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>
23 #include <trace/events/block.h>
25 #define DM_MSG_PREFIX "core"
28 * Cookies are numeric values sent with CHANGE and REMOVE
29 * uevents while resuming, removing or renaming the device.
31 #define DM_COOKIE_ENV_VAR_NAME "DM_COOKIE"
32 #define DM_COOKIE_LENGTH 24
34 static const char *_name = DM_NAME;
36 static unsigned int major = 0;
37 static unsigned int _major = 0;
39 static DEFINE_SPINLOCK(_minor_lock);
42 * One of these is allocated per bio.
45 struct mapped_device *md;
49 unsigned long start_time;
54 * One of these is allocated per target within a bio. Hopefully
55 * this will be simplified out one day.
64 * For request-based dm.
65 * One of these is allocated per request.
67 struct dm_rq_target_io {
68 struct mapped_device *md;
70 struct request *orig, clone;
76 * For request-based dm.
77 * One of these is allocated per bio.
79 struct dm_rq_clone_bio_info {
84 union map_info *dm_get_mapinfo(struct bio *bio)
86 if (bio && bio->bi_private)
87 return &((struct dm_target_io *)bio->bi_private)->info;
91 #define MINOR_ALLOCED ((void *)-1)
94 * Bits for the md->flags field.
96 #define DMF_BLOCK_IO_FOR_SUSPEND 0
97 #define DMF_SUSPENDED 1
100 #define DMF_DELETING 4
101 #define DMF_NOFLUSH_SUSPENDING 5
102 #define DMF_QUEUE_IO_TO_THREAD 6
105 * Work processed by per-device workqueue.
107 struct mapped_device {
108 struct rw_semaphore io_lock;
109 struct mutex suspend_lock;
116 struct request_queue *queue;
117 struct gendisk *disk;
123 * A list of ios that arrived while we were suspended.
126 wait_queue_head_t wait;
127 struct work_struct work;
128 struct bio_list deferred;
129 spinlock_t deferred_lock;
132 * An error from the barrier request currently being processed.
137 * Processing queue (flush/barriers)
139 struct workqueue_struct *wq;
142 * The current mapping.
144 struct dm_table *map;
147 * io objects are allocated from here.
158 wait_queue_head_t eventq;
160 struct list_head uevent_list;
161 spinlock_t uevent_lock; /* Protect access to uevent_list */
164 * freeze/thaw support require holding onto a super block
166 struct super_block *frozen_sb;
167 struct block_device *bdev;
169 /* forced geometry settings */
170 struct hd_geometry geometry;
175 /* zero-length barrier that will be cloned and submitted to targets */
176 struct bio barrier_bio;
180 static struct kmem_cache *_io_cache;
181 static struct kmem_cache *_tio_cache;
182 static struct kmem_cache *_rq_tio_cache;
183 static struct kmem_cache *_rq_bio_info_cache;
185 static int __init local_init(void)
189 /* allocate a slab for the dm_ios */
190 _io_cache = KMEM_CACHE(dm_io, 0);
194 /* allocate a slab for the target ios */
195 _tio_cache = KMEM_CACHE(dm_target_io, 0);
197 goto out_free_io_cache;
199 _rq_tio_cache = KMEM_CACHE(dm_rq_target_io, 0);
201 goto out_free_tio_cache;
203 _rq_bio_info_cache = KMEM_CACHE(dm_rq_clone_bio_info, 0);
204 if (!_rq_bio_info_cache)
205 goto out_free_rq_tio_cache;
207 r = dm_uevent_init();
209 goto out_free_rq_bio_info_cache;
212 r = register_blkdev(_major, _name);
214 goto out_uevent_exit;
223 out_free_rq_bio_info_cache:
224 kmem_cache_destroy(_rq_bio_info_cache);
225 out_free_rq_tio_cache:
226 kmem_cache_destroy(_rq_tio_cache);
228 kmem_cache_destroy(_tio_cache);
230 kmem_cache_destroy(_io_cache);
235 static void local_exit(void)
237 kmem_cache_destroy(_rq_bio_info_cache);
238 kmem_cache_destroy(_rq_tio_cache);
239 kmem_cache_destroy(_tio_cache);
240 kmem_cache_destroy(_io_cache);
241 unregister_blkdev(_major, _name);
246 DMINFO("cleaned up");
249 static int (*_inits[])(void) __initdata = {
258 static void (*_exits[])(void) = {
267 static int __init dm_init(void)
269 const int count = ARRAY_SIZE(_inits);
273 for (i = 0; i < count; i++) {
288 static void __exit dm_exit(void)
290 int i = ARRAY_SIZE(_exits);
297 * Block device functions
299 static int dm_blk_open(struct block_device *bdev, fmode_t mode)
301 struct mapped_device *md;
303 spin_lock(&_minor_lock);
305 md = bdev->bd_disk->private_data;
309 if (test_bit(DMF_FREEING, &md->flags) ||
310 test_bit(DMF_DELETING, &md->flags)) {
316 atomic_inc(&md->open_count);
319 spin_unlock(&_minor_lock);
321 return md ? 0 : -ENXIO;
324 static int dm_blk_close(struct gendisk *disk, fmode_t mode)
326 struct mapped_device *md = disk->private_data;
327 atomic_dec(&md->open_count);
332 int dm_open_count(struct mapped_device *md)
334 return atomic_read(&md->open_count);
338 * Guarantees nothing is using the device before it's deleted.
340 int dm_lock_for_deletion(struct mapped_device *md)
344 spin_lock(&_minor_lock);
346 if (dm_open_count(md))
349 set_bit(DMF_DELETING, &md->flags);
351 spin_unlock(&_minor_lock);
356 static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo)
358 struct mapped_device *md = bdev->bd_disk->private_data;
360 return dm_get_geometry(md, geo);
363 static int dm_blk_ioctl(struct block_device *bdev, fmode_t mode,
364 unsigned int cmd, unsigned long arg)
366 struct mapped_device *md = bdev->bd_disk->private_data;
367 struct dm_table *map = dm_get_table(md);
368 struct dm_target *tgt;
371 if (!map || !dm_table_get_size(map))
374 /* We only support devices that have a single target */
375 if (dm_table_get_num_targets(map) != 1)
378 tgt = dm_table_get_target(map, 0);
380 if (dm_suspended(md)) {
385 if (tgt->type->ioctl)
386 r = tgt->type->ioctl(tgt, cmd, arg);
394 static struct dm_io *alloc_io(struct mapped_device *md)
396 return mempool_alloc(md->io_pool, GFP_NOIO);
399 static void free_io(struct mapped_device *md, struct dm_io *io)
401 mempool_free(io, md->io_pool);
404 static void free_tio(struct mapped_device *md, struct dm_target_io *tio)
406 mempool_free(tio, md->tio_pool);
409 static void start_io_acct(struct dm_io *io)
411 struct mapped_device *md = io->md;
414 io->start_time = jiffies;
416 cpu = part_stat_lock();
417 part_round_stats(cpu, &dm_disk(md)->part0);
419 dm_disk(md)->part0.in_flight = atomic_inc_return(&md->pending);
422 static void end_io_acct(struct dm_io *io)
424 struct mapped_device *md = io->md;
425 struct bio *bio = io->bio;
426 unsigned long duration = jiffies - io->start_time;
428 int rw = bio_data_dir(bio);
430 cpu = part_stat_lock();
431 part_round_stats(cpu, &dm_disk(md)->part0);
432 part_stat_add(cpu, &dm_disk(md)->part0, ticks[rw], duration);
436 * After this is decremented the bio must not be touched if it is
439 dm_disk(md)->part0.in_flight = pending =
440 atomic_dec_return(&md->pending);
442 /* nudge anyone waiting on suspend queue */
448 * Add the bio to the list of deferred io.
450 static void queue_io(struct mapped_device *md, struct bio *bio)
452 down_write(&md->io_lock);
454 spin_lock_irq(&md->deferred_lock);
455 bio_list_add(&md->deferred, bio);
456 spin_unlock_irq(&md->deferred_lock);
458 if (!test_and_set_bit(DMF_QUEUE_IO_TO_THREAD, &md->flags))
459 queue_work(md->wq, &md->work);
461 up_write(&md->io_lock);
465 * Everyone (including functions in this file), should use this
466 * function to access the md->map field, and make sure they call
467 * dm_table_put() when finished.
469 struct dm_table *dm_get_table(struct mapped_device *md)
473 read_lock(&md->map_lock);
477 read_unlock(&md->map_lock);
483 * Get the geometry associated with a dm device
485 int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo)
493 * Set the geometry of a device.
495 int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo)
497 sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors;
499 if (geo->start > sz) {
500 DMWARN("Start sector is beyond the geometry limits.");
509 /*-----------------------------------------------------------------
511 * A more elegant soln is in the works that uses the queue
512 * merge fn, unfortunately there are a couple of changes to
513 * the block layer that I want to make for this. So in the
514 * interests of getting something for people to use I give
515 * you this clearly demarcated crap.
516 *---------------------------------------------------------------*/
518 static int __noflush_suspending(struct mapped_device *md)
520 return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
524 * Decrements the number of outstanding ios that a bio has been
525 * cloned into, completing the original io if necc.
527 static void dec_pending(struct dm_io *io, int error)
532 struct mapped_device *md = io->md;
534 /* Push-back supersedes any I/O errors */
535 if (error && !(io->error > 0 && __noflush_suspending(md)))
538 if (atomic_dec_and_test(&io->io_count)) {
539 if (io->error == DM_ENDIO_REQUEUE) {
541 * Target requested pushing back the I/O.
543 spin_lock_irqsave(&md->deferred_lock, flags);
544 if (__noflush_suspending(md)) {
545 if (!bio_barrier(io->bio))
546 bio_list_add_head(&md->deferred,
549 /* noflush suspend was interrupted. */
551 spin_unlock_irqrestore(&md->deferred_lock, flags);
554 io_error = io->error;
557 if (bio_barrier(bio)) {
559 * There can be just one barrier request so we use
560 * a per-device variable for error reporting.
561 * Note that you can't touch the bio after end_io_acct
563 if (!md->barrier_error && io_error != -EOPNOTSUPP)
564 md->barrier_error = io_error;
569 if (io_error != DM_ENDIO_REQUEUE) {
570 trace_block_bio_complete(md->queue, bio);
572 bio_endio(bio, io_error);
580 static void clone_endio(struct bio *bio, int error)
583 struct dm_target_io *tio = bio->bi_private;
584 struct dm_io *io = tio->io;
585 struct mapped_device *md = tio->io->md;
586 dm_endio_fn endio = tio->ti->type->end_io;
588 if (!bio_flagged(bio, BIO_UPTODATE) && !error)
592 r = endio(tio->ti, bio, error, &tio->info);
593 if (r < 0 || r == DM_ENDIO_REQUEUE)
595 * error and requeue request are handled
599 else if (r == DM_ENDIO_INCOMPLETE)
600 /* The target will handle the io */
603 DMWARN("unimplemented target endio return value: %d", r);
609 * Store md for cleanup instead of tio which is about to get freed.
611 bio->bi_private = md->bs;
615 dec_pending(io, error);
618 static sector_t max_io_len(struct mapped_device *md,
619 sector_t sector, struct dm_target *ti)
621 sector_t offset = sector - ti->begin;
622 sector_t len = ti->len - offset;
625 * Does the target need to split even further ?
629 boundary = ((offset + ti->split_io) & ~(ti->split_io - 1))
638 static void __map_bio(struct dm_target *ti, struct bio *clone,
639 struct dm_target_io *tio)
643 struct mapped_device *md;
645 clone->bi_end_io = clone_endio;
646 clone->bi_private = tio;
649 * Map the clone. If r == 0 we don't need to do
650 * anything, the target has assumed ownership of
653 atomic_inc(&tio->io->io_count);
654 sector = clone->bi_sector;
655 r = ti->type->map(ti, clone, &tio->info);
656 if (r == DM_MAPIO_REMAPPED) {
657 /* the bio has been remapped so dispatch it */
659 trace_block_remap(bdev_get_queue(clone->bi_bdev), clone,
660 tio->io->bio->bi_bdev->bd_dev, sector);
662 generic_make_request(clone);
663 } else if (r < 0 || r == DM_MAPIO_REQUEUE) {
664 /* error the io and bail out, or requeue it if needed */
666 dec_pending(tio->io, r);
668 * Store bio_set for cleanup.
670 clone->bi_private = md->bs;
674 DMWARN("unimplemented target map return value: %d", r);
680 struct mapped_device *md;
681 struct dm_table *map;
685 sector_t sector_count;
689 static void dm_bio_destructor(struct bio *bio)
691 struct bio_set *bs = bio->bi_private;
697 * Creates a little bio that is just does part of a bvec.
699 static struct bio *split_bvec(struct bio *bio, sector_t sector,
700 unsigned short idx, unsigned int offset,
701 unsigned int len, struct bio_set *bs)
704 struct bio_vec *bv = bio->bi_io_vec + idx;
706 clone = bio_alloc_bioset(GFP_NOIO, 1, bs);
707 clone->bi_destructor = dm_bio_destructor;
708 *clone->bi_io_vec = *bv;
710 clone->bi_sector = sector;
711 clone->bi_bdev = bio->bi_bdev;
712 clone->bi_rw = bio->bi_rw & ~(1 << BIO_RW_BARRIER);
714 clone->bi_size = to_bytes(len);
715 clone->bi_io_vec->bv_offset = offset;
716 clone->bi_io_vec->bv_len = clone->bi_size;
717 clone->bi_flags |= 1 << BIO_CLONED;
719 if (bio_integrity(bio)) {
720 bio_integrity_clone(clone, bio, GFP_NOIO);
721 bio_integrity_trim(clone,
722 bio_sector_offset(bio, idx, offset), len);
729 * Creates a bio that consists of range of complete bvecs.
731 static struct bio *clone_bio(struct bio *bio, sector_t sector,
732 unsigned short idx, unsigned short bv_count,
733 unsigned int len, struct bio_set *bs)
737 clone = bio_alloc_bioset(GFP_NOIO, bio->bi_max_vecs, bs);
738 __bio_clone(clone, bio);
739 clone->bi_rw &= ~(1 << BIO_RW_BARRIER);
740 clone->bi_destructor = dm_bio_destructor;
741 clone->bi_sector = sector;
743 clone->bi_vcnt = idx + bv_count;
744 clone->bi_size = to_bytes(len);
745 clone->bi_flags &= ~(1 << BIO_SEG_VALID);
747 if (bio_integrity(bio)) {
748 bio_integrity_clone(clone, bio, GFP_NOIO);
750 if (idx != bio->bi_idx || clone->bi_size < bio->bi_size)
751 bio_integrity_trim(clone,
752 bio_sector_offset(bio, idx, 0), len);
758 static struct dm_target_io *alloc_tio(struct clone_info *ci,
759 struct dm_target *ti)
761 struct dm_target_io *tio = mempool_alloc(ci->md->tio_pool, GFP_NOIO);
765 memset(&tio->info, 0, sizeof(tio->info));
770 static void __flush_target(struct clone_info *ci, struct dm_target *ti,
773 struct dm_target_io *tio = alloc_tio(ci, ti);
776 tio->info.flush_request = flush_nr;
778 clone = bio_alloc_bioset(GFP_NOIO, 0, ci->md->bs);
779 __bio_clone(clone, ci->bio);
780 clone->bi_destructor = dm_bio_destructor;
782 __map_bio(ti, clone, tio);
785 static int __clone_and_map_empty_barrier(struct clone_info *ci)
787 unsigned target_nr = 0, flush_nr;
788 struct dm_target *ti;
790 while ((ti = dm_table_get_target(ci->map, target_nr++)))
791 for (flush_nr = 0; flush_nr < ti->num_flush_requests;
793 __flush_target(ci, ti, flush_nr);
795 ci->sector_count = 0;
800 static int __clone_and_map(struct clone_info *ci)
802 struct bio *clone, *bio = ci->bio;
803 struct dm_target *ti;
804 sector_t len = 0, max;
805 struct dm_target_io *tio;
807 if (unlikely(bio_empty_barrier(bio)))
808 return __clone_and_map_empty_barrier(ci);
810 ti = dm_table_find_target(ci->map, ci->sector);
811 if (!dm_target_is_valid(ti))
814 max = max_io_len(ci->md, ci->sector, ti);
817 * Allocate a target io object.
819 tio = alloc_tio(ci, ti);
821 if (ci->sector_count <= max) {
823 * Optimise for the simple case where we can do all of
824 * the remaining io with a single clone.
826 clone = clone_bio(bio, ci->sector, ci->idx,
827 bio->bi_vcnt - ci->idx, ci->sector_count,
829 __map_bio(ti, clone, tio);
830 ci->sector_count = 0;
832 } else if (to_sector(bio->bi_io_vec[ci->idx].bv_len) <= max) {
834 * There are some bvecs that don't span targets.
835 * Do as many of these as possible.
838 sector_t remaining = max;
841 for (i = ci->idx; remaining && (i < bio->bi_vcnt); i++) {
842 bv_len = to_sector(bio->bi_io_vec[i].bv_len);
844 if (bv_len > remaining)
851 clone = clone_bio(bio, ci->sector, ci->idx, i - ci->idx, len,
853 __map_bio(ti, clone, tio);
856 ci->sector_count -= len;
861 * Handle a bvec that must be split between two or more targets.
863 struct bio_vec *bv = bio->bi_io_vec + ci->idx;
864 sector_t remaining = to_sector(bv->bv_len);
865 unsigned int offset = 0;
869 ti = dm_table_find_target(ci->map, ci->sector);
870 if (!dm_target_is_valid(ti))
873 max = max_io_len(ci->md, ci->sector, ti);
875 tio = alloc_tio(ci, ti);
878 len = min(remaining, max);
880 clone = split_bvec(bio, ci->sector, ci->idx,
881 bv->bv_offset + offset, len,
884 __map_bio(ti, clone, tio);
887 ci->sector_count -= len;
888 offset += to_bytes(len);
889 } while (remaining -= len);
898 * Split the bio into several clones and submit it to targets.
900 static void __split_and_process_bio(struct mapped_device *md, struct bio *bio)
902 struct clone_info ci;
905 ci.map = dm_get_table(md);
906 if (unlikely(!ci.map)) {
907 if (!bio_barrier(bio))
910 if (!md->barrier_error)
911 md->barrier_error = -EIO;
917 ci.io = alloc_io(md);
919 atomic_set(&ci.io->io_count, 1);
922 ci.sector = bio->bi_sector;
923 ci.sector_count = bio_sectors(bio);
924 if (unlikely(bio_empty_barrier(bio)))
926 ci.idx = bio->bi_idx;
928 start_io_acct(ci.io);
929 while (ci.sector_count && !error)
930 error = __clone_and_map(&ci);
932 /* drop the extra reference count */
933 dec_pending(ci.io, error);
934 dm_table_put(ci.map);
936 /*-----------------------------------------------------------------
938 *---------------------------------------------------------------*/
940 static int dm_merge_bvec(struct request_queue *q,
941 struct bvec_merge_data *bvm,
942 struct bio_vec *biovec)
944 struct mapped_device *md = q->queuedata;
945 struct dm_table *map = dm_get_table(md);
946 struct dm_target *ti;
947 sector_t max_sectors;
953 ti = dm_table_find_target(map, bvm->bi_sector);
954 if (!dm_target_is_valid(ti))
958 * Find maximum amount of I/O that won't need splitting
960 max_sectors = min(max_io_len(md, bvm->bi_sector, ti),
961 (sector_t) BIO_MAX_SECTORS);
962 max_size = (max_sectors << SECTOR_SHIFT) - bvm->bi_size;
967 * merge_bvec_fn() returns number of bytes
968 * it can accept at this offset
969 * max is precomputed maximal io size
971 if (max_size && ti->type->merge)
972 max_size = ti->type->merge(ti, bvm, biovec, max_size);
974 * If the target doesn't support merge method and some of the devices
975 * provided their merge_bvec method (we know this by looking at
976 * queue_max_hw_sectors), then we can't allow bios with multiple vector
977 * entries. So always set max_size to 0, and the code below allows
980 else if (queue_max_hw_sectors(q) <= PAGE_SIZE >> 9)
989 * Always allow an entire first page
991 if (max_size <= biovec->bv_len && !(bvm->bi_size >> SECTOR_SHIFT))
992 max_size = biovec->bv_len;
998 * The request function that just remaps the bio built up by
1001 static int dm_request(struct request_queue *q, struct bio *bio)
1003 int rw = bio_data_dir(bio);
1004 struct mapped_device *md = q->queuedata;
1007 down_read(&md->io_lock);
1009 cpu = part_stat_lock();
1010 part_stat_inc(cpu, &dm_disk(md)->part0, ios[rw]);
1011 part_stat_add(cpu, &dm_disk(md)->part0, sectors[rw], bio_sectors(bio));
1015 * If we're suspended or the thread is processing barriers
1016 * we have to queue this io for later.
1018 if (unlikely(test_bit(DMF_QUEUE_IO_TO_THREAD, &md->flags)) ||
1019 unlikely(bio_barrier(bio))) {
1020 up_read(&md->io_lock);
1022 if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) &&
1023 bio_rw(bio) == READA) {
1033 __split_and_process_bio(md, bio);
1034 up_read(&md->io_lock);
1038 static void dm_unplug_all(struct request_queue *q)
1040 struct mapped_device *md = q->queuedata;
1041 struct dm_table *map = dm_get_table(md);
1044 dm_table_unplug_all(map);
1049 static int dm_any_congested(void *congested_data, int bdi_bits)
1052 struct mapped_device *md = congested_data;
1053 struct dm_table *map;
1055 if (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
1056 map = dm_get_table(md);
1058 r = dm_table_any_congested(map, bdi_bits);
1066 /*-----------------------------------------------------------------
1067 * An IDR is used to keep track of allocated minor numbers.
1068 *---------------------------------------------------------------*/
1069 static DEFINE_IDR(_minor_idr);
1071 static void free_minor(int minor)
1073 spin_lock(&_minor_lock);
1074 idr_remove(&_minor_idr, minor);
1075 spin_unlock(&_minor_lock);
1079 * See if the device with a specific minor # is free.
1081 static int specific_minor(int minor)
1085 if (minor >= (1 << MINORBITS))
1088 r = idr_pre_get(&_minor_idr, GFP_KERNEL);
1092 spin_lock(&_minor_lock);
1094 if (idr_find(&_minor_idr, minor)) {
1099 r = idr_get_new_above(&_minor_idr, MINOR_ALLOCED, minor, &m);
1104 idr_remove(&_minor_idr, m);
1110 spin_unlock(&_minor_lock);
1114 static int next_free_minor(int *minor)
1118 r = idr_pre_get(&_minor_idr, GFP_KERNEL);
1122 spin_lock(&_minor_lock);
1124 r = idr_get_new(&_minor_idr, MINOR_ALLOCED, &m);
1128 if (m >= (1 << MINORBITS)) {
1129 idr_remove(&_minor_idr, m);
1137 spin_unlock(&_minor_lock);
1141 static struct block_device_operations dm_blk_dops;
1143 static void dm_wq_work(struct work_struct *work);
1146 * Allocate and initialise a blank device with a given minor.
1148 static struct mapped_device *alloc_dev(int minor)
1151 struct mapped_device *md = kzalloc(sizeof(*md), GFP_KERNEL);
1155 DMWARN("unable to allocate device, out of memory.");
1159 if (!try_module_get(THIS_MODULE))
1160 goto bad_module_get;
1162 /* get a minor number for the dev */
1163 if (minor == DM_ANY_MINOR)
1164 r = next_free_minor(&minor);
1166 r = specific_minor(minor);
1170 init_rwsem(&md->io_lock);
1171 mutex_init(&md->suspend_lock);
1172 spin_lock_init(&md->deferred_lock);
1173 rwlock_init(&md->map_lock);
1174 atomic_set(&md->holders, 1);
1175 atomic_set(&md->open_count, 0);
1176 atomic_set(&md->event_nr, 0);
1177 atomic_set(&md->uevent_seq, 0);
1178 INIT_LIST_HEAD(&md->uevent_list);
1179 spin_lock_init(&md->uevent_lock);
1181 md->queue = blk_alloc_queue(GFP_KERNEL);
1185 md->queue->queuedata = md;
1186 md->queue->backing_dev_info.congested_fn = dm_any_congested;
1187 md->queue->backing_dev_info.congested_data = md;
1188 blk_queue_make_request(md->queue, dm_request);
1189 blk_queue_ordered(md->queue, QUEUE_ORDERED_DRAIN, NULL);
1190 blk_queue_bounce_limit(md->queue, BLK_BOUNCE_ANY);
1191 md->queue->unplug_fn = dm_unplug_all;
1192 blk_queue_merge_bvec(md->queue, dm_merge_bvec);
1194 md->io_pool = mempool_create_slab_pool(MIN_IOS, _io_cache);
1198 md->tio_pool = mempool_create_slab_pool(MIN_IOS, _tio_cache);
1202 md->bs = bioset_create(16, 0);
1206 md->disk = alloc_disk(1);
1210 atomic_set(&md->pending, 0);
1211 init_waitqueue_head(&md->wait);
1212 INIT_WORK(&md->work, dm_wq_work);
1213 init_waitqueue_head(&md->eventq);
1215 md->disk->major = _major;
1216 md->disk->first_minor = minor;
1217 md->disk->fops = &dm_blk_dops;
1218 md->disk->queue = md->queue;
1219 md->disk->private_data = md;
1220 sprintf(md->disk->disk_name, "dm-%d", minor);
1222 format_dev_t(md->name, MKDEV(_major, minor));
1224 md->wq = create_singlethread_workqueue("kdmflush");
1228 md->bdev = bdget_disk(md->disk, 0);
1232 /* Populate the mapping, nobody knows we exist yet */
1233 spin_lock(&_minor_lock);
1234 old_md = idr_replace(&_minor_idr, md, minor);
1235 spin_unlock(&_minor_lock);
1237 BUG_ON(old_md != MINOR_ALLOCED);
1242 destroy_workqueue(md->wq);
1246 bioset_free(md->bs);
1248 mempool_destroy(md->tio_pool);
1250 mempool_destroy(md->io_pool);
1252 blk_cleanup_queue(md->queue);
1256 module_put(THIS_MODULE);
1262 static void unlock_fs(struct mapped_device *md);
1264 static void free_dev(struct mapped_device *md)
1266 int minor = MINOR(disk_devt(md->disk));
1270 destroy_workqueue(md->wq);
1271 mempool_destroy(md->tio_pool);
1272 mempool_destroy(md->io_pool);
1273 bioset_free(md->bs);
1274 blk_integrity_unregister(md->disk);
1275 del_gendisk(md->disk);
1278 spin_lock(&_minor_lock);
1279 md->disk->private_data = NULL;
1280 spin_unlock(&_minor_lock);
1283 blk_cleanup_queue(md->queue);
1284 module_put(THIS_MODULE);
1289 * Bind a table to the device.
1291 static void event_callback(void *context)
1293 unsigned long flags;
1295 struct mapped_device *md = (struct mapped_device *) context;
1297 spin_lock_irqsave(&md->uevent_lock, flags);
1298 list_splice_init(&md->uevent_list, &uevents);
1299 spin_unlock_irqrestore(&md->uevent_lock, flags);
1301 dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj);
1303 atomic_inc(&md->event_nr);
1304 wake_up(&md->eventq);
1307 static void __set_size(struct mapped_device *md, sector_t size)
1309 set_capacity(md->disk, size);
1311 mutex_lock(&md->bdev->bd_inode->i_mutex);
1312 i_size_write(md->bdev->bd_inode, (loff_t)size << SECTOR_SHIFT);
1313 mutex_unlock(&md->bdev->bd_inode->i_mutex);
1316 static int __bind(struct mapped_device *md, struct dm_table *t,
1317 struct queue_limits *limits)
1319 struct request_queue *q = md->queue;
1322 size = dm_table_get_size(t);
1325 * Wipe any geometry if the size of the table changed.
1327 if (size != get_capacity(md->disk))
1328 memset(&md->geometry, 0, sizeof(md->geometry));
1330 __set_size(md, size);
1333 dm_table_destroy(t);
1337 dm_table_event_callback(t, event_callback, md);
1339 write_lock(&md->map_lock);
1341 dm_table_set_restrictions(t, q, limits);
1342 write_unlock(&md->map_lock);
1347 static void __unbind(struct mapped_device *md)
1349 struct dm_table *map = md->map;
1354 dm_table_event_callback(map, NULL, NULL);
1355 write_lock(&md->map_lock);
1357 write_unlock(&md->map_lock);
1358 dm_table_destroy(map);
1362 * Constructor for a new device.
1364 int dm_create(int minor, struct mapped_device **result)
1366 struct mapped_device *md;
1368 md = alloc_dev(minor);
1378 static struct mapped_device *dm_find_md(dev_t dev)
1380 struct mapped_device *md;
1381 unsigned minor = MINOR(dev);
1383 if (MAJOR(dev) != _major || minor >= (1 << MINORBITS))
1386 spin_lock(&_minor_lock);
1388 md = idr_find(&_minor_idr, minor);
1389 if (md && (md == MINOR_ALLOCED ||
1390 (MINOR(disk_devt(dm_disk(md))) != minor) ||
1391 test_bit(DMF_FREEING, &md->flags))) {
1397 spin_unlock(&_minor_lock);
1402 struct mapped_device *dm_get_md(dev_t dev)
1404 struct mapped_device *md = dm_find_md(dev);
1412 void *dm_get_mdptr(struct mapped_device *md)
1414 return md->interface_ptr;
1417 void dm_set_mdptr(struct mapped_device *md, void *ptr)
1419 md->interface_ptr = ptr;
1422 void dm_get(struct mapped_device *md)
1424 atomic_inc(&md->holders);
1427 const char *dm_device_name(struct mapped_device *md)
1431 EXPORT_SYMBOL_GPL(dm_device_name);
1433 void dm_put(struct mapped_device *md)
1435 struct dm_table *map;
1437 BUG_ON(test_bit(DMF_FREEING, &md->flags));
1439 if (atomic_dec_and_lock(&md->holders, &_minor_lock)) {
1440 map = dm_get_table(md);
1441 idr_replace(&_minor_idr, MINOR_ALLOCED,
1442 MINOR(disk_devt(dm_disk(md))));
1443 set_bit(DMF_FREEING, &md->flags);
1444 spin_unlock(&_minor_lock);
1445 if (!dm_suspended(md)) {
1446 dm_table_presuspend_targets(map);
1447 dm_table_postsuspend_targets(map);
1455 EXPORT_SYMBOL_GPL(dm_put);
1457 static int dm_wait_for_completion(struct mapped_device *md, int interruptible)
1460 DECLARE_WAITQUEUE(wait, current);
1462 dm_unplug_all(md->queue);
1464 add_wait_queue(&md->wait, &wait);
1467 set_current_state(interruptible);
1470 if (!atomic_read(&md->pending))
1473 if (interruptible == TASK_INTERRUPTIBLE &&
1474 signal_pending(current)) {
1481 set_current_state(TASK_RUNNING);
1483 remove_wait_queue(&md->wait, &wait);
1488 static void dm_flush(struct mapped_device *md)
1490 dm_wait_for_completion(md, TASK_UNINTERRUPTIBLE);
1492 bio_init(&md->barrier_bio);
1493 md->barrier_bio.bi_bdev = md->bdev;
1494 md->barrier_bio.bi_rw = WRITE_BARRIER;
1495 __split_and_process_bio(md, &md->barrier_bio);
1497 dm_wait_for_completion(md, TASK_UNINTERRUPTIBLE);
1500 static void process_barrier(struct mapped_device *md, struct bio *bio)
1502 md->barrier_error = 0;
1506 if (!bio_empty_barrier(bio)) {
1507 __split_and_process_bio(md, bio);
1511 if (md->barrier_error != DM_ENDIO_REQUEUE)
1512 bio_endio(bio, md->barrier_error);
1514 spin_lock_irq(&md->deferred_lock);
1515 bio_list_add_head(&md->deferred, bio);
1516 spin_unlock_irq(&md->deferred_lock);
1521 * Process the deferred bios
1523 static void dm_wq_work(struct work_struct *work)
1525 struct mapped_device *md = container_of(work, struct mapped_device,
1529 down_write(&md->io_lock);
1531 while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
1532 spin_lock_irq(&md->deferred_lock);
1533 c = bio_list_pop(&md->deferred);
1534 spin_unlock_irq(&md->deferred_lock);
1537 clear_bit(DMF_QUEUE_IO_TO_THREAD, &md->flags);
1541 up_write(&md->io_lock);
1544 process_barrier(md, c);
1546 __split_and_process_bio(md, c);
1548 down_write(&md->io_lock);
1551 up_write(&md->io_lock);
1554 static void dm_queue_flush(struct mapped_device *md)
1556 clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
1557 smp_mb__after_clear_bit();
1558 queue_work(md->wq, &md->work);
1562 * Swap in a new table (destroying old one).
1564 int dm_swap_table(struct mapped_device *md, struct dm_table *table)
1566 struct queue_limits limits;
1569 mutex_lock(&md->suspend_lock);
1571 /* device must be suspended */
1572 if (!dm_suspended(md))
1575 r = dm_calculate_queue_limits(table, &limits);
1580 r = __bind(md, table, &limits);
1583 mutex_unlock(&md->suspend_lock);
1588 * Functions to lock and unlock any filesystem running on the
1591 static int lock_fs(struct mapped_device *md)
1595 WARN_ON(md->frozen_sb);
1597 md->frozen_sb = freeze_bdev(md->bdev);
1598 if (IS_ERR(md->frozen_sb)) {
1599 r = PTR_ERR(md->frozen_sb);
1600 md->frozen_sb = NULL;
1604 set_bit(DMF_FROZEN, &md->flags);
1609 static void unlock_fs(struct mapped_device *md)
1611 if (!test_bit(DMF_FROZEN, &md->flags))
1614 thaw_bdev(md->bdev, md->frozen_sb);
1615 md->frozen_sb = NULL;
1616 clear_bit(DMF_FROZEN, &md->flags);
1620 * We need to be able to change a mapping table under a mounted
1621 * filesystem. For example we might want to move some data in
1622 * the background. Before the table can be swapped with
1623 * dm_bind_table, dm_suspend must be called to flush any in
1624 * flight bios and ensure that any further io gets deferred.
1626 int dm_suspend(struct mapped_device *md, unsigned suspend_flags)
1628 struct dm_table *map = NULL;
1630 int do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG ? 1 : 0;
1631 int noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG ? 1 : 0;
1633 mutex_lock(&md->suspend_lock);
1635 if (dm_suspended(md)) {
1640 map = dm_get_table(md);
1643 * DMF_NOFLUSH_SUSPENDING must be set before presuspend.
1644 * This flag is cleared before dm_suspend returns.
1647 set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
1649 /* This does not get reverted if there's an error later. */
1650 dm_table_presuspend_targets(map);
1653 * Flush I/O to the device. noflush supersedes do_lockfs,
1654 * because lock_fs() needs to flush I/Os.
1656 if (!noflush && do_lockfs) {
1663 * Here we must make sure that no processes are submitting requests
1664 * to target drivers i.e. no one may be executing
1665 * __split_and_process_bio. This is called from dm_request and
1668 * To get all processes out of __split_and_process_bio in dm_request,
1669 * we take the write lock. To prevent any process from reentering
1670 * __split_and_process_bio from dm_request, we set
1671 * DMF_QUEUE_IO_TO_THREAD.
1673 * To quiesce the thread (dm_wq_work), we set DMF_BLOCK_IO_FOR_SUSPEND
1674 * and call flush_workqueue(md->wq). flush_workqueue will wait until
1675 * dm_wq_work exits and DMF_BLOCK_IO_FOR_SUSPEND will prevent any
1676 * further calls to __split_and_process_bio from dm_wq_work.
1678 down_write(&md->io_lock);
1679 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
1680 set_bit(DMF_QUEUE_IO_TO_THREAD, &md->flags);
1681 up_write(&md->io_lock);
1683 flush_workqueue(md->wq);
1686 * At this point no more requests are entering target request routines.
1687 * We call dm_wait_for_completion to wait for all existing requests
1690 r = dm_wait_for_completion(md, TASK_INTERRUPTIBLE);
1692 down_write(&md->io_lock);
1694 clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
1695 up_write(&md->io_lock);
1697 /* were we interrupted ? */
1702 goto out; /* pushback list is already flushed, so skip flush */
1706 * If dm_wait_for_completion returned 0, the device is completely
1707 * quiescent now. There is no request-processing activity. All new
1708 * requests are being added to md->deferred list.
1711 dm_table_postsuspend_targets(map);
1713 set_bit(DMF_SUSPENDED, &md->flags);
1719 mutex_unlock(&md->suspend_lock);
1723 int dm_resume(struct mapped_device *md)
1726 struct dm_table *map = NULL;
1728 mutex_lock(&md->suspend_lock);
1729 if (!dm_suspended(md))
1732 map = dm_get_table(md);
1733 if (!map || !dm_table_get_size(map))
1736 r = dm_table_resume_targets(map);
1744 clear_bit(DMF_SUSPENDED, &md->flags);
1746 dm_table_unplug_all(map);
1750 mutex_unlock(&md->suspend_lock);
1755 /*-----------------------------------------------------------------
1756 * Event notification.
1757 *---------------------------------------------------------------*/
1758 void dm_kobject_uevent(struct mapped_device *md, enum kobject_action action,
1761 char udev_cookie[DM_COOKIE_LENGTH];
1762 char *envp[] = { udev_cookie, NULL };
1765 kobject_uevent(&disk_to_dev(md->disk)->kobj, action);
1767 snprintf(udev_cookie, DM_COOKIE_LENGTH, "%s=%u",
1768 DM_COOKIE_ENV_VAR_NAME, cookie);
1769 kobject_uevent_env(&disk_to_dev(md->disk)->kobj, action, envp);
1773 uint32_t dm_next_uevent_seq(struct mapped_device *md)
1775 return atomic_add_return(1, &md->uevent_seq);
1778 uint32_t dm_get_event_nr(struct mapped_device *md)
1780 return atomic_read(&md->event_nr);
1783 int dm_wait_event(struct mapped_device *md, int event_nr)
1785 return wait_event_interruptible(md->eventq,
1786 (event_nr != atomic_read(&md->event_nr)));
1789 void dm_uevent_add(struct mapped_device *md, struct list_head *elist)
1791 unsigned long flags;
1793 spin_lock_irqsave(&md->uevent_lock, flags);
1794 list_add(elist, &md->uevent_list);
1795 spin_unlock_irqrestore(&md->uevent_lock, flags);
1799 * The gendisk is only valid as long as you have a reference
1802 struct gendisk *dm_disk(struct mapped_device *md)
1807 struct kobject *dm_kobject(struct mapped_device *md)
1813 * struct mapped_device should not be exported outside of dm.c
1814 * so use this check to verify that kobj is part of md structure
1816 struct mapped_device *dm_get_from_kobject(struct kobject *kobj)
1818 struct mapped_device *md;
1820 md = container_of(kobj, struct mapped_device, kobj);
1821 if (&md->kobj != kobj)
1824 if (test_bit(DMF_FREEING, &md->flags) ||
1825 test_bit(DMF_DELETING, &md->flags))
1832 int dm_suspended(struct mapped_device *md)
1834 return test_bit(DMF_SUSPENDED, &md->flags);
1837 int dm_noflush_suspending(struct dm_target *ti)
1839 struct mapped_device *md = dm_table_get_md(ti->table);
1840 int r = __noflush_suspending(md);
1846 EXPORT_SYMBOL_GPL(dm_noflush_suspending);
1848 static struct block_device_operations dm_blk_dops = {
1849 .open = dm_blk_open,
1850 .release = dm_blk_close,
1851 .ioctl = dm_blk_ioctl,
1852 .getgeo = dm_blk_getgeo,
1853 .owner = THIS_MODULE
1856 EXPORT_SYMBOL(dm_get_mapinfo);
1861 module_init(dm_init);
1862 module_exit(dm_exit);
1864 module_param(major, uint, 0);
1865 MODULE_PARM_DESC(major, "The major number of the device mapper");
1866 MODULE_DESCRIPTION(DM_NAME " driver");
1867 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
1868 MODULE_LICENSE("GPL");