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;
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 - the bio clones we allocate are embedded in these
81 * We allocate these with bio_alloc_bioset, using the front_pad parameter when
82 * the bioset is created - this means the bio has to come at the end of the
85 struct dm_rq_clone_bio_info {
87 struct dm_rq_target_io *tio;
91 union map_info *dm_get_mapinfo(struct bio *bio)
93 if (bio && bio->bi_private)
94 return &((struct dm_target_io *)bio->bi_private)->info;
98 union map_info *dm_get_rq_mapinfo(struct request *rq)
100 if (rq && rq->end_io_data)
101 return &((struct dm_rq_target_io *)rq->end_io_data)->info;
104 EXPORT_SYMBOL_GPL(dm_get_rq_mapinfo);
106 #define MINOR_ALLOCED ((void *)-1)
109 * Bits for the md->flags field.
111 #define DMF_BLOCK_IO_FOR_SUSPEND 0
112 #define DMF_SUSPENDED 1
114 #define DMF_FREEING 3
115 #define DMF_DELETING 4
116 #define DMF_NOFLUSH_SUSPENDING 5
117 #define DMF_MERGE_IS_OPTIONAL 6
120 * Work processed by per-device workqueue.
122 struct mapped_device {
123 struct rw_semaphore io_lock;
124 struct mutex suspend_lock;
131 struct request_queue *queue;
133 /* Protect queue and type against concurrent access. */
134 struct mutex type_lock;
136 struct target_type *immutable_target_type;
138 struct gendisk *disk;
144 * A list of ios that arrived while we were suspended.
147 wait_queue_head_t wait;
148 struct work_struct work;
149 struct bio_list deferred;
150 spinlock_t deferred_lock;
153 * Processing queue (flush)
155 struct workqueue_struct *wq;
158 * The current mapping.
160 struct dm_table *map;
163 * io objects are allocated from here.
173 wait_queue_head_t eventq;
175 struct list_head uevent_list;
176 spinlock_t uevent_lock; /* Protect access to uevent_list */
179 * freeze/thaw support require holding onto a super block
181 struct super_block *frozen_sb;
182 struct block_device *bdev;
184 /* forced geometry settings */
185 struct hd_geometry geometry;
190 /* zero-length flush that will be cloned and submitted to targets */
191 struct bio flush_bio;
195 * For mempools pre-allocation at the table loading time.
197 struct dm_md_mempools {
203 static struct kmem_cache *_io_cache;
204 static struct kmem_cache *_rq_tio_cache;
206 static int __init local_init(void)
210 /* allocate a slab for the dm_ios */
211 _io_cache = KMEM_CACHE(dm_io, 0);
215 _rq_tio_cache = KMEM_CACHE(dm_rq_target_io, 0);
217 goto out_free_io_cache;
219 r = dm_uevent_init();
221 goto out_free_rq_tio_cache;
224 r = register_blkdev(_major, _name);
226 goto out_uevent_exit;
235 out_free_rq_tio_cache:
236 kmem_cache_destroy(_rq_tio_cache);
238 kmem_cache_destroy(_io_cache);
243 static void local_exit(void)
245 kmem_cache_destroy(_rq_tio_cache);
246 kmem_cache_destroy(_io_cache);
247 unregister_blkdev(_major, _name);
252 DMINFO("cleaned up");
255 static int (*_inits[])(void) __initdata = {
265 static void (*_exits[])(void) = {
275 static int __init dm_init(void)
277 const int count = ARRAY_SIZE(_inits);
281 for (i = 0; i < count; i++) {
296 static void __exit dm_exit(void)
298 int i = ARRAY_SIZE(_exits);
304 * Should be empty by this point.
306 idr_destroy(&_minor_idr);
310 * Block device functions
312 int dm_deleting_md(struct mapped_device *md)
314 return test_bit(DMF_DELETING, &md->flags);
317 static int dm_blk_open(struct block_device *bdev, fmode_t mode)
319 struct mapped_device *md;
321 spin_lock(&_minor_lock);
323 md = bdev->bd_disk->private_data;
327 if (test_bit(DMF_FREEING, &md->flags) ||
328 dm_deleting_md(md)) {
334 atomic_inc(&md->open_count);
337 spin_unlock(&_minor_lock);
339 return md ? 0 : -ENXIO;
342 static int dm_blk_close(struct gendisk *disk, fmode_t mode)
344 struct mapped_device *md = disk->private_data;
346 spin_lock(&_minor_lock);
348 atomic_dec(&md->open_count);
351 spin_unlock(&_minor_lock);
356 int dm_open_count(struct mapped_device *md)
358 return atomic_read(&md->open_count);
362 * Guarantees nothing is using the device before it's deleted.
364 int dm_lock_for_deletion(struct mapped_device *md)
368 spin_lock(&_minor_lock);
370 if (dm_open_count(md))
373 set_bit(DMF_DELETING, &md->flags);
375 spin_unlock(&_minor_lock);
380 static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo)
382 struct mapped_device *md = bdev->bd_disk->private_data;
384 return dm_get_geometry(md, geo);
387 static int dm_blk_ioctl(struct block_device *bdev, fmode_t mode,
388 unsigned int cmd, unsigned long arg)
390 struct mapped_device *md = bdev->bd_disk->private_data;
391 struct dm_table *map = dm_get_live_table(md);
392 struct dm_target *tgt;
395 if (!map || !dm_table_get_size(map))
398 /* We only support devices that have a single target */
399 if (dm_table_get_num_targets(map) != 1)
402 tgt = dm_table_get_target(map, 0);
404 if (dm_suspended_md(md)) {
409 if (tgt->type->ioctl)
410 r = tgt->type->ioctl(tgt, cmd, arg);
418 static struct dm_io *alloc_io(struct mapped_device *md)
420 return mempool_alloc(md->io_pool, GFP_NOIO);
423 static void free_io(struct mapped_device *md, struct dm_io *io)
425 mempool_free(io, md->io_pool);
428 static void free_tio(struct mapped_device *md, struct dm_target_io *tio)
430 bio_put(&tio->clone);
433 static struct dm_rq_target_io *alloc_rq_tio(struct mapped_device *md,
436 return mempool_alloc(md->io_pool, gfp_mask);
439 static void free_rq_tio(struct dm_rq_target_io *tio)
441 mempool_free(tio, tio->md->io_pool);
444 static int md_in_flight(struct mapped_device *md)
446 return atomic_read(&md->pending[READ]) +
447 atomic_read(&md->pending[WRITE]);
450 static void start_io_acct(struct dm_io *io)
452 struct mapped_device *md = io->md;
454 int rw = bio_data_dir(io->bio);
456 io->start_time = jiffies;
458 cpu = part_stat_lock();
459 part_round_stats(cpu, &dm_disk(md)->part0);
461 atomic_set(&dm_disk(md)->part0.in_flight[rw],
462 atomic_inc_return(&md->pending[rw]));
465 static void end_io_acct(struct dm_io *io)
467 struct mapped_device *md = io->md;
468 struct bio *bio = io->bio;
469 unsigned long duration = jiffies - io->start_time;
471 int rw = bio_data_dir(bio);
473 cpu = part_stat_lock();
474 part_round_stats(cpu, &dm_disk(md)->part0);
475 part_stat_add(cpu, &dm_disk(md)->part0, ticks[rw], duration);
479 * After this is decremented the bio must not be touched if it is
482 pending = atomic_dec_return(&md->pending[rw]);
483 atomic_set(&dm_disk(md)->part0.in_flight[rw], pending);
484 pending += atomic_read(&md->pending[rw^0x1]);
486 /* nudge anyone waiting on suspend queue */
492 * Add the bio to the list of deferred io.
494 static void queue_io(struct mapped_device *md, struct bio *bio)
498 spin_lock_irqsave(&md->deferred_lock, flags);
499 bio_list_add(&md->deferred, bio);
500 spin_unlock_irqrestore(&md->deferred_lock, flags);
501 queue_work(md->wq, &md->work);
505 * Everyone (including functions in this file), should use this
506 * function to access the md->map field, and make sure they call
507 * dm_table_put() when finished.
509 struct dm_table *dm_get_live_table(struct mapped_device *md)
514 read_lock_irqsave(&md->map_lock, flags);
518 read_unlock_irqrestore(&md->map_lock, flags);
524 * Get the geometry associated with a dm device
526 int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo)
534 * Set the geometry of a device.
536 int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo)
538 sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors;
540 if (geo->start > sz) {
541 DMWARN("Start sector is beyond the geometry limits.");
550 /*-----------------------------------------------------------------
552 * A more elegant soln is in the works that uses the queue
553 * merge fn, unfortunately there are a couple of changes to
554 * the block layer that I want to make for this. So in the
555 * interests of getting something for people to use I give
556 * you this clearly demarcated crap.
557 *---------------------------------------------------------------*/
559 static int __noflush_suspending(struct mapped_device *md)
561 return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
565 * Decrements the number of outstanding ios that a bio has been
566 * cloned into, completing the original io if necc.
568 static void dec_pending(struct dm_io *io, int error)
573 struct mapped_device *md = io->md;
575 /* Push-back supersedes any I/O errors */
576 if (unlikely(error)) {
577 spin_lock_irqsave(&io->endio_lock, flags);
578 if (!(io->error > 0 && __noflush_suspending(md)))
580 spin_unlock_irqrestore(&io->endio_lock, flags);
583 if (atomic_dec_and_test(&io->io_count)) {
584 if (io->error == DM_ENDIO_REQUEUE) {
586 * Target requested pushing back the I/O.
588 spin_lock_irqsave(&md->deferred_lock, flags);
589 if (__noflush_suspending(md))
590 bio_list_add_head(&md->deferred, io->bio);
592 /* noflush suspend was interrupted. */
594 spin_unlock_irqrestore(&md->deferred_lock, flags);
597 io_error = io->error;
602 if (io_error == DM_ENDIO_REQUEUE)
605 if ((bio->bi_rw & REQ_FLUSH) && bio->bi_size) {
607 * Preflush done for flush with data, reissue
610 bio->bi_rw &= ~REQ_FLUSH;
613 /* done with normal IO or empty flush */
614 bio_endio(bio, io_error);
619 static void clone_endio(struct bio *bio, int error)
622 struct dm_target_io *tio = bio->bi_private;
623 struct dm_io *io = tio->io;
624 struct mapped_device *md = tio->io->md;
625 dm_endio_fn endio = tio->ti->type->end_io;
627 if (!bio_flagged(bio, BIO_UPTODATE) && !error)
631 r = endio(tio->ti, bio, error);
632 if (r < 0 || r == DM_ENDIO_REQUEUE)
634 * error and requeue request are handled
638 else if (r == DM_ENDIO_INCOMPLETE)
639 /* The target will handle the io */
642 DMWARN("unimplemented target endio return value: %d", r);
648 dec_pending(io, error);
652 * Partial completion handling for request-based dm
654 static void end_clone_bio(struct bio *clone, int error)
656 struct dm_rq_clone_bio_info *info = clone->bi_private;
657 struct dm_rq_target_io *tio = info->tio;
658 struct bio *bio = info->orig;
659 unsigned int nr_bytes = info->orig->bi_size;
665 * An error has already been detected on the request.
666 * Once error occurred, just let clone->end_io() handle
672 * Don't notice the error to the upper layer yet.
673 * The error handling decision is made by the target driver,
674 * when the request is completed.
681 * I/O for the bio successfully completed.
682 * Notice the data completion to the upper layer.
686 * bios are processed from the head of the list.
687 * So the completing bio should always be rq->bio.
688 * If it's not, something wrong is happening.
690 if (tio->orig->bio != bio)
691 DMERR("bio completion is going in the middle of the request");
694 * Update the original request.
695 * Do not use blk_end_request() here, because it may complete
696 * the original request before the clone, and break the ordering.
698 blk_update_request(tio->orig, 0, nr_bytes);
702 * Don't touch any member of the md after calling this function because
703 * the md may be freed in dm_put() at the end of this function.
704 * Or do dm_get() before calling this function and dm_put() later.
706 static void rq_completed(struct mapped_device *md, int rw, int run_queue)
708 atomic_dec(&md->pending[rw]);
710 /* nudge anyone waiting on suspend queue */
711 if (!md_in_flight(md))
715 * Run this off this callpath, as drivers could invoke end_io while
716 * inside their request_fn (and holding the queue lock). Calling
717 * back into ->request_fn() could deadlock attempting to grab the
721 blk_run_queue_async(md->queue);
724 * dm_put() must be at the end of this function. See the comment above
729 static void free_rq_clone(struct request *clone)
731 struct dm_rq_target_io *tio = clone->end_io_data;
733 blk_rq_unprep_clone(clone);
738 * Complete the clone and the original request.
739 * Must be called without queue lock.
741 static void dm_end_request(struct request *clone, int error)
743 int rw = rq_data_dir(clone);
744 struct dm_rq_target_io *tio = clone->end_io_data;
745 struct mapped_device *md = tio->md;
746 struct request *rq = tio->orig;
748 if (rq->cmd_type == REQ_TYPE_BLOCK_PC) {
749 rq->errors = clone->errors;
750 rq->resid_len = clone->resid_len;
754 * We are using the sense buffer of the original
756 * So setting the length of the sense data is enough.
758 rq->sense_len = clone->sense_len;
761 free_rq_clone(clone);
762 blk_end_request_all(rq, error);
763 rq_completed(md, rw, true);
766 static void dm_unprep_request(struct request *rq)
768 struct request *clone = rq->special;
771 rq->cmd_flags &= ~REQ_DONTPREP;
773 free_rq_clone(clone);
777 * Requeue the original request of a clone.
779 void dm_requeue_unmapped_request(struct request *clone)
781 int rw = rq_data_dir(clone);
782 struct dm_rq_target_io *tio = clone->end_io_data;
783 struct mapped_device *md = tio->md;
784 struct request *rq = tio->orig;
785 struct request_queue *q = rq->q;
788 dm_unprep_request(rq);
790 spin_lock_irqsave(q->queue_lock, flags);
791 blk_requeue_request(q, rq);
792 spin_unlock_irqrestore(q->queue_lock, flags);
794 rq_completed(md, rw, 0);
796 EXPORT_SYMBOL_GPL(dm_requeue_unmapped_request);
798 static void __stop_queue(struct request_queue *q)
803 static void stop_queue(struct request_queue *q)
807 spin_lock_irqsave(q->queue_lock, flags);
809 spin_unlock_irqrestore(q->queue_lock, flags);
812 static void __start_queue(struct request_queue *q)
814 if (blk_queue_stopped(q))
818 static void start_queue(struct request_queue *q)
822 spin_lock_irqsave(q->queue_lock, flags);
824 spin_unlock_irqrestore(q->queue_lock, flags);
827 static void dm_done(struct request *clone, int error, bool mapped)
830 struct dm_rq_target_io *tio = clone->end_io_data;
831 dm_request_endio_fn rq_end_io = NULL;
834 rq_end_io = tio->ti->type->rq_end_io;
836 if (mapped && rq_end_io)
837 r = rq_end_io(tio->ti, clone, error, &tio->info);
841 /* The target wants to complete the I/O */
842 dm_end_request(clone, r);
843 else if (r == DM_ENDIO_INCOMPLETE)
844 /* The target will handle the I/O */
846 else if (r == DM_ENDIO_REQUEUE)
847 /* The target wants to requeue the I/O */
848 dm_requeue_unmapped_request(clone);
850 DMWARN("unimplemented target endio return value: %d", r);
856 * Request completion handler for request-based dm
858 static void dm_softirq_done(struct request *rq)
861 struct request *clone = rq->completion_data;
862 struct dm_rq_target_io *tio = clone->end_io_data;
864 if (rq->cmd_flags & REQ_FAILED)
867 dm_done(clone, tio->error, mapped);
871 * Complete the clone and the original request with the error status
872 * through softirq context.
874 static void dm_complete_request(struct request *clone, int error)
876 struct dm_rq_target_io *tio = clone->end_io_data;
877 struct request *rq = tio->orig;
880 rq->completion_data = clone;
881 blk_complete_request(rq);
885 * Complete the not-mapped clone and the original request with the error status
886 * through softirq context.
887 * Target's rq_end_io() function isn't called.
888 * This may be used when the target's map_rq() function fails.
890 void dm_kill_unmapped_request(struct request *clone, int error)
892 struct dm_rq_target_io *tio = clone->end_io_data;
893 struct request *rq = tio->orig;
895 rq->cmd_flags |= REQ_FAILED;
896 dm_complete_request(clone, error);
898 EXPORT_SYMBOL_GPL(dm_kill_unmapped_request);
901 * Called with the queue lock held
903 static void end_clone_request(struct request *clone, int error)
906 * For just cleaning up the information of the queue in which
907 * the clone was dispatched.
908 * The clone is *NOT* freed actually here because it is alloced from
909 * dm own mempool and REQ_ALLOCED isn't set in clone->cmd_flags.
911 __blk_put_request(clone->q, clone);
914 * Actual request completion is done in a softirq context which doesn't
915 * hold the queue lock. Otherwise, deadlock could occur because:
916 * - another request may be submitted by the upper level driver
917 * of the stacking during the completion
918 * - the submission which requires queue lock may be done
921 dm_complete_request(clone, error);
925 * Return maximum size of I/O possible at the supplied sector up to the current
928 static sector_t max_io_len_target_boundary(sector_t sector, struct dm_target *ti)
930 sector_t target_offset = dm_target_offset(ti, sector);
932 return ti->len - target_offset;
935 static sector_t max_io_len(sector_t sector, struct dm_target *ti)
937 sector_t len = max_io_len_target_boundary(sector, ti);
938 sector_t offset, max_len;
941 * Does the target need to split even further?
943 if (ti->max_io_len) {
944 offset = dm_target_offset(ti, sector);
945 if (unlikely(ti->max_io_len & (ti->max_io_len - 1)))
946 max_len = sector_div(offset, ti->max_io_len);
948 max_len = offset & (ti->max_io_len - 1);
949 max_len = ti->max_io_len - max_len;
958 int dm_set_target_max_io_len(struct dm_target *ti, sector_t len)
960 if (len > UINT_MAX) {
961 DMERR("Specified maximum size of target IO (%llu) exceeds limit (%u)",
962 (unsigned long long)len, UINT_MAX);
963 ti->error = "Maximum size of target IO is too large";
967 ti->max_io_len = (uint32_t) len;
971 EXPORT_SYMBOL_GPL(dm_set_target_max_io_len);
973 static void __map_bio(struct dm_target_io *tio)
977 struct mapped_device *md;
978 struct bio *clone = &tio->clone;
979 struct dm_target *ti = tio->ti;
981 clone->bi_end_io = clone_endio;
982 clone->bi_private = tio;
985 * Map the clone. If r == 0 we don't need to do
986 * anything, the target has assumed ownership of
989 atomic_inc(&tio->io->io_count);
990 sector = clone->bi_sector;
991 r = ti->type->map(ti, clone);
992 if (r == DM_MAPIO_REMAPPED) {
993 /* the bio has been remapped so dispatch it */
995 trace_block_bio_remap(bdev_get_queue(clone->bi_bdev), clone,
996 tio->io->bio->bi_bdev->bd_dev, sector);
998 generic_make_request(clone);
999 } else if (r < 0 || r == DM_MAPIO_REQUEUE) {
1000 /* error the io and bail out, or requeue it if needed */
1002 dec_pending(tio->io, r);
1005 DMWARN("unimplemented target map return value: %d", r);
1011 struct mapped_device *md;
1012 struct dm_table *map;
1016 sector_t sector_count;
1020 static void bio_setup_sector(struct bio *bio, sector_t sector, sector_t len)
1022 bio->bi_sector = sector;
1023 bio->bi_size = to_bytes(len);
1026 static void bio_setup_bv(struct bio *bio, unsigned short idx, unsigned short bv_count)
1029 bio->bi_vcnt = idx + bv_count;
1030 bio->bi_flags &= ~(1 << BIO_SEG_VALID);
1033 static void clone_bio_integrity(struct bio *bio, struct bio *clone,
1034 unsigned short idx, unsigned len, unsigned offset,
1037 if (!bio_integrity(bio))
1040 bio_integrity_clone(clone, bio, GFP_NOIO);
1043 bio_integrity_trim(clone, bio_sector_offset(bio, idx, offset), len);
1047 * Creates a little bio that just does part of a bvec.
1049 static void clone_split_bio(struct dm_target_io *tio, struct bio *bio,
1050 sector_t sector, unsigned short idx,
1051 unsigned offset, unsigned len)
1053 struct bio *clone = &tio->clone;
1054 struct bio_vec *bv = bio->bi_io_vec + idx;
1056 *clone->bi_io_vec = *bv;
1058 bio_setup_sector(clone, sector, len);
1060 clone->bi_bdev = bio->bi_bdev;
1061 clone->bi_rw = bio->bi_rw;
1063 clone->bi_io_vec->bv_offset = offset;
1064 clone->bi_io_vec->bv_len = clone->bi_size;
1065 clone->bi_flags |= 1 << BIO_CLONED;
1067 clone_bio_integrity(bio, clone, idx, len, offset, 1);
1071 * Creates a bio that consists of range of complete bvecs.
1073 static void clone_bio(struct dm_target_io *tio, struct bio *bio,
1074 sector_t sector, unsigned short idx,
1075 unsigned short bv_count, unsigned len)
1077 struct bio *clone = &tio->clone;
1080 __bio_clone(clone, bio);
1081 bio_setup_sector(clone, sector, len);
1082 bio_setup_bv(clone, idx, bv_count);
1084 if (idx != bio->bi_idx || clone->bi_size < bio->bi_size)
1086 clone_bio_integrity(bio, clone, idx, len, 0, trim);
1089 static struct dm_target_io *alloc_tio(struct clone_info *ci,
1090 struct dm_target *ti, int nr_iovecs,
1091 unsigned target_bio_nr)
1093 struct dm_target_io *tio;
1096 clone = bio_alloc_bioset(GFP_NOIO, nr_iovecs, ci->md->bs);
1097 tio = container_of(clone, struct dm_target_io, clone);
1101 memset(&tio->info, 0, sizeof(tio->info));
1102 tio->target_bio_nr = target_bio_nr;
1107 static void __clone_and_map_simple_bio(struct clone_info *ci,
1108 struct dm_target *ti,
1109 unsigned target_bio_nr, sector_t len)
1111 struct dm_target_io *tio = alloc_tio(ci, ti, ci->bio->bi_max_vecs, target_bio_nr);
1112 struct bio *clone = &tio->clone;
1115 * Discard requests require the bio's inline iovecs be initialized.
1116 * ci->bio->bi_max_vecs is BIO_INLINE_VECS anyway, for both flush
1117 * and discard, so no need for concern about wasted bvec allocations.
1119 __bio_clone(clone, ci->bio);
1121 bio_setup_sector(clone, ci->sector, len);
1126 static void __send_duplicate_bios(struct clone_info *ci, struct dm_target *ti,
1127 unsigned num_bios, sector_t len)
1129 unsigned target_bio_nr;
1131 for (target_bio_nr = 0; target_bio_nr < num_bios; target_bio_nr++)
1132 __clone_and_map_simple_bio(ci, ti, target_bio_nr, len);
1135 static int __send_empty_flush(struct clone_info *ci)
1137 unsigned target_nr = 0;
1138 struct dm_target *ti;
1140 BUG_ON(bio_has_data(ci->bio));
1141 while ((ti = dm_table_get_target(ci->map, target_nr++)))
1142 __send_duplicate_bios(ci, ti, ti->num_flush_bios, 0);
1147 static void __clone_and_map_data_bio(struct clone_info *ci, struct dm_target *ti,
1148 sector_t sector, int nr_iovecs,
1149 unsigned short idx, unsigned short bv_count,
1150 unsigned offset, unsigned len,
1151 unsigned split_bvec)
1153 struct bio *bio = ci->bio;
1154 struct dm_target_io *tio;
1156 tio = alloc_tio(ci, ti, nr_iovecs, 0);
1159 clone_split_bio(tio, bio, sector, idx, offset, len);
1161 clone_bio(tio, bio, sector, idx, bv_count, len);
1166 typedef unsigned (*get_num_bios_fn)(struct dm_target *ti);
1168 static unsigned get_num_discard_bios(struct dm_target *ti)
1170 return ti->num_discard_bios;
1173 static unsigned get_num_write_same_bios(struct dm_target *ti)
1175 return ti->num_write_same_bios;
1178 typedef bool (*is_split_required_fn)(struct dm_target *ti);
1180 static bool is_split_required_for_discard(struct dm_target *ti)
1182 return ti->split_discard_bios;
1185 static int __send_changing_extent_only(struct clone_info *ci,
1186 get_num_bios_fn get_num_bios,
1187 is_split_required_fn is_split_required)
1189 struct dm_target *ti;
1194 ti = dm_table_find_target(ci->map, ci->sector);
1195 if (!dm_target_is_valid(ti))
1199 * Even though the device advertised support for this type of
1200 * request, that does not mean every target supports it, and
1201 * reconfiguration might also have changed that since the
1202 * check was performed.
1204 num_bios = get_num_bios ? get_num_bios(ti) : 0;
1208 if (is_split_required && !is_split_required(ti))
1209 len = min(ci->sector_count, max_io_len_target_boundary(ci->sector, ti));
1211 len = min(ci->sector_count, max_io_len(ci->sector, ti));
1213 __send_duplicate_bios(ci, ti, num_bios, len);
1216 } while (ci->sector_count -= len);
1221 static int __send_discard(struct clone_info *ci)
1223 return __send_changing_extent_only(ci, get_num_discard_bios,
1224 is_split_required_for_discard);
1227 static int __send_write_same(struct clone_info *ci)
1229 return __send_changing_extent_only(ci, get_num_write_same_bios, NULL);
1233 * Find maximum number of sectors / bvecs we can process with a single bio.
1235 static sector_t __len_within_target(struct clone_info *ci, sector_t max, int *idx)
1237 struct bio *bio = ci->bio;
1238 sector_t bv_len, total_len = 0;
1240 for (*idx = ci->idx; max && (*idx < bio->bi_vcnt); (*idx)++) {
1241 bv_len = to_sector(bio->bi_io_vec[*idx].bv_len);
1247 total_len += bv_len;
1253 static int __split_bvec_across_targets(struct clone_info *ci,
1254 struct dm_target *ti, sector_t max)
1256 struct bio *bio = ci->bio;
1257 struct bio_vec *bv = bio->bi_io_vec + ci->idx;
1258 sector_t remaining = to_sector(bv->bv_len);
1259 unsigned offset = 0;
1264 ti = dm_table_find_target(ci->map, ci->sector);
1265 if (!dm_target_is_valid(ti))
1268 max = max_io_len(ci->sector, ti);
1271 len = min(remaining, max);
1273 __clone_and_map_data_bio(ci, ti, ci->sector, 1, ci->idx, 0,
1274 bv->bv_offset + offset, len, 1);
1277 ci->sector_count -= len;
1278 offset += to_bytes(len);
1279 } while (remaining -= len);
1287 * Select the correct strategy for processing a non-flush bio.
1289 static int __split_and_process_non_flush(struct clone_info *ci)
1291 struct bio *bio = ci->bio;
1292 struct dm_target *ti;
1296 if (unlikely(bio->bi_rw & REQ_DISCARD))
1297 return __send_discard(ci);
1298 else if (unlikely(bio->bi_rw & REQ_WRITE_SAME))
1299 return __send_write_same(ci);
1301 ti = dm_table_find_target(ci->map, ci->sector);
1302 if (!dm_target_is_valid(ti))
1305 max = max_io_len(ci->sector, ti);
1308 * Optimise for the simple case where we can do all of
1309 * the remaining io with a single clone.
1311 if (ci->sector_count <= max) {
1312 __clone_and_map_data_bio(ci, ti, ci->sector, bio->bi_max_vecs,
1313 ci->idx, bio->bi_vcnt - ci->idx, 0,
1314 ci->sector_count, 0);
1315 ci->sector_count = 0;
1320 * There are some bvecs that don't span targets.
1321 * Do as many of these as possible.
1323 if (to_sector(bio->bi_io_vec[ci->idx].bv_len) <= max) {
1324 len = __len_within_target(ci, max, &idx);
1326 __clone_and_map_data_bio(ci, ti, ci->sector, bio->bi_max_vecs,
1327 ci->idx, idx - ci->idx, 0, len, 0);
1330 ci->sector_count -= len;
1337 * Handle a bvec that must be split between two or more targets.
1339 return __split_bvec_across_targets(ci, ti, max);
1343 * Entry point to split a bio into clones and submit them to the targets.
1345 static void __split_and_process_bio(struct mapped_device *md, struct bio *bio)
1347 struct clone_info ci;
1350 ci.map = dm_get_live_table(md);
1351 if (unlikely(!ci.map)) {
1357 ci.io = alloc_io(md);
1359 atomic_set(&ci.io->io_count, 1);
1362 spin_lock_init(&ci.io->endio_lock);
1363 ci.sector = bio->bi_sector;
1364 ci.idx = bio->bi_idx;
1366 start_io_acct(ci.io);
1368 if (bio->bi_rw & REQ_FLUSH) {
1369 ci.bio = &ci.md->flush_bio;
1370 ci.sector_count = 0;
1371 error = __send_empty_flush(&ci);
1372 /* dec_pending submits any data associated with flush */
1375 ci.sector_count = bio_sectors(bio);
1376 while (ci.sector_count && !error)
1377 error = __split_and_process_non_flush(&ci);
1380 /* drop the extra reference count */
1381 dec_pending(ci.io, error);
1382 dm_table_put(ci.map);
1384 /*-----------------------------------------------------------------
1386 *---------------------------------------------------------------*/
1388 static int dm_merge_bvec(struct request_queue *q,
1389 struct bvec_merge_data *bvm,
1390 struct bio_vec *biovec)
1392 struct mapped_device *md = q->queuedata;
1393 struct dm_table *map = dm_get_live_table(md);
1394 struct dm_target *ti;
1395 sector_t max_sectors;
1401 ti = dm_table_find_target(map, bvm->bi_sector);
1402 if (!dm_target_is_valid(ti))
1406 * Find maximum amount of I/O that won't need splitting
1408 max_sectors = min(max_io_len(bvm->bi_sector, ti),
1409 (sector_t) BIO_MAX_SECTORS);
1410 max_size = (max_sectors << SECTOR_SHIFT) - bvm->bi_size;
1415 * merge_bvec_fn() returns number of bytes
1416 * it can accept at this offset
1417 * max is precomputed maximal io size
1419 if (max_size && ti->type->merge)
1420 max_size = ti->type->merge(ti, bvm, biovec, max_size);
1422 * If the target doesn't support merge method and some of the devices
1423 * provided their merge_bvec method (we know this by looking at
1424 * queue_max_hw_sectors), then we can't allow bios with multiple vector
1425 * entries. So always set max_size to 0, and the code below allows
1428 else if (queue_max_hw_sectors(q) <= PAGE_SIZE >> 9)
1437 * Always allow an entire first page
1439 if (max_size <= biovec->bv_len && !(bvm->bi_size >> SECTOR_SHIFT))
1440 max_size = biovec->bv_len;
1446 * The request function that just remaps the bio built up by
1449 static void _dm_request(struct request_queue *q, struct bio *bio)
1451 int rw = bio_data_dir(bio);
1452 struct mapped_device *md = q->queuedata;
1455 down_read(&md->io_lock);
1457 cpu = part_stat_lock();
1458 part_stat_inc(cpu, &dm_disk(md)->part0, ios[rw]);
1459 part_stat_add(cpu, &dm_disk(md)->part0, sectors[rw], bio_sectors(bio));
1462 /* if we're suspended, we have to queue this io for later */
1463 if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))) {
1464 up_read(&md->io_lock);
1466 if (bio_rw(bio) != READA)
1473 __split_and_process_bio(md, bio);
1474 up_read(&md->io_lock);
1478 static int dm_request_based(struct mapped_device *md)
1480 return blk_queue_stackable(md->queue);
1483 static void dm_request(struct request_queue *q, struct bio *bio)
1485 struct mapped_device *md = q->queuedata;
1487 if (dm_request_based(md))
1488 blk_queue_bio(q, bio);
1490 _dm_request(q, bio);
1493 void dm_dispatch_request(struct request *rq)
1497 if (blk_queue_io_stat(rq->q))
1498 rq->cmd_flags |= REQ_IO_STAT;
1500 rq->start_time = jiffies;
1501 r = blk_insert_cloned_request(rq->q, rq);
1503 dm_complete_request(rq, r);
1505 EXPORT_SYMBOL_GPL(dm_dispatch_request);
1507 static int dm_rq_bio_constructor(struct bio *bio, struct bio *bio_orig,
1510 struct dm_rq_target_io *tio = data;
1511 struct dm_rq_clone_bio_info *info =
1512 container_of(bio, struct dm_rq_clone_bio_info, clone);
1514 info->orig = bio_orig;
1516 bio->bi_end_io = end_clone_bio;
1517 bio->bi_private = info;
1522 static int setup_clone(struct request *clone, struct request *rq,
1523 struct dm_rq_target_io *tio)
1527 r = blk_rq_prep_clone(clone, rq, tio->md->bs, GFP_ATOMIC,
1528 dm_rq_bio_constructor, tio);
1532 clone->cmd = rq->cmd;
1533 clone->cmd_len = rq->cmd_len;
1534 clone->sense = rq->sense;
1535 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(rq->special)) {
1577 DMWARN("Already has something in rq->special.");
1578 return BLKPREP_KILL;
1581 clone = clone_rq(rq, md, GFP_ATOMIC);
1583 return BLKPREP_DEFER;
1585 rq->special = clone;
1586 rq->cmd_flags |= REQ_DONTPREP;
1593 * 0 : the request has been processed (not requeued)
1594 * !0 : the request has been requeued
1596 static int map_request(struct dm_target *ti, struct request *clone,
1597 struct mapped_device *md)
1599 int r, requeued = 0;
1600 struct dm_rq_target_io *tio = clone->end_io_data;
1603 r = ti->type->map_rq(ti, clone, &tio->info);
1605 case DM_MAPIO_SUBMITTED:
1606 /* The target has taken the I/O to submit by itself later */
1608 case DM_MAPIO_REMAPPED:
1609 /* The target has remapped the I/O so dispatch it */
1610 trace_block_rq_remap(clone->q, clone, disk_devt(dm_disk(md)),
1611 blk_rq_pos(tio->orig));
1612 dm_dispatch_request(clone);
1614 case DM_MAPIO_REQUEUE:
1615 /* The target wants to requeue the I/O */
1616 dm_requeue_unmapped_request(clone);
1621 DMWARN("unimplemented target map return value: %d", r);
1625 /* The target wants to complete the I/O */
1626 dm_kill_unmapped_request(clone, r);
1633 static struct request *dm_start_request(struct mapped_device *md, struct request *orig)
1635 struct request *clone;
1637 blk_start_request(orig);
1638 clone = orig->special;
1639 atomic_inc(&md->pending[rq_data_dir(clone)]);
1642 * Hold the md reference here for the in-flight I/O.
1643 * We can't rely on the reference count by device opener,
1644 * because the device may be closed during the request completion
1645 * when all bios are completed.
1646 * See the comment in rq_completed() too.
1654 * q->request_fn for request-based dm.
1655 * Called with the queue lock held.
1657 static void dm_request_fn(struct request_queue *q)
1659 struct mapped_device *md = q->queuedata;
1660 struct dm_table *map = dm_get_live_table(md);
1661 struct dm_target *ti;
1662 struct request *rq, *clone;
1666 * For suspend, check blk_queue_stopped() and increment
1667 * ->pending within a single queue_lock not to increment the
1668 * number of in-flight I/Os after the queue is stopped in
1671 while (!blk_queue_stopped(q)) {
1672 rq = blk_peek_request(q);
1676 /* always use block 0 to find the target for flushes for now */
1678 if (!(rq->cmd_flags & REQ_FLUSH))
1679 pos = blk_rq_pos(rq);
1681 ti = dm_table_find_target(map, pos);
1682 if (!dm_target_is_valid(ti)) {
1684 * Must perform setup, that dm_done() requires,
1685 * before calling dm_kill_unmapped_request
1687 DMERR_LIMIT("request attempted access beyond the end of device");
1688 clone = dm_start_request(md, rq);
1689 dm_kill_unmapped_request(clone, -EIO);
1693 if (ti->type->busy && ti->type->busy(ti))
1696 clone = dm_start_request(md, rq);
1698 spin_unlock(q->queue_lock);
1699 if (map_request(ti, clone, md))
1702 BUG_ON(!irqs_disabled());
1703 spin_lock(q->queue_lock);
1709 BUG_ON(!irqs_disabled());
1710 spin_lock(q->queue_lock);
1713 blk_delay_queue(q, HZ / 10);
1718 int dm_underlying_device_busy(struct request_queue *q)
1720 return blk_lld_busy(q);
1722 EXPORT_SYMBOL_GPL(dm_underlying_device_busy);
1724 static int dm_lld_busy(struct request_queue *q)
1727 struct mapped_device *md = q->queuedata;
1728 struct dm_table *map = dm_get_live_table(md);
1730 if (!map || test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))
1733 r = dm_table_any_busy_target(map);
1740 static int dm_any_congested(void *congested_data, int bdi_bits)
1743 struct mapped_device *md = congested_data;
1744 struct dm_table *map;
1746 if (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
1747 map = dm_get_live_table(md);
1750 * Request-based dm cares about only own queue for
1751 * the query about congestion status of request_queue
1753 if (dm_request_based(md))
1754 r = md->queue->backing_dev_info.state &
1757 r = dm_table_any_congested(map, bdi_bits);
1766 /*-----------------------------------------------------------------
1767 * An IDR is used to keep track of allocated minor numbers.
1768 *---------------------------------------------------------------*/
1769 static void free_minor(int minor)
1771 spin_lock(&_minor_lock);
1772 idr_remove(&_minor_idr, minor);
1773 spin_unlock(&_minor_lock);
1777 * See if the device with a specific minor # is free.
1779 static int specific_minor(int minor)
1783 if (minor >= (1 << MINORBITS))
1786 idr_preload(GFP_KERNEL);
1787 spin_lock(&_minor_lock);
1789 r = idr_alloc(&_minor_idr, MINOR_ALLOCED, minor, minor + 1, GFP_NOWAIT);
1791 spin_unlock(&_minor_lock);
1794 return r == -ENOSPC ? -EBUSY : r;
1798 static int next_free_minor(int *minor)
1802 idr_preload(GFP_KERNEL);
1803 spin_lock(&_minor_lock);
1805 r = idr_alloc(&_minor_idr, MINOR_ALLOCED, 0, 1 << MINORBITS, GFP_NOWAIT);
1807 spin_unlock(&_minor_lock);
1815 static const struct block_device_operations dm_blk_dops;
1817 static void dm_wq_work(struct work_struct *work);
1819 static void dm_init_md_queue(struct mapped_device *md)
1822 * Request-based dm devices cannot be stacked on top of bio-based dm
1823 * devices. The type of this dm device has not been decided yet.
1824 * The type is decided at the first table loading time.
1825 * To prevent problematic device stacking, clear the queue flag
1826 * for request stacking support until then.
1828 * This queue is new, so no concurrency on the queue_flags.
1830 queue_flag_clear_unlocked(QUEUE_FLAG_STACKABLE, md->queue);
1832 md->queue->queuedata = md;
1833 md->queue->backing_dev_info.congested_fn = dm_any_congested;
1834 md->queue->backing_dev_info.congested_data = md;
1835 blk_queue_make_request(md->queue, dm_request);
1836 blk_queue_bounce_limit(md->queue, BLK_BOUNCE_ANY);
1837 blk_queue_merge_bvec(md->queue, dm_merge_bvec);
1841 * Allocate and initialise a blank device with a given minor.
1843 static struct mapped_device *alloc_dev(int minor)
1846 struct mapped_device *md = kzalloc(sizeof(*md), GFP_KERNEL);
1850 DMWARN("unable to allocate device, out of memory.");
1854 if (!try_module_get(THIS_MODULE))
1855 goto bad_module_get;
1857 /* get a minor number for the dev */
1858 if (minor == DM_ANY_MINOR)
1859 r = next_free_minor(&minor);
1861 r = specific_minor(minor);
1865 md->type = DM_TYPE_NONE;
1866 init_rwsem(&md->io_lock);
1867 mutex_init(&md->suspend_lock);
1868 mutex_init(&md->type_lock);
1869 spin_lock_init(&md->deferred_lock);
1870 rwlock_init(&md->map_lock);
1871 atomic_set(&md->holders, 1);
1872 atomic_set(&md->open_count, 0);
1873 atomic_set(&md->event_nr, 0);
1874 atomic_set(&md->uevent_seq, 0);
1875 INIT_LIST_HEAD(&md->uevent_list);
1876 spin_lock_init(&md->uevent_lock);
1878 md->queue = blk_alloc_queue(GFP_KERNEL);
1882 dm_init_md_queue(md);
1884 md->disk = alloc_disk(1);
1888 atomic_set(&md->pending[0], 0);
1889 atomic_set(&md->pending[1], 0);
1890 init_waitqueue_head(&md->wait);
1891 INIT_WORK(&md->work, dm_wq_work);
1892 init_waitqueue_head(&md->eventq);
1894 md->disk->major = _major;
1895 md->disk->first_minor = minor;
1896 md->disk->fops = &dm_blk_dops;
1897 md->disk->queue = md->queue;
1898 md->disk->private_data = md;
1899 sprintf(md->disk->disk_name, "dm-%d", minor);
1901 format_dev_t(md->name, MKDEV(_major, minor));
1903 md->wq = alloc_workqueue("kdmflush",
1904 WQ_NON_REENTRANT | WQ_MEM_RECLAIM, 0);
1908 md->bdev = bdget_disk(md->disk, 0);
1912 bio_init(&md->flush_bio);
1913 md->flush_bio.bi_bdev = md->bdev;
1914 md->flush_bio.bi_rw = WRITE_FLUSH;
1916 /* Populate the mapping, nobody knows we exist yet */
1917 spin_lock(&_minor_lock);
1918 old_md = idr_replace(&_minor_idr, md, minor);
1919 spin_unlock(&_minor_lock);
1921 BUG_ON(old_md != MINOR_ALLOCED);
1926 destroy_workqueue(md->wq);
1928 del_gendisk(md->disk);
1931 blk_cleanup_queue(md->queue);
1935 module_put(THIS_MODULE);
1941 static void unlock_fs(struct mapped_device *md);
1943 static void free_dev(struct mapped_device *md)
1945 int minor = MINOR(disk_devt(md->disk));
1949 destroy_workqueue(md->wq);
1951 mempool_destroy(md->io_pool);
1953 bioset_free(md->bs);
1954 blk_integrity_unregister(md->disk);
1955 del_gendisk(md->disk);
1958 spin_lock(&_minor_lock);
1959 md->disk->private_data = NULL;
1960 spin_unlock(&_minor_lock);
1963 blk_cleanup_queue(md->queue);
1964 module_put(THIS_MODULE);
1968 static void __bind_mempools(struct mapped_device *md, struct dm_table *t)
1970 struct dm_md_mempools *p = dm_table_get_md_mempools(t);
1972 if (md->io_pool && md->bs) {
1973 /* The md already has necessary mempools. */
1974 if (dm_table_get_type(t) == DM_TYPE_BIO_BASED) {
1976 * Reload bioset because front_pad may have changed
1977 * because a different table was loaded.
1979 bioset_free(md->bs);
1982 } else if (dm_table_get_type(t) == DM_TYPE_REQUEST_BASED) {
1984 * There's no need to reload with request-based dm
1985 * because the size of front_pad doesn't change.
1986 * Note for future: If you are to reload bioset,
1987 * prep-ed requests in the queue may refer
1988 * to bio from the old bioset, so you must walk
1989 * through the queue to unprep.
1995 BUG_ON(!p || md->io_pool || md->bs);
1997 md->io_pool = p->io_pool;
2003 /* mempool bind completed, now no need any mempools in the table */
2004 dm_table_free_md_mempools(t);
2008 * Bind a table to the device.
2010 static void event_callback(void *context)
2012 unsigned long flags;
2014 struct mapped_device *md = (struct mapped_device *) context;
2016 spin_lock_irqsave(&md->uevent_lock, flags);
2017 list_splice_init(&md->uevent_list, &uevents);
2018 spin_unlock_irqrestore(&md->uevent_lock, flags);
2020 dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj);
2022 atomic_inc(&md->event_nr);
2023 wake_up(&md->eventq);
2027 * Protected by md->suspend_lock obtained by dm_swap_table().
2029 static void __set_size(struct mapped_device *md, sector_t size)
2031 set_capacity(md->disk, size);
2033 i_size_write(md->bdev->bd_inode, (loff_t)size << SECTOR_SHIFT);
2037 * Return 1 if the queue has a compulsory merge_bvec_fn function.
2039 * If this function returns 0, then the device is either a non-dm
2040 * device without a merge_bvec_fn, or it is a dm device that is
2041 * able to split any bios it receives that are too big.
2043 int dm_queue_merge_is_compulsory(struct request_queue *q)
2045 struct mapped_device *dev_md;
2047 if (!q->merge_bvec_fn)
2050 if (q->make_request_fn == dm_request) {
2051 dev_md = q->queuedata;
2052 if (test_bit(DMF_MERGE_IS_OPTIONAL, &dev_md->flags))
2059 static int dm_device_merge_is_compulsory(struct dm_target *ti,
2060 struct dm_dev *dev, sector_t start,
2061 sector_t len, void *data)
2063 struct block_device *bdev = dev->bdev;
2064 struct request_queue *q = bdev_get_queue(bdev);
2066 return dm_queue_merge_is_compulsory(q);
2070 * Return 1 if it is acceptable to ignore merge_bvec_fn based
2071 * on the properties of the underlying devices.
2073 static int dm_table_merge_is_optional(struct dm_table *table)
2076 struct dm_target *ti;
2078 while (i < dm_table_get_num_targets(table)) {
2079 ti = dm_table_get_target(table, i++);
2081 if (ti->type->iterate_devices &&
2082 ti->type->iterate_devices(ti, dm_device_merge_is_compulsory, NULL))
2090 * Returns old map, which caller must destroy.
2092 static struct dm_table *__bind(struct mapped_device *md, struct dm_table *t,
2093 struct queue_limits *limits)
2095 struct dm_table *old_map;
2096 struct request_queue *q = md->queue;
2098 unsigned long flags;
2099 int merge_is_optional;
2101 size = dm_table_get_size(t);
2104 * Wipe any geometry if the size of the table changed.
2106 if (size != get_capacity(md->disk))
2107 memset(&md->geometry, 0, sizeof(md->geometry));
2109 __set_size(md, size);
2111 dm_table_event_callback(t, event_callback, md);
2114 * The queue hasn't been stopped yet, if the old table type wasn't
2115 * for request-based during suspension. So stop it to prevent
2116 * I/O mapping before resume.
2117 * This must be done before setting the queue restrictions,
2118 * because request-based dm may be run just after the setting.
2120 if (dm_table_request_based(t) && !blk_queue_stopped(q))
2123 __bind_mempools(md, t);
2125 merge_is_optional = dm_table_merge_is_optional(t);
2127 write_lock_irqsave(&md->map_lock, flags);
2130 md->immutable_target_type = dm_table_get_immutable_target_type(t);
2132 dm_table_set_restrictions(t, q, limits);
2133 if (merge_is_optional)
2134 set_bit(DMF_MERGE_IS_OPTIONAL, &md->flags);
2136 clear_bit(DMF_MERGE_IS_OPTIONAL, &md->flags);
2137 write_unlock_irqrestore(&md->map_lock, flags);
2143 * Returns unbound table for the caller to free.
2145 static struct dm_table *__unbind(struct mapped_device *md)
2147 struct dm_table *map = md->map;
2148 unsigned long flags;
2153 dm_table_event_callback(map, NULL, NULL);
2154 write_lock_irqsave(&md->map_lock, flags);
2156 write_unlock_irqrestore(&md->map_lock, flags);
2162 * Constructor for a new device.
2164 int dm_create(int minor, struct mapped_device **result)
2166 struct mapped_device *md;
2168 md = alloc_dev(minor);
2179 * Functions to manage md->type.
2180 * All are required to hold md->type_lock.
2182 void dm_lock_md_type(struct mapped_device *md)
2184 mutex_lock(&md->type_lock);
2187 void dm_unlock_md_type(struct mapped_device *md)
2189 mutex_unlock(&md->type_lock);
2192 void dm_set_md_type(struct mapped_device *md, unsigned type)
2197 unsigned dm_get_md_type(struct mapped_device *md)
2202 struct target_type *dm_get_immutable_target_type(struct mapped_device *md)
2204 return md->immutable_target_type;
2208 * Fully initialize a request-based queue (->elevator, ->request_fn, etc).
2210 static int dm_init_request_based_queue(struct mapped_device *md)
2212 struct request_queue *q = NULL;
2214 if (md->queue->elevator)
2217 /* Fully initialize the queue */
2218 q = blk_init_allocated_queue(md->queue, dm_request_fn, NULL);
2223 dm_init_md_queue(md);
2224 blk_queue_softirq_done(md->queue, dm_softirq_done);
2225 blk_queue_prep_rq(md->queue, dm_prep_fn);
2226 blk_queue_lld_busy(md->queue, dm_lld_busy);
2228 elv_register_queue(md->queue);
2234 * Setup the DM device's queue based on md's type
2236 int dm_setup_md_queue(struct mapped_device *md)
2238 if ((dm_get_md_type(md) == DM_TYPE_REQUEST_BASED) &&
2239 !dm_init_request_based_queue(md)) {
2240 DMWARN("Cannot initialize queue for request-based mapped device");
2247 static struct mapped_device *dm_find_md(dev_t dev)
2249 struct mapped_device *md;
2250 unsigned minor = MINOR(dev);
2252 if (MAJOR(dev) != _major || minor >= (1 << MINORBITS))
2255 spin_lock(&_minor_lock);
2257 md = idr_find(&_minor_idr, minor);
2258 if (md && (md == MINOR_ALLOCED ||
2259 (MINOR(disk_devt(dm_disk(md))) != minor) ||
2260 dm_deleting_md(md) ||
2261 test_bit(DMF_FREEING, &md->flags))) {
2267 spin_unlock(&_minor_lock);
2272 struct mapped_device *dm_get_md(dev_t dev)
2274 struct mapped_device *md = dm_find_md(dev);
2281 EXPORT_SYMBOL_GPL(dm_get_md);
2283 void *dm_get_mdptr(struct mapped_device *md)
2285 return md->interface_ptr;
2288 void dm_set_mdptr(struct mapped_device *md, void *ptr)
2290 md->interface_ptr = ptr;
2293 void dm_get(struct mapped_device *md)
2295 atomic_inc(&md->holders);
2296 BUG_ON(test_bit(DMF_FREEING, &md->flags));
2299 const char *dm_device_name(struct mapped_device *md)
2303 EXPORT_SYMBOL_GPL(dm_device_name);
2305 static void __dm_destroy(struct mapped_device *md, bool wait)
2307 struct dm_table *map;
2311 spin_lock(&_minor_lock);
2312 map = dm_get_live_table(md);
2313 idr_replace(&_minor_idr, MINOR_ALLOCED, MINOR(disk_devt(dm_disk(md))));
2314 set_bit(DMF_FREEING, &md->flags);
2315 spin_unlock(&_minor_lock);
2317 if (!dm_suspended_md(md)) {
2318 dm_table_presuspend_targets(map);
2319 dm_table_postsuspend_targets(map);
2323 * Rare, but there may be I/O requests still going to complete,
2324 * for example. Wait for all references to disappear.
2325 * No one should increment the reference count of the mapped_device,
2326 * after the mapped_device state becomes DMF_FREEING.
2329 while (atomic_read(&md->holders))
2331 else if (atomic_read(&md->holders))
2332 DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)",
2333 dm_device_name(md), atomic_read(&md->holders));
2337 dm_table_destroy(__unbind(md));
2341 void dm_destroy(struct mapped_device *md)
2343 __dm_destroy(md, true);
2346 void dm_destroy_immediate(struct mapped_device *md)
2348 __dm_destroy(md, false);
2351 void dm_put(struct mapped_device *md)
2353 atomic_dec(&md->holders);
2355 EXPORT_SYMBOL_GPL(dm_put);
2357 static int dm_wait_for_completion(struct mapped_device *md, int interruptible)
2360 DECLARE_WAITQUEUE(wait, current);
2362 add_wait_queue(&md->wait, &wait);
2365 set_current_state(interruptible);
2367 if (!md_in_flight(md))
2370 if (interruptible == TASK_INTERRUPTIBLE &&
2371 signal_pending(current)) {
2378 set_current_state(TASK_RUNNING);
2380 remove_wait_queue(&md->wait, &wait);
2386 * Process the deferred bios
2388 static void dm_wq_work(struct work_struct *work)
2390 struct mapped_device *md = container_of(work, struct mapped_device,
2394 down_read(&md->io_lock);
2396 while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
2397 spin_lock_irq(&md->deferred_lock);
2398 c = bio_list_pop(&md->deferred);
2399 spin_unlock_irq(&md->deferred_lock);
2404 up_read(&md->io_lock);
2406 if (dm_request_based(md))
2407 generic_make_request(c);
2409 __split_and_process_bio(md, c);
2411 down_read(&md->io_lock);
2414 up_read(&md->io_lock);
2417 static void dm_queue_flush(struct mapped_device *md)
2419 clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2420 smp_mb__after_clear_bit();
2421 queue_work(md->wq, &md->work);
2425 * Swap in a new table, returning the old one for the caller to destroy.
2427 struct dm_table *dm_swap_table(struct mapped_device *md, struct dm_table *table)
2429 struct dm_table *live_map = NULL, *map = ERR_PTR(-EINVAL);
2430 struct queue_limits limits;
2433 mutex_lock(&md->suspend_lock);
2435 /* device must be suspended */
2436 if (!dm_suspended_md(md))
2440 * If the new table has no data devices, retain the existing limits.
2441 * This helps multipath with queue_if_no_path if all paths disappear,
2442 * then new I/O is queued based on these limits, and then some paths
2445 if (dm_table_has_no_data_devices(table)) {
2446 live_map = dm_get_live_table(md);
2448 limits = md->queue->limits;
2449 dm_table_put(live_map);
2453 r = dm_calculate_queue_limits(table, &limits);
2460 map = __bind(md, table, &limits);
2463 mutex_unlock(&md->suspend_lock);
2468 * Functions to lock and unlock any filesystem running on the
2471 static int lock_fs(struct mapped_device *md)
2475 WARN_ON(md->frozen_sb);
2477 md->frozen_sb = freeze_bdev(md->bdev);
2478 if (IS_ERR(md->frozen_sb)) {
2479 r = PTR_ERR(md->frozen_sb);
2480 md->frozen_sb = NULL;
2484 set_bit(DMF_FROZEN, &md->flags);
2489 static void unlock_fs(struct mapped_device *md)
2491 if (!test_bit(DMF_FROZEN, &md->flags))
2494 thaw_bdev(md->bdev, md->frozen_sb);
2495 md->frozen_sb = NULL;
2496 clear_bit(DMF_FROZEN, &md->flags);
2500 * We need to be able to change a mapping table under a mounted
2501 * filesystem. For example we might want to move some data in
2502 * the background. Before the table can be swapped with
2503 * dm_bind_table, dm_suspend must be called to flush any in
2504 * flight bios and ensure that any further io gets deferred.
2507 * Suspend mechanism in request-based dm.
2509 * 1. Flush all I/Os by lock_fs() if needed.
2510 * 2. Stop dispatching any I/O by stopping the request_queue.
2511 * 3. Wait for all in-flight I/Os to be completed or requeued.
2513 * To abort suspend, start the request_queue.
2515 int dm_suspend(struct mapped_device *md, unsigned suspend_flags)
2517 struct dm_table *map = NULL;
2519 int do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG ? 1 : 0;
2520 int noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG ? 1 : 0;
2522 mutex_lock(&md->suspend_lock);
2524 if (dm_suspended_md(md)) {
2529 map = dm_get_live_table(md);
2532 * DMF_NOFLUSH_SUSPENDING must be set before presuspend.
2533 * This flag is cleared before dm_suspend returns.
2536 set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2538 /* This does not get reverted if there's an error later. */
2539 dm_table_presuspend_targets(map);
2542 * Flush I/O to the device.
2543 * Any I/O submitted after lock_fs() may not be flushed.
2544 * noflush takes precedence over do_lockfs.
2545 * (lock_fs() flushes I/Os and waits for them to complete.)
2547 if (!noflush && do_lockfs) {
2554 * Here we must make sure that no processes are submitting requests
2555 * to target drivers i.e. no one may be executing
2556 * __split_and_process_bio. This is called from dm_request and
2559 * To get all processes out of __split_and_process_bio in dm_request,
2560 * we take the write lock. To prevent any process from reentering
2561 * __split_and_process_bio from dm_request and quiesce the thread
2562 * (dm_wq_work), we set BMF_BLOCK_IO_FOR_SUSPEND and call
2563 * flush_workqueue(md->wq).
2565 down_write(&md->io_lock);
2566 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2567 up_write(&md->io_lock);
2570 * Stop md->queue before flushing md->wq in case request-based
2571 * dm defers requests to md->wq from md->queue.
2573 if (dm_request_based(md))
2574 stop_queue(md->queue);
2576 flush_workqueue(md->wq);
2579 * At this point no more requests are entering target request routines.
2580 * We call dm_wait_for_completion to wait for all existing requests
2583 r = dm_wait_for_completion(md, TASK_INTERRUPTIBLE);
2585 down_write(&md->io_lock);
2587 clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2588 up_write(&md->io_lock);
2590 /* were we interrupted ? */
2594 if (dm_request_based(md))
2595 start_queue(md->queue);
2598 goto out; /* pushback list is already flushed, so skip flush */
2602 * If dm_wait_for_completion returned 0, the device is completely
2603 * quiescent now. There is no request-processing activity. All new
2604 * requests are being added to md->deferred list.
2607 set_bit(DMF_SUSPENDED, &md->flags);
2609 dm_table_postsuspend_targets(map);
2615 mutex_unlock(&md->suspend_lock);
2619 int dm_resume(struct mapped_device *md)
2622 struct dm_table *map = NULL;
2624 mutex_lock(&md->suspend_lock);
2625 if (!dm_suspended_md(md))
2628 map = dm_get_live_table(md);
2629 if (!map || !dm_table_get_size(map))
2632 r = dm_table_resume_targets(map);
2639 * Flushing deferred I/Os must be done after targets are resumed
2640 * so that mapping of targets can work correctly.
2641 * Request-based dm is queueing the deferred I/Os in its request_queue.
2643 if (dm_request_based(md))
2644 start_queue(md->queue);
2648 clear_bit(DMF_SUSPENDED, &md->flags);
2653 mutex_unlock(&md->suspend_lock);
2658 /*-----------------------------------------------------------------
2659 * Event notification.
2660 *---------------------------------------------------------------*/
2661 int dm_kobject_uevent(struct mapped_device *md, enum kobject_action action,
2664 char udev_cookie[DM_COOKIE_LENGTH];
2665 char *envp[] = { udev_cookie, NULL };
2668 return kobject_uevent(&disk_to_dev(md->disk)->kobj, action);
2670 snprintf(udev_cookie, DM_COOKIE_LENGTH, "%s=%u",
2671 DM_COOKIE_ENV_VAR_NAME, cookie);
2672 return kobject_uevent_env(&disk_to_dev(md->disk)->kobj,
2677 uint32_t dm_next_uevent_seq(struct mapped_device *md)
2679 return atomic_add_return(1, &md->uevent_seq);
2682 uint32_t dm_get_event_nr(struct mapped_device *md)
2684 return atomic_read(&md->event_nr);
2687 int dm_wait_event(struct mapped_device *md, int event_nr)
2689 return wait_event_interruptible(md->eventq,
2690 (event_nr != atomic_read(&md->event_nr)));
2693 void dm_uevent_add(struct mapped_device *md, struct list_head *elist)
2695 unsigned long flags;
2697 spin_lock_irqsave(&md->uevent_lock, flags);
2698 list_add(elist, &md->uevent_list);
2699 spin_unlock_irqrestore(&md->uevent_lock, flags);
2703 * The gendisk is only valid as long as you have a reference
2706 struct gendisk *dm_disk(struct mapped_device *md)
2711 struct kobject *dm_kobject(struct mapped_device *md)
2717 * struct mapped_device should not be exported outside of dm.c
2718 * so use this check to verify that kobj is part of md structure
2720 struct mapped_device *dm_get_from_kobject(struct kobject *kobj)
2722 struct mapped_device *md;
2724 md = container_of(kobj, struct mapped_device, kobj);
2725 if (&md->kobj != kobj)
2728 if (test_bit(DMF_FREEING, &md->flags) ||
2736 int dm_suspended_md(struct mapped_device *md)
2738 return test_bit(DMF_SUSPENDED, &md->flags);
2741 int dm_suspended(struct dm_target *ti)
2743 return dm_suspended_md(dm_table_get_md(ti->table));
2745 EXPORT_SYMBOL_GPL(dm_suspended);
2747 int dm_noflush_suspending(struct dm_target *ti)
2749 return __noflush_suspending(dm_table_get_md(ti->table));
2751 EXPORT_SYMBOL_GPL(dm_noflush_suspending);
2753 struct dm_md_mempools *dm_alloc_md_mempools(unsigned type, unsigned integrity, unsigned per_bio_data_size)
2755 struct dm_md_mempools *pools = kzalloc(sizeof(*pools), GFP_KERNEL);
2756 struct kmem_cache *cachep;
2757 unsigned int pool_size;
2758 unsigned int front_pad;
2763 if (type == DM_TYPE_BIO_BASED) {
2766 front_pad = roundup(per_bio_data_size, __alignof__(struct dm_target_io)) + offsetof(struct dm_target_io, clone);
2767 } else if (type == DM_TYPE_REQUEST_BASED) {
2768 cachep = _rq_tio_cache;
2769 pool_size = MIN_IOS;
2770 front_pad = offsetof(struct dm_rq_clone_bio_info, clone);
2771 /* per_bio_data_size is not used. See __bind_mempools(). */
2772 WARN_ON(per_bio_data_size != 0);
2776 pools->io_pool = mempool_create_slab_pool(MIN_IOS, cachep);
2777 if (!pools->io_pool)
2780 pools->bs = bioset_create(pool_size, front_pad);
2784 if (integrity && bioset_integrity_create(pools->bs, pool_size))
2790 dm_free_md_mempools(pools);
2795 void dm_free_md_mempools(struct dm_md_mempools *pools)
2801 mempool_destroy(pools->io_pool);
2804 bioset_free(pools->bs);
2809 static const struct block_device_operations dm_blk_dops = {
2810 .open = dm_blk_open,
2811 .release = dm_blk_close,
2812 .ioctl = dm_blk_ioctl,
2813 .getgeo = dm_blk_getgeo,
2814 .owner = THIS_MODULE
2817 EXPORT_SYMBOL(dm_get_mapinfo);
2822 module_init(dm_init);
2823 module_exit(dm_exit);
2825 module_param(major, uint, 0);
2826 MODULE_PARM_DESC(major, "The major number of the device mapper");
2827 MODULE_DESCRIPTION(DM_NAME " driver");
2828 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
2829 MODULE_LICENSE("GPL");