2 * Copyright (C) 2001, 2002 Sistina Software (UK) Limited.
3 * Copyright (C) 2004-2008 Red Hat, Inc. All rights reserved.
5 * This file is released under the GPL.
11 #include <linux/init.h>
12 #include <linux/module.h>
13 #include <linux/mutex.h>
14 #include <linux/moduleparam.h>
15 #include <linux/blkpg.h>
16 #include <linux/bio.h>
17 #include <linux/mempool.h>
18 #include <linux/slab.h>
19 #include <linux/idr.h>
20 #include <linux/hdreg.h>
21 #include <linux/delay.h>
22 #include <linux/wait.h>
23 #include <linux/kthread.h>
24 #include <linux/ktime.h>
25 #include <linux/elevator.h> /* for rq_end_sector() */
26 #include <linux/blk-mq.h>
28 #include <trace/events/block.h>
30 #define DM_MSG_PREFIX "core"
34 * ratelimit state to be used in DMXXX_LIMIT().
36 DEFINE_RATELIMIT_STATE(dm_ratelimit_state,
37 DEFAULT_RATELIMIT_INTERVAL,
38 DEFAULT_RATELIMIT_BURST);
39 EXPORT_SYMBOL(dm_ratelimit_state);
43 * Cookies are numeric values sent with CHANGE and REMOVE
44 * uevents while resuming, removing or renaming the device.
46 #define DM_COOKIE_ENV_VAR_NAME "DM_COOKIE"
47 #define DM_COOKIE_LENGTH 24
49 static const char *_name = DM_NAME;
51 static unsigned int major = 0;
52 static unsigned int _major = 0;
54 static DEFINE_IDR(_minor_idr);
56 static DEFINE_SPINLOCK(_minor_lock);
58 static void do_deferred_remove(struct work_struct *w);
60 static DECLARE_WORK(deferred_remove_work, do_deferred_remove);
62 static struct workqueue_struct *deferred_remove_workqueue;
66 * One of these is allocated per bio.
69 struct mapped_device *md;
73 unsigned long start_time;
74 spinlock_t endio_lock;
75 struct dm_stats_aux stats_aux;
79 * For request-based dm.
80 * One of these is allocated per request.
82 struct dm_rq_target_io {
83 struct mapped_device *md;
85 struct request *orig, *clone;
86 struct kthread_work work;
92 * For request-based dm - the bio clones we allocate are embedded in these
95 * We allocate these with bio_alloc_bioset, using the front_pad parameter when
96 * the bioset is created - this means the bio has to come at the end of the
99 struct dm_rq_clone_bio_info {
101 struct dm_rq_target_io *tio;
105 union map_info *dm_get_rq_mapinfo(struct request *rq)
107 if (rq && rq->end_io_data)
108 return &((struct dm_rq_target_io *)rq->end_io_data)->info;
111 EXPORT_SYMBOL_GPL(dm_get_rq_mapinfo);
113 #define MINOR_ALLOCED ((void *)-1)
116 * Bits for the md->flags field.
118 #define DMF_BLOCK_IO_FOR_SUSPEND 0
119 #define DMF_SUSPENDED 1
121 #define DMF_FREEING 3
122 #define DMF_DELETING 4
123 #define DMF_NOFLUSH_SUSPENDING 5
124 #define DMF_MERGE_IS_OPTIONAL 6
125 #define DMF_DEFERRED_REMOVE 7
126 #define DMF_SUSPENDED_INTERNALLY 8
129 * A dummy definition to make RCU happy.
130 * struct dm_table should never be dereferenced in this file.
137 * Work processed by per-device workqueue.
139 struct mapped_device {
140 struct srcu_struct io_barrier;
141 struct mutex suspend_lock;
146 * The current mapping.
147 * Use dm_get_live_table{_fast} or take suspend_lock for
150 struct dm_table __rcu *map;
152 struct list_head table_devices;
153 struct mutex table_devices_lock;
157 struct request_queue *queue;
159 /* Protect queue and type against concurrent access. */
160 struct mutex type_lock;
162 struct target_type *immutable_target_type;
164 struct gendisk *disk;
170 * A list of ios that arrived while we were suspended.
173 wait_queue_head_t wait;
174 struct work_struct work;
175 struct bio_list deferred;
176 spinlock_t deferred_lock;
179 * Processing queue (flush)
181 struct workqueue_struct *wq;
184 * io objects are allocated from here.
195 wait_queue_head_t eventq;
197 struct list_head uevent_list;
198 spinlock_t uevent_lock; /* Protect access to uevent_list */
201 * freeze/thaw support require holding onto a super block
203 struct super_block *frozen_sb;
204 struct block_device *bdev;
206 /* forced geometry settings */
207 struct hd_geometry geometry;
209 /* kobject and completion */
210 struct dm_kobject_holder kobj_holder;
212 /* zero-length flush that will be cloned and submitted to targets */
213 struct bio flush_bio;
215 /* the number of internal suspends */
216 unsigned internal_suspend_count;
218 struct dm_stats stats;
220 struct kthread_worker kworker;
221 struct task_struct *kworker_task;
223 /* for request-based merge heuristic in dm_request_fn() */
224 unsigned seq_rq_merge_deadline_usecs;
226 sector_t last_rq_pos;
227 ktime_t last_rq_start_time;
229 /* for blk-mq request-based DM support */
230 struct blk_mq_tag_set tag_set;
234 * For mempools pre-allocation at the table loading time.
236 struct dm_md_mempools {
242 struct table_device {
243 struct list_head list;
245 struct dm_dev dm_dev;
248 #define RESERVED_BIO_BASED_IOS 16
249 #define RESERVED_REQUEST_BASED_IOS 256
250 #define RESERVED_MAX_IOS 1024
251 static struct kmem_cache *_io_cache;
252 static struct kmem_cache *_rq_tio_cache;
253 static struct kmem_cache *_rq_cache;
256 * Bio-based DM's mempools' reserved IOs set by the user.
258 static unsigned reserved_bio_based_ios = RESERVED_BIO_BASED_IOS;
261 * Request-based DM's mempools' reserved IOs set by the user.
263 static unsigned reserved_rq_based_ios = RESERVED_REQUEST_BASED_IOS;
265 static unsigned __dm_get_module_param(unsigned *module_param,
266 unsigned def, unsigned max)
268 unsigned param = ACCESS_ONCE(*module_param);
269 unsigned modified_param = 0;
272 modified_param = def;
273 else if (param > max)
274 modified_param = max;
276 if (modified_param) {
277 (void)cmpxchg(module_param, param, modified_param);
278 param = modified_param;
284 unsigned dm_get_reserved_bio_based_ios(void)
286 return __dm_get_module_param(&reserved_bio_based_ios,
287 RESERVED_BIO_BASED_IOS, RESERVED_MAX_IOS);
289 EXPORT_SYMBOL_GPL(dm_get_reserved_bio_based_ios);
291 unsigned dm_get_reserved_rq_based_ios(void)
293 return __dm_get_module_param(&reserved_rq_based_ios,
294 RESERVED_REQUEST_BASED_IOS, RESERVED_MAX_IOS);
296 EXPORT_SYMBOL_GPL(dm_get_reserved_rq_based_ios);
298 static int __init local_init(void)
302 /* allocate a slab for the dm_ios */
303 _io_cache = KMEM_CACHE(dm_io, 0);
307 _rq_tio_cache = KMEM_CACHE(dm_rq_target_io, 0);
309 goto out_free_io_cache;
311 _rq_cache = kmem_cache_create("dm_clone_request", sizeof(struct request),
312 __alignof__(struct request), 0, NULL);
314 goto out_free_rq_tio_cache;
316 r = dm_uevent_init();
318 goto out_free_rq_cache;
320 deferred_remove_workqueue = alloc_workqueue("kdmremove", WQ_UNBOUND, 1);
321 if (!deferred_remove_workqueue) {
323 goto out_uevent_exit;
327 r = register_blkdev(_major, _name);
329 goto out_free_workqueue;
337 destroy_workqueue(deferred_remove_workqueue);
341 kmem_cache_destroy(_rq_cache);
342 out_free_rq_tio_cache:
343 kmem_cache_destroy(_rq_tio_cache);
345 kmem_cache_destroy(_io_cache);
350 static void local_exit(void)
352 flush_scheduled_work();
353 destroy_workqueue(deferred_remove_workqueue);
355 kmem_cache_destroy(_rq_cache);
356 kmem_cache_destroy(_rq_tio_cache);
357 kmem_cache_destroy(_io_cache);
358 unregister_blkdev(_major, _name);
363 DMINFO("cleaned up");
366 static int (*_inits[])(void) __initdata = {
377 static void (*_exits[])(void) = {
388 static int __init dm_init(void)
390 const int count = ARRAY_SIZE(_inits);
394 for (i = 0; i < count; i++) {
409 static void __exit dm_exit(void)
411 int i = ARRAY_SIZE(_exits);
417 * Should be empty by this point.
419 idr_destroy(&_minor_idr);
423 * Block device functions
425 int dm_deleting_md(struct mapped_device *md)
427 return test_bit(DMF_DELETING, &md->flags);
430 static int dm_blk_open(struct block_device *bdev, fmode_t mode)
432 struct mapped_device *md;
434 spin_lock(&_minor_lock);
436 md = bdev->bd_disk->private_data;
440 if (test_bit(DMF_FREEING, &md->flags) ||
441 dm_deleting_md(md)) {
447 atomic_inc(&md->open_count);
449 spin_unlock(&_minor_lock);
451 return md ? 0 : -ENXIO;
454 static void dm_blk_close(struct gendisk *disk, fmode_t mode)
456 struct mapped_device *md;
458 spin_lock(&_minor_lock);
460 md = disk->private_data;
464 if (atomic_dec_and_test(&md->open_count) &&
465 (test_bit(DMF_DEFERRED_REMOVE, &md->flags)))
466 queue_work(deferred_remove_workqueue, &deferred_remove_work);
470 spin_unlock(&_minor_lock);
473 int dm_open_count(struct mapped_device *md)
475 return atomic_read(&md->open_count);
479 * Guarantees nothing is using the device before it's deleted.
481 int dm_lock_for_deletion(struct mapped_device *md, bool mark_deferred, bool only_deferred)
485 spin_lock(&_minor_lock);
487 if (dm_open_count(md)) {
490 set_bit(DMF_DEFERRED_REMOVE, &md->flags);
491 } else if (only_deferred && !test_bit(DMF_DEFERRED_REMOVE, &md->flags))
494 set_bit(DMF_DELETING, &md->flags);
496 spin_unlock(&_minor_lock);
501 int dm_cancel_deferred_remove(struct mapped_device *md)
505 spin_lock(&_minor_lock);
507 if (test_bit(DMF_DELETING, &md->flags))
510 clear_bit(DMF_DEFERRED_REMOVE, &md->flags);
512 spin_unlock(&_minor_lock);
517 static void do_deferred_remove(struct work_struct *w)
519 dm_deferred_remove();
522 sector_t dm_get_size(struct mapped_device *md)
524 return get_capacity(md->disk);
527 struct request_queue *dm_get_md_queue(struct mapped_device *md)
532 struct dm_stats *dm_get_stats(struct mapped_device *md)
537 static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo)
539 struct mapped_device *md = bdev->bd_disk->private_data;
541 return dm_get_geometry(md, geo);
544 static int dm_blk_ioctl(struct block_device *bdev, fmode_t mode,
545 unsigned int cmd, unsigned long arg)
547 struct mapped_device *md = bdev->bd_disk->private_data;
549 struct dm_table *map;
550 struct dm_target *tgt;
554 map = dm_get_live_table(md, &srcu_idx);
556 if (!map || !dm_table_get_size(map))
559 /* We only support devices that have a single target */
560 if (dm_table_get_num_targets(map) != 1)
563 tgt = dm_table_get_target(map, 0);
564 if (!tgt->type->ioctl)
567 if (dm_suspended_md(md)) {
572 r = tgt->type->ioctl(tgt, cmd, arg);
575 dm_put_live_table(md, srcu_idx);
577 if (r == -ENOTCONN) {
585 static struct dm_io *alloc_io(struct mapped_device *md)
587 return mempool_alloc(md->io_pool, GFP_NOIO);
590 static void free_io(struct mapped_device *md, struct dm_io *io)
592 mempool_free(io, md->io_pool);
595 static void free_tio(struct mapped_device *md, struct dm_target_io *tio)
597 bio_put(&tio->clone);
600 static struct dm_rq_target_io *alloc_rq_tio(struct mapped_device *md,
603 return mempool_alloc(md->io_pool, gfp_mask);
606 static void free_rq_tio(struct dm_rq_target_io *tio)
608 mempool_free(tio, tio->md->io_pool);
611 static struct request *alloc_clone_request(struct mapped_device *md,
614 return mempool_alloc(md->rq_pool, gfp_mask);
617 static void free_clone_request(struct mapped_device *md, struct request *rq)
619 mempool_free(rq, md->rq_pool);
622 static int md_in_flight(struct mapped_device *md)
624 return atomic_read(&md->pending[READ]) +
625 atomic_read(&md->pending[WRITE]);
628 static void start_io_acct(struct dm_io *io)
630 struct mapped_device *md = io->md;
631 struct bio *bio = io->bio;
633 int rw = bio_data_dir(bio);
635 io->start_time = jiffies;
637 cpu = part_stat_lock();
638 part_round_stats(cpu, &dm_disk(md)->part0);
640 atomic_set(&dm_disk(md)->part0.in_flight[rw],
641 atomic_inc_return(&md->pending[rw]));
643 if (unlikely(dm_stats_used(&md->stats)))
644 dm_stats_account_io(&md->stats, bio->bi_rw, bio->bi_iter.bi_sector,
645 bio_sectors(bio), false, 0, &io->stats_aux);
648 static void end_io_acct(struct dm_io *io)
650 struct mapped_device *md = io->md;
651 struct bio *bio = io->bio;
652 unsigned long duration = jiffies - io->start_time;
654 int rw = bio_data_dir(bio);
656 generic_end_io_acct(rw, &dm_disk(md)->part0, io->start_time);
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), true, duration, &io->stats_aux);
663 * After this is decremented the bio must not be touched if it is
666 pending = atomic_dec_return(&md->pending[rw]);
667 atomic_set(&dm_disk(md)->part0.in_flight[rw], pending);
668 pending += atomic_read(&md->pending[rw^0x1]);
670 /* nudge anyone waiting on suspend queue */
676 * Add the bio to the list of deferred io.
678 static void queue_io(struct mapped_device *md, struct bio *bio)
682 spin_lock_irqsave(&md->deferred_lock, flags);
683 bio_list_add(&md->deferred, bio);
684 spin_unlock_irqrestore(&md->deferred_lock, flags);
685 queue_work(md->wq, &md->work);
689 * Everyone (including functions in this file), should use this
690 * function to access the md->map field, and make sure they call
691 * dm_put_live_table() when finished.
693 struct dm_table *dm_get_live_table(struct mapped_device *md, int *srcu_idx) __acquires(md->io_barrier)
695 *srcu_idx = srcu_read_lock(&md->io_barrier);
697 return srcu_dereference(md->map, &md->io_barrier);
700 void dm_put_live_table(struct mapped_device *md, int srcu_idx) __releases(md->io_barrier)
702 srcu_read_unlock(&md->io_barrier, srcu_idx);
705 void dm_sync_table(struct mapped_device *md)
707 synchronize_srcu(&md->io_barrier);
708 synchronize_rcu_expedited();
712 * A fast alternative to dm_get_live_table/dm_put_live_table.
713 * The caller must not block between these two functions.
715 static struct dm_table *dm_get_live_table_fast(struct mapped_device *md) __acquires(RCU)
718 return rcu_dereference(md->map);
721 static void dm_put_live_table_fast(struct mapped_device *md) __releases(RCU)
727 * Open a table device so we can use it as a map destination.
729 static int open_table_device(struct table_device *td, dev_t dev,
730 struct mapped_device *md)
732 static char *_claim_ptr = "I belong to device-mapper";
733 struct block_device *bdev;
737 BUG_ON(td->dm_dev.bdev);
739 bdev = blkdev_get_by_dev(dev, td->dm_dev.mode | FMODE_EXCL, _claim_ptr);
741 return PTR_ERR(bdev);
743 r = bd_link_disk_holder(bdev, dm_disk(md));
745 blkdev_put(bdev, td->dm_dev.mode | FMODE_EXCL);
749 td->dm_dev.bdev = bdev;
754 * Close a table device that we've been using.
756 static void close_table_device(struct table_device *td, struct mapped_device *md)
758 if (!td->dm_dev.bdev)
761 bd_unlink_disk_holder(td->dm_dev.bdev, dm_disk(md));
762 blkdev_put(td->dm_dev.bdev, td->dm_dev.mode | FMODE_EXCL);
763 td->dm_dev.bdev = NULL;
766 static struct table_device *find_table_device(struct list_head *l, dev_t dev,
768 struct table_device *td;
770 list_for_each_entry(td, l, list)
771 if (td->dm_dev.bdev->bd_dev == dev && td->dm_dev.mode == mode)
777 int dm_get_table_device(struct mapped_device *md, dev_t dev, fmode_t mode,
778 struct dm_dev **result) {
780 struct table_device *td;
782 mutex_lock(&md->table_devices_lock);
783 td = find_table_device(&md->table_devices, dev, mode);
785 td = kmalloc(sizeof(*td), GFP_KERNEL);
787 mutex_unlock(&md->table_devices_lock);
791 td->dm_dev.mode = mode;
792 td->dm_dev.bdev = NULL;
794 if ((r = open_table_device(td, dev, md))) {
795 mutex_unlock(&md->table_devices_lock);
800 format_dev_t(td->dm_dev.name, dev);
802 atomic_set(&td->count, 0);
803 list_add(&td->list, &md->table_devices);
805 atomic_inc(&td->count);
806 mutex_unlock(&md->table_devices_lock);
808 *result = &td->dm_dev;
811 EXPORT_SYMBOL_GPL(dm_get_table_device);
813 void dm_put_table_device(struct mapped_device *md, struct dm_dev *d)
815 struct table_device *td = container_of(d, struct table_device, dm_dev);
817 mutex_lock(&md->table_devices_lock);
818 if (atomic_dec_and_test(&td->count)) {
819 close_table_device(td, md);
823 mutex_unlock(&md->table_devices_lock);
825 EXPORT_SYMBOL(dm_put_table_device);
827 static void free_table_devices(struct list_head *devices)
829 struct list_head *tmp, *next;
831 list_for_each_safe(tmp, next, devices) {
832 struct table_device *td = list_entry(tmp, struct table_device, list);
834 DMWARN("dm_destroy: %s still exists with %d references",
835 td->dm_dev.name, atomic_read(&td->count));
841 * Get the geometry associated with a dm device
843 int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo)
851 * Set the geometry of a device.
853 int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo)
855 sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors;
857 if (geo->start > sz) {
858 DMWARN("Start sector is beyond the geometry limits.");
867 /*-----------------------------------------------------------------
869 * A more elegant soln is in the works that uses the queue
870 * merge fn, unfortunately there are a couple of changes to
871 * the block layer that I want to make for this. So in the
872 * interests of getting something for people to use I give
873 * you this clearly demarcated crap.
874 *---------------------------------------------------------------*/
876 static int __noflush_suspending(struct mapped_device *md)
878 return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
882 * Decrements the number of outstanding ios that a bio has been
883 * cloned into, completing the original io if necc.
885 static void dec_pending(struct dm_io *io, int error)
890 struct mapped_device *md = io->md;
892 /* Push-back supersedes any I/O errors */
893 if (unlikely(error)) {
894 spin_lock_irqsave(&io->endio_lock, flags);
895 if (!(io->error > 0 && __noflush_suspending(md)))
897 spin_unlock_irqrestore(&io->endio_lock, flags);
900 if (atomic_dec_and_test(&io->io_count)) {
901 if (io->error == DM_ENDIO_REQUEUE) {
903 * Target requested pushing back the I/O.
905 spin_lock_irqsave(&md->deferred_lock, flags);
906 if (__noflush_suspending(md))
907 bio_list_add_head(&md->deferred, io->bio);
909 /* noflush suspend was interrupted. */
911 spin_unlock_irqrestore(&md->deferred_lock, flags);
914 io_error = io->error;
919 if (io_error == DM_ENDIO_REQUEUE)
922 if ((bio->bi_rw & REQ_FLUSH) && bio->bi_iter.bi_size) {
924 * Preflush done for flush with data, reissue
927 bio->bi_rw &= ~REQ_FLUSH;
930 /* done with normal IO or empty flush */
931 trace_block_bio_complete(md->queue, bio, io_error);
932 bio_endio(bio, io_error);
937 static void disable_write_same(struct mapped_device *md)
939 struct queue_limits *limits = dm_get_queue_limits(md);
941 /* device doesn't really support WRITE SAME, disable it */
942 limits->max_write_same_sectors = 0;
945 static void clone_endio(struct bio *bio, int error)
948 struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone);
949 struct dm_io *io = tio->io;
950 struct mapped_device *md = tio->io->md;
951 dm_endio_fn endio = tio->ti->type->end_io;
953 if (!bio_flagged(bio, BIO_UPTODATE) && !error)
957 r = endio(tio->ti, bio, error);
958 if (r < 0 || r == DM_ENDIO_REQUEUE)
960 * error and requeue request are handled
964 else if (r == DM_ENDIO_INCOMPLETE)
965 /* The target will handle the io */
968 DMWARN("unimplemented target endio return value: %d", r);
973 if (unlikely(r == -EREMOTEIO && (bio->bi_rw & REQ_WRITE_SAME) &&
974 !bdev_get_queue(bio->bi_bdev)->limits.max_write_same_sectors))
975 disable_write_same(md);
978 dec_pending(io, error);
982 * Partial completion handling for request-based dm
984 static void end_clone_bio(struct bio *clone, int error)
986 struct dm_rq_clone_bio_info *info =
987 container_of(clone, struct dm_rq_clone_bio_info, clone);
988 struct dm_rq_target_io *tio = info->tio;
989 struct bio *bio = info->orig;
990 unsigned int nr_bytes = info->orig->bi_iter.bi_size;
996 * An error has already been detected on the request.
997 * Once error occurred, just let clone->end_io() handle
1003 * Don't notice the error to the upper layer yet.
1004 * The error handling decision is made by the target driver,
1005 * when the request is completed.
1012 * I/O for the bio successfully completed.
1013 * Notice the data completion to the upper layer.
1017 * bios are processed from the head of the list.
1018 * So the completing bio should always be rq->bio.
1019 * If it's not, something wrong is happening.
1021 if (tio->orig->bio != bio)
1022 DMERR("bio completion is going in the middle of the request");
1025 * Update the original request.
1026 * Do not use blk_end_request() here, because it may complete
1027 * the original request before the clone, and break the ordering.
1029 blk_update_request(tio->orig, 0, nr_bytes);
1032 static struct dm_rq_target_io *tio_from_request(struct request *rq)
1034 return (rq->q->mq_ops ? blk_mq_rq_to_pdu(rq) : rq->special);
1038 * Don't touch any member of the md after calling this function because
1039 * the md may be freed in dm_put() at the end of this function.
1040 * Or do dm_get() before calling this function and dm_put() later.
1042 static void rq_completed(struct mapped_device *md, int rw, bool run_queue)
1044 int nr_requests_pending;
1046 atomic_dec(&md->pending[rw]);
1048 /* nudge anyone waiting on suspend queue */
1049 nr_requests_pending = md_in_flight(md);
1050 if (!nr_requests_pending)
1054 * Run this off this callpath, as drivers could invoke end_io while
1055 * inside their request_fn (and holding the queue lock). Calling
1056 * back into ->request_fn() could deadlock attempting to grab the
1060 if (md->queue->mq_ops)
1061 blk_mq_run_hw_queues(md->queue, true);
1062 else if (!nr_requests_pending ||
1063 (nr_requests_pending >= md->queue->nr_congestion_on))
1064 blk_run_queue_async(md->queue);
1068 * dm_put() must be at the end of this function. See the comment above
1073 static void free_rq_clone(struct request *clone)
1075 struct dm_rq_target_io *tio = clone->end_io_data;
1076 struct mapped_device *md = tio->md;
1078 blk_rq_unprep_clone(clone);
1080 if (clone->q->mq_ops)
1081 tio->ti->type->release_clone_rq(clone);
1082 else if (!md->queue->mq_ops)
1083 /* request_fn queue stacked on request_fn queue(s) */
1084 free_clone_request(md, clone);
1086 if (!md->queue->mq_ops)
1091 * Complete the clone and the original request.
1092 * Must be called without clone's queue lock held,
1093 * see end_clone_request() for more details.
1095 static void dm_end_request(struct request *clone, int error)
1097 int rw = rq_data_dir(clone);
1098 struct dm_rq_target_io *tio = clone->end_io_data;
1099 struct mapped_device *md = tio->md;
1100 struct request *rq = tio->orig;
1102 if (rq->cmd_type == REQ_TYPE_BLOCK_PC) {
1103 rq->errors = clone->errors;
1104 rq->resid_len = clone->resid_len;
1108 * We are using the sense buffer of the original
1110 * So setting the length of the sense data is enough.
1112 rq->sense_len = clone->sense_len;
1115 free_rq_clone(clone);
1117 blk_end_request_all(rq, error);
1119 blk_mq_end_request(rq, error);
1120 rq_completed(md, rw, true);
1123 static void dm_unprep_request(struct request *rq)
1125 struct dm_rq_target_io *tio = tio_from_request(rq);
1126 struct request *clone = tio->clone;
1128 if (!rq->q->mq_ops) {
1130 rq->cmd_flags &= ~REQ_DONTPREP;
1134 free_rq_clone(clone);
1138 * Requeue the original request of a clone.
1140 static void old_requeue_request(struct request *rq)
1142 struct request_queue *q = rq->q;
1143 unsigned long flags;
1145 spin_lock_irqsave(q->queue_lock, flags);
1146 blk_requeue_request(q, rq);
1147 spin_unlock_irqrestore(q->queue_lock, flags);
1150 static void dm_requeue_unmapped_original_request(struct mapped_device *md,
1153 int rw = rq_data_dir(rq);
1155 dm_unprep_request(rq);
1158 old_requeue_request(rq);
1160 blk_mq_requeue_request(rq);
1161 blk_mq_kick_requeue_list(rq->q);
1164 rq_completed(md, rw, false);
1167 static void dm_requeue_unmapped_request(struct request *clone)
1169 struct dm_rq_target_io *tio = clone->end_io_data;
1171 dm_requeue_unmapped_original_request(tio->md, tio->orig);
1174 static void old_stop_queue(struct request_queue *q)
1176 unsigned long flags;
1178 if (blk_queue_stopped(q))
1181 spin_lock_irqsave(q->queue_lock, flags);
1183 spin_unlock_irqrestore(q->queue_lock, flags);
1186 static void stop_queue(struct request_queue *q)
1191 blk_mq_stop_hw_queues(q);
1194 static void old_start_queue(struct request_queue *q)
1196 unsigned long flags;
1198 spin_lock_irqsave(q->queue_lock, flags);
1199 if (blk_queue_stopped(q))
1201 spin_unlock_irqrestore(q->queue_lock, flags);
1204 static void start_queue(struct request_queue *q)
1209 blk_mq_start_stopped_hw_queues(q, true);
1212 static void dm_done(struct request *clone, int error, bool mapped)
1215 struct dm_rq_target_io *tio = clone->end_io_data;
1216 dm_request_endio_fn rq_end_io = NULL;
1219 rq_end_io = tio->ti->type->rq_end_io;
1221 if (mapped && rq_end_io)
1222 r = rq_end_io(tio->ti, clone, error, &tio->info);
1225 if (unlikely(r == -EREMOTEIO && (clone->cmd_flags & REQ_WRITE_SAME) &&
1226 !clone->q->limits.max_write_same_sectors))
1227 disable_write_same(tio->md);
1230 /* The target wants to complete the I/O */
1231 dm_end_request(clone, r);
1232 else if (r == DM_ENDIO_INCOMPLETE)
1233 /* The target will handle the I/O */
1235 else if (r == DM_ENDIO_REQUEUE)
1236 /* The target wants to requeue the I/O */
1237 dm_requeue_unmapped_request(clone);
1239 DMWARN("unimplemented target endio return value: %d", r);
1245 * Request completion handler for request-based dm
1247 static void dm_softirq_done(struct request *rq)
1250 struct dm_rq_target_io *tio = tio_from_request(rq);
1251 struct request *clone = tio->clone;
1255 rw = rq_data_dir(rq);
1256 if (!rq->q->mq_ops) {
1257 blk_end_request_all(rq, tio->error);
1258 rq_completed(tio->md, rw, false);
1261 blk_mq_end_request(rq, tio->error);
1262 rq_completed(tio->md, rw, false);
1267 if (rq->cmd_flags & REQ_FAILED)
1270 dm_done(clone, tio->error, mapped);
1274 * Complete the clone and the original request with the error status
1275 * through softirq context.
1277 static void dm_complete_request(struct request *rq, int error)
1279 struct dm_rq_target_io *tio = tio_from_request(rq);
1282 blk_complete_request(rq);
1286 * Complete the not-mapped clone and the original request with the error status
1287 * through softirq context.
1288 * Target's rq_end_io() function isn't called.
1289 * This may be used when the target's map_rq() or clone_and_map_rq() functions fail.
1291 static void dm_kill_unmapped_request(struct request *rq, int error)
1293 rq->cmd_flags |= REQ_FAILED;
1294 dm_complete_request(rq, error);
1298 * Called with the clone's queue lock held (for non-blk-mq)
1300 static void end_clone_request(struct request *clone, int error)
1302 struct dm_rq_target_io *tio = clone->end_io_data;
1304 if (!clone->q->mq_ops) {
1306 * For just cleaning up the information of the queue in which
1307 * the clone was dispatched.
1308 * The clone is *NOT* freed actually here because it is alloced
1309 * from dm own mempool (REQ_ALLOCED isn't set).
1311 __blk_put_request(clone->q, clone);
1315 * Actual request completion is done in a softirq context which doesn't
1316 * hold the clone's queue lock. Otherwise, deadlock could occur because:
1317 * - another request may be submitted by the upper level driver
1318 * of the stacking during the completion
1319 * - the submission which requires queue lock may be done
1320 * against this clone's queue
1322 dm_complete_request(tio->orig, error);
1326 * Return maximum size of I/O possible at the supplied sector up to the current
1329 static sector_t max_io_len_target_boundary(sector_t sector, struct dm_target *ti)
1331 sector_t target_offset = dm_target_offset(ti, sector);
1333 return ti->len - target_offset;
1336 static sector_t max_io_len(sector_t sector, struct dm_target *ti)
1338 sector_t len = max_io_len_target_boundary(sector, ti);
1339 sector_t offset, max_len;
1342 * Does the target need to split even further?
1344 if (ti->max_io_len) {
1345 offset = dm_target_offset(ti, sector);
1346 if (unlikely(ti->max_io_len & (ti->max_io_len - 1)))
1347 max_len = sector_div(offset, ti->max_io_len);
1349 max_len = offset & (ti->max_io_len - 1);
1350 max_len = ti->max_io_len - max_len;
1359 int dm_set_target_max_io_len(struct dm_target *ti, sector_t len)
1361 if (len > UINT_MAX) {
1362 DMERR("Specified maximum size of target IO (%llu) exceeds limit (%u)",
1363 (unsigned long long)len, UINT_MAX);
1364 ti->error = "Maximum size of target IO is too large";
1368 ti->max_io_len = (uint32_t) len;
1372 EXPORT_SYMBOL_GPL(dm_set_target_max_io_len);
1375 * A target may call dm_accept_partial_bio only from the map routine. It is
1376 * allowed for all bio types except REQ_FLUSH.
1378 * dm_accept_partial_bio informs the dm that the target only wants to process
1379 * additional n_sectors sectors of the bio and the rest of the data should be
1380 * sent in a next bio.
1382 * A diagram that explains the arithmetics:
1383 * +--------------------+---------------+-------+
1385 * +--------------------+---------------+-------+
1387 * <-------------- *tio->len_ptr --------------->
1388 * <------- bi_size ------->
1391 * Region 1 was already iterated over with bio_advance or similar function.
1392 * (it may be empty if the target doesn't use bio_advance)
1393 * Region 2 is the remaining bio size that the target wants to process.
1394 * (it may be empty if region 1 is non-empty, although there is no reason
1396 * The target requires that region 3 is to be sent in the next bio.
1398 * If the target wants to receive multiple copies of the bio (via num_*bios, etc),
1399 * the partially processed part (the sum of regions 1+2) must be the same for all
1400 * copies of the bio.
1402 void dm_accept_partial_bio(struct bio *bio, unsigned n_sectors)
1404 struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone);
1405 unsigned bi_size = bio->bi_iter.bi_size >> SECTOR_SHIFT;
1406 BUG_ON(bio->bi_rw & REQ_FLUSH);
1407 BUG_ON(bi_size > *tio->len_ptr);
1408 BUG_ON(n_sectors > bi_size);
1409 *tio->len_ptr -= bi_size - n_sectors;
1410 bio->bi_iter.bi_size = n_sectors << SECTOR_SHIFT;
1412 EXPORT_SYMBOL_GPL(dm_accept_partial_bio);
1414 static void __map_bio(struct dm_target_io *tio)
1418 struct mapped_device *md;
1419 struct bio *clone = &tio->clone;
1420 struct dm_target *ti = tio->ti;
1422 clone->bi_end_io = clone_endio;
1425 * Map the clone. If r == 0 we don't need to do
1426 * anything, the target has assumed ownership of
1429 atomic_inc(&tio->io->io_count);
1430 sector = clone->bi_iter.bi_sector;
1431 r = ti->type->map(ti, clone);
1432 if (r == DM_MAPIO_REMAPPED) {
1433 /* the bio has been remapped so dispatch it */
1435 trace_block_bio_remap(bdev_get_queue(clone->bi_bdev), clone,
1436 tio->io->bio->bi_bdev->bd_dev, sector);
1438 generic_make_request(clone);
1439 } else if (r < 0 || r == DM_MAPIO_REQUEUE) {
1440 /* error the io and bail out, or requeue it if needed */
1442 dec_pending(tio->io, r);
1445 DMWARN("unimplemented target map return value: %d", r);
1451 struct mapped_device *md;
1452 struct dm_table *map;
1456 unsigned sector_count;
1459 static void bio_setup_sector(struct bio *bio, sector_t sector, unsigned len)
1461 bio->bi_iter.bi_sector = sector;
1462 bio->bi_iter.bi_size = to_bytes(len);
1466 * Creates a bio that consists of range of complete bvecs.
1468 static void clone_bio(struct dm_target_io *tio, struct bio *bio,
1469 sector_t sector, unsigned len)
1471 struct bio *clone = &tio->clone;
1473 __bio_clone_fast(clone, bio);
1475 if (bio_integrity(bio))
1476 bio_integrity_clone(clone, bio, GFP_NOIO);
1478 bio_advance(clone, to_bytes(sector - clone->bi_iter.bi_sector));
1479 clone->bi_iter.bi_size = to_bytes(len);
1481 if (bio_integrity(bio))
1482 bio_integrity_trim(clone, 0, len);
1485 static struct dm_target_io *alloc_tio(struct clone_info *ci,
1486 struct dm_target *ti,
1487 unsigned target_bio_nr)
1489 struct dm_target_io *tio;
1492 clone = bio_alloc_bioset(GFP_NOIO, 0, ci->md->bs);
1493 tio = container_of(clone, struct dm_target_io, clone);
1497 tio->target_bio_nr = target_bio_nr;
1502 static void __clone_and_map_simple_bio(struct clone_info *ci,
1503 struct dm_target *ti,
1504 unsigned target_bio_nr, unsigned *len)
1506 struct dm_target_io *tio = alloc_tio(ci, ti, target_bio_nr);
1507 struct bio *clone = &tio->clone;
1511 __bio_clone_fast(clone, ci->bio);
1513 bio_setup_sector(clone, ci->sector, *len);
1518 static void __send_duplicate_bios(struct clone_info *ci, struct dm_target *ti,
1519 unsigned num_bios, unsigned *len)
1521 unsigned target_bio_nr;
1523 for (target_bio_nr = 0; target_bio_nr < num_bios; target_bio_nr++)
1524 __clone_and_map_simple_bio(ci, ti, target_bio_nr, len);
1527 static int __send_empty_flush(struct clone_info *ci)
1529 unsigned target_nr = 0;
1530 struct dm_target *ti;
1532 BUG_ON(bio_has_data(ci->bio));
1533 while ((ti = dm_table_get_target(ci->map, target_nr++)))
1534 __send_duplicate_bios(ci, ti, ti->num_flush_bios, NULL);
1539 static void __clone_and_map_data_bio(struct clone_info *ci, struct dm_target *ti,
1540 sector_t sector, unsigned *len)
1542 struct bio *bio = ci->bio;
1543 struct dm_target_io *tio;
1544 unsigned target_bio_nr;
1545 unsigned num_target_bios = 1;
1548 * Does the target want to receive duplicate copies of the bio?
1550 if (bio_data_dir(bio) == WRITE && ti->num_write_bios)
1551 num_target_bios = ti->num_write_bios(ti, bio);
1553 for (target_bio_nr = 0; target_bio_nr < num_target_bios; target_bio_nr++) {
1554 tio = alloc_tio(ci, ti, target_bio_nr);
1556 clone_bio(tio, bio, sector, *len);
1561 typedef unsigned (*get_num_bios_fn)(struct dm_target *ti);
1563 static unsigned get_num_discard_bios(struct dm_target *ti)
1565 return ti->num_discard_bios;
1568 static unsigned get_num_write_same_bios(struct dm_target *ti)
1570 return ti->num_write_same_bios;
1573 typedef bool (*is_split_required_fn)(struct dm_target *ti);
1575 static bool is_split_required_for_discard(struct dm_target *ti)
1577 return ti->split_discard_bios;
1580 static int __send_changing_extent_only(struct clone_info *ci,
1581 get_num_bios_fn get_num_bios,
1582 is_split_required_fn is_split_required)
1584 struct dm_target *ti;
1589 ti = dm_table_find_target(ci->map, ci->sector);
1590 if (!dm_target_is_valid(ti))
1594 * Even though the device advertised support for this type of
1595 * request, that does not mean every target supports it, and
1596 * reconfiguration might also have changed that since the
1597 * check was performed.
1599 num_bios = get_num_bios ? get_num_bios(ti) : 0;
1603 if (is_split_required && !is_split_required(ti))
1604 len = min((sector_t)ci->sector_count, max_io_len_target_boundary(ci->sector, ti));
1606 len = min((sector_t)ci->sector_count, max_io_len(ci->sector, ti));
1608 __send_duplicate_bios(ci, ti, num_bios, &len);
1611 } while (ci->sector_count -= len);
1616 static int __send_discard(struct clone_info *ci)
1618 return __send_changing_extent_only(ci, get_num_discard_bios,
1619 is_split_required_for_discard);
1622 static int __send_write_same(struct clone_info *ci)
1624 return __send_changing_extent_only(ci, get_num_write_same_bios, NULL);
1628 * Select the correct strategy for processing a non-flush bio.
1630 static int __split_and_process_non_flush(struct clone_info *ci)
1632 struct bio *bio = ci->bio;
1633 struct dm_target *ti;
1636 if (unlikely(bio->bi_rw & REQ_DISCARD))
1637 return __send_discard(ci);
1638 else if (unlikely(bio->bi_rw & REQ_WRITE_SAME))
1639 return __send_write_same(ci);
1641 ti = dm_table_find_target(ci->map, ci->sector);
1642 if (!dm_target_is_valid(ti))
1645 len = min_t(sector_t, max_io_len(ci->sector, ti), ci->sector_count);
1647 __clone_and_map_data_bio(ci, ti, ci->sector, &len);
1650 ci->sector_count -= len;
1656 * Entry point to split a bio into clones and submit them to the targets.
1658 static void __split_and_process_bio(struct mapped_device *md,
1659 struct dm_table *map, struct bio *bio)
1661 struct clone_info ci;
1664 if (unlikely(!map)) {
1671 ci.io = alloc_io(md);
1673 atomic_set(&ci.io->io_count, 1);
1676 spin_lock_init(&ci.io->endio_lock);
1677 ci.sector = bio->bi_iter.bi_sector;
1679 start_io_acct(ci.io);
1681 if (bio->bi_rw & REQ_FLUSH) {
1682 ci.bio = &ci.md->flush_bio;
1683 ci.sector_count = 0;
1684 error = __send_empty_flush(&ci);
1685 /* dec_pending submits any data associated with flush */
1688 ci.sector_count = bio_sectors(bio);
1689 while (ci.sector_count && !error)
1690 error = __split_and_process_non_flush(&ci);
1693 /* drop the extra reference count */
1694 dec_pending(ci.io, error);
1696 /*-----------------------------------------------------------------
1698 *---------------------------------------------------------------*/
1700 static int dm_merge_bvec(struct request_queue *q,
1701 struct bvec_merge_data *bvm,
1702 struct bio_vec *biovec)
1704 struct mapped_device *md = q->queuedata;
1705 struct dm_table *map = dm_get_live_table_fast(md);
1706 struct dm_target *ti;
1707 sector_t max_sectors;
1713 ti = dm_table_find_target(map, bvm->bi_sector);
1714 if (!dm_target_is_valid(ti))
1718 * Find maximum amount of I/O that won't need splitting
1720 max_sectors = min(max_io_len(bvm->bi_sector, ti),
1721 (sector_t) queue_max_sectors(q));
1722 max_size = (max_sectors << SECTOR_SHIFT) - bvm->bi_size;
1723 if (unlikely(max_size < 0)) /* this shouldn't _ever_ happen */
1727 * merge_bvec_fn() returns number of bytes
1728 * it can accept at this offset
1729 * max is precomputed maximal io size
1731 if (max_size && ti->type->merge)
1732 max_size = ti->type->merge(ti, bvm, biovec, max_size);
1734 * If the target doesn't support merge method and some of the devices
1735 * provided their merge_bvec method (we know this by looking for the
1736 * max_hw_sectors that dm_set_device_limits may set), then we can't
1737 * allow bios with multiple vector entries. So always set max_size
1738 * to 0, and the code below allows just one page.
1740 else if (queue_max_hw_sectors(q) <= PAGE_SIZE >> 9)
1744 dm_put_live_table_fast(md);
1746 * Always allow an entire first page
1748 if (max_size <= biovec->bv_len && !(bvm->bi_size >> SECTOR_SHIFT))
1749 max_size = biovec->bv_len;
1755 * The request function that just remaps the bio built up by
1758 static void dm_make_request(struct request_queue *q, struct bio *bio)
1760 int rw = bio_data_dir(bio);
1761 struct mapped_device *md = q->queuedata;
1763 struct dm_table *map;
1765 map = dm_get_live_table(md, &srcu_idx);
1767 generic_start_io_acct(rw, bio_sectors(bio), &dm_disk(md)->part0);
1769 /* if we're suspended, we have to queue this io for later */
1770 if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))) {
1771 dm_put_live_table(md, srcu_idx);
1773 if (bio_rw(bio) != READA)
1780 __split_and_process_bio(md, map, bio);
1781 dm_put_live_table(md, srcu_idx);
1785 int dm_request_based(struct mapped_device *md)
1787 return blk_queue_stackable(md->queue);
1790 static void dm_dispatch_clone_request(struct request *clone, struct request *rq)
1794 if (blk_queue_io_stat(clone->q))
1795 clone->cmd_flags |= REQ_IO_STAT;
1797 clone->start_time = jiffies;
1798 r = blk_insert_cloned_request(clone->q, clone);
1800 /* must complete clone in terms of original request */
1801 dm_complete_request(rq, r);
1804 static int dm_rq_bio_constructor(struct bio *bio, struct bio *bio_orig,
1807 struct dm_rq_target_io *tio = data;
1808 struct dm_rq_clone_bio_info *info =
1809 container_of(bio, struct dm_rq_clone_bio_info, clone);
1811 info->orig = bio_orig;
1813 bio->bi_end_io = end_clone_bio;
1818 static int setup_clone(struct request *clone, struct request *rq,
1819 struct dm_rq_target_io *tio, gfp_t gfp_mask)
1823 r = blk_rq_prep_clone(clone, rq, tio->md->bs, gfp_mask,
1824 dm_rq_bio_constructor, tio);
1828 clone->cmd = rq->cmd;
1829 clone->cmd_len = rq->cmd_len;
1830 clone->sense = rq->sense;
1831 clone->end_io = end_clone_request;
1832 clone->end_io_data = tio;
1839 static struct request *clone_rq(struct request *rq, struct mapped_device *md,
1840 struct dm_rq_target_io *tio, gfp_t gfp_mask)
1843 * Do not allocate a clone if tio->clone was already set
1844 * (see: dm_mq_queue_rq).
1846 bool alloc_clone = !tio->clone;
1847 struct request *clone;
1850 clone = alloc_clone_request(md, gfp_mask);
1856 blk_rq_init(NULL, clone);
1857 if (setup_clone(clone, rq, tio, gfp_mask)) {
1860 free_clone_request(md, clone);
1867 static void map_tio_request(struct kthread_work *work);
1869 static void init_tio(struct dm_rq_target_io *tio, struct request *rq,
1870 struct mapped_device *md)
1877 memset(&tio->info, 0, sizeof(tio->info));
1878 if (md->kworker_task)
1879 init_kthread_work(&tio->work, map_tio_request);
1882 static struct dm_rq_target_io *prep_tio(struct request *rq,
1883 struct mapped_device *md, gfp_t gfp_mask)
1885 struct dm_rq_target_io *tio;
1887 struct dm_table *table;
1889 tio = alloc_rq_tio(md, gfp_mask);
1893 init_tio(tio, rq, md);
1895 table = dm_get_live_table(md, &srcu_idx);
1896 if (!dm_table_mq_request_based(table)) {
1897 if (!clone_rq(rq, md, tio, gfp_mask)) {
1898 dm_put_live_table(md, srcu_idx);
1903 dm_put_live_table(md, srcu_idx);
1909 * Called with the queue lock held.
1911 static int dm_prep_fn(struct request_queue *q, struct request *rq)
1913 struct mapped_device *md = q->queuedata;
1914 struct dm_rq_target_io *tio;
1916 if (unlikely(rq->special)) {
1917 DMWARN("Already has something in rq->special.");
1918 return BLKPREP_KILL;
1921 tio = prep_tio(rq, md, GFP_ATOMIC);
1923 return BLKPREP_DEFER;
1926 rq->cmd_flags |= REQ_DONTPREP;
1933 * 0 : the request has been processed
1934 * DM_MAPIO_REQUEUE : the original request needs to be requeued
1935 * < 0 : the request was completed due to failure
1937 static int map_request(struct dm_rq_target_io *tio, struct request *rq,
1938 struct mapped_device *md)
1941 struct dm_target *ti = tio->ti;
1942 struct request *clone = NULL;
1946 r = ti->type->map_rq(ti, clone, &tio->info);
1948 r = ti->type->clone_and_map_rq(ti, rq, &tio->info, &clone);
1950 /* The target wants to complete the I/O */
1951 dm_kill_unmapped_request(rq, r);
1955 return DM_MAPIO_REQUEUE;
1956 if (setup_clone(clone, rq, tio, GFP_ATOMIC)) {
1958 ti->type->release_clone_rq(clone);
1959 return DM_MAPIO_REQUEUE;
1964 case DM_MAPIO_SUBMITTED:
1965 /* The target has taken the I/O to submit by itself later */
1967 case DM_MAPIO_REMAPPED:
1968 /* The target has remapped the I/O so dispatch it */
1969 trace_block_rq_remap(clone->q, clone, disk_devt(dm_disk(md)),
1971 dm_dispatch_clone_request(clone, rq);
1973 case DM_MAPIO_REQUEUE:
1974 /* The target wants to requeue the I/O */
1975 dm_requeue_unmapped_request(clone);
1979 DMWARN("unimplemented target map return value: %d", r);
1983 /* The target wants to complete the I/O */
1984 dm_kill_unmapped_request(rq, r);
1991 static void map_tio_request(struct kthread_work *work)
1993 struct dm_rq_target_io *tio = container_of(work, struct dm_rq_target_io, work);
1994 struct request *rq = tio->orig;
1995 struct mapped_device *md = tio->md;
1997 if (map_request(tio, rq, md) == DM_MAPIO_REQUEUE)
1998 dm_requeue_unmapped_original_request(md, rq);
2001 static void dm_start_request(struct mapped_device *md, struct request *orig)
2003 if (!orig->q->mq_ops)
2004 blk_start_request(orig);
2006 blk_mq_start_request(orig);
2007 atomic_inc(&md->pending[rq_data_dir(orig)]);
2009 if (md->seq_rq_merge_deadline_usecs) {
2010 md->last_rq_pos = rq_end_sector(orig);
2011 md->last_rq_rw = rq_data_dir(orig);
2012 md->last_rq_start_time = ktime_get();
2016 * Hold the md reference here for the in-flight I/O.
2017 * We can't rely on the reference count by device opener,
2018 * because the device may be closed during the request completion
2019 * when all bios are completed.
2020 * See the comment in rq_completed() too.
2025 #define MAX_SEQ_RQ_MERGE_DEADLINE_USECS 100000
2027 ssize_t dm_attr_rq_based_seq_io_merge_deadline_show(struct mapped_device *md, char *buf)
2029 return sprintf(buf, "%u\n", md->seq_rq_merge_deadline_usecs);
2032 ssize_t dm_attr_rq_based_seq_io_merge_deadline_store(struct mapped_device *md,
2033 const char *buf, size_t count)
2037 if (!dm_request_based(md))
2040 if (kstrtouint(buf, 10, &deadline))
2043 if (deadline > MAX_SEQ_RQ_MERGE_DEADLINE_USECS)
2044 deadline = MAX_SEQ_RQ_MERGE_DEADLINE_USECS;
2046 md->seq_rq_merge_deadline_usecs = deadline;
2051 static bool dm_request_peeked_before_merge_deadline(struct mapped_device *md)
2053 ktime_t kt_deadline;
2055 if (!md->seq_rq_merge_deadline_usecs)
2058 kt_deadline = ns_to_ktime((u64)md->seq_rq_merge_deadline_usecs * NSEC_PER_USEC);
2059 kt_deadline = ktime_add_safe(md->last_rq_start_time, kt_deadline);
2061 return !ktime_after(ktime_get(), kt_deadline);
2065 * q->request_fn for request-based dm.
2066 * Called with the queue lock held.
2068 static void dm_request_fn(struct request_queue *q)
2070 struct mapped_device *md = q->queuedata;
2072 struct dm_table *map = dm_get_live_table(md, &srcu_idx);
2073 struct dm_target *ti;
2075 struct dm_rq_target_io *tio;
2079 * For suspend, check blk_queue_stopped() and increment
2080 * ->pending within a single queue_lock not to increment the
2081 * number of in-flight I/Os after the queue is stopped in
2084 while (!blk_queue_stopped(q)) {
2085 rq = blk_peek_request(q);
2089 /* always use block 0 to find the target for flushes for now */
2091 if (!(rq->cmd_flags & REQ_FLUSH))
2092 pos = blk_rq_pos(rq);
2094 ti = dm_table_find_target(map, pos);
2095 if (!dm_target_is_valid(ti)) {
2097 * Must perform setup, that rq_completed() requires,
2098 * before calling dm_kill_unmapped_request
2100 DMERR_LIMIT("request attempted access beyond the end of device");
2101 dm_start_request(md, rq);
2102 dm_kill_unmapped_request(rq, -EIO);
2106 if (dm_request_peeked_before_merge_deadline(md) &&
2107 md_in_flight(md) && rq->bio && rq->bio->bi_vcnt == 1 &&
2108 md->last_rq_pos == pos && md->last_rq_rw == rq_data_dir(rq))
2111 if (ti->type->busy && ti->type->busy(ti))
2114 dm_start_request(md, rq);
2116 tio = tio_from_request(rq);
2117 /* Establish tio->ti before queuing work (map_tio_request) */
2119 queue_kthread_work(&md->kworker, &tio->work);
2120 BUG_ON(!irqs_disabled());
2126 blk_delay_queue(q, HZ / 100);
2128 dm_put_live_table(md, srcu_idx);
2131 static int dm_any_congested(void *congested_data, int bdi_bits)
2134 struct mapped_device *md = congested_data;
2135 struct dm_table *map;
2137 if (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
2138 map = dm_get_live_table_fast(md);
2141 * Request-based dm cares about only own queue for
2142 * the query about congestion status of request_queue
2144 if (dm_request_based(md))
2145 r = md->queue->backing_dev_info.state &
2148 r = dm_table_any_congested(map, bdi_bits);
2150 dm_put_live_table_fast(md);
2156 /*-----------------------------------------------------------------
2157 * An IDR is used to keep track of allocated minor numbers.
2158 *---------------------------------------------------------------*/
2159 static void free_minor(int minor)
2161 spin_lock(&_minor_lock);
2162 idr_remove(&_minor_idr, minor);
2163 spin_unlock(&_minor_lock);
2167 * See if the device with a specific minor # is free.
2169 static int specific_minor(int minor)
2173 if (minor >= (1 << MINORBITS))
2176 idr_preload(GFP_KERNEL);
2177 spin_lock(&_minor_lock);
2179 r = idr_alloc(&_minor_idr, MINOR_ALLOCED, minor, minor + 1, GFP_NOWAIT);
2181 spin_unlock(&_minor_lock);
2184 return r == -ENOSPC ? -EBUSY : r;
2188 static int next_free_minor(int *minor)
2192 idr_preload(GFP_KERNEL);
2193 spin_lock(&_minor_lock);
2195 r = idr_alloc(&_minor_idr, MINOR_ALLOCED, 0, 1 << MINORBITS, GFP_NOWAIT);
2197 spin_unlock(&_minor_lock);
2205 static const struct block_device_operations dm_blk_dops;
2207 static void dm_wq_work(struct work_struct *work);
2209 static void dm_init_md_queue(struct mapped_device *md)
2212 * Request-based dm devices cannot be stacked on top of bio-based dm
2213 * devices. The type of this dm device may not have been decided yet.
2214 * The type is decided at the first table loading time.
2215 * To prevent problematic device stacking, clear the queue flag
2216 * for request stacking support until then.
2218 * This queue is new, so no concurrency on the queue_flags.
2220 queue_flag_clear_unlocked(QUEUE_FLAG_STACKABLE, md->queue);
2223 static void dm_init_old_md_queue(struct mapped_device *md)
2225 dm_init_md_queue(md);
2228 * Initialize aspects of queue that aren't relevant for blk-mq
2230 md->queue->queuedata = md;
2231 md->queue->backing_dev_info.congested_fn = dm_any_congested;
2232 md->queue->backing_dev_info.congested_data = md;
2234 blk_queue_bounce_limit(md->queue, BLK_BOUNCE_ANY);
2238 * Allocate and initialise a blank device with a given minor.
2240 static struct mapped_device *alloc_dev(int minor)
2243 struct mapped_device *md = kzalloc(sizeof(*md), GFP_KERNEL);
2247 DMWARN("unable to allocate device, out of memory.");
2251 if (!try_module_get(THIS_MODULE))
2252 goto bad_module_get;
2254 /* get a minor number for the dev */
2255 if (minor == DM_ANY_MINOR)
2256 r = next_free_minor(&minor);
2258 r = specific_minor(minor);
2262 r = init_srcu_struct(&md->io_barrier);
2264 goto bad_io_barrier;
2266 md->type = DM_TYPE_NONE;
2267 mutex_init(&md->suspend_lock);
2268 mutex_init(&md->type_lock);
2269 mutex_init(&md->table_devices_lock);
2270 spin_lock_init(&md->deferred_lock);
2271 atomic_set(&md->holders, 1);
2272 atomic_set(&md->open_count, 0);
2273 atomic_set(&md->event_nr, 0);
2274 atomic_set(&md->uevent_seq, 0);
2275 INIT_LIST_HEAD(&md->uevent_list);
2276 INIT_LIST_HEAD(&md->table_devices);
2277 spin_lock_init(&md->uevent_lock);
2279 md->queue = blk_alloc_queue(GFP_KERNEL);
2283 dm_init_md_queue(md);
2285 md->disk = alloc_disk(1);
2289 atomic_set(&md->pending[0], 0);
2290 atomic_set(&md->pending[1], 0);
2291 init_waitqueue_head(&md->wait);
2292 INIT_WORK(&md->work, dm_wq_work);
2293 init_waitqueue_head(&md->eventq);
2294 init_completion(&md->kobj_holder.completion);
2295 md->kworker_task = NULL;
2297 md->disk->major = _major;
2298 md->disk->first_minor = minor;
2299 md->disk->fops = &dm_blk_dops;
2300 md->disk->queue = md->queue;
2301 md->disk->private_data = md;
2302 sprintf(md->disk->disk_name, "dm-%d", minor);
2304 format_dev_t(md->name, MKDEV(_major, minor));
2306 md->wq = alloc_workqueue("kdmflush", WQ_MEM_RECLAIM, 0);
2310 md->bdev = bdget_disk(md->disk, 0);
2314 bio_init(&md->flush_bio);
2315 md->flush_bio.bi_bdev = md->bdev;
2316 md->flush_bio.bi_rw = WRITE_FLUSH;
2318 dm_stats_init(&md->stats);
2320 /* Populate the mapping, nobody knows we exist yet */
2321 spin_lock(&_minor_lock);
2322 old_md = idr_replace(&_minor_idr, md, minor);
2323 spin_unlock(&_minor_lock);
2325 BUG_ON(old_md != MINOR_ALLOCED);
2330 destroy_workqueue(md->wq);
2332 del_gendisk(md->disk);
2335 blk_cleanup_queue(md->queue);
2337 cleanup_srcu_struct(&md->io_barrier);
2341 module_put(THIS_MODULE);
2347 static void unlock_fs(struct mapped_device *md);
2349 static void free_dev(struct mapped_device *md)
2351 int minor = MINOR(disk_devt(md->disk));
2352 bool using_blk_mq = !!md->queue->mq_ops;
2355 destroy_workqueue(md->wq);
2357 if (md->kworker_task)
2358 kthread_stop(md->kworker_task);
2360 mempool_destroy(md->io_pool);
2362 mempool_destroy(md->rq_pool);
2364 bioset_free(md->bs);
2366 cleanup_srcu_struct(&md->io_barrier);
2367 free_table_devices(&md->table_devices);
2368 dm_stats_cleanup(&md->stats);
2370 spin_lock(&_minor_lock);
2371 md->disk->private_data = NULL;
2372 spin_unlock(&_minor_lock);
2373 if (blk_get_integrity(md->disk))
2374 blk_integrity_unregister(md->disk);
2375 del_gendisk(md->disk);
2377 blk_cleanup_queue(md->queue);
2379 blk_mq_free_tag_set(&md->tag_set);
2383 module_put(THIS_MODULE);
2387 static void __bind_mempools(struct mapped_device *md, struct dm_table *t)
2389 struct dm_md_mempools *p = dm_table_get_md_mempools(t);
2391 if (md->io_pool && md->bs) {
2392 /* The md already has necessary mempools. */
2393 if (dm_table_get_type(t) == DM_TYPE_BIO_BASED) {
2395 * Reload bioset because front_pad may have changed
2396 * because a different table was loaded.
2398 bioset_free(md->bs);
2403 * There's no need to reload with request-based dm
2404 * because the size of front_pad doesn't change.
2405 * Note for future: If you are to reload bioset,
2406 * prep-ed requests in the queue may refer
2407 * to bio from the old bioset, so you must walk
2408 * through the queue to unprep.
2413 BUG_ON(!p || md->io_pool || md->rq_pool || md->bs);
2415 md->io_pool = p->io_pool;
2417 md->rq_pool = p->rq_pool;
2423 /* mempool bind completed, no longer need any mempools in the table */
2424 dm_table_free_md_mempools(t);
2428 * Bind a table to the device.
2430 static void event_callback(void *context)
2432 unsigned long flags;
2434 struct mapped_device *md = (struct mapped_device *) context;
2436 spin_lock_irqsave(&md->uevent_lock, flags);
2437 list_splice_init(&md->uevent_list, &uevents);
2438 spin_unlock_irqrestore(&md->uevent_lock, flags);
2440 dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj);
2442 atomic_inc(&md->event_nr);
2443 wake_up(&md->eventq);
2447 * Protected by md->suspend_lock obtained by dm_swap_table().
2449 static void __set_size(struct mapped_device *md, sector_t size)
2451 set_capacity(md->disk, size);
2453 i_size_write(md->bdev->bd_inode, (loff_t)size << SECTOR_SHIFT);
2457 * Return 1 if the queue has a compulsory merge_bvec_fn function.
2459 * If this function returns 0, then the device is either a non-dm
2460 * device without a merge_bvec_fn, or it is a dm device that is
2461 * able to split any bios it receives that are too big.
2463 int dm_queue_merge_is_compulsory(struct request_queue *q)
2465 struct mapped_device *dev_md;
2467 if (!q->merge_bvec_fn)
2470 if (q->make_request_fn == dm_make_request) {
2471 dev_md = q->queuedata;
2472 if (test_bit(DMF_MERGE_IS_OPTIONAL, &dev_md->flags))
2479 static int dm_device_merge_is_compulsory(struct dm_target *ti,
2480 struct dm_dev *dev, sector_t start,
2481 sector_t len, void *data)
2483 struct block_device *bdev = dev->bdev;
2484 struct request_queue *q = bdev_get_queue(bdev);
2486 return dm_queue_merge_is_compulsory(q);
2490 * Return 1 if it is acceptable to ignore merge_bvec_fn based
2491 * on the properties of the underlying devices.
2493 static int dm_table_merge_is_optional(struct dm_table *table)
2496 struct dm_target *ti;
2498 while (i < dm_table_get_num_targets(table)) {
2499 ti = dm_table_get_target(table, i++);
2501 if (ti->type->iterate_devices &&
2502 ti->type->iterate_devices(ti, dm_device_merge_is_compulsory, NULL))
2510 * Returns old map, which caller must destroy.
2512 static struct dm_table *__bind(struct mapped_device *md, struct dm_table *t,
2513 struct queue_limits *limits)
2515 struct dm_table *old_map;
2516 struct request_queue *q = md->queue;
2518 int merge_is_optional;
2520 size = dm_table_get_size(t);
2523 * Wipe any geometry if the size of the table changed.
2525 if (size != dm_get_size(md))
2526 memset(&md->geometry, 0, sizeof(md->geometry));
2528 __set_size(md, size);
2530 dm_table_event_callback(t, event_callback, md);
2533 * The queue hasn't been stopped yet, if the old table type wasn't
2534 * for request-based during suspension. So stop it to prevent
2535 * I/O mapping before resume.
2536 * This must be done before setting the queue restrictions,
2537 * because request-based dm may be run just after the setting.
2539 if (dm_table_request_based(t))
2542 __bind_mempools(md, t);
2544 merge_is_optional = dm_table_merge_is_optional(t);
2546 old_map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2547 rcu_assign_pointer(md->map, t);
2548 md->immutable_target_type = dm_table_get_immutable_target_type(t);
2550 dm_table_set_restrictions(t, q, limits);
2551 if (merge_is_optional)
2552 set_bit(DMF_MERGE_IS_OPTIONAL, &md->flags);
2554 clear_bit(DMF_MERGE_IS_OPTIONAL, &md->flags);
2562 * Returns unbound table for the caller to free.
2564 static struct dm_table *__unbind(struct mapped_device *md)
2566 struct dm_table *map = rcu_dereference_protected(md->map, 1);
2571 dm_table_event_callback(map, NULL, NULL);
2572 RCU_INIT_POINTER(md->map, NULL);
2579 * Constructor for a new device.
2581 int dm_create(int minor, struct mapped_device **result)
2583 struct mapped_device *md;
2585 md = alloc_dev(minor);
2596 * Functions to manage md->type.
2597 * All are required to hold md->type_lock.
2599 void dm_lock_md_type(struct mapped_device *md)
2601 mutex_lock(&md->type_lock);
2604 void dm_unlock_md_type(struct mapped_device *md)
2606 mutex_unlock(&md->type_lock);
2609 void dm_set_md_type(struct mapped_device *md, unsigned type)
2611 BUG_ON(!mutex_is_locked(&md->type_lock));
2615 unsigned dm_get_md_type(struct mapped_device *md)
2617 BUG_ON(!mutex_is_locked(&md->type_lock));
2621 struct target_type *dm_get_immutable_target_type(struct mapped_device *md)
2623 return md->immutable_target_type;
2627 * The queue_limits are only valid as long as you have a reference
2630 struct queue_limits *dm_get_queue_limits(struct mapped_device *md)
2632 BUG_ON(!atomic_read(&md->holders));
2633 return &md->queue->limits;
2635 EXPORT_SYMBOL_GPL(dm_get_queue_limits);
2637 static void init_rq_based_worker_thread(struct mapped_device *md)
2639 /* Initialize the request-based DM worker thread */
2640 init_kthread_worker(&md->kworker);
2641 md->kworker_task = kthread_run(kthread_worker_fn, &md->kworker,
2642 "kdmwork-%s", dm_device_name(md));
2646 * Fully initialize a request-based queue (->elevator, ->request_fn, etc).
2648 static int dm_init_request_based_queue(struct mapped_device *md)
2650 struct request_queue *q = NULL;
2652 if (md->queue->elevator)
2655 /* Fully initialize the queue */
2656 q = blk_init_allocated_queue(md->queue, dm_request_fn, NULL);
2660 /* disable dm_request_fn's merge heuristic by default */
2661 md->seq_rq_merge_deadline_usecs = 0;
2664 dm_init_old_md_queue(md);
2665 blk_queue_softirq_done(md->queue, dm_softirq_done);
2666 blk_queue_prep_rq(md->queue, dm_prep_fn);
2668 init_rq_based_worker_thread(md);
2670 elv_register_queue(md->queue);
2675 static int dm_mq_init_request(void *data, struct request *rq,
2676 unsigned int hctx_idx, unsigned int request_idx,
2677 unsigned int numa_node)
2679 struct mapped_device *md = data;
2680 struct dm_rq_target_io *tio = blk_mq_rq_to_pdu(rq);
2683 * Must initialize md member of tio, otherwise it won't
2684 * be available in dm_mq_queue_rq.
2691 static int dm_mq_queue_rq(struct blk_mq_hw_ctx *hctx,
2692 const struct blk_mq_queue_data *bd)
2694 struct request *rq = bd->rq;
2695 struct dm_rq_target_io *tio = blk_mq_rq_to_pdu(rq);
2696 struct mapped_device *md = tio->md;
2698 struct dm_table *map = dm_get_live_table(md, &srcu_idx);
2699 struct dm_target *ti;
2702 /* always use block 0 to find the target for flushes for now */
2704 if (!(rq->cmd_flags & REQ_FLUSH))
2705 pos = blk_rq_pos(rq);
2707 ti = dm_table_find_target(map, pos);
2708 if (!dm_target_is_valid(ti)) {
2709 dm_put_live_table(md, srcu_idx);
2710 DMERR_LIMIT("request attempted access beyond the end of device");
2712 * Must perform setup, that rq_completed() requires,
2713 * before returning BLK_MQ_RQ_QUEUE_ERROR
2715 dm_start_request(md, rq);
2716 return BLK_MQ_RQ_QUEUE_ERROR;
2718 dm_put_live_table(md, srcu_idx);
2720 if (ti->type->busy && ti->type->busy(ti))
2721 return BLK_MQ_RQ_QUEUE_BUSY;
2723 dm_start_request(md, rq);
2725 /* Init tio using md established in .init_request */
2726 init_tio(tio, rq, md);
2729 * Establish tio->ti before queuing work (map_tio_request)
2730 * or making direct call to map_request().
2734 /* Clone the request if underlying devices aren't blk-mq */
2735 if (dm_table_get_type(map) == DM_TYPE_REQUEST_BASED) {
2736 /* clone request is allocated at the end of the pdu */
2737 tio->clone = (void *)blk_mq_rq_to_pdu(rq) + sizeof(struct dm_rq_target_io);
2738 if (!clone_rq(rq, md, tio, GFP_ATOMIC))
2739 return BLK_MQ_RQ_QUEUE_BUSY;
2740 queue_kthread_work(&md->kworker, &tio->work);
2742 /* Direct call is fine since .queue_rq allows allocations */
2743 if (map_request(tio, rq, md) == DM_MAPIO_REQUEUE)
2744 dm_requeue_unmapped_original_request(md, rq);
2747 return BLK_MQ_RQ_QUEUE_OK;
2750 static struct blk_mq_ops dm_mq_ops = {
2751 .queue_rq = dm_mq_queue_rq,
2752 .map_queue = blk_mq_map_queue,
2753 .complete = dm_softirq_done,
2754 .init_request = dm_mq_init_request,
2757 static int dm_init_request_based_blk_mq_queue(struct mapped_device *md)
2759 unsigned md_type = dm_get_md_type(md);
2760 struct request_queue *q;
2763 memset(&md->tag_set, 0, sizeof(md->tag_set));
2764 md->tag_set.ops = &dm_mq_ops;
2765 md->tag_set.queue_depth = BLKDEV_MAX_RQ;
2766 md->tag_set.numa_node = NUMA_NO_NODE;
2767 md->tag_set.flags = BLK_MQ_F_SHOULD_MERGE | BLK_MQ_F_SG_MERGE;
2768 md->tag_set.nr_hw_queues = 1;
2769 if (md_type == DM_TYPE_REQUEST_BASED) {
2770 /* make the memory for non-blk-mq clone part of the pdu */
2771 md->tag_set.cmd_size = sizeof(struct dm_rq_target_io) + sizeof(struct request);
2773 md->tag_set.cmd_size = sizeof(struct dm_rq_target_io);
2774 md->tag_set.driver_data = md;
2776 err = blk_mq_alloc_tag_set(&md->tag_set);
2780 q = blk_mq_init_allocated_queue(&md->tag_set, md->queue);
2786 dm_init_md_queue(md);
2788 /* backfill 'mq' sysfs registration normally done in blk_register_queue */
2789 blk_mq_register_disk(md->disk);
2791 if (md_type == DM_TYPE_REQUEST_BASED)
2792 init_rq_based_worker_thread(md);
2797 blk_mq_free_tag_set(&md->tag_set);
2802 * Setup the DM device's queue based on md's type
2804 int dm_setup_md_queue(struct mapped_device *md)
2807 unsigned md_type = dm_get_md_type(md);
2810 case DM_TYPE_REQUEST_BASED:
2811 r = dm_init_request_based_queue(md);
2813 DMWARN("Cannot initialize queue for request-based mapped device");
2817 case DM_TYPE_MQ_REQUEST_BASED:
2818 r = dm_init_request_based_blk_mq_queue(md);
2820 DMWARN("Cannot initialize queue for request-based blk-mq mapped device");
2824 case DM_TYPE_BIO_BASED:
2825 dm_init_old_md_queue(md);
2826 blk_queue_make_request(md->queue, dm_make_request);
2827 blk_queue_merge_bvec(md->queue, dm_merge_bvec);
2834 struct mapped_device *dm_get_md(dev_t dev)
2836 struct mapped_device *md;
2837 unsigned minor = MINOR(dev);
2839 if (MAJOR(dev) != _major || minor >= (1 << MINORBITS))
2842 spin_lock(&_minor_lock);
2844 md = idr_find(&_minor_idr, minor);
2846 if ((md == MINOR_ALLOCED ||
2847 (MINOR(disk_devt(dm_disk(md))) != minor) ||
2848 dm_deleting_md(md) ||
2849 test_bit(DMF_FREEING, &md->flags))) {
2857 spin_unlock(&_minor_lock);
2861 EXPORT_SYMBOL_GPL(dm_get_md);
2863 void *dm_get_mdptr(struct mapped_device *md)
2865 return md->interface_ptr;
2868 void dm_set_mdptr(struct mapped_device *md, void *ptr)
2870 md->interface_ptr = ptr;
2873 void dm_get(struct mapped_device *md)
2875 atomic_inc(&md->holders);
2876 BUG_ON(test_bit(DMF_FREEING, &md->flags));
2879 int dm_hold(struct mapped_device *md)
2881 spin_lock(&_minor_lock);
2882 if (test_bit(DMF_FREEING, &md->flags)) {
2883 spin_unlock(&_minor_lock);
2887 spin_unlock(&_minor_lock);
2890 EXPORT_SYMBOL_GPL(dm_hold);
2892 const char *dm_device_name(struct mapped_device *md)
2896 EXPORT_SYMBOL_GPL(dm_device_name);
2898 static void __dm_destroy(struct mapped_device *md, bool wait)
2900 struct dm_table *map;
2905 map = dm_get_live_table(md, &srcu_idx);
2907 spin_lock(&_minor_lock);
2908 idr_replace(&_minor_idr, MINOR_ALLOCED, MINOR(disk_devt(dm_disk(md))));
2909 set_bit(DMF_FREEING, &md->flags);
2910 spin_unlock(&_minor_lock);
2912 if (dm_request_based(md) && md->kworker_task)
2913 flush_kthread_worker(&md->kworker);
2916 * Take suspend_lock so that presuspend and postsuspend methods
2917 * do not race with internal suspend.
2919 mutex_lock(&md->suspend_lock);
2920 if (!dm_suspended_md(md)) {
2921 dm_table_presuspend_targets(map);
2922 dm_table_postsuspend_targets(map);
2924 mutex_unlock(&md->suspend_lock);
2926 /* dm_put_live_table must be before msleep, otherwise deadlock is possible */
2927 dm_put_live_table(md, srcu_idx);
2930 * Rare, but there may be I/O requests still going to complete,
2931 * for example. Wait for all references to disappear.
2932 * No one should increment the reference count of the mapped_device,
2933 * after the mapped_device state becomes DMF_FREEING.
2936 while (atomic_read(&md->holders))
2938 else if (atomic_read(&md->holders))
2939 DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)",
2940 dm_device_name(md), atomic_read(&md->holders));
2943 dm_table_destroy(__unbind(md));
2947 void dm_destroy(struct mapped_device *md)
2949 __dm_destroy(md, true);
2952 void dm_destroy_immediate(struct mapped_device *md)
2954 __dm_destroy(md, false);
2957 void dm_put(struct mapped_device *md)
2959 atomic_dec(&md->holders);
2961 EXPORT_SYMBOL_GPL(dm_put);
2963 static int dm_wait_for_completion(struct mapped_device *md, int interruptible)
2966 DECLARE_WAITQUEUE(wait, current);
2968 add_wait_queue(&md->wait, &wait);
2971 set_current_state(interruptible);
2973 if (!md_in_flight(md))
2976 if (interruptible == TASK_INTERRUPTIBLE &&
2977 signal_pending(current)) {
2984 set_current_state(TASK_RUNNING);
2986 remove_wait_queue(&md->wait, &wait);
2992 * Process the deferred bios
2994 static void dm_wq_work(struct work_struct *work)
2996 struct mapped_device *md = container_of(work, struct mapped_device,
3000 struct dm_table *map;
3002 map = dm_get_live_table(md, &srcu_idx);
3004 while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
3005 spin_lock_irq(&md->deferred_lock);
3006 c = bio_list_pop(&md->deferred);
3007 spin_unlock_irq(&md->deferred_lock);
3012 if (dm_request_based(md))
3013 generic_make_request(c);
3015 __split_and_process_bio(md, map, c);
3018 dm_put_live_table(md, srcu_idx);
3021 static void dm_queue_flush(struct mapped_device *md)
3023 clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
3024 smp_mb__after_atomic();
3025 queue_work(md->wq, &md->work);
3029 * Swap in a new table, returning the old one for the caller to destroy.
3031 struct dm_table *dm_swap_table(struct mapped_device *md, struct dm_table *table)
3033 struct dm_table *live_map = NULL, *map = ERR_PTR(-EINVAL);
3034 struct queue_limits limits;
3037 mutex_lock(&md->suspend_lock);
3039 /* device must be suspended */
3040 if (!dm_suspended_md(md))
3044 * If the new table has no data devices, retain the existing limits.
3045 * This helps multipath with queue_if_no_path if all paths disappear,
3046 * then new I/O is queued based on these limits, and then some paths
3049 if (dm_table_has_no_data_devices(table)) {
3050 live_map = dm_get_live_table_fast(md);
3052 limits = md->queue->limits;
3053 dm_put_live_table_fast(md);
3057 r = dm_calculate_queue_limits(table, &limits);
3064 map = __bind(md, table, &limits);
3067 mutex_unlock(&md->suspend_lock);
3072 * Functions to lock and unlock any filesystem running on the
3075 static int lock_fs(struct mapped_device *md)
3079 WARN_ON(md->frozen_sb);
3081 md->frozen_sb = freeze_bdev(md->bdev);
3082 if (IS_ERR(md->frozen_sb)) {
3083 r = PTR_ERR(md->frozen_sb);
3084 md->frozen_sb = NULL;
3088 set_bit(DMF_FROZEN, &md->flags);
3093 static void unlock_fs(struct mapped_device *md)
3095 if (!test_bit(DMF_FROZEN, &md->flags))
3098 thaw_bdev(md->bdev, md->frozen_sb);
3099 md->frozen_sb = NULL;
3100 clear_bit(DMF_FROZEN, &md->flags);
3104 * If __dm_suspend returns 0, the device is completely quiescent
3105 * now. There is no request-processing activity. All new requests
3106 * are being added to md->deferred list.
3108 * Caller must hold md->suspend_lock
3110 static int __dm_suspend(struct mapped_device *md, struct dm_table *map,
3111 unsigned suspend_flags, int interruptible)
3113 bool do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG;
3114 bool noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG;
3118 * DMF_NOFLUSH_SUSPENDING must be set before presuspend.
3119 * This flag is cleared before dm_suspend returns.
3122 set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
3125 * This gets reverted if there's an error later and the targets
3126 * provide the .presuspend_undo hook.
3128 dm_table_presuspend_targets(map);
3131 * Flush I/O to the device.
3132 * Any I/O submitted after lock_fs() may not be flushed.
3133 * noflush takes precedence over do_lockfs.
3134 * (lock_fs() flushes I/Os and waits for them to complete.)
3136 if (!noflush && do_lockfs) {
3139 dm_table_presuspend_undo_targets(map);
3145 * Here we must make sure that no processes are submitting requests
3146 * to target drivers i.e. no one may be executing
3147 * __split_and_process_bio. This is called from dm_request and
3150 * To get all processes out of __split_and_process_bio in dm_request,
3151 * we take the write lock. To prevent any process from reentering
3152 * __split_and_process_bio from dm_request and quiesce the thread
3153 * (dm_wq_work), we set BMF_BLOCK_IO_FOR_SUSPEND and call
3154 * flush_workqueue(md->wq).
3156 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
3158 synchronize_srcu(&md->io_barrier);
3161 * Stop md->queue before flushing md->wq in case request-based
3162 * dm defers requests to md->wq from md->queue.
3164 if (dm_request_based(md)) {
3165 stop_queue(md->queue);
3166 if (md->kworker_task)
3167 flush_kthread_worker(&md->kworker);
3170 flush_workqueue(md->wq);
3173 * At this point no more requests are entering target request routines.
3174 * We call dm_wait_for_completion to wait for all existing requests
3177 r = dm_wait_for_completion(md, interruptible);
3180 clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
3182 synchronize_srcu(&md->io_barrier);
3184 /* were we interrupted ? */
3188 if (dm_request_based(md))
3189 start_queue(md->queue);
3192 dm_table_presuspend_undo_targets(map);
3193 /* pushback list is already flushed, so skip flush */
3200 * We need to be able to change a mapping table under a mounted
3201 * filesystem. For example we might want to move some data in
3202 * the background. Before the table can be swapped with
3203 * dm_bind_table, dm_suspend must be called to flush any in
3204 * flight bios and ensure that any further io gets deferred.
3207 * Suspend mechanism in request-based dm.
3209 * 1. Flush all I/Os by lock_fs() if needed.
3210 * 2. Stop dispatching any I/O by stopping the request_queue.
3211 * 3. Wait for all in-flight I/Os to be completed or requeued.
3213 * To abort suspend, start the request_queue.
3215 int dm_suspend(struct mapped_device *md, unsigned suspend_flags)
3217 struct dm_table *map = NULL;
3221 mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING);
3223 if (dm_suspended_md(md)) {
3228 if (dm_suspended_internally_md(md)) {
3229 /* already internally suspended, wait for internal resume */
3230 mutex_unlock(&md->suspend_lock);
3231 r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE);
3237 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
3239 r = __dm_suspend(md, map, suspend_flags, TASK_INTERRUPTIBLE);
3243 set_bit(DMF_SUSPENDED, &md->flags);
3245 dm_table_postsuspend_targets(map);
3248 mutex_unlock(&md->suspend_lock);
3252 static int __dm_resume(struct mapped_device *md, struct dm_table *map)
3255 int r = dm_table_resume_targets(map);
3263 * Flushing deferred I/Os must be done after targets are resumed
3264 * so that mapping of targets can work correctly.
3265 * Request-based dm is queueing the deferred I/Os in its request_queue.
3267 if (dm_request_based(md))
3268 start_queue(md->queue);
3275 int dm_resume(struct mapped_device *md)
3278 struct dm_table *map = NULL;
3281 mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING);
3283 if (!dm_suspended_md(md))
3286 if (dm_suspended_internally_md(md)) {
3287 /* already internally suspended, wait for internal resume */
3288 mutex_unlock(&md->suspend_lock);
3289 r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE);
3295 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
3296 if (!map || !dm_table_get_size(map))
3299 r = __dm_resume(md, map);
3303 clear_bit(DMF_SUSPENDED, &md->flags);
3307 mutex_unlock(&md->suspend_lock);
3313 * Internal suspend/resume works like userspace-driven suspend. It waits
3314 * until all bios finish and prevents issuing new bios to the target drivers.
3315 * It may be used only from the kernel.
3318 static void __dm_internal_suspend(struct mapped_device *md, unsigned suspend_flags)
3320 struct dm_table *map = NULL;
3322 if (md->internal_suspend_count++)
3323 return; /* nested internal suspend */
3325 if (dm_suspended_md(md)) {
3326 set_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
3327 return; /* nest suspend */
3330 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
3333 * Using TASK_UNINTERRUPTIBLE because only NOFLUSH internal suspend is
3334 * supported. Properly supporting a TASK_INTERRUPTIBLE internal suspend
3335 * would require changing .presuspend to return an error -- avoid this
3336 * until there is a need for more elaborate variants of internal suspend.
3338 (void) __dm_suspend(md, map, suspend_flags, TASK_UNINTERRUPTIBLE);
3340 set_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
3342 dm_table_postsuspend_targets(map);
3345 static void __dm_internal_resume(struct mapped_device *md)
3347 BUG_ON(!md->internal_suspend_count);
3349 if (--md->internal_suspend_count)
3350 return; /* resume from nested internal suspend */
3352 if (dm_suspended_md(md))
3353 goto done; /* resume from nested suspend */
3356 * NOTE: existing callers don't need to call dm_table_resume_targets
3357 * (which may fail -- so best to avoid it for now by passing NULL map)
3359 (void) __dm_resume(md, NULL);
3362 clear_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
3363 smp_mb__after_atomic();
3364 wake_up_bit(&md->flags, DMF_SUSPENDED_INTERNALLY);
3367 void dm_internal_suspend_noflush(struct mapped_device *md)
3369 mutex_lock(&md->suspend_lock);
3370 __dm_internal_suspend(md, DM_SUSPEND_NOFLUSH_FLAG);
3371 mutex_unlock(&md->suspend_lock);
3373 EXPORT_SYMBOL_GPL(dm_internal_suspend_noflush);
3375 void dm_internal_resume(struct mapped_device *md)
3377 mutex_lock(&md->suspend_lock);
3378 __dm_internal_resume(md);
3379 mutex_unlock(&md->suspend_lock);
3381 EXPORT_SYMBOL_GPL(dm_internal_resume);
3384 * Fast variants of internal suspend/resume hold md->suspend_lock,
3385 * which prevents interaction with userspace-driven suspend.
3388 void dm_internal_suspend_fast(struct mapped_device *md)
3390 mutex_lock(&md->suspend_lock);
3391 if (dm_suspended_md(md) || dm_suspended_internally_md(md))
3394 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
3395 synchronize_srcu(&md->io_barrier);
3396 flush_workqueue(md->wq);
3397 dm_wait_for_completion(md, TASK_UNINTERRUPTIBLE);
3399 EXPORT_SYMBOL_GPL(dm_internal_suspend_fast);
3401 void dm_internal_resume_fast(struct mapped_device *md)
3403 if (dm_suspended_md(md) || dm_suspended_internally_md(md))
3409 mutex_unlock(&md->suspend_lock);
3411 EXPORT_SYMBOL_GPL(dm_internal_resume_fast);
3413 /*-----------------------------------------------------------------
3414 * Event notification.
3415 *---------------------------------------------------------------*/
3416 int dm_kobject_uevent(struct mapped_device *md, enum kobject_action action,
3419 char udev_cookie[DM_COOKIE_LENGTH];
3420 char *envp[] = { udev_cookie, NULL };
3423 return kobject_uevent(&disk_to_dev(md->disk)->kobj, action);
3425 snprintf(udev_cookie, DM_COOKIE_LENGTH, "%s=%u",
3426 DM_COOKIE_ENV_VAR_NAME, cookie);
3427 return kobject_uevent_env(&disk_to_dev(md->disk)->kobj,
3432 uint32_t dm_next_uevent_seq(struct mapped_device *md)
3434 return atomic_add_return(1, &md->uevent_seq);
3437 uint32_t dm_get_event_nr(struct mapped_device *md)
3439 return atomic_read(&md->event_nr);
3442 int dm_wait_event(struct mapped_device *md, int event_nr)
3444 return wait_event_interruptible(md->eventq,
3445 (event_nr != atomic_read(&md->event_nr)));
3448 void dm_uevent_add(struct mapped_device *md, struct list_head *elist)
3450 unsigned long flags;
3452 spin_lock_irqsave(&md->uevent_lock, flags);
3453 list_add(elist, &md->uevent_list);
3454 spin_unlock_irqrestore(&md->uevent_lock, flags);
3458 * The gendisk is only valid as long as you have a reference
3461 struct gendisk *dm_disk(struct mapped_device *md)
3466 struct kobject *dm_kobject(struct mapped_device *md)
3468 return &md->kobj_holder.kobj;
3471 struct mapped_device *dm_get_from_kobject(struct kobject *kobj)
3473 struct mapped_device *md;
3475 md = container_of(kobj, struct mapped_device, kobj_holder.kobj);
3477 if (test_bit(DMF_FREEING, &md->flags) ||
3485 int dm_suspended_md(struct mapped_device *md)
3487 return test_bit(DMF_SUSPENDED, &md->flags);
3490 int dm_suspended_internally_md(struct mapped_device *md)
3492 return test_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
3495 int dm_test_deferred_remove_flag(struct mapped_device *md)
3497 return test_bit(DMF_DEFERRED_REMOVE, &md->flags);
3500 int dm_suspended(struct dm_target *ti)
3502 return dm_suspended_md(dm_table_get_md(ti->table));
3504 EXPORT_SYMBOL_GPL(dm_suspended);
3506 int dm_noflush_suspending(struct dm_target *ti)
3508 return __noflush_suspending(dm_table_get_md(ti->table));
3510 EXPORT_SYMBOL_GPL(dm_noflush_suspending);
3512 struct dm_md_mempools *dm_alloc_md_mempools(unsigned type, unsigned integrity, unsigned per_bio_data_size)
3514 struct dm_md_mempools *pools = kzalloc(sizeof(*pools), GFP_KERNEL);
3515 struct kmem_cache *cachep;
3516 unsigned int pool_size = 0;
3517 unsigned int front_pad;
3523 case DM_TYPE_BIO_BASED:
3525 pool_size = dm_get_reserved_bio_based_ios();
3526 front_pad = roundup(per_bio_data_size, __alignof__(struct dm_target_io)) + offsetof(struct dm_target_io, clone);
3528 case DM_TYPE_REQUEST_BASED:
3529 pool_size = dm_get_reserved_rq_based_ios();
3530 pools->rq_pool = mempool_create_slab_pool(pool_size, _rq_cache);
3531 if (!pools->rq_pool)
3533 /* fall through to setup remaining rq-based pools */
3534 case DM_TYPE_MQ_REQUEST_BASED:
3535 cachep = _rq_tio_cache;
3537 pool_size = dm_get_reserved_rq_based_ios();
3538 front_pad = offsetof(struct dm_rq_clone_bio_info, clone);
3539 /* per_bio_data_size is not used. See __bind_mempools(). */
3540 WARN_ON(per_bio_data_size != 0);
3546 pools->io_pool = mempool_create_slab_pool(pool_size, cachep);
3547 if (!pools->io_pool)
3550 pools->bs = bioset_create_nobvec(pool_size, front_pad);
3554 if (integrity && bioset_integrity_create(pools->bs, pool_size))
3560 dm_free_md_mempools(pools);
3565 void dm_free_md_mempools(struct dm_md_mempools *pools)
3571 mempool_destroy(pools->io_pool);
3574 mempool_destroy(pools->rq_pool);
3577 bioset_free(pools->bs);
3582 static const struct block_device_operations dm_blk_dops = {
3583 .open = dm_blk_open,
3584 .release = dm_blk_close,
3585 .ioctl = dm_blk_ioctl,
3586 .getgeo = dm_blk_getgeo,
3587 .owner = THIS_MODULE
3593 module_init(dm_init);
3594 module_exit(dm_exit);
3596 module_param(major, uint, 0);
3597 MODULE_PARM_DESC(major, "The major number of the device mapper");
3599 module_param(reserved_bio_based_ios, uint, S_IRUGO | S_IWUSR);
3600 MODULE_PARM_DESC(reserved_bio_based_ios, "Reserved IOs in bio-based mempools");
3602 module_param(reserved_rq_based_ios, uint, S_IRUGO | S_IWUSR);
3603 MODULE_PARM_DESC(reserved_rq_based_ios, "Reserved IOs in request-based mempools");
3605 MODULE_DESCRIPTION(DM_NAME " driver");
3606 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
3607 MODULE_LICENSE("GPL");