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/smp_lock.h>
19 #include <linux/mempool.h>
20 #include <linux/slab.h>
21 #include <linux/idr.h>
22 #include <linux/hdreg.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_SPINLOCK(_minor_lock);
43 * One of these is allocated per bio.
46 struct mapped_device *md;
50 unsigned long start_time;
51 spinlock_t endio_lock;
56 * One of these is allocated per target within a bio. Hopefully
57 * this will be simplified out one day.
66 * For request-based dm.
67 * One of these is allocated per request.
69 struct dm_rq_target_io {
70 struct mapped_device *md;
72 struct request *orig, clone;
78 * For request-based dm.
79 * One of these is allocated per bio.
81 struct dm_rq_clone_bio_info {
83 struct dm_rq_target_io *tio;
86 union map_info *dm_get_mapinfo(struct bio *bio)
88 if (bio && bio->bi_private)
89 return &((struct dm_target_io *)bio->bi_private)->info;
93 union map_info *dm_get_rq_mapinfo(struct request *rq)
95 if (rq && rq->end_io_data)
96 return &((struct dm_rq_target_io *)rq->end_io_data)->info;
99 EXPORT_SYMBOL_GPL(dm_get_rq_mapinfo);
101 #define MINOR_ALLOCED ((void *)-1)
104 * Bits for the md->flags field.
106 #define DMF_BLOCK_IO_FOR_SUSPEND 0
107 #define DMF_SUSPENDED 1
109 #define DMF_FREEING 3
110 #define DMF_DELETING 4
111 #define DMF_NOFLUSH_SUSPENDING 5
112 #define DMF_QUEUE_IO_TO_THREAD 6
115 * Work processed by per-device workqueue.
117 struct mapped_device {
118 struct rw_semaphore io_lock;
119 struct mutex suspend_lock;
126 struct request_queue *queue;
127 struct gendisk *disk;
133 * A list of ios that arrived while we were suspended.
136 wait_queue_head_t wait;
137 struct work_struct work;
138 struct bio_list deferred;
139 spinlock_t deferred_lock;
142 * An error from the barrier request currently being processed.
147 * Protect barrier_error from concurrent endio processing
148 * in request-based dm.
150 spinlock_t barrier_error_lock;
153 * Processing queue (flush/barriers)
155 struct workqueue_struct *wq;
156 struct work_struct barrier_work;
158 /* A pointer to the currently processing pre/post flush request */
159 struct request *flush_request;
162 * The current mapping.
164 struct dm_table *map;
167 * io objects are allocated from here.
178 wait_queue_head_t eventq;
180 struct list_head uevent_list;
181 spinlock_t uevent_lock; /* Protect access to uevent_list */
184 * freeze/thaw support require holding onto a super block
186 struct super_block *frozen_sb;
187 struct block_device *bdev;
189 /* forced geometry settings */
190 struct hd_geometry geometry;
192 /* For saving the address of __make_request for request based dm */
193 make_request_fn *saved_make_request_fn;
198 /* zero-length barrier that will be cloned and submitted to targets */
199 struct bio barrier_bio;
203 * For mempools pre-allocation at the table loading time.
205 struct dm_md_mempools {
212 static struct kmem_cache *_io_cache;
213 static struct kmem_cache *_tio_cache;
214 static struct kmem_cache *_rq_tio_cache;
215 static struct kmem_cache *_rq_bio_info_cache;
217 static int __init local_init(void)
221 /* allocate a slab for the dm_ios */
222 _io_cache = KMEM_CACHE(dm_io, 0);
226 /* allocate a slab for the target ios */
227 _tio_cache = KMEM_CACHE(dm_target_io, 0);
229 goto out_free_io_cache;
231 _rq_tio_cache = KMEM_CACHE(dm_rq_target_io, 0);
233 goto out_free_tio_cache;
235 _rq_bio_info_cache = KMEM_CACHE(dm_rq_clone_bio_info, 0);
236 if (!_rq_bio_info_cache)
237 goto out_free_rq_tio_cache;
239 r = dm_uevent_init();
241 goto out_free_rq_bio_info_cache;
244 r = register_blkdev(_major, _name);
246 goto out_uevent_exit;
255 out_free_rq_bio_info_cache:
256 kmem_cache_destroy(_rq_bio_info_cache);
257 out_free_rq_tio_cache:
258 kmem_cache_destroy(_rq_tio_cache);
260 kmem_cache_destroy(_tio_cache);
262 kmem_cache_destroy(_io_cache);
267 static void local_exit(void)
269 kmem_cache_destroy(_rq_bio_info_cache);
270 kmem_cache_destroy(_rq_tio_cache);
271 kmem_cache_destroy(_tio_cache);
272 kmem_cache_destroy(_io_cache);
273 unregister_blkdev(_major, _name);
278 DMINFO("cleaned up");
281 static int (*_inits[])(void) __initdata = {
291 static void (*_exits[])(void) = {
301 static int __init dm_init(void)
303 const int count = ARRAY_SIZE(_inits);
307 for (i = 0; i < count; i++) {
322 static void __exit dm_exit(void)
324 int i = ARRAY_SIZE(_exits);
331 * Block device functions
333 int dm_deleting_md(struct mapped_device *md)
335 return test_bit(DMF_DELETING, &md->flags);
338 static int dm_blk_open(struct block_device *bdev, fmode_t mode)
340 struct mapped_device *md;
343 spin_lock(&_minor_lock);
345 md = bdev->bd_disk->private_data;
349 if (test_bit(DMF_FREEING, &md->flags) ||
350 dm_deleting_md(md)) {
356 atomic_inc(&md->open_count);
359 spin_unlock(&_minor_lock);
362 return md ? 0 : -ENXIO;
365 static int dm_blk_close(struct gendisk *disk, fmode_t mode)
367 struct mapped_device *md = disk->private_data;
370 atomic_dec(&md->open_count);
377 int dm_open_count(struct mapped_device *md)
379 return atomic_read(&md->open_count);
383 * Guarantees nothing is using the device before it's deleted.
385 int dm_lock_for_deletion(struct mapped_device *md)
389 spin_lock(&_minor_lock);
391 if (dm_open_count(md))
394 set_bit(DMF_DELETING, &md->flags);
396 spin_unlock(&_minor_lock);
401 static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo)
403 struct mapped_device *md = bdev->bd_disk->private_data;
405 return dm_get_geometry(md, geo);
408 static int dm_blk_ioctl(struct block_device *bdev, fmode_t mode,
409 unsigned int cmd, unsigned long arg)
411 struct mapped_device *md = bdev->bd_disk->private_data;
412 struct dm_table *map = dm_get_live_table(md);
413 struct dm_target *tgt;
416 if (!map || !dm_table_get_size(map))
419 /* We only support devices that have a single target */
420 if (dm_table_get_num_targets(map) != 1)
423 tgt = dm_table_get_target(map, 0);
425 if (dm_suspended_md(md)) {
430 if (tgt->type->ioctl)
431 r = tgt->type->ioctl(tgt, cmd, arg);
439 static struct dm_io *alloc_io(struct mapped_device *md)
441 return mempool_alloc(md->io_pool, GFP_NOIO);
444 static void free_io(struct mapped_device *md, struct dm_io *io)
446 mempool_free(io, md->io_pool);
449 static void free_tio(struct mapped_device *md, struct dm_target_io *tio)
451 mempool_free(tio, md->tio_pool);
454 static struct dm_rq_target_io *alloc_rq_tio(struct mapped_device *md,
457 return mempool_alloc(md->tio_pool, gfp_mask);
460 static void free_rq_tio(struct dm_rq_target_io *tio)
462 mempool_free(tio, tio->md->tio_pool);
465 static struct dm_rq_clone_bio_info *alloc_bio_info(struct mapped_device *md)
467 return mempool_alloc(md->io_pool, GFP_ATOMIC);
470 static void free_bio_info(struct dm_rq_clone_bio_info *info)
472 mempool_free(info, info->tio->md->io_pool);
475 static int md_in_flight(struct mapped_device *md)
477 return atomic_read(&md->pending[READ]) +
478 atomic_read(&md->pending[WRITE]);
481 static void start_io_acct(struct dm_io *io)
483 struct mapped_device *md = io->md;
485 int rw = bio_data_dir(io->bio);
487 io->start_time = jiffies;
489 cpu = part_stat_lock();
490 part_round_stats(cpu, &dm_disk(md)->part0);
492 dm_disk(md)->part0.in_flight[rw] = atomic_inc_return(&md->pending[rw]);
495 static void end_io_acct(struct dm_io *io)
497 struct mapped_device *md = io->md;
498 struct bio *bio = io->bio;
499 unsigned long duration = jiffies - io->start_time;
501 int rw = bio_data_dir(bio);
503 cpu = part_stat_lock();
504 part_round_stats(cpu, &dm_disk(md)->part0);
505 part_stat_add(cpu, &dm_disk(md)->part0, ticks[rw], duration);
509 * After this is decremented the bio must not be touched if it is
512 dm_disk(md)->part0.in_flight[rw] = pending =
513 atomic_dec_return(&md->pending[rw]);
514 pending += atomic_read(&md->pending[rw^0x1]);
516 /* nudge anyone waiting on suspend queue */
522 * Add the bio to the list of deferred io.
524 static void queue_io(struct mapped_device *md, struct bio *bio)
526 down_write(&md->io_lock);
528 spin_lock_irq(&md->deferred_lock);
529 bio_list_add(&md->deferred, bio);
530 spin_unlock_irq(&md->deferred_lock);
532 if (!test_and_set_bit(DMF_QUEUE_IO_TO_THREAD, &md->flags))
533 queue_work(md->wq, &md->work);
535 up_write(&md->io_lock);
539 * Everyone (including functions in this file), should use this
540 * function to access the md->map field, and make sure they call
541 * dm_table_put() when finished.
543 struct dm_table *dm_get_live_table(struct mapped_device *md)
548 read_lock_irqsave(&md->map_lock, flags);
552 read_unlock_irqrestore(&md->map_lock, flags);
558 * Get the geometry associated with a dm device
560 int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo)
568 * Set the geometry of a device.
570 int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo)
572 sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors;
574 if (geo->start > sz) {
575 DMWARN("Start sector is beyond the geometry limits.");
584 /*-----------------------------------------------------------------
586 * A more elegant soln is in the works that uses the queue
587 * merge fn, unfortunately there are a couple of changes to
588 * the block layer that I want to make for this. So in the
589 * interests of getting something for people to use I give
590 * you this clearly demarcated crap.
591 *---------------------------------------------------------------*/
593 static int __noflush_suspending(struct mapped_device *md)
595 return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
599 * Decrements the number of outstanding ios that a bio has been
600 * cloned into, completing the original io if necc.
602 static void dec_pending(struct dm_io *io, int error)
607 struct mapped_device *md = io->md;
609 /* Push-back supersedes any I/O errors */
610 if (unlikely(error)) {
611 spin_lock_irqsave(&io->endio_lock, flags);
612 if (!(io->error > 0 && __noflush_suspending(md)))
614 spin_unlock_irqrestore(&io->endio_lock, flags);
617 if (atomic_dec_and_test(&io->io_count)) {
618 if (io->error == DM_ENDIO_REQUEUE) {
620 * Target requested pushing back the I/O.
622 spin_lock_irqsave(&md->deferred_lock, flags);
623 if (__noflush_suspending(md)) {
624 if (!(io->bio->bi_rw & REQ_HARDBARRIER))
625 bio_list_add_head(&md->deferred,
628 /* noflush suspend was interrupted. */
630 spin_unlock_irqrestore(&md->deferred_lock, flags);
633 io_error = io->error;
636 if (bio->bi_rw & REQ_HARDBARRIER) {
638 * There can be just one barrier request so we use
639 * a per-device variable for error reporting.
640 * Note that you can't touch the bio after end_io_acct
642 if (!md->barrier_error && io_error != -EOPNOTSUPP)
643 md->barrier_error = io_error;
650 if (io_error != DM_ENDIO_REQUEUE) {
651 trace_block_bio_complete(md->queue, bio);
653 bio_endio(bio, io_error);
659 static void clone_endio(struct bio *bio, int error)
662 struct dm_target_io *tio = bio->bi_private;
663 struct dm_io *io = tio->io;
664 struct mapped_device *md = tio->io->md;
665 dm_endio_fn endio = tio->ti->type->end_io;
667 if (!bio_flagged(bio, BIO_UPTODATE) && !error)
671 r = endio(tio->ti, bio, error, &tio->info);
672 if (r < 0 || r == DM_ENDIO_REQUEUE)
674 * error and requeue request are handled
678 else if (r == DM_ENDIO_INCOMPLETE)
679 /* The target will handle the io */
682 DMWARN("unimplemented target endio return value: %d", r);
688 * Store md for cleanup instead of tio which is about to get freed.
690 bio->bi_private = md->bs;
694 dec_pending(io, error);
698 * Partial completion handling for request-based dm
700 static void end_clone_bio(struct bio *clone, int error)
702 struct dm_rq_clone_bio_info *info = clone->bi_private;
703 struct dm_rq_target_io *tio = info->tio;
704 struct bio *bio = info->orig;
705 unsigned int nr_bytes = info->orig->bi_size;
711 * An error has already been detected on the request.
712 * Once error occurred, just let clone->end_io() handle
718 * Don't notice the error to the upper layer yet.
719 * The error handling decision is made by the target driver,
720 * when the request is completed.
727 * I/O for the bio successfully completed.
728 * Notice the data completion to the upper layer.
732 * bios are processed from the head of the list.
733 * So the completing bio should always be rq->bio.
734 * If it's not, something wrong is happening.
736 if (tio->orig->bio != bio)
737 DMERR("bio completion is going in the middle of the request");
740 * Update the original request.
741 * Do not use blk_end_request() here, because it may complete
742 * the original request before the clone, and break the ordering.
744 blk_update_request(tio->orig, 0, nr_bytes);
747 static void store_barrier_error(struct mapped_device *md, int error)
751 spin_lock_irqsave(&md->barrier_error_lock, flags);
753 * Basically, the first error is taken, but:
754 * -EOPNOTSUPP supersedes any I/O error.
755 * Requeue request supersedes any I/O error but -EOPNOTSUPP.
757 if (!md->barrier_error || error == -EOPNOTSUPP ||
758 (md->barrier_error != -EOPNOTSUPP &&
759 error == DM_ENDIO_REQUEUE))
760 md->barrier_error = error;
761 spin_unlock_irqrestore(&md->barrier_error_lock, flags);
765 * Don't touch any member of the md after calling this function because
766 * the md may be freed in dm_put() at the end of this function.
767 * Or do dm_get() before calling this function and dm_put() later.
769 static void rq_completed(struct mapped_device *md, int rw, int run_queue)
771 atomic_dec(&md->pending[rw]);
773 /* nudge anyone waiting on suspend queue */
774 if (!md_in_flight(md))
778 blk_run_queue(md->queue);
781 * dm_put() must be at the end of this function. See the comment above
786 static void free_rq_clone(struct request *clone)
788 struct dm_rq_target_io *tio = clone->end_io_data;
790 blk_rq_unprep_clone(clone);
795 * Complete the clone and the original request.
796 * Must be called without queue lock.
798 static void dm_end_request(struct request *clone, int error)
800 int rw = rq_data_dir(clone);
802 bool is_barrier = clone->cmd_flags & REQ_HARDBARRIER;
803 struct dm_rq_target_io *tio = clone->end_io_data;
804 struct mapped_device *md = tio->md;
805 struct request *rq = tio->orig;
807 if (rq->cmd_type == REQ_TYPE_BLOCK_PC && !is_barrier) {
808 rq->errors = clone->errors;
809 rq->resid_len = clone->resid_len;
813 * We are using the sense buffer of the original
815 * So setting the length of the sense data is enough.
817 rq->sense_len = clone->sense_len;
820 free_rq_clone(clone);
822 if (unlikely(is_barrier)) {
824 store_barrier_error(md, error);
827 blk_end_request_all(rq, error);
829 rq_completed(md, rw, run_queue);
832 static void dm_unprep_request(struct request *rq)
834 struct request *clone = rq->special;
837 rq->cmd_flags &= ~REQ_DONTPREP;
839 free_rq_clone(clone);
843 * Requeue the original request of a clone.
845 void dm_requeue_unmapped_request(struct request *clone)
847 int rw = rq_data_dir(clone);
848 struct dm_rq_target_io *tio = clone->end_io_data;
849 struct mapped_device *md = tio->md;
850 struct request *rq = tio->orig;
851 struct request_queue *q = rq->q;
854 if (unlikely(clone->cmd_flags & REQ_HARDBARRIER)) {
856 * Barrier clones share an original request.
857 * Leave it to dm_end_request(), which handles this special
860 dm_end_request(clone, DM_ENDIO_REQUEUE);
864 dm_unprep_request(rq);
866 spin_lock_irqsave(q->queue_lock, flags);
867 if (elv_queue_empty(q))
869 blk_requeue_request(q, rq);
870 spin_unlock_irqrestore(q->queue_lock, flags);
872 rq_completed(md, rw, 0);
874 EXPORT_SYMBOL_GPL(dm_requeue_unmapped_request);
876 static void __stop_queue(struct request_queue *q)
881 static void stop_queue(struct request_queue *q)
885 spin_lock_irqsave(q->queue_lock, flags);
887 spin_unlock_irqrestore(q->queue_lock, flags);
890 static void __start_queue(struct request_queue *q)
892 if (blk_queue_stopped(q))
896 static void start_queue(struct request_queue *q)
900 spin_lock_irqsave(q->queue_lock, flags);
902 spin_unlock_irqrestore(q->queue_lock, flags);
905 static void dm_done(struct request *clone, int error, bool mapped)
908 struct dm_rq_target_io *tio = clone->end_io_data;
909 dm_request_endio_fn rq_end_io = tio->ti->type->rq_end_io;
911 if (mapped && rq_end_io)
912 r = rq_end_io(tio->ti, clone, error, &tio->info);
915 /* The target wants to complete the I/O */
916 dm_end_request(clone, r);
917 else if (r == DM_ENDIO_INCOMPLETE)
918 /* The target will handle the I/O */
920 else if (r == DM_ENDIO_REQUEUE)
921 /* The target wants to requeue the I/O */
922 dm_requeue_unmapped_request(clone);
924 DMWARN("unimplemented target endio return value: %d", r);
930 * Request completion handler for request-based dm
932 static void dm_softirq_done(struct request *rq)
935 struct request *clone = rq->completion_data;
936 struct dm_rq_target_io *tio = clone->end_io_data;
938 if (rq->cmd_flags & REQ_FAILED)
941 dm_done(clone, tio->error, mapped);
945 * Complete the clone and the original request with the error status
946 * through softirq context.
948 static void dm_complete_request(struct request *clone, int error)
950 struct dm_rq_target_io *tio = clone->end_io_data;
951 struct request *rq = tio->orig;
953 if (unlikely(clone->cmd_flags & REQ_HARDBARRIER)) {
955 * Barrier clones share an original request. So can't use
956 * softirq_done with the original.
957 * Pass the clone to dm_done() directly in this special case.
958 * It is safe (even if clone->q->queue_lock is held here)
959 * because there is no I/O dispatching during the completion
962 dm_done(clone, error, true);
967 rq->completion_data = clone;
968 blk_complete_request(rq);
972 * Complete the not-mapped clone and the original request with the error status
973 * through softirq context.
974 * Target's rq_end_io() function isn't called.
975 * This may be used when the target's map_rq() function fails.
977 void dm_kill_unmapped_request(struct request *clone, int error)
979 struct dm_rq_target_io *tio = clone->end_io_data;
980 struct request *rq = tio->orig;
982 if (unlikely(clone->cmd_flags & REQ_HARDBARRIER)) {
984 * Barrier clones share an original request.
985 * Leave it to dm_end_request(), which handles this special
989 dm_end_request(clone, error);
993 rq->cmd_flags |= REQ_FAILED;
994 dm_complete_request(clone, error);
996 EXPORT_SYMBOL_GPL(dm_kill_unmapped_request);
999 * Called with the queue lock held
1001 static void end_clone_request(struct request *clone, int error)
1004 * For just cleaning up the information of the queue in which
1005 * the clone was dispatched.
1006 * The clone is *NOT* freed actually here because it is alloced from
1007 * dm own mempool and REQ_ALLOCED isn't set in clone->cmd_flags.
1009 __blk_put_request(clone->q, clone);
1012 * Actual request completion is done in a softirq context which doesn't
1013 * hold the queue lock. Otherwise, deadlock could occur because:
1014 * - another request may be submitted by the upper level driver
1015 * of the stacking during the completion
1016 * - the submission which requires queue lock may be done
1017 * against this queue
1019 dm_complete_request(clone, error);
1022 static sector_t max_io_len(struct mapped_device *md,
1023 sector_t sector, struct dm_target *ti)
1025 sector_t offset = sector - ti->begin;
1026 sector_t len = ti->len - offset;
1029 * Does the target need to split even further ?
1033 boundary = ((offset + ti->split_io) & ~(ti->split_io - 1))
1042 static void __map_bio(struct dm_target *ti, struct bio *clone,
1043 struct dm_target_io *tio)
1047 struct mapped_device *md;
1049 clone->bi_end_io = clone_endio;
1050 clone->bi_private = tio;
1053 * Map the clone. If r == 0 we don't need to do
1054 * anything, the target has assumed ownership of
1057 atomic_inc(&tio->io->io_count);
1058 sector = clone->bi_sector;
1059 r = ti->type->map(ti, clone, &tio->info);
1060 if (r == DM_MAPIO_REMAPPED) {
1061 /* the bio has been remapped so dispatch it */
1063 trace_block_remap(bdev_get_queue(clone->bi_bdev), clone,
1064 tio->io->bio->bi_bdev->bd_dev, sector);
1066 generic_make_request(clone);
1067 } else if (r < 0 || r == DM_MAPIO_REQUEUE) {
1068 /* error the io and bail out, or requeue it if needed */
1070 dec_pending(tio->io, r);
1072 * Store bio_set for cleanup.
1074 clone->bi_private = md->bs;
1078 DMWARN("unimplemented target map return value: %d", r);
1084 struct mapped_device *md;
1085 struct dm_table *map;
1089 sector_t sector_count;
1093 static void dm_bio_destructor(struct bio *bio)
1095 struct bio_set *bs = bio->bi_private;
1101 * Creates a little bio that is just does part of a bvec.
1103 static struct bio *split_bvec(struct bio *bio, sector_t sector,
1104 unsigned short idx, unsigned int offset,
1105 unsigned int len, struct bio_set *bs)
1108 struct bio_vec *bv = bio->bi_io_vec + idx;
1110 clone = bio_alloc_bioset(GFP_NOIO, 1, bs);
1111 clone->bi_destructor = dm_bio_destructor;
1112 *clone->bi_io_vec = *bv;
1114 clone->bi_sector = sector;
1115 clone->bi_bdev = bio->bi_bdev;
1116 clone->bi_rw = bio->bi_rw & ~REQ_HARDBARRIER;
1118 clone->bi_size = to_bytes(len);
1119 clone->bi_io_vec->bv_offset = offset;
1120 clone->bi_io_vec->bv_len = clone->bi_size;
1121 clone->bi_flags |= 1 << BIO_CLONED;
1123 if (bio_integrity(bio)) {
1124 bio_integrity_clone(clone, bio, GFP_NOIO, bs);
1125 bio_integrity_trim(clone,
1126 bio_sector_offset(bio, idx, offset), len);
1133 * Creates a bio that consists of range of complete bvecs.
1135 static struct bio *clone_bio(struct bio *bio, sector_t sector,
1136 unsigned short idx, unsigned short bv_count,
1137 unsigned int len, struct bio_set *bs)
1141 clone = bio_alloc_bioset(GFP_NOIO, bio->bi_max_vecs, bs);
1142 __bio_clone(clone, bio);
1143 clone->bi_rw &= ~REQ_HARDBARRIER;
1144 clone->bi_destructor = dm_bio_destructor;
1145 clone->bi_sector = sector;
1146 clone->bi_idx = idx;
1147 clone->bi_vcnt = idx + bv_count;
1148 clone->bi_size = to_bytes(len);
1149 clone->bi_flags &= ~(1 << BIO_SEG_VALID);
1151 if (bio_integrity(bio)) {
1152 bio_integrity_clone(clone, bio, GFP_NOIO, bs);
1154 if (idx != bio->bi_idx || clone->bi_size < bio->bi_size)
1155 bio_integrity_trim(clone,
1156 bio_sector_offset(bio, idx, 0), len);
1162 static struct dm_target_io *alloc_tio(struct clone_info *ci,
1163 struct dm_target *ti)
1165 struct dm_target_io *tio = mempool_alloc(ci->md->tio_pool, GFP_NOIO);
1169 memset(&tio->info, 0, sizeof(tio->info));
1174 static void __flush_target(struct clone_info *ci, struct dm_target *ti,
1177 struct dm_target_io *tio = alloc_tio(ci, ti);
1180 tio->info.flush_request = flush_nr;
1182 clone = bio_alloc_bioset(GFP_NOIO, 0, ci->md->bs);
1183 __bio_clone(clone, ci->bio);
1184 clone->bi_destructor = dm_bio_destructor;
1186 __map_bio(ti, clone, tio);
1189 static int __clone_and_map_empty_barrier(struct clone_info *ci)
1191 unsigned target_nr = 0, flush_nr;
1192 struct dm_target *ti;
1194 while ((ti = dm_table_get_target(ci->map, target_nr++)))
1195 for (flush_nr = 0; flush_nr < ti->num_flush_requests;
1197 __flush_target(ci, ti, flush_nr);
1199 ci->sector_count = 0;
1204 static int __clone_and_map(struct clone_info *ci)
1206 struct bio *clone, *bio = ci->bio;
1207 struct dm_target *ti;
1208 sector_t len = 0, max;
1209 struct dm_target_io *tio;
1211 if (unlikely(bio_empty_barrier(bio)))
1212 return __clone_and_map_empty_barrier(ci);
1214 ti = dm_table_find_target(ci->map, ci->sector);
1215 if (!dm_target_is_valid(ti))
1218 max = max_io_len(ci->md, ci->sector, ti);
1221 * Allocate a target io object.
1223 tio = alloc_tio(ci, 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 = clone_bio(bio, ci->sector, ci->idx,
1231 bio->bi_vcnt - ci->idx, ci->sector_count,
1233 __map_bio(ti, clone, tio);
1234 ci->sector_count = 0;
1236 } else if (to_sector(bio->bi_io_vec[ci->idx].bv_len) <= max) {
1238 * There are some bvecs that don't span targets.
1239 * Do as many of these as possible.
1242 sector_t remaining = max;
1245 for (i = ci->idx; remaining && (i < bio->bi_vcnt); i++) {
1246 bv_len = to_sector(bio->bi_io_vec[i].bv_len);
1248 if (bv_len > remaining)
1251 remaining -= bv_len;
1255 clone = clone_bio(bio, ci->sector, ci->idx, i - ci->idx, len,
1257 __map_bio(ti, clone, tio);
1260 ci->sector_count -= len;
1265 * Handle a bvec that must be split between two or more targets.
1267 struct bio_vec *bv = bio->bi_io_vec + ci->idx;
1268 sector_t remaining = to_sector(bv->bv_len);
1269 unsigned int offset = 0;
1273 ti = dm_table_find_target(ci->map, ci->sector);
1274 if (!dm_target_is_valid(ti))
1277 max = max_io_len(ci->md, ci->sector, ti);
1279 tio = alloc_tio(ci, ti);
1282 len = min(remaining, max);
1284 clone = split_bvec(bio, ci->sector, ci->idx,
1285 bv->bv_offset + offset, len,
1288 __map_bio(ti, clone, tio);
1291 ci->sector_count -= len;
1292 offset += to_bytes(len);
1293 } while (remaining -= len);
1302 * Split the bio into several clones and submit it to targets.
1304 static void __split_and_process_bio(struct mapped_device *md, struct bio *bio)
1306 struct clone_info ci;
1309 ci.map = dm_get_live_table(md);
1310 if (unlikely(!ci.map)) {
1311 if (!(bio->bi_rw & REQ_HARDBARRIER))
1314 if (!md->barrier_error)
1315 md->barrier_error = -EIO;
1321 ci.io = alloc_io(md);
1323 atomic_set(&ci.io->io_count, 1);
1326 spin_lock_init(&ci.io->endio_lock);
1327 ci.sector = bio->bi_sector;
1328 ci.sector_count = bio_sectors(bio);
1329 if (unlikely(bio_empty_barrier(bio)))
1330 ci.sector_count = 1;
1331 ci.idx = bio->bi_idx;
1333 start_io_acct(ci.io);
1334 while (ci.sector_count && !error)
1335 error = __clone_and_map(&ci);
1337 /* drop the extra reference count */
1338 dec_pending(ci.io, error);
1339 dm_table_put(ci.map);
1341 /*-----------------------------------------------------------------
1343 *---------------------------------------------------------------*/
1345 static int dm_merge_bvec(struct request_queue *q,
1346 struct bvec_merge_data *bvm,
1347 struct bio_vec *biovec)
1349 struct mapped_device *md = q->queuedata;
1350 struct dm_table *map = dm_get_live_table(md);
1351 struct dm_target *ti;
1352 sector_t max_sectors;
1358 ti = dm_table_find_target(map, bvm->bi_sector);
1359 if (!dm_target_is_valid(ti))
1363 * Find maximum amount of I/O that won't need splitting
1365 max_sectors = min(max_io_len(md, bvm->bi_sector, ti),
1366 (sector_t) BIO_MAX_SECTORS);
1367 max_size = (max_sectors << SECTOR_SHIFT) - bvm->bi_size;
1372 * merge_bvec_fn() returns number of bytes
1373 * it can accept at this offset
1374 * max is precomputed maximal io size
1376 if (max_size && ti->type->merge)
1377 max_size = ti->type->merge(ti, bvm, biovec, max_size);
1379 * If the target doesn't support merge method and some of the devices
1380 * provided their merge_bvec method (we know this by looking at
1381 * queue_max_hw_sectors), then we can't allow bios with multiple vector
1382 * entries. So always set max_size to 0, and the code below allows
1385 else if (queue_max_hw_sectors(q) <= PAGE_SIZE >> 9)
1394 * Always allow an entire first page
1396 if (max_size <= biovec->bv_len && !(bvm->bi_size >> SECTOR_SHIFT))
1397 max_size = biovec->bv_len;
1403 * The request function that just remaps the bio built up by
1406 static int _dm_request(struct request_queue *q, struct bio *bio)
1408 int rw = bio_data_dir(bio);
1409 struct mapped_device *md = q->queuedata;
1412 down_read(&md->io_lock);
1414 cpu = part_stat_lock();
1415 part_stat_inc(cpu, &dm_disk(md)->part0, ios[rw]);
1416 part_stat_add(cpu, &dm_disk(md)->part0, sectors[rw], bio_sectors(bio));
1420 * If we're suspended or the thread is processing barriers
1421 * we have to queue this io for later.
1423 if (unlikely(test_bit(DMF_QUEUE_IO_TO_THREAD, &md->flags)) ||
1424 unlikely(bio->bi_rw & REQ_HARDBARRIER)) {
1425 up_read(&md->io_lock);
1427 if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) &&
1428 bio_rw(bio) == READA) {
1438 __split_and_process_bio(md, bio);
1439 up_read(&md->io_lock);
1443 static int dm_make_request(struct request_queue *q, struct bio *bio)
1445 struct mapped_device *md = q->queuedata;
1447 return md->saved_make_request_fn(q, bio); /* call __make_request() */
1450 static int dm_request_based(struct mapped_device *md)
1452 return blk_queue_stackable(md->queue);
1455 static int dm_request(struct request_queue *q, struct bio *bio)
1457 struct mapped_device *md = q->queuedata;
1459 if (dm_request_based(md))
1460 return dm_make_request(q, bio);
1462 return _dm_request(q, bio);
1465 static bool dm_rq_is_flush_request(struct request *rq)
1467 if (rq->cmd_flags & REQ_FLUSH)
1473 void dm_dispatch_request(struct request *rq)
1477 if (blk_queue_io_stat(rq->q))
1478 rq->cmd_flags |= REQ_IO_STAT;
1480 rq->start_time = jiffies;
1481 r = blk_insert_cloned_request(rq->q, rq);
1483 dm_complete_request(rq, r);
1485 EXPORT_SYMBOL_GPL(dm_dispatch_request);
1487 static void dm_rq_bio_destructor(struct bio *bio)
1489 struct dm_rq_clone_bio_info *info = bio->bi_private;
1490 struct mapped_device *md = info->tio->md;
1492 free_bio_info(info);
1493 bio_free(bio, md->bs);
1496 static int dm_rq_bio_constructor(struct bio *bio, struct bio *bio_orig,
1499 struct dm_rq_target_io *tio = data;
1500 struct mapped_device *md = tio->md;
1501 struct dm_rq_clone_bio_info *info = alloc_bio_info(md);
1506 info->orig = bio_orig;
1508 bio->bi_end_io = end_clone_bio;
1509 bio->bi_private = info;
1510 bio->bi_destructor = dm_rq_bio_destructor;
1515 static int setup_clone(struct request *clone, struct request *rq,
1516 struct dm_rq_target_io *tio)
1520 if (dm_rq_is_flush_request(rq)) {
1521 blk_rq_init(NULL, clone);
1522 clone->cmd_type = REQ_TYPE_FS;
1523 clone->cmd_flags |= (REQ_HARDBARRIER | WRITE);
1525 r = blk_rq_prep_clone(clone, rq, tio->md->bs, GFP_ATOMIC,
1526 dm_rq_bio_constructor, tio);
1530 clone->cmd = rq->cmd;
1531 clone->cmd_len = rq->cmd_len;
1532 clone->sense = rq->sense;
1533 clone->buffer = rq->buffer;
1536 clone->end_io = end_clone_request;
1537 clone->end_io_data = tio;
1542 static struct request *clone_rq(struct request *rq, struct mapped_device *md,
1545 struct request *clone;
1546 struct dm_rq_target_io *tio;
1548 tio = alloc_rq_tio(md, gfp_mask);
1556 memset(&tio->info, 0, sizeof(tio->info));
1558 clone = &tio->clone;
1559 if (setup_clone(clone, rq, tio)) {
1569 * Called with the queue lock held.
1571 static int dm_prep_fn(struct request_queue *q, struct request *rq)
1573 struct mapped_device *md = q->queuedata;
1574 struct request *clone;
1576 if (unlikely(dm_rq_is_flush_request(rq)))
1579 if (unlikely(rq->special)) {
1580 DMWARN("Already has something in rq->special.");
1581 return BLKPREP_KILL;
1584 clone = clone_rq(rq, md, GFP_ATOMIC);
1586 return BLKPREP_DEFER;
1588 rq->special = clone;
1589 rq->cmd_flags |= REQ_DONTPREP;
1596 * 0 : the request has been processed (not requeued)
1597 * !0 : the request has been requeued
1599 static int map_request(struct dm_target *ti, struct request *clone,
1600 struct mapped_device *md)
1602 int r, requeued = 0;
1603 struct dm_rq_target_io *tio = clone->end_io_data;
1606 * Hold the md reference here for the in-flight I/O.
1607 * We can't rely on the reference count by device opener,
1608 * because the device may be closed during the request completion
1609 * when all bios are completed.
1610 * See the comment in rq_completed() too.
1615 r = ti->type->map_rq(ti, clone, &tio->info);
1617 case DM_MAPIO_SUBMITTED:
1618 /* The target has taken the I/O to submit by itself later */
1620 case DM_MAPIO_REMAPPED:
1621 /* The target has remapped the I/O so dispatch it */
1622 trace_block_rq_remap(clone->q, clone, disk_devt(dm_disk(md)),
1623 blk_rq_pos(tio->orig));
1624 dm_dispatch_request(clone);
1626 case DM_MAPIO_REQUEUE:
1627 /* The target wants to requeue the I/O */
1628 dm_requeue_unmapped_request(clone);
1633 DMWARN("unimplemented target map return value: %d", r);
1637 /* The target wants to complete the I/O */
1638 dm_kill_unmapped_request(clone, r);
1646 * q->request_fn for request-based dm.
1647 * Called with the queue lock held.
1649 static void dm_request_fn(struct request_queue *q)
1651 struct mapped_device *md = q->queuedata;
1652 struct dm_table *map = dm_get_live_table(md);
1653 struct dm_target *ti;
1654 struct request *rq, *clone;
1657 * For suspend, check blk_queue_stopped() and increment
1658 * ->pending within a single queue_lock not to increment the
1659 * number of in-flight I/Os after the queue is stopped in
1662 while (!blk_queue_plugged(q) && !blk_queue_stopped(q)) {
1663 rq = blk_peek_request(q);
1667 if (unlikely(dm_rq_is_flush_request(rq))) {
1668 BUG_ON(md->flush_request);
1669 md->flush_request = rq;
1670 blk_start_request(rq);
1671 queue_work(md->wq, &md->barrier_work);
1675 ti = dm_table_find_target(map, blk_rq_pos(rq));
1676 if (ti->type->busy && ti->type->busy(ti))
1679 blk_start_request(rq);
1680 clone = rq->special;
1681 atomic_inc(&md->pending[rq_data_dir(clone)]);
1683 spin_unlock(q->queue_lock);
1684 if (map_request(ti, clone, md))
1687 spin_lock_irq(q->queue_lock);
1693 spin_lock_irq(q->queue_lock);
1696 if (!elv_queue_empty(q))
1697 /* Some requests still remain, retry later */
1706 int dm_underlying_device_busy(struct request_queue *q)
1708 return blk_lld_busy(q);
1710 EXPORT_SYMBOL_GPL(dm_underlying_device_busy);
1712 static int dm_lld_busy(struct request_queue *q)
1715 struct mapped_device *md = q->queuedata;
1716 struct dm_table *map = dm_get_live_table(md);
1718 if (!map || test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))
1721 r = dm_table_any_busy_target(map);
1728 static void dm_unplug_all(struct request_queue *q)
1730 struct mapped_device *md = q->queuedata;
1731 struct dm_table *map = dm_get_live_table(md);
1734 if (dm_request_based(md))
1735 generic_unplug_device(q);
1737 dm_table_unplug_all(map);
1742 static int dm_any_congested(void *congested_data, int bdi_bits)
1745 struct mapped_device *md = congested_data;
1746 struct dm_table *map;
1748 if (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
1749 map = dm_get_live_table(md);
1752 * Request-based dm cares about only own queue for
1753 * the query about congestion status of request_queue
1755 if (dm_request_based(md))
1756 r = md->queue->backing_dev_info.state &
1759 r = dm_table_any_congested(map, bdi_bits);
1768 /*-----------------------------------------------------------------
1769 * An IDR is used to keep track of allocated minor numbers.
1770 *---------------------------------------------------------------*/
1771 static DEFINE_IDR(_minor_idr);
1773 static void free_minor(int minor)
1775 spin_lock(&_minor_lock);
1776 idr_remove(&_minor_idr, minor);
1777 spin_unlock(&_minor_lock);
1781 * See if the device with a specific minor # is free.
1783 static int specific_minor(int minor)
1787 if (minor >= (1 << MINORBITS))
1790 r = idr_pre_get(&_minor_idr, GFP_KERNEL);
1794 spin_lock(&_minor_lock);
1796 if (idr_find(&_minor_idr, minor)) {
1801 r = idr_get_new_above(&_minor_idr, MINOR_ALLOCED, minor, &m);
1806 idr_remove(&_minor_idr, m);
1812 spin_unlock(&_minor_lock);
1816 static int next_free_minor(int *minor)
1820 r = idr_pre_get(&_minor_idr, GFP_KERNEL);
1824 spin_lock(&_minor_lock);
1826 r = idr_get_new(&_minor_idr, MINOR_ALLOCED, &m);
1830 if (m >= (1 << MINORBITS)) {
1831 idr_remove(&_minor_idr, m);
1839 spin_unlock(&_minor_lock);
1843 static const struct block_device_operations dm_blk_dops;
1845 static void dm_wq_work(struct work_struct *work);
1846 static void dm_rq_barrier_work(struct work_struct *work);
1849 * Allocate and initialise a blank device with a given minor.
1851 static struct mapped_device *alloc_dev(int minor)
1854 struct mapped_device *md = kzalloc(sizeof(*md), GFP_KERNEL);
1858 DMWARN("unable to allocate device, out of memory.");
1862 if (!try_module_get(THIS_MODULE))
1863 goto bad_module_get;
1865 /* get a minor number for the dev */
1866 if (minor == DM_ANY_MINOR)
1867 r = next_free_minor(&minor);
1869 r = specific_minor(minor);
1873 init_rwsem(&md->io_lock);
1874 mutex_init(&md->suspend_lock);
1875 spin_lock_init(&md->deferred_lock);
1876 spin_lock_init(&md->barrier_error_lock);
1877 rwlock_init(&md->map_lock);
1878 atomic_set(&md->holders, 1);
1879 atomic_set(&md->open_count, 0);
1880 atomic_set(&md->event_nr, 0);
1881 atomic_set(&md->uevent_seq, 0);
1882 INIT_LIST_HEAD(&md->uevent_list);
1883 spin_lock_init(&md->uevent_lock);
1885 md->queue = blk_init_queue(dm_request_fn, NULL);
1890 * Request-based dm devices cannot be stacked on top of bio-based dm
1891 * devices. The type of this dm device has not been decided yet,
1892 * although we initialized the queue using blk_init_queue().
1893 * The type is decided at the first table loading time.
1894 * To prevent problematic device stacking, clear the queue flag
1895 * for request stacking support until then.
1897 * This queue is new, so no concurrency on the queue_flags.
1899 queue_flag_clear_unlocked(QUEUE_FLAG_STACKABLE, md->queue);
1900 md->saved_make_request_fn = md->queue->make_request_fn;
1901 md->queue->queuedata = md;
1902 md->queue->backing_dev_info.congested_fn = dm_any_congested;
1903 md->queue->backing_dev_info.congested_data = md;
1904 blk_queue_make_request(md->queue, dm_request);
1905 blk_queue_bounce_limit(md->queue, BLK_BOUNCE_ANY);
1906 md->queue->unplug_fn = dm_unplug_all;
1907 blk_queue_merge_bvec(md->queue, dm_merge_bvec);
1908 blk_queue_softirq_done(md->queue, dm_softirq_done);
1909 blk_queue_prep_rq(md->queue, dm_prep_fn);
1910 blk_queue_lld_busy(md->queue, dm_lld_busy);
1911 blk_queue_ordered(md->queue, QUEUE_ORDERED_DRAIN_FLUSH);
1913 md->disk = alloc_disk(1);
1917 atomic_set(&md->pending[0], 0);
1918 atomic_set(&md->pending[1], 0);
1919 init_waitqueue_head(&md->wait);
1920 INIT_WORK(&md->work, dm_wq_work);
1921 INIT_WORK(&md->barrier_work, dm_rq_barrier_work);
1922 init_waitqueue_head(&md->eventq);
1924 md->disk->major = _major;
1925 md->disk->first_minor = minor;
1926 md->disk->fops = &dm_blk_dops;
1927 md->disk->queue = md->queue;
1928 md->disk->private_data = md;
1929 sprintf(md->disk->disk_name, "dm-%d", minor);
1931 format_dev_t(md->name, MKDEV(_major, minor));
1933 md->wq = create_singlethread_workqueue("kdmflush");
1937 md->bdev = bdget_disk(md->disk, 0);
1941 /* Populate the mapping, nobody knows we exist yet */
1942 spin_lock(&_minor_lock);
1943 old_md = idr_replace(&_minor_idr, md, minor);
1944 spin_unlock(&_minor_lock);
1946 BUG_ON(old_md != MINOR_ALLOCED);
1951 destroy_workqueue(md->wq);
1953 del_gendisk(md->disk);
1956 blk_cleanup_queue(md->queue);
1960 module_put(THIS_MODULE);
1966 static void unlock_fs(struct mapped_device *md);
1968 static void free_dev(struct mapped_device *md)
1970 int minor = MINOR(disk_devt(md->disk));
1974 destroy_workqueue(md->wq);
1976 mempool_destroy(md->tio_pool);
1978 mempool_destroy(md->io_pool);
1980 bioset_free(md->bs);
1981 blk_integrity_unregister(md->disk);
1982 del_gendisk(md->disk);
1985 spin_lock(&_minor_lock);
1986 md->disk->private_data = NULL;
1987 spin_unlock(&_minor_lock);
1990 blk_cleanup_queue(md->queue);
1991 module_put(THIS_MODULE);
1995 static void __bind_mempools(struct mapped_device *md, struct dm_table *t)
1997 struct dm_md_mempools *p;
1999 if (md->io_pool && md->tio_pool && md->bs)
2000 /* the md already has necessary mempools */
2003 p = dm_table_get_md_mempools(t);
2004 BUG_ON(!p || md->io_pool || md->tio_pool || md->bs);
2006 md->io_pool = p->io_pool;
2008 md->tio_pool = p->tio_pool;
2014 /* mempool bind completed, now no need any mempools in the table */
2015 dm_table_free_md_mempools(t);
2019 * Bind a table to the device.
2021 static void event_callback(void *context)
2023 unsigned long flags;
2025 struct mapped_device *md = (struct mapped_device *) context;
2027 spin_lock_irqsave(&md->uevent_lock, flags);
2028 list_splice_init(&md->uevent_list, &uevents);
2029 spin_unlock_irqrestore(&md->uevent_lock, flags);
2031 dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj);
2033 atomic_inc(&md->event_nr);
2034 wake_up(&md->eventq);
2037 static void __set_size(struct mapped_device *md, sector_t size)
2039 set_capacity(md->disk, size);
2041 mutex_lock(&md->bdev->bd_inode->i_mutex);
2042 i_size_write(md->bdev->bd_inode, (loff_t)size << SECTOR_SHIFT);
2043 mutex_unlock(&md->bdev->bd_inode->i_mutex);
2047 * Returns old map, which caller must destroy.
2049 static struct dm_table *__bind(struct mapped_device *md, struct dm_table *t,
2050 struct queue_limits *limits)
2052 struct dm_table *old_map;
2053 struct request_queue *q = md->queue;
2055 unsigned long flags;
2057 size = dm_table_get_size(t);
2060 * Wipe any geometry if the size of the table changed.
2062 if (size != get_capacity(md->disk))
2063 memset(&md->geometry, 0, sizeof(md->geometry));
2065 __set_size(md, size);
2067 dm_table_event_callback(t, event_callback, md);
2070 * The queue hasn't been stopped yet, if the old table type wasn't
2071 * for request-based during suspension. So stop it to prevent
2072 * I/O mapping before resume.
2073 * This must be done before setting the queue restrictions,
2074 * because request-based dm may be run just after the setting.
2076 if (dm_table_request_based(t) && !blk_queue_stopped(q))
2079 __bind_mempools(md, t);
2081 write_lock_irqsave(&md->map_lock, flags);
2084 dm_table_set_restrictions(t, q, limits);
2085 write_unlock_irqrestore(&md->map_lock, flags);
2091 * Returns unbound table for the caller to free.
2093 static struct dm_table *__unbind(struct mapped_device *md)
2095 struct dm_table *map = md->map;
2096 unsigned long flags;
2101 dm_table_event_callback(map, NULL, NULL);
2102 write_lock_irqsave(&md->map_lock, flags);
2104 write_unlock_irqrestore(&md->map_lock, flags);
2110 * Constructor for a new device.
2112 int dm_create(int minor, struct mapped_device **result)
2114 struct mapped_device *md;
2116 md = alloc_dev(minor);
2126 static struct mapped_device *dm_find_md(dev_t dev)
2128 struct mapped_device *md;
2129 unsigned minor = MINOR(dev);
2131 if (MAJOR(dev) != _major || minor >= (1 << MINORBITS))
2134 spin_lock(&_minor_lock);
2136 md = idr_find(&_minor_idr, minor);
2137 if (md && (md == MINOR_ALLOCED ||
2138 (MINOR(disk_devt(dm_disk(md))) != minor) ||
2139 test_bit(DMF_FREEING, &md->flags))) {
2145 spin_unlock(&_minor_lock);
2150 struct mapped_device *dm_get_md(dev_t dev)
2152 struct mapped_device *md = dm_find_md(dev);
2160 void *dm_get_mdptr(struct mapped_device *md)
2162 return md->interface_ptr;
2165 void dm_set_mdptr(struct mapped_device *md, void *ptr)
2167 md->interface_ptr = ptr;
2170 void dm_get(struct mapped_device *md)
2172 atomic_inc(&md->holders);
2175 const char *dm_device_name(struct mapped_device *md)
2179 EXPORT_SYMBOL_GPL(dm_device_name);
2181 void dm_put(struct mapped_device *md)
2183 struct dm_table *map;
2185 BUG_ON(test_bit(DMF_FREEING, &md->flags));
2187 if (atomic_dec_and_lock(&md->holders, &_minor_lock)) {
2188 map = dm_get_live_table(md);
2189 idr_replace(&_minor_idr, MINOR_ALLOCED,
2190 MINOR(disk_devt(dm_disk(md))));
2191 set_bit(DMF_FREEING, &md->flags);
2192 spin_unlock(&_minor_lock);
2193 if (!dm_suspended_md(md)) {
2194 dm_table_presuspend_targets(map);
2195 dm_table_postsuspend_targets(map);
2199 dm_table_destroy(__unbind(md));
2203 EXPORT_SYMBOL_GPL(dm_put);
2205 static int dm_wait_for_completion(struct mapped_device *md, int interruptible)
2208 DECLARE_WAITQUEUE(wait, current);
2210 dm_unplug_all(md->queue);
2212 add_wait_queue(&md->wait, &wait);
2215 set_current_state(interruptible);
2218 if (!md_in_flight(md))
2221 if (interruptible == TASK_INTERRUPTIBLE &&
2222 signal_pending(current)) {
2229 set_current_state(TASK_RUNNING);
2231 remove_wait_queue(&md->wait, &wait);
2236 static void dm_flush(struct mapped_device *md)
2238 dm_wait_for_completion(md, TASK_UNINTERRUPTIBLE);
2240 bio_init(&md->barrier_bio);
2241 md->barrier_bio.bi_bdev = md->bdev;
2242 md->barrier_bio.bi_rw = WRITE_BARRIER;
2243 __split_and_process_bio(md, &md->barrier_bio);
2245 dm_wait_for_completion(md, TASK_UNINTERRUPTIBLE);
2248 static void process_barrier(struct mapped_device *md, struct bio *bio)
2250 md->barrier_error = 0;
2254 if (!bio_empty_barrier(bio)) {
2255 __split_and_process_bio(md, bio);
2259 if (md->barrier_error != DM_ENDIO_REQUEUE)
2260 bio_endio(bio, md->barrier_error);
2262 spin_lock_irq(&md->deferred_lock);
2263 bio_list_add_head(&md->deferred, bio);
2264 spin_unlock_irq(&md->deferred_lock);
2269 * Process the deferred bios
2271 static void dm_wq_work(struct work_struct *work)
2273 struct mapped_device *md = container_of(work, struct mapped_device,
2277 down_write(&md->io_lock);
2279 while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
2280 spin_lock_irq(&md->deferred_lock);
2281 c = bio_list_pop(&md->deferred);
2282 spin_unlock_irq(&md->deferred_lock);
2285 clear_bit(DMF_QUEUE_IO_TO_THREAD, &md->flags);
2289 up_write(&md->io_lock);
2291 if (dm_request_based(md))
2292 generic_make_request(c);
2294 if (c->bi_rw & REQ_HARDBARRIER)
2295 process_barrier(md, c);
2297 __split_and_process_bio(md, c);
2300 down_write(&md->io_lock);
2303 up_write(&md->io_lock);
2306 static void dm_queue_flush(struct mapped_device *md)
2308 clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2309 smp_mb__after_clear_bit();
2310 queue_work(md->wq, &md->work);
2313 static void dm_rq_set_flush_nr(struct request *clone, unsigned flush_nr)
2315 struct dm_rq_target_io *tio = clone->end_io_data;
2317 tio->info.flush_request = flush_nr;
2320 /* Issue barrier requests to targets and wait for their completion. */
2321 static int dm_rq_barrier(struct mapped_device *md)
2324 struct dm_table *map = dm_get_live_table(md);
2325 unsigned num_targets = dm_table_get_num_targets(map);
2326 struct dm_target *ti;
2327 struct request *clone;
2329 md->barrier_error = 0;
2331 for (i = 0; i < num_targets; i++) {
2332 ti = dm_table_get_target(map, i);
2333 for (j = 0; j < ti->num_flush_requests; j++) {
2334 clone = clone_rq(md->flush_request, md, GFP_NOIO);
2335 dm_rq_set_flush_nr(clone, j);
2336 atomic_inc(&md->pending[rq_data_dir(clone)]);
2337 map_request(ti, clone, md);
2341 dm_wait_for_completion(md, TASK_UNINTERRUPTIBLE);
2344 return md->barrier_error;
2347 static void dm_rq_barrier_work(struct work_struct *work)
2350 struct mapped_device *md = container_of(work, struct mapped_device,
2352 struct request_queue *q = md->queue;
2354 unsigned long flags;
2357 * Hold the md reference here and leave it at the last part so that
2358 * the md can't be deleted by device opener when the barrier request
2363 error = dm_rq_barrier(md);
2365 rq = md->flush_request;
2366 md->flush_request = NULL;
2368 if (error == DM_ENDIO_REQUEUE) {
2369 spin_lock_irqsave(q->queue_lock, flags);
2370 blk_requeue_request(q, rq);
2371 spin_unlock_irqrestore(q->queue_lock, flags);
2373 blk_end_request_all(rq, error);
2381 * Swap in a new table, returning the old one for the caller to destroy.
2383 struct dm_table *dm_swap_table(struct mapped_device *md, struct dm_table *table)
2385 struct dm_table *map = ERR_PTR(-EINVAL);
2386 struct queue_limits limits;
2389 mutex_lock(&md->suspend_lock);
2391 /* device must be suspended */
2392 if (!dm_suspended_md(md))
2395 r = dm_calculate_queue_limits(table, &limits);
2401 /* cannot change the device type, once a table is bound */
2403 (dm_table_get_type(md->map) != dm_table_get_type(table))) {
2404 DMWARN("can't change the device type after a table is bound");
2408 map = __bind(md, table, &limits);
2411 mutex_unlock(&md->suspend_lock);
2416 * Functions to lock and unlock any filesystem running on the
2419 static int lock_fs(struct mapped_device *md)
2423 WARN_ON(md->frozen_sb);
2425 md->frozen_sb = freeze_bdev(md->bdev);
2426 if (IS_ERR(md->frozen_sb)) {
2427 r = PTR_ERR(md->frozen_sb);
2428 md->frozen_sb = NULL;
2432 set_bit(DMF_FROZEN, &md->flags);
2437 static void unlock_fs(struct mapped_device *md)
2439 if (!test_bit(DMF_FROZEN, &md->flags))
2442 thaw_bdev(md->bdev, md->frozen_sb);
2443 md->frozen_sb = NULL;
2444 clear_bit(DMF_FROZEN, &md->flags);
2448 * We need to be able to change a mapping table under a mounted
2449 * filesystem. For example we might want to move some data in
2450 * the background. Before the table can be swapped with
2451 * dm_bind_table, dm_suspend must be called to flush any in
2452 * flight bios and ensure that any further io gets deferred.
2455 * Suspend mechanism in request-based dm.
2457 * 1. Flush all I/Os by lock_fs() if needed.
2458 * 2. Stop dispatching any I/O by stopping the request_queue.
2459 * 3. Wait for all in-flight I/Os to be completed or requeued.
2461 * To abort suspend, start the request_queue.
2463 int dm_suspend(struct mapped_device *md, unsigned suspend_flags)
2465 struct dm_table *map = NULL;
2467 int do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG ? 1 : 0;
2468 int noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG ? 1 : 0;
2470 mutex_lock(&md->suspend_lock);
2472 if (dm_suspended_md(md)) {
2477 map = dm_get_live_table(md);
2480 * DMF_NOFLUSH_SUSPENDING must be set before presuspend.
2481 * This flag is cleared before dm_suspend returns.
2484 set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2486 /* This does not get reverted if there's an error later. */
2487 dm_table_presuspend_targets(map);
2490 * Flush I/O to the device.
2491 * Any I/O submitted after lock_fs() may not be flushed.
2492 * noflush takes precedence over do_lockfs.
2493 * (lock_fs() flushes I/Os and waits for them to complete.)
2495 if (!noflush && do_lockfs) {
2502 * Here we must make sure that no processes are submitting requests
2503 * to target drivers i.e. no one may be executing
2504 * __split_and_process_bio. This is called from dm_request and
2507 * To get all processes out of __split_and_process_bio in dm_request,
2508 * we take the write lock. To prevent any process from reentering
2509 * __split_and_process_bio from dm_request, we set
2510 * DMF_QUEUE_IO_TO_THREAD.
2512 * To quiesce the thread (dm_wq_work), we set DMF_BLOCK_IO_FOR_SUSPEND
2513 * and call flush_workqueue(md->wq). flush_workqueue will wait until
2514 * dm_wq_work exits and DMF_BLOCK_IO_FOR_SUSPEND will prevent any
2515 * further calls to __split_and_process_bio from dm_wq_work.
2517 down_write(&md->io_lock);
2518 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2519 set_bit(DMF_QUEUE_IO_TO_THREAD, &md->flags);
2520 up_write(&md->io_lock);
2523 * Request-based dm uses md->wq for barrier (dm_rq_barrier_work) which
2524 * can be kicked until md->queue is stopped. So stop md->queue before
2527 if (dm_request_based(md))
2528 stop_queue(md->queue);
2530 flush_workqueue(md->wq);
2533 * At this point no more requests are entering target request routines.
2534 * We call dm_wait_for_completion to wait for all existing requests
2537 r = dm_wait_for_completion(md, TASK_INTERRUPTIBLE);
2539 down_write(&md->io_lock);
2541 clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2542 up_write(&md->io_lock);
2544 /* were we interrupted ? */
2548 if (dm_request_based(md))
2549 start_queue(md->queue);
2552 goto out; /* pushback list is already flushed, so skip flush */
2556 * If dm_wait_for_completion returned 0, the device is completely
2557 * quiescent now. There is no request-processing activity. All new
2558 * requests are being added to md->deferred list.
2561 set_bit(DMF_SUSPENDED, &md->flags);
2563 dm_table_postsuspend_targets(map);
2569 mutex_unlock(&md->suspend_lock);
2573 int dm_resume(struct mapped_device *md)
2576 struct dm_table *map = NULL;
2578 mutex_lock(&md->suspend_lock);
2579 if (!dm_suspended_md(md))
2582 map = dm_get_live_table(md);
2583 if (!map || !dm_table_get_size(map))
2586 r = dm_table_resume_targets(map);
2593 * Flushing deferred I/Os must be done after targets are resumed
2594 * so that mapping of targets can work correctly.
2595 * Request-based dm is queueing the deferred I/Os in its request_queue.
2597 if (dm_request_based(md))
2598 start_queue(md->queue);
2602 clear_bit(DMF_SUSPENDED, &md->flags);
2604 dm_table_unplug_all(map);
2608 mutex_unlock(&md->suspend_lock);
2613 /*-----------------------------------------------------------------
2614 * Event notification.
2615 *---------------------------------------------------------------*/
2616 int dm_kobject_uevent(struct mapped_device *md, enum kobject_action action,
2619 char udev_cookie[DM_COOKIE_LENGTH];
2620 char *envp[] = { udev_cookie, NULL };
2623 return kobject_uevent(&disk_to_dev(md->disk)->kobj, action);
2625 snprintf(udev_cookie, DM_COOKIE_LENGTH, "%s=%u",
2626 DM_COOKIE_ENV_VAR_NAME, cookie);
2627 return kobject_uevent_env(&disk_to_dev(md->disk)->kobj,
2632 uint32_t dm_next_uevent_seq(struct mapped_device *md)
2634 return atomic_add_return(1, &md->uevent_seq);
2637 uint32_t dm_get_event_nr(struct mapped_device *md)
2639 return atomic_read(&md->event_nr);
2642 int dm_wait_event(struct mapped_device *md, int event_nr)
2644 return wait_event_interruptible(md->eventq,
2645 (event_nr != atomic_read(&md->event_nr)));
2648 void dm_uevent_add(struct mapped_device *md, struct list_head *elist)
2650 unsigned long flags;
2652 spin_lock_irqsave(&md->uevent_lock, flags);
2653 list_add(elist, &md->uevent_list);
2654 spin_unlock_irqrestore(&md->uevent_lock, flags);
2658 * The gendisk is only valid as long as you have a reference
2661 struct gendisk *dm_disk(struct mapped_device *md)
2666 struct kobject *dm_kobject(struct mapped_device *md)
2672 * struct mapped_device should not be exported outside of dm.c
2673 * so use this check to verify that kobj is part of md structure
2675 struct mapped_device *dm_get_from_kobject(struct kobject *kobj)
2677 struct mapped_device *md;
2679 md = container_of(kobj, struct mapped_device, kobj);
2680 if (&md->kobj != kobj)
2683 if (test_bit(DMF_FREEING, &md->flags) ||
2691 int dm_suspended_md(struct mapped_device *md)
2693 return test_bit(DMF_SUSPENDED, &md->flags);
2696 int dm_suspended(struct dm_target *ti)
2698 return dm_suspended_md(dm_table_get_md(ti->table));
2700 EXPORT_SYMBOL_GPL(dm_suspended);
2702 int dm_noflush_suspending(struct dm_target *ti)
2704 return __noflush_suspending(dm_table_get_md(ti->table));
2706 EXPORT_SYMBOL_GPL(dm_noflush_suspending);
2708 struct dm_md_mempools *dm_alloc_md_mempools(unsigned type)
2710 struct dm_md_mempools *pools = kmalloc(sizeof(*pools), GFP_KERNEL);
2715 pools->io_pool = (type == DM_TYPE_BIO_BASED) ?
2716 mempool_create_slab_pool(MIN_IOS, _io_cache) :
2717 mempool_create_slab_pool(MIN_IOS, _rq_bio_info_cache);
2718 if (!pools->io_pool)
2719 goto free_pools_and_out;
2721 pools->tio_pool = (type == DM_TYPE_BIO_BASED) ?
2722 mempool_create_slab_pool(MIN_IOS, _tio_cache) :
2723 mempool_create_slab_pool(MIN_IOS, _rq_tio_cache);
2724 if (!pools->tio_pool)
2725 goto free_io_pool_and_out;
2727 pools->bs = (type == DM_TYPE_BIO_BASED) ?
2728 bioset_create(16, 0) : bioset_create(MIN_IOS, 0);
2730 goto free_tio_pool_and_out;
2734 free_tio_pool_and_out:
2735 mempool_destroy(pools->tio_pool);
2737 free_io_pool_and_out:
2738 mempool_destroy(pools->io_pool);
2746 void dm_free_md_mempools(struct dm_md_mempools *pools)
2752 mempool_destroy(pools->io_pool);
2754 if (pools->tio_pool)
2755 mempool_destroy(pools->tio_pool);
2758 bioset_free(pools->bs);
2763 static const struct block_device_operations dm_blk_dops = {
2764 .open = dm_blk_open,
2765 .release = dm_blk_close,
2766 .ioctl = dm_blk_ioctl,
2767 .getgeo = dm_blk_getgeo,
2768 .owner = THIS_MODULE
2771 EXPORT_SYMBOL(dm_get_mapinfo);
2776 module_init(dm_init);
2777 module_exit(dm_exit);
2779 module_param(major, uint, 0);
2780 MODULE_PARM_DESC(major, "The major number of the device mapper");
2781 MODULE_DESCRIPTION(DM_NAME " driver");
2782 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
2783 MODULE_LICENSE("GPL");