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>
22 #include <linux/wait.h>
23 #include <linux/kthread.h>
24 #include <linux/ktime.h>
25 #include <linux/elevator.h> /* for rq_end_sector() */
26 #include <linux/blk-mq.h>
28 #include <trace/events/block.h>
30 #define DM_MSG_PREFIX "core"
34 * ratelimit state to be used in DMXXX_LIMIT().
36 DEFINE_RATELIMIT_STATE(dm_ratelimit_state,
37 DEFAULT_RATELIMIT_INTERVAL,
38 DEFAULT_RATELIMIT_BURST);
39 EXPORT_SYMBOL(dm_ratelimit_state);
43 * Cookies are numeric values sent with CHANGE and REMOVE
44 * uevents while resuming, removing or renaming the device.
46 #define DM_COOKIE_ENV_VAR_NAME "DM_COOKIE"
47 #define DM_COOKIE_LENGTH 24
49 static const char *_name = DM_NAME;
51 static unsigned int major = 0;
52 static unsigned int _major = 0;
54 static DEFINE_IDR(_minor_idr);
56 static DEFINE_SPINLOCK(_minor_lock);
58 static void do_deferred_remove(struct work_struct *w);
60 static DECLARE_WORK(deferred_remove_work, do_deferred_remove);
62 static struct workqueue_struct *deferred_remove_workqueue;
66 * One of these is allocated per bio.
69 struct mapped_device *md;
73 unsigned long start_time;
74 spinlock_t endio_lock;
75 struct dm_stats_aux stats_aux;
79 * For request-based dm.
80 * One of these is allocated per request.
82 struct dm_rq_target_io {
83 struct mapped_device *md;
85 struct request *orig, *clone;
86 struct kthread_work work;
92 * For request-based dm - the bio clones we allocate are embedded in these
95 * We allocate these with bio_alloc_bioset, using the front_pad parameter when
96 * the bioset is created - this means the bio has to come at the end of the
99 struct dm_rq_clone_bio_info {
101 struct dm_rq_target_io *tio;
105 union map_info *dm_get_rq_mapinfo(struct request *rq)
107 if (rq && rq->end_io_data)
108 return &((struct dm_rq_target_io *)rq->end_io_data)->info;
111 EXPORT_SYMBOL_GPL(dm_get_rq_mapinfo);
113 #define MINOR_ALLOCED ((void *)-1)
116 * Bits for the md->flags field.
118 #define DMF_BLOCK_IO_FOR_SUSPEND 0
119 #define DMF_SUSPENDED 1
121 #define DMF_FREEING 3
122 #define DMF_DELETING 4
123 #define DMF_NOFLUSH_SUSPENDING 5
124 #define DMF_MERGE_IS_OPTIONAL 6
125 #define DMF_DEFERRED_REMOVE 7
126 #define DMF_SUSPENDED_INTERNALLY 8
129 * A dummy definition to make RCU happy.
130 * struct dm_table should never be dereferenced in this file.
137 * Work processed by per-device workqueue.
139 struct mapped_device {
140 struct srcu_struct io_barrier;
141 struct mutex suspend_lock;
146 * The current mapping.
147 * Use dm_get_live_table{_fast} or take suspend_lock for
150 struct dm_table __rcu *map;
152 struct list_head table_devices;
153 struct mutex table_devices_lock;
157 struct request_queue *queue;
159 /* Protect queue and type against concurrent access. */
160 struct mutex type_lock;
162 struct target_type *immutable_target_type;
164 struct gendisk *disk;
170 * A list of ios that arrived while we were suspended.
173 wait_queue_head_t wait;
174 struct work_struct work;
175 struct bio_list deferred;
176 spinlock_t deferred_lock;
179 * Processing queue (flush)
181 struct workqueue_struct *wq;
184 * io objects are allocated from here.
195 wait_queue_head_t eventq;
197 struct list_head uevent_list;
198 spinlock_t uevent_lock; /* Protect access to uevent_list */
201 * freeze/thaw support require holding onto a super block
203 struct super_block *frozen_sb;
204 struct block_device *bdev;
206 /* forced geometry settings */
207 struct hd_geometry geometry;
209 /* kobject and completion */
210 struct dm_kobject_holder kobj_holder;
212 /* zero-length flush that will be cloned and submitted to targets */
213 struct bio flush_bio;
215 /* the number of internal suspends */
216 unsigned internal_suspend_count;
218 struct dm_stats stats;
220 struct kthread_worker kworker;
221 struct task_struct *kworker_task;
223 /* for request-based merge heuristic in dm_request_fn() */
224 unsigned seq_rq_merge_deadline_usecs;
226 sector_t last_rq_pos;
227 ktime_t last_rq_start_time;
229 /* for blk-mq request-based DM support */
230 struct blk_mq_tag_set tag_set;
234 #ifdef CONFIG_DM_MQ_DEFAULT
235 static bool use_blk_mq = true;
237 static bool use_blk_mq = false;
240 bool dm_use_blk_mq(struct mapped_device *md)
242 return md->use_blk_mq;
246 * For mempools pre-allocation at the table loading time.
248 struct dm_md_mempools {
254 struct table_device {
255 struct list_head list;
257 struct dm_dev dm_dev;
260 #define RESERVED_BIO_BASED_IOS 16
261 #define RESERVED_REQUEST_BASED_IOS 256
262 #define RESERVED_MAX_IOS 1024
263 static struct kmem_cache *_io_cache;
264 static struct kmem_cache *_rq_tio_cache;
265 static struct kmem_cache *_rq_cache;
268 * Bio-based DM's mempools' reserved IOs set by the user.
270 static unsigned reserved_bio_based_ios = RESERVED_BIO_BASED_IOS;
273 * Request-based DM's mempools' reserved IOs set by the user.
275 static unsigned reserved_rq_based_ios = RESERVED_REQUEST_BASED_IOS;
277 static unsigned __dm_get_module_param(unsigned *module_param,
278 unsigned def, unsigned max)
280 unsigned param = ACCESS_ONCE(*module_param);
281 unsigned modified_param = 0;
284 modified_param = def;
285 else if (param > max)
286 modified_param = max;
288 if (modified_param) {
289 (void)cmpxchg(module_param, param, modified_param);
290 param = modified_param;
296 unsigned dm_get_reserved_bio_based_ios(void)
298 return __dm_get_module_param(&reserved_bio_based_ios,
299 RESERVED_BIO_BASED_IOS, RESERVED_MAX_IOS);
301 EXPORT_SYMBOL_GPL(dm_get_reserved_bio_based_ios);
303 unsigned dm_get_reserved_rq_based_ios(void)
305 return __dm_get_module_param(&reserved_rq_based_ios,
306 RESERVED_REQUEST_BASED_IOS, RESERVED_MAX_IOS);
308 EXPORT_SYMBOL_GPL(dm_get_reserved_rq_based_ios);
310 static int __init local_init(void)
314 /* allocate a slab for the dm_ios */
315 _io_cache = KMEM_CACHE(dm_io, 0);
319 _rq_tio_cache = KMEM_CACHE(dm_rq_target_io, 0);
321 goto out_free_io_cache;
323 _rq_cache = kmem_cache_create("dm_clone_request", sizeof(struct request),
324 __alignof__(struct request), 0, NULL);
326 goto out_free_rq_tio_cache;
328 r = dm_uevent_init();
330 goto out_free_rq_cache;
332 deferred_remove_workqueue = alloc_workqueue("kdmremove", WQ_UNBOUND, 1);
333 if (!deferred_remove_workqueue) {
335 goto out_uevent_exit;
339 r = register_blkdev(_major, _name);
341 goto out_free_workqueue;
349 destroy_workqueue(deferred_remove_workqueue);
353 kmem_cache_destroy(_rq_cache);
354 out_free_rq_tio_cache:
355 kmem_cache_destroy(_rq_tio_cache);
357 kmem_cache_destroy(_io_cache);
362 static void local_exit(void)
364 flush_scheduled_work();
365 destroy_workqueue(deferred_remove_workqueue);
367 kmem_cache_destroy(_rq_cache);
368 kmem_cache_destroy(_rq_tio_cache);
369 kmem_cache_destroy(_io_cache);
370 unregister_blkdev(_major, _name);
375 DMINFO("cleaned up");
378 static int (*_inits[])(void) __initdata = {
389 static void (*_exits[])(void) = {
400 static int __init dm_init(void)
402 const int count = ARRAY_SIZE(_inits);
406 for (i = 0; i < count; i++) {
421 static void __exit dm_exit(void)
423 int i = ARRAY_SIZE(_exits);
429 * Should be empty by this point.
431 idr_destroy(&_minor_idr);
435 * Block device functions
437 int dm_deleting_md(struct mapped_device *md)
439 return test_bit(DMF_DELETING, &md->flags);
442 static int dm_blk_open(struct block_device *bdev, fmode_t mode)
444 struct mapped_device *md;
446 spin_lock(&_minor_lock);
448 md = bdev->bd_disk->private_data;
452 if (test_bit(DMF_FREEING, &md->flags) ||
453 dm_deleting_md(md)) {
459 atomic_inc(&md->open_count);
461 spin_unlock(&_minor_lock);
463 return md ? 0 : -ENXIO;
466 static void dm_blk_close(struct gendisk *disk, fmode_t mode)
468 struct mapped_device *md;
470 spin_lock(&_minor_lock);
472 md = disk->private_data;
476 if (atomic_dec_and_test(&md->open_count) &&
477 (test_bit(DMF_DEFERRED_REMOVE, &md->flags)))
478 queue_work(deferred_remove_workqueue, &deferred_remove_work);
482 spin_unlock(&_minor_lock);
485 int dm_open_count(struct mapped_device *md)
487 return atomic_read(&md->open_count);
491 * Guarantees nothing is using the device before it's deleted.
493 int dm_lock_for_deletion(struct mapped_device *md, bool mark_deferred, bool only_deferred)
497 spin_lock(&_minor_lock);
499 if (dm_open_count(md)) {
502 set_bit(DMF_DEFERRED_REMOVE, &md->flags);
503 } else if (only_deferred && !test_bit(DMF_DEFERRED_REMOVE, &md->flags))
506 set_bit(DMF_DELETING, &md->flags);
508 spin_unlock(&_minor_lock);
513 int dm_cancel_deferred_remove(struct mapped_device *md)
517 spin_lock(&_minor_lock);
519 if (test_bit(DMF_DELETING, &md->flags))
522 clear_bit(DMF_DEFERRED_REMOVE, &md->flags);
524 spin_unlock(&_minor_lock);
529 static void do_deferred_remove(struct work_struct *w)
531 dm_deferred_remove();
534 sector_t dm_get_size(struct mapped_device *md)
536 return get_capacity(md->disk);
539 struct request_queue *dm_get_md_queue(struct mapped_device *md)
544 struct dm_stats *dm_get_stats(struct mapped_device *md)
549 static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo)
551 struct mapped_device *md = bdev->bd_disk->private_data;
553 return dm_get_geometry(md, geo);
556 static int dm_blk_ioctl(struct block_device *bdev, fmode_t mode,
557 unsigned int cmd, unsigned long arg)
559 struct mapped_device *md = bdev->bd_disk->private_data;
561 struct dm_table *map;
562 struct dm_target *tgt;
566 map = dm_get_live_table(md, &srcu_idx);
568 if (!map || !dm_table_get_size(map))
571 /* We only support devices that have a single target */
572 if (dm_table_get_num_targets(map) != 1)
575 tgt = dm_table_get_target(map, 0);
576 if (!tgt->type->ioctl)
579 if (dm_suspended_md(md)) {
584 r = tgt->type->ioctl(tgt, cmd, arg);
587 dm_put_live_table(md, srcu_idx);
589 if (r == -ENOTCONN) {
597 static struct dm_io *alloc_io(struct mapped_device *md)
599 return mempool_alloc(md->io_pool, GFP_NOIO);
602 static void free_io(struct mapped_device *md, struct dm_io *io)
604 mempool_free(io, md->io_pool);
607 static void free_tio(struct mapped_device *md, struct dm_target_io *tio)
609 bio_put(&tio->clone);
612 static struct dm_rq_target_io *alloc_rq_tio(struct mapped_device *md,
615 return mempool_alloc(md->io_pool, gfp_mask);
618 static void free_rq_tio(struct dm_rq_target_io *tio)
620 mempool_free(tio, tio->md->io_pool);
623 static struct request *alloc_clone_request(struct mapped_device *md,
626 return mempool_alloc(md->rq_pool, gfp_mask);
629 static void free_clone_request(struct mapped_device *md, struct request *rq)
631 mempool_free(rq, md->rq_pool);
634 static int md_in_flight(struct mapped_device *md)
636 return atomic_read(&md->pending[READ]) +
637 atomic_read(&md->pending[WRITE]);
640 static void start_io_acct(struct dm_io *io)
642 struct mapped_device *md = io->md;
643 struct bio *bio = io->bio;
645 int rw = bio_data_dir(bio);
647 io->start_time = jiffies;
649 cpu = part_stat_lock();
650 part_round_stats(cpu, &dm_disk(md)->part0);
652 atomic_set(&dm_disk(md)->part0.in_flight[rw],
653 atomic_inc_return(&md->pending[rw]));
655 if (unlikely(dm_stats_used(&md->stats)))
656 dm_stats_account_io(&md->stats, bio->bi_rw, bio->bi_iter.bi_sector,
657 bio_sectors(bio), false, 0, &io->stats_aux);
660 static void end_io_acct(struct dm_io *io)
662 struct mapped_device *md = io->md;
663 struct bio *bio = io->bio;
664 unsigned long duration = jiffies - io->start_time;
666 int rw = bio_data_dir(bio);
668 generic_end_io_acct(rw, &dm_disk(md)->part0, io->start_time);
670 if (unlikely(dm_stats_used(&md->stats)))
671 dm_stats_account_io(&md->stats, bio->bi_rw, bio->bi_iter.bi_sector,
672 bio_sectors(bio), true, duration, &io->stats_aux);
675 * After this is decremented the bio must not be touched if it is
678 pending = atomic_dec_return(&md->pending[rw]);
679 atomic_set(&dm_disk(md)->part0.in_flight[rw], pending);
680 pending += atomic_read(&md->pending[rw^0x1]);
682 /* nudge anyone waiting on suspend queue */
688 * Add the bio to the list of deferred io.
690 static void queue_io(struct mapped_device *md, struct bio *bio)
694 spin_lock_irqsave(&md->deferred_lock, flags);
695 bio_list_add(&md->deferred, bio);
696 spin_unlock_irqrestore(&md->deferred_lock, flags);
697 queue_work(md->wq, &md->work);
701 * Everyone (including functions in this file), should use this
702 * function to access the md->map field, and make sure they call
703 * dm_put_live_table() when finished.
705 struct dm_table *dm_get_live_table(struct mapped_device *md, int *srcu_idx) __acquires(md->io_barrier)
707 *srcu_idx = srcu_read_lock(&md->io_barrier);
709 return srcu_dereference(md->map, &md->io_barrier);
712 void dm_put_live_table(struct mapped_device *md, int srcu_idx) __releases(md->io_barrier)
714 srcu_read_unlock(&md->io_barrier, srcu_idx);
717 void dm_sync_table(struct mapped_device *md)
719 synchronize_srcu(&md->io_barrier);
720 synchronize_rcu_expedited();
724 * A fast alternative to dm_get_live_table/dm_put_live_table.
725 * The caller must not block between these two functions.
727 static struct dm_table *dm_get_live_table_fast(struct mapped_device *md) __acquires(RCU)
730 return rcu_dereference(md->map);
733 static void dm_put_live_table_fast(struct mapped_device *md) __releases(RCU)
739 * Open a table device so we can use it as a map destination.
741 static int open_table_device(struct table_device *td, dev_t dev,
742 struct mapped_device *md)
744 static char *_claim_ptr = "I belong to device-mapper";
745 struct block_device *bdev;
749 BUG_ON(td->dm_dev.bdev);
751 bdev = blkdev_get_by_dev(dev, td->dm_dev.mode | FMODE_EXCL, _claim_ptr);
753 return PTR_ERR(bdev);
755 r = bd_link_disk_holder(bdev, dm_disk(md));
757 blkdev_put(bdev, td->dm_dev.mode | FMODE_EXCL);
761 td->dm_dev.bdev = bdev;
766 * Close a table device that we've been using.
768 static void close_table_device(struct table_device *td, struct mapped_device *md)
770 if (!td->dm_dev.bdev)
773 bd_unlink_disk_holder(td->dm_dev.bdev, dm_disk(md));
774 blkdev_put(td->dm_dev.bdev, td->dm_dev.mode | FMODE_EXCL);
775 td->dm_dev.bdev = NULL;
778 static struct table_device *find_table_device(struct list_head *l, dev_t dev,
780 struct table_device *td;
782 list_for_each_entry(td, l, list)
783 if (td->dm_dev.bdev->bd_dev == dev && td->dm_dev.mode == mode)
789 int dm_get_table_device(struct mapped_device *md, dev_t dev, fmode_t mode,
790 struct dm_dev **result) {
792 struct table_device *td;
794 mutex_lock(&md->table_devices_lock);
795 td = find_table_device(&md->table_devices, dev, mode);
797 td = kmalloc(sizeof(*td), GFP_KERNEL);
799 mutex_unlock(&md->table_devices_lock);
803 td->dm_dev.mode = mode;
804 td->dm_dev.bdev = NULL;
806 if ((r = open_table_device(td, dev, md))) {
807 mutex_unlock(&md->table_devices_lock);
812 format_dev_t(td->dm_dev.name, dev);
814 atomic_set(&td->count, 0);
815 list_add(&td->list, &md->table_devices);
817 atomic_inc(&td->count);
818 mutex_unlock(&md->table_devices_lock);
820 *result = &td->dm_dev;
823 EXPORT_SYMBOL_GPL(dm_get_table_device);
825 void dm_put_table_device(struct mapped_device *md, struct dm_dev *d)
827 struct table_device *td = container_of(d, struct table_device, dm_dev);
829 mutex_lock(&md->table_devices_lock);
830 if (atomic_dec_and_test(&td->count)) {
831 close_table_device(td, md);
835 mutex_unlock(&md->table_devices_lock);
837 EXPORT_SYMBOL(dm_put_table_device);
839 static void free_table_devices(struct list_head *devices)
841 struct list_head *tmp, *next;
843 list_for_each_safe(tmp, next, devices) {
844 struct table_device *td = list_entry(tmp, struct table_device, list);
846 DMWARN("dm_destroy: %s still exists with %d references",
847 td->dm_dev.name, atomic_read(&td->count));
853 * Get the geometry associated with a dm device
855 int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo)
863 * Set the geometry of a device.
865 int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo)
867 sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors;
869 if (geo->start > sz) {
870 DMWARN("Start sector is beyond the geometry limits.");
879 /*-----------------------------------------------------------------
881 * A more elegant soln is in the works that uses the queue
882 * merge fn, unfortunately there are a couple of changes to
883 * the block layer that I want to make for this. So in the
884 * interests of getting something for people to use I give
885 * you this clearly demarcated crap.
886 *---------------------------------------------------------------*/
888 static int __noflush_suspending(struct mapped_device *md)
890 return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
894 * Decrements the number of outstanding ios that a bio has been
895 * cloned into, completing the original io if necc.
897 static void dec_pending(struct dm_io *io, int error)
902 struct mapped_device *md = io->md;
904 /* Push-back supersedes any I/O errors */
905 if (unlikely(error)) {
906 spin_lock_irqsave(&io->endio_lock, flags);
907 if (!(io->error > 0 && __noflush_suspending(md)))
909 spin_unlock_irqrestore(&io->endio_lock, flags);
912 if (atomic_dec_and_test(&io->io_count)) {
913 if (io->error == DM_ENDIO_REQUEUE) {
915 * Target requested pushing back the I/O.
917 spin_lock_irqsave(&md->deferred_lock, flags);
918 if (__noflush_suspending(md))
919 bio_list_add_head(&md->deferred, io->bio);
921 /* noflush suspend was interrupted. */
923 spin_unlock_irqrestore(&md->deferred_lock, flags);
926 io_error = io->error;
931 if (io_error == DM_ENDIO_REQUEUE)
934 if ((bio->bi_rw & REQ_FLUSH) && bio->bi_iter.bi_size) {
936 * Preflush done for flush with data, reissue
939 bio->bi_rw &= ~REQ_FLUSH;
942 /* done with normal IO or empty flush */
943 trace_block_bio_complete(md->queue, bio, io_error);
944 bio_endio(bio, io_error);
949 static void disable_write_same(struct mapped_device *md)
951 struct queue_limits *limits = dm_get_queue_limits(md);
953 /* device doesn't really support WRITE SAME, disable it */
954 limits->max_write_same_sectors = 0;
957 static void clone_endio(struct bio *bio, int error)
960 struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone);
961 struct dm_io *io = tio->io;
962 struct mapped_device *md = tio->io->md;
963 dm_endio_fn endio = tio->ti->type->end_io;
965 if (!bio_flagged(bio, BIO_UPTODATE) && !error)
969 r = endio(tio->ti, bio, error);
970 if (r < 0 || r == DM_ENDIO_REQUEUE)
972 * error and requeue request are handled
976 else if (r == DM_ENDIO_INCOMPLETE)
977 /* The target will handle the io */
980 DMWARN("unimplemented target endio return value: %d", r);
985 if (unlikely(r == -EREMOTEIO && (bio->bi_rw & REQ_WRITE_SAME) &&
986 !bdev_get_queue(bio->bi_bdev)->limits.max_write_same_sectors))
987 disable_write_same(md);
990 dec_pending(io, error);
994 * Partial completion handling for request-based dm
996 static void end_clone_bio(struct bio *clone, int error)
998 struct dm_rq_clone_bio_info *info =
999 container_of(clone, struct dm_rq_clone_bio_info, clone);
1000 struct dm_rq_target_io *tio = info->tio;
1001 struct bio *bio = info->orig;
1002 unsigned int nr_bytes = info->orig->bi_iter.bi_size;
1008 * An error has already been detected on the request.
1009 * Once error occurred, just let clone->end_io() handle
1015 * Don't notice the error to the upper layer yet.
1016 * The error handling decision is made by the target driver,
1017 * when the request is completed.
1024 * I/O for the bio successfully completed.
1025 * Notice the data completion to the upper layer.
1029 * bios are processed from the head of the list.
1030 * So the completing bio should always be rq->bio.
1031 * If it's not, something wrong is happening.
1033 if (tio->orig->bio != bio)
1034 DMERR("bio completion is going in the middle of the request");
1037 * Update the original request.
1038 * Do not use blk_end_request() here, because it may complete
1039 * the original request before the clone, and break the ordering.
1041 blk_update_request(tio->orig, 0, nr_bytes);
1044 static struct dm_rq_target_io *tio_from_request(struct request *rq)
1046 return (rq->q->mq_ops ? blk_mq_rq_to_pdu(rq) : rq->special);
1050 * Don't touch any member of the md after calling this function because
1051 * the md may be freed in dm_put() at the end of this function.
1052 * Or do dm_get() before calling this function and dm_put() later.
1054 static void rq_completed(struct mapped_device *md, int rw, bool run_queue)
1056 int nr_requests_pending;
1058 atomic_dec(&md->pending[rw]);
1060 /* nudge anyone waiting on suspend queue */
1061 nr_requests_pending = md_in_flight(md);
1062 if (!nr_requests_pending)
1066 * Run this off this callpath, as drivers could invoke end_io while
1067 * inside their request_fn (and holding the queue lock). Calling
1068 * back into ->request_fn() could deadlock attempting to grab the
1072 if (md->queue->mq_ops)
1073 blk_mq_run_hw_queues(md->queue, true);
1074 else if (!nr_requests_pending ||
1075 (nr_requests_pending >= md->queue->nr_congestion_on))
1076 blk_run_queue_async(md->queue);
1080 * dm_put() must be at the end of this function. See the comment above
1085 static void free_rq_clone(struct request *clone, bool must_be_mapped)
1087 struct dm_rq_target_io *tio = clone->end_io_data;
1088 struct mapped_device *md = tio->md;
1090 WARN_ON_ONCE(must_be_mapped && !clone->q);
1092 blk_rq_unprep_clone(clone);
1094 if (md->type == DM_TYPE_MQ_REQUEST_BASED)
1095 /* stacked on blk-mq queue(s) */
1096 tio->ti->type->release_clone_rq(clone);
1097 else if (!md->queue->mq_ops)
1098 /* request_fn queue stacked on request_fn queue(s) */
1099 free_clone_request(md, clone);
1101 * NOTE: for the blk-mq queue stacked on request_fn queue(s) case:
1102 * no need to call free_clone_request() because we leverage blk-mq by
1103 * allocating the clone at the end of the blk-mq pdu (see: clone_rq)
1106 if (!md->queue->mq_ops)
1111 * Complete the clone and the original request.
1112 * Must be called without clone's queue lock held,
1113 * see end_clone_request() for more details.
1115 static void dm_end_request(struct request *clone, int error)
1117 int rw = rq_data_dir(clone);
1118 struct dm_rq_target_io *tio = clone->end_io_data;
1119 struct mapped_device *md = tio->md;
1120 struct request *rq = tio->orig;
1122 if (rq->cmd_type == REQ_TYPE_BLOCK_PC) {
1123 rq->errors = clone->errors;
1124 rq->resid_len = clone->resid_len;
1128 * We are using the sense buffer of the original
1130 * So setting the length of the sense data is enough.
1132 rq->sense_len = clone->sense_len;
1135 free_rq_clone(clone, true);
1137 blk_end_request_all(rq, error);
1139 blk_mq_end_request(rq, error);
1140 rq_completed(md, rw, true);
1143 static void dm_unprep_request(struct request *rq)
1145 struct dm_rq_target_io *tio = tio_from_request(rq);
1146 struct request *clone = tio->clone;
1148 if (!rq->q->mq_ops) {
1150 rq->cmd_flags &= ~REQ_DONTPREP;
1154 free_rq_clone(clone, false);
1158 * Requeue the original request of a clone.
1160 static void old_requeue_request(struct request *rq)
1162 struct request_queue *q = rq->q;
1163 unsigned long flags;
1165 spin_lock_irqsave(q->queue_lock, flags);
1166 blk_requeue_request(q, rq);
1167 spin_unlock_irqrestore(q->queue_lock, flags);
1170 static void dm_requeue_unmapped_original_request(struct mapped_device *md,
1173 int rw = rq_data_dir(rq);
1175 dm_unprep_request(rq);
1178 old_requeue_request(rq);
1180 blk_mq_requeue_request(rq);
1181 blk_mq_kick_requeue_list(rq->q);
1184 rq_completed(md, rw, false);
1187 static void dm_requeue_unmapped_request(struct request *clone)
1189 struct dm_rq_target_io *tio = clone->end_io_data;
1191 dm_requeue_unmapped_original_request(tio->md, tio->orig);
1194 static void old_stop_queue(struct request_queue *q)
1196 unsigned long flags;
1198 if (blk_queue_stopped(q))
1201 spin_lock_irqsave(q->queue_lock, flags);
1203 spin_unlock_irqrestore(q->queue_lock, flags);
1206 static void stop_queue(struct request_queue *q)
1211 blk_mq_stop_hw_queues(q);
1214 static void old_start_queue(struct request_queue *q)
1216 unsigned long flags;
1218 spin_lock_irqsave(q->queue_lock, flags);
1219 if (blk_queue_stopped(q))
1221 spin_unlock_irqrestore(q->queue_lock, flags);
1224 static void start_queue(struct request_queue *q)
1229 blk_mq_start_stopped_hw_queues(q, true);
1232 static void dm_done(struct request *clone, int error, bool mapped)
1235 struct dm_rq_target_io *tio = clone->end_io_data;
1236 dm_request_endio_fn rq_end_io = NULL;
1239 rq_end_io = tio->ti->type->rq_end_io;
1241 if (mapped && rq_end_io)
1242 r = rq_end_io(tio->ti, clone, error, &tio->info);
1245 if (unlikely(r == -EREMOTEIO && (clone->cmd_flags & REQ_WRITE_SAME) &&
1246 !clone->q->limits.max_write_same_sectors))
1247 disable_write_same(tio->md);
1250 /* The target wants to complete the I/O */
1251 dm_end_request(clone, r);
1252 else if (r == DM_ENDIO_INCOMPLETE)
1253 /* The target will handle the I/O */
1255 else if (r == DM_ENDIO_REQUEUE)
1256 /* The target wants to requeue the I/O */
1257 dm_requeue_unmapped_request(clone);
1259 DMWARN("unimplemented target endio return value: %d", r);
1265 * Request completion handler for request-based dm
1267 static void dm_softirq_done(struct request *rq)
1270 struct dm_rq_target_io *tio = tio_from_request(rq);
1271 struct request *clone = tio->clone;
1275 rw = rq_data_dir(rq);
1276 if (!rq->q->mq_ops) {
1277 blk_end_request_all(rq, tio->error);
1278 rq_completed(tio->md, rw, false);
1281 blk_mq_end_request(rq, tio->error);
1282 rq_completed(tio->md, rw, false);
1287 if (rq->cmd_flags & REQ_FAILED)
1290 dm_done(clone, tio->error, mapped);
1294 * Complete the clone and the original request with the error status
1295 * through softirq context.
1297 static void dm_complete_request(struct request *rq, int error)
1299 struct dm_rq_target_io *tio = tio_from_request(rq);
1302 blk_complete_request(rq);
1306 * Complete the not-mapped clone and the original request with the error status
1307 * through softirq context.
1308 * Target's rq_end_io() function isn't called.
1309 * This may be used when the target's map_rq() or clone_and_map_rq() functions fail.
1311 static void dm_kill_unmapped_request(struct request *rq, int error)
1313 rq->cmd_flags |= REQ_FAILED;
1314 dm_complete_request(rq, error);
1318 * Called with the clone's queue lock held (for non-blk-mq)
1320 static void end_clone_request(struct request *clone, int error)
1322 struct dm_rq_target_io *tio = clone->end_io_data;
1324 if (!clone->q->mq_ops) {
1326 * For just cleaning up the information of the queue in which
1327 * the clone was dispatched.
1328 * The clone is *NOT* freed actually here because it is alloced
1329 * from dm own mempool (REQ_ALLOCED isn't set).
1331 __blk_put_request(clone->q, clone);
1335 * Actual request completion is done in a softirq context which doesn't
1336 * hold the clone's queue lock. Otherwise, deadlock could occur because:
1337 * - another request may be submitted by the upper level driver
1338 * of the stacking during the completion
1339 * - the submission which requires queue lock may be done
1340 * against this clone's queue
1342 dm_complete_request(tio->orig, error);
1346 * Return maximum size of I/O possible at the supplied sector up to the current
1349 static sector_t max_io_len_target_boundary(sector_t sector, struct dm_target *ti)
1351 sector_t target_offset = dm_target_offset(ti, sector);
1353 return ti->len - target_offset;
1356 static sector_t max_io_len(sector_t sector, struct dm_target *ti)
1358 sector_t len = max_io_len_target_boundary(sector, ti);
1359 sector_t offset, max_len;
1362 * Does the target need to split even further?
1364 if (ti->max_io_len) {
1365 offset = dm_target_offset(ti, sector);
1366 if (unlikely(ti->max_io_len & (ti->max_io_len - 1)))
1367 max_len = sector_div(offset, ti->max_io_len);
1369 max_len = offset & (ti->max_io_len - 1);
1370 max_len = ti->max_io_len - max_len;
1379 int dm_set_target_max_io_len(struct dm_target *ti, sector_t len)
1381 if (len > UINT_MAX) {
1382 DMERR("Specified maximum size of target IO (%llu) exceeds limit (%u)",
1383 (unsigned long long)len, UINT_MAX);
1384 ti->error = "Maximum size of target IO is too large";
1388 ti->max_io_len = (uint32_t) len;
1392 EXPORT_SYMBOL_GPL(dm_set_target_max_io_len);
1395 * A target may call dm_accept_partial_bio only from the map routine. It is
1396 * allowed for all bio types except REQ_FLUSH.
1398 * dm_accept_partial_bio informs the dm that the target only wants to process
1399 * additional n_sectors sectors of the bio and the rest of the data should be
1400 * sent in a next bio.
1402 * A diagram that explains the arithmetics:
1403 * +--------------------+---------------+-------+
1405 * +--------------------+---------------+-------+
1407 * <-------------- *tio->len_ptr --------------->
1408 * <------- bi_size ------->
1411 * Region 1 was already iterated over with bio_advance or similar function.
1412 * (it may be empty if the target doesn't use bio_advance)
1413 * Region 2 is the remaining bio size that the target wants to process.
1414 * (it may be empty if region 1 is non-empty, although there is no reason
1416 * The target requires that region 3 is to be sent in the next bio.
1418 * If the target wants to receive multiple copies of the bio (via num_*bios, etc),
1419 * the partially processed part (the sum of regions 1+2) must be the same for all
1420 * copies of the bio.
1422 void dm_accept_partial_bio(struct bio *bio, unsigned n_sectors)
1424 struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone);
1425 unsigned bi_size = bio->bi_iter.bi_size >> SECTOR_SHIFT;
1426 BUG_ON(bio->bi_rw & REQ_FLUSH);
1427 BUG_ON(bi_size > *tio->len_ptr);
1428 BUG_ON(n_sectors > bi_size);
1429 *tio->len_ptr -= bi_size - n_sectors;
1430 bio->bi_iter.bi_size = n_sectors << SECTOR_SHIFT;
1432 EXPORT_SYMBOL_GPL(dm_accept_partial_bio);
1434 static void __map_bio(struct dm_target_io *tio)
1438 struct mapped_device *md;
1439 struct bio *clone = &tio->clone;
1440 struct dm_target *ti = tio->ti;
1442 clone->bi_end_io = clone_endio;
1445 * Map the clone. If r == 0 we don't need to do
1446 * anything, the target has assumed ownership of
1449 atomic_inc(&tio->io->io_count);
1450 sector = clone->bi_iter.bi_sector;
1451 r = ti->type->map(ti, clone);
1452 if (r == DM_MAPIO_REMAPPED) {
1453 /* the bio has been remapped so dispatch it */
1455 trace_block_bio_remap(bdev_get_queue(clone->bi_bdev), clone,
1456 tio->io->bio->bi_bdev->bd_dev, sector);
1458 generic_make_request(clone);
1459 } else if (r < 0 || r == DM_MAPIO_REQUEUE) {
1460 /* error the io and bail out, or requeue it if needed */
1462 dec_pending(tio->io, r);
1465 DMWARN("unimplemented target map return value: %d", r);
1471 struct mapped_device *md;
1472 struct dm_table *map;
1476 unsigned sector_count;
1479 static void bio_setup_sector(struct bio *bio, sector_t sector, unsigned len)
1481 bio->bi_iter.bi_sector = sector;
1482 bio->bi_iter.bi_size = to_bytes(len);
1486 * Creates a bio that consists of range of complete bvecs.
1488 static void clone_bio(struct dm_target_io *tio, struct bio *bio,
1489 sector_t sector, unsigned len)
1491 struct bio *clone = &tio->clone;
1493 __bio_clone_fast(clone, bio);
1495 if (bio_integrity(bio))
1496 bio_integrity_clone(clone, bio, GFP_NOIO);
1498 bio_advance(clone, to_bytes(sector - clone->bi_iter.bi_sector));
1499 clone->bi_iter.bi_size = to_bytes(len);
1501 if (bio_integrity(bio))
1502 bio_integrity_trim(clone, 0, len);
1505 static struct dm_target_io *alloc_tio(struct clone_info *ci,
1506 struct dm_target *ti,
1507 unsigned target_bio_nr)
1509 struct dm_target_io *tio;
1512 clone = bio_alloc_bioset(GFP_NOIO, 0, ci->md->bs);
1513 tio = container_of(clone, struct dm_target_io, clone);
1517 tio->target_bio_nr = target_bio_nr;
1522 static void __clone_and_map_simple_bio(struct clone_info *ci,
1523 struct dm_target *ti,
1524 unsigned target_bio_nr, unsigned *len)
1526 struct dm_target_io *tio = alloc_tio(ci, ti, target_bio_nr);
1527 struct bio *clone = &tio->clone;
1531 __bio_clone_fast(clone, ci->bio);
1533 bio_setup_sector(clone, ci->sector, *len);
1538 static void __send_duplicate_bios(struct clone_info *ci, struct dm_target *ti,
1539 unsigned num_bios, unsigned *len)
1541 unsigned target_bio_nr;
1543 for (target_bio_nr = 0; target_bio_nr < num_bios; target_bio_nr++)
1544 __clone_and_map_simple_bio(ci, ti, target_bio_nr, len);
1547 static int __send_empty_flush(struct clone_info *ci)
1549 unsigned target_nr = 0;
1550 struct dm_target *ti;
1552 BUG_ON(bio_has_data(ci->bio));
1553 while ((ti = dm_table_get_target(ci->map, target_nr++)))
1554 __send_duplicate_bios(ci, ti, ti->num_flush_bios, NULL);
1559 static void __clone_and_map_data_bio(struct clone_info *ci, struct dm_target *ti,
1560 sector_t sector, unsigned *len)
1562 struct bio *bio = ci->bio;
1563 struct dm_target_io *tio;
1564 unsigned target_bio_nr;
1565 unsigned num_target_bios = 1;
1568 * Does the target want to receive duplicate copies of the bio?
1570 if (bio_data_dir(bio) == WRITE && ti->num_write_bios)
1571 num_target_bios = ti->num_write_bios(ti, bio);
1573 for (target_bio_nr = 0; target_bio_nr < num_target_bios; target_bio_nr++) {
1574 tio = alloc_tio(ci, ti, target_bio_nr);
1576 clone_bio(tio, bio, sector, *len);
1581 typedef unsigned (*get_num_bios_fn)(struct dm_target *ti);
1583 static unsigned get_num_discard_bios(struct dm_target *ti)
1585 return ti->num_discard_bios;
1588 static unsigned get_num_write_same_bios(struct dm_target *ti)
1590 return ti->num_write_same_bios;
1593 typedef bool (*is_split_required_fn)(struct dm_target *ti);
1595 static bool is_split_required_for_discard(struct dm_target *ti)
1597 return ti->split_discard_bios;
1600 static int __send_changing_extent_only(struct clone_info *ci,
1601 get_num_bios_fn get_num_bios,
1602 is_split_required_fn is_split_required)
1604 struct dm_target *ti;
1609 ti = dm_table_find_target(ci->map, ci->sector);
1610 if (!dm_target_is_valid(ti))
1614 * Even though the device advertised support for this type of
1615 * request, that does not mean every target supports it, and
1616 * reconfiguration might also have changed that since the
1617 * check was performed.
1619 num_bios = get_num_bios ? get_num_bios(ti) : 0;
1623 if (is_split_required && !is_split_required(ti))
1624 len = min((sector_t)ci->sector_count, max_io_len_target_boundary(ci->sector, ti));
1626 len = min((sector_t)ci->sector_count, max_io_len(ci->sector, ti));
1628 __send_duplicate_bios(ci, ti, num_bios, &len);
1631 } while (ci->sector_count -= len);
1636 static int __send_discard(struct clone_info *ci)
1638 return __send_changing_extent_only(ci, get_num_discard_bios,
1639 is_split_required_for_discard);
1642 static int __send_write_same(struct clone_info *ci)
1644 return __send_changing_extent_only(ci, get_num_write_same_bios, NULL);
1648 * Select the correct strategy for processing a non-flush bio.
1650 static int __split_and_process_non_flush(struct clone_info *ci)
1652 struct bio *bio = ci->bio;
1653 struct dm_target *ti;
1656 if (unlikely(bio->bi_rw & REQ_DISCARD))
1657 return __send_discard(ci);
1658 else if (unlikely(bio->bi_rw & REQ_WRITE_SAME))
1659 return __send_write_same(ci);
1661 ti = dm_table_find_target(ci->map, ci->sector);
1662 if (!dm_target_is_valid(ti))
1665 len = min_t(sector_t, max_io_len(ci->sector, ti), ci->sector_count);
1667 __clone_and_map_data_bio(ci, ti, ci->sector, &len);
1670 ci->sector_count -= len;
1676 * Entry point to split a bio into clones and submit them to the targets.
1678 static void __split_and_process_bio(struct mapped_device *md,
1679 struct dm_table *map, struct bio *bio)
1681 struct clone_info ci;
1684 if (unlikely(!map)) {
1691 ci.io = alloc_io(md);
1693 atomic_set(&ci.io->io_count, 1);
1696 spin_lock_init(&ci.io->endio_lock);
1697 ci.sector = bio->bi_iter.bi_sector;
1699 start_io_acct(ci.io);
1701 if (bio->bi_rw & REQ_FLUSH) {
1702 ci.bio = &ci.md->flush_bio;
1703 ci.sector_count = 0;
1704 error = __send_empty_flush(&ci);
1705 /* dec_pending submits any data associated with flush */
1708 ci.sector_count = bio_sectors(bio);
1709 while (ci.sector_count && !error)
1710 error = __split_and_process_non_flush(&ci);
1713 /* drop the extra reference count */
1714 dec_pending(ci.io, error);
1716 /*-----------------------------------------------------------------
1718 *---------------------------------------------------------------*/
1720 static int dm_merge_bvec(struct request_queue *q,
1721 struct bvec_merge_data *bvm,
1722 struct bio_vec *biovec)
1724 struct mapped_device *md = q->queuedata;
1725 struct dm_table *map = dm_get_live_table_fast(md);
1726 struct dm_target *ti;
1727 sector_t max_sectors;
1733 ti = dm_table_find_target(map, bvm->bi_sector);
1734 if (!dm_target_is_valid(ti))
1738 * Find maximum amount of I/O that won't need splitting
1740 max_sectors = min(max_io_len(bvm->bi_sector, ti),
1741 (sector_t) queue_max_sectors(q));
1742 max_size = (max_sectors << SECTOR_SHIFT) - bvm->bi_size;
1743 if (unlikely(max_size < 0)) /* this shouldn't _ever_ happen */
1747 * merge_bvec_fn() returns number of bytes
1748 * it can accept at this offset
1749 * max is precomputed maximal io size
1751 if (max_size && ti->type->merge)
1752 max_size = ti->type->merge(ti, bvm, biovec, max_size);
1754 * If the target doesn't support merge method and some of the devices
1755 * provided their merge_bvec method (we know this by looking for the
1756 * max_hw_sectors that dm_set_device_limits may set), then we can't
1757 * allow bios with multiple vector entries. So always set max_size
1758 * to 0, and the code below allows just one page.
1760 else if (queue_max_hw_sectors(q) <= PAGE_SIZE >> 9)
1764 dm_put_live_table_fast(md);
1766 * Always allow an entire first page
1768 if (max_size <= biovec->bv_len && !(bvm->bi_size >> SECTOR_SHIFT))
1769 max_size = biovec->bv_len;
1775 * The request function that just remaps the bio built up by
1778 static void dm_make_request(struct request_queue *q, struct bio *bio)
1780 int rw = bio_data_dir(bio);
1781 struct mapped_device *md = q->queuedata;
1783 struct dm_table *map;
1785 map = dm_get_live_table(md, &srcu_idx);
1787 generic_start_io_acct(rw, bio_sectors(bio), &dm_disk(md)->part0);
1789 /* if we're suspended, we have to queue this io for later */
1790 if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))) {
1791 dm_put_live_table(md, srcu_idx);
1793 if (bio_rw(bio) != READA)
1800 __split_and_process_bio(md, map, bio);
1801 dm_put_live_table(md, srcu_idx);
1805 int dm_request_based(struct mapped_device *md)
1807 return blk_queue_stackable(md->queue);
1810 static void dm_dispatch_clone_request(struct request *clone, struct request *rq)
1814 if (blk_queue_io_stat(clone->q))
1815 clone->cmd_flags |= REQ_IO_STAT;
1817 clone->start_time = jiffies;
1818 r = blk_insert_cloned_request(clone->q, clone);
1820 /* must complete clone in terms of original request */
1821 dm_complete_request(rq, r);
1824 static int dm_rq_bio_constructor(struct bio *bio, struct bio *bio_orig,
1827 struct dm_rq_target_io *tio = data;
1828 struct dm_rq_clone_bio_info *info =
1829 container_of(bio, struct dm_rq_clone_bio_info, clone);
1831 info->orig = bio_orig;
1833 bio->bi_end_io = end_clone_bio;
1838 static int setup_clone(struct request *clone, struct request *rq,
1839 struct dm_rq_target_io *tio, gfp_t gfp_mask)
1843 r = blk_rq_prep_clone(clone, rq, tio->md->bs, gfp_mask,
1844 dm_rq_bio_constructor, tio);
1848 clone->cmd = rq->cmd;
1849 clone->cmd_len = rq->cmd_len;
1850 clone->sense = rq->sense;
1851 clone->end_io = end_clone_request;
1852 clone->end_io_data = tio;
1859 static struct request *clone_rq(struct request *rq, struct mapped_device *md,
1860 struct dm_rq_target_io *tio, gfp_t gfp_mask)
1863 * Do not allocate a clone if tio->clone was already set
1864 * (see: dm_mq_queue_rq).
1866 bool alloc_clone = !tio->clone;
1867 struct request *clone;
1870 clone = alloc_clone_request(md, gfp_mask);
1876 blk_rq_init(NULL, clone);
1877 if (setup_clone(clone, rq, tio, gfp_mask)) {
1880 free_clone_request(md, clone);
1887 static void map_tio_request(struct kthread_work *work);
1889 static void init_tio(struct dm_rq_target_io *tio, struct request *rq,
1890 struct mapped_device *md)
1897 memset(&tio->info, 0, sizeof(tio->info));
1898 if (md->kworker_task)
1899 init_kthread_work(&tio->work, map_tio_request);
1902 static struct dm_rq_target_io *prep_tio(struct request *rq,
1903 struct mapped_device *md, gfp_t gfp_mask)
1905 struct dm_rq_target_io *tio;
1907 struct dm_table *table;
1909 tio = alloc_rq_tio(md, gfp_mask);
1913 init_tio(tio, rq, md);
1915 table = dm_get_live_table(md, &srcu_idx);
1916 if (!dm_table_mq_request_based(table)) {
1917 if (!clone_rq(rq, md, tio, gfp_mask)) {
1918 dm_put_live_table(md, srcu_idx);
1923 dm_put_live_table(md, srcu_idx);
1929 * Called with the queue lock held.
1931 static int dm_prep_fn(struct request_queue *q, struct request *rq)
1933 struct mapped_device *md = q->queuedata;
1934 struct dm_rq_target_io *tio;
1936 if (unlikely(rq->special)) {
1937 DMWARN("Already has something in rq->special.");
1938 return BLKPREP_KILL;
1941 tio = prep_tio(rq, md, GFP_ATOMIC);
1943 return BLKPREP_DEFER;
1946 rq->cmd_flags |= REQ_DONTPREP;
1953 * 0 : the request has been processed
1954 * DM_MAPIO_REQUEUE : the original request needs to be requeued
1955 * < 0 : the request was completed due to failure
1957 static int map_request(struct dm_rq_target_io *tio, struct request *rq,
1958 struct mapped_device *md)
1961 struct dm_target *ti = tio->ti;
1962 struct request *clone = NULL;
1966 r = ti->type->map_rq(ti, clone, &tio->info);
1968 r = ti->type->clone_and_map_rq(ti, rq, &tio->info, &clone);
1970 /* The target wants to complete the I/O */
1971 dm_kill_unmapped_request(rq, r);
1975 return DM_MAPIO_REQUEUE;
1976 if (setup_clone(clone, rq, tio, GFP_ATOMIC)) {
1978 ti->type->release_clone_rq(clone);
1979 return DM_MAPIO_REQUEUE;
1984 case DM_MAPIO_SUBMITTED:
1985 /* The target has taken the I/O to submit by itself later */
1987 case DM_MAPIO_REMAPPED:
1988 /* The target has remapped the I/O so dispatch it */
1989 trace_block_rq_remap(clone->q, clone, disk_devt(dm_disk(md)),
1991 dm_dispatch_clone_request(clone, rq);
1993 case DM_MAPIO_REQUEUE:
1994 /* The target wants to requeue the I/O */
1995 dm_requeue_unmapped_request(clone);
1999 DMWARN("unimplemented target map return value: %d", r);
2003 /* The target wants to complete the I/O */
2004 dm_kill_unmapped_request(rq, r);
2011 static void map_tio_request(struct kthread_work *work)
2013 struct dm_rq_target_io *tio = container_of(work, struct dm_rq_target_io, work);
2014 struct request *rq = tio->orig;
2015 struct mapped_device *md = tio->md;
2017 if (map_request(tio, rq, md) == DM_MAPIO_REQUEUE)
2018 dm_requeue_unmapped_original_request(md, rq);
2021 static void dm_start_request(struct mapped_device *md, struct request *orig)
2023 if (!orig->q->mq_ops)
2024 blk_start_request(orig);
2026 blk_mq_start_request(orig);
2027 atomic_inc(&md->pending[rq_data_dir(orig)]);
2029 if (md->seq_rq_merge_deadline_usecs) {
2030 md->last_rq_pos = rq_end_sector(orig);
2031 md->last_rq_rw = rq_data_dir(orig);
2032 md->last_rq_start_time = ktime_get();
2036 * Hold the md reference here for the in-flight I/O.
2037 * We can't rely on the reference count by device opener,
2038 * because the device may be closed during the request completion
2039 * when all bios are completed.
2040 * See the comment in rq_completed() too.
2045 #define MAX_SEQ_RQ_MERGE_DEADLINE_USECS 100000
2047 ssize_t dm_attr_rq_based_seq_io_merge_deadline_show(struct mapped_device *md, char *buf)
2049 return sprintf(buf, "%u\n", md->seq_rq_merge_deadline_usecs);
2052 ssize_t dm_attr_rq_based_seq_io_merge_deadline_store(struct mapped_device *md,
2053 const char *buf, size_t count)
2057 if (!dm_request_based(md) || md->use_blk_mq)
2060 if (kstrtouint(buf, 10, &deadline))
2063 if (deadline > MAX_SEQ_RQ_MERGE_DEADLINE_USECS)
2064 deadline = MAX_SEQ_RQ_MERGE_DEADLINE_USECS;
2066 md->seq_rq_merge_deadline_usecs = deadline;
2071 static bool dm_request_peeked_before_merge_deadline(struct mapped_device *md)
2073 ktime_t kt_deadline;
2075 if (!md->seq_rq_merge_deadline_usecs)
2078 kt_deadline = ns_to_ktime((u64)md->seq_rq_merge_deadline_usecs * NSEC_PER_USEC);
2079 kt_deadline = ktime_add_safe(md->last_rq_start_time, kt_deadline);
2081 return !ktime_after(ktime_get(), kt_deadline);
2085 * q->request_fn for request-based dm.
2086 * Called with the queue lock held.
2088 static void dm_request_fn(struct request_queue *q)
2090 struct mapped_device *md = q->queuedata;
2092 struct dm_table *map = dm_get_live_table(md, &srcu_idx);
2093 struct dm_target *ti;
2095 struct dm_rq_target_io *tio;
2099 * For suspend, check blk_queue_stopped() and increment
2100 * ->pending within a single queue_lock not to increment the
2101 * number of in-flight I/Os after the queue is stopped in
2104 while (!blk_queue_stopped(q)) {
2105 rq = blk_peek_request(q);
2109 /* always use block 0 to find the target for flushes for now */
2111 if (!(rq->cmd_flags & REQ_FLUSH))
2112 pos = blk_rq_pos(rq);
2114 ti = dm_table_find_target(map, pos);
2115 if (!dm_target_is_valid(ti)) {
2117 * Must perform setup, that rq_completed() requires,
2118 * before calling dm_kill_unmapped_request
2120 DMERR_LIMIT("request attempted access beyond the end of device");
2121 dm_start_request(md, rq);
2122 dm_kill_unmapped_request(rq, -EIO);
2126 if (dm_request_peeked_before_merge_deadline(md) &&
2127 md_in_flight(md) && rq->bio && rq->bio->bi_vcnt == 1 &&
2128 md->last_rq_pos == pos && md->last_rq_rw == rq_data_dir(rq))
2131 if (ti->type->busy && ti->type->busy(ti))
2134 dm_start_request(md, rq);
2136 tio = tio_from_request(rq);
2137 /* Establish tio->ti before queuing work (map_tio_request) */
2139 queue_kthread_work(&md->kworker, &tio->work);
2140 BUG_ON(!irqs_disabled());
2146 blk_delay_queue(q, HZ / 100);
2148 dm_put_live_table(md, srcu_idx);
2151 static int dm_any_congested(void *congested_data, int bdi_bits)
2154 struct mapped_device *md = congested_data;
2155 struct dm_table *map;
2157 if (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
2158 map = dm_get_live_table_fast(md);
2161 * Request-based dm cares about only own queue for
2162 * the query about congestion status of request_queue
2164 if (dm_request_based(md))
2165 r = md->queue->backing_dev_info.state &
2168 r = dm_table_any_congested(map, bdi_bits);
2170 dm_put_live_table_fast(md);
2176 /*-----------------------------------------------------------------
2177 * An IDR is used to keep track of allocated minor numbers.
2178 *---------------------------------------------------------------*/
2179 static void free_minor(int minor)
2181 spin_lock(&_minor_lock);
2182 idr_remove(&_minor_idr, minor);
2183 spin_unlock(&_minor_lock);
2187 * See if the device with a specific minor # is free.
2189 static int specific_minor(int minor)
2193 if (minor >= (1 << MINORBITS))
2196 idr_preload(GFP_KERNEL);
2197 spin_lock(&_minor_lock);
2199 r = idr_alloc(&_minor_idr, MINOR_ALLOCED, minor, minor + 1, GFP_NOWAIT);
2201 spin_unlock(&_minor_lock);
2204 return r == -ENOSPC ? -EBUSY : r;
2208 static int next_free_minor(int *minor)
2212 idr_preload(GFP_KERNEL);
2213 spin_lock(&_minor_lock);
2215 r = idr_alloc(&_minor_idr, MINOR_ALLOCED, 0, 1 << MINORBITS, GFP_NOWAIT);
2217 spin_unlock(&_minor_lock);
2225 static const struct block_device_operations dm_blk_dops;
2227 static void dm_wq_work(struct work_struct *work);
2229 static void dm_init_md_queue(struct mapped_device *md)
2232 * Request-based dm devices cannot be stacked on top of bio-based dm
2233 * devices. The type of this dm device may not have been decided yet.
2234 * The type is decided at the first table loading time.
2235 * To prevent problematic device stacking, clear the queue flag
2236 * for request stacking support until then.
2238 * This queue is new, so no concurrency on the queue_flags.
2240 queue_flag_clear_unlocked(QUEUE_FLAG_STACKABLE, md->queue);
2243 static void dm_init_old_md_queue(struct mapped_device *md)
2245 md->use_blk_mq = false;
2246 dm_init_md_queue(md);
2249 * Initialize aspects of queue that aren't relevant for blk-mq
2251 md->queue->queuedata = md;
2252 md->queue->backing_dev_info.congested_fn = dm_any_congested;
2253 md->queue->backing_dev_info.congested_data = md;
2255 blk_queue_bounce_limit(md->queue, BLK_BOUNCE_ANY);
2259 * Allocate and initialise a blank device with a given minor.
2261 static struct mapped_device *alloc_dev(int minor)
2264 struct mapped_device *md = kzalloc(sizeof(*md), GFP_KERNEL);
2268 DMWARN("unable to allocate device, out of memory.");
2272 if (!try_module_get(THIS_MODULE))
2273 goto bad_module_get;
2275 /* get a minor number for the dev */
2276 if (minor == DM_ANY_MINOR)
2277 r = next_free_minor(&minor);
2279 r = specific_minor(minor);
2283 r = init_srcu_struct(&md->io_barrier);
2285 goto bad_io_barrier;
2287 md->use_blk_mq = use_blk_mq;
2288 md->type = DM_TYPE_NONE;
2289 mutex_init(&md->suspend_lock);
2290 mutex_init(&md->type_lock);
2291 mutex_init(&md->table_devices_lock);
2292 spin_lock_init(&md->deferred_lock);
2293 atomic_set(&md->holders, 1);
2294 atomic_set(&md->open_count, 0);
2295 atomic_set(&md->event_nr, 0);
2296 atomic_set(&md->uevent_seq, 0);
2297 INIT_LIST_HEAD(&md->uevent_list);
2298 INIT_LIST_HEAD(&md->table_devices);
2299 spin_lock_init(&md->uevent_lock);
2301 md->queue = blk_alloc_queue(GFP_KERNEL);
2305 dm_init_md_queue(md);
2307 md->disk = alloc_disk(1);
2311 atomic_set(&md->pending[0], 0);
2312 atomic_set(&md->pending[1], 0);
2313 init_waitqueue_head(&md->wait);
2314 INIT_WORK(&md->work, dm_wq_work);
2315 init_waitqueue_head(&md->eventq);
2316 init_completion(&md->kobj_holder.completion);
2317 md->kworker_task = NULL;
2319 md->disk->major = _major;
2320 md->disk->first_minor = minor;
2321 md->disk->fops = &dm_blk_dops;
2322 md->disk->queue = md->queue;
2323 md->disk->private_data = md;
2324 sprintf(md->disk->disk_name, "dm-%d", minor);
2326 format_dev_t(md->name, MKDEV(_major, minor));
2328 md->wq = alloc_workqueue("kdmflush", WQ_MEM_RECLAIM, 0);
2332 md->bdev = bdget_disk(md->disk, 0);
2336 bio_init(&md->flush_bio);
2337 md->flush_bio.bi_bdev = md->bdev;
2338 md->flush_bio.bi_rw = WRITE_FLUSH;
2340 dm_stats_init(&md->stats);
2342 /* Populate the mapping, nobody knows we exist yet */
2343 spin_lock(&_minor_lock);
2344 old_md = idr_replace(&_minor_idr, md, minor);
2345 spin_unlock(&_minor_lock);
2347 BUG_ON(old_md != MINOR_ALLOCED);
2352 destroy_workqueue(md->wq);
2354 del_gendisk(md->disk);
2357 blk_cleanup_queue(md->queue);
2359 cleanup_srcu_struct(&md->io_barrier);
2363 module_put(THIS_MODULE);
2369 static void unlock_fs(struct mapped_device *md);
2371 static void free_dev(struct mapped_device *md)
2373 int minor = MINOR(disk_devt(md->disk));
2376 destroy_workqueue(md->wq);
2378 if (md->kworker_task)
2379 kthread_stop(md->kworker_task);
2381 mempool_destroy(md->io_pool);
2383 mempool_destroy(md->rq_pool);
2385 bioset_free(md->bs);
2387 cleanup_srcu_struct(&md->io_barrier);
2388 free_table_devices(&md->table_devices);
2389 dm_stats_cleanup(&md->stats);
2391 spin_lock(&_minor_lock);
2392 md->disk->private_data = NULL;
2393 spin_unlock(&_minor_lock);
2394 if (blk_get_integrity(md->disk))
2395 blk_integrity_unregister(md->disk);
2396 del_gendisk(md->disk);
2398 blk_cleanup_queue(md->queue);
2400 blk_mq_free_tag_set(&md->tag_set);
2404 module_put(THIS_MODULE);
2408 static void __bind_mempools(struct mapped_device *md, struct dm_table *t)
2410 struct dm_md_mempools *p = dm_table_get_md_mempools(t);
2413 /* The md already has necessary mempools. */
2414 if (dm_table_get_type(t) == DM_TYPE_BIO_BASED) {
2416 * Reload bioset because front_pad may have changed
2417 * because a different table was loaded.
2419 bioset_free(md->bs);
2424 * There's no need to reload with request-based dm
2425 * because the size of front_pad doesn't change.
2426 * Note for future: If you are to reload bioset,
2427 * prep-ed requests in the queue may refer
2428 * to bio from the old bioset, so you must walk
2429 * through the queue to unprep.
2434 BUG_ON(!p || md->io_pool || md->rq_pool || md->bs);
2436 md->io_pool = p->io_pool;
2438 md->rq_pool = p->rq_pool;
2444 /* mempool bind completed, no longer need any mempools in the table */
2445 dm_table_free_md_mempools(t);
2449 * Bind a table to the device.
2451 static void event_callback(void *context)
2453 unsigned long flags;
2455 struct mapped_device *md = (struct mapped_device *) context;
2457 spin_lock_irqsave(&md->uevent_lock, flags);
2458 list_splice_init(&md->uevent_list, &uevents);
2459 spin_unlock_irqrestore(&md->uevent_lock, flags);
2461 dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj);
2463 atomic_inc(&md->event_nr);
2464 wake_up(&md->eventq);
2468 * Protected by md->suspend_lock obtained by dm_swap_table().
2470 static void __set_size(struct mapped_device *md, sector_t size)
2472 set_capacity(md->disk, size);
2474 i_size_write(md->bdev->bd_inode, (loff_t)size << SECTOR_SHIFT);
2478 * Return 1 if the queue has a compulsory merge_bvec_fn function.
2480 * If this function returns 0, then the device is either a non-dm
2481 * device without a merge_bvec_fn, or it is a dm device that is
2482 * able to split any bios it receives that are too big.
2484 int dm_queue_merge_is_compulsory(struct request_queue *q)
2486 struct mapped_device *dev_md;
2488 if (!q->merge_bvec_fn)
2491 if (q->make_request_fn == dm_make_request) {
2492 dev_md = q->queuedata;
2493 if (test_bit(DMF_MERGE_IS_OPTIONAL, &dev_md->flags))
2500 static int dm_device_merge_is_compulsory(struct dm_target *ti,
2501 struct dm_dev *dev, sector_t start,
2502 sector_t len, void *data)
2504 struct block_device *bdev = dev->bdev;
2505 struct request_queue *q = bdev_get_queue(bdev);
2507 return dm_queue_merge_is_compulsory(q);
2511 * Return 1 if it is acceptable to ignore merge_bvec_fn based
2512 * on the properties of the underlying devices.
2514 static int dm_table_merge_is_optional(struct dm_table *table)
2517 struct dm_target *ti;
2519 while (i < dm_table_get_num_targets(table)) {
2520 ti = dm_table_get_target(table, i++);
2522 if (ti->type->iterate_devices &&
2523 ti->type->iterate_devices(ti, dm_device_merge_is_compulsory, NULL))
2531 * Returns old map, which caller must destroy.
2533 static struct dm_table *__bind(struct mapped_device *md, struct dm_table *t,
2534 struct queue_limits *limits)
2536 struct dm_table *old_map;
2537 struct request_queue *q = md->queue;
2539 int merge_is_optional;
2541 size = dm_table_get_size(t);
2544 * Wipe any geometry if the size of the table changed.
2546 if (size != dm_get_size(md))
2547 memset(&md->geometry, 0, sizeof(md->geometry));
2549 __set_size(md, size);
2551 dm_table_event_callback(t, event_callback, md);
2554 * The queue hasn't been stopped yet, if the old table type wasn't
2555 * for request-based during suspension. So stop it to prevent
2556 * I/O mapping before resume.
2557 * This must be done before setting the queue restrictions,
2558 * because request-based dm may be run just after the setting.
2560 if (dm_table_request_based(t))
2563 __bind_mempools(md, t);
2565 merge_is_optional = dm_table_merge_is_optional(t);
2567 old_map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2568 rcu_assign_pointer(md->map, t);
2569 md->immutable_target_type = dm_table_get_immutable_target_type(t);
2571 dm_table_set_restrictions(t, q, limits);
2572 if (merge_is_optional)
2573 set_bit(DMF_MERGE_IS_OPTIONAL, &md->flags);
2575 clear_bit(DMF_MERGE_IS_OPTIONAL, &md->flags);
2583 * Returns unbound table for the caller to free.
2585 static struct dm_table *__unbind(struct mapped_device *md)
2587 struct dm_table *map = rcu_dereference_protected(md->map, 1);
2592 dm_table_event_callback(map, NULL, NULL);
2593 RCU_INIT_POINTER(md->map, NULL);
2600 * Constructor for a new device.
2602 int dm_create(int minor, struct mapped_device **result)
2604 struct mapped_device *md;
2606 md = alloc_dev(minor);
2617 * Functions to manage md->type.
2618 * All are required to hold md->type_lock.
2620 void dm_lock_md_type(struct mapped_device *md)
2622 mutex_lock(&md->type_lock);
2625 void dm_unlock_md_type(struct mapped_device *md)
2627 mutex_unlock(&md->type_lock);
2630 void dm_set_md_type(struct mapped_device *md, unsigned type)
2632 BUG_ON(!mutex_is_locked(&md->type_lock));
2636 unsigned dm_get_md_type(struct mapped_device *md)
2638 BUG_ON(!mutex_is_locked(&md->type_lock));
2642 struct target_type *dm_get_immutable_target_type(struct mapped_device *md)
2644 return md->immutable_target_type;
2648 * The queue_limits are only valid as long as you have a reference
2651 struct queue_limits *dm_get_queue_limits(struct mapped_device *md)
2653 BUG_ON(!atomic_read(&md->holders));
2654 return &md->queue->limits;
2656 EXPORT_SYMBOL_GPL(dm_get_queue_limits);
2658 static void init_rq_based_worker_thread(struct mapped_device *md)
2660 /* Initialize the request-based DM worker thread */
2661 init_kthread_worker(&md->kworker);
2662 md->kworker_task = kthread_run(kthread_worker_fn, &md->kworker,
2663 "kdmwork-%s", dm_device_name(md));
2667 * Fully initialize a request-based queue (->elevator, ->request_fn, etc).
2669 static int dm_init_request_based_queue(struct mapped_device *md)
2671 struct request_queue *q = NULL;
2673 /* Fully initialize the queue */
2674 q = blk_init_allocated_queue(md->queue, dm_request_fn, NULL);
2678 /* disable dm_request_fn's merge heuristic by default */
2679 md->seq_rq_merge_deadline_usecs = 0;
2682 dm_init_old_md_queue(md);
2683 blk_queue_softirq_done(md->queue, dm_softirq_done);
2684 blk_queue_prep_rq(md->queue, dm_prep_fn);
2686 init_rq_based_worker_thread(md);
2688 elv_register_queue(md->queue);
2693 static int dm_mq_init_request(void *data, struct request *rq,
2694 unsigned int hctx_idx, unsigned int request_idx,
2695 unsigned int numa_node)
2697 struct mapped_device *md = data;
2698 struct dm_rq_target_io *tio = blk_mq_rq_to_pdu(rq);
2701 * Must initialize md member of tio, otherwise it won't
2702 * be available in dm_mq_queue_rq.
2709 static int dm_mq_queue_rq(struct blk_mq_hw_ctx *hctx,
2710 const struct blk_mq_queue_data *bd)
2712 struct request *rq = bd->rq;
2713 struct dm_rq_target_io *tio = blk_mq_rq_to_pdu(rq);
2714 struct mapped_device *md = tio->md;
2716 struct dm_table *map = dm_get_live_table(md, &srcu_idx);
2717 struct dm_target *ti;
2720 /* always use block 0 to find the target for flushes for now */
2722 if (!(rq->cmd_flags & REQ_FLUSH))
2723 pos = blk_rq_pos(rq);
2725 ti = dm_table_find_target(map, pos);
2726 if (!dm_target_is_valid(ti)) {
2727 dm_put_live_table(md, srcu_idx);
2728 DMERR_LIMIT("request attempted access beyond the end of device");
2730 * Must perform setup, that rq_completed() requires,
2731 * before returning BLK_MQ_RQ_QUEUE_ERROR
2733 dm_start_request(md, rq);
2734 return BLK_MQ_RQ_QUEUE_ERROR;
2736 dm_put_live_table(md, srcu_idx);
2738 if (ti->type->busy && ti->type->busy(ti))
2739 return BLK_MQ_RQ_QUEUE_BUSY;
2741 dm_start_request(md, rq);
2743 /* Init tio using md established in .init_request */
2744 init_tio(tio, rq, md);
2747 * Establish tio->ti before queuing work (map_tio_request)
2748 * or making direct call to map_request().
2752 /* Clone the request if underlying devices aren't blk-mq */
2753 if (dm_table_get_type(map) == DM_TYPE_REQUEST_BASED) {
2754 /* clone request is allocated at the end of the pdu */
2755 tio->clone = (void *)blk_mq_rq_to_pdu(rq) + sizeof(struct dm_rq_target_io);
2756 if (!clone_rq(rq, md, tio, GFP_ATOMIC))
2757 return BLK_MQ_RQ_QUEUE_BUSY;
2758 queue_kthread_work(&md->kworker, &tio->work);
2760 /* Direct call is fine since .queue_rq allows allocations */
2761 if (map_request(tio, rq, md) == DM_MAPIO_REQUEUE)
2762 dm_requeue_unmapped_original_request(md, rq);
2765 return BLK_MQ_RQ_QUEUE_OK;
2768 static struct blk_mq_ops dm_mq_ops = {
2769 .queue_rq = dm_mq_queue_rq,
2770 .map_queue = blk_mq_map_queue,
2771 .complete = dm_softirq_done,
2772 .init_request = dm_mq_init_request,
2775 static int dm_init_request_based_blk_mq_queue(struct mapped_device *md)
2777 unsigned md_type = dm_get_md_type(md);
2778 struct request_queue *q;
2781 memset(&md->tag_set, 0, sizeof(md->tag_set));
2782 md->tag_set.ops = &dm_mq_ops;
2783 md->tag_set.queue_depth = BLKDEV_MAX_RQ;
2784 md->tag_set.numa_node = NUMA_NO_NODE;
2785 md->tag_set.flags = BLK_MQ_F_SHOULD_MERGE | BLK_MQ_F_SG_MERGE;
2786 md->tag_set.nr_hw_queues = 1;
2787 if (md_type == DM_TYPE_REQUEST_BASED) {
2788 /* make the memory for non-blk-mq clone part of the pdu */
2789 md->tag_set.cmd_size = sizeof(struct dm_rq_target_io) + sizeof(struct request);
2791 md->tag_set.cmd_size = sizeof(struct dm_rq_target_io);
2792 md->tag_set.driver_data = md;
2794 err = blk_mq_alloc_tag_set(&md->tag_set);
2798 q = blk_mq_init_allocated_queue(&md->tag_set, md->queue);
2804 dm_init_md_queue(md);
2806 /* backfill 'mq' sysfs registration normally done in blk_register_queue */
2807 blk_mq_register_disk(md->disk);
2809 if (md_type == DM_TYPE_REQUEST_BASED)
2810 init_rq_based_worker_thread(md);
2815 blk_mq_free_tag_set(&md->tag_set);
2819 static unsigned filter_md_type(unsigned type, struct mapped_device *md)
2821 if (type == DM_TYPE_BIO_BASED)
2824 return !md->use_blk_mq ? DM_TYPE_REQUEST_BASED : DM_TYPE_MQ_REQUEST_BASED;
2828 * Setup the DM device's queue based on md's type
2830 int dm_setup_md_queue(struct mapped_device *md)
2833 unsigned md_type = filter_md_type(dm_get_md_type(md), md);
2836 case DM_TYPE_REQUEST_BASED:
2837 r = dm_init_request_based_queue(md);
2839 DMWARN("Cannot initialize queue for request-based mapped device");
2843 case DM_TYPE_MQ_REQUEST_BASED:
2844 r = dm_init_request_based_blk_mq_queue(md);
2846 DMWARN("Cannot initialize queue for request-based blk-mq mapped device");
2850 case DM_TYPE_BIO_BASED:
2851 dm_init_old_md_queue(md);
2852 blk_queue_make_request(md->queue, dm_make_request);
2853 blk_queue_merge_bvec(md->queue, dm_merge_bvec);
2860 struct mapped_device *dm_get_md(dev_t dev)
2862 struct mapped_device *md;
2863 unsigned minor = MINOR(dev);
2865 if (MAJOR(dev) != _major || minor >= (1 << MINORBITS))
2868 spin_lock(&_minor_lock);
2870 md = idr_find(&_minor_idr, minor);
2872 if ((md == MINOR_ALLOCED ||
2873 (MINOR(disk_devt(dm_disk(md))) != minor) ||
2874 dm_deleting_md(md) ||
2875 test_bit(DMF_FREEING, &md->flags))) {
2883 spin_unlock(&_minor_lock);
2887 EXPORT_SYMBOL_GPL(dm_get_md);
2889 void *dm_get_mdptr(struct mapped_device *md)
2891 return md->interface_ptr;
2894 void dm_set_mdptr(struct mapped_device *md, void *ptr)
2896 md->interface_ptr = ptr;
2899 void dm_get(struct mapped_device *md)
2901 atomic_inc(&md->holders);
2902 BUG_ON(test_bit(DMF_FREEING, &md->flags));
2905 int dm_hold(struct mapped_device *md)
2907 spin_lock(&_minor_lock);
2908 if (test_bit(DMF_FREEING, &md->flags)) {
2909 spin_unlock(&_minor_lock);
2913 spin_unlock(&_minor_lock);
2916 EXPORT_SYMBOL_GPL(dm_hold);
2918 const char *dm_device_name(struct mapped_device *md)
2922 EXPORT_SYMBOL_GPL(dm_device_name);
2924 static void __dm_destroy(struct mapped_device *md, bool wait)
2926 struct dm_table *map;
2931 map = dm_get_live_table(md, &srcu_idx);
2933 spin_lock(&_minor_lock);
2934 idr_replace(&_minor_idr, MINOR_ALLOCED, MINOR(disk_devt(dm_disk(md))));
2935 set_bit(DMF_FREEING, &md->flags);
2936 spin_unlock(&_minor_lock);
2938 if (dm_request_based(md) && md->kworker_task)
2939 flush_kthread_worker(&md->kworker);
2942 * Take suspend_lock so that presuspend and postsuspend methods
2943 * do not race with internal suspend.
2945 mutex_lock(&md->suspend_lock);
2946 if (!dm_suspended_md(md)) {
2947 dm_table_presuspend_targets(map);
2948 dm_table_postsuspend_targets(map);
2950 mutex_unlock(&md->suspend_lock);
2952 /* dm_put_live_table must be before msleep, otherwise deadlock is possible */
2953 dm_put_live_table(md, srcu_idx);
2956 * Rare, but there may be I/O requests still going to complete,
2957 * for example. Wait for all references to disappear.
2958 * No one should increment the reference count of the mapped_device,
2959 * after the mapped_device state becomes DMF_FREEING.
2962 while (atomic_read(&md->holders))
2964 else if (atomic_read(&md->holders))
2965 DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)",
2966 dm_device_name(md), atomic_read(&md->holders));
2969 dm_table_destroy(__unbind(md));
2973 void dm_destroy(struct mapped_device *md)
2975 __dm_destroy(md, true);
2978 void dm_destroy_immediate(struct mapped_device *md)
2980 __dm_destroy(md, false);
2983 void dm_put(struct mapped_device *md)
2985 atomic_dec(&md->holders);
2987 EXPORT_SYMBOL_GPL(dm_put);
2989 static int dm_wait_for_completion(struct mapped_device *md, int interruptible)
2992 DECLARE_WAITQUEUE(wait, current);
2994 add_wait_queue(&md->wait, &wait);
2997 set_current_state(interruptible);
2999 if (!md_in_flight(md))
3002 if (interruptible == TASK_INTERRUPTIBLE &&
3003 signal_pending(current)) {
3010 set_current_state(TASK_RUNNING);
3012 remove_wait_queue(&md->wait, &wait);
3018 * Process the deferred bios
3020 static void dm_wq_work(struct work_struct *work)
3022 struct mapped_device *md = container_of(work, struct mapped_device,
3026 struct dm_table *map;
3028 map = dm_get_live_table(md, &srcu_idx);
3030 while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
3031 spin_lock_irq(&md->deferred_lock);
3032 c = bio_list_pop(&md->deferred);
3033 spin_unlock_irq(&md->deferred_lock);
3038 if (dm_request_based(md))
3039 generic_make_request(c);
3041 __split_and_process_bio(md, map, c);
3044 dm_put_live_table(md, srcu_idx);
3047 static void dm_queue_flush(struct mapped_device *md)
3049 clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
3050 smp_mb__after_atomic();
3051 queue_work(md->wq, &md->work);
3055 * Swap in a new table, returning the old one for the caller to destroy.
3057 struct dm_table *dm_swap_table(struct mapped_device *md, struct dm_table *table)
3059 struct dm_table *live_map = NULL, *map = ERR_PTR(-EINVAL);
3060 struct queue_limits limits;
3063 mutex_lock(&md->suspend_lock);
3065 /* device must be suspended */
3066 if (!dm_suspended_md(md))
3070 * If the new table has no data devices, retain the existing limits.
3071 * This helps multipath with queue_if_no_path if all paths disappear,
3072 * then new I/O is queued based on these limits, and then some paths
3075 if (dm_table_has_no_data_devices(table)) {
3076 live_map = dm_get_live_table_fast(md);
3078 limits = md->queue->limits;
3079 dm_put_live_table_fast(md);
3083 r = dm_calculate_queue_limits(table, &limits);
3090 map = __bind(md, table, &limits);
3093 mutex_unlock(&md->suspend_lock);
3098 * Functions to lock and unlock any filesystem running on the
3101 static int lock_fs(struct mapped_device *md)
3105 WARN_ON(md->frozen_sb);
3107 md->frozen_sb = freeze_bdev(md->bdev);
3108 if (IS_ERR(md->frozen_sb)) {
3109 r = PTR_ERR(md->frozen_sb);
3110 md->frozen_sb = NULL;
3114 set_bit(DMF_FROZEN, &md->flags);
3119 static void unlock_fs(struct mapped_device *md)
3121 if (!test_bit(DMF_FROZEN, &md->flags))
3124 thaw_bdev(md->bdev, md->frozen_sb);
3125 md->frozen_sb = NULL;
3126 clear_bit(DMF_FROZEN, &md->flags);
3130 * If __dm_suspend returns 0, the device is completely quiescent
3131 * now. There is no request-processing activity. All new requests
3132 * are being added to md->deferred list.
3134 * Caller must hold md->suspend_lock
3136 static int __dm_suspend(struct mapped_device *md, struct dm_table *map,
3137 unsigned suspend_flags, int interruptible)
3139 bool do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG;
3140 bool noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG;
3144 * DMF_NOFLUSH_SUSPENDING must be set before presuspend.
3145 * This flag is cleared before dm_suspend returns.
3148 set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
3151 * This gets reverted if there's an error later and the targets
3152 * provide the .presuspend_undo hook.
3154 dm_table_presuspend_targets(map);
3157 * Flush I/O to the device.
3158 * Any I/O submitted after lock_fs() may not be flushed.
3159 * noflush takes precedence over do_lockfs.
3160 * (lock_fs() flushes I/Os and waits for them to complete.)
3162 if (!noflush && do_lockfs) {
3165 dm_table_presuspend_undo_targets(map);
3171 * Here we must make sure that no processes are submitting requests
3172 * to target drivers i.e. no one may be executing
3173 * __split_and_process_bio. This is called from dm_request and
3176 * To get all processes out of __split_and_process_bio in dm_request,
3177 * we take the write lock. To prevent any process from reentering
3178 * __split_and_process_bio from dm_request and quiesce the thread
3179 * (dm_wq_work), we set BMF_BLOCK_IO_FOR_SUSPEND and call
3180 * flush_workqueue(md->wq).
3182 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
3184 synchronize_srcu(&md->io_barrier);
3187 * Stop md->queue before flushing md->wq in case request-based
3188 * dm defers requests to md->wq from md->queue.
3190 if (dm_request_based(md)) {
3191 stop_queue(md->queue);
3192 if (md->kworker_task)
3193 flush_kthread_worker(&md->kworker);
3196 flush_workqueue(md->wq);
3199 * At this point no more requests are entering target request routines.
3200 * We call dm_wait_for_completion to wait for all existing requests
3203 r = dm_wait_for_completion(md, interruptible);
3206 clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
3208 synchronize_srcu(&md->io_barrier);
3210 /* were we interrupted ? */
3214 if (dm_request_based(md))
3215 start_queue(md->queue);
3218 dm_table_presuspend_undo_targets(map);
3219 /* pushback list is already flushed, so skip flush */
3226 * We need to be able to change a mapping table under a mounted
3227 * filesystem. For example we might want to move some data in
3228 * the background. Before the table can be swapped with
3229 * dm_bind_table, dm_suspend must be called to flush any in
3230 * flight bios and ensure that any further io gets deferred.
3233 * Suspend mechanism in request-based dm.
3235 * 1. Flush all I/Os by lock_fs() if needed.
3236 * 2. Stop dispatching any I/O by stopping the request_queue.
3237 * 3. Wait for all in-flight I/Os to be completed or requeued.
3239 * To abort suspend, start the request_queue.
3241 int dm_suspend(struct mapped_device *md, unsigned suspend_flags)
3243 struct dm_table *map = NULL;
3247 mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING);
3249 if (dm_suspended_md(md)) {
3254 if (dm_suspended_internally_md(md)) {
3255 /* already internally suspended, wait for internal resume */
3256 mutex_unlock(&md->suspend_lock);
3257 r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE);
3263 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
3265 r = __dm_suspend(md, map, suspend_flags, TASK_INTERRUPTIBLE);
3269 set_bit(DMF_SUSPENDED, &md->flags);
3271 dm_table_postsuspend_targets(map);
3274 mutex_unlock(&md->suspend_lock);
3278 static int __dm_resume(struct mapped_device *md, struct dm_table *map)
3281 int r = dm_table_resume_targets(map);
3289 * Flushing deferred I/Os must be done after targets are resumed
3290 * so that mapping of targets can work correctly.
3291 * Request-based dm is queueing the deferred I/Os in its request_queue.
3293 if (dm_request_based(md))
3294 start_queue(md->queue);
3301 int dm_resume(struct mapped_device *md)
3304 struct dm_table *map = NULL;
3307 mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING);
3309 if (!dm_suspended_md(md))
3312 if (dm_suspended_internally_md(md)) {
3313 /* already internally suspended, wait for internal resume */
3314 mutex_unlock(&md->suspend_lock);
3315 r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE);
3321 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
3322 if (!map || !dm_table_get_size(map))
3325 r = __dm_resume(md, map);
3329 clear_bit(DMF_SUSPENDED, &md->flags);
3333 mutex_unlock(&md->suspend_lock);
3339 * Internal suspend/resume works like userspace-driven suspend. It waits
3340 * until all bios finish and prevents issuing new bios to the target drivers.
3341 * It may be used only from the kernel.
3344 static void __dm_internal_suspend(struct mapped_device *md, unsigned suspend_flags)
3346 struct dm_table *map = NULL;
3348 if (md->internal_suspend_count++)
3349 return; /* nested internal suspend */
3351 if (dm_suspended_md(md)) {
3352 set_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
3353 return; /* nest suspend */
3356 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
3359 * Using TASK_UNINTERRUPTIBLE because only NOFLUSH internal suspend is
3360 * supported. Properly supporting a TASK_INTERRUPTIBLE internal suspend
3361 * would require changing .presuspend to return an error -- avoid this
3362 * until there is a need for more elaborate variants of internal suspend.
3364 (void) __dm_suspend(md, map, suspend_flags, TASK_UNINTERRUPTIBLE);
3366 set_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
3368 dm_table_postsuspend_targets(map);
3371 static void __dm_internal_resume(struct mapped_device *md)
3373 BUG_ON(!md->internal_suspend_count);
3375 if (--md->internal_suspend_count)
3376 return; /* resume from nested internal suspend */
3378 if (dm_suspended_md(md))
3379 goto done; /* resume from nested suspend */
3382 * NOTE: existing callers don't need to call dm_table_resume_targets
3383 * (which may fail -- so best to avoid it for now by passing NULL map)
3385 (void) __dm_resume(md, NULL);
3388 clear_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
3389 smp_mb__after_atomic();
3390 wake_up_bit(&md->flags, DMF_SUSPENDED_INTERNALLY);
3393 void dm_internal_suspend_noflush(struct mapped_device *md)
3395 mutex_lock(&md->suspend_lock);
3396 __dm_internal_suspend(md, DM_SUSPEND_NOFLUSH_FLAG);
3397 mutex_unlock(&md->suspend_lock);
3399 EXPORT_SYMBOL_GPL(dm_internal_suspend_noflush);
3401 void dm_internal_resume(struct mapped_device *md)
3403 mutex_lock(&md->suspend_lock);
3404 __dm_internal_resume(md);
3405 mutex_unlock(&md->suspend_lock);
3407 EXPORT_SYMBOL_GPL(dm_internal_resume);
3410 * Fast variants of internal suspend/resume hold md->suspend_lock,
3411 * which prevents interaction with userspace-driven suspend.
3414 void dm_internal_suspend_fast(struct mapped_device *md)
3416 mutex_lock(&md->suspend_lock);
3417 if (dm_suspended_md(md) || dm_suspended_internally_md(md))
3420 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
3421 synchronize_srcu(&md->io_barrier);
3422 flush_workqueue(md->wq);
3423 dm_wait_for_completion(md, TASK_UNINTERRUPTIBLE);
3425 EXPORT_SYMBOL_GPL(dm_internal_suspend_fast);
3427 void dm_internal_resume_fast(struct mapped_device *md)
3429 if (dm_suspended_md(md) || dm_suspended_internally_md(md))
3435 mutex_unlock(&md->suspend_lock);
3437 EXPORT_SYMBOL_GPL(dm_internal_resume_fast);
3439 /*-----------------------------------------------------------------
3440 * Event notification.
3441 *---------------------------------------------------------------*/
3442 int dm_kobject_uevent(struct mapped_device *md, enum kobject_action action,
3445 char udev_cookie[DM_COOKIE_LENGTH];
3446 char *envp[] = { udev_cookie, NULL };
3449 return kobject_uevent(&disk_to_dev(md->disk)->kobj, action);
3451 snprintf(udev_cookie, DM_COOKIE_LENGTH, "%s=%u",
3452 DM_COOKIE_ENV_VAR_NAME, cookie);
3453 return kobject_uevent_env(&disk_to_dev(md->disk)->kobj,
3458 uint32_t dm_next_uevent_seq(struct mapped_device *md)
3460 return atomic_add_return(1, &md->uevent_seq);
3463 uint32_t dm_get_event_nr(struct mapped_device *md)
3465 return atomic_read(&md->event_nr);
3468 int dm_wait_event(struct mapped_device *md, int event_nr)
3470 return wait_event_interruptible(md->eventq,
3471 (event_nr != atomic_read(&md->event_nr)));
3474 void dm_uevent_add(struct mapped_device *md, struct list_head *elist)
3476 unsigned long flags;
3478 spin_lock_irqsave(&md->uevent_lock, flags);
3479 list_add(elist, &md->uevent_list);
3480 spin_unlock_irqrestore(&md->uevent_lock, flags);
3484 * The gendisk is only valid as long as you have a reference
3487 struct gendisk *dm_disk(struct mapped_device *md)
3491 EXPORT_SYMBOL_GPL(dm_disk);
3493 struct kobject *dm_kobject(struct mapped_device *md)
3495 return &md->kobj_holder.kobj;
3498 struct mapped_device *dm_get_from_kobject(struct kobject *kobj)
3500 struct mapped_device *md;
3502 md = container_of(kobj, struct mapped_device, kobj_holder.kobj);
3504 if (test_bit(DMF_FREEING, &md->flags) ||
3512 int dm_suspended_md(struct mapped_device *md)
3514 return test_bit(DMF_SUSPENDED, &md->flags);
3517 int dm_suspended_internally_md(struct mapped_device *md)
3519 return test_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
3522 int dm_test_deferred_remove_flag(struct mapped_device *md)
3524 return test_bit(DMF_DEFERRED_REMOVE, &md->flags);
3527 int dm_suspended(struct dm_target *ti)
3529 return dm_suspended_md(dm_table_get_md(ti->table));
3531 EXPORT_SYMBOL_GPL(dm_suspended);
3533 int dm_noflush_suspending(struct dm_target *ti)
3535 return __noflush_suspending(dm_table_get_md(ti->table));
3537 EXPORT_SYMBOL_GPL(dm_noflush_suspending);
3539 struct dm_md_mempools *dm_alloc_md_mempools(struct mapped_device *md, unsigned type,
3540 unsigned integrity, unsigned per_bio_data_size)
3542 struct dm_md_mempools *pools = kzalloc(sizeof(*pools), GFP_KERNEL);
3543 struct kmem_cache *cachep = NULL;
3544 unsigned int pool_size = 0;
3545 unsigned int front_pad;
3550 type = filter_md_type(type, md);
3553 case DM_TYPE_BIO_BASED:
3555 pool_size = dm_get_reserved_bio_based_ios();
3556 front_pad = roundup(per_bio_data_size, __alignof__(struct dm_target_io)) + offsetof(struct dm_target_io, clone);
3558 case DM_TYPE_REQUEST_BASED:
3559 cachep = _rq_tio_cache;
3560 pool_size = dm_get_reserved_rq_based_ios();
3561 pools->rq_pool = mempool_create_slab_pool(pool_size, _rq_cache);
3562 if (!pools->rq_pool)
3564 /* fall through to setup remaining rq-based pools */
3565 case DM_TYPE_MQ_REQUEST_BASED:
3567 pool_size = dm_get_reserved_rq_based_ios();
3568 front_pad = offsetof(struct dm_rq_clone_bio_info, clone);
3569 /* per_bio_data_size is not used. See __bind_mempools(). */
3570 WARN_ON(per_bio_data_size != 0);
3577 pools->io_pool = mempool_create_slab_pool(pool_size, cachep);
3578 if (!pools->io_pool)
3582 pools->bs = bioset_create_nobvec(pool_size, front_pad);
3586 if (integrity && bioset_integrity_create(pools->bs, pool_size))
3592 dm_free_md_mempools(pools);
3597 void dm_free_md_mempools(struct dm_md_mempools *pools)
3603 mempool_destroy(pools->io_pool);
3606 mempool_destroy(pools->rq_pool);
3609 bioset_free(pools->bs);
3614 static const struct block_device_operations dm_blk_dops = {
3615 .open = dm_blk_open,
3616 .release = dm_blk_close,
3617 .ioctl = dm_blk_ioctl,
3618 .getgeo = dm_blk_getgeo,
3619 .owner = THIS_MODULE
3625 module_init(dm_init);
3626 module_exit(dm_exit);
3628 module_param(major, uint, 0);
3629 MODULE_PARM_DESC(major, "The major number of the device mapper");
3631 module_param(reserved_bio_based_ios, uint, S_IRUGO | S_IWUSR);
3632 MODULE_PARM_DESC(reserved_bio_based_ios, "Reserved IOs in bio-based mempools");
3634 module_param(reserved_rq_based_ios, uint, S_IRUGO | S_IWUSR);
3635 MODULE_PARM_DESC(reserved_rq_based_ios, "Reserved IOs in request-based mempools");
3637 module_param(use_blk_mq, bool, S_IRUGO | S_IWUSR);
3638 MODULE_PARM_DESC(use_blk_mq, "Use block multiqueue for request-based DM devices");
3640 MODULE_DESCRIPTION(DM_NAME " driver");
3641 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
3642 MODULE_LICENSE("GPL");