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>
29 #include <trace/events/block.h>
31 #define DM_MSG_PREFIX "core"
35 * ratelimit state to be used in DMXXX_LIMIT().
37 DEFINE_RATELIMIT_STATE(dm_ratelimit_state,
38 DEFAULT_RATELIMIT_INTERVAL,
39 DEFAULT_RATELIMIT_BURST);
40 EXPORT_SYMBOL(dm_ratelimit_state);
44 * Cookies are numeric values sent with CHANGE and REMOVE
45 * uevents while resuming, removing or renaming the device.
47 #define DM_COOKIE_ENV_VAR_NAME "DM_COOKIE"
48 #define DM_COOKIE_LENGTH 24
50 static const char *_name = DM_NAME;
52 static unsigned int major = 0;
53 static unsigned int _major = 0;
55 static DEFINE_IDR(_minor_idr);
57 static DEFINE_SPINLOCK(_minor_lock);
59 static void do_deferred_remove(struct work_struct *w);
61 static DECLARE_WORK(deferred_remove_work, do_deferred_remove);
63 static struct workqueue_struct *deferred_remove_workqueue;
67 * One of these is allocated per bio.
70 struct mapped_device *md;
74 unsigned long start_time;
75 spinlock_t endio_lock;
76 struct dm_stats_aux stats_aux;
80 * For request-based dm.
81 * One of these is allocated per request.
83 struct dm_rq_target_io {
84 struct mapped_device *md;
86 struct request *orig, *clone;
87 struct kthread_work work;
90 struct dm_stats_aux stats_aux;
91 unsigned long duration_jiffies;
96 * For request-based dm - the bio clones we allocate are embedded in these
99 * We allocate these with bio_alloc_bioset, using the front_pad parameter when
100 * the bioset is created - this means the bio has to come at the end of the
103 struct dm_rq_clone_bio_info {
105 struct dm_rq_target_io *tio;
109 #define MINOR_ALLOCED ((void *)-1)
112 * Bits for the md->flags field.
114 #define DMF_BLOCK_IO_FOR_SUSPEND 0
115 #define DMF_SUSPENDED 1
117 #define DMF_FREEING 3
118 #define DMF_DELETING 4
119 #define DMF_NOFLUSH_SUSPENDING 5
120 #define DMF_DEFERRED_REMOVE 6
121 #define DMF_SUSPENDED_INTERNALLY 7
124 * Work processed by per-device workqueue.
126 struct mapped_device {
127 struct srcu_struct io_barrier;
128 struct mutex suspend_lock;
131 * The current mapping (struct dm_table *).
132 * Use dm_get_live_table{_fast} or take suspend_lock for
137 struct list_head table_devices;
138 struct mutex table_devices_lock;
142 struct request_queue *queue;
146 /* Protect queue and type against concurrent access. */
147 struct mutex type_lock;
152 struct dm_target *immutable_target;
153 struct target_type *immutable_target_type;
155 struct gendisk *disk;
161 * A list of ios that arrived while we were suspended.
164 wait_queue_head_t wait;
165 struct work_struct work;
166 spinlock_t deferred_lock;
167 struct bio_list deferred;
172 wait_queue_head_t eventq;
175 struct list_head uevent_list;
176 spinlock_t uevent_lock; /* Protect access to uevent_list */
178 /* the number of internal suspends */
179 unsigned internal_suspend_count;
182 * Processing queue (flush)
184 struct workqueue_struct *wq;
187 * io objects are allocated from here.
195 * freeze/thaw support require holding onto a super block
197 struct super_block *frozen_sb;
199 /* forced geometry settings */
200 struct hd_geometry geometry;
202 struct block_device *bdev;
204 /* kobject and completion */
205 struct dm_kobject_holder kobj_holder;
207 /* zero-length flush that will be cloned and submitted to targets */
208 struct bio flush_bio;
210 struct dm_stats stats;
212 struct kthread_worker kworker;
213 struct task_struct *kworker_task;
215 /* for request-based merge heuristic in dm_request_fn() */
216 unsigned seq_rq_merge_deadline_usecs;
218 sector_t last_rq_pos;
219 ktime_t last_rq_start_time;
221 /* for blk-mq request-based DM support */
222 struct blk_mq_tag_set *tag_set;
227 #ifdef CONFIG_DM_MQ_DEFAULT
228 static bool use_blk_mq = true;
230 static bool use_blk_mq = false;
233 #define DM_MQ_NR_HW_QUEUES 1
234 #define DM_MQ_QUEUE_DEPTH 2048
235 #define DM_NUMA_NODE NUMA_NO_NODE
237 static unsigned dm_mq_nr_hw_queues = DM_MQ_NR_HW_QUEUES;
238 static unsigned dm_mq_queue_depth = DM_MQ_QUEUE_DEPTH;
239 static int dm_numa_node = DM_NUMA_NODE;
241 bool dm_use_blk_mq(struct mapped_device *md)
243 return md->use_blk_mq;
245 EXPORT_SYMBOL_GPL(dm_use_blk_mq);
248 * For mempools pre-allocation at the table loading time.
250 struct dm_md_mempools {
256 struct table_device {
257 struct list_head list;
259 struct dm_dev dm_dev;
262 #define RESERVED_BIO_BASED_IOS 16
263 #define RESERVED_REQUEST_BASED_IOS 256
264 #define RESERVED_MAX_IOS 1024
265 static struct kmem_cache *_io_cache;
266 static struct kmem_cache *_rq_tio_cache;
267 static struct kmem_cache *_rq_cache;
270 * Bio-based DM's mempools' reserved IOs set by the user.
272 static unsigned reserved_bio_based_ios = RESERVED_BIO_BASED_IOS;
275 * Request-based DM's mempools' reserved IOs set by the user.
277 static unsigned reserved_rq_based_ios = RESERVED_REQUEST_BASED_IOS;
279 static int __dm_get_module_param_int(int *module_param, int min, int max)
281 int param = ACCESS_ONCE(*module_param);
282 int modified_param = 0;
283 bool modified = true;
286 modified_param = min;
287 else if (param > max)
288 modified_param = max;
293 (void)cmpxchg(module_param, param, modified_param);
294 param = modified_param;
300 static unsigned __dm_get_module_param(unsigned *module_param,
301 unsigned def, unsigned max)
303 unsigned param = ACCESS_ONCE(*module_param);
304 unsigned modified_param = 0;
307 modified_param = def;
308 else if (param > max)
309 modified_param = max;
311 if (modified_param) {
312 (void)cmpxchg(module_param, param, modified_param);
313 param = modified_param;
319 unsigned dm_get_reserved_bio_based_ios(void)
321 return __dm_get_module_param(&reserved_bio_based_ios,
322 RESERVED_BIO_BASED_IOS, RESERVED_MAX_IOS);
324 EXPORT_SYMBOL_GPL(dm_get_reserved_bio_based_ios);
326 unsigned dm_get_reserved_rq_based_ios(void)
328 return __dm_get_module_param(&reserved_rq_based_ios,
329 RESERVED_REQUEST_BASED_IOS, RESERVED_MAX_IOS);
331 EXPORT_SYMBOL_GPL(dm_get_reserved_rq_based_ios);
333 static unsigned dm_get_blk_mq_nr_hw_queues(void)
335 return __dm_get_module_param(&dm_mq_nr_hw_queues, 1, 32);
338 static unsigned dm_get_blk_mq_queue_depth(void)
340 return __dm_get_module_param(&dm_mq_queue_depth,
341 DM_MQ_QUEUE_DEPTH, BLK_MQ_MAX_DEPTH);
344 static unsigned dm_get_numa_node(void)
346 return __dm_get_module_param_int(&dm_numa_node,
347 DM_NUMA_NODE, num_online_nodes() - 1);
350 static int __init local_init(void)
354 /* allocate a slab for the dm_ios */
355 _io_cache = KMEM_CACHE(dm_io, 0);
359 _rq_tio_cache = KMEM_CACHE(dm_rq_target_io, 0);
361 goto out_free_io_cache;
363 _rq_cache = kmem_cache_create("dm_old_clone_request", sizeof(struct request),
364 __alignof__(struct request), 0, NULL);
366 goto out_free_rq_tio_cache;
368 r = dm_uevent_init();
370 goto out_free_rq_cache;
372 deferred_remove_workqueue = alloc_workqueue("kdmremove", WQ_UNBOUND, 1);
373 if (!deferred_remove_workqueue) {
375 goto out_uevent_exit;
379 r = register_blkdev(_major, _name);
381 goto out_free_workqueue;
389 destroy_workqueue(deferred_remove_workqueue);
393 kmem_cache_destroy(_rq_cache);
394 out_free_rq_tio_cache:
395 kmem_cache_destroy(_rq_tio_cache);
397 kmem_cache_destroy(_io_cache);
402 static void local_exit(void)
404 flush_scheduled_work();
405 destroy_workqueue(deferred_remove_workqueue);
407 kmem_cache_destroy(_rq_cache);
408 kmem_cache_destroy(_rq_tio_cache);
409 kmem_cache_destroy(_io_cache);
410 unregister_blkdev(_major, _name);
415 DMINFO("cleaned up");
418 static int (*_inits[])(void) __initdata = {
429 static void (*_exits[])(void) = {
440 static int __init dm_init(void)
442 const int count = ARRAY_SIZE(_inits);
446 for (i = 0; i < count; i++) {
461 static void __exit dm_exit(void)
463 int i = ARRAY_SIZE(_exits);
469 * Should be empty by this point.
471 idr_destroy(&_minor_idr);
475 * Block device functions
477 int dm_deleting_md(struct mapped_device *md)
479 return test_bit(DMF_DELETING, &md->flags);
482 static int dm_blk_open(struct block_device *bdev, fmode_t mode)
484 struct mapped_device *md;
486 spin_lock(&_minor_lock);
488 md = bdev->bd_disk->private_data;
492 if (test_bit(DMF_FREEING, &md->flags) ||
493 dm_deleting_md(md)) {
499 atomic_inc(&md->open_count);
501 spin_unlock(&_minor_lock);
503 return md ? 0 : -ENXIO;
506 static void dm_blk_close(struct gendisk *disk, fmode_t mode)
508 struct mapped_device *md;
510 spin_lock(&_minor_lock);
512 md = disk->private_data;
516 if (atomic_dec_and_test(&md->open_count) &&
517 (test_bit(DMF_DEFERRED_REMOVE, &md->flags)))
518 queue_work(deferred_remove_workqueue, &deferred_remove_work);
522 spin_unlock(&_minor_lock);
525 int dm_open_count(struct mapped_device *md)
527 return atomic_read(&md->open_count);
531 * Guarantees nothing is using the device before it's deleted.
533 int dm_lock_for_deletion(struct mapped_device *md, bool mark_deferred, bool only_deferred)
537 spin_lock(&_minor_lock);
539 if (dm_open_count(md)) {
542 set_bit(DMF_DEFERRED_REMOVE, &md->flags);
543 } else if (only_deferred && !test_bit(DMF_DEFERRED_REMOVE, &md->flags))
546 set_bit(DMF_DELETING, &md->flags);
548 spin_unlock(&_minor_lock);
553 int dm_cancel_deferred_remove(struct mapped_device *md)
557 spin_lock(&_minor_lock);
559 if (test_bit(DMF_DELETING, &md->flags))
562 clear_bit(DMF_DEFERRED_REMOVE, &md->flags);
564 spin_unlock(&_minor_lock);
569 static void do_deferred_remove(struct work_struct *w)
571 dm_deferred_remove();
574 sector_t dm_get_size(struct mapped_device *md)
576 return get_capacity(md->disk);
579 struct request_queue *dm_get_md_queue(struct mapped_device *md)
584 struct dm_stats *dm_get_stats(struct mapped_device *md)
589 static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo)
591 struct mapped_device *md = bdev->bd_disk->private_data;
593 return dm_get_geometry(md, geo);
596 static int dm_grab_bdev_for_ioctl(struct mapped_device *md,
597 struct block_device **bdev,
600 struct dm_target *tgt;
601 struct dm_table *map;
606 map = dm_get_live_table(md, &srcu_idx);
607 if (!map || !dm_table_get_size(map))
610 /* We only support devices that have a single target */
611 if (dm_table_get_num_targets(map) != 1)
614 tgt = dm_table_get_target(map, 0);
615 if (!tgt->type->prepare_ioctl)
618 if (dm_suspended_md(md)) {
623 r = tgt->type->prepare_ioctl(tgt, bdev, mode);
628 dm_put_live_table(md, srcu_idx);
632 dm_put_live_table(md, srcu_idx);
633 if (r == -ENOTCONN && !fatal_signal_pending(current)) {
640 static int dm_blk_ioctl(struct block_device *bdev, fmode_t mode,
641 unsigned int cmd, unsigned long arg)
643 struct mapped_device *md = bdev->bd_disk->private_data;
646 r = dm_grab_bdev_for_ioctl(md, &bdev, &mode);
652 * Target determined this ioctl is being issued against
653 * a logical partition of the parent bdev; so extra
654 * validation is needed.
656 r = scsi_verify_blk_ioctl(NULL, cmd);
661 r = __blkdev_driver_ioctl(bdev, mode, cmd, arg);
667 static struct dm_io *alloc_io(struct mapped_device *md)
669 return mempool_alloc(md->io_pool, GFP_NOIO);
672 static void free_io(struct mapped_device *md, struct dm_io *io)
674 mempool_free(io, md->io_pool);
677 static void free_tio(struct mapped_device *md, struct dm_target_io *tio)
679 bio_put(&tio->clone);
682 static struct dm_rq_target_io *alloc_old_rq_tio(struct mapped_device *md,
685 return mempool_alloc(md->io_pool, gfp_mask);
688 static void free_old_rq_tio(struct dm_rq_target_io *tio)
690 mempool_free(tio, tio->md->io_pool);
693 static struct request *alloc_old_clone_request(struct mapped_device *md,
696 return mempool_alloc(md->rq_pool, gfp_mask);
699 static void free_old_clone_request(struct mapped_device *md, struct request *rq)
701 mempool_free(rq, md->rq_pool);
704 static int md_in_flight(struct mapped_device *md)
706 return atomic_read(&md->pending[READ]) +
707 atomic_read(&md->pending[WRITE]);
710 static void start_io_acct(struct dm_io *io)
712 struct mapped_device *md = io->md;
713 struct bio *bio = io->bio;
715 int rw = bio_data_dir(bio);
717 io->start_time = jiffies;
719 cpu = part_stat_lock();
720 part_round_stats(cpu, &dm_disk(md)->part0);
722 atomic_set(&dm_disk(md)->part0.in_flight[rw],
723 atomic_inc_return(&md->pending[rw]));
725 if (unlikely(dm_stats_used(&md->stats)))
726 dm_stats_account_io(&md->stats, bio->bi_rw, bio->bi_iter.bi_sector,
727 bio_sectors(bio), false, 0, &io->stats_aux);
730 static void end_io_acct(struct dm_io *io)
732 struct mapped_device *md = io->md;
733 struct bio *bio = io->bio;
734 unsigned long duration = jiffies - io->start_time;
736 int rw = bio_data_dir(bio);
738 generic_end_io_acct(rw, &dm_disk(md)->part0, io->start_time);
740 if (unlikely(dm_stats_used(&md->stats)))
741 dm_stats_account_io(&md->stats, bio->bi_rw, bio->bi_iter.bi_sector,
742 bio_sectors(bio), true, duration, &io->stats_aux);
745 * After this is decremented the bio must not be touched if it is
748 pending = atomic_dec_return(&md->pending[rw]);
749 atomic_set(&dm_disk(md)->part0.in_flight[rw], pending);
750 pending += atomic_read(&md->pending[rw^0x1]);
752 /* nudge anyone waiting on suspend queue */
758 * Add the bio to the list of deferred io.
760 static void queue_io(struct mapped_device *md, struct bio *bio)
764 spin_lock_irqsave(&md->deferred_lock, flags);
765 bio_list_add(&md->deferred, bio);
766 spin_unlock_irqrestore(&md->deferred_lock, flags);
767 queue_work(md->wq, &md->work);
771 * Everyone (including functions in this file), should use this
772 * function to access the md->map field, and make sure they call
773 * dm_put_live_table() when finished.
775 struct dm_table *dm_get_live_table(struct mapped_device *md, int *srcu_idx) __acquires(md->io_barrier)
777 *srcu_idx = srcu_read_lock(&md->io_barrier);
779 return srcu_dereference(md->map, &md->io_barrier);
782 void dm_put_live_table(struct mapped_device *md, int srcu_idx) __releases(md->io_barrier)
784 srcu_read_unlock(&md->io_barrier, srcu_idx);
787 void dm_sync_table(struct mapped_device *md)
789 synchronize_srcu(&md->io_barrier);
790 synchronize_rcu_expedited();
794 * A fast alternative to dm_get_live_table/dm_put_live_table.
795 * The caller must not block between these two functions.
797 static struct dm_table *dm_get_live_table_fast(struct mapped_device *md) __acquires(RCU)
800 return rcu_dereference(md->map);
803 static void dm_put_live_table_fast(struct mapped_device *md) __releases(RCU)
809 * Open a table device so we can use it as a map destination.
811 static int open_table_device(struct table_device *td, dev_t dev,
812 struct mapped_device *md)
814 static char *_claim_ptr = "I belong to device-mapper";
815 struct block_device *bdev;
819 BUG_ON(td->dm_dev.bdev);
821 bdev = blkdev_get_by_dev(dev, td->dm_dev.mode | FMODE_EXCL, _claim_ptr);
823 return PTR_ERR(bdev);
825 r = bd_link_disk_holder(bdev, dm_disk(md));
827 blkdev_put(bdev, td->dm_dev.mode | FMODE_EXCL);
831 td->dm_dev.bdev = bdev;
836 * Close a table device that we've been using.
838 static void close_table_device(struct table_device *td, struct mapped_device *md)
840 if (!td->dm_dev.bdev)
843 bd_unlink_disk_holder(td->dm_dev.bdev, dm_disk(md));
844 blkdev_put(td->dm_dev.bdev, td->dm_dev.mode | FMODE_EXCL);
845 td->dm_dev.bdev = NULL;
848 static struct table_device *find_table_device(struct list_head *l, dev_t dev,
850 struct table_device *td;
852 list_for_each_entry(td, l, list)
853 if (td->dm_dev.bdev->bd_dev == dev && td->dm_dev.mode == mode)
859 int dm_get_table_device(struct mapped_device *md, dev_t dev, fmode_t mode,
860 struct dm_dev **result) {
862 struct table_device *td;
864 mutex_lock(&md->table_devices_lock);
865 td = find_table_device(&md->table_devices, dev, mode);
867 td = kmalloc_node(sizeof(*td), GFP_KERNEL, md->numa_node_id);
869 mutex_unlock(&md->table_devices_lock);
873 td->dm_dev.mode = mode;
874 td->dm_dev.bdev = NULL;
876 if ((r = open_table_device(td, dev, md))) {
877 mutex_unlock(&md->table_devices_lock);
882 format_dev_t(td->dm_dev.name, dev);
884 atomic_set(&td->count, 0);
885 list_add(&td->list, &md->table_devices);
887 atomic_inc(&td->count);
888 mutex_unlock(&md->table_devices_lock);
890 *result = &td->dm_dev;
893 EXPORT_SYMBOL_GPL(dm_get_table_device);
895 void dm_put_table_device(struct mapped_device *md, struct dm_dev *d)
897 struct table_device *td = container_of(d, struct table_device, dm_dev);
899 mutex_lock(&md->table_devices_lock);
900 if (atomic_dec_and_test(&td->count)) {
901 close_table_device(td, md);
905 mutex_unlock(&md->table_devices_lock);
907 EXPORT_SYMBOL(dm_put_table_device);
909 static void free_table_devices(struct list_head *devices)
911 struct list_head *tmp, *next;
913 list_for_each_safe(tmp, next, devices) {
914 struct table_device *td = list_entry(tmp, struct table_device, list);
916 DMWARN("dm_destroy: %s still exists with %d references",
917 td->dm_dev.name, atomic_read(&td->count));
923 * Get the geometry associated with a dm device
925 int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo)
933 * Set the geometry of a device.
935 int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo)
937 sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors;
939 if (geo->start > sz) {
940 DMWARN("Start sector is beyond the geometry limits.");
949 /*-----------------------------------------------------------------
951 * A more elegant soln is in the works that uses the queue
952 * merge fn, unfortunately there are a couple of changes to
953 * the block layer that I want to make for this. So in the
954 * interests of getting something for people to use I give
955 * you this clearly demarcated crap.
956 *---------------------------------------------------------------*/
958 static int __noflush_suspending(struct mapped_device *md)
960 return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
964 * Decrements the number of outstanding ios that a bio has been
965 * cloned into, completing the original io if necc.
967 static void dec_pending(struct dm_io *io, int error)
972 struct mapped_device *md = io->md;
974 /* Push-back supersedes any I/O errors */
975 if (unlikely(error)) {
976 spin_lock_irqsave(&io->endio_lock, flags);
977 if (!(io->error > 0 && __noflush_suspending(md)))
979 spin_unlock_irqrestore(&io->endio_lock, flags);
982 if (atomic_dec_and_test(&io->io_count)) {
983 if (io->error == DM_ENDIO_REQUEUE) {
985 * Target requested pushing back the I/O.
987 spin_lock_irqsave(&md->deferred_lock, flags);
988 if (__noflush_suspending(md))
989 bio_list_add_head(&md->deferred, io->bio);
991 /* noflush suspend was interrupted. */
993 spin_unlock_irqrestore(&md->deferred_lock, flags);
996 io_error = io->error;
1001 if (io_error == DM_ENDIO_REQUEUE)
1004 if ((bio->bi_rw & REQ_FLUSH) && bio->bi_iter.bi_size) {
1006 * Preflush done for flush with data, reissue
1007 * without REQ_FLUSH.
1009 bio->bi_rw &= ~REQ_FLUSH;
1012 /* done with normal IO or empty flush */
1013 trace_block_bio_complete(md->queue, bio, io_error);
1014 bio->bi_error = io_error;
1020 static void disable_write_same(struct mapped_device *md)
1022 struct queue_limits *limits = dm_get_queue_limits(md);
1024 /* device doesn't really support WRITE SAME, disable it */
1025 limits->max_write_same_sectors = 0;
1028 static void clone_endio(struct bio *bio)
1030 int error = bio->bi_error;
1032 struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone);
1033 struct dm_io *io = tio->io;
1034 struct mapped_device *md = tio->io->md;
1035 dm_endio_fn endio = tio->ti->type->end_io;
1038 r = endio(tio->ti, bio, error);
1039 if (r < 0 || r == DM_ENDIO_REQUEUE)
1041 * error and requeue request are handled
1045 else if (r == DM_ENDIO_INCOMPLETE)
1046 /* The target will handle the io */
1049 DMWARN("unimplemented target endio return value: %d", r);
1054 if (unlikely(r == -EREMOTEIO && (bio->bi_rw & REQ_WRITE_SAME) &&
1055 !bdev_get_queue(bio->bi_bdev)->limits.max_write_same_sectors))
1056 disable_write_same(md);
1059 dec_pending(io, error);
1063 * Partial completion handling for request-based dm
1065 static void end_clone_bio(struct bio *clone)
1067 struct dm_rq_clone_bio_info *info =
1068 container_of(clone, struct dm_rq_clone_bio_info, clone);
1069 struct dm_rq_target_io *tio = info->tio;
1070 struct bio *bio = info->orig;
1071 unsigned int nr_bytes = info->orig->bi_iter.bi_size;
1072 int error = clone->bi_error;
1078 * An error has already been detected on the request.
1079 * Once error occurred, just let clone->end_io() handle
1085 * Don't notice the error to the upper layer yet.
1086 * The error handling decision is made by the target driver,
1087 * when the request is completed.
1094 * I/O for the bio successfully completed.
1095 * Notice the data completion to the upper layer.
1099 * bios are processed from the head of the list.
1100 * So the completing bio should always be rq->bio.
1101 * If it's not, something wrong is happening.
1103 if (tio->orig->bio != bio)
1104 DMERR("bio completion is going in the middle of the request");
1107 * Update the original request.
1108 * Do not use blk_end_request() here, because it may complete
1109 * the original request before the clone, and break the ordering.
1111 blk_update_request(tio->orig, 0, nr_bytes);
1114 static struct dm_rq_target_io *tio_from_request(struct request *rq)
1116 return (rq->q->mq_ops ? blk_mq_rq_to_pdu(rq) : rq->special);
1119 static void rq_end_stats(struct mapped_device *md, struct request *orig)
1121 if (unlikely(dm_stats_used(&md->stats))) {
1122 struct dm_rq_target_io *tio = tio_from_request(orig);
1123 tio->duration_jiffies = jiffies - tio->duration_jiffies;
1124 dm_stats_account_io(&md->stats, orig->cmd_flags, blk_rq_pos(orig),
1125 tio->n_sectors, true, tio->duration_jiffies,
1131 * Don't touch any member of the md after calling this function because
1132 * the md may be freed in dm_put() at the end of this function.
1133 * Or do dm_get() before calling this function and dm_put() later.
1135 static void rq_completed(struct mapped_device *md, int rw, bool run_queue)
1137 atomic_dec(&md->pending[rw]);
1139 /* nudge anyone waiting on suspend queue */
1140 if (!md_in_flight(md))
1144 * Run this off this callpath, as drivers could invoke end_io while
1145 * inside their request_fn (and holding the queue lock). Calling
1146 * back into ->request_fn() could deadlock attempting to grab the
1149 if (!md->queue->mq_ops && run_queue)
1150 blk_run_queue_async(md->queue);
1153 * dm_put() must be at the end of this function. See the comment above
1158 static void free_rq_clone(struct request *clone)
1160 struct dm_rq_target_io *tio = clone->end_io_data;
1161 struct mapped_device *md = tio->md;
1163 blk_rq_unprep_clone(clone);
1165 if (md->type == DM_TYPE_MQ_REQUEST_BASED)
1166 /* stacked on blk-mq queue(s) */
1167 tio->ti->type->release_clone_rq(clone);
1168 else if (!md->queue->mq_ops)
1169 /* request_fn queue stacked on request_fn queue(s) */
1170 free_old_clone_request(md, clone);
1172 if (!md->queue->mq_ops)
1173 free_old_rq_tio(tio);
1177 * Complete the clone and the original request.
1178 * Must be called without clone's queue lock held,
1179 * see end_clone_request() for more details.
1181 static void dm_end_request(struct request *clone, int error)
1183 int rw = rq_data_dir(clone);
1184 struct dm_rq_target_io *tio = clone->end_io_data;
1185 struct mapped_device *md = tio->md;
1186 struct request *rq = tio->orig;
1188 if (rq->cmd_type == REQ_TYPE_BLOCK_PC) {
1189 rq->errors = clone->errors;
1190 rq->resid_len = clone->resid_len;
1194 * We are using the sense buffer of the original
1196 * So setting the length of the sense data is enough.
1198 rq->sense_len = clone->sense_len;
1201 free_rq_clone(clone);
1202 rq_end_stats(md, rq);
1204 blk_end_request_all(rq, error);
1206 blk_mq_end_request(rq, error);
1207 rq_completed(md, rw, true);
1210 static void dm_unprep_request(struct request *rq)
1212 struct dm_rq_target_io *tio = tio_from_request(rq);
1213 struct request *clone = tio->clone;
1215 if (!rq->q->mq_ops) {
1217 rq->cmd_flags &= ~REQ_DONTPREP;
1221 free_rq_clone(clone);
1222 else if (!tio->md->queue->mq_ops)
1223 free_old_rq_tio(tio);
1227 * Requeue the original request of a clone.
1229 static void dm_old_requeue_request(struct request *rq)
1231 struct request_queue *q = rq->q;
1232 unsigned long flags;
1234 spin_lock_irqsave(q->queue_lock, flags);
1235 blk_requeue_request(q, rq);
1236 blk_run_queue_async(q);
1237 spin_unlock_irqrestore(q->queue_lock, flags);
1240 static void dm_mq_requeue_request(struct request *rq)
1242 struct request_queue *q = rq->q;
1243 unsigned long flags;
1245 blk_mq_requeue_request(rq);
1246 spin_lock_irqsave(q->queue_lock, flags);
1247 if (!blk_queue_stopped(q))
1248 blk_mq_kick_requeue_list(q);
1249 spin_unlock_irqrestore(q->queue_lock, flags);
1252 static void dm_requeue_original_request(struct mapped_device *md,
1255 int rw = rq_data_dir(rq);
1257 dm_unprep_request(rq);
1259 rq_end_stats(md, rq);
1261 dm_old_requeue_request(rq);
1263 dm_mq_requeue_request(rq);
1265 rq_completed(md, rw, false);
1268 static void dm_old_stop_queue(struct request_queue *q)
1270 unsigned long flags;
1272 spin_lock_irqsave(q->queue_lock, flags);
1273 if (blk_queue_stopped(q)) {
1274 spin_unlock_irqrestore(q->queue_lock, flags);
1279 spin_unlock_irqrestore(q->queue_lock, flags);
1282 static void dm_stop_queue(struct request_queue *q)
1285 dm_old_stop_queue(q);
1287 blk_mq_stop_hw_queues(q);
1290 static void dm_old_start_queue(struct request_queue *q)
1292 unsigned long flags;
1294 spin_lock_irqsave(q->queue_lock, flags);
1295 if (blk_queue_stopped(q))
1297 spin_unlock_irqrestore(q->queue_lock, flags);
1300 static void dm_start_queue(struct request_queue *q)
1303 dm_old_start_queue(q);
1305 blk_mq_start_stopped_hw_queues(q, true);
1306 blk_mq_kick_requeue_list(q);
1310 static void dm_done(struct request *clone, int error, bool mapped)
1313 struct dm_rq_target_io *tio = clone->end_io_data;
1314 dm_request_endio_fn rq_end_io = NULL;
1317 rq_end_io = tio->ti->type->rq_end_io;
1319 if (mapped && rq_end_io)
1320 r = rq_end_io(tio->ti, clone, error, &tio->info);
1323 if (unlikely(r == -EREMOTEIO && (clone->cmd_flags & REQ_WRITE_SAME) &&
1324 !clone->q->limits.max_write_same_sectors))
1325 disable_write_same(tio->md);
1328 /* The target wants to complete the I/O */
1329 dm_end_request(clone, r);
1330 else if (r == DM_ENDIO_INCOMPLETE)
1331 /* The target will handle the I/O */
1333 else if (r == DM_ENDIO_REQUEUE)
1334 /* The target wants to requeue the I/O */
1335 dm_requeue_original_request(tio->md, tio->orig);
1337 DMWARN("unimplemented target endio return value: %d", r);
1343 * Request completion handler for request-based dm
1345 static void dm_softirq_done(struct request *rq)
1348 struct dm_rq_target_io *tio = tio_from_request(rq);
1349 struct request *clone = tio->clone;
1353 rq_end_stats(tio->md, rq);
1354 rw = rq_data_dir(rq);
1355 if (!rq->q->mq_ops) {
1356 blk_end_request_all(rq, tio->error);
1357 rq_completed(tio->md, rw, false);
1358 free_old_rq_tio(tio);
1360 blk_mq_end_request(rq, tio->error);
1361 rq_completed(tio->md, rw, false);
1366 if (rq->cmd_flags & REQ_FAILED)
1369 dm_done(clone, tio->error, mapped);
1373 * Complete the clone and the original request with the error status
1374 * through softirq context.
1376 static void dm_complete_request(struct request *rq, int error)
1378 struct dm_rq_target_io *tio = tio_from_request(rq);
1382 blk_complete_request(rq);
1384 blk_mq_complete_request(rq, error);
1388 * Complete the not-mapped clone and the original request with the error status
1389 * through softirq context.
1390 * Target's rq_end_io() function isn't called.
1391 * This may be used when the target's map_rq() or clone_and_map_rq() functions fail.
1393 static void dm_kill_unmapped_request(struct request *rq, int error)
1395 rq->cmd_flags |= REQ_FAILED;
1396 dm_complete_request(rq, error);
1400 * Called with the clone's queue lock held (in the case of .request_fn)
1402 static void end_clone_request(struct request *clone, int error)
1404 struct dm_rq_target_io *tio = clone->end_io_data;
1406 if (!clone->q->mq_ops) {
1408 * For just cleaning up the information of the queue in which
1409 * the clone was dispatched.
1410 * The clone is *NOT* freed actually here because it is alloced
1411 * from dm own mempool (REQ_ALLOCED isn't set).
1413 __blk_put_request(clone->q, clone);
1417 * Actual request completion is done in a softirq context which doesn't
1418 * hold the clone's queue lock. Otherwise, deadlock could occur because:
1419 * - another request may be submitted by the upper level driver
1420 * of the stacking during the completion
1421 * - the submission which requires queue lock may be done
1422 * against this clone's queue
1424 dm_complete_request(tio->orig, error);
1428 * Return maximum size of I/O possible at the supplied sector up to the current
1431 static sector_t max_io_len_target_boundary(sector_t sector, struct dm_target *ti)
1433 sector_t target_offset = dm_target_offset(ti, sector);
1435 return ti->len - target_offset;
1438 static sector_t max_io_len(sector_t sector, struct dm_target *ti)
1440 sector_t len = max_io_len_target_boundary(sector, ti);
1441 sector_t offset, max_len;
1444 * Does the target need to split even further?
1446 if (ti->max_io_len) {
1447 offset = dm_target_offset(ti, sector);
1448 if (unlikely(ti->max_io_len & (ti->max_io_len - 1)))
1449 max_len = sector_div(offset, ti->max_io_len);
1451 max_len = offset & (ti->max_io_len - 1);
1452 max_len = ti->max_io_len - max_len;
1461 int dm_set_target_max_io_len(struct dm_target *ti, sector_t len)
1463 if (len > UINT_MAX) {
1464 DMERR("Specified maximum size of target IO (%llu) exceeds limit (%u)",
1465 (unsigned long long)len, UINT_MAX);
1466 ti->error = "Maximum size of target IO is too large";
1470 ti->max_io_len = (uint32_t) len;
1474 EXPORT_SYMBOL_GPL(dm_set_target_max_io_len);
1477 * A target may call dm_accept_partial_bio only from the map routine. It is
1478 * allowed for all bio types except REQ_FLUSH.
1480 * dm_accept_partial_bio informs the dm that the target only wants to process
1481 * additional n_sectors sectors of the bio and the rest of the data should be
1482 * sent in a next bio.
1484 * A diagram that explains the arithmetics:
1485 * +--------------------+---------------+-------+
1487 * +--------------------+---------------+-------+
1489 * <-------------- *tio->len_ptr --------------->
1490 * <------- bi_size ------->
1493 * Region 1 was already iterated over with bio_advance or similar function.
1494 * (it may be empty if the target doesn't use bio_advance)
1495 * Region 2 is the remaining bio size that the target wants to process.
1496 * (it may be empty if region 1 is non-empty, although there is no reason
1498 * The target requires that region 3 is to be sent in the next bio.
1500 * If the target wants to receive multiple copies of the bio (via num_*bios, etc),
1501 * the partially processed part (the sum of regions 1+2) must be the same for all
1502 * copies of the bio.
1504 void dm_accept_partial_bio(struct bio *bio, unsigned n_sectors)
1506 struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone);
1507 unsigned bi_size = bio->bi_iter.bi_size >> SECTOR_SHIFT;
1508 BUG_ON(bio->bi_rw & REQ_FLUSH);
1509 BUG_ON(bi_size > *tio->len_ptr);
1510 BUG_ON(n_sectors > bi_size);
1511 *tio->len_ptr -= bi_size - n_sectors;
1512 bio->bi_iter.bi_size = n_sectors << SECTOR_SHIFT;
1514 EXPORT_SYMBOL_GPL(dm_accept_partial_bio);
1516 static void __map_bio(struct dm_target_io *tio)
1520 struct mapped_device *md;
1521 struct bio *clone = &tio->clone;
1522 struct dm_target *ti = tio->ti;
1524 clone->bi_end_io = clone_endio;
1527 * Map the clone. If r == 0 we don't need to do
1528 * anything, the target has assumed ownership of
1531 atomic_inc(&tio->io->io_count);
1532 sector = clone->bi_iter.bi_sector;
1533 r = ti->type->map(ti, clone);
1534 if (r == DM_MAPIO_REMAPPED) {
1535 /* the bio has been remapped so dispatch it */
1537 trace_block_bio_remap(bdev_get_queue(clone->bi_bdev), clone,
1538 tio->io->bio->bi_bdev->bd_dev, sector);
1540 generic_make_request(clone);
1541 } else if (r < 0 || r == DM_MAPIO_REQUEUE) {
1542 /* error the io and bail out, or requeue it if needed */
1544 dec_pending(tio->io, r);
1546 } else if (r != DM_MAPIO_SUBMITTED) {
1547 DMWARN("unimplemented target map return value: %d", r);
1553 struct mapped_device *md;
1554 struct dm_table *map;
1558 unsigned sector_count;
1561 static void bio_setup_sector(struct bio *bio, sector_t sector, unsigned len)
1563 bio->bi_iter.bi_sector = sector;
1564 bio->bi_iter.bi_size = to_bytes(len);
1568 * Creates a bio that consists of range of complete bvecs.
1570 static void clone_bio(struct dm_target_io *tio, struct bio *bio,
1571 sector_t sector, unsigned len)
1573 struct bio *clone = &tio->clone;
1575 __bio_clone_fast(clone, bio);
1577 if (bio_integrity(bio))
1578 bio_integrity_clone(clone, bio, GFP_NOIO);
1580 bio_advance(clone, to_bytes(sector - clone->bi_iter.bi_sector));
1581 clone->bi_iter.bi_size = to_bytes(len);
1583 if (bio_integrity(bio))
1584 bio_integrity_trim(clone, 0, len);
1587 static struct dm_target_io *alloc_tio(struct clone_info *ci,
1588 struct dm_target *ti,
1589 unsigned target_bio_nr)
1591 struct dm_target_io *tio;
1594 clone = bio_alloc_bioset(GFP_NOIO, 0, ci->md->bs);
1595 tio = container_of(clone, struct dm_target_io, clone);
1599 tio->target_bio_nr = target_bio_nr;
1604 static void __clone_and_map_simple_bio(struct clone_info *ci,
1605 struct dm_target *ti,
1606 unsigned target_bio_nr, unsigned *len)
1608 struct dm_target_io *tio = alloc_tio(ci, ti, target_bio_nr);
1609 struct bio *clone = &tio->clone;
1613 __bio_clone_fast(clone, ci->bio);
1615 bio_setup_sector(clone, ci->sector, *len);
1620 static void __send_duplicate_bios(struct clone_info *ci, struct dm_target *ti,
1621 unsigned num_bios, unsigned *len)
1623 unsigned target_bio_nr;
1625 for (target_bio_nr = 0; target_bio_nr < num_bios; target_bio_nr++)
1626 __clone_and_map_simple_bio(ci, ti, target_bio_nr, len);
1629 static int __send_empty_flush(struct clone_info *ci)
1631 unsigned target_nr = 0;
1632 struct dm_target *ti;
1634 BUG_ON(bio_has_data(ci->bio));
1635 while ((ti = dm_table_get_target(ci->map, target_nr++)))
1636 __send_duplicate_bios(ci, ti, ti->num_flush_bios, NULL);
1641 static void __clone_and_map_data_bio(struct clone_info *ci, struct dm_target *ti,
1642 sector_t sector, unsigned *len)
1644 struct bio *bio = ci->bio;
1645 struct dm_target_io *tio;
1646 unsigned target_bio_nr;
1647 unsigned num_target_bios = 1;
1650 * Does the target want to receive duplicate copies of the bio?
1652 if (bio_data_dir(bio) == WRITE && ti->num_write_bios)
1653 num_target_bios = ti->num_write_bios(ti, bio);
1655 for (target_bio_nr = 0; target_bio_nr < num_target_bios; target_bio_nr++) {
1656 tio = alloc_tio(ci, ti, target_bio_nr);
1658 clone_bio(tio, bio, sector, *len);
1663 typedef unsigned (*get_num_bios_fn)(struct dm_target *ti);
1665 static unsigned get_num_discard_bios(struct dm_target *ti)
1667 return ti->num_discard_bios;
1670 static unsigned get_num_write_same_bios(struct dm_target *ti)
1672 return ti->num_write_same_bios;
1675 typedef bool (*is_split_required_fn)(struct dm_target *ti);
1677 static bool is_split_required_for_discard(struct dm_target *ti)
1679 return ti->split_discard_bios;
1682 static int __send_changing_extent_only(struct clone_info *ci,
1683 get_num_bios_fn get_num_bios,
1684 is_split_required_fn is_split_required)
1686 struct dm_target *ti;
1691 ti = dm_table_find_target(ci->map, ci->sector);
1692 if (!dm_target_is_valid(ti))
1696 * Even though the device advertised support for this type of
1697 * request, that does not mean every target supports it, and
1698 * reconfiguration might also have changed that since the
1699 * check was performed.
1701 num_bios = get_num_bios ? get_num_bios(ti) : 0;
1705 if (is_split_required && !is_split_required(ti))
1706 len = min((sector_t)ci->sector_count, max_io_len_target_boundary(ci->sector, ti));
1708 len = min((sector_t)ci->sector_count, max_io_len(ci->sector, ti));
1710 __send_duplicate_bios(ci, ti, num_bios, &len);
1713 } while (ci->sector_count -= len);
1718 static int __send_discard(struct clone_info *ci)
1720 return __send_changing_extent_only(ci, get_num_discard_bios,
1721 is_split_required_for_discard);
1724 static int __send_write_same(struct clone_info *ci)
1726 return __send_changing_extent_only(ci, get_num_write_same_bios, NULL);
1730 * Select the correct strategy for processing a non-flush bio.
1732 static int __split_and_process_non_flush(struct clone_info *ci)
1734 struct bio *bio = ci->bio;
1735 struct dm_target *ti;
1738 if (unlikely(bio->bi_rw & REQ_DISCARD))
1739 return __send_discard(ci);
1740 else if (unlikely(bio->bi_rw & REQ_WRITE_SAME))
1741 return __send_write_same(ci);
1743 ti = dm_table_find_target(ci->map, ci->sector);
1744 if (!dm_target_is_valid(ti))
1747 len = min_t(sector_t, max_io_len(ci->sector, ti), ci->sector_count);
1749 __clone_and_map_data_bio(ci, ti, ci->sector, &len);
1752 ci->sector_count -= len;
1758 * Entry point to split a bio into clones and submit them to the targets.
1760 static void __split_and_process_bio(struct mapped_device *md,
1761 struct dm_table *map, struct bio *bio)
1763 struct clone_info ci;
1766 if (unlikely(!map)) {
1773 ci.io = alloc_io(md);
1775 atomic_set(&ci.io->io_count, 1);
1778 spin_lock_init(&ci.io->endio_lock);
1779 ci.sector = bio->bi_iter.bi_sector;
1781 start_io_acct(ci.io);
1783 if (bio->bi_rw & REQ_FLUSH) {
1784 ci.bio = &ci.md->flush_bio;
1785 ci.sector_count = 0;
1786 error = __send_empty_flush(&ci);
1787 /* dec_pending submits any data associated with flush */
1790 ci.sector_count = bio_sectors(bio);
1791 while (ci.sector_count && !error)
1792 error = __split_and_process_non_flush(&ci);
1795 /* drop the extra reference count */
1796 dec_pending(ci.io, error);
1798 /*-----------------------------------------------------------------
1800 *---------------------------------------------------------------*/
1803 * The request function that just remaps the bio built up by
1806 static blk_qc_t dm_make_request(struct request_queue *q, struct bio *bio)
1808 int rw = bio_data_dir(bio);
1809 struct mapped_device *md = q->queuedata;
1811 struct dm_table *map;
1813 map = dm_get_live_table(md, &srcu_idx);
1815 generic_start_io_acct(rw, bio_sectors(bio), &dm_disk(md)->part0);
1817 /* if we're suspended, we have to queue this io for later */
1818 if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))) {
1819 dm_put_live_table(md, srcu_idx);
1821 if (bio_rw(bio) != READA)
1825 return BLK_QC_T_NONE;
1828 __split_and_process_bio(md, map, bio);
1829 dm_put_live_table(md, srcu_idx);
1830 return BLK_QC_T_NONE;
1833 int dm_request_based(struct mapped_device *md)
1835 return blk_queue_stackable(md->queue);
1838 static void dm_dispatch_clone_request(struct request *clone, struct request *rq)
1842 if (blk_queue_io_stat(clone->q))
1843 clone->cmd_flags |= REQ_IO_STAT;
1845 clone->start_time = jiffies;
1846 r = blk_insert_cloned_request(clone->q, clone);
1848 /* must complete clone in terms of original request */
1849 dm_complete_request(rq, r);
1852 static int dm_rq_bio_constructor(struct bio *bio, struct bio *bio_orig,
1855 struct dm_rq_target_io *tio = data;
1856 struct dm_rq_clone_bio_info *info =
1857 container_of(bio, struct dm_rq_clone_bio_info, clone);
1859 info->orig = bio_orig;
1861 bio->bi_end_io = end_clone_bio;
1866 static int setup_clone(struct request *clone, struct request *rq,
1867 struct dm_rq_target_io *tio, gfp_t gfp_mask)
1871 r = blk_rq_prep_clone(clone, rq, tio->md->bs, gfp_mask,
1872 dm_rq_bio_constructor, tio);
1876 clone->cmd = rq->cmd;
1877 clone->cmd_len = rq->cmd_len;
1878 clone->sense = rq->sense;
1879 clone->end_io = end_clone_request;
1880 clone->end_io_data = tio;
1887 static struct request *clone_old_rq(struct request *rq, struct mapped_device *md,
1888 struct dm_rq_target_io *tio, gfp_t gfp_mask)
1891 * Create clone for use with .request_fn request_queue
1893 struct request *clone;
1895 clone = alloc_old_clone_request(md, gfp_mask);
1899 blk_rq_init(NULL, clone);
1900 if (setup_clone(clone, rq, tio, gfp_mask)) {
1902 free_old_clone_request(md, clone);
1909 static void map_tio_request(struct kthread_work *work);
1911 static void init_tio(struct dm_rq_target_io *tio, struct request *rq,
1912 struct mapped_device *md)
1920 * Avoid initializing info for blk-mq; it passes
1921 * target-specific data through info.ptr
1922 * (see: dm_mq_init_request)
1924 if (!md->init_tio_pdu)
1925 memset(&tio->info, 0, sizeof(tio->info));
1926 if (md->kworker_task)
1927 init_kthread_work(&tio->work, map_tio_request);
1930 static struct dm_rq_target_io *dm_old_prep_tio(struct request *rq,
1931 struct mapped_device *md,
1934 struct dm_rq_target_io *tio;
1936 struct dm_table *table;
1938 tio = alloc_old_rq_tio(md, gfp_mask);
1942 init_tio(tio, rq, md);
1944 table = dm_get_live_table(md, &srcu_idx);
1946 * Must clone a request if this .request_fn DM device
1947 * is stacked on .request_fn device(s).
1949 if (!dm_table_mq_request_based(table)) {
1950 if (!clone_old_rq(rq, md, tio, gfp_mask)) {
1951 dm_put_live_table(md, srcu_idx);
1952 free_old_rq_tio(tio);
1956 dm_put_live_table(md, srcu_idx);
1962 * Called with the queue lock held.
1964 static int dm_old_prep_fn(struct request_queue *q, struct request *rq)
1966 struct mapped_device *md = q->queuedata;
1967 struct dm_rq_target_io *tio;
1969 if (unlikely(rq->special)) {
1970 DMWARN("Already has something in rq->special.");
1971 return BLKPREP_KILL;
1974 tio = dm_old_prep_tio(rq, md, GFP_ATOMIC);
1976 return BLKPREP_DEFER;
1979 rq->cmd_flags |= REQ_DONTPREP;
1986 * 0 : the request has been processed
1987 * DM_MAPIO_REQUEUE : the original request needs to be requeued
1988 * < 0 : the request was completed due to failure
1990 static int map_request(struct dm_rq_target_io *tio, struct request *rq,
1991 struct mapped_device *md)
1994 struct dm_target *ti = tio->ti;
1995 struct request *clone = NULL;
1999 r = ti->type->map_rq(ti, clone, &tio->info);
2001 r = ti->type->clone_and_map_rq(ti, rq, &tio->info, &clone);
2003 /* The target wants to complete the I/O */
2004 dm_kill_unmapped_request(rq, r);
2007 if (r != DM_MAPIO_REMAPPED)
2009 if (setup_clone(clone, rq, tio, GFP_ATOMIC)) {
2011 ti->type->release_clone_rq(clone);
2012 return DM_MAPIO_REQUEUE;
2017 case DM_MAPIO_SUBMITTED:
2018 /* The target has taken the I/O to submit by itself later */
2020 case DM_MAPIO_REMAPPED:
2021 /* The target has remapped the I/O so dispatch it */
2022 trace_block_rq_remap(clone->q, clone, disk_devt(dm_disk(md)),
2024 dm_dispatch_clone_request(clone, rq);
2026 case DM_MAPIO_REQUEUE:
2027 /* The target wants to requeue the I/O */
2028 dm_requeue_original_request(md, tio->orig);
2032 DMWARN("unimplemented target map return value: %d", r);
2036 /* The target wants to complete the I/O */
2037 dm_kill_unmapped_request(rq, r);
2044 static void map_tio_request(struct kthread_work *work)
2046 struct dm_rq_target_io *tio = container_of(work, struct dm_rq_target_io, work);
2047 struct request *rq = tio->orig;
2048 struct mapped_device *md = tio->md;
2050 if (map_request(tio, rq, md) == DM_MAPIO_REQUEUE)
2051 dm_requeue_original_request(md, rq);
2054 static void dm_start_request(struct mapped_device *md, struct request *orig)
2056 if (!orig->q->mq_ops)
2057 blk_start_request(orig);
2059 blk_mq_start_request(orig);
2060 atomic_inc(&md->pending[rq_data_dir(orig)]);
2062 if (md->seq_rq_merge_deadline_usecs) {
2063 md->last_rq_pos = rq_end_sector(orig);
2064 md->last_rq_rw = rq_data_dir(orig);
2065 md->last_rq_start_time = ktime_get();
2068 if (unlikely(dm_stats_used(&md->stats))) {
2069 struct dm_rq_target_io *tio = tio_from_request(orig);
2070 tio->duration_jiffies = jiffies;
2071 tio->n_sectors = blk_rq_sectors(orig);
2072 dm_stats_account_io(&md->stats, orig->cmd_flags, blk_rq_pos(orig),
2073 tio->n_sectors, false, 0, &tio->stats_aux);
2077 * Hold the md reference here for the in-flight I/O.
2078 * We can't rely on the reference count by device opener,
2079 * because the device may be closed during the request completion
2080 * when all bios are completed.
2081 * See the comment in rq_completed() too.
2086 #define MAX_SEQ_RQ_MERGE_DEADLINE_USECS 100000
2088 ssize_t dm_attr_rq_based_seq_io_merge_deadline_show(struct mapped_device *md, char *buf)
2090 return sprintf(buf, "%u\n", md->seq_rq_merge_deadline_usecs);
2093 ssize_t dm_attr_rq_based_seq_io_merge_deadline_store(struct mapped_device *md,
2094 const char *buf, size_t count)
2098 if (!dm_request_based(md) || md->use_blk_mq)
2101 if (kstrtouint(buf, 10, &deadline))
2104 if (deadline > MAX_SEQ_RQ_MERGE_DEADLINE_USECS)
2105 deadline = MAX_SEQ_RQ_MERGE_DEADLINE_USECS;
2107 md->seq_rq_merge_deadline_usecs = deadline;
2112 static bool dm_request_peeked_before_merge_deadline(struct mapped_device *md)
2114 ktime_t kt_deadline;
2116 if (!md->seq_rq_merge_deadline_usecs)
2119 kt_deadline = ns_to_ktime((u64)md->seq_rq_merge_deadline_usecs * NSEC_PER_USEC);
2120 kt_deadline = ktime_add_safe(md->last_rq_start_time, kt_deadline);
2122 return !ktime_after(ktime_get(), kt_deadline);
2126 * q->request_fn for request-based dm.
2127 * Called with the queue lock held.
2129 static void dm_request_fn(struct request_queue *q)
2131 struct mapped_device *md = q->queuedata;
2132 struct dm_target *ti = md->immutable_target;
2134 struct dm_rq_target_io *tio;
2137 if (unlikely(!ti)) {
2139 struct dm_table *map = dm_get_live_table(md, &srcu_idx);
2141 ti = dm_table_find_target(map, pos);
2142 dm_put_live_table(md, srcu_idx);
2146 * For suspend, check blk_queue_stopped() and increment
2147 * ->pending within a single queue_lock not to increment the
2148 * number of in-flight I/Os after the queue is stopped in
2151 while (!blk_queue_stopped(q)) {
2152 rq = blk_peek_request(q);
2156 /* always use block 0 to find the target for flushes for now */
2158 if (!(rq->cmd_flags & REQ_FLUSH))
2159 pos = blk_rq_pos(rq);
2161 if ((dm_request_peeked_before_merge_deadline(md) &&
2162 md_in_flight(md) && rq->bio && rq->bio->bi_vcnt == 1 &&
2163 md->last_rq_pos == pos && md->last_rq_rw == rq_data_dir(rq)) ||
2164 (ti->type->busy && ti->type->busy(ti))) {
2165 blk_delay_queue(q, HZ / 100);
2169 dm_start_request(md, rq);
2171 tio = tio_from_request(rq);
2172 /* Establish tio->ti before queuing work (map_tio_request) */
2174 queue_kthread_work(&md->kworker, &tio->work);
2175 BUG_ON(!irqs_disabled());
2179 static int dm_any_congested(void *congested_data, int bdi_bits)
2182 struct mapped_device *md = congested_data;
2183 struct dm_table *map;
2185 if (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
2186 if (dm_request_based(md)) {
2188 * With request-based DM we only need to check the
2189 * top-level queue for congestion.
2191 r = md->queue->backing_dev_info.wb.state & bdi_bits;
2193 map = dm_get_live_table_fast(md);
2195 r = dm_table_any_congested(map, bdi_bits);
2196 dm_put_live_table_fast(md);
2203 /*-----------------------------------------------------------------
2204 * An IDR is used to keep track of allocated minor numbers.
2205 *---------------------------------------------------------------*/
2206 static void free_minor(int minor)
2208 spin_lock(&_minor_lock);
2209 idr_remove(&_minor_idr, minor);
2210 spin_unlock(&_minor_lock);
2214 * See if the device with a specific minor # is free.
2216 static int specific_minor(int minor)
2220 if (minor >= (1 << MINORBITS))
2223 idr_preload(GFP_KERNEL);
2224 spin_lock(&_minor_lock);
2226 r = idr_alloc(&_minor_idr, MINOR_ALLOCED, minor, minor + 1, GFP_NOWAIT);
2228 spin_unlock(&_minor_lock);
2231 return r == -ENOSPC ? -EBUSY : r;
2235 static int next_free_minor(int *minor)
2239 idr_preload(GFP_KERNEL);
2240 spin_lock(&_minor_lock);
2242 r = idr_alloc(&_minor_idr, MINOR_ALLOCED, 0, 1 << MINORBITS, GFP_NOWAIT);
2244 spin_unlock(&_minor_lock);
2252 static const struct block_device_operations dm_blk_dops;
2254 static void dm_wq_work(struct work_struct *work);
2256 static void dm_init_md_queue(struct mapped_device *md)
2259 * Request-based dm devices cannot be stacked on top of bio-based dm
2260 * devices. The type of this dm device may not have been decided yet.
2261 * The type is decided at the first table loading time.
2262 * To prevent problematic device stacking, clear the queue flag
2263 * for request stacking support until then.
2265 * This queue is new, so no concurrency on the queue_flags.
2267 queue_flag_clear_unlocked(QUEUE_FLAG_STACKABLE, md->queue);
2270 * Initialize data that will only be used by a non-blk-mq DM queue
2271 * - must do so here (in alloc_dev callchain) before queue is used
2273 md->queue->queuedata = md;
2274 md->queue->backing_dev_info.congested_data = md;
2277 static void dm_init_normal_md_queue(struct mapped_device *md)
2279 md->use_blk_mq = false;
2280 dm_init_md_queue(md);
2283 * Initialize aspects of queue that aren't relevant for blk-mq
2285 md->queue->backing_dev_info.congested_fn = dm_any_congested;
2286 blk_queue_bounce_limit(md->queue, BLK_BOUNCE_ANY);
2289 static void cleanup_mapped_device(struct mapped_device *md)
2292 destroy_workqueue(md->wq);
2293 if (md->kworker_task)
2294 kthread_stop(md->kworker_task);
2295 mempool_destroy(md->io_pool);
2296 mempool_destroy(md->rq_pool);
2298 bioset_free(md->bs);
2300 cleanup_srcu_struct(&md->io_barrier);
2303 spin_lock(&_minor_lock);
2304 md->disk->private_data = NULL;
2305 spin_unlock(&_minor_lock);
2306 del_gendisk(md->disk);
2311 blk_cleanup_queue(md->queue);
2320 * Allocate and initialise a blank device with a given minor.
2322 static struct mapped_device *alloc_dev(int minor)
2324 int r, numa_node_id = dm_get_numa_node();
2325 struct mapped_device *md;
2328 md = kzalloc_node(sizeof(*md), GFP_KERNEL, numa_node_id);
2330 DMWARN("unable to allocate device, out of memory.");
2334 if (!try_module_get(THIS_MODULE))
2335 goto bad_module_get;
2337 /* get a minor number for the dev */
2338 if (minor == DM_ANY_MINOR)
2339 r = next_free_minor(&minor);
2341 r = specific_minor(minor);
2345 r = init_srcu_struct(&md->io_barrier);
2347 goto bad_io_barrier;
2349 md->numa_node_id = numa_node_id;
2350 md->use_blk_mq = use_blk_mq;
2351 md->init_tio_pdu = false;
2352 md->type = DM_TYPE_NONE;
2353 mutex_init(&md->suspend_lock);
2354 mutex_init(&md->type_lock);
2355 mutex_init(&md->table_devices_lock);
2356 spin_lock_init(&md->deferred_lock);
2357 atomic_set(&md->holders, 1);
2358 atomic_set(&md->open_count, 0);
2359 atomic_set(&md->event_nr, 0);
2360 atomic_set(&md->uevent_seq, 0);
2361 INIT_LIST_HEAD(&md->uevent_list);
2362 INIT_LIST_HEAD(&md->table_devices);
2363 spin_lock_init(&md->uevent_lock);
2365 md->queue = blk_alloc_queue_node(GFP_KERNEL, numa_node_id);
2369 dm_init_md_queue(md);
2371 md->disk = alloc_disk_node(1, numa_node_id);
2375 atomic_set(&md->pending[0], 0);
2376 atomic_set(&md->pending[1], 0);
2377 init_waitqueue_head(&md->wait);
2378 INIT_WORK(&md->work, dm_wq_work);
2379 init_waitqueue_head(&md->eventq);
2380 init_completion(&md->kobj_holder.completion);
2381 md->kworker_task = NULL;
2383 md->disk->major = _major;
2384 md->disk->first_minor = minor;
2385 md->disk->fops = &dm_blk_dops;
2386 md->disk->queue = md->queue;
2387 md->disk->private_data = md;
2388 sprintf(md->disk->disk_name, "dm-%d", minor);
2390 format_dev_t(md->name, MKDEV(_major, minor));
2392 md->wq = alloc_workqueue("kdmflush", WQ_MEM_RECLAIM, 0);
2396 md->bdev = bdget_disk(md->disk, 0);
2400 bio_init(&md->flush_bio);
2401 md->flush_bio.bi_bdev = md->bdev;
2402 md->flush_bio.bi_rw = WRITE_FLUSH;
2404 dm_stats_init(&md->stats);
2406 /* Populate the mapping, nobody knows we exist yet */
2407 spin_lock(&_minor_lock);
2408 old_md = idr_replace(&_minor_idr, md, minor);
2409 spin_unlock(&_minor_lock);
2411 BUG_ON(old_md != MINOR_ALLOCED);
2416 cleanup_mapped_device(md);
2420 module_put(THIS_MODULE);
2426 static void unlock_fs(struct mapped_device *md);
2428 static void free_dev(struct mapped_device *md)
2430 int minor = MINOR(disk_devt(md->disk));
2434 cleanup_mapped_device(md);
2436 blk_mq_free_tag_set(md->tag_set);
2440 free_table_devices(&md->table_devices);
2441 dm_stats_cleanup(&md->stats);
2444 module_put(THIS_MODULE);
2448 static void __bind_mempools(struct mapped_device *md, struct dm_table *t)
2450 struct dm_md_mempools *p = dm_table_get_md_mempools(t);
2453 /* The md already has necessary mempools. */
2454 if (dm_table_get_type(t) == DM_TYPE_BIO_BASED) {
2456 * Reload bioset because front_pad may have changed
2457 * because a different table was loaded.
2459 bioset_free(md->bs);
2464 * There's no need to reload with request-based dm
2465 * because the size of front_pad doesn't change.
2466 * Note for future: If you are to reload bioset,
2467 * prep-ed requests in the queue may refer
2468 * to bio from the old bioset, so you must walk
2469 * through the queue to unprep.
2474 BUG_ON(!p || md->io_pool || md->rq_pool || md->bs);
2476 md->io_pool = p->io_pool;
2478 md->rq_pool = p->rq_pool;
2484 /* mempool bind completed, no longer need any mempools in the table */
2485 dm_table_free_md_mempools(t);
2489 * Bind a table to the device.
2491 static void event_callback(void *context)
2493 unsigned long flags;
2495 struct mapped_device *md = (struct mapped_device *) context;
2497 spin_lock_irqsave(&md->uevent_lock, flags);
2498 list_splice_init(&md->uevent_list, &uevents);
2499 spin_unlock_irqrestore(&md->uevent_lock, flags);
2501 dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj);
2503 atomic_inc(&md->event_nr);
2504 wake_up(&md->eventq);
2508 * Protected by md->suspend_lock obtained by dm_swap_table().
2510 static void __set_size(struct mapped_device *md, sector_t size)
2512 set_capacity(md->disk, size);
2514 i_size_write(md->bdev->bd_inode, (loff_t)size << SECTOR_SHIFT);
2518 * Returns old map, which caller must destroy.
2520 static struct dm_table *__bind(struct mapped_device *md, struct dm_table *t,
2521 struct queue_limits *limits)
2523 struct dm_table *old_map;
2524 struct request_queue *q = md->queue;
2527 size = dm_table_get_size(t);
2530 * Wipe any geometry if the size of the table changed.
2532 if (size != dm_get_size(md))
2533 memset(&md->geometry, 0, sizeof(md->geometry));
2535 __set_size(md, size);
2537 dm_table_event_callback(t, event_callback, md);
2540 * The queue hasn't been stopped yet, if the old table type wasn't
2541 * for request-based during suspension. So stop it to prevent
2542 * I/O mapping before resume.
2543 * This must be done before setting the queue restrictions,
2544 * because request-based dm may be run just after the setting.
2546 if (dm_table_request_based(t)) {
2549 * Leverage the fact that request-based DM targets are
2550 * immutable singletons and establish md->immutable_target
2551 * - used to optimize both dm_request_fn and dm_mq_queue_rq
2553 md->immutable_target = dm_table_get_immutable_target(t);
2556 __bind_mempools(md, t);
2558 old_map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2559 rcu_assign_pointer(md->map, (void *)t);
2560 md->immutable_target_type = dm_table_get_immutable_target_type(t);
2562 dm_table_set_restrictions(t, q, limits);
2570 * Returns unbound table for the caller to free.
2572 static struct dm_table *__unbind(struct mapped_device *md)
2574 struct dm_table *map = rcu_dereference_protected(md->map, 1);
2579 dm_table_event_callback(map, NULL, NULL);
2580 RCU_INIT_POINTER(md->map, NULL);
2587 * Constructor for a new device.
2589 int dm_create(int minor, struct mapped_device **result)
2591 struct mapped_device *md;
2593 md = alloc_dev(minor);
2604 * Functions to manage md->type.
2605 * All are required to hold md->type_lock.
2607 void dm_lock_md_type(struct mapped_device *md)
2609 mutex_lock(&md->type_lock);
2612 void dm_unlock_md_type(struct mapped_device *md)
2614 mutex_unlock(&md->type_lock);
2617 void dm_set_md_type(struct mapped_device *md, unsigned type)
2619 BUG_ON(!mutex_is_locked(&md->type_lock));
2623 unsigned dm_get_md_type(struct mapped_device *md)
2628 struct target_type *dm_get_immutable_target_type(struct mapped_device *md)
2630 return md->immutable_target_type;
2634 * The queue_limits are only valid as long as you have a reference
2637 struct queue_limits *dm_get_queue_limits(struct mapped_device *md)
2639 BUG_ON(!atomic_read(&md->holders));
2640 return &md->queue->limits;
2642 EXPORT_SYMBOL_GPL(dm_get_queue_limits);
2644 static void dm_old_init_rq_based_worker_thread(struct mapped_device *md)
2646 /* Initialize the request-based DM worker thread */
2647 init_kthread_worker(&md->kworker);
2648 md->kworker_task = kthread_run(kthread_worker_fn, &md->kworker,
2649 "kdmwork-%s", dm_device_name(md));
2653 * Fully initialize a .request_fn request-based queue.
2655 static int dm_old_init_request_queue(struct mapped_device *md)
2657 /* Fully initialize the queue */
2658 if (!blk_init_allocated_queue(md->queue, dm_request_fn, NULL))
2661 /* disable dm_request_fn's merge heuristic by default */
2662 md->seq_rq_merge_deadline_usecs = 0;
2664 dm_init_normal_md_queue(md);
2665 blk_queue_softirq_done(md->queue, dm_softirq_done);
2666 blk_queue_prep_rq(md->queue, dm_old_prep_fn);
2668 dm_old_init_rq_based_worker_thread(md);
2670 elv_register_queue(md->queue);
2675 static int dm_mq_init_request(void *data, struct request *rq,
2676 unsigned int hctx_idx, unsigned int request_idx,
2677 unsigned int numa_node)
2679 struct mapped_device *md = data;
2680 struct dm_rq_target_io *tio = blk_mq_rq_to_pdu(rq);
2683 * Must initialize md member of tio, otherwise it won't
2684 * be available in dm_mq_queue_rq.
2688 if (md->init_tio_pdu) {
2689 /* target-specific per-io data is immediately after the tio */
2690 tio->info.ptr = tio + 1;
2696 static int dm_mq_queue_rq(struct blk_mq_hw_ctx *hctx,
2697 const struct blk_mq_queue_data *bd)
2699 struct request *rq = bd->rq;
2700 struct dm_rq_target_io *tio = blk_mq_rq_to_pdu(rq);
2701 struct mapped_device *md = tio->md;
2702 struct dm_target *ti = md->immutable_target;
2704 if (unlikely(!ti)) {
2706 struct dm_table *map = dm_get_live_table(md, &srcu_idx);
2708 ti = dm_table_find_target(map, 0);
2709 dm_put_live_table(md, srcu_idx);
2712 if (ti->type->busy && ti->type->busy(ti))
2713 return BLK_MQ_RQ_QUEUE_BUSY;
2715 dm_start_request(md, rq);
2717 /* Init tio using md established in .init_request */
2718 init_tio(tio, rq, md);
2721 * Establish tio->ti before queuing work (map_tio_request)
2722 * or making direct call to map_request().
2726 /* Direct call is fine since .queue_rq allows allocations */
2727 if (map_request(tio, rq, md) == DM_MAPIO_REQUEUE) {
2728 /* Undo dm_start_request() before requeuing */
2729 rq_end_stats(md, rq);
2730 rq_completed(md, rq_data_dir(rq), false);
2731 return BLK_MQ_RQ_QUEUE_BUSY;
2734 return BLK_MQ_RQ_QUEUE_OK;
2737 static struct blk_mq_ops dm_mq_ops = {
2738 .queue_rq = dm_mq_queue_rq,
2739 .map_queue = blk_mq_map_queue,
2740 .complete = dm_softirq_done,
2741 .init_request = dm_mq_init_request,
2744 static int dm_mq_init_request_queue(struct mapped_device *md,
2745 struct dm_target *immutable_tgt)
2747 struct request_queue *q;
2750 if (dm_get_md_type(md) == DM_TYPE_REQUEST_BASED) {
2751 DMERR("request-based dm-mq may only be stacked on blk-mq device(s)");
2755 md->tag_set = kzalloc_node(sizeof(struct blk_mq_tag_set), GFP_KERNEL, md->numa_node_id);
2759 md->tag_set->ops = &dm_mq_ops;
2760 md->tag_set->queue_depth = dm_get_blk_mq_queue_depth();
2761 md->tag_set->numa_node = md->numa_node_id;
2762 md->tag_set->flags = BLK_MQ_F_SHOULD_MERGE | BLK_MQ_F_SG_MERGE;
2763 md->tag_set->nr_hw_queues = dm_get_blk_mq_nr_hw_queues();
2764 md->tag_set->driver_data = md;
2766 md->tag_set->cmd_size = sizeof(struct dm_rq_target_io);
2767 if (immutable_tgt && immutable_tgt->per_io_data_size) {
2768 /* any target-specific per-io data is immediately after the tio */
2769 md->tag_set->cmd_size += immutable_tgt->per_io_data_size;
2770 md->init_tio_pdu = true;
2773 err = blk_mq_alloc_tag_set(md->tag_set);
2775 goto out_kfree_tag_set;
2777 q = blk_mq_init_allocated_queue(md->tag_set, md->queue);
2782 dm_init_md_queue(md);
2784 /* backfill 'mq' sysfs registration normally done in blk_register_queue */
2785 blk_mq_register_disk(md->disk);
2790 blk_mq_free_tag_set(md->tag_set);
2797 static unsigned filter_md_type(unsigned type, struct mapped_device *md)
2799 if (type == DM_TYPE_BIO_BASED)
2802 return !md->use_blk_mq ? DM_TYPE_REQUEST_BASED : DM_TYPE_MQ_REQUEST_BASED;
2806 * Setup the DM device's queue based on md's type
2808 int dm_setup_md_queue(struct mapped_device *md, struct dm_table *t)
2811 unsigned md_type = filter_md_type(dm_get_md_type(md), md);
2814 case DM_TYPE_REQUEST_BASED:
2815 r = dm_old_init_request_queue(md);
2817 DMERR("Cannot initialize queue for request-based mapped device");
2821 case DM_TYPE_MQ_REQUEST_BASED:
2822 r = dm_mq_init_request_queue(md, dm_table_get_immutable_target(t));
2824 DMERR("Cannot initialize queue for request-based dm-mq mapped device");
2828 case DM_TYPE_BIO_BASED:
2829 dm_init_normal_md_queue(md);
2830 blk_queue_make_request(md->queue, dm_make_request);
2832 * DM handles splitting bios as needed. Free the bio_split bioset
2833 * since it won't be used (saves 1 process per bio-based DM device).
2835 bioset_free(md->queue->bio_split);
2836 md->queue->bio_split = NULL;
2843 struct mapped_device *dm_get_md(dev_t dev)
2845 struct mapped_device *md;
2846 unsigned minor = MINOR(dev);
2848 if (MAJOR(dev) != _major || minor >= (1 << MINORBITS))
2851 spin_lock(&_minor_lock);
2853 md = idr_find(&_minor_idr, minor);
2855 if ((md == MINOR_ALLOCED ||
2856 (MINOR(disk_devt(dm_disk(md))) != minor) ||
2857 dm_deleting_md(md) ||
2858 test_bit(DMF_FREEING, &md->flags))) {
2866 spin_unlock(&_minor_lock);
2870 EXPORT_SYMBOL_GPL(dm_get_md);
2872 void *dm_get_mdptr(struct mapped_device *md)
2874 return md->interface_ptr;
2877 void dm_set_mdptr(struct mapped_device *md, void *ptr)
2879 md->interface_ptr = ptr;
2882 void dm_get(struct mapped_device *md)
2884 atomic_inc(&md->holders);
2885 BUG_ON(test_bit(DMF_FREEING, &md->flags));
2888 int dm_hold(struct mapped_device *md)
2890 spin_lock(&_minor_lock);
2891 if (test_bit(DMF_FREEING, &md->flags)) {
2892 spin_unlock(&_minor_lock);
2896 spin_unlock(&_minor_lock);
2899 EXPORT_SYMBOL_GPL(dm_hold);
2901 const char *dm_device_name(struct mapped_device *md)
2905 EXPORT_SYMBOL_GPL(dm_device_name);
2907 static void __dm_destroy(struct mapped_device *md, bool wait)
2909 struct dm_table *map;
2914 spin_lock(&_minor_lock);
2915 idr_replace(&_minor_idr, MINOR_ALLOCED, MINOR(disk_devt(dm_disk(md))));
2916 set_bit(DMF_FREEING, &md->flags);
2917 spin_unlock(&_minor_lock);
2919 if (dm_request_based(md) && md->kworker_task)
2920 flush_kthread_worker(&md->kworker);
2923 * Take suspend_lock so that presuspend and postsuspend methods
2924 * do not race with internal suspend.
2926 mutex_lock(&md->suspend_lock);
2927 map = dm_get_live_table(md, &srcu_idx);
2928 if (!dm_suspended_md(md)) {
2929 dm_table_presuspend_targets(map);
2930 dm_table_postsuspend_targets(map);
2932 /* dm_put_live_table must be before msleep, otherwise deadlock is possible */
2933 dm_put_live_table(md, srcu_idx);
2934 mutex_unlock(&md->suspend_lock);
2937 * Rare, but there may be I/O requests still going to complete,
2938 * for example. Wait for all references to disappear.
2939 * No one should increment the reference count of the mapped_device,
2940 * after the mapped_device state becomes DMF_FREEING.
2943 while (atomic_read(&md->holders))
2945 else if (atomic_read(&md->holders))
2946 DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)",
2947 dm_device_name(md), atomic_read(&md->holders));
2950 dm_table_destroy(__unbind(md));
2954 void dm_destroy(struct mapped_device *md)
2956 __dm_destroy(md, true);
2959 void dm_destroy_immediate(struct mapped_device *md)
2961 __dm_destroy(md, false);
2964 void dm_put(struct mapped_device *md)
2966 atomic_dec(&md->holders);
2968 EXPORT_SYMBOL_GPL(dm_put);
2970 static int dm_wait_for_completion(struct mapped_device *md, int interruptible)
2973 DECLARE_WAITQUEUE(wait, current);
2975 add_wait_queue(&md->wait, &wait);
2978 set_current_state(interruptible);
2980 if (!md_in_flight(md))
2983 if (interruptible == TASK_INTERRUPTIBLE &&
2984 signal_pending(current)) {
2991 set_current_state(TASK_RUNNING);
2993 remove_wait_queue(&md->wait, &wait);
2999 * Process the deferred bios
3001 static void dm_wq_work(struct work_struct *work)
3003 struct mapped_device *md = container_of(work, struct mapped_device,
3007 struct dm_table *map;
3009 map = dm_get_live_table(md, &srcu_idx);
3011 while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
3012 spin_lock_irq(&md->deferred_lock);
3013 c = bio_list_pop(&md->deferred);
3014 spin_unlock_irq(&md->deferred_lock);
3019 if (dm_request_based(md))
3020 generic_make_request(c);
3022 __split_and_process_bio(md, map, c);
3025 dm_put_live_table(md, srcu_idx);
3028 static void dm_queue_flush(struct mapped_device *md)
3030 clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
3031 smp_mb__after_atomic();
3032 queue_work(md->wq, &md->work);
3036 * Swap in a new table, returning the old one for the caller to destroy.
3038 struct dm_table *dm_swap_table(struct mapped_device *md, struct dm_table *table)
3040 struct dm_table *live_map = NULL, *map = ERR_PTR(-EINVAL);
3041 struct queue_limits limits;
3044 mutex_lock(&md->suspend_lock);
3046 /* device must be suspended */
3047 if (!dm_suspended_md(md))
3051 * If the new table has no data devices, retain the existing limits.
3052 * This helps multipath with queue_if_no_path if all paths disappear,
3053 * then new I/O is queued based on these limits, and then some paths
3056 if (dm_table_has_no_data_devices(table)) {
3057 live_map = dm_get_live_table_fast(md);
3059 limits = md->queue->limits;
3060 dm_put_live_table_fast(md);
3064 r = dm_calculate_queue_limits(table, &limits);
3071 map = __bind(md, table, &limits);
3074 mutex_unlock(&md->suspend_lock);
3079 * Functions to lock and unlock any filesystem running on the
3082 static int lock_fs(struct mapped_device *md)
3086 WARN_ON(md->frozen_sb);
3088 md->frozen_sb = freeze_bdev(md->bdev);
3089 if (IS_ERR(md->frozen_sb)) {
3090 r = PTR_ERR(md->frozen_sb);
3091 md->frozen_sb = NULL;
3095 set_bit(DMF_FROZEN, &md->flags);
3100 static void unlock_fs(struct mapped_device *md)
3102 if (!test_bit(DMF_FROZEN, &md->flags))
3105 thaw_bdev(md->bdev, md->frozen_sb);
3106 md->frozen_sb = NULL;
3107 clear_bit(DMF_FROZEN, &md->flags);
3111 * If __dm_suspend returns 0, the device is completely quiescent
3112 * now. There is no request-processing activity. All new requests
3113 * are being added to md->deferred list.
3115 * Caller must hold md->suspend_lock
3117 static int __dm_suspend(struct mapped_device *md, struct dm_table *map,
3118 unsigned suspend_flags, int interruptible)
3120 bool do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG;
3121 bool noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG;
3125 * DMF_NOFLUSH_SUSPENDING must be set before presuspend.
3126 * This flag is cleared before dm_suspend returns.
3129 set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
3132 * This gets reverted if there's an error later and the targets
3133 * provide the .presuspend_undo hook.
3135 dm_table_presuspend_targets(map);
3138 * Flush I/O to the device.
3139 * Any I/O submitted after lock_fs() may not be flushed.
3140 * noflush takes precedence over do_lockfs.
3141 * (lock_fs() flushes I/Os and waits for them to complete.)
3143 if (!noflush && do_lockfs) {
3146 dm_table_presuspend_undo_targets(map);
3152 * Here we must make sure that no processes are submitting requests
3153 * to target drivers i.e. no one may be executing
3154 * __split_and_process_bio. This is called from dm_request and
3157 * To get all processes out of __split_and_process_bio in dm_request,
3158 * we take the write lock. To prevent any process from reentering
3159 * __split_and_process_bio from dm_request and quiesce the thread
3160 * (dm_wq_work), we set BMF_BLOCK_IO_FOR_SUSPEND and call
3161 * flush_workqueue(md->wq).
3163 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
3165 synchronize_srcu(&md->io_barrier);
3168 * Stop md->queue before flushing md->wq in case request-based
3169 * dm defers requests to md->wq from md->queue.
3171 if (dm_request_based(md)) {
3172 dm_stop_queue(md->queue);
3173 if (md->kworker_task)
3174 flush_kthread_worker(&md->kworker);
3177 flush_workqueue(md->wq);
3180 * At this point no more requests are entering target request routines.
3181 * We call dm_wait_for_completion to wait for all existing requests
3184 r = dm_wait_for_completion(md, interruptible);
3187 clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
3189 synchronize_srcu(&md->io_barrier);
3191 /* were we interrupted ? */
3195 if (dm_request_based(md))
3196 dm_start_queue(md->queue);
3199 dm_table_presuspend_undo_targets(map);
3200 /* pushback list is already flushed, so skip flush */
3207 * We need to be able to change a mapping table under a mounted
3208 * filesystem. For example we might want to move some data in
3209 * the background. Before the table can be swapped with
3210 * dm_bind_table, dm_suspend must be called to flush any in
3211 * flight bios and ensure that any further io gets deferred.
3214 * Suspend mechanism in request-based dm.
3216 * 1. Flush all I/Os by lock_fs() if needed.
3217 * 2. Stop dispatching any I/O by stopping the request_queue.
3218 * 3. Wait for all in-flight I/Os to be completed or requeued.
3220 * To abort suspend, start the request_queue.
3222 int dm_suspend(struct mapped_device *md, unsigned suspend_flags)
3224 struct dm_table *map = NULL;
3228 mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING);
3230 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));
3246 r = __dm_suspend(md, map, suspend_flags, TASK_INTERRUPTIBLE);
3250 set_bit(DMF_SUSPENDED, &md->flags);
3252 dm_table_postsuspend_targets(map);
3255 mutex_unlock(&md->suspend_lock);
3259 static int __dm_resume(struct mapped_device *md, struct dm_table *map)
3262 int r = dm_table_resume_targets(map);
3270 * Flushing deferred I/Os must be done after targets are resumed
3271 * so that mapping of targets can work correctly.
3272 * Request-based dm is queueing the deferred I/Os in its request_queue.
3274 if (dm_request_based(md))
3275 dm_start_queue(md->queue);
3282 int dm_resume(struct mapped_device *md)
3285 struct dm_table *map = NULL;
3288 mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING);
3290 if (!dm_suspended_md(md))
3293 if (dm_suspended_internally_md(md)) {
3294 /* already internally suspended, wait for internal resume */
3295 mutex_unlock(&md->suspend_lock);
3296 r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE);
3302 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
3303 if (!map || !dm_table_get_size(map))
3306 r = __dm_resume(md, map);
3310 clear_bit(DMF_SUSPENDED, &md->flags);
3314 mutex_unlock(&md->suspend_lock);
3320 * Internal suspend/resume works like userspace-driven suspend. It waits
3321 * until all bios finish and prevents issuing new bios to the target drivers.
3322 * It may be used only from the kernel.
3325 static void __dm_internal_suspend(struct mapped_device *md, unsigned suspend_flags)
3327 struct dm_table *map = NULL;
3329 if (md->internal_suspend_count++)
3330 return; /* nested internal suspend */
3332 if (dm_suspended_md(md)) {
3333 set_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
3334 return; /* nest suspend */
3337 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
3340 * Using TASK_UNINTERRUPTIBLE because only NOFLUSH internal suspend is
3341 * supported. Properly supporting a TASK_INTERRUPTIBLE internal suspend
3342 * would require changing .presuspend to return an error -- avoid this
3343 * until there is a need for more elaborate variants of internal suspend.
3345 (void) __dm_suspend(md, map, suspend_flags, TASK_UNINTERRUPTIBLE);
3347 set_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
3349 dm_table_postsuspend_targets(map);
3352 static void __dm_internal_resume(struct mapped_device *md)
3354 BUG_ON(!md->internal_suspend_count);
3356 if (--md->internal_suspend_count)
3357 return; /* resume from nested internal suspend */
3359 if (dm_suspended_md(md))
3360 goto done; /* resume from nested suspend */
3363 * NOTE: existing callers don't need to call dm_table_resume_targets
3364 * (which may fail -- so best to avoid it for now by passing NULL map)
3366 (void) __dm_resume(md, NULL);
3369 clear_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
3370 smp_mb__after_atomic();
3371 wake_up_bit(&md->flags, DMF_SUSPENDED_INTERNALLY);
3374 void dm_internal_suspend_noflush(struct mapped_device *md)
3376 mutex_lock(&md->suspend_lock);
3377 __dm_internal_suspend(md, DM_SUSPEND_NOFLUSH_FLAG);
3378 mutex_unlock(&md->suspend_lock);
3380 EXPORT_SYMBOL_GPL(dm_internal_suspend_noflush);
3382 void dm_internal_resume(struct mapped_device *md)
3384 mutex_lock(&md->suspend_lock);
3385 __dm_internal_resume(md);
3386 mutex_unlock(&md->suspend_lock);
3388 EXPORT_SYMBOL_GPL(dm_internal_resume);
3391 * Fast variants of internal suspend/resume hold md->suspend_lock,
3392 * which prevents interaction with userspace-driven suspend.
3395 void dm_internal_suspend_fast(struct mapped_device *md)
3397 mutex_lock(&md->suspend_lock);
3398 if (dm_suspended_md(md) || dm_suspended_internally_md(md))
3401 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
3402 synchronize_srcu(&md->io_barrier);
3403 flush_workqueue(md->wq);
3404 dm_wait_for_completion(md, TASK_UNINTERRUPTIBLE);
3406 EXPORT_SYMBOL_GPL(dm_internal_suspend_fast);
3408 void dm_internal_resume_fast(struct mapped_device *md)
3410 if (dm_suspended_md(md) || dm_suspended_internally_md(md))
3416 mutex_unlock(&md->suspend_lock);
3418 EXPORT_SYMBOL_GPL(dm_internal_resume_fast);
3420 /*-----------------------------------------------------------------
3421 * Event notification.
3422 *---------------------------------------------------------------*/
3423 int dm_kobject_uevent(struct mapped_device *md, enum kobject_action action,
3426 char udev_cookie[DM_COOKIE_LENGTH];
3427 char *envp[] = { udev_cookie, NULL };
3430 return kobject_uevent(&disk_to_dev(md->disk)->kobj, action);
3432 snprintf(udev_cookie, DM_COOKIE_LENGTH, "%s=%u",
3433 DM_COOKIE_ENV_VAR_NAME, cookie);
3434 return kobject_uevent_env(&disk_to_dev(md->disk)->kobj,
3439 uint32_t dm_next_uevent_seq(struct mapped_device *md)
3441 return atomic_add_return(1, &md->uevent_seq);
3444 uint32_t dm_get_event_nr(struct mapped_device *md)
3446 return atomic_read(&md->event_nr);
3449 int dm_wait_event(struct mapped_device *md, int event_nr)
3451 return wait_event_interruptible(md->eventq,
3452 (event_nr != atomic_read(&md->event_nr)));
3455 void dm_uevent_add(struct mapped_device *md, struct list_head *elist)
3457 unsigned long flags;
3459 spin_lock_irqsave(&md->uevent_lock, flags);
3460 list_add(elist, &md->uevent_list);
3461 spin_unlock_irqrestore(&md->uevent_lock, flags);
3465 * The gendisk is only valid as long as you have a reference
3468 struct gendisk *dm_disk(struct mapped_device *md)
3472 EXPORT_SYMBOL_GPL(dm_disk);
3474 struct kobject *dm_kobject(struct mapped_device *md)
3476 return &md->kobj_holder.kobj;
3479 struct mapped_device *dm_get_from_kobject(struct kobject *kobj)
3481 struct mapped_device *md;
3483 md = container_of(kobj, struct mapped_device, kobj_holder.kobj);
3485 if (test_bit(DMF_FREEING, &md->flags) ||
3493 int dm_suspended_md(struct mapped_device *md)
3495 return test_bit(DMF_SUSPENDED, &md->flags);
3498 int dm_suspended_internally_md(struct mapped_device *md)
3500 return test_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
3503 int dm_test_deferred_remove_flag(struct mapped_device *md)
3505 return test_bit(DMF_DEFERRED_REMOVE, &md->flags);
3508 int dm_suspended(struct dm_target *ti)
3510 return dm_suspended_md(dm_table_get_md(ti->table));
3512 EXPORT_SYMBOL_GPL(dm_suspended);
3514 int dm_noflush_suspending(struct dm_target *ti)
3516 return __noflush_suspending(dm_table_get_md(ti->table));
3518 EXPORT_SYMBOL_GPL(dm_noflush_suspending);
3520 struct dm_md_mempools *dm_alloc_md_mempools(struct mapped_device *md, unsigned type,
3521 unsigned integrity, unsigned per_io_data_size)
3523 struct dm_md_mempools *pools = kzalloc_node(sizeof(*pools), GFP_KERNEL, md->numa_node_id);
3524 struct kmem_cache *cachep = NULL;
3525 unsigned int pool_size = 0;
3526 unsigned int front_pad;
3531 type = filter_md_type(type, md);
3534 case DM_TYPE_BIO_BASED:
3536 pool_size = dm_get_reserved_bio_based_ios();
3537 front_pad = roundup(per_io_data_size, __alignof__(struct dm_target_io)) + offsetof(struct dm_target_io, clone);
3539 case DM_TYPE_REQUEST_BASED:
3540 cachep = _rq_tio_cache;
3541 pool_size = dm_get_reserved_rq_based_ios();
3542 pools->rq_pool = mempool_create_slab_pool(pool_size, _rq_cache);
3543 if (!pools->rq_pool)
3545 /* fall through to setup remaining rq-based pools */
3546 case DM_TYPE_MQ_REQUEST_BASED:
3548 pool_size = dm_get_reserved_rq_based_ios();
3549 front_pad = offsetof(struct dm_rq_clone_bio_info, clone);
3550 /* per_io_data_size is used for blk-mq pdu at queue allocation */
3557 pools->io_pool = mempool_create_slab_pool(pool_size, cachep);
3558 if (!pools->io_pool)
3562 pools->bs = bioset_create_nobvec(pool_size, front_pad);
3566 if (integrity && bioset_integrity_create(pools->bs, pool_size))
3572 dm_free_md_mempools(pools);
3577 void dm_free_md_mempools(struct dm_md_mempools *pools)
3582 mempool_destroy(pools->io_pool);
3583 mempool_destroy(pools->rq_pool);
3586 bioset_free(pools->bs);
3591 static int dm_pr_register(struct block_device *bdev, u64 old_key, u64 new_key,
3594 struct mapped_device *md = bdev->bd_disk->private_data;
3595 const struct pr_ops *ops;
3599 r = dm_grab_bdev_for_ioctl(md, &bdev, &mode);
3603 ops = bdev->bd_disk->fops->pr_ops;
3604 if (ops && ops->pr_register)
3605 r = ops->pr_register(bdev, old_key, new_key, flags);
3613 static int dm_pr_reserve(struct block_device *bdev, u64 key, enum pr_type type,
3616 struct mapped_device *md = bdev->bd_disk->private_data;
3617 const struct pr_ops *ops;
3621 r = dm_grab_bdev_for_ioctl(md, &bdev, &mode);
3625 ops = bdev->bd_disk->fops->pr_ops;
3626 if (ops && ops->pr_reserve)
3627 r = ops->pr_reserve(bdev, key, type, flags);
3635 static int dm_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
3637 struct mapped_device *md = bdev->bd_disk->private_data;
3638 const struct pr_ops *ops;
3642 r = dm_grab_bdev_for_ioctl(md, &bdev, &mode);
3646 ops = bdev->bd_disk->fops->pr_ops;
3647 if (ops && ops->pr_release)
3648 r = ops->pr_release(bdev, key, type);
3656 static int dm_pr_preempt(struct block_device *bdev, u64 old_key, u64 new_key,
3657 enum pr_type type, bool abort)
3659 struct mapped_device *md = bdev->bd_disk->private_data;
3660 const struct pr_ops *ops;
3664 r = dm_grab_bdev_for_ioctl(md, &bdev, &mode);
3668 ops = bdev->bd_disk->fops->pr_ops;
3669 if (ops && ops->pr_preempt)
3670 r = ops->pr_preempt(bdev, old_key, new_key, type, abort);
3678 static int dm_pr_clear(struct block_device *bdev, u64 key)
3680 struct mapped_device *md = bdev->bd_disk->private_data;
3681 const struct pr_ops *ops;
3685 r = dm_grab_bdev_for_ioctl(md, &bdev, &mode);
3689 ops = bdev->bd_disk->fops->pr_ops;
3690 if (ops && ops->pr_clear)
3691 r = ops->pr_clear(bdev, key);
3699 static const struct pr_ops dm_pr_ops = {
3700 .pr_register = dm_pr_register,
3701 .pr_reserve = dm_pr_reserve,
3702 .pr_release = dm_pr_release,
3703 .pr_preempt = dm_pr_preempt,
3704 .pr_clear = dm_pr_clear,
3707 static const struct block_device_operations dm_blk_dops = {
3708 .open = dm_blk_open,
3709 .release = dm_blk_close,
3710 .ioctl = dm_blk_ioctl,
3711 .getgeo = dm_blk_getgeo,
3712 .pr_ops = &dm_pr_ops,
3713 .owner = THIS_MODULE
3719 module_init(dm_init);
3720 module_exit(dm_exit);
3722 module_param(major, uint, 0);
3723 MODULE_PARM_DESC(major, "The major number of the device mapper");
3725 module_param(reserved_bio_based_ios, uint, S_IRUGO | S_IWUSR);
3726 MODULE_PARM_DESC(reserved_bio_based_ios, "Reserved IOs in bio-based mempools");
3728 module_param(reserved_rq_based_ios, uint, S_IRUGO | S_IWUSR);
3729 MODULE_PARM_DESC(reserved_rq_based_ios, "Reserved IOs in request-based mempools");
3731 module_param(use_blk_mq, bool, S_IRUGO | S_IWUSR);
3732 MODULE_PARM_DESC(use_blk_mq, "Use block multiqueue for request-based DM devices");
3734 module_param(dm_mq_nr_hw_queues, uint, S_IRUGO | S_IWUSR);
3735 MODULE_PARM_DESC(dm_mq_nr_hw_queues, "Number of hardware queues for request-based dm-mq devices");
3737 module_param(dm_mq_queue_depth, uint, S_IRUGO | S_IWUSR);
3738 MODULE_PARM_DESC(dm_mq_queue_depth, "Queue depth for request-based dm-mq devices");
3740 module_param(dm_numa_node, int, S_IRUGO | S_IWUSR);
3741 MODULE_PARM_DESC(dm_numa_node, "NUMA node for DM device memory allocations");
3743 MODULE_DESCRIPTION(DM_NAME " driver");
3744 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
3745 MODULE_LICENSE("GPL");