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 /* wait until the kobject is released */
199 struct completion kobj_completion;
201 /* zero-length flush that will be cloned and submitted to targets */
202 struct bio flush_bio;
206 * For mempools pre-allocation at the table loading time.
208 struct dm_md_mempools {
215 static struct kmem_cache *_io_cache;
216 static struct kmem_cache *_tio_cache;
217 static struct kmem_cache *_rq_tio_cache;
218 static struct kmem_cache *_rq_bio_info_cache;
220 static int __init local_init(void)
224 /* allocate a slab for the dm_ios */
225 _io_cache = KMEM_CACHE(dm_io, 0);
229 /* allocate a slab for the target ios */
230 _tio_cache = KMEM_CACHE(dm_target_io, 0);
232 goto out_free_io_cache;
234 _rq_tio_cache = KMEM_CACHE(dm_rq_target_io, 0);
236 goto out_free_tio_cache;
238 _rq_bio_info_cache = KMEM_CACHE(dm_rq_clone_bio_info, 0);
239 if (!_rq_bio_info_cache)
240 goto out_free_rq_tio_cache;
242 r = dm_uevent_init();
244 goto out_free_rq_bio_info_cache;
247 r = register_blkdev(_major, _name);
249 goto out_uevent_exit;
258 out_free_rq_bio_info_cache:
259 kmem_cache_destroy(_rq_bio_info_cache);
260 out_free_rq_tio_cache:
261 kmem_cache_destroy(_rq_tio_cache);
263 kmem_cache_destroy(_tio_cache);
265 kmem_cache_destroy(_io_cache);
270 static void local_exit(void)
272 kmem_cache_destroy(_rq_bio_info_cache);
273 kmem_cache_destroy(_rq_tio_cache);
274 kmem_cache_destroy(_tio_cache);
275 kmem_cache_destroy(_io_cache);
276 unregister_blkdev(_major, _name);
281 DMINFO("cleaned up");
284 static int (*_inits[])(void) __initdata = {
294 static void (*_exits[])(void) = {
304 static int __init dm_init(void)
306 const int count = ARRAY_SIZE(_inits);
310 for (i = 0; i < count; i++) {
325 static void __exit dm_exit(void)
327 int i = ARRAY_SIZE(_exits);
333 * Should be empty by this point.
335 idr_remove_all(&_minor_idr);
336 idr_destroy(&_minor_idr);
340 * Block device functions
342 int dm_deleting_md(struct mapped_device *md)
344 return test_bit(DMF_DELETING, &md->flags);
347 static int dm_blk_open(struct block_device *bdev, fmode_t mode)
349 struct mapped_device *md;
351 spin_lock(&_minor_lock);
353 md = bdev->bd_disk->private_data;
357 if (test_bit(DMF_FREEING, &md->flags) ||
358 dm_deleting_md(md)) {
364 atomic_inc(&md->open_count);
367 spin_unlock(&_minor_lock);
369 return md ? 0 : -ENXIO;
372 static int dm_blk_close(struct gendisk *disk, fmode_t mode)
374 struct mapped_device *md = disk->private_data;
376 spin_lock(&_minor_lock);
378 atomic_dec(&md->open_count);
381 spin_unlock(&_minor_lock);
386 int dm_open_count(struct mapped_device *md)
388 return atomic_read(&md->open_count);
392 * Guarantees nothing is using the device before it's deleted.
394 int dm_lock_for_deletion(struct mapped_device *md)
398 spin_lock(&_minor_lock);
400 if (dm_open_count(md))
403 set_bit(DMF_DELETING, &md->flags);
405 spin_unlock(&_minor_lock);
410 static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo)
412 struct mapped_device *md = bdev->bd_disk->private_data;
414 return dm_get_geometry(md, geo);
417 static int dm_blk_ioctl(struct block_device *bdev, fmode_t mode,
418 unsigned int cmd, unsigned long arg)
420 struct mapped_device *md = bdev->bd_disk->private_data;
421 struct dm_table *map = dm_get_live_table(md);
422 struct dm_target *tgt;
425 if (!map || !dm_table_get_size(map))
428 /* We only support devices that have a single target */
429 if (dm_table_get_num_targets(map) != 1)
432 tgt = dm_table_get_target(map, 0);
434 if (dm_suspended_md(md)) {
439 if (tgt->type->ioctl)
440 r = tgt->type->ioctl(tgt, cmd, arg);
448 static struct dm_io *alloc_io(struct mapped_device *md)
450 return mempool_alloc(md->io_pool, GFP_NOIO);
453 static void free_io(struct mapped_device *md, struct dm_io *io)
455 mempool_free(io, md->io_pool);
458 static void free_tio(struct mapped_device *md, struct dm_target_io *tio)
460 mempool_free(tio, md->tio_pool);
463 static struct dm_rq_target_io *alloc_rq_tio(struct mapped_device *md,
466 return mempool_alloc(md->tio_pool, gfp_mask);
469 static void free_rq_tio(struct dm_rq_target_io *tio)
471 mempool_free(tio, tio->md->tio_pool);
474 static struct dm_rq_clone_bio_info *alloc_bio_info(struct mapped_device *md)
476 return mempool_alloc(md->io_pool, GFP_ATOMIC);
479 static void free_bio_info(struct dm_rq_clone_bio_info *info)
481 mempool_free(info, info->tio->md->io_pool);
484 static int md_in_flight(struct mapped_device *md)
486 return atomic_read(&md->pending[READ]) +
487 atomic_read(&md->pending[WRITE]);
490 static void start_io_acct(struct dm_io *io)
492 struct mapped_device *md = io->md;
494 int rw = bio_data_dir(io->bio);
496 io->start_time = jiffies;
498 cpu = part_stat_lock();
499 part_round_stats(cpu, &dm_disk(md)->part0);
501 atomic_set(&dm_disk(md)->part0.in_flight[rw],
502 atomic_inc_return(&md->pending[rw]));
505 static void end_io_acct(struct dm_io *io)
507 struct mapped_device *md = io->md;
508 struct bio *bio = io->bio;
509 unsigned long duration = jiffies - io->start_time;
511 int rw = bio_data_dir(bio);
513 cpu = part_stat_lock();
514 part_round_stats(cpu, &dm_disk(md)->part0);
515 part_stat_add(cpu, &dm_disk(md)->part0, ticks[rw], duration);
519 * After this is decremented the bio must not be touched if it is
522 pending = atomic_dec_return(&md->pending[rw]);
523 atomic_set(&dm_disk(md)->part0.in_flight[rw], pending);
524 pending += atomic_read(&md->pending[rw^0x1]);
526 /* nudge anyone waiting on suspend queue */
532 * Add the bio to the list of deferred io.
534 static void queue_io(struct mapped_device *md, struct bio *bio)
538 spin_lock_irqsave(&md->deferred_lock, flags);
539 bio_list_add(&md->deferred, bio);
540 spin_unlock_irqrestore(&md->deferred_lock, flags);
541 queue_work(md->wq, &md->work);
545 * Everyone (including functions in this file), should use this
546 * function to access the md->map field, and make sure they call
547 * dm_table_put() when finished.
549 struct dm_table *dm_get_live_table(struct mapped_device *md)
554 read_lock_irqsave(&md->map_lock, flags);
558 read_unlock_irqrestore(&md->map_lock, flags);
564 * Get the geometry associated with a dm device
566 int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo)
574 * Set the geometry of a device.
576 int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo)
578 sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors;
580 if (geo->start > sz) {
581 DMWARN("Start sector is beyond the geometry limits.");
590 /*-----------------------------------------------------------------
592 * A more elegant soln is in the works that uses the queue
593 * merge fn, unfortunately there are a couple of changes to
594 * the block layer that I want to make for this. So in the
595 * interests of getting something for people to use I give
596 * you this clearly demarcated crap.
597 *---------------------------------------------------------------*/
599 static int __noflush_suspending(struct mapped_device *md)
601 return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
605 * Decrements the number of outstanding ios that a bio has been
606 * cloned into, completing the original io if necc.
608 static void dec_pending(struct dm_io *io, int error)
613 struct mapped_device *md = io->md;
615 /* Push-back supersedes any I/O errors */
616 if (unlikely(error)) {
617 spin_lock_irqsave(&io->endio_lock, flags);
618 if (!(io->error > 0 && __noflush_suspending(md)))
620 spin_unlock_irqrestore(&io->endio_lock, flags);
623 if (atomic_dec_and_test(&io->io_count)) {
624 if (io->error == DM_ENDIO_REQUEUE) {
626 * Target requested pushing back the I/O.
628 spin_lock_irqsave(&md->deferred_lock, flags);
629 if (__noflush_suspending(md))
630 bio_list_add_head(&md->deferred, io->bio);
632 /* noflush suspend was interrupted. */
634 spin_unlock_irqrestore(&md->deferred_lock, flags);
637 io_error = io->error;
642 if (io_error == DM_ENDIO_REQUEUE)
645 if ((bio->bi_rw & REQ_FLUSH) && bio->bi_size) {
647 * Preflush done for flush with data, reissue
650 bio->bi_rw &= ~REQ_FLUSH;
653 /* done with normal IO or empty flush */
654 trace_block_bio_complete(md->queue, bio, io_error);
655 bio_endio(bio, io_error);
660 static void clone_endio(struct bio *bio, int error)
663 struct dm_target_io *tio = bio->bi_private;
664 struct dm_io *io = tio->io;
665 struct mapped_device *md = tio->io->md;
666 dm_endio_fn endio = tio->ti->type->end_io;
668 if (!bio_flagged(bio, BIO_UPTODATE) && !error)
672 r = endio(tio->ti, bio, error, &tio->info);
673 if (r < 0 || r == DM_ENDIO_REQUEUE)
675 * error and requeue request are handled
679 else if (r == DM_ENDIO_INCOMPLETE)
680 /* The target will handle the io */
683 DMWARN("unimplemented target endio return value: %d", r);
689 * Store md for cleanup instead of tio which is about to get freed.
691 bio->bi_private = md->bs;
695 dec_pending(io, error);
699 * Partial completion handling for request-based dm
701 static void end_clone_bio(struct bio *clone, int error)
703 struct dm_rq_clone_bio_info *info = clone->bi_private;
704 struct dm_rq_target_io *tio = info->tio;
705 struct bio *bio = info->orig;
706 unsigned int nr_bytes = info->orig->bi_size;
712 * An error has already been detected on the request.
713 * Once error occurred, just let clone->end_io() handle
719 * Don't notice the error to the upper layer yet.
720 * The error handling decision is made by the target driver,
721 * when the request is completed.
728 * I/O for the bio successfully completed.
729 * Notice the data completion to the upper layer.
733 * bios are processed from the head of the list.
734 * So the completing bio should always be rq->bio.
735 * If it's not, something wrong is happening.
737 if (tio->orig->bio != bio)
738 DMERR("bio completion is going in the middle of the request");
741 * Update the original request.
742 * Do not use blk_end_request() here, because it may complete
743 * the original request before the clone, and break the ordering.
745 blk_update_request(tio->orig, 0, nr_bytes);
749 * Don't touch any member of the md after calling this function because
750 * the md may be freed in dm_put() at the end of this function.
751 * Or do dm_get() before calling this function and dm_put() later.
753 static void rq_completed(struct mapped_device *md, int rw, int run_queue)
755 atomic_dec(&md->pending[rw]);
757 /* nudge anyone waiting on suspend queue */
758 if (!md_in_flight(md))
762 * Run this off this callpath, as drivers could invoke end_io while
763 * inside their request_fn (and holding the queue lock). Calling
764 * back into ->request_fn() could deadlock attempting to grab the
768 blk_run_queue_async(md->queue);
771 * dm_put() must be at the end of this function. See the comment above
776 static void free_rq_clone(struct request *clone)
778 struct dm_rq_target_io *tio = clone->end_io_data;
780 blk_rq_unprep_clone(clone);
785 * Complete the clone and the original request.
786 * Must be called without queue lock.
788 static void dm_end_request(struct request *clone, int error)
790 int rw = rq_data_dir(clone);
791 struct dm_rq_target_io *tio = clone->end_io_data;
792 struct mapped_device *md = tio->md;
793 struct request *rq = tio->orig;
795 if (rq->cmd_type == REQ_TYPE_BLOCK_PC) {
796 rq->errors = clone->errors;
797 rq->resid_len = clone->resid_len;
801 * We are using the sense buffer of the original
803 * So setting the length of the sense data is enough.
805 rq->sense_len = clone->sense_len;
808 free_rq_clone(clone);
809 blk_end_request_all(rq, error);
810 rq_completed(md, rw, true);
813 static void dm_unprep_request(struct request *rq)
815 struct request *clone = rq->special;
818 rq->cmd_flags &= ~REQ_DONTPREP;
820 free_rq_clone(clone);
824 * Requeue the original request of a clone.
826 void dm_requeue_unmapped_request(struct request *clone)
828 int rw = rq_data_dir(clone);
829 struct dm_rq_target_io *tio = clone->end_io_data;
830 struct mapped_device *md = tio->md;
831 struct request *rq = tio->orig;
832 struct request_queue *q = rq->q;
835 dm_unprep_request(rq);
837 spin_lock_irqsave(q->queue_lock, flags);
838 blk_requeue_request(q, rq);
839 spin_unlock_irqrestore(q->queue_lock, flags);
841 rq_completed(md, rw, 0);
843 EXPORT_SYMBOL_GPL(dm_requeue_unmapped_request);
845 static void __stop_queue(struct request_queue *q)
850 static void stop_queue(struct request_queue *q)
854 spin_lock_irqsave(q->queue_lock, flags);
856 spin_unlock_irqrestore(q->queue_lock, flags);
859 static void __start_queue(struct request_queue *q)
861 if (blk_queue_stopped(q))
865 static void start_queue(struct request_queue *q)
869 spin_lock_irqsave(q->queue_lock, flags);
871 spin_unlock_irqrestore(q->queue_lock, flags);
874 static void dm_done(struct request *clone, int error, bool mapped)
877 struct dm_rq_target_io *tio = clone->end_io_data;
878 dm_request_endio_fn rq_end_io = NULL;
881 rq_end_io = tio->ti->type->rq_end_io;
883 if (mapped && rq_end_io)
884 r = rq_end_io(tio->ti, clone, error, &tio->info);
888 /* The target wants to complete the I/O */
889 dm_end_request(clone, r);
890 else if (r == DM_ENDIO_INCOMPLETE)
891 /* The target will handle the I/O */
893 else if (r == DM_ENDIO_REQUEUE)
894 /* The target wants to requeue the I/O */
895 dm_requeue_unmapped_request(clone);
897 DMWARN("unimplemented target endio return value: %d", r);
903 * Request completion handler for request-based dm
905 static void dm_softirq_done(struct request *rq)
908 struct request *clone = rq->completion_data;
909 struct dm_rq_target_io *tio = clone->end_io_data;
911 if (rq->cmd_flags & REQ_FAILED)
914 dm_done(clone, tio->error, mapped);
918 * Complete the clone and the original request with the error status
919 * through softirq context.
921 static void dm_complete_request(struct request *clone, int error)
923 struct dm_rq_target_io *tio = clone->end_io_data;
924 struct request *rq = tio->orig;
927 rq->completion_data = clone;
928 blk_complete_request(rq);
932 * Complete the not-mapped clone and the original request with the error status
933 * through softirq context.
934 * Target's rq_end_io() function isn't called.
935 * This may be used when the target's map_rq() function fails.
937 void dm_kill_unmapped_request(struct request *clone, int error)
939 struct dm_rq_target_io *tio = clone->end_io_data;
940 struct request *rq = tio->orig;
942 rq->cmd_flags |= REQ_FAILED;
943 dm_complete_request(clone, error);
945 EXPORT_SYMBOL_GPL(dm_kill_unmapped_request);
948 * Called with the queue lock held
950 static void end_clone_request(struct request *clone, int error)
953 * For just cleaning up the information of the queue in which
954 * the clone was dispatched.
955 * The clone is *NOT* freed actually here because it is alloced from
956 * dm own mempool and REQ_ALLOCED isn't set in clone->cmd_flags.
958 __blk_put_request(clone->q, clone);
961 * Actual request completion is done in a softirq context which doesn't
962 * hold the queue lock. Otherwise, deadlock could occur because:
963 * - another request may be submitted by the upper level driver
964 * of the stacking during the completion
965 * - the submission which requires queue lock may be done
968 dm_complete_request(clone, error);
972 * Return maximum size of I/O possible at the supplied sector up to the current
975 static sector_t max_io_len_target_boundary(sector_t sector, struct dm_target *ti)
977 sector_t target_offset = dm_target_offset(ti, sector);
979 return ti->len - target_offset;
982 static sector_t max_io_len(sector_t sector, struct dm_target *ti)
984 sector_t len = max_io_len_target_boundary(sector, ti);
987 * Does the target need to split even further ?
991 sector_t offset = dm_target_offset(ti, sector);
992 boundary = ((offset + ti->split_io) & ~(ti->split_io - 1))
1001 static void __map_bio(struct dm_target *ti, struct bio *clone,
1002 struct dm_target_io *tio)
1006 struct mapped_device *md;
1008 clone->bi_end_io = clone_endio;
1009 clone->bi_private = tio;
1012 * Map the clone. If r == 0 we don't need to do
1013 * anything, the target has assumed ownership of
1016 atomic_inc(&tio->io->io_count);
1017 sector = clone->bi_sector;
1018 r = ti->type->map(ti, clone, &tio->info);
1019 if (r == DM_MAPIO_REMAPPED) {
1020 /* the bio has been remapped so dispatch it */
1022 trace_block_bio_remap(bdev_get_queue(clone->bi_bdev), clone,
1023 tio->io->bio->bi_bdev->bd_dev, sector);
1025 generic_make_request(clone);
1026 } else if (r < 0 || r == DM_MAPIO_REQUEUE) {
1027 /* error the io and bail out, or requeue it if needed */
1029 dec_pending(tio->io, r);
1031 * Store bio_set for cleanup.
1033 clone->bi_private = md->bs;
1037 DMWARN("unimplemented target map return value: %d", r);
1043 struct mapped_device *md;
1044 struct dm_table *map;
1048 sector_t sector_count;
1052 static void dm_bio_destructor(struct bio *bio)
1054 struct bio_set *bs = bio->bi_private;
1060 * Creates a little bio that just does part of a bvec.
1062 static struct bio *split_bvec(struct bio *bio, sector_t sector,
1063 unsigned short idx, unsigned int offset,
1064 unsigned int len, struct bio_set *bs)
1067 struct bio_vec *bv = bio->bi_io_vec + idx;
1069 clone = bio_alloc_bioset(GFP_NOIO, 1, bs);
1070 clone->bi_destructor = dm_bio_destructor;
1071 *clone->bi_io_vec = *bv;
1073 clone->bi_sector = sector;
1074 clone->bi_bdev = bio->bi_bdev;
1075 clone->bi_rw = bio->bi_rw;
1077 clone->bi_size = to_bytes(len);
1078 clone->bi_io_vec->bv_offset = offset;
1079 clone->bi_io_vec->bv_len = clone->bi_size;
1080 clone->bi_flags |= 1 << BIO_CLONED;
1082 if (bio_integrity(bio)) {
1083 bio_integrity_clone(clone, bio, GFP_NOIO, bs);
1084 bio_integrity_trim(clone,
1085 bio_sector_offset(bio, idx, offset), len);
1092 * Creates a bio that consists of range of complete bvecs.
1094 static struct bio *clone_bio(struct bio *bio, sector_t sector,
1095 unsigned short idx, unsigned short bv_count,
1096 unsigned int len, struct bio_set *bs)
1100 clone = bio_alloc_bioset(GFP_NOIO, bio->bi_max_vecs, bs);
1101 __bio_clone(clone, bio);
1102 clone->bi_destructor = dm_bio_destructor;
1103 clone->bi_sector = sector;
1104 clone->bi_idx = idx;
1105 clone->bi_vcnt = idx + bv_count;
1106 clone->bi_size = to_bytes(len);
1107 clone->bi_flags &= ~(1 << BIO_SEG_VALID);
1109 if (bio_integrity(bio)) {
1110 bio_integrity_clone(clone, bio, GFP_NOIO, bs);
1112 if (idx != bio->bi_idx || clone->bi_size < bio->bi_size)
1113 bio_integrity_trim(clone,
1114 bio_sector_offset(bio, idx, 0), len);
1120 static struct dm_target_io *alloc_tio(struct clone_info *ci,
1121 struct dm_target *ti)
1123 struct dm_target_io *tio = mempool_alloc(ci->md->tio_pool, GFP_NOIO);
1127 memset(&tio->info, 0, sizeof(tio->info));
1132 static void __issue_target_request(struct clone_info *ci, struct dm_target *ti,
1133 unsigned request_nr, sector_t len)
1135 struct dm_target_io *tio = alloc_tio(ci, ti);
1138 tio->info.target_request_nr = request_nr;
1141 * Discard requests require the bio's inline iovecs be initialized.
1142 * ci->bio->bi_max_vecs is BIO_INLINE_VECS anyway, for both flush
1143 * and discard, so no need for concern about wasted bvec allocations.
1145 clone = bio_alloc_bioset(GFP_NOIO, ci->bio->bi_max_vecs, ci->md->bs);
1146 __bio_clone(clone, ci->bio);
1147 clone->bi_destructor = dm_bio_destructor;
1149 clone->bi_sector = ci->sector;
1150 clone->bi_size = to_bytes(len);
1153 __map_bio(ti, clone, tio);
1156 static void __issue_target_requests(struct clone_info *ci, struct dm_target *ti,
1157 unsigned num_requests, sector_t len)
1159 unsigned request_nr;
1161 for (request_nr = 0; request_nr < num_requests; request_nr++)
1162 __issue_target_request(ci, ti, request_nr, len);
1165 static int __clone_and_map_empty_flush(struct clone_info *ci)
1167 unsigned target_nr = 0;
1168 struct dm_target *ti;
1170 BUG_ON(bio_has_data(ci->bio));
1171 while ((ti = dm_table_get_target(ci->map, target_nr++)))
1172 __issue_target_requests(ci, ti, ti->num_flush_requests, 0);
1178 * Perform all io with a single clone.
1180 static void __clone_and_map_simple(struct clone_info *ci, struct dm_target *ti)
1182 struct bio *clone, *bio = ci->bio;
1183 struct dm_target_io *tio;
1185 tio = alloc_tio(ci, ti);
1186 clone = clone_bio(bio, ci->sector, ci->idx,
1187 bio->bi_vcnt - ci->idx, ci->sector_count,
1189 __map_bio(ti, clone, tio);
1190 ci->sector_count = 0;
1193 static int __clone_and_map_discard(struct clone_info *ci)
1195 struct dm_target *ti;
1199 ti = dm_table_find_target(ci->map, ci->sector);
1200 if (!dm_target_is_valid(ti))
1204 * Even though the device advertised discard support,
1205 * that does not mean every target supports it, and
1206 * reconfiguration might also have changed that since the
1207 * check was performed.
1209 if (!ti->num_discard_requests)
1212 len = min(ci->sector_count, max_io_len_target_boundary(ci->sector, ti));
1214 __issue_target_requests(ci, ti, ti->num_discard_requests, len);
1217 } while (ci->sector_count -= len);
1222 static int __clone_and_map(struct clone_info *ci)
1224 struct bio *clone, *bio = ci->bio;
1225 struct dm_target *ti;
1226 sector_t len = 0, max;
1227 struct dm_target_io *tio;
1229 if (unlikely(bio->bi_rw & REQ_DISCARD))
1230 return __clone_and_map_discard(ci);
1232 ti = dm_table_find_target(ci->map, ci->sector);
1233 if (!dm_target_is_valid(ti))
1236 max = max_io_len(ci->sector, ti);
1238 if (ci->sector_count <= max) {
1240 * Optimise for the simple case where we can do all of
1241 * the remaining io with a single clone.
1243 __clone_and_map_simple(ci, ti);
1245 } else if (to_sector(bio->bi_io_vec[ci->idx].bv_len) <= max) {
1247 * There are some bvecs that don't span targets.
1248 * Do as many of these as possible.
1251 sector_t remaining = max;
1254 for (i = ci->idx; remaining && (i < bio->bi_vcnt); i++) {
1255 bv_len = to_sector(bio->bi_io_vec[i].bv_len);
1257 if (bv_len > remaining)
1260 remaining -= bv_len;
1264 tio = alloc_tio(ci, ti);
1265 clone = clone_bio(bio, ci->sector, ci->idx, i - ci->idx, len,
1267 __map_bio(ti, clone, tio);
1270 ci->sector_count -= len;
1275 * Handle a bvec that must be split between two or more targets.
1277 struct bio_vec *bv = bio->bi_io_vec + ci->idx;
1278 sector_t remaining = to_sector(bv->bv_len);
1279 unsigned int offset = 0;
1283 ti = dm_table_find_target(ci->map, ci->sector);
1284 if (!dm_target_is_valid(ti))
1287 max = max_io_len(ci->sector, ti);
1290 len = min(remaining, max);
1292 tio = alloc_tio(ci, ti);
1293 clone = split_bvec(bio, ci->sector, ci->idx,
1294 bv->bv_offset + offset, len,
1297 __map_bio(ti, clone, tio);
1300 ci->sector_count -= len;
1301 offset += to_bytes(len);
1302 } while (remaining -= len);
1311 * Split the bio into several clones and submit it to targets.
1313 static void __split_and_process_bio(struct mapped_device *md, struct bio *bio)
1315 struct clone_info ci;
1318 ci.map = dm_get_live_table(md);
1319 if (unlikely(!ci.map)) {
1325 ci.io = alloc_io(md);
1327 atomic_set(&ci.io->io_count, 1);
1330 spin_lock_init(&ci.io->endio_lock);
1331 ci.sector = bio->bi_sector;
1332 ci.idx = bio->bi_idx;
1334 start_io_acct(ci.io);
1335 if (bio->bi_rw & REQ_FLUSH) {
1336 ci.bio = &ci.md->flush_bio;
1337 ci.sector_count = 0;
1338 error = __clone_and_map_empty_flush(&ci);
1339 /* dec_pending submits any data associated with flush */
1342 ci.sector_count = bio_sectors(bio);
1343 while (ci.sector_count && !error)
1344 error = __clone_and_map(&ci);
1347 /* drop the extra reference count */
1348 dec_pending(ci.io, error);
1349 dm_table_put(ci.map);
1351 /*-----------------------------------------------------------------
1353 *---------------------------------------------------------------*/
1355 static int dm_merge_bvec(struct request_queue *q,
1356 struct bvec_merge_data *bvm,
1357 struct bio_vec *biovec)
1359 struct mapped_device *md = q->queuedata;
1360 struct dm_table *map = dm_get_live_table(md);
1361 struct dm_target *ti;
1362 sector_t max_sectors;
1368 ti = dm_table_find_target(map, bvm->bi_sector);
1369 if (!dm_target_is_valid(ti))
1373 * Find maximum amount of I/O that won't need splitting
1375 max_sectors = min(max_io_len(bvm->bi_sector, ti),
1376 (sector_t) BIO_MAX_SECTORS);
1377 max_size = (max_sectors << SECTOR_SHIFT) - bvm->bi_size;
1382 * merge_bvec_fn() returns number of bytes
1383 * it can accept at this offset
1384 * max is precomputed maximal io size
1386 if (max_size && ti->type->merge)
1387 max_size = ti->type->merge(ti, bvm, biovec, max_size);
1389 * If the target doesn't support merge method and some of the devices
1390 * provided their merge_bvec method (we know this by looking at
1391 * queue_max_hw_sectors), then we can't allow bios with multiple vector
1392 * entries. So always set max_size to 0, and the code below allows
1395 else if (queue_max_hw_sectors(q) <= PAGE_SIZE >> 9)
1404 * Always allow an entire first page
1406 if (max_size <= biovec->bv_len && !(bvm->bi_size >> SECTOR_SHIFT))
1407 max_size = biovec->bv_len;
1413 * The request function that just remaps the bio built up by
1416 static void _dm_request(struct request_queue *q, struct bio *bio)
1418 int rw = bio_data_dir(bio);
1419 struct mapped_device *md = q->queuedata;
1422 down_read(&md->io_lock);
1424 cpu = part_stat_lock();
1425 part_stat_inc(cpu, &dm_disk(md)->part0, ios[rw]);
1426 part_stat_add(cpu, &dm_disk(md)->part0, sectors[rw], bio_sectors(bio));
1429 /* if we're suspended, we have to queue this io for later */
1430 if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))) {
1431 up_read(&md->io_lock);
1433 if (bio_rw(bio) != READA)
1440 __split_and_process_bio(md, bio);
1441 up_read(&md->io_lock);
1445 static int dm_request_based(struct mapped_device *md)
1447 return blk_queue_stackable(md->queue);
1450 static void dm_request(struct request_queue *q, struct bio *bio)
1452 struct mapped_device *md = q->queuedata;
1454 if (dm_request_based(md))
1455 blk_queue_bio(q, bio);
1457 _dm_request(q, bio);
1460 void dm_dispatch_request(struct request *rq)
1464 if (blk_queue_io_stat(rq->q))
1465 rq->cmd_flags |= REQ_IO_STAT;
1467 rq->start_time = jiffies;
1468 r = blk_insert_cloned_request(rq->q, rq);
1470 dm_complete_request(rq, r);
1472 EXPORT_SYMBOL_GPL(dm_dispatch_request);
1474 static void dm_rq_bio_destructor(struct bio *bio)
1476 struct dm_rq_clone_bio_info *info = bio->bi_private;
1477 struct mapped_device *md = info->tio->md;
1479 free_bio_info(info);
1480 bio_free(bio, md->bs);
1483 static int dm_rq_bio_constructor(struct bio *bio, struct bio *bio_orig,
1486 struct dm_rq_target_io *tio = data;
1487 struct mapped_device *md = tio->md;
1488 struct dm_rq_clone_bio_info *info = alloc_bio_info(md);
1493 info->orig = bio_orig;
1495 bio->bi_end_io = end_clone_bio;
1496 bio->bi_private = info;
1497 bio->bi_destructor = dm_rq_bio_destructor;
1502 static int setup_clone(struct request *clone, struct request *rq,
1503 struct dm_rq_target_io *tio)
1507 r = blk_rq_prep_clone(clone, rq, tio->md->bs, GFP_ATOMIC,
1508 dm_rq_bio_constructor, tio);
1512 clone->cmd = rq->cmd;
1513 clone->cmd_len = rq->cmd_len;
1514 clone->sense = rq->sense;
1515 clone->buffer = rq->buffer;
1516 clone->end_io = end_clone_request;
1517 clone->end_io_data = tio;
1522 static struct request *clone_rq(struct request *rq, struct mapped_device *md,
1525 struct request *clone;
1526 struct dm_rq_target_io *tio;
1528 tio = alloc_rq_tio(md, gfp_mask);
1536 memset(&tio->info, 0, sizeof(tio->info));
1538 clone = &tio->clone;
1539 if (setup_clone(clone, rq, tio)) {
1549 * Called with the queue lock held.
1551 static int dm_prep_fn(struct request_queue *q, struct request *rq)
1553 struct mapped_device *md = q->queuedata;
1554 struct request *clone;
1556 if (unlikely(rq->special)) {
1557 DMWARN("Already has something in rq->special.");
1558 return BLKPREP_KILL;
1561 clone = clone_rq(rq, md, GFP_ATOMIC);
1563 return BLKPREP_DEFER;
1565 rq->special = clone;
1566 rq->cmd_flags |= REQ_DONTPREP;
1573 * 0 : the request has been processed (not requeued)
1574 * !0 : the request has been requeued
1576 static int map_request(struct dm_target *ti, struct request *clone,
1577 struct mapped_device *md)
1579 int r, requeued = 0;
1580 struct dm_rq_target_io *tio = clone->end_io_data;
1583 r = ti->type->map_rq(ti, clone, &tio->info);
1585 case DM_MAPIO_SUBMITTED:
1586 /* The target has taken the I/O to submit by itself later */
1588 case DM_MAPIO_REMAPPED:
1589 /* The target has remapped the I/O so dispatch it */
1590 trace_block_rq_remap(clone->q, clone, disk_devt(dm_disk(md)),
1591 blk_rq_pos(tio->orig));
1592 dm_dispatch_request(clone);
1594 case DM_MAPIO_REQUEUE:
1595 /* The target wants to requeue the I/O */
1596 dm_requeue_unmapped_request(clone);
1601 DMWARN("unimplemented target map return value: %d", r);
1605 /* The target wants to complete the I/O */
1606 dm_kill_unmapped_request(clone, r);
1613 static struct request *dm_start_request(struct mapped_device *md, struct request *orig)
1615 struct request *clone;
1617 blk_start_request(orig);
1618 clone = orig->special;
1619 atomic_inc(&md->pending[rq_data_dir(clone)]);
1622 * Hold the md reference here for the in-flight I/O.
1623 * We can't rely on the reference count by device opener,
1624 * because the device may be closed during the request completion
1625 * when all bios are completed.
1626 * See the comment in rq_completed() too.
1634 * q->request_fn for request-based dm.
1635 * Called with the queue lock held.
1637 static void dm_request_fn(struct request_queue *q)
1639 struct mapped_device *md = q->queuedata;
1640 struct dm_table *map = dm_get_live_table(md);
1641 struct dm_target *ti;
1642 struct request *rq, *clone;
1646 * For suspend, check blk_queue_stopped() and increment
1647 * ->pending within a single queue_lock not to increment the
1648 * number of in-flight I/Os after the queue is stopped in
1651 while (!blk_queue_stopped(q)) {
1652 rq = blk_peek_request(q);
1656 /* always use block 0 to find the target for flushes for now */
1658 if (!(rq->cmd_flags & REQ_FLUSH))
1659 pos = blk_rq_pos(rq);
1661 ti = dm_table_find_target(map, pos);
1662 if (!dm_target_is_valid(ti)) {
1664 * Must perform setup, that dm_done() requires,
1665 * before calling dm_kill_unmapped_request
1667 DMERR_LIMIT("request attempted access beyond the end of device");
1668 clone = dm_start_request(md, rq);
1669 dm_kill_unmapped_request(clone, -EIO);
1673 if (ti->type->busy && ti->type->busy(ti))
1676 clone = dm_start_request(md, rq);
1678 spin_unlock(q->queue_lock);
1679 if (map_request(ti, clone, md))
1682 BUG_ON(!irqs_disabled());
1683 spin_lock(q->queue_lock);
1689 BUG_ON(!irqs_disabled());
1690 spin_lock(q->queue_lock);
1693 blk_delay_queue(q, HZ / 10);
1698 int dm_underlying_device_busy(struct request_queue *q)
1700 return blk_lld_busy(q);
1702 EXPORT_SYMBOL_GPL(dm_underlying_device_busy);
1704 static int dm_lld_busy(struct request_queue *q)
1707 struct mapped_device *md = q->queuedata;
1708 struct dm_table *map = dm_get_live_table(md);
1710 if (!map || test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))
1713 r = dm_table_any_busy_target(map);
1720 static int dm_any_congested(void *congested_data, int bdi_bits)
1723 struct mapped_device *md = congested_data;
1724 struct dm_table *map;
1726 if (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
1727 map = dm_get_live_table(md);
1730 * Request-based dm cares about only own queue for
1731 * the query about congestion status of request_queue
1733 if (dm_request_based(md))
1734 r = md->queue->backing_dev_info.state &
1737 r = dm_table_any_congested(map, bdi_bits);
1746 /*-----------------------------------------------------------------
1747 * An IDR is used to keep track of allocated minor numbers.
1748 *---------------------------------------------------------------*/
1749 static void free_minor(int minor)
1751 spin_lock(&_minor_lock);
1752 idr_remove(&_minor_idr, minor);
1753 spin_unlock(&_minor_lock);
1757 * See if the device with a specific minor # is free.
1759 static int specific_minor(int minor)
1763 if (minor >= (1 << MINORBITS))
1766 r = idr_pre_get(&_minor_idr, GFP_KERNEL);
1770 spin_lock(&_minor_lock);
1772 if (idr_find(&_minor_idr, minor)) {
1777 r = idr_get_new_above(&_minor_idr, MINOR_ALLOCED, minor, &m);
1782 idr_remove(&_minor_idr, m);
1788 spin_unlock(&_minor_lock);
1792 static int next_free_minor(int *minor)
1796 r = idr_pre_get(&_minor_idr, GFP_KERNEL);
1800 spin_lock(&_minor_lock);
1802 r = idr_get_new(&_minor_idr, MINOR_ALLOCED, &m);
1806 if (m >= (1 << MINORBITS)) {
1807 idr_remove(&_minor_idr, m);
1815 spin_unlock(&_minor_lock);
1819 static const struct block_device_operations dm_blk_dops;
1821 static void dm_wq_work(struct work_struct *work);
1823 static void dm_init_md_queue(struct mapped_device *md)
1826 * Request-based dm devices cannot be stacked on top of bio-based dm
1827 * devices. The type of this dm device has not been decided yet.
1828 * The type is decided at the first table loading time.
1829 * To prevent problematic device stacking, clear the queue flag
1830 * for request stacking support until then.
1832 * This queue is new, so no concurrency on the queue_flags.
1834 queue_flag_clear_unlocked(QUEUE_FLAG_STACKABLE, md->queue);
1836 md->queue->queuedata = md;
1837 md->queue->backing_dev_info.congested_fn = dm_any_congested;
1838 md->queue->backing_dev_info.congested_data = md;
1839 blk_queue_make_request(md->queue, dm_request);
1840 blk_queue_bounce_limit(md->queue, BLK_BOUNCE_ANY);
1841 blk_queue_merge_bvec(md->queue, dm_merge_bvec);
1845 * Allocate and initialise a blank device with a given minor.
1847 static struct mapped_device *alloc_dev(int minor)
1850 struct mapped_device *md = kzalloc(sizeof(*md), GFP_KERNEL);
1854 DMWARN("unable to allocate device, out of memory.");
1858 if (!try_module_get(THIS_MODULE))
1859 goto bad_module_get;
1861 /* get a minor number for the dev */
1862 if (minor == DM_ANY_MINOR)
1863 r = next_free_minor(&minor);
1865 r = specific_minor(minor);
1869 md->type = DM_TYPE_NONE;
1870 init_rwsem(&md->io_lock);
1871 mutex_init(&md->suspend_lock);
1872 mutex_init(&md->type_lock);
1873 spin_lock_init(&md->deferred_lock);
1874 rwlock_init(&md->map_lock);
1875 atomic_set(&md->holders, 1);
1876 atomic_set(&md->open_count, 0);
1877 atomic_set(&md->event_nr, 0);
1878 atomic_set(&md->uevent_seq, 0);
1879 INIT_LIST_HEAD(&md->uevent_list);
1880 spin_lock_init(&md->uevent_lock);
1882 md->queue = blk_alloc_queue(GFP_KERNEL);
1886 dm_init_md_queue(md);
1888 md->disk = alloc_disk(1);
1892 atomic_set(&md->pending[0], 0);
1893 atomic_set(&md->pending[1], 0);
1894 init_waitqueue_head(&md->wait);
1895 INIT_WORK(&md->work, dm_wq_work);
1896 init_waitqueue_head(&md->eventq);
1897 init_completion(&md->kobj_completion);
1899 md->disk->major = _major;
1900 md->disk->first_minor = minor;
1901 md->disk->fops = &dm_blk_dops;
1902 md->disk->queue = md->queue;
1903 md->disk->private_data = md;
1904 sprintf(md->disk->disk_name, "dm-%d", minor);
1906 format_dev_t(md->name, MKDEV(_major, minor));
1908 md->wq = alloc_workqueue("kdmflush",
1909 WQ_NON_REENTRANT | WQ_MEM_RECLAIM, 0);
1913 md->bdev = bdget_disk(md->disk, 0);
1917 bio_init(&md->flush_bio);
1918 md->flush_bio.bi_bdev = md->bdev;
1919 md->flush_bio.bi_rw = WRITE_FLUSH;
1921 /* Populate the mapping, nobody knows we exist yet */
1922 spin_lock(&_minor_lock);
1923 old_md = idr_replace(&_minor_idr, md, minor);
1924 spin_unlock(&_minor_lock);
1926 BUG_ON(old_md != MINOR_ALLOCED);
1931 destroy_workqueue(md->wq);
1933 del_gendisk(md->disk);
1936 blk_cleanup_queue(md->queue);
1940 module_put(THIS_MODULE);
1946 static void unlock_fs(struct mapped_device *md);
1948 static void free_dev(struct mapped_device *md)
1950 int minor = MINOR(disk_devt(md->disk));
1954 destroy_workqueue(md->wq);
1956 mempool_destroy(md->tio_pool);
1958 mempool_destroy(md->io_pool);
1960 bioset_free(md->bs);
1961 blk_integrity_unregister(md->disk);
1962 del_gendisk(md->disk);
1965 spin_lock(&_minor_lock);
1966 md->disk->private_data = NULL;
1967 spin_unlock(&_minor_lock);
1970 blk_cleanup_queue(md->queue);
1971 module_put(THIS_MODULE);
1975 static void __bind_mempools(struct mapped_device *md, struct dm_table *t)
1977 struct dm_md_mempools *p;
1979 if (md->io_pool && md->tio_pool && md->bs)
1980 /* the md already has necessary mempools */
1983 p = dm_table_get_md_mempools(t);
1984 BUG_ON(!p || md->io_pool || md->tio_pool || md->bs);
1986 md->io_pool = p->io_pool;
1988 md->tio_pool = p->tio_pool;
1994 /* mempool bind completed, now no need any mempools in the table */
1995 dm_table_free_md_mempools(t);
1999 * Bind a table to the device.
2001 static void event_callback(void *context)
2003 unsigned long flags;
2005 struct mapped_device *md = (struct mapped_device *) context;
2007 spin_lock_irqsave(&md->uevent_lock, flags);
2008 list_splice_init(&md->uevent_list, &uevents);
2009 spin_unlock_irqrestore(&md->uevent_lock, flags);
2011 dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj);
2013 atomic_inc(&md->event_nr);
2014 wake_up(&md->eventq);
2018 * Protected by md->suspend_lock obtained by dm_swap_table().
2020 static void __set_size(struct mapped_device *md, sector_t size)
2022 set_capacity(md->disk, size);
2024 i_size_write(md->bdev->bd_inode, (loff_t)size << SECTOR_SHIFT);
2028 * Return 1 if the queue has a compulsory merge_bvec_fn function.
2030 * If this function returns 0, then the device is either a non-dm
2031 * device without a merge_bvec_fn, or it is a dm device that is
2032 * able to split any bios it receives that are too big.
2034 int dm_queue_merge_is_compulsory(struct request_queue *q)
2036 struct mapped_device *dev_md;
2038 if (!q->merge_bvec_fn)
2041 if (q->make_request_fn == dm_request) {
2042 dev_md = q->queuedata;
2043 if (test_bit(DMF_MERGE_IS_OPTIONAL, &dev_md->flags))
2050 static int dm_device_merge_is_compulsory(struct dm_target *ti,
2051 struct dm_dev *dev, sector_t start,
2052 sector_t len, void *data)
2054 struct block_device *bdev = dev->bdev;
2055 struct request_queue *q = bdev_get_queue(bdev);
2057 return dm_queue_merge_is_compulsory(q);
2061 * Return 1 if it is acceptable to ignore merge_bvec_fn based
2062 * on the properties of the underlying devices.
2064 static int dm_table_merge_is_optional(struct dm_table *table)
2067 struct dm_target *ti;
2069 while (i < dm_table_get_num_targets(table)) {
2070 ti = dm_table_get_target(table, i++);
2072 if (ti->type->iterate_devices &&
2073 ti->type->iterate_devices(ti, dm_device_merge_is_compulsory, NULL))
2081 * Returns old map, which caller must destroy.
2083 static struct dm_table *__bind(struct mapped_device *md, struct dm_table *t,
2084 struct queue_limits *limits)
2086 struct dm_table *old_map;
2087 struct request_queue *q = md->queue;
2089 unsigned long flags;
2090 int merge_is_optional;
2092 size = dm_table_get_size(t);
2095 * Wipe any geometry if the size of the table changed.
2097 if (size != get_capacity(md->disk))
2098 memset(&md->geometry, 0, sizeof(md->geometry));
2100 __set_size(md, size);
2102 dm_table_event_callback(t, event_callback, md);
2105 * The queue hasn't been stopped yet, if the old table type wasn't
2106 * for request-based during suspension. So stop it to prevent
2107 * I/O mapping before resume.
2108 * This must be done before setting the queue restrictions,
2109 * because request-based dm may be run just after the setting.
2111 if (dm_table_request_based(t) && !blk_queue_stopped(q))
2114 __bind_mempools(md, t);
2116 merge_is_optional = dm_table_merge_is_optional(t);
2118 write_lock_irqsave(&md->map_lock, flags);
2121 md->immutable_target_type = dm_table_get_immutable_target_type(t);
2123 dm_table_set_restrictions(t, q, limits);
2124 if (merge_is_optional)
2125 set_bit(DMF_MERGE_IS_OPTIONAL, &md->flags);
2127 clear_bit(DMF_MERGE_IS_OPTIONAL, &md->flags);
2128 write_unlock_irqrestore(&md->map_lock, flags);
2134 * Returns unbound table for the caller to free.
2136 static struct dm_table *__unbind(struct mapped_device *md)
2138 struct dm_table *map = md->map;
2139 unsigned long flags;
2144 dm_table_event_callback(map, NULL, NULL);
2145 write_lock_irqsave(&md->map_lock, flags);
2147 write_unlock_irqrestore(&md->map_lock, flags);
2153 * Constructor for a new device.
2155 int dm_create(int minor, struct mapped_device **result)
2157 struct mapped_device *md;
2159 md = alloc_dev(minor);
2170 * Functions to manage md->type.
2171 * All are required to hold md->type_lock.
2173 void dm_lock_md_type(struct mapped_device *md)
2175 mutex_lock(&md->type_lock);
2178 void dm_unlock_md_type(struct mapped_device *md)
2180 mutex_unlock(&md->type_lock);
2183 void dm_set_md_type(struct mapped_device *md, unsigned type)
2188 unsigned dm_get_md_type(struct mapped_device *md)
2193 struct target_type *dm_get_immutable_target_type(struct mapped_device *md)
2195 return md->immutable_target_type;
2199 * Fully initialize a request-based queue (->elevator, ->request_fn, etc).
2201 static int dm_init_request_based_queue(struct mapped_device *md)
2203 struct request_queue *q = NULL;
2205 if (md->queue->elevator)
2208 /* Fully initialize the queue */
2209 q = blk_init_allocated_queue(md->queue, dm_request_fn, NULL);
2214 dm_init_md_queue(md);
2215 blk_queue_softirq_done(md->queue, dm_softirq_done);
2216 blk_queue_prep_rq(md->queue, dm_prep_fn);
2217 blk_queue_lld_busy(md->queue, dm_lld_busy);
2219 elv_register_queue(md->queue);
2225 * Setup the DM device's queue based on md's type
2227 int dm_setup_md_queue(struct mapped_device *md)
2229 if ((dm_get_md_type(md) == DM_TYPE_REQUEST_BASED) &&
2230 !dm_init_request_based_queue(md)) {
2231 DMWARN("Cannot initialize queue for request-based mapped device");
2238 static struct mapped_device *dm_find_md(dev_t dev)
2240 struct mapped_device *md;
2241 unsigned minor = MINOR(dev);
2243 if (MAJOR(dev) != _major || minor >= (1 << MINORBITS))
2246 spin_lock(&_minor_lock);
2248 md = idr_find(&_minor_idr, minor);
2249 if (md && (md == MINOR_ALLOCED ||
2250 (MINOR(disk_devt(dm_disk(md))) != minor) ||
2251 dm_deleting_md(md) ||
2252 test_bit(DMF_FREEING, &md->flags))) {
2258 spin_unlock(&_minor_lock);
2263 struct mapped_device *dm_get_md(dev_t dev)
2265 struct mapped_device *md = dm_find_md(dev);
2272 EXPORT_SYMBOL_GPL(dm_get_md);
2274 void *dm_get_mdptr(struct mapped_device *md)
2276 return md->interface_ptr;
2279 void dm_set_mdptr(struct mapped_device *md, void *ptr)
2281 md->interface_ptr = ptr;
2284 void dm_get(struct mapped_device *md)
2286 atomic_inc(&md->holders);
2287 BUG_ON(test_bit(DMF_FREEING, &md->flags));
2290 const char *dm_device_name(struct mapped_device *md)
2294 EXPORT_SYMBOL_GPL(dm_device_name);
2296 static void __dm_destroy(struct mapped_device *md, bool wait)
2298 struct dm_table *map;
2302 spin_lock(&_minor_lock);
2303 map = dm_get_live_table(md);
2304 idr_replace(&_minor_idr, MINOR_ALLOCED, MINOR(disk_devt(dm_disk(md))));
2305 set_bit(DMF_FREEING, &md->flags);
2306 spin_unlock(&_minor_lock);
2308 if (!dm_suspended_md(md)) {
2309 dm_table_presuspend_targets(map);
2310 dm_table_postsuspend_targets(map);
2314 * Rare, but there may be I/O requests still going to complete,
2315 * for example. Wait for all references to disappear.
2316 * No one should increment the reference count of the mapped_device,
2317 * after the mapped_device state becomes DMF_FREEING.
2320 while (atomic_read(&md->holders))
2322 else if (atomic_read(&md->holders))
2323 DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)",
2324 dm_device_name(md), atomic_read(&md->holders));
2328 dm_table_destroy(__unbind(md));
2332 void dm_destroy(struct mapped_device *md)
2334 __dm_destroy(md, true);
2337 void dm_destroy_immediate(struct mapped_device *md)
2339 __dm_destroy(md, false);
2342 void dm_put(struct mapped_device *md)
2344 atomic_dec(&md->holders);
2346 EXPORT_SYMBOL_GPL(dm_put);
2348 static int dm_wait_for_completion(struct mapped_device *md, int interruptible)
2351 DECLARE_WAITQUEUE(wait, current);
2353 add_wait_queue(&md->wait, &wait);
2356 set_current_state(interruptible);
2358 if (!md_in_flight(md))
2361 if (interruptible == TASK_INTERRUPTIBLE &&
2362 signal_pending(current)) {
2369 set_current_state(TASK_RUNNING);
2371 remove_wait_queue(&md->wait, &wait);
2377 * Process the deferred bios
2379 static void dm_wq_work(struct work_struct *work)
2381 struct mapped_device *md = container_of(work, struct mapped_device,
2385 down_read(&md->io_lock);
2387 while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
2388 spin_lock_irq(&md->deferred_lock);
2389 c = bio_list_pop(&md->deferred);
2390 spin_unlock_irq(&md->deferred_lock);
2395 up_read(&md->io_lock);
2397 if (dm_request_based(md))
2398 generic_make_request(c);
2400 __split_and_process_bio(md, c);
2402 down_read(&md->io_lock);
2405 up_read(&md->io_lock);
2408 static void dm_queue_flush(struct mapped_device *md)
2410 clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2411 smp_mb__after_clear_bit();
2412 queue_work(md->wq, &md->work);
2416 * Swap in a new table, returning the old one for the caller to destroy.
2418 struct dm_table *dm_swap_table(struct mapped_device *md, struct dm_table *table)
2420 struct dm_table *map = ERR_PTR(-EINVAL);
2421 struct queue_limits limits;
2424 mutex_lock(&md->suspend_lock);
2426 /* device must be suspended */
2427 if (!dm_suspended_md(md))
2430 r = dm_calculate_queue_limits(table, &limits);
2436 map = __bind(md, table, &limits);
2439 mutex_unlock(&md->suspend_lock);
2444 * Functions to lock and unlock any filesystem running on the
2447 static int lock_fs(struct mapped_device *md)
2451 WARN_ON(md->frozen_sb);
2453 md->frozen_sb = freeze_bdev(md->bdev);
2454 if (IS_ERR(md->frozen_sb)) {
2455 r = PTR_ERR(md->frozen_sb);
2456 md->frozen_sb = NULL;
2460 set_bit(DMF_FROZEN, &md->flags);
2465 static void unlock_fs(struct mapped_device *md)
2467 if (!test_bit(DMF_FROZEN, &md->flags))
2470 thaw_bdev(md->bdev, md->frozen_sb);
2471 md->frozen_sb = NULL;
2472 clear_bit(DMF_FROZEN, &md->flags);
2476 * We need to be able to change a mapping table under a mounted
2477 * filesystem. For example we might want to move some data in
2478 * the background. Before the table can be swapped with
2479 * dm_bind_table, dm_suspend must be called to flush any in
2480 * flight bios and ensure that any further io gets deferred.
2483 * Suspend mechanism in request-based dm.
2485 * 1. Flush all I/Os by lock_fs() if needed.
2486 * 2. Stop dispatching any I/O by stopping the request_queue.
2487 * 3. Wait for all in-flight I/Os to be completed or requeued.
2489 * To abort suspend, start the request_queue.
2491 int dm_suspend(struct mapped_device *md, unsigned suspend_flags)
2493 struct dm_table *map = NULL;
2495 int do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG ? 1 : 0;
2496 int noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG ? 1 : 0;
2498 mutex_lock(&md->suspend_lock);
2500 if (dm_suspended_md(md)) {
2505 map = dm_get_live_table(md);
2508 * DMF_NOFLUSH_SUSPENDING must be set before presuspend.
2509 * This flag is cleared before dm_suspend returns.
2512 set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2514 /* This does not get reverted if there's an error later. */
2515 dm_table_presuspend_targets(map);
2518 * Flush I/O to the device.
2519 * Any I/O submitted after lock_fs() may not be flushed.
2520 * noflush takes precedence over do_lockfs.
2521 * (lock_fs() flushes I/Os and waits for them to complete.)
2523 if (!noflush && do_lockfs) {
2530 * Here we must make sure that no processes are submitting requests
2531 * to target drivers i.e. no one may be executing
2532 * __split_and_process_bio. This is called from dm_request and
2535 * To get all processes out of __split_and_process_bio in dm_request,
2536 * we take the write lock. To prevent any process from reentering
2537 * __split_and_process_bio from dm_request and quiesce the thread
2538 * (dm_wq_work), we set BMF_BLOCK_IO_FOR_SUSPEND and call
2539 * flush_workqueue(md->wq).
2541 down_write(&md->io_lock);
2542 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2543 up_write(&md->io_lock);
2546 * Stop md->queue before flushing md->wq in case request-based
2547 * dm defers requests to md->wq from md->queue.
2549 if (dm_request_based(md))
2550 stop_queue(md->queue);
2552 flush_workqueue(md->wq);
2555 * At this point no more requests are entering target request routines.
2556 * We call dm_wait_for_completion to wait for all existing requests
2559 r = dm_wait_for_completion(md, TASK_INTERRUPTIBLE);
2561 down_write(&md->io_lock);
2563 clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2564 up_write(&md->io_lock);
2566 /* were we interrupted ? */
2570 if (dm_request_based(md))
2571 start_queue(md->queue);
2574 goto out; /* pushback list is already flushed, so skip flush */
2578 * If dm_wait_for_completion returned 0, the device is completely
2579 * quiescent now. There is no request-processing activity. All new
2580 * requests are being added to md->deferred list.
2583 set_bit(DMF_SUSPENDED, &md->flags);
2585 dm_table_postsuspend_targets(map);
2591 mutex_unlock(&md->suspend_lock);
2595 int dm_resume(struct mapped_device *md)
2598 struct dm_table *map = NULL;
2600 mutex_lock(&md->suspend_lock);
2601 if (!dm_suspended_md(md))
2604 map = dm_get_live_table(md);
2605 if (!map || !dm_table_get_size(map))
2608 r = dm_table_resume_targets(map);
2615 * Flushing deferred I/Os must be done after targets are resumed
2616 * so that mapping of targets can work correctly.
2617 * Request-based dm is queueing the deferred I/Os in its request_queue.
2619 if (dm_request_based(md))
2620 start_queue(md->queue);
2624 clear_bit(DMF_SUSPENDED, &md->flags);
2629 mutex_unlock(&md->suspend_lock);
2634 /*-----------------------------------------------------------------
2635 * Event notification.
2636 *---------------------------------------------------------------*/
2637 int dm_kobject_uevent(struct mapped_device *md, enum kobject_action action,
2640 char udev_cookie[DM_COOKIE_LENGTH];
2641 char *envp[] = { udev_cookie, NULL };
2644 return kobject_uevent(&disk_to_dev(md->disk)->kobj, action);
2646 snprintf(udev_cookie, DM_COOKIE_LENGTH, "%s=%u",
2647 DM_COOKIE_ENV_VAR_NAME, cookie);
2648 return kobject_uevent_env(&disk_to_dev(md->disk)->kobj,
2653 uint32_t dm_next_uevent_seq(struct mapped_device *md)
2655 return atomic_add_return(1, &md->uevent_seq);
2658 uint32_t dm_get_event_nr(struct mapped_device *md)
2660 return atomic_read(&md->event_nr);
2663 int dm_wait_event(struct mapped_device *md, int event_nr)
2665 return wait_event_interruptible(md->eventq,
2666 (event_nr != atomic_read(&md->event_nr)));
2669 void dm_uevent_add(struct mapped_device *md, struct list_head *elist)
2671 unsigned long flags;
2673 spin_lock_irqsave(&md->uevent_lock, flags);
2674 list_add(elist, &md->uevent_list);
2675 spin_unlock_irqrestore(&md->uevent_lock, flags);
2679 * The gendisk is only valid as long as you have a reference
2682 struct gendisk *dm_disk(struct mapped_device *md)
2687 struct kobject *dm_kobject(struct mapped_device *md)
2693 * struct mapped_device should not be exported outside of dm.c
2694 * so use this check to verify that kobj is part of md structure
2696 struct mapped_device *dm_get_from_kobject(struct kobject *kobj)
2698 struct mapped_device *md;
2700 md = container_of(kobj, struct mapped_device, kobj);
2701 if (&md->kobj != kobj)
2704 if (test_bit(DMF_FREEING, &md->flags) ||
2712 struct completion *dm_get_completion_from_kobject(struct kobject *kobj)
2714 struct mapped_device *md = container_of(kobj, struct mapped_device, kobj);
2716 return &md->kobj_completion;
2719 int dm_suspended_md(struct mapped_device *md)
2721 return test_bit(DMF_SUSPENDED, &md->flags);
2724 int dm_suspended(struct dm_target *ti)
2726 return dm_suspended_md(dm_table_get_md(ti->table));
2728 EXPORT_SYMBOL_GPL(dm_suspended);
2730 int dm_noflush_suspending(struct dm_target *ti)
2732 return __noflush_suspending(dm_table_get_md(ti->table));
2734 EXPORT_SYMBOL_GPL(dm_noflush_suspending);
2736 struct dm_md_mempools *dm_alloc_md_mempools(unsigned type, unsigned integrity)
2738 struct dm_md_mempools *pools = kmalloc(sizeof(*pools), GFP_KERNEL);
2739 unsigned int pool_size = (type == DM_TYPE_BIO_BASED) ? 16 : MIN_IOS;
2744 pools->io_pool = (type == DM_TYPE_BIO_BASED) ?
2745 mempool_create_slab_pool(MIN_IOS, _io_cache) :
2746 mempool_create_slab_pool(MIN_IOS, _rq_bio_info_cache);
2747 if (!pools->io_pool)
2748 goto free_pools_and_out;
2750 pools->tio_pool = (type == DM_TYPE_BIO_BASED) ?
2751 mempool_create_slab_pool(MIN_IOS, _tio_cache) :
2752 mempool_create_slab_pool(MIN_IOS, _rq_tio_cache);
2753 if (!pools->tio_pool)
2754 goto free_io_pool_and_out;
2756 pools->bs = bioset_create(pool_size, 0);
2758 goto free_tio_pool_and_out;
2760 if (integrity && bioset_integrity_create(pools->bs, pool_size))
2761 goto free_bioset_and_out;
2765 free_bioset_and_out:
2766 bioset_free(pools->bs);
2768 free_tio_pool_and_out:
2769 mempool_destroy(pools->tio_pool);
2771 free_io_pool_and_out:
2772 mempool_destroy(pools->io_pool);
2780 void dm_free_md_mempools(struct dm_md_mempools *pools)
2786 mempool_destroy(pools->io_pool);
2788 if (pools->tio_pool)
2789 mempool_destroy(pools->tio_pool);
2792 bioset_free(pools->bs);
2797 static const struct block_device_operations dm_blk_dops = {
2798 .open = dm_blk_open,
2799 .release = dm_blk_close,
2800 .ioctl = dm_blk_ioctl,
2801 .getgeo = dm_blk_getgeo,
2802 .owner = THIS_MODULE
2805 EXPORT_SYMBOL(dm_get_mapinfo);
2810 module_init(dm_init);
2811 module_exit(dm_exit);
2813 module_param(major, uint, 0);
2814 MODULE_PARM_DESC(major, "The major number of the device mapper");
2815 MODULE_DESCRIPTION(DM_NAME " driver");
2816 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
2817 MODULE_LICENSE("GPL");