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;
67 * One of these is allocated per target within a bio. Hopefully
68 * 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 - the bio clones we allocate are embedded in these
93 * We allocate these with bio_alloc_bioset, using the front_pad parameter when
94 * the bioset is created - this means the bio has to come at the end of the
97 struct dm_rq_clone_bio_info {
99 struct dm_rq_target_io *tio;
103 union map_info *dm_get_mapinfo(struct bio *bio)
105 if (bio && bio->bi_private)
106 return &((struct dm_target_io *)bio->bi_private)->info;
110 union map_info *dm_get_rq_mapinfo(struct request *rq)
112 if (rq && rq->end_io_data)
113 return &((struct dm_rq_target_io *)rq->end_io_data)->info;
116 EXPORT_SYMBOL_GPL(dm_get_rq_mapinfo);
118 #define MINOR_ALLOCED ((void *)-1)
121 * Bits for the md->flags field.
123 #define DMF_BLOCK_IO_FOR_SUSPEND 0
124 #define DMF_SUSPENDED 1
126 #define DMF_FREEING 3
127 #define DMF_DELETING 4
128 #define DMF_NOFLUSH_SUSPENDING 5
129 #define DMF_MERGE_IS_OPTIONAL 6
132 * Work processed by per-device workqueue.
134 struct mapped_device {
135 struct rw_semaphore io_lock;
136 struct mutex suspend_lock;
143 struct request_queue *queue;
145 /* Protect queue and type against concurrent access. */
146 struct mutex type_lock;
148 struct target_type *immutable_target_type;
150 struct gendisk *disk;
156 * A list of ios that arrived while we were suspended.
159 wait_queue_head_t wait;
160 struct work_struct work;
161 struct bio_list deferred;
162 spinlock_t deferred_lock;
165 * Processing queue (flush)
167 struct workqueue_struct *wq;
170 * The current mapping.
172 struct dm_table *map;
175 * io objects are allocated from here.
186 wait_queue_head_t eventq;
188 struct list_head uevent_list;
189 spinlock_t uevent_lock; /* Protect access to uevent_list */
192 * freeze/thaw support require holding onto a super block
194 struct super_block *frozen_sb;
195 struct block_device *bdev;
197 /* forced geometry settings */
198 struct hd_geometry geometry;
203 /* zero-length flush that will be cloned and submitted to targets */
204 struct bio flush_bio;
208 * For mempools pre-allocation at the table loading time.
210 struct dm_md_mempools {
217 static struct kmem_cache *_io_cache;
218 static struct kmem_cache *_rq_tio_cache;
221 * Unused now, and needs to be deleted. But since io_pool is overloaded and it's
222 * still used for _io_cache, I'm leaving this for a later cleanup
224 static struct kmem_cache *_rq_bio_info_cache;
226 static int __init local_init(void)
230 /* allocate a slab for the dm_ios */
231 _io_cache = KMEM_CACHE(dm_io, 0);
235 _rq_tio_cache = KMEM_CACHE(dm_rq_target_io, 0);
237 goto out_free_io_cache;
239 _rq_bio_info_cache = KMEM_CACHE(dm_rq_clone_bio_info, 0);
240 if (!_rq_bio_info_cache)
241 goto out_free_rq_tio_cache;
243 r = dm_uevent_init();
245 goto out_free_rq_bio_info_cache;
248 r = register_blkdev(_major, _name);
250 goto out_uevent_exit;
259 out_free_rq_bio_info_cache:
260 kmem_cache_destroy(_rq_bio_info_cache);
261 out_free_rq_tio_cache:
262 kmem_cache_destroy(_rq_tio_cache);
264 kmem_cache_destroy(_io_cache);
269 static void local_exit(void)
271 kmem_cache_destroy(_rq_bio_info_cache);
272 kmem_cache_destroy(_rq_tio_cache);
273 kmem_cache_destroy(_io_cache);
274 unregister_blkdev(_major, _name);
279 DMINFO("cleaned up");
282 static int (*_inits[])(void) __initdata = {
292 static void (*_exits[])(void) = {
302 static int __init dm_init(void)
304 const int count = ARRAY_SIZE(_inits);
308 for (i = 0; i < count; i++) {
323 static void __exit dm_exit(void)
325 int i = ARRAY_SIZE(_exits);
331 * Should be empty by this point.
333 idr_remove_all(&_minor_idr);
334 idr_destroy(&_minor_idr);
338 * Block device functions
340 int dm_deleting_md(struct mapped_device *md)
342 return test_bit(DMF_DELETING, &md->flags);
345 static int dm_blk_open(struct block_device *bdev, fmode_t mode)
347 struct mapped_device *md;
349 spin_lock(&_minor_lock);
351 md = bdev->bd_disk->private_data;
355 if (test_bit(DMF_FREEING, &md->flags) ||
356 dm_deleting_md(md)) {
362 atomic_inc(&md->open_count);
365 spin_unlock(&_minor_lock);
367 return md ? 0 : -ENXIO;
370 static int dm_blk_close(struct gendisk *disk, fmode_t mode)
372 struct mapped_device *md = disk->private_data;
374 spin_lock(&_minor_lock);
376 atomic_dec(&md->open_count);
379 spin_unlock(&_minor_lock);
384 int dm_open_count(struct mapped_device *md)
386 return atomic_read(&md->open_count);
390 * Guarantees nothing is using the device before it's deleted.
392 int dm_lock_for_deletion(struct mapped_device *md)
396 spin_lock(&_minor_lock);
398 if (dm_open_count(md))
401 set_bit(DMF_DELETING, &md->flags);
403 spin_unlock(&_minor_lock);
408 static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo)
410 struct mapped_device *md = bdev->bd_disk->private_data;
412 return dm_get_geometry(md, geo);
415 static int dm_blk_ioctl(struct block_device *bdev, fmode_t mode,
416 unsigned int cmd, unsigned long arg)
418 struct mapped_device *md = bdev->bd_disk->private_data;
419 struct dm_table *map = dm_get_live_table(md);
420 struct dm_target *tgt;
423 if (!map || !dm_table_get_size(map))
426 /* We only support devices that have a single target */
427 if (dm_table_get_num_targets(map) != 1)
430 tgt = dm_table_get_target(map, 0);
432 if (dm_suspended_md(md)) {
437 if (tgt->type->ioctl)
438 r = tgt->type->ioctl(tgt, cmd, arg);
446 static struct dm_io *alloc_io(struct mapped_device *md)
448 return mempool_alloc(md->io_pool, GFP_NOIO);
451 static void free_io(struct mapped_device *md, struct dm_io *io)
453 mempool_free(io, md->io_pool);
456 static void free_tio(struct mapped_device *md, struct dm_target_io *tio)
458 bio_put(&tio->clone);
461 static struct dm_rq_target_io *alloc_rq_tio(struct mapped_device *md,
464 return mempool_alloc(md->tio_pool, gfp_mask);
467 static void free_rq_tio(struct dm_rq_target_io *tio)
469 mempool_free(tio, tio->md->tio_pool);
472 static int md_in_flight(struct mapped_device *md)
474 return atomic_read(&md->pending[READ]) +
475 atomic_read(&md->pending[WRITE]);
478 static void start_io_acct(struct dm_io *io)
480 struct mapped_device *md = io->md;
482 int rw = bio_data_dir(io->bio);
484 io->start_time = jiffies;
486 cpu = part_stat_lock();
487 part_round_stats(cpu, &dm_disk(md)->part0);
489 atomic_set(&dm_disk(md)->part0.in_flight[rw],
490 atomic_inc_return(&md->pending[rw]));
493 static void end_io_acct(struct dm_io *io)
495 struct mapped_device *md = io->md;
496 struct bio *bio = io->bio;
497 unsigned long duration = jiffies - io->start_time;
499 int rw = bio_data_dir(bio);
501 cpu = part_stat_lock();
502 part_round_stats(cpu, &dm_disk(md)->part0);
503 part_stat_add(cpu, &dm_disk(md)->part0, ticks[rw], duration);
507 * After this is decremented the bio must not be touched if it is
510 pending = atomic_dec_return(&md->pending[rw]);
511 atomic_set(&dm_disk(md)->part0.in_flight[rw], pending);
512 pending += atomic_read(&md->pending[rw^0x1]);
514 /* nudge anyone waiting on suspend queue */
520 * Add the bio to the list of deferred io.
522 static void queue_io(struct mapped_device *md, struct bio *bio)
526 spin_lock_irqsave(&md->deferred_lock, flags);
527 bio_list_add(&md->deferred, bio);
528 spin_unlock_irqrestore(&md->deferred_lock, flags);
529 queue_work(md->wq, &md->work);
533 * Everyone (including functions in this file), should use this
534 * function to access the md->map field, and make sure they call
535 * dm_table_put() when finished.
537 struct dm_table *dm_get_live_table(struct mapped_device *md)
542 read_lock_irqsave(&md->map_lock, flags);
546 read_unlock_irqrestore(&md->map_lock, flags);
552 * Get the geometry associated with a dm device
554 int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo)
562 * Set the geometry of a device.
564 int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo)
566 sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors;
568 if (geo->start > sz) {
569 DMWARN("Start sector is beyond the geometry limits.");
578 /*-----------------------------------------------------------------
580 * A more elegant soln is in the works that uses the queue
581 * merge fn, unfortunately there are a couple of changes to
582 * the block layer that I want to make for this. So in the
583 * interests of getting something for people to use I give
584 * you this clearly demarcated crap.
585 *---------------------------------------------------------------*/
587 static int __noflush_suspending(struct mapped_device *md)
589 return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
593 * Decrements the number of outstanding ios that a bio has been
594 * cloned into, completing the original io if necc.
596 static void dec_pending(struct dm_io *io, int error)
601 struct mapped_device *md = io->md;
603 /* Push-back supersedes any I/O errors */
604 if (unlikely(error)) {
605 spin_lock_irqsave(&io->endio_lock, flags);
606 if (!(io->error > 0 && __noflush_suspending(md)))
608 spin_unlock_irqrestore(&io->endio_lock, flags);
611 if (atomic_dec_and_test(&io->io_count)) {
612 if (io->error == DM_ENDIO_REQUEUE) {
614 * Target requested pushing back the I/O.
616 spin_lock_irqsave(&md->deferred_lock, flags);
617 if (__noflush_suspending(md))
618 bio_list_add_head(&md->deferred, io->bio);
620 /* noflush suspend was interrupted. */
622 spin_unlock_irqrestore(&md->deferred_lock, flags);
625 io_error = io->error;
630 if (io_error == DM_ENDIO_REQUEUE)
633 if ((bio->bi_rw & REQ_FLUSH) && bio->bi_size) {
635 * Preflush done for flush with data, reissue
638 bio->bi_rw &= ~REQ_FLUSH;
641 /* done with normal IO or empty flush */
642 trace_block_bio_complete(md->queue, bio, io_error);
643 bio_endio(bio, io_error);
648 static void clone_endio(struct bio *bio, int error)
651 struct dm_target_io *tio = bio->bi_private;
652 struct dm_io *io = tio->io;
653 struct mapped_device *md = tio->io->md;
654 dm_endio_fn endio = tio->ti->type->end_io;
656 if (!bio_flagged(bio, BIO_UPTODATE) && !error)
660 r = endio(tio->ti, bio, error, &tio->info);
661 if (r < 0 || r == DM_ENDIO_REQUEUE)
663 * error and requeue request are handled
667 else if (r == DM_ENDIO_INCOMPLETE)
668 /* The target will handle the io */
671 DMWARN("unimplemented target endio return value: %d", r);
677 dec_pending(io, error);
681 * Partial completion handling for request-based dm
683 static void end_clone_bio(struct bio *clone, int error)
685 struct dm_rq_clone_bio_info *info = clone->bi_private;
686 struct dm_rq_target_io *tio = info->tio;
687 struct bio *bio = info->orig;
688 unsigned int nr_bytes = info->orig->bi_size;
694 * An error has already been detected on the request.
695 * Once error occurred, just let clone->end_io() handle
701 * Don't notice the error to the upper layer yet.
702 * The error handling decision is made by the target driver,
703 * when the request is completed.
710 * I/O for the bio successfully completed.
711 * Notice the data completion to the upper layer.
715 * bios are processed from the head of the list.
716 * So the completing bio should always be rq->bio.
717 * If it's not, something wrong is happening.
719 if (tio->orig->bio != bio)
720 DMERR("bio completion is going in the middle of the request");
723 * Update the original request.
724 * Do not use blk_end_request() here, because it may complete
725 * the original request before the clone, and break the ordering.
727 blk_update_request(tio->orig, 0, nr_bytes);
731 * Don't touch any member of the md after calling this function because
732 * the md may be freed in dm_put() at the end of this function.
733 * Or do dm_get() before calling this function and dm_put() later.
735 static void rq_completed(struct mapped_device *md, int rw, int run_queue)
737 atomic_dec(&md->pending[rw]);
739 /* nudge anyone waiting on suspend queue */
740 if (!md_in_flight(md))
744 blk_run_queue(md->queue);
747 * dm_put() must be at the end of this function. See the comment above
752 static void free_rq_clone(struct request *clone)
754 struct dm_rq_target_io *tio = clone->end_io_data;
756 blk_rq_unprep_clone(clone);
761 * Complete the clone and the original request.
762 * Must be called without queue lock.
764 static void dm_end_request(struct request *clone, int error)
766 int rw = rq_data_dir(clone);
767 struct dm_rq_target_io *tio = clone->end_io_data;
768 struct mapped_device *md = tio->md;
769 struct request *rq = tio->orig;
771 if (rq->cmd_type == REQ_TYPE_BLOCK_PC) {
772 rq->errors = clone->errors;
773 rq->resid_len = clone->resid_len;
777 * We are using the sense buffer of the original
779 * So setting the length of the sense data is enough.
781 rq->sense_len = clone->sense_len;
784 free_rq_clone(clone);
785 blk_end_request_all(rq, error);
786 rq_completed(md, rw, true);
789 static void dm_unprep_request(struct request *rq)
791 struct request *clone = rq->special;
794 rq->cmd_flags &= ~REQ_DONTPREP;
796 free_rq_clone(clone);
800 * Requeue the original request of a clone.
802 void dm_requeue_unmapped_request(struct request *clone)
804 int rw = rq_data_dir(clone);
805 struct dm_rq_target_io *tio = clone->end_io_data;
806 struct mapped_device *md = tio->md;
807 struct request *rq = tio->orig;
808 struct request_queue *q = rq->q;
811 dm_unprep_request(rq);
813 spin_lock_irqsave(q->queue_lock, flags);
814 blk_requeue_request(q, rq);
815 spin_unlock_irqrestore(q->queue_lock, flags);
817 rq_completed(md, rw, 0);
819 EXPORT_SYMBOL_GPL(dm_requeue_unmapped_request);
821 static void __stop_queue(struct request_queue *q)
826 static void stop_queue(struct request_queue *q)
830 spin_lock_irqsave(q->queue_lock, flags);
832 spin_unlock_irqrestore(q->queue_lock, flags);
835 static void __start_queue(struct request_queue *q)
837 if (blk_queue_stopped(q))
841 static void start_queue(struct request_queue *q)
845 spin_lock_irqsave(q->queue_lock, flags);
847 spin_unlock_irqrestore(q->queue_lock, flags);
850 static void dm_done(struct request *clone, int error, bool mapped)
853 struct dm_rq_target_io *tio = clone->end_io_data;
854 dm_request_endio_fn rq_end_io = NULL;
857 rq_end_io = tio->ti->type->rq_end_io;
859 if (mapped && rq_end_io)
860 r = rq_end_io(tio->ti, clone, error, &tio->info);
864 /* The target wants to complete the I/O */
865 dm_end_request(clone, r);
866 else if (r == DM_ENDIO_INCOMPLETE)
867 /* The target will handle the I/O */
869 else if (r == DM_ENDIO_REQUEUE)
870 /* The target wants to requeue the I/O */
871 dm_requeue_unmapped_request(clone);
873 DMWARN("unimplemented target endio return value: %d", r);
879 * Request completion handler for request-based dm
881 static void dm_softirq_done(struct request *rq)
884 struct request *clone = rq->completion_data;
885 struct dm_rq_target_io *tio = clone->end_io_data;
887 if (rq->cmd_flags & REQ_FAILED)
890 dm_done(clone, tio->error, mapped);
894 * Complete the clone and the original request with the error status
895 * through softirq context.
897 static void dm_complete_request(struct request *clone, int error)
899 struct dm_rq_target_io *tio = clone->end_io_data;
900 struct request *rq = tio->orig;
903 rq->completion_data = clone;
904 blk_complete_request(rq);
908 * Complete the not-mapped clone and the original request with the error status
909 * through softirq context.
910 * Target's rq_end_io() function isn't called.
911 * This may be used when the target's map_rq() function fails.
913 void dm_kill_unmapped_request(struct request *clone, int error)
915 struct dm_rq_target_io *tio = clone->end_io_data;
916 struct request *rq = tio->orig;
918 rq->cmd_flags |= REQ_FAILED;
919 dm_complete_request(clone, error);
921 EXPORT_SYMBOL_GPL(dm_kill_unmapped_request);
924 * Called with the queue lock held
926 static void end_clone_request(struct request *clone, int error)
929 * For just cleaning up the information of the queue in which
930 * the clone was dispatched.
931 * The clone is *NOT* freed actually here because it is alloced from
932 * dm own mempool and REQ_ALLOCED isn't set in clone->cmd_flags.
934 __blk_put_request(clone->q, clone);
937 * Actual request completion is done in a softirq context which doesn't
938 * hold the queue lock. Otherwise, deadlock could occur because:
939 * - another request may be submitted by the upper level driver
940 * of the stacking during the completion
941 * - the submission which requires queue lock may be done
944 dm_complete_request(clone, error);
948 * Return maximum size of I/O possible at the supplied sector up to the current
951 static sector_t max_io_len_target_boundary(sector_t sector, struct dm_target *ti)
953 sector_t target_offset = dm_target_offset(ti, sector);
955 return ti->len - target_offset;
958 static sector_t max_io_len(sector_t sector, struct dm_target *ti)
960 sector_t len = max_io_len_target_boundary(sector, ti);
961 sector_t offset, max_len;
964 * Does the target need to split even further?
966 if (ti->max_io_len) {
967 offset = dm_target_offset(ti, sector);
968 if (unlikely(ti->max_io_len & (ti->max_io_len - 1)))
969 max_len = sector_div(offset, ti->max_io_len);
971 max_len = offset & (ti->max_io_len - 1);
972 max_len = ti->max_io_len - max_len;
981 int dm_set_target_max_io_len(struct dm_target *ti, sector_t len)
983 if (len > UINT_MAX) {
984 DMERR("Specified maximum size of target IO (%llu) exceeds limit (%u)",
985 (unsigned long long)len, UINT_MAX);
986 ti->error = "Maximum size of target IO is too large";
990 ti->max_io_len = (uint32_t) len;
994 EXPORT_SYMBOL_GPL(dm_set_target_max_io_len);
996 static void __map_bio(struct dm_target *ti, struct dm_target_io *tio)
1000 struct mapped_device *md;
1001 struct bio *clone = &tio->clone;
1003 clone->bi_end_io = clone_endio;
1004 clone->bi_private = tio;
1007 * Map the clone. If r == 0 we don't need to do
1008 * anything, the target has assumed ownership of
1011 atomic_inc(&tio->io->io_count);
1012 sector = clone->bi_sector;
1013 r = ti->type->map(ti, clone, &tio->info);
1014 if (r == DM_MAPIO_REMAPPED) {
1015 /* the bio has been remapped so dispatch it */
1017 trace_block_bio_remap(bdev_get_queue(clone->bi_bdev), clone,
1018 tio->io->bio->bi_bdev->bd_dev, sector);
1020 generic_make_request(clone);
1021 } else if (r < 0 || r == DM_MAPIO_REQUEUE) {
1022 /* error the io and bail out, or requeue it if needed */
1024 dec_pending(tio->io, r);
1027 DMWARN("unimplemented target map return value: %d", r);
1033 struct mapped_device *md;
1034 struct dm_table *map;
1038 sector_t sector_count;
1043 * Creates a little bio that just does part of a bvec.
1045 static void split_bvec(struct dm_target_io *tio, struct bio *bio,
1046 sector_t sector, unsigned short idx, unsigned int offset,
1047 unsigned int len, struct bio_set *bs)
1049 struct bio *clone = &tio->clone;
1050 struct bio_vec *bv = bio->bi_io_vec + idx;
1052 *clone->bi_io_vec = *bv;
1054 clone->bi_sector = sector;
1055 clone->bi_bdev = bio->bi_bdev;
1056 clone->bi_rw = bio->bi_rw;
1058 clone->bi_size = to_bytes(len);
1059 clone->bi_io_vec->bv_offset = offset;
1060 clone->bi_io_vec->bv_len = clone->bi_size;
1061 clone->bi_flags |= 1 << BIO_CLONED;
1063 if (bio_integrity(bio)) {
1064 bio_integrity_clone(clone, bio, GFP_NOIO);
1065 bio_integrity_trim(clone,
1066 bio_sector_offset(bio, idx, offset), len);
1071 * Creates a bio that consists of range of complete bvecs.
1073 static void clone_bio(struct dm_target_io *tio, struct bio *bio,
1074 sector_t sector, unsigned short idx,
1075 unsigned short bv_count, unsigned int len,
1078 struct bio *clone = &tio->clone;
1080 __bio_clone(clone, bio);
1081 clone->bi_sector = sector;
1082 clone->bi_idx = idx;
1083 clone->bi_vcnt = idx + bv_count;
1084 clone->bi_size = to_bytes(len);
1085 clone->bi_flags &= ~(1 << BIO_SEG_VALID);
1087 if (bio_integrity(bio)) {
1088 bio_integrity_clone(clone, bio, GFP_NOIO);
1090 if (idx != bio->bi_idx || clone->bi_size < bio->bi_size)
1091 bio_integrity_trim(clone,
1092 bio_sector_offset(bio, idx, 0), len);
1096 static struct dm_target_io *alloc_tio(struct clone_info *ci,
1097 struct dm_target *ti, int nr_iovecs)
1099 struct dm_target_io *tio;
1102 clone = bio_alloc_bioset(GFP_NOIO, nr_iovecs, ci->md->bs);
1103 tio = container_of(clone, struct dm_target_io, clone);
1107 memset(&tio->info, 0, sizeof(tio->info));
1112 static void __issue_target_request(struct clone_info *ci, struct dm_target *ti,
1113 unsigned request_nr, sector_t len)
1115 struct dm_target_io *tio = alloc_tio(ci, ti, ci->bio->bi_max_vecs);
1116 struct bio *clone = &tio->clone;
1118 tio->info.target_request_nr = request_nr;
1121 * Discard requests require the bio's inline iovecs be initialized.
1122 * ci->bio->bi_max_vecs is BIO_INLINE_VECS anyway, for both flush
1123 * and discard, so no need for concern about wasted bvec allocations.
1126 __bio_clone(clone, ci->bio);
1128 clone->bi_sector = ci->sector;
1129 clone->bi_size = to_bytes(len);
1135 static void __issue_target_requests(struct clone_info *ci, struct dm_target *ti,
1136 unsigned num_requests, sector_t len)
1138 unsigned request_nr;
1140 for (request_nr = 0; request_nr < num_requests; request_nr++)
1141 __issue_target_request(ci, ti, request_nr, len);
1144 static int __clone_and_map_empty_flush(struct clone_info *ci)
1146 unsigned target_nr = 0;
1147 struct dm_target *ti;
1149 BUG_ON(bio_has_data(ci->bio));
1150 while ((ti = dm_table_get_target(ci->map, target_nr++)))
1151 __issue_target_requests(ci, ti, ti->num_flush_requests, 0);
1157 * Perform all io with a single clone.
1159 static void __clone_and_map_simple(struct clone_info *ci, struct dm_target *ti)
1161 struct bio *bio = ci->bio;
1162 struct dm_target_io *tio;
1164 tio = alloc_tio(ci, ti, bio->bi_max_vecs);
1165 clone_bio(tio, bio, ci->sector, ci->idx, bio->bi_vcnt - ci->idx,
1166 ci->sector_count, ci->md->bs);
1168 ci->sector_count = 0;
1171 static int __clone_and_map_discard(struct clone_info *ci)
1173 struct dm_target *ti;
1177 ti = dm_table_find_target(ci->map, ci->sector);
1178 if (!dm_target_is_valid(ti))
1182 * Even though the device advertised discard support,
1183 * that does not mean every target supports it, and
1184 * reconfiguration might also have changed that since the
1185 * check was performed.
1187 if (!ti->num_discard_requests)
1190 if (!ti->split_discard_requests)
1191 len = min(ci->sector_count, max_io_len_target_boundary(ci->sector, ti));
1193 len = min(ci->sector_count, max_io_len(ci->sector, ti));
1195 __issue_target_requests(ci, ti, ti->num_discard_requests, len);
1198 } while (ci->sector_count -= len);
1203 static int __clone_and_map(struct clone_info *ci)
1205 struct bio *bio = ci->bio;
1206 struct dm_target *ti;
1207 sector_t len = 0, max;
1208 struct dm_target_io *tio;
1210 if (unlikely(bio->bi_rw & REQ_DISCARD))
1211 return __clone_and_map_discard(ci);
1213 ti = dm_table_find_target(ci->map, ci->sector);
1214 if (!dm_target_is_valid(ti))
1217 max = max_io_len(ci->sector, ti);
1219 if (ci->sector_count <= max) {
1221 * Optimise for the simple case where we can do all of
1222 * the remaining io with a single clone.
1224 __clone_and_map_simple(ci, ti);
1226 } else if (to_sector(bio->bi_io_vec[ci->idx].bv_len) <= max) {
1228 * There are some bvecs that don't span targets.
1229 * Do as many of these as possible.
1232 sector_t remaining = max;
1235 for (i = ci->idx; remaining && (i < bio->bi_vcnt); i++) {
1236 bv_len = to_sector(bio->bi_io_vec[i].bv_len);
1238 if (bv_len > remaining)
1241 remaining -= bv_len;
1245 tio = alloc_tio(ci, ti, bio->bi_max_vecs);
1246 clone_bio(tio, bio, ci->sector, ci->idx, i - ci->idx, len,
1251 ci->sector_count -= len;
1256 * Handle a bvec that must be split between two or more targets.
1258 struct bio_vec *bv = bio->bi_io_vec + ci->idx;
1259 sector_t remaining = to_sector(bv->bv_len);
1260 unsigned int offset = 0;
1264 ti = dm_table_find_target(ci->map, ci->sector);
1265 if (!dm_target_is_valid(ti))
1268 max = max_io_len(ci->sector, ti);
1271 len = min(remaining, max);
1273 tio = alloc_tio(ci, ti, 1);
1274 split_bvec(tio, bio, ci->sector, ci->idx,
1275 bv->bv_offset + offset, len, ci->md->bs);
1280 ci->sector_count -= len;
1281 offset += to_bytes(len);
1282 } while (remaining -= len);
1291 * Split the bio into several clones and submit it to targets.
1293 static void __split_and_process_bio(struct mapped_device *md, struct bio *bio)
1295 struct clone_info ci;
1298 ci.map = dm_get_live_table(md);
1299 if (unlikely(!ci.map)) {
1305 ci.io = alloc_io(md);
1307 atomic_set(&ci.io->io_count, 1);
1310 spin_lock_init(&ci.io->endio_lock);
1311 ci.sector = bio->bi_sector;
1312 ci.idx = bio->bi_idx;
1314 start_io_acct(ci.io);
1315 if (bio->bi_rw & REQ_FLUSH) {
1316 ci.bio = &ci.md->flush_bio;
1317 ci.sector_count = 0;
1318 error = __clone_and_map_empty_flush(&ci);
1319 /* dec_pending submits any data associated with flush */
1322 ci.sector_count = bio_sectors(bio);
1323 while (ci.sector_count && !error)
1324 error = __clone_and_map(&ci);
1327 /* drop the extra reference count */
1328 dec_pending(ci.io, error);
1329 dm_table_put(ci.map);
1331 /*-----------------------------------------------------------------
1333 *---------------------------------------------------------------*/
1335 static int dm_merge_bvec(struct request_queue *q,
1336 struct bvec_merge_data *bvm,
1337 struct bio_vec *biovec)
1339 struct mapped_device *md = q->queuedata;
1340 struct dm_table *map = dm_get_live_table(md);
1341 struct dm_target *ti;
1342 sector_t max_sectors;
1348 ti = dm_table_find_target(map, bvm->bi_sector);
1349 if (!dm_target_is_valid(ti))
1353 * Find maximum amount of I/O that won't need splitting
1355 max_sectors = min(max_io_len(bvm->bi_sector, ti),
1356 (sector_t) BIO_MAX_SECTORS);
1357 max_size = (max_sectors << SECTOR_SHIFT) - bvm->bi_size;
1362 * merge_bvec_fn() returns number of bytes
1363 * it can accept at this offset
1364 * max is precomputed maximal io size
1366 if (max_size && ti->type->merge)
1367 max_size = ti->type->merge(ti, bvm, biovec, max_size);
1369 * If the target doesn't support merge method and some of the devices
1370 * provided their merge_bvec method (we know this by looking at
1371 * queue_max_hw_sectors), then we can't allow bios with multiple vector
1372 * entries. So always set max_size to 0, and the code below allows
1375 else if (queue_max_hw_sectors(q) <= PAGE_SIZE >> 9)
1384 * Always allow an entire first page
1386 if (max_size <= biovec->bv_len && !(bvm->bi_size >> SECTOR_SHIFT))
1387 max_size = biovec->bv_len;
1393 * The request function that just remaps the bio built up by
1396 static void _dm_request(struct request_queue *q, struct bio *bio)
1398 int rw = bio_data_dir(bio);
1399 struct mapped_device *md = q->queuedata;
1402 down_read(&md->io_lock);
1404 cpu = part_stat_lock();
1405 part_stat_inc(cpu, &dm_disk(md)->part0, ios[rw]);
1406 part_stat_add(cpu, &dm_disk(md)->part0, sectors[rw], bio_sectors(bio));
1409 /* if we're suspended, we have to queue this io for later */
1410 if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))) {
1411 up_read(&md->io_lock);
1413 if (bio_rw(bio) != READA)
1420 __split_and_process_bio(md, bio);
1421 up_read(&md->io_lock);
1425 static int dm_request_based(struct mapped_device *md)
1427 return blk_queue_stackable(md->queue);
1430 static void dm_request(struct request_queue *q, struct bio *bio)
1432 struct mapped_device *md = q->queuedata;
1434 if (dm_request_based(md))
1435 blk_queue_bio(q, bio);
1437 _dm_request(q, bio);
1440 void dm_dispatch_request(struct request *rq)
1444 if (blk_queue_io_stat(rq->q))
1445 rq->cmd_flags |= REQ_IO_STAT;
1447 rq->start_time = jiffies;
1448 r = blk_insert_cloned_request(rq->q, rq);
1450 dm_complete_request(rq, r);
1452 EXPORT_SYMBOL_GPL(dm_dispatch_request);
1454 static int dm_rq_bio_constructor(struct bio *bio, struct bio *bio_orig,
1457 struct dm_rq_target_io *tio = data;
1458 struct dm_rq_clone_bio_info *info =
1459 container_of(bio, struct dm_rq_clone_bio_info, clone);
1461 info->orig = bio_orig;
1463 bio->bi_end_io = end_clone_bio;
1464 bio->bi_private = info;
1469 static int setup_clone(struct request *clone, struct request *rq,
1470 struct dm_rq_target_io *tio)
1474 r = blk_rq_prep_clone(clone, rq, tio->md->bs, GFP_ATOMIC,
1475 dm_rq_bio_constructor, tio);
1479 clone->cmd = rq->cmd;
1480 clone->cmd_len = rq->cmd_len;
1481 clone->sense = rq->sense;
1482 clone->buffer = rq->buffer;
1483 clone->end_io = end_clone_request;
1484 clone->end_io_data = tio;
1489 static struct request *clone_rq(struct request *rq, struct mapped_device *md,
1492 struct request *clone;
1493 struct dm_rq_target_io *tio;
1495 tio = alloc_rq_tio(md, gfp_mask);
1503 memset(&tio->info, 0, sizeof(tio->info));
1505 clone = &tio->clone;
1506 if (setup_clone(clone, rq, tio)) {
1516 * Called with the queue lock held.
1518 static int dm_prep_fn(struct request_queue *q, struct request *rq)
1520 struct mapped_device *md = q->queuedata;
1521 struct request *clone;
1523 if (unlikely(rq->special)) {
1524 DMWARN("Already has something in rq->special.");
1525 return BLKPREP_KILL;
1528 clone = clone_rq(rq, md, GFP_ATOMIC);
1530 return BLKPREP_DEFER;
1532 rq->special = clone;
1533 rq->cmd_flags |= REQ_DONTPREP;
1540 * 0 : the request has been processed (not requeued)
1541 * !0 : the request has been requeued
1543 static int map_request(struct dm_target *ti, struct request *clone,
1544 struct mapped_device *md)
1546 int r, requeued = 0;
1547 struct dm_rq_target_io *tio = clone->end_io_data;
1550 r = ti->type->map_rq(ti, clone, &tio->info);
1552 case DM_MAPIO_SUBMITTED:
1553 /* The target has taken the I/O to submit by itself later */
1555 case DM_MAPIO_REMAPPED:
1556 /* The target has remapped the I/O so dispatch it */
1557 trace_block_rq_remap(clone->q, clone, disk_devt(dm_disk(md)),
1558 blk_rq_pos(tio->orig));
1559 dm_dispatch_request(clone);
1561 case DM_MAPIO_REQUEUE:
1562 /* The target wants to requeue the I/O */
1563 dm_requeue_unmapped_request(clone);
1568 DMWARN("unimplemented target map return value: %d", r);
1572 /* The target wants to complete the I/O */
1573 dm_kill_unmapped_request(clone, r);
1580 static struct request *dm_start_request(struct mapped_device *md, struct request *orig)
1582 struct request *clone;
1584 blk_start_request(orig);
1585 clone = orig->special;
1586 atomic_inc(&md->pending[rq_data_dir(clone)]);
1589 * Hold the md reference here for the in-flight I/O.
1590 * We can't rely on the reference count by device opener,
1591 * because the device may be closed during the request completion
1592 * when all bios are completed.
1593 * See the comment in rq_completed() too.
1601 * q->request_fn for request-based dm.
1602 * Called with the queue lock held.
1604 static void dm_request_fn(struct request_queue *q)
1606 struct mapped_device *md = q->queuedata;
1607 struct dm_table *map = dm_get_live_table(md);
1608 struct dm_target *ti;
1609 struct request *rq, *clone;
1613 * For suspend, check blk_queue_stopped() and increment
1614 * ->pending within a single queue_lock not to increment the
1615 * number of in-flight I/Os after the queue is stopped in
1618 while (!blk_queue_stopped(q)) {
1619 rq = blk_peek_request(q);
1623 /* always use block 0 to find the target for flushes for now */
1625 if (!(rq->cmd_flags & REQ_FLUSH))
1626 pos = blk_rq_pos(rq);
1628 ti = dm_table_find_target(map, pos);
1629 if (!dm_target_is_valid(ti)) {
1631 * Must perform setup, that dm_done() requires,
1632 * before calling dm_kill_unmapped_request
1634 DMERR_LIMIT("request attempted access beyond the end of device");
1635 clone = dm_start_request(md, rq);
1636 dm_kill_unmapped_request(clone, -EIO);
1640 if (ti->type->busy && ti->type->busy(ti))
1643 clone = dm_start_request(md, rq);
1645 spin_unlock(q->queue_lock);
1646 if (map_request(ti, clone, md))
1649 BUG_ON(!irqs_disabled());
1650 spin_lock(q->queue_lock);
1656 BUG_ON(!irqs_disabled());
1657 spin_lock(q->queue_lock);
1660 blk_delay_queue(q, HZ / 10);
1665 int dm_underlying_device_busy(struct request_queue *q)
1667 return blk_lld_busy(q);
1669 EXPORT_SYMBOL_GPL(dm_underlying_device_busy);
1671 static int dm_lld_busy(struct request_queue *q)
1674 struct mapped_device *md = q->queuedata;
1675 struct dm_table *map = dm_get_live_table(md);
1677 if (!map || test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))
1680 r = dm_table_any_busy_target(map);
1687 static int dm_any_congested(void *congested_data, int bdi_bits)
1690 struct mapped_device *md = congested_data;
1691 struct dm_table *map;
1693 if (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
1694 map = dm_get_live_table(md);
1697 * Request-based dm cares about only own queue for
1698 * the query about congestion status of request_queue
1700 if (dm_request_based(md))
1701 r = md->queue->backing_dev_info.state &
1704 r = dm_table_any_congested(map, bdi_bits);
1713 /*-----------------------------------------------------------------
1714 * An IDR is used to keep track of allocated minor numbers.
1715 *---------------------------------------------------------------*/
1716 static void free_minor(int minor)
1718 spin_lock(&_minor_lock);
1719 idr_remove(&_minor_idr, minor);
1720 spin_unlock(&_minor_lock);
1724 * See if the device with a specific minor # is free.
1726 static int specific_minor(int minor)
1730 if (minor >= (1 << MINORBITS))
1733 r = idr_pre_get(&_minor_idr, GFP_KERNEL);
1737 spin_lock(&_minor_lock);
1739 if (idr_find(&_minor_idr, minor)) {
1744 r = idr_get_new_above(&_minor_idr, MINOR_ALLOCED, minor, &m);
1749 idr_remove(&_minor_idr, m);
1755 spin_unlock(&_minor_lock);
1759 static int next_free_minor(int *minor)
1763 r = idr_pre_get(&_minor_idr, GFP_KERNEL);
1767 spin_lock(&_minor_lock);
1769 r = idr_get_new(&_minor_idr, MINOR_ALLOCED, &m);
1773 if (m >= (1 << MINORBITS)) {
1774 idr_remove(&_minor_idr, m);
1782 spin_unlock(&_minor_lock);
1786 static const struct block_device_operations dm_blk_dops;
1788 static void dm_wq_work(struct work_struct *work);
1790 static void dm_init_md_queue(struct mapped_device *md)
1793 * Request-based dm devices cannot be stacked on top of bio-based dm
1794 * devices. The type of this dm device has not been decided yet.
1795 * The type is decided at the first table loading time.
1796 * To prevent problematic device stacking, clear the queue flag
1797 * for request stacking support until then.
1799 * This queue is new, so no concurrency on the queue_flags.
1801 queue_flag_clear_unlocked(QUEUE_FLAG_STACKABLE, md->queue);
1803 md->queue->queuedata = md;
1804 md->queue->backing_dev_info.congested_fn = dm_any_congested;
1805 md->queue->backing_dev_info.congested_data = md;
1806 blk_queue_make_request(md->queue, dm_request);
1807 blk_queue_bounce_limit(md->queue, BLK_BOUNCE_ANY);
1808 blk_queue_merge_bvec(md->queue, dm_merge_bvec);
1812 * Allocate and initialise a blank device with a given minor.
1814 static struct mapped_device *alloc_dev(int minor)
1817 struct mapped_device *md = kzalloc(sizeof(*md), GFP_KERNEL);
1821 DMWARN("unable to allocate device, out of memory.");
1825 if (!try_module_get(THIS_MODULE))
1826 goto bad_module_get;
1828 /* get a minor number for the dev */
1829 if (minor == DM_ANY_MINOR)
1830 r = next_free_minor(&minor);
1832 r = specific_minor(minor);
1836 md->type = DM_TYPE_NONE;
1837 init_rwsem(&md->io_lock);
1838 mutex_init(&md->suspend_lock);
1839 mutex_init(&md->type_lock);
1840 spin_lock_init(&md->deferred_lock);
1841 rwlock_init(&md->map_lock);
1842 atomic_set(&md->holders, 1);
1843 atomic_set(&md->open_count, 0);
1844 atomic_set(&md->event_nr, 0);
1845 atomic_set(&md->uevent_seq, 0);
1846 INIT_LIST_HEAD(&md->uevent_list);
1847 spin_lock_init(&md->uevent_lock);
1849 md->queue = blk_alloc_queue(GFP_KERNEL);
1853 dm_init_md_queue(md);
1855 md->disk = alloc_disk(1);
1859 atomic_set(&md->pending[0], 0);
1860 atomic_set(&md->pending[1], 0);
1861 init_waitqueue_head(&md->wait);
1862 INIT_WORK(&md->work, dm_wq_work);
1863 init_waitqueue_head(&md->eventq);
1865 md->disk->major = _major;
1866 md->disk->first_minor = minor;
1867 md->disk->fops = &dm_blk_dops;
1868 md->disk->queue = md->queue;
1869 md->disk->private_data = md;
1870 sprintf(md->disk->disk_name, "dm-%d", minor);
1872 format_dev_t(md->name, MKDEV(_major, minor));
1874 md->wq = alloc_workqueue("kdmflush",
1875 WQ_NON_REENTRANT | WQ_MEM_RECLAIM, 0);
1879 md->bdev = bdget_disk(md->disk, 0);
1883 bio_init(&md->flush_bio);
1884 md->flush_bio.bi_bdev = md->bdev;
1885 md->flush_bio.bi_rw = WRITE_FLUSH;
1887 /* Populate the mapping, nobody knows we exist yet */
1888 spin_lock(&_minor_lock);
1889 old_md = idr_replace(&_minor_idr, md, minor);
1890 spin_unlock(&_minor_lock);
1892 BUG_ON(old_md != MINOR_ALLOCED);
1897 destroy_workqueue(md->wq);
1899 del_gendisk(md->disk);
1902 blk_cleanup_queue(md->queue);
1906 module_put(THIS_MODULE);
1912 static void unlock_fs(struct mapped_device *md);
1914 static void free_dev(struct mapped_device *md)
1916 int minor = MINOR(disk_devt(md->disk));
1920 destroy_workqueue(md->wq);
1922 mempool_destroy(md->tio_pool);
1924 mempool_destroy(md->io_pool);
1926 bioset_free(md->bs);
1927 blk_integrity_unregister(md->disk);
1928 del_gendisk(md->disk);
1931 spin_lock(&_minor_lock);
1932 md->disk->private_data = NULL;
1933 spin_unlock(&_minor_lock);
1936 blk_cleanup_queue(md->queue);
1937 module_put(THIS_MODULE);
1941 static void __bind_mempools(struct mapped_device *md, struct dm_table *t)
1943 struct dm_md_mempools *p;
1945 if (md->io_pool && (md->tio_pool || dm_table_get_type(t) == DM_TYPE_BIO_BASED) && md->bs)
1946 /* the md already has necessary mempools */
1949 p = dm_table_get_md_mempools(t);
1950 BUG_ON(!p || md->io_pool || md->tio_pool || md->bs);
1952 md->io_pool = p->io_pool;
1954 md->tio_pool = p->tio_pool;
1960 /* mempool bind completed, now no need any mempools in the table */
1961 dm_table_free_md_mempools(t);
1965 * Bind a table to the device.
1967 static void event_callback(void *context)
1969 unsigned long flags;
1971 struct mapped_device *md = (struct mapped_device *) context;
1973 spin_lock_irqsave(&md->uevent_lock, flags);
1974 list_splice_init(&md->uevent_list, &uevents);
1975 spin_unlock_irqrestore(&md->uevent_lock, flags);
1977 dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj);
1979 atomic_inc(&md->event_nr);
1980 wake_up(&md->eventq);
1984 * Protected by md->suspend_lock obtained by dm_swap_table().
1986 static void __set_size(struct mapped_device *md, sector_t size)
1988 set_capacity(md->disk, size);
1990 i_size_write(md->bdev->bd_inode, (loff_t)size << SECTOR_SHIFT);
1994 * Return 1 if the queue has a compulsory merge_bvec_fn function.
1996 * If this function returns 0, then the device is either a non-dm
1997 * device without a merge_bvec_fn, or it is a dm device that is
1998 * able to split any bios it receives that are too big.
2000 int dm_queue_merge_is_compulsory(struct request_queue *q)
2002 struct mapped_device *dev_md;
2004 if (!q->merge_bvec_fn)
2007 if (q->make_request_fn == dm_request) {
2008 dev_md = q->queuedata;
2009 if (test_bit(DMF_MERGE_IS_OPTIONAL, &dev_md->flags))
2016 static int dm_device_merge_is_compulsory(struct dm_target *ti,
2017 struct dm_dev *dev, sector_t start,
2018 sector_t len, void *data)
2020 struct block_device *bdev = dev->bdev;
2021 struct request_queue *q = bdev_get_queue(bdev);
2023 return dm_queue_merge_is_compulsory(q);
2027 * Return 1 if it is acceptable to ignore merge_bvec_fn based
2028 * on the properties of the underlying devices.
2030 static int dm_table_merge_is_optional(struct dm_table *table)
2033 struct dm_target *ti;
2035 while (i < dm_table_get_num_targets(table)) {
2036 ti = dm_table_get_target(table, i++);
2038 if (ti->type->iterate_devices &&
2039 ti->type->iterate_devices(ti, dm_device_merge_is_compulsory, NULL))
2047 * Returns old map, which caller must destroy.
2049 static struct dm_table *__bind(struct mapped_device *md, struct dm_table *t,
2050 struct queue_limits *limits)
2052 struct dm_table *old_map;
2053 struct request_queue *q = md->queue;
2055 unsigned long flags;
2056 int merge_is_optional;
2058 size = dm_table_get_size(t);
2061 * Wipe any geometry if the size of the table changed.
2063 if (size != get_capacity(md->disk))
2064 memset(&md->geometry, 0, sizeof(md->geometry));
2066 __set_size(md, size);
2068 dm_table_event_callback(t, event_callback, md);
2071 * The queue hasn't been stopped yet, if the old table type wasn't
2072 * for request-based during suspension. So stop it to prevent
2073 * I/O mapping before resume.
2074 * This must be done before setting the queue restrictions,
2075 * because request-based dm may be run just after the setting.
2077 if (dm_table_request_based(t) && !blk_queue_stopped(q))
2080 __bind_mempools(md, t);
2082 merge_is_optional = dm_table_merge_is_optional(t);
2084 write_lock_irqsave(&md->map_lock, flags);
2087 md->immutable_target_type = dm_table_get_immutable_target_type(t);
2089 dm_table_set_restrictions(t, q, limits);
2090 if (merge_is_optional)
2091 set_bit(DMF_MERGE_IS_OPTIONAL, &md->flags);
2093 clear_bit(DMF_MERGE_IS_OPTIONAL, &md->flags);
2094 write_unlock_irqrestore(&md->map_lock, flags);
2100 * Returns unbound table for the caller to free.
2102 static struct dm_table *__unbind(struct mapped_device *md)
2104 struct dm_table *map = md->map;
2105 unsigned long flags;
2110 dm_table_event_callback(map, NULL, NULL);
2111 write_lock_irqsave(&md->map_lock, flags);
2113 write_unlock_irqrestore(&md->map_lock, flags);
2119 * Constructor for a new device.
2121 int dm_create(int minor, struct mapped_device **result)
2123 struct mapped_device *md;
2125 md = alloc_dev(minor);
2136 * Functions to manage md->type.
2137 * All are required to hold md->type_lock.
2139 void dm_lock_md_type(struct mapped_device *md)
2141 mutex_lock(&md->type_lock);
2144 void dm_unlock_md_type(struct mapped_device *md)
2146 mutex_unlock(&md->type_lock);
2149 void dm_set_md_type(struct mapped_device *md, unsigned type)
2154 unsigned dm_get_md_type(struct mapped_device *md)
2159 struct target_type *dm_get_immutable_target_type(struct mapped_device *md)
2161 return md->immutable_target_type;
2165 * Fully initialize a request-based queue (->elevator, ->request_fn, etc).
2167 static int dm_init_request_based_queue(struct mapped_device *md)
2169 struct request_queue *q = NULL;
2171 if (md->queue->elevator)
2174 /* Fully initialize the queue */
2175 q = blk_init_allocated_queue(md->queue, dm_request_fn, NULL);
2180 dm_init_md_queue(md);
2181 blk_queue_softirq_done(md->queue, dm_softirq_done);
2182 blk_queue_prep_rq(md->queue, dm_prep_fn);
2183 blk_queue_lld_busy(md->queue, dm_lld_busy);
2185 elv_register_queue(md->queue);
2191 * Setup the DM device's queue based on md's type
2193 int dm_setup_md_queue(struct mapped_device *md)
2195 if ((dm_get_md_type(md) == DM_TYPE_REQUEST_BASED) &&
2196 !dm_init_request_based_queue(md)) {
2197 DMWARN("Cannot initialize queue for request-based mapped device");
2204 static struct mapped_device *dm_find_md(dev_t dev)
2206 struct mapped_device *md;
2207 unsigned minor = MINOR(dev);
2209 if (MAJOR(dev) != _major || minor >= (1 << MINORBITS))
2212 spin_lock(&_minor_lock);
2214 md = idr_find(&_minor_idr, minor);
2215 if (md && (md == MINOR_ALLOCED ||
2216 (MINOR(disk_devt(dm_disk(md))) != minor) ||
2217 dm_deleting_md(md) ||
2218 test_bit(DMF_FREEING, &md->flags))) {
2224 spin_unlock(&_minor_lock);
2229 struct mapped_device *dm_get_md(dev_t dev)
2231 struct mapped_device *md = dm_find_md(dev);
2238 EXPORT_SYMBOL_GPL(dm_get_md);
2240 void *dm_get_mdptr(struct mapped_device *md)
2242 return md->interface_ptr;
2245 void dm_set_mdptr(struct mapped_device *md, void *ptr)
2247 md->interface_ptr = ptr;
2250 void dm_get(struct mapped_device *md)
2252 atomic_inc(&md->holders);
2253 BUG_ON(test_bit(DMF_FREEING, &md->flags));
2256 const char *dm_device_name(struct mapped_device *md)
2260 EXPORT_SYMBOL_GPL(dm_device_name);
2262 static void __dm_destroy(struct mapped_device *md, bool wait)
2264 struct dm_table *map;
2268 spin_lock(&_minor_lock);
2269 map = dm_get_live_table(md);
2270 idr_replace(&_minor_idr, MINOR_ALLOCED, MINOR(disk_devt(dm_disk(md))));
2271 set_bit(DMF_FREEING, &md->flags);
2272 spin_unlock(&_minor_lock);
2274 if (!dm_suspended_md(md)) {
2275 dm_table_presuspend_targets(map);
2276 dm_table_postsuspend_targets(map);
2280 * Rare, but there may be I/O requests still going to complete,
2281 * for example. Wait for all references to disappear.
2282 * No one should increment the reference count of the mapped_device,
2283 * after the mapped_device state becomes DMF_FREEING.
2286 while (atomic_read(&md->holders))
2288 else if (atomic_read(&md->holders))
2289 DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)",
2290 dm_device_name(md), atomic_read(&md->holders));
2294 dm_table_destroy(__unbind(md));
2298 void dm_destroy(struct mapped_device *md)
2300 __dm_destroy(md, true);
2303 void dm_destroy_immediate(struct mapped_device *md)
2305 __dm_destroy(md, false);
2308 void dm_put(struct mapped_device *md)
2310 atomic_dec(&md->holders);
2312 EXPORT_SYMBOL_GPL(dm_put);
2314 static int dm_wait_for_completion(struct mapped_device *md, int interruptible)
2317 DECLARE_WAITQUEUE(wait, current);
2319 add_wait_queue(&md->wait, &wait);
2322 set_current_state(interruptible);
2324 if (!md_in_flight(md))
2327 if (interruptible == TASK_INTERRUPTIBLE &&
2328 signal_pending(current)) {
2335 set_current_state(TASK_RUNNING);
2337 remove_wait_queue(&md->wait, &wait);
2343 * Process the deferred bios
2345 static void dm_wq_work(struct work_struct *work)
2347 struct mapped_device *md = container_of(work, struct mapped_device,
2351 down_read(&md->io_lock);
2353 while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
2354 spin_lock_irq(&md->deferred_lock);
2355 c = bio_list_pop(&md->deferred);
2356 spin_unlock_irq(&md->deferred_lock);
2361 up_read(&md->io_lock);
2363 if (dm_request_based(md))
2364 generic_make_request(c);
2366 __split_and_process_bio(md, c);
2368 down_read(&md->io_lock);
2371 up_read(&md->io_lock);
2374 static void dm_queue_flush(struct mapped_device *md)
2376 clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2377 smp_mb__after_clear_bit();
2378 queue_work(md->wq, &md->work);
2382 * Swap in a new table, returning the old one for the caller to destroy.
2384 struct dm_table *dm_swap_table(struct mapped_device *md, struct dm_table *table)
2386 struct dm_table *live_map, *map = ERR_PTR(-EINVAL);
2387 struct queue_limits limits;
2390 mutex_lock(&md->suspend_lock);
2392 /* device must be suspended */
2393 if (!dm_suspended_md(md))
2397 * If the new table has no data devices, retain the existing limits.
2398 * This helps multipath with queue_if_no_path if all paths disappear,
2399 * then new I/O is queued based on these limits, and then some paths
2402 if (dm_table_has_no_data_devices(table)) {
2403 live_map = dm_get_live_table(md);
2405 limits = md->queue->limits;
2406 dm_table_put(live_map);
2409 r = dm_calculate_queue_limits(table, &limits);
2415 map = __bind(md, table, &limits);
2418 mutex_unlock(&md->suspend_lock);
2423 * Functions to lock and unlock any filesystem running on the
2426 static int lock_fs(struct mapped_device *md)
2430 WARN_ON(md->frozen_sb);
2432 md->frozen_sb = freeze_bdev(md->bdev);
2433 if (IS_ERR(md->frozen_sb)) {
2434 r = PTR_ERR(md->frozen_sb);
2435 md->frozen_sb = NULL;
2439 set_bit(DMF_FROZEN, &md->flags);
2444 static void unlock_fs(struct mapped_device *md)
2446 if (!test_bit(DMF_FROZEN, &md->flags))
2449 thaw_bdev(md->bdev, md->frozen_sb);
2450 md->frozen_sb = NULL;
2451 clear_bit(DMF_FROZEN, &md->flags);
2455 * We need to be able to change a mapping table under a mounted
2456 * filesystem. For example we might want to move some data in
2457 * the background. Before the table can be swapped with
2458 * dm_bind_table, dm_suspend must be called to flush any in
2459 * flight bios and ensure that any further io gets deferred.
2462 * Suspend mechanism in request-based dm.
2464 * 1. Flush all I/Os by lock_fs() if needed.
2465 * 2. Stop dispatching any I/O by stopping the request_queue.
2466 * 3. Wait for all in-flight I/Os to be completed or requeued.
2468 * To abort suspend, start the request_queue.
2470 int dm_suspend(struct mapped_device *md, unsigned suspend_flags)
2472 struct dm_table *map = NULL;
2474 int do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG ? 1 : 0;
2475 int noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG ? 1 : 0;
2477 mutex_lock(&md->suspend_lock);
2479 if (dm_suspended_md(md)) {
2484 map = dm_get_live_table(md);
2487 * DMF_NOFLUSH_SUSPENDING must be set before presuspend.
2488 * This flag is cleared before dm_suspend returns.
2491 set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2493 /* This does not get reverted if there's an error later. */
2494 dm_table_presuspend_targets(map);
2497 * Flush I/O to the device.
2498 * Any I/O submitted after lock_fs() may not be flushed.
2499 * noflush takes precedence over do_lockfs.
2500 * (lock_fs() flushes I/Os and waits for them to complete.)
2502 if (!noflush && do_lockfs) {
2509 * Here we must make sure that no processes are submitting requests
2510 * to target drivers i.e. no one may be executing
2511 * __split_and_process_bio. This is called from dm_request and
2514 * To get all processes out of __split_and_process_bio in dm_request,
2515 * we take the write lock. To prevent any process from reentering
2516 * __split_and_process_bio from dm_request and quiesce the thread
2517 * (dm_wq_work), we set BMF_BLOCK_IO_FOR_SUSPEND and call
2518 * flush_workqueue(md->wq).
2520 down_write(&md->io_lock);
2521 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2522 up_write(&md->io_lock);
2525 * Stop md->queue before flushing md->wq in case request-based
2526 * dm defers requests to md->wq from md->queue.
2528 if (dm_request_based(md))
2529 stop_queue(md->queue);
2531 flush_workqueue(md->wq);
2534 * At this point no more requests are entering target request routines.
2535 * We call dm_wait_for_completion to wait for all existing requests
2538 r = dm_wait_for_completion(md, TASK_INTERRUPTIBLE);
2540 down_write(&md->io_lock);
2542 clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2543 up_write(&md->io_lock);
2545 /* were we interrupted ? */
2549 if (dm_request_based(md))
2550 start_queue(md->queue);
2553 goto out; /* pushback list is already flushed, so skip flush */
2557 * If dm_wait_for_completion returned 0, the device is completely
2558 * quiescent now. There is no request-processing activity. All new
2559 * requests are being added to md->deferred list.
2562 set_bit(DMF_SUSPENDED, &md->flags);
2564 dm_table_postsuspend_targets(map);
2570 mutex_unlock(&md->suspend_lock);
2574 int dm_resume(struct mapped_device *md)
2577 struct dm_table *map = NULL;
2579 mutex_lock(&md->suspend_lock);
2580 if (!dm_suspended_md(md))
2583 map = dm_get_live_table(md);
2584 if (!map || !dm_table_get_size(map))
2587 r = dm_table_resume_targets(map);
2594 * Flushing deferred I/Os must be done after targets are resumed
2595 * so that mapping of targets can work correctly.
2596 * Request-based dm is queueing the deferred I/Os in its request_queue.
2598 if (dm_request_based(md))
2599 start_queue(md->queue);
2603 clear_bit(DMF_SUSPENDED, &md->flags);
2608 mutex_unlock(&md->suspend_lock);
2613 /*-----------------------------------------------------------------
2614 * Event notification.
2615 *---------------------------------------------------------------*/
2616 int dm_kobject_uevent(struct mapped_device *md, enum kobject_action action,
2619 char udev_cookie[DM_COOKIE_LENGTH];
2620 char *envp[] = { udev_cookie, NULL };
2623 return kobject_uevent(&disk_to_dev(md->disk)->kobj, action);
2625 snprintf(udev_cookie, DM_COOKIE_LENGTH, "%s=%u",
2626 DM_COOKIE_ENV_VAR_NAME, cookie);
2627 return kobject_uevent_env(&disk_to_dev(md->disk)->kobj,
2632 uint32_t dm_next_uevent_seq(struct mapped_device *md)
2634 return atomic_add_return(1, &md->uevent_seq);
2637 uint32_t dm_get_event_nr(struct mapped_device *md)
2639 return atomic_read(&md->event_nr);
2642 int dm_wait_event(struct mapped_device *md, int event_nr)
2644 return wait_event_interruptible(md->eventq,
2645 (event_nr != atomic_read(&md->event_nr)));
2648 void dm_uevent_add(struct mapped_device *md, struct list_head *elist)
2650 unsigned long flags;
2652 spin_lock_irqsave(&md->uevent_lock, flags);
2653 list_add(elist, &md->uevent_list);
2654 spin_unlock_irqrestore(&md->uevent_lock, flags);
2658 * The gendisk is only valid as long as you have a reference
2661 struct gendisk *dm_disk(struct mapped_device *md)
2666 struct kobject *dm_kobject(struct mapped_device *md)
2672 * struct mapped_device should not be exported outside of dm.c
2673 * so use this check to verify that kobj is part of md structure
2675 struct mapped_device *dm_get_from_kobject(struct kobject *kobj)
2677 struct mapped_device *md;
2679 md = container_of(kobj, struct mapped_device, kobj);
2680 if (&md->kobj != kobj)
2683 if (test_bit(DMF_FREEING, &md->flags) ||
2691 int dm_suspended_md(struct mapped_device *md)
2693 return test_bit(DMF_SUSPENDED, &md->flags);
2696 int dm_suspended(struct dm_target *ti)
2698 return dm_suspended_md(dm_table_get_md(ti->table));
2700 EXPORT_SYMBOL_GPL(dm_suspended);
2702 int dm_noflush_suspending(struct dm_target *ti)
2704 return __noflush_suspending(dm_table_get_md(ti->table));
2706 EXPORT_SYMBOL_GPL(dm_noflush_suspending);
2708 struct dm_md_mempools *dm_alloc_md_mempools(unsigned type, unsigned integrity)
2710 struct dm_md_mempools *pools = kmalloc(sizeof(*pools), GFP_KERNEL);
2711 unsigned int pool_size = (type == DM_TYPE_BIO_BASED) ? 16 : MIN_IOS;
2716 pools->io_pool = (type == DM_TYPE_BIO_BASED) ?
2717 mempool_create_slab_pool(MIN_IOS, _io_cache) :
2718 mempool_create_slab_pool(MIN_IOS, _rq_bio_info_cache);
2719 if (!pools->io_pool)
2720 goto free_pools_and_out;
2722 pools->tio_pool = NULL;
2723 if (type == DM_TYPE_REQUEST_BASED) {
2724 pools->tio_pool = mempool_create_slab_pool(MIN_IOS, _rq_tio_cache);
2725 if (!pools->tio_pool)
2726 goto free_io_pool_and_out;
2729 pools->bs = (type == DM_TYPE_BIO_BASED) ?
2730 bioset_create(pool_size,
2731 offsetof(struct dm_target_io, clone)) :
2732 bioset_create(pool_size,
2733 offsetof(struct dm_rq_clone_bio_info, clone));
2735 goto free_tio_pool_and_out;
2737 if (integrity && bioset_integrity_create(pools->bs, pool_size))
2738 goto free_bioset_and_out;
2742 free_bioset_and_out:
2743 bioset_free(pools->bs);
2745 free_tio_pool_and_out:
2746 if (pools->tio_pool)
2747 mempool_destroy(pools->tio_pool);
2749 free_io_pool_and_out:
2750 mempool_destroy(pools->io_pool);
2758 void dm_free_md_mempools(struct dm_md_mempools *pools)
2764 mempool_destroy(pools->io_pool);
2766 if (pools->tio_pool)
2767 mempool_destroy(pools->tio_pool);
2770 bioset_free(pools->bs);
2775 static const struct block_device_operations dm_blk_dops = {
2776 .open = dm_blk_open,
2777 .release = dm_blk_close,
2778 .ioctl = dm_blk_ioctl,
2779 .getgeo = dm_blk_getgeo,
2780 .owner = THIS_MODULE
2783 EXPORT_SYMBOL(dm_get_mapinfo);
2788 module_init(dm_init);
2789 module_exit(dm_exit);
2791 module_param(major, uint, 0);
2792 MODULE_PARM_DESC(major, "The major number of the device mapper");
2793 MODULE_DESCRIPTION(DM_NAME " driver");
2794 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
2795 MODULE_LICENSE("GPL");