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);
53 static void do_deferred_remove(struct work_struct *w);
55 static DECLARE_WORK(deferred_remove_work, do_deferred_remove);
59 * One of these is allocated per bio.
62 struct mapped_device *md;
66 unsigned long start_time;
67 spinlock_t endio_lock;
68 struct dm_stats_aux stats_aux;
72 * For request-based dm.
73 * One of these is allocated per request.
75 struct dm_rq_target_io {
76 struct mapped_device *md;
78 struct request *orig, clone;
84 * For request-based dm - the bio clones we allocate are embedded in these
87 * We allocate these with bio_alloc_bioset, using the front_pad parameter when
88 * the bioset is created - this means the bio has to come at the end of the
91 struct dm_rq_clone_bio_info {
93 struct dm_rq_target_io *tio;
97 union map_info *dm_get_rq_mapinfo(struct request *rq)
99 if (rq && rq->end_io_data)
100 return &((struct dm_rq_target_io *)rq->end_io_data)->info;
103 EXPORT_SYMBOL_GPL(dm_get_rq_mapinfo);
105 #define MINOR_ALLOCED ((void *)-1)
108 * Bits for the md->flags field.
110 #define DMF_BLOCK_IO_FOR_SUSPEND 0
111 #define DMF_SUSPENDED 1
113 #define DMF_FREEING 3
114 #define DMF_DELETING 4
115 #define DMF_NOFLUSH_SUSPENDING 5
116 #define DMF_MERGE_IS_OPTIONAL 6
117 #define DMF_DEFERRED_REMOVE 7
120 * A dummy definition to make RCU happy.
121 * struct dm_table should never be dereferenced in this file.
128 * Work processed by per-device workqueue.
130 struct mapped_device {
131 struct srcu_struct io_barrier;
132 struct mutex suspend_lock;
137 * The current mapping.
138 * Use dm_get_live_table{_fast} or take suspend_lock for
141 struct dm_table *map;
145 struct request_queue *queue;
147 /* Protect queue and type against concurrent access. */
148 struct mutex type_lock;
150 struct target_type *immutable_target_type;
152 struct gendisk *disk;
158 * A list of ios that arrived while we were suspended.
161 wait_queue_head_t wait;
162 struct work_struct work;
163 struct bio_list deferred;
164 spinlock_t deferred_lock;
167 * Processing queue (flush)
169 struct workqueue_struct *wq;
172 * io objects are allocated from here.
182 wait_queue_head_t eventq;
184 struct list_head uevent_list;
185 spinlock_t uevent_lock; /* Protect access to uevent_list */
188 * freeze/thaw support require holding onto a super block
190 struct super_block *frozen_sb;
191 struct block_device *bdev;
193 /* forced geometry settings */
194 struct hd_geometry geometry;
196 /* kobject and completion */
197 struct dm_kobject_holder kobj_holder;
199 /* zero-length flush that will be cloned and submitted to targets */
200 struct bio flush_bio;
202 struct dm_stats stats;
206 * For mempools pre-allocation at the table loading time.
208 struct dm_md_mempools {
213 #define RESERVED_BIO_BASED_IOS 16
214 #define RESERVED_REQUEST_BASED_IOS 256
215 #define RESERVED_MAX_IOS 1024
216 static struct kmem_cache *_io_cache;
217 static struct kmem_cache *_rq_tio_cache;
220 * Bio-based DM's mempools' reserved IOs set by the user.
222 static unsigned reserved_bio_based_ios = RESERVED_BIO_BASED_IOS;
225 * Request-based DM's mempools' reserved IOs set by the user.
227 static unsigned reserved_rq_based_ios = RESERVED_REQUEST_BASED_IOS;
229 static unsigned __dm_get_reserved_ios(unsigned *reserved_ios,
230 unsigned def, unsigned max)
232 unsigned ios = ACCESS_ONCE(*reserved_ios);
233 unsigned modified_ios = 0;
241 (void)cmpxchg(reserved_ios, ios, modified_ios);
248 unsigned dm_get_reserved_bio_based_ios(void)
250 return __dm_get_reserved_ios(&reserved_bio_based_ios,
251 RESERVED_BIO_BASED_IOS, RESERVED_MAX_IOS);
253 EXPORT_SYMBOL_GPL(dm_get_reserved_bio_based_ios);
255 unsigned dm_get_reserved_rq_based_ios(void)
257 return __dm_get_reserved_ios(&reserved_rq_based_ios,
258 RESERVED_REQUEST_BASED_IOS, RESERVED_MAX_IOS);
260 EXPORT_SYMBOL_GPL(dm_get_reserved_rq_based_ios);
262 static int __init local_init(void)
266 /* allocate a slab for the dm_ios */
267 _io_cache = KMEM_CACHE(dm_io, 0);
271 _rq_tio_cache = KMEM_CACHE(dm_rq_target_io, 0);
273 goto out_free_io_cache;
275 r = dm_uevent_init();
277 goto out_free_rq_tio_cache;
280 r = register_blkdev(_major, _name);
282 goto out_uevent_exit;
291 out_free_rq_tio_cache:
292 kmem_cache_destroy(_rq_tio_cache);
294 kmem_cache_destroy(_io_cache);
299 static void local_exit(void)
301 flush_scheduled_work();
303 kmem_cache_destroy(_rq_tio_cache);
304 kmem_cache_destroy(_io_cache);
305 unregister_blkdev(_major, _name);
310 DMINFO("cleaned up");
313 static int (*_inits[])(void) __initdata = {
324 static void (*_exits[])(void) = {
335 static int __init dm_init(void)
337 const int count = ARRAY_SIZE(_inits);
341 for (i = 0; i < count; i++) {
356 static void __exit dm_exit(void)
358 int i = ARRAY_SIZE(_exits);
364 * Should be empty by this point.
366 idr_destroy(&_minor_idr);
370 * Block device functions
372 int dm_deleting_md(struct mapped_device *md)
374 return test_bit(DMF_DELETING, &md->flags);
377 static int dm_blk_open(struct block_device *bdev, fmode_t mode)
379 struct mapped_device *md;
381 spin_lock(&_minor_lock);
383 md = bdev->bd_disk->private_data;
387 if (test_bit(DMF_FREEING, &md->flags) ||
388 dm_deleting_md(md)) {
394 atomic_inc(&md->open_count);
397 spin_unlock(&_minor_lock);
399 return md ? 0 : -ENXIO;
402 static void dm_blk_close(struct gendisk *disk, fmode_t mode)
404 struct mapped_device *md = disk->private_data;
406 spin_lock(&_minor_lock);
408 if (atomic_dec_and_test(&md->open_count) &&
409 (test_bit(DMF_DEFERRED_REMOVE, &md->flags)))
410 schedule_work(&deferred_remove_work);
414 spin_unlock(&_minor_lock);
417 int dm_open_count(struct mapped_device *md)
419 return atomic_read(&md->open_count);
423 * Guarantees nothing is using the device before it's deleted.
425 int dm_lock_for_deletion(struct mapped_device *md, bool mark_deferred, bool only_deferred)
429 spin_lock(&_minor_lock);
431 if (dm_open_count(md)) {
434 set_bit(DMF_DEFERRED_REMOVE, &md->flags);
435 } else if (only_deferred && !test_bit(DMF_DEFERRED_REMOVE, &md->flags))
438 set_bit(DMF_DELETING, &md->flags);
440 spin_unlock(&_minor_lock);
445 int dm_cancel_deferred_remove(struct mapped_device *md)
449 spin_lock(&_minor_lock);
451 if (test_bit(DMF_DELETING, &md->flags))
454 clear_bit(DMF_DEFERRED_REMOVE, &md->flags);
456 spin_unlock(&_minor_lock);
461 static void do_deferred_remove(struct work_struct *w)
463 dm_deferred_remove();
466 sector_t dm_get_size(struct mapped_device *md)
468 return get_capacity(md->disk);
471 struct request_queue *dm_get_md_queue(struct mapped_device *md)
476 struct dm_stats *dm_get_stats(struct mapped_device *md)
481 static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo)
483 struct mapped_device *md = bdev->bd_disk->private_data;
485 return dm_get_geometry(md, geo);
488 static int dm_blk_ioctl(struct block_device *bdev, fmode_t mode,
489 unsigned int cmd, unsigned long arg)
491 struct mapped_device *md = bdev->bd_disk->private_data;
493 struct dm_table *map;
494 struct dm_target *tgt;
498 map = dm_get_live_table(md, &srcu_idx);
500 if (!map || !dm_table_get_size(map))
503 /* We only support devices that have a single target */
504 if (dm_table_get_num_targets(map) != 1)
507 tgt = dm_table_get_target(map, 0);
509 if (dm_suspended_md(md)) {
514 if (tgt->type->ioctl)
515 r = tgt->type->ioctl(tgt, cmd, arg);
518 dm_put_live_table(md, srcu_idx);
520 if (r == -ENOTCONN) {
528 static struct dm_io *alloc_io(struct mapped_device *md)
530 return mempool_alloc(md->io_pool, GFP_NOIO);
533 static void free_io(struct mapped_device *md, struct dm_io *io)
535 mempool_free(io, md->io_pool);
538 static void free_tio(struct mapped_device *md, struct dm_target_io *tio)
540 bio_put(&tio->clone);
543 static struct dm_rq_target_io *alloc_rq_tio(struct mapped_device *md,
546 return mempool_alloc(md->io_pool, gfp_mask);
549 static void free_rq_tio(struct dm_rq_target_io *tio)
551 mempool_free(tio, tio->md->io_pool);
554 static int md_in_flight(struct mapped_device *md)
556 return atomic_read(&md->pending[READ]) +
557 atomic_read(&md->pending[WRITE]);
560 static void start_io_acct(struct dm_io *io)
562 struct mapped_device *md = io->md;
563 struct bio *bio = io->bio;
565 int rw = bio_data_dir(bio);
567 io->start_time = jiffies;
569 cpu = part_stat_lock();
570 part_round_stats(cpu, &dm_disk(md)->part0);
572 atomic_set(&dm_disk(md)->part0.in_flight[rw],
573 atomic_inc_return(&md->pending[rw]));
575 if (unlikely(dm_stats_used(&md->stats)))
576 dm_stats_account_io(&md->stats, bio->bi_rw, bio->bi_iter.bi_sector,
577 bio_sectors(bio), false, 0, &io->stats_aux);
580 static void end_io_acct(struct dm_io *io)
582 struct mapped_device *md = io->md;
583 struct bio *bio = io->bio;
584 unsigned long duration = jiffies - io->start_time;
586 int rw = bio_data_dir(bio);
588 cpu = part_stat_lock();
589 part_round_stats(cpu, &dm_disk(md)->part0);
590 part_stat_add(cpu, &dm_disk(md)->part0, ticks[rw], duration);
593 if (unlikely(dm_stats_used(&md->stats)))
594 dm_stats_account_io(&md->stats, bio->bi_rw, bio->bi_iter.bi_sector,
595 bio_sectors(bio), true, duration, &io->stats_aux);
598 * After this is decremented the bio must not be touched if it is
601 pending = atomic_dec_return(&md->pending[rw]);
602 atomic_set(&dm_disk(md)->part0.in_flight[rw], pending);
603 pending += atomic_read(&md->pending[rw^0x1]);
605 /* nudge anyone waiting on suspend queue */
611 * Add the bio to the list of deferred io.
613 static void queue_io(struct mapped_device *md, struct bio *bio)
617 spin_lock_irqsave(&md->deferred_lock, flags);
618 bio_list_add(&md->deferred, bio);
619 spin_unlock_irqrestore(&md->deferred_lock, flags);
620 queue_work(md->wq, &md->work);
624 * Everyone (including functions in this file), should use this
625 * function to access the md->map field, and make sure they call
626 * dm_put_live_table() when finished.
628 struct dm_table *dm_get_live_table(struct mapped_device *md, int *srcu_idx) __acquires(md->io_barrier)
630 *srcu_idx = srcu_read_lock(&md->io_barrier);
632 return srcu_dereference(md->map, &md->io_barrier);
635 void dm_put_live_table(struct mapped_device *md, int srcu_idx) __releases(md->io_barrier)
637 srcu_read_unlock(&md->io_barrier, srcu_idx);
640 void dm_sync_table(struct mapped_device *md)
642 synchronize_srcu(&md->io_barrier);
643 synchronize_rcu_expedited();
647 * A fast alternative to dm_get_live_table/dm_put_live_table.
648 * The caller must not block between these two functions.
650 static struct dm_table *dm_get_live_table_fast(struct mapped_device *md) __acquires(RCU)
653 return rcu_dereference(md->map);
656 static void dm_put_live_table_fast(struct mapped_device *md) __releases(RCU)
662 * Get the geometry associated with a dm device
664 int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo)
672 * Set the geometry of a device.
674 int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo)
676 sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors;
678 if (geo->start > sz) {
679 DMWARN("Start sector is beyond the geometry limits.");
688 /*-----------------------------------------------------------------
690 * A more elegant soln is in the works that uses the queue
691 * merge fn, unfortunately there are a couple of changes to
692 * the block layer that I want to make for this. So in the
693 * interests of getting something for people to use I give
694 * you this clearly demarcated crap.
695 *---------------------------------------------------------------*/
697 static int __noflush_suspending(struct mapped_device *md)
699 return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
703 * Decrements the number of outstanding ios that a bio has been
704 * cloned into, completing the original io if necc.
706 static void dec_pending(struct dm_io *io, int error)
711 struct mapped_device *md = io->md;
713 /* Push-back supersedes any I/O errors */
714 if (unlikely(error)) {
715 spin_lock_irqsave(&io->endio_lock, flags);
716 if (!(io->error > 0 && __noflush_suspending(md)))
718 spin_unlock_irqrestore(&io->endio_lock, flags);
721 if (atomic_dec_and_test(&io->io_count)) {
722 if (io->error == DM_ENDIO_REQUEUE) {
724 * Target requested pushing back the I/O.
726 spin_lock_irqsave(&md->deferred_lock, flags);
727 if (__noflush_suspending(md))
728 bio_list_add_head(&md->deferred, io->bio);
730 /* noflush suspend was interrupted. */
732 spin_unlock_irqrestore(&md->deferred_lock, flags);
735 io_error = io->error;
740 if (io_error == DM_ENDIO_REQUEUE)
743 if ((bio->bi_rw & REQ_FLUSH) && bio->bi_iter.bi_size) {
745 * Preflush done for flush with data, reissue
748 bio->bi_rw &= ~REQ_FLUSH;
751 /* done with normal IO or empty flush */
752 trace_block_bio_complete(md->queue, bio, io_error);
753 bio_endio(bio, io_error);
758 static void clone_endio(struct bio *bio, int error)
761 struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone);
762 struct dm_io *io = tio->io;
763 struct mapped_device *md = tio->io->md;
764 dm_endio_fn endio = tio->ti->type->end_io;
766 if (!bio_flagged(bio, BIO_UPTODATE) && !error)
770 r = endio(tio->ti, bio, error);
771 if (r < 0 || r == DM_ENDIO_REQUEUE)
773 * error and requeue request are handled
777 else if (r == DM_ENDIO_INCOMPLETE)
778 /* The target will handle the io */
781 DMWARN("unimplemented target endio return value: %d", r);
787 dec_pending(io, error);
791 * Partial completion handling for request-based dm
793 static void end_clone_bio(struct bio *clone, int error)
795 struct dm_rq_clone_bio_info *info =
796 container_of(clone, struct dm_rq_clone_bio_info, clone);
797 struct dm_rq_target_io *tio = info->tio;
798 struct bio *bio = info->orig;
799 unsigned int nr_bytes = info->orig->bi_iter.bi_size;
805 * An error has already been detected on the request.
806 * Once error occurred, just let clone->end_io() handle
812 * Don't notice the error to the upper layer yet.
813 * The error handling decision is made by the target driver,
814 * when the request is completed.
821 * I/O for the bio successfully completed.
822 * Notice the data completion to the upper layer.
826 * bios are processed from the head of the list.
827 * So the completing bio should always be rq->bio.
828 * If it's not, something wrong is happening.
830 if (tio->orig->bio != bio)
831 DMERR("bio completion is going in the middle of the request");
834 * Update the original request.
835 * Do not use blk_end_request() here, because it may complete
836 * the original request before the clone, and break the ordering.
838 blk_update_request(tio->orig, 0, nr_bytes);
842 * Don't touch any member of the md after calling this function because
843 * the md may be freed in dm_put() at the end of this function.
844 * Or do dm_get() before calling this function and dm_put() later.
846 static void rq_completed(struct mapped_device *md, int rw, int run_queue)
848 atomic_dec(&md->pending[rw]);
850 /* nudge anyone waiting on suspend queue */
851 if (!md_in_flight(md))
855 * Run this off this callpath, as drivers could invoke end_io while
856 * inside their request_fn (and holding the queue lock). Calling
857 * back into ->request_fn() could deadlock attempting to grab the
861 blk_run_queue_async(md->queue);
864 * dm_put() must be at the end of this function. See the comment above
869 static void free_rq_clone(struct request *clone)
871 struct dm_rq_target_io *tio = clone->end_io_data;
873 blk_rq_unprep_clone(clone);
878 * Complete the clone and the original request.
879 * Must be called without queue lock.
881 static void dm_end_request(struct request *clone, int error)
883 int rw = rq_data_dir(clone);
884 struct dm_rq_target_io *tio = clone->end_io_data;
885 struct mapped_device *md = tio->md;
886 struct request *rq = tio->orig;
888 if (rq->cmd_type == REQ_TYPE_BLOCK_PC) {
889 rq->errors = clone->errors;
890 rq->resid_len = clone->resid_len;
894 * We are using the sense buffer of the original
896 * So setting the length of the sense data is enough.
898 rq->sense_len = clone->sense_len;
901 free_rq_clone(clone);
902 blk_end_request_all(rq, error);
903 rq_completed(md, rw, true);
906 static void dm_unprep_request(struct request *rq)
908 struct request *clone = rq->special;
911 rq->cmd_flags &= ~REQ_DONTPREP;
913 free_rq_clone(clone);
917 * Requeue the original request of a clone.
919 void dm_requeue_unmapped_request(struct request *clone)
921 int rw = rq_data_dir(clone);
922 struct dm_rq_target_io *tio = clone->end_io_data;
923 struct mapped_device *md = tio->md;
924 struct request *rq = tio->orig;
925 struct request_queue *q = rq->q;
928 dm_unprep_request(rq);
930 spin_lock_irqsave(q->queue_lock, flags);
931 blk_requeue_request(q, rq);
932 spin_unlock_irqrestore(q->queue_lock, flags);
934 rq_completed(md, rw, 0);
936 EXPORT_SYMBOL_GPL(dm_requeue_unmapped_request);
938 static void __stop_queue(struct request_queue *q)
943 static void stop_queue(struct request_queue *q)
947 spin_lock_irqsave(q->queue_lock, flags);
949 spin_unlock_irqrestore(q->queue_lock, flags);
952 static void __start_queue(struct request_queue *q)
954 if (blk_queue_stopped(q))
958 static void start_queue(struct request_queue *q)
962 spin_lock_irqsave(q->queue_lock, flags);
964 spin_unlock_irqrestore(q->queue_lock, flags);
967 static void dm_done(struct request *clone, int error, bool mapped)
970 struct dm_rq_target_io *tio = clone->end_io_data;
971 dm_request_endio_fn rq_end_io = NULL;
974 rq_end_io = tio->ti->type->rq_end_io;
976 if (mapped && rq_end_io)
977 r = rq_end_io(tio->ti, clone, error, &tio->info);
981 /* The target wants to complete the I/O */
982 dm_end_request(clone, r);
983 else if (r == DM_ENDIO_INCOMPLETE)
984 /* The target will handle the I/O */
986 else if (r == DM_ENDIO_REQUEUE)
987 /* The target wants to requeue the I/O */
988 dm_requeue_unmapped_request(clone);
990 DMWARN("unimplemented target endio return value: %d", r);
996 * Request completion handler for request-based dm
998 static void dm_softirq_done(struct request *rq)
1001 struct request *clone = rq->completion_data;
1002 struct dm_rq_target_io *tio = clone->end_io_data;
1004 if (rq->cmd_flags & REQ_FAILED)
1007 dm_done(clone, tio->error, mapped);
1011 * Complete the clone and the original request with the error status
1012 * through softirq context.
1014 static void dm_complete_request(struct request *clone, int error)
1016 struct dm_rq_target_io *tio = clone->end_io_data;
1017 struct request *rq = tio->orig;
1020 rq->completion_data = clone;
1021 blk_complete_request(rq);
1025 * Complete the not-mapped clone and the original request with the error status
1026 * through softirq context.
1027 * Target's rq_end_io() function isn't called.
1028 * This may be used when the target's map_rq() function fails.
1030 void dm_kill_unmapped_request(struct request *clone, int error)
1032 struct dm_rq_target_io *tio = clone->end_io_data;
1033 struct request *rq = tio->orig;
1035 rq->cmd_flags |= REQ_FAILED;
1036 dm_complete_request(clone, error);
1038 EXPORT_SYMBOL_GPL(dm_kill_unmapped_request);
1041 * Called with the queue lock held
1043 static void end_clone_request(struct request *clone, int error)
1046 * For just cleaning up the information of the queue in which
1047 * the clone was dispatched.
1048 * The clone is *NOT* freed actually here because it is alloced from
1049 * dm own mempool and REQ_ALLOCED isn't set in clone->cmd_flags.
1051 __blk_put_request(clone->q, clone);
1054 * Actual request completion is done in a softirq context which doesn't
1055 * hold the queue lock. Otherwise, deadlock could occur because:
1056 * - another request may be submitted by the upper level driver
1057 * of the stacking during the completion
1058 * - the submission which requires queue lock may be done
1059 * against this queue
1061 dm_complete_request(clone, error);
1065 * Return maximum size of I/O possible at the supplied sector up to the current
1068 static sector_t max_io_len_target_boundary(sector_t sector, struct dm_target *ti)
1070 sector_t target_offset = dm_target_offset(ti, sector);
1072 return ti->len - target_offset;
1075 static sector_t max_io_len(sector_t sector, struct dm_target *ti)
1077 sector_t len = max_io_len_target_boundary(sector, ti);
1078 sector_t offset, max_len;
1081 * Does the target need to split even further?
1083 if (ti->max_io_len) {
1084 offset = dm_target_offset(ti, sector);
1085 if (unlikely(ti->max_io_len & (ti->max_io_len - 1)))
1086 max_len = sector_div(offset, ti->max_io_len);
1088 max_len = offset & (ti->max_io_len - 1);
1089 max_len = ti->max_io_len - max_len;
1098 int dm_set_target_max_io_len(struct dm_target *ti, sector_t len)
1100 if (len > UINT_MAX) {
1101 DMERR("Specified maximum size of target IO (%llu) exceeds limit (%u)",
1102 (unsigned long long)len, UINT_MAX);
1103 ti->error = "Maximum size of target IO is too large";
1107 ti->max_io_len = (uint32_t) len;
1111 EXPORT_SYMBOL_GPL(dm_set_target_max_io_len);
1114 * A target may call dm_accept_partial_bio only from the map routine. It is
1115 * allowed for all bio types except REQ_FLUSH.
1117 * dm_accept_partial_bio informs the dm that the target only wants to process
1118 * additional n_sectors sectors of the bio and the rest of the data should be
1119 * sent in a next bio.
1121 * A diagram that explains the arithmetics:
1122 * +--------------------+---------------+-------+
1124 * +--------------------+---------------+-------+
1126 * <-------------- *tio->len_ptr --------------->
1127 * <------- bi_size ------->
1130 * Region 1 was already iterated over with bio_advance or similar function.
1131 * (it may be empty if the target doesn't use bio_advance)
1132 * Region 2 is the remaining bio size that the target wants to process.
1133 * (it may be empty if region 1 is non-empty, although there is no reason
1135 * The target requires that region 3 is to be sent in the next bio.
1137 * If the target wants to receive multiple copies of the bio (via num_*bios, etc),
1138 * the partially processed part (the sum of regions 1+2) must be the same for all
1139 * copies of the bio.
1141 void dm_accept_partial_bio(struct bio *bio, unsigned n_sectors)
1143 struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone);
1144 unsigned bi_size = bio->bi_iter.bi_size >> SECTOR_SHIFT;
1145 BUG_ON(bio->bi_rw & REQ_FLUSH);
1146 BUG_ON(bi_size > *tio->len_ptr);
1147 BUG_ON(n_sectors > bi_size);
1148 *tio->len_ptr -= bi_size - n_sectors;
1149 bio->bi_iter.bi_size = n_sectors << SECTOR_SHIFT;
1151 EXPORT_SYMBOL_GPL(dm_accept_partial_bio);
1153 static void __map_bio(struct dm_target_io *tio)
1157 struct mapped_device *md;
1158 struct bio *clone = &tio->clone;
1159 struct dm_target *ti = tio->ti;
1161 clone->bi_end_io = clone_endio;
1164 * Map the clone. If r == 0 we don't need to do
1165 * anything, the target has assumed ownership of
1168 atomic_inc(&tio->io->io_count);
1169 sector = clone->bi_iter.bi_sector;
1170 r = ti->type->map(ti, clone);
1171 if (r == DM_MAPIO_REMAPPED) {
1172 /* the bio has been remapped so dispatch it */
1174 trace_block_bio_remap(bdev_get_queue(clone->bi_bdev), clone,
1175 tio->io->bio->bi_bdev->bd_dev, sector);
1177 generic_make_request(clone);
1178 } else if (r < 0 || r == DM_MAPIO_REQUEUE) {
1179 /* error the io and bail out, or requeue it if needed */
1181 dec_pending(tio->io, r);
1184 DMWARN("unimplemented target map return value: %d", r);
1190 struct mapped_device *md;
1191 struct dm_table *map;
1195 unsigned sector_count;
1198 static void bio_setup_sector(struct bio *bio, sector_t sector, unsigned len)
1200 bio->bi_iter.bi_sector = sector;
1201 bio->bi_iter.bi_size = to_bytes(len);
1205 * Creates a bio that consists of range of complete bvecs.
1207 static void clone_bio(struct dm_target_io *tio, struct bio *bio,
1208 sector_t sector, unsigned len)
1210 struct bio *clone = &tio->clone;
1212 __bio_clone_fast(clone, bio);
1214 if (bio_integrity(bio))
1215 bio_integrity_clone(clone, bio, GFP_NOIO);
1217 bio_advance(clone, to_bytes(sector - clone->bi_iter.bi_sector));
1218 clone->bi_iter.bi_size = to_bytes(len);
1220 if (bio_integrity(bio))
1221 bio_integrity_trim(clone, 0, len);
1224 static struct dm_target_io *alloc_tio(struct clone_info *ci,
1225 struct dm_target *ti, int nr_iovecs,
1226 unsigned target_bio_nr)
1228 struct dm_target_io *tio;
1231 clone = bio_alloc_bioset(GFP_NOIO, nr_iovecs, ci->md->bs);
1232 tio = container_of(clone, struct dm_target_io, clone);
1236 tio->target_bio_nr = target_bio_nr;
1241 static void __clone_and_map_simple_bio(struct clone_info *ci,
1242 struct dm_target *ti,
1243 unsigned target_bio_nr, unsigned *len)
1245 struct dm_target_io *tio = alloc_tio(ci, ti, ci->bio->bi_max_vecs, target_bio_nr);
1246 struct bio *clone = &tio->clone;
1251 * Discard requests require the bio's inline iovecs be initialized.
1252 * ci->bio->bi_max_vecs is BIO_INLINE_VECS anyway, for both flush
1253 * and discard, so no need for concern about wasted bvec allocations.
1255 __bio_clone_fast(clone, ci->bio);
1257 bio_setup_sector(clone, ci->sector, *len);
1262 static void __send_duplicate_bios(struct clone_info *ci, struct dm_target *ti,
1263 unsigned num_bios, unsigned *len)
1265 unsigned target_bio_nr;
1267 for (target_bio_nr = 0; target_bio_nr < num_bios; target_bio_nr++)
1268 __clone_and_map_simple_bio(ci, ti, target_bio_nr, len);
1271 static int __send_empty_flush(struct clone_info *ci)
1273 unsigned target_nr = 0;
1274 struct dm_target *ti;
1276 BUG_ON(bio_has_data(ci->bio));
1277 while ((ti = dm_table_get_target(ci->map, target_nr++)))
1278 __send_duplicate_bios(ci, ti, ti->num_flush_bios, NULL);
1283 static void __clone_and_map_data_bio(struct clone_info *ci, struct dm_target *ti,
1284 sector_t sector, unsigned *len)
1286 struct bio *bio = ci->bio;
1287 struct dm_target_io *tio;
1288 unsigned target_bio_nr;
1289 unsigned num_target_bios = 1;
1292 * Does the target want to receive duplicate copies of the bio?
1294 if (bio_data_dir(bio) == WRITE && ti->num_write_bios)
1295 num_target_bios = ti->num_write_bios(ti, bio);
1297 for (target_bio_nr = 0; target_bio_nr < num_target_bios; target_bio_nr++) {
1298 tio = alloc_tio(ci, ti, 0, target_bio_nr);
1300 clone_bio(tio, bio, sector, *len);
1305 typedef unsigned (*get_num_bios_fn)(struct dm_target *ti);
1307 static unsigned get_num_discard_bios(struct dm_target *ti)
1309 return ti->num_discard_bios;
1312 static unsigned get_num_write_same_bios(struct dm_target *ti)
1314 return ti->num_write_same_bios;
1317 typedef bool (*is_split_required_fn)(struct dm_target *ti);
1319 static bool is_split_required_for_discard(struct dm_target *ti)
1321 return ti->split_discard_bios;
1324 static int __send_changing_extent_only(struct clone_info *ci,
1325 get_num_bios_fn get_num_bios,
1326 is_split_required_fn is_split_required)
1328 struct dm_target *ti;
1333 ti = dm_table_find_target(ci->map, ci->sector);
1334 if (!dm_target_is_valid(ti))
1338 * Even though the device advertised support for this type of
1339 * request, that does not mean every target supports it, and
1340 * reconfiguration might also have changed that since the
1341 * check was performed.
1343 num_bios = get_num_bios ? get_num_bios(ti) : 0;
1347 if (is_split_required && !is_split_required(ti))
1348 len = min((sector_t)ci->sector_count, max_io_len_target_boundary(ci->sector, ti));
1350 len = min((sector_t)ci->sector_count, max_io_len(ci->sector, ti));
1352 __send_duplicate_bios(ci, ti, num_bios, &len);
1355 } while (ci->sector_count -= len);
1360 static int __send_discard(struct clone_info *ci)
1362 return __send_changing_extent_only(ci, get_num_discard_bios,
1363 is_split_required_for_discard);
1366 static int __send_write_same(struct clone_info *ci)
1368 return __send_changing_extent_only(ci, get_num_write_same_bios, NULL);
1372 * Select the correct strategy for processing a non-flush bio.
1374 static int __split_and_process_non_flush(struct clone_info *ci)
1376 struct bio *bio = ci->bio;
1377 struct dm_target *ti;
1380 if (unlikely(bio->bi_rw & REQ_DISCARD))
1381 return __send_discard(ci);
1382 else if (unlikely(bio->bi_rw & REQ_WRITE_SAME))
1383 return __send_write_same(ci);
1385 ti = dm_table_find_target(ci->map, ci->sector);
1386 if (!dm_target_is_valid(ti))
1389 len = min_t(sector_t, max_io_len(ci->sector, ti), ci->sector_count);
1391 __clone_and_map_data_bio(ci, ti, ci->sector, &len);
1394 ci->sector_count -= len;
1400 * Entry point to split a bio into clones and submit them to the targets.
1402 static void __split_and_process_bio(struct mapped_device *md,
1403 struct dm_table *map, struct bio *bio)
1405 struct clone_info ci;
1408 if (unlikely(!map)) {
1415 ci.io = alloc_io(md);
1417 atomic_set(&ci.io->io_count, 1);
1420 spin_lock_init(&ci.io->endio_lock);
1421 ci.sector = bio->bi_iter.bi_sector;
1423 start_io_acct(ci.io);
1425 if (bio->bi_rw & REQ_FLUSH) {
1426 ci.bio = &ci.md->flush_bio;
1427 ci.sector_count = 0;
1428 error = __send_empty_flush(&ci);
1429 /* dec_pending submits any data associated with flush */
1432 ci.sector_count = bio_sectors(bio);
1433 while (ci.sector_count && !error)
1434 error = __split_and_process_non_flush(&ci);
1437 /* drop the extra reference count */
1438 dec_pending(ci.io, error);
1440 /*-----------------------------------------------------------------
1442 *---------------------------------------------------------------*/
1444 static int dm_merge_bvec(struct request_queue *q,
1445 struct bvec_merge_data *bvm,
1446 struct bio_vec *biovec)
1448 struct mapped_device *md = q->queuedata;
1449 struct dm_table *map = dm_get_live_table_fast(md);
1450 struct dm_target *ti;
1451 sector_t max_sectors;
1457 ti = dm_table_find_target(map, bvm->bi_sector);
1458 if (!dm_target_is_valid(ti))
1462 * Find maximum amount of I/O that won't need splitting
1464 max_sectors = min(max_io_len(bvm->bi_sector, ti),
1465 (sector_t) BIO_MAX_SECTORS);
1466 max_size = (max_sectors << SECTOR_SHIFT) - bvm->bi_size;
1471 * merge_bvec_fn() returns number of bytes
1472 * it can accept at this offset
1473 * max is precomputed maximal io size
1475 if (max_size && ti->type->merge)
1476 max_size = ti->type->merge(ti, bvm, biovec, max_size);
1478 * If the target doesn't support merge method and some of the devices
1479 * provided their merge_bvec method (we know this by looking at
1480 * queue_max_hw_sectors), then we can't allow bios with multiple vector
1481 * entries. So always set max_size to 0, and the code below allows
1484 else if (queue_max_hw_sectors(q) <= PAGE_SIZE >> 9)
1489 dm_put_live_table_fast(md);
1491 * Always allow an entire first page
1493 if (max_size <= biovec->bv_len && !(bvm->bi_size >> SECTOR_SHIFT))
1494 max_size = biovec->bv_len;
1500 * The request function that just remaps the bio built up by
1503 static void _dm_request(struct request_queue *q, struct bio *bio)
1505 int rw = bio_data_dir(bio);
1506 struct mapped_device *md = q->queuedata;
1509 struct dm_table *map;
1511 map = dm_get_live_table(md, &srcu_idx);
1513 cpu = part_stat_lock();
1514 part_stat_inc(cpu, &dm_disk(md)->part0, ios[rw]);
1515 part_stat_add(cpu, &dm_disk(md)->part0, sectors[rw], bio_sectors(bio));
1518 /* if we're suspended, we have to queue this io for later */
1519 if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))) {
1520 dm_put_live_table(md, srcu_idx);
1522 if (bio_rw(bio) != READA)
1529 __split_and_process_bio(md, map, bio);
1530 dm_put_live_table(md, srcu_idx);
1534 int dm_request_based(struct mapped_device *md)
1536 return blk_queue_stackable(md->queue);
1539 static void dm_request(struct request_queue *q, struct bio *bio)
1541 struct mapped_device *md = q->queuedata;
1543 if (dm_request_based(md))
1544 blk_queue_bio(q, bio);
1546 _dm_request(q, bio);
1549 void dm_dispatch_request(struct request *rq)
1553 if (blk_queue_io_stat(rq->q))
1554 rq->cmd_flags |= REQ_IO_STAT;
1556 rq->start_time = jiffies;
1557 r = blk_insert_cloned_request(rq->q, rq);
1559 dm_complete_request(rq, r);
1561 EXPORT_SYMBOL_GPL(dm_dispatch_request);
1563 static int dm_rq_bio_constructor(struct bio *bio, struct bio *bio_orig,
1566 struct dm_rq_target_io *tio = data;
1567 struct dm_rq_clone_bio_info *info =
1568 container_of(bio, struct dm_rq_clone_bio_info, clone);
1570 info->orig = bio_orig;
1572 bio->bi_end_io = end_clone_bio;
1577 static int setup_clone(struct request *clone, struct request *rq,
1578 struct dm_rq_target_io *tio)
1582 r = blk_rq_prep_clone(clone, rq, tio->md->bs, GFP_ATOMIC,
1583 dm_rq_bio_constructor, tio);
1587 clone->cmd = rq->cmd;
1588 clone->cmd_len = rq->cmd_len;
1589 clone->sense = rq->sense;
1590 clone->buffer = rq->buffer;
1591 clone->end_io = end_clone_request;
1592 clone->end_io_data = tio;
1597 static struct request *clone_rq(struct request *rq, struct mapped_device *md,
1600 struct request *clone;
1601 struct dm_rq_target_io *tio;
1603 tio = alloc_rq_tio(md, gfp_mask);
1611 memset(&tio->info, 0, sizeof(tio->info));
1613 clone = &tio->clone;
1614 if (setup_clone(clone, rq, tio)) {
1624 * Called with the queue lock held.
1626 static int dm_prep_fn(struct request_queue *q, struct request *rq)
1628 struct mapped_device *md = q->queuedata;
1629 struct request *clone;
1631 if (unlikely(rq->special)) {
1632 DMWARN("Already has something in rq->special.");
1633 return BLKPREP_KILL;
1636 clone = clone_rq(rq, md, GFP_ATOMIC);
1638 return BLKPREP_DEFER;
1640 rq->special = clone;
1641 rq->cmd_flags |= REQ_DONTPREP;
1648 * 0 : the request has been processed (not requeued)
1649 * !0 : the request has been requeued
1651 static int map_request(struct dm_target *ti, struct request *clone,
1652 struct mapped_device *md)
1654 int r, requeued = 0;
1655 struct dm_rq_target_io *tio = clone->end_io_data;
1658 r = ti->type->map_rq(ti, clone, &tio->info);
1660 case DM_MAPIO_SUBMITTED:
1661 /* The target has taken the I/O to submit by itself later */
1663 case DM_MAPIO_REMAPPED:
1664 /* The target has remapped the I/O so dispatch it */
1665 trace_block_rq_remap(clone->q, clone, disk_devt(dm_disk(md)),
1666 blk_rq_pos(tio->orig));
1667 dm_dispatch_request(clone);
1669 case DM_MAPIO_REQUEUE:
1670 /* The target wants to requeue the I/O */
1671 dm_requeue_unmapped_request(clone);
1676 DMWARN("unimplemented target map return value: %d", r);
1680 /* The target wants to complete the I/O */
1681 dm_kill_unmapped_request(clone, r);
1688 static struct request *dm_start_request(struct mapped_device *md, struct request *orig)
1690 struct request *clone;
1692 blk_start_request(orig);
1693 clone = orig->special;
1694 atomic_inc(&md->pending[rq_data_dir(clone)]);
1697 * Hold the md reference here for the in-flight I/O.
1698 * We can't rely on the reference count by device opener,
1699 * because the device may be closed during the request completion
1700 * when all bios are completed.
1701 * See the comment in rq_completed() too.
1709 * q->request_fn for request-based dm.
1710 * Called with the queue lock held.
1712 static void dm_request_fn(struct request_queue *q)
1714 struct mapped_device *md = q->queuedata;
1716 struct dm_table *map = dm_get_live_table(md, &srcu_idx);
1717 struct dm_target *ti;
1718 struct request *rq, *clone;
1722 * For suspend, check blk_queue_stopped() and increment
1723 * ->pending within a single queue_lock not to increment the
1724 * number of in-flight I/Os after the queue is stopped in
1727 while (!blk_queue_stopped(q)) {
1728 rq = blk_peek_request(q);
1732 /* always use block 0 to find the target for flushes for now */
1734 if (!(rq->cmd_flags & REQ_FLUSH))
1735 pos = blk_rq_pos(rq);
1737 ti = dm_table_find_target(map, pos);
1738 if (!dm_target_is_valid(ti)) {
1740 * Must perform setup, that dm_done() requires,
1741 * before calling dm_kill_unmapped_request
1743 DMERR_LIMIT("request attempted access beyond the end of device");
1744 clone = dm_start_request(md, rq);
1745 dm_kill_unmapped_request(clone, -EIO);
1749 if (ti->type->busy && ti->type->busy(ti))
1752 clone = dm_start_request(md, rq);
1754 spin_unlock(q->queue_lock);
1755 if (map_request(ti, clone, md))
1758 BUG_ON(!irqs_disabled());
1759 spin_lock(q->queue_lock);
1765 BUG_ON(!irqs_disabled());
1766 spin_lock(q->queue_lock);
1769 blk_delay_queue(q, HZ / 10);
1771 dm_put_live_table(md, srcu_idx);
1774 int dm_underlying_device_busy(struct request_queue *q)
1776 return blk_lld_busy(q);
1778 EXPORT_SYMBOL_GPL(dm_underlying_device_busy);
1780 static int dm_lld_busy(struct request_queue *q)
1783 struct mapped_device *md = q->queuedata;
1784 struct dm_table *map = dm_get_live_table_fast(md);
1786 if (!map || test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))
1789 r = dm_table_any_busy_target(map);
1791 dm_put_live_table_fast(md);
1796 static int dm_any_congested(void *congested_data, int bdi_bits)
1799 struct mapped_device *md = congested_data;
1800 struct dm_table *map;
1802 if (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
1803 map = dm_get_live_table_fast(md);
1806 * Request-based dm cares about only own queue for
1807 * the query about congestion status of request_queue
1809 if (dm_request_based(md))
1810 r = md->queue->backing_dev_info.state &
1813 r = dm_table_any_congested(map, bdi_bits);
1815 dm_put_live_table_fast(md);
1821 /*-----------------------------------------------------------------
1822 * An IDR is used to keep track of allocated minor numbers.
1823 *---------------------------------------------------------------*/
1824 static void free_minor(int minor)
1826 spin_lock(&_minor_lock);
1827 idr_remove(&_minor_idr, minor);
1828 spin_unlock(&_minor_lock);
1832 * See if the device with a specific minor # is free.
1834 static int specific_minor(int minor)
1838 if (minor >= (1 << MINORBITS))
1841 idr_preload(GFP_KERNEL);
1842 spin_lock(&_minor_lock);
1844 r = idr_alloc(&_minor_idr, MINOR_ALLOCED, minor, minor + 1, GFP_NOWAIT);
1846 spin_unlock(&_minor_lock);
1849 return r == -ENOSPC ? -EBUSY : r;
1853 static int next_free_minor(int *minor)
1857 idr_preload(GFP_KERNEL);
1858 spin_lock(&_minor_lock);
1860 r = idr_alloc(&_minor_idr, MINOR_ALLOCED, 0, 1 << MINORBITS, GFP_NOWAIT);
1862 spin_unlock(&_minor_lock);
1870 static const struct block_device_operations dm_blk_dops;
1872 static void dm_wq_work(struct work_struct *work);
1874 static void dm_init_md_queue(struct mapped_device *md)
1877 * Request-based dm devices cannot be stacked on top of bio-based dm
1878 * devices. The type of this dm device has not been decided yet.
1879 * The type is decided at the first table loading time.
1880 * To prevent problematic device stacking, clear the queue flag
1881 * for request stacking support until then.
1883 * This queue is new, so no concurrency on the queue_flags.
1885 queue_flag_clear_unlocked(QUEUE_FLAG_STACKABLE, md->queue);
1887 md->queue->queuedata = md;
1888 md->queue->backing_dev_info.congested_fn = dm_any_congested;
1889 md->queue->backing_dev_info.congested_data = md;
1890 blk_queue_make_request(md->queue, dm_request);
1891 blk_queue_bounce_limit(md->queue, BLK_BOUNCE_ANY);
1892 blk_queue_merge_bvec(md->queue, dm_merge_bvec);
1896 * Allocate and initialise a blank device with a given minor.
1898 static struct mapped_device *alloc_dev(int minor)
1901 struct mapped_device *md = kzalloc(sizeof(*md), GFP_KERNEL);
1905 DMWARN("unable to allocate device, out of memory.");
1909 if (!try_module_get(THIS_MODULE))
1910 goto bad_module_get;
1912 /* get a minor number for the dev */
1913 if (minor == DM_ANY_MINOR)
1914 r = next_free_minor(&minor);
1916 r = specific_minor(minor);
1920 r = init_srcu_struct(&md->io_barrier);
1922 goto bad_io_barrier;
1924 md->type = DM_TYPE_NONE;
1925 mutex_init(&md->suspend_lock);
1926 mutex_init(&md->type_lock);
1927 spin_lock_init(&md->deferred_lock);
1928 atomic_set(&md->holders, 1);
1929 atomic_set(&md->open_count, 0);
1930 atomic_set(&md->event_nr, 0);
1931 atomic_set(&md->uevent_seq, 0);
1932 INIT_LIST_HEAD(&md->uevent_list);
1933 spin_lock_init(&md->uevent_lock);
1935 md->queue = blk_alloc_queue(GFP_KERNEL);
1939 dm_init_md_queue(md);
1941 md->disk = alloc_disk(1);
1945 atomic_set(&md->pending[0], 0);
1946 atomic_set(&md->pending[1], 0);
1947 init_waitqueue_head(&md->wait);
1948 INIT_WORK(&md->work, dm_wq_work);
1949 init_waitqueue_head(&md->eventq);
1950 init_completion(&md->kobj_holder.completion);
1952 md->disk->major = _major;
1953 md->disk->first_minor = minor;
1954 md->disk->fops = &dm_blk_dops;
1955 md->disk->queue = md->queue;
1956 md->disk->private_data = md;
1957 sprintf(md->disk->disk_name, "dm-%d", minor);
1959 format_dev_t(md->name, MKDEV(_major, minor));
1961 md->wq = alloc_workqueue("kdmflush", WQ_MEM_RECLAIM, 0);
1965 md->bdev = bdget_disk(md->disk, 0);
1969 bio_init(&md->flush_bio);
1970 md->flush_bio.bi_bdev = md->bdev;
1971 md->flush_bio.bi_rw = WRITE_FLUSH;
1973 dm_stats_init(&md->stats);
1975 /* Populate the mapping, nobody knows we exist yet */
1976 spin_lock(&_minor_lock);
1977 old_md = idr_replace(&_minor_idr, md, minor);
1978 spin_unlock(&_minor_lock);
1980 BUG_ON(old_md != MINOR_ALLOCED);
1985 destroy_workqueue(md->wq);
1987 del_gendisk(md->disk);
1990 blk_cleanup_queue(md->queue);
1992 cleanup_srcu_struct(&md->io_barrier);
1996 module_put(THIS_MODULE);
2002 static void unlock_fs(struct mapped_device *md);
2004 static void free_dev(struct mapped_device *md)
2006 int minor = MINOR(disk_devt(md->disk));
2010 destroy_workqueue(md->wq);
2012 mempool_destroy(md->io_pool);
2014 bioset_free(md->bs);
2015 blk_integrity_unregister(md->disk);
2016 del_gendisk(md->disk);
2017 cleanup_srcu_struct(&md->io_barrier);
2020 spin_lock(&_minor_lock);
2021 md->disk->private_data = NULL;
2022 spin_unlock(&_minor_lock);
2025 blk_cleanup_queue(md->queue);
2026 dm_stats_cleanup(&md->stats);
2027 module_put(THIS_MODULE);
2031 static void __bind_mempools(struct mapped_device *md, struct dm_table *t)
2033 struct dm_md_mempools *p = dm_table_get_md_mempools(t);
2035 if (md->io_pool && md->bs) {
2036 /* The md already has necessary mempools. */
2037 if (dm_table_get_type(t) == DM_TYPE_BIO_BASED) {
2039 * Reload bioset because front_pad may have changed
2040 * because a different table was loaded.
2042 bioset_free(md->bs);
2045 } else if (dm_table_get_type(t) == DM_TYPE_REQUEST_BASED) {
2047 * There's no need to reload with request-based dm
2048 * because the size of front_pad doesn't change.
2049 * Note for future: If you are to reload bioset,
2050 * prep-ed requests in the queue may refer
2051 * to bio from the old bioset, so you must walk
2052 * through the queue to unprep.
2058 BUG_ON(!p || md->io_pool || md->bs);
2060 md->io_pool = p->io_pool;
2066 /* mempool bind completed, now no need any mempools in the table */
2067 dm_table_free_md_mempools(t);
2071 * Bind a table to the device.
2073 static void event_callback(void *context)
2075 unsigned long flags;
2077 struct mapped_device *md = (struct mapped_device *) context;
2079 spin_lock_irqsave(&md->uevent_lock, flags);
2080 list_splice_init(&md->uevent_list, &uevents);
2081 spin_unlock_irqrestore(&md->uevent_lock, flags);
2083 dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj);
2085 atomic_inc(&md->event_nr);
2086 wake_up(&md->eventq);
2090 * Protected by md->suspend_lock obtained by dm_swap_table().
2092 static void __set_size(struct mapped_device *md, sector_t size)
2094 set_capacity(md->disk, size);
2096 i_size_write(md->bdev->bd_inode, (loff_t)size << SECTOR_SHIFT);
2100 * Return 1 if the queue has a compulsory merge_bvec_fn function.
2102 * If this function returns 0, then the device is either a non-dm
2103 * device without a merge_bvec_fn, or it is a dm device that is
2104 * able to split any bios it receives that are too big.
2106 int dm_queue_merge_is_compulsory(struct request_queue *q)
2108 struct mapped_device *dev_md;
2110 if (!q->merge_bvec_fn)
2113 if (q->make_request_fn == dm_request) {
2114 dev_md = q->queuedata;
2115 if (test_bit(DMF_MERGE_IS_OPTIONAL, &dev_md->flags))
2122 static int dm_device_merge_is_compulsory(struct dm_target *ti,
2123 struct dm_dev *dev, sector_t start,
2124 sector_t len, void *data)
2126 struct block_device *bdev = dev->bdev;
2127 struct request_queue *q = bdev_get_queue(bdev);
2129 return dm_queue_merge_is_compulsory(q);
2133 * Return 1 if it is acceptable to ignore merge_bvec_fn based
2134 * on the properties of the underlying devices.
2136 static int dm_table_merge_is_optional(struct dm_table *table)
2139 struct dm_target *ti;
2141 while (i < dm_table_get_num_targets(table)) {
2142 ti = dm_table_get_target(table, i++);
2144 if (ti->type->iterate_devices &&
2145 ti->type->iterate_devices(ti, dm_device_merge_is_compulsory, NULL))
2153 * Returns old map, which caller must destroy.
2155 static struct dm_table *__bind(struct mapped_device *md, struct dm_table *t,
2156 struct queue_limits *limits)
2158 struct dm_table *old_map;
2159 struct request_queue *q = md->queue;
2161 int merge_is_optional;
2163 size = dm_table_get_size(t);
2166 * Wipe any geometry if the size of the table changed.
2168 if (size != dm_get_size(md))
2169 memset(&md->geometry, 0, sizeof(md->geometry));
2171 __set_size(md, size);
2173 dm_table_event_callback(t, event_callback, md);
2176 * The queue hasn't been stopped yet, if the old table type wasn't
2177 * for request-based during suspension. So stop it to prevent
2178 * I/O mapping before resume.
2179 * This must be done before setting the queue restrictions,
2180 * because request-based dm may be run just after the setting.
2182 if (dm_table_request_based(t) && !blk_queue_stopped(q))
2185 __bind_mempools(md, t);
2187 merge_is_optional = dm_table_merge_is_optional(t);
2190 rcu_assign_pointer(md->map, t);
2191 md->immutable_target_type = dm_table_get_immutable_target_type(t);
2193 dm_table_set_restrictions(t, q, limits);
2194 if (merge_is_optional)
2195 set_bit(DMF_MERGE_IS_OPTIONAL, &md->flags);
2197 clear_bit(DMF_MERGE_IS_OPTIONAL, &md->flags);
2204 * Returns unbound table for the caller to free.
2206 static struct dm_table *__unbind(struct mapped_device *md)
2208 struct dm_table *map = md->map;
2213 dm_table_event_callback(map, NULL, NULL);
2214 RCU_INIT_POINTER(md->map, NULL);
2221 * Constructor for a new device.
2223 int dm_create(int minor, struct mapped_device **result)
2225 struct mapped_device *md;
2227 md = alloc_dev(minor);
2238 * Functions to manage md->type.
2239 * All are required to hold md->type_lock.
2241 void dm_lock_md_type(struct mapped_device *md)
2243 mutex_lock(&md->type_lock);
2246 void dm_unlock_md_type(struct mapped_device *md)
2248 mutex_unlock(&md->type_lock);
2251 void dm_set_md_type(struct mapped_device *md, unsigned type)
2253 BUG_ON(!mutex_is_locked(&md->type_lock));
2257 unsigned dm_get_md_type(struct mapped_device *md)
2259 BUG_ON(!mutex_is_locked(&md->type_lock));
2263 struct target_type *dm_get_immutable_target_type(struct mapped_device *md)
2265 return md->immutable_target_type;
2269 * The queue_limits are only valid as long as you have a reference
2272 struct queue_limits *dm_get_queue_limits(struct mapped_device *md)
2274 BUG_ON(!atomic_read(&md->holders));
2275 return &md->queue->limits;
2277 EXPORT_SYMBOL_GPL(dm_get_queue_limits);
2280 * Fully initialize a request-based queue (->elevator, ->request_fn, etc).
2282 static int dm_init_request_based_queue(struct mapped_device *md)
2284 struct request_queue *q = NULL;
2286 if (md->queue->elevator)
2289 /* Fully initialize the queue */
2290 q = blk_init_allocated_queue(md->queue, dm_request_fn, NULL);
2295 dm_init_md_queue(md);
2296 blk_queue_softirq_done(md->queue, dm_softirq_done);
2297 blk_queue_prep_rq(md->queue, dm_prep_fn);
2298 blk_queue_lld_busy(md->queue, dm_lld_busy);
2300 elv_register_queue(md->queue);
2306 * Setup the DM device's queue based on md's type
2308 int dm_setup_md_queue(struct mapped_device *md)
2310 if ((dm_get_md_type(md) == DM_TYPE_REQUEST_BASED) &&
2311 !dm_init_request_based_queue(md)) {
2312 DMWARN("Cannot initialize queue for request-based mapped device");
2319 static struct mapped_device *dm_find_md(dev_t dev)
2321 struct mapped_device *md;
2322 unsigned minor = MINOR(dev);
2324 if (MAJOR(dev) != _major || minor >= (1 << MINORBITS))
2327 spin_lock(&_minor_lock);
2329 md = idr_find(&_minor_idr, minor);
2330 if (md && (md == MINOR_ALLOCED ||
2331 (MINOR(disk_devt(dm_disk(md))) != minor) ||
2332 dm_deleting_md(md) ||
2333 test_bit(DMF_FREEING, &md->flags))) {
2339 spin_unlock(&_minor_lock);
2344 struct mapped_device *dm_get_md(dev_t dev)
2346 struct mapped_device *md = dm_find_md(dev);
2353 EXPORT_SYMBOL_GPL(dm_get_md);
2355 void *dm_get_mdptr(struct mapped_device *md)
2357 return md->interface_ptr;
2360 void dm_set_mdptr(struct mapped_device *md, void *ptr)
2362 md->interface_ptr = ptr;
2365 void dm_get(struct mapped_device *md)
2367 atomic_inc(&md->holders);
2368 BUG_ON(test_bit(DMF_FREEING, &md->flags));
2371 const char *dm_device_name(struct mapped_device *md)
2375 EXPORT_SYMBOL_GPL(dm_device_name);
2377 static void __dm_destroy(struct mapped_device *md, bool wait)
2379 struct dm_table *map;
2384 spin_lock(&_minor_lock);
2385 map = dm_get_live_table(md, &srcu_idx);
2386 idr_replace(&_minor_idr, MINOR_ALLOCED, MINOR(disk_devt(dm_disk(md))));
2387 set_bit(DMF_FREEING, &md->flags);
2388 spin_unlock(&_minor_lock);
2390 if (!dm_suspended_md(md)) {
2391 dm_table_presuspend_targets(map);
2392 dm_table_postsuspend_targets(map);
2395 /* dm_put_live_table must be before msleep, otherwise deadlock is possible */
2396 dm_put_live_table(md, srcu_idx);
2399 * Rare, but there may be I/O requests still going to complete,
2400 * for example. Wait for all references to disappear.
2401 * No one should increment the reference count of the mapped_device,
2402 * after the mapped_device state becomes DMF_FREEING.
2405 while (atomic_read(&md->holders))
2407 else if (atomic_read(&md->holders))
2408 DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)",
2409 dm_device_name(md), atomic_read(&md->holders));
2412 dm_table_destroy(__unbind(md));
2416 void dm_destroy(struct mapped_device *md)
2418 __dm_destroy(md, true);
2421 void dm_destroy_immediate(struct mapped_device *md)
2423 __dm_destroy(md, false);
2426 void dm_put(struct mapped_device *md)
2428 atomic_dec(&md->holders);
2430 EXPORT_SYMBOL_GPL(dm_put);
2432 static int dm_wait_for_completion(struct mapped_device *md, int interruptible)
2435 DECLARE_WAITQUEUE(wait, current);
2437 add_wait_queue(&md->wait, &wait);
2440 set_current_state(interruptible);
2442 if (!md_in_flight(md))
2445 if (interruptible == TASK_INTERRUPTIBLE &&
2446 signal_pending(current)) {
2453 set_current_state(TASK_RUNNING);
2455 remove_wait_queue(&md->wait, &wait);
2461 * Process the deferred bios
2463 static void dm_wq_work(struct work_struct *work)
2465 struct mapped_device *md = container_of(work, struct mapped_device,
2469 struct dm_table *map;
2471 map = dm_get_live_table(md, &srcu_idx);
2473 while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
2474 spin_lock_irq(&md->deferred_lock);
2475 c = bio_list_pop(&md->deferred);
2476 spin_unlock_irq(&md->deferred_lock);
2481 if (dm_request_based(md))
2482 generic_make_request(c);
2484 __split_and_process_bio(md, map, c);
2487 dm_put_live_table(md, srcu_idx);
2490 static void dm_queue_flush(struct mapped_device *md)
2492 clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2493 smp_mb__after_clear_bit();
2494 queue_work(md->wq, &md->work);
2498 * Swap in a new table, returning the old one for the caller to destroy.
2500 struct dm_table *dm_swap_table(struct mapped_device *md, struct dm_table *table)
2502 struct dm_table *live_map = NULL, *map = ERR_PTR(-EINVAL);
2503 struct queue_limits limits;
2506 mutex_lock(&md->suspend_lock);
2508 /* device must be suspended */
2509 if (!dm_suspended_md(md))
2513 * If the new table has no data devices, retain the existing limits.
2514 * This helps multipath with queue_if_no_path if all paths disappear,
2515 * then new I/O is queued based on these limits, and then some paths
2518 if (dm_table_has_no_data_devices(table)) {
2519 live_map = dm_get_live_table_fast(md);
2521 limits = md->queue->limits;
2522 dm_put_live_table_fast(md);
2526 r = dm_calculate_queue_limits(table, &limits);
2533 map = __bind(md, table, &limits);
2536 mutex_unlock(&md->suspend_lock);
2541 * Functions to lock and unlock any filesystem running on the
2544 static int lock_fs(struct mapped_device *md)
2548 WARN_ON(md->frozen_sb);
2550 md->frozen_sb = freeze_bdev(md->bdev);
2551 if (IS_ERR(md->frozen_sb)) {
2552 r = PTR_ERR(md->frozen_sb);
2553 md->frozen_sb = NULL;
2557 set_bit(DMF_FROZEN, &md->flags);
2562 static void unlock_fs(struct mapped_device *md)
2564 if (!test_bit(DMF_FROZEN, &md->flags))
2567 thaw_bdev(md->bdev, md->frozen_sb);
2568 md->frozen_sb = NULL;
2569 clear_bit(DMF_FROZEN, &md->flags);
2573 * We need to be able to change a mapping table under a mounted
2574 * filesystem. For example we might want to move some data in
2575 * the background. Before the table can be swapped with
2576 * dm_bind_table, dm_suspend must be called to flush any in
2577 * flight bios and ensure that any further io gets deferred.
2580 * Suspend mechanism in request-based dm.
2582 * 1. Flush all I/Os by lock_fs() if needed.
2583 * 2. Stop dispatching any I/O by stopping the request_queue.
2584 * 3. Wait for all in-flight I/Os to be completed or requeued.
2586 * To abort suspend, start the request_queue.
2588 int dm_suspend(struct mapped_device *md, unsigned suspend_flags)
2590 struct dm_table *map = NULL;
2592 int do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG ? 1 : 0;
2593 int noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG ? 1 : 0;
2595 mutex_lock(&md->suspend_lock);
2597 if (dm_suspended_md(md)) {
2605 * DMF_NOFLUSH_SUSPENDING must be set before presuspend.
2606 * This flag is cleared before dm_suspend returns.
2609 set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2611 /* This does not get reverted if there's an error later. */
2612 dm_table_presuspend_targets(map);
2615 * Flush I/O to the device.
2616 * Any I/O submitted after lock_fs() may not be flushed.
2617 * noflush takes precedence over do_lockfs.
2618 * (lock_fs() flushes I/Os and waits for them to complete.)
2620 if (!noflush && do_lockfs) {
2627 * Here we must make sure that no processes are submitting requests
2628 * to target drivers i.e. no one may be executing
2629 * __split_and_process_bio. This is called from dm_request and
2632 * To get all processes out of __split_and_process_bio in dm_request,
2633 * we take the write lock. To prevent any process from reentering
2634 * __split_and_process_bio from dm_request and quiesce the thread
2635 * (dm_wq_work), we set BMF_BLOCK_IO_FOR_SUSPEND and call
2636 * flush_workqueue(md->wq).
2638 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2639 synchronize_srcu(&md->io_barrier);
2642 * Stop md->queue before flushing md->wq in case request-based
2643 * dm defers requests to md->wq from md->queue.
2645 if (dm_request_based(md))
2646 stop_queue(md->queue);
2648 flush_workqueue(md->wq);
2651 * At this point no more requests are entering target request routines.
2652 * We call dm_wait_for_completion to wait for all existing requests
2655 r = dm_wait_for_completion(md, TASK_INTERRUPTIBLE);
2658 clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2659 synchronize_srcu(&md->io_barrier);
2661 /* were we interrupted ? */
2665 if (dm_request_based(md))
2666 start_queue(md->queue);
2669 goto out_unlock; /* pushback list is already flushed, so skip flush */
2673 * If dm_wait_for_completion returned 0, the device is completely
2674 * quiescent now. There is no request-processing activity. All new
2675 * requests are being added to md->deferred list.
2678 set_bit(DMF_SUSPENDED, &md->flags);
2680 dm_table_postsuspend_targets(map);
2683 mutex_unlock(&md->suspend_lock);
2687 int dm_resume(struct mapped_device *md)
2690 struct dm_table *map = NULL;
2692 mutex_lock(&md->suspend_lock);
2693 if (!dm_suspended_md(md))
2697 if (!map || !dm_table_get_size(map))
2700 r = dm_table_resume_targets(map);
2707 * Flushing deferred I/Os must be done after targets are resumed
2708 * so that mapping of targets can work correctly.
2709 * Request-based dm is queueing the deferred I/Os in its request_queue.
2711 if (dm_request_based(md))
2712 start_queue(md->queue);
2716 clear_bit(DMF_SUSPENDED, &md->flags);
2720 mutex_unlock(&md->suspend_lock);
2726 * Internal suspend/resume works like userspace-driven suspend. It waits
2727 * until all bios finish and prevents issuing new bios to the target drivers.
2728 * It may be used only from the kernel.
2730 * Internal suspend holds md->suspend_lock, which prevents interaction with
2731 * userspace-driven suspend.
2734 void dm_internal_suspend(struct mapped_device *md)
2736 mutex_lock(&md->suspend_lock);
2737 if (dm_suspended_md(md))
2740 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2741 synchronize_srcu(&md->io_barrier);
2742 flush_workqueue(md->wq);
2743 dm_wait_for_completion(md, TASK_UNINTERRUPTIBLE);
2746 void dm_internal_resume(struct mapped_device *md)
2748 if (dm_suspended_md(md))
2754 mutex_unlock(&md->suspend_lock);
2757 /*-----------------------------------------------------------------
2758 * Event notification.
2759 *---------------------------------------------------------------*/
2760 int dm_kobject_uevent(struct mapped_device *md, enum kobject_action action,
2763 char udev_cookie[DM_COOKIE_LENGTH];
2764 char *envp[] = { udev_cookie, NULL };
2767 return kobject_uevent(&disk_to_dev(md->disk)->kobj, action);
2769 snprintf(udev_cookie, DM_COOKIE_LENGTH, "%s=%u",
2770 DM_COOKIE_ENV_VAR_NAME, cookie);
2771 return kobject_uevent_env(&disk_to_dev(md->disk)->kobj,
2776 uint32_t dm_next_uevent_seq(struct mapped_device *md)
2778 return atomic_add_return(1, &md->uevent_seq);
2781 uint32_t dm_get_event_nr(struct mapped_device *md)
2783 return atomic_read(&md->event_nr);
2786 int dm_wait_event(struct mapped_device *md, int event_nr)
2788 return wait_event_interruptible(md->eventq,
2789 (event_nr != atomic_read(&md->event_nr)));
2792 void dm_uevent_add(struct mapped_device *md, struct list_head *elist)
2794 unsigned long flags;
2796 spin_lock_irqsave(&md->uevent_lock, flags);
2797 list_add(elist, &md->uevent_list);
2798 spin_unlock_irqrestore(&md->uevent_lock, flags);
2802 * The gendisk is only valid as long as you have a reference
2805 struct gendisk *dm_disk(struct mapped_device *md)
2810 struct kobject *dm_kobject(struct mapped_device *md)
2812 return &md->kobj_holder.kobj;
2815 struct mapped_device *dm_get_from_kobject(struct kobject *kobj)
2817 struct mapped_device *md;
2819 md = container_of(kobj, struct mapped_device, kobj_holder.kobj);
2821 if (test_bit(DMF_FREEING, &md->flags) ||
2829 int dm_suspended_md(struct mapped_device *md)
2831 return test_bit(DMF_SUSPENDED, &md->flags);
2834 int dm_test_deferred_remove_flag(struct mapped_device *md)
2836 return test_bit(DMF_DEFERRED_REMOVE, &md->flags);
2839 int dm_suspended(struct dm_target *ti)
2841 return dm_suspended_md(dm_table_get_md(ti->table));
2843 EXPORT_SYMBOL_GPL(dm_suspended);
2845 int dm_noflush_suspending(struct dm_target *ti)
2847 return __noflush_suspending(dm_table_get_md(ti->table));
2849 EXPORT_SYMBOL_GPL(dm_noflush_suspending);
2851 struct dm_md_mempools *dm_alloc_md_mempools(unsigned type, unsigned integrity, unsigned per_bio_data_size)
2853 struct dm_md_mempools *pools = kzalloc(sizeof(*pools), GFP_KERNEL);
2854 struct kmem_cache *cachep;
2855 unsigned int pool_size;
2856 unsigned int front_pad;
2861 if (type == DM_TYPE_BIO_BASED) {
2863 pool_size = dm_get_reserved_bio_based_ios();
2864 front_pad = roundup(per_bio_data_size, __alignof__(struct dm_target_io)) + offsetof(struct dm_target_io, clone);
2865 } else if (type == DM_TYPE_REQUEST_BASED) {
2866 cachep = _rq_tio_cache;
2867 pool_size = dm_get_reserved_rq_based_ios();
2868 front_pad = offsetof(struct dm_rq_clone_bio_info, clone);
2869 /* per_bio_data_size is not used. See __bind_mempools(). */
2870 WARN_ON(per_bio_data_size != 0);
2874 pools->io_pool = mempool_create_slab_pool(pool_size, cachep);
2875 if (!pools->io_pool)
2878 pools->bs = bioset_create(pool_size, front_pad);
2882 if (integrity && bioset_integrity_create(pools->bs, pool_size))
2888 dm_free_md_mempools(pools);
2893 void dm_free_md_mempools(struct dm_md_mempools *pools)
2899 mempool_destroy(pools->io_pool);
2902 bioset_free(pools->bs);
2907 static const struct block_device_operations dm_blk_dops = {
2908 .open = dm_blk_open,
2909 .release = dm_blk_close,
2910 .ioctl = dm_blk_ioctl,
2911 .getgeo = dm_blk_getgeo,
2912 .owner = THIS_MODULE
2918 module_init(dm_init);
2919 module_exit(dm_exit);
2921 module_param(major, uint, 0);
2922 MODULE_PARM_DESC(major, "The major number of the device mapper");
2924 module_param(reserved_bio_based_ios, uint, S_IRUGO | S_IWUSR);
2925 MODULE_PARM_DESC(reserved_bio_based_ios, "Reserved IOs in bio-based mempools");
2927 module_param(reserved_rq_based_ios, uint, S_IRUGO | S_IWUSR);
2928 MODULE_PARM_DESC(reserved_rq_based_ios, "Reserved IOs in request-based mempools");
2930 MODULE_DESCRIPTION(DM_NAME " driver");
2931 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
2932 MODULE_LICENSE("GPL");