2 * Copyright (C) 2001, 2002 Sistina Software (UK) Limited.
3 * Copyright (C) 2004-2008 Red Hat, Inc. All rights reserved.
5 * This file is released under the GPL.
11 #include <linux/init.h>
12 #include <linux/module.h>
13 #include <linux/mutex.h>
14 #include <linux/moduleparam.h>
15 #include <linux/blkpg.h>
16 #include <linux/bio.h>
17 #include <linux/mempool.h>
18 #include <linux/slab.h>
19 #include <linux/idr.h>
20 #include <linux/hdreg.h>
21 #include <linux/delay.h>
23 #include <trace/events/block.h>
25 #define DM_MSG_PREFIX "core"
29 * ratelimit state to be used in DMXXX_LIMIT().
31 DEFINE_RATELIMIT_STATE(dm_ratelimit_state,
32 DEFAULT_RATELIMIT_INTERVAL,
33 DEFAULT_RATELIMIT_BURST);
34 EXPORT_SYMBOL(dm_ratelimit_state);
38 * Cookies are numeric values sent with CHANGE and REMOVE
39 * uevents while resuming, removing or renaming the device.
41 #define DM_COOKIE_ENV_VAR_NAME "DM_COOKIE"
42 #define DM_COOKIE_LENGTH 24
44 static const char *_name = DM_NAME;
46 static unsigned int major = 0;
47 static unsigned int _major = 0;
49 static DEFINE_IDR(_minor_idr);
51 static DEFINE_SPINLOCK(_minor_lock);
54 * One of these is allocated per bio.
57 struct mapped_device *md;
61 unsigned long start_time;
62 spinlock_t endio_lock;
67 * One of these is allocated per target within a bio. Hopefully
68 * this will be simplified out one day.
77 * For request-based dm.
78 * One of these is allocated per request.
80 struct dm_rq_target_io {
81 struct mapped_device *md;
83 struct request *orig, clone;
89 * For request-based dm - the bio clones we allocate are embedded in these
92 * We allocate these with bio_alloc_bioset, using the front_pad parameter when
93 * the bioset is created - this means the bio has to come at the end of the
96 struct dm_rq_clone_bio_info {
98 struct dm_rq_target_io *tio;
102 union map_info *dm_get_mapinfo(struct bio *bio)
104 if (bio && bio->bi_private)
105 return &((struct dm_target_io *)bio->bi_private)->info;
109 union map_info *dm_get_rq_mapinfo(struct request *rq)
111 if (rq && rq->end_io_data)
112 return &((struct dm_rq_target_io *)rq->end_io_data)->info;
115 EXPORT_SYMBOL_GPL(dm_get_rq_mapinfo);
117 #define MINOR_ALLOCED ((void *)-1)
120 * Bits for the md->flags field.
122 #define DMF_BLOCK_IO_FOR_SUSPEND 0
123 #define DMF_SUSPENDED 1
125 #define DMF_FREEING 3
126 #define DMF_DELETING 4
127 #define DMF_NOFLUSH_SUSPENDING 5
128 #define DMF_MERGE_IS_OPTIONAL 6
131 * Work processed by per-device workqueue.
133 struct mapped_device {
134 struct rw_semaphore io_lock;
135 struct mutex suspend_lock;
142 struct request_queue *queue;
144 /* Protect queue and type against concurrent access. */
145 struct mutex type_lock;
147 struct target_type *immutable_target_type;
149 struct gendisk *disk;
155 * A list of ios that arrived while we were suspended.
158 wait_queue_head_t wait;
159 struct work_struct work;
160 struct bio_list deferred;
161 spinlock_t deferred_lock;
164 * Processing queue (flush)
166 struct workqueue_struct *wq;
169 * The current mapping.
171 struct dm_table *map;
174 * io objects are allocated from here.
185 wait_queue_head_t eventq;
187 struct list_head uevent_list;
188 spinlock_t uevent_lock; /* Protect access to uevent_list */
191 * freeze/thaw support require holding onto a super block
193 struct super_block *frozen_sb;
194 struct block_device *bdev;
196 /* forced geometry settings */
197 struct hd_geometry geometry;
202 /* zero-length flush that will be cloned and submitted to targets */
203 struct bio flush_bio;
207 * For mempools pre-allocation at the table loading time.
209 struct dm_md_mempools {
216 static struct kmem_cache *_io_cache;
217 static struct kmem_cache *_tio_cache;
218 static struct kmem_cache *_rq_tio_cache;
221 * Unused now, and needs to be deleted. But since io_pool is overloaded and it's
222 * still used for _io_cache, I'm leaving this for a later cleanup
224 static struct kmem_cache *_rq_bio_info_cache;
226 static int __init local_init(void)
230 /* allocate a slab for the dm_ios */
231 _io_cache = KMEM_CACHE(dm_io, 0);
235 /* allocate a slab for the target ios */
236 _tio_cache = KMEM_CACHE(dm_target_io, 0);
238 goto out_free_io_cache;
240 _rq_tio_cache = KMEM_CACHE(dm_rq_target_io, 0);
242 goto out_free_tio_cache;
244 _rq_bio_info_cache = KMEM_CACHE(dm_rq_clone_bio_info, 0);
245 if (!_rq_bio_info_cache)
246 goto out_free_rq_tio_cache;
248 r = dm_uevent_init();
250 goto out_free_rq_bio_info_cache;
253 r = register_blkdev(_major, _name);
255 goto out_uevent_exit;
264 out_free_rq_bio_info_cache:
265 kmem_cache_destroy(_rq_bio_info_cache);
266 out_free_rq_tio_cache:
267 kmem_cache_destroy(_rq_tio_cache);
269 kmem_cache_destroy(_tio_cache);
271 kmem_cache_destroy(_io_cache);
276 static void local_exit(void)
278 kmem_cache_destroy(_rq_bio_info_cache);
279 kmem_cache_destroy(_rq_tio_cache);
280 kmem_cache_destroy(_tio_cache);
281 kmem_cache_destroy(_io_cache);
282 unregister_blkdev(_major, _name);
287 DMINFO("cleaned up");
290 static int (*_inits[])(void) __initdata = {
300 static void (*_exits[])(void) = {
310 static int __init dm_init(void)
312 const int count = ARRAY_SIZE(_inits);
316 for (i = 0; i < count; i++) {
331 static void __exit dm_exit(void)
333 int i = ARRAY_SIZE(_exits);
339 * Should be empty by this point.
341 idr_remove_all(&_minor_idr);
342 idr_destroy(&_minor_idr);
346 * Block device functions
348 int dm_deleting_md(struct mapped_device *md)
350 return test_bit(DMF_DELETING, &md->flags);
353 static int dm_blk_open(struct block_device *bdev, fmode_t mode)
355 struct mapped_device *md;
357 spin_lock(&_minor_lock);
359 md = bdev->bd_disk->private_data;
363 if (test_bit(DMF_FREEING, &md->flags) ||
364 dm_deleting_md(md)) {
370 atomic_inc(&md->open_count);
373 spin_unlock(&_minor_lock);
375 return md ? 0 : -ENXIO;
378 static int dm_blk_close(struct gendisk *disk, fmode_t mode)
380 struct mapped_device *md = disk->private_data;
382 spin_lock(&_minor_lock);
384 atomic_dec(&md->open_count);
387 spin_unlock(&_minor_lock);
392 int dm_open_count(struct mapped_device *md)
394 return atomic_read(&md->open_count);
398 * Guarantees nothing is using the device before it's deleted.
400 int dm_lock_for_deletion(struct mapped_device *md)
404 spin_lock(&_minor_lock);
406 if (dm_open_count(md))
409 set_bit(DMF_DELETING, &md->flags);
411 spin_unlock(&_minor_lock);
416 static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo)
418 struct mapped_device *md = bdev->bd_disk->private_data;
420 return dm_get_geometry(md, geo);
423 static int dm_blk_ioctl(struct block_device *bdev, fmode_t mode,
424 unsigned int cmd, unsigned long arg)
426 struct mapped_device *md = bdev->bd_disk->private_data;
427 struct dm_table *map = dm_get_live_table(md);
428 struct dm_target *tgt;
431 if (!map || !dm_table_get_size(map))
434 /* We only support devices that have a single target */
435 if (dm_table_get_num_targets(map) != 1)
438 tgt = dm_table_get_target(map, 0);
440 if (dm_suspended_md(md)) {
445 if (tgt->type->ioctl)
446 r = tgt->type->ioctl(tgt, cmd, arg);
454 static struct dm_io *alloc_io(struct mapped_device *md)
456 return mempool_alloc(md->io_pool, GFP_NOIO);
459 static void free_io(struct mapped_device *md, struct dm_io *io)
461 mempool_free(io, md->io_pool);
464 static void free_tio(struct mapped_device *md, struct dm_target_io *tio)
466 mempool_free(tio, md->tio_pool);
469 static struct dm_rq_target_io *alloc_rq_tio(struct mapped_device *md,
472 return mempool_alloc(md->tio_pool, gfp_mask);
475 static void free_rq_tio(struct dm_rq_target_io *tio)
477 mempool_free(tio, tio->md->tio_pool);
480 static int md_in_flight(struct mapped_device *md)
482 return atomic_read(&md->pending[READ]) +
483 atomic_read(&md->pending[WRITE]);
486 static void start_io_acct(struct dm_io *io)
488 struct mapped_device *md = io->md;
490 int rw = bio_data_dir(io->bio);
492 io->start_time = jiffies;
494 cpu = part_stat_lock();
495 part_round_stats(cpu, &dm_disk(md)->part0);
497 atomic_set(&dm_disk(md)->part0.in_flight[rw],
498 atomic_inc_return(&md->pending[rw]));
501 static void end_io_acct(struct dm_io *io)
503 struct mapped_device *md = io->md;
504 struct bio *bio = io->bio;
505 unsigned long duration = jiffies - io->start_time;
507 int rw = bio_data_dir(bio);
509 cpu = part_stat_lock();
510 part_round_stats(cpu, &dm_disk(md)->part0);
511 part_stat_add(cpu, &dm_disk(md)->part0, ticks[rw], duration);
515 * After this is decremented the bio must not be touched if it is
518 pending = atomic_dec_return(&md->pending[rw]);
519 atomic_set(&dm_disk(md)->part0.in_flight[rw], pending);
520 pending += atomic_read(&md->pending[rw^0x1]);
522 /* nudge anyone waiting on suspend queue */
528 * Add the bio to the list of deferred io.
530 static void queue_io(struct mapped_device *md, struct bio *bio)
534 spin_lock_irqsave(&md->deferred_lock, flags);
535 bio_list_add(&md->deferred, bio);
536 spin_unlock_irqrestore(&md->deferred_lock, flags);
537 queue_work(md->wq, &md->work);
541 * Everyone (including functions in this file), should use this
542 * function to access the md->map field, and make sure they call
543 * dm_table_put() when finished.
545 struct dm_table *dm_get_live_table(struct mapped_device *md)
550 read_lock_irqsave(&md->map_lock, flags);
554 read_unlock_irqrestore(&md->map_lock, flags);
560 * Get the geometry associated with a dm device
562 int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo)
570 * Set the geometry of a device.
572 int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo)
574 sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors;
576 if (geo->start > sz) {
577 DMWARN("Start sector is beyond the geometry limits.");
586 /*-----------------------------------------------------------------
588 * A more elegant soln is in the works that uses the queue
589 * merge fn, unfortunately there are a couple of changes to
590 * the block layer that I want to make for this. So in the
591 * interests of getting something for people to use I give
592 * you this clearly demarcated crap.
593 *---------------------------------------------------------------*/
595 static int __noflush_suspending(struct mapped_device *md)
597 return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
601 * Decrements the number of outstanding ios that a bio has been
602 * cloned into, completing the original io if necc.
604 static void dec_pending(struct dm_io *io, int error)
609 struct mapped_device *md = io->md;
611 /* Push-back supersedes any I/O errors */
612 if (unlikely(error)) {
613 spin_lock_irqsave(&io->endio_lock, flags);
614 if (!(io->error > 0 && __noflush_suspending(md)))
616 spin_unlock_irqrestore(&io->endio_lock, flags);
619 if (atomic_dec_and_test(&io->io_count)) {
620 if (io->error == DM_ENDIO_REQUEUE) {
622 * Target requested pushing back the I/O.
624 spin_lock_irqsave(&md->deferred_lock, flags);
625 if (__noflush_suspending(md))
626 bio_list_add_head(&md->deferred, io->bio);
628 /* noflush suspend was interrupted. */
630 spin_unlock_irqrestore(&md->deferred_lock, flags);
633 io_error = io->error;
638 if (io_error == DM_ENDIO_REQUEUE)
641 if ((bio->bi_rw & REQ_FLUSH) && bio->bi_size) {
643 * Preflush done for flush with data, reissue
646 bio->bi_rw &= ~REQ_FLUSH;
649 /* done with normal IO or empty flush */
650 trace_block_bio_complete(md->queue, bio, io_error);
651 bio_endio(bio, io_error);
656 static void clone_endio(struct bio *bio, int error)
659 struct dm_target_io *tio = bio->bi_private;
660 struct dm_io *io = tio->io;
661 struct mapped_device *md = tio->io->md;
662 dm_endio_fn endio = tio->ti->type->end_io;
664 if (!bio_flagged(bio, BIO_UPTODATE) && !error)
668 r = endio(tio->ti, bio, error, &tio->info);
669 if (r < 0 || r == DM_ENDIO_REQUEUE)
671 * error and requeue request are handled
675 else if (r == DM_ENDIO_INCOMPLETE)
676 /* The target will handle the io */
679 DMWARN("unimplemented target endio return value: %d", r);
686 dec_pending(io, error);
690 * Partial completion handling for request-based dm
692 static void end_clone_bio(struct bio *clone, int error)
694 struct dm_rq_clone_bio_info *info = clone->bi_private;
695 struct dm_rq_target_io *tio = info->tio;
696 struct bio *bio = info->orig;
697 unsigned int nr_bytes = info->orig->bi_size;
703 * An error has already been detected on the request.
704 * Once error occurred, just let clone->end_io() handle
710 * Don't notice the error to the upper layer yet.
711 * The error handling decision is made by the target driver,
712 * when the request is completed.
719 * I/O for the bio successfully completed.
720 * Notice the data completion to the upper layer.
724 * bios are processed from the head of the list.
725 * So the completing bio should always be rq->bio.
726 * If it's not, something wrong is happening.
728 if (tio->orig->bio != bio)
729 DMERR("bio completion is going in the middle of the request");
732 * Update the original request.
733 * Do not use blk_end_request() here, because it may complete
734 * the original request before the clone, and break the ordering.
736 blk_update_request(tio->orig, 0, nr_bytes);
740 * Don't touch any member of the md after calling this function because
741 * the md may be freed in dm_put() at the end of this function.
742 * Or do dm_get() before calling this function and dm_put() later.
744 static void rq_completed(struct mapped_device *md, int rw, int run_queue)
746 atomic_dec(&md->pending[rw]);
748 /* nudge anyone waiting on suspend queue */
749 if (!md_in_flight(md))
753 blk_run_queue(md->queue);
756 * dm_put() must be at the end of this function. See the comment above
761 static void free_rq_clone(struct request *clone)
763 struct dm_rq_target_io *tio = clone->end_io_data;
765 blk_rq_unprep_clone(clone);
770 * Complete the clone and the original request.
771 * Must be called without queue lock.
773 static void dm_end_request(struct request *clone, int error)
775 int rw = rq_data_dir(clone);
776 struct dm_rq_target_io *tio = clone->end_io_data;
777 struct mapped_device *md = tio->md;
778 struct request *rq = tio->orig;
780 if (rq->cmd_type == REQ_TYPE_BLOCK_PC) {
781 rq->errors = clone->errors;
782 rq->resid_len = clone->resid_len;
786 * We are using the sense buffer of the original
788 * So setting the length of the sense data is enough.
790 rq->sense_len = clone->sense_len;
793 free_rq_clone(clone);
794 blk_end_request_all(rq, error);
795 rq_completed(md, rw, true);
798 static void dm_unprep_request(struct request *rq)
800 struct request *clone = rq->special;
803 rq->cmd_flags &= ~REQ_DONTPREP;
805 free_rq_clone(clone);
809 * Requeue the original request of a clone.
811 void dm_requeue_unmapped_request(struct request *clone)
813 int rw = rq_data_dir(clone);
814 struct dm_rq_target_io *tio = clone->end_io_data;
815 struct mapped_device *md = tio->md;
816 struct request *rq = tio->orig;
817 struct request_queue *q = rq->q;
820 dm_unprep_request(rq);
822 spin_lock_irqsave(q->queue_lock, flags);
823 blk_requeue_request(q, rq);
824 spin_unlock_irqrestore(q->queue_lock, flags);
826 rq_completed(md, rw, 0);
828 EXPORT_SYMBOL_GPL(dm_requeue_unmapped_request);
830 static void __stop_queue(struct request_queue *q)
835 static void stop_queue(struct request_queue *q)
839 spin_lock_irqsave(q->queue_lock, flags);
841 spin_unlock_irqrestore(q->queue_lock, flags);
844 static void __start_queue(struct request_queue *q)
846 if (blk_queue_stopped(q))
850 static void start_queue(struct request_queue *q)
854 spin_lock_irqsave(q->queue_lock, flags);
856 spin_unlock_irqrestore(q->queue_lock, flags);
859 static void dm_done(struct request *clone, int error, bool mapped)
862 struct dm_rq_target_io *tio = clone->end_io_data;
863 dm_request_endio_fn rq_end_io = NULL;
866 rq_end_io = tio->ti->type->rq_end_io;
868 if (mapped && rq_end_io)
869 r = rq_end_io(tio->ti, clone, error, &tio->info);
873 /* The target wants to complete the I/O */
874 dm_end_request(clone, r);
875 else if (r == DM_ENDIO_INCOMPLETE)
876 /* The target will handle the I/O */
878 else if (r == DM_ENDIO_REQUEUE)
879 /* The target wants to requeue the I/O */
880 dm_requeue_unmapped_request(clone);
882 DMWARN("unimplemented target endio return value: %d", r);
888 * Request completion handler for request-based dm
890 static void dm_softirq_done(struct request *rq)
893 struct request *clone = rq->completion_data;
894 struct dm_rq_target_io *tio = clone->end_io_data;
896 if (rq->cmd_flags & REQ_FAILED)
899 dm_done(clone, tio->error, mapped);
903 * Complete the clone and the original request with the error status
904 * through softirq context.
906 static void dm_complete_request(struct request *clone, int error)
908 struct dm_rq_target_io *tio = clone->end_io_data;
909 struct request *rq = tio->orig;
912 rq->completion_data = clone;
913 blk_complete_request(rq);
917 * Complete the not-mapped clone and the original request with the error status
918 * through softirq context.
919 * Target's rq_end_io() function isn't called.
920 * This may be used when the target's map_rq() function fails.
922 void dm_kill_unmapped_request(struct request *clone, int error)
924 struct dm_rq_target_io *tio = clone->end_io_data;
925 struct request *rq = tio->orig;
927 rq->cmd_flags |= REQ_FAILED;
928 dm_complete_request(clone, error);
930 EXPORT_SYMBOL_GPL(dm_kill_unmapped_request);
933 * Called with the queue lock held
935 static void end_clone_request(struct request *clone, int error)
938 * For just cleaning up the information of the queue in which
939 * the clone was dispatched.
940 * The clone is *NOT* freed actually here because it is alloced from
941 * dm own mempool and REQ_ALLOCED isn't set in clone->cmd_flags.
943 __blk_put_request(clone->q, clone);
946 * Actual request completion is done in a softirq context which doesn't
947 * hold the queue lock. Otherwise, deadlock could occur because:
948 * - another request may be submitted by the upper level driver
949 * of the stacking during the completion
950 * - the submission which requires queue lock may be done
953 dm_complete_request(clone, error);
957 * Return maximum size of I/O possible at the supplied sector up to the current
960 static sector_t max_io_len_target_boundary(sector_t sector, struct dm_target *ti)
962 sector_t target_offset = dm_target_offset(ti, sector);
964 return ti->len - target_offset;
967 static sector_t max_io_len(sector_t sector, struct dm_target *ti)
969 sector_t len = max_io_len_target_boundary(sector, ti);
970 sector_t offset, max_len;
973 * Does the target need to split even further?
975 if (ti->max_io_len) {
976 offset = dm_target_offset(ti, sector);
977 if (unlikely(ti->max_io_len & (ti->max_io_len - 1)))
978 max_len = sector_div(offset, ti->max_io_len);
980 max_len = offset & (ti->max_io_len - 1);
981 max_len = ti->max_io_len - max_len;
990 int dm_set_target_max_io_len(struct dm_target *ti, sector_t len)
992 if (len > UINT_MAX) {
993 DMERR("Specified maximum size of target IO (%llu) exceeds limit (%u)",
994 (unsigned long long)len, UINT_MAX);
995 ti->error = "Maximum size of target IO is too large";
999 ti->max_io_len = (uint32_t) len;
1003 EXPORT_SYMBOL_GPL(dm_set_target_max_io_len);
1005 static void __map_bio(struct dm_target *ti, struct bio *clone,
1006 struct dm_target_io *tio)
1010 struct mapped_device *md;
1012 clone->bi_end_io = clone_endio;
1013 clone->bi_private = tio;
1016 * Map the clone. If r == 0 we don't need to do
1017 * anything, the target has assumed ownership of
1020 atomic_inc(&tio->io->io_count);
1021 sector = clone->bi_sector;
1022 r = ti->type->map(ti, clone, &tio->info);
1023 if (r == DM_MAPIO_REMAPPED) {
1024 /* the bio has been remapped so dispatch it */
1026 trace_block_bio_remap(bdev_get_queue(clone->bi_bdev), clone,
1027 tio->io->bio->bi_bdev->bd_dev, sector);
1029 generic_make_request(clone);
1030 } else if (r < 0 || r == DM_MAPIO_REQUEUE) {
1031 /* error the io and bail out, or requeue it if needed */
1033 dec_pending(tio->io, r);
1037 DMWARN("unimplemented target map return value: %d", r);
1043 struct mapped_device *md;
1044 struct dm_table *map;
1048 sector_t sector_count;
1053 * Creates a little bio that just does part of a bvec.
1055 static struct bio *split_bvec(struct bio *bio, sector_t sector,
1056 unsigned short idx, unsigned int offset,
1057 unsigned int len, struct bio_set *bs)
1060 struct bio_vec *bv = bio->bi_io_vec + idx;
1062 clone = bio_alloc_bioset(GFP_NOIO, 1, bs);
1063 *clone->bi_io_vec = *bv;
1065 clone->bi_sector = sector;
1066 clone->bi_bdev = bio->bi_bdev;
1067 clone->bi_rw = bio->bi_rw;
1069 clone->bi_size = to_bytes(len);
1070 clone->bi_io_vec->bv_offset = offset;
1071 clone->bi_io_vec->bv_len = clone->bi_size;
1072 clone->bi_flags |= 1 << BIO_CLONED;
1074 if (bio_integrity(bio)) {
1075 bio_integrity_clone(clone, bio, GFP_NOIO);
1076 bio_integrity_trim(clone,
1077 bio_sector_offset(bio, idx, offset), len);
1084 * Creates a bio that consists of range of complete bvecs.
1086 static struct bio *clone_bio(struct bio *bio, sector_t sector,
1087 unsigned short idx, unsigned short bv_count,
1088 unsigned int len, struct bio_set *bs)
1092 clone = bio_alloc_bioset(GFP_NOIO, bio->bi_max_vecs, bs);
1093 __bio_clone(clone, bio);
1094 clone->bi_sector = sector;
1095 clone->bi_idx = idx;
1096 clone->bi_vcnt = idx + bv_count;
1097 clone->bi_size = to_bytes(len);
1098 clone->bi_flags &= ~(1 << BIO_SEG_VALID);
1100 if (bio_integrity(bio)) {
1101 bio_integrity_clone(clone, bio, GFP_NOIO);
1103 if (idx != bio->bi_idx || clone->bi_size < bio->bi_size)
1104 bio_integrity_trim(clone,
1105 bio_sector_offset(bio, idx, 0), len);
1111 static struct dm_target_io *alloc_tio(struct clone_info *ci,
1112 struct dm_target *ti)
1114 struct dm_target_io *tio = mempool_alloc(ci->md->tio_pool, GFP_NOIO);
1118 memset(&tio->info, 0, sizeof(tio->info));
1123 static void __issue_target_request(struct clone_info *ci, struct dm_target *ti,
1124 unsigned request_nr, sector_t len)
1126 struct dm_target_io *tio = alloc_tio(ci, ti);
1129 tio->info.target_request_nr = request_nr;
1132 * Discard requests require the bio's inline iovecs be initialized.
1133 * ci->bio->bi_max_vecs is BIO_INLINE_VECS anyway, for both flush
1134 * and discard, so no need for concern about wasted bvec allocations.
1136 clone = bio_clone_bioset(ci->bio, GFP_NOIO, ci->md->bs);
1139 clone->bi_sector = ci->sector;
1140 clone->bi_size = to_bytes(len);
1143 __map_bio(ti, clone, tio);
1146 static void __issue_target_requests(struct clone_info *ci, struct dm_target *ti,
1147 unsigned num_requests, sector_t len)
1149 unsigned request_nr;
1151 for (request_nr = 0; request_nr < num_requests; request_nr++)
1152 __issue_target_request(ci, ti, request_nr, len);
1155 static int __clone_and_map_empty_flush(struct clone_info *ci)
1157 unsigned target_nr = 0;
1158 struct dm_target *ti;
1160 BUG_ON(bio_has_data(ci->bio));
1161 while ((ti = dm_table_get_target(ci->map, target_nr++)))
1162 __issue_target_requests(ci, ti, ti->num_flush_requests, 0);
1168 * Perform all io with a single clone.
1170 static void __clone_and_map_simple(struct clone_info *ci, struct dm_target *ti)
1172 struct bio *clone, *bio = ci->bio;
1173 struct dm_target_io *tio;
1175 tio = alloc_tio(ci, ti);
1176 clone = clone_bio(bio, ci->sector, ci->idx,
1177 bio->bi_vcnt - ci->idx, ci->sector_count,
1179 __map_bio(ti, clone, tio);
1180 ci->sector_count = 0;
1183 static int __clone_and_map_discard(struct clone_info *ci)
1185 struct dm_target *ti;
1189 ti = dm_table_find_target(ci->map, ci->sector);
1190 if (!dm_target_is_valid(ti))
1194 * Even though the device advertised discard support,
1195 * that does not mean every target supports it, and
1196 * reconfiguration might also have changed that since the
1197 * check was performed.
1199 if (!ti->num_discard_requests)
1202 if (!ti->split_discard_requests)
1203 len = min(ci->sector_count, max_io_len_target_boundary(ci->sector, ti));
1205 len = min(ci->sector_count, max_io_len(ci->sector, ti));
1207 __issue_target_requests(ci, ti, ti->num_discard_requests, len);
1210 } while (ci->sector_count -= len);
1215 static int __clone_and_map(struct clone_info *ci)
1217 struct bio *clone, *bio = ci->bio;
1218 struct dm_target *ti;
1219 sector_t len = 0, max;
1220 struct dm_target_io *tio;
1222 if (unlikely(bio->bi_rw & REQ_DISCARD))
1223 return __clone_and_map_discard(ci);
1225 ti = dm_table_find_target(ci->map, ci->sector);
1226 if (!dm_target_is_valid(ti))
1229 max = max_io_len(ci->sector, ti);
1231 if (ci->sector_count <= max) {
1233 * Optimise for the simple case where we can do all of
1234 * the remaining io with a single clone.
1236 __clone_and_map_simple(ci, ti);
1238 } else if (to_sector(bio->bi_io_vec[ci->idx].bv_len) <= max) {
1240 * There are some bvecs that don't span targets.
1241 * Do as many of these as possible.
1244 sector_t remaining = max;
1247 for (i = ci->idx; remaining && (i < bio->bi_vcnt); i++) {
1248 bv_len = to_sector(bio->bi_io_vec[i].bv_len);
1250 if (bv_len > remaining)
1253 remaining -= bv_len;
1257 tio = alloc_tio(ci, ti);
1258 clone = clone_bio(bio, ci->sector, ci->idx, i - ci->idx, len,
1260 __map_bio(ti, clone, tio);
1263 ci->sector_count -= len;
1268 * Handle a bvec that must be split between two or more targets.
1270 struct bio_vec *bv = bio->bi_io_vec + ci->idx;
1271 sector_t remaining = to_sector(bv->bv_len);
1272 unsigned int offset = 0;
1276 ti = dm_table_find_target(ci->map, ci->sector);
1277 if (!dm_target_is_valid(ti))
1280 max = max_io_len(ci->sector, ti);
1283 len = min(remaining, max);
1285 tio = alloc_tio(ci, ti);
1286 clone = split_bvec(bio, ci->sector, ci->idx,
1287 bv->bv_offset + offset, len,
1290 __map_bio(ti, clone, tio);
1293 ci->sector_count -= len;
1294 offset += to_bytes(len);
1295 } while (remaining -= len);
1304 * Split the bio into several clones and submit it to targets.
1306 static void __split_and_process_bio(struct mapped_device *md, struct bio *bio)
1308 struct clone_info ci;
1311 ci.map = dm_get_live_table(md);
1312 if (unlikely(!ci.map)) {
1318 ci.io = alloc_io(md);
1320 atomic_set(&ci.io->io_count, 1);
1323 spin_lock_init(&ci.io->endio_lock);
1324 ci.sector = bio->bi_sector;
1325 ci.idx = bio->bi_idx;
1327 start_io_acct(ci.io);
1328 if (bio->bi_rw & REQ_FLUSH) {
1329 ci.bio = &ci.md->flush_bio;
1330 ci.sector_count = 0;
1331 error = __clone_and_map_empty_flush(&ci);
1332 /* dec_pending submits any data associated with flush */
1335 ci.sector_count = bio_sectors(bio);
1336 while (ci.sector_count && !error)
1337 error = __clone_and_map(&ci);
1340 /* drop the extra reference count */
1341 dec_pending(ci.io, error);
1342 dm_table_put(ci.map);
1344 /*-----------------------------------------------------------------
1346 *---------------------------------------------------------------*/
1348 static int dm_merge_bvec(struct request_queue *q,
1349 struct bvec_merge_data *bvm,
1350 struct bio_vec *biovec)
1352 struct mapped_device *md = q->queuedata;
1353 struct dm_table *map = dm_get_live_table(md);
1354 struct dm_target *ti;
1355 sector_t max_sectors;
1361 ti = dm_table_find_target(map, bvm->bi_sector);
1362 if (!dm_target_is_valid(ti))
1366 * Find maximum amount of I/O that won't need splitting
1368 max_sectors = min(max_io_len(bvm->bi_sector, ti),
1369 (sector_t) BIO_MAX_SECTORS);
1370 max_size = (max_sectors << SECTOR_SHIFT) - bvm->bi_size;
1375 * merge_bvec_fn() returns number of bytes
1376 * it can accept at this offset
1377 * max is precomputed maximal io size
1379 if (max_size && ti->type->merge)
1380 max_size = ti->type->merge(ti, bvm, biovec, max_size);
1382 * If the target doesn't support merge method and some of the devices
1383 * provided their merge_bvec method (we know this by looking at
1384 * queue_max_hw_sectors), then we can't allow bios with multiple vector
1385 * entries. So always set max_size to 0, and the code below allows
1388 else if (queue_max_hw_sectors(q) <= PAGE_SIZE >> 9)
1397 * Always allow an entire first page
1399 if (max_size <= biovec->bv_len && !(bvm->bi_size >> SECTOR_SHIFT))
1400 max_size = biovec->bv_len;
1406 * The request function that just remaps the bio built up by
1409 static void _dm_request(struct request_queue *q, struct bio *bio)
1411 int rw = bio_data_dir(bio);
1412 struct mapped_device *md = q->queuedata;
1415 down_read(&md->io_lock);
1417 cpu = part_stat_lock();
1418 part_stat_inc(cpu, &dm_disk(md)->part0, ios[rw]);
1419 part_stat_add(cpu, &dm_disk(md)->part0, sectors[rw], bio_sectors(bio));
1422 /* if we're suspended, we have to queue this io for later */
1423 if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))) {
1424 up_read(&md->io_lock);
1426 if (bio_rw(bio) != READA)
1433 __split_and_process_bio(md, bio);
1434 up_read(&md->io_lock);
1438 static int dm_request_based(struct mapped_device *md)
1440 return blk_queue_stackable(md->queue);
1443 static void dm_request(struct request_queue *q, struct bio *bio)
1445 struct mapped_device *md = q->queuedata;
1447 if (dm_request_based(md))
1448 blk_queue_bio(q, bio);
1450 _dm_request(q, bio);
1453 void dm_dispatch_request(struct request *rq)
1457 if (blk_queue_io_stat(rq->q))
1458 rq->cmd_flags |= REQ_IO_STAT;
1460 rq->start_time = jiffies;
1461 r = blk_insert_cloned_request(rq->q, rq);
1463 dm_complete_request(rq, r);
1465 EXPORT_SYMBOL_GPL(dm_dispatch_request);
1467 static int dm_rq_bio_constructor(struct bio *bio, struct bio *bio_orig,
1470 struct dm_rq_target_io *tio = data;
1471 struct dm_rq_clone_bio_info *info =
1472 container_of(bio, struct dm_rq_clone_bio_info, clone);
1474 info->orig = bio_orig;
1476 bio->bi_end_io = end_clone_bio;
1477 bio->bi_private = info;
1482 static int setup_clone(struct request *clone, struct request *rq,
1483 struct dm_rq_target_io *tio)
1487 r = blk_rq_prep_clone(clone, rq, tio->md->bs, GFP_ATOMIC,
1488 dm_rq_bio_constructor, tio);
1492 clone->cmd = rq->cmd;
1493 clone->cmd_len = rq->cmd_len;
1494 clone->sense = rq->sense;
1495 clone->buffer = rq->buffer;
1496 clone->end_io = end_clone_request;
1497 clone->end_io_data = tio;
1502 static struct request *clone_rq(struct request *rq, struct mapped_device *md,
1505 struct request *clone;
1506 struct dm_rq_target_io *tio;
1508 tio = alloc_rq_tio(md, gfp_mask);
1516 memset(&tio->info, 0, sizeof(tio->info));
1518 clone = &tio->clone;
1519 if (setup_clone(clone, rq, tio)) {
1529 * Called with the queue lock held.
1531 static int dm_prep_fn(struct request_queue *q, struct request *rq)
1533 struct mapped_device *md = q->queuedata;
1534 struct request *clone;
1536 if (unlikely(rq->special)) {
1537 DMWARN("Already has something in rq->special.");
1538 return BLKPREP_KILL;
1541 clone = clone_rq(rq, md, GFP_ATOMIC);
1543 return BLKPREP_DEFER;
1545 rq->special = clone;
1546 rq->cmd_flags |= REQ_DONTPREP;
1553 * 0 : the request has been processed (not requeued)
1554 * !0 : the request has been requeued
1556 static int map_request(struct dm_target *ti, struct request *clone,
1557 struct mapped_device *md)
1559 int r, requeued = 0;
1560 struct dm_rq_target_io *tio = clone->end_io_data;
1563 r = ti->type->map_rq(ti, clone, &tio->info);
1565 case DM_MAPIO_SUBMITTED:
1566 /* The target has taken the I/O to submit by itself later */
1568 case DM_MAPIO_REMAPPED:
1569 /* The target has remapped the I/O so dispatch it */
1570 trace_block_rq_remap(clone->q, clone, disk_devt(dm_disk(md)),
1571 blk_rq_pos(tio->orig));
1572 dm_dispatch_request(clone);
1574 case DM_MAPIO_REQUEUE:
1575 /* The target wants to requeue the I/O */
1576 dm_requeue_unmapped_request(clone);
1581 DMWARN("unimplemented target map return value: %d", r);
1585 /* The target wants to complete the I/O */
1586 dm_kill_unmapped_request(clone, r);
1593 static struct request *dm_start_request(struct mapped_device *md, struct request *orig)
1595 struct request *clone;
1597 blk_start_request(orig);
1598 clone = orig->special;
1599 atomic_inc(&md->pending[rq_data_dir(clone)]);
1602 * Hold the md reference here for the in-flight I/O.
1603 * We can't rely on the reference count by device opener,
1604 * because the device may be closed during the request completion
1605 * when all bios are completed.
1606 * See the comment in rq_completed() too.
1614 * q->request_fn for request-based dm.
1615 * Called with the queue lock held.
1617 static void dm_request_fn(struct request_queue *q)
1619 struct mapped_device *md = q->queuedata;
1620 struct dm_table *map = dm_get_live_table(md);
1621 struct dm_target *ti;
1622 struct request *rq, *clone;
1626 * For suspend, check blk_queue_stopped() and increment
1627 * ->pending within a single queue_lock not to increment the
1628 * number of in-flight I/Os after the queue is stopped in
1631 while (!blk_queue_stopped(q)) {
1632 rq = blk_peek_request(q);
1636 /* always use block 0 to find the target for flushes for now */
1638 if (!(rq->cmd_flags & REQ_FLUSH))
1639 pos = blk_rq_pos(rq);
1641 ti = dm_table_find_target(map, pos);
1642 if (!dm_target_is_valid(ti)) {
1644 * Must perform setup, that dm_done() requires,
1645 * before calling dm_kill_unmapped_request
1647 DMERR_LIMIT("request attempted access beyond the end of device");
1648 clone = dm_start_request(md, rq);
1649 dm_kill_unmapped_request(clone, -EIO);
1653 if (ti->type->busy && ti->type->busy(ti))
1656 clone = dm_start_request(md, rq);
1658 spin_unlock(q->queue_lock);
1659 if (map_request(ti, clone, md))
1662 BUG_ON(!irqs_disabled());
1663 spin_lock(q->queue_lock);
1669 BUG_ON(!irqs_disabled());
1670 spin_lock(q->queue_lock);
1673 blk_delay_queue(q, HZ / 10);
1678 int dm_underlying_device_busy(struct request_queue *q)
1680 return blk_lld_busy(q);
1682 EXPORT_SYMBOL_GPL(dm_underlying_device_busy);
1684 static int dm_lld_busy(struct request_queue *q)
1687 struct mapped_device *md = q->queuedata;
1688 struct dm_table *map = dm_get_live_table(md);
1690 if (!map || test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))
1693 r = dm_table_any_busy_target(map);
1700 static int dm_any_congested(void *congested_data, int bdi_bits)
1703 struct mapped_device *md = congested_data;
1704 struct dm_table *map;
1706 if (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
1707 map = dm_get_live_table(md);
1710 * Request-based dm cares about only own queue for
1711 * the query about congestion status of request_queue
1713 if (dm_request_based(md))
1714 r = md->queue->backing_dev_info.state &
1717 r = dm_table_any_congested(map, bdi_bits);
1726 /*-----------------------------------------------------------------
1727 * An IDR is used to keep track of allocated minor numbers.
1728 *---------------------------------------------------------------*/
1729 static void free_minor(int minor)
1731 spin_lock(&_minor_lock);
1732 idr_remove(&_minor_idr, minor);
1733 spin_unlock(&_minor_lock);
1737 * See if the device with a specific minor # is free.
1739 static int specific_minor(int minor)
1743 if (minor >= (1 << MINORBITS))
1746 r = idr_pre_get(&_minor_idr, GFP_KERNEL);
1750 spin_lock(&_minor_lock);
1752 if (idr_find(&_minor_idr, minor)) {
1757 r = idr_get_new_above(&_minor_idr, MINOR_ALLOCED, minor, &m);
1762 idr_remove(&_minor_idr, m);
1768 spin_unlock(&_minor_lock);
1772 static int next_free_minor(int *minor)
1776 r = idr_pre_get(&_minor_idr, GFP_KERNEL);
1780 spin_lock(&_minor_lock);
1782 r = idr_get_new(&_minor_idr, MINOR_ALLOCED, &m);
1786 if (m >= (1 << MINORBITS)) {
1787 idr_remove(&_minor_idr, m);
1795 spin_unlock(&_minor_lock);
1799 static const struct block_device_operations dm_blk_dops;
1801 static void dm_wq_work(struct work_struct *work);
1803 static void dm_init_md_queue(struct mapped_device *md)
1806 * Request-based dm devices cannot be stacked on top of bio-based dm
1807 * devices. The type of this dm device has not been decided yet.
1808 * The type is decided at the first table loading time.
1809 * To prevent problematic device stacking, clear the queue flag
1810 * for request stacking support until then.
1812 * This queue is new, so no concurrency on the queue_flags.
1814 queue_flag_clear_unlocked(QUEUE_FLAG_STACKABLE, md->queue);
1816 md->queue->queuedata = md;
1817 md->queue->backing_dev_info.congested_fn = dm_any_congested;
1818 md->queue->backing_dev_info.congested_data = md;
1819 blk_queue_make_request(md->queue, dm_request);
1820 blk_queue_bounce_limit(md->queue, BLK_BOUNCE_ANY);
1821 blk_queue_merge_bvec(md->queue, dm_merge_bvec);
1825 * Allocate and initialise a blank device with a given minor.
1827 static struct mapped_device *alloc_dev(int minor)
1830 struct mapped_device *md = kzalloc(sizeof(*md), GFP_KERNEL);
1834 DMWARN("unable to allocate device, out of memory.");
1838 if (!try_module_get(THIS_MODULE))
1839 goto bad_module_get;
1841 /* get a minor number for the dev */
1842 if (minor == DM_ANY_MINOR)
1843 r = next_free_minor(&minor);
1845 r = specific_minor(minor);
1849 md->type = DM_TYPE_NONE;
1850 init_rwsem(&md->io_lock);
1851 mutex_init(&md->suspend_lock);
1852 mutex_init(&md->type_lock);
1853 spin_lock_init(&md->deferred_lock);
1854 rwlock_init(&md->map_lock);
1855 atomic_set(&md->holders, 1);
1856 atomic_set(&md->open_count, 0);
1857 atomic_set(&md->event_nr, 0);
1858 atomic_set(&md->uevent_seq, 0);
1859 INIT_LIST_HEAD(&md->uevent_list);
1860 spin_lock_init(&md->uevent_lock);
1862 md->queue = blk_alloc_queue(GFP_KERNEL);
1866 dm_init_md_queue(md);
1868 md->disk = alloc_disk(1);
1872 atomic_set(&md->pending[0], 0);
1873 atomic_set(&md->pending[1], 0);
1874 init_waitqueue_head(&md->wait);
1875 INIT_WORK(&md->work, dm_wq_work);
1876 init_waitqueue_head(&md->eventq);
1878 md->disk->major = _major;
1879 md->disk->first_minor = minor;
1880 md->disk->fops = &dm_blk_dops;
1881 md->disk->queue = md->queue;
1882 md->disk->private_data = md;
1883 sprintf(md->disk->disk_name, "dm-%d", minor);
1885 format_dev_t(md->name, MKDEV(_major, minor));
1887 md->wq = alloc_workqueue("kdmflush",
1888 WQ_NON_REENTRANT | WQ_MEM_RECLAIM, 0);
1892 md->bdev = bdget_disk(md->disk, 0);
1896 bio_init(&md->flush_bio);
1897 md->flush_bio.bi_bdev = md->bdev;
1898 md->flush_bio.bi_rw = WRITE_FLUSH;
1900 /* Populate the mapping, nobody knows we exist yet */
1901 spin_lock(&_minor_lock);
1902 old_md = idr_replace(&_minor_idr, md, minor);
1903 spin_unlock(&_minor_lock);
1905 BUG_ON(old_md != MINOR_ALLOCED);
1910 destroy_workqueue(md->wq);
1912 del_gendisk(md->disk);
1915 blk_cleanup_queue(md->queue);
1919 module_put(THIS_MODULE);
1925 static void unlock_fs(struct mapped_device *md);
1927 static void free_dev(struct mapped_device *md)
1929 int minor = MINOR(disk_devt(md->disk));
1933 destroy_workqueue(md->wq);
1935 mempool_destroy(md->tio_pool);
1937 mempool_destroy(md->io_pool);
1939 bioset_free(md->bs);
1940 blk_integrity_unregister(md->disk);
1941 del_gendisk(md->disk);
1944 spin_lock(&_minor_lock);
1945 md->disk->private_data = NULL;
1946 spin_unlock(&_minor_lock);
1949 blk_cleanup_queue(md->queue);
1950 module_put(THIS_MODULE);
1954 static void __bind_mempools(struct mapped_device *md, struct dm_table *t)
1956 struct dm_md_mempools *p;
1958 if (md->io_pool && md->tio_pool && md->bs)
1959 /* the md already has necessary mempools */
1962 p = dm_table_get_md_mempools(t);
1963 BUG_ON(!p || md->io_pool || md->tio_pool || md->bs);
1965 md->io_pool = p->io_pool;
1967 md->tio_pool = p->tio_pool;
1973 /* mempool bind completed, now no need any mempools in the table */
1974 dm_table_free_md_mempools(t);
1978 * Bind a table to the device.
1980 static void event_callback(void *context)
1982 unsigned long flags;
1984 struct mapped_device *md = (struct mapped_device *) context;
1986 spin_lock_irqsave(&md->uevent_lock, flags);
1987 list_splice_init(&md->uevent_list, &uevents);
1988 spin_unlock_irqrestore(&md->uevent_lock, flags);
1990 dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj);
1992 atomic_inc(&md->event_nr);
1993 wake_up(&md->eventq);
1997 * Protected by md->suspend_lock obtained by dm_swap_table().
1999 static void __set_size(struct mapped_device *md, sector_t size)
2001 set_capacity(md->disk, size);
2003 i_size_write(md->bdev->bd_inode, (loff_t)size << SECTOR_SHIFT);
2007 * Return 1 if the queue has a compulsory merge_bvec_fn function.
2009 * If this function returns 0, then the device is either a non-dm
2010 * device without a merge_bvec_fn, or it is a dm device that is
2011 * able to split any bios it receives that are too big.
2013 int dm_queue_merge_is_compulsory(struct request_queue *q)
2015 struct mapped_device *dev_md;
2017 if (!q->merge_bvec_fn)
2020 if (q->make_request_fn == dm_request) {
2021 dev_md = q->queuedata;
2022 if (test_bit(DMF_MERGE_IS_OPTIONAL, &dev_md->flags))
2029 static int dm_device_merge_is_compulsory(struct dm_target *ti,
2030 struct dm_dev *dev, sector_t start,
2031 sector_t len, void *data)
2033 struct block_device *bdev = dev->bdev;
2034 struct request_queue *q = bdev_get_queue(bdev);
2036 return dm_queue_merge_is_compulsory(q);
2040 * Return 1 if it is acceptable to ignore merge_bvec_fn based
2041 * on the properties of the underlying devices.
2043 static int dm_table_merge_is_optional(struct dm_table *table)
2046 struct dm_target *ti;
2048 while (i < dm_table_get_num_targets(table)) {
2049 ti = dm_table_get_target(table, i++);
2051 if (ti->type->iterate_devices &&
2052 ti->type->iterate_devices(ti, dm_device_merge_is_compulsory, NULL))
2060 * Returns old map, which caller must destroy.
2062 static struct dm_table *__bind(struct mapped_device *md, struct dm_table *t,
2063 struct queue_limits *limits)
2065 struct dm_table *old_map;
2066 struct request_queue *q = md->queue;
2068 unsigned long flags;
2069 int merge_is_optional;
2071 size = dm_table_get_size(t);
2074 * Wipe any geometry if the size of the table changed.
2076 if (size != get_capacity(md->disk))
2077 memset(&md->geometry, 0, sizeof(md->geometry));
2079 __set_size(md, size);
2081 dm_table_event_callback(t, event_callback, md);
2084 * The queue hasn't been stopped yet, if the old table type wasn't
2085 * for request-based during suspension. So stop it to prevent
2086 * I/O mapping before resume.
2087 * This must be done before setting the queue restrictions,
2088 * because request-based dm may be run just after the setting.
2090 if (dm_table_request_based(t) && !blk_queue_stopped(q))
2093 __bind_mempools(md, t);
2095 merge_is_optional = dm_table_merge_is_optional(t);
2097 write_lock_irqsave(&md->map_lock, flags);
2100 md->immutable_target_type = dm_table_get_immutable_target_type(t);
2102 dm_table_set_restrictions(t, q, limits);
2103 if (merge_is_optional)
2104 set_bit(DMF_MERGE_IS_OPTIONAL, &md->flags);
2106 clear_bit(DMF_MERGE_IS_OPTIONAL, &md->flags);
2107 write_unlock_irqrestore(&md->map_lock, flags);
2113 * Returns unbound table for the caller to free.
2115 static struct dm_table *__unbind(struct mapped_device *md)
2117 struct dm_table *map = md->map;
2118 unsigned long flags;
2123 dm_table_event_callback(map, NULL, NULL);
2124 write_lock_irqsave(&md->map_lock, flags);
2126 write_unlock_irqrestore(&md->map_lock, flags);
2132 * Constructor for a new device.
2134 int dm_create(int minor, struct mapped_device **result)
2136 struct mapped_device *md;
2138 md = alloc_dev(minor);
2149 * Functions to manage md->type.
2150 * All are required to hold md->type_lock.
2152 void dm_lock_md_type(struct mapped_device *md)
2154 mutex_lock(&md->type_lock);
2157 void dm_unlock_md_type(struct mapped_device *md)
2159 mutex_unlock(&md->type_lock);
2162 void dm_set_md_type(struct mapped_device *md, unsigned type)
2167 unsigned dm_get_md_type(struct mapped_device *md)
2172 struct target_type *dm_get_immutable_target_type(struct mapped_device *md)
2174 return md->immutable_target_type;
2178 * Fully initialize a request-based queue (->elevator, ->request_fn, etc).
2180 static int dm_init_request_based_queue(struct mapped_device *md)
2182 struct request_queue *q = NULL;
2184 if (md->queue->elevator)
2187 /* Fully initialize the queue */
2188 q = blk_init_allocated_queue(md->queue, dm_request_fn, NULL);
2193 dm_init_md_queue(md);
2194 blk_queue_softirq_done(md->queue, dm_softirq_done);
2195 blk_queue_prep_rq(md->queue, dm_prep_fn);
2196 blk_queue_lld_busy(md->queue, dm_lld_busy);
2198 elv_register_queue(md->queue);
2204 * Setup the DM device's queue based on md's type
2206 int dm_setup_md_queue(struct mapped_device *md)
2208 if ((dm_get_md_type(md) == DM_TYPE_REQUEST_BASED) &&
2209 !dm_init_request_based_queue(md)) {
2210 DMWARN("Cannot initialize queue for request-based mapped device");
2217 static struct mapped_device *dm_find_md(dev_t dev)
2219 struct mapped_device *md;
2220 unsigned minor = MINOR(dev);
2222 if (MAJOR(dev) != _major || minor >= (1 << MINORBITS))
2225 spin_lock(&_minor_lock);
2227 md = idr_find(&_minor_idr, minor);
2228 if (md && (md == MINOR_ALLOCED ||
2229 (MINOR(disk_devt(dm_disk(md))) != minor) ||
2230 dm_deleting_md(md) ||
2231 test_bit(DMF_FREEING, &md->flags))) {
2237 spin_unlock(&_minor_lock);
2242 struct mapped_device *dm_get_md(dev_t dev)
2244 struct mapped_device *md = dm_find_md(dev);
2251 EXPORT_SYMBOL_GPL(dm_get_md);
2253 void *dm_get_mdptr(struct mapped_device *md)
2255 return md->interface_ptr;
2258 void dm_set_mdptr(struct mapped_device *md, void *ptr)
2260 md->interface_ptr = ptr;
2263 void dm_get(struct mapped_device *md)
2265 atomic_inc(&md->holders);
2266 BUG_ON(test_bit(DMF_FREEING, &md->flags));
2269 const char *dm_device_name(struct mapped_device *md)
2273 EXPORT_SYMBOL_GPL(dm_device_name);
2275 static void __dm_destroy(struct mapped_device *md, bool wait)
2277 struct dm_table *map;
2281 spin_lock(&_minor_lock);
2282 map = dm_get_live_table(md);
2283 idr_replace(&_minor_idr, MINOR_ALLOCED, MINOR(disk_devt(dm_disk(md))));
2284 set_bit(DMF_FREEING, &md->flags);
2285 spin_unlock(&_minor_lock);
2287 if (!dm_suspended_md(md)) {
2288 dm_table_presuspend_targets(map);
2289 dm_table_postsuspend_targets(map);
2293 * Rare, but there may be I/O requests still going to complete,
2294 * for example. Wait for all references to disappear.
2295 * No one should increment the reference count of the mapped_device,
2296 * after the mapped_device state becomes DMF_FREEING.
2299 while (atomic_read(&md->holders))
2301 else if (atomic_read(&md->holders))
2302 DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)",
2303 dm_device_name(md), atomic_read(&md->holders));
2307 dm_table_destroy(__unbind(md));
2311 void dm_destroy(struct mapped_device *md)
2313 __dm_destroy(md, true);
2316 void dm_destroy_immediate(struct mapped_device *md)
2318 __dm_destroy(md, false);
2321 void dm_put(struct mapped_device *md)
2323 atomic_dec(&md->holders);
2325 EXPORT_SYMBOL_GPL(dm_put);
2327 static int dm_wait_for_completion(struct mapped_device *md, int interruptible)
2330 DECLARE_WAITQUEUE(wait, current);
2332 add_wait_queue(&md->wait, &wait);
2335 set_current_state(interruptible);
2337 if (!md_in_flight(md))
2340 if (interruptible == TASK_INTERRUPTIBLE &&
2341 signal_pending(current)) {
2348 set_current_state(TASK_RUNNING);
2350 remove_wait_queue(&md->wait, &wait);
2356 * Process the deferred bios
2358 static void dm_wq_work(struct work_struct *work)
2360 struct mapped_device *md = container_of(work, struct mapped_device,
2364 down_read(&md->io_lock);
2366 while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
2367 spin_lock_irq(&md->deferred_lock);
2368 c = bio_list_pop(&md->deferred);
2369 spin_unlock_irq(&md->deferred_lock);
2374 up_read(&md->io_lock);
2376 if (dm_request_based(md))
2377 generic_make_request(c);
2379 __split_and_process_bio(md, c);
2381 down_read(&md->io_lock);
2384 up_read(&md->io_lock);
2387 static void dm_queue_flush(struct mapped_device *md)
2389 clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2390 smp_mb__after_clear_bit();
2391 queue_work(md->wq, &md->work);
2395 * Swap in a new table, returning the old one for the caller to destroy.
2397 struct dm_table *dm_swap_table(struct mapped_device *md, struct dm_table *table)
2399 struct dm_table *live_map, *map = ERR_PTR(-EINVAL);
2400 struct queue_limits limits;
2403 mutex_lock(&md->suspend_lock);
2405 /* device must be suspended */
2406 if (!dm_suspended_md(md))
2410 * If the new table has no data devices, retain the existing limits.
2411 * This helps multipath with queue_if_no_path if all paths disappear,
2412 * then new I/O is queued based on these limits, and then some paths
2415 if (dm_table_has_no_data_devices(table)) {
2416 live_map = dm_get_live_table(md);
2418 limits = md->queue->limits;
2419 dm_table_put(live_map);
2422 r = dm_calculate_queue_limits(table, &limits);
2428 map = __bind(md, table, &limits);
2431 mutex_unlock(&md->suspend_lock);
2436 * Functions to lock and unlock any filesystem running on the
2439 static int lock_fs(struct mapped_device *md)
2443 WARN_ON(md->frozen_sb);
2445 md->frozen_sb = freeze_bdev(md->bdev);
2446 if (IS_ERR(md->frozen_sb)) {
2447 r = PTR_ERR(md->frozen_sb);
2448 md->frozen_sb = NULL;
2452 set_bit(DMF_FROZEN, &md->flags);
2457 static void unlock_fs(struct mapped_device *md)
2459 if (!test_bit(DMF_FROZEN, &md->flags))
2462 thaw_bdev(md->bdev, md->frozen_sb);
2463 md->frozen_sb = NULL;
2464 clear_bit(DMF_FROZEN, &md->flags);
2468 * We need to be able to change a mapping table under a mounted
2469 * filesystem. For example we might want to move some data in
2470 * the background. Before the table can be swapped with
2471 * dm_bind_table, dm_suspend must be called to flush any in
2472 * flight bios and ensure that any further io gets deferred.
2475 * Suspend mechanism in request-based dm.
2477 * 1. Flush all I/Os by lock_fs() if needed.
2478 * 2. Stop dispatching any I/O by stopping the request_queue.
2479 * 3. Wait for all in-flight I/Os to be completed or requeued.
2481 * To abort suspend, start the request_queue.
2483 int dm_suspend(struct mapped_device *md, unsigned suspend_flags)
2485 struct dm_table *map = NULL;
2487 int do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG ? 1 : 0;
2488 int noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG ? 1 : 0;
2490 mutex_lock(&md->suspend_lock);
2492 if (dm_suspended_md(md)) {
2497 map = dm_get_live_table(md);
2500 * DMF_NOFLUSH_SUSPENDING must be set before presuspend.
2501 * This flag is cleared before dm_suspend returns.
2504 set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2506 /* This does not get reverted if there's an error later. */
2507 dm_table_presuspend_targets(map);
2510 * Flush I/O to the device.
2511 * Any I/O submitted after lock_fs() may not be flushed.
2512 * noflush takes precedence over do_lockfs.
2513 * (lock_fs() flushes I/Os and waits for them to complete.)
2515 if (!noflush && do_lockfs) {
2522 * Here we must make sure that no processes are submitting requests
2523 * to target drivers i.e. no one may be executing
2524 * __split_and_process_bio. This is called from dm_request and
2527 * To get all processes out of __split_and_process_bio in dm_request,
2528 * we take the write lock. To prevent any process from reentering
2529 * __split_and_process_bio from dm_request and quiesce the thread
2530 * (dm_wq_work), we set BMF_BLOCK_IO_FOR_SUSPEND and call
2531 * flush_workqueue(md->wq).
2533 down_write(&md->io_lock);
2534 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2535 up_write(&md->io_lock);
2538 * Stop md->queue before flushing md->wq in case request-based
2539 * dm defers requests to md->wq from md->queue.
2541 if (dm_request_based(md))
2542 stop_queue(md->queue);
2544 flush_workqueue(md->wq);
2547 * At this point no more requests are entering target request routines.
2548 * We call dm_wait_for_completion to wait for all existing requests
2551 r = dm_wait_for_completion(md, TASK_INTERRUPTIBLE);
2553 down_write(&md->io_lock);
2555 clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2556 up_write(&md->io_lock);
2558 /* were we interrupted ? */
2562 if (dm_request_based(md))
2563 start_queue(md->queue);
2566 goto out; /* pushback list is already flushed, so skip flush */
2570 * If dm_wait_for_completion returned 0, the device is completely
2571 * quiescent now. There is no request-processing activity. All new
2572 * requests are being added to md->deferred list.
2575 set_bit(DMF_SUSPENDED, &md->flags);
2577 dm_table_postsuspend_targets(map);
2583 mutex_unlock(&md->suspend_lock);
2587 int dm_resume(struct mapped_device *md)
2590 struct dm_table *map = NULL;
2592 mutex_lock(&md->suspend_lock);
2593 if (!dm_suspended_md(md))
2596 map = dm_get_live_table(md);
2597 if (!map || !dm_table_get_size(map))
2600 r = dm_table_resume_targets(map);
2607 * Flushing deferred I/Os must be done after targets are resumed
2608 * so that mapping of targets can work correctly.
2609 * Request-based dm is queueing the deferred I/Os in its request_queue.
2611 if (dm_request_based(md))
2612 start_queue(md->queue);
2616 clear_bit(DMF_SUSPENDED, &md->flags);
2621 mutex_unlock(&md->suspend_lock);
2626 /*-----------------------------------------------------------------
2627 * Event notification.
2628 *---------------------------------------------------------------*/
2629 int dm_kobject_uevent(struct mapped_device *md, enum kobject_action action,
2632 char udev_cookie[DM_COOKIE_LENGTH];
2633 char *envp[] = { udev_cookie, NULL };
2636 return kobject_uevent(&disk_to_dev(md->disk)->kobj, action);
2638 snprintf(udev_cookie, DM_COOKIE_LENGTH, "%s=%u",
2639 DM_COOKIE_ENV_VAR_NAME, cookie);
2640 return kobject_uevent_env(&disk_to_dev(md->disk)->kobj,
2645 uint32_t dm_next_uevent_seq(struct mapped_device *md)
2647 return atomic_add_return(1, &md->uevent_seq);
2650 uint32_t dm_get_event_nr(struct mapped_device *md)
2652 return atomic_read(&md->event_nr);
2655 int dm_wait_event(struct mapped_device *md, int event_nr)
2657 return wait_event_interruptible(md->eventq,
2658 (event_nr != atomic_read(&md->event_nr)));
2661 void dm_uevent_add(struct mapped_device *md, struct list_head *elist)
2663 unsigned long flags;
2665 spin_lock_irqsave(&md->uevent_lock, flags);
2666 list_add(elist, &md->uevent_list);
2667 spin_unlock_irqrestore(&md->uevent_lock, flags);
2671 * The gendisk is only valid as long as you have a reference
2674 struct gendisk *dm_disk(struct mapped_device *md)
2679 struct kobject *dm_kobject(struct mapped_device *md)
2685 * struct mapped_device should not be exported outside of dm.c
2686 * so use this check to verify that kobj is part of md structure
2688 struct mapped_device *dm_get_from_kobject(struct kobject *kobj)
2690 struct mapped_device *md;
2692 md = container_of(kobj, struct mapped_device, kobj);
2693 if (&md->kobj != kobj)
2696 if (test_bit(DMF_FREEING, &md->flags) ||
2704 int dm_suspended_md(struct mapped_device *md)
2706 return test_bit(DMF_SUSPENDED, &md->flags);
2709 int dm_suspended(struct dm_target *ti)
2711 return dm_suspended_md(dm_table_get_md(ti->table));
2713 EXPORT_SYMBOL_GPL(dm_suspended);
2715 int dm_noflush_suspending(struct dm_target *ti)
2717 return __noflush_suspending(dm_table_get_md(ti->table));
2719 EXPORT_SYMBOL_GPL(dm_noflush_suspending);
2721 struct dm_md_mempools *dm_alloc_md_mempools(unsigned type, unsigned integrity)
2723 struct dm_md_mempools *pools = kmalloc(sizeof(*pools), GFP_KERNEL);
2724 unsigned int pool_size = (type == DM_TYPE_BIO_BASED) ? 16 : MIN_IOS;
2729 pools->io_pool = (type == DM_TYPE_BIO_BASED) ?
2730 mempool_create_slab_pool(MIN_IOS, _io_cache) :
2731 mempool_create_slab_pool(MIN_IOS, _rq_bio_info_cache);
2732 if (!pools->io_pool)
2733 goto free_pools_and_out;
2735 pools->tio_pool = (type == DM_TYPE_BIO_BASED) ?
2736 mempool_create_slab_pool(MIN_IOS, _tio_cache) :
2737 mempool_create_slab_pool(MIN_IOS, _rq_tio_cache);
2738 if (!pools->tio_pool)
2739 goto free_io_pool_and_out;
2741 pools->bs = (type == DM_TYPE_BIO_BASED) ?
2742 bioset_create(pool_size, 0) :
2743 bioset_create(pool_size,
2744 offsetof(struct dm_rq_clone_bio_info, clone));
2746 goto free_tio_pool_and_out;
2748 if (integrity && bioset_integrity_create(pools->bs, pool_size))
2749 goto free_bioset_and_out;
2753 free_bioset_and_out:
2754 bioset_free(pools->bs);
2756 free_tio_pool_and_out:
2757 mempool_destroy(pools->tio_pool);
2759 free_io_pool_and_out:
2760 mempool_destroy(pools->io_pool);
2768 void dm_free_md_mempools(struct dm_md_mempools *pools)
2774 mempool_destroy(pools->io_pool);
2776 if (pools->tio_pool)
2777 mempool_destroy(pools->tio_pool);
2780 bioset_free(pools->bs);
2785 static const struct block_device_operations dm_blk_dops = {
2786 .open = dm_blk_open,
2787 .release = dm_blk_close,
2788 .ioctl = dm_blk_ioctl,
2789 .getgeo = dm_blk_getgeo,
2790 .owner = THIS_MODULE
2793 EXPORT_SYMBOL(dm_get_mapinfo);
2798 module_init(dm_init);
2799 module_exit(dm_exit);
2801 module_param(major, uint, 0);
2802 MODULE_PARM_DESC(major, "The major number of the device mapper");
2803 MODULE_DESCRIPTION(DM_NAME " driver");
2804 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
2805 MODULE_LICENSE("GPL");