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;
89 struct dm_stats_aux stats_aux;
90 unsigned long duration_jiffies;
95 * For request-based dm - the bio clones we allocate are embedded in these
98 * We allocate these with bio_alloc_bioset, using the front_pad parameter when
99 * the bioset is created - this means the bio has to come at the end of the
102 struct dm_rq_clone_bio_info {
104 struct dm_rq_target_io *tio;
108 union map_info *dm_get_rq_mapinfo(struct request *rq)
110 if (rq && rq->end_io_data)
111 return &((struct dm_rq_target_io *)rq->end_io_data)->info;
114 EXPORT_SYMBOL_GPL(dm_get_rq_mapinfo);
116 #define MINOR_ALLOCED ((void *)-1)
119 * Bits for the md->flags field.
121 #define DMF_BLOCK_IO_FOR_SUSPEND 0
122 #define DMF_SUSPENDED 1
124 #define DMF_FREEING 3
125 #define DMF_DELETING 4
126 #define DMF_NOFLUSH_SUSPENDING 5
127 #define DMF_MERGE_IS_OPTIONAL 6
128 #define DMF_DEFERRED_REMOVE 7
129 #define DMF_SUSPENDED_INTERNALLY 8
132 * A dummy definition to make RCU happy.
133 * struct dm_table should never be dereferenced in this file.
140 * Work processed by per-device workqueue.
142 struct mapped_device {
143 struct srcu_struct io_barrier;
144 struct mutex suspend_lock;
149 * The current mapping.
150 * Use dm_get_live_table{_fast} or take suspend_lock for
153 struct dm_table __rcu *map;
155 struct list_head table_devices;
156 struct mutex table_devices_lock;
160 struct request_queue *queue;
162 /* Protect queue and type against concurrent access. */
163 struct mutex type_lock;
165 struct target_type *immutable_target_type;
167 struct gendisk *disk;
173 * A list of ios that arrived while we were suspended.
176 wait_queue_head_t wait;
177 struct work_struct work;
178 struct bio_list deferred;
179 spinlock_t deferred_lock;
182 * Processing queue (flush)
184 struct workqueue_struct *wq;
187 * io objects are allocated from here.
198 wait_queue_head_t eventq;
200 struct list_head uevent_list;
201 spinlock_t uevent_lock; /* Protect access to uevent_list */
204 * freeze/thaw support require holding onto a super block
206 struct super_block *frozen_sb;
207 struct block_device *bdev;
209 /* forced geometry settings */
210 struct hd_geometry geometry;
212 /* kobject and completion */
213 struct dm_kobject_holder kobj_holder;
215 /* zero-length flush that will be cloned and submitted to targets */
216 struct bio flush_bio;
218 /* the number of internal suspends */
219 unsigned internal_suspend_count;
221 struct dm_stats stats;
223 struct kthread_worker kworker;
224 struct task_struct *kworker_task;
226 /* for request-based merge heuristic in dm_request_fn() */
227 unsigned seq_rq_merge_deadline_usecs;
229 sector_t last_rq_pos;
230 ktime_t last_rq_start_time;
232 /* for blk-mq request-based DM support */
233 struct blk_mq_tag_set tag_set;
237 #ifdef CONFIG_DM_MQ_DEFAULT
238 static bool use_blk_mq = true;
240 static bool use_blk_mq = false;
243 bool dm_use_blk_mq(struct mapped_device *md)
245 return md->use_blk_mq;
249 * For mempools pre-allocation at the table loading time.
251 struct dm_md_mempools {
257 struct table_device {
258 struct list_head list;
260 struct dm_dev dm_dev;
263 #define RESERVED_BIO_BASED_IOS 16
264 #define RESERVED_REQUEST_BASED_IOS 256
265 #define RESERVED_MAX_IOS 1024
266 static struct kmem_cache *_io_cache;
267 static struct kmem_cache *_rq_tio_cache;
268 static struct kmem_cache *_rq_cache;
271 * Bio-based DM's mempools' reserved IOs set by the user.
273 static unsigned reserved_bio_based_ios = RESERVED_BIO_BASED_IOS;
276 * Request-based DM's mempools' reserved IOs set by the user.
278 static unsigned reserved_rq_based_ios = RESERVED_REQUEST_BASED_IOS;
280 static unsigned __dm_get_module_param(unsigned *module_param,
281 unsigned def, unsigned max)
283 unsigned param = ACCESS_ONCE(*module_param);
284 unsigned modified_param = 0;
287 modified_param = def;
288 else if (param > max)
289 modified_param = max;
291 if (modified_param) {
292 (void)cmpxchg(module_param, param, modified_param);
293 param = modified_param;
299 unsigned dm_get_reserved_bio_based_ios(void)
301 return __dm_get_module_param(&reserved_bio_based_ios,
302 RESERVED_BIO_BASED_IOS, RESERVED_MAX_IOS);
304 EXPORT_SYMBOL_GPL(dm_get_reserved_bio_based_ios);
306 unsigned dm_get_reserved_rq_based_ios(void)
308 return __dm_get_module_param(&reserved_rq_based_ios,
309 RESERVED_REQUEST_BASED_IOS, RESERVED_MAX_IOS);
311 EXPORT_SYMBOL_GPL(dm_get_reserved_rq_based_ios);
313 static int __init local_init(void)
317 /* allocate a slab for the dm_ios */
318 _io_cache = KMEM_CACHE(dm_io, 0);
322 _rq_tio_cache = KMEM_CACHE(dm_rq_target_io, 0);
324 goto out_free_io_cache;
326 _rq_cache = kmem_cache_create("dm_clone_request", sizeof(struct request),
327 __alignof__(struct request), 0, NULL);
329 goto out_free_rq_tio_cache;
331 r = dm_uevent_init();
333 goto out_free_rq_cache;
335 deferred_remove_workqueue = alloc_workqueue("kdmremove", WQ_UNBOUND, 1);
336 if (!deferred_remove_workqueue) {
338 goto out_uevent_exit;
342 r = register_blkdev(_major, _name);
344 goto out_free_workqueue;
352 destroy_workqueue(deferred_remove_workqueue);
356 kmem_cache_destroy(_rq_cache);
357 out_free_rq_tio_cache:
358 kmem_cache_destroy(_rq_tio_cache);
360 kmem_cache_destroy(_io_cache);
365 static void local_exit(void)
367 flush_scheduled_work();
368 destroy_workqueue(deferred_remove_workqueue);
370 kmem_cache_destroy(_rq_cache);
371 kmem_cache_destroy(_rq_tio_cache);
372 kmem_cache_destroy(_io_cache);
373 unregister_blkdev(_major, _name);
378 DMINFO("cleaned up");
381 static int (*_inits[])(void) __initdata = {
392 static void (*_exits[])(void) = {
403 static int __init dm_init(void)
405 const int count = ARRAY_SIZE(_inits);
409 for (i = 0; i < count; i++) {
424 static void __exit dm_exit(void)
426 int i = ARRAY_SIZE(_exits);
432 * Should be empty by this point.
434 idr_destroy(&_minor_idr);
438 * Block device functions
440 int dm_deleting_md(struct mapped_device *md)
442 return test_bit(DMF_DELETING, &md->flags);
445 static int dm_blk_open(struct block_device *bdev, fmode_t mode)
447 struct mapped_device *md;
449 spin_lock(&_minor_lock);
451 md = bdev->bd_disk->private_data;
455 if (test_bit(DMF_FREEING, &md->flags) ||
456 dm_deleting_md(md)) {
462 atomic_inc(&md->open_count);
464 spin_unlock(&_minor_lock);
466 return md ? 0 : -ENXIO;
469 static void dm_blk_close(struct gendisk *disk, fmode_t mode)
471 struct mapped_device *md;
473 spin_lock(&_minor_lock);
475 md = disk->private_data;
479 if (atomic_dec_and_test(&md->open_count) &&
480 (test_bit(DMF_DEFERRED_REMOVE, &md->flags)))
481 queue_work(deferred_remove_workqueue, &deferred_remove_work);
485 spin_unlock(&_minor_lock);
488 int dm_open_count(struct mapped_device *md)
490 return atomic_read(&md->open_count);
494 * Guarantees nothing is using the device before it's deleted.
496 int dm_lock_for_deletion(struct mapped_device *md, bool mark_deferred, bool only_deferred)
500 spin_lock(&_minor_lock);
502 if (dm_open_count(md)) {
505 set_bit(DMF_DEFERRED_REMOVE, &md->flags);
506 } else if (only_deferred && !test_bit(DMF_DEFERRED_REMOVE, &md->flags))
509 set_bit(DMF_DELETING, &md->flags);
511 spin_unlock(&_minor_lock);
516 int dm_cancel_deferred_remove(struct mapped_device *md)
520 spin_lock(&_minor_lock);
522 if (test_bit(DMF_DELETING, &md->flags))
525 clear_bit(DMF_DEFERRED_REMOVE, &md->flags);
527 spin_unlock(&_minor_lock);
532 static void do_deferred_remove(struct work_struct *w)
534 dm_deferred_remove();
537 sector_t dm_get_size(struct mapped_device *md)
539 return get_capacity(md->disk);
542 struct request_queue *dm_get_md_queue(struct mapped_device *md)
547 struct dm_stats *dm_get_stats(struct mapped_device *md)
552 static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo)
554 struct mapped_device *md = bdev->bd_disk->private_data;
556 return dm_get_geometry(md, geo);
559 static int dm_blk_ioctl(struct block_device *bdev, fmode_t mode,
560 unsigned int cmd, unsigned long arg)
562 struct mapped_device *md = bdev->bd_disk->private_data;
564 struct dm_table *map;
565 struct dm_target *tgt;
569 map = dm_get_live_table(md, &srcu_idx);
571 if (!map || !dm_table_get_size(map))
574 /* We only support devices that have a single target */
575 if (dm_table_get_num_targets(map) != 1)
578 tgt = dm_table_get_target(map, 0);
579 if (!tgt->type->ioctl)
582 if (dm_suspended_md(md)) {
587 r = tgt->type->ioctl(tgt, cmd, arg);
590 dm_put_live_table(md, srcu_idx);
592 if (r == -ENOTCONN) {
600 static struct dm_io *alloc_io(struct mapped_device *md)
602 return mempool_alloc(md->io_pool, GFP_NOIO);
605 static void free_io(struct mapped_device *md, struct dm_io *io)
607 mempool_free(io, md->io_pool);
610 static void free_tio(struct mapped_device *md, struct dm_target_io *tio)
612 bio_put(&tio->clone);
615 static struct dm_rq_target_io *alloc_rq_tio(struct mapped_device *md,
618 return mempool_alloc(md->io_pool, gfp_mask);
621 static void free_rq_tio(struct dm_rq_target_io *tio)
623 mempool_free(tio, tio->md->io_pool);
626 static struct request *alloc_clone_request(struct mapped_device *md,
629 return mempool_alloc(md->rq_pool, gfp_mask);
632 static void free_clone_request(struct mapped_device *md, struct request *rq)
634 mempool_free(rq, md->rq_pool);
637 static int md_in_flight(struct mapped_device *md)
639 return atomic_read(&md->pending[READ]) +
640 atomic_read(&md->pending[WRITE]);
643 static void start_io_acct(struct dm_io *io)
645 struct mapped_device *md = io->md;
646 struct bio *bio = io->bio;
648 int rw = bio_data_dir(bio);
650 io->start_time = jiffies;
652 cpu = part_stat_lock();
653 part_round_stats(cpu, &dm_disk(md)->part0);
655 atomic_set(&dm_disk(md)->part0.in_flight[rw],
656 atomic_inc_return(&md->pending[rw]));
658 if (unlikely(dm_stats_used(&md->stats)))
659 dm_stats_account_io(&md->stats, bio->bi_rw, bio->bi_iter.bi_sector,
660 bio_sectors(bio), false, 0, &io->stats_aux);
663 static void end_io_acct(struct dm_io *io)
665 struct mapped_device *md = io->md;
666 struct bio *bio = io->bio;
667 unsigned long duration = jiffies - io->start_time;
669 int rw = bio_data_dir(bio);
671 generic_end_io_acct(rw, &dm_disk(md)->part0, io->start_time);
673 if (unlikely(dm_stats_used(&md->stats)))
674 dm_stats_account_io(&md->stats, bio->bi_rw, bio->bi_iter.bi_sector,
675 bio_sectors(bio), true, duration, &io->stats_aux);
678 * After this is decremented the bio must not be touched if it is
681 pending = atomic_dec_return(&md->pending[rw]);
682 atomic_set(&dm_disk(md)->part0.in_flight[rw], pending);
683 pending += atomic_read(&md->pending[rw^0x1]);
685 /* nudge anyone waiting on suspend queue */
691 * Add the bio to the list of deferred io.
693 static void queue_io(struct mapped_device *md, struct bio *bio)
697 spin_lock_irqsave(&md->deferred_lock, flags);
698 bio_list_add(&md->deferred, bio);
699 spin_unlock_irqrestore(&md->deferred_lock, flags);
700 queue_work(md->wq, &md->work);
704 * Everyone (including functions in this file), should use this
705 * function to access the md->map field, and make sure they call
706 * dm_put_live_table() when finished.
708 struct dm_table *dm_get_live_table(struct mapped_device *md, int *srcu_idx) __acquires(md->io_barrier)
710 *srcu_idx = srcu_read_lock(&md->io_barrier);
712 return srcu_dereference(md->map, &md->io_barrier);
715 void dm_put_live_table(struct mapped_device *md, int srcu_idx) __releases(md->io_barrier)
717 srcu_read_unlock(&md->io_barrier, srcu_idx);
720 void dm_sync_table(struct mapped_device *md)
722 synchronize_srcu(&md->io_barrier);
723 synchronize_rcu_expedited();
727 * A fast alternative to dm_get_live_table/dm_put_live_table.
728 * The caller must not block between these two functions.
730 static struct dm_table *dm_get_live_table_fast(struct mapped_device *md) __acquires(RCU)
733 return rcu_dereference(md->map);
736 static void dm_put_live_table_fast(struct mapped_device *md) __releases(RCU)
742 * Open a table device so we can use it as a map destination.
744 static int open_table_device(struct table_device *td, dev_t dev,
745 struct mapped_device *md)
747 static char *_claim_ptr = "I belong to device-mapper";
748 struct block_device *bdev;
752 BUG_ON(td->dm_dev.bdev);
754 bdev = blkdev_get_by_dev(dev, td->dm_dev.mode | FMODE_EXCL, _claim_ptr);
756 return PTR_ERR(bdev);
758 r = bd_link_disk_holder(bdev, dm_disk(md));
760 blkdev_put(bdev, td->dm_dev.mode | FMODE_EXCL);
764 td->dm_dev.bdev = bdev;
769 * Close a table device that we've been using.
771 static void close_table_device(struct table_device *td, struct mapped_device *md)
773 if (!td->dm_dev.bdev)
776 bd_unlink_disk_holder(td->dm_dev.bdev, dm_disk(md));
777 blkdev_put(td->dm_dev.bdev, td->dm_dev.mode | FMODE_EXCL);
778 td->dm_dev.bdev = NULL;
781 static struct table_device *find_table_device(struct list_head *l, dev_t dev,
783 struct table_device *td;
785 list_for_each_entry(td, l, list)
786 if (td->dm_dev.bdev->bd_dev == dev && td->dm_dev.mode == mode)
792 int dm_get_table_device(struct mapped_device *md, dev_t dev, fmode_t mode,
793 struct dm_dev **result) {
795 struct table_device *td;
797 mutex_lock(&md->table_devices_lock);
798 td = find_table_device(&md->table_devices, dev, mode);
800 td = kmalloc(sizeof(*td), GFP_KERNEL);
802 mutex_unlock(&md->table_devices_lock);
806 td->dm_dev.mode = mode;
807 td->dm_dev.bdev = NULL;
809 if ((r = open_table_device(td, dev, md))) {
810 mutex_unlock(&md->table_devices_lock);
815 format_dev_t(td->dm_dev.name, dev);
817 atomic_set(&td->count, 0);
818 list_add(&td->list, &md->table_devices);
820 atomic_inc(&td->count);
821 mutex_unlock(&md->table_devices_lock);
823 *result = &td->dm_dev;
826 EXPORT_SYMBOL_GPL(dm_get_table_device);
828 void dm_put_table_device(struct mapped_device *md, struct dm_dev *d)
830 struct table_device *td = container_of(d, struct table_device, dm_dev);
832 mutex_lock(&md->table_devices_lock);
833 if (atomic_dec_and_test(&td->count)) {
834 close_table_device(td, md);
838 mutex_unlock(&md->table_devices_lock);
840 EXPORT_SYMBOL(dm_put_table_device);
842 static void free_table_devices(struct list_head *devices)
844 struct list_head *tmp, *next;
846 list_for_each_safe(tmp, next, devices) {
847 struct table_device *td = list_entry(tmp, struct table_device, list);
849 DMWARN("dm_destroy: %s still exists with %d references",
850 td->dm_dev.name, atomic_read(&td->count));
856 * Get the geometry associated with a dm device
858 int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo)
866 * Set the geometry of a device.
868 int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo)
870 sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors;
872 if (geo->start > sz) {
873 DMWARN("Start sector is beyond the geometry limits.");
882 /*-----------------------------------------------------------------
884 * A more elegant soln is in the works that uses the queue
885 * merge fn, unfortunately there are a couple of changes to
886 * the block layer that I want to make for this. So in the
887 * interests of getting something for people to use I give
888 * you this clearly demarcated crap.
889 *---------------------------------------------------------------*/
891 static int __noflush_suspending(struct mapped_device *md)
893 return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
897 * Decrements the number of outstanding ios that a bio has been
898 * cloned into, completing the original io if necc.
900 static void dec_pending(struct dm_io *io, int error)
905 struct mapped_device *md = io->md;
907 /* Push-back supersedes any I/O errors */
908 if (unlikely(error)) {
909 spin_lock_irqsave(&io->endio_lock, flags);
910 if (!(io->error > 0 && __noflush_suspending(md)))
912 spin_unlock_irqrestore(&io->endio_lock, flags);
915 if (atomic_dec_and_test(&io->io_count)) {
916 if (io->error == DM_ENDIO_REQUEUE) {
918 * Target requested pushing back the I/O.
920 spin_lock_irqsave(&md->deferred_lock, flags);
921 if (__noflush_suspending(md))
922 bio_list_add_head(&md->deferred, io->bio);
924 /* noflush suspend was interrupted. */
926 spin_unlock_irqrestore(&md->deferred_lock, flags);
929 io_error = io->error;
934 if (io_error == DM_ENDIO_REQUEUE)
937 if ((bio->bi_rw & REQ_FLUSH) && bio->bi_iter.bi_size) {
939 * Preflush done for flush with data, reissue
942 bio->bi_rw &= ~REQ_FLUSH;
945 /* done with normal IO or empty flush */
946 trace_block_bio_complete(md->queue, bio, io_error);
947 bio_endio(bio, io_error);
952 static void disable_write_same(struct mapped_device *md)
954 struct queue_limits *limits = dm_get_queue_limits(md);
956 /* device doesn't really support WRITE SAME, disable it */
957 limits->max_write_same_sectors = 0;
960 static void clone_endio(struct bio *bio, int error)
963 struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone);
964 struct dm_io *io = tio->io;
965 struct mapped_device *md = tio->io->md;
966 dm_endio_fn endio = tio->ti->type->end_io;
968 if (!bio_flagged(bio, BIO_UPTODATE) && !error)
972 r = endio(tio->ti, bio, error);
973 if (r < 0 || r == DM_ENDIO_REQUEUE)
975 * error and requeue request are handled
979 else if (r == DM_ENDIO_INCOMPLETE)
980 /* The target will handle the io */
983 DMWARN("unimplemented target endio return value: %d", r);
988 if (unlikely(r == -EREMOTEIO && (bio->bi_rw & REQ_WRITE_SAME) &&
989 !bdev_get_queue(bio->bi_bdev)->limits.max_write_same_sectors))
990 disable_write_same(md);
993 dec_pending(io, error);
997 * Partial completion handling for request-based dm
999 static void end_clone_bio(struct bio *clone, int error)
1001 struct dm_rq_clone_bio_info *info =
1002 container_of(clone, struct dm_rq_clone_bio_info, clone);
1003 struct dm_rq_target_io *tio = info->tio;
1004 struct bio *bio = info->orig;
1005 unsigned int nr_bytes = info->orig->bi_iter.bi_size;
1011 * An error has already been detected on the request.
1012 * Once error occurred, just let clone->end_io() handle
1018 * Don't notice the error to the upper layer yet.
1019 * The error handling decision is made by the target driver,
1020 * when the request is completed.
1027 * I/O for the bio successfully completed.
1028 * Notice the data completion to the upper layer.
1032 * bios are processed from the head of the list.
1033 * So the completing bio should always be rq->bio.
1034 * If it's not, something wrong is happening.
1036 if (tio->orig->bio != bio)
1037 DMERR("bio completion is going in the middle of the request");
1040 * Update the original request.
1041 * Do not use blk_end_request() here, because it may complete
1042 * the original request before the clone, and break the ordering.
1044 blk_update_request(tio->orig, 0, nr_bytes);
1047 static struct dm_rq_target_io *tio_from_request(struct request *rq)
1049 return (rq->q->mq_ops ? blk_mq_rq_to_pdu(rq) : rq->special);
1052 static void rq_end_stats(struct mapped_device *md, struct request *orig)
1054 if (unlikely(dm_stats_used(&md->stats))) {
1055 struct dm_rq_target_io *tio = tio_from_request(orig);
1056 tio->duration_jiffies = jiffies - tio->duration_jiffies;
1057 dm_stats_account_io(&md->stats, orig->cmd_flags, blk_rq_pos(orig),
1058 tio->n_sectors, true, tio->duration_jiffies,
1064 * Don't touch any member of the md after calling this function because
1065 * the md may be freed in dm_put() at the end of this function.
1066 * Or do dm_get() before calling this function and dm_put() later.
1068 static void rq_completed(struct mapped_device *md, int rw, bool run_queue)
1070 int nr_requests_pending;
1072 atomic_dec(&md->pending[rw]);
1074 /* nudge anyone waiting on suspend queue */
1075 nr_requests_pending = md_in_flight(md);
1076 if (!nr_requests_pending)
1080 * Run this off this callpath, as drivers could invoke end_io while
1081 * inside their request_fn (and holding the queue lock). Calling
1082 * back into ->request_fn() could deadlock attempting to grab the
1086 if (md->queue->mq_ops)
1087 blk_mq_run_hw_queues(md->queue, true);
1088 else if (!nr_requests_pending ||
1089 (nr_requests_pending >= md->queue->nr_congestion_on))
1090 blk_run_queue_async(md->queue);
1094 * dm_put() must be at the end of this function. See the comment above
1099 static void free_rq_clone(struct request *clone)
1101 struct dm_rq_target_io *tio = clone->end_io_data;
1102 struct mapped_device *md = tio->md;
1104 blk_rq_unprep_clone(clone);
1106 if (md->type == DM_TYPE_MQ_REQUEST_BASED)
1107 /* stacked on blk-mq queue(s) */
1108 tio->ti->type->release_clone_rq(clone);
1109 else if (!md->queue->mq_ops)
1110 /* request_fn queue stacked on request_fn queue(s) */
1111 free_clone_request(md, clone);
1113 * NOTE: for the blk-mq queue stacked on request_fn queue(s) case:
1114 * no need to call free_clone_request() because we leverage blk-mq by
1115 * allocating the clone at the end of the blk-mq pdu (see: clone_rq)
1118 if (!md->queue->mq_ops)
1123 * Complete the clone and the original request.
1124 * Must be called without clone's queue lock held,
1125 * see end_clone_request() for more details.
1127 static void dm_end_request(struct request *clone, int error)
1129 int rw = rq_data_dir(clone);
1130 struct dm_rq_target_io *tio = clone->end_io_data;
1131 struct mapped_device *md = tio->md;
1132 struct request *rq = tio->orig;
1134 if (rq->cmd_type == REQ_TYPE_BLOCK_PC) {
1135 rq->errors = clone->errors;
1136 rq->resid_len = clone->resid_len;
1140 * We are using the sense buffer of the original
1142 * So setting the length of the sense data is enough.
1144 rq->sense_len = clone->sense_len;
1147 free_rq_clone(clone);
1148 rq_end_stats(md, rq);
1150 blk_end_request_all(rq, error);
1152 blk_mq_end_request(rq, error);
1153 rq_completed(md, rw, true);
1156 static void dm_unprep_request(struct request *rq)
1158 struct dm_rq_target_io *tio = tio_from_request(rq);
1159 struct request *clone = tio->clone;
1161 if (!rq->q->mq_ops) {
1163 rq->cmd_flags &= ~REQ_DONTPREP;
1167 free_rq_clone(clone);
1171 * Requeue the original request of a clone.
1173 static void old_requeue_request(struct request *rq)
1175 struct request_queue *q = rq->q;
1176 unsigned long flags;
1178 spin_lock_irqsave(q->queue_lock, flags);
1179 blk_requeue_request(q, rq);
1180 blk_run_queue_async(q);
1181 spin_unlock_irqrestore(q->queue_lock, flags);
1184 static void dm_requeue_original_request(struct mapped_device *md,
1187 int rw = rq_data_dir(rq);
1189 dm_unprep_request(rq);
1191 rq_end_stats(md, rq);
1193 old_requeue_request(rq);
1195 blk_mq_requeue_request(rq);
1196 blk_mq_kick_requeue_list(rq->q);
1199 rq_completed(md, rw, false);
1202 static void old_stop_queue(struct request_queue *q)
1204 unsigned long flags;
1206 if (blk_queue_stopped(q))
1209 spin_lock_irqsave(q->queue_lock, flags);
1211 spin_unlock_irqrestore(q->queue_lock, flags);
1214 static void stop_queue(struct request_queue *q)
1219 blk_mq_stop_hw_queues(q);
1222 static void old_start_queue(struct request_queue *q)
1224 unsigned long flags;
1226 spin_lock_irqsave(q->queue_lock, flags);
1227 if (blk_queue_stopped(q))
1229 spin_unlock_irqrestore(q->queue_lock, flags);
1232 static void start_queue(struct request_queue *q)
1237 blk_mq_start_stopped_hw_queues(q, true);
1240 static void dm_done(struct request *clone, int error, bool mapped)
1243 struct dm_rq_target_io *tio = clone->end_io_data;
1244 dm_request_endio_fn rq_end_io = NULL;
1247 rq_end_io = tio->ti->type->rq_end_io;
1249 if (mapped && rq_end_io)
1250 r = rq_end_io(tio->ti, clone, error, &tio->info);
1253 if (unlikely(r == -EREMOTEIO && (clone->cmd_flags & REQ_WRITE_SAME) &&
1254 !clone->q->limits.max_write_same_sectors))
1255 disable_write_same(tio->md);
1258 /* The target wants to complete the I/O */
1259 dm_end_request(clone, r);
1260 else if (r == DM_ENDIO_INCOMPLETE)
1261 /* The target will handle the I/O */
1263 else if (r == DM_ENDIO_REQUEUE)
1264 /* The target wants to requeue the I/O */
1265 dm_requeue_original_request(tio->md, tio->orig);
1267 DMWARN("unimplemented target endio return value: %d", r);
1273 * Request completion handler for request-based dm
1275 static void dm_softirq_done(struct request *rq)
1278 struct dm_rq_target_io *tio = tio_from_request(rq);
1279 struct request *clone = tio->clone;
1283 rq_end_stats(tio->md, rq);
1284 rw = rq_data_dir(rq);
1285 if (!rq->q->mq_ops) {
1286 blk_end_request_all(rq, tio->error);
1287 rq_completed(tio->md, rw, false);
1290 blk_mq_end_request(rq, tio->error);
1291 rq_completed(tio->md, rw, false);
1296 if (rq->cmd_flags & REQ_FAILED)
1299 dm_done(clone, tio->error, mapped);
1303 * Complete the clone and the original request with the error status
1304 * through softirq context.
1306 static void dm_complete_request(struct request *rq, int error)
1308 struct dm_rq_target_io *tio = tio_from_request(rq);
1311 blk_complete_request(rq);
1315 * Complete the not-mapped clone and the original request with the error status
1316 * through softirq context.
1317 * Target's rq_end_io() function isn't called.
1318 * This may be used when the target's map_rq() or clone_and_map_rq() functions fail.
1320 static void dm_kill_unmapped_request(struct request *rq, int error)
1322 rq->cmd_flags |= REQ_FAILED;
1323 dm_complete_request(rq, error);
1327 * Called with the clone's queue lock held (for non-blk-mq)
1329 static void end_clone_request(struct request *clone, int error)
1331 struct dm_rq_target_io *tio = clone->end_io_data;
1333 if (!clone->q->mq_ops) {
1335 * For just cleaning up the information of the queue in which
1336 * the clone was dispatched.
1337 * The clone is *NOT* freed actually here because it is alloced
1338 * from dm own mempool (REQ_ALLOCED isn't set).
1340 __blk_put_request(clone->q, clone);
1344 * Actual request completion is done in a softirq context which doesn't
1345 * hold the clone's queue lock. Otherwise, deadlock could occur because:
1346 * - another request may be submitted by the upper level driver
1347 * of the stacking during the completion
1348 * - the submission which requires queue lock may be done
1349 * against this clone's queue
1351 dm_complete_request(tio->orig, error);
1355 * Return maximum size of I/O possible at the supplied sector up to the current
1358 static sector_t max_io_len_target_boundary(sector_t sector, struct dm_target *ti)
1360 sector_t target_offset = dm_target_offset(ti, sector);
1362 return ti->len - target_offset;
1365 static sector_t max_io_len(sector_t sector, struct dm_target *ti)
1367 sector_t len = max_io_len_target_boundary(sector, ti);
1368 sector_t offset, max_len;
1371 * Does the target need to split even further?
1373 if (ti->max_io_len) {
1374 offset = dm_target_offset(ti, sector);
1375 if (unlikely(ti->max_io_len & (ti->max_io_len - 1)))
1376 max_len = sector_div(offset, ti->max_io_len);
1378 max_len = offset & (ti->max_io_len - 1);
1379 max_len = ti->max_io_len - max_len;
1388 int dm_set_target_max_io_len(struct dm_target *ti, sector_t len)
1390 if (len > UINT_MAX) {
1391 DMERR("Specified maximum size of target IO (%llu) exceeds limit (%u)",
1392 (unsigned long long)len, UINT_MAX);
1393 ti->error = "Maximum size of target IO is too large";
1397 ti->max_io_len = (uint32_t) len;
1401 EXPORT_SYMBOL_GPL(dm_set_target_max_io_len);
1404 * A target may call dm_accept_partial_bio only from the map routine. It is
1405 * allowed for all bio types except REQ_FLUSH.
1407 * dm_accept_partial_bio informs the dm that the target only wants to process
1408 * additional n_sectors sectors of the bio and the rest of the data should be
1409 * sent in a next bio.
1411 * A diagram that explains the arithmetics:
1412 * +--------------------+---------------+-------+
1414 * +--------------------+---------------+-------+
1416 * <-------------- *tio->len_ptr --------------->
1417 * <------- bi_size ------->
1420 * Region 1 was already iterated over with bio_advance or similar function.
1421 * (it may be empty if the target doesn't use bio_advance)
1422 * Region 2 is the remaining bio size that the target wants to process.
1423 * (it may be empty if region 1 is non-empty, although there is no reason
1425 * The target requires that region 3 is to be sent in the next bio.
1427 * If the target wants to receive multiple copies of the bio (via num_*bios, etc),
1428 * the partially processed part (the sum of regions 1+2) must be the same for all
1429 * copies of the bio.
1431 void dm_accept_partial_bio(struct bio *bio, unsigned n_sectors)
1433 struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone);
1434 unsigned bi_size = bio->bi_iter.bi_size >> SECTOR_SHIFT;
1435 BUG_ON(bio->bi_rw & REQ_FLUSH);
1436 BUG_ON(bi_size > *tio->len_ptr);
1437 BUG_ON(n_sectors > bi_size);
1438 *tio->len_ptr -= bi_size - n_sectors;
1439 bio->bi_iter.bi_size = n_sectors << SECTOR_SHIFT;
1441 EXPORT_SYMBOL_GPL(dm_accept_partial_bio);
1443 static void __map_bio(struct dm_target_io *tio)
1447 struct mapped_device *md;
1448 struct bio *clone = &tio->clone;
1449 struct dm_target *ti = tio->ti;
1451 clone->bi_end_io = clone_endio;
1454 * Map the clone. If r == 0 we don't need to do
1455 * anything, the target has assumed ownership of
1458 atomic_inc(&tio->io->io_count);
1459 sector = clone->bi_iter.bi_sector;
1460 r = ti->type->map(ti, clone);
1461 if (r == DM_MAPIO_REMAPPED) {
1462 /* the bio has been remapped so dispatch it */
1464 trace_block_bio_remap(bdev_get_queue(clone->bi_bdev), clone,
1465 tio->io->bio->bi_bdev->bd_dev, sector);
1467 generic_make_request(clone);
1468 } else if (r < 0 || r == DM_MAPIO_REQUEUE) {
1469 /* error the io and bail out, or requeue it if needed */
1471 dec_pending(tio->io, r);
1474 DMWARN("unimplemented target map return value: %d", r);
1480 struct mapped_device *md;
1481 struct dm_table *map;
1485 unsigned sector_count;
1488 static void bio_setup_sector(struct bio *bio, sector_t sector, unsigned len)
1490 bio->bi_iter.bi_sector = sector;
1491 bio->bi_iter.bi_size = to_bytes(len);
1495 * Creates a bio that consists of range of complete bvecs.
1497 static void clone_bio(struct dm_target_io *tio, struct bio *bio,
1498 sector_t sector, unsigned len)
1500 struct bio *clone = &tio->clone;
1502 __bio_clone_fast(clone, bio);
1504 if (bio_integrity(bio))
1505 bio_integrity_clone(clone, bio, GFP_NOIO);
1507 bio_advance(clone, to_bytes(sector - clone->bi_iter.bi_sector));
1508 clone->bi_iter.bi_size = to_bytes(len);
1510 if (bio_integrity(bio))
1511 bio_integrity_trim(clone, 0, len);
1514 static struct dm_target_io *alloc_tio(struct clone_info *ci,
1515 struct dm_target *ti,
1516 unsigned target_bio_nr)
1518 struct dm_target_io *tio;
1521 clone = bio_alloc_bioset(GFP_NOIO, 0, ci->md->bs);
1522 tio = container_of(clone, struct dm_target_io, clone);
1526 tio->target_bio_nr = target_bio_nr;
1531 static void __clone_and_map_simple_bio(struct clone_info *ci,
1532 struct dm_target *ti,
1533 unsigned target_bio_nr, unsigned *len)
1535 struct dm_target_io *tio = alloc_tio(ci, ti, target_bio_nr);
1536 struct bio *clone = &tio->clone;
1540 __bio_clone_fast(clone, ci->bio);
1542 bio_setup_sector(clone, ci->sector, *len);
1547 static void __send_duplicate_bios(struct clone_info *ci, struct dm_target *ti,
1548 unsigned num_bios, unsigned *len)
1550 unsigned target_bio_nr;
1552 for (target_bio_nr = 0; target_bio_nr < num_bios; target_bio_nr++)
1553 __clone_and_map_simple_bio(ci, ti, target_bio_nr, len);
1556 static int __send_empty_flush(struct clone_info *ci)
1558 unsigned target_nr = 0;
1559 struct dm_target *ti;
1561 BUG_ON(bio_has_data(ci->bio));
1562 while ((ti = dm_table_get_target(ci->map, target_nr++)))
1563 __send_duplicate_bios(ci, ti, ti->num_flush_bios, NULL);
1568 static void __clone_and_map_data_bio(struct clone_info *ci, struct dm_target *ti,
1569 sector_t sector, unsigned *len)
1571 struct bio *bio = ci->bio;
1572 struct dm_target_io *tio;
1573 unsigned target_bio_nr;
1574 unsigned num_target_bios = 1;
1577 * Does the target want to receive duplicate copies of the bio?
1579 if (bio_data_dir(bio) == WRITE && ti->num_write_bios)
1580 num_target_bios = ti->num_write_bios(ti, bio);
1582 for (target_bio_nr = 0; target_bio_nr < num_target_bios; target_bio_nr++) {
1583 tio = alloc_tio(ci, ti, target_bio_nr);
1585 clone_bio(tio, bio, sector, *len);
1590 typedef unsigned (*get_num_bios_fn)(struct dm_target *ti);
1592 static unsigned get_num_discard_bios(struct dm_target *ti)
1594 return ti->num_discard_bios;
1597 static unsigned get_num_write_same_bios(struct dm_target *ti)
1599 return ti->num_write_same_bios;
1602 typedef bool (*is_split_required_fn)(struct dm_target *ti);
1604 static bool is_split_required_for_discard(struct dm_target *ti)
1606 return ti->split_discard_bios;
1609 static int __send_changing_extent_only(struct clone_info *ci,
1610 get_num_bios_fn get_num_bios,
1611 is_split_required_fn is_split_required)
1613 struct dm_target *ti;
1618 ti = dm_table_find_target(ci->map, ci->sector);
1619 if (!dm_target_is_valid(ti))
1623 * Even though the device advertised support for this type of
1624 * request, that does not mean every target supports it, and
1625 * reconfiguration might also have changed that since the
1626 * check was performed.
1628 num_bios = get_num_bios ? get_num_bios(ti) : 0;
1632 if (is_split_required && !is_split_required(ti))
1633 len = min((sector_t)ci->sector_count, max_io_len_target_boundary(ci->sector, ti));
1635 len = min((sector_t)ci->sector_count, max_io_len(ci->sector, ti));
1637 __send_duplicate_bios(ci, ti, num_bios, &len);
1640 } while (ci->sector_count -= len);
1645 static int __send_discard(struct clone_info *ci)
1647 return __send_changing_extent_only(ci, get_num_discard_bios,
1648 is_split_required_for_discard);
1651 static int __send_write_same(struct clone_info *ci)
1653 return __send_changing_extent_only(ci, get_num_write_same_bios, NULL);
1657 * Select the correct strategy for processing a non-flush bio.
1659 static int __split_and_process_non_flush(struct clone_info *ci)
1661 struct bio *bio = ci->bio;
1662 struct dm_target *ti;
1665 if (unlikely(bio->bi_rw & REQ_DISCARD))
1666 return __send_discard(ci);
1667 else if (unlikely(bio->bi_rw & REQ_WRITE_SAME))
1668 return __send_write_same(ci);
1670 ti = dm_table_find_target(ci->map, ci->sector);
1671 if (!dm_target_is_valid(ti))
1674 len = min_t(sector_t, max_io_len(ci->sector, ti), ci->sector_count);
1676 __clone_and_map_data_bio(ci, ti, ci->sector, &len);
1679 ci->sector_count -= len;
1685 * Entry point to split a bio into clones and submit them to the targets.
1687 static void __split_and_process_bio(struct mapped_device *md,
1688 struct dm_table *map, struct bio *bio)
1690 struct clone_info ci;
1693 if (unlikely(!map)) {
1700 ci.io = alloc_io(md);
1702 atomic_set(&ci.io->io_count, 1);
1705 spin_lock_init(&ci.io->endio_lock);
1706 ci.sector = bio->bi_iter.bi_sector;
1708 start_io_acct(ci.io);
1710 if (bio->bi_rw & REQ_FLUSH) {
1711 ci.bio = &ci.md->flush_bio;
1712 ci.sector_count = 0;
1713 error = __send_empty_flush(&ci);
1714 /* dec_pending submits any data associated with flush */
1717 ci.sector_count = bio_sectors(bio);
1718 while (ci.sector_count && !error)
1719 error = __split_and_process_non_flush(&ci);
1722 /* drop the extra reference count */
1723 dec_pending(ci.io, error);
1725 /*-----------------------------------------------------------------
1727 *---------------------------------------------------------------*/
1729 static int dm_merge_bvec(struct request_queue *q,
1730 struct bvec_merge_data *bvm,
1731 struct bio_vec *biovec)
1733 struct mapped_device *md = q->queuedata;
1734 struct dm_table *map = dm_get_live_table_fast(md);
1735 struct dm_target *ti;
1736 sector_t max_sectors, max_size = 0;
1741 ti = dm_table_find_target(map, bvm->bi_sector);
1742 if (!dm_target_is_valid(ti))
1746 * Find maximum amount of I/O that won't need splitting
1748 max_sectors = min(max_io_len(bvm->bi_sector, ti),
1749 (sector_t) queue_max_sectors(q));
1750 max_size = (max_sectors << SECTOR_SHIFT) - bvm->bi_size;
1753 * FIXME: this stop-gap fix _must_ be cleaned up (by passing a sector_t
1754 * to the targets' merge function since it holds sectors not bytes).
1755 * Just doing this as an interim fix for stable@ because the more
1756 * comprehensive cleanup of switching to sector_t will impact every
1757 * DM target that implements a ->merge hook.
1759 if (max_size > INT_MAX)
1763 * merge_bvec_fn() returns number of bytes
1764 * it can accept at this offset
1765 * max is precomputed maximal io size
1767 if (max_size && ti->type->merge)
1768 max_size = ti->type->merge(ti, bvm, biovec, (int) max_size);
1770 * If the target doesn't support merge method and some of the devices
1771 * provided their merge_bvec method (we know this by looking for the
1772 * max_hw_sectors that dm_set_device_limits may set), then we can't
1773 * allow bios with multiple vector entries. So always set max_size
1774 * to 0, and the code below allows just one page.
1776 else if (queue_max_hw_sectors(q) <= PAGE_SIZE >> 9)
1780 dm_put_live_table_fast(md);
1782 * Always allow an entire first page
1784 if (max_size <= biovec->bv_len && !(bvm->bi_size >> SECTOR_SHIFT))
1785 max_size = biovec->bv_len;
1791 * The request function that just remaps the bio built up by
1794 static void dm_make_request(struct request_queue *q, struct bio *bio)
1796 int rw = bio_data_dir(bio);
1797 struct mapped_device *md = q->queuedata;
1799 struct dm_table *map;
1801 map = dm_get_live_table(md, &srcu_idx);
1803 generic_start_io_acct(rw, bio_sectors(bio), &dm_disk(md)->part0);
1805 /* if we're suspended, we have to queue this io for later */
1806 if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))) {
1807 dm_put_live_table(md, srcu_idx);
1809 if (bio_rw(bio) != READA)
1816 __split_and_process_bio(md, map, bio);
1817 dm_put_live_table(md, srcu_idx);
1821 int dm_request_based(struct mapped_device *md)
1823 return blk_queue_stackable(md->queue);
1826 static void dm_dispatch_clone_request(struct request *clone, struct request *rq)
1830 if (blk_queue_io_stat(clone->q))
1831 clone->cmd_flags |= REQ_IO_STAT;
1833 clone->start_time = jiffies;
1834 r = blk_insert_cloned_request(clone->q, clone);
1836 /* must complete clone in terms of original request */
1837 dm_complete_request(rq, r);
1840 static int dm_rq_bio_constructor(struct bio *bio, struct bio *bio_orig,
1843 struct dm_rq_target_io *tio = data;
1844 struct dm_rq_clone_bio_info *info =
1845 container_of(bio, struct dm_rq_clone_bio_info, clone);
1847 info->orig = bio_orig;
1849 bio->bi_end_io = end_clone_bio;
1854 static int setup_clone(struct request *clone, struct request *rq,
1855 struct dm_rq_target_io *tio, gfp_t gfp_mask)
1859 r = blk_rq_prep_clone(clone, rq, tio->md->bs, gfp_mask,
1860 dm_rq_bio_constructor, tio);
1864 clone->cmd = rq->cmd;
1865 clone->cmd_len = rq->cmd_len;
1866 clone->sense = rq->sense;
1867 clone->end_io = end_clone_request;
1868 clone->end_io_data = tio;
1875 static struct request *clone_rq(struct request *rq, struct mapped_device *md,
1876 struct dm_rq_target_io *tio, gfp_t gfp_mask)
1879 * Do not allocate a clone if tio->clone was already set
1880 * (see: dm_mq_queue_rq).
1882 bool alloc_clone = !tio->clone;
1883 struct request *clone;
1886 clone = alloc_clone_request(md, gfp_mask);
1892 blk_rq_init(NULL, clone);
1893 if (setup_clone(clone, rq, tio, gfp_mask)) {
1896 free_clone_request(md, clone);
1903 static void map_tio_request(struct kthread_work *work);
1905 static void init_tio(struct dm_rq_target_io *tio, struct request *rq,
1906 struct mapped_device *md)
1913 memset(&tio->info, 0, sizeof(tio->info));
1914 if (md->kworker_task)
1915 init_kthread_work(&tio->work, map_tio_request);
1918 static struct dm_rq_target_io *prep_tio(struct request *rq,
1919 struct mapped_device *md, gfp_t gfp_mask)
1921 struct dm_rq_target_io *tio;
1923 struct dm_table *table;
1925 tio = alloc_rq_tio(md, gfp_mask);
1929 init_tio(tio, rq, md);
1931 table = dm_get_live_table(md, &srcu_idx);
1932 if (!dm_table_mq_request_based(table)) {
1933 if (!clone_rq(rq, md, tio, gfp_mask)) {
1934 dm_put_live_table(md, srcu_idx);
1939 dm_put_live_table(md, srcu_idx);
1945 * Called with the queue lock held.
1947 static int dm_prep_fn(struct request_queue *q, struct request *rq)
1949 struct mapped_device *md = q->queuedata;
1950 struct dm_rq_target_io *tio;
1952 if (unlikely(rq->special)) {
1953 DMWARN("Already has something in rq->special.");
1954 return BLKPREP_KILL;
1957 tio = prep_tio(rq, md, GFP_ATOMIC);
1959 return BLKPREP_DEFER;
1962 rq->cmd_flags |= REQ_DONTPREP;
1969 * 0 : the request has been processed
1970 * DM_MAPIO_REQUEUE : the original request needs to be requeued
1971 * < 0 : the request was completed due to failure
1973 static int map_request(struct dm_rq_target_io *tio, struct request *rq,
1974 struct mapped_device *md)
1977 struct dm_target *ti = tio->ti;
1978 struct request *clone = NULL;
1982 r = ti->type->map_rq(ti, clone, &tio->info);
1984 r = ti->type->clone_and_map_rq(ti, rq, &tio->info, &clone);
1986 /* The target wants to complete the I/O */
1987 dm_kill_unmapped_request(rq, r);
1990 if (r != DM_MAPIO_REMAPPED)
1992 if (setup_clone(clone, rq, tio, GFP_ATOMIC)) {
1994 ti->type->release_clone_rq(clone);
1995 return DM_MAPIO_REQUEUE;
2000 case DM_MAPIO_SUBMITTED:
2001 /* The target has taken the I/O to submit by itself later */
2003 case DM_MAPIO_REMAPPED:
2004 /* The target has remapped the I/O so dispatch it */
2005 trace_block_rq_remap(clone->q, clone, disk_devt(dm_disk(md)),
2007 dm_dispatch_clone_request(clone, rq);
2009 case DM_MAPIO_REQUEUE:
2010 /* The target wants to requeue the I/O */
2011 dm_requeue_original_request(md, tio->orig);
2015 DMWARN("unimplemented target map return value: %d", r);
2019 /* The target wants to complete the I/O */
2020 dm_kill_unmapped_request(rq, r);
2027 static void map_tio_request(struct kthread_work *work)
2029 struct dm_rq_target_io *tio = container_of(work, struct dm_rq_target_io, work);
2030 struct request *rq = tio->orig;
2031 struct mapped_device *md = tio->md;
2033 if (map_request(tio, rq, md) == DM_MAPIO_REQUEUE)
2034 dm_requeue_original_request(md, rq);
2037 static void dm_start_request(struct mapped_device *md, struct request *orig)
2039 if (!orig->q->mq_ops)
2040 blk_start_request(orig);
2042 blk_mq_start_request(orig);
2043 atomic_inc(&md->pending[rq_data_dir(orig)]);
2045 if (md->seq_rq_merge_deadline_usecs) {
2046 md->last_rq_pos = rq_end_sector(orig);
2047 md->last_rq_rw = rq_data_dir(orig);
2048 md->last_rq_start_time = ktime_get();
2051 if (unlikely(dm_stats_used(&md->stats))) {
2052 struct dm_rq_target_io *tio = tio_from_request(orig);
2053 tio->duration_jiffies = jiffies;
2054 tio->n_sectors = blk_rq_sectors(orig);
2055 dm_stats_account_io(&md->stats, orig->cmd_flags, blk_rq_pos(orig),
2056 tio->n_sectors, false, 0, &tio->stats_aux);
2060 * Hold the md reference here for the in-flight I/O.
2061 * We can't rely on the reference count by device opener,
2062 * because the device may be closed during the request completion
2063 * when all bios are completed.
2064 * See the comment in rq_completed() too.
2069 #define MAX_SEQ_RQ_MERGE_DEADLINE_USECS 100000
2071 ssize_t dm_attr_rq_based_seq_io_merge_deadline_show(struct mapped_device *md, char *buf)
2073 return sprintf(buf, "%u\n", md->seq_rq_merge_deadline_usecs);
2076 ssize_t dm_attr_rq_based_seq_io_merge_deadline_store(struct mapped_device *md,
2077 const char *buf, size_t count)
2081 if (!dm_request_based(md) || md->use_blk_mq)
2084 if (kstrtouint(buf, 10, &deadline))
2087 if (deadline > MAX_SEQ_RQ_MERGE_DEADLINE_USECS)
2088 deadline = MAX_SEQ_RQ_MERGE_DEADLINE_USECS;
2090 md->seq_rq_merge_deadline_usecs = deadline;
2095 static bool dm_request_peeked_before_merge_deadline(struct mapped_device *md)
2097 ktime_t kt_deadline;
2099 if (!md->seq_rq_merge_deadline_usecs)
2102 kt_deadline = ns_to_ktime((u64)md->seq_rq_merge_deadline_usecs * NSEC_PER_USEC);
2103 kt_deadline = ktime_add_safe(md->last_rq_start_time, kt_deadline);
2105 return !ktime_after(ktime_get(), kt_deadline);
2109 * q->request_fn for request-based dm.
2110 * Called with the queue lock held.
2112 static void dm_request_fn(struct request_queue *q)
2114 struct mapped_device *md = q->queuedata;
2116 struct dm_table *map = dm_get_live_table(md, &srcu_idx);
2117 struct dm_target *ti;
2119 struct dm_rq_target_io *tio;
2123 * For suspend, check blk_queue_stopped() and increment
2124 * ->pending within a single queue_lock not to increment the
2125 * number of in-flight I/Os after the queue is stopped in
2128 while (!blk_queue_stopped(q)) {
2129 rq = blk_peek_request(q);
2133 /* always use block 0 to find the target for flushes for now */
2135 if (!(rq->cmd_flags & REQ_FLUSH))
2136 pos = blk_rq_pos(rq);
2138 ti = dm_table_find_target(map, pos);
2139 if (!dm_target_is_valid(ti)) {
2141 * Must perform setup, that rq_completed() requires,
2142 * before calling dm_kill_unmapped_request
2144 DMERR_LIMIT("request attempted access beyond the end of device");
2145 dm_start_request(md, rq);
2146 dm_kill_unmapped_request(rq, -EIO);
2150 if (dm_request_peeked_before_merge_deadline(md) &&
2151 md_in_flight(md) && rq->bio && rq->bio->bi_vcnt == 1 &&
2152 md->last_rq_pos == pos && md->last_rq_rw == rq_data_dir(rq))
2155 if (ti->type->busy && ti->type->busy(ti))
2158 dm_start_request(md, rq);
2160 tio = tio_from_request(rq);
2161 /* Establish tio->ti before queuing work (map_tio_request) */
2163 queue_kthread_work(&md->kworker, &tio->work);
2164 BUG_ON(!irqs_disabled());
2170 blk_delay_queue(q, HZ / 100);
2172 dm_put_live_table(md, srcu_idx);
2175 static int dm_any_congested(void *congested_data, int bdi_bits)
2178 struct mapped_device *md = congested_data;
2179 struct dm_table *map;
2181 if (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
2182 map = dm_get_live_table_fast(md);
2185 * Request-based dm cares about only own queue for
2186 * the query about congestion status of request_queue
2188 if (dm_request_based(md))
2189 r = md->queue->backing_dev_info.wb.state &
2192 r = dm_table_any_congested(map, bdi_bits);
2194 dm_put_live_table_fast(md);
2200 /*-----------------------------------------------------------------
2201 * An IDR is used to keep track of allocated minor numbers.
2202 *---------------------------------------------------------------*/
2203 static void free_minor(int minor)
2205 spin_lock(&_minor_lock);
2206 idr_remove(&_minor_idr, minor);
2207 spin_unlock(&_minor_lock);
2211 * See if the device with a specific minor # is free.
2213 static int specific_minor(int minor)
2217 if (minor >= (1 << MINORBITS))
2220 idr_preload(GFP_KERNEL);
2221 spin_lock(&_minor_lock);
2223 r = idr_alloc(&_minor_idr, MINOR_ALLOCED, minor, minor + 1, GFP_NOWAIT);
2225 spin_unlock(&_minor_lock);
2228 return r == -ENOSPC ? -EBUSY : r;
2232 static int next_free_minor(int *minor)
2236 idr_preload(GFP_KERNEL);
2237 spin_lock(&_minor_lock);
2239 r = idr_alloc(&_minor_idr, MINOR_ALLOCED, 0, 1 << MINORBITS, GFP_NOWAIT);
2241 spin_unlock(&_minor_lock);
2249 static const struct block_device_operations dm_blk_dops;
2251 static void dm_wq_work(struct work_struct *work);
2253 static void dm_init_md_queue(struct mapped_device *md)
2256 * Request-based dm devices cannot be stacked on top of bio-based dm
2257 * devices. The type of this dm device may not have been decided yet.
2258 * The type is decided at the first table loading time.
2259 * To prevent problematic device stacking, clear the queue flag
2260 * for request stacking support until then.
2262 * This queue is new, so no concurrency on the queue_flags.
2264 queue_flag_clear_unlocked(QUEUE_FLAG_STACKABLE, md->queue);
2267 static void dm_init_old_md_queue(struct mapped_device *md)
2269 md->use_blk_mq = false;
2270 dm_init_md_queue(md);
2273 * Initialize aspects of queue that aren't relevant for blk-mq
2275 md->queue->queuedata = md;
2276 md->queue->backing_dev_info.congested_fn = dm_any_congested;
2277 md->queue->backing_dev_info.congested_data = md;
2279 blk_queue_bounce_limit(md->queue, BLK_BOUNCE_ANY);
2282 static void cleanup_mapped_device(struct mapped_device *md)
2284 cleanup_srcu_struct(&md->io_barrier);
2287 destroy_workqueue(md->wq);
2288 if (md->kworker_task)
2289 kthread_stop(md->kworker_task);
2291 mempool_destroy(md->io_pool);
2293 mempool_destroy(md->rq_pool);
2295 bioset_free(md->bs);
2298 spin_lock(&_minor_lock);
2299 md->disk->private_data = NULL;
2300 spin_unlock(&_minor_lock);
2301 if (blk_get_integrity(md->disk))
2302 blk_integrity_unregister(md->disk);
2303 del_gendisk(md->disk);
2308 blk_cleanup_queue(md->queue);
2317 * Allocate and initialise a blank device with a given minor.
2319 static struct mapped_device *alloc_dev(int minor)
2322 struct mapped_device *md = kzalloc(sizeof(*md), GFP_KERNEL);
2326 DMWARN("unable to allocate device, out of memory.");
2330 if (!try_module_get(THIS_MODULE))
2331 goto bad_module_get;
2333 /* get a minor number for the dev */
2334 if (minor == DM_ANY_MINOR)
2335 r = next_free_minor(&minor);
2337 r = specific_minor(minor);
2341 r = init_srcu_struct(&md->io_barrier);
2343 goto bad_io_barrier;
2345 md->use_blk_mq = use_blk_mq;
2346 md->type = DM_TYPE_NONE;
2347 mutex_init(&md->suspend_lock);
2348 mutex_init(&md->type_lock);
2349 mutex_init(&md->table_devices_lock);
2350 spin_lock_init(&md->deferred_lock);
2351 atomic_set(&md->holders, 1);
2352 atomic_set(&md->open_count, 0);
2353 atomic_set(&md->event_nr, 0);
2354 atomic_set(&md->uevent_seq, 0);
2355 INIT_LIST_HEAD(&md->uevent_list);
2356 INIT_LIST_HEAD(&md->table_devices);
2357 spin_lock_init(&md->uevent_lock);
2359 md->queue = blk_alloc_queue(GFP_KERNEL);
2363 dm_init_md_queue(md);
2365 md->disk = alloc_disk(1);
2369 atomic_set(&md->pending[0], 0);
2370 atomic_set(&md->pending[1], 0);
2371 init_waitqueue_head(&md->wait);
2372 INIT_WORK(&md->work, dm_wq_work);
2373 init_waitqueue_head(&md->eventq);
2374 init_completion(&md->kobj_holder.completion);
2375 md->kworker_task = NULL;
2377 md->disk->major = _major;
2378 md->disk->first_minor = minor;
2379 md->disk->fops = &dm_blk_dops;
2380 md->disk->queue = md->queue;
2381 md->disk->private_data = md;
2382 sprintf(md->disk->disk_name, "dm-%d", minor);
2384 format_dev_t(md->name, MKDEV(_major, minor));
2386 md->wq = alloc_workqueue("kdmflush", WQ_MEM_RECLAIM, 0);
2390 md->bdev = bdget_disk(md->disk, 0);
2394 bio_init(&md->flush_bio);
2395 md->flush_bio.bi_bdev = md->bdev;
2396 md->flush_bio.bi_rw = WRITE_FLUSH;
2398 dm_stats_init(&md->stats);
2400 /* Populate the mapping, nobody knows we exist yet */
2401 spin_lock(&_minor_lock);
2402 old_md = idr_replace(&_minor_idr, md, minor);
2403 spin_unlock(&_minor_lock);
2405 BUG_ON(old_md != MINOR_ALLOCED);
2410 cleanup_mapped_device(md);
2414 module_put(THIS_MODULE);
2420 static void unlock_fs(struct mapped_device *md);
2422 static void free_dev(struct mapped_device *md)
2424 int minor = MINOR(disk_devt(md->disk));
2428 cleanup_mapped_device(md);
2430 blk_mq_free_tag_set(&md->tag_set);
2432 free_table_devices(&md->table_devices);
2433 dm_stats_cleanup(&md->stats);
2436 module_put(THIS_MODULE);
2440 static void __bind_mempools(struct mapped_device *md, struct dm_table *t)
2442 struct dm_md_mempools *p = dm_table_get_md_mempools(t);
2445 /* The md already has necessary mempools. */
2446 if (dm_table_get_type(t) == DM_TYPE_BIO_BASED) {
2448 * Reload bioset because front_pad may have changed
2449 * because a different table was loaded.
2451 bioset_free(md->bs);
2456 * There's no need to reload with request-based dm
2457 * because the size of front_pad doesn't change.
2458 * Note for future: If you are to reload bioset,
2459 * prep-ed requests in the queue may refer
2460 * to bio from the old bioset, so you must walk
2461 * through the queue to unprep.
2466 BUG_ON(!p || md->io_pool || md->rq_pool || md->bs);
2468 md->io_pool = p->io_pool;
2470 md->rq_pool = p->rq_pool;
2476 /* mempool bind completed, no longer need any mempools in the table */
2477 dm_table_free_md_mempools(t);
2481 * Bind a table to the device.
2483 static void event_callback(void *context)
2485 unsigned long flags;
2487 struct mapped_device *md = (struct mapped_device *) context;
2489 spin_lock_irqsave(&md->uevent_lock, flags);
2490 list_splice_init(&md->uevent_list, &uevents);
2491 spin_unlock_irqrestore(&md->uevent_lock, flags);
2493 dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj);
2495 atomic_inc(&md->event_nr);
2496 wake_up(&md->eventq);
2500 * Protected by md->suspend_lock obtained by dm_swap_table().
2502 static void __set_size(struct mapped_device *md, sector_t size)
2504 set_capacity(md->disk, size);
2506 i_size_write(md->bdev->bd_inode, (loff_t)size << SECTOR_SHIFT);
2510 * Return 1 if the queue has a compulsory merge_bvec_fn function.
2512 * If this function returns 0, then the device is either a non-dm
2513 * device without a merge_bvec_fn, or it is a dm device that is
2514 * able to split any bios it receives that are too big.
2516 int dm_queue_merge_is_compulsory(struct request_queue *q)
2518 struct mapped_device *dev_md;
2520 if (!q->merge_bvec_fn)
2523 if (q->make_request_fn == dm_make_request) {
2524 dev_md = q->queuedata;
2525 if (test_bit(DMF_MERGE_IS_OPTIONAL, &dev_md->flags))
2532 static int dm_device_merge_is_compulsory(struct dm_target *ti,
2533 struct dm_dev *dev, sector_t start,
2534 sector_t len, void *data)
2536 struct block_device *bdev = dev->bdev;
2537 struct request_queue *q = bdev_get_queue(bdev);
2539 return dm_queue_merge_is_compulsory(q);
2543 * Return 1 if it is acceptable to ignore merge_bvec_fn based
2544 * on the properties of the underlying devices.
2546 static int dm_table_merge_is_optional(struct dm_table *table)
2549 struct dm_target *ti;
2551 while (i < dm_table_get_num_targets(table)) {
2552 ti = dm_table_get_target(table, i++);
2554 if (ti->type->iterate_devices &&
2555 ti->type->iterate_devices(ti, dm_device_merge_is_compulsory, NULL))
2563 * Returns old map, which caller must destroy.
2565 static struct dm_table *__bind(struct mapped_device *md, struct dm_table *t,
2566 struct queue_limits *limits)
2568 struct dm_table *old_map;
2569 struct request_queue *q = md->queue;
2571 int merge_is_optional;
2573 size = dm_table_get_size(t);
2576 * Wipe any geometry if the size of the table changed.
2578 if (size != dm_get_size(md))
2579 memset(&md->geometry, 0, sizeof(md->geometry));
2581 __set_size(md, size);
2583 dm_table_event_callback(t, event_callback, md);
2586 * The queue hasn't been stopped yet, if the old table type wasn't
2587 * for request-based during suspension. So stop it to prevent
2588 * I/O mapping before resume.
2589 * This must be done before setting the queue restrictions,
2590 * because request-based dm may be run just after the setting.
2592 if (dm_table_request_based(t))
2595 __bind_mempools(md, t);
2597 merge_is_optional = dm_table_merge_is_optional(t);
2599 old_map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2600 rcu_assign_pointer(md->map, t);
2601 md->immutable_target_type = dm_table_get_immutable_target_type(t);
2603 dm_table_set_restrictions(t, q, limits);
2604 if (merge_is_optional)
2605 set_bit(DMF_MERGE_IS_OPTIONAL, &md->flags);
2607 clear_bit(DMF_MERGE_IS_OPTIONAL, &md->flags);
2615 * Returns unbound table for the caller to free.
2617 static struct dm_table *__unbind(struct mapped_device *md)
2619 struct dm_table *map = rcu_dereference_protected(md->map, 1);
2624 dm_table_event_callback(map, NULL, NULL);
2625 RCU_INIT_POINTER(md->map, NULL);
2632 * Constructor for a new device.
2634 int dm_create(int minor, struct mapped_device **result)
2636 struct mapped_device *md;
2638 md = alloc_dev(minor);
2649 * Functions to manage md->type.
2650 * All are required to hold md->type_lock.
2652 void dm_lock_md_type(struct mapped_device *md)
2654 mutex_lock(&md->type_lock);
2657 void dm_unlock_md_type(struct mapped_device *md)
2659 mutex_unlock(&md->type_lock);
2662 void dm_set_md_type(struct mapped_device *md, unsigned type)
2664 BUG_ON(!mutex_is_locked(&md->type_lock));
2668 unsigned dm_get_md_type(struct mapped_device *md)
2670 BUG_ON(!mutex_is_locked(&md->type_lock));
2674 struct target_type *dm_get_immutable_target_type(struct mapped_device *md)
2676 return md->immutable_target_type;
2680 * The queue_limits are only valid as long as you have a reference
2683 struct queue_limits *dm_get_queue_limits(struct mapped_device *md)
2685 BUG_ON(!atomic_read(&md->holders));
2686 return &md->queue->limits;
2688 EXPORT_SYMBOL_GPL(dm_get_queue_limits);
2690 static void init_rq_based_worker_thread(struct mapped_device *md)
2692 /* Initialize the request-based DM worker thread */
2693 init_kthread_worker(&md->kworker);
2694 md->kworker_task = kthread_run(kthread_worker_fn, &md->kworker,
2695 "kdmwork-%s", dm_device_name(md));
2699 * Fully initialize a request-based queue (->elevator, ->request_fn, etc).
2701 static int dm_init_request_based_queue(struct mapped_device *md)
2703 struct request_queue *q = NULL;
2705 /* Fully initialize the queue */
2706 q = blk_init_allocated_queue(md->queue, dm_request_fn, NULL);
2710 /* disable dm_request_fn's merge heuristic by default */
2711 md->seq_rq_merge_deadline_usecs = 0;
2714 dm_init_old_md_queue(md);
2715 blk_queue_softirq_done(md->queue, dm_softirq_done);
2716 blk_queue_prep_rq(md->queue, dm_prep_fn);
2718 init_rq_based_worker_thread(md);
2720 elv_register_queue(md->queue);
2725 static int dm_mq_init_request(void *data, struct request *rq,
2726 unsigned int hctx_idx, unsigned int request_idx,
2727 unsigned int numa_node)
2729 struct mapped_device *md = data;
2730 struct dm_rq_target_io *tio = blk_mq_rq_to_pdu(rq);
2733 * Must initialize md member of tio, otherwise it won't
2734 * be available in dm_mq_queue_rq.
2741 static int dm_mq_queue_rq(struct blk_mq_hw_ctx *hctx,
2742 const struct blk_mq_queue_data *bd)
2744 struct request *rq = bd->rq;
2745 struct dm_rq_target_io *tio = blk_mq_rq_to_pdu(rq);
2746 struct mapped_device *md = tio->md;
2748 struct dm_table *map = dm_get_live_table(md, &srcu_idx);
2749 struct dm_target *ti;
2752 /* always use block 0 to find the target for flushes for now */
2754 if (!(rq->cmd_flags & REQ_FLUSH))
2755 pos = blk_rq_pos(rq);
2757 ti = dm_table_find_target(map, pos);
2758 if (!dm_target_is_valid(ti)) {
2759 dm_put_live_table(md, srcu_idx);
2760 DMERR_LIMIT("request attempted access beyond the end of device");
2762 * Must perform setup, that rq_completed() requires,
2763 * before returning BLK_MQ_RQ_QUEUE_ERROR
2765 dm_start_request(md, rq);
2766 return BLK_MQ_RQ_QUEUE_ERROR;
2768 dm_put_live_table(md, srcu_idx);
2770 if (ti->type->busy && ti->type->busy(ti))
2771 return BLK_MQ_RQ_QUEUE_BUSY;
2773 dm_start_request(md, rq);
2775 /* Init tio using md established in .init_request */
2776 init_tio(tio, rq, md);
2779 * Establish tio->ti before queuing work (map_tio_request)
2780 * or making direct call to map_request().
2784 /* Clone the request if underlying devices aren't blk-mq */
2785 if (dm_table_get_type(map) == DM_TYPE_REQUEST_BASED) {
2786 /* clone request is allocated at the end of the pdu */
2787 tio->clone = (void *)blk_mq_rq_to_pdu(rq) + sizeof(struct dm_rq_target_io);
2788 (void) clone_rq(rq, md, tio, GFP_ATOMIC);
2789 queue_kthread_work(&md->kworker, &tio->work);
2791 /* Direct call is fine since .queue_rq allows allocations */
2792 if (map_request(tio, rq, md) == DM_MAPIO_REQUEUE) {
2793 /* Undo dm_start_request() before requeuing */
2794 rq_end_stats(md, rq);
2795 rq_completed(md, rq_data_dir(rq), false);
2796 return BLK_MQ_RQ_QUEUE_BUSY;
2800 return BLK_MQ_RQ_QUEUE_OK;
2803 static struct blk_mq_ops dm_mq_ops = {
2804 .queue_rq = dm_mq_queue_rq,
2805 .map_queue = blk_mq_map_queue,
2806 .complete = dm_softirq_done,
2807 .init_request = dm_mq_init_request,
2810 static int dm_init_request_based_blk_mq_queue(struct mapped_device *md)
2812 unsigned md_type = dm_get_md_type(md);
2813 struct request_queue *q;
2816 memset(&md->tag_set, 0, sizeof(md->tag_set));
2817 md->tag_set.ops = &dm_mq_ops;
2818 md->tag_set.queue_depth = BLKDEV_MAX_RQ;
2819 md->tag_set.numa_node = NUMA_NO_NODE;
2820 md->tag_set.flags = BLK_MQ_F_SHOULD_MERGE | BLK_MQ_F_SG_MERGE;
2821 md->tag_set.nr_hw_queues = 1;
2822 if (md_type == DM_TYPE_REQUEST_BASED) {
2823 /* make the memory for non-blk-mq clone part of the pdu */
2824 md->tag_set.cmd_size = sizeof(struct dm_rq_target_io) + sizeof(struct request);
2826 md->tag_set.cmd_size = sizeof(struct dm_rq_target_io);
2827 md->tag_set.driver_data = md;
2829 err = blk_mq_alloc_tag_set(&md->tag_set);
2833 q = blk_mq_init_allocated_queue(&md->tag_set, md->queue);
2839 dm_init_md_queue(md);
2841 /* backfill 'mq' sysfs registration normally done in blk_register_queue */
2842 blk_mq_register_disk(md->disk);
2844 if (md_type == DM_TYPE_REQUEST_BASED)
2845 init_rq_based_worker_thread(md);
2850 blk_mq_free_tag_set(&md->tag_set);
2854 static unsigned filter_md_type(unsigned type, struct mapped_device *md)
2856 if (type == DM_TYPE_BIO_BASED)
2859 return !md->use_blk_mq ? DM_TYPE_REQUEST_BASED : DM_TYPE_MQ_REQUEST_BASED;
2863 * Setup the DM device's queue based on md's type
2865 int dm_setup_md_queue(struct mapped_device *md)
2868 unsigned md_type = filter_md_type(dm_get_md_type(md), md);
2871 case DM_TYPE_REQUEST_BASED:
2872 r = dm_init_request_based_queue(md);
2874 DMWARN("Cannot initialize queue for request-based mapped device");
2878 case DM_TYPE_MQ_REQUEST_BASED:
2879 r = dm_init_request_based_blk_mq_queue(md);
2881 DMWARN("Cannot initialize queue for request-based blk-mq mapped device");
2885 case DM_TYPE_BIO_BASED:
2886 dm_init_old_md_queue(md);
2887 blk_queue_make_request(md->queue, dm_make_request);
2888 blk_queue_merge_bvec(md->queue, dm_merge_bvec);
2895 struct mapped_device *dm_get_md(dev_t dev)
2897 struct mapped_device *md;
2898 unsigned minor = MINOR(dev);
2900 if (MAJOR(dev) != _major || minor >= (1 << MINORBITS))
2903 spin_lock(&_minor_lock);
2905 md = idr_find(&_minor_idr, minor);
2907 if ((md == MINOR_ALLOCED ||
2908 (MINOR(disk_devt(dm_disk(md))) != minor) ||
2909 dm_deleting_md(md) ||
2910 test_bit(DMF_FREEING, &md->flags))) {
2918 spin_unlock(&_minor_lock);
2922 EXPORT_SYMBOL_GPL(dm_get_md);
2924 void *dm_get_mdptr(struct mapped_device *md)
2926 return md->interface_ptr;
2929 void dm_set_mdptr(struct mapped_device *md, void *ptr)
2931 md->interface_ptr = ptr;
2934 void dm_get(struct mapped_device *md)
2936 atomic_inc(&md->holders);
2937 BUG_ON(test_bit(DMF_FREEING, &md->flags));
2940 int dm_hold(struct mapped_device *md)
2942 spin_lock(&_minor_lock);
2943 if (test_bit(DMF_FREEING, &md->flags)) {
2944 spin_unlock(&_minor_lock);
2948 spin_unlock(&_minor_lock);
2951 EXPORT_SYMBOL_GPL(dm_hold);
2953 const char *dm_device_name(struct mapped_device *md)
2957 EXPORT_SYMBOL_GPL(dm_device_name);
2959 static void __dm_destroy(struct mapped_device *md, bool wait)
2961 struct dm_table *map;
2966 map = dm_get_live_table(md, &srcu_idx);
2968 spin_lock(&_minor_lock);
2969 idr_replace(&_minor_idr, MINOR_ALLOCED, MINOR(disk_devt(dm_disk(md))));
2970 set_bit(DMF_FREEING, &md->flags);
2971 spin_unlock(&_minor_lock);
2973 if (dm_request_based(md) && md->kworker_task)
2974 flush_kthread_worker(&md->kworker);
2977 * Take suspend_lock so that presuspend and postsuspend methods
2978 * do not race with internal suspend.
2980 mutex_lock(&md->suspend_lock);
2981 if (!dm_suspended_md(md)) {
2982 dm_table_presuspend_targets(map);
2983 dm_table_postsuspend_targets(map);
2985 mutex_unlock(&md->suspend_lock);
2987 /* dm_put_live_table must be before msleep, otherwise deadlock is possible */
2988 dm_put_live_table(md, srcu_idx);
2991 * Rare, but there may be I/O requests still going to complete,
2992 * for example. Wait for all references to disappear.
2993 * No one should increment the reference count of the mapped_device,
2994 * after the mapped_device state becomes DMF_FREEING.
2997 while (atomic_read(&md->holders))
2999 else if (atomic_read(&md->holders))
3000 DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)",
3001 dm_device_name(md), atomic_read(&md->holders));
3004 dm_table_destroy(__unbind(md));
3008 void dm_destroy(struct mapped_device *md)
3010 __dm_destroy(md, true);
3013 void dm_destroy_immediate(struct mapped_device *md)
3015 __dm_destroy(md, false);
3018 void dm_put(struct mapped_device *md)
3020 atomic_dec(&md->holders);
3022 EXPORT_SYMBOL_GPL(dm_put);
3024 static int dm_wait_for_completion(struct mapped_device *md, int interruptible)
3027 DECLARE_WAITQUEUE(wait, current);
3029 add_wait_queue(&md->wait, &wait);
3032 set_current_state(interruptible);
3034 if (!md_in_flight(md))
3037 if (interruptible == TASK_INTERRUPTIBLE &&
3038 signal_pending(current)) {
3045 set_current_state(TASK_RUNNING);
3047 remove_wait_queue(&md->wait, &wait);
3053 * Process the deferred bios
3055 static void dm_wq_work(struct work_struct *work)
3057 struct mapped_device *md = container_of(work, struct mapped_device,
3061 struct dm_table *map;
3063 map = dm_get_live_table(md, &srcu_idx);
3065 while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
3066 spin_lock_irq(&md->deferred_lock);
3067 c = bio_list_pop(&md->deferred);
3068 spin_unlock_irq(&md->deferred_lock);
3073 if (dm_request_based(md))
3074 generic_make_request(c);
3076 __split_and_process_bio(md, map, c);
3079 dm_put_live_table(md, srcu_idx);
3082 static void dm_queue_flush(struct mapped_device *md)
3084 clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
3085 smp_mb__after_atomic();
3086 queue_work(md->wq, &md->work);
3090 * Swap in a new table, returning the old one for the caller to destroy.
3092 struct dm_table *dm_swap_table(struct mapped_device *md, struct dm_table *table)
3094 struct dm_table *live_map = NULL, *map = ERR_PTR(-EINVAL);
3095 struct queue_limits limits;
3098 mutex_lock(&md->suspend_lock);
3100 /* device must be suspended */
3101 if (!dm_suspended_md(md))
3105 * If the new table has no data devices, retain the existing limits.
3106 * This helps multipath with queue_if_no_path if all paths disappear,
3107 * then new I/O is queued based on these limits, and then some paths
3110 if (dm_table_has_no_data_devices(table)) {
3111 live_map = dm_get_live_table_fast(md);
3113 limits = md->queue->limits;
3114 dm_put_live_table_fast(md);
3118 r = dm_calculate_queue_limits(table, &limits);
3125 map = __bind(md, table, &limits);
3128 mutex_unlock(&md->suspend_lock);
3133 * Functions to lock and unlock any filesystem running on the
3136 static int lock_fs(struct mapped_device *md)
3140 WARN_ON(md->frozen_sb);
3142 md->frozen_sb = freeze_bdev(md->bdev);
3143 if (IS_ERR(md->frozen_sb)) {
3144 r = PTR_ERR(md->frozen_sb);
3145 md->frozen_sb = NULL;
3149 set_bit(DMF_FROZEN, &md->flags);
3154 static void unlock_fs(struct mapped_device *md)
3156 if (!test_bit(DMF_FROZEN, &md->flags))
3159 thaw_bdev(md->bdev, md->frozen_sb);
3160 md->frozen_sb = NULL;
3161 clear_bit(DMF_FROZEN, &md->flags);
3165 * If __dm_suspend returns 0, the device is completely quiescent
3166 * now. There is no request-processing activity. All new requests
3167 * are being added to md->deferred list.
3169 * Caller must hold md->suspend_lock
3171 static int __dm_suspend(struct mapped_device *md, struct dm_table *map,
3172 unsigned suspend_flags, int interruptible)
3174 bool do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG;
3175 bool noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG;
3179 * DMF_NOFLUSH_SUSPENDING must be set before presuspend.
3180 * This flag is cleared before dm_suspend returns.
3183 set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
3186 * This gets reverted if there's an error later and the targets
3187 * provide the .presuspend_undo hook.
3189 dm_table_presuspend_targets(map);
3192 * Flush I/O to the device.
3193 * Any I/O submitted after lock_fs() may not be flushed.
3194 * noflush takes precedence over do_lockfs.
3195 * (lock_fs() flushes I/Os and waits for them to complete.)
3197 if (!noflush && do_lockfs) {
3200 dm_table_presuspend_undo_targets(map);
3206 * Here we must make sure that no processes are submitting requests
3207 * to target drivers i.e. no one may be executing
3208 * __split_and_process_bio. This is called from dm_request and
3211 * To get all processes out of __split_and_process_bio in dm_request,
3212 * we take the write lock. To prevent any process from reentering
3213 * __split_and_process_bio from dm_request and quiesce the thread
3214 * (dm_wq_work), we set BMF_BLOCK_IO_FOR_SUSPEND and call
3215 * flush_workqueue(md->wq).
3217 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
3219 synchronize_srcu(&md->io_barrier);
3222 * Stop md->queue before flushing md->wq in case request-based
3223 * dm defers requests to md->wq from md->queue.
3225 if (dm_request_based(md)) {
3226 stop_queue(md->queue);
3227 if (md->kworker_task)
3228 flush_kthread_worker(&md->kworker);
3231 flush_workqueue(md->wq);
3234 * At this point no more requests are entering target request routines.
3235 * We call dm_wait_for_completion to wait for all existing requests
3238 r = dm_wait_for_completion(md, interruptible);
3241 clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
3243 synchronize_srcu(&md->io_barrier);
3245 /* were we interrupted ? */
3249 if (dm_request_based(md))
3250 start_queue(md->queue);
3253 dm_table_presuspend_undo_targets(map);
3254 /* pushback list is already flushed, so skip flush */
3261 * We need to be able to change a mapping table under a mounted
3262 * filesystem. For example we might want to move some data in
3263 * the background. Before the table can be swapped with
3264 * dm_bind_table, dm_suspend must be called to flush any in
3265 * flight bios and ensure that any further io gets deferred.
3268 * Suspend mechanism in request-based dm.
3270 * 1. Flush all I/Os by lock_fs() if needed.
3271 * 2. Stop dispatching any I/O by stopping the request_queue.
3272 * 3. Wait for all in-flight I/Os to be completed or requeued.
3274 * To abort suspend, start the request_queue.
3276 int dm_suspend(struct mapped_device *md, unsigned suspend_flags)
3278 struct dm_table *map = NULL;
3282 mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING);
3284 if (dm_suspended_md(md)) {
3289 if (dm_suspended_internally_md(md)) {
3290 /* already internally suspended, wait for internal resume */
3291 mutex_unlock(&md->suspend_lock);
3292 r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE);
3298 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
3300 r = __dm_suspend(md, map, suspend_flags, TASK_INTERRUPTIBLE);
3304 set_bit(DMF_SUSPENDED, &md->flags);
3306 dm_table_postsuspend_targets(map);
3309 mutex_unlock(&md->suspend_lock);
3313 static int __dm_resume(struct mapped_device *md, struct dm_table *map)
3316 int r = dm_table_resume_targets(map);
3324 * Flushing deferred I/Os must be done after targets are resumed
3325 * so that mapping of targets can work correctly.
3326 * Request-based dm is queueing the deferred I/Os in its request_queue.
3328 if (dm_request_based(md))
3329 start_queue(md->queue);
3336 int dm_resume(struct mapped_device *md)
3339 struct dm_table *map = NULL;
3342 mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING);
3344 if (!dm_suspended_md(md))
3347 if (dm_suspended_internally_md(md)) {
3348 /* already internally suspended, wait for internal resume */
3349 mutex_unlock(&md->suspend_lock);
3350 r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE);
3356 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
3357 if (!map || !dm_table_get_size(map))
3360 r = __dm_resume(md, map);
3364 clear_bit(DMF_SUSPENDED, &md->flags);
3368 mutex_unlock(&md->suspend_lock);
3374 * Internal suspend/resume works like userspace-driven suspend. It waits
3375 * until all bios finish and prevents issuing new bios to the target drivers.
3376 * It may be used only from the kernel.
3379 static void __dm_internal_suspend(struct mapped_device *md, unsigned suspend_flags)
3381 struct dm_table *map = NULL;
3383 if (md->internal_suspend_count++)
3384 return; /* nested internal suspend */
3386 if (dm_suspended_md(md)) {
3387 set_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
3388 return; /* nest suspend */
3391 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
3394 * Using TASK_UNINTERRUPTIBLE because only NOFLUSH internal suspend is
3395 * supported. Properly supporting a TASK_INTERRUPTIBLE internal suspend
3396 * would require changing .presuspend to return an error -- avoid this
3397 * until there is a need for more elaborate variants of internal suspend.
3399 (void) __dm_suspend(md, map, suspend_flags, TASK_UNINTERRUPTIBLE);
3401 set_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
3403 dm_table_postsuspend_targets(map);
3406 static void __dm_internal_resume(struct mapped_device *md)
3408 BUG_ON(!md->internal_suspend_count);
3410 if (--md->internal_suspend_count)
3411 return; /* resume from nested internal suspend */
3413 if (dm_suspended_md(md))
3414 goto done; /* resume from nested suspend */
3417 * NOTE: existing callers don't need to call dm_table_resume_targets
3418 * (which may fail -- so best to avoid it for now by passing NULL map)
3420 (void) __dm_resume(md, NULL);
3423 clear_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
3424 smp_mb__after_atomic();
3425 wake_up_bit(&md->flags, DMF_SUSPENDED_INTERNALLY);
3428 void dm_internal_suspend_noflush(struct mapped_device *md)
3430 mutex_lock(&md->suspend_lock);
3431 __dm_internal_suspend(md, DM_SUSPEND_NOFLUSH_FLAG);
3432 mutex_unlock(&md->suspend_lock);
3434 EXPORT_SYMBOL_GPL(dm_internal_suspend_noflush);
3436 void dm_internal_resume(struct mapped_device *md)
3438 mutex_lock(&md->suspend_lock);
3439 __dm_internal_resume(md);
3440 mutex_unlock(&md->suspend_lock);
3442 EXPORT_SYMBOL_GPL(dm_internal_resume);
3445 * Fast variants of internal suspend/resume hold md->suspend_lock,
3446 * which prevents interaction with userspace-driven suspend.
3449 void dm_internal_suspend_fast(struct mapped_device *md)
3451 mutex_lock(&md->suspend_lock);
3452 if (dm_suspended_md(md) || dm_suspended_internally_md(md))
3455 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
3456 synchronize_srcu(&md->io_barrier);
3457 flush_workqueue(md->wq);
3458 dm_wait_for_completion(md, TASK_UNINTERRUPTIBLE);
3460 EXPORT_SYMBOL_GPL(dm_internal_suspend_fast);
3462 void dm_internal_resume_fast(struct mapped_device *md)
3464 if (dm_suspended_md(md) || dm_suspended_internally_md(md))
3470 mutex_unlock(&md->suspend_lock);
3472 EXPORT_SYMBOL_GPL(dm_internal_resume_fast);
3474 /*-----------------------------------------------------------------
3475 * Event notification.
3476 *---------------------------------------------------------------*/
3477 int dm_kobject_uevent(struct mapped_device *md, enum kobject_action action,
3480 char udev_cookie[DM_COOKIE_LENGTH];
3481 char *envp[] = { udev_cookie, NULL };
3484 return kobject_uevent(&disk_to_dev(md->disk)->kobj, action);
3486 snprintf(udev_cookie, DM_COOKIE_LENGTH, "%s=%u",
3487 DM_COOKIE_ENV_VAR_NAME, cookie);
3488 return kobject_uevent_env(&disk_to_dev(md->disk)->kobj,
3493 uint32_t dm_next_uevent_seq(struct mapped_device *md)
3495 return atomic_add_return(1, &md->uevent_seq);
3498 uint32_t dm_get_event_nr(struct mapped_device *md)
3500 return atomic_read(&md->event_nr);
3503 int dm_wait_event(struct mapped_device *md, int event_nr)
3505 return wait_event_interruptible(md->eventq,
3506 (event_nr != atomic_read(&md->event_nr)));
3509 void dm_uevent_add(struct mapped_device *md, struct list_head *elist)
3511 unsigned long flags;
3513 spin_lock_irqsave(&md->uevent_lock, flags);
3514 list_add(elist, &md->uevent_list);
3515 spin_unlock_irqrestore(&md->uevent_lock, flags);
3519 * The gendisk is only valid as long as you have a reference
3522 struct gendisk *dm_disk(struct mapped_device *md)
3526 EXPORT_SYMBOL_GPL(dm_disk);
3528 struct kobject *dm_kobject(struct mapped_device *md)
3530 return &md->kobj_holder.kobj;
3533 struct mapped_device *dm_get_from_kobject(struct kobject *kobj)
3535 struct mapped_device *md;
3537 md = container_of(kobj, struct mapped_device, kobj_holder.kobj);
3539 if (test_bit(DMF_FREEING, &md->flags) ||
3547 int dm_suspended_md(struct mapped_device *md)
3549 return test_bit(DMF_SUSPENDED, &md->flags);
3552 int dm_suspended_internally_md(struct mapped_device *md)
3554 return test_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
3557 int dm_test_deferred_remove_flag(struct mapped_device *md)
3559 return test_bit(DMF_DEFERRED_REMOVE, &md->flags);
3562 int dm_suspended(struct dm_target *ti)
3564 return dm_suspended_md(dm_table_get_md(ti->table));
3566 EXPORT_SYMBOL_GPL(dm_suspended);
3568 int dm_noflush_suspending(struct dm_target *ti)
3570 return __noflush_suspending(dm_table_get_md(ti->table));
3572 EXPORT_SYMBOL_GPL(dm_noflush_suspending);
3574 struct dm_md_mempools *dm_alloc_md_mempools(struct mapped_device *md, unsigned type,
3575 unsigned integrity, unsigned per_bio_data_size)
3577 struct dm_md_mempools *pools = kzalloc(sizeof(*pools), GFP_KERNEL);
3578 struct kmem_cache *cachep = NULL;
3579 unsigned int pool_size = 0;
3580 unsigned int front_pad;
3585 type = filter_md_type(type, md);
3588 case DM_TYPE_BIO_BASED:
3590 pool_size = dm_get_reserved_bio_based_ios();
3591 front_pad = roundup(per_bio_data_size, __alignof__(struct dm_target_io)) + offsetof(struct dm_target_io, clone);
3593 case DM_TYPE_REQUEST_BASED:
3594 cachep = _rq_tio_cache;
3595 pool_size = dm_get_reserved_rq_based_ios();
3596 pools->rq_pool = mempool_create_slab_pool(pool_size, _rq_cache);
3597 if (!pools->rq_pool)
3599 /* fall through to setup remaining rq-based pools */
3600 case DM_TYPE_MQ_REQUEST_BASED:
3602 pool_size = dm_get_reserved_rq_based_ios();
3603 front_pad = offsetof(struct dm_rq_clone_bio_info, clone);
3604 /* per_bio_data_size is not used. See __bind_mempools(). */
3605 WARN_ON(per_bio_data_size != 0);
3612 pools->io_pool = mempool_create_slab_pool(pool_size, cachep);
3613 if (!pools->io_pool)
3617 pools->bs = bioset_create_nobvec(pool_size, front_pad);
3621 if (integrity && bioset_integrity_create(pools->bs, pool_size))
3627 dm_free_md_mempools(pools);
3632 void dm_free_md_mempools(struct dm_md_mempools *pools)
3638 mempool_destroy(pools->io_pool);
3641 mempool_destroy(pools->rq_pool);
3644 bioset_free(pools->bs);
3649 static const struct block_device_operations dm_blk_dops = {
3650 .open = dm_blk_open,
3651 .release = dm_blk_close,
3652 .ioctl = dm_blk_ioctl,
3653 .getgeo = dm_blk_getgeo,
3654 .owner = THIS_MODULE
3660 module_init(dm_init);
3661 module_exit(dm_exit);
3663 module_param(major, uint, 0);
3664 MODULE_PARM_DESC(major, "The major number of the device mapper");
3666 module_param(reserved_bio_based_ios, uint, S_IRUGO | S_IWUSR);
3667 MODULE_PARM_DESC(reserved_bio_based_ios, "Reserved IOs in bio-based mempools");
3669 module_param(reserved_rq_based_ios, uint, S_IRUGO | S_IWUSR);
3670 MODULE_PARM_DESC(reserved_rq_based_ios, "Reserved IOs in request-based mempools");
3672 module_param(use_blk_mq, bool, S_IRUGO | S_IWUSR);
3673 MODULE_PARM_DESC(use_blk_mq, "Use block multiqueue for request-based DM devices");
3675 MODULE_DESCRIPTION(DM_NAME " driver");
3676 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
3677 MODULE_LICENSE("GPL");