2 * Copyright (C) 2001, 2002 Sistina Software (UK) Limited.
3 * Copyright (C) 2004-2008 Red Hat, Inc. All rights reserved.
5 * This file is released under the GPL.
11 #include <linux/init.h>
12 #include <linux/module.h>
13 #include <linux/mutex.h>
14 #include <linux/moduleparam.h>
15 #include <linux/blkpg.h>
16 #include <linux/bio.h>
17 #include <linux/buffer_head.h>
18 #include <linux/mempool.h>
19 #include <linux/slab.h>
20 #include <linux/idr.h>
21 #include <linux/hdreg.h>
22 #include <linux/delay.h>
24 #include <trace/events/block.h>
26 #define DM_MSG_PREFIX "core"
30 * ratelimit state to be used in DMXXX_LIMIT().
32 DEFINE_RATELIMIT_STATE(dm_ratelimit_state,
33 DEFAULT_RATELIMIT_INTERVAL,
34 DEFAULT_RATELIMIT_BURST);
35 EXPORT_SYMBOL(dm_ratelimit_state);
39 * Cookies are numeric values sent with CHANGE and REMOVE
40 * uevents while resuming, removing or renaming the device.
42 #define DM_COOKIE_ENV_VAR_NAME "DM_COOKIE"
43 #define DM_COOKIE_LENGTH 24
45 static const char *_name = DM_NAME;
47 static unsigned int major = 0;
48 static unsigned int _major = 0;
50 static DEFINE_IDR(_minor_idr);
52 static DEFINE_SPINLOCK(_minor_lock);
55 * One of these is allocated per bio.
58 struct mapped_device *md;
62 unsigned long start_time;
63 spinlock_t endio_lock;
68 * One of these is allocated per target within a bio. Hopefully
69 * this will be simplified out one day.
78 * For request-based dm.
79 * One of these is allocated per request.
81 struct dm_rq_target_io {
82 struct mapped_device *md;
84 struct request *orig, clone;
90 * For request-based dm.
91 * One of these is allocated per bio.
93 struct dm_rq_clone_bio_info {
95 struct dm_rq_target_io *tio;
98 union map_info *dm_get_mapinfo(struct bio *bio)
100 if (bio && bio->bi_private)
101 return &((struct dm_target_io *)bio->bi_private)->info;
105 union map_info *dm_get_rq_mapinfo(struct request *rq)
107 if (rq && rq->end_io_data)
108 return &((struct dm_rq_target_io *)rq->end_io_data)->info;
111 EXPORT_SYMBOL_GPL(dm_get_rq_mapinfo);
113 #define MINOR_ALLOCED ((void *)-1)
116 * Bits for the md->flags field.
118 #define DMF_BLOCK_IO_FOR_SUSPEND 0
119 #define DMF_SUSPENDED 1
121 #define DMF_FREEING 3
122 #define DMF_DELETING 4
123 #define DMF_NOFLUSH_SUSPENDING 5
124 #define DMF_MERGE_IS_OPTIONAL 6
127 * Work processed by per-device workqueue.
129 struct mapped_device {
130 struct rw_semaphore io_lock;
131 struct mutex suspend_lock;
138 struct request_queue *queue;
140 /* Protect queue and type against concurrent access. */
141 struct mutex type_lock;
143 struct target_type *immutable_target_type;
145 struct gendisk *disk;
151 * A list of ios that arrived while we were suspended.
154 wait_queue_head_t wait;
155 struct work_struct work;
156 struct bio_list deferred;
157 spinlock_t deferred_lock;
160 * Processing queue (flush)
162 struct workqueue_struct *wq;
165 * The current mapping.
167 struct dm_table *map;
170 * io objects are allocated from here.
181 wait_queue_head_t eventq;
183 struct list_head uevent_list;
184 spinlock_t uevent_lock; /* Protect access to uevent_list */
187 * freeze/thaw support require holding onto a super block
189 struct super_block *frozen_sb;
190 struct block_device *bdev;
192 /* forced geometry settings */
193 struct hd_geometry geometry;
195 /* For saving the address of __make_request for request based dm */
196 make_request_fn *saved_make_request_fn;
201 /* zero-length flush that will be cloned and submitted to targets */
202 struct bio flush_bio;
206 * For mempools pre-allocation at the table loading time.
208 struct dm_md_mempools {
215 static struct kmem_cache *_io_cache;
216 static struct kmem_cache *_tio_cache;
217 static struct kmem_cache *_rq_tio_cache;
218 static struct kmem_cache *_rq_bio_info_cache;
220 static int __init local_init(void)
224 /* allocate a slab for the dm_ios */
225 _io_cache = KMEM_CACHE(dm_io, 0);
229 /* allocate a slab for the target ios */
230 _tio_cache = KMEM_CACHE(dm_target_io, 0);
232 goto out_free_io_cache;
234 _rq_tio_cache = KMEM_CACHE(dm_rq_target_io, 0);
236 goto out_free_tio_cache;
238 _rq_bio_info_cache = KMEM_CACHE(dm_rq_clone_bio_info, 0);
239 if (!_rq_bio_info_cache)
240 goto out_free_rq_tio_cache;
242 r = dm_uevent_init();
244 goto out_free_rq_bio_info_cache;
247 r = register_blkdev(_major, _name);
249 goto out_uevent_exit;
258 out_free_rq_bio_info_cache:
259 kmem_cache_destroy(_rq_bio_info_cache);
260 out_free_rq_tio_cache:
261 kmem_cache_destroy(_rq_tio_cache);
263 kmem_cache_destroy(_tio_cache);
265 kmem_cache_destroy(_io_cache);
270 static void local_exit(void)
272 kmem_cache_destroy(_rq_bio_info_cache);
273 kmem_cache_destroy(_rq_tio_cache);
274 kmem_cache_destroy(_tio_cache);
275 kmem_cache_destroy(_io_cache);
276 unregister_blkdev(_major, _name);
281 DMINFO("cleaned up");
284 static int (*_inits[])(void) __initdata = {
294 static void (*_exits[])(void) = {
304 static int __init dm_init(void)
306 const int count = ARRAY_SIZE(_inits);
310 for (i = 0; i < count; i++) {
325 static void __exit dm_exit(void)
327 int i = ARRAY_SIZE(_exits);
333 * Should be empty by this point.
335 idr_remove_all(&_minor_idr);
336 idr_destroy(&_minor_idr);
340 * Block device functions
342 int dm_deleting_md(struct mapped_device *md)
344 return test_bit(DMF_DELETING, &md->flags);
347 static int dm_blk_open(struct block_device *bdev, fmode_t mode)
349 struct mapped_device *md;
351 spin_lock(&_minor_lock);
353 md = bdev->bd_disk->private_data;
357 if (test_bit(DMF_FREEING, &md->flags) ||
358 dm_deleting_md(md)) {
364 atomic_inc(&md->open_count);
367 spin_unlock(&_minor_lock);
369 return md ? 0 : -ENXIO;
372 static int dm_blk_close(struct gendisk *disk, fmode_t mode)
374 struct mapped_device *md = disk->private_data;
376 spin_lock(&_minor_lock);
378 atomic_dec(&md->open_count);
381 spin_unlock(&_minor_lock);
386 int dm_open_count(struct mapped_device *md)
388 return atomic_read(&md->open_count);
392 * Guarantees nothing is using the device before it's deleted.
394 int dm_lock_for_deletion(struct mapped_device *md)
398 spin_lock(&_minor_lock);
400 if (dm_open_count(md))
403 set_bit(DMF_DELETING, &md->flags);
405 spin_unlock(&_minor_lock);
410 static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo)
412 struct mapped_device *md = bdev->bd_disk->private_data;
414 return dm_get_geometry(md, geo);
417 static int dm_blk_ioctl(struct block_device *bdev, fmode_t mode,
418 unsigned int cmd, unsigned long arg)
420 struct mapped_device *md = bdev->bd_disk->private_data;
421 struct dm_table *map = dm_get_live_table(md);
422 struct dm_target *tgt;
425 if (!map || !dm_table_get_size(map))
428 /* We only support devices that have a single target */
429 if (dm_table_get_num_targets(map) != 1)
432 tgt = dm_table_get_target(map, 0);
434 if (dm_suspended_md(md)) {
439 if (tgt->type->ioctl)
440 r = tgt->type->ioctl(tgt, cmd, arg);
448 static struct dm_io *alloc_io(struct mapped_device *md)
450 return mempool_alloc(md->io_pool, GFP_NOIO);
453 static void free_io(struct mapped_device *md, struct dm_io *io)
455 mempool_free(io, md->io_pool);
458 static void free_tio(struct mapped_device *md, struct dm_target_io *tio)
460 mempool_free(tio, md->tio_pool);
463 static struct dm_rq_target_io *alloc_rq_tio(struct mapped_device *md,
466 return mempool_alloc(md->tio_pool, gfp_mask);
469 static void free_rq_tio(struct dm_rq_target_io *tio)
471 mempool_free(tio, tio->md->tio_pool);
474 static struct dm_rq_clone_bio_info *alloc_bio_info(struct mapped_device *md)
476 return mempool_alloc(md->io_pool, GFP_ATOMIC);
479 static void free_bio_info(struct dm_rq_clone_bio_info *info)
481 mempool_free(info, info->tio->md->io_pool);
484 static int md_in_flight(struct mapped_device *md)
486 return atomic_read(&md->pending[READ]) +
487 atomic_read(&md->pending[WRITE]);
490 static void start_io_acct(struct dm_io *io)
492 struct mapped_device *md = io->md;
494 int rw = bio_data_dir(io->bio);
496 io->start_time = jiffies;
498 cpu = part_stat_lock();
499 part_round_stats(cpu, &dm_disk(md)->part0);
501 atomic_set(&dm_disk(md)->part0.in_flight[rw],
502 atomic_inc_return(&md->pending[rw]));
505 static void end_io_acct(struct dm_io *io)
507 struct mapped_device *md = io->md;
508 struct bio *bio = io->bio;
509 unsigned long duration = jiffies - io->start_time;
511 int rw = bio_data_dir(bio);
513 cpu = part_stat_lock();
514 part_round_stats(cpu, &dm_disk(md)->part0);
515 part_stat_add(cpu, &dm_disk(md)->part0, ticks[rw], duration);
519 * After this is decremented the bio must not be touched if it is
522 pending = atomic_dec_return(&md->pending[rw]);
523 atomic_set(&dm_disk(md)->part0.in_flight[rw], pending);
524 pending += atomic_read(&md->pending[rw^0x1]);
526 /* nudge anyone waiting on suspend queue */
532 * Add the bio to the list of deferred io.
534 static void queue_io(struct mapped_device *md, struct bio *bio)
538 spin_lock_irqsave(&md->deferred_lock, flags);
539 bio_list_add(&md->deferred, bio);
540 spin_unlock_irqrestore(&md->deferred_lock, flags);
541 queue_work(md->wq, &md->work);
545 * Everyone (including functions in this file), should use this
546 * function to access the md->map field, and make sure they call
547 * dm_table_put() when finished.
549 struct dm_table *dm_get_live_table(struct mapped_device *md)
554 read_lock_irqsave(&md->map_lock, flags);
558 read_unlock_irqrestore(&md->map_lock, flags);
564 * Get the geometry associated with a dm device
566 int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo)
574 * Set the geometry of a device.
576 int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo)
578 sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors;
580 if (geo->start > sz) {
581 DMWARN("Start sector is beyond the geometry limits.");
590 /*-----------------------------------------------------------------
592 * A more elegant soln is in the works that uses the queue
593 * merge fn, unfortunately there are a couple of changes to
594 * the block layer that I want to make for this. So in the
595 * interests of getting something for people to use I give
596 * you this clearly demarcated crap.
597 *---------------------------------------------------------------*/
599 static int __noflush_suspending(struct mapped_device *md)
601 return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
605 * Decrements the number of outstanding ios that a bio has been
606 * cloned into, completing the original io if necc.
608 static void dec_pending(struct dm_io *io, int error)
613 struct mapped_device *md = io->md;
615 /* Push-back supersedes any I/O errors */
616 if (unlikely(error)) {
617 spin_lock_irqsave(&io->endio_lock, flags);
618 if (!(io->error > 0 && __noflush_suspending(md)))
620 spin_unlock_irqrestore(&io->endio_lock, flags);
623 if (atomic_dec_and_test(&io->io_count)) {
624 if (io->error == DM_ENDIO_REQUEUE) {
626 * Target requested pushing back the I/O.
628 spin_lock_irqsave(&md->deferred_lock, flags);
629 if (__noflush_suspending(md))
630 bio_list_add_head(&md->deferred, io->bio);
632 /* noflush suspend was interrupted. */
634 spin_unlock_irqrestore(&md->deferred_lock, flags);
637 io_error = io->error;
642 if (io_error == DM_ENDIO_REQUEUE)
645 if ((bio->bi_rw & REQ_FLUSH) && bio->bi_size) {
647 * Preflush done for flush with data, reissue
650 bio->bi_rw &= ~REQ_FLUSH;
653 /* done with normal IO or empty flush */
654 trace_block_bio_complete(md->queue, bio, io_error);
655 bio_endio(bio, io_error);
660 static void clone_endio(struct bio *bio, int error)
663 struct dm_target_io *tio = bio->bi_private;
664 struct dm_io *io = tio->io;
665 struct mapped_device *md = tio->io->md;
666 dm_endio_fn endio = tio->ti->type->end_io;
668 if (!bio_flagged(bio, BIO_UPTODATE) && !error)
672 r = endio(tio->ti, bio, error, &tio->info);
673 if (r < 0 || r == DM_ENDIO_REQUEUE)
675 * error and requeue request are handled
679 else if (r == DM_ENDIO_INCOMPLETE)
680 /* The target will handle the io */
683 DMWARN("unimplemented target endio return value: %d", r);
689 * Store md for cleanup instead of tio which is about to get freed.
691 bio->bi_private = md->bs;
695 dec_pending(io, error);
699 * Partial completion handling for request-based dm
701 static void end_clone_bio(struct bio *clone, int error)
703 struct dm_rq_clone_bio_info *info = clone->bi_private;
704 struct dm_rq_target_io *tio = info->tio;
705 struct bio *bio = info->orig;
706 unsigned int nr_bytes = info->orig->bi_size;
712 * An error has already been detected on the request.
713 * Once error occurred, just let clone->end_io() handle
719 * Don't notice the error to the upper layer yet.
720 * The error handling decision is made by the target driver,
721 * when the request is completed.
728 * I/O for the bio successfully completed.
729 * Notice the data completion to the upper layer.
733 * bios are processed from the head of the list.
734 * So the completing bio should always be rq->bio.
735 * If it's not, something wrong is happening.
737 if (tio->orig->bio != bio)
738 DMERR("bio completion is going in the middle of the request");
741 * Update the original request.
742 * Do not use blk_end_request() here, because it may complete
743 * the original request before the clone, and break the ordering.
745 blk_update_request(tio->orig, 0, nr_bytes);
749 * Don't touch any member of the md after calling this function because
750 * the md may be freed in dm_put() at the end of this function.
751 * Or do dm_get() before calling this function and dm_put() later.
753 static void rq_completed(struct mapped_device *md, int rw, int run_queue)
755 atomic_dec(&md->pending[rw]);
757 /* nudge anyone waiting on suspend queue */
758 if (!md_in_flight(md))
762 blk_run_queue(md->queue);
765 * dm_put() must be at the end of this function. See the comment above
770 static void free_rq_clone(struct request *clone)
772 struct dm_rq_target_io *tio = clone->end_io_data;
774 blk_rq_unprep_clone(clone);
779 * Complete the clone and the original request.
780 * Must be called without queue lock.
782 static void dm_end_request(struct request *clone, int error)
784 int rw = rq_data_dir(clone);
785 struct dm_rq_target_io *tio = clone->end_io_data;
786 struct mapped_device *md = tio->md;
787 struct request *rq = tio->orig;
789 if (rq->cmd_type == REQ_TYPE_BLOCK_PC) {
790 rq->errors = clone->errors;
791 rq->resid_len = clone->resid_len;
795 * We are using the sense buffer of the original
797 * So setting the length of the sense data is enough.
799 rq->sense_len = clone->sense_len;
802 free_rq_clone(clone);
803 blk_end_request_all(rq, error);
804 rq_completed(md, rw, true);
807 static void dm_unprep_request(struct request *rq)
809 struct request *clone = rq->special;
812 rq->cmd_flags &= ~REQ_DONTPREP;
814 free_rq_clone(clone);
818 * Requeue the original request of a clone.
820 void dm_requeue_unmapped_request(struct request *clone)
822 int rw = rq_data_dir(clone);
823 struct dm_rq_target_io *tio = clone->end_io_data;
824 struct mapped_device *md = tio->md;
825 struct request *rq = tio->orig;
826 struct request_queue *q = rq->q;
829 dm_unprep_request(rq);
831 spin_lock_irqsave(q->queue_lock, flags);
832 blk_requeue_request(q, rq);
833 spin_unlock_irqrestore(q->queue_lock, flags);
835 rq_completed(md, rw, 0);
837 EXPORT_SYMBOL_GPL(dm_requeue_unmapped_request);
839 static void __stop_queue(struct request_queue *q)
844 static void stop_queue(struct request_queue *q)
848 spin_lock_irqsave(q->queue_lock, flags);
850 spin_unlock_irqrestore(q->queue_lock, flags);
853 static void __start_queue(struct request_queue *q)
855 if (blk_queue_stopped(q))
859 static void start_queue(struct request_queue *q)
863 spin_lock_irqsave(q->queue_lock, flags);
865 spin_unlock_irqrestore(q->queue_lock, flags);
868 static void dm_done(struct request *clone, int error, bool mapped)
871 struct dm_rq_target_io *tio = clone->end_io_data;
872 dm_request_endio_fn rq_end_io = tio->ti->type->rq_end_io;
874 if (mapped && rq_end_io)
875 r = rq_end_io(tio->ti, clone, error, &tio->info);
878 /* The target wants to complete the I/O */
879 dm_end_request(clone, r);
880 else if (r == DM_ENDIO_INCOMPLETE)
881 /* The target will handle the I/O */
883 else if (r == DM_ENDIO_REQUEUE)
884 /* The target wants to requeue the I/O */
885 dm_requeue_unmapped_request(clone);
887 DMWARN("unimplemented target endio return value: %d", r);
893 * Request completion handler for request-based dm
895 static void dm_softirq_done(struct request *rq)
898 struct request *clone = rq->completion_data;
899 struct dm_rq_target_io *tio = clone->end_io_data;
901 if (rq->cmd_flags & REQ_FAILED)
904 dm_done(clone, tio->error, mapped);
908 * Complete the clone and the original request with the error status
909 * through softirq context.
911 static void dm_complete_request(struct request *clone, int error)
913 struct dm_rq_target_io *tio = clone->end_io_data;
914 struct request *rq = tio->orig;
917 rq->completion_data = clone;
918 blk_complete_request(rq);
922 * Complete the not-mapped clone and the original request with the error status
923 * through softirq context.
924 * Target's rq_end_io() function isn't called.
925 * This may be used when the target's map_rq() function fails.
927 void dm_kill_unmapped_request(struct request *clone, int error)
929 struct dm_rq_target_io *tio = clone->end_io_data;
930 struct request *rq = tio->orig;
932 rq->cmd_flags |= REQ_FAILED;
933 dm_complete_request(clone, error);
935 EXPORT_SYMBOL_GPL(dm_kill_unmapped_request);
938 * Called with the queue lock held
940 static void end_clone_request(struct request *clone, int error)
943 * For just cleaning up the information of the queue in which
944 * the clone was dispatched.
945 * The clone is *NOT* freed actually here because it is alloced from
946 * dm own mempool and REQ_ALLOCED isn't set in clone->cmd_flags.
948 __blk_put_request(clone->q, clone);
951 * Actual request completion is done in a softirq context which doesn't
952 * hold the queue lock. Otherwise, deadlock could occur because:
953 * - another request may be submitted by the upper level driver
954 * of the stacking during the completion
955 * - the submission which requires queue lock may be done
958 dm_complete_request(clone, error);
962 * Return maximum size of I/O possible at the supplied sector up to the current
965 static sector_t max_io_len_target_boundary(sector_t sector, struct dm_target *ti)
967 sector_t target_offset = dm_target_offset(ti, sector);
969 return ti->len - target_offset;
972 static sector_t max_io_len(sector_t sector, struct dm_target *ti)
974 sector_t len = max_io_len_target_boundary(sector, ti);
977 * Does the target need to split even further ?
981 sector_t offset = dm_target_offset(ti, sector);
982 boundary = ((offset + ti->split_io) & ~(ti->split_io - 1))
991 static void __map_bio(struct dm_target *ti, struct bio *clone,
992 struct dm_target_io *tio)
996 struct mapped_device *md;
998 clone->bi_end_io = clone_endio;
999 clone->bi_private = tio;
1002 * Map the clone. If r == 0 we don't need to do
1003 * anything, the target has assumed ownership of
1006 atomic_inc(&tio->io->io_count);
1007 sector = clone->bi_sector;
1008 r = ti->type->map(ti, clone, &tio->info);
1009 if (r == DM_MAPIO_REMAPPED) {
1010 /* the bio has been remapped so dispatch it */
1012 trace_block_bio_remap(bdev_get_queue(clone->bi_bdev), clone,
1013 tio->io->bio->bi_bdev->bd_dev, sector);
1015 generic_make_request(clone);
1016 } else if (r < 0 || r == DM_MAPIO_REQUEUE) {
1017 /* error the io and bail out, or requeue it if needed */
1019 dec_pending(tio->io, r);
1021 * Store bio_set for cleanup.
1023 clone->bi_private = md->bs;
1027 DMWARN("unimplemented target map return value: %d", r);
1033 struct mapped_device *md;
1034 struct dm_table *map;
1038 sector_t sector_count;
1042 static void dm_bio_destructor(struct bio *bio)
1044 struct bio_set *bs = bio->bi_private;
1050 * Creates a little bio that just does part of a bvec.
1052 static struct bio *split_bvec(struct bio *bio, sector_t sector,
1053 unsigned short idx, unsigned int offset,
1054 unsigned int len, struct bio_set *bs)
1057 struct bio_vec *bv = bio->bi_io_vec + idx;
1059 clone = bio_alloc_bioset(GFP_NOIO, 1, bs);
1060 clone->bi_destructor = dm_bio_destructor;
1061 *clone->bi_io_vec = *bv;
1063 clone->bi_sector = sector;
1064 clone->bi_bdev = bio->bi_bdev;
1065 clone->bi_rw = bio->bi_rw;
1067 clone->bi_size = to_bytes(len);
1068 clone->bi_io_vec->bv_offset = offset;
1069 clone->bi_io_vec->bv_len = clone->bi_size;
1070 clone->bi_flags |= 1 << BIO_CLONED;
1072 if (bio_integrity(bio)) {
1073 bio_integrity_clone(clone, bio, GFP_NOIO, bs);
1074 bio_integrity_trim(clone,
1075 bio_sector_offset(bio, idx, offset), len);
1082 * Creates a bio that consists of range of complete bvecs.
1084 static struct bio *clone_bio(struct bio *bio, sector_t sector,
1085 unsigned short idx, unsigned short bv_count,
1086 unsigned int len, struct bio_set *bs)
1090 clone = bio_alloc_bioset(GFP_NOIO, bio->bi_max_vecs, bs);
1091 __bio_clone(clone, bio);
1092 clone->bi_destructor = dm_bio_destructor;
1093 clone->bi_sector = sector;
1094 clone->bi_idx = idx;
1095 clone->bi_vcnt = idx + bv_count;
1096 clone->bi_size = to_bytes(len);
1097 clone->bi_flags &= ~(1 << BIO_SEG_VALID);
1099 if (bio_integrity(bio)) {
1100 bio_integrity_clone(clone, bio, GFP_NOIO, bs);
1102 if (idx != bio->bi_idx || clone->bi_size < bio->bi_size)
1103 bio_integrity_trim(clone,
1104 bio_sector_offset(bio, idx, 0), len);
1110 static struct dm_target_io *alloc_tio(struct clone_info *ci,
1111 struct dm_target *ti)
1113 struct dm_target_io *tio = mempool_alloc(ci->md->tio_pool, GFP_NOIO);
1117 memset(&tio->info, 0, sizeof(tio->info));
1122 static void __issue_target_request(struct clone_info *ci, struct dm_target *ti,
1123 unsigned request_nr, sector_t len)
1125 struct dm_target_io *tio = alloc_tio(ci, ti);
1128 tio->info.target_request_nr = request_nr;
1131 * Discard requests require the bio's inline iovecs be initialized.
1132 * ci->bio->bi_max_vecs is BIO_INLINE_VECS anyway, for both flush
1133 * and discard, so no need for concern about wasted bvec allocations.
1135 clone = bio_alloc_bioset(GFP_NOIO, ci->bio->bi_max_vecs, ci->md->bs);
1136 __bio_clone(clone, ci->bio);
1137 clone->bi_destructor = dm_bio_destructor;
1139 clone->bi_sector = ci->sector;
1140 clone->bi_size = to_bytes(len);
1143 __map_bio(ti, clone, tio);
1146 static void __issue_target_requests(struct clone_info *ci, struct dm_target *ti,
1147 unsigned num_requests, sector_t len)
1149 unsigned request_nr;
1151 for (request_nr = 0; request_nr < num_requests; request_nr++)
1152 __issue_target_request(ci, ti, request_nr, len);
1155 static int __clone_and_map_empty_flush(struct clone_info *ci)
1157 unsigned target_nr = 0;
1158 struct dm_target *ti;
1160 BUG_ON(bio_has_data(ci->bio));
1161 while ((ti = dm_table_get_target(ci->map, target_nr++)))
1162 __issue_target_requests(ci, ti, ti->num_flush_requests, 0);
1168 * Perform all io with a single clone.
1170 static void __clone_and_map_simple(struct clone_info *ci, struct dm_target *ti)
1172 struct bio *clone, *bio = ci->bio;
1173 struct dm_target_io *tio;
1175 tio = alloc_tio(ci, ti);
1176 clone = clone_bio(bio, ci->sector, ci->idx,
1177 bio->bi_vcnt - ci->idx, ci->sector_count,
1179 __map_bio(ti, clone, tio);
1180 ci->sector_count = 0;
1183 static int __clone_and_map_discard(struct clone_info *ci)
1185 struct dm_target *ti;
1189 ti = dm_table_find_target(ci->map, ci->sector);
1190 if (!dm_target_is_valid(ti))
1194 * Even though the device advertised discard support,
1195 * that does not mean every target supports it, and
1196 * reconfiguration might also have changed that since the
1197 * check was performed.
1199 if (!ti->num_discard_requests)
1202 len = min(ci->sector_count, max_io_len_target_boundary(ci->sector, ti));
1204 __issue_target_requests(ci, ti, ti->num_discard_requests, len);
1207 } while (ci->sector_count -= len);
1212 static int __clone_and_map(struct clone_info *ci)
1214 struct bio *clone, *bio = ci->bio;
1215 struct dm_target *ti;
1216 sector_t len = 0, max;
1217 struct dm_target_io *tio;
1219 if (unlikely(bio->bi_rw & REQ_DISCARD))
1220 return __clone_and_map_discard(ci);
1222 ti = dm_table_find_target(ci->map, ci->sector);
1223 if (!dm_target_is_valid(ti))
1226 max = max_io_len(ci->sector, ti);
1228 if (ci->sector_count <= max) {
1230 * Optimise for the simple case where we can do all of
1231 * the remaining io with a single clone.
1233 __clone_and_map_simple(ci, ti);
1235 } else if (to_sector(bio->bi_io_vec[ci->idx].bv_len) <= max) {
1237 * There are some bvecs that don't span targets.
1238 * Do as many of these as possible.
1241 sector_t remaining = max;
1244 for (i = ci->idx; remaining && (i < bio->bi_vcnt); i++) {
1245 bv_len = to_sector(bio->bi_io_vec[i].bv_len);
1247 if (bv_len > remaining)
1250 remaining -= bv_len;
1254 tio = alloc_tio(ci, ti);
1255 clone = clone_bio(bio, ci->sector, ci->idx, i - ci->idx, len,
1257 __map_bio(ti, clone, tio);
1260 ci->sector_count -= len;
1265 * Handle a bvec that must be split between two or more targets.
1267 struct bio_vec *bv = bio->bi_io_vec + ci->idx;
1268 sector_t remaining = to_sector(bv->bv_len);
1269 unsigned int offset = 0;
1273 ti = dm_table_find_target(ci->map, ci->sector);
1274 if (!dm_target_is_valid(ti))
1277 max = max_io_len(ci->sector, ti);
1280 len = min(remaining, max);
1282 tio = alloc_tio(ci, ti);
1283 clone = split_bvec(bio, ci->sector, ci->idx,
1284 bv->bv_offset + offset, len,
1287 __map_bio(ti, clone, tio);
1290 ci->sector_count -= len;
1291 offset += to_bytes(len);
1292 } while (remaining -= len);
1301 * Split the bio into several clones and submit it to targets.
1303 static void __split_and_process_bio(struct mapped_device *md, struct bio *bio)
1305 struct clone_info ci;
1308 ci.map = dm_get_live_table(md);
1309 if (unlikely(!ci.map)) {
1315 ci.io = alloc_io(md);
1317 atomic_set(&ci.io->io_count, 1);
1320 spin_lock_init(&ci.io->endio_lock);
1321 ci.sector = bio->bi_sector;
1322 ci.idx = bio->bi_idx;
1324 start_io_acct(ci.io);
1325 if (bio->bi_rw & REQ_FLUSH) {
1326 ci.bio = &ci.md->flush_bio;
1327 ci.sector_count = 0;
1328 error = __clone_and_map_empty_flush(&ci);
1329 /* dec_pending submits any data associated with flush */
1332 ci.sector_count = bio_sectors(bio);
1333 while (ci.sector_count && !error)
1334 error = __clone_and_map(&ci);
1337 /* drop the extra reference count */
1338 dec_pending(ci.io, error);
1339 dm_table_put(ci.map);
1341 /*-----------------------------------------------------------------
1343 *---------------------------------------------------------------*/
1345 static int dm_merge_bvec(struct request_queue *q,
1346 struct bvec_merge_data *bvm,
1347 struct bio_vec *biovec)
1349 struct mapped_device *md = q->queuedata;
1350 struct dm_table *map = dm_get_live_table(md);
1351 struct dm_target *ti;
1352 sector_t max_sectors;
1358 ti = dm_table_find_target(map, bvm->bi_sector);
1359 if (!dm_target_is_valid(ti))
1363 * Find maximum amount of I/O that won't need splitting
1365 max_sectors = min(max_io_len(bvm->bi_sector, ti),
1366 (sector_t) BIO_MAX_SECTORS);
1367 max_size = (max_sectors << SECTOR_SHIFT) - bvm->bi_size;
1372 * merge_bvec_fn() returns number of bytes
1373 * it can accept at this offset
1374 * max is precomputed maximal io size
1376 if (max_size && ti->type->merge)
1377 max_size = ti->type->merge(ti, bvm, biovec, max_size);
1379 * If the target doesn't support merge method and some of the devices
1380 * provided their merge_bvec method (we know this by looking at
1381 * queue_max_hw_sectors), then we can't allow bios with multiple vector
1382 * entries. So always set max_size to 0, and the code below allows
1385 else if (queue_max_hw_sectors(q) <= PAGE_SIZE >> 9)
1394 * Always allow an entire first page
1396 if (max_size <= biovec->bv_len && !(bvm->bi_size >> SECTOR_SHIFT))
1397 max_size = biovec->bv_len;
1403 * The request function that just remaps the bio built up by
1406 static int _dm_request(struct request_queue *q, struct bio *bio)
1408 int rw = bio_data_dir(bio);
1409 struct mapped_device *md = q->queuedata;
1412 down_read(&md->io_lock);
1414 cpu = part_stat_lock();
1415 part_stat_inc(cpu, &dm_disk(md)->part0, ios[rw]);
1416 part_stat_add(cpu, &dm_disk(md)->part0, sectors[rw], bio_sectors(bio));
1419 /* if we're suspended, we have to queue this io for later */
1420 if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))) {
1421 up_read(&md->io_lock);
1423 if (bio_rw(bio) != READA)
1430 __split_and_process_bio(md, bio);
1431 up_read(&md->io_lock);
1435 static int dm_make_request(struct request_queue *q, struct bio *bio)
1437 struct mapped_device *md = q->queuedata;
1439 return md->saved_make_request_fn(q, bio); /* call __make_request() */
1442 static int dm_request_based(struct mapped_device *md)
1444 return blk_queue_stackable(md->queue);
1447 static int dm_request(struct request_queue *q, struct bio *bio)
1449 struct mapped_device *md = q->queuedata;
1451 if (dm_request_based(md))
1452 return dm_make_request(q, bio);
1454 return _dm_request(q, bio);
1457 void dm_dispatch_request(struct request *rq)
1461 if (blk_queue_io_stat(rq->q))
1462 rq->cmd_flags |= REQ_IO_STAT;
1464 rq->start_time = jiffies;
1465 r = blk_insert_cloned_request(rq->q, rq);
1467 dm_complete_request(rq, r);
1469 EXPORT_SYMBOL_GPL(dm_dispatch_request);
1471 static void dm_rq_bio_destructor(struct bio *bio)
1473 struct dm_rq_clone_bio_info *info = bio->bi_private;
1474 struct mapped_device *md = info->tio->md;
1476 free_bio_info(info);
1477 bio_free(bio, md->bs);
1480 static int dm_rq_bio_constructor(struct bio *bio, struct bio *bio_orig,
1483 struct dm_rq_target_io *tio = data;
1484 struct mapped_device *md = tio->md;
1485 struct dm_rq_clone_bio_info *info = alloc_bio_info(md);
1490 info->orig = bio_orig;
1492 bio->bi_end_io = end_clone_bio;
1493 bio->bi_private = info;
1494 bio->bi_destructor = dm_rq_bio_destructor;
1499 static int setup_clone(struct request *clone, struct request *rq,
1500 struct dm_rq_target_io *tio)
1504 r = blk_rq_prep_clone(clone, rq, tio->md->bs, GFP_ATOMIC,
1505 dm_rq_bio_constructor, tio);
1509 clone->cmd = rq->cmd;
1510 clone->cmd_len = rq->cmd_len;
1511 clone->sense = rq->sense;
1512 clone->buffer = rq->buffer;
1513 clone->end_io = end_clone_request;
1514 clone->end_io_data = tio;
1519 static struct request *clone_rq(struct request *rq, struct mapped_device *md,
1522 struct request *clone;
1523 struct dm_rq_target_io *tio;
1525 tio = alloc_rq_tio(md, gfp_mask);
1533 memset(&tio->info, 0, sizeof(tio->info));
1535 clone = &tio->clone;
1536 if (setup_clone(clone, rq, tio)) {
1546 * Called with the queue lock held.
1548 static int dm_prep_fn(struct request_queue *q, struct request *rq)
1550 struct mapped_device *md = q->queuedata;
1551 struct request *clone;
1553 if (unlikely(rq->special)) {
1554 DMWARN("Already has something in rq->special.");
1555 return BLKPREP_KILL;
1558 clone = clone_rq(rq, md, GFP_ATOMIC);
1560 return BLKPREP_DEFER;
1562 rq->special = clone;
1563 rq->cmd_flags |= REQ_DONTPREP;
1570 * 0 : the request has been processed (not requeued)
1571 * !0 : the request has been requeued
1573 static int map_request(struct dm_target *ti, struct request *clone,
1574 struct mapped_device *md)
1576 int r, requeued = 0;
1577 struct dm_rq_target_io *tio = clone->end_io_data;
1580 * Hold the md reference here for the in-flight I/O.
1581 * We can't rely on the reference count by device opener,
1582 * because the device may be closed during the request completion
1583 * when all bios are completed.
1584 * See the comment in rq_completed() too.
1589 r = ti->type->map_rq(ti, clone, &tio->info);
1591 case DM_MAPIO_SUBMITTED:
1592 /* The target has taken the I/O to submit by itself later */
1594 case DM_MAPIO_REMAPPED:
1595 /* The target has remapped the I/O so dispatch it */
1596 trace_block_rq_remap(clone->q, clone, disk_devt(dm_disk(md)),
1597 blk_rq_pos(tio->orig));
1598 dm_dispatch_request(clone);
1600 case DM_MAPIO_REQUEUE:
1601 /* The target wants to requeue the I/O */
1602 dm_requeue_unmapped_request(clone);
1607 DMWARN("unimplemented target map return value: %d", r);
1611 /* The target wants to complete the I/O */
1612 dm_kill_unmapped_request(clone, r);
1620 * q->request_fn for request-based dm.
1621 * Called with the queue lock held.
1623 static void dm_request_fn(struct request_queue *q)
1625 struct mapped_device *md = q->queuedata;
1626 struct dm_table *map = dm_get_live_table(md);
1627 struct dm_target *ti;
1628 struct request *rq, *clone;
1632 * For suspend, check blk_queue_stopped() and increment
1633 * ->pending within a single queue_lock not to increment the
1634 * number of in-flight I/Os after the queue is stopped in
1637 while (!blk_queue_stopped(q)) {
1638 rq = blk_peek_request(q);
1642 /* always use block 0 to find the target for flushes for now */
1644 if (!(rq->cmd_flags & REQ_FLUSH))
1645 pos = blk_rq_pos(rq);
1647 ti = dm_table_find_target(map, pos);
1648 BUG_ON(!dm_target_is_valid(ti));
1650 if (ti->type->busy && ti->type->busy(ti))
1653 blk_start_request(rq);
1654 clone = rq->special;
1655 atomic_inc(&md->pending[rq_data_dir(clone)]);
1657 spin_unlock(q->queue_lock);
1658 if (map_request(ti, clone, md))
1661 BUG_ON(!irqs_disabled());
1662 spin_lock(q->queue_lock);
1668 BUG_ON(!irqs_disabled());
1669 spin_lock(q->queue_lock);
1672 blk_delay_queue(q, HZ / 10);
1679 int dm_underlying_device_busy(struct request_queue *q)
1681 return blk_lld_busy(q);
1683 EXPORT_SYMBOL_GPL(dm_underlying_device_busy);
1685 static int dm_lld_busy(struct request_queue *q)
1688 struct mapped_device *md = q->queuedata;
1689 struct dm_table *map = dm_get_live_table(md);
1691 if (!map || test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))
1694 r = dm_table_any_busy_target(map);
1701 static int dm_any_congested(void *congested_data, int bdi_bits)
1704 struct mapped_device *md = congested_data;
1705 struct dm_table *map;
1707 if (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
1708 map = dm_get_live_table(md);
1711 * Request-based dm cares about only own queue for
1712 * the query about congestion status of request_queue
1714 if (dm_request_based(md))
1715 r = md->queue->backing_dev_info.state &
1718 r = dm_table_any_congested(map, bdi_bits);
1727 /*-----------------------------------------------------------------
1728 * An IDR is used to keep track of allocated minor numbers.
1729 *---------------------------------------------------------------*/
1730 static void free_minor(int minor)
1732 spin_lock(&_minor_lock);
1733 idr_remove(&_minor_idr, minor);
1734 spin_unlock(&_minor_lock);
1738 * See if the device with a specific minor # is free.
1740 static int specific_minor(int minor)
1744 if (minor >= (1 << MINORBITS))
1747 r = idr_pre_get(&_minor_idr, GFP_KERNEL);
1751 spin_lock(&_minor_lock);
1753 if (idr_find(&_minor_idr, minor)) {
1758 r = idr_get_new_above(&_minor_idr, MINOR_ALLOCED, minor, &m);
1763 idr_remove(&_minor_idr, m);
1769 spin_unlock(&_minor_lock);
1773 static int next_free_minor(int *minor)
1777 r = idr_pre_get(&_minor_idr, GFP_KERNEL);
1781 spin_lock(&_minor_lock);
1783 r = idr_get_new(&_minor_idr, MINOR_ALLOCED, &m);
1787 if (m >= (1 << MINORBITS)) {
1788 idr_remove(&_minor_idr, m);
1796 spin_unlock(&_minor_lock);
1800 static const struct block_device_operations dm_blk_dops;
1802 static void dm_wq_work(struct work_struct *work);
1804 static void dm_init_md_queue(struct mapped_device *md)
1807 * Request-based dm devices cannot be stacked on top of bio-based dm
1808 * devices. The type of this dm device has not been decided yet.
1809 * The type is decided at the first table loading time.
1810 * To prevent problematic device stacking, clear the queue flag
1811 * for request stacking support until then.
1813 * This queue is new, so no concurrency on the queue_flags.
1815 queue_flag_clear_unlocked(QUEUE_FLAG_STACKABLE, md->queue);
1817 md->queue->queuedata = md;
1818 md->queue->backing_dev_info.congested_fn = dm_any_congested;
1819 md->queue->backing_dev_info.congested_data = md;
1820 blk_queue_make_request(md->queue, dm_request);
1821 blk_queue_bounce_limit(md->queue, BLK_BOUNCE_ANY);
1822 blk_queue_merge_bvec(md->queue, dm_merge_bvec);
1826 * Allocate and initialise a blank device with a given minor.
1828 static struct mapped_device *alloc_dev(int minor)
1831 struct mapped_device *md = kzalloc(sizeof(*md), GFP_KERNEL);
1835 DMWARN("unable to allocate device, out of memory.");
1839 if (!try_module_get(THIS_MODULE))
1840 goto bad_module_get;
1842 /* get a minor number for the dev */
1843 if (minor == DM_ANY_MINOR)
1844 r = next_free_minor(&minor);
1846 r = specific_minor(minor);
1850 md->type = DM_TYPE_NONE;
1851 init_rwsem(&md->io_lock);
1852 mutex_init(&md->suspend_lock);
1853 mutex_init(&md->type_lock);
1854 spin_lock_init(&md->deferred_lock);
1855 rwlock_init(&md->map_lock);
1856 atomic_set(&md->holders, 1);
1857 atomic_set(&md->open_count, 0);
1858 atomic_set(&md->event_nr, 0);
1859 atomic_set(&md->uevent_seq, 0);
1860 INIT_LIST_HEAD(&md->uevent_list);
1861 spin_lock_init(&md->uevent_lock);
1863 md->queue = blk_alloc_queue(GFP_KERNEL);
1867 dm_init_md_queue(md);
1869 md->disk = alloc_disk(1);
1873 atomic_set(&md->pending[0], 0);
1874 atomic_set(&md->pending[1], 0);
1875 init_waitqueue_head(&md->wait);
1876 INIT_WORK(&md->work, dm_wq_work);
1877 init_waitqueue_head(&md->eventq);
1879 md->disk->major = _major;
1880 md->disk->first_minor = minor;
1881 md->disk->fops = &dm_blk_dops;
1882 md->disk->queue = md->queue;
1883 md->disk->private_data = md;
1884 sprintf(md->disk->disk_name, "dm-%d", minor);
1886 format_dev_t(md->name, MKDEV(_major, minor));
1888 md->wq = alloc_workqueue("kdmflush",
1889 WQ_NON_REENTRANT | WQ_MEM_RECLAIM, 0);
1893 md->bdev = bdget_disk(md->disk, 0);
1897 bio_init(&md->flush_bio);
1898 md->flush_bio.bi_bdev = md->bdev;
1899 md->flush_bio.bi_rw = WRITE_FLUSH;
1901 /* Populate the mapping, nobody knows we exist yet */
1902 spin_lock(&_minor_lock);
1903 old_md = idr_replace(&_minor_idr, md, minor);
1904 spin_unlock(&_minor_lock);
1906 BUG_ON(old_md != MINOR_ALLOCED);
1911 destroy_workqueue(md->wq);
1913 del_gendisk(md->disk);
1916 blk_cleanup_queue(md->queue);
1920 module_put(THIS_MODULE);
1926 static void unlock_fs(struct mapped_device *md);
1928 static void free_dev(struct mapped_device *md)
1930 int minor = MINOR(disk_devt(md->disk));
1934 destroy_workqueue(md->wq);
1936 mempool_destroy(md->tio_pool);
1938 mempool_destroy(md->io_pool);
1940 bioset_free(md->bs);
1941 blk_integrity_unregister(md->disk);
1942 del_gendisk(md->disk);
1945 spin_lock(&_minor_lock);
1946 md->disk->private_data = NULL;
1947 spin_unlock(&_minor_lock);
1950 blk_cleanup_queue(md->queue);
1951 module_put(THIS_MODULE);
1955 static void __bind_mempools(struct mapped_device *md, struct dm_table *t)
1957 struct dm_md_mempools *p;
1959 if (md->io_pool && md->tio_pool && md->bs)
1960 /* the md already has necessary mempools */
1963 p = dm_table_get_md_mempools(t);
1964 BUG_ON(!p || md->io_pool || md->tio_pool || md->bs);
1966 md->io_pool = p->io_pool;
1968 md->tio_pool = p->tio_pool;
1974 /* mempool bind completed, now no need any mempools in the table */
1975 dm_table_free_md_mempools(t);
1979 * Bind a table to the device.
1981 static void event_callback(void *context)
1983 unsigned long flags;
1985 struct mapped_device *md = (struct mapped_device *) context;
1987 spin_lock_irqsave(&md->uevent_lock, flags);
1988 list_splice_init(&md->uevent_list, &uevents);
1989 spin_unlock_irqrestore(&md->uevent_lock, flags);
1991 dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj);
1993 atomic_inc(&md->event_nr);
1994 wake_up(&md->eventq);
1998 * Protected by md->suspend_lock obtained by dm_swap_table().
2000 static void __set_size(struct mapped_device *md, sector_t size)
2002 set_capacity(md->disk, size);
2004 i_size_write(md->bdev->bd_inode, (loff_t)size << SECTOR_SHIFT);
2008 * Return 1 if the queue has a compulsory merge_bvec_fn function.
2010 * If this function returns 0, then the device is either a non-dm
2011 * device without a merge_bvec_fn, or it is a dm device that is
2012 * able to split any bios it receives that are too big.
2014 int dm_queue_merge_is_compulsory(struct request_queue *q)
2016 struct mapped_device *dev_md;
2018 if (!q->merge_bvec_fn)
2021 if (q->make_request_fn == dm_request) {
2022 dev_md = q->queuedata;
2023 if (test_bit(DMF_MERGE_IS_OPTIONAL, &dev_md->flags))
2030 static int dm_device_merge_is_compulsory(struct dm_target *ti,
2031 struct dm_dev *dev, sector_t start,
2032 sector_t len, void *data)
2034 struct block_device *bdev = dev->bdev;
2035 struct request_queue *q = bdev_get_queue(bdev);
2037 return dm_queue_merge_is_compulsory(q);
2041 * Return 1 if it is acceptable to ignore merge_bvec_fn based
2042 * on the properties of the underlying devices.
2044 static int dm_table_merge_is_optional(struct dm_table *table)
2047 struct dm_target *ti;
2049 while (i < dm_table_get_num_targets(table)) {
2050 ti = dm_table_get_target(table, i++);
2052 if (ti->type->iterate_devices &&
2053 ti->type->iterate_devices(ti, dm_device_merge_is_compulsory, NULL))
2061 * Returns old map, which caller must destroy.
2063 static struct dm_table *__bind(struct mapped_device *md, struct dm_table *t,
2064 struct queue_limits *limits)
2066 struct dm_table *old_map;
2067 struct request_queue *q = md->queue;
2069 unsigned long flags;
2070 int merge_is_optional;
2072 size = dm_table_get_size(t);
2075 * Wipe any geometry if the size of the table changed.
2077 if (size != get_capacity(md->disk))
2078 memset(&md->geometry, 0, sizeof(md->geometry));
2080 __set_size(md, size);
2082 dm_table_event_callback(t, event_callback, md);
2085 * The queue hasn't been stopped yet, if the old table type wasn't
2086 * for request-based during suspension. So stop it to prevent
2087 * I/O mapping before resume.
2088 * This must be done before setting the queue restrictions,
2089 * because request-based dm may be run just after the setting.
2091 if (dm_table_request_based(t) && !blk_queue_stopped(q))
2094 __bind_mempools(md, t);
2096 merge_is_optional = dm_table_merge_is_optional(t);
2098 write_lock_irqsave(&md->map_lock, flags);
2101 md->immutable_target_type = dm_table_get_immutable_target_type(t);
2103 dm_table_set_restrictions(t, q, limits);
2104 if (merge_is_optional)
2105 set_bit(DMF_MERGE_IS_OPTIONAL, &md->flags);
2107 clear_bit(DMF_MERGE_IS_OPTIONAL, &md->flags);
2108 write_unlock_irqrestore(&md->map_lock, flags);
2114 * Returns unbound table for the caller to free.
2116 static struct dm_table *__unbind(struct mapped_device *md)
2118 struct dm_table *map = md->map;
2119 unsigned long flags;
2124 dm_table_event_callback(map, NULL, NULL);
2125 write_lock_irqsave(&md->map_lock, flags);
2127 write_unlock_irqrestore(&md->map_lock, flags);
2133 * Constructor for a new device.
2135 int dm_create(int minor, struct mapped_device **result)
2137 struct mapped_device *md;
2139 md = alloc_dev(minor);
2150 * Functions to manage md->type.
2151 * All are required to hold md->type_lock.
2153 void dm_lock_md_type(struct mapped_device *md)
2155 mutex_lock(&md->type_lock);
2158 void dm_unlock_md_type(struct mapped_device *md)
2160 mutex_unlock(&md->type_lock);
2163 void dm_set_md_type(struct mapped_device *md, unsigned type)
2168 unsigned dm_get_md_type(struct mapped_device *md)
2173 struct target_type *dm_get_immutable_target_type(struct mapped_device *md)
2175 return md->immutable_target_type;
2179 * Fully initialize a request-based queue (->elevator, ->request_fn, etc).
2181 static int dm_init_request_based_queue(struct mapped_device *md)
2183 struct request_queue *q = NULL;
2185 if (md->queue->elevator)
2188 /* Fully initialize the queue */
2189 q = blk_init_allocated_queue(md->queue, dm_request_fn, NULL);
2194 md->saved_make_request_fn = md->queue->make_request_fn;
2195 dm_init_md_queue(md);
2196 blk_queue_softirq_done(md->queue, dm_softirq_done);
2197 blk_queue_prep_rq(md->queue, dm_prep_fn);
2198 blk_queue_lld_busy(md->queue, dm_lld_busy);
2200 elv_register_queue(md->queue);
2206 * Setup the DM device's queue based on md's type
2208 int dm_setup_md_queue(struct mapped_device *md)
2210 if ((dm_get_md_type(md) == DM_TYPE_REQUEST_BASED) &&
2211 !dm_init_request_based_queue(md)) {
2212 DMWARN("Cannot initialize queue for request-based mapped device");
2219 static struct mapped_device *dm_find_md(dev_t dev)
2221 struct mapped_device *md;
2222 unsigned minor = MINOR(dev);
2224 if (MAJOR(dev) != _major || minor >= (1 << MINORBITS))
2227 spin_lock(&_minor_lock);
2229 md = idr_find(&_minor_idr, minor);
2230 if (md && (md == MINOR_ALLOCED ||
2231 (MINOR(disk_devt(dm_disk(md))) != minor) ||
2232 dm_deleting_md(md) ||
2233 test_bit(DMF_FREEING, &md->flags))) {
2239 spin_unlock(&_minor_lock);
2244 struct mapped_device *dm_get_md(dev_t dev)
2246 struct mapped_device *md = dm_find_md(dev);
2253 EXPORT_SYMBOL_GPL(dm_get_md);
2255 void *dm_get_mdptr(struct mapped_device *md)
2257 return md->interface_ptr;
2260 void dm_set_mdptr(struct mapped_device *md, void *ptr)
2262 md->interface_ptr = ptr;
2265 void dm_get(struct mapped_device *md)
2267 atomic_inc(&md->holders);
2268 BUG_ON(test_bit(DMF_FREEING, &md->flags));
2271 const char *dm_device_name(struct mapped_device *md)
2275 EXPORT_SYMBOL_GPL(dm_device_name);
2277 static void __dm_destroy(struct mapped_device *md, bool wait)
2279 struct dm_table *map;
2283 spin_lock(&_minor_lock);
2284 map = dm_get_live_table(md);
2285 idr_replace(&_minor_idr, MINOR_ALLOCED, MINOR(disk_devt(dm_disk(md))));
2286 set_bit(DMF_FREEING, &md->flags);
2287 spin_unlock(&_minor_lock);
2289 if (!dm_suspended_md(md)) {
2290 dm_table_presuspend_targets(map);
2291 dm_table_postsuspend_targets(map);
2295 * Rare, but there may be I/O requests still going to complete,
2296 * for example. Wait for all references to disappear.
2297 * No one should increment the reference count of the mapped_device,
2298 * after the mapped_device state becomes DMF_FREEING.
2301 while (atomic_read(&md->holders))
2303 else if (atomic_read(&md->holders))
2304 DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)",
2305 dm_device_name(md), atomic_read(&md->holders));
2309 dm_table_destroy(__unbind(md));
2313 void dm_destroy(struct mapped_device *md)
2315 __dm_destroy(md, true);
2318 void dm_destroy_immediate(struct mapped_device *md)
2320 __dm_destroy(md, false);
2323 void dm_put(struct mapped_device *md)
2325 atomic_dec(&md->holders);
2327 EXPORT_SYMBOL_GPL(dm_put);
2329 static int dm_wait_for_completion(struct mapped_device *md, int interruptible)
2332 DECLARE_WAITQUEUE(wait, current);
2334 add_wait_queue(&md->wait, &wait);
2337 set_current_state(interruptible);
2339 if (!md_in_flight(md))
2342 if (interruptible == TASK_INTERRUPTIBLE &&
2343 signal_pending(current)) {
2350 set_current_state(TASK_RUNNING);
2352 remove_wait_queue(&md->wait, &wait);
2358 * Process the deferred bios
2360 static void dm_wq_work(struct work_struct *work)
2362 struct mapped_device *md = container_of(work, struct mapped_device,
2366 down_read(&md->io_lock);
2368 while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
2369 spin_lock_irq(&md->deferred_lock);
2370 c = bio_list_pop(&md->deferred);
2371 spin_unlock_irq(&md->deferred_lock);
2376 up_read(&md->io_lock);
2378 if (dm_request_based(md))
2379 generic_make_request(c);
2381 __split_and_process_bio(md, c);
2383 down_read(&md->io_lock);
2386 up_read(&md->io_lock);
2389 static void dm_queue_flush(struct mapped_device *md)
2391 clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2392 smp_mb__after_clear_bit();
2393 queue_work(md->wq, &md->work);
2397 * Swap in a new table, returning the old one for the caller to destroy.
2399 struct dm_table *dm_swap_table(struct mapped_device *md, struct dm_table *table)
2401 struct dm_table *map = ERR_PTR(-EINVAL);
2402 struct queue_limits limits;
2405 mutex_lock(&md->suspend_lock);
2407 /* device must be suspended */
2408 if (!dm_suspended_md(md))
2411 r = dm_calculate_queue_limits(table, &limits);
2417 map = __bind(md, table, &limits);
2420 mutex_unlock(&md->suspend_lock);
2425 * Functions to lock and unlock any filesystem running on the
2428 static int lock_fs(struct mapped_device *md)
2432 WARN_ON(md->frozen_sb);
2434 md->frozen_sb = freeze_bdev(md->bdev);
2435 if (IS_ERR(md->frozen_sb)) {
2436 r = PTR_ERR(md->frozen_sb);
2437 md->frozen_sb = NULL;
2441 set_bit(DMF_FROZEN, &md->flags);
2446 static void unlock_fs(struct mapped_device *md)
2448 if (!test_bit(DMF_FROZEN, &md->flags))
2451 thaw_bdev(md->bdev, md->frozen_sb);
2452 md->frozen_sb = NULL;
2453 clear_bit(DMF_FROZEN, &md->flags);
2457 * We need to be able to change a mapping table under a mounted
2458 * filesystem. For example we might want to move some data in
2459 * the background. Before the table can be swapped with
2460 * dm_bind_table, dm_suspend must be called to flush any in
2461 * flight bios and ensure that any further io gets deferred.
2464 * Suspend mechanism in request-based dm.
2466 * 1. Flush all I/Os by lock_fs() if needed.
2467 * 2. Stop dispatching any I/O by stopping the request_queue.
2468 * 3. Wait for all in-flight I/Os to be completed or requeued.
2470 * To abort suspend, start the request_queue.
2472 int dm_suspend(struct mapped_device *md, unsigned suspend_flags)
2474 struct dm_table *map = NULL;
2476 int do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG ? 1 : 0;
2477 int noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG ? 1 : 0;
2479 mutex_lock(&md->suspend_lock);
2481 if (dm_suspended_md(md)) {
2486 map = dm_get_live_table(md);
2489 * DMF_NOFLUSH_SUSPENDING must be set before presuspend.
2490 * This flag is cleared before dm_suspend returns.
2493 set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2495 /* This does not get reverted if there's an error later. */
2496 dm_table_presuspend_targets(map);
2499 * Flush I/O to the device.
2500 * Any I/O submitted after lock_fs() may not be flushed.
2501 * noflush takes precedence over do_lockfs.
2502 * (lock_fs() flushes I/Os and waits for them to complete.)
2504 if (!noflush && do_lockfs) {
2511 * Here we must make sure that no processes are submitting requests
2512 * to target drivers i.e. no one may be executing
2513 * __split_and_process_bio. This is called from dm_request and
2516 * To get all processes out of __split_and_process_bio in dm_request,
2517 * we take the write lock. To prevent any process from reentering
2518 * __split_and_process_bio from dm_request and quiesce the thread
2519 * (dm_wq_work), we set BMF_BLOCK_IO_FOR_SUSPEND and call
2520 * flush_workqueue(md->wq).
2522 down_write(&md->io_lock);
2523 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2524 up_write(&md->io_lock);
2527 * Stop md->queue before flushing md->wq in case request-based
2528 * dm defers requests to md->wq from md->queue.
2530 if (dm_request_based(md))
2531 stop_queue(md->queue);
2533 flush_workqueue(md->wq);
2536 * At this point no more requests are entering target request routines.
2537 * We call dm_wait_for_completion to wait for all existing requests
2540 r = dm_wait_for_completion(md, TASK_INTERRUPTIBLE);
2542 down_write(&md->io_lock);
2544 clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2545 up_write(&md->io_lock);
2547 /* were we interrupted ? */
2551 if (dm_request_based(md))
2552 start_queue(md->queue);
2555 goto out; /* pushback list is already flushed, so skip flush */
2559 * If dm_wait_for_completion returned 0, the device is completely
2560 * quiescent now. There is no request-processing activity. All new
2561 * requests are being added to md->deferred list.
2564 set_bit(DMF_SUSPENDED, &md->flags);
2566 dm_table_postsuspend_targets(map);
2572 mutex_unlock(&md->suspend_lock);
2576 int dm_resume(struct mapped_device *md)
2579 struct dm_table *map = NULL;
2581 mutex_lock(&md->suspend_lock);
2582 if (!dm_suspended_md(md))
2585 map = dm_get_live_table(md);
2586 if (!map || !dm_table_get_size(map))
2589 r = dm_table_resume_targets(map);
2596 * Flushing deferred I/Os must be done after targets are resumed
2597 * so that mapping of targets can work correctly.
2598 * Request-based dm is queueing the deferred I/Os in its request_queue.
2600 if (dm_request_based(md))
2601 start_queue(md->queue);
2605 clear_bit(DMF_SUSPENDED, &md->flags);
2610 mutex_unlock(&md->suspend_lock);
2615 /*-----------------------------------------------------------------
2616 * Event notification.
2617 *---------------------------------------------------------------*/
2618 int dm_kobject_uevent(struct mapped_device *md, enum kobject_action action,
2621 char udev_cookie[DM_COOKIE_LENGTH];
2622 char *envp[] = { udev_cookie, NULL };
2625 return kobject_uevent(&disk_to_dev(md->disk)->kobj, action);
2627 snprintf(udev_cookie, DM_COOKIE_LENGTH, "%s=%u",
2628 DM_COOKIE_ENV_VAR_NAME, cookie);
2629 return kobject_uevent_env(&disk_to_dev(md->disk)->kobj,
2634 uint32_t dm_next_uevent_seq(struct mapped_device *md)
2636 return atomic_add_return(1, &md->uevent_seq);
2639 uint32_t dm_get_event_nr(struct mapped_device *md)
2641 return atomic_read(&md->event_nr);
2644 int dm_wait_event(struct mapped_device *md, int event_nr)
2646 return wait_event_interruptible(md->eventq,
2647 (event_nr != atomic_read(&md->event_nr)));
2650 void dm_uevent_add(struct mapped_device *md, struct list_head *elist)
2652 unsigned long flags;
2654 spin_lock_irqsave(&md->uevent_lock, flags);
2655 list_add(elist, &md->uevent_list);
2656 spin_unlock_irqrestore(&md->uevent_lock, flags);
2660 * The gendisk is only valid as long as you have a reference
2663 struct gendisk *dm_disk(struct mapped_device *md)
2668 struct kobject *dm_kobject(struct mapped_device *md)
2674 * struct mapped_device should not be exported outside of dm.c
2675 * so use this check to verify that kobj is part of md structure
2677 struct mapped_device *dm_get_from_kobject(struct kobject *kobj)
2679 struct mapped_device *md;
2681 md = container_of(kobj, struct mapped_device, kobj);
2682 if (&md->kobj != kobj)
2685 if (test_bit(DMF_FREEING, &md->flags) ||
2693 int dm_suspended_md(struct mapped_device *md)
2695 return test_bit(DMF_SUSPENDED, &md->flags);
2698 int dm_suspended(struct dm_target *ti)
2700 return dm_suspended_md(dm_table_get_md(ti->table));
2702 EXPORT_SYMBOL_GPL(dm_suspended);
2704 int dm_noflush_suspending(struct dm_target *ti)
2706 return __noflush_suspending(dm_table_get_md(ti->table));
2708 EXPORT_SYMBOL_GPL(dm_noflush_suspending);
2710 struct dm_md_mempools *dm_alloc_md_mempools(unsigned type, unsigned integrity)
2712 struct dm_md_mempools *pools = kmalloc(sizeof(*pools), GFP_KERNEL);
2713 unsigned int pool_size = (type == DM_TYPE_BIO_BASED) ? 16 : MIN_IOS;
2718 pools->io_pool = (type == DM_TYPE_BIO_BASED) ?
2719 mempool_create_slab_pool(MIN_IOS, _io_cache) :
2720 mempool_create_slab_pool(MIN_IOS, _rq_bio_info_cache);
2721 if (!pools->io_pool)
2722 goto free_pools_and_out;
2724 pools->tio_pool = (type == DM_TYPE_BIO_BASED) ?
2725 mempool_create_slab_pool(MIN_IOS, _tio_cache) :
2726 mempool_create_slab_pool(MIN_IOS, _rq_tio_cache);
2727 if (!pools->tio_pool)
2728 goto free_io_pool_and_out;
2730 pools->bs = bioset_create(pool_size, 0);
2732 goto free_tio_pool_and_out;
2734 if (integrity && bioset_integrity_create(pools->bs, pool_size))
2735 goto free_bioset_and_out;
2739 free_bioset_and_out:
2740 bioset_free(pools->bs);
2742 free_tio_pool_and_out:
2743 mempool_destroy(pools->tio_pool);
2745 free_io_pool_and_out:
2746 mempool_destroy(pools->io_pool);
2754 void dm_free_md_mempools(struct dm_md_mempools *pools)
2760 mempool_destroy(pools->io_pool);
2762 if (pools->tio_pool)
2763 mempool_destroy(pools->tio_pool);
2766 bioset_free(pools->bs);
2771 static const struct block_device_operations dm_blk_dops = {
2772 .open = dm_blk_open,
2773 .release = dm_blk_close,
2774 .ioctl = dm_blk_ioctl,
2775 .getgeo = dm_blk_getgeo,
2776 .owner = THIS_MODULE
2779 EXPORT_SYMBOL(dm_get_mapinfo);
2784 module_init(dm_init);
2785 module_exit(dm_exit);
2787 module_param(major, uint, 0);
2788 MODULE_PARM_DESC(major, "The major number of the device mapper");
2789 MODULE_DESCRIPTION(DM_NAME " driver");
2790 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
2791 MODULE_LICENSE("GPL");