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"
10 #include "dm-uevent.h"
12 #include <linux/init.h>
13 #include <linux/module.h>
14 #include <linux/mutex.h>
15 #include <linux/moduleparam.h>
16 #include <linux/blkpg.h>
17 #include <linux/bio.h>
18 #include <linux/buffer_head.h>
19 #include <linux/mempool.h>
20 #include <linux/slab.h>
21 #include <linux/idr.h>
22 #include <linux/hdreg.h>
23 #include <linux/blktrace_api.h>
24 #include <linux/smp_lock.h>
26 #define DM_MSG_PREFIX "core"
28 static const char *_name = DM_NAME;
30 static unsigned int major = 0;
31 static unsigned int _major = 0;
33 static DEFINE_SPINLOCK(_minor_lock);
35 * One of these is allocated per bio.
38 struct mapped_device *md;
42 unsigned long start_time;
46 * One of these is allocated per target within a bio. Hopefully
47 * this will be simplified out one day.
55 union map_info *dm_get_mapinfo(struct bio *bio)
57 if (bio && bio->bi_private)
58 return &((struct dm_target_io *)bio->bi_private)->info;
62 #define MINOR_ALLOCED ((void *)-1)
65 * Bits for the md->flags field.
67 #define DMF_BLOCK_IO 0
68 #define DMF_SUSPENDED 1
71 #define DMF_DELETING 4
72 #define DMF_NOFLUSH_SUSPENDING 5
74 struct mapped_device {
75 struct rw_semaphore io_lock;
76 struct semaphore suspend_lock;
77 spinlock_t pushback_lock;
84 struct request_queue *queue;
91 * A list of ios that arrived while we were suspended.
94 wait_queue_head_t wait;
95 struct bio_list deferred;
96 struct bio_list pushback;
99 * The current mapping.
101 struct dm_table *map;
104 * io objects are allocated from here.
115 wait_queue_head_t eventq;
117 struct list_head uevent_list;
118 spinlock_t uevent_lock; /* Protect access to uevent_list */
121 * freeze/thaw support require holding onto a super block
123 struct super_block *frozen_sb;
124 struct block_device *suspended_bdev;
126 /* forced geometry settings */
127 struct hd_geometry geometry;
131 static struct kmem_cache *_io_cache;
132 static struct kmem_cache *_tio_cache;
134 static int __init local_init(void)
138 /* allocate a slab for the dm_ios */
139 _io_cache = KMEM_CACHE(dm_io, 0);
143 /* allocate a slab for the target ios */
144 _tio_cache = KMEM_CACHE(dm_target_io, 0);
146 kmem_cache_destroy(_io_cache);
150 r = dm_uevent_init();
152 kmem_cache_destroy(_tio_cache);
153 kmem_cache_destroy(_io_cache);
158 r = register_blkdev(_major, _name);
160 kmem_cache_destroy(_tio_cache);
161 kmem_cache_destroy(_io_cache);
172 static void local_exit(void)
174 kmem_cache_destroy(_tio_cache);
175 kmem_cache_destroy(_io_cache);
176 unregister_blkdev(_major, _name);
181 DMINFO("cleaned up");
184 int (*_inits[])(void) __initdata = {
192 void (*_exits[])(void) = {
200 static int __init dm_init(void)
202 const int count = ARRAY_SIZE(_inits);
206 for (i = 0; i < count; i++) {
221 static void __exit dm_exit(void)
223 int i = ARRAY_SIZE(_exits);
230 * Block device functions
232 static int dm_blk_open(struct inode *inode, struct file *file)
234 struct mapped_device *md;
236 spin_lock(&_minor_lock);
238 md = inode->i_bdev->bd_disk->private_data;
242 if (test_bit(DMF_FREEING, &md->flags) ||
243 test_bit(DMF_DELETING, &md->flags)) {
249 atomic_inc(&md->open_count);
252 spin_unlock(&_minor_lock);
254 return md ? 0 : -ENXIO;
257 static int dm_blk_close(struct inode *inode, struct file *file)
259 struct mapped_device *md;
261 md = inode->i_bdev->bd_disk->private_data;
262 atomic_dec(&md->open_count);
267 int dm_open_count(struct mapped_device *md)
269 return atomic_read(&md->open_count);
273 * Guarantees nothing is using the device before it's deleted.
275 int dm_lock_for_deletion(struct mapped_device *md)
279 spin_lock(&_minor_lock);
281 if (dm_open_count(md))
284 set_bit(DMF_DELETING, &md->flags);
286 spin_unlock(&_minor_lock);
291 static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo)
293 struct mapped_device *md = bdev->bd_disk->private_data;
295 return dm_get_geometry(md, geo);
298 static int dm_blk_ioctl(struct inode *inode, struct file *file,
299 unsigned int cmd, unsigned long arg)
301 struct mapped_device *md;
302 struct dm_table *map;
303 struct dm_target *tgt;
306 /* We don't really need this lock, but we do need 'inode'. */
309 md = inode->i_bdev->bd_disk->private_data;
311 map = dm_get_table(md);
313 if (!map || !dm_table_get_size(map))
316 /* We only support devices that have a single target */
317 if (dm_table_get_num_targets(map) != 1)
320 tgt = dm_table_get_target(map, 0);
322 if (dm_suspended(md)) {
327 if (tgt->type->ioctl)
328 r = tgt->type->ioctl(tgt, inode, file, cmd, arg);
337 static struct dm_io *alloc_io(struct mapped_device *md)
339 return mempool_alloc(md->io_pool, GFP_NOIO);
342 static void free_io(struct mapped_device *md, struct dm_io *io)
344 mempool_free(io, md->io_pool);
347 static struct dm_target_io *alloc_tio(struct mapped_device *md)
349 return mempool_alloc(md->tio_pool, GFP_NOIO);
352 static void free_tio(struct mapped_device *md, struct dm_target_io *tio)
354 mempool_free(tio, md->tio_pool);
357 static void start_io_acct(struct dm_io *io)
359 struct mapped_device *md = io->md;
361 io->start_time = jiffies;
364 disk_round_stats(dm_disk(md));
366 dm_disk(md)->in_flight = atomic_inc_return(&md->pending);
369 static int end_io_acct(struct dm_io *io)
371 struct mapped_device *md = io->md;
372 struct bio *bio = io->bio;
373 unsigned long duration = jiffies - io->start_time;
375 int rw = bio_data_dir(bio);
378 disk_round_stats(dm_disk(md));
380 dm_disk(md)->in_flight = pending = atomic_dec_return(&md->pending);
382 disk_stat_add(dm_disk(md), ticks[rw], duration);
388 * Add the bio to the list of deferred io.
390 static int queue_io(struct mapped_device *md, struct bio *bio)
392 down_write(&md->io_lock);
394 if (!test_bit(DMF_BLOCK_IO, &md->flags)) {
395 up_write(&md->io_lock);
399 bio_list_add(&md->deferred, bio);
401 up_write(&md->io_lock);
402 return 0; /* deferred successfully */
406 * Everyone (including functions in this file), should use this
407 * function to access the md->map field, and make sure they call
408 * dm_table_put() when finished.
410 struct dm_table *dm_get_table(struct mapped_device *md)
414 read_lock(&md->map_lock);
418 read_unlock(&md->map_lock);
424 * Get the geometry associated with a dm device
426 int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo)
434 * Set the geometry of a device.
436 int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo)
438 sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors;
440 if (geo->start > sz) {
441 DMWARN("Start sector is beyond the geometry limits.");
450 /*-----------------------------------------------------------------
452 * A more elegant soln is in the works that uses the queue
453 * merge fn, unfortunately there are a couple of changes to
454 * the block layer that I want to make for this. So in the
455 * interests of getting something for people to use I give
456 * you this clearly demarcated crap.
457 *---------------------------------------------------------------*/
459 static int __noflush_suspending(struct mapped_device *md)
461 return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
465 * Decrements the number of outstanding ios that a bio has been
466 * cloned into, completing the original io if necc.
468 static void dec_pending(struct dm_io *io, int error)
472 /* Push-back supersedes any I/O errors */
473 if (error && !(io->error > 0 && __noflush_suspending(io->md)))
476 if (atomic_dec_and_test(&io->io_count)) {
477 if (io->error == DM_ENDIO_REQUEUE) {
479 * Target requested pushing back the I/O.
480 * This must be handled before the sleeper on
481 * suspend queue merges the pushback list.
483 spin_lock_irqsave(&io->md->pushback_lock, flags);
484 if (__noflush_suspending(io->md))
485 bio_list_add(&io->md->pushback, io->bio);
487 /* noflush suspend was interrupted. */
489 spin_unlock_irqrestore(&io->md->pushback_lock, flags);
493 /* nudge anyone waiting on suspend queue */
494 wake_up(&io->md->wait);
496 if (io->error != DM_ENDIO_REQUEUE) {
497 blk_add_trace_bio(io->md->queue, io->bio,
500 bio_endio(io->bio, io->error);
507 static void clone_endio(struct bio *bio, int error)
510 struct dm_target_io *tio = bio->bi_private;
511 struct mapped_device *md = tio->io->md;
512 dm_endio_fn endio = tio->ti->type->end_io;
514 if (!bio_flagged(bio, BIO_UPTODATE) && !error)
518 r = endio(tio->ti, bio, error, &tio->info);
519 if (r < 0 || r == DM_ENDIO_REQUEUE)
521 * error and requeue request are handled
525 else if (r == DM_ENDIO_INCOMPLETE)
526 /* The target will handle the io */
529 DMWARN("unimplemented target endio return value: %d", r);
534 dec_pending(tio->io, error);
537 * Store md for cleanup instead of tio which is about to get freed.
539 bio->bi_private = md->bs;
545 static sector_t max_io_len(struct mapped_device *md,
546 sector_t sector, struct dm_target *ti)
548 sector_t offset = sector - ti->begin;
549 sector_t len = ti->len - offset;
552 * Does the target need to split even further ?
556 boundary = ((offset + ti->split_io) & ~(ti->split_io - 1))
565 static void __map_bio(struct dm_target *ti, struct bio *clone,
566 struct dm_target_io *tio)
570 struct mapped_device *md;
575 BUG_ON(!clone->bi_size);
577 clone->bi_end_io = clone_endio;
578 clone->bi_private = tio;
581 * Map the clone. If r == 0 we don't need to do
582 * anything, the target has assumed ownership of
585 atomic_inc(&tio->io->io_count);
586 sector = clone->bi_sector;
587 r = ti->type->map(ti, clone, &tio->info);
588 if (r == DM_MAPIO_REMAPPED) {
589 /* the bio has been remapped so dispatch it */
591 blk_add_trace_remap(bdev_get_queue(clone->bi_bdev), clone,
592 tio->io->bio->bi_bdev->bd_dev,
593 clone->bi_sector, sector);
595 generic_make_request(clone);
596 } else if (r < 0 || r == DM_MAPIO_REQUEUE) {
597 /* error the io and bail out, or requeue it if needed */
599 dec_pending(tio->io, r);
601 * Store bio_set for cleanup.
603 clone->bi_private = md->bs;
607 DMWARN("unimplemented target map return value: %d", r);
613 struct mapped_device *md;
614 struct dm_table *map;
618 sector_t sector_count;
622 static void dm_bio_destructor(struct bio *bio)
624 struct bio_set *bs = bio->bi_private;
630 * Creates a little bio that is just does part of a bvec.
632 static struct bio *split_bvec(struct bio *bio, sector_t sector,
633 unsigned short idx, unsigned int offset,
634 unsigned int len, struct bio_set *bs)
637 struct bio_vec *bv = bio->bi_io_vec + idx;
639 clone = bio_alloc_bioset(GFP_NOIO, 1, bs);
640 clone->bi_destructor = dm_bio_destructor;
641 *clone->bi_io_vec = *bv;
643 clone->bi_sector = sector;
644 clone->bi_bdev = bio->bi_bdev;
645 clone->bi_rw = bio->bi_rw;
647 clone->bi_size = to_bytes(len);
648 clone->bi_io_vec->bv_offset = offset;
649 clone->bi_io_vec->bv_len = clone->bi_size;
655 * Creates a bio that consists of range of complete bvecs.
657 static struct bio *clone_bio(struct bio *bio, sector_t sector,
658 unsigned short idx, unsigned short bv_count,
659 unsigned int len, struct bio_set *bs)
663 clone = bio_alloc_bioset(GFP_NOIO, bio->bi_max_vecs, bs);
664 __bio_clone(clone, bio);
665 clone->bi_destructor = dm_bio_destructor;
666 clone->bi_sector = sector;
668 clone->bi_vcnt = idx + bv_count;
669 clone->bi_size = to_bytes(len);
670 clone->bi_flags &= ~(1 << BIO_SEG_VALID);
675 static void __clone_and_map(struct clone_info *ci)
677 struct bio *clone, *bio = ci->bio;
678 struct dm_target *ti = dm_table_find_target(ci->map, ci->sector);
679 sector_t len = 0, max = max_io_len(ci->md, ci->sector, ti);
680 struct dm_target_io *tio;
683 * Allocate a target io object.
685 tio = alloc_tio(ci->md);
688 memset(&tio->info, 0, sizeof(tio->info));
690 if (ci->sector_count <= max) {
692 * Optimise for the simple case where we can do all of
693 * the remaining io with a single clone.
695 clone = clone_bio(bio, ci->sector, ci->idx,
696 bio->bi_vcnt - ci->idx, ci->sector_count,
698 __map_bio(ti, clone, tio);
699 ci->sector_count = 0;
701 } else if (to_sector(bio->bi_io_vec[ci->idx].bv_len) <= max) {
703 * There are some bvecs that don't span targets.
704 * Do as many of these as possible.
707 sector_t remaining = max;
710 for (i = ci->idx; remaining && (i < bio->bi_vcnt); i++) {
711 bv_len = to_sector(bio->bi_io_vec[i].bv_len);
713 if (bv_len > remaining)
720 clone = clone_bio(bio, ci->sector, ci->idx, i - ci->idx, len,
722 __map_bio(ti, clone, tio);
725 ci->sector_count -= len;
730 * Handle a bvec that must be split between two or more targets.
732 struct bio_vec *bv = bio->bi_io_vec + ci->idx;
733 sector_t remaining = to_sector(bv->bv_len);
734 unsigned int offset = 0;
738 ti = dm_table_find_target(ci->map, ci->sector);
739 max = max_io_len(ci->md, ci->sector, ti);
741 tio = alloc_tio(ci->md);
744 memset(&tio->info, 0, sizeof(tio->info));
747 len = min(remaining, max);
749 clone = split_bvec(bio, ci->sector, ci->idx,
750 bv->bv_offset + offset, len,
753 __map_bio(ti, clone, tio);
756 ci->sector_count -= len;
757 offset += to_bytes(len);
758 } while (remaining -= len);
765 * Split the bio into several clones.
767 static int __split_bio(struct mapped_device *md, struct bio *bio)
769 struct clone_info ci;
771 ci.map = dm_get_table(md);
772 if (unlikely(!ci.map))
777 ci.io = alloc_io(md);
779 atomic_set(&ci.io->io_count, 1);
782 ci.sector = bio->bi_sector;
783 ci.sector_count = bio_sectors(bio);
784 ci.idx = bio->bi_idx;
786 start_io_acct(ci.io);
787 while (ci.sector_count)
788 __clone_and_map(&ci);
790 /* drop the extra reference count */
791 dec_pending(ci.io, 0);
792 dm_table_put(ci.map);
796 /*-----------------------------------------------------------------
798 *---------------------------------------------------------------*/
801 * The request function that just remaps the bio built up by
804 static int dm_request(struct request_queue *q, struct bio *bio)
807 int rw = bio_data_dir(bio);
808 struct mapped_device *md = q->queuedata;
811 * There is no use in forwarding any barrier request since we can't
812 * guarantee it is (or can be) handled by the targets correctly.
814 if (unlikely(bio_barrier(bio))) {
815 bio_endio(bio, -EOPNOTSUPP);
819 down_read(&md->io_lock);
821 disk_stat_inc(dm_disk(md), ios[rw]);
822 disk_stat_add(dm_disk(md), sectors[rw], bio_sectors(bio));
825 * If we're suspended we have to queue
828 while (test_bit(DMF_BLOCK_IO, &md->flags)) {
829 up_read(&md->io_lock);
831 if (bio_rw(bio) != READA)
832 r = queue_io(md, bio);
838 * We're in a while loop, because someone could suspend
839 * before we get to the following read lock.
841 down_read(&md->io_lock);
844 r = __split_bio(md, bio);
845 up_read(&md->io_lock);
854 static void dm_unplug_all(struct request_queue *q)
856 struct mapped_device *md = q->queuedata;
857 struct dm_table *map = dm_get_table(md);
860 dm_table_unplug_all(map);
865 static int dm_any_congested(void *congested_data, int bdi_bits)
868 struct mapped_device *md = (struct mapped_device *) congested_data;
869 struct dm_table *map = dm_get_table(md);
871 if (!map || test_bit(DMF_BLOCK_IO, &md->flags))
874 r = dm_table_any_congested(map, bdi_bits);
880 /*-----------------------------------------------------------------
881 * An IDR is used to keep track of allocated minor numbers.
882 *---------------------------------------------------------------*/
883 static DEFINE_IDR(_minor_idr);
885 static void free_minor(int minor)
887 spin_lock(&_minor_lock);
888 idr_remove(&_minor_idr, minor);
889 spin_unlock(&_minor_lock);
893 * See if the device with a specific minor # is free.
895 static int specific_minor(struct mapped_device *md, int minor)
899 if (minor >= (1 << MINORBITS))
902 r = idr_pre_get(&_minor_idr, GFP_KERNEL);
906 spin_lock(&_minor_lock);
908 if (idr_find(&_minor_idr, minor)) {
913 r = idr_get_new_above(&_minor_idr, MINOR_ALLOCED, minor, &m);
918 idr_remove(&_minor_idr, m);
924 spin_unlock(&_minor_lock);
928 static int next_free_minor(struct mapped_device *md, int *minor)
932 r = idr_pre_get(&_minor_idr, GFP_KERNEL);
936 spin_lock(&_minor_lock);
938 r = idr_get_new(&_minor_idr, MINOR_ALLOCED, &m);
943 if (m >= (1 << MINORBITS)) {
944 idr_remove(&_minor_idr, m);
952 spin_unlock(&_minor_lock);
956 static struct block_device_operations dm_blk_dops;
959 * Allocate and initialise a blank device with a given minor.
961 static struct mapped_device *alloc_dev(int minor)
964 struct mapped_device *md = kmalloc(sizeof(*md), GFP_KERNEL);
968 DMWARN("unable to allocate device, out of memory.");
972 if (!try_module_get(THIS_MODULE))
975 /* get a minor number for the dev */
976 if (minor == DM_ANY_MINOR)
977 r = next_free_minor(md, &minor);
979 r = specific_minor(md, minor);
983 memset(md, 0, sizeof(*md));
984 init_rwsem(&md->io_lock);
985 init_MUTEX(&md->suspend_lock);
986 spin_lock_init(&md->pushback_lock);
987 rwlock_init(&md->map_lock);
988 atomic_set(&md->holders, 1);
989 atomic_set(&md->open_count, 0);
990 atomic_set(&md->event_nr, 0);
991 atomic_set(&md->uevent_seq, 0);
992 INIT_LIST_HEAD(&md->uevent_list);
993 spin_lock_init(&md->uevent_lock);
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;
1006 md->io_pool = mempool_create_slab_pool(MIN_IOS, _io_cache);
1010 md->tio_pool = mempool_create_slab_pool(MIN_IOS, _tio_cache);
1014 md->bs = bioset_create(16, 16);
1018 md->disk = alloc_disk(1);
1022 atomic_set(&md->pending, 0);
1023 init_waitqueue_head(&md->wait);
1024 init_waitqueue_head(&md->eventq);
1026 md->disk->major = _major;
1027 md->disk->first_minor = minor;
1028 md->disk->fops = &dm_blk_dops;
1029 md->disk->queue = md->queue;
1030 md->disk->private_data = md;
1031 sprintf(md->disk->disk_name, "dm-%d", minor);
1033 format_dev_t(md->name, MKDEV(_major, minor));
1035 /* Populate the mapping, nobody knows we exist yet */
1036 spin_lock(&_minor_lock);
1037 old_md = idr_replace(&_minor_idr, md, minor);
1038 spin_unlock(&_minor_lock);
1040 BUG_ON(old_md != MINOR_ALLOCED);
1045 bioset_free(md->bs);
1047 mempool_destroy(md->tio_pool);
1049 mempool_destroy(md->io_pool);
1051 blk_cleanup_queue(md->queue);
1055 module_put(THIS_MODULE);
1061 static void unlock_fs(struct mapped_device *md);
1063 static void free_dev(struct mapped_device *md)
1065 int minor = md->disk->first_minor;
1067 if (md->suspended_bdev) {
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 unsigned long flags;
1094 struct mapped_device *md = (struct mapped_device *) context;
1096 spin_lock_irqsave(&md->uevent_lock, flags);
1097 list_splice_init(&md->uevent_list, &uevents);
1098 spin_unlock_irqrestore(&md->uevent_lock, flags);
1100 dm_send_uevents(&uevents, &md->disk->kobj);
1102 atomic_inc(&md->event_nr);
1103 wake_up(&md->eventq);
1106 static void __set_size(struct mapped_device *md, sector_t size)
1108 set_capacity(md->disk, size);
1110 mutex_lock(&md->suspended_bdev->bd_inode->i_mutex);
1111 i_size_write(md->suspended_bdev->bd_inode, (loff_t)size << SECTOR_SHIFT);
1112 mutex_unlock(&md->suspended_bdev->bd_inode->i_mutex);
1115 static int __bind(struct mapped_device *md, struct dm_table *t)
1117 struct request_queue *q = md->queue;
1120 size = dm_table_get_size(t);
1123 * Wipe any geometry if the size of the table changed.
1125 if (size != get_capacity(md->disk))
1126 memset(&md->geometry, 0, sizeof(md->geometry));
1128 if (md->suspended_bdev)
1129 __set_size(md, size);
1134 dm_table_event_callback(t, event_callback, md);
1136 write_lock(&md->map_lock);
1138 dm_table_set_restrictions(t, q);
1139 write_unlock(&md->map_lock);
1144 static void __unbind(struct mapped_device *md)
1146 struct dm_table *map = md->map;
1151 dm_table_event_callback(map, NULL, NULL);
1152 write_lock(&md->map_lock);
1154 write_unlock(&md->map_lock);
1159 * Constructor for a new device.
1161 int dm_create(int minor, struct mapped_device **result)
1163 struct mapped_device *md;
1165 md = alloc_dev(minor);
1173 static struct mapped_device *dm_find_md(dev_t dev)
1175 struct mapped_device *md;
1176 unsigned minor = MINOR(dev);
1178 if (MAJOR(dev) != _major || minor >= (1 << MINORBITS))
1181 spin_lock(&_minor_lock);
1183 md = idr_find(&_minor_idr, minor);
1184 if (md && (md == MINOR_ALLOCED ||
1185 (dm_disk(md)->first_minor != minor) ||
1186 test_bit(DMF_FREEING, &md->flags))) {
1192 spin_unlock(&_minor_lock);
1197 struct mapped_device *dm_get_md(dev_t dev)
1199 struct mapped_device *md = dm_find_md(dev);
1207 void *dm_get_mdptr(struct mapped_device *md)
1209 return md->interface_ptr;
1212 void dm_set_mdptr(struct mapped_device *md, void *ptr)
1214 md->interface_ptr = ptr;
1217 void dm_get(struct mapped_device *md)
1219 atomic_inc(&md->holders);
1222 const char *dm_device_name(struct mapped_device *md)
1226 EXPORT_SYMBOL_GPL(dm_device_name);
1228 void dm_put(struct mapped_device *md)
1230 struct dm_table *map;
1232 BUG_ON(test_bit(DMF_FREEING, &md->flags));
1234 if (atomic_dec_and_lock(&md->holders, &_minor_lock)) {
1235 map = dm_get_table(md);
1236 idr_replace(&_minor_idr, MINOR_ALLOCED, dm_disk(md)->first_minor);
1237 set_bit(DMF_FREEING, &md->flags);
1238 spin_unlock(&_minor_lock);
1239 if (!dm_suspended(md)) {
1240 dm_table_presuspend_targets(map);
1241 dm_table_postsuspend_targets(map);
1248 EXPORT_SYMBOL_GPL(dm_put);
1251 * Process the deferred bios
1253 static void __flush_deferred_io(struct mapped_device *md, struct bio *c)
1260 if (__split_bio(md, c))
1267 * Swap in a new table (destroying old one).
1269 int dm_swap_table(struct mapped_device *md, struct dm_table *table)
1273 down(&md->suspend_lock);
1275 /* device must be suspended */
1276 if (!dm_suspended(md))
1279 /* without bdev, the device size cannot be changed */
1280 if (!md->suspended_bdev)
1281 if (get_capacity(md->disk) != dm_table_get_size(table))
1285 r = __bind(md, table);
1288 up(&md->suspend_lock);
1293 * Functions to lock and unlock any filesystem running on the
1296 static int lock_fs(struct mapped_device *md)
1300 WARN_ON(md->frozen_sb);
1302 md->frozen_sb = freeze_bdev(md->suspended_bdev);
1303 if (IS_ERR(md->frozen_sb)) {
1304 r = PTR_ERR(md->frozen_sb);
1305 md->frozen_sb = NULL;
1309 set_bit(DMF_FROZEN, &md->flags);
1311 /* don't bdput right now, we don't want the bdev
1312 * to go away while it is locked.
1317 static void unlock_fs(struct mapped_device *md)
1319 if (!test_bit(DMF_FROZEN, &md->flags))
1322 thaw_bdev(md->suspended_bdev, md->frozen_sb);
1323 md->frozen_sb = NULL;
1324 clear_bit(DMF_FROZEN, &md->flags);
1328 * We need to be able to change a mapping table under a mounted
1329 * filesystem. For example we might want to move some data in
1330 * the background. Before the table can be swapped with
1331 * dm_bind_table, dm_suspend must be called to flush any in
1332 * flight bios and ensure that any further io gets deferred.
1334 int dm_suspend(struct mapped_device *md, unsigned suspend_flags)
1336 struct dm_table *map = NULL;
1337 unsigned long flags;
1338 DECLARE_WAITQUEUE(wait, current);
1341 int do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG ? 1 : 0;
1342 int noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG ? 1 : 0;
1344 down(&md->suspend_lock);
1346 if (dm_suspended(md))
1349 map = dm_get_table(md);
1352 * DMF_NOFLUSH_SUSPENDING must be set before presuspend.
1353 * This flag is cleared before dm_suspend returns.
1356 set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
1358 /* This does not get reverted if there's an error later. */
1359 dm_table_presuspend_targets(map);
1361 /* bdget() can stall if the pending I/Os are not flushed */
1363 md->suspended_bdev = bdget_disk(md->disk, 0);
1364 if (!md->suspended_bdev) {
1365 DMWARN("bdget failed in dm_suspend");
1372 * Flush I/O to the device.
1373 * noflush supersedes do_lockfs, because lock_fs() needs to flush I/Os.
1375 if (do_lockfs && !noflush) {
1382 * First we set the BLOCK_IO flag so no more ios will be mapped.
1384 down_write(&md->io_lock);
1385 set_bit(DMF_BLOCK_IO, &md->flags);
1387 add_wait_queue(&md->wait, &wait);
1388 up_write(&md->io_lock);
1392 dm_table_unplug_all(map);
1395 * Then we wait for the already mapped ios to
1399 set_current_state(TASK_INTERRUPTIBLE);
1401 if (!atomic_read(&md->pending) || signal_pending(current))
1406 set_current_state(TASK_RUNNING);
1408 down_write(&md->io_lock);
1409 remove_wait_queue(&md->wait, &wait);
1412 spin_lock_irqsave(&md->pushback_lock, flags);
1413 clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
1414 bio_list_merge_head(&md->deferred, &md->pushback);
1415 bio_list_init(&md->pushback);
1416 spin_unlock_irqrestore(&md->pushback_lock, flags);
1419 /* were we interrupted ? */
1421 if (atomic_read(&md->pending)) {
1422 clear_bit(DMF_BLOCK_IO, &md->flags);
1423 def = bio_list_get(&md->deferred);
1424 __flush_deferred_io(md, def);
1425 up_write(&md->io_lock);
1427 goto out; /* pushback list is already flushed, so skip flush */
1429 up_write(&md->io_lock);
1431 dm_table_postsuspend_targets(map);
1433 set_bit(DMF_SUSPENDED, &md->flags);
1440 * Because there may be already I/Os in the pushback list,
1441 * flush them before return.
1443 down_write(&md->io_lock);
1445 spin_lock_irqsave(&md->pushback_lock, flags);
1446 clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
1447 bio_list_merge_head(&md->deferred, &md->pushback);
1448 bio_list_init(&md->pushback);
1449 spin_unlock_irqrestore(&md->pushback_lock, flags);
1451 def = bio_list_get(&md->deferred);
1452 __flush_deferred_io(md, def);
1453 up_write(&md->io_lock);
1457 if (r && md->suspended_bdev) {
1458 bdput(md->suspended_bdev);
1459 md->suspended_bdev = NULL;
1465 up(&md->suspend_lock);
1469 int dm_resume(struct mapped_device *md)
1473 struct dm_table *map = NULL;
1475 down(&md->suspend_lock);
1476 if (!dm_suspended(md))
1479 map = dm_get_table(md);
1480 if (!map || !dm_table_get_size(map))
1483 r = dm_table_resume_targets(map);
1487 down_write(&md->io_lock);
1488 clear_bit(DMF_BLOCK_IO, &md->flags);
1490 def = bio_list_get(&md->deferred);
1491 __flush_deferred_io(md, def);
1492 up_write(&md->io_lock);
1496 if (md->suspended_bdev) {
1497 bdput(md->suspended_bdev);
1498 md->suspended_bdev = NULL;
1501 clear_bit(DMF_SUSPENDED, &md->flags);
1503 dm_table_unplug_all(map);
1505 kobject_uevent(&md->disk->kobj, KOBJ_CHANGE);
1511 up(&md->suspend_lock);
1516 /*-----------------------------------------------------------------
1517 * Event notification.
1518 *---------------------------------------------------------------*/
1519 uint32_t dm_next_uevent_seq(struct mapped_device *md)
1521 return atomic_add_return(1, &md->uevent_seq);
1524 uint32_t dm_get_event_nr(struct mapped_device *md)
1526 return atomic_read(&md->event_nr);
1529 int dm_wait_event(struct mapped_device *md, int event_nr)
1531 return wait_event_interruptible(md->eventq,
1532 (event_nr != atomic_read(&md->event_nr)));
1535 void dm_uevent_add(struct mapped_device *md, struct list_head *elist)
1537 unsigned long flags;
1539 spin_lock_irqsave(&md->uevent_lock, flags);
1540 list_add(elist, &md->uevent_list);
1541 spin_unlock_irqrestore(&md->uevent_lock, flags);
1545 * The gendisk is only valid as long as you have a reference
1548 struct gendisk *dm_disk(struct mapped_device *md)
1553 int dm_suspended(struct mapped_device *md)
1555 return test_bit(DMF_SUSPENDED, &md->flags);
1558 int dm_noflush_suspending(struct dm_target *ti)
1560 struct mapped_device *md = dm_table_get_md(ti->table);
1561 int r = __noflush_suspending(md);
1567 EXPORT_SYMBOL_GPL(dm_noflush_suspending);
1569 static struct block_device_operations dm_blk_dops = {
1570 .open = dm_blk_open,
1571 .release = dm_blk_close,
1572 .ioctl = dm_blk_ioctl,
1573 .getgeo = dm_blk_getgeo,
1574 .owner = THIS_MODULE
1577 EXPORT_SYMBOL(dm_get_mapinfo);
1582 module_init(dm_init);
1583 module_exit(dm_exit);
1585 module_param(major, uint, 0);
1586 MODULE_PARM_DESC(major, "The major number of the device mapper");
1587 MODULE_DESCRIPTION(DM_NAME " driver");
1588 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
1589 MODULE_LICENSE("GPL");