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.
11 #include <linux/init.h>
12 #include <linux/module.h>
13 #include <linux/mutex.h>
14 #include <linux/moduleparam.h>
15 #include <linux/blkpg.h>
16 #include <linux/bio.h>
17 #include <linux/buffer_head.h>
18 #include <linux/mempool.h>
19 #include <linux/slab.h>
20 #include <linux/idr.h>
21 #include <linux/hdreg.h>
22 #include <linux/blktrace_api.h>
23 #include <trace/block.h>
25 #define DM_MSG_PREFIX "core"
27 static const char *_name = DM_NAME;
29 static unsigned int major = 0;
30 static unsigned int _major = 0;
32 static DEFINE_SPINLOCK(_minor_lock);
35 * One of these is allocated per bio.
38 struct mapped_device *md;
42 unsigned long start_time;
47 * One of these is allocated per target within a bio. Hopefully
48 * this will be simplified out one day.
56 DEFINE_TRACE(block_bio_complete);
59 * For request-based dm.
60 * One of these is allocated per request.
62 struct dm_rq_target_io {
63 struct mapped_device *md;
65 struct request *orig, clone;
71 * For request-based dm.
72 * One of these is allocated per bio.
74 struct dm_rq_clone_bio_info {
79 union map_info *dm_get_mapinfo(struct bio *bio)
81 if (bio && bio->bi_private)
82 return &((struct dm_target_io *)bio->bi_private)->info;
86 #define MINOR_ALLOCED ((void *)-1)
89 * Bits for the md->flags field.
91 #define DMF_BLOCK_IO_FOR_SUSPEND 0
92 #define DMF_SUSPENDED 1
95 #define DMF_DELETING 4
96 #define DMF_NOFLUSH_SUSPENDING 5
97 #define DMF_QUEUE_IO_TO_THREAD 6
100 * Work processed by per-device workqueue.
102 struct mapped_device {
103 struct rw_semaphore io_lock;
104 struct mutex suspend_lock;
111 struct request_queue *queue;
112 struct gendisk *disk;
118 * A list of ios that arrived while we were suspended.
121 wait_queue_head_t wait;
122 struct work_struct work;
123 struct bio_list deferred;
124 spinlock_t deferred_lock;
127 * An error from the barrier request currently being processed.
132 * Processing queue (flush/barriers)
134 struct workqueue_struct *wq;
137 * The current mapping.
139 struct dm_table *map;
142 * io objects are allocated from here.
153 wait_queue_head_t eventq;
155 struct list_head uevent_list;
156 spinlock_t uevent_lock; /* Protect access to uevent_list */
159 * freeze/thaw support require holding onto a super block
161 struct super_block *frozen_sb;
162 struct block_device *suspended_bdev;
164 /* forced geometry settings */
165 struct hd_geometry geometry;
172 static struct kmem_cache *_io_cache;
173 static struct kmem_cache *_tio_cache;
174 static struct kmem_cache *_rq_tio_cache;
175 static struct kmem_cache *_rq_bio_info_cache;
177 static int __init local_init(void)
181 /* allocate a slab for the dm_ios */
182 _io_cache = KMEM_CACHE(dm_io, 0);
186 /* allocate a slab for the target ios */
187 _tio_cache = KMEM_CACHE(dm_target_io, 0);
189 goto out_free_io_cache;
191 _rq_tio_cache = KMEM_CACHE(dm_rq_target_io, 0);
193 goto out_free_tio_cache;
195 _rq_bio_info_cache = KMEM_CACHE(dm_rq_clone_bio_info, 0);
196 if (!_rq_bio_info_cache)
197 goto out_free_rq_tio_cache;
199 r = dm_uevent_init();
201 goto out_free_rq_bio_info_cache;
204 r = register_blkdev(_major, _name);
206 goto out_uevent_exit;
215 out_free_rq_bio_info_cache:
216 kmem_cache_destroy(_rq_bio_info_cache);
217 out_free_rq_tio_cache:
218 kmem_cache_destroy(_rq_tio_cache);
220 kmem_cache_destroy(_tio_cache);
222 kmem_cache_destroy(_io_cache);
227 static void local_exit(void)
229 kmem_cache_destroy(_rq_bio_info_cache);
230 kmem_cache_destroy(_rq_tio_cache);
231 kmem_cache_destroy(_tio_cache);
232 kmem_cache_destroy(_io_cache);
233 unregister_blkdev(_major, _name);
238 DMINFO("cleaned up");
241 static int (*_inits[])(void) __initdata = {
250 static void (*_exits[])(void) = {
259 static int __init dm_init(void)
261 const int count = ARRAY_SIZE(_inits);
265 for (i = 0; i < count; i++) {
280 static void __exit dm_exit(void)
282 int i = ARRAY_SIZE(_exits);
289 * Block device functions
291 static int dm_blk_open(struct block_device *bdev, fmode_t mode)
293 struct mapped_device *md;
295 spin_lock(&_minor_lock);
297 md = bdev->bd_disk->private_data;
301 if (test_bit(DMF_FREEING, &md->flags) ||
302 test_bit(DMF_DELETING, &md->flags)) {
308 atomic_inc(&md->open_count);
311 spin_unlock(&_minor_lock);
313 return md ? 0 : -ENXIO;
316 static int dm_blk_close(struct gendisk *disk, fmode_t mode)
318 struct mapped_device *md = disk->private_data;
319 atomic_dec(&md->open_count);
324 int dm_open_count(struct mapped_device *md)
326 return atomic_read(&md->open_count);
330 * Guarantees nothing is using the device before it's deleted.
332 int dm_lock_for_deletion(struct mapped_device *md)
336 spin_lock(&_minor_lock);
338 if (dm_open_count(md))
341 set_bit(DMF_DELETING, &md->flags);
343 spin_unlock(&_minor_lock);
348 static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo)
350 struct mapped_device *md = bdev->bd_disk->private_data;
352 return dm_get_geometry(md, geo);
355 static int dm_blk_ioctl(struct block_device *bdev, fmode_t mode,
356 unsigned int cmd, unsigned long arg)
358 struct mapped_device *md = bdev->bd_disk->private_data;
359 struct dm_table *map = dm_get_table(md);
360 struct dm_target *tgt;
363 if (!map || !dm_table_get_size(map))
366 /* We only support devices that have a single target */
367 if (dm_table_get_num_targets(map) != 1)
370 tgt = dm_table_get_target(map, 0);
372 if (dm_suspended(md)) {
377 if (tgt->type->ioctl)
378 r = tgt->type->ioctl(tgt, cmd, arg);
386 static struct dm_io *alloc_io(struct mapped_device *md)
388 return mempool_alloc(md->io_pool, GFP_NOIO);
391 static void free_io(struct mapped_device *md, struct dm_io *io)
393 mempool_free(io, md->io_pool);
396 static struct dm_target_io *alloc_tio(struct mapped_device *md)
398 return mempool_alloc(md->tio_pool, GFP_NOIO);
401 static void free_tio(struct mapped_device *md, struct dm_target_io *tio)
403 mempool_free(tio, md->tio_pool);
406 static void start_io_acct(struct dm_io *io)
408 struct mapped_device *md = io->md;
411 io->start_time = jiffies;
413 cpu = part_stat_lock();
414 part_round_stats(cpu, &dm_disk(md)->part0);
416 dm_disk(md)->part0.in_flight = atomic_inc_return(&md->pending);
419 static void end_io_acct(struct dm_io *io)
421 struct mapped_device *md = io->md;
422 struct bio *bio = io->bio;
423 unsigned long duration = jiffies - io->start_time;
425 int rw = bio_data_dir(bio);
427 cpu = part_stat_lock();
428 part_round_stats(cpu, &dm_disk(md)->part0);
429 part_stat_add(cpu, &dm_disk(md)->part0, ticks[rw], duration);
433 * After this is decremented the bio must not be touched if it is
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 void queue_io(struct mapped_device *md, struct bio *bio)
449 down_write(&md->io_lock);
451 spin_lock_irq(&md->deferred_lock);
452 bio_list_add(&md->deferred, bio);
453 spin_unlock_irq(&md->deferred_lock);
455 if (!test_and_set_bit(DMF_QUEUE_IO_TO_THREAD, &md->flags))
456 queue_work(md->wq, &md->work);
458 up_write(&md->io_lock);
462 * Everyone (including functions in this file), should use this
463 * function to access the md->map field, and make sure they call
464 * dm_table_put() when finished.
466 struct dm_table *dm_get_table(struct mapped_device *md)
470 read_lock(&md->map_lock);
474 read_unlock(&md->map_lock);
480 * Get the geometry associated with a dm device
482 int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo)
490 * Set the geometry of a device.
492 int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo)
494 sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors;
496 if (geo->start > sz) {
497 DMWARN("Start sector is beyond the geometry limits.");
506 /*-----------------------------------------------------------------
508 * A more elegant soln is in the works that uses the queue
509 * merge fn, unfortunately there are a couple of changes to
510 * the block layer that I want to make for this. So in the
511 * interests of getting something for people to use I give
512 * you this clearly demarcated crap.
513 *---------------------------------------------------------------*/
515 static int __noflush_suspending(struct mapped_device *md)
517 return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
521 * Decrements the number of outstanding ios that a bio has been
522 * cloned into, completing the original io if necc.
524 static void dec_pending(struct dm_io *io, int error)
529 struct mapped_device *md = io->md;
531 /* Push-back supersedes any I/O errors */
532 if (error && !(io->error > 0 && __noflush_suspending(md)))
535 if (atomic_dec_and_test(&io->io_count)) {
536 if (io->error == DM_ENDIO_REQUEUE) {
538 * Target requested pushing back the I/O.
540 spin_lock_irqsave(&md->deferred_lock, flags);
541 if (__noflush_suspending(md))
542 bio_list_add_head(&md->deferred, io->bio);
544 /* noflush suspend was interrupted. */
546 spin_unlock_irqrestore(&md->deferred_lock, flags);
549 io_error = io->error;
552 if (bio_barrier(bio)) {
554 * There can be just one barrier request so we use
555 * a per-device variable for error reporting.
556 * Note that you can't touch the bio after end_io_acct
558 md->barrier_error = io_error;
563 if (io_error != DM_ENDIO_REQUEUE) {
564 trace_block_bio_complete(md->queue, bio);
566 bio_endio(bio, io_error);
574 static void clone_endio(struct bio *bio, int error)
577 struct dm_target_io *tio = bio->bi_private;
578 struct dm_io *io = tio->io;
579 struct mapped_device *md = tio->io->md;
580 dm_endio_fn endio = tio->ti->type->end_io;
582 if (!bio_flagged(bio, BIO_UPTODATE) && !error)
586 r = endio(tio->ti, bio, error, &tio->info);
587 if (r < 0 || r == DM_ENDIO_REQUEUE)
589 * error and requeue request are handled
593 else if (r == DM_ENDIO_INCOMPLETE)
594 /* The target will handle the io */
597 DMWARN("unimplemented target endio return value: %d", r);
603 * Store md for cleanup instead of tio which is about to get freed.
605 bio->bi_private = md->bs;
609 dec_pending(io, error);
612 static sector_t max_io_len(struct mapped_device *md,
613 sector_t sector, struct dm_target *ti)
615 sector_t offset = sector - ti->begin;
616 sector_t len = ti->len - offset;
619 * Does the target need to split even further ?
623 boundary = ((offset + ti->split_io) & ~(ti->split_io - 1))
632 static void __map_bio(struct dm_target *ti, struct bio *clone,
633 struct dm_target_io *tio)
637 struct mapped_device *md;
642 BUG_ON(!clone->bi_size);
644 clone->bi_end_io = clone_endio;
645 clone->bi_private = tio;
648 * Map the clone. If r == 0 we don't need to do
649 * anything, the target has assumed ownership of
652 atomic_inc(&tio->io->io_count);
653 sector = clone->bi_sector;
654 r = ti->type->map(ti, clone, &tio->info);
655 if (r == DM_MAPIO_REMAPPED) {
656 /* the bio has been remapped so dispatch it */
658 trace_block_remap(bdev_get_queue(clone->bi_bdev), clone,
659 tio->io->bio->bi_bdev->bd_dev, sector);
661 generic_make_request(clone);
662 } else if (r < 0 || r == DM_MAPIO_REQUEUE) {
663 /* error the io and bail out, or requeue it if needed */
665 dec_pending(tio->io, r);
667 * Store bio_set for cleanup.
669 clone->bi_private = md->bs;
673 DMWARN("unimplemented target map return value: %d", r);
679 struct mapped_device *md;
680 struct dm_table *map;
684 sector_t sector_count;
688 static void dm_bio_destructor(struct bio *bio)
690 struct bio_set *bs = bio->bi_private;
696 * Creates a little bio that is just does part of a bvec.
698 static struct bio *split_bvec(struct bio *bio, sector_t sector,
699 unsigned short idx, unsigned int offset,
700 unsigned int len, struct bio_set *bs)
703 struct bio_vec *bv = bio->bi_io_vec + idx;
705 clone = bio_alloc_bioset(GFP_NOIO, 1, bs);
706 clone->bi_destructor = dm_bio_destructor;
707 *clone->bi_io_vec = *bv;
709 clone->bi_sector = sector;
710 clone->bi_bdev = bio->bi_bdev;
711 clone->bi_rw = bio->bi_rw & ~(1 << BIO_RW_BARRIER);
713 clone->bi_size = to_bytes(len);
714 clone->bi_io_vec->bv_offset = offset;
715 clone->bi_io_vec->bv_len = clone->bi_size;
716 clone->bi_flags |= 1 << BIO_CLONED;
718 if (bio_integrity(bio)) {
719 bio_integrity_clone(clone, bio, GFP_NOIO);
720 bio_integrity_trim(clone,
721 bio_sector_offset(bio, idx, offset), len);
728 * Creates a bio that consists of range of complete bvecs.
730 static struct bio *clone_bio(struct bio *bio, sector_t sector,
731 unsigned short idx, unsigned short bv_count,
732 unsigned int len, struct bio_set *bs)
736 clone = bio_alloc_bioset(GFP_NOIO, bio->bi_max_vecs, bs);
737 __bio_clone(clone, bio);
738 clone->bi_rw &= ~(1 << BIO_RW_BARRIER);
739 clone->bi_destructor = dm_bio_destructor;
740 clone->bi_sector = sector;
742 clone->bi_vcnt = idx + bv_count;
743 clone->bi_size = to_bytes(len);
744 clone->bi_flags &= ~(1 << BIO_SEG_VALID);
746 if (bio_integrity(bio)) {
747 bio_integrity_clone(clone, bio, GFP_NOIO);
749 if (idx != bio->bi_idx || clone->bi_size < bio->bi_size)
750 bio_integrity_trim(clone,
751 bio_sector_offset(bio, idx, 0), len);
757 static int __clone_and_map(struct clone_info *ci)
759 struct bio *clone, *bio = ci->bio;
760 struct dm_target *ti;
761 sector_t len = 0, max;
762 struct dm_target_io *tio;
764 ti = dm_table_find_target(ci->map, ci->sector);
765 if (!dm_target_is_valid(ti))
768 max = max_io_len(ci->md, ci->sector, ti);
771 * Allocate a target io object.
773 tio = alloc_tio(ci->md);
776 memset(&tio->info, 0, sizeof(tio->info));
778 if (ci->sector_count <= max) {
780 * Optimise for the simple case where we can do all of
781 * the remaining io with a single clone.
783 clone = clone_bio(bio, ci->sector, ci->idx,
784 bio->bi_vcnt - ci->idx, ci->sector_count,
786 __map_bio(ti, clone, tio);
787 ci->sector_count = 0;
789 } else if (to_sector(bio->bi_io_vec[ci->idx].bv_len) <= max) {
791 * There are some bvecs that don't span targets.
792 * Do as many of these as possible.
795 sector_t remaining = max;
798 for (i = ci->idx; remaining && (i < bio->bi_vcnt); i++) {
799 bv_len = to_sector(bio->bi_io_vec[i].bv_len);
801 if (bv_len > remaining)
808 clone = clone_bio(bio, ci->sector, ci->idx, i - ci->idx, len,
810 __map_bio(ti, clone, tio);
813 ci->sector_count -= len;
818 * Handle a bvec that must be split between two or more targets.
820 struct bio_vec *bv = bio->bi_io_vec + ci->idx;
821 sector_t remaining = to_sector(bv->bv_len);
822 unsigned int offset = 0;
826 ti = dm_table_find_target(ci->map, ci->sector);
827 if (!dm_target_is_valid(ti))
830 max = max_io_len(ci->md, ci->sector, ti);
832 tio = alloc_tio(ci->md);
835 memset(&tio->info, 0, sizeof(tio->info));
838 len = min(remaining, max);
840 clone = split_bvec(bio, ci->sector, ci->idx,
841 bv->bv_offset + offset, len,
844 __map_bio(ti, clone, tio);
847 ci->sector_count -= len;
848 offset += to_bytes(len);
849 } while (remaining -= len);
858 * Split the bio into several clones and submit it to targets.
860 static void __split_and_process_bio(struct mapped_device *md, struct bio *bio)
862 struct clone_info ci;
865 ci.map = dm_get_table(md);
866 if (unlikely(!ci.map)) {
867 if (!bio_barrier(bio))
870 md->barrier_error = -EIO;
876 ci.io = alloc_io(md);
878 atomic_set(&ci.io->io_count, 1);
881 ci.sector = bio->bi_sector;
882 ci.sector_count = bio_sectors(bio);
883 ci.idx = bio->bi_idx;
885 start_io_acct(ci.io);
886 while (ci.sector_count && !error)
887 error = __clone_and_map(&ci);
889 /* drop the extra reference count */
890 dec_pending(ci.io, error);
891 dm_table_put(ci.map);
893 /*-----------------------------------------------------------------
895 *---------------------------------------------------------------*/
897 static int dm_merge_bvec(struct request_queue *q,
898 struct bvec_merge_data *bvm,
899 struct bio_vec *biovec)
901 struct mapped_device *md = q->queuedata;
902 struct dm_table *map = dm_get_table(md);
903 struct dm_target *ti;
904 sector_t max_sectors;
910 ti = dm_table_find_target(map, bvm->bi_sector);
911 if (!dm_target_is_valid(ti))
915 * Find maximum amount of I/O that won't need splitting
917 max_sectors = min(max_io_len(md, bvm->bi_sector, ti),
918 (sector_t) BIO_MAX_SECTORS);
919 max_size = (max_sectors << SECTOR_SHIFT) - bvm->bi_size;
924 * merge_bvec_fn() returns number of bytes
925 * it can accept at this offset
926 * max is precomputed maximal io size
928 if (max_size && ti->type->merge)
929 max_size = ti->type->merge(ti, bvm, biovec, max_size);
936 * Always allow an entire first page
938 if (max_size <= biovec->bv_len && !(bvm->bi_size >> SECTOR_SHIFT))
939 max_size = biovec->bv_len;
945 * The request function that just remaps the bio built up by
948 static int dm_request(struct request_queue *q, struct bio *bio)
950 int rw = bio_data_dir(bio);
951 struct mapped_device *md = q->queuedata;
954 down_read(&md->io_lock);
956 cpu = part_stat_lock();
957 part_stat_inc(cpu, &dm_disk(md)->part0, ios[rw]);
958 part_stat_add(cpu, &dm_disk(md)->part0, sectors[rw], bio_sectors(bio));
962 * If we're suspended or the thread is processing barriers
963 * we have to queue this io for later.
965 if (unlikely(test_bit(DMF_QUEUE_IO_TO_THREAD, &md->flags)) ||
966 unlikely(bio_barrier(bio))) {
967 up_read(&md->io_lock);
969 if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) &&
970 bio_rw(bio) == READA) {
980 __split_and_process_bio(md, bio);
981 up_read(&md->io_lock);
985 static void dm_unplug_all(struct request_queue *q)
987 struct mapped_device *md = q->queuedata;
988 struct dm_table *map = dm_get_table(md);
991 dm_table_unplug_all(map);
996 static int dm_any_congested(void *congested_data, int bdi_bits)
999 struct mapped_device *md = congested_data;
1000 struct dm_table *map;
1002 if (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
1003 map = dm_get_table(md);
1005 r = dm_table_any_congested(map, bdi_bits);
1013 /*-----------------------------------------------------------------
1014 * An IDR is used to keep track of allocated minor numbers.
1015 *---------------------------------------------------------------*/
1016 static DEFINE_IDR(_minor_idr);
1018 static void free_minor(int minor)
1020 spin_lock(&_minor_lock);
1021 idr_remove(&_minor_idr, minor);
1022 spin_unlock(&_minor_lock);
1026 * See if the device with a specific minor # is free.
1028 static int specific_minor(int minor)
1032 if (minor >= (1 << MINORBITS))
1035 r = idr_pre_get(&_minor_idr, GFP_KERNEL);
1039 spin_lock(&_minor_lock);
1041 if (idr_find(&_minor_idr, minor)) {
1046 r = idr_get_new_above(&_minor_idr, MINOR_ALLOCED, minor, &m);
1051 idr_remove(&_minor_idr, m);
1057 spin_unlock(&_minor_lock);
1061 static int next_free_minor(int *minor)
1065 r = idr_pre_get(&_minor_idr, GFP_KERNEL);
1069 spin_lock(&_minor_lock);
1071 r = idr_get_new(&_minor_idr, MINOR_ALLOCED, &m);
1075 if (m >= (1 << MINORBITS)) {
1076 idr_remove(&_minor_idr, m);
1084 spin_unlock(&_minor_lock);
1088 static struct block_device_operations dm_blk_dops;
1090 static void dm_wq_work(struct work_struct *work);
1093 * Allocate and initialise a blank device with a given minor.
1095 static struct mapped_device *alloc_dev(int minor)
1098 struct mapped_device *md = kzalloc(sizeof(*md), GFP_KERNEL);
1102 DMWARN("unable to allocate device, out of memory.");
1106 if (!try_module_get(THIS_MODULE))
1107 goto bad_module_get;
1109 /* get a minor number for the dev */
1110 if (minor == DM_ANY_MINOR)
1111 r = next_free_minor(&minor);
1113 r = specific_minor(minor);
1117 init_rwsem(&md->io_lock);
1118 mutex_init(&md->suspend_lock);
1119 spin_lock_init(&md->deferred_lock);
1120 rwlock_init(&md->map_lock);
1121 atomic_set(&md->holders, 1);
1122 atomic_set(&md->open_count, 0);
1123 atomic_set(&md->event_nr, 0);
1124 atomic_set(&md->uevent_seq, 0);
1125 INIT_LIST_HEAD(&md->uevent_list);
1126 spin_lock_init(&md->uevent_lock);
1128 md->queue = blk_alloc_queue(GFP_KERNEL);
1132 md->queue->queuedata = md;
1133 md->queue->backing_dev_info.congested_fn = dm_any_congested;
1134 md->queue->backing_dev_info.congested_data = md;
1135 blk_queue_make_request(md->queue, dm_request);
1136 blk_queue_ordered(md->queue, QUEUE_ORDERED_DRAIN, NULL);
1137 blk_queue_bounce_limit(md->queue, BLK_BOUNCE_ANY);
1138 md->queue->unplug_fn = dm_unplug_all;
1139 blk_queue_merge_bvec(md->queue, dm_merge_bvec);
1141 md->io_pool = mempool_create_slab_pool(MIN_IOS, _io_cache);
1145 md->tio_pool = mempool_create_slab_pool(MIN_IOS, _tio_cache);
1149 md->bs = bioset_create(16, 0);
1153 md->disk = alloc_disk(1);
1157 atomic_set(&md->pending, 0);
1158 init_waitqueue_head(&md->wait);
1159 INIT_WORK(&md->work, dm_wq_work);
1160 init_waitqueue_head(&md->eventq);
1162 md->disk->major = _major;
1163 md->disk->first_minor = minor;
1164 md->disk->fops = &dm_blk_dops;
1165 md->disk->queue = md->queue;
1166 md->disk->private_data = md;
1167 sprintf(md->disk->disk_name, "dm-%d", minor);
1169 format_dev_t(md->name, MKDEV(_major, minor));
1171 md->wq = create_singlethread_workqueue("kdmflush");
1175 /* Populate the mapping, nobody knows we exist yet */
1176 spin_lock(&_minor_lock);
1177 old_md = idr_replace(&_minor_idr, md, minor);
1178 spin_unlock(&_minor_lock);
1180 BUG_ON(old_md != MINOR_ALLOCED);
1187 bioset_free(md->bs);
1189 mempool_destroy(md->tio_pool);
1191 mempool_destroy(md->io_pool);
1193 blk_cleanup_queue(md->queue);
1197 module_put(THIS_MODULE);
1203 static void unlock_fs(struct mapped_device *md);
1205 static void free_dev(struct mapped_device *md)
1207 int minor = MINOR(disk_devt(md->disk));
1209 if (md->suspended_bdev) {
1211 bdput(md->suspended_bdev);
1213 destroy_workqueue(md->wq);
1214 mempool_destroy(md->tio_pool);
1215 mempool_destroy(md->io_pool);
1216 bioset_free(md->bs);
1217 blk_integrity_unregister(md->disk);
1218 del_gendisk(md->disk);
1221 spin_lock(&_minor_lock);
1222 md->disk->private_data = NULL;
1223 spin_unlock(&_minor_lock);
1226 blk_cleanup_queue(md->queue);
1227 module_put(THIS_MODULE);
1232 * Bind a table to the device.
1234 static void event_callback(void *context)
1236 unsigned long flags;
1238 struct mapped_device *md = (struct mapped_device *) context;
1240 spin_lock_irqsave(&md->uevent_lock, flags);
1241 list_splice_init(&md->uevent_list, &uevents);
1242 spin_unlock_irqrestore(&md->uevent_lock, flags);
1244 dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj);
1246 atomic_inc(&md->event_nr);
1247 wake_up(&md->eventq);
1250 static void __set_size(struct mapped_device *md, sector_t size)
1252 set_capacity(md->disk, size);
1254 mutex_lock(&md->suspended_bdev->bd_inode->i_mutex);
1255 i_size_write(md->suspended_bdev->bd_inode, (loff_t)size << SECTOR_SHIFT);
1256 mutex_unlock(&md->suspended_bdev->bd_inode->i_mutex);
1259 static int __bind(struct mapped_device *md, struct dm_table *t)
1261 struct request_queue *q = md->queue;
1264 size = dm_table_get_size(t);
1267 * Wipe any geometry if the size of the table changed.
1269 if (size != get_capacity(md->disk))
1270 memset(&md->geometry, 0, sizeof(md->geometry));
1272 if (md->suspended_bdev)
1273 __set_size(md, size);
1276 dm_table_destroy(t);
1280 dm_table_event_callback(t, event_callback, md);
1282 write_lock(&md->map_lock);
1284 dm_table_set_restrictions(t, q);
1285 write_unlock(&md->map_lock);
1290 static void __unbind(struct mapped_device *md)
1292 struct dm_table *map = md->map;
1297 dm_table_event_callback(map, NULL, NULL);
1298 write_lock(&md->map_lock);
1300 write_unlock(&md->map_lock);
1301 dm_table_destroy(map);
1305 * Constructor for a new device.
1307 int dm_create(int minor, struct mapped_device **result)
1309 struct mapped_device *md;
1311 md = alloc_dev(minor);
1321 static struct mapped_device *dm_find_md(dev_t dev)
1323 struct mapped_device *md;
1324 unsigned minor = MINOR(dev);
1326 if (MAJOR(dev) != _major || minor >= (1 << MINORBITS))
1329 spin_lock(&_minor_lock);
1331 md = idr_find(&_minor_idr, minor);
1332 if (md && (md == MINOR_ALLOCED ||
1333 (MINOR(disk_devt(dm_disk(md))) != minor) ||
1334 test_bit(DMF_FREEING, &md->flags))) {
1340 spin_unlock(&_minor_lock);
1345 struct mapped_device *dm_get_md(dev_t dev)
1347 struct mapped_device *md = dm_find_md(dev);
1355 void *dm_get_mdptr(struct mapped_device *md)
1357 return md->interface_ptr;
1360 void dm_set_mdptr(struct mapped_device *md, void *ptr)
1362 md->interface_ptr = ptr;
1365 void dm_get(struct mapped_device *md)
1367 atomic_inc(&md->holders);
1370 const char *dm_device_name(struct mapped_device *md)
1374 EXPORT_SYMBOL_GPL(dm_device_name);
1376 void dm_put(struct mapped_device *md)
1378 struct dm_table *map;
1380 BUG_ON(test_bit(DMF_FREEING, &md->flags));
1382 if (atomic_dec_and_lock(&md->holders, &_minor_lock)) {
1383 map = dm_get_table(md);
1384 idr_replace(&_minor_idr, MINOR_ALLOCED,
1385 MINOR(disk_devt(dm_disk(md))));
1386 set_bit(DMF_FREEING, &md->flags);
1387 spin_unlock(&_minor_lock);
1388 if (!dm_suspended(md)) {
1389 dm_table_presuspend_targets(map);
1390 dm_table_postsuspend_targets(map);
1398 EXPORT_SYMBOL_GPL(dm_put);
1400 static int dm_wait_for_completion(struct mapped_device *md, int interruptible)
1403 DECLARE_WAITQUEUE(wait, current);
1405 dm_unplug_all(md->queue);
1407 add_wait_queue(&md->wait, &wait);
1410 set_current_state(interruptible);
1413 if (!atomic_read(&md->pending))
1416 if (interruptible == TASK_INTERRUPTIBLE &&
1417 signal_pending(current)) {
1424 set_current_state(TASK_RUNNING);
1426 remove_wait_queue(&md->wait, &wait);
1431 static int dm_flush(struct mapped_device *md)
1433 dm_wait_for_completion(md, TASK_UNINTERRUPTIBLE);
1437 static void process_barrier(struct mapped_device *md, struct bio *bio)
1439 int error = dm_flush(md);
1441 if (unlikely(error)) {
1442 bio_endio(bio, error);
1445 if (bio_empty_barrier(bio)) {
1450 __split_and_process_bio(md, bio);
1452 error = dm_flush(md);
1454 if (!error && md->barrier_error)
1455 error = md->barrier_error;
1457 if (md->barrier_error != DM_ENDIO_REQUEUE)
1458 bio_endio(bio, error);
1462 * Process the deferred bios
1464 static void dm_wq_work(struct work_struct *work)
1466 struct mapped_device *md = container_of(work, struct mapped_device,
1470 down_write(&md->io_lock);
1472 while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
1473 spin_lock_irq(&md->deferred_lock);
1474 c = bio_list_pop(&md->deferred);
1475 spin_unlock_irq(&md->deferred_lock);
1478 clear_bit(DMF_QUEUE_IO_TO_THREAD, &md->flags);
1482 up_write(&md->io_lock);
1485 process_barrier(md, c);
1487 __split_and_process_bio(md, c);
1489 down_write(&md->io_lock);
1492 up_write(&md->io_lock);
1495 static void dm_queue_flush(struct mapped_device *md)
1497 clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
1498 smp_mb__after_clear_bit();
1499 queue_work(md->wq, &md->work);
1503 * Swap in a new table (destroying old one).
1505 int dm_swap_table(struct mapped_device *md, struct dm_table *table)
1509 mutex_lock(&md->suspend_lock);
1511 /* device must be suspended */
1512 if (!dm_suspended(md))
1515 /* without bdev, the device size cannot be changed */
1516 if (!md->suspended_bdev)
1517 if (get_capacity(md->disk) != dm_table_get_size(table))
1521 r = __bind(md, table);
1524 mutex_unlock(&md->suspend_lock);
1529 * Functions to lock and unlock any filesystem running on the
1532 static int lock_fs(struct mapped_device *md)
1536 WARN_ON(md->frozen_sb);
1538 md->frozen_sb = freeze_bdev(md->suspended_bdev);
1539 if (IS_ERR(md->frozen_sb)) {
1540 r = PTR_ERR(md->frozen_sb);
1541 md->frozen_sb = NULL;
1545 set_bit(DMF_FROZEN, &md->flags);
1547 /* don't bdput right now, we don't want the bdev
1548 * to go away while it is locked.
1553 static void unlock_fs(struct mapped_device *md)
1555 if (!test_bit(DMF_FROZEN, &md->flags))
1558 thaw_bdev(md->suspended_bdev, md->frozen_sb);
1559 md->frozen_sb = NULL;
1560 clear_bit(DMF_FROZEN, &md->flags);
1564 * We need to be able to change a mapping table under a mounted
1565 * filesystem. For example we might want to move some data in
1566 * the background. Before the table can be swapped with
1567 * dm_bind_table, dm_suspend must be called to flush any in
1568 * flight bios and ensure that any further io gets deferred.
1570 int dm_suspend(struct mapped_device *md, unsigned suspend_flags)
1572 struct dm_table *map = NULL;
1574 int do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG ? 1 : 0;
1575 int noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG ? 1 : 0;
1577 mutex_lock(&md->suspend_lock);
1579 if (dm_suspended(md)) {
1584 map = dm_get_table(md);
1587 * DMF_NOFLUSH_SUSPENDING must be set before presuspend.
1588 * This flag is cleared before dm_suspend returns.
1591 set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
1593 /* This does not get reverted if there's an error later. */
1594 dm_table_presuspend_targets(map);
1596 /* bdget() can stall if the pending I/Os are not flushed */
1598 md->suspended_bdev = bdget_disk(md->disk, 0);
1599 if (!md->suspended_bdev) {
1600 DMWARN("bdget failed in dm_suspend");
1606 * Flush I/O to the device. noflush supersedes do_lockfs,
1607 * because lock_fs() needs to flush I/Os.
1617 * Here we must make sure that no processes are submitting requests
1618 * to target drivers i.e. no one may be executing
1619 * __split_and_process_bio. This is called from dm_request and
1622 * To get all processes out of __split_and_process_bio in dm_request,
1623 * we take the write lock. To prevent any process from reentering
1624 * __split_and_process_bio from dm_request, we set
1625 * DMF_QUEUE_IO_TO_THREAD.
1627 * To quiesce the thread (dm_wq_work), we set DMF_BLOCK_IO_FOR_SUSPEND
1628 * and call flush_workqueue(md->wq). flush_workqueue will wait until
1629 * dm_wq_work exits and DMF_BLOCK_IO_FOR_SUSPEND will prevent any
1630 * further calls to __split_and_process_bio from dm_wq_work.
1632 down_write(&md->io_lock);
1633 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
1634 set_bit(DMF_QUEUE_IO_TO_THREAD, &md->flags);
1635 up_write(&md->io_lock);
1637 flush_workqueue(md->wq);
1640 * At this point no more requests are entering target request routines.
1641 * We call dm_wait_for_completion to wait for all existing requests
1644 r = dm_wait_for_completion(md, TASK_INTERRUPTIBLE);
1646 down_write(&md->io_lock);
1648 clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
1649 up_write(&md->io_lock);
1651 /* were we interrupted ? */
1656 goto out; /* pushback list is already flushed, so skip flush */
1660 * If dm_wait_for_completion returned 0, the device is completely
1661 * quiescent now. There is no request-processing activity. All new
1662 * requests are being added to md->deferred list.
1665 dm_table_postsuspend_targets(map);
1667 set_bit(DMF_SUSPENDED, &md->flags);
1670 if (r && md->suspended_bdev) {
1671 bdput(md->suspended_bdev);
1672 md->suspended_bdev = NULL;
1678 mutex_unlock(&md->suspend_lock);
1682 int dm_resume(struct mapped_device *md)
1685 struct dm_table *map = NULL;
1687 mutex_lock(&md->suspend_lock);
1688 if (!dm_suspended(md))
1691 map = dm_get_table(md);
1692 if (!map || !dm_table_get_size(map))
1695 r = dm_table_resume_targets(map);
1703 if (md->suspended_bdev) {
1704 bdput(md->suspended_bdev);
1705 md->suspended_bdev = NULL;
1708 clear_bit(DMF_SUSPENDED, &md->flags);
1710 dm_table_unplug_all(map);
1712 dm_kobject_uevent(md);
1718 mutex_unlock(&md->suspend_lock);
1723 /*-----------------------------------------------------------------
1724 * Event notification.
1725 *---------------------------------------------------------------*/
1726 void dm_kobject_uevent(struct mapped_device *md)
1728 kobject_uevent(&disk_to_dev(md->disk)->kobj, KOBJ_CHANGE);
1731 uint32_t dm_next_uevent_seq(struct mapped_device *md)
1733 return atomic_add_return(1, &md->uevent_seq);
1736 uint32_t dm_get_event_nr(struct mapped_device *md)
1738 return atomic_read(&md->event_nr);
1741 int dm_wait_event(struct mapped_device *md, int event_nr)
1743 return wait_event_interruptible(md->eventq,
1744 (event_nr != atomic_read(&md->event_nr)));
1747 void dm_uevent_add(struct mapped_device *md, struct list_head *elist)
1749 unsigned long flags;
1751 spin_lock_irqsave(&md->uevent_lock, flags);
1752 list_add(elist, &md->uevent_list);
1753 spin_unlock_irqrestore(&md->uevent_lock, flags);
1757 * The gendisk is only valid as long as you have a reference
1760 struct gendisk *dm_disk(struct mapped_device *md)
1765 struct kobject *dm_kobject(struct mapped_device *md)
1771 * struct mapped_device should not be exported outside of dm.c
1772 * so use this check to verify that kobj is part of md structure
1774 struct mapped_device *dm_get_from_kobject(struct kobject *kobj)
1776 struct mapped_device *md;
1778 md = container_of(kobj, struct mapped_device, kobj);
1779 if (&md->kobj != kobj)
1786 int dm_suspended(struct mapped_device *md)
1788 return test_bit(DMF_SUSPENDED, &md->flags);
1791 int dm_noflush_suspending(struct dm_target *ti)
1793 struct mapped_device *md = dm_table_get_md(ti->table);
1794 int r = __noflush_suspending(md);
1800 EXPORT_SYMBOL_GPL(dm_noflush_suspending);
1802 static struct block_device_operations dm_blk_dops = {
1803 .open = dm_blk_open,
1804 .release = dm_blk_close,
1805 .ioctl = dm_blk_ioctl,
1806 .getgeo = dm_blk_getgeo,
1807 .owner = THIS_MODULE
1810 EXPORT_SYMBOL(dm_get_mapinfo);
1815 module_init(dm_init);
1816 module_exit(dm_exit);
1818 module_param(major, uint, 0);
1819 MODULE_PARM_DESC(major, "The major number of the device mapper");
1820 MODULE_DESCRIPTION(DM_NAME " driver");
1821 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
1822 MODULE_LICENSE("GPL");