0; /* deferred successfully */
832 * We're in a while loop, because someone could suspend
833 * before we get to the following read lock.
835 down_read(&md->io_lock);
838 __split_bio(md, bio);
839 up_read(&md->io_lock);
843 static int dm_flush_all(struct request_queue *q, struct gendisk *disk,
844 sector_t *error_sector)
846 struct mapped_device *md = q->queuedata;
847 struct dm_table *map = dm_get_table(md);
851 ret = dm_table_flush_all(map);
858 static void dm_unplug_all(struct request_queue *q)
860 struct mapped_device *md = q->queuedata;
861 struct dm_table *map = dm_get_table(md);
864 dm_table_unplug_all(map);
869 static int dm_any_congested(void *congested_data, int bdi_bits)
872 struct mapped_device *md = (struct mapped_device *) congested_data;
873 struct dm_table *map = dm_get_table(md);
875 if (!map || test_bit(DMF_BLOCK_IO, &md->flags))
878 r = dm_table_any_congested(map, bdi_bits);
884 /*-----------------------------------------------------------------
885 * An IDR is used to keep track of allocated minor numbers.
886 *---------------------------------------------------------------*/
887 static DEFINE_IDR(_minor_idr);
889 static void free_minor(int minor)
891 spin_lock(&_minor_lock);
892 idr_remove(&_minor_idr, minor);
893 spin_unlock(&_minor_lock);
897 * See if the device with a specific minor # is free.
899 static int specific_minor(struct mapped_device *md, int minor)
903 if (minor >= (1 << MINORBITS))
906 r = idr_pre_get(&_minor_idr, GFP_KERNEL);
910 spin_lock(&_minor_lock);
912 if (idr_find(&_minor_idr, minor)) {
917 r = idr_get_new_above(&_minor_idr, MINOR_ALLOCED, minor, &m);
922 idr_remove(&_minor_idr, m);
928 spin_unlock(&_minor_lock);
932 static int next_free_minor(struct mapped_device *md, int *minor)
936 r = idr_pre_get(&_minor_idr, GFP_KERNEL);
940 spin_lock(&_minor_lock);
942 r = idr_get_new(&_minor_idr, MINOR_ALLOCED, &m);
947 if (m >= (1 << MINORBITS)) {
948 idr_remove(&_minor_idr, m);
956 spin_unlock(&_minor_lock);
960 static struct block_device_operations dm_blk_dops;
963 * Allocate and initialise a blank device with a given minor.
965 static struct mapped_device *alloc_dev(int minor)
968 struct mapped_device *md = kmalloc(sizeof(*md), GFP_KERNEL);
972 DMWARN("unable to allocate device, out of memory.");
976 if (!try_module_get(THIS_MODULE))
979 /* get a minor number for the dev */
980 if (minor == DM_ANY_MINOR)
981 r = next_free_minor(md, &minor);
983 r = specific_minor(md, minor);
987 memset(md, 0, sizeof(*md));
988 init_rwsem(&md->io_lock);
989 init_MUTEX(&md->suspend_lock);
990 spin_lock_init(&md->pushback_lock);
991 rwlock_init(&md->map_lock);
992 atomic_set(&md->holders, 1);
993 atomic_set(&md->open_count, 0);
994 atomic_set(&md->event_nr, 0);
996 md->queue = blk_alloc_queue(GFP_KERNEL);
998 goto bad1_free_minor;
1000 md->queue->queuedata = md;
1001 md->queue->backing_dev_info.congested_fn = dm_any_congested;
1002 md->queue->backing_dev_info.congested_data = md;
1003 blk_queue_make_request(md->queue, dm_request);
1004 blk_queue_bounce_limit(md->queue, BLK_BOUNCE_ANY);
1005 md->queue->unplug_fn = dm_unplug_all;
1006 md->queue->issue_flush_fn = dm_flush_all;
1008 md->io_pool = mempool_create_slab_pool(MIN_IOS, _io_cache);
1012 md->tio_pool = mempool_create_slab_pool(MIN_IOS, _tio_cache);
1016 md->bs = bioset_create(16, 16);
1020 md->disk = alloc_disk(1);
1024 atomic_set(&md->pending, 0);
1025 init_waitqueue_head(&md->wait);
1026 init_waitqueue_head(&md->eventq);
1028 md->disk->major = _major;
1029 md->disk->first_minor = minor;
1030 md->disk->fops = &dm_blk_dops;
1031 md->disk->queue = md->queue;
1032 md->disk->private_data = md;
1033 sprintf(md->disk->disk_name, "dm-%d", minor);
1035 format_dev_t(md->name, MKDEV(_major, minor));
1037 /* Populate the mapping, nobody knows we exist yet */
1038 spin_lock(&_minor_lock);
1039 old_md = idr_replace(&_minor_idr, md, minor);
1040 spin_unlock(&_minor_lock);
1042 BUG_ON(old_md != MINOR_ALLOCED);
1047 bioset_free(md->bs);
1049 mempool_destroy(md->tio_pool);
1051 mempool_destroy(md->io_pool);
1053 blk_cleanup_queue(md->queue);
1057 module_put(THIS_MODULE);
1063 static void free_dev(struct mapped_device *md)
1065 int minor = md->disk->first_minor;
1067 if (md->suspended_bdev) {
1068 thaw_bdev(md->suspended_bdev, NULL);
1069 bdput(md->suspended_bdev);
1071 mempool_destroy(md->tio_pool);
1072 mempool_destroy(md->io_pool);
1073 bioset_free(md->bs);
1074 del_gendisk(md->disk);
1077 spin_lock(&_minor_lock);
1078 md->disk->private_data = NULL;
1079 spin_unlock(&_minor_lock);
1082 blk_cleanup_queue(md->queue);
1083 module_put(THIS_MODULE);
1088 * Bind a table to the device.
1090 static void event_callback(void *context)
1092 struct mapped_device *md = (struct mapped_device *) context;
1094 atomic_inc(&md->event_nr);
1095 wake_up(&md->eventq);
1098 static void __set_size(struct mapped_device *md, sector_t size)
1100 set_capacity(md->disk, size);
1102 mutex_lock(&md->suspended_bdev->bd_inode->i_mutex);
1103 i_size_write(md->suspended_bdev->bd_inode, (loff_t)size << SECTOR_SHIFT);
1104 mutex_unlock(&md->suspended_bdev->bd_inode->i_mutex);
1107 static int __bind(struct mapped_device *md, struct dm_table *t)
1109 struct request_queue *q = md->queue;
1112 size = dm_table_get_size(t);
1115 * Wipe any geometry if the size of the table changed.
1117 if (size != get_capacity(md->disk))
1118 memset(&md->geometry, 0, sizeof(md->geometry));
1120 if (md->suspended_bdev)
1121 __set_size(md, size);
1126 dm_table_event_callback(t, event_callback, md);
1128 write_lock(&md->map_lock);
1130 dm_table_set_restrictions(t, q);
1131 write_unlock(&md->map_lock);
1136 static void __unbind(struct mapped_device *md)
1138 struct dm_table *map = md->map;
1143 dm_table_event_callback(map, NULL, NULL);
1144 write_lock(&md->map_lock);
1146 write_unlock(&md->map_lock);
1151 * Constructor for a new device.
1153 int dm_create(int minor, struct mapped_device **result)
1155 struct mapped_device *md;
1157 md = alloc_dev(minor);
1165 static struct mapped_device *dm_find_md(dev_t dev)
1167 struct mapped_device *md;
1168 unsigned minor = MINOR(dev);
1170 if (MAJOR(dev) != _major || minor >= (1 << MINORBITS))
1173 spin_lock(&_minor_lock);
1175 md = idr_find(&_minor_idr, minor);
1176 if (md && (md == MINOR_ALLOCED ||
1177 (dm_disk(md)->first_minor != minor) ||
1178 test_bit(DMF_FREEING, &md->flags))) {
1184 spin_unlock(&_minor_lock);
1189 struct mapped_device *dm_get_md(dev_t dev)
1191 struct mapped_device *md = dm_find_md(dev);
1199 void *dm_get_mdptr(struct mapped_device *md)
1201 return md->interface_ptr;
1204 void dm_set_mdptr(struct mapped_device *md, void *ptr)
1206 md->interface_ptr = ptr;
1209 void dm_get(struct mapped_device *md)
1211 atomic_inc(&md->holders);
1214 const char *dm_device_name(struct mapped_device *md)
1218 EXPORT_SYMBOL_GPL(dm_device_name);
1220 void dm_put(struct mapped_device *md)
1222 struct dm_table *map;
1224 BUG_ON(test_bit(DMF_FREEING, &md->flags));
1226 if (atomic_dec_and_lock(&md->holders, &_minor_lock)) {
1227 map = dm_get_table(md);
1228 idr_replace(&_minor_idr, MINOR_ALLOCED, dm_disk(md)->first_minor);
1229 set_bit(DMF_FREEING, &md->flags);
1230 spin_unlock(&_minor_lock);
1231 if (!dm_suspended(md)) {
1232 dm_table_presuspend_targets(map);
1233 dm_table_postsuspend_targets(map);
1240 EXPORT_SYMBOL_GPL(dm_put);
1243 * Process the deferred bios
1245 static void __flush_deferred_io(struct mapped_device *md, struct bio *c)
1258 * Swap in a new table (destroying old one).
1260 int dm_swap_table(struct mapped_device *md, struct dm_table *table)
1264 down(&md->suspend_lock);
1266 /* device must be suspended */
1267 if (!dm_suspended(md))
1270 /* without bdev, the device size cannot be changed */
1271 if (!md->suspended_bdev)
1272 if (get_capacity(md->disk) != dm_table_get_size(table))
1276 r = __bind(md, table);
1279 up(&md->suspend_lock);
1284 * Functions to lock and unlock any filesystem running on the
1287 static int lock_fs(struct mapped_device *md)
1291 WARN_ON(md->frozen_sb);
1293 md->frozen_sb = freeze_bdev(md->suspended_bdev);
1294 if (IS_ERR(md->frozen_sb)) {
1295 r = PTR_ERR(md->frozen_sb);
1296 md->frozen_sb = NULL;
1300 set_bit(DMF_FROZEN, &md->flags);
1302 /* don't bdput right now, we don't want the bdev
1303 * to go away while it is locked.
1308 static void unlock_fs(struct mapped_device *md)
1310 if (!test_bit(DMF_FROZEN, &md->flags))
1313 thaw_bdev(md->suspended_bdev, md->frozen_sb);
1314 md->frozen_sb = NULL;
1315 clear_bit(DMF_FROZEN, &md->flags);
1319 * We need to be able to change a mapping table under a mounted
1320 * filesystem. For example we might want to move some data in
1321 * the background. Before the table can be swapped with
1322 * dm_bind_table, dm_suspend must be called to flush any in
1323 * flight bios and ensure that any further io gets deferred.
1325 int dm_suspend(struct mapped_device *md, unsigned suspend_flags)
1327 struct dm_table *map = NULL;
1328 unsigned long flags;
1329 DECLARE_WAITQUEUE(wait, current);
1332 int do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG ? 1 : 0;
1333 int noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG ? 1 : 0;
1335 down(&md->suspend_lock);
1337 if (dm_suspended(md))
1340 map = dm_get_table(md);
1343 * DMF_NOFLUSH_SUSPENDING must be set before presuspend.
1344 * This flag is cleared before dm_suspend returns.
1347 set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
1349 /* This does not get reverted if there's an error later. */
1350 dm_table_presuspend_targets(map);
1352 /* bdget() can stall if the pending I/Os are not flushed */
1354 md->suspended_bdev = bdget_disk(md->disk, 0);
1355 if (!md->suspended_bdev) {
1356 DMWARN("bdget failed in dm_suspend");
1363 * Flush I/O to the device.
1364 * noflush supersedes do_lockfs, because lock_fs() needs to flush I/Os.
1366 if (do_lockfs && !noflush) {
1373 * First we set the BLOCK_IO flag so no more ios will be mapped.
1375 down_write(&md->io_lock);
1376 set_bit(DMF_BLOCK_IO, &md->flags);
1378 add_wait_queue(&md->wait, &wait);
1379 up_write(&md->io_lock);
1383 dm_table_unplug_all(map);
1386 * Then we wait for the already mapped ios to
1390 set_current_state(TASK_INTERRUPTIBLE);
1392 if (!atomic_read(&md->pending) || signal_pending(current))
1397 set_current_state(TASK_RUNNING);
1399 down_write(&md->io_lock);
1400 remove_wait_queue(&md->wait, &wait);
1403 spin_lock_irqsave(&md->pushback_lock, flags);
1404 clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
1405 bio_list_merge_head(&md->deferred, &md->pushback);
1406 bio_list_init(&md->pushback);
1407 spin_unlock_irqrestore(&md->pushback_lock, flags);
1410 /* were we interrupted ? */
1412 if (atomic_read(&md->pending)) {
1413 clear_bit(DMF_BLOCK_IO, &md->flags);
1414 def = bio_list_get(&md->deferred);
1415 __flush_deferred_io(md, def);
1416 up_write(&md->io_lock);
1418 goto out; /* pushback list is already flushed, so skip flush */
1420 up_write(&md->io_lock);
1422 dm_table_postsuspend_targets(map);
1424 set_bit(DMF_SUSPENDED, &md->flags);
1431 * Because there may be already I/Os in the pushback list,
1432 * flush them before return.
1434 down_write(&md->io_lock);
1436 spin_lock_irqsave(&md->pushback_lock, flags);
1437 clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
1438 bio_list_merge_head(&md->deferred, &md->pushback);
1439 bio_list_init(&md->pushback);
1440 spin_unlock_irqrestore(&md->pushback_lock, flags);
1442 def = bio_list_get(&md->deferred);
1443 __flush_deferred_io(md, def);
1444 up_write(&md->io_lock);
1448 if (r && md->suspended_bdev) {
1449 bdput(md->suspended_bdev);
1450 md->suspended_bdev = NULL;
1456 up(&md->suspend_lock);
1460 int dm_resume(struct mapped_device *md)
1464 struct dm_table *map = NULL;
1466 down(&md->suspend_lock);
1467 if (!dm_suspended(md))
1470 map = dm_get_table(md);
1471 if (!map || !dm_table_get_size(map))
1474 r = dm_table_resume_targets(map);
1478 down_write(&md->io_lock);
1479 clear_bit(DMF_BLOCK_IO, &md->flags);
1481 def = bio_list_get(&md->deferred);
1482 __flush_deferred_io(md, def);
1483 up_write(&md->io_lock);
1487 if (md->suspended_bdev) {
1488 bdput(md->suspended_bdev);
1489 md->suspended_bdev = NULL;
1492 clear_bit(DMF_SUSPENDED, &md->flags);
1494 dm_table_unplug_all(map);
1496 kobject_uevent(&md->disk->kobj, KOBJ_CHANGE);
1502 up(&md->suspend_lock);
1507 /*-----------------------------------------------------------------
1508 * Event notification.
1509 *---------------------------------------------------------------*/
1510 uint32_t dm_get_event_nr(struct mapped_device *md)
1512 return atomic_read(&md->event_nr);
1515 int dm_wait_event(struct mapped_device *md, int event_nr)
1517 return wait_event_interruptible(md->eventq,
1518 (event_nr != atomic_read(&md->event_nr)));
1522 * The gendisk is only valid as long as you have a reference
1525 struct gendisk *dm_disk(struct mapped_device *md)
1530 int dm_suspended(struct mapped_device *md)
1532 return test_bit(DMF_SUSPENDED, &md->flags);
1535 int dm_noflush_suspending(struct dm_target *ti)
1537 struct mapped_device *md = dm_table_get_md(ti->table);
1538 int r = __noflush_suspending(md);
1544 EXPORT_SYMBOL_GPL(dm_noflush_suspending);
1546 static struct block_device_operations dm_blk_dops = {
1547 .open = dm_blk_open,
1548 .release = dm_blk_close,
1549 .ioctl = dm_blk_ioctl,
1550 .getgeo = dm_blk_getgeo,
1551 .owner = THIS_MODULE
1554 EXPORT_SYMBOL(dm_get_mapinfo);
1559 module_init(dm_init);
1560 module_exit(dm_exit);
1562 module_param(major, uint, 0);
1563 MODULE_PARM_DESC(major, "The major number of the device mapper");
1564 MODULE_DESCRIPTION(DM_NAME " driver");
1565 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
1566 MODULE_LICENSE("GPL");