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/buffer_head.h>
18 #include <linux/mempool.h>
19 #include <linux/slab.h>
20 #include <linux/idr.h>
21 #include <linux/hdreg.h>
22 #include <linux/delay.h>
24 #include <trace/events/block.h>
26 #define DM_MSG_PREFIX "core"
29 * Cookies are numeric values sent with CHANGE and REMOVE
30 * uevents while resuming, removing or renaming the device.
32 #define DM_COOKIE_ENV_VAR_NAME "DM_COOKIE"
33 #define DM_COOKIE_LENGTH 24
35 static const char *_name = DM_NAME;
37 static unsigned int major = 0;
38 static unsigned int _major = 0;
40 static DEFINE_IDR(_minor_idr);
42 static DEFINE_SPINLOCK(_minor_lock);
45 * One of these is allocated per bio.
48 struct mapped_device *md;
52 unsigned long start_time;
53 spinlock_t endio_lock;
58 * One of these is allocated per target within a bio. Hopefully
59 * this will be simplified out one day.
68 * For request-based dm.
69 * One of these is allocated per request.
71 struct dm_rq_target_io {
72 struct mapped_device *md;
74 struct request *orig, clone;
80 * For request-based dm.
81 * One of these is allocated per bio.
83 struct dm_rq_clone_bio_info {
85 struct dm_rq_target_io *tio;
88 union map_info *dm_get_mapinfo(struct bio *bio)
90 if (bio && bio->bi_private)
91 return &((struct dm_target_io *)bio->bi_private)->info;
95 union map_info *dm_get_rq_mapinfo(struct request *rq)
97 if (rq && rq->end_io_data)
98 return &((struct dm_rq_target_io *)rq->end_io_data)->info;
101 EXPORT_SYMBOL_GPL(dm_get_rq_mapinfo);
103 #define MINOR_ALLOCED ((void *)-1)
106 * Bits for the md->flags field.
108 #define DMF_BLOCK_IO_FOR_SUSPEND 0
109 #define DMF_SUSPENDED 1
111 #define DMF_FREEING 3
112 #define DMF_DELETING 4
113 #define DMF_NOFLUSH_SUSPENDING 5
114 #define DMF_MERGE_IS_OPTIONAL 6
117 * Work processed by per-device workqueue.
119 struct mapped_device {
120 struct rw_semaphore io_lock;
121 struct mutex suspend_lock;
128 struct request_queue *queue;
130 /* Protect queue and type against concurrent access. */
131 struct mutex type_lock;
133 struct target_type *immutable_target_type;
135 struct gendisk *disk;
141 * A list of ios that arrived while we were suspended.
144 wait_queue_head_t wait;
145 struct work_struct work;
146 struct bio_list deferred;
147 spinlock_t deferred_lock;
150 * Processing queue (flush)
152 struct workqueue_struct *wq;
155 * The current mapping.
157 struct dm_table *map;
160 * io objects are allocated from here.
171 wait_queue_head_t eventq;
173 struct list_head uevent_list;
174 spinlock_t uevent_lock; /* Protect access to uevent_list */
177 * freeze/thaw support require holding onto a super block
179 struct super_block *frozen_sb;
180 struct block_device *bdev;
182 /* forced geometry settings */
183 struct hd_geometry geometry;
185 /* kobject and completion */
186 struct dm_kobject_holder kobj_holder;
188 /* zero-length flush that will be cloned and submitted to targets */
189 struct bio flush_bio;
193 * For mempools pre-allocation at the table loading time.
195 struct dm_md_mempools {
202 static struct kmem_cache *_io_cache;
203 static struct kmem_cache *_tio_cache;
204 static struct kmem_cache *_rq_tio_cache;
205 static struct kmem_cache *_rq_bio_info_cache;
207 static int __init local_init(void)
211 /* allocate a slab for the dm_ios */
212 _io_cache = KMEM_CACHE(dm_io, 0);
216 /* allocate a slab for the target ios */
217 _tio_cache = KMEM_CACHE(dm_target_io, 0);
219 goto out_free_io_cache;
221 _rq_tio_cache = KMEM_CACHE(dm_rq_target_io, 0);
223 goto out_free_tio_cache;
225 _rq_bio_info_cache = KMEM_CACHE(dm_rq_clone_bio_info, 0);
226 if (!_rq_bio_info_cache)
227 goto out_free_rq_tio_cache;
229 r = dm_uevent_init();
231 goto out_free_rq_bio_info_cache;
234 r = register_blkdev(_major, _name);
236 goto out_uevent_exit;
245 out_free_rq_bio_info_cache:
246 kmem_cache_destroy(_rq_bio_info_cache);
247 out_free_rq_tio_cache:
248 kmem_cache_destroy(_rq_tio_cache);
250 kmem_cache_destroy(_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(_tio_cache);
262 kmem_cache_destroy(_io_cache);
263 unregister_blkdev(_major, _name);
268 DMINFO("cleaned up");
271 static int (*_inits[])(void) __initdata = {
281 static void (*_exits[])(void) = {
291 static int __init dm_init(void)
293 const int count = ARRAY_SIZE(_inits);
297 for (i = 0; i < count; i++) {
312 static void __exit dm_exit(void)
314 int i = ARRAY_SIZE(_exits);
320 * Should be empty by this point.
322 idr_remove_all(&_minor_idr);
323 idr_destroy(&_minor_idr);
327 * Block device functions
329 int dm_deleting_md(struct mapped_device *md)
331 return test_bit(DMF_DELETING, &md->flags);
334 static int dm_blk_open(struct block_device *bdev, fmode_t mode)
336 struct mapped_device *md;
338 spin_lock(&_minor_lock);
340 md = bdev->bd_disk->private_data;
344 if (test_bit(DMF_FREEING, &md->flags) ||
345 dm_deleting_md(md)) {
351 atomic_inc(&md->open_count);
354 spin_unlock(&_minor_lock);
356 return md ? 0 : -ENXIO;
359 static int dm_blk_close(struct gendisk *disk, fmode_t mode)
361 struct mapped_device *md = disk->private_data;
363 spin_lock(&_minor_lock);
365 atomic_dec(&md->open_count);
368 spin_unlock(&_minor_lock);
373 int dm_open_count(struct mapped_device *md)
375 return atomic_read(&md->open_count);
379 * Guarantees nothing is using the device before it's deleted.
381 int dm_lock_for_deletion(struct mapped_device *md)
385 spin_lock(&_minor_lock);
387 if (dm_open_count(md))
390 set_bit(DMF_DELETING, &md->flags);
392 spin_unlock(&_minor_lock);
397 static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo)
399 struct mapped_device *md = bdev->bd_disk->private_data;
401 return dm_get_geometry(md, geo);
404 static int dm_blk_ioctl(struct block_device *bdev, fmode_t mode,
405 unsigned int cmd, unsigned long arg)
407 struct mapped_device *md = bdev->bd_disk->private_data;
408 struct dm_table *map = dm_get_live_table(md);
409 struct dm_target *tgt;
412 if (!map || !dm_table_get_size(map))
415 /* We only support devices that have a single target */
416 if (dm_table_get_num_targets(map) != 1)
419 tgt = dm_table_get_target(map, 0);
421 if (dm_suspended_md(md)) {
426 if (tgt->type->ioctl)
427 r = tgt->type->ioctl(tgt, cmd, arg);
435 static struct dm_io *alloc_io(struct mapped_device *md)
437 return mempool_alloc(md->io_pool, GFP_NOIO);
440 static void free_io(struct mapped_device *md, struct dm_io *io)
442 mempool_free(io, md->io_pool);
445 static void free_tio(struct mapped_device *md, struct dm_target_io *tio)
447 mempool_free(tio, md->tio_pool);
450 static struct dm_rq_target_io *alloc_rq_tio(struct mapped_device *md,
453 return mempool_alloc(md->tio_pool, gfp_mask);
456 static void free_rq_tio(struct dm_rq_target_io *tio)
458 mempool_free(tio, tio->md->tio_pool);
461 static struct dm_rq_clone_bio_info *alloc_bio_info(struct mapped_device *md)
463 return mempool_alloc(md->io_pool, GFP_ATOMIC);
466 static void free_bio_info(struct dm_rq_clone_bio_info *info)
468 mempool_free(info, info->tio->md->io_pool);
471 static int md_in_flight(struct mapped_device *md)
473 return atomic_read(&md->pending[READ]) +
474 atomic_read(&md->pending[WRITE]);
477 static void start_io_acct(struct dm_io *io)
479 struct mapped_device *md = io->md;
481 int rw = bio_data_dir(io->bio);
483 io->start_time = jiffies;
485 cpu = part_stat_lock();
486 part_round_stats(cpu, &dm_disk(md)->part0);
488 atomic_set(&dm_disk(md)->part0.in_flight[rw],
489 atomic_inc_return(&md->pending[rw]));
492 static void end_io_acct(struct dm_io *io)
494 struct mapped_device *md = io->md;
495 struct bio *bio = io->bio;
496 unsigned long duration = jiffies - io->start_time;
498 int rw = bio_data_dir(bio);
500 cpu = part_stat_lock();
501 part_round_stats(cpu, &dm_disk(md)->part0);
502 part_stat_add(cpu, &dm_disk(md)->part0, ticks[rw], duration);
506 * After this is decremented the bio must not be touched if it is
509 pending = atomic_dec_return(&md->pending[rw]);
510 atomic_set(&dm_disk(md)->part0.in_flight[rw], pending);
511 pending += atomic_read(&md->pending[rw^0x1]);
513 /* nudge anyone waiting on suspend queue */
519 * Add the bio to the list of deferred io.
521 static void queue_io(struct mapped_device *md, struct bio *bio)
525 spin_lock_irqsave(&md->deferred_lock, flags);
526 bio_list_add(&md->deferred, bio);
527 spin_unlock_irqrestore(&md->deferred_lock, flags);
528 queue_work(md->wq, &md->work);
532 * Everyone (including functions in this file), should use this
533 * function to access the md->map field, and make sure they call
534 * dm_table_put() when finished.
536 struct dm_table *dm_get_live_table(struct mapped_device *md)
541 read_lock_irqsave(&md->map_lock, flags);
545 read_unlock_irqrestore(&md->map_lock, flags);
551 * Get the geometry associated with a dm device
553 int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo)
561 * Set the geometry of a device.
563 int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo)
565 sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors;
567 if (geo->start > sz) {
568 DMWARN("Start sector is beyond the geometry limits.");
577 /*-----------------------------------------------------------------
579 * A more elegant soln is in the works that uses the queue
580 * merge fn, unfortunately there are a couple of changes to
581 * the block layer that I want to make for this. So in the
582 * interests of getting something for people to use I give
583 * you this clearly demarcated crap.
584 *---------------------------------------------------------------*/
586 static int __noflush_suspending(struct mapped_device *md)
588 return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
592 * Decrements the number of outstanding ios that a bio has been
593 * cloned into, completing the original io if necc.
595 static void dec_pending(struct dm_io *io, int error)
600 struct mapped_device *md = io->md;
602 /* Push-back supersedes any I/O errors */
603 if (unlikely(error)) {
604 spin_lock_irqsave(&io->endio_lock, flags);
605 if (!(io->error > 0 && __noflush_suspending(md)))
607 spin_unlock_irqrestore(&io->endio_lock, flags);
610 if (atomic_dec_and_test(&io->io_count)) {
611 if (io->error == DM_ENDIO_REQUEUE) {
613 * Target requested pushing back the I/O.
615 spin_lock_irqsave(&md->deferred_lock, flags);
616 if (__noflush_suspending(md))
617 bio_list_add_head(&md->deferred, io->bio);
619 /* noflush suspend was interrupted. */
621 spin_unlock_irqrestore(&md->deferred_lock, flags);
624 io_error = io->error;
629 if (io_error == DM_ENDIO_REQUEUE)
632 if ((bio->bi_rw & REQ_FLUSH) && bio->bi_size) {
634 * Preflush done for flush with data, reissue
637 bio->bi_rw &= ~REQ_FLUSH;
640 /* done with normal IO or empty flush */
641 trace_block_bio_complete(md->queue, bio, io_error);
642 bio_endio(bio, io_error);
647 static void clone_endio(struct bio *bio, int error)
650 struct dm_target_io *tio = bio->bi_private;
651 struct dm_io *io = tio->io;
652 struct mapped_device *md = tio->io->md;
653 dm_endio_fn endio = tio->ti->type->end_io;
655 if (!bio_flagged(bio, BIO_UPTODATE) && !error)
659 r = endio(tio->ti, bio, error, &tio->info);
660 if (r < 0 || r == DM_ENDIO_REQUEUE)
662 * error and requeue request are handled
666 else if (r == DM_ENDIO_INCOMPLETE)
667 /* The target will handle the io */
670 DMWARN("unimplemented target endio return value: %d", r);
676 * Store md for cleanup instead of tio which is about to get freed.
678 bio->bi_private = md->bs;
682 dec_pending(io, error);
686 * Partial completion handling for request-based dm
688 static void end_clone_bio(struct bio *clone, int error)
690 struct dm_rq_clone_bio_info *info = clone->bi_private;
691 struct dm_rq_target_io *tio = info->tio;
692 struct bio *bio = info->orig;
693 unsigned int nr_bytes = info->orig->bi_size;
699 * An error has already been detected on the request.
700 * Once error occurred, just let clone->end_io() handle
706 * Don't notice the error to the upper layer yet.
707 * The error handling decision is made by the target driver,
708 * when the request is completed.
715 * I/O for the bio successfully completed.
716 * Notice the data completion to the upper layer.
720 * bios are processed from the head of the list.
721 * So the completing bio should always be rq->bio.
722 * If it's not, something wrong is happening.
724 if (tio->orig->bio != bio)
725 DMERR("bio completion is going in the middle of the request");
728 * Update the original request.
729 * Do not use blk_end_request() here, because it may complete
730 * the original request before the clone, and break the ordering.
732 blk_update_request(tio->orig, 0, nr_bytes);
736 * Don't touch any member of the md after calling this function because
737 * the md may be freed in dm_put() at the end of this function.
738 * Or do dm_get() before calling this function and dm_put() later.
740 static void rq_completed(struct mapped_device *md, int rw, int run_queue)
742 atomic_dec(&md->pending[rw]);
744 /* nudge anyone waiting on suspend queue */
745 if (!md_in_flight(md))
749 * Run this off this callpath, as drivers could invoke end_io while
750 * inside their request_fn (and holding the queue lock). Calling
751 * back into ->request_fn() could deadlock attempting to grab the
755 blk_run_queue_async(md->queue);
758 * dm_put() must be at the end of this function. See the comment above
763 static void free_rq_clone(struct request *clone)
765 struct dm_rq_target_io *tio = clone->end_io_data;
767 blk_rq_unprep_clone(clone);
772 * Complete the clone and the original request.
773 * Must be called without queue lock.
775 static void dm_end_request(struct request *clone, int error)
777 int rw = rq_data_dir(clone);
778 struct dm_rq_target_io *tio = clone->end_io_data;
779 struct mapped_device *md = tio->md;
780 struct request *rq = tio->orig;
782 if (rq->cmd_type == REQ_TYPE_BLOCK_PC) {
783 rq->errors = clone->errors;
784 rq->resid_len = clone->resid_len;
788 * We are using the sense buffer of the original
790 * So setting the length of the sense data is enough.
792 rq->sense_len = clone->sense_len;
795 free_rq_clone(clone);
796 blk_end_request_all(rq, error);
797 rq_completed(md, rw, true);
800 static void dm_unprep_request(struct request *rq)
802 struct request *clone = rq->special;
805 rq->cmd_flags &= ~REQ_DONTPREP;
807 free_rq_clone(clone);
811 * Requeue the original request of a clone.
813 void dm_requeue_unmapped_request(struct request *clone)
815 int rw = rq_data_dir(clone);
816 struct dm_rq_target_io *tio = clone->end_io_data;
817 struct mapped_device *md = tio->md;
818 struct request *rq = tio->orig;
819 struct request_queue *q = rq->q;
822 dm_unprep_request(rq);
824 spin_lock_irqsave(q->queue_lock, flags);
825 blk_requeue_request(q, rq);
826 spin_unlock_irqrestore(q->queue_lock, flags);
828 rq_completed(md, rw, 0);
830 EXPORT_SYMBOL_GPL(dm_requeue_unmapped_request);
832 static void __stop_queue(struct request_queue *q)
837 static void stop_queue(struct request_queue *q)
841 spin_lock_irqsave(q->queue_lock, flags);
843 spin_unlock_irqrestore(q->queue_lock, flags);
846 static void __start_queue(struct request_queue *q)
848 if (blk_queue_stopped(q))
852 static void start_queue(struct request_queue *q)
856 spin_lock_irqsave(q->queue_lock, flags);
858 spin_unlock_irqrestore(q->queue_lock, flags);
861 static void dm_done(struct request *clone, int error, bool mapped)
864 struct dm_rq_target_io *tio = clone->end_io_data;
865 dm_request_endio_fn rq_end_io = NULL;
868 rq_end_io = tio->ti->type->rq_end_io;
870 if (mapped && rq_end_io)
871 r = rq_end_io(tio->ti, clone, error, &tio->info);
875 /* The target wants to complete the I/O */
876 dm_end_request(clone, r);
877 else if (r == DM_ENDIO_INCOMPLETE)
878 /* The target will handle the I/O */
880 else if (r == DM_ENDIO_REQUEUE)
881 /* The target wants to requeue the I/O */
882 dm_requeue_unmapped_request(clone);
884 DMWARN("unimplemented target endio return value: %d", r);
890 * Request completion handler for request-based dm
892 static void dm_softirq_done(struct request *rq)
895 struct request *clone = rq->completion_data;
896 struct dm_rq_target_io *tio = clone->end_io_data;
898 if (rq->cmd_flags & REQ_FAILED)
901 dm_done(clone, tio->error, mapped);
905 * Complete the clone and the original request with the error status
906 * through softirq context.
908 static void dm_complete_request(struct request *clone, int error)
910 struct dm_rq_target_io *tio = clone->end_io_data;
911 struct request *rq = tio->orig;
914 rq->completion_data = clone;
915 blk_complete_request(rq);
919 * Complete the not-mapped clone and the original request with the error status
920 * through softirq context.
921 * Target's rq_end_io() function isn't called.
922 * This may be used when the target's map_rq() function fails.
924 void dm_kill_unmapped_request(struct request *clone, int error)
926 struct dm_rq_target_io *tio = clone->end_io_data;
927 struct request *rq = tio->orig;
929 rq->cmd_flags |= REQ_FAILED;
930 dm_complete_request(clone, error);
932 EXPORT_SYMBOL_GPL(dm_kill_unmapped_request);
935 * Called with the queue lock held
937 static void end_clone_request(struct request *clone, int error)
940 * For just cleaning up the information of the queue in which
941 * the clone was dispatched.
942 * The clone is *NOT* freed actually here because it is alloced from
943 * dm own mempool and REQ_ALLOCED isn't set in clone->cmd_flags.
945 __blk_put_request(clone->q, clone);
948 * Actual request completion is done in a softirq context which doesn't
949 * hold the queue lock. Otherwise, deadlock could occur because:
950 * - another request may be submitted by the upper level driver
951 * of the stacking during the completion
952 * - the submission which requires queue lock may be done
955 dm_complete_request(clone, error);
959 * Return maximum size of I/O possible at the supplied sector up to the current
962 static sector_t max_io_len_target_boundary(sector_t sector, struct dm_target *ti)
964 sector_t target_offset = dm_target_offset(ti, sector);
966 return ti->len - target_offset;
969 static sector_t max_io_len(sector_t sector, struct dm_target *ti)
971 sector_t len = max_io_len_target_boundary(sector, ti);
974 * Does the target need to split even further ?
978 sector_t offset = dm_target_offset(ti, sector);
979 boundary = ((offset + ti->split_io) & ~(ti->split_io - 1))
988 static void __map_bio(struct dm_target *ti, struct bio *clone,
989 struct dm_target_io *tio)
993 struct mapped_device *md;
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, &tio->info);
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);
1018 * Store bio_set for cleanup.
1020 clone->bi_private = md->bs;
1024 DMWARN("unimplemented target map return value: %d", r);
1030 struct mapped_device *md;
1031 struct dm_table *map;
1035 sector_t sector_count;
1039 static void dm_bio_destructor(struct bio *bio)
1041 struct bio_set *bs = bio->bi_private;
1047 * Creates a little bio that just does part of a bvec.
1049 static struct bio *split_bvec(struct bio *bio, sector_t sector,
1050 unsigned short idx, unsigned int offset,
1051 unsigned int len, struct bio_set *bs)
1054 struct bio_vec *bv = bio->bi_io_vec + idx;
1056 clone = bio_alloc_bioset(GFP_NOIO, 1, bs);
1057 clone->bi_destructor = dm_bio_destructor;
1058 *clone->bi_io_vec = *bv;
1060 clone->bi_sector = sector;
1061 clone->bi_bdev = bio->bi_bdev;
1062 clone->bi_rw = bio->bi_rw;
1064 clone->bi_size = to_bytes(len);
1065 clone->bi_io_vec->bv_offset = offset;
1066 clone->bi_io_vec->bv_len = clone->bi_size;
1067 clone->bi_flags |= 1 << BIO_CLONED;
1069 if (bio_integrity(bio)) {
1070 bio_integrity_clone(clone, bio, GFP_NOIO, bs);
1071 bio_integrity_trim(clone,
1072 bio_sector_offset(bio, idx, offset), len);
1079 * Creates a bio that consists of range of complete bvecs.
1081 static struct bio *clone_bio(struct bio *bio, sector_t sector,
1082 unsigned short idx, unsigned short bv_count,
1083 unsigned int len, struct bio_set *bs)
1087 clone = bio_alloc_bioset(GFP_NOIO, bio->bi_max_vecs, bs);
1088 __bio_clone(clone, bio);
1089 clone->bi_destructor = dm_bio_destructor;
1090 clone->bi_sector = sector;
1091 clone->bi_idx = idx;
1092 clone->bi_vcnt = idx + bv_count;
1093 clone->bi_size = to_bytes(len);
1094 clone->bi_flags &= ~(1 << BIO_SEG_VALID);
1096 if (bio_integrity(bio)) {
1097 bio_integrity_clone(clone, bio, GFP_NOIO, bs);
1099 if (idx != bio->bi_idx || clone->bi_size < bio->bi_size)
1100 bio_integrity_trim(clone,
1101 bio_sector_offset(bio, idx, 0), len);
1107 static struct dm_target_io *alloc_tio(struct clone_info *ci,
1108 struct dm_target *ti)
1110 struct dm_target_io *tio = mempool_alloc(ci->md->tio_pool, GFP_NOIO);
1114 memset(&tio->info, 0, sizeof(tio->info));
1119 static void __issue_target_request(struct clone_info *ci, struct dm_target *ti,
1120 unsigned request_nr, sector_t len)
1122 struct dm_target_io *tio = alloc_tio(ci, ti);
1125 tio->info.target_request_nr = request_nr;
1128 * Discard requests require the bio's inline iovecs be initialized.
1129 * ci->bio->bi_max_vecs is BIO_INLINE_VECS anyway, for both flush
1130 * and discard, so no need for concern about wasted bvec allocations.
1132 clone = bio_alloc_bioset(GFP_NOIO, ci->bio->bi_max_vecs, ci->md->bs);
1133 __bio_clone(clone, ci->bio);
1134 clone->bi_destructor = dm_bio_destructor;
1136 clone->bi_sector = ci->sector;
1137 clone->bi_size = to_bytes(len);
1140 __map_bio(ti, clone, tio);
1143 static void __issue_target_requests(struct clone_info *ci, struct dm_target *ti,
1144 unsigned num_requests, sector_t len)
1146 unsigned request_nr;
1148 for (request_nr = 0; request_nr < num_requests; request_nr++)
1149 __issue_target_request(ci, ti, request_nr, len);
1152 static int __clone_and_map_empty_flush(struct clone_info *ci)
1154 unsigned target_nr = 0;
1155 struct dm_target *ti;
1157 BUG_ON(bio_has_data(ci->bio));
1158 while ((ti = dm_table_get_target(ci->map, target_nr++)))
1159 __issue_target_requests(ci, ti, ti->num_flush_requests, 0);
1165 * Perform all io with a single clone.
1167 static void __clone_and_map_simple(struct clone_info *ci, struct dm_target *ti)
1169 struct bio *clone, *bio = ci->bio;
1170 struct dm_target_io *tio;
1172 tio = alloc_tio(ci, ti);
1173 clone = clone_bio(bio, ci->sector, ci->idx,
1174 bio->bi_vcnt - ci->idx, ci->sector_count,
1176 __map_bio(ti, clone, tio);
1177 ci->sector_count = 0;
1180 static int __clone_and_map_discard(struct clone_info *ci)
1182 struct dm_target *ti;
1186 ti = dm_table_find_target(ci->map, ci->sector);
1187 if (!dm_target_is_valid(ti))
1191 * Even though the device advertised discard support,
1192 * that does not mean every target supports it, and
1193 * reconfiguration might also have changed that since the
1194 * check was performed.
1196 if (!ti->num_discard_requests)
1199 len = min(ci->sector_count, max_io_len_target_boundary(ci->sector, ti));
1201 __issue_target_requests(ci, ti, ti->num_discard_requests, len);
1204 } while (ci->sector_count -= len);
1209 static int __clone_and_map(struct clone_info *ci)
1211 struct bio *clone, *bio = ci->bio;
1212 struct dm_target *ti;
1213 sector_t len = 0, max;
1214 struct dm_target_io *tio;
1216 if (unlikely(bio->bi_rw & REQ_DISCARD))
1217 return __clone_and_map_discard(ci);
1219 ti = dm_table_find_target(ci->map, ci->sector);
1220 if (!dm_target_is_valid(ti))
1223 max = max_io_len(ci->sector, ti);
1225 if (ci->sector_count <= max) {
1227 * Optimise for the simple case where we can do all of
1228 * the remaining io with a single clone.
1230 __clone_and_map_simple(ci, ti);
1232 } else if (to_sector(bio->bi_io_vec[ci->idx].bv_len) <= max) {
1234 * There are some bvecs that don't span targets.
1235 * Do as many of these as possible.
1238 sector_t remaining = max;
1241 for (i = ci->idx; remaining && (i < bio->bi_vcnt); i++) {
1242 bv_len = to_sector(bio->bi_io_vec[i].bv_len);
1244 if (bv_len > remaining)
1247 remaining -= bv_len;
1251 tio = alloc_tio(ci, ti);
1252 clone = clone_bio(bio, ci->sector, ci->idx, i - ci->idx, len,
1254 __map_bio(ti, clone, tio);
1257 ci->sector_count -= len;
1262 * Handle a bvec that must be split between two or more targets.
1264 struct bio_vec *bv = bio->bi_io_vec + ci->idx;
1265 sector_t remaining = to_sector(bv->bv_len);
1266 unsigned int offset = 0;
1270 ti = dm_table_find_target(ci->map, ci->sector);
1271 if (!dm_target_is_valid(ti))
1274 max = max_io_len(ci->sector, ti);
1277 len = min(remaining, max);
1279 tio = alloc_tio(ci, ti);
1280 clone = split_bvec(bio, ci->sector, ci->idx,
1281 bv->bv_offset + offset, len,
1284 __map_bio(ti, clone, tio);
1287 ci->sector_count -= len;
1288 offset += to_bytes(len);
1289 } while (remaining -= len);
1298 * Split the bio into several clones and submit it to targets.
1300 static void __split_and_process_bio(struct mapped_device *md, struct bio *bio)
1302 struct clone_info ci;
1305 ci.map = dm_get_live_table(md);
1306 if (unlikely(!ci.map)) {
1312 ci.io = alloc_io(md);
1314 atomic_set(&ci.io->io_count, 1);
1317 spin_lock_init(&ci.io->endio_lock);
1318 ci.sector = bio->bi_sector;
1319 ci.idx = bio->bi_idx;
1321 start_io_acct(ci.io);
1322 if (bio->bi_rw & REQ_FLUSH) {
1323 ci.bio = &ci.md->flush_bio;
1324 ci.sector_count = 0;
1325 error = __clone_and_map_empty_flush(&ci);
1326 /* dec_pending submits any data associated with flush */
1329 ci.sector_count = bio_sectors(bio);
1330 while (ci.sector_count && !error)
1331 error = __clone_and_map(&ci);
1334 /* drop the extra reference count */
1335 dec_pending(ci.io, error);
1336 dm_table_put(ci.map);
1338 /*-----------------------------------------------------------------
1340 *---------------------------------------------------------------*/
1342 static int dm_merge_bvec(struct request_queue *q,
1343 struct bvec_merge_data *bvm,
1344 struct bio_vec *biovec)
1346 struct mapped_device *md = q->queuedata;
1347 struct dm_table *map = dm_get_live_table(md);
1348 struct dm_target *ti;
1349 sector_t max_sectors;
1355 ti = dm_table_find_target(map, bvm->bi_sector);
1356 if (!dm_target_is_valid(ti))
1360 * Find maximum amount of I/O that won't need splitting
1362 max_sectors = min(max_io_len(bvm->bi_sector, ti),
1363 (sector_t) BIO_MAX_SECTORS);
1364 max_size = (max_sectors << SECTOR_SHIFT) - bvm->bi_size;
1369 * merge_bvec_fn() returns number of bytes
1370 * it can accept at this offset
1371 * max is precomputed maximal io size
1373 if (max_size && ti->type->merge)
1374 max_size = ti->type->merge(ti, bvm, biovec, max_size);
1376 * If the target doesn't support merge method and some of the devices
1377 * provided their merge_bvec method (we know this by looking at
1378 * queue_max_hw_sectors), then we can't allow bios with multiple vector
1379 * entries. So always set max_size to 0, and the code below allows
1382 else if (queue_max_hw_sectors(q) <= PAGE_SIZE >> 9)
1391 * Always allow an entire first page
1393 if (max_size <= biovec->bv_len && !(bvm->bi_size >> SECTOR_SHIFT))
1394 max_size = biovec->bv_len;
1400 * The request function that just remaps the bio built up by
1403 static void _dm_request(struct request_queue *q, struct bio *bio)
1405 int rw = bio_data_dir(bio);
1406 struct mapped_device *md = q->queuedata;
1409 down_read(&md->io_lock);
1411 cpu = part_stat_lock();
1412 part_stat_inc(cpu, &dm_disk(md)->part0, ios[rw]);
1413 part_stat_add(cpu, &dm_disk(md)->part0, sectors[rw], bio_sectors(bio));
1416 /* if we're suspended, we have to queue this io for later */
1417 if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))) {
1418 up_read(&md->io_lock);
1420 if (bio_rw(bio) != READA)
1427 __split_and_process_bio(md, bio);
1428 up_read(&md->io_lock);
1432 static int dm_request_based(struct mapped_device *md)
1434 return blk_queue_stackable(md->queue);
1437 static void dm_request(struct request_queue *q, struct bio *bio)
1439 struct mapped_device *md = q->queuedata;
1441 if (dm_request_based(md))
1442 blk_queue_bio(q, bio);
1444 _dm_request(q, bio);
1447 void dm_dispatch_request(struct request *rq)
1451 if (blk_queue_io_stat(rq->q))
1452 rq->cmd_flags |= REQ_IO_STAT;
1454 rq->start_time = jiffies;
1455 r = blk_insert_cloned_request(rq->q, rq);
1457 dm_complete_request(rq, r);
1459 EXPORT_SYMBOL_GPL(dm_dispatch_request);
1461 static void dm_rq_bio_destructor(struct bio *bio)
1463 struct dm_rq_clone_bio_info *info = bio->bi_private;
1464 struct mapped_device *md = info->tio->md;
1466 free_bio_info(info);
1467 bio_free(bio, md->bs);
1470 static int dm_rq_bio_constructor(struct bio *bio, struct bio *bio_orig,
1473 struct dm_rq_target_io *tio = data;
1474 struct mapped_device *md = tio->md;
1475 struct dm_rq_clone_bio_info *info = alloc_bio_info(md);
1480 info->orig = bio_orig;
1482 bio->bi_end_io = end_clone_bio;
1483 bio->bi_private = info;
1484 bio->bi_destructor = dm_rq_bio_destructor;
1489 static int setup_clone(struct request *clone, struct request *rq,
1490 struct dm_rq_target_io *tio)
1494 r = blk_rq_prep_clone(clone, rq, tio->md->bs, GFP_ATOMIC,
1495 dm_rq_bio_constructor, tio);
1499 clone->cmd = rq->cmd;
1500 clone->cmd_len = rq->cmd_len;
1501 clone->sense = rq->sense;
1502 clone->buffer = rq->buffer;
1503 clone->end_io = end_clone_request;
1504 clone->end_io_data = tio;
1509 static struct request *clone_rq(struct request *rq, struct mapped_device *md,
1512 struct request *clone;
1513 struct dm_rq_target_io *tio;
1515 tio = alloc_rq_tio(md, gfp_mask);
1523 memset(&tio->info, 0, sizeof(tio->info));
1525 clone = &tio->clone;
1526 if (setup_clone(clone, rq, tio)) {
1536 * Called with the queue lock held.
1538 static int dm_prep_fn(struct request_queue *q, struct request *rq)
1540 struct mapped_device *md = q->queuedata;
1541 struct request *clone;
1543 if (unlikely(rq->special)) {
1544 DMWARN("Already has something in rq->special.");
1545 return BLKPREP_KILL;
1548 clone = clone_rq(rq, md, GFP_ATOMIC);
1550 return BLKPREP_DEFER;
1552 rq->special = clone;
1553 rq->cmd_flags |= REQ_DONTPREP;
1560 * 0 : the request has been processed (not requeued)
1561 * !0 : the request has been requeued
1563 static int map_request(struct dm_target *ti, struct request *clone,
1564 struct mapped_device *md)
1566 int r, requeued = 0;
1567 struct dm_rq_target_io *tio = clone->end_io_data;
1570 r = ti->type->map_rq(ti, clone, &tio->info);
1572 case DM_MAPIO_SUBMITTED:
1573 /* The target has taken the I/O to submit by itself later */
1575 case DM_MAPIO_REMAPPED:
1576 /* The target has remapped the I/O so dispatch it */
1577 trace_block_rq_remap(clone->q, clone, disk_devt(dm_disk(md)),
1578 blk_rq_pos(tio->orig));
1579 dm_dispatch_request(clone);
1581 case DM_MAPIO_REQUEUE:
1582 /* The target wants to requeue the I/O */
1583 dm_requeue_unmapped_request(clone);
1588 DMWARN("unimplemented target map return value: %d", r);
1592 /* The target wants to complete the I/O */
1593 dm_kill_unmapped_request(clone, r);
1600 static struct request *dm_start_request(struct mapped_device *md, struct request *orig)
1602 struct request *clone;
1604 blk_start_request(orig);
1605 clone = orig->special;
1606 atomic_inc(&md->pending[rq_data_dir(clone)]);
1609 * Hold the md reference here for the in-flight I/O.
1610 * We can't rely on the reference count by device opener,
1611 * because the device may be closed during the request completion
1612 * when all bios are completed.
1613 * See the comment in rq_completed() too.
1621 * q->request_fn for request-based dm.
1622 * Called with the queue lock held.
1624 static void dm_request_fn(struct request_queue *q)
1626 struct mapped_device *md = q->queuedata;
1627 struct dm_table *map = dm_get_live_table(md);
1628 struct dm_target *ti;
1629 struct request *rq, *clone;
1633 * For suspend, check blk_queue_stopped() and increment
1634 * ->pending within a single queue_lock not to increment the
1635 * number of in-flight I/Os after the queue is stopped in
1638 while (!blk_queue_stopped(q)) {
1639 rq = blk_peek_request(q);
1643 /* always use block 0 to find the target for flushes for now */
1645 if (!(rq->cmd_flags & REQ_FLUSH))
1646 pos = blk_rq_pos(rq);
1648 ti = dm_table_find_target(map, pos);
1649 if (!dm_target_is_valid(ti)) {
1651 * Must perform setup, that dm_done() requires,
1652 * before calling dm_kill_unmapped_request
1654 DMERR_LIMIT("request attempted access beyond the end of device");
1655 clone = dm_start_request(md, rq);
1656 dm_kill_unmapped_request(clone, -EIO);
1660 if (ti->type->busy && ti->type->busy(ti))
1663 clone = dm_start_request(md, rq);
1665 spin_unlock(q->queue_lock);
1666 if (map_request(ti, clone, md))
1669 BUG_ON(!irqs_disabled());
1670 spin_lock(q->queue_lock);
1676 BUG_ON(!irqs_disabled());
1677 spin_lock(q->queue_lock);
1680 blk_delay_queue(q, HZ / 10);
1685 int dm_underlying_device_busy(struct request_queue *q)
1687 return blk_lld_busy(q);
1689 EXPORT_SYMBOL_GPL(dm_underlying_device_busy);
1691 static int dm_lld_busy(struct request_queue *q)
1694 struct mapped_device *md = q->queuedata;
1695 struct dm_table *map = dm_get_live_table(md);
1697 if (!map || test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))
1700 r = dm_table_any_busy_target(map);
1707 static int dm_any_congested(void *congested_data, int bdi_bits)
1710 struct mapped_device *md = congested_data;
1711 struct dm_table *map;
1713 if (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
1714 map = dm_get_live_table(md);
1717 * Request-based dm cares about only own queue for
1718 * the query about congestion status of request_queue
1720 if (dm_request_based(md))
1721 r = md->queue->backing_dev_info.state &
1724 r = dm_table_any_congested(map, bdi_bits);
1733 /*-----------------------------------------------------------------
1734 * An IDR is used to keep track of allocated minor numbers.
1735 *---------------------------------------------------------------*/
1736 static void free_minor(int minor)
1738 spin_lock(&_minor_lock);
1739 idr_remove(&_minor_idr, minor);
1740 spin_unlock(&_minor_lock);
1744 * See if the device with a specific minor # is free.
1746 static int specific_minor(int minor)
1750 if (minor >= (1 << MINORBITS))
1753 r = idr_pre_get(&_minor_idr, GFP_KERNEL);
1757 spin_lock(&_minor_lock);
1759 if (idr_find(&_minor_idr, minor)) {
1764 r = idr_get_new_above(&_minor_idr, MINOR_ALLOCED, minor, &m);
1769 idr_remove(&_minor_idr, m);
1775 spin_unlock(&_minor_lock);
1779 static int next_free_minor(int *minor)
1783 r = idr_pre_get(&_minor_idr, GFP_KERNEL);
1787 spin_lock(&_minor_lock);
1789 r = idr_get_new(&_minor_idr, MINOR_ALLOCED, &m);
1793 if (m >= (1 << MINORBITS)) {
1794 idr_remove(&_minor_idr, m);
1802 spin_unlock(&_minor_lock);
1806 static const struct block_device_operations dm_blk_dops;
1808 static void dm_wq_work(struct work_struct *work);
1810 static void dm_init_md_queue(struct mapped_device *md)
1813 * Request-based dm devices cannot be stacked on top of bio-based dm
1814 * devices. The type of this dm device has not been decided yet.
1815 * The type is decided at the first table loading time.
1816 * To prevent problematic device stacking, clear the queue flag
1817 * for request stacking support until then.
1819 * This queue is new, so no concurrency on the queue_flags.
1821 queue_flag_clear_unlocked(QUEUE_FLAG_STACKABLE, md->queue);
1823 md->queue->queuedata = md;
1824 md->queue->backing_dev_info.congested_fn = dm_any_congested;
1825 md->queue->backing_dev_info.congested_data = md;
1826 blk_queue_make_request(md->queue, dm_request);
1827 blk_queue_bounce_limit(md->queue, BLK_BOUNCE_ANY);
1828 blk_queue_merge_bvec(md->queue, dm_merge_bvec);
1832 * Allocate and initialise a blank device with a given minor.
1834 static struct mapped_device *alloc_dev(int minor)
1837 struct mapped_device *md = kzalloc(sizeof(*md), GFP_KERNEL);
1841 DMWARN("unable to allocate device, out of memory.");
1845 if (!try_module_get(THIS_MODULE))
1846 goto bad_module_get;
1848 /* get a minor number for the dev */
1849 if (minor == DM_ANY_MINOR)
1850 r = next_free_minor(&minor);
1852 r = specific_minor(minor);
1856 md->type = DM_TYPE_NONE;
1857 init_rwsem(&md->io_lock);
1858 mutex_init(&md->suspend_lock);
1859 mutex_init(&md->type_lock);
1860 spin_lock_init(&md->deferred_lock);
1861 rwlock_init(&md->map_lock);
1862 atomic_set(&md->holders, 1);
1863 atomic_set(&md->open_count, 0);
1864 atomic_set(&md->event_nr, 0);
1865 atomic_set(&md->uevent_seq, 0);
1866 INIT_LIST_HEAD(&md->uevent_list);
1867 spin_lock_init(&md->uevent_lock);
1869 md->queue = blk_alloc_queue(GFP_KERNEL);
1873 dm_init_md_queue(md);
1875 md->disk = alloc_disk(1);
1879 atomic_set(&md->pending[0], 0);
1880 atomic_set(&md->pending[1], 0);
1881 init_waitqueue_head(&md->wait);
1882 INIT_WORK(&md->work, dm_wq_work);
1883 init_waitqueue_head(&md->eventq);
1884 init_completion(&md->kobj_holder.completion);
1886 md->disk->major = _major;
1887 md->disk->first_minor = minor;
1888 md->disk->fops = &dm_blk_dops;
1889 md->disk->queue = md->queue;
1890 md->disk->private_data = md;
1891 sprintf(md->disk->disk_name, "dm-%d", minor);
1893 format_dev_t(md->name, MKDEV(_major, minor));
1895 md->wq = alloc_workqueue("kdmflush",
1896 WQ_NON_REENTRANT | WQ_MEM_RECLAIM, 0);
1900 md->bdev = bdget_disk(md->disk, 0);
1904 bio_init(&md->flush_bio);
1905 md->flush_bio.bi_bdev = md->bdev;
1906 md->flush_bio.bi_rw = WRITE_FLUSH;
1908 /* Populate the mapping, nobody knows we exist yet */
1909 spin_lock(&_minor_lock);
1910 old_md = idr_replace(&_minor_idr, md, minor);
1911 spin_unlock(&_minor_lock);
1913 BUG_ON(old_md != MINOR_ALLOCED);
1918 destroy_workqueue(md->wq);
1920 del_gendisk(md->disk);
1923 blk_cleanup_queue(md->queue);
1927 module_put(THIS_MODULE);
1933 static void unlock_fs(struct mapped_device *md);
1935 static void free_dev(struct mapped_device *md)
1937 int minor = MINOR(disk_devt(md->disk));
1941 destroy_workqueue(md->wq);
1943 mempool_destroy(md->tio_pool);
1945 mempool_destroy(md->io_pool);
1947 bioset_free(md->bs);
1948 blk_integrity_unregister(md->disk);
1949 del_gendisk(md->disk);
1952 spin_lock(&_minor_lock);
1953 md->disk->private_data = NULL;
1954 spin_unlock(&_minor_lock);
1957 blk_cleanup_queue(md->queue);
1958 module_put(THIS_MODULE);
1962 static void __bind_mempools(struct mapped_device *md, struct dm_table *t)
1964 struct dm_md_mempools *p;
1966 if (md->io_pool && md->tio_pool && md->bs)
1967 /* the md already has necessary mempools */
1970 p = dm_table_get_md_mempools(t);
1971 BUG_ON(!p || md->io_pool || md->tio_pool || md->bs);
1973 md->io_pool = p->io_pool;
1975 md->tio_pool = p->tio_pool;
1981 /* mempool bind completed, now no need any mempools in the table */
1982 dm_table_free_md_mempools(t);
1986 * Bind a table to the device.
1988 static void event_callback(void *context)
1990 unsigned long flags;
1992 struct mapped_device *md = (struct mapped_device *) context;
1994 spin_lock_irqsave(&md->uevent_lock, flags);
1995 list_splice_init(&md->uevent_list, &uevents);
1996 spin_unlock_irqrestore(&md->uevent_lock, flags);
1998 dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj);
2000 atomic_inc(&md->event_nr);
2001 wake_up(&md->eventq);
2005 * Protected by md->suspend_lock obtained by dm_swap_table().
2007 static void __set_size(struct mapped_device *md, sector_t size)
2009 set_capacity(md->disk, size);
2011 i_size_write(md->bdev->bd_inode, (loff_t)size << SECTOR_SHIFT);
2015 * Return 1 if the queue has a compulsory merge_bvec_fn function.
2017 * If this function returns 0, then the device is either a non-dm
2018 * device without a merge_bvec_fn, or it is a dm device that is
2019 * able to split any bios it receives that are too big.
2021 int dm_queue_merge_is_compulsory(struct request_queue *q)
2023 struct mapped_device *dev_md;
2025 if (!q->merge_bvec_fn)
2028 if (q->make_request_fn == dm_request) {
2029 dev_md = q->queuedata;
2030 if (test_bit(DMF_MERGE_IS_OPTIONAL, &dev_md->flags))
2037 static int dm_device_merge_is_compulsory(struct dm_target *ti,
2038 struct dm_dev *dev, sector_t start,
2039 sector_t len, void *data)
2041 struct block_device *bdev = dev->bdev;
2042 struct request_queue *q = bdev_get_queue(bdev);
2044 return dm_queue_merge_is_compulsory(q);
2048 * Return 1 if it is acceptable to ignore merge_bvec_fn based
2049 * on the properties of the underlying devices.
2051 static int dm_table_merge_is_optional(struct dm_table *table)
2054 struct dm_target *ti;
2056 while (i < dm_table_get_num_targets(table)) {
2057 ti = dm_table_get_target(table, i++);
2059 if (ti->type->iterate_devices &&
2060 ti->type->iterate_devices(ti, dm_device_merge_is_compulsory, NULL))
2068 * Returns old map, which caller must destroy.
2070 static struct dm_table *__bind(struct mapped_device *md, struct dm_table *t,
2071 struct queue_limits *limits)
2073 struct dm_table *old_map;
2074 struct request_queue *q = md->queue;
2076 unsigned long flags;
2077 int merge_is_optional;
2079 size = dm_table_get_size(t);
2082 * Wipe any geometry if the size of the table changed.
2084 if (size != get_capacity(md->disk))
2085 memset(&md->geometry, 0, sizeof(md->geometry));
2087 __set_size(md, size);
2089 dm_table_event_callback(t, event_callback, md);
2092 * The queue hasn't been stopped yet, if the old table type wasn't
2093 * for request-based during suspension. So stop it to prevent
2094 * I/O mapping before resume.
2095 * This must be done before setting the queue restrictions,
2096 * because request-based dm may be run just after the setting.
2098 if (dm_table_request_based(t) && !blk_queue_stopped(q))
2101 __bind_mempools(md, t);
2103 merge_is_optional = dm_table_merge_is_optional(t);
2105 write_lock_irqsave(&md->map_lock, flags);
2108 md->immutable_target_type = dm_table_get_immutable_target_type(t);
2110 dm_table_set_restrictions(t, q, limits);
2111 if (merge_is_optional)
2112 set_bit(DMF_MERGE_IS_OPTIONAL, &md->flags);
2114 clear_bit(DMF_MERGE_IS_OPTIONAL, &md->flags);
2115 write_unlock_irqrestore(&md->map_lock, flags);
2121 * Returns unbound table for the caller to free.
2123 static struct dm_table *__unbind(struct mapped_device *md)
2125 struct dm_table *map = md->map;
2126 unsigned long flags;
2131 dm_table_event_callback(map, NULL, NULL);
2132 write_lock_irqsave(&md->map_lock, flags);
2134 write_unlock_irqrestore(&md->map_lock, flags);
2140 * Constructor for a new device.
2142 int dm_create(int minor, struct mapped_device **result)
2144 struct mapped_device *md;
2146 md = alloc_dev(minor);
2157 * Functions to manage md->type.
2158 * All are required to hold md->type_lock.
2160 void dm_lock_md_type(struct mapped_device *md)
2162 mutex_lock(&md->type_lock);
2165 void dm_unlock_md_type(struct mapped_device *md)
2167 mutex_unlock(&md->type_lock);
2170 void dm_set_md_type(struct mapped_device *md, unsigned type)
2175 unsigned dm_get_md_type(struct mapped_device *md)
2180 struct target_type *dm_get_immutable_target_type(struct mapped_device *md)
2182 return md->immutable_target_type;
2186 * Fully initialize a request-based queue (->elevator, ->request_fn, etc).
2188 static int dm_init_request_based_queue(struct mapped_device *md)
2190 struct request_queue *q = NULL;
2192 if (md->queue->elevator)
2195 /* Fully initialize the queue */
2196 q = blk_init_allocated_queue(md->queue, dm_request_fn, NULL);
2201 dm_init_md_queue(md);
2202 blk_queue_softirq_done(md->queue, dm_softirq_done);
2203 blk_queue_prep_rq(md->queue, dm_prep_fn);
2204 blk_queue_lld_busy(md->queue, dm_lld_busy);
2206 elv_register_queue(md->queue);
2212 * Setup the DM device's queue based on md's type
2214 int dm_setup_md_queue(struct mapped_device *md)
2216 if ((dm_get_md_type(md) == DM_TYPE_REQUEST_BASED) &&
2217 !dm_init_request_based_queue(md)) {
2218 DMWARN("Cannot initialize queue for request-based mapped device");
2225 struct mapped_device *dm_get_md(dev_t dev)
2227 struct mapped_device *md;
2228 unsigned minor = MINOR(dev);
2230 if (MAJOR(dev) != _major || minor >= (1 << MINORBITS))
2233 spin_lock(&_minor_lock);
2235 md = idr_find(&_minor_idr, minor);
2237 if ((md == MINOR_ALLOCED ||
2238 (MINOR(disk_devt(dm_disk(md))) != minor) ||
2239 dm_deleting_md(md) ||
2240 test_bit(DMF_FREEING, &md->flags))) {
2248 spin_unlock(&_minor_lock);
2252 EXPORT_SYMBOL_GPL(dm_get_md);
2254 void *dm_get_mdptr(struct mapped_device *md)
2256 return md->interface_ptr;
2259 void dm_set_mdptr(struct mapped_device *md, void *ptr)
2261 md->interface_ptr = ptr;
2264 void dm_get(struct mapped_device *md)
2266 atomic_inc(&md->holders);
2267 BUG_ON(test_bit(DMF_FREEING, &md->flags));
2270 const char *dm_device_name(struct mapped_device *md)
2274 EXPORT_SYMBOL_GPL(dm_device_name);
2276 static void __dm_destroy(struct mapped_device *md, bool wait)
2278 struct dm_table *map;
2282 spin_lock(&_minor_lock);
2283 map = dm_get_live_table(md);
2284 idr_replace(&_minor_idr, MINOR_ALLOCED, MINOR(disk_devt(dm_disk(md))));
2285 set_bit(DMF_FREEING, &md->flags);
2286 spin_unlock(&_minor_lock);
2289 * Take suspend_lock so that presuspend and postsuspend methods
2290 * do not race with internal suspend.
2292 mutex_lock(&md->suspend_lock);
2293 if (!dm_suspended_md(md)) {
2294 dm_table_presuspend_targets(map);
2295 dm_table_postsuspend_targets(map);
2297 mutex_unlock(&md->suspend_lock);
2300 * Rare, but there may be I/O requests still going to complete,
2301 * for example. Wait for all references to disappear.
2302 * No one should increment the reference count of the mapped_device,
2303 * after the mapped_device state becomes DMF_FREEING.
2306 while (atomic_read(&md->holders))
2308 else if (atomic_read(&md->holders))
2309 DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)",
2310 dm_device_name(md), atomic_read(&md->holders));
2314 dm_table_destroy(__unbind(md));
2318 void dm_destroy(struct mapped_device *md)
2320 __dm_destroy(md, true);
2323 void dm_destroy_immediate(struct mapped_device *md)
2325 __dm_destroy(md, false);
2328 void dm_put(struct mapped_device *md)
2330 atomic_dec(&md->holders);
2332 EXPORT_SYMBOL_GPL(dm_put);
2334 static int dm_wait_for_completion(struct mapped_device *md, int interruptible)
2337 DECLARE_WAITQUEUE(wait, current);
2339 add_wait_queue(&md->wait, &wait);
2342 set_current_state(interruptible);
2344 if (!md_in_flight(md))
2347 if (interruptible == TASK_INTERRUPTIBLE &&
2348 signal_pending(current)) {
2355 set_current_state(TASK_RUNNING);
2357 remove_wait_queue(&md->wait, &wait);
2363 * Process the deferred bios
2365 static void dm_wq_work(struct work_struct *work)
2367 struct mapped_device *md = container_of(work, struct mapped_device,
2371 down_read(&md->io_lock);
2373 while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
2374 spin_lock_irq(&md->deferred_lock);
2375 c = bio_list_pop(&md->deferred);
2376 spin_unlock_irq(&md->deferred_lock);
2381 up_read(&md->io_lock);
2383 if (dm_request_based(md))
2384 generic_make_request(c);
2386 __split_and_process_bio(md, c);
2388 down_read(&md->io_lock);
2391 up_read(&md->io_lock);
2394 static void dm_queue_flush(struct mapped_device *md)
2396 clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2397 smp_mb__after_clear_bit();
2398 queue_work(md->wq, &md->work);
2402 * Swap in a new table, returning the old one for the caller to destroy.
2404 struct dm_table *dm_swap_table(struct mapped_device *md, struct dm_table *table)
2406 struct dm_table *map = ERR_PTR(-EINVAL);
2407 struct queue_limits limits;
2410 mutex_lock(&md->suspend_lock);
2412 /* device must be suspended */
2413 if (!dm_suspended_md(md))
2416 r = dm_calculate_queue_limits(table, &limits);
2422 map = __bind(md, table, &limits);
2425 mutex_unlock(&md->suspend_lock);
2430 * Functions to lock and unlock any filesystem running on the
2433 static int lock_fs(struct mapped_device *md)
2437 WARN_ON(md->frozen_sb);
2439 md->frozen_sb = freeze_bdev(md->bdev);
2440 if (IS_ERR(md->frozen_sb)) {
2441 r = PTR_ERR(md->frozen_sb);
2442 md->frozen_sb = NULL;
2446 set_bit(DMF_FROZEN, &md->flags);
2451 static void unlock_fs(struct mapped_device *md)
2453 if (!test_bit(DMF_FROZEN, &md->flags))
2456 thaw_bdev(md->bdev, md->frozen_sb);
2457 md->frozen_sb = NULL;
2458 clear_bit(DMF_FROZEN, &md->flags);
2462 * We need to be able to change a mapping table under a mounted
2463 * filesystem. For example we might want to move some data in
2464 * the background. Before the table can be swapped with
2465 * dm_bind_table, dm_suspend must be called to flush any in
2466 * flight bios and ensure that any further io gets deferred.
2469 * Suspend mechanism in request-based dm.
2471 * 1. Flush all I/Os by lock_fs() if needed.
2472 * 2. Stop dispatching any I/O by stopping the request_queue.
2473 * 3. Wait for all in-flight I/Os to be completed or requeued.
2475 * To abort suspend, start the request_queue.
2477 int dm_suspend(struct mapped_device *md, unsigned suspend_flags)
2479 struct dm_table *map = NULL;
2481 int do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG ? 1 : 0;
2482 int noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG ? 1 : 0;
2484 mutex_lock(&md->suspend_lock);
2486 if (dm_suspended_md(md)) {
2491 map = dm_get_live_table(md);
2494 * DMF_NOFLUSH_SUSPENDING must be set before presuspend.
2495 * This flag is cleared before dm_suspend returns.
2498 set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2500 /* This does not get reverted if there's an error later. */
2501 dm_table_presuspend_targets(map);
2504 * Flush I/O to the device.
2505 * Any I/O submitted after lock_fs() may not be flushed.
2506 * noflush takes precedence over do_lockfs.
2507 * (lock_fs() flushes I/Os and waits for them to complete.)
2509 if (!noflush && do_lockfs) {
2516 * Here we must make sure that no processes are submitting requests
2517 * to target drivers i.e. no one may be executing
2518 * __split_and_process_bio. This is called from dm_request and
2521 * To get all processes out of __split_and_process_bio in dm_request,
2522 * we take the write lock. To prevent any process from reentering
2523 * __split_and_process_bio from dm_request and quiesce the thread
2524 * (dm_wq_work), we set BMF_BLOCK_IO_FOR_SUSPEND and call
2525 * flush_workqueue(md->wq).
2527 down_write(&md->io_lock);
2528 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2529 up_write(&md->io_lock);
2532 * Stop md->queue before flushing md->wq in case request-based
2533 * dm defers requests to md->wq from md->queue.
2535 if (dm_request_based(md))
2536 stop_queue(md->queue);
2538 flush_workqueue(md->wq);
2541 * At this point no more requests are entering target request routines.
2542 * We call dm_wait_for_completion to wait for all existing requests
2545 r = dm_wait_for_completion(md, TASK_INTERRUPTIBLE);
2547 down_write(&md->io_lock);
2549 clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2550 up_write(&md->io_lock);
2552 /* were we interrupted ? */
2556 if (dm_request_based(md))
2557 start_queue(md->queue);
2560 goto out; /* pushback list is already flushed, so skip flush */
2564 * If dm_wait_for_completion returned 0, the device is completely
2565 * quiescent now. There is no request-processing activity. All new
2566 * requests are being added to md->deferred list.
2569 set_bit(DMF_SUSPENDED, &md->flags);
2571 dm_table_postsuspend_targets(map);
2577 mutex_unlock(&md->suspend_lock);
2581 int dm_resume(struct mapped_device *md)
2584 struct dm_table *map = NULL;
2586 mutex_lock(&md->suspend_lock);
2587 if (!dm_suspended_md(md))
2590 map = dm_get_live_table(md);
2591 if (!map || !dm_table_get_size(map))
2594 r = dm_table_resume_targets(map);
2601 * Flushing deferred I/Os must be done after targets are resumed
2602 * so that mapping of targets can work correctly.
2603 * Request-based dm is queueing the deferred I/Os in its request_queue.
2605 if (dm_request_based(md))
2606 start_queue(md->queue);
2610 clear_bit(DMF_SUSPENDED, &md->flags);
2615 mutex_unlock(&md->suspend_lock);
2620 /*-----------------------------------------------------------------
2621 * Event notification.
2622 *---------------------------------------------------------------*/
2623 int dm_kobject_uevent(struct mapped_device *md, enum kobject_action action,
2626 char udev_cookie[DM_COOKIE_LENGTH];
2627 char *envp[] = { udev_cookie, NULL };
2630 return kobject_uevent(&disk_to_dev(md->disk)->kobj, action);
2632 snprintf(udev_cookie, DM_COOKIE_LENGTH, "%s=%u",
2633 DM_COOKIE_ENV_VAR_NAME, cookie);
2634 return kobject_uevent_env(&disk_to_dev(md->disk)->kobj,
2639 uint32_t dm_next_uevent_seq(struct mapped_device *md)
2641 return atomic_add_return(1, &md->uevent_seq);
2644 uint32_t dm_get_event_nr(struct mapped_device *md)
2646 return atomic_read(&md->event_nr);
2649 int dm_wait_event(struct mapped_device *md, int event_nr)
2651 return wait_event_interruptible(md->eventq,
2652 (event_nr != atomic_read(&md->event_nr)));
2655 void dm_uevent_add(struct mapped_device *md, struct list_head *elist)
2657 unsigned long flags;
2659 spin_lock_irqsave(&md->uevent_lock, flags);
2660 list_add(elist, &md->uevent_list);
2661 spin_unlock_irqrestore(&md->uevent_lock, flags);
2665 * The gendisk is only valid as long as you have a reference
2668 struct gendisk *dm_disk(struct mapped_device *md)
2673 struct kobject *dm_kobject(struct mapped_device *md)
2675 return &md->kobj_holder.kobj;
2679 * struct mapped_device should not be exported outside of dm.c
2680 * so use this check to verify that kobj is part of md structure
2682 struct mapped_device *dm_get_from_kobject(struct kobject *kobj)
2684 struct mapped_device *md;
2686 md = container_of(kobj, struct mapped_device, kobj_holder.kobj);
2688 spin_lock(&_minor_lock);
2689 if (test_bit(DMF_FREEING, &md->flags) || dm_deleting_md(md)) {
2695 spin_unlock(&_minor_lock);
2700 int dm_suspended_md(struct mapped_device *md)
2702 return test_bit(DMF_SUSPENDED, &md->flags);
2705 int dm_suspended(struct dm_target *ti)
2707 return dm_suspended_md(dm_table_get_md(ti->table));
2709 EXPORT_SYMBOL_GPL(dm_suspended);
2711 int dm_noflush_suspending(struct dm_target *ti)
2713 return __noflush_suspending(dm_table_get_md(ti->table));
2715 EXPORT_SYMBOL_GPL(dm_noflush_suspending);
2717 struct dm_md_mempools *dm_alloc_md_mempools(unsigned type, unsigned integrity)
2719 struct dm_md_mempools *pools = kmalloc(sizeof(*pools), GFP_KERNEL);
2720 unsigned int pool_size = (type == DM_TYPE_BIO_BASED) ? 16 : MIN_IOS;
2725 pools->io_pool = (type == DM_TYPE_BIO_BASED) ?
2726 mempool_create_slab_pool(MIN_IOS, _io_cache) :
2727 mempool_create_slab_pool(MIN_IOS, _rq_bio_info_cache);
2728 if (!pools->io_pool)
2729 goto free_pools_and_out;
2731 pools->tio_pool = (type == DM_TYPE_BIO_BASED) ?
2732 mempool_create_slab_pool(MIN_IOS, _tio_cache) :
2733 mempool_create_slab_pool(MIN_IOS, _rq_tio_cache);
2734 if (!pools->tio_pool)
2735 goto free_io_pool_and_out;
2737 pools->bs = bioset_create(pool_size, 0);
2739 goto free_tio_pool_and_out;
2741 if (integrity && bioset_integrity_create(pools->bs, pool_size))
2742 goto free_bioset_and_out;
2746 free_bioset_and_out:
2747 bioset_free(pools->bs);
2749 free_tio_pool_and_out:
2750 mempool_destroy(pools->tio_pool);
2752 free_io_pool_and_out:
2753 mempool_destroy(pools->io_pool);
2761 void dm_free_md_mempools(struct dm_md_mempools *pools)
2767 mempool_destroy(pools->io_pool);
2769 if (pools->tio_pool)
2770 mempool_destroy(pools->tio_pool);
2773 bioset_free(pools->bs);
2778 static const struct block_device_operations dm_blk_dops = {
2779 .open = dm_blk_open,
2780 .release = dm_blk_close,
2781 .ioctl = dm_blk_ioctl,
2782 .getgeo = dm_blk_getgeo,
2783 .owner = THIS_MODULE
2786 EXPORT_SYMBOL(dm_get_mapinfo);
2791 module_init(dm_init);
2792 module_exit(dm_exit);
2794 module_param(major, uint, 0);
2795 MODULE_PARM_DESC(major, "The major number of the device mapper");
2796 MODULE_DESCRIPTION(DM_NAME " driver");
2797 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
2798 MODULE_LICENSE("GPL");