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
75 * Work processed by per-device workqueue.
82 struct work_struct work;
83 struct mapped_device *md;
87 struct mapped_device {
88 struct rw_semaphore io_lock;
89 struct mutex suspend_lock;
90 spinlock_t pushback_lock;
97 struct request_queue *queue;
104 * A list of ios that arrived while we were suspended.
107 wait_queue_head_t wait;
108 struct bio_list deferred;
109 struct bio_list pushback;
112 * Processing queue (flush/barriers)
114 struct workqueue_struct *wq;
117 * The current mapping.
119 struct dm_table *map;
122 * io objects are allocated from here.
133 wait_queue_head_t eventq;
135 struct list_head uevent_list;
136 spinlock_t uevent_lock; /* Protect access to uevent_list */
139 * freeze/thaw support require holding onto a super block
141 struct super_block *frozen_sb;
142 struct block_device *suspended_bdev;
144 /* forced geometry settings */
145 struct hd_geometry geometry;
149 static struct kmem_cache *_io_cache;
150 static struct kmem_cache *_tio_cache;
152 static int __init local_init(void)
156 /* allocate a slab for the dm_ios */
157 _io_cache = KMEM_CACHE(dm_io, 0);
161 /* allocate a slab for the target ios */
162 _tio_cache = KMEM_CACHE(dm_target_io, 0);
164 kmem_cache_destroy(_io_cache);
168 r = dm_uevent_init();
170 kmem_cache_destroy(_tio_cache);
171 kmem_cache_destroy(_io_cache);
176 r = register_blkdev(_major, _name);
178 kmem_cache_destroy(_tio_cache);
179 kmem_cache_destroy(_io_cache);
190 static void local_exit(void)
192 kmem_cache_destroy(_tio_cache);
193 kmem_cache_destroy(_io_cache);
194 unregister_blkdev(_major, _name);
199 DMINFO("cleaned up");
202 static int (*_inits[])(void) __initdata = {
211 static void (*_exits[])(void) = {
220 static int __init dm_init(void)
222 const int count = ARRAY_SIZE(_inits);
226 for (i = 0; i < count; i++) {
241 static void __exit dm_exit(void)
243 int i = ARRAY_SIZE(_exits);
250 * Block device functions
252 static int dm_blk_open(struct inode *inode, struct file *file)
254 struct mapped_device *md;
256 spin_lock(&_minor_lock);
258 md = inode->i_bdev->bd_disk->private_data;
262 if (test_bit(DMF_FREEING, &md->flags) ||
263 test_bit(DMF_DELETING, &md->flags)) {
268 if (test_bit(DMF_FREEING, &md->flags) ||
269 test_bit(DMF_DELETING, &md->flags))
273 atomic_inc(&md->open_count);
276 spin_unlock(&_minor_lock);
278 return md ? 0 : -ENXIO;
281 static int dm_blk_close(struct inode *inode, struct file *file)
283 struct mapped_device *md;
285 md = inode->i_bdev->bd_disk->private_data;
286 atomic_dec(&md->open_count);
291 int dm_open_count(struct mapped_device *md)
293 return atomic_read(&md->open_count);
297 * Guarantees nothing is using the device before it's deleted.
299 int dm_lock_for_deletion(struct mapped_device *md)
303 spin_lock(&_minor_lock);
305 if (dm_open_count(md))
308 set_bit(DMF_DELETING, &md->flags);
310 spin_unlock(&_minor_lock);
315 static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo)
317 struct mapped_device *md = bdev->bd_disk->private_data;
319 return dm_get_geometry(md, geo);
322 static int dm_blk_ioctl(struct inode *inode, struct file *file,
323 unsigned int cmd, unsigned long arg)
325 struct mapped_device *md;
326 struct dm_table *map;
327 struct dm_target *tgt;
330 /* We don't really need this lock, but we do need 'inode'. */
333 md = inode->i_bdev->bd_disk->private_data;
335 map = dm_get_table(md);
337 if (!map || !dm_table_get_size(map))
340 /* We only support devices that have a single target */
341 if (dm_table_get_num_targets(map) != 1)
344 tgt = dm_table_get_target(map, 0);
346 if (dm_suspended(md)) {
351 if (tgt->type->ioctl)
352 r = tgt->type->ioctl(tgt, inode, file, cmd, arg);
361 static struct dm_io *alloc_io(struct mapped_device *md)
363 return mempool_alloc(md->io_pool, GFP_NOIO);
366 static void free_io(struct mapped_device *md, struct dm_io *io)
368 mempool_free(io, md->io_pool);
371 static struct dm_target_io *alloc_tio(struct mapped_device *md)
373 return mempool_alloc(md->tio_pool, GFP_NOIO);
376 static void free_tio(struct mapped_device *md, struct dm_target_io *tio)
378 mempool_free(tio, md->tio_pool);
381 static void start_io_acct(struct dm_io *io)
383 struct mapped_device *md = io->md;
385 io->start_time = jiffies;
388 disk_round_stats(dm_disk(md));
390 dm_disk(md)->in_flight = atomic_inc_return(&md->pending);
393 static int end_io_acct(struct dm_io *io)
395 struct mapped_device *md = io->md;
396 struct bio *bio = io->bio;
397 unsigned long duration = jiffies - io->start_time;
399 int rw = bio_data_dir(bio);
402 disk_round_stats(dm_disk(md));
404 dm_disk(md)->in_flight = pending = atomic_dec_return(&md->pending);
406 disk_stat_add(dm_disk(md), ticks[rw], duration);
412 * Add the bio to the list of deferred io.
414 static int queue_io(struct mapped_device *md, struct bio *bio)
416 down_write(&md->io_lock);
418 if (!test_bit(DMF_BLOCK_IO, &md->flags)) {
419 up_write(&md->io_lock);
423 bio_list_add(&md->deferred, bio);
425 up_write(&md->io_lock);
426 return 0; /* deferred successfully */
430 * Everyone (including functions in this file), should use this
431 * function to access the md->map field, and make sure they call
432 * dm_table_put() when finished.
434 struct dm_table *dm_get_table(struct mapped_device *md)
438 read_lock(&md->map_lock);
442 read_unlock(&md->map_lock);
448 * Get the geometry associated with a dm device
450 int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo)
458 * Set the geometry of a device.
460 int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo)
462 sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors;
464 if (geo->start > sz) {
465 DMWARN("Start sector is beyond the geometry limits.");
474 /*-----------------------------------------------------------------
476 * A more elegant soln is in the works that uses the queue
477 * merge fn, unfortunately there are a couple of changes to
478 * the block layer that I want to make for this. So in the
479 * interests of getting something for people to use I give
480 * you this clearly demarcated crap.
481 *---------------------------------------------------------------*/
483 static int __noflush_suspending(struct mapped_device *md)
485 return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
489 * Decrements the number of outstanding ios that a bio has been
490 * cloned into, completing the original io if necc.
492 static void dec_pending(struct dm_io *io, int error)
496 /* Push-back supersedes any I/O errors */
497 if (error && !(io->error > 0 && __noflush_suspending(io->md)))
500 if (atomic_dec_and_test(&io->io_count)) {
501 if (io->error == DM_ENDIO_REQUEUE) {
503 * Target requested pushing back the I/O.
504 * This must be handled before the sleeper on
505 * suspend queue merges the pushback list.
507 spin_lock_irqsave(&io->md->pushback_lock, flags);
508 if (__noflush_suspending(io->md))
509 bio_list_add(&io->md->pushback, io->bio);
511 /* noflush suspend was interrupted. */
513 spin_unlock_irqrestore(&io->md->pushback_lock, flags);
517 /* nudge anyone waiting on suspend queue */
518 wake_up(&io->md->wait);
520 if (io->error != DM_ENDIO_REQUEUE) {
521 blk_add_trace_bio(io->md->queue, io->bio,
524 bio_endio(io->bio, io->error);
531 static void clone_endio(struct bio *bio, int error)
534 struct dm_target_io *tio = bio->bi_private;
535 struct mapped_device *md = tio->io->md;
536 dm_endio_fn endio = tio->ti->type->end_io;
538 if (!bio_flagged(bio, BIO_UPTODATE) && !error)
542 r = endio(tio->ti, bio, error, &tio->info);
543 if (r < 0 || r == DM_ENDIO_REQUEUE)
545 * error and requeue request are handled
549 else if (r == DM_ENDIO_INCOMPLETE)
550 /* The target will handle the io */
553 DMWARN("unimplemented target endio return value: %d", r);
558 dec_pending(tio->io, error);
561 * Store md for cleanup instead of tio which is about to get freed.
563 bio->bi_private = md->bs;
569 static sector_t max_io_len(struct mapped_device *md,
570 sector_t sector, struct dm_target *ti)
572 sector_t offset = sector - ti->begin;
573 sector_t len = ti->len - offset;
576 * Does the target need to split even further ?
580 boundary = ((offset + ti->split_io) & ~(ti->split_io - 1))
589 static void __map_bio(struct dm_target *ti, struct bio *clone,
590 struct dm_target_io *tio)
594 struct mapped_device *md;
599 BUG_ON(!clone->bi_size);
601 clone->bi_end_io = clone_endio;
602 clone->bi_private = tio;
605 * Map the clone. If r == 0 we don't need to do
606 * anything, the target has assumed ownership of
609 atomic_inc(&tio->io->io_count);
610 sector = clone->bi_sector;
611 r = ti->type->map(ti, clone, &tio->info);
612 if (r == DM_MAPIO_REMAPPED) {
613 /* the bio has been remapped so dispatch it */
615 blk_add_trace_remap(bdev_get_queue(clone->bi_bdev), clone,
616 tio->io->bio->bi_bdev->bd_dev,
617 clone->bi_sector, sector);
619 generic_make_request(clone);
620 } else if (r < 0 || r == DM_MAPIO_REQUEUE) {
621 /* error the io and bail out, or requeue it if needed */
623 dec_pending(tio->io, r);
625 * Store bio_set for cleanup.
627 clone->bi_private = md->bs;
631 DMWARN("unimplemented target map return value: %d", r);
637 struct mapped_device *md;
638 struct dm_table *map;
642 sector_t sector_count;
646 static void dm_bio_destructor(struct bio *bio)
648 struct bio_set *bs = bio->bi_private;
654 * Creates a little bio that is just does part of a bvec.
656 static struct bio *split_bvec(struct bio *bio, sector_t sector,
657 unsigned short idx, unsigned int offset,
658 unsigned int len, struct bio_set *bs)
661 struct bio_vec *bv = bio->bi_io_vec + idx;
663 clone = bio_alloc_bioset(GFP_NOIO, 1, bs);
664 clone->bi_destructor = dm_bio_destructor;
665 *clone->bi_io_vec = *bv;
667 clone->bi_sector = sector;
668 clone->bi_bdev = bio->bi_bdev;
669 clone->bi_rw = bio->bi_rw;
671 clone->bi_size = to_bytes(len);
672 clone->bi_io_vec->bv_offset = offset;
673 clone->bi_io_vec->bv_len = clone->bi_size;
679 * Creates a bio that consists of range of complete bvecs.
681 static struct bio *clone_bio(struct bio *bio, sector_t sector,
682 unsigned short idx, unsigned short bv_count,
683 unsigned int len, struct bio_set *bs)
687 clone = bio_alloc_bioset(GFP_NOIO, bio->bi_max_vecs, bs);
688 __bio_clone(clone, bio);
689 clone->bi_destructor = dm_bio_destructor;
690 clone->bi_sector = sector;
692 clone->bi_vcnt = idx + bv_count;
693 clone->bi_size = to_bytes(len);
694 clone->bi_flags &= ~(1 << BIO_SEG_VALID);
699 static int __clone_and_map(struct clone_info *ci)
701 struct bio *clone, *bio = ci->bio;
702 struct dm_target *ti;
703 sector_t len = 0, max;
704 struct dm_target_io *tio;
706 ti = dm_table_find_target(ci->map, ci->sector);
707 if (!dm_target_is_valid(ti))
710 max = max_io_len(ci->md, ci->sector, ti);
713 * Allocate a target io object.
715 tio = alloc_tio(ci->md);
718 memset(&tio->info, 0, sizeof(tio->info));
720 if (ci->sector_count <= max) {
722 * Optimise for the simple case where we can do all of
723 * the remaining io with a single clone.
725 clone = clone_bio(bio, ci->sector, ci->idx,
726 bio->bi_vcnt - ci->idx, ci->sector_count,
728 __map_bio(ti, clone, tio);
729 ci->sector_count = 0;
731 } else if (to_sector(bio->bi_io_vec[ci->idx].bv_len) <= max) {
733 * There are some bvecs that don't span targets.
734 * Do as many of these as possible.
737 sector_t remaining = max;
740 for (i = ci->idx; remaining && (i < bio->bi_vcnt); i++) {
741 bv_len = to_sector(bio->bi_io_vec[i].bv_len);
743 if (bv_len > remaining)
750 clone = clone_bio(bio, ci->sector, ci->idx, i - ci->idx, len,
752 __map_bio(ti, clone, tio);
755 ci->sector_count -= len;
760 * Handle a bvec that must be split between two or more targets.
762 struct bio_vec *bv = bio->bi_io_vec + ci->idx;
763 sector_t remaining = to_sector(bv->bv_len);
764 unsigned int offset = 0;
768 ti = dm_table_find_target(ci->map, ci->sector);
769 if (!dm_target_is_valid(ti))
772 max = max_io_len(ci->md, ci->sector, ti);
774 tio = alloc_tio(ci->md);
777 memset(&tio->info, 0, sizeof(tio->info));
780 len = min(remaining, max);
782 clone = split_bvec(bio, ci->sector, ci->idx,
783 bv->bv_offset + offset, len,
786 __map_bio(ti, clone, tio);
789 ci->sector_count -= len;
790 offset += to_bytes(len);
791 } while (remaining -= len);
800 * Split the bio into several clones.
802 static int __split_bio(struct mapped_device *md, struct bio *bio)
804 struct clone_info ci;
807 ci.map = dm_get_table(md);
808 if (unlikely(!ci.map))
813 ci.io = alloc_io(md);
815 atomic_set(&ci.io->io_count, 1);
818 ci.sector = bio->bi_sector;
819 ci.sector_count = bio_sectors(bio);
820 ci.idx = bio->bi_idx;
822 start_io_acct(ci.io);
823 while (ci.sector_count && !error)
824 error = __clone_and_map(&ci);
826 /* drop the extra reference count */
827 dec_pending(ci.io, error);
828 dm_table_put(ci.map);
832 /*-----------------------------------------------------------------
834 *---------------------------------------------------------------*/
836 static int dm_merge_bvec(struct request_queue *q,
837 struct bvec_merge_data *bvm,
838 struct bio_vec *biovec)
840 struct mapped_device *md = q->queuedata;
841 struct dm_table *map = dm_get_table(md);
842 struct dm_target *ti;
843 sector_t max_sectors;
849 ti = dm_table_find_target(map, bvm->bi_sector);
850 if (!dm_target_is_valid(ti))
854 * Find maximum amount of I/O that won't need splitting
856 max_sectors = min(max_io_len(md, bvm->bi_sector, ti),
857 (sector_t) BIO_MAX_SECTORS);
858 max_size = (max_sectors << SECTOR_SHIFT) - bvm->bi_size;
863 * merge_bvec_fn() returns number of bytes
864 * it can accept at this offset
865 * max is precomputed maximal io size
867 if (max_size && ti->type->merge)
868 max_size = ti->type->merge(ti, bvm, biovec, max_size);
875 * Always allow an entire first page
877 if (max_size <= biovec->bv_len && !(bvm->bi_size >> SECTOR_SHIFT))
878 max_size = biovec->bv_len;
884 * The request function that just remaps the bio built up by
887 static int dm_request(struct request_queue *q, struct bio *bio)
890 int rw = bio_data_dir(bio);
891 struct mapped_device *md = q->queuedata;
894 * There is no use in forwarding any barrier request since we can't
895 * guarantee it is (or can be) handled by the targets correctly.
897 if (unlikely(bio_barrier(bio))) {
898 bio_endio(bio, -EOPNOTSUPP);
902 down_read(&md->io_lock);
904 disk_stat_inc(dm_disk(md), ios[rw]);
905 disk_stat_add(dm_disk(md), sectors[rw], bio_sectors(bio));
908 * If we're suspended we have to queue
911 while (test_bit(DMF_BLOCK_IO, &md->flags)) {
912 up_read(&md->io_lock);
914 if (bio_rw(bio) != READA)
915 r = queue_io(md, bio);
921 * We're in a while loop, because someone could suspend
922 * before we get to the following read lock.
924 down_read(&md->io_lock);
927 r = __split_bio(md, bio);
928 up_read(&md->io_lock);
937 static void dm_unplug_all(struct request_queue *q)
939 struct mapped_device *md = q->queuedata;
940 struct dm_table *map = dm_get_table(md);
943 dm_table_unplug_all(map);
948 static int dm_any_congested(void *congested_data, int bdi_bits)
951 struct mapped_device *md = (struct mapped_device *) congested_data;
952 struct dm_table *map = dm_get_table(md);
954 if (!map || test_bit(DMF_BLOCK_IO, &md->flags))
957 r = dm_table_any_congested(map, bdi_bits);
963 /*-----------------------------------------------------------------
964 * An IDR is used to keep track of allocated minor numbers.
965 *---------------------------------------------------------------*/
966 static DEFINE_IDR(_minor_idr);
968 static void free_minor(int minor)
970 spin_lock(&_minor_lock);
971 idr_remove(&_minor_idr, minor);
972 spin_unlock(&_minor_lock);
976 * See if the device with a specific minor # is free.
978 static int specific_minor(int minor)
982 if (minor >= (1 << MINORBITS))
985 r = idr_pre_get(&_minor_idr, GFP_KERNEL);
989 spin_lock(&_minor_lock);
991 if (idr_find(&_minor_idr, minor)) {
996 r = idr_get_new_above(&_minor_idr, MINOR_ALLOCED, minor, &m);
1001 idr_remove(&_minor_idr, m);
1007 spin_unlock(&_minor_lock);
1011 static int next_free_minor(int *minor)
1015 r = idr_pre_get(&_minor_idr, GFP_KERNEL);
1019 spin_lock(&_minor_lock);
1021 r = idr_get_new(&_minor_idr, MINOR_ALLOCED, &m);
1025 if (m >= (1 << MINORBITS)) {
1026 idr_remove(&_minor_idr, m);
1034 spin_unlock(&_minor_lock);
1038 static struct block_device_operations dm_blk_dops;
1041 * Allocate and initialise a blank device with a given minor.
1043 static struct mapped_device *alloc_dev(int minor)
1046 struct mapped_device *md = kzalloc(sizeof(*md), GFP_KERNEL);
1050 DMWARN("unable to allocate device, out of memory.");
1054 if (!try_module_get(THIS_MODULE))
1055 goto bad_module_get;
1057 /* get a minor number for the dev */
1058 if (minor == DM_ANY_MINOR)
1059 r = next_free_minor(&minor);
1061 r = specific_minor(minor);
1065 init_rwsem(&md->io_lock);
1066 mutex_init(&md->suspend_lock);
1067 spin_lock_init(&md->pushback_lock);
1068 rwlock_init(&md->map_lock);
1069 atomic_set(&md->holders, 1);
1070 atomic_set(&md->open_count, 0);
1071 atomic_set(&md->event_nr, 0);
1072 atomic_set(&md->uevent_seq, 0);
1073 INIT_LIST_HEAD(&md->uevent_list);
1074 spin_lock_init(&md->uevent_lock);
1076 md->queue = blk_alloc_queue(GFP_KERNEL);
1080 md->queue->queuedata = md;
1081 md->queue->backing_dev_info.congested_fn = dm_any_congested;
1082 md->queue->backing_dev_info.congested_data = md;
1083 blk_queue_make_request(md->queue, dm_request);
1084 blk_queue_bounce_limit(md->queue, BLK_BOUNCE_ANY);
1085 md->queue->unplug_fn = dm_unplug_all;
1086 blk_queue_merge_bvec(md->queue, dm_merge_bvec);
1088 md->io_pool = mempool_create_slab_pool(MIN_IOS, _io_cache);
1092 md->tio_pool = mempool_create_slab_pool(MIN_IOS, _tio_cache);
1096 md->bs = bioset_create(16, 16);
1100 md->disk = alloc_disk(1);
1104 atomic_set(&md->pending, 0);
1105 init_waitqueue_head(&md->wait);
1106 init_waitqueue_head(&md->eventq);
1108 md->disk->major = _major;
1109 md->disk->first_minor = minor;
1110 md->disk->fops = &dm_blk_dops;
1111 md->disk->queue = md->queue;
1112 md->disk->private_data = md;
1113 sprintf(md->disk->disk_name, "dm-%d", minor);
1115 format_dev_t(md->name, MKDEV(_major, minor));
1117 md->wq = create_singlethread_workqueue("kdmflush");
1121 /* Populate the mapping, nobody knows we exist yet */
1122 spin_lock(&_minor_lock);
1123 old_md = idr_replace(&_minor_idr, md, minor);
1124 spin_unlock(&_minor_lock);
1126 BUG_ON(old_md != MINOR_ALLOCED);
1133 bioset_free(md->bs);
1135 mempool_destroy(md->tio_pool);
1137 mempool_destroy(md->io_pool);
1139 blk_cleanup_queue(md->queue);
1143 module_put(THIS_MODULE);
1149 static void unlock_fs(struct mapped_device *md);
1151 static void free_dev(struct mapped_device *md)
1153 int minor = md->disk->first_minor;
1155 if (md->suspended_bdev) {
1157 bdput(md->suspended_bdev);
1159 destroy_workqueue(md->wq);
1160 mempool_destroy(md->tio_pool);
1161 mempool_destroy(md->io_pool);
1162 bioset_free(md->bs);
1163 del_gendisk(md->disk);
1166 spin_lock(&_minor_lock);
1167 md->disk->private_data = NULL;
1168 spin_unlock(&_minor_lock);
1171 blk_cleanup_queue(md->queue);
1172 module_put(THIS_MODULE);
1177 * Bind a table to the device.
1179 static void event_callback(void *context)
1181 unsigned long flags;
1183 struct mapped_device *md = (struct mapped_device *) context;
1185 spin_lock_irqsave(&md->uevent_lock, flags);
1186 list_splice_init(&md->uevent_list, &uevents);
1187 spin_unlock_irqrestore(&md->uevent_lock, flags);
1189 dm_send_uevents(&uevents, &md->disk->dev.kobj);
1191 atomic_inc(&md->event_nr);
1192 wake_up(&md->eventq);
1195 static void __set_size(struct mapped_device *md, sector_t size)
1197 set_capacity(md->disk, size);
1199 mutex_lock(&md->suspended_bdev->bd_inode->i_mutex);
1200 i_size_write(md->suspended_bdev->bd_inode, (loff_t)size << SECTOR_SHIFT);
1201 mutex_unlock(&md->suspended_bdev->bd_inode->i_mutex);
1204 static int __bind(struct mapped_device *md, struct dm_table *t)
1206 struct request_queue *q = md->queue;
1209 size = dm_table_get_size(t);
1212 * Wipe any geometry if the size of the table changed.
1214 if (size != get_capacity(md->disk))
1215 memset(&md->geometry, 0, sizeof(md->geometry));
1217 if (md->suspended_bdev)
1218 __set_size(md, size);
1223 dm_table_event_callback(t, event_callback, md);
1225 write_lock(&md->map_lock);
1227 dm_table_set_restrictions(t, q);
1228 write_unlock(&md->map_lock);
1233 static void __unbind(struct mapped_device *md)
1235 struct dm_table *map = md->map;
1240 dm_table_event_callback(map, NULL, NULL);
1241 write_lock(&md->map_lock);
1243 write_unlock(&md->map_lock);
1248 * Constructor for a new device.
1250 int dm_create(int minor, struct mapped_device **result)
1252 struct mapped_device *md;
1254 md = alloc_dev(minor);
1262 static struct mapped_device *dm_find_md(dev_t dev)
1264 struct mapped_device *md;
1265 unsigned minor = MINOR(dev);
1267 if (MAJOR(dev) != _major || minor >= (1 << MINORBITS))
1270 spin_lock(&_minor_lock);
1272 md = idr_find(&_minor_idr, minor);
1273 if (md && (md == MINOR_ALLOCED ||
1274 (dm_disk(md)->first_minor != minor) ||
1275 test_bit(DMF_FREEING, &md->flags))) {
1281 spin_unlock(&_minor_lock);
1286 struct mapped_device *dm_get_md(dev_t dev)
1288 struct mapped_device *md = dm_find_md(dev);
1296 void *dm_get_mdptr(struct mapped_device *md)
1298 return md->interface_ptr;
1301 void dm_set_mdptr(struct mapped_device *md, void *ptr)
1303 md->interface_ptr = ptr;
1306 void dm_get(struct mapped_device *md)
1308 atomic_inc(&md->holders);
1311 const char *dm_device_name(struct mapped_device *md)
1315 EXPORT_SYMBOL_GPL(dm_device_name);
1317 void dm_put(struct mapped_device *md)
1319 struct dm_table *map;
1321 BUG_ON(test_bit(DMF_FREEING, &md->flags));
1323 if (atomic_dec_and_lock(&md->holders, &_minor_lock)) {
1324 map = dm_get_table(md);
1325 idr_replace(&_minor_idr, MINOR_ALLOCED, dm_disk(md)->first_minor);
1326 set_bit(DMF_FREEING, &md->flags);
1327 spin_unlock(&_minor_lock);
1328 if (!dm_suspended(md)) {
1329 dm_table_presuspend_targets(map);
1330 dm_table_postsuspend_targets(map);
1337 EXPORT_SYMBOL_GPL(dm_put);
1339 static int dm_wait_for_completion(struct mapped_device *md)
1344 set_current_state(TASK_INTERRUPTIBLE);
1347 if (!atomic_read(&md->pending))
1350 if (signal_pending(current)) {
1357 set_current_state(TASK_RUNNING);
1363 * Process the deferred bios
1365 static void __flush_deferred_io(struct mapped_device *md)
1369 while ((c = bio_list_pop(&md->deferred))) {
1370 if (__split_bio(md, c))
1374 clear_bit(DMF_BLOCK_IO, &md->flags);
1377 static void __merge_pushback_list(struct mapped_device *md)
1379 unsigned long flags;
1381 spin_lock_irqsave(&md->pushback_lock, flags);
1382 clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
1383 bio_list_merge_head(&md->deferred, &md->pushback);
1384 bio_list_init(&md->pushback);
1385 spin_unlock_irqrestore(&md->pushback_lock, flags);
1388 static void dm_wq_work(struct work_struct *work)
1390 struct dm_wq_req *req = container_of(work, struct dm_wq_req, work);
1391 struct mapped_device *md = req->md;
1393 down_write(&md->io_lock);
1394 switch (req->type) {
1395 case DM_WQ_FLUSH_ALL:
1396 __merge_pushback_list(md);
1398 case DM_WQ_FLUSH_DEFERRED:
1399 __flush_deferred_io(md);
1402 DMERR("dm_wq_work: unrecognised work type %d", req->type);
1405 up_write(&md->io_lock);
1408 static void dm_wq_queue(struct mapped_device *md, int type, void *context,
1409 struct dm_wq_req *req)
1413 req->context = context;
1414 INIT_WORK(&req->work, dm_wq_work);
1415 queue_work(md->wq, &req->work);
1418 static void dm_queue_flush(struct mapped_device *md, int type, void *context)
1420 struct dm_wq_req req;
1422 dm_wq_queue(md, type, context, &req);
1423 flush_workqueue(md->wq);
1427 * Swap in a new table (destroying old one).
1429 int dm_swap_table(struct mapped_device *md, struct dm_table *table)
1433 mutex_lock(&md->suspend_lock);
1435 /* device must be suspended */
1436 if (!dm_suspended(md))
1439 /* without bdev, the device size cannot be changed */
1440 if (!md->suspended_bdev)
1441 if (get_capacity(md->disk) != dm_table_get_size(table))
1445 r = __bind(md, table);
1448 mutex_unlock(&md->suspend_lock);
1453 * Functions to lock and unlock any filesystem running on the
1456 static int lock_fs(struct mapped_device *md)
1460 WARN_ON(md->frozen_sb);
1462 md->frozen_sb = freeze_bdev(md->suspended_bdev);
1463 if (IS_ERR(md->frozen_sb)) {
1464 r = PTR_ERR(md->frozen_sb);
1465 md->frozen_sb = NULL;
1469 set_bit(DMF_FROZEN, &md->flags);
1471 /* don't bdput right now, we don't want the bdev
1472 * to go away while it is locked.
1477 static void unlock_fs(struct mapped_device *md)
1479 if (!test_bit(DMF_FROZEN, &md->flags))
1482 thaw_bdev(md->suspended_bdev, md->frozen_sb);
1483 md->frozen_sb = NULL;
1484 clear_bit(DMF_FROZEN, &md->flags);
1488 * We need to be able to change a mapping table under a mounted
1489 * filesystem. For example we might want to move some data in
1490 * the background. Before the table can be swapped with
1491 * dm_bind_table, dm_suspend must be called to flush any in
1492 * flight bios and ensure that any further io gets deferred.
1494 int dm_suspend(struct mapped_device *md, unsigned suspend_flags)
1496 struct dm_table *map = NULL;
1497 DECLARE_WAITQUEUE(wait, current);
1499 int do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG ? 1 : 0;
1500 int noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG ? 1 : 0;
1502 mutex_lock(&md->suspend_lock);
1504 if (dm_suspended(md)) {
1509 map = dm_get_table(md);
1512 * DMF_NOFLUSH_SUSPENDING must be set before presuspend.
1513 * This flag is cleared before dm_suspend returns.
1516 set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
1518 /* This does not get reverted if there's an error later. */
1519 dm_table_presuspend_targets(map);
1521 /* bdget() can stall if the pending I/Os are not flushed */
1523 md->suspended_bdev = bdget_disk(md->disk, 0);
1524 if (!md->suspended_bdev) {
1525 DMWARN("bdget failed in dm_suspend");
1531 * Flush I/O to the device. noflush supersedes do_lockfs,
1532 * because lock_fs() needs to flush I/Os.
1542 * First we set the BLOCK_IO flag so no more ios will be mapped.
1544 down_write(&md->io_lock);
1545 set_bit(DMF_BLOCK_IO, &md->flags);
1547 add_wait_queue(&md->wait, &wait);
1548 up_write(&md->io_lock);
1552 dm_table_unplug_all(map);
1555 * Wait for the already-mapped ios to complete.
1557 r = dm_wait_for_completion(md);
1559 down_write(&md->io_lock);
1560 remove_wait_queue(&md->wait, &wait);
1563 __merge_pushback_list(md);
1564 up_write(&md->io_lock);
1566 /* were we interrupted ? */
1568 dm_queue_flush(md, DM_WQ_FLUSH_DEFERRED, NULL);
1571 goto out; /* pushback list is already flushed, so skip flush */
1574 dm_table_postsuspend_targets(map);
1576 set_bit(DMF_SUSPENDED, &md->flags);
1581 * Because there may be already I/Os in the pushback list,
1582 * flush them before return.
1584 dm_queue_flush(md, DM_WQ_FLUSH_ALL, NULL);
1587 if (r && md->suspended_bdev) {
1588 bdput(md->suspended_bdev);
1589 md->suspended_bdev = NULL;
1595 mutex_unlock(&md->suspend_lock);
1599 int dm_resume(struct mapped_device *md)
1602 struct dm_table *map = NULL;
1604 mutex_lock(&md->suspend_lock);
1605 if (!dm_suspended(md))
1608 map = dm_get_table(md);
1609 if (!map || !dm_table_get_size(map))
1612 r = dm_table_resume_targets(map);
1616 dm_queue_flush(md, DM_WQ_FLUSH_DEFERRED, NULL);
1620 if (md->suspended_bdev) {
1621 bdput(md->suspended_bdev);
1622 md->suspended_bdev = NULL;
1625 clear_bit(DMF_SUSPENDED, &md->flags);
1627 dm_table_unplug_all(map);
1629 dm_kobject_uevent(md);
1635 mutex_unlock(&md->suspend_lock);
1640 /*-----------------------------------------------------------------
1641 * Event notification.
1642 *---------------------------------------------------------------*/
1643 void dm_kobject_uevent(struct mapped_device *md)
1645 kobject_uevent(&md->disk->dev.kobj, KOBJ_CHANGE);
1648 uint32_t dm_next_uevent_seq(struct mapped_device *md)
1650 return atomic_add_return(1, &md->uevent_seq);
1653 uint32_t dm_get_event_nr(struct mapped_device *md)
1655 return atomic_read(&md->event_nr);
1658 int dm_wait_event(struct mapped_device *md, int event_nr)
1660 return wait_event_interruptible(md->eventq,
1661 (event_nr != atomic_read(&md->event_nr)));
1664 void dm_uevent_add(struct mapped_device *md, struct list_head *elist)
1666 unsigned long flags;
1668 spin_lock_irqsave(&md->uevent_lock, flags);
1669 list_add(elist, &md->uevent_list);
1670 spin_unlock_irqrestore(&md->uevent_lock, flags);
1674 * The gendisk is only valid as long as you have a reference
1677 struct gendisk *dm_disk(struct mapped_device *md)
1682 int dm_suspended(struct mapped_device *md)
1684 return test_bit(DMF_SUSPENDED, &md->flags);
1687 int dm_noflush_suspending(struct dm_target *ti)
1689 struct mapped_device *md = dm_table_get_md(ti->table);
1690 int r = __noflush_suspending(md);
1696 EXPORT_SYMBOL_GPL(dm_noflush_suspending);
1698 static struct block_device_operations dm_blk_dops = {
1699 .open = dm_blk_open,
1700 .release = dm_blk_close,
1701 .ioctl = dm_blk_ioctl,
1702 .getgeo = dm_blk_getgeo,
1703 .owner = THIS_MODULE
1706 EXPORT_SYMBOL(dm_get_mapinfo);
1711 module_init(dm_init);
1712 module_exit(dm_exit);
1714 module_param(major, uint, 0);
1715 MODULE_PARM_DESC(major, "The major number of the device mapper");
1716 MODULE_DESCRIPTION(DM_NAME " driver");
1717 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
1718 MODULE_LICENSE("GPL");