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, max_size = 0;
1731 ti = dm_table_find_target(map, bvm->bi_sector);
1732 if (!dm_target_is_valid(ti))
1736 * Find maximum amount of I/O that won't need splitting
1738 max_sectors = min(max_io_len(bvm->bi_sector, ti),
1739 (sector_t) queue_max_sectors(q));
1740 max_size = (max_sectors << SECTOR_SHIFT) - bvm->bi_size;
1743 * FIXME: this stop-gap fix _must_ be cleaned up (by passing a sector_t
1744 * to the targets' merge function since it holds sectors not bytes).
1745 * Just doing this as an interim fix for stable@ because the more
1746 * comprehensive cleanup of switching to sector_t will impact every
1747 * DM target that implements a ->merge hook.
1749 if (max_size > INT_MAX)
1753 * merge_bvec_fn() returns number of bytes
1754 * it can accept at this offset
1755 * max is precomputed maximal io size
1757 if (max_size && ti->type->merge)
1758 max_size = ti->type->merge(ti, bvm, biovec, (int) max_size);
1760 * If the target doesn't support merge method and some of the devices
1761 * provided their merge_bvec method (we know this by looking for the
1762 * max_hw_sectors that dm_set_device_limits may set), then we can't
1763 * allow bios with multiple vector entries. So always set max_size
1764 * to 0, and the code below allows just one page.
1766 else if (queue_max_hw_sectors(q) <= PAGE_SIZE >> 9)
1770 dm_put_live_table_fast(md);
1772 * Always allow an entire first page
1774 if (max_size <= biovec->bv_len && !(bvm->bi_size >> SECTOR_SHIFT))
1775 max_size = biovec->bv_len;
1781 * The request function that just remaps the bio built up by
1784 static void dm_make_request(struct request_queue *q, struct bio *bio)
1786 int rw = bio_data_dir(bio);
1787 struct mapped_device *md = q->queuedata;
1789 struct dm_table *map;
1791 map = dm_get_live_table(md, &srcu_idx);
1793 generic_start_io_acct(rw, bio_sectors(bio), &dm_disk(md)->part0);
1795 /* if we're suspended, we have to queue this io for later */
1796 if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))) {
1797 dm_put_live_table(md, srcu_idx);
1799 if (bio_rw(bio) != READA)
1806 __split_and_process_bio(md, map, bio);
1807 dm_put_live_table(md, srcu_idx);
1811 int dm_request_based(struct mapped_device *md)
1813 return blk_queue_stackable(md->queue);
1816 static void dm_dispatch_clone_request(struct request *clone, struct request *rq)
1820 if (blk_queue_io_stat(clone->q))
1821 clone->cmd_flags |= REQ_IO_STAT;
1823 clone->start_time = jiffies;
1824 r = blk_insert_cloned_request(clone->q, clone);
1826 /* must complete clone in terms of original request */
1827 dm_complete_request(rq, r);
1830 static int dm_rq_bio_constructor(struct bio *bio, struct bio *bio_orig,
1833 struct dm_rq_target_io *tio = data;
1834 struct dm_rq_clone_bio_info *info =
1835 container_of(bio, struct dm_rq_clone_bio_info, clone);
1837 info->orig = bio_orig;
1839 bio->bi_end_io = end_clone_bio;
1844 static int setup_clone(struct request *clone, struct request *rq,
1845 struct dm_rq_target_io *tio, gfp_t gfp_mask)
1849 r = blk_rq_prep_clone(clone, rq, tio->md->bs, gfp_mask,
1850 dm_rq_bio_constructor, tio);
1854 clone->cmd = rq->cmd;
1855 clone->cmd_len = rq->cmd_len;
1856 clone->sense = rq->sense;
1857 clone->end_io = end_clone_request;
1858 clone->end_io_data = tio;
1865 static struct request *clone_rq(struct request *rq, struct mapped_device *md,
1866 struct dm_rq_target_io *tio, gfp_t gfp_mask)
1869 * Do not allocate a clone if tio->clone was already set
1870 * (see: dm_mq_queue_rq).
1872 bool alloc_clone = !tio->clone;
1873 struct request *clone;
1876 clone = alloc_clone_request(md, gfp_mask);
1882 blk_rq_init(NULL, clone);
1883 if (setup_clone(clone, rq, tio, gfp_mask)) {
1886 free_clone_request(md, clone);
1893 static void map_tio_request(struct kthread_work *work);
1895 static void init_tio(struct dm_rq_target_io *tio, struct request *rq,
1896 struct mapped_device *md)
1903 memset(&tio->info, 0, sizeof(tio->info));
1904 if (md->kworker_task)
1905 init_kthread_work(&tio->work, map_tio_request);
1908 static struct dm_rq_target_io *prep_tio(struct request *rq,
1909 struct mapped_device *md, gfp_t gfp_mask)
1911 struct dm_rq_target_io *tio;
1913 struct dm_table *table;
1915 tio = alloc_rq_tio(md, gfp_mask);
1919 init_tio(tio, rq, md);
1921 table = dm_get_live_table(md, &srcu_idx);
1922 if (!dm_table_mq_request_based(table)) {
1923 if (!clone_rq(rq, md, tio, gfp_mask)) {
1924 dm_put_live_table(md, srcu_idx);
1929 dm_put_live_table(md, srcu_idx);
1935 * Called with the queue lock held.
1937 static int dm_prep_fn(struct request_queue *q, struct request *rq)
1939 struct mapped_device *md = q->queuedata;
1940 struct dm_rq_target_io *tio;
1942 if (unlikely(rq->special)) {
1943 DMWARN("Already has something in rq->special.");
1944 return BLKPREP_KILL;
1947 tio = prep_tio(rq, md, GFP_ATOMIC);
1949 return BLKPREP_DEFER;
1952 rq->cmd_flags |= REQ_DONTPREP;
1959 * 0 : the request has been processed
1960 * DM_MAPIO_REQUEUE : the original request needs to be requeued
1961 * < 0 : the request was completed due to failure
1963 static int map_request(struct dm_rq_target_io *tio, struct request *rq,
1964 struct mapped_device *md)
1967 struct dm_target *ti = tio->ti;
1968 struct request *clone = NULL;
1972 r = ti->type->map_rq(ti, clone, &tio->info);
1974 r = ti->type->clone_and_map_rq(ti, rq, &tio->info, &clone);
1976 /* The target wants to complete the I/O */
1977 dm_kill_unmapped_request(rq, r);
1980 if (r != DM_MAPIO_REMAPPED)
1982 if (setup_clone(clone, rq, tio, GFP_ATOMIC)) {
1984 ti->type->release_clone_rq(clone);
1985 return DM_MAPIO_REQUEUE;
1990 case DM_MAPIO_SUBMITTED:
1991 /* The target has taken the I/O to submit by itself later */
1993 case DM_MAPIO_REMAPPED:
1994 /* The target has remapped the I/O so dispatch it */
1995 trace_block_rq_remap(clone->q, clone, disk_devt(dm_disk(md)),
1997 dm_dispatch_clone_request(clone, rq);
1999 case DM_MAPIO_REQUEUE:
2000 /* The target wants to requeue the I/O */
2001 dm_requeue_unmapped_request(clone);
2005 DMWARN("unimplemented target map return value: %d", r);
2009 /* The target wants to complete the I/O */
2010 dm_kill_unmapped_request(rq, r);
2017 static void map_tio_request(struct kthread_work *work)
2019 struct dm_rq_target_io *tio = container_of(work, struct dm_rq_target_io, work);
2020 struct request *rq = tio->orig;
2021 struct mapped_device *md = tio->md;
2023 if (map_request(tio, rq, md) == DM_MAPIO_REQUEUE)
2024 dm_requeue_unmapped_original_request(md, rq);
2027 static void dm_start_request(struct mapped_device *md, struct request *orig)
2029 if (!orig->q->mq_ops)
2030 blk_start_request(orig);
2032 blk_mq_start_request(orig);
2033 atomic_inc(&md->pending[rq_data_dir(orig)]);
2035 if (md->seq_rq_merge_deadline_usecs) {
2036 md->last_rq_pos = rq_end_sector(orig);
2037 md->last_rq_rw = rq_data_dir(orig);
2038 md->last_rq_start_time = ktime_get();
2042 * Hold the md reference here for the in-flight I/O.
2043 * We can't rely on the reference count by device opener,
2044 * because the device may be closed during the request completion
2045 * when all bios are completed.
2046 * See the comment in rq_completed() too.
2051 #define MAX_SEQ_RQ_MERGE_DEADLINE_USECS 100000
2053 ssize_t dm_attr_rq_based_seq_io_merge_deadline_show(struct mapped_device *md, char *buf)
2055 return sprintf(buf, "%u\n", md->seq_rq_merge_deadline_usecs);
2058 ssize_t dm_attr_rq_based_seq_io_merge_deadline_store(struct mapped_device *md,
2059 const char *buf, size_t count)
2063 if (!dm_request_based(md) || md->use_blk_mq)
2066 if (kstrtouint(buf, 10, &deadline))
2069 if (deadline > MAX_SEQ_RQ_MERGE_DEADLINE_USECS)
2070 deadline = MAX_SEQ_RQ_MERGE_DEADLINE_USECS;
2072 md->seq_rq_merge_deadline_usecs = deadline;
2077 static bool dm_request_peeked_before_merge_deadline(struct mapped_device *md)
2079 ktime_t kt_deadline;
2081 if (!md->seq_rq_merge_deadline_usecs)
2084 kt_deadline = ns_to_ktime((u64)md->seq_rq_merge_deadline_usecs * NSEC_PER_USEC);
2085 kt_deadline = ktime_add_safe(md->last_rq_start_time, kt_deadline);
2087 return !ktime_after(ktime_get(), kt_deadline);
2091 * q->request_fn for request-based dm.
2092 * Called with the queue lock held.
2094 static void dm_request_fn(struct request_queue *q)
2096 struct mapped_device *md = q->queuedata;
2098 struct dm_table *map = dm_get_live_table(md, &srcu_idx);
2099 struct dm_target *ti;
2101 struct dm_rq_target_io *tio;
2105 * For suspend, check blk_queue_stopped() and increment
2106 * ->pending within a single queue_lock not to increment the
2107 * number of in-flight I/Os after the queue is stopped in
2110 while (!blk_queue_stopped(q)) {
2111 rq = blk_peek_request(q);
2115 /* always use block 0 to find the target for flushes for now */
2117 if (!(rq->cmd_flags & REQ_FLUSH))
2118 pos = blk_rq_pos(rq);
2120 ti = dm_table_find_target(map, pos);
2121 if (!dm_target_is_valid(ti)) {
2123 * Must perform setup, that rq_completed() requires,
2124 * before calling dm_kill_unmapped_request
2126 DMERR_LIMIT("request attempted access beyond the end of device");
2127 dm_start_request(md, rq);
2128 dm_kill_unmapped_request(rq, -EIO);
2132 if (dm_request_peeked_before_merge_deadline(md) &&
2133 md_in_flight(md) && rq->bio && rq->bio->bi_vcnt == 1 &&
2134 md->last_rq_pos == pos && md->last_rq_rw == rq_data_dir(rq))
2137 if (ti->type->busy && ti->type->busy(ti))
2140 dm_start_request(md, rq);
2142 tio = tio_from_request(rq);
2143 /* Establish tio->ti before queuing work (map_tio_request) */
2145 queue_kthread_work(&md->kworker, &tio->work);
2146 BUG_ON(!irqs_disabled());
2152 blk_delay_queue(q, HZ / 100);
2154 dm_put_live_table(md, srcu_idx);
2157 static int dm_any_congested(void *congested_data, int bdi_bits)
2160 struct mapped_device *md = congested_data;
2161 struct dm_table *map;
2163 if (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
2164 map = dm_get_live_table_fast(md);
2167 * Request-based dm cares about only own queue for
2168 * the query about congestion status of request_queue
2170 if (dm_request_based(md))
2171 r = md->queue->backing_dev_info.state &
2174 r = dm_table_any_congested(map, bdi_bits);
2176 dm_put_live_table_fast(md);
2182 /*-----------------------------------------------------------------
2183 * An IDR is used to keep track of allocated minor numbers.
2184 *---------------------------------------------------------------*/
2185 static void free_minor(int minor)
2187 spin_lock(&_minor_lock);
2188 idr_remove(&_minor_idr, minor);
2189 spin_unlock(&_minor_lock);
2193 * See if the device with a specific minor # is free.
2195 static int specific_minor(int minor)
2199 if (minor >= (1 << MINORBITS))
2202 idr_preload(GFP_KERNEL);
2203 spin_lock(&_minor_lock);
2205 r = idr_alloc(&_minor_idr, MINOR_ALLOCED, minor, minor + 1, GFP_NOWAIT);
2207 spin_unlock(&_minor_lock);
2210 return r == -ENOSPC ? -EBUSY : r;
2214 static int next_free_minor(int *minor)
2218 idr_preload(GFP_KERNEL);
2219 spin_lock(&_minor_lock);
2221 r = idr_alloc(&_minor_idr, MINOR_ALLOCED, 0, 1 << MINORBITS, GFP_NOWAIT);
2223 spin_unlock(&_minor_lock);
2231 static const struct block_device_operations dm_blk_dops;
2233 static void dm_wq_work(struct work_struct *work);
2235 static void dm_init_md_queue(struct mapped_device *md)
2238 * Request-based dm devices cannot be stacked on top of bio-based dm
2239 * devices. The type of this dm device may not have been decided yet.
2240 * The type is decided at the first table loading time.
2241 * To prevent problematic device stacking, clear the queue flag
2242 * for request stacking support until then.
2244 * This queue is new, so no concurrency on the queue_flags.
2246 queue_flag_clear_unlocked(QUEUE_FLAG_STACKABLE, md->queue);
2249 static void dm_init_old_md_queue(struct mapped_device *md)
2251 md->use_blk_mq = false;
2252 dm_init_md_queue(md);
2255 * Initialize aspects of queue that aren't relevant for blk-mq
2257 md->queue->queuedata = md;
2258 md->queue->backing_dev_info.congested_fn = dm_any_congested;
2259 md->queue->backing_dev_info.congested_data = md;
2261 blk_queue_bounce_limit(md->queue, BLK_BOUNCE_ANY);
2265 * Allocate and initialise a blank device with a given minor.
2267 static struct mapped_device *alloc_dev(int minor)
2270 struct mapped_device *md = kzalloc(sizeof(*md), GFP_KERNEL);
2274 DMWARN("unable to allocate device, out of memory.");
2278 if (!try_module_get(THIS_MODULE))
2279 goto bad_module_get;
2281 /* get a minor number for the dev */
2282 if (minor == DM_ANY_MINOR)
2283 r = next_free_minor(&minor);
2285 r = specific_minor(minor);
2289 r = init_srcu_struct(&md->io_barrier);
2291 goto bad_io_barrier;
2293 md->use_blk_mq = use_blk_mq;
2294 md->type = DM_TYPE_NONE;
2295 mutex_init(&md->suspend_lock);
2296 mutex_init(&md->type_lock);
2297 mutex_init(&md->table_devices_lock);
2298 spin_lock_init(&md->deferred_lock);
2299 atomic_set(&md->holders, 1);
2300 atomic_set(&md->open_count, 0);
2301 atomic_set(&md->event_nr, 0);
2302 atomic_set(&md->uevent_seq, 0);
2303 INIT_LIST_HEAD(&md->uevent_list);
2304 INIT_LIST_HEAD(&md->table_devices);
2305 spin_lock_init(&md->uevent_lock);
2307 md->queue = blk_alloc_queue(GFP_KERNEL);
2311 dm_init_md_queue(md);
2313 md->disk = alloc_disk(1);
2317 atomic_set(&md->pending[0], 0);
2318 atomic_set(&md->pending[1], 0);
2319 init_waitqueue_head(&md->wait);
2320 INIT_WORK(&md->work, dm_wq_work);
2321 init_waitqueue_head(&md->eventq);
2322 init_completion(&md->kobj_holder.completion);
2323 md->kworker_task = NULL;
2325 md->disk->major = _major;
2326 md->disk->first_minor = minor;
2327 md->disk->fops = &dm_blk_dops;
2328 md->disk->queue = md->queue;
2329 md->disk->private_data = md;
2330 sprintf(md->disk->disk_name, "dm-%d", minor);
2332 format_dev_t(md->name, MKDEV(_major, minor));
2334 md->wq = alloc_workqueue("kdmflush", WQ_MEM_RECLAIM, 0);
2338 md->bdev = bdget_disk(md->disk, 0);
2342 bio_init(&md->flush_bio);
2343 md->flush_bio.bi_bdev = md->bdev;
2344 md->flush_bio.bi_rw = WRITE_FLUSH;
2346 dm_stats_init(&md->stats);
2348 /* Populate the mapping, nobody knows we exist yet */
2349 spin_lock(&_minor_lock);
2350 old_md = idr_replace(&_minor_idr, md, minor);
2351 spin_unlock(&_minor_lock);
2353 BUG_ON(old_md != MINOR_ALLOCED);
2358 destroy_workqueue(md->wq);
2360 del_gendisk(md->disk);
2363 blk_cleanup_queue(md->queue);
2365 cleanup_srcu_struct(&md->io_barrier);
2369 module_put(THIS_MODULE);
2375 static void unlock_fs(struct mapped_device *md);
2377 static void free_dev(struct mapped_device *md)
2379 int minor = MINOR(disk_devt(md->disk));
2382 destroy_workqueue(md->wq);
2384 if (md->kworker_task)
2385 kthread_stop(md->kworker_task);
2387 mempool_destroy(md->io_pool);
2389 mempool_destroy(md->rq_pool);
2391 bioset_free(md->bs);
2393 cleanup_srcu_struct(&md->io_barrier);
2394 free_table_devices(&md->table_devices);
2395 dm_stats_cleanup(&md->stats);
2397 spin_lock(&_minor_lock);
2398 md->disk->private_data = NULL;
2399 spin_unlock(&_minor_lock);
2400 if (blk_get_integrity(md->disk))
2401 blk_integrity_unregister(md->disk);
2402 del_gendisk(md->disk);
2404 blk_cleanup_queue(md->queue);
2406 blk_mq_free_tag_set(&md->tag_set);
2410 module_put(THIS_MODULE);
2414 static void __bind_mempools(struct mapped_device *md, struct dm_table *t)
2416 struct dm_md_mempools *p = dm_table_get_md_mempools(t);
2419 /* The md already has necessary mempools. */
2420 if (dm_table_get_type(t) == DM_TYPE_BIO_BASED) {
2422 * Reload bioset because front_pad may have changed
2423 * because a different table was loaded.
2425 bioset_free(md->bs);
2430 * There's no need to reload with request-based dm
2431 * because the size of front_pad doesn't change.
2432 * Note for future: If you are to reload bioset,
2433 * prep-ed requests in the queue may refer
2434 * to bio from the old bioset, so you must walk
2435 * through the queue to unprep.
2440 BUG_ON(!p || md->io_pool || md->rq_pool || md->bs);
2442 md->io_pool = p->io_pool;
2444 md->rq_pool = p->rq_pool;
2450 /* mempool bind completed, no longer need any mempools in the table */
2451 dm_table_free_md_mempools(t);
2455 * Bind a table to the device.
2457 static void event_callback(void *context)
2459 unsigned long flags;
2461 struct mapped_device *md = (struct mapped_device *) context;
2463 spin_lock_irqsave(&md->uevent_lock, flags);
2464 list_splice_init(&md->uevent_list, &uevents);
2465 spin_unlock_irqrestore(&md->uevent_lock, flags);
2467 dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj);
2469 atomic_inc(&md->event_nr);
2470 wake_up(&md->eventq);
2474 * Protected by md->suspend_lock obtained by dm_swap_table().
2476 static void __set_size(struct mapped_device *md, sector_t size)
2478 set_capacity(md->disk, size);
2480 i_size_write(md->bdev->bd_inode, (loff_t)size << SECTOR_SHIFT);
2484 * Return 1 if the queue has a compulsory merge_bvec_fn function.
2486 * If this function returns 0, then the device is either a non-dm
2487 * device without a merge_bvec_fn, or it is a dm device that is
2488 * able to split any bios it receives that are too big.
2490 int dm_queue_merge_is_compulsory(struct request_queue *q)
2492 struct mapped_device *dev_md;
2494 if (!q->merge_bvec_fn)
2497 if (q->make_request_fn == dm_make_request) {
2498 dev_md = q->queuedata;
2499 if (test_bit(DMF_MERGE_IS_OPTIONAL, &dev_md->flags))
2506 static int dm_device_merge_is_compulsory(struct dm_target *ti,
2507 struct dm_dev *dev, sector_t start,
2508 sector_t len, void *data)
2510 struct block_device *bdev = dev->bdev;
2511 struct request_queue *q = bdev_get_queue(bdev);
2513 return dm_queue_merge_is_compulsory(q);
2517 * Return 1 if it is acceptable to ignore merge_bvec_fn based
2518 * on the properties of the underlying devices.
2520 static int dm_table_merge_is_optional(struct dm_table *table)
2523 struct dm_target *ti;
2525 while (i < dm_table_get_num_targets(table)) {
2526 ti = dm_table_get_target(table, i++);
2528 if (ti->type->iterate_devices &&
2529 ti->type->iterate_devices(ti, dm_device_merge_is_compulsory, NULL))
2537 * Returns old map, which caller must destroy.
2539 static struct dm_table *__bind(struct mapped_device *md, struct dm_table *t,
2540 struct queue_limits *limits)
2542 struct dm_table *old_map;
2543 struct request_queue *q = md->queue;
2545 int merge_is_optional;
2547 size = dm_table_get_size(t);
2550 * Wipe any geometry if the size of the table changed.
2552 if (size != dm_get_size(md))
2553 memset(&md->geometry, 0, sizeof(md->geometry));
2555 __set_size(md, size);
2557 dm_table_event_callback(t, event_callback, md);
2560 * The queue hasn't been stopped yet, if the old table type wasn't
2561 * for request-based during suspension. So stop it to prevent
2562 * I/O mapping before resume.
2563 * This must be done before setting the queue restrictions,
2564 * because request-based dm may be run just after the setting.
2566 if (dm_table_request_based(t))
2569 __bind_mempools(md, t);
2571 merge_is_optional = dm_table_merge_is_optional(t);
2573 old_map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2574 rcu_assign_pointer(md->map, t);
2575 md->immutable_target_type = dm_table_get_immutable_target_type(t);
2577 dm_table_set_restrictions(t, q, limits);
2578 if (merge_is_optional)
2579 set_bit(DMF_MERGE_IS_OPTIONAL, &md->flags);
2581 clear_bit(DMF_MERGE_IS_OPTIONAL, &md->flags);
2589 * Returns unbound table for the caller to free.
2591 static struct dm_table *__unbind(struct mapped_device *md)
2593 struct dm_table *map = rcu_dereference_protected(md->map, 1);
2598 dm_table_event_callback(map, NULL, NULL);
2599 RCU_INIT_POINTER(md->map, NULL);
2606 * Constructor for a new device.
2608 int dm_create(int minor, struct mapped_device **result)
2610 struct mapped_device *md;
2612 md = alloc_dev(minor);
2623 * Functions to manage md->type.
2624 * All are required to hold md->type_lock.
2626 void dm_lock_md_type(struct mapped_device *md)
2628 mutex_lock(&md->type_lock);
2631 void dm_unlock_md_type(struct mapped_device *md)
2633 mutex_unlock(&md->type_lock);
2636 void dm_set_md_type(struct mapped_device *md, unsigned type)
2638 BUG_ON(!mutex_is_locked(&md->type_lock));
2642 unsigned dm_get_md_type(struct mapped_device *md)
2644 BUG_ON(!mutex_is_locked(&md->type_lock));
2648 struct target_type *dm_get_immutable_target_type(struct mapped_device *md)
2650 return md->immutable_target_type;
2654 * The queue_limits are only valid as long as you have a reference
2657 struct queue_limits *dm_get_queue_limits(struct mapped_device *md)
2659 BUG_ON(!atomic_read(&md->holders));
2660 return &md->queue->limits;
2662 EXPORT_SYMBOL_GPL(dm_get_queue_limits);
2664 static void init_rq_based_worker_thread(struct mapped_device *md)
2666 /* Initialize the request-based DM worker thread */
2667 init_kthread_worker(&md->kworker);
2668 md->kworker_task = kthread_run(kthread_worker_fn, &md->kworker,
2669 "kdmwork-%s", dm_device_name(md));
2673 * Fully initialize a request-based queue (->elevator, ->request_fn, etc).
2675 static int dm_init_request_based_queue(struct mapped_device *md)
2677 struct request_queue *q = NULL;
2679 /* Fully initialize the queue */
2680 q = blk_init_allocated_queue(md->queue, dm_request_fn, NULL);
2684 /* disable dm_request_fn's merge heuristic by default */
2685 md->seq_rq_merge_deadline_usecs = 0;
2688 dm_init_old_md_queue(md);
2689 blk_queue_softirq_done(md->queue, dm_softirq_done);
2690 blk_queue_prep_rq(md->queue, dm_prep_fn);
2692 init_rq_based_worker_thread(md);
2694 elv_register_queue(md->queue);
2699 static int dm_mq_init_request(void *data, struct request *rq,
2700 unsigned int hctx_idx, unsigned int request_idx,
2701 unsigned int numa_node)
2703 struct mapped_device *md = data;
2704 struct dm_rq_target_io *tio = blk_mq_rq_to_pdu(rq);
2707 * Must initialize md member of tio, otherwise it won't
2708 * be available in dm_mq_queue_rq.
2715 static int dm_mq_queue_rq(struct blk_mq_hw_ctx *hctx,
2716 const struct blk_mq_queue_data *bd)
2718 struct request *rq = bd->rq;
2719 struct dm_rq_target_io *tio = blk_mq_rq_to_pdu(rq);
2720 struct mapped_device *md = tio->md;
2722 struct dm_table *map = dm_get_live_table(md, &srcu_idx);
2723 struct dm_target *ti;
2726 /* always use block 0 to find the target for flushes for now */
2728 if (!(rq->cmd_flags & REQ_FLUSH))
2729 pos = blk_rq_pos(rq);
2731 ti = dm_table_find_target(map, pos);
2732 if (!dm_target_is_valid(ti)) {
2733 dm_put_live_table(md, srcu_idx);
2734 DMERR_LIMIT("request attempted access beyond the end of device");
2736 * Must perform setup, that rq_completed() requires,
2737 * before returning BLK_MQ_RQ_QUEUE_ERROR
2739 dm_start_request(md, rq);
2740 return BLK_MQ_RQ_QUEUE_ERROR;
2742 dm_put_live_table(md, srcu_idx);
2744 if (ti->type->busy && ti->type->busy(ti))
2745 return BLK_MQ_RQ_QUEUE_BUSY;
2747 dm_start_request(md, rq);
2749 /* Init tio using md established in .init_request */
2750 init_tio(tio, rq, md);
2753 * Establish tio->ti before queuing work (map_tio_request)
2754 * or making direct call to map_request().
2758 /* Clone the request if underlying devices aren't blk-mq */
2759 if (dm_table_get_type(map) == DM_TYPE_REQUEST_BASED) {
2760 /* clone request is allocated at the end of the pdu */
2761 tio->clone = (void *)blk_mq_rq_to_pdu(rq) + sizeof(struct dm_rq_target_io);
2762 (void) clone_rq(rq, md, tio, GFP_ATOMIC);
2763 queue_kthread_work(&md->kworker, &tio->work);
2765 /* Direct call is fine since .queue_rq allows allocations */
2766 if (map_request(tio, rq, md) == DM_MAPIO_REQUEUE) {
2767 /* Undo dm_start_request() before requeuing */
2768 rq_completed(md, rq_data_dir(rq), false);
2769 return BLK_MQ_RQ_QUEUE_BUSY;
2773 return BLK_MQ_RQ_QUEUE_OK;
2776 static struct blk_mq_ops dm_mq_ops = {
2777 .queue_rq = dm_mq_queue_rq,
2778 .map_queue = blk_mq_map_queue,
2779 .complete = dm_softirq_done,
2780 .init_request = dm_mq_init_request,
2783 static int dm_init_request_based_blk_mq_queue(struct mapped_device *md)
2785 unsigned md_type = dm_get_md_type(md);
2786 struct request_queue *q;
2789 memset(&md->tag_set, 0, sizeof(md->tag_set));
2790 md->tag_set.ops = &dm_mq_ops;
2791 md->tag_set.queue_depth = BLKDEV_MAX_RQ;
2792 md->tag_set.numa_node = NUMA_NO_NODE;
2793 md->tag_set.flags = BLK_MQ_F_SHOULD_MERGE | BLK_MQ_F_SG_MERGE;
2794 md->tag_set.nr_hw_queues = 1;
2795 if (md_type == DM_TYPE_REQUEST_BASED) {
2796 /* make the memory for non-blk-mq clone part of the pdu */
2797 md->tag_set.cmd_size = sizeof(struct dm_rq_target_io) + sizeof(struct request);
2799 md->tag_set.cmd_size = sizeof(struct dm_rq_target_io);
2800 md->tag_set.driver_data = md;
2802 err = blk_mq_alloc_tag_set(&md->tag_set);
2806 q = blk_mq_init_allocated_queue(&md->tag_set, md->queue);
2812 dm_init_md_queue(md);
2814 /* backfill 'mq' sysfs registration normally done in blk_register_queue */
2815 blk_mq_register_disk(md->disk);
2817 if (md_type == DM_TYPE_REQUEST_BASED)
2818 init_rq_based_worker_thread(md);
2823 blk_mq_free_tag_set(&md->tag_set);
2827 static unsigned filter_md_type(unsigned type, struct mapped_device *md)
2829 if (type == DM_TYPE_BIO_BASED)
2832 return !md->use_blk_mq ? DM_TYPE_REQUEST_BASED : DM_TYPE_MQ_REQUEST_BASED;
2836 * Setup the DM device's queue based on md's type
2838 int dm_setup_md_queue(struct mapped_device *md)
2841 unsigned md_type = filter_md_type(dm_get_md_type(md), md);
2844 case DM_TYPE_REQUEST_BASED:
2845 r = dm_init_request_based_queue(md);
2847 DMWARN("Cannot initialize queue for request-based mapped device");
2851 case DM_TYPE_MQ_REQUEST_BASED:
2852 r = dm_init_request_based_blk_mq_queue(md);
2854 DMWARN("Cannot initialize queue for request-based blk-mq mapped device");
2858 case DM_TYPE_BIO_BASED:
2859 dm_init_old_md_queue(md);
2860 blk_queue_make_request(md->queue, dm_make_request);
2861 blk_queue_merge_bvec(md->queue, dm_merge_bvec);
2868 struct mapped_device *dm_get_md(dev_t dev)
2870 struct mapped_device *md;
2871 unsigned minor = MINOR(dev);
2873 if (MAJOR(dev) != _major || minor >= (1 << MINORBITS))
2876 spin_lock(&_minor_lock);
2878 md = idr_find(&_minor_idr, minor);
2880 if ((md == MINOR_ALLOCED ||
2881 (MINOR(disk_devt(dm_disk(md))) != minor) ||
2882 dm_deleting_md(md) ||
2883 test_bit(DMF_FREEING, &md->flags))) {
2891 spin_unlock(&_minor_lock);
2895 EXPORT_SYMBOL_GPL(dm_get_md);
2897 void *dm_get_mdptr(struct mapped_device *md)
2899 return md->interface_ptr;
2902 void dm_set_mdptr(struct mapped_device *md, void *ptr)
2904 md->interface_ptr = ptr;
2907 void dm_get(struct mapped_device *md)
2909 atomic_inc(&md->holders);
2910 BUG_ON(test_bit(DMF_FREEING, &md->flags));
2913 int dm_hold(struct mapped_device *md)
2915 spin_lock(&_minor_lock);
2916 if (test_bit(DMF_FREEING, &md->flags)) {
2917 spin_unlock(&_minor_lock);
2921 spin_unlock(&_minor_lock);
2924 EXPORT_SYMBOL_GPL(dm_hold);
2926 const char *dm_device_name(struct mapped_device *md)
2930 EXPORT_SYMBOL_GPL(dm_device_name);
2932 static void __dm_destroy(struct mapped_device *md, bool wait)
2934 struct dm_table *map;
2939 map = dm_get_live_table(md, &srcu_idx);
2941 spin_lock(&_minor_lock);
2942 idr_replace(&_minor_idr, MINOR_ALLOCED, MINOR(disk_devt(dm_disk(md))));
2943 set_bit(DMF_FREEING, &md->flags);
2944 spin_unlock(&_minor_lock);
2946 if (dm_request_based(md) && md->kworker_task)
2947 flush_kthread_worker(&md->kworker);
2950 * Take suspend_lock so that presuspend and postsuspend methods
2951 * do not race with internal suspend.
2953 mutex_lock(&md->suspend_lock);
2954 if (!dm_suspended_md(md)) {
2955 dm_table_presuspend_targets(map);
2956 dm_table_postsuspend_targets(map);
2958 mutex_unlock(&md->suspend_lock);
2960 /* dm_put_live_table must be before msleep, otherwise deadlock is possible */
2961 dm_put_live_table(md, srcu_idx);
2964 * Rare, but there may be I/O requests still going to complete,
2965 * for example. Wait for all references to disappear.
2966 * No one should increment the reference count of the mapped_device,
2967 * after the mapped_device state becomes DMF_FREEING.
2970 while (atomic_read(&md->holders))
2972 else if (atomic_read(&md->holders))
2973 DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)",
2974 dm_device_name(md), atomic_read(&md->holders));
2977 dm_table_destroy(__unbind(md));
2981 void dm_destroy(struct mapped_device *md)
2983 __dm_destroy(md, true);
2986 void dm_destroy_immediate(struct mapped_device *md)
2988 __dm_destroy(md, false);
2991 void dm_put(struct mapped_device *md)
2993 atomic_dec(&md->holders);
2995 EXPORT_SYMBOL_GPL(dm_put);
2997 static int dm_wait_for_completion(struct mapped_device *md, int interruptible)
3000 DECLARE_WAITQUEUE(wait, current);
3002 add_wait_queue(&md->wait, &wait);
3005 set_current_state(interruptible);
3007 if (!md_in_flight(md))
3010 if (interruptible == TASK_INTERRUPTIBLE &&
3011 signal_pending(current)) {
3018 set_current_state(TASK_RUNNING);
3020 remove_wait_queue(&md->wait, &wait);
3026 * Process the deferred bios
3028 static void dm_wq_work(struct work_struct *work)
3030 struct mapped_device *md = container_of(work, struct mapped_device,
3034 struct dm_table *map;
3036 map = dm_get_live_table(md, &srcu_idx);
3038 while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
3039 spin_lock_irq(&md->deferred_lock);
3040 c = bio_list_pop(&md->deferred);
3041 spin_unlock_irq(&md->deferred_lock);
3046 if (dm_request_based(md))
3047 generic_make_request(c);
3049 __split_and_process_bio(md, map, c);
3052 dm_put_live_table(md, srcu_idx);
3055 static void dm_queue_flush(struct mapped_device *md)
3057 clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
3058 smp_mb__after_atomic();
3059 queue_work(md->wq, &md->work);
3063 * Swap in a new table, returning the old one for the caller to destroy.
3065 struct dm_table *dm_swap_table(struct mapped_device *md, struct dm_table *table)
3067 struct dm_table *live_map = NULL, *map = ERR_PTR(-EINVAL);
3068 struct queue_limits limits;
3071 mutex_lock(&md->suspend_lock);
3073 /* device must be suspended */
3074 if (!dm_suspended_md(md))
3078 * If the new table has no data devices, retain the existing limits.
3079 * This helps multipath with queue_if_no_path if all paths disappear,
3080 * then new I/O is queued based on these limits, and then some paths
3083 if (dm_table_has_no_data_devices(table)) {
3084 live_map = dm_get_live_table_fast(md);
3086 limits = md->queue->limits;
3087 dm_put_live_table_fast(md);
3091 r = dm_calculate_queue_limits(table, &limits);
3098 map = __bind(md, table, &limits);
3101 mutex_unlock(&md->suspend_lock);
3106 * Functions to lock and unlock any filesystem running on the
3109 static int lock_fs(struct mapped_device *md)
3113 WARN_ON(md->frozen_sb);
3115 md->frozen_sb = freeze_bdev(md->bdev);
3116 if (IS_ERR(md->frozen_sb)) {
3117 r = PTR_ERR(md->frozen_sb);
3118 md->frozen_sb = NULL;
3122 set_bit(DMF_FROZEN, &md->flags);
3127 static void unlock_fs(struct mapped_device *md)
3129 if (!test_bit(DMF_FROZEN, &md->flags))
3132 thaw_bdev(md->bdev, md->frozen_sb);
3133 md->frozen_sb = NULL;
3134 clear_bit(DMF_FROZEN, &md->flags);
3138 * If __dm_suspend returns 0, the device is completely quiescent
3139 * now. There is no request-processing activity. All new requests
3140 * are being added to md->deferred list.
3142 * Caller must hold md->suspend_lock
3144 static int __dm_suspend(struct mapped_device *md, struct dm_table *map,
3145 unsigned suspend_flags, int interruptible)
3147 bool do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG;
3148 bool noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG;
3152 * DMF_NOFLUSH_SUSPENDING must be set before presuspend.
3153 * This flag is cleared before dm_suspend returns.
3156 set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
3159 * This gets reverted if there's an error later and the targets
3160 * provide the .presuspend_undo hook.
3162 dm_table_presuspend_targets(map);
3165 * Flush I/O to the device.
3166 * Any I/O submitted after lock_fs() may not be flushed.
3167 * noflush takes precedence over do_lockfs.
3168 * (lock_fs() flushes I/Os and waits for them to complete.)
3170 if (!noflush && do_lockfs) {
3173 dm_table_presuspend_undo_targets(map);
3179 * Here we must make sure that no processes are submitting requests
3180 * to target drivers i.e. no one may be executing
3181 * __split_and_process_bio. This is called from dm_request and
3184 * To get all processes out of __split_and_process_bio in dm_request,
3185 * we take the write lock. To prevent any process from reentering
3186 * __split_and_process_bio from dm_request and quiesce the thread
3187 * (dm_wq_work), we set BMF_BLOCK_IO_FOR_SUSPEND and call
3188 * flush_workqueue(md->wq).
3190 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
3192 synchronize_srcu(&md->io_barrier);
3195 * Stop md->queue before flushing md->wq in case request-based
3196 * dm defers requests to md->wq from md->queue.
3198 if (dm_request_based(md)) {
3199 stop_queue(md->queue);
3200 if (md->kworker_task)
3201 flush_kthread_worker(&md->kworker);
3204 flush_workqueue(md->wq);
3207 * At this point no more requests are entering target request routines.
3208 * We call dm_wait_for_completion to wait for all existing requests
3211 r = dm_wait_for_completion(md, interruptible);
3214 clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
3216 synchronize_srcu(&md->io_barrier);
3218 /* were we interrupted ? */
3222 if (dm_request_based(md))
3223 start_queue(md->queue);
3226 dm_table_presuspend_undo_targets(map);
3227 /* pushback list is already flushed, so skip flush */
3234 * We need to be able to change a mapping table under a mounted
3235 * filesystem. For example we might want to move some data in
3236 * the background. Before the table can be swapped with
3237 * dm_bind_table, dm_suspend must be called to flush any in
3238 * flight bios and ensure that any further io gets deferred.
3241 * Suspend mechanism in request-based dm.
3243 * 1. Flush all I/Os by lock_fs() if needed.
3244 * 2. Stop dispatching any I/O by stopping the request_queue.
3245 * 3. Wait for all in-flight I/Os to be completed or requeued.
3247 * To abort suspend, start the request_queue.
3249 int dm_suspend(struct mapped_device *md, unsigned suspend_flags)
3251 struct dm_table *map = NULL;
3255 mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING);
3257 if (dm_suspended_md(md)) {
3262 if (dm_suspended_internally_md(md)) {
3263 /* already internally suspended, wait for internal resume */
3264 mutex_unlock(&md->suspend_lock);
3265 r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE);
3271 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
3273 r = __dm_suspend(md, map, suspend_flags, TASK_INTERRUPTIBLE);
3277 set_bit(DMF_SUSPENDED, &md->flags);
3279 dm_table_postsuspend_targets(map);
3282 mutex_unlock(&md->suspend_lock);
3286 static int __dm_resume(struct mapped_device *md, struct dm_table *map)
3289 int r = dm_table_resume_targets(map);
3297 * Flushing deferred I/Os must be done after targets are resumed
3298 * so that mapping of targets can work correctly.
3299 * Request-based dm is queueing the deferred I/Os in its request_queue.
3301 if (dm_request_based(md))
3302 start_queue(md->queue);
3309 int dm_resume(struct mapped_device *md)
3312 struct dm_table *map = NULL;
3315 mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING);
3317 if (!dm_suspended_md(md))
3320 if (dm_suspended_internally_md(md)) {
3321 /* already internally suspended, wait for internal resume */
3322 mutex_unlock(&md->suspend_lock);
3323 r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE);
3329 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
3330 if (!map || !dm_table_get_size(map))
3333 r = __dm_resume(md, map);
3337 clear_bit(DMF_SUSPENDED, &md->flags);
3341 mutex_unlock(&md->suspend_lock);
3347 * Internal suspend/resume works like userspace-driven suspend. It waits
3348 * until all bios finish and prevents issuing new bios to the target drivers.
3349 * It may be used only from the kernel.
3352 static void __dm_internal_suspend(struct mapped_device *md, unsigned suspend_flags)
3354 struct dm_table *map = NULL;
3356 if (md->internal_suspend_count++)
3357 return; /* nested internal suspend */
3359 if (dm_suspended_md(md)) {
3360 set_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
3361 return; /* nest suspend */
3364 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
3367 * Using TASK_UNINTERRUPTIBLE because only NOFLUSH internal suspend is
3368 * supported. Properly supporting a TASK_INTERRUPTIBLE internal suspend
3369 * would require changing .presuspend to return an error -- avoid this
3370 * until there is a need for more elaborate variants of internal suspend.
3372 (void) __dm_suspend(md, map, suspend_flags, TASK_UNINTERRUPTIBLE);
3374 set_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
3376 dm_table_postsuspend_targets(map);
3379 static void __dm_internal_resume(struct mapped_device *md)
3381 BUG_ON(!md->internal_suspend_count);
3383 if (--md->internal_suspend_count)
3384 return; /* resume from nested internal suspend */
3386 if (dm_suspended_md(md))
3387 goto done; /* resume from nested suspend */
3390 * NOTE: existing callers don't need to call dm_table_resume_targets
3391 * (which may fail -- so best to avoid it for now by passing NULL map)
3393 (void) __dm_resume(md, NULL);
3396 clear_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
3397 smp_mb__after_atomic();
3398 wake_up_bit(&md->flags, DMF_SUSPENDED_INTERNALLY);
3401 void dm_internal_suspend_noflush(struct mapped_device *md)
3403 mutex_lock(&md->suspend_lock);
3404 __dm_internal_suspend(md, DM_SUSPEND_NOFLUSH_FLAG);
3405 mutex_unlock(&md->suspend_lock);
3407 EXPORT_SYMBOL_GPL(dm_internal_suspend_noflush);
3409 void dm_internal_resume(struct mapped_device *md)
3411 mutex_lock(&md->suspend_lock);
3412 __dm_internal_resume(md);
3413 mutex_unlock(&md->suspend_lock);
3415 EXPORT_SYMBOL_GPL(dm_internal_resume);
3418 * Fast variants of internal suspend/resume hold md->suspend_lock,
3419 * which prevents interaction with userspace-driven suspend.
3422 void dm_internal_suspend_fast(struct mapped_device *md)
3424 mutex_lock(&md->suspend_lock);
3425 if (dm_suspended_md(md) || dm_suspended_internally_md(md))
3428 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
3429 synchronize_srcu(&md->io_barrier);
3430 flush_workqueue(md->wq);
3431 dm_wait_for_completion(md, TASK_UNINTERRUPTIBLE);
3433 EXPORT_SYMBOL_GPL(dm_internal_suspend_fast);
3435 void dm_internal_resume_fast(struct mapped_device *md)
3437 if (dm_suspended_md(md) || dm_suspended_internally_md(md))
3443 mutex_unlock(&md->suspend_lock);
3445 EXPORT_SYMBOL_GPL(dm_internal_resume_fast);
3447 /*-----------------------------------------------------------------
3448 * Event notification.
3449 *---------------------------------------------------------------*/
3450 int dm_kobject_uevent(struct mapped_device *md, enum kobject_action action,
3453 char udev_cookie[DM_COOKIE_LENGTH];
3454 char *envp[] = { udev_cookie, NULL };
3457 return kobject_uevent(&disk_to_dev(md->disk)->kobj, action);
3459 snprintf(udev_cookie, DM_COOKIE_LENGTH, "%s=%u",
3460 DM_COOKIE_ENV_VAR_NAME, cookie);
3461 return kobject_uevent_env(&disk_to_dev(md->disk)->kobj,
3466 uint32_t dm_next_uevent_seq(struct mapped_device *md)
3468 return atomic_add_return(1, &md->uevent_seq);
3471 uint32_t dm_get_event_nr(struct mapped_device *md)
3473 return atomic_read(&md->event_nr);
3476 int dm_wait_event(struct mapped_device *md, int event_nr)
3478 return wait_event_interruptible(md->eventq,
3479 (event_nr != atomic_read(&md->event_nr)));
3482 void dm_uevent_add(struct mapped_device *md, struct list_head *elist)
3484 unsigned long flags;
3486 spin_lock_irqsave(&md->uevent_lock, flags);
3487 list_add(elist, &md->uevent_list);
3488 spin_unlock_irqrestore(&md->uevent_lock, flags);
3492 * The gendisk is only valid as long as you have a reference
3495 struct gendisk *dm_disk(struct mapped_device *md)
3499 EXPORT_SYMBOL_GPL(dm_disk);
3501 struct kobject *dm_kobject(struct mapped_device *md)
3503 return &md->kobj_holder.kobj;
3506 struct mapped_device *dm_get_from_kobject(struct kobject *kobj)
3508 struct mapped_device *md;
3510 md = container_of(kobj, struct mapped_device, kobj_holder.kobj);
3512 if (test_bit(DMF_FREEING, &md->flags) ||
3520 int dm_suspended_md(struct mapped_device *md)
3522 return test_bit(DMF_SUSPENDED, &md->flags);
3525 int dm_suspended_internally_md(struct mapped_device *md)
3527 return test_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
3530 int dm_test_deferred_remove_flag(struct mapped_device *md)
3532 return test_bit(DMF_DEFERRED_REMOVE, &md->flags);
3535 int dm_suspended(struct dm_target *ti)
3537 return dm_suspended_md(dm_table_get_md(ti->table));
3539 EXPORT_SYMBOL_GPL(dm_suspended);
3541 int dm_noflush_suspending(struct dm_target *ti)
3543 return __noflush_suspending(dm_table_get_md(ti->table));
3545 EXPORT_SYMBOL_GPL(dm_noflush_suspending);
3547 struct dm_md_mempools *dm_alloc_md_mempools(struct mapped_device *md, unsigned type,
3548 unsigned integrity, unsigned per_bio_data_size)
3550 struct dm_md_mempools *pools = kzalloc(sizeof(*pools), GFP_KERNEL);
3551 struct kmem_cache *cachep = NULL;
3552 unsigned int pool_size = 0;
3553 unsigned int front_pad;
3558 type = filter_md_type(type, md);
3561 case DM_TYPE_BIO_BASED:
3563 pool_size = dm_get_reserved_bio_based_ios();
3564 front_pad = roundup(per_bio_data_size, __alignof__(struct dm_target_io)) + offsetof(struct dm_target_io, clone);
3566 case DM_TYPE_REQUEST_BASED:
3567 cachep = _rq_tio_cache;
3568 pool_size = dm_get_reserved_rq_based_ios();
3569 pools->rq_pool = mempool_create_slab_pool(pool_size, _rq_cache);
3570 if (!pools->rq_pool)
3572 /* fall through to setup remaining rq-based pools */
3573 case DM_TYPE_MQ_REQUEST_BASED:
3575 pool_size = dm_get_reserved_rq_based_ios();
3576 front_pad = offsetof(struct dm_rq_clone_bio_info, clone);
3577 /* per_bio_data_size is not used. See __bind_mempools(). */
3578 WARN_ON(per_bio_data_size != 0);
3585 pools->io_pool = mempool_create_slab_pool(pool_size, cachep);
3586 if (!pools->io_pool)
3590 pools->bs = bioset_create_nobvec(pool_size, front_pad);
3594 if (integrity && bioset_integrity_create(pools->bs, pool_size))
3600 dm_free_md_mempools(pools);
3605 void dm_free_md_mempools(struct dm_md_mempools *pools)
3611 mempool_destroy(pools->io_pool);
3614 mempool_destroy(pools->rq_pool);
3617 bioset_free(pools->bs);
3622 static const struct block_device_operations dm_blk_dops = {
3623 .open = dm_blk_open,
3624 .release = dm_blk_close,
3625 .ioctl = dm_blk_ioctl,
3626 .getgeo = dm_blk_getgeo,
3627 .owner = THIS_MODULE
3633 module_init(dm_init);
3634 module_exit(dm_exit);
3636 module_param(major, uint, 0);
3637 MODULE_PARM_DESC(major, "The major number of the device mapper");
3639 module_param(reserved_bio_based_ios, uint, S_IRUGO | S_IWUSR);
3640 MODULE_PARM_DESC(reserved_bio_based_ios, "Reserved IOs in bio-based mempools");
3642 module_param(reserved_rq_based_ios, uint, S_IRUGO | S_IWUSR);
3643 MODULE_PARM_DESC(reserved_rq_based_ios, "Reserved IOs in request-based mempools");
3645 module_param(use_blk_mq, bool, S_IRUGO | S_IWUSR);
3646 MODULE_PARM_DESC(use_blk_mq, "Use block multiqueue for request-based DM devices");
3648 MODULE_DESCRIPTION(DM_NAME " driver");
3649 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
3650 MODULE_LICENSE("GPL");