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/mempool.h>
18 #include <linux/slab.h>
19 #include <linux/idr.h>
20 #include <linux/hdreg.h>
21 #include <linux/delay.h>
23 #include <trace/events/block.h>
25 #define DM_MSG_PREFIX "core"
29 * ratelimit state to be used in DMXXX_LIMIT().
31 DEFINE_RATELIMIT_STATE(dm_ratelimit_state,
32 DEFAULT_RATELIMIT_INTERVAL,
33 DEFAULT_RATELIMIT_BURST);
34 EXPORT_SYMBOL(dm_ratelimit_state);
38 * Cookies are numeric values sent with CHANGE and REMOVE
39 * uevents while resuming, removing or renaming the device.
41 #define DM_COOKIE_ENV_VAR_NAME "DM_COOKIE"
42 #define DM_COOKIE_LENGTH 24
44 static const char *_name = DM_NAME;
46 static unsigned int major = 0;
47 static unsigned int _major = 0;
49 static DEFINE_IDR(_minor_idr);
51 static DEFINE_SPINLOCK(_minor_lock);
53 static void do_deferred_remove(struct work_struct *w);
55 static DECLARE_WORK(deferred_remove_work, do_deferred_remove);
57 static struct workqueue_struct *deferred_remove_workqueue;
61 * One of these is allocated per bio.
64 struct mapped_device *md;
68 unsigned long start_time;
69 spinlock_t endio_lock;
70 struct dm_stats_aux stats_aux;
74 * For request-based dm.
75 * One of these is allocated per request.
77 struct dm_rq_target_io {
78 struct mapped_device *md;
80 struct request *orig, clone;
86 * For request-based dm - the bio clones we allocate are embedded in these
89 * We allocate these with bio_alloc_bioset, using the front_pad parameter when
90 * the bioset is created - this means the bio has to come at the end of the
93 struct dm_rq_clone_bio_info {
95 struct dm_rq_target_io *tio;
99 union map_info *dm_get_rq_mapinfo(struct request *rq)
101 if (rq && rq->end_io_data)
102 return &((struct dm_rq_target_io *)rq->end_io_data)->info;
105 EXPORT_SYMBOL_GPL(dm_get_rq_mapinfo);
107 #define MINOR_ALLOCED ((void *)-1)
110 * Bits for the md->flags field.
112 #define DMF_BLOCK_IO_FOR_SUSPEND 0
113 #define DMF_SUSPENDED 1
115 #define DMF_FREEING 3
116 #define DMF_DELETING 4
117 #define DMF_NOFLUSH_SUSPENDING 5
118 #define DMF_MERGE_IS_OPTIONAL 6
119 #define DMF_DEFERRED_REMOVE 7
122 * A dummy definition to make RCU happy.
123 * struct dm_table should never be dereferenced in this file.
130 * Work processed by per-device workqueue.
132 struct mapped_device {
133 struct srcu_struct io_barrier;
134 struct mutex suspend_lock;
139 * The current mapping.
140 * Use dm_get_live_table{_fast} or take suspend_lock for
143 struct dm_table *map;
147 struct request_queue *queue;
149 /* Protect queue and type against concurrent access. */
150 struct mutex type_lock;
152 struct target_type *immutable_target_type;
154 struct gendisk *disk;
160 * A list of ios that arrived while we were suspended.
163 wait_queue_head_t wait;
164 struct work_struct work;
165 struct bio_list deferred;
166 spinlock_t deferred_lock;
169 * Processing queue (flush)
171 struct workqueue_struct *wq;
174 * io objects are allocated from here.
184 wait_queue_head_t eventq;
186 struct list_head uevent_list;
187 spinlock_t uevent_lock; /* Protect access to uevent_list */
190 * freeze/thaw support require holding onto a super block
192 struct super_block *frozen_sb;
193 struct block_device *bdev;
195 /* forced geometry settings */
196 struct hd_geometry geometry;
198 /* kobject and completion */
199 struct dm_kobject_holder kobj_holder;
201 /* zero-length flush that will be cloned and submitted to targets */
202 struct bio flush_bio;
204 struct dm_stats stats;
208 * For mempools pre-allocation at the table loading time.
210 struct dm_md_mempools {
215 #define RESERVED_BIO_BASED_IOS 16
216 #define RESERVED_REQUEST_BASED_IOS 256
217 #define RESERVED_MAX_IOS 1024
218 static struct kmem_cache *_io_cache;
219 static struct kmem_cache *_rq_tio_cache;
222 * Bio-based DM's mempools' reserved IOs set by the user.
224 static unsigned reserved_bio_based_ios = RESERVED_BIO_BASED_IOS;
227 * Request-based DM's mempools' reserved IOs set by the user.
229 static unsigned reserved_rq_based_ios = RESERVED_REQUEST_BASED_IOS;
231 static unsigned __dm_get_reserved_ios(unsigned *reserved_ios,
232 unsigned def, unsigned max)
234 unsigned ios = ACCESS_ONCE(*reserved_ios);
235 unsigned modified_ios = 0;
243 (void)cmpxchg(reserved_ios, ios, modified_ios);
250 unsigned dm_get_reserved_bio_based_ios(void)
252 return __dm_get_reserved_ios(&reserved_bio_based_ios,
253 RESERVED_BIO_BASED_IOS, RESERVED_MAX_IOS);
255 EXPORT_SYMBOL_GPL(dm_get_reserved_bio_based_ios);
257 unsigned dm_get_reserved_rq_based_ios(void)
259 return __dm_get_reserved_ios(&reserved_rq_based_ios,
260 RESERVED_REQUEST_BASED_IOS, RESERVED_MAX_IOS);
262 EXPORT_SYMBOL_GPL(dm_get_reserved_rq_based_ios);
264 static int __init local_init(void)
268 /* allocate a slab for the dm_ios */
269 _io_cache = KMEM_CACHE(dm_io, 0);
273 _rq_tio_cache = KMEM_CACHE(dm_rq_target_io, 0);
275 goto out_free_io_cache;
277 r = dm_uevent_init();
279 goto out_free_rq_tio_cache;
281 deferred_remove_workqueue = alloc_workqueue("kdmremove", WQ_UNBOUND, 1);
282 if (!deferred_remove_workqueue) {
284 goto out_uevent_exit;
288 r = register_blkdev(_major, _name);
290 goto out_free_workqueue;
298 destroy_workqueue(deferred_remove_workqueue);
301 out_free_rq_tio_cache:
302 kmem_cache_destroy(_rq_tio_cache);
304 kmem_cache_destroy(_io_cache);
309 static void local_exit(void)
311 flush_scheduled_work();
312 destroy_workqueue(deferred_remove_workqueue);
314 kmem_cache_destroy(_rq_tio_cache);
315 kmem_cache_destroy(_io_cache);
316 unregister_blkdev(_major, _name);
321 DMINFO("cleaned up");
324 static int (*_inits[])(void) __initdata = {
335 static void (*_exits[])(void) = {
346 static int __init dm_init(void)
348 const int count = ARRAY_SIZE(_inits);
352 for (i = 0; i < count; i++) {
367 static void __exit dm_exit(void)
369 int i = ARRAY_SIZE(_exits);
375 * Should be empty by this point.
377 idr_destroy(&_minor_idr);
381 * Block device functions
383 int dm_deleting_md(struct mapped_device *md)
385 return test_bit(DMF_DELETING, &md->flags);
388 static int dm_blk_open(struct block_device *bdev, fmode_t mode)
390 struct mapped_device *md;
392 spin_lock(&_minor_lock);
394 md = bdev->bd_disk->private_data;
398 if (test_bit(DMF_FREEING, &md->flags) ||
399 dm_deleting_md(md)) {
405 atomic_inc(&md->open_count);
408 spin_unlock(&_minor_lock);
410 return md ? 0 : -ENXIO;
413 static void dm_blk_close(struct gendisk *disk, fmode_t mode)
415 struct mapped_device *md = disk->private_data;
417 spin_lock(&_minor_lock);
419 if (atomic_dec_and_test(&md->open_count) &&
420 (test_bit(DMF_DEFERRED_REMOVE, &md->flags)))
421 queue_work(deferred_remove_workqueue, &deferred_remove_work);
425 spin_unlock(&_minor_lock);
428 int dm_open_count(struct mapped_device *md)
430 return atomic_read(&md->open_count);
434 * Guarantees nothing is using the device before it's deleted.
436 int dm_lock_for_deletion(struct mapped_device *md, bool mark_deferred, bool only_deferred)
440 spin_lock(&_minor_lock);
442 if (dm_open_count(md)) {
445 set_bit(DMF_DEFERRED_REMOVE, &md->flags);
446 } else if (only_deferred && !test_bit(DMF_DEFERRED_REMOVE, &md->flags))
449 set_bit(DMF_DELETING, &md->flags);
451 spin_unlock(&_minor_lock);
456 int dm_cancel_deferred_remove(struct mapped_device *md)
460 spin_lock(&_minor_lock);
462 if (test_bit(DMF_DELETING, &md->flags))
465 clear_bit(DMF_DEFERRED_REMOVE, &md->flags);
467 spin_unlock(&_minor_lock);
472 static void do_deferred_remove(struct work_struct *w)
474 dm_deferred_remove();
477 sector_t dm_get_size(struct mapped_device *md)
479 return get_capacity(md->disk);
482 struct request_queue *dm_get_md_queue(struct mapped_device *md)
487 struct dm_stats *dm_get_stats(struct mapped_device *md)
492 static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo)
494 struct mapped_device *md = bdev->bd_disk->private_data;
496 return dm_get_geometry(md, geo);
499 static int dm_blk_ioctl(struct block_device *bdev, fmode_t mode,
500 unsigned int cmd, unsigned long arg)
502 struct mapped_device *md = bdev->bd_disk->private_data;
504 struct dm_table *map;
505 struct dm_target *tgt;
509 map = dm_get_live_table(md, &srcu_idx);
511 if (!map || !dm_table_get_size(map))
514 /* We only support devices that have a single target */
515 if (dm_table_get_num_targets(map) != 1)
518 tgt = dm_table_get_target(map, 0);
520 if (dm_suspended_md(md)) {
525 if (tgt->type->ioctl)
526 r = tgt->type->ioctl(tgt, cmd, arg);
529 dm_put_live_table(md, srcu_idx);
531 if (r == -ENOTCONN) {
539 static struct dm_io *alloc_io(struct mapped_device *md)
541 return mempool_alloc(md->io_pool, GFP_NOIO);
544 static void free_io(struct mapped_device *md, struct dm_io *io)
546 mempool_free(io, md->io_pool);
549 static void free_tio(struct mapped_device *md, struct dm_target_io *tio)
551 bio_put(&tio->clone);
554 static struct dm_rq_target_io *alloc_rq_tio(struct mapped_device *md,
557 return mempool_alloc(md->io_pool, gfp_mask);
560 static void free_rq_tio(struct dm_rq_target_io *tio)
562 mempool_free(tio, tio->md->io_pool);
565 static int md_in_flight(struct mapped_device *md)
567 return atomic_read(&md->pending[READ]) +
568 atomic_read(&md->pending[WRITE]);
571 static void start_io_acct(struct dm_io *io)
573 struct mapped_device *md = io->md;
574 struct bio *bio = io->bio;
576 int rw = bio_data_dir(bio);
578 io->start_time = jiffies;
580 cpu = part_stat_lock();
581 part_round_stats(cpu, &dm_disk(md)->part0);
583 atomic_set(&dm_disk(md)->part0.in_flight[rw],
584 atomic_inc_return(&md->pending[rw]));
586 if (unlikely(dm_stats_used(&md->stats)))
587 dm_stats_account_io(&md->stats, bio->bi_rw, bio->bi_iter.bi_sector,
588 bio_sectors(bio), false, 0, &io->stats_aux);
591 static void end_io_acct(struct dm_io *io)
593 struct mapped_device *md = io->md;
594 struct bio *bio = io->bio;
595 unsigned long duration = jiffies - io->start_time;
597 int rw = bio_data_dir(bio);
599 cpu = part_stat_lock();
600 part_round_stats(cpu, &dm_disk(md)->part0);
601 part_stat_add(cpu, &dm_disk(md)->part0, ticks[rw], duration);
604 if (unlikely(dm_stats_used(&md->stats)))
605 dm_stats_account_io(&md->stats, bio->bi_rw, bio->bi_iter.bi_sector,
606 bio_sectors(bio), true, duration, &io->stats_aux);
609 * After this is decremented the bio must not be touched if it is
612 pending = atomic_dec_return(&md->pending[rw]);
613 atomic_set(&dm_disk(md)->part0.in_flight[rw], pending);
614 pending += atomic_read(&md->pending[rw^0x1]);
616 /* nudge anyone waiting on suspend queue */
622 * Add the bio to the list of deferred io.
624 static void queue_io(struct mapped_device *md, struct bio *bio)
628 spin_lock_irqsave(&md->deferred_lock, flags);
629 bio_list_add(&md->deferred, bio);
630 spin_unlock_irqrestore(&md->deferred_lock, flags);
631 queue_work(md->wq, &md->work);
635 * Everyone (including functions in this file), should use this
636 * function to access the md->map field, and make sure they call
637 * dm_put_live_table() when finished.
639 struct dm_table *dm_get_live_table(struct mapped_device *md, int *srcu_idx) __acquires(md->io_barrier)
641 *srcu_idx = srcu_read_lock(&md->io_barrier);
643 return srcu_dereference(md->map, &md->io_barrier);
646 void dm_put_live_table(struct mapped_device *md, int srcu_idx) __releases(md->io_barrier)
648 srcu_read_unlock(&md->io_barrier, srcu_idx);
651 void dm_sync_table(struct mapped_device *md)
653 synchronize_srcu(&md->io_barrier);
654 synchronize_rcu_expedited();
658 * A fast alternative to dm_get_live_table/dm_put_live_table.
659 * The caller must not block between these two functions.
661 static struct dm_table *dm_get_live_table_fast(struct mapped_device *md) __acquires(RCU)
664 return rcu_dereference(md->map);
667 static void dm_put_live_table_fast(struct mapped_device *md) __releases(RCU)
673 * Get the geometry associated with a dm device
675 int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo)
683 * Set the geometry of a device.
685 int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo)
687 sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors;
689 if (geo->start > sz) {
690 DMWARN("Start sector is beyond the geometry limits.");
699 /*-----------------------------------------------------------------
701 * A more elegant soln is in the works that uses the queue
702 * merge fn, unfortunately there are a couple of changes to
703 * the block layer that I want to make for this. So in the
704 * interests of getting something for people to use I give
705 * you this clearly demarcated crap.
706 *---------------------------------------------------------------*/
708 static int __noflush_suspending(struct mapped_device *md)
710 return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
714 * Decrements the number of outstanding ios that a bio has been
715 * cloned into, completing the original io if necc.
717 static void dec_pending(struct dm_io *io, int error)
722 struct mapped_device *md = io->md;
724 /* Push-back supersedes any I/O errors */
725 if (unlikely(error)) {
726 spin_lock_irqsave(&io->endio_lock, flags);
727 if (!(io->error > 0 && __noflush_suspending(md)))
729 spin_unlock_irqrestore(&io->endio_lock, flags);
732 if (atomic_dec_and_test(&io->io_count)) {
733 if (io->error == DM_ENDIO_REQUEUE) {
735 * Target requested pushing back the I/O.
737 spin_lock_irqsave(&md->deferred_lock, flags);
738 if (__noflush_suspending(md))
739 bio_list_add_head(&md->deferred, io->bio);
741 /* noflush suspend was interrupted. */
743 spin_unlock_irqrestore(&md->deferred_lock, flags);
746 io_error = io->error;
751 if (io_error == DM_ENDIO_REQUEUE)
754 if ((bio->bi_rw & REQ_FLUSH) && bio->bi_iter.bi_size) {
756 * Preflush done for flush with data, reissue
759 bio->bi_rw &= ~REQ_FLUSH;
762 /* done with normal IO or empty flush */
763 trace_block_bio_complete(md->queue, bio, io_error);
764 bio_endio(bio, io_error);
769 static void disable_write_same(struct mapped_device *md)
771 struct queue_limits *limits = dm_get_queue_limits(md);
773 /* device doesn't really support WRITE SAME, disable it */
774 limits->max_write_same_sectors = 0;
777 static void clone_endio(struct bio *bio, int error)
780 struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone);
781 struct dm_io *io = tio->io;
782 struct mapped_device *md = tio->io->md;
783 dm_endio_fn endio = tio->ti->type->end_io;
785 if (!bio_flagged(bio, BIO_UPTODATE) && !error)
789 r = endio(tio->ti, bio, error);
790 if (r < 0 || r == DM_ENDIO_REQUEUE)
792 * error and requeue request are handled
796 else if (r == DM_ENDIO_INCOMPLETE)
797 /* The target will handle the io */
800 DMWARN("unimplemented target endio return value: %d", r);
805 if (unlikely(r == -EREMOTEIO && (bio->bi_rw & REQ_WRITE_SAME) &&
806 !bdev_get_queue(bio->bi_bdev)->limits.max_write_same_sectors))
807 disable_write_same(md);
810 dec_pending(io, error);
814 * Partial completion handling for request-based dm
816 static void end_clone_bio(struct bio *clone, int error)
818 struct dm_rq_clone_bio_info *info =
819 container_of(clone, struct dm_rq_clone_bio_info, clone);
820 struct dm_rq_target_io *tio = info->tio;
821 struct bio *bio = info->orig;
822 unsigned int nr_bytes = info->orig->bi_iter.bi_size;
828 * An error has already been detected on the request.
829 * Once error occurred, just let clone->end_io() handle
835 * Don't notice the error to the upper layer yet.
836 * The error handling decision is made by the target driver,
837 * when the request is completed.
844 * I/O for the bio successfully completed.
845 * Notice the data completion to the upper layer.
849 * bios are processed from the head of the list.
850 * So the completing bio should always be rq->bio.
851 * If it's not, something wrong is happening.
853 if (tio->orig->bio != bio)
854 DMERR("bio completion is going in the middle of the request");
857 * Update the original request.
858 * Do not use blk_end_request() here, because it may complete
859 * the original request before the clone, and break the ordering.
861 blk_update_request(tio->orig, 0, nr_bytes);
865 * Don't touch any member of the md after calling this function because
866 * the md may be freed in dm_put() at the end of this function.
867 * Or do dm_get() before calling this function and dm_put() later.
869 static void rq_completed(struct mapped_device *md, int rw, int run_queue)
871 atomic_dec(&md->pending[rw]);
873 /* nudge anyone waiting on suspend queue */
874 if (!md_in_flight(md))
878 * Run this off this callpath, as drivers could invoke end_io while
879 * inside their request_fn (and holding the queue lock). Calling
880 * back into ->request_fn() could deadlock attempting to grab the
884 blk_run_queue_async(md->queue);
887 * dm_put() must be at the end of this function. See the comment above
892 static void free_rq_clone(struct request *clone)
894 struct dm_rq_target_io *tio = clone->end_io_data;
896 blk_rq_unprep_clone(clone);
901 * Complete the clone and the original request.
902 * Must be called without queue lock.
904 static void dm_end_request(struct request *clone, int error)
906 int rw = rq_data_dir(clone);
907 struct dm_rq_target_io *tio = clone->end_io_data;
908 struct mapped_device *md = tio->md;
909 struct request *rq = tio->orig;
911 if (rq->cmd_type == REQ_TYPE_BLOCK_PC) {
912 rq->errors = clone->errors;
913 rq->resid_len = clone->resid_len;
917 * We are using the sense buffer of the original
919 * So setting the length of the sense data is enough.
921 rq->sense_len = clone->sense_len;
924 free_rq_clone(clone);
925 blk_end_request_all(rq, error);
926 rq_completed(md, rw, true);
929 static void dm_unprep_request(struct request *rq)
931 struct request *clone = rq->special;
934 rq->cmd_flags &= ~REQ_DONTPREP;
936 free_rq_clone(clone);
940 * Requeue the original request of a clone.
942 void dm_requeue_unmapped_request(struct request *clone)
944 int rw = rq_data_dir(clone);
945 struct dm_rq_target_io *tio = clone->end_io_data;
946 struct mapped_device *md = tio->md;
947 struct request *rq = tio->orig;
948 struct request_queue *q = rq->q;
951 dm_unprep_request(rq);
953 spin_lock_irqsave(q->queue_lock, flags);
954 blk_requeue_request(q, rq);
955 spin_unlock_irqrestore(q->queue_lock, flags);
957 rq_completed(md, rw, 0);
959 EXPORT_SYMBOL_GPL(dm_requeue_unmapped_request);
961 static void __stop_queue(struct request_queue *q)
966 static void stop_queue(struct request_queue *q)
970 spin_lock_irqsave(q->queue_lock, flags);
972 spin_unlock_irqrestore(q->queue_lock, flags);
975 static void __start_queue(struct request_queue *q)
977 if (blk_queue_stopped(q))
981 static void start_queue(struct request_queue *q)
985 spin_lock_irqsave(q->queue_lock, flags);
987 spin_unlock_irqrestore(q->queue_lock, flags);
990 static void dm_done(struct request *clone, int error, bool mapped)
993 struct dm_rq_target_io *tio = clone->end_io_data;
994 dm_request_endio_fn rq_end_io = NULL;
997 rq_end_io = tio->ti->type->rq_end_io;
999 if (mapped && rq_end_io)
1000 r = rq_end_io(tio->ti, clone, error, &tio->info);
1003 if (unlikely(r == -EREMOTEIO && (clone->cmd_flags & REQ_WRITE_SAME) &&
1004 !clone->q->limits.max_write_same_sectors))
1005 disable_write_same(tio->md);
1008 /* The target wants to complete the I/O */
1009 dm_end_request(clone, r);
1010 else if (r == DM_ENDIO_INCOMPLETE)
1011 /* The target will handle the I/O */
1013 else if (r == DM_ENDIO_REQUEUE)
1014 /* The target wants to requeue the I/O */
1015 dm_requeue_unmapped_request(clone);
1017 DMWARN("unimplemented target endio return value: %d", r);
1023 * Request completion handler for request-based dm
1025 static void dm_softirq_done(struct request *rq)
1028 struct request *clone = rq->completion_data;
1029 struct dm_rq_target_io *tio = clone->end_io_data;
1031 if (rq->cmd_flags & REQ_FAILED)
1034 dm_done(clone, tio->error, mapped);
1038 * Complete the clone and the original request with the error status
1039 * through softirq context.
1041 static void dm_complete_request(struct request *clone, int error)
1043 struct dm_rq_target_io *tio = clone->end_io_data;
1044 struct request *rq = tio->orig;
1047 rq->completion_data = clone;
1048 blk_complete_request(rq);
1052 * Complete the not-mapped clone and the original request with the error status
1053 * through softirq context.
1054 * Target's rq_end_io() function isn't called.
1055 * This may be used when the target's map_rq() function fails.
1057 void dm_kill_unmapped_request(struct request *clone, int error)
1059 struct dm_rq_target_io *tio = clone->end_io_data;
1060 struct request *rq = tio->orig;
1062 rq->cmd_flags |= REQ_FAILED;
1063 dm_complete_request(clone, error);
1065 EXPORT_SYMBOL_GPL(dm_kill_unmapped_request);
1068 * Called with the queue lock held
1070 static void end_clone_request(struct request *clone, int error)
1073 * For just cleaning up the information of the queue in which
1074 * the clone was dispatched.
1075 * The clone is *NOT* freed actually here because it is alloced from
1076 * dm own mempool and REQ_ALLOCED isn't set in clone->cmd_flags.
1078 __blk_put_request(clone->q, clone);
1081 * Actual request completion is done in a softirq context which doesn't
1082 * hold the queue lock. Otherwise, deadlock could occur because:
1083 * - another request may be submitted by the upper level driver
1084 * of the stacking during the completion
1085 * - the submission which requires queue lock may be done
1086 * against this queue
1088 dm_complete_request(clone, error);
1092 * Return maximum size of I/O possible at the supplied sector up to the current
1095 static sector_t max_io_len_target_boundary(sector_t sector, struct dm_target *ti)
1097 sector_t target_offset = dm_target_offset(ti, sector);
1099 return ti->len - target_offset;
1102 static sector_t max_io_len(sector_t sector, struct dm_target *ti)
1104 sector_t len = max_io_len_target_boundary(sector, ti);
1105 sector_t offset, max_len;
1108 * Does the target need to split even further?
1110 if (ti->max_io_len) {
1111 offset = dm_target_offset(ti, sector);
1112 if (unlikely(ti->max_io_len & (ti->max_io_len - 1)))
1113 max_len = sector_div(offset, ti->max_io_len);
1115 max_len = offset & (ti->max_io_len - 1);
1116 max_len = ti->max_io_len - max_len;
1125 int dm_set_target_max_io_len(struct dm_target *ti, sector_t len)
1127 if (len > UINT_MAX) {
1128 DMERR("Specified maximum size of target IO (%llu) exceeds limit (%u)",
1129 (unsigned long long)len, UINT_MAX);
1130 ti->error = "Maximum size of target IO is too large";
1134 ti->max_io_len = (uint32_t) len;
1138 EXPORT_SYMBOL_GPL(dm_set_target_max_io_len);
1141 * A target may call dm_accept_partial_bio only from the map routine. It is
1142 * allowed for all bio types except REQ_FLUSH.
1144 * dm_accept_partial_bio informs the dm that the target only wants to process
1145 * additional n_sectors sectors of the bio and the rest of the data should be
1146 * sent in a next bio.
1148 * A diagram that explains the arithmetics:
1149 * +--------------------+---------------+-------+
1151 * +--------------------+---------------+-------+
1153 * <-------------- *tio->len_ptr --------------->
1154 * <------- bi_size ------->
1157 * Region 1 was already iterated over with bio_advance or similar function.
1158 * (it may be empty if the target doesn't use bio_advance)
1159 * Region 2 is the remaining bio size that the target wants to process.
1160 * (it may be empty if region 1 is non-empty, although there is no reason
1162 * The target requires that region 3 is to be sent in the next bio.
1164 * If the target wants to receive multiple copies of the bio (via num_*bios, etc),
1165 * the partially processed part (the sum of regions 1+2) must be the same for all
1166 * copies of the bio.
1168 void dm_accept_partial_bio(struct bio *bio, unsigned n_sectors)
1170 struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone);
1171 unsigned bi_size = bio->bi_iter.bi_size >> SECTOR_SHIFT;
1172 BUG_ON(bio->bi_rw & REQ_FLUSH);
1173 BUG_ON(bi_size > *tio->len_ptr);
1174 BUG_ON(n_sectors > bi_size);
1175 *tio->len_ptr -= bi_size - n_sectors;
1176 bio->bi_iter.bi_size = n_sectors << SECTOR_SHIFT;
1178 EXPORT_SYMBOL_GPL(dm_accept_partial_bio);
1180 static void __map_bio(struct dm_target_io *tio)
1184 struct mapped_device *md;
1185 struct bio *clone = &tio->clone;
1186 struct dm_target *ti = tio->ti;
1188 clone->bi_end_io = clone_endio;
1191 * Map the clone. If r == 0 we don't need to do
1192 * anything, the target has assumed ownership of
1195 atomic_inc(&tio->io->io_count);
1196 sector = clone->bi_iter.bi_sector;
1197 r = ti->type->map(ti, clone);
1198 if (r == DM_MAPIO_REMAPPED) {
1199 /* the bio has been remapped so dispatch it */
1201 trace_block_bio_remap(bdev_get_queue(clone->bi_bdev), clone,
1202 tio->io->bio->bi_bdev->bd_dev, sector);
1204 generic_make_request(clone);
1205 } else if (r < 0 || r == DM_MAPIO_REQUEUE) {
1206 /* error the io and bail out, or requeue it if needed */
1208 dec_pending(tio->io, r);
1211 DMWARN("unimplemented target map return value: %d", r);
1217 struct mapped_device *md;
1218 struct dm_table *map;
1222 unsigned sector_count;
1225 static void bio_setup_sector(struct bio *bio, sector_t sector, unsigned len)
1227 bio->bi_iter.bi_sector = sector;
1228 bio->bi_iter.bi_size = to_bytes(len);
1232 * Creates a bio that consists of range of complete bvecs.
1234 static void clone_bio(struct dm_target_io *tio, struct bio *bio,
1235 sector_t sector, unsigned len)
1237 struct bio *clone = &tio->clone;
1239 __bio_clone_fast(clone, bio);
1241 if (bio_integrity(bio))
1242 bio_integrity_clone(clone, bio, GFP_NOIO);
1244 bio_advance(clone, to_bytes(sector - clone->bi_iter.bi_sector));
1245 clone->bi_iter.bi_size = to_bytes(len);
1247 if (bio_integrity(bio))
1248 bio_integrity_trim(clone, 0, len);
1251 static struct dm_target_io *alloc_tio(struct clone_info *ci,
1252 struct dm_target *ti, int nr_iovecs,
1253 unsigned target_bio_nr)
1255 struct dm_target_io *tio;
1258 clone = bio_alloc_bioset(GFP_NOIO, nr_iovecs, ci->md->bs);
1259 tio = container_of(clone, struct dm_target_io, clone);
1263 tio->target_bio_nr = target_bio_nr;
1268 static void __clone_and_map_simple_bio(struct clone_info *ci,
1269 struct dm_target *ti,
1270 unsigned target_bio_nr, unsigned *len)
1272 struct dm_target_io *tio = alloc_tio(ci, ti, ci->bio->bi_max_vecs, target_bio_nr);
1273 struct bio *clone = &tio->clone;
1278 * Discard requests require the bio's inline iovecs be initialized.
1279 * ci->bio->bi_max_vecs is BIO_INLINE_VECS anyway, for both flush
1280 * and discard, so no need for concern about wasted bvec allocations.
1282 __bio_clone_fast(clone, ci->bio);
1284 bio_setup_sector(clone, ci->sector, *len);
1289 static void __send_duplicate_bios(struct clone_info *ci, struct dm_target *ti,
1290 unsigned num_bios, unsigned *len)
1292 unsigned target_bio_nr;
1294 for (target_bio_nr = 0; target_bio_nr < num_bios; target_bio_nr++)
1295 __clone_and_map_simple_bio(ci, ti, target_bio_nr, len);
1298 static int __send_empty_flush(struct clone_info *ci)
1300 unsigned target_nr = 0;
1301 struct dm_target *ti;
1303 BUG_ON(bio_has_data(ci->bio));
1304 while ((ti = dm_table_get_target(ci->map, target_nr++)))
1305 __send_duplicate_bios(ci, ti, ti->num_flush_bios, NULL);
1310 static void __clone_and_map_data_bio(struct clone_info *ci, struct dm_target *ti,
1311 sector_t sector, unsigned *len)
1313 struct bio *bio = ci->bio;
1314 struct dm_target_io *tio;
1315 unsigned target_bio_nr;
1316 unsigned num_target_bios = 1;
1319 * Does the target want to receive duplicate copies of the bio?
1321 if (bio_data_dir(bio) == WRITE && ti->num_write_bios)
1322 num_target_bios = ti->num_write_bios(ti, bio);
1324 for (target_bio_nr = 0; target_bio_nr < num_target_bios; target_bio_nr++) {
1325 tio = alloc_tio(ci, ti, 0, target_bio_nr);
1327 clone_bio(tio, bio, sector, *len);
1332 typedef unsigned (*get_num_bios_fn)(struct dm_target *ti);
1334 static unsigned get_num_discard_bios(struct dm_target *ti)
1336 return ti->num_discard_bios;
1339 static unsigned get_num_write_same_bios(struct dm_target *ti)
1341 return ti->num_write_same_bios;
1344 typedef bool (*is_split_required_fn)(struct dm_target *ti);
1346 static bool is_split_required_for_discard(struct dm_target *ti)
1348 return ti->split_discard_bios;
1351 static int __send_changing_extent_only(struct clone_info *ci,
1352 get_num_bios_fn get_num_bios,
1353 is_split_required_fn is_split_required)
1355 struct dm_target *ti;
1360 ti = dm_table_find_target(ci->map, ci->sector);
1361 if (!dm_target_is_valid(ti))
1365 * Even though the device advertised support for this type of
1366 * request, that does not mean every target supports it, and
1367 * reconfiguration might also have changed that since the
1368 * check was performed.
1370 num_bios = get_num_bios ? get_num_bios(ti) : 0;
1374 if (is_split_required && !is_split_required(ti))
1375 len = min((sector_t)ci->sector_count, max_io_len_target_boundary(ci->sector, ti));
1377 len = min((sector_t)ci->sector_count, max_io_len(ci->sector, ti));
1379 __send_duplicate_bios(ci, ti, num_bios, &len);
1382 } while (ci->sector_count -= len);
1387 static int __send_discard(struct clone_info *ci)
1389 return __send_changing_extent_only(ci, get_num_discard_bios,
1390 is_split_required_for_discard);
1393 static int __send_write_same(struct clone_info *ci)
1395 return __send_changing_extent_only(ci, get_num_write_same_bios, NULL);
1399 * Select the correct strategy for processing a non-flush bio.
1401 static int __split_and_process_non_flush(struct clone_info *ci)
1403 struct bio *bio = ci->bio;
1404 struct dm_target *ti;
1407 if (unlikely(bio->bi_rw & REQ_DISCARD))
1408 return __send_discard(ci);
1409 else if (unlikely(bio->bi_rw & REQ_WRITE_SAME))
1410 return __send_write_same(ci);
1412 ti = dm_table_find_target(ci->map, ci->sector);
1413 if (!dm_target_is_valid(ti))
1416 len = min_t(sector_t, max_io_len(ci->sector, ti), ci->sector_count);
1418 __clone_and_map_data_bio(ci, ti, ci->sector, &len);
1421 ci->sector_count -= len;
1427 * Entry point to split a bio into clones and submit them to the targets.
1429 static void __split_and_process_bio(struct mapped_device *md,
1430 struct dm_table *map, struct bio *bio)
1432 struct clone_info ci;
1435 if (unlikely(!map)) {
1442 ci.io = alloc_io(md);
1444 atomic_set(&ci.io->io_count, 1);
1447 spin_lock_init(&ci.io->endio_lock);
1448 ci.sector = bio->bi_iter.bi_sector;
1450 start_io_acct(ci.io);
1452 if (bio->bi_rw & REQ_FLUSH) {
1453 ci.bio = &ci.md->flush_bio;
1454 ci.sector_count = 0;
1455 error = __send_empty_flush(&ci);
1456 /* dec_pending submits any data associated with flush */
1459 ci.sector_count = bio_sectors(bio);
1460 while (ci.sector_count && !error)
1461 error = __split_and_process_non_flush(&ci);
1464 /* drop the extra reference count */
1465 dec_pending(ci.io, error);
1467 /*-----------------------------------------------------------------
1469 *---------------------------------------------------------------*/
1471 static int dm_merge_bvec(struct request_queue *q,
1472 struct bvec_merge_data *bvm,
1473 struct bio_vec *biovec)
1475 struct mapped_device *md = q->queuedata;
1476 struct dm_table *map = dm_get_live_table_fast(md);
1477 struct dm_target *ti;
1478 sector_t max_sectors;
1484 ti = dm_table_find_target(map, bvm->bi_sector);
1485 if (!dm_target_is_valid(ti))
1489 * Find maximum amount of I/O that won't need splitting
1491 max_sectors = min(max_io_len(bvm->bi_sector, ti),
1492 (sector_t) BIO_MAX_SECTORS);
1493 max_size = (max_sectors << SECTOR_SHIFT) - bvm->bi_size;
1498 * merge_bvec_fn() returns number of bytes
1499 * it can accept at this offset
1500 * max is precomputed maximal io size
1502 if (max_size && ti->type->merge)
1503 max_size = ti->type->merge(ti, bvm, biovec, max_size);
1505 * If the target doesn't support merge method and some of the devices
1506 * provided their merge_bvec method (we know this by looking at
1507 * queue_max_hw_sectors), then we can't allow bios with multiple vector
1508 * entries. So always set max_size to 0, and the code below allows
1511 else if (queue_max_hw_sectors(q) <= PAGE_SIZE >> 9)
1515 dm_put_live_table_fast(md);
1517 * Always allow an entire first page
1519 if (max_size <= biovec->bv_len && !(bvm->bi_size >> SECTOR_SHIFT))
1520 max_size = biovec->bv_len;
1526 * The request function that just remaps the bio built up by
1529 static void _dm_request(struct request_queue *q, struct bio *bio)
1531 int rw = bio_data_dir(bio);
1532 struct mapped_device *md = q->queuedata;
1535 struct dm_table *map;
1537 map = dm_get_live_table(md, &srcu_idx);
1539 cpu = part_stat_lock();
1540 part_stat_inc(cpu, &dm_disk(md)->part0, ios[rw]);
1541 part_stat_add(cpu, &dm_disk(md)->part0, sectors[rw], bio_sectors(bio));
1544 /* if we're suspended, we have to queue this io for later */
1545 if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))) {
1546 dm_put_live_table(md, srcu_idx);
1548 if (bio_rw(bio) != READA)
1555 __split_and_process_bio(md, map, bio);
1556 dm_put_live_table(md, srcu_idx);
1560 int dm_request_based(struct mapped_device *md)
1562 return blk_queue_stackable(md->queue);
1565 static void dm_request(struct request_queue *q, struct bio *bio)
1567 struct mapped_device *md = q->queuedata;
1569 if (dm_request_based(md))
1570 blk_queue_bio(q, bio);
1572 _dm_request(q, bio);
1575 void dm_dispatch_request(struct request *rq)
1579 if (blk_queue_io_stat(rq->q))
1580 rq->cmd_flags |= REQ_IO_STAT;
1582 rq->start_time = jiffies;
1583 r = blk_insert_cloned_request(rq->q, rq);
1585 dm_complete_request(rq, r);
1587 EXPORT_SYMBOL_GPL(dm_dispatch_request);
1589 static int dm_rq_bio_constructor(struct bio *bio, struct bio *bio_orig,
1592 struct dm_rq_target_io *tio = data;
1593 struct dm_rq_clone_bio_info *info =
1594 container_of(bio, struct dm_rq_clone_bio_info, clone);
1596 info->orig = bio_orig;
1598 bio->bi_end_io = end_clone_bio;
1603 static int setup_clone(struct request *clone, struct request *rq,
1604 struct dm_rq_target_io *tio)
1608 r = blk_rq_prep_clone(clone, rq, tio->md->bs, GFP_ATOMIC,
1609 dm_rq_bio_constructor, tio);
1613 clone->cmd = rq->cmd;
1614 clone->cmd_len = rq->cmd_len;
1615 clone->sense = rq->sense;
1616 clone->end_io = end_clone_request;
1617 clone->end_io_data = tio;
1622 static struct request *clone_rq(struct request *rq, struct mapped_device *md,
1625 struct request *clone;
1626 struct dm_rq_target_io *tio;
1628 tio = alloc_rq_tio(md, gfp_mask);
1636 memset(&tio->info, 0, sizeof(tio->info));
1638 clone = &tio->clone;
1639 if (setup_clone(clone, rq, tio)) {
1649 * Called with the queue lock held.
1651 static int dm_prep_fn(struct request_queue *q, struct request *rq)
1653 struct mapped_device *md = q->queuedata;
1654 struct request *clone;
1656 if (unlikely(rq->special)) {
1657 DMWARN("Already has something in rq->special.");
1658 return BLKPREP_KILL;
1661 clone = clone_rq(rq, md, GFP_ATOMIC);
1663 return BLKPREP_DEFER;
1665 rq->special = clone;
1666 rq->cmd_flags |= REQ_DONTPREP;
1673 * 0 : the request has been processed (not requeued)
1674 * !0 : the request has been requeued
1676 static int map_request(struct dm_target *ti, struct request *clone,
1677 struct mapped_device *md)
1679 int r, requeued = 0;
1680 struct dm_rq_target_io *tio = clone->end_io_data;
1683 r = ti->type->map_rq(ti, clone, &tio->info);
1685 case DM_MAPIO_SUBMITTED:
1686 /* The target has taken the I/O to submit by itself later */
1688 case DM_MAPIO_REMAPPED:
1689 /* The target has remapped the I/O so dispatch it */
1690 trace_block_rq_remap(clone->q, clone, disk_devt(dm_disk(md)),
1691 blk_rq_pos(tio->orig));
1692 dm_dispatch_request(clone);
1694 case DM_MAPIO_REQUEUE:
1695 /* The target wants to requeue the I/O */
1696 dm_requeue_unmapped_request(clone);
1701 DMWARN("unimplemented target map return value: %d", r);
1705 /* The target wants to complete the I/O */
1706 dm_kill_unmapped_request(clone, r);
1713 static struct request *dm_start_request(struct mapped_device *md, struct request *orig)
1715 struct request *clone;
1717 blk_start_request(orig);
1718 clone = orig->special;
1719 atomic_inc(&md->pending[rq_data_dir(clone)]);
1722 * Hold the md reference here for the in-flight I/O.
1723 * We can't rely on the reference count by device opener,
1724 * because the device may be closed during the request completion
1725 * when all bios are completed.
1726 * See the comment in rq_completed() too.
1734 * q->request_fn for request-based dm.
1735 * Called with the queue lock held.
1737 static void dm_request_fn(struct request_queue *q)
1739 struct mapped_device *md = q->queuedata;
1741 struct dm_table *map = dm_get_live_table(md, &srcu_idx);
1742 struct dm_target *ti;
1743 struct request *rq, *clone;
1747 * For suspend, check blk_queue_stopped() and increment
1748 * ->pending within a single queue_lock not to increment the
1749 * number of in-flight I/Os after the queue is stopped in
1752 while (!blk_queue_stopped(q)) {
1753 rq = blk_peek_request(q);
1757 /* always use block 0 to find the target for flushes for now */
1759 if (!(rq->cmd_flags & REQ_FLUSH))
1760 pos = blk_rq_pos(rq);
1762 ti = dm_table_find_target(map, pos);
1763 if (!dm_target_is_valid(ti)) {
1765 * Must perform setup, that dm_done() requires,
1766 * before calling dm_kill_unmapped_request
1768 DMERR_LIMIT("request attempted access beyond the end of device");
1769 clone = dm_start_request(md, rq);
1770 dm_kill_unmapped_request(clone, -EIO);
1774 if (ti->type->busy && ti->type->busy(ti))
1777 clone = dm_start_request(md, rq);
1779 spin_unlock(q->queue_lock);
1780 if (map_request(ti, clone, md))
1783 BUG_ON(!irqs_disabled());
1784 spin_lock(q->queue_lock);
1790 BUG_ON(!irqs_disabled());
1791 spin_lock(q->queue_lock);
1794 blk_delay_queue(q, HZ / 10);
1796 dm_put_live_table(md, srcu_idx);
1799 int dm_underlying_device_busy(struct request_queue *q)
1801 return blk_lld_busy(q);
1803 EXPORT_SYMBOL_GPL(dm_underlying_device_busy);
1805 static int dm_lld_busy(struct request_queue *q)
1808 struct mapped_device *md = q->queuedata;
1809 struct dm_table *map = dm_get_live_table_fast(md);
1811 if (!map || test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))
1814 r = dm_table_any_busy_target(map);
1816 dm_put_live_table_fast(md);
1821 static int dm_any_congested(void *congested_data, int bdi_bits)
1824 struct mapped_device *md = congested_data;
1825 struct dm_table *map;
1827 if (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
1828 map = dm_get_live_table_fast(md);
1831 * Request-based dm cares about only own queue for
1832 * the query about congestion status of request_queue
1834 if (dm_request_based(md))
1835 r = md->queue->backing_dev_info.state &
1838 r = dm_table_any_congested(map, bdi_bits);
1840 dm_put_live_table_fast(md);
1846 /*-----------------------------------------------------------------
1847 * An IDR is used to keep track of allocated minor numbers.
1848 *---------------------------------------------------------------*/
1849 static void free_minor(int minor)
1851 spin_lock(&_minor_lock);
1852 idr_remove(&_minor_idr, minor);
1853 spin_unlock(&_minor_lock);
1857 * See if the device with a specific minor # is free.
1859 static int specific_minor(int minor)
1863 if (minor >= (1 << MINORBITS))
1866 idr_preload(GFP_KERNEL);
1867 spin_lock(&_minor_lock);
1869 r = idr_alloc(&_minor_idr, MINOR_ALLOCED, minor, minor + 1, GFP_NOWAIT);
1871 spin_unlock(&_minor_lock);
1874 return r == -ENOSPC ? -EBUSY : r;
1878 static int next_free_minor(int *minor)
1882 idr_preload(GFP_KERNEL);
1883 spin_lock(&_minor_lock);
1885 r = idr_alloc(&_minor_idr, MINOR_ALLOCED, 0, 1 << MINORBITS, GFP_NOWAIT);
1887 spin_unlock(&_minor_lock);
1895 static const struct block_device_operations dm_blk_dops;
1897 static void dm_wq_work(struct work_struct *work);
1899 static void dm_init_md_queue(struct mapped_device *md)
1902 * Request-based dm devices cannot be stacked on top of bio-based dm
1903 * devices. The type of this dm device has not been decided yet.
1904 * The type is decided at the first table loading time.
1905 * To prevent problematic device stacking, clear the queue flag
1906 * for request stacking support until then.
1908 * This queue is new, so no concurrency on the queue_flags.
1910 queue_flag_clear_unlocked(QUEUE_FLAG_STACKABLE, md->queue);
1912 md->queue->queuedata = md;
1913 md->queue->backing_dev_info.congested_fn = dm_any_congested;
1914 md->queue->backing_dev_info.congested_data = md;
1915 blk_queue_make_request(md->queue, dm_request);
1916 blk_queue_bounce_limit(md->queue, BLK_BOUNCE_ANY);
1917 blk_queue_merge_bvec(md->queue, dm_merge_bvec);
1921 * Allocate and initialise a blank device with a given minor.
1923 static struct mapped_device *alloc_dev(int minor)
1926 struct mapped_device *md = kzalloc(sizeof(*md), GFP_KERNEL);
1930 DMWARN("unable to allocate device, out of memory.");
1934 if (!try_module_get(THIS_MODULE))
1935 goto bad_module_get;
1937 /* get a minor number for the dev */
1938 if (minor == DM_ANY_MINOR)
1939 r = next_free_minor(&minor);
1941 r = specific_minor(minor);
1945 r = init_srcu_struct(&md->io_barrier);
1947 goto bad_io_barrier;
1949 md->type = DM_TYPE_NONE;
1950 mutex_init(&md->suspend_lock);
1951 mutex_init(&md->type_lock);
1952 spin_lock_init(&md->deferred_lock);
1953 atomic_set(&md->holders, 1);
1954 atomic_set(&md->open_count, 0);
1955 atomic_set(&md->event_nr, 0);
1956 atomic_set(&md->uevent_seq, 0);
1957 INIT_LIST_HEAD(&md->uevent_list);
1958 spin_lock_init(&md->uevent_lock);
1960 md->queue = blk_alloc_queue(GFP_KERNEL);
1964 dm_init_md_queue(md);
1966 md->disk = alloc_disk(1);
1970 atomic_set(&md->pending[0], 0);
1971 atomic_set(&md->pending[1], 0);
1972 init_waitqueue_head(&md->wait);
1973 INIT_WORK(&md->work, dm_wq_work);
1974 init_waitqueue_head(&md->eventq);
1975 init_completion(&md->kobj_holder.completion);
1977 md->disk->major = _major;
1978 md->disk->first_minor = minor;
1979 md->disk->fops = &dm_blk_dops;
1980 md->disk->queue = md->queue;
1981 md->disk->private_data = md;
1982 sprintf(md->disk->disk_name, "dm-%d", minor);
1984 format_dev_t(md->name, MKDEV(_major, minor));
1986 md->wq = alloc_workqueue("kdmflush", WQ_MEM_RECLAIM, 0);
1990 md->bdev = bdget_disk(md->disk, 0);
1994 bio_init(&md->flush_bio);
1995 md->flush_bio.bi_bdev = md->bdev;
1996 md->flush_bio.bi_rw = WRITE_FLUSH;
1998 dm_stats_init(&md->stats);
2000 /* Populate the mapping, nobody knows we exist yet */
2001 spin_lock(&_minor_lock);
2002 old_md = idr_replace(&_minor_idr, md, minor);
2003 spin_unlock(&_minor_lock);
2005 BUG_ON(old_md != MINOR_ALLOCED);
2010 destroy_workqueue(md->wq);
2012 del_gendisk(md->disk);
2015 blk_cleanup_queue(md->queue);
2017 cleanup_srcu_struct(&md->io_barrier);
2021 module_put(THIS_MODULE);
2027 static void unlock_fs(struct mapped_device *md);
2029 static void free_dev(struct mapped_device *md)
2031 int minor = MINOR(disk_devt(md->disk));
2035 destroy_workqueue(md->wq);
2037 mempool_destroy(md->io_pool);
2039 bioset_free(md->bs);
2040 blk_integrity_unregister(md->disk);
2041 del_gendisk(md->disk);
2042 cleanup_srcu_struct(&md->io_barrier);
2045 spin_lock(&_minor_lock);
2046 md->disk->private_data = NULL;
2047 spin_unlock(&_minor_lock);
2050 blk_cleanup_queue(md->queue);
2051 dm_stats_cleanup(&md->stats);
2052 module_put(THIS_MODULE);
2056 static void __bind_mempools(struct mapped_device *md, struct dm_table *t)
2058 struct dm_md_mempools *p = dm_table_get_md_mempools(t);
2060 if (md->io_pool && md->bs) {
2061 /* The md already has necessary mempools. */
2062 if (dm_table_get_type(t) == DM_TYPE_BIO_BASED) {
2064 * Reload bioset because front_pad may have changed
2065 * because a different table was loaded.
2067 bioset_free(md->bs);
2070 } else if (dm_table_get_type(t) == DM_TYPE_REQUEST_BASED) {
2072 * There's no need to reload with request-based dm
2073 * because the size of front_pad doesn't change.
2074 * Note for future: If you are to reload bioset,
2075 * prep-ed requests in the queue may refer
2076 * to bio from the old bioset, so you must walk
2077 * through the queue to unprep.
2083 BUG_ON(!p || md->io_pool || md->bs);
2085 md->io_pool = p->io_pool;
2091 /* mempool bind completed, now no need any mempools in the table */
2092 dm_table_free_md_mempools(t);
2096 * Bind a table to the device.
2098 static void event_callback(void *context)
2100 unsigned long flags;
2102 struct mapped_device *md = (struct mapped_device *) context;
2104 spin_lock_irqsave(&md->uevent_lock, flags);
2105 list_splice_init(&md->uevent_list, &uevents);
2106 spin_unlock_irqrestore(&md->uevent_lock, flags);
2108 dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj);
2110 atomic_inc(&md->event_nr);
2111 wake_up(&md->eventq);
2115 * Protected by md->suspend_lock obtained by dm_swap_table().
2117 static void __set_size(struct mapped_device *md, sector_t size)
2119 set_capacity(md->disk, size);
2121 i_size_write(md->bdev->bd_inode, (loff_t)size << SECTOR_SHIFT);
2125 * Return 1 if the queue has a compulsory merge_bvec_fn function.
2127 * If this function returns 0, then the device is either a non-dm
2128 * device without a merge_bvec_fn, or it is a dm device that is
2129 * able to split any bios it receives that are too big.
2131 int dm_queue_merge_is_compulsory(struct request_queue *q)
2133 struct mapped_device *dev_md;
2135 if (!q->merge_bvec_fn)
2138 if (q->make_request_fn == dm_request) {
2139 dev_md = q->queuedata;
2140 if (test_bit(DMF_MERGE_IS_OPTIONAL, &dev_md->flags))
2147 static int dm_device_merge_is_compulsory(struct dm_target *ti,
2148 struct dm_dev *dev, sector_t start,
2149 sector_t len, void *data)
2151 struct block_device *bdev = dev->bdev;
2152 struct request_queue *q = bdev_get_queue(bdev);
2154 return dm_queue_merge_is_compulsory(q);
2158 * Return 1 if it is acceptable to ignore merge_bvec_fn based
2159 * on the properties of the underlying devices.
2161 static int dm_table_merge_is_optional(struct dm_table *table)
2164 struct dm_target *ti;
2166 while (i < dm_table_get_num_targets(table)) {
2167 ti = dm_table_get_target(table, i++);
2169 if (ti->type->iterate_devices &&
2170 ti->type->iterate_devices(ti, dm_device_merge_is_compulsory, NULL))
2178 * Returns old map, which caller must destroy.
2180 static struct dm_table *__bind(struct mapped_device *md, struct dm_table *t,
2181 struct queue_limits *limits)
2183 struct dm_table *old_map;
2184 struct request_queue *q = md->queue;
2186 int merge_is_optional;
2188 size = dm_table_get_size(t);
2191 * Wipe any geometry if the size of the table changed.
2193 if (size != dm_get_size(md))
2194 memset(&md->geometry, 0, sizeof(md->geometry));
2196 __set_size(md, size);
2198 dm_table_event_callback(t, event_callback, md);
2201 * The queue hasn't been stopped yet, if the old table type wasn't
2202 * for request-based during suspension. So stop it to prevent
2203 * I/O mapping before resume.
2204 * This must be done before setting the queue restrictions,
2205 * because request-based dm may be run just after the setting.
2207 if (dm_table_request_based(t) && !blk_queue_stopped(q))
2210 __bind_mempools(md, t);
2212 merge_is_optional = dm_table_merge_is_optional(t);
2215 rcu_assign_pointer(md->map, t);
2216 md->immutable_target_type = dm_table_get_immutable_target_type(t);
2218 dm_table_set_restrictions(t, q, limits);
2219 if (merge_is_optional)
2220 set_bit(DMF_MERGE_IS_OPTIONAL, &md->flags);
2222 clear_bit(DMF_MERGE_IS_OPTIONAL, &md->flags);
2229 * Returns unbound table for the caller to free.
2231 static struct dm_table *__unbind(struct mapped_device *md)
2233 struct dm_table *map = md->map;
2238 dm_table_event_callback(map, NULL, NULL);
2239 RCU_INIT_POINTER(md->map, NULL);
2246 * Constructor for a new device.
2248 int dm_create(int minor, struct mapped_device **result)
2250 struct mapped_device *md;
2252 md = alloc_dev(minor);
2263 * Functions to manage md->type.
2264 * All are required to hold md->type_lock.
2266 void dm_lock_md_type(struct mapped_device *md)
2268 mutex_lock(&md->type_lock);
2271 void dm_unlock_md_type(struct mapped_device *md)
2273 mutex_unlock(&md->type_lock);
2276 void dm_set_md_type(struct mapped_device *md, unsigned type)
2278 BUG_ON(!mutex_is_locked(&md->type_lock));
2282 unsigned dm_get_md_type(struct mapped_device *md)
2284 BUG_ON(!mutex_is_locked(&md->type_lock));
2288 struct target_type *dm_get_immutable_target_type(struct mapped_device *md)
2290 return md->immutable_target_type;
2294 * The queue_limits are only valid as long as you have a reference
2297 struct queue_limits *dm_get_queue_limits(struct mapped_device *md)
2299 BUG_ON(!atomic_read(&md->holders));
2300 return &md->queue->limits;
2302 EXPORT_SYMBOL_GPL(dm_get_queue_limits);
2305 * Fully initialize a request-based queue (->elevator, ->request_fn, etc).
2307 static int dm_init_request_based_queue(struct mapped_device *md)
2309 struct request_queue *q = NULL;
2311 if (md->queue->elevator)
2314 /* Fully initialize the queue */
2315 q = blk_init_allocated_queue(md->queue, dm_request_fn, NULL);
2320 dm_init_md_queue(md);
2321 blk_queue_softirq_done(md->queue, dm_softirq_done);
2322 blk_queue_prep_rq(md->queue, dm_prep_fn);
2323 blk_queue_lld_busy(md->queue, dm_lld_busy);
2325 elv_register_queue(md->queue);
2331 * Setup the DM device's queue based on md's type
2333 int dm_setup_md_queue(struct mapped_device *md)
2335 if ((dm_get_md_type(md) == DM_TYPE_REQUEST_BASED) &&
2336 !dm_init_request_based_queue(md)) {
2337 DMWARN("Cannot initialize queue for request-based mapped device");
2344 static struct mapped_device *dm_find_md(dev_t dev)
2346 struct mapped_device *md;
2347 unsigned minor = MINOR(dev);
2349 if (MAJOR(dev) != _major || minor >= (1 << MINORBITS))
2352 spin_lock(&_minor_lock);
2354 md = idr_find(&_minor_idr, minor);
2355 if (md && (md == MINOR_ALLOCED ||
2356 (MINOR(disk_devt(dm_disk(md))) != minor) ||
2357 dm_deleting_md(md) ||
2358 test_bit(DMF_FREEING, &md->flags))) {
2364 spin_unlock(&_minor_lock);
2369 struct mapped_device *dm_get_md(dev_t dev)
2371 struct mapped_device *md = dm_find_md(dev);
2378 EXPORT_SYMBOL_GPL(dm_get_md);
2380 void *dm_get_mdptr(struct mapped_device *md)
2382 return md->interface_ptr;
2385 void dm_set_mdptr(struct mapped_device *md, void *ptr)
2387 md->interface_ptr = ptr;
2390 void dm_get(struct mapped_device *md)
2392 atomic_inc(&md->holders);
2393 BUG_ON(test_bit(DMF_FREEING, &md->flags));
2396 const char *dm_device_name(struct mapped_device *md)
2400 EXPORT_SYMBOL_GPL(dm_device_name);
2402 static void __dm_destroy(struct mapped_device *md, bool wait)
2404 struct dm_table *map;
2409 spin_lock(&_minor_lock);
2410 map = dm_get_live_table(md, &srcu_idx);
2411 idr_replace(&_minor_idr, MINOR_ALLOCED, MINOR(disk_devt(dm_disk(md))));
2412 set_bit(DMF_FREEING, &md->flags);
2413 spin_unlock(&_minor_lock);
2415 if (!dm_suspended_md(md)) {
2416 dm_table_presuspend_targets(map);
2417 dm_table_postsuspend_targets(map);
2420 /* dm_put_live_table must be before msleep, otherwise deadlock is possible */
2421 dm_put_live_table(md, srcu_idx);
2424 * Rare, but there may be I/O requests still going to complete,
2425 * for example. Wait for all references to disappear.
2426 * No one should increment the reference count of the mapped_device,
2427 * after the mapped_device state becomes DMF_FREEING.
2430 while (atomic_read(&md->holders))
2432 else if (atomic_read(&md->holders))
2433 DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)",
2434 dm_device_name(md), atomic_read(&md->holders));
2437 dm_table_destroy(__unbind(md));
2441 void dm_destroy(struct mapped_device *md)
2443 __dm_destroy(md, true);
2446 void dm_destroy_immediate(struct mapped_device *md)
2448 __dm_destroy(md, false);
2451 void dm_put(struct mapped_device *md)
2453 atomic_dec(&md->holders);
2455 EXPORT_SYMBOL_GPL(dm_put);
2457 static int dm_wait_for_completion(struct mapped_device *md, int interruptible)
2460 DECLARE_WAITQUEUE(wait, current);
2462 add_wait_queue(&md->wait, &wait);
2465 set_current_state(interruptible);
2467 if (!md_in_flight(md))
2470 if (interruptible == TASK_INTERRUPTIBLE &&
2471 signal_pending(current)) {
2478 set_current_state(TASK_RUNNING);
2480 remove_wait_queue(&md->wait, &wait);
2486 * Process the deferred bios
2488 static void dm_wq_work(struct work_struct *work)
2490 struct mapped_device *md = container_of(work, struct mapped_device,
2494 struct dm_table *map;
2496 map = dm_get_live_table(md, &srcu_idx);
2498 while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
2499 spin_lock_irq(&md->deferred_lock);
2500 c = bio_list_pop(&md->deferred);
2501 spin_unlock_irq(&md->deferred_lock);
2506 if (dm_request_based(md))
2507 generic_make_request(c);
2509 __split_and_process_bio(md, map, c);
2512 dm_put_live_table(md, srcu_idx);
2515 static void dm_queue_flush(struct mapped_device *md)
2517 clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2518 smp_mb__after_atomic();
2519 queue_work(md->wq, &md->work);
2523 * Swap in a new table, returning the old one for the caller to destroy.
2525 struct dm_table *dm_swap_table(struct mapped_device *md, struct dm_table *table)
2527 struct dm_table *live_map = NULL, *map = ERR_PTR(-EINVAL);
2528 struct queue_limits limits;
2531 mutex_lock(&md->suspend_lock);
2533 /* device must be suspended */
2534 if (!dm_suspended_md(md))
2538 * If the new table has no data devices, retain the existing limits.
2539 * This helps multipath with queue_if_no_path if all paths disappear,
2540 * then new I/O is queued based on these limits, and then some paths
2543 if (dm_table_has_no_data_devices(table)) {
2544 live_map = dm_get_live_table_fast(md);
2546 limits = md->queue->limits;
2547 dm_put_live_table_fast(md);
2551 r = dm_calculate_queue_limits(table, &limits);
2558 map = __bind(md, table, &limits);
2561 mutex_unlock(&md->suspend_lock);
2566 * Functions to lock and unlock any filesystem running on the
2569 static int lock_fs(struct mapped_device *md)
2573 WARN_ON(md->frozen_sb);
2575 md->frozen_sb = freeze_bdev(md->bdev);
2576 if (IS_ERR(md->frozen_sb)) {
2577 r = PTR_ERR(md->frozen_sb);
2578 md->frozen_sb = NULL;
2582 set_bit(DMF_FROZEN, &md->flags);
2587 static void unlock_fs(struct mapped_device *md)
2589 if (!test_bit(DMF_FROZEN, &md->flags))
2592 thaw_bdev(md->bdev, md->frozen_sb);
2593 md->frozen_sb = NULL;
2594 clear_bit(DMF_FROZEN, &md->flags);
2598 * We need to be able to change a mapping table under a mounted
2599 * filesystem. For example we might want to move some data in
2600 * the background. Before the table can be swapped with
2601 * dm_bind_table, dm_suspend must be called to flush any in
2602 * flight bios and ensure that any further io gets deferred.
2605 * Suspend mechanism in request-based dm.
2607 * 1. Flush all I/Os by lock_fs() if needed.
2608 * 2. Stop dispatching any I/O by stopping the request_queue.
2609 * 3. Wait for all in-flight I/Os to be completed or requeued.
2611 * To abort suspend, start the request_queue.
2613 int dm_suspend(struct mapped_device *md, unsigned suspend_flags)
2615 struct dm_table *map = NULL;
2617 int do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG ? 1 : 0;
2618 int noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG ? 1 : 0;
2620 mutex_lock(&md->suspend_lock);
2622 if (dm_suspended_md(md)) {
2630 * DMF_NOFLUSH_SUSPENDING must be set before presuspend.
2631 * This flag is cleared before dm_suspend returns.
2634 set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2636 /* This does not get reverted if there's an error later. */
2637 dm_table_presuspend_targets(map);
2640 * Flush I/O to the device.
2641 * Any I/O submitted after lock_fs() may not be flushed.
2642 * noflush takes precedence over do_lockfs.
2643 * (lock_fs() flushes I/Os and waits for them to complete.)
2645 if (!noflush && do_lockfs) {
2652 * Here we must make sure that no processes are submitting requests
2653 * to target drivers i.e. no one may be executing
2654 * __split_and_process_bio. This is called from dm_request and
2657 * To get all processes out of __split_and_process_bio in dm_request,
2658 * we take the write lock. To prevent any process from reentering
2659 * __split_and_process_bio from dm_request and quiesce the thread
2660 * (dm_wq_work), we set BMF_BLOCK_IO_FOR_SUSPEND and call
2661 * flush_workqueue(md->wq).
2663 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2664 synchronize_srcu(&md->io_barrier);
2667 * Stop md->queue before flushing md->wq in case request-based
2668 * dm defers requests to md->wq from md->queue.
2670 if (dm_request_based(md))
2671 stop_queue(md->queue);
2673 flush_workqueue(md->wq);
2676 * At this point no more requests are entering target request routines.
2677 * We call dm_wait_for_completion to wait for all existing requests
2680 r = dm_wait_for_completion(md, TASK_INTERRUPTIBLE);
2683 clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2684 synchronize_srcu(&md->io_barrier);
2686 /* were we interrupted ? */
2690 if (dm_request_based(md))
2691 start_queue(md->queue);
2694 goto out_unlock; /* pushback list is already flushed, so skip flush */
2698 * If dm_wait_for_completion returned 0, the device is completely
2699 * quiescent now. There is no request-processing activity. All new
2700 * requests are being added to md->deferred list.
2703 set_bit(DMF_SUSPENDED, &md->flags);
2705 dm_table_postsuspend_targets(map);
2708 mutex_unlock(&md->suspend_lock);
2712 int dm_resume(struct mapped_device *md)
2715 struct dm_table *map = NULL;
2717 mutex_lock(&md->suspend_lock);
2718 if (!dm_suspended_md(md))
2722 if (!map || !dm_table_get_size(map))
2725 r = dm_table_resume_targets(map);
2732 * Flushing deferred I/Os must be done after targets are resumed
2733 * so that mapping of targets can work correctly.
2734 * Request-based dm is queueing the deferred I/Os in its request_queue.
2736 if (dm_request_based(md))
2737 start_queue(md->queue);
2741 clear_bit(DMF_SUSPENDED, &md->flags);
2745 mutex_unlock(&md->suspend_lock);
2751 * Internal suspend/resume works like userspace-driven suspend. It waits
2752 * until all bios finish and prevents issuing new bios to the target drivers.
2753 * It may be used only from the kernel.
2755 * Internal suspend holds md->suspend_lock, which prevents interaction with
2756 * userspace-driven suspend.
2759 void dm_internal_suspend(struct mapped_device *md)
2761 mutex_lock(&md->suspend_lock);
2762 if (dm_suspended_md(md))
2765 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2766 synchronize_srcu(&md->io_barrier);
2767 flush_workqueue(md->wq);
2768 dm_wait_for_completion(md, TASK_UNINTERRUPTIBLE);
2771 void dm_internal_resume(struct mapped_device *md)
2773 if (dm_suspended_md(md))
2779 mutex_unlock(&md->suspend_lock);
2782 /*-----------------------------------------------------------------
2783 * Event notification.
2784 *---------------------------------------------------------------*/
2785 int dm_kobject_uevent(struct mapped_device *md, enum kobject_action action,
2788 char udev_cookie[DM_COOKIE_LENGTH];
2789 char *envp[] = { udev_cookie, NULL };
2792 return kobject_uevent(&disk_to_dev(md->disk)->kobj, action);
2794 snprintf(udev_cookie, DM_COOKIE_LENGTH, "%s=%u",
2795 DM_COOKIE_ENV_VAR_NAME, cookie);
2796 return kobject_uevent_env(&disk_to_dev(md->disk)->kobj,
2801 uint32_t dm_next_uevent_seq(struct mapped_device *md)
2803 return atomic_add_return(1, &md->uevent_seq);
2806 uint32_t dm_get_event_nr(struct mapped_device *md)
2808 return atomic_read(&md->event_nr);
2811 int dm_wait_event(struct mapped_device *md, int event_nr)
2813 return wait_event_interruptible(md->eventq,
2814 (event_nr != atomic_read(&md->event_nr)));
2817 void dm_uevent_add(struct mapped_device *md, struct list_head *elist)
2819 unsigned long flags;
2821 spin_lock_irqsave(&md->uevent_lock, flags);
2822 list_add(elist, &md->uevent_list);
2823 spin_unlock_irqrestore(&md->uevent_lock, flags);
2827 * The gendisk is only valid as long as you have a reference
2830 struct gendisk *dm_disk(struct mapped_device *md)
2835 struct kobject *dm_kobject(struct mapped_device *md)
2837 return &md->kobj_holder.kobj;
2840 struct mapped_device *dm_get_from_kobject(struct kobject *kobj)
2842 struct mapped_device *md;
2844 md = container_of(kobj, struct mapped_device, kobj_holder.kobj);
2846 if (test_bit(DMF_FREEING, &md->flags) ||
2854 int dm_suspended_md(struct mapped_device *md)
2856 return test_bit(DMF_SUSPENDED, &md->flags);
2859 int dm_test_deferred_remove_flag(struct mapped_device *md)
2861 return test_bit(DMF_DEFERRED_REMOVE, &md->flags);
2864 int dm_suspended(struct dm_target *ti)
2866 return dm_suspended_md(dm_table_get_md(ti->table));
2868 EXPORT_SYMBOL_GPL(dm_suspended);
2870 int dm_noflush_suspending(struct dm_target *ti)
2872 return __noflush_suspending(dm_table_get_md(ti->table));
2874 EXPORT_SYMBOL_GPL(dm_noflush_suspending);
2876 struct dm_md_mempools *dm_alloc_md_mempools(unsigned type, unsigned integrity, unsigned per_bio_data_size)
2878 struct dm_md_mempools *pools = kzalloc(sizeof(*pools), GFP_KERNEL);
2879 struct kmem_cache *cachep;
2880 unsigned int pool_size;
2881 unsigned int front_pad;
2886 if (type == DM_TYPE_BIO_BASED) {
2888 pool_size = dm_get_reserved_bio_based_ios();
2889 front_pad = roundup(per_bio_data_size, __alignof__(struct dm_target_io)) + offsetof(struct dm_target_io, clone);
2890 } else if (type == DM_TYPE_REQUEST_BASED) {
2891 cachep = _rq_tio_cache;
2892 pool_size = dm_get_reserved_rq_based_ios();
2893 front_pad = offsetof(struct dm_rq_clone_bio_info, clone);
2894 /* per_bio_data_size is not used. See __bind_mempools(). */
2895 WARN_ON(per_bio_data_size != 0);
2899 pools->io_pool = mempool_create_slab_pool(pool_size, cachep);
2900 if (!pools->io_pool)
2903 pools->bs = bioset_create(pool_size, front_pad);
2907 if (integrity && bioset_integrity_create(pools->bs, pool_size))
2913 dm_free_md_mempools(pools);
2918 void dm_free_md_mempools(struct dm_md_mempools *pools)
2924 mempool_destroy(pools->io_pool);
2927 bioset_free(pools->bs);
2932 static const struct block_device_operations dm_blk_dops = {
2933 .open = dm_blk_open,
2934 .release = dm_blk_close,
2935 .ioctl = dm_blk_ioctl,
2936 .getgeo = dm_blk_getgeo,
2937 .owner = THIS_MODULE
2943 module_init(dm_init);
2944 module_exit(dm_exit);
2946 module_param(major, uint, 0);
2947 MODULE_PARM_DESC(major, "The major number of the device mapper");
2949 module_param(reserved_bio_based_ios, uint, S_IRUGO | S_IWUSR);
2950 MODULE_PARM_DESC(reserved_bio_based_ios, "Reserved IOs in bio-based mempools");
2952 module_param(reserved_rq_based_ios, uint, S_IRUGO | S_IWUSR);
2953 MODULE_PARM_DESC(reserved_rq_based_ios, "Reserved IOs in request-based mempools");
2955 MODULE_DESCRIPTION(DM_NAME " driver");
2956 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
2957 MODULE_LICENSE("GPL");