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>
29 #include <trace/events/block.h>
31 #define DM_MSG_PREFIX "core"
35 * ratelimit state to be used in DMXXX_LIMIT().
37 DEFINE_RATELIMIT_STATE(dm_ratelimit_state,
38 DEFAULT_RATELIMIT_INTERVAL,
39 DEFAULT_RATELIMIT_BURST);
40 EXPORT_SYMBOL(dm_ratelimit_state);
44 * Cookies are numeric values sent with CHANGE and REMOVE
45 * uevents while resuming, removing or renaming the device.
47 #define DM_COOKIE_ENV_VAR_NAME "DM_COOKIE"
48 #define DM_COOKIE_LENGTH 24
50 static const char *_name = DM_NAME;
52 static unsigned int major = 0;
53 static unsigned int _major = 0;
55 static DEFINE_IDR(_minor_idr);
57 static DEFINE_SPINLOCK(_minor_lock);
59 static void do_deferred_remove(struct work_struct *w);
61 static DECLARE_WORK(deferred_remove_work, do_deferred_remove);
63 static struct workqueue_struct *deferred_remove_workqueue;
67 * One of these is allocated per bio.
70 struct mapped_device *md;
74 unsigned long start_time;
75 spinlock_t endio_lock;
76 struct dm_stats_aux stats_aux;
80 * For request-based dm.
81 * One of these is allocated per request.
83 struct dm_rq_target_io {
84 struct mapped_device *md;
86 struct request *orig, *clone;
87 struct kthread_work work;
90 struct dm_stats_aux stats_aux;
91 unsigned long duration_jiffies;
96 * For request-based dm - the bio clones we allocate are embedded in these
99 * We allocate these with bio_alloc_bioset, using the front_pad parameter when
100 * the bioset is created - this means the bio has to come at the end of the
103 struct dm_rq_clone_bio_info {
105 struct dm_rq_target_io *tio;
109 #define MINOR_ALLOCED ((void *)-1)
112 * Bits for the md->flags field.
114 #define DMF_BLOCK_IO_FOR_SUSPEND 0
115 #define DMF_SUSPENDED 1
117 #define DMF_FREEING 3
118 #define DMF_DELETING 4
119 #define DMF_NOFLUSH_SUSPENDING 5
120 #define DMF_DEFERRED_REMOVE 6
121 #define DMF_SUSPENDED_INTERNALLY 7
124 * A dummy definition to make RCU happy.
125 * struct dm_table should never be dereferenced in this file.
132 * Work processed by per-device workqueue.
134 struct mapped_device {
135 struct srcu_struct io_barrier;
136 struct mutex suspend_lock;
141 * The current mapping.
142 * Use dm_get_live_table{_fast} or take suspend_lock for
145 struct dm_table __rcu *map;
147 struct list_head table_devices;
148 struct mutex table_devices_lock;
152 struct request_queue *queue;
154 /* Protect queue and type against concurrent access. */
155 struct mutex type_lock;
157 struct dm_target *immutable_target;
158 struct target_type *immutable_target_type;
160 struct gendisk *disk;
166 * A list of ios that arrived while we were suspended.
169 wait_queue_head_t wait;
170 struct work_struct work;
171 struct bio_list deferred;
172 spinlock_t deferred_lock;
175 * Processing queue (flush)
177 struct workqueue_struct *wq;
180 * io objects are allocated from here.
191 wait_queue_head_t eventq;
193 struct list_head uevent_list;
194 spinlock_t uevent_lock; /* Protect access to uevent_list */
197 * freeze/thaw support require holding onto a super block
199 struct super_block *frozen_sb;
200 struct block_device *bdev;
202 /* forced geometry settings */
203 struct hd_geometry geometry;
205 /* kobject and completion */
206 struct dm_kobject_holder kobj_holder;
208 /* zero-length flush that will be cloned and submitted to targets */
209 struct bio flush_bio;
211 /* the number of internal suspends */
212 unsigned internal_suspend_count;
214 struct dm_stats stats;
216 struct kthread_worker kworker;
217 struct task_struct *kworker_task;
219 /* for request-based merge heuristic in dm_request_fn() */
220 unsigned seq_rq_merge_deadline_usecs;
222 sector_t last_rq_pos;
223 ktime_t last_rq_start_time;
225 /* for blk-mq request-based DM support */
226 struct blk_mq_tag_set *tag_set;
230 #ifdef CONFIG_DM_MQ_DEFAULT
231 static bool use_blk_mq = true;
233 static bool use_blk_mq = false;
236 #define DM_MQ_NR_HW_QUEUES 1
237 #define DM_MQ_QUEUE_DEPTH 2048
239 static unsigned dm_mq_nr_hw_queues = DM_MQ_NR_HW_QUEUES;
240 static unsigned dm_mq_queue_depth = DM_MQ_QUEUE_DEPTH;
242 bool dm_use_blk_mq(struct mapped_device *md)
244 return md->use_blk_mq;
248 * For mempools pre-allocation at the table loading time.
250 struct dm_md_mempools {
256 struct table_device {
257 struct list_head list;
259 struct dm_dev dm_dev;
262 #define RESERVED_BIO_BASED_IOS 16
263 #define RESERVED_REQUEST_BASED_IOS 256
264 #define RESERVED_MAX_IOS 1024
265 static struct kmem_cache *_io_cache;
266 static struct kmem_cache *_rq_tio_cache;
267 static struct kmem_cache *_rq_cache;
270 * Bio-based DM's mempools' reserved IOs set by the user.
272 static unsigned reserved_bio_based_ios = RESERVED_BIO_BASED_IOS;
275 * Request-based DM's mempools' reserved IOs set by the user.
277 static unsigned reserved_rq_based_ios = RESERVED_REQUEST_BASED_IOS;
279 static unsigned __dm_get_module_param(unsigned *module_param,
280 unsigned def, unsigned max)
282 unsigned param = ACCESS_ONCE(*module_param);
283 unsigned modified_param = 0;
286 modified_param = def;
287 else if (param > max)
288 modified_param = max;
290 if (modified_param) {
291 (void)cmpxchg(module_param, param, modified_param);
292 param = modified_param;
298 unsigned dm_get_reserved_bio_based_ios(void)
300 return __dm_get_module_param(&reserved_bio_based_ios,
301 RESERVED_BIO_BASED_IOS, RESERVED_MAX_IOS);
303 EXPORT_SYMBOL_GPL(dm_get_reserved_bio_based_ios);
305 unsigned dm_get_reserved_rq_based_ios(void)
307 return __dm_get_module_param(&reserved_rq_based_ios,
308 RESERVED_REQUEST_BASED_IOS, RESERVED_MAX_IOS);
310 EXPORT_SYMBOL_GPL(dm_get_reserved_rq_based_ios);
312 static unsigned dm_get_blk_mq_nr_hw_queues(void)
314 return __dm_get_module_param(&dm_mq_nr_hw_queues, 1, 32);
317 static unsigned dm_get_blk_mq_queue_depth(void)
319 return __dm_get_module_param(&dm_mq_queue_depth,
320 DM_MQ_QUEUE_DEPTH, BLK_MQ_MAX_DEPTH);
323 static int __init local_init(void)
327 /* allocate a slab for the dm_ios */
328 _io_cache = KMEM_CACHE(dm_io, 0);
332 _rq_tio_cache = KMEM_CACHE(dm_rq_target_io, 0);
334 goto out_free_io_cache;
336 _rq_cache = kmem_cache_create("dm_clone_request", sizeof(struct request),
337 __alignof__(struct request), 0, NULL);
339 goto out_free_rq_tio_cache;
341 r = dm_uevent_init();
343 goto out_free_rq_cache;
345 deferred_remove_workqueue = alloc_workqueue("kdmremove", WQ_UNBOUND, 1);
346 if (!deferred_remove_workqueue) {
348 goto out_uevent_exit;
352 r = register_blkdev(_major, _name);
354 goto out_free_workqueue;
362 destroy_workqueue(deferred_remove_workqueue);
366 kmem_cache_destroy(_rq_cache);
367 out_free_rq_tio_cache:
368 kmem_cache_destroy(_rq_tio_cache);
370 kmem_cache_destroy(_io_cache);
375 static void local_exit(void)
377 flush_scheduled_work();
378 destroy_workqueue(deferred_remove_workqueue);
380 kmem_cache_destroy(_rq_cache);
381 kmem_cache_destroy(_rq_tio_cache);
382 kmem_cache_destroy(_io_cache);
383 unregister_blkdev(_major, _name);
388 DMINFO("cleaned up");
391 static int (*_inits[])(void) __initdata = {
402 static void (*_exits[])(void) = {
413 static int __init dm_init(void)
415 const int count = ARRAY_SIZE(_inits);
419 for (i = 0; i < count; i++) {
434 static void __exit dm_exit(void)
436 int i = ARRAY_SIZE(_exits);
442 * Should be empty by this point.
444 idr_destroy(&_minor_idr);
448 * Block device functions
450 int dm_deleting_md(struct mapped_device *md)
452 return test_bit(DMF_DELETING, &md->flags);
455 static int dm_blk_open(struct block_device *bdev, fmode_t mode)
457 struct mapped_device *md;
459 spin_lock(&_minor_lock);
461 md = bdev->bd_disk->private_data;
465 if (test_bit(DMF_FREEING, &md->flags) ||
466 dm_deleting_md(md)) {
472 atomic_inc(&md->open_count);
474 spin_unlock(&_minor_lock);
476 return md ? 0 : -ENXIO;
479 static void dm_blk_close(struct gendisk *disk, fmode_t mode)
481 struct mapped_device *md;
483 spin_lock(&_minor_lock);
485 md = disk->private_data;
489 if (atomic_dec_and_test(&md->open_count) &&
490 (test_bit(DMF_DEFERRED_REMOVE, &md->flags)))
491 queue_work(deferred_remove_workqueue, &deferred_remove_work);
495 spin_unlock(&_minor_lock);
498 int dm_open_count(struct mapped_device *md)
500 return atomic_read(&md->open_count);
504 * Guarantees nothing is using the device before it's deleted.
506 int dm_lock_for_deletion(struct mapped_device *md, bool mark_deferred, bool only_deferred)
510 spin_lock(&_minor_lock);
512 if (dm_open_count(md)) {
515 set_bit(DMF_DEFERRED_REMOVE, &md->flags);
516 } else if (only_deferred && !test_bit(DMF_DEFERRED_REMOVE, &md->flags))
519 set_bit(DMF_DELETING, &md->flags);
521 spin_unlock(&_minor_lock);
526 int dm_cancel_deferred_remove(struct mapped_device *md)
530 spin_lock(&_minor_lock);
532 if (test_bit(DMF_DELETING, &md->flags))
535 clear_bit(DMF_DEFERRED_REMOVE, &md->flags);
537 spin_unlock(&_minor_lock);
542 static void do_deferred_remove(struct work_struct *w)
544 dm_deferred_remove();
547 sector_t dm_get_size(struct mapped_device *md)
549 return get_capacity(md->disk);
552 struct request_queue *dm_get_md_queue(struct mapped_device *md)
557 struct dm_stats *dm_get_stats(struct mapped_device *md)
562 static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo)
564 struct mapped_device *md = bdev->bd_disk->private_data;
566 return dm_get_geometry(md, geo);
569 static int dm_grab_bdev_for_ioctl(struct mapped_device *md,
570 struct block_device **bdev,
573 struct dm_target *tgt;
574 struct dm_table *map;
579 map = dm_get_live_table(md, &srcu_idx);
580 if (!map || !dm_table_get_size(map))
583 /* We only support devices that have a single target */
584 if (dm_table_get_num_targets(map) != 1)
587 tgt = dm_table_get_target(map, 0);
588 if (!tgt->type->prepare_ioctl)
591 if (dm_suspended_md(md)) {
596 r = tgt->type->prepare_ioctl(tgt, bdev, mode);
601 dm_put_live_table(md, srcu_idx);
605 dm_put_live_table(md, srcu_idx);
606 if (r == -ENOTCONN && !fatal_signal_pending(current)) {
613 static int dm_blk_ioctl(struct block_device *bdev, fmode_t mode,
614 unsigned int cmd, unsigned long arg)
616 struct mapped_device *md = bdev->bd_disk->private_data;
619 r = dm_grab_bdev_for_ioctl(md, &bdev, &mode);
625 * Target determined this ioctl is being issued against
626 * a logical partition of the parent bdev; so extra
627 * validation is needed.
629 r = scsi_verify_blk_ioctl(NULL, cmd);
634 r = __blkdev_driver_ioctl(bdev, mode, cmd, arg);
640 static struct dm_io *alloc_io(struct mapped_device *md)
642 return mempool_alloc(md->io_pool, GFP_NOIO);
645 static void free_io(struct mapped_device *md, struct dm_io *io)
647 mempool_free(io, md->io_pool);
650 static void free_tio(struct mapped_device *md, struct dm_target_io *tio)
652 bio_put(&tio->clone);
655 static struct dm_rq_target_io *alloc_rq_tio(struct mapped_device *md,
658 return mempool_alloc(md->io_pool, gfp_mask);
661 static void free_rq_tio(struct dm_rq_target_io *tio)
663 mempool_free(tio, tio->md->io_pool);
666 static struct request *alloc_clone_request(struct mapped_device *md,
669 return mempool_alloc(md->rq_pool, gfp_mask);
672 static void free_clone_request(struct mapped_device *md, struct request *rq)
674 mempool_free(rq, md->rq_pool);
677 static int md_in_flight(struct mapped_device *md)
679 return atomic_read(&md->pending[READ]) +
680 atomic_read(&md->pending[WRITE]);
683 static void start_io_acct(struct dm_io *io)
685 struct mapped_device *md = io->md;
686 struct bio *bio = io->bio;
688 int rw = bio_data_dir(bio);
690 io->start_time = jiffies;
692 cpu = part_stat_lock();
693 part_round_stats(cpu, &dm_disk(md)->part0);
695 atomic_set(&dm_disk(md)->part0.in_flight[rw],
696 atomic_inc_return(&md->pending[rw]));
698 if (unlikely(dm_stats_used(&md->stats)))
699 dm_stats_account_io(&md->stats, bio->bi_rw, bio->bi_iter.bi_sector,
700 bio_sectors(bio), false, 0, &io->stats_aux);
703 static void end_io_acct(struct dm_io *io)
705 struct mapped_device *md = io->md;
706 struct bio *bio = io->bio;
707 unsigned long duration = jiffies - io->start_time;
709 int rw = bio_data_dir(bio);
711 generic_end_io_acct(rw, &dm_disk(md)->part0, io->start_time);
713 if (unlikely(dm_stats_used(&md->stats)))
714 dm_stats_account_io(&md->stats, bio->bi_rw, bio->bi_iter.bi_sector,
715 bio_sectors(bio), true, duration, &io->stats_aux);
718 * After this is decremented the bio must not be touched if it is
721 pending = atomic_dec_return(&md->pending[rw]);
722 atomic_set(&dm_disk(md)->part0.in_flight[rw], pending);
723 pending += atomic_read(&md->pending[rw^0x1]);
725 /* nudge anyone waiting on suspend queue */
731 * Add the bio to the list of deferred io.
733 static void queue_io(struct mapped_device *md, struct bio *bio)
737 spin_lock_irqsave(&md->deferred_lock, flags);
738 bio_list_add(&md->deferred, bio);
739 spin_unlock_irqrestore(&md->deferred_lock, flags);
740 queue_work(md->wq, &md->work);
744 * Everyone (including functions in this file), should use this
745 * function to access the md->map field, and make sure they call
746 * dm_put_live_table() when finished.
748 struct dm_table *dm_get_live_table(struct mapped_device *md, int *srcu_idx) __acquires(md->io_barrier)
750 *srcu_idx = srcu_read_lock(&md->io_barrier);
752 return srcu_dereference(md->map, &md->io_barrier);
755 void dm_put_live_table(struct mapped_device *md, int srcu_idx) __releases(md->io_barrier)
757 srcu_read_unlock(&md->io_barrier, srcu_idx);
760 void dm_sync_table(struct mapped_device *md)
762 synchronize_srcu(&md->io_barrier);
763 synchronize_rcu_expedited();
767 * A fast alternative to dm_get_live_table/dm_put_live_table.
768 * The caller must not block between these two functions.
770 static struct dm_table *dm_get_live_table_fast(struct mapped_device *md) __acquires(RCU)
773 return rcu_dereference(md->map);
776 static void dm_put_live_table_fast(struct mapped_device *md) __releases(RCU)
782 * Open a table device so we can use it as a map destination.
784 static int open_table_device(struct table_device *td, dev_t dev,
785 struct mapped_device *md)
787 static char *_claim_ptr = "I belong to device-mapper";
788 struct block_device *bdev;
792 BUG_ON(td->dm_dev.bdev);
794 bdev = blkdev_get_by_dev(dev, td->dm_dev.mode | FMODE_EXCL, _claim_ptr);
796 return PTR_ERR(bdev);
798 r = bd_link_disk_holder(bdev, dm_disk(md));
800 blkdev_put(bdev, td->dm_dev.mode | FMODE_EXCL);
804 td->dm_dev.bdev = bdev;
809 * Close a table device that we've been using.
811 static void close_table_device(struct table_device *td, struct mapped_device *md)
813 if (!td->dm_dev.bdev)
816 bd_unlink_disk_holder(td->dm_dev.bdev, dm_disk(md));
817 blkdev_put(td->dm_dev.bdev, td->dm_dev.mode | FMODE_EXCL);
818 td->dm_dev.bdev = NULL;
821 static struct table_device *find_table_device(struct list_head *l, dev_t dev,
823 struct table_device *td;
825 list_for_each_entry(td, l, list)
826 if (td->dm_dev.bdev->bd_dev == dev && td->dm_dev.mode == mode)
832 int dm_get_table_device(struct mapped_device *md, dev_t dev, fmode_t mode,
833 struct dm_dev **result) {
835 struct table_device *td;
837 mutex_lock(&md->table_devices_lock);
838 td = find_table_device(&md->table_devices, dev, mode);
840 td = kmalloc(sizeof(*td), GFP_KERNEL);
842 mutex_unlock(&md->table_devices_lock);
846 td->dm_dev.mode = mode;
847 td->dm_dev.bdev = NULL;
849 if ((r = open_table_device(td, dev, md))) {
850 mutex_unlock(&md->table_devices_lock);
855 format_dev_t(td->dm_dev.name, dev);
857 atomic_set(&td->count, 0);
858 list_add(&td->list, &md->table_devices);
860 atomic_inc(&td->count);
861 mutex_unlock(&md->table_devices_lock);
863 *result = &td->dm_dev;
866 EXPORT_SYMBOL_GPL(dm_get_table_device);
868 void dm_put_table_device(struct mapped_device *md, struct dm_dev *d)
870 struct table_device *td = container_of(d, struct table_device, dm_dev);
872 mutex_lock(&md->table_devices_lock);
873 if (atomic_dec_and_test(&td->count)) {
874 close_table_device(td, md);
878 mutex_unlock(&md->table_devices_lock);
880 EXPORT_SYMBOL(dm_put_table_device);
882 static void free_table_devices(struct list_head *devices)
884 struct list_head *tmp, *next;
886 list_for_each_safe(tmp, next, devices) {
887 struct table_device *td = list_entry(tmp, struct table_device, list);
889 DMWARN("dm_destroy: %s still exists with %d references",
890 td->dm_dev.name, atomic_read(&td->count));
896 * Get the geometry associated with a dm device
898 int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo)
906 * Set the geometry of a device.
908 int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo)
910 sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors;
912 if (geo->start > sz) {
913 DMWARN("Start sector is beyond the geometry limits.");
922 /*-----------------------------------------------------------------
924 * A more elegant soln is in the works that uses the queue
925 * merge fn, unfortunately there are a couple of changes to
926 * the block layer that I want to make for this. So in the
927 * interests of getting something for people to use I give
928 * you this clearly demarcated crap.
929 *---------------------------------------------------------------*/
931 static int __noflush_suspending(struct mapped_device *md)
933 return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
937 * Decrements the number of outstanding ios that a bio has been
938 * cloned into, completing the original io if necc.
940 static void dec_pending(struct dm_io *io, int error)
945 struct mapped_device *md = io->md;
947 /* Push-back supersedes any I/O errors */
948 if (unlikely(error)) {
949 spin_lock_irqsave(&io->endio_lock, flags);
950 if (!(io->error > 0 && __noflush_suspending(md)))
952 spin_unlock_irqrestore(&io->endio_lock, flags);
955 if (atomic_dec_and_test(&io->io_count)) {
956 if (io->error == DM_ENDIO_REQUEUE) {
958 * Target requested pushing back the I/O.
960 spin_lock_irqsave(&md->deferred_lock, flags);
961 if (__noflush_suspending(md))
962 bio_list_add_head(&md->deferred, io->bio);
964 /* noflush suspend was interrupted. */
966 spin_unlock_irqrestore(&md->deferred_lock, flags);
969 io_error = io->error;
974 if (io_error == DM_ENDIO_REQUEUE)
977 if ((bio->bi_rw & REQ_FLUSH) && bio->bi_iter.bi_size) {
979 * Preflush done for flush with data, reissue
982 bio->bi_rw &= ~REQ_FLUSH;
985 /* done with normal IO or empty flush */
986 trace_block_bio_complete(md->queue, bio, io_error);
987 bio->bi_error = io_error;
993 static void disable_write_same(struct mapped_device *md)
995 struct queue_limits *limits = dm_get_queue_limits(md);
997 /* device doesn't really support WRITE SAME, disable it */
998 limits->max_write_same_sectors = 0;
1001 static void clone_endio(struct bio *bio)
1003 int error = bio->bi_error;
1005 struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone);
1006 struct dm_io *io = tio->io;
1007 struct mapped_device *md = tio->io->md;
1008 dm_endio_fn endio = tio->ti->type->end_io;
1011 r = endio(tio->ti, bio, error);
1012 if (r < 0 || r == DM_ENDIO_REQUEUE)
1014 * error and requeue request are handled
1018 else if (r == DM_ENDIO_INCOMPLETE)
1019 /* The target will handle the io */
1022 DMWARN("unimplemented target endio return value: %d", r);
1027 if (unlikely(r == -EREMOTEIO && (bio->bi_rw & REQ_WRITE_SAME) &&
1028 !bdev_get_queue(bio->bi_bdev)->limits.max_write_same_sectors))
1029 disable_write_same(md);
1032 dec_pending(io, error);
1036 * Partial completion handling for request-based dm
1038 static void end_clone_bio(struct bio *clone)
1040 struct dm_rq_clone_bio_info *info =
1041 container_of(clone, struct dm_rq_clone_bio_info, clone);
1042 struct dm_rq_target_io *tio = info->tio;
1043 struct bio *bio = info->orig;
1044 unsigned int nr_bytes = info->orig->bi_iter.bi_size;
1045 int error = clone->bi_error;
1051 * An error has already been detected on the request.
1052 * Once error occurred, just let clone->end_io() handle
1058 * Don't notice the error to the upper layer yet.
1059 * The error handling decision is made by the target driver,
1060 * when the request is completed.
1067 * I/O for the bio successfully completed.
1068 * Notice the data completion to the upper layer.
1072 * bios are processed from the head of the list.
1073 * So the completing bio should always be rq->bio.
1074 * If it's not, something wrong is happening.
1076 if (tio->orig->bio != bio)
1077 DMERR("bio completion is going in the middle of the request");
1080 * Update the original request.
1081 * Do not use blk_end_request() here, because it may complete
1082 * the original request before the clone, and break the ordering.
1084 blk_update_request(tio->orig, 0, nr_bytes);
1087 static struct dm_rq_target_io *tio_from_request(struct request *rq)
1089 return (rq->q->mq_ops ? blk_mq_rq_to_pdu(rq) : rq->special);
1092 static void rq_end_stats(struct mapped_device *md, struct request *orig)
1094 if (unlikely(dm_stats_used(&md->stats))) {
1095 struct dm_rq_target_io *tio = tio_from_request(orig);
1096 tio->duration_jiffies = jiffies - tio->duration_jiffies;
1097 dm_stats_account_io(&md->stats, orig->cmd_flags, blk_rq_pos(orig),
1098 tio->n_sectors, true, tio->duration_jiffies,
1104 * Don't touch any member of the md after calling this function because
1105 * the md may be freed in dm_put() at the end of this function.
1106 * Or do dm_get() before calling this function and dm_put() later.
1108 static void rq_completed(struct mapped_device *md, int rw, bool run_queue)
1110 atomic_dec(&md->pending[rw]);
1112 /* nudge anyone waiting on suspend queue */
1113 if (!md_in_flight(md))
1117 * Run this off this callpath, as drivers could invoke end_io while
1118 * inside their request_fn (and holding the queue lock). Calling
1119 * back into ->request_fn() could deadlock attempting to grab the
1122 if (!md->queue->mq_ops && run_queue)
1123 blk_run_queue_async(md->queue);
1126 * dm_put() must be at the end of this function. See the comment above
1131 static void free_rq_clone(struct request *clone)
1133 struct dm_rq_target_io *tio = clone->end_io_data;
1134 struct mapped_device *md = tio->md;
1136 blk_rq_unprep_clone(clone);
1138 if (md->type == DM_TYPE_MQ_REQUEST_BASED)
1139 /* stacked on blk-mq queue(s) */
1140 tio->ti->type->release_clone_rq(clone);
1141 else if (!md->queue->mq_ops)
1142 /* request_fn queue stacked on request_fn queue(s) */
1143 free_clone_request(md, clone);
1145 * NOTE: for the blk-mq queue stacked on request_fn queue(s) case:
1146 * no need to call free_clone_request() because we leverage blk-mq by
1147 * allocating the clone at the end of the blk-mq pdu (see: clone_rq)
1150 if (!md->queue->mq_ops)
1155 * Complete the clone and the original request.
1156 * Must be called without clone's queue lock held,
1157 * see end_clone_request() for more details.
1159 static void dm_end_request(struct request *clone, int error)
1161 int rw = rq_data_dir(clone);
1162 struct dm_rq_target_io *tio = clone->end_io_data;
1163 struct mapped_device *md = tio->md;
1164 struct request *rq = tio->orig;
1166 if (rq->cmd_type == REQ_TYPE_BLOCK_PC) {
1167 rq->errors = clone->errors;
1168 rq->resid_len = clone->resid_len;
1172 * We are using the sense buffer of the original
1174 * So setting the length of the sense data is enough.
1176 rq->sense_len = clone->sense_len;
1179 free_rq_clone(clone);
1180 rq_end_stats(md, rq);
1182 blk_end_request_all(rq, error);
1184 blk_mq_end_request(rq, error);
1185 rq_completed(md, rw, true);
1188 static void dm_unprep_request(struct request *rq)
1190 struct dm_rq_target_io *tio = tio_from_request(rq);
1191 struct request *clone = tio->clone;
1193 if (!rq->q->mq_ops) {
1195 rq->cmd_flags &= ~REQ_DONTPREP;
1199 free_rq_clone(clone);
1200 else if (!tio->md->queue->mq_ops)
1205 * Requeue the original request of a clone.
1207 static void old_requeue_request(struct request *rq)
1209 struct request_queue *q = rq->q;
1210 unsigned long flags;
1212 spin_lock_irqsave(q->queue_lock, flags);
1213 blk_requeue_request(q, rq);
1214 blk_run_queue_async(q);
1215 spin_unlock_irqrestore(q->queue_lock, flags);
1218 static void dm_mq_requeue_request(struct request *rq)
1220 struct request_queue *q = rq->q;
1221 unsigned long flags;
1223 blk_mq_requeue_request(rq);
1224 spin_lock_irqsave(q->queue_lock, flags);
1225 if (!blk_queue_stopped(q))
1226 blk_mq_kick_requeue_list(q);
1227 spin_unlock_irqrestore(q->queue_lock, flags);
1230 static void dm_requeue_original_request(struct mapped_device *md,
1233 int rw = rq_data_dir(rq);
1235 dm_unprep_request(rq);
1237 rq_end_stats(md, rq);
1239 old_requeue_request(rq);
1241 dm_mq_requeue_request(rq);
1243 rq_completed(md, rw, false);
1246 static void old_stop_queue(struct request_queue *q)
1248 unsigned long flags;
1250 spin_lock_irqsave(q->queue_lock, flags);
1251 if (blk_queue_stopped(q)) {
1252 spin_unlock_irqrestore(q->queue_lock, flags);
1257 spin_unlock_irqrestore(q->queue_lock, flags);
1260 static void stop_queue(struct request_queue *q)
1265 blk_mq_stop_hw_queues(q);
1268 static void old_start_queue(struct request_queue *q)
1270 unsigned long flags;
1272 spin_lock_irqsave(q->queue_lock, flags);
1273 if (blk_queue_stopped(q))
1275 spin_unlock_irqrestore(q->queue_lock, flags);
1278 static void start_queue(struct request_queue *q)
1283 blk_mq_start_stopped_hw_queues(q, true);
1284 blk_mq_kick_requeue_list(q);
1288 static void dm_done(struct request *clone, int error, bool mapped)
1291 struct dm_rq_target_io *tio = clone->end_io_data;
1292 dm_request_endio_fn rq_end_io = NULL;
1295 rq_end_io = tio->ti->type->rq_end_io;
1297 if (mapped && rq_end_io)
1298 r = rq_end_io(tio->ti, clone, error, &tio->info);
1301 if (unlikely(r == -EREMOTEIO && (clone->cmd_flags & REQ_WRITE_SAME) &&
1302 !clone->q->limits.max_write_same_sectors))
1303 disable_write_same(tio->md);
1306 /* The target wants to complete the I/O */
1307 dm_end_request(clone, r);
1308 else if (r == DM_ENDIO_INCOMPLETE)
1309 /* The target will handle the I/O */
1311 else if (r == DM_ENDIO_REQUEUE)
1312 /* The target wants to requeue the I/O */
1313 dm_requeue_original_request(tio->md, tio->orig);
1315 DMWARN("unimplemented target endio return value: %d", r);
1321 * Request completion handler for request-based dm
1323 static void dm_softirq_done(struct request *rq)
1326 struct dm_rq_target_io *tio = tio_from_request(rq);
1327 struct request *clone = tio->clone;
1331 rq_end_stats(tio->md, rq);
1332 rw = rq_data_dir(rq);
1333 if (!rq->q->mq_ops) {
1334 blk_end_request_all(rq, tio->error);
1335 rq_completed(tio->md, rw, false);
1338 blk_mq_end_request(rq, tio->error);
1339 rq_completed(tio->md, rw, false);
1344 if (rq->cmd_flags & REQ_FAILED)
1347 dm_done(clone, tio->error, mapped);
1351 * Complete the clone and the original request with the error status
1352 * through softirq context.
1354 static void dm_complete_request(struct request *rq, int error)
1356 struct dm_rq_target_io *tio = tio_from_request(rq);
1360 blk_complete_request(rq);
1362 blk_mq_complete_request(rq, error);
1366 * Complete the not-mapped clone and the original request with the error status
1367 * through softirq context.
1368 * Target's rq_end_io() function isn't called.
1369 * This may be used when the target's map_rq() or clone_and_map_rq() functions fail.
1371 static void dm_kill_unmapped_request(struct request *rq, int error)
1373 rq->cmd_flags |= REQ_FAILED;
1374 dm_complete_request(rq, error);
1378 * Called with the clone's queue lock held (for non-blk-mq)
1380 static void end_clone_request(struct request *clone, int error)
1382 struct dm_rq_target_io *tio = clone->end_io_data;
1384 if (!clone->q->mq_ops) {
1386 * For just cleaning up the information of the queue in which
1387 * the clone was dispatched.
1388 * The clone is *NOT* freed actually here because it is alloced
1389 * from dm own mempool (REQ_ALLOCED isn't set).
1391 __blk_put_request(clone->q, clone);
1395 * Actual request completion is done in a softirq context which doesn't
1396 * hold the clone's queue lock. Otherwise, deadlock could occur because:
1397 * - another request may be submitted by the upper level driver
1398 * of the stacking during the completion
1399 * - the submission which requires queue lock may be done
1400 * against this clone's queue
1402 dm_complete_request(tio->orig, error);
1406 * Return maximum size of I/O possible at the supplied sector up to the current
1409 static sector_t max_io_len_target_boundary(sector_t sector, struct dm_target *ti)
1411 sector_t target_offset = dm_target_offset(ti, sector);
1413 return ti->len - target_offset;
1416 static sector_t max_io_len(sector_t sector, struct dm_target *ti)
1418 sector_t len = max_io_len_target_boundary(sector, ti);
1419 sector_t offset, max_len;
1422 * Does the target need to split even further?
1424 if (ti->max_io_len) {
1425 offset = dm_target_offset(ti, sector);
1426 if (unlikely(ti->max_io_len & (ti->max_io_len - 1)))
1427 max_len = sector_div(offset, ti->max_io_len);
1429 max_len = offset & (ti->max_io_len - 1);
1430 max_len = ti->max_io_len - max_len;
1439 int dm_set_target_max_io_len(struct dm_target *ti, sector_t len)
1441 if (len > UINT_MAX) {
1442 DMERR("Specified maximum size of target IO (%llu) exceeds limit (%u)",
1443 (unsigned long long)len, UINT_MAX);
1444 ti->error = "Maximum size of target IO is too large";
1448 ti->max_io_len = (uint32_t) len;
1452 EXPORT_SYMBOL_GPL(dm_set_target_max_io_len);
1455 * A target may call dm_accept_partial_bio only from the map routine. It is
1456 * allowed for all bio types except REQ_FLUSH.
1458 * dm_accept_partial_bio informs the dm that the target only wants to process
1459 * additional n_sectors sectors of the bio and the rest of the data should be
1460 * sent in a next bio.
1462 * A diagram that explains the arithmetics:
1463 * +--------------------+---------------+-------+
1465 * +--------------------+---------------+-------+
1467 * <-------------- *tio->len_ptr --------------->
1468 * <------- bi_size ------->
1471 * Region 1 was already iterated over with bio_advance or similar function.
1472 * (it may be empty if the target doesn't use bio_advance)
1473 * Region 2 is the remaining bio size that the target wants to process.
1474 * (it may be empty if region 1 is non-empty, although there is no reason
1476 * The target requires that region 3 is to be sent in the next bio.
1478 * If the target wants to receive multiple copies of the bio (via num_*bios, etc),
1479 * the partially processed part (the sum of regions 1+2) must be the same for all
1480 * copies of the bio.
1482 void dm_accept_partial_bio(struct bio *bio, unsigned n_sectors)
1484 struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone);
1485 unsigned bi_size = bio->bi_iter.bi_size >> SECTOR_SHIFT;
1486 BUG_ON(bio->bi_rw & REQ_FLUSH);
1487 BUG_ON(bi_size > *tio->len_ptr);
1488 BUG_ON(n_sectors > bi_size);
1489 *tio->len_ptr -= bi_size - n_sectors;
1490 bio->bi_iter.bi_size = n_sectors << SECTOR_SHIFT;
1492 EXPORT_SYMBOL_GPL(dm_accept_partial_bio);
1494 static void __map_bio(struct dm_target_io *tio)
1498 struct mapped_device *md;
1499 struct bio *clone = &tio->clone;
1500 struct dm_target *ti = tio->ti;
1502 clone->bi_end_io = clone_endio;
1505 * Map the clone. If r == 0 we don't need to do
1506 * anything, the target has assumed ownership of
1509 atomic_inc(&tio->io->io_count);
1510 sector = clone->bi_iter.bi_sector;
1511 r = ti->type->map(ti, clone);
1512 if (r == DM_MAPIO_REMAPPED) {
1513 /* the bio has been remapped so dispatch it */
1515 trace_block_bio_remap(bdev_get_queue(clone->bi_bdev), clone,
1516 tio->io->bio->bi_bdev->bd_dev, sector);
1518 generic_make_request(clone);
1519 } else if (r < 0 || r == DM_MAPIO_REQUEUE) {
1520 /* error the io and bail out, or requeue it if needed */
1522 dec_pending(tio->io, r);
1524 } else if (r != DM_MAPIO_SUBMITTED) {
1525 DMWARN("unimplemented target map return value: %d", r);
1531 struct mapped_device *md;
1532 struct dm_table *map;
1536 unsigned sector_count;
1539 static void bio_setup_sector(struct bio *bio, sector_t sector, unsigned len)
1541 bio->bi_iter.bi_sector = sector;
1542 bio->bi_iter.bi_size = to_bytes(len);
1546 * Creates a bio that consists of range of complete bvecs.
1548 static void clone_bio(struct dm_target_io *tio, struct bio *bio,
1549 sector_t sector, unsigned len)
1551 struct bio *clone = &tio->clone;
1553 __bio_clone_fast(clone, bio);
1555 if (bio_integrity(bio))
1556 bio_integrity_clone(clone, bio, GFP_NOIO);
1558 bio_advance(clone, to_bytes(sector - clone->bi_iter.bi_sector));
1559 clone->bi_iter.bi_size = to_bytes(len);
1561 if (bio_integrity(bio))
1562 bio_integrity_trim(clone, 0, len);
1565 static struct dm_target_io *alloc_tio(struct clone_info *ci,
1566 struct dm_target *ti,
1567 unsigned target_bio_nr)
1569 struct dm_target_io *tio;
1572 clone = bio_alloc_bioset(GFP_NOIO, 0, ci->md->bs);
1573 tio = container_of(clone, struct dm_target_io, clone);
1577 tio->target_bio_nr = target_bio_nr;
1582 static void __clone_and_map_simple_bio(struct clone_info *ci,
1583 struct dm_target *ti,
1584 unsigned target_bio_nr, unsigned *len)
1586 struct dm_target_io *tio = alloc_tio(ci, ti, target_bio_nr);
1587 struct bio *clone = &tio->clone;
1591 __bio_clone_fast(clone, ci->bio);
1593 bio_setup_sector(clone, ci->sector, *len);
1598 static void __send_duplicate_bios(struct clone_info *ci, struct dm_target *ti,
1599 unsigned num_bios, unsigned *len)
1601 unsigned target_bio_nr;
1603 for (target_bio_nr = 0; target_bio_nr < num_bios; target_bio_nr++)
1604 __clone_and_map_simple_bio(ci, ti, target_bio_nr, len);
1607 static int __send_empty_flush(struct clone_info *ci)
1609 unsigned target_nr = 0;
1610 struct dm_target *ti;
1612 BUG_ON(bio_has_data(ci->bio));
1613 while ((ti = dm_table_get_target(ci->map, target_nr++)))
1614 __send_duplicate_bios(ci, ti, ti->num_flush_bios, NULL);
1619 static void __clone_and_map_data_bio(struct clone_info *ci, struct dm_target *ti,
1620 sector_t sector, unsigned *len)
1622 struct bio *bio = ci->bio;
1623 struct dm_target_io *tio;
1624 unsigned target_bio_nr;
1625 unsigned num_target_bios = 1;
1628 * Does the target want to receive duplicate copies of the bio?
1630 if (bio_data_dir(bio) == WRITE && ti->num_write_bios)
1631 num_target_bios = ti->num_write_bios(ti, bio);
1633 for (target_bio_nr = 0; target_bio_nr < num_target_bios; target_bio_nr++) {
1634 tio = alloc_tio(ci, ti, target_bio_nr);
1636 clone_bio(tio, bio, sector, *len);
1641 typedef unsigned (*get_num_bios_fn)(struct dm_target *ti);
1643 static unsigned get_num_discard_bios(struct dm_target *ti)
1645 return ti->num_discard_bios;
1648 static unsigned get_num_write_same_bios(struct dm_target *ti)
1650 return ti->num_write_same_bios;
1653 typedef bool (*is_split_required_fn)(struct dm_target *ti);
1655 static bool is_split_required_for_discard(struct dm_target *ti)
1657 return ti->split_discard_bios;
1660 static int __send_changing_extent_only(struct clone_info *ci,
1661 get_num_bios_fn get_num_bios,
1662 is_split_required_fn is_split_required)
1664 struct dm_target *ti;
1669 ti = dm_table_find_target(ci->map, ci->sector);
1670 if (!dm_target_is_valid(ti))
1674 * Even though the device advertised support for this type of
1675 * request, that does not mean every target supports it, and
1676 * reconfiguration might also have changed that since the
1677 * check was performed.
1679 num_bios = get_num_bios ? get_num_bios(ti) : 0;
1683 if (is_split_required && !is_split_required(ti))
1684 len = min((sector_t)ci->sector_count, max_io_len_target_boundary(ci->sector, ti));
1686 len = min((sector_t)ci->sector_count, max_io_len(ci->sector, ti));
1688 __send_duplicate_bios(ci, ti, num_bios, &len);
1691 } while (ci->sector_count -= len);
1696 static int __send_discard(struct clone_info *ci)
1698 return __send_changing_extent_only(ci, get_num_discard_bios,
1699 is_split_required_for_discard);
1702 static int __send_write_same(struct clone_info *ci)
1704 return __send_changing_extent_only(ci, get_num_write_same_bios, NULL);
1708 * Select the correct strategy for processing a non-flush bio.
1710 static int __split_and_process_non_flush(struct clone_info *ci)
1712 struct bio *bio = ci->bio;
1713 struct dm_target *ti;
1716 if (unlikely(bio->bi_rw & REQ_DISCARD))
1717 return __send_discard(ci);
1718 else if (unlikely(bio->bi_rw & REQ_WRITE_SAME))
1719 return __send_write_same(ci);
1721 ti = dm_table_find_target(ci->map, ci->sector);
1722 if (!dm_target_is_valid(ti))
1725 len = min_t(sector_t, max_io_len(ci->sector, ti), ci->sector_count);
1727 __clone_and_map_data_bio(ci, ti, ci->sector, &len);
1730 ci->sector_count -= len;
1736 * Entry point to split a bio into clones and submit them to the targets.
1738 static void __split_and_process_bio(struct mapped_device *md,
1739 struct dm_table *map, struct bio *bio)
1741 struct clone_info ci;
1744 if (unlikely(!map)) {
1751 ci.io = alloc_io(md);
1753 atomic_set(&ci.io->io_count, 1);
1756 spin_lock_init(&ci.io->endio_lock);
1757 ci.sector = bio->bi_iter.bi_sector;
1759 start_io_acct(ci.io);
1761 if (bio->bi_rw & REQ_FLUSH) {
1762 ci.bio = &ci.md->flush_bio;
1763 ci.sector_count = 0;
1764 error = __send_empty_flush(&ci);
1765 /* dec_pending submits any data associated with flush */
1768 ci.sector_count = bio_sectors(bio);
1769 while (ci.sector_count && !error)
1770 error = __split_and_process_non_flush(&ci);
1773 /* drop the extra reference count */
1774 dec_pending(ci.io, error);
1776 /*-----------------------------------------------------------------
1778 *---------------------------------------------------------------*/
1781 * The request function that just remaps the bio built up by
1784 static blk_qc_t dm_make_request(struct request_queue *q, struct bio *bio)
1786 int rw = bio_data_dir(bio);
1787 struct mapped_device *md = q->queuedata;
1789 struct dm_table *map;
1791 map = dm_get_live_table(md, &srcu_idx);
1793 generic_start_io_acct(rw, bio_sectors(bio), &dm_disk(md)->part0);
1795 /* if we're suspended, we have to queue this io for later */
1796 if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))) {
1797 dm_put_live_table(md, srcu_idx);
1799 if (bio_rw(bio) != READA)
1803 return BLK_QC_T_NONE;
1806 __split_and_process_bio(md, map, bio);
1807 dm_put_live_table(md, srcu_idx);
1808 return BLK_QC_T_NONE;
1811 int dm_request_based(struct mapped_device *md)
1813 return blk_queue_stackable(md->queue);
1816 static void dm_dispatch_clone_request(struct request *clone, struct request *rq)
1820 if (blk_queue_io_stat(clone->q))
1821 clone->cmd_flags |= REQ_IO_STAT;
1823 clone->start_time = jiffies;
1824 r = blk_insert_cloned_request(clone->q, clone);
1826 /* must complete clone in terms of original request */
1827 dm_complete_request(rq, r);
1830 static int dm_rq_bio_constructor(struct bio *bio, struct bio *bio_orig,
1833 struct dm_rq_target_io *tio = data;
1834 struct dm_rq_clone_bio_info *info =
1835 container_of(bio, struct dm_rq_clone_bio_info, clone);
1837 info->orig = bio_orig;
1839 bio->bi_end_io = end_clone_bio;
1844 static int setup_clone(struct request *clone, struct request *rq,
1845 struct dm_rq_target_io *tio, gfp_t gfp_mask)
1849 r = blk_rq_prep_clone(clone, rq, tio->md->bs, gfp_mask,
1850 dm_rq_bio_constructor, tio);
1854 clone->cmd = rq->cmd;
1855 clone->cmd_len = rq->cmd_len;
1856 clone->sense = rq->sense;
1857 clone->end_io = end_clone_request;
1858 clone->end_io_data = tio;
1865 static struct request *clone_rq(struct request *rq, struct mapped_device *md,
1866 struct dm_rq_target_io *tio, gfp_t gfp_mask)
1869 * Do not allocate a clone if tio->clone was already set
1870 * (see: dm_mq_queue_rq).
1872 bool alloc_clone = !tio->clone;
1873 struct request *clone;
1876 clone = alloc_clone_request(md, gfp_mask);
1882 blk_rq_init(NULL, clone);
1883 if (setup_clone(clone, rq, tio, gfp_mask)) {
1886 free_clone_request(md, clone);
1893 static void map_tio_request(struct kthread_work *work);
1895 static void init_tio(struct dm_rq_target_io *tio, struct request *rq,
1896 struct mapped_device *md)
1903 memset(&tio->info, 0, sizeof(tio->info));
1904 if (md->kworker_task)
1905 init_kthread_work(&tio->work, map_tio_request);
1908 static struct dm_rq_target_io *prep_tio(struct request *rq,
1909 struct mapped_device *md, gfp_t gfp_mask)
1911 struct dm_rq_target_io *tio;
1913 struct dm_table *table;
1915 tio = alloc_rq_tio(md, gfp_mask);
1919 init_tio(tio, rq, md);
1921 table = dm_get_live_table(md, &srcu_idx);
1922 if (!dm_table_mq_request_based(table)) {
1923 if (!clone_rq(rq, md, tio, gfp_mask)) {
1924 dm_put_live_table(md, srcu_idx);
1929 dm_put_live_table(md, srcu_idx);
1935 * Called with the queue lock held.
1937 static int dm_prep_fn(struct request_queue *q, struct request *rq)
1939 struct mapped_device *md = q->queuedata;
1940 struct dm_rq_target_io *tio;
1942 if (unlikely(rq->special)) {
1943 DMWARN("Already has something in rq->special.");
1944 return BLKPREP_KILL;
1947 tio = prep_tio(rq, md, GFP_ATOMIC);
1949 return BLKPREP_DEFER;
1952 rq->cmd_flags |= REQ_DONTPREP;
1959 * 0 : the request has been processed
1960 * DM_MAPIO_REQUEUE : the original request needs to be requeued
1961 * < 0 : the request was completed due to failure
1963 static int map_request(struct dm_rq_target_io *tio, struct request *rq,
1964 struct mapped_device *md)
1967 struct dm_target *ti = tio->ti;
1968 struct request *clone = NULL;
1972 r = ti->type->map_rq(ti, clone, &tio->info);
1974 r = ti->type->clone_and_map_rq(ti, rq, &tio->info, &clone);
1976 /* The target wants to complete the I/O */
1977 dm_kill_unmapped_request(rq, r);
1980 if (r != DM_MAPIO_REMAPPED)
1982 if (setup_clone(clone, rq, tio, GFP_ATOMIC)) {
1984 ti->type->release_clone_rq(clone);
1985 return DM_MAPIO_REQUEUE;
1990 case DM_MAPIO_SUBMITTED:
1991 /* The target has taken the I/O to submit by itself later */
1993 case DM_MAPIO_REMAPPED:
1994 /* The target has remapped the I/O so dispatch it */
1995 trace_block_rq_remap(clone->q, clone, disk_devt(dm_disk(md)),
1997 dm_dispatch_clone_request(clone, rq);
1999 case DM_MAPIO_REQUEUE:
2000 /* The target wants to requeue the I/O */
2001 dm_requeue_original_request(md, tio->orig);
2005 DMWARN("unimplemented target map return value: %d", r);
2009 /* The target wants to complete the I/O */
2010 dm_kill_unmapped_request(rq, r);
2017 static void map_tio_request(struct kthread_work *work)
2019 struct dm_rq_target_io *tio = container_of(work, struct dm_rq_target_io, work);
2020 struct request *rq = tio->orig;
2021 struct mapped_device *md = tio->md;
2023 if (map_request(tio, rq, md) == DM_MAPIO_REQUEUE)
2024 dm_requeue_original_request(md, rq);
2027 static void dm_start_request(struct mapped_device *md, struct request *orig)
2029 if (!orig->q->mq_ops)
2030 blk_start_request(orig);
2032 blk_mq_start_request(orig);
2033 atomic_inc(&md->pending[rq_data_dir(orig)]);
2035 if (md->seq_rq_merge_deadline_usecs) {
2036 md->last_rq_pos = rq_end_sector(orig);
2037 md->last_rq_rw = rq_data_dir(orig);
2038 md->last_rq_start_time = ktime_get();
2041 if (unlikely(dm_stats_used(&md->stats))) {
2042 struct dm_rq_target_io *tio = tio_from_request(orig);
2043 tio->duration_jiffies = jiffies;
2044 tio->n_sectors = blk_rq_sectors(orig);
2045 dm_stats_account_io(&md->stats, orig->cmd_flags, blk_rq_pos(orig),
2046 tio->n_sectors, false, 0, &tio->stats_aux);
2050 * Hold the md reference here for the in-flight I/O.
2051 * We can't rely on the reference count by device opener,
2052 * because the device may be closed during the request completion
2053 * when all bios are completed.
2054 * See the comment in rq_completed() too.
2059 #define MAX_SEQ_RQ_MERGE_DEADLINE_USECS 100000
2061 ssize_t dm_attr_rq_based_seq_io_merge_deadline_show(struct mapped_device *md, char *buf)
2063 return sprintf(buf, "%u\n", md->seq_rq_merge_deadline_usecs);
2066 ssize_t dm_attr_rq_based_seq_io_merge_deadline_store(struct mapped_device *md,
2067 const char *buf, size_t count)
2071 if (!dm_request_based(md) || md->use_blk_mq)
2074 if (kstrtouint(buf, 10, &deadline))
2077 if (deadline > MAX_SEQ_RQ_MERGE_DEADLINE_USECS)
2078 deadline = MAX_SEQ_RQ_MERGE_DEADLINE_USECS;
2080 md->seq_rq_merge_deadline_usecs = deadline;
2085 static bool dm_request_peeked_before_merge_deadline(struct mapped_device *md)
2087 ktime_t kt_deadline;
2089 if (!md->seq_rq_merge_deadline_usecs)
2092 kt_deadline = ns_to_ktime((u64)md->seq_rq_merge_deadline_usecs * NSEC_PER_USEC);
2093 kt_deadline = ktime_add_safe(md->last_rq_start_time, kt_deadline);
2095 return !ktime_after(ktime_get(), kt_deadline);
2099 * q->request_fn for request-based dm.
2100 * Called with the queue lock held.
2102 static void dm_request_fn(struct request_queue *q)
2104 struct mapped_device *md = q->queuedata;
2105 struct dm_target *ti = md->immutable_target;
2107 struct dm_rq_target_io *tio;
2110 if (unlikely(!ti)) {
2112 struct dm_table *map = dm_get_live_table(md, &srcu_idx);
2114 ti = dm_table_find_target(map, pos);
2115 dm_put_live_table(md, srcu_idx);
2119 * For suspend, check blk_queue_stopped() and increment
2120 * ->pending within a single queue_lock not to increment the
2121 * number of in-flight I/Os after the queue is stopped in
2124 while (!blk_queue_stopped(q)) {
2125 rq = blk_peek_request(q);
2129 /* always use block 0 to find the target for flushes for now */
2131 if (!(rq->cmd_flags & REQ_FLUSH))
2132 pos = blk_rq_pos(rq);
2134 if ((dm_request_peeked_before_merge_deadline(md) &&
2135 md_in_flight(md) && rq->bio && rq->bio->bi_vcnt == 1 &&
2136 md->last_rq_pos == pos && md->last_rq_rw == rq_data_dir(rq)) ||
2137 (ti->type->busy && ti->type->busy(ti))) {
2138 blk_delay_queue(q, HZ / 100);
2142 dm_start_request(md, rq);
2144 tio = tio_from_request(rq);
2145 /* Establish tio->ti before queuing work (map_tio_request) */
2147 queue_kthread_work(&md->kworker, &tio->work);
2148 BUG_ON(!irqs_disabled());
2152 static int dm_any_congested(void *congested_data, int bdi_bits)
2155 struct mapped_device *md = congested_data;
2156 struct dm_table *map;
2158 if (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
2159 if (dm_request_based(md)) {
2161 * With request-based DM we only need to check the
2162 * top-level queue for congestion.
2164 r = md->queue->backing_dev_info.wb.state & bdi_bits;
2166 map = dm_get_live_table_fast(md);
2168 r = dm_table_any_congested(map, bdi_bits);
2169 dm_put_live_table_fast(md);
2176 /*-----------------------------------------------------------------
2177 * An IDR is used to keep track of allocated minor numbers.
2178 *---------------------------------------------------------------*/
2179 static void free_minor(int minor)
2181 spin_lock(&_minor_lock);
2182 idr_remove(&_minor_idr, minor);
2183 spin_unlock(&_minor_lock);
2187 * See if the device with a specific minor # is free.
2189 static int specific_minor(int minor)
2193 if (minor >= (1 << MINORBITS))
2196 idr_preload(GFP_KERNEL);
2197 spin_lock(&_minor_lock);
2199 r = idr_alloc(&_minor_idr, MINOR_ALLOCED, minor, minor + 1, GFP_NOWAIT);
2201 spin_unlock(&_minor_lock);
2204 return r == -ENOSPC ? -EBUSY : r;
2208 static int next_free_minor(int *minor)
2212 idr_preload(GFP_KERNEL);
2213 spin_lock(&_minor_lock);
2215 r = idr_alloc(&_minor_idr, MINOR_ALLOCED, 0, 1 << MINORBITS, GFP_NOWAIT);
2217 spin_unlock(&_minor_lock);
2225 static const struct block_device_operations dm_blk_dops;
2227 static void dm_wq_work(struct work_struct *work);
2229 static void dm_init_md_queue(struct mapped_device *md)
2232 * Request-based dm devices cannot be stacked on top of bio-based dm
2233 * devices. The type of this dm device may not have been decided yet.
2234 * The type is decided at the first table loading time.
2235 * To prevent problematic device stacking, clear the queue flag
2236 * for request stacking support until then.
2238 * This queue is new, so no concurrency on the queue_flags.
2240 queue_flag_clear_unlocked(QUEUE_FLAG_STACKABLE, md->queue);
2243 * Initialize data that will only be used by a non-blk-mq DM queue
2244 * - must do so here (in alloc_dev callchain) before queue is used
2246 md->queue->queuedata = md;
2247 md->queue->backing_dev_info.congested_data = md;
2250 static void dm_init_old_md_queue(struct mapped_device *md)
2252 md->use_blk_mq = false;
2253 dm_init_md_queue(md);
2256 * Initialize aspects of queue that aren't relevant for blk-mq
2258 md->queue->backing_dev_info.congested_fn = dm_any_congested;
2259 blk_queue_bounce_limit(md->queue, BLK_BOUNCE_ANY);
2262 static void cleanup_mapped_device(struct mapped_device *md)
2265 destroy_workqueue(md->wq);
2266 if (md->kworker_task)
2267 kthread_stop(md->kworker_task);
2268 mempool_destroy(md->io_pool);
2269 mempool_destroy(md->rq_pool);
2271 bioset_free(md->bs);
2273 cleanup_srcu_struct(&md->io_barrier);
2276 spin_lock(&_minor_lock);
2277 md->disk->private_data = NULL;
2278 spin_unlock(&_minor_lock);
2279 del_gendisk(md->disk);
2284 blk_cleanup_queue(md->queue);
2293 * Allocate and initialise a blank device with a given minor.
2295 static struct mapped_device *alloc_dev(int minor)
2298 struct mapped_device *md = kzalloc(sizeof(*md), GFP_KERNEL);
2302 DMWARN("unable to allocate device, out of memory.");
2306 if (!try_module_get(THIS_MODULE))
2307 goto bad_module_get;
2309 /* get a minor number for the dev */
2310 if (minor == DM_ANY_MINOR)
2311 r = next_free_minor(&minor);
2313 r = specific_minor(minor);
2317 r = init_srcu_struct(&md->io_barrier);
2319 goto bad_io_barrier;
2321 md->use_blk_mq = use_blk_mq;
2322 md->type = DM_TYPE_NONE;
2323 mutex_init(&md->suspend_lock);
2324 mutex_init(&md->type_lock);
2325 mutex_init(&md->table_devices_lock);
2326 spin_lock_init(&md->deferred_lock);
2327 atomic_set(&md->holders, 1);
2328 atomic_set(&md->open_count, 0);
2329 atomic_set(&md->event_nr, 0);
2330 atomic_set(&md->uevent_seq, 0);
2331 INIT_LIST_HEAD(&md->uevent_list);
2332 INIT_LIST_HEAD(&md->table_devices);
2333 spin_lock_init(&md->uevent_lock);
2335 md->queue = blk_alloc_queue(GFP_KERNEL);
2339 dm_init_md_queue(md);
2341 md->disk = alloc_disk(1);
2345 atomic_set(&md->pending[0], 0);
2346 atomic_set(&md->pending[1], 0);
2347 init_waitqueue_head(&md->wait);
2348 INIT_WORK(&md->work, dm_wq_work);
2349 init_waitqueue_head(&md->eventq);
2350 init_completion(&md->kobj_holder.completion);
2351 md->kworker_task = NULL;
2353 md->disk->major = _major;
2354 md->disk->first_minor = minor;
2355 md->disk->fops = &dm_blk_dops;
2356 md->disk->queue = md->queue;
2357 md->disk->private_data = md;
2358 sprintf(md->disk->disk_name, "dm-%d", minor);
2360 format_dev_t(md->name, MKDEV(_major, minor));
2362 md->wq = alloc_workqueue("kdmflush", WQ_MEM_RECLAIM, 0);
2366 md->bdev = bdget_disk(md->disk, 0);
2370 bio_init(&md->flush_bio);
2371 md->flush_bio.bi_bdev = md->bdev;
2372 md->flush_bio.bi_rw = WRITE_FLUSH;
2374 dm_stats_init(&md->stats);
2376 /* Populate the mapping, nobody knows we exist yet */
2377 spin_lock(&_minor_lock);
2378 old_md = idr_replace(&_minor_idr, md, minor);
2379 spin_unlock(&_minor_lock);
2381 BUG_ON(old_md != MINOR_ALLOCED);
2386 cleanup_mapped_device(md);
2390 module_put(THIS_MODULE);
2396 static void unlock_fs(struct mapped_device *md);
2398 static void free_dev(struct mapped_device *md)
2400 int minor = MINOR(disk_devt(md->disk));
2404 cleanup_mapped_device(md);
2406 blk_mq_free_tag_set(md->tag_set);
2410 free_table_devices(&md->table_devices);
2411 dm_stats_cleanup(&md->stats);
2414 module_put(THIS_MODULE);
2418 static void __bind_mempools(struct mapped_device *md, struct dm_table *t)
2420 struct dm_md_mempools *p = dm_table_get_md_mempools(t);
2423 /* The md already has necessary mempools. */
2424 if (dm_table_get_type(t) == DM_TYPE_BIO_BASED) {
2426 * Reload bioset because front_pad may have changed
2427 * because a different table was loaded.
2429 bioset_free(md->bs);
2434 * There's no need to reload with request-based dm
2435 * because the size of front_pad doesn't change.
2436 * Note for future: If you are to reload bioset,
2437 * prep-ed requests in the queue may refer
2438 * to bio from the old bioset, so you must walk
2439 * through the queue to unprep.
2444 BUG_ON(!p || md->io_pool || md->rq_pool || md->bs);
2446 md->io_pool = p->io_pool;
2448 md->rq_pool = p->rq_pool;
2454 /* mempool bind completed, no longer need any mempools in the table */
2455 dm_table_free_md_mempools(t);
2459 * Bind a table to the device.
2461 static void event_callback(void *context)
2463 unsigned long flags;
2465 struct mapped_device *md = (struct mapped_device *) context;
2467 spin_lock_irqsave(&md->uevent_lock, flags);
2468 list_splice_init(&md->uevent_list, &uevents);
2469 spin_unlock_irqrestore(&md->uevent_lock, flags);
2471 dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj);
2473 atomic_inc(&md->event_nr);
2474 wake_up(&md->eventq);
2478 * Protected by md->suspend_lock obtained by dm_swap_table().
2480 static void __set_size(struct mapped_device *md, sector_t size)
2482 set_capacity(md->disk, size);
2484 i_size_write(md->bdev->bd_inode, (loff_t)size << SECTOR_SHIFT);
2488 * Returns old map, which caller must destroy.
2490 static struct dm_table *__bind(struct mapped_device *md, struct dm_table *t,
2491 struct queue_limits *limits)
2493 struct dm_table *old_map;
2494 struct request_queue *q = md->queue;
2497 size = dm_table_get_size(t);
2500 * Wipe any geometry if the size of the table changed.
2502 if (size != dm_get_size(md))
2503 memset(&md->geometry, 0, sizeof(md->geometry));
2505 __set_size(md, size);
2507 dm_table_event_callback(t, event_callback, md);
2510 * The queue hasn't been stopped yet, if the old table type wasn't
2511 * for request-based during suspension. So stop it to prevent
2512 * I/O mapping before resume.
2513 * This must be done before setting the queue restrictions,
2514 * because request-based dm may be run just after the setting.
2516 if (dm_table_request_based(t)) {
2519 * Leverage the fact that request-based DM targets are
2520 * immutable singletons and establish md->immutable_target
2521 * - used to optimize both dm_request_fn and dm_mq_queue_rq
2523 md->immutable_target = dm_table_get_immutable_target(t);
2526 __bind_mempools(md, t);
2528 old_map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2529 rcu_assign_pointer(md->map, t);
2530 md->immutable_target_type = dm_table_get_immutable_target_type(t);
2532 dm_table_set_restrictions(t, q, limits);
2540 * Returns unbound table for the caller to free.
2542 static struct dm_table *__unbind(struct mapped_device *md)
2544 struct dm_table *map = rcu_dereference_protected(md->map, 1);
2549 dm_table_event_callback(map, NULL, NULL);
2550 RCU_INIT_POINTER(md->map, NULL);
2557 * Constructor for a new device.
2559 int dm_create(int minor, struct mapped_device **result)
2561 struct mapped_device *md;
2563 md = alloc_dev(minor);
2574 * Functions to manage md->type.
2575 * All are required to hold md->type_lock.
2577 void dm_lock_md_type(struct mapped_device *md)
2579 mutex_lock(&md->type_lock);
2582 void dm_unlock_md_type(struct mapped_device *md)
2584 mutex_unlock(&md->type_lock);
2587 void dm_set_md_type(struct mapped_device *md, unsigned type)
2589 BUG_ON(!mutex_is_locked(&md->type_lock));
2593 unsigned dm_get_md_type(struct mapped_device *md)
2598 struct target_type *dm_get_immutable_target_type(struct mapped_device *md)
2600 return md->immutable_target_type;
2604 * The queue_limits are only valid as long as you have a reference
2607 struct queue_limits *dm_get_queue_limits(struct mapped_device *md)
2609 BUG_ON(!atomic_read(&md->holders));
2610 return &md->queue->limits;
2612 EXPORT_SYMBOL_GPL(dm_get_queue_limits);
2614 static void init_rq_based_worker_thread(struct mapped_device *md)
2616 /* Initialize the request-based DM worker thread */
2617 init_kthread_worker(&md->kworker);
2618 md->kworker_task = kthread_run(kthread_worker_fn, &md->kworker,
2619 "kdmwork-%s", dm_device_name(md));
2623 * Fully initialize a request-based queue (->elevator, ->request_fn, etc).
2625 static int dm_init_request_based_queue(struct mapped_device *md)
2627 struct request_queue *q = NULL;
2629 /* Fully initialize the queue */
2630 q = blk_init_allocated_queue(md->queue, dm_request_fn, NULL);
2634 /* disable dm_request_fn's merge heuristic by default */
2635 md->seq_rq_merge_deadline_usecs = 0;
2638 dm_init_old_md_queue(md);
2639 blk_queue_softirq_done(md->queue, dm_softirq_done);
2640 blk_queue_prep_rq(md->queue, dm_prep_fn);
2642 init_rq_based_worker_thread(md);
2644 elv_register_queue(md->queue);
2649 static int dm_mq_init_request(void *data, struct request *rq,
2650 unsigned int hctx_idx, unsigned int request_idx,
2651 unsigned int numa_node)
2653 struct mapped_device *md = data;
2654 struct dm_rq_target_io *tio = blk_mq_rq_to_pdu(rq);
2657 * Must initialize md member of tio, otherwise it won't
2658 * be available in dm_mq_queue_rq.
2665 static int dm_mq_queue_rq(struct blk_mq_hw_ctx *hctx,
2666 const struct blk_mq_queue_data *bd)
2668 struct request *rq = bd->rq;
2669 struct dm_rq_target_io *tio = blk_mq_rq_to_pdu(rq);
2670 struct mapped_device *md = tio->md;
2671 struct dm_target *ti = md->immutable_target;
2673 if (unlikely(!ti)) {
2675 struct dm_table *map = dm_get_live_table(md, &srcu_idx);
2677 ti = dm_table_find_target(map, 0);
2678 dm_put_live_table(md, srcu_idx);
2681 if (ti->type->busy && ti->type->busy(ti))
2682 return BLK_MQ_RQ_QUEUE_BUSY;
2684 dm_start_request(md, rq);
2686 /* Init tio using md established in .init_request */
2687 init_tio(tio, rq, md);
2690 * Establish tio->ti before queuing work (map_tio_request)
2691 * or making direct call to map_request().
2696 * Both the table and md type cannot change after initial table load
2698 if (dm_get_md_type(md) == DM_TYPE_REQUEST_BASED) {
2699 /* clone request is allocated at the end of the pdu */
2700 tio->clone = (void *)blk_mq_rq_to_pdu(rq) + sizeof(struct dm_rq_target_io);
2701 (void) clone_rq(rq, md, tio, GFP_ATOMIC);
2702 queue_kthread_work(&md->kworker, &tio->work);
2704 /* Direct call is fine since .queue_rq allows allocations */
2705 if (map_request(tio, rq, md) == DM_MAPIO_REQUEUE) {
2706 /* Undo dm_start_request() before requeuing */
2707 rq_end_stats(md, rq);
2708 rq_completed(md, rq_data_dir(rq), false);
2709 return BLK_MQ_RQ_QUEUE_BUSY;
2713 return BLK_MQ_RQ_QUEUE_OK;
2716 static struct blk_mq_ops dm_mq_ops = {
2717 .queue_rq = dm_mq_queue_rq,
2718 .map_queue = blk_mq_map_queue,
2719 .complete = dm_softirq_done,
2720 .init_request = dm_mq_init_request,
2723 static int dm_init_request_based_blk_mq_queue(struct mapped_device *md)
2725 unsigned md_type = dm_get_md_type(md);
2726 struct request_queue *q;
2729 md->tag_set = kzalloc(sizeof(struct blk_mq_tag_set), GFP_KERNEL);
2733 md->tag_set->ops = &dm_mq_ops;
2734 md->tag_set->queue_depth = dm_get_blk_mq_queue_depth();
2735 md->tag_set->numa_node = NUMA_NO_NODE;
2736 md->tag_set->flags = BLK_MQ_F_SHOULD_MERGE | BLK_MQ_F_SG_MERGE;
2737 md->tag_set->nr_hw_queues = dm_get_blk_mq_nr_hw_queues();
2738 md->tag_set->driver_data = md;
2740 md->tag_set->cmd_size = sizeof(struct dm_rq_target_io);
2741 if (md_type == DM_TYPE_REQUEST_BASED) {
2742 /* put the memory for non-blk-mq clone at the end of the pdu */
2743 md->tag_set->cmd_size += sizeof(struct request);
2746 err = blk_mq_alloc_tag_set(md->tag_set);
2748 goto out_kfree_tag_set;
2750 q = blk_mq_init_allocated_queue(md->tag_set, md->queue);
2756 dm_init_md_queue(md);
2758 /* backfill 'mq' sysfs registration normally done in blk_register_queue */
2759 blk_mq_register_disk(md->disk);
2761 if (md_type == DM_TYPE_REQUEST_BASED)
2762 init_rq_based_worker_thread(md);
2767 blk_mq_free_tag_set(md->tag_set);
2774 static unsigned filter_md_type(unsigned type, struct mapped_device *md)
2776 if (type == DM_TYPE_BIO_BASED)
2779 return !md->use_blk_mq ? DM_TYPE_REQUEST_BASED : DM_TYPE_MQ_REQUEST_BASED;
2783 * Setup the DM device's queue based on md's type
2785 int dm_setup_md_queue(struct mapped_device *md)
2788 unsigned md_type = filter_md_type(dm_get_md_type(md), md);
2791 case DM_TYPE_REQUEST_BASED:
2792 r = dm_init_request_based_queue(md);
2794 DMWARN("Cannot initialize queue for request-based mapped device");
2798 case DM_TYPE_MQ_REQUEST_BASED:
2799 r = dm_init_request_based_blk_mq_queue(md);
2801 DMWARN("Cannot initialize queue for request-based blk-mq mapped device");
2805 case DM_TYPE_BIO_BASED:
2806 dm_init_old_md_queue(md);
2807 blk_queue_make_request(md->queue, dm_make_request);
2809 * DM handles splitting bios as needed. Free the bio_split bioset
2810 * since it won't be used (saves 1 process per bio-based DM device).
2812 bioset_free(md->queue->bio_split);
2813 md->queue->bio_split = NULL;
2820 struct mapped_device *dm_get_md(dev_t dev)
2822 struct mapped_device *md;
2823 unsigned minor = MINOR(dev);
2825 if (MAJOR(dev) != _major || minor >= (1 << MINORBITS))
2828 spin_lock(&_minor_lock);
2830 md = idr_find(&_minor_idr, minor);
2832 if ((md == MINOR_ALLOCED ||
2833 (MINOR(disk_devt(dm_disk(md))) != minor) ||
2834 dm_deleting_md(md) ||
2835 test_bit(DMF_FREEING, &md->flags))) {
2843 spin_unlock(&_minor_lock);
2847 EXPORT_SYMBOL_GPL(dm_get_md);
2849 void *dm_get_mdptr(struct mapped_device *md)
2851 return md->interface_ptr;
2854 void dm_set_mdptr(struct mapped_device *md, void *ptr)
2856 md->interface_ptr = ptr;
2859 void dm_get(struct mapped_device *md)
2861 atomic_inc(&md->holders);
2862 BUG_ON(test_bit(DMF_FREEING, &md->flags));
2865 int dm_hold(struct mapped_device *md)
2867 spin_lock(&_minor_lock);
2868 if (test_bit(DMF_FREEING, &md->flags)) {
2869 spin_unlock(&_minor_lock);
2873 spin_unlock(&_minor_lock);
2876 EXPORT_SYMBOL_GPL(dm_hold);
2878 const char *dm_device_name(struct mapped_device *md)
2882 EXPORT_SYMBOL_GPL(dm_device_name);
2884 static void __dm_destroy(struct mapped_device *md, bool wait)
2886 struct dm_table *map;
2891 spin_lock(&_minor_lock);
2892 idr_replace(&_minor_idr, MINOR_ALLOCED, MINOR(disk_devt(dm_disk(md))));
2893 set_bit(DMF_FREEING, &md->flags);
2894 spin_unlock(&_minor_lock);
2896 if (dm_request_based(md) && md->kworker_task)
2897 flush_kthread_worker(&md->kworker);
2900 * Take suspend_lock so that presuspend and postsuspend methods
2901 * do not race with internal suspend.
2903 mutex_lock(&md->suspend_lock);
2904 map = dm_get_live_table(md, &srcu_idx);
2905 if (!dm_suspended_md(md)) {
2906 dm_table_presuspend_targets(map);
2907 dm_table_postsuspend_targets(map);
2909 /* dm_put_live_table must be before msleep, otherwise deadlock is possible */
2910 dm_put_live_table(md, srcu_idx);
2911 mutex_unlock(&md->suspend_lock);
2914 * Rare, but there may be I/O requests still going to complete,
2915 * for example. Wait for all references to disappear.
2916 * No one should increment the reference count of the mapped_device,
2917 * after the mapped_device state becomes DMF_FREEING.
2920 while (atomic_read(&md->holders))
2922 else if (atomic_read(&md->holders))
2923 DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)",
2924 dm_device_name(md), atomic_read(&md->holders));
2927 dm_table_destroy(__unbind(md));
2931 void dm_destroy(struct mapped_device *md)
2933 __dm_destroy(md, true);
2936 void dm_destroy_immediate(struct mapped_device *md)
2938 __dm_destroy(md, false);
2941 void dm_put(struct mapped_device *md)
2943 atomic_dec(&md->holders);
2945 EXPORT_SYMBOL_GPL(dm_put);
2947 static int dm_wait_for_completion(struct mapped_device *md, int interruptible)
2950 DECLARE_WAITQUEUE(wait, current);
2952 add_wait_queue(&md->wait, &wait);
2955 set_current_state(interruptible);
2957 if (!md_in_flight(md))
2960 if (interruptible == TASK_INTERRUPTIBLE &&
2961 signal_pending(current)) {
2968 set_current_state(TASK_RUNNING);
2970 remove_wait_queue(&md->wait, &wait);
2976 * Process the deferred bios
2978 static void dm_wq_work(struct work_struct *work)
2980 struct mapped_device *md = container_of(work, struct mapped_device,
2984 struct dm_table *map;
2986 map = dm_get_live_table(md, &srcu_idx);
2988 while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
2989 spin_lock_irq(&md->deferred_lock);
2990 c = bio_list_pop(&md->deferred);
2991 spin_unlock_irq(&md->deferred_lock);
2996 if (dm_request_based(md))
2997 generic_make_request(c);
2999 __split_and_process_bio(md, map, c);
3002 dm_put_live_table(md, srcu_idx);
3005 static void dm_queue_flush(struct mapped_device *md)
3007 clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
3008 smp_mb__after_atomic();
3009 queue_work(md->wq, &md->work);
3013 * Swap in a new table, returning the old one for the caller to destroy.
3015 struct dm_table *dm_swap_table(struct mapped_device *md, struct dm_table *table)
3017 struct dm_table *live_map = NULL, *map = ERR_PTR(-EINVAL);
3018 struct queue_limits limits;
3021 mutex_lock(&md->suspend_lock);
3023 /* device must be suspended */
3024 if (!dm_suspended_md(md))
3028 * If the new table has no data devices, retain the existing limits.
3029 * This helps multipath with queue_if_no_path if all paths disappear,
3030 * then new I/O is queued based on these limits, and then some paths
3033 if (dm_table_has_no_data_devices(table)) {
3034 live_map = dm_get_live_table_fast(md);
3036 limits = md->queue->limits;
3037 dm_put_live_table_fast(md);
3041 r = dm_calculate_queue_limits(table, &limits);
3048 map = __bind(md, table, &limits);
3051 mutex_unlock(&md->suspend_lock);
3056 * Functions to lock and unlock any filesystem running on the
3059 static int lock_fs(struct mapped_device *md)
3063 WARN_ON(md->frozen_sb);
3065 md->frozen_sb = freeze_bdev(md->bdev);
3066 if (IS_ERR(md->frozen_sb)) {
3067 r = PTR_ERR(md->frozen_sb);
3068 md->frozen_sb = NULL;
3072 set_bit(DMF_FROZEN, &md->flags);
3077 static void unlock_fs(struct mapped_device *md)
3079 if (!test_bit(DMF_FROZEN, &md->flags))
3082 thaw_bdev(md->bdev, md->frozen_sb);
3083 md->frozen_sb = NULL;
3084 clear_bit(DMF_FROZEN, &md->flags);
3088 * If __dm_suspend returns 0, the device is completely quiescent
3089 * now. There is no request-processing activity. All new requests
3090 * are being added to md->deferred list.
3092 * Caller must hold md->suspend_lock
3094 static int __dm_suspend(struct mapped_device *md, struct dm_table *map,
3095 unsigned suspend_flags, int interruptible)
3097 bool do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG;
3098 bool noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG;
3102 * DMF_NOFLUSH_SUSPENDING must be set before presuspend.
3103 * This flag is cleared before dm_suspend returns.
3106 set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
3109 * This gets reverted if there's an error later and the targets
3110 * provide the .presuspend_undo hook.
3112 dm_table_presuspend_targets(map);
3115 * Flush I/O to the device.
3116 * Any I/O submitted after lock_fs() may not be flushed.
3117 * noflush takes precedence over do_lockfs.
3118 * (lock_fs() flushes I/Os and waits for them to complete.)
3120 if (!noflush && do_lockfs) {
3123 dm_table_presuspend_undo_targets(map);
3129 * Here we must make sure that no processes are submitting requests
3130 * to target drivers i.e. no one may be executing
3131 * __split_and_process_bio. This is called from dm_request and
3134 * To get all processes out of __split_and_process_bio in dm_request,
3135 * we take the write lock. To prevent any process from reentering
3136 * __split_and_process_bio from dm_request and quiesce the thread
3137 * (dm_wq_work), we set BMF_BLOCK_IO_FOR_SUSPEND and call
3138 * flush_workqueue(md->wq).
3140 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
3142 synchronize_srcu(&md->io_barrier);
3145 * Stop md->queue before flushing md->wq in case request-based
3146 * dm defers requests to md->wq from md->queue.
3148 if (dm_request_based(md)) {
3149 stop_queue(md->queue);
3150 if (md->kworker_task)
3151 flush_kthread_worker(&md->kworker);
3154 flush_workqueue(md->wq);
3157 * At this point no more requests are entering target request routines.
3158 * We call dm_wait_for_completion to wait for all existing requests
3161 r = dm_wait_for_completion(md, interruptible);
3164 clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
3166 synchronize_srcu(&md->io_barrier);
3168 /* were we interrupted ? */
3172 if (dm_request_based(md))
3173 start_queue(md->queue);
3176 dm_table_presuspend_undo_targets(map);
3177 /* pushback list is already flushed, so skip flush */
3184 * We need to be able to change a mapping table under a mounted
3185 * filesystem. For example we might want to move some data in
3186 * the background. Before the table can be swapped with
3187 * dm_bind_table, dm_suspend must be called to flush any in
3188 * flight bios and ensure that any further io gets deferred.
3191 * Suspend mechanism in request-based dm.
3193 * 1. Flush all I/Os by lock_fs() if needed.
3194 * 2. Stop dispatching any I/O by stopping the request_queue.
3195 * 3. Wait for all in-flight I/Os to be completed or requeued.
3197 * To abort suspend, start the request_queue.
3199 int dm_suspend(struct mapped_device *md, unsigned suspend_flags)
3201 struct dm_table *map = NULL;
3205 mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING);
3207 if (dm_suspended_md(md)) {
3212 if (dm_suspended_internally_md(md)) {
3213 /* already internally suspended, wait for internal resume */
3214 mutex_unlock(&md->suspend_lock);
3215 r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE);
3221 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
3223 r = __dm_suspend(md, map, suspend_flags, TASK_INTERRUPTIBLE);
3227 set_bit(DMF_SUSPENDED, &md->flags);
3229 dm_table_postsuspend_targets(map);
3232 mutex_unlock(&md->suspend_lock);
3236 static int __dm_resume(struct mapped_device *md, struct dm_table *map)
3239 int r = dm_table_resume_targets(map);
3247 * Flushing deferred I/Os must be done after targets are resumed
3248 * so that mapping of targets can work correctly.
3249 * Request-based dm is queueing the deferred I/Os in its request_queue.
3251 if (dm_request_based(md))
3252 start_queue(md->queue);
3259 int dm_resume(struct mapped_device *md)
3262 struct dm_table *map = NULL;
3265 mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING);
3267 if (!dm_suspended_md(md))
3270 if (dm_suspended_internally_md(md)) {
3271 /* already internally suspended, wait for internal resume */
3272 mutex_unlock(&md->suspend_lock);
3273 r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE);
3279 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
3280 if (!map || !dm_table_get_size(map))
3283 r = __dm_resume(md, map);
3287 clear_bit(DMF_SUSPENDED, &md->flags);
3291 mutex_unlock(&md->suspend_lock);
3297 * Internal suspend/resume works like userspace-driven suspend. It waits
3298 * until all bios finish and prevents issuing new bios to the target drivers.
3299 * It may be used only from the kernel.
3302 static void __dm_internal_suspend(struct mapped_device *md, unsigned suspend_flags)
3304 struct dm_table *map = NULL;
3306 if (md->internal_suspend_count++)
3307 return; /* nested internal suspend */
3309 if (dm_suspended_md(md)) {
3310 set_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
3311 return; /* nest suspend */
3314 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
3317 * Using TASK_UNINTERRUPTIBLE because only NOFLUSH internal suspend is
3318 * supported. Properly supporting a TASK_INTERRUPTIBLE internal suspend
3319 * would require changing .presuspend to return an error -- avoid this
3320 * until there is a need for more elaborate variants of internal suspend.
3322 (void) __dm_suspend(md, map, suspend_flags, TASK_UNINTERRUPTIBLE);
3324 set_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
3326 dm_table_postsuspend_targets(map);
3329 static void __dm_internal_resume(struct mapped_device *md)
3331 BUG_ON(!md->internal_suspend_count);
3333 if (--md->internal_suspend_count)
3334 return; /* resume from nested internal suspend */
3336 if (dm_suspended_md(md))
3337 goto done; /* resume from nested suspend */
3340 * NOTE: existing callers don't need to call dm_table_resume_targets
3341 * (which may fail -- so best to avoid it for now by passing NULL map)
3343 (void) __dm_resume(md, NULL);
3346 clear_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
3347 smp_mb__after_atomic();
3348 wake_up_bit(&md->flags, DMF_SUSPENDED_INTERNALLY);
3351 void dm_internal_suspend_noflush(struct mapped_device *md)
3353 mutex_lock(&md->suspend_lock);
3354 __dm_internal_suspend(md, DM_SUSPEND_NOFLUSH_FLAG);
3355 mutex_unlock(&md->suspend_lock);
3357 EXPORT_SYMBOL_GPL(dm_internal_suspend_noflush);
3359 void dm_internal_resume(struct mapped_device *md)
3361 mutex_lock(&md->suspend_lock);
3362 __dm_internal_resume(md);
3363 mutex_unlock(&md->suspend_lock);
3365 EXPORT_SYMBOL_GPL(dm_internal_resume);
3368 * Fast variants of internal suspend/resume hold md->suspend_lock,
3369 * which prevents interaction with userspace-driven suspend.
3372 void dm_internal_suspend_fast(struct mapped_device *md)
3374 mutex_lock(&md->suspend_lock);
3375 if (dm_suspended_md(md) || dm_suspended_internally_md(md))
3378 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
3379 synchronize_srcu(&md->io_barrier);
3380 flush_workqueue(md->wq);
3381 dm_wait_for_completion(md, TASK_UNINTERRUPTIBLE);
3383 EXPORT_SYMBOL_GPL(dm_internal_suspend_fast);
3385 void dm_internal_resume_fast(struct mapped_device *md)
3387 if (dm_suspended_md(md) || dm_suspended_internally_md(md))
3393 mutex_unlock(&md->suspend_lock);
3395 EXPORT_SYMBOL_GPL(dm_internal_resume_fast);
3397 /*-----------------------------------------------------------------
3398 * Event notification.
3399 *---------------------------------------------------------------*/
3400 int dm_kobject_uevent(struct mapped_device *md, enum kobject_action action,
3403 char udev_cookie[DM_COOKIE_LENGTH];
3404 char *envp[] = { udev_cookie, NULL };
3407 return kobject_uevent(&disk_to_dev(md->disk)->kobj, action);
3409 snprintf(udev_cookie, DM_COOKIE_LENGTH, "%s=%u",
3410 DM_COOKIE_ENV_VAR_NAME, cookie);
3411 return kobject_uevent_env(&disk_to_dev(md->disk)->kobj,
3416 uint32_t dm_next_uevent_seq(struct mapped_device *md)
3418 return atomic_add_return(1, &md->uevent_seq);
3421 uint32_t dm_get_event_nr(struct mapped_device *md)
3423 return atomic_read(&md->event_nr);
3426 int dm_wait_event(struct mapped_device *md, int event_nr)
3428 return wait_event_interruptible(md->eventq,
3429 (event_nr != atomic_read(&md->event_nr)));
3432 void dm_uevent_add(struct mapped_device *md, struct list_head *elist)
3434 unsigned long flags;
3436 spin_lock_irqsave(&md->uevent_lock, flags);
3437 list_add(elist, &md->uevent_list);
3438 spin_unlock_irqrestore(&md->uevent_lock, flags);
3442 * The gendisk is only valid as long as you have a reference
3445 struct gendisk *dm_disk(struct mapped_device *md)
3449 EXPORT_SYMBOL_GPL(dm_disk);
3451 struct kobject *dm_kobject(struct mapped_device *md)
3453 return &md->kobj_holder.kobj;
3456 struct mapped_device *dm_get_from_kobject(struct kobject *kobj)
3458 struct mapped_device *md;
3460 md = container_of(kobj, struct mapped_device, kobj_holder.kobj);
3462 if (test_bit(DMF_FREEING, &md->flags) ||
3470 int dm_suspended_md(struct mapped_device *md)
3472 return test_bit(DMF_SUSPENDED, &md->flags);
3475 int dm_suspended_internally_md(struct mapped_device *md)
3477 return test_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
3480 int dm_test_deferred_remove_flag(struct mapped_device *md)
3482 return test_bit(DMF_DEFERRED_REMOVE, &md->flags);
3485 int dm_suspended(struct dm_target *ti)
3487 return dm_suspended_md(dm_table_get_md(ti->table));
3489 EXPORT_SYMBOL_GPL(dm_suspended);
3491 int dm_noflush_suspending(struct dm_target *ti)
3493 return __noflush_suspending(dm_table_get_md(ti->table));
3495 EXPORT_SYMBOL_GPL(dm_noflush_suspending);
3497 struct dm_md_mempools *dm_alloc_md_mempools(struct mapped_device *md, unsigned type,
3498 unsigned integrity, unsigned per_bio_data_size)
3500 struct dm_md_mempools *pools = kzalloc(sizeof(*pools), GFP_KERNEL);
3501 struct kmem_cache *cachep = NULL;
3502 unsigned int pool_size = 0;
3503 unsigned int front_pad;
3508 type = filter_md_type(type, md);
3511 case DM_TYPE_BIO_BASED:
3513 pool_size = dm_get_reserved_bio_based_ios();
3514 front_pad = roundup(per_bio_data_size, __alignof__(struct dm_target_io)) + offsetof(struct dm_target_io, clone);
3516 case DM_TYPE_REQUEST_BASED:
3517 cachep = _rq_tio_cache;
3518 pool_size = dm_get_reserved_rq_based_ios();
3519 pools->rq_pool = mempool_create_slab_pool(pool_size, _rq_cache);
3520 if (!pools->rq_pool)
3522 /* fall through to setup remaining rq-based pools */
3523 case DM_TYPE_MQ_REQUEST_BASED:
3525 pool_size = dm_get_reserved_rq_based_ios();
3526 front_pad = offsetof(struct dm_rq_clone_bio_info, clone);
3527 /* per_bio_data_size is not used. See __bind_mempools(). */
3528 WARN_ON(per_bio_data_size != 0);
3535 pools->io_pool = mempool_create_slab_pool(pool_size, cachep);
3536 if (!pools->io_pool)
3540 pools->bs = bioset_create_nobvec(pool_size, front_pad);
3544 if (integrity && bioset_integrity_create(pools->bs, pool_size))
3550 dm_free_md_mempools(pools);
3555 void dm_free_md_mempools(struct dm_md_mempools *pools)
3560 mempool_destroy(pools->io_pool);
3561 mempool_destroy(pools->rq_pool);
3564 bioset_free(pools->bs);
3569 static int dm_pr_register(struct block_device *bdev, u64 old_key, u64 new_key,
3572 struct mapped_device *md = bdev->bd_disk->private_data;
3573 const struct pr_ops *ops;
3577 r = dm_grab_bdev_for_ioctl(md, &bdev, &mode);
3581 ops = bdev->bd_disk->fops->pr_ops;
3582 if (ops && ops->pr_register)
3583 r = ops->pr_register(bdev, old_key, new_key, flags);
3591 static int dm_pr_reserve(struct block_device *bdev, u64 key, enum pr_type type,
3594 struct mapped_device *md = bdev->bd_disk->private_data;
3595 const struct pr_ops *ops;
3599 r = dm_grab_bdev_for_ioctl(md, &bdev, &mode);
3603 ops = bdev->bd_disk->fops->pr_ops;
3604 if (ops && ops->pr_reserve)
3605 r = ops->pr_reserve(bdev, key, type, flags);
3613 static int dm_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
3615 struct mapped_device *md = bdev->bd_disk->private_data;
3616 const struct pr_ops *ops;
3620 r = dm_grab_bdev_for_ioctl(md, &bdev, &mode);
3624 ops = bdev->bd_disk->fops->pr_ops;
3625 if (ops && ops->pr_release)
3626 r = ops->pr_release(bdev, key, type);
3634 static int dm_pr_preempt(struct block_device *bdev, u64 old_key, u64 new_key,
3635 enum pr_type type, bool abort)
3637 struct mapped_device *md = bdev->bd_disk->private_data;
3638 const struct pr_ops *ops;
3642 r = dm_grab_bdev_for_ioctl(md, &bdev, &mode);
3646 ops = bdev->bd_disk->fops->pr_ops;
3647 if (ops && ops->pr_preempt)
3648 r = ops->pr_preempt(bdev, old_key, new_key, type, abort);
3656 static int dm_pr_clear(struct block_device *bdev, u64 key)
3658 struct mapped_device *md = bdev->bd_disk->private_data;
3659 const struct pr_ops *ops;
3663 r = dm_grab_bdev_for_ioctl(md, &bdev, &mode);
3667 ops = bdev->bd_disk->fops->pr_ops;
3668 if (ops && ops->pr_clear)
3669 r = ops->pr_clear(bdev, key);
3677 static const struct pr_ops dm_pr_ops = {
3678 .pr_register = dm_pr_register,
3679 .pr_reserve = dm_pr_reserve,
3680 .pr_release = dm_pr_release,
3681 .pr_preempt = dm_pr_preempt,
3682 .pr_clear = dm_pr_clear,
3685 static const struct block_device_operations dm_blk_dops = {
3686 .open = dm_blk_open,
3687 .release = dm_blk_close,
3688 .ioctl = dm_blk_ioctl,
3689 .getgeo = dm_blk_getgeo,
3690 .pr_ops = &dm_pr_ops,
3691 .owner = THIS_MODULE
3697 module_init(dm_init);
3698 module_exit(dm_exit);
3700 module_param(major, uint, 0);
3701 MODULE_PARM_DESC(major, "The major number of the device mapper");
3703 module_param(reserved_bio_based_ios, uint, S_IRUGO | S_IWUSR);
3704 MODULE_PARM_DESC(reserved_bio_based_ios, "Reserved IOs in bio-based mempools");
3706 module_param(reserved_rq_based_ios, uint, S_IRUGO | S_IWUSR);
3707 MODULE_PARM_DESC(reserved_rq_based_ios, "Reserved IOs in request-based mempools");
3709 module_param(use_blk_mq, bool, S_IRUGO | S_IWUSR);
3710 MODULE_PARM_DESC(use_blk_mq, "Use block multiqueue for request-based DM devices");
3712 module_param(dm_mq_nr_hw_queues, uint, S_IRUGO | S_IWUSR);
3713 MODULE_PARM_DESC(dm_mq_nr_hw_queues, "Number of hardware queues for request-based dm-mq devices");
3715 module_param(dm_mq_queue_depth, uint, S_IRUGO | S_IWUSR);
3716 MODULE_PARM_DESC(dm_mq_queue_depth, "Queue depth for request-based dm-mq devices");
3718 MODULE_DESCRIPTION(DM_NAME " driver");
3719 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
3720 MODULE_LICENSE("GPL");