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);
59 * One of these is allocated per bio.
62 struct mapped_device *md;
66 unsigned long start_time;
67 spinlock_t endio_lock;
68 struct dm_stats_aux stats_aux;
72 * For request-based dm.
73 * One of these is allocated per request.
75 struct dm_rq_target_io {
76 struct mapped_device *md;
78 struct request *orig, clone;
84 * For request-based dm - the bio clones we allocate are embedded in these
87 * We allocate these with bio_alloc_bioset, using the front_pad parameter when
88 * the bioset is created - this means the bio has to come at the end of the
91 struct dm_rq_clone_bio_info {
93 struct dm_rq_target_io *tio;
97 union map_info *dm_get_rq_mapinfo(struct request *rq)
99 if (rq && rq->end_io_data)
100 return &((struct dm_rq_target_io *)rq->end_io_data)->info;
103 EXPORT_SYMBOL_GPL(dm_get_rq_mapinfo);
105 #define MINOR_ALLOCED ((void *)-1)
108 * Bits for the md->flags field.
110 #define DMF_BLOCK_IO_FOR_SUSPEND 0
111 #define DMF_SUSPENDED 1
113 #define DMF_FREEING 3
114 #define DMF_DELETING 4
115 #define DMF_NOFLUSH_SUSPENDING 5
116 #define DMF_MERGE_IS_OPTIONAL 6
117 #define DMF_DEFERRED_REMOVE 7
120 * A dummy definition to make RCU happy.
121 * struct dm_table should never be dereferenced in this file.
128 * Work processed by per-device workqueue.
130 struct mapped_device {
131 struct srcu_struct io_barrier;
132 struct mutex suspend_lock;
137 * The current mapping.
138 * Use dm_get_live_table{_fast} or take suspend_lock for
141 struct dm_table *map;
145 struct request_queue *queue;
147 /* Protect queue and type against concurrent access. */
148 struct mutex type_lock;
150 struct target_type *immutable_target_type;
152 struct gendisk *disk;
158 * A list of ios that arrived while we were suspended.
161 wait_queue_head_t wait;
162 struct work_struct work;
163 struct bio_list deferred;
164 spinlock_t deferred_lock;
167 * Processing queue (flush)
169 struct workqueue_struct *wq;
172 * io objects are allocated from here.
182 wait_queue_head_t eventq;
184 struct list_head uevent_list;
185 spinlock_t uevent_lock; /* Protect access to uevent_list */
188 * freeze/thaw support require holding onto a super block
190 struct super_block *frozen_sb;
191 struct block_device *bdev;
193 /* forced geometry settings */
194 struct hd_geometry geometry;
196 /* kobject and completion */
197 struct dm_kobject_holder kobj_holder;
199 /* zero-length flush that will be cloned and submitted to targets */
200 struct bio flush_bio;
202 struct dm_stats stats;
206 * For mempools pre-allocation at the table loading time.
208 struct dm_md_mempools {
213 #define RESERVED_BIO_BASED_IOS 16
214 #define RESERVED_REQUEST_BASED_IOS 256
215 #define RESERVED_MAX_IOS 1024
216 static struct kmem_cache *_io_cache;
217 static struct kmem_cache *_rq_tio_cache;
220 * Bio-based DM's mempools' reserved IOs set by the user.
222 static unsigned reserved_bio_based_ios = RESERVED_BIO_BASED_IOS;
225 * Request-based DM's mempools' reserved IOs set by the user.
227 static unsigned reserved_rq_based_ios = RESERVED_REQUEST_BASED_IOS;
229 static unsigned __dm_get_reserved_ios(unsigned *reserved_ios,
230 unsigned def, unsigned max)
232 unsigned ios = ACCESS_ONCE(*reserved_ios);
233 unsigned modified_ios = 0;
241 (void)cmpxchg(reserved_ios, ios, modified_ios);
248 unsigned dm_get_reserved_bio_based_ios(void)
250 return __dm_get_reserved_ios(&reserved_bio_based_ios,
251 RESERVED_BIO_BASED_IOS, RESERVED_MAX_IOS);
253 EXPORT_SYMBOL_GPL(dm_get_reserved_bio_based_ios);
255 unsigned dm_get_reserved_rq_based_ios(void)
257 return __dm_get_reserved_ios(&reserved_rq_based_ios,
258 RESERVED_REQUEST_BASED_IOS, RESERVED_MAX_IOS);
260 EXPORT_SYMBOL_GPL(dm_get_reserved_rq_based_ios);
262 static int __init local_init(void)
266 /* allocate a slab for the dm_ios */
267 _io_cache = KMEM_CACHE(dm_io, 0);
271 _rq_tio_cache = KMEM_CACHE(dm_rq_target_io, 0);
273 goto out_free_io_cache;
275 r = dm_uevent_init();
277 goto out_free_rq_tio_cache;
280 r = register_blkdev(_major, _name);
282 goto out_uevent_exit;
291 out_free_rq_tio_cache:
292 kmem_cache_destroy(_rq_tio_cache);
294 kmem_cache_destroy(_io_cache);
299 static void local_exit(void)
301 flush_scheduled_work();
303 kmem_cache_destroy(_rq_tio_cache);
304 kmem_cache_destroy(_io_cache);
305 unregister_blkdev(_major, _name);
310 DMINFO("cleaned up");
313 static int (*_inits[])(void) __initdata = {
324 static void (*_exits[])(void) = {
335 static int __init dm_init(void)
337 const int count = ARRAY_SIZE(_inits);
341 for (i = 0; i < count; i++) {
356 static void __exit dm_exit(void)
358 int i = ARRAY_SIZE(_exits);
364 * Should be empty by this point.
366 idr_destroy(&_minor_idr);
370 * Block device functions
372 int dm_deleting_md(struct mapped_device *md)
374 return test_bit(DMF_DELETING, &md->flags);
377 static int dm_blk_open(struct block_device *bdev, fmode_t mode)
379 struct mapped_device *md;
381 spin_lock(&_minor_lock);
383 md = bdev->bd_disk->private_data;
387 if (test_bit(DMF_FREEING, &md->flags) ||
388 dm_deleting_md(md)) {
394 atomic_inc(&md->open_count);
397 spin_unlock(&_minor_lock);
399 return md ? 0 : -ENXIO;
402 static void dm_blk_close(struct gendisk *disk, fmode_t mode)
404 struct mapped_device *md = disk->private_data;
406 spin_lock(&_minor_lock);
408 if (atomic_dec_and_test(&md->open_count) &&
409 (test_bit(DMF_DEFERRED_REMOVE, &md->flags)))
410 schedule_work(&deferred_remove_work);
414 spin_unlock(&_minor_lock);
417 int dm_open_count(struct mapped_device *md)
419 return atomic_read(&md->open_count);
423 * Guarantees nothing is using the device before it's deleted.
425 int dm_lock_for_deletion(struct mapped_device *md, bool mark_deferred, bool only_deferred)
429 spin_lock(&_minor_lock);
431 if (dm_open_count(md)) {
434 set_bit(DMF_DEFERRED_REMOVE, &md->flags);
435 } else if (only_deferred && !test_bit(DMF_DEFERRED_REMOVE, &md->flags))
438 set_bit(DMF_DELETING, &md->flags);
440 spin_unlock(&_minor_lock);
445 int dm_cancel_deferred_remove(struct mapped_device *md)
449 spin_lock(&_minor_lock);
451 if (test_bit(DMF_DELETING, &md->flags))
454 clear_bit(DMF_DEFERRED_REMOVE, &md->flags);
456 spin_unlock(&_minor_lock);
461 static void do_deferred_remove(struct work_struct *w)
463 dm_deferred_remove();
466 sector_t dm_get_size(struct mapped_device *md)
468 return get_capacity(md->disk);
471 struct request_queue *dm_get_md_queue(struct mapped_device *md)
476 struct dm_stats *dm_get_stats(struct mapped_device *md)
481 static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo)
483 struct mapped_device *md = bdev->bd_disk->private_data;
485 return dm_get_geometry(md, geo);
488 static int dm_blk_ioctl(struct block_device *bdev, fmode_t mode,
489 unsigned int cmd, unsigned long arg)
491 struct mapped_device *md = bdev->bd_disk->private_data;
493 struct dm_table *map;
494 struct dm_target *tgt;
498 map = dm_get_live_table(md, &srcu_idx);
500 if (!map || !dm_table_get_size(map))
503 /* We only support devices that have a single target */
504 if (dm_table_get_num_targets(map) != 1)
507 tgt = dm_table_get_target(map, 0);
509 if (dm_suspended_md(md)) {
514 if (tgt->type->ioctl)
515 r = tgt->type->ioctl(tgt, cmd, arg);
518 dm_put_live_table(md, srcu_idx);
520 if (r == -ENOTCONN) {
528 static struct dm_io *alloc_io(struct mapped_device *md)
530 return mempool_alloc(md->io_pool, GFP_NOIO);
533 static void free_io(struct mapped_device *md, struct dm_io *io)
535 mempool_free(io, md->io_pool);
538 static void free_tio(struct mapped_device *md, struct dm_target_io *tio)
540 bio_put(&tio->clone);
543 static struct dm_rq_target_io *alloc_rq_tio(struct mapped_device *md,
546 return mempool_alloc(md->io_pool, gfp_mask);
549 static void free_rq_tio(struct dm_rq_target_io *tio)
551 mempool_free(tio, tio->md->io_pool);
554 static int md_in_flight(struct mapped_device *md)
556 return atomic_read(&md->pending[READ]) +
557 atomic_read(&md->pending[WRITE]);
560 static void start_io_acct(struct dm_io *io)
562 struct mapped_device *md = io->md;
563 struct bio *bio = io->bio;
565 int rw = bio_data_dir(bio);
567 io->start_time = jiffies;
569 cpu = part_stat_lock();
570 part_round_stats(cpu, &dm_disk(md)->part0);
572 atomic_set(&dm_disk(md)->part0.in_flight[rw],
573 atomic_inc_return(&md->pending[rw]));
575 if (unlikely(dm_stats_used(&md->stats)))
576 dm_stats_account_io(&md->stats, bio->bi_rw, bio->bi_iter.bi_sector,
577 bio_sectors(bio), false, 0, &io->stats_aux);
580 static void end_io_acct(struct dm_io *io)
582 struct mapped_device *md = io->md;
583 struct bio *bio = io->bio;
584 unsigned long duration = jiffies - io->start_time;
586 int rw = bio_data_dir(bio);
588 cpu = part_stat_lock();
589 part_round_stats(cpu, &dm_disk(md)->part0);
590 part_stat_add(cpu, &dm_disk(md)->part0, ticks[rw], duration);
593 if (unlikely(dm_stats_used(&md->stats)))
594 dm_stats_account_io(&md->stats, bio->bi_rw, bio->bi_iter.bi_sector,
595 bio_sectors(bio), true, duration, &io->stats_aux);
598 * After this is decremented the bio must not be touched if it is
601 pending = atomic_dec_return(&md->pending[rw]);
602 atomic_set(&dm_disk(md)->part0.in_flight[rw], pending);
603 pending += atomic_read(&md->pending[rw^0x1]);
605 /* nudge anyone waiting on suspend queue */
611 * Add the bio to the list of deferred io.
613 static void queue_io(struct mapped_device *md, struct bio *bio)
617 spin_lock_irqsave(&md->deferred_lock, flags);
618 bio_list_add(&md->deferred, bio);
619 spin_unlock_irqrestore(&md->deferred_lock, flags);
620 queue_work(md->wq, &md->work);
624 * Everyone (including functions in this file), should use this
625 * function to access the md->map field, and make sure they call
626 * dm_put_live_table() when finished.
628 struct dm_table *dm_get_live_table(struct mapped_device *md, int *srcu_idx) __acquires(md->io_barrier)
630 *srcu_idx = srcu_read_lock(&md->io_barrier);
632 return srcu_dereference(md->map, &md->io_barrier);
635 void dm_put_live_table(struct mapped_device *md, int srcu_idx) __releases(md->io_barrier)
637 srcu_read_unlock(&md->io_barrier, srcu_idx);
640 void dm_sync_table(struct mapped_device *md)
642 synchronize_srcu(&md->io_barrier);
643 synchronize_rcu_expedited();
647 * A fast alternative to dm_get_live_table/dm_put_live_table.
648 * The caller must not block between these two functions.
650 static struct dm_table *dm_get_live_table_fast(struct mapped_device *md) __acquires(RCU)
653 return rcu_dereference(md->map);
656 static void dm_put_live_table_fast(struct mapped_device *md) __releases(RCU)
662 * Get the geometry associated with a dm device
664 int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo)
672 * Set the geometry of a device.
674 int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo)
676 sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors;
678 if (geo->start > sz) {
679 DMWARN("Start sector is beyond the geometry limits.");
688 /*-----------------------------------------------------------------
690 * A more elegant soln is in the works that uses the queue
691 * merge fn, unfortunately there are a couple of changes to
692 * the block layer that I want to make for this. So in the
693 * interests of getting something for people to use I give
694 * you this clearly demarcated crap.
695 *---------------------------------------------------------------*/
697 static int __noflush_suspending(struct mapped_device *md)
699 return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
703 * Decrements the number of outstanding ios that a bio has been
704 * cloned into, completing the original io if necc.
706 static void dec_pending(struct dm_io *io, int error)
711 struct mapped_device *md = io->md;
713 /* Push-back supersedes any I/O errors */
714 if (unlikely(error)) {
715 spin_lock_irqsave(&io->endio_lock, flags);
716 if (!(io->error > 0 && __noflush_suspending(md)))
718 spin_unlock_irqrestore(&io->endio_lock, flags);
721 if (atomic_dec_and_test(&io->io_count)) {
722 if (io->error == DM_ENDIO_REQUEUE) {
724 * Target requested pushing back the I/O.
726 spin_lock_irqsave(&md->deferred_lock, flags);
727 if (__noflush_suspending(md))
728 bio_list_add_head(&md->deferred, io->bio);
730 /* noflush suspend was interrupted. */
732 spin_unlock_irqrestore(&md->deferred_lock, flags);
735 io_error = io->error;
740 if (io_error == DM_ENDIO_REQUEUE)
743 if ((bio->bi_rw & REQ_FLUSH) && bio->bi_iter.bi_size) {
745 * Preflush done for flush with data, reissue
748 bio->bi_rw &= ~REQ_FLUSH;
751 /* done with normal IO or empty flush */
752 trace_block_bio_complete(md->queue, bio, io_error);
753 bio_endio(bio, io_error);
758 static void disable_write_same(struct mapped_device *md)
760 struct queue_limits *limits = dm_get_queue_limits(md);
762 /* device doesn't really support WRITE SAME, disable it */
763 limits->max_write_same_sectors = 0;
766 static void clone_endio(struct bio *bio, int error)
769 struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone);
770 struct dm_io *io = tio->io;
771 struct mapped_device *md = tio->io->md;
772 dm_endio_fn endio = tio->ti->type->end_io;
774 if (!bio_flagged(bio, BIO_UPTODATE) && !error)
778 r = endio(tio->ti, bio, error);
779 if (r < 0 || r == DM_ENDIO_REQUEUE)
781 * error and requeue request are handled
785 else if (r == DM_ENDIO_INCOMPLETE)
786 /* The target will handle the io */
789 DMWARN("unimplemented target endio return value: %d", r);
794 if (unlikely(r == -EREMOTEIO && (bio->bi_rw & REQ_WRITE_SAME) &&
795 !bdev_get_queue(bio->bi_bdev)->limits.max_write_same_sectors))
796 disable_write_same(md);
799 dec_pending(io, error);
803 * Partial completion handling for request-based dm
805 static void end_clone_bio(struct bio *clone, int error)
807 struct dm_rq_clone_bio_info *info =
808 container_of(clone, struct dm_rq_clone_bio_info, clone);
809 struct dm_rq_target_io *tio = info->tio;
810 struct bio *bio = info->orig;
811 unsigned int nr_bytes = info->orig->bi_iter.bi_size;
817 * An error has already been detected on the request.
818 * Once error occurred, just let clone->end_io() handle
824 * Don't notice the error to the upper layer yet.
825 * The error handling decision is made by the target driver,
826 * when the request is completed.
833 * I/O for the bio successfully completed.
834 * Notice the data completion to the upper layer.
838 * bios are processed from the head of the list.
839 * So the completing bio should always be rq->bio.
840 * If it's not, something wrong is happening.
842 if (tio->orig->bio != bio)
843 DMERR("bio completion is going in the middle of the request");
846 * Update the original request.
847 * Do not use blk_end_request() here, because it may complete
848 * the original request before the clone, and break the ordering.
850 blk_update_request(tio->orig, 0, nr_bytes);
854 * Don't touch any member of the md after calling this function because
855 * the md may be freed in dm_put() at the end of this function.
856 * Or do dm_get() before calling this function and dm_put() later.
858 static void rq_completed(struct mapped_device *md, int rw, int run_queue)
860 atomic_dec(&md->pending[rw]);
862 /* nudge anyone waiting on suspend queue */
863 if (!md_in_flight(md))
867 * Run this off this callpath, as drivers could invoke end_io while
868 * inside their request_fn (and holding the queue lock). Calling
869 * back into ->request_fn() could deadlock attempting to grab the
873 blk_run_queue_async(md->queue);
876 * dm_put() must be at the end of this function. See the comment above
881 static void free_rq_clone(struct request *clone)
883 struct dm_rq_target_io *tio = clone->end_io_data;
885 blk_rq_unprep_clone(clone);
890 * Complete the clone and the original request.
891 * Must be called without queue lock.
893 static void dm_end_request(struct request *clone, int error)
895 int rw = rq_data_dir(clone);
896 struct dm_rq_target_io *tio = clone->end_io_data;
897 struct mapped_device *md = tio->md;
898 struct request *rq = tio->orig;
900 if (rq->cmd_type == REQ_TYPE_BLOCK_PC) {
901 rq->errors = clone->errors;
902 rq->resid_len = clone->resid_len;
906 * We are using the sense buffer of the original
908 * So setting the length of the sense data is enough.
910 rq->sense_len = clone->sense_len;
913 free_rq_clone(clone);
914 blk_end_request_all(rq, error);
915 rq_completed(md, rw, true);
918 static void dm_unprep_request(struct request *rq)
920 struct request *clone = rq->special;
923 rq->cmd_flags &= ~REQ_DONTPREP;
925 free_rq_clone(clone);
929 * Requeue the original request of a clone.
931 void dm_requeue_unmapped_request(struct request *clone)
933 int rw = rq_data_dir(clone);
934 struct dm_rq_target_io *tio = clone->end_io_data;
935 struct mapped_device *md = tio->md;
936 struct request *rq = tio->orig;
937 struct request_queue *q = rq->q;
940 dm_unprep_request(rq);
942 spin_lock_irqsave(q->queue_lock, flags);
943 blk_requeue_request(q, rq);
944 spin_unlock_irqrestore(q->queue_lock, flags);
946 rq_completed(md, rw, 0);
948 EXPORT_SYMBOL_GPL(dm_requeue_unmapped_request);
950 static void __stop_queue(struct request_queue *q)
955 static void stop_queue(struct request_queue *q)
959 spin_lock_irqsave(q->queue_lock, flags);
961 spin_unlock_irqrestore(q->queue_lock, flags);
964 static void __start_queue(struct request_queue *q)
966 if (blk_queue_stopped(q))
970 static void start_queue(struct request_queue *q)
974 spin_lock_irqsave(q->queue_lock, flags);
976 spin_unlock_irqrestore(q->queue_lock, flags);
979 static void dm_done(struct request *clone, int error, bool mapped)
982 struct dm_rq_target_io *tio = clone->end_io_data;
983 dm_request_endio_fn rq_end_io = NULL;
986 rq_end_io = tio->ti->type->rq_end_io;
988 if (mapped && rq_end_io)
989 r = rq_end_io(tio->ti, clone, error, &tio->info);
992 if (unlikely(r == -EREMOTEIO && (clone->cmd_flags & REQ_WRITE_SAME) &&
993 !clone->q->limits.max_write_same_sectors))
994 disable_write_same(tio->md);
997 /* The target wants to complete the I/O */
998 dm_end_request(clone, r);
999 else if (r == DM_ENDIO_INCOMPLETE)
1000 /* The target will handle the I/O */
1002 else if (r == DM_ENDIO_REQUEUE)
1003 /* The target wants to requeue the I/O */
1004 dm_requeue_unmapped_request(clone);
1006 DMWARN("unimplemented target endio return value: %d", r);
1012 * Request completion handler for request-based dm
1014 static void dm_softirq_done(struct request *rq)
1017 struct request *clone = rq->completion_data;
1018 struct dm_rq_target_io *tio = clone->end_io_data;
1020 if (rq->cmd_flags & REQ_FAILED)
1023 dm_done(clone, tio->error, mapped);
1027 * Complete the clone and the original request with the error status
1028 * through softirq context.
1030 static void dm_complete_request(struct request *clone, int error)
1032 struct dm_rq_target_io *tio = clone->end_io_data;
1033 struct request *rq = tio->orig;
1036 rq->completion_data = clone;
1037 blk_complete_request(rq);
1041 * Complete the not-mapped clone and the original request with the error status
1042 * through softirq context.
1043 * Target's rq_end_io() function isn't called.
1044 * This may be used when the target's map_rq() function fails.
1046 void dm_kill_unmapped_request(struct request *clone, int error)
1048 struct dm_rq_target_io *tio = clone->end_io_data;
1049 struct request *rq = tio->orig;
1051 rq->cmd_flags |= REQ_FAILED;
1052 dm_complete_request(clone, error);
1054 EXPORT_SYMBOL_GPL(dm_kill_unmapped_request);
1057 * Called with the queue lock held
1059 static void end_clone_request(struct request *clone, int error)
1062 * For just cleaning up the information of the queue in which
1063 * the clone was dispatched.
1064 * The clone is *NOT* freed actually here because it is alloced from
1065 * dm own mempool and REQ_ALLOCED isn't set in clone->cmd_flags.
1067 __blk_put_request(clone->q, clone);
1070 * Actual request completion is done in a softirq context which doesn't
1071 * hold the queue lock. Otherwise, deadlock could occur because:
1072 * - another request may be submitted by the upper level driver
1073 * of the stacking during the completion
1074 * - the submission which requires queue lock may be done
1075 * against this queue
1077 dm_complete_request(clone, error);
1081 * Return maximum size of I/O possible at the supplied sector up to the current
1084 static sector_t max_io_len_target_boundary(sector_t sector, struct dm_target *ti)
1086 sector_t target_offset = dm_target_offset(ti, sector);
1088 return ti->len - target_offset;
1091 static sector_t max_io_len(sector_t sector, struct dm_target *ti)
1093 sector_t len = max_io_len_target_boundary(sector, ti);
1094 sector_t offset, max_len;
1097 * Does the target need to split even further?
1099 if (ti->max_io_len) {
1100 offset = dm_target_offset(ti, sector);
1101 if (unlikely(ti->max_io_len & (ti->max_io_len - 1)))
1102 max_len = sector_div(offset, ti->max_io_len);
1104 max_len = offset & (ti->max_io_len - 1);
1105 max_len = ti->max_io_len - max_len;
1114 int dm_set_target_max_io_len(struct dm_target *ti, sector_t len)
1116 if (len > UINT_MAX) {
1117 DMERR("Specified maximum size of target IO (%llu) exceeds limit (%u)",
1118 (unsigned long long)len, UINT_MAX);
1119 ti->error = "Maximum size of target IO is too large";
1123 ti->max_io_len = (uint32_t) len;
1127 EXPORT_SYMBOL_GPL(dm_set_target_max_io_len);
1130 * A target may call dm_accept_partial_bio only from the map routine. It is
1131 * allowed for all bio types except REQ_FLUSH.
1133 * dm_accept_partial_bio informs the dm that the target only wants to process
1134 * additional n_sectors sectors of the bio and the rest of the data should be
1135 * sent in a next bio.
1137 * A diagram that explains the arithmetics:
1138 * +--------------------+---------------+-------+
1140 * +--------------------+---------------+-------+
1142 * <-------------- *tio->len_ptr --------------->
1143 * <------- bi_size ------->
1146 * Region 1 was already iterated over with bio_advance or similar function.
1147 * (it may be empty if the target doesn't use bio_advance)
1148 * Region 2 is the remaining bio size that the target wants to process.
1149 * (it may be empty if region 1 is non-empty, although there is no reason
1151 * The target requires that region 3 is to be sent in the next bio.
1153 * If the target wants to receive multiple copies of the bio (via num_*bios, etc),
1154 * the partially processed part (the sum of regions 1+2) must be the same for all
1155 * copies of the bio.
1157 void dm_accept_partial_bio(struct bio *bio, unsigned n_sectors)
1159 struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone);
1160 unsigned bi_size = bio->bi_iter.bi_size >> SECTOR_SHIFT;
1161 BUG_ON(bio->bi_rw & REQ_FLUSH);
1162 BUG_ON(bi_size > *tio->len_ptr);
1163 BUG_ON(n_sectors > bi_size);
1164 *tio->len_ptr -= bi_size - n_sectors;
1165 bio->bi_iter.bi_size = n_sectors << SECTOR_SHIFT;
1167 EXPORT_SYMBOL_GPL(dm_accept_partial_bio);
1169 static void __map_bio(struct dm_target_io *tio)
1173 struct mapped_device *md;
1174 struct bio *clone = &tio->clone;
1175 struct dm_target *ti = tio->ti;
1177 clone->bi_end_io = clone_endio;
1180 * Map the clone. If r == 0 we don't need to do
1181 * anything, the target has assumed ownership of
1184 atomic_inc(&tio->io->io_count);
1185 sector = clone->bi_iter.bi_sector;
1186 r = ti->type->map(ti, clone);
1187 if (r == DM_MAPIO_REMAPPED) {
1188 /* the bio has been remapped so dispatch it */
1190 trace_block_bio_remap(bdev_get_queue(clone->bi_bdev), clone,
1191 tio->io->bio->bi_bdev->bd_dev, sector);
1193 generic_make_request(clone);
1194 } else if (r < 0 || r == DM_MAPIO_REQUEUE) {
1195 /* error the io and bail out, or requeue it if needed */
1197 dec_pending(tio->io, r);
1200 DMWARN("unimplemented target map return value: %d", r);
1206 struct mapped_device *md;
1207 struct dm_table *map;
1211 unsigned sector_count;
1214 static void bio_setup_sector(struct bio *bio, sector_t sector, unsigned len)
1216 bio->bi_iter.bi_sector = sector;
1217 bio->bi_iter.bi_size = to_bytes(len);
1221 * Creates a bio that consists of range of complete bvecs.
1223 static void clone_bio(struct dm_target_io *tio, struct bio *bio,
1224 sector_t sector, unsigned len)
1226 struct bio *clone = &tio->clone;
1228 __bio_clone_fast(clone, bio);
1230 if (bio_integrity(bio))
1231 bio_integrity_clone(clone, bio, GFP_NOIO);
1233 bio_advance(clone, to_bytes(sector - clone->bi_iter.bi_sector));
1234 clone->bi_iter.bi_size = to_bytes(len);
1236 if (bio_integrity(bio))
1237 bio_integrity_trim(clone, 0, len);
1240 static struct dm_target_io *alloc_tio(struct clone_info *ci,
1241 struct dm_target *ti, int nr_iovecs,
1242 unsigned target_bio_nr)
1244 struct dm_target_io *tio;
1247 clone = bio_alloc_bioset(GFP_NOIO, nr_iovecs, ci->md->bs);
1248 tio = container_of(clone, struct dm_target_io, clone);
1252 tio->target_bio_nr = target_bio_nr;
1257 static void __clone_and_map_simple_bio(struct clone_info *ci,
1258 struct dm_target *ti,
1259 unsigned target_bio_nr, unsigned *len)
1261 struct dm_target_io *tio = alloc_tio(ci, ti, ci->bio->bi_max_vecs, target_bio_nr);
1262 struct bio *clone = &tio->clone;
1267 * Discard requests require the bio's inline iovecs be initialized.
1268 * ci->bio->bi_max_vecs is BIO_INLINE_VECS anyway, for both flush
1269 * and discard, so no need for concern about wasted bvec allocations.
1271 __bio_clone_fast(clone, ci->bio);
1273 bio_setup_sector(clone, ci->sector, *len);
1278 static void __send_duplicate_bios(struct clone_info *ci, struct dm_target *ti,
1279 unsigned num_bios, unsigned *len)
1281 unsigned target_bio_nr;
1283 for (target_bio_nr = 0; target_bio_nr < num_bios; target_bio_nr++)
1284 __clone_and_map_simple_bio(ci, ti, target_bio_nr, len);
1287 static int __send_empty_flush(struct clone_info *ci)
1289 unsigned target_nr = 0;
1290 struct dm_target *ti;
1292 BUG_ON(bio_has_data(ci->bio));
1293 while ((ti = dm_table_get_target(ci->map, target_nr++)))
1294 __send_duplicate_bios(ci, ti, ti->num_flush_bios, NULL);
1299 static void __clone_and_map_data_bio(struct clone_info *ci, struct dm_target *ti,
1300 sector_t sector, unsigned *len)
1302 struct bio *bio = ci->bio;
1303 struct dm_target_io *tio;
1304 unsigned target_bio_nr;
1305 unsigned num_target_bios = 1;
1308 * Does the target want to receive duplicate copies of the bio?
1310 if (bio_data_dir(bio) == WRITE && ti->num_write_bios)
1311 num_target_bios = ti->num_write_bios(ti, bio);
1313 for (target_bio_nr = 0; target_bio_nr < num_target_bios; target_bio_nr++) {
1314 tio = alloc_tio(ci, ti, 0, target_bio_nr);
1316 clone_bio(tio, bio, sector, *len);
1321 typedef unsigned (*get_num_bios_fn)(struct dm_target *ti);
1323 static unsigned get_num_discard_bios(struct dm_target *ti)
1325 return ti->num_discard_bios;
1328 static unsigned get_num_write_same_bios(struct dm_target *ti)
1330 return ti->num_write_same_bios;
1333 typedef bool (*is_split_required_fn)(struct dm_target *ti);
1335 static bool is_split_required_for_discard(struct dm_target *ti)
1337 return ti->split_discard_bios;
1340 static int __send_changing_extent_only(struct clone_info *ci,
1341 get_num_bios_fn get_num_bios,
1342 is_split_required_fn is_split_required)
1344 struct dm_target *ti;
1349 ti = dm_table_find_target(ci->map, ci->sector);
1350 if (!dm_target_is_valid(ti))
1354 * Even though the device advertised support for this type of
1355 * request, that does not mean every target supports it, and
1356 * reconfiguration might also have changed that since the
1357 * check was performed.
1359 num_bios = get_num_bios ? get_num_bios(ti) : 0;
1363 if (is_split_required && !is_split_required(ti))
1364 len = min((sector_t)ci->sector_count, max_io_len_target_boundary(ci->sector, ti));
1366 len = min((sector_t)ci->sector_count, max_io_len(ci->sector, ti));
1368 __send_duplicate_bios(ci, ti, num_bios, &len);
1371 } while (ci->sector_count -= len);
1376 static int __send_discard(struct clone_info *ci)
1378 return __send_changing_extent_only(ci, get_num_discard_bios,
1379 is_split_required_for_discard);
1382 static int __send_write_same(struct clone_info *ci)
1384 return __send_changing_extent_only(ci, get_num_write_same_bios, NULL);
1388 * Select the correct strategy for processing a non-flush bio.
1390 static int __split_and_process_non_flush(struct clone_info *ci)
1392 struct bio *bio = ci->bio;
1393 struct dm_target *ti;
1396 if (unlikely(bio->bi_rw & REQ_DISCARD))
1397 return __send_discard(ci);
1398 else if (unlikely(bio->bi_rw & REQ_WRITE_SAME))
1399 return __send_write_same(ci);
1401 ti = dm_table_find_target(ci->map, ci->sector);
1402 if (!dm_target_is_valid(ti))
1405 len = min_t(sector_t, max_io_len(ci->sector, ti), ci->sector_count);
1407 __clone_and_map_data_bio(ci, ti, ci->sector, &len);
1410 ci->sector_count -= len;
1416 * Entry point to split a bio into clones and submit them to the targets.
1418 static void __split_and_process_bio(struct mapped_device *md,
1419 struct dm_table *map, struct bio *bio)
1421 struct clone_info ci;
1424 if (unlikely(!map)) {
1431 ci.io = alloc_io(md);
1433 atomic_set(&ci.io->io_count, 1);
1436 spin_lock_init(&ci.io->endio_lock);
1437 ci.sector = bio->bi_iter.bi_sector;
1439 start_io_acct(ci.io);
1441 if (bio->bi_rw & REQ_FLUSH) {
1442 ci.bio = &ci.md->flush_bio;
1443 ci.sector_count = 0;
1444 error = __send_empty_flush(&ci);
1445 /* dec_pending submits any data associated with flush */
1448 ci.sector_count = bio_sectors(bio);
1449 while (ci.sector_count && !error)
1450 error = __split_and_process_non_flush(&ci);
1453 /* drop the extra reference count */
1454 dec_pending(ci.io, error);
1456 /*-----------------------------------------------------------------
1458 *---------------------------------------------------------------*/
1460 static int dm_merge_bvec(struct request_queue *q,
1461 struct bvec_merge_data *bvm,
1462 struct bio_vec *biovec)
1464 struct mapped_device *md = q->queuedata;
1465 struct dm_table *map = dm_get_live_table_fast(md);
1466 struct dm_target *ti;
1467 sector_t max_sectors;
1473 ti = dm_table_find_target(map, bvm->bi_sector);
1474 if (!dm_target_is_valid(ti))
1478 * Find maximum amount of I/O that won't need splitting
1480 max_sectors = min(max_io_len(bvm->bi_sector, ti),
1481 (sector_t) BIO_MAX_SECTORS);
1482 max_size = (max_sectors << SECTOR_SHIFT) - bvm->bi_size;
1487 * merge_bvec_fn() returns number of bytes
1488 * it can accept at this offset
1489 * max is precomputed maximal io size
1491 if (max_size && ti->type->merge)
1492 max_size = ti->type->merge(ti, bvm, biovec, max_size);
1494 * If the target doesn't support merge method and some of the devices
1495 * provided their merge_bvec method (we know this by looking at
1496 * queue_max_hw_sectors), then we can't allow bios with multiple vector
1497 * entries. So always set max_size to 0, and the code below allows
1500 else if (queue_max_hw_sectors(q) <= PAGE_SIZE >> 9)
1505 dm_put_live_table_fast(md);
1507 * Always allow an entire first page
1509 if (max_size <= biovec->bv_len && !(bvm->bi_size >> SECTOR_SHIFT))
1510 max_size = biovec->bv_len;
1516 * The request function that just remaps the bio built up by
1519 static void _dm_request(struct request_queue *q, struct bio *bio)
1521 int rw = bio_data_dir(bio);
1522 struct mapped_device *md = q->queuedata;
1525 struct dm_table *map;
1527 map = dm_get_live_table(md, &srcu_idx);
1529 cpu = part_stat_lock();
1530 part_stat_inc(cpu, &dm_disk(md)->part0, ios[rw]);
1531 part_stat_add(cpu, &dm_disk(md)->part0, sectors[rw], bio_sectors(bio));
1534 /* if we're suspended, we have to queue this io for later */
1535 if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))) {
1536 dm_put_live_table(md, srcu_idx);
1538 if (bio_rw(bio) != READA)
1545 __split_and_process_bio(md, map, bio);
1546 dm_put_live_table(md, srcu_idx);
1550 int dm_request_based(struct mapped_device *md)
1552 return blk_queue_stackable(md->queue);
1555 static void dm_request(struct request_queue *q, struct bio *bio)
1557 struct mapped_device *md = q->queuedata;
1559 if (dm_request_based(md))
1560 blk_queue_bio(q, bio);
1562 _dm_request(q, bio);
1565 void dm_dispatch_request(struct request *rq)
1569 if (blk_queue_io_stat(rq->q))
1570 rq->cmd_flags |= REQ_IO_STAT;
1572 rq->start_time = jiffies;
1573 r = blk_insert_cloned_request(rq->q, rq);
1575 dm_complete_request(rq, r);
1577 EXPORT_SYMBOL_GPL(dm_dispatch_request);
1579 static int dm_rq_bio_constructor(struct bio *bio, struct bio *bio_orig,
1582 struct dm_rq_target_io *tio = data;
1583 struct dm_rq_clone_bio_info *info =
1584 container_of(bio, struct dm_rq_clone_bio_info, clone);
1586 info->orig = bio_orig;
1588 bio->bi_end_io = end_clone_bio;
1593 static int setup_clone(struct request *clone, struct request *rq,
1594 struct dm_rq_target_io *tio)
1598 r = blk_rq_prep_clone(clone, rq, tio->md->bs, GFP_ATOMIC,
1599 dm_rq_bio_constructor, tio);
1603 clone->cmd = rq->cmd;
1604 clone->cmd_len = rq->cmd_len;
1605 clone->sense = rq->sense;
1606 clone->buffer = rq->buffer;
1607 clone->end_io = end_clone_request;
1608 clone->end_io_data = tio;
1613 static struct request *clone_rq(struct request *rq, struct mapped_device *md,
1616 struct request *clone;
1617 struct dm_rq_target_io *tio;
1619 tio = alloc_rq_tio(md, gfp_mask);
1627 memset(&tio->info, 0, sizeof(tio->info));
1629 clone = &tio->clone;
1630 if (setup_clone(clone, rq, tio)) {
1640 * Called with the queue lock held.
1642 static int dm_prep_fn(struct request_queue *q, struct request *rq)
1644 struct mapped_device *md = q->queuedata;
1645 struct request *clone;
1647 if (unlikely(rq->special)) {
1648 DMWARN("Already has something in rq->special.");
1649 return BLKPREP_KILL;
1652 clone = clone_rq(rq, md, GFP_ATOMIC);
1654 return BLKPREP_DEFER;
1656 rq->special = clone;
1657 rq->cmd_flags |= REQ_DONTPREP;
1664 * 0 : the request has been processed (not requeued)
1665 * !0 : the request has been requeued
1667 static int map_request(struct dm_target *ti, struct request *clone,
1668 struct mapped_device *md)
1670 int r, requeued = 0;
1671 struct dm_rq_target_io *tio = clone->end_io_data;
1674 r = ti->type->map_rq(ti, clone, &tio->info);
1676 case DM_MAPIO_SUBMITTED:
1677 /* The target has taken the I/O to submit by itself later */
1679 case DM_MAPIO_REMAPPED:
1680 /* The target has remapped the I/O so dispatch it */
1681 trace_block_rq_remap(clone->q, clone, disk_devt(dm_disk(md)),
1682 blk_rq_pos(tio->orig));
1683 dm_dispatch_request(clone);
1685 case DM_MAPIO_REQUEUE:
1686 /* The target wants to requeue the I/O */
1687 dm_requeue_unmapped_request(clone);
1692 DMWARN("unimplemented target map return value: %d", r);
1696 /* The target wants to complete the I/O */
1697 dm_kill_unmapped_request(clone, r);
1704 static struct request *dm_start_request(struct mapped_device *md, struct request *orig)
1706 struct request *clone;
1708 blk_start_request(orig);
1709 clone = orig->special;
1710 atomic_inc(&md->pending[rq_data_dir(clone)]);
1713 * Hold the md reference here for the in-flight I/O.
1714 * We can't rely on the reference count by device opener,
1715 * because the device may be closed during the request completion
1716 * when all bios are completed.
1717 * See the comment in rq_completed() too.
1725 * q->request_fn for request-based dm.
1726 * Called with the queue lock held.
1728 static void dm_request_fn(struct request_queue *q)
1730 struct mapped_device *md = q->queuedata;
1732 struct dm_table *map = dm_get_live_table(md, &srcu_idx);
1733 struct dm_target *ti;
1734 struct request *rq, *clone;
1738 * For suspend, check blk_queue_stopped() and increment
1739 * ->pending within a single queue_lock not to increment the
1740 * number of in-flight I/Os after the queue is stopped in
1743 while (!blk_queue_stopped(q)) {
1744 rq = blk_peek_request(q);
1748 /* always use block 0 to find the target for flushes for now */
1750 if (!(rq->cmd_flags & REQ_FLUSH))
1751 pos = blk_rq_pos(rq);
1753 ti = dm_table_find_target(map, pos);
1754 if (!dm_target_is_valid(ti)) {
1756 * Must perform setup, that dm_done() requires,
1757 * before calling dm_kill_unmapped_request
1759 DMERR_LIMIT("request attempted access beyond the end of device");
1760 clone = dm_start_request(md, rq);
1761 dm_kill_unmapped_request(clone, -EIO);
1765 if (ti->type->busy && ti->type->busy(ti))
1768 clone = dm_start_request(md, rq);
1770 spin_unlock(q->queue_lock);
1771 if (map_request(ti, clone, md))
1774 BUG_ON(!irqs_disabled());
1775 spin_lock(q->queue_lock);
1781 BUG_ON(!irqs_disabled());
1782 spin_lock(q->queue_lock);
1785 blk_delay_queue(q, HZ / 10);
1787 dm_put_live_table(md, srcu_idx);
1790 int dm_underlying_device_busy(struct request_queue *q)
1792 return blk_lld_busy(q);
1794 EXPORT_SYMBOL_GPL(dm_underlying_device_busy);
1796 static int dm_lld_busy(struct request_queue *q)
1799 struct mapped_device *md = q->queuedata;
1800 struct dm_table *map = dm_get_live_table_fast(md);
1802 if (!map || test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))
1805 r = dm_table_any_busy_target(map);
1807 dm_put_live_table_fast(md);
1812 static int dm_any_congested(void *congested_data, int bdi_bits)
1815 struct mapped_device *md = congested_data;
1816 struct dm_table *map;
1818 if (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
1819 map = dm_get_live_table_fast(md);
1822 * Request-based dm cares about only own queue for
1823 * the query about congestion status of request_queue
1825 if (dm_request_based(md))
1826 r = md->queue->backing_dev_info.state &
1829 r = dm_table_any_congested(map, bdi_bits);
1831 dm_put_live_table_fast(md);
1837 /*-----------------------------------------------------------------
1838 * An IDR is used to keep track of allocated minor numbers.
1839 *---------------------------------------------------------------*/
1840 static void free_minor(int minor)
1842 spin_lock(&_minor_lock);
1843 idr_remove(&_minor_idr, minor);
1844 spin_unlock(&_minor_lock);
1848 * See if the device with a specific minor # is free.
1850 static int specific_minor(int minor)
1854 if (minor >= (1 << MINORBITS))
1857 idr_preload(GFP_KERNEL);
1858 spin_lock(&_minor_lock);
1860 r = idr_alloc(&_minor_idr, MINOR_ALLOCED, minor, minor + 1, GFP_NOWAIT);
1862 spin_unlock(&_minor_lock);
1865 return r == -ENOSPC ? -EBUSY : r;
1869 static int next_free_minor(int *minor)
1873 idr_preload(GFP_KERNEL);
1874 spin_lock(&_minor_lock);
1876 r = idr_alloc(&_minor_idr, MINOR_ALLOCED, 0, 1 << MINORBITS, GFP_NOWAIT);
1878 spin_unlock(&_minor_lock);
1886 static const struct block_device_operations dm_blk_dops;
1888 static void dm_wq_work(struct work_struct *work);
1890 static void dm_init_md_queue(struct mapped_device *md)
1893 * Request-based dm devices cannot be stacked on top of bio-based dm
1894 * devices. The type of this dm device has not been decided yet.
1895 * The type is decided at the first table loading time.
1896 * To prevent problematic device stacking, clear the queue flag
1897 * for request stacking support until then.
1899 * This queue is new, so no concurrency on the queue_flags.
1901 queue_flag_clear_unlocked(QUEUE_FLAG_STACKABLE, md->queue);
1903 md->queue->queuedata = md;
1904 md->queue->backing_dev_info.congested_fn = dm_any_congested;
1905 md->queue->backing_dev_info.congested_data = md;
1906 blk_queue_make_request(md->queue, dm_request);
1907 blk_queue_bounce_limit(md->queue, BLK_BOUNCE_ANY);
1908 blk_queue_merge_bvec(md->queue, dm_merge_bvec);
1912 * Allocate and initialise a blank device with a given minor.
1914 static struct mapped_device *alloc_dev(int minor)
1917 struct mapped_device *md = kzalloc(sizeof(*md), GFP_KERNEL);
1921 DMWARN("unable to allocate device, out of memory.");
1925 if (!try_module_get(THIS_MODULE))
1926 goto bad_module_get;
1928 /* get a minor number for the dev */
1929 if (minor == DM_ANY_MINOR)
1930 r = next_free_minor(&minor);
1932 r = specific_minor(minor);
1936 r = init_srcu_struct(&md->io_barrier);
1938 goto bad_io_barrier;
1940 md->type = DM_TYPE_NONE;
1941 mutex_init(&md->suspend_lock);
1942 mutex_init(&md->type_lock);
1943 spin_lock_init(&md->deferred_lock);
1944 atomic_set(&md->holders, 1);
1945 atomic_set(&md->open_count, 0);
1946 atomic_set(&md->event_nr, 0);
1947 atomic_set(&md->uevent_seq, 0);
1948 INIT_LIST_HEAD(&md->uevent_list);
1949 spin_lock_init(&md->uevent_lock);
1951 md->queue = blk_alloc_queue(GFP_KERNEL);
1955 dm_init_md_queue(md);
1957 md->disk = alloc_disk(1);
1961 atomic_set(&md->pending[0], 0);
1962 atomic_set(&md->pending[1], 0);
1963 init_waitqueue_head(&md->wait);
1964 INIT_WORK(&md->work, dm_wq_work);
1965 init_waitqueue_head(&md->eventq);
1966 init_completion(&md->kobj_holder.completion);
1968 md->disk->major = _major;
1969 md->disk->first_minor = minor;
1970 md->disk->fops = &dm_blk_dops;
1971 md->disk->queue = md->queue;
1972 md->disk->private_data = md;
1973 sprintf(md->disk->disk_name, "dm-%d", minor);
1975 format_dev_t(md->name, MKDEV(_major, minor));
1977 md->wq = alloc_workqueue("kdmflush", WQ_MEM_RECLAIM, 0);
1981 md->bdev = bdget_disk(md->disk, 0);
1985 bio_init(&md->flush_bio);
1986 md->flush_bio.bi_bdev = md->bdev;
1987 md->flush_bio.bi_rw = WRITE_FLUSH;
1989 dm_stats_init(&md->stats);
1991 /* Populate the mapping, nobody knows we exist yet */
1992 spin_lock(&_minor_lock);
1993 old_md = idr_replace(&_minor_idr, md, minor);
1994 spin_unlock(&_minor_lock);
1996 BUG_ON(old_md != MINOR_ALLOCED);
2001 destroy_workqueue(md->wq);
2003 del_gendisk(md->disk);
2006 blk_cleanup_queue(md->queue);
2008 cleanup_srcu_struct(&md->io_barrier);
2012 module_put(THIS_MODULE);
2018 static void unlock_fs(struct mapped_device *md);
2020 static void free_dev(struct mapped_device *md)
2022 int minor = MINOR(disk_devt(md->disk));
2026 destroy_workqueue(md->wq);
2028 mempool_destroy(md->io_pool);
2030 bioset_free(md->bs);
2031 blk_integrity_unregister(md->disk);
2032 del_gendisk(md->disk);
2033 cleanup_srcu_struct(&md->io_barrier);
2036 spin_lock(&_minor_lock);
2037 md->disk->private_data = NULL;
2038 spin_unlock(&_minor_lock);
2041 blk_cleanup_queue(md->queue);
2042 dm_stats_cleanup(&md->stats);
2043 module_put(THIS_MODULE);
2047 static void __bind_mempools(struct mapped_device *md, struct dm_table *t)
2049 struct dm_md_mempools *p = dm_table_get_md_mempools(t);
2051 if (md->io_pool && md->bs) {
2052 /* The md already has necessary mempools. */
2053 if (dm_table_get_type(t) == DM_TYPE_BIO_BASED) {
2055 * Reload bioset because front_pad may have changed
2056 * because a different table was loaded.
2058 bioset_free(md->bs);
2061 } else if (dm_table_get_type(t) == DM_TYPE_REQUEST_BASED) {
2063 * There's no need to reload with request-based dm
2064 * because the size of front_pad doesn't change.
2065 * Note for future: If you are to reload bioset,
2066 * prep-ed requests in the queue may refer
2067 * to bio from the old bioset, so you must walk
2068 * through the queue to unprep.
2074 BUG_ON(!p || md->io_pool || md->bs);
2076 md->io_pool = p->io_pool;
2082 /* mempool bind completed, now no need any mempools in the table */
2083 dm_table_free_md_mempools(t);
2087 * Bind a table to the device.
2089 static void event_callback(void *context)
2091 unsigned long flags;
2093 struct mapped_device *md = (struct mapped_device *) context;
2095 spin_lock_irqsave(&md->uevent_lock, flags);
2096 list_splice_init(&md->uevent_list, &uevents);
2097 spin_unlock_irqrestore(&md->uevent_lock, flags);
2099 dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj);
2101 atomic_inc(&md->event_nr);
2102 wake_up(&md->eventq);
2106 * Protected by md->suspend_lock obtained by dm_swap_table().
2108 static void __set_size(struct mapped_device *md, sector_t size)
2110 set_capacity(md->disk, size);
2112 i_size_write(md->bdev->bd_inode, (loff_t)size << SECTOR_SHIFT);
2116 * Return 1 if the queue has a compulsory merge_bvec_fn function.
2118 * If this function returns 0, then the device is either a non-dm
2119 * device without a merge_bvec_fn, or it is a dm device that is
2120 * able to split any bios it receives that are too big.
2122 int dm_queue_merge_is_compulsory(struct request_queue *q)
2124 struct mapped_device *dev_md;
2126 if (!q->merge_bvec_fn)
2129 if (q->make_request_fn == dm_request) {
2130 dev_md = q->queuedata;
2131 if (test_bit(DMF_MERGE_IS_OPTIONAL, &dev_md->flags))
2138 static int dm_device_merge_is_compulsory(struct dm_target *ti,
2139 struct dm_dev *dev, sector_t start,
2140 sector_t len, void *data)
2142 struct block_device *bdev = dev->bdev;
2143 struct request_queue *q = bdev_get_queue(bdev);
2145 return dm_queue_merge_is_compulsory(q);
2149 * Return 1 if it is acceptable to ignore merge_bvec_fn based
2150 * on the properties of the underlying devices.
2152 static int dm_table_merge_is_optional(struct dm_table *table)
2155 struct dm_target *ti;
2157 while (i < dm_table_get_num_targets(table)) {
2158 ti = dm_table_get_target(table, i++);
2160 if (ti->type->iterate_devices &&
2161 ti->type->iterate_devices(ti, dm_device_merge_is_compulsory, NULL))
2169 * Returns old map, which caller must destroy.
2171 static struct dm_table *__bind(struct mapped_device *md, struct dm_table *t,
2172 struct queue_limits *limits)
2174 struct dm_table *old_map;
2175 struct request_queue *q = md->queue;
2177 int merge_is_optional;
2179 size = dm_table_get_size(t);
2182 * Wipe any geometry if the size of the table changed.
2184 if (size != dm_get_size(md))
2185 memset(&md->geometry, 0, sizeof(md->geometry));
2187 __set_size(md, size);
2189 dm_table_event_callback(t, event_callback, md);
2192 * The queue hasn't been stopped yet, if the old table type wasn't
2193 * for request-based during suspension. So stop it to prevent
2194 * I/O mapping before resume.
2195 * This must be done before setting the queue restrictions,
2196 * because request-based dm may be run just after the setting.
2198 if (dm_table_request_based(t) && !blk_queue_stopped(q))
2201 __bind_mempools(md, t);
2203 merge_is_optional = dm_table_merge_is_optional(t);
2206 rcu_assign_pointer(md->map, t);
2207 md->immutable_target_type = dm_table_get_immutable_target_type(t);
2209 dm_table_set_restrictions(t, q, limits);
2210 if (merge_is_optional)
2211 set_bit(DMF_MERGE_IS_OPTIONAL, &md->flags);
2213 clear_bit(DMF_MERGE_IS_OPTIONAL, &md->flags);
2220 * Returns unbound table for the caller to free.
2222 static struct dm_table *__unbind(struct mapped_device *md)
2224 struct dm_table *map = md->map;
2229 dm_table_event_callback(map, NULL, NULL);
2230 RCU_INIT_POINTER(md->map, NULL);
2237 * Constructor for a new device.
2239 int dm_create(int minor, struct mapped_device **result)
2241 struct mapped_device *md;
2243 md = alloc_dev(minor);
2254 * Functions to manage md->type.
2255 * All are required to hold md->type_lock.
2257 void dm_lock_md_type(struct mapped_device *md)
2259 mutex_lock(&md->type_lock);
2262 void dm_unlock_md_type(struct mapped_device *md)
2264 mutex_unlock(&md->type_lock);
2267 void dm_set_md_type(struct mapped_device *md, unsigned type)
2269 BUG_ON(!mutex_is_locked(&md->type_lock));
2273 unsigned dm_get_md_type(struct mapped_device *md)
2275 BUG_ON(!mutex_is_locked(&md->type_lock));
2279 struct target_type *dm_get_immutable_target_type(struct mapped_device *md)
2281 return md->immutable_target_type;
2285 * The queue_limits are only valid as long as you have a reference
2288 struct queue_limits *dm_get_queue_limits(struct mapped_device *md)
2290 BUG_ON(!atomic_read(&md->holders));
2291 return &md->queue->limits;
2293 EXPORT_SYMBOL_GPL(dm_get_queue_limits);
2296 * Fully initialize a request-based queue (->elevator, ->request_fn, etc).
2298 static int dm_init_request_based_queue(struct mapped_device *md)
2300 struct request_queue *q = NULL;
2302 if (md->queue->elevator)
2305 /* Fully initialize the queue */
2306 q = blk_init_allocated_queue(md->queue, dm_request_fn, NULL);
2311 dm_init_md_queue(md);
2312 blk_queue_softirq_done(md->queue, dm_softirq_done);
2313 blk_queue_prep_rq(md->queue, dm_prep_fn);
2314 blk_queue_lld_busy(md->queue, dm_lld_busy);
2316 elv_register_queue(md->queue);
2322 * Setup the DM device's queue based on md's type
2324 int dm_setup_md_queue(struct mapped_device *md)
2326 if ((dm_get_md_type(md) == DM_TYPE_REQUEST_BASED) &&
2327 !dm_init_request_based_queue(md)) {
2328 DMWARN("Cannot initialize queue for request-based mapped device");
2335 static struct mapped_device *dm_find_md(dev_t dev)
2337 struct mapped_device *md;
2338 unsigned minor = MINOR(dev);
2340 if (MAJOR(dev) != _major || minor >= (1 << MINORBITS))
2343 spin_lock(&_minor_lock);
2345 md = idr_find(&_minor_idr, minor);
2346 if (md && (md == MINOR_ALLOCED ||
2347 (MINOR(disk_devt(dm_disk(md))) != minor) ||
2348 dm_deleting_md(md) ||
2349 test_bit(DMF_FREEING, &md->flags))) {
2355 spin_unlock(&_minor_lock);
2360 struct mapped_device *dm_get_md(dev_t dev)
2362 struct mapped_device *md = dm_find_md(dev);
2369 EXPORT_SYMBOL_GPL(dm_get_md);
2371 void *dm_get_mdptr(struct mapped_device *md)
2373 return md->interface_ptr;
2376 void dm_set_mdptr(struct mapped_device *md, void *ptr)
2378 md->interface_ptr = ptr;
2381 void dm_get(struct mapped_device *md)
2383 atomic_inc(&md->holders);
2384 BUG_ON(test_bit(DMF_FREEING, &md->flags));
2387 const char *dm_device_name(struct mapped_device *md)
2391 EXPORT_SYMBOL_GPL(dm_device_name);
2393 static void __dm_destroy(struct mapped_device *md, bool wait)
2395 struct dm_table *map;
2400 spin_lock(&_minor_lock);
2401 map = dm_get_live_table(md, &srcu_idx);
2402 idr_replace(&_minor_idr, MINOR_ALLOCED, MINOR(disk_devt(dm_disk(md))));
2403 set_bit(DMF_FREEING, &md->flags);
2404 spin_unlock(&_minor_lock);
2406 if (!dm_suspended_md(md)) {
2407 dm_table_presuspend_targets(map);
2408 dm_table_postsuspend_targets(map);
2411 /* dm_put_live_table must be before msleep, otherwise deadlock is possible */
2412 dm_put_live_table(md, srcu_idx);
2415 * Rare, but there may be I/O requests still going to complete,
2416 * for example. Wait for all references to disappear.
2417 * No one should increment the reference count of the mapped_device,
2418 * after the mapped_device state becomes DMF_FREEING.
2421 while (atomic_read(&md->holders))
2423 else if (atomic_read(&md->holders))
2424 DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)",
2425 dm_device_name(md), atomic_read(&md->holders));
2428 dm_table_destroy(__unbind(md));
2432 void dm_destroy(struct mapped_device *md)
2434 __dm_destroy(md, true);
2437 void dm_destroy_immediate(struct mapped_device *md)
2439 __dm_destroy(md, false);
2442 void dm_put(struct mapped_device *md)
2444 atomic_dec(&md->holders);
2446 EXPORT_SYMBOL_GPL(dm_put);
2448 static int dm_wait_for_completion(struct mapped_device *md, int interruptible)
2451 DECLARE_WAITQUEUE(wait, current);
2453 add_wait_queue(&md->wait, &wait);
2456 set_current_state(interruptible);
2458 if (!md_in_flight(md))
2461 if (interruptible == TASK_INTERRUPTIBLE &&
2462 signal_pending(current)) {
2469 set_current_state(TASK_RUNNING);
2471 remove_wait_queue(&md->wait, &wait);
2477 * Process the deferred bios
2479 static void dm_wq_work(struct work_struct *work)
2481 struct mapped_device *md = container_of(work, struct mapped_device,
2485 struct dm_table *map;
2487 map = dm_get_live_table(md, &srcu_idx);
2489 while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
2490 spin_lock_irq(&md->deferred_lock);
2491 c = bio_list_pop(&md->deferred);
2492 spin_unlock_irq(&md->deferred_lock);
2497 if (dm_request_based(md))
2498 generic_make_request(c);
2500 __split_and_process_bio(md, map, c);
2503 dm_put_live_table(md, srcu_idx);
2506 static void dm_queue_flush(struct mapped_device *md)
2508 clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2509 smp_mb__after_clear_bit();
2510 queue_work(md->wq, &md->work);
2514 * Swap in a new table, returning the old one for the caller to destroy.
2516 struct dm_table *dm_swap_table(struct mapped_device *md, struct dm_table *table)
2518 struct dm_table *live_map = NULL, *map = ERR_PTR(-EINVAL);
2519 struct queue_limits limits;
2522 mutex_lock(&md->suspend_lock);
2524 /* device must be suspended */
2525 if (!dm_suspended_md(md))
2529 * If the new table has no data devices, retain the existing limits.
2530 * This helps multipath with queue_if_no_path if all paths disappear,
2531 * then new I/O is queued based on these limits, and then some paths
2534 if (dm_table_has_no_data_devices(table)) {
2535 live_map = dm_get_live_table_fast(md);
2537 limits = md->queue->limits;
2538 dm_put_live_table_fast(md);
2542 r = dm_calculate_queue_limits(table, &limits);
2549 map = __bind(md, table, &limits);
2552 mutex_unlock(&md->suspend_lock);
2557 * Functions to lock and unlock any filesystem running on the
2560 static int lock_fs(struct mapped_device *md)
2564 WARN_ON(md->frozen_sb);
2566 md->frozen_sb = freeze_bdev(md->bdev);
2567 if (IS_ERR(md->frozen_sb)) {
2568 r = PTR_ERR(md->frozen_sb);
2569 md->frozen_sb = NULL;
2573 set_bit(DMF_FROZEN, &md->flags);
2578 static void unlock_fs(struct mapped_device *md)
2580 if (!test_bit(DMF_FROZEN, &md->flags))
2583 thaw_bdev(md->bdev, md->frozen_sb);
2584 md->frozen_sb = NULL;
2585 clear_bit(DMF_FROZEN, &md->flags);
2589 * We need to be able to change a mapping table under a mounted
2590 * filesystem. For example we might want to move some data in
2591 * the background. Before the table can be swapped with
2592 * dm_bind_table, dm_suspend must be called to flush any in
2593 * flight bios and ensure that any further io gets deferred.
2596 * Suspend mechanism in request-based dm.
2598 * 1. Flush all I/Os by lock_fs() if needed.
2599 * 2. Stop dispatching any I/O by stopping the request_queue.
2600 * 3. Wait for all in-flight I/Os to be completed or requeued.
2602 * To abort suspend, start the request_queue.
2604 int dm_suspend(struct mapped_device *md, unsigned suspend_flags)
2606 struct dm_table *map = NULL;
2608 int do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG ? 1 : 0;
2609 int noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG ? 1 : 0;
2611 mutex_lock(&md->suspend_lock);
2613 if (dm_suspended_md(md)) {
2621 * DMF_NOFLUSH_SUSPENDING must be set before presuspend.
2622 * This flag is cleared before dm_suspend returns.
2625 set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2627 /* This does not get reverted if there's an error later. */
2628 dm_table_presuspend_targets(map);
2631 * Flush I/O to the device.
2632 * Any I/O submitted after lock_fs() may not be flushed.
2633 * noflush takes precedence over do_lockfs.
2634 * (lock_fs() flushes I/Os and waits for them to complete.)
2636 if (!noflush && do_lockfs) {
2643 * Here we must make sure that no processes are submitting requests
2644 * to target drivers i.e. no one may be executing
2645 * __split_and_process_bio. This is called from dm_request and
2648 * To get all processes out of __split_and_process_bio in dm_request,
2649 * we take the write lock. To prevent any process from reentering
2650 * __split_and_process_bio from dm_request and quiesce the thread
2651 * (dm_wq_work), we set BMF_BLOCK_IO_FOR_SUSPEND and call
2652 * flush_workqueue(md->wq).
2654 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2655 synchronize_srcu(&md->io_barrier);
2658 * Stop md->queue before flushing md->wq in case request-based
2659 * dm defers requests to md->wq from md->queue.
2661 if (dm_request_based(md))
2662 stop_queue(md->queue);
2664 flush_workqueue(md->wq);
2667 * At this point no more requests are entering target request routines.
2668 * We call dm_wait_for_completion to wait for all existing requests
2671 r = dm_wait_for_completion(md, TASK_INTERRUPTIBLE);
2674 clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2675 synchronize_srcu(&md->io_barrier);
2677 /* were we interrupted ? */
2681 if (dm_request_based(md))
2682 start_queue(md->queue);
2685 goto out_unlock; /* pushback list is already flushed, so skip flush */
2689 * If dm_wait_for_completion returned 0, the device is completely
2690 * quiescent now. There is no request-processing activity. All new
2691 * requests are being added to md->deferred list.
2694 set_bit(DMF_SUSPENDED, &md->flags);
2696 dm_table_postsuspend_targets(map);
2699 mutex_unlock(&md->suspend_lock);
2703 int dm_resume(struct mapped_device *md)
2706 struct dm_table *map = NULL;
2708 mutex_lock(&md->suspend_lock);
2709 if (!dm_suspended_md(md))
2713 if (!map || !dm_table_get_size(map))
2716 r = dm_table_resume_targets(map);
2723 * Flushing deferred I/Os must be done after targets are resumed
2724 * so that mapping of targets can work correctly.
2725 * Request-based dm is queueing the deferred I/Os in its request_queue.
2727 if (dm_request_based(md))
2728 start_queue(md->queue);
2732 clear_bit(DMF_SUSPENDED, &md->flags);
2736 mutex_unlock(&md->suspend_lock);
2742 * Internal suspend/resume works like userspace-driven suspend. It waits
2743 * until all bios finish and prevents issuing new bios to the target drivers.
2744 * It may be used only from the kernel.
2746 * Internal suspend holds md->suspend_lock, which prevents interaction with
2747 * userspace-driven suspend.
2750 void dm_internal_suspend(struct mapped_device *md)
2752 mutex_lock(&md->suspend_lock);
2753 if (dm_suspended_md(md))
2756 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2757 synchronize_srcu(&md->io_barrier);
2758 flush_workqueue(md->wq);
2759 dm_wait_for_completion(md, TASK_UNINTERRUPTIBLE);
2762 void dm_internal_resume(struct mapped_device *md)
2764 if (dm_suspended_md(md))
2770 mutex_unlock(&md->suspend_lock);
2773 /*-----------------------------------------------------------------
2774 * Event notification.
2775 *---------------------------------------------------------------*/
2776 int dm_kobject_uevent(struct mapped_device *md, enum kobject_action action,
2779 char udev_cookie[DM_COOKIE_LENGTH];
2780 char *envp[] = { udev_cookie, NULL };
2783 return kobject_uevent(&disk_to_dev(md->disk)->kobj, action);
2785 snprintf(udev_cookie, DM_COOKIE_LENGTH, "%s=%u",
2786 DM_COOKIE_ENV_VAR_NAME, cookie);
2787 return kobject_uevent_env(&disk_to_dev(md->disk)->kobj,
2792 uint32_t dm_next_uevent_seq(struct mapped_device *md)
2794 return atomic_add_return(1, &md->uevent_seq);
2797 uint32_t dm_get_event_nr(struct mapped_device *md)
2799 return atomic_read(&md->event_nr);
2802 int dm_wait_event(struct mapped_device *md, int event_nr)
2804 return wait_event_interruptible(md->eventq,
2805 (event_nr != atomic_read(&md->event_nr)));
2808 void dm_uevent_add(struct mapped_device *md, struct list_head *elist)
2810 unsigned long flags;
2812 spin_lock_irqsave(&md->uevent_lock, flags);
2813 list_add(elist, &md->uevent_list);
2814 spin_unlock_irqrestore(&md->uevent_lock, flags);
2818 * The gendisk is only valid as long as you have a reference
2821 struct gendisk *dm_disk(struct mapped_device *md)
2826 struct kobject *dm_kobject(struct mapped_device *md)
2828 return &md->kobj_holder.kobj;
2831 struct mapped_device *dm_get_from_kobject(struct kobject *kobj)
2833 struct mapped_device *md;
2835 md = container_of(kobj, struct mapped_device, kobj_holder.kobj);
2837 if (test_bit(DMF_FREEING, &md->flags) ||
2845 int dm_suspended_md(struct mapped_device *md)
2847 return test_bit(DMF_SUSPENDED, &md->flags);
2850 int dm_test_deferred_remove_flag(struct mapped_device *md)
2852 return test_bit(DMF_DEFERRED_REMOVE, &md->flags);
2855 int dm_suspended(struct dm_target *ti)
2857 return dm_suspended_md(dm_table_get_md(ti->table));
2859 EXPORT_SYMBOL_GPL(dm_suspended);
2861 int dm_noflush_suspending(struct dm_target *ti)
2863 return __noflush_suspending(dm_table_get_md(ti->table));
2865 EXPORT_SYMBOL_GPL(dm_noflush_suspending);
2867 struct dm_md_mempools *dm_alloc_md_mempools(unsigned type, unsigned integrity, unsigned per_bio_data_size)
2869 struct dm_md_mempools *pools = kzalloc(sizeof(*pools), GFP_KERNEL);
2870 struct kmem_cache *cachep;
2871 unsigned int pool_size;
2872 unsigned int front_pad;
2877 if (type == DM_TYPE_BIO_BASED) {
2879 pool_size = dm_get_reserved_bio_based_ios();
2880 front_pad = roundup(per_bio_data_size, __alignof__(struct dm_target_io)) + offsetof(struct dm_target_io, clone);
2881 } else if (type == DM_TYPE_REQUEST_BASED) {
2882 cachep = _rq_tio_cache;
2883 pool_size = dm_get_reserved_rq_based_ios();
2884 front_pad = offsetof(struct dm_rq_clone_bio_info, clone);
2885 /* per_bio_data_size is not used. See __bind_mempools(). */
2886 WARN_ON(per_bio_data_size != 0);
2890 pools->io_pool = mempool_create_slab_pool(pool_size, cachep);
2891 if (!pools->io_pool)
2894 pools->bs = bioset_create(pool_size, front_pad);
2898 if (integrity && bioset_integrity_create(pools->bs, pool_size))
2904 dm_free_md_mempools(pools);
2909 void dm_free_md_mempools(struct dm_md_mempools *pools)
2915 mempool_destroy(pools->io_pool);
2918 bioset_free(pools->bs);
2923 static const struct block_device_operations dm_blk_dops = {
2924 .open = dm_blk_open,
2925 .release = dm_blk_close,
2926 .ioctl = dm_blk_ioctl,
2927 .getgeo = dm_blk_getgeo,
2928 .owner = THIS_MODULE
2934 module_init(dm_init);
2935 module_exit(dm_exit);
2937 module_param(major, uint, 0);
2938 MODULE_PARM_DESC(major, "The major number of the device mapper");
2940 module_param(reserved_bio_based_ios, uint, S_IRUGO | S_IWUSR);
2941 MODULE_PARM_DESC(reserved_bio_based_ios, "Reserved IOs in bio-based mempools");
2943 module_param(reserved_rq_based_ios, uint, S_IRUGO | S_IWUSR);
2944 MODULE_PARM_DESC(reserved_rq_based_ios, "Reserved IOs in request-based mempools");
2946 MODULE_DESCRIPTION(DM_NAME " driver");
2947 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
2948 MODULE_LICENSE("GPL");