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 atomic_dec(&md->pending[rw]);
1058 /* nudge anyone waiting on suspend queue */
1059 if (!md_in_flight(md))
1063 * Run this off this callpath, as drivers could invoke end_io while
1064 * inside their request_fn (and holding the queue lock). Calling
1065 * back into ->request_fn() could deadlock attempting to grab the
1069 if (md->queue->mq_ops)
1070 blk_mq_run_hw_queues(md->queue, true);
1072 blk_run_queue_async(md->queue);
1076 * dm_put() must be at the end of this function. See the comment above
1081 static void free_rq_clone(struct request *clone)
1083 struct dm_rq_target_io *tio = clone->end_io_data;
1084 struct mapped_device *md = tio->md;
1086 blk_rq_unprep_clone(clone);
1088 if (md->type == DM_TYPE_MQ_REQUEST_BASED)
1089 /* stacked on blk-mq queue(s) */
1090 tio->ti->type->release_clone_rq(clone);
1091 else if (!md->queue->mq_ops)
1092 /* request_fn queue stacked on request_fn queue(s) */
1093 free_clone_request(md, clone);
1095 * NOTE: for the blk-mq queue stacked on request_fn queue(s) case:
1096 * no need to call free_clone_request() because we leverage blk-mq by
1097 * allocating the clone at the end of the blk-mq pdu (see: clone_rq)
1100 if (!md->queue->mq_ops)
1105 * Complete the clone and the original request.
1106 * Must be called without clone's queue lock held,
1107 * see end_clone_request() for more details.
1109 static void dm_end_request(struct request *clone, int error)
1111 int rw = rq_data_dir(clone);
1112 struct dm_rq_target_io *tio = clone->end_io_data;
1113 struct mapped_device *md = tio->md;
1114 struct request *rq = tio->orig;
1116 if (rq->cmd_type == REQ_TYPE_BLOCK_PC) {
1117 rq->errors = clone->errors;
1118 rq->resid_len = clone->resid_len;
1122 * We are using the sense buffer of the original
1124 * So setting the length of the sense data is enough.
1126 rq->sense_len = clone->sense_len;
1129 free_rq_clone(clone);
1131 blk_end_request_all(rq, error);
1133 blk_mq_end_request(rq, error);
1134 rq_completed(md, rw, true);
1137 static void dm_unprep_request(struct request *rq)
1139 struct dm_rq_target_io *tio = tio_from_request(rq);
1140 struct request *clone = tio->clone;
1142 if (!rq->q->mq_ops) {
1144 rq->cmd_flags &= ~REQ_DONTPREP;
1148 free_rq_clone(clone);
1152 * Requeue the original request of a clone.
1154 static void old_requeue_request(struct request *rq)
1156 struct request_queue *q = rq->q;
1157 unsigned long flags;
1159 spin_lock_irqsave(q->queue_lock, flags);
1160 blk_requeue_request(q, rq);
1161 blk_run_queue_async(q);
1162 spin_unlock_irqrestore(q->queue_lock, flags);
1165 static void dm_requeue_unmapped_original_request(struct mapped_device *md,
1168 int rw = rq_data_dir(rq);
1170 dm_unprep_request(rq);
1173 old_requeue_request(rq);
1175 blk_mq_requeue_request(rq);
1176 blk_mq_kick_requeue_list(rq->q);
1179 rq_completed(md, rw, false);
1182 static void dm_requeue_unmapped_request(struct request *clone)
1184 struct dm_rq_target_io *tio = clone->end_io_data;
1186 dm_requeue_unmapped_original_request(tio->md, tio->orig);
1189 static void old_stop_queue(struct request_queue *q)
1191 unsigned long flags;
1193 if (blk_queue_stopped(q))
1196 spin_lock_irqsave(q->queue_lock, flags);
1198 spin_unlock_irqrestore(q->queue_lock, flags);
1201 static void stop_queue(struct request_queue *q)
1206 blk_mq_stop_hw_queues(q);
1209 static void old_start_queue(struct request_queue *q)
1211 unsigned long flags;
1213 spin_lock_irqsave(q->queue_lock, flags);
1214 if (blk_queue_stopped(q))
1216 spin_unlock_irqrestore(q->queue_lock, flags);
1219 static void start_queue(struct request_queue *q)
1224 blk_mq_start_stopped_hw_queues(q, true);
1227 static void dm_done(struct request *clone, int error, bool mapped)
1230 struct dm_rq_target_io *tio = clone->end_io_data;
1231 dm_request_endio_fn rq_end_io = NULL;
1234 rq_end_io = tio->ti->type->rq_end_io;
1236 if (mapped && rq_end_io)
1237 r = rq_end_io(tio->ti, clone, error, &tio->info);
1240 if (unlikely(r == -EREMOTEIO && (clone->cmd_flags & REQ_WRITE_SAME) &&
1241 !clone->q->limits.max_write_same_sectors))
1242 disable_write_same(tio->md);
1245 /* The target wants to complete the I/O */
1246 dm_end_request(clone, r);
1247 else if (r == DM_ENDIO_INCOMPLETE)
1248 /* The target will handle the I/O */
1250 else if (r == DM_ENDIO_REQUEUE)
1251 /* The target wants to requeue the I/O */
1252 dm_requeue_unmapped_request(clone);
1254 DMWARN("unimplemented target endio return value: %d", r);
1260 * Request completion handler for request-based dm
1262 static void dm_softirq_done(struct request *rq)
1265 struct dm_rq_target_io *tio = tio_from_request(rq);
1266 struct request *clone = tio->clone;
1270 rw = rq_data_dir(rq);
1271 if (!rq->q->mq_ops) {
1272 blk_end_request_all(rq, tio->error);
1273 rq_completed(tio->md, rw, false);
1276 blk_mq_end_request(rq, tio->error);
1277 rq_completed(tio->md, rw, false);
1282 if (rq->cmd_flags & REQ_FAILED)
1285 dm_done(clone, tio->error, mapped);
1289 * Complete the clone and the original request with the error status
1290 * through softirq context.
1292 static void dm_complete_request(struct request *rq, int error)
1294 struct dm_rq_target_io *tio = tio_from_request(rq);
1297 blk_complete_request(rq);
1301 * Complete the not-mapped clone and the original request with the error status
1302 * through softirq context.
1303 * Target's rq_end_io() function isn't called.
1304 * This may be used when the target's map_rq() or clone_and_map_rq() functions fail.
1306 static void dm_kill_unmapped_request(struct request *rq, int error)
1308 rq->cmd_flags |= REQ_FAILED;
1309 dm_complete_request(rq, error);
1313 * Called with the clone's queue lock held (for non-blk-mq)
1315 static void end_clone_request(struct request *clone, int error)
1317 struct dm_rq_target_io *tio = clone->end_io_data;
1319 if (!clone->q->mq_ops) {
1321 * For just cleaning up the information of the queue in which
1322 * the clone was dispatched.
1323 * The clone is *NOT* freed actually here because it is alloced
1324 * from dm own mempool (REQ_ALLOCED isn't set).
1326 __blk_put_request(clone->q, clone);
1330 * Actual request completion is done in a softirq context which doesn't
1331 * hold the clone's queue lock. Otherwise, deadlock could occur because:
1332 * - another request may be submitted by the upper level driver
1333 * of the stacking during the completion
1334 * - the submission which requires queue lock may be done
1335 * against this clone's queue
1337 dm_complete_request(tio->orig, error);
1341 * Return maximum size of I/O possible at the supplied sector up to the current
1344 static sector_t max_io_len_target_boundary(sector_t sector, struct dm_target *ti)
1346 sector_t target_offset = dm_target_offset(ti, sector);
1348 return ti->len - target_offset;
1351 static sector_t max_io_len(sector_t sector, struct dm_target *ti)
1353 sector_t len = max_io_len_target_boundary(sector, ti);
1354 sector_t offset, max_len;
1357 * Does the target need to split even further?
1359 if (ti->max_io_len) {
1360 offset = dm_target_offset(ti, sector);
1361 if (unlikely(ti->max_io_len & (ti->max_io_len - 1)))
1362 max_len = sector_div(offset, ti->max_io_len);
1364 max_len = offset & (ti->max_io_len - 1);
1365 max_len = ti->max_io_len - max_len;
1374 int dm_set_target_max_io_len(struct dm_target *ti, sector_t len)
1376 if (len > UINT_MAX) {
1377 DMERR("Specified maximum size of target IO (%llu) exceeds limit (%u)",
1378 (unsigned long long)len, UINT_MAX);
1379 ti->error = "Maximum size of target IO is too large";
1383 ti->max_io_len = (uint32_t) len;
1387 EXPORT_SYMBOL_GPL(dm_set_target_max_io_len);
1390 * A target may call dm_accept_partial_bio only from the map routine. It is
1391 * allowed for all bio types except REQ_FLUSH.
1393 * dm_accept_partial_bio informs the dm that the target only wants to process
1394 * additional n_sectors sectors of the bio and the rest of the data should be
1395 * sent in a next bio.
1397 * A diagram that explains the arithmetics:
1398 * +--------------------+---------------+-------+
1400 * +--------------------+---------------+-------+
1402 * <-------------- *tio->len_ptr --------------->
1403 * <------- bi_size ------->
1406 * Region 1 was already iterated over with bio_advance or similar function.
1407 * (it may be empty if the target doesn't use bio_advance)
1408 * Region 2 is the remaining bio size that the target wants to process.
1409 * (it may be empty if region 1 is non-empty, although there is no reason
1411 * The target requires that region 3 is to be sent in the next bio.
1413 * If the target wants to receive multiple copies of the bio (via num_*bios, etc),
1414 * the partially processed part (the sum of regions 1+2) must be the same for all
1415 * copies of the bio.
1417 void dm_accept_partial_bio(struct bio *bio, unsigned n_sectors)
1419 struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone);
1420 unsigned bi_size = bio->bi_iter.bi_size >> SECTOR_SHIFT;
1421 BUG_ON(bio->bi_rw & REQ_FLUSH);
1422 BUG_ON(bi_size > *tio->len_ptr);
1423 BUG_ON(n_sectors > bi_size);
1424 *tio->len_ptr -= bi_size - n_sectors;
1425 bio->bi_iter.bi_size = n_sectors << SECTOR_SHIFT;
1427 EXPORT_SYMBOL_GPL(dm_accept_partial_bio);
1429 static void __map_bio(struct dm_target_io *tio)
1433 struct mapped_device *md;
1434 struct bio *clone = &tio->clone;
1435 struct dm_target *ti = tio->ti;
1437 clone->bi_end_io = clone_endio;
1440 * Map the clone. If r == 0 we don't need to do
1441 * anything, the target has assumed ownership of
1444 atomic_inc(&tio->io->io_count);
1445 sector = clone->bi_iter.bi_sector;
1446 r = ti->type->map(ti, clone);
1447 if (r == DM_MAPIO_REMAPPED) {
1448 /* the bio has been remapped so dispatch it */
1450 trace_block_bio_remap(bdev_get_queue(clone->bi_bdev), clone,
1451 tio->io->bio->bi_bdev->bd_dev, sector);
1453 generic_make_request(clone);
1454 } else if (r < 0 || r == DM_MAPIO_REQUEUE) {
1455 /* error the io and bail out, or requeue it if needed */
1457 dec_pending(tio->io, r);
1460 DMWARN("unimplemented target map return value: %d", r);
1466 struct mapped_device *md;
1467 struct dm_table *map;
1471 unsigned sector_count;
1474 static void bio_setup_sector(struct bio *bio, sector_t sector, unsigned len)
1476 bio->bi_iter.bi_sector = sector;
1477 bio->bi_iter.bi_size = to_bytes(len);
1481 * Creates a bio that consists of range of complete bvecs.
1483 static void clone_bio(struct dm_target_io *tio, struct bio *bio,
1484 sector_t sector, unsigned len)
1486 struct bio *clone = &tio->clone;
1488 __bio_clone_fast(clone, bio);
1490 if (bio_integrity(bio))
1491 bio_integrity_clone(clone, bio, GFP_NOIO);
1493 bio_advance(clone, to_bytes(sector - clone->bi_iter.bi_sector));
1494 clone->bi_iter.bi_size = to_bytes(len);
1496 if (bio_integrity(bio))
1497 bio_integrity_trim(clone, 0, len);
1500 static struct dm_target_io *alloc_tio(struct clone_info *ci,
1501 struct dm_target *ti,
1502 unsigned target_bio_nr)
1504 struct dm_target_io *tio;
1507 clone = bio_alloc_bioset(GFP_NOIO, 0, ci->md->bs);
1508 tio = container_of(clone, struct dm_target_io, clone);
1512 tio->target_bio_nr = target_bio_nr;
1517 static void __clone_and_map_simple_bio(struct clone_info *ci,
1518 struct dm_target *ti,
1519 unsigned target_bio_nr, unsigned *len)
1521 struct dm_target_io *tio = alloc_tio(ci, ti, target_bio_nr);
1522 struct bio *clone = &tio->clone;
1526 __bio_clone_fast(clone, ci->bio);
1528 bio_setup_sector(clone, ci->sector, *len);
1533 static void __send_duplicate_bios(struct clone_info *ci, struct dm_target *ti,
1534 unsigned num_bios, unsigned *len)
1536 unsigned target_bio_nr;
1538 for (target_bio_nr = 0; target_bio_nr < num_bios; target_bio_nr++)
1539 __clone_and_map_simple_bio(ci, ti, target_bio_nr, len);
1542 static int __send_empty_flush(struct clone_info *ci)
1544 unsigned target_nr = 0;
1545 struct dm_target *ti;
1547 BUG_ON(bio_has_data(ci->bio));
1548 while ((ti = dm_table_get_target(ci->map, target_nr++)))
1549 __send_duplicate_bios(ci, ti, ti->num_flush_bios, NULL);
1554 static void __clone_and_map_data_bio(struct clone_info *ci, struct dm_target *ti,
1555 sector_t sector, unsigned *len)
1557 struct bio *bio = ci->bio;
1558 struct dm_target_io *tio;
1559 unsigned target_bio_nr;
1560 unsigned num_target_bios = 1;
1563 * Does the target want to receive duplicate copies of the bio?
1565 if (bio_data_dir(bio) == WRITE && ti->num_write_bios)
1566 num_target_bios = ti->num_write_bios(ti, bio);
1568 for (target_bio_nr = 0; target_bio_nr < num_target_bios; target_bio_nr++) {
1569 tio = alloc_tio(ci, ti, target_bio_nr);
1571 clone_bio(tio, bio, sector, *len);
1576 typedef unsigned (*get_num_bios_fn)(struct dm_target *ti);
1578 static unsigned get_num_discard_bios(struct dm_target *ti)
1580 return ti->num_discard_bios;
1583 static unsigned get_num_write_same_bios(struct dm_target *ti)
1585 return ti->num_write_same_bios;
1588 typedef bool (*is_split_required_fn)(struct dm_target *ti);
1590 static bool is_split_required_for_discard(struct dm_target *ti)
1592 return ti->split_discard_bios;
1595 static int __send_changing_extent_only(struct clone_info *ci,
1596 get_num_bios_fn get_num_bios,
1597 is_split_required_fn is_split_required)
1599 struct dm_target *ti;
1604 ti = dm_table_find_target(ci->map, ci->sector);
1605 if (!dm_target_is_valid(ti))
1609 * Even though the device advertised support for this type of
1610 * request, that does not mean every target supports it, and
1611 * reconfiguration might also have changed that since the
1612 * check was performed.
1614 num_bios = get_num_bios ? get_num_bios(ti) : 0;
1618 if (is_split_required && !is_split_required(ti))
1619 len = min((sector_t)ci->sector_count, max_io_len_target_boundary(ci->sector, ti));
1621 len = min((sector_t)ci->sector_count, max_io_len(ci->sector, ti));
1623 __send_duplicate_bios(ci, ti, num_bios, &len);
1626 } while (ci->sector_count -= len);
1631 static int __send_discard(struct clone_info *ci)
1633 return __send_changing_extent_only(ci, get_num_discard_bios,
1634 is_split_required_for_discard);
1637 static int __send_write_same(struct clone_info *ci)
1639 return __send_changing_extent_only(ci, get_num_write_same_bios, NULL);
1643 * Select the correct strategy for processing a non-flush bio.
1645 static int __split_and_process_non_flush(struct clone_info *ci)
1647 struct bio *bio = ci->bio;
1648 struct dm_target *ti;
1651 if (unlikely(bio->bi_rw & REQ_DISCARD))
1652 return __send_discard(ci);
1653 else if (unlikely(bio->bi_rw & REQ_WRITE_SAME))
1654 return __send_write_same(ci);
1656 ti = dm_table_find_target(ci->map, ci->sector);
1657 if (!dm_target_is_valid(ti))
1660 len = min_t(sector_t, max_io_len(ci->sector, ti), ci->sector_count);
1662 __clone_and_map_data_bio(ci, ti, ci->sector, &len);
1665 ci->sector_count -= len;
1671 * Entry point to split a bio into clones and submit them to the targets.
1673 static void __split_and_process_bio(struct mapped_device *md,
1674 struct dm_table *map, struct bio *bio)
1676 struct clone_info ci;
1679 if (unlikely(!map)) {
1686 ci.io = alloc_io(md);
1688 atomic_set(&ci.io->io_count, 1);
1691 spin_lock_init(&ci.io->endio_lock);
1692 ci.sector = bio->bi_iter.bi_sector;
1694 start_io_acct(ci.io);
1696 if (bio->bi_rw & REQ_FLUSH) {
1697 ci.bio = &ci.md->flush_bio;
1698 ci.sector_count = 0;
1699 error = __send_empty_flush(&ci);
1700 /* dec_pending submits any data associated with flush */
1703 ci.sector_count = bio_sectors(bio);
1704 while (ci.sector_count && !error)
1705 error = __split_and_process_non_flush(&ci);
1708 /* drop the extra reference count */
1709 dec_pending(ci.io, error);
1711 /*-----------------------------------------------------------------
1713 *---------------------------------------------------------------*/
1715 static int dm_merge_bvec(struct request_queue *q,
1716 struct bvec_merge_data *bvm,
1717 struct bio_vec *biovec)
1719 struct mapped_device *md = q->queuedata;
1720 struct dm_table *map = dm_get_live_table_fast(md);
1721 struct dm_target *ti;
1722 sector_t max_sectors;
1728 ti = dm_table_find_target(map, bvm->bi_sector);
1729 if (!dm_target_is_valid(ti))
1733 * Find maximum amount of I/O that won't need splitting
1735 max_sectors = min(max_io_len(bvm->bi_sector, ti),
1736 (sector_t) BIO_MAX_SECTORS);
1737 max_size = (max_sectors << SECTOR_SHIFT) - bvm->bi_size;
1742 * merge_bvec_fn() returns number of bytes
1743 * it can accept at this offset
1744 * max is precomputed maximal io size
1746 if (max_size && ti->type->merge)
1747 max_size = ti->type->merge(ti, bvm, biovec, max_size);
1749 * If the target doesn't support merge method and some of the devices
1750 * provided their merge_bvec method (we know this by looking at
1751 * queue_max_hw_sectors), then we can't allow bios with multiple vector
1752 * entries. So always set max_size to 0, and the code below allows
1755 else if (queue_max_hw_sectors(q) <= PAGE_SIZE >> 9)
1759 dm_put_live_table_fast(md);
1761 * Always allow an entire first page
1763 if (max_size <= biovec->bv_len && !(bvm->bi_size >> SECTOR_SHIFT))
1764 max_size = biovec->bv_len;
1770 * The request function that just remaps the bio built up by
1773 static void dm_make_request(struct request_queue *q, struct bio *bio)
1775 int rw = bio_data_dir(bio);
1776 struct mapped_device *md = q->queuedata;
1778 struct dm_table *map;
1780 map = dm_get_live_table(md, &srcu_idx);
1782 generic_start_io_acct(rw, bio_sectors(bio), &dm_disk(md)->part0);
1784 /* if we're suspended, we have to queue this io for later */
1785 if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))) {
1786 dm_put_live_table(md, srcu_idx);
1788 if (bio_rw(bio) != READA)
1795 __split_and_process_bio(md, map, bio);
1796 dm_put_live_table(md, srcu_idx);
1800 int dm_request_based(struct mapped_device *md)
1802 return blk_queue_stackable(md->queue);
1805 static void dm_dispatch_clone_request(struct request *clone, struct request *rq)
1809 if (blk_queue_io_stat(clone->q))
1810 clone->cmd_flags |= REQ_IO_STAT;
1812 clone->start_time = jiffies;
1813 r = blk_insert_cloned_request(clone->q, clone);
1815 /* must complete clone in terms of original request */
1816 dm_complete_request(rq, r);
1819 static int dm_rq_bio_constructor(struct bio *bio, struct bio *bio_orig,
1822 struct dm_rq_target_io *tio = data;
1823 struct dm_rq_clone_bio_info *info =
1824 container_of(bio, struct dm_rq_clone_bio_info, clone);
1826 info->orig = bio_orig;
1828 bio->bi_end_io = end_clone_bio;
1833 static int setup_clone(struct request *clone, struct request *rq,
1834 struct dm_rq_target_io *tio, gfp_t gfp_mask)
1838 r = blk_rq_prep_clone(clone, rq, tio->md->bs, gfp_mask,
1839 dm_rq_bio_constructor, tio);
1843 clone->cmd = rq->cmd;
1844 clone->cmd_len = rq->cmd_len;
1845 clone->sense = rq->sense;
1846 clone->end_io = end_clone_request;
1847 clone->end_io_data = tio;
1854 static struct request *clone_rq(struct request *rq, struct mapped_device *md,
1855 struct dm_rq_target_io *tio, gfp_t gfp_mask)
1858 * Do not allocate a clone if tio->clone was already set
1859 * (see: dm_mq_queue_rq).
1861 bool alloc_clone = !tio->clone;
1862 struct request *clone;
1865 clone = alloc_clone_request(md, gfp_mask);
1871 blk_rq_init(NULL, clone);
1872 if (setup_clone(clone, rq, tio, gfp_mask)) {
1875 free_clone_request(md, clone);
1882 static void map_tio_request(struct kthread_work *work);
1884 static void init_tio(struct dm_rq_target_io *tio, struct request *rq,
1885 struct mapped_device *md)
1892 memset(&tio->info, 0, sizeof(tio->info));
1893 if (md->kworker_task)
1894 init_kthread_work(&tio->work, map_tio_request);
1897 static struct dm_rq_target_io *prep_tio(struct request *rq,
1898 struct mapped_device *md, gfp_t gfp_mask)
1900 struct dm_rq_target_io *tio;
1902 struct dm_table *table;
1904 tio = alloc_rq_tio(md, gfp_mask);
1908 init_tio(tio, rq, md);
1910 table = dm_get_live_table(md, &srcu_idx);
1911 if (!dm_table_mq_request_based(table)) {
1912 if (!clone_rq(rq, md, tio, gfp_mask)) {
1913 dm_put_live_table(md, srcu_idx);
1918 dm_put_live_table(md, srcu_idx);
1924 * Called with the queue lock held.
1926 static int dm_prep_fn(struct request_queue *q, struct request *rq)
1928 struct mapped_device *md = q->queuedata;
1929 struct dm_rq_target_io *tio;
1931 if (unlikely(rq->special)) {
1932 DMWARN("Already has something in rq->special.");
1933 return BLKPREP_KILL;
1936 tio = prep_tio(rq, md, GFP_ATOMIC);
1938 return BLKPREP_DEFER;
1941 rq->cmd_flags |= REQ_DONTPREP;
1948 * 0 : the request has been processed
1949 * DM_MAPIO_REQUEUE : the original request needs to be requeued
1950 * < 0 : the request was completed due to failure
1952 static int map_request(struct dm_rq_target_io *tio, struct request *rq,
1953 struct mapped_device *md)
1956 struct dm_target *ti = tio->ti;
1957 struct request *clone = NULL;
1961 r = ti->type->map_rq(ti, clone, &tio->info);
1963 r = ti->type->clone_and_map_rq(ti, rq, &tio->info, &clone);
1965 /* The target wants to complete the I/O */
1966 dm_kill_unmapped_request(rq, r);
1969 if (r != DM_MAPIO_REMAPPED)
1971 if (setup_clone(clone, rq, tio, GFP_ATOMIC)) {
1973 ti->type->release_clone_rq(clone);
1974 return DM_MAPIO_REQUEUE;
1979 case DM_MAPIO_SUBMITTED:
1980 /* The target has taken the I/O to submit by itself later */
1982 case DM_MAPIO_REMAPPED:
1983 /* The target has remapped the I/O so dispatch it */
1984 trace_block_rq_remap(clone->q, clone, disk_devt(dm_disk(md)),
1986 dm_dispatch_clone_request(clone, rq);
1988 case DM_MAPIO_REQUEUE:
1989 /* The target wants to requeue the I/O */
1990 dm_requeue_unmapped_request(clone);
1994 DMWARN("unimplemented target map return value: %d", r);
1998 /* The target wants to complete the I/O */
1999 dm_kill_unmapped_request(rq, r);
2006 static void map_tio_request(struct kthread_work *work)
2008 struct dm_rq_target_io *tio = container_of(work, struct dm_rq_target_io, work);
2009 struct request *rq = tio->orig;
2010 struct mapped_device *md = tio->md;
2012 if (map_request(tio, rq, md) == DM_MAPIO_REQUEUE)
2013 dm_requeue_unmapped_original_request(md, rq);
2016 static void dm_start_request(struct mapped_device *md, struct request *orig)
2018 if (!orig->q->mq_ops)
2019 blk_start_request(orig);
2021 blk_mq_start_request(orig);
2022 atomic_inc(&md->pending[rq_data_dir(orig)]);
2024 if (md->seq_rq_merge_deadline_usecs) {
2025 md->last_rq_pos = rq_end_sector(orig);
2026 md->last_rq_rw = rq_data_dir(orig);
2027 md->last_rq_start_time = ktime_get();
2031 * Hold the md reference here for the in-flight I/O.
2032 * We can't rely on the reference count by device opener,
2033 * because the device may be closed during the request completion
2034 * when all bios are completed.
2035 * See the comment in rq_completed() too.
2040 #define MAX_SEQ_RQ_MERGE_DEADLINE_USECS 100000
2042 ssize_t dm_attr_rq_based_seq_io_merge_deadline_show(struct mapped_device *md, char *buf)
2044 return sprintf(buf, "%u\n", md->seq_rq_merge_deadline_usecs);
2047 ssize_t dm_attr_rq_based_seq_io_merge_deadline_store(struct mapped_device *md,
2048 const char *buf, size_t count)
2052 if (!dm_request_based(md) || md->use_blk_mq)
2055 if (kstrtouint(buf, 10, &deadline))
2058 if (deadline > MAX_SEQ_RQ_MERGE_DEADLINE_USECS)
2059 deadline = MAX_SEQ_RQ_MERGE_DEADLINE_USECS;
2061 md->seq_rq_merge_deadline_usecs = deadline;
2066 static bool dm_request_peeked_before_merge_deadline(struct mapped_device *md)
2068 ktime_t kt_deadline;
2070 if (!md->seq_rq_merge_deadline_usecs)
2073 kt_deadline = ns_to_ktime((u64)md->seq_rq_merge_deadline_usecs * NSEC_PER_USEC);
2074 kt_deadline = ktime_add_safe(md->last_rq_start_time, kt_deadline);
2076 return !ktime_after(ktime_get(), kt_deadline);
2080 * q->request_fn for request-based dm.
2081 * Called with the queue lock held.
2083 static void dm_request_fn(struct request_queue *q)
2085 struct mapped_device *md = q->queuedata;
2087 struct dm_table *map = dm_get_live_table(md, &srcu_idx);
2088 struct dm_target *ti;
2090 struct dm_rq_target_io *tio;
2094 * For suspend, check blk_queue_stopped() and increment
2095 * ->pending within a single queue_lock not to increment the
2096 * number of in-flight I/Os after the queue is stopped in
2099 while (!blk_queue_stopped(q)) {
2100 rq = blk_peek_request(q);
2104 /* always use block 0 to find the target for flushes for now */
2106 if (!(rq->cmd_flags & REQ_FLUSH))
2107 pos = blk_rq_pos(rq);
2109 ti = dm_table_find_target(map, pos);
2110 if (!dm_target_is_valid(ti)) {
2112 * Must perform setup, that rq_completed() requires,
2113 * before calling dm_kill_unmapped_request
2115 DMERR_LIMIT("request attempted access beyond the end of device");
2116 dm_start_request(md, rq);
2117 dm_kill_unmapped_request(rq, -EIO);
2121 if (dm_request_peeked_before_merge_deadline(md) &&
2122 md_in_flight(md) && rq->bio && rq->bio->bi_vcnt == 1 &&
2123 md->last_rq_pos == pos && md->last_rq_rw == rq_data_dir(rq))
2126 if (ti->type->busy && ti->type->busy(ti))
2129 dm_start_request(md, rq);
2131 tio = tio_from_request(rq);
2132 /* Establish tio->ti before queuing work (map_tio_request) */
2134 queue_kthread_work(&md->kworker, &tio->work);
2135 BUG_ON_NONRT(!irqs_disabled());
2141 blk_delay_queue(q, HZ / 100);
2143 dm_put_live_table(md, srcu_idx);
2146 static int dm_any_congested(void *congested_data, int bdi_bits)
2149 struct mapped_device *md = congested_data;
2150 struct dm_table *map;
2152 if (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
2153 map = dm_get_live_table_fast(md);
2156 * Request-based dm cares about only own queue for
2157 * the query about congestion status of request_queue
2159 if (dm_request_based(md))
2160 r = md->queue->backing_dev_info.state &
2163 r = dm_table_any_congested(map, bdi_bits);
2165 dm_put_live_table_fast(md);
2171 /*-----------------------------------------------------------------
2172 * An IDR is used to keep track of allocated minor numbers.
2173 *---------------------------------------------------------------*/
2174 static void free_minor(int minor)
2176 spin_lock(&_minor_lock);
2177 idr_remove(&_minor_idr, minor);
2178 spin_unlock(&_minor_lock);
2182 * See if the device with a specific minor # is free.
2184 static int specific_minor(int minor)
2188 if (minor >= (1 << MINORBITS))
2191 idr_preload(GFP_KERNEL);
2192 spin_lock(&_minor_lock);
2194 r = idr_alloc(&_minor_idr, MINOR_ALLOCED, minor, minor + 1, GFP_NOWAIT);
2196 spin_unlock(&_minor_lock);
2199 return r == -ENOSPC ? -EBUSY : r;
2203 static int next_free_minor(int *minor)
2207 idr_preload(GFP_KERNEL);
2208 spin_lock(&_minor_lock);
2210 r = idr_alloc(&_minor_idr, MINOR_ALLOCED, 0, 1 << MINORBITS, GFP_NOWAIT);
2212 spin_unlock(&_minor_lock);
2220 static const struct block_device_operations dm_blk_dops;
2222 static void dm_wq_work(struct work_struct *work);
2224 static void dm_init_md_queue(struct mapped_device *md)
2227 * Request-based dm devices cannot be stacked on top of bio-based dm
2228 * devices. The type of this dm device may not have been decided yet.
2229 * The type is decided at the first table loading time.
2230 * To prevent problematic device stacking, clear the queue flag
2231 * for request stacking support until then.
2233 * This queue is new, so no concurrency on the queue_flags.
2235 queue_flag_clear_unlocked(QUEUE_FLAG_STACKABLE, md->queue);
2238 static void dm_init_old_md_queue(struct mapped_device *md)
2240 md->use_blk_mq = false;
2241 dm_init_md_queue(md);
2244 * Initialize aspects of queue that aren't relevant for blk-mq
2246 md->queue->queuedata = md;
2247 md->queue->backing_dev_info.congested_fn = dm_any_congested;
2248 md->queue->backing_dev_info.congested_data = md;
2250 blk_queue_bounce_limit(md->queue, BLK_BOUNCE_ANY);
2254 * Allocate and initialise a blank device with a given minor.
2256 static struct mapped_device *alloc_dev(int minor)
2259 struct mapped_device *md = kzalloc(sizeof(*md), GFP_KERNEL);
2263 DMWARN("unable to allocate device, out of memory.");
2267 if (!try_module_get(THIS_MODULE))
2268 goto bad_module_get;
2270 /* get a minor number for the dev */
2271 if (minor == DM_ANY_MINOR)
2272 r = next_free_minor(&minor);
2274 r = specific_minor(minor);
2278 r = init_srcu_struct(&md->io_barrier);
2280 goto bad_io_barrier;
2282 md->use_blk_mq = use_blk_mq;
2283 md->type = DM_TYPE_NONE;
2284 mutex_init(&md->suspend_lock);
2285 mutex_init(&md->type_lock);
2286 mutex_init(&md->table_devices_lock);
2287 spin_lock_init(&md->deferred_lock);
2288 atomic_set(&md->holders, 1);
2289 atomic_set(&md->open_count, 0);
2290 atomic_set(&md->event_nr, 0);
2291 atomic_set(&md->uevent_seq, 0);
2292 INIT_LIST_HEAD(&md->uevent_list);
2293 INIT_LIST_HEAD(&md->table_devices);
2294 spin_lock_init(&md->uevent_lock);
2296 md->queue = blk_alloc_queue(GFP_KERNEL);
2300 dm_init_md_queue(md);
2302 md->disk = alloc_disk(1);
2306 atomic_set(&md->pending[0], 0);
2307 atomic_set(&md->pending[1], 0);
2308 init_waitqueue_head(&md->wait);
2309 INIT_WORK(&md->work, dm_wq_work);
2310 init_waitqueue_head(&md->eventq);
2311 init_completion(&md->kobj_holder.completion);
2312 md->kworker_task = NULL;
2314 md->disk->major = _major;
2315 md->disk->first_minor = minor;
2316 md->disk->fops = &dm_blk_dops;
2317 md->disk->queue = md->queue;
2318 md->disk->private_data = md;
2319 sprintf(md->disk->disk_name, "dm-%d", minor);
2321 format_dev_t(md->name, MKDEV(_major, minor));
2323 md->wq = alloc_workqueue("kdmflush", WQ_MEM_RECLAIM, 0);
2327 md->bdev = bdget_disk(md->disk, 0);
2331 bio_init(&md->flush_bio);
2332 md->flush_bio.bi_bdev = md->bdev;
2333 md->flush_bio.bi_rw = WRITE_FLUSH;
2335 dm_stats_init(&md->stats);
2337 /* Populate the mapping, nobody knows we exist yet */
2338 spin_lock(&_minor_lock);
2339 old_md = idr_replace(&_minor_idr, md, minor);
2340 spin_unlock(&_minor_lock);
2342 BUG_ON(old_md != MINOR_ALLOCED);
2347 destroy_workqueue(md->wq);
2349 del_gendisk(md->disk);
2352 blk_cleanup_queue(md->queue);
2354 cleanup_srcu_struct(&md->io_barrier);
2358 module_put(THIS_MODULE);
2364 static void unlock_fs(struct mapped_device *md);
2366 static void free_dev(struct mapped_device *md)
2368 int minor = MINOR(disk_devt(md->disk));
2371 destroy_workqueue(md->wq);
2373 if (md->kworker_task)
2374 kthread_stop(md->kworker_task);
2376 mempool_destroy(md->io_pool);
2378 mempool_destroy(md->rq_pool);
2380 bioset_free(md->bs);
2382 cleanup_srcu_struct(&md->io_barrier);
2383 free_table_devices(&md->table_devices);
2384 dm_stats_cleanup(&md->stats);
2386 spin_lock(&_minor_lock);
2387 md->disk->private_data = NULL;
2388 spin_unlock(&_minor_lock);
2389 if (blk_get_integrity(md->disk))
2390 blk_integrity_unregister(md->disk);
2391 del_gendisk(md->disk);
2393 blk_cleanup_queue(md->queue);
2395 blk_mq_free_tag_set(&md->tag_set);
2399 module_put(THIS_MODULE);
2403 static void __bind_mempools(struct mapped_device *md, struct dm_table *t)
2405 struct dm_md_mempools *p = dm_table_get_md_mempools(t);
2408 /* The md already has necessary mempools. */
2409 if (dm_table_get_type(t) == DM_TYPE_BIO_BASED) {
2411 * Reload bioset because front_pad may have changed
2412 * because a different table was loaded.
2414 bioset_free(md->bs);
2419 * There's no need to reload with request-based dm
2420 * because the size of front_pad doesn't change.
2421 * Note for future: If you are to reload bioset,
2422 * prep-ed requests in the queue may refer
2423 * to bio from the old bioset, so you must walk
2424 * through the queue to unprep.
2429 BUG_ON(!p || md->io_pool || md->rq_pool || md->bs);
2431 md->io_pool = p->io_pool;
2433 md->rq_pool = p->rq_pool;
2439 /* mempool bind completed, no longer need any mempools in the table */
2440 dm_table_free_md_mempools(t);
2444 * Bind a table to the device.
2446 static void event_callback(void *context)
2448 unsigned long flags;
2450 struct mapped_device *md = (struct mapped_device *) context;
2452 spin_lock_irqsave(&md->uevent_lock, flags);
2453 list_splice_init(&md->uevent_list, &uevents);
2454 spin_unlock_irqrestore(&md->uevent_lock, flags);
2456 dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj);
2458 atomic_inc(&md->event_nr);
2459 wake_up(&md->eventq);
2463 * Protected by md->suspend_lock obtained by dm_swap_table().
2465 static void __set_size(struct mapped_device *md, sector_t size)
2467 set_capacity(md->disk, size);
2469 i_size_write(md->bdev->bd_inode, (loff_t)size << SECTOR_SHIFT);
2473 * Return 1 if the queue has a compulsory merge_bvec_fn function.
2475 * If this function returns 0, then the device is either a non-dm
2476 * device without a merge_bvec_fn, or it is a dm device that is
2477 * able to split any bios it receives that are too big.
2479 int dm_queue_merge_is_compulsory(struct request_queue *q)
2481 struct mapped_device *dev_md;
2483 if (!q->merge_bvec_fn)
2486 if (q->make_request_fn == dm_make_request) {
2487 dev_md = q->queuedata;
2488 if (test_bit(DMF_MERGE_IS_OPTIONAL, &dev_md->flags))
2495 static int dm_device_merge_is_compulsory(struct dm_target *ti,
2496 struct dm_dev *dev, sector_t start,
2497 sector_t len, void *data)
2499 struct block_device *bdev = dev->bdev;
2500 struct request_queue *q = bdev_get_queue(bdev);
2502 return dm_queue_merge_is_compulsory(q);
2506 * Return 1 if it is acceptable to ignore merge_bvec_fn based
2507 * on the properties of the underlying devices.
2509 static int dm_table_merge_is_optional(struct dm_table *table)
2512 struct dm_target *ti;
2514 while (i < dm_table_get_num_targets(table)) {
2515 ti = dm_table_get_target(table, i++);
2517 if (ti->type->iterate_devices &&
2518 ti->type->iterate_devices(ti, dm_device_merge_is_compulsory, NULL))
2526 * Returns old map, which caller must destroy.
2528 static struct dm_table *__bind(struct mapped_device *md, struct dm_table *t,
2529 struct queue_limits *limits)
2531 struct dm_table *old_map;
2532 struct request_queue *q = md->queue;
2534 int merge_is_optional;
2536 size = dm_table_get_size(t);
2539 * Wipe any geometry if the size of the table changed.
2541 if (size != dm_get_size(md))
2542 memset(&md->geometry, 0, sizeof(md->geometry));
2544 __set_size(md, size);
2546 dm_table_event_callback(t, event_callback, md);
2549 * The queue hasn't been stopped yet, if the old table type wasn't
2550 * for request-based during suspension. So stop it to prevent
2551 * I/O mapping before resume.
2552 * This must be done before setting the queue restrictions,
2553 * because request-based dm may be run just after the setting.
2555 if (dm_table_request_based(t))
2558 __bind_mempools(md, t);
2560 merge_is_optional = dm_table_merge_is_optional(t);
2562 old_map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2563 rcu_assign_pointer(md->map, t);
2564 md->immutable_target_type = dm_table_get_immutable_target_type(t);
2566 dm_table_set_restrictions(t, q, limits);
2567 if (merge_is_optional)
2568 set_bit(DMF_MERGE_IS_OPTIONAL, &md->flags);
2570 clear_bit(DMF_MERGE_IS_OPTIONAL, &md->flags);
2578 * Returns unbound table for the caller to free.
2580 static struct dm_table *__unbind(struct mapped_device *md)
2582 struct dm_table *map = rcu_dereference_protected(md->map, 1);
2587 dm_table_event_callback(map, NULL, NULL);
2588 RCU_INIT_POINTER(md->map, NULL);
2595 * Constructor for a new device.
2597 int dm_create(int minor, struct mapped_device **result)
2599 struct mapped_device *md;
2601 md = alloc_dev(minor);
2612 * Functions to manage md->type.
2613 * All are required to hold md->type_lock.
2615 void dm_lock_md_type(struct mapped_device *md)
2617 mutex_lock(&md->type_lock);
2620 void dm_unlock_md_type(struct mapped_device *md)
2622 mutex_unlock(&md->type_lock);
2625 void dm_set_md_type(struct mapped_device *md, unsigned type)
2627 BUG_ON(!mutex_is_locked(&md->type_lock));
2631 unsigned dm_get_md_type(struct mapped_device *md)
2633 BUG_ON(!mutex_is_locked(&md->type_lock));
2637 struct target_type *dm_get_immutable_target_type(struct mapped_device *md)
2639 return md->immutable_target_type;
2643 * The queue_limits are only valid as long as you have a reference
2646 struct queue_limits *dm_get_queue_limits(struct mapped_device *md)
2648 BUG_ON(!atomic_read(&md->holders));
2649 return &md->queue->limits;
2651 EXPORT_SYMBOL_GPL(dm_get_queue_limits);
2653 static void init_rq_based_worker_thread(struct mapped_device *md)
2655 /* Initialize the request-based DM worker thread */
2656 init_kthread_worker(&md->kworker);
2657 md->kworker_task = kthread_run(kthread_worker_fn, &md->kworker,
2658 "kdmwork-%s", dm_device_name(md));
2662 * Fully initialize a request-based queue (->elevator, ->request_fn, etc).
2664 static int dm_init_request_based_queue(struct mapped_device *md)
2666 struct request_queue *q = NULL;
2668 /* Fully initialize the queue */
2669 q = blk_init_allocated_queue(md->queue, dm_request_fn, NULL);
2673 /* disable dm_request_fn's merge heuristic by default */
2674 md->seq_rq_merge_deadline_usecs = 0;
2677 dm_init_old_md_queue(md);
2678 blk_queue_softirq_done(md->queue, dm_softirq_done);
2679 blk_queue_prep_rq(md->queue, dm_prep_fn);
2681 init_rq_based_worker_thread(md);
2683 elv_register_queue(md->queue);
2688 static int dm_mq_init_request(void *data, struct request *rq,
2689 unsigned int hctx_idx, unsigned int request_idx,
2690 unsigned int numa_node)
2692 struct mapped_device *md = data;
2693 struct dm_rq_target_io *tio = blk_mq_rq_to_pdu(rq);
2696 * Must initialize md member of tio, otherwise it won't
2697 * be available in dm_mq_queue_rq.
2704 static int dm_mq_queue_rq(struct blk_mq_hw_ctx *hctx,
2705 const struct blk_mq_queue_data *bd)
2707 struct request *rq = bd->rq;
2708 struct dm_rq_target_io *tio = blk_mq_rq_to_pdu(rq);
2709 struct mapped_device *md = tio->md;
2711 struct dm_table *map = dm_get_live_table(md, &srcu_idx);
2712 struct dm_target *ti;
2715 /* always use block 0 to find the target for flushes for now */
2717 if (!(rq->cmd_flags & REQ_FLUSH))
2718 pos = blk_rq_pos(rq);
2720 ti = dm_table_find_target(map, pos);
2721 if (!dm_target_is_valid(ti)) {
2722 dm_put_live_table(md, srcu_idx);
2723 DMERR_LIMIT("request attempted access beyond the end of device");
2725 * Must perform setup, that rq_completed() requires,
2726 * before returning BLK_MQ_RQ_QUEUE_ERROR
2728 dm_start_request(md, rq);
2729 return BLK_MQ_RQ_QUEUE_ERROR;
2731 dm_put_live_table(md, srcu_idx);
2733 if (ti->type->busy && ti->type->busy(ti))
2734 return BLK_MQ_RQ_QUEUE_BUSY;
2736 dm_start_request(md, rq);
2738 /* Init tio using md established in .init_request */
2739 init_tio(tio, rq, md);
2742 * Establish tio->ti before queuing work (map_tio_request)
2743 * or making direct call to map_request().
2747 /* Clone the request if underlying devices aren't blk-mq */
2748 if (dm_table_get_type(map) == DM_TYPE_REQUEST_BASED) {
2749 /* clone request is allocated at the end of the pdu */
2750 tio->clone = (void *)blk_mq_rq_to_pdu(rq) + sizeof(struct dm_rq_target_io);
2751 (void) clone_rq(rq, md, tio, GFP_ATOMIC);
2752 queue_kthread_work(&md->kworker, &tio->work);
2754 /* Direct call is fine since .queue_rq allows allocations */
2755 if (map_request(tio, rq, md) == DM_MAPIO_REQUEUE) {
2756 /* Undo dm_start_request() before requeuing */
2757 rq_completed(md, rq_data_dir(rq), false);
2758 return BLK_MQ_RQ_QUEUE_BUSY;
2762 return BLK_MQ_RQ_QUEUE_OK;
2765 static struct blk_mq_ops dm_mq_ops = {
2766 .queue_rq = dm_mq_queue_rq,
2767 .map_queue = blk_mq_map_queue,
2768 .complete = dm_softirq_done,
2769 .init_request = dm_mq_init_request,
2772 static int dm_init_request_based_blk_mq_queue(struct mapped_device *md)
2774 unsigned md_type = dm_get_md_type(md);
2775 struct request_queue *q;
2778 memset(&md->tag_set, 0, sizeof(md->tag_set));
2779 md->tag_set.ops = &dm_mq_ops;
2780 md->tag_set.queue_depth = BLKDEV_MAX_RQ;
2781 md->tag_set.numa_node = NUMA_NO_NODE;
2782 md->tag_set.flags = BLK_MQ_F_SHOULD_MERGE | BLK_MQ_F_SG_MERGE;
2783 md->tag_set.nr_hw_queues = 1;
2784 if (md_type == DM_TYPE_REQUEST_BASED) {
2785 /* make the memory for non-blk-mq clone part of the pdu */
2786 md->tag_set.cmd_size = sizeof(struct dm_rq_target_io) + sizeof(struct request);
2788 md->tag_set.cmd_size = sizeof(struct dm_rq_target_io);
2789 md->tag_set.driver_data = md;
2791 err = blk_mq_alloc_tag_set(&md->tag_set);
2795 q = blk_mq_init_allocated_queue(&md->tag_set, md->queue);
2801 dm_init_md_queue(md);
2803 /* backfill 'mq' sysfs registration normally done in blk_register_queue */
2804 blk_mq_register_disk(md->disk);
2806 if (md_type == DM_TYPE_REQUEST_BASED)
2807 init_rq_based_worker_thread(md);
2812 blk_mq_free_tag_set(&md->tag_set);
2816 static unsigned filter_md_type(unsigned type, struct mapped_device *md)
2818 if (type == DM_TYPE_BIO_BASED)
2821 return !md->use_blk_mq ? DM_TYPE_REQUEST_BASED : DM_TYPE_MQ_REQUEST_BASED;
2825 * Setup the DM device's queue based on md's type
2827 int dm_setup_md_queue(struct mapped_device *md)
2830 unsigned md_type = filter_md_type(dm_get_md_type(md), md);
2833 case DM_TYPE_REQUEST_BASED:
2834 r = dm_init_request_based_queue(md);
2836 DMWARN("Cannot initialize queue for request-based mapped device");
2840 case DM_TYPE_MQ_REQUEST_BASED:
2841 r = dm_init_request_based_blk_mq_queue(md);
2843 DMWARN("Cannot initialize queue for request-based blk-mq mapped device");
2847 case DM_TYPE_BIO_BASED:
2848 dm_init_old_md_queue(md);
2849 blk_queue_make_request(md->queue, dm_make_request);
2850 blk_queue_merge_bvec(md->queue, dm_merge_bvec);
2857 struct mapped_device *dm_get_md(dev_t dev)
2859 struct mapped_device *md;
2860 unsigned minor = MINOR(dev);
2862 if (MAJOR(dev) != _major || minor >= (1 << MINORBITS))
2865 spin_lock(&_minor_lock);
2867 md = idr_find(&_minor_idr, minor);
2869 if ((md == MINOR_ALLOCED ||
2870 (MINOR(disk_devt(dm_disk(md))) != minor) ||
2871 dm_deleting_md(md) ||
2872 test_bit(DMF_FREEING, &md->flags))) {
2880 spin_unlock(&_minor_lock);
2884 EXPORT_SYMBOL_GPL(dm_get_md);
2886 void *dm_get_mdptr(struct mapped_device *md)
2888 return md->interface_ptr;
2891 void dm_set_mdptr(struct mapped_device *md, void *ptr)
2893 md->interface_ptr = ptr;
2896 void dm_get(struct mapped_device *md)
2898 atomic_inc(&md->holders);
2899 BUG_ON(test_bit(DMF_FREEING, &md->flags));
2902 int dm_hold(struct mapped_device *md)
2904 spin_lock(&_minor_lock);
2905 if (test_bit(DMF_FREEING, &md->flags)) {
2906 spin_unlock(&_minor_lock);
2910 spin_unlock(&_minor_lock);
2913 EXPORT_SYMBOL_GPL(dm_hold);
2915 const char *dm_device_name(struct mapped_device *md)
2919 EXPORT_SYMBOL_GPL(dm_device_name);
2921 static void __dm_destroy(struct mapped_device *md, bool wait)
2923 struct dm_table *map;
2928 map = dm_get_live_table(md, &srcu_idx);
2930 spin_lock(&_minor_lock);
2931 idr_replace(&_minor_idr, MINOR_ALLOCED, MINOR(disk_devt(dm_disk(md))));
2932 set_bit(DMF_FREEING, &md->flags);
2933 spin_unlock(&_minor_lock);
2935 if (dm_request_based(md) && md->kworker_task)
2936 flush_kthread_worker(&md->kworker);
2939 * Take suspend_lock so that presuspend and postsuspend methods
2940 * do not race with internal suspend.
2942 mutex_lock(&md->suspend_lock);
2943 if (!dm_suspended_md(md)) {
2944 dm_table_presuspend_targets(map);
2945 dm_table_postsuspend_targets(map);
2947 mutex_unlock(&md->suspend_lock);
2949 /* dm_put_live_table must be before msleep, otherwise deadlock is possible */
2950 dm_put_live_table(md, srcu_idx);
2953 * Rare, but there may be I/O requests still going to complete,
2954 * for example. Wait for all references to disappear.
2955 * No one should increment the reference count of the mapped_device,
2956 * after the mapped_device state becomes DMF_FREEING.
2959 while (atomic_read(&md->holders))
2961 else if (atomic_read(&md->holders))
2962 DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)",
2963 dm_device_name(md), atomic_read(&md->holders));
2966 dm_table_destroy(__unbind(md));
2970 void dm_destroy(struct mapped_device *md)
2972 __dm_destroy(md, true);
2975 void dm_destroy_immediate(struct mapped_device *md)
2977 __dm_destroy(md, false);
2980 void dm_put(struct mapped_device *md)
2982 atomic_dec(&md->holders);
2984 EXPORT_SYMBOL_GPL(dm_put);
2986 static int dm_wait_for_completion(struct mapped_device *md, int interruptible)
2989 DECLARE_WAITQUEUE(wait, current);
2991 add_wait_queue(&md->wait, &wait);
2994 set_current_state(interruptible);
2996 if (!md_in_flight(md))
2999 if (interruptible == TASK_INTERRUPTIBLE &&
3000 signal_pending(current)) {
3007 set_current_state(TASK_RUNNING);
3009 remove_wait_queue(&md->wait, &wait);
3015 * Process the deferred bios
3017 static void dm_wq_work(struct work_struct *work)
3019 struct mapped_device *md = container_of(work, struct mapped_device,
3023 struct dm_table *map;
3025 map = dm_get_live_table(md, &srcu_idx);
3027 while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
3028 spin_lock_irq(&md->deferred_lock);
3029 c = bio_list_pop(&md->deferred);
3030 spin_unlock_irq(&md->deferred_lock);
3035 if (dm_request_based(md))
3036 generic_make_request(c);
3038 __split_and_process_bio(md, map, c);
3041 dm_put_live_table(md, srcu_idx);
3044 static void dm_queue_flush(struct mapped_device *md)
3046 clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
3047 smp_mb__after_atomic();
3048 queue_work(md->wq, &md->work);
3052 * Swap in a new table, returning the old one for the caller to destroy.
3054 struct dm_table *dm_swap_table(struct mapped_device *md, struct dm_table *table)
3056 struct dm_table *live_map = NULL, *map = ERR_PTR(-EINVAL);
3057 struct queue_limits limits;
3060 mutex_lock(&md->suspend_lock);
3062 /* device must be suspended */
3063 if (!dm_suspended_md(md))
3067 * If the new table has no data devices, retain the existing limits.
3068 * This helps multipath with queue_if_no_path if all paths disappear,
3069 * then new I/O is queued based on these limits, and then some paths
3072 if (dm_table_has_no_data_devices(table)) {
3073 live_map = dm_get_live_table_fast(md);
3075 limits = md->queue->limits;
3076 dm_put_live_table_fast(md);
3080 r = dm_calculate_queue_limits(table, &limits);
3087 map = __bind(md, table, &limits);
3090 mutex_unlock(&md->suspend_lock);
3095 * Functions to lock and unlock any filesystem running on the
3098 static int lock_fs(struct mapped_device *md)
3102 WARN_ON(md->frozen_sb);
3104 md->frozen_sb = freeze_bdev(md->bdev);
3105 if (IS_ERR(md->frozen_sb)) {
3106 r = PTR_ERR(md->frozen_sb);
3107 md->frozen_sb = NULL;
3111 set_bit(DMF_FROZEN, &md->flags);
3116 static void unlock_fs(struct mapped_device *md)
3118 if (!test_bit(DMF_FROZEN, &md->flags))
3121 thaw_bdev(md->bdev, md->frozen_sb);
3122 md->frozen_sb = NULL;
3123 clear_bit(DMF_FROZEN, &md->flags);
3127 * If __dm_suspend returns 0, the device is completely quiescent
3128 * now. There is no request-processing activity. All new requests
3129 * are being added to md->deferred list.
3131 * Caller must hold md->suspend_lock
3133 static int __dm_suspend(struct mapped_device *md, struct dm_table *map,
3134 unsigned suspend_flags, int interruptible)
3136 bool do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG;
3137 bool noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG;
3141 * DMF_NOFLUSH_SUSPENDING must be set before presuspend.
3142 * This flag is cleared before dm_suspend returns.
3145 set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
3148 * This gets reverted if there's an error later and the targets
3149 * provide the .presuspend_undo hook.
3151 dm_table_presuspend_targets(map);
3154 * Flush I/O to the device.
3155 * Any I/O submitted after lock_fs() may not be flushed.
3156 * noflush takes precedence over do_lockfs.
3157 * (lock_fs() flushes I/Os and waits for them to complete.)
3159 if (!noflush && do_lockfs) {
3162 dm_table_presuspend_undo_targets(map);
3168 * Here we must make sure that no processes are submitting requests
3169 * to target drivers i.e. no one may be executing
3170 * __split_and_process_bio. This is called from dm_request and
3173 * To get all processes out of __split_and_process_bio in dm_request,
3174 * we take the write lock. To prevent any process from reentering
3175 * __split_and_process_bio from dm_request and quiesce the thread
3176 * (dm_wq_work), we set BMF_BLOCK_IO_FOR_SUSPEND and call
3177 * flush_workqueue(md->wq).
3179 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
3181 synchronize_srcu(&md->io_barrier);
3184 * Stop md->queue before flushing md->wq in case request-based
3185 * dm defers requests to md->wq from md->queue.
3187 if (dm_request_based(md)) {
3188 stop_queue(md->queue);
3189 if (md->kworker_task)
3190 flush_kthread_worker(&md->kworker);
3193 flush_workqueue(md->wq);
3196 * At this point no more requests are entering target request routines.
3197 * We call dm_wait_for_completion to wait for all existing requests
3200 r = dm_wait_for_completion(md, interruptible);
3203 clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
3205 synchronize_srcu(&md->io_barrier);
3207 /* were we interrupted ? */
3211 if (dm_request_based(md))
3212 start_queue(md->queue);
3215 dm_table_presuspend_undo_targets(map);
3216 /* pushback list is already flushed, so skip flush */
3223 * We need to be able to change a mapping table under a mounted
3224 * filesystem. For example we might want to move some data in
3225 * the background. Before the table can be swapped with
3226 * dm_bind_table, dm_suspend must be called to flush any in
3227 * flight bios and ensure that any further io gets deferred.
3230 * Suspend mechanism in request-based dm.
3232 * 1. Flush all I/Os by lock_fs() if needed.
3233 * 2. Stop dispatching any I/O by stopping the request_queue.
3234 * 3. Wait for all in-flight I/Os to be completed or requeued.
3236 * To abort suspend, start the request_queue.
3238 int dm_suspend(struct mapped_device *md, unsigned suspend_flags)
3240 struct dm_table *map = NULL;
3244 mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING);
3246 if (dm_suspended_md(md)) {
3251 if (dm_suspended_internally_md(md)) {
3252 /* already internally suspended, wait for internal resume */
3253 mutex_unlock(&md->suspend_lock);
3254 r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE);
3260 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
3262 r = __dm_suspend(md, map, suspend_flags, TASK_INTERRUPTIBLE);
3266 set_bit(DMF_SUSPENDED, &md->flags);
3268 dm_table_postsuspend_targets(map);
3271 mutex_unlock(&md->suspend_lock);
3275 static int __dm_resume(struct mapped_device *md, struct dm_table *map)
3278 int r = dm_table_resume_targets(map);
3286 * Flushing deferred I/Os must be done after targets are resumed
3287 * so that mapping of targets can work correctly.
3288 * Request-based dm is queueing the deferred I/Os in its request_queue.
3290 if (dm_request_based(md))
3291 start_queue(md->queue);
3298 int dm_resume(struct mapped_device *md)
3301 struct dm_table *map = NULL;
3304 mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING);
3306 if (!dm_suspended_md(md))
3309 if (dm_suspended_internally_md(md)) {
3310 /* already internally suspended, wait for internal resume */
3311 mutex_unlock(&md->suspend_lock);
3312 r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE);
3318 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
3319 if (!map || !dm_table_get_size(map))
3322 r = __dm_resume(md, map);
3326 clear_bit(DMF_SUSPENDED, &md->flags);
3330 mutex_unlock(&md->suspend_lock);
3336 * Internal suspend/resume works like userspace-driven suspend. It waits
3337 * until all bios finish and prevents issuing new bios to the target drivers.
3338 * It may be used only from the kernel.
3341 static void __dm_internal_suspend(struct mapped_device *md, unsigned suspend_flags)
3343 struct dm_table *map = NULL;
3345 if (md->internal_suspend_count++)
3346 return; /* nested internal suspend */
3348 if (dm_suspended_md(md)) {
3349 set_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
3350 return; /* nest suspend */
3353 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
3356 * Using TASK_UNINTERRUPTIBLE because only NOFLUSH internal suspend is
3357 * supported. Properly supporting a TASK_INTERRUPTIBLE internal suspend
3358 * would require changing .presuspend to return an error -- avoid this
3359 * until there is a need for more elaborate variants of internal suspend.
3361 (void) __dm_suspend(md, map, suspend_flags, TASK_UNINTERRUPTIBLE);
3363 set_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
3365 dm_table_postsuspend_targets(map);
3368 static void __dm_internal_resume(struct mapped_device *md)
3370 BUG_ON(!md->internal_suspend_count);
3372 if (--md->internal_suspend_count)
3373 return; /* resume from nested internal suspend */
3375 if (dm_suspended_md(md))
3376 goto done; /* resume from nested suspend */
3379 * NOTE: existing callers don't need to call dm_table_resume_targets
3380 * (which may fail -- so best to avoid it for now by passing NULL map)
3382 (void) __dm_resume(md, NULL);
3385 clear_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
3386 smp_mb__after_atomic();
3387 wake_up_bit(&md->flags, DMF_SUSPENDED_INTERNALLY);
3390 void dm_internal_suspend_noflush(struct mapped_device *md)
3392 mutex_lock(&md->suspend_lock);
3393 __dm_internal_suspend(md, DM_SUSPEND_NOFLUSH_FLAG);
3394 mutex_unlock(&md->suspend_lock);
3396 EXPORT_SYMBOL_GPL(dm_internal_suspend_noflush);
3398 void dm_internal_resume(struct mapped_device *md)
3400 mutex_lock(&md->suspend_lock);
3401 __dm_internal_resume(md);
3402 mutex_unlock(&md->suspend_lock);
3404 EXPORT_SYMBOL_GPL(dm_internal_resume);
3407 * Fast variants of internal suspend/resume hold md->suspend_lock,
3408 * which prevents interaction with userspace-driven suspend.
3411 void dm_internal_suspend_fast(struct mapped_device *md)
3413 mutex_lock(&md->suspend_lock);
3414 if (dm_suspended_md(md) || dm_suspended_internally_md(md))
3417 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
3418 synchronize_srcu(&md->io_barrier);
3419 flush_workqueue(md->wq);
3420 dm_wait_for_completion(md, TASK_UNINTERRUPTIBLE);
3422 EXPORT_SYMBOL_GPL(dm_internal_suspend_fast);
3424 void dm_internal_resume_fast(struct mapped_device *md)
3426 if (dm_suspended_md(md) || dm_suspended_internally_md(md))
3432 mutex_unlock(&md->suspend_lock);
3434 EXPORT_SYMBOL_GPL(dm_internal_resume_fast);
3436 /*-----------------------------------------------------------------
3437 * Event notification.
3438 *---------------------------------------------------------------*/
3439 int dm_kobject_uevent(struct mapped_device *md, enum kobject_action action,
3442 char udev_cookie[DM_COOKIE_LENGTH];
3443 char *envp[] = { udev_cookie, NULL };
3446 return kobject_uevent(&disk_to_dev(md->disk)->kobj, action);
3448 snprintf(udev_cookie, DM_COOKIE_LENGTH, "%s=%u",
3449 DM_COOKIE_ENV_VAR_NAME, cookie);
3450 return kobject_uevent_env(&disk_to_dev(md->disk)->kobj,
3455 uint32_t dm_next_uevent_seq(struct mapped_device *md)
3457 return atomic_add_return(1, &md->uevent_seq);
3460 uint32_t dm_get_event_nr(struct mapped_device *md)
3462 return atomic_read(&md->event_nr);
3465 int dm_wait_event(struct mapped_device *md, int event_nr)
3467 return wait_event_interruptible(md->eventq,
3468 (event_nr != atomic_read(&md->event_nr)));
3471 void dm_uevent_add(struct mapped_device *md, struct list_head *elist)
3473 unsigned long flags;
3475 spin_lock_irqsave(&md->uevent_lock, flags);
3476 list_add(elist, &md->uevent_list);
3477 spin_unlock_irqrestore(&md->uevent_lock, flags);
3481 * The gendisk is only valid as long as you have a reference
3484 struct gendisk *dm_disk(struct mapped_device *md)
3488 EXPORT_SYMBOL_GPL(dm_disk);
3490 struct kobject *dm_kobject(struct mapped_device *md)
3492 return &md->kobj_holder.kobj;
3495 struct mapped_device *dm_get_from_kobject(struct kobject *kobj)
3497 struct mapped_device *md;
3499 md = container_of(kobj, struct mapped_device, kobj_holder.kobj);
3501 if (test_bit(DMF_FREEING, &md->flags) ||
3509 int dm_suspended_md(struct mapped_device *md)
3511 return test_bit(DMF_SUSPENDED, &md->flags);
3514 int dm_suspended_internally_md(struct mapped_device *md)
3516 return test_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
3519 int dm_test_deferred_remove_flag(struct mapped_device *md)
3521 return test_bit(DMF_DEFERRED_REMOVE, &md->flags);
3524 int dm_suspended(struct dm_target *ti)
3526 return dm_suspended_md(dm_table_get_md(ti->table));
3528 EXPORT_SYMBOL_GPL(dm_suspended);
3530 int dm_noflush_suspending(struct dm_target *ti)
3532 return __noflush_suspending(dm_table_get_md(ti->table));
3534 EXPORT_SYMBOL_GPL(dm_noflush_suspending);
3536 struct dm_md_mempools *dm_alloc_md_mempools(struct mapped_device *md, unsigned type,
3537 unsigned integrity, unsigned per_bio_data_size)
3539 struct dm_md_mempools *pools = kzalloc(sizeof(*pools), GFP_KERNEL);
3540 struct kmem_cache *cachep = NULL;
3541 unsigned int pool_size = 0;
3542 unsigned int front_pad;
3547 type = filter_md_type(type, md);
3550 case DM_TYPE_BIO_BASED:
3552 pool_size = dm_get_reserved_bio_based_ios();
3553 front_pad = roundup(per_bio_data_size, __alignof__(struct dm_target_io)) + offsetof(struct dm_target_io, clone);
3555 case DM_TYPE_REQUEST_BASED:
3556 cachep = _rq_tio_cache;
3557 pool_size = dm_get_reserved_rq_based_ios();
3558 pools->rq_pool = mempool_create_slab_pool(pool_size, _rq_cache);
3559 if (!pools->rq_pool)
3561 /* fall through to setup remaining rq-based pools */
3562 case DM_TYPE_MQ_REQUEST_BASED:
3564 pool_size = dm_get_reserved_rq_based_ios();
3565 front_pad = offsetof(struct dm_rq_clone_bio_info, clone);
3566 /* per_bio_data_size is not used. See __bind_mempools(). */
3567 WARN_ON(per_bio_data_size != 0);
3574 pools->io_pool = mempool_create_slab_pool(pool_size, cachep);
3575 if (!pools->io_pool)
3579 pools->bs = bioset_create_nobvec(pool_size, front_pad);
3583 if (integrity && bioset_integrity_create(pools->bs, pool_size))
3589 dm_free_md_mempools(pools);
3594 void dm_free_md_mempools(struct dm_md_mempools *pools)
3600 mempool_destroy(pools->io_pool);
3603 mempool_destroy(pools->rq_pool);
3606 bioset_free(pools->bs);
3611 static const struct block_device_operations dm_blk_dops = {
3612 .open = dm_blk_open,
3613 .release = dm_blk_close,
3614 .ioctl = dm_blk_ioctl,
3615 .getgeo = dm_blk_getgeo,
3616 .owner = THIS_MODULE
3622 module_init(dm_init);
3623 module_exit(dm_exit);
3625 module_param(major, uint, 0);
3626 MODULE_PARM_DESC(major, "The major number of the device mapper");
3628 module_param(reserved_bio_based_ios, uint, S_IRUGO | S_IWUSR);
3629 MODULE_PARM_DESC(reserved_bio_based_ios, "Reserved IOs in bio-based mempools");
3631 module_param(reserved_rq_based_ios, uint, S_IRUGO | S_IWUSR);
3632 MODULE_PARM_DESC(reserved_rq_based_ios, "Reserved IOs in request-based mempools");
3634 module_param(use_blk_mq, bool, S_IRUGO | S_IWUSR);
3635 MODULE_PARM_DESC(use_blk_mq, "Use block multiqueue for request-based DM devices");
3637 MODULE_DESCRIPTION(DM_NAME " driver");
3638 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
3639 MODULE_LICENSE("GPL");