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 dm_stats *dm_get_stats(struct mapped_device *md)
476 static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo)
478 struct mapped_device *md = bdev->bd_disk->private_data;
480 return dm_get_geometry(md, geo);
483 static int dm_blk_ioctl(struct block_device *bdev, fmode_t mode,
484 unsigned int cmd, unsigned long arg)
486 struct mapped_device *md = bdev->bd_disk->private_data;
488 struct dm_table *map;
489 struct dm_target *tgt;
493 map = dm_get_live_table(md, &srcu_idx);
495 if (!map || !dm_table_get_size(map))
498 /* We only support devices that have a single target */
499 if (dm_table_get_num_targets(map) != 1)
502 tgt = dm_table_get_target(map, 0);
504 if (dm_suspended_md(md)) {
509 if (tgt->type->ioctl)
510 r = tgt->type->ioctl(tgt, cmd, arg);
513 dm_put_live_table(md, srcu_idx);
515 if (r == -ENOTCONN) {
523 static struct dm_io *alloc_io(struct mapped_device *md)
525 return mempool_alloc(md->io_pool, GFP_NOIO);
528 static void free_io(struct mapped_device *md, struct dm_io *io)
530 mempool_free(io, md->io_pool);
533 static void free_tio(struct mapped_device *md, struct dm_target_io *tio)
535 bio_put(&tio->clone);
538 static struct dm_rq_target_io *alloc_rq_tio(struct mapped_device *md,
541 return mempool_alloc(md->io_pool, gfp_mask);
544 static void free_rq_tio(struct dm_rq_target_io *tio)
546 mempool_free(tio, tio->md->io_pool);
549 static int md_in_flight(struct mapped_device *md)
551 return atomic_read(&md->pending[READ]) +
552 atomic_read(&md->pending[WRITE]);
555 static void start_io_acct(struct dm_io *io)
557 struct mapped_device *md = io->md;
558 struct bio *bio = io->bio;
560 int rw = bio_data_dir(bio);
562 io->start_time = jiffies;
564 cpu = part_stat_lock();
565 part_round_stats(cpu, &dm_disk(md)->part0);
567 atomic_set(&dm_disk(md)->part0.in_flight[rw],
568 atomic_inc_return(&md->pending[rw]));
570 if (unlikely(dm_stats_used(&md->stats)))
571 dm_stats_account_io(&md->stats, bio->bi_rw, bio->bi_iter.bi_sector,
572 bio_sectors(bio), false, 0, &io->stats_aux);
575 static void end_io_acct(struct dm_io *io)
577 struct mapped_device *md = io->md;
578 struct bio *bio = io->bio;
579 unsigned long duration = jiffies - io->start_time;
581 int rw = bio_data_dir(bio);
583 cpu = part_stat_lock();
584 part_round_stats(cpu, &dm_disk(md)->part0);
585 part_stat_add(cpu, &dm_disk(md)->part0, ticks[rw], duration);
588 if (unlikely(dm_stats_used(&md->stats)))
589 dm_stats_account_io(&md->stats, bio->bi_rw, bio->bi_iter.bi_sector,
590 bio_sectors(bio), true, duration, &io->stats_aux);
593 * After this is decremented the bio must not be touched if it is
596 pending = atomic_dec_return(&md->pending[rw]);
597 atomic_set(&dm_disk(md)->part0.in_flight[rw], pending);
598 pending += atomic_read(&md->pending[rw^0x1]);
600 /* nudge anyone waiting on suspend queue */
606 * Add the bio to the list of deferred io.
608 static void queue_io(struct mapped_device *md, struct bio *bio)
612 spin_lock_irqsave(&md->deferred_lock, flags);
613 bio_list_add(&md->deferred, bio);
614 spin_unlock_irqrestore(&md->deferred_lock, flags);
615 queue_work(md->wq, &md->work);
619 * Everyone (including functions in this file), should use this
620 * function to access the md->map field, and make sure they call
621 * dm_put_live_table() when finished.
623 struct dm_table *dm_get_live_table(struct mapped_device *md, int *srcu_idx) __acquires(md->io_barrier)
625 *srcu_idx = srcu_read_lock(&md->io_barrier);
627 return srcu_dereference(md->map, &md->io_barrier);
630 void dm_put_live_table(struct mapped_device *md, int srcu_idx) __releases(md->io_barrier)
632 srcu_read_unlock(&md->io_barrier, srcu_idx);
635 void dm_sync_table(struct mapped_device *md)
637 synchronize_srcu(&md->io_barrier);
638 synchronize_rcu_expedited();
642 * A fast alternative to dm_get_live_table/dm_put_live_table.
643 * The caller must not block between these two functions.
645 static struct dm_table *dm_get_live_table_fast(struct mapped_device *md) __acquires(RCU)
648 return rcu_dereference(md->map);
651 static void dm_put_live_table_fast(struct mapped_device *md) __releases(RCU)
657 * Get the geometry associated with a dm device
659 int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo)
667 * Set the geometry of a device.
669 int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo)
671 sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors;
673 if (geo->start > sz) {
674 DMWARN("Start sector is beyond the geometry limits.");
683 /*-----------------------------------------------------------------
685 * A more elegant soln is in the works that uses the queue
686 * merge fn, unfortunately there are a couple of changes to
687 * the block layer that I want to make for this. So in the
688 * interests of getting something for people to use I give
689 * you this clearly demarcated crap.
690 *---------------------------------------------------------------*/
692 static int __noflush_suspending(struct mapped_device *md)
694 return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
698 * Decrements the number of outstanding ios that a bio has been
699 * cloned into, completing the original io if necc.
701 static void dec_pending(struct dm_io *io, int error)
706 struct mapped_device *md = io->md;
708 /* Push-back supersedes any I/O errors */
709 if (unlikely(error)) {
710 spin_lock_irqsave(&io->endio_lock, flags);
711 if (!(io->error > 0 && __noflush_suspending(md)))
713 spin_unlock_irqrestore(&io->endio_lock, flags);
716 if (atomic_dec_and_test(&io->io_count)) {
717 if (io->error == DM_ENDIO_REQUEUE) {
719 * Target requested pushing back the I/O.
721 spin_lock_irqsave(&md->deferred_lock, flags);
722 if (__noflush_suspending(md))
723 bio_list_add_head(&md->deferred, io->bio);
725 /* noflush suspend was interrupted. */
727 spin_unlock_irqrestore(&md->deferred_lock, flags);
730 io_error = io->error;
735 if (io_error == DM_ENDIO_REQUEUE)
738 if ((bio->bi_rw & REQ_FLUSH) && bio->bi_iter.bi_size) {
740 * Preflush done for flush with data, reissue
743 bio->bi_rw &= ~REQ_FLUSH;
746 /* done with normal IO or empty flush */
747 trace_block_bio_complete(md->queue, bio, io_error);
748 bio_endio(bio, io_error);
753 static void clone_endio(struct bio *bio, int error)
756 struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone);
757 struct dm_io *io = tio->io;
758 struct mapped_device *md = tio->io->md;
759 dm_endio_fn endio = tio->ti->type->end_io;
761 if (!bio_flagged(bio, BIO_UPTODATE) && !error)
765 r = endio(tio->ti, bio, error);
766 if (r < 0 || r == DM_ENDIO_REQUEUE)
768 * error and requeue request are handled
772 else if (r == DM_ENDIO_INCOMPLETE)
773 /* The target will handle the io */
776 DMWARN("unimplemented target endio return value: %d", r);
782 dec_pending(io, error);
786 * Partial completion handling for request-based dm
788 static void end_clone_bio(struct bio *clone, int error)
790 struct dm_rq_clone_bio_info *info =
791 container_of(clone, struct dm_rq_clone_bio_info, clone);
792 struct dm_rq_target_io *tio = info->tio;
793 struct bio *bio = info->orig;
794 unsigned int nr_bytes = info->orig->bi_iter.bi_size;
800 * An error has already been detected on the request.
801 * Once error occurred, just let clone->end_io() handle
807 * Don't notice the error to the upper layer yet.
808 * The error handling decision is made by the target driver,
809 * when the request is completed.
816 * I/O for the bio successfully completed.
817 * Notice the data completion to the upper layer.
821 * bios are processed from the head of the list.
822 * So the completing bio should always be rq->bio.
823 * If it's not, something wrong is happening.
825 if (tio->orig->bio != bio)
826 DMERR("bio completion is going in the middle of the request");
829 * Update the original request.
830 * Do not use blk_end_request() here, because it may complete
831 * the original request before the clone, and break the ordering.
833 blk_update_request(tio->orig, 0, nr_bytes);
837 * Don't touch any member of the md after calling this function because
838 * the md may be freed in dm_put() at the end of this function.
839 * Or do dm_get() before calling this function and dm_put() later.
841 static void rq_completed(struct mapped_device *md, int rw, int run_queue)
843 atomic_dec(&md->pending[rw]);
845 /* nudge anyone waiting on suspend queue */
846 if (!md_in_flight(md))
850 * Run this off this callpath, as drivers could invoke end_io while
851 * inside their request_fn (and holding the queue lock). Calling
852 * back into ->request_fn() could deadlock attempting to grab the
856 blk_run_queue_async(md->queue);
859 * dm_put() must be at the end of this function. See the comment above
864 static void free_rq_clone(struct request *clone)
866 struct dm_rq_target_io *tio = clone->end_io_data;
868 blk_rq_unprep_clone(clone);
873 * Complete the clone and the original request.
874 * Must be called without queue lock.
876 static void dm_end_request(struct request *clone, int error)
878 int rw = rq_data_dir(clone);
879 struct dm_rq_target_io *tio = clone->end_io_data;
880 struct mapped_device *md = tio->md;
881 struct request *rq = tio->orig;
883 if (rq->cmd_type == REQ_TYPE_BLOCK_PC) {
884 rq->errors = clone->errors;
885 rq->resid_len = clone->resid_len;
889 * We are using the sense buffer of the original
891 * So setting the length of the sense data is enough.
893 rq->sense_len = clone->sense_len;
896 free_rq_clone(clone);
897 blk_end_request_all(rq, error);
898 rq_completed(md, rw, true);
901 static void dm_unprep_request(struct request *rq)
903 struct request *clone = rq->special;
906 rq->cmd_flags &= ~REQ_DONTPREP;
908 free_rq_clone(clone);
912 * Requeue the original request of a clone.
914 void dm_requeue_unmapped_request(struct request *clone)
916 int rw = rq_data_dir(clone);
917 struct dm_rq_target_io *tio = clone->end_io_data;
918 struct mapped_device *md = tio->md;
919 struct request *rq = tio->orig;
920 struct request_queue *q = rq->q;
923 dm_unprep_request(rq);
925 spin_lock_irqsave(q->queue_lock, flags);
926 blk_requeue_request(q, rq);
927 spin_unlock_irqrestore(q->queue_lock, flags);
929 rq_completed(md, rw, 0);
931 EXPORT_SYMBOL_GPL(dm_requeue_unmapped_request);
933 static void __stop_queue(struct request_queue *q)
938 static void stop_queue(struct request_queue *q)
942 spin_lock_irqsave(q->queue_lock, flags);
944 spin_unlock_irqrestore(q->queue_lock, flags);
947 static void __start_queue(struct request_queue *q)
949 if (blk_queue_stopped(q))
953 static void start_queue(struct request_queue *q)
957 spin_lock_irqsave(q->queue_lock, flags);
959 spin_unlock_irqrestore(q->queue_lock, flags);
962 static void dm_done(struct request *clone, int error, bool mapped)
965 struct dm_rq_target_io *tio = clone->end_io_data;
966 dm_request_endio_fn rq_end_io = NULL;
969 rq_end_io = tio->ti->type->rq_end_io;
971 if (mapped && rq_end_io)
972 r = rq_end_io(tio->ti, clone, error, &tio->info);
976 /* The target wants to complete the I/O */
977 dm_end_request(clone, r);
978 else if (r == DM_ENDIO_INCOMPLETE)
979 /* The target will handle the I/O */
981 else if (r == DM_ENDIO_REQUEUE)
982 /* The target wants to requeue the I/O */
983 dm_requeue_unmapped_request(clone);
985 DMWARN("unimplemented target endio return value: %d", r);
991 * Request completion handler for request-based dm
993 static void dm_softirq_done(struct request *rq)
996 struct request *clone = rq->completion_data;
997 struct dm_rq_target_io *tio = clone->end_io_data;
999 if (rq->cmd_flags & REQ_FAILED)
1002 dm_done(clone, tio->error, mapped);
1006 * Complete the clone and the original request with the error status
1007 * through softirq context.
1009 static void dm_complete_request(struct request *clone, int error)
1011 struct dm_rq_target_io *tio = clone->end_io_data;
1012 struct request *rq = tio->orig;
1015 rq->completion_data = clone;
1016 blk_complete_request(rq);
1020 * Complete the not-mapped clone and the original request with the error status
1021 * through softirq context.
1022 * Target's rq_end_io() function isn't called.
1023 * This may be used when the target's map_rq() function fails.
1025 void dm_kill_unmapped_request(struct request *clone, int error)
1027 struct dm_rq_target_io *tio = clone->end_io_data;
1028 struct request *rq = tio->orig;
1030 rq->cmd_flags |= REQ_FAILED;
1031 dm_complete_request(clone, error);
1033 EXPORT_SYMBOL_GPL(dm_kill_unmapped_request);
1036 * Called with the queue lock held
1038 static void end_clone_request(struct request *clone, int error)
1041 * For just cleaning up the information of the queue in which
1042 * the clone was dispatched.
1043 * The clone is *NOT* freed actually here because it is alloced from
1044 * dm own mempool and REQ_ALLOCED isn't set in clone->cmd_flags.
1046 __blk_put_request(clone->q, clone);
1049 * Actual request completion is done in a softirq context which doesn't
1050 * hold the queue lock. Otherwise, deadlock could occur because:
1051 * - another request may be submitted by the upper level driver
1052 * of the stacking during the completion
1053 * - the submission which requires queue lock may be done
1054 * against this queue
1056 dm_complete_request(clone, error);
1060 * Return maximum size of I/O possible at the supplied sector up to the current
1063 static sector_t max_io_len_target_boundary(sector_t sector, struct dm_target *ti)
1065 sector_t target_offset = dm_target_offset(ti, sector);
1067 return ti->len - target_offset;
1070 static sector_t max_io_len(sector_t sector, struct dm_target *ti)
1072 sector_t len = max_io_len_target_boundary(sector, ti);
1073 sector_t offset, max_len;
1076 * Does the target need to split even further?
1078 if (ti->max_io_len) {
1079 offset = dm_target_offset(ti, sector);
1080 if (unlikely(ti->max_io_len & (ti->max_io_len - 1)))
1081 max_len = sector_div(offset, ti->max_io_len);
1083 max_len = offset & (ti->max_io_len - 1);
1084 max_len = ti->max_io_len - max_len;
1093 int dm_set_target_max_io_len(struct dm_target *ti, sector_t len)
1095 if (len > UINT_MAX) {
1096 DMERR("Specified maximum size of target IO (%llu) exceeds limit (%u)",
1097 (unsigned long long)len, UINT_MAX);
1098 ti->error = "Maximum size of target IO is too large";
1102 ti->max_io_len = (uint32_t) len;
1106 EXPORT_SYMBOL_GPL(dm_set_target_max_io_len);
1108 static void __map_bio(struct dm_target_io *tio)
1112 struct mapped_device *md;
1113 struct bio *clone = &tio->clone;
1114 struct dm_target *ti = tio->ti;
1116 clone->bi_end_io = clone_endio;
1119 * Map the clone. If r == 0 we don't need to do
1120 * anything, the target has assumed ownership of
1123 atomic_inc(&tio->io->io_count);
1124 sector = clone->bi_iter.bi_sector;
1125 r = ti->type->map(ti, clone);
1126 if (r == DM_MAPIO_REMAPPED) {
1127 /* the bio has been remapped so dispatch it */
1129 trace_block_bio_remap(bdev_get_queue(clone->bi_bdev), clone,
1130 tio->io->bio->bi_bdev->bd_dev, sector);
1132 generic_make_request(clone);
1133 } else if (r < 0 || r == DM_MAPIO_REQUEUE) {
1134 /* error the io and bail out, or requeue it if needed */
1136 dec_pending(tio->io, r);
1139 DMWARN("unimplemented target map return value: %d", r);
1145 struct mapped_device *md;
1146 struct dm_table *map;
1150 sector_t sector_count;
1153 static void bio_setup_sector(struct bio *bio, sector_t sector, sector_t len)
1155 bio->bi_iter.bi_sector = sector;
1156 bio->bi_iter.bi_size = to_bytes(len);
1160 * Creates a bio that consists of range of complete bvecs.
1162 static void clone_bio(struct dm_target_io *tio, struct bio *bio,
1163 sector_t sector, unsigned len)
1165 struct bio *clone = &tio->clone;
1167 __bio_clone_fast(clone, bio);
1169 if (bio_integrity(bio))
1170 bio_integrity_clone(clone, bio, GFP_NOIO);
1172 bio_advance(clone, to_bytes(sector - clone->bi_iter.bi_sector));
1173 clone->bi_iter.bi_size = to_bytes(len);
1175 if (bio_integrity(bio))
1176 bio_integrity_trim(clone, 0, len);
1179 static struct dm_target_io *alloc_tio(struct clone_info *ci,
1180 struct dm_target *ti, int nr_iovecs,
1181 unsigned target_bio_nr)
1183 struct dm_target_io *tio;
1186 clone = bio_alloc_bioset(GFP_NOIO, nr_iovecs, ci->md->bs);
1187 tio = container_of(clone, struct dm_target_io, clone);
1191 tio->target_bio_nr = target_bio_nr;
1196 static void __clone_and_map_simple_bio(struct clone_info *ci,
1197 struct dm_target *ti,
1198 unsigned target_bio_nr, sector_t len)
1200 struct dm_target_io *tio = alloc_tio(ci, ti, ci->bio->bi_max_vecs, target_bio_nr);
1201 struct bio *clone = &tio->clone;
1204 * Discard requests require the bio's inline iovecs be initialized.
1205 * ci->bio->bi_max_vecs is BIO_INLINE_VECS anyway, for both flush
1206 * and discard, so no need for concern about wasted bvec allocations.
1208 __bio_clone_fast(clone, ci->bio);
1210 bio_setup_sector(clone, ci->sector, len);
1215 static void __send_duplicate_bios(struct clone_info *ci, struct dm_target *ti,
1216 unsigned num_bios, sector_t len)
1218 unsigned target_bio_nr;
1220 for (target_bio_nr = 0; target_bio_nr < num_bios; target_bio_nr++)
1221 __clone_and_map_simple_bio(ci, ti, target_bio_nr, len);
1224 static int __send_empty_flush(struct clone_info *ci)
1226 unsigned target_nr = 0;
1227 struct dm_target *ti;
1229 BUG_ON(bio_has_data(ci->bio));
1230 while ((ti = dm_table_get_target(ci->map, target_nr++)))
1231 __send_duplicate_bios(ci, ti, ti->num_flush_bios, 0);
1236 static void __clone_and_map_data_bio(struct clone_info *ci, struct dm_target *ti,
1237 sector_t sector, unsigned len)
1239 struct bio *bio = ci->bio;
1240 struct dm_target_io *tio;
1241 unsigned target_bio_nr;
1242 unsigned num_target_bios = 1;
1245 * Does the target want to receive duplicate copies of the bio?
1247 if (bio_data_dir(bio) == WRITE && ti->num_write_bios)
1248 num_target_bios = ti->num_write_bios(ti, bio);
1250 for (target_bio_nr = 0; target_bio_nr < num_target_bios; target_bio_nr++) {
1251 tio = alloc_tio(ci, ti, 0, target_bio_nr);
1252 clone_bio(tio, bio, sector, len);
1257 typedef unsigned (*get_num_bios_fn)(struct dm_target *ti);
1259 static unsigned get_num_discard_bios(struct dm_target *ti)
1261 return ti->num_discard_bios;
1264 static unsigned get_num_write_same_bios(struct dm_target *ti)
1266 return ti->num_write_same_bios;
1269 typedef bool (*is_split_required_fn)(struct dm_target *ti);
1271 static bool is_split_required_for_discard(struct dm_target *ti)
1273 return ti->split_discard_bios;
1276 static int __send_changing_extent_only(struct clone_info *ci,
1277 get_num_bios_fn get_num_bios,
1278 is_split_required_fn is_split_required)
1280 struct dm_target *ti;
1285 ti = dm_table_find_target(ci->map, ci->sector);
1286 if (!dm_target_is_valid(ti))
1290 * Even though the device advertised support for this type of
1291 * request, that does not mean every target supports it, and
1292 * reconfiguration might also have changed that since the
1293 * check was performed.
1295 num_bios = get_num_bios ? get_num_bios(ti) : 0;
1299 if (is_split_required && !is_split_required(ti))
1300 len = min(ci->sector_count, max_io_len_target_boundary(ci->sector, ti));
1302 len = min(ci->sector_count, max_io_len(ci->sector, ti));
1304 __send_duplicate_bios(ci, ti, num_bios, len);
1307 } while (ci->sector_count -= len);
1312 static int __send_discard(struct clone_info *ci)
1314 return __send_changing_extent_only(ci, get_num_discard_bios,
1315 is_split_required_for_discard);
1318 static int __send_write_same(struct clone_info *ci)
1320 return __send_changing_extent_only(ci, get_num_write_same_bios, NULL);
1324 * Select the correct strategy for processing a non-flush bio.
1326 static int __split_and_process_non_flush(struct clone_info *ci)
1328 struct bio *bio = ci->bio;
1329 struct dm_target *ti;
1332 if (unlikely(bio->bi_rw & REQ_DISCARD))
1333 return __send_discard(ci);
1334 else if (unlikely(bio->bi_rw & REQ_WRITE_SAME))
1335 return __send_write_same(ci);
1337 ti = dm_table_find_target(ci->map, ci->sector);
1338 if (!dm_target_is_valid(ti))
1341 len = min_t(sector_t, max_io_len(ci->sector, ti), ci->sector_count);
1343 __clone_and_map_data_bio(ci, ti, ci->sector, len);
1346 ci->sector_count -= len;
1352 * Entry point to split a bio into clones and submit them to the targets.
1354 static void __split_and_process_bio(struct mapped_device *md,
1355 struct dm_table *map, struct bio *bio)
1357 struct clone_info ci;
1360 if (unlikely(!map)) {
1367 ci.io = alloc_io(md);
1369 atomic_set(&ci.io->io_count, 1);
1372 spin_lock_init(&ci.io->endio_lock);
1373 ci.sector = bio->bi_iter.bi_sector;
1375 start_io_acct(ci.io);
1377 if (bio->bi_rw & REQ_FLUSH) {
1378 ci.bio = &ci.md->flush_bio;
1379 ci.sector_count = 0;
1380 error = __send_empty_flush(&ci);
1381 /* dec_pending submits any data associated with flush */
1384 ci.sector_count = bio_sectors(bio);
1385 while (ci.sector_count && !error)
1386 error = __split_and_process_non_flush(&ci);
1389 /* drop the extra reference count */
1390 dec_pending(ci.io, error);
1392 /*-----------------------------------------------------------------
1394 *---------------------------------------------------------------*/
1396 static int dm_merge_bvec(struct request_queue *q,
1397 struct bvec_merge_data *bvm,
1398 struct bio_vec *biovec)
1400 struct mapped_device *md = q->queuedata;
1401 struct dm_table *map = dm_get_live_table_fast(md);
1402 struct dm_target *ti;
1403 sector_t max_sectors;
1409 ti = dm_table_find_target(map, bvm->bi_sector);
1410 if (!dm_target_is_valid(ti))
1414 * Find maximum amount of I/O that won't need splitting
1416 max_sectors = min(max_io_len(bvm->bi_sector, ti),
1417 (sector_t) BIO_MAX_SECTORS);
1418 max_size = (max_sectors << SECTOR_SHIFT) - bvm->bi_size;
1423 * merge_bvec_fn() returns number of bytes
1424 * it can accept at this offset
1425 * max is precomputed maximal io size
1427 if (max_size && ti->type->merge)
1428 max_size = ti->type->merge(ti, bvm, biovec, max_size);
1430 * If the target doesn't support merge method and some of the devices
1431 * provided their merge_bvec method (we know this by looking at
1432 * queue_max_hw_sectors), then we can't allow bios with multiple vector
1433 * entries. So always set max_size to 0, and the code below allows
1436 else if (queue_max_hw_sectors(q) <= PAGE_SIZE >> 9)
1441 dm_put_live_table_fast(md);
1443 * Always allow an entire first page
1445 if (max_size <= biovec->bv_len && !(bvm->bi_size >> SECTOR_SHIFT))
1446 max_size = biovec->bv_len;
1452 * The request function that just remaps the bio built up by
1455 static void _dm_request(struct request_queue *q, struct bio *bio)
1457 int rw = bio_data_dir(bio);
1458 struct mapped_device *md = q->queuedata;
1461 struct dm_table *map;
1463 map = dm_get_live_table(md, &srcu_idx);
1465 cpu = part_stat_lock();
1466 part_stat_inc(cpu, &dm_disk(md)->part0, ios[rw]);
1467 part_stat_add(cpu, &dm_disk(md)->part0, sectors[rw], bio_sectors(bio));
1470 /* if we're suspended, we have to queue this io for later */
1471 if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))) {
1472 dm_put_live_table(md, srcu_idx);
1474 if (bio_rw(bio) != READA)
1481 __split_and_process_bio(md, map, bio);
1482 dm_put_live_table(md, srcu_idx);
1486 int dm_request_based(struct mapped_device *md)
1488 return blk_queue_stackable(md->queue);
1491 static void dm_request(struct request_queue *q, struct bio *bio)
1493 struct mapped_device *md = q->queuedata;
1495 if (dm_request_based(md))
1496 blk_queue_bio(q, bio);
1498 _dm_request(q, bio);
1501 void dm_dispatch_request(struct request *rq)
1505 if (blk_queue_io_stat(rq->q))
1506 rq->cmd_flags |= REQ_IO_STAT;
1508 rq->start_time = jiffies;
1509 r = blk_insert_cloned_request(rq->q, rq);
1511 dm_complete_request(rq, r);
1513 EXPORT_SYMBOL_GPL(dm_dispatch_request);
1515 static int dm_rq_bio_constructor(struct bio *bio, struct bio *bio_orig,
1518 struct dm_rq_target_io *tio = data;
1519 struct dm_rq_clone_bio_info *info =
1520 container_of(bio, struct dm_rq_clone_bio_info, clone);
1522 info->orig = bio_orig;
1524 bio->bi_end_io = end_clone_bio;
1529 static int setup_clone(struct request *clone, struct request *rq,
1530 struct dm_rq_target_io *tio)
1534 r = blk_rq_prep_clone(clone, rq, tio->md->bs, GFP_ATOMIC,
1535 dm_rq_bio_constructor, tio);
1539 clone->cmd = rq->cmd;
1540 clone->cmd_len = rq->cmd_len;
1541 clone->sense = rq->sense;
1542 clone->buffer = rq->buffer;
1543 clone->end_io = end_clone_request;
1544 clone->end_io_data = tio;
1549 static struct request *clone_rq(struct request *rq, struct mapped_device *md,
1552 struct request *clone;
1553 struct dm_rq_target_io *tio;
1555 tio = alloc_rq_tio(md, gfp_mask);
1563 memset(&tio->info, 0, sizeof(tio->info));
1565 clone = &tio->clone;
1566 if (setup_clone(clone, rq, tio)) {
1576 * Called with the queue lock held.
1578 static int dm_prep_fn(struct request_queue *q, struct request *rq)
1580 struct mapped_device *md = q->queuedata;
1581 struct request *clone;
1583 if (unlikely(rq->special)) {
1584 DMWARN("Already has something in rq->special.");
1585 return BLKPREP_KILL;
1588 clone = clone_rq(rq, md, GFP_ATOMIC);
1590 return BLKPREP_DEFER;
1592 rq->special = clone;
1593 rq->cmd_flags |= REQ_DONTPREP;
1600 * 0 : the request has been processed (not requeued)
1601 * !0 : the request has been requeued
1603 static int map_request(struct dm_target *ti, struct request *clone,
1604 struct mapped_device *md)
1606 int r, requeued = 0;
1607 struct dm_rq_target_io *tio = clone->end_io_data;
1610 r = ti->type->map_rq(ti, clone, &tio->info);
1612 case DM_MAPIO_SUBMITTED:
1613 /* The target has taken the I/O to submit by itself later */
1615 case DM_MAPIO_REMAPPED:
1616 /* The target has remapped the I/O so dispatch it */
1617 trace_block_rq_remap(clone->q, clone, disk_devt(dm_disk(md)),
1618 blk_rq_pos(tio->orig));
1619 dm_dispatch_request(clone);
1621 case DM_MAPIO_REQUEUE:
1622 /* The target wants to requeue the I/O */
1623 dm_requeue_unmapped_request(clone);
1628 DMWARN("unimplemented target map return value: %d", r);
1632 /* The target wants to complete the I/O */
1633 dm_kill_unmapped_request(clone, r);
1640 static struct request *dm_start_request(struct mapped_device *md, struct request *orig)
1642 struct request *clone;
1644 blk_start_request(orig);
1645 clone = orig->special;
1646 atomic_inc(&md->pending[rq_data_dir(clone)]);
1649 * Hold the md reference here for the in-flight I/O.
1650 * We can't rely on the reference count by device opener,
1651 * because the device may be closed during the request completion
1652 * when all bios are completed.
1653 * See the comment in rq_completed() too.
1661 * q->request_fn for request-based dm.
1662 * Called with the queue lock held.
1664 static void dm_request_fn(struct request_queue *q)
1666 struct mapped_device *md = q->queuedata;
1668 struct dm_table *map = dm_get_live_table(md, &srcu_idx);
1669 struct dm_target *ti;
1670 struct request *rq, *clone;
1674 * For suspend, check blk_queue_stopped() and increment
1675 * ->pending within a single queue_lock not to increment the
1676 * number of in-flight I/Os after the queue is stopped in
1679 while (!blk_queue_stopped(q)) {
1680 rq = blk_peek_request(q);
1684 /* always use block 0 to find the target for flushes for now */
1686 if (!(rq->cmd_flags & REQ_FLUSH))
1687 pos = blk_rq_pos(rq);
1689 ti = dm_table_find_target(map, pos);
1690 if (!dm_target_is_valid(ti)) {
1692 * Must perform setup, that dm_done() requires,
1693 * before calling dm_kill_unmapped_request
1695 DMERR_LIMIT("request attempted access beyond the end of device");
1696 clone = dm_start_request(md, rq);
1697 dm_kill_unmapped_request(clone, -EIO);
1701 if (ti->type->busy && ti->type->busy(ti))
1704 clone = dm_start_request(md, rq);
1706 spin_unlock(q->queue_lock);
1707 if (map_request(ti, clone, md))
1710 BUG_ON(!irqs_disabled());
1711 spin_lock(q->queue_lock);
1717 BUG_ON(!irqs_disabled());
1718 spin_lock(q->queue_lock);
1721 blk_delay_queue(q, HZ / 10);
1723 dm_put_live_table(md, srcu_idx);
1726 int dm_underlying_device_busy(struct request_queue *q)
1728 return blk_lld_busy(q);
1730 EXPORT_SYMBOL_GPL(dm_underlying_device_busy);
1732 static int dm_lld_busy(struct request_queue *q)
1735 struct mapped_device *md = q->queuedata;
1736 struct dm_table *map = dm_get_live_table_fast(md);
1738 if (!map || test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))
1741 r = dm_table_any_busy_target(map);
1743 dm_put_live_table_fast(md);
1748 static int dm_any_congested(void *congested_data, int bdi_bits)
1751 struct mapped_device *md = congested_data;
1752 struct dm_table *map;
1754 if (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
1755 map = dm_get_live_table_fast(md);
1758 * Request-based dm cares about only own queue for
1759 * the query about congestion status of request_queue
1761 if (dm_request_based(md))
1762 r = md->queue->backing_dev_info.state &
1765 r = dm_table_any_congested(map, bdi_bits);
1767 dm_put_live_table_fast(md);
1773 /*-----------------------------------------------------------------
1774 * An IDR is used to keep track of allocated minor numbers.
1775 *---------------------------------------------------------------*/
1776 static void free_minor(int minor)
1778 spin_lock(&_minor_lock);
1779 idr_remove(&_minor_idr, minor);
1780 spin_unlock(&_minor_lock);
1784 * See if the device with a specific minor # is free.
1786 static int specific_minor(int minor)
1790 if (minor >= (1 << MINORBITS))
1793 idr_preload(GFP_KERNEL);
1794 spin_lock(&_minor_lock);
1796 r = idr_alloc(&_minor_idr, MINOR_ALLOCED, minor, minor + 1, GFP_NOWAIT);
1798 spin_unlock(&_minor_lock);
1801 return r == -ENOSPC ? -EBUSY : r;
1805 static int next_free_minor(int *minor)
1809 idr_preload(GFP_KERNEL);
1810 spin_lock(&_minor_lock);
1812 r = idr_alloc(&_minor_idr, MINOR_ALLOCED, 0, 1 << MINORBITS, GFP_NOWAIT);
1814 spin_unlock(&_minor_lock);
1822 static const struct block_device_operations dm_blk_dops;
1824 static void dm_wq_work(struct work_struct *work);
1826 static void dm_init_md_queue(struct mapped_device *md)
1829 * Request-based dm devices cannot be stacked on top of bio-based dm
1830 * devices. The type of this dm device has not been decided yet.
1831 * The type is decided at the first table loading time.
1832 * To prevent problematic device stacking, clear the queue flag
1833 * for request stacking support until then.
1835 * This queue is new, so no concurrency on the queue_flags.
1837 queue_flag_clear_unlocked(QUEUE_FLAG_STACKABLE, md->queue);
1839 md->queue->queuedata = md;
1840 md->queue->backing_dev_info.congested_fn = dm_any_congested;
1841 md->queue->backing_dev_info.congested_data = md;
1842 blk_queue_make_request(md->queue, dm_request);
1843 blk_queue_bounce_limit(md->queue, BLK_BOUNCE_ANY);
1844 blk_queue_merge_bvec(md->queue, dm_merge_bvec);
1848 * Allocate and initialise a blank device with a given minor.
1850 static struct mapped_device *alloc_dev(int minor)
1853 struct mapped_device *md = kzalloc(sizeof(*md), GFP_KERNEL);
1857 DMWARN("unable to allocate device, out of memory.");
1861 if (!try_module_get(THIS_MODULE))
1862 goto bad_module_get;
1864 /* get a minor number for the dev */
1865 if (minor == DM_ANY_MINOR)
1866 r = next_free_minor(&minor);
1868 r = specific_minor(minor);
1872 r = init_srcu_struct(&md->io_barrier);
1874 goto bad_io_barrier;
1876 md->type = DM_TYPE_NONE;
1877 mutex_init(&md->suspend_lock);
1878 mutex_init(&md->type_lock);
1879 spin_lock_init(&md->deferred_lock);
1880 atomic_set(&md->holders, 1);
1881 atomic_set(&md->open_count, 0);
1882 atomic_set(&md->event_nr, 0);
1883 atomic_set(&md->uevent_seq, 0);
1884 INIT_LIST_HEAD(&md->uevent_list);
1885 spin_lock_init(&md->uevent_lock);
1887 md->queue = blk_alloc_queue(GFP_KERNEL);
1891 dm_init_md_queue(md);
1893 md->disk = alloc_disk(1);
1897 atomic_set(&md->pending[0], 0);
1898 atomic_set(&md->pending[1], 0);
1899 init_waitqueue_head(&md->wait);
1900 INIT_WORK(&md->work, dm_wq_work);
1901 init_waitqueue_head(&md->eventq);
1902 init_completion(&md->kobj_holder.completion);
1904 md->disk->major = _major;
1905 md->disk->first_minor = minor;
1906 md->disk->fops = &dm_blk_dops;
1907 md->disk->queue = md->queue;
1908 md->disk->private_data = md;
1909 sprintf(md->disk->disk_name, "dm-%d", minor);
1911 format_dev_t(md->name, MKDEV(_major, minor));
1913 md->wq = alloc_workqueue("kdmflush", WQ_MEM_RECLAIM, 0);
1917 md->bdev = bdget_disk(md->disk, 0);
1921 bio_init(&md->flush_bio);
1922 md->flush_bio.bi_bdev = md->bdev;
1923 md->flush_bio.bi_rw = WRITE_FLUSH;
1925 dm_stats_init(&md->stats);
1927 /* Populate the mapping, nobody knows we exist yet */
1928 spin_lock(&_minor_lock);
1929 old_md = idr_replace(&_minor_idr, md, minor);
1930 spin_unlock(&_minor_lock);
1932 BUG_ON(old_md != MINOR_ALLOCED);
1937 destroy_workqueue(md->wq);
1939 del_gendisk(md->disk);
1942 blk_cleanup_queue(md->queue);
1944 cleanup_srcu_struct(&md->io_barrier);
1948 module_put(THIS_MODULE);
1954 static void unlock_fs(struct mapped_device *md);
1956 static void free_dev(struct mapped_device *md)
1958 int minor = MINOR(disk_devt(md->disk));
1962 destroy_workqueue(md->wq);
1964 mempool_destroy(md->io_pool);
1966 bioset_free(md->bs);
1967 blk_integrity_unregister(md->disk);
1968 del_gendisk(md->disk);
1969 cleanup_srcu_struct(&md->io_barrier);
1972 spin_lock(&_minor_lock);
1973 md->disk->private_data = NULL;
1974 spin_unlock(&_minor_lock);
1977 blk_cleanup_queue(md->queue);
1978 dm_stats_cleanup(&md->stats);
1979 module_put(THIS_MODULE);
1983 static void __bind_mempools(struct mapped_device *md, struct dm_table *t)
1985 struct dm_md_mempools *p = dm_table_get_md_mempools(t);
1987 if (md->io_pool && md->bs) {
1988 /* The md already has necessary mempools. */
1989 if (dm_table_get_type(t) == DM_TYPE_BIO_BASED) {
1991 * Reload bioset because front_pad may have changed
1992 * because a different table was loaded.
1994 bioset_free(md->bs);
1997 } else if (dm_table_get_type(t) == DM_TYPE_REQUEST_BASED) {
1999 * There's no need to reload with request-based dm
2000 * because the size of front_pad doesn't change.
2001 * Note for future: If you are to reload bioset,
2002 * prep-ed requests in the queue may refer
2003 * to bio from the old bioset, so you must walk
2004 * through the queue to unprep.
2010 BUG_ON(!p || md->io_pool || md->bs);
2012 md->io_pool = p->io_pool;
2018 /* mempool bind completed, now no need any mempools in the table */
2019 dm_table_free_md_mempools(t);
2023 * Bind a table to the device.
2025 static void event_callback(void *context)
2027 unsigned long flags;
2029 struct mapped_device *md = (struct mapped_device *) context;
2031 spin_lock_irqsave(&md->uevent_lock, flags);
2032 list_splice_init(&md->uevent_list, &uevents);
2033 spin_unlock_irqrestore(&md->uevent_lock, flags);
2035 dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj);
2037 atomic_inc(&md->event_nr);
2038 wake_up(&md->eventq);
2042 * Protected by md->suspend_lock obtained by dm_swap_table().
2044 static void __set_size(struct mapped_device *md, sector_t size)
2046 set_capacity(md->disk, size);
2048 i_size_write(md->bdev->bd_inode, (loff_t)size << SECTOR_SHIFT);
2052 * Return 1 if the queue has a compulsory merge_bvec_fn function.
2054 * If this function returns 0, then the device is either a non-dm
2055 * device without a merge_bvec_fn, or it is a dm device that is
2056 * able to split any bios it receives that are too big.
2058 int dm_queue_merge_is_compulsory(struct request_queue *q)
2060 struct mapped_device *dev_md;
2062 if (!q->merge_bvec_fn)
2065 if (q->make_request_fn == dm_request) {
2066 dev_md = q->queuedata;
2067 if (test_bit(DMF_MERGE_IS_OPTIONAL, &dev_md->flags))
2074 static int dm_device_merge_is_compulsory(struct dm_target *ti,
2075 struct dm_dev *dev, sector_t start,
2076 sector_t len, void *data)
2078 struct block_device *bdev = dev->bdev;
2079 struct request_queue *q = bdev_get_queue(bdev);
2081 return dm_queue_merge_is_compulsory(q);
2085 * Return 1 if it is acceptable to ignore merge_bvec_fn based
2086 * on the properties of the underlying devices.
2088 static int dm_table_merge_is_optional(struct dm_table *table)
2091 struct dm_target *ti;
2093 while (i < dm_table_get_num_targets(table)) {
2094 ti = dm_table_get_target(table, i++);
2096 if (ti->type->iterate_devices &&
2097 ti->type->iterate_devices(ti, dm_device_merge_is_compulsory, NULL))
2105 * Returns old map, which caller must destroy.
2107 static struct dm_table *__bind(struct mapped_device *md, struct dm_table *t,
2108 struct queue_limits *limits)
2110 struct dm_table *old_map;
2111 struct request_queue *q = md->queue;
2113 int merge_is_optional;
2115 size = dm_table_get_size(t);
2118 * Wipe any geometry if the size of the table changed.
2120 if (size != dm_get_size(md))
2121 memset(&md->geometry, 0, sizeof(md->geometry));
2123 __set_size(md, size);
2125 dm_table_event_callback(t, event_callback, md);
2128 * The queue hasn't been stopped yet, if the old table type wasn't
2129 * for request-based during suspension. So stop it to prevent
2130 * I/O mapping before resume.
2131 * This must be done before setting the queue restrictions,
2132 * because request-based dm may be run just after the setting.
2134 if (dm_table_request_based(t) && !blk_queue_stopped(q))
2137 __bind_mempools(md, t);
2139 merge_is_optional = dm_table_merge_is_optional(t);
2142 rcu_assign_pointer(md->map, t);
2143 md->immutable_target_type = dm_table_get_immutable_target_type(t);
2145 dm_table_set_restrictions(t, q, limits);
2146 if (merge_is_optional)
2147 set_bit(DMF_MERGE_IS_OPTIONAL, &md->flags);
2149 clear_bit(DMF_MERGE_IS_OPTIONAL, &md->flags);
2156 * Returns unbound table for the caller to free.
2158 static struct dm_table *__unbind(struct mapped_device *md)
2160 struct dm_table *map = md->map;
2165 dm_table_event_callback(map, NULL, NULL);
2166 rcu_assign_pointer(md->map, NULL);
2173 * Constructor for a new device.
2175 int dm_create(int minor, struct mapped_device **result)
2177 struct mapped_device *md;
2179 md = alloc_dev(minor);
2190 * Functions to manage md->type.
2191 * All are required to hold md->type_lock.
2193 void dm_lock_md_type(struct mapped_device *md)
2195 mutex_lock(&md->type_lock);
2198 void dm_unlock_md_type(struct mapped_device *md)
2200 mutex_unlock(&md->type_lock);
2203 void dm_set_md_type(struct mapped_device *md, unsigned type)
2205 BUG_ON(!mutex_is_locked(&md->type_lock));
2209 unsigned dm_get_md_type(struct mapped_device *md)
2211 BUG_ON(!mutex_is_locked(&md->type_lock));
2215 struct target_type *dm_get_immutable_target_type(struct mapped_device *md)
2217 return md->immutable_target_type;
2221 * The queue_limits are only valid as long as you have a reference
2224 struct queue_limits *dm_get_queue_limits(struct mapped_device *md)
2226 BUG_ON(!atomic_read(&md->holders));
2227 return &md->queue->limits;
2229 EXPORT_SYMBOL_GPL(dm_get_queue_limits);
2232 * Fully initialize a request-based queue (->elevator, ->request_fn, etc).
2234 static int dm_init_request_based_queue(struct mapped_device *md)
2236 struct request_queue *q = NULL;
2238 if (md->queue->elevator)
2241 /* Fully initialize the queue */
2242 q = blk_init_allocated_queue(md->queue, dm_request_fn, NULL);
2247 dm_init_md_queue(md);
2248 blk_queue_softirq_done(md->queue, dm_softirq_done);
2249 blk_queue_prep_rq(md->queue, dm_prep_fn);
2250 blk_queue_lld_busy(md->queue, dm_lld_busy);
2252 elv_register_queue(md->queue);
2258 * Setup the DM device's queue based on md's type
2260 int dm_setup_md_queue(struct mapped_device *md)
2262 if ((dm_get_md_type(md) == DM_TYPE_REQUEST_BASED) &&
2263 !dm_init_request_based_queue(md)) {
2264 DMWARN("Cannot initialize queue for request-based mapped device");
2271 static struct mapped_device *dm_find_md(dev_t dev)
2273 struct mapped_device *md;
2274 unsigned minor = MINOR(dev);
2276 if (MAJOR(dev) != _major || minor >= (1 << MINORBITS))
2279 spin_lock(&_minor_lock);
2281 md = idr_find(&_minor_idr, minor);
2282 if (md && (md == MINOR_ALLOCED ||
2283 (MINOR(disk_devt(dm_disk(md))) != minor) ||
2284 dm_deleting_md(md) ||
2285 test_bit(DMF_FREEING, &md->flags))) {
2291 spin_unlock(&_minor_lock);
2296 struct mapped_device *dm_get_md(dev_t dev)
2298 struct mapped_device *md = dm_find_md(dev);
2305 EXPORT_SYMBOL_GPL(dm_get_md);
2307 void *dm_get_mdptr(struct mapped_device *md)
2309 return md->interface_ptr;
2312 void dm_set_mdptr(struct mapped_device *md, void *ptr)
2314 md->interface_ptr = ptr;
2317 void dm_get(struct mapped_device *md)
2319 atomic_inc(&md->holders);
2320 BUG_ON(test_bit(DMF_FREEING, &md->flags));
2323 const char *dm_device_name(struct mapped_device *md)
2327 EXPORT_SYMBOL_GPL(dm_device_name);
2329 static void __dm_destroy(struct mapped_device *md, bool wait)
2331 struct dm_table *map;
2336 spin_lock(&_minor_lock);
2337 map = dm_get_live_table(md, &srcu_idx);
2338 idr_replace(&_minor_idr, MINOR_ALLOCED, MINOR(disk_devt(dm_disk(md))));
2339 set_bit(DMF_FREEING, &md->flags);
2340 spin_unlock(&_minor_lock);
2342 if (!dm_suspended_md(md)) {
2343 dm_table_presuspend_targets(map);
2344 dm_table_postsuspend_targets(map);
2347 /* dm_put_live_table must be before msleep, otherwise deadlock is possible */
2348 dm_put_live_table(md, srcu_idx);
2351 * Rare, but there may be I/O requests still going to complete,
2352 * for example. Wait for all references to disappear.
2353 * No one should increment the reference count of the mapped_device,
2354 * after the mapped_device state becomes DMF_FREEING.
2357 while (atomic_read(&md->holders))
2359 else if (atomic_read(&md->holders))
2360 DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)",
2361 dm_device_name(md), atomic_read(&md->holders));
2364 dm_table_destroy(__unbind(md));
2368 void dm_destroy(struct mapped_device *md)
2370 __dm_destroy(md, true);
2373 void dm_destroy_immediate(struct mapped_device *md)
2375 __dm_destroy(md, false);
2378 void dm_put(struct mapped_device *md)
2380 atomic_dec(&md->holders);
2382 EXPORT_SYMBOL_GPL(dm_put);
2384 static int dm_wait_for_completion(struct mapped_device *md, int interruptible)
2387 DECLARE_WAITQUEUE(wait, current);
2389 add_wait_queue(&md->wait, &wait);
2392 set_current_state(interruptible);
2394 if (!md_in_flight(md))
2397 if (interruptible == TASK_INTERRUPTIBLE &&
2398 signal_pending(current)) {
2405 set_current_state(TASK_RUNNING);
2407 remove_wait_queue(&md->wait, &wait);
2413 * Process the deferred bios
2415 static void dm_wq_work(struct work_struct *work)
2417 struct mapped_device *md = container_of(work, struct mapped_device,
2421 struct dm_table *map;
2423 map = dm_get_live_table(md, &srcu_idx);
2425 while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
2426 spin_lock_irq(&md->deferred_lock);
2427 c = bio_list_pop(&md->deferred);
2428 spin_unlock_irq(&md->deferred_lock);
2433 if (dm_request_based(md))
2434 generic_make_request(c);
2436 __split_and_process_bio(md, map, c);
2439 dm_put_live_table(md, srcu_idx);
2442 static void dm_queue_flush(struct mapped_device *md)
2444 clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2445 smp_mb__after_clear_bit();
2446 queue_work(md->wq, &md->work);
2450 * Swap in a new table, returning the old one for the caller to destroy.
2452 struct dm_table *dm_swap_table(struct mapped_device *md, struct dm_table *table)
2454 struct dm_table *live_map = NULL, *map = ERR_PTR(-EINVAL);
2455 struct queue_limits limits;
2458 mutex_lock(&md->suspend_lock);
2460 /* device must be suspended */
2461 if (!dm_suspended_md(md))
2465 * If the new table has no data devices, retain the existing limits.
2466 * This helps multipath with queue_if_no_path if all paths disappear,
2467 * then new I/O is queued based on these limits, and then some paths
2470 if (dm_table_has_no_data_devices(table)) {
2471 live_map = dm_get_live_table_fast(md);
2473 limits = md->queue->limits;
2474 dm_put_live_table_fast(md);
2478 r = dm_calculate_queue_limits(table, &limits);
2485 map = __bind(md, table, &limits);
2488 mutex_unlock(&md->suspend_lock);
2493 * Functions to lock and unlock any filesystem running on the
2496 static int lock_fs(struct mapped_device *md)
2500 WARN_ON(md->frozen_sb);
2502 md->frozen_sb = freeze_bdev(md->bdev);
2503 if (IS_ERR(md->frozen_sb)) {
2504 r = PTR_ERR(md->frozen_sb);
2505 md->frozen_sb = NULL;
2509 set_bit(DMF_FROZEN, &md->flags);
2514 static void unlock_fs(struct mapped_device *md)
2516 if (!test_bit(DMF_FROZEN, &md->flags))
2519 thaw_bdev(md->bdev, md->frozen_sb);
2520 md->frozen_sb = NULL;
2521 clear_bit(DMF_FROZEN, &md->flags);
2525 * We need to be able to change a mapping table under a mounted
2526 * filesystem. For example we might want to move some data in
2527 * the background. Before the table can be swapped with
2528 * dm_bind_table, dm_suspend must be called to flush any in
2529 * flight bios and ensure that any further io gets deferred.
2532 * Suspend mechanism in request-based dm.
2534 * 1. Flush all I/Os by lock_fs() if needed.
2535 * 2. Stop dispatching any I/O by stopping the request_queue.
2536 * 3. Wait for all in-flight I/Os to be completed or requeued.
2538 * To abort suspend, start the request_queue.
2540 int dm_suspend(struct mapped_device *md, unsigned suspend_flags)
2542 struct dm_table *map = NULL;
2544 int do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG ? 1 : 0;
2545 int noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG ? 1 : 0;
2547 mutex_lock(&md->suspend_lock);
2549 if (dm_suspended_md(md)) {
2557 * DMF_NOFLUSH_SUSPENDING must be set before presuspend.
2558 * This flag is cleared before dm_suspend returns.
2561 set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2563 /* This does not get reverted if there's an error later. */
2564 dm_table_presuspend_targets(map);
2567 * Flush I/O to the device.
2568 * Any I/O submitted after lock_fs() may not be flushed.
2569 * noflush takes precedence over do_lockfs.
2570 * (lock_fs() flushes I/Os and waits for them to complete.)
2572 if (!noflush && do_lockfs) {
2579 * Here we must make sure that no processes are submitting requests
2580 * to target drivers i.e. no one may be executing
2581 * __split_and_process_bio. This is called from dm_request and
2584 * To get all processes out of __split_and_process_bio in dm_request,
2585 * we take the write lock. To prevent any process from reentering
2586 * __split_and_process_bio from dm_request and quiesce the thread
2587 * (dm_wq_work), we set BMF_BLOCK_IO_FOR_SUSPEND and call
2588 * flush_workqueue(md->wq).
2590 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2591 synchronize_srcu(&md->io_barrier);
2594 * Stop md->queue before flushing md->wq in case request-based
2595 * dm defers requests to md->wq from md->queue.
2597 if (dm_request_based(md))
2598 stop_queue(md->queue);
2600 flush_workqueue(md->wq);
2603 * At this point no more requests are entering target request routines.
2604 * We call dm_wait_for_completion to wait for all existing requests
2607 r = dm_wait_for_completion(md, TASK_INTERRUPTIBLE);
2610 clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2611 synchronize_srcu(&md->io_barrier);
2613 /* were we interrupted ? */
2617 if (dm_request_based(md))
2618 start_queue(md->queue);
2621 goto out_unlock; /* pushback list is already flushed, so skip flush */
2625 * If dm_wait_for_completion returned 0, the device is completely
2626 * quiescent now. There is no request-processing activity. All new
2627 * requests are being added to md->deferred list.
2630 set_bit(DMF_SUSPENDED, &md->flags);
2632 dm_table_postsuspend_targets(map);
2635 mutex_unlock(&md->suspend_lock);
2639 int dm_resume(struct mapped_device *md)
2642 struct dm_table *map = NULL;
2644 mutex_lock(&md->suspend_lock);
2645 if (!dm_suspended_md(md))
2649 if (!map || !dm_table_get_size(map))
2652 r = dm_table_resume_targets(map);
2659 * Flushing deferred I/Os must be done after targets are resumed
2660 * so that mapping of targets can work correctly.
2661 * Request-based dm is queueing the deferred I/Os in its request_queue.
2663 if (dm_request_based(md))
2664 start_queue(md->queue);
2668 clear_bit(DMF_SUSPENDED, &md->flags);
2672 mutex_unlock(&md->suspend_lock);
2678 * Internal suspend/resume works like userspace-driven suspend. It waits
2679 * until all bios finish and prevents issuing new bios to the target drivers.
2680 * It may be used only from the kernel.
2682 * Internal suspend holds md->suspend_lock, which prevents interaction with
2683 * userspace-driven suspend.
2686 void dm_internal_suspend(struct mapped_device *md)
2688 mutex_lock(&md->suspend_lock);
2689 if (dm_suspended_md(md))
2692 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2693 synchronize_srcu(&md->io_barrier);
2694 flush_workqueue(md->wq);
2695 dm_wait_for_completion(md, TASK_UNINTERRUPTIBLE);
2698 void dm_internal_resume(struct mapped_device *md)
2700 if (dm_suspended_md(md))
2706 mutex_unlock(&md->suspend_lock);
2709 /*-----------------------------------------------------------------
2710 * Event notification.
2711 *---------------------------------------------------------------*/
2712 int dm_kobject_uevent(struct mapped_device *md, enum kobject_action action,
2715 char udev_cookie[DM_COOKIE_LENGTH];
2716 char *envp[] = { udev_cookie, NULL };
2719 return kobject_uevent(&disk_to_dev(md->disk)->kobj, action);
2721 snprintf(udev_cookie, DM_COOKIE_LENGTH, "%s=%u",
2722 DM_COOKIE_ENV_VAR_NAME, cookie);
2723 return kobject_uevent_env(&disk_to_dev(md->disk)->kobj,
2728 uint32_t dm_next_uevent_seq(struct mapped_device *md)
2730 return atomic_add_return(1, &md->uevent_seq);
2733 uint32_t dm_get_event_nr(struct mapped_device *md)
2735 return atomic_read(&md->event_nr);
2738 int dm_wait_event(struct mapped_device *md, int event_nr)
2740 return wait_event_interruptible(md->eventq,
2741 (event_nr != atomic_read(&md->event_nr)));
2744 void dm_uevent_add(struct mapped_device *md, struct list_head *elist)
2746 unsigned long flags;
2748 spin_lock_irqsave(&md->uevent_lock, flags);
2749 list_add(elist, &md->uevent_list);
2750 spin_unlock_irqrestore(&md->uevent_lock, flags);
2754 * The gendisk is only valid as long as you have a reference
2757 struct gendisk *dm_disk(struct mapped_device *md)
2762 struct kobject *dm_kobject(struct mapped_device *md)
2764 return &md->kobj_holder.kobj;
2767 struct mapped_device *dm_get_from_kobject(struct kobject *kobj)
2769 struct mapped_device *md;
2771 md = container_of(kobj, struct mapped_device, kobj_holder.kobj);
2773 if (test_bit(DMF_FREEING, &md->flags) ||
2781 int dm_suspended_md(struct mapped_device *md)
2783 return test_bit(DMF_SUSPENDED, &md->flags);
2786 int dm_test_deferred_remove_flag(struct mapped_device *md)
2788 return test_bit(DMF_DEFERRED_REMOVE, &md->flags);
2791 int dm_suspended(struct dm_target *ti)
2793 return dm_suspended_md(dm_table_get_md(ti->table));
2795 EXPORT_SYMBOL_GPL(dm_suspended);
2797 int dm_noflush_suspending(struct dm_target *ti)
2799 return __noflush_suspending(dm_table_get_md(ti->table));
2801 EXPORT_SYMBOL_GPL(dm_noflush_suspending);
2803 struct dm_md_mempools *dm_alloc_md_mempools(unsigned type, unsigned integrity, unsigned per_bio_data_size)
2805 struct dm_md_mempools *pools = kzalloc(sizeof(*pools), GFP_KERNEL);
2806 struct kmem_cache *cachep;
2807 unsigned int pool_size;
2808 unsigned int front_pad;
2813 if (type == DM_TYPE_BIO_BASED) {
2815 pool_size = dm_get_reserved_bio_based_ios();
2816 front_pad = roundup(per_bio_data_size, __alignof__(struct dm_target_io)) + offsetof(struct dm_target_io, clone);
2817 } else if (type == DM_TYPE_REQUEST_BASED) {
2818 cachep = _rq_tio_cache;
2819 pool_size = dm_get_reserved_rq_based_ios();
2820 front_pad = offsetof(struct dm_rq_clone_bio_info, clone);
2821 /* per_bio_data_size is not used. See __bind_mempools(). */
2822 WARN_ON(per_bio_data_size != 0);
2826 pools->io_pool = mempool_create_slab_pool(pool_size, cachep);
2827 if (!pools->io_pool)
2830 pools->bs = bioset_create(pool_size, front_pad);
2834 if (integrity && bioset_integrity_create(pools->bs, pool_size))
2840 dm_free_md_mempools(pools);
2845 void dm_free_md_mempools(struct dm_md_mempools *pools)
2851 mempool_destroy(pools->io_pool);
2854 bioset_free(pools->bs);
2859 static const struct block_device_operations dm_blk_dops = {
2860 .open = dm_blk_open,
2861 .release = dm_blk_close,
2862 .ioctl = dm_blk_ioctl,
2863 .getgeo = dm_blk_getgeo,
2864 .owner = THIS_MODULE
2870 module_init(dm_init);
2871 module_exit(dm_exit);
2873 module_param(major, uint, 0);
2874 MODULE_PARM_DESC(major, "The major number of the device mapper");
2876 module_param(reserved_bio_based_ios, uint, S_IRUGO | S_IWUSR);
2877 MODULE_PARM_DESC(reserved_bio_based_ios, "Reserved IOs in bio-based mempools");
2879 module_param(reserved_rq_based_ios, uint, S_IRUGO | S_IWUSR);
2880 MODULE_PARM_DESC(reserved_rq_based_ios, "Reserved IOs in request-based mempools");
2882 MODULE_DESCRIPTION(DM_NAME " driver");
2883 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
2884 MODULE_LICENSE("GPL");