2 * Copyright (C) 2001, 2002 Sistina Software (UK) Limited.
3 * Copyright (C) 2004-2008 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 <trace/block.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);
36 * One of these is allocated per bio.
39 struct mapped_device *md;
43 unsigned long start_time;
48 * One of these is allocated per target within a bio. Hopefully
49 * this will be simplified out one day.
57 DEFINE_TRACE(block_bio_complete);
60 * For request-based dm.
61 * One of these is allocated per request.
63 struct dm_rq_target_io {
64 struct mapped_device *md;
66 struct request *orig, clone;
72 * For request-based dm.
73 * One of these is allocated per bio.
75 struct dm_rq_clone_bio_info {
80 union map_info *dm_get_mapinfo(struct bio *bio)
82 if (bio && bio->bi_private)
83 return &((struct dm_target_io *)bio->bi_private)->info;
87 #define MINOR_ALLOCED ((void *)-1)
90 * Bits for the md->flags field.
92 #define DMF_BLOCK_IO 0
93 #define DMF_SUSPENDED 1
96 #define DMF_DELETING 4
97 #define DMF_NOFLUSH_SUSPENDING 5
100 * Work processed by per-device workqueue.
104 DM_WQ_FLUSH_DEFERRED,
106 struct work_struct work;
107 struct mapped_device *md;
111 struct mapped_device {
112 struct rw_semaphore io_lock;
113 struct mutex suspend_lock;
114 spinlock_t pushback_lock;
121 struct request_queue *queue;
122 struct gendisk *disk;
128 * A list of ios that arrived while we were suspended.
131 wait_queue_head_t wait;
132 struct bio_list deferred;
133 struct bio_list pushback;
136 * Processing queue (flush/barriers)
138 struct workqueue_struct *wq;
141 * The current mapping.
143 struct dm_table *map;
146 * io objects are allocated from here.
157 wait_queue_head_t eventq;
159 struct list_head uevent_list;
160 spinlock_t uevent_lock; /* Protect access to uevent_list */
163 * freeze/thaw support require holding onto a super block
165 struct super_block *frozen_sb;
166 struct block_device *suspended_bdev;
168 /* forced geometry settings */
169 struct hd_geometry geometry;
176 static struct kmem_cache *_io_cache;
177 static struct kmem_cache *_tio_cache;
178 static struct kmem_cache *_rq_tio_cache;
179 static struct kmem_cache *_rq_bio_info_cache;
181 static int __init local_init(void)
185 /* allocate a slab for the dm_ios */
186 _io_cache = KMEM_CACHE(dm_io, 0);
190 /* allocate a slab for the target ios */
191 _tio_cache = KMEM_CACHE(dm_target_io, 0);
193 goto out_free_io_cache;
195 _rq_tio_cache = KMEM_CACHE(dm_rq_target_io, 0);
197 goto out_free_tio_cache;
199 _rq_bio_info_cache = KMEM_CACHE(dm_rq_clone_bio_info, 0);
200 if (!_rq_bio_info_cache)
201 goto out_free_rq_tio_cache;
203 r = dm_uevent_init();
205 goto out_free_rq_bio_info_cache;
208 r = register_blkdev(_major, _name);
210 goto out_uevent_exit;
219 out_free_rq_bio_info_cache:
220 kmem_cache_destroy(_rq_bio_info_cache);
221 out_free_rq_tio_cache:
222 kmem_cache_destroy(_rq_tio_cache);
224 kmem_cache_destroy(_tio_cache);
226 kmem_cache_destroy(_io_cache);
231 static void local_exit(void)
233 kmem_cache_destroy(_rq_bio_info_cache);
234 kmem_cache_destroy(_rq_tio_cache);
235 kmem_cache_destroy(_tio_cache);
236 kmem_cache_destroy(_io_cache);
237 unregister_blkdev(_major, _name);
242 DMINFO("cleaned up");
245 static int (*_inits[])(void) __initdata = {
254 static void (*_exits[])(void) = {
263 static int __init dm_init(void)
265 const int count = ARRAY_SIZE(_inits);
269 for (i = 0; i < count; i++) {
284 static void __exit dm_exit(void)
286 int i = ARRAY_SIZE(_exits);
293 * Block device functions
295 static int dm_blk_open(struct block_device *bdev, fmode_t mode)
297 struct mapped_device *md;
299 spin_lock(&_minor_lock);
301 md = bdev->bd_disk->private_data;
305 if (test_bit(DMF_FREEING, &md->flags) ||
306 test_bit(DMF_DELETING, &md->flags)) {
312 atomic_inc(&md->open_count);
315 spin_unlock(&_minor_lock);
317 return md ? 0 : -ENXIO;
320 static int dm_blk_close(struct gendisk *disk, fmode_t mode)
322 struct mapped_device *md = disk->private_data;
323 atomic_dec(&md->open_count);
328 int dm_open_count(struct mapped_device *md)
330 return atomic_read(&md->open_count);
334 * Guarantees nothing is using the device before it's deleted.
336 int dm_lock_for_deletion(struct mapped_device *md)
340 spin_lock(&_minor_lock);
342 if (dm_open_count(md))
345 set_bit(DMF_DELETING, &md->flags);
347 spin_unlock(&_minor_lock);
352 static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo)
354 struct mapped_device *md = bdev->bd_disk->private_data;
356 return dm_get_geometry(md, geo);
359 static int dm_blk_ioctl(struct block_device *bdev, fmode_t mode,
360 unsigned int cmd, unsigned long arg)
362 struct mapped_device *md = bdev->bd_disk->private_data;
363 struct dm_table *map = dm_get_table(md);
364 struct dm_target *tgt;
367 if (!map || !dm_table_get_size(map))
370 /* We only support devices that have a single target */
371 if (dm_table_get_num_targets(map) != 1)
374 tgt = dm_table_get_target(map, 0);
376 if (dm_suspended(md)) {
381 if (tgt->type->ioctl)
382 r = tgt->type->ioctl(tgt, cmd, arg);
390 static struct dm_io *alloc_io(struct mapped_device *md)
392 return mempool_alloc(md->io_pool, GFP_NOIO);
395 static void free_io(struct mapped_device *md, struct dm_io *io)
397 mempool_free(io, md->io_pool);
400 static struct dm_target_io *alloc_tio(struct mapped_device *md)
402 return mempool_alloc(md->tio_pool, GFP_NOIO);
405 static void free_tio(struct mapped_device *md, struct dm_target_io *tio)
407 mempool_free(tio, md->tio_pool);
410 static void start_io_acct(struct dm_io *io)
412 struct mapped_device *md = io->md;
415 io->start_time = jiffies;
417 cpu = part_stat_lock();
418 part_round_stats(cpu, &dm_disk(md)->part0);
420 dm_disk(md)->part0.in_flight = atomic_inc_return(&md->pending);
423 static void end_io_acct(struct dm_io *io)
425 struct mapped_device *md = io->md;
426 struct bio *bio = io->bio;
427 unsigned long duration = jiffies - io->start_time;
429 int rw = bio_data_dir(bio);
431 cpu = part_stat_lock();
432 part_round_stats(cpu, &dm_disk(md)->part0);
433 part_stat_add(cpu, &dm_disk(md)->part0, ticks[rw], duration);
436 dm_disk(md)->part0.in_flight = pending =
437 atomic_dec_return(&md->pending);
439 /* nudge anyone waiting on suspend queue */
445 * Add the bio to the list of deferred io.
447 static int queue_io(struct mapped_device *md, struct bio *bio)
449 down_write(&md->io_lock);
451 if (!test_bit(DMF_BLOCK_IO, &md->flags)) {
452 up_write(&md->io_lock);
456 bio_list_add(&md->deferred, bio);
458 up_write(&md->io_lock);
459 return 0; /* deferred successfully */
463 * Everyone (including functions in this file), should use this
464 * function to access the md->map field, and make sure they call
465 * dm_table_put() when finished.
467 struct dm_table *dm_get_table(struct mapped_device *md)
471 read_lock(&md->map_lock);
475 read_unlock(&md->map_lock);
481 * Get the geometry associated with a dm device
483 int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo)
491 * Set the geometry of a device.
493 int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo)
495 sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors;
497 if (geo->start > sz) {
498 DMWARN("Start sector is beyond the geometry limits.");
507 /*-----------------------------------------------------------------
509 * A more elegant soln is in the works that uses the queue
510 * merge fn, unfortunately there are a couple of changes to
511 * the block layer that I want to make for this. So in the
512 * interests of getting something for people to use I give
513 * you this clearly demarcated crap.
514 *---------------------------------------------------------------*/
516 static int __noflush_suspending(struct mapped_device *md)
518 return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
522 * Decrements the number of outstanding ios that a bio has been
523 * cloned into, completing the original io if necc.
525 static void dec_pending(struct dm_io *io, int error)
530 struct mapped_device *md = io->md;
532 /* Push-back supersedes any I/O errors */
533 if (error && !(io->error > 0 && __noflush_suspending(md)))
536 if (atomic_dec_and_test(&io->io_count)) {
537 if (io->error == DM_ENDIO_REQUEUE) {
539 * Target requested pushing back the I/O.
540 * This must be handled before the sleeper on
541 * suspend queue merges the pushback list.
543 spin_lock_irqsave(&md->pushback_lock, flags);
544 if (__noflush_suspending(md))
545 bio_list_add(&md->pushback, io->bio);
547 /* noflush suspend was interrupted. */
549 spin_unlock_irqrestore(&md->pushback_lock, flags);
554 io_error = io->error;
559 if (io_error != DM_ENDIO_REQUEUE) {
560 trace_block_bio_complete(md->queue, bio);
562 bio_endio(bio, io_error);
567 static void clone_endio(struct bio *bio, int error)
570 struct dm_target_io *tio = bio->bi_private;
571 struct dm_io *io = tio->io;
572 struct mapped_device *md = tio->io->md;
573 dm_endio_fn endio = tio->ti->type->end_io;
575 if (!bio_flagged(bio, BIO_UPTODATE) && !error)
579 r = endio(tio->ti, bio, error, &tio->info);
580 if (r < 0 || r == DM_ENDIO_REQUEUE)
582 * error and requeue request are handled
586 else if (r == DM_ENDIO_INCOMPLETE)
587 /* The target will handle the io */
590 DMWARN("unimplemented target endio return value: %d", r);
596 * Store md for cleanup instead of tio which is about to get freed.
598 bio->bi_private = md->bs;
602 dec_pending(io, error);
605 static sector_t max_io_len(struct mapped_device *md,
606 sector_t sector, struct dm_target *ti)
608 sector_t offset = sector - ti->begin;
609 sector_t len = ti->len - offset;
612 * Does the target need to split even further ?
616 boundary = ((offset + ti->split_io) & ~(ti->split_io - 1))
625 static void __map_bio(struct dm_target *ti, struct bio *clone,
626 struct dm_target_io *tio)
630 struct mapped_device *md;
635 BUG_ON(!clone->bi_size);
637 clone->bi_end_io = clone_endio;
638 clone->bi_private = tio;
641 * Map the clone. If r == 0 we don't need to do
642 * anything, the target has assumed ownership of
645 atomic_inc(&tio->io->io_count);
646 sector = clone->bi_sector;
647 r = ti->type->map(ti, clone, &tio->info);
648 if (r == DM_MAPIO_REMAPPED) {
649 /* the bio has been remapped so dispatch it */
651 trace_block_remap(bdev_get_queue(clone->bi_bdev), clone,
652 tio->io->bio->bi_bdev->bd_dev,
653 clone->bi_sector, sector);
655 generic_make_request(clone);
656 } else if (r < 0 || r == DM_MAPIO_REQUEUE) {
657 /* error the io and bail out, or requeue it if needed */
659 dec_pending(tio->io, r);
661 * Store bio_set for cleanup.
663 clone->bi_private = md->bs;
667 DMWARN("unimplemented target map return value: %d", r);
673 struct mapped_device *md;
674 struct dm_table *map;
678 sector_t sector_count;
682 static void dm_bio_destructor(struct bio *bio)
684 struct bio_set *bs = bio->bi_private;
690 * Creates a little bio that is just does part of a bvec.
692 static struct bio *split_bvec(struct bio *bio, sector_t sector,
693 unsigned short idx, unsigned int offset,
694 unsigned int len, struct bio_set *bs)
697 struct bio_vec *bv = bio->bi_io_vec + idx;
699 clone = bio_alloc_bioset(GFP_NOIO, 1, bs);
700 clone->bi_destructor = dm_bio_destructor;
701 *clone->bi_io_vec = *bv;
703 clone->bi_sector = sector;
704 clone->bi_bdev = bio->bi_bdev;
705 clone->bi_rw = bio->bi_rw;
707 clone->bi_size = to_bytes(len);
708 clone->bi_io_vec->bv_offset = offset;
709 clone->bi_io_vec->bv_len = clone->bi_size;
710 clone->bi_flags |= 1 << BIO_CLONED;
716 * Creates a bio that consists of range of complete bvecs.
718 static struct bio *clone_bio(struct bio *bio, sector_t sector,
719 unsigned short idx, unsigned short bv_count,
720 unsigned int len, struct bio_set *bs)
724 clone = bio_alloc_bioset(GFP_NOIO, bio->bi_max_vecs, bs);
725 __bio_clone(clone, bio);
726 clone->bi_destructor = dm_bio_destructor;
727 clone->bi_sector = sector;
729 clone->bi_vcnt = idx + bv_count;
730 clone->bi_size = to_bytes(len);
731 clone->bi_flags &= ~(1 << BIO_SEG_VALID);
736 static int __clone_and_map(struct clone_info *ci)
738 struct bio *clone, *bio = ci->bio;
739 struct dm_target *ti;
740 sector_t len = 0, max;
741 struct dm_target_io *tio;
743 ti = dm_table_find_target(ci->map, ci->sector);
744 if (!dm_target_is_valid(ti))
747 max = max_io_len(ci->md, ci->sector, ti);
750 * Allocate a target io object.
752 tio = alloc_tio(ci->md);
755 memset(&tio->info, 0, sizeof(tio->info));
757 if (ci->sector_count <= max) {
759 * Optimise for the simple case where we can do all of
760 * the remaining io with a single clone.
762 clone = clone_bio(bio, ci->sector, ci->idx,
763 bio->bi_vcnt - ci->idx, ci->sector_count,
765 __map_bio(ti, clone, tio);
766 ci->sector_count = 0;
768 } else if (to_sector(bio->bi_io_vec[ci->idx].bv_len) <= max) {
770 * There are some bvecs that don't span targets.
771 * Do as many of these as possible.
774 sector_t remaining = max;
777 for (i = ci->idx; remaining && (i < bio->bi_vcnt); i++) {
778 bv_len = to_sector(bio->bi_io_vec[i].bv_len);
780 if (bv_len > remaining)
787 clone = clone_bio(bio, ci->sector, ci->idx, i - ci->idx, len,
789 __map_bio(ti, clone, tio);
792 ci->sector_count -= len;
797 * Handle a bvec that must be split between two or more targets.
799 struct bio_vec *bv = bio->bi_io_vec + ci->idx;
800 sector_t remaining = to_sector(bv->bv_len);
801 unsigned int offset = 0;
805 ti = dm_table_find_target(ci->map, ci->sector);
806 if (!dm_target_is_valid(ti))
809 max = max_io_len(ci->md, ci->sector, ti);
811 tio = alloc_tio(ci->md);
814 memset(&tio->info, 0, sizeof(tio->info));
817 len = min(remaining, max);
819 clone = split_bvec(bio, ci->sector, ci->idx,
820 bv->bv_offset + offset, len,
823 __map_bio(ti, clone, tio);
826 ci->sector_count -= len;
827 offset += to_bytes(len);
828 } while (remaining -= len);
837 * Split the bio into several clones.
839 static int __split_bio(struct mapped_device *md, struct bio *bio)
841 struct clone_info ci;
844 ci.map = dm_get_table(md);
845 if (unlikely(!ci.map))
847 if (unlikely(bio_barrier(bio) && !dm_table_barrier_ok(ci.map))) {
848 dm_table_put(ci.map);
849 bio_endio(bio, -EOPNOTSUPP);
854 ci.io = alloc_io(md);
856 atomic_set(&ci.io->io_count, 1);
859 ci.sector = bio->bi_sector;
860 ci.sector_count = bio_sectors(bio);
861 ci.idx = bio->bi_idx;
863 start_io_acct(ci.io);
864 while (ci.sector_count && !error)
865 error = __clone_and_map(&ci);
867 /* drop the extra reference count */
868 dec_pending(ci.io, error);
869 dm_table_put(ci.map);
873 /*-----------------------------------------------------------------
875 *---------------------------------------------------------------*/
877 static int dm_merge_bvec(struct request_queue *q,
878 struct bvec_merge_data *bvm,
879 struct bio_vec *biovec)
881 struct mapped_device *md = q->queuedata;
882 struct dm_table *map = dm_get_table(md);
883 struct dm_target *ti;
884 sector_t max_sectors;
890 ti = dm_table_find_target(map, bvm->bi_sector);
891 if (!dm_target_is_valid(ti))
895 * Find maximum amount of I/O that won't need splitting
897 max_sectors = min(max_io_len(md, bvm->bi_sector, ti),
898 (sector_t) BIO_MAX_SECTORS);
899 max_size = (max_sectors << SECTOR_SHIFT) - bvm->bi_size;
904 * merge_bvec_fn() returns number of bytes
905 * it can accept at this offset
906 * max is precomputed maximal io size
908 if (max_size && ti->type->merge)
909 max_size = ti->type->merge(ti, bvm, biovec, max_size);
916 * Always allow an entire first page
918 if (max_size <= biovec->bv_len && !(bvm->bi_size >> SECTOR_SHIFT))
919 max_size = biovec->bv_len;
925 * The request function that just remaps the bio built up by
928 static int dm_request(struct request_queue *q, struct bio *bio)
931 int rw = bio_data_dir(bio);
932 struct mapped_device *md = q->queuedata;
935 down_read(&md->io_lock);
937 cpu = part_stat_lock();
938 part_stat_inc(cpu, &dm_disk(md)->part0, ios[rw]);
939 part_stat_add(cpu, &dm_disk(md)->part0, sectors[rw], bio_sectors(bio));
943 * If we're suspended we have to queue
946 while (test_bit(DMF_BLOCK_IO, &md->flags)) {
947 up_read(&md->io_lock);
949 if (bio_rw(bio) != READA)
950 r = queue_io(md, bio);
956 * We're in a while loop, because someone could suspend
957 * before we get to the following read lock.
959 down_read(&md->io_lock);
962 r = __split_bio(md, bio);
963 up_read(&md->io_lock);
972 static void dm_unplug_all(struct request_queue *q)
974 struct mapped_device *md = q->queuedata;
975 struct dm_table *map = dm_get_table(md);
978 dm_table_unplug_all(map);
983 static int dm_any_congested(void *congested_data, int bdi_bits)
986 struct mapped_device *md = congested_data;
987 struct dm_table *map;
989 if (!test_bit(DMF_BLOCK_IO, &md->flags)) {
990 map = dm_get_table(md);
992 r = dm_table_any_congested(map, bdi_bits);
1000 /*-----------------------------------------------------------------
1001 * An IDR is used to keep track of allocated minor numbers.
1002 *---------------------------------------------------------------*/
1003 static DEFINE_IDR(_minor_idr);
1005 static void free_minor(int minor)
1007 spin_lock(&_minor_lock);
1008 idr_remove(&_minor_idr, minor);
1009 spin_unlock(&_minor_lock);
1013 * See if the device with a specific minor # is free.
1015 static int specific_minor(int minor)
1019 if (minor >= (1 << MINORBITS))
1022 r = idr_pre_get(&_minor_idr, GFP_KERNEL);
1026 spin_lock(&_minor_lock);
1028 if (idr_find(&_minor_idr, minor)) {
1033 r = idr_get_new_above(&_minor_idr, MINOR_ALLOCED, minor, &m);
1038 idr_remove(&_minor_idr, m);
1044 spin_unlock(&_minor_lock);
1048 static int next_free_minor(int *minor)
1052 r = idr_pre_get(&_minor_idr, GFP_KERNEL);
1056 spin_lock(&_minor_lock);
1058 r = idr_get_new(&_minor_idr, MINOR_ALLOCED, &m);
1062 if (m >= (1 << MINORBITS)) {
1063 idr_remove(&_minor_idr, m);
1071 spin_unlock(&_minor_lock);
1075 static struct block_device_operations dm_blk_dops;
1078 * Allocate and initialise a blank device with a given minor.
1080 static struct mapped_device *alloc_dev(int minor)
1083 struct mapped_device *md = kzalloc(sizeof(*md), GFP_KERNEL);
1087 DMWARN("unable to allocate device, out of memory.");
1091 if (!try_module_get(THIS_MODULE))
1092 goto bad_module_get;
1094 /* get a minor number for the dev */
1095 if (minor == DM_ANY_MINOR)
1096 r = next_free_minor(&minor);
1098 r = specific_minor(minor);
1102 init_rwsem(&md->io_lock);
1103 mutex_init(&md->suspend_lock);
1104 spin_lock_init(&md->pushback_lock);
1105 rwlock_init(&md->map_lock);
1106 atomic_set(&md->holders, 1);
1107 atomic_set(&md->open_count, 0);
1108 atomic_set(&md->event_nr, 0);
1109 atomic_set(&md->uevent_seq, 0);
1110 INIT_LIST_HEAD(&md->uevent_list);
1111 spin_lock_init(&md->uevent_lock);
1113 md->queue = blk_alloc_queue(GFP_KERNEL);
1117 md->queue->queuedata = md;
1118 md->queue->backing_dev_info.congested_fn = dm_any_congested;
1119 md->queue->backing_dev_info.congested_data = md;
1120 blk_queue_make_request(md->queue, dm_request);
1121 blk_queue_bounce_limit(md->queue, BLK_BOUNCE_ANY);
1122 md->queue->unplug_fn = dm_unplug_all;
1123 blk_queue_merge_bvec(md->queue, dm_merge_bvec);
1125 md->io_pool = mempool_create_slab_pool(MIN_IOS, _io_cache);
1129 md->tio_pool = mempool_create_slab_pool(MIN_IOS, _tio_cache);
1133 md->bs = bioset_create(16, 0);
1137 md->disk = alloc_disk(1);
1141 atomic_set(&md->pending, 0);
1142 init_waitqueue_head(&md->wait);
1143 init_waitqueue_head(&md->eventq);
1145 md->disk->major = _major;
1146 md->disk->first_minor = minor;
1147 md->disk->fops = &dm_blk_dops;
1148 md->disk->queue = md->queue;
1149 md->disk->private_data = md;
1150 sprintf(md->disk->disk_name, "dm-%d", minor);
1152 format_dev_t(md->name, MKDEV(_major, minor));
1154 md->wq = create_singlethread_workqueue("kdmflush");
1158 /* Populate the mapping, nobody knows we exist yet */
1159 spin_lock(&_minor_lock);
1160 old_md = idr_replace(&_minor_idr, md, minor);
1161 spin_unlock(&_minor_lock);
1163 BUG_ON(old_md != MINOR_ALLOCED);
1170 bioset_free(md->bs);
1172 mempool_destroy(md->tio_pool);
1174 mempool_destroy(md->io_pool);
1176 blk_cleanup_queue(md->queue);
1180 module_put(THIS_MODULE);
1186 static void unlock_fs(struct mapped_device *md);
1188 static void free_dev(struct mapped_device *md)
1190 int minor = MINOR(disk_devt(md->disk));
1192 if (md->suspended_bdev) {
1194 bdput(md->suspended_bdev);
1196 destroy_workqueue(md->wq);
1197 mempool_destroy(md->tio_pool);
1198 mempool_destroy(md->io_pool);
1199 bioset_free(md->bs);
1200 del_gendisk(md->disk);
1203 spin_lock(&_minor_lock);
1204 md->disk->private_data = NULL;
1205 spin_unlock(&_minor_lock);
1208 blk_cleanup_queue(md->queue);
1209 module_put(THIS_MODULE);
1214 * Bind a table to the device.
1216 static void event_callback(void *context)
1218 unsigned long flags;
1220 struct mapped_device *md = (struct mapped_device *) context;
1222 spin_lock_irqsave(&md->uevent_lock, flags);
1223 list_splice_init(&md->uevent_list, &uevents);
1224 spin_unlock_irqrestore(&md->uevent_lock, flags);
1226 dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj);
1228 atomic_inc(&md->event_nr);
1229 wake_up(&md->eventq);
1232 static void __set_size(struct mapped_device *md, sector_t size)
1234 set_capacity(md->disk, size);
1236 mutex_lock(&md->suspended_bdev->bd_inode->i_mutex);
1237 i_size_write(md->suspended_bdev->bd_inode, (loff_t)size << SECTOR_SHIFT);
1238 mutex_unlock(&md->suspended_bdev->bd_inode->i_mutex);
1241 static int __bind(struct mapped_device *md, struct dm_table *t)
1243 struct request_queue *q = md->queue;
1246 size = dm_table_get_size(t);
1249 * Wipe any geometry if the size of the table changed.
1251 if (size != get_capacity(md->disk))
1252 memset(&md->geometry, 0, sizeof(md->geometry));
1254 if (md->suspended_bdev)
1255 __set_size(md, size);
1258 dm_table_destroy(t);
1262 dm_table_event_callback(t, event_callback, md);
1264 write_lock(&md->map_lock);
1266 dm_table_set_restrictions(t, q);
1267 write_unlock(&md->map_lock);
1272 static void __unbind(struct mapped_device *md)
1274 struct dm_table *map = md->map;
1279 dm_table_event_callback(map, NULL, NULL);
1280 write_lock(&md->map_lock);
1282 write_unlock(&md->map_lock);
1283 dm_table_destroy(map);
1287 * Constructor for a new device.
1289 int dm_create(int minor, struct mapped_device **result)
1291 struct mapped_device *md;
1293 md = alloc_dev(minor);
1303 static struct mapped_device *dm_find_md(dev_t dev)
1305 struct mapped_device *md;
1306 unsigned minor = MINOR(dev);
1308 if (MAJOR(dev) != _major || minor >= (1 << MINORBITS))
1311 spin_lock(&_minor_lock);
1313 md = idr_find(&_minor_idr, minor);
1314 if (md && (md == MINOR_ALLOCED ||
1315 (MINOR(disk_devt(dm_disk(md))) != minor) ||
1316 test_bit(DMF_FREEING, &md->flags))) {
1322 spin_unlock(&_minor_lock);
1327 struct mapped_device *dm_get_md(dev_t dev)
1329 struct mapped_device *md = dm_find_md(dev);
1337 void *dm_get_mdptr(struct mapped_device *md)
1339 return md->interface_ptr;
1342 void dm_set_mdptr(struct mapped_device *md, void *ptr)
1344 md->interface_ptr = ptr;
1347 void dm_get(struct mapped_device *md)
1349 atomic_inc(&md->holders);
1352 const char *dm_device_name(struct mapped_device *md)
1356 EXPORT_SYMBOL_GPL(dm_device_name);
1358 void dm_put(struct mapped_device *md)
1360 struct dm_table *map;
1362 BUG_ON(test_bit(DMF_FREEING, &md->flags));
1364 if (atomic_dec_and_lock(&md->holders, &_minor_lock)) {
1365 map = dm_get_table(md);
1366 idr_replace(&_minor_idr, MINOR_ALLOCED,
1367 MINOR(disk_devt(dm_disk(md))));
1368 set_bit(DMF_FREEING, &md->flags);
1369 spin_unlock(&_minor_lock);
1370 if (!dm_suspended(md)) {
1371 dm_table_presuspend_targets(map);
1372 dm_table_postsuspend_targets(map);
1380 EXPORT_SYMBOL_GPL(dm_put);
1382 static int dm_wait_for_completion(struct mapped_device *md)
1387 set_current_state(TASK_INTERRUPTIBLE);
1390 if (!atomic_read(&md->pending))
1393 if (signal_pending(current)) {
1400 set_current_state(TASK_RUNNING);
1406 * Process the deferred bios
1408 static void __flush_deferred_io(struct mapped_device *md)
1412 while ((c = bio_list_pop(&md->deferred))) {
1413 if (__split_bio(md, c))
1417 clear_bit(DMF_BLOCK_IO, &md->flags);
1420 static void __merge_pushback_list(struct mapped_device *md)
1422 unsigned long flags;
1424 spin_lock_irqsave(&md->pushback_lock, flags);
1425 clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
1426 bio_list_merge_head(&md->deferred, &md->pushback);
1427 bio_list_init(&md->pushback);
1428 spin_unlock_irqrestore(&md->pushback_lock, flags);
1431 static void dm_wq_work(struct work_struct *work)
1433 struct dm_wq_req *req = container_of(work, struct dm_wq_req, work);
1434 struct mapped_device *md = req->md;
1436 down_write(&md->io_lock);
1437 switch (req->type) {
1438 case DM_WQ_FLUSH_DEFERRED:
1439 __flush_deferred_io(md);
1442 DMERR("dm_wq_work: unrecognised work type %d", req->type);
1445 up_write(&md->io_lock);
1448 static void dm_wq_queue(struct mapped_device *md, int type, void *context,
1449 struct dm_wq_req *req)
1453 req->context = context;
1454 INIT_WORK(&req->work, dm_wq_work);
1455 queue_work(md->wq, &req->work);
1458 static void dm_queue_flush(struct mapped_device *md, int type, void *context)
1460 struct dm_wq_req req;
1462 dm_wq_queue(md, type, context, &req);
1463 flush_workqueue(md->wq);
1467 * Swap in a new table (destroying old one).
1469 int dm_swap_table(struct mapped_device *md, struct dm_table *table)
1473 mutex_lock(&md->suspend_lock);
1475 /* device must be suspended */
1476 if (!dm_suspended(md))
1479 /* without bdev, the device size cannot be changed */
1480 if (!md->suspended_bdev)
1481 if (get_capacity(md->disk) != dm_table_get_size(table))
1485 r = __bind(md, table);
1488 mutex_unlock(&md->suspend_lock);
1493 * Functions to lock and unlock any filesystem running on the
1496 static int lock_fs(struct mapped_device *md)
1500 WARN_ON(md->frozen_sb);
1502 md->frozen_sb = freeze_bdev(md->suspended_bdev);
1503 if (IS_ERR(md->frozen_sb)) {
1504 r = PTR_ERR(md->frozen_sb);
1505 md->frozen_sb = NULL;
1509 set_bit(DMF_FROZEN, &md->flags);
1511 /* don't bdput right now, we don't want the bdev
1512 * to go away while it is locked.
1517 static void unlock_fs(struct mapped_device *md)
1519 if (!test_bit(DMF_FROZEN, &md->flags))
1522 thaw_bdev(md->suspended_bdev, md->frozen_sb);
1523 md->frozen_sb = NULL;
1524 clear_bit(DMF_FROZEN, &md->flags);
1528 * We need to be able to change a mapping table under a mounted
1529 * filesystem. For example we might want to move some data in
1530 * the background. Before the table can be swapped with
1531 * dm_bind_table, dm_suspend must be called to flush any in
1532 * flight bios and ensure that any further io gets deferred.
1534 int dm_suspend(struct mapped_device *md, unsigned suspend_flags)
1536 struct dm_table *map = NULL;
1537 DECLARE_WAITQUEUE(wait, current);
1539 int do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG ? 1 : 0;
1540 int noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG ? 1 : 0;
1542 mutex_lock(&md->suspend_lock);
1544 if (dm_suspended(md)) {
1549 map = dm_get_table(md);
1552 * DMF_NOFLUSH_SUSPENDING must be set before presuspend.
1553 * This flag is cleared before dm_suspend returns.
1556 set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
1558 /* This does not get reverted if there's an error later. */
1559 dm_table_presuspend_targets(map);
1561 /* bdget() can stall if the pending I/Os are not flushed */
1563 md->suspended_bdev = bdget_disk(md->disk, 0);
1564 if (!md->suspended_bdev) {
1565 DMWARN("bdget failed in dm_suspend");
1571 * Flush I/O to the device. noflush supersedes do_lockfs,
1572 * because lock_fs() needs to flush I/Os.
1582 * First we set the BLOCK_IO flag so no more ios will be mapped.
1584 down_write(&md->io_lock);
1585 set_bit(DMF_BLOCK_IO, &md->flags);
1587 add_wait_queue(&md->wait, &wait);
1588 up_write(&md->io_lock);
1592 dm_table_unplug_all(map);
1595 * Wait for the already-mapped ios to complete.
1597 r = dm_wait_for_completion(md);
1599 down_write(&md->io_lock);
1600 remove_wait_queue(&md->wait, &wait);
1603 __merge_pushback_list(md);
1604 up_write(&md->io_lock);
1606 /* were we interrupted ? */
1608 dm_queue_flush(md, DM_WQ_FLUSH_DEFERRED, NULL);
1611 goto out; /* pushback list is already flushed, so skip flush */
1614 dm_table_postsuspend_targets(map);
1616 set_bit(DMF_SUSPENDED, &md->flags);
1619 if (r && md->suspended_bdev) {
1620 bdput(md->suspended_bdev);
1621 md->suspended_bdev = NULL;
1627 mutex_unlock(&md->suspend_lock);
1631 int dm_resume(struct mapped_device *md)
1634 struct dm_table *map = NULL;
1636 mutex_lock(&md->suspend_lock);
1637 if (!dm_suspended(md))
1640 map = dm_get_table(md);
1641 if (!map || !dm_table_get_size(map))
1644 r = dm_table_resume_targets(map);
1648 dm_queue_flush(md, DM_WQ_FLUSH_DEFERRED, NULL);
1652 if (md->suspended_bdev) {
1653 bdput(md->suspended_bdev);
1654 md->suspended_bdev = NULL;
1657 clear_bit(DMF_SUSPENDED, &md->flags);
1659 dm_table_unplug_all(map);
1661 dm_kobject_uevent(md);
1667 mutex_unlock(&md->suspend_lock);
1672 /*-----------------------------------------------------------------
1673 * Event notification.
1674 *---------------------------------------------------------------*/
1675 void dm_kobject_uevent(struct mapped_device *md)
1677 kobject_uevent(&disk_to_dev(md->disk)->kobj, KOBJ_CHANGE);
1680 uint32_t dm_next_uevent_seq(struct mapped_device *md)
1682 return atomic_add_return(1, &md->uevent_seq);
1685 uint32_t dm_get_event_nr(struct mapped_device *md)
1687 return atomic_read(&md->event_nr);
1690 int dm_wait_event(struct mapped_device *md, int event_nr)
1692 return wait_event_interruptible(md->eventq,
1693 (event_nr != atomic_read(&md->event_nr)));
1696 void dm_uevent_add(struct mapped_device *md, struct list_head *elist)
1698 unsigned long flags;
1700 spin_lock_irqsave(&md->uevent_lock, flags);
1701 list_add(elist, &md->uevent_list);
1702 spin_unlock_irqrestore(&md->uevent_lock, flags);
1706 * The gendisk is only valid as long as you have a reference
1709 struct gendisk *dm_disk(struct mapped_device *md)
1714 struct kobject *dm_kobject(struct mapped_device *md)
1720 * struct mapped_device should not be exported outside of dm.c
1721 * so use this check to verify that kobj is part of md structure
1723 struct mapped_device *dm_get_from_kobject(struct kobject *kobj)
1725 struct mapped_device *md;
1727 md = container_of(kobj, struct mapped_device, kobj);
1728 if (&md->kobj != kobj)
1735 int dm_suspended(struct mapped_device *md)
1737 return test_bit(DMF_SUSPENDED, &md->flags);
1740 int dm_noflush_suspending(struct dm_target *ti)
1742 struct mapped_device *md = dm_table_get_md(ti->table);
1743 int r = __noflush_suspending(md);
1749 EXPORT_SYMBOL_GPL(dm_noflush_suspending);
1751 static struct block_device_operations dm_blk_dops = {
1752 .open = dm_blk_open,
1753 .release = dm_blk_close,
1754 .ioctl = dm_blk_ioctl,
1755 .getgeo = dm_blk_getgeo,
1756 .owner = THIS_MODULE
1759 EXPORT_SYMBOL(dm_get_mapinfo);
1764 module_init(dm_init);
1765 module_exit(dm_exit);
1767 module_param(major, uint, 0);
1768 MODULE_PARM_DESC(major, "The major number of the device mapper");
1769 MODULE_DESCRIPTION(DM_NAME " driver");
1770 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
1771 MODULE_LICENSE("GPL");