2 * Copyright (C) 2001, 2002 Sistina Software (UK) Limited.
3 * Copyright (C) 2004-2008 Red Hat, Inc. All rights reserved.
5 * This file is released under the GPL.
11 #include <linux/init.h>
12 #include <linux/module.h>
13 #include <linux/mutex.h>
14 #include <linux/moduleparam.h>
15 #include <linux/blkpg.h>
16 #include <linux/bio.h>
17 #include <linux/mempool.h>
18 #include <linux/slab.h>
19 #include <linux/idr.h>
20 #include <linux/hdreg.h>
21 #include <linux/delay.h>
22 #include <linux/wait.h>
23 #include <linux/kthread.h>
24 #include <linux/ktime.h>
25 #include <linux/elevator.h> /* for rq_end_sector() */
26 #include <linux/blk-mq.h>
28 #include <trace/events/block.h>
30 #define DM_MSG_PREFIX "core"
34 * ratelimit state to be used in DMXXX_LIMIT().
36 DEFINE_RATELIMIT_STATE(dm_ratelimit_state,
37 DEFAULT_RATELIMIT_INTERVAL,
38 DEFAULT_RATELIMIT_BURST);
39 EXPORT_SYMBOL(dm_ratelimit_state);
43 * Cookies are numeric values sent with CHANGE and REMOVE
44 * uevents while resuming, removing or renaming the device.
46 #define DM_COOKIE_ENV_VAR_NAME "DM_COOKIE"
47 #define DM_COOKIE_LENGTH 24
49 static const char *_name = DM_NAME;
51 static unsigned int major = 0;
52 static unsigned int _major = 0;
54 static DEFINE_IDR(_minor_idr);
56 static DEFINE_SPINLOCK(_minor_lock);
58 static void do_deferred_remove(struct work_struct *w);
60 static DECLARE_WORK(deferred_remove_work, do_deferred_remove);
62 static struct workqueue_struct *deferred_remove_workqueue;
66 * One of these is allocated per bio.
69 struct mapped_device *md;
73 unsigned long start_time;
74 spinlock_t endio_lock;
75 struct dm_stats_aux stats_aux;
79 * For request-based dm.
80 * One of these is allocated per request.
82 struct dm_rq_target_io {
83 struct mapped_device *md;
85 struct request *orig, *clone;
86 struct kthread_work work;
92 * For request-based dm - the bio clones we allocate are embedded in these
95 * We allocate these with bio_alloc_bioset, using the front_pad parameter when
96 * the bioset is created - this means the bio has to come at the end of the
99 struct dm_rq_clone_bio_info {
101 struct dm_rq_target_io *tio;
105 union map_info *dm_get_rq_mapinfo(struct request *rq)
107 if (rq && rq->end_io_data)
108 return &((struct dm_rq_target_io *)rq->end_io_data)->info;
111 EXPORT_SYMBOL_GPL(dm_get_rq_mapinfo);
113 #define MINOR_ALLOCED ((void *)-1)
116 * Bits for the md->flags field.
118 #define DMF_BLOCK_IO_FOR_SUSPEND 0
119 #define DMF_SUSPENDED 1
121 #define DMF_FREEING 3
122 #define DMF_DELETING 4
123 #define DMF_NOFLUSH_SUSPENDING 5
124 #define DMF_MERGE_IS_OPTIONAL 6
125 #define DMF_DEFERRED_REMOVE 7
126 #define DMF_SUSPENDED_INTERNALLY 8
129 * A dummy definition to make RCU happy.
130 * struct dm_table should never be dereferenced in this file.
137 * Work processed by per-device workqueue.
139 struct mapped_device {
140 struct srcu_struct io_barrier;
141 struct mutex suspend_lock;
146 * The current mapping.
147 * Use dm_get_live_table{_fast} or take suspend_lock for
150 struct dm_table __rcu *map;
152 struct list_head table_devices;
153 struct mutex table_devices_lock;
157 struct request_queue *queue;
159 /* Protect queue and type against concurrent access. */
160 struct mutex type_lock;
162 struct target_type *immutable_target_type;
164 struct gendisk *disk;
170 * A list of ios that arrived while we were suspended.
173 wait_queue_head_t wait;
174 struct work_struct work;
175 struct bio_list deferred;
176 spinlock_t deferred_lock;
179 * Processing queue (flush)
181 struct workqueue_struct *wq;
184 * io objects are allocated from here.
195 wait_queue_head_t eventq;
197 struct list_head uevent_list;
198 spinlock_t uevent_lock; /* Protect access to uevent_list */
201 * freeze/thaw support require holding onto a super block
203 struct super_block *frozen_sb;
204 struct block_device *bdev;
206 /* forced geometry settings */
207 struct hd_geometry geometry;
209 /* kobject and completion */
210 struct dm_kobject_holder kobj_holder;
212 /* zero-length flush that will be cloned and submitted to targets */
213 struct bio flush_bio;
215 /* the number of internal suspends */
216 unsigned internal_suspend_count;
218 struct dm_stats stats;
220 struct kthread_worker kworker;
221 struct task_struct *kworker_task;
223 /* for request-based merge heuristic in dm_request_fn() */
224 unsigned seq_rq_merge_deadline_usecs;
226 sector_t last_rq_pos;
227 ktime_t last_rq_start_time;
229 /* for blk-mq request-based DM support */
230 struct blk_mq_tag_set tag_set;
234 #ifdef CONFIG_DM_MQ_DEFAULT
235 static bool use_blk_mq = true;
237 static bool use_blk_mq = false;
240 bool dm_use_blk_mq(struct mapped_device *md)
242 return md->use_blk_mq;
246 * For mempools pre-allocation at the table loading time.
248 struct dm_md_mempools {
254 struct table_device {
255 struct list_head list;
257 struct dm_dev dm_dev;
260 #define RESERVED_BIO_BASED_IOS 16
261 #define RESERVED_REQUEST_BASED_IOS 256
262 #define RESERVED_MAX_IOS 1024
263 static struct kmem_cache *_io_cache;
264 static struct kmem_cache *_rq_tio_cache;
265 static struct kmem_cache *_rq_cache;
268 * Bio-based DM's mempools' reserved IOs set by the user.
270 static unsigned reserved_bio_based_ios = RESERVED_BIO_BASED_IOS;
273 * Request-based DM's mempools' reserved IOs set by the user.
275 static unsigned reserved_rq_based_ios = RESERVED_REQUEST_BASED_IOS;
277 static unsigned __dm_get_module_param(unsigned *module_param,
278 unsigned def, unsigned max)
280 unsigned param = ACCESS_ONCE(*module_param);
281 unsigned modified_param = 0;
284 modified_param = def;
285 else if (param > max)
286 modified_param = max;
288 if (modified_param) {
289 (void)cmpxchg(module_param, param, modified_param);
290 param = modified_param;
296 unsigned dm_get_reserved_bio_based_ios(void)
298 return __dm_get_module_param(&reserved_bio_based_ios,
299 RESERVED_BIO_BASED_IOS, RESERVED_MAX_IOS);
301 EXPORT_SYMBOL_GPL(dm_get_reserved_bio_based_ios);
303 unsigned dm_get_reserved_rq_based_ios(void)
305 return __dm_get_module_param(&reserved_rq_based_ios,
306 RESERVED_REQUEST_BASED_IOS, RESERVED_MAX_IOS);
308 EXPORT_SYMBOL_GPL(dm_get_reserved_rq_based_ios);
310 static int __init local_init(void)
314 /* allocate a slab for the dm_ios */
315 _io_cache = KMEM_CACHE(dm_io, 0);
319 _rq_tio_cache = KMEM_CACHE(dm_rq_target_io, 0);
321 goto out_free_io_cache;
323 _rq_cache = kmem_cache_create("dm_clone_request", sizeof(struct request),
324 __alignof__(struct request), 0, NULL);
326 goto out_free_rq_tio_cache;
328 r = dm_uevent_init();
330 goto out_free_rq_cache;
332 deferred_remove_workqueue = alloc_workqueue("kdmremove", WQ_UNBOUND, 1);
333 if (!deferred_remove_workqueue) {
335 goto out_uevent_exit;
339 r = register_blkdev(_major, _name);
341 goto out_free_workqueue;
349 destroy_workqueue(deferred_remove_workqueue);
353 kmem_cache_destroy(_rq_cache);
354 out_free_rq_tio_cache:
355 kmem_cache_destroy(_rq_tio_cache);
357 kmem_cache_destroy(_io_cache);
362 static void local_exit(void)
364 flush_scheduled_work();
365 destroy_workqueue(deferred_remove_workqueue);
367 kmem_cache_destroy(_rq_cache);
368 kmem_cache_destroy(_rq_tio_cache);
369 kmem_cache_destroy(_io_cache);
370 unregister_blkdev(_major, _name);
375 DMINFO("cleaned up");
378 static int (*_inits[])(void) __initdata = {
389 static void (*_exits[])(void) = {
400 static int __init dm_init(void)
402 const int count = ARRAY_SIZE(_inits);
406 for (i = 0; i < count; i++) {
421 static void __exit dm_exit(void)
423 int i = ARRAY_SIZE(_exits);
429 * Should be empty by this point.
431 idr_destroy(&_minor_idr);
435 * Block device functions
437 int dm_deleting_md(struct mapped_device *md)
439 return test_bit(DMF_DELETING, &md->flags);
442 static int dm_blk_open(struct block_device *bdev, fmode_t mode)
444 struct mapped_device *md;
446 spin_lock(&_minor_lock);
448 md = bdev->bd_disk->private_data;
452 if (test_bit(DMF_FREEING, &md->flags) ||
453 dm_deleting_md(md)) {
459 atomic_inc(&md->open_count);
461 spin_unlock(&_minor_lock);
463 return md ? 0 : -ENXIO;
466 static void dm_blk_close(struct gendisk *disk, fmode_t mode)
468 struct mapped_device *md;
470 spin_lock(&_minor_lock);
472 md = disk->private_data;
476 if (atomic_dec_and_test(&md->open_count) &&
477 (test_bit(DMF_DEFERRED_REMOVE, &md->flags)))
478 queue_work(deferred_remove_workqueue, &deferred_remove_work);
482 spin_unlock(&_minor_lock);
485 int dm_open_count(struct mapped_device *md)
487 return atomic_read(&md->open_count);
491 * Guarantees nothing is using the device before it's deleted.
493 int dm_lock_for_deletion(struct mapped_device *md, bool mark_deferred, bool only_deferred)
497 spin_lock(&_minor_lock);
499 if (dm_open_count(md)) {
502 set_bit(DMF_DEFERRED_REMOVE, &md->flags);
503 } else if (only_deferred && !test_bit(DMF_DEFERRED_REMOVE, &md->flags))
506 set_bit(DMF_DELETING, &md->flags);
508 spin_unlock(&_minor_lock);
513 int dm_cancel_deferred_remove(struct mapped_device *md)
517 spin_lock(&_minor_lock);
519 if (test_bit(DMF_DELETING, &md->flags))
522 clear_bit(DMF_DEFERRED_REMOVE, &md->flags);
524 spin_unlock(&_minor_lock);
529 static void do_deferred_remove(struct work_struct *w)
531 dm_deferred_remove();
534 sector_t dm_get_size(struct mapped_device *md)
536 return get_capacity(md->disk);
539 struct request_queue *dm_get_md_queue(struct mapped_device *md)
544 struct dm_stats *dm_get_stats(struct mapped_device *md)
549 static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo)
551 struct mapped_device *md = bdev->bd_disk->private_data;
553 return dm_get_geometry(md, geo);
556 static int dm_blk_ioctl(struct block_device *bdev, fmode_t mode,
557 unsigned int cmd, unsigned long arg)
559 struct mapped_device *md = bdev->bd_disk->private_data;
561 struct dm_table *map;
562 struct dm_target *tgt;
566 map = dm_get_live_table(md, &srcu_idx);
568 if (!map || !dm_table_get_size(map))
571 /* We only support devices that have a single target */
572 if (dm_table_get_num_targets(map) != 1)
575 tgt = dm_table_get_target(map, 0);
576 if (!tgt->type->ioctl)
579 if (dm_suspended_md(md)) {
584 r = tgt->type->ioctl(tgt, cmd, arg);
587 dm_put_live_table(md, srcu_idx);
589 if (r == -ENOTCONN) {
597 static struct dm_io *alloc_io(struct mapped_device *md)
599 return mempool_alloc(md->io_pool, GFP_NOIO);
602 static void free_io(struct mapped_device *md, struct dm_io *io)
604 mempool_free(io, md->io_pool);
607 static void free_tio(struct mapped_device *md, struct dm_target_io *tio)
609 bio_put(&tio->clone);
612 static struct dm_rq_target_io *alloc_rq_tio(struct mapped_device *md,
615 return mempool_alloc(md->io_pool, gfp_mask);
618 static void free_rq_tio(struct dm_rq_target_io *tio)
620 mempool_free(tio, tio->md->io_pool);
623 static struct request *alloc_clone_request(struct mapped_device *md,
626 return mempool_alloc(md->rq_pool, gfp_mask);
629 static void free_clone_request(struct mapped_device *md, struct request *rq)
631 mempool_free(rq, md->rq_pool);
634 static int md_in_flight(struct mapped_device *md)
636 return atomic_read(&md->pending[READ]) +
637 atomic_read(&md->pending[WRITE]);
640 static void start_io_acct(struct dm_io *io)
642 struct mapped_device *md = io->md;
643 struct bio *bio = io->bio;
645 int rw = bio_data_dir(bio);
647 io->start_time = jiffies;
649 cpu = part_stat_lock();
650 part_round_stats(cpu, &dm_disk(md)->part0);
652 atomic_set(&dm_disk(md)->part0.in_flight[rw],
653 atomic_inc_return(&md->pending[rw]));
655 if (unlikely(dm_stats_used(&md->stats)))
656 dm_stats_account_io(&md->stats, bio->bi_rw, bio->bi_iter.bi_sector,
657 bio_sectors(bio), false, 0, &io->stats_aux);
660 static void end_io_acct(struct dm_io *io)
662 struct mapped_device *md = io->md;
663 struct bio *bio = io->bio;
664 unsigned long duration = jiffies - io->start_time;
666 int rw = bio_data_dir(bio);
668 generic_end_io_acct(rw, &dm_disk(md)->part0, io->start_time);
670 if (unlikely(dm_stats_used(&md->stats)))
671 dm_stats_account_io(&md->stats, bio->bi_rw, bio->bi_iter.bi_sector,
672 bio_sectors(bio), true, duration, &io->stats_aux);
675 * After this is decremented the bio must not be touched if it is
678 pending = atomic_dec_return(&md->pending[rw]);
679 atomic_set(&dm_disk(md)->part0.in_flight[rw], pending);
680 pending += atomic_read(&md->pending[rw^0x1]);
682 /* nudge anyone waiting on suspend queue */
688 * Add the bio to the list of deferred io.
690 static void queue_io(struct mapped_device *md, struct bio *bio)
694 spin_lock_irqsave(&md->deferred_lock, flags);
695 bio_list_add(&md->deferred, bio);
696 spin_unlock_irqrestore(&md->deferred_lock, flags);
697 queue_work(md->wq, &md->work);
701 * Everyone (including functions in this file), should use this
702 * function to access the md->map field, and make sure they call
703 * dm_put_live_table() when finished.
705 struct dm_table *dm_get_live_table(struct mapped_device *md, int *srcu_idx) __acquires(md->io_barrier)
707 *srcu_idx = srcu_read_lock(&md->io_barrier);
709 return srcu_dereference(md->map, &md->io_barrier);
712 void dm_put_live_table(struct mapped_device *md, int srcu_idx) __releases(md->io_barrier)
714 srcu_read_unlock(&md->io_barrier, srcu_idx);
717 void dm_sync_table(struct mapped_device *md)
719 synchronize_srcu(&md->io_barrier);
720 synchronize_rcu_expedited();
724 * A fast alternative to dm_get_live_table/dm_put_live_table.
725 * The caller must not block between these two functions.
727 static struct dm_table *dm_get_live_table_fast(struct mapped_device *md) __acquires(RCU)
730 return rcu_dereference(md->map);
733 static void dm_put_live_table_fast(struct mapped_device *md) __releases(RCU)
739 * Open a table device so we can use it as a map destination.
741 static int open_table_device(struct table_device *td, dev_t dev,
742 struct mapped_device *md)
744 static char *_claim_ptr = "I belong to device-mapper";
745 struct block_device *bdev;
749 BUG_ON(td->dm_dev.bdev);
751 bdev = blkdev_get_by_dev(dev, td->dm_dev.mode | FMODE_EXCL, _claim_ptr);
753 return PTR_ERR(bdev);
755 r = bd_link_disk_holder(bdev, dm_disk(md));
757 blkdev_put(bdev, td->dm_dev.mode | FMODE_EXCL);
761 td->dm_dev.bdev = bdev;
766 * Close a table device that we've been using.
768 static void close_table_device(struct table_device *td, struct mapped_device *md)
770 if (!td->dm_dev.bdev)
773 bd_unlink_disk_holder(td->dm_dev.bdev, dm_disk(md));
774 blkdev_put(td->dm_dev.bdev, td->dm_dev.mode | FMODE_EXCL);
775 td->dm_dev.bdev = NULL;
778 static struct table_device *find_table_device(struct list_head *l, dev_t dev,
780 struct table_device *td;
782 list_for_each_entry(td, l, list)
783 if (td->dm_dev.bdev->bd_dev == dev && td->dm_dev.mode == mode)
789 int dm_get_table_device(struct mapped_device *md, dev_t dev, fmode_t mode,
790 struct dm_dev **result) {
792 struct table_device *td;
794 mutex_lock(&md->table_devices_lock);
795 td = find_table_device(&md->table_devices, dev, mode);
797 td = kmalloc(sizeof(*td), GFP_KERNEL);
799 mutex_unlock(&md->table_devices_lock);
803 td->dm_dev.mode = mode;
804 td->dm_dev.bdev = NULL;
806 if ((r = open_table_device(td, dev, md))) {
807 mutex_unlock(&md->table_devices_lock);
812 format_dev_t(td->dm_dev.name, dev);
814 atomic_set(&td->count, 0);
815 list_add(&td->list, &md->table_devices);
817 atomic_inc(&td->count);
818 mutex_unlock(&md->table_devices_lock);
820 *result = &td->dm_dev;
823 EXPORT_SYMBOL_GPL(dm_get_table_device);
825 void dm_put_table_device(struct mapped_device *md, struct dm_dev *d)
827 struct table_device *td = container_of(d, struct table_device, dm_dev);
829 mutex_lock(&md->table_devices_lock);
830 if (atomic_dec_and_test(&td->count)) {
831 close_table_device(td, md);
835 mutex_unlock(&md->table_devices_lock);
837 EXPORT_SYMBOL(dm_put_table_device);
839 static void free_table_devices(struct list_head *devices)
841 struct list_head *tmp, *next;
843 list_for_each_safe(tmp, next, devices) {
844 struct table_device *td = list_entry(tmp, struct table_device, list);
846 DMWARN("dm_destroy: %s still exists with %d references",
847 td->dm_dev.name, atomic_read(&td->count));
853 * Get the geometry associated with a dm device
855 int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo)
863 * Set the geometry of a device.
865 int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo)
867 sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors;
869 if (geo->start > sz) {
870 DMWARN("Start sector is beyond the geometry limits.");
879 /*-----------------------------------------------------------------
881 * A more elegant soln is in the works that uses the queue
882 * merge fn, unfortunately there are a couple of changes to
883 * the block layer that I want to make for this. So in the
884 * interests of getting something for people to use I give
885 * you this clearly demarcated crap.
886 *---------------------------------------------------------------*/
888 static int __noflush_suspending(struct mapped_device *md)
890 return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
894 * Decrements the number of outstanding ios that a bio has been
895 * cloned into, completing the original io if necc.
897 static void dec_pending(struct dm_io *io, int error)
902 struct mapped_device *md = io->md;
904 /* Push-back supersedes any I/O errors */
905 if (unlikely(error)) {
906 spin_lock_irqsave(&io->endio_lock, flags);
907 if (!(io->error > 0 && __noflush_suspending(md)))
909 spin_unlock_irqrestore(&io->endio_lock, flags);
912 if (atomic_dec_and_test(&io->io_count)) {
913 if (io->error == DM_ENDIO_REQUEUE) {
915 * Target requested pushing back the I/O.
917 spin_lock_irqsave(&md->deferred_lock, flags);
918 if (__noflush_suspending(md))
919 bio_list_add_head(&md->deferred, io->bio);
921 /* noflush suspend was interrupted. */
923 spin_unlock_irqrestore(&md->deferred_lock, flags);
926 io_error = io->error;
931 if (io_error == DM_ENDIO_REQUEUE)
934 if ((bio->bi_rw & REQ_FLUSH) && bio->bi_iter.bi_size) {
936 * Preflush done for flush with data, reissue
939 bio->bi_rw &= ~REQ_FLUSH;
942 /* done with normal IO or empty flush */
943 trace_block_bio_complete(md->queue, bio, io_error);
944 bio_endio(bio, io_error);
949 static void disable_write_same(struct mapped_device *md)
951 struct queue_limits *limits = dm_get_queue_limits(md);
953 /* device doesn't really support WRITE SAME, disable it */
954 limits->max_write_same_sectors = 0;
957 static void clone_endio(struct bio *bio, int error)
960 struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone);
961 struct dm_io *io = tio->io;
962 struct mapped_device *md = tio->io->md;
963 dm_endio_fn endio = tio->ti->type->end_io;
965 if (!bio_flagged(bio, BIO_UPTODATE) && !error)
969 r = endio(tio->ti, bio, error);
970 if (r < 0 || r == DM_ENDIO_REQUEUE)
972 * error and requeue request are handled
976 else if (r == DM_ENDIO_INCOMPLETE)
977 /* The target will handle the io */
980 DMWARN("unimplemented target endio return value: %d", r);
985 if (unlikely(r == -EREMOTEIO && (bio->bi_rw & REQ_WRITE_SAME) &&
986 !bdev_get_queue(bio->bi_bdev)->limits.max_write_same_sectors))
987 disable_write_same(md);
990 dec_pending(io, error);
994 * Partial completion handling for request-based dm
996 static void end_clone_bio(struct bio *clone, int error)
998 struct dm_rq_clone_bio_info *info =
999 container_of(clone, struct dm_rq_clone_bio_info, clone);
1000 struct dm_rq_target_io *tio = info->tio;
1001 struct bio *bio = info->orig;
1002 unsigned int nr_bytes = info->orig->bi_iter.bi_size;
1008 * An error has already been detected on the request.
1009 * Once error occurred, just let clone->end_io() handle
1015 * Don't notice the error to the upper layer yet.
1016 * The error handling decision is made by the target driver,
1017 * when the request is completed.
1024 * I/O for the bio successfully completed.
1025 * Notice the data completion to the upper layer.
1029 * bios are processed from the head of the list.
1030 * So the completing bio should always be rq->bio.
1031 * If it's not, something wrong is happening.
1033 if (tio->orig->bio != bio)
1034 DMERR("bio completion is going in the middle of the request");
1037 * Update the original request.
1038 * Do not use blk_end_request() here, because it may complete
1039 * the original request before the clone, and break the ordering.
1041 blk_update_request(tio->orig, 0, nr_bytes);
1044 static struct dm_rq_target_io *tio_from_request(struct request *rq)
1046 return (rq->q->mq_ops ? blk_mq_rq_to_pdu(rq) : rq->special);
1050 * Don't touch any member of the md after calling this function because
1051 * the md may be freed in dm_put() at the end of this function.
1052 * Or do dm_get() before calling this function and dm_put() later.
1054 static void rq_completed(struct mapped_device *md, int rw, bool run_queue)
1056 int nr_requests_pending;
1058 atomic_dec(&md->pending[rw]);
1060 /* nudge anyone waiting on suspend queue */
1061 nr_requests_pending = md_in_flight(md);
1062 if (!nr_requests_pending)
1066 * Run this off this callpath, as drivers could invoke end_io while
1067 * inside their request_fn (and holding the queue lock). Calling
1068 * back into ->request_fn() could deadlock attempting to grab the
1072 if (md->queue->mq_ops)
1073 blk_mq_run_hw_queues(md->queue, true);
1074 else if (!nr_requests_pending ||
1075 (nr_requests_pending >= md->queue->nr_congestion_on))
1076 blk_run_queue_async(md->queue);
1080 * dm_put() must be at the end of this function. See the comment above
1085 static void free_rq_clone(struct request *clone)
1087 struct dm_rq_target_io *tio = clone->end_io_data;
1088 struct mapped_device *md = tio->md;
1090 blk_rq_unprep_clone(clone);
1092 if (md->type == DM_TYPE_MQ_REQUEST_BASED)
1093 /* stacked on blk-mq queue(s) */
1094 tio->ti->type->release_clone_rq(clone);
1095 else if (!md->queue->mq_ops)
1096 /* request_fn queue stacked on request_fn queue(s) */
1097 free_clone_request(md, clone);
1099 * NOTE: for the blk-mq queue stacked on request_fn queue(s) case:
1100 * no need to call free_clone_request() because we leverage blk-mq by
1101 * allocating the clone at the end of the blk-mq pdu (see: clone_rq)
1104 if (!md->queue->mq_ops)
1109 * Complete the clone and the original request.
1110 * Must be called without clone's queue lock held,
1111 * see end_clone_request() for more details.
1113 static void dm_end_request(struct request *clone, int error)
1115 int rw = rq_data_dir(clone);
1116 struct dm_rq_target_io *tio = clone->end_io_data;
1117 struct mapped_device *md = tio->md;
1118 struct request *rq = tio->orig;
1120 if (rq->cmd_type == REQ_TYPE_BLOCK_PC) {
1121 rq->errors = clone->errors;
1122 rq->resid_len = clone->resid_len;
1126 * We are using the sense buffer of the original
1128 * So setting the length of the sense data is enough.
1130 rq->sense_len = clone->sense_len;
1133 free_rq_clone(clone);
1135 blk_end_request_all(rq, error);
1137 blk_mq_end_request(rq, error);
1138 rq_completed(md, rw, true);
1141 static void dm_unprep_request(struct request *rq)
1143 struct dm_rq_target_io *tio = tio_from_request(rq);
1144 struct request *clone = tio->clone;
1146 if (!rq->q->mq_ops) {
1148 rq->cmd_flags &= ~REQ_DONTPREP;
1152 free_rq_clone(clone);
1156 * Requeue the original request of a clone.
1158 static void old_requeue_request(struct request *rq)
1160 struct request_queue *q = rq->q;
1161 unsigned long flags;
1163 spin_lock_irqsave(q->queue_lock, flags);
1164 blk_requeue_request(q, rq);
1165 blk_run_queue_async(q);
1166 spin_unlock_irqrestore(q->queue_lock, flags);
1169 static void dm_requeue_unmapped_original_request(struct mapped_device *md,
1172 int rw = rq_data_dir(rq);
1174 dm_unprep_request(rq);
1177 old_requeue_request(rq);
1179 blk_mq_requeue_request(rq);
1180 blk_mq_kick_requeue_list(rq->q);
1183 rq_completed(md, rw, false);
1186 static void dm_requeue_unmapped_request(struct request *clone)
1188 struct dm_rq_target_io *tio = clone->end_io_data;
1190 dm_requeue_unmapped_original_request(tio->md, tio->orig);
1193 static void old_stop_queue(struct request_queue *q)
1195 unsigned long flags;
1197 if (blk_queue_stopped(q))
1200 spin_lock_irqsave(q->queue_lock, flags);
1202 spin_unlock_irqrestore(q->queue_lock, flags);
1205 static void stop_queue(struct request_queue *q)
1210 blk_mq_stop_hw_queues(q);
1213 static void old_start_queue(struct request_queue *q)
1215 unsigned long flags;
1217 spin_lock_irqsave(q->queue_lock, flags);
1218 if (blk_queue_stopped(q))
1220 spin_unlock_irqrestore(q->queue_lock, flags);
1223 static void start_queue(struct request_queue *q)
1228 blk_mq_start_stopped_hw_queues(q, true);
1231 static void dm_done(struct request *clone, int error, bool mapped)
1234 struct dm_rq_target_io *tio = clone->end_io_data;
1235 dm_request_endio_fn rq_end_io = NULL;
1238 rq_end_io = tio->ti->type->rq_end_io;
1240 if (mapped && rq_end_io)
1241 r = rq_end_io(tio->ti, clone, error, &tio->info);
1244 if (unlikely(r == -EREMOTEIO && (clone->cmd_flags & REQ_WRITE_SAME) &&
1245 !clone->q->limits.max_write_same_sectors))
1246 disable_write_same(tio->md);
1249 /* The target wants to complete the I/O */
1250 dm_end_request(clone, r);
1251 else if (r == DM_ENDIO_INCOMPLETE)
1252 /* The target will handle the I/O */
1254 else if (r == DM_ENDIO_REQUEUE)
1255 /* The target wants to requeue the I/O */
1256 dm_requeue_unmapped_request(clone);
1258 DMWARN("unimplemented target endio return value: %d", r);
1264 * Request completion handler for request-based dm
1266 static void dm_softirq_done(struct request *rq)
1269 struct dm_rq_target_io *tio = tio_from_request(rq);
1270 struct request *clone = tio->clone;
1274 rw = rq_data_dir(rq);
1275 if (!rq->q->mq_ops) {
1276 blk_end_request_all(rq, tio->error);
1277 rq_completed(tio->md, rw, false);
1280 blk_mq_end_request(rq, tio->error);
1281 rq_completed(tio->md, rw, false);
1286 if (rq->cmd_flags & REQ_FAILED)
1289 dm_done(clone, tio->error, mapped);
1293 * Complete the clone and the original request with the error status
1294 * through softirq context.
1296 static void dm_complete_request(struct request *rq, int error)
1298 struct dm_rq_target_io *tio = tio_from_request(rq);
1301 blk_complete_request(rq);
1305 * Complete the not-mapped clone and the original request with the error status
1306 * through softirq context.
1307 * Target's rq_end_io() function isn't called.
1308 * This may be used when the target's map_rq() or clone_and_map_rq() functions fail.
1310 static void dm_kill_unmapped_request(struct request *rq, int error)
1312 rq->cmd_flags |= REQ_FAILED;
1313 dm_complete_request(rq, error);
1317 * Called with the clone's queue lock held (for non-blk-mq)
1319 static void end_clone_request(struct request *clone, int error)
1321 struct dm_rq_target_io *tio = clone->end_io_data;
1323 if (!clone->q->mq_ops) {
1325 * For just cleaning up the information of the queue in which
1326 * the clone was dispatched.
1327 * The clone is *NOT* freed actually here because it is alloced
1328 * from dm own mempool (REQ_ALLOCED isn't set).
1330 __blk_put_request(clone->q, clone);
1334 * Actual request completion is done in a softirq context which doesn't
1335 * hold the clone's queue lock. Otherwise, deadlock could occur because:
1336 * - another request may be submitted by the upper level driver
1337 * of the stacking during the completion
1338 * - the submission which requires queue lock may be done
1339 * against this clone's queue
1341 dm_complete_request(tio->orig, error);
1345 * Return maximum size of I/O possible at the supplied sector up to the current
1348 static sector_t max_io_len_target_boundary(sector_t sector, struct dm_target *ti)
1350 sector_t target_offset = dm_target_offset(ti, sector);
1352 return ti->len - target_offset;
1355 static sector_t max_io_len(sector_t sector, struct dm_target *ti)
1357 sector_t len = max_io_len_target_boundary(sector, ti);
1358 sector_t offset, max_len;
1361 * Does the target need to split even further?
1363 if (ti->max_io_len) {
1364 offset = dm_target_offset(ti, sector);
1365 if (unlikely(ti->max_io_len & (ti->max_io_len - 1)))
1366 max_len = sector_div(offset, ti->max_io_len);
1368 max_len = offset & (ti->max_io_len - 1);
1369 max_len = ti->max_io_len - max_len;
1378 int dm_set_target_max_io_len(struct dm_target *ti, sector_t len)
1380 if (len > UINT_MAX) {
1381 DMERR("Specified maximum size of target IO (%llu) exceeds limit (%u)",
1382 (unsigned long long)len, UINT_MAX);
1383 ti->error = "Maximum size of target IO is too large";
1387 ti->max_io_len = (uint32_t) len;
1391 EXPORT_SYMBOL_GPL(dm_set_target_max_io_len);
1394 * A target may call dm_accept_partial_bio only from the map routine. It is
1395 * allowed for all bio types except REQ_FLUSH.
1397 * dm_accept_partial_bio informs the dm that the target only wants to process
1398 * additional n_sectors sectors of the bio and the rest of the data should be
1399 * sent in a next bio.
1401 * A diagram that explains the arithmetics:
1402 * +--------------------+---------------+-------+
1404 * +--------------------+---------------+-------+
1406 * <-------------- *tio->len_ptr --------------->
1407 * <------- bi_size ------->
1410 * Region 1 was already iterated over with bio_advance or similar function.
1411 * (it may be empty if the target doesn't use bio_advance)
1412 * Region 2 is the remaining bio size that the target wants to process.
1413 * (it may be empty if region 1 is non-empty, although there is no reason
1415 * The target requires that region 3 is to be sent in the next bio.
1417 * If the target wants to receive multiple copies of the bio (via num_*bios, etc),
1418 * the partially processed part (the sum of regions 1+2) must be the same for all
1419 * copies of the bio.
1421 void dm_accept_partial_bio(struct bio *bio, unsigned n_sectors)
1423 struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone);
1424 unsigned bi_size = bio->bi_iter.bi_size >> SECTOR_SHIFT;
1425 BUG_ON(bio->bi_rw & REQ_FLUSH);
1426 BUG_ON(bi_size > *tio->len_ptr);
1427 BUG_ON(n_sectors > bi_size);
1428 *tio->len_ptr -= bi_size - n_sectors;
1429 bio->bi_iter.bi_size = n_sectors << SECTOR_SHIFT;
1431 EXPORT_SYMBOL_GPL(dm_accept_partial_bio);
1433 static void __map_bio(struct dm_target_io *tio)
1437 struct mapped_device *md;
1438 struct bio *clone = &tio->clone;
1439 struct dm_target *ti = tio->ti;
1441 clone->bi_end_io = clone_endio;
1444 * Map the clone. If r == 0 we don't need to do
1445 * anything, the target has assumed ownership of
1448 atomic_inc(&tio->io->io_count);
1449 sector = clone->bi_iter.bi_sector;
1450 r = ti->type->map(ti, clone);
1451 if (r == DM_MAPIO_REMAPPED) {
1452 /* the bio has been remapped so dispatch it */
1454 trace_block_bio_remap(bdev_get_queue(clone->bi_bdev), clone,
1455 tio->io->bio->bi_bdev->bd_dev, sector);
1457 generic_make_request(clone);
1458 } else if (r < 0 || r == DM_MAPIO_REQUEUE) {
1459 /* error the io and bail out, or requeue it if needed */
1461 dec_pending(tio->io, r);
1464 DMWARN("unimplemented target map return value: %d", r);
1470 struct mapped_device *md;
1471 struct dm_table *map;
1475 unsigned sector_count;
1478 static void bio_setup_sector(struct bio *bio, sector_t sector, unsigned len)
1480 bio->bi_iter.bi_sector = sector;
1481 bio->bi_iter.bi_size = to_bytes(len);
1485 * Creates a bio that consists of range of complete bvecs.
1487 static void clone_bio(struct dm_target_io *tio, struct bio *bio,
1488 sector_t sector, unsigned len)
1490 struct bio *clone = &tio->clone;
1492 __bio_clone_fast(clone, bio);
1494 if (bio_integrity(bio))
1495 bio_integrity_clone(clone, bio, GFP_NOIO);
1497 bio_advance(clone, to_bytes(sector - clone->bi_iter.bi_sector));
1498 clone->bi_iter.bi_size = to_bytes(len);
1500 if (bio_integrity(bio))
1501 bio_integrity_trim(clone, 0, len);
1504 static struct dm_target_io *alloc_tio(struct clone_info *ci,
1505 struct dm_target *ti,
1506 unsigned target_bio_nr)
1508 struct dm_target_io *tio;
1511 clone = bio_alloc_bioset(GFP_NOIO, 0, ci->md->bs);
1512 tio = container_of(clone, struct dm_target_io, clone);
1516 tio->target_bio_nr = target_bio_nr;
1521 static void __clone_and_map_simple_bio(struct clone_info *ci,
1522 struct dm_target *ti,
1523 unsigned target_bio_nr, unsigned *len)
1525 struct dm_target_io *tio = alloc_tio(ci, ti, target_bio_nr);
1526 struct bio *clone = &tio->clone;
1530 __bio_clone_fast(clone, ci->bio);
1532 bio_setup_sector(clone, ci->sector, *len);
1537 static void __send_duplicate_bios(struct clone_info *ci, struct dm_target *ti,
1538 unsigned num_bios, unsigned *len)
1540 unsigned target_bio_nr;
1542 for (target_bio_nr = 0; target_bio_nr < num_bios; target_bio_nr++)
1543 __clone_and_map_simple_bio(ci, ti, target_bio_nr, len);
1546 static int __send_empty_flush(struct clone_info *ci)
1548 unsigned target_nr = 0;
1549 struct dm_target *ti;
1551 BUG_ON(bio_has_data(ci->bio));
1552 while ((ti = dm_table_get_target(ci->map, target_nr++)))
1553 __send_duplicate_bios(ci, ti, ti->num_flush_bios, NULL);
1558 static void __clone_and_map_data_bio(struct clone_info *ci, struct dm_target *ti,
1559 sector_t sector, unsigned *len)
1561 struct bio *bio = ci->bio;
1562 struct dm_target_io *tio;
1563 unsigned target_bio_nr;
1564 unsigned num_target_bios = 1;
1567 * Does the target want to receive duplicate copies of the bio?
1569 if (bio_data_dir(bio) == WRITE && ti->num_write_bios)
1570 num_target_bios = ti->num_write_bios(ti, bio);
1572 for (target_bio_nr = 0; target_bio_nr < num_target_bios; target_bio_nr++) {
1573 tio = alloc_tio(ci, ti, target_bio_nr);
1575 clone_bio(tio, bio, sector, *len);
1580 typedef unsigned (*get_num_bios_fn)(struct dm_target *ti);
1582 static unsigned get_num_discard_bios(struct dm_target *ti)
1584 return ti->num_discard_bios;
1587 static unsigned get_num_write_same_bios(struct dm_target *ti)
1589 return ti->num_write_same_bios;
1592 typedef bool (*is_split_required_fn)(struct dm_target *ti);
1594 static bool is_split_required_for_discard(struct dm_target *ti)
1596 return ti->split_discard_bios;
1599 static int __send_changing_extent_only(struct clone_info *ci,
1600 get_num_bios_fn get_num_bios,
1601 is_split_required_fn is_split_required)
1603 struct dm_target *ti;
1608 ti = dm_table_find_target(ci->map, ci->sector);
1609 if (!dm_target_is_valid(ti))
1613 * Even though the device advertised support for this type of
1614 * request, that does not mean every target supports it, and
1615 * reconfiguration might also have changed that since the
1616 * check was performed.
1618 num_bios = get_num_bios ? get_num_bios(ti) : 0;
1622 if (is_split_required && !is_split_required(ti))
1623 len = min((sector_t)ci->sector_count, max_io_len_target_boundary(ci->sector, ti));
1625 len = min((sector_t)ci->sector_count, max_io_len(ci->sector, ti));
1627 __send_duplicate_bios(ci, ti, num_bios, &len);
1630 } while (ci->sector_count -= len);
1635 static int __send_discard(struct clone_info *ci)
1637 return __send_changing_extent_only(ci, get_num_discard_bios,
1638 is_split_required_for_discard);
1641 static int __send_write_same(struct clone_info *ci)
1643 return __send_changing_extent_only(ci, get_num_write_same_bios, NULL);
1647 * Select the correct strategy for processing a non-flush bio.
1649 static int __split_and_process_non_flush(struct clone_info *ci)
1651 struct bio *bio = ci->bio;
1652 struct dm_target *ti;
1655 if (unlikely(bio->bi_rw & REQ_DISCARD))
1656 return __send_discard(ci);
1657 else if (unlikely(bio->bi_rw & REQ_WRITE_SAME))
1658 return __send_write_same(ci);
1660 ti = dm_table_find_target(ci->map, ci->sector);
1661 if (!dm_target_is_valid(ti))
1664 len = min_t(sector_t, max_io_len(ci->sector, ti), ci->sector_count);
1666 __clone_and_map_data_bio(ci, ti, ci->sector, &len);
1669 ci->sector_count -= len;
1675 * Entry point to split a bio into clones and submit them to the targets.
1677 static void __split_and_process_bio(struct mapped_device *md,
1678 struct dm_table *map, struct bio *bio)
1680 struct clone_info ci;
1683 if (unlikely(!map)) {
1690 ci.io = alloc_io(md);
1692 atomic_set(&ci.io->io_count, 1);
1695 spin_lock_init(&ci.io->endio_lock);
1696 ci.sector = bio->bi_iter.bi_sector;
1698 start_io_acct(ci.io);
1700 if (bio->bi_rw & REQ_FLUSH) {
1701 ci.bio = &ci.md->flush_bio;
1702 ci.sector_count = 0;
1703 error = __send_empty_flush(&ci);
1704 /* dec_pending submits any data associated with flush */
1707 ci.sector_count = bio_sectors(bio);
1708 while (ci.sector_count && !error)
1709 error = __split_and_process_non_flush(&ci);
1712 /* drop the extra reference count */
1713 dec_pending(ci.io, error);
1715 /*-----------------------------------------------------------------
1717 *---------------------------------------------------------------*/
1719 static int dm_merge_bvec(struct request_queue *q,
1720 struct bvec_merge_data *bvm,
1721 struct bio_vec *biovec)
1723 struct mapped_device *md = q->queuedata;
1724 struct dm_table *map = dm_get_live_table_fast(md);
1725 struct dm_target *ti;
1726 sector_t max_sectors;
1732 ti = dm_table_find_target(map, bvm->bi_sector);
1733 if (!dm_target_is_valid(ti))
1737 * Find maximum amount of I/O that won't need splitting
1739 max_sectors = min(max_io_len(bvm->bi_sector, ti),
1740 (sector_t) queue_max_sectors(q));
1741 max_size = (max_sectors << SECTOR_SHIFT) - bvm->bi_size;
1742 if (unlikely(max_size < 0)) /* this shouldn't _ever_ happen */
1746 * merge_bvec_fn() returns number of bytes
1747 * it can accept at this offset
1748 * max is precomputed maximal io size
1750 if (max_size && ti->type->merge)
1751 max_size = ti->type->merge(ti, bvm, biovec, max_size);
1753 * If the target doesn't support merge method and some of the devices
1754 * provided their merge_bvec method (we know this by looking for the
1755 * max_hw_sectors that dm_set_device_limits may set), then we can't
1756 * allow bios with multiple vector entries. So always set max_size
1757 * to 0, and the code below allows just one page.
1759 else if (queue_max_hw_sectors(q) <= PAGE_SIZE >> 9)
1763 dm_put_live_table_fast(md);
1765 * Always allow an entire first page
1767 if (max_size <= biovec->bv_len && !(bvm->bi_size >> SECTOR_SHIFT))
1768 max_size = biovec->bv_len;
1774 * The request function that just remaps the bio built up by
1777 static void dm_make_request(struct request_queue *q, struct bio *bio)
1779 int rw = bio_data_dir(bio);
1780 struct mapped_device *md = q->queuedata;
1782 struct dm_table *map;
1784 map = dm_get_live_table(md, &srcu_idx);
1786 generic_start_io_acct(rw, bio_sectors(bio), &dm_disk(md)->part0);
1788 /* if we're suspended, we have to queue this io for later */
1789 if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))) {
1790 dm_put_live_table(md, srcu_idx);
1792 if (bio_rw(bio) != READA)
1799 __split_and_process_bio(md, map, bio);
1800 dm_put_live_table(md, srcu_idx);
1804 int dm_request_based(struct mapped_device *md)
1806 return blk_queue_stackable(md->queue);
1809 static void dm_dispatch_clone_request(struct request *clone, struct request *rq)
1813 if (blk_queue_io_stat(clone->q))
1814 clone->cmd_flags |= REQ_IO_STAT;
1816 clone->start_time = jiffies;
1817 r = blk_insert_cloned_request(clone->q, clone);
1819 /* must complete clone in terms of original request */
1820 dm_complete_request(rq, r);
1823 static int dm_rq_bio_constructor(struct bio *bio, struct bio *bio_orig,
1826 struct dm_rq_target_io *tio = data;
1827 struct dm_rq_clone_bio_info *info =
1828 container_of(bio, struct dm_rq_clone_bio_info, clone);
1830 info->orig = bio_orig;
1832 bio->bi_end_io = end_clone_bio;
1837 static int setup_clone(struct request *clone, struct request *rq,
1838 struct dm_rq_target_io *tio, gfp_t gfp_mask)
1842 r = blk_rq_prep_clone(clone, rq, tio->md->bs, gfp_mask,
1843 dm_rq_bio_constructor, tio);
1847 clone->cmd = rq->cmd;
1848 clone->cmd_len = rq->cmd_len;
1849 clone->sense = rq->sense;
1850 clone->end_io = end_clone_request;
1851 clone->end_io_data = tio;
1858 static struct request *clone_rq(struct request *rq, struct mapped_device *md,
1859 struct dm_rq_target_io *tio, gfp_t gfp_mask)
1862 * Do not allocate a clone if tio->clone was already set
1863 * (see: dm_mq_queue_rq).
1865 bool alloc_clone = !tio->clone;
1866 struct request *clone;
1869 clone = alloc_clone_request(md, gfp_mask);
1875 blk_rq_init(NULL, clone);
1876 if (setup_clone(clone, rq, tio, gfp_mask)) {
1879 free_clone_request(md, clone);
1886 static void map_tio_request(struct kthread_work *work);
1888 static void init_tio(struct dm_rq_target_io *tio, struct request *rq,
1889 struct mapped_device *md)
1896 memset(&tio->info, 0, sizeof(tio->info));
1897 if (md->kworker_task)
1898 init_kthread_work(&tio->work, map_tio_request);
1901 static struct dm_rq_target_io *prep_tio(struct request *rq,
1902 struct mapped_device *md, gfp_t gfp_mask)
1904 struct dm_rq_target_io *tio;
1906 struct dm_table *table;
1908 tio = alloc_rq_tio(md, gfp_mask);
1912 init_tio(tio, rq, md);
1914 table = dm_get_live_table(md, &srcu_idx);
1915 if (!dm_table_mq_request_based(table)) {
1916 if (!clone_rq(rq, md, tio, gfp_mask)) {
1917 dm_put_live_table(md, srcu_idx);
1922 dm_put_live_table(md, srcu_idx);
1928 * Called with the queue lock held.
1930 static int dm_prep_fn(struct request_queue *q, struct request *rq)
1932 struct mapped_device *md = q->queuedata;
1933 struct dm_rq_target_io *tio;
1935 if (unlikely(rq->special)) {
1936 DMWARN("Already has something in rq->special.");
1937 return BLKPREP_KILL;
1940 tio = prep_tio(rq, md, GFP_ATOMIC);
1942 return BLKPREP_DEFER;
1945 rq->cmd_flags |= REQ_DONTPREP;
1952 * 0 : the request has been processed
1953 * DM_MAPIO_REQUEUE : the original request needs to be requeued
1954 * < 0 : the request was completed due to failure
1956 static int map_request(struct dm_rq_target_io *tio, struct request *rq,
1957 struct mapped_device *md)
1960 struct dm_target *ti = tio->ti;
1961 struct request *clone = NULL;
1965 r = ti->type->map_rq(ti, clone, &tio->info);
1967 r = ti->type->clone_and_map_rq(ti, rq, &tio->info, &clone);
1969 /* The target wants to complete the I/O */
1970 dm_kill_unmapped_request(rq, r);
1973 if (r != DM_MAPIO_REMAPPED)
1975 if (setup_clone(clone, rq, tio, GFP_ATOMIC)) {
1977 ti->type->release_clone_rq(clone);
1978 return DM_MAPIO_REQUEUE;
1983 case DM_MAPIO_SUBMITTED:
1984 /* The target has taken the I/O to submit by itself later */
1986 case DM_MAPIO_REMAPPED:
1987 /* The target has remapped the I/O so dispatch it */
1988 trace_block_rq_remap(clone->q, clone, disk_devt(dm_disk(md)),
1990 dm_dispatch_clone_request(clone, rq);
1992 case DM_MAPIO_REQUEUE:
1993 /* The target wants to requeue the I/O */
1994 dm_requeue_unmapped_request(clone);
1998 DMWARN("unimplemented target map return value: %d", r);
2002 /* The target wants to complete the I/O */
2003 dm_kill_unmapped_request(rq, r);
2010 static void map_tio_request(struct kthread_work *work)
2012 struct dm_rq_target_io *tio = container_of(work, struct dm_rq_target_io, work);
2013 struct request *rq = tio->orig;
2014 struct mapped_device *md = tio->md;
2016 if (map_request(tio, rq, md) == DM_MAPIO_REQUEUE)
2017 dm_requeue_unmapped_original_request(md, rq);
2020 static void dm_start_request(struct mapped_device *md, struct request *orig)
2022 if (!orig->q->mq_ops)
2023 blk_start_request(orig);
2025 blk_mq_start_request(orig);
2026 atomic_inc(&md->pending[rq_data_dir(orig)]);
2028 if (md->seq_rq_merge_deadline_usecs) {
2029 md->last_rq_pos = rq_end_sector(orig);
2030 md->last_rq_rw = rq_data_dir(orig);
2031 md->last_rq_start_time = ktime_get();
2035 * Hold the md reference here for the in-flight I/O.
2036 * We can't rely on the reference count by device opener,
2037 * because the device may be closed during the request completion
2038 * when all bios are completed.
2039 * See the comment in rq_completed() too.
2044 #define MAX_SEQ_RQ_MERGE_DEADLINE_USECS 100000
2046 ssize_t dm_attr_rq_based_seq_io_merge_deadline_show(struct mapped_device *md, char *buf)
2048 return sprintf(buf, "%u\n", md->seq_rq_merge_deadline_usecs);
2051 ssize_t dm_attr_rq_based_seq_io_merge_deadline_store(struct mapped_device *md,
2052 const char *buf, size_t count)
2056 if (!dm_request_based(md) || md->use_blk_mq)
2059 if (kstrtouint(buf, 10, &deadline))
2062 if (deadline > MAX_SEQ_RQ_MERGE_DEADLINE_USECS)
2063 deadline = MAX_SEQ_RQ_MERGE_DEADLINE_USECS;
2065 md->seq_rq_merge_deadline_usecs = deadline;
2070 static bool dm_request_peeked_before_merge_deadline(struct mapped_device *md)
2072 ktime_t kt_deadline;
2074 if (!md->seq_rq_merge_deadline_usecs)
2077 kt_deadline = ns_to_ktime((u64)md->seq_rq_merge_deadline_usecs * NSEC_PER_USEC);
2078 kt_deadline = ktime_add_safe(md->last_rq_start_time, kt_deadline);
2080 return !ktime_after(ktime_get(), kt_deadline);
2084 * q->request_fn for request-based dm.
2085 * Called with the queue lock held.
2087 static void dm_request_fn(struct request_queue *q)
2089 struct mapped_device *md = q->queuedata;
2091 struct dm_table *map = dm_get_live_table(md, &srcu_idx);
2092 struct dm_target *ti;
2094 struct dm_rq_target_io *tio;
2098 * For suspend, check blk_queue_stopped() and increment
2099 * ->pending within a single queue_lock not to increment the
2100 * number of in-flight I/Os after the queue is stopped in
2103 while (!blk_queue_stopped(q)) {
2104 rq = blk_peek_request(q);
2108 /* always use block 0 to find the target for flushes for now */
2110 if (!(rq->cmd_flags & REQ_FLUSH))
2111 pos = blk_rq_pos(rq);
2113 ti = dm_table_find_target(map, pos);
2114 if (!dm_target_is_valid(ti)) {
2116 * Must perform setup, that rq_completed() requires,
2117 * before calling dm_kill_unmapped_request
2119 DMERR_LIMIT("request attempted access beyond the end of device");
2120 dm_start_request(md, rq);
2121 dm_kill_unmapped_request(rq, -EIO);
2125 if (dm_request_peeked_before_merge_deadline(md) &&
2126 md_in_flight(md) && rq->bio && rq->bio->bi_vcnt == 1 &&
2127 md->last_rq_pos == pos && md->last_rq_rw == rq_data_dir(rq))
2130 if (ti->type->busy && ti->type->busy(ti))
2133 dm_start_request(md, rq);
2135 tio = tio_from_request(rq);
2136 /* Establish tio->ti before queuing work (map_tio_request) */
2138 queue_kthread_work(&md->kworker, &tio->work);
2139 BUG_ON(!irqs_disabled());
2145 blk_delay_queue(q, HZ / 100);
2147 dm_put_live_table(md, srcu_idx);
2150 static int dm_any_congested(void *congested_data, int bdi_bits)
2153 struct mapped_device *md = congested_data;
2154 struct dm_table *map;
2156 if (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
2157 map = dm_get_live_table_fast(md);
2160 * Request-based dm cares about only own queue for
2161 * the query about congestion status of request_queue
2163 if (dm_request_based(md))
2164 r = md->queue->backing_dev_info.state &
2167 r = dm_table_any_congested(map, bdi_bits);
2169 dm_put_live_table_fast(md);
2175 /*-----------------------------------------------------------------
2176 * An IDR is used to keep track of allocated minor numbers.
2177 *---------------------------------------------------------------*/
2178 static void free_minor(int minor)
2180 spin_lock(&_minor_lock);
2181 idr_remove(&_minor_idr, minor);
2182 spin_unlock(&_minor_lock);
2186 * See if the device with a specific minor # is free.
2188 static int specific_minor(int minor)
2192 if (minor >= (1 << MINORBITS))
2195 idr_preload(GFP_KERNEL);
2196 spin_lock(&_minor_lock);
2198 r = idr_alloc(&_minor_idr, MINOR_ALLOCED, minor, minor + 1, GFP_NOWAIT);
2200 spin_unlock(&_minor_lock);
2203 return r == -ENOSPC ? -EBUSY : r;
2207 static int next_free_minor(int *minor)
2211 idr_preload(GFP_KERNEL);
2212 spin_lock(&_minor_lock);
2214 r = idr_alloc(&_minor_idr, MINOR_ALLOCED, 0, 1 << MINORBITS, GFP_NOWAIT);
2216 spin_unlock(&_minor_lock);
2224 static const struct block_device_operations dm_blk_dops;
2226 static void dm_wq_work(struct work_struct *work);
2228 static void dm_init_md_queue(struct mapped_device *md)
2231 * Request-based dm devices cannot be stacked on top of bio-based dm
2232 * devices. The type of this dm device may not have been decided yet.
2233 * The type is decided at the first table loading time.
2234 * To prevent problematic device stacking, clear the queue flag
2235 * for request stacking support until then.
2237 * This queue is new, so no concurrency on the queue_flags.
2239 queue_flag_clear_unlocked(QUEUE_FLAG_STACKABLE, md->queue);
2242 static void dm_init_old_md_queue(struct mapped_device *md)
2244 md->use_blk_mq = false;
2245 dm_init_md_queue(md);
2248 * Initialize aspects of queue that aren't relevant for blk-mq
2250 md->queue->queuedata = md;
2251 md->queue->backing_dev_info.congested_fn = dm_any_congested;
2252 md->queue->backing_dev_info.congested_data = md;
2254 blk_queue_bounce_limit(md->queue, BLK_BOUNCE_ANY);
2258 * Allocate and initialise a blank device with a given minor.
2260 static struct mapped_device *alloc_dev(int minor)
2263 struct mapped_device *md = kzalloc(sizeof(*md), GFP_KERNEL);
2267 DMWARN("unable to allocate device, out of memory.");
2271 if (!try_module_get(THIS_MODULE))
2272 goto bad_module_get;
2274 /* get a minor number for the dev */
2275 if (minor == DM_ANY_MINOR)
2276 r = next_free_minor(&minor);
2278 r = specific_minor(minor);
2282 r = init_srcu_struct(&md->io_barrier);
2284 goto bad_io_barrier;
2286 md->use_blk_mq = use_blk_mq;
2287 md->type = DM_TYPE_NONE;
2288 mutex_init(&md->suspend_lock);
2289 mutex_init(&md->type_lock);
2290 mutex_init(&md->table_devices_lock);
2291 spin_lock_init(&md->deferred_lock);
2292 atomic_set(&md->holders, 1);
2293 atomic_set(&md->open_count, 0);
2294 atomic_set(&md->event_nr, 0);
2295 atomic_set(&md->uevent_seq, 0);
2296 INIT_LIST_HEAD(&md->uevent_list);
2297 INIT_LIST_HEAD(&md->table_devices);
2298 spin_lock_init(&md->uevent_lock);
2300 md->queue = blk_alloc_queue(GFP_KERNEL);
2304 dm_init_md_queue(md);
2306 md->disk = alloc_disk(1);
2310 atomic_set(&md->pending[0], 0);
2311 atomic_set(&md->pending[1], 0);
2312 init_waitqueue_head(&md->wait);
2313 INIT_WORK(&md->work, dm_wq_work);
2314 init_waitqueue_head(&md->eventq);
2315 init_completion(&md->kobj_holder.completion);
2316 md->kworker_task = NULL;
2318 md->disk->major = _major;
2319 md->disk->first_minor = minor;
2320 md->disk->fops = &dm_blk_dops;
2321 md->disk->queue = md->queue;
2322 md->disk->private_data = md;
2323 sprintf(md->disk->disk_name, "dm-%d", minor);
2325 format_dev_t(md->name, MKDEV(_major, minor));
2327 md->wq = alloc_workqueue("kdmflush", WQ_MEM_RECLAIM, 0);
2331 md->bdev = bdget_disk(md->disk, 0);
2335 bio_init(&md->flush_bio);
2336 md->flush_bio.bi_bdev = md->bdev;
2337 md->flush_bio.bi_rw = WRITE_FLUSH;
2339 dm_stats_init(&md->stats);
2341 /* Populate the mapping, nobody knows we exist yet */
2342 spin_lock(&_minor_lock);
2343 old_md = idr_replace(&_minor_idr, md, minor);
2344 spin_unlock(&_minor_lock);
2346 BUG_ON(old_md != MINOR_ALLOCED);
2351 destroy_workqueue(md->wq);
2353 del_gendisk(md->disk);
2356 blk_cleanup_queue(md->queue);
2358 cleanup_srcu_struct(&md->io_barrier);
2362 module_put(THIS_MODULE);
2368 static void unlock_fs(struct mapped_device *md);
2370 static void free_dev(struct mapped_device *md)
2372 int minor = MINOR(disk_devt(md->disk));
2375 destroy_workqueue(md->wq);
2377 if (md->kworker_task)
2378 kthread_stop(md->kworker_task);
2380 mempool_destroy(md->io_pool);
2382 mempool_destroy(md->rq_pool);
2384 bioset_free(md->bs);
2386 cleanup_srcu_struct(&md->io_barrier);
2387 free_table_devices(&md->table_devices);
2388 dm_stats_cleanup(&md->stats);
2390 spin_lock(&_minor_lock);
2391 md->disk->private_data = NULL;
2392 spin_unlock(&_minor_lock);
2393 if (blk_get_integrity(md->disk))
2394 blk_integrity_unregister(md->disk);
2395 del_gendisk(md->disk);
2397 blk_cleanup_queue(md->queue);
2399 blk_mq_free_tag_set(&md->tag_set);
2403 module_put(THIS_MODULE);
2407 static void __bind_mempools(struct mapped_device *md, struct dm_table *t)
2409 struct dm_md_mempools *p = dm_table_get_md_mempools(t);
2412 /* The md already has necessary mempools. */
2413 if (dm_table_get_type(t) == DM_TYPE_BIO_BASED) {
2415 * Reload bioset because front_pad may have changed
2416 * because a different table was loaded.
2418 bioset_free(md->bs);
2423 * There's no need to reload with request-based dm
2424 * because the size of front_pad doesn't change.
2425 * Note for future: If you are to reload bioset,
2426 * prep-ed requests in the queue may refer
2427 * to bio from the old bioset, so you must walk
2428 * through the queue to unprep.
2433 BUG_ON(!p || md->io_pool || md->rq_pool || md->bs);
2435 md->io_pool = p->io_pool;
2437 md->rq_pool = p->rq_pool;
2443 /* mempool bind completed, no longer need any mempools in the table */
2444 dm_table_free_md_mempools(t);
2448 * Bind a table to the device.
2450 static void event_callback(void *context)
2452 unsigned long flags;
2454 struct mapped_device *md = (struct mapped_device *) context;
2456 spin_lock_irqsave(&md->uevent_lock, flags);
2457 list_splice_init(&md->uevent_list, &uevents);
2458 spin_unlock_irqrestore(&md->uevent_lock, flags);
2460 dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj);
2462 atomic_inc(&md->event_nr);
2463 wake_up(&md->eventq);
2467 * Protected by md->suspend_lock obtained by dm_swap_table().
2469 static void __set_size(struct mapped_device *md, sector_t size)
2471 set_capacity(md->disk, size);
2473 i_size_write(md->bdev->bd_inode, (loff_t)size << SECTOR_SHIFT);
2477 * Return 1 if the queue has a compulsory merge_bvec_fn function.
2479 * If this function returns 0, then the device is either a non-dm
2480 * device without a merge_bvec_fn, or it is a dm device that is
2481 * able to split any bios it receives that are too big.
2483 int dm_queue_merge_is_compulsory(struct request_queue *q)
2485 struct mapped_device *dev_md;
2487 if (!q->merge_bvec_fn)
2490 if (q->make_request_fn == dm_make_request) {
2491 dev_md = q->queuedata;
2492 if (test_bit(DMF_MERGE_IS_OPTIONAL, &dev_md->flags))
2499 static int dm_device_merge_is_compulsory(struct dm_target *ti,
2500 struct dm_dev *dev, sector_t start,
2501 sector_t len, void *data)
2503 struct block_device *bdev = dev->bdev;
2504 struct request_queue *q = bdev_get_queue(bdev);
2506 return dm_queue_merge_is_compulsory(q);
2510 * Return 1 if it is acceptable to ignore merge_bvec_fn based
2511 * on the properties of the underlying devices.
2513 static int dm_table_merge_is_optional(struct dm_table *table)
2516 struct dm_target *ti;
2518 while (i < dm_table_get_num_targets(table)) {
2519 ti = dm_table_get_target(table, i++);
2521 if (ti->type->iterate_devices &&
2522 ti->type->iterate_devices(ti, dm_device_merge_is_compulsory, NULL))
2530 * Returns old map, which caller must destroy.
2532 static struct dm_table *__bind(struct mapped_device *md, struct dm_table *t,
2533 struct queue_limits *limits)
2535 struct dm_table *old_map;
2536 struct request_queue *q = md->queue;
2538 int merge_is_optional;
2540 size = dm_table_get_size(t);
2543 * Wipe any geometry if the size of the table changed.
2545 if (size != dm_get_size(md))
2546 memset(&md->geometry, 0, sizeof(md->geometry));
2548 __set_size(md, size);
2550 dm_table_event_callback(t, event_callback, md);
2553 * The queue hasn't been stopped yet, if the old table type wasn't
2554 * for request-based during suspension. So stop it to prevent
2555 * I/O mapping before resume.
2556 * This must be done before setting the queue restrictions,
2557 * because request-based dm may be run just after the setting.
2559 if (dm_table_request_based(t))
2562 __bind_mempools(md, t);
2564 merge_is_optional = dm_table_merge_is_optional(t);
2566 old_map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2567 rcu_assign_pointer(md->map, t);
2568 md->immutable_target_type = dm_table_get_immutable_target_type(t);
2570 dm_table_set_restrictions(t, q, limits);
2571 if (merge_is_optional)
2572 set_bit(DMF_MERGE_IS_OPTIONAL, &md->flags);
2574 clear_bit(DMF_MERGE_IS_OPTIONAL, &md->flags);
2582 * Returns unbound table for the caller to free.
2584 static struct dm_table *__unbind(struct mapped_device *md)
2586 struct dm_table *map = rcu_dereference_protected(md->map, 1);
2591 dm_table_event_callback(map, NULL, NULL);
2592 RCU_INIT_POINTER(md->map, NULL);
2599 * Constructor for a new device.
2601 int dm_create(int minor, struct mapped_device **result)
2603 struct mapped_device *md;
2605 md = alloc_dev(minor);
2616 * Functions to manage md->type.
2617 * All are required to hold md->type_lock.
2619 void dm_lock_md_type(struct mapped_device *md)
2621 mutex_lock(&md->type_lock);
2624 void dm_unlock_md_type(struct mapped_device *md)
2626 mutex_unlock(&md->type_lock);
2629 void dm_set_md_type(struct mapped_device *md, unsigned type)
2631 BUG_ON(!mutex_is_locked(&md->type_lock));
2635 unsigned dm_get_md_type(struct mapped_device *md)
2637 BUG_ON(!mutex_is_locked(&md->type_lock));
2641 struct target_type *dm_get_immutable_target_type(struct mapped_device *md)
2643 return md->immutable_target_type;
2647 * The queue_limits are only valid as long as you have a reference
2650 struct queue_limits *dm_get_queue_limits(struct mapped_device *md)
2652 BUG_ON(!atomic_read(&md->holders));
2653 return &md->queue->limits;
2655 EXPORT_SYMBOL_GPL(dm_get_queue_limits);
2657 static void init_rq_based_worker_thread(struct mapped_device *md)
2659 /* Initialize the request-based DM worker thread */
2660 init_kthread_worker(&md->kworker);
2661 md->kworker_task = kthread_run(kthread_worker_fn, &md->kworker,
2662 "kdmwork-%s", dm_device_name(md));
2666 * Fully initialize a request-based queue (->elevator, ->request_fn, etc).
2668 static int dm_init_request_based_queue(struct mapped_device *md)
2670 struct request_queue *q = NULL;
2672 /* Fully initialize the queue */
2673 q = blk_init_allocated_queue(md->queue, dm_request_fn, NULL);
2677 /* disable dm_request_fn's merge heuristic by default */
2678 md->seq_rq_merge_deadline_usecs = 0;
2681 dm_init_old_md_queue(md);
2682 blk_queue_softirq_done(md->queue, dm_softirq_done);
2683 blk_queue_prep_rq(md->queue, dm_prep_fn);
2685 init_rq_based_worker_thread(md);
2687 elv_register_queue(md->queue);
2692 static int dm_mq_init_request(void *data, struct request *rq,
2693 unsigned int hctx_idx, unsigned int request_idx,
2694 unsigned int numa_node)
2696 struct mapped_device *md = data;
2697 struct dm_rq_target_io *tio = blk_mq_rq_to_pdu(rq);
2700 * Must initialize md member of tio, otherwise it won't
2701 * be available in dm_mq_queue_rq.
2708 static int dm_mq_queue_rq(struct blk_mq_hw_ctx *hctx,
2709 const struct blk_mq_queue_data *bd)
2711 struct request *rq = bd->rq;
2712 struct dm_rq_target_io *tio = blk_mq_rq_to_pdu(rq);
2713 struct mapped_device *md = tio->md;
2715 struct dm_table *map = dm_get_live_table(md, &srcu_idx);
2716 struct dm_target *ti;
2719 /* always use block 0 to find the target for flushes for now */
2721 if (!(rq->cmd_flags & REQ_FLUSH))
2722 pos = blk_rq_pos(rq);
2724 ti = dm_table_find_target(map, pos);
2725 if (!dm_target_is_valid(ti)) {
2726 dm_put_live_table(md, srcu_idx);
2727 DMERR_LIMIT("request attempted access beyond the end of device");
2729 * Must perform setup, that rq_completed() requires,
2730 * before returning BLK_MQ_RQ_QUEUE_ERROR
2732 dm_start_request(md, rq);
2733 return BLK_MQ_RQ_QUEUE_ERROR;
2735 dm_put_live_table(md, srcu_idx);
2737 if (ti->type->busy && ti->type->busy(ti))
2738 return BLK_MQ_RQ_QUEUE_BUSY;
2740 dm_start_request(md, rq);
2742 /* Init tio using md established in .init_request */
2743 init_tio(tio, rq, md);
2746 * Establish tio->ti before queuing work (map_tio_request)
2747 * or making direct call to map_request().
2751 /* Clone the request if underlying devices aren't blk-mq */
2752 if (dm_table_get_type(map) == DM_TYPE_REQUEST_BASED) {
2753 /* clone request is allocated at the end of the pdu */
2754 tio->clone = (void *)blk_mq_rq_to_pdu(rq) + sizeof(struct dm_rq_target_io);
2755 (void) clone_rq(rq, md, tio, GFP_ATOMIC);
2756 queue_kthread_work(&md->kworker, &tio->work);
2758 /* Direct call is fine since .queue_rq allows allocations */
2759 if (map_request(tio, rq, md) == DM_MAPIO_REQUEUE) {
2760 /* Undo dm_start_request() before requeuing */
2761 rq_completed(md, rq_data_dir(rq), false);
2762 return BLK_MQ_RQ_QUEUE_BUSY;
2766 return BLK_MQ_RQ_QUEUE_OK;
2769 static struct blk_mq_ops dm_mq_ops = {
2770 .queue_rq = dm_mq_queue_rq,
2771 .map_queue = blk_mq_map_queue,
2772 .complete = dm_softirq_done,
2773 .init_request = dm_mq_init_request,
2776 static int dm_init_request_based_blk_mq_queue(struct mapped_device *md)
2778 unsigned md_type = dm_get_md_type(md);
2779 struct request_queue *q;
2782 memset(&md->tag_set, 0, sizeof(md->tag_set));
2783 md->tag_set.ops = &dm_mq_ops;
2784 md->tag_set.queue_depth = BLKDEV_MAX_RQ;
2785 md->tag_set.numa_node = NUMA_NO_NODE;
2786 md->tag_set.flags = BLK_MQ_F_SHOULD_MERGE | BLK_MQ_F_SG_MERGE;
2787 md->tag_set.nr_hw_queues = 1;
2788 if (md_type == DM_TYPE_REQUEST_BASED) {
2789 /* make the memory for non-blk-mq clone part of the pdu */
2790 md->tag_set.cmd_size = sizeof(struct dm_rq_target_io) + sizeof(struct request);
2792 md->tag_set.cmd_size = sizeof(struct dm_rq_target_io);
2793 md->tag_set.driver_data = md;
2795 err = blk_mq_alloc_tag_set(&md->tag_set);
2799 q = blk_mq_init_allocated_queue(&md->tag_set, md->queue);
2805 dm_init_md_queue(md);
2807 /* backfill 'mq' sysfs registration normally done in blk_register_queue */
2808 blk_mq_register_disk(md->disk);
2810 if (md_type == DM_TYPE_REQUEST_BASED)
2811 init_rq_based_worker_thread(md);
2816 blk_mq_free_tag_set(&md->tag_set);
2820 static unsigned filter_md_type(unsigned type, struct mapped_device *md)
2822 if (type == DM_TYPE_BIO_BASED)
2825 return !md->use_blk_mq ? DM_TYPE_REQUEST_BASED : DM_TYPE_MQ_REQUEST_BASED;
2829 * Setup the DM device's queue based on md's type
2831 int dm_setup_md_queue(struct mapped_device *md)
2834 unsigned md_type = filter_md_type(dm_get_md_type(md), md);
2837 case DM_TYPE_REQUEST_BASED:
2838 r = dm_init_request_based_queue(md);
2840 DMWARN("Cannot initialize queue for request-based mapped device");
2844 case DM_TYPE_MQ_REQUEST_BASED:
2845 r = dm_init_request_based_blk_mq_queue(md);
2847 DMWARN("Cannot initialize queue for request-based blk-mq mapped device");
2851 case DM_TYPE_BIO_BASED:
2852 dm_init_old_md_queue(md);
2853 blk_queue_make_request(md->queue, dm_make_request);
2854 blk_queue_merge_bvec(md->queue, dm_merge_bvec);
2861 struct mapped_device *dm_get_md(dev_t dev)
2863 struct mapped_device *md;
2864 unsigned minor = MINOR(dev);
2866 if (MAJOR(dev) != _major || minor >= (1 << MINORBITS))
2869 spin_lock(&_minor_lock);
2871 md = idr_find(&_minor_idr, minor);
2873 if ((md == MINOR_ALLOCED ||
2874 (MINOR(disk_devt(dm_disk(md))) != minor) ||
2875 dm_deleting_md(md) ||
2876 test_bit(DMF_FREEING, &md->flags))) {
2884 spin_unlock(&_minor_lock);
2888 EXPORT_SYMBOL_GPL(dm_get_md);
2890 void *dm_get_mdptr(struct mapped_device *md)
2892 return md->interface_ptr;
2895 void dm_set_mdptr(struct mapped_device *md, void *ptr)
2897 md->interface_ptr = ptr;
2900 void dm_get(struct mapped_device *md)
2902 atomic_inc(&md->holders);
2903 BUG_ON(test_bit(DMF_FREEING, &md->flags));
2906 int dm_hold(struct mapped_device *md)
2908 spin_lock(&_minor_lock);
2909 if (test_bit(DMF_FREEING, &md->flags)) {
2910 spin_unlock(&_minor_lock);
2914 spin_unlock(&_minor_lock);
2917 EXPORT_SYMBOL_GPL(dm_hold);
2919 const char *dm_device_name(struct mapped_device *md)
2923 EXPORT_SYMBOL_GPL(dm_device_name);
2925 static void __dm_destroy(struct mapped_device *md, bool wait)
2927 struct dm_table *map;
2932 map = dm_get_live_table(md, &srcu_idx);
2934 spin_lock(&_minor_lock);
2935 idr_replace(&_minor_idr, MINOR_ALLOCED, MINOR(disk_devt(dm_disk(md))));
2936 set_bit(DMF_FREEING, &md->flags);
2937 spin_unlock(&_minor_lock);
2939 if (dm_request_based(md) && md->kworker_task)
2940 flush_kthread_worker(&md->kworker);
2943 * Take suspend_lock so that presuspend and postsuspend methods
2944 * do not race with internal suspend.
2946 mutex_lock(&md->suspend_lock);
2947 if (!dm_suspended_md(md)) {
2948 dm_table_presuspend_targets(map);
2949 dm_table_postsuspend_targets(map);
2951 mutex_unlock(&md->suspend_lock);
2953 /* dm_put_live_table must be before msleep, otherwise deadlock is possible */
2954 dm_put_live_table(md, srcu_idx);
2957 * Rare, but there may be I/O requests still going to complete,
2958 * for example. Wait for all references to disappear.
2959 * No one should increment the reference count of the mapped_device,
2960 * after the mapped_device state becomes DMF_FREEING.
2963 while (atomic_read(&md->holders))
2965 else if (atomic_read(&md->holders))
2966 DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)",
2967 dm_device_name(md), atomic_read(&md->holders));
2970 dm_table_destroy(__unbind(md));
2974 void dm_destroy(struct mapped_device *md)
2976 __dm_destroy(md, true);
2979 void dm_destroy_immediate(struct mapped_device *md)
2981 __dm_destroy(md, false);
2984 void dm_put(struct mapped_device *md)
2986 atomic_dec(&md->holders);
2988 EXPORT_SYMBOL_GPL(dm_put);
2990 static int dm_wait_for_completion(struct mapped_device *md, int interruptible)
2993 DECLARE_WAITQUEUE(wait, current);
2995 add_wait_queue(&md->wait, &wait);
2998 set_current_state(interruptible);
3000 if (!md_in_flight(md))
3003 if (interruptible == TASK_INTERRUPTIBLE &&
3004 signal_pending(current)) {
3011 set_current_state(TASK_RUNNING);
3013 remove_wait_queue(&md->wait, &wait);
3019 * Process the deferred bios
3021 static void dm_wq_work(struct work_struct *work)
3023 struct mapped_device *md = container_of(work, struct mapped_device,
3027 struct dm_table *map;
3029 map = dm_get_live_table(md, &srcu_idx);
3031 while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
3032 spin_lock_irq(&md->deferred_lock);
3033 c = bio_list_pop(&md->deferred);
3034 spin_unlock_irq(&md->deferred_lock);
3039 if (dm_request_based(md))
3040 generic_make_request(c);
3042 __split_and_process_bio(md, map, c);
3045 dm_put_live_table(md, srcu_idx);
3048 static void dm_queue_flush(struct mapped_device *md)
3050 clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
3051 smp_mb__after_atomic();
3052 queue_work(md->wq, &md->work);
3056 * Swap in a new table, returning the old one for the caller to destroy.
3058 struct dm_table *dm_swap_table(struct mapped_device *md, struct dm_table *table)
3060 struct dm_table *live_map = NULL, *map = ERR_PTR(-EINVAL);
3061 struct queue_limits limits;
3064 mutex_lock(&md->suspend_lock);
3066 /* device must be suspended */
3067 if (!dm_suspended_md(md))
3071 * If the new table has no data devices, retain the existing limits.
3072 * This helps multipath with queue_if_no_path if all paths disappear,
3073 * then new I/O is queued based on these limits, and then some paths
3076 if (dm_table_has_no_data_devices(table)) {
3077 live_map = dm_get_live_table_fast(md);
3079 limits = md->queue->limits;
3080 dm_put_live_table_fast(md);
3084 r = dm_calculate_queue_limits(table, &limits);
3091 map = __bind(md, table, &limits);
3094 mutex_unlock(&md->suspend_lock);
3099 * Functions to lock and unlock any filesystem running on the
3102 static int lock_fs(struct mapped_device *md)
3106 WARN_ON(md->frozen_sb);
3108 md->frozen_sb = freeze_bdev(md->bdev);
3109 if (IS_ERR(md->frozen_sb)) {
3110 r = PTR_ERR(md->frozen_sb);
3111 md->frozen_sb = NULL;
3115 set_bit(DMF_FROZEN, &md->flags);
3120 static void unlock_fs(struct mapped_device *md)
3122 if (!test_bit(DMF_FROZEN, &md->flags))
3125 thaw_bdev(md->bdev, md->frozen_sb);
3126 md->frozen_sb = NULL;
3127 clear_bit(DMF_FROZEN, &md->flags);
3131 * If __dm_suspend returns 0, the device is completely quiescent
3132 * now. There is no request-processing activity. All new requests
3133 * are being added to md->deferred list.
3135 * Caller must hold md->suspend_lock
3137 static int __dm_suspend(struct mapped_device *md, struct dm_table *map,
3138 unsigned suspend_flags, int interruptible)
3140 bool do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG;
3141 bool noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG;
3145 * DMF_NOFLUSH_SUSPENDING must be set before presuspend.
3146 * This flag is cleared before dm_suspend returns.
3149 set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
3152 * This gets reverted if there's an error later and the targets
3153 * provide the .presuspend_undo hook.
3155 dm_table_presuspend_targets(map);
3158 * Flush I/O to the device.
3159 * Any I/O submitted after lock_fs() may not be flushed.
3160 * noflush takes precedence over do_lockfs.
3161 * (lock_fs() flushes I/Os and waits for them to complete.)
3163 if (!noflush && do_lockfs) {
3166 dm_table_presuspend_undo_targets(map);
3172 * Here we must make sure that no processes are submitting requests
3173 * to target drivers i.e. no one may be executing
3174 * __split_and_process_bio. This is called from dm_request and
3177 * To get all processes out of __split_and_process_bio in dm_request,
3178 * we take the write lock. To prevent any process from reentering
3179 * __split_and_process_bio from dm_request and quiesce the thread
3180 * (dm_wq_work), we set BMF_BLOCK_IO_FOR_SUSPEND and call
3181 * flush_workqueue(md->wq).
3183 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
3185 synchronize_srcu(&md->io_barrier);
3188 * Stop md->queue before flushing md->wq in case request-based
3189 * dm defers requests to md->wq from md->queue.
3191 if (dm_request_based(md)) {
3192 stop_queue(md->queue);
3193 if (md->kworker_task)
3194 flush_kthread_worker(&md->kworker);
3197 flush_workqueue(md->wq);
3200 * At this point no more requests are entering target request routines.
3201 * We call dm_wait_for_completion to wait for all existing requests
3204 r = dm_wait_for_completion(md, interruptible);
3207 clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
3209 synchronize_srcu(&md->io_barrier);
3211 /* were we interrupted ? */
3215 if (dm_request_based(md))
3216 start_queue(md->queue);
3219 dm_table_presuspend_undo_targets(map);
3220 /* pushback list is already flushed, so skip flush */
3227 * We need to be able to change a mapping table under a mounted
3228 * filesystem. For example we might want to move some data in
3229 * the background. Before the table can be swapped with
3230 * dm_bind_table, dm_suspend must be called to flush any in
3231 * flight bios and ensure that any further io gets deferred.
3234 * Suspend mechanism in request-based dm.
3236 * 1. Flush all I/Os by lock_fs() if needed.
3237 * 2. Stop dispatching any I/O by stopping the request_queue.
3238 * 3. Wait for all in-flight I/Os to be completed or requeued.
3240 * To abort suspend, start the request_queue.
3242 int dm_suspend(struct mapped_device *md, unsigned suspend_flags)
3244 struct dm_table *map = NULL;
3248 mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING);
3250 if (dm_suspended_md(md)) {
3255 if (dm_suspended_internally_md(md)) {
3256 /* already internally suspended, wait for internal resume */
3257 mutex_unlock(&md->suspend_lock);
3258 r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE);
3264 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
3266 r = __dm_suspend(md, map, suspend_flags, TASK_INTERRUPTIBLE);
3270 set_bit(DMF_SUSPENDED, &md->flags);
3272 dm_table_postsuspend_targets(map);
3275 mutex_unlock(&md->suspend_lock);
3279 static int __dm_resume(struct mapped_device *md, struct dm_table *map)
3282 int r = dm_table_resume_targets(map);
3290 * Flushing deferred I/Os must be done after targets are resumed
3291 * so that mapping of targets can work correctly.
3292 * Request-based dm is queueing the deferred I/Os in its request_queue.
3294 if (dm_request_based(md))
3295 start_queue(md->queue);
3302 int dm_resume(struct mapped_device *md)
3305 struct dm_table *map = NULL;
3308 mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING);
3310 if (!dm_suspended_md(md))
3313 if (dm_suspended_internally_md(md)) {
3314 /* already internally suspended, wait for internal resume */
3315 mutex_unlock(&md->suspend_lock);
3316 r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE);
3322 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
3323 if (!map || !dm_table_get_size(map))
3326 r = __dm_resume(md, map);
3330 clear_bit(DMF_SUSPENDED, &md->flags);
3334 mutex_unlock(&md->suspend_lock);
3340 * Internal suspend/resume works like userspace-driven suspend. It waits
3341 * until all bios finish and prevents issuing new bios to the target drivers.
3342 * It may be used only from the kernel.
3345 static void __dm_internal_suspend(struct mapped_device *md, unsigned suspend_flags)
3347 struct dm_table *map = NULL;
3349 if (md->internal_suspend_count++)
3350 return; /* nested internal suspend */
3352 if (dm_suspended_md(md)) {
3353 set_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
3354 return; /* nest suspend */
3357 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
3360 * Using TASK_UNINTERRUPTIBLE because only NOFLUSH internal suspend is
3361 * supported. Properly supporting a TASK_INTERRUPTIBLE internal suspend
3362 * would require changing .presuspend to return an error -- avoid this
3363 * until there is a need for more elaborate variants of internal suspend.
3365 (void) __dm_suspend(md, map, suspend_flags, TASK_UNINTERRUPTIBLE);
3367 set_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
3369 dm_table_postsuspend_targets(map);
3372 static void __dm_internal_resume(struct mapped_device *md)
3374 BUG_ON(!md->internal_suspend_count);
3376 if (--md->internal_suspend_count)
3377 return; /* resume from nested internal suspend */
3379 if (dm_suspended_md(md))
3380 goto done; /* resume from nested suspend */
3383 * NOTE: existing callers don't need to call dm_table_resume_targets
3384 * (which may fail -- so best to avoid it for now by passing NULL map)
3386 (void) __dm_resume(md, NULL);
3389 clear_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
3390 smp_mb__after_atomic();
3391 wake_up_bit(&md->flags, DMF_SUSPENDED_INTERNALLY);
3394 void dm_internal_suspend_noflush(struct mapped_device *md)
3396 mutex_lock(&md->suspend_lock);
3397 __dm_internal_suspend(md, DM_SUSPEND_NOFLUSH_FLAG);
3398 mutex_unlock(&md->suspend_lock);
3400 EXPORT_SYMBOL_GPL(dm_internal_suspend_noflush);
3402 void dm_internal_resume(struct mapped_device *md)
3404 mutex_lock(&md->suspend_lock);
3405 __dm_internal_resume(md);
3406 mutex_unlock(&md->suspend_lock);
3408 EXPORT_SYMBOL_GPL(dm_internal_resume);
3411 * Fast variants of internal suspend/resume hold md->suspend_lock,
3412 * which prevents interaction with userspace-driven suspend.
3415 void dm_internal_suspend_fast(struct mapped_device *md)
3417 mutex_lock(&md->suspend_lock);
3418 if (dm_suspended_md(md) || dm_suspended_internally_md(md))
3421 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
3422 synchronize_srcu(&md->io_barrier);
3423 flush_workqueue(md->wq);
3424 dm_wait_for_completion(md, TASK_UNINTERRUPTIBLE);
3426 EXPORT_SYMBOL_GPL(dm_internal_suspend_fast);
3428 void dm_internal_resume_fast(struct mapped_device *md)
3430 if (dm_suspended_md(md) || dm_suspended_internally_md(md))
3436 mutex_unlock(&md->suspend_lock);
3438 EXPORT_SYMBOL_GPL(dm_internal_resume_fast);
3440 /*-----------------------------------------------------------------
3441 * Event notification.
3442 *---------------------------------------------------------------*/
3443 int dm_kobject_uevent(struct mapped_device *md, enum kobject_action action,
3446 char udev_cookie[DM_COOKIE_LENGTH];
3447 char *envp[] = { udev_cookie, NULL };
3450 return kobject_uevent(&disk_to_dev(md->disk)->kobj, action);
3452 snprintf(udev_cookie, DM_COOKIE_LENGTH, "%s=%u",
3453 DM_COOKIE_ENV_VAR_NAME, cookie);
3454 return kobject_uevent_env(&disk_to_dev(md->disk)->kobj,
3459 uint32_t dm_next_uevent_seq(struct mapped_device *md)
3461 return atomic_add_return(1, &md->uevent_seq);
3464 uint32_t dm_get_event_nr(struct mapped_device *md)
3466 return atomic_read(&md->event_nr);
3469 int dm_wait_event(struct mapped_device *md, int event_nr)
3471 return wait_event_interruptible(md->eventq,
3472 (event_nr != atomic_read(&md->event_nr)));
3475 void dm_uevent_add(struct mapped_device *md, struct list_head *elist)
3477 unsigned long flags;
3479 spin_lock_irqsave(&md->uevent_lock, flags);
3480 list_add(elist, &md->uevent_list);
3481 spin_unlock_irqrestore(&md->uevent_lock, flags);
3485 * The gendisk is only valid as long as you have a reference
3488 struct gendisk *dm_disk(struct mapped_device *md)
3492 EXPORT_SYMBOL_GPL(dm_disk);
3494 struct kobject *dm_kobject(struct mapped_device *md)
3496 return &md->kobj_holder.kobj;
3499 struct mapped_device *dm_get_from_kobject(struct kobject *kobj)
3501 struct mapped_device *md;
3503 md = container_of(kobj, struct mapped_device, kobj_holder.kobj);
3505 if (test_bit(DMF_FREEING, &md->flags) ||
3513 int dm_suspended_md(struct mapped_device *md)
3515 return test_bit(DMF_SUSPENDED, &md->flags);
3518 int dm_suspended_internally_md(struct mapped_device *md)
3520 return test_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
3523 int dm_test_deferred_remove_flag(struct mapped_device *md)
3525 return test_bit(DMF_DEFERRED_REMOVE, &md->flags);
3528 int dm_suspended(struct dm_target *ti)
3530 return dm_suspended_md(dm_table_get_md(ti->table));
3532 EXPORT_SYMBOL_GPL(dm_suspended);
3534 int dm_noflush_suspending(struct dm_target *ti)
3536 return __noflush_suspending(dm_table_get_md(ti->table));
3538 EXPORT_SYMBOL_GPL(dm_noflush_suspending);
3540 struct dm_md_mempools *dm_alloc_md_mempools(struct mapped_device *md, unsigned type,
3541 unsigned integrity, unsigned per_bio_data_size)
3543 struct dm_md_mempools *pools = kzalloc(sizeof(*pools), GFP_KERNEL);
3544 struct kmem_cache *cachep = NULL;
3545 unsigned int pool_size = 0;
3546 unsigned int front_pad;
3551 type = filter_md_type(type, md);
3554 case DM_TYPE_BIO_BASED:
3556 pool_size = dm_get_reserved_bio_based_ios();
3557 front_pad = roundup(per_bio_data_size, __alignof__(struct dm_target_io)) + offsetof(struct dm_target_io, clone);
3559 case DM_TYPE_REQUEST_BASED:
3560 cachep = _rq_tio_cache;
3561 pool_size = dm_get_reserved_rq_based_ios();
3562 pools->rq_pool = mempool_create_slab_pool(pool_size, _rq_cache);
3563 if (!pools->rq_pool)
3565 /* fall through to setup remaining rq-based pools */
3566 case DM_TYPE_MQ_REQUEST_BASED:
3568 pool_size = dm_get_reserved_rq_based_ios();
3569 front_pad = offsetof(struct dm_rq_clone_bio_info, clone);
3570 /* per_bio_data_size is not used. See __bind_mempools(). */
3571 WARN_ON(per_bio_data_size != 0);
3578 pools->io_pool = mempool_create_slab_pool(pool_size, cachep);
3579 if (!pools->io_pool)
3583 pools->bs = bioset_create_nobvec(pool_size, front_pad);
3587 if (integrity && bioset_integrity_create(pools->bs, pool_size))
3593 dm_free_md_mempools(pools);
3598 void dm_free_md_mempools(struct dm_md_mempools *pools)
3604 mempool_destroy(pools->io_pool);
3607 mempool_destroy(pools->rq_pool);
3610 bioset_free(pools->bs);
3615 static const struct block_device_operations dm_blk_dops = {
3616 .open = dm_blk_open,
3617 .release = dm_blk_close,
3618 .ioctl = dm_blk_ioctl,
3619 .getgeo = dm_blk_getgeo,
3620 .owner = THIS_MODULE
3626 module_init(dm_init);
3627 module_exit(dm_exit);
3629 module_param(major, uint, 0);
3630 MODULE_PARM_DESC(major, "The major number of the device mapper");
3632 module_param(reserved_bio_based_ios, uint, S_IRUGO | S_IWUSR);
3633 MODULE_PARM_DESC(reserved_bio_based_ios, "Reserved IOs in bio-based mempools");
3635 module_param(reserved_rq_based_ios, uint, S_IRUGO | S_IWUSR);
3636 MODULE_PARM_DESC(reserved_rq_based_ios, "Reserved IOs in request-based mempools");
3638 module_param(use_blk_mq, bool, S_IRUGO | S_IWUSR);
3639 MODULE_PARM_DESC(use_blk_mq, "Use block multiqueue for request-based DM devices");
3641 MODULE_DESCRIPTION(DM_NAME " driver");
3642 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
3643 MODULE_LICENSE("GPL");