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 = {
210 static void (*_exits[])(void) = {
218 static int __init dm_init(void)
220 const int count = ARRAY_SIZE(_inits);
224 for (i = 0; i < count; i++) {
239 static void __exit dm_exit(void)
241 int i = ARRAY_SIZE(_exits);
248 * Block device functions
250 static int dm_blk_open(struct inode *inode, struct file *file)
252 struct mapped_device *md;
254 spin_lock(&_minor_lock);
256 md = inode->i_bdev->bd_disk->private_data;
260 if (test_bit(DMF_FREEING, &md->flags) ||
261 test_bit(DMF_DELETING, &md->flags)) {
267 atomic_inc(&md->open_count);
270 spin_unlock(&_minor_lock);
272 return md ? 0 : -ENXIO;
275 static int dm_blk_close(struct inode *inode, struct file *file)
277 struct mapped_device *md;
279 md = inode->i_bdev->bd_disk->private_data;
280 atomic_dec(&md->open_count);
285 int dm_open_count(struct mapped_device *md)
287 return atomic_read(&md->open_count);
291 * Guarantees nothing is using the device before it's deleted.
293 int dm_lock_for_deletion(struct mapped_device *md)
297 spin_lock(&_minor_lock);
299 if (dm_open_count(md))
302 set_bit(DMF_DELETING, &md->flags);
304 spin_unlock(&_minor_lock);
309 static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo)
311 struct mapped_device *md = bdev->bd_disk->private_data;
313 return dm_get_geometry(md, geo);
316 static int dm_blk_ioctl(struct inode *inode, struct file *file,
317 unsigned int cmd, unsigned long arg)
319 struct mapped_device *md;
320 struct dm_table *map;
321 struct dm_target *tgt;
324 /* We don't really need this lock, but we do need 'inode'. */
327 md = inode->i_bdev->bd_disk->private_data;
329 map = dm_get_table(md);
331 if (!map || !dm_table_get_size(map))
334 /* We only support devices that have a single target */
335 if (dm_table_get_num_targets(map) != 1)
338 tgt = dm_table_get_target(map, 0);
340 if (dm_suspended(md)) {
345 if (tgt->type->ioctl)
346 r = tgt->type->ioctl(tgt, inode, file, cmd, arg);
355 static struct dm_io *alloc_io(struct mapped_device *md)
357 return mempool_alloc(md->io_pool, GFP_NOIO);
360 static void free_io(struct mapped_device *md, struct dm_io *io)
362 mempool_free(io, md->io_pool);
365 static struct dm_target_io *alloc_tio(struct mapped_device *md)
367 return mempool_alloc(md->tio_pool, GFP_NOIO);
370 static void free_tio(struct mapped_device *md, struct dm_target_io *tio)
372 mempool_free(tio, md->tio_pool);
375 static void start_io_acct(struct dm_io *io)
377 struct mapped_device *md = io->md;
379 io->start_time = jiffies;
382 disk_round_stats(dm_disk(md));
384 dm_disk(md)->in_flight = atomic_inc_return(&md->pending);
387 static int end_io_acct(struct dm_io *io)
389 struct mapped_device *md = io->md;
390 struct bio *bio = io->bio;
391 unsigned long duration = jiffies - io->start_time;
393 int rw = bio_data_dir(bio);
396 disk_round_stats(dm_disk(md));
398 dm_disk(md)->in_flight = pending = atomic_dec_return(&md->pending);
400 disk_stat_add(dm_disk(md), ticks[rw], duration);
406 * Add the bio to the list of deferred io.
408 static int queue_io(struct mapped_device *md, struct bio *bio)
410 down_write(&md->io_lock);
412 if (!test_bit(DMF_BLOCK_IO, &md->flags)) {
413 up_write(&md->io_lock);
417 bio_list_add(&md->deferred, bio);
419 up_write(&md->io_lock);
420 return 0; /* deferred successfully */
424 * Everyone (including functions in this file), should use this
425 * function to access the md->map field, and make sure they call
426 * dm_table_put() when finished.
428 struct dm_table *dm_get_table(struct mapped_device *md)
432 read_lock(&md->map_lock);
436 read_unlock(&md->map_lock);
442 * Get the geometry associated with a dm device
444 int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo)
452 * Set the geometry of a device.
454 int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo)
456 sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors;
458 if (geo->start > sz) {
459 DMWARN("Start sector is beyond the geometry limits.");
468 /*-----------------------------------------------------------------
470 * A more elegant soln is in the works that uses the queue
471 * merge fn, unfortunately there are a couple of changes to
472 * the block layer that I want to make for this. So in the
473 * interests of getting something for people to use I give
474 * you this clearly demarcated crap.
475 *---------------------------------------------------------------*/
477 static int __noflush_suspending(struct mapped_device *md)
479 return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
483 * Decrements the number of outstanding ios that a bio has been
484 * cloned into, completing the original io if necc.
486 static void dec_pending(struct dm_io *io, int error)
490 /* Push-back supersedes any I/O errors */
491 if (error && !(io->error > 0 && __noflush_suspending(io->md)))
494 if (atomic_dec_and_test(&io->io_count)) {
495 if (io->error == DM_ENDIO_REQUEUE) {
497 * Target requested pushing back the I/O.
498 * This must be handled before the sleeper on
499 * suspend queue merges the pushback list.
501 spin_lock_irqsave(&io->md->pushback_lock, flags);
502 if (__noflush_suspending(io->md))
503 bio_list_add(&io->md->pushback, io->bio);
505 /* noflush suspend was interrupted. */
507 spin_unlock_irqrestore(&io->md->pushback_lock, flags);
511 /* nudge anyone waiting on suspend queue */
512 wake_up(&io->md->wait);
514 if (io->error != DM_ENDIO_REQUEUE) {
515 blk_add_trace_bio(io->md->queue, io->bio,
518 bio_endio(io->bio, io->error);
525 static void clone_endio(struct bio *bio, int error)
528 struct dm_target_io *tio = bio->bi_private;
529 struct mapped_device *md = tio->io->md;
530 dm_endio_fn endio = tio->ti->type->end_io;
532 if (!bio_flagged(bio, BIO_UPTODATE) && !error)
536 r = endio(tio->ti, bio, error, &tio->info);
537 if (r < 0 || r == DM_ENDIO_REQUEUE)
539 * error and requeue request are handled
543 else if (r == DM_ENDIO_INCOMPLETE)
544 /* The target will handle the io */
547 DMWARN("unimplemented target endio return value: %d", r);
552 dec_pending(tio->io, error);
555 * Store md for cleanup instead of tio which is about to get freed.
557 bio->bi_private = md->bs;
563 static sector_t max_io_len(struct mapped_device *md,
564 sector_t sector, struct dm_target *ti)
566 sector_t offset = sector - ti->begin;
567 sector_t len = ti->len - offset;
570 * Does the target need to split even further ?
574 boundary = ((offset + ti->split_io) & ~(ti->split_io - 1))
583 static void __map_bio(struct dm_target *ti, struct bio *clone,
584 struct dm_target_io *tio)
588 struct mapped_device *md;
593 BUG_ON(!clone->bi_size);
595 clone->bi_end_io = clone_endio;
596 clone->bi_private = tio;
599 * Map the clone. If r == 0 we don't need to do
600 * anything, the target has assumed ownership of
603 atomic_inc(&tio->io->io_count);
604 sector = clone->bi_sector;
605 r = ti->type->map(ti, clone, &tio->info);
606 if (r == DM_MAPIO_REMAPPED) {
607 /* the bio has been remapped so dispatch it */
609 blk_add_trace_remap(bdev_get_queue(clone->bi_bdev), clone,
610 tio->io->bio->bi_bdev->bd_dev,
611 clone->bi_sector, sector);
613 generic_make_request(clone);
614 } else if (r < 0 || r == DM_MAPIO_REQUEUE) {
615 /* error the io and bail out, or requeue it if needed */
617 dec_pending(tio->io, r);
619 * Store bio_set for cleanup.
621 clone->bi_private = md->bs;
625 DMWARN("unimplemented target map return value: %d", r);
631 struct mapped_device *md;
632 struct dm_table *map;
636 sector_t sector_count;
640 static void dm_bio_destructor(struct bio *bio)
642 struct bio_set *bs = bio->bi_private;
648 * Creates a little bio that is just does part of a bvec.
650 static struct bio *split_bvec(struct bio *bio, sector_t sector,
651 unsigned short idx, unsigned int offset,
652 unsigned int len, struct bio_set *bs)
655 struct bio_vec *bv = bio->bi_io_vec + idx;
657 clone = bio_alloc_bioset(GFP_NOIO, 1, bs);
658 clone->bi_destructor = dm_bio_destructor;
659 *clone->bi_io_vec = *bv;
661 clone->bi_sector = sector;
662 clone->bi_bdev = bio->bi_bdev;
663 clone->bi_rw = bio->bi_rw;
665 clone->bi_size = to_bytes(len);
666 clone->bi_io_vec->bv_offset = offset;
667 clone->bi_io_vec->bv_len = clone->bi_size;
673 * Creates a bio that consists of range of complete bvecs.
675 static struct bio *clone_bio(struct bio *bio, sector_t sector,
676 unsigned short idx, unsigned short bv_count,
677 unsigned int len, struct bio_set *bs)
681 clone = bio_alloc_bioset(GFP_NOIO, bio->bi_max_vecs, bs);
682 __bio_clone(clone, bio);
683 clone->bi_destructor = dm_bio_destructor;
684 clone->bi_sector = sector;
686 clone->bi_vcnt = idx + bv_count;
687 clone->bi_size = to_bytes(len);
688 clone->bi_flags &= ~(1 << BIO_SEG_VALID);
693 static int __clone_and_map(struct clone_info *ci)
695 struct bio *clone, *bio = ci->bio;
696 struct dm_target *ti;
697 sector_t len = 0, max;
698 struct dm_target_io *tio;
700 ti = dm_table_find_target(ci->map, ci->sector);
701 if (!dm_target_is_valid(ti))
704 max = max_io_len(ci->md, ci->sector, ti);
707 * Allocate a target io object.
709 tio = alloc_tio(ci->md);
712 memset(&tio->info, 0, sizeof(tio->info));
714 if (ci->sector_count <= max) {
716 * Optimise for the simple case where we can do all of
717 * the remaining io with a single clone.
719 clone = clone_bio(bio, ci->sector, ci->idx,
720 bio->bi_vcnt - ci->idx, ci->sector_count,
722 __map_bio(ti, clone, tio);
723 ci->sector_count = 0;
725 } else if (to_sector(bio->bi_io_vec[ci->idx].bv_len) <= max) {
727 * There are some bvecs that don't span targets.
728 * Do as many of these as possible.
731 sector_t remaining = max;
734 for (i = ci->idx; remaining && (i < bio->bi_vcnt); i++) {
735 bv_len = to_sector(bio->bi_io_vec[i].bv_len);
737 if (bv_len > remaining)
744 clone = clone_bio(bio, ci->sector, ci->idx, i - ci->idx, len,
746 __map_bio(ti, clone, tio);
749 ci->sector_count -= len;
754 * Handle a bvec that must be split between two or more targets.
756 struct bio_vec *bv = bio->bi_io_vec + ci->idx;
757 sector_t remaining = to_sector(bv->bv_len);
758 unsigned int offset = 0;
762 ti = dm_table_find_target(ci->map, ci->sector);
763 if (!dm_target_is_valid(ti))
766 max = max_io_len(ci->md, ci->sector, ti);
768 tio = alloc_tio(ci->md);
771 memset(&tio->info, 0, sizeof(tio->info));
774 len = min(remaining, max);
776 clone = split_bvec(bio, ci->sector, ci->idx,
777 bv->bv_offset + offset, len,
780 __map_bio(ti, clone, tio);
783 ci->sector_count -= len;
784 offset += to_bytes(len);
785 } while (remaining -= len);
794 * Split the bio into several clones.
796 static int __split_bio(struct mapped_device *md, struct bio *bio)
798 struct clone_info ci;
801 ci.map = dm_get_table(md);
802 if (unlikely(!ci.map))
807 ci.io = alloc_io(md);
809 atomic_set(&ci.io->io_count, 1);
812 ci.sector = bio->bi_sector;
813 ci.sector_count = bio_sectors(bio);
814 ci.idx = bio->bi_idx;
816 start_io_acct(ci.io);
817 while (ci.sector_count && !error)
818 error = __clone_and_map(&ci);
820 /* drop the extra reference count */
821 dec_pending(ci.io, error);
822 dm_table_put(ci.map);
826 /*-----------------------------------------------------------------
828 *---------------------------------------------------------------*/
831 * The request function that just remaps the bio built up by
834 static int dm_request(struct request_queue *q, struct bio *bio)
837 int rw = bio_data_dir(bio);
838 struct mapped_device *md = q->queuedata;
841 * There is no use in forwarding any barrier request since we can't
842 * guarantee it is (or can be) handled by the targets correctly.
844 if (unlikely(bio_barrier(bio))) {
845 bio_endio(bio, -EOPNOTSUPP);
849 down_read(&md->io_lock);
851 disk_stat_inc(dm_disk(md), ios[rw]);
852 disk_stat_add(dm_disk(md), sectors[rw], bio_sectors(bio));
855 * If we're suspended we have to queue
858 while (test_bit(DMF_BLOCK_IO, &md->flags)) {
859 up_read(&md->io_lock);
861 if (bio_rw(bio) != READA)
862 r = queue_io(md, bio);
868 * We're in a while loop, because someone could suspend
869 * before we get to the following read lock.
871 down_read(&md->io_lock);
874 r = __split_bio(md, bio);
875 up_read(&md->io_lock);
884 static void dm_unplug_all(struct request_queue *q)
886 struct mapped_device *md = q->queuedata;
887 struct dm_table *map = dm_get_table(md);
890 dm_table_unplug_all(map);
895 static int dm_any_congested(void *congested_data, int bdi_bits)
898 struct mapped_device *md = (struct mapped_device *) congested_data;
899 struct dm_table *map = dm_get_table(md);
901 if (!map || test_bit(DMF_BLOCK_IO, &md->flags))
904 r = dm_table_any_congested(map, bdi_bits);
910 /*-----------------------------------------------------------------
911 * An IDR is used to keep track of allocated minor numbers.
912 *---------------------------------------------------------------*/
913 static DEFINE_IDR(_minor_idr);
915 static void free_minor(int minor)
917 spin_lock(&_minor_lock);
918 idr_remove(&_minor_idr, minor);
919 spin_unlock(&_minor_lock);
923 * See if the device with a specific minor # is free.
925 static int specific_minor(struct mapped_device *md, int minor)
929 if (minor >= (1 << MINORBITS))
932 r = idr_pre_get(&_minor_idr, GFP_KERNEL);
936 spin_lock(&_minor_lock);
938 if (idr_find(&_minor_idr, minor)) {
943 r = idr_get_new_above(&_minor_idr, MINOR_ALLOCED, minor, &m);
948 idr_remove(&_minor_idr, m);
954 spin_unlock(&_minor_lock);
958 static int next_free_minor(struct mapped_device *md, int *minor)
962 r = idr_pre_get(&_minor_idr, GFP_KERNEL);
966 spin_lock(&_minor_lock);
968 r = idr_get_new(&_minor_idr, MINOR_ALLOCED, &m);
973 if (m >= (1 << MINORBITS)) {
974 idr_remove(&_minor_idr, m);
982 spin_unlock(&_minor_lock);
986 static struct block_device_operations dm_blk_dops;
989 * Allocate and initialise a blank device with a given minor.
991 static struct mapped_device *alloc_dev(int minor)
994 struct mapped_device *md = kmalloc(sizeof(*md), GFP_KERNEL);
998 DMWARN("unable to allocate device, out of memory.");
1002 if (!try_module_get(THIS_MODULE))
1003 goto bad_module_get;
1005 /* get a minor number for the dev */
1006 if (minor == DM_ANY_MINOR)
1007 r = next_free_minor(md, &minor);
1009 r = specific_minor(md, minor);
1013 memset(md, 0, sizeof(*md));
1014 init_rwsem(&md->io_lock);
1015 mutex_init(&md->suspend_lock);
1016 spin_lock_init(&md->pushback_lock);
1017 rwlock_init(&md->map_lock);
1018 atomic_set(&md->holders, 1);
1019 atomic_set(&md->open_count, 0);
1020 atomic_set(&md->event_nr, 0);
1021 atomic_set(&md->uevent_seq, 0);
1022 INIT_LIST_HEAD(&md->uevent_list);
1023 spin_lock_init(&md->uevent_lock);
1025 md->queue = blk_alloc_queue(GFP_KERNEL);
1029 md->queue->queuedata = md;
1030 md->queue->backing_dev_info.congested_fn = dm_any_congested;
1031 md->queue->backing_dev_info.congested_data = md;
1032 blk_queue_make_request(md->queue, dm_request);
1033 blk_queue_bounce_limit(md->queue, BLK_BOUNCE_ANY);
1034 md->queue->unplug_fn = dm_unplug_all;
1036 md->io_pool = mempool_create_slab_pool(MIN_IOS, _io_cache);
1040 md->tio_pool = mempool_create_slab_pool(MIN_IOS, _tio_cache);
1044 md->bs = bioset_create(16, 16);
1048 md->disk = alloc_disk(1);
1052 atomic_set(&md->pending, 0);
1053 init_waitqueue_head(&md->wait);
1054 init_waitqueue_head(&md->eventq);
1056 md->disk->major = _major;
1057 md->disk->first_minor = minor;
1058 md->disk->fops = &dm_blk_dops;
1059 md->disk->queue = md->queue;
1060 md->disk->private_data = md;
1061 sprintf(md->disk->disk_name, "dm-%d", minor);
1063 format_dev_t(md->name, MKDEV(_major, minor));
1065 md->wq = create_singlethread_workqueue("kdmflush");
1069 /* Populate the mapping, nobody knows we exist yet */
1070 spin_lock(&_minor_lock);
1071 old_md = idr_replace(&_minor_idr, md, minor);
1072 spin_unlock(&_minor_lock);
1074 BUG_ON(old_md != MINOR_ALLOCED);
1081 bioset_free(md->bs);
1083 mempool_destroy(md->tio_pool);
1085 mempool_destroy(md->io_pool);
1087 blk_cleanup_queue(md->queue);
1091 module_put(THIS_MODULE);
1097 static void unlock_fs(struct mapped_device *md);
1099 static void free_dev(struct mapped_device *md)
1101 int minor = md->disk->first_minor;
1103 if (md->suspended_bdev) {
1105 bdput(md->suspended_bdev);
1107 destroy_workqueue(md->wq);
1108 mempool_destroy(md->tio_pool);
1109 mempool_destroy(md->io_pool);
1110 bioset_free(md->bs);
1111 del_gendisk(md->disk);
1114 spin_lock(&_minor_lock);
1115 md->disk->private_data = NULL;
1116 spin_unlock(&_minor_lock);
1119 blk_cleanup_queue(md->queue);
1120 module_put(THIS_MODULE);
1125 * Bind a table to the device.
1127 static void event_callback(void *context)
1129 unsigned long flags;
1131 struct mapped_device *md = (struct mapped_device *) context;
1133 spin_lock_irqsave(&md->uevent_lock, flags);
1134 list_splice_init(&md->uevent_list, &uevents);
1135 spin_unlock_irqrestore(&md->uevent_lock, flags);
1137 dm_send_uevents(&uevents, &md->disk->dev.kobj);
1139 atomic_inc(&md->event_nr);
1140 wake_up(&md->eventq);
1143 static void __set_size(struct mapped_device *md, sector_t size)
1145 set_capacity(md->disk, size);
1147 mutex_lock(&md->suspended_bdev->bd_inode->i_mutex);
1148 i_size_write(md->suspended_bdev->bd_inode, (loff_t)size << SECTOR_SHIFT);
1149 mutex_unlock(&md->suspended_bdev->bd_inode->i_mutex);
1152 static int __bind(struct mapped_device *md, struct dm_table *t)
1154 struct request_queue *q = md->queue;
1157 size = dm_table_get_size(t);
1160 * Wipe any geometry if the size of the table changed.
1162 if (size != get_capacity(md->disk))
1163 memset(&md->geometry, 0, sizeof(md->geometry));
1165 if (md->suspended_bdev)
1166 __set_size(md, size);
1171 dm_table_event_callback(t, event_callback, md);
1173 write_lock(&md->map_lock);
1175 dm_table_set_restrictions(t, q);
1176 write_unlock(&md->map_lock);
1181 static void __unbind(struct mapped_device *md)
1183 struct dm_table *map = md->map;
1188 dm_table_event_callback(map, NULL, NULL);
1189 write_lock(&md->map_lock);
1191 write_unlock(&md->map_lock);
1196 * Constructor for a new device.
1198 int dm_create(int minor, struct mapped_device **result)
1200 struct mapped_device *md;
1202 md = alloc_dev(minor);
1210 static struct mapped_device *dm_find_md(dev_t dev)
1212 struct mapped_device *md;
1213 unsigned minor = MINOR(dev);
1215 if (MAJOR(dev) != _major || minor >= (1 << MINORBITS))
1218 spin_lock(&_minor_lock);
1220 md = idr_find(&_minor_idr, minor);
1221 if (md && (md == MINOR_ALLOCED ||
1222 (dm_disk(md)->first_minor != minor) ||
1223 test_bit(DMF_FREEING, &md->flags))) {
1229 spin_unlock(&_minor_lock);
1234 struct mapped_device *dm_get_md(dev_t dev)
1236 struct mapped_device *md = dm_find_md(dev);
1244 void *dm_get_mdptr(struct mapped_device *md)
1246 return md->interface_ptr;
1249 void dm_set_mdptr(struct mapped_device *md, void *ptr)
1251 md->interface_ptr = ptr;
1254 void dm_get(struct mapped_device *md)
1256 atomic_inc(&md->holders);
1259 const char *dm_device_name(struct mapped_device *md)
1263 EXPORT_SYMBOL_GPL(dm_device_name);
1265 void dm_put(struct mapped_device *md)
1267 struct dm_table *map;
1269 BUG_ON(test_bit(DMF_FREEING, &md->flags));
1271 if (atomic_dec_and_lock(&md->holders, &_minor_lock)) {
1272 map = dm_get_table(md);
1273 idr_replace(&_minor_idr, MINOR_ALLOCED, dm_disk(md)->first_minor);
1274 set_bit(DMF_FREEING, &md->flags);
1275 spin_unlock(&_minor_lock);
1276 if (!dm_suspended(md)) {
1277 dm_table_presuspend_targets(map);
1278 dm_table_postsuspend_targets(map);
1285 EXPORT_SYMBOL_GPL(dm_put);
1287 static int dm_wait_for_completion(struct mapped_device *md)
1292 set_current_state(TASK_INTERRUPTIBLE);
1295 if (!atomic_read(&md->pending))
1298 if (signal_pending(current)) {
1305 set_current_state(TASK_RUNNING);
1311 * Process the deferred bios
1313 static void __flush_deferred_io(struct mapped_device *md)
1317 while ((c = bio_list_pop(&md->deferred))) {
1318 if (__split_bio(md, c))
1322 clear_bit(DMF_BLOCK_IO, &md->flags);
1325 static void __merge_pushback_list(struct mapped_device *md)
1327 unsigned long flags;
1329 spin_lock_irqsave(&md->pushback_lock, flags);
1330 clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
1331 bio_list_merge_head(&md->deferred, &md->pushback);
1332 bio_list_init(&md->pushback);
1333 spin_unlock_irqrestore(&md->pushback_lock, flags);
1336 static void dm_wq_work(struct work_struct *work)
1338 struct dm_wq_req *req = container_of(work, struct dm_wq_req, work);
1339 struct mapped_device *md = req->md;
1341 down_write(&md->io_lock);
1342 switch (req->type) {
1343 case DM_WQ_FLUSH_ALL:
1344 __merge_pushback_list(md);
1346 case DM_WQ_FLUSH_DEFERRED:
1347 __flush_deferred_io(md);
1350 DMERR("dm_wq_work: unrecognised work type %d", req->type);
1353 up_write(&md->io_lock);
1356 static void dm_wq_queue(struct mapped_device *md, int type, void *context,
1357 struct dm_wq_req *req)
1361 req->context = context;
1362 INIT_WORK(&req->work, dm_wq_work);
1363 queue_work(md->wq, &req->work);
1366 static void dm_queue_flush(struct mapped_device *md, int type, void *context)
1368 struct dm_wq_req req;
1370 dm_wq_queue(md, type, context, &req);
1371 flush_workqueue(md->wq);
1375 * Swap in a new table (destroying old one).
1377 int dm_swap_table(struct mapped_device *md, struct dm_table *table)
1381 mutex_lock(&md->suspend_lock);
1383 /* device must be suspended */
1384 if (!dm_suspended(md))
1387 /* without bdev, the device size cannot be changed */
1388 if (!md->suspended_bdev)
1389 if (get_capacity(md->disk) != dm_table_get_size(table))
1393 r = __bind(md, table);
1396 mutex_unlock(&md->suspend_lock);
1401 * Functions to lock and unlock any filesystem running on the
1404 static int lock_fs(struct mapped_device *md)
1408 WARN_ON(md->frozen_sb);
1410 md->frozen_sb = freeze_bdev(md->suspended_bdev);
1411 if (IS_ERR(md->frozen_sb)) {
1412 r = PTR_ERR(md->frozen_sb);
1413 md->frozen_sb = NULL;
1417 set_bit(DMF_FROZEN, &md->flags);
1419 /* don't bdput right now, we don't want the bdev
1420 * to go away while it is locked.
1425 static void unlock_fs(struct mapped_device *md)
1427 if (!test_bit(DMF_FROZEN, &md->flags))
1430 thaw_bdev(md->suspended_bdev, md->frozen_sb);
1431 md->frozen_sb = NULL;
1432 clear_bit(DMF_FROZEN, &md->flags);
1436 * We need to be able to change a mapping table under a mounted
1437 * filesystem. For example we might want to move some data in
1438 * the background. Before the table can be swapped with
1439 * dm_bind_table, dm_suspend must be called to flush any in
1440 * flight bios and ensure that any further io gets deferred.
1442 int dm_suspend(struct mapped_device *md, unsigned suspend_flags)
1444 struct dm_table *map = NULL;
1445 DECLARE_WAITQUEUE(wait, current);
1447 int do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG ? 1 : 0;
1448 int noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG ? 1 : 0;
1450 mutex_lock(&md->suspend_lock);
1452 if (dm_suspended(md)) {
1457 map = dm_get_table(md);
1460 * DMF_NOFLUSH_SUSPENDING must be set before presuspend.
1461 * This flag is cleared before dm_suspend returns.
1464 set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
1466 /* This does not get reverted if there's an error later. */
1467 dm_table_presuspend_targets(map);
1469 /* bdget() can stall if the pending I/Os are not flushed */
1471 md->suspended_bdev = bdget_disk(md->disk, 0);
1472 if (!md->suspended_bdev) {
1473 DMWARN("bdget failed in dm_suspend");
1479 * Flush I/O to the device. noflush supersedes do_lockfs,
1480 * because lock_fs() needs to flush I/Os.
1490 * First we set the BLOCK_IO flag so no more ios will be mapped.
1492 down_write(&md->io_lock);
1493 set_bit(DMF_BLOCK_IO, &md->flags);
1495 add_wait_queue(&md->wait, &wait);
1496 up_write(&md->io_lock);
1500 dm_table_unplug_all(map);
1503 * Wait for the already-mapped ios to complete.
1505 r = dm_wait_for_completion(md);
1507 down_write(&md->io_lock);
1508 remove_wait_queue(&md->wait, &wait);
1511 __merge_pushback_list(md);
1512 up_write(&md->io_lock);
1514 /* were we interrupted ? */
1516 dm_queue_flush(md, DM_WQ_FLUSH_DEFERRED, NULL);
1519 goto out; /* pushback list is already flushed, so skip flush */
1522 dm_table_postsuspend_targets(map);
1524 set_bit(DMF_SUSPENDED, &md->flags);
1529 * Because there may be already I/Os in the pushback list,
1530 * flush them before return.
1532 dm_queue_flush(md, DM_WQ_FLUSH_ALL, NULL);
1535 if (r && md->suspended_bdev) {
1536 bdput(md->suspended_bdev);
1537 md->suspended_bdev = NULL;
1543 mutex_unlock(&md->suspend_lock);
1547 int dm_resume(struct mapped_device *md)
1550 struct dm_table *map = NULL;
1552 mutex_lock(&md->suspend_lock);
1553 if (!dm_suspended(md))
1556 map = dm_get_table(md);
1557 if (!map || !dm_table_get_size(map))
1560 r = dm_table_resume_targets(map);
1564 dm_queue_flush(md, DM_WQ_FLUSH_DEFERRED, NULL);
1568 if (md->suspended_bdev) {
1569 bdput(md->suspended_bdev);
1570 md->suspended_bdev = NULL;
1573 clear_bit(DMF_SUSPENDED, &md->flags);
1575 dm_table_unplug_all(map);
1577 dm_kobject_uevent(md);
1583 mutex_unlock(&md->suspend_lock);
1588 /*-----------------------------------------------------------------
1589 * Event notification.
1590 *---------------------------------------------------------------*/
1591 void dm_kobject_uevent(struct mapped_device *md)
1593 kobject_uevent(&md->disk->dev.kobj, KOBJ_CHANGE);
1596 uint32_t dm_next_uevent_seq(struct mapped_device *md)
1598 return atomic_add_return(1, &md->uevent_seq);
1601 uint32_t dm_get_event_nr(struct mapped_device *md)
1603 return atomic_read(&md->event_nr);
1606 int dm_wait_event(struct mapped_device *md, int event_nr)
1608 return wait_event_interruptible(md->eventq,
1609 (event_nr != atomic_read(&md->event_nr)));
1612 void dm_uevent_add(struct mapped_device *md, struct list_head *elist)
1614 unsigned long flags;
1616 spin_lock_irqsave(&md->uevent_lock, flags);
1617 list_add(elist, &md->uevent_list);
1618 spin_unlock_irqrestore(&md->uevent_lock, flags);
1622 * The gendisk is only valid as long as you have a reference
1625 struct gendisk *dm_disk(struct mapped_device *md)
1630 int dm_suspended(struct mapped_device *md)
1632 return test_bit(DMF_SUSPENDED, &md->flags);
1635 int dm_noflush_suspending(struct dm_target *ti)
1637 struct mapped_device *md = dm_table_get_md(ti->table);
1638 int r = __noflush_suspending(md);
1644 EXPORT_SYMBOL_GPL(dm_noflush_suspending);
1646 static struct block_device_operations dm_blk_dops = {
1647 .open = dm_blk_open,
1648 .release = dm_blk_close,
1649 .ioctl = dm_blk_ioctl,
1650 .getgeo = dm_blk_getgeo,
1651 .owner = THIS_MODULE
1654 EXPORT_SYMBOL(dm_get_mapinfo);
1659 module_init(dm_init);
1660 module_exit(dm_exit);
1662 module_param(major, uint, 0);
1663 MODULE_PARM_DESC(major, "The major number of the device mapper");
1664 MODULE_DESCRIPTION(DM_NAME " driver");
1665 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
1666 MODULE_LICENSE("GPL");