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/delay.h>
24 #include <trace/events/block.h>
26 #define DM_MSG_PREFIX "core"
30 * ratelimit state to be used in DMXXX_LIMIT().
32 DEFINE_RATELIMIT_STATE(dm_ratelimit_state,
33 DEFAULT_RATELIMIT_INTERVAL,
34 DEFAULT_RATELIMIT_BURST);
35 EXPORT_SYMBOL(dm_ratelimit_state);
39 * Cookies are numeric values sent with CHANGE and REMOVE
40 * uevents while resuming, removing or renaming the device.
42 #define DM_COOKIE_ENV_VAR_NAME "DM_COOKIE"
43 #define DM_COOKIE_LENGTH 24
45 static const char *_name = DM_NAME;
47 static unsigned int major = 0;
48 static unsigned int _major = 0;
50 static DEFINE_IDR(_minor_idr);
52 static DEFINE_SPINLOCK(_minor_lock);
55 * One of these is allocated per bio.
58 struct mapped_device *md;
62 unsigned long start_time;
63 spinlock_t endio_lock;
68 * One of these is allocated per target within a bio. Hopefully
69 * this will be simplified out one day.
78 * For request-based dm.
79 * One of these is allocated per request.
81 struct dm_rq_target_io {
82 struct mapped_device *md;
84 struct request *orig, clone;
90 * For request-based dm.
91 * One of these is allocated per bio.
93 struct dm_rq_clone_bio_info {
95 struct dm_rq_target_io *tio;
98 union map_info *dm_get_mapinfo(struct bio *bio)
100 if (bio && bio->bi_private)
101 return &((struct dm_target_io *)bio->bi_private)->info;
105 union map_info *dm_get_rq_mapinfo(struct request *rq)
107 if (rq && rq->end_io_data)
108 return &((struct dm_rq_target_io *)rq->end_io_data)->info;
111 EXPORT_SYMBOL_GPL(dm_get_rq_mapinfo);
113 #define MINOR_ALLOCED ((void *)-1)
116 * Bits for the md->flags field.
118 #define DMF_BLOCK_IO_FOR_SUSPEND 0
119 #define DMF_SUSPENDED 1
121 #define DMF_FREEING 3
122 #define DMF_DELETING 4
123 #define DMF_NOFLUSH_SUSPENDING 5
124 #define DMF_MERGE_IS_OPTIONAL 6
127 * Work processed by per-device workqueue.
129 struct mapped_device {
130 struct rw_semaphore io_lock;
131 struct mutex suspend_lock;
138 struct request_queue *queue;
140 /* Protect queue and type against concurrent access. */
141 struct mutex type_lock;
143 struct target_type *immutable_target_type;
145 struct gendisk *disk;
151 * A list of ios that arrived while we were suspended.
154 wait_queue_head_t wait;
155 struct work_struct work;
156 struct bio_list deferred;
157 spinlock_t deferred_lock;
160 * Processing queue (flush)
162 struct workqueue_struct *wq;
165 * The current mapping.
167 struct dm_table *map;
170 * io objects are allocated from here.
181 wait_queue_head_t eventq;
183 struct list_head uevent_list;
184 spinlock_t uevent_lock; /* Protect access to uevent_list */
187 * freeze/thaw support require holding onto a super block
189 struct super_block *frozen_sb;
190 struct block_device *bdev;
192 /* forced geometry settings */
193 struct hd_geometry geometry;
198 /* zero-length flush that will be cloned and submitted to targets */
199 struct bio flush_bio;
203 * For mempools pre-allocation at the table loading time.
205 struct dm_md_mempools {
212 static struct kmem_cache *_io_cache;
213 static struct kmem_cache *_tio_cache;
214 static struct kmem_cache *_rq_tio_cache;
215 static struct kmem_cache *_rq_bio_info_cache;
217 static int __init local_init(void)
221 /* allocate a slab for the dm_ios */
222 _io_cache = KMEM_CACHE(dm_io, 0);
226 /* allocate a slab for the target ios */
227 _tio_cache = KMEM_CACHE(dm_target_io, 0);
229 goto out_free_io_cache;
231 _rq_tio_cache = KMEM_CACHE(dm_rq_target_io, 0);
233 goto out_free_tio_cache;
235 _rq_bio_info_cache = KMEM_CACHE(dm_rq_clone_bio_info, 0);
236 if (!_rq_bio_info_cache)
237 goto out_free_rq_tio_cache;
239 r = dm_uevent_init();
241 goto out_free_rq_bio_info_cache;
244 r = register_blkdev(_major, _name);
246 goto out_uevent_exit;
255 out_free_rq_bio_info_cache:
256 kmem_cache_destroy(_rq_bio_info_cache);
257 out_free_rq_tio_cache:
258 kmem_cache_destroy(_rq_tio_cache);
260 kmem_cache_destroy(_tio_cache);
262 kmem_cache_destroy(_io_cache);
267 static void local_exit(void)
269 kmem_cache_destroy(_rq_bio_info_cache);
270 kmem_cache_destroy(_rq_tio_cache);
271 kmem_cache_destroy(_tio_cache);
272 kmem_cache_destroy(_io_cache);
273 unregister_blkdev(_major, _name);
278 DMINFO("cleaned up");
281 static int (*_inits[])(void) __initdata = {
291 static void (*_exits[])(void) = {
301 static int __init dm_init(void)
303 const int count = ARRAY_SIZE(_inits);
307 for (i = 0; i < count; i++) {
322 static void __exit dm_exit(void)
324 int i = ARRAY_SIZE(_exits);
330 * Should be empty by this point.
332 idr_remove_all(&_minor_idr);
333 idr_destroy(&_minor_idr);
337 * Block device functions
339 int dm_deleting_md(struct mapped_device *md)
341 return test_bit(DMF_DELETING, &md->flags);
344 static int dm_blk_open(struct block_device *bdev, fmode_t mode)
346 struct mapped_device *md;
348 spin_lock(&_minor_lock);
350 md = bdev->bd_disk->private_data;
354 if (test_bit(DMF_FREEING, &md->flags) ||
355 dm_deleting_md(md)) {
361 atomic_inc(&md->open_count);
364 spin_unlock(&_minor_lock);
366 return md ? 0 : -ENXIO;
369 static int dm_blk_close(struct gendisk *disk, fmode_t mode)
371 struct mapped_device *md = disk->private_data;
373 spin_lock(&_minor_lock);
375 atomic_dec(&md->open_count);
378 spin_unlock(&_minor_lock);
383 int dm_open_count(struct mapped_device *md)
385 return atomic_read(&md->open_count);
389 * Guarantees nothing is using the device before it's deleted.
391 int dm_lock_for_deletion(struct mapped_device *md)
395 spin_lock(&_minor_lock);
397 if (dm_open_count(md))
400 set_bit(DMF_DELETING, &md->flags);
402 spin_unlock(&_minor_lock);
407 static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo)
409 struct mapped_device *md = bdev->bd_disk->private_data;
411 return dm_get_geometry(md, geo);
414 static int dm_blk_ioctl(struct block_device *bdev, fmode_t mode,
415 unsigned int cmd, unsigned long arg)
417 struct mapped_device *md = bdev->bd_disk->private_data;
418 struct dm_table *map = dm_get_live_table(md);
419 struct dm_target *tgt;
422 if (!map || !dm_table_get_size(map))
425 /* We only support devices that have a single target */
426 if (dm_table_get_num_targets(map) != 1)
429 tgt = dm_table_get_target(map, 0);
431 if (dm_suspended_md(md)) {
436 if (tgt->type->ioctl)
437 r = tgt->type->ioctl(tgt, cmd, arg);
445 static struct dm_io *alloc_io(struct mapped_device *md)
447 return mempool_alloc(md->io_pool, GFP_NOIO);
450 static void free_io(struct mapped_device *md, struct dm_io *io)
452 mempool_free(io, md->io_pool);
455 static void free_tio(struct mapped_device *md, struct dm_target_io *tio)
457 mempool_free(tio, md->tio_pool);
460 static struct dm_rq_target_io *alloc_rq_tio(struct mapped_device *md,
463 return mempool_alloc(md->tio_pool, gfp_mask);
466 static void free_rq_tio(struct dm_rq_target_io *tio)
468 mempool_free(tio, tio->md->tio_pool);
471 static struct dm_rq_clone_bio_info *alloc_bio_info(struct mapped_device *md)
473 return mempool_alloc(md->io_pool, GFP_ATOMIC);
476 static void free_bio_info(struct dm_rq_clone_bio_info *info)
478 mempool_free(info, info->tio->md->io_pool);
481 static int md_in_flight(struct mapped_device *md)
483 return atomic_read(&md->pending[READ]) +
484 atomic_read(&md->pending[WRITE]);
487 static void start_io_acct(struct dm_io *io)
489 struct mapped_device *md = io->md;
491 int rw = bio_data_dir(io->bio);
493 io->start_time = jiffies;
495 cpu = part_stat_lock();
496 part_round_stats(cpu, &dm_disk(md)->part0);
498 atomic_set(&dm_disk(md)->part0.in_flight[rw],
499 atomic_inc_return(&md->pending[rw]));
502 static void end_io_acct(struct dm_io *io)
504 struct mapped_device *md = io->md;
505 struct bio *bio = io->bio;
506 unsigned long duration = jiffies - io->start_time;
508 int rw = bio_data_dir(bio);
510 cpu = part_stat_lock();
511 part_round_stats(cpu, &dm_disk(md)->part0);
512 part_stat_add(cpu, &dm_disk(md)->part0, ticks[rw], duration);
516 * After this is decremented the bio must not be touched if it is
519 pending = atomic_dec_return(&md->pending[rw]);
520 atomic_set(&dm_disk(md)->part0.in_flight[rw], pending);
521 pending += atomic_read(&md->pending[rw^0x1]);
523 /* nudge anyone waiting on suspend queue */
529 * Add the bio to the list of deferred io.
531 static void queue_io(struct mapped_device *md, struct bio *bio)
535 spin_lock_irqsave(&md->deferred_lock, flags);
536 bio_list_add(&md->deferred, bio);
537 spin_unlock_irqrestore(&md->deferred_lock, flags);
538 queue_work(md->wq, &md->work);
542 * Everyone (including functions in this file), should use this
543 * function to access the md->map field, and make sure they call
544 * dm_table_put() when finished.
546 struct dm_table *dm_get_live_table(struct mapped_device *md)
551 read_lock_irqsave(&md->map_lock, flags);
555 read_unlock_irqrestore(&md->map_lock, flags);
561 * Get the geometry associated with a dm device
563 int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo)
571 * Set the geometry of a device.
573 int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo)
575 sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors;
577 if (geo->start > sz) {
578 DMWARN("Start sector is beyond the geometry limits.");
587 /*-----------------------------------------------------------------
589 * A more elegant soln is in the works that uses the queue
590 * merge fn, unfortunately there are a couple of changes to
591 * the block layer that I want to make for this. So in the
592 * interests of getting something for people to use I give
593 * you this clearly demarcated crap.
594 *---------------------------------------------------------------*/
596 static int __noflush_suspending(struct mapped_device *md)
598 return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
602 * Decrements the number of outstanding ios that a bio has been
603 * cloned into, completing the original io if necc.
605 static void dec_pending(struct dm_io *io, int error)
610 struct mapped_device *md = io->md;
612 /* Push-back supersedes any I/O errors */
613 if (unlikely(error)) {
614 spin_lock_irqsave(&io->endio_lock, flags);
615 if (!(io->error > 0 && __noflush_suspending(md)))
617 spin_unlock_irqrestore(&io->endio_lock, flags);
620 if (atomic_dec_and_test(&io->io_count)) {
621 if (io->error == DM_ENDIO_REQUEUE) {
623 * Target requested pushing back the I/O.
625 spin_lock_irqsave(&md->deferred_lock, flags);
626 if (__noflush_suspending(md))
627 bio_list_add_head(&md->deferred, io->bio);
629 /* noflush suspend was interrupted. */
631 spin_unlock_irqrestore(&md->deferred_lock, flags);
634 io_error = io->error;
639 if (io_error == DM_ENDIO_REQUEUE)
642 if ((bio->bi_rw & REQ_FLUSH) && bio->bi_size) {
644 * Preflush done for flush with data, reissue
647 bio->bi_rw &= ~REQ_FLUSH;
650 /* done with normal IO or empty flush */
651 trace_block_bio_complete(md->queue, bio, io_error);
652 bio_endio(bio, io_error);
657 static void clone_endio(struct bio *bio, int error)
660 struct dm_target_io *tio = bio->bi_private;
661 struct dm_io *io = tio->io;
662 struct mapped_device *md = tio->io->md;
663 dm_endio_fn endio = tio->ti->type->end_io;
665 if (!bio_flagged(bio, BIO_UPTODATE) && !error)
669 r = endio(tio->ti, bio, error, &tio->info);
670 if (r < 0 || r == DM_ENDIO_REQUEUE)
672 * error and requeue request are handled
676 else if (r == DM_ENDIO_INCOMPLETE)
677 /* The target will handle the io */
680 DMWARN("unimplemented target endio return value: %d", r);
686 * Store md for cleanup instead of tio which is about to get freed.
688 bio->bi_private = md->bs;
692 dec_pending(io, error);
696 * Partial completion handling for request-based dm
698 static void end_clone_bio(struct bio *clone, int error)
700 struct dm_rq_clone_bio_info *info = clone->bi_private;
701 struct dm_rq_target_io *tio = info->tio;
702 struct bio *bio = info->orig;
703 unsigned int nr_bytes = info->orig->bi_size;
709 * An error has already been detected on the request.
710 * Once error occurred, just let clone->end_io() handle
716 * Don't notice the error to the upper layer yet.
717 * The error handling decision is made by the target driver,
718 * when the request is completed.
725 * I/O for the bio successfully completed.
726 * Notice the data completion to the upper layer.
730 * bios are processed from the head of the list.
731 * So the completing bio should always be rq->bio.
732 * If it's not, something wrong is happening.
734 if (tio->orig->bio != bio)
735 DMERR("bio completion is going in the middle of the request");
738 * Update the original request.
739 * Do not use blk_end_request() here, because it may complete
740 * the original request before the clone, and break the ordering.
742 blk_update_request(tio->orig, 0, nr_bytes);
746 * Don't touch any member of the md after calling this function because
747 * the md may be freed in dm_put() at the end of this function.
748 * Or do dm_get() before calling this function and dm_put() later.
750 static void rq_completed(struct mapped_device *md, int rw, int run_queue)
752 atomic_dec(&md->pending[rw]);
754 /* nudge anyone waiting on suspend queue */
755 if (!md_in_flight(md))
759 * Run this off this callpath, as drivers could invoke end_io while
760 * inside their request_fn (and holding the queue lock). Calling
761 * back into ->request_fn() could deadlock attempting to grab the
765 blk_run_queue_async(md->queue);
768 * dm_put() must be at the end of this function. See the comment above
773 static void free_rq_clone(struct request *clone)
775 struct dm_rq_target_io *tio = clone->end_io_data;
777 blk_rq_unprep_clone(clone);
782 * Complete the clone and the original request.
783 * Must be called without queue lock.
785 static void dm_end_request(struct request *clone, int error)
787 int rw = rq_data_dir(clone);
788 struct dm_rq_target_io *tio = clone->end_io_data;
789 struct mapped_device *md = tio->md;
790 struct request *rq = tio->orig;
792 if (rq->cmd_type == REQ_TYPE_BLOCK_PC) {
793 rq->errors = clone->errors;
794 rq->resid_len = clone->resid_len;
798 * We are using the sense buffer of the original
800 * So setting the length of the sense data is enough.
802 rq->sense_len = clone->sense_len;
805 free_rq_clone(clone);
806 blk_end_request_all(rq, error);
807 rq_completed(md, rw, true);
810 static void dm_unprep_request(struct request *rq)
812 struct request *clone = rq->special;
815 rq->cmd_flags &= ~REQ_DONTPREP;
817 free_rq_clone(clone);
821 * Requeue the original request of a clone.
823 void dm_requeue_unmapped_request(struct request *clone)
825 int rw = rq_data_dir(clone);
826 struct dm_rq_target_io *tio = clone->end_io_data;
827 struct mapped_device *md = tio->md;
828 struct request *rq = tio->orig;
829 struct request_queue *q = rq->q;
832 dm_unprep_request(rq);
834 spin_lock_irqsave(q->queue_lock, flags);
835 blk_requeue_request(q, rq);
836 spin_unlock_irqrestore(q->queue_lock, flags);
838 rq_completed(md, rw, 0);
840 EXPORT_SYMBOL_GPL(dm_requeue_unmapped_request);
842 static void __stop_queue(struct request_queue *q)
847 static void stop_queue(struct request_queue *q)
851 spin_lock_irqsave(q->queue_lock, flags);
853 spin_unlock_irqrestore(q->queue_lock, flags);
856 static void __start_queue(struct request_queue *q)
858 if (blk_queue_stopped(q))
862 static void start_queue(struct request_queue *q)
866 spin_lock_irqsave(q->queue_lock, flags);
868 spin_unlock_irqrestore(q->queue_lock, flags);
871 static void dm_done(struct request *clone, int error, bool mapped)
874 struct dm_rq_target_io *tio = clone->end_io_data;
875 dm_request_endio_fn rq_end_io = NULL;
878 rq_end_io = tio->ti->type->rq_end_io;
880 if (mapped && rq_end_io)
881 r = rq_end_io(tio->ti, clone, error, &tio->info);
885 /* The target wants to complete the I/O */
886 dm_end_request(clone, r);
887 else if (r == DM_ENDIO_INCOMPLETE)
888 /* The target will handle the I/O */
890 else if (r == DM_ENDIO_REQUEUE)
891 /* The target wants to requeue the I/O */
892 dm_requeue_unmapped_request(clone);
894 DMWARN("unimplemented target endio return value: %d", r);
900 * Request completion handler for request-based dm
902 static void dm_softirq_done(struct request *rq)
905 struct request *clone = rq->completion_data;
906 struct dm_rq_target_io *tio = clone->end_io_data;
908 if (rq->cmd_flags & REQ_FAILED)
911 dm_done(clone, tio->error, mapped);
915 * Complete the clone and the original request with the error status
916 * through softirq context.
918 static void dm_complete_request(struct request *clone, int error)
920 struct dm_rq_target_io *tio = clone->end_io_data;
921 struct request *rq = tio->orig;
924 rq->completion_data = clone;
925 blk_complete_request(rq);
929 * Complete the not-mapped clone and the original request with the error status
930 * through softirq context.
931 * Target's rq_end_io() function isn't called.
932 * This may be used when the target's map_rq() function fails.
934 void dm_kill_unmapped_request(struct request *clone, int error)
936 struct dm_rq_target_io *tio = clone->end_io_data;
937 struct request *rq = tio->orig;
939 rq->cmd_flags |= REQ_FAILED;
940 dm_complete_request(clone, error);
942 EXPORT_SYMBOL_GPL(dm_kill_unmapped_request);
945 * Called with the queue lock held
947 static void end_clone_request(struct request *clone, int error)
950 * For just cleaning up the information of the queue in which
951 * the clone was dispatched.
952 * The clone is *NOT* freed actually here because it is alloced from
953 * dm own mempool and REQ_ALLOCED isn't set in clone->cmd_flags.
955 __blk_put_request(clone->q, clone);
958 * Actual request completion is done in a softirq context which doesn't
959 * hold the queue lock. Otherwise, deadlock could occur because:
960 * - another request may be submitted by the upper level driver
961 * of the stacking during the completion
962 * - the submission which requires queue lock may be done
965 dm_complete_request(clone, error);
969 * Return maximum size of I/O possible at the supplied sector up to the current
972 static sector_t max_io_len_target_boundary(sector_t sector, struct dm_target *ti)
974 sector_t target_offset = dm_target_offset(ti, sector);
976 return ti->len - target_offset;
979 static sector_t max_io_len(sector_t sector, struct dm_target *ti)
981 sector_t len = max_io_len_target_boundary(sector, ti);
984 * Does the target need to split even further ?
988 sector_t offset = dm_target_offset(ti, sector);
989 boundary = ((offset + ti->split_io) & ~(ti->split_io - 1))
998 static void __map_bio(struct dm_target *ti, struct bio *clone,
999 struct dm_target_io *tio)
1003 struct mapped_device *md;
1005 clone->bi_end_io = clone_endio;
1006 clone->bi_private = tio;
1009 * Map the clone. If r == 0 we don't need to do
1010 * anything, the target has assumed ownership of
1013 atomic_inc(&tio->io->io_count);
1014 sector = clone->bi_sector;
1015 r = ti->type->map(ti, clone, &tio->info);
1016 if (r == DM_MAPIO_REMAPPED) {
1017 /* the bio has been remapped so dispatch it */
1019 trace_block_bio_remap(bdev_get_queue(clone->bi_bdev), clone,
1020 tio->io->bio->bi_bdev->bd_dev, sector);
1022 generic_make_request(clone);
1023 } else if (r < 0 || r == DM_MAPIO_REQUEUE) {
1024 /* error the io and bail out, or requeue it if needed */
1026 dec_pending(tio->io, r);
1028 * Store bio_set for cleanup.
1030 clone->bi_private = md->bs;
1034 DMWARN("unimplemented target map return value: %d", r);
1040 struct mapped_device *md;
1041 struct dm_table *map;
1045 sector_t sector_count;
1049 static void dm_bio_destructor(struct bio *bio)
1051 struct bio_set *bs = bio->bi_private;
1057 * Creates a little bio that just does part of a bvec.
1059 static struct bio *split_bvec(struct bio *bio, sector_t sector,
1060 unsigned short idx, unsigned int offset,
1061 unsigned int len, struct bio_set *bs)
1064 struct bio_vec *bv = bio->bi_io_vec + idx;
1066 clone = bio_alloc_bioset(GFP_NOIO, 1, bs);
1067 clone->bi_destructor = dm_bio_destructor;
1068 *clone->bi_io_vec = *bv;
1070 clone->bi_sector = sector;
1071 clone->bi_bdev = bio->bi_bdev;
1072 clone->bi_rw = bio->bi_rw;
1074 clone->bi_size = to_bytes(len);
1075 clone->bi_io_vec->bv_offset = offset;
1076 clone->bi_io_vec->bv_len = clone->bi_size;
1077 clone->bi_flags |= 1 << BIO_CLONED;
1079 if (bio_integrity(bio)) {
1080 bio_integrity_clone(clone, bio, GFP_NOIO, bs);
1081 bio_integrity_trim(clone,
1082 bio_sector_offset(bio, idx, offset), len);
1089 * Creates a bio that consists of range of complete bvecs.
1091 static struct bio *clone_bio(struct bio *bio, sector_t sector,
1092 unsigned short idx, unsigned short bv_count,
1093 unsigned int len, struct bio_set *bs)
1097 clone = bio_alloc_bioset(GFP_NOIO, bio->bi_max_vecs, bs);
1098 __bio_clone(clone, bio);
1099 clone->bi_destructor = dm_bio_destructor;
1100 clone->bi_sector = sector;
1101 clone->bi_idx = idx;
1102 clone->bi_vcnt = idx + bv_count;
1103 clone->bi_size = to_bytes(len);
1104 clone->bi_flags &= ~(1 << BIO_SEG_VALID);
1106 if (bio_integrity(bio)) {
1107 bio_integrity_clone(clone, bio, GFP_NOIO, bs);
1109 if (idx != bio->bi_idx || clone->bi_size < bio->bi_size)
1110 bio_integrity_trim(clone,
1111 bio_sector_offset(bio, idx, 0), len);
1117 static struct dm_target_io *alloc_tio(struct clone_info *ci,
1118 struct dm_target *ti)
1120 struct dm_target_io *tio = mempool_alloc(ci->md->tio_pool, GFP_NOIO);
1124 memset(&tio->info, 0, sizeof(tio->info));
1129 static void __issue_target_request(struct clone_info *ci, struct dm_target *ti,
1130 unsigned request_nr, sector_t len)
1132 struct dm_target_io *tio = alloc_tio(ci, ti);
1135 tio->info.target_request_nr = request_nr;
1138 * Discard requests require the bio's inline iovecs be initialized.
1139 * ci->bio->bi_max_vecs is BIO_INLINE_VECS anyway, for both flush
1140 * and discard, so no need for concern about wasted bvec allocations.
1142 clone = bio_alloc_bioset(GFP_NOIO, ci->bio->bi_max_vecs, ci->md->bs);
1143 __bio_clone(clone, ci->bio);
1144 clone->bi_destructor = dm_bio_destructor;
1146 clone->bi_sector = ci->sector;
1147 clone->bi_size = to_bytes(len);
1150 __map_bio(ti, clone, tio);
1153 static void __issue_target_requests(struct clone_info *ci, struct dm_target *ti,
1154 unsigned num_requests, sector_t len)
1156 unsigned request_nr;
1158 for (request_nr = 0; request_nr < num_requests; request_nr++)
1159 __issue_target_request(ci, ti, request_nr, len);
1162 static int __clone_and_map_empty_flush(struct clone_info *ci)
1164 unsigned target_nr = 0;
1165 struct dm_target *ti;
1167 BUG_ON(bio_has_data(ci->bio));
1168 while ((ti = dm_table_get_target(ci->map, target_nr++)))
1169 __issue_target_requests(ci, ti, ti->num_flush_requests, 0);
1175 * Perform all io with a single clone.
1177 static void __clone_and_map_simple(struct clone_info *ci, struct dm_target *ti)
1179 struct bio *clone, *bio = ci->bio;
1180 struct dm_target_io *tio;
1182 tio = alloc_tio(ci, ti);
1183 clone = clone_bio(bio, ci->sector, ci->idx,
1184 bio->bi_vcnt - ci->idx, ci->sector_count,
1186 __map_bio(ti, clone, tio);
1187 ci->sector_count = 0;
1190 static int __clone_and_map_discard(struct clone_info *ci)
1192 struct dm_target *ti;
1196 ti = dm_table_find_target(ci->map, ci->sector);
1197 if (!dm_target_is_valid(ti))
1201 * Even though the device advertised discard support,
1202 * that does not mean every target supports it, and
1203 * reconfiguration might also have changed that since the
1204 * check was performed.
1206 if (!ti->num_discard_requests)
1209 len = min(ci->sector_count, max_io_len_target_boundary(ci->sector, ti));
1211 __issue_target_requests(ci, ti, ti->num_discard_requests, len);
1214 } while (ci->sector_count -= len);
1219 static int __clone_and_map(struct clone_info *ci)
1221 struct bio *clone, *bio = ci->bio;
1222 struct dm_target *ti;
1223 sector_t len = 0, max;
1224 struct dm_target_io *tio;
1226 if (unlikely(bio->bi_rw & REQ_DISCARD))
1227 return __clone_and_map_discard(ci);
1229 ti = dm_table_find_target(ci->map, ci->sector);
1230 if (!dm_target_is_valid(ti))
1233 max = max_io_len(ci->sector, ti);
1235 if (ci->sector_count <= max) {
1237 * Optimise for the simple case where we can do all of
1238 * the remaining io with a single clone.
1240 __clone_and_map_simple(ci, ti);
1242 } else if (to_sector(bio->bi_io_vec[ci->idx].bv_len) <= max) {
1244 * There are some bvecs that don't span targets.
1245 * Do as many of these as possible.
1248 sector_t remaining = max;
1251 for (i = ci->idx; remaining && (i < bio->bi_vcnt); i++) {
1252 bv_len = to_sector(bio->bi_io_vec[i].bv_len);
1254 if (bv_len > remaining)
1257 remaining -= bv_len;
1261 tio = alloc_tio(ci, ti);
1262 clone = clone_bio(bio, ci->sector, ci->idx, i - ci->idx, len,
1264 __map_bio(ti, clone, tio);
1267 ci->sector_count -= len;
1272 * Handle a bvec that must be split between two or more targets.
1274 struct bio_vec *bv = bio->bi_io_vec + ci->idx;
1275 sector_t remaining = to_sector(bv->bv_len);
1276 unsigned int offset = 0;
1280 ti = dm_table_find_target(ci->map, ci->sector);
1281 if (!dm_target_is_valid(ti))
1284 max = max_io_len(ci->sector, ti);
1287 len = min(remaining, max);
1289 tio = alloc_tio(ci, ti);
1290 clone = split_bvec(bio, ci->sector, ci->idx,
1291 bv->bv_offset + offset, len,
1294 __map_bio(ti, clone, tio);
1297 ci->sector_count -= len;
1298 offset += to_bytes(len);
1299 } while (remaining -= len);
1308 * Split the bio into several clones and submit it to targets.
1310 static void __split_and_process_bio(struct mapped_device *md, struct bio *bio)
1312 struct clone_info ci;
1315 ci.map = dm_get_live_table(md);
1316 if (unlikely(!ci.map)) {
1322 ci.io = alloc_io(md);
1324 atomic_set(&ci.io->io_count, 1);
1327 spin_lock_init(&ci.io->endio_lock);
1328 ci.sector = bio->bi_sector;
1329 ci.idx = bio->bi_idx;
1331 start_io_acct(ci.io);
1332 if (bio->bi_rw & REQ_FLUSH) {
1333 ci.bio = &ci.md->flush_bio;
1334 ci.sector_count = 0;
1335 error = __clone_and_map_empty_flush(&ci);
1336 /* dec_pending submits any data associated with flush */
1339 ci.sector_count = bio_sectors(bio);
1340 while (ci.sector_count && !error)
1341 error = __clone_and_map(&ci);
1344 /* drop the extra reference count */
1345 dec_pending(ci.io, error);
1346 dm_table_put(ci.map);
1348 /*-----------------------------------------------------------------
1350 *---------------------------------------------------------------*/
1352 static int dm_merge_bvec(struct request_queue *q,
1353 struct bvec_merge_data *bvm,
1354 struct bio_vec *biovec)
1356 struct mapped_device *md = q->queuedata;
1357 struct dm_table *map = dm_get_live_table(md);
1358 struct dm_target *ti;
1359 sector_t max_sectors;
1365 ti = dm_table_find_target(map, bvm->bi_sector);
1366 if (!dm_target_is_valid(ti))
1370 * Find maximum amount of I/O that won't need splitting
1372 max_sectors = min(max_io_len(bvm->bi_sector, ti),
1373 (sector_t) BIO_MAX_SECTORS);
1374 max_size = (max_sectors << SECTOR_SHIFT) - bvm->bi_size;
1379 * merge_bvec_fn() returns number of bytes
1380 * it can accept at this offset
1381 * max is precomputed maximal io size
1383 if (max_size && ti->type->merge)
1384 max_size = ti->type->merge(ti, bvm, biovec, max_size);
1386 * If the target doesn't support merge method and some of the devices
1387 * provided their merge_bvec method (we know this by looking at
1388 * queue_max_hw_sectors), then we can't allow bios with multiple vector
1389 * entries. So always set max_size to 0, and the code below allows
1392 else if (queue_max_hw_sectors(q) <= PAGE_SIZE >> 9)
1401 * Always allow an entire first page
1403 if (max_size <= biovec->bv_len && !(bvm->bi_size >> SECTOR_SHIFT))
1404 max_size = biovec->bv_len;
1410 * The request function that just remaps the bio built up by
1413 static void _dm_request(struct request_queue *q, struct bio *bio)
1415 int rw = bio_data_dir(bio);
1416 struct mapped_device *md = q->queuedata;
1419 down_read(&md->io_lock);
1421 cpu = part_stat_lock();
1422 part_stat_inc(cpu, &dm_disk(md)->part0, ios[rw]);
1423 part_stat_add(cpu, &dm_disk(md)->part0, sectors[rw], bio_sectors(bio));
1426 /* if we're suspended, we have to queue this io for later */
1427 if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))) {
1428 up_read(&md->io_lock);
1430 if (bio_rw(bio) != READA)
1437 __split_and_process_bio(md, bio);
1438 up_read(&md->io_lock);
1442 static int dm_request_based(struct mapped_device *md)
1444 return blk_queue_stackable(md->queue);
1447 static void dm_request(struct request_queue *q, struct bio *bio)
1449 struct mapped_device *md = q->queuedata;
1451 if (dm_request_based(md))
1452 blk_queue_bio(q, bio);
1454 _dm_request(q, bio);
1457 void dm_dispatch_request(struct request *rq)
1461 if (blk_queue_io_stat(rq->q))
1462 rq->cmd_flags |= REQ_IO_STAT;
1464 rq->start_time = jiffies;
1465 r = blk_insert_cloned_request(rq->q, rq);
1467 dm_complete_request(rq, r);
1469 EXPORT_SYMBOL_GPL(dm_dispatch_request);
1471 static void dm_rq_bio_destructor(struct bio *bio)
1473 struct dm_rq_clone_bio_info *info = bio->bi_private;
1474 struct mapped_device *md = info->tio->md;
1476 free_bio_info(info);
1477 bio_free(bio, md->bs);
1480 static int dm_rq_bio_constructor(struct bio *bio, struct bio *bio_orig,
1483 struct dm_rq_target_io *tio = data;
1484 struct mapped_device *md = tio->md;
1485 struct dm_rq_clone_bio_info *info = alloc_bio_info(md);
1490 info->orig = bio_orig;
1492 bio->bi_end_io = end_clone_bio;
1493 bio->bi_private = info;
1494 bio->bi_destructor = dm_rq_bio_destructor;
1499 static int setup_clone(struct request *clone, struct request *rq,
1500 struct dm_rq_target_io *tio)
1504 r = blk_rq_prep_clone(clone, rq, tio->md->bs, GFP_ATOMIC,
1505 dm_rq_bio_constructor, tio);
1509 clone->cmd = rq->cmd;
1510 clone->cmd_len = rq->cmd_len;
1511 clone->sense = rq->sense;
1512 clone->buffer = rq->buffer;
1513 clone->end_io = end_clone_request;
1514 clone->end_io_data = tio;
1519 static struct request *clone_rq(struct request *rq, struct mapped_device *md,
1522 struct request *clone;
1523 struct dm_rq_target_io *tio;
1525 tio = alloc_rq_tio(md, gfp_mask);
1533 memset(&tio->info, 0, sizeof(tio->info));
1535 clone = &tio->clone;
1536 if (setup_clone(clone, rq, tio)) {
1546 * Called with the queue lock held.
1548 static int dm_prep_fn(struct request_queue *q, struct request *rq)
1550 struct mapped_device *md = q->queuedata;
1551 struct request *clone;
1553 if (unlikely(rq->special)) {
1554 DMWARN("Already has something in rq->special.");
1555 return BLKPREP_KILL;
1558 clone = clone_rq(rq, md, GFP_ATOMIC);
1560 return BLKPREP_DEFER;
1562 rq->special = clone;
1563 rq->cmd_flags |= REQ_DONTPREP;
1570 * 0 : the request has been processed (not requeued)
1571 * !0 : the request has been requeued
1573 static int map_request(struct dm_target *ti, struct request *clone,
1574 struct mapped_device *md)
1576 int r, requeued = 0;
1577 struct dm_rq_target_io *tio = clone->end_io_data;
1580 r = ti->type->map_rq(ti, clone, &tio->info);
1582 case DM_MAPIO_SUBMITTED:
1583 /* The target has taken the I/O to submit by itself later */
1585 case DM_MAPIO_REMAPPED:
1586 /* The target has remapped the I/O so dispatch it */
1587 trace_block_rq_remap(clone->q, clone, disk_devt(dm_disk(md)),
1588 blk_rq_pos(tio->orig));
1589 dm_dispatch_request(clone);
1591 case DM_MAPIO_REQUEUE:
1592 /* The target wants to requeue the I/O */
1593 dm_requeue_unmapped_request(clone);
1598 DMWARN("unimplemented target map return value: %d", r);
1602 /* The target wants to complete the I/O */
1603 dm_kill_unmapped_request(clone, r);
1610 static struct request *dm_start_request(struct mapped_device *md, struct request *orig)
1612 struct request *clone;
1614 blk_start_request(orig);
1615 clone = orig->special;
1616 atomic_inc(&md->pending[rq_data_dir(clone)]);
1619 * Hold the md reference here for the in-flight I/O.
1620 * We can't rely on the reference count by device opener,
1621 * because the device may be closed during the request completion
1622 * when all bios are completed.
1623 * See the comment in rq_completed() too.
1631 * q->request_fn for request-based dm.
1632 * Called with the queue lock held.
1634 static void dm_request_fn(struct request_queue *q)
1636 struct mapped_device *md = q->queuedata;
1637 struct dm_table *map = dm_get_live_table(md);
1638 struct dm_target *ti;
1639 struct request *rq, *clone;
1643 * For suspend, check blk_queue_stopped() and increment
1644 * ->pending within a single queue_lock not to increment the
1645 * number of in-flight I/Os after the queue is stopped in
1648 while (!blk_queue_stopped(q)) {
1649 rq = blk_peek_request(q);
1653 /* always use block 0 to find the target for flushes for now */
1655 if (!(rq->cmd_flags & REQ_FLUSH))
1656 pos = blk_rq_pos(rq);
1658 ti = dm_table_find_target(map, pos);
1659 if (!dm_target_is_valid(ti)) {
1661 * Must perform setup, that dm_done() requires,
1662 * before calling dm_kill_unmapped_request
1664 DMERR_LIMIT("request attempted access beyond the end of device");
1665 clone = dm_start_request(md, rq);
1666 dm_kill_unmapped_request(clone, -EIO);
1670 if (ti->type->busy && ti->type->busy(ti))
1673 clone = dm_start_request(md, rq);
1675 spin_unlock(q->queue_lock);
1676 if (map_request(ti, clone, md))
1679 BUG_ON(!irqs_disabled());
1680 spin_lock(q->queue_lock);
1686 BUG_ON(!irqs_disabled());
1687 spin_lock(q->queue_lock);
1690 blk_delay_queue(q, HZ / 10);
1695 int dm_underlying_device_busy(struct request_queue *q)
1697 return blk_lld_busy(q);
1699 EXPORT_SYMBOL_GPL(dm_underlying_device_busy);
1701 static int dm_lld_busy(struct request_queue *q)
1704 struct mapped_device *md = q->queuedata;
1705 struct dm_table *map = dm_get_live_table(md);
1707 if (!map || test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))
1710 r = dm_table_any_busy_target(map);
1717 static int dm_any_congested(void *congested_data, int bdi_bits)
1720 struct mapped_device *md = congested_data;
1721 struct dm_table *map;
1723 if (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
1724 map = dm_get_live_table(md);
1727 * Request-based dm cares about only own queue for
1728 * the query about congestion status of request_queue
1730 if (dm_request_based(md))
1731 r = md->queue->backing_dev_info.state &
1734 r = dm_table_any_congested(map, bdi_bits);
1743 /*-----------------------------------------------------------------
1744 * An IDR is used to keep track of allocated minor numbers.
1745 *---------------------------------------------------------------*/
1746 static void free_minor(int minor)
1748 spin_lock(&_minor_lock);
1749 idr_remove(&_minor_idr, minor);
1750 spin_unlock(&_minor_lock);
1754 * See if the device with a specific minor # is free.
1756 static int specific_minor(int minor)
1760 if (minor >= (1 << MINORBITS))
1763 r = idr_pre_get(&_minor_idr, GFP_KERNEL);
1767 spin_lock(&_minor_lock);
1769 if (idr_find(&_minor_idr, minor)) {
1774 r = idr_get_new_above(&_minor_idr, MINOR_ALLOCED, minor, &m);
1779 idr_remove(&_minor_idr, m);
1785 spin_unlock(&_minor_lock);
1789 static int next_free_minor(int *minor)
1793 r = idr_pre_get(&_minor_idr, GFP_KERNEL);
1797 spin_lock(&_minor_lock);
1799 r = idr_get_new(&_minor_idr, MINOR_ALLOCED, &m);
1803 if (m >= (1 << MINORBITS)) {
1804 idr_remove(&_minor_idr, m);
1812 spin_unlock(&_minor_lock);
1816 static const struct block_device_operations dm_blk_dops;
1818 static void dm_wq_work(struct work_struct *work);
1820 static void dm_init_md_queue(struct mapped_device *md)
1823 * Request-based dm devices cannot be stacked on top of bio-based dm
1824 * devices. The type of this dm device has not been decided yet.
1825 * The type is decided at the first table loading time.
1826 * To prevent problematic device stacking, clear the queue flag
1827 * for request stacking support until then.
1829 * This queue is new, so no concurrency on the queue_flags.
1831 queue_flag_clear_unlocked(QUEUE_FLAG_STACKABLE, md->queue);
1833 md->queue->queuedata = md;
1834 md->queue->backing_dev_info.congested_fn = dm_any_congested;
1835 md->queue->backing_dev_info.congested_data = md;
1836 blk_queue_make_request(md->queue, dm_request);
1837 blk_queue_bounce_limit(md->queue, BLK_BOUNCE_ANY);
1838 blk_queue_merge_bvec(md->queue, dm_merge_bvec);
1842 * Allocate and initialise a blank device with a given minor.
1844 static struct mapped_device *alloc_dev(int minor)
1847 struct mapped_device *md = kzalloc(sizeof(*md), GFP_KERNEL);
1851 DMWARN("unable to allocate device, out of memory.");
1855 if (!try_module_get(THIS_MODULE))
1856 goto bad_module_get;
1858 /* get a minor number for the dev */
1859 if (minor == DM_ANY_MINOR)
1860 r = next_free_minor(&minor);
1862 r = specific_minor(minor);
1866 md->type = DM_TYPE_NONE;
1867 init_rwsem(&md->io_lock);
1868 mutex_init(&md->suspend_lock);
1869 mutex_init(&md->type_lock);
1870 spin_lock_init(&md->deferred_lock);
1871 rwlock_init(&md->map_lock);
1872 atomic_set(&md->holders, 1);
1873 atomic_set(&md->open_count, 0);
1874 atomic_set(&md->event_nr, 0);
1875 atomic_set(&md->uevent_seq, 0);
1876 INIT_LIST_HEAD(&md->uevent_list);
1877 spin_lock_init(&md->uevent_lock);
1879 md->queue = blk_alloc_queue(GFP_KERNEL);
1883 dm_init_md_queue(md);
1885 md->disk = alloc_disk(1);
1889 atomic_set(&md->pending[0], 0);
1890 atomic_set(&md->pending[1], 0);
1891 init_waitqueue_head(&md->wait);
1892 INIT_WORK(&md->work, dm_wq_work);
1893 init_waitqueue_head(&md->eventq);
1895 md->disk->major = _major;
1896 md->disk->first_minor = minor;
1897 md->disk->fops = &dm_blk_dops;
1898 md->disk->queue = md->queue;
1899 md->disk->private_data = md;
1900 sprintf(md->disk->disk_name, "dm-%d", minor);
1902 format_dev_t(md->name, MKDEV(_major, minor));
1904 md->wq = alloc_workqueue("kdmflush",
1905 WQ_NON_REENTRANT | WQ_MEM_RECLAIM, 0);
1909 md->bdev = bdget_disk(md->disk, 0);
1913 bio_init(&md->flush_bio);
1914 md->flush_bio.bi_bdev = md->bdev;
1915 md->flush_bio.bi_rw = WRITE_FLUSH;
1917 /* Populate the mapping, nobody knows we exist yet */
1918 spin_lock(&_minor_lock);
1919 old_md = idr_replace(&_minor_idr, md, minor);
1920 spin_unlock(&_minor_lock);
1922 BUG_ON(old_md != MINOR_ALLOCED);
1927 destroy_workqueue(md->wq);
1929 del_gendisk(md->disk);
1932 blk_cleanup_queue(md->queue);
1936 module_put(THIS_MODULE);
1942 static void unlock_fs(struct mapped_device *md);
1944 static void free_dev(struct mapped_device *md)
1946 int minor = MINOR(disk_devt(md->disk));
1950 destroy_workqueue(md->wq);
1952 mempool_destroy(md->tio_pool);
1954 mempool_destroy(md->io_pool);
1956 bioset_free(md->bs);
1957 blk_integrity_unregister(md->disk);
1958 del_gendisk(md->disk);
1961 spin_lock(&_minor_lock);
1962 md->disk->private_data = NULL;
1963 spin_unlock(&_minor_lock);
1966 blk_cleanup_queue(md->queue);
1967 module_put(THIS_MODULE);
1971 static void __bind_mempools(struct mapped_device *md, struct dm_table *t)
1973 struct dm_md_mempools *p;
1975 if (md->io_pool && md->tio_pool && md->bs)
1976 /* the md already has necessary mempools */
1979 p = dm_table_get_md_mempools(t);
1980 BUG_ON(!p || md->io_pool || md->tio_pool || md->bs);
1982 md->io_pool = p->io_pool;
1984 md->tio_pool = p->tio_pool;
1990 /* mempool bind completed, now no need any mempools in the table */
1991 dm_table_free_md_mempools(t);
1995 * Bind a table to the device.
1997 static void event_callback(void *context)
1999 unsigned long flags;
2001 struct mapped_device *md = (struct mapped_device *) context;
2003 spin_lock_irqsave(&md->uevent_lock, flags);
2004 list_splice_init(&md->uevent_list, &uevents);
2005 spin_unlock_irqrestore(&md->uevent_lock, flags);
2007 dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj);
2009 atomic_inc(&md->event_nr);
2010 wake_up(&md->eventq);
2014 * Protected by md->suspend_lock obtained by dm_swap_table().
2016 static void __set_size(struct mapped_device *md, sector_t size)
2018 set_capacity(md->disk, size);
2020 i_size_write(md->bdev->bd_inode, (loff_t)size << SECTOR_SHIFT);
2024 * Return 1 if the queue has a compulsory merge_bvec_fn function.
2026 * If this function returns 0, then the device is either a non-dm
2027 * device without a merge_bvec_fn, or it is a dm device that is
2028 * able to split any bios it receives that are too big.
2030 int dm_queue_merge_is_compulsory(struct request_queue *q)
2032 struct mapped_device *dev_md;
2034 if (!q->merge_bvec_fn)
2037 if (q->make_request_fn == dm_request) {
2038 dev_md = q->queuedata;
2039 if (test_bit(DMF_MERGE_IS_OPTIONAL, &dev_md->flags))
2046 static int dm_device_merge_is_compulsory(struct dm_target *ti,
2047 struct dm_dev *dev, sector_t start,
2048 sector_t len, void *data)
2050 struct block_device *bdev = dev->bdev;
2051 struct request_queue *q = bdev_get_queue(bdev);
2053 return dm_queue_merge_is_compulsory(q);
2057 * Return 1 if it is acceptable to ignore merge_bvec_fn based
2058 * on the properties of the underlying devices.
2060 static int dm_table_merge_is_optional(struct dm_table *table)
2063 struct dm_target *ti;
2065 while (i < dm_table_get_num_targets(table)) {
2066 ti = dm_table_get_target(table, i++);
2068 if (ti->type->iterate_devices &&
2069 ti->type->iterate_devices(ti, dm_device_merge_is_compulsory, NULL))
2077 * Returns old map, which caller must destroy.
2079 static struct dm_table *__bind(struct mapped_device *md, struct dm_table *t,
2080 struct queue_limits *limits)
2082 struct dm_table *old_map;
2083 struct request_queue *q = md->queue;
2085 unsigned long flags;
2086 int merge_is_optional;
2088 size = dm_table_get_size(t);
2091 * Wipe any geometry if the size of the table changed.
2093 if (size != get_capacity(md->disk))
2094 memset(&md->geometry, 0, sizeof(md->geometry));
2096 __set_size(md, size);
2098 dm_table_event_callback(t, event_callback, md);
2101 * The queue hasn't been stopped yet, if the old table type wasn't
2102 * for request-based during suspension. So stop it to prevent
2103 * I/O mapping before resume.
2104 * This must be done before setting the queue restrictions,
2105 * because request-based dm may be run just after the setting.
2107 if (dm_table_request_based(t) && !blk_queue_stopped(q))
2110 __bind_mempools(md, t);
2112 merge_is_optional = dm_table_merge_is_optional(t);
2114 write_lock_irqsave(&md->map_lock, flags);
2117 md->immutable_target_type = dm_table_get_immutable_target_type(t);
2119 dm_table_set_restrictions(t, q, limits);
2120 if (merge_is_optional)
2121 set_bit(DMF_MERGE_IS_OPTIONAL, &md->flags);
2123 clear_bit(DMF_MERGE_IS_OPTIONAL, &md->flags);
2124 write_unlock_irqrestore(&md->map_lock, flags);
2130 * Returns unbound table for the caller to free.
2132 static struct dm_table *__unbind(struct mapped_device *md)
2134 struct dm_table *map = md->map;
2135 unsigned long flags;
2140 dm_table_event_callback(map, NULL, NULL);
2141 write_lock_irqsave(&md->map_lock, flags);
2143 write_unlock_irqrestore(&md->map_lock, flags);
2149 * Constructor for a new device.
2151 int dm_create(int minor, struct mapped_device **result)
2153 struct mapped_device *md;
2155 md = alloc_dev(minor);
2166 * Functions to manage md->type.
2167 * All are required to hold md->type_lock.
2169 void dm_lock_md_type(struct mapped_device *md)
2171 mutex_lock(&md->type_lock);
2174 void dm_unlock_md_type(struct mapped_device *md)
2176 mutex_unlock(&md->type_lock);
2179 void dm_set_md_type(struct mapped_device *md, unsigned type)
2184 unsigned dm_get_md_type(struct mapped_device *md)
2189 struct target_type *dm_get_immutable_target_type(struct mapped_device *md)
2191 return md->immutable_target_type;
2195 * Fully initialize a request-based queue (->elevator, ->request_fn, etc).
2197 static int dm_init_request_based_queue(struct mapped_device *md)
2199 struct request_queue *q = NULL;
2201 if (md->queue->elevator)
2204 /* Fully initialize the queue */
2205 q = blk_init_allocated_queue(md->queue, dm_request_fn, NULL);
2210 dm_init_md_queue(md);
2211 blk_queue_softirq_done(md->queue, dm_softirq_done);
2212 blk_queue_prep_rq(md->queue, dm_prep_fn);
2213 blk_queue_lld_busy(md->queue, dm_lld_busy);
2215 elv_register_queue(md->queue);
2221 * Setup the DM device's queue based on md's type
2223 int dm_setup_md_queue(struct mapped_device *md)
2225 if ((dm_get_md_type(md) == DM_TYPE_REQUEST_BASED) &&
2226 !dm_init_request_based_queue(md)) {
2227 DMWARN("Cannot initialize queue for request-based mapped device");
2234 static struct mapped_device *dm_find_md(dev_t dev)
2236 struct mapped_device *md;
2237 unsigned minor = MINOR(dev);
2239 if (MAJOR(dev) != _major || minor >= (1 << MINORBITS))
2242 spin_lock(&_minor_lock);
2244 md = idr_find(&_minor_idr, minor);
2245 if (md && (md == MINOR_ALLOCED ||
2246 (MINOR(disk_devt(dm_disk(md))) != minor) ||
2247 dm_deleting_md(md) ||
2248 test_bit(DMF_FREEING, &md->flags))) {
2254 spin_unlock(&_minor_lock);
2259 struct mapped_device *dm_get_md(dev_t dev)
2261 struct mapped_device *md = dm_find_md(dev);
2268 EXPORT_SYMBOL_GPL(dm_get_md);
2270 void *dm_get_mdptr(struct mapped_device *md)
2272 return md->interface_ptr;
2275 void dm_set_mdptr(struct mapped_device *md, void *ptr)
2277 md->interface_ptr = ptr;
2280 void dm_get(struct mapped_device *md)
2282 atomic_inc(&md->holders);
2283 BUG_ON(test_bit(DMF_FREEING, &md->flags));
2286 const char *dm_device_name(struct mapped_device *md)
2290 EXPORT_SYMBOL_GPL(dm_device_name);
2292 static void __dm_destroy(struct mapped_device *md, bool wait)
2294 struct dm_table *map;
2298 spin_lock(&_minor_lock);
2299 map = dm_get_live_table(md);
2300 idr_replace(&_minor_idr, MINOR_ALLOCED, MINOR(disk_devt(dm_disk(md))));
2301 set_bit(DMF_FREEING, &md->flags);
2302 spin_unlock(&_minor_lock);
2304 if (!dm_suspended_md(md)) {
2305 dm_table_presuspend_targets(map);
2306 dm_table_postsuspend_targets(map);
2310 * Rare, but there may be I/O requests still going to complete,
2311 * for example. Wait for all references to disappear.
2312 * No one should increment the reference count of the mapped_device,
2313 * after the mapped_device state becomes DMF_FREEING.
2316 while (atomic_read(&md->holders))
2318 else if (atomic_read(&md->holders))
2319 DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)",
2320 dm_device_name(md), atomic_read(&md->holders));
2324 dm_table_destroy(__unbind(md));
2328 void dm_destroy(struct mapped_device *md)
2330 __dm_destroy(md, true);
2333 void dm_destroy_immediate(struct mapped_device *md)
2335 __dm_destroy(md, false);
2338 void dm_put(struct mapped_device *md)
2340 atomic_dec(&md->holders);
2342 EXPORT_SYMBOL_GPL(dm_put);
2344 static int dm_wait_for_completion(struct mapped_device *md, int interruptible)
2347 DECLARE_WAITQUEUE(wait, current);
2349 add_wait_queue(&md->wait, &wait);
2352 set_current_state(interruptible);
2354 if (!md_in_flight(md))
2357 if (interruptible == TASK_INTERRUPTIBLE &&
2358 signal_pending(current)) {
2365 set_current_state(TASK_RUNNING);
2367 remove_wait_queue(&md->wait, &wait);
2373 * Process the deferred bios
2375 static void dm_wq_work(struct work_struct *work)
2377 struct mapped_device *md = container_of(work, struct mapped_device,
2381 down_read(&md->io_lock);
2383 while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
2384 spin_lock_irq(&md->deferred_lock);
2385 c = bio_list_pop(&md->deferred);
2386 spin_unlock_irq(&md->deferred_lock);
2391 up_read(&md->io_lock);
2393 if (dm_request_based(md))
2394 generic_make_request(c);
2396 __split_and_process_bio(md, c);
2398 down_read(&md->io_lock);
2401 up_read(&md->io_lock);
2404 static void dm_queue_flush(struct mapped_device *md)
2406 clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2407 smp_mb__after_clear_bit();
2408 queue_work(md->wq, &md->work);
2412 * Swap in a new table, returning the old one for the caller to destroy.
2414 struct dm_table *dm_swap_table(struct mapped_device *md, struct dm_table *table)
2416 struct dm_table *map = ERR_PTR(-EINVAL);
2417 struct queue_limits limits;
2420 mutex_lock(&md->suspend_lock);
2422 /* device must be suspended */
2423 if (!dm_suspended_md(md))
2426 r = dm_calculate_queue_limits(table, &limits);
2432 map = __bind(md, table, &limits);
2435 mutex_unlock(&md->suspend_lock);
2440 * Functions to lock and unlock any filesystem running on the
2443 static int lock_fs(struct mapped_device *md)
2447 WARN_ON(md->frozen_sb);
2449 md->frozen_sb = freeze_bdev(md->bdev);
2450 if (IS_ERR(md->frozen_sb)) {
2451 r = PTR_ERR(md->frozen_sb);
2452 md->frozen_sb = NULL;
2456 set_bit(DMF_FROZEN, &md->flags);
2461 static void unlock_fs(struct mapped_device *md)
2463 if (!test_bit(DMF_FROZEN, &md->flags))
2466 thaw_bdev(md->bdev, md->frozen_sb);
2467 md->frozen_sb = NULL;
2468 clear_bit(DMF_FROZEN, &md->flags);
2472 * We need to be able to change a mapping table under a mounted
2473 * filesystem. For example we might want to move some data in
2474 * the background. Before the table can be swapped with
2475 * dm_bind_table, dm_suspend must be called to flush any in
2476 * flight bios and ensure that any further io gets deferred.
2479 * Suspend mechanism in request-based dm.
2481 * 1. Flush all I/Os by lock_fs() if needed.
2482 * 2. Stop dispatching any I/O by stopping the request_queue.
2483 * 3. Wait for all in-flight I/Os to be completed or requeued.
2485 * To abort suspend, start the request_queue.
2487 int dm_suspend(struct mapped_device *md, unsigned suspend_flags)
2489 struct dm_table *map = NULL;
2491 int do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG ? 1 : 0;
2492 int noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG ? 1 : 0;
2494 mutex_lock(&md->suspend_lock);
2496 if (dm_suspended_md(md)) {
2501 map = dm_get_live_table(md);
2504 * DMF_NOFLUSH_SUSPENDING must be set before presuspend.
2505 * This flag is cleared before dm_suspend returns.
2508 set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2510 /* This does not get reverted if there's an error later. */
2511 dm_table_presuspend_targets(map);
2514 * Flush I/O to the device.
2515 * Any I/O submitted after lock_fs() may not be flushed.
2516 * noflush takes precedence over do_lockfs.
2517 * (lock_fs() flushes I/Os and waits for them to complete.)
2519 if (!noflush && do_lockfs) {
2526 * Here we must make sure that no processes are submitting requests
2527 * to target drivers i.e. no one may be executing
2528 * __split_and_process_bio. This is called from dm_request and
2531 * To get all processes out of __split_and_process_bio in dm_request,
2532 * we take the write lock. To prevent any process from reentering
2533 * __split_and_process_bio from dm_request and quiesce the thread
2534 * (dm_wq_work), we set BMF_BLOCK_IO_FOR_SUSPEND and call
2535 * flush_workqueue(md->wq).
2537 down_write(&md->io_lock);
2538 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2539 up_write(&md->io_lock);
2542 * Stop md->queue before flushing md->wq in case request-based
2543 * dm defers requests to md->wq from md->queue.
2545 if (dm_request_based(md))
2546 stop_queue(md->queue);
2548 flush_workqueue(md->wq);
2551 * At this point no more requests are entering target request routines.
2552 * We call dm_wait_for_completion to wait for all existing requests
2555 r = dm_wait_for_completion(md, TASK_INTERRUPTIBLE);
2557 down_write(&md->io_lock);
2559 clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2560 up_write(&md->io_lock);
2562 /* were we interrupted ? */
2566 if (dm_request_based(md))
2567 start_queue(md->queue);
2570 goto out; /* pushback list is already flushed, so skip flush */
2574 * If dm_wait_for_completion returned 0, the device is completely
2575 * quiescent now. There is no request-processing activity. All new
2576 * requests are being added to md->deferred list.
2579 set_bit(DMF_SUSPENDED, &md->flags);
2581 dm_table_postsuspend_targets(map);
2587 mutex_unlock(&md->suspend_lock);
2591 int dm_resume(struct mapped_device *md)
2594 struct dm_table *map = NULL;
2596 mutex_lock(&md->suspend_lock);
2597 if (!dm_suspended_md(md))
2600 map = dm_get_live_table(md);
2601 if (!map || !dm_table_get_size(map))
2604 r = dm_table_resume_targets(map);
2611 * Flushing deferred I/Os must be done after targets are resumed
2612 * so that mapping of targets can work correctly.
2613 * Request-based dm is queueing the deferred I/Os in its request_queue.
2615 if (dm_request_based(md))
2616 start_queue(md->queue);
2620 clear_bit(DMF_SUSPENDED, &md->flags);
2625 mutex_unlock(&md->suspend_lock);
2630 /*-----------------------------------------------------------------
2631 * Event notification.
2632 *---------------------------------------------------------------*/
2633 int dm_kobject_uevent(struct mapped_device *md, enum kobject_action action,
2636 char udev_cookie[DM_COOKIE_LENGTH];
2637 char *envp[] = { udev_cookie, NULL };
2640 return kobject_uevent(&disk_to_dev(md->disk)->kobj, action);
2642 snprintf(udev_cookie, DM_COOKIE_LENGTH, "%s=%u",
2643 DM_COOKIE_ENV_VAR_NAME, cookie);
2644 return kobject_uevent_env(&disk_to_dev(md->disk)->kobj,
2649 uint32_t dm_next_uevent_seq(struct mapped_device *md)
2651 return atomic_add_return(1, &md->uevent_seq);
2654 uint32_t dm_get_event_nr(struct mapped_device *md)
2656 return atomic_read(&md->event_nr);
2659 int dm_wait_event(struct mapped_device *md, int event_nr)
2661 return wait_event_interruptible(md->eventq,
2662 (event_nr != atomic_read(&md->event_nr)));
2665 void dm_uevent_add(struct mapped_device *md, struct list_head *elist)
2667 unsigned long flags;
2669 spin_lock_irqsave(&md->uevent_lock, flags);
2670 list_add(elist, &md->uevent_list);
2671 spin_unlock_irqrestore(&md->uevent_lock, flags);
2675 * The gendisk is only valid as long as you have a reference
2678 struct gendisk *dm_disk(struct mapped_device *md)
2683 struct kobject *dm_kobject(struct mapped_device *md)
2689 * struct mapped_device should not be exported outside of dm.c
2690 * so use this check to verify that kobj is part of md structure
2692 struct mapped_device *dm_get_from_kobject(struct kobject *kobj)
2694 struct mapped_device *md;
2696 md = container_of(kobj, struct mapped_device, kobj);
2697 if (&md->kobj != kobj)
2700 if (test_bit(DMF_FREEING, &md->flags) ||
2708 int dm_suspended_md(struct mapped_device *md)
2710 return test_bit(DMF_SUSPENDED, &md->flags);
2713 int dm_suspended(struct dm_target *ti)
2715 return dm_suspended_md(dm_table_get_md(ti->table));
2717 EXPORT_SYMBOL_GPL(dm_suspended);
2719 int dm_noflush_suspending(struct dm_target *ti)
2721 return __noflush_suspending(dm_table_get_md(ti->table));
2723 EXPORT_SYMBOL_GPL(dm_noflush_suspending);
2725 struct dm_md_mempools *dm_alloc_md_mempools(unsigned type, unsigned integrity)
2727 struct dm_md_mempools *pools = kmalloc(sizeof(*pools), GFP_KERNEL);
2728 unsigned int pool_size = (type == DM_TYPE_BIO_BASED) ? 16 : MIN_IOS;
2733 pools->io_pool = (type == DM_TYPE_BIO_BASED) ?
2734 mempool_create_slab_pool(MIN_IOS, _io_cache) :
2735 mempool_create_slab_pool(MIN_IOS, _rq_bio_info_cache);
2736 if (!pools->io_pool)
2737 goto free_pools_and_out;
2739 pools->tio_pool = (type == DM_TYPE_BIO_BASED) ?
2740 mempool_create_slab_pool(MIN_IOS, _tio_cache) :
2741 mempool_create_slab_pool(MIN_IOS, _rq_tio_cache);
2742 if (!pools->tio_pool)
2743 goto free_io_pool_and_out;
2745 pools->bs = bioset_create(pool_size, 0);
2747 goto free_tio_pool_and_out;
2749 if (integrity && bioset_integrity_create(pools->bs, pool_size))
2750 goto free_bioset_and_out;
2754 free_bioset_and_out:
2755 bioset_free(pools->bs);
2757 free_tio_pool_and_out:
2758 mempool_destroy(pools->tio_pool);
2760 free_io_pool_and_out:
2761 mempool_destroy(pools->io_pool);
2769 void dm_free_md_mempools(struct dm_md_mempools *pools)
2775 mempool_destroy(pools->io_pool);
2777 if (pools->tio_pool)
2778 mempool_destroy(pools->tio_pool);
2781 bioset_free(pools->bs);
2786 static const struct block_device_operations dm_blk_dops = {
2787 .open = dm_blk_open,
2788 .release = dm_blk_close,
2789 .ioctl = dm_blk_ioctl,
2790 .getgeo = dm_blk_getgeo,
2791 .owner = THIS_MODULE
2794 EXPORT_SYMBOL(dm_get_mapinfo);
2799 module_init(dm_init);
2800 module_exit(dm_exit);
2802 module_param(major, uint, 0);
2803 MODULE_PARM_DESC(major, "The major number of the device mapper");
2804 MODULE_DESCRIPTION(DM_NAME " driver");
2805 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
2806 MODULE_LICENSE("GPL");