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>
23 #include <trace/events/block.h>
25 #define DM_MSG_PREFIX "core"
28 * Cookies are numeric values sent with CHANGE and REMOVE
29 * uevents while resuming, removing or renaming the device.
31 #define DM_COOKIE_ENV_VAR_NAME "DM_COOKIE"
32 #define DM_COOKIE_LENGTH 24
34 static const char *_name = DM_NAME;
36 static unsigned int major = 0;
37 static unsigned int _major = 0;
39 static DEFINE_SPINLOCK(_minor_lock);
42 * One of these is allocated per bio.
45 struct mapped_device *md;
49 unsigned long start_time;
50 spinlock_t endio_lock;
55 * One of these is allocated per target within a bio. Hopefully
56 * this will be simplified out one day.
65 * For request-based dm.
66 * One of these is allocated per request.
68 struct dm_rq_target_io {
69 struct mapped_device *md;
71 struct request *orig, clone;
77 * For request-based dm.
78 * One of these is allocated per bio.
80 struct dm_rq_clone_bio_info {
82 struct dm_rq_target_io *tio;
85 union map_info *dm_get_mapinfo(struct bio *bio)
87 if (bio && bio->bi_private)
88 return &((struct dm_target_io *)bio->bi_private)->info;
92 union map_info *dm_get_rq_mapinfo(struct request *rq)
94 if (rq && rq->end_io_data)
95 return &((struct dm_rq_target_io *)rq->end_io_data)->info;
98 EXPORT_SYMBOL_GPL(dm_get_rq_mapinfo);
100 #define MINOR_ALLOCED ((void *)-1)
103 * Bits for the md->flags field.
105 #define DMF_BLOCK_IO_FOR_SUSPEND 0
106 #define DMF_SUSPENDED 1
108 #define DMF_FREEING 3
109 #define DMF_DELETING 4
110 #define DMF_NOFLUSH_SUSPENDING 5
111 #define DMF_QUEUE_IO_TO_THREAD 6
114 * Work processed by per-device workqueue.
116 struct mapped_device {
117 struct rw_semaphore io_lock;
118 struct mutex suspend_lock;
125 struct request_queue *queue;
126 struct gendisk *disk;
132 * A list of ios that arrived while we were suspended.
135 wait_queue_head_t wait;
136 struct work_struct work;
137 struct bio_list deferred;
138 spinlock_t deferred_lock;
141 * An error from the barrier request currently being processed.
146 * Processing queue (flush/barriers)
148 struct workqueue_struct *wq;
151 * The current mapping.
153 struct dm_table *map;
156 * io objects are allocated from here.
167 wait_queue_head_t eventq;
169 struct list_head uevent_list;
170 spinlock_t uevent_lock; /* Protect access to uevent_list */
173 * freeze/thaw support require holding onto a super block
175 struct super_block *frozen_sb;
176 struct block_device *bdev;
178 /* forced geometry settings */
179 struct hd_geometry geometry;
181 /* For saving the address of __make_request for request based dm */
182 make_request_fn *saved_make_request_fn;
187 /* zero-length barrier that will be cloned and submitted to targets */
188 struct bio barrier_bio;
192 * For mempools pre-allocation at the table loading time.
194 struct dm_md_mempools {
201 static struct kmem_cache *_io_cache;
202 static struct kmem_cache *_tio_cache;
203 static struct kmem_cache *_rq_tio_cache;
204 static struct kmem_cache *_rq_bio_info_cache;
206 static int __init local_init(void)
210 /* allocate a slab for the dm_ios */
211 _io_cache = KMEM_CACHE(dm_io, 0);
215 /* allocate a slab for the target ios */
216 _tio_cache = KMEM_CACHE(dm_target_io, 0);
218 goto out_free_io_cache;
220 _rq_tio_cache = KMEM_CACHE(dm_rq_target_io, 0);
222 goto out_free_tio_cache;
224 _rq_bio_info_cache = KMEM_CACHE(dm_rq_clone_bio_info, 0);
225 if (!_rq_bio_info_cache)
226 goto out_free_rq_tio_cache;
228 r = dm_uevent_init();
230 goto out_free_rq_bio_info_cache;
233 r = register_blkdev(_major, _name);
235 goto out_uevent_exit;
244 out_free_rq_bio_info_cache:
245 kmem_cache_destroy(_rq_bio_info_cache);
246 out_free_rq_tio_cache:
247 kmem_cache_destroy(_rq_tio_cache);
249 kmem_cache_destroy(_tio_cache);
251 kmem_cache_destroy(_io_cache);
256 static void local_exit(void)
258 kmem_cache_destroy(_rq_bio_info_cache);
259 kmem_cache_destroy(_rq_tio_cache);
260 kmem_cache_destroy(_tio_cache);
261 kmem_cache_destroy(_io_cache);
262 unregister_blkdev(_major, _name);
267 DMINFO("cleaned up");
270 static int (*_inits[])(void) __initdata = {
280 static void (*_exits[])(void) = {
290 static int __init dm_init(void)
292 const int count = ARRAY_SIZE(_inits);
296 for (i = 0; i < count; i++) {
311 static void __exit dm_exit(void)
313 int i = ARRAY_SIZE(_exits);
320 * Block device functions
322 static int dm_blk_open(struct block_device *bdev, fmode_t mode)
324 struct mapped_device *md;
326 spin_lock(&_minor_lock);
328 md = bdev->bd_disk->private_data;
332 if (test_bit(DMF_FREEING, &md->flags) ||
333 test_bit(DMF_DELETING, &md->flags)) {
339 atomic_inc(&md->open_count);
342 spin_unlock(&_minor_lock);
344 return md ? 0 : -ENXIO;
347 static int dm_blk_close(struct gendisk *disk, fmode_t mode)
349 struct mapped_device *md = disk->private_data;
350 atomic_dec(&md->open_count);
355 int dm_open_count(struct mapped_device *md)
357 return atomic_read(&md->open_count);
361 * Guarantees nothing is using the device before it's deleted.
363 int dm_lock_for_deletion(struct mapped_device *md)
367 spin_lock(&_minor_lock);
369 if (dm_open_count(md))
372 set_bit(DMF_DELETING, &md->flags);
374 spin_unlock(&_minor_lock);
379 static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo)
381 struct mapped_device *md = bdev->bd_disk->private_data;
383 return dm_get_geometry(md, geo);
386 static int dm_blk_ioctl(struct block_device *bdev, fmode_t mode,
387 unsigned int cmd, unsigned long arg)
389 struct mapped_device *md = bdev->bd_disk->private_data;
390 struct dm_table *map = dm_get_table(md);
391 struct dm_target *tgt;
394 if (!map || !dm_table_get_size(map))
397 /* We only support devices that have a single target */
398 if (dm_table_get_num_targets(map) != 1)
401 tgt = dm_table_get_target(map, 0);
403 if (dm_suspended(md)) {
408 if (tgt->type->ioctl)
409 r = tgt->type->ioctl(tgt, cmd, arg);
417 static struct dm_io *alloc_io(struct mapped_device *md)
419 return mempool_alloc(md->io_pool, GFP_NOIO);
422 static void free_io(struct mapped_device *md, struct dm_io *io)
424 mempool_free(io, md->io_pool);
427 static void free_tio(struct mapped_device *md, struct dm_target_io *tio)
429 mempool_free(tio, md->tio_pool);
432 static struct dm_rq_target_io *alloc_rq_tio(struct mapped_device *md,
435 return mempool_alloc(md->tio_pool, gfp_mask);
438 static void free_rq_tio(struct dm_rq_target_io *tio)
440 mempool_free(tio, tio->md->tio_pool);
443 static struct dm_rq_clone_bio_info *alloc_bio_info(struct mapped_device *md)
445 return mempool_alloc(md->io_pool, GFP_ATOMIC);
448 static void free_bio_info(struct dm_rq_clone_bio_info *info)
450 mempool_free(info, info->tio->md->io_pool);
453 static int md_in_flight(struct mapped_device *md)
455 return atomic_read(&md->pending[READ]) +
456 atomic_read(&md->pending[WRITE]);
459 static void start_io_acct(struct dm_io *io)
461 struct mapped_device *md = io->md;
463 int rw = bio_data_dir(io->bio);
465 io->start_time = jiffies;
467 cpu = part_stat_lock();
468 part_round_stats(cpu, &dm_disk(md)->part0);
470 dm_disk(md)->part0.in_flight[rw] = atomic_inc_return(&md->pending[rw]);
473 static void end_io_acct(struct dm_io *io)
475 struct mapped_device *md = io->md;
476 struct bio *bio = io->bio;
477 unsigned long duration = jiffies - io->start_time;
479 int rw = bio_data_dir(bio);
481 cpu = part_stat_lock();
482 part_round_stats(cpu, &dm_disk(md)->part0);
483 part_stat_add(cpu, &dm_disk(md)->part0, ticks[rw], duration);
487 * After this is decremented the bio must not be touched if it is
490 dm_disk(md)->part0.in_flight[rw] = pending =
491 atomic_dec_return(&md->pending[rw]);
492 pending += atomic_read(&md->pending[rw^0x1]);
494 /* nudge anyone waiting on suspend queue */
500 * Add the bio to the list of deferred io.
502 static void queue_io(struct mapped_device *md, struct bio *bio)
504 down_write(&md->io_lock);
506 spin_lock_irq(&md->deferred_lock);
507 bio_list_add(&md->deferred, bio);
508 spin_unlock_irq(&md->deferred_lock);
510 if (!test_and_set_bit(DMF_QUEUE_IO_TO_THREAD, &md->flags))
511 queue_work(md->wq, &md->work);
513 up_write(&md->io_lock);
517 * Everyone (including functions in this file), should use this
518 * function to access the md->map field, and make sure they call
519 * dm_table_put() when finished.
521 struct dm_table *dm_get_table(struct mapped_device *md)
526 read_lock_irqsave(&md->map_lock, flags);
530 read_unlock_irqrestore(&md->map_lock, flags);
536 * Get the geometry associated with a dm device
538 int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo)
546 * Set the geometry of a device.
548 int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo)
550 sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors;
552 if (geo->start > sz) {
553 DMWARN("Start sector is beyond the geometry limits.");
562 /*-----------------------------------------------------------------
564 * A more elegant soln is in the works that uses the queue
565 * merge fn, unfortunately there are a couple of changes to
566 * the block layer that I want to make for this. So in the
567 * interests of getting something for people to use I give
568 * you this clearly demarcated crap.
569 *---------------------------------------------------------------*/
571 static int __noflush_suspending(struct mapped_device *md)
573 return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
577 * Decrements the number of outstanding ios that a bio has been
578 * cloned into, completing the original io if necc.
580 static void dec_pending(struct dm_io *io, int error)
585 struct mapped_device *md = io->md;
587 /* Push-back supersedes any I/O errors */
588 if (unlikely(error)) {
589 spin_lock_irqsave(&io->endio_lock, flags);
590 if (!(io->error > 0 && __noflush_suspending(md)))
592 spin_unlock_irqrestore(&io->endio_lock, flags);
595 if (atomic_dec_and_test(&io->io_count)) {
596 if (io->error == DM_ENDIO_REQUEUE) {
598 * Target requested pushing back the I/O.
600 spin_lock_irqsave(&md->deferred_lock, flags);
601 if (__noflush_suspending(md)) {
602 if (!bio_rw_flagged(io->bio, BIO_RW_BARRIER))
603 bio_list_add_head(&md->deferred,
606 /* noflush suspend was interrupted. */
608 spin_unlock_irqrestore(&md->deferred_lock, flags);
611 io_error = io->error;
614 if (bio_rw_flagged(bio, BIO_RW_BARRIER)) {
616 * There can be just one barrier request so we use
617 * a per-device variable for error reporting.
618 * Note that you can't touch the bio after end_io_acct
620 if (!md->barrier_error && io_error != -EOPNOTSUPP)
621 md->barrier_error = io_error;
626 if (io_error != DM_ENDIO_REQUEUE) {
627 trace_block_bio_complete(md->queue, bio);
629 bio_endio(bio, io_error);
637 static void clone_endio(struct bio *bio, int error)
640 struct dm_target_io *tio = bio->bi_private;
641 struct dm_io *io = tio->io;
642 struct mapped_device *md = tio->io->md;
643 dm_endio_fn endio = tio->ti->type->end_io;
645 if (!bio_flagged(bio, BIO_UPTODATE) && !error)
649 r = endio(tio->ti, bio, error, &tio->info);
650 if (r < 0 || r == DM_ENDIO_REQUEUE)
652 * error and requeue request are handled
656 else if (r == DM_ENDIO_INCOMPLETE)
657 /* The target will handle the io */
660 DMWARN("unimplemented target endio return value: %d", r);
666 * Store md for cleanup instead of tio which is about to get freed.
668 bio->bi_private = md->bs;
672 dec_pending(io, error);
676 * Partial completion handling for request-based dm
678 static void end_clone_bio(struct bio *clone, int error)
680 struct dm_rq_clone_bio_info *info = clone->bi_private;
681 struct dm_rq_target_io *tio = info->tio;
682 struct bio *bio = info->orig;
683 unsigned int nr_bytes = info->orig->bi_size;
689 * An error has already been detected on the request.
690 * Once error occurred, just let clone->end_io() handle
696 * Don't notice the error to the upper layer yet.
697 * The error handling decision is made by the target driver,
698 * when the request is completed.
705 * I/O for the bio successfully completed.
706 * Notice the data completion to the upper layer.
710 * bios are processed from the head of the list.
711 * So the completing bio should always be rq->bio.
712 * If it's not, something wrong is happening.
714 if (tio->orig->bio != bio)
715 DMERR("bio completion is going in the middle of the request");
718 * Update the original request.
719 * Do not use blk_end_request() here, because it may complete
720 * the original request before the clone, and break the ordering.
722 blk_update_request(tio->orig, 0, nr_bytes);
726 * Don't touch any member of the md after calling this function because
727 * the md may be freed in dm_put() at the end of this function.
728 * Or do dm_get() before calling this function and dm_put() later.
730 static void rq_completed(struct mapped_device *md, int run_queue)
732 int wakeup_waiters = 0;
733 struct request_queue *q = md->queue;
736 spin_lock_irqsave(q->queue_lock, flags);
737 if (!queue_in_flight(q))
739 spin_unlock_irqrestore(q->queue_lock, flags);
741 /* nudge anyone waiting on suspend queue */
749 * dm_put() must be at the end of this function. See the comment above
754 static void free_rq_clone(struct request *clone)
756 struct dm_rq_target_io *tio = clone->end_io_data;
758 blk_rq_unprep_clone(clone);
762 static void dm_unprep_request(struct request *rq)
764 struct request *clone = rq->special;
767 rq->cmd_flags &= ~REQ_DONTPREP;
769 free_rq_clone(clone);
773 * Requeue the original request of a clone.
775 void dm_requeue_unmapped_request(struct request *clone)
777 struct dm_rq_target_io *tio = clone->end_io_data;
778 struct mapped_device *md = tio->md;
779 struct request *rq = tio->orig;
780 struct request_queue *q = rq->q;
783 dm_unprep_request(rq);
785 spin_lock_irqsave(q->queue_lock, flags);
786 if (elv_queue_empty(q))
788 blk_requeue_request(q, rq);
789 spin_unlock_irqrestore(q->queue_lock, flags);
793 EXPORT_SYMBOL_GPL(dm_requeue_unmapped_request);
795 static void __stop_queue(struct request_queue *q)
800 static void stop_queue(struct request_queue *q)
804 spin_lock_irqsave(q->queue_lock, flags);
806 spin_unlock_irqrestore(q->queue_lock, flags);
809 static void __start_queue(struct request_queue *q)
811 if (blk_queue_stopped(q))
815 static void start_queue(struct request_queue *q)
819 spin_lock_irqsave(q->queue_lock, flags);
821 spin_unlock_irqrestore(q->queue_lock, flags);
825 * Complete the clone and the original request.
826 * Must be called without queue lock.
828 static void dm_end_request(struct request *clone, int error)
830 struct dm_rq_target_io *tio = clone->end_io_data;
831 struct mapped_device *md = tio->md;
832 struct request *rq = tio->orig;
834 if (blk_pc_request(rq)) {
835 rq->errors = clone->errors;
836 rq->resid_len = clone->resid_len;
840 * We are using the sense buffer of the original
842 * So setting the length of the sense data is enough.
844 rq->sense_len = clone->sense_len;
847 free_rq_clone(clone);
849 blk_end_request_all(rq, error);
855 * Request completion handler for request-based dm
857 static void dm_softirq_done(struct request *rq)
859 struct request *clone = rq->completion_data;
860 struct dm_rq_target_io *tio = clone->end_io_data;
861 dm_request_endio_fn rq_end_io = tio->ti->type->rq_end_io;
862 int error = tio->error;
864 if (!(rq->cmd_flags & REQ_FAILED) && rq_end_io)
865 error = rq_end_io(tio->ti, clone, error, &tio->info);
868 /* The target wants to complete the I/O */
869 dm_end_request(clone, error);
870 else if (error == DM_ENDIO_INCOMPLETE)
871 /* The target will handle the I/O */
873 else if (error == DM_ENDIO_REQUEUE)
874 /* The target wants to requeue the I/O */
875 dm_requeue_unmapped_request(clone);
877 DMWARN("unimplemented target endio return value: %d", error);
883 * Complete the clone and the original request with the error status
884 * through softirq context.
886 static void dm_complete_request(struct request *clone, int error)
888 struct dm_rq_target_io *tio = clone->end_io_data;
889 struct request *rq = tio->orig;
892 rq->completion_data = clone;
893 blk_complete_request(rq);
897 * Complete the not-mapped clone and the original request with the error status
898 * through softirq context.
899 * Target's rq_end_io() function isn't called.
900 * This may be used when the target's map_rq() function fails.
902 void dm_kill_unmapped_request(struct request *clone, int error)
904 struct dm_rq_target_io *tio = clone->end_io_data;
905 struct request *rq = tio->orig;
907 rq->cmd_flags |= REQ_FAILED;
908 dm_complete_request(clone, error);
910 EXPORT_SYMBOL_GPL(dm_kill_unmapped_request);
913 * Called with the queue lock held
915 static void end_clone_request(struct request *clone, int error)
918 * For just cleaning up the information of the queue in which
919 * the clone was dispatched.
920 * The clone is *NOT* freed actually here because it is alloced from
921 * dm own mempool and REQ_ALLOCED isn't set in clone->cmd_flags.
923 __blk_put_request(clone->q, clone);
926 * Actual request completion is done in a softirq context which doesn't
927 * hold the queue lock. Otherwise, deadlock could occur because:
928 * - another request may be submitted by the upper level driver
929 * of the stacking during the completion
930 * - the submission which requires queue lock may be done
933 dm_complete_request(clone, error);
936 static sector_t max_io_len(struct mapped_device *md,
937 sector_t sector, struct dm_target *ti)
939 sector_t offset = sector - ti->begin;
940 sector_t len = ti->len - offset;
943 * Does the target need to split even further ?
947 boundary = ((offset + ti->split_io) & ~(ti->split_io - 1))
956 static void __map_bio(struct dm_target *ti, struct bio *clone,
957 struct dm_target_io *tio)
961 struct mapped_device *md;
963 clone->bi_end_io = clone_endio;
964 clone->bi_private = tio;
967 * Map the clone. If r == 0 we don't need to do
968 * anything, the target has assumed ownership of
971 atomic_inc(&tio->io->io_count);
972 sector = clone->bi_sector;
973 r = ti->type->map(ti, clone, &tio->info);
974 if (r == DM_MAPIO_REMAPPED) {
975 /* the bio has been remapped so dispatch it */
977 trace_block_remap(bdev_get_queue(clone->bi_bdev), clone,
978 tio->io->bio->bi_bdev->bd_dev, sector);
980 generic_make_request(clone);
981 } else if (r < 0 || r == DM_MAPIO_REQUEUE) {
982 /* error the io and bail out, or requeue it if needed */
984 dec_pending(tio->io, r);
986 * Store bio_set for cleanup.
988 clone->bi_private = md->bs;
992 DMWARN("unimplemented target map return value: %d", r);
998 struct mapped_device *md;
999 struct dm_table *map;
1003 sector_t sector_count;
1007 static void dm_bio_destructor(struct bio *bio)
1009 struct bio_set *bs = bio->bi_private;
1015 * Creates a little bio that is just does part of a bvec.
1017 static struct bio *split_bvec(struct bio *bio, sector_t sector,
1018 unsigned short idx, unsigned int offset,
1019 unsigned int len, struct bio_set *bs)
1022 struct bio_vec *bv = bio->bi_io_vec + idx;
1024 clone = bio_alloc_bioset(GFP_NOIO, 1, bs);
1025 clone->bi_destructor = dm_bio_destructor;
1026 *clone->bi_io_vec = *bv;
1028 clone->bi_sector = sector;
1029 clone->bi_bdev = bio->bi_bdev;
1030 clone->bi_rw = bio->bi_rw & ~(1 << BIO_RW_BARRIER);
1032 clone->bi_size = to_bytes(len);
1033 clone->bi_io_vec->bv_offset = offset;
1034 clone->bi_io_vec->bv_len = clone->bi_size;
1035 clone->bi_flags |= 1 << BIO_CLONED;
1037 if (bio_integrity(bio)) {
1038 bio_integrity_clone(clone, bio, GFP_NOIO, bs);
1039 bio_integrity_trim(clone,
1040 bio_sector_offset(bio, idx, offset), len);
1047 * Creates a bio that consists of range of complete bvecs.
1049 static struct bio *clone_bio(struct bio *bio, sector_t sector,
1050 unsigned short idx, unsigned short bv_count,
1051 unsigned int len, struct bio_set *bs)
1055 clone = bio_alloc_bioset(GFP_NOIO, bio->bi_max_vecs, bs);
1056 __bio_clone(clone, bio);
1057 clone->bi_rw &= ~(1 << BIO_RW_BARRIER);
1058 clone->bi_destructor = dm_bio_destructor;
1059 clone->bi_sector = sector;
1060 clone->bi_idx = idx;
1061 clone->bi_vcnt = idx + bv_count;
1062 clone->bi_size = to_bytes(len);
1063 clone->bi_flags &= ~(1 << BIO_SEG_VALID);
1065 if (bio_integrity(bio)) {
1066 bio_integrity_clone(clone, bio, GFP_NOIO, bs);
1068 if (idx != bio->bi_idx || clone->bi_size < bio->bi_size)
1069 bio_integrity_trim(clone,
1070 bio_sector_offset(bio, idx, 0), len);
1076 static struct dm_target_io *alloc_tio(struct clone_info *ci,
1077 struct dm_target *ti)
1079 struct dm_target_io *tio = mempool_alloc(ci->md->tio_pool, GFP_NOIO);
1083 memset(&tio->info, 0, sizeof(tio->info));
1088 static void __flush_target(struct clone_info *ci, struct dm_target *ti,
1091 struct dm_target_io *tio = alloc_tio(ci, ti);
1094 tio->info.flush_request = flush_nr;
1096 clone = bio_alloc_bioset(GFP_NOIO, 0, ci->md->bs);
1097 __bio_clone(clone, ci->bio);
1098 clone->bi_destructor = dm_bio_destructor;
1100 __map_bio(ti, clone, tio);
1103 static int __clone_and_map_empty_barrier(struct clone_info *ci)
1105 unsigned target_nr = 0, flush_nr;
1106 struct dm_target *ti;
1108 while ((ti = dm_table_get_target(ci->map, target_nr++)))
1109 for (flush_nr = 0; flush_nr < ti->num_flush_requests;
1111 __flush_target(ci, ti, flush_nr);
1113 ci->sector_count = 0;
1118 static int __clone_and_map(struct clone_info *ci)
1120 struct bio *clone, *bio = ci->bio;
1121 struct dm_target *ti;
1122 sector_t len = 0, max;
1123 struct dm_target_io *tio;
1125 if (unlikely(bio_empty_barrier(bio)))
1126 return __clone_and_map_empty_barrier(ci);
1128 ti = dm_table_find_target(ci->map, ci->sector);
1129 if (!dm_target_is_valid(ti))
1132 max = max_io_len(ci->md, ci->sector, ti);
1135 * Allocate a target io object.
1137 tio = alloc_tio(ci, ti);
1139 if (ci->sector_count <= max) {
1141 * Optimise for the simple case where we can do all of
1142 * the remaining io with a single clone.
1144 clone = clone_bio(bio, ci->sector, ci->idx,
1145 bio->bi_vcnt - ci->idx, ci->sector_count,
1147 __map_bio(ti, clone, tio);
1148 ci->sector_count = 0;
1150 } else if (to_sector(bio->bi_io_vec[ci->idx].bv_len) <= max) {
1152 * There are some bvecs that don't span targets.
1153 * Do as many of these as possible.
1156 sector_t remaining = max;
1159 for (i = ci->idx; remaining && (i < bio->bi_vcnt); i++) {
1160 bv_len = to_sector(bio->bi_io_vec[i].bv_len);
1162 if (bv_len > remaining)
1165 remaining -= bv_len;
1169 clone = clone_bio(bio, ci->sector, ci->idx, i - ci->idx, len,
1171 __map_bio(ti, clone, tio);
1174 ci->sector_count -= len;
1179 * Handle a bvec that must be split between two or more targets.
1181 struct bio_vec *bv = bio->bi_io_vec + ci->idx;
1182 sector_t remaining = to_sector(bv->bv_len);
1183 unsigned int offset = 0;
1187 ti = dm_table_find_target(ci->map, ci->sector);
1188 if (!dm_target_is_valid(ti))
1191 max = max_io_len(ci->md, ci->sector, ti);
1193 tio = alloc_tio(ci, ti);
1196 len = min(remaining, max);
1198 clone = split_bvec(bio, ci->sector, ci->idx,
1199 bv->bv_offset + offset, len,
1202 __map_bio(ti, clone, tio);
1205 ci->sector_count -= len;
1206 offset += to_bytes(len);
1207 } while (remaining -= len);
1216 * Split the bio into several clones and submit it to targets.
1218 static void __split_and_process_bio(struct mapped_device *md, struct bio *bio)
1220 struct clone_info ci;
1223 ci.map = dm_get_table(md);
1224 if (unlikely(!ci.map)) {
1225 if (!bio_rw_flagged(bio, BIO_RW_BARRIER))
1228 if (!md->barrier_error)
1229 md->barrier_error = -EIO;
1235 ci.io = alloc_io(md);
1237 atomic_set(&ci.io->io_count, 1);
1240 spin_lock_init(&ci.io->endio_lock);
1241 ci.sector = bio->bi_sector;
1242 ci.sector_count = bio_sectors(bio);
1243 if (unlikely(bio_empty_barrier(bio)))
1244 ci.sector_count = 1;
1245 ci.idx = bio->bi_idx;
1247 start_io_acct(ci.io);
1248 while (ci.sector_count && !error)
1249 error = __clone_and_map(&ci);
1251 /* drop the extra reference count */
1252 dec_pending(ci.io, error);
1253 dm_table_put(ci.map);
1255 /*-----------------------------------------------------------------
1257 *---------------------------------------------------------------*/
1259 static int dm_merge_bvec(struct request_queue *q,
1260 struct bvec_merge_data *bvm,
1261 struct bio_vec *biovec)
1263 struct mapped_device *md = q->queuedata;
1264 struct dm_table *map = dm_get_table(md);
1265 struct dm_target *ti;
1266 sector_t max_sectors;
1272 ti = dm_table_find_target(map, bvm->bi_sector);
1273 if (!dm_target_is_valid(ti))
1277 * Find maximum amount of I/O that won't need splitting
1279 max_sectors = min(max_io_len(md, bvm->bi_sector, ti),
1280 (sector_t) BIO_MAX_SECTORS);
1281 max_size = (max_sectors << SECTOR_SHIFT) - bvm->bi_size;
1286 * merge_bvec_fn() returns number of bytes
1287 * it can accept at this offset
1288 * max is precomputed maximal io size
1290 if (max_size && ti->type->merge)
1291 max_size = ti->type->merge(ti, bvm, biovec, max_size);
1293 * If the target doesn't support merge method and some of the devices
1294 * provided their merge_bvec method (we know this by looking at
1295 * queue_max_hw_sectors), then we can't allow bios with multiple vector
1296 * entries. So always set max_size to 0, and the code below allows
1299 else if (queue_max_hw_sectors(q) <= PAGE_SIZE >> 9)
1308 * Always allow an entire first page
1310 if (max_size <= biovec->bv_len && !(bvm->bi_size >> SECTOR_SHIFT))
1311 max_size = biovec->bv_len;
1317 * The request function that just remaps the bio built up by
1320 static int _dm_request(struct request_queue *q, struct bio *bio)
1322 int rw = bio_data_dir(bio);
1323 struct mapped_device *md = q->queuedata;
1326 down_read(&md->io_lock);
1328 cpu = part_stat_lock();
1329 part_stat_inc(cpu, &dm_disk(md)->part0, ios[rw]);
1330 part_stat_add(cpu, &dm_disk(md)->part0, sectors[rw], bio_sectors(bio));
1334 * If we're suspended or the thread is processing barriers
1335 * we have to queue this io for later.
1337 if (unlikely(test_bit(DMF_QUEUE_IO_TO_THREAD, &md->flags)) ||
1338 unlikely(bio_rw_flagged(bio, BIO_RW_BARRIER))) {
1339 up_read(&md->io_lock);
1341 if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) &&
1342 bio_rw(bio) == READA) {
1352 __split_and_process_bio(md, bio);
1353 up_read(&md->io_lock);
1357 static int dm_make_request(struct request_queue *q, struct bio *bio)
1359 struct mapped_device *md = q->queuedata;
1361 if (unlikely(bio_rw_flagged(bio, BIO_RW_BARRIER))) {
1362 bio_endio(bio, -EOPNOTSUPP);
1366 return md->saved_make_request_fn(q, bio); /* call __make_request() */
1369 static int dm_request_based(struct mapped_device *md)
1371 return blk_queue_stackable(md->queue);
1374 static int dm_request(struct request_queue *q, struct bio *bio)
1376 struct mapped_device *md = q->queuedata;
1378 if (dm_request_based(md))
1379 return dm_make_request(q, bio);
1381 return _dm_request(q, bio);
1384 void dm_dispatch_request(struct request *rq)
1388 if (blk_queue_io_stat(rq->q))
1389 rq->cmd_flags |= REQ_IO_STAT;
1391 rq->start_time = jiffies;
1392 r = blk_insert_cloned_request(rq->q, rq);
1394 dm_complete_request(rq, r);
1396 EXPORT_SYMBOL_GPL(dm_dispatch_request);
1398 static void dm_rq_bio_destructor(struct bio *bio)
1400 struct dm_rq_clone_bio_info *info = bio->bi_private;
1401 struct mapped_device *md = info->tio->md;
1403 free_bio_info(info);
1404 bio_free(bio, md->bs);
1407 static int dm_rq_bio_constructor(struct bio *bio, struct bio *bio_orig,
1410 struct dm_rq_target_io *tio = data;
1411 struct mapped_device *md = tio->md;
1412 struct dm_rq_clone_bio_info *info = alloc_bio_info(md);
1417 info->orig = bio_orig;
1419 bio->bi_end_io = end_clone_bio;
1420 bio->bi_private = info;
1421 bio->bi_destructor = dm_rq_bio_destructor;
1426 static int setup_clone(struct request *clone, struct request *rq,
1427 struct dm_rq_target_io *tio)
1429 int r = blk_rq_prep_clone(clone, rq, tio->md->bs, GFP_ATOMIC,
1430 dm_rq_bio_constructor, tio);
1435 clone->cmd = rq->cmd;
1436 clone->cmd_len = rq->cmd_len;
1437 clone->sense = rq->sense;
1438 clone->buffer = rq->buffer;
1439 clone->end_io = end_clone_request;
1440 clone->end_io_data = tio;
1445 static struct request *clone_rq(struct request *rq, struct mapped_device *md,
1448 struct request *clone;
1449 struct dm_rq_target_io *tio;
1451 tio = alloc_rq_tio(md, gfp_mask);
1459 memset(&tio->info, 0, sizeof(tio->info));
1461 clone = &tio->clone;
1462 if (setup_clone(clone, rq, tio)) {
1472 * Called with the queue lock held.
1474 static int dm_prep_fn(struct request_queue *q, struct request *rq)
1476 struct mapped_device *md = q->queuedata;
1477 struct request *clone;
1479 if (unlikely(rq->special)) {
1480 DMWARN("Already has something in rq->special.");
1481 return BLKPREP_KILL;
1484 clone = clone_rq(rq, md, GFP_ATOMIC);
1486 return BLKPREP_DEFER;
1488 rq->special = clone;
1489 rq->cmd_flags |= REQ_DONTPREP;
1494 static void map_request(struct dm_target *ti, struct request *clone,
1495 struct mapped_device *md)
1498 struct dm_rq_target_io *tio = clone->end_io_data;
1501 * Hold the md reference here for the in-flight I/O.
1502 * We can't rely on the reference count by device opener,
1503 * because the device may be closed during the request completion
1504 * when all bios are completed.
1505 * See the comment in rq_completed() too.
1510 r = ti->type->map_rq(ti, clone, &tio->info);
1512 case DM_MAPIO_SUBMITTED:
1513 /* The target has taken the I/O to submit by itself later */
1515 case DM_MAPIO_REMAPPED:
1516 /* The target has remapped the I/O so dispatch it */
1517 dm_dispatch_request(clone);
1519 case DM_MAPIO_REQUEUE:
1520 /* The target wants to requeue the I/O */
1521 dm_requeue_unmapped_request(clone);
1525 DMWARN("unimplemented target map return value: %d", r);
1529 /* The target wants to complete the I/O */
1530 dm_kill_unmapped_request(clone, r);
1536 * q->request_fn for request-based dm.
1537 * Called with the queue lock held.
1539 static void dm_request_fn(struct request_queue *q)
1541 struct mapped_device *md = q->queuedata;
1542 struct dm_table *map = dm_get_table(md);
1543 struct dm_target *ti;
1547 * For suspend, check blk_queue_stopped() and don't increment
1548 * the number of in-flight I/Os after the queue is stopped
1551 while (!blk_queue_plugged(q) && !blk_queue_stopped(q)) {
1552 rq = blk_peek_request(q);
1556 ti = dm_table_find_target(map, blk_rq_pos(rq));
1557 if (ti->type->busy && ti->type->busy(ti))
1560 blk_start_request(rq);
1561 spin_unlock(q->queue_lock);
1562 map_request(ti, rq->special, md);
1563 spin_lock_irq(q->queue_lock);
1569 if (!elv_queue_empty(q))
1570 /* Some requests still remain, retry later */
1579 int dm_underlying_device_busy(struct request_queue *q)
1581 return blk_lld_busy(q);
1583 EXPORT_SYMBOL_GPL(dm_underlying_device_busy);
1585 static int dm_lld_busy(struct request_queue *q)
1588 struct mapped_device *md = q->queuedata;
1589 struct dm_table *map = dm_get_table(md);
1591 if (!map || test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))
1594 r = dm_table_any_busy_target(map);
1601 static void dm_unplug_all(struct request_queue *q)
1603 struct mapped_device *md = q->queuedata;
1604 struct dm_table *map = dm_get_table(md);
1607 if (dm_request_based(md))
1608 generic_unplug_device(q);
1610 dm_table_unplug_all(map);
1615 static int dm_any_congested(void *congested_data, int bdi_bits)
1618 struct mapped_device *md = congested_data;
1619 struct dm_table *map;
1621 if (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
1622 map = dm_get_table(md);
1625 * Request-based dm cares about only own queue for
1626 * the query about congestion status of request_queue
1628 if (dm_request_based(md))
1629 r = md->queue->backing_dev_info.state &
1632 r = dm_table_any_congested(map, bdi_bits);
1641 /*-----------------------------------------------------------------
1642 * An IDR is used to keep track of allocated minor numbers.
1643 *---------------------------------------------------------------*/
1644 static DEFINE_IDR(_minor_idr);
1646 static void free_minor(int minor)
1648 spin_lock(&_minor_lock);
1649 idr_remove(&_minor_idr, minor);
1650 spin_unlock(&_minor_lock);
1654 * See if the device with a specific minor # is free.
1656 static int specific_minor(int minor)
1660 if (minor >= (1 << MINORBITS))
1663 r = idr_pre_get(&_minor_idr, GFP_KERNEL);
1667 spin_lock(&_minor_lock);
1669 if (idr_find(&_minor_idr, minor)) {
1674 r = idr_get_new_above(&_minor_idr, MINOR_ALLOCED, minor, &m);
1679 idr_remove(&_minor_idr, m);
1685 spin_unlock(&_minor_lock);
1689 static int next_free_minor(int *minor)
1693 r = idr_pre_get(&_minor_idr, GFP_KERNEL);
1697 spin_lock(&_minor_lock);
1699 r = idr_get_new(&_minor_idr, MINOR_ALLOCED, &m);
1703 if (m >= (1 << MINORBITS)) {
1704 idr_remove(&_minor_idr, m);
1712 spin_unlock(&_minor_lock);
1716 static const struct block_device_operations dm_blk_dops;
1718 static void dm_wq_work(struct work_struct *work);
1721 * Allocate and initialise a blank device with a given minor.
1723 static struct mapped_device *alloc_dev(int minor)
1726 struct mapped_device *md = kzalloc(sizeof(*md), GFP_KERNEL);
1730 DMWARN("unable to allocate device, out of memory.");
1734 if (!try_module_get(THIS_MODULE))
1735 goto bad_module_get;
1737 /* get a minor number for the dev */
1738 if (minor == DM_ANY_MINOR)
1739 r = next_free_minor(&minor);
1741 r = specific_minor(minor);
1745 init_rwsem(&md->io_lock);
1746 mutex_init(&md->suspend_lock);
1747 spin_lock_init(&md->deferred_lock);
1748 rwlock_init(&md->map_lock);
1749 atomic_set(&md->holders, 1);
1750 atomic_set(&md->open_count, 0);
1751 atomic_set(&md->event_nr, 0);
1752 atomic_set(&md->uevent_seq, 0);
1753 INIT_LIST_HEAD(&md->uevent_list);
1754 spin_lock_init(&md->uevent_lock);
1756 md->queue = blk_init_queue(dm_request_fn, NULL);
1761 * Request-based dm devices cannot be stacked on top of bio-based dm
1762 * devices. The type of this dm device has not been decided yet,
1763 * although we initialized the queue using blk_init_queue().
1764 * The type is decided at the first table loading time.
1765 * To prevent problematic device stacking, clear the queue flag
1766 * for request stacking support until then.
1768 * This queue is new, so no concurrency on the queue_flags.
1770 queue_flag_clear_unlocked(QUEUE_FLAG_STACKABLE, md->queue);
1771 md->saved_make_request_fn = md->queue->make_request_fn;
1772 md->queue->queuedata = md;
1773 md->queue->backing_dev_info.congested_fn = dm_any_congested;
1774 md->queue->backing_dev_info.congested_data = md;
1775 blk_queue_make_request(md->queue, dm_request);
1776 blk_queue_bounce_limit(md->queue, BLK_BOUNCE_ANY);
1777 md->queue->unplug_fn = dm_unplug_all;
1778 blk_queue_merge_bvec(md->queue, dm_merge_bvec);
1779 blk_queue_softirq_done(md->queue, dm_softirq_done);
1780 blk_queue_prep_rq(md->queue, dm_prep_fn);
1781 blk_queue_lld_busy(md->queue, dm_lld_busy);
1783 md->disk = alloc_disk(1);
1787 atomic_set(&md->pending[0], 0);
1788 atomic_set(&md->pending[1], 0);
1789 init_waitqueue_head(&md->wait);
1790 INIT_WORK(&md->work, dm_wq_work);
1791 init_waitqueue_head(&md->eventq);
1793 md->disk->major = _major;
1794 md->disk->first_minor = minor;
1795 md->disk->fops = &dm_blk_dops;
1796 md->disk->queue = md->queue;
1797 md->disk->private_data = md;
1798 sprintf(md->disk->disk_name, "dm-%d", minor);
1800 format_dev_t(md->name, MKDEV(_major, minor));
1802 md->wq = create_singlethread_workqueue("kdmflush");
1806 md->bdev = bdget_disk(md->disk, 0);
1810 /* Populate the mapping, nobody knows we exist yet */
1811 spin_lock(&_minor_lock);
1812 old_md = idr_replace(&_minor_idr, md, minor);
1813 spin_unlock(&_minor_lock);
1815 BUG_ON(old_md != MINOR_ALLOCED);
1820 destroy_workqueue(md->wq);
1822 del_gendisk(md->disk);
1825 blk_cleanup_queue(md->queue);
1829 module_put(THIS_MODULE);
1835 static void unlock_fs(struct mapped_device *md);
1837 static void free_dev(struct mapped_device *md)
1839 int minor = MINOR(disk_devt(md->disk));
1843 destroy_workqueue(md->wq);
1845 mempool_destroy(md->tio_pool);
1847 mempool_destroy(md->io_pool);
1849 bioset_free(md->bs);
1850 blk_integrity_unregister(md->disk);
1851 del_gendisk(md->disk);
1854 spin_lock(&_minor_lock);
1855 md->disk->private_data = NULL;
1856 spin_unlock(&_minor_lock);
1859 blk_cleanup_queue(md->queue);
1860 module_put(THIS_MODULE);
1864 static void __bind_mempools(struct mapped_device *md, struct dm_table *t)
1866 struct dm_md_mempools *p;
1868 if (md->io_pool && md->tio_pool && md->bs)
1869 /* the md already has necessary mempools */
1872 p = dm_table_get_md_mempools(t);
1873 BUG_ON(!p || md->io_pool || md->tio_pool || md->bs);
1875 md->io_pool = p->io_pool;
1877 md->tio_pool = p->tio_pool;
1883 /* mempool bind completed, now no need any mempools in the table */
1884 dm_table_free_md_mempools(t);
1888 * Bind a table to the device.
1890 static void event_callback(void *context)
1892 unsigned long flags;
1894 struct mapped_device *md = (struct mapped_device *) context;
1896 spin_lock_irqsave(&md->uevent_lock, flags);
1897 list_splice_init(&md->uevent_list, &uevents);
1898 spin_unlock_irqrestore(&md->uevent_lock, flags);
1900 dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj);
1902 atomic_inc(&md->event_nr);
1903 wake_up(&md->eventq);
1906 static void __set_size(struct mapped_device *md, sector_t size)
1908 set_capacity(md->disk, size);
1910 mutex_lock(&md->bdev->bd_inode->i_mutex);
1911 i_size_write(md->bdev->bd_inode, (loff_t)size << SECTOR_SHIFT);
1912 mutex_unlock(&md->bdev->bd_inode->i_mutex);
1915 static int __bind(struct mapped_device *md, struct dm_table *t,
1916 struct queue_limits *limits)
1918 struct request_queue *q = md->queue;
1920 unsigned long flags;
1922 size = dm_table_get_size(t);
1925 * Wipe any geometry if the size of the table changed.
1927 if (size != get_capacity(md->disk))
1928 memset(&md->geometry, 0, sizeof(md->geometry));
1930 __set_size(md, size);
1933 dm_table_destroy(t);
1937 dm_table_event_callback(t, event_callback, md);
1940 * The queue hasn't been stopped yet, if the old table type wasn't
1941 * for request-based during suspension. So stop it to prevent
1942 * I/O mapping before resume.
1943 * This must be done before setting the queue restrictions,
1944 * because request-based dm may be run just after the setting.
1946 if (dm_table_request_based(t) && !blk_queue_stopped(q))
1949 __bind_mempools(md, t);
1951 write_lock_irqsave(&md->map_lock, flags);
1953 dm_table_set_restrictions(t, q, limits);
1954 write_unlock_irqrestore(&md->map_lock, flags);
1959 static void __unbind(struct mapped_device *md)
1961 struct dm_table *map = md->map;
1962 unsigned long flags;
1967 dm_table_event_callback(map, NULL, NULL);
1968 write_lock_irqsave(&md->map_lock, flags);
1970 write_unlock_irqrestore(&md->map_lock, flags);
1971 dm_table_destroy(map);
1975 * Constructor for a new device.
1977 int dm_create(int minor, struct mapped_device **result)
1979 struct mapped_device *md;
1981 md = alloc_dev(minor);
1991 static struct mapped_device *dm_find_md(dev_t dev)
1993 struct mapped_device *md;
1994 unsigned minor = MINOR(dev);
1996 if (MAJOR(dev) != _major || minor >= (1 << MINORBITS))
1999 spin_lock(&_minor_lock);
2001 md = idr_find(&_minor_idr, minor);
2002 if (md && (md == MINOR_ALLOCED ||
2003 (MINOR(disk_devt(dm_disk(md))) != minor) ||
2004 test_bit(DMF_FREEING, &md->flags))) {
2010 spin_unlock(&_minor_lock);
2015 struct mapped_device *dm_get_md(dev_t dev)
2017 struct mapped_device *md = dm_find_md(dev);
2025 void *dm_get_mdptr(struct mapped_device *md)
2027 return md->interface_ptr;
2030 void dm_set_mdptr(struct mapped_device *md, void *ptr)
2032 md->interface_ptr = ptr;
2035 void dm_get(struct mapped_device *md)
2037 atomic_inc(&md->holders);
2040 const char *dm_device_name(struct mapped_device *md)
2044 EXPORT_SYMBOL_GPL(dm_device_name);
2046 void dm_put(struct mapped_device *md)
2048 struct dm_table *map;
2050 BUG_ON(test_bit(DMF_FREEING, &md->flags));
2052 if (atomic_dec_and_lock(&md->holders, &_minor_lock)) {
2053 map = dm_get_table(md);
2054 idr_replace(&_minor_idr, MINOR_ALLOCED,
2055 MINOR(disk_devt(dm_disk(md))));
2056 set_bit(DMF_FREEING, &md->flags);
2057 spin_unlock(&_minor_lock);
2058 if (!dm_suspended(md)) {
2059 dm_table_presuspend_targets(map);
2060 dm_table_postsuspend_targets(map);
2068 EXPORT_SYMBOL_GPL(dm_put);
2070 static int dm_wait_for_completion(struct mapped_device *md, int interruptible)
2073 DECLARE_WAITQUEUE(wait, current);
2074 struct request_queue *q = md->queue;
2075 unsigned long flags;
2077 dm_unplug_all(md->queue);
2079 add_wait_queue(&md->wait, &wait);
2082 set_current_state(interruptible);
2085 if (dm_request_based(md)) {
2086 spin_lock_irqsave(q->queue_lock, flags);
2087 if (!queue_in_flight(q)) {
2088 spin_unlock_irqrestore(q->queue_lock, flags);
2091 spin_unlock_irqrestore(q->queue_lock, flags);
2092 } else if (!md_in_flight(md))
2095 if (interruptible == TASK_INTERRUPTIBLE &&
2096 signal_pending(current)) {
2103 set_current_state(TASK_RUNNING);
2105 remove_wait_queue(&md->wait, &wait);
2110 static void dm_flush(struct mapped_device *md)
2112 dm_wait_for_completion(md, TASK_UNINTERRUPTIBLE);
2114 bio_init(&md->barrier_bio);
2115 md->barrier_bio.bi_bdev = md->bdev;
2116 md->barrier_bio.bi_rw = WRITE_BARRIER;
2117 __split_and_process_bio(md, &md->barrier_bio);
2119 dm_wait_for_completion(md, TASK_UNINTERRUPTIBLE);
2122 static void process_barrier(struct mapped_device *md, struct bio *bio)
2124 md->barrier_error = 0;
2128 if (!bio_empty_barrier(bio)) {
2129 __split_and_process_bio(md, bio);
2133 if (md->barrier_error != DM_ENDIO_REQUEUE)
2134 bio_endio(bio, md->barrier_error);
2136 spin_lock_irq(&md->deferred_lock);
2137 bio_list_add_head(&md->deferred, bio);
2138 spin_unlock_irq(&md->deferred_lock);
2143 * Process the deferred bios
2145 static void dm_wq_work(struct work_struct *work)
2147 struct mapped_device *md = container_of(work, struct mapped_device,
2151 down_write(&md->io_lock);
2153 while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
2154 spin_lock_irq(&md->deferred_lock);
2155 c = bio_list_pop(&md->deferred);
2156 spin_unlock_irq(&md->deferred_lock);
2159 clear_bit(DMF_QUEUE_IO_TO_THREAD, &md->flags);
2163 up_write(&md->io_lock);
2165 if (dm_request_based(md))
2166 generic_make_request(c);
2168 if (bio_rw_flagged(c, BIO_RW_BARRIER))
2169 process_barrier(md, c);
2171 __split_and_process_bio(md, c);
2174 down_write(&md->io_lock);
2177 up_write(&md->io_lock);
2180 static void dm_queue_flush(struct mapped_device *md)
2182 clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2183 smp_mb__after_clear_bit();
2184 queue_work(md->wq, &md->work);
2188 * Swap in a new table (destroying old one).
2190 int dm_swap_table(struct mapped_device *md, struct dm_table *table)
2192 struct queue_limits limits;
2195 mutex_lock(&md->suspend_lock);
2197 /* device must be suspended */
2198 if (!dm_suspended(md))
2201 r = dm_calculate_queue_limits(table, &limits);
2205 /* cannot change the device type, once a table is bound */
2207 (dm_table_get_type(md->map) != dm_table_get_type(table))) {
2208 DMWARN("can't change the device type after a table is bound");
2213 r = __bind(md, table, &limits);
2216 mutex_unlock(&md->suspend_lock);
2221 * Functions to lock and unlock any filesystem running on the
2224 static int lock_fs(struct mapped_device *md)
2228 WARN_ON(md->frozen_sb);
2230 md->frozen_sb = freeze_bdev(md->bdev);
2231 if (IS_ERR(md->frozen_sb)) {
2232 r = PTR_ERR(md->frozen_sb);
2233 md->frozen_sb = NULL;
2237 set_bit(DMF_FROZEN, &md->flags);
2242 static void unlock_fs(struct mapped_device *md)
2244 if (!test_bit(DMF_FROZEN, &md->flags))
2247 thaw_bdev(md->bdev, md->frozen_sb);
2248 md->frozen_sb = NULL;
2249 clear_bit(DMF_FROZEN, &md->flags);
2253 * We need to be able to change a mapping table under a mounted
2254 * filesystem. For example we might want to move some data in
2255 * the background. Before the table can be swapped with
2256 * dm_bind_table, dm_suspend must be called to flush any in
2257 * flight bios and ensure that any further io gets deferred.
2260 * Suspend mechanism in request-based dm.
2262 * 1. Flush all I/Os by lock_fs() if needed.
2263 * 2. Stop dispatching any I/O by stopping the request_queue.
2264 * 3. Wait for all in-flight I/Os to be completed or requeued.
2266 * To abort suspend, start the request_queue.
2268 int dm_suspend(struct mapped_device *md, unsigned suspend_flags)
2270 struct dm_table *map = NULL;
2272 int do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG ? 1 : 0;
2273 int noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG ? 1 : 0;
2275 mutex_lock(&md->suspend_lock);
2277 if (dm_suspended(md)) {
2282 map = dm_get_table(md);
2285 * DMF_NOFLUSH_SUSPENDING must be set before presuspend.
2286 * This flag is cleared before dm_suspend returns.
2289 set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2291 /* This does not get reverted if there's an error later. */
2292 dm_table_presuspend_targets(map);
2295 * Flush I/O to the device.
2296 * Any I/O submitted after lock_fs() may not be flushed.
2297 * noflush takes precedence over do_lockfs.
2298 * (lock_fs() flushes I/Os and waits for them to complete.)
2300 if (!noflush && do_lockfs) {
2307 * Here we must make sure that no processes are submitting requests
2308 * to target drivers i.e. no one may be executing
2309 * __split_and_process_bio. This is called from dm_request and
2312 * To get all processes out of __split_and_process_bio in dm_request,
2313 * we take the write lock. To prevent any process from reentering
2314 * __split_and_process_bio from dm_request, we set
2315 * DMF_QUEUE_IO_TO_THREAD.
2317 * To quiesce the thread (dm_wq_work), we set DMF_BLOCK_IO_FOR_SUSPEND
2318 * and call flush_workqueue(md->wq). flush_workqueue will wait until
2319 * dm_wq_work exits and DMF_BLOCK_IO_FOR_SUSPEND will prevent any
2320 * further calls to __split_and_process_bio from dm_wq_work.
2322 down_write(&md->io_lock);
2323 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2324 set_bit(DMF_QUEUE_IO_TO_THREAD, &md->flags);
2325 up_write(&md->io_lock);
2327 flush_workqueue(md->wq);
2329 if (dm_request_based(md))
2330 stop_queue(md->queue);
2333 * At this point no more requests are entering target request routines.
2334 * We call dm_wait_for_completion to wait for all existing requests
2337 r = dm_wait_for_completion(md, TASK_INTERRUPTIBLE);
2339 down_write(&md->io_lock);
2341 clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2342 up_write(&md->io_lock);
2344 /* were we interrupted ? */
2348 if (dm_request_based(md))
2349 start_queue(md->queue);
2352 goto out; /* pushback list is already flushed, so skip flush */
2356 * If dm_wait_for_completion returned 0, the device is completely
2357 * quiescent now. There is no request-processing activity. All new
2358 * requests are being added to md->deferred list.
2361 dm_table_postsuspend_targets(map);
2363 set_bit(DMF_SUSPENDED, &md->flags);
2369 mutex_unlock(&md->suspend_lock);
2373 int dm_resume(struct mapped_device *md)
2376 struct dm_table *map = NULL;
2378 mutex_lock(&md->suspend_lock);
2379 if (!dm_suspended(md))
2382 map = dm_get_table(md);
2383 if (!map || !dm_table_get_size(map))
2386 r = dm_table_resume_targets(map);
2393 * Flushing deferred I/Os must be done after targets are resumed
2394 * so that mapping of targets can work correctly.
2395 * Request-based dm is queueing the deferred I/Os in its request_queue.
2397 if (dm_request_based(md))
2398 start_queue(md->queue);
2402 clear_bit(DMF_SUSPENDED, &md->flags);
2404 dm_table_unplug_all(map);
2408 mutex_unlock(&md->suspend_lock);
2413 /*-----------------------------------------------------------------
2414 * Event notification.
2415 *---------------------------------------------------------------*/
2416 void dm_kobject_uevent(struct mapped_device *md, enum kobject_action action,
2419 char udev_cookie[DM_COOKIE_LENGTH];
2420 char *envp[] = { udev_cookie, NULL };
2423 kobject_uevent(&disk_to_dev(md->disk)->kobj, action);
2425 snprintf(udev_cookie, DM_COOKIE_LENGTH, "%s=%u",
2426 DM_COOKIE_ENV_VAR_NAME, cookie);
2427 kobject_uevent_env(&disk_to_dev(md->disk)->kobj, action, envp);
2431 uint32_t dm_next_uevent_seq(struct mapped_device *md)
2433 return atomic_add_return(1, &md->uevent_seq);
2436 uint32_t dm_get_event_nr(struct mapped_device *md)
2438 return atomic_read(&md->event_nr);
2441 int dm_wait_event(struct mapped_device *md, int event_nr)
2443 return wait_event_interruptible(md->eventq,
2444 (event_nr != atomic_read(&md->event_nr)));
2447 void dm_uevent_add(struct mapped_device *md, struct list_head *elist)
2449 unsigned long flags;
2451 spin_lock_irqsave(&md->uevent_lock, flags);
2452 list_add(elist, &md->uevent_list);
2453 spin_unlock_irqrestore(&md->uevent_lock, flags);
2457 * The gendisk is only valid as long as you have a reference
2460 struct gendisk *dm_disk(struct mapped_device *md)
2465 struct kobject *dm_kobject(struct mapped_device *md)
2471 * struct mapped_device should not be exported outside of dm.c
2472 * so use this check to verify that kobj is part of md structure
2474 struct mapped_device *dm_get_from_kobject(struct kobject *kobj)
2476 struct mapped_device *md;
2478 md = container_of(kobj, struct mapped_device, kobj);
2479 if (&md->kobj != kobj)
2482 if (test_bit(DMF_FREEING, &md->flags) ||
2483 test_bit(DMF_DELETING, &md->flags))
2490 int dm_suspended(struct mapped_device *md)
2492 return test_bit(DMF_SUSPENDED, &md->flags);
2495 int dm_noflush_suspending(struct dm_target *ti)
2497 struct mapped_device *md = dm_table_get_md(ti->table);
2498 int r = __noflush_suspending(md);
2504 EXPORT_SYMBOL_GPL(dm_noflush_suspending);
2506 struct dm_md_mempools *dm_alloc_md_mempools(unsigned type)
2508 struct dm_md_mempools *pools = kmalloc(sizeof(*pools), GFP_KERNEL);
2513 pools->io_pool = (type == DM_TYPE_BIO_BASED) ?
2514 mempool_create_slab_pool(MIN_IOS, _io_cache) :
2515 mempool_create_slab_pool(MIN_IOS, _rq_bio_info_cache);
2516 if (!pools->io_pool)
2517 goto free_pools_and_out;
2519 pools->tio_pool = (type == DM_TYPE_BIO_BASED) ?
2520 mempool_create_slab_pool(MIN_IOS, _tio_cache) :
2521 mempool_create_slab_pool(MIN_IOS, _rq_tio_cache);
2522 if (!pools->tio_pool)
2523 goto free_io_pool_and_out;
2525 pools->bs = (type == DM_TYPE_BIO_BASED) ?
2526 bioset_create(16, 0) : bioset_create(MIN_IOS, 0);
2528 goto free_tio_pool_and_out;
2532 free_tio_pool_and_out:
2533 mempool_destroy(pools->tio_pool);
2535 free_io_pool_and_out:
2536 mempool_destroy(pools->io_pool);
2544 void dm_free_md_mempools(struct dm_md_mempools *pools)
2550 mempool_destroy(pools->io_pool);
2552 if (pools->tio_pool)
2553 mempool_destroy(pools->tio_pool);
2556 bioset_free(pools->bs);
2561 static const struct block_device_operations dm_blk_dops = {
2562 .open = dm_blk_open,
2563 .release = dm_blk_close,
2564 .ioctl = dm_blk_ioctl,
2565 .getgeo = dm_blk_getgeo,
2566 .owner = THIS_MODULE
2569 EXPORT_SYMBOL(dm_get_mapinfo);
2574 module_init(dm_init);
2575 module_exit(dm_exit);
2577 module_param(major, uint, 0);
2578 MODULE_PARM_DESC(major, "The major number of the device mapper");
2579 MODULE_DESCRIPTION(DM_NAME " driver");
2580 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
2581 MODULE_LICENSE("GPL");