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>
25 #include <trace/events/block.h>
27 #define DM_MSG_PREFIX "core"
31 * ratelimit state to be used in DMXXX_LIMIT().
33 DEFINE_RATELIMIT_STATE(dm_ratelimit_state,
34 DEFAULT_RATELIMIT_INTERVAL,
35 DEFAULT_RATELIMIT_BURST);
36 EXPORT_SYMBOL(dm_ratelimit_state);
40 * Cookies are numeric values sent with CHANGE and REMOVE
41 * uevents while resuming, removing or renaming the device.
43 #define DM_COOKIE_ENV_VAR_NAME "DM_COOKIE"
44 #define DM_COOKIE_LENGTH 24
46 static const char *_name = DM_NAME;
48 static unsigned int major = 0;
49 static unsigned int _major = 0;
51 static DEFINE_IDR(_minor_idr);
53 static DEFINE_SPINLOCK(_minor_lock);
55 static void do_deferred_remove(struct work_struct *w);
57 static DECLARE_WORK(deferred_remove_work, do_deferred_remove);
59 static struct workqueue_struct *deferred_remove_workqueue;
63 * One of these is allocated per bio.
66 struct mapped_device *md;
70 unsigned long start_time;
71 spinlock_t endio_lock;
72 struct dm_stats_aux stats_aux;
76 * For request-based dm.
77 * One of these is allocated per request.
79 struct dm_rq_target_io {
80 struct mapped_device *md;
82 struct request *orig, *clone;
83 struct kthread_work work;
89 * For request-based dm - the bio clones we allocate are embedded in these
92 * We allocate these with bio_alloc_bioset, using the front_pad parameter when
93 * the bioset is created - this means the bio has to come at the end of the
96 struct dm_rq_clone_bio_info {
98 struct dm_rq_target_io *tio;
102 union map_info *dm_get_rq_mapinfo(struct request *rq)
104 if (rq && rq->end_io_data)
105 return &((struct dm_rq_target_io *)rq->end_io_data)->info;
108 EXPORT_SYMBOL_GPL(dm_get_rq_mapinfo);
110 #define MINOR_ALLOCED ((void *)-1)
113 * Bits for the md->flags field.
115 #define DMF_BLOCK_IO_FOR_SUSPEND 0
116 #define DMF_SUSPENDED 1
118 #define DMF_FREEING 3
119 #define DMF_DELETING 4
120 #define DMF_NOFLUSH_SUSPENDING 5
121 #define DMF_MERGE_IS_OPTIONAL 6
122 #define DMF_DEFERRED_REMOVE 7
123 #define DMF_SUSPENDED_INTERNALLY 8
126 * A dummy definition to make RCU happy.
127 * struct dm_table should never be dereferenced in this file.
134 * Work processed by per-device workqueue.
136 struct mapped_device {
137 struct srcu_struct io_barrier;
138 struct mutex suspend_lock;
143 * The current mapping.
144 * Use dm_get_live_table{_fast} or take suspend_lock for
147 struct dm_table __rcu *map;
149 struct list_head table_devices;
150 struct mutex table_devices_lock;
154 struct request_queue *queue;
156 /* Protect queue and type against concurrent access. */
157 struct mutex type_lock;
159 struct target_type *immutable_target_type;
161 struct gendisk *disk;
167 * A list of ios that arrived while we were suspended.
170 wait_queue_head_t wait;
171 struct work_struct work;
172 struct bio_list deferred;
173 spinlock_t deferred_lock;
176 * Processing queue (flush)
178 struct workqueue_struct *wq;
181 * io objects are allocated from here.
192 wait_queue_head_t eventq;
194 struct list_head uevent_list;
195 spinlock_t uevent_lock; /* Protect access to uevent_list */
198 * freeze/thaw support require holding onto a super block
200 struct super_block *frozen_sb;
201 struct block_device *bdev;
203 /* forced geometry settings */
204 struct hd_geometry geometry;
206 /* kobject and completion */
207 struct dm_kobject_holder kobj_holder;
209 /* zero-length flush that will be cloned and submitted to targets */
210 struct bio flush_bio;
212 struct dm_stats stats;
214 struct kthread_worker kworker;
215 struct task_struct *kworker_task;
219 * For mempools pre-allocation at the table loading time.
221 struct dm_md_mempools {
227 struct table_device {
228 struct list_head list;
230 struct dm_dev dm_dev;
233 #define RESERVED_BIO_BASED_IOS 16
234 #define RESERVED_REQUEST_BASED_IOS 256
235 #define RESERVED_MAX_IOS 1024
236 static struct kmem_cache *_io_cache;
237 static struct kmem_cache *_rq_tio_cache;
238 static struct kmem_cache *_rq_cache;
241 * Bio-based DM's mempools' reserved IOs set by the user.
243 static unsigned reserved_bio_based_ios = RESERVED_BIO_BASED_IOS;
246 * Request-based DM's mempools' reserved IOs set by the user.
248 static unsigned reserved_rq_based_ios = RESERVED_REQUEST_BASED_IOS;
250 static unsigned __dm_get_reserved_ios(unsigned *reserved_ios,
251 unsigned def, unsigned max)
253 unsigned ios = ACCESS_ONCE(*reserved_ios);
254 unsigned modified_ios = 0;
262 (void)cmpxchg(reserved_ios, ios, modified_ios);
269 unsigned dm_get_reserved_bio_based_ios(void)
271 return __dm_get_reserved_ios(&reserved_bio_based_ios,
272 RESERVED_BIO_BASED_IOS, RESERVED_MAX_IOS);
274 EXPORT_SYMBOL_GPL(dm_get_reserved_bio_based_ios);
276 unsigned dm_get_reserved_rq_based_ios(void)
278 return __dm_get_reserved_ios(&reserved_rq_based_ios,
279 RESERVED_REQUEST_BASED_IOS, RESERVED_MAX_IOS);
281 EXPORT_SYMBOL_GPL(dm_get_reserved_rq_based_ios);
283 static int __init local_init(void)
287 /* allocate a slab for the dm_ios */
288 _io_cache = KMEM_CACHE(dm_io, 0);
292 _rq_tio_cache = KMEM_CACHE(dm_rq_target_io, 0);
294 goto out_free_io_cache;
296 _rq_cache = kmem_cache_create("dm_clone_request", sizeof(struct request),
297 __alignof__(struct request), 0, NULL);
299 goto out_free_rq_tio_cache;
301 r = dm_uevent_init();
303 goto out_free_rq_cache;
305 deferred_remove_workqueue = alloc_workqueue("kdmremove", WQ_UNBOUND, 1);
306 if (!deferred_remove_workqueue) {
308 goto out_uevent_exit;
312 r = register_blkdev(_major, _name);
314 goto out_free_workqueue;
322 destroy_workqueue(deferred_remove_workqueue);
326 kmem_cache_destroy(_rq_cache);
327 out_free_rq_tio_cache:
328 kmem_cache_destroy(_rq_tio_cache);
330 kmem_cache_destroy(_io_cache);
335 static void local_exit(void)
337 flush_scheduled_work();
338 destroy_workqueue(deferred_remove_workqueue);
340 kmem_cache_destroy(_rq_cache);
341 kmem_cache_destroy(_rq_tio_cache);
342 kmem_cache_destroy(_io_cache);
343 unregister_blkdev(_major, _name);
348 DMINFO("cleaned up");
351 static int (*_inits[])(void) __initdata = {
362 static void (*_exits[])(void) = {
373 static int __init dm_init(void)
375 const int count = ARRAY_SIZE(_inits);
379 for (i = 0; i < count; i++) {
394 static void __exit dm_exit(void)
396 int i = ARRAY_SIZE(_exits);
402 * Should be empty by this point.
404 idr_destroy(&_minor_idr);
408 * Block device functions
410 int dm_deleting_md(struct mapped_device *md)
412 return test_bit(DMF_DELETING, &md->flags);
415 static int dm_blk_open(struct block_device *bdev, fmode_t mode)
417 struct mapped_device *md;
419 spin_lock(&_minor_lock);
421 md = bdev->bd_disk->private_data;
425 if (test_bit(DMF_FREEING, &md->flags) ||
426 dm_deleting_md(md)) {
432 atomic_inc(&md->open_count);
435 spin_unlock(&_minor_lock);
437 return md ? 0 : -ENXIO;
440 static void dm_blk_close(struct gendisk *disk, fmode_t mode)
442 struct mapped_device *md = disk->private_data;
444 spin_lock(&_minor_lock);
446 if (atomic_dec_and_test(&md->open_count) &&
447 (test_bit(DMF_DEFERRED_REMOVE, &md->flags)))
448 queue_work(deferred_remove_workqueue, &deferred_remove_work);
452 spin_unlock(&_minor_lock);
455 int dm_open_count(struct mapped_device *md)
457 return atomic_read(&md->open_count);
461 * Guarantees nothing is using the device before it's deleted.
463 int dm_lock_for_deletion(struct mapped_device *md, bool mark_deferred, bool only_deferred)
467 spin_lock(&_minor_lock);
469 if (dm_open_count(md)) {
472 set_bit(DMF_DEFERRED_REMOVE, &md->flags);
473 } else if (only_deferred && !test_bit(DMF_DEFERRED_REMOVE, &md->flags))
476 set_bit(DMF_DELETING, &md->flags);
478 spin_unlock(&_minor_lock);
483 int dm_cancel_deferred_remove(struct mapped_device *md)
487 spin_lock(&_minor_lock);
489 if (test_bit(DMF_DELETING, &md->flags))
492 clear_bit(DMF_DEFERRED_REMOVE, &md->flags);
494 spin_unlock(&_minor_lock);
499 static void do_deferred_remove(struct work_struct *w)
501 dm_deferred_remove();
504 sector_t dm_get_size(struct mapped_device *md)
506 return get_capacity(md->disk);
509 struct request_queue *dm_get_md_queue(struct mapped_device *md)
514 struct dm_stats *dm_get_stats(struct mapped_device *md)
519 static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo)
521 struct mapped_device *md = bdev->bd_disk->private_data;
523 return dm_get_geometry(md, geo);
526 static int dm_blk_ioctl(struct block_device *bdev, fmode_t mode,
527 unsigned int cmd, unsigned long arg)
529 struct mapped_device *md = bdev->bd_disk->private_data;
531 struct dm_table *map;
532 struct dm_target *tgt;
536 map = dm_get_live_table(md, &srcu_idx);
538 if (!map || !dm_table_get_size(map))
541 /* We only support devices that have a single target */
542 if (dm_table_get_num_targets(map) != 1)
545 tgt = dm_table_get_target(map, 0);
546 if (!tgt->type->ioctl)
549 if (dm_suspended_md(md)) {
554 r = tgt->type->ioctl(tgt, cmd, arg);
557 dm_put_live_table(md, srcu_idx);
559 if (r == -ENOTCONN) {
567 static struct dm_io *alloc_io(struct mapped_device *md)
569 return mempool_alloc(md->io_pool, GFP_NOIO);
572 static void free_io(struct mapped_device *md, struct dm_io *io)
574 mempool_free(io, md->io_pool);
577 static void free_tio(struct mapped_device *md, struct dm_target_io *tio)
579 bio_put(&tio->clone);
582 static struct dm_rq_target_io *alloc_rq_tio(struct mapped_device *md,
585 return mempool_alloc(md->io_pool, gfp_mask);
588 static void free_rq_tio(struct dm_rq_target_io *tio)
590 mempool_free(tio, tio->md->io_pool);
593 static struct request *alloc_clone_request(struct mapped_device *md,
596 return mempool_alloc(md->rq_pool, gfp_mask);
599 static void free_clone_request(struct mapped_device *md, struct request *rq)
601 mempool_free(rq, md->rq_pool);
604 static int md_in_flight(struct mapped_device *md)
606 return atomic_read(&md->pending[READ]) +
607 atomic_read(&md->pending[WRITE]);
610 static void start_io_acct(struct dm_io *io)
612 struct mapped_device *md = io->md;
613 struct bio *bio = io->bio;
615 int rw = bio_data_dir(bio);
617 io->start_time = jiffies;
619 cpu = part_stat_lock();
620 part_round_stats(cpu, &dm_disk(md)->part0);
622 atomic_set(&dm_disk(md)->part0.in_flight[rw],
623 atomic_inc_return(&md->pending[rw]));
625 if (unlikely(dm_stats_used(&md->stats)))
626 dm_stats_account_io(&md->stats, bio->bi_rw, bio->bi_iter.bi_sector,
627 bio_sectors(bio), false, 0, &io->stats_aux);
630 static void end_io_acct(struct dm_io *io)
632 struct mapped_device *md = io->md;
633 struct bio *bio = io->bio;
634 unsigned long duration = jiffies - io->start_time;
636 int rw = bio_data_dir(bio);
638 generic_end_io_acct(rw, &dm_disk(md)->part0, io->start_time);
640 if (unlikely(dm_stats_used(&md->stats)))
641 dm_stats_account_io(&md->stats, bio->bi_rw, bio->bi_iter.bi_sector,
642 bio_sectors(bio), true, duration, &io->stats_aux);
645 * After this is decremented the bio must not be touched if it is
648 pending = atomic_dec_return(&md->pending[rw]);
649 atomic_set(&dm_disk(md)->part0.in_flight[rw], pending);
650 pending += atomic_read(&md->pending[rw^0x1]);
652 /* nudge anyone waiting on suspend queue */
658 * Add the bio to the list of deferred io.
660 static void queue_io(struct mapped_device *md, struct bio *bio)
664 spin_lock_irqsave(&md->deferred_lock, flags);
665 bio_list_add(&md->deferred, bio);
666 spin_unlock_irqrestore(&md->deferred_lock, flags);
667 queue_work(md->wq, &md->work);
671 * Everyone (including functions in this file), should use this
672 * function to access the md->map field, and make sure they call
673 * dm_put_live_table() when finished.
675 struct dm_table *dm_get_live_table(struct mapped_device *md, int *srcu_idx) __acquires(md->io_barrier)
677 *srcu_idx = srcu_read_lock(&md->io_barrier);
679 return srcu_dereference(md->map, &md->io_barrier);
682 void dm_put_live_table(struct mapped_device *md, int srcu_idx) __releases(md->io_barrier)
684 srcu_read_unlock(&md->io_barrier, srcu_idx);
687 void dm_sync_table(struct mapped_device *md)
689 synchronize_srcu(&md->io_barrier);
690 synchronize_rcu_expedited();
694 * A fast alternative to dm_get_live_table/dm_put_live_table.
695 * The caller must not block between these two functions.
697 static struct dm_table *dm_get_live_table_fast(struct mapped_device *md) __acquires(RCU)
700 return rcu_dereference(md->map);
703 static void dm_put_live_table_fast(struct mapped_device *md) __releases(RCU)
709 * Open a table device so we can use it as a map destination.
711 static int open_table_device(struct table_device *td, dev_t dev,
712 struct mapped_device *md)
714 static char *_claim_ptr = "I belong to device-mapper";
715 struct block_device *bdev;
719 BUG_ON(td->dm_dev.bdev);
721 bdev = blkdev_get_by_dev(dev, td->dm_dev.mode | FMODE_EXCL, _claim_ptr);
723 return PTR_ERR(bdev);
725 r = bd_link_disk_holder(bdev, dm_disk(md));
727 blkdev_put(bdev, td->dm_dev.mode | FMODE_EXCL);
731 td->dm_dev.bdev = bdev;
736 * Close a table device that we've been using.
738 static void close_table_device(struct table_device *td, struct mapped_device *md)
740 if (!td->dm_dev.bdev)
743 bd_unlink_disk_holder(td->dm_dev.bdev, dm_disk(md));
744 blkdev_put(td->dm_dev.bdev, td->dm_dev.mode | FMODE_EXCL);
745 td->dm_dev.bdev = NULL;
748 static struct table_device *find_table_device(struct list_head *l, dev_t dev,
750 struct table_device *td;
752 list_for_each_entry(td, l, list)
753 if (td->dm_dev.bdev->bd_dev == dev && td->dm_dev.mode == mode)
759 int dm_get_table_device(struct mapped_device *md, dev_t dev, fmode_t mode,
760 struct dm_dev **result) {
762 struct table_device *td;
764 mutex_lock(&md->table_devices_lock);
765 td = find_table_device(&md->table_devices, dev, mode);
767 td = kmalloc(sizeof(*td), GFP_KERNEL);
769 mutex_unlock(&md->table_devices_lock);
773 td->dm_dev.mode = mode;
774 td->dm_dev.bdev = NULL;
776 if ((r = open_table_device(td, dev, md))) {
777 mutex_unlock(&md->table_devices_lock);
782 format_dev_t(td->dm_dev.name, dev);
784 atomic_set(&td->count, 0);
785 list_add(&td->list, &md->table_devices);
787 atomic_inc(&td->count);
788 mutex_unlock(&md->table_devices_lock);
790 *result = &td->dm_dev;
793 EXPORT_SYMBOL_GPL(dm_get_table_device);
795 void dm_put_table_device(struct mapped_device *md, struct dm_dev *d)
797 struct table_device *td = container_of(d, struct table_device, dm_dev);
799 mutex_lock(&md->table_devices_lock);
800 if (atomic_dec_and_test(&td->count)) {
801 close_table_device(td, md);
805 mutex_unlock(&md->table_devices_lock);
807 EXPORT_SYMBOL(dm_put_table_device);
809 static void free_table_devices(struct list_head *devices)
811 struct list_head *tmp, *next;
813 list_for_each_safe(tmp, next, devices) {
814 struct table_device *td = list_entry(tmp, struct table_device, list);
816 DMWARN("dm_destroy: %s still exists with %d references",
817 td->dm_dev.name, atomic_read(&td->count));
823 * Get the geometry associated with a dm device
825 int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo)
833 * Set the geometry of a device.
835 int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo)
837 sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors;
839 if (geo->start > sz) {
840 DMWARN("Start sector is beyond the geometry limits.");
849 /*-----------------------------------------------------------------
851 * A more elegant soln is in the works that uses the queue
852 * merge fn, unfortunately there are a couple of changes to
853 * the block layer that I want to make for this. So in the
854 * interests of getting something for people to use I give
855 * you this clearly demarcated crap.
856 *---------------------------------------------------------------*/
858 static int __noflush_suspending(struct mapped_device *md)
860 return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
864 * Decrements the number of outstanding ios that a bio has been
865 * cloned into, completing the original io if necc.
867 static void dec_pending(struct dm_io *io, int error)
872 struct mapped_device *md = io->md;
874 /* Push-back supersedes any I/O errors */
875 if (unlikely(error)) {
876 spin_lock_irqsave(&io->endio_lock, flags);
877 if (!(io->error > 0 && __noflush_suspending(md)))
879 spin_unlock_irqrestore(&io->endio_lock, flags);
882 if (atomic_dec_and_test(&io->io_count)) {
883 if (io->error == DM_ENDIO_REQUEUE) {
885 * Target requested pushing back the I/O.
887 spin_lock_irqsave(&md->deferred_lock, flags);
888 if (__noflush_suspending(md))
889 bio_list_add_head(&md->deferred, io->bio);
891 /* noflush suspend was interrupted. */
893 spin_unlock_irqrestore(&md->deferred_lock, flags);
896 io_error = io->error;
901 if (io_error == DM_ENDIO_REQUEUE)
904 if ((bio->bi_rw & REQ_FLUSH) && bio->bi_iter.bi_size) {
906 * Preflush done for flush with data, reissue
909 bio->bi_rw &= ~REQ_FLUSH;
912 /* done with normal IO or empty flush */
913 trace_block_bio_complete(md->queue, bio, io_error);
914 bio_endio(bio, io_error);
919 static void disable_write_same(struct mapped_device *md)
921 struct queue_limits *limits = dm_get_queue_limits(md);
923 /* device doesn't really support WRITE SAME, disable it */
924 limits->max_write_same_sectors = 0;
927 static void clone_endio(struct bio *bio, int error)
930 struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone);
931 struct dm_io *io = tio->io;
932 struct mapped_device *md = tio->io->md;
933 dm_endio_fn endio = tio->ti->type->end_io;
935 if (!bio_flagged(bio, BIO_UPTODATE) && !error)
939 r = endio(tio->ti, bio, error);
940 if (r < 0 || r == DM_ENDIO_REQUEUE)
942 * error and requeue request are handled
946 else if (r == DM_ENDIO_INCOMPLETE)
947 /* The target will handle the io */
950 DMWARN("unimplemented target endio return value: %d", r);
955 if (unlikely(r == -EREMOTEIO && (bio->bi_rw & REQ_WRITE_SAME) &&
956 !bdev_get_queue(bio->bi_bdev)->limits.max_write_same_sectors))
957 disable_write_same(md);
960 dec_pending(io, error);
964 * Partial completion handling for request-based dm
966 static void end_clone_bio(struct bio *clone, int error)
968 struct dm_rq_clone_bio_info *info =
969 container_of(clone, struct dm_rq_clone_bio_info, clone);
970 struct dm_rq_target_io *tio = info->tio;
971 struct bio *bio = info->orig;
972 unsigned int nr_bytes = info->orig->bi_iter.bi_size;
978 * An error has already been detected on the request.
979 * Once error occurred, just let clone->end_io() handle
985 * Don't notice the error to the upper layer yet.
986 * The error handling decision is made by the target driver,
987 * when the request is completed.
994 * I/O for the bio successfully completed.
995 * Notice the data completion to the upper layer.
999 * bios are processed from the head of the list.
1000 * So the completing bio should always be rq->bio.
1001 * If it's not, something wrong is happening.
1003 if (tio->orig->bio != bio)
1004 DMERR("bio completion is going in the middle of the request");
1007 * Update the original request.
1008 * Do not use blk_end_request() here, because it may complete
1009 * the original request before the clone, and break the ordering.
1011 blk_update_request(tio->orig, 0, nr_bytes);
1015 * Don't touch any member of the md after calling this function because
1016 * the md may be freed in dm_put() at the end of this function.
1017 * Or do dm_get() before calling this function and dm_put() later.
1019 static void rq_completed(struct mapped_device *md, int rw, bool run_queue)
1021 atomic_dec(&md->pending[rw]);
1023 /* nudge anyone waiting on suspend queue */
1024 if (!md_in_flight(md))
1028 * Run this off this callpath, as drivers could invoke end_io while
1029 * inside their request_fn (and holding the queue lock). Calling
1030 * back into ->request_fn() could deadlock attempting to grab the
1034 blk_run_queue_async(md->queue);
1037 * dm_put() must be at the end of this function. See the comment above
1042 static void free_rq_clone(struct request *clone)
1044 struct dm_rq_target_io *tio = clone->end_io_data;
1046 blk_rq_unprep_clone(clone);
1047 if (clone->q && clone->q->mq_ops)
1048 tio->ti->type->release_clone_rq(clone);
1050 free_clone_request(tio->md, clone);
1055 * Complete the clone and the original request.
1056 * Must be called without clone's queue lock held,
1057 * see end_clone_request() for more details.
1059 static void dm_end_request(struct request *clone, int error)
1061 int rw = rq_data_dir(clone);
1062 struct dm_rq_target_io *tio = clone->end_io_data;
1063 struct mapped_device *md = tio->md;
1064 struct request *rq = tio->orig;
1066 if (rq->cmd_type == REQ_TYPE_BLOCK_PC) {
1067 rq->errors = clone->errors;
1068 rq->resid_len = clone->resid_len;
1072 * We are using the sense buffer of the original
1074 * So setting the length of the sense data is enough.
1076 rq->sense_len = clone->sense_len;
1079 free_rq_clone(clone);
1080 blk_end_request_all(rq, error);
1081 rq_completed(md, rw, true);
1084 static void dm_unprep_request(struct request *rq)
1086 struct dm_rq_target_io *tio = rq->special;
1087 struct request *clone = tio->clone;
1090 rq->cmd_flags &= ~REQ_DONTPREP;
1093 free_rq_clone(clone);
1097 * Requeue the original request of a clone.
1099 static void dm_requeue_unmapped_original_request(struct mapped_device *md,
1102 int rw = rq_data_dir(rq);
1103 struct request_queue *q = rq->q;
1104 unsigned long flags;
1106 dm_unprep_request(rq);
1108 spin_lock_irqsave(q->queue_lock, flags);
1109 blk_requeue_request(q, rq);
1110 spin_unlock_irqrestore(q->queue_lock, flags);
1112 rq_completed(md, rw, false);
1115 static void dm_requeue_unmapped_request(struct request *clone)
1117 struct dm_rq_target_io *tio = clone->end_io_data;
1119 dm_requeue_unmapped_original_request(tio->md, tio->orig);
1122 static void __stop_queue(struct request_queue *q)
1127 static void stop_queue(struct request_queue *q)
1129 unsigned long flags;
1131 spin_lock_irqsave(q->queue_lock, flags);
1133 spin_unlock_irqrestore(q->queue_lock, flags);
1136 static void __start_queue(struct request_queue *q)
1138 if (blk_queue_stopped(q))
1142 static void start_queue(struct request_queue *q)
1144 unsigned long flags;
1146 spin_lock_irqsave(q->queue_lock, flags);
1148 spin_unlock_irqrestore(q->queue_lock, flags);
1151 static void dm_done(struct request *clone, int error, bool mapped)
1154 struct dm_rq_target_io *tio = clone->end_io_data;
1155 dm_request_endio_fn rq_end_io = NULL;
1158 rq_end_io = tio->ti->type->rq_end_io;
1160 if (mapped && rq_end_io)
1161 r = rq_end_io(tio->ti, clone, error, &tio->info);
1164 if (unlikely(r == -EREMOTEIO && (clone->cmd_flags & REQ_WRITE_SAME) &&
1165 !clone->q->limits.max_write_same_sectors))
1166 disable_write_same(tio->md);
1169 /* The target wants to complete the I/O */
1170 dm_end_request(clone, r);
1171 else if (r == DM_ENDIO_INCOMPLETE)
1172 /* The target will handle the I/O */
1174 else if (r == DM_ENDIO_REQUEUE)
1175 /* The target wants to requeue the I/O */
1176 dm_requeue_unmapped_request(clone);
1178 DMWARN("unimplemented target endio return value: %d", r);
1184 * Request completion handler for request-based dm
1186 static void dm_softirq_done(struct request *rq)
1189 struct dm_rq_target_io *tio = rq->special;
1190 struct request *clone = tio->clone;
1193 blk_end_request_all(rq, tio->error);
1194 rq_completed(tio->md, rq_data_dir(rq), false);
1199 if (rq->cmd_flags & REQ_FAILED)
1202 dm_done(clone, tio->error, mapped);
1206 * Complete the clone and the original request with the error status
1207 * through softirq context.
1209 static void dm_complete_request(struct request *rq, int error)
1211 struct dm_rq_target_io *tio = rq->special;
1214 blk_complete_request(rq);
1218 * Complete the not-mapped clone and the original request with the error status
1219 * through softirq context.
1220 * Target's rq_end_io() function isn't called.
1221 * This may be used when the target's map_rq() or clone_and_map_rq() functions fail.
1223 static void dm_kill_unmapped_request(struct request *rq, int error)
1225 rq->cmd_flags |= REQ_FAILED;
1226 dm_complete_request(rq, error);
1230 * Called with the clone's queue lock held
1232 static void end_clone_request(struct request *clone, int error)
1234 struct dm_rq_target_io *tio = clone->end_io_data;
1236 if (!clone->q->mq_ops) {
1238 * For just cleaning up the information of the queue in which
1239 * the clone was dispatched.
1240 * The clone is *NOT* freed actually here because it is alloced
1241 * from dm own mempool (REQ_ALLOCED isn't set).
1243 __blk_put_request(clone->q, clone);
1247 * Actual request completion is done in a softirq context which doesn't
1248 * hold the clone's queue lock. Otherwise, deadlock could occur because:
1249 * - another request may be submitted by the upper level driver
1250 * of the stacking during the completion
1251 * - the submission which requires queue lock may be done
1252 * against this clone's queue
1254 dm_complete_request(tio->orig, error);
1258 * Return maximum size of I/O possible at the supplied sector up to the current
1261 static sector_t max_io_len_target_boundary(sector_t sector, struct dm_target *ti)
1263 sector_t target_offset = dm_target_offset(ti, sector);
1265 return ti->len - target_offset;
1268 static sector_t max_io_len(sector_t sector, struct dm_target *ti)
1270 sector_t len = max_io_len_target_boundary(sector, ti);
1271 sector_t offset, max_len;
1274 * Does the target need to split even further?
1276 if (ti->max_io_len) {
1277 offset = dm_target_offset(ti, sector);
1278 if (unlikely(ti->max_io_len & (ti->max_io_len - 1)))
1279 max_len = sector_div(offset, ti->max_io_len);
1281 max_len = offset & (ti->max_io_len - 1);
1282 max_len = ti->max_io_len - max_len;
1291 int dm_set_target_max_io_len(struct dm_target *ti, sector_t len)
1293 if (len > UINT_MAX) {
1294 DMERR("Specified maximum size of target IO (%llu) exceeds limit (%u)",
1295 (unsigned long long)len, UINT_MAX);
1296 ti->error = "Maximum size of target IO is too large";
1300 ti->max_io_len = (uint32_t) len;
1304 EXPORT_SYMBOL_GPL(dm_set_target_max_io_len);
1307 * A target may call dm_accept_partial_bio only from the map routine. It is
1308 * allowed for all bio types except REQ_FLUSH.
1310 * dm_accept_partial_bio informs the dm that the target only wants to process
1311 * additional n_sectors sectors of the bio and the rest of the data should be
1312 * sent in a next bio.
1314 * A diagram that explains the arithmetics:
1315 * +--------------------+---------------+-------+
1317 * +--------------------+---------------+-------+
1319 * <-------------- *tio->len_ptr --------------->
1320 * <------- bi_size ------->
1323 * Region 1 was already iterated over with bio_advance or similar function.
1324 * (it may be empty if the target doesn't use bio_advance)
1325 * Region 2 is the remaining bio size that the target wants to process.
1326 * (it may be empty if region 1 is non-empty, although there is no reason
1328 * The target requires that region 3 is to be sent in the next bio.
1330 * If the target wants to receive multiple copies of the bio (via num_*bios, etc),
1331 * the partially processed part (the sum of regions 1+2) must be the same for all
1332 * copies of the bio.
1334 void dm_accept_partial_bio(struct bio *bio, unsigned n_sectors)
1336 struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone);
1337 unsigned bi_size = bio->bi_iter.bi_size >> SECTOR_SHIFT;
1338 BUG_ON(bio->bi_rw & REQ_FLUSH);
1339 BUG_ON(bi_size > *tio->len_ptr);
1340 BUG_ON(n_sectors > bi_size);
1341 *tio->len_ptr -= bi_size - n_sectors;
1342 bio->bi_iter.bi_size = n_sectors << SECTOR_SHIFT;
1344 EXPORT_SYMBOL_GPL(dm_accept_partial_bio);
1346 static void __map_bio(struct dm_target_io *tio)
1350 struct mapped_device *md;
1351 struct bio *clone = &tio->clone;
1352 struct dm_target *ti = tio->ti;
1354 clone->bi_end_io = clone_endio;
1357 * Map the clone. If r == 0 we don't need to do
1358 * anything, the target has assumed ownership of
1361 atomic_inc(&tio->io->io_count);
1362 sector = clone->bi_iter.bi_sector;
1363 r = ti->type->map(ti, clone);
1364 if (r == DM_MAPIO_REMAPPED) {
1365 /* the bio has been remapped so dispatch it */
1367 trace_block_bio_remap(bdev_get_queue(clone->bi_bdev), clone,
1368 tio->io->bio->bi_bdev->bd_dev, sector);
1370 generic_make_request(clone);
1371 } else if (r < 0 || r == DM_MAPIO_REQUEUE) {
1372 /* error the io and bail out, or requeue it if needed */
1374 dec_pending(tio->io, r);
1377 DMWARN("unimplemented target map return value: %d", r);
1383 struct mapped_device *md;
1384 struct dm_table *map;
1388 unsigned sector_count;
1391 static void bio_setup_sector(struct bio *bio, sector_t sector, unsigned len)
1393 bio->bi_iter.bi_sector = sector;
1394 bio->bi_iter.bi_size = to_bytes(len);
1398 * Creates a bio that consists of range of complete bvecs.
1400 static void clone_bio(struct dm_target_io *tio, struct bio *bio,
1401 sector_t sector, unsigned len)
1403 struct bio *clone = &tio->clone;
1405 __bio_clone_fast(clone, bio);
1407 if (bio_integrity(bio))
1408 bio_integrity_clone(clone, bio, GFP_NOIO);
1410 bio_advance(clone, to_bytes(sector - clone->bi_iter.bi_sector));
1411 clone->bi_iter.bi_size = to_bytes(len);
1413 if (bio_integrity(bio))
1414 bio_integrity_trim(clone, 0, len);
1417 static struct dm_target_io *alloc_tio(struct clone_info *ci,
1418 struct dm_target *ti,
1419 unsigned target_bio_nr)
1421 struct dm_target_io *tio;
1424 clone = bio_alloc_bioset(GFP_NOIO, 0, ci->md->bs);
1425 tio = container_of(clone, struct dm_target_io, clone);
1429 tio->target_bio_nr = target_bio_nr;
1434 static void __clone_and_map_simple_bio(struct clone_info *ci,
1435 struct dm_target *ti,
1436 unsigned target_bio_nr, unsigned *len)
1438 struct dm_target_io *tio = alloc_tio(ci, ti, target_bio_nr);
1439 struct bio *clone = &tio->clone;
1443 __bio_clone_fast(clone, ci->bio);
1445 bio_setup_sector(clone, ci->sector, *len);
1450 static void __send_duplicate_bios(struct clone_info *ci, struct dm_target *ti,
1451 unsigned num_bios, unsigned *len)
1453 unsigned target_bio_nr;
1455 for (target_bio_nr = 0; target_bio_nr < num_bios; target_bio_nr++)
1456 __clone_and_map_simple_bio(ci, ti, target_bio_nr, len);
1459 static int __send_empty_flush(struct clone_info *ci)
1461 unsigned target_nr = 0;
1462 struct dm_target *ti;
1464 BUG_ON(bio_has_data(ci->bio));
1465 while ((ti = dm_table_get_target(ci->map, target_nr++)))
1466 __send_duplicate_bios(ci, ti, ti->num_flush_bios, NULL);
1471 static void __clone_and_map_data_bio(struct clone_info *ci, struct dm_target *ti,
1472 sector_t sector, unsigned *len)
1474 struct bio *bio = ci->bio;
1475 struct dm_target_io *tio;
1476 unsigned target_bio_nr;
1477 unsigned num_target_bios = 1;
1480 * Does the target want to receive duplicate copies of the bio?
1482 if (bio_data_dir(bio) == WRITE && ti->num_write_bios)
1483 num_target_bios = ti->num_write_bios(ti, bio);
1485 for (target_bio_nr = 0; target_bio_nr < num_target_bios; target_bio_nr++) {
1486 tio = alloc_tio(ci, ti, target_bio_nr);
1488 clone_bio(tio, bio, sector, *len);
1493 typedef unsigned (*get_num_bios_fn)(struct dm_target *ti);
1495 static unsigned get_num_discard_bios(struct dm_target *ti)
1497 return ti->num_discard_bios;
1500 static unsigned get_num_write_same_bios(struct dm_target *ti)
1502 return ti->num_write_same_bios;
1505 typedef bool (*is_split_required_fn)(struct dm_target *ti);
1507 static bool is_split_required_for_discard(struct dm_target *ti)
1509 return ti->split_discard_bios;
1512 static int __send_changing_extent_only(struct clone_info *ci,
1513 get_num_bios_fn get_num_bios,
1514 is_split_required_fn is_split_required)
1516 struct dm_target *ti;
1521 ti = dm_table_find_target(ci->map, ci->sector);
1522 if (!dm_target_is_valid(ti))
1526 * Even though the device advertised support for this type of
1527 * request, that does not mean every target supports it, and
1528 * reconfiguration might also have changed that since the
1529 * check was performed.
1531 num_bios = get_num_bios ? get_num_bios(ti) : 0;
1535 if (is_split_required && !is_split_required(ti))
1536 len = min((sector_t)ci->sector_count, max_io_len_target_boundary(ci->sector, ti));
1538 len = min((sector_t)ci->sector_count, max_io_len(ci->sector, ti));
1540 __send_duplicate_bios(ci, ti, num_bios, &len);
1543 } while (ci->sector_count -= len);
1548 static int __send_discard(struct clone_info *ci)
1550 return __send_changing_extent_only(ci, get_num_discard_bios,
1551 is_split_required_for_discard);
1554 static int __send_write_same(struct clone_info *ci)
1556 return __send_changing_extent_only(ci, get_num_write_same_bios, NULL);
1560 * Select the correct strategy for processing a non-flush bio.
1562 static int __split_and_process_non_flush(struct clone_info *ci)
1564 struct bio *bio = ci->bio;
1565 struct dm_target *ti;
1568 if (unlikely(bio->bi_rw & REQ_DISCARD))
1569 return __send_discard(ci);
1570 else if (unlikely(bio->bi_rw & REQ_WRITE_SAME))
1571 return __send_write_same(ci);
1573 ti = dm_table_find_target(ci->map, ci->sector);
1574 if (!dm_target_is_valid(ti))
1577 len = min_t(sector_t, max_io_len(ci->sector, ti), ci->sector_count);
1579 __clone_and_map_data_bio(ci, ti, ci->sector, &len);
1582 ci->sector_count -= len;
1588 * Entry point to split a bio into clones and submit them to the targets.
1590 static void __split_and_process_bio(struct mapped_device *md,
1591 struct dm_table *map, struct bio *bio)
1593 struct clone_info ci;
1596 if (unlikely(!map)) {
1603 ci.io = alloc_io(md);
1605 atomic_set(&ci.io->io_count, 1);
1608 spin_lock_init(&ci.io->endio_lock);
1609 ci.sector = bio->bi_iter.bi_sector;
1611 start_io_acct(ci.io);
1613 if (bio->bi_rw & REQ_FLUSH) {
1614 ci.bio = &ci.md->flush_bio;
1615 ci.sector_count = 0;
1616 error = __send_empty_flush(&ci);
1617 /* dec_pending submits any data associated with flush */
1620 ci.sector_count = bio_sectors(bio);
1621 while (ci.sector_count && !error)
1622 error = __split_and_process_non_flush(&ci);
1625 /* drop the extra reference count */
1626 dec_pending(ci.io, error);
1628 /*-----------------------------------------------------------------
1630 *---------------------------------------------------------------*/
1632 static int dm_merge_bvec(struct request_queue *q,
1633 struct bvec_merge_data *bvm,
1634 struct bio_vec *biovec)
1636 struct mapped_device *md = q->queuedata;
1637 struct dm_table *map = dm_get_live_table_fast(md);
1638 struct dm_target *ti;
1639 sector_t max_sectors;
1645 ti = dm_table_find_target(map, bvm->bi_sector);
1646 if (!dm_target_is_valid(ti))
1650 * Find maximum amount of I/O that won't need splitting
1652 max_sectors = min(max_io_len(bvm->bi_sector, ti),
1653 (sector_t) queue_max_sectors(q));
1654 max_size = (max_sectors << SECTOR_SHIFT) - bvm->bi_size;
1655 if (unlikely(max_size < 0)) /* this shouldn't _ever_ happen */
1659 * merge_bvec_fn() returns number of bytes
1660 * it can accept at this offset
1661 * max is precomputed maximal io size
1663 if (max_size && ti->type->merge)
1664 max_size = ti->type->merge(ti, bvm, biovec, max_size);
1666 * If the target doesn't support merge method and some of the devices
1667 * provided their merge_bvec method (we know this by looking for the
1668 * max_hw_sectors that dm_set_device_limits may set), then we can't
1669 * allow bios with multiple vector entries. So always set max_size
1670 * to 0, and the code below allows just one page.
1672 else if (queue_max_hw_sectors(q) <= PAGE_SIZE >> 9)
1676 dm_put_live_table_fast(md);
1678 * Always allow an entire first page
1680 if (max_size <= biovec->bv_len && !(bvm->bi_size >> SECTOR_SHIFT))
1681 max_size = biovec->bv_len;
1687 * The request function that just remaps the bio built up by
1690 static void _dm_request(struct request_queue *q, struct bio *bio)
1692 int rw = bio_data_dir(bio);
1693 struct mapped_device *md = q->queuedata;
1695 struct dm_table *map;
1697 map = dm_get_live_table(md, &srcu_idx);
1699 generic_start_io_acct(rw, bio_sectors(bio), &dm_disk(md)->part0);
1701 /* if we're suspended, we have to queue this io for later */
1702 if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))) {
1703 dm_put_live_table(md, srcu_idx);
1705 if (bio_rw(bio) != READA)
1712 __split_and_process_bio(md, map, bio);
1713 dm_put_live_table(md, srcu_idx);
1717 int dm_request_based(struct mapped_device *md)
1719 return blk_queue_stackable(md->queue);
1722 static void dm_request(struct request_queue *q, struct bio *bio)
1724 struct mapped_device *md = q->queuedata;
1726 if (dm_request_based(md))
1727 blk_queue_bio(q, bio);
1729 _dm_request(q, bio);
1732 static void dm_dispatch_clone_request(struct request *clone, struct request *rq)
1736 if (blk_queue_io_stat(clone->q))
1737 clone->cmd_flags |= REQ_IO_STAT;
1739 clone->start_time = jiffies;
1740 r = blk_insert_cloned_request(clone->q, clone);
1742 /* must complete clone in terms of original request */
1743 dm_complete_request(rq, r);
1746 static int dm_rq_bio_constructor(struct bio *bio, struct bio *bio_orig,
1749 struct dm_rq_target_io *tio = data;
1750 struct dm_rq_clone_bio_info *info =
1751 container_of(bio, struct dm_rq_clone_bio_info, clone);
1753 info->orig = bio_orig;
1755 bio->bi_end_io = end_clone_bio;
1760 static int setup_clone(struct request *clone, struct request *rq,
1761 struct dm_rq_target_io *tio, gfp_t gfp_mask)
1765 r = blk_rq_prep_clone(clone, rq, tio->md->bs, gfp_mask,
1766 dm_rq_bio_constructor, tio);
1770 clone->cmd = rq->cmd;
1771 clone->cmd_len = rq->cmd_len;
1772 clone->sense = rq->sense;
1773 clone->end_io = end_clone_request;
1774 clone->end_io_data = tio;
1781 static struct request *clone_rq(struct request *rq, struct mapped_device *md,
1782 struct dm_rq_target_io *tio, gfp_t gfp_mask)
1784 struct request *clone = alloc_clone_request(md, gfp_mask);
1789 blk_rq_init(NULL, clone);
1790 if (setup_clone(clone, rq, tio, gfp_mask)) {
1792 free_clone_request(md, clone);
1799 static void map_tio_request(struct kthread_work *work);
1801 static struct dm_rq_target_io *prep_tio(struct request *rq,
1802 struct mapped_device *md, gfp_t gfp_mask)
1804 struct dm_rq_target_io *tio;
1806 struct dm_table *table;
1808 tio = alloc_rq_tio(md, gfp_mask);
1817 memset(&tio->info, 0, sizeof(tio->info));
1818 init_kthread_work(&tio->work, map_tio_request);
1820 table = dm_get_live_table(md, &srcu_idx);
1821 if (!dm_table_mq_request_based(table)) {
1822 if (!clone_rq(rq, md, tio, gfp_mask)) {
1823 dm_put_live_table(md, srcu_idx);
1828 dm_put_live_table(md, srcu_idx);
1834 * Called with the queue lock held.
1836 static int dm_prep_fn(struct request_queue *q, struct request *rq)
1838 struct mapped_device *md = q->queuedata;
1839 struct dm_rq_target_io *tio;
1841 if (unlikely(rq->special)) {
1842 DMWARN("Already has something in rq->special.");
1843 return BLKPREP_KILL;
1846 tio = prep_tio(rq, md, GFP_ATOMIC);
1848 return BLKPREP_DEFER;
1851 rq->cmd_flags |= REQ_DONTPREP;
1858 * 0 : the request has been processed
1859 * DM_MAPIO_REQUEUE : the original request needs to be requeued
1860 * < 0 : the request was completed due to failure
1862 static int map_request(struct dm_target *ti, struct request *rq,
1863 struct mapped_device *md)
1866 struct dm_rq_target_io *tio = rq->special;
1867 struct request *clone = NULL;
1871 r = ti->type->map_rq(ti, clone, &tio->info);
1873 r = ti->type->clone_and_map_rq(ti, rq, &tio->info, &clone);
1875 /* The target wants to complete the I/O */
1876 dm_kill_unmapped_request(rq, r);
1880 return DM_MAPIO_REQUEUE;
1881 if (setup_clone(clone, rq, tio, GFP_KERNEL)) {
1883 ti->type->release_clone_rq(clone);
1884 return DM_MAPIO_REQUEUE;
1889 case DM_MAPIO_SUBMITTED:
1890 /* The target has taken the I/O to submit by itself later */
1892 case DM_MAPIO_REMAPPED:
1893 /* The target has remapped the I/O so dispatch it */
1894 trace_block_rq_remap(clone->q, clone, disk_devt(dm_disk(md)),
1896 dm_dispatch_clone_request(clone, rq);
1898 case DM_MAPIO_REQUEUE:
1899 /* The target wants to requeue the I/O */
1900 dm_requeue_unmapped_request(clone);
1904 DMWARN("unimplemented target map return value: %d", r);
1908 /* The target wants to complete the I/O */
1909 dm_kill_unmapped_request(rq, r);
1916 static void map_tio_request(struct kthread_work *work)
1918 struct dm_rq_target_io *tio = container_of(work, struct dm_rq_target_io, work);
1919 struct request *rq = tio->orig;
1920 struct mapped_device *md = tio->md;
1922 if (map_request(tio->ti, rq, md) == DM_MAPIO_REQUEUE)
1923 dm_requeue_unmapped_original_request(md, rq);
1926 static void dm_start_request(struct mapped_device *md, struct request *orig)
1928 blk_start_request(orig);
1929 atomic_inc(&md->pending[rq_data_dir(orig)]);
1932 * Hold the md reference here for the in-flight I/O.
1933 * We can't rely on the reference count by device opener,
1934 * because the device may be closed during the request completion
1935 * when all bios are completed.
1936 * See the comment in rq_completed() too.
1942 * q->request_fn for request-based dm.
1943 * Called with the queue lock held.
1945 static void dm_request_fn(struct request_queue *q)
1947 struct mapped_device *md = q->queuedata;
1949 struct dm_table *map = dm_get_live_table(md, &srcu_idx);
1950 struct dm_target *ti;
1952 struct dm_rq_target_io *tio;
1956 * For suspend, check blk_queue_stopped() and increment
1957 * ->pending within a single queue_lock not to increment the
1958 * number of in-flight I/Os after the queue is stopped in
1961 while (!blk_queue_stopped(q)) {
1962 rq = blk_peek_request(q);
1966 /* always use block 0 to find the target for flushes for now */
1968 if (!(rq->cmd_flags & REQ_FLUSH))
1969 pos = blk_rq_pos(rq);
1971 ti = dm_table_find_target(map, pos);
1972 if (!dm_target_is_valid(ti)) {
1974 * Must perform setup, that rq_completed() requires,
1975 * before calling dm_kill_unmapped_request
1977 DMERR_LIMIT("request attempted access beyond the end of device");
1978 dm_start_request(md, rq);
1979 dm_kill_unmapped_request(rq, -EIO);
1983 if (ti->type->busy && ti->type->busy(ti))
1986 dm_start_request(md, rq);
1989 /* Establish tio->ti before queuing work (map_tio_request) */
1991 queue_kthread_work(&md->kworker, &tio->work);
1992 BUG_ON(!irqs_disabled());
1998 blk_delay_queue(q, HZ / 10);
2000 dm_put_live_table(md, srcu_idx);
2003 int dm_underlying_device_busy(struct request_queue *q)
2005 return blk_lld_busy(q);
2007 EXPORT_SYMBOL_GPL(dm_underlying_device_busy);
2009 static int dm_lld_busy(struct request_queue *q)
2012 struct mapped_device *md = q->queuedata;
2013 struct dm_table *map = dm_get_live_table_fast(md);
2015 if (!map || test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))
2018 r = dm_table_any_busy_target(map);
2020 dm_put_live_table_fast(md);
2025 static int dm_any_congested(void *congested_data, int bdi_bits)
2028 struct mapped_device *md = congested_data;
2029 struct dm_table *map;
2031 if (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
2032 map = dm_get_live_table_fast(md);
2035 * Request-based dm cares about only own queue for
2036 * the query about congestion status of request_queue
2038 if (dm_request_based(md))
2039 r = md->queue->backing_dev_info.state &
2042 r = dm_table_any_congested(map, bdi_bits);
2044 dm_put_live_table_fast(md);
2050 /*-----------------------------------------------------------------
2051 * An IDR is used to keep track of allocated minor numbers.
2052 *---------------------------------------------------------------*/
2053 static void free_minor(int minor)
2055 spin_lock(&_minor_lock);
2056 idr_remove(&_minor_idr, minor);
2057 spin_unlock(&_minor_lock);
2061 * See if the device with a specific minor # is free.
2063 static int specific_minor(int minor)
2067 if (minor >= (1 << MINORBITS))
2070 idr_preload(GFP_KERNEL);
2071 spin_lock(&_minor_lock);
2073 r = idr_alloc(&_minor_idr, MINOR_ALLOCED, minor, minor + 1, GFP_NOWAIT);
2075 spin_unlock(&_minor_lock);
2078 return r == -ENOSPC ? -EBUSY : r;
2082 static int next_free_minor(int *minor)
2086 idr_preload(GFP_KERNEL);
2087 spin_lock(&_minor_lock);
2089 r = idr_alloc(&_minor_idr, MINOR_ALLOCED, 0, 1 << MINORBITS, GFP_NOWAIT);
2091 spin_unlock(&_minor_lock);
2099 static const struct block_device_operations dm_blk_dops;
2101 static void dm_wq_work(struct work_struct *work);
2103 static void dm_init_md_queue(struct mapped_device *md)
2106 * Request-based dm devices cannot be stacked on top of bio-based dm
2107 * devices. The type of this dm device has not been decided yet.
2108 * The type is decided at the first table loading time.
2109 * To prevent problematic device stacking, clear the queue flag
2110 * for request stacking support until then.
2112 * This queue is new, so no concurrency on the queue_flags.
2114 queue_flag_clear_unlocked(QUEUE_FLAG_STACKABLE, md->queue);
2116 md->queue->queuedata = md;
2117 md->queue->backing_dev_info.congested_fn = dm_any_congested;
2118 md->queue->backing_dev_info.congested_data = md;
2119 blk_queue_make_request(md->queue, dm_request);
2120 blk_queue_bounce_limit(md->queue, BLK_BOUNCE_ANY);
2121 blk_queue_merge_bvec(md->queue, dm_merge_bvec);
2125 * Allocate and initialise a blank device with a given minor.
2127 static struct mapped_device *alloc_dev(int minor)
2130 struct mapped_device *md = kzalloc(sizeof(*md), GFP_KERNEL);
2134 DMWARN("unable to allocate device, out of memory.");
2138 if (!try_module_get(THIS_MODULE))
2139 goto bad_module_get;
2141 /* get a minor number for the dev */
2142 if (minor == DM_ANY_MINOR)
2143 r = next_free_minor(&minor);
2145 r = specific_minor(minor);
2149 r = init_srcu_struct(&md->io_barrier);
2151 goto bad_io_barrier;
2153 md->type = DM_TYPE_NONE;
2154 mutex_init(&md->suspend_lock);
2155 mutex_init(&md->type_lock);
2156 mutex_init(&md->table_devices_lock);
2157 spin_lock_init(&md->deferred_lock);
2158 atomic_set(&md->holders, 1);
2159 atomic_set(&md->open_count, 0);
2160 atomic_set(&md->event_nr, 0);
2161 atomic_set(&md->uevent_seq, 0);
2162 INIT_LIST_HEAD(&md->uevent_list);
2163 INIT_LIST_HEAD(&md->table_devices);
2164 spin_lock_init(&md->uevent_lock);
2166 md->queue = blk_alloc_queue(GFP_KERNEL);
2170 dm_init_md_queue(md);
2172 md->disk = alloc_disk(1);
2176 atomic_set(&md->pending[0], 0);
2177 atomic_set(&md->pending[1], 0);
2178 init_waitqueue_head(&md->wait);
2179 INIT_WORK(&md->work, dm_wq_work);
2180 init_waitqueue_head(&md->eventq);
2181 init_completion(&md->kobj_holder.completion);
2182 md->kworker_task = NULL;
2184 md->disk->major = _major;
2185 md->disk->first_minor = minor;
2186 md->disk->fops = &dm_blk_dops;
2187 md->disk->queue = md->queue;
2188 md->disk->private_data = md;
2189 sprintf(md->disk->disk_name, "dm-%d", minor);
2191 format_dev_t(md->name, MKDEV(_major, minor));
2193 md->wq = alloc_workqueue("kdmflush", WQ_MEM_RECLAIM, 0);
2197 md->bdev = bdget_disk(md->disk, 0);
2201 bio_init(&md->flush_bio);
2202 md->flush_bio.bi_bdev = md->bdev;
2203 md->flush_bio.bi_rw = WRITE_FLUSH;
2205 dm_stats_init(&md->stats);
2207 /* Populate the mapping, nobody knows we exist yet */
2208 spin_lock(&_minor_lock);
2209 old_md = idr_replace(&_minor_idr, md, minor);
2210 spin_unlock(&_minor_lock);
2212 BUG_ON(old_md != MINOR_ALLOCED);
2217 destroy_workqueue(md->wq);
2219 del_gendisk(md->disk);
2222 blk_cleanup_queue(md->queue);
2224 cleanup_srcu_struct(&md->io_barrier);
2228 module_put(THIS_MODULE);
2234 static void unlock_fs(struct mapped_device *md);
2236 static void free_dev(struct mapped_device *md)
2238 int minor = MINOR(disk_devt(md->disk));
2242 destroy_workqueue(md->wq);
2244 if (md->kworker_task)
2245 kthread_stop(md->kworker_task);
2247 mempool_destroy(md->io_pool);
2249 mempool_destroy(md->rq_pool);
2251 bioset_free(md->bs);
2252 blk_integrity_unregister(md->disk);
2253 del_gendisk(md->disk);
2254 cleanup_srcu_struct(&md->io_barrier);
2255 free_table_devices(&md->table_devices);
2258 spin_lock(&_minor_lock);
2259 md->disk->private_data = NULL;
2260 spin_unlock(&_minor_lock);
2263 blk_cleanup_queue(md->queue);
2264 dm_stats_cleanup(&md->stats);
2265 module_put(THIS_MODULE);
2269 static void __bind_mempools(struct mapped_device *md, struct dm_table *t)
2271 struct dm_md_mempools *p = dm_table_get_md_mempools(t);
2273 if (md->io_pool && md->bs) {
2274 /* The md already has necessary mempools. */
2275 if (dm_table_get_type(t) == DM_TYPE_BIO_BASED) {
2277 * Reload bioset because front_pad may have changed
2278 * because a different table was loaded.
2280 bioset_free(md->bs);
2285 * There's no need to reload with request-based dm
2286 * because the size of front_pad doesn't change.
2287 * Note for future: If you are to reload bioset,
2288 * prep-ed requests in the queue may refer
2289 * to bio from the old bioset, so you must walk
2290 * through the queue to unprep.
2295 BUG_ON(!p || md->io_pool || md->rq_pool || md->bs);
2297 md->io_pool = p->io_pool;
2299 md->rq_pool = p->rq_pool;
2305 /* mempool bind completed, now no need any mempools in the table */
2306 dm_table_free_md_mempools(t);
2310 * Bind a table to the device.
2312 static void event_callback(void *context)
2314 unsigned long flags;
2316 struct mapped_device *md = (struct mapped_device *) context;
2318 spin_lock_irqsave(&md->uevent_lock, flags);
2319 list_splice_init(&md->uevent_list, &uevents);
2320 spin_unlock_irqrestore(&md->uevent_lock, flags);
2322 dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj);
2324 atomic_inc(&md->event_nr);
2325 wake_up(&md->eventq);
2329 * Protected by md->suspend_lock obtained by dm_swap_table().
2331 static void __set_size(struct mapped_device *md, sector_t size)
2333 set_capacity(md->disk, size);
2335 i_size_write(md->bdev->bd_inode, (loff_t)size << SECTOR_SHIFT);
2339 * Return 1 if the queue has a compulsory merge_bvec_fn function.
2341 * If this function returns 0, then the device is either a non-dm
2342 * device without a merge_bvec_fn, or it is a dm device that is
2343 * able to split any bios it receives that are too big.
2345 int dm_queue_merge_is_compulsory(struct request_queue *q)
2347 struct mapped_device *dev_md;
2349 if (!q->merge_bvec_fn)
2352 if (q->make_request_fn == dm_request) {
2353 dev_md = q->queuedata;
2354 if (test_bit(DMF_MERGE_IS_OPTIONAL, &dev_md->flags))
2361 static int dm_device_merge_is_compulsory(struct dm_target *ti,
2362 struct dm_dev *dev, sector_t start,
2363 sector_t len, void *data)
2365 struct block_device *bdev = dev->bdev;
2366 struct request_queue *q = bdev_get_queue(bdev);
2368 return dm_queue_merge_is_compulsory(q);
2372 * Return 1 if it is acceptable to ignore merge_bvec_fn based
2373 * on the properties of the underlying devices.
2375 static int dm_table_merge_is_optional(struct dm_table *table)
2378 struct dm_target *ti;
2380 while (i < dm_table_get_num_targets(table)) {
2381 ti = dm_table_get_target(table, i++);
2383 if (ti->type->iterate_devices &&
2384 ti->type->iterate_devices(ti, dm_device_merge_is_compulsory, NULL))
2392 * Returns old map, which caller must destroy.
2394 static struct dm_table *__bind(struct mapped_device *md, struct dm_table *t,
2395 struct queue_limits *limits)
2397 struct dm_table *old_map;
2398 struct request_queue *q = md->queue;
2400 int merge_is_optional;
2402 size = dm_table_get_size(t);
2405 * Wipe any geometry if the size of the table changed.
2407 if (size != dm_get_size(md))
2408 memset(&md->geometry, 0, sizeof(md->geometry));
2410 __set_size(md, size);
2412 dm_table_event_callback(t, event_callback, md);
2415 * The queue hasn't been stopped yet, if the old table type wasn't
2416 * for request-based during suspension. So stop it to prevent
2417 * I/O mapping before resume.
2418 * This must be done before setting the queue restrictions,
2419 * because request-based dm may be run just after the setting.
2421 if (dm_table_request_based(t) && !blk_queue_stopped(q))
2424 __bind_mempools(md, t);
2426 merge_is_optional = dm_table_merge_is_optional(t);
2428 old_map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2429 rcu_assign_pointer(md->map, t);
2430 md->immutable_target_type = dm_table_get_immutable_target_type(t);
2432 dm_table_set_restrictions(t, q, limits);
2433 if (merge_is_optional)
2434 set_bit(DMF_MERGE_IS_OPTIONAL, &md->flags);
2436 clear_bit(DMF_MERGE_IS_OPTIONAL, &md->flags);
2444 * Returns unbound table for the caller to free.
2446 static struct dm_table *__unbind(struct mapped_device *md)
2448 struct dm_table *map = rcu_dereference_protected(md->map, 1);
2453 dm_table_event_callback(map, NULL, NULL);
2454 RCU_INIT_POINTER(md->map, NULL);
2461 * Constructor for a new device.
2463 int dm_create(int minor, struct mapped_device **result)
2465 struct mapped_device *md;
2467 md = alloc_dev(minor);
2478 * Functions to manage md->type.
2479 * All are required to hold md->type_lock.
2481 void dm_lock_md_type(struct mapped_device *md)
2483 mutex_lock(&md->type_lock);
2486 void dm_unlock_md_type(struct mapped_device *md)
2488 mutex_unlock(&md->type_lock);
2491 void dm_set_md_type(struct mapped_device *md, unsigned type)
2493 BUG_ON(!mutex_is_locked(&md->type_lock));
2497 unsigned dm_get_md_type(struct mapped_device *md)
2499 BUG_ON(!mutex_is_locked(&md->type_lock));
2503 static bool dm_md_type_request_based(struct mapped_device *md)
2505 unsigned table_type = dm_get_md_type(md);
2507 return (table_type == DM_TYPE_REQUEST_BASED ||
2508 table_type == DM_TYPE_MQ_REQUEST_BASED);
2511 struct target_type *dm_get_immutable_target_type(struct mapped_device *md)
2513 return md->immutable_target_type;
2517 * The queue_limits are only valid as long as you have a reference
2520 struct queue_limits *dm_get_queue_limits(struct mapped_device *md)
2522 BUG_ON(!atomic_read(&md->holders));
2523 return &md->queue->limits;
2525 EXPORT_SYMBOL_GPL(dm_get_queue_limits);
2528 * Fully initialize a request-based queue (->elevator, ->request_fn, etc).
2530 static int dm_init_request_based_queue(struct mapped_device *md)
2532 struct request_queue *q = NULL;
2534 if (md->queue->elevator)
2537 /* Fully initialize the queue */
2538 q = blk_init_allocated_queue(md->queue, dm_request_fn, NULL);
2543 dm_init_md_queue(md);
2544 blk_queue_softirq_done(md->queue, dm_softirq_done);
2545 blk_queue_prep_rq(md->queue, dm_prep_fn);
2546 blk_queue_lld_busy(md->queue, dm_lld_busy);
2548 /* Also initialize the request-based DM worker thread */
2549 init_kthread_worker(&md->kworker);
2550 md->kworker_task = kthread_run(kthread_worker_fn, &md->kworker,
2551 "kdmwork-%s", dm_device_name(md));
2553 elv_register_queue(md->queue);
2559 * Setup the DM device's queue based on md's type
2561 int dm_setup_md_queue(struct mapped_device *md)
2563 if (dm_md_type_request_based(md) && !dm_init_request_based_queue(md)) {
2564 DMWARN("Cannot initialize queue for request-based mapped device");
2571 static struct mapped_device *dm_find_md(dev_t dev)
2573 struct mapped_device *md;
2574 unsigned minor = MINOR(dev);
2576 if (MAJOR(dev) != _major || minor >= (1 << MINORBITS))
2579 spin_lock(&_minor_lock);
2581 md = idr_find(&_minor_idr, minor);
2582 if (md && (md == MINOR_ALLOCED ||
2583 (MINOR(disk_devt(dm_disk(md))) != minor) ||
2584 dm_deleting_md(md) ||
2585 test_bit(DMF_FREEING, &md->flags))) {
2591 spin_unlock(&_minor_lock);
2596 struct mapped_device *dm_get_md(dev_t dev)
2598 struct mapped_device *md = dm_find_md(dev);
2605 EXPORT_SYMBOL_GPL(dm_get_md);
2607 void *dm_get_mdptr(struct mapped_device *md)
2609 return md->interface_ptr;
2612 void dm_set_mdptr(struct mapped_device *md, void *ptr)
2614 md->interface_ptr = ptr;
2617 void dm_get(struct mapped_device *md)
2619 atomic_inc(&md->holders);
2620 BUG_ON(test_bit(DMF_FREEING, &md->flags));
2623 const char *dm_device_name(struct mapped_device *md)
2627 EXPORT_SYMBOL_GPL(dm_device_name);
2629 static void __dm_destroy(struct mapped_device *md, bool wait)
2631 struct dm_table *map;
2636 spin_lock(&_minor_lock);
2637 map = dm_get_live_table(md, &srcu_idx);
2638 idr_replace(&_minor_idr, MINOR_ALLOCED, MINOR(disk_devt(dm_disk(md))));
2639 set_bit(DMF_FREEING, &md->flags);
2640 spin_unlock(&_minor_lock);
2642 if (dm_request_based(md))
2643 flush_kthread_worker(&md->kworker);
2645 if (!dm_suspended_md(md)) {
2646 dm_table_presuspend_targets(map);
2647 dm_table_postsuspend_targets(map);
2650 /* dm_put_live_table must be before msleep, otherwise deadlock is possible */
2651 dm_put_live_table(md, srcu_idx);
2654 * Rare, but there may be I/O requests still going to complete,
2655 * for example. Wait for all references to disappear.
2656 * No one should increment the reference count of the mapped_device,
2657 * after the mapped_device state becomes DMF_FREEING.
2660 while (atomic_read(&md->holders))
2662 else if (atomic_read(&md->holders))
2663 DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)",
2664 dm_device_name(md), atomic_read(&md->holders));
2667 dm_table_destroy(__unbind(md));
2671 void dm_destroy(struct mapped_device *md)
2673 __dm_destroy(md, true);
2676 void dm_destroy_immediate(struct mapped_device *md)
2678 __dm_destroy(md, false);
2681 void dm_put(struct mapped_device *md)
2683 atomic_dec(&md->holders);
2685 EXPORT_SYMBOL_GPL(dm_put);
2687 static int dm_wait_for_completion(struct mapped_device *md, int interruptible)
2690 DECLARE_WAITQUEUE(wait, current);
2692 add_wait_queue(&md->wait, &wait);
2695 set_current_state(interruptible);
2697 if (!md_in_flight(md))
2700 if (interruptible == TASK_INTERRUPTIBLE &&
2701 signal_pending(current)) {
2708 set_current_state(TASK_RUNNING);
2710 remove_wait_queue(&md->wait, &wait);
2716 * Process the deferred bios
2718 static void dm_wq_work(struct work_struct *work)
2720 struct mapped_device *md = container_of(work, struct mapped_device,
2724 struct dm_table *map;
2726 map = dm_get_live_table(md, &srcu_idx);
2728 while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
2729 spin_lock_irq(&md->deferred_lock);
2730 c = bio_list_pop(&md->deferred);
2731 spin_unlock_irq(&md->deferred_lock);
2736 if (dm_request_based(md))
2737 generic_make_request(c);
2739 __split_and_process_bio(md, map, c);
2742 dm_put_live_table(md, srcu_idx);
2745 static void dm_queue_flush(struct mapped_device *md)
2747 clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2748 smp_mb__after_atomic();
2749 queue_work(md->wq, &md->work);
2753 * Swap in a new table, returning the old one for the caller to destroy.
2755 struct dm_table *dm_swap_table(struct mapped_device *md, struct dm_table *table)
2757 struct dm_table *live_map = NULL, *map = ERR_PTR(-EINVAL);
2758 struct queue_limits limits;
2761 mutex_lock(&md->suspend_lock);
2763 /* device must be suspended */
2764 if (!dm_suspended_md(md))
2768 * If the new table has no data devices, retain the existing limits.
2769 * This helps multipath with queue_if_no_path if all paths disappear,
2770 * then new I/O is queued based on these limits, and then some paths
2773 if (dm_table_has_no_data_devices(table)) {
2774 live_map = dm_get_live_table_fast(md);
2776 limits = md->queue->limits;
2777 dm_put_live_table_fast(md);
2781 r = dm_calculate_queue_limits(table, &limits);
2788 map = __bind(md, table, &limits);
2791 mutex_unlock(&md->suspend_lock);
2796 * Functions to lock and unlock any filesystem running on the
2799 static int lock_fs(struct mapped_device *md)
2803 WARN_ON(md->frozen_sb);
2805 md->frozen_sb = freeze_bdev(md->bdev);
2806 if (IS_ERR(md->frozen_sb)) {
2807 r = PTR_ERR(md->frozen_sb);
2808 md->frozen_sb = NULL;
2812 set_bit(DMF_FROZEN, &md->flags);
2817 static void unlock_fs(struct mapped_device *md)
2819 if (!test_bit(DMF_FROZEN, &md->flags))
2822 thaw_bdev(md->bdev, md->frozen_sb);
2823 md->frozen_sb = NULL;
2824 clear_bit(DMF_FROZEN, &md->flags);
2828 * If __dm_suspend returns 0, the device is completely quiescent
2829 * now. There is no request-processing activity. All new requests
2830 * are being added to md->deferred list.
2832 * Caller must hold md->suspend_lock
2834 static int __dm_suspend(struct mapped_device *md, struct dm_table *map,
2835 unsigned suspend_flags, int interruptible)
2837 bool do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG;
2838 bool noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG;
2842 * DMF_NOFLUSH_SUSPENDING must be set before presuspend.
2843 * This flag is cleared before dm_suspend returns.
2846 set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2849 * This gets reverted if there's an error later and the targets
2850 * provide the .presuspend_undo hook.
2852 dm_table_presuspend_targets(map);
2855 * Flush I/O to the device.
2856 * Any I/O submitted after lock_fs() may not be flushed.
2857 * noflush takes precedence over do_lockfs.
2858 * (lock_fs() flushes I/Os and waits for them to complete.)
2860 if (!noflush && do_lockfs) {
2863 dm_table_presuspend_undo_targets(map);
2869 * Here we must make sure that no processes are submitting requests
2870 * to target drivers i.e. no one may be executing
2871 * __split_and_process_bio. This is called from dm_request and
2874 * To get all processes out of __split_and_process_bio in dm_request,
2875 * we take the write lock. To prevent any process from reentering
2876 * __split_and_process_bio from dm_request and quiesce the thread
2877 * (dm_wq_work), we set BMF_BLOCK_IO_FOR_SUSPEND and call
2878 * flush_workqueue(md->wq).
2880 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2882 synchronize_srcu(&md->io_barrier);
2885 * Stop md->queue before flushing md->wq in case request-based
2886 * dm defers requests to md->wq from md->queue.
2888 if (dm_request_based(md)) {
2889 stop_queue(md->queue);
2890 flush_kthread_worker(&md->kworker);
2893 flush_workqueue(md->wq);
2896 * At this point no more requests are entering target request routines.
2897 * We call dm_wait_for_completion to wait for all existing requests
2900 r = dm_wait_for_completion(md, interruptible);
2903 clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2905 synchronize_srcu(&md->io_barrier);
2907 /* were we interrupted ? */
2911 if (dm_request_based(md))
2912 start_queue(md->queue);
2915 dm_table_presuspend_undo_targets(map);
2916 /* pushback list is already flushed, so skip flush */
2923 * We need to be able to change a mapping table under a mounted
2924 * filesystem. For example we might want to move some data in
2925 * the background. Before the table can be swapped with
2926 * dm_bind_table, dm_suspend must be called to flush any in
2927 * flight bios and ensure that any further io gets deferred.
2930 * Suspend mechanism in request-based dm.
2932 * 1. Flush all I/Os by lock_fs() if needed.
2933 * 2. Stop dispatching any I/O by stopping the request_queue.
2934 * 3. Wait for all in-flight I/Os to be completed or requeued.
2936 * To abort suspend, start the request_queue.
2938 int dm_suspend(struct mapped_device *md, unsigned suspend_flags)
2940 struct dm_table *map = NULL;
2944 mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING);
2946 if (dm_suspended_md(md)) {
2951 if (dm_suspended_internally_md(md)) {
2952 /* already internally suspended, wait for internal resume */
2953 mutex_unlock(&md->suspend_lock);
2954 r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE);
2960 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2962 r = __dm_suspend(md, map, suspend_flags, TASK_INTERRUPTIBLE);
2966 set_bit(DMF_SUSPENDED, &md->flags);
2968 dm_table_postsuspend_targets(map);
2971 mutex_unlock(&md->suspend_lock);
2975 static int __dm_resume(struct mapped_device *md, struct dm_table *map)
2978 int r = dm_table_resume_targets(map);
2986 * Flushing deferred I/Os must be done after targets are resumed
2987 * so that mapping of targets can work correctly.
2988 * Request-based dm is queueing the deferred I/Os in its request_queue.
2990 if (dm_request_based(md))
2991 start_queue(md->queue);
2998 int dm_resume(struct mapped_device *md)
3001 struct dm_table *map = NULL;
3004 mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING);
3006 if (!dm_suspended_md(md))
3009 if (dm_suspended_internally_md(md)) {
3010 /* already internally suspended, wait for internal resume */
3011 mutex_unlock(&md->suspend_lock);
3012 r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE);
3018 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
3019 if (!map || !dm_table_get_size(map))
3022 r = __dm_resume(md, map);
3026 clear_bit(DMF_SUSPENDED, &md->flags);
3030 mutex_unlock(&md->suspend_lock);
3036 * Internal suspend/resume works like userspace-driven suspend. It waits
3037 * until all bios finish and prevents issuing new bios to the target drivers.
3038 * It may be used only from the kernel.
3041 static void __dm_internal_suspend(struct mapped_device *md, unsigned suspend_flags)
3043 struct dm_table *map = NULL;
3045 if (dm_suspended_internally_md(md))
3046 return; /* nested internal suspend */
3048 if (dm_suspended_md(md)) {
3049 set_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
3050 return; /* nest suspend */
3053 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
3056 * Using TASK_UNINTERRUPTIBLE because only NOFLUSH internal suspend is
3057 * supported. Properly supporting a TASK_INTERRUPTIBLE internal suspend
3058 * would require changing .presuspend to return an error -- avoid this
3059 * until there is a need for more elaborate variants of internal suspend.
3061 (void) __dm_suspend(md, map, suspend_flags, TASK_UNINTERRUPTIBLE);
3063 set_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
3065 dm_table_postsuspend_targets(map);
3068 static void __dm_internal_resume(struct mapped_device *md)
3070 if (!dm_suspended_internally_md(md))
3071 return; /* resume from nested internal suspend */
3073 if (dm_suspended_md(md))
3074 goto done; /* resume from nested suspend */
3077 * NOTE: existing callers don't need to call dm_table_resume_targets
3078 * (which may fail -- so best to avoid it for now by passing NULL map)
3080 (void) __dm_resume(md, NULL);
3083 clear_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
3084 smp_mb__after_atomic();
3085 wake_up_bit(&md->flags, DMF_SUSPENDED_INTERNALLY);
3088 void dm_internal_suspend_noflush(struct mapped_device *md)
3090 mutex_lock(&md->suspend_lock);
3091 __dm_internal_suspend(md, DM_SUSPEND_NOFLUSH_FLAG);
3092 mutex_unlock(&md->suspend_lock);
3094 EXPORT_SYMBOL_GPL(dm_internal_suspend_noflush);
3096 void dm_internal_resume(struct mapped_device *md)
3098 mutex_lock(&md->suspend_lock);
3099 __dm_internal_resume(md);
3100 mutex_unlock(&md->suspend_lock);
3102 EXPORT_SYMBOL_GPL(dm_internal_resume);
3105 * Fast variants of internal suspend/resume hold md->suspend_lock,
3106 * which prevents interaction with userspace-driven suspend.
3109 void dm_internal_suspend_fast(struct mapped_device *md)
3111 mutex_lock(&md->suspend_lock);
3112 if (dm_suspended_md(md) || dm_suspended_internally_md(md))
3115 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
3116 synchronize_srcu(&md->io_barrier);
3117 flush_workqueue(md->wq);
3118 dm_wait_for_completion(md, TASK_UNINTERRUPTIBLE);
3121 void dm_internal_resume_fast(struct mapped_device *md)
3123 if (dm_suspended_md(md) || dm_suspended_internally_md(md))
3129 mutex_unlock(&md->suspend_lock);
3132 /*-----------------------------------------------------------------
3133 * Event notification.
3134 *---------------------------------------------------------------*/
3135 int dm_kobject_uevent(struct mapped_device *md, enum kobject_action action,
3138 char udev_cookie[DM_COOKIE_LENGTH];
3139 char *envp[] = { udev_cookie, NULL };
3142 return kobject_uevent(&disk_to_dev(md->disk)->kobj, action);
3144 snprintf(udev_cookie, DM_COOKIE_LENGTH, "%s=%u",
3145 DM_COOKIE_ENV_VAR_NAME, cookie);
3146 return kobject_uevent_env(&disk_to_dev(md->disk)->kobj,
3151 uint32_t dm_next_uevent_seq(struct mapped_device *md)
3153 return atomic_add_return(1, &md->uevent_seq);
3156 uint32_t dm_get_event_nr(struct mapped_device *md)
3158 return atomic_read(&md->event_nr);
3161 int dm_wait_event(struct mapped_device *md, int event_nr)
3163 return wait_event_interruptible(md->eventq,
3164 (event_nr != atomic_read(&md->event_nr)));
3167 void dm_uevent_add(struct mapped_device *md, struct list_head *elist)
3169 unsigned long flags;
3171 spin_lock_irqsave(&md->uevent_lock, flags);
3172 list_add(elist, &md->uevent_list);
3173 spin_unlock_irqrestore(&md->uevent_lock, flags);
3177 * The gendisk is only valid as long as you have a reference
3180 struct gendisk *dm_disk(struct mapped_device *md)
3185 struct kobject *dm_kobject(struct mapped_device *md)
3187 return &md->kobj_holder.kobj;
3190 struct mapped_device *dm_get_from_kobject(struct kobject *kobj)
3192 struct mapped_device *md;
3194 md = container_of(kobj, struct mapped_device, kobj_holder.kobj);
3196 if (test_bit(DMF_FREEING, &md->flags) ||
3204 int dm_suspended_md(struct mapped_device *md)
3206 return test_bit(DMF_SUSPENDED, &md->flags);
3209 int dm_suspended_internally_md(struct mapped_device *md)
3211 return test_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
3214 int dm_test_deferred_remove_flag(struct mapped_device *md)
3216 return test_bit(DMF_DEFERRED_REMOVE, &md->flags);
3219 int dm_suspended(struct dm_target *ti)
3221 return dm_suspended_md(dm_table_get_md(ti->table));
3223 EXPORT_SYMBOL_GPL(dm_suspended);
3225 int dm_noflush_suspending(struct dm_target *ti)
3227 return __noflush_suspending(dm_table_get_md(ti->table));
3229 EXPORT_SYMBOL_GPL(dm_noflush_suspending);
3231 struct dm_md_mempools *dm_alloc_md_mempools(unsigned type, unsigned integrity, unsigned per_bio_data_size)
3233 struct dm_md_mempools *pools = kzalloc(sizeof(*pools), GFP_KERNEL);
3234 struct kmem_cache *cachep;
3235 unsigned int pool_size = 0;
3236 unsigned int front_pad;
3242 case DM_TYPE_BIO_BASED:
3244 pool_size = dm_get_reserved_bio_based_ios();
3245 front_pad = roundup(per_bio_data_size, __alignof__(struct dm_target_io)) + offsetof(struct dm_target_io, clone);
3247 case DM_TYPE_REQUEST_BASED:
3248 pool_size = dm_get_reserved_rq_based_ios();
3249 pools->rq_pool = mempool_create_slab_pool(pool_size, _rq_cache);
3250 if (!pools->rq_pool)
3252 /* fall through to setup remaining rq-based pools */
3253 case DM_TYPE_MQ_REQUEST_BASED:
3254 cachep = _rq_tio_cache;
3256 pool_size = dm_get_reserved_rq_based_ios();
3257 front_pad = offsetof(struct dm_rq_clone_bio_info, clone);
3258 /* per_bio_data_size is not used. See __bind_mempools(). */
3259 WARN_ON(per_bio_data_size != 0);
3265 pools->io_pool = mempool_create_slab_pool(pool_size, cachep);
3266 if (!pools->io_pool)
3269 pools->bs = bioset_create_nobvec(pool_size, front_pad);
3273 if (integrity && bioset_integrity_create(pools->bs, pool_size))
3279 dm_free_md_mempools(pools);
3284 void dm_free_md_mempools(struct dm_md_mempools *pools)
3290 mempool_destroy(pools->io_pool);
3293 mempool_destroy(pools->rq_pool);
3296 bioset_free(pools->bs);
3301 static const struct block_device_operations dm_blk_dops = {
3302 .open = dm_blk_open,
3303 .release = dm_blk_close,
3304 .ioctl = dm_blk_ioctl,
3305 .getgeo = dm_blk_getgeo,
3306 .owner = THIS_MODULE
3312 module_init(dm_init);
3313 module_exit(dm_exit);
3315 module_param(major, uint, 0);
3316 MODULE_PARM_DESC(major, "The major number of the device mapper");
3318 module_param(reserved_bio_based_ios, uint, S_IRUGO | S_IWUSR);
3319 MODULE_PARM_DESC(reserved_bio_based_ios, "Reserved IOs in bio-based mempools");
3321 module_param(reserved_rq_based_ios, uint, S_IRUGO | S_IWUSR);
3322 MODULE_PARM_DESC(reserved_rq_based_ios, "Reserved IOs in request-based mempools");
3324 MODULE_DESCRIPTION(DM_NAME " driver");
3325 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
3326 MODULE_LICENSE("GPL");