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);
993 static struct dm_rq_target_io *tio_from_request(struct request *rq)
995 return (rq->q->mq_ops ? blk_mq_rq_to_pdu(rq) : rq->special);
999 * Don't touch any member of the md after calling this function because
1000 * the md may be freed in dm_put() at the end of this function.
1001 * Or do dm_get() before calling this function and dm_put() later.
1003 static void rq_completed(struct mapped_device *md, int rw, bool run_queue)
1005 int nr_requests_pending;
1007 atomic_dec(&md->pending[rw]);
1009 /* nudge anyone waiting on suspend queue */
1010 nr_requests_pending = md_in_flight(md);
1011 if (!nr_requests_pending)
1015 * Run this off this callpath, as drivers could invoke end_io while
1016 * inside their request_fn (and holding the queue lock). Calling
1017 * back into ->request_fn() could deadlock attempting to grab the
1021 if (md->queue->mq_ops)
1022 blk_mq_run_hw_queues(md->queue, true);
1023 else if (!nr_requests_pending ||
1024 (nr_requests_pending >= md->queue->nr_congestion_on))
1025 blk_run_queue_async(md->queue);
1029 * dm_put() must be at the end of this function. See the comment above
1034 static void free_rq_clone(struct request *clone)
1036 struct dm_rq_target_io *tio = clone->end_io_data;
1037 struct mapped_device *md = tio->md;
1039 if (md->type == DM_TYPE_MQ_REQUEST_BASED)
1040 /* stacked on blk-mq queue(s) */
1041 tio->ti->type->release_clone_rq(clone);
1042 else if (!md->queue->mq_ops)
1043 /* request_fn queue stacked on request_fn queue(s) */
1044 free_clone_request(md, clone);
1046 * NOTE: for the blk-mq queue stacked on request_fn queue(s) case:
1047 * no need to call free_clone_request() because we leverage blk-mq by
1048 * allocating the clone at the end of the blk-mq pdu (see: clone_rq)
1051 if (!md->queue->mq_ops)
1056 * Complete the clone and the original request.
1057 * Must be called without clone's queue lock held,
1058 * see end_clone_request() for more details.
1060 static void dm_end_request(struct request *clone, int error)
1062 int rw = rq_data_dir(clone);
1063 struct dm_rq_target_io *tio = clone->end_io_data;
1064 struct mapped_device *md = tio->md;
1065 struct request *rq = tio->orig;
1067 if (rq->cmd_type == REQ_TYPE_BLOCK_PC) {
1068 rq->errors = clone->errors;
1069 rq->resid_len = clone->resid_len;
1073 * We are using the sense buffer of the original
1075 * So setting the length of the sense data is enough.
1077 rq->sense_len = clone->sense_len;
1080 free_rq_clone(clone);
1082 blk_end_request_all(rq, error);
1084 blk_mq_end_request(rq, error);
1085 rq_completed(md, rw, true);
1088 static void dm_unprep_request(struct request *rq)
1090 struct dm_rq_target_io *tio = tio_from_request(rq);
1091 struct request *clone = tio->clone;
1093 if (!rq->q->mq_ops) {
1095 rq->cmd_flags &= ~REQ_DONTPREP;
1099 free_rq_clone(clone);
1103 * Requeue the original request of a clone.
1105 static void old_requeue_request(struct request *rq)
1107 struct request_queue *q = rq->q;
1108 unsigned long flags;
1110 spin_lock_irqsave(q->queue_lock, flags);
1111 blk_requeue_request(q, rq);
1112 blk_run_queue_async(q);
1113 spin_unlock_irqrestore(q->queue_lock, flags);
1116 static void dm_requeue_unmapped_original_request(struct mapped_device *md,
1119 int rw = rq_data_dir(rq);
1121 dm_unprep_request(rq);
1124 old_requeue_request(rq);
1126 blk_mq_requeue_request(rq);
1127 blk_mq_kick_requeue_list(rq->q);
1130 rq_completed(md, rw, false);
1133 static void dm_requeue_unmapped_request(struct request *clone)
1135 struct dm_rq_target_io *tio = clone->end_io_data;
1137 dm_requeue_unmapped_original_request(tio->md, tio->orig);
1140 static void old_stop_queue(struct request_queue *q)
1142 unsigned long flags;
1144 if (blk_queue_stopped(q))
1147 spin_lock_irqsave(q->queue_lock, flags);
1149 spin_unlock_irqrestore(q->queue_lock, flags);
1152 static void stop_queue(struct request_queue *q)
1157 blk_mq_stop_hw_queues(q);
1160 static void old_start_queue(struct request_queue *q)
1162 unsigned long flags;
1164 spin_lock_irqsave(q->queue_lock, flags);
1165 if (blk_queue_stopped(q))
1167 spin_unlock_irqrestore(q->queue_lock, flags);
1170 static void start_queue(struct request_queue *q)
1175 blk_mq_start_stopped_hw_queues(q, true);
1178 static void dm_done(struct request *clone, int error, bool mapped)
1181 struct dm_rq_target_io *tio = clone->end_io_data;
1182 dm_request_endio_fn rq_end_io = NULL;
1185 rq_end_io = tio->ti->type->rq_end_io;
1187 if (mapped && rq_end_io)
1188 r = rq_end_io(tio->ti, clone, error, &tio->info);
1191 if (unlikely(r == -EREMOTEIO && (clone->cmd_flags & REQ_WRITE_SAME) &&
1192 !clone->q->limits.max_write_same_sectors))
1193 disable_write_same(tio->md);
1196 /* The target wants to complete the I/O */
1197 dm_end_request(clone, r);
1198 else if (r == DM_ENDIO_INCOMPLETE)
1199 /* The target will handle the I/O */
1201 else if (r == DM_ENDIO_REQUEUE)
1202 /* The target wants to requeue the I/O */
1203 dm_requeue_unmapped_request(clone);
1205 DMWARN("unimplemented target endio return value: %d", r);
1211 * Request completion handler for request-based dm
1213 static void dm_softirq_done(struct request *rq)
1216 struct dm_rq_target_io *tio = tio_from_request(rq);
1217 struct request *clone = tio->clone;
1221 rw = rq_data_dir(rq);
1222 if (!rq->q->mq_ops) {
1223 blk_end_request_all(rq, tio->error);
1224 rq_completed(tio->md, rw, false);
1227 blk_mq_end_request(rq, tio->error);
1228 rq_completed(tio->md, rw, false);
1233 if (rq->cmd_flags & REQ_FAILED)
1236 dm_done(clone, tio->error, mapped);
1240 * Complete the clone and the original request with the error status
1241 * through softirq context.
1243 static void dm_complete_request(struct request *rq, int error)
1245 struct dm_rq_target_io *tio = tio_from_request(rq);
1248 blk_complete_request(rq);
1252 * Complete the not-mapped clone and the original request with the error status
1253 * through softirq context.
1254 * Target's rq_end_io() function isn't called.
1255 * This may be used when the target's map_rq() or clone_and_map_rq() functions fail.
1257 static void dm_kill_unmapped_request(struct request *rq, int error)
1259 rq->cmd_flags |= REQ_FAILED;
1260 dm_complete_request(rq, error);
1264 * Called with the clone's queue lock held (for non-blk-mq)
1266 static void end_clone_request(struct request *clone, int error)
1268 struct dm_rq_target_io *tio = clone->end_io_data;
1270 if (!clone->q->mq_ops) {
1272 * For just cleaning up the information of the queue in which
1273 * the clone was dispatched.
1274 * The clone is *NOT* freed actually here because it is alloced
1275 * from dm own mempool (REQ_ALLOCED isn't set).
1277 __blk_put_request(clone->q, clone);
1281 * Actual request completion is done in a softirq context which doesn't
1282 * hold the clone's queue lock. Otherwise, deadlock could occur because:
1283 * - another request may be submitted by the upper level driver
1284 * of the stacking during the completion
1285 * - the submission which requires queue lock may be done
1286 * against this clone's queue
1288 dm_complete_request(tio->orig, error);
1292 * Return maximum size of I/O possible at the supplied sector up to the current
1295 static sector_t max_io_len_target_boundary(sector_t sector, struct dm_target *ti)
1297 sector_t target_offset = dm_target_offset(ti, sector);
1299 return ti->len - target_offset;
1302 static sector_t max_io_len(sector_t sector, struct dm_target *ti)
1304 sector_t len = max_io_len_target_boundary(sector, ti);
1305 sector_t offset, max_len;
1308 * Does the target need to split even further?
1310 if (ti->max_io_len) {
1311 offset = dm_target_offset(ti, sector);
1312 if (unlikely(ti->max_io_len & (ti->max_io_len - 1)))
1313 max_len = sector_div(offset, ti->max_io_len);
1315 max_len = offset & (ti->max_io_len - 1);
1316 max_len = ti->max_io_len - max_len;
1325 int dm_set_target_max_io_len(struct dm_target *ti, sector_t len)
1327 if (len > UINT_MAX) {
1328 DMERR("Specified maximum size of target IO (%llu) exceeds limit (%u)",
1329 (unsigned long long)len, UINT_MAX);
1330 ti->error = "Maximum size of target IO is too large";
1334 ti->max_io_len = (uint32_t) len;
1338 EXPORT_SYMBOL_GPL(dm_set_target_max_io_len);
1341 * A target may call dm_accept_partial_bio only from the map routine. It is
1342 * allowed for all bio types except REQ_FLUSH.
1344 * dm_accept_partial_bio informs the dm that the target only wants to process
1345 * additional n_sectors sectors of the bio and the rest of the data should be
1346 * sent in a next bio.
1348 * A diagram that explains the arithmetics:
1349 * +--------------------+---------------+-------+
1351 * +--------------------+---------------+-------+
1353 * <-------------- *tio->len_ptr --------------->
1354 * <------- bi_size ------->
1357 * Region 1 was already iterated over with bio_advance or similar function.
1358 * (it may be empty if the target doesn't use bio_advance)
1359 * Region 2 is the remaining bio size that the target wants to process.
1360 * (it may be empty if region 1 is non-empty, although there is no reason
1362 * The target requires that region 3 is to be sent in the next bio.
1364 * If the target wants to receive multiple copies of the bio (via num_*bios, etc),
1365 * the partially processed part (the sum of regions 1+2) must be the same for all
1366 * copies of the bio.
1368 void dm_accept_partial_bio(struct bio *bio, unsigned n_sectors)
1370 struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone);
1371 unsigned bi_size = bio->bi_iter.bi_size >> SECTOR_SHIFT;
1372 BUG_ON(bio->bi_rw & REQ_FLUSH);
1373 BUG_ON(bi_size > *tio->len_ptr);
1374 BUG_ON(n_sectors > bi_size);
1375 *tio->len_ptr -= bi_size - n_sectors;
1376 bio->bi_iter.bi_size = n_sectors << SECTOR_SHIFT;
1378 EXPORT_SYMBOL_GPL(dm_accept_partial_bio);
1380 static void __map_bio(struct dm_target_io *tio)
1384 struct mapped_device *md;
1385 struct bio *clone = &tio->clone;
1386 struct dm_target *ti = tio->ti;
1388 clone->bi_end_io = clone_endio;
1391 * Map the clone. If r == 0 we don't need to do
1392 * anything, the target has assumed ownership of
1395 atomic_inc(&tio->io->io_count);
1396 sector = clone->bi_iter.bi_sector;
1397 r = ti->type->map(ti, clone);
1398 if (r == DM_MAPIO_REMAPPED) {
1399 /* the bio has been remapped so dispatch it */
1401 trace_block_bio_remap(bdev_get_queue(clone->bi_bdev), clone,
1402 tio->io->bio->bi_bdev->bd_dev, sector);
1404 generic_make_request(clone);
1405 } else if (r < 0 || r == DM_MAPIO_REQUEUE) {
1406 /* error the io and bail out, or requeue it if needed */
1408 dec_pending(tio->io, r);
1411 DMWARN("unimplemented target map return value: %d", r);
1417 struct mapped_device *md;
1418 struct dm_table *map;
1422 unsigned sector_count;
1425 static void bio_setup_sector(struct bio *bio, sector_t sector, unsigned len)
1427 bio->bi_iter.bi_sector = sector;
1428 bio->bi_iter.bi_size = to_bytes(len);
1432 * Creates a bio that consists of range of complete bvecs.
1434 static void clone_bio(struct dm_target_io *tio, struct bio *bio,
1435 sector_t sector, unsigned len)
1437 struct bio *clone = &tio->clone;
1439 __bio_clone_fast(clone, bio);
1441 if (bio_integrity(bio))
1442 bio_integrity_clone(clone, bio, GFP_NOIO);
1444 bio_advance(clone, to_bytes(sector - clone->bi_iter.bi_sector));
1445 clone->bi_iter.bi_size = to_bytes(len);
1447 if (bio_integrity(bio))
1448 bio_integrity_trim(clone, 0, len);
1451 static struct dm_target_io *alloc_tio(struct clone_info *ci,
1452 struct dm_target *ti,
1453 unsigned target_bio_nr)
1455 struct dm_target_io *tio;
1458 clone = bio_alloc_bioset(GFP_NOIO, 0, ci->md->bs);
1459 tio = container_of(clone, struct dm_target_io, clone);
1463 tio->target_bio_nr = target_bio_nr;
1468 static void __clone_and_map_simple_bio(struct clone_info *ci,
1469 struct dm_target *ti,
1470 unsigned target_bio_nr, unsigned *len)
1472 struct dm_target_io *tio = alloc_tio(ci, ti, target_bio_nr);
1473 struct bio *clone = &tio->clone;
1477 __bio_clone_fast(clone, ci->bio);
1479 bio_setup_sector(clone, ci->sector, *len);
1484 static void __send_duplicate_bios(struct clone_info *ci, struct dm_target *ti,
1485 unsigned num_bios, unsigned *len)
1487 unsigned target_bio_nr;
1489 for (target_bio_nr = 0; target_bio_nr < num_bios; target_bio_nr++)
1490 __clone_and_map_simple_bio(ci, ti, target_bio_nr, len);
1493 static int __send_empty_flush(struct clone_info *ci)
1495 unsigned target_nr = 0;
1496 struct dm_target *ti;
1498 BUG_ON(bio_has_data(ci->bio));
1499 while ((ti = dm_table_get_target(ci->map, target_nr++)))
1500 __send_duplicate_bios(ci, ti, ti->num_flush_bios, NULL);
1505 static void __clone_and_map_data_bio(struct clone_info *ci, struct dm_target *ti,
1506 sector_t sector, unsigned *len)
1508 struct bio *bio = ci->bio;
1509 struct dm_target_io *tio;
1510 unsigned target_bio_nr;
1511 unsigned num_target_bios = 1;
1514 * Does the target want to receive duplicate copies of the bio?
1516 if (bio_data_dir(bio) == WRITE && ti->num_write_bios)
1517 num_target_bios = ti->num_write_bios(ti, bio);
1519 for (target_bio_nr = 0; target_bio_nr < num_target_bios; target_bio_nr++) {
1520 tio = alloc_tio(ci, ti, target_bio_nr);
1522 clone_bio(tio, bio, sector, *len);
1527 typedef unsigned (*get_num_bios_fn)(struct dm_target *ti);
1529 static unsigned get_num_discard_bios(struct dm_target *ti)
1531 return ti->num_discard_bios;
1534 static unsigned get_num_write_same_bios(struct dm_target *ti)
1536 return ti->num_write_same_bios;
1539 typedef bool (*is_split_required_fn)(struct dm_target *ti);
1541 static bool is_split_required_for_discard(struct dm_target *ti)
1543 return ti->split_discard_bios;
1546 static int __send_changing_extent_only(struct clone_info *ci,
1547 get_num_bios_fn get_num_bios,
1548 is_split_required_fn is_split_required)
1550 struct dm_target *ti;
1555 ti = dm_table_find_target(ci->map, ci->sector);
1556 if (!dm_target_is_valid(ti))
1560 * Even though the device advertised support for this type of
1561 * request, that does not mean every target supports it, and
1562 * reconfiguration might also have changed that since the
1563 * check was performed.
1565 num_bios = get_num_bios ? get_num_bios(ti) : 0;
1569 if (is_split_required && !is_split_required(ti))
1570 len = min((sector_t)ci->sector_count, max_io_len_target_boundary(ci->sector, ti));
1572 len = min((sector_t)ci->sector_count, max_io_len(ci->sector, ti));
1574 __send_duplicate_bios(ci, ti, num_bios, &len);
1577 } while (ci->sector_count -= len);
1582 static int __send_discard(struct clone_info *ci)
1584 return __send_changing_extent_only(ci, get_num_discard_bios,
1585 is_split_required_for_discard);
1588 static int __send_write_same(struct clone_info *ci)
1590 return __send_changing_extent_only(ci, get_num_write_same_bios, NULL);
1594 * Select the correct strategy for processing a non-flush bio.
1596 static int __split_and_process_non_flush(struct clone_info *ci)
1598 struct bio *bio = ci->bio;
1599 struct dm_target *ti;
1602 if (unlikely(bio->bi_rw & REQ_DISCARD))
1603 return __send_discard(ci);
1604 else if (unlikely(bio->bi_rw & REQ_WRITE_SAME))
1605 return __send_write_same(ci);
1607 ti = dm_table_find_target(ci->map, ci->sector);
1608 if (!dm_target_is_valid(ti))
1611 len = min_t(sector_t, max_io_len(ci->sector, ti), ci->sector_count);
1613 __clone_and_map_data_bio(ci, ti, ci->sector, &len);
1616 ci->sector_count -= len;
1622 * Entry point to split a bio into clones and submit them to the targets.
1624 static void __split_and_process_bio(struct mapped_device *md,
1625 struct dm_table *map, struct bio *bio)
1627 struct clone_info ci;
1630 if (unlikely(!map)) {
1637 ci.io = alloc_io(md);
1639 atomic_set(&ci.io->io_count, 1);
1642 spin_lock_init(&ci.io->endio_lock);
1643 ci.sector = bio->bi_iter.bi_sector;
1645 start_io_acct(ci.io);
1647 if (bio->bi_rw & REQ_FLUSH) {
1648 ci.bio = &ci.md->flush_bio;
1649 ci.sector_count = 0;
1650 error = __send_empty_flush(&ci);
1651 /* dec_pending submits any data associated with flush */
1654 ci.sector_count = bio_sectors(bio);
1655 while (ci.sector_count && !error)
1656 error = __split_and_process_non_flush(&ci);
1659 /* drop the extra reference count */
1660 dec_pending(ci.io, error);
1662 /*-----------------------------------------------------------------
1664 *---------------------------------------------------------------*/
1666 static int dm_merge_bvec(struct request_queue *q,
1667 struct bvec_merge_data *bvm,
1668 struct bio_vec *biovec)
1670 struct mapped_device *md = q->queuedata;
1671 struct dm_table *map = dm_get_live_table_fast(md);
1672 struct dm_target *ti;
1673 sector_t max_sectors, max_size = 0;
1678 ti = dm_table_find_target(map, bvm->bi_sector);
1679 if (!dm_target_is_valid(ti))
1683 * Find maximum amount of I/O that won't need splitting
1685 max_sectors = min(max_io_len(bvm->bi_sector, ti),
1686 (sector_t) queue_max_sectors(q));
1687 max_size = (max_sectors << SECTOR_SHIFT) - bvm->bi_size;
1690 * FIXME: this stop-gap fix _must_ be cleaned up (by passing a sector_t
1691 * to the targets' merge function since it holds sectors not bytes).
1692 * Just doing this as an interim fix for stable@ because the more
1693 * comprehensive cleanup of switching to sector_t will impact every
1694 * DM target that implements a ->merge hook.
1696 if (max_size > INT_MAX)
1700 * merge_bvec_fn() returns number of bytes
1701 * it can accept at this offset
1702 * max is precomputed maximal io size
1704 if (max_size && ti->type->merge)
1705 max_size = ti->type->merge(ti, bvm, biovec, (int) max_size);
1707 * If the target doesn't support merge method and some of the devices
1708 * provided their merge_bvec method (we know this by looking for the
1709 * max_hw_sectors that dm_set_device_limits may set), then we can't
1710 * allow bios with multiple vector entries. So always set max_size
1711 * to 0, and the code below allows just one page.
1713 else if (queue_max_hw_sectors(q) <= PAGE_SIZE >> 9)
1717 dm_put_live_table_fast(md);
1719 * Always allow an entire first page
1721 if (max_size <= biovec->bv_len && !(bvm->bi_size >> SECTOR_SHIFT))
1722 max_size = biovec->bv_len;
1728 * The request function that just remaps the bio built up by
1731 static void dm_make_request(struct request_queue *q, struct bio *bio)
1733 int rw = bio_data_dir(bio);
1734 struct mapped_device *md = q->queuedata;
1736 struct dm_table *map;
1738 map = dm_get_live_table(md, &srcu_idx);
1740 generic_start_io_acct(rw, bio_sectors(bio), &dm_disk(md)->part0);
1742 /* if we're suspended, we have to queue this io for later */
1743 if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))) {
1744 dm_put_live_table(md, srcu_idx);
1746 if (bio_rw(bio) != READA)
1753 __split_and_process_bio(md, map, bio);
1754 dm_put_live_table(md, srcu_idx);
1758 int dm_request_based(struct mapped_device *md)
1760 return blk_queue_stackable(md->queue);
1763 static void dm_dispatch_clone_request(struct request *clone, struct request *rq)
1767 if (blk_queue_io_stat(clone->q))
1768 clone->cmd_flags |= REQ_IO_STAT;
1770 clone->start_time = jiffies;
1771 r = blk_insert_cloned_request(clone->q, clone);
1773 /* must complete clone in terms of original request */
1774 dm_complete_request(rq, r);
1777 static void setup_clone(struct request *clone, struct request *rq,
1778 struct dm_rq_target_io *tio)
1780 blk_rq_prep_clone(clone, rq);
1781 clone->end_io = end_clone_request;
1782 clone->end_io_data = tio;
1786 static struct request *clone_rq(struct request *rq, struct mapped_device *md,
1787 struct dm_rq_target_io *tio, gfp_t gfp_mask)
1790 * Do not allocate a clone if tio->clone was already set
1791 * (see: dm_mq_queue_rq).
1793 bool alloc_clone = !tio->clone;
1794 struct request *clone;
1797 clone = alloc_clone_request(md, gfp_mask);
1803 blk_rq_init(NULL, clone);
1804 setup_clone(clone, rq, tio);
1809 static void map_tio_request(struct kthread_work *work);
1811 static void init_tio(struct dm_rq_target_io *tio, struct request *rq,
1812 struct mapped_device *md)
1819 memset(&tio->info, 0, sizeof(tio->info));
1820 if (md->kworker_task)
1821 init_kthread_work(&tio->work, map_tio_request);
1824 static struct dm_rq_target_io *prep_tio(struct request *rq,
1825 struct mapped_device *md, gfp_t gfp_mask)
1827 struct dm_rq_target_io *tio;
1829 struct dm_table *table;
1831 tio = alloc_rq_tio(md, gfp_mask);
1835 init_tio(tio, rq, md);
1837 table = dm_get_live_table(md, &srcu_idx);
1838 if (!dm_table_mq_request_based(table)) {
1839 if (!clone_rq(rq, md, tio, gfp_mask)) {
1840 dm_put_live_table(md, srcu_idx);
1845 dm_put_live_table(md, srcu_idx);
1851 * Called with the queue lock held.
1853 static int dm_prep_fn(struct request_queue *q, struct request *rq)
1855 struct mapped_device *md = q->queuedata;
1856 struct dm_rq_target_io *tio;
1858 if (unlikely(rq->special)) {
1859 DMWARN("Already has something in rq->special.");
1860 return BLKPREP_KILL;
1863 tio = prep_tio(rq, md, GFP_ATOMIC);
1865 return BLKPREP_DEFER;
1868 rq->cmd_flags |= REQ_DONTPREP;
1875 * 0 : the request has been processed
1876 * DM_MAPIO_REQUEUE : the original request needs to be requeued
1877 * < 0 : the request was completed due to failure
1879 static int map_request(struct dm_rq_target_io *tio, struct request *rq,
1880 struct mapped_device *md)
1883 struct dm_target *ti = tio->ti;
1884 struct request *clone = NULL;
1888 r = ti->type->map_rq(ti, clone, &tio->info);
1890 r = ti->type->clone_and_map_rq(ti, rq, &tio->info, &clone);
1892 /* The target wants to complete the I/O */
1893 dm_kill_unmapped_request(rq, r);
1896 if (r != DM_MAPIO_REMAPPED)
1898 setup_clone(clone, rq, tio);
1902 case DM_MAPIO_SUBMITTED:
1903 /* The target has taken the I/O to submit by itself later */
1905 case DM_MAPIO_REMAPPED:
1906 /* The target has remapped the I/O so dispatch it */
1907 trace_block_rq_remap(clone->q, clone, disk_devt(dm_disk(md)),
1909 dm_dispatch_clone_request(clone, rq);
1911 case DM_MAPIO_REQUEUE:
1912 /* The target wants to requeue the I/O */
1913 dm_requeue_unmapped_request(clone);
1917 DMWARN("unimplemented target map return value: %d", r);
1921 /* The target wants to complete the I/O */
1922 dm_kill_unmapped_request(rq, r);
1929 static void map_tio_request(struct kthread_work *work)
1931 struct dm_rq_target_io *tio = container_of(work, struct dm_rq_target_io, work);
1932 struct request *rq = tio->orig;
1933 struct mapped_device *md = tio->md;
1935 if (map_request(tio, rq, md) == DM_MAPIO_REQUEUE)
1936 dm_requeue_unmapped_original_request(md, rq);
1939 static void dm_start_request(struct mapped_device *md, struct request *orig)
1941 if (!orig->q->mq_ops)
1942 blk_start_request(orig);
1944 blk_mq_start_request(orig);
1945 atomic_inc(&md->pending[rq_data_dir(orig)]);
1947 if (md->seq_rq_merge_deadline_usecs) {
1948 md->last_rq_pos = rq_end_sector(orig);
1949 md->last_rq_rw = rq_data_dir(orig);
1950 md->last_rq_start_time = ktime_get();
1954 * Hold the md reference here for the in-flight I/O.
1955 * We can't rely on the reference count by device opener,
1956 * because the device may be closed during the request completion
1957 * when all bios are completed.
1958 * See the comment in rq_completed() too.
1963 #define MAX_SEQ_RQ_MERGE_DEADLINE_USECS 100000
1965 ssize_t dm_attr_rq_based_seq_io_merge_deadline_show(struct mapped_device *md, char *buf)
1967 return sprintf(buf, "%u\n", md->seq_rq_merge_deadline_usecs);
1970 ssize_t dm_attr_rq_based_seq_io_merge_deadline_store(struct mapped_device *md,
1971 const char *buf, size_t count)
1975 if (!dm_request_based(md) || md->use_blk_mq)
1978 if (kstrtouint(buf, 10, &deadline))
1981 if (deadline > MAX_SEQ_RQ_MERGE_DEADLINE_USECS)
1982 deadline = MAX_SEQ_RQ_MERGE_DEADLINE_USECS;
1984 md->seq_rq_merge_deadline_usecs = deadline;
1989 static bool dm_request_peeked_before_merge_deadline(struct mapped_device *md)
1991 ktime_t kt_deadline;
1993 if (!md->seq_rq_merge_deadline_usecs)
1996 kt_deadline = ns_to_ktime((u64)md->seq_rq_merge_deadline_usecs * NSEC_PER_USEC);
1997 kt_deadline = ktime_add_safe(md->last_rq_start_time, kt_deadline);
1999 return !ktime_after(ktime_get(), kt_deadline);
2003 * q->request_fn for request-based dm.
2004 * Called with the queue lock held.
2006 static void dm_request_fn(struct request_queue *q)
2008 struct mapped_device *md = q->queuedata;
2010 struct dm_table *map = dm_get_live_table(md, &srcu_idx);
2011 struct dm_target *ti;
2013 struct dm_rq_target_io *tio;
2017 * For suspend, check blk_queue_stopped() and increment
2018 * ->pending within a single queue_lock not to increment the
2019 * number of in-flight I/Os after the queue is stopped in
2022 while (!blk_queue_stopped(q)) {
2023 rq = blk_peek_request(q);
2027 /* always use block 0 to find the target for flushes for now */
2029 if (!(rq->cmd_flags & REQ_FLUSH))
2030 pos = blk_rq_pos(rq);
2032 ti = dm_table_find_target(map, pos);
2033 if (!dm_target_is_valid(ti)) {
2035 * Must perform setup, that rq_completed() requires,
2036 * before calling dm_kill_unmapped_request
2038 DMERR_LIMIT("request attempted access beyond the end of device");
2039 dm_start_request(md, rq);
2040 dm_kill_unmapped_request(rq, -EIO);
2044 if (dm_request_peeked_before_merge_deadline(md) &&
2045 md_in_flight(md) && rq->bio && rq->bio->bi_vcnt == 1 &&
2046 md->last_rq_pos == pos && md->last_rq_rw == rq_data_dir(rq))
2049 if (ti->type->busy && ti->type->busy(ti))
2052 dm_start_request(md, rq);
2054 tio = tio_from_request(rq);
2055 /* Establish tio->ti before queuing work (map_tio_request) */
2057 queue_kthread_work(&md->kworker, &tio->work);
2058 BUG_ON(!irqs_disabled());
2064 blk_delay_queue(q, HZ / 100);
2066 dm_put_live_table(md, srcu_idx);
2069 static int dm_any_congested(void *congested_data, int bdi_bits)
2072 struct mapped_device *md = congested_data;
2073 struct dm_table *map;
2075 if (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
2076 map = dm_get_live_table_fast(md);
2079 * Request-based dm cares about only own queue for
2080 * the query about congestion status of request_queue
2082 if (dm_request_based(md))
2083 r = md->queue->backing_dev_info.state &
2086 r = dm_table_any_congested(map, bdi_bits);
2088 dm_put_live_table_fast(md);
2094 /*-----------------------------------------------------------------
2095 * An IDR is used to keep track of allocated minor numbers.
2096 *---------------------------------------------------------------*/
2097 static void free_minor(int minor)
2099 spin_lock(&_minor_lock);
2100 idr_remove(&_minor_idr, minor);
2101 spin_unlock(&_minor_lock);
2105 * See if the device with a specific minor # is free.
2107 static int specific_minor(int minor)
2111 if (minor >= (1 << MINORBITS))
2114 idr_preload(GFP_KERNEL);
2115 spin_lock(&_minor_lock);
2117 r = idr_alloc(&_minor_idr, MINOR_ALLOCED, minor, minor + 1, GFP_NOWAIT);
2119 spin_unlock(&_minor_lock);
2122 return r == -ENOSPC ? -EBUSY : r;
2126 static int next_free_minor(int *minor)
2130 idr_preload(GFP_KERNEL);
2131 spin_lock(&_minor_lock);
2133 r = idr_alloc(&_minor_idr, MINOR_ALLOCED, 0, 1 << MINORBITS, GFP_NOWAIT);
2135 spin_unlock(&_minor_lock);
2143 static const struct block_device_operations dm_blk_dops;
2145 static void dm_wq_work(struct work_struct *work);
2147 static void dm_init_md_queue(struct mapped_device *md)
2150 * Request-based dm devices cannot be stacked on top of bio-based dm
2151 * devices. The type of this dm device may not have been decided yet.
2152 * The type is decided at the first table loading time.
2153 * To prevent problematic device stacking, clear the queue flag
2154 * for request stacking support until then.
2156 * This queue is new, so no concurrency on the queue_flags.
2158 queue_flag_clear_unlocked(QUEUE_FLAG_STACKABLE, md->queue);
2161 static void dm_init_old_md_queue(struct mapped_device *md)
2163 md->use_blk_mq = false;
2164 dm_init_md_queue(md);
2167 * Initialize aspects of queue that aren't relevant for blk-mq
2169 md->queue->queuedata = md;
2170 md->queue->backing_dev_info.congested_fn = dm_any_congested;
2171 md->queue->backing_dev_info.congested_data = md;
2173 blk_queue_bounce_limit(md->queue, BLK_BOUNCE_ANY);
2177 * Allocate and initialise a blank device with a given minor.
2179 static struct mapped_device *alloc_dev(int minor)
2182 struct mapped_device *md = kzalloc(sizeof(*md), GFP_KERNEL);
2186 DMWARN("unable to allocate device, out of memory.");
2190 if (!try_module_get(THIS_MODULE))
2191 goto bad_module_get;
2193 /* get a minor number for the dev */
2194 if (minor == DM_ANY_MINOR)
2195 r = next_free_minor(&minor);
2197 r = specific_minor(minor);
2201 r = init_srcu_struct(&md->io_barrier);
2203 goto bad_io_barrier;
2205 md->use_blk_mq = use_blk_mq;
2206 md->type = DM_TYPE_NONE;
2207 mutex_init(&md->suspend_lock);
2208 mutex_init(&md->type_lock);
2209 mutex_init(&md->table_devices_lock);
2210 spin_lock_init(&md->deferred_lock);
2211 atomic_set(&md->holders, 1);
2212 atomic_set(&md->open_count, 0);
2213 atomic_set(&md->event_nr, 0);
2214 atomic_set(&md->uevent_seq, 0);
2215 INIT_LIST_HEAD(&md->uevent_list);
2216 INIT_LIST_HEAD(&md->table_devices);
2217 spin_lock_init(&md->uevent_lock);
2219 md->queue = blk_alloc_queue(GFP_KERNEL);
2223 dm_init_md_queue(md);
2225 md->disk = alloc_disk(1);
2229 atomic_set(&md->pending[0], 0);
2230 atomic_set(&md->pending[1], 0);
2231 init_waitqueue_head(&md->wait);
2232 INIT_WORK(&md->work, dm_wq_work);
2233 init_waitqueue_head(&md->eventq);
2234 init_completion(&md->kobj_holder.completion);
2235 md->kworker_task = NULL;
2237 md->disk->major = _major;
2238 md->disk->first_minor = minor;
2239 md->disk->fops = &dm_blk_dops;
2240 md->disk->queue = md->queue;
2241 md->disk->private_data = md;
2242 sprintf(md->disk->disk_name, "dm-%d", minor);
2244 format_dev_t(md->name, MKDEV(_major, minor));
2246 md->wq = alloc_workqueue("kdmflush", WQ_MEM_RECLAIM, 0);
2250 md->bdev = bdget_disk(md->disk, 0);
2254 bio_init(&md->flush_bio);
2255 md->flush_bio.bi_bdev = md->bdev;
2256 md->flush_bio.bi_rw = WRITE_FLUSH;
2258 dm_stats_init(&md->stats);
2260 /* Populate the mapping, nobody knows we exist yet */
2261 spin_lock(&_minor_lock);
2262 old_md = idr_replace(&_minor_idr, md, minor);
2263 spin_unlock(&_minor_lock);
2265 BUG_ON(old_md != MINOR_ALLOCED);
2270 destroy_workqueue(md->wq);
2272 del_gendisk(md->disk);
2275 blk_cleanup_queue(md->queue);
2277 cleanup_srcu_struct(&md->io_barrier);
2281 module_put(THIS_MODULE);
2287 static void unlock_fs(struct mapped_device *md);
2289 static void free_dev(struct mapped_device *md)
2291 int minor = MINOR(disk_devt(md->disk));
2294 destroy_workqueue(md->wq);
2296 if (md->kworker_task)
2297 kthread_stop(md->kworker_task);
2299 mempool_destroy(md->io_pool);
2301 mempool_destroy(md->rq_pool);
2303 bioset_free(md->bs);
2305 cleanup_srcu_struct(&md->io_barrier);
2306 free_table_devices(&md->table_devices);
2307 dm_stats_cleanup(&md->stats);
2309 spin_lock(&_minor_lock);
2310 md->disk->private_data = NULL;
2311 spin_unlock(&_minor_lock);
2312 if (blk_get_integrity(md->disk))
2313 blk_integrity_unregister(md->disk);
2314 del_gendisk(md->disk);
2316 blk_cleanup_queue(md->queue);
2318 blk_mq_free_tag_set(&md->tag_set);
2322 module_put(THIS_MODULE);
2326 static unsigned filter_md_type(unsigned type, struct mapped_device *md)
2328 if (type == DM_TYPE_BIO_BASED)
2331 return !md->use_blk_mq ? DM_TYPE_REQUEST_BASED : DM_TYPE_MQ_REQUEST_BASED;
2334 static void __bind_mempools(struct mapped_device *md, struct dm_table *t)
2336 struct dm_md_mempools *p = dm_table_get_md_mempools(t);
2338 switch (filter_md_type(dm_table_get_type(t), md)) {
2339 case DM_TYPE_BIO_BASED:
2340 if (md->bs && md->io_pool) {
2342 * This bio-based md already has necessary mempools.
2343 * Reload bioset because front_pad may have changed
2344 * because a different table was loaded.
2346 bioset_free(md->bs);
2352 case DM_TYPE_REQUEST_BASED:
2353 if (md->rq_pool && md->io_pool)
2355 * This request-based md already has necessary mempools.
2359 case DM_TYPE_MQ_REQUEST_BASED:
2360 BUG_ON(p); /* No mempools needed */
2364 BUG_ON(!p || md->io_pool || md->rq_pool || md->bs);
2366 md->io_pool = p->io_pool;
2368 md->rq_pool = p->rq_pool;
2373 /* mempool bind completed, no longer need any mempools in the table */
2374 dm_table_free_md_mempools(t);
2378 * Bind a table to the device.
2380 static void event_callback(void *context)
2382 unsigned long flags;
2384 struct mapped_device *md = (struct mapped_device *) context;
2386 spin_lock_irqsave(&md->uevent_lock, flags);
2387 list_splice_init(&md->uevent_list, &uevents);
2388 spin_unlock_irqrestore(&md->uevent_lock, flags);
2390 dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj);
2392 atomic_inc(&md->event_nr);
2393 wake_up(&md->eventq);
2397 * Protected by md->suspend_lock obtained by dm_swap_table().
2399 static void __set_size(struct mapped_device *md, sector_t size)
2401 set_capacity(md->disk, size);
2403 i_size_write(md->bdev->bd_inode, (loff_t)size << SECTOR_SHIFT);
2407 * Return 1 if the queue has a compulsory merge_bvec_fn function.
2409 * If this function returns 0, then the device is either a non-dm
2410 * device without a merge_bvec_fn, or it is a dm device that is
2411 * able to split any bios it receives that are too big.
2413 int dm_queue_merge_is_compulsory(struct request_queue *q)
2415 struct mapped_device *dev_md;
2417 if (!q->merge_bvec_fn)
2420 if (q->make_request_fn == dm_make_request) {
2421 dev_md = q->queuedata;
2422 if (test_bit(DMF_MERGE_IS_OPTIONAL, &dev_md->flags))
2429 static int dm_device_merge_is_compulsory(struct dm_target *ti,
2430 struct dm_dev *dev, sector_t start,
2431 sector_t len, void *data)
2433 struct block_device *bdev = dev->bdev;
2434 struct request_queue *q = bdev_get_queue(bdev);
2436 return dm_queue_merge_is_compulsory(q);
2440 * Return 1 if it is acceptable to ignore merge_bvec_fn based
2441 * on the properties of the underlying devices.
2443 static int dm_table_merge_is_optional(struct dm_table *table)
2446 struct dm_target *ti;
2448 while (i < dm_table_get_num_targets(table)) {
2449 ti = dm_table_get_target(table, i++);
2451 if (ti->type->iterate_devices &&
2452 ti->type->iterate_devices(ti, dm_device_merge_is_compulsory, NULL))
2460 * Returns old map, which caller must destroy.
2462 static struct dm_table *__bind(struct mapped_device *md, struct dm_table *t,
2463 struct queue_limits *limits)
2465 struct dm_table *old_map;
2466 struct request_queue *q = md->queue;
2468 int merge_is_optional;
2470 size = dm_table_get_size(t);
2473 * Wipe any geometry if the size of the table changed.
2475 if (size != dm_get_size(md))
2476 memset(&md->geometry, 0, sizeof(md->geometry));
2478 __set_size(md, size);
2480 dm_table_event_callback(t, event_callback, md);
2483 * The queue hasn't been stopped yet, if the old table type wasn't
2484 * for request-based during suspension. So stop it to prevent
2485 * I/O mapping before resume.
2486 * This must be done before setting the queue restrictions,
2487 * because request-based dm may be run just after the setting.
2489 if (dm_table_request_based(t))
2492 __bind_mempools(md, t);
2494 merge_is_optional = dm_table_merge_is_optional(t);
2496 old_map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2497 rcu_assign_pointer(md->map, t);
2498 md->immutable_target_type = dm_table_get_immutable_target_type(t);
2500 dm_table_set_restrictions(t, q, limits);
2501 if (merge_is_optional)
2502 set_bit(DMF_MERGE_IS_OPTIONAL, &md->flags);
2504 clear_bit(DMF_MERGE_IS_OPTIONAL, &md->flags);
2512 * Returns unbound table for the caller to free.
2514 static struct dm_table *__unbind(struct mapped_device *md)
2516 struct dm_table *map = rcu_dereference_protected(md->map, 1);
2521 dm_table_event_callback(map, NULL, NULL);
2522 RCU_INIT_POINTER(md->map, NULL);
2529 * Constructor for a new device.
2531 int dm_create(int minor, struct mapped_device **result)
2533 struct mapped_device *md;
2535 md = alloc_dev(minor);
2546 * Functions to manage md->type.
2547 * All are required to hold md->type_lock.
2549 void dm_lock_md_type(struct mapped_device *md)
2551 mutex_lock(&md->type_lock);
2554 void dm_unlock_md_type(struct mapped_device *md)
2556 mutex_unlock(&md->type_lock);
2559 void dm_set_md_type(struct mapped_device *md, unsigned type)
2561 BUG_ON(!mutex_is_locked(&md->type_lock));
2565 unsigned dm_get_md_type(struct mapped_device *md)
2567 BUG_ON(!mutex_is_locked(&md->type_lock));
2571 struct target_type *dm_get_immutable_target_type(struct mapped_device *md)
2573 return md->immutable_target_type;
2577 * The queue_limits are only valid as long as you have a reference
2580 struct queue_limits *dm_get_queue_limits(struct mapped_device *md)
2582 BUG_ON(!atomic_read(&md->holders));
2583 return &md->queue->limits;
2585 EXPORT_SYMBOL_GPL(dm_get_queue_limits);
2587 static void init_rq_based_worker_thread(struct mapped_device *md)
2589 /* Initialize the request-based DM worker thread */
2590 init_kthread_worker(&md->kworker);
2591 md->kworker_task = kthread_run(kthread_worker_fn, &md->kworker,
2592 "kdmwork-%s", dm_device_name(md));
2596 * Fully initialize a request-based queue (->elevator, ->request_fn, etc).
2598 static int dm_init_request_based_queue(struct mapped_device *md)
2600 struct request_queue *q = NULL;
2602 /* Fully initialize the queue */
2603 q = blk_init_allocated_queue(md->queue, dm_request_fn, NULL);
2607 /* disable dm_request_fn's merge heuristic by default */
2608 md->seq_rq_merge_deadline_usecs = 0;
2611 dm_init_old_md_queue(md);
2612 blk_queue_softirq_done(md->queue, dm_softirq_done);
2613 blk_queue_prep_rq(md->queue, dm_prep_fn);
2615 init_rq_based_worker_thread(md);
2617 elv_register_queue(md->queue);
2622 static int dm_mq_init_request(void *data, struct request *rq,
2623 unsigned int hctx_idx, unsigned int request_idx,
2624 unsigned int numa_node)
2626 struct mapped_device *md = data;
2627 struct dm_rq_target_io *tio = blk_mq_rq_to_pdu(rq);
2630 * Must initialize md member of tio, otherwise it won't
2631 * be available in dm_mq_queue_rq.
2638 static int dm_mq_queue_rq(struct blk_mq_hw_ctx *hctx,
2639 const struct blk_mq_queue_data *bd)
2641 struct request *rq = bd->rq;
2642 struct dm_rq_target_io *tio = blk_mq_rq_to_pdu(rq);
2643 struct mapped_device *md = tio->md;
2645 struct dm_table *map = dm_get_live_table(md, &srcu_idx);
2646 struct dm_target *ti;
2649 /* always use block 0 to find the target for flushes for now */
2651 if (!(rq->cmd_flags & REQ_FLUSH))
2652 pos = blk_rq_pos(rq);
2654 ti = dm_table_find_target(map, pos);
2655 if (!dm_target_is_valid(ti)) {
2656 dm_put_live_table(md, srcu_idx);
2657 DMERR_LIMIT("request attempted access beyond the end of device");
2659 * Must perform setup, that rq_completed() requires,
2660 * before returning BLK_MQ_RQ_QUEUE_ERROR
2662 dm_start_request(md, rq);
2663 return BLK_MQ_RQ_QUEUE_ERROR;
2665 dm_put_live_table(md, srcu_idx);
2667 if (ti->type->busy && ti->type->busy(ti))
2668 return BLK_MQ_RQ_QUEUE_BUSY;
2670 dm_start_request(md, rq);
2672 /* Init tio using md established in .init_request */
2673 init_tio(tio, rq, md);
2676 * Establish tio->ti before queuing work (map_tio_request)
2677 * or making direct call to map_request().
2681 /* Clone the request if underlying devices aren't blk-mq */
2682 if (dm_table_get_type(map) == DM_TYPE_REQUEST_BASED) {
2683 /* clone request is allocated at the end of the pdu */
2684 tio->clone = (void *)blk_mq_rq_to_pdu(rq) + sizeof(struct dm_rq_target_io);
2685 (void) clone_rq(rq, md, tio, GFP_ATOMIC);
2686 queue_kthread_work(&md->kworker, &tio->work);
2688 /* Direct call is fine since .queue_rq allows allocations */
2689 if (map_request(tio, rq, md) == DM_MAPIO_REQUEUE) {
2690 /* Undo dm_start_request() before requeuing */
2691 rq_completed(md, rq_data_dir(rq), false);
2692 return BLK_MQ_RQ_QUEUE_BUSY;
2696 return BLK_MQ_RQ_QUEUE_OK;
2699 static struct blk_mq_ops dm_mq_ops = {
2700 .queue_rq = dm_mq_queue_rq,
2701 .map_queue = blk_mq_map_queue,
2702 .complete = dm_softirq_done,
2703 .init_request = dm_mq_init_request,
2706 static int dm_init_request_based_blk_mq_queue(struct mapped_device *md)
2708 unsigned md_type = dm_get_md_type(md);
2709 struct request_queue *q;
2712 memset(&md->tag_set, 0, sizeof(md->tag_set));
2713 md->tag_set.ops = &dm_mq_ops;
2714 md->tag_set.queue_depth = BLKDEV_MAX_RQ;
2715 md->tag_set.numa_node = NUMA_NO_NODE;
2716 md->tag_set.flags = BLK_MQ_F_SHOULD_MERGE | BLK_MQ_F_SG_MERGE;
2717 md->tag_set.nr_hw_queues = 1;
2718 if (md_type == DM_TYPE_REQUEST_BASED) {
2719 /* make the memory for non-blk-mq clone part of the pdu */
2720 md->tag_set.cmd_size = sizeof(struct dm_rq_target_io) + sizeof(struct request);
2722 md->tag_set.cmd_size = sizeof(struct dm_rq_target_io);
2723 md->tag_set.driver_data = md;
2725 err = blk_mq_alloc_tag_set(&md->tag_set);
2729 q = blk_mq_init_allocated_queue(&md->tag_set, md->queue);
2735 dm_init_md_queue(md);
2737 /* backfill 'mq' sysfs registration normally done in blk_register_queue */
2738 blk_mq_register_disk(md->disk);
2740 if (md_type == DM_TYPE_REQUEST_BASED)
2741 init_rq_based_worker_thread(md);
2746 blk_mq_free_tag_set(&md->tag_set);
2751 * Setup the DM device's queue based on md's type
2753 int dm_setup_md_queue(struct mapped_device *md)
2756 unsigned md_type = filter_md_type(dm_get_md_type(md), md);
2759 case DM_TYPE_REQUEST_BASED:
2760 r = dm_init_request_based_queue(md);
2762 DMWARN("Cannot initialize queue for request-based mapped device");
2766 case DM_TYPE_MQ_REQUEST_BASED:
2767 r = dm_init_request_based_blk_mq_queue(md);
2769 DMWARN("Cannot initialize queue for request-based blk-mq mapped device");
2773 case DM_TYPE_BIO_BASED:
2774 dm_init_old_md_queue(md);
2775 blk_queue_make_request(md->queue, dm_make_request);
2776 blk_queue_merge_bvec(md->queue, dm_merge_bvec);
2783 struct mapped_device *dm_get_md(dev_t dev)
2785 struct mapped_device *md;
2786 unsigned minor = MINOR(dev);
2788 if (MAJOR(dev) != _major || minor >= (1 << MINORBITS))
2791 spin_lock(&_minor_lock);
2793 md = idr_find(&_minor_idr, minor);
2795 if ((md == MINOR_ALLOCED ||
2796 (MINOR(disk_devt(dm_disk(md))) != minor) ||
2797 dm_deleting_md(md) ||
2798 test_bit(DMF_FREEING, &md->flags))) {
2806 spin_unlock(&_minor_lock);
2810 EXPORT_SYMBOL_GPL(dm_get_md);
2812 void *dm_get_mdptr(struct mapped_device *md)
2814 return md->interface_ptr;
2817 void dm_set_mdptr(struct mapped_device *md, void *ptr)
2819 md->interface_ptr = ptr;
2822 void dm_get(struct mapped_device *md)
2824 atomic_inc(&md->holders);
2825 BUG_ON(test_bit(DMF_FREEING, &md->flags));
2828 int dm_hold(struct mapped_device *md)
2830 spin_lock(&_minor_lock);
2831 if (test_bit(DMF_FREEING, &md->flags)) {
2832 spin_unlock(&_minor_lock);
2836 spin_unlock(&_minor_lock);
2839 EXPORT_SYMBOL_GPL(dm_hold);
2841 const char *dm_device_name(struct mapped_device *md)
2845 EXPORT_SYMBOL_GPL(dm_device_name);
2847 static void __dm_destroy(struct mapped_device *md, bool wait)
2849 struct dm_table *map;
2854 map = dm_get_live_table(md, &srcu_idx);
2856 spin_lock(&_minor_lock);
2857 idr_replace(&_minor_idr, MINOR_ALLOCED, MINOR(disk_devt(dm_disk(md))));
2858 set_bit(DMF_FREEING, &md->flags);
2859 spin_unlock(&_minor_lock);
2861 if (dm_request_based(md) && md->kworker_task)
2862 flush_kthread_worker(&md->kworker);
2865 * Take suspend_lock so that presuspend and postsuspend methods
2866 * do not race with internal suspend.
2868 mutex_lock(&md->suspend_lock);
2869 if (!dm_suspended_md(md)) {
2870 dm_table_presuspend_targets(map);
2871 dm_table_postsuspend_targets(map);
2873 mutex_unlock(&md->suspend_lock);
2875 /* dm_put_live_table must be before msleep, otherwise deadlock is possible */
2876 dm_put_live_table(md, srcu_idx);
2879 * Rare, but there may be I/O requests still going to complete,
2880 * for example. Wait for all references to disappear.
2881 * No one should increment the reference count of the mapped_device,
2882 * after the mapped_device state becomes DMF_FREEING.
2885 while (atomic_read(&md->holders))
2887 else if (atomic_read(&md->holders))
2888 DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)",
2889 dm_device_name(md), atomic_read(&md->holders));
2892 dm_table_destroy(__unbind(md));
2896 void dm_destroy(struct mapped_device *md)
2898 __dm_destroy(md, true);
2901 void dm_destroy_immediate(struct mapped_device *md)
2903 __dm_destroy(md, false);
2906 void dm_put(struct mapped_device *md)
2908 atomic_dec(&md->holders);
2910 EXPORT_SYMBOL_GPL(dm_put);
2912 static int dm_wait_for_completion(struct mapped_device *md, int interruptible)
2915 DECLARE_WAITQUEUE(wait, current);
2917 add_wait_queue(&md->wait, &wait);
2920 set_current_state(interruptible);
2922 if (!md_in_flight(md))
2925 if (interruptible == TASK_INTERRUPTIBLE &&
2926 signal_pending(current)) {
2933 set_current_state(TASK_RUNNING);
2935 remove_wait_queue(&md->wait, &wait);
2941 * Process the deferred bios
2943 static void dm_wq_work(struct work_struct *work)
2945 struct mapped_device *md = container_of(work, struct mapped_device,
2949 struct dm_table *map;
2951 map = dm_get_live_table(md, &srcu_idx);
2953 while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
2954 spin_lock_irq(&md->deferred_lock);
2955 c = bio_list_pop(&md->deferred);
2956 spin_unlock_irq(&md->deferred_lock);
2961 if (dm_request_based(md))
2962 generic_make_request(c);
2964 __split_and_process_bio(md, map, c);
2967 dm_put_live_table(md, srcu_idx);
2970 static void dm_queue_flush(struct mapped_device *md)
2972 clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2973 smp_mb__after_atomic();
2974 queue_work(md->wq, &md->work);
2978 * Swap in a new table, returning the old one for the caller to destroy.
2980 struct dm_table *dm_swap_table(struct mapped_device *md, struct dm_table *table)
2982 struct dm_table *live_map = NULL, *map = ERR_PTR(-EINVAL);
2983 struct queue_limits limits;
2986 mutex_lock(&md->suspend_lock);
2988 /* device must be suspended */
2989 if (!dm_suspended_md(md))
2993 * If the new table has no data devices, retain the existing limits.
2994 * This helps multipath with queue_if_no_path if all paths disappear,
2995 * then new I/O is queued based on these limits, and then some paths
2998 if (dm_table_has_no_data_devices(table)) {
2999 live_map = dm_get_live_table_fast(md);
3001 limits = md->queue->limits;
3002 dm_put_live_table_fast(md);
3006 r = dm_calculate_queue_limits(table, &limits);
3013 map = __bind(md, table, &limits);
3016 mutex_unlock(&md->suspend_lock);
3021 * Functions to lock and unlock any filesystem running on the
3024 static int lock_fs(struct mapped_device *md)
3028 WARN_ON(md->frozen_sb);
3030 md->frozen_sb = freeze_bdev(md->bdev);
3031 if (IS_ERR(md->frozen_sb)) {
3032 r = PTR_ERR(md->frozen_sb);
3033 md->frozen_sb = NULL;
3037 set_bit(DMF_FROZEN, &md->flags);
3042 static void unlock_fs(struct mapped_device *md)
3044 if (!test_bit(DMF_FROZEN, &md->flags))
3047 thaw_bdev(md->bdev, md->frozen_sb);
3048 md->frozen_sb = NULL;
3049 clear_bit(DMF_FROZEN, &md->flags);
3053 * If __dm_suspend returns 0, the device is completely quiescent
3054 * now. There is no request-processing activity. All new requests
3055 * are being added to md->deferred list.
3057 * Caller must hold md->suspend_lock
3059 static int __dm_suspend(struct mapped_device *md, struct dm_table *map,
3060 unsigned suspend_flags, int interruptible)
3062 bool do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG;
3063 bool noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG;
3067 * DMF_NOFLUSH_SUSPENDING must be set before presuspend.
3068 * This flag is cleared before dm_suspend returns.
3071 set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
3074 * This gets reverted if there's an error later and the targets
3075 * provide the .presuspend_undo hook.
3077 dm_table_presuspend_targets(map);
3080 * Flush I/O to the device.
3081 * Any I/O submitted after lock_fs() may not be flushed.
3082 * noflush takes precedence over do_lockfs.
3083 * (lock_fs() flushes I/Os and waits for them to complete.)
3085 if (!noflush && do_lockfs) {
3088 dm_table_presuspend_undo_targets(map);
3094 * Here we must make sure that no processes are submitting requests
3095 * to target drivers i.e. no one may be executing
3096 * __split_and_process_bio. This is called from dm_request and
3099 * To get all processes out of __split_and_process_bio in dm_request,
3100 * we take the write lock. To prevent any process from reentering
3101 * __split_and_process_bio from dm_request and quiesce the thread
3102 * (dm_wq_work), we set BMF_BLOCK_IO_FOR_SUSPEND and call
3103 * flush_workqueue(md->wq).
3105 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
3107 synchronize_srcu(&md->io_barrier);
3110 * Stop md->queue before flushing md->wq in case request-based
3111 * dm defers requests to md->wq from md->queue.
3113 if (dm_request_based(md)) {
3114 stop_queue(md->queue);
3115 if (md->kworker_task)
3116 flush_kthread_worker(&md->kworker);
3119 flush_workqueue(md->wq);
3122 * At this point no more requests are entering target request routines.
3123 * We call dm_wait_for_completion to wait for all existing requests
3126 r = dm_wait_for_completion(md, interruptible);
3129 clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
3131 synchronize_srcu(&md->io_barrier);
3133 /* were we interrupted ? */
3137 if (dm_request_based(md))
3138 start_queue(md->queue);
3141 dm_table_presuspend_undo_targets(map);
3142 /* pushback list is already flushed, so skip flush */
3149 * We need to be able to change a mapping table under a mounted
3150 * filesystem. For example we might want to move some data in
3151 * the background. Before the table can be swapped with
3152 * dm_bind_table, dm_suspend must be called to flush any in
3153 * flight bios and ensure that any further io gets deferred.
3156 * Suspend mechanism in request-based dm.
3158 * 1. Flush all I/Os by lock_fs() if needed.
3159 * 2. Stop dispatching any I/O by stopping the request_queue.
3160 * 3. Wait for all in-flight I/Os to be completed or requeued.
3162 * To abort suspend, start the request_queue.
3164 int dm_suspend(struct mapped_device *md, unsigned suspend_flags)
3166 struct dm_table *map = NULL;
3170 mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING);
3172 if (dm_suspended_md(md)) {
3177 if (dm_suspended_internally_md(md)) {
3178 /* already internally suspended, wait for internal resume */
3179 mutex_unlock(&md->suspend_lock);
3180 r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE);
3186 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
3188 r = __dm_suspend(md, map, suspend_flags, TASK_INTERRUPTIBLE);
3192 set_bit(DMF_SUSPENDED, &md->flags);
3194 dm_table_postsuspend_targets(map);
3197 mutex_unlock(&md->suspend_lock);
3201 static int __dm_resume(struct mapped_device *md, struct dm_table *map)
3204 int r = dm_table_resume_targets(map);
3212 * Flushing deferred I/Os must be done after targets are resumed
3213 * so that mapping of targets can work correctly.
3214 * Request-based dm is queueing the deferred I/Os in its request_queue.
3216 if (dm_request_based(md))
3217 start_queue(md->queue);
3224 int dm_resume(struct mapped_device *md)
3227 struct dm_table *map = NULL;
3230 mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING);
3232 if (!dm_suspended_md(md))
3235 if (dm_suspended_internally_md(md)) {
3236 /* already internally suspended, wait for internal resume */
3237 mutex_unlock(&md->suspend_lock);
3238 r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE);
3244 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
3245 if (!map || !dm_table_get_size(map))
3248 r = __dm_resume(md, map);
3252 clear_bit(DMF_SUSPENDED, &md->flags);
3256 mutex_unlock(&md->suspend_lock);
3262 * Internal suspend/resume works like userspace-driven suspend. It waits
3263 * until all bios finish and prevents issuing new bios to the target drivers.
3264 * It may be used only from the kernel.
3267 static void __dm_internal_suspend(struct mapped_device *md, unsigned suspend_flags)
3269 struct dm_table *map = NULL;
3271 if (md->internal_suspend_count++)
3272 return; /* nested internal suspend */
3274 if (dm_suspended_md(md)) {
3275 set_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
3276 return; /* nest suspend */
3279 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
3282 * Using TASK_UNINTERRUPTIBLE because only NOFLUSH internal suspend is
3283 * supported. Properly supporting a TASK_INTERRUPTIBLE internal suspend
3284 * would require changing .presuspend to return an error -- avoid this
3285 * until there is a need for more elaborate variants of internal suspend.
3287 (void) __dm_suspend(md, map, suspend_flags, TASK_UNINTERRUPTIBLE);
3289 set_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
3291 dm_table_postsuspend_targets(map);
3294 static void __dm_internal_resume(struct mapped_device *md)
3296 BUG_ON(!md->internal_suspend_count);
3298 if (--md->internal_suspend_count)
3299 return; /* resume from nested internal suspend */
3301 if (dm_suspended_md(md))
3302 goto done; /* resume from nested suspend */
3305 * NOTE: existing callers don't need to call dm_table_resume_targets
3306 * (which may fail -- so best to avoid it for now by passing NULL map)
3308 (void) __dm_resume(md, NULL);
3311 clear_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
3312 smp_mb__after_atomic();
3313 wake_up_bit(&md->flags, DMF_SUSPENDED_INTERNALLY);
3316 void dm_internal_suspend_noflush(struct mapped_device *md)
3318 mutex_lock(&md->suspend_lock);
3319 __dm_internal_suspend(md, DM_SUSPEND_NOFLUSH_FLAG);
3320 mutex_unlock(&md->suspend_lock);
3322 EXPORT_SYMBOL_GPL(dm_internal_suspend_noflush);
3324 void dm_internal_resume(struct mapped_device *md)
3326 mutex_lock(&md->suspend_lock);
3327 __dm_internal_resume(md);
3328 mutex_unlock(&md->suspend_lock);
3330 EXPORT_SYMBOL_GPL(dm_internal_resume);
3333 * Fast variants of internal suspend/resume hold md->suspend_lock,
3334 * which prevents interaction with userspace-driven suspend.
3337 void dm_internal_suspend_fast(struct mapped_device *md)
3339 mutex_lock(&md->suspend_lock);
3340 if (dm_suspended_md(md) || dm_suspended_internally_md(md))
3343 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
3344 synchronize_srcu(&md->io_barrier);
3345 flush_workqueue(md->wq);
3346 dm_wait_for_completion(md, TASK_UNINTERRUPTIBLE);
3348 EXPORT_SYMBOL_GPL(dm_internal_suspend_fast);
3350 void dm_internal_resume_fast(struct mapped_device *md)
3352 if (dm_suspended_md(md) || dm_suspended_internally_md(md))
3358 mutex_unlock(&md->suspend_lock);
3360 EXPORT_SYMBOL_GPL(dm_internal_resume_fast);
3362 /*-----------------------------------------------------------------
3363 * Event notification.
3364 *---------------------------------------------------------------*/
3365 int dm_kobject_uevent(struct mapped_device *md, enum kobject_action action,
3368 char udev_cookie[DM_COOKIE_LENGTH];
3369 char *envp[] = { udev_cookie, NULL };
3372 return kobject_uevent(&disk_to_dev(md->disk)->kobj, action);
3374 snprintf(udev_cookie, DM_COOKIE_LENGTH, "%s=%u",
3375 DM_COOKIE_ENV_VAR_NAME, cookie);
3376 return kobject_uevent_env(&disk_to_dev(md->disk)->kobj,
3381 uint32_t dm_next_uevent_seq(struct mapped_device *md)
3383 return atomic_add_return(1, &md->uevent_seq);
3386 uint32_t dm_get_event_nr(struct mapped_device *md)
3388 return atomic_read(&md->event_nr);
3391 int dm_wait_event(struct mapped_device *md, int event_nr)
3393 return wait_event_interruptible(md->eventq,
3394 (event_nr != atomic_read(&md->event_nr)));
3397 void dm_uevent_add(struct mapped_device *md, struct list_head *elist)
3399 unsigned long flags;
3401 spin_lock_irqsave(&md->uevent_lock, flags);
3402 list_add(elist, &md->uevent_list);
3403 spin_unlock_irqrestore(&md->uevent_lock, flags);
3407 * The gendisk is only valid as long as you have a reference
3410 struct gendisk *dm_disk(struct mapped_device *md)
3414 EXPORT_SYMBOL_GPL(dm_disk);
3416 struct kobject *dm_kobject(struct mapped_device *md)
3418 return &md->kobj_holder.kobj;
3421 struct mapped_device *dm_get_from_kobject(struct kobject *kobj)
3423 struct mapped_device *md;
3425 md = container_of(kobj, struct mapped_device, kobj_holder.kobj);
3427 if (test_bit(DMF_FREEING, &md->flags) ||
3435 int dm_suspended_md(struct mapped_device *md)
3437 return test_bit(DMF_SUSPENDED, &md->flags);
3440 int dm_suspended_internally_md(struct mapped_device *md)
3442 return test_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
3445 int dm_test_deferred_remove_flag(struct mapped_device *md)
3447 return test_bit(DMF_DEFERRED_REMOVE, &md->flags);
3450 int dm_suspended(struct dm_target *ti)
3452 return dm_suspended_md(dm_table_get_md(ti->table));
3454 EXPORT_SYMBOL_GPL(dm_suspended);
3456 int dm_noflush_suspending(struct dm_target *ti)
3458 return __noflush_suspending(dm_table_get_md(ti->table));
3460 EXPORT_SYMBOL_GPL(dm_noflush_suspending);
3462 struct dm_md_mempools *dm_alloc_bio_mempools(unsigned integrity,
3463 unsigned per_bio_data_size)
3465 struct dm_md_mempools *pools;
3466 unsigned int pool_size = dm_get_reserved_bio_based_ios();
3467 unsigned int front_pad;
3469 pools = kzalloc(sizeof(*pools), GFP_KERNEL);
3471 return ERR_PTR(-ENOMEM);
3473 front_pad = roundup(per_bio_data_size, __alignof__(struct dm_target_io)) +
3474 offsetof(struct dm_target_io, clone);
3476 pools->io_pool = mempool_create_slab_pool(pool_size, _io_cache);
3477 if (!pools->io_pool)
3480 pools->bs = bioset_create_nobvec(pool_size, front_pad);
3484 if (integrity && bioset_integrity_create(pools->bs, pool_size))
3489 dm_free_md_mempools(pools);
3490 return ERR_PTR(-ENOMEM);
3493 struct dm_md_mempools *dm_alloc_rq_mempools(struct mapped_device *md,
3496 unsigned int pool_size;
3497 struct dm_md_mempools *pools;
3499 if (filter_md_type(type, md) == DM_TYPE_MQ_REQUEST_BASED)
3500 return NULL; /* No mempools needed */
3502 pool_size = dm_get_reserved_rq_based_ios();
3503 pools = kzalloc(sizeof(*pools), GFP_KERNEL);
3505 return ERR_PTR(-ENOMEM);
3507 pools->rq_pool = mempool_create_slab_pool(pool_size, _rq_cache);
3508 if (!pools->rq_pool)
3511 pools->io_pool = mempool_create_slab_pool(pool_size, _rq_tio_cache);
3512 if (!pools->io_pool)
3517 dm_free_md_mempools(pools);
3518 return ERR_PTR(-ENOMEM);
3521 void dm_free_md_mempools(struct dm_md_mempools *pools)
3527 mempool_destroy(pools->io_pool);
3530 mempool_destroy(pools->rq_pool);
3533 bioset_free(pools->bs);
3538 static const struct block_device_operations dm_blk_dops = {
3539 .open = dm_blk_open,
3540 .release = dm_blk_close,
3541 .ioctl = dm_blk_ioctl,
3542 .getgeo = dm_blk_getgeo,
3543 .owner = THIS_MODULE
3549 module_init(dm_init);
3550 module_exit(dm_exit);
3552 module_param(major, uint, 0);
3553 MODULE_PARM_DESC(major, "The major number of the device mapper");
3555 module_param(reserved_bio_based_ios, uint, S_IRUGO | S_IWUSR);
3556 MODULE_PARM_DESC(reserved_bio_based_ios, "Reserved IOs in bio-based mempools");
3558 module_param(reserved_rq_based_ios, uint, S_IRUGO | S_IWUSR);
3559 MODULE_PARM_DESC(reserved_rq_based_ios, "Reserved IOs in request-based mempools");
3561 module_param(use_blk_mq, bool, S_IRUGO | S_IWUSR);
3562 MODULE_PARM_DESC(use_blk_mq, "Use block multiqueue for request-based DM devices");
3564 MODULE_DESCRIPTION(DM_NAME " driver");
3565 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
3566 MODULE_LICENSE("GPL");