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);
1113 static void __map_bio(struct dm_target_io *tio)
1117 struct mapped_device *md;
1118 struct bio *clone = &tio->clone;
1119 struct dm_target *ti = tio->ti;
1121 clone->bi_end_io = clone_endio;
1124 * Map the clone. If r == 0 we don't need to do
1125 * anything, the target has assumed ownership of
1128 atomic_inc(&tio->io->io_count);
1129 sector = clone->bi_iter.bi_sector;
1130 r = ti->type->map(ti, clone);
1131 if (r == DM_MAPIO_REMAPPED) {
1132 /* the bio has been remapped so dispatch it */
1134 trace_block_bio_remap(bdev_get_queue(clone->bi_bdev), clone,
1135 tio->io->bio->bi_bdev->bd_dev, sector);
1137 generic_make_request(clone);
1138 } else if (r < 0 || r == DM_MAPIO_REQUEUE) {
1139 /* error the io and bail out, or requeue it if needed */
1141 dec_pending(tio->io, r);
1144 DMWARN("unimplemented target map return value: %d", r);
1150 struct mapped_device *md;
1151 struct dm_table *map;
1155 sector_t sector_count;
1158 static void bio_setup_sector(struct bio *bio, sector_t sector, sector_t len)
1160 bio->bi_iter.bi_sector = sector;
1161 bio->bi_iter.bi_size = to_bytes(len);
1165 * Creates a bio that consists of range of complete bvecs.
1167 static void clone_bio(struct dm_target_io *tio, struct bio *bio,
1168 sector_t sector, unsigned len)
1170 struct bio *clone = &tio->clone;
1172 __bio_clone_fast(clone, bio);
1174 if (bio_integrity(bio))
1175 bio_integrity_clone(clone, bio, GFP_NOIO);
1177 bio_advance(clone, to_bytes(sector - clone->bi_iter.bi_sector));
1178 clone->bi_iter.bi_size = to_bytes(len);
1180 if (bio_integrity(bio))
1181 bio_integrity_trim(clone, 0, len);
1184 static struct dm_target_io *alloc_tio(struct clone_info *ci,
1185 struct dm_target *ti, int nr_iovecs,
1186 unsigned target_bio_nr)
1188 struct dm_target_io *tio;
1191 clone = bio_alloc_bioset(GFP_NOIO, nr_iovecs, ci->md->bs);
1192 tio = container_of(clone, struct dm_target_io, clone);
1196 tio->target_bio_nr = target_bio_nr;
1201 static void __clone_and_map_simple_bio(struct clone_info *ci,
1202 struct dm_target *ti,
1203 unsigned target_bio_nr, sector_t len)
1205 struct dm_target_io *tio = alloc_tio(ci, ti, ci->bio->bi_max_vecs, target_bio_nr);
1206 struct bio *clone = &tio->clone;
1209 * Discard requests require the bio's inline iovecs be initialized.
1210 * ci->bio->bi_max_vecs is BIO_INLINE_VECS anyway, for both flush
1211 * and discard, so no need for concern about wasted bvec allocations.
1213 __bio_clone_fast(clone, ci->bio);
1215 bio_setup_sector(clone, ci->sector, len);
1220 static void __send_duplicate_bios(struct clone_info *ci, struct dm_target *ti,
1221 unsigned num_bios, sector_t len)
1223 unsigned target_bio_nr;
1225 for (target_bio_nr = 0; target_bio_nr < num_bios; target_bio_nr++)
1226 __clone_and_map_simple_bio(ci, ti, target_bio_nr, len);
1229 static int __send_empty_flush(struct clone_info *ci)
1231 unsigned target_nr = 0;
1232 struct dm_target *ti;
1234 BUG_ON(bio_has_data(ci->bio));
1235 while ((ti = dm_table_get_target(ci->map, target_nr++)))
1236 __send_duplicate_bios(ci, ti, ti->num_flush_bios, 0);
1241 static void __clone_and_map_data_bio(struct clone_info *ci, struct dm_target *ti,
1242 sector_t sector, unsigned len)
1244 struct bio *bio = ci->bio;
1245 struct dm_target_io *tio;
1246 unsigned target_bio_nr;
1247 unsigned num_target_bios = 1;
1250 * Does the target want to receive duplicate copies of the bio?
1252 if (bio_data_dir(bio) == WRITE && ti->num_write_bios)
1253 num_target_bios = ti->num_write_bios(ti, bio);
1255 for (target_bio_nr = 0; target_bio_nr < num_target_bios; target_bio_nr++) {
1256 tio = alloc_tio(ci, ti, 0, target_bio_nr);
1257 clone_bio(tio, bio, sector, len);
1262 typedef unsigned (*get_num_bios_fn)(struct dm_target *ti);
1264 static unsigned get_num_discard_bios(struct dm_target *ti)
1266 return ti->num_discard_bios;
1269 static unsigned get_num_write_same_bios(struct dm_target *ti)
1271 return ti->num_write_same_bios;
1274 typedef bool (*is_split_required_fn)(struct dm_target *ti);
1276 static bool is_split_required_for_discard(struct dm_target *ti)
1278 return ti->split_discard_bios;
1281 static int __send_changing_extent_only(struct clone_info *ci,
1282 get_num_bios_fn get_num_bios,
1283 is_split_required_fn is_split_required)
1285 struct dm_target *ti;
1290 ti = dm_table_find_target(ci->map, ci->sector);
1291 if (!dm_target_is_valid(ti))
1295 * Even though the device advertised support for this type of
1296 * request, that does not mean every target supports it, and
1297 * reconfiguration might also have changed that since the
1298 * check was performed.
1300 num_bios = get_num_bios ? get_num_bios(ti) : 0;
1304 if (is_split_required && !is_split_required(ti))
1305 len = min(ci->sector_count, max_io_len_target_boundary(ci->sector, ti));
1307 len = min(ci->sector_count, max_io_len(ci->sector, ti));
1309 __send_duplicate_bios(ci, ti, num_bios, len);
1312 } while (ci->sector_count -= len);
1317 static int __send_discard(struct clone_info *ci)
1319 return __send_changing_extent_only(ci, get_num_discard_bios,
1320 is_split_required_for_discard);
1323 static int __send_write_same(struct clone_info *ci)
1325 return __send_changing_extent_only(ci, get_num_write_same_bios, NULL);
1329 * Select the correct strategy for processing a non-flush bio.
1331 static int __split_and_process_non_flush(struct clone_info *ci)
1333 struct bio *bio = ci->bio;
1334 struct dm_target *ti;
1337 if (unlikely(bio->bi_rw & REQ_DISCARD))
1338 return __send_discard(ci);
1339 else if (unlikely(bio->bi_rw & REQ_WRITE_SAME))
1340 return __send_write_same(ci);
1342 ti = dm_table_find_target(ci->map, ci->sector);
1343 if (!dm_target_is_valid(ti))
1346 len = min_t(sector_t, max_io_len(ci->sector, ti), ci->sector_count);
1348 __clone_and_map_data_bio(ci, ti, ci->sector, len);
1351 ci->sector_count -= len;
1357 * Entry point to split a bio into clones and submit them to the targets.
1359 static void __split_and_process_bio(struct mapped_device *md,
1360 struct dm_table *map, struct bio *bio)
1362 struct clone_info ci;
1365 if (unlikely(!map)) {
1372 ci.io = alloc_io(md);
1374 atomic_set(&ci.io->io_count, 1);
1377 spin_lock_init(&ci.io->endio_lock);
1378 ci.sector = bio->bi_iter.bi_sector;
1380 start_io_acct(ci.io);
1382 if (bio->bi_rw & REQ_FLUSH) {
1383 ci.bio = &ci.md->flush_bio;
1384 ci.sector_count = 0;
1385 error = __send_empty_flush(&ci);
1386 /* dec_pending submits any data associated with flush */
1389 ci.sector_count = bio_sectors(bio);
1390 while (ci.sector_count && !error)
1391 error = __split_and_process_non_flush(&ci);
1394 /* drop the extra reference count */
1395 dec_pending(ci.io, error);
1397 /*-----------------------------------------------------------------
1399 *---------------------------------------------------------------*/
1401 static int dm_merge_bvec(struct request_queue *q,
1402 struct bvec_merge_data *bvm,
1403 struct bio_vec *biovec)
1405 struct mapped_device *md = q->queuedata;
1406 struct dm_table *map = dm_get_live_table_fast(md);
1407 struct dm_target *ti;
1408 sector_t max_sectors;
1414 ti = dm_table_find_target(map, bvm->bi_sector);
1415 if (!dm_target_is_valid(ti))
1419 * Find maximum amount of I/O that won't need splitting
1421 max_sectors = min(max_io_len(bvm->bi_sector, ti),
1422 (sector_t) BIO_MAX_SECTORS);
1423 max_size = (max_sectors << SECTOR_SHIFT) - bvm->bi_size;
1428 * merge_bvec_fn() returns number of bytes
1429 * it can accept at this offset
1430 * max is precomputed maximal io size
1432 if (max_size && ti->type->merge)
1433 max_size = ti->type->merge(ti, bvm, biovec, max_size);
1435 * If the target doesn't support merge method and some of the devices
1436 * provided their merge_bvec method (we know this by looking at
1437 * queue_max_hw_sectors), then we can't allow bios with multiple vector
1438 * entries. So always set max_size to 0, and the code below allows
1441 else if (queue_max_hw_sectors(q) <= PAGE_SIZE >> 9)
1446 dm_put_live_table_fast(md);
1448 * Always allow an entire first page
1450 if (max_size <= biovec->bv_len && !(bvm->bi_size >> SECTOR_SHIFT))
1451 max_size = biovec->bv_len;
1457 * The request function that just remaps the bio built up by
1460 static void _dm_request(struct request_queue *q, struct bio *bio)
1462 int rw = bio_data_dir(bio);
1463 struct mapped_device *md = q->queuedata;
1466 struct dm_table *map;
1468 map = dm_get_live_table(md, &srcu_idx);
1470 cpu = part_stat_lock();
1471 part_stat_inc(cpu, &dm_disk(md)->part0, ios[rw]);
1472 part_stat_add(cpu, &dm_disk(md)->part0, sectors[rw], bio_sectors(bio));
1475 /* if we're suspended, we have to queue this io for later */
1476 if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))) {
1477 dm_put_live_table(md, srcu_idx);
1479 if (bio_rw(bio) != READA)
1486 __split_and_process_bio(md, map, bio);
1487 dm_put_live_table(md, srcu_idx);
1491 int dm_request_based(struct mapped_device *md)
1493 return blk_queue_stackable(md->queue);
1496 static void dm_request(struct request_queue *q, struct bio *bio)
1498 struct mapped_device *md = q->queuedata;
1500 if (dm_request_based(md))
1501 blk_queue_bio(q, bio);
1503 _dm_request(q, bio);
1506 void dm_dispatch_request(struct request *rq)
1510 if (blk_queue_io_stat(rq->q))
1511 rq->cmd_flags |= REQ_IO_STAT;
1513 rq->start_time = jiffies;
1514 r = blk_insert_cloned_request(rq->q, rq);
1516 dm_complete_request(rq, r);
1518 EXPORT_SYMBOL_GPL(dm_dispatch_request);
1520 static int dm_rq_bio_constructor(struct bio *bio, struct bio *bio_orig,
1523 struct dm_rq_target_io *tio = data;
1524 struct dm_rq_clone_bio_info *info =
1525 container_of(bio, struct dm_rq_clone_bio_info, clone);
1527 info->orig = bio_orig;
1529 bio->bi_end_io = end_clone_bio;
1534 static int setup_clone(struct request *clone, struct request *rq,
1535 struct dm_rq_target_io *tio)
1539 r = blk_rq_prep_clone(clone, rq, tio->md->bs, GFP_ATOMIC,
1540 dm_rq_bio_constructor, tio);
1544 clone->cmd = rq->cmd;
1545 clone->cmd_len = rq->cmd_len;
1546 clone->sense = rq->sense;
1547 clone->end_io = end_clone_request;
1548 clone->end_io_data = tio;
1553 static struct request *clone_rq(struct request *rq, struct mapped_device *md,
1556 struct request *clone;
1557 struct dm_rq_target_io *tio;
1559 tio = alloc_rq_tio(md, gfp_mask);
1567 memset(&tio->info, 0, sizeof(tio->info));
1569 clone = &tio->clone;
1570 if (setup_clone(clone, rq, tio)) {
1580 * Called with the queue lock held.
1582 static int dm_prep_fn(struct request_queue *q, struct request *rq)
1584 struct mapped_device *md = q->queuedata;
1585 struct request *clone;
1587 if (unlikely(rq->special)) {
1588 DMWARN("Already has something in rq->special.");
1589 return BLKPREP_KILL;
1592 clone = clone_rq(rq, md, GFP_ATOMIC);
1594 return BLKPREP_DEFER;
1596 rq->special = clone;
1597 rq->cmd_flags |= REQ_DONTPREP;
1604 * 0 : the request has been processed (not requeued)
1605 * !0 : the request has been requeued
1607 static int map_request(struct dm_target *ti, struct request *clone,
1608 struct mapped_device *md)
1610 int r, requeued = 0;
1611 struct dm_rq_target_io *tio = clone->end_io_data;
1614 r = ti->type->map_rq(ti, clone, &tio->info);
1616 case DM_MAPIO_SUBMITTED:
1617 /* The target has taken the I/O to submit by itself later */
1619 case DM_MAPIO_REMAPPED:
1620 /* The target has remapped the I/O so dispatch it */
1621 trace_block_rq_remap(clone->q, clone, disk_devt(dm_disk(md)),
1622 blk_rq_pos(tio->orig));
1623 dm_dispatch_request(clone);
1625 case DM_MAPIO_REQUEUE:
1626 /* The target wants to requeue the I/O */
1627 dm_requeue_unmapped_request(clone);
1632 DMWARN("unimplemented target map return value: %d", r);
1636 /* The target wants to complete the I/O */
1637 dm_kill_unmapped_request(clone, r);
1644 static struct request *dm_start_request(struct mapped_device *md, struct request *orig)
1646 struct request *clone;
1648 blk_start_request(orig);
1649 clone = orig->special;
1650 atomic_inc(&md->pending[rq_data_dir(clone)]);
1653 * Hold the md reference here for the in-flight I/O.
1654 * We can't rely on the reference count by device opener,
1655 * because the device may be closed during the request completion
1656 * when all bios are completed.
1657 * See the comment in rq_completed() too.
1665 * q->request_fn for request-based dm.
1666 * Called with the queue lock held.
1668 static void dm_request_fn(struct request_queue *q)
1670 struct mapped_device *md = q->queuedata;
1672 struct dm_table *map = dm_get_live_table(md, &srcu_idx);
1673 struct dm_target *ti;
1674 struct request *rq, *clone;
1678 * For suspend, check blk_queue_stopped() and increment
1679 * ->pending within a single queue_lock not to increment the
1680 * number of in-flight I/Os after the queue is stopped in
1683 while (!blk_queue_stopped(q)) {
1684 rq = blk_peek_request(q);
1688 /* always use block 0 to find the target for flushes for now */
1690 if (!(rq->cmd_flags & REQ_FLUSH))
1691 pos = blk_rq_pos(rq);
1693 ti = dm_table_find_target(map, pos);
1694 if (!dm_target_is_valid(ti)) {
1696 * Must perform setup, that dm_done() requires,
1697 * before calling dm_kill_unmapped_request
1699 DMERR_LIMIT("request attempted access beyond the end of device");
1700 clone = dm_start_request(md, rq);
1701 dm_kill_unmapped_request(clone, -EIO);
1705 if (ti->type->busy && ti->type->busy(ti))
1708 clone = dm_start_request(md, rq);
1710 spin_unlock(q->queue_lock);
1711 if (map_request(ti, clone, md))
1714 BUG_ON(!irqs_disabled());
1715 spin_lock(q->queue_lock);
1721 BUG_ON(!irqs_disabled());
1722 spin_lock(q->queue_lock);
1725 blk_delay_queue(q, HZ / 10);
1727 dm_put_live_table(md, srcu_idx);
1730 int dm_underlying_device_busy(struct request_queue *q)
1732 return blk_lld_busy(q);
1734 EXPORT_SYMBOL_GPL(dm_underlying_device_busy);
1736 static int dm_lld_busy(struct request_queue *q)
1739 struct mapped_device *md = q->queuedata;
1740 struct dm_table *map = dm_get_live_table_fast(md);
1742 if (!map || test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))
1745 r = dm_table_any_busy_target(map);
1747 dm_put_live_table_fast(md);
1752 static int dm_any_congested(void *congested_data, int bdi_bits)
1755 struct mapped_device *md = congested_data;
1756 struct dm_table *map;
1758 if (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
1759 map = dm_get_live_table_fast(md);
1762 * Request-based dm cares about only own queue for
1763 * the query about congestion status of request_queue
1765 if (dm_request_based(md))
1766 r = md->queue->backing_dev_info.state &
1769 r = dm_table_any_congested(map, bdi_bits);
1771 dm_put_live_table_fast(md);
1777 /*-----------------------------------------------------------------
1778 * An IDR is used to keep track of allocated minor numbers.
1779 *---------------------------------------------------------------*/
1780 static void free_minor(int minor)
1782 spin_lock(&_minor_lock);
1783 idr_remove(&_minor_idr, minor);
1784 spin_unlock(&_minor_lock);
1788 * See if the device with a specific minor # is free.
1790 static int specific_minor(int minor)
1794 if (minor >= (1 << MINORBITS))
1797 idr_preload(GFP_KERNEL);
1798 spin_lock(&_minor_lock);
1800 r = idr_alloc(&_minor_idr, MINOR_ALLOCED, minor, minor + 1, GFP_NOWAIT);
1802 spin_unlock(&_minor_lock);
1805 return r == -ENOSPC ? -EBUSY : r;
1809 static int next_free_minor(int *minor)
1813 idr_preload(GFP_KERNEL);
1814 spin_lock(&_minor_lock);
1816 r = idr_alloc(&_minor_idr, MINOR_ALLOCED, 0, 1 << MINORBITS, GFP_NOWAIT);
1818 spin_unlock(&_minor_lock);
1826 static const struct block_device_operations dm_blk_dops;
1828 static void dm_wq_work(struct work_struct *work);
1830 static void dm_init_md_queue(struct mapped_device *md)
1833 * Request-based dm devices cannot be stacked on top of bio-based dm
1834 * devices. The type of this dm device has not been decided yet.
1835 * The type is decided at the first table loading time.
1836 * To prevent problematic device stacking, clear the queue flag
1837 * for request stacking support until then.
1839 * This queue is new, so no concurrency on the queue_flags.
1841 queue_flag_clear_unlocked(QUEUE_FLAG_STACKABLE, md->queue);
1843 md->queue->queuedata = md;
1844 md->queue->backing_dev_info.congested_fn = dm_any_congested;
1845 md->queue->backing_dev_info.congested_data = md;
1846 blk_queue_make_request(md->queue, dm_request);
1847 blk_queue_bounce_limit(md->queue, BLK_BOUNCE_ANY);
1848 blk_queue_merge_bvec(md->queue, dm_merge_bvec);
1852 * Allocate and initialise a blank device with a given minor.
1854 static struct mapped_device *alloc_dev(int minor)
1857 struct mapped_device *md = kzalloc(sizeof(*md), GFP_KERNEL);
1861 DMWARN("unable to allocate device, out of memory.");
1865 if (!try_module_get(THIS_MODULE))
1866 goto bad_module_get;
1868 /* get a minor number for the dev */
1869 if (minor == DM_ANY_MINOR)
1870 r = next_free_minor(&minor);
1872 r = specific_minor(minor);
1876 r = init_srcu_struct(&md->io_barrier);
1878 goto bad_io_barrier;
1880 md->type = DM_TYPE_NONE;
1881 mutex_init(&md->suspend_lock);
1882 mutex_init(&md->type_lock);
1883 spin_lock_init(&md->deferred_lock);
1884 atomic_set(&md->holders, 1);
1885 atomic_set(&md->open_count, 0);
1886 atomic_set(&md->event_nr, 0);
1887 atomic_set(&md->uevent_seq, 0);
1888 INIT_LIST_HEAD(&md->uevent_list);
1889 spin_lock_init(&md->uevent_lock);
1891 md->queue = blk_alloc_queue(GFP_KERNEL);
1895 dm_init_md_queue(md);
1897 md->disk = alloc_disk(1);
1901 atomic_set(&md->pending[0], 0);
1902 atomic_set(&md->pending[1], 0);
1903 init_waitqueue_head(&md->wait);
1904 INIT_WORK(&md->work, dm_wq_work);
1905 init_waitqueue_head(&md->eventq);
1906 init_completion(&md->kobj_holder.completion);
1908 md->disk->major = _major;
1909 md->disk->first_minor = minor;
1910 md->disk->fops = &dm_blk_dops;
1911 md->disk->queue = md->queue;
1912 md->disk->private_data = md;
1913 sprintf(md->disk->disk_name, "dm-%d", minor);
1915 format_dev_t(md->name, MKDEV(_major, minor));
1917 md->wq = alloc_workqueue("kdmflush", WQ_MEM_RECLAIM, 0);
1921 md->bdev = bdget_disk(md->disk, 0);
1925 bio_init(&md->flush_bio);
1926 md->flush_bio.bi_bdev = md->bdev;
1927 md->flush_bio.bi_rw = WRITE_FLUSH;
1929 dm_stats_init(&md->stats);
1931 /* Populate the mapping, nobody knows we exist yet */
1932 spin_lock(&_minor_lock);
1933 old_md = idr_replace(&_minor_idr, md, minor);
1934 spin_unlock(&_minor_lock);
1936 BUG_ON(old_md != MINOR_ALLOCED);
1941 destroy_workqueue(md->wq);
1943 del_gendisk(md->disk);
1946 blk_cleanup_queue(md->queue);
1948 cleanup_srcu_struct(&md->io_barrier);
1952 module_put(THIS_MODULE);
1958 static void unlock_fs(struct mapped_device *md);
1960 static void free_dev(struct mapped_device *md)
1962 int minor = MINOR(disk_devt(md->disk));
1966 destroy_workqueue(md->wq);
1968 mempool_destroy(md->io_pool);
1970 bioset_free(md->bs);
1971 blk_integrity_unregister(md->disk);
1972 del_gendisk(md->disk);
1973 cleanup_srcu_struct(&md->io_barrier);
1976 spin_lock(&_minor_lock);
1977 md->disk->private_data = NULL;
1978 spin_unlock(&_minor_lock);
1981 blk_cleanup_queue(md->queue);
1982 dm_stats_cleanup(&md->stats);
1983 module_put(THIS_MODULE);
1987 static void __bind_mempools(struct mapped_device *md, struct dm_table *t)
1989 struct dm_md_mempools *p = dm_table_get_md_mempools(t);
1991 if (md->io_pool && md->bs) {
1992 /* The md already has necessary mempools. */
1993 if (dm_table_get_type(t) == DM_TYPE_BIO_BASED) {
1995 * Reload bioset because front_pad may have changed
1996 * because a different table was loaded.
1998 bioset_free(md->bs);
2001 } else if (dm_table_get_type(t) == DM_TYPE_REQUEST_BASED) {
2003 * There's no need to reload with request-based dm
2004 * because the size of front_pad doesn't change.
2005 * Note for future: If you are to reload bioset,
2006 * prep-ed requests in the queue may refer
2007 * to bio from the old bioset, so you must walk
2008 * through the queue to unprep.
2014 BUG_ON(!p || md->io_pool || md->bs);
2016 md->io_pool = p->io_pool;
2022 /* mempool bind completed, now no need any mempools in the table */
2023 dm_table_free_md_mempools(t);
2027 * Bind a table to the device.
2029 static void event_callback(void *context)
2031 unsigned long flags;
2033 struct mapped_device *md = (struct mapped_device *) context;
2035 spin_lock_irqsave(&md->uevent_lock, flags);
2036 list_splice_init(&md->uevent_list, &uevents);
2037 spin_unlock_irqrestore(&md->uevent_lock, flags);
2039 dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj);
2041 atomic_inc(&md->event_nr);
2042 wake_up(&md->eventq);
2046 * Protected by md->suspend_lock obtained by dm_swap_table().
2048 static void __set_size(struct mapped_device *md, sector_t size)
2050 set_capacity(md->disk, size);
2052 i_size_write(md->bdev->bd_inode, (loff_t)size << SECTOR_SHIFT);
2056 * Return 1 if the queue has a compulsory merge_bvec_fn function.
2058 * If this function returns 0, then the device is either a non-dm
2059 * device without a merge_bvec_fn, or it is a dm device that is
2060 * able to split any bios it receives that are too big.
2062 int dm_queue_merge_is_compulsory(struct request_queue *q)
2064 struct mapped_device *dev_md;
2066 if (!q->merge_bvec_fn)
2069 if (q->make_request_fn == dm_request) {
2070 dev_md = q->queuedata;
2071 if (test_bit(DMF_MERGE_IS_OPTIONAL, &dev_md->flags))
2078 static int dm_device_merge_is_compulsory(struct dm_target *ti,
2079 struct dm_dev *dev, sector_t start,
2080 sector_t len, void *data)
2082 struct block_device *bdev = dev->bdev;
2083 struct request_queue *q = bdev_get_queue(bdev);
2085 return dm_queue_merge_is_compulsory(q);
2089 * Return 1 if it is acceptable to ignore merge_bvec_fn based
2090 * on the properties of the underlying devices.
2092 static int dm_table_merge_is_optional(struct dm_table *table)
2095 struct dm_target *ti;
2097 while (i < dm_table_get_num_targets(table)) {
2098 ti = dm_table_get_target(table, i++);
2100 if (ti->type->iterate_devices &&
2101 ti->type->iterate_devices(ti, dm_device_merge_is_compulsory, NULL))
2109 * Returns old map, which caller must destroy.
2111 static struct dm_table *__bind(struct mapped_device *md, struct dm_table *t,
2112 struct queue_limits *limits)
2114 struct dm_table *old_map;
2115 struct request_queue *q = md->queue;
2117 int merge_is_optional;
2119 size = dm_table_get_size(t);
2122 * Wipe any geometry if the size of the table changed.
2124 if (size != dm_get_size(md))
2125 memset(&md->geometry, 0, sizeof(md->geometry));
2127 __set_size(md, size);
2129 dm_table_event_callback(t, event_callback, md);
2132 * The queue hasn't been stopped yet, if the old table type wasn't
2133 * for request-based during suspension. So stop it to prevent
2134 * I/O mapping before resume.
2135 * This must be done before setting the queue restrictions,
2136 * because request-based dm may be run just after the setting.
2138 if (dm_table_request_based(t) && !blk_queue_stopped(q))
2141 __bind_mempools(md, t);
2143 merge_is_optional = dm_table_merge_is_optional(t);
2146 rcu_assign_pointer(md->map, t);
2147 md->immutable_target_type = dm_table_get_immutable_target_type(t);
2149 dm_table_set_restrictions(t, q, limits);
2150 if (merge_is_optional)
2151 set_bit(DMF_MERGE_IS_OPTIONAL, &md->flags);
2153 clear_bit(DMF_MERGE_IS_OPTIONAL, &md->flags);
2160 * Returns unbound table for the caller to free.
2162 static struct dm_table *__unbind(struct mapped_device *md)
2164 struct dm_table *map = md->map;
2169 dm_table_event_callback(map, NULL, NULL);
2170 RCU_INIT_POINTER(md->map, NULL);
2177 * Constructor for a new device.
2179 int dm_create(int minor, struct mapped_device **result)
2181 struct mapped_device *md;
2183 md = alloc_dev(minor);
2194 * Functions to manage md->type.
2195 * All are required to hold md->type_lock.
2197 void dm_lock_md_type(struct mapped_device *md)
2199 mutex_lock(&md->type_lock);
2202 void dm_unlock_md_type(struct mapped_device *md)
2204 mutex_unlock(&md->type_lock);
2207 void dm_set_md_type(struct mapped_device *md, unsigned type)
2209 BUG_ON(!mutex_is_locked(&md->type_lock));
2213 unsigned dm_get_md_type(struct mapped_device *md)
2215 BUG_ON(!mutex_is_locked(&md->type_lock));
2219 struct target_type *dm_get_immutable_target_type(struct mapped_device *md)
2221 return md->immutable_target_type;
2225 * The queue_limits are only valid as long as you have a reference
2228 struct queue_limits *dm_get_queue_limits(struct mapped_device *md)
2230 BUG_ON(!atomic_read(&md->holders));
2231 return &md->queue->limits;
2233 EXPORT_SYMBOL_GPL(dm_get_queue_limits);
2236 * Fully initialize a request-based queue (->elevator, ->request_fn, etc).
2238 static int dm_init_request_based_queue(struct mapped_device *md)
2240 struct request_queue *q = NULL;
2242 if (md->queue->elevator)
2245 /* Fully initialize the queue */
2246 q = blk_init_allocated_queue(md->queue, dm_request_fn, NULL);
2251 dm_init_md_queue(md);
2252 blk_queue_softirq_done(md->queue, dm_softirq_done);
2253 blk_queue_prep_rq(md->queue, dm_prep_fn);
2254 blk_queue_lld_busy(md->queue, dm_lld_busy);
2256 elv_register_queue(md->queue);
2262 * Setup the DM device's queue based on md's type
2264 int dm_setup_md_queue(struct mapped_device *md)
2266 if ((dm_get_md_type(md) == DM_TYPE_REQUEST_BASED) &&
2267 !dm_init_request_based_queue(md)) {
2268 DMWARN("Cannot initialize queue for request-based mapped device");
2275 static struct mapped_device *dm_find_md(dev_t dev)
2277 struct mapped_device *md;
2278 unsigned minor = MINOR(dev);
2280 if (MAJOR(dev) != _major || minor >= (1 << MINORBITS))
2283 spin_lock(&_minor_lock);
2285 md = idr_find(&_minor_idr, minor);
2286 if (md && (md == MINOR_ALLOCED ||
2287 (MINOR(disk_devt(dm_disk(md))) != minor) ||
2288 dm_deleting_md(md) ||
2289 test_bit(DMF_FREEING, &md->flags))) {
2295 spin_unlock(&_minor_lock);
2300 struct mapped_device *dm_get_md(dev_t dev)
2302 struct mapped_device *md = dm_find_md(dev);
2309 EXPORT_SYMBOL_GPL(dm_get_md);
2311 void *dm_get_mdptr(struct mapped_device *md)
2313 return md->interface_ptr;
2316 void dm_set_mdptr(struct mapped_device *md, void *ptr)
2318 md->interface_ptr = ptr;
2321 void dm_get(struct mapped_device *md)
2323 atomic_inc(&md->holders);
2324 BUG_ON(test_bit(DMF_FREEING, &md->flags));
2327 const char *dm_device_name(struct mapped_device *md)
2331 EXPORT_SYMBOL_GPL(dm_device_name);
2333 static void __dm_destroy(struct mapped_device *md, bool wait)
2335 struct dm_table *map;
2340 spin_lock(&_minor_lock);
2341 map = dm_get_live_table(md, &srcu_idx);
2342 idr_replace(&_minor_idr, MINOR_ALLOCED, MINOR(disk_devt(dm_disk(md))));
2343 set_bit(DMF_FREEING, &md->flags);
2344 spin_unlock(&_minor_lock);
2346 if (!dm_suspended_md(md)) {
2347 dm_table_presuspend_targets(map);
2348 dm_table_postsuspend_targets(map);
2351 /* dm_put_live_table must be before msleep, otherwise deadlock is possible */
2352 dm_put_live_table(md, srcu_idx);
2355 * Rare, but there may be I/O requests still going to complete,
2356 * for example. Wait for all references to disappear.
2357 * No one should increment the reference count of the mapped_device,
2358 * after the mapped_device state becomes DMF_FREEING.
2361 while (atomic_read(&md->holders))
2363 else if (atomic_read(&md->holders))
2364 DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)",
2365 dm_device_name(md), atomic_read(&md->holders));
2368 dm_table_destroy(__unbind(md));
2372 void dm_destroy(struct mapped_device *md)
2374 __dm_destroy(md, true);
2377 void dm_destroy_immediate(struct mapped_device *md)
2379 __dm_destroy(md, false);
2382 void dm_put(struct mapped_device *md)
2384 atomic_dec(&md->holders);
2386 EXPORT_SYMBOL_GPL(dm_put);
2388 static int dm_wait_for_completion(struct mapped_device *md, int interruptible)
2391 DECLARE_WAITQUEUE(wait, current);
2393 add_wait_queue(&md->wait, &wait);
2396 set_current_state(interruptible);
2398 if (!md_in_flight(md))
2401 if (interruptible == TASK_INTERRUPTIBLE &&
2402 signal_pending(current)) {
2409 set_current_state(TASK_RUNNING);
2411 remove_wait_queue(&md->wait, &wait);
2417 * Process the deferred bios
2419 static void dm_wq_work(struct work_struct *work)
2421 struct mapped_device *md = container_of(work, struct mapped_device,
2425 struct dm_table *map;
2427 map = dm_get_live_table(md, &srcu_idx);
2429 while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
2430 spin_lock_irq(&md->deferred_lock);
2431 c = bio_list_pop(&md->deferred);
2432 spin_unlock_irq(&md->deferred_lock);
2437 if (dm_request_based(md))
2438 generic_make_request(c);
2440 __split_and_process_bio(md, map, c);
2443 dm_put_live_table(md, srcu_idx);
2446 static void dm_queue_flush(struct mapped_device *md)
2448 clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2449 smp_mb__after_atomic();
2450 queue_work(md->wq, &md->work);
2454 * Swap in a new table, returning the old one for the caller to destroy.
2456 struct dm_table *dm_swap_table(struct mapped_device *md, struct dm_table *table)
2458 struct dm_table *live_map = NULL, *map = ERR_PTR(-EINVAL);
2459 struct queue_limits limits;
2462 mutex_lock(&md->suspend_lock);
2464 /* device must be suspended */
2465 if (!dm_suspended_md(md))
2469 * If the new table has no data devices, retain the existing limits.
2470 * This helps multipath with queue_if_no_path if all paths disappear,
2471 * then new I/O is queued based on these limits, and then some paths
2474 if (dm_table_has_no_data_devices(table)) {
2475 live_map = dm_get_live_table_fast(md);
2477 limits = md->queue->limits;
2478 dm_put_live_table_fast(md);
2482 r = dm_calculate_queue_limits(table, &limits);
2489 map = __bind(md, table, &limits);
2492 mutex_unlock(&md->suspend_lock);
2497 * Functions to lock and unlock any filesystem running on the
2500 static int lock_fs(struct mapped_device *md)
2504 WARN_ON(md->frozen_sb);
2506 md->frozen_sb = freeze_bdev(md->bdev);
2507 if (IS_ERR(md->frozen_sb)) {
2508 r = PTR_ERR(md->frozen_sb);
2509 md->frozen_sb = NULL;
2513 set_bit(DMF_FROZEN, &md->flags);
2518 static void unlock_fs(struct mapped_device *md)
2520 if (!test_bit(DMF_FROZEN, &md->flags))
2523 thaw_bdev(md->bdev, md->frozen_sb);
2524 md->frozen_sb = NULL;
2525 clear_bit(DMF_FROZEN, &md->flags);
2529 * We need to be able to change a mapping table under a mounted
2530 * filesystem. For example we might want to move some data in
2531 * the background. Before the table can be swapped with
2532 * dm_bind_table, dm_suspend must be called to flush any in
2533 * flight bios and ensure that any further io gets deferred.
2536 * Suspend mechanism in request-based dm.
2538 * 1. Flush all I/Os by lock_fs() if needed.
2539 * 2. Stop dispatching any I/O by stopping the request_queue.
2540 * 3. Wait for all in-flight I/Os to be completed or requeued.
2542 * To abort suspend, start the request_queue.
2544 int dm_suspend(struct mapped_device *md, unsigned suspend_flags)
2546 struct dm_table *map = NULL;
2548 int do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG ? 1 : 0;
2549 int noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG ? 1 : 0;
2551 mutex_lock(&md->suspend_lock);
2553 if (dm_suspended_md(md)) {
2561 * DMF_NOFLUSH_SUSPENDING must be set before presuspend.
2562 * This flag is cleared before dm_suspend returns.
2565 set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2567 /* This does not get reverted if there's an error later. */
2568 dm_table_presuspend_targets(map);
2571 * Flush I/O to the device.
2572 * Any I/O submitted after lock_fs() may not be flushed.
2573 * noflush takes precedence over do_lockfs.
2574 * (lock_fs() flushes I/Os and waits for them to complete.)
2576 if (!noflush && do_lockfs) {
2583 * Here we must make sure that no processes are submitting requests
2584 * to target drivers i.e. no one may be executing
2585 * __split_and_process_bio. This is called from dm_request and
2588 * To get all processes out of __split_and_process_bio in dm_request,
2589 * we take the write lock. To prevent any process from reentering
2590 * __split_and_process_bio from dm_request and quiesce the thread
2591 * (dm_wq_work), we set BMF_BLOCK_IO_FOR_SUSPEND and call
2592 * flush_workqueue(md->wq).
2594 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2595 synchronize_srcu(&md->io_barrier);
2598 * Stop md->queue before flushing md->wq in case request-based
2599 * dm defers requests to md->wq from md->queue.
2601 if (dm_request_based(md))
2602 stop_queue(md->queue);
2604 flush_workqueue(md->wq);
2607 * At this point no more requests are entering target request routines.
2608 * We call dm_wait_for_completion to wait for all existing requests
2611 r = dm_wait_for_completion(md, TASK_INTERRUPTIBLE);
2614 clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2615 synchronize_srcu(&md->io_barrier);
2617 /* were we interrupted ? */
2621 if (dm_request_based(md))
2622 start_queue(md->queue);
2625 goto out_unlock; /* pushback list is already flushed, so skip flush */
2629 * If dm_wait_for_completion returned 0, the device is completely
2630 * quiescent now. There is no request-processing activity. All new
2631 * requests are being added to md->deferred list.
2634 set_bit(DMF_SUSPENDED, &md->flags);
2636 dm_table_postsuspend_targets(map);
2639 mutex_unlock(&md->suspend_lock);
2643 int dm_resume(struct mapped_device *md)
2646 struct dm_table *map = NULL;
2648 mutex_lock(&md->suspend_lock);
2649 if (!dm_suspended_md(md))
2653 if (!map || !dm_table_get_size(map))
2656 r = dm_table_resume_targets(map);
2663 * Flushing deferred I/Os must be done after targets are resumed
2664 * so that mapping of targets can work correctly.
2665 * Request-based dm is queueing the deferred I/Os in its request_queue.
2667 if (dm_request_based(md))
2668 start_queue(md->queue);
2672 clear_bit(DMF_SUSPENDED, &md->flags);
2676 mutex_unlock(&md->suspend_lock);
2682 * Internal suspend/resume works like userspace-driven suspend. It waits
2683 * until all bios finish and prevents issuing new bios to the target drivers.
2684 * It may be used only from the kernel.
2686 * Internal suspend holds md->suspend_lock, which prevents interaction with
2687 * userspace-driven suspend.
2690 void dm_internal_suspend(struct mapped_device *md)
2692 mutex_lock(&md->suspend_lock);
2693 if (dm_suspended_md(md))
2696 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2697 synchronize_srcu(&md->io_barrier);
2698 flush_workqueue(md->wq);
2699 dm_wait_for_completion(md, TASK_UNINTERRUPTIBLE);
2702 void dm_internal_resume(struct mapped_device *md)
2704 if (dm_suspended_md(md))
2710 mutex_unlock(&md->suspend_lock);
2713 /*-----------------------------------------------------------------
2714 * Event notification.
2715 *---------------------------------------------------------------*/
2716 int dm_kobject_uevent(struct mapped_device *md, enum kobject_action action,
2719 char udev_cookie[DM_COOKIE_LENGTH];
2720 char *envp[] = { udev_cookie, NULL };
2723 return kobject_uevent(&disk_to_dev(md->disk)->kobj, action);
2725 snprintf(udev_cookie, DM_COOKIE_LENGTH, "%s=%u",
2726 DM_COOKIE_ENV_VAR_NAME, cookie);
2727 return kobject_uevent_env(&disk_to_dev(md->disk)->kobj,
2732 uint32_t dm_next_uevent_seq(struct mapped_device *md)
2734 return atomic_add_return(1, &md->uevent_seq);
2737 uint32_t dm_get_event_nr(struct mapped_device *md)
2739 return atomic_read(&md->event_nr);
2742 int dm_wait_event(struct mapped_device *md, int event_nr)
2744 return wait_event_interruptible(md->eventq,
2745 (event_nr != atomic_read(&md->event_nr)));
2748 void dm_uevent_add(struct mapped_device *md, struct list_head *elist)
2750 unsigned long flags;
2752 spin_lock_irqsave(&md->uevent_lock, flags);
2753 list_add(elist, &md->uevent_list);
2754 spin_unlock_irqrestore(&md->uevent_lock, flags);
2758 * The gendisk is only valid as long as you have a reference
2761 struct gendisk *dm_disk(struct mapped_device *md)
2766 struct kobject *dm_kobject(struct mapped_device *md)
2768 return &md->kobj_holder.kobj;
2771 struct mapped_device *dm_get_from_kobject(struct kobject *kobj)
2773 struct mapped_device *md;
2775 md = container_of(kobj, struct mapped_device, kobj_holder.kobj);
2777 if (test_bit(DMF_FREEING, &md->flags) ||
2785 int dm_suspended_md(struct mapped_device *md)
2787 return test_bit(DMF_SUSPENDED, &md->flags);
2790 int dm_test_deferred_remove_flag(struct mapped_device *md)
2792 return test_bit(DMF_DEFERRED_REMOVE, &md->flags);
2795 int dm_suspended(struct dm_target *ti)
2797 return dm_suspended_md(dm_table_get_md(ti->table));
2799 EXPORT_SYMBOL_GPL(dm_suspended);
2801 int dm_noflush_suspending(struct dm_target *ti)
2803 return __noflush_suspending(dm_table_get_md(ti->table));
2805 EXPORT_SYMBOL_GPL(dm_noflush_suspending);
2807 struct dm_md_mempools *dm_alloc_md_mempools(unsigned type, unsigned integrity, unsigned per_bio_data_size)
2809 struct dm_md_mempools *pools = kzalloc(sizeof(*pools), GFP_KERNEL);
2810 struct kmem_cache *cachep;
2811 unsigned int pool_size;
2812 unsigned int front_pad;
2817 if (type == DM_TYPE_BIO_BASED) {
2819 pool_size = dm_get_reserved_bio_based_ios();
2820 front_pad = roundup(per_bio_data_size, __alignof__(struct dm_target_io)) + offsetof(struct dm_target_io, clone);
2821 } else if (type == DM_TYPE_REQUEST_BASED) {
2822 cachep = _rq_tio_cache;
2823 pool_size = dm_get_reserved_rq_based_ios();
2824 front_pad = offsetof(struct dm_rq_clone_bio_info, clone);
2825 /* per_bio_data_size is not used. See __bind_mempools(). */
2826 WARN_ON(per_bio_data_size != 0);
2830 pools->io_pool = mempool_create_slab_pool(pool_size, cachep);
2831 if (!pools->io_pool)
2834 pools->bs = bioset_create(pool_size, front_pad);
2838 if (integrity && bioset_integrity_create(pools->bs, pool_size))
2844 dm_free_md_mempools(pools);
2849 void dm_free_md_mempools(struct dm_md_mempools *pools)
2855 mempool_destroy(pools->io_pool);
2858 bioset_free(pools->bs);
2863 static const struct block_device_operations dm_blk_dops = {
2864 .open = dm_blk_open,
2865 .release = dm_blk_close,
2866 .ioctl = dm_blk_ioctl,
2867 .getgeo = dm_blk_getgeo,
2868 .owner = THIS_MODULE
2874 module_init(dm_init);
2875 module_exit(dm_exit);
2877 module_param(major, uint, 0);
2878 MODULE_PARM_DESC(major, "The major number of the device mapper");
2880 module_param(reserved_bio_based_ios, uint, S_IRUGO | S_IWUSR);
2881 MODULE_PARM_DESC(reserved_bio_based_ios, "Reserved IOs in bio-based mempools");
2883 module_param(reserved_rq_based_ios, uint, S_IRUGO | S_IWUSR);
2884 MODULE_PARM_DESC(reserved_rq_based_ios, "Reserved IOs in request-based mempools");
2886 MODULE_DESCRIPTION(DM_NAME " driver");
2887 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
2888 MODULE_LICENSE("GPL");