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 * A dummy definition to make RCU happy.
121 * struct dm_table should never be dereferenced in this file.
128 * Work processed by per-device workqueue.
130 struct mapped_device {
131 struct srcu_struct io_barrier;
132 struct mutex suspend_lock;
137 * The current mapping.
138 * Use dm_get_live_table{_fast} or take suspend_lock for
141 struct dm_table *map;
145 struct request_queue *queue;
147 /* Protect queue and type against concurrent access. */
148 struct mutex type_lock;
150 struct target_type *immutable_target_type;
152 struct gendisk *disk;
158 * A list of ios that arrived while we were suspended.
161 wait_queue_head_t wait;
162 struct work_struct work;
163 struct bio_list deferred;
164 spinlock_t deferred_lock;
167 * Processing queue (flush)
169 struct workqueue_struct *wq;
172 * io objects are allocated from here.
182 wait_queue_head_t eventq;
184 struct list_head uevent_list;
185 spinlock_t uevent_lock; /* Protect access to uevent_list */
188 * freeze/thaw support require holding onto a super block
190 struct super_block *frozen_sb;
191 struct block_device *bdev;
193 /* forced geometry settings */
194 struct hd_geometry geometry;
199 /* zero-length flush that will be cloned and submitted to targets */
200 struct bio flush_bio;
204 * For mempools pre-allocation at the table loading time.
206 struct dm_md_mempools {
212 static struct kmem_cache *_io_cache;
213 static struct kmem_cache *_rq_tio_cache;
215 static int __init local_init(void)
219 /* allocate a slab for the dm_ios */
220 _io_cache = KMEM_CACHE(dm_io, 0);
224 _rq_tio_cache = KMEM_CACHE(dm_rq_target_io, 0);
226 goto out_free_io_cache;
228 r = dm_uevent_init();
230 goto out_free_rq_tio_cache;
233 r = register_blkdev(_major, _name);
235 goto out_uevent_exit;
244 out_free_rq_tio_cache:
245 kmem_cache_destroy(_rq_tio_cache);
247 kmem_cache_destroy(_io_cache);
252 static void local_exit(void)
254 kmem_cache_destroy(_rq_tio_cache);
255 kmem_cache_destroy(_io_cache);
256 unregister_blkdev(_major, _name);
261 DMINFO("cleaned up");
264 static int (*_inits[])(void) __initdata = {
274 static void (*_exits[])(void) = {
284 static int __init dm_init(void)
286 const int count = ARRAY_SIZE(_inits);
290 for (i = 0; i < count; i++) {
305 static void __exit dm_exit(void)
307 int i = ARRAY_SIZE(_exits);
313 * Should be empty by this point.
315 idr_destroy(&_minor_idr);
319 * Block device functions
321 int dm_deleting_md(struct mapped_device *md)
323 return test_bit(DMF_DELETING, &md->flags);
326 static int dm_blk_open(struct block_device *bdev, fmode_t mode)
328 struct mapped_device *md;
330 spin_lock(&_minor_lock);
332 md = bdev->bd_disk->private_data;
336 if (test_bit(DMF_FREEING, &md->flags) ||
337 dm_deleting_md(md)) {
343 atomic_inc(&md->open_count);
346 spin_unlock(&_minor_lock);
348 return md ? 0 : -ENXIO;
351 static void dm_blk_close(struct gendisk *disk, fmode_t mode)
353 struct mapped_device *md = disk->private_data;
355 spin_lock(&_minor_lock);
357 atomic_dec(&md->open_count);
360 spin_unlock(&_minor_lock);
363 int dm_open_count(struct mapped_device *md)
365 return atomic_read(&md->open_count);
369 * Guarantees nothing is using the device before it's deleted.
371 int dm_lock_for_deletion(struct mapped_device *md)
375 spin_lock(&_minor_lock);
377 if (dm_open_count(md))
380 set_bit(DMF_DELETING, &md->flags);
382 spin_unlock(&_minor_lock);
387 static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo)
389 struct mapped_device *md = bdev->bd_disk->private_data;
391 return dm_get_geometry(md, geo);
394 static int dm_blk_ioctl(struct block_device *bdev, fmode_t mode,
395 unsigned int cmd, unsigned long arg)
397 struct mapped_device *md = bdev->bd_disk->private_data;
399 struct dm_table *map;
400 struct dm_target *tgt;
404 map = dm_get_live_table(md, &srcu_idx);
406 if (!map || !dm_table_get_size(map))
409 /* We only support devices that have a single target */
410 if (dm_table_get_num_targets(map) != 1)
413 tgt = dm_table_get_target(map, 0);
415 if (dm_suspended_md(md)) {
420 if (tgt->type->ioctl)
421 r = tgt->type->ioctl(tgt, cmd, arg);
424 dm_put_live_table(md, srcu_idx);
426 if (r == -ENOTCONN) {
434 static struct dm_io *alloc_io(struct mapped_device *md)
436 return mempool_alloc(md->io_pool, GFP_NOIO);
439 static void free_io(struct mapped_device *md, struct dm_io *io)
441 mempool_free(io, md->io_pool);
444 static void free_tio(struct mapped_device *md, struct dm_target_io *tio)
446 bio_put(&tio->clone);
449 static struct dm_rq_target_io *alloc_rq_tio(struct mapped_device *md,
452 return mempool_alloc(md->io_pool, gfp_mask);
455 static void free_rq_tio(struct dm_rq_target_io *tio)
457 mempool_free(tio, tio->md->io_pool);
460 static int md_in_flight(struct mapped_device *md)
462 return atomic_read(&md->pending[READ]) +
463 atomic_read(&md->pending[WRITE]);
466 static void start_io_acct(struct dm_io *io)
468 struct mapped_device *md = io->md;
470 int rw = bio_data_dir(io->bio);
472 io->start_time = jiffies;
474 cpu = part_stat_lock();
475 part_round_stats(cpu, &dm_disk(md)->part0);
477 atomic_set(&dm_disk(md)->part0.in_flight[rw],
478 atomic_inc_return(&md->pending[rw]));
481 static void end_io_acct(struct dm_io *io)
483 struct mapped_device *md = io->md;
484 struct bio *bio = io->bio;
485 unsigned long duration = jiffies - io->start_time;
487 int rw = bio_data_dir(bio);
489 cpu = part_stat_lock();
490 part_round_stats(cpu, &dm_disk(md)->part0);
491 part_stat_add(cpu, &dm_disk(md)->part0, ticks[rw], duration);
495 * After this is decremented the bio must not be touched if it is
498 pending = atomic_dec_return(&md->pending[rw]);
499 atomic_set(&dm_disk(md)->part0.in_flight[rw], pending);
500 pending += atomic_read(&md->pending[rw^0x1]);
502 /* nudge anyone waiting on suspend queue */
508 * Add the bio to the list of deferred io.
510 static void queue_io(struct mapped_device *md, struct bio *bio)
514 spin_lock_irqsave(&md->deferred_lock, flags);
515 bio_list_add(&md->deferred, bio);
516 spin_unlock_irqrestore(&md->deferred_lock, flags);
517 queue_work(md->wq, &md->work);
521 * Everyone (including functions in this file), should use this
522 * function to access the md->map field, and make sure they call
523 * dm_put_live_table() when finished.
525 struct dm_table *dm_get_live_table(struct mapped_device *md, int *srcu_idx) __acquires(md->io_barrier)
527 *srcu_idx = srcu_read_lock(&md->io_barrier);
529 return srcu_dereference(md->map, &md->io_barrier);
532 void dm_put_live_table(struct mapped_device *md, int srcu_idx) __releases(md->io_barrier)
534 srcu_read_unlock(&md->io_barrier, srcu_idx);
537 void dm_sync_table(struct mapped_device *md)
539 synchronize_srcu(&md->io_barrier);
540 synchronize_rcu_expedited();
544 * A fast alternative to dm_get_live_table/dm_put_live_table.
545 * The caller must not block between these two functions.
547 static struct dm_table *dm_get_live_table_fast(struct mapped_device *md) __acquires(RCU)
550 return rcu_dereference(md->map);
553 static void dm_put_live_table_fast(struct mapped_device *md) __releases(RCU)
559 * Get the geometry associated with a dm device
561 int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo)
569 * Set the geometry of a device.
571 int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo)
573 sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors;
575 if (geo->start > sz) {
576 DMWARN("Start sector is beyond the geometry limits.");
585 /*-----------------------------------------------------------------
587 * A more elegant soln is in the works that uses the queue
588 * merge fn, unfortunately there are a couple of changes to
589 * the block layer that I want to make for this. So in the
590 * interests of getting something for people to use I give
591 * you this clearly demarcated crap.
592 *---------------------------------------------------------------*/
594 static int __noflush_suspending(struct mapped_device *md)
596 return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
600 * Decrements the number of outstanding ios that a bio has been
601 * cloned into, completing the original io if necc.
603 static void dec_pending(struct dm_io *io, int error)
608 struct mapped_device *md = io->md;
610 /* Push-back supersedes any I/O errors */
611 if (unlikely(error)) {
612 spin_lock_irqsave(&io->endio_lock, flags);
613 if (!(io->error > 0 && __noflush_suspending(md)))
615 spin_unlock_irqrestore(&io->endio_lock, flags);
618 if (atomic_dec_and_test(&io->io_count)) {
619 if (io->error == DM_ENDIO_REQUEUE) {
621 * Target requested pushing back the I/O.
623 spin_lock_irqsave(&md->deferred_lock, flags);
624 if (__noflush_suspending(md))
625 bio_list_add_head(&md->deferred, io->bio);
627 /* noflush suspend was interrupted. */
629 spin_unlock_irqrestore(&md->deferred_lock, flags);
632 io_error = io->error;
637 if (io_error == DM_ENDIO_REQUEUE)
640 if ((bio->bi_rw & REQ_FLUSH) && bio->bi_size) {
642 * Preflush done for flush with data, reissue
645 bio->bi_rw &= ~REQ_FLUSH;
648 /* done with normal IO or empty flush */
649 trace_block_bio_complete(md->queue, bio, io_error);
650 bio_endio(bio, io_error);
655 static void clone_endio(struct bio *bio, int error)
658 struct dm_target_io *tio = bio->bi_private;
659 struct dm_io *io = tio->io;
660 struct mapped_device *md = tio->io->md;
661 dm_endio_fn endio = tio->ti->type->end_io;
663 if (!bio_flagged(bio, BIO_UPTODATE) && !error)
667 r = endio(tio->ti, bio, error);
668 if (r < 0 || r == DM_ENDIO_REQUEUE)
670 * error and requeue request are handled
674 else if (r == DM_ENDIO_INCOMPLETE)
675 /* The target will handle the io */
678 DMWARN("unimplemented target endio return value: %d", r);
684 dec_pending(io, error);
688 * Partial completion handling for request-based dm
690 static void end_clone_bio(struct bio *clone, int error)
692 struct dm_rq_clone_bio_info *info = clone->bi_private;
693 struct dm_rq_target_io *tio = info->tio;
694 struct bio *bio = info->orig;
695 unsigned int nr_bytes = info->orig->bi_size;
701 * An error has already been detected on the request.
702 * Once error occurred, just let clone->end_io() handle
708 * Don't notice the error to the upper layer yet.
709 * The error handling decision is made by the target driver,
710 * when the request is completed.
717 * I/O for the bio successfully completed.
718 * Notice the data completion to the upper layer.
722 * bios are processed from the head of the list.
723 * So the completing bio should always be rq->bio.
724 * If it's not, something wrong is happening.
726 if (tio->orig->bio != bio)
727 DMERR("bio completion is going in the middle of the request");
730 * Update the original request.
731 * Do not use blk_end_request() here, because it may complete
732 * the original request before the clone, and break the ordering.
734 blk_update_request(tio->orig, 0, nr_bytes);
738 * Don't touch any member of the md after calling this function because
739 * the md may be freed in dm_put() at the end of this function.
740 * Or do dm_get() before calling this function and dm_put() later.
742 static void rq_completed(struct mapped_device *md, int rw, int run_queue)
744 atomic_dec(&md->pending[rw]);
746 /* nudge anyone waiting on suspend queue */
747 if (!md_in_flight(md))
751 * Run this off this callpath, as drivers could invoke end_io while
752 * inside their request_fn (and holding the queue lock). Calling
753 * back into ->request_fn() could deadlock attempting to grab the
757 blk_run_queue_async(md->queue);
760 * dm_put() must be at the end of this function. See the comment above
765 static void free_rq_clone(struct request *clone)
767 struct dm_rq_target_io *tio = clone->end_io_data;
769 blk_rq_unprep_clone(clone);
774 * Complete the clone and the original request.
775 * Must be called without queue lock.
777 static void dm_end_request(struct request *clone, int error)
779 int rw = rq_data_dir(clone);
780 struct dm_rq_target_io *tio = clone->end_io_data;
781 struct mapped_device *md = tio->md;
782 struct request *rq = tio->orig;
784 if (rq->cmd_type == REQ_TYPE_BLOCK_PC) {
785 rq->errors = clone->errors;
786 rq->resid_len = clone->resid_len;
790 * We are using the sense buffer of the original
792 * So setting the length of the sense data is enough.
794 rq->sense_len = clone->sense_len;
797 free_rq_clone(clone);
798 blk_end_request_all(rq, error);
799 rq_completed(md, rw, true);
802 static void dm_unprep_request(struct request *rq)
804 struct request *clone = rq->special;
807 rq->cmd_flags &= ~REQ_DONTPREP;
809 free_rq_clone(clone);
813 * Requeue the original request of a clone.
815 void dm_requeue_unmapped_request(struct request *clone)
817 int rw = rq_data_dir(clone);
818 struct dm_rq_target_io *tio = clone->end_io_data;
819 struct mapped_device *md = tio->md;
820 struct request *rq = tio->orig;
821 struct request_queue *q = rq->q;
824 dm_unprep_request(rq);
826 spin_lock_irqsave(q->queue_lock, flags);
827 blk_requeue_request(q, rq);
828 spin_unlock_irqrestore(q->queue_lock, flags);
830 rq_completed(md, rw, 0);
832 EXPORT_SYMBOL_GPL(dm_requeue_unmapped_request);
834 static void __stop_queue(struct request_queue *q)
839 static void stop_queue(struct request_queue *q)
843 spin_lock_irqsave(q->queue_lock, flags);
845 spin_unlock_irqrestore(q->queue_lock, flags);
848 static void __start_queue(struct request_queue *q)
850 if (blk_queue_stopped(q))
854 static void start_queue(struct request_queue *q)
858 spin_lock_irqsave(q->queue_lock, flags);
860 spin_unlock_irqrestore(q->queue_lock, flags);
863 static void dm_done(struct request *clone, int error, bool mapped)
866 struct dm_rq_target_io *tio = clone->end_io_data;
867 dm_request_endio_fn rq_end_io = NULL;
870 rq_end_io = tio->ti->type->rq_end_io;
872 if (mapped && rq_end_io)
873 r = rq_end_io(tio->ti, clone, error, &tio->info);
877 /* The target wants to complete the I/O */
878 dm_end_request(clone, r);
879 else if (r == DM_ENDIO_INCOMPLETE)
880 /* The target will handle the I/O */
882 else if (r == DM_ENDIO_REQUEUE)
883 /* The target wants to requeue the I/O */
884 dm_requeue_unmapped_request(clone);
886 DMWARN("unimplemented target endio return value: %d", r);
892 * Request completion handler for request-based dm
894 static void dm_softirq_done(struct request *rq)
897 struct request *clone = rq->completion_data;
898 struct dm_rq_target_io *tio = clone->end_io_data;
900 if (rq->cmd_flags & REQ_FAILED)
903 dm_done(clone, tio->error, mapped);
907 * Complete the clone and the original request with the error status
908 * through softirq context.
910 static void dm_complete_request(struct request *clone, int error)
912 struct dm_rq_target_io *tio = clone->end_io_data;
913 struct request *rq = tio->orig;
916 rq->completion_data = clone;
917 blk_complete_request(rq);
921 * Complete the not-mapped clone and the original request with the error status
922 * through softirq context.
923 * Target's rq_end_io() function isn't called.
924 * This may be used when the target's map_rq() function fails.
926 void dm_kill_unmapped_request(struct request *clone, int error)
928 struct dm_rq_target_io *tio = clone->end_io_data;
929 struct request *rq = tio->orig;
931 rq->cmd_flags |= REQ_FAILED;
932 dm_complete_request(clone, error);
934 EXPORT_SYMBOL_GPL(dm_kill_unmapped_request);
937 * Called with the queue lock held
939 static void end_clone_request(struct request *clone, int error)
942 * For just cleaning up the information of the queue in which
943 * the clone was dispatched.
944 * The clone is *NOT* freed actually here because it is alloced from
945 * dm own mempool and REQ_ALLOCED isn't set in clone->cmd_flags.
947 __blk_put_request(clone->q, clone);
950 * Actual request completion is done in a softirq context which doesn't
951 * hold the queue lock. Otherwise, deadlock could occur because:
952 * - another request may be submitted by the upper level driver
953 * of the stacking during the completion
954 * - the submission which requires queue lock may be done
957 dm_complete_request(clone, error);
961 * Return maximum size of I/O possible at the supplied sector up to the current
964 static sector_t max_io_len_target_boundary(sector_t sector, struct dm_target *ti)
966 sector_t target_offset = dm_target_offset(ti, sector);
968 return ti->len - target_offset;
971 static sector_t max_io_len(sector_t sector, struct dm_target *ti)
973 sector_t len = max_io_len_target_boundary(sector, ti);
974 sector_t offset, max_len;
977 * Does the target need to split even further?
979 if (ti->max_io_len) {
980 offset = dm_target_offset(ti, sector);
981 if (unlikely(ti->max_io_len & (ti->max_io_len - 1)))
982 max_len = sector_div(offset, ti->max_io_len);
984 max_len = offset & (ti->max_io_len - 1);
985 max_len = ti->max_io_len - max_len;
994 int dm_set_target_max_io_len(struct dm_target *ti, sector_t len)
996 if (len > UINT_MAX) {
997 DMERR("Specified maximum size of target IO (%llu) exceeds limit (%u)",
998 (unsigned long long)len, UINT_MAX);
999 ti->error = "Maximum size of target IO is too large";
1003 ti->max_io_len = (uint32_t) len;
1007 EXPORT_SYMBOL_GPL(dm_set_target_max_io_len);
1009 static void __map_bio(struct dm_target_io *tio)
1013 struct mapped_device *md;
1014 struct bio *clone = &tio->clone;
1015 struct dm_target *ti = tio->ti;
1017 clone->bi_end_io = clone_endio;
1018 clone->bi_private = tio;
1021 * Map the clone. If r == 0 we don't need to do
1022 * anything, the target has assumed ownership of
1025 atomic_inc(&tio->io->io_count);
1026 sector = clone->bi_sector;
1027 r = ti->type->map(ti, clone);
1028 if (r == DM_MAPIO_REMAPPED) {
1029 /* the bio has been remapped so dispatch it */
1031 trace_block_bio_remap(bdev_get_queue(clone->bi_bdev), clone,
1032 tio->io->bio->bi_bdev->bd_dev, sector);
1034 generic_make_request(clone);
1035 } else if (r < 0 || r == DM_MAPIO_REQUEUE) {
1036 /* error the io and bail out, or requeue it if needed */
1038 dec_pending(tio->io, r);
1041 DMWARN("unimplemented target map return value: %d", r);
1047 struct mapped_device *md;
1048 struct dm_table *map;
1052 sector_t sector_count;
1056 static void bio_setup_sector(struct bio *bio, sector_t sector, sector_t len)
1058 bio->bi_sector = sector;
1059 bio->bi_size = to_bytes(len);
1062 static void bio_setup_bv(struct bio *bio, unsigned short idx, unsigned short bv_count)
1065 bio->bi_vcnt = idx + bv_count;
1066 bio->bi_flags &= ~(1 << BIO_SEG_VALID);
1069 static void clone_bio_integrity(struct bio *bio, struct bio *clone,
1070 unsigned short idx, unsigned len, unsigned offset,
1073 if (!bio_integrity(bio))
1076 bio_integrity_clone(clone, bio, GFP_NOIO);
1079 bio_integrity_trim(clone, bio_sector_offset(bio, idx, offset), len);
1083 * Creates a little bio that just does part of a bvec.
1085 static void clone_split_bio(struct dm_target_io *tio, struct bio *bio,
1086 sector_t sector, unsigned short idx,
1087 unsigned offset, unsigned len)
1089 struct bio *clone = &tio->clone;
1090 struct bio_vec *bv = bio->bi_io_vec + idx;
1092 *clone->bi_io_vec = *bv;
1094 bio_setup_sector(clone, sector, len);
1096 clone->bi_bdev = bio->bi_bdev;
1097 clone->bi_rw = bio->bi_rw;
1099 clone->bi_io_vec->bv_offset = offset;
1100 clone->bi_io_vec->bv_len = clone->bi_size;
1101 clone->bi_flags |= 1 << BIO_CLONED;
1103 clone_bio_integrity(bio, clone, idx, len, offset, 1);
1107 * Creates a bio that consists of range of complete bvecs.
1109 static void clone_bio(struct dm_target_io *tio, struct bio *bio,
1110 sector_t sector, unsigned short idx,
1111 unsigned short bv_count, unsigned len)
1113 struct bio *clone = &tio->clone;
1116 __bio_clone(clone, bio);
1117 bio_setup_sector(clone, sector, len);
1118 bio_setup_bv(clone, idx, bv_count);
1120 if (idx != bio->bi_idx || clone->bi_size < bio->bi_size)
1122 clone_bio_integrity(bio, clone, idx, len, 0, trim);
1125 static struct dm_target_io *alloc_tio(struct clone_info *ci,
1126 struct dm_target *ti, int nr_iovecs,
1127 unsigned target_bio_nr)
1129 struct dm_target_io *tio;
1132 clone = bio_alloc_bioset(GFP_NOIO, nr_iovecs, ci->md->bs);
1133 tio = container_of(clone, struct dm_target_io, clone);
1137 memset(&tio->info, 0, sizeof(tio->info));
1138 tio->target_bio_nr = target_bio_nr;
1143 static void __clone_and_map_simple_bio(struct clone_info *ci,
1144 struct dm_target *ti,
1145 unsigned target_bio_nr, sector_t len)
1147 struct dm_target_io *tio = alloc_tio(ci, ti, ci->bio->bi_max_vecs, target_bio_nr);
1148 struct bio *clone = &tio->clone;
1151 * Discard requests require the bio's inline iovecs be initialized.
1152 * ci->bio->bi_max_vecs is BIO_INLINE_VECS anyway, for both flush
1153 * and discard, so no need for concern about wasted bvec allocations.
1155 __bio_clone(clone, ci->bio);
1157 bio_setup_sector(clone, ci->sector, len);
1162 static void __send_duplicate_bios(struct clone_info *ci, struct dm_target *ti,
1163 unsigned num_bios, sector_t len)
1165 unsigned target_bio_nr;
1167 for (target_bio_nr = 0; target_bio_nr < num_bios; target_bio_nr++)
1168 __clone_and_map_simple_bio(ci, ti, target_bio_nr, len);
1171 static int __send_empty_flush(struct clone_info *ci)
1173 unsigned target_nr = 0;
1174 struct dm_target *ti;
1176 BUG_ON(bio_has_data(ci->bio));
1177 while ((ti = dm_table_get_target(ci->map, target_nr++)))
1178 __send_duplicate_bios(ci, ti, ti->num_flush_bios, 0);
1183 static void __clone_and_map_data_bio(struct clone_info *ci, struct dm_target *ti,
1184 sector_t sector, int nr_iovecs,
1185 unsigned short idx, unsigned short bv_count,
1186 unsigned offset, unsigned len,
1187 unsigned split_bvec)
1189 struct bio *bio = ci->bio;
1190 struct dm_target_io *tio;
1191 unsigned target_bio_nr;
1192 unsigned num_target_bios = 1;
1195 * Does the target want to receive duplicate copies of the bio?
1197 if (bio_data_dir(bio) == WRITE && ti->num_write_bios)
1198 num_target_bios = ti->num_write_bios(ti, bio);
1200 for (target_bio_nr = 0; target_bio_nr < num_target_bios; target_bio_nr++) {
1201 tio = alloc_tio(ci, ti, nr_iovecs, target_bio_nr);
1203 clone_split_bio(tio, bio, sector, idx, offset, len);
1205 clone_bio(tio, bio, sector, idx, bv_count, len);
1210 typedef unsigned (*get_num_bios_fn)(struct dm_target *ti);
1212 static unsigned get_num_discard_bios(struct dm_target *ti)
1214 return ti->num_discard_bios;
1217 static unsigned get_num_write_same_bios(struct dm_target *ti)
1219 return ti->num_write_same_bios;
1222 typedef bool (*is_split_required_fn)(struct dm_target *ti);
1224 static bool is_split_required_for_discard(struct dm_target *ti)
1226 return ti->split_discard_bios;
1229 static int __send_changing_extent_only(struct clone_info *ci,
1230 get_num_bios_fn get_num_bios,
1231 is_split_required_fn is_split_required)
1233 struct dm_target *ti;
1238 ti = dm_table_find_target(ci->map, ci->sector);
1239 if (!dm_target_is_valid(ti))
1243 * Even though the device advertised support for this type of
1244 * request, that does not mean every target supports it, and
1245 * reconfiguration might also have changed that since the
1246 * check was performed.
1248 num_bios = get_num_bios ? get_num_bios(ti) : 0;
1252 if (is_split_required && !is_split_required(ti))
1253 len = min(ci->sector_count, max_io_len_target_boundary(ci->sector, ti));
1255 len = min(ci->sector_count, max_io_len(ci->sector, ti));
1257 __send_duplicate_bios(ci, ti, num_bios, len);
1260 } while (ci->sector_count -= len);
1265 static int __send_discard(struct clone_info *ci)
1267 return __send_changing_extent_only(ci, get_num_discard_bios,
1268 is_split_required_for_discard);
1271 static int __send_write_same(struct clone_info *ci)
1273 return __send_changing_extent_only(ci, get_num_write_same_bios, NULL);
1277 * Find maximum number of sectors / bvecs we can process with a single bio.
1279 static sector_t __len_within_target(struct clone_info *ci, sector_t max, int *idx)
1281 struct bio *bio = ci->bio;
1282 sector_t bv_len, total_len = 0;
1284 for (*idx = ci->idx; max && (*idx < bio->bi_vcnt); (*idx)++) {
1285 bv_len = to_sector(bio->bi_io_vec[*idx].bv_len);
1291 total_len += bv_len;
1297 static int __split_bvec_across_targets(struct clone_info *ci,
1298 struct dm_target *ti, sector_t max)
1300 struct bio *bio = ci->bio;
1301 struct bio_vec *bv = bio->bi_io_vec + ci->idx;
1302 sector_t remaining = to_sector(bv->bv_len);
1303 unsigned offset = 0;
1308 ti = dm_table_find_target(ci->map, ci->sector);
1309 if (!dm_target_is_valid(ti))
1312 max = max_io_len(ci->sector, ti);
1315 len = min(remaining, max);
1317 __clone_and_map_data_bio(ci, ti, ci->sector, 1, ci->idx, 0,
1318 bv->bv_offset + offset, len, 1);
1321 ci->sector_count -= len;
1322 offset += to_bytes(len);
1323 } while (remaining -= len);
1331 * Select the correct strategy for processing a non-flush bio.
1333 static int __split_and_process_non_flush(struct clone_info *ci)
1335 struct bio *bio = ci->bio;
1336 struct dm_target *ti;
1340 if (unlikely(bio->bi_rw & REQ_DISCARD))
1341 return __send_discard(ci);
1342 else if (unlikely(bio->bi_rw & REQ_WRITE_SAME))
1343 return __send_write_same(ci);
1345 ti = dm_table_find_target(ci->map, ci->sector);
1346 if (!dm_target_is_valid(ti))
1349 max = max_io_len(ci->sector, ti);
1352 * Optimise for the simple case where we can do all of
1353 * the remaining io with a single clone.
1355 if (ci->sector_count <= max) {
1356 __clone_and_map_data_bio(ci, ti, ci->sector, bio->bi_max_vecs,
1357 ci->idx, bio->bi_vcnt - ci->idx, 0,
1358 ci->sector_count, 0);
1359 ci->sector_count = 0;
1364 * There are some bvecs that don't span targets.
1365 * Do as many of these as possible.
1367 if (to_sector(bio->bi_io_vec[ci->idx].bv_len) <= max) {
1368 len = __len_within_target(ci, max, &idx);
1370 __clone_and_map_data_bio(ci, ti, ci->sector, bio->bi_max_vecs,
1371 ci->idx, idx - ci->idx, 0, len, 0);
1374 ci->sector_count -= len;
1381 * Handle a bvec that must be split between two or more targets.
1383 return __split_bvec_across_targets(ci, ti, max);
1387 * Entry point to split a bio into clones and submit them to the targets.
1389 static void __split_and_process_bio(struct mapped_device *md,
1390 struct dm_table *map, struct bio *bio)
1392 struct clone_info ci;
1395 if (unlikely(!map)) {
1402 ci.io = alloc_io(md);
1404 atomic_set(&ci.io->io_count, 1);
1407 spin_lock_init(&ci.io->endio_lock);
1408 ci.sector = bio->bi_sector;
1409 ci.idx = bio->bi_idx;
1411 start_io_acct(ci.io);
1413 if (bio->bi_rw & REQ_FLUSH) {
1414 ci.bio = &ci.md->flush_bio;
1415 ci.sector_count = 0;
1416 error = __send_empty_flush(&ci);
1417 /* dec_pending submits any data associated with flush */
1420 ci.sector_count = bio_sectors(bio);
1421 while (ci.sector_count && !error)
1422 error = __split_and_process_non_flush(&ci);
1425 /* drop the extra reference count */
1426 dec_pending(ci.io, error);
1428 /*-----------------------------------------------------------------
1430 *---------------------------------------------------------------*/
1432 static int dm_merge_bvec(struct request_queue *q,
1433 struct bvec_merge_data *bvm,
1434 struct bio_vec *biovec)
1436 struct mapped_device *md = q->queuedata;
1437 struct dm_table *map = dm_get_live_table_fast(md);
1438 struct dm_target *ti;
1439 sector_t max_sectors;
1445 ti = dm_table_find_target(map, bvm->bi_sector);
1446 if (!dm_target_is_valid(ti))
1450 * Find maximum amount of I/O that won't need splitting
1452 max_sectors = min(max_io_len(bvm->bi_sector, ti),
1453 (sector_t) BIO_MAX_SECTORS);
1454 max_size = (max_sectors << SECTOR_SHIFT) - bvm->bi_size;
1459 * merge_bvec_fn() returns number of bytes
1460 * it can accept at this offset
1461 * max is precomputed maximal io size
1463 if (max_size && ti->type->merge)
1464 max_size = ti->type->merge(ti, bvm, biovec, max_size);
1466 * If the target doesn't support merge method and some of the devices
1467 * provided their merge_bvec method (we know this by looking at
1468 * queue_max_hw_sectors), then we can't allow bios with multiple vector
1469 * entries. So always set max_size to 0, and the code below allows
1472 else if (queue_max_hw_sectors(q) <= PAGE_SIZE >> 9)
1477 dm_put_live_table_fast(md);
1479 * Always allow an entire first page
1481 if (max_size <= biovec->bv_len && !(bvm->bi_size >> SECTOR_SHIFT))
1482 max_size = biovec->bv_len;
1488 * The request function that just remaps the bio built up by
1491 static void _dm_request(struct request_queue *q, struct bio *bio)
1493 int rw = bio_data_dir(bio);
1494 struct mapped_device *md = q->queuedata;
1497 struct dm_table *map;
1499 map = dm_get_live_table(md, &srcu_idx);
1501 cpu = part_stat_lock();
1502 part_stat_inc(cpu, &dm_disk(md)->part0, ios[rw]);
1503 part_stat_add(cpu, &dm_disk(md)->part0, sectors[rw], bio_sectors(bio));
1506 /* if we're suspended, we have to queue this io for later */
1507 if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))) {
1508 dm_put_live_table(md, srcu_idx);
1510 if (bio_rw(bio) != READA)
1517 __split_and_process_bio(md, map, bio);
1518 dm_put_live_table(md, srcu_idx);
1522 static int dm_request_based(struct mapped_device *md)
1524 return blk_queue_stackable(md->queue);
1527 static void dm_request(struct request_queue *q, struct bio *bio)
1529 struct mapped_device *md = q->queuedata;
1531 if (dm_request_based(md))
1532 blk_queue_bio(q, bio);
1534 _dm_request(q, bio);
1537 void dm_dispatch_request(struct request *rq)
1541 if (blk_queue_io_stat(rq->q))
1542 rq->cmd_flags |= REQ_IO_STAT;
1544 rq->start_time = jiffies;
1545 r = blk_insert_cloned_request(rq->q, rq);
1547 dm_complete_request(rq, r);
1549 EXPORT_SYMBOL_GPL(dm_dispatch_request);
1551 static int dm_rq_bio_constructor(struct bio *bio, struct bio *bio_orig,
1554 struct dm_rq_target_io *tio = data;
1555 struct dm_rq_clone_bio_info *info =
1556 container_of(bio, struct dm_rq_clone_bio_info, clone);
1558 info->orig = bio_orig;
1560 bio->bi_end_io = end_clone_bio;
1561 bio->bi_private = info;
1566 static int setup_clone(struct request *clone, struct request *rq,
1567 struct dm_rq_target_io *tio)
1571 r = blk_rq_prep_clone(clone, rq, tio->md->bs, GFP_ATOMIC,
1572 dm_rq_bio_constructor, tio);
1576 clone->cmd = rq->cmd;
1577 clone->cmd_len = rq->cmd_len;
1578 clone->sense = rq->sense;
1579 clone->buffer = rq->buffer;
1580 clone->end_io = end_clone_request;
1581 clone->end_io_data = tio;
1586 static struct request *clone_rq(struct request *rq, struct mapped_device *md,
1589 struct request *clone;
1590 struct dm_rq_target_io *tio;
1592 tio = alloc_rq_tio(md, gfp_mask);
1600 memset(&tio->info, 0, sizeof(tio->info));
1602 clone = &tio->clone;
1603 if (setup_clone(clone, rq, tio)) {
1613 * Called with the queue lock held.
1615 static int dm_prep_fn(struct request_queue *q, struct request *rq)
1617 struct mapped_device *md = q->queuedata;
1618 struct request *clone;
1620 if (unlikely(rq->special)) {
1621 DMWARN("Already has something in rq->special.");
1622 return BLKPREP_KILL;
1625 clone = clone_rq(rq, md, GFP_ATOMIC);
1627 return BLKPREP_DEFER;
1629 rq->special = clone;
1630 rq->cmd_flags |= REQ_DONTPREP;
1637 * 0 : the request has been processed (not requeued)
1638 * !0 : the request has been requeued
1640 static int map_request(struct dm_target *ti, struct request *clone,
1641 struct mapped_device *md)
1643 int r, requeued = 0;
1644 struct dm_rq_target_io *tio = clone->end_io_data;
1647 r = ti->type->map_rq(ti, clone, &tio->info);
1649 case DM_MAPIO_SUBMITTED:
1650 /* The target has taken the I/O to submit by itself later */
1652 case DM_MAPIO_REMAPPED:
1653 /* The target has remapped the I/O so dispatch it */
1654 trace_block_rq_remap(clone->q, clone, disk_devt(dm_disk(md)),
1655 blk_rq_pos(tio->orig));
1656 dm_dispatch_request(clone);
1658 case DM_MAPIO_REQUEUE:
1659 /* The target wants to requeue the I/O */
1660 dm_requeue_unmapped_request(clone);
1665 DMWARN("unimplemented target map return value: %d", r);
1669 /* The target wants to complete the I/O */
1670 dm_kill_unmapped_request(clone, r);
1677 static struct request *dm_start_request(struct mapped_device *md, struct request *orig)
1679 struct request *clone;
1681 blk_start_request(orig);
1682 clone = orig->special;
1683 atomic_inc(&md->pending[rq_data_dir(clone)]);
1686 * Hold the md reference here for the in-flight I/O.
1687 * We can't rely on the reference count by device opener,
1688 * because the device may be closed during the request completion
1689 * when all bios are completed.
1690 * See the comment in rq_completed() too.
1698 * q->request_fn for request-based dm.
1699 * Called with the queue lock held.
1701 static void dm_request_fn(struct request_queue *q)
1703 struct mapped_device *md = q->queuedata;
1705 struct dm_table *map = dm_get_live_table(md, &srcu_idx);
1706 struct dm_target *ti;
1707 struct request *rq, *clone;
1711 * For suspend, check blk_queue_stopped() and increment
1712 * ->pending within a single queue_lock not to increment the
1713 * number of in-flight I/Os after the queue is stopped in
1716 while (!blk_queue_stopped(q)) {
1717 rq = blk_peek_request(q);
1721 /* always use block 0 to find the target for flushes for now */
1723 if (!(rq->cmd_flags & REQ_FLUSH))
1724 pos = blk_rq_pos(rq);
1726 ti = dm_table_find_target(map, pos);
1727 if (!dm_target_is_valid(ti)) {
1729 * Must perform setup, that dm_done() requires,
1730 * before calling dm_kill_unmapped_request
1732 DMERR_LIMIT("request attempted access beyond the end of device");
1733 clone = dm_start_request(md, rq);
1734 dm_kill_unmapped_request(clone, -EIO);
1738 if (ti->type->busy && ti->type->busy(ti))
1741 clone = dm_start_request(md, rq);
1743 spin_unlock(q->queue_lock);
1744 if (map_request(ti, clone, md))
1747 BUG_ON(!irqs_disabled());
1748 spin_lock(q->queue_lock);
1754 BUG_ON(!irqs_disabled());
1755 spin_lock(q->queue_lock);
1758 blk_delay_queue(q, HZ / 10);
1760 dm_put_live_table(md, srcu_idx);
1763 int dm_underlying_device_busy(struct request_queue *q)
1765 return blk_lld_busy(q);
1767 EXPORT_SYMBOL_GPL(dm_underlying_device_busy);
1769 static int dm_lld_busy(struct request_queue *q)
1772 struct mapped_device *md = q->queuedata;
1773 struct dm_table *map = dm_get_live_table_fast(md);
1775 if (!map || test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))
1778 r = dm_table_any_busy_target(map);
1780 dm_put_live_table_fast(md);
1785 static int dm_any_congested(void *congested_data, int bdi_bits)
1788 struct mapped_device *md = congested_data;
1789 struct dm_table *map;
1791 if (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
1792 map = dm_get_live_table_fast(md);
1795 * Request-based dm cares about only own queue for
1796 * the query about congestion status of request_queue
1798 if (dm_request_based(md))
1799 r = md->queue->backing_dev_info.state &
1802 r = dm_table_any_congested(map, bdi_bits);
1804 dm_put_live_table_fast(md);
1810 /*-----------------------------------------------------------------
1811 * An IDR is used to keep track of allocated minor numbers.
1812 *---------------------------------------------------------------*/
1813 static void free_minor(int minor)
1815 spin_lock(&_minor_lock);
1816 idr_remove(&_minor_idr, minor);
1817 spin_unlock(&_minor_lock);
1821 * See if the device with a specific minor # is free.
1823 static int specific_minor(int minor)
1827 if (minor >= (1 << MINORBITS))
1830 idr_preload(GFP_KERNEL);
1831 spin_lock(&_minor_lock);
1833 r = idr_alloc(&_minor_idr, MINOR_ALLOCED, minor, minor + 1, GFP_NOWAIT);
1835 spin_unlock(&_minor_lock);
1838 return r == -ENOSPC ? -EBUSY : r;
1842 static int next_free_minor(int *minor)
1846 idr_preload(GFP_KERNEL);
1847 spin_lock(&_minor_lock);
1849 r = idr_alloc(&_minor_idr, MINOR_ALLOCED, 0, 1 << MINORBITS, GFP_NOWAIT);
1851 spin_unlock(&_minor_lock);
1859 static const struct block_device_operations dm_blk_dops;
1861 static void dm_wq_work(struct work_struct *work);
1863 static void dm_init_md_queue(struct mapped_device *md)
1866 * Request-based dm devices cannot be stacked on top of bio-based dm
1867 * devices. The type of this dm device has not been decided yet.
1868 * The type is decided at the first table loading time.
1869 * To prevent problematic device stacking, clear the queue flag
1870 * for request stacking support until then.
1872 * This queue is new, so no concurrency on the queue_flags.
1874 queue_flag_clear_unlocked(QUEUE_FLAG_STACKABLE, md->queue);
1876 md->queue->queuedata = md;
1877 md->queue->backing_dev_info.congested_fn = dm_any_congested;
1878 md->queue->backing_dev_info.congested_data = md;
1879 blk_queue_make_request(md->queue, dm_request);
1880 blk_queue_bounce_limit(md->queue, BLK_BOUNCE_ANY);
1881 blk_queue_merge_bvec(md->queue, dm_merge_bvec);
1885 * Allocate and initialise a blank device with a given minor.
1887 static struct mapped_device *alloc_dev(int minor)
1890 struct mapped_device *md = kzalloc(sizeof(*md), GFP_KERNEL);
1894 DMWARN("unable to allocate device, out of memory.");
1898 if (!try_module_get(THIS_MODULE))
1899 goto bad_module_get;
1901 /* get a minor number for the dev */
1902 if (minor == DM_ANY_MINOR)
1903 r = next_free_minor(&minor);
1905 r = specific_minor(minor);
1909 r = init_srcu_struct(&md->io_barrier);
1911 goto bad_io_barrier;
1913 md->type = DM_TYPE_NONE;
1914 mutex_init(&md->suspend_lock);
1915 mutex_init(&md->type_lock);
1916 spin_lock_init(&md->deferred_lock);
1917 atomic_set(&md->holders, 1);
1918 atomic_set(&md->open_count, 0);
1919 atomic_set(&md->event_nr, 0);
1920 atomic_set(&md->uevent_seq, 0);
1921 INIT_LIST_HEAD(&md->uevent_list);
1922 spin_lock_init(&md->uevent_lock);
1924 md->queue = blk_alloc_queue(GFP_KERNEL);
1928 dm_init_md_queue(md);
1930 md->disk = alloc_disk(1);
1934 atomic_set(&md->pending[0], 0);
1935 atomic_set(&md->pending[1], 0);
1936 init_waitqueue_head(&md->wait);
1937 INIT_WORK(&md->work, dm_wq_work);
1938 init_waitqueue_head(&md->eventq);
1940 md->disk->major = _major;
1941 md->disk->first_minor = minor;
1942 md->disk->fops = &dm_blk_dops;
1943 md->disk->queue = md->queue;
1944 md->disk->private_data = md;
1945 sprintf(md->disk->disk_name, "dm-%d", minor);
1947 format_dev_t(md->name, MKDEV(_major, minor));
1949 md->wq = alloc_workqueue("kdmflush", WQ_MEM_RECLAIM, 0);
1953 md->bdev = bdget_disk(md->disk, 0);
1957 bio_init(&md->flush_bio);
1958 md->flush_bio.bi_bdev = md->bdev;
1959 md->flush_bio.bi_rw = WRITE_FLUSH;
1961 /* Populate the mapping, nobody knows we exist yet */
1962 spin_lock(&_minor_lock);
1963 old_md = idr_replace(&_minor_idr, md, minor);
1964 spin_unlock(&_minor_lock);
1966 BUG_ON(old_md != MINOR_ALLOCED);
1971 destroy_workqueue(md->wq);
1973 del_gendisk(md->disk);
1976 blk_cleanup_queue(md->queue);
1978 cleanup_srcu_struct(&md->io_barrier);
1982 module_put(THIS_MODULE);
1988 static void unlock_fs(struct mapped_device *md);
1990 static void free_dev(struct mapped_device *md)
1992 int minor = MINOR(disk_devt(md->disk));
1996 destroy_workqueue(md->wq);
1998 mempool_destroy(md->io_pool);
2000 bioset_free(md->bs);
2001 blk_integrity_unregister(md->disk);
2002 del_gendisk(md->disk);
2003 cleanup_srcu_struct(&md->io_barrier);
2006 spin_lock(&_minor_lock);
2007 md->disk->private_data = NULL;
2008 spin_unlock(&_minor_lock);
2011 blk_cleanup_queue(md->queue);
2012 module_put(THIS_MODULE);
2016 static void __bind_mempools(struct mapped_device *md, struct dm_table *t)
2018 struct dm_md_mempools *p = dm_table_get_md_mempools(t);
2020 if (md->io_pool && md->bs) {
2021 /* The md already has necessary mempools. */
2022 if (dm_table_get_type(t) == DM_TYPE_BIO_BASED) {
2024 * Reload bioset because front_pad may have changed
2025 * because a different table was loaded.
2027 bioset_free(md->bs);
2030 } else if (dm_table_get_type(t) == DM_TYPE_REQUEST_BASED) {
2032 * There's no need to reload with request-based dm
2033 * because the size of front_pad doesn't change.
2034 * Note for future: If you are to reload bioset,
2035 * prep-ed requests in the queue may refer
2036 * to bio from the old bioset, so you must walk
2037 * through the queue to unprep.
2043 BUG_ON(!p || md->io_pool || md->bs);
2045 md->io_pool = p->io_pool;
2051 /* mempool bind completed, now no need any mempools in the table */
2052 dm_table_free_md_mempools(t);
2056 * Bind a table to the device.
2058 static void event_callback(void *context)
2060 unsigned long flags;
2062 struct mapped_device *md = (struct mapped_device *) context;
2064 spin_lock_irqsave(&md->uevent_lock, flags);
2065 list_splice_init(&md->uevent_list, &uevents);
2066 spin_unlock_irqrestore(&md->uevent_lock, flags);
2068 dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj);
2070 atomic_inc(&md->event_nr);
2071 wake_up(&md->eventq);
2075 * Protected by md->suspend_lock obtained by dm_swap_table().
2077 static void __set_size(struct mapped_device *md, sector_t size)
2079 set_capacity(md->disk, size);
2081 i_size_write(md->bdev->bd_inode, (loff_t)size << SECTOR_SHIFT);
2085 * Return 1 if the queue has a compulsory merge_bvec_fn function.
2087 * If this function returns 0, then the device is either a non-dm
2088 * device without a merge_bvec_fn, or it is a dm device that is
2089 * able to split any bios it receives that are too big.
2091 int dm_queue_merge_is_compulsory(struct request_queue *q)
2093 struct mapped_device *dev_md;
2095 if (!q->merge_bvec_fn)
2098 if (q->make_request_fn == dm_request) {
2099 dev_md = q->queuedata;
2100 if (test_bit(DMF_MERGE_IS_OPTIONAL, &dev_md->flags))
2107 static int dm_device_merge_is_compulsory(struct dm_target *ti,
2108 struct dm_dev *dev, sector_t start,
2109 sector_t len, void *data)
2111 struct block_device *bdev = dev->bdev;
2112 struct request_queue *q = bdev_get_queue(bdev);
2114 return dm_queue_merge_is_compulsory(q);
2118 * Return 1 if it is acceptable to ignore merge_bvec_fn based
2119 * on the properties of the underlying devices.
2121 static int dm_table_merge_is_optional(struct dm_table *table)
2124 struct dm_target *ti;
2126 while (i < dm_table_get_num_targets(table)) {
2127 ti = dm_table_get_target(table, i++);
2129 if (ti->type->iterate_devices &&
2130 ti->type->iterate_devices(ti, dm_device_merge_is_compulsory, NULL))
2138 * Returns old map, which caller must destroy.
2140 static struct dm_table *__bind(struct mapped_device *md, struct dm_table *t,
2141 struct queue_limits *limits)
2143 struct dm_table *old_map;
2144 struct request_queue *q = md->queue;
2146 int merge_is_optional;
2148 size = dm_table_get_size(t);
2151 * Wipe any geometry if the size of the table changed.
2153 if (size != get_capacity(md->disk))
2154 memset(&md->geometry, 0, sizeof(md->geometry));
2156 __set_size(md, size);
2158 dm_table_event_callback(t, event_callback, md);
2161 * The queue hasn't been stopped yet, if the old table type wasn't
2162 * for request-based during suspension. So stop it to prevent
2163 * I/O mapping before resume.
2164 * This must be done before setting the queue restrictions,
2165 * because request-based dm may be run just after the setting.
2167 if (dm_table_request_based(t) && !blk_queue_stopped(q))
2170 __bind_mempools(md, t);
2172 merge_is_optional = dm_table_merge_is_optional(t);
2175 rcu_assign_pointer(md->map, t);
2176 md->immutable_target_type = dm_table_get_immutable_target_type(t);
2178 dm_table_set_restrictions(t, q, limits);
2179 if (merge_is_optional)
2180 set_bit(DMF_MERGE_IS_OPTIONAL, &md->flags);
2182 clear_bit(DMF_MERGE_IS_OPTIONAL, &md->flags);
2189 * Returns unbound table for the caller to free.
2191 static struct dm_table *__unbind(struct mapped_device *md)
2193 struct dm_table *map = md->map;
2198 dm_table_event_callback(map, NULL, NULL);
2199 rcu_assign_pointer(md->map, NULL);
2206 * Constructor for a new device.
2208 int dm_create(int minor, struct mapped_device **result)
2210 struct mapped_device *md;
2212 md = alloc_dev(minor);
2223 * Functions to manage md->type.
2224 * All are required to hold md->type_lock.
2226 void dm_lock_md_type(struct mapped_device *md)
2228 mutex_lock(&md->type_lock);
2231 void dm_unlock_md_type(struct mapped_device *md)
2233 mutex_unlock(&md->type_lock);
2236 void dm_set_md_type(struct mapped_device *md, unsigned type)
2241 unsigned dm_get_md_type(struct mapped_device *md)
2246 struct target_type *dm_get_immutable_target_type(struct mapped_device *md)
2248 return md->immutable_target_type;
2252 * Fully initialize a request-based queue (->elevator, ->request_fn, etc).
2254 static int dm_init_request_based_queue(struct mapped_device *md)
2256 struct request_queue *q = NULL;
2258 if (md->queue->elevator)
2261 /* Fully initialize the queue */
2262 q = blk_init_allocated_queue(md->queue, dm_request_fn, NULL);
2267 dm_init_md_queue(md);
2268 blk_queue_softirq_done(md->queue, dm_softirq_done);
2269 blk_queue_prep_rq(md->queue, dm_prep_fn);
2270 blk_queue_lld_busy(md->queue, dm_lld_busy);
2272 elv_register_queue(md->queue);
2278 * Setup the DM device's queue based on md's type
2280 int dm_setup_md_queue(struct mapped_device *md)
2282 if ((dm_get_md_type(md) == DM_TYPE_REQUEST_BASED) &&
2283 !dm_init_request_based_queue(md)) {
2284 DMWARN("Cannot initialize queue for request-based mapped device");
2291 static struct mapped_device *dm_find_md(dev_t dev)
2293 struct mapped_device *md;
2294 unsigned minor = MINOR(dev);
2296 if (MAJOR(dev) != _major || minor >= (1 << MINORBITS))
2299 spin_lock(&_minor_lock);
2301 md = idr_find(&_minor_idr, minor);
2302 if (md && (md == MINOR_ALLOCED ||
2303 (MINOR(disk_devt(dm_disk(md))) != minor) ||
2304 dm_deleting_md(md) ||
2305 test_bit(DMF_FREEING, &md->flags))) {
2311 spin_unlock(&_minor_lock);
2316 struct mapped_device *dm_get_md(dev_t dev)
2318 struct mapped_device *md = dm_find_md(dev);
2325 EXPORT_SYMBOL_GPL(dm_get_md);
2327 void *dm_get_mdptr(struct mapped_device *md)
2329 return md->interface_ptr;
2332 void dm_set_mdptr(struct mapped_device *md, void *ptr)
2334 md->interface_ptr = ptr;
2337 void dm_get(struct mapped_device *md)
2339 atomic_inc(&md->holders);
2340 BUG_ON(test_bit(DMF_FREEING, &md->flags));
2343 const char *dm_device_name(struct mapped_device *md)
2347 EXPORT_SYMBOL_GPL(dm_device_name);
2349 static void __dm_destroy(struct mapped_device *md, bool wait)
2351 struct dm_table *map;
2356 spin_lock(&_minor_lock);
2357 map = dm_get_live_table(md, &srcu_idx);
2358 idr_replace(&_minor_idr, MINOR_ALLOCED, MINOR(disk_devt(dm_disk(md))));
2359 set_bit(DMF_FREEING, &md->flags);
2360 spin_unlock(&_minor_lock);
2362 if (!dm_suspended_md(md)) {
2363 dm_table_presuspend_targets(map);
2364 dm_table_postsuspend_targets(map);
2367 /* dm_put_live_table must be before msleep, otherwise deadlock is possible */
2368 dm_put_live_table(md, srcu_idx);
2371 * Rare, but there may be I/O requests still going to complete,
2372 * for example. Wait for all references to disappear.
2373 * No one should increment the reference count of the mapped_device,
2374 * after the mapped_device state becomes DMF_FREEING.
2377 while (atomic_read(&md->holders))
2379 else if (atomic_read(&md->holders))
2380 DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)",
2381 dm_device_name(md), atomic_read(&md->holders));
2384 dm_table_destroy(__unbind(md));
2388 void dm_destroy(struct mapped_device *md)
2390 __dm_destroy(md, true);
2393 void dm_destroy_immediate(struct mapped_device *md)
2395 __dm_destroy(md, false);
2398 void dm_put(struct mapped_device *md)
2400 atomic_dec(&md->holders);
2402 EXPORT_SYMBOL_GPL(dm_put);
2404 static int dm_wait_for_completion(struct mapped_device *md, int interruptible)
2407 DECLARE_WAITQUEUE(wait, current);
2409 add_wait_queue(&md->wait, &wait);
2412 set_current_state(interruptible);
2414 if (!md_in_flight(md))
2417 if (interruptible == TASK_INTERRUPTIBLE &&
2418 signal_pending(current)) {
2425 set_current_state(TASK_RUNNING);
2427 remove_wait_queue(&md->wait, &wait);
2433 * Process the deferred bios
2435 static void dm_wq_work(struct work_struct *work)
2437 struct mapped_device *md = container_of(work, struct mapped_device,
2441 struct dm_table *map;
2443 map = dm_get_live_table(md, &srcu_idx);
2445 while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
2446 spin_lock_irq(&md->deferred_lock);
2447 c = bio_list_pop(&md->deferred);
2448 spin_unlock_irq(&md->deferred_lock);
2453 if (dm_request_based(md))
2454 generic_make_request(c);
2456 __split_and_process_bio(md, map, c);
2459 dm_put_live_table(md, srcu_idx);
2462 static void dm_queue_flush(struct mapped_device *md)
2464 clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2465 smp_mb__after_clear_bit();
2466 queue_work(md->wq, &md->work);
2470 * Swap in a new table, returning the old one for the caller to destroy.
2472 struct dm_table *dm_swap_table(struct mapped_device *md, struct dm_table *table)
2474 struct dm_table *live_map = NULL, *map = ERR_PTR(-EINVAL);
2475 struct queue_limits limits;
2478 mutex_lock(&md->suspend_lock);
2480 /* device must be suspended */
2481 if (!dm_suspended_md(md))
2485 * If the new table has no data devices, retain the existing limits.
2486 * This helps multipath with queue_if_no_path if all paths disappear,
2487 * then new I/O is queued based on these limits, and then some paths
2490 if (dm_table_has_no_data_devices(table)) {
2491 live_map = dm_get_live_table_fast(md);
2493 limits = md->queue->limits;
2494 dm_put_live_table_fast(md);
2498 r = dm_calculate_queue_limits(table, &limits);
2505 map = __bind(md, table, &limits);
2508 mutex_unlock(&md->suspend_lock);
2513 * Functions to lock and unlock any filesystem running on the
2516 static int lock_fs(struct mapped_device *md)
2520 WARN_ON(md->frozen_sb);
2522 md->frozen_sb = freeze_bdev(md->bdev);
2523 if (IS_ERR(md->frozen_sb)) {
2524 r = PTR_ERR(md->frozen_sb);
2525 md->frozen_sb = NULL;
2529 set_bit(DMF_FROZEN, &md->flags);
2534 static void unlock_fs(struct mapped_device *md)
2536 if (!test_bit(DMF_FROZEN, &md->flags))
2539 thaw_bdev(md->bdev, md->frozen_sb);
2540 md->frozen_sb = NULL;
2541 clear_bit(DMF_FROZEN, &md->flags);
2545 * We need to be able to change a mapping table under a mounted
2546 * filesystem. For example we might want to move some data in
2547 * the background. Before the table can be swapped with
2548 * dm_bind_table, dm_suspend must be called to flush any in
2549 * flight bios and ensure that any further io gets deferred.
2552 * Suspend mechanism in request-based dm.
2554 * 1. Flush all I/Os by lock_fs() if needed.
2555 * 2. Stop dispatching any I/O by stopping the request_queue.
2556 * 3. Wait for all in-flight I/Os to be completed or requeued.
2558 * To abort suspend, start the request_queue.
2560 int dm_suspend(struct mapped_device *md, unsigned suspend_flags)
2562 struct dm_table *map = NULL;
2564 int do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG ? 1 : 0;
2565 int noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG ? 1 : 0;
2567 mutex_lock(&md->suspend_lock);
2569 if (dm_suspended_md(md)) {
2577 * DMF_NOFLUSH_SUSPENDING must be set before presuspend.
2578 * This flag is cleared before dm_suspend returns.
2581 set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2583 /* This does not get reverted if there's an error later. */
2584 dm_table_presuspend_targets(map);
2587 * Flush I/O to the device.
2588 * Any I/O submitted after lock_fs() may not be flushed.
2589 * noflush takes precedence over do_lockfs.
2590 * (lock_fs() flushes I/Os and waits for them to complete.)
2592 if (!noflush && do_lockfs) {
2599 * Here we must make sure that no processes are submitting requests
2600 * to target drivers i.e. no one may be executing
2601 * __split_and_process_bio. This is called from dm_request and
2604 * To get all processes out of __split_and_process_bio in dm_request,
2605 * we take the write lock. To prevent any process from reentering
2606 * __split_and_process_bio from dm_request and quiesce the thread
2607 * (dm_wq_work), we set BMF_BLOCK_IO_FOR_SUSPEND and call
2608 * flush_workqueue(md->wq).
2610 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2611 synchronize_srcu(&md->io_barrier);
2614 * Stop md->queue before flushing md->wq in case request-based
2615 * dm defers requests to md->wq from md->queue.
2617 if (dm_request_based(md))
2618 stop_queue(md->queue);
2620 flush_workqueue(md->wq);
2623 * At this point no more requests are entering target request routines.
2624 * We call dm_wait_for_completion to wait for all existing requests
2627 r = dm_wait_for_completion(md, TASK_INTERRUPTIBLE);
2630 clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2631 synchronize_srcu(&md->io_barrier);
2633 /* were we interrupted ? */
2637 if (dm_request_based(md))
2638 start_queue(md->queue);
2641 goto out_unlock; /* pushback list is already flushed, so skip flush */
2645 * If dm_wait_for_completion returned 0, the device is completely
2646 * quiescent now. There is no request-processing activity. All new
2647 * requests are being added to md->deferred list.
2650 set_bit(DMF_SUSPENDED, &md->flags);
2652 dm_table_postsuspend_targets(map);
2655 mutex_unlock(&md->suspend_lock);
2659 int dm_resume(struct mapped_device *md)
2662 struct dm_table *map = NULL;
2664 mutex_lock(&md->suspend_lock);
2665 if (!dm_suspended_md(md))
2669 if (!map || !dm_table_get_size(map))
2672 r = dm_table_resume_targets(map);
2679 * Flushing deferred I/Os must be done after targets are resumed
2680 * so that mapping of targets can work correctly.
2681 * Request-based dm is queueing the deferred I/Os in its request_queue.
2683 if (dm_request_based(md))
2684 start_queue(md->queue);
2688 clear_bit(DMF_SUSPENDED, &md->flags);
2692 mutex_unlock(&md->suspend_lock);
2697 /*-----------------------------------------------------------------
2698 * Event notification.
2699 *---------------------------------------------------------------*/
2700 int dm_kobject_uevent(struct mapped_device *md, enum kobject_action action,
2703 char udev_cookie[DM_COOKIE_LENGTH];
2704 char *envp[] = { udev_cookie, NULL };
2707 return kobject_uevent(&disk_to_dev(md->disk)->kobj, action);
2709 snprintf(udev_cookie, DM_COOKIE_LENGTH, "%s=%u",
2710 DM_COOKIE_ENV_VAR_NAME, cookie);
2711 return kobject_uevent_env(&disk_to_dev(md->disk)->kobj,
2716 uint32_t dm_next_uevent_seq(struct mapped_device *md)
2718 return atomic_add_return(1, &md->uevent_seq);
2721 uint32_t dm_get_event_nr(struct mapped_device *md)
2723 return atomic_read(&md->event_nr);
2726 int dm_wait_event(struct mapped_device *md, int event_nr)
2728 return wait_event_interruptible(md->eventq,
2729 (event_nr != atomic_read(&md->event_nr)));
2732 void dm_uevent_add(struct mapped_device *md, struct list_head *elist)
2734 unsigned long flags;
2736 spin_lock_irqsave(&md->uevent_lock, flags);
2737 list_add(elist, &md->uevent_list);
2738 spin_unlock_irqrestore(&md->uevent_lock, flags);
2742 * The gendisk is only valid as long as you have a reference
2745 struct gendisk *dm_disk(struct mapped_device *md)
2750 struct kobject *dm_kobject(struct mapped_device *md)
2756 * struct mapped_device should not be exported outside of dm.c
2757 * so use this check to verify that kobj is part of md structure
2759 struct mapped_device *dm_get_from_kobject(struct kobject *kobj)
2761 struct mapped_device *md;
2763 md = container_of(kobj, struct mapped_device, kobj);
2764 if (&md->kobj != kobj)
2767 if (test_bit(DMF_FREEING, &md->flags) ||
2775 int dm_suspended_md(struct mapped_device *md)
2777 return test_bit(DMF_SUSPENDED, &md->flags);
2780 int dm_suspended(struct dm_target *ti)
2782 return dm_suspended_md(dm_table_get_md(ti->table));
2784 EXPORT_SYMBOL_GPL(dm_suspended);
2786 int dm_noflush_suspending(struct dm_target *ti)
2788 return __noflush_suspending(dm_table_get_md(ti->table));
2790 EXPORT_SYMBOL_GPL(dm_noflush_suspending);
2792 struct dm_md_mempools *dm_alloc_md_mempools(unsigned type, unsigned integrity, unsigned per_bio_data_size)
2794 struct dm_md_mempools *pools = kzalloc(sizeof(*pools), GFP_KERNEL);
2795 struct kmem_cache *cachep;
2796 unsigned int pool_size;
2797 unsigned int front_pad;
2802 if (type == DM_TYPE_BIO_BASED) {
2805 front_pad = roundup(per_bio_data_size, __alignof__(struct dm_target_io)) + offsetof(struct dm_target_io, clone);
2806 } else if (type == DM_TYPE_REQUEST_BASED) {
2807 cachep = _rq_tio_cache;
2808 pool_size = MIN_IOS;
2809 front_pad = offsetof(struct dm_rq_clone_bio_info, clone);
2810 /* per_bio_data_size is not used. See __bind_mempools(). */
2811 WARN_ON(per_bio_data_size != 0);
2815 pools->io_pool = mempool_create_slab_pool(MIN_IOS, cachep);
2816 if (!pools->io_pool)
2819 pools->bs = bioset_create(pool_size, front_pad);
2823 if (integrity && bioset_integrity_create(pools->bs, pool_size))
2829 dm_free_md_mempools(pools);
2834 void dm_free_md_mempools(struct dm_md_mempools *pools)
2840 mempool_destroy(pools->io_pool);
2843 bioset_free(pools->bs);
2848 static const struct block_device_operations dm_blk_dops = {
2849 .open = dm_blk_open,
2850 .release = dm_blk_close,
2851 .ioctl = dm_blk_ioctl,
2852 .getgeo = dm_blk_getgeo,
2853 .owner = THIS_MODULE
2856 EXPORT_SYMBOL(dm_get_mapinfo);
2861 module_init(dm_init);
2862 module_exit(dm_exit);
2864 module_param(major, uint, 0);
2865 MODULE_PARM_DESC(major, "The major number of the device mapper");
2866 MODULE_DESCRIPTION(DM_NAME " driver");
2867 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
2868 MODULE_LICENSE("GPL");