2 * Copyright (C) 2001, 2002 Sistina Software (UK) Limited.
3 * Copyright (C) 2004-2008 Red Hat, Inc. All rights reserved.
5 * This file is released under the GPL.
11 #include <linux/init.h>
12 #include <linux/module.h>
13 #include <linux/mutex.h>
14 #include <linux/moduleparam.h>
15 #include <linux/blkpg.h>
16 #include <linux/bio.h>
17 #include <linux/mempool.h>
18 #include <linux/slab.h>
19 #include <linux/idr.h>
20 #include <linux/hdreg.h>
21 #include <linux/delay.h>
23 #include <trace/events/block.h>
25 #define DM_MSG_PREFIX "core"
29 * ratelimit state to be used in DMXXX_LIMIT().
31 DEFINE_RATELIMIT_STATE(dm_ratelimit_state,
32 DEFAULT_RATELIMIT_INTERVAL,
33 DEFAULT_RATELIMIT_BURST);
34 EXPORT_SYMBOL(dm_ratelimit_state);
38 * Cookies are numeric values sent with CHANGE and REMOVE
39 * uevents while resuming, removing or renaming the device.
41 #define DM_COOKIE_ENV_VAR_NAME "DM_COOKIE"
42 #define DM_COOKIE_LENGTH 24
44 static const char *_name = DM_NAME;
46 static unsigned int major = 0;
47 static unsigned int _major = 0;
49 static DEFINE_IDR(_minor_idr);
51 static DEFINE_SPINLOCK(_minor_lock);
54 * One of these is allocated per bio.
57 struct mapped_device *md;
61 unsigned long start_time;
62 spinlock_t endio_lock;
66 * For request-based dm.
67 * One of these is allocated per request.
69 struct dm_rq_target_io {
70 struct mapped_device *md;
72 struct request *orig, clone;
78 * For request-based dm - the bio clones we allocate are embedded in these
81 * We allocate these with bio_alloc_bioset, using the front_pad parameter when
82 * the bioset is created - this means the bio has to come at the end of the
85 struct dm_rq_clone_bio_info {
87 struct dm_rq_target_io *tio;
91 union map_info *dm_get_mapinfo(struct bio *bio)
93 if (bio && bio->bi_private)
94 return &((struct dm_target_io *)bio->bi_private)->info;
98 union map_info *dm_get_rq_mapinfo(struct request *rq)
100 if (rq && rq->end_io_data)
101 return &((struct dm_rq_target_io *)rq->end_io_data)->info;
104 EXPORT_SYMBOL_GPL(dm_get_rq_mapinfo);
106 #define MINOR_ALLOCED ((void *)-1)
109 * Bits for the md->flags field.
111 #define DMF_BLOCK_IO_FOR_SUSPEND 0
112 #define DMF_SUSPENDED 1
114 #define DMF_FREEING 3
115 #define DMF_DELETING 4
116 #define DMF_NOFLUSH_SUSPENDING 5
117 #define DMF_MERGE_IS_OPTIONAL 6
120 * Work processed by per-device workqueue.
122 struct mapped_device {
123 struct rw_semaphore io_lock;
124 struct mutex suspend_lock;
131 struct request_queue *queue;
133 /* Protect queue and type against concurrent access. */
134 struct mutex type_lock;
136 struct target_type *immutable_target_type;
138 struct gendisk *disk;
144 * A list of ios that arrived while we were suspended.
147 wait_queue_head_t wait;
148 struct work_struct work;
149 struct bio_list deferred;
150 spinlock_t deferred_lock;
153 * Processing queue (flush)
155 struct workqueue_struct *wq;
158 * The current mapping.
160 struct dm_table *map;
163 * io objects are allocated from here.
174 wait_queue_head_t eventq;
176 struct list_head uevent_list;
177 spinlock_t uevent_lock; /* Protect access to uevent_list */
180 * freeze/thaw support require holding onto a super block
182 struct super_block *frozen_sb;
183 struct block_device *bdev;
185 /* forced geometry settings */
186 struct hd_geometry geometry;
191 /* zero-length flush that will be cloned and submitted to targets */
192 struct bio flush_bio;
196 * For mempools pre-allocation at the table loading time.
198 struct dm_md_mempools {
205 static struct kmem_cache *_io_cache;
206 static struct kmem_cache *_rq_tio_cache;
209 * Unused now, and needs to be deleted. But since io_pool is overloaded and it's
210 * still used for _io_cache, I'm leaving this for a later cleanup
212 static struct kmem_cache *_rq_bio_info_cache;
214 static int __init local_init(void)
218 /* allocate a slab for the dm_ios */
219 _io_cache = KMEM_CACHE(dm_io, 0);
223 _rq_tio_cache = KMEM_CACHE(dm_rq_target_io, 0);
225 goto out_free_io_cache;
227 _rq_bio_info_cache = KMEM_CACHE(dm_rq_clone_bio_info, 0);
228 if (!_rq_bio_info_cache)
229 goto out_free_rq_tio_cache;
231 r = dm_uevent_init();
233 goto out_free_rq_bio_info_cache;
236 r = register_blkdev(_major, _name);
238 goto out_uevent_exit;
247 out_free_rq_bio_info_cache:
248 kmem_cache_destroy(_rq_bio_info_cache);
249 out_free_rq_tio_cache:
250 kmem_cache_destroy(_rq_tio_cache);
252 kmem_cache_destroy(_io_cache);
257 static void local_exit(void)
259 kmem_cache_destroy(_rq_bio_info_cache);
260 kmem_cache_destroy(_rq_tio_cache);
261 kmem_cache_destroy(_io_cache);
262 unregister_blkdev(_major, _name);
267 DMINFO("cleaned up");
270 static int (*_inits[])(void) __initdata = {
280 static void (*_exits[])(void) = {
290 static int __init dm_init(void)
292 const int count = ARRAY_SIZE(_inits);
296 for (i = 0; i < count; i++) {
311 static void __exit dm_exit(void)
313 int i = ARRAY_SIZE(_exits);
319 * Should be empty by this point.
321 idr_destroy(&_minor_idr);
325 * Block device functions
327 int dm_deleting_md(struct mapped_device *md)
329 return test_bit(DMF_DELETING, &md->flags);
332 static int dm_blk_open(struct block_device *bdev, fmode_t mode)
334 struct mapped_device *md;
336 spin_lock(&_minor_lock);
338 md = bdev->bd_disk->private_data;
342 if (test_bit(DMF_FREEING, &md->flags) ||
343 dm_deleting_md(md)) {
349 atomic_inc(&md->open_count);
352 spin_unlock(&_minor_lock);
354 return md ? 0 : -ENXIO;
357 static int dm_blk_close(struct gendisk *disk, fmode_t mode)
359 struct mapped_device *md = disk->private_data;
361 spin_lock(&_minor_lock);
363 atomic_dec(&md->open_count);
366 spin_unlock(&_minor_lock);
371 int dm_open_count(struct mapped_device *md)
373 return atomic_read(&md->open_count);
377 * Guarantees nothing is using the device before it's deleted.
379 int dm_lock_for_deletion(struct mapped_device *md)
383 spin_lock(&_minor_lock);
385 if (dm_open_count(md))
388 set_bit(DMF_DELETING, &md->flags);
390 spin_unlock(&_minor_lock);
395 static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo)
397 struct mapped_device *md = bdev->bd_disk->private_data;
399 return dm_get_geometry(md, geo);
402 static int dm_blk_ioctl(struct block_device *bdev, fmode_t mode,
403 unsigned int cmd, unsigned long arg)
405 struct mapped_device *md = bdev->bd_disk->private_data;
406 struct dm_table *map = dm_get_live_table(md);
407 struct dm_target *tgt;
410 if (!map || !dm_table_get_size(map))
413 /* We only support devices that have a single target */
414 if (dm_table_get_num_targets(map) != 1)
417 tgt = dm_table_get_target(map, 0);
419 if (dm_suspended_md(md)) {
424 if (tgt->type->ioctl)
425 r = tgt->type->ioctl(tgt, cmd, arg);
433 static struct dm_io *alloc_io(struct mapped_device *md)
435 return mempool_alloc(md->io_pool, GFP_NOIO);
438 static void free_io(struct mapped_device *md, struct dm_io *io)
440 mempool_free(io, md->io_pool);
443 static void free_tio(struct mapped_device *md, struct dm_target_io *tio)
445 bio_put(&tio->clone);
448 static struct dm_rq_target_io *alloc_rq_tio(struct mapped_device *md,
451 return mempool_alloc(md->tio_pool, gfp_mask);
454 static void free_rq_tio(struct dm_rq_target_io *tio)
456 mempool_free(tio, tio->md->tio_pool);
459 static int md_in_flight(struct mapped_device *md)
461 return atomic_read(&md->pending[READ]) +
462 atomic_read(&md->pending[WRITE]);
465 static void start_io_acct(struct dm_io *io)
467 struct mapped_device *md = io->md;
469 int rw = bio_data_dir(io->bio);
471 io->start_time = jiffies;
473 cpu = part_stat_lock();
474 part_round_stats(cpu, &dm_disk(md)->part0);
476 atomic_set(&dm_disk(md)->part0.in_flight[rw],
477 atomic_inc_return(&md->pending[rw]));
480 static void end_io_acct(struct dm_io *io)
482 struct mapped_device *md = io->md;
483 struct bio *bio = io->bio;
484 unsigned long duration = jiffies - io->start_time;
486 int rw = bio_data_dir(bio);
488 cpu = part_stat_lock();
489 part_round_stats(cpu, &dm_disk(md)->part0);
490 part_stat_add(cpu, &dm_disk(md)->part0, ticks[rw], duration);
494 * After this is decremented the bio must not be touched if it is
497 pending = atomic_dec_return(&md->pending[rw]);
498 atomic_set(&dm_disk(md)->part0.in_flight[rw], pending);
499 pending += atomic_read(&md->pending[rw^0x1]);
501 /* nudge anyone waiting on suspend queue */
507 * Add the bio to the list of deferred io.
509 static void queue_io(struct mapped_device *md, struct bio *bio)
513 spin_lock_irqsave(&md->deferred_lock, flags);
514 bio_list_add(&md->deferred, bio);
515 spin_unlock_irqrestore(&md->deferred_lock, flags);
516 queue_work(md->wq, &md->work);
520 * Everyone (including functions in this file), should use this
521 * function to access the md->map field, and make sure they call
522 * dm_table_put() when finished.
524 struct dm_table *dm_get_live_table(struct mapped_device *md)
529 read_lock_irqsave(&md->map_lock, flags);
533 read_unlock_irqrestore(&md->map_lock, flags);
539 * Get the geometry associated with a dm device
541 int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo)
549 * Set the geometry of a device.
551 int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo)
553 sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors;
555 if (geo->start > sz) {
556 DMWARN("Start sector is beyond the geometry limits.");
565 /*-----------------------------------------------------------------
567 * A more elegant soln is in the works that uses the queue
568 * merge fn, unfortunately there are a couple of changes to
569 * the block layer that I want to make for this. So in the
570 * interests of getting something for people to use I give
571 * you this clearly demarcated crap.
572 *---------------------------------------------------------------*/
574 static int __noflush_suspending(struct mapped_device *md)
576 return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
580 * Decrements the number of outstanding ios that a bio has been
581 * cloned into, completing the original io if necc.
583 static void dec_pending(struct dm_io *io, int error)
588 struct mapped_device *md = io->md;
590 /* Push-back supersedes any I/O errors */
591 if (unlikely(error)) {
592 spin_lock_irqsave(&io->endio_lock, flags);
593 if (!(io->error > 0 && __noflush_suspending(md)))
595 spin_unlock_irqrestore(&io->endio_lock, flags);
598 if (atomic_dec_and_test(&io->io_count)) {
599 if (io->error == DM_ENDIO_REQUEUE) {
601 * Target requested pushing back the I/O.
603 spin_lock_irqsave(&md->deferred_lock, flags);
604 if (__noflush_suspending(md))
605 bio_list_add_head(&md->deferred, io->bio);
607 /* noflush suspend was interrupted. */
609 spin_unlock_irqrestore(&md->deferred_lock, flags);
612 io_error = io->error;
617 if (io_error == DM_ENDIO_REQUEUE)
620 if ((bio->bi_rw & REQ_FLUSH) && bio->bi_size) {
622 * Preflush done for flush with data, reissue
625 bio->bi_rw &= ~REQ_FLUSH;
628 /* done with normal IO or empty flush */
629 bio_endio(bio, io_error);
634 static void clone_endio(struct bio *bio, int error)
637 struct dm_target_io *tio = bio->bi_private;
638 struct dm_io *io = tio->io;
639 struct mapped_device *md = tio->io->md;
640 dm_endio_fn endio = tio->ti->type->end_io;
642 if (!bio_flagged(bio, BIO_UPTODATE) && !error)
646 r = endio(tio->ti, bio, error);
647 if (r < 0 || r == DM_ENDIO_REQUEUE)
649 * error and requeue request are handled
653 else if (r == DM_ENDIO_INCOMPLETE)
654 /* The target will handle the io */
657 DMWARN("unimplemented target endio return value: %d", r);
663 dec_pending(io, error);
667 * Partial completion handling for request-based dm
669 static void end_clone_bio(struct bio *clone, int error)
671 struct dm_rq_clone_bio_info *info = clone->bi_private;
672 struct dm_rq_target_io *tio = info->tio;
673 struct bio *bio = info->orig;
674 unsigned int nr_bytes = info->orig->bi_size;
680 * An error has already been detected on the request.
681 * Once error occurred, just let clone->end_io() handle
687 * Don't notice the error to the upper layer yet.
688 * The error handling decision is made by the target driver,
689 * when the request is completed.
696 * I/O for the bio successfully completed.
697 * Notice the data completion to the upper layer.
701 * bios are processed from the head of the list.
702 * So the completing bio should always be rq->bio.
703 * If it's not, something wrong is happening.
705 if (tio->orig->bio != bio)
706 DMERR("bio completion is going in the middle of the request");
709 * Update the original request.
710 * Do not use blk_end_request() here, because it may complete
711 * the original request before the clone, and break the ordering.
713 blk_update_request(tio->orig, 0, nr_bytes);
717 * Don't touch any member of the md after calling this function because
718 * the md may be freed in dm_put() at the end of this function.
719 * Or do dm_get() before calling this function and dm_put() later.
721 static void rq_completed(struct mapped_device *md, int rw, int run_queue)
723 atomic_dec(&md->pending[rw]);
725 /* nudge anyone waiting on suspend queue */
726 if (!md_in_flight(md))
730 * Run this off this callpath, as drivers could invoke end_io while
731 * inside their request_fn (and holding the queue lock). Calling
732 * back into ->request_fn() could deadlock attempting to grab the
736 blk_run_queue_async(md->queue);
739 * dm_put() must be at the end of this function. See the comment above
744 static void free_rq_clone(struct request *clone)
746 struct dm_rq_target_io *tio = clone->end_io_data;
748 blk_rq_unprep_clone(clone);
753 * Complete the clone and the original request.
754 * Must be called without queue lock.
756 static void dm_end_request(struct request *clone, int error)
758 int rw = rq_data_dir(clone);
759 struct dm_rq_target_io *tio = clone->end_io_data;
760 struct mapped_device *md = tio->md;
761 struct request *rq = tio->orig;
763 if (rq->cmd_type == REQ_TYPE_BLOCK_PC) {
764 rq->errors = clone->errors;
765 rq->resid_len = clone->resid_len;
769 * We are using the sense buffer of the original
771 * So setting the length of the sense data is enough.
773 rq->sense_len = clone->sense_len;
776 free_rq_clone(clone);
777 blk_end_request_all(rq, error);
778 rq_completed(md, rw, true);
781 static void dm_unprep_request(struct request *rq)
783 struct request *clone = rq->special;
786 rq->cmd_flags &= ~REQ_DONTPREP;
788 free_rq_clone(clone);
792 * Requeue the original request of a clone.
794 void dm_requeue_unmapped_request(struct request *clone)
796 int rw = rq_data_dir(clone);
797 struct dm_rq_target_io *tio = clone->end_io_data;
798 struct mapped_device *md = tio->md;
799 struct request *rq = tio->orig;
800 struct request_queue *q = rq->q;
803 dm_unprep_request(rq);
805 spin_lock_irqsave(q->queue_lock, flags);
806 blk_requeue_request(q, rq);
807 spin_unlock_irqrestore(q->queue_lock, flags);
809 rq_completed(md, rw, 0);
811 EXPORT_SYMBOL_GPL(dm_requeue_unmapped_request);
813 static void __stop_queue(struct request_queue *q)
818 static void stop_queue(struct request_queue *q)
822 spin_lock_irqsave(q->queue_lock, flags);
824 spin_unlock_irqrestore(q->queue_lock, flags);
827 static void __start_queue(struct request_queue *q)
829 if (blk_queue_stopped(q))
833 static void start_queue(struct request_queue *q)
837 spin_lock_irqsave(q->queue_lock, flags);
839 spin_unlock_irqrestore(q->queue_lock, flags);
842 static void dm_done(struct request *clone, int error, bool mapped)
845 struct dm_rq_target_io *tio = clone->end_io_data;
846 dm_request_endio_fn rq_end_io = NULL;
849 rq_end_io = tio->ti->type->rq_end_io;
851 if (mapped && rq_end_io)
852 r = rq_end_io(tio->ti, clone, error, &tio->info);
856 /* The target wants to complete the I/O */
857 dm_end_request(clone, r);
858 else if (r == DM_ENDIO_INCOMPLETE)
859 /* The target will handle the I/O */
861 else if (r == DM_ENDIO_REQUEUE)
862 /* The target wants to requeue the I/O */
863 dm_requeue_unmapped_request(clone);
865 DMWARN("unimplemented target endio return value: %d", r);
871 * Request completion handler for request-based dm
873 static void dm_softirq_done(struct request *rq)
876 struct request *clone = rq->completion_data;
877 struct dm_rq_target_io *tio = clone->end_io_data;
879 if (rq->cmd_flags & REQ_FAILED)
882 dm_done(clone, tio->error, mapped);
886 * Complete the clone and the original request with the error status
887 * through softirq context.
889 static void dm_complete_request(struct request *clone, int error)
891 struct dm_rq_target_io *tio = clone->end_io_data;
892 struct request *rq = tio->orig;
895 rq->completion_data = clone;
896 blk_complete_request(rq);
900 * Complete the not-mapped clone and the original request with the error status
901 * through softirq context.
902 * Target's rq_end_io() function isn't called.
903 * This may be used when the target's map_rq() function fails.
905 void dm_kill_unmapped_request(struct request *clone, int error)
907 struct dm_rq_target_io *tio = clone->end_io_data;
908 struct request *rq = tio->orig;
910 rq->cmd_flags |= REQ_FAILED;
911 dm_complete_request(clone, error);
913 EXPORT_SYMBOL_GPL(dm_kill_unmapped_request);
916 * Called with the queue lock held
918 static void end_clone_request(struct request *clone, int error)
921 * For just cleaning up the information of the queue in which
922 * the clone was dispatched.
923 * The clone is *NOT* freed actually here because it is alloced from
924 * dm own mempool and REQ_ALLOCED isn't set in clone->cmd_flags.
926 __blk_put_request(clone->q, clone);
929 * Actual request completion is done in a softirq context which doesn't
930 * hold the queue lock. Otherwise, deadlock could occur because:
931 * - another request may be submitted by the upper level driver
932 * of the stacking during the completion
933 * - the submission which requires queue lock may be done
936 dm_complete_request(clone, error);
940 * Return maximum size of I/O possible at the supplied sector up to the current
943 static sector_t max_io_len_target_boundary(sector_t sector, struct dm_target *ti)
945 sector_t target_offset = dm_target_offset(ti, sector);
947 return ti->len - target_offset;
950 static sector_t max_io_len(sector_t sector, struct dm_target *ti)
952 sector_t len = max_io_len_target_boundary(sector, ti);
953 sector_t offset, max_len;
956 * Does the target need to split even further?
958 if (ti->max_io_len) {
959 offset = dm_target_offset(ti, sector);
960 if (unlikely(ti->max_io_len & (ti->max_io_len - 1)))
961 max_len = sector_div(offset, ti->max_io_len);
963 max_len = offset & (ti->max_io_len - 1);
964 max_len = ti->max_io_len - max_len;
973 int dm_set_target_max_io_len(struct dm_target *ti, sector_t len)
975 if (len > UINT_MAX) {
976 DMERR("Specified maximum size of target IO (%llu) exceeds limit (%u)",
977 (unsigned long long)len, UINT_MAX);
978 ti->error = "Maximum size of target IO is too large";
982 ti->max_io_len = (uint32_t) len;
986 EXPORT_SYMBOL_GPL(dm_set_target_max_io_len);
988 static void __map_bio(struct dm_target *ti, struct dm_target_io *tio)
992 struct mapped_device *md;
993 struct bio *clone = &tio->clone;
995 clone->bi_end_io = clone_endio;
996 clone->bi_private = tio;
999 * Map the clone. If r == 0 we don't need to do
1000 * anything, the target has assumed ownership of
1003 atomic_inc(&tio->io->io_count);
1004 sector = clone->bi_sector;
1005 r = ti->type->map(ti, clone);
1006 if (r == DM_MAPIO_REMAPPED) {
1007 /* the bio has been remapped so dispatch it */
1009 trace_block_bio_remap(bdev_get_queue(clone->bi_bdev), clone,
1010 tio->io->bio->bi_bdev->bd_dev, sector);
1012 generic_make_request(clone);
1013 } else if (r < 0 || r == DM_MAPIO_REQUEUE) {
1014 /* error the io and bail out, or requeue it if needed */
1016 dec_pending(tio->io, r);
1019 DMWARN("unimplemented target map return value: %d", r);
1025 struct mapped_device *md;
1026 struct dm_table *map;
1030 sector_t sector_count;
1035 * Creates a little bio that just does part of a bvec.
1037 static void split_bvec(struct dm_target_io *tio, struct bio *bio,
1038 sector_t sector, unsigned short idx, unsigned int offset,
1039 unsigned int len, struct bio_set *bs)
1041 struct bio *clone = &tio->clone;
1042 struct bio_vec *bv = bio->bi_io_vec + idx;
1044 *clone->bi_io_vec = *bv;
1046 clone->bi_sector = sector;
1047 clone->bi_bdev = bio->bi_bdev;
1048 clone->bi_rw = bio->bi_rw;
1050 clone->bi_size = to_bytes(len);
1051 clone->bi_io_vec->bv_offset = offset;
1052 clone->bi_io_vec->bv_len = clone->bi_size;
1053 clone->bi_flags |= 1 << BIO_CLONED;
1055 if (bio_integrity(bio)) {
1056 bio_integrity_clone(clone, bio, GFP_NOIO);
1057 bio_integrity_trim(clone,
1058 bio_sector_offset(bio, idx, offset), len);
1063 * Creates a bio that consists of range of complete bvecs.
1065 static void clone_bio(struct dm_target_io *tio, struct bio *bio,
1066 sector_t sector, unsigned short idx,
1067 unsigned short bv_count, unsigned int len,
1070 struct bio *clone = &tio->clone;
1072 __bio_clone(clone, bio);
1073 clone->bi_sector = sector;
1074 clone->bi_idx = idx;
1075 clone->bi_vcnt = idx + bv_count;
1076 clone->bi_size = to_bytes(len);
1077 clone->bi_flags &= ~(1 << BIO_SEG_VALID);
1079 if (bio_integrity(bio)) {
1080 bio_integrity_clone(clone, bio, GFP_NOIO);
1082 if (idx != bio->bi_idx || clone->bi_size < bio->bi_size)
1083 bio_integrity_trim(clone,
1084 bio_sector_offset(bio, idx, 0), len);
1088 static struct dm_target_io *alloc_tio(struct clone_info *ci,
1089 struct dm_target *ti, int nr_iovecs)
1091 struct dm_target_io *tio;
1094 clone = bio_alloc_bioset(GFP_NOIO, nr_iovecs, ci->md->bs);
1095 tio = container_of(clone, struct dm_target_io, clone);
1099 memset(&tio->info, 0, sizeof(tio->info));
1100 tio->target_request_nr = 0;
1105 static void __issue_target_request(struct clone_info *ci, struct dm_target *ti,
1106 unsigned request_nr, sector_t len)
1108 struct dm_target_io *tio = alloc_tio(ci, ti, ci->bio->bi_max_vecs);
1109 struct bio *clone = &tio->clone;
1111 tio->target_request_nr = request_nr;
1114 * Discard requests require the bio's inline iovecs be initialized.
1115 * ci->bio->bi_max_vecs is BIO_INLINE_VECS anyway, for both flush
1116 * and discard, so no need for concern about wasted bvec allocations.
1119 __bio_clone(clone, ci->bio);
1121 clone->bi_sector = ci->sector;
1122 clone->bi_size = to_bytes(len);
1128 static void __issue_target_requests(struct clone_info *ci, struct dm_target *ti,
1129 unsigned num_requests, sector_t len)
1131 unsigned request_nr;
1133 for (request_nr = 0; request_nr < num_requests; request_nr++)
1134 __issue_target_request(ci, ti, request_nr, len);
1137 static int __clone_and_map_empty_flush(struct clone_info *ci)
1139 unsigned target_nr = 0;
1140 struct dm_target *ti;
1142 BUG_ON(bio_has_data(ci->bio));
1143 while ((ti = dm_table_get_target(ci->map, target_nr++)))
1144 __issue_target_requests(ci, ti, ti->num_flush_requests, 0);
1150 * Perform all io with a single clone.
1152 static void __clone_and_map_simple(struct clone_info *ci, struct dm_target *ti)
1154 struct bio *bio = ci->bio;
1155 struct dm_target_io *tio;
1157 tio = alloc_tio(ci, ti, bio->bi_max_vecs);
1158 clone_bio(tio, bio, ci->sector, ci->idx, bio->bi_vcnt - ci->idx,
1159 ci->sector_count, ci->md->bs);
1161 ci->sector_count = 0;
1164 typedef unsigned (*get_num_requests_fn)(struct dm_target *ti);
1166 static unsigned get_num_discard_requests(struct dm_target *ti)
1168 return ti->num_discard_requests;
1171 static unsigned get_num_write_same_requests(struct dm_target *ti)
1173 return ti->num_write_same_requests;
1176 typedef bool (*is_split_required_fn)(struct dm_target *ti);
1178 static bool is_split_required_for_discard(struct dm_target *ti)
1180 return ti->split_discard_requests;
1183 static int __clone_and_map_changing_extent_only(struct clone_info *ci,
1184 get_num_requests_fn get_num_requests,
1185 is_split_required_fn is_split_required)
1187 struct dm_target *ti;
1189 unsigned num_requests;
1192 ti = dm_table_find_target(ci->map, ci->sector);
1193 if (!dm_target_is_valid(ti))
1197 * Even though the device advertised support for this type of
1198 * request, that does not mean every target supports it, and
1199 * reconfiguration might also have changed that since the
1200 * check was performed.
1202 num_requests = get_num_requests ? get_num_requests(ti) : 0;
1206 if (is_split_required && !is_split_required(ti))
1207 len = min(ci->sector_count, max_io_len_target_boundary(ci->sector, ti));
1209 len = min(ci->sector_count, max_io_len(ci->sector, ti));
1211 __issue_target_requests(ci, ti, num_requests, len);
1214 } while (ci->sector_count -= len);
1219 static int __clone_and_map_discard(struct clone_info *ci)
1221 return __clone_and_map_changing_extent_only(ci, get_num_discard_requests,
1222 is_split_required_for_discard);
1225 static int __clone_and_map_write_same(struct clone_info *ci)
1227 return __clone_and_map_changing_extent_only(ci, get_num_write_same_requests, NULL);
1230 static int __clone_and_map(struct clone_info *ci)
1232 struct bio *bio = ci->bio;
1233 struct dm_target *ti;
1234 sector_t len = 0, max;
1235 struct dm_target_io *tio;
1237 if (unlikely(bio->bi_rw & REQ_DISCARD))
1238 return __clone_and_map_discard(ci);
1239 else if (unlikely(bio->bi_rw & REQ_WRITE_SAME))
1240 return __clone_and_map_write_same(ci);
1242 ti = dm_table_find_target(ci->map, ci->sector);
1243 if (!dm_target_is_valid(ti))
1246 max = max_io_len(ci->sector, ti);
1248 if (ci->sector_count <= max) {
1250 * Optimise for the simple case where we can do all of
1251 * the remaining io with a single clone.
1253 __clone_and_map_simple(ci, ti);
1255 } else if (to_sector(bio->bi_io_vec[ci->idx].bv_len) <= max) {
1257 * There are some bvecs that don't span targets.
1258 * Do as many of these as possible.
1261 sector_t remaining = max;
1264 for (i = ci->idx; remaining && (i < bio->bi_vcnt); i++) {
1265 bv_len = to_sector(bio->bi_io_vec[i].bv_len);
1267 if (bv_len > remaining)
1270 remaining -= bv_len;
1274 tio = alloc_tio(ci, ti, bio->bi_max_vecs);
1275 clone_bio(tio, bio, ci->sector, ci->idx, i - ci->idx, len,
1280 ci->sector_count -= len;
1285 * Handle a bvec that must be split between two or more targets.
1287 struct bio_vec *bv = bio->bi_io_vec + ci->idx;
1288 sector_t remaining = to_sector(bv->bv_len);
1289 unsigned int offset = 0;
1293 ti = dm_table_find_target(ci->map, ci->sector);
1294 if (!dm_target_is_valid(ti))
1297 max = max_io_len(ci->sector, ti);
1300 len = min(remaining, max);
1302 tio = alloc_tio(ci, ti, 1);
1303 split_bvec(tio, bio, ci->sector, ci->idx,
1304 bv->bv_offset + offset, len, ci->md->bs);
1309 ci->sector_count -= len;
1310 offset += to_bytes(len);
1311 } while (remaining -= len);
1320 * Split the bio into several clones and submit it to targets.
1322 static void __split_and_process_bio(struct mapped_device *md, struct bio *bio)
1324 struct clone_info ci;
1327 ci.map = dm_get_live_table(md);
1328 if (unlikely(!ci.map)) {
1334 ci.io = alloc_io(md);
1336 atomic_set(&ci.io->io_count, 1);
1339 spin_lock_init(&ci.io->endio_lock);
1340 ci.sector = bio->bi_sector;
1341 ci.idx = bio->bi_idx;
1343 start_io_acct(ci.io);
1344 if (bio->bi_rw & REQ_FLUSH) {
1345 ci.bio = &ci.md->flush_bio;
1346 ci.sector_count = 0;
1347 error = __clone_and_map_empty_flush(&ci);
1348 /* dec_pending submits any data associated with flush */
1351 ci.sector_count = bio_sectors(bio);
1352 while (ci.sector_count && !error)
1353 error = __clone_and_map(&ci);
1356 /* drop the extra reference count */
1357 dec_pending(ci.io, error);
1358 dm_table_put(ci.map);
1360 /*-----------------------------------------------------------------
1362 *---------------------------------------------------------------*/
1364 static int dm_merge_bvec(struct request_queue *q,
1365 struct bvec_merge_data *bvm,
1366 struct bio_vec *biovec)
1368 struct mapped_device *md = q->queuedata;
1369 struct dm_table *map = dm_get_live_table(md);
1370 struct dm_target *ti;
1371 sector_t max_sectors;
1377 ti = dm_table_find_target(map, bvm->bi_sector);
1378 if (!dm_target_is_valid(ti))
1382 * Find maximum amount of I/O that won't need splitting
1384 max_sectors = min(max_io_len(bvm->bi_sector, ti),
1385 (sector_t) BIO_MAX_SECTORS);
1386 max_size = (max_sectors << SECTOR_SHIFT) - bvm->bi_size;
1391 * merge_bvec_fn() returns number of bytes
1392 * it can accept at this offset
1393 * max is precomputed maximal io size
1395 if (max_size && ti->type->merge)
1396 max_size = ti->type->merge(ti, bvm, biovec, max_size);
1398 * If the target doesn't support merge method and some of the devices
1399 * provided their merge_bvec method (we know this by looking at
1400 * queue_max_hw_sectors), then we can't allow bios with multiple vector
1401 * entries. So always set max_size to 0, and the code below allows
1404 else if (queue_max_hw_sectors(q) <= PAGE_SIZE >> 9)
1413 * Always allow an entire first page
1415 if (max_size <= biovec->bv_len && !(bvm->bi_size >> SECTOR_SHIFT))
1416 max_size = biovec->bv_len;
1422 * The request function that just remaps the bio built up by
1425 static void _dm_request(struct request_queue *q, struct bio *bio)
1427 int rw = bio_data_dir(bio);
1428 struct mapped_device *md = q->queuedata;
1431 down_read(&md->io_lock);
1433 cpu = part_stat_lock();
1434 part_stat_inc(cpu, &dm_disk(md)->part0, ios[rw]);
1435 part_stat_add(cpu, &dm_disk(md)->part0, sectors[rw], bio_sectors(bio));
1438 /* if we're suspended, we have to queue this io for later */
1439 if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))) {
1440 up_read(&md->io_lock);
1442 if (bio_rw(bio) != READA)
1449 __split_and_process_bio(md, bio);
1450 up_read(&md->io_lock);
1454 static int dm_request_based(struct mapped_device *md)
1456 return blk_queue_stackable(md->queue);
1459 static void dm_request(struct request_queue *q, struct bio *bio)
1461 struct mapped_device *md = q->queuedata;
1463 if (dm_request_based(md))
1464 blk_queue_bio(q, bio);
1466 _dm_request(q, bio);
1469 void dm_dispatch_request(struct request *rq)
1473 if (blk_queue_io_stat(rq->q))
1474 rq->cmd_flags |= REQ_IO_STAT;
1476 rq->start_time = jiffies;
1477 r = blk_insert_cloned_request(rq->q, rq);
1479 dm_complete_request(rq, r);
1481 EXPORT_SYMBOL_GPL(dm_dispatch_request);
1483 static int dm_rq_bio_constructor(struct bio *bio, struct bio *bio_orig,
1486 struct dm_rq_target_io *tio = data;
1487 struct dm_rq_clone_bio_info *info =
1488 container_of(bio, struct dm_rq_clone_bio_info, clone);
1490 info->orig = bio_orig;
1492 bio->bi_end_io = end_clone_bio;
1493 bio->bi_private = info;
1498 static int setup_clone(struct request *clone, struct request *rq,
1499 struct dm_rq_target_io *tio)
1503 r = blk_rq_prep_clone(clone, rq, tio->md->bs, GFP_ATOMIC,
1504 dm_rq_bio_constructor, tio);
1508 clone->cmd = rq->cmd;
1509 clone->cmd_len = rq->cmd_len;
1510 clone->sense = rq->sense;
1511 clone->buffer = rq->buffer;
1512 clone->end_io = end_clone_request;
1513 clone->end_io_data = tio;
1518 static struct request *clone_rq(struct request *rq, struct mapped_device *md,
1521 struct request *clone;
1522 struct dm_rq_target_io *tio;
1524 tio = alloc_rq_tio(md, gfp_mask);
1532 memset(&tio->info, 0, sizeof(tio->info));
1534 clone = &tio->clone;
1535 if (setup_clone(clone, rq, tio)) {
1545 * Called with the queue lock held.
1547 static int dm_prep_fn(struct request_queue *q, struct request *rq)
1549 struct mapped_device *md = q->queuedata;
1550 struct request *clone;
1552 if (unlikely(rq->special)) {
1553 DMWARN("Already has something in rq->special.");
1554 return BLKPREP_KILL;
1557 clone = clone_rq(rq, md, GFP_ATOMIC);
1559 return BLKPREP_DEFER;
1561 rq->special = clone;
1562 rq->cmd_flags |= REQ_DONTPREP;
1569 * 0 : the request has been processed (not requeued)
1570 * !0 : the request has been requeued
1572 static int map_request(struct dm_target *ti, struct request *clone,
1573 struct mapped_device *md)
1575 int r, requeued = 0;
1576 struct dm_rq_target_io *tio = clone->end_io_data;
1579 r = ti->type->map_rq(ti, clone, &tio->info);
1581 case DM_MAPIO_SUBMITTED:
1582 /* The target has taken the I/O to submit by itself later */
1584 case DM_MAPIO_REMAPPED:
1585 /* The target has remapped the I/O so dispatch it */
1586 trace_block_rq_remap(clone->q, clone, disk_devt(dm_disk(md)),
1587 blk_rq_pos(tio->orig));
1588 dm_dispatch_request(clone);
1590 case DM_MAPIO_REQUEUE:
1591 /* The target wants to requeue the I/O */
1592 dm_requeue_unmapped_request(clone);
1597 DMWARN("unimplemented target map return value: %d", r);
1601 /* The target wants to complete the I/O */
1602 dm_kill_unmapped_request(clone, r);
1609 static struct request *dm_start_request(struct mapped_device *md, struct request *orig)
1611 struct request *clone;
1613 blk_start_request(orig);
1614 clone = orig->special;
1615 atomic_inc(&md->pending[rq_data_dir(clone)]);
1618 * Hold the md reference here for the in-flight I/O.
1619 * We can't rely on the reference count by device opener,
1620 * because the device may be closed during the request completion
1621 * when all bios are completed.
1622 * See the comment in rq_completed() too.
1630 * q->request_fn for request-based dm.
1631 * Called with the queue lock held.
1633 static void dm_request_fn(struct request_queue *q)
1635 struct mapped_device *md = q->queuedata;
1636 struct dm_table *map = dm_get_live_table(md);
1637 struct dm_target *ti;
1638 struct request *rq, *clone;
1642 * For suspend, check blk_queue_stopped() and increment
1643 * ->pending within a single queue_lock not to increment the
1644 * number of in-flight I/Os after the queue is stopped in
1647 while (!blk_queue_stopped(q)) {
1648 rq = blk_peek_request(q);
1652 /* always use block 0 to find the target for flushes for now */
1654 if (!(rq->cmd_flags & REQ_FLUSH))
1655 pos = blk_rq_pos(rq);
1657 ti = dm_table_find_target(map, pos);
1658 if (!dm_target_is_valid(ti)) {
1660 * Must perform setup, that dm_done() requires,
1661 * before calling dm_kill_unmapped_request
1663 DMERR_LIMIT("request attempted access beyond the end of device");
1664 clone = dm_start_request(md, rq);
1665 dm_kill_unmapped_request(clone, -EIO);
1669 if (ti->type->busy && ti->type->busy(ti))
1672 clone = dm_start_request(md, rq);
1674 spin_unlock(q->queue_lock);
1675 if (map_request(ti, clone, md))
1678 BUG_ON(!irqs_disabled());
1679 spin_lock(q->queue_lock);
1685 BUG_ON(!irqs_disabled());
1686 spin_lock(q->queue_lock);
1689 blk_delay_queue(q, HZ / 10);
1694 int dm_underlying_device_busy(struct request_queue *q)
1696 return blk_lld_busy(q);
1698 EXPORT_SYMBOL_GPL(dm_underlying_device_busy);
1700 static int dm_lld_busy(struct request_queue *q)
1703 struct mapped_device *md = q->queuedata;
1704 struct dm_table *map = dm_get_live_table(md);
1706 if (!map || test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))
1709 r = dm_table_any_busy_target(map);
1716 static int dm_any_congested(void *congested_data, int bdi_bits)
1719 struct mapped_device *md = congested_data;
1720 struct dm_table *map;
1722 if (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
1723 map = dm_get_live_table(md);
1726 * Request-based dm cares about only own queue for
1727 * the query about congestion status of request_queue
1729 if (dm_request_based(md))
1730 r = md->queue->backing_dev_info.state &
1733 r = dm_table_any_congested(map, bdi_bits);
1742 /*-----------------------------------------------------------------
1743 * An IDR is used to keep track of allocated minor numbers.
1744 *---------------------------------------------------------------*/
1745 static void free_minor(int minor)
1747 spin_lock(&_minor_lock);
1748 idr_remove(&_minor_idr, minor);
1749 spin_unlock(&_minor_lock);
1753 * See if the device with a specific minor # is free.
1755 static int specific_minor(int minor)
1759 if (minor >= (1 << MINORBITS))
1762 idr_preload(GFP_KERNEL);
1763 spin_lock(&_minor_lock);
1765 r = idr_alloc(&_minor_idr, MINOR_ALLOCED, minor, minor + 1, GFP_NOWAIT);
1767 spin_unlock(&_minor_lock);
1770 return r == -ENOSPC ? -EBUSY : r;
1774 static int next_free_minor(int *minor)
1778 idr_preload(GFP_KERNEL);
1779 spin_lock(&_minor_lock);
1781 r = idr_alloc(&_minor_idr, MINOR_ALLOCED, 0, 1 << MINORBITS, GFP_NOWAIT);
1783 spin_unlock(&_minor_lock);
1791 static const struct block_device_operations dm_blk_dops;
1793 static void dm_wq_work(struct work_struct *work);
1795 static void dm_init_md_queue(struct mapped_device *md)
1798 * Request-based dm devices cannot be stacked on top of bio-based dm
1799 * devices. The type of this dm device has not been decided yet.
1800 * The type is decided at the first table loading time.
1801 * To prevent problematic device stacking, clear the queue flag
1802 * for request stacking support until then.
1804 * This queue is new, so no concurrency on the queue_flags.
1806 queue_flag_clear_unlocked(QUEUE_FLAG_STACKABLE, md->queue);
1808 md->queue->queuedata = md;
1809 md->queue->backing_dev_info.congested_fn = dm_any_congested;
1810 md->queue->backing_dev_info.congested_data = md;
1811 blk_queue_make_request(md->queue, dm_request);
1812 blk_queue_bounce_limit(md->queue, BLK_BOUNCE_ANY);
1813 blk_queue_merge_bvec(md->queue, dm_merge_bvec);
1817 * Allocate and initialise a blank device with a given minor.
1819 static struct mapped_device *alloc_dev(int minor)
1822 struct mapped_device *md = kzalloc(sizeof(*md), GFP_KERNEL);
1826 DMWARN("unable to allocate device, out of memory.");
1830 if (!try_module_get(THIS_MODULE))
1831 goto bad_module_get;
1833 /* get a minor number for the dev */
1834 if (minor == DM_ANY_MINOR)
1835 r = next_free_minor(&minor);
1837 r = specific_minor(minor);
1841 md->type = DM_TYPE_NONE;
1842 init_rwsem(&md->io_lock);
1843 mutex_init(&md->suspend_lock);
1844 mutex_init(&md->type_lock);
1845 spin_lock_init(&md->deferred_lock);
1846 rwlock_init(&md->map_lock);
1847 atomic_set(&md->holders, 1);
1848 atomic_set(&md->open_count, 0);
1849 atomic_set(&md->event_nr, 0);
1850 atomic_set(&md->uevent_seq, 0);
1851 INIT_LIST_HEAD(&md->uevent_list);
1852 spin_lock_init(&md->uevent_lock);
1854 md->queue = blk_alloc_queue(GFP_KERNEL);
1858 dm_init_md_queue(md);
1860 md->disk = alloc_disk(1);
1864 atomic_set(&md->pending[0], 0);
1865 atomic_set(&md->pending[1], 0);
1866 init_waitqueue_head(&md->wait);
1867 INIT_WORK(&md->work, dm_wq_work);
1868 init_waitqueue_head(&md->eventq);
1870 md->disk->major = _major;
1871 md->disk->first_minor = minor;
1872 md->disk->fops = &dm_blk_dops;
1873 md->disk->queue = md->queue;
1874 md->disk->private_data = md;
1875 sprintf(md->disk->disk_name, "dm-%d", minor);
1877 format_dev_t(md->name, MKDEV(_major, minor));
1879 md->wq = alloc_workqueue("kdmflush",
1880 WQ_NON_REENTRANT | WQ_MEM_RECLAIM, 0);
1884 md->bdev = bdget_disk(md->disk, 0);
1888 bio_init(&md->flush_bio);
1889 md->flush_bio.bi_bdev = md->bdev;
1890 md->flush_bio.bi_rw = WRITE_FLUSH;
1892 /* Populate the mapping, nobody knows we exist yet */
1893 spin_lock(&_minor_lock);
1894 old_md = idr_replace(&_minor_idr, md, minor);
1895 spin_unlock(&_minor_lock);
1897 BUG_ON(old_md != MINOR_ALLOCED);
1902 destroy_workqueue(md->wq);
1904 del_gendisk(md->disk);
1907 blk_cleanup_queue(md->queue);
1911 module_put(THIS_MODULE);
1917 static void unlock_fs(struct mapped_device *md);
1919 static void free_dev(struct mapped_device *md)
1921 int minor = MINOR(disk_devt(md->disk));
1925 destroy_workqueue(md->wq);
1927 mempool_destroy(md->tio_pool);
1929 mempool_destroy(md->io_pool);
1931 bioset_free(md->bs);
1932 blk_integrity_unregister(md->disk);
1933 del_gendisk(md->disk);
1936 spin_lock(&_minor_lock);
1937 md->disk->private_data = NULL;
1938 spin_unlock(&_minor_lock);
1941 blk_cleanup_queue(md->queue);
1942 module_put(THIS_MODULE);
1946 static void __bind_mempools(struct mapped_device *md, struct dm_table *t)
1948 struct dm_md_mempools *p = dm_table_get_md_mempools(t);
1950 if (md->io_pool && (md->tio_pool || dm_table_get_type(t) == DM_TYPE_BIO_BASED) && md->bs) {
1952 * The md already has necessary mempools. Reload just the
1953 * bioset because front_pad may have changed because
1954 * a different table was loaded.
1956 bioset_free(md->bs);
1962 BUG_ON(!p || md->io_pool || md->tio_pool || md->bs);
1964 md->io_pool = p->io_pool;
1966 md->tio_pool = p->tio_pool;
1972 /* mempool bind completed, now no need any mempools in the table */
1973 dm_table_free_md_mempools(t);
1977 * Bind a table to the device.
1979 static void event_callback(void *context)
1981 unsigned long flags;
1983 struct mapped_device *md = (struct mapped_device *) context;
1985 spin_lock_irqsave(&md->uevent_lock, flags);
1986 list_splice_init(&md->uevent_list, &uevents);
1987 spin_unlock_irqrestore(&md->uevent_lock, flags);
1989 dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj);
1991 atomic_inc(&md->event_nr);
1992 wake_up(&md->eventq);
1996 * Protected by md->suspend_lock obtained by dm_swap_table().
1998 static void __set_size(struct mapped_device *md, sector_t size)
2000 set_capacity(md->disk, size);
2002 i_size_write(md->bdev->bd_inode, (loff_t)size << SECTOR_SHIFT);
2006 * Return 1 if the queue has a compulsory merge_bvec_fn function.
2008 * If this function returns 0, then the device is either a non-dm
2009 * device without a merge_bvec_fn, or it is a dm device that is
2010 * able to split any bios it receives that are too big.
2012 int dm_queue_merge_is_compulsory(struct request_queue *q)
2014 struct mapped_device *dev_md;
2016 if (!q->merge_bvec_fn)
2019 if (q->make_request_fn == dm_request) {
2020 dev_md = q->queuedata;
2021 if (test_bit(DMF_MERGE_IS_OPTIONAL, &dev_md->flags))
2028 static int dm_device_merge_is_compulsory(struct dm_target *ti,
2029 struct dm_dev *dev, sector_t start,
2030 sector_t len, void *data)
2032 struct block_device *bdev = dev->bdev;
2033 struct request_queue *q = bdev_get_queue(bdev);
2035 return dm_queue_merge_is_compulsory(q);
2039 * Return 1 if it is acceptable to ignore merge_bvec_fn based
2040 * on the properties of the underlying devices.
2042 static int dm_table_merge_is_optional(struct dm_table *table)
2045 struct dm_target *ti;
2047 while (i < dm_table_get_num_targets(table)) {
2048 ti = dm_table_get_target(table, i++);
2050 if (ti->type->iterate_devices &&
2051 ti->type->iterate_devices(ti, dm_device_merge_is_compulsory, NULL))
2059 * Returns old map, which caller must destroy.
2061 static struct dm_table *__bind(struct mapped_device *md, struct dm_table *t,
2062 struct queue_limits *limits)
2064 struct dm_table *old_map;
2065 struct request_queue *q = md->queue;
2067 unsigned long flags;
2068 int merge_is_optional;
2070 size = dm_table_get_size(t);
2073 * Wipe any geometry if the size of the table changed.
2075 if (size != get_capacity(md->disk))
2076 memset(&md->geometry, 0, sizeof(md->geometry));
2078 __set_size(md, size);
2080 dm_table_event_callback(t, event_callback, md);
2083 * The queue hasn't been stopped yet, if the old table type wasn't
2084 * for request-based during suspension. So stop it to prevent
2085 * I/O mapping before resume.
2086 * This must be done before setting the queue restrictions,
2087 * because request-based dm may be run just after the setting.
2089 if (dm_table_request_based(t) && !blk_queue_stopped(q))
2092 __bind_mempools(md, t);
2094 merge_is_optional = dm_table_merge_is_optional(t);
2096 write_lock_irqsave(&md->map_lock, flags);
2099 md->immutable_target_type = dm_table_get_immutable_target_type(t);
2101 dm_table_set_restrictions(t, q, limits);
2102 if (merge_is_optional)
2103 set_bit(DMF_MERGE_IS_OPTIONAL, &md->flags);
2105 clear_bit(DMF_MERGE_IS_OPTIONAL, &md->flags);
2106 write_unlock_irqrestore(&md->map_lock, flags);
2112 * Returns unbound table for the caller to free.
2114 static struct dm_table *__unbind(struct mapped_device *md)
2116 struct dm_table *map = md->map;
2117 unsigned long flags;
2122 dm_table_event_callback(map, NULL, NULL);
2123 write_lock_irqsave(&md->map_lock, flags);
2125 write_unlock_irqrestore(&md->map_lock, flags);
2131 * Constructor for a new device.
2133 int dm_create(int minor, struct mapped_device **result)
2135 struct mapped_device *md;
2137 md = alloc_dev(minor);
2148 * Functions to manage md->type.
2149 * All are required to hold md->type_lock.
2151 void dm_lock_md_type(struct mapped_device *md)
2153 mutex_lock(&md->type_lock);
2156 void dm_unlock_md_type(struct mapped_device *md)
2158 mutex_unlock(&md->type_lock);
2161 void dm_set_md_type(struct mapped_device *md, unsigned type)
2166 unsigned dm_get_md_type(struct mapped_device *md)
2171 struct target_type *dm_get_immutable_target_type(struct mapped_device *md)
2173 return md->immutable_target_type;
2177 * Fully initialize a request-based queue (->elevator, ->request_fn, etc).
2179 static int dm_init_request_based_queue(struct mapped_device *md)
2181 struct request_queue *q = NULL;
2183 if (md->queue->elevator)
2186 /* Fully initialize the queue */
2187 q = blk_init_allocated_queue(md->queue, dm_request_fn, NULL);
2192 dm_init_md_queue(md);
2193 blk_queue_softirq_done(md->queue, dm_softirq_done);
2194 blk_queue_prep_rq(md->queue, dm_prep_fn);
2195 blk_queue_lld_busy(md->queue, dm_lld_busy);
2197 elv_register_queue(md->queue);
2203 * Setup the DM device's queue based on md's type
2205 int dm_setup_md_queue(struct mapped_device *md)
2207 if ((dm_get_md_type(md) == DM_TYPE_REQUEST_BASED) &&
2208 !dm_init_request_based_queue(md)) {
2209 DMWARN("Cannot initialize queue for request-based mapped device");
2216 static struct mapped_device *dm_find_md(dev_t dev)
2218 struct mapped_device *md;
2219 unsigned minor = MINOR(dev);
2221 if (MAJOR(dev) != _major || minor >= (1 << MINORBITS))
2224 spin_lock(&_minor_lock);
2226 md = idr_find(&_minor_idr, minor);
2227 if (md && (md == MINOR_ALLOCED ||
2228 (MINOR(disk_devt(dm_disk(md))) != minor) ||
2229 dm_deleting_md(md) ||
2230 test_bit(DMF_FREEING, &md->flags))) {
2236 spin_unlock(&_minor_lock);
2241 struct mapped_device *dm_get_md(dev_t dev)
2243 struct mapped_device *md = dm_find_md(dev);
2250 EXPORT_SYMBOL_GPL(dm_get_md);
2252 void *dm_get_mdptr(struct mapped_device *md)
2254 return md->interface_ptr;
2257 void dm_set_mdptr(struct mapped_device *md, void *ptr)
2259 md->interface_ptr = ptr;
2262 void dm_get(struct mapped_device *md)
2264 atomic_inc(&md->holders);
2265 BUG_ON(test_bit(DMF_FREEING, &md->flags));
2268 const char *dm_device_name(struct mapped_device *md)
2272 EXPORT_SYMBOL_GPL(dm_device_name);
2274 static void __dm_destroy(struct mapped_device *md, bool wait)
2276 struct dm_table *map;
2280 spin_lock(&_minor_lock);
2281 map = dm_get_live_table(md);
2282 idr_replace(&_minor_idr, MINOR_ALLOCED, MINOR(disk_devt(dm_disk(md))));
2283 set_bit(DMF_FREEING, &md->flags);
2284 spin_unlock(&_minor_lock);
2286 if (!dm_suspended_md(md)) {
2287 dm_table_presuspend_targets(map);
2288 dm_table_postsuspend_targets(map);
2292 * Rare, but there may be I/O requests still going to complete,
2293 * for example. Wait for all references to disappear.
2294 * No one should increment the reference count of the mapped_device,
2295 * after the mapped_device state becomes DMF_FREEING.
2298 while (atomic_read(&md->holders))
2300 else if (atomic_read(&md->holders))
2301 DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)",
2302 dm_device_name(md), atomic_read(&md->holders));
2306 dm_table_destroy(__unbind(md));
2310 void dm_destroy(struct mapped_device *md)
2312 __dm_destroy(md, true);
2315 void dm_destroy_immediate(struct mapped_device *md)
2317 __dm_destroy(md, false);
2320 void dm_put(struct mapped_device *md)
2322 atomic_dec(&md->holders);
2324 EXPORT_SYMBOL_GPL(dm_put);
2326 static int dm_wait_for_completion(struct mapped_device *md, int interruptible)
2329 DECLARE_WAITQUEUE(wait, current);
2331 add_wait_queue(&md->wait, &wait);
2334 set_current_state(interruptible);
2336 if (!md_in_flight(md))
2339 if (interruptible == TASK_INTERRUPTIBLE &&
2340 signal_pending(current)) {
2347 set_current_state(TASK_RUNNING);
2349 remove_wait_queue(&md->wait, &wait);
2355 * Process the deferred bios
2357 static void dm_wq_work(struct work_struct *work)
2359 struct mapped_device *md = container_of(work, struct mapped_device,
2363 down_read(&md->io_lock);
2365 while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
2366 spin_lock_irq(&md->deferred_lock);
2367 c = bio_list_pop(&md->deferred);
2368 spin_unlock_irq(&md->deferred_lock);
2373 up_read(&md->io_lock);
2375 if (dm_request_based(md))
2376 generic_make_request(c);
2378 __split_and_process_bio(md, c);
2380 down_read(&md->io_lock);
2383 up_read(&md->io_lock);
2386 static void dm_queue_flush(struct mapped_device *md)
2388 clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2389 smp_mb__after_clear_bit();
2390 queue_work(md->wq, &md->work);
2394 * Swap in a new table, returning the old one for the caller to destroy.
2396 struct dm_table *dm_swap_table(struct mapped_device *md, struct dm_table *table)
2398 struct dm_table *live_map, *map = ERR_PTR(-EINVAL);
2399 struct queue_limits limits;
2402 mutex_lock(&md->suspend_lock);
2404 /* device must be suspended */
2405 if (!dm_suspended_md(md))
2409 * If the new table has no data devices, retain the existing limits.
2410 * This helps multipath with queue_if_no_path if all paths disappear,
2411 * then new I/O is queued based on these limits, and then some paths
2414 if (dm_table_has_no_data_devices(table)) {
2415 live_map = dm_get_live_table(md);
2417 limits = md->queue->limits;
2418 dm_table_put(live_map);
2421 r = dm_calculate_queue_limits(table, &limits);
2427 map = __bind(md, table, &limits);
2430 mutex_unlock(&md->suspend_lock);
2435 * Functions to lock and unlock any filesystem running on the
2438 static int lock_fs(struct mapped_device *md)
2442 WARN_ON(md->frozen_sb);
2444 md->frozen_sb = freeze_bdev(md->bdev);
2445 if (IS_ERR(md->frozen_sb)) {
2446 r = PTR_ERR(md->frozen_sb);
2447 md->frozen_sb = NULL;
2451 set_bit(DMF_FROZEN, &md->flags);
2456 static void unlock_fs(struct mapped_device *md)
2458 if (!test_bit(DMF_FROZEN, &md->flags))
2461 thaw_bdev(md->bdev, md->frozen_sb);
2462 md->frozen_sb = NULL;
2463 clear_bit(DMF_FROZEN, &md->flags);
2467 * We need to be able to change a mapping table under a mounted
2468 * filesystem. For example we might want to move some data in
2469 * the background. Before the table can be swapped with
2470 * dm_bind_table, dm_suspend must be called to flush any in
2471 * flight bios and ensure that any further io gets deferred.
2474 * Suspend mechanism in request-based dm.
2476 * 1. Flush all I/Os by lock_fs() if needed.
2477 * 2. Stop dispatching any I/O by stopping the request_queue.
2478 * 3. Wait for all in-flight I/Os to be completed or requeued.
2480 * To abort suspend, start the request_queue.
2482 int dm_suspend(struct mapped_device *md, unsigned suspend_flags)
2484 struct dm_table *map = NULL;
2486 int do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG ? 1 : 0;
2487 int noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG ? 1 : 0;
2489 mutex_lock(&md->suspend_lock);
2491 if (dm_suspended_md(md)) {
2496 map = dm_get_live_table(md);
2499 * DMF_NOFLUSH_SUSPENDING must be set before presuspend.
2500 * This flag is cleared before dm_suspend returns.
2503 set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2505 /* This does not get reverted if there's an error later. */
2506 dm_table_presuspend_targets(map);
2509 * Flush I/O to the device.
2510 * Any I/O submitted after lock_fs() may not be flushed.
2511 * noflush takes precedence over do_lockfs.
2512 * (lock_fs() flushes I/Os and waits for them to complete.)
2514 if (!noflush && do_lockfs) {
2521 * Here we must make sure that no processes are submitting requests
2522 * to target drivers i.e. no one may be executing
2523 * __split_and_process_bio. This is called from dm_request and
2526 * To get all processes out of __split_and_process_bio in dm_request,
2527 * we take the write lock. To prevent any process from reentering
2528 * __split_and_process_bio from dm_request and quiesce the thread
2529 * (dm_wq_work), we set BMF_BLOCK_IO_FOR_SUSPEND and call
2530 * flush_workqueue(md->wq).
2532 down_write(&md->io_lock);
2533 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2534 up_write(&md->io_lock);
2537 * Stop md->queue before flushing md->wq in case request-based
2538 * dm defers requests to md->wq from md->queue.
2540 if (dm_request_based(md))
2541 stop_queue(md->queue);
2543 flush_workqueue(md->wq);
2546 * At this point no more requests are entering target request routines.
2547 * We call dm_wait_for_completion to wait for all existing requests
2550 r = dm_wait_for_completion(md, TASK_INTERRUPTIBLE);
2552 down_write(&md->io_lock);
2554 clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2555 up_write(&md->io_lock);
2557 /* were we interrupted ? */
2561 if (dm_request_based(md))
2562 start_queue(md->queue);
2565 goto out; /* pushback list is already flushed, so skip flush */
2569 * If dm_wait_for_completion returned 0, the device is completely
2570 * quiescent now. There is no request-processing activity. All new
2571 * requests are being added to md->deferred list.
2574 set_bit(DMF_SUSPENDED, &md->flags);
2576 dm_table_postsuspend_targets(map);
2582 mutex_unlock(&md->suspend_lock);
2586 int dm_resume(struct mapped_device *md)
2589 struct dm_table *map = NULL;
2591 mutex_lock(&md->suspend_lock);
2592 if (!dm_suspended_md(md))
2595 map = dm_get_live_table(md);
2596 if (!map || !dm_table_get_size(map))
2599 r = dm_table_resume_targets(map);
2606 * Flushing deferred I/Os must be done after targets are resumed
2607 * so that mapping of targets can work correctly.
2608 * Request-based dm is queueing the deferred I/Os in its request_queue.
2610 if (dm_request_based(md))
2611 start_queue(md->queue);
2615 clear_bit(DMF_SUSPENDED, &md->flags);
2620 mutex_unlock(&md->suspend_lock);
2625 /*-----------------------------------------------------------------
2626 * Event notification.
2627 *---------------------------------------------------------------*/
2628 int dm_kobject_uevent(struct mapped_device *md, enum kobject_action action,
2631 char udev_cookie[DM_COOKIE_LENGTH];
2632 char *envp[] = { udev_cookie, NULL };
2635 return kobject_uevent(&disk_to_dev(md->disk)->kobj, action);
2637 snprintf(udev_cookie, DM_COOKIE_LENGTH, "%s=%u",
2638 DM_COOKIE_ENV_VAR_NAME, cookie);
2639 return kobject_uevent_env(&disk_to_dev(md->disk)->kobj,
2644 uint32_t dm_next_uevent_seq(struct mapped_device *md)
2646 return atomic_add_return(1, &md->uevent_seq);
2649 uint32_t dm_get_event_nr(struct mapped_device *md)
2651 return atomic_read(&md->event_nr);
2654 int dm_wait_event(struct mapped_device *md, int event_nr)
2656 return wait_event_interruptible(md->eventq,
2657 (event_nr != atomic_read(&md->event_nr)));
2660 void dm_uevent_add(struct mapped_device *md, struct list_head *elist)
2662 unsigned long flags;
2664 spin_lock_irqsave(&md->uevent_lock, flags);
2665 list_add(elist, &md->uevent_list);
2666 spin_unlock_irqrestore(&md->uevent_lock, flags);
2670 * The gendisk is only valid as long as you have a reference
2673 struct gendisk *dm_disk(struct mapped_device *md)
2678 struct kobject *dm_kobject(struct mapped_device *md)
2684 * struct mapped_device should not be exported outside of dm.c
2685 * so use this check to verify that kobj is part of md structure
2687 struct mapped_device *dm_get_from_kobject(struct kobject *kobj)
2689 struct mapped_device *md;
2691 md = container_of(kobj, struct mapped_device, kobj);
2692 if (&md->kobj != kobj)
2695 if (test_bit(DMF_FREEING, &md->flags) ||
2703 int dm_suspended_md(struct mapped_device *md)
2705 return test_bit(DMF_SUSPENDED, &md->flags);
2708 int dm_suspended(struct dm_target *ti)
2710 return dm_suspended_md(dm_table_get_md(ti->table));
2712 EXPORT_SYMBOL_GPL(dm_suspended);
2714 int dm_noflush_suspending(struct dm_target *ti)
2716 return __noflush_suspending(dm_table_get_md(ti->table));
2718 EXPORT_SYMBOL_GPL(dm_noflush_suspending);
2720 struct dm_md_mempools *dm_alloc_md_mempools(unsigned type, unsigned integrity, unsigned per_bio_data_size)
2722 struct dm_md_mempools *pools = kmalloc(sizeof(*pools), GFP_KERNEL);
2723 unsigned int pool_size = (type == DM_TYPE_BIO_BASED) ? 16 : MIN_IOS;
2728 per_bio_data_size = roundup(per_bio_data_size, __alignof__(struct dm_target_io));
2730 pools->io_pool = (type == DM_TYPE_BIO_BASED) ?
2731 mempool_create_slab_pool(MIN_IOS, _io_cache) :
2732 mempool_create_slab_pool(MIN_IOS, _rq_bio_info_cache);
2733 if (!pools->io_pool)
2734 goto free_pools_and_out;
2736 pools->tio_pool = NULL;
2737 if (type == DM_TYPE_REQUEST_BASED) {
2738 pools->tio_pool = mempool_create_slab_pool(MIN_IOS, _rq_tio_cache);
2739 if (!pools->tio_pool)
2740 goto free_io_pool_and_out;
2743 pools->bs = (type == DM_TYPE_BIO_BASED) ?
2744 bioset_create(pool_size,
2745 per_bio_data_size + offsetof(struct dm_target_io, clone)) :
2746 bioset_create(pool_size,
2747 offsetof(struct dm_rq_clone_bio_info, clone));
2749 goto free_tio_pool_and_out;
2751 if (integrity && bioset_integrity_create(pools->bs, pool_size))
2752 goto free_bioset_and_out;
2756 free_bioset_and_out:
2757 bioset_free(pools->bs);
2759 free_tio_pool_and_out:
2760 if (pools->tio_pool)
2761 mempool_destroy(pools->tio_pool);
2763 free_io_pool_and_out:
2764 mempool_destroy(pools->io_pool);
2772 void dm_free_md_mempools(struct dm_md_mempools *pools)
2778 mempool_destroy(pools->io_pool);
2780 if (pools->tio_pool)
2781 mempool_destroy(pools->tio_pool);
2784 bioset_free(pools->bs);
2789 static const struct block_device_operations dm_blk_dops = {
2790 .open = dm_blk_open,
2791 .release = dm_blk_close,
2792 .ioctl = dm_blk_ioctl,
2793 .getgeo = dm_blk_getgeo,
2794 .owner = THIS_MODULE
2797 EXPORT_SYMBOL(dm_get_mapinfo);
2802 module_init(dm_init);
2803 module_exit(dm_exit);
2805 module_param(major, uint, 0);
2806 MODULE_PARM_DESC(major, "The major number of the device mapper");
2807 MODULE_DESCRIPTION(DM_NAME " driver");
2808 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
2809 MODULE_LICENSE("GPL");