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 <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;
173 static struct kmem_cache *_io_cache;
174 static struct kmem_cache *_tio_cache;
175 static struct kmem_cache *_rq_tio_cache;
176 static struct kmem_cache *_rq_bio_info_cache;
178 static int __init local_init(void)
182 /* allocate a slab for the dm_ios */
183 _io_cache = KMEM_CACHE(dm_io, 0);
187 /* allocate a slab for the target ios */
188 _tio_cache = KMEM_CACHE(dm_target_io, 0);
190 goto out_free_io_cache;
192 _rq_tio_cache = KMEM_CACHE(dm_rq_target_io, 0);
194 goto out_free_tio_cache;
196 _rq_bio_info_cache = KMEM_CACHE(dm_rq_clone_bio_info, 0);
197 if (!_rq_bio_info_cache)
198 goto out_free_rq_tio_cache;
200 r = dm_uevent_init();
202 goto out_free_rq_bio_info_cache;
205 r = register_blkdev(_major, _name);
207 goto out_uevent_exit;
216 out_free_rq_bio_info_cache:
217 kmem_cache_destroy(_rq_bio_info_cache);
218 out_free_rq_tio_cache:
219 kmem_cache_destroy(_rq_tio_cache);
221 kmem_cache_destroy(_tio_cache);
223 kmem_cache_destroy(_io_cache);
228 static void local_exit(void)
230 kmem_cache_destroy(_rq_bio_info_cache);
231 kmem_cache_destroy(_rq_tio_cache);
232 kmem_cache_destroy(_tio_cache);
233 kmem_cache_destroy(_io_cache);
234 unregister_blkdev(_major, _name);
239 DMINFO("cleaned up");
242 static int (*_inits[])(void) __initdata = {
251 static void (*_exits[])(void) = {
260 static int __init dm_init(void)
262 const int count = ARRAY_SIZE(_inits);
266 for (i = 0; i < count; i++) {
281 static void __exit dm_exit(void)
283 int i = ARRAY_SIZE(_exits);
290 * Block device functions
292 static int dm_blk_open(struct block_device *bdev, fmode_t mode)
294 struct mapped_device *md;
296 spin_lock(&_minor_lock);
298 md = bdev->bd_disk->private_data;
302 if (test_bit(DMF_FREEING, &md->flags) ||
303 test_bit(DMF_DELETING, &md->flags)) {
309 atomic_inc(&md->open_count);
312 spin_unlock(&_minor_lock);
314 return md ? 0 : -ENXIO;
317 static int dm_blk_close(struct gendisk *disk, fmode_t mode)
319 struct mapped_device *md = disk->private_data;
320 atomic_dec(&md->open_count);
325 int dm_open_count(struct mapped_device *md)
327 return atomic_read(&md->open_count);
331 * Guarantees nothing is using the device before it's deleted.
333 int dm_lock_for_deletion(struct mapped_device *md)
337 spin_lock(&_minor_lock);
339 if (dm_open_count(md))
342 set_bit(DMF_DELETING, &md->flags);
344 spin_unlock(&_minor_lock);
349 static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo)
351 struct mapped_device *md = bdev->bd_disk->private_data;
353 return dm_get_geometry(md, geo);
356 static int dm_blk_ioctl(struct block_device *bdev, fmode_t mode,
357 unsigned int cmd, unsigned long arg)
359 struct mapped_device *md = bdev->bd_disk->private_data;
360 struct dm_table *map = dm_get_table(md);
361 struct dm_target *tgt;
364 if (!map || !dm_table_get_size(map))
367 /* We only support devices that have a single target */
368 if (dm_table_get_num_targets(map) != 1)
371 tgt = dm_table_get_target(map, 0);
373 if (dm_suspended(md)) {
378 if (tgt->type->ioctl)
379 r = tgt->type->ioctl(tgt, cmd, arg);
387 static struct dm_io *alloc_io(struct mapped_device *md)
389 return mempool_alloc(md->io_pool, GFP_NOIO);
392 static void free_io(struct mapped_device *md, struct dm_io *io)
394 mempool_free(io, md->io_pool);
397 static struct dm_target_io *alloc_tio(struct mapped_device *md)
399 return mempool_alloc(md->tio_pool, GFP_NOIO);
402 static void free_tio(struct mapped_device *md, struct dm_target_io *tio)
404 mempool_free(tio, md->tio_pool);
407 static void start_io_acct(struct dm_io *io)
409 struct mapped_device *md = io->md;
412 io->start_time = jiffies;
414 cpu = part_stat_lock();
415 part_round_stats(cpu, &dm_disk(md)->part0);
417 dm_disk(md)->part0.in_flight = atomic_inc_return(&md->pending);
420 static void end_io_acct(struct dm_io *io)
422 struct mapped_device *md = io->md;
423 struct bio *bio = io->bio;
424 unsigned long duration = jiffies - io->start_time;
426 int rw = bio_data_dir(bio);
428 cpu = part_stat_lock();
429 part_round_stats(cpu, &dm_disk(md)->part0);
430 part_stat_add(cpu, &dm_disk(md)->part0, ticks[rw], duration);
433 dm_disk(md)->part0.in_flight = pending =
434 atomic_dec_return(&md->pending);
436 /* nudge anyone waiting on suspend queue */
442 * Add the bio to the list of deferred io.
444 static int queue_io(struct mapped_device *md, struct bio *bio)
446 down_write(&md->io_lock);
448 if (!test_bit(DMF_BLOCK_IO, &md->flags)) {
449 up_write(&md->io_lock);
453 bio_list_add(&md->deferred, bio);
455 up_write(&md->io_lock);
456 return 0; /* deferred successfully */
460 * Everyone (including functions in this file), should use this
461 * function to access the md->map field, and make sure they call
462 * dm_table_put() when finished.
464 struct dm_table *dm_get_table(struct mapped_device *md)
468 read_lock(&md->map_lock);
472 read_unlock(&md->map_lock);
478 * Get the geometry associated with a dm device
480 int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo)
488 * Set the geometry of a device.
490 int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo)
492 sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors;
494 if (geo->start > sz) {
495 DMWARN("Start sector is beyond the geometry limits.");
504 /*-----------------------------------------------------------------
506 * A more elegant soln is in the works that uses the queue
507 * merge fn, unfortunately there are a couple of changes to
508 * the block layer that I want to make for this. So in the
509 * interests of getting something for people to use I give
510 * you this clearly demarcated crap.
511 *---------------------------------------------------------------*/
513 static int __noflush_suspending(struct mapped_device *md)
515 return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
519 * Decrements the number of outstanding ios that a bio has been
520 * cloned into, completing the original io if necc.
522 static void dec_pending(struct dm_io *io, int error)
526 /* Push-back supersedes any I/O errors */
527 if (error && !(io->error > 0 && __noflush_suspending(io->md)))
530 if (atomic_dec_and_test(&io->io_count)) {
531 if (io->error == DM_ENDIO_REQUEUE) {
533 * Target requested pushing back the I/O.
534 * This must be handled before the sleeper on
535 * suspend queue merges the pushback list.
537 spin_lock_irqsave(&io->md->pushback_lock, flags);
538 if (__noflush_suspending(io->md))
539 bio_list_add(&io->md->pushback, io->bio);
541 /* noflush suspend was interrupted. */
543 spin_unlock_irqrestore(&io->md->pushback_lock, flags);
548 if (io->error != DM_ENDIO_REQUEUE) {
549 trace_block_bio_complete(io->md->queue, io->bio);
551 bio_endio(io->bio, io->error);
558 static void clone_endio(struct bio *bio, int error)
561 struct dm_target_io *tio = bio->bi_private;
562 struct mapped_device *md = tio->io->md;
563 dm_endio_fn endio = tio->ti->type->end_io;
565 if (!bio_flagged(bio, BIO_UPTODATE) && !error)
569 r = endio(tio->ti, bio, error, &tio->info);
570 if (r < 0 || r == DM_ENDIO_REQUEUE)
572 * error and requeue request are handled
576 else if (r == DM_ENDIO_INCOMPLETE)
577 /* The target will handle the io */
580 DMWARN("unimplemented target endio return value: %d", r);
585 dec_pending(tio->io, error);
588 * Store md for cleanup instead of tio which is about to get freed.
590 bio->bi_private = md->bs;
596 static sector_t max_io_len(struct mapped_device *md,
597 sector_t sector, struct dm_target *ti)
599 sector_t offset = sector - ti->begin;
600 sector_t len = ti->len - offset;
603 * Does the target need to split even further ?
607 boundary = ((offset + ti->split_io) & ~(ti->split_io - 1))
616 static void __map_bio(struct dm_target *ti, struct bio *clone,
617 struct dm_target_io *tio)
621 struct mapped_device *md;
626 BUG_ON(!clone->bi_size);
628 clone->bi_end_io = clone_endio;
629 clone->bi_private = tio;
632 * Map the clone. If r == 0 we don't need to do
633 * anything, the target has assumed ownership of
636 atomic_inc(&tio->io->io_count);
637 sector = clone->bi_sector;
638 r = ti->type->map(ti, clone, &tio->info);
639 if (r == DM_MAPIO_REMAPPED) {
640 /* the bio has been remapped so dispatch it */
642 trace_block_remap(bdev_get_queue(clone->bi_bdev), clone,
643 tio->io->bio->bi_bdev->bd_dev,
644 clone->bi_sector, sector);
646 generic_make_request(clone);
647 } else if (r < 0 || r == DM_MAPIO_REQUEUE) {
648 /* error the io and bail out, or requeue it if needed */
650 dec_pending(tio->io, r);
652 * Store bio_set for cleanup.
654 clone->bi_private = md->bs;
658 DMWARN("unimplemented target map return value: %d", r);
664 struct mapped_device *md;
665 struct dm_table *map;
669 sector_t sector_count;
673 static void dm_bio_destructor(struct bio *bio)
675 struct bio_set *bs = bio->bi_private;
681 * Creates a little bio that is just does part of a bvec.
683 static struct bio *split_bvec(struct bio *bio, sector_t sector,
684 unsigned short idx, unsigned int offset,
685 unsigned int len, struct bio_set *bs)
688 struct bio_vec *bv = bio->bi_io_vec + idx;
690 clone = bio_alloc_bioset(GFP_NOIO, 1, bs);
691 clone->bi_destructor = dm_bio_destructor;
692 *clone->bi_io_vec = *bv;
694 clone->bi_sector = sector;
695 clone->bi_bdev = bio->bi_bdev;
696 clone->bi_rw = bio->bi_rw;
698 clone->bi_size = to_bytes(len);
699 clone->bi_io_vec->bv_offset = offset;
700 clone->bi_io_vec->bv_len = clone->bi_size;
701 clone->bi_flags |= 1 << BIO_CLONED;
707 * Creates a bio that consists of range of complete bvecs.
709 static struct bio *clone_bio(struct bio *bio, sector_t sector,
710 unsigned short idx, unsigned short bv_count,
711 unsigned int len, struct bio_set *bs)
715 clone = bio_alloc_bioset(GFP_NOIO, bio->bi_max_vecs, bs);
716 __bio_clone(clone, bio);
717 clone->bi_destructor = dm_bio_destructor;
718 clone->bi_sector = sector;
720 clone->bi_vcnt = idx + bv_count;
721 clone->bi_size = to_bytes(len);
722 clone->bi_flags &= ~(1 << BIO_SEG_VALID);
727 static int __clone_and_map(struct clone_info *ci)
729 struct bio *clone, *bio = ci->bio;
730 struct dm_target *ti;
731 sector_t len = 0, max;
732 struct dm_target_io *tio;
734 ti = dm_table_find_target(ci->map, ci->sector);
735 if (!dm_target_is_valid(ti))
738 max = max_io_len(ci->md, ci->sector, ti);
741 * Allocate a target io object.
743 tio = alloc_tio(ci->md);
746 memset(&tio->info, 0, sizeof(tio->info));
748 if (ci->sector_count <= max) {
750 * Optimise for the simple case where we can do all of
751 * the remaining io with a single clone.
753 clone = clone_bio(bio, ci->sector, ci->idx,
754 bio->bi_vcnt - ci->idx, ci->sector_count,
756 __map_bio(ti, clone, tio);
757 ci->sector_count = 0;
759 } else if (to_sector(bio->bi_io_vec[ci->idx].bv_len) <= max) {
761 * There are some bvecs that don't span targets.
762 * Do as many of these as possible.
765 sector_t remaining = max;
768 for (i = ci->idx; remaining && (i < bio->bi_vcnt); i++) {
769 bv_len = to_sector(bio->bi_io_vec[i].bv_len);
771 if (bv_len > remaining)
778 clone = clone_bio(bio, ci->sector, ci->idx, i - ci->idx, len,
780 __map_bio(ti, clone, tio);
783 ci->sector_count -= len;
788 * Handle a bvec that must be split between two or more targets.
790 struct bio_vec *bv = bio->bi_io_vec + ci->idx;
791 sector_t remaining = to_sector(bv->bv_len);
792 unsigned int offset = 0;
796 ti = dm_table_find_target(ci->map, ci->sector);
797 if (!dm_target_is_valid(ti))
800 max = max_io_len(ci->md, ci->sector, ti);
802 tio = alloc_tio(ci->md);
805 memset(&tio->info, 0, sizeof(tio->info));
808 len = min(remaining, max);
810 clone = split_bvec(bio, ci->sector, ci->idx,
811 bv->bv_offset + offset, len,
814 __map_bio(ti, clone, tio);
817 ci->sector_count -= len;
818 offset += to_bytes(len);
819 } while (remaining -= len);
828 * Split the bio into several clones.
830 static int __split_bio(struct mapped_device *md, struct bio *bio)
832 struct clone_info ci;
835 ci.map = dm_get_table(md);
836 if (unlikely(!ci.map))
838 if (unlikely(bio_barrier(bio) && !dm_table_barrier_ok(ci.map))) {
839 dm_table_put(ci.map);
840 bio_endio(bio, -EOPNOTSUPP);
845 ci.io = alloc_io(md);
847 atomic_set(&ci.io->io_count, 1);
850 ci.sector = bio->bi_sector;
851 ci.sector_count = bio_sectors(bio);
852 ci.idx = bio->bi_idx;
854 start_io_acct(ci.io);
855 while (ci.sector_count && !error)
856 error = __clone_and_map(&ci);
858 /* drop the extra reference count */
859 dec_pending(ci.io, error);
860 dm_table_put(ci.map);
864 /*-----------------------------------------------------------------
866 *---------------------------------------------------------------*/
868 static int dm_merge_bvec(struct request_queue *q,
869 struct bvec_merge_data *bvm,
870 struct bio_vec *biovec)
872 struct mapped_device *md = q->queuedata;
873 struct dm_table *map = dm_get_table(md);
874 struct dm_target *ti;
875 sector_t max_sectors;
881 ti = dm_table_find_target(map, bvm->bi_sector);
882 if (!dm_target_is_valid(ti))
886 * Find maximum amount of I/O that won't need splitting
888 max_sectors = min(max_io_len(md, bvm->bi_sector, ti),
889 (sector_t) BIO_MAX_SECTORS);
890 max_size = (max_sectors << SECTOR_SHIFT) - bvm->bi_size;
895 * merge_bvec_fn() returns number of bytes
896 * it can accept at this offset
897 * max is precomputed maximal io size
899 if (max_size && ti->type->merge)
900 max_size = ti->type->merge(ti, bvm, biovec, max_size);
907 * Always allow an entire first page
909 if (max_size <= biovec->bv_len && !(bvm->bi_size >> SECTOR_SHIFT))
910 max_size = biovec->bv_len;
916 * The request function that just remaps the bio built up by
919 static int dm_request(struct request_queue *q, struct bio *bio)
922 int rw = bio_data_dir(bio);
923 struct mapped_device *md = q->queuedata;
926 down_read(&md->io_lock);
928 cpu = part_stat_lock();
929 part_stat_inc(cpu, &dm_disk(md)->part0, ios[rw]);
930 part_stat_add(cpu, &dm_disk(md)->part0, sectors[rw], bio_sectors(bio));
934 * If we're suspended we have to queue
937 while (test_bit(DMF_BLOCK_IO, &md->flags)) {
938 up_read(&md->io_lock);
940 if (bio_rw(bio) != READA)
941 r = queue_io(md, bio);
947 * We're in a while loop, because someone could suspend
948 * before we get to the following read lock.
950 down_read(&md->io_lock);
953 r = __split_bio(md, bio);
954 up_read(&md->io_lock);
963 static void dm_unplug_all(struct request_queue *q)
965 struct mapped_device *md = q->queuedata;
966 struct dm_table *map = dm_get_table(md);
969 dm_table_unplug_all(map);
974 static int dm_any_congested(void *congested_data, int bdi_bits)
977 struct mapped_device *md = congested_data;
978 struct dm_table *map;
980 if (!test_bit(DMF_BLOCK_IO, &md->flags)) {
981 map = dm_get_table(md);
983 r = dm_table_any_congested(map, bdi_bits);
991 /*-----------------------------------------------------------------
992 * An IDR is used to keep track of allocated minor numbers.
993 *---------------------------------------------------------------*/
994 static DEFINE_IDR(_minor_idr);
996 static void free_minor(int minor)
998 spin_lock(&_minor_lock);
999 idr_remove(&_minor_idr, minor);
1000 spin_unlock(&_minor_lock);
1004 * See if the device with a specific minor # is free.
1006 static int specific_minor(int minor)
1010 if (minor >= (1 << MINORBITS))
1013 r = idr_pre_get(&_minor_idr, GFP_KERNEL);
1017 spin_lock(&_minor_lock);
1019 if (idr_find(&_minor_idr, minor)) {
1024 r = idr_get_new_above(&_minor_idr, MINOR_ALLOCED, minor, &m);
1029 idr_remove(&_minor_idr, m);
1035 spin_unlock(&_minor_lock);
1039 static int next_free_minor(int *minor)
1043 r = idr_pre_get(&_minor_idr, GFP_KERNEL);
1047 spin_lock(&_minor_lock);
1049 r = idr_get_new(&_minor_idr, MINOR_ALLOCED, &m);
1053 if (m >= (1 << MINORBITS)) {
1054 idr_remove(&_minor_idr, m);
1062 spin_unlock(&_minor_lock);
1066 static struct block_device_operations dm_blk_dops;
1069 * Allocate and initialise a blank device with a given minor.
1071 static struct mapped_device *alloc_dev(int minor)
1074 struct mapped_device *md = kzalloc(sizeof(*md), GFP_KERNEL);
1078 DMWARN("unable to allocate device, out of memory.");
1082 if (!try_module_get(THIS_MODULE))
1083 goto bad_module_get;
1085 /* get a minor number for the dev */
1086 if (minor == DM_ANY_MINOR)
1087 r = next_free_minor(&minor);
1089 r = specific_minor(minor);
1093 init_rwsem(&md->io_lock);
1094 mutex_init(&md->suspend_lock);
1095 spin_lock_init(&md->pushback_lock);
1096 rwlock_init(&md->map_lock);
1097 atomic_set(&md->holders, 1);
1098 atomic_set(&md->open_count, 0);
1099 atomic_set(&md->event_nr, 0);
1100 atomic_set(&md->uevent_seq, 0);
1101 INIT_LIST_HEAD(&md->uevent_list);
1102 spin_lock_init(&md->uevent_lock);
1104 md->queue = blk_alloc_queue(GFP_KERNEL);
1108 md->queue->queuedata = md;
1109 md->queue->backing_dev_info.congested_fn = dm_any_congested;
1110 md->queue->backing_dev_info.congested_data = md;
1111 blk_queue_make_request(md->queue, dm_request);
1112 blk_queue_bounce_limit(md->queue, BLK_BOUNCE_ANY);
1113 md->queue->unplug_fn = dm_unplug_all;
1114 blk_queue_merge_bvec(md->queue, dm_merge_bvec);
1116 md->io_pool = mempool_create_slab_pool(MIN_IOS, _io_cache);
1120 md->tio_pool = mempool_create_slab_pool(MIN_IOS, _tio_cache);
1124 md->bs = bioset_create(16, 0);
1128 md->disk = alloc_disk(1);
1132 atomic_set(&md->pending, 0);
1133 init_waitqueue_head(&md->wait);
1134 init_waitqueue_head(&md->eventq);
1136 md->disk->major = _major;
1137 md->disk->first_minor = minor;
1138 md->disk->fops = &dm_blk_dops;
1139 md->disk->queue = md->queue;
1140 md->disk->private_data = md;
1141 sprintf(md->disk->disk_name, "dm-%d", minor);
1143 format_dev_t(md->name, MKDEV(_major, minor));
1145 md->wq = create_singlethread_workqueue("kdmflush");
1149 /* Populate the mapping, nobody knows we exist yet */
1150 spin_lock(&_minor_lock);
1151 old_md = idr_replace(&_minor_idr, md, minor);
1152 spin_unlock(&_minor_lock);
1154 BUG_ON(old_md != MINOR_ALLOCED);
1161 bioset_free(md->bs);
1163 mempool_destroy(md->tio_pool);
1165 mempool_destroy(md->io_pool);
1167 blk_cleanup_queue(md->queue);
1171 module_put(THIS_MODULE);
1177 static void unlock_fs(struct mapped_device *md);
1179 static void free_dev(struct mapped_device *md)
1181 int minor = MINOR(disk_devt(md->disk));
1183 if (md->suspended_bdev) {
1185 bdput(md->suspended_bdev);
1187 destroy_workqueue(md->wq);
1188 mempool_destroy(md->tio_pool);
1189 mempool_destroy(md->io_pool);
1190 bioset_free(md->bs);
1191 del_gendisk(md->disk);
1194 spin_lock(&_minor_lock);
1195 md->disk->private_data = NULL;
1196 spin_unlock(&_minor_lock);
1199 blk_cleanup_queue(md->queue);
1200 module_put(THIS_MODULE);
1205 * Bind a table to the device.
1207 static void event_callback(void *context)
1209 unsigned long flags;
1211 struct mapped_device *md = (struct mapped_device *) context;
1213 spin_lock_irqsave(&md->uevent_lock, flags);
1214 list_splice_init(&md->uevent_list, &uevents);
1215 spin_unlock_irqrestore(&md->uevent_lock, flags);
1217 dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj);
1219 atomic_inc(&md->event_nr);
1220 wake_up(&md->eventq);
1223 static void __set_size(struct mapped_device *md, sector_t size)
1225 set_capacity(md->disk, size);
1227 mutex_lock(&md->suspended_bdev->bd_inode->i_mutex);
1228 i_size_write(md->suspended_bdev->bd_inode, (loff_t)size << SECTOR_SHIFT);
1229 mutex_unlock(&md->suspended_bdev->bd_inode->i_mutex);
1232 static int __bind(struct mapped_device *md, struct dm_table *t)
1234 struct request_queue *q = md->queue;
1237 size = dm_table_get_size(t);
1240 * Wipe any geometry if the size of the table changed.
1242 if (size != get_capacity(md->disk))
1243 memset(&md->geometry, 0, sizeof(md->geometry));
1245 if (md->suspended_bdev)
1246 __set_size(md, size);
1249 dm_table_destroy(t);
1253 dm_table_event_callback(t, event_callback, md);
1255 write_lock(&md->map_lock);
1257 dm_table_set_restrictions(t, q);
1258 write_unlock(&md->map_lock);
1263 static void __unbind(struct mapped_device *md)
1265 struct dm_table *map = md->map;
1270 dm_table_event_callback(map, NULL, NULL);
1271 write_lock(&md->map_lock);
1273 write_unlock(&md->map_lock);
1274 dm_table_destroy(map);
1278 * Constructor for a new device.
1280 int dm_create(int minor, struct mapped_device **result)
1282 struct mapped_device *md;
1284 md = alloc_dev(minor);
1292 static struct mapped_device *dm_find_md(dev_t dev)
1294 struct mapped_device *md;
1295 unsigned minor = MINOR(dev);
1297 if (MAJOR(dev) != _major || minor >= (1 << MINORBITS))
1300 spin_lock(&_minor_lock);
1302 md = idr_find(&_minor_idr, minor);
1303 if (md && (md == MINOR_ALLOCED ||
1304 (MINOR(disk_devt(dm_disk(md))) != minor) ||
1305 test_bit(DMF_FREEING, &md->flags))) {
1311 spin_unlock(&_minor_lock);
1316 struct mapped_device *dm_get_md(dev_t dev)
1318 struct mapped_device *md = dm_find_md(dev);
1326 void *dm_get_mdptr(struct mapped_device *md)
1328 return md->interface_ptr;
1331 void dm_set_mdptr(struct mapped_device *md, void *ptr)
1333 md->interface_ptr = ptr;
1336 void dm_get(struct mapped_device *md)
1338 atomic_inc(&md->holders);
1341 const char *dm_device_name(struct mapped_device *md)
1345 EXPORT_SYMBOL_GPL(dm_device_name);
1347 void dm_put(struct mapped_device *md)
1349 struct dm_table *map;
1351 BUG_ON(test_bit(DMF_FREEING, &md->flags));
1353 if (atomic_dec_and_lock(&md->holders, &_minor_lock)) {
1354 map = dm_get_table(md);
1355 idr_replace(&_minor_idr, MINOR_ALLOCED,
1356 MINOR(disk_devt(dm_disk(md))));
1357 set_bit(DMF_FREEING, &md->flags);
1358 spin_unlock(&_minor_lock);
1359 if (!dm_suspended(md)) {
1360 dm_table_presuspend_targets(map);
1361 dm_table_postsuspend_targets(map);
1368 EXPORT_SYMBOL_GPL(dm_put);
1370 static int dm_wait_for_completion(struct mapped_device *md)
1375 set_current_state(TASK_INTERRUPTIBLE);
1378 if (!atomic_read(&md->pending))
1381 if (signal_pending(current)) {
1388 set_current_state(TASK_RUNNING);
1394 * Process the deferred bios
1396 static void __flush_deferred_io(struct mapped_device *md)
1400 while ((c = bio_list_pop(&md->deferred))) {
1401 if (__split_bio(md, c))
1405 clear_bit(DMF_BLOCK_IO, &md->flags);
1408 static void __merge_pushback_list(struct mapped_device *md)
1410 unsigned long flags;
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 static void dm_wq_work(struct work_struct *work)
1421 struct dm_wq_req *req = container_of(work, struct dm_wq_req, work);
1422 struct mapped_device *md = req->md;
1424 down_write(&md->io_lock);
1425 switch (req->type) {
1426 case DM_WQ_FLUSH_DEFERRED:
1427 __flush_deferred_io(md);
1430 DMERR("dm_wq_work: unrecognised work type %d", req->type);
1433 up_write(&md->io_lock);
1436 static void dm_wq_queue(struct mapped_device *md, int type, void *context,
1437 struct dm_wq_req *req)
1441 req->context = context;
1442 INIT_WORK(&req->work, dm_wq_work);
1443 queue_work(md->wq, &req->work);
1446 static void dm_queue_flush(struct mapped_device *md, int type, void *context)
1448 struct dm_wq_req req;
1450 dm_wq_queue(md, type, context, &req);
1451 flush_workqueue(md->wq);
1455 * Swap in a new table (destroying old one).
1457 int dm_swap_table(struct mapped_device *md, struct dm_table *table)
1461 mutex_lock(&md->suspend_lock);
1463 /* device must be suspended */
1464 if (!dm_suspended(md))
1467 /* without bdev, the device size cannot be changed */
1468 if (!md->suspended_bdev)
1469 if (get_capacity(md->disk) != dm_table_get_size(table))
1473 r = __bind(md, table);
1476 mutex_unlock(&md->suspend_lock);
1481 * Functions to lock and unlock any filesystem running on the
1484 static int lock_fs(struct mapped_device *md)
1488 WARN_ON(md->frozen_sb);
1490 md->frozen_sb = freeze_bdev(md->suspended_bdev);
1491 if (IS_ERR(md->frozen_sb)) {
1492 r = PTR_ERR(md->frozen_sb);
1493 md->frozen_sb = NULL;
1497 set_bit(DMF_FROZEN, &md->flags);
1499 /* don't bdput right now, we don't want the bdev
1500 * to go away while it is locked.
1505 static void unlock_fs(struct mapped_device *md)
1507 if (!test_bit(DMF_FROZEN, &md->flags))
1510 thaw_bdev(md->suspended_bdev, md->frozen_sb);
1511 md->frozen_sb = NULL;
1512 clear_bit(DMF_FROZEN, &md->flags);
1516 * We need to be able to change a mapping table under a mounted
1517 * filesystem. For example we might want to move some data in
1518 * the background. Before the table can be swapped with
1519 * dm_bind_table, dm_suspend must be called to flush any in
1520 * flight bios and ensure that any further io gets deferred.
1522 int dm_suspend(struct mapped_device *md, unsigned suspend_flags)
1524 struct dm_table *map = NULL;
1525 DECLARE_WAITQUEUE(wait, current);
1527 int do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG ? 1 : 0;
1528 int noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG ? 1 : 0;
1530 mutex_lock(&md->suspend_lock);
1532 if (dm_suspended(md)) {
1537 map = dm_get_table(md);
1540 * DMF_NOFLUSH_SUSPENDING must be set before presuspend.
1541 * This flag is cleared before dm_suspend returns.
1544 set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
1546 /* This does not get reverted if there's an error later. */
1547 dm_table_presuspend_targets(map);
1549 /* bdget() can stall if the pending I/Os are not flushed */
1551 md->suspended_bdev = bdget_disk(md->disk, 0);
1552 if (!md->suspended_bdev) {
1553 DMWARN("bdget failed in dm_suspend");
1559 * Flush I/O to the device. noflush supersedes do_lockfs,
1560 * because lock_fs() needs to flush I/Os.
1570 * First we set the BLOCK_IO flag so no more ios will be mapped.
1572 down_write(&md->io_lock);
1573 set_bit(DMF_BLOCK_IO, &md->flags);
1575 add_wait_queue(&md->wait, &wait);
1576 up_write(&md->io_lock);
1580 dm_table_unplug_all(map);
1583 * Wait for the already-mapped ios to complete.
1585 r = dm_wait_for_completion(md);
1587 down_write(&md->io_lock);
1588 remove_wait_queue(&md->wait, &wait);
1591 __merge_pushback_list(md);
1592 up_write(&md->io_lock);
1594 /* were we interrupted ? */
1596 dm_queue_flush(md, DM_WQ_FLUSH_DEFERRED, NULL);
1599 goto out; /* pushback list is already flushed, so skip flush */
1602 dm_table_postsuspend_targets(map);
1604 set_bit(DMF_SUSPENDED, &md->flags);
1607 if (r && md->suspended_bdev) {
1608 bdput(md->suspended_bdev);
1609 md->suspended_bdev = NULL;
1615 mutex_unlock(&md->suspend_lock);
1619 int dm_resume(struct mapped_device *md)
1622 struct dm_table *map = NULL;
1624 mutex_lock(&md->suspend_lock);
1625 if (!dm_suspended(md))
1628 map = dm_get_table(md);
1629 if (!map || !dm_table_get_size(map))
1632 r = dm_table_resume_targets(map);
1636 dm_queue_flush(md, DM_WQ_FLUSH_DEFERRED, NULL);
1640 if (md->suspended_bdev) {
1641 bdput(md->suspended_bdev);
1642 md->suspended_bdev = NULL;
1645 clear_bit(DMF_SUSPENDED, &md->flags);
1647 dm_table_unplug_all(map);
1649 dm_kobject_uevent(md);
1655 mutex_unlock(&md->suspend_lock);
1660 /*-----------------------------------------------------------------
1661 * Event notification.
1662 *---------------------------------------------------------------*/
1663 void dm_kobject_uevent(struct mapped_device *md)
1665 kobject_uevent(&disk_to_dev(md->disk)->kobj, KOBJ_CHANGE);
1668 uint32_t dm_next_uevent_seq(struct mapped_device *md)
1670 return atomic_add_return(1, &md->uevent_seq);
1673 uint32_t dm_get_event_nr(struct mapped_device *md)
1675 return atomic_read(&md->event_nr);
1678 int dm_wait_event(struct mapped_device *md, int event_nr)
1680 return wait_event_interruptible(md->eventq,
1681 (event_nr != atomic_read(&md->event_nr)));
1684 void dm_uevent_add(struct mapped_device *md, struct list_head *elist)
1686 unsigned long flags;
1688 spin_lock_irqsave(&md->uevent_lock, flags);
1689 list_add(elist, &md->uevent_list);
1690 spin_unlock_irqrestore(&md->uevent_lock, flags);
1694 * The gendisk is only valid as long as you have a reference
1697 struct gendisk *dm_disk(struct mapped_device *md)
1702 int dm_suspended(struct mapped_device *md)
1704 return test_bit(DMF_SUSPENDED, &md->flags);
1707 int dm_noflush_suspending(struct dm_target *ti)
1709 struct mapped_device *md = dm_table_get_md(ti->table);
1710 int r = __noflush_suspending(md);
1716 EXPORT_SYMBOL_GPL(dm_noflush_suspending);
1718 static struct block_device_operations dm_blk_dops = {
1719 .open = dm_blk_open,
1720 .release = dm_blk_close,
1721 .ioctl = dm_blk_ioctl,
1722 .getgeo = dm_blk_getgeo,
1723 .owner = THIS_MODULE
1726 EXPORT_SYMBOL(dm_get_mapinfo);
1731 module_init(dm_init);
1732 module_exit(dm_exit);
1734 module_param(major, uint, 0);
1735 MODULE_PARM_DESC(major, "The major number of the device mapper");
1736 MODULE_DESCRIPTION(DM_NAME " driver");
1737 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
1738 MODULE_LICENSE("GPL");