2 * Copyright (C) 2001, 2002 Sistina Software (UK) Limited.
3 * Copyright (C) 2004-2008 Red Hat, Inc. All rights reserved.
5 * This file is released under the GPL.
11 #include <linux/init.h>
12 #include <linux/module.h>
13 #include <linux/mutex.h>
14 #include <linux/moduleparam.h>
15 #include <linux/blkpg.h>
16 #include <linux/bio.h>
17 #include <linux/mempool.h>
18 #include <linux/slab.h>
19 #include <linux/idr.h>
20 #include <linux/hdreg.h>
21 #include <linux/delay.h>
23 #include <trace/events/block.h>
25 #define DM_MSG_PREFIX "core"
29 * ratelimit state to be used in DMXXX_LIMIT().
31 DEFINE_RATELIMIT_STATE(dm_ratelimit_state,
32 DEFAULT_RATELIMIT_INTERVAL,
33 DEFAULT_RATELIMIT_BURST);
34 EXPORT_SYMBOL(dm_ratelimit_state);
38 * Cookies are numeric values sent with CHANGE and REMOVE
39 * uevents while resuming, removing or renaming the device.
41 #define DM_COOKIE_ENV_VAR_NAME "DM_COOKIE"
42 #define DM_COOKIE_LENGTH 24
44 static const char *_name = DM_NAME;
46 static unsigned int major = 0;
47 static unsigned int _major = 0;
49 static DEFINE_IDR(_minor_idr);
51 static DEFINE_SPINLOCK(_minor_lock);
54 * One of these is allocated per bio.
57 struct mapped_device *md;
61 unsigned long start_time;
62 spinlock_t endio_lock;
67 * One of these is allocated per target within a bio. Hopefully
68 * this will be simplified out one day.
77 * For request-based dm.
78 * One of these is allocated per request.
80 struct dm_rq_target_io {
81 struct mapped_device *md;
83 struct request *orig, clone;
89 * For request-based dm.
90 * One of these is allocated per bio.
92 struct dm_rq_clone_bio_info {
94 struct dm_rq_target_io *tio;
97 union map_info *dm_get_mapinfo(struct bio *bio)
99 if (bio && bio->bi_private)
100 return &((struct dm_target_io *)bio->bi_private)->info;
104 union map_info *dm_get_rq_mapinfo(struct request *rq)
106 if (rq && rq->end_io_data)
107 return &((struct dm_rq_target_io *)rq->end_io_data)->info;
110 EXPORT_SYMBOL_GPL(dm_get_rq_mapinfo);
112 #define MINOR_ALLOCED ((void *)-1)
115 * Bits for the md->flags field.
117 #define DMF_BLOCK_IO_FOR_SUSPEND 0
118 #define DMF_SUSPENDED 1
120 #define DMF_FREEING 3
121 #define DMF_DELETING 4
122 #define DMF_NOFLUSH_SUSPENDING 5
123 #define DMF_MERGE_IS_OPTIONAL 6
126 * Work processed by per-device workqueue.
128 struct mapped_device {
129 struct rw_semaphore io_lock;
130 struct mutex suspend_lock;
137 struct request_queue *queue;
139 /* Protect queue and type against concurrent access. */
140 struct mutex type_lock;
142 struct target_type *immutable_target_type;
144 struct gendisk *disk;
150 * A list of ios that arrived while we were suspended.
153 wait_queue_head_t wait;
154 struct work_struct work;
155 struct bio_list deferred;
156 spinlock_t deferred_lock;
159 * Processing queue (flush)
161 struct workqueue_struct *wq;
164 * The current mapping.
166 struct dm_table *map;
169 * io objects are allocated from here.
180 wait_queue_head_t eventq;
182 struct list_head uevent_list;
183 spinlock_t uevent_lock; /* Protect access to uevent_list */
186 * freeze/thaw support require holding onto a super block
188 struct super_block *frozen_sb;
189 struct block_device *bdev;
191 /* forced geometry settings */
192 struct hd_geometry geometry;
197 /* zero-length flush that will be cloned and submitted to targets */
198 struct bio flush_bio;
202 * For mempools pre-allocation at the table loading time.
204 struct dm_md_mempools {
211 static struct kmem_cache *_io_cache;
212 static struct kmem_cache *_tio_cache;
213 static struct kmem_cache *_rq_tio_cache;
214 static struct kmem_cache *_rq_bio_info_cache;
216 static int __init local_init(void)
220 /* allocate a slab for the dm_ios */
221 _io_cache = KMEM_CACHE(dm_io, 0);
225 /* allocate a slab for the target ios */
226 _tio_cache = KMEM_CACHE(dm_target_io, 0);
228 goto out_free_io_cache;
230 _rq_tio_cache = KMEM_CACHE(dm_rq_target_io, 0);
232 goto out_free_tio_cache;
234 _rq_bio_info_cache = KMEM_CACHE(dm_rq_clone_bio_info, 0);
235 if (!_rq_bio_info_cache)
236 goto out_free_rq_tio_cache;
238 r = dm_uevent_init();
240 goto out_free_rq_bio_info_cache;
243 r = register_blkdev(_major, _name);
245 goto out_uevent_exit;
254 out_free_rq_bio_info_cache:
255 kmem_cache_destroy(_rq_bio_info_cache);
256 out_free_rq_tio_cache:
257 kmem_cache_destroy(_rq_tio_cache);
259 kmem_cache_destroy(_tio_cache);
261 kmem_cache_destroy(_io_cache);
266 static void local_exit(void)
268 kmem_cache_destroy(_rq_bio_info_cache);
269 kmem_cache_destroy(_rq_tio_cache);
270 kmem_cache_destroy(_tio_cache);
271 kmem_cache_destroy(_io_cache);
272 unregister_blkdev(_major, _name);
277 DMINFO("cleaned up");
280 static int (*_inits[])(void) __initdata = {
290 static void (*_exits[])(void) = {
300 static int __init dm_init(void)
302 const int count = ARRAY_SIZE(_inits);
306 for (i = 0; i < count; i++) {
321 static void __exit dm_exit(void)
323 int i = ARRAY_SIZE(_exits);
329 * Should be empty by this point.
331 idr_remove_all(&_minor_idr);
332 idr_destroy(&_minor_idr);
336 * Block device functions
338 int dm_deleting_md(struct mapped_device *md)
340 return test_bit(DMF_DELETING, &md->flags);
343 static int dm_blk_open(struct block_device *bdev, fmode_t mode)
345 struct mapped_device *md;
347 spin_lock(&_minor_lock);
349 md = bdev->bd_disk->private_data;
353 if (test_bit(DMF_FREEING, &md->flags) ||
354 dm_deleting_md(md)) {
360 atomic_inc(&md->open_count);
363 spin_unlock(&_minor_lock);
365 return md ? 0 : -ENXIO;
368 static int dm_blk_close(struct gendisk *disk, fmode_t mode)
370 struct mapped_device *md = disk->private_data;
372 spin_lock(&_minor_lock);
374 atomic_dec(&md->open_count);
377 spin_unlock(&_minor_lock);
382 int dm_open_count(struct mapped_device *md)
384 return atomic_read(&md->open_count);
388 * Guarantees nothing is using the device before it's deleted.
390 int dm_lock_for_deletion(struct mapped_device *md)
394 spin_lock(&_minor_lock);
396 if (dm_open_count(md))
399 set_bit(DMF_DELETING, &md->flags);
401 spin_unlock(&_minor_lock);
406 static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo)
408 struct mapped_device *md = bdev->bd_disk->private_data;
410 return dm_get_geometry(md, geo);
413 static int dm_blk_ioctl(struct block_device *bdev, fmode_t mode,
414 unsigned int cmd, unsigned long arg)
416 struct mapped_device *md = bdev->bd_disk->private_data;
417 struct dm_table *map = dm_get_live_table(md);
418 struct dm_target *tgt;
421 if (!map || !dm_table_get_size(map))
424 /* We only support devices that have a single target */
425 if (dm_table_get_num_targets(map) != 1)
428 tgt = dm_table_get_target(map, 0);
430 if (dm_suspended_md(md)) {
435 if (tgt->type->ioctl)
436 r = tgt->type->ioctl(tgt, cmd, arg);
444 static struct dm_io *alloc_io(struct mapped_device *md)
446 return mempool_alloc(md->io_pool, GFP_NOIO);
449 static void free_io(struct mapped_device *md, struct dm_io *io)
451 mempool_free(io, md->io_pool);
454 static void free_tio(struct mapped_device *md, struct dm_target_io *tio)
456 mempool_free(tio, md->tio_pool);
459 static struct dm_rq_target_io *alloc_rq_tio(struct mapped_device *md,
462 return mempool_alloc(md->tio_pool, gfp_mask);
465 static void free_rq_tio(struct dm_rq_target_io *tio)
467 mempool_free(tio, tio->md->tio_pool);
470 static struct dm_rq_clone_bio_info *alloc_bio_info(struct mapped_device *md)
472 return mempool_alloc(md->io_pool, GFP_ATOMIC);
475 static void free_bio_info(struct dm_rq_clone_bio_info *info)
477 mempool_free(info, info->tio->md->io_pool);
480 static int md_in_flight(struct mapped_device *md)
482 return atomic_read(&md->pending[READ]) +
483 atomic_read(&md->pending[WRITE]);
486 static void start_io_acct(struct dm_io *io)
488 struct mapped_device *md = io->md;
490 int rw = bio_data_dir(io->bio);
492 io->start_time = jiffies;
494 cpu = part_stat_lock();
495 part_round_stats(cpu, &dm_disk(md)->part0);
497 atomic_set(&dm_disk(md)->part0.in_flight[rw],
498 atomic_inc_return(&md->pending[rw]));
501 static void end_io_acct(struct dm_io *io)
503 struct mapped_device *md = io->md;
504 struct bio *bio = io->bio;
505 unsigned long duration = jiffies - io->start_time;
507 int rw = bio_data_dir(bio);
509 cpu = part_stat_lock();
510 part_round_stats(cpu, &dm_disk(md)->part0);
511 part_stat_add(cpu, &dm_disk(md)->part0, ticks[rw], duration);
515 * After this is decremented the bio must not be touched if it is
518 pending = atomic_dec_return(&md->pending[rw]);
519 atomic_set(&dm_disk(md)->part0.in_flight[rw], pending);
520 pending += atomic_read(&md->pending[rw^0x1]);
522 /* nudge anyone waiting on suspend queue */
528 * Add the bio to the list of deferred io.
530 static void queue_io(struct mapped_device *md, struct bio *bio)
534 spin_lock_irqsave(&md->deferred_lock, flags);
535 bio_list_add(&md->deferred, bio);
536 spin_unlock_irqrestore(&md->deferred_lock, flags);
537 queue_work(md->wq, &md->work);
541 * Everyone (including functions in this file), should use this
542 * function to access the md->map field, and make sure they call
543 * dm_table_put() when finished.
545 struct dm_table *dm_get_live_table(struct mapped_device *md)
550 read_lock_irqsave(&md->map_lock, flags);
554 read_unlock_irqrestore(&md->map_lock, flags);
560 * Get the geometry associated with a dm device
562 int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo)
570 * Set the geometry of a device.
572 int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo)
574 sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors;
576 if (geo->start > sz) {
577 DMWARN("Start sector is beyond the geometry limits.");
586 /*-----------------------------------------------------------------
588 * A more elegant soln is in the works that uses the queue
589 * merge fn, unfortunately there are a couple of changes to
590 * the block layer that I want to make for this. So in the
591 * interests of getting something for people to use I give
592 * you this clearly demarcated crap.
593 *---------------------------------------------------------------*/
595 static int __noflush_suspending(struct mapped_device *md)
597 return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
601 * Decrements the number of outstanding ios that a bio has been
602 * cloned into, completing the original io if necc.
604 static void dec_pending(struct dm_io *io, int error)
609 struct mapped_device *md = io->md;
611 /* Push-back supersedes any I/O errors */
612 if (unlikely(error)) {
613 spin_lock_irqsave(&io->endio_lock, flags);
614 if (!(io->error > 0 && __noflush_suspending(md)))
616 spin_unlock_irqrestore(&io->endio_lock, flags);
619 if (atomic_dec_and_test(&io->io_count)) {
620 if (io->error == DM_ENDIO_REQUEUE) {
622 * Target requested pushing back the I/O.
624 spin_lock_irqsave(&md->deferred_lock, flags);
625 if (__noflush_suspending(md))
626 bio_list_add_head(&md->deferred, io->bio);
628 /* noflush suspend was interrupted. */
630 spin_unlock_irqrestore(&md->deferred_lock, flags);
633 io_error = io->error;
638 if (io_error == DM_ENDIO_REQUEUE)
641 if ((bio->bi_rw & REQ_FLUSH) && bio->bi_size) {
643 * Preflush done for flush with data, reissue
646 bio->bi_rw &= ~REQ_FLUSH;
649 /* done with normal IO or empty flush */
650 trace_block_bio_complete(md->queue, bio, io_error);
651 bio_endio(bio, io_error);
656 static void clone_endio(struct bio *bio, int error)
659 struct dm_target_io *tio = bio->bi_private;
660 struct dm_io *io = tio->io;
661 struct mapped_device *md = tio->io->md;
662 dm_endio_fn endio = tio->ti->type->end_io;
664 if (!bio_flagged(bio, BIO_UPTODATE) && !error)
668 r = endio(tio->ti, bio, error, &tio->info);
669 if (r < 0 || r == DM_ENDIO_REQUEUE)
671 * error and requeue request are handled
675 else if (r == DM_ENDIO_INCOMPLETE)
676 /* The target will handle the io */
679 DMWARN("unimplemented target endio return value: %d", r);
685 * Store md for cleanup instead of tio which is about to get freed.
687 bio->bi_private = md->bs;
691 dec_pending(io, error);
695 * Partial completion handling for request-based dm
697 static void end_clone_bio(struct bio *clone, int error)
699 struct dm_rq_clone_bio_info *info = clone->bi_private;
700 struct dm_rq_target_io *tio = info->tio;
701 struct bio *bio = info->orig;
702 unsigned int nr_bytes = info->orig->bi_size;
708 * An error has already been detected on the request.
709 * Once error occurred, just let clone->end_io() handle
715 * Don't notice the error to the upper layer yet.
716 * The error handling decision is made by the target driver,
717 * when the request is completed.
724 * I/O for the bio successfully completed.
725 * Notice the data completion to the upper layer.
729 * bios are processed from the head of the list.
730 * So the completing bio should always be rq->bio.
731 * If it's not, something wrong is happening.
733 if (tio->orig->bio != bio)
734 DMERR("bio completion is going in the middle of the request");
737 * Update the original request.
738 * Do not use blk_end_request() here, because it may complete
739 * the original request before the clone, and break the ordering.
741 blk_update_request(tio->orig, 0, nr_bytes);
745 * Don't touch any member of the md after calling this function because
746 * the md may be freed in dm_put() at the end of this function.
747 * Or do dm_get() before calling this function and dm_put() later.
749 static void rq_completed(struct mapped_device *md, int rw, int run_queue)
751 atomic_dec(&md->pending[rw]);
753 /* nudge anyone waiting on suspend queue */
754 if (!md_in_flight(md))
758 blk_run_queue(md->queue);
761 * dm_put() must be at the end of this function. See the comment above
766 static void free_rq_clone(struct request *clone)
768 struct dm_rq_target_io *tio = clone->end_io_data;
770 blk_rq_unprep_clone(clone);
775 * Complete the clone and the original request.
776 * Must be called without queue lock.
778 static void dm_end_request(struct request *clone, int error)
780 int rw = rq_data_dir(clone);
781 struct dm_rq_target_io *tio = clone->end_io_data;
782 struct mapped_device *md = tio->md;
783 struct request *rq = tio->orig;
785 if (rq->cmd_type == REQ_TYPE_BLOCK_PC) {
786 rq->errors = clone->errors;
787 rq->resid_len = clone->resid_len;
791 * We are using the sense buffer of the original
793 * So setting the length of the sense data is enough.
795 rq->sense_len = clone->sense_len;
798 free_rq_clone(clone);
799 blk_end_request_all(rq, error);
800 rq_completed(md, rw, true);
803 static void dm_unprep_request(struct request *rq)
805 struct request *clone = rq->special;
808 rq->cmd_flags &= ~REQ_DONTPREP;
810 free_rq_clone(clone);
814 * Requeue the original request of a clone.
816 void dm_requeue_unmapped_request(struct request *clone)
818 int rw = rq_data_dir(clone);
819 struct dm_rq_target_io *tio = clone->end_io_data;
820 struct mapped_device *md = tio->md;
821 struct request *rq = tio->orig;
822 struct request_queue *q = rq->q;
825 dm_unprep_request(rq);
827 spin_lock_irqsave(q->queue_lock, flags);
828 blk_requeue_request(q, rq);
829 spin_unlock_irqrestore(q->queue_lock, flags);
831 rq_completed(md, rw, 0);
833 EXPORT_SYMBOL_GPL(dm_requeue_unmapped_request);
835 static void __stop_queue(struct request_queue *q)
840 static void stop_queue(struct request_queue *q)
844 spin_lock_irqsave(q->queue_lock, flags);
846 spin_unlock_irqrestore(q->queue_lock, flags);
849 static void __start_queue(struct request_queue *q)
851 if (blk_queue_stopped(q))
855 static void start_queue(struct request_queue *q)
859 spin_lock_irqsave(q->queue_lock, flags);
861 spin_unlock_irqrestore(q->queue_lock, flags);
864 static void dm_done(struct request *clone, int error, bool mapped)
867 struct dm_rq_target_io *tio = clone->end_io_data;
868 dm_request_endio_fn rq_end_io = tio->ti->type->rq_end_io;
870 if (mapped && rq_end_io)
871 r = rq_end_io(tio->ti, clone, error, &tio->info);
874 /* The target wants to complete the I/O */
875 dm_end_request(clone, r);
876 else if (r == DM_ENDIO_INCOMPLETE)
877 /* The target will handle the I/O */
879 else if (r == DM_ENDIO_REQUEUE)
880 /* The target wants to requeue the I/O */
881 dm_requeue_unmapped_request(clone);
883 DMWARN("unimplemented target endio return value: %d", r);
889 * Request completion handler for request-based dm
891 static void dm_softirq_done(struct request *rq)
894 struct request *clone = rq->completion_data;
895 struct dm_rq_target_io *tio = clone->end_io_data;
897 if (rq->cmd_flags & REQ_FAILED)
900 dm_done(clone, tio->error, mapped);
904 * Complete the clone and the original request with the error status
905 * through softirq context.
907 static void dm_complete_request(struct request *clone, int error)
909 struct dm_rq_target_io *tio = clone->end_io_data;
910 struct request *rq = tio->orig;
913 rq->completion_data = clone;
914 blk_complete_request(rq);
918 * Complete the not-mapped clone and the original request with the error status
919 * through softirq context.
920 * Target's rq_end_io() function isn't called.
921 * This may be used when the target's map_rq() function fails.
923 void dm_kill_unmapped_request(struct request *clone, int error)
925 struct dm_rq_target_io *tio = clone->end_io_data;
926 struct request *rq = tio->orig;
928 rq->cmd_flags |= REQ_FAILED;
929 dm_complete_request(clone, error);
931 EXPORT_SYMBOL_GPL(dm_kill_unmapped_request);
934 * Called with the queue lock held
936 static void end_clone_request(struct request *clone, int error)
939 * For just cleaning up the information of the queue in which
940 * the clone was dispatched.
941 * The clone is *NOT* freed actually here because it is alloced from
942 * dm own mempool and REQ_ALLOCED isn't set in clone->cmd_flags.
944 __blk_put_request(clone->q, clone);
947 * Actual request completion is done in a softirq context which doesn't
948 * hold the queue lock. Otherwise, deadlock could occur because:
949 * - another request may be submitted by the upper level driver
950 * of the stacking during the completion
951 * - the submission which requires queue lock may be done
954 dm_complete_request(clone, error);
958 * Return maximum size of I/O possible at the supplied sector up to the current
961 static sector_t max_io_len_target_boundary(sector_t sector, struct dm_target *ti)
963 sector_t target_offset = dm_target_offset(ti, sector);
965 return ti->len - target_offset;
968 static sector_t max_io_len(sector_t sector, struct dm_target *ti)
970 sector_t len = max_io_len_target_boundary(sector, ti);
971 sector_t offset, max_len;
974 * Does the target need to split even further?
976 if (ti->max_io_len) {
977 offset = dm_target_offset(ti, sector);
978 if (unlikely(ti->max_io_len & (ti->max_io_len - 1)))
979 max_len = sector_div(offset, ti->max_io_len);
981 max_len = offset & (ti->max_io_len - 1);
982 max_len = ti->max_io_len - max_len;
991 int dm_set_target_max_io_len(struct dm_target *ti, sector_t len)
993 if (len > UINT_MAX) {
994 DMERR("Specified maximum size of target IO (%llu) exceeds limit (%u)",
995 (unsigned long long)len, UINT_MAX);
996 ti->error = "Maximum size of target IO is too large";
1000 ti->max_io_len = (uint32_t) len;
1004 EXPORT_SYMBOL_GPL(dm_set_target_max_io_len);
1006 static void __map_bio(struct dm_target *ti, struct bio *clone,
1007 struct dm_target_io *tio)
1011 struct mapped_device *md;
1013 clone->bi_end_io = clone_endio;
1014 clone->bi_private = tio;
1017 * Map the clone. If r == 0 we don't need to do
1018 * anything, the target has assumed ownership of
1021 atomic_inc(&tio->io->io_count);
1022 sector = clone->bi_sector;
1023 r = ti->type->map(ti, clone, &tio->info);
1024 if (r == DM_MAPIO_REMAPPED) {
1025 /* the bio has been remapped so dispatch it */
1027 trace_block_bio_remap(bdev_get_queue(clone->bi_bdev), clone,
1028 tio->io->bio->bi_bdev->bd_dev, sector);
1030 generic_make_request(clone);
1031 } else if (r < 0 || r == DM_MAPIO_REQUEUE) {
1032 /* error the io and bail out, or requeue it if needed */
1034 dec_pending(tio->io, r);
1036 * Store bio_set for cleanup.
1038 clone->bi_end_io = NULL;
1039 clone->bi_private = md->bs;
1043 DMWARN("unimplemented target map return value: %d", r);
1049 struct mapped_device *md;
1050 struct dm_table *map;
1054 sector_t sector_count;
1058 static void dm_bio_destructor(struct bio *bio)
1060 struct bio_set *bs = bio->bi_private;
1066 * Creates a little bio that just does part of a bvec.
1068 static struct bio *split_bvec(struct bio *bio, sector_t sector,
1069 unsigned short idx, unsigned int offset,
1070 unsigned int len, struct bio_set *bs)
1073 struct bio_vec *bv = bio->bi_io_vec + idx;
1075 clone = bio_alloc_bioset(GFP_NOIO, 1, bs);
1076 clone->bi_destructor = dm_bio_destructor;
1077 *clone->bi_io_vec = *bv;
1079 clone->bi_sector = sector;
1080 clone->bi_bdev = bio->bi_bdev;
1081 clone->bi_rw = bio->bi_rw;
1083 clone->bi_size = to_bytes(len);
1084 clone->bi_io_vec->bv_offset = offset;
1085 clone->bi_io_vec->bv_len = clone->bi_size;
1086 clone->bi_flags |= 1 << BIO_CLONED;
1088 if (bio_integrity(bio)) {
1089 bio_integrity_clone(clone, bio, GFP_NOIO, bs);
1090 bio_integrity_trim(clone,
1091 bio_sector_offset(bio, idx, offset), len);
1098 * Creates a bio that consists of range of complete bvecs.
1100 static struct bio *clone_bio(struct bio *bio, sector_t sector,
1101 unsigned short idx, unsigned short bv_count,
1102 unsigned int len, struct bio_set *bs)
1106 clone = bio_alloc_bioset(GFP_NOIO, bio->bi_max_vecs, bs);
1107 __bio_clone(clone, bio);
1108 clone->bi_destructor = dm_bio_destructor;
1109 clone->bi_sector = sector;
1110 clone->bi_idx = idx;
1111 clone->bi_vcnt = idx + bv_count;
1112 clone->bi_size = to_bytes(len);
1113 clone->bi_flags &= ~(1 << BIO_SEG_VALID);
1115 if (bio_integrity(bio)) {
1116 bio_integrity_clone(clone, bio, GFP_NOIO, bs);
1118 if (idx != bio->bi_idx || clone->bi_size < bio->bi_size)
1119 bio_integrity_trim(clone,
1120 bio_sector_offset(bio, idx, 0), len);
1126 static struct dm_target_io *alloc_tio(struct clone_info *ci,
1127 struct dm_target *ti)
1129 struct dm_target_io *tio = mempool_alloc(ci->md->tio_pool, GFP_NOIO);
1133 memset(&tio->info, 0, sizeof(tio->info));
1138 static void __issue_target_request(struct clone_info *ci, struct dm_target *ti,
1139 unsigned request_nr, sector_t len)
1141 struct dm_target_io *tio = alloc_tio(ci, ti);
1144 tio->info.target_request_nr = request_nr;
1147 * Discard requests require the bio's inline iovecs be initialized.
1148 * ci->bio->bi_max_vecs is BIO_INLINE_VECS anyway, for both flush
1149 * and discard, so no need for concern about wasted bvec allocations.
1151 clone = bio_alloc_bioset(GFP_NOIO, ci->bio->bi_max_vecs, ci->md->bs);
1152 __bio_clone(clone, ci->bio);
1153 clone->bi_destructor = dm_bio_destructor;
1155 clone->bi_sector = ci->sector;
1156 clone->bi_size = to_bytes(len);
1159 __map_bio(ti, clone, tio);
1162 static void __issue_target_requests(struct clone_info *ci, struct dm_target *ti,
1163 unsigned num_requests, sector_t len)
1165 unsigned request_nr;
1167 for (request_nr = 0; request_nr < num_requests; request_nr++)
1168 __issue_target_request(ci, ti, request_nr, len);
1171 static int __clone_and_map_empty_flush(struct clone_info *ci)
1173 unsigned target_nr = 0;
1174 struct dm_target *ti;
1176 BUG_ON(bio_has_data(ci->bio));
1177 while ((ti = dm_table_get_target(ci->map, target_nr++)))
1178 __issue_target_requests(ci, ti, ti->num_flush_requests, 0);
1184 * Perform all io with a single clone.
1186 static void __clone_and_map_simple(struct clone_info *ci, struct dm_target *ti)
1188 struct bio *clone, *bio = ci->bio;
1189 struct dm_target_io *tio;
1191 tio = alloc_tio(ci, ti);
1192 clone = clone_bio(bio, ci->sector, ci->idx,
1193 bio->bi_vcnt - ci->idx, ci->sector_count,
1195 __map_bio(ti, clone, tio);
1196 ci->sector_count = 0;
1199 static int __clone_and_map_discard(struct clone_info *ci)
1201 struct dm_target *ti;
1205 ti = dm_table_find_target(ci->map, ci->sector);
1206 if (!dm_target_is_valid(ti))
1210 * Even though the device advertised discard support,
1211 * that does not mean every target supports it, and
1212 * reconfiguration might also have changed that since the
1213 * check was performed.
1215 if (!ti->num_discard_requests)
1218 if (!ti->split_discard_requests)
1219 len = min(ci->sector_count, max_io_len_target_boundary(ci->sector, ti));
1221 len = min(ci->sector_count, max_io_len(ci->sector, ti));
1223 __issue_target_requests(ci, ti, ti->num_discard_requests, len);
1226 } while (ci->sector_count -= len);
1231 static int __clone_and_map(struct clone_info *ci)
1233 struct bio *clone, *bio = ci->bio;
1234 struct dm_target *ti;
1235 sector_t len = 0, max;
1236 struct dm_target_io *tio;
1238 if (unlikely(bio->bi_rw & REQ_DISCARD))
1239 return __clone_and_map_discard(ci);
1241 ti = dm_table_find_target(ci->map, ci->sector);
1242 if (!dm_target_is_valid(ti))
1245 max = max_io_len(ci->sector, ti);
1247 if (ci->sector_count <= max) {
1249 * Optimise for the simple case where we can do all of
1250 * the remaining io with a single clone.
1252 __clone_and_map_simple(ci, ti);
1254 } else if (to_sector(bio->bi_io_vec[ci->idx].bv_len) <= max) {
1256 * There are some bvecs that don't span targets.
1257 * Do as many of these as possible.
1260 sector_t remaining = max;
1263 for (i = ci->idx; remaining && (i < bio->bi_vcnt); i++) {
1264 bv_len = to_sector(bio->bi_io_vec[i].bv_len);
1266 if (bv_len > remaining)
1269 remaining -= bv_len;
1273 tio = alloc_tio(ci, ti);
1274 clone = clone_bio(bio, ci->sector, ci->idx, i - ci->idx, len,
1276 __map_bio(ti, clone, tio);
1279 ci->sector_count -= len;
1284 * Handle a bvec that must be split between two or more targets.
1286 struct bio_vec *bv = bio->bi_io_vec + ci->idx;
1287 sector_t remaining = to_sector(bv->bv_len);
1288 unsigned int offset = 0;
1292 ti = dm_table_find_target(ci->map, ci->sector);
1293 if (!dm_target_is_valid(ti))
1296 max = max_io_len(ci->sector, ti);
1299 len = min(remaining, max);
1301 tio = alloc_tio(ci, ti);
1302 clone = split_bvec(bio, ci->sector, ci->idx,
1303 bv->bv_offset + offset, len,
1306 __map_bio(ti, clone, tio);
1309 ci->sector_count -= len;
1310 offset += to_bytes(len);
1311 } while (remaining -= len);
1320 * Split the bio into several clones and submit it to targets.
1322 static void __split_and_process_bio(struct mapped_device *md, struct bio *bio)
1324 struct clone_info ci;
1327 ci.map = dm_get_live_table(md);
1328 if (unlikely(!ci.map)) {
1334 ci.io = alloc_io(md);
1336 atomic_set(&ci.io->io_count, 1);
1339 spin_lock_init(&ci.io->endio_lock);
1340 ci.sector = bio->bi_sector;
1341 ci.idx = bio->bi_idx;
1343 start_io_acct(ci.io);
1344 if (bio->bi_rw & REQ_FLUSH) {
1345 ci.bio = &ci.md->flush_bio;
1346 ci.sector_count = 0;
1347 error = __clone_and_map_empty_flush(&ci);
1348 /* dec_pending submits any data associated with flush */
1351 ci.sector_count = bio_sectors(bio);
1352 while (ci.sector_count && !error)
1353 error = __clone_and_map(&ci);
1356 /* drop the extra reference count */
1357 dec_pending(ci.io, error);
1358 dm_table_put(ci.map);
1360 /*-----------------------------------------------------------------
1362 *---------------------------------------------------------------*/
1364 static int dm_merge_bvec(struct request_queue *q,
1365 struct bvec_merge_data *bvm,
1366 struct bio_vec *biovec)
1368 struct mapped_device *md = q->queuedata;
1369 struct dm_table *map = dm_get_live_table(md);
1370 struct dm_target *ti;
1371 sector_t max_sectors;
1377 ti = dm_table_find_target(map, bvm->bi_sector);
1378 if (!dm_target_is_valid(ti))
1382 * Find maximum amount of I/O that won't need splitting
1384 max_sectors = min(max_io_len(bvm->bi_sector, ti),
1385 (sector_t) BIO_MAX_SECTORS);
1386 max_size = (max_sectors << SECTOR_SHIFT) - bvm->bi_size;
1391 * merge_bvec_fn() returns number of bytes
1392 * it can accept at this offset
1393 * max is precomputed maximal io size
1395 if (max_size && ti->type->merge)
1396 max_size = ti->type->merge(ti, bvm, biovec, max_size);
1398 * If the target doesn't support merge method and some of the devices
1399 * provided their merge_bvec method (we know this by looking at
1400 * queue_max_hw_sectors), then we can't allow bios with multiple vector
1401 * entries. So always set max_size to 0, and the code below allows
1404 else if (queue_max_hw_sectors(q) <= PAGE_SIZE >> 9)
1413 * Always allow an entire first page
1415 if (max_size <= biovec->bv_len && !(bvm->bi_size >> SECTOR_SHIFT))
1416 max_size = biovec->bv_len;
1422 * The request function that just remaps the bio built up by
1425 static void _dm_request(struct request_queue *q, struct bio *bio)
1427 int rw = bio_data_dir(bio);
1428 struct mapped_device *md = q->queuedata;
1431 down_read(&md->io_lock);
1433 cpu = part_stat_lock();
1434 part_stat_inc(cpu, &dm_disk(md)->part0, ios[rw]);
1435 part_stat_add(cpu, &dm_disk(md)->part0, sectors[rw], bio_sectors(bio));
1438 /* if we're suspended, we have to queue this io for later */
1439 if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))) {
1440 up_read(&md->io_lock);
1442 if (bio_rw(bio) != READA)
1449 __split_and_process_bio(md, bio);
1450 up_read(&md->io_lock);
1454 static int dm_request_based(struct mapped_device *md)
1456 return blk_queue_stackable(md->queue);
1459 static void dm_request(struct request_queue *q, struct bio *bio)
1461 struct mapped_device *md = q->queuedata;
1463 if (dm_request_based(md))
1464 blk_queue_bio(q, bio);
1466 _dm_request(q, bio);
1469 void dm_dispatch_request(struct request *rq)
1473 if (blk_queue_io_stat(rq->q))
1474 rq->cmd_flags |= REQ_IO_STAT;
1476 rq->start_time = jiffies;
1477 r = blk_insert_cloned_request(rq->q, rq);
1479 dm_complete_request(rq, r);
1481 EXPORT_SYMBOL_GPL(dm_dispatch_request);
1483 static void dm_rq_bio_destructor(struct bio *bio)
1485 struct dm_rq_clone_bio_info *info = bio->bi_private;
1486 struct mapped_device *md = info->tio->md;
1488 free_bio_info(info);
1489 bio_free(bio, md->bs);
1492 static int dm_rq_bio_constructor(struct bio *bio, struct bio *bio_orig,
1495 struct dm_rq_target_io *tio = data;
1496 struct mapped_device *md = tio->md;
1497 struct dm_rq_clone_bio_info *info = alloc_bio_info(md);
1502 info->orig = bio_orig;
1504 bio->bi_end_io = end_clone_bio;
1505 bio->bi_private = info;
1506 bio->bi_destructor = dm_rq_bio_destructor;
1511 static int setup_clone(struct request *clone, struct request *rq,
1512 struct dm_rq_target_io *tio)
1516 r = blk_rq_prep_clone(clone, rq, tio->md->bs, GFP_ATOMIC,
1517 dm_rq_bio_constructor, tio);
1521 clone->cmd = rq->cmd;
1522 clone->cmd_len = rq->cmd_len;
1523 clone->sense = rq->sense;
1524 clone->buffer = rq->buffer;
1525 clone->end_io = end_clone_request;
1526 clone->end_io_data = tio;
1531 static struct request *clone_rq(struct request *rq, struct mapped_device *md,
1534 struct request *clone;
1535 struct dm_rq_target_io *tio;
1537 tio = alloc_rq_tio(md, gfp_mask);
1545 memset(&tio->info, 0, sizeof(tio->info));
1547 clone = &tio->clone;
1548 if (setup_clone(clone, rq, tio)) {
1558 * Called with the queue lock held.
1560 static int dm_prep_fn(struct request_queue *q, struct request *rq)
1562 struct mapped_device *md = q->queuedata;
1563 struct request *clone;
1565 if (unlikely(rq->special)) {
1566 DMWARN("Already has something in rq->special.");
1567 return BLKPREP_KILL;
1570 clone = clone_rq(rq, md, GFP_ATOMIC);
1572 return BLKPREP_DEFER;
1574 rq->special = clone;
1575 rq->cmd_flags |= REQ_DONTPREP;
1582 * 0 : the request has been processed (not requeued)
1583 * !0 : the request has been requeued
1585 static int map_request(struct dm_target *ti, struct request *clone,
1586 struct mapped_device *md)
1588 int r, requeued = 0;
1589 struct dm_rq_target_io *tio = clone->end_io_data;
1592 * Hold the md reference here for the in-flight I/O.
1593 * We can't rely on the reference count by device opener,
1594 * because the device may be closed during the request completion
1595 * when all bios are completed.
1596 * See the comment in rq_completed() too.
1601 r = ti->type->map_rq(ti, clone, &tio->info);
1603 case DM_MAPIO_SUBMITTED:
1604 /* The target has taken the I/O to submit by itself later */
1606 case DM_MAPIO_REMAPPED:
1607 /* The target has remapped the I/O so dispatch it */
1608 trace_block_rq_remap(clone->q, clone, disk_devt(dm_disk(md)),
1609 blk_rq_pos(tio->orig));
1610 dm_dispatch_request(clone);
1612 case DM_MAPIO_REQUEUE:
1613 /* The target wants to requeue the I/O */
1614 dm_requeue_unmapped_request(clone);
1619 DMWARN("unimplemented target map return value: %d", r);
1623 /* The target wants to complete the I/O */
1624 dm_kill_unmapped_request(clone, r);
1632 * q->request_fn for request-based dm.
1633 * Called with the queue lock held.
1635 static void dm_request_fn(struct request_queue *q)
1637 struct mapped_device *md = q->queuedata;
1638 struct dm_table *map = dm_get_live_table(md);
1639 struct dm_target *ti;
1640 struct request *rq, *clone;
1644 * For suspend, check blk_queue_stopped() and increment
1645 * ->pending within a single queue_lock not to increment the
1646 * number of in-flight I/Os after the queue is stopped in
1649 while (!blk_queue_stopped(q)) {
1650 rq = blk_peek_request(q);
1654 /* always use block 0 to find the target for flushes for now */
1656 if (!(rq->cmd_flags & REQ_FLUSH))
1657 pos = blk_rq_pos(rq);
1659 ti = dm_table_find_target(map, pos);
1660 BUG_ON(!dm_target_is_valid(ti));
1662 if (ti->type->busy && ti->type->busy(ti))
1665 blk_start_request(rq);
1666 clone = rq->special;
1667 atomic_inc(&md->pending[rq_data_dir(clone)]);
1669 spin_unlock(q->queue_lock);
1670 if (map_request(ti, clone, md))
1673 BUG_ON(!irqs_disabled());
1674 spin_lock(q->queue_lock);
1680 BUG_ON(!irqs_disabled());
1681 spin_lock(q->queue_lock);
1684 blk_delay_queue(q, HZ / 10);
1691 int dm_underlying_device_busy(struct request_queue *q)
1693 return blk_lld_busy(q);
1695 EXPORT_SYMBOL_GPL(dm_underlying_device_busy);
1697 static int dm_lld_busy(struct request_queue *q)
1700 struct mapped_device *md = q->queuedata;
1701 struct dm_table *map = dm_get_live_table(md);
1703 if (!map || test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))
1706 r = dm_table_any_busy_target(map);
1713 static int dm_any_congested(void *congested_data, int bdi_bits)
1716 struct mapped_device *md = congested_data;
1717 struct dm_table *map;
1719 if (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
1720 map = dm_get_live_table(md);
1723 * Request-based dm cares about only own queue for
1724 * the query about congestion status of request_queue
1726 if (dm_request_based(md))
1727 r = md->queue->backing_dev_info.state &
1730 r = dm_table_any_congested(map, bdi_bits);
1739 /*-----------------------------------------------------------------
1740 * An IDR is used to keep track of allocated minor numbers.
1741 *---------------------------------------------------------------*/
1742 static void free_minor(int minor)
1744 spin_lock(&_minor_lock);
1745 idr_remove(&_minor_idr, minor);
1746 spin_unlock(&_minor_lock);
1750 * See if the device with a specific minor # is free.
1752 static int specific_minor(int minor)
1756 if (minor >= (1 << MINORBITS))
1759 r = idr_pre_get(&_minor_idr, GFP_KERNEL);
1763 spin_lock(&_minor_lock);
1765 if (idr_find(&_minor_idr, minor)) {
1770 r = idr_get_new_above(&_minor_idr, MINOR_ALLOCED, minor, &m);
1775 idr_remove(&_minor_idr, m);
1781 spin_unlock(&_minor_lock);
1785 static int next_free_minor(int *minor)
1789 r = idr_pre_get(&_minor_idr, GFP_KERNEL);
1793 spin_lock(&_minor_lock);
1795 r = idr_get_new(&_minor_idr, MINOR_ALLOCED, &m);
1799 if (m >= (1 << MINORBITS)) {
1800 idr_remove(&_minor_idr, m);
1808 spin_unlock(&_minor_lock);
1812 static const struct block_device_operations dm_blk_dops;
1814 static void dm_wq_work(struct work_struct *work);
1816 static void dm_init_md_queue(struct mapped_device *md)
1819 * Request-based dm devices cannot be stacked on top of bio-based dm
1820 * devices. The type of this dm device has not been decided yet.
1821 * The type is decided at the first table loading time.
1822 * To prevent problematic device stacking, clear the queue flag
1823 * for request stacking support until then.
1825 * This queue is new, so no concurrency on the queue_flags.
1827 queue_flag_clear_unlocked(QUEUE_FLAG_STACKABLE, md->queue);
1829 md->queue->queuedata = md;
1830 md->queue->backing_dev_info.congested_fn = dm_any_congested;
1831 md->queue->backing_dev_info.congested_data = md;
1832 blk_queue_make_request(md->queue, dm_request);
1833 blk_queue_bounce_limit(md->queue, BLK_BOUNCE_ANY);
1834 blk_queue_merge_bvec(md->queue, dm_merge_bvec);
1838 * Allocate and initialise a blank device with a given minor.
1840 static struct mapped_device *alloc_dev(int minor)
1843 struct mapped_device *md = kzalloc(sizeof(*md), GFP_KERNEL);
1847 DMWARN("unable to allocate device, out of memory.");
1851 if (!try_module_get(THIS_MODULE))
1852 goto bad_module_get;
1854 /* get a minor number for the dev */
1855 if (minor == DM_ANY_MINOR)
1856 r = next_free_minor(&minor);
1858 r = specific_minor(minor);
1862 md->type = DM_TYPE_NONE;
1863 init_rwsem(&md->io_lock);
1864 mutex_init(&md->suspend_lock);
1865 mutex_init(&md->type_lock);
1866 spin_lock_init(&md->deferred_lock);
1867 rwlock_init(&md->map_lock);
1868 atomic_set(&md->holders, 1);
1869 atomic_set(&md->open_count, 0);
1870 atomic_set(&md->event_nr, 0);
1871 atomic_set(&md->uevent_seq, 0);
1872 INIT_LIST_HEAD(&md->uevent_list);
1873 spin_lock_init(&md->uevent_lock);
1875 md->queue = blk_alloc_queue(GFP_KERNEL);
1879 dm_init_md_queue(md);
1881 md->disk = alloc_disk(1);
1885 atomic_set(&md->pending[0], 0);
1886 atomic_set(&md->pending[1], 0);
1887 init_waitqueue_head(&md->wait);
1888 INIT_WORK(&md->work, dm_wq_work);
1889 init_waitqueue_head(&md->eventq);
1891 md->disk->major = _major;
1892 md->disk->first_minor = minor;
1893 md->disk->fops = &dm_blk_dops;
1894 md->disk->queue = md->queue;
1895 md->disk->private_data = md;
1896 sprintf(md->disk->disk_name, "dm-%d", minor);
1898 format_dev_t(md->name, MKDEV(_major, minor));
1900 md->wq = alloc_workqueue("kdmflush",
1901 WQ_NON_REENTRANT | WQ_MEM_RECLAIM, 0);
1905 md->bdev = bdget_disk(md->disk, 0);
1909 bio_init(&md->flush_bio);
1910 md->flush_bio.bi_bdev = md->bdev;
1911 md->flush_bio.bi_rw = WRITE_FLUSH;
1913 /* Populate the mapping, nobody knows we exist yet */
1914 spin_lock(&_minor_lock);
1915 old_md = idr_replace(&_minor_idr, md, minor);
1916 spin_unlock(&_minor_lock);
1918 BUG_ON(old_md != MINOR_ALLOCED);
1923 destroy_workqueue(md->wq);
1925 del_gendisk(md->disk);
1928 blk_cleanup_queue(md->queue);
1932 module_put(THIS_MODULE);
1938 static void unlock_fs(struct mapped_device *md);
1940 static void free_dev(struct mapped_device *md)
1942 int minor = MINOR(disk_devt(md->disk));
1946 destroy_workqueue(md->wq);
1948 mempool_destroy(md->tio_pool);
1950 mempool_destroy(md->io_pool);
1952 bioset_free(md->bs);
1953 blk_integrity_unregister(md->disk);
1954 del_gendisk(md->disk);
1957 spin_lock(&_minor_lock);
1958 md->disk->private_data = NULL;
1959 spin_unlock(&_minor_lock);
1962 blk_cleanup_queue(md->queue);
1963 module_put(THIS_MODULE);
1967 static void __bind_mempools(struct mapped_device *md, struct dm_table *t)
1969 struct dm_md_mempools *p;
1971 if (md->io_pool && md->tio_pool && md->bs)
1972 /* the md already has necessary mempools */
1975 p = dm_table_get_md_mempools(t);
1976 BUG_ON(!p || md->io_pool || md->tio_pool || md->bs);
1978 md->io_pool = p->io_pool;
1980 md->tio_pool = p->tio_pool;
1986 /* mempool bind completed, now no need any mempools in the table */
1987 dm_table_free_md_mempools(t);
1991 * Bind a table to the device.
1993 static void event_callback(void *context)
1995 unsigned long flags;
1997 struct mapped_device *md = (struct mapped_device *) context;
1999 spin_lock_irqsave(&md->uevent_lock, flags);
2000 list_splice_init(&md->uevent_list, &uevents);
2001 spin_unlock_irqrestore(&md->uevent_lock, flags);
2003 dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj);
2005 atomic_inc(&md->event_nr);
2006 wake_up(&md->eventq);
2010 * Protected by md->suspend_lock obtained by dm_swap_table().
2012 static void __set_size(struct mapped_device *md, sector_t size)
2014 set_capacity(md->disk, size);
2016 i_size_write(md->bdev->bd_inode, (loff_t)size << SECTOR_SHIFT);
2020 * Return 1 if the queue has a compulsory merge_bvec_fn function.
2022 * If this function returns 0, then the device is either a non-dm
2023 * device without a merge_bvec_fn, or it is a dm device that is
2024 * able to split any bios it receives that are too big.
2026 int dm_queue_merge_is_compulsory(struct request_queue *q)
2028 struct mapped_device *dev_md;
2030 if (!q->merge_bvec_fn)
2033 if (q->make_request_fn == dm_request) {
2034 dev_md = q->queuedata;
2035 if (test_bit(DMF_MERGE_IS_OPTIONAL, &dev_md->flags))
2042 static int dm_device_merge_is_compulsory(struct dm_target *ti,
2043 struct dm_dev *dev, sector_t start,
2044 sector_t len, void *data)
2046 struct block_device *bdev = dev->bdev;
2047 struct request_queue *q = bdev_get_queue(bdev);
2049 return dm_queue_merge_is_compulsory(q);
2053 * Return 1 if it is acceptable to ignore merge_bvec_fn based
2054 * on the properties of the underlying devices.
2056 static int dm_table_merge_is_optional(struct dm_table *table)
2059 struct dm_target *ti;
2061 while (i < dm_table_get_num_targets(table)) {
2062 ti = dm_table_get_target(table, i++);
2064 if (ti->type->iterate_devices &&
2065 ti->type->iterate_devices(ti, dm_device_merge_is_compulsory, NULL))
2073 * Returns old map, which caller must destroy.
2075 static struct dm_table *__bind(struct mapped_device *md, struct dm_table *t,
2076 struct queue_limits *limits)
2078 struct dm_table *old_map;
2079 struct request_queue *q = md->queue;
2081 unsigned long flags;
2082 int merge_is_optional;
2084 size = dm_table_get_size(t);
2087 * Wipe any geometry if the size of the table changed.
2089 if (size != get_capacity(md->disk))
2090 memset(&md->geometry, 0, sizeof(md->geometry));
2092 __set_size(md, size);
2094 dm_table_event_callback(t, event_callback, md);
2097 * The queue hasn't been stopped yet, if the old table type wasn't
2098 * for request-based during suspension. So stop it to prevent
2099 * I/O mapping before resume.
2100 * This must be done before setting the queue restrictions,
2101 * because request-based dm may be run just after the setting.
2103 if (dm_table_request_based(t) && !blk_queue_stopped(q))
2106 __bind_mempools(md, t);
2108 merge_is_optional = dm_table_merge_is_optional(t);
2110 write_lock_irqsave(&md->map_lock, flags);
2113 md->immutable_target_type = dm_table_get_immutable_target_type(t);
2115 dm_table_set_restrictions(t, q, limits);
2116 if (merge_is_optional)
2117 set_bit(DMF_MERGE_IS_OPTIONAL, &md->flags);
2119 clear_bit(DMF_MERGE_IS_OPTIONAL, &md->flags);
2120 write_unlock_irqrestore(&md->map_lock, flags);
2126 * Returns unbound table for the caller to free.
2128 static struct dm_table *__unbind(struct mapped_device *md)
2130 struct dm_table *map = md->map;
2131 unsigned long flags;
2136 dm_table_event_callback(map, NULL, NULL);
2137 write_lock_irqsave(&md->map_lock, flags);
2139 write_unlock_irqrestore(&md->map_lock, flags);
2145 * Constructor for a new device.
2147 int dm_create(int minor, struct mapped_device **result)
2149 struct mapped_device *md;
2151 md = alloc_dev(minor);
2162 * Functions to manage md->type.
2163 * All are required to hold md->type_lock.
2165 void dm_lock_md_type(struct mapped_device *md)
2167 mutex_lock(&md->type_lock);
2170 void dm_unlock_md_type(struct mapped_device *md)
2172 mutex_unlock(&md->type_lock);
2175 void dm_set_md_type(struct mapped_device *md, unsigned type)
2180 unsigned dm_get_md_type(struct mapped_device *md)
2185 struct target_type *dm_get_immutable_target_type(struct mapped_device *md)
2187 return md->immutable_target_type;
2191 * Fully initialize a request-based queue (->elevator, ->request_fn, etc).
2193 static int dm_init_request_based_queue(struct mapped_device *md)
2195 struct request_queue *q = NULL;
2197 if (md->queue->elevator)
2200 /* Fully initialize the queue */
2201 q = blk_init_allocated_queue(md->queue, dm_request_fn, NULL);
2206 dm_init_md_queue(md);
2207 blk_queue_softirq_done(md->queue, dm_softirq_done);
2208 blk_queue_prep_rq(md->queue, dm_prep_fn);
2209 blk_queue_lld_busy(md->queue, dm_lld_busy);
2211 elv_register_queue(md->queue);
2217 * Setup the DM device's queue based on md's type
2219 int dm_setup_md_queue(struct mapped_device *md)
2221 if ((dm_get_md_type(md) == DM_TYPE_REQUEST_BASED) &&
2222 !dm_init_request_based_queue(md)) {
2223 DMWARN("Cannot initialize queue for request-based mapped device");
2230 static struct mapped_device *dm_find_md(dev_t dev)
2232 struct mapped_device *md;
2233 unsigned minor = MINOR(dev);
2235 if (MAJOR(dev) != _major || minor >= (1 << MINORBITS))
2238 spin_lock(&_minor_lock);
2240 md = idr_find(&_minor_idr, minor);
2241 if (md && (md == MINOR_ALLOCED ||
2242 (MINOR(disk_devt(dm_disk(md))) != minor) ||
2243 dm_deleting_md(md) ||
2244 test_bit(DMF_FREEING, &md->flags))) {
2250 spin_unlock(&_minor_lock);
2255 struct mapped_device *dm_get_md(dev_t dev)
2257 struct mapped_device *md = dm_find_md(dev);
2264 EXPORT_SYMBOL_GPL(dm_get_md);
2266 void *dm_get_mdptr(struct mapped_device *md)
2268 return md->interface_ptr;
2271 void dm_set_mdptr(struct mapped_device *md, void *ptr)
2273 md->interface_ptr = ptr;
2276 void dm_get(struct mapped_device *md)
2278 atomic_inc(&md->holders);
2279 BUG_ON(test_bit(DMF_FREEING, &md->flags));
2282 const char *dm_device_name(struct mapped_device *md)
2286 EXPORT_SYMBOL_GPL(dm_device_name);
2288 static void __dm_destroy(struct mapped_device *md, bool wait)
2290 struct dm_table *map;
2294 spin_lock(&_minor_lock);
2295 map = dm_get_live_table(md);
2296 idr_replace(&_minor_idr, MINOR_ALLOCED, MINOR(disk_devt(dm_disk(md))));
2297 set_bit(DMF_FREEING, &md->flags);
2298 spin_unlock(&_minor_lock);
2300 if (!dm_suspended_md(md)) {
2301 dm_table_presuspend_targets(map);
2302 dm_table_postsuspend_targets(map);
2306 * Rare, but there may be I/O requests still going to complete,
2307 * for example. Wait for all references to disappear.
2308 * No one should increment the reference count of the mapped_device,
2309 * after the mapped_device state becomes DMF_FREEING.
2312 while (atomic_read(&md->holders))
2314 else if (atomic_read(&md->holders))
2315 DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)",
2316 dm_device_name(md), atomic_read(&md->holders));
2320 dm_table_destroy(__unbind(md));
2324 void dm_destroy(struct mapped_device *md)
2326 __dm_destroy(md, true);
2329 void dm_destroy_immediate(struct mapped_device *md)
2331 __dm_destroy(md, false);
2334 void dm_put(struct mapped_device *md)
2336 atomic_dec(&md->holders);
2338 EXPORT_SYMBOL_GPL(dm_put);
2340 static int dm_wait_for_completion(struct mapped_device *md, int interruptible)
2343 DECLARE_WAITQUEUE(wait, current);
2345 add_wait_queue(&md->wait, &wait);
2348 set_current_state(interruptible);
2350 if (!md_in_flight(md))
2353 if (interruptible == TASK_INTERRUPTIBLE &&
2354 signal_pending(current)) {
2361 set_current_state(TASK_RUNNING);
2363 remove_wait_queue(&md->wait, &wait);
2369 * Process the deferred bios
2371 static void dm_wq_work(struct work_struct *work)
2373 struct mapped_device *md = container_of(work, struct mapped_device,
2377 down_read(&md->io_lock);
2379 while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
2380 spin_lock_irq(&md->deferred_lock);
2381 c = bio_list_pop(&md->deferred);
2382 spin_unlock_irq(&md->deferred_lock);
2387 up_read(&md->io_lock);
2389 if (dm_request_based(md))
2390 generic_make_request(c);
2392 __split_and_process_bio(md, c);
2394 down_read(&md->io_lock);
2397 up_read(&md->io_lock);
2400 static void dm_queue_flush(struct mapped_device *md)
2402 clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2403 smp_mb__after_clear_bit();
2404 queue_work(md->wq, &md->work);
2408 * Swap in a new table, returning the old one for the caller to destroy.
2410 struct dm_table *dm_swap_table(struct mapped_device *md, struct dm_table *table)
2412 struct dm_table *map = ERR_PTR(-EINVAL);
2413 struct queue_limits limits;
2416 mutex_lock(&md->suspend_lock);
2418 /* device must be suspended */
2419 if (!dm_suspended_md(md))
2422 r = dm_calculate_queue_limits(table, &limits);
2428 map = __bind(md, table, &limits);
2431 mutex_unlock(&md->suspend_lock);
2436 * Functions to lock and unlock any filesystem running on the
2439 static int lock_fs(struct mapped_device *md)
2443 WARN_ON(md->frozen_sb);
2445 md->frozen_sb = freeze_bdev(md->bdev);
2446 if (IS_ERR(md->frozen_sb)) {
2447 r = PTR_ERR(md->frozen_sb);
2448 md->frozen_sb = NULL;
2452 set_bit(DMF_FROZEN, &md->flags);
2457 static void unlock_fs(struct mapped_device *md)
2459 if (!test_bit(DMF_FROZEN, &md->flags))
2462 thaw_bdev(md->bdev, md->frozen_sb);
2463 md->frozen_sb = NULL;
2464 clear_bit(DMF_FROZEN, &md->flags);
2468 * We need to be able to change a mapping table under a mounted
2469 * filesystem. For example we might want to move some data in
2470 * the background. Before the table can be swapped with
2471 * dm_bind_table, dm_suspend must be called to flush any in
2472 * flight bios and ensure that any further io gets deferred.
2475 * Suspend mechanism in request-based dm.
2477 * 1. Flush all I/Os by lock_fs() if needed.
2478 * 2. Stop dispatching any I/O by stopping the request_queue.
2479 * 3. Wait for all in-flight I/Os to be completed or requeued.
2481 * To abort suspend, start the request_queue.
2483 int dm_suspend(struct mapped_device *md, unsigned suspend_flags)
2485 struct dm_table *map = NULL;
2487 int do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG ? 1 : 0;
2488 int noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG ? 1 : 0;
2490 mutex_lock(&md->suspend_lock);
2492 if (dm_suspended_md(md)) {
2497 map = dm_get_live_table(md);
2500 * DMF_NOFLUSH_SUSPENDING must be set before presuspend.
2501 * This flag is cleared before dm_suspend returns.
2504 set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2506 /* This does not get reverted if there's an error later. */
2507 dm_table_presuspend_targets(map);
2510 * Flush I/O to the device.
2511 * Any I/O submitted after lock_fs() may not be flushed.
2512 * noflush takes precedence over do_lockfs.
2513 * (lock_fs() flushes I/Os and waits for them to complete.)
2515 if (!noflush && do_lockfs) {
2522 * Here we must make sure that no processes are submitting requests
2523 * to target drivers i.e. no one may be executing
2524 * __split_and_process_bio. This is called from dm_request and
2527 * To get all processes out of __split_and_process_bio in dm_request,
2528 * we take the write lock. To prevent any process from reentering
2529 * __split_and_process_bio from dm_request and quiesce the thread
2530 * (dm_wq_work), we set BMF_BLOCK_IO_FOR_SUSPEND and call
2531 * flush_workqueue(md->wq).
2533 down_write(&md->io_lock);
2534 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2535 up_write(&md->io_lock);
2538 * Stop md->queue before flushing md->wq in case request-based
2539 * dm defers requests to md->wq from md->queue.
2541 if (dm_request_based(md))
2542 stop_queue(md->queue);
2544 flush_workqueue(md->wq);
2547 * At this point no more requests are entering target request routines.
2548 * We call dm_wait_for_completion to wait for all existing requests
2551 r = dm_wait_for_completion(md, TASK_INTERRUPTIBLE);
2553 down_write(&md->io_lock);
2555 clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2556 up_write(&md->io_lock);
2558 /* were we interrupted ? */
2562 if (dm_request_based(md))
2563 start_queue(md->queue);
2566 goto out; /* pushback list is already flushed, so skip flush */
2570 * If dm_wait_for_completion returned 0, the device is completely
2571 * quiescent now. There is no request-processing activity. All new
2572 * requests are being added to md->deferred list.
2575 set_bit(DMF_SUSPENDED, &md->flags);
2577 dm_table_postsuspend_targets(map);
2583 mutex_unlock(&md->suspend_lock);
2587 int dm_resume(struct mapped_device *md)
2590 struct dm_table *map = NULL;
2592 mutex_lock(&md->suspend_lock);
2593 if (!dm_suspended_md(md))
2596 map = dm_get_live_table(md);
2597 if (!map || !dm_table_get_size(map))
2600 r = dm_table_resume_targets(map);
2607 * Flushing deferred I/Os must be done after targets are resumed
2608 * so that mapping of targets can work correctly.
2609 * Request-based dm is queueing the deferred I/Os in its request_queue.
2611 if (dm_request_based(md))
2612 start_queue(md->queue);
2616 clear_bit(DMF_SUSPENDED, &md->flags);
2621 mutex_unlock(&md->suspend_lock);
2626 /*-----------------------------------------------------------------
2627 * Event notification.
2628 *---------------------------------------------------------------*/
2629 int dm_kobject_uevent(struct mapped_device *md, enum kobject_action action,
2632 char udev_cookie[DM_COOKIE_LENGTH];
2633 char *envp[] = { udev_cookie, NULL };
2636 return kobject_uevent(&disk_to_dev(md->disk)->kobj, action);
2638 snprintf(udev_cookie, DM_COOKIE_LENGTH, "%s=%u",
2639 DM_COOKIE_ENV_VAR_NAME, cookie);
2640 return kobject_uevent_env(&disk_to_dev(md->disk)->kobj,
2645 uint32_t dm_next_uevent_seq(struct mapped_device *md)
2647 return atomic_add_return(1, &md->uevent_seq);
2650 uint32_t dm_get_event_nr(struct mapped_device *md)
2652 return atomic_read(&md->event_nr);
2655 int dm_wait_event(struct mapped_device *md, int event_nr)
2657 return wait_event_interruptible(md->eventq,
2658 (event_nr != atomic_read(&md->event_nr)));
2661 void dm_uevent_add(struct mapped_device *md, struct list_head *elist)
2663 unsigned long flags;
2665 spin_lock_irqsave(&md->uevent_lock, flags);
2666 list_add(elist, &md->uevent_list);
2667 spin_unlock_irqrestore(&md->uevent_lock, flags);
2671 * The gendisk is only valid as long as you have a reference
2674 struct gendisk *dm_disk(struct mapped_device *md)
2679 struct kobject *dm_kobject(struct mapped_device *md)
2685 * struct mapped_device should not be exported outside of dm.c
2686 * so use this check to verify that kobj is part of md structure
2688 struct mapped_device *dm_get_from_kobject(struct kobject *kobj)
2690 struct mapped_device *md;
2692 md = container_of(kobj, struct mapped_device, kobj);
2693 if (&md->kobj != kobj)
2696 if (test_bit(DMF_FREEING, &md->flags) ||
2704 int dm_suspended_md(struct mapped_device *md)
2706 return test_bit(DMF_SUSPENDED, &md->flags);
2709 int dm_suspended(struct dm_target *ti)
2711 return dm_suspended_md(dm_table_get_md(ti->table));
2713 EXPORT_SYMBOL_GPL(dm_suspended);
2715 int dm_noflush_suspending(struct dm_target *ti)
2717 return __noflush_suspending(dm_table_get_md(ti->table));
2719 EXPORT_SYMBOL_GPL(dm_noflush_suspending);
2721 struct dm_md_mempools *dm_alloc_md_mempools(unsigned type, unsigned integrity)
2723 struct dm_md_mempools *pools = kmalloc(sizeof(*pools), GFP_KERNEL);
2724 unsigned int pool_size = (type == DM_TYPE_BIO_BASED) ? 16 : MIN_IOS;
2729 pools->io_pool = (type == DM_TYPE_BIO_BASED) ?
2730 mempool_create_slab_pool(MIN_IOS, _io_cache) :
2731 mempool_create_slab_pool(MIN_IOS, _rq_bio_info_cache);
2732 if (!pools->io_pool)
2733 goto free_pools_and_out;
2735 pools->tio_pool = (type == DM_TYPE_BIO_BASED) ?
2736 mempool_create_slab_pool(MIN_IOS, _tio_cache) :
2737 mempool_create_slab_pool(MIN_IOS, _rq_tio_cache);
2738 if (!pools->tio_pool)
2739 goto free_io_pool_and_out;
2741 pools->bs = bioset_create(pool_size, 0);
2743 goto free_tio_pool_and_out;
2745 if (integrity && bioset_integrity_create(pools->bs, pool_size))
2746 goto free_bioset_and_out;
2750 free_bioset_and_out:
2751 bioset_free(pools->bs);
2753 free_tio_pool_and_out:
2754 mempool_destroy(pools->tio_pool);
2756 free_io_pool_and_out:
2757 mempool_destroy(pools->io_pool);
2765 void dm_free_md_mempools(struct dm_md_mempools *pools)
2771 mempool_destroy(pools->io_pool);
2773 if (pools->tio_pool)
2774 mempool_destroy(pools->tio_pool);
2777 bioset_free(pools->bs);
2782 static const struct block_device_operations dm_blk_dops = {
2783 .open = dm_blk_open,
2784 .release = dm_blk_close,
2785 .ioctl = dm_blk_ioctl,
2786 .getgeo = dm_blk_getgeo,
2787 .owner = THIS_MODULE
2790 EXPORT_SYMBOL(dm_get_mapinfo);
2795 module_init(dm_init);
2796 module_exit(dm_exit);
2798 module_param(major, uint, 0);
2799 MODULE_PARM_DESC(major, "The major number of the device mapper");
2800 MODULE_DESCRIPTION(DM_NAME " driver");
2801 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
2802 MODULE_LICENSE("GPL");