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);
54 * One of these is allocated per bio.
57 struct mapped_device *md;
61 unsigned long start_time;
62 spinlock_t endio_lock;
66 * For request-based dm.
67 * One of these is allocated per request.
69 struct dm_rq_target_io {
70 struct mapped_device *md;
72 struct request *orig, clone;
78 * For request-based dm - the bio clones we allocate are embedded in these
81 * We allocate these with bio_alloc_bioset, using the front_pad parameter when
82 * the bioset is created - this means the bio has to come at the end of the
85 struct dm_rq_clone_bio_info {
87 struct dm_rq_target_io *tio;
91 union map_info *dm_get_mapinfo(struct bio *bio)
93 if (bio && bio->bi_private)
94 return &((struct dm_target_io *)bio->bi_private)->info;
98 union map_info *dm_get_rq_mapinfo(struct request *rq)
100 if (rq && rq->end_io_data)
101 return &((struct dm_rq_target_io *)rq->end_io_data)->info;
104 EXPORT_SYMBOL_GPL(dm_get_rq_mapinfo);
106 #define MINOR_ALLOCED ((void *)-1)
109 * Bits for the md->flags field.
111 #define DMF_BLOCK_IO_FOR_SUSPEND 0
112 #define DMF_SUSPENDED 1
114 #define DMF_FREEING 3
115 #define DMF_DELETING 4
116 #define DMF_NOFLUSH_SUSPENDING 5
117 #define DMF_MERGE_IS_OPTIONAL 6
120 * Work processed by per-device workqueue.
122 struct mapped_device {
123 struct rw_semaphore io_lock;
124 struct mutex suspend_lock;
131 struct request_queue *queue;
133 /* Protect queue and type against concurrent access. */
134 struct mutex type_lock;
136 struct target_type *immutable_target_type;
138 struct gendisk *disk;
144 * A list of ios that arrived while we were suspended.
147 wait_queue_head_t wait;
148 struct work_struct work;
149 struct bio_list deferred;
150 spinlock_t deferred_lock;
153 * Processing queue (flush)
155 struct workqueue_struct *wq;
158 * The current mapping.
160 struct dm_table *map;
163 * io objects are allocated from here.
174 wait_queue_head_t eventq;
176 struct list_head uevent_list;
177 spinlock_t uevent_lock; /* Protect access to uevent_list */
180 * freeze/thaw support require holding onto a super block
182 struct super_block *frozen_sb;
183 struct block_device *bdev;
185 /* forced geometry settings */
186 struct hd_geometry geometry;
191 /* zero-length flush that will be cloned and submitted to targets */
192 struct bio flush_bio;
196 * For mempools pre-allocation at the table loading time.
198 struct dm_md_mempools {
205 static struct kmem_cache *_io_cache;
206 static struct kmem_cache *_rq_tio_cache;
209 * Unused now, and needs to be deleted. But since io_pool is overloaded and it's
210 * still used for _io_cache, I'm leaving this for a later cleanup
212 static struct kmem_cache *_rq_bio_info_cache;
214 static int __init local_init(void)
218 /* allocate a slab for the dm_ios */
219 _io_cache = KMEM_CACHE(dm_io, 0);
223 _rq_tio_cache = KMEM_CACHE(dm_rq_target_io, 0);
225 goto out_free_io_cache;
227 _rq_bio_info_cache = KMEM_CACHE(dm_rq_clone_bio_info, 0);
228 if (!_rq_bio_info_cache)
229 goto out_free_rq_tio_cache;
231 r = dm_uevent_init();
233 goto out_free_rq_bio_info_cache;
236 r = register_blkdev(_major, _name);
238 goto out_uevent_exit;
247 out_free_rq_bio_info_cache:
248 kmem_cache_destroy(_rq_bio_info_cache);
249 out_free_rq_tio_cache:
250 kmem_cache_destroy(_rq_tio_cache);
252 kmem_cache_destroy(_io_cache);
257 static void local_exit(void)
259 kmem_cache_destroy(_rq_bio_info_cache);
260 kmem_cache_destroy(_rq_tio_cache);
261 kmem_cache_destroy(_io_cache);
262 unregister_blkdev(_major, _name);
267 DMINFO("cleaned up");
270 static int (*_inits[])(void) __initdata = {
280 static void (*_exits[])(void) = {
290 static int __init dm_init(void)
292 const int count = ARRAY_SIZE(_inits);
296 for (i = 0; i < count; i++) {
311 static void __exit dm_exit(void)
313 int i = ARRAY_SIZE(_exits);
319 * Should be empty by this point.
321 idr_destroy(&_minor_idr);
325 * Block device functions
327 int dm_deleting_md(struct mapped_device *md)
329 return test_bit(DMF_DELETING, &md->flags);
332 static int dm_blk_open(struct block_device *bdev, fmode_t mode)
334 struct mapped_device *md;
336 spin_lock(&_minor_lock);
338 md = bdev->bd_disk->private_data;
342 if (test_bit(DMF_FREEING, &md->flags) ||
343 dm_deleting_md(md)) {
349 atomic_inc(&md->open_count);
352 spin_unlock(&_minor_lock);
354 return md ? 0 : -ENXIO;
357 static int dm_blk_close(struct gendisk *disk, fmode_t mode)
359 struct mapped_device *md = disk->private_data;
361 spin_lock(&_minor_lock);
363 atomic_dec(&md->open_count);
366 spin_unlock(&_minor_lock);
371 int dm_open_count(struct mapped_device *md)
373 return atomic_read(&md->open_count);
377 * Guarantees nothing is using the device before it's deleted.
379 int dm_lock_for_deletion(struct mapped_device *md)
383 spin_lock(&_minor_lock);
385 if (dm_open_count(md))
388 set_bit(DMF_DELETING, &md->flags);
390 spin_unlock(&_minor_lock);
395 static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo)
397 struct mapped_device *md = bdev->bd_disk->private_data;
399 return dm_get_geometry(md, geo);
402 static int dm_blk_ioctl(struct block_device *bdev, fmode_t mode,
403 unsigned int cmd, unsigned long arg)
405 struct mapped_device *md = bdev->bd_disk->private_data;
406 struct dm_table *map = dm_get_live_table(md);
407 struct dm_target *tgt;
410 if (!map || !dm_table_get_size(map))
413 /* We only support devices that have a single target */
414 if (dm_table_get_num_targets(map) != 1)
417 tgt = dm_table_get_target(map, 0);
419 if (dm_suspended_md(md)) {
424 if (tgt->type->ioctl)
425 r = tgt->type->ioctl(tgt, cmd, arg);
433 static struct dm_io *alloc_io(struct mapped_device *md)
435 return mempool_alloc(md->io_pool, GFP_NOIO);
438 static void free_io(struct mapped_device *md, struct dm_io *io)
440 mempool_free(io, md->io_pool);
443 static void free_tio(struct mapped_device *md, struct dm_target_io *tio)
445 bio_put(&tio->clone);
448 static struct dm_rq_target_io *alloc_rq_tio(struct mapped_device *md,
451 return mempool_alloc(md->tio_pool, gfp_mask);
454 static void free_rq_tio(struct dm_rq_target_io *tio)
456 mempool_free(tio, tio->md->tio_pool);
459 static int md_in_flight(struct mapped_device *md)
461 return atomic_read(&md->pending[READ]) +
462 atomic_read(&md->pending[WRITE]);
465 static void start_io_acct(struct dm_io *io)
467 struct mapped_device *md = io->md;
469 int rw = bio_data_dir(io->bio);
471 io->start_time = jiffies;
473 cpu = part_stat_lock();
474 part_round_stats(cpu, &dm_disk(md)->part0);
476 atomic_set(&dm_disk(md)->part0.in_flight[rw],
477 atomic_inc_return(&md->pending[rw]));
480 static void end_io_acct(struct dm_io *io)
482 struct mapped_device *md = io->md;
483 struct bio *bio = io->bio;
484 unsigned long duration = jiffies - io->start_time;
486 int rw = bio_data_dir(bio);
488 cpu = part_stat_lock();
489 part_round_stats(cpu, &dm_disk(md)->part0);
490 part_stat_add(cpu, &dm_disk(md)->part0, ticks[rw], duration);
494 * After this is decremented the bio must not be touched if it is
497 pending = atomic_dec_return(&md->pending[rw]);
498 atomic_set(&dm_disk(md)->part0.in_flight[rw], pending);
499 pending += atomic_read(&md->pending[rw^0x1]);
501 /* nudge anyone waiting on suspend queue */
507 * Add the bio to the list of deferred io.
509 static void queue_io(struct mapped_device *md, struct bio *bio)
513 spin_lock_irqsave(&md->deferred_lock, flags);
514 bio_list_add(&md->deferred, bio);
515 spin_unlock_irqrestore(&md->deferred_lock, flags);
516 queue_work(md->wq, &md->work);
520 * Everyone (including functions in this file), should use this
521 * function to access the md->map field, and make sure they call
522 * dm_table_put() when finished.
524 struct dm_table *dm_get_live_table(struct mapped_device *md)
529 read_lock_irqsave(&md->map_lock, flags);
533 read_unlock_irqrestore(&md->map_lock, flags);
539 * Get the geometry associated with a dm device
541 int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo)
549 * Set the geometry of a device.
551 int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo)
553 sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors;
555 if (geo->start > sz) {
556 DMWARN("Start sector is beyond the geometry limits.");
565 /*-----------------------------------------------------------------
567 * A more elegant soln is in the works that uses the queue
568 * merge fn, unfortunately there are a couple of changes to
569 * the block layer that I want to make for this. So in the
570 * interests of getting something for people to use I give
571 * you this clearly demarcated crap.
572 *---------------------------------------------------------------*/
574 static int __noflush_suspending(struct mapped_device *md)
576 return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
580 * Decrements the number of outstanding ios that a bio has been
581 * cloned into, completing the original io if necc.
583 static void dec_pending(struct dm_io *io, int error)
588 struct mapped_device *md = io->md;
590 /* Push-back supersedes any I/O errors */
591 if (unlikely(error)) {
592 spin_lock_irqsave(&io->endio_lock, flags);
593 if (!(io->error > 0 && __noflush_suspending(md)))
595 spin_unlock_irqrestore(&io->endio_lock, flags);
598 if (atomic_dec_and_test(&io->io_count)) {
599 if (io->error == DM_ENDIO_REQUEUE) {
601 * Target requested pushing back the I/O.
603 spin_lock_irqsave(&md->deferred_lock, flags);
604 if (__noflush_suspending(md))
605 bio_list_add_head(&md->deferred, io->bio);
607 /* noflush suspend was interrupted. */
609 spin_unlock_irqrestore(&md->deferred_lock, flags);
612 io_error = io->error;
617 if (io_error == DM_ENDIO_REQUEUE)
620 if ((bio->bi_rw & REQ_FLUSH) && bio->bi_size) {
622 * Preflush done for flush with data, reissue
625 bio->bi_rw &= ~REQ_FLUSH;
628 /* done with normal IO or empty flush */
629 bio_endio(bio, io_error);
634 static void clone_endio(struct bio *bio, int error)
637 struct dm_target_io *tio = bio->bi_private;
638 struct dm_io *io = tio->io;
639 struct mapped_device *md = tio->io->md;
640 dm_endio_fn endio = tio->ti->type->end_io;
642 if (!bio_flagged(bio, BIO_UPTODATE) && !error)
646 r = endio(tio->ti, bio, error);
647 if (r < 0 || r == DM_ENDIO_REQUEUE)
649 * error and requeue request are handled
653 else if (r == DM_ENDIO_INCOMPLETE)
654 /* The target will handle the io */
657 DMWARN("unimplemented target endio return value: %d", r);
663 dec_pending(io, error);
667 * Partial completion handling for request-based dm
669 static void end_clone_bio(struct bio *clone, int error)
671 struct dm_rq_clone_bio_info *info = clone->bi_private;
672 struct dm_rq_target_io *tio = info->tio;
673 struct bio *bio = info->orig;
674 unsigned int nr_bytes = info->orig->bi_size;
680 * An error has already been detected on the request.
681 * Once error occurred, just let clone->end_io() handle
687 * Don't notice the error to the upper layer yet.
688 * The error handling decision is made by the target driver,
689 * when the request is completed.
696 * I/O for the bio successfully completed.
697 * Notice the data completion to the upper layer.
701 * bios are processed from the head of the list.
702 * So the completing bio should always be rq->bio.
703 * If it's not, something wrong is happening.
705 if (tio->orig->bio != bio)
706 DMERR("bio completion is going in the middle of the request");
709 * Update the original request.
710 * Do not use blk_end_request() here, because it may complete
711 * the original request before the clone, and break the ordering.
713 blk_update_request(tio->orig, 0, nr_bytes);
717 * Don't touch any member of the md after calling this function because
718 * the md may be freed in dm_put() at the end of this function.
719 * Or do dm_get() before calling this function and dm_put() later.
721 static void rq_completed(struct mapped_device *md, int rw, int run_queue)
723 atomic_dec(&md->pending[rw]);
725 /* nudge anyone waiting on suspend queue */
726 if (!md_in_flight(md))
730 * Run this off this callpath, as drivers could invoke end_io while
731 * inside their request_fn (and holding the queue lock). Calling
732 * back into ->request_fn() could deadlock attempting to grab the
736 blk_run_queue_async(md->queue);
739 * dm_put() must be at the end of this function. See the comment above
744 static void free_rq_clone(struct request *clone)
746 struct dm_rq_target_io *tio = clone->end_io_data;
748 blk_rq_unprep_clone(clone);
753 * Complete the clone and the original request.
754 * Must be called without queue lock.
756 static void dm_end_request(struct request *clone, int error)
758 int rw = rq_data_dir(clone);
759 struct dm_rq_target_io *tio = clone->end_io_data;
760 struct mapped_device *md = tio->md;
761 struct request *rq = tio->orig;
763 if (rq->cmd_type == REQ_TYPE_BLOCK_PC) {
764 rq->errors = clone->errors;
765 rq->resid_len = clone->resid_len;
769 * We are using the sense buffer of the original
771 * So setting the length of the sense data is enough.
773 rq->sense_len = clone->sense_len;
776 free_rq_clone(clone);
777 blk_end_request_all(rq, error);
778 rq_completed(md, rw, true);
781 static void dm_unprep_request(struct request *rq)
783 struct request *clone = rq->special;
786 rq->cmd_flags &= ~REQ_DONTPREP;
788 free_rq_clone(clone);
792 * Requeue the original request of a clone.
794 void dm_requeue_unmapped_request(struct request *clone)
796 int rw = rq_data_dir(clone);
797 struct dm_rq_target_io *tio = clone->end_io_data;
798 struct mapped_device *md = tio->md;
799 struct request *rq = tio->orig;
800 struct request_queue *q = rq->q;
803 dm_unprep_request(rq);
805 spin_lock_irqsave(q->queue_lock, flags);
806 blk_requeue_request(q, rq);
807 spin_unlock_irqrestore(q->queue_lock, flags);
809 rq_completed(md, rw, 0);
811 EXPORT_SYMBOL_GPL(dm_requeue_unmapped_request);
813 static void __stop_queue(struct request_queue *q)
818 static void stop_queue(struct request_queue *q)
822 spin_lock_irqsave(q->queue_lock, flags);
824 spin_unlock_irqrestore(q->queue_lock, flags);
827 static void __start_queue(struct request_queue *q)
829 if (blk_queue_stopped(q))
833 static void start_queue(struct request_queue *q)
837 spin_lock_irqsave(q->queue_lock, flags);
839 spin_unlock_irqrestore(q->queue_lock, flags);
842 static void dm_done(struct request *clone, int error, bool mapped)
845 struct dm_rq_target_io *tio = clone->end_io_data;
846 dm_request_endio_fn rq_end_io = NULL;
849 rq_end_io = tio->ti->type->rq_end_io;
851 if (mapped && rq_end_io)
852 r = rq_end_io(tio->ti, clone, error, &tio->info);
856 /* The target wants to complete the I/O */
857 dm_end_request(clone, r);
858 else if (r == DM_ENDIO_INCOMPLETE)
859 /* The target will handle the I/O */
861 else if (r == DM_ENDIO_REQUEUE)
862 /* The target wants to requeue the I/O */
863 dm_requeue_unmapped_request(clone);
865 DMWARN("unimplemented target endio return value: %d", r);
871 * Request completion handler for request-based dm
873 static void dm_softirq_done(struct request *rq)
876 struct request *clone = rq->completion_data;
877 struct dm_rq_target_io *tio = clone->end_io_data;
879 if (rq->cmd_flags & REQ_FAILED)
882 dm_done(clone, tio->error, mapped);
886 * Complete the clone and the original request with the error status
887 * through softirq context.
889 static void dm_complete_request(struct request *clone, int error)
891 struct dm_rq_target_io *tio = clone->end_io_data;
892 struct request *rq = tio->orig;
895 rq->completion_data = clone;
896 blk_complete_request(rq);
900 * Complete the not-mapped clone and the original request with the error status
901 * through softirq context.
902 * Target's rq_end_io() function isn't called.
903 * This may be used when the target's map_rq() function fails.
905 void dm_kill_unmapped_request(struct request *clone, int error)
907 struct dm_rq_target_io *tio = clone->end_io_data;
908 struct request *rq = tio->orig;
910 rq->cmd_flags |= REQ_FAILED;
911 dm_complete_request(clone, error);
913 EXPORT_SYMBOL_GPL(dm_kill_unmapped_request);
916 * Called with the queue lock held
918 static void end_clone_request(struct request *clone, int error)
921 * For just cleaning up the information of the queue in which
922 * the clone was dispatched.
923 * The clone is *NOT* freed actually here because it is alloced from
924 * dm own mempool and REQ_ALLOCED isn't set in clone->cmd_flags.
926 __blk_put_request(clone->q, clone);
929 * Actual request completion is done in a softirq context which doesn't
930 * hold the queue lock. Otherwise, deadlock could occur because:
931 * - another request may be submitted by the upper level driver
932 * of the stacking during the completion
933 * - the submission which requires queue lock may be done
936 dm_complete_request(clone, error);
940 * Return maximum size of I/O possible at the supplied sector up to the current
943 static sector_t max_io_len_target_boundary(sector_t sector, struct dm_target *ti)
945 sector_t target_offset = dm_target_offset(ti, sector);
947 return ti->len - target_offset;
950 static sector_t max_io_len(sector_t sector, struct dm_target *ti)
952 sector_t len = max_io_len_target_boundary(sector, ti);
953 sector_t offset, max_len;
956 * Does the target need to split even further?
958 if (ti->max_io_len) {
959 offset = dm_target_offset(ti, sector);
960 if (unlikely(ti->max_io_len & (ti->max_io_len - 1)))
961 max_len = sector_div(offset, ti->max_io_len);
963 max_len = offset & (ti->max_io_len - 1);
964 max_len = ti->max_io_len - max_len;
973 int dm_set_target_max_io_len(struct dm_target *ti, sector_t len)
975 if (len > UINT_MAX) {
976 DMERR("Specified maximum size of target IO (%llu) exceeds limit (%u)",
977 (unsigned long long)len, UINT_MAX);
978 ti->error = "Maximum size of target IO is too large";
982 ti->max_io_len = (uint32_t) len;
986 EXPORT_SYMBOL_GPL(dm_set_target_max_io_len);
988 static void __map_bio(struct dm_target_io *tio)
992 struct mapped_device *md;
993 struct bio *clone = &tio->clone;
994 struct dm_target *ti = tio->ti;
996 clone->bi_end_io = clone_endio;
997 clone->bi_private = tio;
1000 * Map the clone. If r == 0 we don't need to do
1001 * anything, the target has assumed ownership of
1004 atomic_inc(&tio->io->io_count);
1005 sector = clone->bi_sector;
1006 r = ti->type->map(ti, clone);
1007 if (r == DM_MAPIO_REMAPPED) {
1008 /* the bio has been remapped so dispatch it */
1010 trace_block_bio_remap(bdev_get_queue(clone->bi_bdev), clone,
1011 tio->io->bio->bi_bdev->bd_dev, sector);
1013 generic_make_request(clone);
1014 } else if (r < 0 || r == DM_MAPIO_REQUEUE) {
1015 /* error the io and bail out, or requeue it if needed */
1017 dec_pending(tio->io, r);
1020 DMWARN("unimplemented target map return value: %d", r);
1026 struct mapped_device *md;
1027 struct dm_table *map;
1031 sector_t sector_count;
1035 static void bio_setup_sector(struct bio *bio, sector_t sector, sector_t len)
1037 bio->bi_sector = sector;
1038 bio->bi_size = to_bytes(len);
1041 static void bio_setup_bv(struct bio *bio, unsigned short idx, unsigned short bv_count)
1044 bio->bi_vcnt = idx + bv_count;
1045 bio->bi_flags &= ~(1 << BIO_SEG_VALID);
1048 static void clone_bio_integrity(struct bio *bio, struct bio *clone,
1049 unsigned short idx, unsigned len, unsigned offset,
1052 if (!bio_integrity(bio))
1055 bio_integrity_clone(clone, bio, GFP_NOIO);
1058 bio_integrity_trim(clone, bio_sector_offset(bio, idx, offset), len);
1062 * Creates a little bio that just does part of a bvec.
1064 static void clone_split_bio(struct dm_target_io *tio, struct bio *bio,
1065 sector_t sector, unsigned short idx,
1066 unsigned offset, unsigned len)
1068 struct bio *clone = &tio->clone;
1069 struct bio_vec *bv = bio->bi_io_vec + idx;
1071 *clone->bi_io_vec = *bv;
1073 bio_setup_sector(clone, sector, len);
1075 clone->bi_bdev = bio->bi_bdev;
1076 clone->bi_rw = bio->bi_rw;
1078 clone->bi_io_vec->bv_offset = offset;
1079 clone->bi_io_vec->bv_len = clone->bi_size;
1080 clone->bi_flags |= 1 << BIO_CLONED;
1082 clone_bio_integrity(bio, clone, idx, len, offset, 1);
1086 * Creates a bio that consists of range of complete bvecs.
1088 static void clone_bio(struct dm_target_io *tio, struct bio *bio,
1089 sector_t sector, unsigned short idx,
1090 unsigned short bv_count, unsigned int len)
1092 struct bio *clone = &tio->clone;
1095 __bio_clone(clone, bio);
1096 bio_setup_sector(clone, sector, len);
1097 bio_setup_bv(clone, idx, bv_count);
1099 if (idx != bio->bi_idx || clone->bi_size < bio->bi_size)
1101 clone_bio_integrity(bio, clone, idx, len, 0, trim);
1104 static struct dm_target_io *alloc_tio(struct clone_info *ci,
1105 struct dm_target *ti, int nr_iovecs,
1106 unsigned target_bio_nr)
1108 struct dm_target_io *tio;
1111 clone = bio_alloc_bioset(GFP_NOIO, nr_iovecs, ci->md->bs);
1112 tio = container_of(clone, struct dm_target_io, clone);
1116 memset(&tio->info, 0, sizeof(tio->info));
1117 tio->target_bio_nr = target_bio_nr;
1122 static void __clone_and_map_simple_bio(struct clone_info *ci,
1123 struct dm_target *ti,
1124 unsigned target_bio_nr, sector_t len)
1126 struct dm_target_io *tio = alloc_tio(ci, ti, ci->bio->bi_max_vecs, target_bio_nr);
1127 struct bio *clone = &tio->clone;
1130 * Discard requests require the bio's inline iovecs be initialized.
1131 * ci->bio->bi_max_vecs is BIO_INLINE_VECS anyway, for both flush
1132 * and discard, so no need for concern about wasted bvec allocations.
1134 __bio_clone(clone, ci->bio);
1136 bio_setup_sector(clone, ci->sector, len);
1141 static void __send_duplicate_bios(struct clone_info *ci, struct dm_target *ti,
1142 unsigned num_bios, sector_t len)
1144 unsigned target_bio_nr;
1146 for (target_bio_nr = 0; target_bio_nr < num_bios; target_bio_nr++)
1147 __clone_and_map_simple_bio(ci, ti, target_bio_nr, len);
1150 static int __send_empty_flush(struct clone_info *ci)
1152 unsigned target_nr = 0;
1153 struct dm_target *ti;
1155 BUG_ON(bio_has_data(ci->bio));
1156 while ((ti = dm_table_get_target(ci->map, target_nr++)))
1157 __send_duplicate_bios(ci, ti, ti->num_flush_bios, 0);
1162 static void __clone_and_map_data_bio(struct clone_info *ci,
1163 struct dm_target *ti)
1165 struct bio *bio = ci->bio;
1166 struct dm_target_io *tio;
1168 tio = alloc_tio(ci, ti, bio->bi_max_vecs, 0);
1169 clone_bio(tio, bio, ci->sector, ci->idx, bio->bi_vcnt - ci->idx,
1172 ci->sector_count = 0;
1175 typedef unsigned (*get_num_bios_fn)(struct dm_target *ti);
1177 static unsigned get_num_discard_bios(struct dm_target *ti)
1179 return ti->num_discard_bios;
1182 static unsigned get_num_write_same_bios(struct dm_target *ti)
1184 return ti->num_write_same_bios;
1187 typedef bool (*is_split_required_fn)(struct dm_target *ti);
1189 static bool is_split_required_for_discard(struct dm_target *ti)
1191 return ti->split_discard_bios;
1194 static int __send_changing_extent_only(struct clone_info *ci,
1195 get_num_bios_fn get_num_bios,
1196 is_split_required_fn is_split_required)
1198 struct dm_target *ti;
1203 ti = dm_table_find_target(ci->map, ci->sector);
1204 if (!dm_target_is_valid(ti))
1208 * Even though the device advertised support for this type of
1209 * request, that does not mean every target supports it, and
1210 * reconfiguration might also have changed that since the
1211 * check was performed.
1213 num_bios = get_num_bios ? get_num_bios(ti) : 0;
1217 if (is_split_required && !is_split_required(ti))
1218 len = min(ci->sector_count, max_io_len_target_boundary(ci->sector, ti));
1220 len = min(ci->sector_count, max_io_len(ci->sector, ti));
1222 __send_duplicate_bios(ci, ti, num_bios, len);
1225 } while (ci->sector_count -= len);
1230 static int __send_discard(struct clone_info *ci)
1232 return __send_changing_extent_only(ci, get_num_discard_bios,
1233 is_split_required_for_discard);
1236 static int __send_write_same(struct clone_info *ci)
1238 return __send_changing_extent_only(ci, get_num_write_same_bios, NULL);
1241 static int __split_and_process_non_flush(struct clone_info *ci)
1243 struct bio *bio = ci->bio;
1244 struct dm_target *ti;
1245 sector_t len = 0, max;
1246 struct dm_target_io *tio;
1248 if (unlikely(bio->bi_rw & REQ_DISCARD))
1249 return __send_discard(ci);
1250 else if (unlikely(bio->bi_rw & REQ_WRITE_SAME))
1251 return __send_write_same(ci);
1253 ti = dm_table_find_target(ci->map, ci->sector);
1254 if (!dm_target_is_valid(ti))
1257 max = max_io_len(ci->sector, ti);
1259 if (ci->sector_count <= max) {
1261 * Optimise for the simple case where we can do all of
1262 * the remaining io with a single clone.
1264 __clone_and_map_data_bio(ci, ti);
1266 } else if (to_sector(bio->bi_io_vec[ci->idx].bv_len) <= max) {
1268 * There are some bvecs that don't span targets.
1269 * Do as many of these as possible.
1272 sector_t remaining = max;
1275 for (i = ci->idx; remaining && (i < bio->bi_vcnt); i++) {
1276 bv_len = to_sector(bio->bi_io_vec[i].bv_len);
1278 if (bv_len > remaining)
1281 remaining -= bv_len;
1285 tio = alloc_tio(ci, ti, bio->bi_max_vecs, 0);
1286 clone_bio(tio, bio, ci->sector, ci->idx, i - ci->idx, len);
1290 ci->sector_count -= len;
1295 * Handle a bvec that must be split between two or more targets.
1297 struct bio_vec *bv = bio->bi_io_vec + ci->idx;
1298 sector_t remaining = to_sector(bv->bv_len);
1299 unsigned int offset = 0;
1303 ti = dm_table_find_target(ci->map, ci->sector);
1304 if (!dm_target_is_valid(ti))
1307 max = max_io_len(ci->sector, ti);
1310 len = min(remaining, max);
1312 tio = alloc_tio(ci, ti, 1, 0);
1313 clone_split_bio(tio, bio, ci->sector, ci->idx,
1314 bv->bv_offset + offset, len);
1319 ci->sector_count -= len;
1320 offset += to_bytes(len);
1321 } while (remaining -= len);
1330 * Entry point to split a bio into clones and submit them to the targets.
1332 static void __split_and_process_bio(struct mapped_device *md, struct bio *bio)
1334 struct clone_info ci;
1337 ci.map = dm_get_live_table(md);
1338 if (unlikely(!ci.map)) {
1344 ci.io = alloc_io(md);
1346 atomic_set(&ci.io->io_count, 1);
1349 spin_lock_init(&ci.io->endio_lock);
1350 ci.sector = bio->bi_sector;
1351 ci.idx = bio->bi_idx;
1353 start_io_acct(ci.io);
1355 if (bio->bi_rw & REQ_FLUSH) {
1356 ci.bio = &ci.md->flush_bio;
1357 ci.sector_count = 0;
1358 error = __send_empty_flush(&ci);
1359 /* dec_pending submits any data associated with flush */
1362 ci.sector_count = bio_sectors(bio);
1363 while (ci.sector_count && !error)
1364 error = __split_and_process_non_flush(&ci);
1367 /* drop the extra reference count */
1368 dec_pending(ci.io, error);
1369 dm_table_put(ci.map);
1371 /*-----------------------------------------------------------------
1373 *---------------------------------------------------------------*/
1375 static int dm_merge_bvec(struct request_queue *q,
1376 struct bvec_merge_data *bvm,
1377 struct bio_vec *biovec)
1379 struct mapped_device *md = q->queuedata;
1380 struct dm_table *map = dm_get_live_table(md);
1381 struct dm_target *ti;
1382 sector_t max_sectors;
1388 ti = dm_table_find_target(map, bvm->bi_sector);
1389 if (!dm_target_is_valid(ti))
1393 * Find maximum amount of I/O that won't need splitting
1395 max_sectors = min(max_io_len(bvm->bi_sector, ti),
1396 (sector_t) BIO_MAX_SECTORS);
1397 max_size = (max_sectors << SECTOR_SHIFT) - bvm->bi_size;
1402 * merge_bvec_fn() returns number of bytes
1403 * it can accept at this offset
1404 * max is precomputed maximal io size
1406 if (max_size && ti->type->merge)
1407 max_size = ti->type->merge(ti, bvm, biovec, max_size);
1409 * If the target doesn't support merge method and some of the devices
1410 * provided their merge_bvec method (we know this by looking at
1411 * queue_max_hw_sectors), then we can't allow bios with multiple vector
1412 * entries. So always set max_size to 0, and the code below allows
1415 else if (queue_max_hw_sectors(q) <= PAGE_SIZE >> 9)
1424 * Always allow an entire first page
1426 if (max_size <= biovec->bv_len && !(bvm->bi_size >> SECTOR_SHIFT))
1427 max_size = biovec->bv_len;
1433 * The request function that just remaps the bio built up by
1436 static void _dm_request(struct request_queue *q, struct bio *bio)
1438 int rw = bio_data_dir(bio);
1439 struct mapped_device *md = q->queuedata;
1442 down_read(&md->io_lock);
1444 cpu = part_stat_lock();
1445 part_stat_inc(cpu, &dm_disk(md)->part0, ios[rw]);
1446 part_stat_add(cpu, &dm_disk(md)->part0, sectors[rw], bio_sectors(bio));
1449 /* if we're suspended, we have to queue this io for later */
1450 if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))) {
1451 up_read(&md->io_lock);
1453 if (bio_rw(bio) != READA)
1460 __split_and_process_bio(md, bio);
1461 up_read(&md->io_lock);
1465 static int dm_request_based(struct mapped_device *md)
1467 return blk_queue_stackable(md->queue);
1470 static void dm_request(struct request_queue *q, struct bio *bio)
1472 struct mapped_device *md = q->queuedata;
1474 if (dm_request_based(md))
1475 blk_queue_bio(q, bio);
1477 _dm_request(q, bio);
1480 void dm_dispatch_request(struct request *rq)
1484 if (blk_queue_io_stat(rq->q))
1485 rq->cmd_flags |= REQ_IO_STAT;
1487 rq->start_time = jiffies;
1488 r = blk_insert_cloned_request(rq->q, rq);
1490 dm_complete_request(rq, r);
1492 EXPORT_SYMBOL_GPL(dm_dispatch_request);
1494 static int dm_rq_bio_constructor(struct bio *bio, struct bio *bio_orig,
1497 struct dm_rq_target_io *tio = data;
1498 struct dm_rq_clone_bio_info *info =
1499 container_of(bio, struct dm_rq_clone_bio_info, clone);
1501 info->orig = bio_orig;
1503 bio->bi_end_io = end_clone_bio;
1504 bio->bi_private = info;
1509 static int setup_clone(struct request *clone, struct request *rq,
1510 struct dm_rq_target_io *tio)
1514 r = blk_rq_prep_clone(clone, rq, tio->md->bs, GFP_ATOMIC,
1515 dm_rq_bio_constructor, tio);
1519 clone->cmd = rq->cmd;
1520 clone->cmd_len = rq->cmd_len;
1521 clone->sense = rq->sense;
1522 clone->buffer = rq->buffer;
1523 clone->end_io = end_clone_request;
1524 clone->end_io_data = tio;
1529 static struct request *clone_rq(struct request *rq, struct mapped_device *md,
1532 struct request *clone;
1533 struct dm_rq_target_io *tio;
1535 tio = alloc_rq_tio(md, gfp_mask);
1543 memset(&tio->info, 0, sizeof(tio->info));
1545 clone = &tio->clone;
1546 if (setup_clone(clone, rq, tio)) {
1556 * Called with the queue lock held.
1558 static int dm_prep_fn(struct request_queue *q, struct request *rq)
1560 struct mapped_device *md = q->queuedata;
1561 struct request *clone;
1563 if (unlikely(rq->special)) {
1564 DMWARN("Already has something in rq->special.");
1565 return BLKPREP_KILL;
1568 clone = clone_rq(rq, md, GFP_ATOMIC);
1570 return BLKPREP_DEFER;
1572 rq->special = clone;
1573 rq->cmd_flags |= REQ_DONTPREP;
1580 * 0 : the request has been processed (not requeued)
1581 * !0 : the request has been requeued
1583 static int map_request(struct dm_target *ti, struct request *clone,
1584 struct mapped_device *md)
1586 int r, requeued = 0;
1587 struct dm_rq_target_io *tio = clone->end_io_data;
1590 r = ti->type->map_rq(ti, clone, &tio->info);
1592 case DM_MAPIO_SUBMITTED:
1593 /* The target has taken the I/O to submit by itself later */
1595 case DM_MAPIO_REMAPPED:
1596 /* The target has remapped the I/O so dispatch it */
1597 trace_block_rq_remap(clone->q, clone, disk_devt(dm_disk(md)),
1598 blk_rq_pos(tio->orig));
1599 dm_dispatch_request(clone);
1601 case DM_MAPIO_REQUEUE:
1602 /* The target wants to requeue the I/O */
1603 dm_requeue_unmapped_request(clone);
1608 DMWARN("unimplemented target map return value: %d", r);
1612 /* The target wants to complete the I/O */
1613 dm_kill_unmapped_request(clone, r);
1620 static struct request *dm_start_request(struct mapped_device *md, struct request *orig)
1622 struct request *clone;
1624 blk_start_request(orig);
1625 clone = orig->special;
1626 atomic_inc(&md->pending[rq_data_dir(clone)]);
1629 * Hold the md reference here for the in-flight I/O.
1630 * We can't rely on the reference count by device opener,
1631 * because the device may be closed during the request completion
1632 * when all bios are completed.
1633 * See the comment in rq_completed() too.
1641 * q->request_fn for request-based dm.
1642 * Called with the queue lock held.
1644 static void dm_request_fn(struct request_queue *q)
1646 struct mapped_device *md = q->queuedata;
1647 struct dm_table *map = dm_get_live_table(md);
1648 struct dm_target *ti;
1649 struct request *rq, *clone;
1653 * For suspend, check blk_queue_stopped() and increment
1654 * ->pending within a single queue_lock not to increment the
1655 * number of in-flight I/Os after the queue is stopped in
1658 while (!blk_queue_stopped(q)) {
1659 rq = blk_peek_request(q);
1663 /* always use block 0 to find the target for flushes for now */
1665 if (!(rq->cmd_flags & REQ_FLUSH))
1666 pos = blk_rq_pos(rq);
1668 ti = dm_table_find_target(map, pos);
1669 if (!dm_target_is_valid(ti)) {
1671 * Must perform setup, that dm_done() requires,
1672 * before calling dm_kill_unmapped_request
1674 DMERR_LIMIT("request attempted access beyond the end of device");
1675 clone = dm_start_request(md, rq);
1676 dm_kill_unmapped_request(clone, -EIO);
1680 if (ti->type->busy && ti->type->busy(ti))
1683 clone = dm_start_request(md, rq);
1685 spin_unlock(q->queue_lock);
1686 if (map_request(ti, clone, md))
1689 BUG_ON(!irqs_disabled());
1690 spin_lock(q->queue_lock);
1696 BUG_ON(!irqs_disabled());
1697 spin_lock(q->queue_lock);
1700 blk_delay_queue(q, HZ / 10);
1705 int dm_underlying_device_busy(struct request_queue *q)
1707 return blk_lld_busy(q);
1709 EXPORT_SYMBOL_GPL(dm_underlying_device_busy);
1711 static int dm_lld_busy(struct request_queue *q)
1714 struct mapped_device *md = q->queuedata;
1715 struct dm_table *map = dm_get_live_table(md);
1717 if (!map || test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))
1720 r = dm_table_any_busy_target(map);
1727 static int dm_any_congested(void *congested_data, int bdi_bits)
1730 struct mapped_device *md = congested_data;
1731 struct dm_table *map;
1733 if (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
1734 map = dm_get_live_table(md);
1737 * Request-based dm cares about only own queue for
1738 * the query about congestion status of request_queue
1740 if (dm_request_based(md))
1741 r = md->queue->backing_dev_info.state &
1744 r = dm_table_any_congested(map, bdi_bits);
1753 /*-----------------------------------------------------------------
1754 * An IDR is used to keep track of allocated minor numbers.
1755 *---------------------------------------------------------------*/
1756 static void free_minor(int minor)
1758 spin_lock(&_minor_lock);
1759 idr_remove(&_minor_idr, minor);
1760 spin_unlock(&_minor_lock);
1764 * See if the device with a specific minor # is free.
1766 static int specific_minor(int minor)
1770 if (minor >= (1 << MINORBITS))
1773 idr_preload(GFP_KERNEL);
1774 spin_lock(&_minor_lock);
1776 r = idr_alloc(&_minor_idr, MINOR_ALLOCED, minor, minor + 1, GFP_NOWAIT);
1778 spin_unlock(&_minor_lock);
1781 return r == -ENOSPC ? -EBUSY : r;
1785 static int next_free_minor(int *minor)
1789 idr_preload(GFP_KERNEL);
1790 spin_lock(&_minor_lock);
1792 r = idr_alloc(&_minor_idr, MINOR_ALLOCED, 0, 1 << MINORBITS, GFP_NOWAIT);
1794 spin_unlock(&_minor_lock);
1802 static const struct block_device_operations dm_blk_dops;
1804 static void dm_wq_work(struct work_struct *work);
1806 static void dm_init_md_queue(struct mapped_device *md)
1809 * Request-based dm devices cannot be stacked on top of bio-based dm
1810 * devices. The type of this dm device has not been decided yet.
1811 * The type is decided at the first table loading time.
1812 * To prevent problematic device stacking, clear the queue flag
1813 * for request stacking support until then.
1815 * This queue is new, so no concurrency on the queue_flags.
1817 queue_flag_clear_unlocked(QUEUE_FLAG_STACKABLE, md->queue);
1819 md->queue->queuedata = md;
1820 md->queue->backing_dev_info.congested_fn = dm_any_congested;
1821 md->queue->backing_dev_info.congested_data = md;
1822 blk_queue_make_request(md->queue, dm_request);
1823 blk_queue_bounce_limit(md->queue, BLK_BOUNCE_ANY);
1824 blk_queue_merge_bvec(md->queue, dm_merge_bvec);
1828 * Allocate and initialise a blank device with a given minor.
1830 static struct mapped_device *alloc_dev(int minor)
1833 struct mapped_device *md = kzalloc(sizeof(*md), GFP_KERNEL);
1837 DMWARN("unable to allocate device, out of memory.");
1841 if (!try_module_get(THIS_MODULE))
1842 goto bad_module_get;
1844 /* get a minor number for the dev */
1845 if (minor == DM_ANY_MINOR)
1846 r = next_free_minor(&minor);
1848 r = specific_minor(minor);
1852 md->type = DM_TYPE_NONE;
1853 init_rwsem(&md->io_lock);
1854 mutex_init(&md->suspend_lock);
1855 mutex_init(&md->type_lock);
1856 spin_lock_init(&md->deferred_lock);
1857 rwlock_init(&md->map_lock);
1858 atomic_set(&md->holders, 1);
1859 atomic_set(&md->open_count, 0);
1860 atomic_set(&md->event_nr, 0);
1861 atomic_set(&md->uevent_seq, 0);
1862 INIT_LIST_HEAD(&md->uevent_list);
1863 spin_lock_init(&md->uevent_lock);
1865 md->queue = blk_alloc_queue(GFP_KERNEL);
1869 dm_init_md_queue(md);
1871 md->disk = alloc_disk(1);
1875 atomic_set(&md->pending[0], 0);
1876 atomic_set(&md->pending[1], 0);
1877 init_waitqueue_head(&md->wait);
1878 INIT_WORK(&md->work, dm_wq_work);
1879 init_waitqueue_head(&md->eventq);
1881 md->disk->major = _major;
1882 md->disk->first_minor = minor;
1883 md->disk->fops = &dm_blk_dops;
1884 md->disk->queue = md->queue;
1885 md->disk->private_data = md;
1886 sprintf(md->disk->disk_name, "dm-%d", minor);
1888 format_dev_t(md->name, MKDEV(_major, minor));
1890 md->wq = alloc_workqueue("kdmflush",
1891 WQ_NON_REENTRANT | WQ_MEM_RECLAIM, 0);
1895 md->bdev = bdget_disk(md->disk, 0);
1899 bio_init(&md->flush_bio);
1900 md->flush_bio.bi_bdev = md->bdev;
1901 md->flush_bio.bi_rw = WRITE_FLUSH;
1903 /* Populate the mapping, nobody knows we exist yet */
1904 spin_lock(&_minor_lock);
1905 old_md = idr_replace(&_minor_idr, md, minor);
1906 spin_unlock(&_minor_lock);
1908 BUG_ON(old_md != MINOR_ALLOCED);
1913 destroy_workqueue(md->wq);
1915 del_gendisk(md->disk);
1918 blk_cleanup_queue(md->queue);
1922 module_put(THIS_MODULE);
1928 static void unlock_fs(struct mapped_device *md);
1930 static void free_dev(struct mapped_device *md)
1932 int minor = MINOR(disk_devt(md->disk));
1936 destroy_workqueue(md->wq);
1938 mempool_destroy(md->tio_pool);
1940 mempool_destroy(md->io_pool);
1942 bioset_free(md->bs);
1943 blk_integrity_unregister(md->disk);
1944 del_gendisk(md->disk);
1947 spin_lock(&_minor_lock);
1948 md->disk->private_data = NULL;
1949 spin_unlock(&_minor_lock);
1952 blk_cleanup_queue(md->queue);
1953 module_put(THIS_MODULE);
1957 static void __bind_mempools(struct mapped_device *md, struct dm_table *t)
1959 struct dm_md_mempools *p = dm_table_get_md_mempools(t);
1961 if (md->io_pool && md->bs) {
1962 /* The md already has necessary mempools. */
1963 if (dm_table_get_type(t) == DM_TYPE_BIO_BASED) {
1965 * Reload bioset because front_pad may have changed
1966 * because a different table was loaded.
1968 bioset_free(md->bs);
1971 } else if (dm_table_get_type(t) == DM_TYPE_REQUEST_BASED) {
1972 BUG_ON(!md->tio_pool);
1974 * There's no need to reload with request-based dm
1975 * because the size of front_pad doesn't change.
1976 * Note for future: If you are to reload bioset,
1977 * prep-ed requests in the queue may refer
1978 * to bio from the old bioset, so you must walk
1979 * through the queue to unprep.
1985 BUG_ON(!p || md->io_pool || md->tio_pool || md->bs);
1987 md->io_pool = p->io_pool;
1989 md->tio_pool = p->tio_pool;
1995 /* mempool bind completed, now no need any mempools in the table */
1996 dm_table_free_md_mempools(t);
2000 * Bind a table to the device.
2002 static void event_callback(void *context)
2004 unsigned long flags;
2006 struct mapped_device *md = (struct mapped_device *) context;
2008 spin_lock_irqsave(&md->uevent_lock, flags);
2009 list_splice_init(&md->uevent_list, &uevents);
2010 spin_unlock_irqrestore(&md->uevent_lock, flags);
2012 dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj);
2014 atomic_inc(&md->event_nr);
2015 wake_up(&md->eventq);
2019 * Protected by md->suspend_lock obtained by dm_swap_table().
2021 static void __set_size(struct mapped_device *md, sector_t size)
2023 set_capacity(md->disk, size);
2025 i_size_write(md->bdev->bd_inode, (loff_t)size << SECTOR_SHIFT);
2029 * Return 1 if the queue has a compulsory merge_bvec_fn function.
2031 * If this function returns 0, then the device is either a non-dm
2032 * device without a merge_bvec_fn, or it is a dm device that is
2033 * able to split any bios it receives that are too big.
2035 int dm_queue_merge_is_compulsory(struct request_queue *q)
2037 struct mapped_device *dev_md;
2039 if (!q->merge_bvec_fn)
2042 if (q->make_request_fn == dm_request) {
2043 dev_md = q->queuedata;
2044 if (test_bit(DMF_MERGE_IS_OPTIONAL, &dev_md->flags))
2051 static int dm_device_merge_is_compulsory(struct dm_target *ti,
2052 struct dm_dev *dev, sector_t start,
2053 sector_t len, void *data)
2055 struct block_device *bdev = dev->bdev;
2056 struct request_queue *q = bdev_get_queue(bdev);
2058 return dm_queue_merge_is_compulsory(q);
2062 * Return 1 if it is acceptable to ignore merge_bvec_fn based
2063 * on the properties of the underlying devices.
2065 static int dm_table_merge_is_optional(struct dm_table *table)
2068 struct dm_target *ti;
2070 while (i < dm_table_get_num_targets(table)) {
2071 ti = dm_table_get_target(table, i++);
2073 if (ti->type->iterate_devices &&
2074 ti->type->iterate_devices(ti, dm_device_merge_is_compulsory, NULL))
2082 * Returns old map, which caller must destroy.
2084 static struct dm_table *__bind(struct mapped_device *md, struct dm_table *t,
2085 struct queue_limits *limits)
2087 struct dm_table *old_map;
2088 struct request_queue *q = md->queue;
2090 unsigned long flags;
2091 int merge_is_optional;
2093 size = dm_table_get_size(t);
2096 * Wipe any geometry if the size of the table changed.
2098 if (size != get_capacity(md->disk))
2099 memset(&md->geometry, 0, sizeof(md->geometry));
2101 __set_size(md, size);
2103 dm_table_event_callback(t, event_callback, md);
2106 * The queue hasn't been stopped yet, if the old table type wasn't
2107 * for request-based during suspension. So stop it to prevent
2108 * I/O mapping before resume.
2109 * This must be done before setting the queue restrictions,
2110 * because request-based dm may be run just after the setting.
2112 if (dm_table_request_based(t) && !blk_queue_stopped(q))
2115 __bind_mempools(md, t);
2117 merge_is_optional = dm_table_merge_is_optional(t);
2119 write_lock_irqsave(&md->map_lock, flags);
2122 md->immutable_target_type = dm_table_get_immutable_target_type(t);
2124 dm_table_set_restrictions(t, q, limits);
2125 if (merge_is_optional)
2126 set_bit(DMF_MERGE_IS_OPTIONAL, &md->flags);
2128 clear_bit(DMF_MERGE_IS_OPTIONAL, &md->flags);
2129 write_unlock_irqrestore(&md->map_lock, flags);
2135 * Returns unbound table for the caller to free.
2137 static struct dm_table *__unbind(struct mapped_device *md)
2139 struct dm_table *map = md->map;
2140 unsigned long flags;
2145 dm_table_event_callback(map, NULL, NULL);
2146 write_lock_irqsave(&md->map_lock, flags);
2148 write_unlock_irqrestore(&md->map_lock, flags);
2154 * Constructor for a new device.
2156 int dm_create(int minor, struct mapped_device **result)
2158 struct mapped_device *md;
2160 md = alloc_dev(minor);
2171 * Functions to manage md->type.
2172 * All are required to hold md->type_lock.
2174 void dm_lock_md_type(struct mapped_device *md)
2176 mutex_lock(&md->type_lock);
2179 void dm_unlock_md_type(struct mapped_device *md)
2181 mutex_unlock(&md->type_lock);
2184 void dm_set_md_type(struct mapped_device *md, unsigned type)
2189 unsigned dm_get_md_type(struct mapped_device *md)
2194 struct target_type *dm_get_immutable_target_type(struct mapped_device *md)
2196 return md->immutable_target_type;
2200 * Fully initialize a request-based queue (->elevator, ->request_fn, etc).
2202 static int dm_init_request_based_queue(struct mapped_device *md)
2204 struct request_queue *q = NULL;
2206 if (md->queue->elevator)
2209 /* Fully initialize the queue */
2210 q = blk_init_allocated_queue(md->queue, dm_request_fn, NULL);
2215 dm_init_md_queue(md);
2216 blk_queue_softirq_done(md->queue, dm_softirq_done);
2217 blk_queue_prep_rq(md->queue, dm_prep_fn);
2218 blk_queue_lld_busy(md->queue, dm_lld_busy);
2220 elv_register_queue(md->queue);
2226 * Setup the DM device's queue based on md's type
2228 int dm_setup_md_queue(struct mapped_device *md)
2230 if ((dm_get_md_type(md) == DM_TYPE_REQUEST_BASED) &&
2231 !dm_init_request_based_queue(md)) {
2232 DMWARN("Cannot initialize queue for request-based mapped device");
2239 static struct mapped_device *dm_find_md(dev_t dev)
2241 struct mapped_device *md;
2242 unsigned minor = MINOR(dev);
2244 if (MAJOR(dev) != _major || minor >= (1 << MINORBITS))
2247 spin_lock(&_minor_lock);
2249 md = idr_find(&_minor_idr, minor);
2250 if (md && (md == MINOR_ALLOCED ||
2251 (MINOR(disk_devt(dm_disk(md))) != minor) ||
2252 dm_deleting_md(md) ||
2253 test_bit(DMF_FREEING, &md->flags))) {
2259 spin_unlock(&_minor_lock);
2264 struct mapped_device *dm_get_md(dev_t dev)
2266 struct mapped_device *md = dm_find_md(dev);
2273 EXPORT_SYMBOL_GPL(dm_get_md);
2275 void *dm_get_mdptr(struct mapped_device *md)
2277 return md->interface_ptr;
2280 void dm_set_mdptr(struct mapped_device *md, void *ptr)
2282 md->interface_ptr = ptr;
2285 void dm_get(struct mapped_device *md)
2287 atomic_inc(&md->holders);
2288 BUG_ON(test_bit(DMF_FREEING, &md->flags));
2291 const char *dm_device_name(struct mapped_device *md)
2295 EXPORT_SYMBOL_GPL(dm_device_name);
2297 static void __dm_destroy(struct mapped_device *md, bool wait)
2299 struct dm_table *map;
2303 spin_lock(&_minor_lock);
2304 map = dm_get_live_table(md);
2305 idr_replace(&_minor_idr, MINOR_ALLOCED, MINOR(disk_devt(dm_disk(md))));
2306 set_bit(DMF_FREEING, &md->flags);
2307 spin_unlock(&_minor_lock);
2309 if (!dm_suspended_md(md)) {
2310 dm_table_presuspend_targets(map);
2311 dm_table_postsuspend_targets(map);
2315 * Rare, but there may be I/O requests still going to complete,
2316 * for example. Wait for all references to disappear.
2317 * No one should increment the reference count of the mapped_device,
2318 * after the mapped_device state becomes DMF_FREEING.
2321 while (atomic_read(&md->holders))
2323 else if (atomic_read(&md->holders))
2324 DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)",
2325 dm_device_name(md), atomic_read(&md->holders));
2329 dm_table_destroy(__unbind(md));
2333 void dm_destroy(struct mapped_device *md)
2335 __dm_destroy(md, true);
2338 void dm_destroy_immediate(struct mapped_device *md)
2340 __dm_destroy(md, false);
2343 void dm_put(struct mapped_device *md)
2345 atomic_dec(&md->holders);
2347 EXPORT_SYMBOL_GPL(dm_put);
2349 static int dm_wait_for_completion(struct mapped_device *md, int interruptible)
2352 DECLARE_WAITQUEUE(wait, current);
2354 add_wait_queue(&md->wait, &wait);
2357 set_current_state(interruptible);
2359 if (!md_in_flight(md))
2362 if (interruptible == TASK_INTERRUPTIBLE &&
2363 signal_pending(current)) {
2370 set_current_state(TASK_RUNNING);
2372 remove_wait_queue(&md->wait, &wait);
2378 * Process the deferred bios
2380 static void dm_wq_work(struct work_struct *work)
2382 struct mapped_device *md = container_of(work, struct mapped_device,
2386 down_read(&md->io_lock);
2388 while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
2389 spin_lock_irq(&md->deferred_lock);
2390 c = bio_list_pop(&md->deferred);
2391 spin_unlock_irq(&md->deferred_lock);
2396 up_read(&md->io_lock);
2398 if (dm_request_based(md))
2399 generic_make_request(c);
2401 __split_and_process_bio(md, c);
2403 down_read(&md->io_lock);
2406 up_read(&md->io_lock);
2409 static void dm_queue_flush(struct mapped_device *md)
2411 clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2412 smp_mb__after_clear_bit();
2413 queue_work(md->wq, &md->work);
2417 * Swap in a new table, returning the old one for the caller to destroy.
2419 struct dm_table *dm_swap_table(struct mapped_device *md, struct dm_table *table)
2421 struct dm_table *live_map, *map = ERR_PTR(-EINVAL);
2422 struct queue_limits limits;
2425 mutex_lock(&md->suspend_lock);
2427 /* device must be suspended */
2428 if (!dm_suspended_md(md))
2432 * If the new table has no data devices, retain the existing limits.
2433 * This helps multipath with queue_if_no_path if all paths disappear,
2434 * then new I/O is queued based on these limits, and then some paths
2437 if (dm_table_has_no_data_devices(table)) {
2438 live_map = dm_get_live_table(md);
2440 limits = md->queue->limits;
2441 dm_table_put(live_map);
2444 r = dm_calculate_queue_limits(table, &limits);
2450 map = __bind(md, table, &limits);
2453 mutex_unlock(&md->suspend_lock);
2458 * Functions to lock and unlock any filesystem running on the
2461 static int lock_fs(struct mapped_device *md)
2465 WARN_ON(md->frozen_sb);
2467 md->frozen_sb = freeze_bdev(md->bdev);
2468 if (IS_ERR(md->frozen_sb)) {
2469 r = PTR_ERR(md->frozen_sb);
2470 md->frozen_sb = NULL;
2474 set_bit(DMF_FROZEN, &md->flags);
2479 static void unlock_fs(struct mapped_device *md)
2481 if (!test_bit(DMF_FROZEN, &md->flags))
2484 thaw_bdev(md->bdev, md->frozen_sb);
2485 md->frozen_sb = NULL;
2486 clear_bit(DMF_FROZEN, &md->flags);
2490 * We need to be able to change a mapping table under a mounted
2491 * filesystem. For example we might want to move some data in
2492 * the background. Before the table can be swapped with
2493 * dm_bind_table, dm_suspend must be called to flush any in
2494 * flight bios and ensure that any further io gets deferred.
2497 * Suspend mechanism in request-based dm.
2499 * 1. Flush all I/Os by lock_fs() if needed.
2500 * 2. Stop dispatching any I/O by stopping the request_queue.
2501 * 3. Wait for all in-flight I/Os to be completed or requeued.
2503 * To abort suspend, start the request_queue.
2505 int dm_suspend(struct mapped_device *md, unsigned suspend_flags)
2507 struct dm_table *map = NULL;
2509 int do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG ? 1 : 0;
2510 int noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG ? 1 : 0;
2512 mutex_lock(&md->suspend_lock);
2514 if (dm_suspended_md(md)) {
2519 map = dm_get_live_table(md);
2522 * DMF_NOFLUSH_SUSPENDING must be set before presuspend.
2523 * This flag is cleared before dm_suspend returns.
2526 set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2528 /* This does not get reverted if there's an error later. */
2529 dm_table_presuspend_targets(map);
2532 * Flush I/O to the device.
2533 * Any I/O submitted after lock_fs() may not be flushed.
2534 * noflush takes precedence over do_lockfs.
2535 * (lock_fs() flushes I/Os and waits for them to complete.)
2537 if (!noflush && do_lockfs) {
2544 * Here we must make sure that no processes are submitting requests
2545 * to target drivers i.e. no one may be executing
2546 * __split_and_process_bio. This is called from dm_request and
2549 * To get all processes out of __split_and_process_bio in dm_request,
2550 * we take the write lock. To prevent any process from reentering
2551 * __split_and_process_bio from dm_request and quiesce the thread
2552 * (dm_wq_work), we set BMF_BLOCK_IO_FOR_SUSPEND and call
2553 * flush_workqueue(md->wq).
2555 down_write(&md->io_lock);
2556 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2557 up_write(&md->io_lock);
2560 * Stop md->queue before flushing md->wq in case request-based
2561 * dm defers requests to md->wq from md->queue.
2563 if (dm_request_based(md))
2564 stop_queue(md->queue);
2566 flush_workqueue(md->wq);
2569 * At this point no more requests are entering target request routines.
2570 * We call dm_wait_for_completion to wait for all existing requests
2573 r = dm_wait_for_completion(md, TASK_INTERRUPTIBLE);
2575 down_write(&md->io_lock);
2577 clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2578 up_write(&md->io_lock);
2580 /* were we interrupted ? */
2584 if (dm_request_based(md))
2585 start_queue(md->queue);
2588 goto out; /* pushback list is already flushed, so skip flush */
2592 * If dm_wait_for_completion returned 0, the device is completely
2593 * quiescent now. There is no request-processing activity. All new
2594 * requests are being added to md->deferred list.
2597 set_bit(DMF_SUSPENDED, &md->flags);
2599 dm_table_postsuspend_targets(map);
2605 mutex_unlock(&md->suspend_lock);
2609 int dm_resume(struct mapped_device *md)
2612 struct dm_table *map = NULL;
2614 mutex_lock(&md->suspend_lock);
2615 if (!dm_suspended_md(md))
2618 map = dm_get_live_table(md);
2619 if (!map || !dm_table_get_size(map))
2622 r = dm_table_resume_targets(map);
2629 * Flushing deferred I/Os must be done after targets are resumed
2630 * so that mapping of targets can work correctly.
2631 * Request-based dm is queueing the deferred I/Os in its request_queue.
2633 if (dm_request_based(md))
2634 start_queue(md->queue);
2638 clear_bit(DMF_SUSPENDED, &md->flags);
2643 mutex_unlock(&md->suspend_lock);
2648 /*-----------------------------------------------------------------
2649 * Event notification.
2650 *---------------------------------------------------------------*/
2651 int dm_kobject_uevent(struct mapped_device *md, enum kobject_action action,
2654 char udev_cookie[DM_COOKIE_LENGTH];
2655 char *envp[] = { udev_cookie, NULL };
2658 return kobject_uevent(&disk_to_dev(md->disk)->kobj, action);
2660 snprintf(udev_cookie, DM_COOKIE_LENGTH, "%s=%u",
2661 DM_COOKIE_ENV_VAR_NAME, cookie);
2662 return kobject_uevent_env(&disk_to_dev(md->disk)->kobj,
2667 uint32_t dm_next_uevent_seq(struct mapped_device *md)
2669 return atomic_add_return(1, &md->uevent_seq);
2672 uint32_t dm_get_event_nr(struct mapped_device *md)
2674 return atomic_read(&md->event_nr);
2677 int dm_wait_event(struct mapped_device *md, int event_nr)
2679 return wait_event_interruptible(md->eventq,
2680 (event_nr != atomic_read(&md->event_nr)));
2683 void dm_uevent_add(struct mapped_device *md, struct list_head *elist)
2685 unsigned long flags;
2687 spin_lock_irqsave(&md->uevent_lock, flags);
2688 list_add(elist, &md->uevent_list);
2689 spin_unlock_irqrestore(&md->uevent_lock, flags);
2693 * The gendisk is only valid as long as you have a reference
2696 struct gendisk *dm_disk(struct mapped_device *md)
2701 struct kobject *dm_kobject(struct mapped_device *md)
2707 * struct mapped_device should not be exported outside of dm.c
2708 * so use this check to verify that kobj is part of md structure
2710 struct mapped_device *dm_get_from_kobject(struct kobject *kobj)
2712 struct mapped_device *md;
2714 md = container_of(kobj, struct mapped_device, kobj);
2715 if (&md->kobj != kobj)
2718 if (test_bit(DMF_FREEING, &md->flags) ||
2726 int dm_suspended_md(struct mapped_device *md)
2728 return test_bit(DMF_SUSPENDED, &md->flags);
2731 int dm_suspended(struct dm_target *ti)
2733 return dm_suspended_md(dm_table_get_md(ti->table));
2735 EXPORT_SYMBOL_GPL(dm_suspended);
2737 int dm_noflush_suspending(struct dm_target *ti)
2739 return __noflush_suspending(dm_table_get_md(ti->table));
2741 EXPORT_SYMBOL_GPL(dm_noflush_suspending);
2743 struct dm_md_mempools *dm_alloc_md_mempools(unsigned type, unsigned integrity, unsigned per_bio_data_size)
2745 struct dm_md_mempools *pools = kmalloc(sizeof(*pools), GFP_KERNEL);
2746 unsigned int pool_size = (type == DM_TYPE_BIO_BASED) ? 16 : MIN_IOS;
2751 per_bio_data_size = roundup(per_bio_data_size, __alignof__(struct dm_target_io));
2753 pools->io_pool = (type == DM_TYPE_BIO_BASED) ?
2754 mempool_create_slab_pool(MIN_IOS, _io_cache) :
2755 mempool_create_slab_pool(MIN_IOS, _rq_bio_info_cache);
2756 if (!pools->io_pool)
2757 goto free_pools_and_out;
2759 pools->tio_pool = NULL;
2760 if (type == DM_TYPE_REQUEST_BASED) {
2761 pools->tio_pool = mempool_create_slab_pool(MIN_IOS, _rq_tio_cache);
2762 if (!pools->tio_pool)
2763 goto free_io_pool_and_out;
2766 pools->bs = (type == DM_TYPE_BIO_BASED) ?
2767 bioset_create(pool_size,
2768 per_bio_data_size + offsetof(struct dm_target_io, clone)) :
2769 bioset_create(pool_size,
2770 offsetof(struct dm_rq_clone_bio_info, clone));
2772 goto free_tio_pool_and_out;
2774 if (integrity && bioset_integrity_create(pools->bs, pool_size))
2775 goto free_bioset_and_out;
2779 free_bioset_and_out:
2780 bioset_free(pools->bs);
2782 free_tio_pool_and_out:
2783 if (pools->tio_pool)
2784 mempool_destroy(pools->tio_pool);
2786 free_io_pool_and_out:
2787 mempool_destroy(pools->io_pool);
2795 void dm_free_md_mempools(struct dm_md_mempools *pools)
2801 mempool_destroy(pools->io_pool);
2803 if (pools->tio_pool)
2804 mempool_destroy(pools->tio_pool);
2807 bioset_free(pools->bs);
2812 static const struct block_device_operations dm_blk_dops = {
2813 .open = dm_blk_open,
2814 .release = dm_blk_close,
2815 .ioctl = dm_blk_ioctl,
2816 .getgeo = dm_blk_getgeo,
2817 .owner = THIS_MODULE
2820 EXPORT_SYMBOL(dm_get_mapinfo);
2825 module_init(dm_init);
2826 module_exit(dm_exit);
2828 module_param(major, uint, 0);
2829 MODULE_PARM_DESC(major, "The major number of the device mapper");
2830 MODULE_DESCRIPTION(DM_NAME " driver");
2831 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
2832 MODULE_LICENSE("GPL");